blob: 7c0f0d191eebcde5c9352eeee2ffef4c949dfd9a [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
Neeraj Soni36c65122018-04-18 21:04:46 +0530568static inline void bio_clone_crypt_key(struct bio *dst, const struct bio *src)
569{
570#ifdef CONFIG_PFK
Neeraj Soni36c65122018-04-18 21:04:46 +0530571 dst->bi_iter.bi_dun = src->bi_iter.bi_dun;
Jaegeuk Kim3bd012f2018-04-20 19:26:09 -0700572#ifdef CONFIG_DM_DEFAULT_KEY
573 dst->bi_crypt_key = src->bi_crypt_key;
574 dst->bi_crypt_skip = src->bi_crypt_skip;
575#endif
Neeraj Soni36c65122018-04-18 21:04:46 +0530576 dst->bi_dio_inode = src->bi_dio_inode;
Jaegeuk Kim5190c812018-04-22 16:05:02 -0700577#endif
Neeraj Soni36c65122018-04-18 21:04:46 +0530578}
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800581 * __bio_clone_fast - clone a bio that shares the original bio's biovec
582 * @bio: destination bio
583 * @bio_src: bio to clone
584 *
585 * Clone a &bio. Caller will own the returned bio, but not
586 * the actual data it points to. Reference count of returned
587 * bio will be one.
588 *
589 * Caller must ensure that @bio_src is not freed before @bio.
590 */
591void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
592{
Christoph Hellwiged996a52016-07-19 11:28:42 +0200593 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
Kent Overstreet59d276f2013-11-23 18:19:27 -0800594
595 /*
596 * most users will be overriding ->bi_bdev with a new target,
597 * so we don't set nor calculate new physical/hw segment counts here
598 */
599 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600600 bio_set_flag(bio, BIO_CLONED);
Jens Axboe1eff9d32016-08-05 15:35:16 -0600601 bio->bi_opf = bio_src->bi_opf;
Kent Overstreet59d276f2013-11-23 18:19:27 -0800602 bio->bi_iter = bio_src->bi_iter;
603 bio->bi_io_vec = bio_src->bi_io_vec;
Jaegeuk Kim5190c812018-04-22 16:05:02 -0700604#ifdef CONFIG_PFK
Neeraj Sonic692cb92018-04-18 17:20:22 +0530605 bio->bi_dio_inode = bio_src->bi_dio_inode;
Jaegeuk Kim5190c812018-04-22 16:05:02 -0700606#endif
Neeraj Soni36c65122018-04-18 21:04:46 +0530607 bio_clone_crypt_key(bio, bio_src);
Paolo Valente20bd7232016-07-27 07:22:05 +0200608 bio_clone_blkcg_association(bio, bio_src);
Kent Overstreet59d276f2013-11-23 18:19:27 -0800609}
610EXPORT_SYMBOL(__bio_clone_fast);
611
612/**
613 * bio_clone_fast - clone a bio that shares the original bio's biovec
614 * @bio: bio to clone
615 * @gfp_mask: allocation priority
616 * @bs: bio_set to allocate from
617 *
618 * Like __bio_clone_fast, only also allocates the returned bio
619 */
620struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
621{
622 struct bio *b;
623
624 b = bio_alloc_bioset(gfp_mask, 0, bs);
625 if (!b)
626 return NULL;
627
628 __bio_clone_fast(b, bio);
629
630 if (bio_integrity(bio)) {
631 int ret;
632
633 ret = bio_integrity_clone(b, bio, gfp_mask);
634
635 if (ret < 0) {
636 bio_put(b);
637 return NULL;
638 }
639 }
640
641 return b;
642}
643EXPORT_SYMBOL(bio_clone_fast);
644
645/**
Kent Overstreetbdb53202013-11-23 17:26:46 -0800646 * bio_clone_bioset - clone a bio
647 * @bio_src: bio to clone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700649 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 *
Kent Overstreetbdb53202013-11-23 17:26:46 -0800651 * Clone bio. Caller will own the returned bio, but not the actual data it
652 * points to. Reference count of returned bio will be one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 */
Kent Overstreetbdb53202013-11-23 17:26:46 -0800654struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700655 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Kent Overstreetbdb53202013-11-23 17:26:46 -0800657 struct bvec_iter iter;
658 struct bio_vec bv;
659 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Kent Overstreetbdb53202013-11-23 17:26:46 -0800661 /*
662 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
663 * bio_src->bi_io_vec to bio->bi_io_vec.
664 *
665 * We can't do that anymore, because:
666 *
667 * - The point of cloning the biovec is to produce a bio with a biovec
668 * the caller can modify: bi_idx and bi_bvec_done should be 0.
669 *
670 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
671 * we tried to clone the whole thing bio_alloc_bioset() would fail.
672 * But the clone should succeed as long as the number of biovecs we
673 * actually need to allocate is fewer than BIO_MAX_PAGES.
674 *
675 * - Lastly, bi_vcnt should not be looked at or relied upon by code
676 * that does not own the bio - reason being drivers don't use it for
677 * iterating over the biovec anymore, so expecting it to be kept up
678 * to date (i.e. for clones that share the parent biovec) is just
679 * asking for trouble and would force extra work on
680 * __bio_clone_fast() anyways.
