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> |
49d1ec85 | 22 | #include <linux/xarray.h> |
1da177e4 | 23 | |
55782138 | 24 | #include <trace/events/block.h> |
9e234eea | 25 | #include "blk.h" |
67b42d0b | 26 | #include "blk-rq-qos.h" |
0bfc2455 | 27 | |
de76fd89 | 28 | static struct biovec_slab { |
6ac0b715 CH |
29 | int nr_vecs; |
30 | char *name; | |
31 | struct kmem_cache *slab; | |
de76fd89 CH |
32 | } bvec_slabs[] __read_mostly = { |
33 | { .nr_vecs = 16, .name = "biovec-16" }, | |
34 | { .nr_vecs = 64, .name = "biovec-64" }, | |
35 | { .nr_vecs = 128, .name = "biovec-128" }, | |
36 | { .nr_vecs = BIO_MAX_PAGES, .name = "biovec-max" }, | |
6ac0b715 CH |
37 | }; |
38 | ||
1da177e4 LT |
39 | /* |
40 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
41 | * IO code that does not need private memory pools. | |
42 | */ | |
f4f8154a | 43 | struct bio_set fs_bio_set; |
3f86a82a | 44 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 45 | |
bb799ca0 JA |
46 | /* |
47 | * Our slab pool management | |
48 | */ | |
49 | struct bio_slab { | |
50 | struct kmem_cache *slab; | |
51 | unsigned int slab_ref; | |
52 | unsigned int slab_size; | |
53 | char name[8]; | |
54 | }; | |
55 | static DEFINE_MUTEX(bio_slab_lock); | |
49d1ec85 | 56 | static DEFINE_XARRAY(bio_slabs); |
bb799ca0 | 57 | |
49d1ec85 | 58 | static struct bio_slab *create_bio_slab(unsigned int size) |
bb799ca0 | 59 | { |
49d1ec85 | 60 | struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); |
bb799ca0 | 61 | |
49d1ec85 ML |
62 | if (!bslab) |
63 | return NULL; | |
bb799ca0 | 64 | |
49d1ec85 ML |
65 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); |
66 | bslab->slab = kmem_cache_create(bslab->name, size, | |
67 | ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL); | |
68 | if (!bslab->slab) | |
69 | goto fail_alloc_slab; | |
bb799ca0 | 70 | |
49d1ec85 ML |
71 | bslab->slab_ref = 1; |
72 | bslab->slab_size = size; | |
bb799ca0 | 73 | |
49d1ec85 ML |
74 | if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) |
75 | return bslab; | |
bb799ca0 | 76 | |
49d1ec85 | 77 | kmem_cache_destroy(bslab->slab); |
bb799ca0 | 78 | |
49d1ec85 ML |
79 | fail_alloc_slab: |
80 | kfree(bslab); | |
81 | return NULL; | |
82 | } | |
bb799ca0 | 83 | |
49d1ec85 ML |
84 | static inline unsigned int bs_bio_slab_size(struct bio_set *bs) |
85 | { | |
9f180e31 | 86 | return bs->front_pad + sizeof(struct bio) + bs->back_pad; |
49d1ec85 ML |
87 | } |
88 | ||
89 | static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) | |
90 | { | |
91 | unsigned int size = bs_bio_slab_size(bs); | |
92 | struct bio_slab *bslab; | |
93 | ||
94 | mutex_lock(&bio_slab_lock); | |
95 | bslab = xa_load(&bio_slabs, size); | |
96 | if (bslab) | |
97 | bslab->slab_ref++; | |
98 | else | |
99 | bslab = create_bio_slab(size); | |
bb799ca0 | 100 | mutex_unlock(&bio_slab_lock); |
49d1ec85 ML |
101 | |
102 | if (bslab) | |
103 | return bslab->slab; | |
104 | return NULL; | |
bb799ca0 JA |
105 | } |
106 | ||
107 | static void bio_put_slab(struct bio_set *bs) | |
108 | { | |
109 | struct bio_slab *bslab = NULL; | |
49d1ec85 | 110 | unsigned int slab_size = bs_bio_slab_size(bs); |
bb799ca0 JA |
111 | |
112 | mutex_lock(&bio_slab_lock); | |
113 | ||
49d1ec85 | 114 | bslab = xa_load(&bio_slabs, slab_size); |
bb799ca0 JA |
115 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
116 | goto out; | |
117 | ||
49d1ec85 ML |
118 | WARN_ON_ONCE(bslab->slab != bs->bio_slab); |
119 | ||
bb799ca0 JA |
120 | WARN_ON(!bslab->slab_ref); |
121 | ||
122 | if (--bslab->slab_ref) | |
123 | goto out; | |
124 | ||
49d1ec85 ML |
125 | xa_erase(&bio_slabs, slab_size); |
126 | ||
bb799ca0 | 127 | kmem_cache_destroy(bslab->slab); |
49d1ec85 | 128 | kfree(bslab); |
bb799ca0 JA |
129 | |
130 | out: | |
131 | mutex_unlock(&bio_slab_lock); | |
132 | } | |
133 | ||
7ba1ba12 MP |
134 | unsigned int bvec_nr_vecs(unsigned short idx) |
135 | { | |
d6c02a9b | 136 | return bvec_slabs[--idx].nr_vecs; |
7ba1ba12 MP |
137 | } |
138 | ||
9f060e22 | 139 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 | 140 | { |
ed996a52 CH |
141 | if (!idx) |
142 | return; | |
143 | idx--; | |
144 | ||
145 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
bb799ca0 | 146 | |
ed996a52 | 147 | if (idx == BVEC_POOL_MAX) { |
9f060e22 | 148 | mempool_free(bv, pool); |
ed996a52 | 149 | } else { |
bb799ca0 JA |
150 | struct biovec_slab *bvs = bvec_slabs + idx; |
151 | ||
152 | kmem_cache_free(bvs->slab, bv); | |
153 | } | |
154 | } | |
155 | ||
f2c3eb9b CH |
156 | /* |
157 | * Make the first allocation restricted and don't dump info on allocation | |
158 | * failures, since we'll fall back to the mempool in case of failure. | |
159 | */ | |
160 | static inline gfp_t bvec_alloc_gfp(gfp_t gfp) | |
161 | { | |
162 | return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | | |
163 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
164 | } | |
165 | ||
9f060e22 KO |
166 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
167 | mempool_t *pool) | |
1da177e4 | 168 | { |
7ff9345f JA |
169 | /* |
170 | * see comment near bvec_array define! | |
171 | */ | |
172 | switch (nr) { | |
de76fd89 | 173 | /* smaller bios use inline vecs */ |
7ff9345f JA |
174 | case 5 ... 16: |
175 | *idx = 2; | |
176 | break; | |
177 | case 17 ... 64: | |
178 | *idx = 3; | |
179 | break; | |
180 | case 65 ... 128: | |
181 | *idx = 4; | |
182 | break; | |
183 | case 129 ... BIO_MAX_PAGES: | |
184 | *idx = 5; | |
185 | break; | |
186 | default: | |
187 | return NULL; | |
188 | } | |
189 | ||
190 | /* | |
f007a3d6 CH |
191 | * Try a slab allocation first for all smaller allocations. If that |
192 | * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. | |
193 | * The mempool is sized to handle up to BIO_MAX_PAGES entries. | |
7ff9345f | 194 | */ |
f007a3d6 | 195 | if (*idx < BVEC_POOL_MAX) { |
7ff9345f | 196 | struct biovec_slab *bvs = bvec_slabs + *idx; |
f007a3d6 | 197 | struct bio_vec *bvl; |
1da177e4 | 198 | |
f2c3eb9b | 199 | bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); |
f007a3d6 CH |
200 | if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) { |
201 | (*idx)++; | |
202 | return bvl; | |
7ff9345f | 203 | } |
f007a3d6 | 204 | *idx = BVEC_POOL_MAX; |
7ff9345f JA |
205 | } |
206 | ||
ed996a52 | 207 | (*idx)++; |
f007a3d6 | 208 | return mempool_alloc(pool, gfp_mask); |
1da177e4 LT |
209 | } |
210 | ||
9ae3b3f5 | 211 | void bio_uninit(struct bio *bio) |
1da177e4 | 212 | { |
db9819c7 CH |
213 | #ifdef CONFIG_BLK_CGROUP |
214 | if (bio->bi_blkg) { | |
215 | blkg_put(bio->bi_blkg); | |
216 | bio->bi_blkg = NULL; | |
217 | } | |
218 | #endif | |
ece841ab JT |
219 | if (bio_integrity(bio)) |
220 | bio_integrity_free(bio); | |
a892c8d5 ST |
221 | |
222 | bio_crypt_free_ctx(bio); | |
4254bba1 | 223 | } |
9ae3b3f5 | 224 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 225 | |
4254bba1 KO |
226 | static void bio_free(struct bio *bio) |
227 | { | |
228 | struct bio_set *bs = bio->bi_pool; | |
229 | void *p; | |
230 | ||
9ae3b3f5 | 231 | bio_uninit(bio); |
4254bba1 KO |
232 | |
233 | if (bs) { | |
8aa6ba2f | 234 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
4254bba1 KO |
235 | |
236 | /* | |
237 | * If we have front padding, adjust the bio pointer before freeing | |
238 | */ | |
239 | p = bio; | |
bb799ca0 JA |
240 | p -= bs->front_pad; |
241 | ||
8aa6ba2f | 242 | mempool_free(p, &bs->bio_pool); |
4254bba1 KO |
243 | } else { |
244 | /* Bio was allocated by bio_kmalloc() */ | |
245 | kfree(bio); | |
246 | } | |
3676347a PO |
247 | } |
248 | ||
9ae3b3f5 JA |
249 | /* |
250 | * Users of this function have their own bio allocation. Subsequently, | |
251 | * they must remember to pair any call to bio_init() with bio_uninit() | |
