blob: 006bcc52a77ebef4c0e792c7eb7c19564ce9482f [file] [log] [blame]
Christoph Hellwig8c165672019-04-30 14:42:39 -04001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Jens Axboe0fe23472006-09-04 15:41:16 +02003 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 */
5#include <linux/mm.h>
6#include <linux/swap.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -07009#include <linux/uio.h>
Tejun Heo852c7882012-03-05 13:15:27 -080010#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/slab.h>
12#include <linux/init.h>
13#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050014#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mempool.h>
16#include <linux/workqueue.h>
Tejun Heo852c7882012-03-05 13:15:27 -080017#include <linux/cgroup.h>
Josef Bacik08e18ea2018-07-03 11:14:50 -040018#include <linux/blk-cgroup.h>
Damien Le Moalb4c58752019-07-01 14:09:15 +090019#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Li Zefan55782132009-06-09 13:43:05 +080021#include <trace/events/block.h>
Shaohua Li9e234ee2017-03-27 10:51:41 -070022#include "blk.h"
Josef Bacik67b42d02018-07-03 11:15:00 -040023#include "blk-rq-qos.h"
Ingo Molnar0bfc2452008-11-26 11:59:56 +010024
Jens Axboe392ddc32008-12-23 12:42:54 +010025/*
26 * Test patch to inline a certain number of bi_io_vec's inside the bio
27 * itself, to shrink a bio data allocation from two mempool calls to one
28 */
29#define BIO_INLINE_VECS 4
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031/*
32 * if you change this list, also change bvec_alloc or things will
33 * break badly! cannot be bigger than what you can fit into an
34 * unsigned short
35 */
Mikulas Patockabd5c4fa2018-03-21 12:49:29 -040036#define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
Christoph Hellwiged996a52016-07-19 11:28:42 +020037static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
Mikulas Patockabd5c4fa2018-03-21 12:49:29 -040038 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 -070039};
40#undef BV
41
42/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
44 * IO code that does not need private memory pools.
45 */
Kent Overstreetf4f81542018-05-08 21:33:52 -040046struct bio_set fs_bio_set;
Kent Overstreet3f86a822012-09-06 15:35:01 -070047EXPORT_SYMBOL(fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Jens Axboebb799ca2008-12-10 15:35:05 +010049/*
50 * Our slab pool management
51 */
52struct bio_slab {
53 struct kmem_cache *slab;
54 unsigned int slab_ref;
55 unsigned int slab_size;
56 char name[8];
57};
58static DEFINE_MUTEX(bio_slab_lock);
59static struct bio_slab *bio_slabs;
60static unsigned int bio_slab_nr, bio_slab_max;
61
62static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
63{
64 unsigned int sz = sizeof(struct bio) + extra_size;
65 struct kmem_cache *slab = NULL;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020066 struct bio_slab *bslab, *new_bio_slabs;
Anna Leuschner386bc352012-10-22 21:53:36 +020067 unsigned int new_bio_slab_max;
Jens Axboebb799ca2008-12-10 15:35:05 +010068 unsigned int i, entry = -1;
69
70 mutex_lock(&bio_slab_lock);
71
72 i = 0;
73 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010074 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010075
76 if (!bslab->slab && entry == -1)
77 entry = i;
78 else if (bslab->slab_size == sz) {
79 slab = bslab->slab;
80 bslab->slab_ref++;
81 break;
82 }
83 i++;
84 }
85
86 if (slab)
87 goto out_unlock;
88
89 if (bio_slab_nr == bio_slab_max && entry == -1) {
Anna Leuschner386bc352012-10-22 21:53:36 +020090 new_bio_slab_max = bio_slab_max << 1;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020091 new_bio_slabs = krealloc(bio_slabs,
Anna Leuschner386bc352012-10-22 21:53:36 +020092 new_bio_slab_max * sizeof(struct bio_slab),
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020093 GFP_KERNEL);
94 if (!new_bio_slabs)
Jens Axboebb799ca2008-12-10 15:35:05 +010095 goto out_unlock;
Anna Leuschner386bc352012-10-22 21:53:36 +020096 bio_slab_max = new_bio_slab_max;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020097 bio_slabs = new_bio_slabs;
Jens Axboebb799ca2008-12-10 15:35:05 +010098 }
99 if (entry == -1)
100 entry = bio_slab_nr++;
101
102 bslab = &bio_slabs[entry];
103
104 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
Mikulas Patocka6a241482014-03-28 15:51:55 -0400105 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
106 SLAB_HWCACHE_ALIGN, NULL);
Jens Axboebb799ca2008-12-10 15:35:05 +0100107 if (!slab)
108 goto out_unlock;
109
Jens Axboebb799ca2008-12-10 15:35:05 +0100110 bslab->slab = slab;
111 bslab->slab_ref = 1;
112 bslab->slab_size = sz;
113out_unlock:
114 mutex_unlock(&bio_slab_lock);
115 return slab;
116}
117
118static void bio_put_slab(struct bio_set *bs)
119{
120 struct bio_slab *bslab = NULL;
121 unsigned int i;
122
123 mutex_lock(&bio_slab_lock);
124
125 for (i = 0; i < bio_slab_nr; i++) {
126 if (bs->bio_slab == bio_slabs[i].slab) {
127 bslab = &bio_slabs[i];
128 break;
129 }
130 }
131
132 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
133 goto out;
134
135 WARN_ON(!bslab->slab_ref);
136
137 if (--bslab->slab_ref)
138 goto out;
139
140 kmem_cache_destroy(bslab->slab);
141 bslab->slab = NULL;
142
143out:
144 mutex_unlock(&bio_slab_lock);
145}
146
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200147unsigned int bvec_nr_vecs(unsigned short idx)
148{
Greg Edwardsd6c02a92018-08-08 13:27:53 -0600149 return bvec_slabs[--idx].nr_vecs;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200150}
151
Kent Overstreet9f060e22012-10-12 15:29:33 -0700152void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
Jens Axboebb799ca2008-12-10 15:35:05 +0100153{
Christoph Hellwiged996a52016-07-19 11:28:42 +0200154 if (!idx)
155 return;
156 idx--;
Jens Axboebb799ca2008-12-10 15:35:05 +0100157
Christoph Hellwiged996a52016-07-19 11:28:42 +0200158 BIO_BUG_ON(idx >= BVEC_POOL_NR);
159
160 if (idx == BVEC_POOL_MAX) {
Kent Overstreet9f060e22012-10-12 15:29:33 -0700161 mempool_free(bv, pool);
Christoph Hellwiged996a52016-07-19 11:28:42 +0200162 } else {
Jens Axboebb799ca2008-12-10 15:35:05 +0100163 struct biovec_slab *bvs = bvec_slabs + idx;
164
165 kmem_cache_free(bvs->slab, bv);
166 }
167}
168
Kent Overstreet9f060e22012-10-12 15:29:33 -0700169struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
170 mempool_t *pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
172 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100175 * see comment near bvec_array define!
176 */
177 switch (nr) {
178 case 1:
179 *idx = 0;
180 break;
181 case 2 ... 4:
182 *idx = 1;
183 break;
184 case 5 ... 16:
185 *idx = 2;
186 break;
187 case 17 ... 64:
188 *idx = 3;
189 break;
190 case 65 ... 128:
191 *idx = 4;
192 break;
193 case 129 ... BIO_MAX_PAGES:
194 *idx = 5;
195 break;
196 default:
197 return NULL;
198 }
199
200 /*
201 * idx now points to the pool we want to allocate from. only the
202 * 1-vec entry pool is mempool backed.
203 */
Christoph Hellwiged996a52016-07-19 11:28:42 +0200204 if (*idx == BVEC_POOL_MAX) {
Jens Axboe7ff93452008-12-11 11:53:43 +0100205fallback:
Kent Overstreet9f060e22012-10-12 15:29:33 -0700206 bvl = mempool_alloc(pool, gfp_mask);
Jens Axboe7ff93452008-12-11 11:53:43 +0100207 } else {
208 struct biovec_slab *bvs = bvec_slabs + *idx;
Mel Gormand0164ad2015-11-06 16:28:21 -0800209 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200211 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100212 * Make this allocation restricted and don't dump info on
213 * allocation failures, since we'll fallback to the mempool
214 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200215 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100216 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
217
218 /*
Mel Gormand0164ad2015-11-06 16:28:21 -0800219 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
Jens Axboe7ff93452008-12-11 11:53:43 +0100220 * is set, retry with the 1-entry mempool
221 */
222 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
Mel Gormand0164ad2015-11-06 16:28:21 -0800223 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
Christoph Hellwiged996a52016-07-19 11:28:42 +0200224 *idx = BVEC_POOL_MAX;
Jens Axboe7ff93452008-12-11 11:53:43 +0100225 goto fallback;
226 }
227 }
228
Christoph Hellwiged996a52016-07-19 11:28:42 +0200229 (*idx)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return bvl;
231}
232
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600233void bio_uninit(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
Dennis Zhou6f70fb62018-12-05 12:10:37 -0500235 bio_disassociate_blkg(bio);
Justin Teeece841a2019-12-05 10:09:01 +0800236
237 if (bio_integrity(bio))
238 bio_integrity_free(bio);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700239}
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600240EXPORT_SYMBOL(bio_uninit);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200241
Kent Overstreet4254bba2012-09-06 15:35:00 -0700242static void bio_free(struct bio *bio)
243{
244 struct bio_set *bs = bio->bi_pool;
245 void *p;
246
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600247 bio_uninit(bio);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700248
249 if (bs) {
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400250 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
Kent Overstreet4254bba2012-09-06 15:35:00 -0700251
252 /*
253 * If we have front padding, adjust the bio pointer before freeing
254 */
255 p = bio;
Jens Axboebb799ca2008-12-10 15:35:05 +0100256 p -= bs->front_pad;
257
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400258 mempool_free(p, &bs->bio_pool);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700259 } else {
260 /* Bio was allocated by bio_kmalloc() */
261 kfree(bio);
262 }
Peter Osterlund36763472005-09-06 15:16:42 -0700263}
264
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600265/*
266 * Users of this function have their own bio allocation. Subsequently,
267 * they must remember to pair any call to bio_init() with bio_uninit()
268 * when IO has completed, or when the bio is released.
