blob: 4c18a68913deb45fc6df162a7be68f66e2635b2f [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 */
Mikulas Patocka6bae9122018-03-21 12:49:29 -040045#define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
Christoph Hellwiged996a52016-07-19 11:28:42 +020046static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
Mikulas Patocka6bae9122018-03-21 12:49:29 -040047 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
Linus Torvalds1da177e2005-04-16 15:20:36 -070048};
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{
Greg Edwardsbbab86e2018-08-08 13:27:53 -0600158 return bvec_slabs[--idx].nr_vecs;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200159}
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
NeilBrown5959cde2017-03-10 17:00:47 +1100375 while ((bio = bio_list_pop(&current->bio_list[0])))
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700376 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
NeilBrown5959cde2017-03-10 17:00:47 +1100377 current->bio_list[0] = nopunt;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700378
NeilBrown5959cde2017-03-10 17:00:47 +1100379 bio_list_init(&nopunt);
380 while ((bio = bio_list_pop(&current->bio_list[1])))
381 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
382 current->bio_list[1] = nopunt;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700383
384 spin_lock(&bs->rescue_lock);
385 bio_list_merge(&bs->rescue_list, &punt);
386 spin_unlock(&bs->rescue_lock);
387
388 queue_work(bs->rescue_workqueue, &bs->rescue_work);
389}
390
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700391/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 * bio_alloc_bioset - allocate a bio for I/O
393 * @gfp_mask: the GFP_ mask given to the slab allocator
394 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200395 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 *
397 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700398 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
399 * backed by the @bs's mempool.
400 *
Mel Gormand0164ad2015-11-06 16:28:21 -0800401 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
402 * always be able to allocate a bio. This is due to the mempool guarantees.
403 * To make this work, callers must never allocate more than 1 bio at a time
404 * from this pool. Callers that need to allocate more than 1 bio must always
405 * submit the previously allocated bio for IO before attempting to allocate
406 * a new one. Failure to do so can cause deadlocks under memory pressure.
Kent Overstreet3f86a822012-09-06 15:35:01 -0700407 *
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700408 * Note that when running under generic_make_request() (i.e. any block
409 * driver), bios are not submitted until after you return - see the code in
410 * generic_make_request() that converts recursion into iteration, to prevent
411 * stack overflows.
412 *
413 * This would normally mean allocating multiple bios under
414 * generic_make_request() would be susceptible to deadlocks, but we have
415 * deadlock avoidance code that resubmits any blocked bios from a rescuer
416 * thread.
417 *
418 * However, we do not guarantee forward progress for allocations from other
419 * mempools. Doing multiple allocations from the same mempool under
420 * generic_make_request() should be avoided - instead, use bio_set's front_pad
421 * for per bio allocations.
422 *
Kent Overstreet3f86a822012-09-06 15:35:01 -0700423 * RETURNS:
424 * Pointer to new bio on success, NULL on failure.
425 */
Al Virodd0fc662005-10-07 07:46:04 +0100426struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700428 gfp_t saved_gfp = gfp_mask;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700429 unsigned front_pad;
430 unsigned inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100431 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200432 struct bio *bio;
433 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200434
Kent Overstreet3f86a822012-09-06 15:35:01 -0700435 if (!bs) {
436 if (nr_iovecs > UIO_MAXIOV)
437 return NULL;
438
439 p = kmalloc(sizeof(struct bio) +
440 nr_iovecs * sizeof(struct bio_vec),
441 gfp_mask);
442 front_pad = 0;
443 inline_vecs = nr_iovecs;
444 } else {
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400445 /* should not use nobvec bioset for nr_iovecs > 0 */
446 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
447 return NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700448 /*
449 * generic_make_request() converts recursion to iteration; this
450 * means if we're running beneath it, any bios we allocate and
451 * submit will not be submitted (and thus freed) until after we
452 * return.
453 *
454 * This exposes us to a potential deadlock if we allocate
455 * multiple bios from the same bio_set() while running
456 * underneath generic_make_request(). If we were to allocate
457 * multiple bios (say a stacking block driver that was splitting
458 * bios), we would deadlock if we exhausted the mempool's
459 * reserve.
460 *
461 * We solve this, and guarantee forward progress, with a rescuer
462 * workqueue per bio_set. If we go to allocate and there are
463 * bios on current->bio_list, we first try the allocation
Mel Gormand0164ad2015-11-06 16:28:21 -0800464 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
465 * bios we would be blocking to the rescuer workqueue before
466 * we retry with the original gfp_flags.
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700467 */
468
NeilBrown5959cde2017-03-10 17:00:47 +1100469 if (current->bio_list &&
470 (!bio_list_empty(&current->bio_list[0]) ||
471 !bio_list_empty(&current->bio_list[1])))
Mel Gormand0164ad2015-11-06 16:28:21 -0800472 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700473
Kent Overstreet3f86a822012-09-06 15:35:01 -0700474 p = mempool_alloc(bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700475 if (!p && gfp_mask != saved_gfp) {
476 punt_bios_to_rescuer(bs);
477 gfp_mask = saved_gfp;
478 p = mempool_alloc(bs->bio_pool, gfp_mask);
479 }
480
Kent Overstreet3f86a822012-09-06 15:35:01 -0700481 front_pad = bs->front_pad;
482 inline_vecs = BIO_INLINE_VECS;
483 }
484
Tejun Heo451a9eb2009-04-15 19:50:51 +0200485 if (unlikely(!p))
486 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100487
Kent Overstreet3f86a822012-09-06 15:35:01 -0700488 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100489 bio_init(bio);
490
Kent Overstreet3f86a822012-09-06 15:35:01 -0700491 if (nr_iovecs > inline_vecs) {
Christoph Hellwiged996a52016-07-19 11:28:42 +0200492 unsigned long idx = 0;
493
Kent Overstreet9f060e22012-10-12 15:29:33 -0700494 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700495 if (!bvl && gfp_mask != saved_gfp) {
496 punt_bios_to_rescuer(bs);
497 gfp_mask = saved_gfp;
Kent Overstreet9f060e22012-10-12 15:29:33 -0700498 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700499 }
500
Ingo Molnar34053972009-02-21 11:16:36 +0100501 if (unlikely(!bvl))
502 goto err_free;
Kent Overstreeta38352e2012-05-25 13:03:11 -0700503
Christoph Hellwiged996a52016-07-19 11:28:42 +0200504 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700505 } else if (nr_iovecs) {
506 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100507 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700508
509 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100510 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100511 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100513
514err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200515 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100516 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200518EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520void zero_fill_bio(struct bio *bio)
521{
522 unsigned long flags;
Kent Overstreet79886132013-11-23 17:19:00 -0800523 struct bio_vec bv;
524 struct bvec_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Kent Overstreet79886132013-11-23 17:19:00 -0800526 bio_for_each_segment(bv, bio, iter) {
527 char *data = bvec_kmap_irq(&bv, &flags);
528 memset(data, 0, bv.bv_len);
529 flush_dcache_page(bv.bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 bvec_kunmap_irq(data, &flags);
531 }
532}
533EXPORT_SYMBOL(zero_fill_bio);
534
535/**
536 * bio_put - release a reference to a bio
537 * @bio: bio to release reference to
538 *
539 * Description:
540 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100541 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 **/
543void bio_put(struct bio *bio)
544{
Jens Axboedac56212015-04-17 16:23:59 -0600545 if (!bio_flagged(bio, BIO_REFFED))
Kent Overstreet4254bba2012-09-06 15:35:00 -0700546 bio_free(bio);
Jens Axboedac56212015-04-17 16:23:59 -0600547 else {
548 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
549
550 /*
551 * last put frees it
552 */
553 if (atomic_dec_and_test(&bio->__bi_cnt))
554 bio_free(bio);
555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200557EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Jens Axboe165125e2007-07-24 09:28:11 +0200559inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
561 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
562 blk_recount_segments(q, bio);
563
564 return bio->bi_phys_segments;
565}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200566EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800569 * __bio_clone_fast - clone a bio that shares the original bio's biovec
570 * @bio: destination bio
571 * @bio_src: bio to clone
572 *
573 * Clone a &bio. Caller will own the returned bio, but not
574 * the actual data it points to. Reference count of returned
575 * bio will be one.
