]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
0fe23479 | 2 | * Copyright (C) 2001 Jens Axboe <[email protected]> |
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> | |
a27bb332 | 22 | #include <linux/uio.h> |
852c788f | 23 | #include <linux/iocontext.h> |
1da177e4 LT |
24 | #include <linux/slab.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/kernel.h> | |
630d9c47 | 27 | #include <linux/export.h> |
1da177e4 LT |
28 | #include <linux/mempool.h> |
29 | #include <linux/workqueue.h> | |
852c788f | 30 | #include <linux/cgroup.h> |
1da177e4 | 31 | |
55782138 | 32 | #include <trace/events/block.h> |
0bfc2455 | 33 | |
392ddc32 JA |
34 | /* |
35 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
36 | * itself, to shrink a bio data allocation from two mempool calls to one | |
37 | */ | |
38 | #define BIO_INLINE_VECS 4 | |
39 | ||
1da177e4 LT |
40 | /* |
41 | * if you change this list, also change bvec_alloc or things will | |
42 | * break badly! cannot be bigger than what you can fit into an | |
43 | * unsigned short | |
44 | */ | |
1da177e4 | 45 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
df677140 | 46 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
1da177e4 LT |
47 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
48 | }; | |
49 | #undef BV | |
50 | ||
1da177e4 LT |
51 | /* |
52 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
53 | * IO code that does not need private memory pools. | |
54 | */ | |
51d654e1 | 55 | struct bio_set *fs_bio_set; |
3f86a82a | 56 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 57 | |
bb799ca0 JA |
58 | /* |
59 | * Our slab pool management | |
60 | */ | |
61 | struct bio_slab { | |
62 | struct kmem_cache *slab; | |
63 | unsigned int slab_ref; | |
64 | unsigned int slab_size; | |
65 | char name[8]; | |
66 | }; | |
67 | static DEFINE_MUTEX(bio_slab_lock); | |
68 | static struct bio_slab *bio_slabs; | |
69 | static unsigned int bio_slab_nr, bio_slab_max; | |
70 | ||
71 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
72 | { | |
73 | unsigned int sz = sizeof(struct bio) + extra_size; | |
74 | struct kmem_cache *slab = NULL; | |
389d7b26 | 75 | struct bio_slab *bslab, *new_bio_slabs; |
386bc35a | 76 | unsigned int new_bio_slab_max; |
bb799ca0 JA |
77 | unsigned int i, entry = -1; |
78 | ||
79 | mutex_lock(&bio_slab_lock); | |
80 | ||
81 | i = 0; | |
82 | while (i < bio_slab_nr) { | |
f06f135d | 83 | bslab = &bio_slabs[i]; |
bb799ca0 JA |
84 | |
85 | if (!bslab->slab && entry == -1) | |
86 | entry = i; | |
87 | else if (bslab->slab_size == sz) { | |
88 | slab = bslab->slab; | |
89 | bslab->slab_ref++; | |
90 | break; | |
91 | } | |
92 | i++; | |
93 | } | |
94 | ||
95 | if (slab) | |
96 | goto out_unlock; | |
97 | ||
98 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
386bc35a | 99 | new_bio_slab_max = bio_slab_max << 1; |
389d7b26 | 100 | new_bio_slabs = krealloc(bio_slabs, |
386bc35a | 101 | new_bio_slab_max * sizeof(struct bio_slab), |
389d7b26 AK |
102 | GFP_KERNEL); |
103 | if (!new_bio_slabs) | |
bb799ca0 | 104 | goto out_unlock; |
386bc35a | 105 | bio_slab_max = new_bio_slab_max; |
389d7b26 | 106 | bio_slabs = new_bio_slabs; |
bb799ca0 JA |
107 | } |
108 | if (entry == -1) | |
109 | entry = bio_slab_nr++; | |
110 | ||
111 | bslab = &bio_slabs[entry]; | |
112 | ||
113 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
6a241483 MP |
114 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
115 | SLAB_HWCACHE_ALIGN, NULL); | |
bb799ca0 JA |
116 | if (!slab) |
117 | goto out_unlock; | |
118 | ||
bb799ca0 JA |
119 | bslab->slab = slab; |
120 | bslab->slab_ref = 1; | |
121 | bslab->slab_size = sz; | |
122 | out_unlock: | |
123 | mutex_unlock(&bio_slab_lock); | |
124 | return slab; | |
125 | } | |
126 | ||
127 | static void bio_put_slab(struct bio_set *bs) | |
128 | { | |
129 | struct bio_slab *bslab = NULL; | |
130 | unsigned int i; | |
131 | ||
132 | mutex_lock(&bio_slab_lock); | |
133 | ||
134 | for (i = 0; i < bio_slab_nr; i++) { | |
135 | if (bs->bio_slab == bio_slabs[i].slab) { | |
136 | bslab = &bio_slabs[i]; | |
137 | break; | |
138 | } | |
139 | } | |
140 | ||
141 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
142 | goto out; | |
143 | ||
144 | WARN_ON(!bslab->slab_ref); | |
145 | ||
146 | if (--bslab->slab_ref) | |
147 | goto out; | |
148 | ||
149 | kmem_cache_destroy(bslab->slab); | |
150 | bslab->slab = NULL; | |
151 | ||
152 | out: | |
153 | mutex_unlock(&bio_slab_lock); | |
154 | } | |
155 | ||
7ba1ba12 MP |
156 | unsigned int bvec_nr_vecs(unsigned short idx) |
157 | { | |
158 | return bvec_slabs[idx].nr_vecs; | |
159 | } | |
160 | ||
9f060e22 | 161 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 JA |
162 | { |
163 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); | |
164 | ||
165 | if (idx == BIOVEC_MAX_IDX) | |
9f060e22 | 166 | mempool_free(bv, pool); |
bb799ca0 JA |
167 | else { |
168 | struct biovec_slab *bvs = bvec_slabs + idx; | |
169 | ||
170 | kmem_cache_free(bvs->slab, bv); | |
171 | } | |
172 | } | |
173 | ||
9f060e22 KO |
174 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
175 | mempool_t *pool) | |
1da177e4 LT |
176 | { |
177 | struct bio_vec *bvl; | |
1da177e4 | 178 | |
7ff9345f JA |
179 | /* |
180 | * see comment near bvec_array define! | |
181 | */ | |
182 | switch (nr) { | |
183 | case 1: | |
184 | *idx = 0; | |
185 | break; | |
186 | case 2 ... 4: | |
187 | *idx = 1; | |
188 | break; | |
189 | case 5 ... 16: | |
190 | *idx = 2; | |
191 | break; | |
192 | case 17 ... 64: | |
193 | *idx = 3; | |
194 | break; | |
195 | case 65 ... 128: | |
196 | *idx = 4; | |
197 | break; | |
198 | case 129 ... BIO_MAX_PAGES: | |
199 | *idx = 5; | |
200 | break; | |
201 | default: | |
202 | return NULL; | |
203 | } | |
204 | ||
205 | /* | |
206 | * idx now points to the pool we want to allocate from. only the | |
207 | * 1-vec entry pool is mempool backed. | |
208 | */ | |
209 | if (*idx == BIOVEC_MAX_IDX) { | |
210 | fallback: | |
9f060e22 | 211 | bvl = mempool_alloc(pool, gfp_mask); |
7ff9345f JA |
212 | } else { |
213 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
214 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); | |
215 | ||
0a0d96b0 | 216 | /* |
7ff9345f JA |
217 | * Make this allocation restricted and don't dump info on |
218 | * allocation failures, since we'll fallback to the mempool | |
219 | * in case of failure. | |
0a0d96b0 | 220 | */ |
7ff9345f | 221 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 222 | |
0a0d96b0 | 223 | /* |
7ff9345f JA |
224 | * Try a slab allocation. If this fails and __GFP_WAIT |
225 | * is set, retry with the 1-entry mempool | |
0a0d96b0 | 226 | */ |
7ff9345f JA |
227 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
228 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { | |
229 | *idx = BIOVEC_MAX_IDX; | |
230 | goto fallback; | |
231 | } | |
232 | } | |
233 | ||
1da177e4 LT |
234 | return bvl; |
235 | } | |
236 | ||
4254bba1 | 237 | static void __bio_free(struct bio *bio) |
1da177e4 | 238 | { |
4254bba1 | 239 | bio_disassociate_task(bio); |
1da177e4 | 240 | |
7ba1ba12 | 241 | if (bio_integrity(bio)) |
1e2a410f | 242 | bio_integrity_free(bio); |
4254bba1 | 243 | } |
7ba1ba12 | 244 | |
4254bba1 KO |
245 | static void bio_free(struct bio *bio) |
246 | { | |
247 | struct bio_set *bs = bio->bi_pool; | |
248 | void *p; | |
249 | ||
250 | __bio_free(bio); | |
251 | ||
252 | if (bs) { | |
a38352e0 | 253 | if (bio_flagged(bio, BIO_OWNS_VEC)) |
9f060e22 | 254 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio)); |
4254bba1 KO |
255 | |
256 | /* | |
257 | * If we have front padding, adjust the bio pointer before freeing | |
258 | */ | |
259 | p = bio; | |
bb799ca0 JA |
260 | p -= bs->front_pad; |
261 | ||
4254bba1 KO |
262 | mempool_free(p, bs->bio_pool); |
263 | } else { | |
264 | /* Bio was allocated by bio_kmalloc() */ | |
265 | kfree(bio); | |
266 | } | |
3676347a PO |
267 | } |
268 | ||
858119e1 | 269 | void bio_init(struct bio *bio) |
1da177e4 | 270 | { |
2b94de55 | 271 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 272 | bio->bi_flags = 1 << BIO_UPTODATE; |
196d38bc | 273 | atomic_set(&bio->bi_remaining, 1); |
1da177e4 | 274 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 | 275 | } |
a112a71d | 276 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 277 | |
f44b48c7 KO |
278 | /** |
279 | * bio_reset - reinitialize a bio | |
280 | * @bio: bio to reset | |
281 | * | |
282 | * Description: | |
283 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
284 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
285 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
286 | * comment in struct bio. | |
287 | */ | |
288 | void bio_reset(struct bio *bio) | |
289 | { | |
290 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
291 | ||
4254bba1 | 292 | __bio_free(bio); |
f44b48c7 KO |
293 | |
294 | memset(bio, 0, BIO_RESET_BYTES); | |
295 | bio->bi_flags = flags|(1 << BIO_UPTODATE); | |
196d38bc | 296 | atomic_set(&bio->bi_remaining, 1); |
f44b48c7 KO |
297 | } |
298 | EXPORT_SYMBOL(bio_reset); | |
