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" }, | |
a8affc03 | 36 | { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" }, |
1da177e4 | 37 | }; |
6ac0b715 | 38 | |
7a800a20 CH |
39 | static struct biovec_slab *biovec_slab(unsigned short nr_vecs) |
40 | { | |
41 | switch (nr_vecs) { | |
42 | /* smaller bios use inline vecs */ | |
43 | case 5 ... 16: | |
44 | return &bvec_slabs[0]; | |
45 | case 17 ... 64: | |
46 | return &bvec_slabs[1]; | |
47 | case 65 ... 128: | |
48 | return &bvec_slabs[2]; | |
a8affc03 | 49 | case 129 ... BIO_MAX_VECS: |
7a800a20 CH |
50 | return &bvec_slabs[3]; |
51 | default: | |
52 | BUG(); | |
53 | return NULL; | |
54 | } | |
55 | } | |
1da177e4 | 56 | |
1da177e4 LT |
57 | /* |
58 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
59 | * IO code that does not need private memory pools. | |
60 | */ | |
f4f8154a | 61 | struct bio_set fs_bio_set; |
3f86a82a | 62 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 63 | |
bb799ca0 JA |
64 | /* |
65 | * Our slab pool management | |
66 | */ | |
67 | struct bio_slab { | |
68 | struct kmem_cache *slab; | |
69 | unsigned int slab_ref; | |
70 | unsigned int slab_size; | |
71 | char name[8]; | |
72 | }; | |
73 | static DEFINE_MUTEX(bio_slab_lock); | |
49d1ec85 | 74 | static DEFINE_XARRAY(bio_slabs); |
bb799ca0 | 75 | |
49d1ec85 | 76 | static struct bio_slab *create_bio_slab(unsigned int size) |
bb799ca0 | 77 | { |
49d1ec85 | 78 | struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); |
bb799ca0 | 79 | |
49d1ec85 ML |
80 | if (!bslab) |
81 | return NULL; | |
bb799ca0 | 82 | |
49d1ec85 ML |
83 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); |
84 | bslab->slab = kmem_cache_create(bslab->name, size, | |
85 | ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL); | |
86 | if (!bslab->slab) | |
87 | goto fail_alloc_slab; | |
bb799ca0 | 88 | |
49d1ec85 ML |
89 | bslab->slab_ref = 1; |
90 | bslab->slab_size = size; | |
bb799ca0 | 91 | |
49d1ec85 ML |
92 | if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) |
93 | return bslab; | |
bb799ca0 | 94 | |
49d1ec85 | 95 | kmem_cache_destroy(bslab->slab); |
bb799ca0 | 96 | |
49d1ec85 ML |
97 | fail_alloc_slab: |
98 | kfree(bslab); | |
99 | return NULL; | |
100 | } | |
bb799ca0 | 101 | |
49d1ec85 ML |
102 | static inline unsigned int bs_bio_slab_size(struct bio_set *bs) |
103 | { | |
9f180e31 | 104 | return bs->front_pad + sizeof(struct bio) + bs->back_pad; |
49d1ec85 | 105 | } |
bb799ca0 | 106 | |
49d1ec85 ML |
107 | static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) |
108 | { | |
109 | unsigned int size = bs_bio_slab_size(bs); | |
110 | struct bio_slab *bslab; | |
bb799ca0 | 111 | |
49d1ec85 ML |
112 | mutex_lock(&bio_slab_lock); |
113 | bslab = xa_load(&bio_slabs, size); | |
114 | if (bslab) | |
115 | bslab->slab_ref++; | |
116 | else | |
117 | bslab = create_bio_slab(size); | |
bb799ca0 | 118 | mutex_unlock(&bio_slab_lock); |
49d1ec85 ML |
119 | |
120 | if (bslab) | |
121 | return bslab->slab; | |
122 | return NULL; | |
bb799ca0 JA |
123 | } |
124 | ||
125 | static void bio_put_slab(struct bio_set *bs) | |
126 | { | |
127 | struct bio_slab *bslab = NULL; | |
49d1ec85 | 128 | unsigned int slab_size = bs_bio_slab_size(bs); |
bb799ca0 JA |
129 | |
130 | mutex_lock(&bio_slab_lock); | |
131 | ||
49d1ec85 | 132 | bslab = xa_load(&bio_slabs, slab_size); |
bb799ca0 JA |
133 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
134 | goto out; | |
135 | ||
49d1ec85 ML |
136 | WARN_ON_ONCE(bslab->slab != bs->bio_slab); |
137 | ||
bb799ca0 JA |
138 | WARN_ON(!bslab->slab_ref); |
139 | ||
140 | if (--bslab->slab_ref) | |
141 | goto out; | |
142 | ||
49d1ec85 ML |
143 | xa_erase(&bio_slabs, slab_size); |
144 | ||
bb799ca0 | 145 | kmem_cache_destroy(bslab->slab); |
49d1ec85 | 146 | kfree(bslab); |
bb799ca0 JA |
147 | |
148 | out: | |
149 | mutex_unlock(&bio_slab_lock); | |
150 | } | |
151 | ||
7a800a20 | 152 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs) |
7ba1ba12 | 153 | { |
a8affc03 | 154 | BIO_BUG_ON(nr_vecs > BIO_MAX_VECS); |
ed996a52 | 155 | |
a8affc03 | 156 | if (nr_vecs == BIO_MAX_VECS) |
9f060e22 | 157 | mempool_free(bv, pool); |
7a800a20 CH |
158 | else if (nr_vecs > BIO_INLINE_VECS) |
159 | kmem_cache_free(biovec_slab(nr_vecs)->slab, bv); | |
bb799ca0 | 160 | } |
bb799ca0 | 161 | |
f2c3eb9b CH |
162 | /* |
163 | * Make the first allocation restricted and don't dump info on allocation | |
164 | * failures, since we'll fall back to the mempool in case of failure. | |
165 | */ | |
166 | static inline gfp_t bvec_alloc_gfp(gfp_t gfp) | |
167 | { | |
168 | return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | | |
169 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
bb799ca0 JA |
170 | } |
171 | ||
7a800a20 CH |
172 | struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, |
173 | gfp_t gfp_mask) | |
1da177e4 | 174 | { |
7a800a20 | 175 | struct biovec_slab *bvs = biovec_slab(*nr_vecs); |
1da177e4 | 176 | |
7a800a20 | 177 | if (WARN_ON_ONCE(!bvs)) |
7ff9345f | 178 | return NULL; |
7ff9345f JA |
179 | |
180 | /* | |
7a800a20 CH |
181 | * Upgrade the nr_vecs request to take full advantage of the allocation. |
182 | * We also rely on this in the bvec_free path. | |
7ff9345f | 183 | */ |
7a800a20 | 184 | *nr_vecs = bvs->nr_vecs; |
7ff9345f | 185 | |
7ff9345f | 186 | /* |
f007a3d6 CH |
187 | * Try a slab allocation first for all smaller allocations. If that |
188 | * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. | |
a8affc03 | 189 | * The mempool is sized to handle up to BIO_MAX_VECS entries. |
7ff9345f | 190 | */ |
a8affc03 | 191 | if (*nr_vecs < BIO_MAX_VECS) { |
f007a3d6 | 192 | struct bio_vec *bvl; |
1da177e4 | 193 | |
f2c3eb9b | 194 | bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); |
7a800a20 | 195 | if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
f007a3d6 | 196 | return bvl; |
a8affc03 | 197 | *nr_vecs = BIO_MAX_VECS; |
7ff9345f JA |
198 | } |
199 | ||
f007a3d6 | 200 | return mempool_alloc(pool, gfp_mask); |
1da177e4 LT |
201 | } |
202 | ||
9ae3b3f5 | 203 | void bio_uninit(struct bio *bio) |
1da177e4 | 204 | { |
db9819c7 CH |
205 | #ifdef CONFIG_BLK_CGROUP |
206 | if (bio->bi_blkg) { | |
207 | blkg_put(bio->bi_blkg); | |
208 | bio->bi_blkg = NULL; | |
209 | } | |
210 | #endif | |
ece841ab JT |
211 | if (bio_integrity(bio)) |
212 | bio_integrity_free(bio); | |
a892c8d5 ST |
213 | |
214 | bio_crypt_free_ctx(bio); | |
4254bba1 | 215 | } |
9ae3b3f5 | 216 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 217 | |
4254bba1 KO |
218 | static void bio_free(struct bio *bio) |
219 | { | |
220 | struct bio_set *bs = bio->bi_pool; | |
221 | void *p; | |
222 | ||
9ae3b3f5 | 223 | bio_uninit(bio); |
4254bba1 KO |
224 | |
225 | if (bs) { | |
7a800a20 | 226 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); |
4254bba1 KO |
227 | |
228 | /* | |
229 | * If we have front padding, adjust the bio pointer before freeing | |
230 | */ | |
231 | p = bio; | |
bb799ca0 JA |
232 | p -= bs->front_pad; |
233 | ||
8aa6ba2f | 234 | mempool_free(p, &bs->bio_pool); |
4254bba1 KO |
235 | } else { |
236 | /* Bio was allocated by bio_kmalloc() */ | |
237 | kfree(bio); | |
238 | } | |
3676347a PO |
239 | } |
240 | ||
9ae3b3f5 JA |
241 | /* |
242 | * Users of this function have their own bio allocation. Subsequently, | |
243 | * they must remember to pair any call to bio_init() with bio_uninit() | |
244 | * when IO has completed, or when the bio is released. | |
245 | */ | |
3a83f467 ML |
246 | void bio_init(struct bio *bio, struct bio_vec *table, |
247 | unsigned short max_vecs) | |
1da177e4 | 248 | { |
da521626 JA |
249 | bio->bi_next = NULL; |
250 | bio->bi_bdev = NULL; | |
251 | bio->bi_opf = 0; | |
