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> |
1da177e4 | 21 | |
55782138 | 22 | #include <trace/events/block.h> |
9e234eea | 23 | #include "blk.h" |
67b42d0b | 24 | #include "blk-rq-qos.h" |
0bfc2455 | 25 | |
392ddc32 JA |
26 | /* |
27 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
28 | * itself, to shrink a bio data allocation from two mempool calls to one | |
29 | */ | |
30 | #define BIO_INLINE_VECS 4 | |
31 | ||
1da177e4 LT |
32 | /* |
33 | * if you change this list, also change bvec_alloc or things will | |
34 | * break badly! cannot be bigger than what you can fit into an | |
35 | * unsigned short | |
36 | */ | |
bd5c4fac | 37 | #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } |
ed996a52 | 38 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { |
bd5c4fac | 39 | BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), |
1da177e4 LT |
40 | }; |
41 | #undef BV | |
42 | ||
1da177e4 LT |
43 | /* |
44 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
45 | * IO code that does not need private memory pools. | |
46 | */ | |
f4f8154a | 47 | struct bio_set fs_bio_set; |
3f86a82a | 48 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 49 | |
bb799ca0 JA |
50 | /* |
51 | * Our slab pool management | |
52 | */ | |
53 | struct bio_slab { | |
54 | struct kmem_cache *slab; | |
55 | unsigned int slab_ref; | |
56 | unsigned int slab_size; | |
57 | char name[8]; | |
58 | }; | |
59 | static DEFINE_MUTEX(bio_slab_lock); | |
60 | static struct bio_slab *bio_slabs; | |
61 | static unsigned int bio_slab_nr, bio_slab_max; | |
62 | ||
63 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
64 | { | |
65 | unsigned int sz = sizeof(struct bio) + extra_size; | |
66 | struct kmem_cache *slab = NULL; | |
389d7b26 | 67 | struct bio_slab *bslab, *new_bio_slabs; |
386bc35a | 68 | unsigned int new_bio_slab_max; |
bb799ca0 JA |
69 | unsigned int i, entry = -1; |
70 | ||
71 | mutex_lock(&bio_slab_lock); | |
72 | ||
73 | i = 0; | |
74 | while (i < bio_slab_nr) { | |
f06f135d | 75 | bslab = &bio_slabs[i]; |
bb799ca0 JA |
76 | |
77 | if (!bslab->slab && entry == -1) | |
78 | entry = i; | |
79 | else if (bslab->slab_size == sz) { | |
80 | slab = bslab->slab; | |
81 | bslab->slab_ref++; | |
82 | break; | |
83 | } | |
84 | i++; | |
85 | } | |
86 | ||
87 | if (slab) | |
88 | goto out_unlock; | |
89 | ||
90 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
386bc35a | 91 | new_bio_slab_max = bio_slab_max << 1; |
389d7b26 | 92 | new_bio_slabs = krealloc(bio_slabs, |
386bc35a | 93 | new_bio_slab_max * sizeof(struct bio_slab), |
389d7b26 AK |
94 | GFP_KERNEL); |
95 | if (!new_bio_slabs) | |
bb799ca0 | 96 | goto out_unlock; |
386bc35a | 97 | bio_slab_max = new_bio_slab_max; |
389d7b26 | 98 | bio_slabs = new_bio_slabs; |
bb799ca0 JA |
99 | } |
100 | if (entry == -1) | |
101 | entry = bio_slab_nr++; | |
102 | ||
103 | bslab = &bio_slabs[entry]; | |
104 | ||
105 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
6a241483 MP |
106 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
107 | SLAB_HWCACHE_ALIGN, NULL); | |
bb799ca0 JA |
108 | if (!slab) |
109 | goto out_unlock; | |
110 | ||
bb799ca0 JA |
111 | bslab->slab = slab; |
112 | bslab->slab_ref = 1; | |
113 | bslab->slab_size = sz; | |
114 | out_unlock: | |
115 | mutex_unlock(&bio_slab_lock); | |
116 | return slab; | |
117 | } | |
118 | ||
119 | static void bio_put_slab(struct bio_set *bs) | |
120 | { | |
121 | struct bio_slab *bslab = NULL; | |
122 | unsigned int i; | |
123 | ||
124 | mutex_lock(&bio_slab_lock); | |
125 | ||
126 | for (i = 0; i < bio_slab_nr; i++) { | |
127 | if (bs->bio_slab == bio_slabs[i].slab) { | |
128 | bslab = &bio_slabs[i]; | |
129 | break; | |
130 | } | |
131 | } | |
132 | ||
133 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
134 | goto out; | |
135 | ||
136 | WARN_ON(!bslab->slab_ref); | |
137 | ||
138 | if (--bslab->slab_ref) | |
139 | goto out; | |
140 | ||
141 | kmem_cache_destroy(bslab->slab); | |
142 | bslab->slab = NULL; | |
143 | ||
144 | out: | |
145 | mutex_unlock(&bio_slab_lock); | |
146 | } | |
147 | ||
7ba1ba12 MP |
148 | unsigned int bvec_nr_vecs(unsigned short idx) |
149 | { | |
d6c02a9b | 150 | return bvec_slabs[--idx].nr_vecs; |
7ba1ba12 MP |
151 | } |
152 | ||
9f060e22 | 153 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 | 154 | { |
ed996a52 CH |
155 | if (!idx) |
156 | return; | |
157 | idx--; | |
158 | ||
159 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
bb799ca0 | 160 | |
ed996a52 | 161 | if (idx == BVEC_POOL_MAX) { |
9f060e22 | 162 | mempool_free(bv, pool); |
ed996a52 | 163 | } else { |
bb799ca0 JA |
164 | struct biovec_slab *bvs = bvec_slabs + idx; |
165 | ||
166 | kmem_cache_free(bvs->slab, bv); | |
167 | } | |
168 | } | |
169 | ||
9f060e22 KO |
170 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
171 | mempool_t *pool) | |
1da177e4 LT |
172 | { |
173 | struct bio_vec *bvl; | |
1da177e4 | 174 | |
7ff9345f JA |
175 | /* |
176 | * see comment near bvec_array define! | |
177 | */ | |
178 | switch (nr) { | |
179 | case 1: | |
180 | *idx = 0; | |
181 | break; | |
182 | case 2 ... 4: | |
183 | *idx = 1; | |
184 | break; | |
185 | case 5 ... 16: | |
186 | *idx = 2; | |
187 | break; | |
188 | case 17 ... 64: | |
189 | *idx = 3; | |
190 | break; | |
191 | case 65 ... 128: | |
192 | *idx = 4; | |
193 | break; | |
194 | case 129 ... BIO_MAX_PAGES: | |
195 | *idx = 5; | |
196 | break; | |
197 | default: | |
198 | return NULL; | |
199 | } | |
200 | ||
201 | /* | |
202 | * idx now points to the pool we want to allocate from. only the | |
203 | * 1-vec entry pool is mempool backed. | |
204 | */ | |
ed996a52 | 205 | if (*idx == BVEC_POOL_MAX) { |
7ff9345f | 206 | fallback: |
9f060e22 | 207 | bvl = mempool_alloc(pool, gfp_mask); |
7ff9345f JA |
208 | } else { |
209 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
d0164adc | 210 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
7ff9345f | 211 | |
0a0d96b0 | 212 | /* |
7ff9345f JA |
213 | * Make this allocation restricted and don't dump info on |
214 | * allocation failures, since we'll fallback to the mempool | |
215 | * in case of failure. | |
0a0d96b0 | 216 | */ |
7ff9345f | 217 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 218 | |
0a0d96b0 | 219 | /* |
d0164adc | 220 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM |
7ff9345f | 221 | * is set, retry with the 1-entry mempool |
0a0d96b0 | 222 | */ |
7ff9345f | 223 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
d0164adc | 224 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { |
ed996a52 | 225 | *idx = BVEC_POOL_MAX; |
7ff9345f JA |
226 | goto fallback; |
227 | } | |
228 | } | |
229 | ||
ed996a52 | 230 | (*idx)++; |
1da177e4 LT |
231 | return bvl; |
232 | } | |
233 | ||
9ae3b3f5 | 234 | void bio_uninit(struct bio *bio) |
1da177e4 | 235 | { |
6f70fb66 | 236 | bio_disassociate_blkg(bio); |
ece841ab JT |
237 | |
238 | if (bio_integrity(bio)) | |
239 | bio_integrity_free(bio); | |
4254bba1 | 240 | } |
9ae3b3f5 | 241 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 242 | |
4254bba1 KO |
243 | static void bio_free(struct bio *bio) |
244 | { | |
245 | struct bio_set *bs = bio->bi_pool; | |
246 | void *p; | |
247 | ||
9ae3b3f5 | 248 | bio_uninit(bio); |
4254bba1 KO |
249 | |
250 | if (bs) { | |
8aa6ba2f | 251 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
4254bba1 KO |
252 | |
253 | /* | |
254 | * If we have front padding, adjust the bio pointer before freeing | |
255 | */ | |
256 | p = bio; | |
bb799ca0 JA |
257 | p -= bs->front_pad; |
258 | ||
8aa6ba2f | 259 | mempool_free(p, &bs->bio_pool); |
4254bba1 KO |
260 | } else { |
261 | /* Bio was allocated by bio_kmalloc() */ | |
262 | kfree(bio); | |
263 | } | |
3676347a PO |
264 | } |
265 | ||
9ae3b3f5 JA |
266 | /* |
267 | * Users of this function have their own bio allocation. Subsequently, | |
268 | * they must remember to pair any call to bio_init() with bio_uninit() | |
269 | * when IO has completed, or when the bio is released. | |
270 | */ | |
3a83f467 ML |
271 | void bio_init(struct bio *bio, struct bio_vec *table, |
272 | unsigned short max_vecs) | |
1da177e4 | 273 | { |
2b94de55 | 274 | memset(bio, 0, sizeof(*bio)); |
c4cf5261 | 275 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 276 | atomic_set(&bio->__bi_cnt, 1); |
3a83f467 ML |
277 | |
278 | bio->bi_io_vec = table; | |
279 | bio->bi_max_vecs = max_vecs; | |
1da177e4 | 280 | } |
a112a71d | 281 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 282 | |
f44b48c7 KO |
283 | /** |
284 | * bio_reset - reinitialize a bio | |
285 | * @bio: bio to reset | |
286 | * | |
287 | * Description: | |
288 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
289 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
290 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
291 | * comment in struct bio. | |
292 | */ | |
293 | void bio_reset(struct bio *bio) | |
294 | { | |
295 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
296 | ||
9ae3b3f5 | 297 | bio_uninit(bio); |
f44b48c7 KO |
298 | |
299 | memset(bio, 0, BIO_RESET_BYTES); | |
4246a0b6 | 300 | bio->bi_flags = flags; |
c4cf5261 | 301 | atomic_set(&bio->__bi_remaining, 1); |
f44b48c7 KO |
302 | } |
303 | EXPORT_SYMBOL(bio_reset); | |
304 | ||
38f8baae | 305 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 306 | { |
4246a0b6 CH |
307 | struct bio *parent = bio->bi_private; |
308 | ||
4e4cbee9 CH |
309 | if (!parent->bi_status) |
310 | parent->bi_status = bio->bi_status; | |
196d38bc | 311 | bio_put(bio); |
38f8baae CH |
312 | return parent; |
313 | } | |
314 | ||
315 | static void bio_chain_endio(struct bio *bio) | |
316 | { | |
317 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
318 | } |
319 | ||
320 | /** | |
321 | * bio_chain - chain bio completions | |
1051a902 RD |
322 | * @bio: the target bio |
323 | * @parent: the @bio's parent bio | |
196d38bc KO |
324 | * |
325 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
326 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
327 | * completed; the chained bio will also be freed when it completes. | |
328 | * | |
329 | * The caller must not set bi_private or bi_end_io in @bio. | |
330 | */ | |
331 | void bio_chain(struct bio *bio, struct bio *parent) | |
332 | { | |
333 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
334 | ||
335 | bio->bi_private = parent; | |
