blob: aa7354088008ba5f73d3ba1af83eb902b0629ec6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
16 *
17 */
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070022#include <linux/uio.h>
Tejun Heo852c7882012-03-05 13:15:27 -080023#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050027#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/mempool.h>
29#include <linux/workqueue.h>
Tejun Heo852c7882012-03-05 13:15:27 -080030#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Li Zefan55782132009-06-09 13:43:05 +080032#include <trace/events/block.h>
Ingo Molnar0bfc2452008-11-26 11:59:56 +010033
Jens Axboe392ddc32008-12-23 12:42:54 +010034/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
43 * unsigned short
44 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Christoph Hellwiged996a52016-07-19 11:28:42 +020046static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48};
49#undef BV
50
51/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
54 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020055struct bio_set *fs_bio_set;
Kent Overstreet3f86a822012-09-06 15:35:01 -070056EXPORT_SYMBOL(fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboebb799ca2008-12-10 15:35:05 +010058/*
59 * Our slab pool management
60 */
61struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66};
67static DEFINE_MUTEX(bio_slab_lock);
68static struct bio_slab *bio_slabs;
69static unsigned int bio_slab_nr, bio_slab_max;
70
71static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72{
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020075 struct bio_slab *bslab, *new_bio_slabs;
Anna Leuschner386bc352012-10-22 21:53:36 +020076 unsigned int new_bio_slab_max;
Jens Axboebb799ca2008-12-10 15:35:05 +010077 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010083 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010084
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
Anna Leuschner386bc352012-10-22 21:53:36 +020099 new_bio_slab_max = bio_slab_max << 1;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200100 new_bio_slabs = krealloc(bio_slabs,
Anna Leuschner386bc352012-10-22 21:53:36 +0200101 new_bio_slab_max * sizeof(struct bio_slab),
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200102 GFP_KERNEL);
103 if (!new_bio_slabs)
Jens Axboebb799ca2008-12-10 15:35:05 +0100104 goto out_unlock;
Anna Leuschner386bc352012-10-22 21:53:36 +0200105 bio_slab_max = new_bio_slab_max;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200106 bio_slabs = new_bio_slabs;
Jens Axboebb799ca2008-12-10 15:35:05 +0100107 }
108 if (entry == -1)
109 entry = bio_slab_nr++;
110
111 bslab = &bio_slabs[entry];
112
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
Mikulas Patocka6a241482014-03-28 15:51:55 -0400114 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 SLAB_HWCACHE_ALIGN, NULL);
Jens Axboebb799ca2008-12-10 15:35:05 +0100116 if (!slab)
117 goto out_unlock;
118
Jens Axboebb799ca2008-12-10 15:35:05 +0100119 bslab->slab = slab;
120 bslab->slab_ref = 1;
121 bslab->slab_size = sz;
122out_unlock:
123 mutex_unlock(&bio_slab_lock);
124 return slab;
125}
126
127static void bio_put_slab(struct bio_set *bs)
128{
129 struct bio_slab *bslab = NULL;
130 unsigned int i;
131
132 mutex_lock(&bio_slab_lock);
133
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
137 break;
138 }
139 }
140
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 goto out;
143
144 WARN_ON(!bslab->slab_ref);
145
146 if (--bslab->slab_ref)
147 goto out;
148
149 kmem_cache_destroy(bslab->slab);
150 bslab->slab = NULL;
151
152out:
153 mutex_unlock(&bio_slab_lock);
154}
155
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200156unsigned int bvec_nr_vecs(unsigned short idx)
157{
158 return bvec_slabs[idx].nr_vecs;
159}
160
Kent Overstreet9f060e22012-10-12 15:29:33 -0700161void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
Jens Axboebb799ca2008-12-10 15:35:05 +0100162{
Christoph Hellwiged996a52016-07-19 11:28:42 +0200163 if (!idx)
164 return;
165 idx--;
Jens Axboebb799ca2008-12-10 15:35:05 +0100166
Christoph Hellwiged996a52016-07-19 11:28:42 +0200167 BIO_BUG_ON(idx >= BVEC_POOL_NR);
168
169 if (idx == BVEC_POOL_MAX) {
Kent Overstreet9f060e22012-10-12 15:29:33 -0700170 mempool_free(bv, pool);
Christoph Hellwiged996a52016-07-19 11:28:42 +0200171 } else {
Jens Axboebb799ca2008-12-10 15:35:05 +0100172 struct biovec_slab *bvs = bvec_slabs + idx;
173
174 kmem_cache_free(bvs->slab, bv);
175 }
176}
177
Kent Overstreet9f060e22012-10-12 15:29:33 -0700178struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
179 mempool_t *pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
181 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
183 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100184 * see comment near bvec_array define!
185 */
186 switch (nr) {
187 case 1:
188 *idx = 0;
189 break;
190 case 2 ... 4:
191 *idx = 1;
192 break;
193 case 5 ... 16:
194 *idx = 2;
195 break;
196 case 17 ... 64:
197 *idx = 3;
198 break;
199 case 65 ... 128:
200 *idx = 4;
201 break;
202 case 129 ... BIO_MAX_PAGES:
203 *idx = 5;
204 break;
205 default:
206 return NULL;
207 }
208
209 /*
210 * idx now points to the pool we want to allocate from. only the
211 * 1-vec entry pool is mempool backed.
212 */
Christoph Hellwiged996a52016-07-19 11:28:42 +0200213 if (*idx == BVEC_POOL_MAX) {
Jens Axboe7ff93452008-12-11 11:53:43 +0100214fallback:
Kent Overstreet9f060e22012-10-12 15:29:33 -0700215 bvl = mempool_alloc(pool, gfp_mask);
Jens Axboe7ff93452008-12-11 11:53:43 +0100216 } else {
217 struct biovec_slab *bvs = bvec_slabs + *idx;
Mel Gormand0164ad2015-11-06 16:28:21 -0800218 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200220 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100221 * Make this allocation restricted and don't dump info on
222 * allocation failures, since we'll fallback to the mempool
223 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200224 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100225 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
226
227 /*
Mel Gormand0164ad2015-11-06 16:28:21 -0800228 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
Jens Axboe7ff93452008-12-11 11:53:43 +0100229 * is set, retry with the 1-entry mempool
230 */
231 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
Mel Gormand0164ad2015-11-06 16:28:21 -0800232 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
Christoph Hellwiged996a52016-07-19 11:28:42 +0200233 *idx = BVEC_POOL_MAX;
Jens Axboe7ff93452008-12-11 11:53:43 +0100234 goto fallback;
235 }
236 }
237
Christoph Hellwiged996a52016-07-19 11:28:42 +0200238 (*idx)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return bvl;
240}
241
Kent Overstreet4254bba2012-09-06 15:35:00 -0700242static void __bio_free(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
Kent Overstreet4254bba2012-09-06 15:35:00 -0700244 bio_disassociate_task(bio);
Jens Axboe992c5dd2007-07-18 13:18:08 +0200245
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200246 if (bio_integrity(bio))
Kent Overstreet1e2a410f2012-09-06 15:34:56 -0700247 bio_integrity_free(bio);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700248}
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200249
Kent Overstreet4254bba2012-09-06 15:35:00 -0700250static void bio_free(struct bio *bio)
251{
252 struct bio_set *bs = bio->bi_pool;
253 void *p;
254
255 __bio_free(bio);
256
257 if (bs) {
Christoph Hellwiged996a52016-07-19 11:28:42 +0200258 bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
Kent Overstreet4254bba2012-09-06 15:35:00 -0700259
260 /*
261 * If we have front padding, adjust the bio pointer before freeing
262 */
263 p = bio;
Jens Axboebb799ca2008-12-10 15:35:05 +0100264 p -= bs->front_pad;
265
Kent Overstreet4254bba2012-09-06 15:35:00 -0700266 mempool_free(p, bs->bio_pool);
267 } else {
268 /* Bio was allocated by bio_kmalloc() */
269 kfree(bio);
270 }
Peter Osterlund36763472005-09-06 15:16:42 -0700271}
272
Arjan van de Ven858119e2006-01-14 13:20:43 -0800273void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Jens Axboe2b94de52007-07-18 13:14:03 +0200275 memset(bio, 0, sizeof(*bio));
Jens Axboec4cf5262015-04-17 16:15:18 -0600276 atomic_set(&bio->__bi_remaining, 1);
Jens Axboedac56212015-04-17 16:23:59 -0600277 atomic_set(&bio->__bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200279EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281/**
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700282 * bio_reset - reinitialize a bio
283 * @bio: bio to reset
284 *
285 * Description:
286 * After calling bio_reset(), @bio will be in the same state as a freshly
287 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
288 * preserved are the ones that are initialized by bio_alloc_bioset(). See
289 * comment in struct bio.