252 | bio->bi_flags = 0; | |
253 | bio->bi_ioprio = 0; | |
254 | bio->bi_write_hint = 0; | |
255 | bio->bi_status = 0; | |
256 | bio->bi_iter.bi_sector = 0; | |
257 | bio->bi_iter.bi_size = 0; | |
258 | bio->bi_iter.bi_idx = 0; | |
259 | bio->bi_iter.bi_bvec_done = 0; | |
260 | bio->bi_end_io = NULL; | |
261 | bio->bi_private = NULL; | |
262 | #ifdef CONFIG_BLK_CGROUP | |
263 | bio->bi_blkg = NULL; | |
264 | bio->bi_issue.value = 0; | |
265 | #ifdef CONFIG_BLK_CGROUP_IOCOST | |
266 | bio->bi_iocost_cost = 0; | |
267 | #endif | |
268 | #endif | |
269 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION | |
270 | bio->bi_crypt_context = NULL; | |
271 | #endif | |
272 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
273 | bio->bi_integrity = NULL; | |
274 | #endif | |
275 | bio->bi_vcnt = 0; | |
276 | ||
c4cf5261 | 277 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 278 | atomic_set(&bio->__bi_cnt, 1); |
3a83f467 | 279 | |
3a83f467 | 280 | bio->bi_max_vecs = max_vecs; |
da521626 JA |
281 | bio->bi_io_vec = table; |
282 | bio->bi_pool = NULL; | |
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 | { | |
9ae3b3f5 | 298 | bio_uninit(bio); |
f44b48c7 | 299 | memset(bio, 0, BIO_RESET_BYTES); |
c4cf5261 | 300 | atomic_set(&bio->__bi_remaining, 1); |
f44b48c7 KO |
301 | } |
302 | EXPORT_SYMBOL(bio_reset); | |
303 | ||
38f8baae | 304 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 305 | { |
4246a0b6 CH |
306 | struct bio *parent = bio->bi_private; |
307 | ||
3edf5346 | 308 | if (bio->bi_status && !parent->bi_status) |
4e4cbee9 | 309 | parent->bi_status = bio->bi_status; |
196d38bc | 310 | bio_put(bio); |
38f8baae CH |
311 | return parent; |
312 | } | |
313 | ||
314 | static void bio_chain_endio(struct bio *bio) | |
315 | { | |
316 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
317 | } |
318 | ||
319 | /** | |
320 | * bio_chain - chain bio completions | |
1051a902 | 321 | * @bio: the target bio |
5b874af6 | 322 | * @parent: the parent bio of @bio |
196d38bc KO |
323 | * |
324 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
325 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
326 | * completed; the chained bio will also be freed when it completes. | |
327 | * | |
328 | * The caller must not set bi_private or bi_end_io in @bio. | |
329 | */ | |
330 | void bio_chain(struct bio *bio, struct bio *parent) | |
331 | { | |
332 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
333 | ||
334 | bio->bi_private = parent; | |
335 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 336 | bio_inc_remaining(parent); |
196d38bc KO |
337 | } |
338 | EXPORT_SYMBOL(bio_chain); | |
339 | ||
df2cb6da KO |
340 | static void bio_alloc_rescue(struct work_struct *work) |
341 | { | |
342 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
343 | struct bio *bio; | |
344 | ||
345 | while (1) { | |
346 | spin_lock(&bs->rescue_lock); | |
347 | bio = bio_list_pop(&bs->rescue_list); | |
348 | spin_unlock(&bs->rescue_lock); | |
349 | ||
350 | if (!bio) | |
351 | break; | |
352 | ||
ed00aabd | 353 | submit_bio_noacct(bio); |
df2cb6da KO |
354 | } |
355 | } | |
356 | ||
357 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
358 | { | |
359 | struct bio_list punt, nopunt; | |
360 | struct bio *bio; | |
361 | ||
47e0fb46 N |
362 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
363 | return; | |
df2cb6da KO |
364 | /* |
365 | * In order to guarantee forward progress we must punt only bios that | |
366 | * were allocated from this bio_set; otherwise, if there was a bio on | |
367 | * there for a stacking driver higher up in the stack, processing it | |
368 | * could require allocating bios from this bio_set, and doing that from | |
369 | * our own rescuer would be bad. | |
370 | * | |
371 | * Since bio lists are singly linked, pop them all instead of trying to | |
372 | * remove from the middle of the list: | |
373 | */ | |
374 | ||
375 | bio_list_init(&punt); | |
376 | bio_list_init(&nopunt); | |
377 | ||
f5fe1b51 | 378 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 379 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 380 | current->bio_list[0] = nopunt; |
df2cb6da | 381 | |
f5fe1b51 N |
382 | bio_list_init(&nopunt); |
383 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
384 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
385 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
386 | |
387 | spin_lock(&bs->rescue_lock); | |
388 | bio_list_merge(&bs->rescue_list, &punt); | |
389 | spin_unlock(&bs->rescue_lock); | |
390 | ||
391 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
392 | } | |
393 | ||
1da177e4 LT |
394 | /** |
395 | * bio_alloc_bioset - allocate a bio for I/O | |
519c8e9f | 396 | * @gfp_mask: the GFP_* mask given to the slab allocator |
1da177e4 | 397 | * @nr_iovecs: number of iovecs to pre-allocate |
db18efac | 398 | * @bs: the bio_set to allocate from. |
1da177e4 | 399 | * |
3175199a | 400 | * Allocate a bio from the mempools in @bs. |
3f86a82a | 401 | * |
3175199a CH |
402 | * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to |
403 | * allocate a bio. This is due to the mempool guarantees. To make this work, | |
404 | * callers must never allocate more than 1 bio at a time from the general pool. | |
405 | * Callers that need to allocate more than 1 bio must always submit the | |
406 | * previously allocated bio for IO before attempting to allocate a new one. | |
407 | * Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 408 | * |
3175199a CH |
409 | * Note that when running under submit_bio_noacct() (i.e. any block driver), |
410 | * bios are not submitted until after you return - see the code in | |
411 | * submit_bio_noacct() that converts recursion into iteration, to prevent | |
412 | * stack overflows. | |
df2cb6da | 413 | * |
3175199a CH |
414 | * This would normally mean allocating multiple bios under submit_bio_noacct() |
415 | * would be susceptible to deadlocks, but we have | |
416 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
417 | * thread. | |
df2cb6da | 418 | * |
3175199a CH |
419 | * However, we do not guarantee forward progress for allocations from other |
420 | * mempools. Doing multiple allocations from the same mempool under | |
421 | * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad | |
422 | * for per bio allocations. | |
df2cb6da | 423 | * |
3175199a | 424 | * Returns: Pointer to new bio on success, NULL on failure. |
3f86a82a | 425 | */ |
0f2e6ab8 | 426 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, |
7a88fa19 | 427 | struct bio_set *bs) |
1da177e4 | 428 | { |
df2cb6da | 429 | gfp_t saved_gfp = gfp_mask; |
451a9ebf TH |
430 | struct bio *bio; |
431 | void *p; | |
432 | ||
3175199a CH |
433 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
434 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) | |
435 | return NULL; | |
df2cb6da | 436 | |
3175199a CH |
437 | /* |
438 | * submit_bio_noacct() converts recursion to iteration; this means if | |
439 | * we're running beneath it, any bios we allocate and submit will not be | |
440 | * submitted (and thus freed) until after we return. | |
441 | * | |
442 | * This exposes us to a potential deadlock if we allocate multiple bios | |
443 | * from the same bio_set() while running underneath submit_bio_noacct(). | |
444 | * If we were to allocate multiple bios (say a stacking block driver | |
445 | * that was splitting bios), we would deadlock if we exhausted the | |
446 | * mempool's reserve. | |
447 | * | |
448 | * We solve this, and guarantee forward progress, with a rescuer | |
449 | * workqueue per bio_set. If we go to allocate and there are bios on | |
450 | * current->bio_list, we first try the allocation without | |
