blob: 848cd351513bad7fb44b176c0205a614fa314a52 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
16 *
17 */
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070022#include <linux/uio.h>
Tejun Heo852c7882012-03-05 13:15:27 -080023#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050027#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/mempool.h>
29#include <linux/workqueue.h>
Tejun Heo852c7882012-03-05 13:15:27 -080030#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Li Zefan55782132009-06-09 13:43:05 +080032#include <trace/events/block.h>
Ingo Molnar0bfc2452008-11-26 11:59:56 +010033
Jens Axboe392ddc32008-12-23 12:42:54 +010034/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
43 * unsigned short
44 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Martin K. Petersendf677142011-03-08 08:28:01 +010046static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48};
49#undef BV
50
51/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
54 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020055struct bio_set *fs_bio_set;
Kent Overstreet3f86a822012-09-06 15:35:01 -070056EXPORT_SYMBOL(fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboebb799ca2008-12-10 15:35:05 +010058/*
59 * Our slab pool management
60 */
61struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66};
67static DEFINE_MUTEX(bio_slab_lock);
68static struct bio_slab *bio_slabs;
69static unsigned int bio_slab_nr, bio_slab_max;
70
71static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72{
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020075 struct bio_slab *bslab, *new_bio_slabs;
Anna Leuschner386bc352012-10-22 21:53:36 +020076 unsigned int new_bio_slab_max;
Jens Axboebb799ca2008-12-10 15:35:05 +010077 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010083 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010084
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
Anna Leuschner386bc352012-10-22 21:53:36 +020099 new_bio_slab_max = bio_slab_max << 1;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200100 new_bio_slabs = krealloc(bio_slabs,
Anna Leuschner386bc352012-10-22 21:53:36 +0200101 new_bio_slab_max * sizeof(struct bio_slab),
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200102 GFP_KERNEL);
103 if (!new_bio_slabs)
Jens Axboebb799ca2008-12-10 15:35:05 +0100104 goto out_unlock;
Anna Leuschner386bc352012-10-22 21:53:36 +0200105 bio_slab_max = new_bio_slab_max;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200106 bio_slabs = new_bio_slabs;
Jens Axboebb799ca2008-12-10 15:35:05 +0100107 }
108 if (entry == -1)
109 entry = bio_slab_nr++;
110
111 bslab = &bio_slabs[entry];
112
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
Mikulas Patocka6a241482014-03-28 15:51:55 -0400114 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 SLAB_HWCACHE_ALIGN, NULL);
Jens Axboebb799ca2008-12-10 15:35:05 +0100116 if (!slab)
117 goto out_unlock;
118
Jens Axboebb799ca2008-12-10 15:35:05 +0100119 bslab->slab = slab;
120 bslab->slab_ref = 1;
121 bslab->slab_size = sz;
122out_unlock:
123 mutex_unlock(&bio_slab_lock);
124 return slab;
125}
126
127static void bio_put_slab(struct bio_set *bs)
128{
129 struct bio_slab *bslab = NULL;
130 unsigned int i;
131
132 mutex_lock(&bio_slab_lock);
133
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
137 break;
138 }
139 }
140
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 goto out;
143
144 WARN_ON(!bslab->slab_ref);
145
146 if (--bslab->slab_ref)
147 goto out;
148
149 kmem_cache_destroy(bslab->slab);
150 bslab->slab = NULL;
151
152out:
153 mutex_unlock(&bio_slab_lock);
154}
155
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200156unsigned int bvec_nr_vecs(unsigned short idx)
157{
158 return bvec_slabs[idx].nr_vecs;
159}
160
Kent Overstreet9f060e22012-10-12 15:29:33 -0700161void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
Jens Axboebb799ca2008-12-10 15:35:05 +0100162{
163 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
164
165 if (idx == BIOVEC_MAX_IDX)
Kent Overstreet9f060e22012-10-12 15:29:33 -0700166 mempool_free(bv, pool);
Jens Axboebb799ca2008-12-10 15:35:05 +0100167 else {
168 struct biovec_slab *bvs = bvec_slabs + idx;
169
170 kmem_cache_free(bvs->slab, bv);
171 }
172}
173
Kent Overstreet9f060e22012-10-12 15:29:33 -0700174struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
175 mempool_t *pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
177 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100180 * see comment near bvec_array define!
181 */
182 switch (nr) {
183 case 1:
184 *idx = 0;
185 break;
186 case 2 ... 4:
187 *idx = 1;
188 break;
189 case 5 ... 16:
190 *idx = 2;
191 break;
192 case 17 ... 64:
193 *idx = 3;
194 break;
195 case 65 ... 128:
196 *idx = 4;
197 break;
198 case 129 ... BIO_MAX_PAGES:
199 *idx = 5;
200 break;
201 default:
202 return NULL;
203 }
204
205 /*
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
208 */
209 if (*idx == BIOVEC_MAX_IDX) {
210fallback:
Kent Overstreet9f060e22012-10-12 15:29:33 -0700211 bvl = mempool_alloc(pool, gfp_mask);
Jens Axboe7ff93452008-12-11 11:53:43 +0100212 } else {
213 struct biovec_slab *bvs = bvec_slabs + *idx;
Mel Gormand0164ad2015-11-06 16:28:21 -0800214 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200216 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200220 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100221 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
222
223 /*
Mel Gormand0164ad2015-11-06 16:28:21 -0800224 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
Jens Axboe7ff93452008-12-11 11:53:43 +0100225 * is set, retry with the 1-entry mempool
226 */
227 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
Mel Gormand0164ad2015-11-06 16:28:21 -0800228 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
Jens Axboe7ff93452008-12-11 11:53:43 +0100229 *idx = BIOVEC_MAX_IDX;
230 goto fallback;
231 }
232 }
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return bvl;
235}
236
Kent Overstreet4254bba2012-09-06 15:35:00 -0700237static void __bio_free(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Kent Overstreet4254bba2012-09-06 15:35:00 -0700239 bio_disassociate_task(bio);
Jens Axboe992c5dd2007-07-18 13:18:08 +0200240
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200241 if (bio_integrity(bio))
Kent Overstreet1e2a410f2012-09-06 15:34:56 -0700242 bio_integrity_free(bio);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700243}
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200244
Kent Overstreet4254bba2012-09-06 15:35:00 -0700245static void bio_free(struct bio *bio)
246{
247 struct bio_set *bs = bio->bi_pool;
248 void *p;
249
250 __bio_free(bio);
251
252 if (bs) {
Kent Overstreeta38352e2012-05-25 13:03:11 -0700253 if (bio_flagged(bio, BIO_OWNS_VEC))
Kent Overstreet9f060e22012-10-12 15:29:33 -0700254 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
Kent Overstreet4254bba2012-09-06 15:35:00 -0700255
256 /*
257 * If we have front padding, adjust the bio pointer before freeing
258 */
259 p = bio;
Jens Axboebb799ca2008-12-10 15:35:05 +0100260 p -= bs->front_pad;
261
Kent Overstreet4254bba2012-09-06 15:35:00 -0700262 mempool_free(p, bs->bio_pool);
263 } else {
264 /* Bio was allocated by bio_kmalloc() */
265 kfree(bio);
266 }
Peter Osterlund36763472005-09-06 15:16:42 -0700267}
268
Arjan van de Ven858119e2006-01-14 13:20:43 -0800269void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Jens Axboe2b94de52007-07-18 13:14:03 +0200271 memset(bio, 0, sizeof(*bio));
Jens Axboec4cf5262015-04-17 16:15:18 -0600272 atomic_set(&bio->__bi_remaining, 1);
Jens Axboedac56212015-04-17 16:23:59 -0600273 atomic_set(&bio->__bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200275EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277/**
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700278 * bio_reset - reinitialize a bio
279 * @bio: bio to reset
280 *
281 * Description:
282 * After calling bio_reset(), @bio will be in the same state as a freshly
283 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
284 * preserved are the ones that are initialized by bio_alloc_bioset(). See
285 * comment in struct bio.