290 */
291void bio_reset(struct bio *bio)
292{
293 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
294
Kent Overstreet4254bba2012-09-06 15:35:00 -0700295 __bio_free(bio);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700296
297 memset(bio, 0, BIO_RESET_BYTES);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200298 bio->bi_flags = flags;
Jens Axboec4cf5262015-04-17 16:15:18 -0600299 atomic_set(&bio->__bi_remaining, 1);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700300}
301EXPORT_SYMBOL(bio_reset);
302
Christoph Hellwig38f8baa2016-03-11 17:34:51 +0100303static struct bio *__bio_chain_endio(struct bio *bio)
Kent Overstreet196d38b2013-11-23 18:34:15 -0800304{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200305 struct bio *parent = bio->bi_private;
306
Christoph Hellwigaf3e3a52016-03-11 17:34:50 +0100307 if (!parent->bi_error)
308 parent->bi_error = bio->bi_error;
Kent Overstreet196d38b2013-11-23 18:34:15 -0800309 bio_put(bio);
Christoph Hellwig38f8baa2016-03-11 17:34:51 +0100310 return parent;
311}
312
313static void bio_chain_endio(struct bio *bio)
314{
315 bio_endio(__bio_chain_endio(bio));
Kent Overstreet196d38b2013-11-23 18:34:15 -0800316}
317
318/**
319 * bio_chain - chain bio completions
Randy Dunlap1051a902014-04-20 16:03:31 -0700320 * @bio: the target bio
321 * @parent: the @bio's parent bio
Kent Overstreet196d38b2013-11-23 18:34:15 -0800322 *
323 * The caller won't have a bi_end_io called when @bio completes - instead,
324 * @parent's bi_end_io won't be called until both @parent and @bio have
325 * completed; the chained bio will also be freed when it completes.
326 *
327 * The caller must not set bi_private or bi_end_io in @bio.
328 */
329void bio_chain(struct bio *bio, struct bio *parent)
330{
331 BUG_ON(bio->bi_private || bio->bi_end_io);
332
333 bio->bi_private = parent;
334 bio->bi_end_io = bio_chain_endio;
Jens Axboec4cf5262015-04-17 16:15:18 -0600335 bio_inc_remaining(parent);
Kent Overstreet196d38b2013-11-23 18:34:15 -0800336}
337EXPORT_SYMBOL(bio_chain);
338
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700339static void bio_alloc_rescue(struct work_struct *work)
340{
341 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
342 struct bio *bio;
343
344 while (1) {
345 spin_lock(&bs->rescue_lock);
346 bio = bio_list_pop(&bs->rescue_list);
347 spin_unlock(&bs->rescue_lock);
348
349 if (!bio)
350 break;
351
352 generic_make_request(bio);
353 }
354}
355
356static void punt_bios_to_rescuer(struct bio_set *bs)
357{
358 struct bio_list punt, nopunt;
359 struct bio *bio;
360
361 /*
362 * In order to guarantee forward progress we must punt only bios that
363 * were allocated from this bio_set; otherwise, if there was a bio on
364 * there for a stacking driver higher up in the stack, processing it
365 * could require allocating bios from this bio_set, and doing that from
366 * our own rescuer would be bad.
367 *
368 * Since bio lists are singly linked, pop them all instead of trying to
369 * remove from the middle of the list:
370 */
371
372 bio_list_init(&punt);
373 bio_list_init(&nopunt);
374
375 while ((bio = bio_list_pop(current->bio_list)))
376 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
377
378 *current->bio_list = nopunt;
379
380 spin_lock(&bs->rescue_lock);
381 bio_list_merge(&bs->rescue_list, &punt);
382 spin_unlock(&bs->rescue_lock);
383
384 queue_work(bs->rescue_workqueue, &bs->rescue_work);
385}
386
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700387/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 * bio_alloc_bioset - allocate a bio for I/O
389 * @gfp_mask: the GFP_ mask given to the slab allocator
390 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200391 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 *
393 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700394 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
395 * backed by the @bs's mempool.
396 *
Mel Gormand0164ad2015-11-06 16:28:21 -0800397 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
398 * always be able to allocate a bio. This is due to the mempool guarantees.
399 * To make this work, callers must never allocate more than 1 bio at a time
400 * from this pool. Callers that need to allocate more than 1 bio must always
401 * submit the previously allocated bio for IO before attempting to allocate
402 * a new one. Failure to do so can cause deadlocks under memory pressure.
Kent Overstreet3f86a822012-09-06 15:35:01 -0700403 *
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700404 * Note that when running under generic_make_request() (i.e. any block
405 * driver), bios are not submitted until after you return - see the code in
406 * generic_make_request() that converts recursion into iteration, to prevent
407 * stack overflows.
408 *
409 * This would normally mean allocating multiple bios under
410 * generic_make_request() would be susceptible to deadlocks, but we have
411 * deadlock avoidance code that resubmits any blocked bios from a rescuer
412 * thread.
413 *
414 * However, we do not guarantee forward progress for allocations from other
415 * mempools. Doing multiple allocations from the same mempool under
416 * generic_make_request() should be avoided - instead, use bio_set's front_pad
417 * for per bio allocations.
418 *
Kent Overstreet3f86a822012-09-06 15:35:01 -0700419 * RETURNS:
420 * Pointer to new bio on success, NULL on failure.
421 */
Al Virodd0fc662005-10-07 07:46:04 +0100422struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700424 gfp_t saved_gfp = gfp_mask;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700425 unsigned front_pad;
426 unsigned inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100427 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200428 struct bio *bio;
429 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200430
Kent Overstreet3f86a822012-09-06 15:35:01 -0700431 if (!bs) {
432 if (nr_iovecs > UIO_MAXIOV)
433 return NULL;
434
435 p = kmalloc(sizeof(struct bio) +
436 nr_iovecs * sizeof(struct bio_vec),
437 gfp_mask);
438 front_pad = 0;
439 inline_vecs = nr_iovecs;
440 } else {
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400441 /* should not use nobvec bioset for nr_iovecs > 0 */
442 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
443 return NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700444 /*
445 * generic_make_request() converts recursion to iteration; this
446 * means if we're running beneath it, any bios we allocate and
447 * submit will not be submitted (and thus freed) until after we
448 * return.
449 *
450 * This exposes us to a potential deadlock if we allocate
451 * multiple bios from the same bio_set() while running
452 * underneath generic_make_request(). If we were to allocate
453 * multiple bios (say a stacking block driver that was splitting
454 * bios), we would deadlock if we exhausted the mempool's
455 * reserve.
456 *
457 * We solve this, and guarantee forward progress, with a rescuer
458 * workqueue per bio_set. If we go to allocate and there are
459 * bios on current->bio_list, we first try the allocation
Mel Gormand0164ad2015-11-06 16:28:21 -0800460 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
461 * bios we would be blocking to the rescuer workqueue before
462 * we retry with the original gfp_flags.