299 | ||
196d38bc KO |
300 | static void bio_chain_endio(struct bio *bio, int error) |
301 | { | |
302 | bio_endio(bio->bi_private, error); | |
303 | bio_put(bio); | |
304 | } | |
305 | ||
306 | /** | |
307 | * bio_chain - chain bio completions | |
1051a902 RD |
308 | * @bio: the target bio |
309 | * @parent: the @bio's parent bio | |
196d38bc KO |
310 | * |
311 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
312 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
313 | * completed; the chained bio will also be freed when it completes. | |
314 | * | |
315 | * The caller must not set bi_private or bi_end_io in @bio. | |
316 | */ | |
317 | void bio_chain(struct bio *bio, struct bio *parent) | |
318 | { | |
319 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
320 | ||
321 | bio->bi_private = parent; | |
322 | bio->bi_end_io = bio_chain_endio; | |
323 | atomic_inc(&parent->bi_remaining); | |
324 | } | |
325 | EXPORT_SYMBOL(bio_chain); | |
326 | ||
df2cb6da KO |
327 | static void bio_alloc_rescue(struct work_struct *work) |
328 | { | |
329 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
330 | struct bio *bio; | |
331 | ||
332 | while (1) { | |
333 | spin_lock(&bs->rescue_lock); | |
334 | bio = bio_list_pop(&bs->rescue_list); | |
335 | spin_unlock(&bs->rescue_lock); | |
336 | ||
337 | if (!bio) | |
338 | break; | |
339 | ||
340 | generic_make_request(bio); | |
341 | } | |
342 | } | |
343 | ||
344 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
345 | { | |
346 | struct bio_list punt, nopunt; | |
347 | struct bio *bio; | |
348 | ||
349 | /* | |
350 | * In order to guarantee forward progress we must punt only bios that | |
351 | * were allocated from this bio_set; otherwise, if there was a bio on | |
352 | * there for a stacking driver higher up in the stack, processing it | |
353 | * could require allocating bios from this bio_set, and doing that from | |
354 | * our own rescuer would be bad. | |
355 | * | |
356 | * Since bio lists are singly linked, pop them all instead of trying to | |
357 | * remove from the middle of the list: | |
358 | */ | |
359 | ||
360 | bio_list_init(&punt); | |
361 | bio_list_init(&nopunt); | |
362 | ||
363 | while ((bio = bio_list_pop(current->bio_list))) | |
364 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
365 | ||
366 | *current->bio_list = nopunt; | |
367 | ||
368 | spin_lock(&bs->rescue_lock); | |
369 | bio_list_merge(&bs->rescue_list, &punt); | |
370 | spin_unlock(&bs->rescue_lock); | |
371 | ||
372 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
373 | } | |
374 | ||
1da177e4 LT |
375 | /** |
376 | * bio_alloc_bioset - allocate a bio for I/O | |
377 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
378 | * @nr_iovecs: number of iovecs to pre-allocate | |
db18efac | 379 | * @bs: the bio_set to allocate from. |
1da177e4 LT |
380 | * |
381 | * Description: | |
3f86a82a KO |
382 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
383 | * backed by the @bs's mempool. | |
384 | * | |
385 | * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be | |
386 | * able to allocate a bio. This is due to the mempool guarantees. To make this | |
387 | * work, callers must never allocate more than 1 bio at a time from this pool. | |
388 | * Callers that need to allocate more than 1 bio must always submit the | |
389 | * previously allocated bio for IO before attempting to allocate a new one. | |
390 | * Failure to do so can cause deadlocks under memory pressure. | |
391 | * | |
df2cb6da KO |
392 | * Note that when running under generic_make_request() (i.e. any block |
393 | * driver), bios are not submitted until after you return - see the code in | |
394 | * generic_make_request() that converts recursion into iteration, to prevent | |
395 | * stack overflows. | |
396 | * | |
397 | * This would normally mean allocating multiple bios under | |
398 | * generic_make_request() would be susceptible to deadlocks, but we have | |
399 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
400 | * thread. | |
401 | * | |
402 | * However, we do not guarantee forward progress for allocations from other | |
403 | * mempools. Doing multiple allocations from the same mempool under | |
404 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
405 | * for per bio allocations. | |
406 | * | |
3f86a82a KO |
407 | * RETURNS: |
408 | * Pointer to new bio on success, NULL on failure. | |
409 | */ | |
dd0fc66f | 410 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 | 411 | { |
df2cb6da | 412 | gfp_t saved_gfp = gfp_mask; |
3f86a82a KO |
413 | unsigned front_pad; |
414 | unsigned inline_vecs; | |
451a9ebf | 415 | unsigned long idx = BIO_POOL_NONE; |
34053979 | 416 | struct bio_vec *bvl = NULL; |
451a9ebf TH |
417 | struct bio *bio; |
418 | void *p; | |
419 | ||
3f86a82a KO |
420 | if (!bs) { |
421 | if (nr_iovecs > UIO_MAXIOV) | |
422 | return NULL; | |
423 | ||
424 | p = kmalloc(sizeof(struct bio) + | |
425 | nr_iovecs * sizeof(struct bio_vec), | |
426 | gfp_mask); | |
427 | front_pad = 0; | |
428 | inline_vecs = nr_iovecs; | |
429 | } else { | |
d8f429e1 JN |
430 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
431 | if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0)) | |
432 | return NULL; | |
df2cb6da KO |
433 | /* |
434 | * generic_make_request() converts recursion to iteration; this | |
435 | * means if we're running beneath it, any bios we allocate and | |
436 | * submit will not be submitted (and thus freed) until after we | |
437 | * return. | |
438 | * | |
439 | * This exposes us to a potential deadlock if we allocate | |
440 | * multiple bios from the same bio_set() while running | |
441 | * underneath generic_make_request(). If we were to allocate | |
442 | * multiple bios (say a stacking block driver that was splitting | |
443 | * bios), we would deadlock if we exhausted the mempool's | |
444 | * reserve. | |
445 | * | |
446 | * We solve this, and guarantee forward progress, with a rescuer | |
447 | * workqueue per bio_set. If we go to allocate and there are | |
448 | * bios on current->bio_list, we first try the allocation | |
449 | * without __GFP_WAIT; if that fails, we punt those bios we | |
450 | * would be blocking to the rescuer workqueue before we retry | |
451 | * with the original gfp_flags. | |
452 | */ | |
453 | ||
454 | if (current->bio_list && !bio_list_empty(current->bio_list)) | |
455 | gfp_mask &= ~__GFP_WAIT; | |
456 | ||
3f86a82a | 457 | p = mempool_alloc(bs->bio_pool, gfp_mask); |
df2cb6da KO |
458 | if (!p && gfp_mask != saved_gfp) { |
459 | punt_bios_to_rescuer(bs); | |
460 | gfp_mask = saved_gfp; | |
461 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
462 | } | |
463 | ||
3f86a82a KO |
464 | front_pad = bs->front_pad; |
465 | inline_vecs = BIO_INLINE_VECS; | |
466 | } | |
467 | ||
451a9ebf TH |
468 | if (unlikely(!p)) |
469 | return NULL; | |
1da177e4 | 470 | |
3f86a82a | 471 | bio = p + front_pad; |
34053979 IM |
472 | bio_init(bio); |
473 | ||
3f86a82a | 474 | if (nr_iovecs > inline_vecs) { |
9f060e22 | 475 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); |
df2cb6da KO |
476 | if (!bvl && gfp_mask != saved_gfp) { |
477 | punt_bios_to_rescuer(bs); | |
478 | gfp_mask = saved_gfp; | |
9f060e22 | 479 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); |
df2cb6da KO |
480 | } |
481 | ||
34053979 IM |
482 | if (unlikely(!bvl)) |
483 | goto err_free; | |
a38352e0 KO |
484 | |
485 | bio->bi_flags |= 1 << BIO_OWNS_VEC; | |
3f86a82a KO |
486 | } else if (nr_iovecs) { |
487 | bvl = bio->bi_inline_vecs; | |
1da177e4 | 488 | } |
3f86a82a KO |
489 | |
490 | bio->bi_pool = bs; | |
34053979 IM |
491 | bio->bi_flags |= idx << BIO_POOL_OFFSET; |
492 | bio->bi_max_vecs = nr_iovecs; | |
34053979 | 493 | bio->bi_io_vec = bvl; |
1da177e4 | 494 | return bio; |
34053979 IM |
495 | |
496 | err_free: | |
451a9ebf | 497 | mempool_free(p, bs->bio_pool); |
34053979 | 498 | return NULL; |
1da177e4 | 499 | } |
a112a71d | 500 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 501 | |
1da177e4 LT |
502 | void zero_fill_bio(struct bio *bio) |
503 | { | |
504 | unsigned long flags; | |
7988613b KO |
505 | struct bio_vec bv; |
506 | struct bvec_iter iter; | |
1da177e4 | 507 | |
7988613b KO |
508 | bio_for_each_segment(bv, bio, iter) { |
509 | char *data = bvec_kmap_irq(&bv, &flags); | |
510 | memset(data, 0, bv.bv_len); | |
511 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
512 | bvec_kunmap_irq(data, &flags); |
513 | } | |
514 | } | |
515 | EXPORT_SYMBOL(zero_fill_bio); | |
516 | ||
517 | /** | |
518 | * bio_put - release a reference to a bio | |
519 | * @bio: bio to release reference to | |
520 | * | |
521 | * Description: | |
522 | * Put a reference to a &struct bio, either one you have gotten with | |
ad0bf110 | 523 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. |
1da177e4 LT |
524 | **/ |
525 | void bio_put(struct bio *bio) | |
526 | { | |
527 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
528 | ||
529 | /* | |
530 | * last put frees it | |
531 | */ | |
4254bba1 KO |