336 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 337 | bio_inc_remaining(parent); |
196d38bc KO |
338 | } |
339 | EXPORT_SYMBOL(bio_chain); | |
340 | ||
df2cb6da KO |
341 | static void bio_alloc_rescue(struct work_struct *work) |
342 | { | |
343 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
344 | struct bio *bio; | |
345 | ||
346 | while (1) { | |
347 | spin_lock(&bs->rescue_lock); | |
348 | bio = bio_list_pop(&bs->rescue_list); | |
349 | spin_unlock(&bs->rescue_lock); | |
350 | ||
351 | if (!bio) | |
352 | break; | |
353 | ||
354 | generic_make_request(bio); | |
355 | } | |
356 | } | |
357 | ||
358 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
359 | { | |
360 | struct bio_list punt, nopunt; | |
361 | struct bio *bio; | |
362 | ||
47e0fb46 N |
363 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
364 | return; | |
df2cb6da KO |
365 | /* |
366 | * In order to guarantee forward progress we must punt only bios that | |
367 | * were allocated from this bio_set; otherwise, if there was a bio on | |
368 | * there for a stacking driver higher up in the stack, processing it | |
369 | * could require allocating bios from this bio_set, and doing that from | |
370 | * our own rescuer would be bad. | |
371 | * | |
372 | * Since bio lists are singly linked, pop them all instead of trying to | |
373 | * remove from the middle of the list: | |
374 | */ | |
375 | ||
376 | bio_list_init(&punt); | |
377 | bio_list_init(&nopunt); | |
378 | ||
f5fe1b51 | 379 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 380 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 381 | current->bio_list[0] = nopunt; |
df2cb6da | 382 | |
f5fe1b51 N |
383 | bio_list_init(&nopunt); |
384 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
385 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
386 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
387 | |
388 | spin_lock(&bs->rescue_lock); | |
389 | bio_list_merge(&bs->rescue_list, &punt); | |
390 | spin_unlock(&bs->rescue_lock); | |
391 | ||
392 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
393 | } | |
394 | ||
1da177e4 LT |
395 | /** |
396 | * bio_alloc_bioset - allocate a bio for I/O | |
519c8e9f | 397 | * @gfp_mask: the GFP_* mask given to the slab allocator |
1da177e4 | 398 | * @nr_iovecs: number of iovecs to pre-allocate |
db18efac | 399 | * @bs: the bio_set to allocate from. |
1da177e4 LT |
400 | * |
401 | * Description: | |
3f86a82a KO |
402 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
403 | * backed by the @bs's mempool. | |
404 | * | |
d0164adc MG |
405 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will |
406 | * always be able to allocate a bio. This is due to the mempool guarantees. | |
407 | * To make this work, callers must never allocate more than 1 bio at a time | |
408 | * from this pool. Callers that need to allocate more than 1 bio must always | |
409 | * submit the previously allocated bio for IO before attempting to allocate | |
410 | * a new one. Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 411 | * |
df2cb6da KO |
412 | * Note that when running under generic_make_request() (i.e. any block |
413 | * driver), bios are not submitted until after you return - see the code in | |
414 | * generic_make_request() that converts recursion into iteration, to prevent | |
415 | * stack overflows. | |
416 | * | |
417 | * This would normally mean allocating multiple bios under | |
418 | * generic_make_request() would be susceptible to deadlocks, but we have | |
419 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
420 | * thread. | |
421 | * | |
422 | * However, we do not guarantee forward progress for allocations from other | |
423 | * mempools. Doing multiple allocations from the same mempool under | |
424 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
425 | * for per bio allocations. | |
426 | * | |
3f86a82a KO |
427 | * RETURNS: |
428 | * Pointer to new bio on success, NULL on failure. | |
429 | */ | |
7a88fa19 DC |
430 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, |
431 | struct bio_set *bs) | |
1da177e4 | 432 | { |
df2cb6da | 433 | gfp_t saved_gfp = gfp_mask; |
3f86a82a KO |
434 | unsigned front_pad; |
435 | unsigned inline_vecs; | |
34053979 | 436 | struct bio_vec *bvl = NULL; |
451a9ebf TH |
437 | struct bio *bio; |
438 | void *p; | |
439 | ||
3f86a82a KO |
440 | if (!bs) { |
441 | if (nr_iovecs > UIO_MAXIOV) | |
442 | return NULL; | |
443 | ||
444 | p = kmalloc(sizeof(struct bio) + | |
445 | nr_iovecs * sizeof(struct bio_vec), | |
446 | gfp_mask); | |
447 | front_pad = 0; | |
448 | inline_vecs = nr_iovecs; | |
449 | } else { | |
d8f429e1 | 450 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
8aa6ba2f KO |
451 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && |
452 | nr_iovecs > 0)) | |
d8f429e1 | 453 | return NULL; |
df2cb6da KO |
454 | /* |
455 | * generic_make_request() converts recursion to iteration; this | |
456 | * means if we're running beneath it, any bios we allocate and | |
457 | * submit will not be submitted (and thus freed) until after we | |
458 | * return. | |
459 | * | |
460 | * This exposes us to a potential deadlock if we allocate | |
461 | * multiple bios from the same bio_set() while running | |
462 | * underneath generic_make_request(). If we were to allocate | |
463 | * multiple bios (say a stacking block driver that was splitting | |
464 | * bios), we would deadlock if we exhausted the mempool's | |
465 | * reserve. | |
466 | * | |
467 | * We solve this, and guarantee forward progress, with a rescuer | |
468 | * workqueue per bio_set. If we go to allocate and there are | |
469 | * bios on current->bio_list, we first try the allocation | |
d0164adc MG |
470 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those |
471 | * bios we would be blocking to the rescuer workqueue before | |
472 | * we retry with the original gfp_flags. | |
df2cb6da KO |
473 | */ |
474 | ||
f5fe1b51 N |
475 | if (current->bio_list && |
476 | (!bio_list_empty(¤t->bio_list[0]) || | |
47e0fb46 N |
477 | !bio_list_empty(¤t->bio_list[1])) && |
478 | bs->rescue_workqueue) | |
d0164adc | 479 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; |
df2cb6da | 480 | |
8aa6ba2f | 481 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
482 | if (!p && gfp_mask != saved_gfp) { |
483 | punt_bios_to_rescuer(bs); | |
484 | gfp_mask = saved_gfp; | |
8aa6ba2f | 485 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
486 | } |
487 | ||
3f86a82a KO |
488 | front_pad = bs->front_pad; |
489 | inline_vecs = BIO_INLINE_VECS; | |
490 | } | |
491 | ||
451a9ebf TH |
492 | if (unlikely(!p)) |
493 | return NULL; | |
1da177e4 | 494 | |
3f86a82a | 495 | bio = p + front_pad; |
3a83f467 | 496 | bio_init(bio, NULL, 0); |
34053979 | 497 | |
3f86a82a | 498 | if (nr_iovecs > inline_vecs) { |
ed996a52 CH |
499 | unsigned long idx = 0; |
500 | ||
8aa6ba2f | 501 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
502 | if (!bvl && gfp_mask != saved_gfp) { |
503 | punt_bios_to_rescuer(bs); | |
504 | gfp_mask = saved_gfp; | |
8aa6ba2f | 505 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
506 | } |
507 | ||
34053979 IM |
508 | if (unlikely(!bvl)) |
509 | goto err_free; | |
a38352e0 | 510 | |
ed996a52 | 511 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
3f86a82a KO |
512 | } else if (nr_iovecs) { |
513 | bvl = bio->bi_inline_vecs; | |
1da177e4 | 514 | } |
3f86a82a KO |
515 | |
516 | bio->bi_pool = bs; | |
34053979 | 517 | bio->bi_max_vecs = nr_iovecs; |
34053979 | 518 | bio->bi_io_vec = bvl; |
1da177e4 | 519 | return bio; |
34053979 IM |
520 | |
521 | err_free: | |
8aa6ba2f | 522 | mempool_free(p, &bs->bio_pool); |
34053979 | 523 | return NULL; |
1da177e4 | 524 | } |
a112a71d | 525 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 526 | |
38a72dac | 527 | void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) |
1da177e4 LT |
528 | { |
529 | unsigned long flags; | |
7988613b KO |
530 | struct bio_vec bv; |
531 | struct bvec_iter iter; | |
1da177e4 | 532 | |
38a72dac | 533 | __bio_for_each_segment(bv, bio, iter, start) { |
7988613b KO |
534 | char *data = bvec_kmap_irq(&bv, &flags); |
535 | memset(data, 0, bv.bv_len); | |
536 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
537 | bvec_kunmap_irq(data, &flags); |
538 | } | |
539 | } | |
38a72dac | 540 | EXPORT_SYMBOL(zero_fill_bio_iter); |
1da177e4 | 541 | |
83c9c547 ML |
542 | /** |
543 | * bio_truncate - truncate the bio to small size of @new_size | |
544 | * @bio: the bio to be truncated | |
545 | * @new_size: new size for truncating the bio | |
546 | * | |
547 | * Description: | |
548 | * Truncate the bio to new size of @new_size. If bio_op(bio) is | |
549 | * REQ_OP_READ, zero the truncated part. This function should only | |
550 | * be used for handling corner cases, such as bio eod. | |
551 | */ | |
85a8ce62 ML |
552 | void bio_truncate(struct bio *bio, unsigned new_size) |
553 | { | |
554 | struct bio_vec bv; | |
555 | struct bvec_iter iter; | |
556 | unsigned int done = 0; | |
557 | bool truncated = false; | |
558 | ||
559 | if (new_size >= bio->bi_iter.bi_size) | |
560 | return; | |
561 | ||
83c9c547 | 562 | if (bio_op(bio) != REQ_OP_READ) |
85a8ce62 ML |
563 | goto exit; |
564 | ||
565 | bio_for_each_segment(bv, bio, iter) { | |
566 | if (done + bv.bv_len > new_size) { | |
567 | unsigned offset; | |
568 | ||
569 | if (!truncated) | |
570 | offset = new_size - done; | |
571 | else | |
572 | offset = 0; | |
573 | zero_user(bv.bv_page, offset, bv.bv_len - offset); | |
574 | truncated = true; | |
575 | } | |
576 | done += bv.bv_len; | |
577 | } | |
578 | ||
579 | exit: | |
580 | /* | |
581 | * Don't touch bvec table here and make it really immutable, since | |
582 | * fs bio user has to retrieve all pages via bio_for_each_segment_all | |
583 | * in its .end_bio() callback. | |
584 | * | |
585 | * It is enough to truncate bio by updating .bi_size since we can make | |
586 | * correct bvec with the updated .bi_size for drivers. | |
587 | */ | |
588 | bio->bi_iter.bi_size = new_size; | |
589 | } | |
590 | ||
1da177e4 LT |
591 | /** |
592 | * bio_put - release a reference to a bio | |
593 | * @bio: bio to release reference to | |
594 | * | |
595 | * Description: | |
596 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 597 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
598 | **/ |
