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> |
1da177e4 | 20 | |
55782138 | 21 | #include <trace/events/block.h> |
9e234eea | 22 | #include "blk.h" |
67b42d0b | 23 | #include "blk-rq-qos.h" |
0bfc2455 | 24 | |
392ddc32 JA |
25 | /* |
26 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
27 | * itself, to shrink a bio data allocation from two mempool calls to one | |
28 | */ | |
29 | #define BIO_INLINE_VECS 4 | |
30 | ||
1da177e4 LT |
31 | /* |
32 | * if you change this list, also change bvec_alloc or things will | |
33 | * break badly! cannot be bigger than what you can fit into an | |
34 | * unsigned short | |
35 | */ | |
bd5c4fac | 36 | #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } |
ed996a52 | 37 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { |
bd5c4fac | 38 | BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), |
1da177e4 LT |
39 | }; |
40 | #undef BV | |
41 | ||
1da177e4 LT |
42 | /* |
43 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
44 | * IO code that does not need private memory pools. | |
45 | */ | |
f4f8154a | 46 | struct bio_set fs_bio_set; |
3f86a82a | 47 | EXPORT_SYMBOL(fs_bio_set); |
1da177e4 | 48 | |
bb799ca0 JA |
49 | /* |
50 | * Our slab pool management | |
51 | */ | |
52 | struct bio_slab { | |
53 | struct kmem_cache *slab; | |
54 | unsigned int slab_ref; | |
55 | unsigned int slab_size; | |
56 | char name[8]; | |
57 | }; | |
58 | static DEFINE_MUTEX(bio_slab_lock); | |
59 | static struct bio_slab *bio_slabs; | |
60 | static unsigned int bio_slab_nr, bio_slab_max; | |
61 | ||
62 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
63 | { | |
64 | unsigned int sz = sizeof(struct bio) + extra_size; | |
65 | struct kmem_cache *slab = NULL; | |
389d7b26 | 66 | struct bio_slab *bslab, *new_bio_slabs; |
386bc35a | 67 | unsigned int new_bio_slab_max; |
bb799ca0 JA |
68 | unsigned int i, entry = -1; |
69 | ||
70 | mutex_lock(&bio_slab_lock); | |
71 | ||
72 | i = 0; | |
73 | while (i < bio_slab_nr) { | |
f06f135d | 74 | bslab = &bio_slabs[i]; |
bb799ca0 JA |
75 | |
76 | if (!bslab->slab && entry == -1) | |
77 | entry = i; | |
78 | else if (bslab->slab_size == sz) { | |
79 | slab = bslab->slab; | |
80 | bslab->slab_ref++; | |
81 | break; | |
82 | } | |
83 | i++; | |
84 | } | |
85 | ||
86 | if (slab) | |
87 | goto out_unlock; | |
88 | ||
89 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
386bc35a | 90 | new_bio_slab_max = bio_slab_max << 1; |
389d7b26 | 91 | new_bio_slabs = krealloc(bio_slabs, |
386bc35a | 92 | new_bio_slab_max * sizeof(struct bio_slab), |
389d7b26 AK |
93 | GFP_KERNEL); |
94 | if (!new_bio_slabs) | |
bb799ca0 | 95 | goto out_unlock; |
386bc35a | 96 | bio_slab_max = new_bio_slab_max; |
389d7b26 | 97 | bio_slabs = new_bio_slabs; |
bb799ca0 JA |
98 | } |
99 | if (entry == -1) | |
100 | entry = bio_slab_nr++; | |
101 | ||
102 | bslab = &bio_slabs[entry]; | |
103 | ||
104 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
6a241483 MP |
105 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
106 | SLAB_HWCACHE_ALIGN, NULL); | |
bb799ca0 JA |
107 | if (!slab) |
108 | goto out_unlock; | |
109 | ||
bb799ca0 JA |
110 | bslab->slab = slab; |
111 | bslab->slab_ref = 1; | |
112 | bslab->slab_size = sz; | |
113 | out_unlock: | |
114 | mutex_unlock(&bio_slab_lock); | |
115 | return slab; | |
116 | } | |
117 | ||
118 | static void bio_put_slab(struct bio_set *bs) | |
119 | { | |
120 | struct bio_slab *bslab = NULL; | |
121 | unsigned int i; | |
122 | ||
123 | mutex_lock(&bio_slab_lock); | |
124 | ||
125 | for (i = 0; i < bio_slab_nr; i++) { | |
126 | if (bs->bio_slab == bio_slabs[i].slab) { | |
127 | bslab = &bio_slabs[i]; | |
128 | break; | |
129 | } | |
130 | } | |
131 | ||
132 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
133 | goto out; | |
134 | ||
135 | WARN_ON(!bslab->slab_ref); | |
136 | ||
137 | if (--bslab->slab_ref) | |
138 | goto out; | |
139 | ||
140 | kmem_cache_destroy(bslab->slab); | |
141 | bslab->slab = NULL; | |
142 | ||
143 | out: | |
144 | mutex_unlock(&bio_slab_lock); | |
145 | } | |
146 | ||
7ba1ba12 MP |
147 | unsigned int bvec_nr_vecs(unsigned short idx) |
148 | { | |
d6c02a9b | 149 | return bvec_slabs[--idx].nr_vecs; |
7ba1ba12 MP |
150 | } |
151 | ||
9f060e22 | 152 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
bb799ca0 | 153 | { |
ed996a52 CH |
154 | if (!idx) |
155 | return; | |
156 | idx--; | |
157 | ||
158 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
bb799ca0 | 159 | |
ed996a52 | 160 | if (idx == BVEC_POOL_MAX) { |
9f060e22 | 161 | mempool_free(bv, pool); |
ed996a52 | 162 | } else { |
bb799ca0 JA |
163 | struct biovec_slab *bvs = bvec_slabs + idx; |
164 | ||
165 | kmem_cache_free(bvs->slab, bv); | |
166 | } | |
167 | } | |
168 | ||
9f060e22 KO |
169 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
170 | mempool_t *pool) | |
1da177e4 LT |
171 | { |
172 | struct bio_vec *bvl; | |
1da177e4 | 173 | |
7ff9345f JA |
174 | /* |
175 | * see comment near bvec_array define! | |
176 | */ | |
177 | switch (nr) { | |
178 | case 1: | |
179 | *idx = 0; | |
180 | break; | |
181 | case 2 ... 4: | |
182 | *idx = 1; | |
183 | break; | |
184 | case 5 ... 16: | |
185 | *idx = 2; | |
186 | break; | |
187 | case 17 ... 64: | |
188 | *idx = 3; | |
189 | break; | |
190 | case 65 ... 128: | |
191 | *idx = 4; | |
192 | break; | |
193 | case 129 ... BIO_MAX_PAGES: | |
194 | *idx = 5; | |
195 | break; | |
196 | default: | |
197 | return NULL; | |
198 | } | |
199 | ||
200 | /* | |
201 | * idx now points to the pool we want to allocate from. only the | |
202 | * 1-vec entry pool is mempool backed. | |
203 | */ | |
ed996a52 | 204 | if (*idx == BVEC_POOL_MAX) { |
7ff9345f | 205 | fallback: |
9f060e22 | 206 | bvl = mempool_alloc(pool, gfp_mask); |
7ff9345f JA |
207 | } else { |
208 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
d0164adc | 209 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
7ff9345f | 210 | |
0a0d96b0 | 211 | /* |
7ff9345f JA |
212 | * Make this allocation restricted and don't dump info on |
213 | * allocation failures, since we'll fallback to the mempool | |
214 | * in case of failure. | |
0a0d96b0 | 215 | */ |
7ff9345f | 216 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
1da177e4 | 217 | |
0a0d96b0 | 218 | /* |
d0164adc | 219 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM |
7ff9345f | 220 | * is set, retry with the 1-entry mempool |
0a0d96b0 | 221 | */ |
7ff9345f | 222 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
d0164adc | 223 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { |
ed996a52 | 224 | *idx = BVEC_POOL_MAX; |
7ff9345f JA |
225 | goto fallback; |
226 | } | |
227 | } | |
228 | ||
ed996a52 | 229 | (*idx)++; |
1da177e4 LT |
230 | return bvl; |
231 | } | |
232 | ||
9ae3b3f5 | 233 | void bio_uninit(struct bio *bio) |
1da177e4 | 234 | { |
6f70fb66 | 235 | bio_disassociate_blkg(bio); |
ece841ab JT |
236 | |
237 | if (bio_integrity(bio)) | |
238 | bio_integrity_free(bio); | |
4254bba1 | 239 | } |
9ae3b3f5 | 240 | EXPORT_SYMBOL(bio_uninit); |
7ba1ba12 | 241 | |
4254bba1 KO |
242 | static void bio_free(struct bio *bio) |
243 | { | |
244 | struct bio_set *bs = bio->bi_pool; | |
245 | void *p; | |
246 | ||
9ae3b3f5 | 247 | bio_uninit(bio); |
4254bba1 KO |
248 | |
249 | if (bs) { | |
8aa6ba2f | 250 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
4254bba1 KO |
251 | |
252 | /* | |
253 | * If we have front padding, adjust the bio pointer before freeing | |
254 | */ | |
255 | p = bio; | |
bb799ca0 JA |
256 | p -= bs->front_pad; |
257 | ||
8aa6ba2f | 258 | mempool_free(p, &bs->bio_pool); |
4254bba1 KO |
259 | } else { |
260 | /* Bio was allocated by bio_kmalloc() */ | |
261 | kfree(bio); | |
262 | } | |
3676347a PO |
263 | } |
264 | ||
9ae3b3f5 JA |
265 | /* |
266 | * Users of this function have their own bio allocation. Subsequently, | |
267 | * they must remember to pair any call to bio_init() with bio_uninit() | |
268 | * when IO has completed, or when the bio is released. | |
269 | */ | |
3a83f467 ML |
270 | void bio_init(struct bio *bio, struct bio_vec *table, |
271 | unsigned short max_vecs) | |
1da177e4 | 272 | { |
2b94de55 | 273 | memset(bio, 0, sizeof(*bio)); |
c4cf5261 | 274 | atomic_set(&bio->__bi_remaining, 1); |
dac56212 | 275 | atomic_set(&bio->__bi_cnt, 1); |
3a83f467 ML |
276 | |
277 | bio->bi_io_vec = table; | |
278 | bio->bi_max_vecs = max_vecs; | |
1da177e4 | 279 | } |
a112a71d | 280 | EXPORT_SYMBOL(bio_init); |
1da177e4 | 281 | |
f44b48c7 KO |
282 | /** |
283 | * bio_reset - reinitialize a bio | |
284 | * @bio: bio to reset | |
285 | * | |
286 | * Description: | |
287 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
288 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
289 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
290 | * comment in struct bio. | |
291 | */ | |
292 | void bio_reset(struct bio *bio) | |
293 | { | |
294 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
295 | ||
9ae3b3f5 | 296 | bio_uninit(bio); |
f44b48c7 KO |
297 | |
298 | memset(bio, 0, BIO_RESET_BYTES); | |
4246a0b6 | 299 | bio->bi_flags = flags; |
c4cf5261 | 300 | atomic_set(&bio->__bi_remaining, 1); |
f44b48c7 KO |
301 | } |
302 | EXPORT_SYMBOL(bio_reset); | |
303 | ||
38f8baae | 304 | static struct bio *__bio_chain_endio(struct bio *bio) |
196d38bc | 305 | { |
4246a0b6 CH |
306 | struct bio *parent = bio->bi_private; |
307 | ||
4e4cbee9 CH |
308 | if (!parent->bi_status) |
309 | parent->bi_status = bio->bi_status; | |
196d38bc | 310 | bio_put(bio); |
38f8baae CH |
311 | return parent; |
312 | } | |
313 | ||
314 | static void bio_chain_endio(struct bio *bio) | |
315 | { | |
316 | bio_endio(__bio_chain_endio(bio)); | |
196d38bc KO |
317 | } |
318 | ||
319 | /** | |
320 | * bio_chain - chain bio completions | |
1051a902 RD |
321 | * @bio: the target bio |
322 | * @parent: the @bio's parent bio | |
196d38bc KO |
323 | * |
324 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
325 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
326 | * completed; the chained bio will also be freed when it completes. | |
327 | * | |
328 | * The caller must not set bi_private or bi_end_io in @bio. | |
329 | */ | |
