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> |
b4c5875d | 18 | #include <linux/highmem.h> |
de6a78b6 | 19 | #include <linux/sched/sysctl.h> |
a892c8d5 | 20 | #include <linux/blk-crypto.h> |
49d1ec85 | 21 | #include <linux/xarray.h> |
1da177e4 | 22 | |
55782138 | 23 | #include <trace/events/block.h> |
9e234eea | 24 | #include "blk.h" |
67b42d0b | 25 | #include "blk-rq-qos.h" |
672fdcf0 | 26 | #include "blk-cgroup.h" |
0bfc2455 | 27 | |
be4d234d | 28 | struct bio_alloc_cache { |
fcade2ce | 29 | struct bio *free_list; |
be4d234d JA |
30 | unsigned int nr; |
31 | }; | |
32 | ||
de76fd89 | 33 | static struct biovec_slab { |
6ac0b715 CH |
34 | int nr_vecs; |
35 | char *name; | |
36 | struct kmem_cache *slab; | |
de76fd89 CH |
37 | } bvec_slabs[] __read_mostly = { |
38 | { .nr_vecs = 16, .name = "biovec-16" }, | |
39 | { .nr_vecs = 64, .name = "biovec-64" }, | |
40 | { .nr_vecs = 128, .name = "biovec-128" }, | |
a8affc03 | 41 | { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" }, |
1da177e4 | 42 | }; |
6ac0b715 | 43 | |
7a800a20 CH |
44 | static struct biovec_slab *biovec_slab(unsigned short nr_vecs) |
45 | { | |
46 | switch (nr_vecs) { | |
47 | /* smaller bios use inline vecs */ | |
48 | case 5 ... 16: | |
49 | return &bvec_slabs[0]; | |
50 | case 17 ... 64: | |
51 | return &bvec_slabs[1]; | |
52 | case 65 ... 128: | |
53 | return &bvec_slabs[2]; | |
a8affc03 | 54 | case 129 ... BIO_MAX_VECS: |
7a800a20 CH |
55 | return &bvec_slabs[3]; |
56 | default: | |
57 | BUG(); | |
58 | return NULL; | |
59 | } | |
60 | } | |
1da177e4 | 61 | |
1da177e4 LT |
62 | /* |
63 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
64 | * IO code that does not need private memory pools. | |
65 | */ | |
f4f8154a | 66 | struct bio_set fs_bio_set; |
3f86a82a | 67 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 68 | |
bb799ca0 JA |
69 | /* |
70 | * Our slab pool management | |
71 | */ | |
72 | struct bio_slab { | |
73 | struct kmem_cache *slab; | |
74 | unsigned int slab_ref; | |
75 | unsigned int slab_size; | |
76 | char name[8]; | |
77 | }; | |
78 | static DEFINE_MUTEX(bio_slab_lock); | |
49d1ec85 | 79 | static DEFINE_XARRAY(bio_slabs); |
bb799ca0 | 80 | |
49d1ec85 | 81 | static struct bio_slab *create_bio_slab(unsigned int size) |
bb799ca0 | 82 | { |
49d1ec85 | 83 | struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); |
bb799ca0 | 84 | |
49d1ec85 ML |
85 | if (!bslab) |
86 | return NULL; | |
bb799ca0 | 87 | |
49d1ec85 ML |
88 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); |
89 | bslab->slab = kmem_cache_create(bslab->name, size, | |
1a7e76e4 CH |
90 | ARCH_KMALLOC_MINALIGN, |
91 | SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL); | |
49d1ec85 ML |
92 | if (!bslab->slab) |
93 | goto fail_alloc_slab; | |
bb799ca0 | 94 | |
49d1ec85 ML |
95 | bslab->slab_ref = 1; |
96 | bslab->slab_size = size; | |
bb799ca0 | 97 | |
49d1ec85 ML |
98 | if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) |
99 | return bslab; | |
bb799ca0 | 100 | |
49d1ec85 | 101 | kmem_cache_destroy(bslab->slab); |
bb799ca0 | 102 | |
49d1ec85 ML |
103 | fail_alloc_slab: |
104 | kfree(bslab); | |
105 | return NULL; | |
106 | } | |
bb799ca0 | 107 | |
49d1ec85 ML |
108 | static inline unsigned int bs_bio_slab_size(struct bio_set *bs) |
109 | { | |
9f180e31 | 110 | return bs->front_pad + sizeof(struct bio) + bs->back_pad; |
49d1ec85 | 111 | } |
bb799ca0 | 112 | |
49d1ec85 ML |
113 | static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) |
114 | { | |
115 | unsigned int size = bs_bio_slab_size(bs); | |
116 | struct bio_slab *bslab; | |
bb799ca0 | 117 | |
49d1ec85 ML |
118 | mutex_lock(&bio_slab_lock); |
119 | bslab = xa_load(&bio_slabs, size); | |
120 | if (bslab) | |
121 | bslab->slab_ref++; | |
122 | else | |
123 | bslab = create_bio_slab(size); | |
bb799ca0 | 124 | mutex_unlock(&bio_slab_lock); |
49d1ec85 ML |
125 | |
126 | if (bslab) | |
127 | return bslab->slab; | |
128 | return NULL; | |
bb799ca0 JA |
129 | } |
130 | ||
131 | static void bio_put_slab(struct bio_set *bs) | |
132 | { | |
133 | struct bio_slab *bslab = NULL; | |
49d1ec85 | 134 | unsigned int slab_size = bs_bio_slab_size(bs); |
bb799ca0 JA |
135 | |
136 | mutex_lock(&bio_slab_lock); | |
137 | ||
49d1ec85 | 138 | bslab = xa_load(&bio_slabs, slab_size); |
bb799ca0 JA |
139 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
140 | goto out; | |
141 | ||
49d1ec85 ML |
142 | WARN_ON_ONCE(bslab->slab != bs->bio_slab); |
143 | ||
bb799ca0 JA |
144 | WARN_ON(!bslab->slab_ref); |
145 | ||
146 | if (--bslab->slab_ref) | |
147 | goto out; | |
148 | ||
49d1ec85 ML |
149 | xa_erase(&bio_slabs, slab_size); |
150 | ||
bb799ca0 | 151 | kmem_cache_destroy(bslab->slab); |
49d1ec85 | 152 | kfree(bslab); |
bb799ca0 JA |
153 | |
154 | out: | |
155 | mutex_unlock(&bio_slab_lock); | |
156 | } | |
157 | ||
7a800a20 | 158 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs) |
7ba1ba12 | 159 | { |
9e8c0d0d | 160 | BUG_ON(nr_vecs > BIO_MAX_VECS); |
ed996a52 | 161 | |
a8affc03 | 162 | if (nr_vecs == BIO_MAX_VECS) |
9f060e22 | 163 | mempool_free(bv, pool); |
7a800a20 CH |
164 | else if (nr_vecs > BIO_INLINE_VECS) |
165 | kmem_cache_free(biovec_slab(nr_vecs)->slab, bv); | |
bb799ca0 | 166 | } |
bb799ca0 | 167 | |
f2c3eb9b CH |
168 | /* |
169 | * Make the first allocation restricted and don't dump info on allocation | |
170 | * failures, since we'll fall back to the mempool in case of failure. | |
171 | */ | |
172 | static inline gfp_t bvec_alloc_gfp(gfp_t gfp) | |
173 | { | |
174 | return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | | |
175 | __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
bb799ca0 JA |
176 | } |
177 | ||
7a800a20 CH |
178 | struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, |
179 | gfp_t gfp_mask) | |
1da177e4 | 180 | { |
7a800a20 | 181 | struct biovec_slab *bvs = biovec_slab(*nr_vecs); |
1da177e4 | 182 | |
7a800a20 | 183 | if (WARN_ON_ONCE(!bvs)) |
7ff9345f | 184 | return NULL; |
7ff9345f JA |
185 | |
186 | /* | |
7a800a20 CH |
187 | * Upgrade the nr_vecs request to take full advantage of the allocation. |
188 | * We also rely on this in the bvec_free path. | |
7ff9345f | 189 | */ |
7a800a20 | 190 | *nr_vecs = bvs->nr_vecs; |
7ff9345f | 191 | |
7ff9345f | 192 | /* |
f007a3d6 CH |
193 | * Try a slab allocation first for all smaller allocations. If that |
194 | * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. | |
a8affc03 | 195 | * The mempool is sized to handle up to BIO_MAX_VECS entries. |
7ff9345f | 196 | */ |
a8affc03 | 197 | if (*nr_vecs < BIO_MAX_VECS) { |
f007a3d6 | 198 | struct bio_vec *bvl; |
1da177e4 | 199 | |
f2c3eb9b | 200 | bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); |
7a800a20 | 201 | if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) |
f007a3d6 | 202 | return bvl; |
a8affc03 | 203 | *nr_vecs = BIO_MAX_VECS; |
7ff9345f JA |
204 | } |
205 | ||
f007a3d6 | 206 | return mempool_alloc(pool, gfp_mask); |
1da177e4 LT |
207 | } |
208 | ||
9ae3b3f5 | 209 | void bio_uninit(struct bio *bio) |
1da177e4 | 210 | { |
db9819c7 CH |
211 | #ifdef CONFIG_BLK_CGROUP |
212 | if (bio->bi_blkg) { | |
213 | blkg_put(bio->bi_blkg); | |
214 | bio->bi_blkg = NULL; | |
215 | } | |
216 | #endif | |
ece841ab JT |
217 | if (bio_integrity(bio)) |
218 | bio_integrity_free(bio); | |
a892c8d5 ST |
219 | |
220 | bio_crypt_free_ctx(bio); | |
4254bba1 | 221 | } |
9ae3b3f5 | 222 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 223 | |
4254bba1 KO |
224 | static void bio_free(struct bio *bio) |
225 | { | |
226 | struct bio_set *bs = bio->bi_pool; | |
066ff571 | 227 | void *p = bio; |
4254bba1 | 228 | |
066ff571 | 229 | WARN_ON_ONCE(!bs); |
4254bba1 | 230 | |
066ff571 CH |
231 | bio_uninit(bio); |
232 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); | |
233 | mempool_free(p - bs->front_pad, &bs->bio_pool); | |
3676347a PO |
234 | } |
235 | ||
9ae3b3f5 JA |
236 | /* |
237 | * Users of this function have their own bio allocation. Subsequently, | |
238 | * they must remember to pair any call to bio_init() with bio_uninit() | |
239 | * when IO has completed, or when the bio is released. | |
240 | */ | |
49add496 | 241 | void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table, |
16458cf3 | 242 | unsigned short max_vecs, blk_opf_t opf) |
1da177e4 | 243 | { |
da521626 | 244 | bio->bi_next = NULL; |
49add496 CH |
245 | bio->bi_bdev = bdev; |
246 | bio->bi_opf = opf; | |
da521626 JA |
247 | bio->bi_flags = 0; |
248 | bio->bi_ioprio = 0; | |
da521626 JA |
249 | bio->bi_status = 0; |
250 | bio->bi_iter.bi_sector = 0; | |
251 | bio->bi_iter.bi_size = 0; | |
252 | bio->bi_iter.bi_idx = 0; | |
253 | bio->bi_iter.bi_bvec_done = 0; | |
254 | bio->bi_end_io = NULL; | |
