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> |
de6a78b6 | 20 | #include <linux/sched/sysctl.h> |
a892c8d5 | 21 | #include <linux/blk-crypto.h> |
1da177e4 | 22 | |
55782138 | 23 | #include <trace/events/block.h> |
9e234eea | 24 | #include "blk.h" |
67b42d0b | 25 | #include "blk-rq-qos.h" |
0bfc2455 | 26 | |
392ddc32 JA |
27 | /* |
28 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
29 | * itself, to shrink a bio data allocation from two mempool calls to one | |
30 | */ | |
31 | #define BIO_INLINE_VECS 4 | |
32 | ||
1da177e4 LT |
33 | /* |
34 | * if you change this list, also change bvec_alloc or things will | |
35 | * break badly! cannot be bigger than what you can fit into an | |
36 | * unsigned short | |
37 | */ | |
bd5c4fac | 38 | #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } |
ed996a52 | 39 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { |
bd5c4fac | 40 | BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), |
1da177e4 LT |
41 | }; |
42 | #undef BV | |
43 | ||
1da177e4 LT |
44 | /* |
45 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
46 | * IO code that does not need private memory pools. | |
47 | */ | |
f4f8154a | 48 | struct bio_set fs_bio_set; |
3f86a82a | 49 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 50 | |
bb799ca0 JA |
51 | /* |
52 | * Our slab pool management | |
53 | */ | |
54 | struct bio_slab { | |
55 | struct kmem_cache *slab; | |
56 | unsigned int slab_ref; | |
57 | unsigned int slab_size; | |
58 | char name[8]; | |
59 | }; | |
60 | static DEFINE_MUTEX(bio_slab_lock); | |
61 | static struct bio_slab *bio_slabs; | |
62 | static unsigned int bio_slab_nr, bio_slab_max; | |
63 | ||
64 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
65 | { | |
66 | unsigned int sz = sizeof(struct bio) + extra_size; | |
67 | struct kmem_cache *slab = NULL; | |
389d7b26 | 68 | struct bio_slab *bslab, *new_bio_slabs; |
386bc35a | 69 | unsigned int new_bio_slab_max; |
bb799ca0 JA |
70 | unsigned int i, entry = -1; |
71 | ||
72 | mutex_lock(&bio_slab_lock); | |
73 | ||
74 | i = 0; | |
75 | while (i < bio_slab_nr) { | |
f06f135d | 76 | bslab = &bio_slabs[i]; |
bb799ca0 JA |
77 | |
78 | if (!bslab->slab && entry == -1) | |
79 | entry = i; | |
80 | else if (bslab->slab_size == sz) { | |
81 | slab = bslab->slab; | |
82 | bslab->slab_ref++; | |
83 | break; | |
84 | } | |
85 | i++; | |
86 | } | |
87 | ||
88 | if (slab) | |
89 | goto out_unlock; | |
90 | ||
91 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
386bc35a | 92 | new_bio_slab_max = bio_slab_max << 1; |
389d7b26 | 93 | new_bio_slabs = krealloc(bio_slabs, |
386bc35a | 94 | new_bio_slab_max * sizeof(struct bio_slab), |
389d7b26 AK |
95 | GFP_KERNEL); |
96 | if (!new_bio_slabs) | |
bb799ca0 | 97 | goto out_unlock; |
386bc35a | 98 | bio_slab_max = new_bio_slab_max; |
389d7b26 | 99 | bio_slabs = new_bio_slabs; |
bb799ca0 JA |
100 | } |
101 | if (entry == -1) | |
102 | entry = bio_slab_nr++; | |
103 | ||
104 | bslab = &bio_slabs[entry]; | |
105 | ||
106 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
6a241483 MP |
107 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
108 | SLAB_HWCACHE_ALIGN, NULL); | |
bb799ca0 JA |
109 | if (!slab) |
110 | goto out_unlock; | |
111 | ||
bb799ca0 JA |
112 | bslab->slab = slab; |
113 | bslab->slab_ref = 1; | |
114 | bslab->slab_size = sz; | |
115 | out_unlock: | |
116 | mutex_unlock(&bio_slab_lock); | |
117 | return slab; | |
118 | } | |
119 | ||
120 | static void bio_put_slab(struct bio_set *bs) | |
121 | { | |
122 | struct bio_slab *bslab = NULL; | |
123 | unsigned int i; | |
124 | ||
125 | mutex_lock(&bio_slab_lock); | |
126 | ||
127 | for (i = 0; i < bio_slab_nr; i++) { | |
128 | if (bs->bio_slab == bio_slabs[i].slab) { | |
129 | bslab = &bio_slabs[i]; | |
130 | break; | |
131 | } | |
132 | } | |
133 | ||
134 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
135 | goto out; | |
136 | ||
137 | WARN_ON(!bslab->slab_ref); | |
138 | ||
139 | if (--bslab->slab_ref) | |
140 | goto out; | |
141 | ||
142 | kmem_cache_destroy(bslab->slab); | |
143 | bslab->slab = NULL; | |
144 | ||
145 | out: | |
146 | mutex_unlock(&bio_slab_lock); | |
147 | } | |
148 | ||
7ba1ba12 MP |
149 | unsigned int bvec_nr_vecs(unsigned short idx) |
150 | { | |
d6c02a9b | 151 | return bvec_slabs[--idx].nr_vecs; |
7ba1ba12 MP |
152 | } |
153 | ||
9f060e22 | 154 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 | 155 | { |
ed996a52 CH |
156 | if (!idx) |
157 | return; | |
158 | idx--; | |
159 | ||
160 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
bb799ca0 | 161 | |
ed996a52 | 162 | if (idx == BVEC_POOL_MAX) { |
9f060e22 | 163 | mempool_free(bv, pool); |
ed996a52 | 164 | } else { |
bb799ca0 JA |
165 | struct biovec_slab *bvs = bvec_slabs + idx; |
166 | ||
167 | kmem_cache_free(bvs->slab, bv); | |
168 | } | |
169 | } | |
170 | ||
9f060e22 KO |
171 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
172 | mempool_t *pool) | |
1da177e4 LT |
173 | { |
174 | struct bio_vec *bvl; | |
1da177e4 | 175 | |
7ff9345f JA |
176 | /* |
177 | * see comment near bvec_array define! | |
178 | */ | |
179 | switch (nr) { | |
180 | case 1: | |
181 | *idx = 0; | |
182 | break; | |
183 | case 2 ... 4: | |
184 | *idx = 1; | |
185 | break; | |
186 | case 5 ... 16: | |
187 | *idx = 2; | |
188 | break; | |
189 | case 17 ... 64: | |
190 | *idx = 3; | |
191 | break; | |
192 | case 65 ... 128: | |
193 | *idx = 4; | |
194 | break; | |
195 | case 129 ... BIO_MAX_PAGES: | |
196 | *idx = 5; | |
197 | break; | |
198 | default: | |
199 | return NULL; | |
200 | } | |
201 | ||
202 | /* | |
203 | * idx now points to the pool we want to allocate from. only the | |
204 | * 1-vec entry pool is mempool backed. | |
205 | */ | |
ed996a52 | 206 | if (*idx == BVEC_POOL_MAX) { |
7ff9345f | 207 | fallback: |
9f060e22 | 208 | bvl = mempool_alloc(pool, gfp_mask); |
7ff9345f JA |
209 | } else { |
210 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
d0164adc | 211 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
7ff9345f | 212 | |
0a0d96b0 | 213 | /* |
7ff9345f JA |
214 | * Make this allocation restricted and don't dump info on |
215 | * allocation failures, since we'll fallback to the mempool | |
216 | * in case of failure. | |
0a0d96b0 | 217 | */ |
7ff9345f | 218 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 219 | |
0a0d96b0 | 220 | /* |
d0164adc | 221 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM |
7ff9345f | 222 | * is set, retry with the 1-entry mempool |
0a0d96b0 | 223 | */ |
7ff9345f | 224 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
d0164adc | 225 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { |
ed996a52 | 226 | *idx = BVEC_POOL_MAX; |
7ff9345f JA |
227 | goto fallback; |
228 | } | |
229 | } | |
230 | ||
ed996a52 | 231 | (*idx)++; |
1da177e4 LT |
232 | return bvl; |
233 | } | |
234 | ||
9ae3b3f5 | 235 | void bio_uninit(struct bio *bio) |
1da177e4 | 236 | { |
6f70fb66 | 237 | bio_disassociate_blkg(bio); |
ece841ab JT |
238 | |
239 | if (bio_integrity(bio)) | |
240 | bio_integrity_free(bio); | |
a892c8d5 ST |
241 | |
242 | bio_crypt_free_ctx(bio); | |
4254bba1 | 243 | } |
9ae3b3f5 | 244 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 245 | |
4254bba1 KO |
246 | static void bio_free(struct bio *bio) |
247 | { | |
248 | struct bio_set *bs = bio->bi_pool; | |
249 | void *p; | |
250 | ||
9ae3b3f5 | 251 | bio_uninit(bio); |
4254bba1 KO |
252 | |
253 | if (bs) { | |
8aa6ba2f | 254 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
4254bba1 KO |
255 | |
256 | /* | |
257 | * If we have front padding, adjust the bio pointer before freeing | |
258 | */ | |
259 | p = bio; | |
bb799ca0 JA |
260 | p -= bs->front_pad; |
261 | ||
8aa6ba2f | 262 | mempool_free(p, &bs->bio_pool); |
4254bba1 KO |
263 | } else { |
264 | /* Bio was allocated by bio_kmalloc() */ | |
265 | kfree(bio); | |
266 | } | |
3676347a PO |
267 | } |
268 | ||
9ae3b3f5 JA |
269 | /* |
270 | * Users of this function have their own bio allocation. Subsequently, | |
271 | * they must remember to pair any call to bio_init() with bio_uninit() | |
272 | * when IO has completed, or when the bio is released. | |
273 | */ | |
3a83f467 ML |
274 | void bio_init(struct bio *bio, struct bio_vec *table, |
275 | unsigned short max_vecs) | |
1da177e4 | 276 | { |
2b94de55 | 277 | memset(bio, 0, sizeof(*bio)); |
c4cf5261 | 278 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 279 | atomic_set(&bio->__bi_cnt, 1); |
3a83f467 ML |
280 | |