532 | if (atomic_dec_and_test(&bio->bi_cnt)) |
533 | bio_free(bio); | |
1da177e4 | 534 | } |
a112a71d | 535 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 536 | |
165125e1 | 537 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
538 | { |
539 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
540 | blk_recount_segments(q, bio); | |
541 | ||
542 | return bio->bi_phys_segments; | |
543 | } | |
a112a71d | 544 | EXPORT_SYMBOL(bio_phys_segments); |
1da177e4 | 545 | |
59d276fe KO |
546 | /** |
547 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
548 | * @bio: destination bio | |
549 | * @bio_src: bio to clone | |
550 | * | |
551 | * Clone a &bio. Caller will own the returned bio, but not | |
552 | * the actual data it points to. Reference count of returned | |
553 | * bio will be one. | |
554 | * | |
555 | * Caller must ensure that @bio_src is not freed before @bio. | |
556 | */ | |
557 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
558 | { | |
559 | BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE); | |
560 | ||
561 | /* | |
562 | * most users will be overriding ->bi_bdev with a new target, | |
563 | * so we don't set nor calculate new physical/hw segment counts here | |
564 | */ | |
565 | bio->bi_bdev = bio_src->bi_bdev; | |
566 | bio->bi_flags |= 1 << BIO_CLONED; | |
567 | bio->bi_rw = bio_src->bi_rw; | |
568 | bio->bi_iter = bio_src->bi_iter; | |
569 | bio->bi_io_vec = bio_src->bi_io_vec; | |
570 | } | |
571 | EXPORT_SYMBOL(__bio_clone_fast); | |
572 | ||
573 | /** | |
574 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
575 | * @bio: bio to clone | |
576 | * @gfp_mask: allocation priority | |
577 | * @bs: bio_set to allocate from | |
578 | * | |
579 | * Like __bio_clone_fast, only also allocates the returned bio | |
580 | */ | |
581 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
582 | { | |
583 | struct bio *b; | |
584 | ||
585 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
586 | if (!b) | |
587 | return NULL; | |
588 | ||
589 | __bio_clone_fast(b, bio); | |
590 | ||
591 | if (bio_integrity(bio)) { | |
592 | int ret; | |
593 | ||
594 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
595 | ||
596 | if (ret < 0) { | |
597 | bio_put(b); | |
598 | return NULL; | |
599 | } | |
600 | } | |
601 | ||
602 | return b; | |
603 | } | |
604 | EXPORT_SYMBOL(bio_clone_fast); | |
605 | ||
1da177e4 | 606 | /** |
bdb53207 KO |
607 | * bio_clone_bioset - clone a bio |
608 | * @bio_src: bio to clone | |
1da177e4 | 609 | * @gfp_mask: allocation priority |
bf800ef1 | 610 | * @bs: bio_set to allocate from |
1da177e4 | 611 | * |
bdb53207 KO |
612 | * Clone bio. Caller will own the returned bio, but not the actual data it |
613 | * points to. Reference count of returned bio will be one. | |
1da177e4 | 614 | */ |
bdb53207 | 615 | struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, |
bf800ef1 | 616 | struct bio_set *bs) |
1da177e4 | 617 | { |
bdb53207 KO |
618 | struct bvec_iter iter; |
619 | struct bio_vec bv; | |
620 | struct bio *bio; | |
1da177e4 | 621 | |
bdb53207 KO |
622 | /* |
623 | * Pre immutable biovecs, __bio_clone() used to just do a memcpy from | |
624 | * bio_src->bi_io_vec to bio->bi_io_vec. | |
625 | * | |
626 | * We can't do that anymore, because: | |
627 | * | |
628 | * - The point of cloning the biovec is to produce a bio with a biovec | |
629 | * the caller can modify: bi_idx and bi_bvec_done should be 0. | |
630 | * | |
631 | * - The original bio could've had more than BIO_MAX_PAGES biovecs; if | |
632 | * we tried to clone the whole thing bio_alloc_bioset() would fail. | |
633 | * But the clone should succeed as long as the number of biovecs we | |
634 | * actually need to allocate is fewer than BIO_MAX_PAGES. | |
635 | * | |
636 | * - Lastly, bi_vcnt should not be looked at or relied upon by code | |
637 | * that does not own the bio - reason being drivers don't use it for | |
638 | * iterating over the biovec anymore, so expecting it to be kept up | |
639 | * to date (i.e. for clones that share the parent biovec) is just | |
640 | * asking for trouble and would force extra work on | |
641 | * __bio_clone_fast() anyways. | |
642 | */ | |
643 | ||
8423ae3d | 644 | bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); |
bdb53207 | 645 | if (!bio) |
7ba1ba12 MP |
646 | return NULL; |
647 | ||
bdb53207 KO |
648 | bio->bi_bdev = bio_src->bi_bdev; |
649 | bio->bi_rw = bio_src->bi_rw; | |
650 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; | |
651 | bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; | |
7ba1ba12 | 652 | |
8423ae3d KO |
653 | if (bio->bi_rw & REQ_DISCARD) |
654 | goto integrity_clone; | |
655 | ||
656 | if (bio->bi_rw & REQ_WRITE_SAME) { | |
657 | bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; | |
658 | goto integrity_clone; | |
659 | } | |
660 | ||
bdb53207 KO |
661 | bio_for_each_segment(bv, bio_src, iter) |
662 | bio->bi_io_vec[bio->bi_vcnt++] = bv; | |
7ba1ba12 | 663 | |
8423ae3d | 664 | integrity_clone: |
bdb53207 KO |
665 | if (bio_integrity(bio_src)) { |
666 | int ret; | |
7ba1ba12 | 667 | |
bdb53207 | 668 | ret = bio_integrity_clone(bio, bio_src, gfp_mask); |
059ea331 | 669 | if (ret < 0) { |
bdb53207 | 670 | bio_put(bio); |
7ba1ba12 | 671 | return NULL; |
059ea331 | 672 | } |
3676347a | 673 | } |
1da177e4 | 674 | |
bdb53207 | 675 | return bio; |
1da177e4 | 676 | } |
bf800ef1 | 677 | EXPORT_SYMBOL(bio_clone_bioset); |
1da177e4 LT |
678 | |
679 | /** | |
680 | * bio_get_nr_vecs - return approx number of vecs | |
681 | * @bdev: I/O target | |
682 | * | |
683 | * Return the approximate number of pages we can send to this target. | |
684 | * There's no guarantee that you will be able to fit this number of pages | |
685 | * into a bio, it does not account for dynamic restrictions that vary | |
686 | * on offset. | |
687 | */ | |
688 | int bio_get_nr_vecs(struct block_device *bdev) | |
689 | { | |
165125e1 | 690 | struct request_queue *q = bdev_get_queue(bdev); |
f908ee94 BS |
691 | int nr_pages; |
692 | ||
693 | nr_pages = min_t(unsigned, | |
5abebfdd KO |
694 | queue_max_segments(q), |
695 | queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); | |
f908ee94 BS |
696 | |
697 | return min_t(unsigned, nr_pages, BIO_MAX_PAGES); | |
698 | ||
1da177e4 | 699 | } |
a112a71d | 700 | EXPORT_SYMBOL(bio_get_nr_vecs); |
1da177e4 | 701 | |
165125e1 | 702 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 | 703 | *page, unsigned int len, unsigned int offset, |
34f2fd8d | 704 | unsigned int max_sectors) |
1da177e4 LT |
705 | { |
706 | int retried_segments = 0; | |
707 | struct bio_vec *bvec; | |
708 | ||
709 | /* | |
710 | * cloned bio must not modify vec list | |
711 | */ | |
712 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
713 | return 0; | |
714 | ||
4f024f37 | 715 | if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
716 | return 0; |
717 | ||
80cfd548 JA |
718 | /* |
719 | * For filesystems with a blocksize smaller than the pagesize | |
720 | * we will often be called with the same page as last time and | |
721 | * a consecutive offset. Optimize this special case. | |
722 | */ | |
723 | if (bio->bi_vcnt > 0) { | |
724 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
725 | ||
726 | if (page == prev->bv_page && | |
727 | offset == prev->bv_offset + prev->bv_len) { | |
1d616585 | 728 | unsigned int prev_bv_len = prev->bv_len; |
80cfd548 | 729 | prev->bv_len += len; |
cc371e66 AK |
730 | |
731 | if (q->merge_bvec_fn) { | |
732 | struct bvec_merge_data bvm = { | |
1d616585 DM |
733 | /* prev_bvec is already charged in |
734 | bi_size, discharge it in order to | |
735 | simulate merging updated prev_bvec | |
736 | as new bvec. */ | |
cc371e66 | 737 | .bi_bdev = bio->bi_bdev, |
4f024f37 KO |
738 | .bi_sector = bio->bi_iter.bi_sector, |
739 | .bi_size = bio->bi_iter.bi_size - | |
740 | prev_bv_len, | |
cc371e66 AK |
741 | .bi_rw = bio->bi_rw, |
742 | }; | |
743 | ||
8bf8c376 | 744 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { |
cc371e66 AK |
745 | prev->bv_len -= len; |
746 | return 0; | |
747 | } | |
80cfd548 JA |
748 | } |
749 | ||
fcbf6a08 | 750 | bio->bi_iter.bi_size += len; |
80cfd548 JA |
751 | goto done; |
752 | } | |
66cb45aa JA |
753 | |
754 | /* | |
755 | * If the queue doesn't support SG gaps and adding this | |
756 | * offset would create a gap, disallow it. | |
757 | */ | |
758 | if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) && | |
759 | bvec_gap_to_prev(prev, offset)) | |
760 | return 0; | |
80cfd548 JA |
761 | } |
762 | ||
763 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
764 | return 0; |
765 | ||
766 | /* | |
fcbf6a08 ML |
767 | * setup the new entry, we might clear it again later if we |
768 | * cannot add the page | |
769 | */ | |
770 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
771 | bvec->bv_page = page; | |
772 | bvec->bv_len = len; | |
773 | bvec->bv_offset = offset; | |
774 | bio->bi_vcnt++; | |
775 | bio->bi_phys_segments++; | |
776 | bio->bi_iter.bi_size += len; | |
777 | ||
778 | /* | |
779 | * Perform a recount if the number of segments is greater | |