255 | bio->bi_private = NULL; | |
256 | #ifdef CONFIG_BLK_CGROUP | |
257 | bio->bi_blkg = NULL; | |
258 | bio->bi_issue.value = 0; | |
49add496 CH |
259 | if (bdev) |
260 | bio_associate_blkg(bio); | |
da521626 JA |
261 | #ifdef CONFIG_BLK_CGROUP_IOCOST |
262 | bio->bi_iocost_cost = 0; | |
263 | #endif | |
264 | #endif | |
265 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION | |
266 | bio->bi_crypt_context = NULL; | |
267 | #endif | |
268 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
269 | bio->bi_integrity = NULL; | |
270 | #endif | |
271 | bio->bi_vcnt = 0; | |
272 | ||
c4cf5261 | 273 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 274 | atomic_set(&bio->__bi_cnt, 1); |
3e08773c | 275 | bio->bi_cookie = BLK_QC_T_NONE; |
3a83f467 | 276 | |
3a83f467 | 277 | bio->bi_max_vecs = max_vecs; |
da521626 JA |
278 | bio->bi_io_vec = table; |
279 | bio->bi_pool = NULL; | |
1da177e4 | 280 | } |
a112a71d | 281 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 282 | |
f44b48c7 KO |
283 | /** |
284 | * bio_reset - reinitialize a bio | |
285 | * @bio: bio to reset | |
a7c50c94 CH |
286 | * @bdev: block device to use the bio for |
287 | * @opf: operation and flags for bio | |
f44b48c7 KO |
288 | * |
289 | * Description: | |
290 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
291 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
292 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
293 | * comment in struct bio. | |
294 | */ | |
16458cf3 | 295 | void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf) |
f44b48c7 | 296 | { |
9ae3b3f5 | 297 | bio_uninit(bio); |
f44b48c7 | 298 | memset(bio, 0, BIO_RESET_BYTES); |
c4cf5261 | 299 | atomic_set(&bio->__bi_remaining, 1); |
a7c50c94 | 300 | bio->bi_bdev = bdev; |
78e34374 CH |
301 | if (bio->bi_bdev) |
302 | bio_associate_blkg(bio); | |
a7c50c94 | 303 | bio->bi_opf = opf; |
f44b48c7 KO |
304 | } |
305 | EXPORT_SYMBOL(bio_reset); | |
306 | ||
38f8baae | 307 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 308 | { |
4246a0b6 CH |
309 | struct bio *parent = bio->bi_private; |
310 | ||
3edf5346 | 311 | if (bio->bi_status && !parent->bi_status) |
4e4cbee9 | 312 | parent->bi_status = bio->bi_status; |
196d38bc | 313 | bio_put(bio); |
38f8baae CH |
314 | return parent; |
315 | } | |
316 | ||
317 | static void bio_chain_endio(struct bio *bio) | |
318 | { | |
319 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
320 | } |
321 | ||
322 | /** | |
323 | * bio_chain - chain bio completions | |
1051a902 | 324 | * @bio: the target bio |
5b874af6 | 325 | * @parent: the parent bio of @bio |
196d38bc KO |
326 | * |
327 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
328 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
329 | * completed; the chained bio will also be freed when it completes. | |
330 | * | |
331 | * The caller must not set bi_private or bi_end_io in @bio. | |
332 | */ | |
333 | void bio_chain(struct bio *bio, struct bio *parent) | |
334 | { | |
335 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
336 | ||
337 | bio->bi_private = parent; | |
338 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 339 | bio_inc_remaining(parent); |
196d38bc KO |
340 | } |
341 | EXPORT_SYMBOL(bio_chain); | |
342 | ||
0a3140ea | 343 | struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev, |
16458cf3 | 344 | unsigned int nr_pages, blk_opf_t opf, gfp_t gfp) |
3b005bf6 | 345 | { |
07888c66 | 346 | struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp); |
0a3140ea | 347 | |
3b005bf6 CH |
348 | if (bio) { |
349 | bio_chain(bio, new); | |
350 | submit_bio(bio); | |
351 | } | |
352 | ||
353 | return new; | |
354 | } | |
355 | EXPORT_SYMBOL_GPL(blk_next_bio); | |
356 | ||
df2cb6da KO |
357 | static void bio_alloc_rescue(struct work_struct *work) |
358 | { | |
359 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
360 | struct bio *bio; | |
361 | ||
362 | while (1) { | |
363 | spin_lock(&bs->rescue_lock); | |
364 | bio = bio_list_pop(&bs->rescue_list); | |
365 | spin_unlock(&bs->rescue_lock); | |
366 | ||
367 | if (!bio) | |
368 | break; | |
369 | ||
ed00aabd | 370 | submit_bio_noacct(bio); |
df2cb6da KO |
371 | } |
372 | } | |
373 | ||
374 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
375 | { | |
376 | struct bio_list punt, nopunt; | |
377 | struct bio *bio; | |
378 | ||
47e0fb46 N |
379 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
380 | return; | |
df2cb6da KO |
381 | /* |
382 | * In order to guarantee forward progress we must punt only bios that | |
383 | * were allocated from this bio_set; otherwise, if there was a bio on | |
384 | * there for a stacking driver higher up in the stack, processing it | |
385 | * could require allocating bios from this bio_set, and doing that from | |
386 | * our own rescuer would be bad. | |
387 | * | |
388 | * Since bio lists are singly linked, pop them all instead of trying to | |
389 | * remove from the middle of the list: | |
390 | */ | |
391 | ||
392 | bio_list_init(&punt); | |
393 | bio_list_init(&nopunt); | |
394 | ||
f5fe1b51 | 395 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 396 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 397 | current->bio_list[0] = nopunt; |
df2cb6da | 398 | |
f5fe1b51 N |
399 | bio_list_init(&nopunt); |
400 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
401 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
402 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
403 | |
404 | spin_lock(&bs->rescue_lock); | |
405 | bio_list_merge(&bs->rescue_list, &punt); | |
406 | spin_unlock(&bs->rescue_lock); | |
407 | ||
408 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
409 | } | |
410 | ||
0df71650 | 411 | static struct bio *bio_alloc_percpu_cache(struct block_device *bdev, |
16458cf3 | 412 | unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp, |
0df71650 MS |
413 | struct bio_set *bs) |
414 | { | |
415 | struct bio_alloc_cache *cache; | |
416 | struct bio *bio; | |
417 | ||
418 | cache = per_cpu_ptr(bs->cache, get_cpu()); | |
419 | if (!cache->free_list) { | |
420 | put_cpu(); | |
421 | return NULL; | |
422 | } | |
423 | bio = cache->free_list; | |
424 | cache->free_list = bio->bi_next; | |
425 | cache->nr--; | |
426 | put_cpu(); | |
427 | ||
428 | bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf); | |
429 | bio->bi_pool = bs; | |
430 | return bio; | |
431 | } | |
432 | ||
1da177e4 LT |
433 | /** |
434 | * bio_alloc_bioset - allocate a bio for I/O | |
609be106 CH |
435 | * @bdev: block device to allocate the bio for (can be %NULL) |
436 | * @nr_vecs: number of bvecs to pre-allocate | |
437 | * @opf: operation and flags for bio | |
519c8e9f | 438 | * @gfp_mask: the GFP_* mask given to the slab allocator |
db18efac | 439 | * @bs: the bio_set to allocate from. |
1da177e4 | 440 | * |
3175199a | 441 | * Allocate a bio from the mempools in @bs. |
3f86a82a | 442 | * |
3175199a CH |
443 | * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to |
444 | * allocate a bio. This is due to the mempool guarantees. To make this work, | |
445 | * callers must never allocate more than 1 bio at a time from the general pool. | |
446 | * Callers that need to allocate more than 1 bio must always submit the | |
447 | * previously allocated bio for IO before attempting to allocate a new one. | |
448 | * Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 449 | * |
3175199a CH |
450 | * Note that when running under submit_bio_noacct() (i.e. any block driver), |
451 | * bios are not submitted until after you return - see the code in | |
452 | * submit_bio_noacct() that converts recursion into iteration, to prevent | |
453 | * stack overflows. | |
df2cb6da | 454 | * |
3175199a CH |
455 | * This would normally mean allocating multiple bios under submit_bio_noacct() |
456 | * would be susceptible to deadlocks, but we have | |
457 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
458 | * thread. | |
df2cb6da | 459 | * |
3175199a CH |
460 | * However, we do not guarantee forward progress for allocations from other |
461 | * mempools. Doing multiple allocations from the same mempool under | |
462 | * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad | |
463 | * for per bio allocations. | |
df2cb6da | 464 | * |
0df71650 MS |
465 | * If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done from process |
466 | * context, not hard/soft IRQ. | |
467 | * | |
3175199a | 468 | * Returns: Pointer to new bio on success, NULL on failure. |
3f86a82a | 469 | */ |
609be106 | 470 | struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs, |
16458cf3 | 471 | blk_opf_t opf, gfp_t gfp_mask, |
7a88fa19 | 472 | struct bio_set *bs) |
1da177e4 | 473 | { |