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700463 */
464
465 if (current->bio_list && !bio_list_empty(current->bio_list))
Mel Gormand0164ad2015-11-06 16:28:21 -0800466 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700467
Kent Overstreet3f86a822012-09-06 15:35:01 -0700468 p = mempool_alloc(bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700469 if (!p && gfp_mask != saved_gfp) {
470 punt_bios_to_rescuer(bs);
471 gfp_mask = saved_gfp;
472 p = mempool_alloc(bs->bio_pool, gfp_mask);
473 }
474
Kent Overstreet3f86a822012-09-06 15:35:01 -0700475 front_pad = bs->front_pad;
476 inline_vecs = BIO_INLINE_VECS;
477 }
478
Tejun Heo451a9eb2009-04-15 19:50:51 +0200479 if (unlikely(!p))
480 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100481
Kent Overstreet3f86a822012-09-06 15:35:01 -0700482 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100483 bio_init(bio);
484
Kent Overstreet3f86a822012-09-06 15:35:01 -0700485 if (nr_iovecs > inline_vecs) {
Christoph Hellwiged996a52016-07-19 11:28:42 +0200486 unsigned long idx = 0;
487
Kent Overstreet9f060e22012-10-12 15:29:33 -0700488 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700489 if (!bvl && gfp_mask != saved_gfp) {
490 punt_bios_to_rescuer(bs);
491 gfp_mask = saved_gfp;
Kent Overstreet9f060e22012-10-12 15:29:33 -0700492 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700493 }
494
Ingo Molnar34053972009-02-21 11:16:36 +0100495 if (unlikely(!bvl))
496 goto err_free;
Kent Overstreeta38352e2012-05-25 13:03:11 -0700497
Christoph Hellwiged996a52016-07-19 11:28:42 +0200498 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700499 } else if (nr_iovecs) {
500 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100501 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700502
503 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100504 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100505 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100507
508err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200509 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100510 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200512EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514void zero_fill_bio(struct bio *bio)
515{
516 unsigned long flags;
Kent Overstreet79886132013-11-23 17:19:00 -0800517 struct bio_vec bv;
518 struct bvec_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Kent Overstreet79886132013-11-23 17:19:00 -0800520 bio_for_each_segment(bv, bio, iter) {
521 char *data = bvec_kmap_irq(&bv, &flags);
522 memset(data, 0, bv.bv_len);
523 flush_dcache_page(bv.bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 bvec_kunmap_irq(data, &flags);
525 }
526}
527EXPORT_SYMBOL(zero_fill_bio);
528
529/**
530 * bio_put - release a reference to a bio
531 * @bio: bio to release reference to
532 *
533 * Description:
534 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100535 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 **/
537void bio_put(struct bio *bio)
538{
Jens Axboedac56212015-04-17 16:23:59 -0600539 if (!bio_flagged(bio, BIO_REFFED))
Kent Overstreet4254bba2012-09-06 15:35:00 -0700540 bio_free(bio);
Jens Axboedac56212015-04-17 16:23:59 -0600541 else {
542 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
543
544 /*
545 * last put frees it
546 */
547 if (atomic_dec_and_test(&bio->__bi_cnt))
548 bio_free(bio);
549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200551EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Jens Axboe165125e2007-07-24 09:28:11 +0200553inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
556 blk_recount_segments(q, bio);
557
558 return bio->bi_phys_segments;
559}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200560EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800563 * __bio_clone_fast - clone a bio that shares the original bio's biovec
564 * @bio: destination bio
565 * @bio_src: bio to clone
566 *
567 * Clone a &bio. Caller will own the returned bio, but not
568 * the actual data it points to. Reference count of returned
569 * bio will be one.
570 *
571 * Caller must ensure that @bio_src is not freed before @bio.
572 */
573void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
574{
Christoph Hellwiged996a52016-07-19 11:28:42 +0200575 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
Kent Overstreet59d276f2013-11-23 18:19:27 -0800576
577 /*
578 * most users will be overriding ->bi_bdev with a new target,
579 * so we don't set nor calculate new physical/hw segment counts here
580 */
581 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600582 bio_set_flag(bio, BIO_CLONED);
Jens Axboe1eff9d32016-08-05 15:35:16 -0600583 bio->bi_opf = bio_src->bi_opf;
Kent Overstreet59d276f2013-11-23 18:19:27 -0800584 bio->bi_iter = bio_src->bi_iter;
585 bio->bi_io_vec = bio_src->bi_io_vec;
Paolo Valente20bd7232016-07-27 07:22:05 +0200586
587 bio_clone_blkcg_association(bio, bio_src);
Kent Overstreet59d276f2013-11-23 18:19:27 -0800588}
589EXPORT_SYMBOL(__bio_clone_fast);
590
591/**
592 * bio_clone_fast - clone a bio that shares the original bio's biovec
593 * @bio: bio to clone
594 * @gfp_mask: allocation priority
595 * @bs: bio_set to allocate from
596 *
597 * Like __bio_clone_fast, only also allocates the returned bio
598 */
599struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
600{
601 struct bio *b;
602
603 b = bio_alloc_bioset(gfp_mask, 0, bs);
604 if (!b)
605 return NULL;
606
607 __bio_clone_fast(b, bio);
608
609 if (bio_integrity(bio)) {
610 int ret;
611
612 ret = bio_integrity_clone(b, bio, gfp_mask);
613
614 if (ret < 0) {
615 bio_put(b);
616 return NULL;
617 }
618 }
619
620 return b;
621}
622EXPORT_SYMBOL(bio_clone_fast);
623
624/**
Kent Overstreetbdb53202013-11-23 17:26:46 -0800625 * bio_clone_bioset - clone a bio
626 * @bio_src: bio to clone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700628 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 *
Kent Overstreetbdb53202013-11-23 17:26:46 -0800630 * Clone bio. Caller will own the returned bio, but not the actual data it
631 * points to. Reference count of returned bio will be one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 */
Kent Overstreetbdb53202013-11-23 17:26:46 -0800633struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700634 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
Kent Overstreetbdb53202013-11-23 17:26:46 -0800636 struct bvec_iter iter;
637 struct bio_vec bv;
638 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Kent Overstreetbdb53202013-11-23 17:26:46 -0800640 /*
641 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
642 * bio_src->bi_io_vec to bio->bi_io_vec.
643 *
644 * We can't do that anymore, because:
645 *
646 * - The point of cloning the biovec is to produce a bio with a biovec
647 * the caller can modify: bi_idx and bi_bvec_done should be 0.
648 *
649 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
650 * we tried to clone the whole thing bio_alloc_bioset() would fail.
651 * But the clone should succeed as long as the number of biovecs we
652 * actually need to allocate is fewer than BIO_MAX_PAGES.
653 *
654 * - Lastly, bi_vcnt should not be looked at or relied upon by code
655 * that does not own the bio - reason being drivers don't use it for
656 * iterating over the biovec anymore, so expecting it to be kept up
657 * to date (i.e. for clones that share the parent biovec) is just
658 * asking for trouble and would force extra work on
659 * __bio_clone_fast() anyways.
660 */
661
Kent Overstreet8423ae32014-02-10 17:45:50 -0800662 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
Kent Overstreetbdb53202013-11-23 17:26:46 -0800663 if (!bio)
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200664 return NULL;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800665 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600666 bio->bi_opf = bio_src->bi_opf;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800667 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
668 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200669
Adrian Hunter7afafc82016-08-16 10:59:35 +0300670 switch (bio_op(bio)) {
671 case REQ_OP_DISCARD:
672 case REQ_OP_SECURE_ERASE:
673 break;
674 case REQ_OP_WRITE_SAME:
Kent Overstreet8423ae32014-02-10 17:45:50 -0800675 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
Adrian Hunter7afafc82016-08-16 10:59:35 +0300676 break;
677 default:
678 bio_for_each_segment(bv, bio_src, iter)
679 bio->bi_io_vec[bio->bi_vcnt++] = bv;
680 break;
Kent Overstreet8423ae32014-02-10 17:45:50 -0800681 }
682
Kent Overstreetbdb53202013-11-23 17:26:46 -0800683 if (bio_integrity(bio_src)) {
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200684 int ret;
685
Kent Overstreetbdb53202013-11-23 17:26:46 -0800686 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
Li Zefan059ea332009-03-09 10:42:45 +0100687 if (ret < 0) {
Kent Overstreetbdb53202013-11-23 17:26:46 -0800688 bio_put(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200689 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100690 }
Peter Osterlund36763472005-09-06 15:16:42 -0700691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Paolo Valente20bd7232016-07-27 07:22:05 +0200693 bio_clone_blkcg_association(bio, bio_src);
694
Kent Overstreetbdb53202013-11-23 17:26:46 -0800695 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700697EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699/**
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800700 * bio_add_pc_page - attempt to add page to bio
701 * @q: the target queue
702 * @bio: destination bio
703 * @page: page to add
704 * @len: vec entry length
705 * @offset: vec entry offset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800707 * Attempt to add a page to the bio_vec maplist. This can fail for a
708 * number of reasons, such as the bio being full or target block device
709 * limitations. The target block device must allow bio's up to PAGE_SIZE,
710 * so it is always possible to add a single page to an empty bio.
