Commit | Line | Data |
---|---|---|
8c16567d | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
0fe23479 | 3 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
1da177e4 LT |
4 | */ |
5 | #include <linux/mm.h> | |
6 | #include <linux/swap.h> | |
7 | #include <linux/bio.h> | |
8 | #include <linux/blkdev.h> | |
a27bb332 | 9 | #include <linux/uio.h> |
852c788f | 10 | #include <linux/iocontext.h> |
1da177e4 LT |
11 | #include <linux/slab.h> |
12 | #include <linux/init.h> | |
13 | #include <linux/kernel.h> | |
630d9c47 | 14 | #include <linux/export.h> |
1da177e4 LT |
15 | #include <linux/mempool.h> |
16 | #include <linux/workqueue.h> | |
852c788f | 17 | #include <linux/cgroup.h> |
08e18eab | 18 | #include <linux/blk-cgroup.h> |
b4c5875d | 19 | #include <linux/highmem.h> |
1da177e4 | 20 | |
55782138 | 21 | #include <trace/events/block.h> |
9e234eea | 22 | #include "blk.h" |
67b42d0b | 23 | #include "blk-rq-qos.h" |
0bfc2455 | 24 | |
392ddc32 JA |
25 | /* |
26 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
27 | * itself, to shrink a bio data allocation from two mempool calls to one | |
28 | */ | |
29 | #define BIO_INLINE_VECS 4 | |
30 | ||
1da177e4 LT |
31 | /* |
32 | * if you change this list, also change bvec_alloc or things will | |
33 | * break badly! cannot be bigger than what you can fit into an | |
34 | * unsigned short | |
35 | */ | |
bd5c4fac | 36 | #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } |
ed996a52 | 37 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { |
bd5c4fac | 38 | BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), |
1da177e4 LT |
39 | }; |
40 | #undef BV | |
41 | ||
1da177e4 LT |
42 | /* |
43 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
44 | * IO code that does not need private memory pools. | |
45 | */ | |
f4f8154a | 46 | struct bio_set fs_bio_set; |
3f86a82a | 47 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 48 | |
bb799ca0 JA |
49 | /* |
50 | * Our slab pool management | |
51 | */ | |
52 | struct bio_slab { | |
53 | struct kmem_cache *slab; | |
54 | unsigned int slab_ref; | |
55 | unsigned int slab_size; | |
56 | char name[8]; | |
57 | }; | |
58 | static DEFINE_MUTEX(bio_slab_lock); | |
59 | static struct bio_slab *bio_slabs; | |
60 | static unsigned int bio_slab_nr, bio_slab_max; | |
61 | ||
62 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
63 | { | |
64 | unsigned int sz = sizeof(struct bio) + extra_size; | |
65 | struct kmem_cache *slab = NULL; | |
389d7b26 | 66 | struct bio_slab *bslab, *new_bio_slabs; |
386bc35a | 67 | unsigned int new_bio_slab_max; |
bb799ca0 JA |
68 | unsigned int i, entry = -1; |
69 | ||
70 | mutex_lock(&bio_slab_lock); | |
71 | ||
72 | i = 0; | |
73 | while (i < bio_slab_nr) { | |
f06f135d | 74 | bslab = &bio_slabs[i]; |
bb799ca0 JA |
75 | |
76 | if (!bslab->slab && entry == -1) | |
77 | entry = i; | |
78 | else if (bslab->slab_size == sz) { | |
79 | slab = bslab->slab; | |
80 | bslab->slab_ref++; | |
81 | break; | |
82 | } | |
83 | i++; | |
84 | } | |
85 | ||
86 | if (slab) | |
87 | goto out_unlock; | |
88 | ||
89 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
386bc35a | 90 | new_bio_slab_max = bio_slab_max << 1; |
389d7b26 | 91 | new_bio_slabs = krealloc(bio_slabs, |
386bc35a | 92 | new_bio_slab_max * sizeof(struct bio_slab), |
389d7b26 AK |
93 | GFP_KERNEL); |
94 | if (!new_bio_slabs) | |
bb799ca0 | 95 | goto out_unlock; |
386bc35a | 96 | bio_slab_max = new_bio_slab_max; |
389d7b26 | 97 | bio_slabs = new_bio_slabs; |
bb799ca0 JA |
98 | } |
99 | if (entry == -1) | |
100 | entry = bio_slab_nr++; | |
101 | ||
102 | bslab = &bio_slabs[entry]; | |
103 | ||
104 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
6a241483 MP |
105 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
106 | SLAB_HWCACHE_ALIGN, NULL); | |
bb799ca0 JA |
107 | if (!slab) |
108 | goto out_unlock; | |
109 | ||
bb799ca0 JA |
110 | bslab->slab = slab; |
111 | bslab->slab_ref = 1; | |
112 | bslab->slab_size = sz; | |
113 | out_unlock: | |
114 | mutex_unlock(&bio_slab_lock); | |
115 | return slab; | |
116 | } | |
117 | ||
118 | static void bio_put_slab(struct bio_set *bs) | |
119 | { | |
120 | struct bio_slab *bslab = NULL; | |
121 | unsigned int i; | |
122 | ||
123 | mutex_lock(&bio_slab_lock); | |
124 | ||
125 | for (i = 0; i < bio_slab_nr; i++) { | |
126 | if (bs->bio_slab == bio_slabs[i].slab) { | |
127 | bslab = &bio_slabs[i]; | |
128 | break; | |
129 | } | |
130 | } | |
131 | ||
132 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
133 | goto out; | |
134 | ||
135 | WARN_ON(!bslab->slab_ref); | |
136 | ||
137 | if (--bslab->slab_ref) | |
138 | goto out; | |
139 | ||
140 | kmem_cache_destroy(bslab->slab); | |
141 | bslab->slab = NULL; | |
142 | ||
143 | out: | |
144 | mutex_unlock(&bio_slab_lock); | |
145 | } | |
146 | ||
7ba1ba12 MP |
147 | unsigned int bvec_nr_vecs(unsigned short idx) |
148 | { | |
d6c02a9b | 149 | return bvec_slabs[--idx].nr_vecs; |
7ba1ba12 MP |
150 | } |
151 | ||
9f060e22 | 152 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 | 153 | { |
ed996a52 CH |
154 | if (!idx) |
155 | return; | |
156 | idx--; | |
157 | ||
158 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
bb799ca0 | 159 | |
ed996a52 | 160 | if (idx == BVEC_POOL_MAX) { |
9f060e22 | 161 | mempool_free(bv, pool); |
ed996a52 | 162 | } else { |
bb799ca0 JA |
163 | struct biovec_slab *bvs = bvec_slabs + idx; |
164 | ||
165 | kmem_cache_free(bvs->slab, bv); | |
166 | } | |
167 | } | |
168 | ||
9f060e22 KO |
169 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
170 | mempool_t *pool) | |
1da177e4 LT |
171 | { |
172 | struct bio_vec *bvl; | |
1da177e4 | 173 | |
7ff9345f JA |
174 | /* |
175 | * see comment near bvec_array define! | |
176 | */ | |
177 | switch (nr) { | |
178 | case 1: | |
179 | *idx = 0; | |
180 | break; | |
181 | case 2 ... 4: | |
182 | *idx = 1; | |
183 | break; | |
184 | case 5 ... 16: | |
185 | *idx = 2; | |
186 | break; | |
187 | case 17 ... 64: | |
188 | *idx = 3; | |
189 | break; | |
190 | case 65 ... 128: | |
191 | *idx = 4; | |
192 | break; | |
193 | case 129 ... BIO_MAX_PAGES: | |
194 | *idx = 5; | |
195 | break; | |
196 | default: | |
197 | return NULL; | |
198 | } | |
199 | ||
200 | /* | |
201 | * idx now points to the pool we want to allocate from. only the | |
202 | * 1-vec entry pool is mempool backed. | |
203 | */ | |
ed996a52 | 204 | if (*idx == BVEC_POOL_MAX) { |
7ff9345f | 205 | fallback: |
9f060e22 | 206 | bvl = mempool_alloc(pool, gfp_mask); |
7ff9345f JA |
207 | } else { |
208 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
d0164adc | 209 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
7ff9345f | 210 | |
0a0d96b0 | 211 | /* |
7ff9345f JA |
212 | * Make this allocation restricted and don't dump info on |
213 | * allocation failures, since we'll fallback to the mempool | |
214 | * in case of failure. | |
0a0d96b0 | 215 | */ |
7ff9345f | 216 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 217 | |
0a0d96b0 | 218 | /* |
d0164adc | 219 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM |
7ff9345f | 220 | * is set, retry with the 1-entry mempool |
0a0d96b0 | 221 | */ |
7ff9345f | 222 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
d0164adc | 223 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { |
ed996a52 | 224 | *idx = BVEC_POOL_MAX; |
7ff9345f JA |
225 | goto fallback; |
226 | } | |
227 | } | |
228 | ||
ed996a52 | 229 | (*idx)++; |
1da177e4 LT |
230 | return bvl; |
231 | } | |
232 | ||
9ae3b3f5 | 233 | void bio_uninit(struct bio *bio) |
1da177e4 | 234 | { |
6f70fb66 | 235 | bio_disassociate_blkg(bio); |
4254bba1 | 236 | } |
9ae3b3f5 | 237 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 238 | |
4254bba1 KO |
239 | static void bio_free(struct bio *bio) |
240 | { | |
241 | struct bio_set *bs = bio->bi_pool; | |
242 | void *p; | |
243 | ||
9ae3b3f5 | 244 | bio_uninit(bio); |
4254bba1 KO |
245 | |
246 | if (bs) { | |
8aa6ba2f | 247 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
4254bba1 KO |
248 | |
249 | /* | |
250 | * If we have front padding, adjust the bio pointer before freeing | |
251 | */ | |
252 | p = bio; | |
bb799ca0 JA |
253 | p -= bs->front_pad; |
254 | ||
8aa6ba2f | 255 | mempool_free(p, &bs->bio_pool); |
4254bba1 KO |
256 | } else { |
257 | /* Bio was allocated by bio_kmalloc() */ | |
258 | kfree(bio); | |
259 | } | |
3676347a PO |
260 | } |
261 | ||
9ae3b3f5 JA |
262 | /* |
263 | * Users of this function have their own bio allocation. Subsequently, | |
264 | * they must remember to pair any call to bio_init() with bio_uninit() | |
265 | * when IO has completed, or when the bio is released. | |
266 | */ | |
3a83f467 ML |
267 | void bio_init(struct bio *bio, struct bio_vec *table, |
268 | unsigned short max_vecs) | |
1da177e4 | 269 | { |
2b94de55 | 270 | memset(bio, 0, sizeof(*bio)); |
c4cf5261 | 271 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 272 | atomic_set(&bio->__bi_cnt, 1); |
3a83f467 ML |
273 | |
274 | bio->bi_io_vec = table; | |
275 | bio->bi_max_vecs = max_vecs; | |
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 | ||
9ae3b3f5 | 293 | bio_uninit(bio); |
f44b48c7 KO |
294 | |
295 | memset(bio, 0, BIO_RESET_BYTES); | |
4246a0b6 | 296 | bio->bi_flags = flags; |
c4cf5261 | 297 | atomic_set(&bio->__bi_remaining, 1); |
f44b48c7 KO |
298 | } |
299 | EXPORT_SYMBOL(bio_reset); | |
300 | ||
38f8baae | 301 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 302 | { |
4246a0b6 CH |
303 | struct bio *parent = bio->bi_private; |
304 | ||
4e4cbee9 CH |
305 | if (!parent->bi_status) |
306 | parent->bi_status = bio->bi_status; | |
196d38bc | 307 | bio_put(bio); |
38f8baae CH |
308 | return parent; |
309 | } | |
310 | ||
311 | static void bio_chain_endio(struct bio *bio) | |
312 | { | |
313 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
314 | } |
315 | ||
316 | /** | |
317 | * bio_chain - chain bio completions | |
1051a902 RD |
318 | * @bio: the target bio |
319 | * @parent: the @bio's parent bio | |
196d38bc KO |
320 | * |
321 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
322 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