576 *
577 * Caller must ensure that @bio_src is not freed before @bio.
578 */
579void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
580{
Christoph Hellwiged996a52016-07-19 11:28:42 +0200581 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
Kent Overstreet59d276f2013-11-23 18:19:27 -0800582
583 /*
584 * most users will be overriding ->bi_bdev with a new target,
585 * so we don't set nor calculate new physical/hw segment counts here
586 */
587 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600588 bio_set_flag(bio, BIO_CLONED);
Jens Axboe1eff9d32016-08-05 15:35:16 -0600589 bio->bi_opf = bio_src->bi_opf;
Kent Overstreet59d276f2013-11-23 18:19:27 -0800590 bio->bi_iter = bio_src->bi_iter;
591 bio->bi_io_vec = bio_src->bi_io_vec;
Paolo Valente20bd7232016-07-27 07:22:05 +0200592
593 bio_clone_blkcg_association(bio, bio_src);
Kent Overstreet59d276f2013-11-23 18:19:27 -0800594}
595EXPORT_SYMBOL(__bio_clone_fast);
596
597/**
598 * bio_clone_fast - clone a bio that shares the original bio's biovec
599 * @bio: bio to clone
600 * @gfp_mask: allocation priority
601 * @bs: bio_set to allocate from
602 *
603 * Like __bio_clone_fast, only also allocates the returned bio
604 */
605struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
606{
607 struct bio *b;
608
609 b = bio_alloc_bioset(gfp_mask, 0, bs);
610 if (!b)
611 return NULL;
612
613 __bio_clone_fast(b, bio);
614
615 if (bio_integrity(bio)) {
616 int ret;
617
618 ret = bio_integrity_clone(b, bio, gfp_mask);
619
620 if (ret < 0) {
621 bio_put(b);
622 return NULL;
623 }
624 }
625
626 return b;
627}
628EXPORT_SYMBOL(bio_clone_fast);
629
630/**
Kent Overstreetbdb53202013-11-23 17:26:46 -0800631 * bio_clone_bioset - clone a bio
632 * @bio_src: bio to clone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700634 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 *
Kent Overstreetbdb53202013-11-23 17:26:46 -0800636 * Clone bio. Caller will own the returned bio, but not the actual data it
637 * points to. Reference count of returned bio will be one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 */
Kent Overstreetbdb53202013-11-23 17:26:46 -0800639struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700640 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Kent Overstreetbdb53202013-11-23 17:26:46 -0800642 struct bvec_iter iter;
643 struct bio_vec bv;
644 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Kent Overstreetbdb53202013-11-23 17:26:46 -0800646 /*
647 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
648 * bio_src->bi_io_vec to bio->bi_io_vec.
649 *
650 * We can't do that anymore, because:
651 *
652 * - The point of cloning the biovec is to produce a bio with a biovec
653 * the caller can modify: bi_idx and bi_bvec_done should be 0.
654 *
655 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
656 * we tried to clone the whole thing bio_alloc_bioset() would fail.
657 * But the clone should succeed as long as the number of biovecs we
658 * actually need to allocate is fewer than BIO_MAX_PAGES.
659 *
660 * - Lastly, bi_vcnt should not be looked at or relied upon by code
661 * that does not own the bio - reason being drivers don't use it for
662 * iterating over the biovec anymore, so expecting it to be kept up
663 * to date (i.e. for clones that share the parent biovec) is just
664 * asking for trouble and would force extra work on
665 * __bio_clone_fast() anyways.
666 */
667
Kent Overstreet8423ae32014-02-10 17:45:50 -0800668 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
Kent Overstreetbdb53202013-11-23 17:26:46 -0800669 if (!bio)
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200670 return NULL;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800671 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600672 bio->bi_opf = bio_src->bi_opf;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800673 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
674 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200675
Adrian Hunter7afafc82016-08-16 10:59:35 +0300676 switch (bio_op(bio)) {
677 case REQ_OP_DISCARD:
678 case REQ_OP_SECURE_ERASE:
679 break;
680 case REQ_OP_WRITE_SAME:
Kent Overstreet8423ae32014-02-10 17:45:50 -0800681 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
Adrian Hunter7afafc82016-08-16 10:59:35 +0300682 break;
683 default:
684 bio_for_each_segment(bv, bio_src, iter)
685 bio->bi_io_vec[bio->bi_vcnt++] = bv;
686 break;
Kent Overstreet8423ae32014-02-10 17:45:50 -0800687 }
688
Kent Overstreetbdb53202013-11-23 17:26:46 -0800689 if (bio_integrity(bio_src)) {
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200690 int ret;
691
Kent Overstreetbdb53202013-11-23 17:26:46 -0800692 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
Li Zefan059ea332009-03-09 10:42:45 +0100693 if (ret < 0) {
Kent Overstreetbdb53202013-11-23 17:26:46 -0800694 bio_put(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200695 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100696 }
Peter Osterlund36763472005-09-06 15:16:42 -0700697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Paolo Valente20bd7232016-07-27 07:22:05 +0200699 bio_clone_blkcg_association(bio, bio_src);
700
Kent Overstreetbdb53202013-11-23 17:26:46 -0800701 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700703EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
705/**
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800706 * bio_add_pc_page - attempt to add page to bio
707 * @q: the target queue
708 * @bio: destination bio
709 * @page: page to add
710 * @len: vec entry length
711 * @offset: vec entry offset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800713 * Attempt to add a page to the bio_vec maplist. This can fail for a
714 * number of reasons, such as the bio being full or target block device
715 * limitations. The target block device must allow bio's up to PAGE_SIZE,
716 * so it is always possible to add a single page to an empty bio.