df2cb6da | 474 | gfp_t saved_gfp = gfp_mask; |
451a9ebf TH |
475 | struct bio *bio; |
476 | void *p; | |
477 | ||
609be106 CH |
478 | /* should not use nobvec bioset for nr_vecs > 0 */ |
479 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0)) | |
3175199a | 480 | return NULL; |
df2cb6da | 481 | |
0df71650 MS |
482 | if (opf & REQ_ALLOC_CACHE) { |
483 | if (bs->cache && nr_vecs <= BIO_INLINE_VECS) { | |
484 | bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf, | |
485 | gfp_mask, bs); | |
486 | if (bio) | |
487 | return bio; | |
488 | /* | |
489 | * No cached bio available, bio returned below marked with | |
490 | * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache. | |
491 | */ | |
492 | } else { | |
493 | opf &= ~REQ_ALLOC_CACHE; | |
494 | } | |
495 | } | |
496 | ||
3175199a CH |
497 | /* |
498 | * submit_bio_noacct() converts recursion to iteration; this means if | |
499 | * we're running beneath it, any bios we allocate and submit will not be | |
500 | * submitted (and thus freed) until after we return. | |
501 | * | |
502 | * This exposes us to a potential deadlock if we allocate multiple bios | |
503 | * from the same bio_set() while running underneath submit_bio_noacct(). | |
504 | * If we were to allocate multiple bios (say a stacking block driver | |
505 | * that was splitting bios), we would deadlock if we exhausted the | |
506 | * mempool's reserve. | |
507 | * | |
508 | * We solve this, and guarantee forward progress, with a rescuer | |
509 | * workqueue per bio_set. If we go to allocate and there are bios on | |
510 | * current->bio_list, we first try the allocation without | |
511 | * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be | |
512 | * blocking to the rescuer workqueue before we retry with the original | |
513 | * gfp_flags. | |
514 | */ | |
515 | if (current->bio_list && | |
516 | (!bio_list_empty(¤t->bio_list[0]) || | |
517 | !bio_list_empty(¤t->bio_list[1])) && | |
518 | bs->rescue_workqueue) | |
519 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; | |
520 | ||
521 | p = mempool_alloc(&bs->bio_pool, gfp_mask); | |
522 | if (!p && gfp_mask != saved_gfp) { | |
523 | punt_bios_to_rescuer(bs); | |
524 | gfp_mask = saved_gfp; | |
8aa6ba2f | 525 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
3f86a82a | 526 | } |
451a9ebf TH |
527 | if (unlikely(!p)) |
528 | return NULL; | |
759aa12f PB |
529 | if (!mempool_is_saturated(&bs->bio_pool)) |
530 | opf &= ~REQ_ALLOC_CACHE; | |
1da177e4 | 531 | |
3175199a | 532 | bio = p + bs->front_pad; |
609be106 | 533 | if (nr_vecs > BIO_INLINE_VECS) { |
3175199a | 534 | struct bio_vec *bvl = NULL; |
34053979 | 535 | |
609be106 | 536 | bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); |
df2cb6da KO |
537 | if (!bvl && gfp_mask != saved_gfp) { |
538 | punt_bios_to_rescuer(bs); | |
539 | gfp_mask = saved_gfp; | |
609be106 | 540 | bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); |
df2cb6da | 541 | } |
34053979 IM |
542 | if (unlikely(!bvl)) |
543 | goto err_free; | |
a38352e0 | 544 | |
49add496 | 545 | bio_init(bio, bdev, bvl, nr_vecs, opf); |
609be106 | 546 | } else if (nr_vecs) { |
49add496 | 547 | bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf); |
3175199a | 548 | } else { |
49add496 | 549 | bio_init(bio, bdev, NULL, 0, opf); |
1da177e4 | 550 | } |
3f86a82a KO |
551 | |
552 | bio->bi_pool = bs; | |
1da177e4 | 553 | return bio; |
34053979 IM |
554 | |
555 | err_free: | |
8aa6ba2f | 556 | mempool_free(p, &bs->bio_pool); |
34053979 | 557 | return NULL; |
1da177e4 | 558 | } |
a112a71d | 559 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 560 | |
3175199a | 561 | /** |
066ff571 CH |
562 | * bio_kmalloc - kmalloc a bio |
563 | * @nr_vecs: number of bio_vecs to allocate | |
3175199a | 564 | * @gfp_mask: the GFP_* mask given to the slab allocator |
3175199a | 565 | * |
066ff571 CH |
566 | * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized |
567 | * using bio_init() before use. To free a bio returned from this function use | |
568 | * kfree() after calling bio_uninit(). A bio returned from this function can | |
569 | * be reused by calling bio_uninit() before calling bio_init() again. | |
570 | * | |
571 | * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this | |
340e1347 | 572 | * function are not backed by a mempool can fail. Do not use this function |
066ff571 | 573 | * for allocations in the file system I/O path. |
3175199a CH |
574 | * |
575 | * Returns: Pointer to new bio on success, NULL on failure. | |
576 | */ | |
066ff571 | 577 | struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask) |
3175199a CH |
578 | { |
579 | struct bio *bio; | |
580 | ||
066ff571 | 581 | if (nr_vecs > UIO_MAXIOV) |
3175199a | 582 | return NULL; |
066ff571 | 583 | return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask); |
3175199a CH |
584 | } |
585 | EXPORT_SYMBOL(bio_kmalloc); | |
586 | ||
6f822e1b | 587 | void zero_fill_bio(struct bio *bio) |
1da177e4 | 588 | { |
7988613b KO |
589 | struct bio_vec bv; |
590 | struct bvec_iter iter; | |
1da177e4 | 591 | |
ab6c340e CH |
592 | bio_for_each_segment(bv, bio, iter) |
593 | memzero_bvec(&bv); | |
1da177e4 | 594 | } |
6f822e1b | 595 | EXPORT_SYMBOL(zero_fill_bio); |
1da177e4 | 596 | |
83c9c547 ML |
597 | /** |
598 | * bio_truncate - truncate the bio to small size of @new_size | |
599 | * @bio: the bio to be truncated | |
600 | * @new_size: new size for truncating the bio | |
601 | * | |
602 | * Description: | |
603 | * Truncate the bio to new size of @new_size. If bio_op(bio) is | |
604 | * REQ_OP_READ, zero the truncated part. This function should only | |
605 | * be used for handling corner cases, such as bio eod. | |
606 | */ | |
4f7ab09a | 607 | static void bio_truncate(struct bio *bio, unsigned new_size) |
85a8ce62 ML |
608 | { |
609 | struct bio_vec bv; | |
610 | struct bvec_iter iter; | |
611 | unsigned int done = 0; | |
612 | bool truncated = false; | |
613 | ||
614 | if (new_size >= bio->bi_iter.bi_size) | |
615 | return; | |
616 | ||
83c9c547 | 617 | if (bio_op(bio) != REQ_OP_READ) |
85a8ce62 ML |
618 | goto exit; |
619 | ||
620 | bio_for_each_segment(bv, bio, iter) { | |
621 | if (done + bv.bv_len > new_size) { | |
622 | unsigned offset; | |
623 | ||
624 | if (!truncated) | |
625 | offset = new_size - done; | |
626 | else | |
627 | offset = 0; | |
3ee859e3 OH |
628 | zero_user(bv.bv_page, bv.bv_offset + offset, |
629 | bv.bv_len - offset); | |
85a8ce62 ML |
630 | truncated = true; |
631 | } | |
632 | done += bv.bv_len; | |
633 | } | |
634 | ||
635 | exit: | |
636 | /* | |
637 | * Don't touch bvec table here and make it really immutable, since | |
638 | * fs bio user has to retrieve all pages via bio_for_each_segment_all | |
639 | * in its .end_bio() callback. | |
640 | * | |
641 | * It is enough to truncate bio by updating .bi_size since we can make | |
642 | * correct bvec with the updated .bi_size for drivers. | |
643 | */ | |
644 | bio->bi_iter.bi_size = new_size; | |
645 | } | |
646 | ||
29125ed6 CH |
647 | /** |
648 | * guard_bio_eod - truncate a BIO to fit the block device | |
649 | * @bio: bio to truncate | |
650 | * | |
651 | * This allows us to do IO even on the odd last sectors of a device, even if the | |
652 | * block size is some multiple of the physical sector size. | |
653 | * | |
654 | * We'll just truncate the bio to the size of the device, and clear the end of | |
655 | * the buffer head manually. Truly out-of-range accesses will turn into actual | |
656 | * I/O errors, this only handles the "we need to be able to do I/O at the final | |
657 | * sector" case. | |
658 | */ | |
659 | void guard_bio_eod(struct bio *bio) | |
660 | { | |
309dca30 | 661 | sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); |
29125ed6 CH |
662 | |
663 | if (!maxsector) | |
664 | return; | |
665 | ||
666 | /* | |
667 | * If the *whole* IO is past the end of the device, | |
668 | * let it through, and the IO layer will turn it into | |
669 | * an EIO. | |
670 | */ | |
671 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) | |
672 | return; | |
673 | ||
674 | maxsector -= bio->bi_iter.bi_sector; | |
675 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) | |
676 | return; | |
677 | ||
678 | bio_truncate(bio, maxsector << 9); | |
679 | } | |
680 | ||
be4d234d JA |
681 | #define ALLOC_CACHE_MAX 512 |
682 | #define ALLOC_CACHE_SLACK 64 | |
683 | ||
684 | static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, | |
685 | unsigned int nr) | |
686 | { | |
687 | unsigned int i = 0; | |
688 | struct bio *bio; | |
689 | ||
fcade2ce JA |
690 | while ((bio = cache->free_list) != NULL) { |
691 | cache->free_list = bio->bi_next; | |
be4d234d JA |
692 | cache->nr--; |