681 */
682
Kent Overstreet8423ae32014-02-10 17:45:50 -0800683 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
Kent Overstreetbdb53202013-11-23 17:26:46 -0800684 if (!bio)
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200685 return NULL;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800686 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600687 bio->bi_opf = bio_src->bi_opf;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800688 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
689 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200690
Adrian Hunter7afafc82016-08-16 10:59:35 +0300691 switch (bio_op(bio)) {
692 case REQ_OP_DISCARD:
693 case REQ_OP_SECURE_ERASE:
694 break;
695 case REQ_OP_WRITE_SAME:
Kent Overstreet8423ae32014-02-10 17:45:50 -0800696 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
Adrian Hunter7afafc82016-08-16 10:59:35 +0300697 break;
698 default:
699 bio_for_each_segment(bv, bio_src, iter)
700 bio->bi_io_vec[bio->bi_vcnt++] = bv;
701 break;
Kent Overstreet8423ae32014-02-10 17:45:50 -0800702 }
703
Kent Overstreetbdb53202013-11-23 17:26:46 -0800704 if (bio_integrity(bio_src)) {
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200705 int ret;
706
Kent Overstreetbdb53202013-11-23 17:26:46 -0800707 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
Li Zefan059ea332009-03-09 10:42:45 +0100708 if (ret < 0) {
Kent Overstreetbdb53202013-11-23 17:26:46 -0800709 bio_put(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200710 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100711 }
Peter Osterlund36763472005-09-06 15:16:42 -0700712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Neeraj Soni69e981b2018-08-16 15:23:25 +0530714 bio_clone_crypt_key(bio, bio_src);
Paolo Valente20bd7232016-07-27 07:22:05 +0200715 bio_clone_blkcg_association(bio, bio_src);
716
Kent Overstreetbdb53202013-11-23 17:26:46 -0800717 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700719EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
721/**
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800722 * bio_add_pc_page - attempt to add page to bio
723 * @q: the target queue
724 * @bio: destination bio
725 * @page: page to add
726 * @len: vec entry length
727 * @offset: vec entry offset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800729 * Attempt to add a page to the bio_vec maplist. This can fail for a
730 * number of reasons, such as the bio being full or target block device
731 * limitations. The target block device must allow bio's up to PAGE_SIZE,
732 * so it is always possible to add a single page to an empty bio.
733 *
734 * This should only be used by REQ_PC bios.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800736int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
737 *page, unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
739 int retried_segments = 0;
740 struct bio_vec *bvec;
741
742 /*
743 * cloned bio must not modify vec list
744 */
745 if (unlikely(bio_flagged(bio, BIO_CLONED)))
746 return 0;
747
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800748 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 return 0;
750
Jens Axboe80cfd542006-01-06 09:43:28 +0100751 /*
752 * For filesystems with a blocksize smaller than the pagesize
753 * we will often be called with the same page as last time and
754 * a consecutive offset. Optimize this special case.
755 */
756 if (bio->bi_vcnt > 0) {
757 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
758
759 if (page == prev->bv_page &&
760 offset == prev->bv_offset + prev->bv_len) {
761 prev->bv_len += len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800762 bio->bi_iter.bi_size += len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100763 goto done;
764 }
Jens Axboe66cb45a2014-06-24 16:22:24 -0600765
766 /*
767 * If the queue doesn't support SG gaps and adding this
768 * offset would create a gap, disallow it.
769 */
Keith Busch03100aa2015-08-19 14:24:05 -0700770 if (bvec_gap_to_prev(q, prev, offset))
Jens Axboe66cb45a2014-06-24 16:22:24 -0600771 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100772 }
773
774 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 return 0;
776
777 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 * setup the new entry, we might clear it again later if we
779 * cannot add the page
780 */
781 bvec = &bio->bi_io_vec[bio->bi_vcnt];
782 bvec->bv_page = page;
783 bvec->bv_len = len;
784 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800785 bio->bi_vcnt++;
786 bio->bi_phys_segments++;
787 bio->bi_iter.bi_size += len;
788
789 /*
790 * Perform a recount if the number of segments is greater
791 * than queue_max_segments(q).
792 */
793
794 while (bio->bi_phys_segments > queue_max_segments(q)) {
795
796 if (retried_segments)
797 goto failed;
798
799 retried_segments = 1;
800 blk_recount_segments(q, bio);
801 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 /* If we may be able to merge these biovecs, force a recount */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800804 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600805 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Jens Axboe80cfd542006-01-06 09:43:28 +0100807 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 return len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800809
810 failed:
811 bvec->bv_page = NULL;
812 bvec->bv_len = 0;
813 bvec->bv_offset = 0;
814 bio->bi_vcnt--;
815 bio->bi_iter.bi_size -= len;
816 blk_recount_segments(q, bio);
817 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200819EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600820
821/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 * bio_add_page - attempt to add page to bio
823 * @bio: destination bio
824 * @page: page to add
825 * @len: vec entry length
826 * @offset: vec entry offset
827 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800828 * Attempt to add a page to the bio_vec maplist. This will only fail
829 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800831int bio_add_page(struct bio *bio, struct page *page,
832 unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800834 struct bio_vec *bv;
Jens Axboe762380a2014-06-05 13:38:39 -0600835
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800836 /*
837 * cloned bio must not modify vec list
838 */
839 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
840 return 0;
Jens Axboe58a49152014-06-10 12:53:56 -0600841
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800842 /*
843 * For filesystems with a blocksize smaller than the pagesize
844 * we will often be called with the same page as last time and
845 * a consecutive offset. Optimize this special case.
846 */
847 if (bio->bi_vcnt > 0) {
848 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
849
850 if (page == bv->bv_page &&
851 offset == bv->bv_offset + bv->bv_len) {
852 bv->bv_len += len;
853 goto done;
854 }
855 }
856
857 if (bio->bi_vcnt >= bio->bi_max_vecs)
858 return 0;
859
860 bv = &bio->bi_io_vec[bio->bi_vcnt];
861 bv->bv_page = page;
862 bv->bv_len = len;
863 bv->bv_offset = offset;
864
865 bio->bi_vcnt++;
866done:
867 bio->bi_iter.bi_size += len;
868 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200870EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Kent Overstreet9e882242012-09-10 14:41:12 -0700872struct submit_bio_ret {
873 struct completion event;
874 int error;
875};
876
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200877static void submit_bio_wait_endio(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700878{
879 struct submit_bio_ret *ret = bio->bi_private;
880
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200881 ret->error = bio->bi_error;
Kent Overstreet9e882242012-09-10 14:41:12 -0700882 complete(&ret->event);
883}
884
885/**
886 * submit_bio_wait - submit a bio, and wait until it completes
Kent Overstreet9e882242012-09-10 14:41:12 -0700887 * @bio: The &struct bio which describes the I/O
888 *
889 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
890 * bio_endio() on failure.