281 | bio->bi_io_vec = table; | |
282 | bio->bi_max_vecs = max_vecs; | |
1da177e4 | 283 | } |
a112a71d | 284 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 285 | |
f44b48c7 KO |
286 | /** |
287 | * bio_reset - reinitialize a bio | |
288 | * @bio: bio to reset | |
289 | * | |
290 | * Description: | |
291 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
292 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
293 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
294 | * comment in struct bio. | |
295 | */ | |
296 | void bio_reset(struct bio *bio) | |
297 | { | |
298 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
299 | ||
9ae3b3f5 | 300 | bio_uninit(bio); |
f44b48c7 KO |
301 | |
302 | memset(bio, 0, BIO_RESET_BYTES); | |
4246a0b6 | 303 | bio->bi_flags = flags; |
c4cf5261 | 304 | atomic_set(&bio->__bi_remaining, 1); |
f44b48c7 KO |
305 | } |
306 | EXPORT_SYMBOL(bio_reset); | |
307 | ||
38f8baae | 308 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 309 | { |
4246a0b6 CH |
310 | struct bio *parent = bio->bi_private; |
311 | ||
4e4cbee9 CH |
312 | if (!parent->bi_status) |
313 | parent->bi_status = bio->bi_status; | |
196d38bc | 314 | bio_put(bio); |
38f8baae CH |
315 | return parent; |
316 | } | |
317 | ||
318 | static void bio_chain_endio(struct bio *bio) | |
319 | { | |
320 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
321 | } |
322 | ||
323 | /** | |
324 | * bio_chain - chain bio completions | |
1051a902 RD |
325 | * @bio: the target bio |
326 | * @parent: the @bio's parent bio | |
196d38bc KO |
327 | * |
328 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
329 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
330 | * completed; the chained bio will also be freed when it completes. | |
331 | * | |
332 | * The caller must not set bi_private or bi_end_io in @bio. | |
333 | */ | |
334 | void bio_chain(struct bio *bio, struct bio *parent) | |
335 | { | |
336 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
337 | ||
338 | bio->bi_private = parent; | |
339 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 340 | bio_inc_remaining(parent); |
196d38bc KO |
341 | } |
342 | EXPORT_SYMBOL(bio_chain); | |
343 | ||
df2cb6da KO |
344 | static void bio_alloc_rescue(struct work_struct *work) |
345 | { | |
346 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
347 | struct bio *bio; | |
348 | ||
349 | while (1) { | |
350 | spin_lock(&bs->rescue_lock); | |
351 | bio = bio_list_pop(&bs->rescue_list); | |
352 | spin_unlock(&bs->rescue_lock); | |
353 | ||
354 | if (!bio) | |
355 | break; | |
356 | ||
357 | generic_make_request(bio); | |
358 | } | |
359 | } | |
360 | ||
361 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
362 | { | |
363 | struct bio_list punt, nopunt; | |
364 | struct bio *bio; | |
365 | ||
47e0fb46 N |
366 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
367 | return; | |
df2cb6da KO |
368 | /* |
369 | * In order to guarantee forward progress we must punt only bios that | |
370 | * were allocated from this bio_set; otherwise, if there was a bio on | |
371 | * there for a stacking driver higher up in the stack, processing it | |
372 | * could require allocating bios from this bio_set, and doing that from | |
373 | * our own rescuer would be bad. | |
374 | * | |
375 | * Since bio lists are singly linked, pop them all instead of trying to | |
376 | * remove from the middle of the list: | |
377 | */ | |
378 | ||
379 | bio_list_init(&punt); | |
380 | bio_list_init(&nopunt); | |
381 | ||
f5fe1b51 | 382 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 383 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 384 | current->bio_list[0] = nopunt; |
df2cb6da | 385 | |
f5fe1b51 N |
386 | bio_list_init(&nopunt); |
387 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
388 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
389 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
390 | |
391 | spin_lock(&bs->rescue_lock); | |
392 | bio_list_merge(&bs->rescue_list, &punt); | |
393 | spin_unlock(&bs->rescue_lock); | |
394 | ||
395 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
396 | } | |
397 | ||
1da177e4 LT |
398 | /** |
399 | * bio_alloc_bioset - allocate a bio for I/O | |
519c8e9f | 400 | * @gfp_mask: the GFP_* mask given to the slab allocator |
1da177e4 | 401 | * @nr_iovecs: number of iovecs to pre-allocate |
db18efac | 402 | * @bs: the bio_set to allocate from. |
1da177e4 LT |
403 | * |
404 | * Description: | |
3f86a82a KO |
405 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
406 | * backed by the @bs's mempool. | |
407 | * | |
d0164adc MG |
408 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will |
409 | * always be able to allocate a bio. This is due to the mempool guarantees. | |
410 | * To make this work, callers must never allocate more than 1 bio at a time | |
411 | * from this pool. Callers that need to allocate more than 1 bio must always | |
412 | * submit the previously allocated bio for IO before attempting to allocate | |
413 | * a new one. Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 414 | * |
df2cb6da KO |
415 | * Note that when running under generic_make_request() (i.e. any block |
416 | * driver), bios are not submitted until after you return - see the code in | |
417 | * generic_make_request() that converts recursion into iteration, to prevent | |
418 | * stack overflows. | |
419 | * | |
420 | * This would normally mean allocating multiple bios under | |
421 | * generic_make_request() would be susceptible to deadlocks, but we have | |
422 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
423 | * thread. | |
424 | * | |
425 | * However, we do not guarantee forward progress for allocations from other | |
426 | * mempools. Doing multiple allocations from the same mempool under | |
427 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
428 | * for per bio allocations. | |
429 | * | |
3f86a82a KO |
430 | * RETURNS: |
431 | * Pointer to new bio on success, NULL on failure. | |
432 | */ | |
7a88fa19 DC |
433 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, |
434 | struct bio_set *bs) | |
1da177e4 | 435 | { |
df2cb6da | 436 | gfp_t saved_gfp = gfp_mask; |
3f86a82a KO |
437 | unsigned front_pad; |
438 | unsigned inline_vecs; | |
34053979 | 439 | struct bio_vec *bvl = NULL; |
451a9ebf TH |
440 | struct bio *bio; |
441 | void *p; | |
442 | ||
3f86a82a KO |
443 | if (!bs) { |
444 | if (nr_iovecs > UIO_MAXIOV) | |
445 | return NULL; | |
446 | ||
1f4fe21c | 447 | p = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); |
3f86a82a KO |
448 | front_pad = 0; |
449 | inline_vecs = nr_iovecs; | |
450 | } else { | |
d8f429e1 | 451 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
8aa6ba2f KO |
452 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && |
453 | nr_iovecs > 0)) | |
d8f429e1 | 454 | return NULL; |
df2cb6da KO |
455 | /* |
456 | * generic_make_request() converts recursion to iteration; this | |
457 | * means if we're running beneath it, any bios we allocate and | |
458 | * submit will not be submitted (and thus freed) until after we | |
459 | * return. | |
460 | * | |
461 | * This exposes us to a potential deadlock if we allocate | |
462 | * multiple bios from the same bio_set() while running | |
463 | * underneath generic_make_request(). If we were to allocate | |
464 | * multiple bios (say a stacking block driver that was splitting | |
465 | * bios), we would deadlock if we exhausted the mempool's | |
466 | * reserve. | |
467 | * | |
468 | * We solve this, and guarantee forward progress, with a rescuer | |
469 | * workqueue per bio_set. If we go to allocate and there are | |
470 | * bios on current->bio_list, we first try the allocation | |
d0164adc MG |
471 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those |
472 | * bios we would be blocking to the rescuer workqueue before | |
473 | * we retry with the original gfp_flags. | |
df2cb6da KO |
474 | */ |
475 | ||
f5fe1b51 N |
476 | if (current->bio_list && |
477 | (!bio_list_empty(¤t->bio_list[0]) || | |
47e0fb46 N |
478 | !bio_list_empty(¤t->bio_list[1])) && |
479 | bs->rescue_workqueue) | |
d0164adc | 480 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; |
df2cb6da | 481 | |
8aa6ba2f | 482 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
483 | if (!p && gfp_mask != saved_gfp) { |
484 | punt_bios_to_rescuer(bs); | |
485 | gfp_mask = saved_gfp; | |
8aa6ba2f | 486 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
487 | } |
488 | ||
3f86a82a KO |