780 | * than queue_max_segments(q). | |
1da177e4 LT |
781 | */ |
782 | ||
fcbf6a08 | 783 | while (bio->bi_phys_segments > queue_max_segments(q)) { |
1da177e4 LT |
784 | |
785 | if (retried_segments) | |
fcbf6a08 | 786 | goto failed; |
1da177e4 LT |
787 | |
788 | retried_segments = 1; | |
789 | blk_recount_segments(q, bio); | |
790 | } | |
791 | ||
1da177e4 LT |
792 | /* |
793 | * if queue has other restrictions (eg varying max sector size | |
794 | * depending on offset), it can specify a merge_bvec_fn in the | |
795 | * queue to get further control | |
796 | */ | |
797 | if (q->merge_bvec_fn) { | |
cc371e66 AK |
798 | struct bvec_merge_data bvm = { |
799 | .bi_bdev = bio->bi_bdev, | |
4f024f37 | 800 | .bi_sector = bio->bi_iter.bi_sector, |
fcbf6a08 | 801 | .bi_size = bio->bi_iter.bi_size - len, |
cc371e66 AK |
802 | .bi_rw = bio->bi_rw, |
803 | }; | |
804 | ||
1da177e4 LT |
805 | /* |
806 | * merge_bvec_fn() returns number of bytes it can accept | |
807 | * at this offset | |
808 | */ | |
fcbf6a08 ML |
809 | if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) |
810 | goto failed; | |
1da177e4 LT |
811 | } |
812 | ||
813 | /* If we may be able to merge these biovecs, force a recount */ | |
fcbf6a08 | 814 | if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
1da177e4 LT |
815 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
816 | ||
80cfd548 | 817 | done: |
1da177e4 | 818 | return len; |
fcbf6a08 ML |
819 | |
820 | failed: | |
821 | bvec->bv_page = NULL; | |
822 | bvec->bv_len = 0; | |
823 | bvec->bv_offset = 0; | |
824 | bio->bi_vcnt--; | |
825 | bio->bi_iter.bi_size -= len; | |
826 | blk_recount_segments(q, bio); | |
827 | return 0; | |
1da177e4 LT |
828 | } |
829 | ||
6e68af66 MC |
830 | /** |
831 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 832 | * @q: the target queue |
6e68af66 MC |
833 | * @bio: destination bio |
834 | * @page: page to add | |
835 | * @len: vec entry length | |
836 | * @offset: vec entry offset | |
837 | * | |
838 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
c6428084 AG |
839 | * number of reasons, such as the bio being full or target block device |
840 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
841 | * so it is always possible to add a single page to an empty bio. | |
842 | * | |
843 | * This should only be used by REQ_PC bios. | |
6e68af66 | 844 | */ |
165125e1 | 845 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
846 | unsigned int len, unsigned int offset) |
847 | { | |
ae03bf63 MP |
848 | return __bio_add_page(q, bio, page, len, offset, |
849 | queue_max_hw_sectors(q)); | |
6e68af66 | 850 | } |
a112a71d | 851 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 852 | |
1da177e4 LT |
853 | /** |
854 | * bio_add_page - attempt to add page to bio | |
855 | * @bio: destination bio | |
856 | * @page: page to add | |
857 | * @len: vec entry length | |
858 | * @offset: vec entry offset | |
859 | * | |
860 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
c6428084 AG |
861 | * number of reasons, such as the bio being full or target block device |
862 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
863 | * so it is always possible to add a single page to an empty bio. | |
1da177e4 LT |
864 | */ |
865 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
866 | unsigned int offset) | |
867 | { | |
defd94b7 | 868 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
58a4915a | 869 | unsigned int max_sectors; |
762380ad | 870 | |
58a4915a JA |
871 | max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); |
872 | if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size) | |
873 | max_sectors = len >> 9; | |
874 | ||
875 | return __bio_add_page(q, bio, page, len, offset, max_sectors); | |
1da177e4 | 876 | } |
a112a71d | 877 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 878 | |
9e882242 KO |
879 | struct submit_bio_ret { |
880 | struct completion event; | |
881 | int error; | |
882 | }; | |
883 | ||
884 | static void submit_bio_wait_endio(struct bio *bio, int error) | |
885 | { | |
886 | struct submit_bio_ret *ret = bio->bi_private; | |
887 | ||
888 | ret->error = error; | |
889 | complete(&ret->event); | |
890 | } | |
891 | ||
892 | /** | |
893 | * submit_bio_wait - submit a bio, and wait until it completes | |
894 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
895 | * @bio: The &struct bio which describes the I/O | |
896 | * | |
897 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
898 | * bio_endio() on failure. | |
899 | */ | |
900 | int submit_bio_wait(int rw, struct bio *bio) | |
901 | { | |
902 | struct submit_bio_ret ret; | |
903 | ||
904 | rw |= REQ_SYNC; | |
905 | init_completion(&ret.event); | |
906 | bio->bi_private = &ret; | |
907 | bio->bi_end_io = submit_bio_wait_endio; | |
908 | submit_bio(rw, bio); | |
909 | wait_for_completion(&ret.event); | |
910 | ||
911 | return ret.error; | |
912 | } | |
913 | EXPORT_SYMBOL(submit_bio_wait); | |
914 | ||
054bdf64 KO |
915 | /** |
916 | * bio_advance - increment/complete a bio by some number of bytes | |
917 | * @bio: bio to advance | |
918 | * @bytes: number of bytes to complete | |
919 | * | |
920 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
921 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
922 | * be updated on the last bvec as well. | |
923 | * | |
924 | * @bio will then represent the remaining, uncompleted portion of the io. | |
925 | */ | |
926 | void bio_advance(struct bio *bio, unsigned bytes) | |
927 | { | |
928 | if (bio_integrity(bio)) | |
929 | bio_integrity_advance(bio, bytes); | |
930 | ||
4550dd6c | 931 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
932 | } |
933 | EXPORT_SYMBOL(bio_advance); | |
934 | ||
a0787606 KO |
935 | /** |
936 | * bio_alloc_pages - allocates a single page for each bvec in a bio | |
937 | * @bio: bio to allocate pages for | |
938 | * @gfp_mask: flags for allocation | |
939 | * | |
940 | * Allocates pages up to @bio->bi_vcnt. | |
941 | * | |
942 | * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are | |
943 | * freed. | |
944 | */ | |
945 | int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) | |
946 | { | |
947 | int i; | |
948 | struct bio_vec *bv; | |
949 | ||
950 | bio_for_each_segment_all(bv, bio, i) { | |
951 | bv->bv_page = alloc_page(gfp_mask); | |
952 | if (!bv->bv_page) { | |
953 | while (--bv >= bio->bi_io_vec) | |
954 | __free_page(bv->bv_page); | |
955 | return -ENOMEM; | |
956 | } | |
957 | } | |
958 | ||
959 | return 0; | |
960 | } | |
961 | EXPORT_SYMBOL(bio_alloc_pages); | |
962 | ||
16ac3d63 KO |
963 | /** |
964 | * bio_copy_data - copy contents of data buffers from one chain of bios to | |
965 | * another | |
966 | * @src: source bio list | |
967 | * @dst: destination bio list | |
968 | * | |
969 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats | |
970 | * @src and @dst as linked lists of bios. | |
971 | * | |
972 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
973 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
974 | */ | |
975 | void bio_copy_data(struct bio *dst, struct bio *src) | |
976 | { | |
1cb9dda4 KO |
977 | struct bvec_iter src_iter, dst_iter; |
978 | struct bio_vec src_bv, dst_bv; | |
16ac3d63 | 979 | void *src_p, *dst_p; |
1cb9dda4 | 980 | unsigned bytes; |
16ac3d63 | 981 | |
1cb9dda4 KO |
982 | src_iter = src->bi_iter; |
983 | dst_iter = dst->bi_iter; | |
16ac3d63 KO |
984 | |
985 | while (1) { | |
1cb9dda4 KO |
986 | if (!src_iter.bi_size) { |
987 | src = src->bi_next; | |
988 | if (!src) | |
989 | break; | |
16ac3d63 | 990 | |
1cb9dda4 | 991 | src_iter = src->bi_iter; |
16ac3d63 KO |
992 | } |
993 | ||
1cb9dda4 KO |
994 | if (!dst_iter.bi_size) { |
995 | dst = dst->bi_next; | |
996 | if (!dst) | |
997 | break; | |
16ac3d63 | 998 | |
1cb9dda4 | 999 | dst_iter = dst->bi_iter; |
16ac3d63 KO |
1000 | } |
1001 | ||
1cb9dda4 KO |
1002 | src_bv = bio_iter_iovec(src, src_iter); |
1003 | dst_bv = bio_iter_iovec(dst, dst_iter); | |
1004 | ||
1005 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1006 | |
1cb9dda4 KO |
1007 | src_p = kmap_atomic(src_bv.bv_page); |
1008 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1009 | |
1cb9dda4 KO |
1010 | memcpy(dst_p + dst_bv.bv_offset, |
1011 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1012 | bytes); |
1013 | ||
1014 | kunmap_atomic(dst_p); | |
1015 | kunmap_atomic(src_p); | |
1016 | ||
1cb9dda4 KO |
1017 | bio_advance_iter(src, &src_iter, bytes); |
1018 | bio_advance_iter(dst, &dst_iter, bytes); | |
16ac3d63 KO |
1019 | } |
1020 | } | |
1021 | EXPORT_SYMBOL(bio_copy_data); | |
1022 | ||
1da177e4 | 1023 | struct bio_map_data { |
152e283f | 1024 | int is_our_pages; |
26e49cfc KO |
1025 | struct iov_iter iter; |
1026 | struct iovec iov[]; | |
1da177e4 LT |
1027 | }; |
1028 | ||
7410b3c6 | 1029 | static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count, |
76029ff3 | 1030 | gfp_t gfp_mask) |
1da177e4 | 1031 | { |
f3f63c1c JA |
1032 | if (iov_count > UIO_MAXIOV) |
1033 | return NULL; | |
1da177e4 | 1034 | |
c8db4448 | 1035 | return kmalloc(sizeof(struct bio_map_data) + |