286 */
287void bio_reset(struct bio *bio)
288{
289 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
290
Kent Overstreet4254bba2012-09-06 15:35:00 -0700291 __bio_free(bio);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700292
293 memset(bio, 0, BIO_RESET_BYTES);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200294 bio->bi_flags = flags;
Jens Axboec4cf5262015-04-17 16:15:18 -0600295 atomic_set(&bio->__bi_remaining, 1);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700296}
297EXPORT_SYMBOL(bio_reset);
298
Christoph Hellwig38f8baa2016-03-11 17:34:51 +0100299static struct bio *__bio_chain_endio(struct bio *bio)
Kent Overstreet196d38b2013-11-23 18:34:15 -0800300{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200301 struct bio *parent = bio->bi_private;
302
Christoph Hellwigaf3e3a52016-03-11 17:34:50 +0100303 if (!parent->bi_error)
304 parent->bi_error = bio->bi_error;
Kent Overstreet196d38b2013-11-23 18:34:15 -0800305 bio_put(bio);
Christoph Hellwig38f8baa2016-03-11 17:34:51 +0100306 return parent;
307}
308
309static void bio_chain_endio(struct bio *bio)
310{
311 bio_endio(__bio_chain_endio(bio));
Kent Overstreet196d38b2013-11-23 18:34:15 -0800312}
313
314/**
315 * bio_chain - chain bio completions
Randy Dunlap1051a902014-04-20 16:03:31 -0700316 * @bio: the target bio
317 * @parent: the @bio's parent bio
Kent Overstreet196d38b2013-11-23 18:34:15 -0800318 *
319 * The caller won't have a bi_end_io called when @bio completes - instead,
320 * @parent's bi_end_io won't be called until both @parent and @bio have
321 * completed; the chained bio will also be freed when it completes.
322 *
323 * The caller must not set bi_private or bi_end_io in @bio.
324 */
325void bio_chain(struct bio *bio, struct bio *parent)
326{
327 BUG_ON(bio->bi_private || bio->bi_end_io);
328
329 bio->bi_private = parent;
330 bio->bi_end_io = bio_chain_endio;
Jens Axboec4cf5262015-04-17 16:15:18 -0600331 bio_inc_remaining(parent);
Kent Overstreet196d38b2013-11-23 18:34:15 -0800332}
333EXPORT_SYMBOL(bio_chain);
334
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700335static void bio_alloc_rescue(struct work_struct *work)
336{
337 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
338 struct bio *bio;
339
340 while (1) {
341 spin_lock(&bs->rescue_lock);
342 bio = bio_list_pop(&bs->rescue_list);
343 spin_unlock(&bs->rescue_lock);
344
345 if (!bio)
346 break;
347
348 generic_make_request(bio);
349 }
350}
351
352static void punt_bios_to_rescuer(struct bio_set *bs)
353{
354 struct bio_list punt, nopunt;
355 struct bio *bio;
356
357 /*
358 * In order to guarantee forward progress we must punt only bios that
359 * were allocated from this bio_set; otherwise, if there was a bio on
360 * there for a stacking driver higher up in the stack, processing it
361 * could require allocating bios from this bio_set, and doing that from
362 * our own rescuer would be bad.
363 *
364 * Since bio lists are singly linked, pop them all instead of trying to
365 * remove from the middle of the list:
366 */
367
368 bio_list_init(&punt);
369 bio_list_init(&nopunt);
370
371 while ((bio = bio_list_pop(current->bio_list)))
372 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
373
374 *current->bio_list = nopunt;
375
376 spin_lock(&bs->rescue_lock);
377 bio_list_merge(&bs->rescue_list, &punt);
378 spin_unlock(&bs->rescue_lock);
379
380 queue_work(bs->rescue_workqueue, &bs->rescue_work);
381}
382
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700383/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 * bio_alloc_bioset - allocate a bio for I/O
385 * @gfp_mask: the GFP_ mask given to the slab allocator
386 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200387 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 *
389 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700390 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
391 * backed by the @bs's mempool.
392 *
Mel Gormand0164ad2015-11-06 16:28:21 -0800393 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
394 * always be able to allocate a bio. This is due to the mempool guarantees.
395 * To make this work, callers must never allocate more than 1 bio at a time
396 * from this pool. Callers that need to allocate more than 1 bio must always
397 * submit the previously allocated bio for IO before attempting to allocate
398 * a new one. Failure to do so can cause deadlocks under memory pressure.
Kent Overstreet3f86a822012-09-06 15:35:01 -0700399 *
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700400 * Note that when running under generic_make_request() (i.e. any block
401 * driver), bios are not submitted until after you return - see the code in
402 * generic_make_request() that converts recursion into iteration, to prevent
403 * stack overflows.
404 *
405 * This would normally mean allocating multiple bios under
406 * generic_make_request() would be susceptible to deadlocks, but we have
407 * deadlock avoidance code that resubmits any blocked bios from a rescuer
408 * thread.
409 *
410 * However, we do not guarantee forward progress for allocations from other
411 * mempools. Doing multiple allocations from the same mempool under
412 * generic_make_request() should be avoided - instead, use bio_set's front_pad
413 * for per bio allocations.
414 *
Kent Overstreet3f86a822012-09-06 15:35:01 -0700415 * RETURNS:
416 * Pointer to new bio on success, NULL on failure.
417 */
Al Virodd0fc662005-10-07 07:46:04 +0100418struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700420 gfp_t saved_gfp = gfp_mask;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700421 unsigned front_pad;
422 unsigned inline_vecs;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200423 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100424 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200425 struct bio *bio;
426 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200427
Kent Overstreet3f86a822012-09-06 15:35:01 -0700428 if (!bs) {
429 if (nr_iovecs > UIO_MAXIOV)
430 return NULL;
431
432 p = kmalloc(sizeof(struct bio) +
433 nr_iovecs * sizeof(struct bio_vec),
434 gfp_mask);
435 front_pad = 0;
436 inline_vecs = nr_iovecs;
437 } else {
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400438 /* should not use nobvec bioset for nr_iovecs > 0 */
439 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
440 return NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700441 /*
442 * generic_make_request() converts recursion to iteration; this
443 * means if we're running beneath it, any bios we allocate and
444 * submit will not be submitted (and thus freed) until after we
445 * return.
446 *
447 * This exposes us to a potential deadlock if we allocate
448 * multiple bios from the same bio_set() while running
449 * underneath generic_make_request(). If we were to allocate
450 * multiple bios (say a stacking block driver that was splitting
451 * bios), we would deadlock if we exhausted the mempool's
452 * reserve.
453 *
454 * We solve this, and guarantee forward progress, with a rescuer
455 * workqueue per bio_set. If we go to allocate and there are
456 * bios on current->bio_list, we first try the allocation
Mel Gormand0164ad2015-11-06 16:28:21 -0800457 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
458 * bios we would be blocking to the rescuer workqueue before
459 * we retry with the original gfp_flags.