269 */
Ming Lei3a83f462016-11-22 08:57:21 -0700270void bio_init(struct bio *bio, struct bio_vec *table,
271 unsigned short max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Jens Axboe2b94de52007-07-18 13:14:03 +0200273 memset(bio, 0, sizeof(*bio));
Jens Axboec4cf5262015-04-17 16:15:18 -0600274 atomic_set(&bio->__bi_remaining, 1);
Jens Axboedac56212015-04-17 16:23:59 -0600275 atomic_set(&bio->__bi_cnt, 1);
Ming Lei3a83f462016-11-22 08:57:21 -0700276
277 bio->bi_io_vec = table;
278 bio->bi_max_vecs = max_vecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200280EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282/**
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700283 * bio_reset - reinitialize a bio
284 * @bio: bio to reset
285 *
286 * Description:
287 * After calling bio_reset(), @bio will be in the same state as a freshly
288 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
289 * preserved are the ones that are initialized by bio_alloc_bioset(). See
290 * comment in struct bio.
291 */
292void bio_reset(struct bio *bio)
293{
294 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
295
Jens Axboe9ae3b3f52017-06-28 15:30:13 -0600296 bio_uninit(bio);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700297
298 memset(bio, 0, BIO_RESET_BYTES);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200299 bio->bi_flags = flags;
Jens Axboec4cf5262015-04-17 16:15:18 -0600300 atomic_set(&bio->__bi_remaining, 1);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700301}
302EXPORT_SYMBOL(bio_reset);
303
Christoph Hellwig38f8baa2016-03-11 17:34:51 +0100304static struct bio *__bio_chain_endio(struct bio *bio)
Kent Overstreet196d38bc2013-11-23 18:34:15 -0800305{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200306 struct bio *parent = bio->bi_private;
307
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200308 if (!parent->bi_status)
309 parent->bi_status = bio->bi_status;
Kent Overstreet196d38bc2013-11-23 18:34:15 -0800310 bio_put(bio);
Christoph Hellwig38f8baa2016-03-11 17:34:51 +0100311 return parent;
312}
313
314static void bio_chain_endio(struct bio *bio)
315{
316 bio_endio(__bio_chain_endio(bio));
Kent Overstreet196d38bc2013-11-23 18:34:15 -0800317}
318
319/**
320 * bio_chain - chain bio completions
Randy Dunlap1051a902014-04-20 16:03:31 -0700321 * @bio: the target bio
322 * @parent: the @bio's parent bio
Kent Overstreet196d38bc2013-11-23 18:34:15 -0800323 *
324 * The caller won't have a bi_end_io called when @bio completes - instead,
325 * @parent's bi_end_io won't be called until both @parent and @bio have
326 * completed; the chained bio will also be freed when it completes.
327 *
328 * The caller must not set bi_private or bi_end_io in @bio.
329 */
330void bio_chain(struct bio *bio, struct bio *parent)
331{
332 BUG_ON(bio->bi_private || bio->bi_end_io);
333
334 bio->bi_private = parent;
335 bio->bi_end_io = bio_chain_endio;
Jens Axboec4cf5262015-04-17 16:15:18 -0600336 bio_inc_remaining(parent);
Kent Overstreet196d38bc2013-11-23 18:34:15 -0800337}
338EXPORT_SYMBOL(bio_chain);
339
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700340static void bio_alloc_rescue(struct work_struct *work)
341{
342 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
343 struct bio *bio;
344
345 while (1) {
346 spin_lock(&bs->rescue_lock);
347 bio = bio_list_pop(&bs->rescue_list);
348 spin_unlock(&bs->rescue_lock);
349
350 if (!bio)
351 break;
352
353 generic_make_request(bio);
354 }
355}
356
357static void punt_bios_to_rescuer(struct bio_set *bs)
358{
359 struct bio_list punt, nopunt;
360 struct bio *bio;
361
NeilBrown47e0fb42017-06-18 14:38:57 +1000362 if (WARN_ON_ONCE(!bs->rescue_workqueue))
363 return;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700364 /*
365 * In order to guarantee forward progress we must punt only bios that
366 * were allocated from this bio_set; otherwise, if there was a bio on
367 * there for a stacking driver higher up in the stack, processing it
368 * could require allocating bios from this bio_set, and doing that from
369 * our own rescuer would be bad.
370 *
371 * Since bio lists are singly linked, pop them all instead of trying to
372 * remove from the middle of the list:
373 */
374
375 bio_list_init(&punt);
376 bio_list_init(&nopunt);
377
NeilBrownf5fe1b52017-03-10 17:00:47 +1100378 while ((bio = bio_list_pop(&current->bio_list[0])))
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700379 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
NeilBrownf5fe1b52017-03-10 17:00:47 +1100380 current->bio_list[0] = nopunt;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700381
NeilBrownf5fe1b52017-03-10 17:00:47 +1100382 bio_list_init(&nopunt);
383 while ((bio = bio_list_pop(&current->bio_list[1])))
384 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
385 current->bio_list[1] = nopunt;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700386
387 spin_lock(&bs->rescue_lock);
388 bio_list_merge(&bs->rescue_list, &punt);
389 spin_unlock(&bs->rescue_lock);
390
391 queue_work(bs->rescue_workqueue, &bs->rescue_work);
392}
393
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700394/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 * bio_alloc_bioset - allocate a bio for I/O
Randy Dunlap519c8e92017-10-16 11:01:00 -0700396 * @gfp_mask: the GFP_* mask given to the slab allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200398 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 *
400 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700401 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
402 * backed by the @bs's mempool.
403 *
Mel Gormand0164ad2015-11-06 16:28:21 -0800404 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
405 * always be able to allocate a bio. This is due to the mempool guarantees.
406 * To make this work, callers must never allocate more than 1 bio at a time
407 * from this pool. Callers that need to allocate more than 1 bio must always
408 * submit the previously allocated bio for IO before attempting to allocate
409 * a new one. Failure to do so can cause deadlocks under memory pressure.
Kent Overstreet3f86a822012-09-06 15:35:01 -0700410 *
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700411 * Note that when running under generic_make_request() (i.e. any block
412 * driver), bios are not submitted until after you return - see the code in
413 * generic_make_request() that converts recursion into iteration, to prevent
414 * stack overflows.
415 *
416 * This would normally mean allocating multiple bios under
417 * generic_make_request() would be susceptible to deadlocks, but we have
418 * deadlock avoidance code that resubmits any blocked bios from a rescuer
419 * thread.
420 *
421 * However, we do not guarantee forward progress for allocations from other
422 * mempools. Doing multiple allocations from the same mempool under
423 * generic_make_request() should be avoided - instead, use bio_set's front_pad
424 * for per bio allocations.
425 *
Kent Overstreet3f86a822012-09-06 15:35:01 -0700426 * RETURNS:
427 * Pointer to new bio on success, NULL on failure.
428 */
Dan Carpenter7a88fa12017-03-23 13:24:55 +0300429struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
430 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700432 gfp_t saved_gfp = gfp_mask;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700433 unsigned front_pad;
434 unsigned inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100435 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200436 struct bio *bio;
437 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200438
Kent Overstreet3f86a822012-09-06 15:35:01 -0700439 if (!bs) {
440 if (nr_iovecs > UIO_MAXIOV)
441 return NULL;
442
443 p = kmalloc(sizeof(struct bio) +
444 nr_iovecs * sizeof(struct bio_vec),
445 gfp_mask);
446 front_pad = 0;
447 inline_vecs = nr_iovecs;
448 } else {
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400449 /* should not use nobvec bioset for nr_iovecs > 0 */
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400450 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
451 nr_iovecs > 0))
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400452 return NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700453 /*
454 * generic_make_request() converts recursion to iteration; this
455 * means if we're running beneath it, any bios we allocate and
456 * submit will not be submitted (and thus freed) until after we
457 * return.
458 *
459 * This exposes us to a potential deadlock if we allocate
460 * multiple bios from the same bio_set() while running
461 * underneath generic_make_request(). If we were to allocate
462 * multiple bios (say a stacking block driver that was splitting
463 * bios), we would deadlock if we exhausted the mempool's
464 * reserve.
465 *
466 * We solve this, and guarantee forward progress, with a rescuer
467 * workqueue per bio_set. If we go to allocate and there are
468 * bios on current->bio_list, we first try the allocation
Mel Gormand0164ad2015-11-06 16:28:21 -0800469 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
470 * bios we would be blocking to the rescuer workqueue before
471 * we retry with the original gfp_flags.
