]>
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 | |
9e882242 KO |
755 | struct submit_bio_ret { |
756 | struct completion event; | |
757 | int error; | |
758 | }; | |
759 | ||
760 | static void submit_bio_wait_endio(struct bio *bio, int error) | |
761 | { | |
762 | struct submit_bio_ret *ret = bio->bi_private; | |
763 | ||
764 | ret->error = error; | |
765 | complete(&ret->event); | |
766 | } | |
767 | ||
768 | /** | |
769 | * submit_bio_wait - submit a bio, and wait until it completes | |
770 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
771 | * @bio: The &struct bio which describes the I/O | |
772 | * | |
773 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
774 | * bio_endio() on failure. | |
775 | */ | |
776 | int submit_bio_wait(int rw, struct bio *bio) | |
777 | { | |
778 | struct submit_bio_ret ret; | |
779 | ||
780 | rw |= REQ_SYNC; | |
781 | init_completion(&ret.event); | |
782 | bio->bi_private = &ret; | |
783 | bio->bi_end_io = submit_bio_wait_endio; | |
784 | submit_bio(rw, bio); | |
785 | wait_for_completion(&ret.event); | |
786 | ||
787 | return ret.error; | |
788 | } | |
789 | EXPORT_SYMBOL(submit_bio_wait); | |
790 | ||
054bdf64 KO |
791 | /** |
792 | * bio_advance - increment/complete a bio by some number of bytes | |
793 | * @bio: bio to advance | |
794 | * @bytes: number of bytes to complete | |
795 | * | |
796 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
797 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
798 | * be updated on the last bvec as well. | |
799 | * | |
800 | * @bio will then represent the remaining, uncompleted portion of the io. | |
801 | */ | |
802 | void bio_advance(struct bio *bio, unsigned bytes) | |
803 | { | |
804 | if (bio_integrity(bio)) | |
805 | bio_integrity_advance(bio, bytes); | |
806 | ||
807 | bio->bi_sector += bytes >> 9; | |
808 | bio->bi_size -= bytes; | |
809 | ||
810 | if (bio->bi_rw & BIO_NO_ADVANCE_ITER_MASK) | |
811 | return; | |
812 | ||
813 | while (bytes) { | |
814 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
815 | WARN_ONCE(1, "bio idx %d >= vcnt %d\n", | |
816 | bio->bi_idx, bio->bi_vcnt); | |
817 | break; | |
818 | } | |
819 | ||
820 | if (bytes >= bio_iovec(bio)->bv_len) { | |
821 | bytes -= bio_iovec(bio)->bv_len; | |
822 | bio->bi_idx++; | |
823 | } else { | |
824 | bio_iovec(bio)->bv_len -= bytes; | |
825 | bio_iovec(bio)->bv_offset += bytes; | |
826 | bytes = 0; | |
827 | } | |
828 | } | |
829 | } | |
830 | EXPORT_SYMBOL(bio_advance); | |
831 | ||
16ac3d63 KO |
832 | /** |
833 | * bio_copy_data - copy contents of data buffers from one chain of bios to | |
834 | * another | |
835 | * @src: source bio list | |
836 | * @dst: destination bio list | |
837 | * | |
838 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats | |
839 | * @src and @dst as linked lists of bios. | |
840 | * | |
841 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
842 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
843 | */ | |
844 | void bio_copy_data(struct bio *dst, struct bio *src) | |
845 | { | |
846 | struct bio_vec *src_bv, *dst_bv; | |
847 | unsigned src_offset, dst_offset, bytes; | |
848 | void *src_p, *dst_p; | |
849 | ||
850 | src_bv = bio_iovec(src); | |
851 | dst_bv = bio_iovec(dst); | |
852 | ||
853 | src_offset = src_bv->bv_offset; | |
854 | dst_offset = dst_bv->bv_offset; | |
855 | ||
856 | while (1) { | |
857 | if (src_offset == src_bv->bv_offset + src_bv->bv_len) { | |
858 | src_bv++; | |
859 | if (src_bv == bio_iovec_idx(src, src->bi_vcnt)) { | |
860 | src = src->bi_next; | |
861 | if (!src) | |
862 | break; | |
863 | ||
864 | src_bv = bio_iovec(src); | |
865 | } | |
866 | ||
867 | src_offset = src_bv->bv_offset; | |
868 | } | |
869 | ||
870 | if (dst_offset == dst_bv->bv_offset + dst_bv->bv_len) { | |
871 | dst_bv++; | |
872 | if (dst_bv == bio_iovec_idx(dst, dst->bi_vcnt)) { | |
873 | dst = dst->bi_next; | |
874 | if (!dst) | |
875 | break; | |
876 | ||
877 | dst_bv = bio_iovec(dst); | |
878 | } | |
879 | ||
880 | dst_offset = dst_bv->bv_offset; | |
881 | } | |
882 | ||
883 | bytes = min(dst_bv->bv_offset + dst_bv->bv_len - dst_offset, | |
884 | src_bv->bv_offset + src_bv->bv_len - src_offset); | |
885 | ||
886 | src_p = kmap_atomic(src_bv->bv_page); | |
887 | dst_p = kmap_atomic(dst_bv->bv_page); | |
888 | ||
889 | memcpy(dst_p + dst_bv->bv_offset, | |
890 | src_p + src_bv->bv_offset, | |
891 | bytes); | |
892 | ||
893 | kunmap_atomic(dst_p); | |
894 | kunmap_atomic(src_p); | |
895 | ||
896 | src_offset += bytes; | |
897 | dst_offset += bytes; | |
898 | } | |
899 | } | |
900 | EXPORT_SYMBOL(bio_copy_data); | |
901 | ||
1da177e4 LT |
902 | struct bio_map_data { |
903 | struct bio_vec *iovecs; | |
c5dec1c3 | 904 | struct sg_iovec *sgvecs; |
152e283f FT |
905 | int nr_sgvecs; |
906 | int is_our_pages; | |
1da177e4 LT |
907 | }; |
908 | ||
c5dec1c3 | 909 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
152e283f FT |
910 | struct sg_iovec *iov, int iov_count, |
911 | int is_our_pages) | |
1da177e4 LT |
912 | { |
913 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
914 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
915 | bmd->nr_sgvecs = iov_count; | |
152e283f | 916 | bmd->is_our_pages = is_our_pages; |
1da177e4 LT |
917 | bio->bi_private = bmd; |
918 | } | |
919 | ||
920 | static void bio_free_map_data(struct bio_map_data *bmd) | |