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700460 */
461
462 if (current->bio_list && !bio_list_empty(current->bio_list))
Mel Gormand0164ad2015-11-06 16:28:21 -0800463 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700464
Kent Overstreet3f86a822012-09-06 15:35:01 -0700465 p = mempool_alloc(bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700466 if (!p && gfp_mask != saved_gfp) {
467 punt_bios_to_rescuer(bs);
468 gfp_mask = saved_gfp;
469 p = mempool_alloc(bs->bio_pool, gfp_mask);
470 }
471
Kent Overstreet3f86a822012-09-06 15:35:01 -0700472 front_pad = bs->front_pad;
473 inline_vecs = BIO_INLINE_VECS;
474 }
475
Tejun Heo451a9eb2009-04-15 19:50:51 +0200476 if (unlikely(!p))
477 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100478
Kent Overstreet3f86a822012-09-06 15:35:01 -0700479 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100480 bio_init(bio);
481
Kent Overstreet3f86a822012-09-06 15:35:01 -0700482 if (nr_iovecs > inline_vecs) {
Kent Overstreet9f060e22012-10-12 15:29:33 -0700483 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700484 if (!bvl && gfp_mask != saved_gfp) {
485 punt_bios_to_rescuer(bs);
486 gfp_mask = saved_gfp;
Kent Overstreet9f060e22012-10-12 15:29:33 -0700487 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700488 }
489
Ingo Molnar34053972009-02-21 11:16:36 +0100490 if (unlikely(!bvl))
491 goto err_free;
Kent Overstreeta38352e2012-05-25 13:03:11 -0700492
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600493 bio_set_flag(bio, BIO_OWNS_VEC);
Kent Overstreet3f86a822012-09-06 15:35:01 -0700494 } else if (nr_iovecs) {
495 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100496 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700497
498 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100499 bio->bi_flags |= idx << BIO_POOL_OFFSET;
500 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100501 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100503
504err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200505 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100506 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200508EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510void zero_fill_bio(struct bio *bio)
511{
512 unsigned long flags;
Kent Overstreet79886132013-11-23 17:19:00 -0800513 struct bio_vec bv;
514 struct bvec_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Kent Overstreet79886132013-11-23 17:19:00 -0800516 bio_for_each_segment(bv, bio, iter) {
517 char *data = bvec_kmap_irq(&bv, &flags);
518 memset(data, 0, bv.bv_len);
519 flush_dcache_page(bv.bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 bvec_kunmap_irq(data, &flags);
521 }
522}
523EXPORT_SYMBOL(zero_fill_bio);
524
525/**
526 * bio_put - release a reference to a bio
527 * @bio: bio to release reference to
528 *
529 * Description:
530 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100531 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 **/
533void bio_put(struct bio *bio)
534{
Jens Axboedac56212015-04-17 16:23:59 -0600535 if (!bio_flagged(bio, BIO_REFFED))
Kent Overstreet4254bba2012-09-06 15:35:00 -0700536 bio_free(bio);
Jens Axboedac56212015-04-17 16:23:59 -0600537 else {
538 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
539
540 /*
541 * last put frees it
542 */
543 if (atomic_dec_and_test(&bio->__bi_cnt))
544 bio_free(bio);
545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200547EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Jens Axboe165125e2007-07-24 09:28:11 +0200549inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
551 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
552 blk_recount_segments(q, bio);
553
554 return bio->bi_phys_segments;
555}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200556EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800559 * __bio_clone_fast - clone a bio that shares the original bio's biovec
560 * @bio: destination bio
561 * @bio_src: bio to clone
562 *
563 * Clone a &bio. Caller will own the returned bio, but not
564 * the actual data it points to. Reference count of returned
565 * bio will be one.
566 *
567 * Caller must ensure that @bio_src is not freed before @bio.
568 */
569void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
570{
571 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
572
573 /*
574 * most users will be overriding ->bi_bdev with a new target,
575 * so we don't set nor calculate new physical/hw segment counts here
576 */
577 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600578 bio_set_flag(bio, BIO_CLONED);
Kent Overstreet59d276f2013-11-23 18:19:27 -0800579 bio->bi_rw = bio_src->bi_rw;
580 bio->bi_iter = bio_src->bi_iter;
581 bio->bi_io_vec = bio_src->bi_io_vec;
582}
583EXPORT_SYMBOL(__bio_clone_fast);
584
585/**
586 * bio_clone_fast - clone a bio that shares the original bio's biovec
587 * @bio: bio to clone
588 * @gfp_mask: allocation priority
589 * @bs: bio_set to allocate from
590 *
591 * Like __bio_clone_fast, only also allocates the returned bio
592 */
593struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
594{
595 struct bio *b;
596
597 b = bio_alloc_bioset(gfp_mask, 0, bs);
598 if (!b)
599 return NULL;
600
601 __bio_clone_fast(b, bio);
602
603 if (bio_integrity(bio)) {
604 int ret;
605
606 ret = bio_integrity_clone(b, bio, gfp_mask);
607
608 if (ret < 0) {
609 bio_put(b);
610 return NULL;
611 }
612 }
613
614 return b;
615}
616EXPORT_SYMBOL(bio_clone_fast);
617
618/**
Kent Overstreetbdb53202013-11-23 17:26:46 -0800619 * bio_clone_bioset - clone a bio
620 * @bio_src: bio to clone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700622 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 *
Kent Overstreetbdb53202013-11-23 17:26:46 -0800624 * Clone bio. Caller will own the returned bio, but not the actual data it
625 * points to. Reference count of returned bio will be one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 */
Kent Overstreetbdb53202013-11-23 17:26:46 -0800627struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700628 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Kent Overstreetbdb53202013-11-23 17:26:46 -0800630 struct bvec_iter iter;
631 struct bio_vec bv;
632 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Kent Overstreetbdb53202013-11-23 17:26:46 -0800634 /*
635 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
636 * bio_src->bi_io_vec to bio->bi_io_vec.
637 *
638 * We can't do that anymore, because:
639 *
640 * - The point of cloning the biovec is to produce a bio with a biovec
641 * the caller can modify: bi_idx and bi_bvec_done should be 0.
642 *
643 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
644 * we tried to clone the whole thing bio_alloc_bioset() would fail.
645 * But the clone should succeed as long as the number of biovecs we
646 * actually need to allocate is fewer than BIO_MAX_PAGES.
647 *
648 * - Lastly, bi_vcnt should not be looked at or relied upon by code
649 * that does not own the bio - reason being drivers don't use it for
650 * iterating over the biovec anymore, so expecting it to be kept up
651 * to date (i.e. for clones that share the parent biovec) is just
652 * asking for trouble and would force extra work on
653 * __bio_clone_fast() anyways.
654 */
655
Kent Overstreet8423ae32014-02-10 17:45:50 -0800656 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
Kent Overstreetbdb53202013-11-23 17:26:46 -0800657 if (!bio)
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200658 return NULL;
Kent Overstreetbdb53202013-11-23 17:26:46 -0800659 bio->bi_bdev = bio_src->bi_bdev;
660 bio->bi_rw = bio_src->bi_rw;
661 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
662 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200663
Mike Christie95fe6c12016-06-05 14:31:48 -0500664 if (bio_op(bio) == REQ_OP_DISCARD)
Kent Overstreet8423ae32014-02-10 17:45:50 -0800665 goto integrity_clone;
666
Mike Christie95fe6c12016-06-05 14:31:48 -0500667 if (bio_op(bio) == REQ_OP_WRITE_SAME) {
Kent Overstreet8423ae32014-02-10 17:45:50 -0800668 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
669 goto integrity_clone;
670 }
671
Kent Overstreetbdb53202013-11-23 17:26:46 -0800672 bio_for_each_segment(bv, bio_src, iter)
673 bio->bi_io_vec[bio->bi_vcnt++] = bv;
674
Kent Overstreet8423ae32014-02-10 17:45:50 -0800675integrity_clone:
Kent Overstreetbdb53202013-11-23 17:26:46 -0800676 if (bio_integrity(bio_src)) {
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200677 int ret;
678
Kent Overstreetbdb53202013-11-23 17:26:46 -0800679 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
Li Zefan059ea332009-03-09 10:42:45 +0100680 if (ret < 0) {
Kent Overstreetbdb53202013-11-23 17:26:46 -0800681 bio_put(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200682 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100683 }
Peter Osterlund36763472005-09-06 15:16:42 -0700684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Kent Overstreetbdb53202013-11-23 17:26:46 -0800686 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700688EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690/**
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800691 * bio_add_pc_page - attempt to add page to bio
692 * @q: the target queue
693 * @bio: destination bio
694 * @page: page to add
695 * @len: vec entry length
696 * @offset: vec entry offset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800698 * Attempt to add a page to the bio_vec maplist. This can fail for a
699 * number of reasons, such as the bio being full or target block device
700 * limitations. The target block device must allow bio's up to PAGE_SIZE,
701 * so it is always possible to add a single page to an empty bio.
702 *
703 * This should only be used by REQ_PC bios.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800705int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
706 *page, unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708 int retried_segments = 0;
709 struct bio_vec *bvec;
710
711 /*
712 * cloned bio must not modify vec list
713 */
714 if (unlikely(bio_flagged(bio, BIO_CLONED)))
715 return 0;
716
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800717 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 return 0;
719
Jens Axboe80cfd542006-01-06 09:43:28 +0100720 /*
721 * For filesystems with a blocksize smaller than the pagesize
722 * we will often be called with the same page as last time and
723 * a consecutive offset. Optimize this special case.