891 */
Mike Christie4e49ea42016-06-05 14:31:41 -0500892int submit_bio_wait(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700893{
894 struct submit_bio_ret ret;
895
Kent Overstreet9e882242012-09-10 14:41:12 -0700896 init_completion(&ret.event);
897 bio->bi_private = &ret;
898 bio->bi_end_io = submit_bio_wait_endio;
Jens Axboe1eff9d32016-08-05 15:35:16 -0600899 bio->bi_opf |= REQ_SYNC;
Mike Christie4e49ea42016-06-05 14:31:41 -0500900 submit_bio(bio);
Stephane Gasparinid57d6112016-02-09 17:07:38 +0100901 wait_for_completion_io(&ret.event);
Kent Overstreet9e882242012-09-10 14:41:12 -0700902
903 return ret.error;
904}
905EXPORT_SYMBOL(submit_bio_wait);
906
Kent Overstreet054bdf62012-09-28 13:17:55 -0700907/**
908 * bio_advance - increment/complete a bio by some number of bytes
909 * @bio: bio to advance
910 * @bytes: number of bytes to complete
911 *
912 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
913 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
914 * be updated on the last bvec as well.
915 *
916 * @bio will then represent the remaining, uncompleted portion of the io.
917 */
918void bio_advance(struct bio *bio, unsigned bytes)
919{
920 if (bio_integrity(bio))
921 bio_integrity_advance(bio, bytes);
922
Kent Overstreet4550dd62013-08-07 14:26:21 -0700923 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -0700924}
925EXPORT_SYMBOL(bio_advance);
926
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700927/**
Kent Overstreeta0787602012-09-10 14:03:28 -0700928 * bio_alloc_pages - allocates a single page for each bvec in a bio
929 * @bio: bio to allocate pages for
930 * @gfp_mask: flags for allocation
931 *
932 * Allocates pages up to @bio->bi_vcnt.
933 *
934 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
935 * freed.
936 */
937int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
938{
939 int i;
940 struct bio_vec *bv;
941
942 bio_for_each_segment_all(bv, bio, i) {
943 bv->bv_page = alloc_page(gfp_mask);
944 if (!bv->bv_page) {
945 while (--bv >= bio->bi_io_vec)
946 __free_page(bv->bv_page);
947 return -ENOMEM;
948 }
949 }
950
951 return 0;
952}
953EXPORT_SYMBOL(bio_alloc_pages);
954
955/**
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700956 * bio_copy_data - copy contents of data buffers from one chain of bios to
957 * another
958 * @src: source bio list
959 * @dst: destination bio list
960 *
961 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
962 * @src and @dst as linked lists of bios.
963 *
964 * Stops when it reaches the end of either @src or @dst - that is, copies
965 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
966 */
967void bio_copy_data(struct bio *dst, struct bio *src)
968{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700969 struct bvec_iter src_iter, dst_iter;
970 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700971 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700972 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700973
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700974 src_iter = src->bi_iter;
975 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700976
977 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700978 if (!src_iter.bi_size) {
979 src = src->bi_next;
980 if (!src)
981 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700982
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700983 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700984 }
985
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700986 if (!dst_iter.bi_size) {
987 dst = dst->bi_next;
988 if (!dst)
989 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700990
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700991 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700992 }
993
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700994 src_bv = bio_iter_iovec(src, src_iter);
995 dst_bv = bio_iter_iovec(dst, dst_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700996
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700997 bytes = min(src_bv.bv_len, dst_bv.bv_len);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700998
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700999 src_p = kmap_atomic(src_bv.bv_page);
1000 dst_p = kmap_atomic(dst_bv.bv_page);
1001
1002 memcpy(dst_p + dst_bv.bv_offset,
1003 src_p + src_bv.bv_offset,
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001004 bytes);
1005
1006 kunmap_atomic(dst_p);
1007 kunmap_atomic(src_p);
1008
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001009 bio_advance_iter(src, &src_iter, bytes);
1010 bio_advance_iter(dst, &dst_iter, bytes);
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001011 }
1012}
1013EXPORT_SYMBOL(bio_copy_data);
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001016 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001017 struct iov_iter iter;
1018 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019};
1020
Fabian Frederick7410b3c2014-04-22 15:09:07 -06001021static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001022 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Jens Axboef3f63c12010-10-29 11:46:56 -06001024 if (iov_count > UIO_MAXIOV)
1025 return NULL;
1026
Kent Overstreetc8db4442013-11-22 19:39:06 -08001027 return kmalloc(sizeof(struct bio_map_data) +
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001028 sizeof(struct iovec) * iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
Dongsu Park9124d3f2015-01-18 16:16:34 +01001031/**
1032 * bio_copy_from_iter - copy all pages from iov_iter to bio
1033 * @bio: The &struct bio which describes the I/O as destination
1034 * @iter: iov_iter as source
1035 *
1036 * Copy all pages from iov_iter to bio.
1037 * Returns 0 on success, or error on failure.
1038 */
1039static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001040{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001041 int i;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001042 struct bio_vec *bvec;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001043
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001044 bio_for_each_segment_all(bvec, bio, i) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001045 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001046
Dongsu Park9124d3f2015-01-18 16:16:34 +01001047 ret = copy_page_from_iter(bvec->bv_page,
1048 bvec->bv_offset,
1049 bvec->bv_len,
1050 &iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001051
Dongsu Park9124d3f2015-01-18 16:16:34 +01001052 if (!iov_iter_count(&iter))
1053 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001054
Dongsu Park9124d3f2015-01-18 16:16:34 +01001055 if (ret < bvec->bv_len)
1056 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001057 }
1058
Dongsu Park9124d3f2015-01-18 16:16:34 +01001059 return 0;
1060}
1061
1062/**
1063 * bio_copy_to_iter - copy all pages from bio to iov_iter
1064 * @bio: The &struct bio which describes the I/O as source
1065 * @iter: iov_iter as destination
1066 *
1067 * Copy all pages from bio to iov_iter.
1068 * Returns 0 on success, or error on failure.
1069 */
1070static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1071{
1072 int i;
1073 struct bio_vec *bvec;
1074
1075 bio_for_each_segment_all(bvec, bio, i) {
1076 ssize_t ret;
1077
1078 ret = copy_page_to_iter(bvec->bv_page,
1079 bvec->bv_offset,
1080 bvec->bv_len,
1081 &iter);
1082
1083 if (!iov_iter_count(&iter))
1084 break;
1085
1086 if (ret < bvec->bv_len)
1087 return -EFAULT;
1088 }
1089
1090 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001091}
1092
Guoqing Jiang491221f2016-09-22 03:10:01 -04001093void bio_free_pages(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001094{
1095 struct bio_vec *bvec;
1096 int i;
1097
1098 bio_for_each_segment_all(bvec, bio, i)
1099 __free_page(bvec->bv_page);
1100}
Guoqing Jiang491221f2016-09-22 03:10:01 -04001101EXPORT_SYMBOL(bio_free_pages);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103/**
1104 * bio_uncopy_user - finish previously mapped bio
1105 * @bio: bio being terminated
1106 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001107 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 * to user space in case of a read.