921 | { | |
922 | kfree(bmd->iovecs); | |
c5dec1c3 | 923 | kfree(bmd->sgvecs); |
1da177e4 LT |
924 | kfree(bmd); |
925 | } | |
926 | ||
121f0994 DC |
927 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, |
928 | unsigned int iov_count, | |
76029ff3 | 929 | gfp_t gfp_mask) |
1da177e4 | 930 | { |
f3f63c1c JA |
931 | struct bio_map_data *bmd; |
932 | ||
933 | if (iov_count > UIO_MAXIOV) | |
934 | return NULL; | |
1da177e4 | 935 | |
f3f63c1c | 936 | bmd = kmalloc(sizeof(*bmd), gfp_mask); |
1da177e4 LT |
937 | if (!bmd) |
938 | return NULL; | |
939 | ||
76029ff3 | 940 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
c5dec1c3 FT |
941 | if (!bmd->iovecs) { |
942 | kfree(bmd); | |
943 | return NULL; | |
944 | } | |
945 | ||
76029ff3 | 946 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
c5dec1c3 | 947 | if (bmd->sgvecs) |
1da177e4 LT |
948 | return bmd; |
949 | ||
c5dec1c3 | 950 | kfree(bmd->iovecs); |
1da177e4 LT |
951 | kfree(bmd); |
952 | return NULL; | |
953 | } | |
954 | ||
aefcc28a | 955 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
ecb554a8 FT |
956 | struct sg_iovec *iov, int iov_count, |
957 | int to_user, int from_user, int do_free_page) | |
c5dec1c3 FT |
958 | { |
959 | int ret = 0, i; | |
960 | struct bio_vec *bvec; | |
961 | int iov_idx = 0; | |
962 | unsigned int iov_off = 0; | |
c5dec1c3 | 963 | |
d74c6d51 | 964 | bio_for_each_segment_all(bvec, bio, i) { |
c5dec1c3 | 965 | char *bv_addr = page_address(bvec->bv_page); |
aefcc28a | 966 | unsigned int bv_len = iovecs[i].bv_len; |
c5dec1c3 FT |
967 | |
968 | while (bv_len && iov_idx < iov_count) { | |
969 | unsigned int bytes; | |
0e0c6212 | 970 | char __user *iov_addr; |
c5dec1c3 FT |
971 | |
972 | bytes = min_t(unsigned int, | |
973 | iov[iov_idx].iov_len - iov_off, bv_len); | |
974 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
975 | ||
976 | if (!ret) { | |
ecb554a8 | 977 | if (to_user) |
c5dec1c3 FT |
978 | ret = copy_to_user(iov_addr, bv_addr, |
979 | bytes); | |
980 | ||
ecb554a8 FT |
981 | if (from_user) |
982 | ret = copy_from_user(bv_addr, iov_addr, | |
983 | bytes); | |
984 | ||
c5dec1c3 FT |
985 | if (ret) |
986 | ret = -EFAULT; | |
987 | } | |
988 | ||
989 | bv_len -= bytes; | |
990 | bv_addr += bytes; | |
991 | iov_addr += bytes; | |
992 | iov_off += bytes; | |
993 | ||
994 | if (iov[iov_idx].iov_len == iov_off) { | |
995 | iov_idx++; | |
996 | iov_off = 0; | |
997 | } | |
998 | } | |
999 | ||
152e283f | 1000 | if (do_free_page) |
c5dec1c3 FT |
1001 | __free_page(bvec->bv_page); |
1002 | } | |
1003 | ||
1004 | return ret; | |
1005 | } | |
1006 | ||
1da177e4 LT |
1007 | /** |
1008 | * bio_uncopy_user - finish previously mapped bio | |
1009 | * @bio: bio being terminated | |
1010 | * | |
1011 | * Free pages allocated from bio_copy_user() and write back data | |
1012 | * to user space in case of a read. | |
1013 | */ | |
1014 | int bio_uncopy_user(struct bio *bio) | |
1015 | { | |
1016 | struct bio_map_data *bmd = bio->bi_private; | |
81882766 | 1017 | int ret = 0; |
1da177e4 | 1018 | |
81882766 FT |
1019 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
1020 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, | |
ecb554a8 FT |
1021 | bmd->nr_sgvecs, bio_data_dir(bio) == READ, |
1022 | 0, bmd->is_our_pages); | |
1da177e4 LT |
1023 | bio_free_map_data(bmd); |
1024 | bio_put(bio); | |
1025 | return ret; | |
1026 | } | |
a112a71d | 1027 | EXPORT_SYMBOL(bio_uncopy_user); |
1da177e4 LT |
1028 | |
1029 | /** | |
c5dec1c3 | 1030 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 1031 | * @q: destination block queue |
152e283f | 1032 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
1033 | * @iov: the iovec. |
1034 | * @iov_count: number of elements in the iovec | |
1da177e4 | 1035 | * @write_to_vm: bool indicating writing to pages or not |
a3bce90e | 1036 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
1037 | * |
1038 | * Prepares and returns a bio for indirect user io, bouncing data | |
1039 | * to/from kernel pages as necessary. Must be paired with | |
1040 | * call bio_uncopy_user() on io completion. | |
1041 | */ | |
152e283f FT |
1042 | struct bio *bio_copy_user_iov(struct request_queue *q, |
1043 | struct rq_map_data *map_data, | |
1044 | struct sg_iovec *iov, int iov_count, | |
1045 | int write_to_vm, gfp_t gfp_mask) | |
1da177e4 | 1046 | { |
1da177e4 LT |
1047 | struct bio_map_data *bmd; |
1048 | struct bio_vec *bvec; | |
1049 | struct page *page; | |
1050 | struct bio *bio; | |
1051 | int i, ret; | |
c5dec1c3 FT |
1052 | int nr_pages = 0; |
1053 | unsigned int len = 0; | |
56c451f4 | 1054 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; |
1da177e4 | 1055 | |
c5dec1c3 FT |
1056 | for (i = 0; i < iov_count; i++) { |
1057 | unsigned long uaddr; | |
1058 | unsigned long end; | |
1059 | unsigned long start; | |
1060 | ||
1061 | uaddr = (unsigned long)iov[i].iov_base; | |
1062 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1063 | start = uaddr >> PAGE_SHIFT; | |
1064 | ||
cb4644ca JA |
1065 | /* |
1066 | * Overflow, abort | |
1067 | */ | |
1068 | if (end < start) | |
1069 | return ERR_PTR(-EINVAL); | |
1070 | ||
c5dec1c3 FT |
1071 | nr_pages += end - start; |
1072 | len += iov[i].iov_len; | |
1073 | } | |
1074 | ||
69838727 FT |
1075 | if (offset) |
1076 | nr_pages++; | |
1077 | ||