724 */
725 if (bio->bi_vcnt > 0) {
726 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
727
728 if (page == prev->bv_page &&
729 offset == prev->bv_offset + prev->bv_len) {
730 prev->bv_len += len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800731 bio->bi_iter.bi_size += len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100732 goto done;
733 }
Jens Axboe66cb45a2014-06-24 16:22:24 -0600734
735 /*
736 * If the queue doesn't support SG gaps and adding this
737 * offset would create a gap, disallow it.
738 */
Keith Busch03100aa2015-08-19 14:24:05 -0700739 if (bvec_gap_to_prev(q, prev, offset))
Jens Axboe66cb45a2014-06-24 16:22:24 -0600740 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100741 }
742
743 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 return 0;
745
746 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 * setup the new entry, we might clear it again later if we
748 * cannot add the page
749 */
750 bvec = &bio->bi_io_vec[bio->bi_vcnt];
751 bvec->bv_page = page;
752 bvec->bv_len = len;
753 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800754 bio->bi_vcnt++;
755 bio->bi_phys_segments++;
756 bio->bi_iter.bi_size += len;
757
758 /*
759 * Perform a recount if the number of segments is greater
760 * than queue_max_segments(q).
761 */
762
763 while (bio->bi_phys_segments > queue_max_segments(q)) {
764
765 if (retried_segments)
766 goto failed;
767
768 retried_segments = 1;
769 blk_recount_segments(q, bio);
770 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 /* If we may be able to merge these biovecs, force a recount */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800773 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600774 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Jens Axboe80cfd542006-01-06 09:43:28 +0100776 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800778
779 failed:
780 bvec->bv_page = NULL;
781 bvec->bv_len = 0;
782 bvec->bv_offset = 0;
783 bio->bi_vcnt--;
784 bio->bi_iter.bi_size -= len;
785 blk_recount_segments(q, bio);
786 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200788EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600789
790/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 * bio_add_page - attempt to add page to bio
792 * @bio: destination bio
793 * @page: page to add
794 * @len: vec entry length
795 * @offset: vec entry offset
796 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800797 * Attempt to add a page to the bio_vec maplist. This will only fail
798 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800800int bio_add_page(struct bio *bio, struct page *page,
801 unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800803 struct bio_vec *bv;
Jens Axboe762380a2014-06-05 13:38:39 -0600804
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800805 /*
806 * cloned bio must not modify vec list
807 */
808 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
809 return 0;
Jens Axboe58a49152014-06-10 12:53:56 -0600810
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800811 /*
812 * For filesystems with a blocksize smaller than the pagesize
813 * we will often be called with the same page as last time and
814 * a consecutive offset. Optimize this special case.
815 */
816 if (bio->bi_vcnt > 0) {
817 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
818
819 if (page == bv->bv_page &&
820 offset == bv->bv_offset + bv->bv_len) {
821 bv->bv_len += len;
822 goto done;
823 }
824 }
825
826 if (bio->bi_vcnt >= bio->bi_max_vecs)
827 return 0;
828
829 bv = &bio->bi_io_vec[bio->bi_vcnt];
830 bv->bv_page = page;
831 bv->bv_len = len;
832 bv->bv_offset = offset;
833
834 bio->bi_vcnt++;
835done:
836 bio->bi_iter.bi_size += len;
837 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200839EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Kent Overstreet9e882242012-09-10 14:41:12 -0700841struct submit_bio_ret {
842 struct completion event;
843 int error;
844};
845
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200846static void submit_bio_wait_endio(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700847{
848 struct submit_bio_ret *ret = bio->bi_private;
849
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200850 ret->error = bio->bi_error;
Kent Overstreet9e882242012-09-10 14:41:12 -0700851 complete(&ret->event);
852}
853
854/**
855 * submit_bio_wait - submit a bio, and wait until it completes
Kent Overstreet9e882242012-09-10 14:41:12 -0700856 * @bio: The &struct bio which describes the I/O
857 *
858 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
859 * bio_endio() on failure.
860 */
Mike Christie4e49ea42016-06-05 14:31:41 -0500861int submit_bio_wait(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700862{
863 struct submit_bio_ret ret;
864
Kent Overstreet9e882242012-09-10 14:41:12 -0700865 init_completion(&ret.event);
866 bio->bi_private = &ret;
867 bio->bi_end_io = submit_bio_wait_endio;
Mike Christie4e49ea42016-06-05 14:31:41 -0500868 bio->bi_rw |= REQ_SYNC;
869 submit_bio(bio);
Stephane Gasparinid57d6112016-02-09 17:07:38 +0100870 wait_for_completion_io(&ret.event);
Kent Overstreet9e882242012-09-10 14:41:12 -0700871
872 return ret.error;
873}
874EXPORT_SYMBOL(submit_bio_wait);
875
Kent Overstreet054bdf62012-09-28 13:17:55 -0700876/**
877 * bio_advance - increment/complete a bio by some number of bytes
878 * @bio: bio to advance
879 * @bytes: number of bytes to complete
880 *
881 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
882 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
883 * be updated on the last bvec as well.
884 *
885 * @bio will then represent the remaining, uncompleted portion of the io.
886 */
887void bio_advance(struct bio *bio, unsigned bytes)
888{
889 if (bio_integrity(bio))
890 bio_integrity_advance(bio, bytes);
891
Kent Overstreet4550dd62013-08-07 14:26:21 -0700892 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -0700893}
894EXPORT_SYMBOL(bio_advance);
895
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700896/**
Kent Overstreeta0787602012-09-10 14:03:28 -0700897 * bio_alloc_pages - allocates a single page for each bvec in a bio
898 * @bio: bio to allocate pages for
899 * @gfp_mask: flags for allocation
900 *
901 * Allocates pages up to @bio->bi_vcnt.
902 *
903 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
904 * freed.
905 */
906int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
907{
908 int i;
909 struct bio_vec *bv;
910
911 bio_for_each_segment_all(bv, bio, i) {
912 bv->bv_page = alloc_page(gfp_mask);
913 if (!bv->bv_page) {
914 while (--bv >= bio->bi_io_vec)
915 __free_page(bv->bv_page);
916 return -ENOMEM;
917 }
918 }
919
920 return 0;
921}
922EXPORT_SYMBOL(bio_alloc_pages);
923
924/**
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700925 * bio_copy_data - copy contents of data buffers from one chain of bios to
926 * another
927 * @src: source bio list
928 * @dst: destination bio list
929 *
930 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
931 * @src and @dst as linked lists of bios.
932 *
933 * Stops when it reaches the end of either @src or @dst - that is, copies
934 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
935 */
936void bio_copy_data(struct bio *dst, struct bio *src)
937{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700938 struct bvec_iter src_iter, dst_iter;
939 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700940 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700941 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700942
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700943 src_iter = src->bi_iter;
944 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700945
946 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700947 if (!src_iter.bi_size) {
948 src = src->bi_next;
949 if (!src)
950 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700951
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700952 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700953 }
954
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700955 if (!dst_iter.bi_size) {
956 dst = dst->bi_next;
957 if (!dst)
958 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700959
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700960 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700961 }
962
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700963 src_bv = bio_iter_iovec(src, src_iter);
964 dst_bv = bio_iter_iovec(dst, dst_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700965
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700966 bytes = min(src_bv.bv_len, dst_bv.bv_len);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700967
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700968 src_p = kmap_atomic(src_bv.bv_page);
969 dst_p = kmap_atomic(dst_bv.bv_page);
970
971 memcpy(dst_p + dst_bv.bv_offset,
972 src_p + src_bv.bv_offset,
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700973 bytes);
974
975 kunmap_atomic(dst_p);
976 kunmap_atomic(src_p);
977
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700978 bio_advance_iter(src, &src_iter, bytes);
979 bio_advance_iter(dst, &dst_iter, bytes);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700980 }
981}
982EXPORT_SYMBOL(bio_copy_data);
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900985 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +0100986 struct iov_iter iter;
987 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988};
989
Fabian Frederick7410b3c2014-04-22 15:09:07 -0600990static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200991 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992{
Jens Axboef3f63c12010-10-29 11:46:56 -0600993 if (iov_count > UIO_MAXIOV)
994 return NULL;
995
Kent Overstreetc8db4442013-11-22 19:39:06 -0800996 return kmalloc(sizeof(struct bio_map_data) +
Kent Overstreet26e49cf2015-01-18 16:16:31 +0100997 sizeof(struct iovec) * iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998}
999
Dongsu Park9124d3f2015-01-18 16:16:34 +01001000/**
1001 * bio_copy_from_iter - copy all pages from iov_iter to bio
1002 * @bio: The &struct bio which describes the I/O as destination
1003 * @iter: iov_iter as source
1004 *
1005 * Copy all pages from iov_iter to bio.