489 | front_pad = bs->front_pad; |
490 | inline_vecs = BIO_INLINE_VECS; | |
491 | } | |
492 | ||
451a9ebf TH |
493 | if (unlikely(!p)) |
494 | return NULL; | |
1da177e4 | 495 | |
3f86a82a | 496 | bio = p + front_pad; |
3a83f467 | 497 | bio_init(bio, NULL, 0); |
34053979 | 498 | |
3f86a82a | 499 | if (nr_iovecs > inline_vecs) { |
ed996a52 CH |
500 | unsigned long idx = 0; |
501 | ||
8aa6ba2f | 502 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
503 | if (!bvl && gfp_mask != saved_gfp) { |
504 | punt_bios_to_rescuer(bs); | |
505 | gfp_mask = saved_gfp; | |
8aa6ba2f | 506 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
507 | } |
508 | ||
34053979 IM |
509 | if (unlikely(!bvl)) |
510 | goto err_free; | |
a38352e0 | 511 | |
ed996a52 | 512 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
3f86a82a KO |
513 | } else if (nr_iovecs) { |
514 | bvl = bio->bi_inline_vecs; | |
1da177e4 | 515 | } |
3f86a82a KO |
516 | |
517 | bio->bi_pool = bs; | |
34053979 | 518 | bio->bi_max_vecs = nr_iovecs; |
34053979 | 519 | bio->bi_io_vec = bvl; |
1da177e4 | 520 | return bio; |
34053979 IM |
521 | |
522 | err_free: | |
8aa6ba2f | 523 | mempool_free(p, &bs->bio_pool); |
34053979 | 524 | return NULL; |
1da177e4 | 525 | } |
a112a71d | 526 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 527 | |
38a72dac | 528 | void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) |
1da177e4 LT |
529 | { |
530 | unsigned long flags; | |
7988613b KO |
531 | struct bio_vec bv; |
532 | struct bvec_iter iter; | |
1da177e4 | 533 | |
38a72dac | 534 | __bio_for_each_segment(bv, bio, iter, start) { |
7988613b KO |
535 | char *data = bvec_kmap_irq(&bv, &flags); |
536 | memset(data, 0, bv.bv_len); | |
537 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
538 | bvec_kunmap_irq(data, &flags); |
539 | } | |
540 | } | |
38a72dac | 541 | EXPORT_SYMBOL(zero_fill_bio_iter); |
1da177e4 | 542 | |
83c9c547 ML |
543 | /** |
544 | * bio_truncate - truncate the bio to small size of @new_size | |
545 | * @bio: the bio to be truncated | |
546 | * @new_size: new size for truncating the bio | |
547 | * | |
548 | * Description: | |
549 | * Truncate the bio to new size of @new_size. If bio_op(bio) is | |
550 | * REQ_OP_READ, zero the truncated part. This function should only | |
551 | * be used for handling corner cases, such as bio eod. | |
552 | */ | |
85a8ce62 ML |
553 | void bio_truncate(struct bio *bio, unsigned new_size) |
554 | { | |
555 | struct bio_vec bv; | |
556 | struct bvec_iter iter; | |
557 | unsigned int done = 0; | |
558 | bool truncated = false; | |
559 | ||
560 | if (new_size >= bio->bi_iter.bi_size) | |
561 | return; | |
562 | ||
83c9c547 | 563 | if (bio_op(bio) != REQ_OP_READ) |
85a8ce62 ML |
564 | goto exit; |
565 | ||
566 | bio_for_each_segment(bv, bio, iter) { | |
567 | if (done + bv.bv_len > new_size) { | |
568 | unsigned offset; | |
569 | ||
570 | if (!truncated) | |
571 | offset = new_size - done; | |
572 | else | |
573 | offset = 0; | |
574 | zero_user(bv.bv_page, offset, bv.bv_len - offset); | |
575 | truncated = true; | |
576 | } | |
577 | done += bv.bv_len; | |
578 | } | |
579 | ||
580 | exit: | |
581 | /* | |
582 | * Don't touch bvec table here and make it really immutable, since | |
583 | * fs bio user has to retrieve all pages via bio_for_each_segment_all | |
584 | * in its .end_bio() callback. | |
585 | * | |
586 | * It is enough to truncate bio by updating .bi_size since we can make | |
587 | * correct bvec with the updated .bi_size for drivers. | |
588 | */ | |
589 | bio->bi_iter.bi_size = new_size; | |
590 | } | |
591 | ||
29125ed6 CH |
592 | /** |
593 | * guard_bio_eod - truncate a BIO to fit the block device | |
594 | * @bio: bio to truncate | |
595 | * | |
596 | * This allows us to do IO even on the odd last sectors of a device, even if the | |
597 | * block size is some multiple of the physical sector size. | |
598 | * | |
599 | * We'll just truncate the bio to the size of the device, and clear the end of | |
600 | * the buffer head manually. Truly out-of-range accesses will turn into actual | |
601 | * I/O errors, this only handles the "we need to be able to do I/O at the final | |
602 | * sector" case. | |
603 | */ | |
604 | void guard_bio_eod(struct bio *bio) | |
605 | { | |
606 | sector_t maxsector; | |
607 | struct hd_struct *part; | |
608 | ||
609 | rcu_read_lock(); | |
610 | part = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
611 | if (part) | |
612 | maxsector = part_nr_sects_read(part); | |
613 | else | |
614 | maxsector = get_capacity(bio->bi_disk); | |
615 | rcu_read_unlock(); | |
616 | ||
617 | if (!maxsector) | |
618 | return; | |
619 | ||
620 | /* | |
621 | * If the *whole* IO is past the end of the device, | |
622 | * let it through, and the IO layer will turn it into | |
623 | * an EIO. | |
624 | */ | |
625 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) | |
626 | return; | |
627 | ||
628 | maxsector -= bio->bi_iter.bi_sector; | |
629 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) | |
630 | return; | |
631 | ||
632 | bio_truncate(bio, maxsector << 9); | |
633 | } | |
634 | ||
1da177e4 LT |
635 | /** |
636 | * bio_put - release a reference to a bio | |
637 | * @bio: bio to release reference to | |
638 | * | |
639 | * Description: | |
640 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 641 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
642 | **/ |
643 | void bio_put(struct bio *bio) | |
644 | { | |
dac56212 | 645 | if (!bio_flagged(bio, BIO_REFFED)) |
4254bba1 | 646 | bio_free(bio); |
dac56212 JA |
647 | else { |
648 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
649 | ||
650 | /* | |
651 | * last put frees it | |
652 | */ | |
653 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
654 | bio_free(bio); | |
655 | } | |
1da177e4 | 656 | } |
a112a71d | 657 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 658 | |
59d276fe KO |
659 | /** |
660 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
661 | * @bio: destination bio | |
662 | * @bio_src: bio to clone | |
663 | * | |
664 | * Clone a &bio. Caller will own the returned bio, but not | |
665 | * the actual data it points to. Reference count of returned | |
666 | * bio will be one. | |
667 | * | |
668 | * Caller must ensure that @bio_src is not freed before @bio. | |
669 | */ | |
670 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
671 | { | |
ed996a52 | 672 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
59d276fe KO |
673 | |
674 | /* | |
74d46992 | 675 | * most users will be overriding ->bi_disk with a new target, |
59d276fe KO |
676 | * so we don't set nor calculate new physical/hw segment counts here |
677 | */ | |
74d46992 | 678 | bio->bi_disk = bio_src->bi_disk; |
62530ed8 | 679 | bio->bi_partno = bio_src->bi_partno; |
b7c44ed9 | 680 | bio_set_flag(bio, BIO_CLONED); |
111be883 SL |
681 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
682 | bio_set_flag(bio, BIO_THROTTLED); | |
1eff9d32 | 683 | bio->bi_opf = bio_src->bi_opf; |
ca474b73 | 684 | bio->bi_ioprio = bio_src->bi_ioprio; |
cb6934f8 | 685 | bio->bi_write_hint = bio_src->bi_write_hint; |
59d276fe KO |
686 | bio->bi_iter = bio_src->bi_iter; |
687 | bio->bi_io_vec = bio_src->bi_io_vec; | |
20bd723e | 688 | |
db6638d7 | 689 | bio_clone_blkg_association(bio, bio_src); |
e439bedf | 690 | blkcg_bio_issue_init(bio); |
59d276fe KO |
691 | } |
692 | EXPORT_SYMBOL(__bio_clone_fast); | |
693 | ||
694 | /** | |
695 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
696 | * @bio: bio to clone | |
697 | * @gfp_mask: allocation priority | |
698 | * @bs: bio_set to allocate from | |
699 | * | |
700 | * Like __bio_clone_fast, only also allocates the returned bio | |
701 | */ | |
702 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
703 | { | |
704 | struct bio *b; | |
705 | ||
706 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
707 | if (!b) | |
708 | return NULL; | |
709 | ||
710 | __bio_clone_fast(b, bio); | |
711 | ||
a892c8d5 ST |
712 | bio_crypt_clone(b, bio, gfp_mask); |
713 | ||
59d276fe KO |
714 | if (bio_integrity(bio)) { |
715 | int ret; | |
716 | ||
717 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
718 | ||
719 | if (ret < 0) { | |
720 | bio_put(b); | |
721 | return NULL; | |
722 | } | |
723 | } | |
724 | ||
725 | return b; | |
726 | } | |
727 | EXPORT_SYMBOL(bio_clone_fast); | |
728 | ||
5cbd28e3 CH |
729 | const char *bio_devname(struct bio *bio, char *buf) |
730 | { | |
731 | return disk_name(bio->bi_disk, bio->bi_partno, buf); | |