451 | * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be | |
452 | * blocking to the rescuer workqueue before we retry with the original | |
453 | * gfp_flags. | |
454 | */ | |
455 | if (current->bio_list && | |
456 | (!bio_list_empty(¤t->bio_list[0]) || | |
457 | !bio_list_empty(¤t->bio_list[1])) && | |
458 | bs->rescue_workqueue) | |
459 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; | |
460 | ||
461 | p = mempool_alloc(&bs->bio_pool, gfp_mask); | |
462 | if (!p && gfp_mask != saved_gfp) { | |
463 | punt_bios_to_rescuer(bs); | |
464 | gfp_mask = saved_gfp; | |
8aa6ba2f | 465 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
3f86a82a | 466 | } |
451a9ebf TH |
467 | if (unlikely(!p)) |
468 | return NULL; | |
1da177e4 | 469 | |
3175199a CH |
470 | bio = p + bs->front_pad; |
471 | if (nr_iovecs > BIO_INLINE_VECS) { | |
3175199a | 472 | struct bio_vec *bvl = NULL; |
34053979 | 473 | |
7a800a20 | 474 | bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); |
df2cb6da KO |
475 | if (!bvl && gfp_mask != saved_gfp) { |
476 | punt_bios_to_rescuer(bs); | |
477 | gfp_mask = saved_gfp; | |
7a800a20 | 478 | bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); |
df2cb6da | 479 | } |
34053979 IM |
480 | if (unlikely(!bvl)) |
481 | goto err_free; | |
a38352e0 | 482 | |
7a800a20 | 483 | bio_init(bio, bvl, nr_iovecs); |
3f86a82a | 484 | } else if (nr_iovecs) { |
3175199a CH |
485 | bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); |
486 | } else { | |
487 | bio_init(bio, NULL, 0); | |
1da177e4 | 488 | } |
3f86a82a KO |
489 | |
490 | bio->bi_pool = bs; | |
1da177e4 | 491 | return bio; |
34053979 IM |
492 | |
493 | err_free: | |
8aa6ba2f | 494 | mempool_free(p, &bs->bio_pool); |
34053979 | 495 | return NULL; |
1da177e4 | 496 | } |
a112a71d | 497 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 498 | |
3175199a CH |
499 | /** |
500 | * bio_kmalloc - kmalloc a bio for I/O | |
501 | * @gfp_mask: the GFP_* mask given to the slab allocator | |
502 | * @nr_iovecs: number of iovecs to pre-allocate | |
503 | * | |
504 | * Use kmalloc to allocate and initialize a bio. | |
505 | * | |
506 | * Returns: Pointer to new bio on success, NULL on failure. | |
507 | */ | |
0f2e6ab8 | 508 | struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) |
3175199a CH |
509 | { |
510 | struct bio *bio; | |
511 | ||
512 | if (nr_iovecs > UIO_MAXIOV) | |
513 | return NULL; | |
514 | ||
515 | bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); | |
516 | if (unlikely(!bio)) | |
517 | return NULL; | |
518 | bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); | |
519 | bio->bi_pool = NULL; | |
520 | return bio; | |
521 | } | |
522 | EXPORT_SYMBOL(bio_kmalloc); | |
523 | ||
6f822e1b | 524 | void zero_fill_bio(struct bio *bio) |
1da177e4 LT |
525 | { |
526 | unsigned long flags; | |
7988613b KO |
527 | struct bio_vec bv; |
528 | struct bvec_iter iter; | |
1da177e4 | 529 | |
6f822e1b | 530 | bio_for_each_segment(bv, bio, iter) { |
7988613b KO |
531 | char *data = bvec_kmap_irq(&bv, &flags); |
532 | memset(data, 0, bv.bv_len); | |
533 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
534 | bvec_kunmap_irq(data, &flags); |
535 | } | |
536 | } | |
6f822e1b | 537 | EXPORT_SYMBOL(zero_fill_bio); |
1da177e4 | 538 | |
83c9c547 ML |
539 | /** |
540 | * bio_truncate - truncate the bio to small size of @new_size | |
541 | * @bio: the bio to be truncated | |
542 | * @new_size: new size for truncating the bio | |
543 | * | |
544 | * Description: | |
545 | * Truncate the bio to new size of @new_size. If bio_op(bio) is | |
546 | * REQ_OP_READ, zero the truncated part. This function should only | |
547 | * be used for handling corner cases, such as bio eod. | |
548 | */ | |
85a8ce62 ML |
549 | void bio_truncate(struct bio *bio, unsigned new_size) |
550 | { | |
551 | struct bio_vec bv; | |
552 | struct bvec_iter iter; | |
553 | unsigned int done = 0; | |
554 | bool truncated = false; | |
555 | ||
556 | if (new_size >= bio->bi_iter.bi_size) | |
557 | return; | |
558 | ||
83c9c547 | 559 | if (bio_op(bio) != REQ_OP_READ) |
85a8ce62 ML |
560 | goto exit; |
561 | ||
562 | bio_for_each_segment(bv, bio, iter) { | |
563 | if (done + bv.bv_len > new_size) { | |
564 | unsigned offset; | |
565 | ||
566 | if (!truncated) | |
567 | offset = new_size - done; | |
568 | else | |
569 | offset = 0; | |
570 | zero_user(bv.bv_page, offset, bv.bv_len - offset); | |
571 | truncated = true; | |
572 | } | |
573 | done += bv.bv_len; | |
574 | } | |
575 | ||
576 | exit: | |
577 | /* | |
578 | * Don't touch bvec table here and make it really immutable, since | |
579 | * fs bio user has to retrieve all pages via bio_for_each_segment_all | |
580 | * in its .end_bio() callback. | |
581 | * | |
582 | * It is enough to truncate bio by updating .bi_size since we can make | |
583 | * correct bvec with the updated .bi_size for drivers. | |
584 | */ | |
585 | bio->bi_iter.bi_size = new_size; | |
586 | } | |
587 | ||
29125ed6 CH |
588 | /** |
589 | * guard_bio_eod - truncate a BIO to fit the block device | |
590 | * @bio: bio to truncate | |
591 | * | |
592 | * This allows us to do IO even on the odd last sectors of a device, even if the | |
593 | * block size is some multiple of the physical sector size. | |
594 | * | |
595 | * We'll just truncate the bio to the size of the device, and clear the end of | |
596 | * the buffer head manually. Truly out-of-range accesses will turn into actual | |
597 | * I/O errors, this only handles the "we need to be able to do I/O at the final | |
598 | * sector" case. | |
599 | */ | |
600 | void guard_bio_eod(struct bio *bio) | |
601 | { | |
309dca30 | 602 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
29125ed6 CH |
603 | |
604 | if (!maxsector) | |
605 | return; | |
606 | ||
607 | /* | |
608 | * If the *whole* IO is past the end of the device, | |
609 | * let it through, and the IO layer will turn it into | |
610 | * an EIO. | |
611 | */ | |
612 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) | |
613 | return; | |
614 | ||
615 | maxsector -= bio->bi_iter.bi_sector; | |
616 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) | |
617 | return; | |
618 | ||
619 | bio_truncate(bio, maxsector << 9); | |
620 | } | |
621 | ||
1da177e4 LT |
622 | /** |
623 | * bio_put - release a reference to a bio | |
624 | * @bio: bio to release reference to | |
625 | * | |
626 | * Description: | |
627 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 628 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
629 | **/ |
630 | void bio_put(struct bio *bio) | |
631 | { | |
dac56212 | 632 | if (!bio_flagged(bio, BIO_REFFED)) |
4254bba1 | 633 | bio_free(bio); |
dac56212 JA |
634 | else { |
635 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
636 | ||
637 | /* | |
638 | * last put frees it | |
639 | */ | |
640 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
641 | bio_free(bio); | |
642 | } | |
1da177e4 | 643 | } |
a112a71d | 644 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 645 | |
59d276fe KO |
646 | /** |
647 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
648 | * @bio: destination bio | |
649 | * @bio_src: bio to clone | |
650 | * | |
651 | * Clone a &bio. Caller will own the returned bio, but not | |
652 | * the actual data it points to. Reference count of returned | |
653 | * bio will be one. | |
654 | * | |
655 | * Caller must ensure that @bio_src is not freed before @bio. | |
656 | */ | |
657 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
658 | { | |
7a800a20 | 659 | WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); |
59d276fe KO |
660 | |
661 | /* | |
309dca30 | 662 | * most users will be overriding ->bi_bdev with a new target, |