252 | * when IO has completed, or when the bio is released. | |
253 | */ | |
3a83f467 ML |
254 | void bio_init(struct bio *bio, struct bio_vec *table, |
255 | unsigned short max_vecs) | |
1da177e4 | 256 | { |
2b94de55 | 257 | memset(bio, 0, sizeof(*bio)); |
c4cf5261 | 258 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 259 | atomic_set(&bio->__bi_cnt, 1); |
3a83f467 ML |
260 | |
261 | bio->bi_io_vec = table; | |
262 | bio->bi_max_vecs = max_vecs; | |
1da177e4 | 263 | } |
a112a71d | 264 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 265 | |
f44b48c7 KO |
266 | /** |
267 | * bio_reset - reinitialize a bio | |
268 | * @bio: bio to reset | |
269 | * | |
270 | * Description: | |
271 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
272 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
273 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
274 | * comment in struct bio. | |
275 | */ | |
276 | void bio_reset(struct bio *bio) | |
277 | { | |
278 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
279 | ||
9ae3b3f5 | 280 | bio_uninit(bio); |
f44b48c7 KO |
281 | |
282 | memset(bio, 0, BIO_RESET_BYTES); | |
4246a0b6 | 283 | bio->bi_flags = flags; |
c4cf5261 | 284 | atomic_set(&bio->__bi_remaining, 1); |
f44b48c7 KO |
285 | } |
286 | EXPORT_SYMBOL(bio_reset); | |
287 | ||
38f8baae | 288 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 289 | { |
4246a0b6 CH |
290 | struct bio *parent = bio->bi_private; |
291 | ||
4e4cbee9 CH |
292 | if (!parent->bi_status) |
293 | parent->bi_status = bio->bi_status; | |
196d38bc | 294 | bio_put(bio); |
38f8baae CH |
295 | return parent; |
296 | } | |
297 | ||
298 | static void bio_chain_endio(struct bio *bio) | |
299 | { | |
300 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
301 | } |
302 | ||
303 | /** | |
304 | * bio_chain - chain bio completions | |
1051a902 | 305 | * @bio: the target bio |
5b874af6 | 306 | * @parent: the parent bio of @bio |
196d38bc KO |
307 | * |
308 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
309 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
310 | * completed; the chained bio will also be freed when it completes. | |
311 | * | |
312 | * The caller must not set bi_private or bi_end_io in @bio. | |
313 | */ | |
314 | void bio_chain(struct bio *bio, struct bio *parent) | |
315 | { | |
316 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
317 | ||
318 | bio->bi_private = parent; | |
319 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 320 | bio_inc_remaining(parent); |
196d38bc KO |
321 | } |
322 | EXPORT_SYMBOL(bio_chain); | |
323 | ||
df2cb6da KO |
324 | static void bio_alloc_rescue(struct work_struct *work) |
325 | { | |
326 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
327 | struct bio *bio; | |
328 | ||
329 | while (1) { | |
330 | spin_lock(&bs->rescue_lock); | |
331 | bio = bio_list_pop(&bs->rescue_list); | |
332 | spin_unlock(&bs->rescue_lock); | |
333 | ||
334 | if (!bio) | |
335 | break; | |
336 | ||
ed00aabd | 337 | submit_bio_noacct(bio); |
df2cb6da KO |
338 | } |
339 | } | |
340 | ||
341 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
342 | { | |
343 | struct bio_list punt, nopunt; | |
344 | struct bio *bio; | |
345 | ||
47e0fb46 N |
346 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
347 | return; | |
df2cb6da KO |
348 | /* |
349 | * In order to guarantee forward progress we must punt only bios that | |
350 | * were allocated from this bio_set; otherwise, if there was a bio on | |
351 | * there for a stacking driver higher up in the stack, processing it | |
352 | * could require allocating bios from this bio_set, and doing that from | |
353 | * our own rescuer would be bad. | |
354 | * | |
355 | * Since bio lists are singly linked, pop them all instead of trying to | |
356 | * remove from the middle of the list: | |
357 | */ | |
358 | ||
359 | bio_list_init(&punt); | |
360 | bio_list_init(&nopunt); | |
361 | ||
f5fe1b51 | 362 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 363 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 364 | current->bio_list[0] = nopunt; |
df2cb6da | 365 | |
f5fe1b51 N |
366 | bio_list_init(&nopunt); |
367 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
368 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
369 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
370 | |
371 | spin_lock(&bs->rescue_lock); | |
372 | bio_list_merge(&bs->rescue_list, &punt); | |
373 | spin_unlock(&bs->rescue_lock); | |
374 | ||
375 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
376 | } | |
377 | ||
1da177e4 LT |
378 | /** |
379 | * bio_alloc_bioset - allocate a bio for I/O | |
519c8e9f | 380 | * @gfp_mask: the GFP_* mask given to the slab allocator |
1da177e4 | 381 | * @nr_iovecs: number of iovecs to pre-allocate |
db18efac | 382 | * @bs: the bio_set to allocate from. |
1da177e4 | 383 | * |
3175199a | 384 | * Allocate a bio from the mempools in @bs. |
3f86a82a | 385 | * |
3175199a CH |
386 | * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to |
387 | * allocate a bio. This is due to the mempool guarantees. To make this work, | |
388 | * callers must never allocate more than 1 bio at a time from the general pool. | |
389 | * Callers that need to allocate more than 1 bio must always submit the | |
390 | * previously allocated bio for IO before attempting to allocate a new one. | |
391 | * Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 392 | * |
3175199a CH |
393 | * Note that when running under submit_bio_noacct() (i.e. any block driver), |
394 | * bios are not submitted until after you return - see the code in | |
395 | * submit_bio_noacct() that converts recursion into iteration, to prevent | |
396 | * stack overflows. | |
df2cb6da | 397 | * |
3175199a CH |
398 | * This would normally mean allocating multiple bios under submit_bio_noacct() |
399 | * would be susceptible to deadlocks, but we have | |
400 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
401 | * thread. | |
df2cb6da | 402 | * |
3175199a CH |
403 | * However, we do not guarantee forward progress for allocations from other |
404 | * mempools. Doing multiple allocations from the same mempool under | |
405 | * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad | |
406 | * for per bio allocations. | |
df2cb6da | 407 | * |
3175199a | 408 | * Returns: Pointer to new bio on success, NULL on failure. |
3f86a82a | 409 | */ |
0f2e6ab8 | 410 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, |
7a88fa19 | 411 | struct bio_set *bs) |
1da177e4 | 412 | { |
df2cb6da | 413 | gfp_t saved_gfp = gfp_mask; |
451a9ebf TH |
414 | struct bio *bio; |
415 | void *p; | |
416 | ||
3175199a CH |
417 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
418 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) | |
419 | return NULL; | |
df2cb6da | 420 | |
3175199a CH |
421 | /* |
422 | * submit_bio_noacct() converts recursion to iteration; this means if | |
423 | * we're running beneath it, any bios we allocate and submit will not be | |
424 | * submitted (and thus freed) until after we return. | |
425 | * | |
426 | * This exposes us to a potential deadlock if we allocate multiple bios | |
427 | * from the same bio_set() while running underneath submit_bio_noacct(). | |
428 | * If we were to allocate multiple bios (say a stacking block driver | |
429 | * that was splitting bios), we would deadlock if we exhausted the | |
430 | * mempool's reserve. | |
431 | * | |
432 | * We solve this, and guarantee forward progress, with a rescuer | |
433 | * workqueue per bio_set. If we go to allocate and there are bios on | |
434 | * current->bio_list, we first try the allocation without | |
435 | * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be | |
436 | * blocking to the rescuer workqueue before we retry with the original | |