a3bce90e | 1078 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
1da177e4 LT |
1079 | if (!bmd) |
1080 | return ERR_PTR(-ENOMEM); | |
1081 | ||
1da177e4 | 1082 | ret = -ENOMEM; |
a9e9dc24 | 1083 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1084 | if (!bio) |
1085 | goto out_bmd; | |
1086 | ||
7b6d91da CH |
1087 | if (!write_to_vm) |
1088 | bio->bi_rw |= REQ_WRITE; | |
1da177e4 LT |
1089 | |
1090 | ret = 0; | |
56c451f4 FT |
1091 | |
1092 | if (map_data) { | |
e623ddb4 | 1093 | nr_pages = 1 << map_data->page_order; |
56c451f4 FT |
1094 | i = map_data->offset / PAGE_SIZE; |
1095 | } | |
1da177e4 | 1096 | while (len) { |
e623ddb4 | 1097 | unsigned int bytes = PAGE_SIZE; |
1da177e4 | 1098 | |
56c451f4 FT |
1099 | bytes -= offset; |
1100 | ||
1da177e4 LT |
1101 | if (bytes > len) |
1102 | bytes = len; | |
1103 | ||
152e283f | 1104 | if (map_data) { |
e623ddb4 | 1105 | if (i == map_data->nr_entries * nr_pages) { |
152e283f FT |
1106 | ret = -ENOMEM; |
1107 | break; | |
1108 | } | |
e623ddb4 FT |
1109 | |
1110 | page = map_data->pages[i / nr_pages]; | |
1111 | page += (i % nr_pages); | |
1112 | ||
1113 | i++; | |
1114 | } else { | |
152e283f | 1115 | page = alloc_page(q->bounce_gfp | gfp_mask); |
e623ddb4 FT |
1116 | if (!page) { |
1117 | ret = -ENOMEM; | |
1118 | break; | |
1119 | } | |
1da177e4 LT |
1120 | } |
1121 | ||
56c451f4 | 1122 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) |
1da177e4 | 1123 | break; |
1da177e4 LT |
1124 | |
1125 | len -= bytes; | |
56c451f4 | 1126 | offset = 0; |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | if (ret) | |
1130 | goto cleanup; | |
1131 | ||
1132 | /* | |
1133 | * success | |
1134 | */ | |
ecb554a8 FT |
1135 | if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || |
1136 | (map_data && map_data->from_user)) { | |
1137 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0); | |
c5dec1c3 FT |
1138 | if (ret) |
1139 | goto cleanup; | |
1da177e4 LT |
1140 | } |
1141 | ||
152e283f | 1142 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
1da177e4 LT |
1143 | return bio; |
1144 | cleanup: | |
152e283f | 1145 | if (!map_data) |
d74c6d51 | 1146 | bio_for_each_segment_all(bvec, bio, i) |
152e283f | 1147 | __free_page(bvec->bv_page); |
1da177e4 LT |
1148 | |
1149 | bio_put(bio); | |
1150 | out_bmd: | |
1151 | bio_free_map_data(bmd); | |
1152 | return ERR_PTR(ret); | |
1153 | } | |
1154 | ||
c5dec1c3 FT |
1155 | /** |
1156 | * bio_copy_user - copy user data to bio | |
1157 | * @q: destination block queue | |
152e283f | 1158 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
c5dec1c3 FT |
1159 | * @uaddr: start of user address |
1160 | * @len: length in bytes | |
1161 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 1162 | * @gfp_mask: memory allocation flags |
c5dec1c3 FT |
1163 | * |
1164 | * Prepares and returns a bio for indirect user io, bouncing data | |
1165 | * to/from kernel pages as necessary. Must be paired with | |
1166 | * call bio_uncopy_user() on io completion. | |
1167 | */ | |
152e283f FT |
1168 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
1169 | unsigned long uaddr, unsigned int len, | |
1170 | int write_to_vm, gfp_t gfp_mask) | |
c5dec1c3 FT |
1171 | { |
1172 | struct sg_iovec iov; | |
1173 | ||
1174 | iov.iov_base = (void __user *)uaddr; | |
1175 | iov.iov_len = len; | |
1176 | ||
152e283f | 1177 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
c5dec1c3 | 1178 | } |
a112a71d | 1179 | EXPORT_SYMBOL(bio_copy_user); |
c5dec1c3 | 1180 | |
165125e1 | 1181 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
1182 | struct block_device *bdev, |
1183 | struct sg_iovec *iov, int iov_count, | |
a3bce90e | 1184 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 | 1185 | { |
f1970baf JB |
1186 | int i, j; |
1187 | int nr_pages = 0; | |
1da177e4 LT |
1188 | struct page **pages; |
1189 | struct bio *bio; | |
f1970baf JB |
1190 | int cur_page = 0; |
1191 | int ret, offset; | |
1da177e4 | 1192 | |
f1970baf JB |
1193 | for (i = 0; i < iov_count; i++) { |
1194 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1195 | unsigned long len = iov[i].iov_len; | |
1196 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1197 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1198 | ||
cb4644ca JA |
1199 | /* |
1200 | * Overflow, abort | |
1201 | */ | |
1202 | if (end < start) | |
1203 | return ERR_PTR(-EINVAL); | |
1204 | ||
f1970baf JB |
1205 | nr_pages += end - start; |
1206 | /* | |
ad2d7225 | 1207 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 1208 | */ |
ad2d7225 | 1209 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
1210 | return ERR_PTR(-EINVAL); |
1211 | } | |
1212 | ||
1213 | if (!nr_pages) | |
1da177e4 LT |
1214 | return ERR_PTR(-EINVAL); |
1215 | ||
a9e9dc24 | 1216 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1217 | if (!bio) |
1218 | return ERR_PTR(-ENOMEM); | |
1219 | ||
1220 | ret = -ENOMEM; | |
a3bce90e | 1221 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
1da177e4 LT |
1222 | if (!pages) |
1223 | goto out; | |
1224 | ||
f1970baf JB |
1225 | for (i = 0; i < iov_count; i++) { |
1226 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1227 | unsigned long len = iov[i].iov_len; | |
1228 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1229 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1230 | const int local_nr_pages = end - start; | |