711 *
712 * This should only be used by REQ_PC bios.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800714int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
715 *page, unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
717 int retried_segments = 0;
718 struct bio_vec *bvec;
719
720 /*
721 * cloned bio must not modify vec list
722 */
723 if (unlikely(bio_flagged(bio, BIO_CLONED)))
724 return 0;
725
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800726 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return 0;
728
Jens Axboe80cfd542006-01-06 09:43:28 +0100729 /*
730 * For filesystems with a blocksize smaller than the pagesize
731 * we will often be called with the same page as last time and
732 * a consecutive offset. Optimize this special case.
733 */
734 if (bio->bi_vcnt > 0) {
735 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
736
737 if (page == prev->bv_page &&
738 offset == prev->bv_offset + prev->bv_len) {
739 prev->bv_len += len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800740 bio->bi_iter.bi_size += len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100741 goto done;
742 }
Jens Axboe66cb45a2014-06-24 16:22:24 -0600743
744 /*
745 * If the queue doesn't support SG gaps and adding this
746 * offset would create a gap, disallow it.
747 */
Keith Busch03100aa2015-08-19 14:24:05 -0700748 if (bvec_gap_to_prev(q, prev, offset))
Jens Axboe66cb45a2014-06-24 16:22:24 -0600749 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100750 }
751
752 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 return 0;
754
755 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 * setup the new entry, we might clear it again later if we
757 * cannot add the page
758 */
759 bvec = &bio->bi_io_vec[bio->bi_vcnt];
760 bvec->bv_page = page;
761 bvec->bv_len = len;
762 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800763 bio->bi_vcnt++;
764 bio->bi_phys_segments++;
765 bio->bi_iter.bi_size += len;
766
767 /*
768 * Perform a recount if the number of segments is greater
769 * than queue_max_segments(q).
770 */
771
772 while (bio->bi_phys_segments > queue_max_segments(q)) {
773
774 if (retried_segments)
775 goto failed;
776
777 retried_segments = 1;
778 blk_recount_segments(q, bio);
779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 /* If we may be able to merge these biovecs, force a recount */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800782 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600783 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Jens Axboe80cfd542006-01-06 09:43:28 +0100785 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 return len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800787
788 failed:
789 bvec->bv_page = NULL;
790 bvec->bv_len = 0;
791 bvec->bv_offset = 0;
792 bio->bi_vcnt--;
793 bio->bi_iter.bi_size -= len;
794 blk_recount_segments(q, bio);
795 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200797EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600798
799/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 * bio_add_page - attempt to add page to bio
801 * @bio: destination bio
802 * @page: page to add
803 * @len: vec entry length
804 * @offset: vec entry offset
805 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800806 * Attempt to add a page to the bio_vec maplist. This will only fail
807 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800809int bio_add_page(struct bio *bio, struct page *page,
810 unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811{
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800812 struct bio_vec *bv;
Jens Axboe762380a2014-06-05 13:38:39 -0600813
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800814 /*
815 * cloned bio must not modify vec list
816 */
817 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
818 return 0;
Jens Axboe58a49152014-06-10 12:53:56 -0600819
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800820 /*
821 * For filesystems with a blocksize smaller than the pagesize
822 * we will often be called with the same page as last time and
823 * a consecutive offset. Optimize this special case.
824 */
825 if (bio->bi_vcnt > 0) {
826 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
827
828 if (page == bv->bv_page &&
829 offset == bv->bv_offset + bv->bv_len) {
830 bv->bv_len += len;
831 goto done;
832 }
833 }
834
835 if (bio->bi_vcnt >= bio->bi_max_vecs)
836 return 0;
837
838 bv = &bio->bi_io_vec[bio->bi_vcnt];
839 bv->bv_page = page;
840 bv->bv_len = len;
841 bv->bv_offset = offset;
842
843 bio->bi_vcnt++;
844done:
845 bio->bi_iter.bi_size += len;
846 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200848EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Kent Overstreet9e882242012-09-10 14:41:12 -0700850struct submit_bio_ret {
851 struct completion event;
852 int error;
853};
854
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200855static void submit_bio_wait_endio(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700856{
857 struct submit_bio_ret *ret = bio->bi_private;
858
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200859 ret->error = bio->bi_error;
Kent Overstreet9e882242012-09-10 14:41:12 -0700860 complete(&ret->event);
861}
862
863/**
864 * submit_bio_wait - submit a bio, and wait until it completes
Kent Overstreet9e882242012-09-10 14:41:12 -0700865 * @bio: The &struct bio which describes the I/O
866 *
867 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
868 * bio_endio() on failure.
869 */
Mike Christie4e49ea42016-06-05 14:31:41 -0500870int submit_bio_wait(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700871{
872 struct submit_bio_ret ret;
873
Kent Overstreet9e882242012-09-10 14:41:12 -0700874 init_completion(&ret.event);
875 bio->bi_private = &ret;
876 bio->bi_end_io = submit_bio_wait_endio;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600877 bio->bi_opf |= REQ_SYNC;
Mike Christie4e49ea42016-06-05 14:31:41 -0500878 submit_bio(bio);
Stephane Gasparinid57d6112016-02-09 17:07:38 +0100879 wait_for_completion_io(&ret.event);
Kent Overstreet9e882242012-09-10 14:41:12 -0700880
881 return ret.error;
882}
883EXPORT_SYMBOL(submit_bio_wait);
884
Kent Overstreet054bdf62012-09-28 13:17:55 -0700885/**
886 * bio_advance - increment/complete a bio by some number of bytes
887 * @bio: bio to advance
888 * @bytes: number of bytes to complete
889 *
890 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
891 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
892 * be updated on the last bvec as well.
893 *
894 * @bio will then represent the remaining, uncompleted portion of the io.
895 */
896void bio_advance(struct bio *bio, unsigned bytes)
897{
898 if (bio_integrity(bio))
899 bio_integrity_advance(bio, bytes);
900
Kent Overstreet4550dd62013-08-07 14:26:21 -0700901 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -0700902}
903EXPORT_SYMBOL(bio_advance);
904
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700905/**
Kent Overstreeta0787602012-09-10 14:03:28 -0700906 * bio_alloc_pages - allocates a single page for each bvec in a bio
907 * @bio: bio to allocate pages for
908 * @gfp_mask: flags for allocation
909 *
910 * Allocates pages up to @bio->bi_vcnt.
911 *
912 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
913 * freed.
914 */
915int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
916{
917 int i;
918 struct bio_vec *bv;
919
920 bio_for_each_segment_all(bv, bio, i) {
921 bv->bv_page = alloc_page(gfp_mask);
922 if (!bv->bv_page) {
923 while (--bv >= bio->bi_io_vec)
924 __free_page(bv->bv_page);
925 return -ENOMEM;
926 }
927 }
928
929 return 0;
930}
931EXPORT_SYMBOL(bio_alloc_pages);
932
933/**
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700934 * bio_copy_data - copy contents of data buffers from one chain of bios to
935 * another
936 * @src: source bio list
937 * @dst: destination bio list
938 *
939 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
940 * @src and @dst as linked lists of bios.