437 | * gfp_flags. | |
438 | */ | |
439 | if (current->bio_list && | |
440 | (!bio_list_empty(¤t->bio_list[0]) || | |
441 | !bio_list_empty(¤t->bio_list[1])) && | |
442 | bs->rescue_workqueue) | |
443 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; | |
444 | ||
445 | p = mempool_alloc(&bs->bio_pool, gfp_mask); | |
446 | if (!p && gfp_mask != saved_gfp) { | |
447 | punt_bios_to_rescuer(bs); | |
448 | gfp_mask = saved_gfp; | |
8aa6ba2f | 449 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
3f86a82a | 450 | } |
451a9ebf TH |
451 | if (unlikely(!p)) |
452 | return NULL; | |
1da177e4 | 453 | |
3175199a CH |
454 | bio = p + bs->front_pad; |
455 | if (nr_iovecs > BIO_INLINE_VECS) { | |
ed996a52 | 456 | unsigned long idx = 0; |
3175199a | 457 | struct bio_vec *bvl = NULL; |
ed996a52 | 458 | |
8aa6ba2f | 459 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
460 | if (!bvl && gfp_mask != saved_gfp) { |
461 | punt_bios_to_rescuer(bs); | |
462 | gfp_mask = saved_gfp; | |
3175199a CH |
463 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, |
464 | &bs->bvec_pool); | |
df2cb6da KO |
465 | } |
466 | ||
34053979 IM |
467 | if (unlikely(!bvl)) |
468 | goto err_free; | |
a38352e0 | 469 | |
3175199a | 470 | bio_init(bio, bvl, bvec_nr_vecs(idx)); |
8358c28a | 471 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
3f86a82a | 472 | } else if (nr_iovecs) { |
3175199a CH |
473 | bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); |
474 | } else { | |
475 | bio_init(bio, NULL, 0); | |
1da177e4 | 476 | } |
3f86a82a KO |
477 | |
478 | bio->bi_pool = bs; | |
1da177e4 | 479 | return bio; |
34053979 IM |
480 | |
481 | err_free: | |
8aa6ba2f | 482 | mempool_free(p, &bs->bio_pool); |
34053979 | 483 | return NULL; |
1da177e4 | 484 | } |
a112a71d | 485 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 486 | |
3175199a CH |
487 | /** |
488 | * bio_kmalloc - kmalloc a bio for I/O | |
489 | * @gfp_mask: the GFP_* mask given to the slab allocator | |
490 | * @nr_iovecs: number of iovecs to pre-allocate | |
491 | * | |
492 | * Use kmalloc to allocate and initialize a bio. | |
493 | * | |
494 | * Returns: Pointer to new bio on success, NULL on failure. | |
495 | */ | |
0f2e6ab8 | 496 | struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) |
3175199a CH |
497 | { |
498 | struct bio *bio; | |
499 | ||
500 | if (nr_iovecs > UIO_MAXIOV) | |
501 | return NULL; | |
502 | ||
503 | bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); | |
504 | if (unlikely(!bio)) | |
505 | return NULL; | |
506 | bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); | |
507 | bio->bi_pool = NULL; | |
508 | return bio; | |
509 | } | |
510 | EXPORT_SYMBOL(bio_kmalloc); | |
511 | ||
38a72dac | 512 | void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) |
1da177e4 LT |
513 | { |
514 | unsigned long flags; | |
7988613b KO |
515 | struct bio_vec bv; |
516 | struct bvec_iter iter; | |
1da177e4 | 517 | |
38a72dac | 518 | __bio_for_each_segment(bv, bio, iter, start) { |
7988613b KO |
519 | char *data = bvec_kmap_irq(&bv, &flags); |
520 | memset(data, 0, bv.bv_len); | |
521 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
522 | bvec_kunmap_irq(data, &flags); |
523 | } | |
524 | } | |
38a72dac | 525 | EXPORT_SYMBOL(zero_fill_bio_iter); |
1da177e4 | 526 | |
83c9c547 ML |
527 | /** |
528 | * bio_truncate - truncate the bio to small size of @new_size | |
529 | * @bio: the bio to be truncated | |
530 | * @new_size: new size for truncating the bio | |
531 | * | |
532 | * Description: | |
533 | * Truncate the bio to new size of @new_size. If bio_op(bio) is | |
534 | * REQ_OP_READ, zero the truncated part. This function should only | |
535 | * be used for handling corner cases, such as bio eod. | |
536 | */ | |
85a8ce62 ML |
537 | void bio_truncate(struct bio *bio, unsigned new_size) |
538 | { | |
539 | struct bio_vec bv; | |
540 | struct bvec_iter iter; | |
541 | unsigned int done = 0; | |
542 | bool truncated = false; | |
543 | ||
544 | if (new_size >= bio->bi_iter.bi_size) | |
545 | return; | |
546 | ||
83c9c547 | 547 | if (bio_op(bio) != REQ_OP_READ) |
85a8ce62 ML |
548 | goto exit; |
549 | ||
550 | bio_for_each_segment(bv, bio, iter) { | |
551 | if (done + bv.bv_len > new_size) { | |
552 | unsigned offset; | |
553 | ||
554 | if (!truncated) | |
555 | offset = new_size - done; | |
556 | else | |
557 | offset = 0; | |
558 | zero_user(bv.bv_page, offset, bv.bv_len - offset); | |
559 | truncated = true; | |
560 | } | |
561 | done += bv.bv_len; | |
562 | } | |
563 | ||
564 | exit: | |
565 | /* | |
566 | * Don't touch bvec table here and make it really immutable, since | |
567 | * fs bio user has to retrieve all pages via bio_for_each_segment_all | |
568 | * in its .end_bio() callback. | |
569 | * | |
570 | * It is enough to truncate bio by updating .bi_size since we can make | |
571 | * correct bvec with the updated .bi_size for drivers. | |
572 | */ | |
573 | bio->bi_iter.bi_size = new_size; | |
574 | } | |
575 | ||
29125ed6 CH |
576 | /** |
577 | * guard_bio_eod - truncate a BIO to fit the block device | |
578 | * @bio: bio to truncate | |
579 | * | |
580 | * This allows us to do IO even on the odd last sectors of a device, even if the | |
581 | * block size is some multiple of the physical sector size. | |
582 | * | |
583 | * We'll just truncate the bio to the size of the device, and clear the end of | |
584 | * the buffer head manually. Truly out-of-range accesses will turn into actual | |
585 | * I/O errors, this only handles the "we need to be able to do I/O at the final | |
586 | * sector" case. | |
587 | */ | |
588 | void guard_bio_eod(struct bio *bio) | |
589 | { | |
309dca30 | 590 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
29125ed6 CH |
591 | |
592 | if (!maxsector) | |
593 | return; | |
594 | ||
595 | /* | |
596 | * If the *whole* IO is past the end of the device, | |
597 | * let it through, and the IO layer will turn it into | |
598 | * an EIO. | |
599 | */ | |
600 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) | |
601 | return; | |
602 | ||
603 | maxsector -= bio->bi_iter.bi_sector; | |
604 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) | |
605 | return; | |
606 | ||
607 | bio_truncate(bio, maxsector << 9); | |
608 | } | |
609 | ||
1da177e4 LT |
610 | /** |
611 | * bio_put - release a reference to a bio | |
612 | * @bio: bio to release reference to | |
613 | * | |
614 | * Description: | |
615 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 616 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
617 | **/ |
618 | void bio_put(struct bio *bio) | |
619 | { | |
dac56212 | 620 | if (!bio_flagged(bio, BIO_REFFED)) |
4254bba1 | 621 | bio_free(bio); |
dac56212 JA |
622 | else { |
623 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
624 | ||
625 | /* | |
626 | * last put frees it | |
627 | */ | |
628 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
629 | bio_free(bio); | |
630 | } | |
1da177e4 | 631 | } |
a112a71d | 632 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 633 | |
59d276fe KO |
634 | /** |
635 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
636 | * @bio: destination bio | |
637 | * @bio_src: bio to clone | |
638 | * | |
639 | * Clone a &bio. Caller will own the returned bio, but not | |
640 | * the actual data it points to. Reference count of returned | |
641 | * bio will be one. | |
642 | * | |
643 | * Caller must ensure that @bio_src is not freed before @bio. | |
644 | */ | |
645 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
646 | { | |
ed996a52 | 647 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
59d276fe KO |
648 | |
649 | /* | |
309dca30 | 650 | * most users will be overriding ->bi_bdev with a new target, |
59d276fe KO |
651 | * so we don't set nor calculate new physical/hw segment counts here |