59d276fe KO |
663 | * so we don't set nor calculate new physical/hw segment counts here |
664 | */ | |
309dca30 | 665 | bio->bi_bdev = bio_src->bi_bdev; |
b7c44ed9 | 666 | bio_set_flag(bio, BIO_CLONED); |
111be883 SL |
667 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
668 | bio_set_flag(bio, BIO_THROTTLED); | |
46bbf653 CH |
669 | if (bio_flagged(bio_src, BIO_REMAPPED)) |
670 | bio_set_flag(bio, BIO_REMAPPED); | |
1eff9d32 | 671 | bio->bi_opf = bio_src->bi_opf; |
ca474b73 | 672 | bio->bi_ioprio = bio_src->bi_ioprio; |
cb6934f8 | 673 | bio->bi_write_hint = bio_src->bi_write_hint; |
59d276fe KO |
674 | bio->bi_iter = bio_src->bi_iter; |
675 | bio->bi_io_vec = bio_src->bi_io_vec; | |
20bd723e | 676 | |
db6638d7 | 677 | bio_clone_blkg_association(bio, bio_src); |
e439bedf | 678 | blkcg_bio_issue_init(bio); |
59d276fe KO |
679 | } |
680 | EXPORT_SYMBOL(__bio_clone_fast); | |
681 | ||
682 | /** | |
683 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
684 | * @bio: bio to clone | |
685 | * @gfp_mask: allocation priority | |
686 | * @bs: bio_set to allocate from | |
687 | * | |
688 | * Like __bio_clone_fast, only also allocates the returned bio | |
689 | */ | |
690 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
691 | { | |
692 | struct bio *b; | |
693 | ||
694 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
695 | if (!b) | |
696 | return NULL; | |
697 | ||
698 | __bio_clone_fast(b, bio); | |
699 | ||
07560151 EB |
700 | if (bio_crypt_clone(b, bio, gfp_mask) < 0) |
701 | goto err_put; | |
a892c8d5 | 702 | |
07560151 EB |
703 | if (bio_integrity(bio) && |
704 | bio_integrity_clone(b, bio, gfp_mask) < 0) | |
705 | goto err_put; | |
59d276fe KO |
706 | |
707 | return b; | |
07560151 EB |
708 | |
709 | err_put: | |
710 | bio_put(b); | |
711 | return NULL; | |
59d276fe KO |
712 | } |
713 | EXPORT_SYMBOL(bio_clone_fast); | |
714 | ||
5cbd28e3 CH |
715 | const char *bio_devname(struct bio *bio, char *buf) |
716 | { | |
309dca30 | 717 | return bdevname(bio->bi_bdev, buf); |
5cbd28e3 CH |
718 | } |
719 | EXPORT_SYMBOL(bio_devname); | |
720 | ||
5919482e ML |
721 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
722 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 723 | bool *same_page) |
5919482e | 724 | { |
d8166519 MWO |
725 | size_t bv_end = bv->bv_offset + bv->bv_len; |
726 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; | |
5919482e ML |
727 | phys_addr_t page_addr = page_to_phys(page); |
728 | ||
729 | if (vec_end_addr + 1 != page_addr + off) | |
730 | return false; | |
731 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
732 | return false; | |
52d52d1c | 733 | |
ff896738 | 734 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
d8166519 MWO |
735 | if (*same_page) |
736 | return true; | |
737 | return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); | |
5919482e ML |
738 | } |
739 | ||
e4581105 CH |
740 | /* |
741 | * Try to merge a page into a segment, while obeying the hardware segment | |
742 | * size limit. This is not for normal read/write bios, but for passthrough | |
743 | * or Zone Append operations that we can't split. | |
744 | */ | |
745 | static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, | |
746 | struct page *page, unsigned len, | |
747 | unsigned offset, bool *same_page) | |
489fbbcb | 748 | { |
384209cd | 749 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
489fbbcb ML |
750 | unsigned long mask = queue_segment_boundary(q); |
751 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
752 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
753 | ||
754 | if ((addr1 | mask) != (addr2 | mask)) | |
755 | return false; | |
489fbbcb ML |
756 | if (bv->bv_len + len > queue_max_segment_size(q)) |
757 | return false; | |
384209cd | 758 | return __bio_try_merge_page(bio, page, len, offset, same_page); |
489fbbcb ML |
759 | } |
760 | ||
1da177e4 | 761 | /** |
e4581105 CH |
762 | * bio_add_hw_page - attempt to add a page to a bio with hw constraints |
763 | * @q: the target queue | |
764 | * @bio: destination bio | |
765 | * @page: page to add | |
766 | * @len: vec entry length | |
767 | * @offset: vec entry offset | |
768 | * @max_sectors: maximum number of sectors that can be added | |
769 | * @same_page: return if the segment has been merged inside the same page | |
c66a14d0 | 770 | * |
e4581105 CH |
771 | * Add a page to a bio while respecting the hardware max_sectors, max_segment |
772 | * and gap limitations. | |
1da177e4 | 773 | */ |
e4581105 | 774 | int bio_add_hw_page(struct request_queue *q, struct bio *bio, |
19047087 | 775 | struct page *page, unsigned int len, unsigned int offset, |
e4581105 | 776 | unsigned int max_sectors, bool *same_page) |
1da177e4 | 777 | { |
1da177e4 LT |
778 | struct bio_vec *bvec; |
779 | ||
e4581105 | 780 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
1da177e4 LT |
781 | return 0; |
782 | ||
e4581105 | 783 | if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
784 | return 0; |
785 | ||
80cfd548 | 786 | if (bio->bi_vcnt > 0) { |
e4581105 | 787 | if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) |
384209cd | 788 | return len; |
320ea869 CH |
789 | |
790 | /* | |
791 | * If the queue doesn't support SG gaps and adding this segment | |
792 | * would create a gap, disallow it. | |
793 | */ | |
384209cd | 794 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
320ea869 CH |
795 | if (bvec_gap_to_prev(q, bvec, offset)) |
796 | return 0; | |
80cfd548 JA |
797 | } |
798 | ||
79d08f89 | 799 | if (bio_full(bio, len)) |
1da177e4 LT |
800 | return 0; |
801 | ||
14ccb66b | 802 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
803 | return 0; |
804 | ||
fcbf6a08 ML |
805 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
806 | bvec->bv_page = page; | |
807 | bvec->bv_len = len; | |
808 | bvec->bv_offset = offset; | |
809 | bio->bi_vcnt++; | |
dcdca753 | 810 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
811 | return len; |
812 | } | |
19047087 | 813 | |
e4581105 CH |
814 | /** |
815 | * bio_add_pc_page - attempt to add page to passthrough bio | |
816 | * @q: the target queue | |
817 | * @bio: destination bio | |
818 | * @page: page to add | |
819 | * @len: vec entry length | |
820 | * @offset: vec entry offset | |
821 | * | |
822 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
823 | * number of reasons, such as the bio being full or target block device | |
824 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
825 | * so it is always possible to add a single page to an empty bio. | |
826 | * | |
827 | * This should only be used by passthrough bios. | |
828 | */ | |
19047087 ML |
829 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, |
830 | struct page *page, unsigned int len, unsigned int offset) | |
831 | { | |
d1916c86 | 832 | bool same_page = false; |
e4581105 CH |
833 | return bio_add_hw_page(q, bio, page, len, offset, |
834 | queue_max_hw_sectors(q), &same_page); | |
19047087 | 835 | } |
a112a71d | 836 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 837 | |
ae29333f JT |
838 | /** |
839 | * bio_add_zone_append_page - attempt to add page to zone-append bio | |
840 | * @bio: destination bio | |
841 | * @page: page to add | |
842 | * @len: vec entry length | |
843 | * @offset: vec entry offset | |
844 | * | |
845 | * Attempt to add a page to the bio_vec maplist of a bio that will be submitted | |
846 | * for a zone-append request. This can fail for a number of reasons, such as the | |
847 | * bio being full or the target block device is not a zoned block device or | |
848 | * other limitations of the target block device. The target block device must | |
849 | * allow bio's up to PAGE_SIZE, so it is always possible to add a single page | |