1109 */
1110int bio_uncopy_user(struct bio *bio)
1111{
1112 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001113 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Roland Dreier35dc2482013-08-05 17:55:01 -07001115 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1116 /*
1117 * if we're in a workqueue, the request is orphaned, so
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001118 * don't copy into a random user address space, just free
1119 * and return -EINTR so user space doesn't expect any data.
Roland Dreier35dc2482013-08-05 17:55:01 -07001120 */
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001121 if (!current->mm)
1122 ret = -EINTR;
1123 else if (bio_data_dir(bio) == READ)
Dongsu Park9124d3f2015-01-18 16:16:34 +01001124 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001125 if (bmd->is_our_pages)
1126 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001127 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001128 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 bio_put(bio);
1130 return ret;
1131}
1132
1133/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001134 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001135 * @q: destination block queue
1136 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1137 * @iter: iovec iterator
1138 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 *
1140 * Prepares and returns a bio for indirect user io, bouncing data
1141 * to/from kernel pages as necessary. Must be paired with
1142 * call bio_uncopy_user() on io completion.
1143 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001144struct bio *bio_copy_user_iov(struct request_queue *q,
1145 struct rq_map_data *map_data,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001146 const struct iov_iter *iter,
1147 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 struct page *page;
1151 struct bio *bio;
1152 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001153 int nr_pages = 0;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001154 unsigned int len = iter->count;
Geliang Tangbd5cece2015-11-21 17:27:31 +08001155 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001157 for (i = 0; i < iter->nr_segs; i++) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001158 unsigned long uaddr;
1159 unsigned long end;
1160 unsigned long start;
1161
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001162 uaddr = (unsigned long) iter->iov[i].iov_base;
1163 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1164 >> PAGE_SHIFT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001165 start = uaddr >> PAGE_SHIFT;
1166
Jens Axboecb4644c2010-11-10 14:36:25 +01001167 /*
1168 * Overflow, abort
1169 */
1170 if (end < start)
1171 return ERR_PTR(-EINVAL);
1172
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001173 nr_pages += end - start;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001174 }
1175
FUJITA Tomonori69838722009-04-28 20:24:29 +02001176 if (offset)
1177 nr_pages++;
1178
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001179 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 if (!bmd)
1181 return ERR_PTR(-ENOMEM);
1182
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001183 /*
1184 * We need to do a deep copy of the iov_iter including the iovecs.
1185 * The caller provided iov might point to an on-stack or otherwise
1186 * shortlived one.
1187 */
1188 bmd->is_our_pages = map_data ? 0 : 1;
1189 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
Al Viroed35ded2017-09-24 10:21:15 -04001190 bmd->iter = *iter;
1191 bmd->iter.iov = bmd->iov;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001194 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 if (!bio)
1196 goto out_bmd;
1197
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001198 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001199 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001202
1203 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001204 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001205 i = map_data->offset / PAGE_SIZE;
1206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001208 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001210 bytes -= offset;
1211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (bytes > len)
1213 bytes = len;
1214
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001215 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001216 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001217 ret = -ENOMEM;
1218 break;
1219 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001220
1221 page = map_data->pages[i / nr_pages];
1222 page += (i % nr_pages);
1223
1224 i++;
1225 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001226 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001227 if (!page) {
1228 ret = -ENOMEM;
1229 break;
1230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 }
1232
Jérôme Glisse056066d2019-04-10 16:27:51 -04001233 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) {
1234 if (!map_data)
1235 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 break;
Jérôme Glisse056066d2019-04-10 16:27:51 -04001237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
1239 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001240 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 }
1242
1243 if (ret)
1244 goto cleanup;
1245
1246 /*
1247 * success
1248 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001249 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001250 (map_data && map_data->from_user)) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001251 ret = bio_copy_from_iter(bio, *iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001252 if (ret)
1253 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
1255
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001256 bio->bi_private = bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 return bio;
1258cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001259 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001260 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 bio_put(bio);
1262out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001263 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return ERR_PTR(ret);
1265}
1266
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001267/**
1268 * bio_map_user_iov - map user iovec into bio
1269 * @q: the struct request_queue for the bio
1270 * @iter: iovec iterator
1271 * @gfp_mask: memory allocation flags
1272 *
1273 * Map the user space address into a bio suitable for io to a block
1274 * device. Returns an error pointer in case of error.
1275 */
1276struct bio *bio_map_user_iov(struct request_queue *q,
1277 const struct iov_iter *iter,
1278 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001280 int j;
James Bottomley f1970ba2005-06-20 14:06:52 +02001281 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 struct page **pages;
1283 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001284 int cur_page = 0;
1285 int ret, offset;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001286 struct iov_iter i;
1287 struct iovec iov;
Al Viroe67dfe72017-09-23 15:51:23 -04001288 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001290 iov_for_each(iov, i, *iter) {
1291 unsigned long uaddr = (unsigned long) iov.iov_base;
1292 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001293 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1294 unsigned long start = uaddr >> PAGE_SHIFT;
1295
Jens Axboecb4644c2010-11-10 14:36:25 +01001296 /*
1297 * Overflow, abort
1298 */
1299 if (end < start)
1300 return ERR_PTR(-EINVAL);
1301
James Bottomley f1970ba2005-06-20 14:06:52 +02001302 nr_pages += end - start;
1303 /*
Linus Walleija441b0d2016-09-14 14:32:52 +02001304 * buffer must be aligned to at least logical block size for now
James Bottomley f1970ba2005-06-20 14:06:52 +02001305 */
Mike Christiead2d7222006-12-01 10:40:20 +01001306 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +02001307 return ERR_PTR(-EINVAL);
1308 }
1309
1310 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 return ERR_PTR(-EINVAL);
1312
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001313 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (!bio)
1315 return ERR_PTR(-ENOMEM);
1316
1317 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001318 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 if (!pages)
1320 goto out;
1321
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001322 iov_for_each(iov, i, *iter) {
1323 unsigned long uaddr = (unsigned long) iov.iov_base;
1324 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001325 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1326 unsigned long start = uaddr >> PAGE_SHIFT;
1327 const int local_nr_pages = end - start;
1328 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001329
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001330 ret = get_user_pages_fast(uaddr, local_nr_pages,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001331 (iter->type & WRITE) != WRITE,
1332 &pages[cur_page]);
Al Viroe67dfe72017-09-23 15:51:23 -04001333 if (unlikely(ret < local_nr_pages)) {
1334 for (j = cur_page; j < page_limit; j++) {
1335 if (!pages[j])
1336 break;
1337 put_page(pages[j]);
1338 }
Jens Axboe99172152006-06-16 13:02:29 +02001339 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001340 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Geliang Tangbd5cece2015-11-21 17:27:31 +08001343 offset = offset_in_page(uaddr);
James Bottomley f1970ba2005-06-20 14:06:52 +02001344 for (j = cur_page; j < page_limit; j++) {
1345 unsigned int bytes = PAGE_SIZE - offset;
Vitaly Mayatskikh5444d8a2017-09-22 01:18:39 -04001346 unsigned short prev_bi_vcnt = bio->bi_vcnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
James Bottomley f1970ba2005-06-20 14:06:52 +02001348 if (len <= 0)
1349 break;
1350
1351 if (bytes > len)
1352 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
James Bottomley f1970ba2005-06-20 14:06:52 +02001354 /*
1355 * sorry...