1006 * Returns 0 on success, or error on failure.
1007 */
1008static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001009{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001010 int i;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001011 struct bio_vec *bvec;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001012
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001013 bio_for_each_segment_all(bvec, bio, i) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001014 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001015
Dongsu Park9124d3f2015-01-18 16:16:34 +01001016 ret = copy_page_from_iter(bvec->bv_page,
1017 bvec->bv_offset,
1018 bvec->bv_len,
1019 &iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001020
Dongsu Park9124d3f2015-01-18 16:16:34 +01001021 if (!iov_iter_count(&iter))
1022 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001023
Dongsu Park9124d3f2015-01-18 16:16:34 +01001024 if (ret < bvec->bv_len)
1025 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001026 }
1027
Dongsu Park9124d3f2015-01-18 16:16:34 +01001028 return 0;
1029}
1030
1031/**
1032 * bio_copy_to_iter - copy all pages from bio to iov_iter
1033 * @bio: The &struct bio which describes the I/O as source
1034 * @iter: iov_iter as destination
1035 *
1036 * Copy all pages from bio to iov_iter.
1037 * Returns 0 on success, or error on failure.
1038 */
1039static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1040{
1041 int i;
1042 struct bio_vec *bvec;
1043
1044 bio_for_each_segment_all(bvec, bio, i) {
1045 ssize_t ret;
1046
1047 ret = copy_page_to_iter(bvec->bv_page,
1048 bvec->bv_offset,
1049 bvec->bv_len,
1050 &iter);
1051
1052 if (!iov_iter_count(&iter))
1053 break;
1054
1055 if (ret < bvec->bv_len)
1056 return -EFAULT;
1057 }
1058
1059 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001060}
1061
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001062static void bio_free_pages(struct bio *bio)
1063{
1064 struct bio_vec *bvec;
1065 int i;
1066
1067 bio_for_each_segment_all(bvec, bio, i)
1068 __free_page(bvec->bv_page);
1069}
1070
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071/**
1072 * bio_uncopy_user - finish previously mapped bio
1073 * @bio: bio being terminated
1074 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001075 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 * to user space in case of a read.
1077 */
1078int bio_uncopy_user(struct bio *bio)
1079{
1080 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001081 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Roland Dreier35dc2482013-08-05 17:55:01 -07001083 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1084 /*
1085 * if we're in a workqueue, the request is orphaned, so
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001086 * don't copy into a random user address space, just free
1087 * and return -EINTR so user space doesn't expect any data.
Roland Dreier35dc2482013-08-05 17:55:01 -07001088 */
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001089 if (!current->mm)
1090 ret = -EINTR;
1091 else if (bio_data_dir(bio) == READ)
Dongsu Park9124d3f2015-01-18 16:16:34 +01001092 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001093 if (bmd->is_our_pages)
1094 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001095 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001096 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 bio_put(bio);
1098 return ret;
1099}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001100EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001103 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001104 * @q: destination block queue
1105 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1106 * @iter: iovec iterator
1107 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 *
1109 * Prepares and returns a bio for indirect user io, bouncing data
1110 * to/from kernel pages as necessary. Must be paired with
1111 * call bio_uncopy_user() on io completion.
1112 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001113struct bio *bio_copy_user_iov(struct request_queue *q,
1114 struct rq_map_data *map_data,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001115 const struct iov_iter *iter,
1116 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 struct page *page;
1120 struct bio *bio;
1121 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001122 int nr_pages = 0;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001123 unsigned int len = iter->count;
Geliang Tangbd5cece2015-11-21 17:27:31 +08001124 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001126 for (i = 0; i < iter->nr_segs; i++) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001127 unsigned long uaddr;
1128 unsigned long end;
1129 unsigned long start;
1130
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001131 uaddr = (unsigned long) iter->iov[i].iov_base;
1132 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1133 >> PAGE_SHIFT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001134 start = uaddr >> PAGE_SHIFT;
1135
Jens Axboecb4644c2010-11-10 14:36:25 +01001136 /*
1137 * Overflow, abort
1138 */
1139 if (end < start)
1140 return ERR_PTR(-EINVAL);
1141
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001142 nr_pages += end - start;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001143 }
1144
FUJITA Tomonori69838722009-04-28 20:24:29 +02001145 if (offset)
1146 nr_pages++;
1147
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001148 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 if (!bmd)
1150 return ERR_PTR(-ENOMEM);
1151
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001152 /*
1153 * We need to do a deep copy of the iov_iter including the iovecs.
1154 * The caller provided iov might point to an on-stack or otherwise
1155 * shortlived one.
1156 */
1157 bmd->is_our_pages = map_data ? 0 : 1;
1158 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1159 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1160 iter->nr_segs, iter->count);
1161
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001163 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (!bio)
1165 goto out_bmd;
1166
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001167 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001168 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001171
1172 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001173 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001174 i = map_data->offset / PAGE_SIZE;
1175 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001177 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001179 bytes -= offset;
1180
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 if (bytes > len)
1182 bytes = len;
1183
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001184 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001185 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001186 ret = -ENOMEM;
1187 break;
1188 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001189
1190 page = map_data->pages[i / nr_pages];
1191 page += (i % nr_pages);
1192
1193 i++;
1194 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001195 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001196 if (!page) {
1197 ret = -ENOMEM;
1198 break;
1199 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
1201
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001202 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
1205 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001206 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
1208
1209 if (ret)
1210 goto cleanup;
1211
1212 /*
1213 * success
1214 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001215 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001216 (map_data && map_data->from_user)) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001217 ret = bio_copy_from_iter(bio, *iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001218 if (ret)
1219 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
1221
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001222 bio->bi_private = bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 return bio;
1224cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001225 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001226 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 bio_put(bio);
1228out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001229 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 return ERR_PTR(ret);
1231}
1232
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001233/**
1234 * bio_map_user_iov - map user iovec into bio
1235 * @q: the struct request_queue for the bio
1236 * @iter: iovec iterator
1237 * @gfp_mask: memory allocation flags
1238 *
1239 * Map the user space address into a bio suitable for io to a block
1240 * device. Returns an error pointer in case of error.