693 | bio_free(bio); | |
694 | if (++i == nr) | |
695 | break; | |
696 | } | |
697 | } | |
698 | ||
699 | static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) | |
700 | { | |
701 | struct bio_set *bs; | |
702 | ||
703 | bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); | |
704 | if (bs->cache) { | |
705 | struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); | |
706 | ||
707 | bio_alloc_cache_prune(cache, -1U); | |
708 | } | |
709 | return 0; | |
710 | } | |
711 | ||
712 | static void bio_alloc_cache_destroy(struct bio_set *bs) | |
713 | { | |
714 | int cpu; | |
715 | ||
716 | if (!bs->cache) | |
717 | return; | |
718 | ||
719 | cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); | |
720 | for_each_possible_cpu(cpu) { | |
721 | struct bio_alloc_cache *cache; | |
722 | ||
723 | cache = per_cpu_ptr(bs->cache, cpu); | |
724 | bio_alloc_cache_prune(cache, -1U); | |
725 | } | |
726 | free_percpu(bs->cache); | |
605f7415 | 727 | bs->cache = NULL; |
be4d234d JA |
728 | } |
729 | ||
1da177e4 LT |
730 | /** |
731 | * bio_put - release a reference to a bio | |
732 | * @bio: bio to release reference to | |
733 | * | |
734 | * Description: | |
735 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 736 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
737 | **/ |
738 | void bio_put(struct bio *bio) | |
739 | { | |
be4d234d | 740 | if (unlikely(bio_flagged(bio, BIO_REFFED))) { |
9e8c0d0d | 741 | BUG_ON(!atomic_read(&bio->__bi_cnt)); |
be4d234d JA |
742 | if (!atomic_dec_and_test(&bio->__bi_cnt)) |
743 | return; | |
744 | } | |
dac56212 | 745 | |
d4347d50 | 746 | if ((bio->bi_opf & REQ_ALLOC_CACHE) && !WARN_ON_ONCE(in_interrupt())) { |
be4d234d JA |
747 | struct bio_alloc_cache *cache; |
748 | ||
749 | bio_uninit(bio); | |
750 | cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); | |
fcade2ce JA |
751 | bio->bi_next = cache->free_list; |
752 | cache->free_list = bio; | |
be4d234d JA |
753 | if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) |
754 | bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); | |
755 | put_cpu(); | |
756 | } else { | |
757 | bio_free(bio); | |
dac56212 | 758 | } |
1da177e4 | 759 | } |
a112a71d | 760 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 761 | |
a0e8de79 | 762 | static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp) |
59d276fe | 763 | { |
b7c44ed9 | 764 | bio_set_flag(bio, BIO_CLONED); |
ca474b73 | 765 | bio->bi_ioprio = bio_src->bi_ioprio; |
59d276fe | 766 | bio->bi_iter = bio_src->bi_iter; |
20bd723e | 767 | |
7ecc56c6 CH |
768 | if (bio->bi_bdev) { |
769 | if (bio->bi_bdev == bio_src->bi_bdev && | |
770 | bio_flagged(bio_src, BIO_REMAPPED)) | |
771 | bio_set_flag(bio, BIO_REMAPPED); | |
772 | bio_clone_blkg_association(bio, bio_src); | |
773 | } | |
56b4b5ab CH |
774 | |
775 | if (bio_crypt_clone(bio, bio_src, gfp) < 0) | |
776 | return -ENOMEM; | |
777 | if (bio_integrity(bio_src) && | |
778 | bio_integrity_clone(bio, bio_src, gfp) < 0) | |
779 | return -ENOMEM; | |
780 | return 0; | |
59d276fe | 781 | } |
59d276fe KO |
782 | |
783 | /** | |
abfc426d CH |
784 | * bio_alloc_clone - clone a bio that shares the original bio's biovec |
785 | * @bdev: block_device to clone onto | |
a0e8de79 CH |
786 | * @bio_src: bio to clone from |
787 | * @gfp: allocation priority | |
788 | * @bs: bio_set to allocate from | |
59d276fe | 789 | * |
a0e8de79 CH |
790 | * Allocate a new bio that is a clone of @bio_src. The caller owns the returned |
791 | * bio, but not the actual data it points to. | |
792 | * | |
793 | * The caller must ensure that the return bio is not freed before @bio_src. | |
59d276fe | 794 | */ |
abfc426d CH |
795 | struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src, |
796 | gfp_t gfp, struct bio_set *bs) | |
59d276fe | 797 | { |
a0e8de79 | 798 | struct bio *bio; |
59d276fe | 799 | |
abfc426d | 800 | bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs); |
a0e8de79 | 801 | if (!bio) |
59d276fe KO |
802 | return NULL; |
803 | ||
a0e8de79 CH |
804 | if (__bio_clone(bio, bio_src, gfp) < 0) { |
805 | bio_put(bio); | |
56b4b5ab CH |
806 | return NULL; |
807 | } | |
a0e8de79 | 808 | bio->bi_io_vec = bio_src->bi_io_vec; |
59d276fe | 809 | |
a0e8de79 | 810 | return bio; |
59d276fe | 811 | } |
abfc426d | 812 | EXPORT_SYMBOL(bio_alloc_clone); |
59d276fe | 813 | |
a0e8de79 | 814 | /** |
abfc426d CH |
815 | * bio_init_clone - clone a bio that shares the original bio's biovec |
816 | * @bdev: block_device to clone onto | |
a0e8de79 CH |
817 | * @bio: bio to clone into |
818 | * @bio_src: bio to clone from | |
819 | * @gfp: allocation priority | |
820 | * | |
821 | * Initialize a new bio in caller provided memory that is a clone of @bio_src. | |
822 | * The caller owns the returned bio, but not the actual data it points to. | |
823 | * | |
824 | * The caller must ensure that @bio_src is not freed before @bio. | |
825 | */ | |
abfc426d CH |
826 | int bio_init_clone(struct block_device *bdev, struct bio *bio, |
827 | struct bio *bio_src, gfp_t gfp) | |
a0e8de79 CH |
828 | { |
829 | int ret; | |
830 | ||
abfc426d | 831 | bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf); |
a0e8de79 CH |
832 | ret = __bio_clone(bio, bio_src, gfp); |
833 | if (ret) | |
834 | bio_uninit(bio); | |
835 | return ret; | |
836 | } | |
abfc426d | 837 | EXPORT_SYMBOL(bio_init_clone); |
a0e8de79 | 838 | |
9a6083be CH |
839 | /** |
840 | * bio_full - check if the bio is full | |
841 | * @bio: bio to check | |
842 | * @len: length of one segment to be added | |
843 | * | |
844 | * Return true if @bio is full and one segment with @len bytes can't be | |
845 | * added to the bio, otherwise return false | |
846 | */ | |
847 | static inline bool bio_full(struct bio *bio, unsigned len) | |
848 | { | |
849 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
850 | return true; | |
851 | if (bio->bi_iter.bi_size > UINT_MAX - len) | |
852 | return true; | |
853 | return false; | |
854 | } | |
855 | ||
5919482e ML |
856 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
857 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 858 | bool *same_page) |
5919482e | 859 | { |
d8166519 MWO |
860 | size_t bv_end = bv->bv_offset + bv->bv_len; |
861 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; | |
5919482e ML |
862 | phys_addr_t page_addr = page_to_phys(page); |
863 | ||
864 | if (vec_end_addr + 1 != page_addr + off) | |
865 | return false; | |
866 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
867 | return false; | |
49580e69 LG |
868 | if (!zone_device_pages_have_same_pgmap(bv->bv_page, page)) |
869 | return false; | |
52d52d1c | 870 | |
ff896738 | 871 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
d8166519 MWO |
872 | if (*same_page) |
873 | return true; | |
11b331f8 AP |
874 | else if (IS_ENABLED(CONFIG_KMSAN)) |
875 | return false; | |
d8166519 | 876 | return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); |
5919482e ML |
877 | } |
878 | ||
9774b391 CH |
879 | /** |
880 | * __bio_try_merge_page - try appending data to an existing bvec. | |
881 | * @bio: destination bio | |
882 | * @page: start page to add | |
883 | * @len: length of the data to add | |
884 | * @off: offset of the data relative to @page | |
885 | * @same_page: return if the segment has been merged inside the same page | |
886 | * | |
887 | * Try to add the data at @page + @off to the last bvec of @bio. This is a | |
888 | * useful optimisation for file systems with a block size smaller than the | |
889 | * page size. | |
890 | * | |
891 | * Warn if (@len, @off) crosses pages in case that @same_page is true. | |
892 | * | |
893 | * Return %true on success or %false on failure. | |
894 | */ | |
895 | static bool __bio_try_merge_page(struct bio *bio, struct page *page, | |
896 | unsigned int len, unsigned int off, bool *same_page) | |
897 | { | |
898 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) | |
899 | return false; | |
900 | ||
901 | if (bio->bi_vcnt > 0) { | |
902 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
903 | ||
904 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
905 | if (bio->bi_iter.bi_size > UINT_MAX - len) { | |
906 | *same_page = false; | |
907 | return false; | |
908 | } | |
909 | bv->bv_len += len; | |
910 | bio->bi_iter.bi_size += len; | |
911 | return true; | |
912 | } | |
913 | } | |
914 | return false; | |
915 | } | |
916 | ||
e4581105 CH |
917 | /* |
918 | * Try to merge a page into a segment, while obeying the hardware segment | |
919 | * size limit. This is not for normal read/write bios, but for passthrough | |
920 | * or Zone Append operations that we can't split. | |
921 | */ | |
922 | static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, | |