330 | void bio_chain(struct bio *bio, struct bio *parent) | |
331 | { | |
332 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
333 | ||
334 | bio->bi_private = parent; | |
335 | bio->bi_end_io = bio_chain_endio; | |
c4cf5261 | 336 | bio_inc_remaining(parent); |
196d38bc KO |
337 | } |
338 | EXPORT_SYMBOL(bio_chain); | |
339 | ||
df2cb6da KO |
340 | static void bio_alloc_rescue(struct work_struct *work) |
341 | { | |
342 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
343 | struct bio *bio; | |
344 | ||
345 | while (1) { | |
346 | spin_lock(&bs->rescue_lock); | |
347 | bio = bio_list_pop(&bs->rescue_list); | |
348 | spin_unlock(&bs->rescue_lock); | |
349 | ||
350 | if (!bio) | |
351 | break; | |
352 | ||
353 | generic_make_request(bio); | |
354 | } | |
355 | } | |
356 | ||
357 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
358 | { | |
359 | struct bio_list punt, nopunt; | |
360 | struct bio *bio; | |
361 | ||
47e0fb46 N |
362 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
363 | return; | |
df2cb6da KO |
364 | /* |
365 | * In order to guarantee forward progress we must punt only bios that | |
366 | * were allocated from this bio_set; otherwise, if there was a bio on | |
367 | * there for a stacking driver higher up in the stack, processing it | |
368 | * could require allocating bios from this bio_set, and doing that from | |
369 | * our own rescuer would be bad. | |
370 | * | |
371 | * Since bio lists are singly linked, pop them all instead of trying to | |
372 | * remove from the middle of the list: | |
373 | */ | |
374 | ||
375 | bio_list_init(&punt); | |
376 | bio_list_init(&nopunt); | |
377 | ||
f5fe1b51 | 378 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
df2cb6da | 379 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
f5fe1b51 | 380 | current->bio_list[0] = nopunt; |
df2cb6da | 381 | |
f5fe1b51 N |
382 | bio_list_init(&nopunt); |
383 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
384 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
385 | current->bio_list[1] = nopunt; | |
df2cb6da KO |
386 | |
387 | spin_lock(&bs->rescue_lock); | |
388 | bio_list_merge(&bs->rescue_list, &punt); | |
389 | spin_unlock(&bs->rescue_lock); | |
390 | ||
391 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
392 | } | |
393 | ||
1da177e4 LT |
394 | /** |
395 | * bio_alloc_bioset - allocate a bio for I/O | |
519c8e9f | 396 | * @gfp_mask: the GFP_* mask given to the slab allocator |
1da177e4 | 397 | * @nr_iovecs: number of iovecs to pre-allocate |
db18efac | 398 | * @bs: the bio_set to allocate from. |
1da177e4 LT |
399 | * |
400 | * Description: | |
3f86a82a KO |
401 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
402 | * backed by the @bs's mempool. | |
403 | * | |
d0164adc MG |
404 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will |
405 | * always be able to allocate a bio. This is due to the mempool guarantees. | |
406 | * To make this work, callers must never allocate more than 1 bio at a time | |
407 | * from this pool. Callers that need to allocate more than 1 bio must always | |
408 | * submit the previously allocated bio for IO before attempting to allocate | |
409 | * a new one. Failure to do so can cause deadlocks under memory pressure. | |
3f86a82a | 410 | * |
df2cb6da KO |
411 | * Note that when running under generic_make_request() (i.e. any block |
412 | * driver), bios are not submitted until after you return - see the code in | |
413 | * generic_make_request() that converts recursion into iteration, to prevent | |
414 | * stack overflows. | |
415 | * | |
416 | * This would normally mean allocating multiple bios under | |
417 | * generic_make_request() would be susceptible to deadlocks, but we have | |
418 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
419 | * thread. | |
420 | * | |
421 | * However, we do not guarantee forward progress for allocations from other | |
422 | * mempools. Doing multiple allocations from the same mempool under | |
423 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
424 | * for per bio allocations. | |
425 | * | |
3f86a82a KO |
426 | * RETURNS: |
427 | * Pointer to new bio on success, NULL on failure. | |
428 | */ | |
7a88fa19 DC |
429 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, |
430 | struct bio_set *bs) | |
1da177e4 | 431 | { |
df2cb6da | 432 | gfp_t saved_gfp = gfp_mask; |
3f86a82a KO |
433 | unsigned front_pad; |
434 | unsigned inline_vecs; | |
34053979 | 435 | struct bio_vec *bvl = NULL; |
451a9ebf TH |
436 | struct bio *bio; |
437 | void *p; | |
438 | ||
3f86a82a KO |
439 | if (!bs) { |
440 | if (nr_iovecs > UIO_MAXIOV) | |
441 | return NULL; | |
442 | ||
443 | p = kmalloc(sizeof(struct bio) + | |
444 | nr_iovecs * sizeof(struct bio_vec), | |
445 | gfp_mask); | |
446 | front_pad = 0; | |
447 | inline_vecs = nr_iovecs; | |
448 | } else { | |
d8f429e1 | 449 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
8aa6ba2f KO |
450 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && |
451 | nr_iovecs > 0)) | |
d8f429e1 | 452 | return NULL; |
df2cb6da KO |
453 | /* |
454 | * generic_make_request() converts recursion to iteration; this | |
455 | * means if we're running beneath it, any bios we allocate and | |
456 | * submit will not be submitted (and thus freed) until after we | |
457 | * return. | |
458 | * | |
459 | * This exposes us to a potential deadlock if we allocate | |
460 | * multiple bios from the same bio_set() while running | |
461 | * underneath generic_make_request(). If we were to allocate | |
462 | * multiple bios (say a stacking block driver that was splitting | |
463 | * bios), we would deadlock if we exhausted the mempool's | |
464 | * reserve. | |
465 | * | |
466 | * We solve this, and guarantee forward progress, with a rescuer | |
467 | * workqueue per bio_set. If we go to allocate and there are | |
468 | * bios on current->bio_list, we first try the allocation | |
d0164adc MG |
469 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those |
470 | * bios we would be blocking to the rescuer workqueue before | |
471 | * we retry with the original gfp_flags. | |
df2cb6da KO |
472 | */ |
473 | ||
f5fe1b51 N |
474 | if (current->bio_list && |
475 | (!bio_list_empty(¤t->bio_list[0]) || | |
47e0fb46 N |
476 | !bio_list_empty(¤t->bio_list[1])) && |
477 | bs->rescue_workqueue) | |
d0164adc | 478 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; |
df2cb6da | 479 | |
8aa6ba2f | 480 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
481 | if (!p && gfp_mask != saved_gfp) { |
482 | punt_bios_to_rescuer(bs); | |
483 | gfp_mask = saved_gfp; | |
8aa6ba2f | 484 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
df2cb6da KO |
485 | } |
486 | ||
3f86a82a KO |
487 | front_pad = bs->front_pad; |
488 | inline_vecs = BIO_INLINE_VECS; | |
489 | } | |
490 | ||
451a9ebf TH |
491 | if (unlikely(!p)) |
492 | return NULL; | |
1da177e4 | 493 | |
3f86a82a | 494 | bio = p + front_pad; |
3a83f467 | 495 | bio_init(bio, NULL, 0); |
34053979 | 496 | |
3f86a82a | 497 | if (nr_iovecs > inline_vecs) { |
ed996a52 CH |
498 | unsigned long idx = 0; |
499 | ||
8aa6ba2f | 500 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
501 | if (!bvl && gfp_mask != saved_gfp) { |
502 | punt_bios_to_rescuer(bs); | |
503 | gfp_mask = saved_gfp; | |
8aa6ba2f | 504 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
df2cb6da KO |
505 | } |
506 | ||
34053979 IM |
507 | if (unlikely(!bvl)) |
508 | goto err_free; | |
a38352e0 | 509 | |
ed996a52 | 510 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
3f86a82a KO |
511 | } else if (nr_iovecs) { |
512 | bvl = bio->bi_inline_vecs; | |
1da177e4 | 513 | } |
3f86a82a KO |
514 | |
515 | bio->bi_pool = bs; | |
34053979 | 516 | bio->bi_max_vecs = nr_iovecs; |
34053979 | 517 | bio->bi_io_vec = bvl; |
1da177e4 | 518 | return bio; |
34053979 IM |
519 | |
520 | err_free: | |
8aa6ba2f | 521 | mempool_free(p, &bs->bio_pool); |
34053979 | 522 | return NULL; |
1da177e4 | 523 | } |
a112a71d | 524 | EXPORT_SYMBOL(bio_alloc_bioset); |
1da177e4 | 525 | |
38a72dac | 526 | void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) |
1da177e4 LT |
527 | { |
528 | unsigned long flags; | |
7988613b KO |
529 | struct bio_vec bv; |
530 | struct bvec_iter iter; | |
1da177e4 | 531 | |
38a72dac | 532 | __bio_for_each_segment(bv, bio, iter, start) { |
7988613b KO |
533 | char *data = bvec_kmap_irq(&bv, &flags); |
534 | memset(data, 0, bv.bv_len); | |
535 | flush_dcache_page(bv.bv_page); | |
1da177e4 LT |
536 | bvec_kunmap_irq(data, &flags); |
537 | } | |
538 | } | |
38a72dac | 539 | EXPORT_SYMBOL(zero_fill_bio_iter); |
1da177e4 LT |
540 | |
541 | /** | |
542 | * bio_put - release a reference to a bio | |
543 | * @bio: bio to release reference to | |
544 | * | |
545 | * Description: | |
546 | * Put a reference to a &struct bio, either one you have gotten with | |
9b10f6a9 | 547 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
1da177e4 LT |
548 | **/ |
549 | void bio_put(struct bio *bio) | |
550 | { | |
dac56212 | 551 | if (!bio_flagged(bio, BIO_REFFED)) |
4254bba1 | 552 | bio_free(bio); |
dac56212 JA |
553 | else { |
554 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
555 | ||
556 | /* | |
557 | * last put frees it | |
558 | */ | |
559 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
560 | bio_free(bio); | |
561 | } | |
1da177e4 | 562 | } |
a112a71d | 563 | EXPORT_SYMBOL(bio_put); |
1da177e4 | 564 | |
59d276fe KO |
565 | /** |
566 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
567 | * @bio: destination bio | |
568 | * @bio_src: bio to clone | |
569 | * | |
570 | * Clone a &bio. Caller will own the returned bio, but not | |
571 | * the actual data it points to. Reference count of returned | |
572 | * bio will be one. | |
573 | * | |
574 | * Caller must ensure that @bio_src is not freed before @bio. | |
575 | */ | |
576 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
577 | { | |
ed996a52 | 578 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
59d276fe KO |
579 | |
580 | /* | |
74d46992 | 581 | * most users will be overriding ->bi_disk with a new target, |
59d276fe KO |
582 | * so we don't set nor calculate new physical/hw segment counts here |
