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> | |
852c788f | 22 | #include <linux/iocontext.h> |
1da177e4 LT |
23 | #include <linux/slab.h> |
24 | #include <linux/init.h> | |
25 | #include <linux/kernel.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/mempool.h> |
28 | #include <linux/workqueue.h> | |
852c788f | 29 | #include <linux/cgroup.h> |
f1970baf | 30 | #include <scsi/sg.h> /* for struct sg_iovec */ |
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 | ||
6feef531 | 40 | static mempool_t *bio_split_pool __read_mostly; |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* |
43 | * if you change this list, also change bvec_alloc or things will | |
44 | * break badly! cannot be bigger than what you can fit into an | |
45 | * unsigned short | |
46 | */ | |
1da177e4 | 47 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
df677140 | 48 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
1da177e4 LT |
49 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
50 | }; | |
51 | #undef BV | |
52 | ||
1da177e4 LT |
53 | /* |
54 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
55 | * IO code that does not need private memory pools. | |
56 | */ | |
51d654e1 | 57 | struct bio_set *fs_bio_set; |
3f86a82a | 58 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 59 | |
bb799ca0 JA |
60 | /* |
61 | * Our slab pool management | |
62 | */ | |
63 | struct bio_slab { | |
64 | struct kmem_cache *slab; | |
65 | unsigned int slab_ref; | |
66 | unsigned int slab_size; | |
67 | char name[8]; | |
68 | }; | |
69 | static DEFINE_MUTEX(bio_slab_lock); | |
70 | static struct bio_slab *bio_slabs; | |
71 | static unsigned int bio_slab_nr, bio_slab_max; | |
72 | ||
73 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
74 | { | |
75 | unsigned int sz = sizeof(struct bio) + extra_size; | |
76 | struct kmem_cache *slab = NULL; | |
389d7b26 | 77 | struct bio_slab *bslab, *new_bio_slabs; |
386bc35a | 78 | unsigned int new_bio_slab_max; |
bb799ca0 JA |
79 | unsigned int i, entry = -1; |
80 | ||
81 | mutex_lock(&bio_slab_lock); | |
82 | ||
83 | i = 0; | |
84 | while (i < bio_slab_nr) { | |
f06f135d | 85 | bslab = &bio_slabs[i]; |
bb799ca0 JA |
86 | |
87 | if (!bslab->slab && entry == -1) | |
88 | entry = i; | |
89 | else if (bslab->slab_size == sz) { | |
90 | slab = bslab->slab; | |
91 | bslab->slab_ref++; | |
92 | break; | |
93 | } | |
94 | i++; | |
95 | } | |
96 | ||
97 | if (slab) | |
98 | goto out_unlock; | |
99 | ||
100 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
386bc35a | 101 | new_bio_slab_max = bio_slab_max << 1; |
389d7b26 | 102 | new_bio_slabs = krealloc(bio_slabs, |
386bc35a | 103 | new_bio_slab_max * sizeof(struct bio_slab), |
389d7b26 AK |
104 | GFP_KERNEL); |
105 | if (!new_bio_slabs) | |
bb799ca0 | 106 | goto out_unlock; |
386bc35a | 107 | bio_slab_max = new_bio_slab_max; |
389d7b26 | 108 | bio_slabs = new_bio_slabs; |
bb799ca0 JA |
109 | } |
110 | if (entry == -1) | |
111 | entry = bio_slab_nr++; | |
112 | ||
113 | bslab = &bio_slabs[entry]; | |
114 | ||
115 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
116 | slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); | |
117 | if (!slab) | |
118 | goto out_unlock; | |
119 | ||
80cdc6da | 120 | printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry); |
bb799ca0 JA |
121 | bslab->slab = slab; |
122 | bslab->slab_ref = 1; | |
123 | bslab->slab_size = sz; | |
124 | out_unlock: | |
125 | mutex_unlock(&bio_slab_lock); | |
126 | return slab; | |
127 | } | |
128 | ||
129 | static void bio_put_slab(struct bio_set *bs) | |
130 | { | |
131 | struct bio_slab *bslab = NULL; | |
132 | unsigned int i; | |
133 | ||
134 | mutex_lock(&bio_slab_lock); | |
135 | ||
136 | for (i = 0; i < bio_slab_nr; i++) { | |
137 | if (bs->bio_slab == bio_slabs[i].slab) { | |
138 | bslab = &bio_slabs[i]; | |
139 | break; | |
140 | } | |
141 | } | |
142 | ||
143 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
144 | goto out; | |
145 | ||
146 | WARN_ON(!bslab->slab_ref); | |
147 | ||
148 | if (--bslab->slab_ref) | |
149 | goto out; | |
150 | ||
151 | kmem_cache_destroy(bslab->slab); | |
152 | bslab->slab = NULL; | |
153 | ||
154 | out: | |
155 | mutex_unlock(&bio_slab_lock); | |
156 | } | |
157 | ||
7ba1ba12 MP |
158 | unsigned int bvec_nr_vecs(unsigned short idx) |
159 | { | |
160 | return bvec_slabs[idx].nr_vecs; | |
161 | } | |
162 | ||
9f060e22 | 163 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 JA |
164 | { |
165 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); | |
166 | ||
167 | if (idx == BIOVEC_MAX_IDX) | |
9f060e22 | 168 | mempool_free(bv, pool); |
bb799ca0 JA |
169 | else { |
170 | struct biovec_slab *bvs = bvec_slabs + idx; | |
171 | ||
172 | kmem_cache_free(bvs->slab, bv); | |
173 | } | |
174 | } | |
175 | ||
9f060e22 KO |
176 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
177 | mempool_t *pool) | |
1da177e4 LT |
178 | { |
179 | struct bio_vec *bvl; | |
1da177e4 | 180 | |
7ff9345f JA |
181 | /* |
182 | * see comment near bvec_array define! | |
183 | */ | |
184 | switch (nr) { | |
185 | case 1: | |
186 | *idx = 0; | |
187 | break; | |
188 | case 2 ... 4: | |
189 | *idx = 1; | |
190 | break; | |
191 | case 5 ... 16: | |
192 | *idx = 2; | |
193 | break; | |
194 | case 17 ... 64: | |
195 | *idx = 3; | |
196 | break; | |
197 | case 65 ... 128: | |
198 | *idx = 4; | |
199 | break; | |
200 | case 129 ... BIO_MAX_PAGES: | |
201 | *idx = 5; | |
202 | break; | |
203 | default: | |
204 | return NULL; | |
205 | } | |
206 | ||
207 | /* | |
208 | * idx now points to the pool we want to allocate from. only the | |
209 | * 1-vec entry pool is mempool backed. | |
210 | */ | |
211 | if (*idx == BIOVEC_MAX_IDX) { | |
212 | fallback: | |
9f060e22 | 213 | bvl = mempool_alloc(pool, gfp_mask); |
7ff9345f JA |
214 | } else { |
215 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
216 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); | |
217 | ||
0a0d96b0 | 218 | /* |
7ff9345f JA |
219 | * Make this allocation restricted and don't dump info on |
220 | * allocation failures, since we'll fallback to the mempool | |
221 | * in case of failure. | |
0a0d96b0 | 222 | */ |
7ff9345f | 223 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 224 | |
0a0d96b0 | 225 | /* |
7ff9345f JA |
226 | * Try a slab allocation. If this fails and __GFP_WAIT |
227 | * is set, retry with the 1-entry mempool | |
0a0d96b0 | 228 | */ |
7ff9345f JA |
229 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
230 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { | |
231 | *idx = BIOVEC_MAX_IDX; | |
232 | goto fallback; | |
233 | } | |
234 | } | |
235 | ||
1da177e4 LT |
236 | return bvl; |
237 | } | |
238 | ||
4254bba1 | 239 | static void __bio_free(struct bio *bio) |
1da177e4 | 240 | { |
4254bba1 | 241 | bio_disassociate_task(bio); |
1da177e4 | 242 | |
7ba1ba12 | 243 | if (bio_integrity(bio)) |
1e2a410f | 244 | bio_integrity_free(bio); |
4254bba1 | 245 | } |
7ba1ba12 | 246 | |
4254bba1 KO |
247 | static void bio_free(struct bio *bio) |
248 | { | |
249 | struct bio_set *bs = bio->bi_pool; | |
250 | void *p; | |
251 | ||
252 | __bio_free(bio); | |
253 | ||
254 | if (bs) { | |
255 | if (bio_has_allocated_vec(bio)) | |
9f060e22 | 256 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio)); |
4254bba1 KO |
257 | |
258 | /* | |
259 | * If we have front padding, adjust the bio pointer before freeing | |
260 | */ | |
261 | p = bio; | |
bb799ca0 JA |
262 | p -= bs->front_pad; |
263 | ||
4254bba1 KO |
264 | mempool_free(p, bs->bio_pool); |
265 | } else { | |
266 | /* Bio was allocated by bio_kmalloc() */ | |
267 | kfree(bio); | |
268 | } | |
3676347a PO |
269 | } |
270 | ||
858119e1 | 271 | void bio_init(struct bio *bio) |
1da177e4 | 272 | { |
2b94de55 | 273 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 274 | bio->bi_flags = 1 << BIO_UPTODATE; |
1da177e4 | 275 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 | 276 | } |
a112a71d | 277 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 278 | |
f44b48c7 KO |
279 | /** |
280 | * bio_reset - reinitialize a bio | |
281 | * @bio: bio to reset | |