1356 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001357 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1358 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001359 break;
1360
Vitaly Mayatskikh5444d8a2017-09-22 01:18:39 -04001361 /*
1362 * check if vector was merged with previous
1363 * drop page reference if needed
1364 */
1365 if (bio->bi_vcnt == prev_bi_vcnt)
1366 put_page(pages[j]);
1367
James Bottomley f1970ba2005-06-20 14:06:52 +02001368 len -= bytes;
1369 offset = 0;
1370 }
1371
1372 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001374 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001376 while (j < page_limit)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001377 put_page(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 }
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 kfree(pages);
1381
1382 /*
1383 * set data direction, and check if mapped pages need bouncing
1384 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001385 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001386 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001388 bio_set_flag(bio, BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001389
1390 /*
1391 * subtle -- if __bio_map_user() ended up bouncing a bio,
1392 * it would normally disappear when its bi_end_io is run.
1393 * however, we need it for the unmap, so grab an extra
1394 * reference to it
1395 */
1396 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001398
1399 out_unmap:
Al Viroe67dfe72017-09-23 15:51:23 -04001400 bio_for_each_segment_all(bvec, bio, j) {
1401 put_page(bvec->bv_page);
James Bottomley f1970ba2005-06-20 14:06:52 +02001402 }
1403 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 kfree(pages);
1405 bio_put(bio);
1406 return ERR_PTR(ret);
1407}
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409static void __bio_unmap_user(struct bio *bio)
1410{
1411 struct bio_vec *bvec;
1412 int i;
1413
1414 /*
1415 * make sure we dirty pages we wrote to
1416 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001417 bio_for_each_segment_all(bvec, bio, i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 if (bio_data_dir(bio) == READ)
1419 set_page_dirty_lock(bvec->bv_page);
1420
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001421 put_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423
1424 bio_put(bio);
1425}
1426
1427/**
1428 * bio_unmap_user - unmap a bio
1429 * @bio: the bio being unmapped
1430 *
1431 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1432 * a process context.
1433 *
1434 * bio_unmap_user() may sleep.
1435 */
1436void bio_unmap_user(struct bio *bio)
1437{
1438 __bio_unmap_user(bio);
1439 bio_put(bio);
1440}
1441
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001442static void bio_map_kern_endio(struct bio *bio)
Jens Axboeb8238252005-06-20 14:05:27 +02001443{
Jens Axboeb8238252005-06-20 14:05:27 +02001444 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001445}
1446
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001447/**
1448 * bio_map_kern - map kernel address into bio
1449 * @q: the struct request_queue for the bio
1450 * @data: pointer to buffer to map
1451 * @len: length in bytes
1452 * @gfp_mask: allocation flags for bio allocation
1453 *
1454 * Map the kernel address into a bio suitable for io to a block
1455 * device. Returns an error pointer in case of error.
1456 */
1457struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1458 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001459{
1460 unsigned long kaddr = (unsigned long)data;
1461 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1462 unsigned long start = kaddr >> PAGE_SHIFT;
1463 const int nr_pages = end - start;
1464 int offset, i;
1465 struct bio *bio;
1466
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001467 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001468 if (!bio)
1469 return ERR_PTR(-ENOMEM);
1470
1471 offset = offset_in_page(kaddr);
1472 for (i = 0; i < nr_pages; i++) {
1473 unsigned int bytes = PAGE_SIZE - offset;
1474
1475 if (len <= 0)
1476 break;
1477
1478 if (bytes > len)
1479 bytes = len;
1480
Mike Christiedefd94b2005-12-05 02:37:06 -06001481 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001482 offset) < bytes) {
1483 /* we don't support partial mappings */
1484 bio_put(bio);
1485 return ERR_PTR(-EINVAL);
1486 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001487
1488 data += bytes;
1489 len -= bytes;
1490 offset = 0;
1491 }
1492
Jens Axboeb8238252005-06-20 14:05:27 +02001493 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001494 return bio;
1495}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001496EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001497
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001498static void bio_copy_kern_endio(struct bio *bio)
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001499{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001500 bio_free_pages(bio);
1501 bio_put(bio);
1502}
1503
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001504static void bio_copy_kern_endio_read(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001505{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001506 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001507 struct bio_vec *bvec;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001508 int i;
1509
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001510 bio_for_each_segment_all(bvec, bio, i) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001511 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001512 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001513 }
1514
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001515 bio_copy_kern_endio(bio);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001516}
1517
1518/**
1519 * bio_copy_kern - copy kernel address into bio
1520 * @q: the struct request_queue for the bio
1521 * @data: pointer to buffer to copy
1522 * @len: length in bytes
1523 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001524 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001525 *
1526 * copy the kernel address into a bio suitable for io to a block
1527 * device. Returns an error pointer in case of error.