941 *
942 * Stops when it reaches the end of either @src or @dst - that is, copies
943 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
944 */
945void bio_copy_data(struct bio *dst, struct bio *src)
946{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700947 struct bvec_iter src_iter, dst_iter;
948 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700949 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700950 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700951
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700952 src_iter = src->bi_iter;
953 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700954
955 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700956 if (!src_iter.bi_size) {
957 src = src->bi_next;
958 if (!src)
959 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700960
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700961 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700962 }
963
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700964 if (!dst_iter.bi_size) {
965 dst = dst->bi_next;
966 if (!dst)
967 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700968
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700969 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700970 }
971
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700972 src_bv = bio_iter_iovec(src, src_iter);
973 dst_bv = bio_iter_iovec(dst, dst_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700974
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700975 bytes = min(src_bv.bv_len, dst_bv.bv_len);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700976
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700977 src_p = kmap_atomic(src_bv.bv_page);
978 dst_p = kmap_atomic(dst_bv.bv_page);
979
980 memcpy(dst_p + dst_bv.bv_offset,
981 src_p + src_bv.bv_offset,
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700982 bytes);
983
984 kunmap_atomic(dst_p);
985 kunmap_atomic(src_p);
986
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700987 bio_advance_iter(src, &src_iter, bytes);
988 bio_advance_iter(dst, &dst_iter, bytes);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700989 }
990}
991EXPORT_SYMBOL(bio_copy_data);
992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900994 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +0100995 struct iov_iter iter;
996 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997};
998
Fabian Frederick7410b3c2014-04-22 15:09:07 -0600999static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001000 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001{
Jens Axboef3f63c12010-10-29 11:46:56 -06001002 if (iov_count > UIO_MAXIOV)
1003 return NULL;
1004
Kent Overstreetc8db4442013-11-22 19:39:06 -08001005 return kmalloc(sizeof(struct bio_map_data) +
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001006 sizeof(struct iovec) * iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007}
1008
Dongsu Park9124d3f2015-01-18 16:16:34 +01001009/**
1010 * bio_copy_from_iter - copy all pages from iov_iter to bio
1011 * @bio: The &struct bio which describes the I/O as destination
1012 * @iter: iov_iter as source
1013 *
1014 * Copy all pages from iov_iter to bio.
1015 * Returns 0 on success, or error on failure.
1016 */
1017static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001018{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001019 int i;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001020 struct bio_vec *bvec;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001021
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001022 bio_for_each_segment_all(bvec, bio, i) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001023 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001024
Dongsu Park9124d3f2015-01-18 16:16:34 +01001025 ret = copy_page_from_iter(bvec->bv_page,
1026 bvec->bv_offset,
1027 bvec->bv_len,
1028 &iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001029
Dongsu Park9124d3f2015-01-18 16:16:34 +01001030 if (!iov_iter_count(&iter))
1031 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001032
Dongsu Park9124d3f2015-01-18 16:16:34 +01001033 if (ret < bvec->bv_len)
1034 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001035 }
1036
Dongsu Park9124d3f2015-01-18 16:16:34 +01001037 return 0;
1038}
1039
1040/**
1041 * bio_copy_to_iter - copy all pages from bio to iov_iter
1042 * @bio: The &struct bio which describes the I/O as source
1043 * @iter: iov_iter as destination
1044 *
1045 * Copy all pages from bio to iov_iter.
1046 * Returns 0 on success, or error on failure.
1047 */
1048static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1049{
1050 int i;
1051 struct bio_vec *bvec;
1052
1053 bio_for_each_segment_all(bvec, bio, i) {
1054 ssize_t ret;
1055
1056 ret = copy_page_to_iter(bvec->bv_page,
1057 bvec->bv_offset,
1058 bvec->bv_len,
1059 &iter);
1060
1061 if (!iov_iter_count(&iter))
1062 break;
1063
1064 if (ret < bvec->bv_len)
1065 return -EFAULT;
1066 }
1067
1068 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001069}
1070
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001071static void bio_free_pages(struct bio *bio)
1072{
1073 struct bio_vec *bvec;
1074 int i;
1075
1076 bio_for_each_segment_all(bvec, bio, i)
1077 __free_page(bvec->bv_page);
1078}
1079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080/**
1081 * bio_uncopy_user - finish previously mapped bio
1082 * @bio: bio being terminated
1083 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001084 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 * to user space in case of a read.
1086 */
1087int bio_uncopy_user(struct bio *bio)
1088{
1089 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001090 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
Roland Dreier35dc2482013-08-05 17:55:01 -07001092 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1093 /*
1094 * if we're in a workqueue, the request is orphaned, so
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001095 * don't copy into a random user address space, just free
1096 * and return -EINTR so user space doesn't expect any data.
Roland Dreier35dc2482013-08-05 17:55:01 -07001097 */
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001098 if (!current->mm)
1099 ret = -EINTR;
1100 else if (bio_data_dir(bio) == READ)
Dongsu Park9124d3f2015-01-18 16:16:34 +01001101 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001102 if (bmd->is_our_pages)
1103 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001104 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001105 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 bio_put(bio);
1107 return ret;
1108}
1109
1110/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001111 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001112 * @q: destination block queue
1113 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1114 * @iter: iovec iterator
1115 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 *
1117 * Prepares and returns a bio for indirect user io, bouncing data
1118 * to/from kernel pages as necessary. Must be paired with
1119 * call bio_uncopy_user() on io completion.
1120 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001121struct bio *bio_copy_user_iov(struct request_queue *q,
1122 struct rq_map_data *map_data,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001123 const struct iov_iter *iter,
1124 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct page *page;
1128 struct bio *bio;
1129 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001130 int nr_pages = 0;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001131 unsigned int len = iter->count;
Geliang Tangbd5cece2015-11-21 17:27:31 +08001132 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001134 for (i = 0; i < iter->nr_segs; i++) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001135 unsigned long uaddr;
1136 unsigned long end;
1137 unsigned long start;
1138
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001139 uaddr = (unsigned long) iter->iov[i].iov_base;
1140 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1141 >> PAGE_SHIFT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001142 start = uaddr >> PAGE_SHIFT;
1143
Jens Axboecb4644c2010-11-10 14:36:25 +01001144 /*
1145 * Overflow, abort
1146 */
1147 if (end < start)
1148 return ERR_PTR(-EINVAL);
1149
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001150 nr_pages += end - start;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001151 }
1152
FUJITA Tomonori69838722009-04-28 20:24:29 +02001153 if (offset)
1154 nr_pages++;
1155
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001156 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (!bmd)
1158 return ERR_PTR(-ENOMEM);
1159
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001160 /*
1161 * We need to do a deep copy of the iov_iter including the iovecs.
1162 * The caller provided iov might point to an on-stack or otherwise
1163 * shortlived one.
1164 */
1165 bmd->is_our_pages = map_data ? 0 : 1;
1166 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1167 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1168 iter->nr_segs, iter->count);
1169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001171 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 if (!bio)
1173 goto out_bmd;
1174
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001175 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001176 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
1178 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001179
1180 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001181 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001182 i = map_data->offset / PAGE_SIZE;
1183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001185 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001187 bytes -= offset;
1188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 if (bytes > len)
1190 bytes = len;
1191
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001192 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001193 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001194 ret = -ENOMEM;
1195 break;
1196 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001197
1198 page = map_data->pages[i / nr_pages];
1199 page += (i % nr_pages);
1200
1201 i++;
1202 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001203 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001204 if (!page) {
1205 ret = -ENOMEM;
1206 break;
1207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
1209
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001210 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001214 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
1217 if (ret)
1218 goto cleanup;
1219
1220 /*
1221 * success
1222 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001223 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001224 (map_data && map_data->from_user)) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001225 ret = bio_copy_from_iter(bio, *iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001226 if (ret)
1227 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 }
1229
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001230 bio->bi_private = bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 return bio;
1232cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001233 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001234 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 bio_put(bio);
1236out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001237 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return ERR_PTR(ret);
1239}
1240
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001241/**
1242 * bio_map_user_iov - map user iovec into bio
1243 * @q: the struct request_queue for the bio
1244 * @iter: iovec iterator
1245 * @gfp_mask: memory allocation flags
1246 *
1247 * Map the user space address into a bio suitable for io to a block
1248 * device. Returns an error pointer in case of error.