599 | void bio_put(struct bio *bio) | |
600 | { | |
dac56212 | 601 | if (!bio_flagged(bio, BIO_REFFED)) |
4254bba1 | 602 | bio_free(bio); |
dac56212 JA |
603 | else { |
604 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
605 | ||
606 | /* | |
607 | * last put frees it | |
608 | */ | |
609 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
610 | bio_free(bio); | |
611 | } | |
1da177e4 | 612 | } |
a112a71d | 613 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 614 | |
59d276fe KO |
615 | /** |
616 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
617 | * @bio: destination bio | |
618 | * @bio_src: bio to clone | |
619 | * | |
620 | * Clone a &bio. Caller will own the returned bio, but not | |
621 | * the actual data it points to. Reference count of returned | |
622 | * bio will be one. | |
623 | * | |
624 | * Caller must ensure that @bio_src is not freed before @bio. | |
625 | */ | |
626 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
627 | { | |
ed996a52 | 628 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
59d276fe KO |
629 | |
630 | /* | |
74d46992 | 631 | * most users will be overriding ->bi_disk with a new target, |
59d276fe KO |
632 | * so we don't set nor calculate new physical/hw segment counts here |
633 | */ | |
74d46992 | 634 | bio->bi_disk = bio_src->bi_disk; |
62530ed8 | 635 | bio->bi_partno = bio_src->bi_partno; |
b7c44ed9 | 636 | bio_set_flag(bio, BIO_CLONED); |
111be883 SL |
637 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
638 | bio_set_flag(bio, BIO_THROTTLED); | |
1eff9d32 | 639 | bio->bi_opf = bio_src->bi_opf; |
ca474b73 | 640 | bio->bi_ioprio = bio_src->bi_ioprio; |
cb6934f8 | 641 | bio->bi_write_hint = bio_src->bi_write_hint; |
59d276fe KO |
642 | bio->bi_iter = bio_src->bi_iter; |
643 | bio->bi_io_vec = bio_src->bi_io_vec; | |
20bd723e | 644 | |
db6638d7 | 645 | bio_clone_blkg_association(bio, bio_src); |
e439bedf | 646 | blkcg_bio_issue_init(bio); |
59d276fe KO |
647 | } |
648 | EXPORT_SYMBOL(__bio_clone_fast); | |
649 | ||
650 | /** | |
651 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
652 | * @bio: bio to clone | |
653 | * @gfp_mask: allocation priority | |
654 | * @bs: bio_set to allocate from | |
655 | * | |
656 | * Like __bio_clone_fast, only also allocates the returned bio | |
657 | */ | |
658 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
659 | { | |
660 | struct bio *b; | |
661 | ||
662 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
663 | if (!b) | |
664 | return NULL; | |
665 | ||
666 | __bio_clone_fast(b, bio); | |
667 | ||
668 | if (bio_integrity(bio)) { | |
669 | int ret; | |
670 | ||
671 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
672 | ||
673 | if (ret < 0) { | |
674 | bio_put(b); | |
675 | return NULL; | |
676 | } | |
677 | } | |
678 | ||
679 | return b; | |
680 | } | |
681 | EXPORT_SYMBOL(bio_clone_fast); | |
682 | ||
5cbd28e3 CH |
683 | const char *bio_devname(struct bio *bio, char *buf) |
684 | { | |
685 | return disk_name(bio->bi_disk, bio->bi_partno, buf); | |
686 | } | |
687 | EXPORT_SYMBOL(bio_devname); | |
688 | ||
5919482e ML |
689 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
690 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 691 | bool *same_page) |
5919482e ML |
692 | { |
693 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + | |
694 | bv->bv_offset + bv->bv_len - 1; | |
695 | phys_addr_t page_addr = page_to_phys(page); | |
696 | ||
697 | if (vec_end_addr + 1 != page_addr + off) | |
698 | return false; | |
699 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
700 | return false; | |
52d52d1c | 701 | |
ff896738 CH |
702 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
703 | if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) | |
704 | return false; | |
5919482e ML |
705 | return true; |
706 | } | |
707 | ||
384209cd CH |
708 | static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, |
709 | struct page *page, unsigned len, unsigned offset, | |
710 | bool *same_page) | |
489fbbcb | 711 | { |
384209cd | 712 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
489fbbcb ML |
713 | unsigned long mask = queue_segment_boundary(q); |
714 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
715 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
716 | ||
717 | if ((addr1 | mask) != (addr2 | mask)) | |
718 | return false; | |
489fbbcb ML |
719 | if (bv->bv_len + len > queue_max_segment_size(q)) |
720 | return false; | |
384209cd | 721 | return __bio_try_merge_page(bio, page, len, offset, same_page); |
489fbbcb ML |
722 | } |
723 | ||
1da177e4 | 724 | /** |
19047087 | 725 | * __bio_add_pc_page - attempt to add page to passthrough bio |
c66a14d0 KO |
726 | * @q: the target queue |
727 | * @bio: destination bio | |
728 | * @page: page to add | |
729 | * @len: vec entry length | |
730 | * @offset: vec entry offset | |
d1916c86 | 731 | * @same_page: return if the merge happen inside the same page |
1da177e4 | 732 | * |
c66a14d0 KO |
733 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
734 | * number of reasons, such as the bio being full or target block device | |
735 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
736 | * so it is always possible to add a single page to an empty bio. | |
737 | * | |
5a8ce240 | 738 | * This should only be used by passthrough bios. |
1da177e4 | 739 | */ |
4713839d | 740 | static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, |
19047087 | 741 | struct page *page, unsigned int len, unsigned int offset, |
d1916c86 | 742 | bool *same_page) |
1da177e4 | 743 | { |
1da177e4 LT |
744 | struct bio_vec *bvec; |
745 | ||
746 | /* | |
747 | * cloned bio must not modify vec list | |
748 | */ | |
749 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
750 | return 0; | |
751 | ||
c66a14d0 | 752 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) |
1da177e4 LT |
753 | return 0; |
754 | ||
80cfd548 | 755 | if (bio->bi_vcnt > 0) { |
d1916c86 | 756 | if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) |
384209cd | 757 | return len; |
320ea869 CH |
758 | |
759 | /* | |
760 | * If the queue doesn't support SG gaps and adding this segment | |
761 | * would create a gap, disallow it. | |
762 | */ | |
384209cd | 763 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
320ea869 CH |
764 | if (bvec_gap_to_prev(q, bvec, offset)) |
765 | return 0; | |
80cfd548 JA |
766 | } |
767 | ||
79d08f89 | 768 | if (bio_full(bio, len)) |
1da177e4 LT |
769 | return 0; |
770 | ||
14ccb66b | 771 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
772 | return 0; |
773 | ||
fcbf6a08 ML |
774 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
775 | bvec->bv_page = page; | |
776 | bvec->bv_len = len; | |
777 | bvec->bv_offset = offset; | |
778 | bio->bi_vcnt++; | |
dcdca753 | 779 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
780 | return len; |
781 | } | |
19047087 ML |
782 | |
783 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, | |
784 | struct page *page, unsigned int len, unsigned int offset) | |
785 | { | |
d1916c86 CH |
786 | bool same_page = false; |
787 | return __bio_add_pc_page(q, bio, page, len, offset, &same_page); | |
19047087 | 788 | } |
a112a71d | 789 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 790 | |
1da177e4 | 791 | /** |
0aa69fd3 CH |
792 | * __bio_try_merge_page - try appending data to an existing bvec. |
793 | * @bio: destination bio | |
551879a4 | 794 | * @page: start page to add |
0aa69fd3 | 795 | * @len: length of the data to add |
551879a4 | 796 | * @off: offset of the data relative to @page |
ff896738 | 797 | * @same_page: return if the segment has been merged inside the same page |
1da177e4 | 798 | * |
0aa69fd3 CH |
799 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
800 | * a useful optimisation for file systems with a block size smaller than the | |
801 | * page size. | |
802 | * | |
551879a4 ML |
803 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
804 | * | |
0aa69fd3 | 805 | * Return %true on success or %false on failure. |
1da177e4 | 806 | */ |
0aa69fd3 | 807 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
ff896738 | 808 | unsigned int len, unsigned int off, bool *same_page) |
1da177e4 | 809 | { |
c66a14d0 | 810 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
0aa69fd3 | 811 | return false; |
762380ad | 812 | |
cc90bc68 | 813 | if (bio->bi_vcnt > 0) { |
0aa69fd3 | 814 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
5919482e ML |
815 | |
816 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
cc90bc68 AG |
817 | if (bio->bi_iter.bi_size > UINT_MAX - len) |
818 | return false; | |
5919482e ML |
819 | bv->bv_len += len; |
820 | bio->bi_iter.bi_size += len; | |
821 | return true; | |
822 | } | |
c66a14d0 | 823 | } |
0aa69fd3 CH |
824 | return false; |
825 | } | |
826 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); | |
c66a14d0 | 827 | |
0aa69fd3 | 828 | /** |
551879a4 | 829 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 830 | * @bio: destination bio |
551879a4 ML |
831 | * @page: start page to add |
832 | * @len: length of the data to add, may cross pages | |
833 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
834 | * |
835 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
836 | * that @bio has space for another bvec. | |
837 | */ | |
838 | void __bio_add_page(struct bio *bio, struct page *page, | |
839 | unsigned int len, unsigned int off) | |
840 | { | |
841 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 842 | |
0aa69fd3 | 843 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 844 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
845 | |
846 | bv->bv_page = page; | |
847 | bv->bv_offset = off; | |
848 | bv->bv_len = len; | |
c66a14d0 | 849 | |
c66a14d0 | 850 | bio->bi_iter.bi_size += len; |
0aa69fd3 | 851 | bio->bi_vcnt++; |
b8e24a93 JW |
852 | |
853 | if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) | |
854 | bio_set_flag(bio, BIO_WORKINGSET); | |
0aa69fd3 CH |
855 | } |
856 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
857 | ||
858 | /** | |
551879a4 | 859 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 860 | * @bio: destination bio |
551879a4 ML |
861 | * @page: start page to add |
862 | * @len: vec entry length, may cross pages | |
863 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 864 | * |
551879a4 | 865 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