282 | * | |
283 | * Description: | |
284 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
285 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
286 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
287 | * comment in struct bio. | |
288 | */ | |
289 | void bio_reset(struct bio *bio) | |
290 | { | |
291 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
292 | ||
4254bba1 | 293 | __bio_free(bio); |
f44b48c7 KO |
294 | |
295 | memset(bio, 0, BIO_RESET_BYTES); | |
296 | bio->bi_flags = flags|(1 << BIO_UPTODATE); | |
297 | } | |
298 | EXPORT_SYMBOL(bio_reset); | |
299 | ||
df2cb6da KO |
300 | static void bio_alloc_rescue(struct work_struct *work) |
301 | { | |
302 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
303 | struct bio *bio; | |
304 | ||
305 | while (1) { | |
306 | spin_lock(&bs->rescue_lock); | |
307 | bio = bio_list_pop(&bs->rescue_list); | |
308 | spin_unlock(&bs->rescue_lock); | |
309 | ||
310 | if (!bio) | |
311 | break; | |
312 | ||
313 | generic_make_request(bio); | |
314 | } | |
315 | } | |
316 | ||
317 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
318 | { | |
319 | struct bio_list punt, nopunt; | |
320 | struct bio *bio; | |
321 | ||
322 | /* | |
323 | * In order to guarantee forward progress we must punt only bios that | |
324 | * were allocated from this bio_set; otherwise, if there was a bio on | |
325 | * there for a stacking driver higher up in the stack, processing it | |
326 | * could require allocating bios from this bio_set, and doing that from | |
327 | * our own rescuer would be bad. | |
328 | * | |
329 | * Since bio lists are singly linked, pop them all instead of trying to | |
330 | * remove from the middle of the list: | |
331 | */ | |
332 | ||
333 | bio_list_init(&punt); | |
334 | bio_list_init(&nopunt); | |
335 | ||
336 | while ((bio = bio_list_pop(current->bio_list))) | |
337 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
338 | ||
339 | *current->bio_list = nopunt; | |
340 | ||
341 | spin_lock(&bs->rescue_lock); | |
342 | bio_list_merge(&bs->rescue_list, &punt); | |
343 | spin_unlock(&bs->rescue_lock); | |
344 | ||
345 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
346 | } | |
347 | ||
1da177e4 LT |
348 | /** |
349 | * bio_alloc_bioset - allocate a bio for I/O | |
350 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
351 | * @nr_iovecs: number of iovecs to pre-allocate | |
db18efac | 352 | * @bs: the bio_set to allocate from. |
1da177e4 LT |
353 | * |
354 | * Description: | |
3f86a82a KO |
355 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
356 | * backed by the @bs's mempool. | |
357 | * | |
358 | * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be | |
359 | * able to allocate a bio. This is due to the mempool guarantees. To make this | |
360 | * work, callers must never allocate more than 1 bio at a time from this pool. | |
361 | * Callers that need to allocate more than 1 bio must always submit the | |
362 | * previously allocated bio for IO before attempting to allocate a new one. | |
363 | * Failure to do so can cause deadlocks under memory pressure. | |
364 | * | |
df2cb6da KO |
365 | * Note that when running under generic_make_request() (i.e. any block |
366 | * driver), bios are not submitted until after you return - see the code in | |
367 | * generic_make_request() that converts recursion into iteration, to prevent | |
368 | * stack overflows. | |
369 | * | |
370 | * This would normally mean allocating multiple bios under | |
371 | * generic_make_request() would be susceptible to deadlocks, but we have | |
372 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
373 | * thread. | |
374 | * | |
375 | * However, we do not guarantee forward progress for allocations from other | |
376 | * mempools. Doing multiple allocations from the same mempool under | |
377 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
378 | * for per bio allocations. | |
379 | * | |
3f86a82a KO |
380 | * RETURNS: |
381 | * Pointer to new bio on success, NULL on failure. | |
382 | */ | |
dd0fc66f | 383 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 | 384 | { |
df2cb6da | 385 | gfp_t saved_gfp = gfp_mask; |
3f86a82a KO |
386 | unsigned front_pad; |
387 | unsigned inline_vecs; | |
451a9ebf | 388 | unsigned long idx = BIO_POOL_NONE; |
34053979 | 389 | struct bio_vec *bvl = NULL; |
451a9ebf TH |
390 | struct bio *bio; |
391 | void *p; | |
392 | ||
3f86a82a KO |
393 | if (!bs) { |
394 | if (nr_iovecs > UIO_MAXIOV) | |
395 | return NULL; | |
396 | ||
397 | p = kmalloc(sizeof(struct bio) + | |
398 | nr_iovecs * sizeof(struct bio_vec), | |
399 | gfp_mask); | |
400 | front_pad = 0; | |
401 | inline_vecs = nr_iovecs; | |
402 | } else { | |
df2cb6da KO |
403 | /* |
404 | * generic_make_request() converts recursion to iteration; this | |
405 | * means if we're running beneath it, any bios we allocate and | |
406 | * submit will not be submitted (and thus freed) until after we | |
407 | * return. | |
408 | * | |
409 | * This exposes us to a potential deadlock if we allocate | |
410 | * multiple bios from the same bio_set() while running | |
411 | * underneath generic_make_request(). If we were to allocate | |
412 | * multiple bios (say a stacking block driver that was splitting | |
413 | * bios), we would deadlock if we exhausted the mempool's | |
414 | * reserve. | |
415 | * | |
416 | * We solve this, and guarantee forward progress, with a rescuer | |
417 | * workqueue per bio_set. If we go to allocate and there are | |
418 | * bios on current->bio_list, we first try the allocation | |
419 | * without __GFP_WAIT; if that fails, we punt those bios we | |
420 | * would be blocking to the rescuer workqueue before we retry | |
421 | * with the original gfp_flags. | |
422 | */ | |
423 | ||
424 | if (current->bio_list && !bio_list_empty(current->bio_list)) | |
425 | gfp_mask &= ~__GFP_WAIT; | |
426 | ||
3f86a82a | 427 | p = mempool_alloc(bs->bio_pool, gfp_mask); |
df2cb6da KO |
428 | if (!p && gfp_mask != saved_gfp) { |
429 | punt_bios_to_rescuer(bs); | |
430 | gfp_mask = saved_gfp; | |
431 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
432 | } | |
433 | ||
3f86a82a KO |
434 | front_pad = bs->front_pad; |
435 | inline_vecs = BIO_INLINE_VECS; | |
436 | } | |
437 | ||
451a9ebf TH |
438 | if (unlikely(!p)) |
439 | return NULL; | |
1da177e4 | 440 | |
3f86a82a | 441 | bio = p + front_pad; |
34053979 IM |
442 | bio_init(bio); |
443 | ||
3f86a82a | 444 | if (nr_iovecs > inline_vecs) { |
9f060e22 | 445 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); |
df2cb6da KO |
446 | if (!bvl && gfp_mask != saved_gfp) { |
447 | punt_bios_to_rescuer(bs); | |
448 | gfp_mask = saved_gfp; | |
9f060e22 | 449 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); |
df2cb6da KO |
450 | } |
451 | ||
34053979 IM |
452 | if (unlikely(!bvl)) |
453 | goto err_free; | |
3f86a82a KO |
454 | } else if (nr_iovecs) { |
455 | bvl = bio->bi_inline_vecs; | |
1da177e4 | 456 | } |
3f86a82a KO |
457 | |
458 | bio->bi_pool = bs; | |
34053979 IM |
459 | bio->bi_flags |= idx << BIO_POOL_OFFSET; |
460 | bio->bi_max_vecs = nr_iovecs; | |
34053979 | 461 | bio->bi_io_vec = bvl; |
1da177e4 | 462 | return bio; |
34053979 IM |
463 | |
464 | err_free: | |
451a9ebf | 465 | mempool_free(p, bs->bio_pool); |
34053979 | 466 | return NULL; |
1da177e4 | 467 | } |
a112a71d | 468 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 469 | |
1da177e4 LT |
470 | void zero_fill_bio(struct bio *bio) |
471 | { | |
472 | unsigned long flags; | |
473 | struct bio_vec *bv; | |
474 | int i; | |
475 | ||
476 | bio_for_each_segment(bv, bio, i) { | |
477 | char *data = bvec_kmap_irq(bv, &flags); | |
478 | memset(data, 0, bv->bv_len); | |
479 | flush_dcache_page(bv->bv_page); | |
480 | bvec_kunmap_irq(data, &flags); | |
481 | } | |
482 | } | |
483 | EXPORT_SYMBOL(zero_fill_bio); | |
484 | ||
485 | /** | |
486 | * bio_put - release a reference to a bio | |
487 | * @bio: bio to release reference to | |
488 | * | |
489 | * Description: | |
490 | * Put a reference to a &struct bio, either one you have gotten with | |