923 | struct page *page, unsigned len, | |
924 | unsigned offset, bool *same_page) | |
489fbbcb | 925 | { |
384209cd | 926 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
489fbbcb ML |
927 | unsigned long mask = queue_segment_boundary(q); |
928 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
929 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
930 | ||
931 | if ((addr1 | mask) != (addr2 | mask)) | |
932 | return false; | |
489fbbcb ML |
933 | if (bv->bv_len + len > queue_max_segment_size(q)) |
934 | return false; | |
384209cd | 935 | return __bio_try_merge_page(bio, page, len, offset, same_page); |
489fbbcb ML |
936 | } |
937 | ||
1da177e4 | 938 | /** |
e4581105 CH |
939 | * bio_add_hw_page - attempt to add a page to a bio with hw constraints |
940 | * @q: the target queue | |
941 | * @bio: destination bio | |
942 | * @page: page to add | |
943 | * @len: vec entry length | |
944 | * @offset: vec entry offset | |
945 | * @max_sectors: maximum number of sectors that can be added | |
946 | * @same_page: return if the segment has been merged inside the same page | |
c66a14d0 | 947 | * |
e4581105 CH |
948 | * Add a page to a bio while respecting the hardware max_sectors, max_segment |
949 | * and gap limitations. | |
1da177e4 | 950 | */ |
e4581105 | 951 | int bio_add_hw_page(struct request_queue *q, struct bio *bio, |
19047087 | 952 | struct page *page, unsigned int len, unsigned int offset, |
e4581105 | 953 | unsigned int max_sectors, bool *same_page) |
1da177e4 | 954 | { |
1da177e4 LT |
955 | struct bio_vec *bvec; |
956 | ||
e4581105 | 957 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
1da177e4 LT |
958 | return 0; |
959 | ||
e4581105 | 960 | if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
961 | return 0; |
962 | ||
80cfd548 | 963 | if (bio->bi_vcnt > 0) { |
e4581105 | 964 | if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) |
384209cd | 965 | return len; |
320ea869 CH |
966 | |
967 | /* | |
968 | * If the queue doesn't support SG gaps and adding this segment | |
969 | * would create a gap, disallow it. | |
970 | */ | |
384209cd | 971 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
c55ddd90 | 972 | if (bvec_gap_to_prev(&q->limits, bvec, offset)) |
320ea869 | 973 | return 0; |
80cfd548 JA |
974 | } |
975 | ||
79d08f89 | 976 | if (bio_full(bio, len)) |
1da177e4 LT |
977 | return 0; |
978 | ||
14ccb66b | 979 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
980 | return 0; |
981 | ||
fcbf6a08 ML |
982 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
983 | bvec->bv_page = page; | |
984 | bvec->bv_len = len; | |
985 | bvec->bv_offset = offset; | |
986 | bio->bi_vcnt++; | |
dcdca753 | 987 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
988 | return len; |
989 | } | |
19047087 | 990 | |
e4581105 CH |
991 | /** |
992 | * bio_add_pc_page - attempt to add page to passthrough bio | |
993 | * @q: the target queue | |
994 | * @bio: destination bio | |
995 | * @page: page to add | |
996 | * @len: vec entry length | |
997 | * @offset: vec entry offset | |
998 | * | |
999 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
1000 | * number of reasons, such as the bio being full or target block device | |
1001 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
1002 | * so it is always possible to add a single page to an empty bio. | |
1003 | * | |
1004 | * This should only be used by passthrough bios. | |
1005 | */ | |
19047087 ML |
1006 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, |
1007 | struct page *page, unsigned int len, unsigned int offset) | |
1008 | { | |
d1916c86 | 1009 | bool same_page = false; |
e4581105 CH |
1010 | return bio_add_hw_page(q, bio, page, len, offset, |
1011 | queue_max_hw_sectors(q), &same_page); | |
19047087 | 1012 | } |
a112a71d | 1013 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 1014 | |
ae29333f JT |
1015 | /** |
1016 | * bio_add_zone_append_page - attempt to add page to zone-append bio | |
1017 | * @bio: destination bio | |
1018 | * @page: page to add | |
1019 | * @len: vec entry length | |
1020 | * @offset: vec entry offset | |
1021 | * | |
1022 | * Attempt to add a page to the bio_vec maplist of a bio that will be submitted | |
1023 | * for a zone-append request. This can fail for a number of reasons, such as the | |
1024 | * bio being full or the target block device is not a zoned block device or | |
1025 | * other limitations of the target block device. The target block device must | |
1026 | * allow bio's up to PAGE_SIZE, so it is always possible to add a single page | |
1027 | * to an empty bio. | |
1028 | * | |
1029 | * Returns: number of bytes added to the bio, or 0 in case of a failure. | |
1030 | */ | |
1031 | int bio_add_zone_append_page(struct bio *bio, struct page *page, | |
1032 | unsigned int len, unsigned int offset) | |
1033 | { | |
3caee463 | 1034 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
ae29333f JT |
1035 | bool same_page = false; |
1036 | ||
1037 | if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) | |
1038 | return 0; | |
1039 | ||
edd1dbc8 | 1040 | if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev))) |
ae29333f JT |
1041 | return 0; |
1042 | ||
1043 | return bio_add_hw_page(q, bio, page, len, offset, | |
1044 | queue_max_zone_append_sectors(q), &same_page); | |
1045 | } | |
1046 | EXPORT_SYMBOL_GPL(bio_add_zone_append_page); | |
1047 | ||
0aa69fd3 | 1048 | /** |
551879a4 | 1049 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 1050 | * @bio: destination bio |
551879a4 ML |
1051 | * @page: start page to add |
1052 | * @len: length of the data to add, may cross pages | |
1053 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
1054 | * |
1055 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
1056 | * that @bio has space for another bvec. | |
1057 | */ | |
1058 | void __bio_add_page(struct bio *bio, struct page *page, | |
1059 | unsigned int len, unsigned int off) | |
1060 | { | |
1061 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 1062 | |
0aa69fd3 | 1063 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 1064 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
1065 | |
1066 | bv->bv_page = page; | |
1067 | bv->bv_offset = off; | |
1068 | bv->bv_len = len; | |
c66a14d0 | 1069 | |
c66a14d0 | 1070 | bio->bi_iter.bi_size += len; |
0aa69fd3 CH |
1071 | bio->bi_vcnt++; |
1072 | } | |
1073 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
1074 | ||
1075 | /** | |
551879a4 | 1076 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 1077 | * @bio: destination bio |
551879a4 ML |
1078 | * @page: start page to add |
1079 | * @len: vec entry length, may cross pages | |
1080 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 1081 | * |
551879a4 | 1082 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
1083 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
1084 | */ | |
1085 | int bio_add_page(struct bio *bio, struct page *page, | |
1086 | unsigned int len, unsigned int offset) | |
1087 | { | |
ff896738 CH |
1088 | bool same_page = false; |
1089 | ||
1090 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 1091 | if (bio_full(bio, len)) |
0aa69fd3 CH |
1092 | return 0; |
1093 | __bio_add_page(bio, page, len, offset); | |
1094 | } | |
c66a14d0 | 1095 | return len; |
1da177e4 | 1096 | } |
a112a71d | 1097 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 1098 | |
85f5a74c MWO |
1099 | /** |
1100 | * bio_add_folio - Attempt to add part of a folio to a bio. | |
1101 | * @bio: BIO to add to. | |
1102 | * @folio: Folio to add. | |
1103 | * @len: How many bytes from the folio to add. | |
1104 | * @off: First byte in this folio to add. | |
1105 | * | |
1106 | * Filesystems that use folios can call this function instead of calling | |
1107 | * bio_add_page() for each page in the folio. If @off is bigger than | |
1108 | * PAGE_SIZE, this function can create a bio_vec that starts in a page | |
1109 | * after the bv_page. BIOs do not support folios that are 4GiB or larger. | |
1110 | * | |
1111 | * Return: Whether the addition was successful. | |
1112 | */ | |
1113 | bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, | |
1114 | size_t off) | |
1115 | { | |
1116 | if (len > UINT_MAX || off > UINT_MAX) | |
455a844d | 1117 | return false; |
85f5a74c MWO |
1118 | return bio_add_page(bio, &folio->page, len, off) > 0; |
1119 | } | |
1120 | ||
c809084a | 1121 | void __bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
1122 | { |
1123 | struct bvec_iter_all iter_all; | |
1124 | struct bio_vec *bvec; | |
7321ecbf | 1125 | |
d241a95f CH |
1126 | bio_for_each_segment_all(bvec, bio, iter_all) { |
1127 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