866 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
867 | */ | |
868 | int bio_add_page(struct bio *bio, struct page *page, | |
869 | unsigned int len, unsigned int offset) | |
870 | { | |
ff896738 CH |
871 | bool same_page = false; |
872 | ||
873 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 874 | if (bio_full(bio, len)) |
0aa69fd3 CH |
875 | return 0; |
876 | __bio_add_page(bio, page, len, offset); | |
877 | } | |
c66a14d0 | 878 | return len; |
1da177e4 | 879 | } |
a112a71d | 880 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 881 | |
d241a95f | 882 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
883 | { |
884 | struct bvec_iter_all iter_all; | |
885 | struct bio_vec *bvec; | |
7321ecbf | 886 | |
b2d0d991 CH |
887 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
888 | return; | |
889 | ||
d241a95f CH |
890 | bio_for_each_segment_all(bvec, bio, iter_all) { |
891 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
892 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 893 | put_page(bvec->bv_page); |
d241a95f | 894 | } |
7321ecbf CH |
895 | } |
896 | ||
6d0c48ae JA |
897 | static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) |
898 | { | |
899 | const struct bio_vec *bv = iter->bvec; | |
900 | unsigned int len; | |
901 | size_t size; | |
902 | ||
903 | if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) | |
904 | return -EINVAL; | |
905 | ||
906 | len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); | |
907 | size = bio_add_page(bio, bv->bv_page, len, | |
908 | bv->bv_offset + iter->iov_offset); | |
a10584c3 CH |
909 | if (unlikely(size != len)) |
910 | return -EINVAL; | |
a10584c3 CH |
911 | iov_iter_advance(iter, size); |
912 | return 0; | |
6d0c48ae JA |
913 | } |
914 | ||
576ed913 CH |
915 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
916 | ||
2cefe4db | 917 | /** |
17d51b10 | 918 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
919 | * @bio: bio to add pages to |
920 | * @iter: iov iterator describing the region to be mapped | |
921 | * | |
17d51b10 | 922 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 923 | * pages will have to be released using put_page() when done. |
17d51b10 MW |
924 | * For multi-segment *iter, this function only adds pages from the |
925 | * the next non-empty segment of the iov iterator. | |
2cefe4db | 926 | */ |
17d51b10 | 927 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 928 | { |
576ed913 CH |
929 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
930 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
931 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
932 | struct page **pages = (struct page **)bv; | |
45691804 | 933 | bool same_page = false; |
576ed913 CH |
934 | ssize_t size, left; |
935 | unsigned len, i; | |
b403ea24 | 936 | size_t offset; |
576ed913 CH |
937 | |
938 | /* | |
939 | * Move page array up in the allocated memory for the bio vecs as far as | |
940 | * possible so that we can start filling biovecs from the beginning | |
941 | * without overwriting the temporary page array. | |
942 | */ | |
943 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
944 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db KO |
945 | |
946 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
947 | if (unlikely(size <= 0)) | |
948 | return size ? size : -EFAULT; | |
2cefe4db | 949 | |
576ed913 CH |
950 | for (left = size, i = 0; left > 0; left -= len, i++) { |
951 | struct page *page = pages[i]; | |
2cefe4db | 952 | |
576ed913 | 953 | len = min_t(size_t, PAGE_SIZE - offset, left); |
45691804 CH |
954 | |
955 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
956 | if (same_page) | |
957 | put_page(page); | |
958 | } else { | |
79d08f89 | 959 | if (WARN_ON_ONCE(bio_full(bio, len))) |
45691804 CH |
960 | return -EINVAL; |
961 | __bio_add_page(bio, page, len, offset); | |
962 | } | |
576ed913 | 963 | offset = 0; |
2cefe4db KO |
964 | } |
965 | ||
2cefe4db KO |
966 | iov_iter_advance(iter, size); |
967 | return 0; | |
968 | } | |
17d51b10 MW |
969 | |
970 | /** | |
6d0c48ae | 971 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 972 | * @bio: bio to add pages to |
6d0c48ae JA |
973 | * @iter: iov iterator describing the region to be added |
974 | * | |
975 | * This takes either an iterator pointing to user memory, or one pointing to | |
976 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
977 | * map them into the kernel. On IO completion, the caller should put those | |
399254aa JA |
978 | * pages. If we're adding kernel pages, and the caller told us it's safe to |
979 | * do so, we just have to add the pages to the bio directly. We don't grab an | |
980 | * extra reference to those pages (the user should already have that), and we | |
981 | * don't put the page on IO completion. The caller needs to check if the bio is | |
982 | * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be | |
983 | * released. | |
17d51b10 | 984 | * |
17d51b10 | 985 | * The function tries, but does not guarantee, to pin as many pages as |
6d0c48ae JA |
986 | * fit into the bio, or are requested in *iter, whatever is smaller. If |
987 | * MM encounters an error pinning the requested pages, it stops. Error | |
988 | * is returned only if 0 pages could be pinned. | |
17d51b10 MW |
989 | */ |
990 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
991 | { | |
6d0c48ae | 992 | const bool is_bvec = iov_iter_is_bvec(iter); |
14eacf12 CH |
993 | int ret; |
994 | ||
995 | if (WARN_ON_ONCE(bio->bi_vcnt)) | |
996 | return -EINVAL; | |
17d51b10 MW |
997 | |
998 | do { | |
6d0c48ae JA |
999 | if (is_bvec) |
1000 | ret = __bio_iov_bvec_add_pages(bio, iter); | |
1001 | else | |
1002 | ret = __bio_iov_iter_get_pages(bio, iter); | |
79d08f89 | 1003 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
17d51b10 | 1004 | |
b6207430 | 1005 | if (is_bvec) |
7321ecbf | 1006 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
14eacf12 | 1007 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 1008 | } |
2cefe4db | 1009 | |
4246a0b6 | 1010 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 1011 | { |
65e53aab | 1012 | complete(bio->bi_private); |
9e882242 KO |
1013 | } |
1014 | ||
1015 | /** | |
1016 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
1017 | * @bio: The &struct bio which describes the I/O |
1018 | * | |
1019 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
1020 | * bio_endio() on failure. | |
3d289d68 JK |
1021 | * |
1022 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
1023 | * result in bio reference to be consumed. The caller must drop the reference | |
1024 | * on his own. | |
9e882242 | 1025 | */ |
4e49ea4a | 1026 | int submit_bio_wait(struct bio *bio) |
9e882242 | 1027 | { |
e319e1fb | 1028 | DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); |
de6a78b6 | 1029 | unsigned long hang_check; |
9e882242 | 1030 | |
65e53aab | 1031 | bio->bi_private = &done; |
9e882242 | 1032 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 1033 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 1034 | submit_bio(bio); |
de6a78b6 ML |
1035 | |
1036 | /* Prevent hang_check timer from firing at us during very long I/O */ | |
1037 | hang_check = sysctl_hung_task_timeout_secs; | |
1038 | if (hang_check) | |
1039 | while (!wait_for_completion_io_timeout(&done, | |
1040 | hang_check * (HZ/2))) | |
1041 | ; | |
1042 | else | |
1043 | wait_for_completion_io(&done); | |
9e882242 | 1044 | |
65e53aab | 1045 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
1046 | } |
1047 | EXPORT_SYMBOL(submit_bio_wait); | |
1048 | ||
054bdf64 KO |
1049 | /** |
1050 | * bio_advance - increment/complete a bio by some number of bytes | |
1051 | * @bio: bio to advance | |
1052 | * @bytes: number of bytes to complete | |
1053 | * | |
1054 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
1055 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
1056 | * be updated on the last bvec as well. | |
1057 | * | |
1058 | * @bio will then represent the remaining, uncompleted portion of the io. | |
1059 | */ | |
1060 | void bio_advance(struct bio *bio, unsigned bytes) | |
1061 | { | |
1062 | if (bio_integrity(bio)) | |
1063 | bio_integrity_advance(bio, bytes); | |
1064 | ||
4550dd6c | 1065 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
1066 | } |
1067 | EXPORT_SYMBOL(bio_advance); | |
1068 | ||
45db54d5 KO |
1069 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1070 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1071 | { |
1cb9dda4 | 1072 | struct bio_vec src_bv, dst_bv; |
16ac3d63 | 1073 | void *src_p, *dst_p; |
1cb9dda4 | 1074 | unsigned bytes; |
16ac3d63 | 1075 | |
45db54d5 KO |
1076 | while (src_iter->bi_size && dst_iter->bi_size) { |
1077 | src_bv = bio_iter_iovec(src, *src_iter); | |
1078 | dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1cb9dda4 KO |
1079 | |
1080 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1081 | |
1cb9dda4 KO |
1082 | src_p = kmap_atomic(src_bv.bv_page); |
1083 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1084 | |
1cb9dda4 KO |
1085 | memcpy(dst_p + dst_bv.bv_offset, |
1086 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1087 | bytes); |
1088 | ||
1089 | kunmap_atomic(dst_p); | |
1090 | kunmap_atomic(src_p); | |
1091 | ||
6e6e811d KO |
1092 | flush_dcache_page(dst_bv.bv_page); |
1093 | ||
45db54d5 KO |
1094 | bio_advance_iter(src, src_iter, bytes); |
1095 | bio_advance_iter(dst, dst_iter, bytes); | |
16ac3d63 KO |
1096 | } |
1097 | } | |
38a72dac KO |
1098 | EXPORT_SYMBOL(bio_copy_data_iter); |
1099 | ||
1100 | /** | |
45db54d5 KO |
1101 | * bio_copy_data - copy contents of data buffers from one bio to another |
1102 | * @src: source bio | |
1103 | * @dst: destination bio | |
38a72dac KO |
1104 | * |
1105 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1106 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1107 | */ | |
1108 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1109 | { | |
45db54d5 KO |
1110 | struct bvec_iter src_iter = src->bi_iter; |
1111 | struct bvec_iter dst_iter = dst->bi_iter; | |
1112 | ||
1113 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1114 | } |
16ac3d63 KO |
1115 | EXPORT_SYMBOL(bio_copy_data); |
1116 | ||
45db54d5 KO |
1117 | /** |
1118 | * bio_list_copy_data - copy contents of data buffers from one chain of bios to | |
1119 | * another | |