850 | * to an empty bio. | |
851 | * | |
852 | * Returns: number of bytes added to the bio, or 0 in case of a failure. | |
853 | */ | |
854 | int bio_add_zone_append_page(struct bio *bio, struct page *page, | |
855 | unsigned int len, unsigned int offset) | |
856 | { | |
582cd91f | 857 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
ae29333f JT |
858 | bool same_page = false; |
859 | ||
860 | if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) | |
861 | return 0; | |
862 | ||
863 | if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) | |
864 | return 0; | |
865 | ||
866 | return bio_add_hw_page(q, bio, page, len, offset, | |
867 | queue_max_zone_append_sectors(q), &same_page); | |
868 | } | |
869 | EXPORT_SYMBOL_GPL(bio_add_zone_append_page); | |
870 | ||
1da177e4 | 871 | /** |
0aa69fd3 CH |
872 | * __bio_try_merge_page - try appending data to an existing bvec. |
873 | * @bio: destination bio | |
551879a4 | 874 | * @page: start page to add |
0aa69fd3 | 875 | * @len: length of the data to add |
551879a4 | 876 | * @off: offset of the data relative to @page |
ff896738 | 877 | * @same_page: return if the segment has been merged inside the same page |
1da177e4 | 878 | * |
0aa69fd3 | 879 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
3cf14889 | 880 | * useful optimisation for file systems with a block size smaller than the |
0aa69fd3 CH |
881 | * page size. |
882 | * | |
551879a4 ML |
883 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
884 | * | |
0aa69fd3 | 885 | * Return %true on success or %false on failure. |
1da177e4 | 886 | */ |
0aa69fd3 | 887 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
ff896738 | 888 | unsigned int len, unsigned int off, bool *same_page) |
1da177e4 | 889 | { |
c66a14d0 | 890 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
0aa69fd3 | 891 | return false; |
762380ad | 892 | |
cc90bc68 | 893 | if (bio->bi_vcnt > 0) { |
0aa69fd3 | 894 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
5919482e ML |
895 | |
896 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
35c820e7 | 897 | if (bio->bi_iter.bi_size > UINT_MAX - len) { |
2cd896a5 | 898 | *same_page = false; |
cc90bc68 | 899 | return false; |
2cd896a5 | 900 | } |
5919482e ML |
901 | bv->bv_len += len; |
902 | bio->bi_iter.bi_size += len; | |
903 | return true; | |
904 | } | |
c66a14d0 | 905 | } |
0aa69fd3 CH |
906 | return false; |
907 | } | |
908 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); | |
c66a14d0 | 909 | |
0aa69fd3 | 910 | /** |
551879a4 | 911 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 912 | * @bio: destination bio |
551879a4 ML |
913 | * @page: start page to add |
914 | * @len: length of the data to add, may cross pages | |
915 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
916 | * |
917 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
918 | * that @bio has space for another bvec. | |
919 | */ | |
920 | void __bio_add_page(struct bio *bio, struct page *page, | |
921 | unsigned int len, unsigned int off) | |
922 | { | |
923 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 924 | |
0aa69fd3 | 925 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 926 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
927 | |
928 | bv->bv_page = page; | |
929 | bv->bv_offset = off; | |
930 | bv->bv_len = len; | |
c66a14d0 | 931 | |
c66a14d0 | 932 | bio->bi_iter.bi_size += len; |
0aa69fd3 | 933 | bio->bi_vcnt++; |
b8e24a93 JW |
934 | |
935 | if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) | |
936 | bio_set_flag(bio, BIO_WORKINGSET); | |
0aa69fd3 CH |
937 | } |
938 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
939 | ||
940 | /** | |
551879a4 | 941 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 942 | * @bio: destination bio |
551879a4 ML |
943 | * @page: start page to add |
944 | * @len: vec entry length, may cross pages | |
945 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 946 | * |
551879a4 | 947 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
948 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
949 | */ | |
950 | int bio_add_page(struct bio *bio, struct page *page, | |
951 | unsigned int len, unsigned int offset) | |
952 | { | |
ff896738 CH |
953 | bool same_page = false; |
954 | ||
955 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 956 | if (bio_full(bio, len)) |
0aa69fd3 CH |
957 | return 0; |
958 | __bio_add_page(bio, page, len, offset); | |
959 | } | |
c66a14d0 | 960 | return len; |
1da177e4 | 961 | } |
a112a71d | 962 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 963 | |
d241a95f | 964 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
965 | { |
966 | struct bvec_iter_all iter_all; | |
967 | struct bio_vec *bvec; | |
7321ecbf | 968 | |
b2d0d991 CH |
969 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
970 | return; | |
971 | ||
d241a95f CH |
972 | bio_for_each_segment_all(bvec, bio, iter_all) { |
973 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
974 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 975 | put_page(bvec->bv_page); |
d241a95f | 976 | } |
7321ecbf | 977 | } |
29b2a3aa | 978 | EXPORT_SYMBOL_GPL(bio_release_pages); |
7321ecbf | 979 | |
7de55b7d | 980 | static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) |
6d0c48ae | 981 | { |
7a800a20 | 982 | WARN_ON_ONCE(bio->bi_max_vecs); |
c42bca92 PB |
983 | |
984 | bio->bi_vcnt = iter->nr_segs; | |
c42bca92 PB |
985 | bio->bi_io_vec = (struct bio_vec *)iter->bvec; |
986 | bio->bi_iter.bi_bvec_done = iter->iov_offset; | |
987 | bio->bi_iter.bi_size = iter->count; | |
ed97ce5e | 988 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
977be012 | 989 | bio_set_flag(bio, BIO_CLONED); |
7de55b7d | 990 | } |
c42bca92 | 991 | |
7de55b7d JT |
992 | static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) |
993 | { | |
994 | __bio_iov_bvec_set(bio, iter); | |
c42bca92 | 995 | iov_iter_advance(iter, iter->count); |
a10584c3 | 996 | return 0; |
6d0c48ae JA |
997 | } |
998 | ||
7de55b7d JT |
999 | static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter) |
1000 | { | |
1001 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; | |
1002 | struct iov_iter i = *iter; | |
1003 | ||
1004 | iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9); | |
1005 | __bio_iov_bvec_set(bio, &i); | |
1006 | iov_iter_advance(iter, i.count); | |
1007 | return 0; | |
1008 | } | |
1009 | ||
576ed913 CH |
1010 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
1011 | ||
2cefe4db | 1012 | /** |
17d51b10 | 1013 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
1014 | * @bio: bio to add pages to |
1015 | * @iter: iov iterator describing the region to be mapped | |
1016 | * | |
17d51b10 | 1017 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 1018 | * pages will have to be released using put_page() when done. |
17d51b10 | 1019 | * For multi-segment *iter, this function only adds pages from the |
3cf14889 | 1020 | * next non-empty segment of the iov iterator. |
2cefe4db | 1021 | */ |
17d51b10 | 1022 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 1023 | { |
576ed913 CH |
1024 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
1025 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
1026 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
1027 | struct page **pages = (struct page **)bv; | |
45691804 | 1028 | bool same_page = false; |
576ed913 CH |
1029 | ssize_t size, left; |
1030 | unsigned len, i; | |
b403ea24 | 1031 | size_t offset; |
576ed913 CH |
1032 | |
1033 | /* | |
1034 | * Move page array up in the allocated memory for the bio vecs as far as | |
1035 | * possible so that we can start filling biovecs from the beginning | |
1036 | * without overwriting the temporary page array. | |