652 | */ | |
309dca30 | 653 | bio->bi_bdev = bio_src->bi_bdev; |
b7c44ed9 | 654 | bio_set_flag(bio, BIO_CLONED); |
111be883 SL |
655 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
656 | bio_set_flag(bio, BIO_THROTTLED); | |
46bbf653 CH |
657 | if (bio_flagged(bio_src, BIO_REMAPPED)) |
658 | bio_set_flag(bio, BIO_REMAPPED); | |
1eff9d32 | 659 | bio->bi_opf = bio_src->bi_opf; |
ca474b73 | 660 | bio->bi_ioprio = bio_src->bi_ioprio; |
cb6934f8 | 661 | bio->bi_write_hint = bio_src->bi_write_hint; |
59d276fe KO |
662 | bio->bi_iter = bio_src->bi_iter; |
663 | bio->bi_io_vec = bio_src->bi_io_vec; | |
20bd723e | 664 | |
db6638d7 | 665 | bio_clone_blkg_association(bio, bio_src); |
e439bedf | 666 | blkcg_bio_issue_init(bio); |
59d276fe KO |
667 | } |
668 | EXPORT_SYMBOL(__bio_clone_fast); | |
669 | ||
670 | /** | |
671 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
672 | * @bio: bio to clone | |
673 | * @gfp_mask: allocation priority | |
674 | * @bs: bio_set to allocate from | |
675 | * | |
676 | * Like __bio_clone_fast, only also allocates the returned bio | |
677 | */ | |
678 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
679 | { | |
680 | struct bio *b; | |
681 | ||
682 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
683 | if (!b) | |
684 | return NULL; | |
685 | ||
686 | __bio_clone_fast(b, bio); | |
687 | ||
07560151 EB |
688 | if (bio_crypt_clone(b, bio, gfp_mask) < 0) |
689 | goto err_put; | |
a892c8d5 | 690 | |
07560151 EB |
691 | if (bio_integrity(bio) && |
692 | bio_integrity_clone(b, bio, gfp_mask) < 0) | |
693 | goto err_put; | |
59d276fe KO |
694 | |
695 | return b; | |
07560151 EB |
696 | |
697 | err_put: | |
698 | bio_put(b); | |
699 | return NULL; | |
59d276fe KO |
700 | } |
701 | EXPORT_SYMBOL(bio_clone_fast); | |
702 | ||
5cbd28e3 CH |
703 | const char *bio_devname(struct bio *bio, char *buf) |
704 | { | |
309dca30 | 705 | return bdevname(bio->bi_bdev, buf); |
5cbd28e3 CH |
706 | } |
707 | EXPORT_SYMBOL(bio_devname); | |
708 | ||
5919482e ML |
709 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
710 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 711 | bool *same_page) |
5919482e | 712 | { |
d8166519 MWO |
713 | size_t bv_end = bv->bv_offset + bv->bv_len; |
714 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; | |
5919482e ML |
715 | phys_addr_t page_addr = page_to_phys(page); |
716 | ||
717 | if (vec_end_addr + 1 != page_addr + off) | |
718 | return false; | |
719 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
720 | return false; | |
52d52d1c | 721 | |
ff896738 | 722 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
d8166519 MWO |
723 | if (*same_page) |
724 | return true; | |
725 | return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); | |
5919482e ML |
726 | } |
727 | ||
e4581105 CH |
728 | /* |
729 | * Try to merge a page into a segment, while obeying the hardware segment | |
730 | * size limit. This is not for normal read/write bios, but for passthrough | |
731 | * or Zone Append operations that we can't split. | |
732 | */ | |
733 | static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, | |
734 | struct page *page, unsigned len, | |
735 | unsigned offset, bool *same_page) | |
489fbbcb | 736 | { |
384209cd | 737 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
489fbbcb ML |
738 | unsigned long mask = queue_segment_boundary(q); |
739 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
740 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
741 | ||
742 | if ((addr1 | mask) != (addr2 | mask)) | |
743 | return false; | |
489fbbcb ML |
744 | if (bv->bv_len + len > queue_max_segment_size(q)) |
745 | return false; | |
384209cd | 746 | return __bio_try_merge_page(bio, page, len, offset, same_page); |
489fbbcb ML |
747 | } |
748 | ||
1da177e4 | 749 | /** |
e4581105 CH |
750 | * bio_add_hw_page - attempt to add a page to a bio with hw constraints |
751 | * @q: the target queue | |
752 | * @bio: destination bio | |
753 | * @page: page to add | |
754 | * @len: vec entry length | |
755 | * @offset: vec entry offset | |
756 | * @max_sectors: maximum number of sectors that can be added | |
757 | * @same_page: return if the segment has been merged inside the same page | |
c66a14d0 | 758 | * |
e4581105 CH |
759 | * Add a page to a bio while respecting the hardware max_sectors, max_segment |
760 | * and gap limitations. | |
1da177e4 | 761 | */ |
e4581105 | 762 | int bio_add_hw_page(struct request_queue *q, struct bio *bio, |
19047087 | 763 | struct page *page, unsigned int len, unsigned int offset, |
e4581105 | 764 | unsigned int max_sectors, bool *same_page) |
1da177e4 | 765 | { |
1da177e4 LT |
766 | struct bio_vec *bvec; |
767 | ||
e4581105 | 768 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
1da177e4 LT |
769 | return 0; |
770 | ||
e4581105 | 771 | if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
772 | return 0; |
773 | ||
80cfd548 | 774 | if (bio->bi_vcnt > 0) { |
e4581105 | 775 | if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) |
384209cd | 776 | return len; |
320ea869 CH |
777 | |
778 | /* | |
779 | * If the queue doesn't support SG gaps and adding this segment | |
780 | * would create a gap, disallow it. | |
781 | */ | |
384209cd | 782 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
320ea869 CH |
783 | if (bvec_gap_to_prev(q, bvec, offset)) |
784 | return 0; | |
80cfd548 JA |
785 | } |
786 | ||
79d08f89 | 787 | if (bio_full(bio, len)) |
1da177e4 LT |
788 | return 0; |
789 | ||
14ccb66b | 790 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
791 | return 0; |
792 | ||
fcbf6a08 ML |
793 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
794 | bvec->bv_page = page; | |
795 | bvec->bv_len = len; | |
796 | bvec->bv_offset = offset; | |
797 | bio->bi_vcnt++; | |
dcdca753 | 798 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
799 | return len; |
800 | } | |
19047087 | 801 | |
e4581105 CH |
802 | /** |
803 | * bio_add_pc_page - attempt to add page to passthrough bio | |
804 | * @q: the target queue | |
805 | * @bio: destination bio | |
806 | * @page: page to add | |
807 | * @len: vec entry length | |
808 | * @offset: vec entry offset | |
809 | * | |
810 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
811 | * number of reasons, such as the bio being full or target block device | |
812 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
813 | * so it is always possible to add a single page to an empty bio. | |
814 | * | |
815 | * This should only be used by passthrough bios. | |
816 | */ | |
19047087 ML |
817 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, |
818 | struct page *page, unsigned int len, unsigned int offset) | |
819 | { | |
d1916c86 | 820 | bool same_page = false; |
e4581105 CH |
821 | return bio_add_hw_page(q, bio, page, len, offset, |
822 | queue_max_hw_sectors(q), &same_page); | |
19047087 | 823 | } |
a112a71d | 824 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 825 | |
1da177e4 | 826 | /** |
0aa69fd3 CH |
827 | * __bio_try_merge_page - try appending data to an existing bvec. |
828 | * @bio: destination bio | |
551879a4 | 829 | * @page: start page to add |
0aa69fd3 | 830 | * @len: length of the data to add |
551879a4 | 831 | * @off: offset of the data relative to @page |
ff896738 | 832 | * @same_page: return if the segment has been merged inside the same page |
1da177e4 | 833 | * |
0aa69fd3 | 834 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
3cf14889 | 835 | * useful optimisation for file systems with a block size smaller than the |
0aa69fd3 CH |
836 | * page size. |
837 | * | |
551879a4 ML |
838 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
839 | * | |
0aa69fd3 | 840 | * Return %true on success or %false on failure. |
1da177e4 | 841 | */ |