583 | */ | |
74d46992 | 584 | bio->bi_disk = bio_src->bi_disk; |
62530ed8 | 585 | bio->bi_partno = bio_src->bi_partno; |
b7c44ed9 | 586 | bio_set_flag(bio, BIO_CLONED); |
111be883 SL |
587 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
588 | bio_set_flag(bio, BIO_THROTTLED); | |
1eff9d32 | 589 | bio->bi_opf = bio_src->bi_opf; |
ca474b73 | 590 | bio->bi_ioprio = bio_src->bi_ioprio; |
cb6934f8 | 591 | bio->bi_write_hint = bio_src->bi_write_hint; |
59d276fe KO |
592 | bio->bi_iter = bio_src->bi_iter; |
593 | bio->bi_io_vec = bio_src->bi_io_vec; | |
20bd723e | 594 | |
db6638d7 | 595 | bio_clone_blkg_association(bio, bio_src); |
e439bedf | 596 | blkcg_bio_issue_init(bio); |
59d276fe KO |
597 | } |
598 | EXPORT_SYMBOL(__bio_clone_fast); | |
599 | ||
600 | /** | |
601 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
602 | * @bio: bio to clone | |
603 | * @gfp_mask: allocation priority | |
604 | * @bs: bio_set to allocate from | |
605 | * | |
606 | * Like __bio_clone_fast, only also allocates the returned bio | |
607 | */ | |
608 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
609 | { | |
610 | struct bio *b; | |
611 | ||
612 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
613 | if (!b) | |
614 | return NULL; | |
615 | ||
616 | __bio_clone_fast(b, bio); | |
617 | ||
618 | if (bio_integrity(bio)) { | |
619 | int ret; | |
620 | ||
621 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
622 | ||
623 | if (ret < 0) { | |
624 | bio_put(b); | |
625 | return NULL; | |
626 | } | |
627 | } | |
628 | ||
629 | return b; | |
630 | } | |
631 | EXPORT_SYMBOL(bio_clone_fast); | |
632 | ||
5919482e ML |
633 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
634 | struct page *page, unsigned int len, unsigned int off, | |
ff896738 | 635 | bool *same_page) |
5919482e ML |
636 | { |
637 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + | |
638 | bv->bv_offset + bv->bv_len - 1; | |
639 | phys_addr_t page_addr = page_to_phys(page); | |
640 | ||
641 | if (vec_end_addr + 1 != page_addr + off) | |
642 | return false; | |
643 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) | |
644 | return false; | |
52d52d1c | 645 | |
ff896738 CH |
646 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
647 | if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) | |
648 | return false; | |
5919482e ML |
649 | return true; |
650 | } | |
651 | ||
384209cd CH |
652 | static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, |
653 | struct page *page, unsigned len, unsigned offset, | |
654 | bool *same_page) | |
489fbbcb | 655 | { |
384209cd | 656 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
489fbbcb ML |
657 | unsigned long mask = queue_segment_boundary(q); |
658 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; | |
659 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; | |
660 | ||
661 | if ((addr1 | mask) != (addr2 | mask)) | |
662 | return false; | |
489fbbcb ML |
663 | if (bv->bv_len + len > queue_max_segment_size(q)) |
664 | return false; | |
384209cd | 665 | return __bio_try_merge_page(bio, page, len, offset, same_page); |
489fbbcb ML |
666 | } |
667 | ||
1da177e4 | 668 | /** |
19047087 | 669 | * __bio_add_pc_page - attempt to add page to passthrough bio |
c66a14d0 KO |
670 | * @q: the target queue |
671 | * @bio: destination bio | |
672 | * @page: page to add | |
673 | * @len: vec entry length | |
674 | * @offset: vec entry offset | |
d1916c86 | 675 | * @same_page: return if the merge happen inside the same page |
1da177e4 | 676 | * |
c66a14d0 KO |
677 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
678 | * number of reasons, such as the bio being full or target block device | |
679 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
680 | * so it is always possible to add a single page to an empty bio. | |
681 | * | |
5a8ce240 | 682 | * This should only be used by passthrough bios. |
1da177e4 | 683 | */ |
4713839d | 684 | static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, |
19047087 | 685 | struct page *page, unsigned int len, unsigned int offset, |
d1916c86 | 686 | bool *same_page) |
1da177e4 | 687 | { |
1da177e4 LT |
688 | struct bio_vec *bvec; |
689 | ||
690 | /* | |
691 | * cloned bio must not modify vec list | |
692 | */ | |
693 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
694 | return 0; | |
695 | ||
c66a14d0 | 696 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) |
1da177e4 LT |
697 | return 0; |
698 | ||
80cfd548 | 699 | if (bio->bi_vcnt > 0) { |
d1916c86 | 700 | if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) |
384209cd | 701 | return len; |
320ea869 CH |
702 | |
703 | /* | |
704 | * If the queue doesn't support SG gaps and adding this segment | |
705 | * would create a gap, disallow it. | |
706 | */ | |
384209cd | 707 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
320ea869 CH |
708 | if (bvec_gap_to_prev(q, bvec, offset)) |
709 | return 0; | |
80cfd548 JA |
710 | } |
711 | ||
79d08f89 | 712 | if (bio_full(bio, len)) |
1da177e4 LT |
713 | return 0; |
714 | ||
14ccb66b | 715 | if (bio->bi_vcnt >= queue_max_segments(q)) |
489fbbcb ML |
716 | return 0; |
717 | ||
fcbf6a08 ML |
718 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
719 | bvec->bv_page = page; | |
720 | bvec->bv_len = len; | |
721 | bvec->bv_offset = offset; | |
722 | bio->bi_vcnt++; | |
dcdca753 | 723 | bio->bi_iter.bi_size += len; |
1da177e4 LT |
724 | return len; |
725 | } | |
19047087 ML |
726 | |
727 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, | |
728 | struct page *page, unsigned int len, unsigned int offset) | |
729 | { | |
d1916c86 CH |
730 | bool same_page = false; |
731 | return __bio_add_pc_page(q, bio, page, len, offset, &same_page); | |
19047087 | 732 | } |
a112a71d | 733 | EXPORT_SYMBOL(bio_add_pc_page); |
6e68af66 | 734 | |
1da177e4 | 735 | /** |
0aa69fd3 CH |
736 | * __bio_try_merge_page - try appending data to an existing bvec. |
737 | * @bio: destination bio | |
551879a4 | 738 | * @page: start page to add |
0aa69fd3 | 739 | * @len: length of the data to add |
551879a4 | 740 | * @off: offset of the data relative to @page |
ff896738 | 741 | * @same_page: return if the segment has been merged inside the same page |
1da177e4 | 742 | * |
0aa69fd3 CH |
743 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
744 | * a useful optimisation for file systems with a block size smaller than the | |
745 | * page size. | |
746 | * | |
551879a4 ML |
747 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
748 | * | |
0aa69fd3 | 749 | * Return %true on success or %false on failure. |
1da177e4 | 750 | */ |
0aa69fd3 | 751 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
ff896738 | 752 | unsigned int len, unsigned int off, bool *same_page) |
1da177e4 | 753 | { |
c66a14d0 | 754 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
0aa69fd3 | 755 | return false; |
762380ad | 756 | |
e3a5d8e3 | 757 | if (bio->bi_vcnt > 0 && !bio_full(bio, len)) { |
0aa69fd3 | 758 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
5919482e ML |
759 | |
760 | if (page_is_mergeable(bv, page, len, off, same_page)) { | |
761 | bv->bv_len += len; | |
762 | bio->bi_iter.bi_size += len; | |
763 | return true; | |
764 | } | |
c66a14d0 | 765 | } |
0aa69fd3 CH |
766 | return false; |
767 | } | |
768 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); | |
c66a14d0 | 769 | |
0aa69fd3 | 770 | /** |
551879a4 | 771 | * __bio_add_page - add page(s) to a bio in a new segment |
0aa69fd3 | 772 | * @bio: destination bio |
551879a4 ML |
773 | * @page: start page to add |
774 | * @len: length of the data to add, may cross pages | |
775 | * @off: offset of the data relative to @page, may cross pages | |
0aa69fd3 CH |
776 | * |
777 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure | |
778 | * that @bio has space for another bvec. | |
779 | */ | |
780 | void __bio_add_page(struct bio *bio, struct page *page, | |
781 | unsigned int len, unsigned int off) | |
782 | { | |
783 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
c66a14d0 | 784 | |
0aa69fd3 | 785 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
79d08f89 | 786 | WARN_ON_ONCE(bio_full(bio, len)); |
0aa69fd3 CH |
787 | |
788 | bv->bv_page = page; | |
789 | bv->bv_offset = off; | |
790 | bv->bv_len = len; | |
c66a14d0 | 791 | |
c66a14d0 | 792 | bio->bi_iter.bi_size += len; |
0aa69fd3 | 793 | bio->bi_vcnt++; |
b8e24a93 JW |
794 | |
795 | if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) | |
796 | bio_set_flag(bio, BIO_WORKINGSET); | |
0aa69fd3 CH |
797 | } |
798 | EXPORT_SYMBOL_GPL(__bio_add_page); | |
799 | ||
800 | /** | |
551879a4 | 801 | * bio_add_page - attempt to add page(s) to bio |
0aa69fd3 | 802 | * @bio: destination bio |
551879a4 ML |
803 | * @page: start page to add |
804 | * @len: vec entry length, may cross pages | |
805 | * @offset: vec entry offset relative to @page, may cross pages | |
0aa69fd3 | 806 | * |
551879a4 | 807 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
0aa69fd3 CH |
808 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
809 | */ | |
810 | int bio_add_page(struct bio *bio, struct page *page, | |
811 | unsigned int len, unsigned int offset) | |
812 | { | |
ff896738 CH |
813 | bool same_page = false; |
814 | ||
815 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
79d08f89 | 816 | if (bio_full(bio, len)) |
0aa69fd3 CH |
817 | return 0; |
818 | __bio_add_page(bio, page, len, offset); | |
819 | } | |
c66a14d0 | 820 | return len; |
1da177e4 | 821 | } |
a112a71d | 822 | EXPORT_SYMBOL(bio_add_page); |
1da177e4 | 823 | |
d241a95f | 824 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
7321ecbf CH |
825 | { |
826 | struct bvec_iter_all iter_all; | |
827 | struct bio_vec *bvec; | |
7321ecbf | 828 | |
b2d0d991 CH |
829 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
830 | return; | |
831 | ||
d241a95f CH |
832 | bio_for_each_segment_all(bvec, bio, iter_all) { |
833 | if (mark_dirty && !PageCompound(bvec->bv_page)) | |
834 | set_page_dirty_lock(bvec->bv_page); | |
7321ecbf | 835 | put_page(bvec->bv_page); |
d241a95f | 836 | } |
7321ecbf CH |
837 | } |
838 | ||
6d0c48ae JA |