732 | } | |
733 | EXPORT_SYMBOL(bio_devname); | |
734 | ||
5919482e ML |
735 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
736 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 737 | bool *same_page) |
5919482e ML |
738 | { |
739 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + | |
740 | bv->bv_offset + bv->bv_len - 1; | |
741 | phys_addr_t page_addr = page_to_phys(page); | |
742 | ||
743 | if (vec_end_addr + 1 != page_addr + off) | |
744 | return false; | |
745 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
746 | return false; | |
52d52d1c | 747 | |
ff896738 CH |
748 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
749 | if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) | |
750 | return false; | |
5919482e ML |
751 | return true; |
752 | } | |
753 | ||
e4581105 CH |
754 | /* |
755 | * Try to merge a page into a segment, while obeying the hardware segment | |
756 | * size limit. This is not for normal read/write bios, but for passthrough | |
757 | * or Zone Append operations that we can't split. | |
758 | */ | |
759 | static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, | |
760 | struct page *page, unsigned len, | |
761 | unsigned offset, bool *same_page) | |
489fbbcb | 762 | { |
384209cd | 763 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
489fbbcb ML |
764 | unsigned long mask = queue_segment_boundary(q); |
765 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
766 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
767 | ||
768 | if ((addr1 | mask) != (addr2 | mask)) | |
769 | return false; | |
489fbbcb ML |
770 | if (bv->bv_len + len > queue_max_segment_size(q)) |
771 | return false; | |
384209cd | 772 | return __bio_try_merge_page(bio, page, len, offset, same_page); |
489fbbcb ML |
773 | } |
774 | ||
1da177e4 | 775 | /** |
e4581105 CH |
776 | * bio_add_hw_page - attempt to add a page to a bio with hw constraints |
777 | * @q: the target queue | |
778 | * @bio: destination bio | |
779 | * @page: page to add | |
780 | * @len: vec entry length | |
781 | * @offset: vec entry offset | |
782 | * @max_sectors: maximum number of sectors that can be added | |
783 | * @same_page: return if the segment has been merged inside the same page | |
c66a14d0 | 784 | * |
e4581105 CH |
785 | * Add a page to a bio while respecting the hardware max_sectors, max_segment |
786 | * and gap limitations. | |
1da177e4 | 787 | */ |
e4581105 | 788 | int bio_add_hw_page(struct request_queue *q, struct bio *bio, |
19047087 | 789 | struct page *page, unsigned int len, unsigned int offset, |
e4581105 | 790 | unsigned int max_sectors, bool *same_page) |
1da177e4 | 791 | { |
1da177e4 LT |
792 | struct bio_vec *bvec; |
793 | ||
e4581105 | 794 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
1da177e4 LT |
795 | return 0; |
796 | ||
e4581105 | 797 | if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
798 | return 0; |
799 | ||
80cfd548 | 800 | if (bio->bi_vcnt > 0) { |
e4581105 | 801 | if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) |
384209cd | 802 | return len; |
320ea869 CH |
803 | |
804 | /* | |
805 | * If the queue doesn't support SG gaps and adding this segment | |
806 | * would create a gap, disallow it. | |
807 | */ | |
384209cd | 808 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
320ea869 CH |
809 | if (bvec_gap_to_prev(q, bvec, offset)) |
810 | return 0; | |
80cfd548 JA |
811 | } |
812 | ||
79d08f89 | 813 | if (bio_full(bio, len)) |
1da177e4 LT |
814 | return 0; |
815 | ||
14ccb66b | 816 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
817 | return 0; |
818 | ||
fcbf6a08 ML |
819 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
820 | bvec->bv_page = page; | |
821 | bvec->bv_len = len; | |
822 | bvec->bv_offset = offset; | |
823 | bio->bi_vcnt++; | |
dcdca753 | 824 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
825 | return len; |
826 | } | |
19047087 | 827 | |
e4581105 CH |
828 | /** |
829 | * bio_add_pc_page - attempt to add page to passthrough bio | |
830 | * @q: the target queue | |
831 | * @bio: destination bio | |
832 | * @page: page to add | |
833 | * @len: vec entry length | |
834 | * @offset: vec entry offset | |
835 | * | |
836 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
837 | * number of reasons, such as the bio being full or target block device | |
838 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
839 | * so it is always possible to add a single page to an empty bio. | |
840 | * | |
841 | * This should only be used by passthrough bios. | |
842 | */ | |
19047087 ML |
843 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, |
844 | struct page *page, unsigned int len, unsigned int offset) | |
845 | { | |
d1916c86 | 846 | bool same_page = false; |
e4581105 CH |
847 | return bio_add_hw_page(q, bio, page, len, offset, |
848 | queue_max_hw_sectors(q), &same_page); | |
19047087 | 849 | } |
a112a71d | 850 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 851 | |
1da177e4 | 852 | /** |
0aa69fd3 CH |
853 | * __bio_try_merge_page - try appending data to an existing bvec. |
854 | * @bio: destination bio | |
551879a4 | 855 | * @page: start page to add |
0aa69fd3 | 856 | * @len: length of the data to add |
551879a4 | 857 | * @off: offset of the data relative to @page |
ff896738 | 858 | * @same_page: return if the segment has been merged inside the same page |
1da177e4 | 859 | * |
0aa69fd3 CH |
860 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
861 | * a useful optimisation for file systems with a block size smaller than the | |
862 | * page size. | |
863 | * | |
551879a4 ML |
864 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
865 | * | |
0aa69fd3 | 866 | * Return %true on success or %false on failure. |
1da177e4 | 867 | */ |
0aa69fd3 | 868 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
ff896738 | 869 | unsigned int len, unsigned int off, bool *same_page) |
1da177e4 | 870 | { |
c66a14d0 | 871 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
0aa69fd3 | 872 | return false; |
762380ad | 873 | |
cc90bc68 | 874 | if (bio->bi_vcnt > 0) { |
0aa69fd3 | 875 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
5919482e ML |
876 | |
877 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
cc90bc68 AG |
878 | if (bio->bi_iter.bi_size > UINT_MAX - len) |
879 | return false; | |
5919482e ML |
880 | bv->bv_len += len; |
881 | bio->bi_iter.bi_size += len; | |
882 | return true; | |
883 | } | |
c66a14d0 | 884 | } |
0aa69fd3 CH |
885 | return false; |
886 | } | |
887 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); | |
c66a14d0 | 888 | |
0aa69fd3 | 889 | /** |
551879a4 | 890 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 891 | * @bio: destination bio |
551879a4 ML |
892 | * @page: start page to add |
893 | * @len: length of the data to add, may cross pages | |
894 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
895 | * |
896 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
897 | * that @bio has space for another bvec. | |
898 | */ | |
899 | void __bio_add_page(struct bio *bio, struct page *page, | |
900 | unsigned int len, unsigned int off) | |
901 | { | |
902 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 903 | |
0aa69fd3 | 904 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 905 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
906 | |
907 | bv->bv_page = page; | |
908 | bv->bv_offset = off; | |
909 | bv->bv_len = len; | |
c66a14d0 | 910 | |
c66a14d0 | 911 | bio->bi_iter.bi_size += len; |
0aa69fd3 | 912 | bio->bi_vcnt++; |
b8e24a93 JW |
913 | |
914 | if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) | |
915 | bio_set_flag(bio, BIO_WORKINGSET); | |
0aa69fd3 CH |
916 | } |
917 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
918 | ||
919 | /** | |
551879a4 | 920 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 921 | * @bio: destination bio |
551879a4 ML |
922 | * @page: start page to add |
923 | * @len: vec entry length, may cross pages | |
924 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 925 | * |
551879a4 | 926 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
927 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
928 | */ | |
929 | int bio_add_page(struct bio *bio, struct page *page, | |
930 | unsigned int len, unsigned int offset) | |
931 | { | |
ff896738 CH |
932 | bool same_page = false; |
933 | ||
934 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 935 | if (bio_full(bio, len)) |
0aa69fd3 CH |
936 | return 0; |