717 *
718 * This should only be used by REQ_PC bios.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800720int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
721 *page, unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722{
723 int retried_segments = 0;
724 struct bio_vec *bvec;
725
726 /*
727 * cloned bio must not modify vec list
728 */
729 if (unlikely(bio_flagged(bio, BIO_CLONED)))
730 return 0;
731
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800732 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return 0;
734
Jens Axboe80cfd542006-01-06 09:43:28 +0100735 /*
736 * For filesystems with a blocksize smaller than the pagesize
737 * we will often be called with the same page as last time and
738 * a consecutive offset. Optimize this special case.
739 */
740 if (bio->bi_vcnt > 0) {
741 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
742
743 if (page == prev->bv_page &&
744 offset == prev->bv_offset + prev->bv_len) {
745 prev->bv_len += len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800746 bio->bi_iter.bi_size += len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100747 goto done;
748 }
Jens Axboe66cb45a2014-06-24 16:22:24 -0600749
750 /*
751 * If the queue doesn't support SG gaps and adding this
752 * offset would create a gap, disallow it.
753 */
Keith Busch03100aa2015-08-19 14:24:05 -0700754 if (bvec_gap_to_prev(q, prev, offset))
Jens Axboe66cb45a2014-06-24 16:22:24 -0600755 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100756 }
757
758 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return 0;
760
761 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 * setup the new entry, we might clear it again later if we
763 * cannot add the page
764 */
765 bvec = &bio->bi_io_vec[bio->bi_vcnt];
766 bvec->bv_page = page;
767 bvec->bv_len = len;
768 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800769 bio->bi_vcnt++;
770 bio->bi_phys_segments++;
771 bio->bi_iter.bi_size += len;
772
773 /*
774 * Perform a recount if the number of segments is greater
775 * than queue_max_segments(q).
776 */
777
778 while (bio->bi_phys_segments > queue_max_segments(q)) {
779
780 if (retried_segments)
781 goto failed;
782
783 retried_segments = 1;
784 blk_recount_segments(q, bio);
785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 /* If we may be able to merge these biovecs, force a recount */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800788 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600789 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Jens Axboe80cfd542006-01-06 09:43:28 +0100791 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800793
794 failed:
795 bvec->bv_page = NULL;
796 bvec->bv_len = 0;
797 bvec->bv_offset = 0;
798 bio->bi_vcnt--;
799 bio->bi_iter.bi_size -= len;
800 blk_recount_segments(q, bio);
801 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200803EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600804
805/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 * bio_add_page - attempt to add page to bio
807 * @bio: destination bio
808 * @page: page to add
809 * @len: vec entry length
810 * @offset: vec entry offset
811 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800812 * Attempt to add a page to the bio_vec maplist. This will only fail
813 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800815int bio_add_page(struct bio *bio, struct page *page,
816 unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817{
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800818 struct bio_vec *bv;
Jens Axboe762380a2014-06-05 13:38:39 -0600819
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800820 /*
821 * cloned bio must not modify vec list
822 */
823 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
824 return 0;
Jens Axboe58a49152014-06-10 12:53:56 -0600825
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800826 /*
827 * For filesystems with a blocksize smaller than the pagesize
828 * we will often be called with the same page as last time and
829 * a consecutive offset. Optimize this special case.
830 */
831 if (bio->bi_vcnt > 0) {
832 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
833
834 if (page == bv->bv_page &&
835 offset == bv->bv_offset + bv->bv_len) {
836 bv->bv_len += len;
837 goto done;
838 }
839 }
840
841 if (bio->bi_vcnt >= bio->bi_max_vecs)
842 return 0;
843
844 bv = &bio->bi_io_vec[bio->bi_vcnt];
845 bv->bv_page = page;
846 bv->bv_len = len;
847 bv->bv_offset = offset;
848
849 bio->bi_vcnt++;
850done:
851 bio->bi_iter.bi_size += len;
852 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200854EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Kent Overstreet9e882242012-09-10 14:41:12 -0700856struct submit_bio_ret {
857 struct completion event;
858 int error;
859};
860
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200861static void submit_bio_wait_endio(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700862{
863 struct submit_bio_ret *ret = bio->bi_private;
864
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200865 ret->error = bio->bi_error;
Kent Overstreet9e882242012-09-10 14:41:12 -0700866 complete(&ret->event);
867}
868
869/**
870 * submit_bio_wait - submit a bio, and wait until it completes
Kent Overstreet9e882242012-09-10 14:41:12 -0700871 * @bio: The &struct bio which describes the I/O
872 *
873 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
874 * bio_endio() on failure.
875 */
Mike Christie4e49ea42016-06-05 14:31:41 -0500876int submit_bio_wait(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700877{
878 struct submit_bio_ret ret;
879
Kent Overstreet9e882242012-09-10 14:41:12 -0700880 init_completion(&ret.event);
881 bio->bi_private = &ret;
882 bio->bi_end_io = submit_bio_wait_endio;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600883 bio->bi_opf |= REQ_SYNC;
Mike Christie4e49ea42016-06-05 14:31:41 -0500884 submit_bio(bio);
Stephane Gasparinid57d6112016-02-09 17:07:38 +0100885 wait_for_completion_io(&ret.event);
Kent Overstreet9e882242012-09-10 14:41:12 -0700886
887 return ret.error;
888}
889EXPORT_SYMBOL(submit_bio_wait);
890
Kent Overstreet054bdf62012-09-28 13:17:55 -0700891/**
892 * bio_advance - increment/complete a bio by some number of bytes
893 * @bio: bio to advance
894 * @bytes: number of bytes to complete
895 *
896 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
897 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
898 * be updated on the last bvec as well.
899 *
900 * @bio will then represent the remaining, uncompleted portion of the io.
901 */
902void bio_advance(struct bio *bio, unsigned bytes)
903{
904 if (bio_integrity(bio))
905 bio_integrity_advance(bio, bytes);
906
Kent Overstreet4550dd62013-08-07 14:26:21 -0700907 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -0700908}
909EXPORT_SYMBOL(bio_advance);
910
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700911/**
Kent Overstreeta0787602012-09-10 14:03:28 -0700912 * bio_alloc_pages - allocates a single page for each bvec in a bio
913 * @bio: bio to allocate pages for
914 * @gfp_mask: flags for allocation
915 *
916 * Allocates pages up to @bio->bi_vcnt.
917 *
918 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
919 * freed.
920 */
921int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
922{
923 int i;
924 struct bio_vec *bv;
925
926 bio_for_each_segment_all(bv, bio, i) {
927 bv->bv_page = alloc_page(gfp_mask);
928 if (!bv->bv_page) {
929 while (--bv >= bio->bi_io_vec)
930 __free_page(bv->bv_page);
931 return -ENOMEM;
932 }
933 }
934
935 return 0;
936}
937EXPORT_SYMBOL(bio_alloc_pages);
938
939/**
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700940 * bio_copy_data - copy contents of data buffers from one chain of bios to
941 * another
942 * @src: source bio list
943 * @dst: destination bio list
944 *
945 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
946 * @src and @dst as linked lists of bios.