1241 */
1242struct bio *bio_map_user_iov(struct request_queue *q,
1243 const struct iov_iter *iter,
1244 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001246 int j;
James Bottomley f1970ba2005-06-20 14:06:52 +02001247 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 struct page **pages;
1249 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001250 int cur_page = 0;
1251 int ret, offset;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001252 struct iov_iter i;
1253 struct iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001255 iov_for_each(iov, i, *iter) {
1256 unsigned long uaddr = (unsigned long) iov.iov_base;
1257 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001258 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1259 unsigned long start = uaddr >> PAGE_SHIFT;
1260
Jens Axboecb4644c2010-11-10 14:36:25 +01001261 /*
1262 * Overflow, abort
1263 */
1264 if (end < start)
1265 return ERR_PTR(-EINVAL);
1266
James Bottomley f1970ba2005-06-20 14:06:52 +02001267 nr_pages += end - start;
1268 /*
Mike Christiead2d7222006-12-01 10:40:20 +01001269 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +02001270 */
Mike Christiead2d7222006-12-01 10:40:20 +01001271 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +02001272 return ERR_PTR(-EINVAL);
1273 }
1274
1275 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 return ERR_PTR(-EINVAL);
1277
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001278 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (!bio)
1280 return ERR_PTR(-ENOMEM);
1281
1282 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001283 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (!pages)
1285 goto out;
1286
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001287 iov_for_each(iov, i, *iter) {
1288 unsigned long uaddr = (unsigned long) iov.iov_base;
1289 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001290 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1291 unsigned long start = uaddr >> PAGE_SHIFT;
1292 const int local_nr_pages = end - start;
1293 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001294
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001295 ret = get_user_pages_fast(uaddr, local_nr_pages,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001296 (iter->type & WRITE) != WRITE,
1297 &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +02001298 if (ret < local_nr_pages) {
1299 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001300 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Geliang Tangbd5cece2015-11-21 17:27:31 +08001303 offset = offset_in_page(uaddr);
James Bottomley f1970ba2005-06-20 14:06:52 +02001304 for (j = cur_page; j < page_limit; j++) {
1305 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
James Bottomley f1970ba2005-06-20 14:06:52 +02001307 if (len <= 0)
1308 break;
1309
1310 if (bytes > len)
1311 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
James Bottomley f1970ba2005-06-20 14:06:52 +02001313 /*
1314 * sorry...
1315 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001316 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1317 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001318 break;
1319
1320 len -= bytes;
1321 offset = 0;
1322 }
1323
1324 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001326 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001328 while (j < page_limit)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001329 put_page(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 }
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 kfree(pages);
1333
1334 /*
1335 * set data direction, and check if mapped pages need bouncing
1336 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001337 if (iter->type & WRITE)
Mike Christie95fe6c12016-06-05 14:31:48 -05001338 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001340 bio_set_flag(bio, BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001341
1342 /*
1343 * subtle -- if __bio_map_user() ended up bouncing a bio,
1344 * it would normally disappear when its bi_end_io is run.
1345 * however, we need it for the unmap, so grab an extra
1346 * reference to it
1347 */
1348 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001350
1351 out_unmap:
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001352 for (j = 0; j < nr_pages; j++) {
1353 if (!pages[j])
James Bottomley f1970ba2005-06-20 14:06:52 +02001354 break;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001355 put_page(pages[j]);
James Bottomley f1970ba2005-06-20 14:06:52 +02001356 }
1357 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 kfree(pages);
1359 bio_put(bio);
1360 return ERR_PTR(ret);
1361}
1362
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363static void __bio_unmap_user(struct bio *bio)
1364{
1365 struct bio_vec *bvec;
1366 int i;
1367
1368 /*
1369 * make sure we dirty pages we wrote to
1370 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001371 bio_for_each_segment_all(bvec, bio, i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (bio_data_dir(bio) == READ)
1373 set_page_dirty_lock(bvec->bv_page);
1374
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001375 put_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
1377
1378 bio_put(bio);
1379}
1380
1381/**
1382 * bio_unmap_user - unmap a bio
1383 * @bio: the bio being unmapped
1384 *
1385 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1386 * a process context.
1387 *
1388 * bio_unmap_user() may sleep.
1389 */
1390void bio_unmap_user(struct bio *bio)
1391{
1392 __bio_unmap_user(bio);
1393 bio_put(bio);
1394}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001395EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001397static void bio_map_kern_endio(struct bio *bio)
Jens Axboeb8238252005-06-20 14:05:27 +02001398{
Jens Axboeb8238252005-06-20 14:05:27 +02001399 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001400}
1401
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001402/**
1403 * bio_map_kern - map kernel address into bio
1404 * @q: the struct request_queue for the bio
1405 * @data: pointer to buffer to map
1406 * @len: length in bytes
1407 * @gfp_mask: allocation flags for bio allocation
1408 *
1409 * Map the kernel address into a bio suitable for io to a block
1410 * device. Returns an error pointer in case of error.
1411 */
1412struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1413 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001414{
1415 unsigned long kaddr = (unsigned long)data;
1416 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1417 unsigned long start = kaddr >> PAGE_SHIFT;
1418 const int nr_pages = end - start;
1419 int offset, i;
1420 struct bio *bio;
1421
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001422 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001423 if (!bio)
1424 return ERR_PTR(-ENOMEM);
1425
1426 offset = offset_in_page(kaddr);
1427 for (i = 0; i < nr_pages; i++) {
1428 unsigned int bytes = PAGE_SIZE - offset;
1429
1430 if (len <= 0)
1431 break;
1432
1433 if (bytes > len)
1434 bytes = len;
1435
Mike Christiedefd94b2005-12-05 02:37:06 -06001436 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001437 offset) < bytes) {
1438 /* we don't support partial mappings */
1439 bio_put(bio);
1440 return ERR_PTR(-EINVAL);
1441 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001442
1443 data += bytes;
1444 len -= bytes;
1445 offset = 0;
1446 }
1447
Jens Axboeb8238252005-06-20 14:05:27 +02001448 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001449 return bio;
1450}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001451EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001452
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001453static void bio_copy_kern_endio(struct bio *bio)
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001454{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001455 bio_free_pages(bio);
1456 bio_put(bio);
1457}
1458
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001459static void bio_copy_kern_endio_read(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001460{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001461 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001462 struct bio_vec *bvec;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001463 int i;
1464
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001465 bio_for_each_segment_all(bvec, bio, i) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001466 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001467 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001468 }
1469
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001470 bio_copy_kern_endio(bio);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001471}
1472
1473/**
1474 * bio_copy_kern - copy kernel address into bio
1475 * @q: the struct request_queue for the bio
1476 * @data: pointer to buffer to copy
1477 * @len: length in bytes
1478 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001479 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001480 *
1481 * copy the kernel address into a bio suitable for io to a block
1482 * device. Returns an error pointer in case of error.
1483 */
1484struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1485 gfp_t gfp_mask, int reading)
1486{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001487 unsigned long kaddr = (unsigned long)data;
1488 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1489 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001490 struct bio *bio;
1491 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001492 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001493
Christoph Hellwig42d26832015-01-18 16:16:28 +01001494 /*
1495 * Overflow, abort
1496 */
1497 if (end < start)
1498 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001499
Christoph Hellwig42d26832015-01-18 16:16:28 +01001500 nr_pages = end - start;
1501 bio = bio_kmalloc(gfp_mask, nr_pages);
1502 if (!bio)
1503 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001504
Christoph Hellwig42d26832015-01-18 16:16:28 +01001505 while (len) {
1506 struct page *page;
1507 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001508
Christoph Hellwig42d26832015-01-18 16:16:28 +01001509 if (bytes > len)
1510 bytes = len;
1511
1512 page = alloc_page(q->bounce_gfp | gfp_mask);
1513 if (!page)
1514 goto cleanup;
1515
1516 if (!reading)
1517 memcpy(page_address(page), p, bytes);
1518
1519 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1520 break;
1521
1522 len -= bytes;
1523 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001524 }
1525
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001526 if (reading) {
1527 bio->bi_end_io = bio_copy_kern_endio_read;
1528 bio->bi_private = data;
1529 } else {
1530 bio->bi_end_io = bio_copy_kern_endio;
Mike Christie95fe6c12016-06-05 14:31:48 -05001531 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001532 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001533
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001534 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001535
1536cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001537 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001538 bio_put(bio);
1539 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001540}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001541EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543/*
1544 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1545 * for performing direct-IO in BIOs.
1546 *
1547 * The problem is that we cannot run set_page_dirty() from interrupt context
1548 * because the required locks are not interrupt-safe. So what we can do is to
1549 * mark the pages dirty _before_ performing IO. And in interrupt context,
1550 * check that the pages are still dirty. If so, fine. If not, redirty them
1551 * in process context.
1552 *
1553 * We special-case compound pages here: normally this means reads into hugetlb
1554 * pages. The logic in here doesn't really work right for compound pages
1555 * because the VM does not uniformly chase down the head page in all cases.
1556 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1557 * handle them at all. So we skip compound pages here at an early stage.