1120 | * @src: source bio list | |
1121 | * @dst: destination bio list | |
1122 | * | |
1123 | * Stops when it reaches the end of either the @src list or @dst list - that is, | |
1124 | * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of | |
1125 | * bios). | |
1126 | */ | |
1127 | void bio_list_copy_data(struct bio *dst, struct bio *src) | |
1128 | { | |
1129 | struct bvec_iter src_iter = src->bi_iter; | |
1130 | struct bvec_iter dst_iter = dst->bi_iter; | |
1131 | ||
1132 | while (1) { | |
1133 | if (!src_iter.bi_size) { | |
1134 | src = src->bi_next; | |
1135 | if (!src) | |
1136 | break; | |
1137 | ||
1138 | src_iter = src->bi_iter; | |
1139 | } | |
1140 | ||
1141 | if (!dst_iter.bi_size) { | |
1142 | dst = dst->bi_next; | |
1143 | if (!dst) | |
1144 | break; | |
1145 | ||
1146 | dst_iter = dst->bi_iter; | |
1147 | } | |
1148 | ||
1149 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
1150 | } | |
1151 | } | |
1152 | EXPORT_SYMBOL(bio_list_copy_data); | |
1153 | ||
1da177e4 | 1154 | struct bio_map_data { |
152e283f | 1155 | int is_our_pages; |
26e49cfc KO |
1156 | struct iov_iter iter; |
1157 | struct iovec iov[]; | |
1da177e4 LT |
1158 | }; |
1159 | ||
0e5b935d | 1160 | static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, |
76029ff3 | 1161 | gfp_t gfp_mask) |
1da177e4 | 1162 | { |
0e5b935d AV |
1163 | struct bio_map_data *bmd; |
1164 | if (data->nr_segs > UIO_MAXIOV) | |
f3f63c1c | 1165 | return NULL; |
1da177e4 | 1166 | |
f1f8f292 | 1167 | bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); |
0e5b935d AV |
1168 | if (!bmd) |
1169 | return NULL; | |
1170 | memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); | |
1171 | bmd->iter = *data; | |
1172 | bmd->iter.iov = bmd->iov; | |
1173 | return bmd; | |
1da177e4 LT |
1174 | } |
1175 | ||
9124d3fe DP |
1176 | /** |
1177 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
1178 | * @bio: The &struct bio which describes the I/O as destination | |
1179 | * @iter: iov_iter as source | |
1180 | * | |
1181 | * Copy all pages from iov_iter to bio. | |
1182 | * Returns 0 on success, or error on failure. | |
1183 | */ | |
98a09d61 | 1184 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) |
c5dec1c3 | 1185 | { |
c5dec1c3 | 1186 | struct bio_vec *bvec; |
6dc4f100 | 1187 | struct bvec_iter_all iter_all; |
c5dec1c3 | 1188 | |
2b070cfe | 1189 | bio_for_each_segment_all(bvec, bio, iter_all) { |
9124d3fe | 1190 | ssize_t ret; |
c5dec1c3 | 1191 | |
9124d3fe DP |
1192 | ret = copy_page_from_iter(bvec->bv_page, |
1193 | bvec->bv_offset, | |
1194 | bvec->bv_len, | |
98a09d61 | 1195 | iter); |
9124d3fe | 1196 | |
98a09d61 | 1197 | if (!iov_iter_count(iter)) |
9124d3fe DP |
1198 | break; |
1199 | ||
1200 | if (ret < bvec->bv_len) | |
1201 | return -EFAULT; | |
c5dec1c3 FT |
1202 | } |
1203 | ||
9124d3fe DP |
1204 | return 0; |
1205 | } | |
1206 | ||
1207 | /** | |
1208 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
1209 | * @bio: The &struct bio which describes the I/O as source | |
1210 | * @iter: iov_iter as destination | |
1211 | * | |
1212 | * Copy all pages from bio to iov_iter. | |
1213 | * Returns 0 on success, or error on failure. | |
1214 | */ | |
1215 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
1216 | { | |
9124d3fe | 1217 | struct bio_vec *bvec; |
6dc4f100 | 1218 | struct bvec_iter_all iter_all; |
9124d3fe | 1219 | |
2b070cfe | 1220 | bio_for_each_segment_all(bvec, bio, iter_all) { |
9124d3fe DP |
1221 | ssize_t ret; |
1222 | ||
1223 | ret = copy_page_to_iter(bvec->bv_page, | |
1224 | bvec->bv_offset, | |
1225 | bvec->bv_len, | |
1226 | &iter); | |
1227 | ||
1228 | if (!iov_iter_count(&iter)) | |
1229 | break; | |
1230 | ||
1231 | if (ret < bvec->bv_len) | |
1232 | return -EFAULT; | |
1233 | } | |
1234 | ||
1235 | return 0; | |
c5dec1c3 FT |
1236 | } |
1237 | ||
491221f8 | 1238 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1239 | { |
1240 | struct bio_vec *bvec; | |
6dc4f100 | 1241 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1242 | |
2b070cfe | 1243 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1244 | __free_page(bvec->bv_page); |
1245 | } | |
491221f8 | 1246 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1247 | |
1da177e4 LT |
1248 | /** |
1249 | * bio_uncopy_user - finish previously mapped bio | |
1250 | * @bio: bio being terminated | |
1251 | * | |
ddad8dd0 | 1252 | * Free pages allocated from bio_copy_user_iov() and write back data |
1da177e4 LT |
1253 | * to user space in case of a read. |
1254 | */ | |
1255 | int bio_uncopy_user(struct bio *bio) | |
1256 | { | |
1257 | struct bio_map_data *bmd = bio->bi_private; | |
1dfa0f68 | 1258 | int ret = 0; |
1da177e4 | 1259 | |
35dc2483 RD |
1260 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { |
1261 | /* | |
1262 | * if we're in a workqueue, the request is orphaned, so | |
2d99b55d HR |
1263 | * don't copy into a random user address space, just free |
1264 | * and return -EINTR so user space doesn't expect any data. | |
35dc2483 | 1265 | */ |
2d99b55d HR |
1266 | if (!current->mm) |
1267 | ret = -EINTR; | |
1268 | else if (bio_data_dir(bio) == READ) | |
9124d3fe | 1269 | ret = bio_copy_to_iter(bio, bmd->iter); |
1dfa0f68 CH |
1270 | if (bmd->is_our_pages) |
1271 | bio_free_pages(bio); | |
35dc2483 | 1272 | } |
c8db4448 | 1273 | kfree(bmd); |
1da177e4 LT |
1274 | bio_put(bio); |
1275 | return ret; | |
1276 | } | |
1277 | ||
1278 | /** | |
c5dec1c3 | 1279 | * bio_copy_user_iov - copy user data to bio |
26e49cfc KO |
1280 | * @q: destination block queue |
1281 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1282 | * @iter: iovec iterator | |
1283 | * @gfp_mask: memory allocation flags | |
1da177e4 LT |
1284 | * |
1285 | * Prepares and returns a bio for indirect user io, bouncing data | |
1286 | * to/from kernel pages as necessary. Must be paired with | |
1287 | * call bio_uncopy_user() on io completion. | |
1288 | */ | |
152e283f FT |
1289 | struct bio *bio_copy_user_iov(struct request_queue *q, |
1290 | struct rq_map_data *map_data, | |
e81cef5d | 1291 | struct iov_iter *iter, |
26e49cfc | 1292 | gfp_t gfp_mask) |
1da177e4 | 1293 | { |
1da177e4 | 1294 | struct bio_map_data *bmd; |
1da177e4 LT |
1295 | struct page *page; |
1296 | struct bio *bio; | |
d16d44eb AV |
1297 | int i = 0, ret; |
1298 | int nr_pages; | |
26e49cfc | 1299 | unsigned int len = iter->count; |
bd5cecea | 1300 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; |
1da177e4 | 1301 | |
0e5b935d | 1302 | bmd = bio_alloc_map_data(iter, gfp_mask); |
1da177e4 LT |
1303 | if (!bmd) |
1304 | return ERR_PTR(-ENOMEM); | |
1305 | ||
26e49cfc KO |
1306 | /* |
1307 | * We need to do a deep copy of the iov_iter including the iovecs. | |
1308 | * The caller provided iov might point to an on-stack or otherwise | |
1309 | * shortlived one. | |
1310 | */ | |
1311 | bmd->is_our_pages = map_data ? 0 : 1; | |
26e49cfc | 1312 | |
d16d44eb AV |
1313 | nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); |
1314 | if (nr_pages > BIO_MAX_PAGES) | |
1315 | nr_pages = BIO_MAX_PAGES; | |
26e49cfc | 1316 | |
1da177e4 | 1317 | ret = -ENOMEM; |
a9e9dc24 | 1318 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1319 | if (!bio) |
1320 | goto out_bmd; | |
1321 | ||
1da177e4 | 1322 | ret = 0; |
56c451f4 FT |
1323 | |
1324 | if (map_data) { | |
e623ddb4 | 1325 | nr_pages = 1 << map_data->page_order; |
56c451f4 FT |
1326 | i = map_data->offset / PAGE_SIZE; |
1327 | } | |
1da177e4 | 1328 | while (len) { |
e623ddb4 | 1329 | unsigned int bytes = PAGE_SIZE; |
1da177e4 | 1330 | |
56c451f4 FT |
1331 | bytes -= offset; |
1332 | ||
1da177e4 LT |
1333 | if (bytes > len) |
1334 | bytes = len; | |
1335 | ||
152e283f | 1336 | if (map_data) { |
e623ddb4 | 1337 | if (i == map_data->nr_entries * nr_pages) { |
152e283f FT |
1338 | ret = -ENOMEM; |
1339 | break; | |
1340 | } | |
e623ddb4 FT |
1341 | |
1342 | page = map_data->pages[i / nr_pages]; | |
1343 | page += (i % nr_pages); | |
1344 | ||
1345 | i++; | |
1346 | } else { | |
152e283f | 1347 | page = alloc_page(q->bounce_gfp | gfp_mask); |
e623ddb4 FT |
1348 | if (!page) { |
1349 | ret = -ENOMEM; | |
1350 | break; | |
1351 | } | |
1da177e4 LT |
1352 | } |
1353 | ||
a3761c3c JG |
1354 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { |
1355 | if (!map_data) | |
1356 | __free_page(page); | |
1da177e4 | 1357 | break; |
a3761c3c | 1358 | } |
1da177e4 LT |
1359 | |
1360 | len -= bytes; | |
56c451f4 | 1361 | offset = 0; |
1da177e4 LT |
1362 | } |
1363 | ||
1364 | if (ret) | |
1365 | goto cleanup; | |
1366 | ||
2884d0be AV |
1367 | if (map_data) |
1368 | map_data->offset += bio->bi_iter.bi_size; | |
1369 | ||
1da177e4 LT |
1370 | /* |
1371 | * success | |
1372 | */ | |
00e23707 | 1373 | if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || |
ecb554a8 | 1374 | (map_data && map_data->from_user)) { |
98a09d61 | 1375 | ret = bio_copy_from_iter(bio, iter); |
c5dec1c3 FT |
1376 | if (ret) |
1377 | goto cleanup; | |
98a09d61 | 1378 | } else { |
f55adad6 KB |
1379 | if (bmd->is_our_pages) |
1380 | zero_fill_bio(bio); | |
98a09d61 | 1381 | iov_iter_advance(iter, bio->bi_iter.bi_size); |
1da177e4 LT |
1382 | } |
1383 | ||
26e49cfc | 1384 | bio->bi_private = bmd; |
2884d0be AV |
1385 | if (map_data && map_data->null_mapped) |
1386 | bio_set_flag(bio, BIO_NULL_MAPPED); | |
1da177e4 LT |
1387 | return bio; |
1388 | cleanup: | |
152e283f | 1389 | if (!map_data) |
1dfa0f68 | 1390 | bio_free_pages(bio); |
1da177e4 LT |
1391 | bio_put(bio); |
1392 | out_bmd: | |
c8db4448 | 1393 | kfree(bmd); |
1da177e4 LT |
1394 | return ERR_PTR(ret); |
1395 | } | |
1396 | ||
37f19e57 CH |
1397 | /** |
1398 | * bio_map_user_iov - map user iovec into bio | |
1399 | * @q: the struct request_queue for the bio | |
1400 | * @iter: iovec iterator | |
1401 | * @gfp_mask: memory allocation flags | |
1402 | * | |
1403 | * Map the user space address into a bio suitable for io to a block | |
1404 | * device. Returns an error pointer in case of error. | |
1405 | */ | |
1406 | struct bio *bio_map_user_iov(struct request_queue *q, | |
e81cef5d | 1407 | struct iov_iter *iter, |
37f19e57 | 1408 | gfp_t gfp_mask) |
1da177e4 | 1409 | { |
26e49cfc | 1410 | int j; |
1da177e4 | 1411 | struct bio *bio; |
076098e5 | 1412 | int ret; |
1da177e4 | 1413 | |
b282cc76 | 1414 | if (!iov_iter_count(iter)) |
1da177e4 LT |
1415 | return ERR_PTR(-EINVAL); |
1416 | ||