1231 | const int page_limit = cur_page + local_nr_pages; | |
cb4644ca | 1232 | |
f5dd33c4 NP |
1233 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
1234 | write_to_vm, &pages[cur_page]); | |
99172157 JA |
1235 | if (ret < local_nr_pages) { |
1236 | ret = -EFAULT; | |
f1970baf | 1237 | goto out_unmap; |
99172157 | 1238 | } |
f1970baf JB |
1239 | |
1240 | offset = uaddr & ~PAGE_MASK; | |
1241 | for (j = cur_page; j < page_limit; j++) { | |
1242 | unsigned int bytes = PAGE_SIZE - offset; | |
1243 | ||
1244 | if (len <= 0) | |
1245 | break; | |
1246 | ||
1247 | if (bytes > len) | |
1248 | bytes = len; | |
1249 | ||
1250 | /* | |
1251 | * sorry... | |
1252 | */ | |
defd94b7 MC |
1253 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
1254 | bytes) | |
f1970baf JB |
1255 | break; |
1256 | ||
1257 | len -= bytes; | |
1258 | offset = 0; | |
1259 | } | |
1da177e4 | 1260 | |
f1970baf | 1261 | cur_page = j; |
1da177e4 | 1262 | /* |
f1970baf | 1263 | * release the pages we didn't map into the bio, if any |
1da177e4 | 1264 | */ |
f1970baf JB |
1265 | while (j < page_limit) |
1266 | page_cache_release(pages[j++]); | |
1da177e4 LT |
1267 | } |
1268 | ||
1da177e4 LT |
1269 | kfree(pages); |
1270 | ||
1271 | /* | |
1272 | * set data direction, and check if mapped pages need bouncing | |
1273 | */ | |
1274 | if (!write_to_vm) | |
7b6d91da | 1275 | bio->bi_rw |= REQ_WRITE; |
1da177e4 | 1276 | |
f1970baf | 1277 | bio->bi_bdev = bdev; |
1da177e4 LT |
1278 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
1279 | return bio; | |
f1970baf JB |
1280 | |
1281 | out_unmap: | |
1282 | for (i = 0; i < nr_pages; i++) { | |
1283 | if(!pages[i]) | |
1284 | break; | |
1285 | page_cache_release(pages[i]); | |
1286 | } | |
1287 | out: | |
1da177e4 LT |
1288 | kfree(pages); |
1289 | bio_put(bio); | |
1290 | return ERR_PTR(ret); | |
1291 | } | |
1292 | ||
1293 | /** | |
1294 | * bio_map_user - map user address into bio | |
165125e1 | 1295 | * @q: the struct request_queue for the bio |
1da177e4 LT |
1296 | * @bdev: destination block device |
1297 | * @uaddr: start of user address | |
1298 | * @len: length in bytes | |
1299 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 1300 | * @gfp_mask: memory allocation flags |
1da177e4 LT |
1301 | * |
1302 | * Map the user space address into a bio suitable for io to a block | |
1303 | * device. Returns an error pointer in case of error. | |
1304 | */ | |
165125e1 | 1305 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
a3bce90e FT |
1306 | unsigned long uaddr, unsigned int len, int write_to_vm, |
1307 | gfp_t gfp_mask) | |
f1970baf JB |
1308 | { |
1309 | struct sg_iovec iov; | |
1310 | ||
3f70353e | 1311 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
1312 | iov.iov_len = len; |
1313 | ||
a3bce90e | 1314 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
f1970baf | 1315 | } |
a112a71d | 1316 | EXPORT_SYMBOL(bio_map_user); |
f1970baf JB |
1317 | |
1318 | /** | |
1319 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 1320 | * @q: the struct request_queue for the bio |
f1970baf JB |
1321 | * @bdev: destination block device |
1322 | * @iov: the iovec. | |
1323 | * @iov_count: number of elements in the iovec | |
1324 | * @write_to_vm: bool indicating writing to pages or not | |
a3bce90e | 1325 | * @gfp_mask: memory allocation flags |
f1970baf JB |
1326 | * |
1327 | * Map the user space address into a bio suitable for io to a block | |
1328 | * device. Returns an error pointer in case of error. | |
1329 | */ | |
165125e1 | 1330 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf | 1331 | struct sg_iovec *iov, int iov_count, |
a3bce90e | 1332 | int write_to_vm, gfp_t gfp_mask) |
1da177e4 LT |
1333 | { |
1334 | struct bio *bio; | |
1335 | ||
a3bce90e FT |
1336 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
1337 | gfp_mask); | |
1da177e4 LT |
1338 | if (IS_ERR(bio)) |
1339 | return bio; | |
1340 | ||
1341 | /* | |
1342 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
1343 | * it would normally disappear when its bi_end_io is run. | |
1344 | * however, we need it for the unmap, so grab an extra | |
1345 | * reference to it | |
1346 | */ | |
1347 | bio_get(bio); | |
1348 | ||
0e75f906 | 1349 | return bio; |
1da177e4 LT |
1350 | } |
1351 | ||
1352 | static void __bio_unmap_user(struct bio *bio) | |
1353 | { | |
1354 | struct bio_vec *bvec; | |
1355 | int i; | |
1356 | ||
1357 | /* | |
1358 | * make sure we dirty pages we wrote to | |
1359 | */ | |
d74c6d51 | 1360 | bio_for_each_segment_all(bvec, bio, i) { |
1da177e4 LT |
1361 | if (bio_data_dir(bio) == READ) |
1362 | set_page_dirty_lock(bvec->bv_page); | |
1363 | ||
1364 | page_cache_release(bvec->bv_page); | |
1365 | } | |
1366 | ||
1367 | bio_put(bio); | |
1368 | } | |
1369 | ||
1370 | /** | |
1371 | * bio_unmap_user - unmap a bio | |
1372 | * @bio: the bio being unmapped | |
1373 | * | |
1374 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
1375 | * a process context. | |
1376 | * | |
1377 | * bio_unmap_user() may sleep. | |
1378 | */ | |
1379 | void bio_unmap_user(struct bio *bio) | |
1380 | { | |
1381 | __bio_unmap_user(bio); | |
1382 | bio_put(bio); | |
1383 | } | |
a112a71d | 1384 | EXPORT_SYMBOL(bio_unmap_user); |
1da177e4 | 1385 | |
6712ecf8 | 1386 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 1387 | { |