1558 *
1559 * Note that this code is very hard to test under normal circumstances because
1560 * direct-io pins the pages with get_user_pages(). This makes
1561 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001562 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 * pagecache.
1564 *
1565 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1566 * deferred bio dirtying paths.
1567 */
1568
1569/*
1570 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1571 */
1572void bio_set_pages_dirty(struct bio *bio)
1573{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001574 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 int i;
1576
Kent Overstreetcb34e052012-09-05 15:22:02 -07001577 bio_for_each_segment_all(bvec, bio, i) {
1578 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 if (page && !PageCompound(page))
1581 set_page_dirty_lock(page);
1582 }
1583}
1584
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001585static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001587 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 int i;
1589
Kent Overstreetcb34e052012-09-05 15:22:02 -07001590 bio_for_each_segment_all(bvec, bio, i) {
1591 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
1593 if (page)
1594 put_page(page);
1595 }
1596}
1597
1598/*
1599 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1600 * If they are, then fine. If, however, some pages are clean then they must
1601 * have been written out during the direct-IO read. So we take another ref on
1602 * the BIO and the offending pages and re-dirty the pages in process context.
1603 *
1604 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +03001605 * here on. It will run one put_page() against each page and will run one
1606 * bio_put() against the BIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 */
1608
David Howells65f27f32006-11-22 14:55:48 +00001609static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
David Howells65f27f32006-11-22 14:55:48 +00001611static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612static DEFINE_SPINLOCK(bio_dirty_lock);
1613static struct bio *bio_dirty_list;
1614
1615/*
1616 * This runs in process context
1617 */
David Howells65f27f32006-11-22 14:55:48 +00001618static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
1620 unsigned long flags;
1621 struct bio *bio;
1622
1623 spin_lock_irqsave(&bio_dirty_lock, flags);
1624 bio = bio_dirty_list;
1625 bio_dirty_list = NULL;
1626 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1627
1628 while (bio) {
1629 struct bio *next = bio->bi_private;
1630
1631 bio_set_pages_dirty(bio);
1632 bio_release_pages(bio);
1633 bio_put(bio);
1634 bio = next;
1635 }
1636}
1637
1638void bio_check_pages_dirty(struct bio *bio)
1639{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001640 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 int nr_clean_pages = 0;
1642 int i;
1643
Kent Overstreetcb34e052012-09-05 15:22:02 -07001644 bio_for_each_segment_all(bvec, bio, i) {
1645 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
1647 if (PageDirty(page) || PageCompound(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001648 put_page(page);
Kent Overstreetcb34e052012-09-05 15:22:02 -07001649 bvec->bv_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 } else {
1651 nr_clean_pages++;
1652 }
1653 }
1654
1655 if (nr_clean_pages) {
1656 unsigned long flags;
1657
1658 spin_lock_irqsave(&bio_dirty_lock, flags);
1659 bio->bi_private = bio_dirty_list;
1660 bio_dirty_list = bio;
1661 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1662 schedule_work(&bio_dirty_work);
1663 } else {
1664 bio_put(bio);
1665 }
1666}
1667
Gu Zheng394ffa52014-11-24 11:05:22 +08001668void generic_start_io_acct(int rw, unsigned long sectors,
1669 struct hd_struct *part)
1670{
1671 int cpu = part_stat_lock();
1672
1673 part_round_stats(cpu, part);
1674 part_stat_inc(cpu, part, ios[rw]);
1675 part_stat_add(cpu, part, sectors[rw], sectors);
1676 part_inc_in_flight(part, rw);
1677
1678 part_stat_unlock();
1679}
1680EXPORT_SYMBOL(generic_start_io_acct);
1681
1682void generic_end_io_acct(int rw, struct hd_struct *part,
1683 unsigned long start_time)
1684{
1685 unsigned long duration = jiffies - start_time;
1686 int cpu = part_stat_lock();
1687
1688 part_stat_add(cpu, part, ticks[rw], duration);
1689 part_round_stats(cpu, part);
1690 part_dec_in_flight(part, rw);
1691
1692 part_stat_unlock();
1693}
1694EXPORT_SYMBOL(generic_end_io_acct);
1695
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001696#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1697void bio_flush_dcache_pages(struct bio *bi)
1698{
Kent Overstreet79886132013-11-23 17:19:00 -08001699 struct bio_vec bvec;
1700 struct bvec_iter iter;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001701
Kent Overstreet79886132013-11-23 17:19:00 -08001702 bio_for_each_segment(bvec, bi, iter)
1703 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001704}
1705EXPORT_SYMBOL(bio_flush_dcache_pages);
1706#endif
1707
Jens Axboec4cf5262015-04-17 16:15:18 -06001708static inline bool bio_remaining_done(struct bio *bio)
1709{
1710 /*
1711 * If we're not chaining, then ->__bi_remaining is always 1 and
1712 * we always end io on the first invocation.
1713 */
1714 if (!bio_flagged(bio, BIO_CHAIN))
1715 return true;
1716
1717 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1718
Mike Snitzer326e1db2015-05-22 09:14:03 -04001719 if (atomic_dec_and_test(&bio->__bi_remaining)) {
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001720 bio_clear_flag(bio, BIO_CHAIN);
Jens Axboec4cf5262015-04-17 16:15:18 -06001721 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001722 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001723
1724 return false;
1725}
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727/**
1728 * bio_endio - end I/O on a bio
1729 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 *
1731 * Description:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001732 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1733 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1734 * bio unless they own it and thus know that it has an end_io function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 **/
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001736void bio_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737{
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001738again:
Christoph Hellwig2b885512016-03-11 17:34:53 +01001739 if (!bio_remaining_done(bio))
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001740 return;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001741
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001742 /*
1743 * Need to have a real endio function for chained bios, otherwise
1744 * various corner cases will break (like stacking block devices that
1745 * save/restore bi_end_io) - however, we want to avoid unbounded
1746 * recursion and blowing the stack. Tail call optimization would
1747 * handle this, but compiling with frame pointers also disables
1748 * gcc's sibling call optimization.
1749 */
1750 if (bio->bi_end_io == bio_chain_endio) {
1751 bio = __bio_chain_endio(bio);
1752 goto again;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001753 }
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001754
1755 if (bio->bi_end_io)
1756 bio->bi_end_io(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001758EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
Kent Overstreet196d38b2013-11-23 18:34:15 -08001760/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001761 * bio_split - split a bio
1762 * @bio: bio to split
1763 * @sectors: number of sectors to split from the front of @bio
1764 * @gfp: gfp mask
1765 * @bs: bio set to allocate from
1766 *
1767 * Allocates and returns a new bio which represents @sectors from the start of
1768 * @bio, and updates @bio to represent the remaining sectors.
1769 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001770 * Unless this is a discard request the newly allocated bio will point
1771 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1772 * @bio is not freed before the split.
Kent Overstreet20d01892013-11-23 18:21:01 -08001773 */
1774struct bio *bio_split(struct bio *bio, int sectors,
1775 gfp_t gfp, struct bio_set *bs)
1776{
1777 struct bio *split = NULL;
1778
1779 BUG_ON(sectors <= 0);
1780 BUG_ON(sectors >= bio_sectors(bio));
1781
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001782 /*
1783 * Discards need a mutable bio_vec to accommodate the payload
1784 * required by the DSM TRIM and UNMAP commands.