ad0bf110 | 491 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. |
1da177e4 LT |
492 | **/ |
493 | void bio_put(struct bio *bio) | |
494 | { | |
495 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
496 | ||
497 | /* | |
498 | * last put frees it | |
499 | */ | |
4254bba1 KO |
500 | if (atomic_dec_and_test(&bio->bi_cnt)) |
501 | bio_free(bio); | |
1da177e4 | 502 | } |
a112a71d | 503 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 504 | |
165125e1 | 505 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
506 | { |
507 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
508 | blk_recount_segments(q, bio); | |
509 | ||
510 | return bio->bi_phys_segments; | |
511 | } | |
a112a71d | 512 | EXPORT_SYMBOL(bio_phys_segments); |
1da177e4 | 513 | |
1da177e4 LT |
514 | /** |
515 | * __bio_clone - clone a bio | |
516 | * @bio: destination bio | |
517 | * @bio_src: bio to clone | |
518 | * | |
519 | * Clone a &bio. Caller will own the returned bio, but not | |
520 | * the actual data it points to. Reference count of returned | |
521 | * bio will be one. | |
522 | */ | |
858119e1 | 523 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
1da177e4 | 524 | { |
e525e153 AM |
525 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
526 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
1da177e4 | 527 | |
5d84070e JA |
528 | /* |
529 | * most users will be overriding ->bi_bdev with a new target, | |
530 | * so we don't set nor calculate new physical/hw segment counts here | |
531 | */ | |
1da177e4 LT |
532 | bio->bi_sector = bio_src->bi_sector; |
533 | bio->bi_bdev = bio_src->bi_bdev; | |
534 | bio->bi_flags |= 1 << BIO_CLONED; | |
535 | bio->bi_rw = bio_src->bi_rw; | |
1da177e4 LT |
536 | bio->bi_vcnt = bio_src->bi_vcnt; |
537 | bio->bi_size = bio_src->bi_size; | |
a5453be4 | 538 | bio->bi_idx = bio_src->bi_idx; |
1da177e4 | 539 | } |
a112a71d | 540 | EXPORT_SYMBOL(__bio_clone); |
1da177e4 LT |
541 | |
542 | /** | |
bf800ef1 | 543 | * bio_clone_bioset - clone a bio |
1da177e4 LT |
544 | * @bio: bio to clone |
545 | * @gfp_mask: allocation priority | |
bf800ef1 | 546 | * @bs: bio_set to allocate from |
1da177e4 LT |
547 | * |
548 | * Like __bio_clone, only also allocates the returned bio | |
549 | */ | |
bf800ef1 KO |
550 | struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask, |
551 | struct bio_set *bs) | |
1da177e4 | 552 | { |
bf800ef1 | 553 | struct bio *b; |
1da177e4 | 554 | |
bf800ef1 | 555 | b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs); |
7ba1ba12 MP |
556 | if (!b) |
557 | return NULL; | |
558 | ||
7ba1ba12 MP |
559 | __bio_clone(b, bio); |
560 | ||
561 | if (bio_integrity(bio)) { | |
562 | int ret; | |
563 | ||
1e2a410f | 564 | ret = bio_integrity_clone(b, bio, gfp_mask); |
7ba1ba12 | 565 | |
059ea331 LZ |
566 | if (ret < 0) { |
567 | bio_put(b); | |
7ba1ba12 | 568 | return NULL; |
059ea331 | 569 | } |
3676347a | 570 | } |
1da177e4 LT |
571 | |
572 | return b; | |
573 | } | |
bf800ef1 | 574 | EXPORT_SYMBOL(bio_clone_bioset); |
1da177e4 LT |
575 | |
576 | /** | |
577 | * bio_get_nr_vecs - return approx number of vecs | |
578 | * @bdev: I/O target | |
579 | * | |
580 | * Return the approximate number of pages we can send to this target. | |
581 | * There's no guarantee that you will be able to fit this number of pages | |
582 | * into a bio, it does not account for dynamic restrictions that vary | |
583 | * on offset. | |
584 | */ | |
585 | int bio_get_nr_vecs(struct block_device *bdev) | |
586 | { | |
165125e1 | 587 | struct request_queue *q = bdev_get_queue(bdev); |
f908ee94 BS |
588 | int nr_pages; |
589 | ||
590 | nr_pages = min_t(unsigned, | |
5abebfdd KO |
591 | queue_max_segments(q), |
592 | queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); | |
f908ee94 BS |
593 | |
594 | return min_t(unsigned, nr_pages, BIO_MAX_PAGES); | |
595 | ||
1da177e4 | 596 | } |
a112a71d | 597 | EXPORT_SYMBOL(bio_get_nr_vecs); |
1da177e4 | 598 | |
165125e1 | 599 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 MC |
600 | *page, unsigned int len, unsigned int offset, |
601 | unsigned short max_sectors) | |
1da177e4 LT |
602 | { |
603 | int retried_segments = 0; | |
604 | struct bio_vec *bvec; | |
605 | ||
606 | /* | |
607 | * cloned bio must not modify vec list | |
608 | */ | |
609 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
610 | return 0; | |
611 | ||
80cfd548 | 612 | if (((bio->bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
613 | return 0; |
614 | ||
80cfd548 JA |
615 | /* |
616 | * For filesystems with a blocksize smaller than the pagesize | |
617 | * we will often be called with the same page as last time and | |
618 | * a consecutive offset. Optimize this special case. | |
619 | */ | |
620 | if (bio->bi_vcnt > 0) { | |
621 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
622 | ||
623 | if (page == prev->bv_page && | |
624 | offset == prev->bv_offset + prev->bv_len) { | |
1d616585 | 625 | unsigned int prev_bv_len = prev->bv_len; |
80cfd548 | 626 | prev->bv_len += len; |
cc371e66 AK |
627 | |
628 | if (q->merge_bvec_fn) { | |
629 | struct bvec_merge_data bvm = { | |
1d616585 DM |
630 | /* prev_bvec is already charged in |
631 | bi_size, discharge it in order to | |
632 | simulate merging updated prev_bvec | |
633 | as new bvec. */ | |
cc371e66 AK |
634 | .bi_bdev = bio->bi_bdev, |
635 | .bi_sector = bio->bi_sector, | |
1d616585 | 636 | .bi_size = bio->bi_size - prev_bv_len, |
cc371e66 AK |
637 | .bi_rw = bio->bi_rw, |
638 | }; | |
639 | ||
8bf8c376 | 640 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { |
cc371e66 AK |
641 | prev->bv_len -= len; |
642 | return 0; | |
643 | } | |
80cfd548 JA |
644 | } |
645 | ||
646 | goto done; | |
647 | } | |
648 | } | |
649 | ||
650 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
651 | return 0; |
652 | ||
653 | /* | |
654 | * we might lose a segment or two here, but rather that than | |
655 | * make this too complex. | |
656 | */ | |
657 | ||
8a78362c | 658 | while (bio->bi_phys_segments >= queue_max_segments(q)) { |
1da177e4 LT |
659 | |
660 | if (retried_segments) | |
661 | return 0; | |
662 | ||
663 | retried_segments = 1; | |
664 | blk_recount_segments(q, bio); | |
665 | } | |
666 | ||
667 | /* | |
668 | * setup the new entry, we might clear it again later if we | |
669 | * cannot add the page | |
670 | */ | |
671 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
672 | bvec->bv_page = page; | |
673 | bvec->bv_len = len; | |
674 | bvec->bv_offset = offset; | |
675 | ||
676 | /* | |
677 | * if queue has other restrictions (eg varying max sector size | |
678 | * depending on offset), it can specify a merge_bvec_fn in the | |
679 | * queue to get further control | |
680 | */ | |
681 | if (q->merge_bvec_fn) { | |
cc371e66 AK |
682 | struct bvec_merge_data bvm = { |
683 | .bi_bdev = bio->bi_bdev, | |
684 | .bi_sector = bio->bi_sector, | |
685 | .bi_size = bio->bi_size, | |
686 | .bi_rw = bio->bi_rw, | |
687 | }; | |
688 | ||
1da177e4 LT |
689 | /* |
690 | * merge_bvec_fn() returns number of bytes it can accept | |
691 | * at this offset | |
692 | */ | |
8bf8c376 | 693 | if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) { |
1da177e4 LT |
694 | bvec->bv_page = NULL; |
695 | bvec->bv_len = 0; | |
696 | bvec->bv_offset = 0; | |
697 | return 0; | |
698 | } | |
699 | } | |
700 | ||
701 | /* If we may be able to merge these biovecs, force a recount */ | |
b8b3e16c | 702 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
1da177e4 LT |
703 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
704 | ||
705 | bio->bi_vcnt++; | |
706 | bio->bi_phys_segments++; | |
80cfd548 | 707 | done: |
1da177e4 LT |
708 | bio->bi_size += len; |
709 | return len; | |
710 | } | |
711 | ||
6e68af66 MC |
712 | /** |
713 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 714 | * @q: the target queue |
6e68af66 MC |
715 | * @bio: destination bio |
716 | * @page: page to add | |
717 | * @len: vec entry length | |
718 | * @offset: vec entry offset | |
719 | * | |
720 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
c6428084 AG |
721 | * number of reasons, such as the bio being full or target block device |