1037 | */ | |
1038 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
1039 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db | 1040 | |
35c820e7 | 1041 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); |
2cefe4db KO |
1042 | if (unlikely(size <= 0)) |
1043 | return size ? size : -EFAULT; | |
2cefe4db | 1044 | |
576ed913 CH |
1045 | for (left = size, i = 0; left > 0; left -= len, i++) { |
1046 | struct page *page = pages[i]; | |
2cefe4db | 1047 | |
576ed913 | 1048 | len = min_t(size_t, PAGE_SIZE - offset, left); |
45691804 CH |
1049 | |
1050 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
1051 | if (same_page) | |
1052 | put_page(page); | |
1053 | } else { | |
79d08f89 | 1054 | if (WARN_ON_ONCE(bio_full(bio, len))) |
45691804 CH |
1055 | return -EINVAL; |
1056 | __bio_add_page(bio, page, len, offset); | |
1057 | } | |
576ed913 | 1058 | offset = 0; |
2cefe4db KO |
1059 | } |
1060 | ||
2cefe4db KO |
1061 | iov_iter_advance(iter, size); |
1062 | return 0; | |
1063 | } | |
17d51b10 | 1064 | |
0512a75b KB |
1065 | static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) |
1066 | { | |
1067 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; | |
1068 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
309dca30 | 1069 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
0512a75b KB |
1070 | unsigned int max_append_sectors = queue_max_zone_append_sectors(q); |
1071 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; | |
1072 | struct page **pages = (struct page **)bv; | |
1073 | ssize_t size, left; | |
1074 | unsigned len, i; | |
1075 | size_t offset; | |
4977d121 | 1076 | int ret = 0; |
0512a75b KB |
1077 | |
1078 | if (WARN_ON_ONCE(!max_append_sectors)) | |
1079 | return 0; | |
1080 | ||
1081 | /* | |
1082 | * Move page array up in the allocated memory for the bio vecs as far as | |
1083 | * possible so that we can start filling biovecs from the beginning | |
1084 | * without overwriting the temporary page array. | |
1085 | */ | |
1086 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
1087 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
1088 | ||
1089 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
1090 | if (unlikely(size <= 0)) | |
1091 | return size ? size : -EFAULT; | |
1092 | ||
1093 | for (left = size, i = 0; left > 0; left -= len, i++) { | |
1094 | struct page *page = pages[i]; | |
1095 | bool same_page = false; | |
1096 | ||
1097 | len = min_t(size_t, PAGE_SIZE - offset, left); | |
1098 | if (bio_add_hw_page(q, bio, page, len, offset, | |
4977d121 NA |
1099 | max_append_sectors, &same_page) != len) { |
1100 | ret = -EINVAL; | |
1101 | break; | |
1102 | } | |
0512a75b KB |
1103 | if (same_page) |
1104 | put_page(page); | |
1105 | offset = 0; | |
1106 | } | |
1107 | ||
4977d121 NA |
1108 | iov_iter_advance(iter, size - left); |
1109 | return ret; | |
0512a75b KB |
1110 | } |
1111 | ||
17d51b10 | 1112 | /** |
6d0c48ae | 1113 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 1114 | * @bio: bio to add pages to |
6d0c48ae JA |
1115 | * @iter: iov iterator describing the region to be added |
1116 | * | |
1117 | * This takes either an iterator pointing to user memory, or one pointing to | |
1118 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
1119 | * map them into the kernel. On IO completion, the caller should put those | |
c42bca92 PB |
1120 | * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided |
1121 | * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs | |
1122 | * to ensure the bvecs and pages stay referenced until the submitted I/O is | |
1123 | * completed by a call to ->ki_complete() or returns with an error other than | |
1124 | * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF | |
1125 | * on IO completion. If it isn't, then pages should be released. | |
17d51b10 | 1126 | * |
17d51b10 | 1127 | * The function tries, but does not guarantee, to pin as many pages as |
5cd3ddc1 | 1128 | * fit into the bio, or are requested in @iter, whatever is smaller. If |
6d0c48ae JA |
1129 | * MM encounters an error pinning the requested pages, it stops. Error |
1130 | * is returned only if 0 pages could be pinned. | |
0cf41e5e PB |
1131 | * |
1132 | * It's intended for direct IO, so doesn't do PSI tracking, the caller is | |
1133 | * responsible for setting BIO_WORKINGSET if necessary. | |
17d51b10 MW |
1134 | */ |
1135 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
1136 | { | |
c42bca92 | 1137 | int ret = 0; |
14eacf12 | 1138 | |
c42bca92 | 1139 | if (iov_iter_is_bvec(iter)) { |
7de55b7d JT |
1140 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) |
1141 | return bio_iov_bvec_set_append(bio, iter); | |
ed97ce5e | 1142 | return bio_iov_bvec_set(bio, iter); |
c42bca92 | 1143 | } |
17d51b10 MW |
1144 | |
1145 | do { | |
86004515 | 1146 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) |
0512a75b | 1147 | ret = __bio_iov_append_get_pages(bio, iter); |
86004515 CH |
1148 | else |
1149 | ret = __bio_iov_iter_get_pages(bio, iter); | |
79d08f89 | 1150 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
17d51b10 | 1151 | |
0cf41e5e PB |
1152 | /* don't account direct I/O as memory stall */ |
1153 | bio_clear_flag(bio, BIO_WORKINGSET); | |
14eacf12 | 1154 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 1155 | } |
29b2a3aa | 1156 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); |
2cefe4db | 1157 | |
4246a0b6 | 1158 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 1159 | { |
65e53aab | 1160 | complete(bio->bi_private); |
9e882242 KO |
1161 | } |
1162 | ||
1163 | /** | |
1164 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
1165 | * @bio: The &struct bio which describes the I/O |
1166 | * | |
1167 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
1168 | * bio_endio() on failure. | |
3d289d68 JK |
1169 | * |
1170 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
1171 | * result in bio reference to be consumed. The caller must drop the reference | |
1172 | * on his own. | |
9e882242 | 1173 | */ |
4e49ea4a | 1174 | int submit_bio_wait(struct bio *bio) |
9e882242 | 1175 | { |
309dca30 CH |
1176 | DECLARE_COMPLETION_ONSTACK_MAP(done, |
1177 | bio->bi_bdev->bd_disk->lockdep_map); | |
de6a78b6 | 1178 | unsigned long hang_check; |
9e882242 | 1179 | |
65e53aab | 1180 | bio->bi_private = &done; |
9e882242 | 1181 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 1182 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 1183 | submit_bio(bio); |
de6a78b6 ML |
1184 | |
1185 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1186 | hang_check = sysctl_hung_task_timeout_secs; | |
1187 | if (hang_check) | |
1188 | while (!wait_for_completion_io_timeout(&done, | |
1189 | hang_check * (HZ/2))) | |
1190 | ; | |
1191 | else | |
1192 | wait_for_completion_io(&done); | |
9e882242 | 1193 | |
65e53aab | 1194 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
1195 | } |
1196 | EXPORT_SYMBOL(submit_bio_wait); | |
1197 | ||
054bdf64 KO |
1198 | /** |
1199 | * bio_advance - increment/complete a bio by some number of bytes | |
1200 | * @bio: bio to advance | |
1201 | * @bytes: number of bytes to complete | |
1202 | * | |
1203 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
1204 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
1205 | * be updated on the last bvec as well. | |
1206 | * | |
1207 | * @bio will then represent the remaining, uncompleted portion of the io. | |
1208 | */ | |
1209 | void bio_advance(struct bio *bio, unsigned bytes) | |
1210 | { | |
1211 | if (bio_integrity(bio)) | |
1212 | bio_integrity_advance(bio, bytes); | |
1213 | ||
a892c8d5 | 1214 | bio_crypt_advance(bio, bytes); |
4550dd6c | 1215 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