839 | static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) |
840 | { | |
841 | const struct bio_vec *bv = iter->bvec; | |
842 | unsigned int len; | |
843 | size_t size; | |
844 | ||
845 | if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) | |
846 | return -EINVAL; | |
847 | ||
848 | len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); | |
849 | size = bio_add_page(bio, bv->bv_page, len, | |
850 | bv->bv_offset + iter->iov_offset); | |
a10584c3 CH |
851 | if (unlikely(size != len)) |
852 | return -EINVAL; | |
a10584c3 CH |
853 | iov_iter_advance(iter, size); |
854 | return 0; | |
6d0c48ae JA |
855 | } |
856 | ||
576ed913 CH |
857 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
858 | ||
2cefe4db | 859 | /** |
17d51b10 | 860 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
2cefe4db KO |
861 | * @bio: bio to add pages to |
862 | * @iter: iov iterator describing the region to be mapped | |
863 | * | |
17d51b10 | 864 | * Pins pages from *iter and appends them to @bio's bvec array. The |
2cefe4db | 865 | * pages will have to be released using put_page() when done. |
17d51b10 MW |
866 | * For multi-segment *iter, this function only adds pages from the |
867 | * the next non-empty segment of the iov iterator. | |
2cefe4db | 868 | */ |
17d51b10 | 869 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
2cefe4db | 870 | { |
576ed913 CH |
871 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
872 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; | |
2cefe4db KO |
873 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
874 | struct page **pages = (struct page **)bv; | |
45691804 | 875 | bool same_page = false; |
576ed913 CH |
876 | ssize_t size, left; |
877 | unsigned len, i; | |
b403ea24 | 878 | size_t offset; |
576ed913 CH |
879 | |
880 | /* | |
881 | * Move page array up in the allocated memory for the bio vecs as far as | |
882 | * possible so that we can start filling biovecs from the beginning | |
883 | * without overwriting the temporary page array. | |
884 | */ | |
885 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); | |
886 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); | |
2cefe4db KO |
887 | |
888 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
889 | if (unlikely(size <= 0)) | |
890 | return size ? size : -EFAULT; | |
2cefe4db | 891 | |
576ed913 CH |
892 | for (left = size, i = 0; left > 0; left -= len, i++) { |
893 | struct page *page = pages[i]; | |
2cefe4db | 894 | |
576ed913 | 895 | len = min_t(size_t, PAGE_SIZE - offset, left); |
45691804 CH |
896 | |
897 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { | |
898 | if (same_page) | |
899 | put_page(page); | |
900 | } else { | |
79d08f89 | 901 | if (WARN_ON_ONCE(bio_full(bio, len))) |
45691804 CH |
902 | return -EINVAL; |
903 | __bio_add_page(bio, page, len, offset); | |
904 | } | |
576ed913 | 905 | offset = 0; |
2cefe4db KO |
906 | } |
907 | ||
2cefe4db KO |
908 | iov_iter_advance(iter, size); |
909 | return 0; | |
910 | } | |
17d51b10 MW |
911 | |
912 | /** | |
6d0c48ae | 913 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
17d51b10 | 914 | * @bio: bio to add pages to |
6d0c48ae JA |
915 | * @iter: iov iterator describing the region to be added |
916 | * | |
917 | * This takes either an iterator pointing to user memory, or one pointing to | |
918 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and | |
919 | * map them into the kernel. On IO completion, the caller should put those | |
399254aa JA |
920 | * pages. If we're adding kernel pages, and the caller told us it's safe to |
921 | * do so, we just have to add the pages to the bio directly. We don't grab an | |
922 | * extra reference to those pages (the user should already have that), and we | |
923 | * don't put the page on IO completion. The caller needs to check if the bio is | |
924 | * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be | |
925 | * released. | |
17d51b10 | 926 | * |
17d51b10 | 927 | * The function tries, but does not guarantee, to pin as many pages as |
6d0c48ae JA |
928 | * fit into the bio, or are requested in *iter, whatever is smaller. If |
929 | * MM encounters an error pinning the requested pages, it stops. Error | |
930 | * is returned only if 0 pages could be pinned. | |
17d51b10 MW |
931 | */ |
932 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
933 | { | |
6d0c48ae | 934 | const bool is_bvec = iov_iter_is_bvec(iter); |
14eacf12 CH |
935 | int ret; |
936 | ||
937 | if (WARN_ON_ONCE(bio->bi_vcnt)) | |
938 | return -EINVAL; | |
17d51b10 MW |
939 | |
940 | do { | |
6d0c48ae JA |
941 | if (is_bvec) |
942 | ret = __bio_iov_bvec_add_pages(bio, iter); | |
943 | else | |
944 | ret = __bio_iov_iter_get_pages(bio, iter); | |
79d08f89 | 945 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
17d51b10 | 946 | |
b6207430 | 947 | if (is_bvec) |
7321ecbf | 948 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
14eacf12 | 949 | return bio->bi_vcnt ? 0 : ret; |
17d51b10 | 950 | } |
2cefe4db | 951 | |
4246a0b6 | 952 | static void submit_bio_wait_endio(struct bio *bio) |
9e882242 | 953 | { |
65e53aab | 954 | complete(bio->bi_private); |
9e882242 KO |
955 | } |
956 | ||
957 | /** | |
958 | * submit_bio_wait - submit a bio, and wait until it completes | |
9e882242 KO |
959 | * @bio: The &struct bio which describes the I/O |
960 | * | |
961 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
962 | * bio_endio() on failure. | |
3d289d68 JK |
963 | * |
964 | * WARNING: Unlike to how submit_bio() is usually used, this function does not | |
965 | * result in bio reference to be consumed. The caller must drop the reference | |
966 | * on his own. | |
9e882242 | 967 | */ |
4e49ea4a | 968 | int submit_bio_wait(struct bio *bio) |
9e882242 | 969 | { |
e319e1fb | 970 | DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); |
9e882242 | 971 | |
65e53aab | 972 | bio->bi_private = &done; |
9e882242 | 973 | bio->bi_end_io = submit_bio_wait_endio; |
1eff9d32 | 974 | bio->bi_opf |= REQ_SYNC; |
4e49ea4a | 975 | submit_bio(bio); |
65e53aab | 976 | wait_for_completion_io(&done); |
9e882242 | 977 | |
65e53aab | 978 | return blk_status_to_errno(bio->bi_status); |
9e882242 KO |
979 | } |
980 | EXPORT_SYMBOL(submit_bio_wait); | |
981 | ||
054bdf64 KO |
982 | /** |
983 | * bio_advance - increment/complete a bio by some number of bytes | |
984 | * @bio: bio to advance | |
985 | * @bytes: number of bytes to complete | |
986 | * | |
987 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
988 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
989 | * be updated on the last bvec as well. | |
990 | * | |
991 | * @bio will then represent the remaining, uncompleted portion of the io. | |
992 | */ | |
993 | void bio_advance(struct bio *bio, unsigned bytes) | |
994 | { | |
995 | if (bio_integrity(bio)) | |
996 | bio_integrity_advance(bio, bytes); | |
997 | ||
4550dd6c | 998 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
054bdf64 KO |
999 | } |
1000 | EXPORT_SYMBOL(bio_advance); | |
1001 | ||
45db54d5 KO |
1002 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
1003 | struct bio *src, struct bvec_iter *src_iter) | |
16ac3d63 | 1004 | { |
1cb9dda4 | 1005 | struct bio_vec src_bv, dst_bv; |
16ac3d63 | 1006 | void *src_p, *dst_p; |
1cb9dda4 | 1007 | unsigned bytes; |
16ac3d63 | 1008 | |
45db54d5 KO |
1009 | while (src_iter->bi_size && dst_iter->bi_size) { |
1010 | src_bv = bio_iter_iovec(src, *src_iter); | |
1011 | dst_bv = bio_iter_iovec(dst, *dst_iter); | |
1cb9dda4 KO |
1012 | |
1013 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
16ac3d63 | 1014 | |
1cb9dda4 KO |
1015 | src_p = kmap_atomic(src_bv.bv_page); |
1016 | dst_p = kmap_atomic(dst_bv.bv_page); | |
16ac3d63 | 1017 | |
1cb9dda4 KO |
1018 | memcpy(dst_p + dst_bv.bv_offset, |
1019 | src_p + src_bv.bv_offset, | |
16ac3d63 KO |
1020 | bytes); |
1021 | ||
1022 | kunmap_atomic(dst_p); | |
1023 | kunmap_atomic(src_p); | |
1024 | ||
6e6e811d KO |
1025 | flush_dcache_page(dst_bv.bv_page); |
1026 | ||
45db54d5 KO |
1027 | bio_advance_iter(src, src_iter, bytes); |
1028 | bio_advance_iter(dst, dst_iter, bytes); | |
16ac3d63 KO |
1029 | } |
1030 | } | |
38a72dac KO |
1031 | EXPORT_SYMBOL(bio_copy_data_iter); |
1032 | ||
1033 | /** | |
45db54d5 KO |
1034 | * bio_copy_data - copy contents of data buffers from one bio to another |
1035 | * @src: source bio | |
1036 | * @dst: destination bio | |
38a72dac KO |
1037 | * |
1038 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1039 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1040 | */ | |
1041 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1042 | { | |
45db54d5 KO |
1043 | struct bvec_iter src_iter = src->bi_iter; |
1044 | struct bvec_iter dst_iter = dst->bi_iter; | |
1045 | ||
1046 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
38a72dac | 1047 | } |
16ac3d63 KO |
1048 | EXPORT_SYMBOL(bio_copy_data); |
1049 | ||
45db54d5 KO |
1050 | /** |
1051 | * bio_list_copy_data - copy contents of data buffers from one chain of bios to | |
1052 | * another | |
1053 | * @src: source bio list | |
1054 | * @dst: destination bio list | |
1055 | * | |
1056 | * Stops when it reaches the end of either the @src list or @dst list - that is, | |
1057 | * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of | |
1058 | * bios). | |
1059 | */ | |
1060 | void bio_list_copy_data(struct bio *dst, struct bio *src) | |
1061 | { | |
1062 | struct bvec_iter src_iter = src->bi_iter; | |
1063 | struct bvec_iter dst_iter = dst->bi_iter; | |
1064 | ||
1065 | while (1) { | |
1066 | if (!src_iter.bi_size) { | |
1067 | src = src->bi_next; | |
1068 | if (!src) | |
1069 | break; | |
1070 | ||
1071 | src_iter = src->bi_iter; | |
1072 | } | |
1073 | ||
1074 | if (!dst_iter.bi_size) { | |
1075 | dst = dst->bi_next; | |
1076 | if (!dst) | |
1077 | break; | |
1078 | ||
1079 | dst_iter = dst->bi_iter; | |
1080 | } | |
1081 | ||
1082 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); | |
1083 | } | |
1084 | } | |
1085 | EXPORT_SYMBOL(bio_list_copy_data); | |
1086 | ||
1da177e4 | 1087 | struct bio_map_data { |
152e283f | 1088 | int is_our_pages; |
26e49cfc KO |
1089 | struct iov_iter iter; |
1090 | struct iovec iov[]; | |