b282cc76 | 1417 | bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); |
1da177e4 LT |
1418 | if (!bio) |
1419 | return ERR_PTR(-ENOMEM); | |
1420 | ||
0a0f1513 | 1421 | while (iov_iter_count(iter)) { |
629e42bc | 1422 | struct page **pages; |
076098e5 AV |
1423 | ssize_t bytes; |
1424 | size_t offs, added = 0; | |
1425 | int npages; | |
1da177e4 | 1426 | |
0a0f1513 | 1427 | bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); |
076098e5 AV |
1428 | if (unlikely(bytes <= 0)) { |
1429 | ret = bytes ? bytes : -EFAULT; | |
f1970baf | 1430 | goto out_unmap; |
99172157 | 1431 | } |
f1970baf | 1432 | |
076098e5 | 1433 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); |
f1970baf | 1434 | |
98f0bc99 AV |
1435 | if (unlikely(offs & queue_dma_alignment(q))) { |
1436 | ret = -EINVAL; | |
1437 | j = 0; | |
1438 | } else { | |
1439 | for (j = 0; j < npages; j++) { | |
1440 | struct page *page = pages[j]; | |
1441 | unsigned int n = PAGE_SIZE - offs; | |
d1916c86 | 1442 | bool same_page = false; |
f1970baf | 1443 | |
98f0bc99 AV |
1444 | if (n > bytes) |
1445 | n = bytes; | |
95d78c28 | 1446 | |
19047087 | 1447 | if (!__bio_add_pc_page(q, bio, page, n, offs, |
d1916c86 CH |
1448 | &same_page)) { |
1449 | if (same_page) | |
1450 | put_page(page); | |
98f0bc99 | 1451 | break; |
d1916c86 | 1452 | } |
1da177e4 | 1453 | |
98f0bc99 AV |
1454 | added += n; |
1455 | bytes -= n; | |
1456 | offs = 0; | |
1457 | } | |
0a0f1513 | 1458 | iov_iter_advance(iter, added); |
f1970baf | 1459 | } |
1da177e4 | 1460 | /* |
f1970baf | 1461 | * release the pages we didn't map into the bio, if any |
1da177e4 | 1462 | */ |
629e42bc | 1463 | while (j < npages) |
09cbfeaf | 1464 | put_page(pages[j++]); |
629e42bc | 1465 | kvfree(pages); |
e2e115d1 AV |
1466 | /* couldn't stuff something into bio? */ |
1467 | if (bytes) | |
1468 | break; | |
1da177e4 LT |
1469 | } |
1470 | ||
b7c44ed9 | 1471 | bio_set_flag(bio, BIO_USER_MAPPED); |
37f19e57 CH |
1472 | |
1473 | /* | |
5fad1b64 | 1474 | * subtle -- if bio_map_user_iov() ended up bouncing a bio, |
37f19e57 CH |
1475 | * it would normally disappear when its bi_end_io is run. |
1476 | * however, we need it for the unmap, so grab an extra | |
1477 | * reference to it | |
1478 | */ | |
1479 | bio_get(bio); | |
1da177e4 | 1480 | return bio; |
f1970baf JB |
1481 | |
1482 | out_unmap: | |
506e0798 | 1483 | bio_release_pages(bio, false); |
1da177e4 LT |
1484 | bio_put(bio); |
1485 | return ERR_PTR(ret); | |
1486 | } | |
1487 | ||
1da177e4 LT |
1488 | /** |
1489 | * bio_unmap_user - unmap a bio | |
1490 | * @bio: the bio being unmapped | |
1491 | * | |
5fad1b64 BVA |
1492 | * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from |
1493 | * process context. | |
1da177e4 LT |
1494 | * |
1495 | * bio_unmap_user() may sleep. | |
1496 | */ | |
1497 | void bio_unmap_user(struct bio *bio) | |
1498 | { | |
163cc2d3 CH |
1499 | bio_release_pages(bio, bio_data_dir(bio) == READ); |
1500 | bio_put(bio); | |
1da177e4 LT |
1501 | bio_put(bio); |
1502 | } | |
1503 | ||
b4c5875d DLM |
1504 | static void bio_invalidate_vmalloc_pages(struct bio *bio) |
1505 | { | |
1506 | #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE | |
1507 | if (bio->bi_private && !op_is_write(bio_op(bio))) { | |
1508 | unsigned long i, len = 0; | |
1509 | ||
1510 | for (i = 0; i < bio->bi_vcnt; i++) | |
1511 | len += bio->bi_io_vec[i].bv_len; | |
1512 | invalidate_kernel_vmap_range(bio->bi_private, len); | |
1513 | } | |
1514 | #endif | |
1515 | } | |
1516 | ||
4246a0b6 | 1517 | static void bio_map_kern_endio(struct bio *bio) |
b823825e | 1518 | { |
b4c5875d | 1519 | bio_invalidate_vmalloc_pages(bio); |
b823825e | 1520 | bio_put(bio); |
b823825e JA |
1521 | } |
1522 | ||
75c72b83 CH |
1523 | /** |
1524 | * bio_map_kern - map kernel address into bio | |
1525 | * @q: the struct request_queue for the bio | |
1526 | * @data: pointer to buffer to map | |
1527 | * @len: length in bytes | |
1528 | * @gfp_mask: allocation flags for bio allocation | |
1529 | * | |
1530 | * Map the kernel address into a bio suitable for io to a block | |
1531 | * device. Returns an error pointer in case of error. | |
1532 | */ | |
1533 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1534 | gfp_t gfp_mask) | |
df46b9a4 MC |
1535 | { |
1536 | unsigned long kaddr = (unsigned long)data; | |
1537 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1538 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1539 | const int nr_pages = end - start; | |
b4c5875d DLM |
1540 | bool is_vmalloc = is_vmalloc_addr(data); |
1541 | struct page *page; | |
df46b9a4 MC |
1542 | int offset, i; |
1543 | struct bio *bio; | |
1544 | ||
a9e9dc24 | 1545 | bio = bio_kmalloc(gfp_mask, nr_pages); |
df46b9a4 MC |
1546 | if (!bio) |
1547 | return ERR_PTR(-ENOMEM); | |
1548 | ||
b4c5875d DLM |
1549 | if (is_vmalloc) { |
1550 | flush_kernel_vmap_range(data, len); | |
1551 | bio->bi_private = data; | |
1552 | } | |
1553 | ||
df46b9a4 MC |
1554 | offset = offset_in_page(kaddr); |
1555 | for (i = 0; i < nr_pages; i++) { | |
1556 | unsigned int bytes = PAGE_SIZE - offset; | |
1557 | ||
1558 | if (len <= 0) | |
1559 | break; | |
1560 | ||
1561 | if (bytes > len) | |
1562 | bytes = len; | |
1563 | ||
b4c5875d DLM |
1564 | if (!is_vmalloc) |
1565 | page = virt_to_page(data); | |
1566 | else | |
1567 | page = vmalloc_to_page(data); | |
1568 | if (bio_add_pc_page(q, bio, page, bytes, | |
75c72b83 CH |
1569 | offset) < bytes) { |
1570 | /* we don't support partial mappings */ | |
1571 | bio_put(bio); | |
1572 | return ERR_PTR(-EINVAL); | |
1573 | } | |
df46b9a4 MC |
1574 | |
1575 | data += bytes; | |
1576 | len -= bytes; | |
1577 | offset = 0; | |
1578 | } | |
1579 | ||
b823825e | 1580 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1581 | return bio; |
1582 | } | |
df46b9a4 | 1583 | |
4246a0b6 | 1584 | static void bio_copy_kern_endio(struct bio *bio) |
68154e90 | 1585 | { |
1dfa0f68 CH |
1586 | bio_free_pages(bio); |
1587 | bio_put(bio); | |
1588 | } | |
1589 | ||
4246a0b6 | 1590 | static void bio_copy_kern_endio_read(struct bio *bio) |
1dfa0f68 | 1591 | { |
42d2683a | 1592 | char *p = bio->bi_private; |
1dfa0f68 | 1593 | struct bio_vec *bvec; |
6dc4f100 | 1594 | struct bvec_iter_all iter_all; |
68154e90 | 1595 | |
2b070cfe | 1596 | bio_for_each_segment_all(bvec, bio, iter_all) { |
1dfa0f68 | 1597 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); |
c8db4448 | 1598 | p += bvec->bv_len; |
68154e90 FT |
1599 | } |
1600 | ||
4246a0b6 | 1601 | bio_copy_kern_endio(bio); |
68154e90 FT |
1602 | } |
1603 | ||
1604 | /** | |
1605 | * bio_copy_kern - copy kernel address into bio | |
1606 | * @q: the struct request_queue for the bio | |
1607 | * @data: pointer to buffer to copy | |
1608 | * @len: length in bytes | |
1609 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1610 | * @reading: data direction is READ |
68154e90 FT |
1611 | * |
1612 | * copy the kernel address into a bio suitable for io to a block | |
1613 | * device. Returns an error pointer in case of error. | |
1614 | */ | |
1615 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1616 | gfp_t gfp_mask, int reading) | |
1617 | { | |
42d2683a CH |
1618 | unsigned long kaddr = (unsigned long)data; |
1619 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1620 | unsigned long start = kaddr >> PAGE_SHIFT; | |
42d2683a CH |
1621 | struct bio *bio; |
1622 | void *p = data; | |
1dfa0f68 | 1623 | int nr_pages = 0; |
68154e90 | 1624 | |
42d2683a CH |
1625 | /* |
1626 | * Overflow, abort | |
1627 | */ | |
1628 | if (end < start) | |
1629 | return ERR_PTR(-EINVAL); | |
68154e90 | 1630 | |
42d2683a CH |
1631 | nr_pages = end - start; |
1632 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1633 | if (!bio) | |
1634 | return ERR_PTR(-ENOMEM); | |
68154e90 | 1635 | |
42d2683a CH |
1636 | while (len) { |
1637 | struct page *page; | |
1638 | unsigned int bytes = PAGE_SIZE; | |
68154e90 | 1639 | |
42d2683a CH |
1640 | if (bytes > len) |
1641 | bytes = len; | |
1642 | ||
1643 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1644 | if (!page) | |
1645 | goto cleanup; | |
1646 | ||
1647 | if (!reading) | |
1648 | memcpy(page_address(page), p, bytes); | |
1649 | ||
1650 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
1651 | break; | |
1652 | ||
1653 | len -= bytes; | |
1654 | p += bytes; | |
68154e90 FT |
1655 | } |
1656 | ||
1dfa0f68 CH |
1657 | if (reading) { |
1658 | bio->bi_end_io = bio_copy_kern_endio_read; | |
1659 | bio->bi_private = data; | |
1660 | } else { | |
1661 | bio->bi_end_io = bio_copy_kern_endio; | |
1dfa0f68 | 1662 | } |
76029ff3 | 1663 | |
68154e90 | 1664 | return bio; |
42d2683a CH |
1665 | |
1666 | cleanup: | |
1dfa0f68 | 1667 | bio_free_pages(bio); |
42d2683a CH |
1668 | bio_put(bio); |
1669 | return ERR_PTR(-ENOMEM); | |
68154e90 FT |
1670 | } |
1671 | ||
1da177e4 LT |
1672 | /* |
1673 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1674 | * for performing direct-IO in BIOs. | |
1675 | * | |
1676 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1677 | * because the required locks are not interrupt-safe. So what we can do is to | |
1678 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1679 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1680 | * in process context. | |
1681 | * | |
1682 | * We special-case compound pages here: normally this means reads into hugetlb | |
1683 | * pages. The logic in here doesn't really work right for compound pages | |
1684 | * because the VM does not uniformly chase down the head page in all cases. | |
1685 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1686 | * handle them at all. So we skip compound pages here at an early stage. | |
1687 | * | |
1688 | * Note that this code is very hard to test under normal circumstances because | |
1689 | * direct-io pins the pages with get_user_pages(). This makes | |
1690 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1691 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1692 | * pagecache. |
1693 | * | |
1694 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1695 | * deferred bio dirtying paths. | |
1696 | */ | |
1697 | ||
1698 | /* | |
1699 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1700 | */ | |
1701 | void bio_set_pages_dirty(struct bio *bio) | |
1702 | { | |
cb34e057 | 1703 | struct bio_vec *bvec; |
6dc4f100 | 1704 | struct bvec_iter_all iter_all; |