937 | __bio_add_page(bio, page, len, offset); | |
938 | } | |
c66a14d0 | 939 | return len; |
1da177e4 | 940 | } |
a112a71d | 941 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 942 | |
d241a95f | 943 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
944 | { |
945 | struct bvec_iter_all iter_all; | |
946 | struct bio_vec *bvec; | |
7321ecbf | 947 | |
b2d0d991 CH |
948 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
949 | return; | |
950 | ||
d241a95f CH |
951 | bio_for_each_segment_all(bvec, bio, iter_all) { |
952 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
953 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 954 | put_page(bvec->bv_page); |
d241a95f | 955 | } |
7321ecbf | 956 | } |
29b2a3aa | 957 | EXPORT_SYMBOL_GPL(bio_release_pages); |
7321ecbf | 958 | |
6d0c48ae JA |
959 | static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) |
960 | { | |
961 | const struct bio_vec *bv = iter->bvec; | |
962 | unsigned int len; | |
963 | size_t size; | |
964 | ||
965 | if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) | |
966 | return -EINVAL; | |
967 | ||
968 | len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); | |
969 | size = bio_add_page(bio, bv->bv_page, len, | |
970 | bv->bv_offset + iter->iov_offset); | |
a10584c3 CH |
971 | if (unlikely(size != len)) |
972 | return -EINVAL; | |
a10584c3 CH |
973 | iov_iter_advance(iter, size); |
974 | return 0; | |
6d0c48ae JA |
975 | } |
976 | ||
576ed913 CH |
977 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
978 | ||
2cefe4db | 979 | /** |
17d51b10 | 980 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
981 | * @bio: bio to add pages to |
982 | * @iter: iov iterator describing the region to be mapped | |
983 | * | |
17d51b10 | 984 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 985 | * pages will have to be released using put_page() when done. |
17d51b10 MW |
986 | * For multi-segment *iter, this function only adds pages from the |
987 | * the next non-empty segment of the iov iterator. | |
2cefe4db | 988 | */ |
17d51b10 | 989 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 990 | { |
576ed913 CH |
991 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
992 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
993 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
994 | struct page **pages = (struct page **)bv; | |
45691804 | 995 | bool same_page = false; |
576ed913 CH |
996 | ssize_t size, left; |
997 | unsigned len, i; | |
b403ea24 | 998 | size_t offset; |
576ed913 CH |
999 | |
1000 | /* | |
1001 | * Move page array up in the allocated memory for the bio vecs as far as | |
1002 | * possible so that we can start filling biovecs from the beginning | |
1003 | * without overwriting the temporary page array. | |
1004 | */ | |
1005 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
1006 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db KO |
1007 | |
1008 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
1009 | if (unlikely(size <= 0)) | |
1010 | return size ? size : -EFAULT; | |
2cefe4db | 1011 | |
576ed913 CH |
1012 | for (left = size, i = 0; left > 0; left -= len, i++) { |
1013 | struct page *page = pages[i]; | |
2cefe4db | 1014 | |
576ed913 | 1015 | len = min_t(size_t, PAGE_SIZE - offset, left); |
45691804 CH |
1016 | |
1017 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
1018 | if (same_page) | |
1019 | put_page(page); | |
1020 | } else { | |
79d08f89 | 1021 | if (WARN_ON_ONCE(bio_full(bio, len))) |
45691804 CH |
1022 | return -EINVAL; |
1023 | __bio_add_page(bio, page, len, offset); | |
1024 | } | |
576ed913 | 1025 | offset = 0; |
2cefe4db KO |
1026 | } |
1027 | ||
2cefe4db KO |
1028 | iov_iter_advance(iter, size); |
1029 | return 0; | |
1030 | } | |
17d51b10 | 1031 | |
0512a75b KB |
1032 | static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) |
1033 | { | |
1034 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; | |
1035 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
1036 | struct request_queue *q = bio->bi_disk->queue; | |
1037 | unsigned int max_append_sectors = queue_max_zone_append_sectors(q); | |
1038 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; | |
1039 | struct page **pages = (struct page **)bv; | |
1040 | ssize_t size, left; | |
1041 | unsigned len, i; | |
1042 | size_t offset; | |
1043 | ||
1044 | if (WARN_ON_ONCE(!max_append_sectors)) | |
1045 | return 0; | |
1046 | ||
1047 | /* | |
1048 | * Move page array up in the allocated memory for the bio vecs as far as | |
1049 | * possible so that we can start filling biovecs from the beginning | |
1050 | * without overwriting the temporary page array. | |
1051 | */ | |
1052 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
1053 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
1054 | ||
1055 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
1056 | if (unlikely(size <= 0)) | |
1057 | return size ? size : -EFAULT; | |
1058 | ||
1059 | for (left = size, i = 0; left > 0; left -= len, i++) { | |
1060 | struct page *page = pages[i]; | |
1061 | bool same_page = false; | |
1062 | ||
1063 | len = min_t(size_t, PAGE_SIZE - offset, left); | |
1064 | if (bio_add_hw_page(q, bio, page, len, offset, | |
1065 | max_append_sectors, &same_page) != len) | |
1066 | return -EINVAL; | |
1067 | if (same_page) | |
1068 | put_page(page); | |
1069 | offset = 0; | |
1070 | } | |
1071 | ||
1072 | iov_iter_advance(iter, size); | |
1073 | return 0; | |
1074 | } | |
1075 | ||
17d51b10 | 1076 | /** |
6d0c48ae | 1077 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 1078 | * @bio: bio to add pages to |
6d0c48ae JA |
1079 | * @iter: iov iterator describing the region to be added |
1080 | * | |
1081 | * This takes either an iterator pointing to user memory, or one pointing to | |
1082 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
1083 | * map them into the kernel. On IO completion, the caller should put those | |
399254aa JA |
1084 | * pages. If we're adding kernel pages, and the caller told us it's safe to |
1085 | * do so, we just have to add the pages to the bio directly. We don't grab an | |
1086 | * extra reference to those pages (the user should already have that), and we | |
1087 | * don't put the page on IO completion. The caller needs to check if the bio is | |
1088 | * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be | |
1089 | * released. | |
17d51b10 | 1090 | * |
17d51b10 | 1091 | * The function tries, but does not guarantee, to pin as many pages as |
6d0c48ae JA |
1092 | * fit into the bio, or are requested in *iter, whatever is smaller. If |
1093 | * MM encounters an error pinning the requested pages, it stops. Error | |
1094 | * is returned only if 0 pages could be pinned. | |
17d51b10 MW |
1095 | */ |
1096 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
1097 | { | |
6d0c48ae | 1098 | const bool is_bvec = iov_iter_is_bvec(iter); |
14eacf12 CH |
1099 | int ret; |
1100 | ||
1101 | if (WARN_ON_ONCE(bio->bi_vcnt)) | |
1102 | return -EINVAL; | |
17d51b10 MW |
1103 | |
1104 | do { | |
0512a75b KB |
1105 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
1106 | if (WARN_ON_ONCE(is_bvec)) | |
1107 | return -EINVAL; | |
1108 | ret = __bio_iov_append_get_pages(bio, iter); | |
1109 | } else { | |
1110 | if (is_bvec) | |
1111 | ret = __bio_iov_bvec_add_pages(bio, iter); | |
1112 | else | |
1113 | ret = __bio_iov_iter_get_pages(bio, iter); | |
1114 | } | |
79d08f89 | 1115 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
17d51b10 | 1116 | |
b6207430 | 1117 | if (is_bvec) |
7321ecbf | 1118 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
14eacf12 | 1119 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 1120 | } |
29b2a3aa | 1121 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); |
2cefe4db | 1122 | |
4246a0b6 | 1123 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 1124 | { |
65e53aab | 1125 | complete(bio->bi_private); |
9e882242 KO |
1126 | } |
1127 | ||
1128 | /** | |
1129 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
1130 | * @bio: The &struct bio which describes the I/O |
1131 | * | |
1132 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
1133 | * bio_endio() on failure. | |
3d289d68 JK |
1134 | * |
1135 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
1136 | * result in bio reference to be consumed. The caller must drop the reference | |
1137 | * on his own. | |
9e882242 | 1138 | */ |
4e49ea4a | 1139 | int submit_bio_wait(struct bio *bio) |