722 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
723 | * so it is always possible to add a single page to an empty bio. | |
724 | * | |
725 | * This should only be used by REQ_PC bios. | |
6e68af66 | 726 | */ |
165125e1 | 727 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
728 | unsigned int len, unsigned int offset) |
729 | { | |
ae03bf63 MP |
730 | return __bio_add_page(q, bio, page, len, offset, |
731 | queue_max_hw_sectors(q)); | |
6e68af66 | 732 | } |
a112a71d | 733 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 734 | |
1da177e4 LT |
735 | /** |
736 | * bio_add_page - attempt to add page to bio | |
737 | * @bio: destination bio | |
738 | * @page: page to add | |
739 | * @len: vec entry length | |
740 | * @offset: vec entry offset | |
741 | * | |
742 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
c6428084 AG |
743 | * number of reasons, such as the bio being full or target block device |
744 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
745 | * so it is always possible to add a single page to an empty bio. | |
1da177e4 LT |
746 | */ |
747 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
748 | unsigned int offset) | |
749 | { | |
defd94b7 | 750 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
ae03bf63 | 751 | return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q)); |
1da177e4 | 752 | } |
a112a71d | 753 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 754 | |
054bdf64 KO |
755 | /** |
756 | * bio_advance - increment/complete a bio by some number of bytes | |
757 | * @bio: bio to advance | |
758 | * @bytes: number of bytes to complete | |
759 | * | |
760 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
761 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
762 | * be updated on the last bvec as well. | |
763 | * | |
764 | * @bio will then represent the remaining, uncompleted portion of the io. | |
765 | */ | |
766 | void bio_advance(struct bio *bio, unsigned bytes) | |
767 | { | |
768 | if (bio_integrity(bio)) | |
769 | bio_integrity_advance(bio, bytes); | |
770 | ||
771 | bio->bi_sector += bytes >> 9; | |
772 | bio->bi_size -= bytes; | |
773 | ||
774 | if (bio->bi_rw & BIO_NO_ADVANCE_ITER_MASK) | |
775 | return; | |
776 | ||
777 | while (bytes) { | |
778 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
779 | WARN_ONCE(1, "bio idx %d >= vcnt %d\n", | |
780 | bio->bi_idx, bio->bi_vcnt); | |
781 | break; | |
782 | } | |
783 | ||
784 | if (bytes >= bio_iovec(bio)->bv_len) { | |
785 | bytes -= bio_iovec(bio)->bv_len; | |
786 | bio->bi_idx++; | |
787 | } else { | |
788 | bio_iovec(bio)->bv_len -= bytes; | |
789 | bio_iovec(bio)->bv_offset += bytes; | |
790 | bytes = 0; | |
791 | } | |
792 | } | |
793 | } | |
794 | EXPORT_SYMBOL(bio_advance); | |
795 | ||
1da177e4 LT |
796 | struct bio_map_data { |
797 | struct bio_vec *iovecs; | |
c5dec1c3 | 798 | struct sg_iovec *sgvecs; |
152e283f FT |
799 | int nr_sgvecs; |
800 | int is_our_pages; | |
1da177e4 LT |
801 | }; |
802 | ||
c5dec1c3 | 803 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
152e283f FT |
804 | struct sg_iovec *iov, int iov_count, |
805 | int is_our_pages) | |
1da177e4 LT |
806 | { |
807 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
808 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
809 | bmd->nr_sgvecs = iov_count; | |
152e283f | 810 | bmd->is_our_pages = is_our_pages; |
1da177e4 LT |
811 | bio->bi_private = bmd; |
812 | } | |
813 | ||
814 | static void bio_free_map_data(struct bio_map_data *bmd) | |
815 | { | |
816 | kfree(bmd->iovecs); | |
c5dec1c3 | 817 | kfree(bmd->sgvecs); |
1da177e4 LT |
818 | kfree(bmd); |
819 | } | |
820 | ||
121f0994 DC |
821 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, |
822 | unsigned int iov_count, | |
76029ff3 | 823 | gfp_t gfp_mask) |
1da177e4 | 824 | { |
f3f63c1c JA |
825 | struct bio_map_data *bmd; |
826 | ||
827 | if (iov_count > UIO_MAXIOV) | |
828 | return NULL; | |
1da177e4 | 829 | |
f3f63c1c | 830 | bmd = kmalloc(sizeof(*bmd), gfp_mask); |
1da177e4 LT |
831 | if (!bmd) |
832 | return NULL; | |
833 | ||
76029ff3 | 834 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
c5dec1c3 FT |
835 | if (!bmd->iovecs) { |
836 | kfree(bmd); | |
837 | return NULL; | |
838 | } | |
839 | ||
76029ff3 | 840 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
c5dec1c3 | 841 | if (bmd->sgvecs) |
1da177e4 LT |
842 | return bmd; |
843 | ||
c5dec1c3 | 844 | kfree(bmd->iovecs); |
1da177e4 LT |
845 | kfree(bmd); |
846 | return NULL; | |
847 | } | |
848 | ||
aefcc28a | 849 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
ecb554a8 FT |
850 | struct sg_iovec *iov, int iov_count, |
851 | int to_user, int from_user, int do_free_page) | |
c5dec1c3 FT |
852 | { |
853 | int ret = 0, i; | |
854 | struct bio_vec *bvec; | |
855 | int iov_idx = 0; | |
856 | unsigned int iov_off = 0; | |
c5dec1c3 FT |
857 | |
858 | __bio_for_each_segment(bvec, bio, i, 0) { | |
859 | char *bv_addr = page_address(bvec->bv_page); | |
aefcc28a | 860 | unsigned int bv_len = iovecs[i].bv_len; |
c5dec1c3 FT |
861 | |
862 | while (bv_len && iov_idx < iov_count) { | |
863 | unsigned int bytes; | |
0e0c6212 | 864 | char __user *iov_addr; |
c5dec1c3 FT |
865 | |
866 | bytes = min_t(unsigned int, | |
867 | iov[iov_idx].iov_len - iov_off, bv_len); | |
868 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
869 | ||
870 | if (!ret) { | |
ecb554a8 | 871 | if (to_user) |
c5dec1c3 FT |
872 | ret = copy_to_user(iov_addr, bv_addr, |
873 | bytes); | |
874 | ||
ecb554a8 FT |
875 | if (from_user) |
876 | ret = copy_from_user(bv_addr, iov_addr, | |
877 | bytes); | |
878 | ||
c5dec1c3 FT |
879 | if (ret) |
880 | ret = -EFAULT; | |
881 | } | |
882 | ||
883 | bv_len -= bytes; | |
884 | bv_addr += bytes; | |
885 | iov_addr += bytes; | |
886 | iov_off += bytes; | |
887 | ||
888 | if (iov[iov_idx].iov_len == iov_off) { | |
889 | iov_idx++; | |
890 | iov_off = 0; | |
891 | } | |
892 | } | |
893 | ||
152e283f | 894 | if (do_free_page) |
c5dec1c3 FT |
895 | __free_page(bvec->bv_page); |
896 | } | |
897 | ||
898 | return ret; | |
899 | } | |
900 | ||
1da177e4 LT |
901 | /** |
902 | * bio_uncopy_user - finish previously mapped bio | |
903 | * @bio: bio being terminated | |
904 | * | |
905 | * Free pages allocated from bio_copy_user() and write back data | |
906 | * to user space in case of a read. | |
907 | */ | |
908 | int bio_uncopy_user(struct bio *bio) | |
909 | { | |
910 | struct bio_map_data *bmd = bio->bi_private; | |
81882766 | 911 | int ret = 0; |
1da177e4 | 912 | |
81882766 FT |
913 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
914 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, | |
ecb554a8 FT |
915 | bmd->nr_sgvecs, bio_data_dir(bio) == READ, |
916 | 0, bmd->is_our_pages); | |
1da177e4 LT |
917 | bio_free_map_data(bmd); |
918 | bio_put(bio); | |
919 | return ret; | |
920 | } | |
a112a71d | 921 | EXPORT_SYMBOL(bio_uncopy_user); |
1da177e4 LT |
922 | |
923 | /** | |
c5dec1c3 | 924 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 925 | * @q: destination block queue |
152e283f | 926 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
927 | * @iov: the iovec. |
928 | * @iov_count: number of elements in the iovec | |
1da177e4 | 929 | * @write_to_vm: bool indicating writing to pages or not |
a3bce90e | 930 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
931 | * |
932 | * Prepares and returns a bio for indirect user io, bouncing data | |
933 | * to/from kernel pages as necessary. Must be paired with | |
934 | * call bio_uncopy_user() on io completion. | |
935 | */ | |
152e283f FT |
936 | struct bio *bio_copy_user_iov(struct request_queue *q, |
937 | struct rq_map_data *map_data, | |
938 | struct sg_iovec *iov, int iov_count, | |
939 | int write_to_vm, gfp_t gfp_mask) | |
1da177e4 | 940 | { |
1da177e4 LT |
941 | struct bio_map_data *bmd; |
942 | struct bio_vec *bvec; | |
943 | struct page *page; | |
944 | struct bio *bio; | |
945 | int i, ret; | |
c5dec1c3 FT |
946 | int nr_pages = 0; |
947 | unsigned int len = 0; | |
56c451f4 | 948 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; |