0aa69fd3 | 842 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
ff896738 | 843 | unsigned int len, unsigned int off, bool *same_page) |
1da177e4 | 844 | { |
c66a14d0 | 845 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
0aa69fd3 | 846 | return false; |
762380ad | 847 | |
cc90bc68 | 848 | if (bio->bi_vcnt > 0) { |
0aa69fd3 | 849 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
5919482e ML |
850 | |
851 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
2cd896a5 RH |
852 | if (bio->bi_iter.bi_size > UINT_MAX - len) { |
853 | *same_page = false; | |
cc90bc68 | 854 | return false; |
2cd896a5 | 855 | } |
5919482e ML |
856 | bv->bv_len += len; |
857 | bio->bi_iter.bi_size += len; | |
858 | return true; | |
859 | } | |
c66a14d0 | 860 | } |
0aa69fd3 CH |
861 | return false; |
862 | } | |
863 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); | |
c66a14d0 | 864 | |
0aa69fd3 | 865 | /** |
551879a4 | 866 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 867 | * @bio: destination bio |
551879a4 ML |
868 | * @page: start page to add |
869 | * @len: length of the data to add, may cross pages | |
870 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
871 | * |
872 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
873 | * that @bio has space for another bvec. | |
874 | */ | |
875 | void __bio_add_page(struct bio *bio, struct page *page, | |
876 | unsigned int len, unsigned int off) | |
877 | { | |
878 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 879 | |
0aa69fd3 | 880 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 881 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
882 | |
883 | bv->bv_page = page; | |
884 | bv->bv_offset = off; | |
885 | bv->bv_len = len; | |
c66a14d0 | 886 | |
c66a14d0 | 887 | bio->bi_iter.bi_size += len; |
0aa69fd3 | 888 | bio->bi_vcnt++; |
b8e24a93 JW |
889 | |
890 | if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) | |
891 | bio_set_flag(bio, BIO_WORKINGSET); | |
0aa69fd3 CH |
892 | } |
893 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
894 | ||
895 | /** | |
551879a4 | 896 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 897 | * @bio: destination bio |
551879a4 ML |
898 | * @page: start page to add |
899 | * @len: vec entry length, may cross pages | |
900 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 901 | * |
551879a4 | 902 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
903 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
904 | */ | |
905 | int bio_add_page(struct bio *bio, struct page *page, | |
906 | unsigned int len, unsigned int offset) | |
907 | { | |
ff896738 CH |
908 | bool same_page = false; |
909 | ||
910 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 911 | if (bio_full(bio, len)) |
0aa69fd3 CH |
912 | return 0; |
913 | __bio_add_page(bio, page, len, offset); | |
914 | } | |
c66a14d0 | 915 | return len; |
1da177e4 | 916 | } |
a112a71d | 917 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 918 | |
d241a95f | 919 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
920 | { |
921 | struct bvec_iter_all iter_all; | |
922 | struct bio_vec *bvec; | |
7321ecbf | 923 | |
b2d0d991 CH |
924 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
925 | return; | |
926 | ||
d241a95f CH |
927 | bio_for_each_segment_all(bvec, bio, iter_all) { |
928 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
929 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 930 | put_page(bvec->bv_page); |
d241a95f | 931 | } |
7321ecbf | 932 | } |
29b2a3aa | 933 | EXPORT_SYMBOL_GPL(bio_release_pages); |
7321ecbf | 934 | |
c42bca92 | 935 | static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) |
6d0c48ae | 936 | { |
c42bca92 PB |
937 | WARN_ON_ONCE(BVEC_POOL_IDX(bio) != 0); |
938 | ||
939 | bio->bi_vcnt = iter->nr_segs; | |
940 | bio->bi_max_vecs = iter->nr_segs; | |
941 | bio->bi_io_vec = (struct bio_vec *)iter->bvec; | |
942 | bio->bi_iter.bi_bvec_done = iter->iov_offset; | |
943 | bio->bi_iter.bi_size = iter->count; | |
ed97ce5e | 944 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
c42bca92 PB |
945 | |
946 | iov_iter_advance(iter, iter->count); | |
a10584c3 | 947 | return 0; |
6d0c48ae JA |
948 | } |
949 | ||
576ed913 CH |
950 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
951 | ||
2cefe4db | 952 | /** |
17d51b10 | 953 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
954 | * @bio: bio to add pages to |
955 | * @iter: iov iterator describing the region to be mapped | |
956 | * | |
17d51b10 | 957 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 958 | * pages will have to be released using put_page() when done. |
17d51b10 | 959 | * For multi-segment *iter, this function only adds pages from the |
3cf14889 | 960 | * next non-empty segment of the iov iterator. |
2cefe4db | 961 | */ |
17d51b10 | 962 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 963 | { |
576ed913 CH |
964 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
965 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
966 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
967 | struct page **pages = (struct page **)bv; | |
45691804 | 968 | bool same_page = false; |
576ed913 CH |
969 | ssize_t size, left; |
970 | unsigned len, i; | |
b403ea24 | 971 | size_t offset; |
576ed913 CH |
972 | |
973 | /* | |
974 | * Move page array up in the allocated memory for the bio vecs as far as | |
975 | * possible so that we can start filling biovecs from the beginning | |
976 | * without overwriting the temporary page array. | |
977 | */ | |
978 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
979 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db KO |
980 | |
981 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
982 | if (unlikely(size <= 0)) | |
983 | return size ? size : -EFAULT; | |
2cefe4db | 984 | |
576ed913 CH |
985 | for (left = size, i = 0; left > 0; left -= len, i++) { |
986 | struct page *page = pages[i]; | |
2cefe4db | 987 | |
576ed913 | 988 | len = min_t(size_t, PAGE_SIZE - offset, left); |
45691804 CH |
989 | |
990 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
991 | if (same_page) | |
992 | put_page(page); | |
993 | } else { | |
79d08f89 | 994 | if (WARN_ON_ONCE(bio_full(bio, len))) |
45691804 CH |
995 | return -EINVAL; |
996 | __bio_add_page(bio, page, len, offset); | |
997 | } | |
576ed913 | 998 | offset = 0; |
2cefe4db KO |
999 | } |
1000 | ||
2cefe4db KO |
1001 | iov_iter_advance(iter, size); |
1002 | return 0; | |
1003 | } | |
17d51b10 | 1004 | |
0512a75b KB |
1005 | static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) |
1006 | { | |
1007 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; | |
1008 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
309dca30 | 1009 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
0512a75b KB |
1010 | unsigned int max_append_sectors = queue_max_zone_append_sectors(q); |
1011 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; | |
1012 | struct page **pages = (struct page **)bv; | |
1013 | ssize_t size, left; | |
1014 | unsigned len, i; | |
1015 | size_t offset; | |
4977d121 | 1016 | int ret = 0; |
0512a75b KB |
1017 | |
1018 | if (WARN_ON_ONCE(!max_append_sectors)) | |
1019 | return 0; | |
1020 | ||
1021 | /* | |
1022 | * Move page array up in the allocated memory for the bio vecs as far as | |
1023 | * possible so that we can start filling biovecs from the beginning | |
1024 | * without overwriting the temporary page array. | |
1025 | */ | |
1026 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
1027 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
1028 | ||
1029 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
1030 | if (unlikely(size <= 0)) | |
1031 | return size ? size : -EFAULT; | |
1032 | ||
1033 | for (left = size, i = 0; left > 0; left -= len, i++) { | |