9e882242 | 1140 | { |
e319e1fb | 1141 | DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); |
de6a78b6 | 1142 | unsigned long hang_check; |
9e882242 | 1143 | |
65e53aab | 1144 | bio->bi_private = &done; |
9e882242 | 1145 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 1146 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 1147 | submit_bio(bio); |
de6a78b6 ML |
1148 | |
1149 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1150 | hang_check = sysctl_hung_task_timeout_secs; | |
1151 | if (hang_check) | |
1152 | while (!wait_for_completion_io_timeout(&done, | |
1153 | hang_check * (HZ/2))) | |
1154 | ; | |
1155 | else | |
1156 | wait_for_completion_io(&done); | |
9e882242 | 1157 | |
65e53aab | 1158 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
1159 | } |
1160 | EXPORT_SYMBOL(submit_bio_wait); | |
1161 | ||
054bdf64 KO |
1162 | /** |
1163 | * bio_advance - increment/complete a bio by some number of bytes | |
1164 | * @bio: bio to advance | |
1165 | * @bytes: number of bytes to complete | |
1166 | * | |
1167 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
1168 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
1169 | * be updated on the last bvec as well. | |
1170 | * | |
1171 | * @bio will then represent the remaining, uncompleted portion of the io. | |
1172 | */ | |
1173 | void bio_advance(struct bio *bio, unsigned bytes) | |
1174 | { | |
1175 | if (bio_integrity(bio)) | |
1176 | bio_integrity_advance(bio, bytes); | |
1177 | ||
a892c8d5 | 1178 | bio_crypt_advance(bio, bytes); |
4550dd6c | 1179 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
1180 | } |
1181 | EXPORT_SYMBOL(bio_advance); | |
1182 | ||
45db54d5 KO |
1183 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1184 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1185 | { |
1cb9dda4 | 1186 | struct bio_vec src_bv, dst_bv; |
16ac3d63 | 1187 | void *src_p, *dst_p; |
1cb9dda4 | 1188 | unsigned bytes; |
16ac3d63 | 1189 | |
45db54d5 KO |
1190 | while (src_iter->bi_size && dst_iter->bi_size) { |
1191 | src_bv = bio_iter_iovec(src, *src_iter); | |
1192 | dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1cb9dda4 KO |
1193 | |
1194 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1195 | |
1cb9dda4 KO |
1196 | src_p = kmap_atomic(src_bv.bv_page); |
1197 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1198 | |
1cb9dda4 KO |
1199 | memcpy(dst_p + dst_bv.bv_offset, |
1200 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1201 | bytes); |
1202 | ||
1203 | kunmap_atomic(dst_p); | |
1204 | kunmap_atomic(src_p); | |
1205 | ||
6e6e811d KO |
1206 | flush_dcache_page(dst_bv.bv_page); |
1207 | ||
45db54d5 KO |
1208 | bio_advance_iter(src, src_iter, bytes); |
1209 | bio_advance_iter(dst, dst_iter, bytes); | |
16ac3d63 KO |
1210 | } |
1211 | } | |
38a72dac KO |
1212 | EXPORT_SYMBOL(bio_copy_data_iter); |
1213 | ||
1214 | /** | |
45db54d5 KO |
1215 | * bio_copy_data - copy contents of data buffers from one bio to another |
1216 | * @src: source bio | |
1217 | * @dst: destination bio | |
38a72dac KO |
1218 | * |
1219 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1220 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1221 | */ | |
1222 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1223 | { | |
45db54d5 KO |
1224 | struct bvec_iter src_iter = src->bi_iter; |
1225 | struct bvec_iter dst_iter = dst->bi_iter; | |
1226 | ||
1227 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1228 | } |
16ac3d63 KO |
1229 | EXPORT_SYMBOL(bio_copy_data); |
1230 | ||
45db54d5 KO |
1231 | /** |
1232 | * bio_list_copy_data - copy contents of data buffers from one chain of bios to | |
1233 | * another | |
1234 | * @src: source bio list | |
1235 | * @dst: destination bio list | |
1236 | * | |
1237 | * Stops when it reaches the end of either the @src list or @dst list - that is, | |
1238 | * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of | |
1239 | * bios). | |
1240 | */ | |
1241 | void bio_list_copy_data(struct bio *dst, struct bio *src) | |
1242 | { | |
1243 | struct bvec_iter src_iter = src->bi_iter; | |
1244 | struct bvec_iter dst_iter = dst->bi_iter; | |
1245 | ||
1246 | while (1) { | |
1247 | if (!src_iter.bi_size) { | |
1248 | src = src->bi_next; | |
1249 | if (!src) | |
1250 | break; | |
1251 | ||
1252 | src_iter = src->bi_iter; | |
1253 | } | |
1254 | ||
1255 | if (!dst_iter.bi_size) { | |
1256 | dst = dst->bi_next; | |
1257 | if (!dst) | |
1258 | break; | |
1259 | ||
1260 | dst_iter = dst->bi_iter; | |
1261 | } | |
1262 | ||
1263 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
1264 | } | |
1265 | } | |
1266 | EXPORT_SYMBOL(bio_list_copy_data); | |
1267 | ||
491221f8 | 1268 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1269 | { |
1270 | struct bio_vec *bvec; | |
6dc4f100 | 1271 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1272 | |
2b070cfe | 1273 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1274 | __free_page(bvec->bv_page); |
1275 | } | |
491221f8 | 1276 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1277 | |
1da177e4 LT |
1278 | /* |
1279 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1280 | * for performing direct-IO in BIOs. | |
1281 | * | |
1282 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1283 | * because the required locks are not interrupt-safe. So what we can do is to | |
1284 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1285 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1286 | * in process context. | |
1287 | * | |
1288 | * We special-case compound pages here: normally this means reads into hugetlb | |
1289 | * pages. The logic in here doesn't really work right for compound pages | |
1290 | * because the VM does not uniformly chase down the head page in all cases. | |
1291 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1292 | * handle them at all. So we skip compound pages here at an early stage. | |
1293 | * | |
1294 | * Note that this code is very hard to test under normal circumstances because | |
1295 | * direct-io pins the pages with get_user_pages(). This makes | |
1296 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1297 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1298 | * pagecache. |
1299 | * | |
1300 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1301 | * deferred bio dirtying paths. | |
1302 | */ | |
1303 | ||
1304 | /* | |
1305 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1306 | */ | |
1307 | void bio_set_pages_dirty(struct bio *bio) | |
1308 | { | |
cb34e057 | 1309 | struct bio_vec *bvec; |
6dc4f100 | 1310 | struct bvec_iter_all iter_all; |
1da177e4 | 1311 | |
2b070cfe | 1312 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1313 | if (!PageCompound(bvec->bv_page)) |
1314 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1315 | } |
1316 | } | |
1317 | ||
1da177e4 LT |
1318 | /* |
1319 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1320 | * If they are, then fine. If, however, some pages are clean then they must | |
1321 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1322 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1323 | * |
1324 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1325 | * here on. It will run one put_page() against each page and will run one |
1326 | * bio_put() against the BIO. | |
1da177e4 LT |
1327 | */ |
1328 | ||
65f27f38 | 1329 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1330 | |
65f27f38 | 1331 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1332 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1333 | static struct bio *bio_dirty_list; | |
1334 | ||
1335 | /* | |
1336 | * This runs in process context | |
1337 | */ | |
65f27f38 | 1338 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1339 | { |
24d5493f | 1340 | struct bio *bio, *next; |
1da177e4 | 1341 | |
24d5493f CH |
1342 | spin_lock_irq(&bio_dirty_lock); |
1343 | next = bio_dirty_list; | |
1da177e4 | 1344 | bio_dirty_list = NULL; |
24d5493f | 1345 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1346 | |
24d5493f CH |
1347 | while ((bio = next) != NULL) { |
1348 | next = bio->bi_private; | |
1da177e4 | 1349 | |
d241a95f | 1350 | bio_release_pages(bio, true); |
1da177e4 | 1351 | bio_put(bio); |
1da177e4 LT |
1352 | } |
1353 | } | |
1354 | ||
1355 | void bio_check_pages_dirty(struct bio *bio) | |
1356 | { | |
cb34e057 | 1357 | struct bio_vec *bvec; |
24d5493f | 1358 | unsigned long flags; |
6dc4f100 | 1359 | struct bvec_iter_all iter_all; |
1da177e4 | 1360 | |