323 | * completed; the chained bio will also be freed when it completes. | |
324 | * | |
325 | * The caller must not set bi_private or bi_end_io in @bio. | |
326 | */ | |
327 | void bio_chain(struct bio *bio, struct bio *parent) | |
328 | { | |
329 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
330 | ||
331 | bio->bi_private = parent; | |
332 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 333 | bio_inc_remaining(parent); |
196d38bc KO |
334 | } |
335 | EXPORT_SYMBOL(bio_chain); | |
336 | ||
df2cb6da KO |
337 | static void bio_alloc_rescue(struct work_struct *work) |
338 | { | |
339 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
340 | struct bio *bio; | |
341 | ||
342 | while (1) { | |
343 | spin_lock(&bs->rescue_lock); | |
344 | bio = bio_list_pop(&bs->rescue_list); | |
345 | spin_unlock(&bs->rescue_lock); | |
346 | ||
347 | if (!bio) | |
348 | break; | |
349 | ||
350 | generic_make_request(bio); | |
351 | } | |
352 | } | |
353 | ||
354 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
355 | { | |
356 | struct bio_list punt, nopunt; | |
357 | struct bio *bio; | |
358 | ||
47e0fb46 N |
359 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
360 | return; | |
df2cb6da KO |
361 | /* |
362 | * In order to guarantee forward progress we must punt only bios that | |
363 | * were allocated from this bio_set; otherwise, if there was a bio on | |
364 | * there for a stacking driver higher up in the stack, processing it | |
365 | * could require allocating bios from this bio_set, and doing that from | |
366 | * our own rescuer would be bad. | |
367 | * | |
368 | * Since bio lists are singly linked, pop them all instead of trying to | |
369 | * remove from the middle of the list: | |
370 | */ | |
371 | ||
372 | bio_list_init(&punt); | |
373 | bio_list_init(&nopunt); | |
374 | ||
f5fe1b51 | 375 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 376 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 377 | current->bio_list[0] = nopunt; |
df2cb6da | 378 | |
f5fe1b51 N |
379 | bio_list_init(&nopunt); |
380 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
381 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
382 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
383 | |
384 | spin_lock(&bs->rescue_lock); | |
385 | bio_list_merge(&bs->rescue_list, &punt); | |
386 | spin_unlock(&bs->rescue_lock); | |
387 | ||
388 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
389 | } | |
390 | ||
1da177e4 LT |
391 | /** |
392 | * bio_alloc_bioset - allocate a bio for I/O | |
519c8e9f | 393 | * @gfp_mask: the GFP_* mask given to the slab allocator |
1da177e4 | 394 | * @nr_iovecs: number of iovecs to pre-allocate |
db18efac | 395 | * @bs: the bio_set to allocate from. |
1da177e4 LT |
396 | * |
397 | * Description: | |
3f86a82a KO |
398 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
399 | * backed by the @bs's mempool. | |
400 | * | |
d0164adc MG |
401 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will |
402 | * always be able to allocate a bio. This is due to the mempool guarantees. | |
403 | * To make this work, callers must never allocate more than 1 bio at a time | |
404 | * from this pool. Callers that need to allocate more than 1 bio must always | |
405 | * submit the previously allocated bio for IO before attempting to allocate | |
406 | * a new one. Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 407 | * |
df2cb6da KO |
408 | * Note that when running under generic_make_request() (i.e. any block |
409 | * driver), bios are not submitted until after you return - see the code in | |
410 | * generic_make_request() that converts recursion into iteration, to prevent | |
411 | * stack overflows. | |
412 | * | |
413 | * This would normally mean allocating multiple bios under | |
414 | * generic_make_request() would be susceptible to deadlocks, but we have | |
415 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
416 | * thread. | |
417 | * | |
418 | * However, we do not guarantee forward progress for allocations from other | |
419 | * mempools. Doing multiple allocations from the same mempool under | |
420 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
421 | * for per bio allocations. | |
422 | * | |
3f86a82a KO |
423 | * RETURNS: |
424 | * Pointer to new bio on success, NULL on failure. | |
425 | */ | |
7a88fa19 DC |
426 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, |
427 | struct bio_set *bs) | |
1da177e4 | 428 | { |
df2cb6da | 429 | gfp_t saved_gfp = gfp_mask; |
3f86a82a KO |
430 | unsigned front_pad; |
431 | unsigned inline_vecs; | |
34053979 | 432 | struct bio_vec *bvl = NULL; |
451a9ebf TH |
433 | struct bio *bio; |
434 | void *p; | |
435 | ||
3f86a82a KO |
436 | if (!bs) { |
437 | if (nr_iovecs > UIO_MAXIOV) | |
438 | return NULL; | |
439 | ||
440 | p = kmalloc(sizeof(struct bio) + | |
441 | nr_iovecs * sizeof(struct bio_vec), | |
442 | gfp_mask); | |
443 | front_pad = 0; | |
444 | inline_vecs = nr_iovecs; | |
445 | } else { | |
d8f429e1 | 446 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
8aa6ba2f KO |
447 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && |
448 | nr_iovecs > 0)) | |
d8f429e1 | 449 | return NULL; |
df2cb6da KO |
450 | /* |
451 | * generic_make_request() converts recursion to iteration; this | |
452 | * means if we're running beneath it, any bios we allocate and | |
453 | * submit will not be submitted (and thus freed) until after we | |
454 | * return. | |
455 | * | |
456 | * This exposes us to a potential deadlock if we allocate | |
457 | * multiple bios from the same bio_set() while running | |
458 | * underneath generic_make_request(). If we were to allocate | |
459 | * multiple bios (say a stacking block driver that was splitting | |
460 | * bios), we would deadlock if we exhausted the mempool's | |
461 | * reserve. | |
462 | * | |
463 | * We solve this, and guarantee forward progress, with a rescuer | |
464 | * workqueue per bio_set. If we go to allocate and there are | |
465 | * bios on current->bio_list, we first try the allocation | |
d0164adc MG |
466 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those |
467 | * bios we would be blocking to the rescuer workqueue before | |
468 | * we retry with the original gfp_flags. | |
df2cb6da KO |
469 | */ |
470 | ||
f5fe1b51 N |
471 | if (current->bio_list && |
472 | (!bio_list_empty(¤t->bio_list[0]) || | |
47e0fb46 N |
473 | !bio_list_empty(¤t->bio_list[1])) && |
474 | bs->rescue_workqueue) | |
d0164adc | 475 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; |
df2cb6da | 476 | |
8aa6ba2f | 477 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
478 | if (!p && gfp_mask != saved_gfp) { |
479 | punt_bios_to_rescuer(bs); | |
480 | gfp_mask = saved_gfp; | |
8aa6ba2f | 481 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
482 | } |
483 | ||
3f86a82a KO |
484 | front_pad = bs->front_pad; |
485 | inline_vecs = BIO_INLINE_VECS; | |
486 | } | |
487 | ||
451a9ebf TH |
488 | if (unlikely(!p)) |
489 | return NULL; | |
1da177e4 | 490 | |
3f86a82a | 491 | bio = p + front_pad; |
3a83f467 | 492 | bio_init(bio, NULL, 0); |
34053979 | 493 | |
3f86a82a | 494 | if (nr_iovecs > inline_vecs) { |
ed996a52 CH |
495 | unsigned long idx = 0; |
496 | ||
8aa6ba2f | 497 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
498 | if (!bvl && gfp_mask != saved_gfp) { |
499 | punt_bios_to_rescuer(bs); | |
500 | gfp_mask = saved_gfp; | |
8aa6ba2f | 501 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
502 | } |
503 | ||
34053979 IM |
504 | if (unlikely(!bvl)) |
505 | goto err_free; | |
a38352e0 | 506 | |
ed996a52 | 507 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
3f86a82a KO |
508 | } else if (nr_iovecs) { |
509 | bvl = bio->bi_inline_vecs; | |
1da177e4 | 510 | } |
3f86a82a KO |
511 | |
512 | bio->bi_pool = bs; | |
34053979 | 513 | bio->bi_max_vecs = nr_iovecs; |
34053979 | 514 | bio->bi_io_vec = bvl; |
1da177e4 | 515 | return bio; |
34053979 IM |
516 | |
517 | err_free: | |
8aa6ba2f | 518 | mempool_free(p, &bs->bio_pool); |
34053979 | 519 | return NULL; |
1da177e4 | 520 | } |
a112a71d | 521 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 522 | |
38a72dac | 523 | void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) |
1da177e4 LT |
524 | { |
525 | unsigned long flags; | |
7988613b KO |
526 | struct bio_vec bv; |
527 | struct bvec_iter iter; | |
1da177e4 | 528 | |
38a72dac | 529 | __bio_for_each_segment(bv, bio, iter, start) { |
7988613b KO |
530 | char *data = bvec_kmap_irq(&bv, &flags); |
531 | memset(data, 0, bv.bv_len); | |
532 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
533 | bvec_kunmap_irq(data, &flags); |
534 | } | |
535 | } | |
38a72dac | 536 | EXPORT_SYMBOL(zero_fill_bio_iter); |
1da177e4 LT |
537 | |
538 | /** | |
539 | * bio_put - release a reference to a bio | |
540 | * @bio: bio to release reference to | |
541 | * | |
542 | * Description: | |
543 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 544 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
545 | **/ |
546 | void bio_put(struct bio *bio) | |
547 | { | |
dac56212 | 548 | if (!bio_flagged(bio, BIO_REFFED)) |
4254bba1 | 549 | bio_free(bio); |
dac56212 JA |
550 | else { |
551 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
552 | ||
553 | /* | |
554 | * last put frees it | |
555 | */ | |
556 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
557 | bio_free(bio); | |
558 | } | |
1da177e4 | 559 | } |
a112a71d | 560 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 561 | |
59d276fe KO |
562 | /** |
563 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
564 | * @bio: destination bio | |
565 | * @bio_src: bio to clone | |
566 | * | |
567 | * Clone a &bio. Caller will own the returned bio, but not | |
568 | * the actual data it points to. Reference count of returned | |
569 | * bio will be one. | |
570 | * | |
571 | * Caller must ensure that @bio_src is not freed before @bio. | |
572 | */ | |
573 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
574 | { | |
ed996a52 | 575 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
59d276fe KO |
576 | |
577 | /* | |