1528 */
1529struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1530 gfp_t gfp_mask, int reading)
1531{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001532 unsigned long kaddr = (unsigned long)data;
1533 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1534 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001535 struct bio *bio;
1536 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001537 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001538
Christoph Hellwig42d26832015-01-18 16:16:28 +01001539 /*
1540 * Overflow, abort
1541 */
1542 if (end < start)
1543 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001544
Christoph Hellwig42d26832015-01-18 16:16:28 +01001545 nr_pages = end - start;
1546 bio = bio_kmalloc(gfp_mask, nr_pages);
1547 if (!bio)
1548 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001549
Christoph Hellwig42d26832015-01-18 16:16:28 +01001550 while (len) {
1551 struct page *page;
1552 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001553
Christoph Hellwig42d26832015-01-18 16:16:28 +01001554 if (bytes > len)
1555 bytes = len;
1556
1557 page = alloc_page(q->bounce_gfp | gfp_mask);
1558 if (!page)
1559 goto cleanup;
1560
1561 if (!reading)
1562 memcpy(page_address(page), p, bytes);
1563
1564 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1565 break;
1566
1567 len -= bytes;
1568 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001569 }
1570
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001571 if (reading) {
1572 bio->bi_end_io = bio_copy_kern_endio_read;
1573 bio->bi_private = data;
1574 } else {
1575 bio->bi_end_io = bio_copy_kern_endio;
Mike Christie95fe6c12016-06-05 14:31:48 -05001576 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001577 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001578
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001579 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001580
1581cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001582 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001583 bio_put(bio);
1584 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001585}
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587/*
1588 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1589 * for performing direct-IO in BIOs.
1590 *
1591 * The problem is that we cannot run set_page_dirty() from interrupt context
1592 * because the required locks are not interrupt-safe. So what we can do is to
1593 * mark the pages dirty _before_ performing IO. And in interrupt context,
1594 * check that the pages are still dirty. If so, fine. If not, redirty them
1595 * in process context.
1596 *
1597 * We special-case compound pages here: normally this means reads into hugetlb
1598 * pages. The logic in here doesn't really work right for compound pages
1599 * because the VM does not uniformly chase down the head page in all cases.
1600 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1601 * handle them at all. So we skip compound pages here at an early stage.
1602 *
1603 * Note that this code is very hard to test under normal circumstances because
1604 * direct-io pins the pages with get_user_pages(). This makes
1605 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001606 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 * pagecache.
1608 *
1609 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1610 * deferred bio dirtying paths.
1611 */
1612
1613/*
1614 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1615 */
1616void bio_set_pages_dirty(struct bio *bio)
1617{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001618 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 int i;
1620
Kent Overstreetcb34e052012-09-05 15:22:02 -07001621 bio_for_each_segment_all(bvec, bio, i) {
1622 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624 if (page && !PageCompound(page))
1625 set_page_dirty_lock(page);
1626 }
1627}
1628
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001629static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001631 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 int i;
1633
Kent Overstreetcb34e052012-09-05 15:22:02 -07001634 bio_for_each_segment_all(bvec, bio, i) {
1635 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 if (page)
1638 put_page(page);
1639 }
1640}
1641
1642/*
1643 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1644 * If they are, then fine. If, however, some pages are clean then they must
1645 * have been written out during the direct-IO read. So we take another ref on
1646 * the BIO and the offending pages and re-dirty the pages in process context.
1647 *
1648 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +03001649 * here on. It will run one put_page() against each page and will run one
1650 * bio_put() against the BIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 */
1652
David Howells65f27f32006-11-22 14:55:48 +00001653static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
David Howells65f27f32006-11-22 14:55:48 +00001655static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656static DEFINE_SPINLOCK(bio_dirty_lock);
1657static struct bio *bio_dirty_list;
1658
1659/*
1660 * This runs in process context
1661 */
David Howells65f27f32006-11-22 14:55:48 +00001662static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663{
1664 unsigned long flags;
1665 struct bio *bio;
1666
1667 spin_lock_irqsave(&bio_dirty_lock, flags);
1668 bio = bio_dirty_list;
1669 bio_dirty_list = NULL;
1670 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1671
1672 while (bio) {
1673 struct bio *next = bio->bi_private;
1674
1675 bio_set_pages_dirty(bio);
1676 bio_release_pages(bio);
1677 bio_put(bio);
1678 bio = next;
1679 }
1680}
1681
1682void bio_check_pages_dirty(struct bio *bio)
1683{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001684 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 int nr_clean_pages = 0;
1686 int i;
1687
Kent Overstreetcb34e052012-09-05 15:22:02 -07001688 bio_for_each_segment_all(bvec, bio, i) {
1689 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
1691 if (PageDirty(page) || PageCompound(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001692 put_page(page);
Kent Overstreetcb34e052012-09-05 15:22:02 -07001693 bvec->bv_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 } else {
1695 nr_clean_pages++;
1696 }
1697 }
1698
1699 if (nr_clean_pages) {
1700 unsigned long flags;
1701
1702 spin_lock_irqsave(&bio_dirty_lock, flags);
1703 bio->bi_private = bio_dirty_list;
1704 bio_dirty_list = bio;
1705 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1706 schedule_work(&bio_dirty_work);
1707 } else {
1708 bio_put(bio);
1709 }
1710}
1711
Gu Zheng394ffa52014-11-24 11:05:22 +08001712void generic_start_io_acct(int rw, unsigned long sectors,
1713 struct hd_struct *part)
1714{
1715 int cpu = part_stat_lock();
1716
1717 part_round_stats(cpu, part);
1718 part_stat_inc(cpu, part, ios[rw]);
1719 part_stat_add(cpu, part, sectors[rw], sectors);
1720 part_inc_in_flight(part, rw);
1721
1722 part_stat_unlock();
1723}
1724EXPORT_SYMBOL(generic_start_io_acct);
1725
1726void generic_end_io_acct(int rw, struct hd_struct *part,
1727 unsigned long start_time)
1728{
1729 unsigned long duration = jiffies - start_time;
1730 int cpu = part_stat_lock();
1731
1732 part_stat_add(cpu, part, ticks[rw], duration);
1733 part_round_stats(cpu, part);
1734 part_dec_in_flight(part, rw);
1735
1736 part_stat_unlock();
1737}
1738EXPORT_SYMBOL(generic_end_io_acct);
1739
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001740#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1741void bio_flush_dcache_pages(struct bio *bi)
1742{
Kent Overstreet79886132013-11-23 17:19:00 -08001743 struct bio_vec bvec;
1744 struct bvec_iter iter;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001745
Kent Overstreet79886132013-11-23 17:19:00 -08001746 bio_for_each_segment(bvec, bi, iter)
1747 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001748}
1749EXPORT_SYMBOL(bio_flush_dcache_pages);
1750#endif
1751
Jens Axboec4cf5262015-04-17 16:15:18 -06001752static inline bool bio_remaining_done(struct bio *bio)
1753{
1754 /*
1755 * If we're not chaining, then ->__bi_remaining is always 1 and
1756 * we always end io on the first invocation.