b823825e | 1388 | bio_put(bio); |
b823825e JA |
1389 | } |
1390 | ||
165125e1 | 1391 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 1392 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
1393 | { |
1394 | unsigned long kaddr = (unsigned long)data; | |
1395 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1396 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1397 | const int nr_pages = end - start; | |
1398 | int offset, i; | |
1399 | struct bio *bio; | |
1400 | ||
a9e9dc24 | 1401 | bio = bio_kmalloc(gfp_mask, nr_pages); |
df46b9a4 MC |
1402 | if (!bio) |
1403 | return ERR_PTR(-ENOMEM); | |
1404 | ||
1405 | offset = offset_in_page(kaddr); | |
1406 | for (i = 0; i < nr_pages; i++) { | |
1407 | unsigned int bytes = PAGE_SIZE - offset; | |
1408 | ||
1409 | if (len <= 0) | |
1410 | break; | |
1411 | ||
1412 | if (bytes > len) | |
1413 | bytes = len; | |
1414 | ||
defd94b7 MC |
1415 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
1416 | offset) < bytes) | |
df46b9a4 MC |
1417 | break; |
1418 | ||
1419 | data += bytes; | |
1420 | len -= bytes; | |
1421 | offset = 0; | |
1422 | } | |
1423 | ||
b823825e | 1424 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1425 | return bio; |
1426 | } | |
1427 | ||
1428 | /** | |
1429 | * bio_map_kern - map kernel address into bio | |
165125e1 | 1430 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
1431 | * @data: pointer to buffer to map |
1432 | * @len: length in bytes | |
1433 | * @gfp_mask: allocation flags for bio allocation | |
1434 | * | |
1435 | * Map the kernel address into a bio suitable for io to a block | |
1436 | * device. Returns an error pointer in case of error. | |
1437 | */ | |
165125e1 | 1438 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 1439 | gfp_t gfp_mask) |
df46b9a4 MC |
1440 | { |
1441 | struct bio *bio; | |
1442 | ||
1443 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
1444 | if (IS_ERR(bio)) | |
1445 | return bio; | |
1446 | ||
1447 | if (bio->bi_size == len) | |
1448 | return bio; | |
1449 | ||
1450 | /* | |
1451 | * Don't support partial mappings. | |
1452 | */ | |
1453 | bio_put(bio); | |
1454 | return ERR_PTR(-EINVAL); | |
1455 | } | |
a112a71d | 1456 | EXPORT_SYMBOL(bio_map_kern); |
df46b9a4 | 1457 | |
68154e90 FT |
1458 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1459 | { | |
1460 | struct bio_vec *bvec; | |
1461 | const int read = bio_data_dir(bio) == READ; | |
76029ff3 | 1462 | struct bio_map_data *bmd = bio->bi_private; |
68154e90 | 1463 | int i; |
76029ff3 | 1464 | char *p = bmd->sgvecs[0].iov_base; |
68154e90 | 1465 | |
d74c6d51 | 1466 | bio_for_each_segment_all(bvec, bio, i) { |
68154e90 | 1467 | char *addr = page_address(bvec->bv_page); |
76029ff3 | 1468 | int len = bmd->iovecs[i].bv_len; |
68154e90 | 1469 | |
4fc981ef | 1470 | if (read) |
76029ff3 | 1471 | memcpy(p, addr, len); |
68154e90 FT |
1472 | |
1473 | __free_page(bvec->bv_page); | |
76029ff3 | 1474 | p += len; |
68154e90 FT |
1475 | } |
1476 | ||
76029ff3 | 1477 | bio_free_map_data(bmd); |
68154e90 FT |
1478 | bio_put(bio); |
1479 | } | |
1480 | ||
1481 | /** | |
1482 | * bio_copy_kern - copy kernel address into bio | |
1483 | * @q: the struct request_queue for the bio | |
1484 | * @data: pointer to buffer to copy | |
1485 | * @len: length in bytes | |
1486 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1487 | * @reading: data direction is READ |
68154e90 FT |
1488 | * |
1489 | * copy the kernel address into a bio suitable for io to a block | |
1490 | * device. Returns an error pointer in case of error. | |
1491 | */ | |
1492 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1493 | gfp_t gfp_mask, int reading) | |
1494 | { | |
68154e90 FT |
1495 | struct bio *bio; |
1496 | struct bio_vec *bvec; | |
4d8ab62e | 1497 | int i; |
68154e90 | 1498 | |
4d8ab62e FT |
1499 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
1500 | if (IS_ERR(bio)) | |
1501 | return bio; | |
68154e90 FT |
1502 | |
1503 | if (!reading) { | |
1504 | void *p = data; | |
1505 | ||
d74c6d51 | 1506 | bio_for_each_segment_all(bvec, bio, i) { |
68154e90 FT |
1507 | char *addr = page_address(bvec->bv_page); |
1508 | ||
1509 | memcpy(addr, p, bvec->bv_len); | |
1510 | p += bvec->bv_len; | |
1511 | } | |
1512 | } | |
1513 | ||
68154e90 | 1514 | bio->bi_end_io = bio_copy_kern_endio; |
76029ff3 | 1515 | |
68154e90 | 1516 | return bio; |
68154e90 | 1517 | } |
a112a71d | 1518 | EXPORT_SYMBOL(bio_copy_kern); |
68154e90 | 1519 | |
1da177e4 LT |
1520 | /* |
1521 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1522 | * for performing direct-IO in BIOs. | |
1523 | * | |
1524 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1525 | * because the required locks are not interrupt-safe. So what we can do is to | |
1526 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1527 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1528 | * in process context. | |
1529 | * | |
1530 | * We special-case compound pages here: normally this means reads into hugetlb | |
1531 | * pages. The logic in here doesn't really work right for compound pages | |
1532 | * because the VM does not uniformly chase down the head page in all cases. | |
1533 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1534 | * handle them at all. So we skip compound pages here at an early stage. | |
1535 | * | |