947 *
948 * Stops when it reaches the end of either @src or @dst - that is, copies
949 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
950 */
951void bio_copy_data(struct bio *dst, struct bio *src)
952{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700953 struct bvec_iter src_iter, dst_iter;
954 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700955 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700956 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700957
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700958 src_iter = src->bi_iter;
959 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700960
961 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700962 if (!src_iter.bi_size) {
963 src = src->bi_next;
964 if (!src)
965 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700966
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700967 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700968 }
969
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700970 if (!dst_iter.bi_size) {
971 dst = dst->bi_next;
972 if (!dst)
973 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700974
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700975 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700976 }
977
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700978 src_bv = bio_iter_iovec(src, src_iter);
979 dst_bv = bio_iter_iovec(dst, dst_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700980
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700981 bytes = min(src_bv.bv_len, dst_bv.bv_len);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700982
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700983 src_p = kmap_atomic(src_bv.bv_page);
984 dst_p = kmap_atomic(dst_bv.bv_page);
985
986 memcpy(dst_p + dst_bv.bv_offset,
987 src_p + src_bv.bv_offset,
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700988 bytes);
989
990 kunmap_atomic(dst_p);
991 kunmap_atomic(src_p);
992
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700993 bio_advance_iter(src, &src_iter, bytes);
994 bio_advance_iter(dst, &dst_iter, bytes);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700995 }
996}
997EXPORT_SYMBOL(bio_copy_data);
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001000 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001001 struct iov_iter iter;
1002 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003};
1004
Fabian Frederick7410b3c2014-04-22 15:09:07 -06001005static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001006 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007{
Jens Axboef3f63c12010-10-29 11:46:56 -06001008 if (iov_count > UIO_MAXIOV)
1009 return NULL;
1010
Kent Overstreetc8db4442013-11-22 19:39:06 -08001011 return kmalloc(sizeof(struct bio_map_data) +
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001012 sizeof(struct iovec) * iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013}
1014
Dongsu Park9124d3f2015-01-18 16:16:34 +01001015/**
1016 * bio_copy_from_iter - copy all pages from iov_iter to bio
1017 * @bio: The &struct bio which describes the I/O as destination
1018 * @iter: iov_iter as source
1019 *
1020 * Copy all pages from iov_iter to bio.
1021 * Returns 0 on success, or error on failure.
1022 */
1023static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001024{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001025 int i;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001026 struct bio_vec *bvec;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001027
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001028 bio_for_each_segment_all(bvec, bio, i) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001029 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001030
Dongsu Park9124d3f2015-01-18 16:16:34 +01001031 ret = copy_page_from_iter(bvec->bv_page,
1032 bvec->bv_offset,
1033 bvec->bv_len,
1034 &iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001035
Dongsu Park9124d3f2015-01-18 16:16:34 +01001036 if (!iov_iter_count(&iter))
1037 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001038
Dongsu Park9124d3f2015-01-18 16:16:34 +01001039 if (ret < bvec->bv_len)
1040 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001041 }
1042
Dongsu Park9124d3f2015-01-18 16:16:34 +01001043 return 0;
1044}
1045
1046/**
1047 * bio_copy_to_iter - copy all pages from bio to iov_iter
1048 * @bio: The &struct bio which describes the I/O as source
1049 * @iter: iov_iter as destination
1050 *
1051 * Copy all pages from bio to iov_iter.
1052 * Returns 0 on success, or error on failure.
1053 */
1054static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1055{
1056 int i;
1057 struct bio_vec *bvec;
1058
1059 bio_for_each_segment_all(bvec, bio, i) {
1060 ssize_t ret;
1061
1062 ret = copy_page_to_iter(bvec->bv_page,
1063 bvec->bv_offset,
1064 bvec->bv_len,
1065 &iter);
1066
1067 if (!iov_iter_count(&iter))
1068 break;
1069
1070 if (ret < bvec->bv_len)
1071 return -EFAULT;
1072 }
1073
1074 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001075}
1076
Guoqing Jiang491221f2016-09-22 03:10:01 -04001077void bio_free_pages(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001078{
1079 struct bio_vec *bvec;
1080 int i;
1081
1082 bio_for_each_segment_all(bvec, bio, i)
1083 __free_page(bvec->bv_page);
1084}
Guoqing Jiang491221f2016-09-22 03:10:01 -04001085EXPORT_SYMBOL(bio_free_pages);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087/**
1088 * bio_uncopy_user - finish previously mapped bio
1089 * @bio: bio being terminated
1090 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001091 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 * to user space in case of a read.
1093 */
1094int bio_uncopy_user(struct bio *bio)
1095{
1096 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001097 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Roland Dreier35dc2482013-08-05 17:55:01 -07001099 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1100 /*
1101 * if we're in a workqueue, the request is orphaned, so
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001102 * don't copy into a random user address space, just free
1103 * and return -EINTR so user space doesn't expect any data.
Roland Dreier35dc2482013-08-05 17:55:01 -07001104 */
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001105 if (!current->mm)
1106 ret = -EINTR;
1107 else if (bio_data_dir(bio) == READ)
Dongsu Park9124d3f2015-01-18 16:16:34 +01001108 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001109 if (bmd->is_our_pages)
1110 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001111 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001112 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 bio_put(bio);
1114 return ret;
1115}
1116
1117/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001118 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001119 * @q: destination block queue
1120 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1121 * @iter: iovec iterator
1122 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 *
1124 * Prepares and returns a bio for indirect user io, bouncing data
1125 * to/from kernel pages as necessary. Must be paired with
1126 * call bio_uncopy_user() on io completion.
1127 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001128struct bio *bio_copy_user_iov(struct request_queue *q,
1129 struct rq_map_data *map_data,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001130 const struct iov_iter *iter,
1131 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 struct page *page;
1135 struct bio *bio;
1136 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001137 int nr_pages = 0;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001138 unsigned int len = iter->count;
Geliang Tangbd5cece2015-11-21 17:27:31 +08001139 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001141 for (i = 0; i < iter->nr_segs; i++) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001142 unsigned long uaddr;
1143 unsigned long end;
1144 unsigned long start;
1145
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001146 uaddr = (unsigned long) iter->iov[i].iov_base;
1147 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1148 >> PAGE_SHIFT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001149 start = uaddr >> PAGE_SHIFT;
1150
Jens Axboecb4644c2010-11-10 14:36:25 +01001151 /*
1152 * Overflow, abort
1153 */
1154 if (end < start)
1155 return ERR_PTR(-EINVAL);
1156
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001157 nr_pages += end - start;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001158 }
1159
FUJITA Tomonori69838722009-04-28 20:24:29 +02001160 if (offset)
1161 nr_pages++;
1162
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001163 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (!bmd)
1165 return ERR_PTR(-ENOMEM);
1166
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001167 /*
1168 * We need to do a deep copy of the iov_iter including the iovecs.
1169 * The caller provided iov might point to an on-stack or otherwise
1170 * shortlived one.