1da177e4 LT |
1091 | }; |
1092 | ||
0e5b935d | 1093 | static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, |
76029ff3 | 1094 | gfp_t gfp_mask) |
1da177e4 | 1095 | { |
0e5b935d AV |
1096 | struct bio_map_data *bmd; |
1097 | if (data->nr_segs > UIO_MAXIOV) | |
f3f63c1c | 1098 | return NULL; |
1da177e4 | 1099 | |
f1f8f292 | 1100 | bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); |
0e5b935d AV |
1101 | if (!bmd) |
1102 | return NULL; | |
1103 | memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); | |
1104 | bmd->iter = *data; | |
1105 | bmd->iter.iov = bmd->iov; | |
1106 | return bmd; | |
1da177e4 LT |
1107 | } |
1108 | ||
9124d3fe DP |
1109 | /** |
1110 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
1111 | * @bio: The &struct bio which describes the I/O as destination | |
1112 | * @iter: iov_iter as source | |
1113 | * | |
1114 | * Copy all pages from iov_iter to bio. | |
1115 | * Returns 0 on success, or error on failure. | |
1116 | */ | |
98a09d61 | 1117 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) |
c5dec1c3 | 1118 | { |
c5dec1c3 | 1119 | struct bio_vec *bvec; |
6dc4f100 | 1120 | struct bvec_iter_all iter_all; |
c5dec1c3 | 1121 | |
2b070cfe | 1122 | bio_for_each_segment_all(bvec, bio, iter_all) { |
9124d3fe | 1123 | ssize_t ret; |
c5dec1c3 | 1124 | |
9124d3fe DP |
1125 | ret = copy_page_from_iter(bvec->bv_page, |
1126 | bvec->bv_offset, | |
1127 | bvec->bv_len, | |
98a09d61 | 1128 | iter); |
9124d3fe | 1129 | |
98a09d61 | 1130 | if (!iov_iter_count(iter)) |
9124d3fe DP |
1131 | break; |
1132 | ||
1133 | if (ret < bvec->bv_len) | |
1134 | return -EFAULT; | |
c5dec1c3 FT |
1135 | } |
1136 | ||
9124d3fe DP |
1137 | return 0; |
1138 | } | |
1139 | ||
1140 | /** | |
1141 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
1142 | * @bio: The &struct bio which describes the I/O as source | |
1143 | * @iter: iov_iter as destination | |
1144 | * | |
1145 | * Copy all pages from bio to iov_iter. | |
1146 | * Returns 0 on success, or error on failure. | |
1147 | */ | |
1148 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
1149 | { | |
9124d3fe | 1150 | struct bio_vec *bvec; |
6dc4f100 | 1151 | struct bvec_iter_all iter_all; |
9124d3fe | 1152 | |
2b070cfe | 1153 | bio_for_each_segment_all(bvec, bio, iter_all) { |
9124d3fe DP |
1154 | ssize_t ret; |
1155 | ||
1156 | ret = copy_page_to_iter(bvec->bv_page, | |
1157 | bvec->bv_offset, | |
1158 | bvec->bv_len, | |
1159 | &iter); | |
1160 | ||
1161 | if (!iov_iter_count(&iter)) | |
1162 | break; | |
1163 | ||
1164 | if (ret < bvec->bv_len) | |
1165 | return -EFAULT; | |
1166 | } | |
1167 | ||
1168 | return 0; | |
c5dec1c3 FT |
1169 | } |
1170 | ||
491221f8 | 1171 | void bio_free_pages(struct bio *bio) |
1dfa0f68 CH |
1172 | { |
1173 | struct bio_vec *bvec; | |
6dc4f100 | 1174 | struct bvec_iter_all iter_all; |
1dfa0f68 | 1175 | |
2b070cfe | 1176 | bio_for_each_segment_all(bvec, bio, iter_all) |
1dfa0f68 CH |
1177 | __free_page(bvec->bv_page); |
1178 | } | |
491221f8 | 1179 | EXPORT_SYMBOL(bio_free_pages); |
1dfa0f68 | 1180 | |
1da177e4 LT |
1181 | /** |
1182 | * bio_uncopy_user - finish previously mapped bio | |
1183 | * @bio: bio being terminated | |
1184 | * | |
ddad8dd0 | 1185 | * Free pages allocated from bio_copy_user_iov() and write back data |
1da177e4 LT |
1186 | * to user space in case of a read. |
1187 | */ | |
1188 | int bio_uncopy_user(struct bio *bio) | |
1189 | { | |
1190 | struct bio_map_data *bmd = bio->bi_private; | |
1dfa0f68 | 1191 | int ret = 0; |
1da177e4 | 1192 | |
35dc2483 RD |
1193 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { |
1194 | /* | |
1195 | * if we're in a workqueue, the request is orphaned, so | |
2d99b55d HR |
1196 | * don't copy into a random user address space, just free |
1197 | * and return -EINTR so user space doesn't expect any data. | |
35dc2483 | 1198 | */ |
2d99b55d HR |
1199 | if (!current->mm) |
1200 | ret = -EINTR; | |
1201 | else if (bio_data_dir(bio) == READ) | |
9124d3fe | 1202 | ret = bio_copy_to_iter(bio, bmd->iter); |
1dfa0f68 CH |
1203 | if (bmd->is_our_pages) |
1204 | bio_free_pages(bio); | |
35dc2483 | 1205 | } |
c8db4448 | 1206 | kfree(bmd); |
1da177e4 LT |
1207 | bio_put(bio); |
1208 | return ret; | |
1209 | } | |
1210 | ||
1211 | /** | |
c5dec1c3 | 1212 | * bio_copy_user_iov - copy user data to bio |
26e49cfc KO |
1213 | * @q: destination block queue |
1214 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1215 | * @iter: iovec iterator | |
1216 | * @gfp_mask: memory allocation flags | |
1da177e4 LT |
1217 | * |
1218 | * Prepares and returns a bio for indirect user io, bouncing data | |
1219 | * to/from kernel pages as necessary. Must be paired with | |
1220 | * call bio_uncopy_user() on io completion. | |
1221 | */ | |
152e283f FT |
1222 | struct bio *bio_copy_user_iov(struct request_queue *q, |
1223 | struct rq_map_data *map_data, | |
e81cef5d | 1224 | struct iov_iter *iter, |
26e49cfc | 1225 | gfp_t gfp_mask) |
1da177e4 | 1226 | { |
1da177e4 | 1227 | struct bio_map_data *bmd; |
1da177e4 LT |
1228 | struct page *page; |
1229 | struct bio *bio; | |
d16d44eb AV |
1230 | int i = 0, ret; |
1231 | int nr_pages; | |
26e49cfc | 1232 | unsigned int len = iter->count; |
bd5cecea | 1233 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; |
1da177e4 | 1234 | |
0e5b935d | 1235 | bmd = bio_alloc_map_data(iter, gfp_mask); |
1da177e4 LT |
1236 | if (!bmd) |
1237 | return ERR_PTR(-ENOMEM); | |
1238 | ||
26e49cfc KO |
1239 | /* |
1240 | * We need to do a deep copy of the iov_iter including the iovecs. | |
1241 | * The caller provided iov might point to an on-stack or otherwise | |
1242 | * shortlived one. | |
1243 | */ | |
1244 | bmd->is_our_pages = map_data ? 0 : 1; | |
26e49cfc | 1245 | |
d16d44eb AV |
1246 | nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); |
1247 | if (nr_pages > BIO_MAX_PAGES) | |
1248 | nr_pages = BIO_MAX_PAGES; | |
26e49cfc | 1249 | |
1da177e4 | 1250 | ret = -ENOMEM; |
a9e9dc24 | 1251 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1da177e4 LT |
1252 | if (!bio) |
1253 | goto out_bmd; | |
1254 | ||
1da177e4 | 1255 | ret = 0; |
56c451f4 FT |
1256 | |
1257 | if (map_data) { | |
e623ddb4 | 1258 | nr_pages = 1 << map_data->page_order; |
56c451f4 FT |
1259 | i = map_data->offset / PAGE_SIZE; |
1260 | } | |
1da177e4 | 1261 | while (len) { |
e623ddb4 | 1262 | unsigned int bytes = PAGE_SIZE; |
1da177e4 | 1263 | |
56c451f4 FT |
1264 | bytes -= offset; |
1265 | ||
1da177e4 LT |
1266 | if (bytes > len) |
1267 | bytes = len; | |
1268 | ||
152e283f | 1269 | if (map_data) { |
e623ddb4 | 1270 | if (i == map_data->nr_entries * nr_pages) { |
152e283f FT |
1271 | ret = -ENOMEM; |
1272 | break; | |
1273 | } | |
e623ddb4 FT |
1274 | |
1275 | page = map_data->pages[i / nr_pages]; | |
1276 | page += (i % nr_pages); | |
1277 | ||
1278 | i++; | |
1279 | } else { | |
152e283f | 1280 | page = alloc_page(q->bounce_gfp | gfp_mask); |
e623ddb4 FT |
1281 | if (!page) { |
1282 | ret = -ENOMEM; | |
1283 | break; | |
1284 | } | |
1da177e4 LT |
1285 | } |
1286 | ||
a3761c3c JG |
1287 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { |
1288 | if (!map_data) | |
1289 | __free_page(page); | |
1da177e4 | 1290 | break; |
a3761c3c | 1291 | } |
1da177e4 LT |
1292 | |
1293 | len -= bytes; | |
56c451f4 | 1294 | offset = 0; |
1da177e4 LT |
1295 | } |
1296 | ||
1297 | if (ret) | |
1298 | goto cleanup; | |
1299 | ||
2884d0be AV |
1300 | if (map_data) |
1301 | map_data->offset += bio->bi_iter.bi_size; | |
1302 | ||
1da177e4 LT |
1303 | /* |
1304 | * success | |
1305 | */ | |
00e23707 | 1306 | if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || |
ecb554a8 | 1307 | (map_data && map_data->from_user)) { |
98a09d61 | 1308 | ret = bio_copy_from_iter(bio, iter); |
c5dec1c3 FT |
1309 | if (ret) |
1310 | goto cleanup; | |
98a09d61 | 1311 | } else { |
f55adad6 KB |
1312 | if (bmd->is_our_pages) |
1313 | zero_fill_bio(bio); | |
98a09d61 | 1314 | iov_iter_advance(iter, bio->bi_iter.bi_size); |
1da177e4 LT |
1315 | } |
1316 | ||
26e49cfc | 1317 | bio->bi_private = bmd; |
2884d0be AV |
1318 | if (map_data && map_data->null_mapped) |
1319 | bio_set_flag(bio, BIO_NULL_MAPPED); | |
1da177e4 LT |
1320 | return bio; |
1321 | cleanup: | |
152e283f | 1322 | if (!map_data) |
1dfa0f68 | 1323 | bio_free_pages(bio); |
1da177e4 LT |
1324 | bio_put(bio); |
1325 | out_bmd: | |
c8db4448 | 1326 | kfree(bmd); |
1da177e4 LT |
1327 | return ERR_PTR(ret); |
1328 | } | |
1329 | ||
37f19e57 CH |
1330 | /** |
1331 | * bio_map_user_iov - map user iovec into bio | |
1332 | * @q: the struct request_queue for the bio | |
1333 | * @iter: iovec iterator | |
1334 | * @gfp_mask: memory allocation flags | |
1335 | * | |
1336 | * Map the user space address into a bio suitable for io to a block | |
1337 | * device. Returns an error pointer in case of error. | |
1338 | */ | |
1339 | struct bio *bio_map_user_iov(struct request_queue *q, | |
e81cef5d | 1340 | struct iov_iter *iter, |
37f19e57 | 1341 | gfp_t gfp_mask) |
1da177e4 | 1342 | { |
26e49cfc | 1343 | int j; |
1da177e4 | 1344 | struct bio *bio; |
076098e5 | 1345 | int ret; |
1da177e4 | 1346 | |
b282cc76 | 1347 | if (!iov_iter_count(iter)) |
1da177e4 LT |
1348 | return ERR_PTR(-EINVAL); |
1349 | ||
b282cc76 | 1350 | bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); |
1da177e4 LT |
1351 | if (!bio) |
1352 | return ERR_PTR(-ENOMEM); | |
1353 | ||
0a0f1513 | 1354 | while (iov_iter_count(iter)) { |
629e42bc | 1355 | struct page **pages; |
076098e5 AV |
1356 | ssize_t bytes; |
1357 | size_t offs, added = 0; | |
1358 | int npages; | |
1da177e4 | 1359 | |
0a0f1513 | 1360 | bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); |
076098e5 AV |
1361 | if (unlikely(bytes <= 0)) { |
1362 | ret = bytes ? bytes : -EFAULT; | |
f1970baf | 1363 | goto out_unmap; |
99172157 | 1364 | } |
f1970baf | 1365 | |
076098e5 | 1366 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); |
f1970baf | 1367 | |
98f0bc99 AV |
1368 | if (unlikely(offs & queue_dma_alignment(q))) { |
1369 | ret = -EINVAL; | |
1370 | j = 0; | |
1371 | } else { | |
1372 | for (j = 0; j < npages; j++) { | |
1373 | struct page *page = pages[j]; | |