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700472 */
473
NeilBrownf5fe1b52017-03-10 17:00:47 +1100474 if (current->bio_list &&
475 (!bio_list_empty(&current->bio_list[0]) ||
NeilBrown47e0fb42017-06-18 14:38:57 +1000476 !bio_list_empty(&current->bio_list[1])) &&
477 bs->rescue_workqueue)
Mel Gormand0164ad2015-11-06 16:28:21 -0800478 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700479
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400480 p = mempool_alloc(&bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700481 if (!p && gfp_mask != saved_gfp) {
482 punt_bios_to_rescuer(bs);
483 gfp_mask = saved_gfp;
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400484 p = mempool_alloc(&bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700485 }
486
Kent Overstreet3f86a822012-09-06 15:35:01 -0700487 front_pad = bs->front_pad;
488 inline_vecs = BIO_INLINE_VECS;
489 }
490
Tejun Heo451a9eb2009-04-15 19:50:51 +0200491 if (unlikely(!p))
492 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100493
Kent Overstreet3f86a822012-09-06 15:35:01 -0700494 bio = p + front_pad;
Ming Lei3a83f462016-11-22 08:57:21 -0700495 bio_init(bio, NULL, 0);
Ingo Molnar34053972009-02-21 11:16:36 +0100496
Kent Overstreet3f86a822012-09-06 15:35:01 -0700497 if (nr_iovecs > inline_vecs) {
Christoph Hellwiged996a52016-07-19 11:28:42 +0200498 unsigned long idx = 0;
499
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400500 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700501 if (!bvl && gfp_mask != saved_gfp) {
502 punt_bios_to_rescuer(bs);
503 gfp_mask = saved_gfp;
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400504 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700505 }
506
Ingo Molnar34053972009-02-21 11:16:36 +0100507 if (unlikely(!bvl))
508 goto err_free;
Kent Overstreeta38352e2012-05-25 13:03:11 -0700509
Christoph Hellwiged996a52016-07-19 11:28:42 +0200510 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700511 } else if (nr_iovecs) {
512 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100513 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700514
515 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100516 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100517 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100519
520err_free:
Kent Overstreet8aa6ba22018-05-08 21:33:50 -0400521 mempool_free(p, &bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100522 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200524EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Kent Overstreet38a72da2018-05-08 21:33:53 -0400526void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527{
528 unsigned long flags;
Kent Overstreet79886132013-11-23 17:19:00 -0800529 struct bio_vec bv;
530 struct bvec_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Kent Overstreet38a72da2018-05-08 21:33:53 -0400532 __bio_for_each_segment(bv, bio, iter, start) {
Kent Overstreet79886132013-11-23 17:19:00 -0800533 char *data = bvec_kmap_irq(&bv, &flags);
534 memset(data, 0, bv.bv_len);
535 flush_dcache_page(bv.bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 bvec_kunmap_irq(data, &flags);
537 }
538}
Kent Overstreet38a72da2018-05-08 21:33:53 -0400539EXPORT_SYMBOL(zero_fill_bio_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Ming Lei85a8ce62019-12-28 07:05:48 +0800541void bio_truncate(struct bio *bio, unsigned new_size)
542{
543 struct bio_vec bv;
544 struct bvec_iter iter;
545 unsigned int done = 0;
546 bool truncated = false;
547
548 if (new_size >= bio->bi_iter.bi_size)
549 return;
550
551 if (bio_data_dir(bio) != READ)
552 goto exit;
553
554 bio_for_each_segment(bv, bio, iter) {
555 if (done + bv.bv_len > new_size) {
556 unsigned offset;
557
558 if (!truncated)
559 offset = new_size - done;
560 else
561 offset = 0;
562 zero_user(bv.bv_page, offset, bv.bv_len - offset);
563 truncated = true;
564 }
565 done += bv.bv_len;
566 }
567
568 exit:
569 /*
570 * Don't touch bvec table here and make it really immutable, since
571 * fs bio user has to retrieve all pages via bio_for_each_segment_all
572 * in its .end_bio() callback.
573 *
574 * It is enough to truncate bio by updating .bi_size since we can make
575 * correct bvec with the updated .bi_size for drivers.
576 */
577 bio->bi_iter.bi_size = new_size;
578}
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/**
581 * bio_put - release a reference to a bio
582 * @bio: bio to release reference to
583 *
584 * Description:
585 * Put a reference to a &struct bio, either one you have gotten with
NeilBrown9b10f6a2017-06-18 14:38:59 +1000586 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 **/
588void bio_put(struct bio *bio)
589{
Jens Axboedac56212015-04-17 16:23:59 -0600590 if (!bio_flagged(bio, BIO_REFFED))
Kent Overstreet4254bba2012-09-06 15:35:00 -0700591 bio_free(bio);
Jens Axboedac56212015-04-17 16:23:59 -0600592 else {
593 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
594
595 /*
596 * last put frees it
597 */
598 if (atomic_dec_and_test(&bio->__bi_cnt))
599 bio_free(bio);
600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200602EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800605 * __bio_clone_fast - clone a bio that shares the original bio's biovec
606 * @bio: destination bio
607 * @bio_src: bio to clone
608 *
609 * Clone a &bio. Caller will own the returned bio, but not
610 * the actual data it points to. Reference count of returned
611 * bio will be one.
612 *
613 * Caller must ensure that @bio_src is not freed before @bio.
614 */
615void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
616{
Christoph Hellwiged996a52016-07-19 11:28:42 +0200617 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
Kent Overstreet59d276f2013-11-23 18:19:27 -0800618
619 /*
Christoph Hellwig74d46992017-08-23 19:10:32 +0200620 * most users will be overriding ->bi_disk with a new target,
Kent Overstreet59d276f2013-11-23 18:19:27 -0800621 * so we don't set nor calculate new physical/hw segment counts here
622 */
Christoph Hellwig74d46992017-08-23 19:10:32 +0200623 bio->bi_disk = bio_src->bi_disk;
Michael Lyle62530ed2017-11-16 23:47:25 -0800624 bio->bi_partno = bio_src->bi_partno;
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600625 bio_set_flag(bio, BIO_CLONED);
Shaohua Li111be882017-12-20 11:10:17 -0700626 if (bio_flagged(bio_src, BIO_THROTTLED))
627 bio_set_flag(bio, BIO_THROTTLED);
Jens Axboe1eff9d32016-08-05 15:35:16 -0600628 bio->bi_opf = bio_src->bi_opf;
Hannes Reineckeca474b72018-11-12 10:35:25 -0700629 bio->bi_ioprio = bio_src->bi_ioprio;
Jens Axboecb6934f2017-06-27 09:22:02 -0600630 bio->bi_write_hint = bio_src->bi_write_hint;
Kent Overstreet59d276f2013-11-23 18:19:27 -0800631 bio->bi_iter = bio_src->bi_iter;
632 bio->bi_io_vec = bio_src->bi_io_vec;
Paolo Valente20bd7232016-07-27 07:22:05 +0200633
Dennis Zhoudb6638d2018-12-05 12:10:35 -0500634 bio_clone_blkg_association(bio, bio_src);
Dennis Zhoue439bed2018-12-05 12:10:32 -0500635 blkcg_bio_issue_init(bio);
Kent Overstreet59d276f2013-11-23 18:19:27 -0800636}
637EXPORT_SYMBOL(__bio_clone_fast);
638
639/**
640 * bio_clone_fast - clone a bio that shares the original bio's biovec
641 * @bio: bio to clone
642 * @gfp_mask: allocation priority
643 * @bs: bio_set to allocate from
644 *
645 * Like __bio_clone_fast, only also allocates the returned bio
646 */
647struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
648{
649 struct bio *b;
650
651 b = bio_alloc_bioset(gfp_mask, 0, bs);
652 if (!b)
653 return NULL;
654
655 __bio_clone_fast(b, bio);
656
657 if (bio_integrity(bio)) {
658 int ret;
659
660 ret = bio_integrity_clone(b, bio, gfp_mask);
661
662 if (ret < 0) {
663 bio_put(b);
664 return NULL;
665 }
666 }
667
668 return b;
669}
670EXPORT_SYMBOL(bio_clone_fast);
671
Ming Lei59194822019-03-17 18:01:07 +0800672static inline bool page_is_mergeable(const struct bio_vec *bv,
673 struct page *page, unsigned int len, unsigned int off,
Christoph Hellwigff896732019-06-17 11:14:11 +0200674 bool *same_page)
Ming Lei59194822019-03-17 18:01:07 +0800675{
676 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) +
677 bv->bv_offset + bv->bv_len - 1;
678 phys_addr_t page_addr = page_to_phys(page);
679
680 if (vec_end_addr + 1 != page_addr + off)
681 return false;
682 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
683 return false;
Christoph Hellwig52d52d12019-04-11 08:23:31 +0200684
Christoph Hellwigff896732019-06-17 11:14:11 +0200685 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
686 if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page)
687 return false;
Ming Lei59194822019-03-17 18:01:07 +0800688 return true;
689}
690
Christoph Hellwig384209c2019-08-12 17:39:57 +0200691static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio,
692 struct page *page, unsigned len, unsigned offset,
693 bool *same_page)
Ming Lei489fbbc2019-03-29 15:08:00 +0800694{
Christoph Hellwig384209c2019-08-12 17:39:57 +0200695 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
Ming Lei489fbbc2019-03-29 15:08:00 +0800696 unsigned long mask = queue_segment_boundary(q);
697 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
698 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
699
700 if ((addr1 | mask) != (addr2 | mask))
701 return false;
Ming Lei489fbbc2019-03-29 15:08:00 +0800702 if (bv->bv_len + len > queue_max_segment_size(q))
703 return false;
Christoph Hellwig384209c2019-08-12 17:39:57 +0200704 return __bio_try_merge_page(bio, page, len, offset, same_page);
Ming Lei489fbbc2019-03-29 15:08:00 +0800705}
706
Shaohua Lif4595872017-03-24 10:34:43 -0700707/**
Ming Lei19047082019-03-17 18:01:08 +0800708 * __bio_add_pc_page - attempt to add page to passthrough bio
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800709 * @q: the target queue
710 * @bio: destination bio
711 * @page: page to add
712 * @len: vec entry length
713 * @offset: vec entry offset
Christoph Hellwigd1916c82019-08-12 17:39:58 +0200714 * @same_page: return if the merge happen inside the same page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800716 * Attempt to add a page to the bio_vec maplist. This can fail for a
717 * number of reasons, such as the bio being full or target block device
718 * limitations. The target block device must allow bio's up to PAGE_SIZE,
719 * so it is always possible to add a single page to an empty bio.
720 *
Ming Lei5a8ce242019-03-17 18:01:06 +0800721 * This should only be used by passthrough bios.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 */
Christoph Hellwig47138392019-04-25 09:04:33 +0200723static int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
Ming Lei19047082019-03-17 18:01:08 +0800724 struct page *page, unsigned int len, unsigned int offset,
Christoph Hellwigd1916c82019-08-12 17:39:58 +0200725 bool *same_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 struct bio_vec *bvec;
728
729 /*
730 * cloned bio must not modify vec list
731 */
732 if (unlikely(bio_flagged(bio, BIO_CLONED)))
733 return 0;
734
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800735 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 return 0;
737
Jens Axboe80cfd542006-01-06 09:43:28 +0100738 if (bio->bi_vcnt > 0) {
Christoph Hellwigd1916c82019-08-12 17:39:58 +0200739 if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page))
Christoph Hellwig384209c2019-08-12 17:39:57 +0200740 return len;
Christoph Hellwig320ea862019-08-12 17:39:56 +0200741
742 /*
743 * If the queue doesn't support SG gaps and adding this segment
744 * would create a gap, disallow it.