1da177e4 | 949 | |
c5dec1c3 FT |
950 | for (i = 0; i < iov_count; i++) { |
951 | unsigned long uaddr; | |
952 | unsigned long end; | |
953 | unsigned long start; | |
954 | ||
955 | uaddr = (unsigned long)iov[i].iov_base; | |
956 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
957 | start = uaddr >> PAGE_SHIFT; | |
958 | ||
cb4644ca JA |
959 | /* |
960 | * Overflow, abort | |
961 | */ | |
962 | if (end < start) | |
963 | return ERR_PTR(-EINVAL); | |
964 | ||
c5dec1c3 FT |
965 | nr_pages += end - start; |
966 | len += iov[i].iov_len; | |
967 | } | |
968 | ||
69838727 FT |
969 | if (offset) |
970 | nr_pages++; | |
971 | ||
a3bce90e | 972 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
1da177e4 LT |
973 | if (!bmd) |
974 | return ERR_PTR(-ENOMEM); | |
975 | ||
1da177e4 | 976 | ret = -ENOMEM; |
a9e9dc24 | 977 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
978 | if (!bio) |
979 | goto out_bmd; | |
980 | ||
7b6d91da CH |
981 | if (!write_to_vm) |
982 | bio->bi_rw |= REQ_WRITE; | |
1da177e4 LT |
983 | |
984 | ret = 0; | |
56c451f4 FT |
985 | |
986 | if (map_data) { | |
e623ddb4 | 987 | nr_pages = 1 << map_data->page_order; |
56c451f4 FT |
988 | i = map_data->offset / PAGE_SIZE; |
989 | } | |
1da177e4 | 990 | while (len) { |
e623ddb4 | 991 | unsigned int bytes = PAGE_SIZE; |
1da177e4 | 992 | |
56c451f4 FT |
993 | bytes -= offset; |
994 | ||
1da177e4 LT |
995 | if (bytes > len) |
996 | bytes = len; | |
997 | ||
152e283f | 998 | if (map_data) { |
e623ddb4 | 999 | if (i == map_data->nr_entries * nr_pages) { |
152e283f FT |
1000 | ret = -ENOMEM; |
1001 | break; | |
1002 | } | |
e623ddb4 FT |
1003 | |
1004 | page = map_data->pages[i / nr_pages]; | |
1005 | page += (i % nr_pages); | |
1006 | ||
1007 | i++; | |
1008 | } else { | |
152e283f | 1009 | page = alloc_page(q->bounce_gfp | gfp_mask); |
e623ddb4 FT |
1010 | if (!page) { |
1011 | ret = -ENOMEM; | |
1012 | break; | |
1013 | } | |
1da177e4 LT |
1014 | } |
1015 | ||
56c451f4 | 1016 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) |
1da177e4 | 1017 | break; |
1da177e4 LT |
1018 | |
1019 | len -= bytes; | |
56c451f4 | 1020 | offset = 0; |
1da177e4 LT |
1021 | } |
1022 | ||
1023 | if (ret) | |
1024 | goto cleanup; | |
1025 | ||
1026 | /* | |
1027 | * success | |
1028 | */ | |
ecb554a8 FT |
1029 | if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || |
1030 | (map_data && map_data->from_user)) { | |
1031 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0); | |
c5dec1c3 FT |
1032 | if (ret) |
1033 | goto cleanup; | |
1da177e4 LT |
1034 | } |
1035 | ||
152e283f | 1036 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
1da177e4 LT |
1037 | return bio; |
1038 | cleanup: | |
152e283f FT |
1039 | if (!map_data) |
1040 | bio_for_each_segment(bvec, bio, i) | |
1041 | __free_page(bvec->bv_page); | |
1da177e4 LT |
1042 | |
1043 | bio_put(bio); | |
1044 | out_bmd: | |
1045 | bio_free_map_data(bmd); | |
1046 | return ERR_PTR(ret); | |
1047 | } | |
1048 | ||
c5dec1c3 FT |
1049 | /** |
1050 | * bio_copy_user - copy user data to bio | |
1051 | * @q: destination block queue | |
152e283f | 1052 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
1053 | * @uaddr: start of user address |
1054 | * @len: length in bytes | |
1055 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 1056 | * @gfp_mask: memory allocation flags |
c5dec1c3 FT |
1057 | * |
1058 | * Prepares and returns a bio for indirect user io, bouncing data | |
1059 | * to/from kernel pages as necessary. Must be paired with | |
1060 | * call bio_uncopy_user() on io completion. | |
1061 | */ | |
152e283f FT |
1062 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
1063 | unsigned long uaddr, unsigned int len, | |
1064 | int write_to_vm, gfp_t gfp_mask) | |
c5dec1c3 FT |
1065 | { |
1066 | struct sg_iovec iov; | |
1067 | ||
1068 | iov.iov_base = (void __user *)uaddr; | |
1069 | iov.iov_len = len; | |
1070 | ||
152e283f | 1071 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
c5dec1c3 | 1072 | } |
a112a71d | 1073 | EXPORT_SYMBOL(bio_copy_user); |
c5dec1c3 | 1074 | |
165125e1 | 1075 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
1076 | struct block_device *bdev, |
1077 | struct sg_iovec *iov, int iov_count, | |
a3bce90e | 1078 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 | 1079 | { |
f1970baf JB |
1080 | int i, j; |
1081 | int nr_pages = 0; | |
1da177e4 LT |
1082 | struct page **pages; |
1083 | struct bio *bio; | |
f1970baf JB |
1084 | int cur_page = 0; |
1085 | int ret, offset; | |
1da177e4 | 1086 | |
f1970baf JB |
1087 | for (i = 0; i < iov_count; i++) { |
1088 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1089 | unsigned long len = iov[i].iov_len; | |
1090 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1091 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1092 | ||
cb4644ca JA |
1093 | /* |
1094 | * Overflow, abort | |
1095 | */ | |
1096 | if (end < start) | |
1097 | return ERR_PTR(-EINVAL); | |
1098 | ||
f1970baf JB |
1099 | nr_pages += end - start; |
1100 | /* | |
ad2d7225 | 1101 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 1102 | */ |
ad2d7225 | 1103 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
1104 | return ERR_PTR(-EINVAL); |
1105 | } | |
1106 | ||
1107 | if (!nr_pages) | |
1da177e4 LT |
1108 | return ERR_PTR(-EINVAL); |
1109 | ||
a9e9dc24 | 1110 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1111 | if (!bio) |
1112 | return ERR_PTR(-ENOMEM); | |
1113 | ||
1114 | ret = -ENOMEM; | |
a3bce90e | 1115 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
1da177e4 LT |
1116 | if (!pages) |
1117 | goto out; | |
1118 | ||
f1970baf JB |
1119 | for (i = 0; i < iov_count; i++) { |
1120 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1121 | unsigned long len = iov[i].iov_len; | |
1122 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1123 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1124 | const int local_nr_pages = end - start; | |
1125 | const int page_limit = cur_page + local_nr_pages; | |
cb4644ca | 1126 | |
f5dd33c4 NP |
1127 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
1128 | write_to_vm, &pages[cur_page]); | |
99172157 JA |
1129 | if (ret < local_nr_pages) { |
1130 | ret = -EFAULT; | |
f1970baf | 1131 | goto out_unmap; |
99172157 | 1132 | } |
f1970baf JB |
1133 | |
1134 | offset = uaddr & ~PAGE_MASK; | |
1135 | for (j = cur_page; j < page_limit; j++) { | |
1136 | unsigned int bytes = PAGE_SIZE - offset; | |
1137 | ||
1138 | if (len <= 0) | |
1139 | break; | |
1140 | ||
1141 | if (bytes > len) | |
1142 | bytes = len; | |
1143 | ||
1144 | /* | |
1145 | * sorry... | |
1146 | */ | |
defd94b7 MC |
1147 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
1148 | bytes) | |
f1970baf JB |
1149 | break; |
1150 | ||
1151 | len -= bytes; | |
1152 | offset = 0; | |
1153 | } | |
1da177e4 | 1154 | |
f1970baf | 1155 | cur_page = j; |
1da177e4 | 1156 | /* |
f1970baf | 1157 | * release the pages we didn't map into the bio, if any |
1da177e4 | 1158 | */ |
f1970baf JB |
1159 | while (j < page_limit) |
1160 | page_cache_release(pages[j++]); | |
1da177e4 LT |
1161 | } |
1162 | ||
1da177e4 LT |
1163 | kfree(pages); |
1164 | ||
1165 | /* | |
1166 | * set data direction, and check if mapped pages need bouncing | |
1167 | */ | |
1168 | if (!write_to_vm) | |
7b6d91da | 1169 | bio->bi_rw |= REQ_WRITE; |
1da177e4 | 1170 | |
f1970baf | 1171 | bio->bi_bdev = bdev; |
1da177e4 LT |
1172 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
1173 | return bio; | |
f1970baf JB |
1174 | |
1175 | out_unmap: | |
1176 | for (i = 0; i < nr_pages; i++) { | |
1177 | if(!pages[i]) | |
1178 | break; | |
1179 | page_cache_release(pages[i]); | |
1180 | } | |
1181 | out: | |
1da177e4 LT |
1182 | kfree(pages); |
1183 | bio_put(bio); | |
1184 | return ERR_PTR(ret); | |
1185 | } | |
1186 | ||
1187 | /** | |
1188 | * bio_map_user - map user address into bio | |
165125e1 | 1189 | * @q: the struct request_queue for the bio |
1da177e4 LT |
1190 | * @bdev: destination block device |
1191 | * @uaddr: start of user address | |
1192 | * @len: length in bytes | |