1da177e4 | 1705 | |
2b070cfe | 1706 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1707 | if (!PageCompound(bvec->bv_page)) |
1708 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1709 | } |
1710 | } | |
1711 | ||
1da177e4 LT |
1712 | /* |
1713 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1714 | * If they are, then fine. If, however, some pages are clean then they must | |
1715 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1716 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1717 | * |
1718 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1719 | * here on. It will run one put_page() against each page and will run one |
1720 | * bio_put() against the BIO. | |
1da177e4 LT |
1721 | */ |
1722 | ||
65f27f38 | 1723 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1724 | |
65f27f38 | 1725 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1726 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1727 | static struct bio *bio_dirty_list; | |
1728 | ||
1729 | /* | |
1730 | * This runs in process context | |
1731 | */ | |
65f27f38 | 1732 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1733 | { |
24d5493f | 1734 | struct bio *bio, *next; |
1da177e4 | 1735 | |
24d5493f CH |
1736 | spin_lock_irq(&bio_dirty_lock); |
1737 | next = bio_dirty_list; | |
1da177e4 | 1738 | bio_dirty_list = NULL; |
24d5493f | 1739 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1740 | |
24d5493f CH |
1741 | while ((bio = next) != NULL) { |
1742 | next = bio->bi_private; | |
1da177e4 | 1743 | |
d241a95f | 1744 | bio_release_pages(bio, true); |
1da177e4 | 1745 | bio_put(bio); |
1da177e4 LT |
1746 | } |
1747 | } | |
1748 | ||
1749 | void bio_check_pages_dirty(struct bio *bio) | |
1750 | { | |
cb34e057 | 1751 | struct bio_vec *bvec; |
24d5493f | 1752 | unsigned long flags; |
6dc4f100 | 1753 | struct bvec_iter_all iter_all; |
1da177e4 | 1754 | |
2b070cfe | 1755 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1756 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1757 | goto defer; | |
1da177e4 LT |
1758 | } |
1759 | ||
d241a95f | 1760 | bio_release_pages(bio, false); |
24d5493f CH |
1761 | bio_put(bio); |
1762 | return; | |
1763 | defer: | |
1764 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1765 | bio->bi_private = bio_dirty_list; | |
1766 | bio_dirty_list = bio; | |
1767 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1768 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1769 | } |
1770 | ||
2b8bd423 | 1771 | void update_io_ticks(struct hd_struct *part, unsigned long now, bool end) |
5b18b5a7 MP |
1772 | { |
1773 | unsigned long stamp; | |
1774 | again: | |
1775 | stamp = READ_ONCE(part->stamp); | |
1776 | if (unlikely(stamp != now)) { | |
1777 | if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { | |
2b8bd423 | 1778 | __part_stat_add(part, io_ticks, end ? now - stamp : 1); |
5b18b5a7 MP |
1779 | } |
1780 | } | |
1781 | if (part->partno) { | |
1782 | part = &part_to_disk(part)->part0; | |
1783 | goto again; | |
1784 | } | |
1785 | } | |
1da177e4 | 1786 | |
ddcf35d3 | 1787 | void generic_start_io_acct(struct request_queue *q, int op, |
d62e26b3 | 1788 | unsigned long sectors, struct hd_struct *part) |
394ffa50 | 1789 | { |
ddcf35d3 | 1790 | const int sgrp = op_stat_group(op); |
394ffa50 | 1791 | |
112f158f MS |
1792 | part_stat_lock(); |
1793 | ||
2b8bd423 | 1794 | update_io_ticks(part, jiffies, false); |
112f158f MS |
1795 | part_stat_inc(part, ios[sgrp]); |
1796 | part_stat_add(part, sectors[sgrp], sectors); | |
ddcf35d3 | 1797 | part_inc_in_flight(q, part, op_is_write(op)); |
394ffa50 GZ |
1798 | |
1799 | part_stat_unlock(); | |
1800 | } | |
1801 | EXPORT_SYMBOL(generic_start_io_acct); | |
1802 | ||
ddcf35d3 | 1803 | void generic_end_io_acct(struct request_queue *q, int req_op, |
d62e26b3 | 1804 | struct hd_struct *part, unsigned long start_time) |
394ffa50 | 1805 | { |
5b18b5a7 MP |
1806 | unsigned long now = jiffies; |
1807 | unsigned long duration = now - start_time; | |
ddcf35d3 | 1808 | const int sgrp = op_stat_group(req_op); |
394ffa50 | 1809 | |
112f158f MS |
1810 | part_stat_lock(); |
1811 | ||
2b8bd423 | 1812 | update_io_ticks(part, now, true); |
112f158f | 1813 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); |
ddcf35d3 | 1814 | part_dec_in_flight(q, part, op_is_write(req_op)); |
394ffa50 GZ |
1815 | |
1816 | part_stat_unlock(); | |
1817 | } | |
1818 | EXPORT_SYMBOL(generic_end_io_acct); | |
1819 | ||
c4cf5261 JA |
1820 | static inline bool bio_remaining_done(struct bio *bio) |
1821 | { | |
1822 | /* | |
1823 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1824 | * we always end io on the first invocation. | |
1825 | */ | |
1826 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1827 | return true; | |
1828 | ||
1829 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1830 | ||
326e1dbb | 1831 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1832 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1833 | return true; |
326e1dbb | 1834 | } |
c4cf5261 JA |
1835 | |
1836 | return false; | |
1837 | } | |
1838 | ||
1da177e4 LT |
1839 | /** |
1840 | * bio_endio - end I/O on a bio | |
1841 | * @bio: bio | |
1da177e4 LT |
1842 | * |
1843 | * Description: | |
4246a0b6 CH |
1844 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1845 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1846 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1847 | * |
1848 | * bio_endio() can be called several times on a bio that has been chained | |
1849 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
1850 | * last time. At this point the BLK_TA_COMPLETE tracing event will be | |
1851 | * generated if BIO_TRACE_COMPLETION is set. | |
1da177e4 | 1852 | **/ |
4246a0b6 | 1853 | void bio_endio(struct bio *bio) |
1da177e4 | 1854 | { |
ba8c6967 | 1855 | again: |
2b885517 | 1856 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1857 | return; |
7c20f116 CH |
1858 | if (!bio_integrity_endio(bio)) |
1859 | return; | |
1da177e4 | 1860 | |
67b42d0b JB |
1861 | if (bio->bi_disk) |
1862 | rq_qos_done_bio(bio->bi_disk->queue, bio); | |
1863 | ||
ba8c6967 CH |
1864 | /* |
1865 | * Need to have a real endio function for chained bios, otherwise | |
1866 | * various corner cases will break (like stacking block devices that | |
1867 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1868 | * recursion and blowing the stack. Tail call optimization would | |
1869 | * handle this, but compiling with frame pointers also disables | |
1870 | * gcc's sibling call optimization. | |
1871 | */ | |
1872 | if (bio->bi_end_io == bio_chain_endio) { | |
1873 | bio = __bio_chain_endio(bio); | |
1874 | goto again; | |
196d38bc | 1875 | } |
ba8c6967 | 1876 | |
74d46992 CH |
1877 | if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
1878 | trace_block_bio_complete(bio->bi_disk->queue, bio, | |
a462b950 | 1879 | blk_status_to_errno(bio->bi_status)); |
fbbaf700 N |
1880 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
1881 | } | |
1882 | ||
9e234eea | 1883 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1884 | /* release cgroup info */ |
1885 | bio_uninit(bio); | |
ba8c6967 CH |
1886 | if (bio->bi_end_io) |
1887 | bio->bi_end_io(bio); | |
1da177e4 | 1888 | } |
a112a71d | 1889 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1890 | |
20d0189b KO |
1891 | /** |
1892 | * bio_split - split a bio | |
1893 | * @bio: bio to split | |
1894 | * @sectors: number of sectors to split from the front of @bio | |
1895 | * @gfp: gfp mask | |
1896 | * @bs: bio set to allocate from | |
1897 | * | |
1898 | * Allocates and returns a new bio which represents @sectors from the start of | |
1899 | * @bio, and updates @bio to represent the remaining sectors. | |
1900 | * | |
f3f5da62 | 1901 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1902 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1903 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1904 | */ |
1905 | struct bio *bio_split(struct bio *bio, int sectors, | |
1906 | gfp_t gfp, struct bio_set *bs) | |
1907 | { | |
f341a4d3 | 1908 | struct bio *split; |
20d0189b KO |
1909 | |
1910 | BUG_ON(sectors <= 0); | |
1911 | BUG_ON(sectors >= bio_sectors(bio)); | |
1912 | ||
f9d03f96 | 1913 | split = bio_clone_fast(bio, gfp, bs); |
20d0189b KO |
1914 | if (!split) |
1915 | return NULL; | |
1916 | ||
1917 | split->bi_iter.bi_size = sectors << 9; | |
1918 | ||
1919 | if (bio_integrity(split)) | |
fbd08e76 | 1920 | bio_integrity_trim(split); |
20d0189b KO |
1921 | |
1922 | bio_advance(bio, split->bi_iter.bi_size); | |
1923 | ||
fbbaf700 | 1924 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1925 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1926 | |
20d0189b KO |
1927 | return split; |
1928 | } | |
1929 | EXPORT_SYMBOL(bio_split); | |
1930 | ||
6678d83f KO |
1931 | /** |
1932 | * bio_trim - trim a bio | |
1933 | * @bio: bio to trim | |
1934 | * @offset: number of sectors to trim from the front of @bio | |
1935 | * @size: size we want to trim @bio to, in sectors | |
1936 | */ | |
1937 | void bio_trim(struct bio *bio, int offset, int size) | |
1938 | { | |
1939 | /* 'bio' is a cloned bio which we need to trim to match | |
1940 | * the given offset and size. | |
6678d83f | 1941 | */ |
6678d83f KO |
1942 | |
1943 | size <<= 9; | |
4f024f37 | 1944 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1945 | return; |
1946 | ||
6678d83f | 1947 | bio_advance(bio, offset << 9); |
4f024f37 | 1948 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1949 | |
1950 | if (bio_integrity(bio)) | |
fbd08e76 | 1951 | bio_integrity_trim(bio); |
376a78ab | 1952 | |
6678d83f KO |
1953 | } |
1954 | EXPORT_SYMBOL_GPL(bio_trim); | |
1955 | ||
1da177e4 LT |
1956 | /* |
1957 | * create memory pools for biovec's in a bio_set. | |
1958 | * use the global biovec slabs created for general use. | |
1959 | */ | |
8aa6ba2f | 1960 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1961 | { |
ed996a52 | 1962 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
1da177e4 | 1963 | |
8aa6ba2f | 1964 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1965 | } |
1966 | ||
917a38c7 KO |
1967 | /* |
1968 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1969 | * | |
1970 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1971 | * kzalloc()). | |
1972 | */ | |
1973 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1974 | { |