1536 | * Note that this code is very hard to test under normal circumstances because | |
1537 | * direct-io pins the pages with get_user_pages(). This makes | |
1538 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1539 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1540 | * pagecache. |
1541 | * | |
1542 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1543 | * deferred bio dirtying paths. | |
1544 | */ | |
1545 | ||
1546 | /* | |
1547 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1548 | */ | |
1549 | void bio_set_pages_dirty(struct bio *bio) | |
1550 | { | |
cb34e057 | 1551 | struct bio_vec *bvec; |
1da177e4 LT |
1552 | int i; |
1553 | ||
cb34e057 KO |
1554 | bio_for_each_segment_all(bvec, bio, i) { |
1555 | struct page *page = bvec->bv_page; | |
1da177e4 LT |
1556 | |
1557 | if (page && !PageCompound(page)) | |
1558 | set_page_dirty_lock(page); | |
1559 | } | |
1560 | } | |
1561 | ||
86b6c7a7 | 1562 | static void bio_release_pages(struct bio *bio) |
1da177e4 | 1563 | { |
cb34e057 | 1564 | struct bio_vec *bvec; |
1da177e4 LT |
1565 | int i; |
1566 | ||
cb34e057 KO |
1567 | bio_for_each_segment_all(bvec, bio, i) { |
1568 | struct page *page = bvec->bv_page; | |
1da177e4 LT |
1569 | |
1570 | if (page) | |
1571 | put_page(page); | |
1572 | } | |
1573 | } | |
1574 | ||
1575 | /* | |
1576 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1577 | * If they are, then fine. If, however, some pages are clean then they must | |
1578 | * have been written out during the direct-IO read. So we take another ref on | |
1579 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1580 | * | |
1581 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1582 | * here on. It will run one page_cache_release() against each page and will | |
1583 | * run one bio_put() against the BIO. | |
1584 | */ | |
1585 | ||
65f27f38 | 1586 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1587 | |
65f27f38 | 1588 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1589 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1590 | static struct bio *bio_dirty_list; | |
1591 | ||
1592 | /* | |
1593 | * This runs in process context | |
1594 | */ | |
65f27f38 | 1595 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1596 | { |
1597 | unsigned long flags; | |
1598 | struct bio *bio; | |
1599 | ||
1600 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1601 | bio = bio_dirty_list; | |
1602 | bio_dirty_list = NULL; | |
1603 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1604 | ||
1605 | while (bio) { | |
1606 | struct bio *next = bio->bi_private; | |
1607 | ||
1608 | bio_set_pages_dirty(bio); | |
1609 | bio_release_pages(bio); | |
1610 | bio_put(bio); | |
1611 | bio = next; | |
1612 | } | |
1613 | } | |
1614 | ||
1615 | void bio_check_pages_dirty(struct bio *bio) | |
1616 | { | |
cb34e057 | 1617 | struct bio_vec *bvec; |
1da177e4 LT |
1618 | int nr_clean_pages = 0; |
1619 | int i; | |
1620 | ||
cb34e057 KO |
1621 | bio_for_each_segment_all(bvec, bio, i) { |
1622 | struct page *page = bvec->bv_page; | |
1da177e4 LT |
1623 | |
1624 | if (PageDirty(page) || PageCompound(page)) { | |
1625 | page_cache_release(page); | |
cb34e057 | 1626 | bvec->bv_page = NULL; |
1da177e4 LT |
1627 | } else { |
1628 | nr_clean_pages++; | |
1629 | } | |
1630 | } | |
1631 | ||
1632 | if (nr_clean_pages) { | |
1633 | unsigned long flags; | |
1634 | ||
1635 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1636 | bio->bi_private = bio_dirty_list; | |
1637 | bio_dirty_list = bio; | |
1638 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1639 | schedule_work(&bio_dirty_work); | |
1640 | } else { | |
1641 | bio_put(bio); | |
1642 | } | |
1643 | } | |
1644 | ||
2d4dc890 IL |
1645 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1646 | void bio_flush_dcache_pages(struct bio *bi) | |
1647 | { | |
1648 | int i; | |
1649 | struct bio_vec *bvec; | |
1650 | ||
1651 | bio_for_each_segment(bvec, bi, i) | |
1652 | flush_dcache_page(bvec->bv_page); | |
1653 | } | |
1654 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1655 | #endif | |
1656 | ||
1da177e4 LT |
1657 | /** |
1658 | * bio_endio - end I/O on a bio | |
1659 | * @bio: bio | |
1da177e4 LT |
1660 | * @error: error, if any |
1661 | * | |
1662 | * Description: | |
6712ecf8 | 1663 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1664 | * preferred way to end I/O on a bio, it takes care of clearing |
1665 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1666 | * established -Exxxx (-EIO, for instance) error values in case | |
25985edc | 1667 | * something went wrong. No one should call bi_end_io() directly on a |
5bb23a68 N |
1668 | * bio unless they own it and thus know that it has an end_io |
1669 | * function. | |
1da177e4 | 1670 | **/ |
6712ecf8 | 1671 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1672 | { |
1673 | if (error) | |
1674 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1675 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1676 | error = -EIO; | |
1da177e4 | 1677 | |
3a366e61 TH |
1678 | trace_block_bio_complete(bio, error); |
1679 | ||
5bb23a68 | 1680 | if (bio->bi_end_io) |
6712ecf8 | 1681 | bio->bi_end_io(bio, error); |
1da177e4 | 1682 | } |
a112a71d | 1683 | EXPORT_SYMBOL(bio_endio); |
1da177e4 LT |
1684 | |
1685 | void bio_pair_release(struct bio_pair *bp) | |
1686 | { | |