1128 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 1129 | put_page(bvec->bv_page); |
d241a95f | 1130 | } |
7321ecbf | 1131 | } |
c809084a | 1132 | EXPORT_SYMBOL_GPL(__bio_release_pages); |
7321ecbf | 1133 | |
1bb6b810 | 1134 | void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) |
6d0c48ae | 1135 | { |
fa5fa8ec PB |
1136 | size_t size = iov_iter_count(iter); |
1137 | ||
7a800a20 | 1138 | WARN_ON_ONCE(bio->bi_max_vecs); |
c42bca92 | 1139 | |
fa5fa8ec PB |
1140 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
1141 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); | |
1142 | size_t max_sectors = queue_max_zone_append_sectors(q); | |
1143 | ||
1144 | size = min(size, max_sectors << SECTOR_SHIFT); | |
1145 | } | |
1146 | ||
c42bca92 | 1147 | bio->bi_vcnt = iter->nr_segs; |
c42bca92 PB |
1148 | bio->bi_io_vec = (struct bio_vec *)iter->bvec; |
1149 | bio->bi_iter.bi_bvec_done = iter->iov_offset; | |
fa5fa8ec | 1150 | bio->bi_iter.bi_size = size; |
ed97ce5e | 1151 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
977be012 | 1152 | bio_set_flag(bio, BIO_CLONED); |
7de55b7d | 1153 | } |
c42bca92 | 1154 | |
c58c0074 KB |
1155 | static int bio_iov_add_page(struct bio *bio, struct page *page, |
1156 | unsigned int len, unsigned int offset) | |
1157 | { | |
1158 | bool same_page = false; | |
1159 | ||
1160 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
c58c0074 KB |
1161 | __bio_add_page(bio, page, len, offset); |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | if (same_page) | |
1166 | put_page(page); | |
1167 | return 0; | |
1168 | } | |
1169 | ||
1170 | static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page, | |
1171 | unsigned int len, unsigned int offset) | |
1172 | { | |
1173 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); | |
1174 | bool same_page = false; | |
1175 | ||
1176 | if (bio_add_hw_page(q, bio, page, len, offset, | |
1177 | queue_max_zone_append_sectors(q), &same_page) != len) | |
1178 | return -EINVAL; | |
1179 | if (same_page) | |
1180 | put_page(page); | |
1181 | return 0; | |
1182 | } | |
1183 | ||
576ed913 CH |
1184 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
1185 | ||
2cefe4db | 1186 | /** |
17d51b10 | 1187 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
1188 | * @bio: bio to add pages to |
1189 | * @iter: iov iterator describing the region to be mapped | |
1190 | * | |
17d51b10 | 1191 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 1192 | * pages will have to be released using put_page() when done. |
17d51b10 | 1193 | * For multi-segment *iter, this function only adds pages from the |
3cf14889 | 1194 | * next non-empty segment of the iov iterator. |
2cefe4db | 1195 | */ |
17d51b10 | 1196 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 1197 | { |
576ed913 CH |
1198 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
1199 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
1200 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
1201 | struct page **pages = (struct page **)bv; | |
5e3e3f2e | 1202 | unsigned int gup_flags = 0; |
576ed913 | 1203 | ssize_t size, left; |
e97424fd | 1204 | unsigned len, i = 0; |
480cb846 | 1205 | size_t offset, trim; |
325347d9 | 1206 | int ret = 0; |
576ed913 CH |
1207 | |
1208 | /* | |
1209 | * Move page array up in the allocated memory for the bio vecs as far as | |
1210 | * possible so that we can start filling biovecs from the beginning | |
1211 | * without overwriting the temporary page array. | |
c58c0074 | 1212 | */ |
576ed913 CH |
1213 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); |
1214 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db | 1215 | |
5e3e3f2e LG |
1216 | if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue)) |
1217 | gup_flags |= FOLL_PCI_P2PDMA; | |
1218 | ||
b1a000d3 KB |
1219 | /* |
1220 | * Each segment in the iov is required to be a block size multiple. | |
1221 | * However, we may not be able to get the entire segment if it spans | |
1222 | * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the | |
1223 | * result to ensure the bio's total size is correct. The remainder of | |
1224 | * the iov data will be picked up in the next bio iteration. | |
1225 | */ | |
5e3e3f2e LG |
1226 | size = iov_iter_get_pages(iter, pages, |
1227 | UINT_MAX - bio->bi_iter.bi_size, | |
1228 | nr_pages, &offset, gup_flags); | |
480cb846 AV |
1229 | if (unlikely(size <= 0)) |
1230 | return size ? size : -EFAULT; | |
1231 | ||
1232 | nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE); | |
1233 | ||
1234 | trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1); | |
1235 | iov_iter_revert(iter, trim); | |
1236 | ||
1237 | size -= trim; | |
1238 | if (unlikely(!size)) { | |
1239 | ret = -EFAULT; | |
e97424fd KB |
1240 | goto out; |
1241 | } | |
2cefe4db | 1242 | |
576ed913 CH |
1243 | for (left = size, i = 0; left > 0; left -= len, i++) { |
1244 | struct page *page = pages[i]; | |
2cefe4db | 1245 | |
576ed913 | 1246 | len = min_t(size_t, PAGE_SIZE - offset, left); |
34cdb8c8 | 1247 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { |
c58c0074 KB |
1248 | ret = bio_iov_add_zone_append_page(bio, page, len, |
1249 | offset); | |
e97424fd | 1250 | if (ret) |
34cdb8c8 | 1251 | break; |
34cdb8c8 KB |
1252 | } else |
1253 | bio_iov_add_page(bio, page, len, offset); | |
45691804 | 1254 | |
576ed913 | 1255 | offset = 0; |
2cefe4db KO |
1256 | } |
1257 | ||
480cb846 | 1258 | iov_iter_revert(iter, left); |
e97424fd KB |
1259 | out: |
1260 | while (i < nr_pages) | |
1261 | put_page(pages[i++]); | |
1262 | ||
325347d9 | 1263 | return ret; |
2cefe4db | 1264 | } |
17d51b10 MW |
1265 | |
1266 | /** | |
6d0c48ae | 1267 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 1268 | * @bio: bio to add pages to |
6d0c48ae JA |
1269 | * @iter: iov iterator describing the region to be added |
1270 | * | |
1271 | * This takes either an iterator pointing to user memory, or one pointing to | |
1272 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
1273 | * map them into the kernel. On IO completion, the caller should put those | |
c42bca92 PB |
1274 | * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided |
1275 | * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs | |
1276 | * to ensure the bvecs and pages stay referenced until the submitted I/O is | |
1277 | * completed by a call to ->ki_complete() or returns with an error other than | |
1278 | * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF | |
1279 | * on IO completion. If it isn't, then pages should be released. | |
17d51b10 | 1280 | * |
17d51b10 | 1281 | * The function tries, but does not guarantee, to pin as many pages as |
5cd3ddc1 | 1282 | * fit into the bio, or are requested in @iter, whatever is smaller. If |
6d0c48ae JA |
1283 | * MM encounters an error pinning the requested pages, it stops. Error |
1284 | * is returned only if 0 pages could be pinned. | |
17d51b10 MW |
1285 | */ |
1286 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
1287 | { | |
c42bca92 | 1288 | int ret = 0; |
14eacf12 | 1289 | |
c42bca92 | 1290 | if (iov_iter_is_bvec(iter)) { |
fa5fa8ec PB |
1291 | bio_iov_bvec_set(bio, iter); |
1292 | iov_iter_advance(iter, bio->bi_iter.bi_size); | |
1293 | return 0; | |
c42bca92 | 1294 | } |
17d51b10 MW |
1295 | |
1296 | do { | |
c58c0074 | 1297 | ret = __bio_iov_iter_get_pages(bio, iter); |
79d08f89 | 1298 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
17d51b10 | 1299 | |
14eacf12 | 1300 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 1301 | } |
29b2a3aa | 1302 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); |
2cefe4db | 1303 | |
4246a0b6 | 1304 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 1305 | { |
65e53aab | 1306 | complete(bio->bi_private); |
9e882242 KO |
1307 | } |
1308 | ||
1309 | /** | |
1310 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
1311 | * @bio: The &struct bio which describes the I/O |
1312 | * | |
1313 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
1314 | * bio_endio() on failure. | |
3d289d68 JK |
1315 | * |
1316 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
1317 | * result in bio reference to be consumed. The caller must drop the reference | |
1318 | * on his own. | |
9e882242 | 1319 | */ |
4e49ea4a | 1320 | int submit_bio_wait(struct bio *bio) |
9e882242 | 1321 | { |
309dca30 CH |
1322 | DECLARE_COMPLETION_ONSTACK_MAP(done, |
1323 | bio->bi_bdev->bd_disk->lockdep_map); | |
de6a78b6 | 1324 | unsigned long hang_check; |
9e882242 | 1325 | |
65e53aab | 1326 | bio->bi_private = &done; |
9e882242 | 1327 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 1328 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 1329 | submit_bio(bio); |
de6a78b6 ML |