2b070cfe | 1361 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1362 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1363 | goto defer; | |
1da177e4 LT |
1364 | } |
1365 | ||
d241a95f | 1366 | bio_release_pages(bio, false); |
24d5493f CH |
1367 | bio_put(bio); |
1368 | return; | |
1369 | defer: | |
1370 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1371 | bio->bi_private = bio_dirty_list; | |
1372 | bio_dirty_list = bio; | |
1373 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1374 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1375 | } |
1376 | ||
c4cf5261 JA |
1377 | static inline bool bio_remaining_done(struct bio *bio) |
1378 | { | |
1379 | /* | |
1380 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1381 | * we always end io on the first invocation. | |
1382 | */ | |
1383 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1384 | return true; | |
1385 | ||
1386 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1387 | ||
326e1dbb | 1388 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1389 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1390 | return true; |
326e1dbb | 1391 | } |
c4cf5261 JA |
1392 | |
1393 | return false; | |
1394 | } | |
1395 | ||
1da177e4 LT |
1396 | /** |
1397 | * bio_endio - end I/O on a bio | |
1398 | * @bio: bio | |
1da177e4 LT |
1399 | * |
1400 | * Description: | |
4246a0b6 CH |
1401 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1402 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1403 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1404 | * |
1405 | * bio_endio() can be called several times on a bio that has been chained | |
1406 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
1407 | * last time. At this point the BLK_TA_COMPLETE tracing event will be | |
1408 | * generated if BIO_TRACE_COMPLETION is set. | |
1da177e4 | 1409 | **/ |
4246a0b6 | 1410 | void bio_endio(struct bio *bio) |
1da177e4 | 1411 | { |
ba8c6967 | 1412 | again: |
2b885517 | 1413 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1414 | return; |
7c20f116 CH |
1415 | if (!bio_integrity_endio(bio)) |
1416 | return; | |
1da177e4 | 1417 | |
67b42d0b JB |
1418 | if (bio->bi_disk) |
1419 | rq_qos_done_bio(bio->bi_disk->queue, bio); | |
1420 | ||
ba8c6967 CH |
1421 | /* |
1422 | * Need to have a real endio function for chained bios, otherwise | |
1423 | * various corner cases will break (like stacking block devices that | |
1424 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1425 | * recursion and blowing the stack. Tail call optimization would | |
1426 | * handle this, but compiling with frame pointers also disables | |
1427 | * gcc's sibling call optimization. | |
1428 | */ | |
1429 | if (bio->bi_end_io == bio_chain_endio) { | |
1430 | bio = __bio_chain_endio(bio); | |
1431 | goto again; | |
196d38bc | 1432 | } |
ba8c6967 | 1433 | |
74d46992 | 1434 | if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
d24de76a | 1435 | trace_block_bio_complete(bio->bi_disk->queue, bio); |
fbbaf700 N |
1436 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
1437 | } | |
1438 | ||
9e234eea | 1439 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1440 | /* release cgroup info */ |
1441 | bio_uninit(bio); | |
ba8c6967 CH |
1442 | if (bio->bi_end_io) |
1443 | bio->bi_end_io(bio); | |
1da177e4 | 1444 | } |
a112a71d | 1445 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1446 | |
20d0189b KO |
1447 | /** |
1448 | * bio_split - split a bio | |
1449 | * @bio: bio to split | |
1450 | * @sectors: number of sectors to split from the front of @bio | |
1451 | * @gfp: gfp mask | |
1452 | * @bs: bio set to allocate from | |
1453 | * | |
1454 | * Allocates and returns a new bio which represents @sectors from the start of | |
1455 | * @bio, and updates @bio to represent the remaining sectors. | |
1456 | * | |
f3f5da62 | 1457 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1458 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1459 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1460 | */ |
1461 | struct bio *bio_split(struct bio *bio, int sectors, | |
1462 | gfp_t gfp, struct bio_set *bs) | |
1463 | { | |
f341a4d3 | 1464 | struct bio *split; |
20d0189b KO |
1465 | |
1466 | BUG_ON(sectors <= 0); | |
1467 | BUG_ON(sectors >= bio_sectors(bio)); | |
1468 | ||
0512a75b KB |
1469 | /* Zone append commands cannot be split */ |
1470 | if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) | |
1471 | return NULL; | |
1472 | ||
f9d03f96 | 1473 | split = bio_clone_fast(bio, gfp, bs); |
20d0189b KO |
1474 | if (!split) |
1475 | return NULL; | |
1476 | ||
1477 | split->bi_iter.bi_size = sectors << 9; | |
1478 | ||
1479 | if (bio_integrity(split)) | |
fbd08e76 | 1480 | bio_integrity_trim(split); |
20d0189b KO |
1481 | |
1482 | bio_advance(bio, split->bi_iter.bi_size); | |
1483 | ||
fbbaf700 | 1484 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1485 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1486 | |
20d0189b KO |
1487 | return split; |
1488 | } | |
1489 | EXPORT_SYMBOL(bio_split); | |
1490 | ||
6678d83f KO |
1491 | /** |
1492 | * bio_trim - trim a bio | |
1493 | * @bio: bio to trim | |
1494 | * @offset: number of sectors to trim from the front of @bio | |
1495 | * @size: size we want to trim @bio to, in sectors | |
1496 | */ | |
1497 | void bio_trim(struct bio *bio, int offset, int size) | |
1498 | { | |
1499 | /* 'bio' is a cloned bio which we need to trim to match | |
1500 | * the given offset and size. | |
6678d83f | 1501 | */ |
6678d83f KO |
1502 | |
1503 | size <<= 9; | |
4f024f37 | 1504 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1505 | return; |
1506 | ||
6678d83f | 1507 | bio_advance(bio, offset << 9); |
4f024f37 | 1508 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1509 | |
1510 | if (bio_integrity(bio)) | |
fbd08e76 | 1511 | bio_integrity_trim(bio); |
376a78ab | 1512 | |
6678d83f KO |
1513 | } |
1514 | EXPORT_SYMBOL_GPL(bio_trim); | |
1515 | ||
1da177e4 LT |
1516 | /* |
1517 | * create memory pools for biovec's in a bio_set. | |
1518 | * use the global biovec slabs created for general use. | |
1519 | */ | |
8aa6ba2f | 1520 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1521 | { |
ed996a52 | 1522 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
1da177e4 | 1523 | |
8aa6ba2f | 1524 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1525 | } |
1526 | ||
917a38c7 KO |
1527 | /* |
1528 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1529 | * | |
1530 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1531 | * kzalloc()). | |
1532 | */ | |
1533 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1534 | { |
df2cb6da KO |
1535 | if (bs->rescue_workqueue) |
1536 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1537 | bs->rescue_workqueue = NULL; |
df2cb6da | 1538 | |
8aa6ba2f KO |
1539 | mempool_exit(&bs->bio_pool); |
1540 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1541 | |
7878cba9 | 1542 | bioset_integrity_free(bs); |
917a38c7 KO |
1543 | if (bs->bio_slab) |
1544 | bio_put_slab(bs); | |
1545 | bs->bio_slab = NULL; | |
1546 | } | |
1547 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1548 | |
917a38c7 KO |
1549 | /** |
1550 | * bioset_init - Initialize a bio_set | |
dad08527 | 1551 | * @bs: pool to initialize |
917a38c7 KO |
1552 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1553 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1554 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1555 | * and %BIOSET_NEED_RESCUER | |
1556 | * | |
dad08527 KO |
1557 | * Description: |
1558 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1559 | * to ask for a number of bytes to be allocated in front of the bio. | |
1560 | * Front pad allocation is useful for embedding the bio inside | |
1561 | * another structure, to avoid allocating extra data to go with the bio. | |
1562 | * Note that the bio must be embedded at the END of that structure always, | |
1563 | * or things will break badly. | |
1564 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
1565 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
1566 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
1567 | * dispatch queued requests when the mempool runs out of space. | |
1568 | * | |
917a38c7 KO |
1569 | */ |
1570 | int bioset_init(struct bio_set *bs, | |
1571 | unsigned int pool_size, | |
1572 | unsigned int front_pad, | |
1573 | int flags) | |
1574 | { | |
1575 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1576 | ||