26e49cfc | 1036 | sizeof(struct iovec) * iov_count, gfp_mask); |
1da177e4 LT |
1037 | } |
1038 | ||
9124d3fe DP |
1039 | /** |
1040 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
1041 | * @bio: The &struct bio which describes the I/O as destination | |
1042 | * @iter: iov_iter as source | |
1043 | * | |
1044 | * Copy all pages from iov_iter to bio. | |
1045 | * Returns 0 on success, or error on failure. | |
1046 | */ | |
1047 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter) | |
c5dec1c3 | 1048 | { |
9124d3fe | 1049 | int i; |
c5dec1c3 | 1050 | struct bio_vec *bvec; |
c5dec1c3 | 1051 | |
d74c6d51 | 1052 | bio_for_each_segment_all(bvec, bio, i) { |
9124d3fe | 1053 | ssize_t ret; |
c5dec1c3 | 1054 | |
9124d3fe DP |
1055 | ret = copy_page_from_iter(bvec->bv_page, |
1056 | bvec->bv_offset, | |
1057 | bvec->bv_len, | |
1058 | &iter); | |
1059 | ||
1060 | if (!iov_iter_count(&iter)) | |
1061 | break; | |
1062 | ||
1063 | if (ret < bvec->bv_len) | |
1064 | return -EFAULT; | |
c5dec1c3 FT |
1065 | } |
1066 | ||
9124d3fe DP |
1067 | return 0; |
1068 | } | |
1069 | ||
1070 | /** | |
1071 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
1072 | * @bio: The &struct bio which describes the I/O as source | |
1073 | * @iter: iov_iter as destination | |
1074 | * | |
1075 | * Copy all pages from bio to iov_iter. | |
1076 | * Returns 0 on success, or error on failure. | |
1077 | */ | |
1078 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
1079 | { | |
1080 | int i; | |
1081 | struct bio_vec *bvec; | |
1082 | ||
1083 | bio_for_each_segment_all(bvec, bio, i) { | |
1084 | ssize_t ret; | |
1085 | ||
1086 | ret = copy_page_to_iter(bvec->bv_page, | |
1087 | bvec->bv_offset, | |
1088 | bvec->bv_len, | |
1089 | &iter); | |
1090 | ||
1091 | if (!iov_iter_count(&iter)) | |
1092 | break; | |
1093 | ||
1094 | if (ret < bvec->bv_len) | |
1095 | return -EFAULT; | |
1096 | } | |
1097 | ||
1098 | return 0; | |
c5dec1c3 FT |
1099 | } |
1100 | ||
1dfa0f68 CH |
1101 | static void bio_free_pages(struct bio *bio) |
1102 | { | |
1103 | struct bio_vec *bvec; | |
1104 | int i; | |
1105 | ||
1106 | bio_for_each_segment_all(bvec, bio, i) | |
1107 | __free_page(bvec->bv_page); | |
1108 | } | |
1109 | ||
1da177e4 LT |
1110 | /** |
1111 | * bio_uncopy_user - finish previously mapped bio | |
1112 | * @bio: bio being terminated | |
1113 | * | |
ddad8dd0 | 1114 | * Free pages allocated from bio_copy_user_iov() and write back data |
1da177e4 LT |
1115 | * to user space in case of a read. |
1116 | */ | |
1117 | int bio_uncopy_user(struct bio *bio) | |
1118 | { | |
1119 | struct bio_map_data *bmd = bio->bi_private; | |
1dfa0f68 | 1120 | int ret = 0; |
1da177e4 | 1121 | |
35dc2483 RD |
1122 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { |
1123 | /* | |
1124 | * if we're in a workqueue, the request is orphaned, so | |
1125 | * don't copy into a random user address space, just free. | |
1126 | */ | |
9124d3fe DP |
1127 | if (current->mm && bio_data_dir(bio) == READ) |
1128 | ret = bio_copy_to_iter(bio, bmd->iter); | |
1dfa0f68 CH |
1129 | if (bmd->is_our_pages) |
1130 | bio_free_pages(bio); | |
35dc2483 | 1131 | } |
c8db4448 | 1132 | kfree(bmd); |
1da177e4 LT |
1133 | bio_put(bio); |
1134 | return ret; | |
1135 | } | |
a112a71d | 1136 | EXPORT_SYMBOL(bio_uncopy_user); |
1da177e4 LT |
1137 | |
1138 | /** | |
c5dec1c3 | 1139 | * bio_copy_user_iov - copy user data to bio |
26e49cfc KO |
1140 | * @q: destination block queue |
1141 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1142 | * @iter: iovec iterator | |
1143 | * @gfp_mask: memory allocation flags | |
1da177e4 LT |
1144 | * |
1145 | * Prepares and returns a bio for indirect user io, bouncing data | |
1146 | * to/from kernel pages as necessary. Must be paired with | |
1147 | * call bio_uncopy_user() on io completion. | |
1148 | */ | |
152e283f FT |
1149 | struct bio *bio_copy_user_iov(struct request_queue *q, |
1150 | struct rq_map_data *map_data, | |
26e49cfc KO |
1151 | const struct iov_iter *iter, |
1152 | gfp_t gfp_mask) | |
1da177e4 | 1153 | { |
1da177e4 | 1154 | struct bio_map_data *bmd; |
1da177e4 LT |
1155 | struct page *page; |
1156 | struct bio *bio; | |
1157 | int i, ret; | |
c5dec1c3 | 1158 | int nr_pages = 0; |
26e49cfc | 1159 | unsigned int len = iter->count; |
56c451f4 | 1160 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; |
1da177e4 | 1161 | |
26e49cfc | 1162 | for (i = 0; i < iter->nr_segs; i++) { |
c5dec1c3 FT |
1163 | unsigned long uaddr; |
1164 | unsigned long end; | |
1165 | unsigned long start; | |
1166 | ||
26e49cfc KO |
1167 | uaddr = (unsigned long) iter->iov[i].iov_base; |
1168 | end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1) | |
1169 | >> PAGE_SHIFT; | |
c5dec1c3 FT |
1170 | start = uaddr >> PAGE_SHIFT; |
1171 | ||
cb4644ca JA |
1172 | /* |
1173 | * Overflow, abort | |
1174 | */ | |
1175 | if (end < start) | |
1176 | return ERR_PTR(-EINVAL); | |
1177 | ||
c5dec1c3 | 1178 | nr_pages += end - start; |
c5dec1c3 FT |
1179 | } |
1180 | ||
69838727 FT |
1181 | if (offset) |
1182 | nr_pages++; | |
1183 | ||
26e49cfc | 1184 | bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask); |
1da177e4 LT |
1185 | if (!bmd) |
1186 | return ERR_PTR(-ENOMEM); | |
1187 | ||
26e49cfc KO |
1188 | /* |
1189 | * We need to do a deep copy of the iov_iter including the iovecs. | |
1190 | * The caller provided iov might point to an on-stack or otherwise | |
1191 | * shortlived one. | |
1192 | */ | |
1193 | bmd->is_our_pages = map_data ? 0 : 1; | |
1194 | memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs); | |
1195 | iov_iter_init(&bmd->iter, iter->type, bmd->iov, | |
1196 | iter->nr_segs, iter->count); | |
1197 | ||
1da177e4 | 1198 | ret = -ENOMEM; |
a9e9dc24 | 1199 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1200 | if (!bio) |
1201 | goto out_bmd; | |
1202 | ||
26e49cfc | 1203 | if (iter->type & WRITE) |
7b6d91da | 1204 | bio->bi_rw |= REQ_WRITE; |
1da177e4 LT |
1205 | |
1206 | ret = 0; | |
56c451f4 FT |
1207 | |
1208 | if (map_data) { | |
e623ddb4 | 1209 | nr_pages = 1 << map_data->page_order; |
56c451f4 FT |
1210 | i = map_data->offset / PAGE_SIZE; |
1211 | } | |
1da177e4 | 1212 | while (len) { |
e623ddb4 | 1213 | unsigned int bytes = PAGE_SIZE; |
1da177e4 | 1214 | |
56c451f4 FT |
1215 | bytes -= offset; |
1216 | ||
1da177e4 LT |
1217 | if (bytes > len) |
1218 | bytes = len; | |
1219 | ||
152e283f | 1220 | if (map_data) { |
e623ddb4 | 1221 | if (i == map_data->nr_entries * nr_pages) { |
152e283f FT |
1222 | ret = -ENOMEM; |
1223 | break; | |
1224 | } | |
e623ddb4 FT |
1225 | |
1226 | page = map_data->pages[i / nr_pages]; | |
1227 | page += (i % nr_pages); | |
1228 | ||
1229 | i++; | |
1230 | } else { | |
152e283f | 1231 | page = alloc_page(q->bounce_gfp | gfp_mask); |
e623ddb4 FT |
1232 | if (!page) { |
1233 | ret = -ENOMEM; | |
1234 | break; | |
1235 | } | |
1da177e4 LT |
1236 | } |
1237 | ||
56c451f4 | 1238 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) |
1da177e4 | 1239 | break; |
1da177e4 LT |
1240 | |
1241 | len -= bytes; | |
56c451f4 | 1242 | offset = 0; |
1da177e4 LT |
1243 | } |
1244 | ||
1245 | if (ret) | |
1246 | goto cleanup; | |
1247 | ||
1248 | /* | |
1249 | * success | |
1250 | */ | |
26e49cfc | 1251 | if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || |
ecb554a8 | 1252 | (map_data && map_data->from_user)) { |
9124d3fe | 1253 | ret = bio_copy_from_iter(bio, *iter); |
c5dec1c3 FT |
1254 | if (ret) |
1255 | goto cleanup; | |
1da177e4 LT |
1256 | } |
1257 | ||
26e49cfc | 1258 | bio->bi_private = bmd; |
1da177e4 LT |
1259 | return bio; |
1260 | cleanup: | |
152e283f | 1261 | if (!map_data) |
1dfa0f68 | 1262 | bio_free_pages(bio); |
1da177e4 LT |
1263 | bio_put(bio); |
1264 | out_bmd: | |
c8db4448 | 1265 | kfree(bmd); |
1da177e4 LT |
1266 | return ERR_PTR(ret); |
1267 | } | |
1268 | ||
37f19e57 CH |
1269 | /** |
1270 | * bio_map_user_iov - map user iovec into bio | |
1271 | * @q: the struct request_queue for the bio | |
1272 | * @iter: iovec iterator | |
1273 | * @gfp_mask: memory allocation flags | |
1274 | * | |
1275 | * Map the user space address into a bio suitable for io to a block | |
1276 | * device. Returns an error pointer in case of error. | |
1277 | */ | |
1278 | struct bio *bio_map_user_iov(struct request_queue *q, | |
1279 | const struct iov_iter *iter, | |
1280 | gfp_t gfp_mask) | |
1da177e4 | 1281 | { |
26e49cfc | 1282 | int j; |
f1970baf | 1283 | int nr_pages = 0; |
1da177e4 LT |
1284 | struct page **pages; |
1285 | struct bio *bio; | |
f1970baf JB |
1286 | int cur_page = 0; |
1287 | int ret, offset; | |
26e49cfc KO |
1288 | struct iov_iter i; |
1289 | struct iovec iov; | |
1da177e4 | 1290 | |
26e49cfc KO |
1291 | iov_for_each(iov, i, *iter) { |
1292 | unsigned long uaddr = (unsigned long) iov.iov_base; | |
1293 | unsigned long len = iov.iov_len; | |
f1970baf JB |
1294 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1295 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1296 | ||
cb4644ca JA |
1297 | /* |
1298 | * Overflow, abort | |