1249 */
1250struct bio *bio_map_user_iov(struct request_queue *q,
1251 const struct iov_iter *iter,
1252 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001254 int j;
James Bottomley f1970ba2005-06-20 14:06:52 +02001255 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 struct page **pages;
1257 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001258 int cur_page = 0;
1259 int ret, offset;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001260 struct iov_iter i;
1261 struct iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001263 iov_for_each(iov, i, *iter) {
1264 unsigned long uaddr = (unsigned long) iov.iov_base;
1265 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001266 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1267 unsigned long start = uaddr >> PAGE_SHIFT;
1268
Jens Axboecb4644c2010-11-10 14:36:25 +01001269 /*
1270 * Overflow, abort
1271 */
1272 if (end < start)
1273 return ERR_PTR(-EINVAL);
1274
James Bottomley f1970ba2005-06-20 14:06:52 +02001275 nr_pages += end - start;
1276 /*
Mike Christiead2d7222006-12-01 10:40:20 +01001277 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +02001278 */
Mike Christiead2d7222006-12-01 10:40:20 +01001279 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +02001280 return ERR_PTR(-EINVAL);
1281 }
1282
1283 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return ERR_PTR(-EINVAL);
1285
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001286 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (!bio)
1288 return ERR_PTR(-ENOMEM);
1289
1290 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001291 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 if (!pages)
1293 goto out;
1294
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001295 iov_for_each(iov, i, *iter) {
1296 unsigned long uaddr = (unsigned long) iov.iov_base;
1297 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001298 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1299 unsigned long start = uaddr >> PAGE_SHIFT;
1300 const int local_nr_pages = end - start;
1301 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001302
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001303 ret = get_user_pages_fast(uaddr, local_nr_pages,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001304 (iter->type & WRITE) != WRITE,
1305 &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +02001306 if (ret < local_nr_pages) {
1307 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001308 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Geliang Tangbd5cece2015-11-21 17:27:31 +08001311 offset = offset_in_page(uaddr);
James Bottomley f1970ba2005-06-20 14:06:52 +02001312 for (j = cur_page; j < page_limit; j++) {
1313 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
James Bottomley f1970ba2005-06-20 14:06:52 +02001315 if (len <= 0)
1316 break;
1317
1318 if (bytes > len)
1319 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
James Bottomley f1970ba2005-06-20 14:06:52 +02001321 /*
1322 * sorry...
1323 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001324 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1325 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001326 break;
1327
1328 len -= bytes;
1329 offset = 0;
1330 }
1331
1332 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001334 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001336 while (j < page_limit)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001337 put_page(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 kfree(pages);
1341
1342 /*
1343 * set data direction, and check if mapped pages need bouncing
1344 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001345 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001346 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001348 bio_set_flag(bio, BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001349
1350 /*
1351 * subtle -- if __bio_map_user() ended up bouncing a bio,
1352 * it would normally disappear when its bi_end_io is run.
1353 * however, we need it for the unmap, so grab an extra
1354 * reference to it
1355 */
1356 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001358
1359 out_unmap:
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001360 for (j = 0; j < nr_pages; j++) {
1361 if (!pages[j])
James Bottomley f1970ba2005-06-20 14:06:52 +02001362 break;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001363 put_page(pages[j]);
James Bottomley f1970ba2005-06-20 14:06:52 +02001364 }
1365 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 kfree(pages);
1367 bio_put(bio);
1368 return ERR_PTR(ret);
1369}
1370
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371static void __bio_unmap_user(struct bio *bio)
1372{
1373 struct bio_vec *bvec;
1374 int i;
1375
1376 /*
1377 * make sure we dirty pages we wrote to
1378 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001379 bio_for_each_segment_all(bvec, bio, i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 if (bio_data_dir(bio) == READ)
1381 set_page_dirty_lock(bvec->bv_page);
1382
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001383 put_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 }
1385
1386 bio_put(bio);
1387}
1388
1389/**
1390 * bio_unmap_user - unmap a bio
1391 * @bio: the bio being unmapped
1392 *
1393 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1394 * a process context.
1395 *
1396 * bio_unmap_user() may sleep.
1397 */
1398void bio_unmap_user(struct bio *bio)
1399{
1400 __bio_unmap_user(bio);
1401 bio_put(bio);
1402}
1403
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001404static void bio_map_kern_endio(struct bio *bio)
Jens Axboeb8238252005-06-20 14:05:27 +02001405{
Jens Axboeb8238252005-06-20 14:05:27 +02001406 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001407}
1408
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001409/**
1410 * bio_map_kern - map kernel address into bio
1411 * @q: the struct request_queue for the bio
1412 * @data: pointer to buffer to map
1413 * @len: length in bytes
1414 * @gfp_mask: allocation flags for bio allocation
1415 *
1416 * Map the kernel address into a bio suitable for io to a block
1417 * device. Returns an error pointer in case of error.
1418 */
1419struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1420 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001421{
1422 unsigned long kaddr = (unsigned long)data;
1423 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1424 unsigned long start = kaddr >> PAGE_SHIFT;
1425 const int nr_pages = end - start;
1426 int offset, i;
1427 struct bio *bio;
1428
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001429 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001430 if (!bio)
1431 return ERR_PTR(-ENOMEM);
1432
1433 offset = offset_in_page(kaddr);
1434 for (i = 0; i < nr_pages; i++) {
1435 unsigned int bytes = PAGE_SIZE - offset;
1436
1437 if (len <= 0)
1438 break;
1439
1440 if (bytes > len)
1441 bytes = len;
1442
Mike Christiedefd94b2005-12-05 02:37:06 -06001443 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001444 offset) < bytes) {
1445 /* we don't support partial mappings */
1446 bio_put(bio);
1447 return ERR_PTR(-EINVAL);
1448 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001449
1450 data += bytes;
1451 len -= bytes;
1452 offset = 0;
1453 }
1454
Jens Axboeb8238252005-06-20 14:05:27 +02001455 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001456 return bio;
1457}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001458EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001459
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001460static void bio_copy_kern_endio(struct bio *bio)
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001461{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001462 bio_free_pages(bio);
1463 bio_put(bio);
1464}
1465
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001466static void bio_copy_kern_endio_read(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001467{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001468 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001469 struct bio_vec *bvec;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001470 int i;
1471
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001472 bio_for_each_segment_all(bvec, bio, i) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001473 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001474 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001475 }
1476
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001477 bio_copy_kern_endio(bio);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001478}
1479
1480/**
1481 * bio_copy_kern - copy kernel address into bio
1482 * @q: the struct request_queue for the bio
1483 * @data: pointer to buffer to copy
1484 * @len: length in bytes
1485 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001486 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001487 *
1488 * copy the kernel address into a bio suitable for io to a block
1489 * device. Returns an error pointer in case of error.
1490 */
1491struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1492 gfp_t gfp_mask, int reading)
1493{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001494 unsigned long kaddr = (unsigned long)data;
1495 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1496 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001497 struct bio *bio;
1498 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001499 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001500
Christoph Hellwig42d26832015-01-18 16:16:28 +01001501 /*
1502 * Overflow, abort
1503 */
1504 if (end < start)
1505 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001506
Christoph Hellwig42d26832015-01-18 16:16:28 +01001507 nr_pages = end - start;
1508 bio = bio_kmalloc(gfp_mask, nr_pages);
1509 if (!bio)
1510 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001511
Christoph Hellwig42d26832015-01-18 16:16:28 +01001512 while (len) {
1513 struct page *page;
1514 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001515
Christoph Hellwig42d26832015-01-18 16:16:28 +01001516 if (bytes > len)
1517 bytes = len;
1518
1519 page = alloc_page(q->bounce_gfp | gfp_mask);
1520 if (!page)
1521 goto cleanup;
1522
1523 if (!reading)
1524 memcpy(page_address(page), p, bytes);
1525
1526 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1527 break;
1528
1529 len -= bytes;
1530 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001531 }
1532
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001533 if (reading) {
1534 bio->bi_end_io = bio_copy_kern_endio_read;
1535 bio->bi_private = data;
1536 } else {
1537 bio->bi_end_io = bio_copy_kern_endio;
Mike Christie95fe6c12016-06-05 14:31:48 -05001538 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001539 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001540
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001541 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001542
1543cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001544 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001545 bio_put(bio);
1546 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001547}
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549/*
1550 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1551 * for performing direct-IO in BIOs.
1552 *
1553 * The problem is that we cannot run set_page_dirty() from interrupt context
1554 * because the required locks are not interrupt-safe. So what we can do is to
1555 * mark the pages dirty _before_ performing IO. And in interrupt context,
1556 * check that the pages are still dirty. If so, fine. If not, redirty them
1557 * in process context.
1558 *
1559 * We special-case compound pages here: normally this means reads into hugetlb
1560 * pages. The logic in here doesn't really work right for compound pages
1561 * because the VM does not uniformly chase down the head page in all cases.
1562 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1563 * handle them at all. So we skip compound pages here at an early stage.
1564 *
1565 * Note that this code is very hard to test under normal circumstances because
1566 * direct-io pins the pages with get_user_pages(). This makes
1567 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001568 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 * pagecache.