74d46992 | 578 | * most users will be overriding ->bi_disk with a new target, |
59d276fe KO |
579 | * so we don't set nor calculate new physical/hw segment counts here |
580 | */ | |
74d46992 | 581 | bio->bi_disk = bio_src->bi_disk; |
62530ed8 | 582 | bio->bi_partno = bio_src->bi_partno; |
b7c44ed9 | 583 | bio_set_flag(bio, BIO_CLONED); |
111be883 SL |
584 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
585 | bio_set_flag(bio, BIO_THROTTLED); | |
1eff9d32 | 586 | bio->bi_opf = bio_src->bi_opf; |
ca474b73 | 587 | bio->bi_ioprio = bio_src->bi_ioprio; |
cb6934f8 | 588 | bio->bi_write_hint = bio_src->bi_write_hint; |
59d276fe KO |
589 | bio->bi_iter = bio_src->bi_iter; |
590 | bio->bi_io_vec = bio_src->bi_io_vec; | |
20bd723e | 591 | |
db6638d7 | 592 | bio_clone_blkg_association(bio, bio_src); |
e439bedf | 593 | blkcg_bio_issue_init(bio); |
59d276fe KO |
594 | } |
595 | EXPORT_SYMBOL(__bio_clone_fast); | |
596 | ||
597 | /** | |
598 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
599 | * @bio: bio to clone | |
600 | * @gfp_mask: allocation priority | |
601 | * @bs: bio_set to allocate from | |
602 | * | |
603 | * Like __bio_clone_fast, only also allocates the returned bio | |
604 | */ | |
605 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
606 | { | |
607 | struct bio *b; | |
608 | ||
609 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
610 | if (!b) | |
611 | return NULL; | |
612 | ||
613 | __bio_clone_fast(b, bio); | |
614 | ||
615 | if (bio_integrity(bio)) { | |
616 | int ret; | |
617 | ||
618 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
619 | ||
620 | if (ret < 0) { | |
621 | bio_put(b); | |
622 | return NULL; | |
623 | } | |
624 | } | |
625 | ||
626 | return b; | |
627 | } | |
628 | EXPORT_SYMBOL(bio_clone_fast); | |
629 | ||
5919482e ML |
630 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
631 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 632 | bool *same_page) |
5919482e ML |
633 | { |
634 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + | |
635 | bv->bv_offset + bv->bv_len - 1; | |
636 | phys_addr_t page_addr = page_to_phys(page); | |
637 | ||
638 | if (vec_end_addr + 1 != page_addr + off) | |
639 | return false; | |
640 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
641 | return false; | |
52d52d1c | 642 | |
ff896738 CH |
643 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
644 | if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) | |
645 | return false; | |
5919482e ML |
646 | return true; |
647 | } | |
648 | ||
489fbbcb ML |
649 | /* |
650 | * Check if the @page can be added to the current segment(@bv), and make | |
651 | * sure to call it only if page_is_mergeable(@bv, @page) is true | |
652 | */ | |
653 | static bool can_add_page_to_seg(struct request_queue *q, | |
654 | struct bio_vec *bv, struct page *page, unsigned len, | |
655 | unsigned offset) | |
656 | { | |
657 | unsigned long mask = queue_segment_boundary(q); | |
658 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
659 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
660 | ||
661 | if ((addr1 | mask) != (addr2 | mask)) | |
662 | return false; | |
663 | ||
664 | if (bv->bv_len + len > queue_max_segment_size(q)) | |
665 | return false; | |
666 | ||
667 | return true; | |
668 | } | |
669 | ||
1da177e4 | 670 | /** |
19047087 | 671 | * __bio_add_pc_page - attempt to add page to passthrough bio |
c66a14d0 KO |
672 | * @q: the target queue |
673 | * @bio: destination bio | |
674 | * @page: page to add | |
675 | * @len: vec entry length | |
676 | * @offset: vec entry offset | |
19047087 | 677 | * @put_same_page: put the page if it is same with last added page |
1da177e4 | 678 | * |
c66a14d0 KO |
679 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
680 | * number of reasons, such as the bio being full or target block device | |
681 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
682 | * so it is always possible to add a single page to an empty bio. | |
683 | * | |
5a8ce240 | 684 | * This should only be used by passthrough bios. |
1da177e4 | 685 | */ |
4713839d | 686 | static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, |
19047087 ML |
687 | struct page *page, unsigned int len, unsigned int offset, |
688 | bool put_same_page) | |
1da177e4 | 689 | { |
1da177e4 | 690 | struct bio_vec *bvec; |
ff896738 | 691 | bool same_page = false; |
1da177e4 LT |
692 | |
693 | /* | |
694 | * cloned bio must not modify vec list | |
695 | */ | |
696 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
697 | return 0; | |
698 | ||
c66a14d0 | 699 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) |
1da177e4 LT |
700 | return 0; |
701 | ||
80cfd548 | 702 | if (bio->bi_vcnt > 0) { |
5a8ce240 | 703 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
80cfd548 | 704 | |
5a8ce240 ML |
705 | if (page == bvec->bv_page && |
706 | offset == bvec->bv_offset + bvec->bv_len) { | |
19047087 ML |
707 | if (put_same_page) |
708 | put_page(page); | |
5a8ce240 | 709 | bvec->bv_len += len; |
80cfd548 JA |
710 | goto done; |
711 | } | |
66cb45aa JA |
712 | |
713 | /* | |
714 | * If the queue doesn't support SG gaps and adding this | |
715 | * offset would create a gap, disallow it. | |
716 | */ | |
5a8ce240 | 717 | if (bvec_gap_to_prev(q, bvec, offset)) |
66cb45aa | 718 | return 0; |
489fbbcb | 719 | |
ff896738 | 720 | if (page_is_mergeable(bvec, page, len, offset, &same_page) && |
dcdca753 CH |
721 | can_add_page_to_seg(q, bvec, page, len, offset)) { |
722 | bvec->bv_len += len; | |
723 | goto done; | |
724 | } | |
80cfd548 JA |
725 | } |
726 | ||
79d08f89 | 727 | if (bio_full(bio, len)) |
1da177e4 LT |
728 | return 0; |
729 | ||
14ccb66b | 730 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
731 | return 0; |
732 | ||
fcbf6a08 ML |
733 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
734 | bvec->bv_page = page; | |
735 | bvec->bv_len = len; | |
736 | bvec->bv_offset = offset; | |
737 | bio->bi_vcnt++; | |
80cfd548 | 738 | done: |
dcdca753 | 739 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
740 | return len; |
741 | } | |
19047087 ML |
742 | |
743 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, | |
744 | struct page *page, unsigned int len, unsigned int offset) | |
745 | { | |
746 | return __bio_add_pc_page(q, bio, page, len, offset, false); | |
747 | } | |
a112a71d | 748 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 749 | |
1da177e4 | 750 | /** |
0aa69fd3 CH |
751 | * __bio_try_merge_page - try appending data to an existing bvec. |
752 | * @bio: destination bio | |
551879a4 | 753 | * @page: start page to add |
0aa69fd3 | 754 | * @len: length of the data to add |
551879a4 | 755 | * @off: offset of the data relative to @page |
ff896738 | 756 | * @same_page: return if the segment has been merged inside the same page |
1da177e4 | 757 | * |
0aa69fd3 CH |
758 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
759 | * a useful optimisation for file systems with a block size smaller than the | |
760 | * page size. | |
761 | * | |
551879a4 ML |
762 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
763 | * | |
0aa69fd3 | 764 | * Return %true on success or %false on failure. |
1da177e4 | 765 | */ |
0aa69fd3 | 766 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
ff896738 | 767 | unsigned int len, unsigned int off, bool *same_page) |
1da177e4 | 768 | { |
c66a14d0 | 769 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
0aa69fd3 | 770 | return false; |
762380ad | 771 | |
c66a14d0 | 772 | if (bio->bi_vcnt > 0) { |
0aa69fd3 | 773 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
5919482e ML |
774 | |
775 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
776 | bv->bv_len += len; | |
777 | bio->bi_iter.bi_size += len; | |
778 | return true; | |
779 | } | |
c66a14d0 | 780 | } |
0aa69fd3 CH |
781 | return false; |
782 | } | |
783 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); | |
c66a14d0 | 784 | |
0aa69fd3 | 785 | /** |
551879a4 | 786 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 787 | * @bio: destination bio |
551879a4 ML |
788 | * @page: start page to add |
789 | * @len: length of the data to add, may cross pages | |
790 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
791 | * |
792 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
793 | * that @bio has space for another bvec. | |
794 | */ | |
795 | void __bio_add_page(struct bio *bio, struct page *page, | |
796 | unsigned int len, unsigned int off) | |
797 | { | |
798 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 799 | |
0aa69fd3 | 800 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 801 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
802 | |
803 | bv->bv_page = page; | |
804 | bv->bv_offset = off; | |
805 | bv->bv_len = len; | |
c66a14d0 | 806 | |
c66a14d0 | 807 | bio->bi_iter.bi_size += len; |
0aa69fd3 | 808 | bio->bi_vcnt++; |
b8e24a93 JW |
809 | |
810 | if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) | |
811 | bio_set_flag(bio, BIO_WORKINGSET); | |
0aa69fd3 CH |
812 | } |
813 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
814 | ||
815 | /** | |
551879a4 | 816 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 817 | * @bio: destination bio |
551879a4 ML |
818 | * @page: start page to add |
819 | * @len: vec entry length, may cross pages | |
820 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 821 | * |
551879a4 | 822 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
823 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
824 | */ | |
825 | int bio_add_page(struct bio *bio, struct page *page, | |
826 | unsigned int len, unsigned int offset) | |
827 | { | |
ff896738 CH |
828 | bool same_page = false; |
829 | ||
830 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 831 | if (bio_full(bio, len)) |
0aa69fd3 CH |
832 | return 0; |
833 | __bio_add_page(bio, page, len, offset); | |
834 | } | |
c66a14d0 | 835 | return len; |
1da177e4 | 836 | } |
a112a71d | 837 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 838 | |
d241a95f | 839 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
840 | { |
841 | struct bvec_iter_all iter_all; | |
842 | struct bio_vec *bvec; | |