1577 | bs->front_pad = front_pad; | |
1578 | ||
1579 | spin_lock_init(&bs->rescue_lock); | |
1580 | bio_list_init(&bs->rescue_list); | |
1581 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1582 | ||
1583 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1584 | if (!bs->bio_slab) | |
1585 | return -ENOMEM; | |
1586 | ||
1587 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) | |
1588 | goto bad; | |
1589 | ||
1590 | if ((flags & BIOSET_NEED_BVECS) && | |
1591 | biovec_init_pool(&bs->bvec_pool, pool_size)) | |
1592 | goto bad; | |
1593 | ||
1594 | if (!(flags & BIOSET_NEED_RESCUER)) | |
1595 | return 0; | |
1596 | ||
1597 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1598 | if (!bs->rescue_workqueue) | |
1599 | goto bad; | |
1600 | ||
1601 | return 0; | |
1602 | bad: | |
1603 | bioset_exit(bs); | |
1604 | return -ENOMEM; | |
1605 | } | |
1606 | EXPORT_SYMBOL(bioset_init); | |
1607 | ||
28e89fd9 JA |
1608 | /* |
1609 | * Initialize and setup a new bio_set, based on the settings from | |
1610 | * another bio_set. | |
1611 | */ | |
1612 | int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) | |
1613 | { | |
1614 | int flags; | |
1615 | ||
1616 | flags = 0; | |
1617 | if (src->bvec_pool.min_nr) | |
1618 | flags |= BIOSET_NEED_BVECS; | |
1619 | if (src->rescue_workqueue) | |
1620 | flags |= BIOSET_NEED_RESCUER; | |
1621 | ||
1622 | return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); | |
1623 | } | |
1624 | EXPORT_SYMBOL(bioset_init_from_src); | |
1625 | ||
852c788f | 1626 | #ifdef CONFIG_BLK_CGROUP |
1d933cf0 | 1627 | |
74b7c02a | 1628 | /** |
2268c0fe | 1629 | * bio_disassociate_blkg - puts back the blkg reference if associated |
74b7c02a | 1630 | * @bio: target bio |
74b7c02a | 1631 | * |
2268c0fe | 1632 | * Helper to disassociate the blkg from @bio if a blkg is associated. |
74b7c02a | 1633 | */ |
2268c0fe | 1634 | void bio_disassociate_blkg(struct bio *bio) |
74b7c02a | 1635 | { |
2268c0fe DZ |
1636 | if (bio->bi_blkg) { |
1637 | blkg_put(bio->bi_blkg); | |
1638 | bio->bi_blkg = NULL; | |
1639 | } | |
74b7c02a | 1640 | } |
892ad71f | 1641 | EXPORT_SYMBOL_GPL(bio_disassociate_blkg); |
74b7c02a | 1642 | |
a7b39b4e | 1643 | /** |
2268c0fe | 1644 | * __bio_associate_blkg - associate a bio with the a blkg |
a7b39b4e | 1645 | * @bio: target bio |
b5f2954d | 1646 | * @blkg: the blkg to associate |
b5f2954d | 1647 | * |
beea9da0 DZ |
1648 | * This tries to associate @bio with the specified @blkg. Association failure |
1649 | * is handled by walking up the blkg tree. Therefore, the blkg associated can | |
1650 | * be anything between @blkg and the root_blkg. This situation only happens | |
1651 | * when a cgroup is dying and then the remaining bios will spill to the closest | |
1652 | * alive blkg. | |
a7b39b4e | 1653 | * |
beea9da0 DZ |
1654 | * A reference will be taken on the @blkg and will be released when @bio is |
1655 | * freed. | |
a7b39b4e | 1656 | */ |
2268c0fe | 1657 | static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) |
a7b39b4e | 1658 | { |
2268c0fe DZ |
1659 | bio_disassociate_blkg(bio); |
1660 | ||
7754f669 | 1661 | bio->bi_blkg = blkg_tryget_closest(blkg); |
a7b39b4e DZF |
1662 | } |
1663 | ||
d459d853 | 1664 | /** |
fd42df30 | 1665 | * bio_associate_blkg_from_css - associate a bio with a specified css |
d459d853 | 1666 | * @bio: target bio |
fd42df30 | 1667 | * @css: target css |
d459d853 | 1668 | * |
fd42df30 | 1669 | * Associate @bio with the blkg found by combining the css's blkg and the |
fc5a828b DZ |
1670 | * request_queue of the @bio. This falls back to the queue's root_blkg if |
1671 | * the association fails with the css. | |
d459d853 | 1672 | */ |
fd42df30 DZ |
1673 | void bio_associate_blkg_from_css(struct bio *bio, |
1674 | struct cgroup_subsys_state *css) | |
d459d853 | 1675 | { |
fc5a828b DZ |
1676 | struct request_queue *q = bio->bi_disk->queue; |
1677 | struct blkcg_gq *blkg; | |
1678 | ||
1679 | rcu_read_lock(); | |
1680 | ||
1681 | if (!css || !css->parent) | |
1682 | blkg = q->root_blkg; | |
1683 | else | |
1684 | blkg = blkg_lookup_create(css_to_blkcg(css), q); | |
1685 | ||
1686 | __bio_associate_blkg(bio, blkg); | |
1687 | ||
1688 | rcu_read_unlock(); | |
d459d853 | 1689 | } |
fd42df30 | 1690 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); |
d459d853 | 1691 | |
6a7f6d86 | 1692 | #ifdef CONFIG_MEMCG |
852c788f | 1693 | /** |
6a7f6d86 | 1694 | * bio_associate_blkg_from_page - associate a bio with the page's blkg |
852c788f | 1695 | * @bio: target bio |
6a7f6d86 DZ |
1696 | * @page: the page to lookup the blkcg from |
1697 | * | |
1698 | * Associate @bio with the blkg from @page's owning memcg and the respective | |
fc5a828b DZ |
1699 | * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's |
1700 | * root_blkg. | |
852c788f | 1701 | */ |
6a7f6d86 | 1702 | void bio_associate_blkg_from_page(struct bio *bio, struct page *page) |
852c788f | 1703 | { |
6a7f6d86 DZ |
1704 | struct cgroup_subsys_state *css; |
1705 | ||
6a7f6d86 DZ |
1706 | if (!page->mem_cgroup) |
1707 | return; | |
1708 | ||
fc5a828b DZ |
1709 | rcu_read_lock(); |
1710 | ||
1711 | css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); | |
1712 | bio_associate_blkg_from_css(bio, css); | |
1713 | ||
1714 | rcu_read_unlock(); | |
6a7f6d86 DZ |
1715 | } |
1716 | #endif /* CONFIG_MEMCG */ | |
1717 | ||
2268c0fe DZ |
1718 | /** |
1719 | * bio_associate_blkg - associate a bio with a blkg | |
1720 | * @bio: target bio | |
1721 | * | |
1722 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1723 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1724 | * already associated, the css is reused and association redone as the | |
1725 | * request_queue may have changed. | |
1726 | */ | |
1727 | void bio_associate_blkg(struct bio *bio) | |
1728 | { | |
fc5a828b | 1729 | struct cgroup_subsys_state *css; |
2268c0fe DZ |
1730 | |
1731 | rcu_read_lock(); | |
1732 | ||
db6638d7 | 1733 | if (bio->bi_blkg) |
fc5a828b | 1734 | css = &bio_blkcg(bio)->css; |
db6638d7 | 1735 | else |
fc5a828b | 1736 | css = blkcg_css(); |
2268c0fe | 1737 | |
fc5a828b | 1738 | bio_associate_blkg_from_css(bio, css); |
2268c0fe DZ |
1739 | |
1740 | rcu_read_unlock(); | |
852c788f | 1741 | } |
5cdf2e3f | 1742 | EXPORT_SYMBOL_GPL(bio_associate_blkg); |
852c788f | 1743 | |
20bd723e | 1744 | /** |
db6638d7 | 1745 | * bio_clone_blkg_association - clone blkg association from src to dst bio |
20bd723e PV |
1746 | * @dst: destination bio |
1747 | * @src: source bio | |
1748 | */ | |
db6638d7 | 1749 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) |
20bd723e | 1750 | { |
6ab21879 DZ |
1751 | rcu_read_lock(); |
1752 | ||
fc5a828b | 1753 | if (src->bi_blkg) |
2268c0fe | 1754 | __bio_associate_blkg(dst, src->bi_blkg); |
6ab21879 DZ |
1755 | |
1756 | rcu_read_unlock(); | |
20bd723e | 1757 | } |
db6638d7 | 1758 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); |
852c788f TH |
1759 | #endif /* CONFIG_BLK_CGROUP */ |
1760 | ||
1da177e4 LT |
1761 | static void __init biovec_init_slabs(void) |
1762 | { | |
1763 | int i; | |
1764 | ||
ed996a52 | 1765 | for (i = 0; i < BVEC_POOL_NR; i++) { |
1da177e4 LT |
1766 | int size; |
1767 | struct biovec_slab *bvs = bvec_slabs + i; | |
1768 | ||
a7fcd37c JA |
1769 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
1770 | bvs->slab = NULL; | |
1771 | continue; | |
1772 | } | |
a7fcd37c | 1773 | |
1da177e4 LT |
1774 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
1775 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1776 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1777 | } |
1778 | } | |
1779 | ||
1780 | static int __init init_bio(void) | |
1781 | { | |
bb799ca0 JA |
1782 | bio_slab_max = 2; |
1783 | bio_slab_nr = 0; | |
6396bb22 KC |
1784 | bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), |
1785 | GFP_KERNEL); | |
2b24e6f6 JT |
1786 | |
1787 | BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); | |
1788 | ||
bb799ca0 JA |
1789 | if (!bio_slabs) |
1790 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 1791 | |
7878cba9 | 1792 | bio_integrity_init(); |
1da177e4 LT |
1793 | biovec_init_slabs(); |
1794 | ||
f4f8154a | 1795 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
1da177e4 LT |
1796 | panic("bio: can't allocate bios\n"); |
1797 | ||
f4f8154a | 1798 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
1799 | panic("bio: can't create integrity pool\n"); |
1800 | ||
1da177e4 LT |
1801 | return 0; |
1802 | } | |
1da177e4 | 1803 | subsys_initcall(init_bio); |