1034 | struct page *page = pages[i]; | |
1035 | bool same_page = false; | |
1036 | ||
1037 | len = min_t(size_t, PAGE_SIZE - offset, left); | |
1038 | if (bio_add_hw_page(q, bio, page, len, offset, | |
4977d121 NA |
1039 | max_append_sectors, &same_page) != len) { |
1040 | ret = -EINVAL; | |
1041 | break; | |
1042 | } | |
0512a75b KB |
1043 | if (same_page) |
1044 | put_page(page); | |
1045 | offset = 0; | |
1046 | } | |
1047 | ||
4977d121 NA |
1048 | iov_iter_advance(iter, size - left); |
1049 | return ret; | |
0512a75b KB |
1050 | } |
1051 | ||
17d51b10 | 1052 | /** |
6d0c48ae | 1053 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 1054 | * @bio: bio to add pages to |
6d0c48ae JA |
1055 | * @iter: iov iterator describing the region to be added |
1056 | * | |
1057 | * This takes either an iterator pointing to user memory, or one pointing to | |
1058 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
1059 | * map them into the kernel. On IO completion, the caller should put those | |
c42bca92 PB |
1060 | * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided |
1061 | * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs | |
1062 | * to ensure the bvecs and pages stay referenced until the submitted I/O is | |
1063 | * completed by a call to ->ki_complete() or returns with an error other than | |
1064 | * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF | |
1065 | * on IO completion. If it isn't, then pages should be released. | |
17d51b10 | 1066 | * |
17d51b10 | 1067 | * The function tries, but does not guarantee, to pin as many pages as |
5cd3ddc1 | 1068 | * fit into the bio, or are requested in @iter, whatever is smaller. If |
6d0c48ae JA |
1069 | * MM encounters an error pinning the requested pages, it stops. Error |
1070 | * is returned only if 0 pages could be pinned. | |
0cf41e5e PB |
1071 | * |
1072 | * It's intended for direct IO, so doesn't do PSI tracking, the caller is | |
1073 | * responsible for setting BIO_WORKINGSET if necessary. | |
17d51b10 MW |
1074 | */ |
1075 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
1076 | { | |
c42bca92 | 1077 | int ret = 0; |
17d51b10 | 1078 | |
c42bca92 PB |
1079 | if (iov_iter_is_bvec(iter)) { |
1080 | if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) | |
1081 | return -EINVAL; | |
ed97ce5e | 1082 | return bio_iov_bvec_set(bio, iter); |
c42bca92 | 1083 | } |
0cf41e5e | 1084 | |
86004515 CH |
1085 | do { |
1086 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) | |
1087 | ret = __bio_iov_append_get_pages(bio, iter); | |
1088 | else | |
1089 | ret = __bio_iov_iter_get_pages(bio, iter); | |
1090 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); | |
1091 | ||
0cf41e5e PB |
1092 | /* don't account direct I/O as memory stall */ |
1093 | bio_clear_flag(bio, BIO_WORKINGSET); | |
14eacf12 | 1094 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 1095 | } |
29b2a3aa | 1096 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); |
2cefe4db | 1097 | |
4246a0b6 | 1098 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 1099 | { |
65e53aab | 1100 | complete(bio->bi_private); |
9e882242 KO |
1101 | } |
1102 | ||
1103 | /** | |
1104 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
1105 | * @bio: The &struct bio which describes the I/O |
1106 | * | |
1107 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
1108 | * bio_endio() on failure. | |
3d289d68 JK |
1109 | * |
1110 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
1111 | * result in bio reference to be consumed. The caller must drop the reference | |
1112 | * on his own. | |
9e882242 | 1113 | */ |
4e49ea4a | 1114 | int submit_bio_wait(struct bio *bio) |
9e882242 | 1115 | { |
309dca30 CH |
1116 | DECLARE_COMPLETION_ONSTACK_MAP(done, |
1117 | bio->bi_bdev->bd_disk->lockdep_map); | |
de6a78b6 | 1118 | unsigned long hang_check; |
9e882242 | 1119 | |
65e53aab | 1120 | bio->bi_private = &done; |
9e882242 | 1121 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 1122 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 1123 | submit_bio(bio); |
de6a78b6 ML |
1124 | |
1125 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1126 | hang_check = sysctl_hung_task_timeout_secs; | |
1127 | if (hang_check) | |
1128 | while (!wait_for_completion_io_timeout(&done, | |
1129 | hang_check * (HZ/2))) | |
1130 | ; | |
1131 | else | |
1132 | wait_for_completion_io(&done); | |
9e882242 | 1133 | |
65e53aab | 1134 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
1135 | } |
1136 | EXPORT_SYMBOL(submit_bio_wait); | |
1137 | ||
054bdf64 KO |
1138 | /** |
1139 | * bio_advance - increment/complete a bio by some number of bytes | |
1140 | * @bio: bio to advance | |
1141 | * @bytes: number of bytes to complete | |
1142 | * | |
1143 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
1144 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
1145 | * be updated on the last bvec as well. | |
1146 | * | |
1147 | * @bio will then represent the remaining, uncompleted portion of the io. | |
1148 | */ | |
1149 | void bio_advance(struct bio *bio, unsigned bytes) | |
1150 | { | |
1151 | if (bio_integrity(bio)) | |
1152 | bio_integrity_advance(bio, bytes); | |
1153 | ||
a892c8d5 | 1154 | bio_crypt_advance(bio, bytes); |
4550dd6c | 1155 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
1156 | } |
1157 | EXPORT_SYMBOL(bio_advance); | |
1158 | ||
45db54d5 KO |
1159 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1160 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1161 | { |
1cb9dda4 | 1162 | struct bio_vec src_bv, dst_bv; |
16ac3d63 | 1163 | void *src_p, *dst_p; |
1cb9dda4 | 1164 | unsigned bytes; |
16ac3d63 | 1165 | |
45db54d5 KO |
1166 | while (src_iter->bi_size && dst_iter->bi_size) { |
1167 | src_bv = bio_iter_iovec(src, *src_iter); | |
1168 | dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1cb9dda4 KO |
1169 | |
1170 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1171 | |
1cb9dda4 KO |
1172 | src_p = kmap_atomic(src_bv.bv_page); |
1173 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1174 | |
1cb9dda4 KO |
1175 | memcpy(dst_p + dst_bv.bv_offset, |
1176 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1177 | bytes); |
1178 | ||
1179 | kunmap_atomic(dst_p); | |
1180 | kunmap_atomic(src_p); | |
1181 | ||
6e6e811d KO |
1182 | flush_dcache_page(dst_bv.bv_page); |
1183 | ||
22b56c29 PB |
1184 | bio_advance_iter_single(src, src_iter, bytes); |
1185 | bio_advance_iter_single(dst, dst_iter, bytes); | |
16ac3d63 KO |
1186 | } |
1187 | } | |
38a72dac KO |
1188 | EXPORT_SYMBOL(bio_copy_data_iter); |
1189 | ||
1190 | /** | |
45db54d5 KO |
1191 | * bio_copy_data - copy contents of data buffers from one bio to another |
1192 | * @src: source bio | |
1193 | * @dst: destination bio | |
38a72dac KO |
1194 | * |
1195 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1196 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1197 | */ | |
1198 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1199 | { | |
45db54d5 KO |
1200 | struct bvec_iter src_iter = src->bi_iter; |
1201 | struct bvec_iter dst_iter = dst->bi_iter; | |
1202 | ||
1203 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1204 | } |
16ac3d63 KO |
1205 | EXPORT_SYMBOL(bio_copy_data); |
1206 | ||
45db54d5 KO |
1207 | /** |
1208 | * bio_list_copy_data - copy contents of data buffers from one chain of bios to | |
1209 | * another | |
1210 | * @src: source bio list | |
1211 | * @dst: destination bio list | |
1212 | * | |
1213 | * Stops when it reaches the end of either the @src list or @dst list - that is, | |
1214 | * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of | |
1215 | * bios). | |
1216 | */ | |
1217 | void bio_list_copy_data(struct bio *dst, struct bio *src) | |
1218 | { | |