1216 | } |
1217 | EXPORT_SYMBOL(bio_advance); | |
1218 | ||
45db54d5 KO |
1219 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1220 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1221 | { |
1cb9dda4 | 1222 | struct bio_vec src_bv, dst_bv; |
16ac3d63 | 1223 | void *src_p, *dst_p; |
1cb9dda4 | 1224 | unsigned bytes; |
16ac3d63 | 1225 | |
45db54d5 KO |
1226 | while (src_iter->bi_size && dst_iter->bi_size) { |
1227 | src_bv = bio_iter_iovec(src, *src_iter); | |
1228 | dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1cb9dda4 KO |
1229 | |
1230 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1231 | |
1cb9dda4 KO |
1232 | src_p = kmap_atomic(src_bv.bv_page); |
1233 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1234 | |
1cb9dda4 KO |
1235 | memcpy(dst_p + dst_bv.bv_offset, |
1236 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1237 | bytes); |
1238 | ||
1239 | kunmap_atomic(dst_p); | |
1240 | kunmap_atomic(src_p); | |
1241 | ||
6e6e811d KO |
1242 | flush_dcache_page(dst_bv.bv_page); |
1243 | ||
22b56c29 PB |
1244 | bio_advance_iter_single(src, src_iter, bytes); |
1245 | bio_advance_iter_single(dst, dst_iter, bytes); | |
16ac3d63 KO |
1246 | } |
1247 | } | |
38a72dac KO |
1248 | EXPORT_SYMBOL(bio_copy_data_iter); |
1249 | ||
1250 | /** | |
45db54d5 KO |
1251 | * bio_copy_data - copy contents of data buffers from one bio to another |
1252 | * @src: source bio | |
1253 | * @dst: destination bio | |
38a72dac KO |
1254 | * |
1255 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1256 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1257 | */ | |
1258 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1259 | { | |
45db54d5 KO |
1260 | struct bvec_iter src_iter = src->bi_iter; |
1261 | struct bvec_iter dst_iter = dst->bi_iter; | |
1262 | ||
1263 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1264 | } |
16ac3d63 KO |
1265 | EXPORT_SYMBOL(bio_copy_data); |
1266 | ||
491221f8 | 1267 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1268 | { |
1269 | struct bio_vec *bvec; | |
6dc4f100 | 1270 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1271 | |
2b070cfe | 1272 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1273 | __free_page(bvec->bv_page); |
1274 | } | |
491221f8 | 1275 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1276 | |
1da177e4 LT |
1277 | /* |
1278 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1279 | * for performing direct-IO in BIOs. | |
1280 | * | |
1281 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1282 | * because the required locks are not interrupt-safe. So what we can do is to | |
1283 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1284 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1285 | * in process context. | |
1286 | * | |
1287 | * We special-case compound pages here: normally this means reads into hugetlb | |
1288 | * pages. The logic in here doesn't really work right for compound pages | |
1289 | * because the VM does not uniformly chase down the head page in all cases. | |
1290 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1291 | * handle them at all. So we skip compound pages here at an early stage. | |
1292 | * | |
1293 | * Note that this code is very hard to test under normal circumstances because | |
1294 | * direct-io pins the pages with get_user_pages(). This makes | |
1295 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1296 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1297 | * pagecache. |
1298 | * | |
1299 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1300 | * deferred bio dirtying paths. | |
1301 | */ | |
1302 | ||
1303 | /* | |
1304 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1305 | */ | |
1306 | void bio_set_pages_dirty(struct bio *bio) | |
1307 | { | |
cb34e057 | 1308 | struct bio_vec *bvec; |
6dc4f100 | 1309 | struct bvec_iter_all iter_all; |
1da177e4 | 1310 | |
2b070cfe | 1311 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1312 | if (!PageCompound(bvec->bv_page)) |
1313 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1314 | } |
1315 | } | |
1316 | ||
1da177e4 LT |
1317 | /* |
1318 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1319 | * If they are, then fine. If, however, some pages are clean then they must | |
1320 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1321 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1322 | * |
1323 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1324 | * here on. It will run one put_page() against each page and will run one |
1325 | * bio_put() against the BIO. | |
1da177e4 LT |
1326 | */ |
1327 | ||
65f27f38 | 1328 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1329 | |
65f27f38 | 1330 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1331 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1332 | static struct bio *bio_dirty_list; | |
1333 | ||
1334 | /* | |
1335 | * This runs in process context | |
1336 | */ | |
65f27f38 | 1337 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1338 | { |
24d5493f | 1339 | struct bio *bio, *next; |
1da177e4 | 1340 | |
24d5493f CH |
1341 | spin_lock_irq(&bio_dirty_lock); |
1342 | next = bio_dirty_list; | |
1da177e4 | 1343 | bio_dirty_list = NULL; |
24d5493f | 1344 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1345 | |
24d5493f CH |
1346 | while ((bio = next) != NULL) { |
1347 | next = bio->bi_private; | |
1da177e4 | 1348 | |
d241a95f | 1349 | bio_release_pages(bio, true); |
1da177e4 | 1350 | bio_put(bio); |
1da177e4 LT |
1351 | } |
1352 | } | |
1353 | ||
1354 | void bio_check_pages_dirty(struct bio *bio) | |
1355 | { | |
cb34e057 | 1356 | struct bio_vec *bvec; |
24d5493f | 1357 | unsigned long flags; |
6dc4f100 | 1358 | struct bvec_iter_all iter_all; |
1da177e4 | 1359 | |
2b070cfe | 1360 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1361 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1362 | goto defer; | |
1da177e4 LT |
1363 | } |
1364 | ||
d241a95f | 1365 | bio_release_pages(bio, false); |
24d5493f CH |
1366 | bio_put(bio); |
1367 | return; | |
1368 | defer: | |
1369 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1370 | bio->bi_private = bio_dirty_list; | |
1371 | bio_dirty_list = bio; | |
1372 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1373 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1374 | } |
1375 | ||
c4cf5261 JA |
1376 | static inline bool bio_remaining_done(struct bio *bio) |
1377 | { | |
1378 | /* | |
1379 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1380 | * we always end io on the first invocation. | |
1381 | */ | |
1382 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1383 | return true; | |
1384 | ||
1385 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1386 | ||
326e1dbb | 1387 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1388 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1389 | return true; |
326e1dbb | 1390 | } |
c4cf5261 JA |
1391 | |
1392 | return false; | |
1393 | } | |
1394 | ||
1da177e4 LT |
1395 | /** |
1396 | * bio_endio - end I/O on a bio | |
1397 | * @bio: bio | |
1da177e4 LT |
1398 | * |
1399 | * Description: | |
4246a0b6 CH |
1400 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1401 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1402 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1403 | * |
1404 | * bio_endio() can be called several times on a bio that has been chained | |
1405 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
60b6a7e6 | 1406 | * last time. |
1da177e4 | 1407 | **/ |
4246a0b6 | 1408 | void bio_endio(struct bio *bio) |
1da177e4 | 1409 | { |
ba8c6967 | 1410 | again: |
2b885517 | 1411 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1412 | return; |