7321ecbf | 843 | |
b2d0d991 CH |
844 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
845 | return; | |
846 | ||
d241a95f CH |
847 | bio_for_each_segment_all(bvec, bio, iter_all) { |
848 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
849 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 850 | put_page(bvec->bv_page); |
d241a95f | 851 | } |
7321ecbf CH |
852 | } |
853 | ||
6d0c48ae JA |
854 | static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) |
855 | { | |
856 | const struct bio_vec *bv = iter->bvec; | |
857 | unsigned int len; | |
858 | size_t size; | |
859 | ||
860 | if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) | |
861 | return -EINVAL; | |
862 | ||
863 | len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); | |
864 | size = bio_add_page(bio, bv->bv_page, len, | |
865 | bv->bv_offset + iter->iov_offset); | |
a10584c3 CH |
866 | if (unlikely(size != len)) |
867 | return -EINVAL; | |
a10584c3 CH |
868 | iov_iter_advance(iter, size); |
869 | return 0; | |
6d0c48ae JA |
870 | } |
871 | ||
576ed913 CH |
872 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
873 | ||
2cefe4db | 874 | /** |
17d51b10 | 875 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
876 | * @bio: bio to add pages to |
877 | * @iter: iov iterator describing the region to be mapped | |
878 | * | |
17d51b10 | 879 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 880 | * pages will have to be released using put_page() when done. |
17d51b10 MW |
881 | * For multi-segment *iter, this function only adds pages from the |
882 | * the next non-empty segment of the iov iterator. | |
2cefe4db | 883 | */ |
17d51b10 | 884 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 885 | { |
576ed913 CH |
886 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
887 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
888 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
889 | struct page **pages = (struct page **)bv; | |
45691804 | 890 | bool same_page = false; |
576ed913 CH |
891 | ssize_t size, left; |
892 | unsigned len, i; | |
b403ea24 | 893 | size_t offset; |
576ed913 CH |
894 | |
895 | /* | |
896 | * Move page array up in the allocated memory for the bio vecs as far as | |
897 | * possible so that we can start filling biovecs from the beginning | |
898 | * without overwriting the temporary page array. | |
899 | */ | |
900 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
901 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db KO |
902 | |
903 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
904 | if (unlikely(size <= 0)) | |
905 | return size ? size : -EFAULT; | |
2cefe4db | 906 | |
576ed913 CH |
907 | for (left = size, i = 0; left > 0; left -= len, i++) { |
908 | struct page *page = pages[i]; | |
2cefe4db | 909 | |
576ed913 | 910 | len = min_t(size_t, PAGE_SIZE - offset, left); |
45691804 CH |
911 | |
912 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
913 | if (same_page) | |
914 | put_page(page); | |
915 | } else { | |
79d08f89 | 916 | if (WARN_ON_ONCE(bio_full(bio, len))) |
45691804 CH |
917 | return -EINVAL; |
918 | __bio_add_page(bio, page, len, offset); | |
919 | } | |
576ed913 | 920 | offset = 0; |
2cefe4db KO |
921 | } |
922 | ||
2cefe4db KO |
923 | iov_iter_advance(iter, size); |
924 | return 0; | |
925 | } | |
17d51b10 MW |
926 | |
927 | /** | |
6d0c48ae | 928 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 929 | * @bio: bio to add pages to |
6d0c48ae JA |
930 | * @iter: iov iterator describing the region to be added |
931 | * | |
932 | * This takes either an iterator pointing to user memory, or one pointing to | |
933 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
934 | * map them into the kernel. On IO completion, the caller should put those | |
399254aa JA |
935 | * pages. If we're adding kernel pages, and the caller told us it's safe to |
936 | * do so, we just have to add the pages to the bio directly. We don't grab an | |
937 | * extra reference to those pages (the user should already have that), and we | |
938 | * don't put the page on IO completion. The caller needs to check if the bio is | |
939 | * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be | |
940 | * released. | |
17d51b10 | 941 | * |
17d51b10 | 942 | * The function tries, but does not guarantee, to pin as many pages as |
6d0c48ae JA |
943 | * fit into the bio, or are requested in *iter, whatever is smaller. If |
944 | * MM encounters an error pinning the requested pages, it stops. Error | |
945 | * is returned only if 0 pages could be pinned. | |
17d51b10 MW |
946 | */ |
947 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
948 | { | |
6d0c48ae | 949 | const bool is_bvec = iov_iter_is_bvec(iter); |
14eacf12 CH |
950 | int ret; |
951 | ||
952 | if (WARN_ON_ONCE(bio->bi_vcnt)) | |
953 | return -EINVAL; | |
17d51b10 MW |
954 | |
955 | do { | |
6d0c48ae JA |
956 | if (is_bvec) |
957 | ret = __bio_iov_bvec_add_pages(bio, iter); | |
958 | else | |
959 | ret = __bio_iov_iter_get_pages(bio, iter); | |
79d08f89 | 960 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
17d51b10 | 961 | |
b6207430 | 962 | if (is_bvec) |
7321ecbf | 963 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
14eacf12 | 964 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 965 | } |
2cefe4db | 966 | |
4246a0b6 | 967 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 968 | { |
65e53aab | 969 | complete(bio->bi_private); |
9e882242 KO |
970 | } |
971 | ||
972 | /** | |
973 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
974 | * @bio: The &struct bio which describes the I/O |
975 | * | |
976 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
977 | * bio_endio() on failure. | |
3d289d68 JK |
978 | * |
979 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
980 | * result in bio reference to be consumed. The caller must drop the reference | |
981 | * on his own. | |
9e882242 | 982 | */ |
4e49ea4a | 983 | int submit_bio_wait(struct bio *bio) |
9e882242 | 984 | { |
e319e1fb | 985 | DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); |
9e882242 | 986 | |
65e53aab | 987 | bio->bi_private = &done; |
9e882242 | 988 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 989 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 990 | submit_bio(bio); |
65e53aab | 991 | wait_for_completion_io(&done); |
9e882242 | 992 | |
65e53aab | 993 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
994 | } |
995 | EXPORT_SYMBOL(submit_bio_wait); | |
996 | ||
054bdf64 KO |
997 | /** |
998 | * bio_advance - increment/complete a bio by some number of bytes | |
999 | * @bio: bio to advance | |
1000 | * @bytes: number of bytes to complete | |
1001 | * | |
1002 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
1003 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
1004 | * be updated on the last bvec as well. | |
1005 | * | |
1006 | * @bio will then represent the remaining, uncompleted portion of the io. | |
1007 | */ | |
1008 | void bio_advance(struct bio *bio, unsigned bytes) | |
1009 | { | |
1010 | if (bio_integrity(bio)) | |
1011 | bio_integrity_advance(bio, bytes); | |
1012 | ||
4550dd6c | 1013 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
1014 | } |
1015 | EXPORT_SYMBOL(bio_advance); | |
1016 | ||
45db54d5 KO |
1017 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1018 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1019 | { |
1cb9dda4 | 1020 | struct bio_vec src_bv, dst_bv; |
16ac3d63 | 1021 | void *src_p, *dst_p; |
1cb9dda4 | 1022 | unsigned bytes; |
16ac3d63 | 1023 | |
45db54d5 KO |
1024 | while (src_iter->bi_size && dst_iter->bi_size) { |
1025 | src_bv = bio_iter_iovec(src, *src_iter); | |
1026 | dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1cb9dda4 KO |
1027 | |
1028 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1029 | |
1cb9dda4 KO |
1030 | src_p = kmap_atomic(src_bv.bv_page); |
1031 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1032 | |
1cb9dda4 KO |
1033 | memcpy(dst_p + dst_bv.bv_offset, |
1034 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1035 | bytes); |
1036 | ||
1037 | kunmap_atomic(dst_p); | |
1038 | kunmap_atomic(src_p); | |
1039 | ||
6e6e811d KO |
1040 | flush_dcache_page(dst_bv.bv_page); |
1041 | ||
45db54d5 KO |
1042 | bio_advance_iter(src, src_iter, bytes); |
1043 | bio_advance_iter(dst, dst_iter, bytes); | |
16ac3d63 KO |
1044 | } |
1045 | } | |
38a72dac KO |
1046 | EXPORT_SYMBOL(bio_copy_data_iter); |
1047 | ||
1048 | /** | |
45db54d5 KO |
1049 | * bio_copy_data - copy contents of data buffers from one bio to another |
1050 | * @src: source bio | |
1051 | * @dst: destination bio | |
38a72dac KO |
1052 | * |
1053 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1054 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1055 | */ | |
1056 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1057 | { | |
45db54d5 KO |
1058 | struct bvec_iter src_iter = src->bi_iter; |
1059 | struct bvec_iter dst_iter = dst->bi_iter; | |
1060 | ||
1061 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1062 | } |
16ac3d63 KO |
1063 | EXPORT_SYMBOL(bio_copy_data); |
1064 | ||
45db54d5 KO |
1065 | /** |
1066 | * bio_list_copy_data - copy contents of data buffers from one chain of bios to | |
1067 | * another | |
1068 | * @src: source bio list | |
1069 | * @dst: destination bio list | |
1070 | * | |
1071 | * Stops when it reaches the end of either the @src list or @dst list - that is, | |
1072 | * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of | |
1073 | * bios). | |
1074 | */ | |
1075 | void bio_list_copy_data(struct bio *dst, struct bio *src) | |
1076 | { | |
1077 | struct bvec_iter src_iter = src->bi_iter; | |
1078 | struct bvec_iter dst_iter = dst->bi_iter; | |
1079 | ||
1080 | while (1) { | |
1081 | if (!src_iter.bi_size) { | |
1082 | src = src->bi_next; | |
1083 | if (!src) | |
1084 | break; | |
1085 | ||
1086 | src_iter = src->bi_iter; | |
1087 | } | |
1088 | ||
1089 | if (!dst_iter.bi_size) { | |
1090 | dst = dst->bi_next; | |
1091 | if (!dst) | |
1092 | break; | |
1093 | ||
1094 | dst_iter = dst->bi_iter; | |
1095 | } | |
1096 | ||