1374 | unsigned int n = PAGE_SIZE - offs; | |
d1916c86 | 1375 | bool same_page = false; |
f1970baf | 1376 | |
98f0bc99 AV |
1377 | if (n > bytes) |
1378 | n = bytes; | |
95d78c28 | 1379 | |
19047087 | 1380 | if (!__bio_add_pc_page(q, bio, page, n, offs, |
d1916c86 CH |
1381 | &same_page)) { |
1382 | if (same_page) | |
1383 | put_page(page); | |
98f0bc99 | 1384 | break; |
d1916c86 | 1385 | } |
1da177e4 | 1386 | |
98f0bc99 AV |
1387 | added += n; |
1388 | bytes -= n; | |
1389 | offs = 0; | |
1390 | } | |
0a0f1513 | 1391 | iov_iter_advance(iter, added); |
f1970baf | 1392 | } |
1da177e4 | 1393 | /* |
f1970baf | 1394 | * release the pages we didn't map into the bio, if any |
1da177e4 | 1395 | */ |
629e42bc | 1396 | while (j < npages) |
09cbfeaf | 1397 | put_page(pages[j++]); |
629e42bc | 1398 | kvfree(pages); |
e2e115d1 AV |
1399 | /* couldn't stuff something into bio? */ |
1400 | if (bytes) | |
1401 | break; | |
1da177e4 LT |
1402 | } |
1403 | ||
b7c44ed9 | 1404 | bio_set_flag(bio, BIO_USER_MAPPED); |
37f19e57 CH |
1405 | |
1406 | /* | |
5fad1b64 | 1407 | * subtle -- if bio_map_user_iov() ended up bouncing a bio, |
37f19e57 CH |
1408 | * it would normally disappear when its bi_end_io is run. |
1409 | * however, we need it for the unmap, so grab an extra | |
1410 | * reference to it | |
1411 | */ | |
1412 | bio_get(bio); | |
1da177e4 | 1413 | return bio; |
f1970baf JB |
1414 | |
1415 | out_unmap: | |
506e0798 | 1416 | bio_release_pages(bio, false); |
1da177e4 LT |
1417 | bio_put(bio); |
1418 | return ERR_PTR(ret); | |
1419 | } | |
1420 | ||
1da177e4 LT |
1421 | /** |
1422 | * bio_unmap_user - unmap a bio | |
1423 | * @bio: the bio being unmapped | |
1424 | * | |
5fad1b64 BVA |
1425 | * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from |
1426 | * process context. | |
1da177e4 LT |
1427 | * |
1428 | * bio_unmap_user() may sleep. | |
1429 | */ | |
1430 | void bio_unmap_user(struct bio *bio) | |
1431 | { | |
163cc2d3 CH |
1432 | bio_release_pages(bio, bio_data_dir(bio) == READ); |
1433 | bio_put(bio); | |
1da177e4 LT |
1434 | bio_put(bio); |
1435 | } | |
1436 | ||
b4c5875d DLM |
1437 | static void bio_invalidate_vmalloc_pages(struct bio *bio) |
1438 | { | |
1439 | #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE | |
1440 | if (bio->bi_private && !op_is_write(bio_op(bio))) { | |
1441 | unsigned long i, len = 0; | |
1442 | ||
1443 | for (i = 0; i < bio->bi_vcnt; i++) | |
1444 | len += bio->bi_io_vec[i].bv_len; | |
1445 | invalidate_kernel_vmap_range(bio->bi_private, len); | |
1446 | } | |
1447 | #endif | |
1448 | } | |
1449 | ||
4246a0b6 | 1450 | static void bio_map_kern_endio(struct bio *bio) |
b823825e | 1451 | { |
b4c5875d | 1452 | bio_invalidate_vmalloc_pages(bio); |
b823825e | 1453 | bio_put(bio); |
b823825e JA |
1454 | } |
1455 | ||
75c72b83 CH |
1456 | /** |
1457 | * bio_map_kern - map kernel address into bio | |
1458 | * @q: the struct request_queue for the bio | |
1459 | * @data: pointer to buffer to map | |
1460 | * @len: length in bytes | |
1461 | * @gfp_mask: allocation flags for bio allocation | |
1462 | * | |
1463 | * Map the kernel address into a bio suitable for io to a block | |
1464 | * device. Returns an error pointer in case of error. | |
1465 | */ | |
1466 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1467 | gfp_t gfp_mask) | |
df46b9a4 MC |
1468 | { |
1469 | unsigned long kaddr = (unsigned long)data; | |
1470 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1471 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1472 | const int nr_pages = end - start; | |
b4c5875d DLM |
1473 | bool is_vmalloc = is_vmalloc_addr(data); |
1474 | struct page *page; | |
df46b9a4 MC |
1475 | int offset, i; |
1476 | struct bio *bio; | |
1477 | ||
a9e9dc24 | 1478 | bio = bio_kmalloc(gfp_mask, nr_pages); |
df46b9a4 MC |
1479 | if (!bio) |
1480 | return ERR_PTR(-ENOMEM); | |
1481 | ||
b4c5875d DLM |
1482 | if (is_vmalloc) { |
1483 | flush_kernel_vmap_range(data, len); | |
1484 | bio->bi_private = data; | |
1485 | } | |
1486 | ||
df46b9a4 MC |
1487 | offset = offset_in_page(kaddr); |
1488 | for (i = 0; i < nr_pages; i++) { | |
1489 | unsigned int bytes = PAGE_SIZE - offset; | |
1490 | ||
1491 | if (len <= 0) | |
1492 | break; | |
1493 | ||
1494 | if (bytes > len) | |
1495 | bytes = len; | |
1496 | ||
b4c5875d DLM |
1497 | if (!is_vmalloc) |
1498 | page = virt_to_page(data); | |
1499 | else | |
1500 | page = vmalloc_to_page(data); | |
1501 | if (bio_add_pc_page(q, bio, page, bytes, | |
75c72b83 CH |
1502 | offset) < bytes) { |
1503 | /* we don't support partial mappings */ | |
1504 | bio_put(bio); | |
1505 | return ERR_PTR(-EINVAL); | |
1506 | } | |
df46b9a4 MC |
1507 | |
1508 | data += bytes; | |
1509 | len -= bytes; | |
1510 | offset = 0; | |
1511 | } | |
1512 | ||
b823825e | 1513 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
1514 | return bio; |
1515 | } | |
df46b9a4 | 1516 | |
4246a0b6 | 1517 | static void bio_copy_kern_endio(struct bio *bio) |
68154e90 | 1518 | { |
1dfa0f68 CH |
1519 | bio_free_pages(bio); |
1520 | bio_put(bio); | |
1521 | } | |
1522 | ||
4246a0b6 | 1523 | static void bio_copy_kern_endio_read(struct bio *bio) |
1dfa0f68 | 1524 | { |
42d2683a | 1525 | char *p = bio->bi_private; |
1dfa0f68 | 1526 | struct bio_vec *bvec; |
6dc4f100 | 1527 | struct bvec_iter_all iter_all; |
68154e90 | 1528 | |
2b070cfe | 1529 | bio_for_each_segment_all(bvec, bio, iter_all) { |
1dfa0f68 | 1530 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); |
c8db4448 | 1531 | p += bvec->bv_len; |
68154e90 FT |
1532 | } |
1533 | ||
4246a0b6 | 1534 | bio_copy_kern_endio(bio); |
68154e90 FT |
1535 | } |
1536 | ||
1537 | /** | |
1538 | * bio_copy_kern - copy kernel address into bio | |
1539 | * @q: the struct request_queue for the bio | |
1540 | * @data: pointer to buffer to copy | |
1541 | * @len: length in bytes | |
1542 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 1543 | * @reading: data direction is READ |
68154e90 FT |
1544 | * |
1545 | * copy the kernel address into a bio suitable for io to a block | |
1546 | * device. Returns an error pointer in case of error. | |
1547 | */ | |
1548 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1549 | gfp_t gfp_mask, int reading) | |
1550 | { | |
42d2683a CH |
1551 | unsigned long kaddr = (unsigned long)data; |
1552 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1553 | unsigned long start = kaddr >> PAGE_SHIFT; | |
42d2683a CH |
1554 | struct bio *bio; |
1555 | void *p = data; | |
1dfa0f68 | 1556 | int nr_pages = 0; |
68154e90 | 1557 | |
42d2683a CH |
1558 | /* |
1559 | * Overflow, abort | |
1560 | */ | |
1561 | if (end < start) | |
1562 | return ERR_PTR(-EINVAL); | |
68154e90 | 1563 | |
42d2683a CH |
1564 | nr_pages = end - start; |
1565 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1566 | if (!bio) | |
1567 | return ERR_PTR(-ENOMEM); | |
68154e90 | 1568 | |
42d2683a CH |
1569 | while (len) { |
1570 | struct page *page; | |
1571 | unsigned int bytes = PAGE_SIZE; | |
68154e90 | 1572 | |
42d2683a CH |
1573 | if (bytes > len) |
1574 | bytes = len; | |
1575 | ||
1576 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1577 | if (!page) | |
1578 | goto cleanup; | |
1579 | ||
1580 | if (!reading) | |
1581 | memcpy(page_address(page), p, bytes); | |
1582 | ||
1583 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
1584 | break; | |
1585 | ||
1586 | len -= bytes; | |
1587 | p += bytes; | |
68154e90 FT |
1588 | } |
1589 | ||
1dfa0f68 CH |
1590 | if (reading) { |
1591 | bio->bi_end_io = bio_copy_kern_endio_read; | |
1592 | bio->bi_private = data; | |
1593 | } else { | |
1594 | bio->bi_end_io = bio_copy_kern_endio; | |
1dfa0f68 | 1595 | } |
76029ff3 | 1596 | |
68154e90 | 1597 | return bio; |
42d2683a CH |
1598 | |
1599 | cleanup: | |
1dfa0f68 | 1600 | bio_free_pages(bio); |
42d2683a CH |
1601 | bio_put(bio); |
1602 | return ERR_PTR(-ENOMEM); | |
68154e90 FT |
1603 | } |
1604 | ||
1da177e4 LT |
1605 | /* |
1606 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1607 | * for performing direct-IO in BIOs. | |
1608 | * | |
1609 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1610 | * because the required locks are not interrupt-safe. So what we can do is to | |
1611 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1612 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1613 | * in process context. | |
1614 | * | |
1615 | * We special-case compound pages here: normally this means reads into hugetlb | |
1616 | * pages. The logic in here doesn't really work right for compound pages | |
1617 | * because the VM does not uniformly chase down the head page in all cases. | |
1618 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1619 | * handle them at all. So we skip compound pages here at an early stage. | |
1620 | * | |
1621 | * Note that this code is very hard to test under normal circumstances because | |
1622 | * direct-io pins the pages with get_user_pages(). This makes | |
1623 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
0d5c3eba | 1624 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1da177e4 LT |
1625 | * pagecache. |
1626 | * | |
1627 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1628 | * deferred bio dirtying paths. | |
1629 | */ | |
1630 | ||
1631 | /* | |
1632 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1633 | */ | |
1634 | void bio_set_pages_dirty(struct bio *bio) | |
1635 | { | |
cb34e057 | 1636 | struct bio_vec *bvec; |
6dc4f100 | 1637 | struct bvec_iter_all iter_all; |
1da177e4 | 1638 | |
2b070cfe | 1639 | bio_for_each_segment_all(bvec, bio, iter_all) { |
3bb50983 CH |
1640 | if (!PageCompound(bvec->bv_page)) |
1641 | set_page_dirty_lock(bvec->bv_page); | |
1da177e4 LT |
1642 | } |
1643 | } | |
1644 | ||
1da177e4 LT |
1645 | /* |
1646 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1647 | * If they are, then fine. If, however, some pages are clean then they must | |
1648 | * have been written out during the direct-IO read. So we take another ref on | |
24d5493f | 1649 | * the BIO and re-dirty the pages in process context. |
1da177e4 LT |
1650 | * |
1651 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
ea1754a0 KS |