1193 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 1194 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
1195 | * |
1196 | * Map the user space address into a bio suitable for io to a block | |
1197 | * device. Returns an error pointer in case of error. | |
1198 | */ | |
165125e1 | 1199 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
a3bce90e FT |
1200 | unsigned long uaddr, unsigned int len, int write_to_vm, |
1201 | gfp_t gfp_mask) | |
f1970baf JB |
1202 | { |
1203 | struct sg_iovec iov; | |
1204 | ||
3f70353e | 1205 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
1206 | iov.iov_len = len; |
1207 | ||
a3bce90e | 1208 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
f1970baf | 1209 | } |
a112a71d | 1210 | EXPORT_SYMBOL(bio_map_user); |
f1970baf JB |
1211 | |
1212 | /** | |
1213 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 1214 | * @q: the struct request_queue for the bio |
f1970baf JB |
1215 | * @bdev: destination block device |
1216 | * @iov: the iovec. | |
1217 | * @iov_count: number of elements in the iovec | |
1218 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 1219 | * @gfp_mask: memory allocation flags |
f1970baf JB |
1220 | * |
1221 | * Map the user space address into a bio suitable for io to a block | |
1222 | * device. Returns an error pointer in case of error. | |
1223 | */ | |
165125e1 | 1224 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf | 1225 | struct sg_iovec *iov, int iov_count, |
a3bce90e | 1226 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 LT |
1227 | { |
1228 | struct bio *bio; | |
1229 | ||
a3bce90e FT |
1230 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
1231 | gfp_mask); | |
1da177e4 LT |
1232 | if (IS_ERR(bio)) |
1233 | return bio; | |
1234 | ||
1235 | /* | |
1236 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
1237 | * it would normally disappear when its bi_end_io is run. | |
1238 | * however, we need it for the unmap, so grab an extra | |
1239 | * reference to it | |
1240 | */ | |
1241 | bio_get(bio); | |
1242 | ||
0e75f906 | 1243 | return bio; |
1da177e4 LT |
1244 | } |
1245 | ||
1246 | static void __bio_unmap_user(struct bio *bio) | |
1247 | { | |
1248 | struct bio_vec *bvec; | |
1249 | int i; | |
1250 | ||
1251 | /* | |
1252 | * make sure we dirty pages we wrote to | |
1253 | */ | |
1254 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1255 | if (bio_data_dir(bio) == READ) | |
1256 | set_page_dirty_lock(bvec->bv_page); | |
1257 | ||
1258 | page_cache_release(bvec->bv_page); | |
1259 | } | |
1260 | ||
1261 | bio_put(bio); | |
1262 | } | |
1263 | ||
1264 | /** | |
1265 | * bio_unmap_user - unmap a bio | |
1266 | * @bio: the bio being unmapped | |
1267 | * | |
1268 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
1269 | * a process context. | |
1270 | * | |
1271 | * bio_unmap_user() may sleep. | |
1272 | */ | |
1273 | void bio_unmap_user(struct bio *bio) | |
1274 | { | |
1275 | __bio_unmap_user(bio); | |
1276 | bio_put(bio); | |
1277 | } | |
a112a71d | 1278 | EXPORT_SYMBOL(bio_unmap_user); |
1da177e4 | 1279 | |
6712ecf8 | 1280 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 1281 | { |
b823825e | 1282 | bio_put(bio); |
b823825e JA |
1283 | } |
1284 | ||
165125e1 | 1285 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 1286 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
1287 | { |
1288 | unsigned long kaddr = (unsigned long)data; | |
1289 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1290 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1291 | const int nr_pages = end - start; | |
1292 | int offset, i; | |
1293 | struct bio *bio; | |
1294 | ||
a9e9dc24 | 1295 | bio = bio_kmalloc(gfp_mask, nr_pages); |
df46b9a4 MC |
1296 | if (!bio) |
1297 | return ERR_PTR(-ENOMEM); | |
1298 | ||
1299 | offset = offset_in_page(kaddr); | |
1300 | for (i = 0; i < nr_pages; i++) { | |
1301 | unsigned int bytes = PAGE_SIZE - offset; | |
1302 | ||
1303 | if (len <= 0) | |
1304 | break; | |
1305 | ||
1306 | if (bytes > len) | |
1307 | bytes = len; | |
1308 | ||
defd94b7 MC |
1309 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
1310 | offset) < bytes) | |
df46b9a4 MC |
1311 | break; |
1312 | ||
1313 | data += bytes; | |
1314 | len -= bytes; | |
1315 | offset = 0; | |
1316 | } | |
1317 | ||
b823825e | 1318 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1319 | return bio; |
1320 | } | |
1321 | ||
1322 | /** | |
1323 | * bio_map_kern - map kernel address into bio | |
165125e1 | 1324 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
1325 | * @data: pointer to buffer to map |
1326 | * @len: length in bytes | |
1327 | * @gfp_mask: allocation flags for bio allocation | |
1328 | * | |
1329 | * Map the kernel address into a bio suitable for io to a block | |
1330 | * device. Returns an error pointer in case of error. | |
1331 | */ | |
165125e1 | 1332 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 1333 | gfp_t gfp_mask) |
df46b9a4 MC |
1334 | { |
1335 | struct bio *bio; | |
1336 | ||
1337 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
1338 | if (IS_ERR(bio)) | |
1339 | return bio; | |
1340 | ||
1341 | if (bio->bi_size == len) | |
1342 | return bio; | |
1343 | ||
1344 | /* | |
1345 | * Don't support partial mappings. | |
1346 | */ | |
1347 | bio_put(bio); | |
1348 | return ERR_PTR(-EINVAL); | |
1349 | } | |
a112a71d | 1350 | EXPORT_SYMBOL(bio_map_kern); |
df46b9a4 | 1351 | |
68154e90 FT |
1352 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1353 | { | |
1354 | struct bio_vec *bvec; | |
1355 | const int read = bio_data_dir(bio) == READ; | |
76029ff3 | 1356 | struct bio_map_data *bmd = bio->bi_private; |
68154e90 | 1357 | int i; |
76029ff3 | 1358 | char *p = bmd->sgvecs[0].iov_base; |
68154e90 FT |
1359 | |
1360 | __bio_for_each_segment(bvec, bio, i, 0) { | |
1361 | char *addr = page_address(bvec->bv_page); | |
76029ff3 | 1362 | int len = bmd->iovecs[i].bv_len; |
68154e90 | 1363 | |
4fc981ef | 1364 | if (read) |
76029ff3 | 1365 | memcpy(p, addr, len); |
68154e90 FT |
1366 | |
1367 | __free_page(bvec->bv_page); | |
76029ff3 | 1368 | p += len; |
68154e90 FT |
1369 | } |
1370 | ||
76029ff3 | 1371 | bio_free_map_data(bmd); |
68154e90 FT |
1372 | bio_put(bio); |
1373 | } | |
1374 | ||
1375 | /** | |
1376 | * bio_copy_kern - copy kernel address into bio | |
1377 | * @q: the struct request_queue for the bio | |
1378 | * @data: pointer to buffer to copy | |
1379 | * @len: length in bytes | |
1380 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1381 | * @reading: data direction is READ |
68154e90 FT |
1382 | * |
1383 | * copy the kernel address into a bio suitable for io to a block | |
1384 | * device. Returns an error pointer in case of error. | |
1385 | */ | |
1386 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1387 | gfp_t gfp_mask, int reading) | |
1388 | { | |
68154e90 FT |
1389 | struct bio *bio; |
1390 | struct bio_vec *bvec; | |
4d8ab62e | 1391 | int i; |
68154e90 | 1392 | |
4d8ab62e FT |
1393 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
1394 | if (IS_ERR(bio)) | |
1395 | return bio; | |
68154e90 FT |
1396 | |
1397 | if (!reading) { | |
1398 | void *p = data; | |
1399 | ||
1400 | bio_for_each_segment(bvec, bio, i) { | |
1401 | char *addr = page_address(bvec->bv_page); | |
1402 | ||
1403 | memcpy(addr, p, bvec->bv_len); | |
1404 | p += bvec->bv_len; | |
1405 | } | |
1406 | } | |
1407 | ||
68154e90 | 1408 | bio->bi_end_io = bio_copy_kern_endio; |
76029ff3 | 1409 | |
68154e90 | 1410 | return bio; |
68154e90 | 1411 | } |
a112a71d | 1412 | EXPORT_SYMBOL(bio_copy_kern); |
68154e90 | 1413 | |
1da177e4 LT |
1414 | /* |
1415 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1416 | * for performing direct-IO in BIOs. | |
1417 | * | |
1418 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1419 | * because the required locks are not interrupt-safe. So what we can do is to | |
1420 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1421 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1422 | * in process context. | |
1423 | * | |
1424 | * We special-case compound pages here: normally this means reads into hugetlb | |