df2cb6da KO |
1975 | if (bs->rescue_workqueue) |
1976 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1977 | bs->rescue_workqueue = NULL; |
df2cb6da | 1978 | |
8aa6ba2f KO |
1979 | mempool_exit(&bs->bio_pool); |
1980 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1981 | |
7878cba9 | 1982 | bioset_integrity_free(bs); |
917a38c7 KO |
1983 | if (bs->bio_slab) |
1984 | bio_put_slab(bs); | |
1985 | bs->bio_slab = NULL; | |
1986 | } | |
1987 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1988 | |
917a38c7 KO |
1989 | /** |
1990 | * bioset_init - Initialize a bio_set | |
dad08527 | 1991 | * @bs: pool to initialize |
917a38c7 KO |
1992 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1993 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1994 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1995 | * and %BIOSET_NEED_RESCUER | |
1996 | * | |
dad08527 KO |
1997 | * Description: |
1998 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1999 | * to ask for a number of bytes to be allocated in front of the bio. | |
2000 | * Front pad allocation is useful for embedding the bio inside | |
2001 | * another structure, to avoid allocating extra data to go with the bio. | |
2002 | * Note that the bio must be embedded at the END of that structure always, | |
2003 | * or things will break badly. | |
2004 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
2005 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
2006 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
2007 | * dispatch queued requests when the mempool runs out of space. | |
2008 | * | |
917a38c7 KO |
2009 | */ |
2010 | int bioset_init(struct bio_set *bs, | |
2011 | unsigned int pool_size, | |
2012 | unsigned int front_pad, | |
2013 | int flags) | |
2014 | { | |
2015 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
2016 | ||
2017 | bs->front_pad = front_pad; | |
2018 | ||
2019 | spin_lock_init(&bs->rescue_lock); | |
2020 | bio_list_init(&bs->rescue_list); | |
2021 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
2022 | ||
2023 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
2024 | if (!bs->bio_slab) | |
2025 | return -ENOMEM; | |
2026 | ||
2027 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) | |
2028 | goto bad; | |
2029 | ||
2030 | if ((flags & BIOSET_NEED_BVECS) && | |
2031 | biovec_init_pool(&bs->bvec_pool, pool_size)) | |
2032 | goto bad; | |
2033 | ||
2034 | if (!(flags & BIOSET_NEED_RESCUER)) | |
2035 | return 0; | |
2036 | ||
2037 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
2038 | if (!bs->rescue_workqueue) | |
2039 | goto bad; | |
2040 | ||
2041 | return 0; | |
2042 | bad: | |
2043 | bioset_exit(bs); | |
2044 | return -ENOMEM; | |
2045 | } | |
2046 | EXPORT_SYMBOL(bioset_init); | |
2047 | ||
28e89fd9 JA |
2048 | /* |
2049 | * Initialize and setup a new bio_set, based on the settings from | |
2050 | * another bio_set. | |
2051 | */ | |
2052 | int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) | |
2053 | { | |
2054 | int flags; | |
2055 | ||
2056 | flags = 0; | |
2057 | if (src->bvec_pool.min_nr) | |
2058 | flags |= BIOSET_NEED_BVECS; | |
2059 | if (src->rescue_workqueue) | |
2060 | flags |= BIOSET_NEED_RESCUER; | |
2061 | ||
2062 | return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); | |
2063 | } | |
2064 | EXPORT_SYMBOL(bioset_init_from_src); | |
2065 | ||
852c788f | 2066 | #ifdef CONFIG_BLK_CGROUP |
1d933cf0 | 2067 | |
74b7c02a | 2068 | /** |
2268c0fe | 2069 | * bio_disassociate_blkg - puts back the blkg reference if associated |
74b7c02a | 2070 | * @bio: target bio |
74b7c02a | 2071 | * |
2268c0fe | 2072 | * Helper to disassociate the blkg from @bio if a blkg is associated. |
74b7c02a | 2073 | */ |
2268c0fe | 2074 | void bio_disassociate_blkg(struct bio *bio) |
74b7c02a | 2075 | { |
2268c0fe DZ |
2076 | if (bio->bi_blkg) { |
2077 | blkg_put(bio->bi_blkg); | |
2078 | bio->bi_blkg = NULL; | |
2079 | } | |
74b7c02a | 2080 | } |
892ad71f | 2081 | EXPORT_SYMBOL_GPL(bio_disassociate_blkg); |
74b7c02a | 2082 | |
a7b39b4e | 2083 | /** |
2268c0fe | 2084 | * __bio_associate_blkg - associate a bio with the a blkg |
a7b39b4e | 2085 | * @bio: target bio |
b5f2954d | 2086 | * @blkg: the blkg to associate |
b5f2954d | 2087 | * |
beea9da0 DZ |
2088 | * This tries to associate @bio with the specified @blkg. Association failure |
2089 | * is handled by walking up the blkg tree. Therefore, the blkg associated can | |
2090 | * be anything between @blkg and the root_blkg. This situation only happens | |
2091 | * when a cgroup is dying and then the remaining bios will spill to the closest | |
2092 | * alive blkg. | |
a7b39b4e | 2093 | * |
beea9da0 DZ |
2094 | * A reference will be taken on the @blkg and will be released when @bio is |
2095 | * freed. | |
a7b39b4e | 2096 | */ |
2268c0fe | 2097 | static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) |
a7b39b4e | 2098 | { |
2268c0fe DZ |
2099 | bio_disassociate_blkg(bio); |
2100 | ||
7754f669 | 2101 | bio->bi_blkg = blkg_tryget_closest(blkg); |
a7b39b4e DZF |
2102 | } |
2103 | ||
d459d853 | 2104 | /** |
fd42df30 | 2105 | * bio_associate_blkg_from_css - associate a bio with a specified css |
d459d853 | 2106 | * @bio: target bio |
fd42df30 | 2107 | * @css: target css |
d459d853 | 2108 | * |
fd42df30 | 2109 | * Associate @bio with the blkg found by combining the css's blkg and the |
fc5a828b DZ |
2110 | * request_queue of the @bio. This falls back to the queue's root_blkg if |
2111 | * the association fails with the css. | |
d459d853 | 2112 | */ |
fd42df30 DZ |
2113 | void bio_associate_blkg_from_css(struct bio *bio, |
2114 | struct cgroup_subsys_state *css) | |
d459d853 | 2115 | { |
fc5a828b DZ |
2116 | struct request_queue *q = bio->bi_disk->queue; |
2117 | struct blkcg_gq *blkg; | |
2118 | ||
2119 | rcu_read_lock(); | |
2120 | ||
2121 | if (!css || !css->parent) | |
2122 | blkg = q->root_blkg; | |
2123 | else | |
2124 | blkg = blkg_lookup_create(css_to_blkcg(css), q); | |
2125 | ||
2126 | __bio_associate_blkg(bio, blkg); | |
2127 | ||
2128 | rcu_read_unlock(); | |
d459d853 | 2129 | } |
fd42df30 | 2130 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); |
d459d853 | 2131 | |
6a7f6d86 | 2132 | #ifdef CONFIG_MEMCG |
852c788f | 2133 | /** |
6a7f6d86 | 2134 | * bio_associate_blkg_from_page - associate a bio with the page's blkg |
852c788f | 2135 | * @bio: target bio |
6a7f6d86 DZ |
2136 | * @page: the page to lookup the blkcg from |
2137 | * | |
2138 | * Associate @bio with the blkg from @page's owning memcg and the respective | |
fc5a828b DZ |
2139 | * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's |
2140 | * root_blkg. | |
852c788f | 2141 | */ |
6a7f6d86 | 2142 | void bio_associate_blkg_from_page(struct bio *bio, struct page *page) |
852c788f | 2143 | { |
6a7f6d86 DZ |
2144 | struct cgroup_subsys_state *css; |
2145 | ||
6a7f6d86 DZ |
2146 | if (!page->mem_cgroup) |
2147 | return; | |
2148 | ||
fc5a828b DZ |
2149 | rcu_read_lock(); |
2150 | ||
2151 | css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); | |
2152 | bio_associate_blkg_from_css(bio, css); | |
2153 | ||
2154 | rcu_read_unlock(); | |
6a7f6d86 DZ |
2155 | } |
2156 | #endif /* CONFIG_MEMCG */ | |
2157 | ||
2268c0fe DZ |
2158 | /** |
2159 | * bio_associate_blkg - associate a bio with a blkg | |
2160 | * @bio: target bio | |
2161 | * | |
2162 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
2163 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
2164 | * already associated, the css is reused and association redone as the | |
2165 | * request_queue may have changed. | |
2166 | */ | |
2167 | void bio_associate_blkg(struct bio *bio) | |
2168 | { | |
fc5a828b | 2169 | struct cgroup_subsys_state *css; |
2268c0fe DZ |
2170 | |
2171 | rcu_read_lock(); | |
2172 | ||
db6638d7 | 2173 | if (bio->bi_blkg) |
fc5a828b | 2174 | css = &bio_blkcg(bio)->css; |
db6638d7 | 2175 | else |
fc5a828b | 2176 | css = blkcg_css(); |
2268c0fe | 2177 | |
fc5a828b | 2178 | bio_associate_blkg_from_css(bio, css); |
2268c0fe DZ |
2179 | |
2180 | rcu_read_unlock(); | |
852c788f | 2181 | } |
5cdf2e3f | 2182 | EXPORT_SYMBOL_GPL(bio_associate_blkg); |
852c788f | 2183 | |
20bd723e | 2184 | /** |
db6638d7 | 2185 | * bio_clone_blkg_association - clone blkg association from src to dst bio |
20bd723e PV |
2186 | * @dst: destination bio |
2187 | * @src: source bio | |
2188 | */ | |
db6638d7 | 2189 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) |
20bd723e | 2190 | { |
6ab21879 DZ |
2191 | rcu_read_lock(); |
2192 | ||
fc5a828b | 2193 | if (src->bi_blkg) |
2268c0fe | 2194 | __bio_associate_blkg(dst, src->bi_blkg); |
6ab21879 DZ |
2195 | |
2196 | rcu_read_unlock(); | |
20bd723e | 2197 | } |
db6638d7 | 2198 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); |
852c788f TH |
2199 | #endif /* CONFIG_BLK_CGROUP */ |
2200 | ||
1da177e4 LT |
2201 | static void __init biovec_init_slabs(void) |
2202 | { | |
2203 | int i; | |
2204 | ||
ed996a52 | 2205 | for (i = 0; i < BVEC_POOL_NR; i++) { |
1da177e4 LT |
2206 | int size; |
2207 | struct biovec_slab *bvs = bvec_slabs + i; | |
2208 | ||
a7fcd37c JA |
2209 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
2210 | bvs->slab = NULL; | |
2211 | continue; | |
2212 | } | |
a7fcd37c | 2213 | |
1da177e4 LT |
2214 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
2215 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 2216 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
2217 | } |
2218 | } | |
2219 | ||
2220 | static int __init init_bio(void) | |
2221 | { | |
bb799ca0 JA |
2222 | bio_slab_max = 2; |
2223 | bio_slab_nr = 0; | |
6396bb22 KC |
2224 | bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), |
2225 | GFP_KERNEL); | |
2b24e6f6 JT |
2226 | |
2227 | BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); | |
2228 | ||
bb799ca0 JA |
2229 | if (!bio_slabs) |
2230 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 2231 | |
7878cba9 | 2232 | bio_integrity_init(); |
1da177e4 LT |
2233 | biovec_init_slabs(); |
2234 | ||
f4f8154a | 2235 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
1da177e4 LT |
2236 | panic("bio: can't allocate bios\n"); |
2237 | ||
f4f8154a | 2238 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
2239 | panic("bio: can't create integrity pool\n"); |
2240 | ||
1da177e4 LT |
2241 | return 0; |
2242 | } | |
1da177e4 | 2243 | subsys_initcall(init_bio); |