745 */
Christoph Hellwig384209c2019-08-12 17:39:57 +0200746 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
Christoph Hellwig320ea862019-08-12 17:39:56 +0200747 if (bvec_gap_to_prev(q, bvec, offset))
748 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100749 }
750
Ming Lei79d08f82019-07-01 15:14:46 +0800751 if (bio_full(bio, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 return 0;
753
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200754 if (bio->bi_vcnt >= queue_max_segments(q))
Ming Lei489fbbc2019-03-29 15:08:00 +0800755 return 0;
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 bvec = &bio->bi_io_vec[bio->bi_vcnt];
758 bvec->bv_page = page;
759 bvec->bv_len = len;
760 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800761 bio->bi_vcnt++;
Christoph Hellwigdcdca752019-04-25 09:04:35 +0200762 bio->bi_iter.bi_size += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return len;
764}
Ming Lei19047082019-03-17 18:01:08 +0800765
766int bio_add_pc_page(struct request_queue *q, struct bio *bio,
767 struct page *page, unsigned int len, unsigned int offset)
768{
Christoph Hellwigd1916c82019-08-12 17:39:58 +0200769 bool same_page = false;
770 return __bio_add_pc_page(q, bio, page, len, offset, &same_page);
Ming Lei19047082019-03-17 18:01:08 +0800771}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200772EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600773
774/**
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700775 * __bio_try_merge_page - try appending data to an existing bvec.
776 * @bio: destination bio
Ming Lei551879a2019-04-23 10:51:04 +0800777 * @page: start page to add
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700778 * @len: length of the data to add
Ming Lei551879a2019-04-23 10:51:04 +0800779 * @off: offset of the data relative to @page
Christoph Hellwigff896732019-06-17 11:14:11 +0200780 * @same_page: return if the segment has been merged inside the same page
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700781 *
782 * Try to add the data at @page + @off to the last bvec of @bio. This is a
783 * a useful optimisation for file systems with a block size smaller than the
784 * page size.
785 *
Ming Lei551879a2019-04-23 10:51:04 +0800786 * Warn if (@len, @off) crosses pages in case that @same_page is true.
787 *
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700788 * Return %true on success or %false on failure.
789 */
790bool __bio_try_merge_page(struct bio *bio, struct page *page,
Christoph Hellwigff896732019-06-17 11:14:11 +0200791 unsigned int len, unsigned int off, bool *same_page)
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700792{
793 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
794 return false;
795
Andreas Gruenbachercc90bc62019-12-09 20:11:14 +0100796 if (bio->bi_vcnt > 0) {
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700797 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
798
Ming Lei59194822019-03-17 18:01:07 +0800799 if (page_is_mergeable(bv, page, len, off, same_page)) {
Andreas Gruenbachercc90bc62019-12-09 20:11:14 +0100800 if (bio->bi_iter.bi_size > UINT_MAX - len)
801 return false;
Ming Lei59194822019-03-17 18:01:07 +0800802 bv->bv_len += len;
803 bio->bi_iter.bi_size += len;
804 return true;
805 }
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700806 }
807 return false;
808}
809EXPORT_SYMBOL_GPL(__bio_try_merge_page);
810
811/**
Ming Lei551879a2019-04-23 10:51:04 +0800812 * __bio_add_page - add page(s) to a bio in a new segment
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700813 * @bio: destination bio
Ming Lei551879a2019-04-23 10:51:04 +0800814 * @page: start page to add
815 * @len: length of the data to add, may cross pages
816 * @off: offset of the data relative to @page, may cross pages
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700817 *
818 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
819 * that @bio has space for another bvec.
820 */
821void __bio_add_page(struct bio *bio, struct page *page,
822 unsigned int len, unsigned int off)
823{
824 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
825
826 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
Ming Lei79d08f82019-07-01 15:14:46 +0800827 WARN_ON_ONCE(bio_full(bio, len));
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700828
829 bv->bv_page = page;
830 bv->bv_offset = off;
831 bv->bv_len = len;
832
833 bio->bi_iter.bi_size += len;
834 bio->bi_vcnt++;
Johannes Weinerb8e24a92019-08-08 15:03:00 -0400835
836 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
837 bio_set_flag(bio, BIO_WORKINGSET);
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700838}
839EXPORT_SYMBOL_GPL(__bio_add_page);
840
841/**
Ming Lei551879a2019-04-23 10:51:04 +0800842 * bio_add_page - attempt to add page(s) to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 * @bio: destination bio
Ming Lei551879a2019-04-23 10:51:04 +0800844 * @page: start page to add
845 * @len: vec entry length, may cross pages
846 * @offset: vec entry offset relative to @page, may cross pages
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 *
Ming Lei551879a2019-04-23 10:51:04 +0800848 * Attempt to add page(s) to the bio_vec maplist. This will only fail
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800849 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800851int bio_add_page(struct bio *bio, struct page *page,
852 unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853{
Christoph Hellwigff896732019-06-17 11:14:11 +0200854 bool same_page = false;
855
856 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
Ming Lei79d08f82019-07-01 15:14:46 +0800857 if (bio_full(bio, len))
Christoph Hellwig0aa69fd2018-06-01 09:03:05 -0700858 return 0;
859 __bio_add_page(bio, page, len, offset);
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800860 }
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800861 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200863EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Christoph Hellwigd241a952019-06-26 15:49:21 +0200865void bio_release_pages(struct bio *bio, bool mark_dirty)
Christoph Hellwig7321ecb2019-04-11 08:23:30 +0200866{
867 struct bvec_iter_all iter_all;
868 struct bio_vec *bvec;
Christoph Hellwig7321ecb2019-04-11 08:23:30 +0200869
Christoph Hellwigb2d0d992019-06-26 15:49:20 +0200870 if (bio_flagged(bio, BIO_NO_PAGE_REF))
871 return;
872
Christoph Hellwigd241a952019-06-26 15:49:21 +0200873 bio_for_each_segment_all(bvec, bio, iter_all) {
874 if (mark_dirty && !PageCompound(bvec->bv_page))
875 set_page_dirty_lock(bvec->bv_page);
Christoph Hellwig7321ecb2019-04-11 08:23:30 +0200876 put_page(bvec->bv_page);
Christoph Hellwigd241a952019-06-26 15:49:21 +0200877 }
Christoph Hellwig7321ecb2019-04-11 08:23:30 +0200878}
879
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700880static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
881{
882 const struct bio_vec *bv = iter->bvec;
883 unsigned int len;
884 size_t size;
885
886 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len))
887 return -EINVAL;
888
889 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count);
890 size = bio_add_page(bio, bv->bv_page, len,
891 bv->bv_offset + iter->iov_offset);
Christoph Hellwiga10584c2019-04-11 08:23:28 +0200892 if (unlikely(size != len))
893 return -EINVAL;
Christoph Hellwiga10584c2019-04-11 08:23:28 +0200894 iov_iter_advance(iter, size);
895 return 0;
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700896}
897
Christoph Hellwig576ed912018-09-20 08:28:21 +0200898#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
899
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600900/**
Martin Wilck17d51b12018-07-25 23:15:09 +0200901 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600902 * @bio: bio to add pages to
903 * @iter: iov iterator describing the region to be mapped
904 *
Martin Wilck17d51b12018-07-25 23:15:09 +0200905 * Pins pages from *iter and appends them to @bio's bvec array. The
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600906 * pages will have to be released using put_page() when done.
Martin Wilck17d51b12018-07-25 23:15:09 +0200907 * For multi-segment *iter, this function only adds pages from the
908 * the next non-empty segment of the iov iterator.
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600909 */
Martin Wilck17d51b12018-07-25 23:15:09 +0200910static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600911{
Christoph Hellwig576ed912018-09-20 08:28:21 +0200912 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
913 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600914 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
915 struct page **pages = (struct page **)bv;
Christoph Hellwig45691802019-06-17 11:14:12 +0200916 bool same_page = false;
Christoph Hellwig576ed912018-09-20 08:28:21 +0200917 ssize_t size, left;
918 unsigned len, i;
Martin Wilckb403ea22018-07-25 23:15:07 +0200919 size_t offset;
Christoph Hellwig576ed912018-09-20 08:28:21 +0200920
921 /*
922 * Move page array up in the allocated memory for the bio vecs as far as
923 * possible so that we can start filling biovecs from the beginning
924 * without overwriting the temporary page array.
925 */
926 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
927 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600928
929 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
930 if (unlikely(size <= 0))
931 return size ? size : -EFAULT;
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600932
Christoph Hellwig576ed912018-09-20 08:28:21 +0200933 for (left = size, i = 0; left > 0; left -= len, i++) {
934 struct page *page = pages[i];
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600935
Christoph Hellwig576ed912018-09-20 08:28:21 +0200936 len = min_t(size_t, PAGE_SIZE - offset, left);
Christoph Hellwig45691802019-06-17 11:14:12 +0200937
938 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
939 if (same_page)
940 put_page(page);
941 } else {
Ming Lei79d08f82019-07-01 15:14:46 +0800942 if (WARN_ON_ONCE(bio_full(bio, len)))
Christoph Hellwig45691802019-06-17 11:14:12 +0200943 return -EINVAL;
944 __bio_add_page(bio, page, len, offset);
945 }
Christoph Hellwig576ed912018-09-20 08:28:21 +0200946 offset = 0;
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600947 }
948
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600949 iov_iter_advance(iter, size);
950 return 0;
951}
Martin Wilck17d51b12018-07-25 23:15:09 +0200952
953/**
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700954 * bio_iov_iter_get_pages - add user or kernel pages to a bio
Martin Wilck17d51b12018-07-25 23:15:09 +0200955 * @bio: bio to add pages to
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700956 * @iter: iov iterator describing the region to be added
Martin Wilck17d51b12018-07-25 23:15:09 +0200957 *
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700958 * This takes either an iterator pointing to user memory, or one pointing to
959 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
960 * map them into the kernel. On IO completion, the caller should put those
Jens Axboe399254a2019-02-27 13:13:23 -0700961 * pages. If we're adding kernel pages, and the caller told us it's safe to
962 * do so, we just have to add the pages to the bio directly. We don't grab an
963 * extra reference to those pages (the user should already have that), and we
964 * don't put the page on IO completion. The caller needs to check if the bio is
965 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
966 * released.