1687 | if (atomic_dec_and_test(&bp->cnt)) { | |
1688 | struct bio *master = bp->bio1.bi_private; | |
1689 | ||
6712ecf8 | 1690 | bio_endio(master, bp->error); |
1da177e4 LT |
1691 | mempool_free(bp, bp->bio2.bi_private); |
1692 | } | |
1693 | } | |
a112a71d | 1694 | EXPORT_SYMBOL(bio_pair_release); |
1da177e4 | 1695 | |
6712ecf8 | 1696 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1697 | { |
1698 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1699 | ||
1700 | if (err) | |
1701 | bp->error = err; | |
1702 | ||
1da177e4 | 1703 | bio_pair_release(bp); |
1da177e4 LT |
1704 | } |
1705 | ||
6712ecf8 | 1706 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1707 | { |
1708 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1709 | ||
1710 | if (err) | |
1711 | bp->error = err; | |
1712 | ||
1da177e4 | 1713 | bio_pair_release(bp); |
1da177e4 LT |
1714 | } |
1715 | ||
1716 | /* | |
c7eee1b8 | 1717 | * split a bio - only worry about a bio with a single page in its iovec |
1da177e4 | 1718 | */ |
6feef531 | 1719 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
1da177e4 | 1720 | { |
6feef531 | 1721 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
1da177e4 LT |
1722 | |
1723 | if (!bp) | |
1724 | return bp; | |
1725 | ||
5f3ea37c | 1726 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
2056a782 JA |
1727 | bi->bi_sector + first_sectors); |
1728 | ||
5b83636a | 1729 | BUG_ON(bio_segments(bi) > 1); |
1da177e4 LT |
1730 | atomic_set(&bp->cnt, 3); |
1731 | bp->error = 0; | |
1732 | bp->bio1 = *bi; | |
1733 | bp->bio2 = *bi; | |
1734 | bp->bio2.bi_sector += first_sectors; | |
1735 | bp->bio2.bi_size -= first_sectors << 9; | |
1736 | bp->bio1.bi_size = first_sectors << 9; | |
1737 | ||
02f3939e | 1738 | if (bi->bi_vcnt != 0) { |
5b83636a KO |
1739 | bp->bv1 = *bio_iovec(bi); |
1740 | bp->bv2 = *bio_iovec(bi); | |
4363ac7c | 1741 | |
02f3939e SL |
1742 | if (bio_is_rw(bi)) { |
1743 | bp->bv2.bv_offset += first_sectors << 9; | |
1744 | bp->bv2.bv_len -= first_sectors << 9; | |
1745 | bp->bv1.bv_len = first_sectors << 9; | |
1746 | } | |
1da177e4 | 1747 | |
02f3939e SL |
1748 | bp->bio1.bi_io_vec = &bp->bv1; |
1749 | bp->bio2.bi_io_vec = &bp->bv2; | |
1da177e4 | 1750 | |
02f3939e SL |
1751 | bp->bio1.bi_max_vecs = 1; |
1752 | bp->bio2.bi_max_vecs = 1; | |
1753 | } | |
a2eb0c10 | 1754 | |
1da177e4 LT |
1755 | bp->bio1.bi_end_io = bio_pair_end_1; |
1756 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1757 | ||
1758 | bp->bio1.bi_private = bi; | |
6feef531 | 1759 | bp->bio2.bi_private = bio_split_pool; |
1da177e4 | 1760 | |
7ba1ba12 MP |
1761 | if (bio_integrity(bi)) |
1762 | bio_integrity_split(bi, bp, first_sectors); | |
1763 | ||
1da177e4 LT |
1764 | return bp; |
1765 | } | |
a112a71d | 1766 | EXPORT_SYMBOL(bio_split); |
1da177e4 | 1767 | |
ad3316bf MP |
1768 | /** |
1769 | * bio_sector_offset - Find hardware sector offset in bio | |
1770 | * @bio: bio to inspect | |
1771 | * @index: bio_vec index | |
1772 | * @offset: offset in bv_page | |
1773 | * | |
1774 | * Return the number of hardware sectors between beginning of bio | |
1775 | * and an end point indicated by a bio_vec index and an offset | |
1776 | * within that vector's page. | |
1777 | */ | |
1778 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, | |
1779 | unsigned int offset) | |
1780 | { | |
e1defc4f | 1781 | unsigned int sector_sz; |
ad3316bf MP |
1782 | struct bio_vec *bv; |
1783 | sector_t sectors; | |
1784 | int i; | |
1785 | ||
e1defc4f | 1786 | sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue); |
ad3316bf MP |
1787 | sectors = 0; |
1788 | ||
1789 | if (index >= bio->bi_idx) | |
1790 | index = bio->bi_vcnt - 1; | |
1791 | ||
d74c6d51 | 1792 | bio_for_each_segment_all(bv, bio, i) { |
ad3316bf MP |
1793 | if (i == index) { |
1794 | if (offset > bv->bv_offset) | |
1795 | sectors += (offset - bv->bv_offset) / sector_sz; | |
1796 | break; | |
1797 | } | |
1798 | ||
1799 | sectors += bv->bv_len / sector_sz; | |
1800 | } | |
1801 | ||
1802 | return sectors; | |
1803 | } | |
1804 | EXPORT_SYMBOL(bio_sector_offset); | |
1da177e4 LT |
1805 | |
1806 | /* | |
1807 | * create memory pools for biovec's in a bio_set. | |
1808 | * use the global biovec slabs created for general use. | |
1809 | */ | |
9f060e22 | 1810 | mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries) |
1da177e4 | 1811 | { |
7ff9345f | 1812 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
1da177e4 | 1813 | |
9f060e22 | 1814 | return mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1815 | } |
1816 | ||
1817 | void bioset_free(struct bio_set *bs) | |
1818 | { | |
df2cb6da KO |
1819 | if (bs->rescue_workqueue) |
1820 | destroy_workqueue(bs->rescue_workqueue); | |
1821 | ||
1da177e4 LT |
1822 | if (bs->bio_pool) |
1823 | mempool_destroy(bs->bio_pool); | |
1824 | ||
9f060e22 KO |
1825 | if (bs->bvec_pool) |
1826 | mempool_destroy(bs->bvec_pool); | |
1827 | ||
7878cba9 | 1828 | bioset_integrity_free(bs); |
bb799ca0 | 1829 | bio_put_slab(bs); |
1da177e4 LT |
1830 | |
1831 | kfree(bs); | |
1832 | } | |
a112a71d | 1833 | EXPORT_SYMBOL(bioset_free); |
1da177e4 | 1834 | |
bb799ca0 JA |
1835 | /** |
1836 | * bioset_create - Create a bio_set | |
1837 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1838 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1839 | * | |
1840 | * Description: | |