7c20f116 CH |
1413 | if (!bio_integrity_endio(bio)) |
1414 | return; | |
1da177e4 | 1415 | |
309dca30 CH |
1416 | if (bio->bi_bdev) |
1417 | rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); | |
67b42d0b | 1418 | |
60b6a7e6 EH |
1419 | if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
1420 | trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); | |
1421 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
1422 | } | |
1423 | ||
ba8c6967 CH |
1424 | /* |
1425 | * Need to have a real endio function for chained bios, otherwise | |
1426 | * various corner cases will break (like stacking block devices that | |
1427 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1428 | * recursion and blowing the stack. Tail call optimization would | |
1429 | * handle this, but compiling with frame pointers also disables | |
1430 | * gcc's sibling call optimization. | |
1431 | */ | |
1432 | if (bio->bi_end_io == bio_chain_endio) { | |
1433 | bio = __bio_chain_endio(bio); | |
1434 | goto again; | |
196d38bc | 1435 | } |
ba8c6967 | 1436 | |
9e234eea | 1437 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1438 | /* release cgroup info */ |
1439 | bio_uninit(bio); | |
ba8c6967 CH |
1440 | if (bio->bi_end_io) |
1441 | bio->bi_end_io(bio); | |
1da177e4 | 1442 | } |
a112a71d | 1443 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1444 | |
20d0189b KO |
1445 | /** |
1446 | * bio_split - split a bio | |
1447 | * @bio: bio to split | |
1448 | * @sectors: number of sectors to split from the front of @bio | |
1449 | * @gfp: gfp mask | |
1450 | * @bs: bio set to allocate from | |
1451 | * | |
1452 | * Allocates and returns a new bio which represents @sectors from the start of | |
1453 | * @bio, and updates @bio to represent the remaining sectors. | |
1454 | * | |
f3f5da62 | 1455 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1456 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1457 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1458 | */ |
1459 | struct bio *bio_split(struct bio *bio, int sectors, | |
1460 | gfp_t gfp, struct bio_set *bs) | |
1461 | { | |
f341a4d3 | 1462 | struct bio *split; |
20d0189b KO |
1463 | |
1464 | BUG_ON(sectors <= 0); | |
1465 | BUG_ON(sectors >= bio_sectors(bio)); | |
1466 | ||
0512a75b KB |
1467 | /* Zone append commands cannot be split */ |
1468 | if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) | |
1469 | return NULL; | |
1470 | ||
f9d03f96 | 1471 | split = bio_clone_fast(bio, gfp, bs); |
20d0189b KO |
1472 | if (!split) |
1473 | return NULL; | |
1474 | ||
1475 | split->bi_iter.bi_size = sectors << 9; | |
1476 | ||
1477 | if (bio_integrity(split)) | |
fbd08e76 | 1478 | bio_integrity_trim(split); |
20d0189b KO |
1479 | |
1480 | bio_advance(bio, split->bi_iter.bi_size); | |
1481 | ||
fbbaf700 | 1482 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1483 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1484 | |
20d0189b KO |
1485 | return split; |
1486 | } | |
1487 | EXPORT_SYMBOL(bio_split); | |
1488 | ||
6678d83f KO |
1489 | /** |
1490 | * bio_trim - trim a bio | |
1491 | * @bio: bio to trim | |
1492 | * @offset: number of sectors to trim from the front of @bio | |
1493 | * @size: size we want to trim @bio to, in sectors | |
1494 | */ | |
1495 | void bio_trim(struct bio *bio, int offset, int size) | |
1496 | { | |
1497 | /* 'bio' is a cloned bio which we need to trim to match | |
1498 | * the given offset and size. | |
6678d83f | 1499 | */ |
6678d83f KO |
1500 | |
1501 | size <<= 9; | |
4f024f37 | 1502 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1503 | return; |
1504 | ||
6678d83f | 1505 | bio_advance(bio, offset << 9); |
4f024f37 | 1506 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1507 | |
1508 | if (bio_integrity(bio)) | |
fbd08e76 | 1509 | bio_integrity_trim(bio); |
376a78ab | 1510 | |
6678d83f KO |
1511 | } |
1512 | EXPORT_SYMBOL_GPL(bio_trim); | |
1513 | ||
1da177e4 LT |
1514 | /* |
1515 | * create memory pools for biovec's in a bio_set. | |
1516 | * use the global biovec slabs created for general use. | |
1517 | */ | |
8aa6ba2f | 1518 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1519 | { |
7a800a20 | 1520 | struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; |
1da177e4 | 1521 | |
8aa6ba2f | 1522 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1523 | } |
1524 | ||
917a38c7 KO |
1525 | /* |
1526 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1527 | * | |
1528 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1529 | * kzalloc()). | |
1530 | */ | |
1531 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1532 | { |
df2cb6da KO |
1533 | if (bs->rescue_workqueue) |
1534 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1535 | bs->rescue_workqueue = NULL; |
df2cb6da | 1536 | |
8aa6ba2f KO |
1537 | mempool_exit(&bs->bio_pool); |
1538 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1539 | |
7878cba9 | 1540 | bioset_integrity_free(bs); |
917a38c7 KO |
1541 | if (bs->bio_slab) |
1542 | bio_put_slab(bs); | |
1543 | bs->bio_slab = NULL; | |
1544 | } | |
1545 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1546 | |
917a38c7 KO |
1547 | /** |
1548 | * bioset_init - Initialize a bio_set | |
dad08527 | 1549 | * @bs: pool to initialize |
917a38c7 KO |
1550 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1551 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1552 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1553 | * and %BIOSET_NEED_RESCUER | |
1554 | * | |
dad08527 KO |
1555 | * Description: |
1556 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1557 | * to ask for a number of bytes to be allocated in front of the bio. | |
1558 | * Front pad allocation is useful for embedding the bio inside | |
1559 | * another structure, to avoid allocating extra data to go with the bio. | |
1560 | * Note that the bio must be embedded at the END of that structure always, | |
1561 | * or things will break badly. | |
1562 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
1563 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
1564 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
1565 | * dispatch queued requests when the mempool runs out of space. | |
1566 | * | |
917a38c7 KO |
1567 | */ |
1568 | int bioset_init(struct bio_set *bs, | |
1569 | unsigned int pool_size, | |
1570 | unsigned int front_pad, | |
1571 | int flags) | |
1572 | { | |
917a38c7 | 1573 | bs->front_pad = front_pad; |
9f180e31 ML |
1574 | if (flags & BIOSET_NEED_BVECS) |
1575 | bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1576 | else | |
1577 | bs->back_pad = 0; | |
917a38c7 KO |
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 | ||
49d1ec85 | 1583 | bs->bio_slab = bio_find_or_create_slab(bs); |
917a38c7 KO |
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 | ||
de76fd89 | 1626 | static int __init init_bio(void) |
1da177e4 LT |
1627 | { |
1628 | int i; | |
1629 | ||
7878cba9 | 1630 | bio_integrity_init(); |
1da177e4 | 1631 | |
de76fd89 CH |
1632 | for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { |
1633 | struct biovec_slab *bvs = bvec_slabs + i; | |
a7fcd37c | 1634 | |
de76fd89 CH |
1635 | bvs->slab = kmem_cache_create(bvs->name, |
1636 | bvs->nr_vecs * sizeof(struct bio_vec), 0, | |
1637 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
1da177e4 | 1638 | } |
1da177e4 | 1639 | |
f4f8154a | 1640 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
1da177e4 LT |
1641 | panic("bio: can't allocate bios\n"); |
1642 | ||
f4f8154a | 1643 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
1644 | panic("bio: can't create integrity pool\n"); |
1645 | ||
1da177e4 LT |
1646 | return 0; |
1647 | } | |
1da177e4 | 1648 | subsys_initcall(init_bio); |