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700967 *
Martin Wilck17d51b12018-07-25 23:15:09 +0200968 * The function tries, but does not guarantee, to pin as many pages as
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700969 * fit into the bio, or are requested in *iter, whatever is smaller. If
970 * MM encounters an error pinning the requested pages, it stops. Error
971 * is returned only if 0 pages could be pinned.
Martin Wilck17d51b12018-07-25 23:15:09 +0200972 */
973int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
974{
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700975 const bool is_bvec = iov_iter_is_bvec(iter);
Christoph Hellwig14eacf12019-04-11 08:23:29 +0200976 int ret;
977
978 if (WARN_ON_ONCE(bio->bi_vcnt))
979 return -EINVAL;
Martin Wilck17d51b12018-07-25 23:15:09 +0200980
981 do {
Jens Axboe6d0c48a2018-11-30 08:47:03 -0700982 if (is_bvec)
983 ret = __bio_iov_bvec_add_pages(bio, iter);
984 else
985 ret = __bio_iov_iter_get_pages(bio, iter);
Ming Lei79d08f82019-07-01 15:14:46 +0800986 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
Martin Wilck17d51b12018-07-25 23:15:09 +0200987
Christoph Hellwigb6207432019-06-26 15:49:28 +0200988 if (is_bvec)
Christoph Hellwig7321ecb2019-04-11 08:23:30 +0200989 bio_set_flag(bio, BIO_NO_PAGE_REF);
Christoph Hellwig14eacf12019-04-11 08:23:29 +0200990 return bio->bi_vcnt ? 0 : ret;
Martin Wilck17d51b12018-07-25 23:15:09 +0200991}
Kent Overstreet2cefe4d2016-10-31 11:59:24 -0600992
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200993static void submit_bio_wait_endio(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700994{
Christoph Hellwig65e53aa2017-10-25 17:55:57 +0900995 complete(bio->bi_private);
Kent Overstreet9e882242012-09-10 14:41:12 -0700996}
997
998/**
999 * submit_bio_wait - submit a bio, and wait until it completes
Kent Overstreet9e882242012-09-10 14:41:12 -07001000 * @bio: The &struct bio which describes the I/O
1001 *
1002 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1003 * bio_endio() on failure.
Jan Kara3d289d62017-08-02 10:25:21 +02001004 *
1005 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1006 * result in bio reference to be consumed. The caller must drop the reference
1007 * on his own.
Kent Overstreet9e882242012-09-10 14:41:12 -07001008 */
Mike Christie4e49ea42016-06-05 14:31:41 -05001009int submit_bio_wait(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -07001010{
Byungchul Parke319e1f2017-10-25 17:56:05 +09001011 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
Kent Overstreet9e882242012-09-10 14:41:12 -07001012
Christoph Hellwig65e53aa2017-10-25 17:55:57 +09001013 bio->bi_private = &done;
Kent Overstreet9e882242012-09-10 14:41:12 -07001014 bio->bi_end_io = submit_bio_wait_endio;
Jens Axboe1eff9d32016-08-05 15:35:16 -06001015 bio->bi_opf |= REQ_SYNC;
Mike Christie4e49ea42016-06-05 14:31:41 -05001016 submit_bio(bio);
Christoph Hellwig65e53aa2017-10-25 17:55:57 +09001017 wait_for_completion_io(&done);
Kent Overstreet9e882242012-09-10 14:41:12 -07001018
Christoph Hellwig65e53aa2017-10-25 17:55:57 +09001019 return blk_status_to_errno(bio->bi_status);
Kent Overstreet9e882242012-09-10 14:41:12 -07001020}
1021EXPORT_SYMBOL(submit_bio_wait);
1022
Kent Overstreet054bdf62012-09-28 13:17:55 -07001023/**
1024 * bio_advance - increment/complete a bio by some number of bytes
1025 * @bio: bio to advance
1026 * @bytes: number of bytes to complete
1027 *
1028 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1029 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1030 * be updated on the last bvec as well.
1031 *
1032 * @bio will then represent the remaining, uncompleted portion of the io.
1033 */
1034void bio_advance(struct bio *bio, unsigned bytes)
1035{
1036 if (bio_integrity(bio))
1037 bio_integrity_advance(bio, bytes);
1038
Kent Overstreet4550dd62013-08-07 14:26:21 -07001039 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -07001040}
1041EXPORT_SYMBOL(bio_advance);
1042
Kent Overstreet45db54d2018-05-08 21:33:54 -04001043void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1044 struct bio *src, struct bvec_iter *src_iter)
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001045{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001046 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001047 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001048 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001049
Kent Overstreet45db54d2018-05-08 21:33:54 -04001050 while (src_iter->bi_size && dst_iter->bi_size) {
1051 src_bv = bio_iter_iovec(src, *src_iter);
1052 dst_bv = bio_iter_iovec(dst, *dst_iter);
1053
1054 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1055
1056 src_p = kmap_atomic(src_bv.bv_page);
1057 dst_p = kmap_atomic(dst_bv.bv_page);
1058
1059 memcpy(dst_p + dst_bv.bv_offset,
1060 src_p + src_bv.bv_offset,
1061 bytes);
1062
1063 kunmap_atomic(dst_p);
1064 kunmap_atomic(src_p);
1065
Kent Overstreet6e6e8112018-05-08 21:33:55 -04001066 flush_dcache_page(dst_bv.bv_page);
1067
Kent Overstreet45db54d2018-05-08 21:33:54 -04001068 bio_advance_iter(src, src_iter, bytes);
1069 bio_advance_iter(dst, dst_iter, bytes);
1070 }
1071}
1072EXPORT_SYMBOL(bio_copy_data_iter);
1073
1074/**
1075 * bio_copy_data - copy contents of data buffers from one bio to another
1076 * @src: source bio
1077 * @dst: destination bio
1078 *
1079 * Stops when it reaches the end of either @src or @dst - that is, copies
1080 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1081 */
1082void bio_copy_data(struct bio *dst, struct bio *src)
1083{
1084 struct bvec_iter src_iter = src->bi_iter;
1085 struct bvec_iter dst_iter = dst->bi_iter;
1086
1087 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1088}
1089EXPORT_SYMBOL(bio_copy_data);
1090
1091/**
1092 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1093 * another
1094 * @src: source bio list
1095 * @dst: destination bio list
1096 *
1097 * Stops when it reaches the end of either the @src list or @dst list - that is,
1098 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1099 * bios).
1100 */
1101void bio_list_copy_data(struct bio *dst, struct bio *src)
1102{
1103 struct bvec_iter src_iter = src->bi_iter;
1104 struct bvec_iter dst_iter = dst->bi_iter;
1105
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001106 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001107 if (!src_iter.bi_size) {
1108 src = src->bi_next;
1109 if (!src)
1110 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001111
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001112 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001113 }
1114
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001115 if (!dst_iter.bi_size) {
1116 dst = dst->bi_next;
1117 if (!dst)
1118 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001119
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001120 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001121 }
1122
Kent Overstreet45db54d2018-05-08 21:33:54 -04001123 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001124 }
1125}
Kent Overstreet45db54d2018-05-08 21:33:54 -04001126EXPORT_SYMBOL(bio_list_copy_data);
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001129 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001130 struct iov_iter iter;
1131 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132};
1133
Al Viro0e5b9352017-09-24 13:14:35 -04001134static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001135 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Al Viro0e5b9352017-09-24 13:14:35 -04001137 struct bio_map_data *bmd;
1138 if (data->nr_segs > UIO_MAXIOV)
Jens Axboef3f63c12010-10-29 11:46:56 -06001139 return NULL;
1140
Gustavo A. R. Silvaf1f8f292019-06-10 10:04:12 -05001141 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask);
Al Viro0e5b9352017-09-24 13:14:35 -04001142 if (!bmd)
1143 return NULL;
1144 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1145 bmd->iter = *data;
1146 bmd->iter.iov = bmd->iov;
1147 return bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148}
1149
Dongsu Park9124d3f2015-01-18 16:16:34 +01001150/**
1151 * bio_copy_from_iter - copy all pages from iov_iter to bio
1152 * @bio: The &struct bio which describes the I/O as destination
1153 * @iter: iov_iter as source
1154 *
1155 * Copy all pages from iov_iter to bio.
1156 * Returns 0 on success, or error on failure.
1157 */
Al Viro98a09d62017-09-24 12:14:36 -04001158static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001159{
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001160 struct bio_vec *bvec;
Ming Lei6dc4f102019-02-15 19:13:19 +08001161 struct bvec_iter_all iter_all;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001162
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02001163 bio_for_each_segment_all(bvec, bio, iter_all) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001164 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001165
Dongsu Park9124d3f2015-01-18 16:16:34 +01001166 ret = copy_page_from_iter(bvec->bv_page,
1167 bvec->bv_offset,
1168 bvec->bv_len,
Al Viro98a09d62017-09-24 12:14:36 -04001169 iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001170
Al Viro98a09d62017-09-24 12:14:36 -04001171 if (!iov_iter_count(iter))
Dongsu Park9124d3f2015-01-18 16:16:34 +01001172 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001173
Dongsu Park9124d3f2015-01-18 16:16:34 +01001174 if (ret < bvec->bv_len)
1175 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001176 }
1177
Dongsu Park9124d3f2015-01-18 16:16:34 +01001178 return 0;
1179}
1180
1181/**
1182 * bio_copy_to_iter - copy all pages from bio to iov_iter
1183 * @bio: The &struct bio which describes the I/O as source
1184 * @iter: iov_iter as destination
1185 *
1186 * Copy all pages from bio to iov_iter.
1187 * Returns 0 on success, or error on failure.