1097 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
1098 | } | |
1099 | } | |
1100 | EXPORT_SYMBOL(bio_list_copy_data); | |
1101 | ||
1da177e4 | 1102 | struct bio_map_data { |
152e283f | 1103 | int is_our_pages; |
26e49cfc KO |
1104 | struct iov_iter iter; |
1105 | struct iovec iov[]; | |
1da177e4 LT |
1106 | }; |
1107 | ||
0e5b935d | 1108 | static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, |
76029ff3 | 1109 | gfp_t gfp_mask) |
1da177e4 | 1110 | { |
0e5b935d AV |
1111 | struct bio_map_data *bmd; |
1112 | if (data->nr_segs > UIO_MAXIOV) | |
f3f63c1c | 1113 | return NULL; |
1da177e4 | 1114 | |
f1f8f292 | 1115 | bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); |
0e5b935d AV |
1116 | if (!bmd) |
1117 | return NULL; | |
1118 | memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); | |
1119 | bmd->iter = *data; | |
1120 | bmd->iter.iov = bmd->iov; | |
1121 | return bmd; | |
1da177e4 LT |
1122 | } |
1123 | ||
9124d3fe DP |
1124 | /** |
1125 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
1126 | * @bio: The &struct bio which describes the I/O as destination | |
1127 | * @iter: iov_iter as source | |
1128 | * | |
1129 | * Copy all pages from iov_iter to bio. | |
1130 | * Returns 0 on success, or error on failure. | |
1131 | */ | |
98a09d61 | 1132 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) |
c5dec1c3 | 1133 | { |
c5dec1c3 | 1134 | struct bio_vec *bvec; |
6dc4f100 | 1135 | struct bvec_iter_all iter_all; |
c5dec1c3 | 1136 | |
2b070cfe | 1137 | bio_for_each_segment_all(bvec, bio, iter_all) { |
9124d3fe | 1138 | ssize_t ret; |
c5dec1c3 | 1139 | |
9124d3fe DP |
1140 | ret = copy_page_from_iter(bvec->bv_page, |
1141 | bvec->bv_offset, | |
1142 | bvec->bv_len, | |
98a09d61 | 1143 | iter); |
9124d3fe | 1144 | |
98a09d61 | 1145 | if (!iov_iter_count(iter)) |
9124d3fe DP |
1146 | break; |
1147 | ||
1148 | if (ret < bvec->bv_len) | |
1149 | return -EFAULT; | |
c5dec1c3 FT |
1150 | } |
1151 | ||
9124d3fe DP |
1152 | return 0; |
1153 | } | |
1154 | ||
1155 | /** | |
1156 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
1157 | * @bio: The &struct bio which describes the I/O as source | |
1158 | * @iter: iov_iter as destination | |
1159 | * | |
1160 | * Copy all pages from bio to iov_iter. | |
1161 | * Returns 0 on success, or error on failure. | |
1162 | */ | |
1163 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
1164 | { | |
9124d3fe | 1165 | struct bio_vec *bvec; |
6dc4f100 | 1166 | struct bvec_iter_all iter_all; |
9124d3fe | 1167 | |
2b070cfe | 1168 | bio_for_each_segment_all(bvec, bio, iter_all) { |
9124d3fe DP |
1169 | ssize_t ret; |
1170 | ||
1171 | ret = copy_page_to_iter(bvec->bv_page, | |
1172 | bvec->bv_offset, | |
1173 | bvec->bv_len, | |
1174 | &iter); | |
1175 | ||
1176 | if (!iov_iter_count(&iter)) | |
1177 | break; | |
1178 | ||
1179 | if (ret < bvec->bv_len) | |
1180 | return -EFAULT; | |
1181 | } | |
1182 | ||
1183 | return 0; | |
c5dec1c3 FT |
1184 | } |
1185 | ||
491221f8 | 1186 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1187 | { |
1188 | struct bio_vec *bvec; | |
6dc4f100 | 1189 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1190 | |
2b070cfe | 1191 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1192 | __free_page(bvec->bv_page); |
1193 | } | |
491221f8 | 1194 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1195 | |
1da177e4 LT |
1196 | /** |
1197 | * bio_uncopy_user - finish previously mapped bio | |
1198 | * @bio: bio being terminated | |
1199 | * | |
ddad8dd0 | 1200 | * Free pages allocated from bio_copy_user_iov() and write back data |
1da177e4 LT |
1201 | * to user space in case of a read. |
1202 | */ | |
1203 | int bio_uncopy_user(struct bio *bio) | |
1204 | { | |
1205 | struct bio_map_data *bmd = bio->bi_private; | |
1dfa0f68 | 1206 | int ret = 0; |
1da177e4 | 1207 | |
35dc2483 RD |
1208 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { |
1209 | /* | |
1210 | * if we're in a workqueue, the request is orphaned, so | |
2d99b55d HR |
1211 | * don't copy into a random user address space, just free |
1212 | * and return -EINTR so user space doesn't expect any data. | |
35dc2483 | 1213 | */ |
2d99b55d HR |
1214 | if (!current->mm) |
1215 | ret = -EINTR; | |
1216 | else if (bio_data_dir(bio) == READ) | |
9124d3fe | 1217 | ret = bio_copy_to_iter(bio, bmd->iter); |
1dfa0f68 CH |
1218 | if (bmd->is_our_pages) |
1219 | bio_free_pages(bio); | |
35dc2483 | 1220 | } |
c8db4448 | 1221 | kfree(bmd); |
1da177e4 LT |
1222 | bio_put(bio); |
1223 | return ret; | |
1224 | } | |
1225 | ||
1226 | /** | |
c5dec1c3 | 1227 | * bio_copy_user_iov - copy user data to bio |
26e49cfc KO |
1228 | * @q: destination block queue |
1229 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1230 | * @iter: iovec iterator | |
1231 | * @gfp_mask: memory allocation flags | |
1da177e4 LT |
1232 | * |
1233 | * Prepares and returns a bio for indirect user io, bouncing data | |
1234 | * to/from kernel pages as necessary. Must be paired with | |
1235 | * call bio_uncopy_user() on io completion. | |
1236 | */ | |
152e283f FT |
1237 | struct bio *bio_copy_user_iov(struct request_queue *q, |
1238 | struct rq_map_data *map_data, | |
e81cef5d | 1239 | struct iov_iter *iter, |
26e49cfc | 1240 | gfp_t gfp_mask) |
1da177e4 | 1241 | { |
1da177e4 | 1242 | struct bio_map_data *bmd; |
1da177e4 LT |
1243 | struct page *page; |
1244 | struct bio *bio; | |
d16d44eb AV |
1245 | int i = 0, ret; |
1246 | int nr_pages; | |
26e49cfc | 1247 | unsigned int len = iter->count; |
bd5cecea | 1248 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; |
1da177e4 | 1249 | |
0e5b935d | 1250 | bmd = bio_alloc_map_data(iter, gfp_mask); |
1da177e4 LT |
1251 | if (!bmd) |
1252 | return ERR_PTR(-ENOMEM); | |
1253 | ||
26e49cfc KO |
1254 | /* |
1255 | * We need to do a deep copy of the iov_iter including the iovecs. | |
1256 | * The caller provided iov might point to an on-stack or otherwise | |
1257 | * shortlived one. | |
1258 | */ | |
1259 | bmd->is_our_pages = map_data ? 0 : 1; | |
26e49cfc | 1260 | |
d16d44eb AV |
1261 | nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); |
1262 | if (nr_pages > BIO_MAX_PAGES) | |
1263 | nr_pages = BIO_MAX_PAGES; | |
26e49cfc | 1264 | |
1da177e4 | 1265 | ret = -ENOMEM; |
a9e9dc24 | 1266 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1267 | if (!bio) |
1268 | goto out_bmd; | |
1269 | ||
1da177e4 | 1270 | ret = 0; |
56c451f4 FT |
1271 | |
1272 | if (map_data) { | |
e623ddb4 | 1273 | nr_pages = 1 << map_data->page_order; |
56c451f4 FT |
1274 | i = map_data->offset / PAGE_SIZE; |
1275 | } | |
1da177e4 | 1276 | while (len) { |
e623ddb4 | 1277 | unsigned int bytes = PAGE_SIZE; |
1da177e4 | 1278 | |
56c451f4 FT |
1279 | bytes -= offset; |
1280 | ||
1da177e4 LT |
1281 | if (bytes > len) |
1282 | bytes = len; | |
1283 | ||
152e283f | 1284 | if (map_data) { |
e623ddb4 | 1285 | if (i == map_data->nr_entries * nr_pages) { |
152e283f FT |
1286 | ret = -ENOMEM; |
1287 | break; | |
1288 | } | |
e623ddb4 FT |
1289 | |
1290 | page = map_data->pages[i / nr_pages]; | |
1291 | page += (i % nr_pages); | |
1292 | ||
1293 | i++; | |
1294 | } else { | |
152e283f | 1295 | page = alloc_page(q->bounce_gfp | gfp_mask); |
e623ddb4 FT |
1296 | if (!page) { |
1297 | ret = -ENOMEM; | |
1298 | break; | |
1299 | } | |
1da177e4 LT |
1300 | } |
1301 | ||
a3761c3c JG |
1302 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { |
1303 | if (!map_data) | |
1304 | __free_page(page); | |
1da177e4 | 1305 | break; |
a3761c3c | 1306 | } |
1da177e4 LT |
1307 | |
1308 | len -= bytes; | |
56c451f4 | 1309 | offset = 0; |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | if (ret) | |
1313 | goto cleanup; | |
1314 | ||
2884d0be AV |
1315 | if (map_data) |
1316 | map_data->offset += bio->bi_iter.bi_size; | |
1317 | ||
1da177e4 LT |
1318 | /* |
1319 | * success | |
1320 | */ | |
00e23707 | 1321 | if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || |
ecb554a8 | 1322 | (map_data && map_data->from_user)) { |
98a09d61 | 1323 | ret = bio_copy_from_iter(bio, iter); |
c5dec1c3 FT |
1324 | if (ret) |
1325 | goto cleanup; | |
98a09d61 | 1326 | } else { |
f55adad6 KB |
1327 | if (bmd->is_our_pages) |
1328 | zero_fill_bio(bio); | |
98a09d61 | 1329 | iov_iter_advance(iter, bio->bi_iter.bi_size); |
1da177e4 LT |
1330 | } |
1331 | ||
26e49cfc | 1332 | bio->bi_private = bmd; |
2884d0be AV |
1333 | if (map_data && map_data->null_mapped) |
1334 | bio_set_flag(bio, BIO_NULL_MAPPED); | |
1da177e4 LT |
1335 | return bio; |
1336 | cleanup: | |
152e283f | 1337 | if (!map_data) |
1dfa0f68 | 1338 | bio_free_pages(bio); |
1da177e4 LT |
1339 | bio_put(bio); |
1340 | out_bmd: | |
c8db4448 | 1341 | kfree(bmd); |
1da177e4 LT |
1342 | return ERR_PTR(ret); |
1343 | } | |
1344 | ||
37f19e57 CH |
1345 | /** |
1346 | * bio_map_user_iov - map user iovec into bio | |
1347 | * @q: the struct request_queue for the bio | |
1348 | * @iter: iovec iterator | |
1349 | * @gfp_mask: memory allocation flags | |
1350 | * | |
1351 | * Map the user space address into a bio suitable for io to a block | |
1352 | * device. Returns an error pointer in case of error. | |
1353 | */ | |
1354 | struct bio *bio_map_user_iov(struct request_queue *q, | |
e81cef5d | 1355 | struct iov_iter *iter, |
37f19e57 | 1356 | gfp_t gfp_mask) |
1da177e4 | 1357 | { |
26e49cfc | 1358 | int j; |
1da177e4 | 1359 | struct bio *bio; |
076098e5 | 1360 | int ret; |
1da177e4 | 1361 | |
b282cc76 | 1362 | if (!iov_iter_count(iter)) |
1da177e4 LT |
1363 | return ERR_PTR(-EINVAL); |
1364 | ||
b282cc76 | 1365 | bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); |
1da177e4 LT |
1366 | if (!bio) |
1367 | return ERR_PTR(-ENOMEM); | |
1368 | ||
0a0f1513 | 1369 | while (iov_iter_count(iter)) { |
629e42bc | 1370 | struct page **pages; |
076098e5 AV |
1371 | ssize_t bytes; |
1372 | size_t offs, added = 0; | |
1373 | int npages; | |
1da177e4 | 1374 | |
0a0f1513 | 1375 | bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); |
076098e5 AV |
1376 | if (unlikely(bytes <= 0)) { |
1377 | ret = bytes ? bytes : -EFAULT; | |
f1970baf | 1378 | goto out_unmap; |
99172157 | 1379 | } |