1570 *
1571 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1572 * deferred bio dirtying paths.
1573 */
1574
1575/*
1576 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1577 */
1578void bio_set_pages_dirty(struct bio *bio)
1579{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001580 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 int i;
1582
Kent Overstreetcb34e052012-09-05 15:22:02 -07001583 bio_for_each_segment_all(bvec, bio, i) {
1584 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
1586 if (page && !PageCompound(page))
1587 set_page_dirty_lock(page);
1588 }
1589}
1590
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001591static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001593 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 int i;
1595
Kent Overstreetcb34e052012-09-05 15:22:02 -07001596 bio_for_each_segment_all(bvec, bio, i) {
1597 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
1599 if (page)
1600 put_page(page);
1601 }
1602}
1603
1604/*
1605 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1606 * If they are, then fine. If, however, some pages are clean then they must
1607 * have been written out during the direct-IO read. So we take another ref on
1608 * the BIO and the offending pages and re-dirty the pages in process context.
1609 *
1610 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +03001611 * here on. It will run one put_page() against each page and will run one
1612 * bio_put() against the BIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 */
1614
David Howells65f27f32006-11-22 14:55:48 +00001615static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
David Howells65f27f32006-11-22 14:55:48 +00001617static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618static DEFINE_SPINLOCK(bio_dirty_lock);
1619static struct bio *bio_dirty_list;
1620
1621/*
1622 * This runs in process context
1623 */
David Howells65f27f32006-11-22 14:55:48 +00001624static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625{
1626 unsigned long flags;
1627 struct bio *bio;
1628
1629 spin_lock_irqsave(&bio_dirty_lock, flags);
1630 bio = bio_dirty_list;
1631 bio_dirty_list = NULL;
1632 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1633
1634 while (bio) {
1635 struct bio *next = bio->bi_private;
1636
1637 bio_set_pages_dirty(bio);
1638 bio_release_pages(bio);
1639 bio_put(bio);
1640 bio = next;
1641 }
1642}
1643
1644void bio_check_pages_dirty(struct bio *bio)
1645{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001646 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 int nr_clean_pages = 0;
1648 int i;
1649
Kent Overstreetcb34e052012-09-05 15:22:02 -07001650 bio_for_each_segment_all(bvec, bio, i) {
1651 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
1653 if (PageDirty(page) || PageCompound(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001654 put_page(page);
Kent Overstreetcb34e052012-09-05 15:22:02 -07001655 bvec->bv_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 } else {
1657 nr_clean_pages++;
1658 }
1659 }
1660
1661 if (nr_clean_pages) {
1662 unsigned long flags;
1663
1664 spin_lock_irqsave(&bio_dirty_lock, flags);
1665 bio->bi_private = bio_dirty_list;
1666 bio_dirty_list = bio;
1667 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1668 schedule_work(&bio_dirty_work);
1669 } else {
1670 bio_put(bio);
1671 }
1672}
1673
Gu Zheng394ffa52014-11-24 11:05:22 +08001674void generic_start_io_acct(int rw, unsigned long sectors,
1675 struct hd_struct *part)
1676{
1677 int cpu = part_stat_lock();
1678
1679 part_round_stats(cpu, part);
1680 part_stat_inc(cpu, part, ios[rw]);
1681 part_stat_add(cpu, part, sectors[rw], sectors);
1682 part_inc_in_flight(part, rw);
1683
1684 part_stat_unlock();
1685}
1686EXPORT_SYMBOL(generic_start_io_acct);
1687
1688void generic_end_io_acct(int rw, struct hd_struct *part,
1689 unsigned long start_time)
1690{
1691 unsigned long duration = jiffies - start_time;
1692 int cpu = part_stat_lock();
1693
1694 part_stat_add(cpu, part, ticks[rw], duration);
1695 part_round_stats(cpu, part);
1696 part_dec_in_flight(part, rw);
1697
1698 part_stat_unlock();
1699}
1700EXPORT_SYMBOL(generic_end_io_acct);
1701
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001702#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1703void bio_flush_dcache_pages(struct bio *bi)
1704{
Kent Overstreet79886132013-11-23 17:19:00 -08001705 struct bio_vec bvec;
1706 struct bvec_iter iter;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001707
Kent Overstreet79886132013-11-23 17:19:00 -08001708 bio_for_each_segment(bvec, bi, iter)
1709 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001710}
1711EXPORT_SYMBOL(bio_flush_dcache_pages);
1712#endif
1713
Jens Axboec4cf5262015-04-17 16:15:18 -06001714static inline bool bio_remaining_done(struct bio *bio)
1715{
1716 /*
1717 * If we're not chaining, then ->__bi_remaining is always 1 and
1718 * we always end io on the first invocation.
1719 */
1720 if (!bio_flagged(bio, BIO_CHAIN))
1721 return true;
1722
1723 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1724
Mike Snitzer326e1db2015-05-22 09:14:03 -04001725 if (atomic_dec_and_test(&bio->__bi_remaining)) {
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001726 bio_clear_flag(bio, BIO_CHAIN);
Jens Axboec4cf5262015-04-17 16:15:18 -06001727 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001728 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001729
1730 return false;
1731}
1732
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733/**
1734 * bio_endio - end I/O on a bio
1735 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 *
1737 * Description:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001738 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1739 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1740 * bio unless they own it and thus know that it has an end_io function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 **/
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001742void bio_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743{
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001744again:
Christoph Hellwig2b885512016-03-11 17:34:53 +01001745 if (!bio_remaining_done(bio))
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001746 return;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001747
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001748 /*
1749 * Need to have a real endio function for chained bios, otherwise
1750 * various corner cases will break (like stacking block devices that
1751 * save/restore bi_end_io) - however, we want to avoid unbounded
1752 * recursion and blowing the stack. Tail call optimization would
1753 * handle this, but compiling with frame pointers also disables
1754 * gcc's sibling call optimization.
1755 */
1756 if (bio->bi_end_io == bio_chain_endio) {
1757 bio = __bio_chain_endio(bio);
1758 goto again;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001759 }
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001760
1761 if (bio->bi_end_io)
1762 bio->bi_end_io(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001764EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
Kent Overstreet196d38b2013-11-23 18:34:15 -08001766/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001767 * bio_split - split a bio
1768 * @bio: bio to split
1769 * @sectors: number of sectors to split from the front of @bio
1770 * @gfp: gfp mask
1771 * @bs: bio set to allocate from
1772 *
1773 * Allocates and returns a new bio which represents @sectors from the start of
1774 * @bio, and updates @bio to represent the remaining sectors.
1775 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001776 * Unless this is a discard request the newly allocated bio will point
1777 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1778 * @bio is not freed before the split.
Kent Overstreet20d01892013-11-23 18:21:01 -08001779 */
1780struct bio *bio_split(struct bio *bio, int sectors,
1781 gfp_t gfp, struct bio_set *bs)
1782{
1783 struct bio *split = NULL;
1784
1785 BUG_ON(sectors <= 0);
1786 BUG_ON(sectors >= bio_sectors(bio));
1787
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001788 /*
1789 * Discards need a mutable bio_vec to accommodate the payload
1790 * required by the DSM TRIM and UNMAP commands.
1791 */
Adrian Hunter7afafc82016-08-16 10:59:35 +03001792 if (bio_op(bio) == REQ_OP_DISCARD || bio_op(bio) == REQ_OP_SECURE_ERASE)
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001793 split = bio_clone_bioset(bio, gfp, bs);
1794 else
1795 split = bio_clone_fast(bio, gfp, bs);
1796
Kent Overstreet20d01892013-11-23 18:21:01 -08001797 if (!split)
1798 return NULL;
1799
1800 split->bi_iter.bi_size = sectors << 9;
1801
1802 if (bio_integrity(split))
1803 bio_integrity_trim(split, 0, sectors);
1804
1805 bio_advance(bio, split->bi_iter.bi_size);
1806
1807 return split;
1808}
1809EXPORT_SYMBOL(bio_split);
1810
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001811/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001812 * bio_trim - trim a bio
1813 * @bio: bio to trim
1814 * @offset: number of sectors to trim from the front of @bio
1815 * @size: size we want to trim @bio to, in sectors
1816 */
1817void bio_trim(struct bio *bio, int offset, int size)
1818{
1819 /* 'bio' is a cloned bio which we need to trim to match
1820 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001821 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001822
1823 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001824 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001825 return;
1826
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001827 bio_clear_flag(bio, BIO_SEG_VALID);
Kent Overstreet6678d832013-08-07 11:14:32 -07001828
1829 bio_advance(bio, offset << 9);
1830
Kent Overstreet4f024f32013-10-11 15:44:27 -07001831 bio->bi_iter.bi_size = size;
Kent Overstreet6678d832013-08-07 11:14:32 -07001832}
1833EXPORT_SYMBOL_GPL(bio_trim);
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835/*
1836 * create memory pools for biovec's in a bio_set.