1757 */
1758 if (!bio_flagged(bio, BIO_CHAIN))
1759 return true;
1760
1761 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1762
Mike Snitzer326e1db2015-05-22 09:14:03 -04001763 if (atomic_dec_and_test(&bio->__bi_remaining)) {
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001764 bio_clear_flag(bio, BIO_CHAIN);
Jens Axboec4cf5262015-04-17 16:15:18 -06001765 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001766 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001767
1768 return false;
1769}
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771/**
1772 * bio_endio - end I/O on a bio
1773 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 *
1775 * Description:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001776 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1777 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1778 * bio unless they own it and thus know that it has an end_io function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 **/
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001780void bio_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001782again:
Christoph Hellwig2b885512016-03-11 17:34:53 +01001783 if (!bio_remaining_done(bio))
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001784 return;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001785
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001786 /*
1787 * Need to have a real endio function for chained bios, otherwise
1788 * various corner cases will break (like stacking block devices that
1789 * save/restore bi_end_io) - however, we want to avoid unbounded
1790 * recursion and blowing the stack. Tail call optimization would
1791 * handle this, but compiling with frame pointers also disables
1792 * gcc's sibling call optimization.
1793 */
1794 if (bio->bi_end_io == bio_chain_endio) {
1795 bio = __bio_chain_endio(bio);
1796 goto again;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001797 }
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001798
1799 if (bio->bi_end_io)
1800 bio->bi_end_io(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001802EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Kent Overstreet196d38b2013-11-23 18:34:15 -08001804/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001805 * bio_split - split a bio
1806 * @bio: bio to split
1807 * @sectors: number of sectors to split from the front of @bio
1808 * @gfp: gfp mask
1809 * @bs: bio set to allocate from
1810 *
1811 * Allocates and returns a new bio which represents @sectors from the start of
1812 * @bio, and updates @bio to represent the remaining sectors.
1813 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001814 * Unless this is a discard request the newly allocated bio will point
1815 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1816 * @bio is not freed before the split.
Kent Overstreet20d01892013-11-23 18:21:01 -08001817 */
1818struct bio *bio_split(struct bio *bio, int sectors,
1819 gfp_t gfp, struct bio_set *bs)
1820{
1821 struct bio *split = NULL;
1822
1823 BUG_ON(sectors <= 0);
1824 BUG_ON(sectors >= bio_sectors(bio));
1825
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001826 /*
1827 * Discards need a mutable bio_vec to accommodate the payload
1828 * required by the DSM TRIM and UNMAP commands.
1829 */
Adrian Hunter7afafc82016-08-16 10:59:35 +03001830 if (bio_op(bio) == REQ_OP_DISCARD || bio_op(bio) == REQ_OP_SECURE_ERASE)
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001831 split = bio_clone_bioset(bio, gfp, bs);
1832 else
1833 split = bio_clone_fast(bio, gfp, bs);
1834
Kent Overstreet20d01892013-11-23 18:21:01 -08001835 if (!split)
1836 return NULL;
1837
1838 split->bi_iter.bi_size = sectors << 9;
1839
1840 if (bio_integrity(split))
1841 bio_integrity_trim(split, 0, sectors);
1842
1843 bio_advance(bio, split->bi_iter.bi_size);
1844
1845 return split;
1846}
1847EXPORT_SYMBOL(bio_split);
1848
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001849/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001850 * bio_trim - trim a bio
1851 * @bio: bio to trim
1852 * @offset: number of sectors to trim from the front of @bio
1853 * @size: size we want to trim @bio to, in sectors
1854 */
1855void bio_trim(struct bio *bio, int offset, int size)
1856{
1857 /* 'bio' is a cloned bio which we need to trim to match
1858 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001859 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001860
1861 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001862 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001863 return;
1864
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001865 bio_clear_flag(bio, BIO_SEG_VALID);
Kent Overstreet6678d832013-08-07 11:14:32 -07001866
1867 bio_advance(bio, offset << 9);
1868
Kent Overstreet4f024f32013-10-11 15:44:27 -07001869 bio->bi_iter.bi_size = size;
Kent Overstreet6678d832013-08-07 11:14:32 -07001870}
1871EXPORT_SYMBOL_GPL(bio_trim);
1872
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873/*
1874 * create memory pools for biovec's in a bio_set.
1875 * use the global biovec slabs created for general use.
1876 */
Fabian Fredericka6c39cb4f2014-04-22 15:09:05 -06001877mempool_t *biovec_create_pool(int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
Christoph Hellwiged996a52016-07-19 11:28:42 +02001879 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Kent Overstreet9f060e22012-10-12 15:29:33 -07001881 return mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882}
1883
1884void bioset_free(struct bio_set *bs)
1885{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001886 if (bs->rescue_workqueue)
1887 destroy_workqueue(bs->rescue_workqueue);
1888
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 if (bs->bio_pool)
1890 mempool_destroy(bs->bio_pool);
1891
Kent Overstreet9f060e22012-10-12 15:29:33 -07001892 if (bs->bvec_pool)
1893 mempool_destroy(bs->bvec_pool);
1894
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001895 bioset_integrity_free(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001896 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
1898 kfree(bs);
1899}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001900EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001902static struct bio_set *__bioset_create(unsigned int pool_size,
1903 unsigned int front_pad,
1904 bool create_bvec_pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
Jens Axboe392ddc32008-12-23 12:42:54 +01001906 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001907 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Jens Axboe1b434492008-10-22 20:32:58 +02001909 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 if (!bs)
1911 return NULL;
1912
Jens Axboebb799ca2008-12-10 15:35:05 +01001913 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001914
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001915 spin_lock_init(&bs->rescue_lock);
1916 bio_list_init(&bs->rescue_list);
1917 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1918
Jens Axboe392ddc32008-12-23 12:42:54 +01001919 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001920 if (!bs->bio_slab) {
1921 kfree(bs);
1922 return NULL;
1923 }
1924
1925 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (!bs->bio_pool)
1927 goto bad;
1928
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001929 if (create_bvec_pool) {
1930 bs->bvec_pool = biovec_create_pool(pool_size);
1931 if (!bs->bvec_pool)
1932 goto bad;
1933 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001935 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1936 if (!bs->rescue_workqueue)
1937 goto bad;
1938
1939 return bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940bad:
1941 bioset_free(bs);
1942 return NULL;
1943}
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001944
1945/**
1946 * bioset_create - Create a bio_set
1947 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1948 * @front_pad: Number of bytes to allocate in front of the returned bio
1949 *
1950 * Description:
1951 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1952 * to ask for a number of bytes to be allocated in front of the bio.