1188 */
1189static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1190{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001191 struct bio_vec *bvec;
Ming Lei6dc4f102019-02-15 19:13:19 +08001192 struct bvec_iter_all iter_all;
Dongsu Park9124d3f2015-01-18 16:16:34 +01001193
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02001194 bio_for_each_segment_all(bvec, bio, iter_all) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001195 ssize_t ret;
1196
1197 ret = copy_page_to_iter(bvec->bv_page,
1198 bvec->bv_offset,
1199 bvec->bv_len,
1200 &iter);
1201
1202 if (!iov_iter_count(&iter))
1203 break;
1204
1205 if (ret < bvec->bv_len)
1206 return -EFAULT;
1207 }
1208
1209 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001210}
1211
Guoqing Jiang491221f2016-09-22 03:10:01 -04001212void bio_free_pages(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001213{
1214 struct bio_vec *bvec;
Ming Lei6dc4f102019-02-15 19:13:19 +08001215 struct bvec_iter_all iter_all;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001216
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02001217 bio_for_each_segment_all(bvec, bio, iter_all)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001218 __free_page(bvec->bv_page);
1219}
Guoqing Jiang491221f2016-09-22 03:10:01 -04001220EXPORT_SYMBOL(bio_free_pages);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222/**
1223 * bio_uncopy_user - finish previously mapped bio
1224 * @bio: bio being terminated
1225 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001226 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 * to user space in case of a read.
1228 */
1229int bio_uncopy_user(struct bio *bio)
1230{
1231 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001232 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Roland Dreier35dc2482013-08-05 17:55:01 -07001234 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1235 /*
1236 * if we're in a workqueue, the request is orphaned, so
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001237 * don't copy into a random user address space, just free
1238 * and return -EINTR so user space doesn't expect any data.
Roland Dreier35dc2482013-08-05 17:55:01 -07001239 */
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001240 if (!current->mm)
1241 ret = -EINTR;
1242 else if (bio_data_dir(bio) == READ)
Dongsu Park9124d3f2015-01-18 16:16:34 +01001243 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001244 if (bmd->is_our_pages)
1245 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001246 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001247 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 bio_put(bio);
1249 return ret;
1250}
1251
1252/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001253 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001254 * @q: destination block queue
1255 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1256 * @iter: iovec iterator
1257 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 *
1259 * Prepares and returns a bio for indirect user io, bouncing data
1260 * to/from kernel pages as necessary. Must be paired with
1261 * call bio_uncopy_user() on io completion.
1262 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001263struct bio *bio_copy_user_iov(struct request_queue *q,
1264 struct rq_map_data *map_data,
Al Viroe81cef52017-09-24 09:25:39 -04001265 struct iov_iter *iter,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001266 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 struct page *page;
1270 struct bio *bio;
Al Virod16d44e2017-09-24 13:09:18 -04001271 int i = 0, ret;
1272 int nr_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001273 unsigned int len = iter->count;
Geliang Tangbd5cece2015-11-21 17:27:31 +08001274 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Al Viro0e5b9352017-09-24 13:14:35 -04001276 bmd = bio_alloc_map_data(iter, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 if (!bmd)
1278 return ERR_PTR(-ENOMEM);
1279
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001280 /*
1281 * We need to do a deep copy of the iov_iter including the iovecs.
1282 * The caller provided iov might point to an on-stack or otherwise
1283 * shortlived one.
1284 */
1285 bmd->is_our_pages = map_data ? 0 : 1;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001286
Al Virod16d44e2017-09-24 13:09:18 -04001287 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1288 if (nr_pages > BIO_MAX_PAGES)
1289 nr_pages = BIO_MAX_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001292 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if (!bio)
1294 goto out_bmd;
1295
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001297
1298 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001299 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001300 i = map_data->offset / PAGE_SIZE;
1301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001303 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001305 bytes -= offset;
1306
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 if (bytes > len)
1308 bytes = len;
1309
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001310 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001311 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001312 ret = -ENOMEM;
1313 break;
1314 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001315
1316 page = map_data->pages[i / nr_pages];
1317 page += (i % nr_pages);
1318
1319 i++;
1320 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001321 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001322 if (!page) {
1323 ret = -ENOMEM;
1324 break;
1325 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 }
1327
Jérôme Glissea3761c32019-04-10 16:27:51 -04001328 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) {
1329 if (!map_data)
1330 __free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 break;
Jérôme Glissea3761c32019-04-10 16:27:51 -04001332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
1334 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001335 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 }
1337
1338 if (ret)
1339 goto cleanup;
1340
Al Viro2884d0b2017-09-24 12:09:21 -04001341 if (map_data)
1342 map_data->offset += bio->bi_iter.bi_size;
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 /*
1345 * success
1346 */
David Howells00e23702018-10-22 13:07:28 +01001347 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001348 (map_data && map_data->from_user)) {
Al Viro98a09d62017-09-24 12:14:36 -04001349 ret = bio_copy_from_iter(bio, iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001350 if (ret)
1351 goto cleanup;
Al Viro98a09d62017-09-24 12:14:36 -04001352 } else {
Keith Buschf55adad2018-12-10 08:44:42 -07001353 if (bmd->is_our_pages)
1354 zero_fill_bio(bio);
Al Viro98a09d62017-09-24 12:14:36 -04001355 iov_iter_advance(iter, bio->bi_iter.bi_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 }
1357
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001358 bio->bi_private = bmd;
Al Viro2884d0b2017-09-24 12:09:21 -04001359 if (map_data && map_data->null_mapped)
1360 bio_set_flag(bio, BIO_NULL_MAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 return bio;
1362cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001363 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001364 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 bio_put(bio);
1366out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001367 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 return ERR_PTR(ret);
1369}
1370
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001371/**
1372 * bio_map_user_iov - map user iovec into bio
1373 * @q: the struct request_queue for the bio
1374 * @iter: iovec iterator
1375 * @gfp_mask: memory allocation flags
1376 *
1377 * Map the user space address into a bio suitable for io to a block
1378 * device. Returns an error pointer in case of error.
1379 */
1380struct bio *bio_map_user_iov(struct request_queue *q,
Al Viroe81cef52017-09-24 09:25:39 -04001381 struct iov_iter *iter,
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001382 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001384 int j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 struct bio *bio;
Al Viro076098e2017-09-23 16:08:57 -04001386 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Al Virob282cc72017-09-23 16:24:59 -04001388 if (!iov_iter_count(iter))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 return ERR_PTR(-EINVAL);
1390
Al Virob282cc72017-09-23 16:24:59 -04001391 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 if (!bio)
1393 return ERR_PTR(-ENOMEM);
1394
Al Viro0a0f1512017-09-24 12:30:17 -04001395 while (iov_iter_count(iter)) {
Al Viro629e42b2017-09-23 16:13:10 -04001396 struct page **pages;
Al Viro076098e2017-09-23 16:08:57 -04001397 ssize_t bytes;
1398 size_t offs, added = 0;
1399 int npages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Al Viro0a0f1512017-09-24 12:30:17 -04001401 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
Al Viro076098e2017-09-23 16:08:57 -04001402 if (unlikely(bytes <= 0)) {
1403 ret = bytes ? bytes : -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001404 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
Al Viro076098e2017-09-23 16:08:57 -04001407 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Al Viro98f0bc92017-09-23 16:23:18 -04001409 if (unlikely(offs & queue_dma_alignment(q))) {
1410 ret = -EINVAL;
1411 j = 0;
1412 } else {
1413 for (j = 0; j < npages; j++) {
1414 struct page *page = pages[j];
1415 unsigned int n = PAGE_SIZE - offs;
Christoph Hellwigd1916c82019-08-12 17:39:58 +02001416 bool same_page = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Al Viro98f0bc92017-09-23 16:23:18 -04001418 if (n > bytes)
1419 n = bytes;
James Bottomley f1970ba2005-06-20 14:06:52 +02001420
Ming Lei19047082019-03-17 18:01:08 +08001421 if (!__bio_add_pc_page(q, bio, page, n, offs,
Christoph Hellwigd1916c82019-08-12 17:39:58 +02001422 &same_page)) {
1423 if (same_page)
1424 put_page(page);
Al Viro98f0bc92017-09-23 16:23:18 -04001425 break;
Christoph Hellwigd1916c82019-08-12 17:39:58 +02001426 }
Vitaly Mayatskikh95d78c22017-09-22 01:18:39 -04001427
Al Viro98f0bc92017-09-23 16:23:18 -04001428 added += n;
1429 bytes -= n;
1430 offs = 0;
1431 }
Al Viro0a0f1512017-09-24 12:30:17 -04001432 iov_iter_advance(iter, added);
James Bottomley f1970ba2005-06-20 14:06:52 +02001433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001435 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 */
Al Viro629e42b2017-09-23 16:13:10 -04001437 while (j < npages)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001438 put_page(pages[j++]);
Al Viro629e42b2017-09-23 16:13:10 -04001439 kvfree(pages);
Al Viroe2e115d2017-09-23 16:16:06 -04001440 /* couldn't stuff something into bio? */
1441 if (bytes)
1442 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
1444
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001445 bio_set_flag(bio, BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001446
1447 /*
Bart Van Assche5fad1b62017-02-01 08:20:08 -08001448 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001449 * it would normally disappear when its bi_end_io is run.
1450 * however, we need it for the unmap, so grab an extra
1451 * reference to it
1452 */
1453 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001455
1456 out_unmap:
Christoph Hellwig506e0792019-06-26 15:49:23 +02001457 bio_release_pages(bio, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 bio_put(bio);
1459 return ERR_PTR(ret);
1460}
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462/**
1463 * bio_unmap_user - unmap a bio
1464 * @bio: the bio being unmapped
1465 *
Bart Van Assche5fad1b62017-02-01 08:20:08 -08001466 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1467 * process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 *
1469 * bio_unmap_user() may sleep.