1785 */
Mike Christie95fe6c12016-06-05 14:31:48 -05001786 if (bio_op(bio) == REQ_OP_DISCARD)
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001787 split = bio_clone_bioset(bio, gfp, bs);
1788 else
1789 split = bio_clone_fast(bio, gfp, bs);
1790
Kent Overstreet20d01892013-11-23 18:21:01 -08001791 if (!split)
1792 return NULL;
1793
1794 split->bi_iter.bi_size = sectors << 9;
1795
1796 if (bio_integrity(split))
1797 bio_integrity_trim(split, 0, sectors);
1798
1799 bio_advance(bio, split->bi_iter.bi_size);
1800
1801 return split;
1802}
1803EXPORT_SYMBOL(bio_split);
1804
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001805/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001806 * bio_trim - trim a bio
1807 * @bio: bio to trim
1808 * @offset: number of sectors to trim from the front of @bio
1809 * @size: size we want to trim @bio to, in sectors
1810 */
1811void bio_trim(struct bio *bio, int offset, int size)
1812{
1813 /* 'bio' is a cloned bio which we need to trim to match
1814 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001815 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001816
1817 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001818 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001819 return;
1820
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001821 bio_clear_flag(bio, BIO_SEG_VALID);
Kent Overstreet6678d832013-08-07 11:14:32 -07001822
1823 bio_advance(bio, offset << 9);
1824
Kent Overstreet4f024f32013-10-11 15:44:27 -07001825 bio->bi_iter.bi_size = size;
Kent Overstreet6678d832013-08-07 11:14:32 -07001826}
1827EXPORT_SYMBOL_GPL(bio_trim);
1828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829/*
1830 * create memory pools for biovec's in a bio_set.
1831 * use the global biovec slabs created for general use.
1832 */
Fabian Fredericka6c39cb4f2014-04-22 15:09:05 -06001833mempool_t *biovec_create_pool(int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834{
Jens Axboe7ff93452008-12-11 11:53:43 +01001835 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Kent Overstreet9f060e22012-10-12 15:29:33 -07001837 return mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838}
1839
1840void bioset_free(struct bio_set *bs)
1841{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001842 if (bs->rescue_workqueue)
1843 destroy_workqueue(bs->rescue_workqueue);
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 if (bs->bio_pool)
1846 mempool_destroy(bs->bio_pool);
1847
Kent Overstreet9f060e22012-10-12 15:29:33 -07001848 if (bs->bvec_pool)
1849 mempool_destroy(bs->bvec_pool);
1850
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001851 bioset_integrity_free(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001852 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
1854 kfree(bs);
1855}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001856EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001858static struct bio_set *__bioset_create(unsigned int pool_size,
1859 unsigned int front_pad,
1860 bool create_bvec_pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
Jens Axboe392ddc32008-12-23 12:42:54 +01001862 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001863 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Jens Axboe1b434492008-10-22 20:32:58 +02001865 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 if (!bs)
1867 return NULL;
1868
Jens Axboebb799ca2008-12-10 15:35:05 +01001869 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001870
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001871 spin_lock_init(&bs->rescue_lock);
1872 bio_list_init(&bs->rescue_list);
1873 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1874
Jens Axboe392ddc32008-12-23 12:42:54 +01001875 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001876 if (!bs->bio_slab) {
1877 kfree(bs);
1878 return NULL;
1879 }
1880
1881 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 if (!bs->bio_pool)
1883 goto bad;
1884
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001885 if (create_bvec_pool) {
1886 bs->bvec_pool = biovec_create_pool(pool_size);
1887 if (!bs->bvec_pool)
1888 goto bad;
1889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001891 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1892 if (!bs->rescue_workqueue)
1893 goto bad;
1894
1895 return bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896bad:
1897 bioset_free(bs);
1898 return NULL;
1899}
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001900
1901/**
1902 * bioset_create - Create a bio_set
1903 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1904 * @front_pad: Number of bytes to allocate in front of the returned bio
1905 *
1906 * Description:
1907 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1908 * to ask for a number of bytes to be allocated in front of the bio.
1909 * Front pad allocation is useful for embedding the bio inside
1910 * another structure, to avoid allocating extra data to go with the bio.
1911 * Note that the bio must be embedded at the END of that structure always,
1912 * or things will break badly.
1913 */
1914struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1915{
1916 return __bioset_create(pool_size, front_pad, true);
1917}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001918EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001920/**
1921 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1922 * @pool_size: Number of bio to cache in the mempool
1923 * @front_pad: Number of bytes to allocate in front of the returned bio
1924 *
1925 * Description:
1926 * Same functionality as bioset_create() except that mempool is not
1927 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1928 */
1929struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1930{
1931 return __bioset_create(pool_size, front_pad, false);
1932}
1933EXPORT_SYMBOL(bioset_create_nobvec);
1934
Tejun Heo852c7882012-03-05 13:15:27 -08001935#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04001936
1937/**
1938 * bio_associate_blkcg - associate a bio with the specified blkcg
1939 * @bio: target bio
1940 * @blkcg_css: css of the blkcg to associate
1941 *
1942 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1943 * treat @bio as if it were issued by a task which belongs to the blkcg.
1944 *
1945 * This function takes an extra reference of @blkcg_css which will be put
1946 * when @bio is released. The caller must own @bio and is responsible for
1947 * synchronizing calls to this function.
1948 */
1949int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1950{
1951 if (unlikely(bio->bi_css))
1952 return -EBUSY;
1953 css_get(blkcg_css);
1954 bio->bi_css = blkcg_css;
1955 return 0;
1956}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001957EXPORT_SYMBOL_GPL(bio_associate_blkcg);
Tejun Heo1d933cf2015-05-22 17:13:24 -04001958
Tejun Heo852c7882012-03-05 13:15:27 -08001959/**
1960 * bio_associate_current - associate a bio with %current
1961 * @bio: target bio
1962 *
1963 * Associate @bio with %current if it hasn't been associated yet. Block
1964 * layer will treat @bio as if it were issued by %current no matter which
1965 * task actually issues it.
1966 *
1967 * This function takes an extra reference of @task's io_context and blkcg
1968 * which will be put when @bio is released. The caller must own @bio,
1969 * ensure %current->io_context exists, and is responsible for synchronizing
1970 * calls to this function.
1971 */
1972int bio_associate_current(struct bio *bio)
1973{
1974 struct io_context *ioc;
Tejun Heo852c7882012-03-05 13:15:27 -08001975
Tejun Heo1d933cf2015-05-22 17:13:24 -04001976 if (bio->bi_css)
Tejun Heo852c7882012-03-05 13:15:27 -08001977 return -EBUSY;
1978
1979 ioc = current->io_context;
1980 if (!ioc)
1981 return -ENOENT;
1982
Tejun Heo852c7882012-03-05 13:15:27 -08001983 get_io_context_active(ioc);
1984 bio->bi_ioc = ioc;
Tejun Heoc165b3e2015-08-18 14:55:29 -07001985 bio->bi_css = task_get_css(current, io_cgrp_id);
Tejun Heo852c7882012-03-05 13:15:27 -08001986 return 0;
1987}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001988EXPORT_SYMBOL_GPL(bio_associate_current);
Tejun Heo852c7882012-03-05 13:15:27 -08001989
1990/**
1991 * bio_disassociate_task - undo bio_associate_current()
1992 * @bio: target bio
1993 */
1994void bio_disassociate_task(struct bio *bio)
1995{
1996 if (bio->bi_ioc) {
1997 put_io_context(bio->bi_ioc);
1998 bio->bi_ioc = NULL;
1999 }
2000 if (bio->bi_css) {
2001 css_put(bio->bi_css);
2002 bio->bi_css = NULL;
2003 }
2004}
2005
2006#endif /* CONFIG_BLK_CGROUP */
2007
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008static void __init biovec_init_slabs(void)
2009{
2010 int i;
2011
2012 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2013 int size;
2014 struct biovec_slab *bvs = bvec_slabs + i;
2015
Jens Axboea7fcd372008-12-05 16:10:29 +01002016 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2017 bvs->slab = NULL;
2018 continue;
2019 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002020
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 size = bvs->nr_vecs * sizeof(struct bio_vec);
2022 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002023 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 }
2025}
2026
2027static int __init init_bio(void)
2028{
Jens Axboebb799ca2008-12-10 15:35:05 +01002029 bio_slab_max = 2;
2030 bio_slab_nr = 0;
2031 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2032 if (!bio_slabs)
2033 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002035 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 biovec_init_slabs();
2037
Jens Axboebb799ca2008-12-10 15:35:05 +01002038 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 if (!fs_bio_set)
2040 panic("bio: can't allocate bios\n");
2041
Martin K. Petersena91a2782011-03-17 11:11:05 +01002042 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2043 panic("bio: can't create integrity pool\n");
2044
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 return 0;
2046}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047subsys_initcall(init_bio);