1299 | */ | |
1300 | if (end < start) | |
1301 | return ERR_PTR(-EINVAL); | |
1302 | ||
f1970baf JB |
1303 | nr_pages += end - start; |
1304 | /* | |
ad2d7225 | 1305 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 1306 | */ |
ad2d7225 | 1307 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
1308 | return ERR_PTR(-EINVAL); |
1309 | } | |
1310 | ||
1311 | if (!nr_pages) | |
1da177e4 LT |
1312 | return ERR_PTR(-EINVAL); |
1313 | ||
a9e9dc24 | 1314 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1315 | if (!bio) |
1316 | return ERR_PTR(-ENOMEM); | |
1317 | ||
1318 | ret = -ENOMEM; | |
a3bce90e | 1319 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
1da177e4 LT |
1320 | if (!pages) |
1321 | goto out; | |
1322 | ||
26e49cfc KO |
1323 | iov_for_each(iov, i, *iter) { |
1324 | unsigned long uaddr = (unsigned long) iov.iov_base; | |
1325 | unsigned long len = iov.iov_len; | |
f1970baf JB |
1326 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1327 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1328 | const int local_nr_pages = end - start; | |
1329 | const int page_limit = cur_page + local_nr_pages; | |
cb4644ca | 1330 | |
f5dd33c4 | 1331 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
26e49cfc KO |
1332 | (iter->type & WRITE) != WRITE, |
1333 | &pages[cur_page]); | |
99172157 JA |
1334 | if (ret < local_nr_pages) { |
1335 | ret = -EFAULT; | |
f1970baf | 1336 | goto out_unmap; |
99172157 | 1337 | } |
f1970baf JB |
1338 | |
1339 | offset = uaddr & ~PAGE_MASK; | |
1340 | for (j = cur_page; j < page_limit; j++) { | |
1341 | unsigned int bytes = PAGE_SIZE - offset; | |
1342 | ||
1343 | if (len <= 0) | |
1344 | break; | |
1345 | ||
1346 | if (bytes > len) | |
1347 | bytes = len; | |
1348 | ||
1349 | /* | |
1350 | * sorry... | |
1351 | */ | |
defd94b7 MC |
1352 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
1353 | bytes) | |
f1970baf JB |
1354 | break; |
1355 | ||
1356 | len -= bytes; | |
1357 | offset = 0; | |
1358 | } | |
1da177e4 | 1359 | |
f1970baf | 1360 | cur_page = j; |
1da177e4 | 1361 | /* |
f1970baf | 1362 | * release the pages we didn't map into the bio, if any |
1da177e4 | 1363 | */ |
f1970baf JB |
1364 | while (j < page_limit) |
1365 | page_cache_release(pages[j++]); | |
1da177e4 LT |
1366 | } |
1367 | ||
1da177e4 LT |
1368 | kfree(pages); |
1369 | ||
1370 | /* | |
1371 | * set data direction, and check if mapped pages need bouncing | |
1372 | */ | |
26e49cfc | 1373 | if (iter->type & WRITE) |
7b6d91da | 1374 | bio->bi_rw |= REQ_WRITE; |
1da177e4 LT |
1375 | |
1376 | bio->bi_flags |= (1 << BIO_USER_MAPPED); | |
37f19e57 CH |
1377 | |
1378 | /* | |
1379 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
1380 | * it would normally disappear when its bi_end_io is run. | |
1381 | * however, we need it for the unmap, so grab an extra | |
1382 | * reference to it | |
1383 | */ | |
1384 | bio_get(bio); | |
1da177e4 | 1385 | return bio; |
f1970baf JB |
1386 | |
1387 | out_unmap: | |
26e49cfc KO |
1388 | for (j = 0; j < nr_pages; j++) { |
1389 | if (!pages[j]) | |
f1970baf | 1390 | break; |
26e49cfc | 1391 | page_cache_release(pages[j]); |
f1970baf JB |
1392 | } |
1393 | out: | |
1da177e4 LT |
1394 | kfree(pages); |
1395 | bio_put(bio); | |
1396 | return ERR_PTR(ret); | |
1397 | } | |
1398 | ||
1da177e4 LT |
1399 | static void __bio_unmap_user(struct bio *bio) |
1400 | { | |
1401 | struct bio_vec *bvec; | |
1402 | int i; | |
1403 | ||
1404 | /* | |
1405 | * make sure we dirty pages we wrote to | |
1406 | */ | |
d74c6d51 | 1407 | bio_for_each_segment_all(bvec, bio, i) { |
1da177e4 LT |
1408 | if (bio_data_dir(bio) == READ) |
1409 | set_page_dirty_lock(bvec->bv_page); | |
1410 | ||
1411 | page_cache_release(bvec->bv_page); | |
1412 | } | |
1413 | ||
1414 | bio_put(bio); | |
1415 | } | |
1416 | ||
1417 | /** | |
1418 | * bio_unmap_user - unmap a bio | |
1419 | * @bio: the bio being unmapped | |
1420 | * | |
1421 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
1422 | * a process context. | |
1423 | * | |
1424 | * bio_unmap_user() may sleep. | |
1425 | */ | |
1426 | void bio_unmap_user(struct bio *bio) | |
1427 | { | |
1428 | __bio_unmap_user(bio); | |
1429 | bio_put(bio); | |
1430 | } | |
a112a71d | 1431 | EXPORT_SYMBOL(bio_unmap_user); |
1da177e4 | 1432 | |
6712ecf8 | 1433 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 1434 | { |
b823825e | 1435 | bio_put(bio); |
b823825e JA |
1436 | } |
1437 | ||
75c72b83 CH |
1438 | /** |
1439 | * bio_map_kern - map kernel address into bio | |
1440 | * @q: the struct request_queue for the bio | |
1441 | * @data: pointer to buffer to map | |
1442 | * @len: length in bytes | |
1443 | * @gfp_mask: allocation flags for bio allocation | |
1444 | * | |
1445 | * Map the kernel address into a bio suitable for io to a block | |
1446 | * device. Returns an error pointer in case of error. | |
1447 | */ | |
1448 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1449 | gfp_t gfp_mask) | |
df46b9a4 MC |
1450 | { |
1451 | unsigned long kaddr = (unsigned long)data; | |
1452 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1453 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1454 | const int nr_pages = end - start; | |
1455 | int offset, i; | |
1456 | struct bio *bio; | |
1457 | ||
a9e9dc24 | 1458 | bio = bio_kmalloc(gfp_mask, nr_pages); |
df46b9a4 MC |
1459 | if (!bio) |
1460 | return ERR_PTR(-ENOMEM); | |
1461 | ||
1462 | offset = offset_in_page(kaddr); | |
1463 | for (i = 0; i < nr_pages; i++) { | |
1464 | unsigned int bytes = PAGE_SIZE - offset; | |
1465 | ||
1466 | if (len <= 0) | |
1467 | break; | |
1468 | ||
1469 | if (bytes > len) | |
1470 | bytes = len; | |
1471 | ||
defd94b7 | 1472 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
75c72b83 CH |
1473 | offset) < bytes) { |
1474 | /* we don't support partial mappings */ | |
1475 | bio_put(bio); | |
1476 | return ERR_PTR(-EINVAL); | |
1477 | } | |
df46b9a4 MC |
1478 | |
1479 | data += bytes; | |
1480 | len -= bytes; | |
1481 | offset = 0; | |
1482 | } | |
1483 | ||
b823825e | 1484 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1485 | return bio; |
1486 | } | |
a112a71d | 1487 | EXPORT_SYMBOL(bio_map_kern); |
df46b9a4 | 1488 | |
68154e90 FT |
1489 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1490 | { | |
1dfa0f68 CH |
1491 | bio_free_pages(bio); |
1492 | bio_put(bio); | |
1493 | } | |
1494 | ||
1495 | static void bio_copy_kern_endio_read(struct bio *bio, int err) | |
1496 | { | |
42d2683a | 1497 | char *p = bio->bi_private; |
1dfa0f68 | 1498 | struct bio_vec *bvec; |
68154e90 FT |
1499 | int i; |
1500 | ||
d74c6d51 | 1501 | bio_for_each_segment_all(bvec, bio, i) { |
1dfa0f68 | 1502 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); |
c8db4448 | 1503 | p += bvec->bv_len; |
68154e90 FT |
1504 | } |
1505 | ||
1dfa0f68 | 1506 | bio_copy_kern_endio(bio, err); |
68154e90 FT |
1507 | } |
1508 | ||
1509 | /** | |
1510 | * bio_copy_kern - copy kernel address into bio | |
1511 | * @q: the struct request_queue for the bio | |
1512 | * @data: pointer to buffer to copy | |
1513 | * @len: length in bytes | |
1514 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1515 | * @reading: data direction is READ |
68154e90 FT |
1516 | * |
1517 | * copy the kernel address into a bio suitable for io to a block | |
1518 | * device. Returns an error pointer in case of error. | |
1519 | */ | |
1520 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1521 | gfp_t gfp_mask, int reading) | |
1522 | { | |
42d2683a CH |
1523 | unsigned long kaddr = (unsigned long)data; |
1524 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1525 | unsigned long start = kaddr >> PAGE_SHIFT; | |
42d2683a CH |
1526 | struct bio *bio; |
1527 | void *p = data; | |
1dfa0f68 | 1528 | int nr_pages = 0; |
68154e90 | 1529 | |
42d2683a CH |
1530 | /* |
1531 | * Overflow, abort | |
1532 | */ | |
1533 | if (end < start) | |
1534 | return ERR_PTR(-EINVAL); | |
68154e90 | 1535 | |
42d2683a CH |
1536 | nr_pages = end - start; |
1537 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1538 | if (!bio) | |
1539 | return ERR_PTR(-ENOMEM); | |
68154e90 | 1540 | |
42d2683a CH |
1541 | while (len) { |
1542 | struct page *page; | |
1543 | unsigned int bytes = PAGE_SIZE; | |
68154e90 | 1544 | |
42d2683a CH |
1545 | if (bytes > len) |
1546 | bytes = len; | |
1547 | ||
1548 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1549 | if (!page) | |
1550 | goto cleanup; | |
1551 | ||
1552 | if (!reading) | |
1553 | memcpy(page_address(page), p, bytes); | |
1554 | ||
1555 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
1556 | break; | |
1557 | ||
1558 | len -= bytes; | |
1559 | p += bytes; | |
68154e90 FT |
1560 | } |
1561 | ||
1dfa0f68 CH |
1562 | if (reading) { |
1563 | bio->bi_end_io = bio_copy_kern_endio_read; | |
1564 | bio->bi_private = data; | |
1565 | } else { | |
1566 | bio->bi_end_io = bio_copy_kern_endio; | |
42d2683a | 1567 | bio->bi_rw |= REQ_WRITE; |
1dfa0f68 | 1568 | } |
76029ff3 | 1569 | |
68154e90 | 1570 | return bio; |
42d2683a CH |
1571 | |
1572 | cleanup: | |