1171 */
1172 bmd->is_our_pages = map_data ? 0 : 1;
1173 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
Al Viroed35ded2017-09-24 10:21:15 -04001174 bmd->iter = *iter;
1175 bmd->iter.iov = bmd->iov;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001178 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (!bio)
1180 goto out_bmd;
1181
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001182 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001183 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
1185 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001186
1187 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001188 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001189 i = map_data->offset / PAGE_SIZE;
1190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001192 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001194 bytes -= offset;
1195
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 if (bytes > len)
1197 bytes = len;
1198
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001199 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001200 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001201 ret = -ENOMEM;
1202 break;
1203 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001204
1205 page = map_data->pages[i / nr_pages];
1206 page += (i % nr_pages);
1207
1208 i++;
1209 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001210 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001211 if (!page) {
1212 ret = -ENOMEM;
1213 break;
1214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
Jérôme Glisse056066d2019-04-10 16:27:51 -04001217 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) {
1218 if (!map_data)
1219 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 break;
Jérôme Glisse056066d2019-04-10 16:27:51 -04001221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
1223 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001224 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 }
1226
1227 if (ret)
1228 goto cleanup;
1229
1230 /*
1231 * success
1232 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001233 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001234 (map_data && map_data->from_user)) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001235 ret = bio_copy_from_iter(bio, *iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001236 if (ret)
1237 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 }
1239
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001240 bio->bi_private = bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 return bio;
1242cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001243 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001244 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 bio_put(bio);
1246out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001247 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 return ERR_PTR(ret);
1249}
1250
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001251/**
1252 * bio_map_user_iov - map user iovec into bio
1253 * @q: the struct request_queue for the bio
1254 * @iter: iovec iterator
1255 * @gfp_mask: memory allocation flags
1256 *
1257 * Map the user space address into a bio suitable for io to a block
1258 * device. Returns an error pointer in case of error.
1259 */
1260struct bio *bio_map_user_iov(struct request_queue *q,
1261 const struct iov_iter *iter,
1262 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001264 int j;
James Bottomley f1970ba2005-06-20 14:06:52 +02001265 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 struct page **pages;
1267 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001268 int cur_page = 0;
1269 int ret, offset;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001270 struct iov_iter i;
1271 struct iovec iov;
Al Viroe67dfe72017-09-23 15:51:23 -04001272 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001274 iov_for_each(iov, i, *iter) {
1275 unsigned long uaddr = (unsigned long) iov.iov_base;
1276 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001277 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1278 unsigned long start = uaddr >> PAGE_SHIFT;
1279
Jens Axboecb4644c2010-11-10 14:36:25 +01001280 /*
1281 * Overflow, abort
1282 */
1283 if (end < start)
1284 return ERR_PTR(-EINVAL);
1285
James Bottomley f1970ba2005-06-20 14:06:52 +02001286 nr_pages += end - start;
1287 /*
Linus Walleija441b0d2016-09-14 14:32:52 +02001288 * buffer must be aligned to at least logical block size for now
James Bottomley f1970ba2005-06-20 14:06:52 +02001289 */
Mike Christiead2d7222006-12-01 10:40:20 +01001290 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +02001291 return ERR_PTR(-EINVAL);
1292 }
1293
1294 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 return ERR_PTR(-EINVAL);
1296
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001297 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if (!bio)
1299 return ERR_PTR(-ENOMEM);
1300
1301 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001302 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 if (!pages)
1304 goto out;
1305
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001306 iov_for_each(iov, i, *iter) {
1307 unsigned long uaddr = (unsigned long) iov.iov_base;
1308 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001309 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1310 unsigned long start = uaddr >> PAGE_SHIFT;
1311 const int local_nr_pages = end - start;
1312 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001313
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001314 ret = get_user_pages_fast(uaddr, local_nr_pages,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001315 (iter->type & WRITE) != WRITE,
1316 &pages[cur_page]);
Al Viroe67dfe72017-09-23 15:51:23 -04001317 if (unlikely(ret < local_nr_pages)) {
1318 for (j = cur_page; j < page_limit; j++) {
1319 if (!pages[j])
1320 break;
1321 put_page(pages[j]);
1322 }
Jens Axboe99172152006-06-16 13:02:29 +02001323 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001324 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Geliang Tangbd5cece2015-11-21 17:27:31 +08001327 offset = offset_in_page(uaddr);
James Bottomley f1970ba2005-06-20 14:06:52 +02001328 for (j = cur_page; j < page_limit; j++) {
1329 unsigned int bytes = PAGE_SIZE - offset;
Vitaly Mayatskikh5444d8a2017-09-22 01:18:39 -04001330 unsigned short prev_bi_vcnt = bio->bi_vcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
James Bottomley f1970ba2005-06-20 14:06:52 +02001332 if (len <= 0)
1333 break;
1334
1335 if (bytes > len)
1336 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
James Bottomley f1970ba2005-06-20 14:06:52 +02001338 /*
1339 * sorry...
1340 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001341 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1342 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001343 break;
1344
Vitaly Mayatskikh5444d8a2017-09-22 01:18:39 -04001345 /*
1346 * check if vector was merged with previous
1347 * drop page reference if needed
1348 */
1349 if (bio->bi_vcnt == prev_bi_vcnt)
1350 put_page(pages[j]);
1351
James Bottomley f1970ba2005-06-20 14:06:52 +02001352 len -= bytes;
1353 offset = 0;
1354 }
1355
1356 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001358 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001360 while (j < page_limit)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001361 put_page(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 }
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 kfree(pages);
1365
1366 /*
1367 * set data direction, and check if mapped pages need bouncing
1368 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001369 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001370 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001372 bio_set_flag(bio, BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001373
1374 /*
1375 * subtle -- if __bio_map_user() ended up bouncing a bio,
1376 * it would normally disappear when its bi_end_io is run.
1377 * however, we need it for the unmap, so grab an extra
1378 * reference to it
1379 */
1380 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001382
1383 out_unmap:
Al Viroe67dfe72017-09-23 15:51:23 -04001384 bio_for_each_segment_all(bvec, bio, j) {
1385 put_page(bvec->bv_page);
James Bottomley f1970ba2005-06-20 14:06:52 +02001386 }
1387 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 kfree(pages);
1389 bio_put(bio);
1390 return ERR_PTR(ret);
1391}
1392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393static void __bio_unmap_user(struct bio *bio)
1394{
1395 struct bio_vec *bvec;
1396 int i;
1397
1398 /*
1399 * make sure we dirty pages we wrote to
1400 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001401 bio_for_each_segment_all(bvec, bio, i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 if (bio_data_dir(bio) == READ)
1403 set_page_dirty_lock(bvec->bv_page);
1404
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001405 put_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 }
1407
1408 bio_put(bio);
1409}
1410
1411/**
1412 * bio_unmap_user - unmap a bio
1413 * @bio: the bio being unmapped
1414 *
1415 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1416 * a process context.
1417 *
1418 * bio_unmap_user() may sleep.
1419 */
1420void bio_unmap_user(struct bio *bio)
1421{
1422 __bio_unmap_user(bio);
1423 bio_put(bio);
1424}
1425
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001426static void bio_map_kern_endio(struct bio *bio)
Jens Axboeb8238252005-06-20 14:05:27 +02001427{
Jens Axboeb8238252005-06-20 14:05:27 +02001428 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001429}
1430
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001431/**
1432 * bio_map_kern - map kernel address into bio
1433 * @q: the struct request_queue for the bio
1434 * @data: pointer to buffer to map
1435 * @len: length in bytes
1436 * @gfp_mask: allocation flags for bio allocation
1437 *
1438 * Map the kernel address into a bio suitable for io to a block
1439 * device. Returns an error pointer in case of error.