1219 | struct bvec_iter src_iter = src->bi_iter; | |
1220 | struct bvec_iter dst_iter = dst->bi_iter; | |
1221 | ||
1222 | while (1) { | |
1223 | if (!src_iter.bi_size) { | |
1224 | src = src->bi_next; | |
1225 | if (!src) | |
1226 | break; | |
1227 | ||
1228 | src_iter = src->bi_iter; | |
1229 | } | |
1230 | ||
1231 | if (!dst_iter.bi_size) { | |
1232 | dst = dst->bi_next; | |
1233 | if (!dst) | |
1234 | break; | |
1235 | ||
1236 | dst_iter = dst->bi_iter; | |
1237 | } | |
1238 | ||
1239 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
1240 | } | |
1241 | } | |
1242 | EXPORT_SYMBOL(bio_list_copy_data); | |
1243 | ||
491221f8 | 1244 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1245 | { |
1246 | struct bio_vec *bvec; | |
6dc4f100 | 1247 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1248 | |
2b070cfe | 1249 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1250 | __free_page(bvec->bv_page); |
1251 | } | |
491221f8 | 1252 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1253 | |
1da177e4 LT |
1254 | /* |
1255 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1256 | * for performing direct-IO in BIOs. | |
1257 | * | |
1258 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1259 | * because the required locks are not interrupt-safe. So what we can do is to | |
1260 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1261 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1262 | * in process context. | |
1263 | * | |
1264 | * We special-case compound pages here: normally this means reads into hugetlb | |
1265 | * pages. The logic in here doesn't really work right for compound pages | |
1266 | * because the VM does not uniformly chase down the head page in all cases. | |
1267 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1268 | * handle them at all. So we skip compound pages here at an early stage. | |
1269 | * | |
1270 | * Note that this code is very hard to test under normal circumstances because | |
1271 | * direct-io pins the pages with get_user_pages(). This makes | |
1272 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1273 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1274 | * pagecache. |
1275 | * | |
1276 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1277 | * deferred bio dirtying paths. | |
1278 | */ | |
1279 | ||
1280 | /* | |
1281 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1282 | */ | |
1283 | void bio_set_pages_dirty(struct bio *bio) | |
1284 | { | |
cb34e057 | 1285 | struct bio_vec *bvec; |
6dc4f100 | 1286 | struct bvec_iter_all iter_all; |
1da177e4 | 1287 | |
2b070cfe | 1288 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1289 | if (!PageCompound(bvec->bv_page)) |
1290 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1291 | } |
1292 | } | |
1293 | ||
1da177e4 LT |
1294 | /* |
1295 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1296 | * If they are, then fine. If, however, some pages are clean then they must | |
1297 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1298 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1299 | * |
1300 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1301 | * here on. It will run one put_page() against each page and will run one |
1302 | * bio_put() against the BIO. | |
1da177e4 LT |
1303 | */ |
1304 | ||
65f27f38 | 1305 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1306 | |
65f27f38 | 1307 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1308 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1309 | static struct bio *bio_dirty_list; | |
1310 | ||
1311 | /* | |
1312 | * This runs in process context | |
1313 | */ | |
65f27f38 | 1314 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1315 | { |
24d5493f | 1316 | struct bio *bio, *next; |
1da177e4 | 1317 | |
24d5493f CH |
1318 | spin_lock_irq(&bio_dirty_lock); |
1319 | next = bio_dirty_list; | |
1da177e4 | 1320 | bio_dirty_list = NULL; |
24d5493f | 1321 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1322 | |
24d5493f CH |
1323 | while ((bio = next) != NULL) { |
1324 | next = bio->bi_private; | |
1da177e4 | 1325 | |
d241a95f | 1326 | bio_release_pages(bio, true); |
1da177e4 | 1327 | bio_put(bio); |
1da177e4 LT |
1328 | } |
1329 | } | |
1330 | ||
1331 | void bio_check_pages_dirty(struct bio *bio) | |
1332 | { | |
cb34e057 | 1333 | struct bio_vec *bvec; |
24d5493f | 1334 | unsigned long flags; |
6dc4f100 | 1335 | struct bvec_iter_all iter_all; |
1da177e4 | 1336 | |
2b070cfe | 1337 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1338 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1339 | goto defer; | |
1da177e4 LT |
1340 | } |
1341 | ||
d241a95f | 1342 | bio_release_pages(bio, false); |
24d5493f CH |
1343 | bio_put(bio); |
1344 | return; | |
1345 | defer: | |
1346 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1347 | bio->bi_private = bio_dirty_list; | |
1348 | bio_dirty_list = bio; | |
1349 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1350 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1351 | } |
1352 | ||
c4cf5261 JA |
1353 | static inline bool bio_remaining_done(struct bio *bio) |
1354 | { | |
1355 | /* | |
1356 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1357 | * we always end io on the first invocation. | |
1358 | */ | |
1359 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1360 | return true; | |
1361 | ||
1362 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1363 | ||
326e1dbb | 1364 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1365 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1366 | return true; |
326e1dbb | 1367 | } |
c4cf5261 JA |
1368 | |
1369 | return false; | |
1370 | } | |
1371 | ||
1da177e4 LT |
1372 | /** |
1373 | * bio_endio - end I/O on a bio | |
1374 | * @bio: bio | |
1da177e4 LT |
1375 | * |
1376 | * Description: | |
4246a0b6 CH |
1377 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1378 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1379 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1380 | * |
1381 | * bio_endio() can be called several times on a bio that has been chained | |
1382 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
1383 | * last time. At this point the BLK_TA_COMPLETE tracing event will be | |
1384 | * generated if BIO_TRACE_COMPLETION is set. | |
1da177e4 | 1385 | **/ |
4246a0b6 | 1386 | void bio_endio(struct bio *bio) |
1da177e4 | 1387 | { |
ba8c6967 | 1388 | again: |
2b885517 | 1389 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1390 | return; |
7c20f116 CH |
1391 | if (!bio_integrity_endio(bio)) |
1392 | return; | |
1da177e4 | 1393 | |
309dca30 CH |
1394 | if (bio->bi_bdev) |
1395 | rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); | |
67b42d0b | 1396 | |
ba8c6967 CH |
1397 | /* |
1398 | * Need to have a real endio function for chained bios, otherwise | |
1399 | * various corner cases will break (like stacking block devices that | |
1400 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1401 | * recursion and blowing the stack. Tail call optimization would | |
1402 | * handle this, but compiling with frame pointers also disables | |
1403 | * gcc's sibling call optimization. | |
1404 | */ | |
1405 | if (bio->bi_end_io == bio_chain_endio) { | |
1406 | bio = __bio_chain_endio(bio); | |
1407 | goto again; | |
196d38bc | 1408 | } |
ba8c6967 | 1409 | |
309dca30 CH |
1410 | if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
1411 | trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); | |
fbbaf700 N |
1412 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
1413 | } | |
1414 | ||
9e234eea | 1415 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1416 | /* release cgroup info */ |
1417 | bio_uninit(bio); | |
ba8c6967 CH |
1418 | if (bio->bi_end_io) |
1419 | bio->bi_end_io(bio); | |