1dfa0f68 | 1573 | bio_free_pages(bio); |
42d2683a CH |
1574 | bio_put(bio); |
1575 | return ERR_PTR(-ENOMEM); | |
68154e90 | 1576 | } |
a112a71d | 1577 | EXPORT_SYMBOL(bio_copy_kern); |
68154e90 | 1578 | |
1da177e4 LT |
1579 | /* |
1580 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1581 | * for performing direct-IO in BIOs. | |
1582 | * | |
1583 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1584 | * because the required locks are not interrupt-safe. So what we can do is to | |
1585 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1586 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1587 | * in process context. | |
1588 | * | |
1589 | * We special-case compound pages here: normally this means reads into hugetlb | |
1590 | * pages. The logic in here doesn't really work right for compound pages | |
1591 | * because the VM does not uniformly chase down the head page in all cases. | |
1592 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1593 | * handle them at all. So we skip compound pages here at an early stage. | |
1594 | * | |
1595 | * Note that this code is very hard to test under normal circumstances because | |
1596 | * direct-io pins the pages with get_user_pages(). This makes | |
1597 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1598 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1599 | * pagecache. |
1600 | * | |
1601 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1602 | * deferred bio dirtying paths. | |
1603 | */ | |
1604 | ||
1605 | /* | |
1606 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1607 | */ | |
1608 | void bio_set_pages_dirty(struct bio *bio) | |
1609 | { | |
cb34e057 | 1610 | struct bio_vec *bvec; |
1da177e4 LT |
1611 | int i; |
1612 | ||
cb34e057 KO |
1613 | bio_for_each_segment_all(bvec, bio, i) { |
1614 | struct page *page = bvec->bv_page; | |
1da177e4 LT |
1615 | |
1616 | if (page && !PageCompound(page)) | |
1617 | set_page_dirty_lock(page); | |
1618 | } | |
1619 | } | |
1620 | ||
86b6c7a7 | 1621 | static void bio_release_pages(struct bio *bio) |
1da177e4 | 1622 | { |
cb34e057 | 1623 | struct bio_vec *bvec; |
1da177e4 LT |
1624 | int i; |
1625 | ||
cb34e057 KO |
1626 | bio_for_each_segment_all(bvec, bio, i) { |
1627 | struct page *page = bvec->bv_page; | |
1da177e4 LT |
1628 | |
1629 | if (page) | |
1630 | put_page(page); | |
1631 | } | |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1636 | * If they are, then fine. If, however, some pages are clean then they must | |
1637 | * have been written out during the direct-IO read. So we take another ref on | |
1638 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1639 | * | |
1640 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1641 | * here on. It will run one page_cache_release() against each page and will | |
1642 | * run one bio_put() against the BIO. | |
1643 | */ | |
1644 | ||
65f27f38 | 1645 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1646 | |
65f27f38 | 1647 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1648 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1649 | static struct bio *bio_dirty_list; | |
1650 | ||
1651 | /* | |
1652 | * This runs in process context | |
1653 | */ | |
65f27f38 | 1654 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1655 | { |
1656 | unsigned long flags; | |
1657 | struct bio *bio; | |
1658 | ||
1659 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1660 | bio = bio_dirty_list; | |
1661 | bio_dirty_list = NULL; | |
1662 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1663 | ||
1664 | while (bio) { | |
1665 | struct bio *next = bio->bi_private; | |
1666 | ||
1667 | bio_set_pages_dirty(bio); | |
1668 | bio_release_pages(bio); | |
1669 | bio_put(bio); | |
1670 | bio = next; | |
1671 | } | |
1672 | } | |
1673 | ||
1674 | void bio_check_pages_dirty(struct bio *bio) | |
1675 | { | |
cb34e057 | 1676 | struct bio_vec *bvec; |
1da177e4 LT |
1677 | int nr_clean_pages = 0; |
1678 | int i; | |
1679 | ||
cb34e057 KO |
1680 | bio_for_each_segment_all(bvec, bio, i) { |
1681 | struct page *page = bvec->bv_page; | |
1da177e4 LT |
1682 | |
1683 | if (PageDirty(page) || PageCompound(page)) { | |
1684 | page_cache_release(page); | |
cb34e057 | 1685 | bvec->bv_page = NULL; |
1da177e4 LT |
1686 | } else { |
1687 | nr_clean_pages++; | |
1688 | } | |
1689 | } | |
1690 | ||
1691 | if (nr_clean_pages) { | |
1692 | unsigned long flags; | |
1693 | ||
1694 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1695 | bio->bi_private = bio_dirty_list; | |
1696 | bio_dirty_list = bio; | |
1697 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1698 | schedule_work(&bio_dirty_work); | |
1699 | } else { | |
1700 | bio_put(bio); | |
1701 | } | |
1702 | } | |
1703 | ||
394ffa50 GZ |
1704 | void generic_start_io_acct(int rw, unsigned long sectors, |
1705 | struct hd_struct *part) | |
1706 | { | |
1707 | int cpu = part_stat_lock(); | |
1708 | ||
1709 | part_round_stats(cpu, part); | |
1710 | part_stat_inc(cpu, part, ios[rw]); | |
1711 | part_stat_add(cpu, part, sectors[rw], sectors); | |
1712 | part_inc_in_flight(part, rw); | |
1713 | ||
1714 | part_stat_unlock(); | |
1715 | } | |
1716 | EXPORT_SYMBOL(generic_start_io_acct); | |
1717 | ||
1718 | void generic_end_io_acct(int rw, struct hd_struct *part, | |
1719 | unsigned long start_time) | |
1720 | { | |
1721 | unsigned long duration = jiffies - start_time; | |
1722 | int cpu = part_stat_lock(); | |
1723 | ||
1724 | part_stat_add(cpu, part, ticks[rw], duration); | |
1725 | part_round_stats(cpu, part); | |
1726 | part_dec_in_flight(part, rw); | |
1727 | ||
1728 | part_stat_unlock(); | |
1729 | } | |
1730 | EXPORT_SYMBOL(generic_end_io_acct); | |
1731 | ||
2d4dc890 IL |
1732 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1733 | void bio_flush_dcache_pages(struct bio *bi) | |
1734 | { | |
7988613b KO |
1735 | struct bio_vec bvec; |
1736 | struct bvec_iter iter; | |
2d4dc890 | 1737 | |
7988613b KO |
1738 | bio_for_each_segment(bvec, bi, iter) |
1739 | flush_dcache_page(bvec.bv_page); | |
2d4dc890 IL |
1740 | } |
1741 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1742 | #endif | |
1743 | ||
1da177e4 LT |
1744 | /** |
1745 | * bio_endio - end I/O on a bio | |
1746 | * @bio: bio | |
1da177e4 LT |
1747 | * @error: error, if any |
1748 | * | |
1749 | * Description: | |
6712ecf8 | 1750 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1751 | * preferred way to end I/O on a bio, it takes care of clearing |
1752 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1753 | * established -Exxxx (-EIO, for instance) error values in case | |
25985edc | 1754 | * something went wrong. No one should call bi_end_io() directly on a |
5bb23a68 N |
1755 | * bio unless they own it and thus know that it has an end_io |
1756 | * function. | |
1da177e4 | 1757 | **/ |
6712ecf8 | 1758 | void bio_endio(struct bio *bio, int error) |
1da177e4 | 1759 | { |
196d38bc KO |
1760 | while (bio) { |
1761 | BUG_ON(atomic_read(&bio->bi_remaining) <= 0); | |
1762 | ||
1763 | if (error) | |
1764 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
1765 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1766 | error = -EIO; | |
1767 | ||
1768 | if (!atomic_dec_and_test(&bio->bi_remaining)) | |
1769 | return; | |
1da177e4 | 1770 | |
196d38bc KO |
1771 | /* |
1772 | * Need to have a real endio function for chained bios, | |
1773 | * otherwise various corner cases will break (like stacking | |
1774 | * block devices that save/restore bi_end_io) - however, we want | |
1775 | * to avoid unbounded recursion and blowing the stack. Tail call | |
1776 | * optimization would handle this, but compiling with frame | |
1777 | * pointers also disables gcc's sibling call optimization. | |
1778 | */ | |
1779 | if (bio->bi_end_io == bio_chain_endio) { | |
1780 | struct bio *parent = bio->bi_private; | |
1781 | bio_put(bio); | |
1782 | bio = parent; | |
1783 | } else { | |
1784 | if (bio->bi_end_io) | |
1785 | bio->bi_end_io(bio, error); | |
1786 | bio = NULL; | |
1787 | } | |
1788 | } | |
1da177e4 | 1789 | } |
a112a71d | 1790 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1791 | |
196d38bc KO |
1792 | /** |
1793 | * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining | |
1794 | * @bio: bio | |
1795 | * @error: error, if any | |
1796 | * | |
1797 | * For code that has saved and restored bi_end_io; thing hard before using this | |
1798 | * function, probably you should've cloned the entire bio. | |
1799 | **/ | |
1800 | void bio_endio_nodec(struct bio *bio, int error) | |
1801 | { | |
1802 | atomic_inc(&bio->bi_remaining); | |
1803 | bio_endio(bio, error); | |
1804 | } | |
1805 | EXPORT_SYMBOL(bio_endio_nodec); | |
1806 | ||
20d0189b KO |
1807 | /** |
1808 | * bio_split - split a bio | |
1809 | * @bio: bio to split | |
1810 | * @sectors: number of sectors to split from the front of @bio | |
1811 | * @gfp: gfp mask | |
1812 | * @bs: bio set to allocate from | |
1813 | * | |
1814 | * Allocates and returns a new bio which represents @sectors from the start of | |
1815 | * @bio, and updates @bio to represent the remaining sectors. | |