1652 | * here on. It will run one put_page() against each page and will run one |
1653 | * bio_put() against the BIO. | |
1da177e4 LT |
1654 | */ |
1655 | ||
65f27f38 | 1656 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1657 | |
65f27f38 | 1658 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1659 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1660 | static struct bio *bio_dirty_list; | |
1661 | ||
1662 | /* | |
1663 | * This runs in process context | |
1664 | */ | |
65f27f38 | 1665 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 | 1666 | { |
24d5493f | 1667 | struct bio *bio, *next; |
1da177e4 | 1668 | |
24d5493f CH |
1669 | spin_lock_irq(&bio_dirty_lock); |
1670 | next = bio_dirty_list; | |
1da177e4 | 1671 | bio_dirty_list = NULL; |
24d5493f | 1672 | spin_unlock_irq(&bio_dirty_lock); |
1da177e4 | 1673 | |
24d5493f CH |
1674 | while ((bio = next) != NULL) { |
1675 | next = bio->bi_private; | |
1da177e4 | 1676 | |
d241a95f | 1677 | bio_release_pages(bio, true); |
1da177e4 | 1678 | bio_put(bio); |
1da177e4 LT |
1679 | } |
1680 | } | |
1681 | ||
1682 | void bio_check_pages_dirty(struct bio *bio) | |
1683 | { | |
cb34e057 | 1684 | struct bio_vec *bvec; |
24d5493f | 1685 | unsigned long flags; |
6dc4f100 | 1686 | struct bvec_iter_all iter_all; |
1da177e4 | 1687 | |
2b070cfe | 1688 | bio_for_each_segment_all(bvec, bio, iter_all) { |
24d5493f CH |
1689 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
1690 | goto defer; | |
1da177e4 LT |
1691 | } |
1692 | ||
d241a95f | 1693 | bio_release_pages(bio, false); |
24d5493f CH |
1694 | bio_put(bio); |
1695 | return; | |
1696 | defer: | |
1697 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1698 | bio->bi_private = bio_dirty_list; | |
1699 | bio_dirty_list = bio; | |
1700 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1701 | schedule_work(&bio_dirty_work); | |
1da177e4 LT |
1702 | } |
1703 | ||
5b18b5a7 MP |
1704 | void update_io_ticks(struct hd_struct *part, unsigned long now) |
1705 | { | |
1706 | unsigned long stamp; | |
1707 | again: | |
1708 | stamp = READ_ONCE(part->stamp); | |
1709 | if (unlikely(stamp != now)) { | |
1710 | if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { | |
1711 | __part_stat_add(part, io_ticks, 1); | |
1712 | } | |
1713 | } | |
1714 | if (part->partno) { | |
1715 | part = &part_to_disk(part)->part0; | |
1716 | goto again; | |
1717 | } | |
1718 | } | |
1da177e4 | 1719 | |
ddcf35d3 | 1720 | void generic_start_io_acct(struct request_queue *q, int op, |
d62e26b3 | 1721 | unsigned long sectors, struct hd_struct *part) |
394ffa50 | 1722 | { |
ddcf35d3 | 1723 | const int sgrp = op_stat_group(op); |
394ffa50 | 1724 | |
112f158f MS |
1725 | part_stat_lock(); |
1726 | ||
5b18b5a7 | 1727 | update_io_ticks(part, jiffies); |
112f158f MS |
1728 | part_stat_inc(part, ios[sgrp]); |
1729 | part_stat_add(part, sectors[sgrp], sectors); | |
ddcf35d3 | 1730 | part_inc_in_flight(q, part, op_is_write(op)); |
394ffa50 GZ |
1731 | |
1732 | part_stat_unlock(); | |
1733 | } | |
1734 | EXPORT_SYMBOL(generic_start_io_acct); | |
1735 | ||
ddcf35d3 | 1736 | void generic_end_io_acct(struct request_queue *q, int req_op, |
d62e26b3 | 1737 | struct hd_struct *part, unsigned long start_time) |
394ffa50 | 1738 | { |
5b18b5a7 MP |
1739 | unsigned long now = jiffies; |
1740 | unsigned long duration = now - start_time; | |
ddcf35d3 | 1741 | const int sgrp = op_stat_group(req_op); |
394ffa50 | 1742 | |
112f158f MS |
1743 | part_stat_lock(); |
1744 | ||
5b18b5a7 | 1745 | update_io_ticks(part, now); |
112f158f | 1746 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); |
5b18b5a7 | 1747 | part_stat_add(part, time_in_queue, duration); |
ddcf35d3 | 1748 | part_dec_in_flight(q, part, op_is_write(req_op)); |
394ffa50 GZ |
1749 | |
1750 | part_stat_unlock(); | |
1751 | } | |
1752 | EXPORT_SYMBOL(generic_end_io_acct); | |
1753 | ||
c4cf5261 JA |
1754 | static inline bool bio_remaining_done(struct bio *bio) |
1755 | { | |
1756 | /* | |
1757 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1758 | * we always end io on the first invocation. | |
1759 | */ | |
1760 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1761 | return true; | |
1762 | ||
1763 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1764 | ||
326e1dbb | 1765 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
b7c44ed9 | 1766 | bio_clear_flag(bio, BIO_CHAIN); |
c4cf5261 | 1767 | return true; |
326e1dbb | 1768 | } |
c4cf5261 JA |
1769 | |
1770 | return false; | |
1771 | } | |
1772 | ||
1da177e4 LT |
1773 | /** |
1774 | * bio_endio - end I/O on a bio | |
1775 | * @bio: bio | |
1da177e4 LT |
1776 | * |
1777 | * Description: | |
4246a0b6 CH |
1778 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
1779 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1780 | * bio unless they own it and thus know that it has an end_io function. | |
fbbaf700 N |
1781 | * |
1782 | * bio_endio() can be called several times on a bio that has been chained | |
1783 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
1784 | * last time. At this point the BLK_TA_COMPLETE tracing event will be | |
1785 | * generated if BIO_TRACE_COMPLETION is set. | |
1da177e4 | 1786 | **/ |
4246a0b6 | 1787 | void bio_endio(struct bio *bio) |
1da177e4 | 1788 | { |
ba8c6967 | 1789 | again: |
2b885517 | 1790 | if (!bio_remaining_done(bio)) |
ba8c6967 | 1791 | return; |
7c20f116 CH |
1792 | if (!bio_integrity_endio(bio)) |
1793 | return; | |
1da177e4 | 1794 | |
67b42d0b JB |
1795 | if (bio->bi_disk) |
1796 | rq_qos_done_bio(bio->bi_disk->queue, bio); | |
1797 | ||
ba8c6967 CH |
1798 | /* |
1799 | * Need to have a real endio function for chained bios, otherwise | |
1800 | * various corner cases will break (like stacking block devices that | |
1801 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1802 | * recursion and blowing the stack. Tail call optimization would | |
1803 | * handle this, but compiling with frame pointers also disables | |
1804 | * gcc's sibling call optimization. | |
1805 | */ | |
1806 | if (bio->bi_end_io == bio_chain_endio) { | |
1807 | bio = __bio_chain_endio(bio); | |
1808 | goto again; | |
196d38bc | 1809 | } |
ba8c6967 | 1810 | |
74d46992 CH |
1811 | if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
1812 | trace_block_bio_complete(bio->bi_disk->queue, bio, | |
a462b950 | 1813 | blk_status_to_errno(bio->bi_status)); |
fbbaf700 N |
1814 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
1815 | } | |
1816 | ||
9e234eea | 1817 | blk_throtl_bio_endio(bio); |
b222dd2f SL |
1818 | /* release cgroup info */ |
1819 | bio_uninit(bio); | |
ba8c6967 CH |
1820 | if (bio->bi_end_io) |
1821 | bio->bi_end_io(bio); | |
1da177e4 | 1822 | } |
a112a71d | 1823 | EXPORT_SYMBOL(bio_endio); |
1da177e4 | 1824 | |
20d0189b KO |
1825 | /** |
1826 | * bio_split - split a bio | |
1827 | * @bio: bio to split | |
1828 | * @sectors: number of sectors to split from the front of @bio | |
1829 | * @gfp: gfp mask | |
1830 | * @bs: bio set to allocate from | |
1831 | * | |
1832 | * Allocates and returns a new bio which represents @sectors from the start of | |
1833 | * @bio, and updates @bio to represent the remaining sectors. | |
1834 | * | |
f3f5da62 | 1835 | * Unless this is a discard request the newly allocated bio will point |
dad77584 BVA |
1836 | * to @bio's bi_io_vec. It is the caller's responsibility to ensure that |
1837 | * neither @bio nor @bs are freed before the split bio. | |
20d0189b KO |
1838 | */ |
1839 | struct bio *bio_split(struct bio *bio, int sectors, | |
1840 | gfp_t gfp, struct bio_set *bs) | |
1841 | { | |
f341a4d3 | 1842 | struct bio *split; |
20d0189b KO |
1843 | |
1844 | BUG_ON(sectors <= 0); | |
1845 | BUG_ON(sectors >= bio_sectors(bio)); | |
1846 | ||
f9d03f96 | 1847 | split = bio_clone_fast(bio, gfp, bs); |
20d0189b KO |
1848 | if (!split) |
1849 | return NULL; | |
1850 | ||
1851 | split->bi_iter.bi_size = sectors << 9; | |
1852 | ||
1853 | if (bio_integrity(split)) | |
fbd08e76 | 1854 | bio_integrity_trim(split); |
20d0189b KO |
1855 | |
1856 | bio_advance(bio, split->bi_iter.bi_size); | |
1857 | ||
fbbaf700 | 1858 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
20d59023 | 1859 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
fbbaf700 | 1860 | |
20d0189b KO |
1861 | return split; |
1862 | } | |
1863 | EXPORT_SYMBOL(bio_split); | |
1864 | ||
6678d83f KO |
1865 | /** |
1866 | * bio_trim - trim a bio | |
1867 | * @bio: bio to trim | |
1868 | * @offset: number of sectors to trim from the front of @bio | |
1869 | * @size: size we want to trim @bio to, in sectors | |
1870 | */ | |
1871 | void bio_trim(struct bio *bio, int offset, int size) | |
1872 | { | |
1873 | /* 'bio' is a cloned bio which we need to trim to match | |
1874 | * the given offset and size. | |
6678d83f | 1875 | */ |
6678d83f KO |
1876 | |
1877 | size <<= 9; | |
4f024f37 | 1878 | if (offset == 0 && size == bio->bi_iter.bi_size) |
6678d83f KO |
1879 | return; |
1880 | ||
6678d83f | 1881 | bio_advance(bio, offset << 9); |
4f024f37 | 1882 | bio->bi_iter.bi_size = size; |
376a78ab DM |
1883 | |
1884 | if (bio_integrity(bio)) | |
fbd08e76 | 1885 | bio_integrity_trim(bio); |
376a78ab | 1886 | |
6678d83f KO |
1887 | } |
1888 | EXPORT_SYMBOL_GPL(bio_trim); | |
1889 | ||
1da177e4 LT |
1890 | /* |
1891 | * create memory pools for biovec's in a bio_set. | |
1892 | * use the global biovec slabs created for general use. | |
1893 | */ | |
8aa6ba2f | 1894 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
1da177e4 | 1895 | { |
ed996a52 | 1896 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
1da177e4 | 1897 | |
8aa6ba2f | 1898 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
1da177e4 LT |
1899 | } |
1900 | ||
917a38c7 KO |
1901 | /* |
1902 | * bioset_exit - exit a bioset initialized with bioset_init() | |
1903 | * | |
1904 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with | |
1905 | * kzalloc()). | |
1906 | */ | |
1907 | void bioset_exit(struct bio_set *bs) | |
1da177e4 | 1908 | { |
df2cb6da KO |
1909 | if (bs->rescue_workqueue) |
1910 | destroy_workqueue(bs->rescue_workqueue); | |
917a38c7 | 1911 | bs->rescue_workqueue = NULL; |
df2cb6da | 1912 | |
8aa6ba2f KO |