f1970baf | 1380 | |
076098e5 | 1381 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); |
f1970baf | 1382 | |
98f0bc99 AV |
1383 | if (unlikely(offs & queue_dma_alignment(q))) { |
1384 | ret = -EINVAL; | |
1385 | j = 0; | |
1386 | } else { | |
1387 | for (j = 0; j < npages; j++) { | |
1388 | struct page *page = pages[j]; | |
1389 | unsigned int n = PAGE_SIZE - offs; | |
f1970baf | 1390 | |
98f0bc99 AV |
1391 | if (n > bytes) |
1392 | n = bytes; | |
95d78c28 | 1393 | |
19047087 ML |
1394 | if (!__bio_add_pc_page(q, bio, page, n, offs, |
1395 | true)) | |
98f0bc99 | 1396 | break; |
1da177e4 | 1397 | |
98f0bc99 AV |
1398 | added += n; |
1399 | bytes -= n; | |
1400 | offs = 0; | |
1401 | } | |
0a0f1513 | 1402 | iov_iter_advance(iter, added); |
f1970baf | 1403 | } |
1da177e4 | 1404 | /* |
f1970baf | 1405 | * release the pages we didn't map into the bio, if any |
1da177e4 | 1406 | */ |
629e42bc | 1407 | while (j < npages) |
09cbfeaf | 1408 | put_page(pages[j++]); |
629e42bc | 1409 | kvfree(pages); |
e2e115d1 AV |
1410 | /* couldn't stuff something into bio? */ |
1411 | if (bytes) | |
1412 | break; | |
1da177e4 LT |
1413 | } |
1414 | ||
b7c44ed9 | 1415 | bio_set_flag(bio, BIO_USER_MAPPED); |
37f19e57 CH |
1416 | |
1417 | /* | |
5fad1b64 | 1418 | * subtle -- if bio_map_user_iov() ended up bouncing a bio, |
37f19e57 CH |
1419 | * it would normally disappear when its bi_end_io is run. |
1420 | * however, we need it for the unmap, so grab an extra | |
1421 | * reference to it | |
1422 | */ | |
1423 | bio_get(bio); | |
1da177e4 | 1424 | return bio; |
f1970baf JB |
1425 | |
1426 | out_unmap: | |
506e0798 | 1427 | bio_release_pages(bio, false); |
1da177e4 LT |
1428 | bio_put(bio); |
1429 | return ERR_PTR(ret); | |
1430 | } | |
1431 | ||
1da177e4 LT |
1432 | /** |
1433 | * bio_unmap_user - unmap a bio | |
1434 | * @bio: the bio being unmapped | |
1435 | * | |
5fad1b64 BVA |
1436 | * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from |
1437 | * process context. | |
1da177e4 LT |
1438 | * |
1439 | * bio_unmap_user() may sleep. | |
1440 | */ | |
1441 | void bio_unmap_user(struct bio *bio) | |
1442 | { | |
163cc2d3 CH |
1443 | bio_release_pages(bio, bio_data_dir(bio) == READ); |
1444 | bio_put(bio); | |
1da177e4 LT |
1445 | bio_put(bio); |
1446 | } | |
1447 | ||
b4c5875d DLM |
1448 | static void bio_invalidate_vmalloc_pages(struct bio *bio) |
1449 | { | |
1450 | #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE | |
1451 | if (bio->bi_private && !op_is_write(bio_op(bio))) { | |
1452 | unsigned long i, len = 0; | |
1453 | ||
1454 | for (i = 0; i < bio->bi_vcnt; i++) | |
1455 | len += bio->bi_io_vec[i].bv_len; | |
1456 | invalidate_kernel_vmap_range(bio->bi_private, len); | |
1457 | } | |
1458 | #endif | |
1459 | } | |
1460 | ||
4246a0b6 | 1461 | static void bio_map_kern_endio(struct bio *bio) |
b823825e | 1462 | { |
b4c5875d | 1463 | bio_invalidate_vmalloc_pages(bio); |
b823825e | 1464 | bio_put(bio); |
b823825e JA |
1465 | } |
1466 | ||
75c72b83 CH |
1467 | /** |
1468 | * bio_map_kern - map kernel address into bio | |
1469 | * @q: the struct request_queue for the bio | |
1470 | * @data: pointer to buffer to map | |
1471 | * @len: length in bytes | |
1472 | * @gfp_mask: allocation flags for bio allocation | |
1473 | * | |
1474 | * Map the kernel address into a bio suitable for io to a block | |
1475 | * device. Returns an error pointer in case of error. | |
1476 | */ | |
1477 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1478 | gfp_t gfp_mask) | |
df46b9a4 MC |
1479 | { |
1480 | unsigned long kaddr = (unsigned long)data; | |
1481 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1482 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1483 | const int nr_pages = end - start; | |
b4c5875d DLM |
1484 | bool is_vmalloc = is_vmalloc_addr(data); |
1485 | struct page *page; | |
df46b9a4 MC |
1486 | int offset, i; |
1487 | struct bio *bio; | |
1488 | ||
a9e9dc24 | 1489 | bio = bio_kmalloc(gfp_mask, nr_pages); |
df46b9a4 MC |
1490 | if (!bio) |
1491 | return ERR_PTR(-ENOMEM); | |
1492 | ||
b4c5875d DLM |
1493 | if (is_vmalloc) { |
1494 | flush_kernel_vmap_range(data, len); | |
1495 | bio->bi_private = data; | |
1496 | } | |
1497 | ||
df46b9a4 MC |
1498 | offset = offset_in_page(kaddr); |
1499 | for (i = 0; i < nr_pages; i++) { | |
1500 | unsigned int bytes = PAGE_SIZE - offset; | |
1501 | ||
1502 | if (len <= 0) | |
1503 | break; | |
1504 | ||
1505 | if (bytes > len) | |
1506 | bytes = len; | |
1507 | ||
b4c5875d DLM |
1508 | if (!is_vmalloc) |
1509 | page = virt_to_page(data); | |
1510 | else | |
1511 | page = vmalloc_to_page(data); | |
1512 | if (bio_add_pc_page(q, bio, page, bytes, | |
75c72b83 CH |
1513 | offset) < bytes) { |
1514 | /* we don't support partial mappings */ | |
1515 | bio_put(bio); | |
1516 | return ERR_PTR(-EINVAL); | |
1517 | } | |
df46b9a4 MC |
1518 | |
1519 | data += bytes; | |
1520 | len -= bytes; | |
1521 | offset = 0; | |
1522 | } | |
1523 | ||
b823825e | 1524 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1525 | return bio; |
1526 | } | |
df46b9a4 | 1527 | |
4246a0b6 | 1528 | static void bio_copy_kern_endio(struct bio *bio) |
68154e90 | 1529 | { |
1dfa0f68 CH |
1530 | bio_free_pages(bio); |
1531 | bio_put(bio); | |
1532 | } | |
1533 | ||
4246a0b6 | 1534 | static void bio_copy_kern_endio_read(struct bio *bio) |
1dfa0f68 | 1535 | { |
42d2683a | 1536 | char *p = bio->bi_private; |
1dfa0f68 | 1537 | struct bio_vec *bvec; |
6dc4f100 | 1538 | struct bvec_iter_all iter_all; |
68154e90 | 1539 | |
2b070cfe | 1540 | bio_for_each_segment_all(bvec, bio, iter_all) { |
1dfa0f68 | 1541 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); |
c8db4448 | 1542 | p += bvec->bv_len; |
68154e90 FT |
1543 | } |
1544 | ||
4246a0b6 | 1545 | bio_copy_kern_endio(bio); |
68154e90 FT |
1546 | } |
1547 | ||
1548 | /** | |
1549 | * bio_copy_kern - copy kernel address into bio | |
1550 | * @q: the struct request_queue for the bio | |
1551 | * @data: pointer to buffer to copy | |
1552 | * @len: length in bytes | |
1553 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1554 | * @reading: data direction is READ |
68154e90 FT |
1555 | * |
1556 | * copy the kernel address into a bio suitable for io to a block | |
1557 | * device. Returns an error pointer in case of error. | |
1558 | */ | |
1559 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1560 | gfp_t gfp_mask, int reading) | |
1561 | { | |
42d2683a CH |
1562 | unsigned long kaddr = (unsigned long)data; |
1563 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1564 | unsigned long start = kaddr >> PAGE_SHIFT; | |
42d2683a CH |
1565 | struct bio *bio; |
1566 | void *p = data; | |
1dfa0f68 | 1567 | int nr_pages = 0; |
68154e90 | 1568 | |
42d2683a CH |
1569 | /* |
1570 | * Overflow, abort | |
1571 | */ | |
1572 | if (end < start) | |
1573 | return ERR_PTR(-EINVAL); | |
68154e90 | 1574 | |
42d2683a CH |
1575 | nr_pages = end - start; |
1576 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1577 | if (!bio) | |
1578 | return ERR_PTR(-ENOMEM); | |
68154e90 | 1579 | |
42d2683a CH |
1580 | while (len) { |
1581 | struct page *page; | |
1582 | unsigned int bytes = PAGE_SIZE; | |
68154e90 | 1583 | |
42d2683a CH |
1584 | if (bytes > len) |
1585 | bytes = len; | |
1586 | ||
1587 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1588 | if (!page) | |
1589 | goto cleanup; | |
1590 | ||
1591 | if (!reading) | |
1592 | memcpy(page_address(page), p, bytes); | |
1593 | ||
1594 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
1595 | break; | |
1596 | ||
1597 | len -= bytes; | |
1598 | p += bytes; | |
68154e90 FT |
1599 | } |
1600 | ||
1dfa0f68 CH |
1601 | if (reading) { |
1602 | bio->bi_end_io = bio_copy_kern_endio_read; | |
1603 | bio->bi_private = data; | |
1604 | } else { | |
1605 | bio->bi_end_io = bio_copy_kern_endio; | |
1dfa0f68 | 1606 | } |
76029ff3 | 1607 | |
68154e90 | 1608 | return bio; |
42d2683a CH |
1609 | |
1610 | cleanup: | |
1dfa0f68 | 1611 | bio_free_pages(bio); |
42d2683a CH |
1612 | bio_put(bio); |
1613 | return ERR_PTR(-ENOMEM); | |
68154e90 FT |
1614 | } |
1615 | ||
1da177e4 LT |
1616 | /* |
1617 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1618 | * for performing direct-IO in BIOs. | |
1619 | * | |
1620 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1621 | * because the required locks are not interrupt-safe. So what we can do is to | |
1622 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1623 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1624 | * in process context. | |
1625 | * | |
1626 | * We special-case compound pages here: normally this means reads into hugetlb | |
1627 | * pages. The logic in here doesn't really work right for compound pages | |
1628 | * because the VM does not uniformly chase down the head page in all cases. | |
1629 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1630 | * handle them at all. So we skip compound pages here at an early stage. | |
1631 | * | |
1632 | * Note that this code is very hard to test under normal circumstances because | |
1633 | * direct-io pins the pages with get_user_pages(). This makes | |
1634 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1635 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1636 | * pagecache. |
1637 | * | |
1638 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1639 | * deferred bio dirtying paths. | |
1640 | */ | |
1641 | ||
1642 | /* | |
1643 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1644 | */ | |
1645 | void bio_set_pages_dirty(struct bio *bio) | |
1646 | { | |
cb34e057 | 1647 | struct bio_vec *bvec; |
6dc4f100 | 1648 | struct bvec_iter_all iter_all; |
1da177e4 | 1649 | |
2b070cfe | 1650 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1651 | if (!PageCompound(bvec->bv_page)) |
1652 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1653 | } |
1654 | } | |
1655 | ||
1da177e4 LT |
1656 | /* |
1657 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1658 | * If they are, then fine. If, however, some pages are clean then they must | |
1659 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1660 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1661 | * |