1440 */
1441struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1442 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001443{
1444 unsigned long kaddr = (unsigned long)data;
1445 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1446 unsigned long start = kaddr >> PAGE_SHIFT;
1447 const int nr_pages = end - start;
1448 int offset, i;
1449 struct bio *bio;
1450
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001451 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001452 if (!bio)
1453 return ERR_PTR(-ENOMEM);
1454
1455 offset = offset_in_page(kaddr);
1456 for (i = 0; i < nr_pages; i++) {
1457 unsigned int bytes = PAGE_SIZE - offset;
1458
1459 if (len <= 0)
1460 break;
1461
1462 if (bytes > len)
1463 bytes = len;
1464
Mike Christiedefd94b2005-12-05 02:37:06 -06001465 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001466 offset) < bytes) {
1467 /* we don't support partial mappings */
1468 bio_put(bio);
1469 return ERR_PTR(-EINVAL);
1470 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001471
1472 data += bytes;
1473 len -= bytes;
1474 offset = 0;
1475 }
1476
Jens Axboeb8238252005-06-20 14:05:27 +02001477 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001478 return bio;
1479}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001480EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001481
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001482static void bio_copy_kern_endio(struct bio *bio)
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001483{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001484 bio_free_pages(bio);
1485 bio_put(bio);
1486}
1487
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001488static void bio_copy_kern_endio_read(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001489{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001490 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001491 struct bio_vec *bvec;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001492 int i;
1493
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001494 bio_for_each_segment_all(bvec, bio, i) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001495 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001496 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001497 }
1498
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001499 bio_copy_kern_endio(bio);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001500}
1501
1502/**
1503 * bio_copy_kern - copy kernel address into bio
1504 * @q: the struct request_queue for the bio
1505 * @data: pointer to buffer to copy
1506 * @len: length in bytes
1507 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001508 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001509 *
1510 * copy the kernel address into a bio suitable for io to a block
1511 * device. Returns an error pointer in case of error.
1512 */
1513struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1514 gfp_t gfp_mask, int reading)
1515{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001516 unsigned long kaddr = (unsigned long)data;
1517 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1518 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001519 struct bio *bio;
1520 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001521 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001522
Christoph Hellwig42d26832015-01-18 16:16:28 +01001523 /*
1524 * Overflow, abort
1525 */
1526 if (end < start)
1527 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001528
Christoph Hellwig42d26832015-01-18 16:16:28 +01001529 nr_pages = end - start;
1530 bio = bio_kmalloc(gfp_mask, nr_pages);
1531 if (!bio)
1532 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001533
Christoph Hellwig42d26832015-01-18 16:16:28 +01001534 while (len) {
1535 struct page *page;
1536 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001537
Christoph Hellwig42d26832015-01-18 16:16:28 +01001538 if (bytes > len)
1539 bytes = len;
1540
1541 page = alloc_page(q->bounce_gfp | gfp_mask);
1542 if (!page)
1543 goto cleanup;
1544
1545 if (!reading)
1546 memcpy(page_address(page), p, bytes);
1547
1548 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1549 break;
1550
1551 len -= bytes;
1552 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001553 }
1554
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001555 if (reading) {
1556 bio->bi_end_io = bio_copy_kern_endio_read;
1557 bio->bi_private = data;
1558 } else {
1559 bio->bi_end_io = bio_copy_kern_endio;
Mike Christie95fe6c12016-06-05 14:31:48 -05001560 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001561 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001562
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001563 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001564
1565cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001566 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001567 bio_put(bio);
1568 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001569}
1570
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571/*
1572 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1573 * for performing direct-IO in BIOs.
1574 *
1575 * The problem is that we cannot run set_page_dirty() from interrupt context
1576 * because the required locks are not interrupt-safe. So what we can do is to
1577 * mark the pages dirty _before_ performing IO. And in interrupt context,
1578 * check that the pages are still dirty. If so, fine. If not, redirty them
1579 * in process context.
1580 *
1581 * We special-case compound pages here: normally this means reads into hugetlb
1582 * pages. The logic in here doesn't really work right for compound pages
1583 * because the VM does not uniformly chase down the head page in all cases.
1584 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1585 * handle them at all. So we skip compound pages here at an early stage.
1586 *
1587 * Note that this code is very hard to test under normal circumstances because
1588 * direct-io pins the pages with get_user_pages(). This makes
1589 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001590 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 * pagecache.
1592 *
1593 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1594 * deferred bio dirtying paths.
1595 */
1596
1597/*
1598 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1599 */
1600void bio_set_pages_dirty(struct bio *bio)
1601{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001602 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 int i;
1604
Kent Overstreetcb34e052012-09-05 15:22:02 -07001605 bio_for_each_segment_all(bvec, bio, i) {
1606 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608 if (page && !PageCompound(page))
1609 set_page_dirty_lock(page);
1610 }
1611}
1612
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001613static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001615 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 int i;
1617
Kent Overstreetcb34e052012-09-05 15:22:02 -07001618 bio_for_each_segment_all(bvec, bio, i) {
1619 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
1621 if (page)
1622 put_page(page);
1623 }
1624}
1625
1626/*
1627 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1628 * If they are, then fine. If, however, some pages are clean then they must
1629 * have been written out during the direct-IO read. So we take another ref on
1630 * the BIO and the offending pages and re-dirty the pages in process context.