1470 */
1471void bio_unmap_user(struct bio *bio)
1472{
Christoph Hellwig163cc2d2019-06-26 15:49:22 +02001473 bio_release_pages(bio, bio_data_dir(bio) == READ);
1474 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 bio_put(bio);
1476}
1477
Damien Le Moalb4c58752019-07-01 14:09:15 +09001478static void bio_invalidate_vmalloc_pages(struct bio *bio)
1479{
1480#ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
1481 if (bio->bi_private && !op_is_write(bio_op(bio))) {
1482 unsigned long i, len = 0;
1483
1484 for (i = 0; i < bio->bi_vcnt; i++)
1485 len += bio->bi_io_vec[i].bv_len;
1486 invalidate_kernel_vmap_range(bio->bi_private, len);
1487 }
1488#endif
1489}
1490
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001491static void bio_map_kern_endio(struct bio *bio)
Jens Axboeb8238252005-06-20 14:05:27 +02001492{
Damien Le Moalb4c58752019-07-01 14:09:15 +09001493 bio_invalidate_vmalloc_pages(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001494 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001495}
1496
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001497/**
1498 * bio_map_kern - map kernel address into bio
1499 * @q: the struct request_queue for the bio
1500 * @data: pointer to buffer to map
1501 * @len: length in bytes
1502 * @gfp_mask: allocation flags for bio allocation
1503 *
1504 * Map the kernel address into a bio suitable for io to a block
1505 * device. Returns an error pointer in case of error.
1506 */
1507struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1508 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001509{
1510 unsigned long kaddr = (unsigned long)data;
1511 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1512 unsigned long start = kaddr >> PAGE_SHIFT;
1513 const int nr_pages = end - start;
Damien Le Moalb4c58752019-07-01 14:09:15 +09001514 bool is_vmalloc = is_vmalloc_addr(data);
1515 struct page *page;
Mike Christie df46b9a2005-06-20 14:04:44 +02001516 int offset, i;
1517 struct bio *bio;
1518
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001519 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001520 if (!bio)
1521 return ERR_PTR(-ENOMEM);
1522
Damien Le Moalb4c58752019-07-01 14:09:15 +09001523 if (is_vmalloc) {
1524 flush_kernel_vmap_range(data, len);
1525 bio->bi_private = data;
1526 }
1527
Mike Christie df46b9a2005-06-20 14:04:44 +02001528 offset = offset_in_page(kaddr);
1529 for (i = 0; i < nr_pages; i++) {
1530 unsigned int bytes = PAGE_SIZE - offset;
1531
1532 if (len <= 0)
1533 break;
1534
1535 if (bytes > len)
1536 bytes = len;
1537
Damien Le Moalb4c58752019-07-01 14:09:15 +09001538 if (!is_vmalloc)
1539 page = virt_to_page(data);
1540 else
1541 page = vmalloc_to_page(data);
1542 if (bio_add_pc_page(q, bio, page, bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001543 offset) < bytes) {
1544 /* we don't support partial mappings */
1545 bio_put(bio);
1546 return ERR_PTR(-EINVAL);
1547 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001548
1549 data += bytes;
1550 len -= bytes;
1551 offset = 0;
1552 }
1553
Jens Axboeb8238252005-06-20 14:05:27 +02001554 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001555 return bio;
1556}
Mike Christie df46b9a2005-06-20 14:04:44 +02001557
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001558static void bio_copy_kern_endio(struct bio *bio)
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001559{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001560 bio_free_pages(bio);
1561 bio_put(bio);
1562}
1563
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001564static void bio_copy_kern_endio_read(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001565{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001566 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001567 struct bio_vec *bvec;
Ming Lei6dc4f102019-02-15 19:13:19 +08001568 struct bvec_iter_all iter_all;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001569
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02001570 bio_for_each_segment_all(bvec, bio, iter_all) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001571 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001572 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001573 }
1574
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001575 bio_copy_kern_endio(bio);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001576}
1577
1578/**
1579 * bio_copy_kern - copy kernel address into bio
1580 * @q: the struct request_queue for the bio
1581 * @data: pointer to buffer to copy
1582 * @len: length in bytes
1583 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001584 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001585 *
1586 * copy the kernel address into a bio suitable for io to a block
1587 * device. Returns an error pointer in case of error.
1588 */
1589struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1590 gfp_t gfp_mask, int reading)
1591{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001592 unsigned long kaddr = (unsigned long)data;
1593 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1594 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001595 struct bio *bio;
1596 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001597 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001598
Christoph Hellwig42d26832015-01-18 16:16:28 +01001599 /*
1600 * Overflow, abort
1601 */
1602 if (end < start)
1603 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001604
Christoph Hellwig42d26832015-01-18 16:16:28 +01001605 nr_pages = end - start;
1606 bio = bio_kmalloc(gfp_mask, nr_pages);
1607 if (!bio)
1608 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001609
Christoph Hellwig42d26832015-01-18 16:16:28 +01001610 while (len) {
1611 struct page *page;
1612 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001613
Christoph Hellwig42d26832015-01-18 16:16:28 +01001614 if (bytes > len)
1615 bytes = len;
1616
1617 page = alloc_page(q->bounce_gfp | gfp_mask);
1618 if (!page)
1619 goto cleanup;
1620
1621 if (!reading)
1622 memcpy(page_address(page), p, bytes);
1623
1624 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1625 break;
1626
1627 len -= bytes;
1628 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001629 }
1630
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001631 if (reading) {
1632 bio->bi_end_io = bio_copy_kern_endio_read;
1633 bio->bi_private = data;
1634 } else {
1635 bio->bi_end_io = bio_copy_kern_endio;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001636 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001637
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001638 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001639
1640cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001641 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001642 bio_put(bio);
1643 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001644}
1645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646/*
1647 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1648 * for performing direct-IO in BIOs.
1649 *
1650 * The problem is that we cannot run set_page_dirty() from interrupt context
1651 * because the required locks are not interrupt-safe. So what we can do is to
1652 * mark the pages dirty _before_ performing IO. And in interrupt context,
1653 * check that the pages are still dirty. If so, fine. If not, redirty them
1654 * in process context.
1655 *
1656 * We special-case compound pages here: normally this means reads into hugetlb
1657 * pages. The logic in here doesn't really work right for compound pages
1658 * because the VM does not uniformly chase down the head page in all cases.
1659 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1660 * handle them at all. So we skip compound pages here at an early stage.
1661 *
1662 * Note that this code is very hard to test under normal circumstances because
1663 * direct-io pins the pages with get_user_pages(). This makes
1664 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001665 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 * pagecache.
1667 *
1668 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1669 * deferred bio dirtying paths.
1670 */
1671
1672/*
1673 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1674 */
1675void bio_set_pages_dirty(struct bio *bio)
1676{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001677 struct bio_vec *bvec;
Ming Lei6dc4f102019-02-15 19:13:19 +08001678 struct bvec_iter_all iter_all;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02001680 bio_for_each_segment_all(bvec, bio, iter_all) {
Christoph Hellwig3bb50982018-07-24 14:04:13 +02001681 if (!PageCompound(bvec->bv_page))
1682 set_page_dirty_lock(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 }
1684}
1685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686/*
1687 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1688 * If they are, then fine. If, however, some pages are clean then they must
1689 * have been written out during the direct-IO read. So we take another ref on
Christoph Hellwig24d54932018-07-24 14:04:12 +02001690 * the BIO and re-dirty the pages in process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 *
1692 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +03001693 * here on. It will run one put_page() against each page and will run one
1694 * bio_put() against the BIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 */
1696
David Howells65f27f32006-11-22 14:55:48 +00001697static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
David Howells65f27f32006-11-22 14:55:48 +00001699static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700static DEFINE_SPINLOCK(bio_dirty_lock);
1701static struct bio *bio_dirty_list;
1702
1703/*
1704 * This runs in process context
1705 */
David Howells65f27f32006-11-22 14:55:48 +00001706static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707{
Christoph Hellwig24d54932018-07-24 14:04:12 +02001708 struct bio *bio, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
Christoph Hellwig24d54932018-07-24 14:04:12 +02001710 spin_lock_irq(&bio_dirty_lock);
1711 next = bio_dirty_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 bio_dirty_list = NULL;
Christoph Hellwig24d54932018-07-24 14:04:12 +02001713 spin_unlock_irq(&bio_dirty_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Christoph Hellwig24d54932018-07-24 14:04:12 +02001715 while ((bio = next) != NULL) {
1716 next = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
Christoph Hellwigd241a952019-06-26 15:49:21 +02001718 bio_release_pages(bio, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 }
1721}
1722
1723void bio_check_pages_dirty(struct bio *bio)
1724{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001725 struct bio_vec *bvec;
Christoph Hellwig24d54932018-07-24 14:04:12 +02001726 unsigned long flags;
Ming Lei6dc4f102019-02-15 19:13:19 +08001727 struct bvec_iter_all iter_all;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Christoph Hellwig2b070cf2019-04-25 09:03:00 +02001729 bio_for_each_segment_all(bvec, bio, iter_all) {
Christoph Hellwig24d54932018-07-24 14:04:12 +02001730 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1731 goto defer;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 }
1733
Christoph Hellwigd241a952019-06-26 15:49:21 +02001734 bio_release_pages(bio, false);
Christoph Hellwig24d54932018-07-24 14:04:12 +02001735 bio_put(bio);
1736 return;
1737defer:
1738 spin_lock_irqsave(&bio_dirty_lock, flags);
1739 bio->bi_private = bio_dirty_list;
1740 bio_dirty_list = bio;
1741 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1742 schedule_work(&bio_dirty_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001745void update_io_ticks(struct hd_struct *part, unsigned long now)
1746{
1747 unsigned long stamp;
1748again:
1749 stamp = READ_ONCE(part->stamp);
1750 if (unlikely(stamp != now)) {
1751 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) {
1752 __part_stat_add(part, io_ticks, 1);
1753 }
1754 }
1755 if (part->partno) {
1756 part = &part_to_disk(part)->part0;
1757 goto again;
1758 }
1759}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Michael Callahanddcf35d2018-07-18 04:47:39 -07001761void generic_start_io_acct(struct request_queue *q, int op,
Jens Axboed62e26b2017-06-30 21:55:08 -06001762 unsigned long sectors, struct hd_struct *part)
Gu Zheng394ffa52014-11-24 11:05:22 +08001763{
Michael Callahanddcf35d2018-07-18 04:47:39 -07001764 const int sgrp = op_stat_group(op);
Gu Zheng394ffa52014-11-24 11:05:22 +08001765
Mike Snitzer112f1582018-12-06 11:41:18 -05001766 part_stat_lock();
1767
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001768 update_io_ticks(part, jiffies);
Mike Snitzer112f1582018-12-06 11:41:18 -05001769 part_stat_inc(part, ios[sgrp]);
1770 part_stat_add(part, sectors[sgrp], sectors);
Michael Callahanddcf35d2018-07-18 04:47:39 -07001771 part_inc_in_flight(q, part, op_is_write(op));
Gu Zheng394ffa52014-11-24 11:05:22 +08001772
1773 part_stat_unlock();
1774}
1775EXPORT_SYMBOL(generic_start_io_acct);
1776
Michael Callahanddcf35d2018-07-18 04:47:39 -07001777void generic_end_io_acct(struct request_queue *q, int req_op,
Jens Axboed62e26b2017-06-30 21:55:08 -06001778 struct hd_struct *part, unsigned long start_time)
Gu Zheng394ffa52014-11-24 11:05:22 +08001779{
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001780 unsigned long now = jiffies;
1781 unsigned long duration = now - start_time;
Michael Callahanddcf35d2018-07-18 04:47:39 -07001782 const int sgrp = op_stat_group(req_op);
Gu Zheng394ffa52014-11-24 11:05:22 +08001783
Mike Snitzer112f1582018-12-06 11:41:18 -05001784 part_stat_lock();
1785
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001786 update_io_ticks(part, now);
Mike Snitzer112f1582018-12-06 11:41:18 -05001787 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001788 part_stat_add(part, time_in_queue, duration);
Michael Callahanddcf35d2018-07-18 04:47:39 -07001789 part_dec_in_flight(q, part, op_is_write(req_op));
Gu Zheng394ffa52014-11-24 11:05:22 +08001790
1791 part_stat_unlock();
1792}
1793EXPORT_SYMBOL(generic_end_io_acct);
1794
Jens Axboec4cf5262015-04-17 16:15:18 -06001795static inline bool bio_remaining_done(struct bio *bio)
1796{
1797 /*
1798 * If we're not chaining, then ->__bi_remaining is always 1 and
1799 * we always end io on the first invocation.