1425 | * pages. The logic in here doesn't really work right for compound pages | |
1426 | * because the VM does not uniformly chase down the head page in all cases. | |
1427 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1428 | * handle them at all. So we skip compound pages here at an early stage. | |
1429 | * | |
1430 | * Note that this code is very hard to test under normal circumstances because | |
1431 | * direct-io pins the pages with get_user_pages(). This makes | |
1432 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1433 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1434 | * pagecache. |
1435 | * | |
1436 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1437 | * deferred bio dirtying paths. | |
1438 | */ | |
1439 | ||
1440 | /* | |
1441 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1442 | */ | |
1443 | void bio_set_pages_dirty(struct bio *bio) | |
1444 | { | |
1445 | struct bio_vec *bvec = bio->bi_io_vec; | |
1446 | int i; | |
1447 | ||
1448 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1449 | struct page *page = bvec[i].bv_page; | |
1450 | ||
1451 | if (page && !PageCompound(page)) | |
1452 | set_page_dirty_lock(page); | |
1453 | } | |
1454 | } | |
1455 | ||
86b6c7a7 | 1456 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1457 | { |
1458 | struct bio_vec *bvec = bio->bi_io_vec; | |
1459 | int i; | |
1460 | ||
1461 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1462 | struct page *page = bvec[i].bv_page; | |
1463 | ||
1464 | if (page) | |
1465 | put_page(page); | |
1466 | } | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1471 | * If they are, then fine. If, however, some pages are clean then they must | |
1472 | * have been written out during the direct-IO read. So we take another ref on | |
1473 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1474 | * | |
1475 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1476 | * here on. It will run one page_cache_release() against each page and will | |
1477 | * run one bio_put() against the BIO. | |
1478 | */ | |
1479 | ||
65f27f38 | 1480 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1481 | |
65f27f38 | 1482 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1483 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1484 | static struct bio *bio_dirty_list; | |
1485 | ||
1486 | /* | |
1487 | * This runs in process context | |
1488 | */ | |
65f27f38 | 1489 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1490 | { |
1491 | unsigned long flags; | |
1492 | struct bio *bio; | |
1493 | ||
1494 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1495 | bio = bio_dirty_list; | |
1496 | bio_dirty_list = NULL; | |
1497 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1498 | ||
1499 | while (bio) { | |
1500 | struct bio *next = bio->bi_private; | |
1501 | ||
1502 | bio_set_pages_dirty(bio); | |
1503 | bio_release_pages(bio); | |
1504 | bio_put(bio); | |
1505 | bio = next; | |
1506 | } | |
1507 | } | |
1508 | ||
1509 | void bio_check_pages_dirty(struct bio *bio) | |
1510 | { | |
1511 | struct bio_vec *bvec = bio->bi_io_vec; | |
1512 | int nr_clean_pages = 0; | |
1513 | int i; | |
1514 | ||
1515 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1516 | struct page *page = bvec[i].bv_page; | |
1517 | ||
1518 | if (PageDirty(page) || PageCompound(page)) { | |
1519 | page_cache_release(page); | |
1520 | bvec[i].bv_page = NULL; | |
1521 | } else { | |
1522 | nr_clean_pages++; | |
1523 | } | |
1524 | } | |
1525 | ||
1526 | if (nr_clean_pages) { | |
1527 | unsigned long flags; | |
1528 | ||
1529 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1530 | bio->bi_private = bio_dirty_list; | |
1531 | bio_dirty_list = bio; | |
1532 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1533 | schedule_work(&bio_dirty_work); | |
1534 | } else { | |
1535 | bio_put(bio); | |
1536 | } | |
1537 | } | |
1538 | ||
2d4dc890 IL |
1539 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1540 | void bio_flush_dcache_pages(struct bio *bi) | |
1541 | { | |
1542 | int i; | |
1543 | struct bio_vec *bvec; | |
1544 | ||
1545 | bio_for_each_segment(bvec, bi, i) | |
1546 | flush_dcache_page(bvec->bv_page); | |
1547 | } | |
1548 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1549 | #endif | |
1550 | ||
1da177e4 LT |
1551 | /** |
1552 | * bio_endio - end I/O on a bio | |
1553 | * @bio: bio | |
1da177e4 LT |
1554 | * @error: error, if any |
1555 | * | |
1556 | * Description: | |
6712ecf8 | 1557 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1558 | * preferred way to end I/O on a bio, it takes care of clearing |
1559 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1560 | * established -Exxxx (-EIO, for instance) error values in case | |
25985edc | 1561 | * something went wrong. No one should call bi_end_io() directly on a |
5bb23a68 N |
1562 | * bio unless they own it and thus know that it has an end_io |
1563 | * function. | |
1da177e4 | 1564 | **/ |
6712ecf8 | 1565 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1566 | { |
1567 | if (error) | |
1568 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1569 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1570 | error = -EIO; | |
1da177e4 | 1571 | |
3a366e61 TH |
1572 | trace_block_bio_complete(bio, error); |
1573 | ||
5bb23a68 | 1574 | if (bio->bi_end_io) |
6712ecf8 | 1575 | bio->bi_end_io(bio, error); |
1da177e4 | 1576 | } |
a112a71d | 1577 | EXPORT_SYMBOL(bio_endio); |
1da177e4 LT |
1578 | |
1579 | void bio_pair_release(struct bio_pair *bp) | |
1580 | { | |
1581 | if (atomic_dec_and_test(&bp->cnt)) { | |
1582 | struct bio *master = bp->bio1.bi_private; | |
1583 | ||
6712ecf8 | 1584 | bio_endio(master, bp->error); |
1da177e4 LT |
1585 | mempool_free(bp, bp->bio2.bi_private); |
1586 | } | |
1587 | } | |
a112a71d | 1588 | EXPORT_SYMBOL(bio_pair_release); |
1da177e4 | 1589 | |
6712ecf8 | 1590 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1591 | { |
1592 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1593 | ||
1594 | if (err) | |
1595 | bp->error = err; | |
1596 | ||
1da177e4 | 1597 | bio_pair_release(bp); |
1da177e4 LT |
1598 | } |
1599 | ||
6712ecf8 | 1600 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1601 | { |
1602 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1603 | ||
1604 | if (err) | |
1605 | bp->error = err; | |
1606 | ||
1da177e4 | 1607 | bio_pair_release(bp); |
1da177e4 LT |
1608 | } |
1609 | ||
1610 | /* | |
c7eee1b8 | 1611 | * split a bio - only worry about a bio with a single page in its iovec |
1da177e4 | 1612 | */ |
6feef531 | 1613 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
1da177e4 | 1614 | { |
6feef531 | 1615 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
1da177e4 LT |
1616 | |
1617 | if (!bp) | |
1618 | return bp; | |
1619 | ||
5f3ea37c | 1620 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
2056a782 JA |
1621 | bi->bi_sector + first_sectors); |
1622 | ||
5b83636a | 1623 | BUG_ON(bio_segments(bi) > 1); |
1da177e4 LT |
1624 | atomic_set(&bp->cnt, 3); |
1625 | bp->error = 0; | |
1626 | bp->bio1 = *bi; | |
1627 | bp->bio2 = *bi; | |
1628 | bp->bio2.bi_sector += first_sectors; | |
1629 | bp->bio2.bi_size -= first_sectors << 9; | |
1630 | bp->bio1.bi_size = first_sectors << 9; | |
1631 | ||
02f3939e | 1632 | if (bi->bi_vcnt != 0) { |
5b83636a KO |
1633 | bp->bv1 = *bio_iovec(bi); |
1634 | bp->bv2 = *bio_iovec(bi); | |
4363ac7c | 1635 | |
02f3939e SL |
1636 | if (bio_is_rw(bi)) { |
1637 | bp->bv2.bv_offset += first_sectors << 9; | |
1638 | bp->bv2.bv_len -= first_sectors << 9; | |
1639 | bp->bv1.bv_len = first_sectors << 9; | |
1640 | } | |
1da177e4 | 1641 | |
02f3939e SL |
1642 | bp->bio1.bi_io_vec = &bp->bv1; |
1643 | bp->bio2.bi_io_vec = &bp->bv2; | |
1da177e4 | 1644 | |
02f3939e SL |
1645 | bp->bio1.bi_max_vecs = 1; |
1646 | bp->bio2.bi_max_vecs = 1; | |
1647 | } | |
a2eb0c10 | 1648 | |
1da177e4 LT |
1649 | bp->bio1.bi_end_io = bio_pair_end_1; |
1650 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1651 | ||
1652 | bp->bio1.bi_private = bi; | |
6feef531 | 1653 | bp->bio2.bi_private = bio_split_pool; |
1da177e4 | 1654 | |