1da177e4 | 1420 | } |
a112a71d | 1421 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1422 | |
20d0189b KO |
1423 | /** |
1424 | * bio_split - split a bio | |
1425 | * @bio: bio to split | |
1426 | * @sectors: number of sectors to split from the front of @bio | |
1427 | * @gfp: gfp mask | |
1428 | * @bs: bio set to allocate from | |
1429 | * | |
1430 | * Allocates and returns a new bio which represents @sectors from the start of | |
1431 | * @bio, and updates @bio to represent the remaining sectors. | |
1432 | * | |
f3f5da62 | 1433 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1434 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1435 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1436 | */ |
1437 | struct bio *bio_split(struct bio *bio, int sectors, | |
1438 | gfp_t gfp, struct bio_set *bs) | |
1439 | { | |
f341a4d3 | 1440 | struct bio *split; |
20d0189b KO |
1441 | |
1442 | BUG_ON(sectors <= 0); | |
1443 | BUG_ON(sectors >= bio_sectors(bio)); | |
1444 | ||
0512a75b KB |
1445 | /* Zone append commands cannot be split */ |
1446 | if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) | |
1447 | return NULL; | |
1448 | ||
f9d03f96 | 1449 | split = bio_clone_fast(bio, gfp, bs); |
20d0189b KO |
1450 | if (!split) |
1451 | return NULL; | |
1452 | ||
1453 | split->bi_iter.bi_size = sectors << 9; | |
1454 | ||
1455 | if (bio_integrity(split)) | |
fbd08e76 | 1456 | bio_integrity_trim(split); |
20d0189b KO |
1457 | |
1458 | bio_advance(bio, split->bi_iter.bi_size); | |
1459 | ||
fbbaf700 | 1460 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1461 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1462 | |
20d0189b KO |
1463 | return split; |
1464 | } | |
1465 | EXPORT_SYMBOL(bio_split); | |
1466 | ||
6678d83f KO |
1467 | /** |
1468 | * bio_trim - trim a bio | |
1469 | * @bio: bio to trim | |
1470 | * @offset: number of sectors to trim from the front of @bio | |
1471 | * @size: size we want to trim @bio to, in sectors | |
1472 | */ | |
1473 | void bio_trim(struct bio *bio, int offset, int size) | |
1474 | { | |
1475 | /* 'bio' is a cloned bio which we need to trim to match | |
1476 | * the given offset and size. | |
6678d83f | 1477 | */ |
6678d83f KO |
1478 | |
1479 | size <<= 9; | |
4f024f37 | 1480 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1481 | return; |
1482 | ||
6678d83f | 1483 | bio_advance(bio, offset << 9); |
4f024f37 | 1484 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1485 | |
1486 | if (bio_integrity(bio)) | |
fbd08e76 | 1487 | bio_integrity_trim(bio); |
376a78ab | 1488 | |
6678d83f KO |
1489 | } |
1490 | EXPORT_SYMBOL_GPL(bio_trim); | |
1491 | ||
1da177e4 LT |
1492 | /* |
1493 | * create memory pools for biovec's in a bio_set. | |
1494 | * use the global biovec slabs created for general use. | |
1495 | */ | |
8aa6ba2f | 1496 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1497 | { |
ed996a52 | 1498 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
1da177e4 | 1499 | |
8aa6ba2f | 1500 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1501 | } |
1502 | ||
917a38c7 KO |
1503 | /* |
1504 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1505 | * | |
1506 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1507 | * kzalloc()). | |
1508 | */ | |
1509 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1510 | { |
df2cb6da KO |
1511 | if (bs->rescue_workqueue) |
1512 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1513 | bs->rescue_workqueue = NULL; |
df2cb6da | 1514 | |
8aa6ba2f KO |
1515 | mempool_exit(&bs->bio_pool); |
1516 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1517 | |
7878cba9 | 1518 | bioset_integrity_free(bs); |
917a38c7 KO |
1519 | if (bs->bio_slab) |
1520 | bio_put_slab(bs); | |
1521 | bs->bio_slab = NULL; | |
1522 | } | |
1523 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1524 | |
917a38c7 KO |
1525 | /** |
1526 | * bioset_init - Initialize a bio_set | |
dad08527 | 1527 | * @bs: pool to initialize |
917a38c7 KO |
1528 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1529 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1530 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1531 | * and %BIOSET_NEED_RESCUER | |
1532 | * | |
dad08527 KO |
1533 | * Description: |
1534 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1535 | * to ask for a number of bytes to be allocated in front of the bio. | |
1536 | * Front pad allocation is useful for embedding the bio inside | |
1537 | * another structure, to avoid allocating extra data to go with the bio. | |
1538 | * Note that the bio must be embedded at the END of that structure always, | |
1539 | * or things will break badly. | |
1540 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
1541 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
1542 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
1543 | * dispatch queued requests when the mempool runs out of space. | |
1544 | * | |
917a38c7 KO |
1545 | */ |
1546 | int bioset_init(struct bio_set *bs, | |
1547 | unsigned int pool_size, | |
1548 | unsigned int front_pad, | |
1549 | int flags) | |
1550 | { | |
917a38c7 | 1551 | bs->front_pad = front_pad; |
9f180e31 ML |
1552 | if (flags & BIOSET_NEED_BVECS) |
1553 | bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1554 | else | |
1555 | bs->back_pad = 0; | |
917a38c7 KO |
1556 | |
1557 | spin_lock_init(&bs->rescue_lock); | |
1558 | bio_list_init(&bs->rescue_list); | |
1559 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1560 | ||
49d1ec85 | 1561 | bs->bio_slab = bio_find_or_create_slab(bs); |
917a38c7 KO |
1562 | if (!bs->bio_slab) |
1563 | return -ENOMEM; | |
1564 | ||
1565 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) | |
1566 | goto bad; | |
1567 | ||
1568 | if ((flags & BIOSET_NEED_BVECS) && | |
1569 | biovec_init_pool(&bs->bvec_pool, pool_size)) | |
1570 | goto bad; | |
1571 | ||
1572 | if (!(flags & BIOSET_NEED_RESCUER)) | |
1573 | return 0; | |
1574 | ||
1575 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1576 | if (!bs->rescue_workqueue) | |
1577 | goto bad; | |
1578 | ||
1579 | return 0; | |
1580 | bad: | |
1581 | bioset_exit(bs); | |
1582 | return -ENOMEM; | |
1583 | } | |
1584 | EXPORT_SYMBOL(bioset_init); | |
1585 | ||
28e89fd9 JA |
1586 | /* |
1587 | * Initialize and setup a new bio_set, based on the settings from | |
1588 | * another bio_set. | |
1589 | */ | |
1590 | int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) | |
1591 | { | |
1592 | int flags; | |
1593 | ||
1594 | flags = 0; | |
1595 | if (src->bvec_pool.min_nr) | |
1596 | flags |= BIOSET_NEED_BVECS; | |
1597 | if (src->rescue_workqueue) | |
1598 | flags |= BIOSET_NEED_RESCUER; | |
1599 | ||
1600 | return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); | |
1601 | } | |
1602 | EXPORT_SYMBOL(bioset_init_from_src); | |
1603 | ||
de76fd89 | 1604 | static int __init init_bio(void) |
1da177e4 LT |
1605 | { |
1606 | int i; | |
1607 | ||
2b24e6f6 JT |
1608 | BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); |
1609 | ||
7878cba9 | 1610 | bio_integrity_init(); |
de76fd89 CH |
1611 | |
1612 | for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { | |
1613 | struct biovec_slab *bvs = bvec_slabs + i; | |
1614 | ||
1615 | bvs->slab = kmem_cache_create(bvs->name, | |
1616 | bvs->nr_vecs * sizeof(struct bio_vec), 0, | |
1617 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
1618 | } | |
1da177e4 | 1619 | |
f4f8154a | 1620 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
1da177e4 LT |
1621 | panic("bio: can't allocate bios\n"); |
1622 | ||
f4f8154a | 1623 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
1624 | panic("bio: can't create integrity pool\n"); |
1625 | ||
1da177e4 LT |
1626 | return 0; |
1627 | } | |
1da177e4 | 1628 | subsys_initcall(init_bio); |