1913 | mempool_exit(&bs->bio_pool); |
1914 | mempool_exit(&bs->bvec_pool); | |
9f060e22 | 1915 | |
7878cba9 | 1916 | bioset_integrity_free(bs); |
917a38c7 KO |
1917 | if (bs->bio_slab) |
1918 | bio_put_slab(bs); | |
1919 | bs->bio_slab = NULL; | |
1920 | } | |
1921 | EXPORT_SYMBOL(bioset_exit); | |
1da177e4 | 1922 | |
917a38c7 KO |
1923 | /** |
1924 | * bioset_init - Initialize a bio_set | |
dad08527 | 1925 | * @bs: pool to initialize |
917a38c7 KO |
1926 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1927 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1928 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1929 | * and %BIOSET_NEED_RESCUER | |
1930 | * | |
dad08527 KO |
1931 | * Description: |
1932 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1933 | * to ask for a number of bytes to be allocated in front of the bio. | |
1934 | * Front pad allocation is useful for embedding the bio inside | |
1935 | * another structure, to avoid allocating extra data to go with the bio. | |
1936 | * Note that the bio must be embedded at the END of that structure always, | |
1937 | * or things will break badly. | |
1938 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
1939 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
1940 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
1941 | * dispatch queued requests when the mempool runs out of space. | |
1942 | * | |
917a38c7 KO |
1943 | */ |
1944 | int bioset_init(struct bio_set *bs, | |
1945 | unsigned int pool_size, | |
1946 | unsigned int front_pad, | |
1947 | int flags) | |
1948 | { | |
1949 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1950 | ||
1951 | bs->front_pad = front_pad; | |
1952 | ||
1953 | spin_lock_init(&bs->rescue_lock); | |
1954 | bio_list_init(&bs->rescue_list); | |
1955 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1956 | ||
1957 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1958 | if (!bs->bio_slab) | |
1959 | return -ENOMEM; | |
1960 | ||
1961 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) | |
1962 | goto bad; | |
1963 | ||
1964 | if ((flags & BIOSET_NEED_BVECS) && | |
1965 | biovec_init_pool(&bs->bvec_pool, pool_size)) | |
1966 | goto bad; | |
1967 | ||
1968 | if (!(flags & BIOSET_NEED_RESCUER)) | |
1969 | return 0; | |
1970 | ||
1971 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1972 | if (!bs->rescue_workqueue) | |
1973 | goto bad; | |
1974 | ||
1975 | return 0; | |
1976 | bad: | |
1977 | bioset_exit(bs); | |
1978 | return -ENOMEM; | |
1979 | } | |
1980 | EXPORT_SYMBOL(bioset_init); | |
1981 | ||
28e89fd9 JA |
1982 | /* |
1983 | * Initialize and setup a new bio_set, based on the settings from | |
1984 | * another bio_set. | |
1985 | */ | |
1986 | int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) | |
1987 | { | |
1988 | int flags; | |
1989 | ||
1990 | flags = 0; | |
1991 | if (src->bvec_pool.min_nr) | |
1992 | flags |= BIOSET_NEED_BVECS; | |
1993 | if (src->rescue_workqueue) | |
1994 | flags |= BIOSET_NEED_RESCUER; | |
1995 | ||
1996 | return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); | |
1997 | } | |
1998 | EXPORT_SYMBOL(bioset_init_from_src); | |
1999 | ||
852c788f | 2000 | #ifdef CONFIG_BLK_CGROUP |
1d933cf0 | 2001 | |
74b7c02a | 2002 | /** |
2268c0fe | 2003 | * bio_disassociate_blkg - puts back the blkg reference if associated |
74b7c02a | 2004 | * @bio: target bio |
74b7c02a | 2005 | * |
2268c0fe | 2006 | * Helper to disassociate the blkg from @bio if a blkg is associated. |
74b7c02a | 2007 | */ |
2268c0fe | 2008 | void bio_disassociate_blkg(struct bio *bio) |
74b7c02a | 2009 | { |
2268c0fe DZ |
2010 | if (bio->bi_blkg) { |
2011 | blkg_put(bio->bi_blkg); | |
2012 | bio->bi_blkg = NULL; | |
2013 | } | |
74b7c02a | 2014 | } |
892ad71f | 2015 | EXPORT_SYMBOL_GPL(bio_disassociate_blkg); |
74b7c02a | 2016 | |
a7b39b4e | 2017 | /** |
2268c0fe | 2018 | * __bio_associate_blkg - associate a bio with the a blkg |
a7b39b4e | 2019 | * @bio: target bio |
b5f2954d | 2020 | * @blkg: the blkg to associate |
b5f2954d | 2021 | * |
beea9da0 DZ |
2022 | * This tries to associate @bio with the specified @blkg. Association failure |
2023 | * is handled by walking up the blkg tree. Therefore, the blkg associated can | |
2024 | * be anything between @blkg and the root_blkg. This situation only happens | |
2025 | * when a cgroup is dying and then the remaining bios will spill to the closest | |
2026 | * alive blkg. | |
a7b39b4e | 2027 | * |
beea9da0 DZ |
2028 | * A reference will be taken on the @blkg and will be released when @bio is |
2029 | * freed. | |
a7b39b4e | 2030 | */ |
2268c0fe | 2031 | static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) |
a7b39b4e | 2032 | { |
2268c0fe DZ |
2033 | bio_disassociate_blkg(bio); |
2034 | ||
7754f669 | 2035 | bio->bi_blkg = blkg_tryget_closest(blkg); |
a7b39b4e DZF |
2036 | } |
2037 | ||
d459d853 | 2038 | /** |
fd42df30 | 2039 | * bio_associate_blkg_from_css - associate a bio with a specified css |
d459d853 | 2040 | * @bio: target bio |
fd42df30 | 2041 | * @css: target css |
d459d853 | 2042 | * |
fd42df30 | 2043 | * Associate @bio with the blkg found by combining the css's blkg and the |
fc5a828b DZ |
2044 | * request_queue of the @bio. This falls back to the queue's root_blkg if |
2045 | * the association fails with the css. | |
d459d853 | 2046 | */ |
fd42df30 DZ |
2047 | void bio_associate_blkg_from_css(struct bio *bio, |
2048 | struct cgroup_subsys_state *css) | |
d459d853 | 2049 | { |
fc5a828b DZ |
2050 | struct request_queue *q = bio->bi_disk->queue; |
2051 | struct blkcg_gq *blkg; | |
2052 | ||
2053 | rcu_read_lock(); | |
2054 | ||
2055 | if (!css || !css->parent) | |
2056 | blkg = q->root_blkg; | |
2057 | else | |
2058 | blkg = blkg_lookup_create(css_to_blkcg(css), q); | |
2059 | ||
2060 | __bio_associate_blkg(bio, blkg); | |
2061 | ||
2062 | rcu_read_unlock(); | |
d459d853 | 2063 | } |
fd42df30 | 2064 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); |
d459d853 | 2065 | |
6a7f6d86 | 2066 | #ifdef CONFIG_MEMCG |
852c788f | 2067 | /** |
6a7f6d86 | 2068 | * bio_associate_blkg_from_page - associate a bio with the page's blkg |
852c788f | 2069 | * @bio: target bio |
6a7f6d86 DZ |
2070 | * @page: the page to lookup the blkcg from |
2071 | * | |
2072 | * Associate @bio with the blkg from @page's owning memcg and the respective | |
fc5a828b DZ |
2073 | * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's |
2074 | * root_blkg. | |
852c788f | 2075 | */ |
6a7f6d86 | 2076 | void bio_associate_blkg_from_page(struct bio *bio, struct page *page) |
852c788f | 2077 | { |
6a7f6d86 DZ |
2078 | struct cgroup_subsys_state *css; |
2079 | ||
6a7f6d86 DZ |
2080 | if (!page->mem_cgroup) |
2081 | return; | |
2082 | ||
fc5a828b DZ |
2083 | rcu_read_lock(); |
2084 | ||
2085 | css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); | |
2086 | bio_associate_blkg_from_css(bio, css); | |
2087 | ||
2088 | rcu_read_unlock(); | |
6a7f6d86 DZ |
2089 | } |
2090 | #endif /* CONFIG_MEMCG */ | |
2091 | ||
2268c0fe DZ |
2092 | /** |
2093 | * bio_associate_blkg - associate a bio with a blkg | |
2094 | * @bio: target bio | |
2095 | * | |
2096 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
2097 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
2098 | * already associated, the css is reused and association redone as the | |
2099 | * request_queue may have changed. | |
2100 | */ | |
2101 | void bio_associate_blkg(struct bio *bio) | |
2102 | { | |
fc5a828b | 2103 | struct cgroup_subsys_state *css; |
2268c0fe DZ |
2104 | |
2105 | rcu_read_lock(); | |
2106 | ||
db6638d7 | 2107 | if (bio->bi_blkg) |
fc5a828b | 2108 | css = &bio_blkcg(bio)->css; |
db6638d7 | 2109 | else |
fc5a828b | 2110 | css = blkcg_css(); |
2268c0fe | 2111 | |
fc5a828b | 2112 | bio_associate_blkg_from_css(bio, css); |
2268c0fe DZ |
2113 | |
2114 | rcu_read_unlock(); | |
852c788f | 2115 | } |
5cdf2e3f | 2116 | EXPORT_SYMBOL_GPL(bio_associate_blkg); |
852c788f | 2117 | |
20bd723e | 2118 | /** |
db6638d7 | 2119 | * bio_clone_blkg_association - clone blkg association from src to dst bio |
20bd723e PV |
2120 | * @dst: destination bio |
2121 | * @src: source bio | |
2122 | */ | |
db6638d7 | 2123 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) |
20bd723e | 2124 | { |
6ab21879 DZ |
2125 | rcu_read_lock(); |
2126 | ||
fc5a828b | 2127 | if (src->bi_blkg) |
2268c0fe | 2128 | __bio_associate_blkg(dst, src->bi_blkg); |
6ab21879 DZ |
2129 | |
2130 | rcu_read_unlock(); | |
20bd723e | 2131 | } |
db6638d7 | 2132 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); |
852c788f TH |
2133 | #endif /* CONFIG_BLK_CGROUP */ |
2134 | ||
1da177e4 LT |
2135 | static void __init biovec_init_slabs(void) |
2136 | { | |
2137 | int i; | |
2138 | ||
ed996a52 | 2139 | for (i = 0; i < BVEC_POOL_NR; i++) { |
1da177e4 LT |
2140 | int size; |
2141 | struct biovec_slab *bvs = bvec_slabs + i; | |
2142 | ||
a7fcd37c JA |
2143 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
2144 | bvs->slab = NULL; | |
2145 | continue; | |
2146 | } | |
a7fcd37c | 2147 | |
1da177e4 LT |
2148 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
2149 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 2150 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
2151 | } |
2152 | } | |
2153 | ||
2154 | static int __init init_bio(void) | |
2155 | { | |
bb799ca0 JA |
2156 | bio_slab_max = 2; |
2157 | bio_slab_nr = 0; | |
6396bb22 KC |
2158 | bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), |
2159 | GFP_KERNEL); | |
2b24e6f6 JT |
2160 | |
2161 | BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); | |
2162 | ||
bb799ca0 JA |
2163 | if (!bio_slabs) |
2164 | panic("bio: can't allocate bios\n"); | |
1da177e4 | 2165 | |
7878cba9 | 2166 | bio_integrity_init(); |
1da177e4 LT |
2167 | biovec_init_slabs(); |
2168 | ||
f4f8154a | 2169 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
1da177e4 LT |
2170 | panic("bio: can't allocate bios\n"); |
2171 | ||
f4f8154a | 2172 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
a91a2785 MP |
2173 | panic("bio: can't create integrity pool\n"); |
2174 | ||
1da177e4 LT |
2175 | return 0; |
2176 | } | |
1da177e4 | 2177 | subsys_initcall(init_bio); |