1631 *
1632 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +03001633 * here on. It will run one put_page() against each page and will run one
1634 * bio_put() against the BIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 */
1636
David Howells65f27f32006-11-22 14:55:48 +00001637static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
David Howells65f27f32006-11-22 14:55:48 +00001639static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640static DEFINE_SPINLOCK(bio_dirty_lock);
1641static struct bio *bio_dirty_list;
1642
1643/*
1644 * This runs in process context
1645 */
David Howells65f27f32006-11-22 14:55:48 +00001646static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 unsigned long flags;
1649 struct bio *bio;
1650
1651 spin_lock_irqsave(&bio_dirty_lock, flags);
1652 bio = bio_dirty_list;
1653 bio_dirty_list = NULL;
1654 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1655
1656 while (bio) {
1657 struct bio *next = bio->bi_private;
1658
1659 bio_set_pages_dirty(bio);
1660 bio_release_pages(bio);
1661 bio_put(bio);
1662 bio = next;
1663 }
1664}
1665
1666void bio_check_pages_dirty(struct bio *bio)
1667{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001668 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 int nr_clean_pages = 0;
1670 int i;
1671
Kent Overstreetcb34e052012-09-05 15:22:02 -07001672 bio_for_each_segment_all(bvec, bio, i) {
1673 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
1675 if (PageDirty(page) || PageCompound(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001676 put_page(page);
Kent Overstreetcb34e052012-09-05 15:22:02 -07001677 bvec->bv_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 } else {
1679 nr_clean_pages++;
1680 }
1681 }
1682
1683 if (nr_clean_pages) {
1684 unsigned long flags;
1685
1686 spin_lock_irqsave(&bio_dirty_lock, flags);
1687 bio->bi_private = bio_dirty_list;
1688 bio_dirty_list = bio;
1689 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1690 schedule_work(&bio_dirty_work);
1691 } else {
1692 bio_put(bio);
1693 }
1694}
1695
Gu Zheng394ffa52014-11-24 11:05:22 +08001696void generic_start_io_acct(int rw, unsigned long sectors,
1697 struct hd_struct *part)
1698{
1699 int cpu = part_stat_lock();
1700
1701 part_round_stats(cpu, part);
1702 part_stat_inc(cpu, part, ios[rw]);
1703 part_stat_add(cpu, part, sectors[rw], sectors);
1704 part_inc_in_flight(part, rw);
1705
1706 part_stat_unlock();
1707}
1708EXPORT_SYMBOL(generic_start_io_acct);
1709
1710void generic_end_io_acct(int rw, struct hd_struct *part,
1711 unsigned long start_time)
1712{
1713 unsigned long duration = jiffies - start_time;
1714 int cpu = part_stat_lock();
1715
1716 part_stat_add(cpu, part, ticks[rw], duration);
1717 part_round_stats(cpu, part);
1718 part_dec_in_flight(part, rw);
1719
1720 part_stat_unlock();
1721}
1722EXPORT_SYMBOL(generic_end_io_acct);
1723
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001724#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1725void bio_flush_dcache_pages(struct bio *bi)
1726{
Kent Overstreet79886132013-11-23 17:19:00 -08001727 struct bio_vec bvec;
1728 struct bvec_iter iter;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001729
Kent Overstreet79886132013-11-23 17:19:00 -08001730 bio_for_each_segment(bvec, bi, iter)
1731 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001732}
1733EXPORT_SYMBOL(bio_flush_dcache_pages);
1734#endif
1735
Jens Axboec4cf5262015-04-17 16:15:18 -06001736static inline bool bio_remaining_done(struct bio *bio)
1737{
1738 /*
1739 * If we're not chaining, then ->__bi_remaining is always 1 and
1740 * we always end io on the first invocation.
1741 */
1742 if (!bio_flagged(bio, BIO_CHAIN))
1743 return true;
1744
1745 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1746
Mike Snitzer326e1db2015-05-22 09:14:03 -04001747 if (atomic_dec_and_test(&bio->__bi_remaining)) {
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001748 bio_clear_flag(bio, BIO_CHAIN);
Jens Axboec4cf5262015-04-17 16:15:18 -06001749 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001750 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001751
1752 return false;
1753}
1754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755/**
1756 * bio_endio - end I/O on a bio
1757 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 *
1759 * Description:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001760 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1761 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1762 * bio unless they own it and thus know that it has an end_io function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 **/
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001764void bio_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765{
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001766again:
Christoph Hellwig2b885512016-03-11 17:34:53 +01001767 if (!bio_remaining_done(bio))
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001768 return;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001769
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001770 /*
1771 * Need to have a real endio function for chained bios, otherwise
1772 * various corner cases will break (like stacking block devices that
1773 * save/restore bi_end_io) - however, we want to avoid unbounded
1774 * recursion and blowing the stack. Tail call optimization would
1775 * handle this, but compiling with frame pointers also disables
1776 * gcc's sibling call optimization.
1777 */
1778 if (bio->bi_end_io == bio_chain_endio) {
1779 bio = __bio_chain_endio(bio);
1780 goto again;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001781 }
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001782
1783 if (bio->bi_end_io)
1784 bio->bi_end_io(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001786EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
Kent Overstreet196d38b2013-11-23 18:34:15 -08001788/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001789 * bio_split - split a bio
1790 * @bio: bio to split
1791 * @sectors: number of sectors to split from the front of @bio
1792 * @gfp: gfp mask
1793 * @bs: bio set to allocate from
1794 *
1795 * Allocates and returns a new bio which represents @sectors from the start of
1796 * @bio, and updates @bio to represent the remaining sectors.
1797 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001798 * Unless this is a discard request the newly allocated bio will point
1799 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1800 * @bio is not freed before the split.
Kent Overstreet20d01892013-11-23 18:21:01 -08001801 */
1802struct bio *bio_split(struct bio *bio, int sectors,
1803 gfp_t gfp, struct bio_set *bs)
1804{
1805 struct bio *split = NULL;
1806
1807 BUG_ON(sectors <= 0);
1808 BUG_ON(sectors >= bio_sectors(bio));
1809
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001810 /*
1811 * Discards need a mutable bio_vec to accommodate the payload
1812 * required by the DSM TRIM and UNMAP commands.
1813 */
Adrian Hunter7afafc82016-08-16 10:59:35 +03001814 if (bio_op(bio) == REQ_OP_DISCARD || bio_op(bio) == REQ_OP_SECURE_ERASE)
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001815 split = bio_clone_bioset(bio, gfp, bs);
1816 else
1817 split = bio_clone_fast(bio, gfp, bs);
1818
Kent Overstreet20d01892013-11-23 18:21:01 -08001819 if (!split)
1820 return NULL;
1821
1822 split->bi_iter.bi_size = sectors << 9;
1823
1824 if (bio_integrity(split))
1825 bio_integrity_trim(split, 0, sectors);
1826
1827 bio_advance(bio, split->bi_iter.bi_size);
1828
1829 return split;
1830}
1831EXPORT_SYMBOL(bio_split);
1832
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001833/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001834 * bio_trim - trim a bio
1835 * @bio: bio to trim
1836 * @offset: number of sectors to trim from the front of @bio
1837 * @size: size we want to trim @bio to, in sectors
1838 */
1839void bio_trim(struct bio *bio, int offset, int size)
1840{
1841 /* 'bio' is a cloned bio which we need to trim to match
1842 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001843 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001844
1845 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001846 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001847 return;
1848
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001849 bio_clear_flag(bio, BIO_SEG_VALID);
Kent Overstreet6678d832013-08-07 11:14:32 -07001850
1851 bio_advance(bio, offset << 9);
1852
Kent Overstreet4f024f32013-10-11 15:44:27 -07001853 bio->bi_iter.bi_size = size;
Kent Overstreet6678d832013-08-07 11:14:32 -07001854}
1855EXPORT_SYMBOL_GPL(bio_trim);
1856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857/*
1858 * create memory pools for biovec's in a bio_set.
1859 * use the global biovec slabs created for general use.