1662 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1663 | * here on. It will run one put_page() against each page and will run one |
1664 | * bio_put() against the BIO. | |
1da177e4 LT |
1665 | */ |
1666 | ||
65f27f38 | 1667 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1668 | |
65f27f38 | 1669 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1670 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1671 | static struct bio *bio_dirty_list; | |
1672 | ||
1673 | /* | |
1674 | * This runs in process context | |
1675 | */ | |
65f27f38 | 1676 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1677 | { |
24d5493f | 1678 | struct bio *bio, *next; |
1da177e4 | 1679 | |
24d5493f CH |
1680 | spin_lock_irq(&bio_dirty_lock); |
1681 | next = bio_dirty_list; | |
1da177e4 | 1682 | bio_dirty_list = NULL; |
24d5493f | 1683 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1684 | |
24d5493f CH |
1685 | while ((bio = next) != NULL) { |
1686 | next = bio->bi_private; | |
1da177e4 | 1687 | |
d241a95f | 1688 | bio_release_pages(bio, true); |
1da177e4 | 1689 | bio_put(bio); |
1da177e4 LT |
1690 | } |
1691 | } | |
1692 | ||
1693 | void bio_check_pages_dirty(struct bio *bio) | |
1694 | { | |
cb34e057 | 1695 | struct bio_vec *bvec; |
24d5493f | 1696 | unsigned long flags; |
6dc4f100 | 1697 | struct bvec_iter_all iter_all; |
1da177e4 | 1698 | |
2b070cfe | 1699 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1700 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1701 | goto defer; | |
1da177e4 LT |
1702 | } |
1703 | ||
d241a95f | 1704 | bio_release_pages(bio, false); |
24d5493f CH |
1705 | bio_put(bio); |
1706 | return; | |
1707 | defer: | |
1708 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1709 | bio->bi_private = bio_dirty_list; | |
1710 | bio_dirty_list = bio; | |
1711 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1712 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1713 | } |
1714 | ||
5b18b5a7 MP |
1715 | void update_io_ticks(struct hd_struct *part, unsigned long now) |
1716 | { | |
1717 | unsigned long stamp; | |
1718 | again: | |
1719 | stamp = READ_ONCE(part->stamp); | |
1720 | if (unlikely(stamp != now)) { | |
1721 | if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { | |
1722 | __part_stat_add(part, io_ticks, 1); | |
1723 | } | |
1724 | } | |
1725 | if (part->partno) { | |
1726 | part = &part_to_disk(part)->part0; | |
1727 | goto again; | |
1728 | } | |
1729 | } | |
1da177e4 | 1730 | |
ddcf35d3 | 1731 | void generic_start_io_acct(struct request_queue *q, int op, |
d62e26b3 | 1732 | unsigned long sectors, struct hd_struct *part) |
394ffa50 | 1733 | { |
ddcf35d3 | 1734 | const int sgrp = op_stat_group(op); |
394ffa50 | 1735 | |
112f158f MS |
1736 | part_stat_lock(); |
1737 | ||
5b18b5a7 | 1738 | update_io_ticks(part, jiffies); |
112f158f MS |
1739 | part_stat_inc(part, ios[sgrp]); |
1740 | part_stat_add(part, sectors[sgrp], sectors); | |
ddcf35d3 | 1741 | part_inc_in_flight(q, part, op_is_write(op)); |
394ffa50 GZ |
1742 | |
1743 | part_stat_unlock(); | |
1744 | } | |
1745 | EXPORT_SYMBOL(generic_start_io_acct); | |
1746 | ||
ddcf35d3 | 1747 | void generic_end_io_acct(struct request_queue *q, int req_op, |
d62e26b3 | 1748 | struct hd_struct *part, unsigned long start_time) |
394ffa50 | 1749 | { |
5b18b5a7 MP |
1750 | unsigned long now = jiffies; |
1751 | unsigned long duration = now - start_time; | |
ddcf35d3 | 1752 | const int sgrp = op_stat_group(req_op); |
394ffa50 | 1753 | |
112f158f MS |
1754 | part_stat_lock(); |
1755 | ||
5b18b5a7 | 1756 | update_io_ticks(part, now); |
112f158f | 1757 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); |
5b18b5a7 | 1758 | part_stat_add(part, time_in_queue, duration); |
ddcf35d3 | 1759 | part_dec_in_flight(q, part, op_is_write(req_op)); |
394ffa50 GZ |
1760 | |
1761 | part_stat_unlock(); | |
1762 | } | |
1763 | EXPORT_SYMBOL(generic_end_io_acct); | |
1764 | ||
c4cf5261 JA |
1765 | static inline bool bio_remaining_done(struct bio *bio) |
1766 | { | |
1767 | /* | |
1768 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1769 | * we always end io on the first invocation. | |
1770 | */ | |
1771 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1772 | return true; | |
1773 | ||
1774 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1775 | ||
326e1dbb | 1776 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1777 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1778 | return true; |
326e1dbb | 1779 | } |
c4cf5261 JA |
1780 | |
1781 | return false; | |
1782 | } | |
1783 | ||
1da177e4 LT |
1784 | /** |
1785 | * bio_endio - end I/O on a bio | |
1786 | * @bio: bio | |
1da177e4 LT |
1787 | * |
1788 | * Description: | |
4246a0b6 CH |
1789 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1790 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1791 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1792 | * |
1793 | * bio_endio() can be called several times on a bio that has been chained | |
1794 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
1795 | * last time. At this point the BLK_TA_COMPLETE tracing event will be | |
1796 | * generated if BIO_TRACE_COMPLETION is set. | |
1da177e4 | 1797 | **/ |
4246a0b6 | 1798 | void bio_endio(struct bio *bio) |
1da177e4 | 1799 | { |
ba8c6967 | 1800 | again: |
2b885517 | 1801 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1802 | return; |
7c20f116 CH |
1803 | if (!bio_integrity_endio(bio)) |
1804 | return; | |
1da177e4 | 1805 | |
67b42d0b JB |
1806 | if (bio->bi_disk) |
1807 | rq_qos_done_bio(bio->bi_disk->queue, bio); | |
1808 | ||
ba8c6967 CH |
1809 | /* |
1810 | * Need to have a real endio function for chained bios, otherwise | |
1811 | * various corner cases will break (like stacking block devices that | |
1812 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1813 | * recursion and blowing the stack. Tail call optimization would | |
1814 | * handle this, but compiling with frame pointers also disables | |
1815 | * gcc's sibling call optimization. | |
1816 | */ | |
1817 | if (bio->bi_end_io == bio_chain_endio) { | |
1818 | bio = __bio_chain_endio(bio); | |
1819 | goto again; | |
196d38bc | 1820 | } |
ba8c6967 | 1821 | |
74d46992 CH |
1822 | if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
1823 | trace_block_bio_complete(bio->bi_disk->queue, bio, | |
a462b950 | 1824 | blk_status_to_errno(bio->bi_status)); |
fbbaf700 N |
1825 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
1826 | } | |
1827 | ||
9e234eea | 1828 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1829 | /* release cgroup info */ |
1830 | bio_uninit(bio); | |
ba8c6967 CH |
1831 | if (bio->bi_end_io) |
1832 | bio->bi_end_io(bio); | |
1da177e4 | 1833 | } |
a112a71d | 1834 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1835 | |
20d0189b KO |
1836 | /** |
1837 | * bio_split - split a bio | |
1838 | * @bio: bio to split | |
1839 | * @sectors: number of sectors to split from the front of @bio | |
1840 | * @gfp: gfp mask | |
1841 | * @bs: bio set to allocate from | |
1842 | * | |
1843 | * Allocates and returns a new bio which represents @sectors from the start of | |
1844 | * @bio, and updates @bio to represent the remaining sectors. | |
1845 | * | |
f3f5da62 | 1846 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1847 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1848 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1849 | */ |
1850 | struct bio *bio_split(struct bio *bio, int sectors, | |
1851 | gfp_t gfp, struct bio_set *bs) | |
1852 | { | |
f341a4d3 | 1853 | struct bio *split; |
20d0189b KO |
1854 | |
1855 | BUG_ON(sectors <= 0); | |
1856 | BUG_ON(sectors >= bio_sectors(bio)); | |
1857 | ||
f9d03f96 | 1858 | split = bio_clone_fast(bio, gfp, bs); |
20d0189b KO |
1859 | if (!split) |
1860 | return NULL; | |
1861 | ||
1862 | split->bi_iter.bi_size = sectors << 9; | |
1863 | ||
1864 | if (bio_integrity(split)) | |
fbd08e76 | 1865 | bio_integrity_trim(split); |
20d0189b KO |
1866 | |
1867 | bio_advance(bio, split->bi_iter.bi_size); | |
1868 | ||
fbbaf700 | 1869 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1870 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1871 | |
20d0189b KO |
1872 | return split; |
1873 | } | |
1874 | EXPORT_SYMBOL(bio_split); | |
1875 | ||
6678d83f KO |
1876 | /** |
1877 | * bio_trim - trim a bio | |
1878 | * @bio: bio to trim | |
1879 | * @offset: number of sectors to trim from the front of @bio | |
1880 | * @size: size we want to trim @bio to, in sectors | |
1881 | */ | |
1882 | void bio_trim(struct bio *bio, int offset, int size) | |
1883 | { | |
1884 | /* 'bio' is a cloned bio which we need to trim to match | |
1885 | * the given offset and size. | |
6678d83f | 1886 | */ |
6678d83f KO |
1887 | |
1888 | size <<= 9; | |
4f024f37 | 1889 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1890 | return; |
1891 | ||
6678d83f | 1892 | bio_advance(bio, offset << 9); |
4f024f37 | 1893 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1894 | |
1895 | if (bio_integrity(bio)) | |
fbd08e76 | 1896 | bio_integrity_trim(bio); |
376a78ab | 1897 | |
6678d83f KO |
1898 | } |
1899 | EXPORT_SYMBOL_GPL(bio_trim); | |
1900 | ||
1da177e4 LT |
1901 | /* |
1902 | * create memory pools for biovec's in a bio_set. | |
1903 | * use the global biovec slabs created for general use. | |
1904 | */ | |
8aa6ba2f | 1905 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1906 | { |
ed996a52 | 1907 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
1da177e4 | 1908 | |
8aa6ba2f | 1909 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1910 | } |
1911 | ||
917a38c7 KO |
1912 | /* |
1913 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1914 | * | |
1915 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1916 | * kzalloc()). | |
1917 | */ | |
1918 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1919 | { |
df2cb6da KO |
1920 | if (bs->rescue_workqueue) |
1921 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1922 | bs->rescue_workqueue = NULL; |
df2cb6da | 1923 | |
8aa6ba2f KO |
1924 | mempool_exit(&bs->bio_pool); |