1841 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1842 | * to ask for a number of bytes to be allocated in front of the bio. | |
1843 | * Front pad allocation is useful for embedding the bio inside | |
1844 | * another structure, to avoid allocating extra data to go with the bio. | |
1845 | * Note that the bio must be embedded at the END of that structure always, | |
1846 | * or things will break badly. | |
1847 | */ | |
1848 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) | |
1da177e4 | 1849 | { |
392ddc32 | 1850 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
1b434498 | 1851 | struct bio_set *bs; |
1da177e4 | 1852 | |
1b434498 | 1853 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1854 | if (!bs) |
1855 | return NULL; | |
1856 | ||
bb799ca0 | 1857 | bs->front_pad = front_pad; |
1b434498 | 1858 | |
df2cb6da KO |
1859 | spin_lock_init(&bs->rescue_lock); |
1860 | bio_list_init(&bs->rescue_list); | |
1861 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1862 | ||
392ddc32 | 1863 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
bb799ca0 JA |
1864 | if (!bs->bio_slab) { |
1865 | kfree(bs); | |
1866 | return NULL; | |
1867 | } | |
1868 | ||
1869 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1da177e4 LT |
1870 | if (!bs->bio_pool) |
1871 | goto bad; | |
1872 | ||
9f060e22 KO |
1873 | bs->bvec_pool = biovec_create_pool(bs, pool_size); |
1874 | if (!bs->bvec_pool) | |
df2cb6da KO |
1875 | goto bad; |
1876 | ||
1877 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1878 | if (!bs->rescue_workqueue) | |
1879 | goto bad; | |
1da177e4 | 1880 | |
df2cb6da | 1881 | return bs; |
1da177e4 LT |
1882 | bad: |
1883 | bioset_free(bs); | |
1884 | return NULL; | |
1885 | } | |
a112a71d | 1886 | EXPORT_SYMBOL(bioset_create); |
1da177e4 | 1887 | |
852c788f TH |
1888 | #ifdef CONFIG_BLK_CGROUP |
1889 | /** | |
1890 | * bio_associate_current - associate a bio with %current | |
1891 | * @bio: target bio | |
1892 | * | |
1893 | * Associate @bio with %current if it hasn't been associated yet. Block | |
1894 | * layer will treat @bio as if it were issued by %current no matter which | |
1895 | * task actually issues it. | |
1896 | * | |
1897 | * This function takes an extra reference of @task's io_context and blkcg | |
1898 | * which will be put when @bio is released. The caller must own @bio, | |
1899 | * ensure %current->io_context exists, and is responsible for synchronizing | |
1900 | * calls to this function. | |
1901 | */ | |
1902 | int bio_associate_current(struct bio *bio) | |
1903 | { | |
1904 | struct io_context *ioc; | |
1905 | struct cgroup_subsys_state *css; | |
1906 | ||
1907 | if (bio->bi_ioc) | |
1908 | return -EBUSY; | |
1909 | ||
1910 | ioc = current->io_context; | |
1911 | if (!ioc) | |
1912 | return -ENOENT; | |
1913 | ||
1914 | /* acquire active ref on @ioc and associate */ | |
1915 | get_io_context_active(ioc); | |
1916 | bio->bi_ioc = ioc; | |
1917 | ||
1918 | /* associate blkcg if exists */ | |
1919 | rcu_read_lock(); | |
1920 | css = task_subsys_state(current, blkio_subsys_id); | |
1921 | if (css && css_tryget(css)) | |
1922 | bio->bi_css = css; | |
1923 | rcu_read_unlock(); | |
1924 | ||
1925 | return 0; | |
1926 | } | |
1927 | ||
1928 | /** | |
1929 | * bio_disassociate_task - undo bio_associate_current() | |
1930 | * @bio: target bio | |
1931 | */ | |
1932 | void bio_disassociate_task(struct bio *bio) | |
1933 | { | |
1934 | if (bio->bi_ioc) { | |
1935 | put_io_context(bio->bi_ioc); | |
1936 | bio->bi_ioc = NULL; | |
1937 | } | |
1938 | if (bio->bi_css) { | |
1939 | css_put(bio->bi_css); | |
1940 | bio->bi_css = NULL; | |
1941 | } | |
1942 | } | |
1943 | ||
1944 | #endif /* CONFIG_BLK_CGROUP */ | |
1945 | ||
1da177e4 LT |
1946 | static void __init biovec_init_slabs(void) |
1947 | { | |
1948 | int i; | |
1949 | ||
1950 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1951 | int size; | |
1952 | struct biovec_slab *bvs = bvec_slabs + i; | |
1953 | ||
a7fcd37c JA |
1954 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
1955 | bvs->slab = NULL; | |
1956 | continue; | |
1957 | } | |
a7fcd37c | 1958 | |
1da177e4 LT |
1959 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
1960 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1961 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1962 | } |
1963 | } | |
1964 | ||
1965 | static int __init init_bio(void) | |
1966 | { | |
bb799ca0 JA |
1967 | bio_slab_max = 2; |
1968 | bio_slab_nr = 0; | |
1969 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
1970 | if (!bio_slabs) | |
1971 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 1972 | |
7878cba9 | 1973 | bio_integrity_init(); |
1da177e4 LT |
1974 | biovec_init_slabs(); |
1975 | ||
bb799ca0 | 1976 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); |
1da177e4 LT |
1977 | if (!fs_bio_set) |
1978 | panic("bio: can't allocate bios\n"); | |
1979 | ||
a91a2785 MP |
1980 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) |
1981 | panic("bio: can't create integrity pool\n"); | |
1982 | ||
0eaae62a MD |
1983 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1984 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1985 | if (!bio_split_pool) |
1986 | panic("bio: can't create split pool\n"); | |
1987 | ||
1988 | return 0; | |
1989 | } | |
1da177e4 | 1990 | subsys_initcall(init_bio); |