1330 | |
1331 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1332 | hang_check = sysctl_hung_task_timeout_secs; | |
1333 | if (hang_check) | |
1334 | while (!wait_for_completion_io_timeout(&done, | |
1335 | hang_check * (HZ/2))) | |
1336 | ; | |
1337 | else | |
1338 | wait_for_completion_io(&done); | |
9e882242 | 1339 | |
65e53aab | 1340 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
1341 | } |
1342 | EXPORT_SYMBOL(submit_bio_wait); | |
1343 | ||
d4aa57a1 | 1344 | void __bio_advance(struct bio *bio, unsigned bytes) |
054bdf64 KO |
1345 | { |
1346 | if (bio_integrity(bio)) | |
1347 | bio_integrity_advance(bio, bytes); | |
1348 | ||
a892c8d5 | 1349 | bio_crypt_advance(bio, bytes); |
4550dd6c | 1350 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 | 1351 | } |
d4aa57a1 | 1352 | EXPORT_SYMBOL(__bio_advance); |
054bdf64 | 1353 | |
45db54d5 KO |
1354 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1355 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1356 | { |
45db54d5 | 1357 | while (src_iter->bi_size && dst_iter->bi_size) { |
f8b679a0 CH |
1358 | struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); |
1359 | struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1360 | unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
403d5034 CH |
1361 | void *src_buf = bvec_kmap_local(&src_bv); |
1362 | void *dst_buf = bvec_kmap_local(&dst_bv); | |
f8b679a0 | 1363 | |
403d5034 CH |
1364 | memcpy(dst_buf, src_buf, bytes); |
1365 | ||
1366 | kunmap_local(dst_buf); | |
f8b679a0 | 1367 | kunmap_local(src_buf); |
6e6e811d | 1368 | |
22b56c29 PB |
1369 | bio_advance_iter_single(src, src_iter, bytes); |
1370 | bio_advance_iter_single(dst, dst_iter, bytes); | |
16ac3d63 KO |
1371 | } |
1372 | } | |
38a72dac KO |
1373 | EXPORT_SYMBOL(bio_copy_data_iter); |
1374 | ||
1375 | /** | |
45db54d5 KO |
1376 | * bio_copy_data - copy contents of data buffers from one bio to another |
1377 | * @src: source bio | |
1378 | * @dst: destination bio | |
38a72dac KO |
1379 | * |
1380 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1381 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1382 | */ | |
1383 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1384 | { | |
45db54d5 KO |
1385 | struct bvec_iter src_iter = src->bi_iter; |
1386 | struct bvec_iter dst_iter = dst->bi_iter; | |
1387 | ||
1388 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1389 | } |
16ac3d63 KO |
1390 | EXPORT_SYMBOL(bio_copy_data); |
1391 | ||
491221f8 | 1392 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1393 | { |
1394 | struct bio_vec *bvec; | |
6dc4f100 | 1395 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1396 | |
2b070cfe | 1397 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1398 | __free_page(bvec->bv_page); |
1399 | } | |
491221f8 | 1400 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1401 | |
1da177e4 LT |
1402 | /* |
1403 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1404 | * for performing direct-IO in BIOs. | |
1405 | * | |
1406 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1407 | * because the required locks are not interrupt-safe. So what we can do is to | |
1408 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1409 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1410 | * in process context. | |
1411 | * | |
1412 | * We special-case compound pages here: normally this means reads into hugetlb | |
1413 | * pages. The logic in here doesn't really work right for compound pages | |
1414 | * because the VM does not uniformly chase down the head page in all cases. | |
1415 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1416 | * handle them at all. So we skip compound pages here at an early stage. | |
1417 | * | |
1418 | * Note that this code is very hard to test under normal circumstances because | |
1419 | * direct-io pins the pages with get_user_pages(). This makes | |
1420 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1421 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1422 | * pagecache. |
1423 | * | |
1424 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1425 | * deferred bio dirtying paths. | |
1426 | */ | |
1427 | ||
1428 | /* | |
1429 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1430 | */ | |
1431 | void bio_set_pages_dirty(struct bio *bio) | |
1432 | { | |
cb34e057 | 1433 | struct bio_vec *bvec; |
6dc4f100 | 1434 | struct bvec_iter_all iter_all; |
1da177e4 | 1435 | |
2b070cfe | 1436 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1437 | if (!PageCompound(bvec->bv_page)) |
1438 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1439 | } |
1440 | } | |
1441 | ||
1da177e4 LT |
1442 | /* |
1443 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1444 | * If they are, then fine. If, however, some pages are clean then they must | |
1445 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1446 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1447 | * |
1448 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1449 | * here on. It will run one put_page() against each page and will run one |
1450 | * bio_put() against the BIO. | |
1da177e4 LT |
1451 | */ |
1452 | ||
65f27f38 | 1453 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1454 | |
65f27f38 | 1455 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1456 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1457 | static struct bio *bio_dirty_list; | |
1458 | ||
1459 | /* | |
1460 | * This runs in process context | |
1461 | */ | |
65f27f38 | 1462 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1463 | { |
24d5493f | 1464 | struct bio *bio, *next; |
1da177e4 | 1465 | |
24d5493f CH |
1466 | spin_lock_irq(&bio_dirty_lock); |
1467 | next = bio_dirty_list; | |
1da177e4 | 1468 | bio_dirty_list = NULL; |
24d5493f | 1469 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1470 | |
24d5493f CH |
1471 | while ((bio = next) != NULL) { |
1472 | next = bio->bi_private; | |
1da177e4 | 1473 | |
d241a95f | 1474 | bio_release_pages(bio, true); |
1da177e4 | 1475 | bio_put(bio); |
1da177e4 LT |
1476 | } |
1477 | } | |
1478 | ||
1479 | void bio_check_pages_dirty(struct bio *bio) | |
1480 | { | |
cb34e057 | 1481 | struct bio_vec *bvec; |
24d5493f | 1482 | unsigned long flags; |
6dc4f100 | 1483 | struct bvec_iter_all iter_all; |
1da177e4 | 1484 | |
2b070cfe | 1485 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1486 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1487 | goto defer; | |
1da177e4 LT |
1488 | } |
1489 | ||
d241a95f | 1490 | bio_release_pages(bio, false); |
24d5493f CH |
1491 | bio_put(bio); |
1492 | return; | |
1493 | defer: | |
1494 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1495 | bio->bi_private = bio_dirty_list; | |
1496 | bio_dirty_list = bio; | |
1497 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1498 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1499 | } |
1500 | ||
c4cf5261 JA |
1501 | static inline bool bio_remaining_done(struct bio *bio) |
1502 | { | |
1503 | /* | |
1504 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1505 | * we always end io on the first invocation. | |
1506 | */ | |
1507 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1508 | return true; | |
1509 | ||
1510 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1511 | ||
326e1dbb | 1512 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1513 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1514 | return true; |
326e1dbb | 1515 | } |
c4cf5261 JA |
1516 | |
1517 | return false; | |
1518 | } | |
1519 | ||
1da177e4 LT |
1520 | /** |
1521 | * bio_endio - end I/O on a bio | |
1522 | * @bio: bio | |
1da177e4 LT |
1523 | * |
1524 | * Description: | |
4246a0b6 CH |
1525 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1526 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1527 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1528 | * |
1529 | * bio_endio() can be called several times on a bio that has been chained | |
1530 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
60b6a7e6 | 1531 | * last time. |
1da177e4 | 1532 | **/ |
4246a0b6 | 1533 | void bio_endio(struct bio *bio) |
1da177e4 | 1534 | { |
ba8c6967 | 1535 | again: |
2b885517 | 1536 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1537 | return; |
7c20f116 CH |
1538 | if (!bio_integrity_endio(bio)) |
1539 | return; | |
1da177e4 | 1540 | |
aa1b46dc | 1541 | rq_qos_done_bio(bio); |
67b42d0b | 1542 | |
60b6a7e6 | 1543 | if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
3caee463 | 1544 | trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); |
60b6a7e6 EH |
1545 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
1546 | } | |
1547 | ||
ba8c6967 CH |
1548 | /* |