1925 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1926 | |
7878cba9 | 1927 | bioset_integrity_free(bs); |
917a38c7 KO |
1928 | if (bs->bio_slab) |
1929 | bio_put_slab(bs); | |
1930 | bs->bio_slab = NULL; | |
1931 | } | |
1932 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1933 | |
917a38c7 KO |
1934 | /** |
1935 | * bioset_init - Initialize a bio_set | |
dad08527 | 1936 | * @bs: pool to initialize |
917a38c7 KO |
1937 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1938 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1939 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1940 | * and %BIOSET_NEED_RESCUER | |
1941 | * | |
dad08527 KO |
1942 | * Description: |
1943 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1944 | * to ask for a number of bytes to be allocated in front of the bio. | |
1945 | * Front pad allocation is useful for embedding the bio inside | |
1946 | * another structure, to avoid allocating extra data to go with the bio. | |
1947 | * Note that the bio must be embedded at the END of that structure always, | |
1948 | * or things will break badly. | |
1949 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
1950 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
1951 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
1952 | * dispatch queued requests when the mempool runs out of space. | |
1953 | * | |
917a38c7 KO |
1954 | */ |
1955 | int bioset_init(struct bio_set *bs, | |
1956 | unsigned int pool_size, | |
1957 | unsigned int front_pad, | |
1958 | int flags) | |
1959 | { | |
1960 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1961 | ||
1962 | bs->front_pad = front_pad; | |
1963 | ||
1964 | spin_lock_init(&bs->rescue_lock); | |
1965 | bio_list_init(&bs->rescue_list); | |
1966 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1967 | ||
1968 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1969 | if (!bs->bio_slab) | |
1970 | return -ENOMEM; | |
1971 | ||
1972 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) | |
1973 | goto bad; | |
1974 | ||
1975 | if ((flags & BIOSET_NEED_BVECS) && | |
1976 | biovec_init_pool(&bs->bvec_pool, pool_size)) | |
1977 | goto bad; | |
1978 | ||
1979 | if (!(flags & BIOSET_NEED_RESCUER)) | |
1980 | return 0; | |
1981 | ||
1982 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1983 | if (!bs->rescue_workqueue) | |
1984 | goto bad; | |
1985 | ||
1986 | return 0; | |
1987 | bad: | |
1988 | bioset_exit(bs); | |
1989 | return -ENOMEM; | |
1990 | } | |
1991 | EXPORT_SYMBOL(bioset_init); | |
1992 | ||
28e89fd9 JA |
1993 | /* |
1994 | * Initialize and setup a new bio_set, based on the settings from | |
1995 | * another bio_set. | |
1996 | */ | |
1997 | int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) | |
1998 | { | |
1999 | int flags; | |
2000 | ||
2001 | flags = 0; | |
2002 | if (src->bvec_pool.min_nr) | |
2003 | flags |= BIOSET_NEED_BVECS; | |
2004 | if (src->rescue_workqueue) | |
2005 | flags |= BIOSET_NEED_RESCUER; | |
2006 | ||
2007 | return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); | |
2008 | } | |
2009 | EXPORT_SYMBOL(bioset_init_from_src); | |
2010 | ||
852c788f | 2011 | #ifdef CONFIG_BLK_CGROUP |
1d933cf0 | 2012 | |
74b7c02a | 2013 | /** |
2268c0fe | 2014 | * bio_disassociate_blkg - puts back the blkg reference if associated |
74b7c02a | 2015 | * @bio: target bio |
74b7c02a | 2016 | * |
2268c0fe | 2017 | * Helper to disassociate the blkg from @bio if a blkg is associated. |
74b7c02a | 2018 | */ |
2268c0fe | 2019 | void bio_disassociate_blkg(struct bio *bio) |
74b7c02a | 2020 | { |
2268c0fe DZ |
2021 | if (bio->bi_blkg) { |
2022 | blkg_put(bio->bi_blkg); | |
2023 | bio->bi_blkg = NULL; | |
2024 | } | |
74b7c02a | 2025 | } |
892ad71f | 2026 | EXPORT_SYMBOL_GPL(bio_disassociate_blkg); |
74b7c02a | 2027 | |
a7b39b4e | 2028 | /** |
2268c0fe | 2029 | * __bio_associate_blkg - associate a bio with the a blkg |
a7b39b4e | 2030 | * @bio: target bio |
b5f2954d | 2031 | * @blkg: the blkg to associate |
b5f2954d | 2032 | * |
beea9da0 DZ |
2033 | * This tries to associate @bio with the specified @blkg. Association failure |
2034 | * is handled by walking up the blkg tree. Therefore, the blkg associated can | |
2035 | * be anything between @blkg and the root_blkg. This situation only happens | |
2036 | * when a cgroup is dying and then the remaining bios will spill to the closest | |
2037 | * alive blkg. | |
a7b39b4e | 2038 | * |
beea9da0 DZ |
2039 | * A reference will be taken on the @blkg and will be released when @bio is |
2040 | * freed. | |
a7b39b4e | 2041 | */ |
2268c0fe | 2042 | static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) |
a7b39b4e | 2043 | { |
2268c0fe DZ |
2044 | bio_disassociate_blkg(bio); |
2045 | ||
7754f669 | 2046 | bio->bi_blkg = blkg_tryget_closest(blkg); |
a7b39b4e DZF |
2047 | } |
2048 | ||
d459d853 | 2049 | /** |
fd42df30 | 2050 | * bio_associate_blkg_from_css - associate a bio with a specified css |
d459d853 | 2051 | * @bio: target bio |
fd42df30 | 2052 | * @css: target css |
d459d853 | 2053 | * |
fd42df30 | 2054 | * Associate @bio with the blkg found by combining the css's blkg and the |
fc5a828b DZ |
2055 | * request_queue of the @bio. This falls back to the queue's root_blkg if |
2056 | * the association fails with the css. | |
d459d853 | 2057 | */ |
fd42df30 DZ |
2058 | void bio_associate_blkg_from_css(struct bio *bio, |
2059 | struct cgroup_subsys_state *css) | |
d459d853 | 2060 | { |
fc5a828b DZ |
2061 | struct request_queue *q = bio->bi_disk->queue; |
2062 | struct blkcg_gq *blkg; | |
2063 | ||
2064 | rcu_read_lock(); | |
2065 | ||
2066 | if (!css || !css->parent) | |
2067 | blkg = q->root_blkg; | |
2068 | else | |
2069 | blkg = blkg_lookup_create(css_to_blkcg(css), q); | |
2070 | ||
2071 | __bio_associate_blkg(bio, blkg); | |
2072 | ||
2073 | rcu_read_unlock(); | |
d459d853 | 2074 | } |
fd42df30 | 2075 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); |
d459d853 | 2076 | |
6a7f6d86 | 2077 | #ifdef CONFIG_MEMCG |
852c788f | 2078 | /** |
6a7f6d86 | 2079 | * bio_associate_blkg_from_page - associate a bio with the page's blkg |
852c788f | 2080 | * @bio: target bio |
6a7f6d86 DZ |
2081 | * @page: the page to lookup the blkcg from |
2082 | * | |
2083 | * Associate @bio with the blkg from @page's owning memcg and the respective | |
fc5a828b DZ |
2084 | * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's |
2085 | * root_blkg. | |
852c788f | 2086 | */ |
6a7f6d86 | 2087 | void bio_associate_blkg_from_page(struct bio *bio, struct page *page) |
852c788f | 2088 | { |
6a7f6d86 DZ |
2089 | struct cgroup_subsys_state *css; |
2090 | ||
6a7f6d86 DZ |
2091 | if (!page->mem_cgroup) |
2092 | return; | |
2093 | ||
fc5a828b DZ |
2094 | rcu_read_lock(); |
2095 | ||
2096 | css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); | |
2097 | bio_associate_blkg_from_css(bio, css); | |
2098 | ||
2099 | rcu_read_unlock(); | |
6a7f6d86 DZ |
2100 | } |
2101 | #endif /* CONFIG_MEMCG */ | |
2102 | ||
2268c0fe DZ |
2103 | /** |
2104 | * bio_associate_blkg - associate a bio with a blkg | |
2105 | * @bio: target bio | |
2106 | * | |
2107 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
2108 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
2109 | * already associated, the css is reused and association redone as the | |
2110 | * request_queue may have changed. | |
2111 | */ | |
2112 | void bio_associate_blkg(struct bio *bio) | |
2113 | { | |
fc5a828b | 2114 | struct cgroup_subsys_state *css; |
2268c0fe DZ |
2115 | |
2116 | rcu_read_lock(); | |
2117 | ||
db6638d7 | 2118 | if (bio->bi_blkg) |
fc5a828b | 2119 | css = &bio_blkcg(bio)->css; |
db6638d7 | 2120 | else |
fc5a828b | 2121 | css = blkcg_css(); |
2268c0fe | 2122 | |
fc5a828b | 2123 | bio_associate_blkg_from_css(bio, css); |
2268c0fe DZ |
2124 | |
2125 | rcu_read_unlock(); | |
852c788f | 2126 | } |
5cdf2e3f | 2127 | EXPORT_SYMBOL_GPL(bio_associate_blkg); |
852c788f | 2128 | |
20bd723e | 2129 | /** |
db6638d7 | 2130 | * bio_clone_blkg_association - clone blkg association from src to dst bio |
20bd723e PV |
2131 | * @dst: destination bio |
2132 | * @src: source bio | |
2133 | */ | |
db6638d7 | 2134 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) |
20bd723e | 2135 | { |
6ab21879 DZ |
2136 | rcu_read_lock(); |
2137 | ||
fc5a828b | 2138 | if (src->bi_blkg) |
2268c0fe | 2139 | __bio_associate_blkg(dst, src->bi_blkg); |
6ab21879 DZ |
2140 | |
2141 | rcu_read_unlock(); | |
20bd723e | 2142 | } |
db6638d7 | 2143 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); |
852c788f TH |
2144 | #endif /* CONFIG_BLK_CGROUP */ |
2145 | ||
1da177e4 LT |
2146 | static void __init biovec_init_slabs(void) |
2147 | { | |
2148 | int i; | |
2149 | ||
ed996a52 | 2150 | for (i = 0; i < BVEC_POOL_NR; i++) { |
1da177e4 LT |
2151 | int size; |
2152 | struct biovec_slab *bvs = bvec_slabs + i; | |
2153 | ||
a7fcd37c JA |
2154 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
2155 | bvs->slab = NULL; | |
2156 | continue; | |
2157 | } | |
a7fcd37c | 2158 | |
1da177e4 LT |
2159 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
2160 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 2161 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
2162 | } |
2163 | } | |
2164 | ||
2165 | static int __init init_bio(void) | |
2166 | { | |
bb799ca0 JA |
2167 | bio_slab_max = 2; |
2168 | bio_slab_nr = 0; | |
6396bb22 KC |
2169 | bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), |
2170 | GFP_KERNEL); | |
2b24e6f6 JT |
2171 | |
2172 | BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); | |
2173 | ||
bb799ca0 JA |
2174 | if (!bio_slabs) |
2175 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 2176 | |
7878cba9 | 2177 | bio_integrity_init(); |
1da177e4 LT |
2178 | biovec_init_slabs(); |
2179 | ||
f4f8154a | 2180 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
1da177e4 LT |
2181 | panic("bio: can't allocate bios\n"); |
2182 | ||
f4f8154a | 2183 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
2184 | panic("bio: can't create integrity pool\n"); |
2185 | ||
1da177e4 LT |
2186 | return 0; |
2187 | } | |
1da177e4 | 2188 | subsys_initcall(init_bio); |