1816 | * | |
1817 | * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's | |
1818 | * responsibility to ensure that @bio is not freed before the split. | |
1819 | */ | |
1820 | struct bio *bio_split(struct bio *bio, int sectors, | |
1821 | gfp_t gfp, struct bio_set *bs) | |
1822 | { | |
1823 | struct bio *split = NULL; | |
1824 | ||
1825 | BUG_ON(sectors <= 0); | |
1826 | BUG_ON(sectors >= bio_sectors(bio)); | |
1827 | ||
1828 | split = bio_clone_fast(bio, gfp, bs); | |
1829 | if (!split) | |
1830 | return NULL; | |
1831 | ||
1832 | split->bi_iter.bi_size = sectors << 9; | |
1833 | ||
1834 | if (bio_integrity(split)) | |
1835 | bio_integrity_trim(split, 0, sectors); | |
1836 | ||
1837 | bio_advance(bio, split->bi_iter.bi_size); | |
1838 | ||
1839 | return split; | |
1840 | } | |
1841 | EXPORT_SYMBOL(bio_split); | |
1842 | ||
6678d83f KO |
1843 | /** |
1844 | * bio_trim - trim a bio | |
1845 | * @bio: bio to trim | |
1846 | * @offset: number of sectors to trim from the front of @bio | |
1847 | * @size: size we want to trim @bio to, in sectors | |
1848 | */ | |
1849 | void bio_trim(struct bio *bio, int offset, int size) | |
1850 | { | |
1851 | /* 'bio' is a cloned bio which we need to trim to match | |
1852 | * the given offset and size. | |
6678d83f | 1853 | */ |
6678d83f KO |
1854 | |
1855 | size <<= 9; | |
4f024f37 | 1856 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1857 | return; |
1858 | ||
1859 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
1860 | ||
1861 | bio_advance(bio, offset << 9); | |
1862 | ||
4f024f37 | 1863 | bio->bi_iter.bi_size = size; |
6678d83f KO |
1864 | } |
1865 | EXPORT_SYMBOL_GPL(bio_trim); | |
1866 | ||
1da177e4 LT |
1867 | /* |
1868 | * create memory pools for biovec's in a bio_set. | |
1869 | * use the global biovec slabs created for general use. | |
1870 | */ | |
a6c39cb4 | 1871 | mempool_t *biovec_create_pool(int pool_entries) |
1da177e4 | 1872 | { |
7ff9345f | 1873 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
1da177e4 | 1874 | |
9f060e22 | 1875 | return mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1876 | } |
1877 | ||
1878 | void bioset_free(struct bio_set *bs) | |
1879 | { | |
df2cb6da KO |
1880 | if (bs->rescue_workqueue) |
1881 | destroy_workqueue(bs->rescue_workqueue); | |
1882 | ||
1da177e4 LT |
1883 | if (bs->bio_pool) |
1884 | mempool_destroy(bs->bio_pool); | |
1885 | ||
9f060e22 KO |
1886 | if (bs->bvec_pool) |
1887 | mempool_destroy(bs->bvec_pool); | |
1888 | ||
7878cba9 | 1889 | bioset_integrity_free(bs); |
bb799ca0 | 1890 | bio_put_slab(bs); |
1da177e4 LT |
1891 | |
1892 | kfree(bs); | |
1893 | } | |
a112a71d | 1894 | EXPORT_SYMBOL(bioset_free); |
1da177e4 | 1895 | |
d8f429e1 JN |
1896 | static struct bio_set *__bioset_create(unsigned int pool_size, |
1897 | unsigned int front_pad, | |
1898 | bool create_bvec_pool) | |
1da177e4 | 1899 | { |
392ddc32 | 1900 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
1b434498 | 1901 | struct bio_set *bs; |
1da177e4 | 1902 | |
1b434498 | 1903 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1904 | if (!bs) |
1905 | return NULL; | |
1906 | ||
bb799ca0 | 1907 | bs->front_pad = front_pad; |
1b434498 | 1908 | |
df2cb6da KO |
1909 | spin_lock_init(&bs->rescue_lock); |
1910 | bio_list_init(&bs->rescue_list); | |
1911 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1912 | ||
392ddc32 | 1913 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
bb799ca0 JA |
1914 | if (!bs->bio_slab) { |
1915 | kfree(bs); | |
1916 | return NULL; | |
1917 | } | |
1918 | ||
1919 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1da177e4 LT |
1920 | if (!bs->bio_pool) |
1921 | goto bad; | |
1922 | ||
d8f429e1 JN |
1923 | if (create_bvec_pool) { |
1924 | bs->bvec_pool = biovec_create_pool(pool_size); | |
1925 | if (!bs->bvec_pool) | |
1926 | goto bad; | |
1927 | } | |
df2cb6da KO |
1928 | |
1929 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1930 | if (!bs->rescue_workqueue) | |
1931 | goto bad; | |
1da177e4 | 1932 | |
df2cb6da | 1933 | return bs; |
1da177e4 LT |
1934 | bad: |
1935 | bioset_free(bs); | |
1936 | return NULL; | |
1937 | } | |
d8f429e1 JN |
1938 | |
1939 | /** | |
1940 | * bioset_create - Create a bio_set | |
1941 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1942 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1943 | * | |
1944 | * Description: | |
1945 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1946 | * to ask for a number of bytes to be allocated in front of the bio. | |
1947 | * Front pad allocation is useful for embedding the bio inside | |
1948 | * another structure, to avoid allocating extra data to go with the bio. | |
1949 | * Note that the bio must be embedded at the END of that structure always, | |
1950 | * or things will break badly. | |
1951 | */ | |
1952 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) | |
1953 | { | |
1954 | return __bioset_create(pool_size, front_pad, true); | |
1955 | } | |
a112a71d | 1956 | EXPORT_SYMBOL(bioset_create); |
1da177e4 | 1957 | |
d8f429e1 JN |
1958 | /** |
1959 | * bioset_create_nobvec - Create a bio_set without bio_vec mempool | |
1960 | * @pool_size: Number of bio to cache in the mempool | |
1961 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1962 | * | |
1963 | * Description: | |
1964 | * Same functionality as bioset_create() except that mempool is not | |
1965 | * created for bio_vecs. Saving some memory for bio_clone_fast() users. | |
1966 | */ | |
1967 | struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad) | |
1968 | { | |
1969 | return __bioset_create(pool_size, front_pad, false); | |
1970 | } | |
1971 | EXPORT_SYMBOL(bioset_create_nobvec); | |
1972 | ||
852c788f TH |
1973 | #ifdef CONFIG_BLK_CGROUP |
1974 | /** | |
1975 | * bio_associate_current - associate a bio with %current | |
1976 | * @bio: target bio | |
1977 | * | |
1978 | * Associate @bio with %current if it hasn't been associated yet. Block | |
1979 | * layer will treat @bio as if it were issued by %current no matter which | |
1980 | * task actually issues it. | |
1981 | * | |
1982 | * This function takes an extra reference of @task's io_context and blkcg | |
1983 | * which will be put when @bio is released. The caller must own @bio, | |
1984 | * ensure %current->io_context exists, and is responsible for synchronizing | |
1985 | * calls to this function. | |
1986 | */ | |
1987 | int bio_associate_current(struct bio *bio) | |
1988 | { | |
1989 | struct io_context *ioc; | |
1990 | struct cgroup_subsys_state *css; | |
1991 | ||
1992 | if (bio->bi_ioc) | |
1993 | return -EBUSY; | |
1994 | ||
1995 | ioc = current->io_context; | |
1996 | if (!ioc) | |
1997 | return -ENOENT; | |
1998 | ||
1999 | /* acquire active ref on @ioc and associate */ | |
2000 | get_io_context_active(ioc); | |
2001 | bio->bi_ioc = ioc; | |
2002 | ||
2003 | /* associate blkcg if exists */ | |
2004 | rcu_read_lock(); | |
073219e9 | 2005 | css = task_css(current, blkio_cgrp_id); |
ec903c0c | 2006 | if (css && css_tryget_online(css)) |
852c788f TH |
2007 | bio->bi_css = css; |
2008 | rcu_read_unlock(); | |
2009 | ||
2010 | return 0; | |
2011 | } | |
2012 | ||
2013 | /** | |
2014 | * bio_disassociate_task - undo bio_associate_current() | |
2015 | * @bio: target bio | |
2016 | */ | |
2017 | void bio_disassociate_task(struct bio *bio) | |
2018 | { | |
2019 | if (bio->bi_ioc) { | |
2020 | put_io_context(bio->bi_ioc); | |
2021 | bio->bi_ioc = NULL; | |
2022 | } | |
2023 | if (bio->bi_css) { | |
2024 | css_put(bio->bi_css); | |
2025 | bio->bi_css = NULL; | |
2026 | } | |
2027 | } | |
2028 | ||
2029 | #endif /* CONFIG_BLK_CGROUP */ | |
2030 | ||
1da177e4 LT |
2031 | static void __init biovec_init_slabs(void) |
2032 | { | |
2033 | int i; | |
2034 | ||
2035 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
2036 | int size; | |
2037 | struct biovec_slab *bvs = bvec_slabs + i; | |
2038 | ||
a7fcd37c JA |
2039 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
2040 | bvs->slab = NULL; | |
2041 | continue; | |
2042 | } | |
a7fcd37c | 2043 | |
1da177e4 LT |
2044 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
2045 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 2046 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
2047 | } |
2048 | } | |
2049 | ||
2050 | static int __init init_bio(void) | |
2051 | { | |
bb799ca0 JA |
2052 | bio_slab_max = 2; |
2053 | bio_slab_nr = 0; | |
2054 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
2055 | if (!bio_slabs) | |
2056 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 2057 | |
7878cba9 | 2058 | bio_integrity_init(); |
1da177e4 LT |
2059 | biovec_init_slabs(); |
2060 | ||
bb799ca0 | 2061 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); |
1da177e4 LT |
2062 | if (!fs_bio_set) |
2063 | panic("bio: can't allocate bios\n"); | |
2064 | ||
a91a2785 MP |
2065 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) |
2066 | panic("bio: can't create integrity pool\n"); | |
2067 | ||
1da177e4 LT |
2068 | return 0; |
2069 | } | |
1da177e4 | 2070 | subsys_initcall(init_bio); |