1549 | * Need to have a real endio function for chained bios, otherwise | |
1550 | * various corner cases will break (like stacking block devices that | |
1551 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1552 | * recursion and blowing the stack. Tail call optimization would | |
1553 | * handle this, but compiling with frame pointers also disables | |
1554 | * gcc's sibling call optimization. | |
1555 | */ | |
1556 | if (bio->bi_end_io == bio_chain_endio) { | |
1557 | bio = __bio_chain_endio(bio); | |
1558 | goto again; | |
196d38bc | 1559 | } |
ba8c6967 | 1560 | |
9e234eea | 1561 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1562 | /* release cgroup info */ |
1563 | bio_uninit(bio); | |
ba8c6967 CH |
1564 | if (bio->bi_end_io) |
1565 | bio->bi_end_io(bio); | |
1da177e4 | 1566 | } |
a112a71d | 1567 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1568 | |
20d0189b KO |
1569 | /** |
1570 | * bio_split - split a bio | |
1571 | * @bio: bio to split | |
1572 | * @sectors: number of sectors to split from the front of @bio | |
1573 | * @gfp: gfp mask | |
1574 | * @bs: bio set to allocate from | |
1575 | * | |
1576 | * Allocates and returns a new bio which represents @sectors from the start of | |
1577 | * @bio, and updates @bio to represent the remaining sectors. | |
1578 | * | |
f3f5da62 | 1579 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1580 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1581 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1582 | */ |
1583 | struct bio *bio_split(struct bio *bio, int sectors, | |
1584 | gfp_t gfp, struct bio_set *bs) | |
1585 | { | |
f341a4d3 | 1586 | struct bio *split; |
20d0189b KO |
1587 | |
1588 | BUG_ON(sectors <= 0); | |
1589 | BUG_ON(sectors >= bio_sectors(bio)); | |
1590 | ||
0512a75b KB |
1591 | /* Zone append commands cannot be split */ |
1592 | if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) | |
1593 | return NULL; | |
1594 | ||
abfc426d | 1595 | split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs); |
20d0189b KO |
1596 | if (!split) |
1597 | return NULL; | |
1598 | ||
1599 | split->bi_iter.bi_size = sectors << 9; | |
1600 | ||
1601 | if (bio_integrity(split)) | |
fbd08e76 | 1602 | bio_integrity_trim(split); |
20d0189b KO |
1603 | |
1604 | bio_advance(bio, split->bi_iter.bi_size); | |
1605 | ||
fbbaf700 | 1606 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1607 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1608 | |
20d0189b KO |
1609 | return split; |
1610 | } | |
1611 | EXPORT_SYMBOL(bio_split); | |
1612 | ||
6678d83f KO |
1613 | /** |
1614 | * bio_trim - trim a bio | |
1615 | * @bio: bio to trim | |
1616 | * @offset: number of sectors to trim from the front of @bio | |
1617 | * @size: size we want to trim @bio to, in sectors | |
e83502ca CK |
1618 | * |
1619 | * This function is typically used for bios that are cloned and submitted | |
1620 | * to the underlying device in parts. | |
6678d83f | 1621 | */ |
e83502ca | 1622 | void bio_trim(struct bio *bio, sector_t offset, sector_t size) |
6678d83f | 1623 | { |
e83502ca | 1624 | if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || |
8535c018 | 1625 | offset + size > bio_sectors(bio))) |
e83502ca | 1626 | return; |
6678d83f KO |
1627 | |
1628 | size <<= 9; | |
4f024f37 | 1629 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1630 | return; |
1631 | ||
6678d83f | 1632 | bio_advance(bio, offset << 9); |
4f024f37 | 1633 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1634 | |
1635 | if (bio_integrity(bio)) | |
fbd08e76 | 1636 | bio_integrity_trim(bio); |
6678d83f KO |
1637 | } |
1638 | EXPORT_SYMBOL_GPL(bio_trim); | |
1639 | ||
1da177e4 LT |
1640 | /* |
1641 | * create memory pools for biovec's in a bio_set. | |
1642 | * use the global biovec slabs created for general use. | |
1643 | */ | |
8aa6ba2f | 1644 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1645 | { |
7a800a20 | 1646 | struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; |
1da177e4 | 1647 | |
8aa6ba2f | 1648 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1649 | } |
1650 | ||
917a38c7 KO |
1651 | /* |
1652 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1653 | * | |
1654 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1655 | * kzalloc()). | |
1656 | */ | |
1657 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1658 | { |
be4d234d | 1659 | bio_alloc_cache_destroy(bs); |
df2cb6da KO |
1660 | if (bs->rescue_workqueue) |
1661 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1662 | bs->rescue_workqueue = NULL; |
df2cb6da | 1663 | |
8aa6ba2f KO |
1664 | mempool_exit(&bs->bio_pool); |
1665 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1666 | |
7878cba9 | 1667 | bioset_integrity_free(bs); |
917a38c7 KO |
1668 | if (bs->bio_slab) |
1669 | bio_put_slab(bs); | |
1670 | bs->bio_slab = NULL; | |
1671 | } | |
1672 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1673 | |
917a38c7 KO |
1674 | /** |
1675 | * bioset_init - Initialize a bio_set | |
dad08527 | 1676 | * @bs: pool to initialize |
917a38c7 KO |
1677 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1678 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1679 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1680 | * and %BIOSET_NEED_RESCUER | |
1681 | * | |
dad08527 KO |
1682 | * Description: |
1683 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1684 | * to ask for a number of bytes to be allocated in front of the bio. | |
1685 | * Front pad allocation is useful for embedding the bio inside | |
1686 | * another structure, to avoid allocating extra data to go with the bio. | |
1687 | * Note that the bio must be embedded at the END of that structure always, | |
1688 | * or things will break badly. | |
1689 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
abfc426d CH |
1690 | * for allocating iovecs. This pool is not needed e.g. for bio_init_clone(). |
1691 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used | |
1692 | * to dispatch queued requests when the mempool runs out of space. | |
dad08527 | 1693 | * |
917a38c7 KO |
1694 | */ |
1695 | int bioset_init(struct bio_set *bs, | |
1696 | unsigned int pool_size, | |
1697 | unsigned int front_pad, | |
1698 | int flags) | |
1699 | { | |
917a38c7 | 1700 | bs->front_pad = front_pad; |
9f180e31 ML |
1701 | if (flags & BIOSET_NEED_BVECS) |
1702 | bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1703 | else | |
1704 | bs->back_pad = 0; | |
917a38c7 KO |
1705 | |
1706 | spin_lock_init(&bs->rescue_lock); | |
1707 | bio_list_init(&bs->rescue_list); | |
1708 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1709 | ||
49d1ec85 | 1710 | bs->bio_slab = bio_find_or_create_slab(bs); |
917a38c7 KO |
1711 | if (!bs->bio_slab) |
1712 | return -ENOMEM; | |
1713 | ||
1714 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) | |
1715 | goto bad; | |
1716 | ||
1717 | if ((flags & BIOSET_NEED_BVECS) && | |
1718 | biovec_init_pool(&bs->bvec_pool, pool_size)) | |
1719 | goto bad; | |
1720 | ||
be4d234d JA |
1721 | if (flags & BIOSET_NEED_RESCUER) { |
1722 | bs->rescue_workqueue = alloc_workqueue("bioset", | |
1723 | WQ_MEM_RECLAIM, 0); | |
1724 | if (!bs->rescue_workqueue) | |
1725 | goto bad; | |
1726 | } | |
1727 | if (flags & BIOSET_PERCPU_CACHE) { | |
1728 | bs->cache = alloc_percpu(struct bio_alloc_cache); | |
1729 | if (!bs->cache) | |
1730 | goto bad; | |
1731 | cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); | |
1732 | } | |
917a38c7 KO |
1733 | |
1734 | return 0; | |
1735 | bad: | |
1736 | bioset_exit(bs); | |
1737 | return -ENOMEM; | |
1738 | } | |
1739 | EXPORT_SYMBOL(bioset_init); | |
1740 | ||
de76fd89 | 1741 | static int __init init_bio(void) |
1da177e4 LT |
1742 | { |
1743 | int i; | |
1744 | ||
7878cba9 | 1745 | bio_integrity_init(); |
1da177e4 | 1746 | |
de76fd89 CH |
1747 | for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { |
1748 | struct biovec_slab *bvs = bvec_slabs + i; | |
a7fcd37c | 1749 | |
de76fd89 CH |
1750 | bvs->slab = kmem_cache_create(bvs->name, |
1751 | bvs->nr_vecs * sizeof(struct bio_vec), 0, | |
1752 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
1da177e4 | 1753 | } |
1da177e4 | 1754 | |
be4d234d JA |
1755 | cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, |
1756 | bio_cpu_dead); | |
1757 | ||
12c5b70c JA |
1758 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, |
1759 | BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE)) | |
1da177e4 LT |
1760 | panic("bio: can't allocate bios\n"); |
1761 | ||
f4f8154a | 1762 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
1763 | panic("bio: can't create integrity pool\n"); |
1764 | ||
1da177e4 LT |
1765 | return 0; |
1766 | } | |
1da177e4 | 1767 | subsys_initcall(init_bio); |