7ba1ba12 MP |
1655 | if (bio_integrity(bi)) |
1656 | bio_integrity_split(bi, bp, first_sectors); | |
1657 | ||
1da177e4 LT |
1658 | return bp; |
1659 | } | |
a112a71d | 1660 | EXPORT_SYMBOL(bio_split); |
1da177e4 | 1661 | |
ad3316bf MP |
1662 | /** |
1663 | * bio_sector_offset - Find hardware sector offset in bio | |
1664 | * @bio: bio to inspect | |
1665 | * @index: bio_vec index | |
1666 | * @offset: offset in bv_page | |
1667 | * | |
1668 | * Return the number of hardware sectors between beginning of bio | |
1669 | * and an end point indicated by a bio_vec index and an offset | |
1670 | * within that vector's page. | |
1671 | */ | |
1672 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, | |
1673 | unsigned int offset) | |
1674 | { | |
e1defc4f | 1675 | unsigned int sector_sz; |
ad3316bf MP |
1676 | struct bio_vec *bv; |
1677 | sector_t sectors; | |
1678 | int i; | |
1679 | ||
e1defc4f | 1680 | sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue); |
ad3316bf MP |
1681 | sectors = 0; |
1682 | ||
1683 | if (index >= bio->bi_idx) | |
1684 | index = bio->bi_vcnt - 1; | |
1685 | ||
1686 | __bio_for_each_segment(bv, bio, i, 0) { | |
1687 | if (i == index) { | |
1688 | if (offset > bv->bv_offset) | |
1689 | sectors += (offset - bv->bv_offset) / sector_sz; | |
1690 | break; | |
1691 | } | |
1692 | ||
1693 | sectors += bv->bv_len / sector_sz; | |
1694 | } | |
1695 | ||
1696 | return sectors; | |
1697 | } | |
1698 | EXPORT_SYMBOL(bio_sector_offset); | |
1da177e4 LT |
1699 | |
1700 | /* | |
1701 | * create memory pools for biovec's in a bio_set. | |
1702 | * use the global biovec slabs created for general use. | |
1703 | */ | |
9f060e22 | 1704 | mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries) |
1da177e4 | 1705 | { |
7ff9345f | 1706 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
1da177e4 | 1707 | |
9f060e22 | 1708 | return mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1709 | } |
1710 | ||
1711 | void bioset_free(struct bio_set *bs) | |
1712 | { | |
df2cb6da KO |
1713 | if (bs->rescue_workqueue) |
1714 | destroy_workqueue(bs->rescue_workqueue); | |
1715 | ||
1da177e4 LT |
1716 | if (bs->bio_pool) |
1717 | mempool_destroy(bs->bio_pool); | |
1718 | ||
9f060e22 KO |
1719 | if (bs->bvec_pool) |
1720 | mempool_destroy(bs->bvec_pool); | |
1721 | ||
7878cba9 | 1722 | bioset_integrity_free(bs); |
bb799ca0 | 1723 | bio_put_slab(bs); |
1da177e4 LT |
1724 | |
1725 | kfree(bs); | |
1726 | } | |
a112a71d | 1727 | EXPORT_SYMBOL(bioset_free); |
1da177e4 | 1728 | |
bb799ca0 JA |
1729 | /** |
1730 | * bioset_create - Create a bio_set | |
1731 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1732 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1733 | * | |
1734 | * Description: | |
1735 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1736 | * to ask for a number of bytes to be allocated in front of the bio. | |
1737 | * Front pad allocation is useful for embedding the bio inside | |
1738 | * another structure, to avoid allocating extra data to go with the bio. | |
1739 | * Note that the bio must be embedded at the END of that structure always, | |
1740 | * or things will break badly. | |
1741 | */ | |
1742 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) | |
1da177e4 | 1743 | { |
392ddc32 | 1744 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
1b434498 | 1745 | struct bio_set *bs; |
1da177e4 | 1746 | |
1b434498 | 1747 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1748 | if (!bs) |
1749 | return NULL; | |
1750 | ||
bb799ca0 | 1751 | bs->front_pad = front_pad; |
1b434498 | 1752 | |
df2cb6da KO |
1753 | spin_lock_init(&bs->rescue_lock); |
1754 | bio_list_init(&bs->rescue_list); | |
1755 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1756 | ||
392ddc32 | 1757 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
bb799ca0 JA |
1758 | if (!bs->bio_slab) { |
1759 | kfree(bs); | |
1760 | return NULL; | |
1761 | } | |
1762 | ||
1763 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1da177e4 LT |
1764 | if (!bs->bio_pool) |
1765 | goto bad; | |
1766 | ||
9f060e22 KO |
1767 | bs->bvec_pool = biovec_create_pool(bs, pool_size); |
1768 | if (!bs->bvec_pool) | |
df2cb6da KO |
1769 | goto bad; |
1770 | ||
1771 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1772 | if (!bs->rescue_workqueue) | |
1773 | goto bad; | |
1da177e4 | 1774 | |
df2cb6da | 1775 | return bs; |
1da177e4 LT |
1776 | bad: |
1777 | bioset_free(bs); | |
1778 | return NULL; | |
1779 | } | |
a112a71d | 1780 | EXPORT_SYMBOL(bioset_create); |
1da177e4 | 1781 | |
852c788f TH |
1782 | #ifdef CONFIG_BLK_CGROUP |
1783 | /** | |
1784 | * bio_associate_current - associate a bio with %current | |
1785 | * @bio: target bio | |
1786 | * | |
1787 | * Associate @bio with %current if it hasn't been associated yet. Block | |
1788 | * layer will treat @bio as if it were issued by %current no matter which | |
1789 | * task actually issues it. | |
1790 | * | |
1791 | * This function takes an extra reference of @task's io_context and blkcg | |
1792 | * which will be put when @bio is released. The caller must own @bio, | |
1793 | * ensure %current->io_context exists, and is responsible for synchronizing | |
1794 | * calls to this function. | |
1795 | */ | |
1796 | int bio_associate_current(struct bio *bio) | |
1797 | { | |
1798 | struct io_context *ioc; | |
1799 | struct cgroup_subsys_state *css; | |
1800 | ||
1801 | if (bio->bi_ioc) | |
1802 | return -EBUSY; | |
1803 | ||
1804 | ioc = current->io_context; | |
1805 | if (!ioc) | |
1806 | return -ENOENT; | |
1807 | ||
1808 | /* acquire active ref on @ioc and associate */ | |
1809 | get_io_context_active(ioc); | |
1810 | bio->bi_ioc = ioc; | |
1811 | ||
1812 | /* associate blkcg if exists */ | |
1813 | rcu_read_lock(); | |
1814 | css = task_subsys_state(current, blkio_subsys_id); | |
1815 | if (css && css_tryget(css)) | |
1816 | bio->bi_css = css; | |
1817 | rcu_read_unlock(); | |
1818 | ||
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | /** | |
1823 | * bio_disassociate_task - undo bio_associate_current() | |
1824 | * @bio: target bio | |
1825 | */ | |
1826 | void bio_disassociate_task(struct bio *bio) | |
1827 | { | |
1828 | if (bio->bi_ioc) { | |
1829 | put_io_context(bio->bi_ioc); | |
1830 | bio->bi_ioc = NULL; | |
1831 | } | |
1832 | if (bio->bi_css) { | |
1833 | css_put(bio->bi_css); | |
1834 | bio->bi_css = NULL; | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | #endif /* CONFIG_BLK_CGROUP */ | |
1839 | ||
1da177e4 LT |
1840 | static void __init biovec_init_slabs(void) |
1841 | { | |
1842 | int i; | |
1843 | ||
1844 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1845 | int size; | |
1846 | struct biovec_slab *bvs = bvec_slabs + i; | |
1847 | ||
a7fcd37c JA |
1848 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
1849 | bvs->slab = NULL; | |
1850 | continue; | |
1851 | } | |
a7fcd37c | 1852 | |
1da177e4 LT |
1853 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
1854 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1855 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1856 | } |
1857 | } | |
1858 | ||
1859 | static int __init init_bio(void) | |
1860 | { | |
bb799ca0 JA |
1861 | bio_slab_max = 2; |
1862 | bio_slab_nr = 0; | |
1863 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
1864 | if (!bio_slabs) | |
1865 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 1866 | |
7878cba9 | 1867 | bio_integrity_init(); |
1da177e4 LT |
1868 | biovec_init_slabs(); |
1869 | ||
bb799ca0 | 1870 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); |
1da177e4 LT |
1871 | if (!fs_bio_set) |
1872 | panic("bio: can't allocate bios\n"); | |
1873 | ||
a91a2785 MP |
1874 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) |
1875 | panic("bio: can't create integrity pool\n"); | |
1876 | ||
0eaae62a MD |
1877 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1878 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1879 | if (!bio_split_pool) |
1880 | panic("bio: can't create split pool\n"); | |
1881 | ||
1882 | return 0; | |
1883 | } | |
1da177e4 | 1884 | subsys_initcall(init_bio); |