1860 */
Fabian Fredericka6c39cb4f2014-04-22 15:09:05 -06001861mempool_t *biovec_create_pool(int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862{
Christoph Hellwiged996a52016-07-19 11:28:42 +02001863 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Kent Overstreet9f060e22012-10-12 15:29:33 -07001865 return mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
1868void bioset_free(struct bio_set *bs)
1869{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001870 if (bs->rescue_workqueue)
1871 destroy_workqueue(bs->rescue_workqueue);
1872
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 if (bs->bio_pool)
1874 mempool_destroy(bs->bio_pool);
1875
Kent Overstreet9f060e22012-10-12 15:29:33 -07001876 if (bs->bvec_pool)
1877 mempool_destroy(bs->bvec_pool);
1878
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001879 bioset_integrity_free(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001880 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882 kfree(bs);
1883}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001884EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001886static struct bio_set *__bioset_create(unsigned int pool_size,
1887 unsigned int front_pad,
1888 bool create_bvec_pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
Jens Axboe392ddc32008-12-23 12:42:54 +01001890 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001891 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Jens Axboe1b434492008-10-22 20:32:58 +02001893 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 if (!bs)
1895 return NULL;
1896
Jens Axboebb799ca2008-12-10 15:35:05 +01001897 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001898
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001899 spin_lock_init(&bs->rescue_lock);
1900 bio_list_init(&bs->rescue_list);
1901 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1902
Jens Axboe392ddc32008-12-23 12:42:54 +01001903 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001904 if (!bs->bio_slab) {
1905 kfree(bs);
1906 return NULL;
1907 }
1908
1909 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 if (!bs->bio_pool)
1911 goto bad;
1912
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001913 if (create_bvec_pool) {
1914 bs->bvec_pool = biovec_create_pool(pool_size);
1915 if (!bs->bvec_pool)
1916 goto bad;
1917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001919 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1920 if (!bs->rescue_workqueue)
1921 goto bad;
1922
1923 return bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924bad:
1925 bioset_free(bs);
1926 return NULL;
1927}
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001928
1929/**
1930 * bioset_create - Create a bio_set
1931 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1932 * @front_pad: Number of bytes to allocate in front of the returned bio
1933 *
1934 * Description:
1935 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1936 * to ask for a number of bytes to be allocated in front of the bio.
1937 * Front pad allocation is useful for embedding the bio inside
1938 * another structure, to avoid allocating extra data to go with the bio.
1939 * Note that the bio must be embedded at the END of that structure always,
1940 * or things will break badly.
1941 */
1942struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1943{
1944 return __bioset_create(pool_size, front_pad, true);
1945}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001946EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001948/**
1949 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1950 * @pool_size: Number of bio to cache in the mempool
1951 * @front_pad: Number of bytes to allocate in front of the returned bio
1952 *
1953 * Description:
1954 * Same functionality as bioset_create() except that mempool is not
1955 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1956 */
1957struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1958{
1959 return __bioset_create(pool_size, front_pad, false);
1960}
1961EXPORT_SYMBOL(bioset_create_nobvec);
1962
Tejun Heo852c7882012-03-05 13:15:27 -08001963#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04001964
1965/**
1966 * bio_associate_blkcg - associate a bio with the specified blkcg
1967 * @bio: target bio
1968 * @blkcg_css: css of the blkcg to associate
1969 *
1970 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1971 * treat @bio as if it were issued by a task which belongs to the blkcg.
1972 *
1973 * This function takes an extra reference of @blkcg_css which will be put
1974 * when @bio is released. The caller must own @bio and is responsible for
1975 * synchronizing calls to this function.
1976 */
1977int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1978{
1979 if (unlikely(bio->bi_css))
1980 return -EBUSY;
1981 css_get(blkcg_css);
1982 bio->bi_css = blkcg_css;
1983 return 0;
1984}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001985EXPORT_SYMBOL_GPL(bio_associate_blkcg);
Tejun Heo1d933cf2015-05-22 17:13:24 -04001986
Tejun Heo852c7882012-03-05 13:15:27 -08001987/**
1988 * bio_associate_current - associate a bio with %current
1989 * @bio: target bio
1990 *
1991 * Associate @bio with %current if it hasn't been associated yet. Block
1992 * layer will treat @bio as if it were issued by %current no matter which
1993 * task actually issues it.
1994 *
1995 * This function takes an extra reference of @task's io_context and blkcg
1996 * which will be put when @bio is released. The caller must own @bio,
1997 * ensure %current->io_context exists, and is responsible for synchronizing
1998 * calls to this function.
1999 */
2000int bio_associate_current(struct bio *bio)
2001{
2002 struct io_context *ioc;
Tejun Heo852c7882012-03-05 13:15:27 -08002003
Tejun Heo1d933cf2015-05-22 17:13:24 -04002004 if (bio->bi_css)
Tejun Heo852c7882012-03-05 13:15:27 -08002005 return -EBUSY;
2006
2007 ioc = current->io_context;
2008 if (!ioc)
2009 return -ENOENT;
2010
Tejun Heo852c7882012-03-05 13:15:27 -08002011 get_io_context_active(ioc);
2012 bio->bi_ioc = ioc;
Tejun Heoc165b3e2015-08-18 14:55:29 -07002013 bio->bi_css = task_get_css(current, io_cgrp_id);
Tejun Heo852c7882012-03-05 13:15:27 -08002014 return 0;
2015}
Tejun Heo5aa2a962015-07-23 14:27:09 -04002016EXPORT_SYMBOL_GPL(bio_associate_current);
Tejun Heo852c7882012-03-05 13:15:27 -08002017
2018/**
2019 * bio_disassociate_task - undo bio_associate_current()
2020 * @bio: target bio
2021 */
2022void bio_disassociate_task(struct bio *bio)
2023{
2024 if (bio->bi_ioc) {
2025 put_io_context(bio->bi_ioc);
2026 bio->bi_ioc = NULL;
2027 }
2028 if (bio->bi_css) {
2029 css_put(bio->bi_css);
2030 bio->bi_css = NULL;
2031 }
2032}
2033
Paolo Valente20bd7232016-07-27 07:22:05 +02002034/**
2035 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2036 * @dst: destination bio
2037 * @src: source bio
2038 */
2039void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2040{
2041 if (src->bi_css)
2042 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2043}
2044
Tejun Heo852c7882012-03-05 13:15:27 -08002045#endif /* CONFIG_BLK_CGROUP */
2046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047static void __init biovec_init_slabs(void)
2048{
2049 int i;
2050
Christoph Hellwiged996a52016-07-19 11:28:42 +02002051 for (i = 0; i < BVEC_POOL_NR; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 int size;
2053 struct biovec_slab *bvs = bvec_slabs + i;
2054
Jens Axboea7fcd372008-12-05 16:10:29 +01002055 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2056 bvs->slab = NULL;
2057 continue;
2058 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 size = bvs->nr_vecs * sizeof(struct bio_vec);
2061 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002062 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
2064}
2065
2066static int __init init_bio(void)
2067{
Jens Axboebb799ca2008-12-10 15:35:05 +01002068 bio_slab_max = 2;
2069 bio_slab_nr = 0;
2070 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2071 if (!bio_slabs)
2072 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002074 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 biovec_init_slabs();
2076
Jens Axboebb799ca2008-12-10 15:35:05 +01002077 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (!fs_bio_set)
2079 panic("bio: can't allocate bios\n");
2080
Martin K. Petersena91a2782011-03-17 11:11:05 +01002081 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2082 panic("bio: can't create integrity pool\n");
2083
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 return 0;
2085}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086subsys_initcall(init_bio);