1837 * use the global biovec slabs created for general use.
1838 */
Fabian Fredericka6c39cb4f2014-04-22 15:09:05 -06001839mempool_t *biovec_create_pool(int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840{
Christoph Hellwiged996a52016-07-19 11:28:42 +02001841 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
Kent Overstreet9f060e22012-10-12 15:29:33 -07001843 return mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844}
1845
1846void bioset_free(struct bio_set *bs)
1847{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001848 if (bs->rescue_workqueue)
1849 destroy_workqueue(bs->rescue_workqueue);
1850
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 if (bs->bio_pool)
1852 mempool_destroy(bs->bio_pool);
1853
Kent Overstreet9f060e22012-10-12 15:29:33 -07001854 if (bs->bvec_pool)
1855 mempool_destroy(bs->bvec_pool);
1856
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001857 bioset_integrity_free(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001858 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
1860 kfree(bs);
1861}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001862EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001864static struct bio_set *__bioset_create(unsigned int pool_size,
1865 unsigned int front_pad,
1866 bool create_bvec_pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867{
Jens Axboe392ddc32008-12-23 12:42:54 +01001868 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001869 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Jens Axboe1b434492008-10-22 20:32:58 +02001871 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 if (!bs)
1873 return NULL;
1874
Jens Axboebb799ca2008-12-10 15:35:05 +01001875 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001876
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001877 spin_lock_init(&bs->rescue_lock);
1878 bio_list_init(&bs->rescue_list);
1879 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1880
Jens Axboe392ddc32008-12-23 12:42:54 +01001881 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001882 if (!bs->bio_slab) {
1883 kfree(bs);
1884 return NULL;
1885 }
1886
1887 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 if (!bs->bio_pool)
1889 goto bad;
1890
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001891 if (create_bvec_pool) {
1892 bs->bvec_pool = biovec_create_pool(pool_size);
1893 if (!bs->bvec_pool)
1894 goto bad;
1895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001897 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1898 if (!bs->rescue_workqueue)
1899 goto bad;
1900
1901 return bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902bad:
1903 bioset_free(bs);
1904 return NULL;
1905}
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001906
1907/**
1908 * bioset_create - Create a bio_set
1909 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1910 * @front_pad: Number of bytes to allocate in front of the returned bio
1911 *
1912 * Description:
1913 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1914 * to ask for a number of bytes to be allocated in front of the bio.
1915 * Front pad allocation is useful for embedding the bio inside
1916 * another structure, to avoid allocating extra data to go with the bio.
1917 * Note that the bio must be embedded at the END of that structure always,
1918 * or things will break badly.
1919 */
1920struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1921{
1922 return __bioset_create(pool_size, front_pad, true);
1923}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001924EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001926/**
1927 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1928 * @pool_size: Number of bio to cache in the mempool
1929 * @front_pad: Number of bytes to allocate in front of the returned bio
1930 *
1931 * Description:
1932 * Same functionality as bioset_create() except that mempool is not
1933 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1934 */
1935struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1936{
1937 return __bioset_create(pool_size, front_pad, false);
1938}
1939EXPORT_SYMBOL(bioset_create_nobvec);
1940
Tejun Heo852c7882012-03-05 13:15:27 -08001941#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04001942
1943/**
1944 * bio_associate_blkcg - associate a bio with the specified blkcg
1945 * @bio: target bio
1946 * @blkcg_css: css of the blkcg to associate
1947 *
1948 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1949 * treat @bio as if it were issued by a task which belongs to the blkcg.
1950 *
1951 * This function takes an extra reference of @blkcg_css which will be put
1952 * when @bio is released. The caller must own @bio and is responsible for
1953 * synchronizing calls to this function.
1954 */
1955int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1956{
1957 if (unlikely(bio->bi_css))
1958 return -EBUSY;
1959 css_get(blkcg_css);
1960 bio->bi_css = blkcg_css;
1961 return 0;
1962}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001963EXPORT_SYMBOL_GPL(bio_associate_blkcg);
Tejun Heo1d933cf2015-05-22 17:13:24 -04001964
Tejun Heo852c7882012-03-05 13:15:27 -08001965/**
1966 * bio_associate_current - associate a bio with %current
1967 * @bio: target bio
1968 *
1969 * Associate @bio with %current if it hasn't been associated yet. Block
1970 * layer will treat @bio as if it were issued by %current no matter which
1971 * task actually issues it.
1972 *
1973 * This function takes an extra reference of @task's io_context and blkcg
1974 * which will be put when @bio is released. The caller must own @bio,
1975 * ensure %current->io_context exists, and is responsible for synchronizing
1976 * calls to this function.
1977 */
1978int bio_associate_current(struct bio *bio)
1979{
1980 struct io_context *ioc;
Tejun Heo852c7882012-03-05 13:15:27 -08001981
Tejun Heo1d933cf2015-05-22 17:13:24 -04001982 if (bio->bi_css)
Tejun Heo852c7882012-03-05 13:15:27 -08001983 return -EBUSY;
1984
1985 ioc = current->io_context;
1986 if (!ioc)
1987 return -ENOENT;
1988
Tejun Heo852c7882012-03-05 13:15:27 -08001989 get_io_context_active(ioc);
1990 bio->bi_ioc = ioc;
Tejun Heoc165b3e2015-08-18 14:55:29 -07001991 bio->bi_css = task_get_css(current, io_cgrp_id);
Tejun Heo852c7882012-03-05 13:15:27 -08001992 return 0;
1993}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001994EXPORT_SYMBOL_GPL(bio_associate_current);
Tejun Heo852c7882012-03-05 13:15:27 -08001995
1996/**
1997 * bio_disassociate_task - undo bio_associate_current()
1998 * @bio: target bio
1999 */
2000void bio_disassociate_task(struct bio *bio)
2001{
2002 if (bio->bi_ioc) {
2003 put_io_context(bio->bi_ioc);
2004 bio->bi_ioc = NULL;
2005 }
2006 if (bio->bi_css) {
2007 css_put(bio->bi_css);
2008 bio->bi_css = NULL;
2009 }
2010}
2011
Paolo Valente20bd7232016-07-27 07:22:05 +02002012/**
2013 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2014 * @dst: destination bio
2015 * @src: source bio
2016 */
2017void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2018{
2019 if (src->bi_css)
2020 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2021}
2022
Tejun Heo852c7882012-03-05 13:15:27 -08002023#endif /* CONFIG_BLK_CGROUP */
2024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025static void __init biovec_init_slabs(void)
2026{
2027 int i;
2028
Christoph Hellwiged996a52016-07-19 11:28:42 +02002029 for (i = 0; i < BVEC_POOL_NR; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 int size;
2031 struct biovec_slab *bvs = bvec_slabs + i;
2032
Jens Axboea7fcd372008-12-05 16:10:29 +01002033 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2034 bvs->slab = NULL;
2035 continue;
2036 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 size = bvs->nr_vecs * sizeof(struct bio_vec);
2039 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002040 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042}
2043
2044static int __init init_bio(void)
2045{
Jens Axboebb799ca2008-12-10 15:35:05 +01002046 bio_slab_max = 2;
2047 bio_slab_nr = 0;
2048 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2049 if (!bio_slabs)
2050 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002052 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 biovec_init_slabs();
2054
Jens Axboebb799ca2008-12-10 15:35:05 +01002055 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (!fs_bio_set)
2057 panic("bio: can't allocate bios\n");
2058
Martin K. Petersena91a2782011-03-17 11:11:05 +01002059 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2060 panic("bio: can't create integrity pool\n");
2061
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 return 0;
2063}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064subsys_initcall(init_bio);