1953 * Front pad allocation is useful for embedding the bio inside
1954 * another structure, to avoid allocating extra data to go with the bio.
1955 * Note that the bio must be embedded at the END of that structure always,
1956 * or things will break badly.
1957 */
1958struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1959{
1960 return __bioset_create(pool_size, front_pad, true);
1961}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001962EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001964/**
1965 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1966 * @pool_size: Number of bio to cache in the mempool
1967 * @front_pad: Number of bytes to allocate in front of the returned bio
1968 *
1969 * Description:
1970 * Same functionality as bioset_create() except that mempool is not
1971 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1972 */
1973struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1974{
1975 return __bioset_create(pool_size, front_pad, false);
1976}
1977EXPORT_SYMBOL(bioset_create_nobvec);
1978
Tejun Heo852c7882012-03-05 13:15:27 -08001979#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04001980
1981/**
1982 * bio_associate_blkcg - associate a bio with the specified blkcg
1983 * @bio: target bio
1984 * @blkcg_css: css of the blkcg to associate
1985 *
1986 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1987 * treat @bio as if it were issued by a task which belongs to the blkcg.
1988 *
1989 * This function takes an extra reference of @blkcg_css which will be put
1990 * when @bio is released. The caller must own @bio and is responsible for
1991 * synchronizing calls to this function.
1992 */
1993int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1994{
1995 if (unlikely(bio->bi_css))
1996 return -EBUSY;
1997 css_get(blkcg_css);
1998 bio->bi_css = blkcg_css;
1999 return 0;
2000}
Tejun Heo5aa2a962015-07-23 14:27:09 -04002001EXPORT_SYMBOL_GPL(bio_associate_blkcg);
Tejun Heo1d933cf2015-05-22 17:13:24 -04002002
Tejun Heo852c7882012-03-05 13:15:27 -08002003/**
2004 * bio_associate_current - associate a bio with %current
2005 * @bio: target bio
2006 *
2007 * Associate @bio with %current if it hasn't been associated yet. Block
2008 * layer will treat @bio as if it were issued by %current no matter which
2009 * task actually issues it.
2010 *
2011 * This function takes an extra reference of @task's io_context and blkcg
2012 * which will be put when @bio is released. The caller must own @bio,
2013 * ensure %current->io_context exists, and is responsible for synchronizing
2014 * calls to this function.
2015 */
2016int bio_associate_current(struct bio *bio)
2017{
2018 struct io_context *ioc;
Tejun Heo852c7882012-03-05 13:15:27 -08002019
Tejun Heo1d933cf2015-05-22 17:13:24 -04002020 if (bio->bi_css)
Tejun Heo852c7882012-03-05 13:15:27 -08002021 return -EBUSY;
2022
2023 ioc = current->io_context;
2024 if (!ioc)
2025 return -ENOENT;
2026
Tejun Heo852c7882012-03-05 13:15:27 -08002027 get_io_context_active(ioc);
2028 bio->bi_ioc = ioc;
Tejun Heoc165b3e2015-08-18 14:55:29 -07002029 bio->bi_css = task_get_css(current, io_cgrp_id);
Tejun Heo852c7882012-03-05 13:15:27 -08002030 return 0;
2031}
Tejun Heo5aa2a962015-07-23 14:27:09 -04002032EXPORT_SYMBOL_GPL(bio_associate_current);
Tejun Heo852c7882012-03-05 13:15:27 -08002033
2034/**
2035 * bio_disassociate_task - undo bio_associate_current()
2036 * @bio: target bio
2037 */
2038void bio_disassociate_task(struct bio *bio)
2039{
2040 if (bio->bi_ioc) {
2041 put_io_context(bio->bi_ioc);
2042 bio->bi_ioc = NULL;
2043 }
2044 if (bio->bi_css) {
2045 css_put(bio->bi_css);
2046 bio->bi_css = NULL;
2047 }
2048}
2049
Paolo Valente20bd7232016-07-27 07:22:05 +02002050/**
2051 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2052 * @dst: destination bio
2053 * @src: source bio
2054 */
2055void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2056{
2057 if (src->bi_css)
2058 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2059}
2060
Tejun Heo852c7882012-03-05 13:15:27 -08002061#endif /* CONFIG_BLK_CGROUP */
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063static void __init biovec_init_slabs(void)
2064{
2065 int i;
2066
Christoph Hellwiged996a52016-07-19 11:28:42 +02002067 for (i = 0; i < BVEC_POOL_NR; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 int size;
2069 struct biovec_slab *bvs = bvec_slabs + i;
2070
Jens Axboea7fcd372008-12-05 16:10:29 +01002071 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2072 bvs->slab = NULL;
2073 continue;
2074 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002075
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 size = bvs->nr_vecs * sizeof(struct bio_vec);
2077 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002078 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 }
2080}
2081
2082static int __init init_bio(void)
2083{
Jens Axboebb799ca2008-12-10 15:35:05 +01002084 bio_slab_max = 2;
2085 bio_slab_nr = 0;
2086 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2087 if (!bio_slabs)
2088 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002090 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 biovec_init_slabs();
2092
Jens Axboebb799ca2008-12-10 15:35:05 +01002093 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 if (!fs_bio_set)
2095 panic("bio: can't allocate bios\n");
2096
Martin K. Petersena91a2782011-03-17 11:11:05 +01002097 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2098 panic("bio: can't create integrity pool\n");
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 return 0;
2101}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102subsys_initcall(init_bio);