1800 */
1801 if (!bio_flagged(bio, BIO_CHAIN))
1802 return true;
1803
1804 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1805
Mike Snitzer326e1db2015-05-22 09:14:03 -04001806 if (atomic_dec_and_test(&bio->__bi_remaining)) {
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001807 bio_clear_flag(bio, BIO_CHAIN);
Jens Axboec4cf5262015-04-17 16:15:18 -06001808 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001809 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001810
1811 return false;
1812}
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814/**
1815 * bio_endio - end I/O on a bio
1816 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 *
1818 * Description:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001819 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1820 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1821 * bio unless they own it and thus know that it has an end_io function.
NeilBrownfbbaf702017-04-07 09:40:52 -06001822 *
1823 * bio_endio() can be called several times on a bio that has been chained
1824 * using bio_chain(). The ->bi_end_io() function will only be called the
1825 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1826 * generated if BIO_TRACE_COMPLETION is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 **/
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001828void bio_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829{
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001830again:
Christoph Hellwig2b885512016-03-11 17:34:53 +01001831 if (!bio_remaining_done(bio))
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001832 return;
Christoph Hellwig7c20f112017-07-03 16:58:43 -06001833 if (!bio_integrity_endio(bio))
1834 return;
Kent Overstreet196d38bc2013-11-23 18:34:15 -08001835
Josef Bacik67b42d02018-07-03 11:15:00 -04001836 if (bio->bi_disk)
1837 rq_qos_done_bio(bio->bi_disk->queue, bio);
1838
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001839 /*
1840 * Need to have a real endio function for chained bios, otherwise
1841 * various corner cases will break (like stacking block devices that
1842 * save/restore bi_end_io) - however, we want to avoid unbounded
1843 * recursion and blowing the stack. Tail call optimization would
1844 * handle this, but compiling with frame pointers also disables
1845 * gcc's sibling call optimization.
1846 */
1847 if (bio->bi_end_io == bio_chain_endio) {
1848 bio = __bio_chain_endio(bio);
1849 goto again;
Kent Overstreet196d38bc2013-11-23 18:34:15 -08001850 }
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001851
Christoph Hellwig74d46992017-08-23 19:10:32 +02001852 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1853 trace_block_bio_complete(bio->bi_disk->queue, bio,
Bart Van Asschea462b952017-06-13 08:07:33 -07001854 blk_status_to_errno(bio->bi_status));
NeilBrownfbbaf702017-04-07 09:40:52 -06001855 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1856 }
1857
Shaohua Li9e234ee2017-03-27 10:51:41 -07001858 blk_throtl_bio_endio(bio);
Shaohua Lib222dd22017-07-10 11:40:17 -07001859 /* release cgroup info */
1860 bio_uninit(bio);
Christoph Hellwigba8c6962016-03-11 17:34:52 +01001861 if (bio->bi_end_io)
1862 bio->bi_end_io(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001864EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
Kent Overstreet196d38bc2013-11-23 18:34:15 -08001866/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001867 * bio_split - split a bio
1868 * @bio: bio to split
1869 * @sectors: number of sectors to split from the front of @bio
1870 * @gfp: gfp mask
1871 * @bs: bio set to allocate from
1872 *
1873 * Allocates and returns a new bio which represents @sectors from the start of
1874 * @bio, and updates @bio to represent the remaining sectors.
1875 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001876 * Unless this is a discard request the newly allocated bio will point
Bart Van Asschedad77582019-08-01 15:50:41 -07001877 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1878 * neither @bio nor @bs are freed before the split bio.
Kent Overstreet20d01892013-11-23 18:21:01 -08001879 */
1880struct bio *bio_split(struct bio *bio, int sectors,
1881 gfp_t gfp, struct bio_set *bs)
1882{
Mikulas Patockaf341a4d2017-11-22 13:18:05 -05001883 struct bio *split;
Kent Overstreet20d01892013-11-23 18:21:01 -08001884
1885 BUG_ON(sectors <= 0);
1886 BUG_ON(sectors >= bio_sectors(bio));
1887
Christoph Hellwigf9d03f92016-12-08 15:20:32 -07001888 split = bio_clone_fast(bio, gfp, bs);
Kent Overstreet20d01892013-11-23 18:21:01 -08001889 if (!split)
1890 return NULL;
1891
1892 split->bi_iter.bi_size = sectors << 9;
1893
1894 if (bio_integrity(split))
Dmitry Monakhovfbd08e72017-06-29 11:31:10 -07001895 bio_integrity_trim(split);
Kent Overstreet20d01892013-11-23 18:21:01 -08001896
1897 bio_advance(bio, split->bi_iter.bi_size);
1898
NeilBrownfbbaf702017-04-07 09:40:52 -06001899 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
Goldwyn Rodrigues20d59022018-01-23 09:10:19 -07001900 bio_set_flag(split, BIO_TRACE_COMPLETION);
NeilBrownfbbaf702017-04-07 09:40:52 -06001901
Kent Overstreet20d01892013-11-23 18:21:01 -08001902 return split;
1903}
1904EXPORT_SYMBOL(bio_split);
1905
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001906/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001907 * bio_trim - trim a bio
1908 * @bio: bio to trim
1909 * @offset: number of sectors to trim from the front of @bio
1910 * @size: size we want to trim @bio to, in sectors
1911 */
1912void bio_trim(struct bio *bio, int offset, int size)
1913{
1914 /* 'bio' is a cloned bio which we need to trim to match
1915 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001916 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001917
1918 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001919 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001920 return;
1921
Kent Overstreet6678d832013-08-07 11:14:32 -07001922 bio_advance(bio, offset << 9);
Kent Overstreet4f024f32013-10-11 15:44:27 -07001923 bio->bi_iter.bi_size = size;
Dmitry Monakhov376a78a2017-06-29 11:31:08 -07001924
1925 if (bio_integrity(bio))
Dmitry Monakhovfbd08e72017-06-29 11:31:10 -07001926 bio_integrity_trim(bio);
Dmitry Monakhov376a78a2017-06-29 11:31:08 -07001927
Kent Overstreet6678d832013-08-07 11:14:32 -07001928}
1929EXPORT_SYMBOL_GPL(bio_trim);
1930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931/*
1932 * create memory pools for biovec's in a bio_set.
1933 * use the global biovec slabs created for general use.
1934 */
Kent Overstreet8aa6ba22018-05-08 21:33:50 -04001935int biovec_init_pool(mempool_t *pool, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936{
Christoph Hellwiged996a52016-07-19 11:28:42 +02001937 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Kent Overstreet8aa6ba22018-05-08 21:33:50 -04001939 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940}
1941
Kent Overstreet917a38c2018-05-08 21:33:51 -04001942/*
1943 * bioset_exit - exit a bioset initialized with bioset_init()
1944 *
1945 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1946 * kzalloc()).
1947 */
1948void bioset_exit(struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001950 if (bs->rescue_workqueue)
1951 destroy_workqueue(bs->rescue_workqueue);
Kent Overstreet917a38c2018-05-08 21:33:51 -04001952 bs->rescue_workqueue = NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001953
Kent Overstreet8aa6ba22018-05-08 21:33:50 -04001954 mempool_exit(&bs->bio_pool);
1955 mempool_exit(&bs->bvec_pool);
Kent Overstreet9f060e22012-10-12 15:29:33 -07001956
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001957 bioset_integrity_free(bs);
Kent Overstreet917a38c2018-05-08 21:33:51 -04001958 if (bs->bio_slab)
1959 bio_put_slab(bs);
1960 bs->bio_slab = NULL;
1961}
1962EXPORT_SYMBOL(bioset_exit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
NeilBrown011067b2017-06-18 14:38:57 +10001964/**
Kent Overstreet917a38c2018-05-08 21:33:51 -04001965 * bioset_init - Initialize a bio_set
Kent Overstreetdad08522018-05-20 18:25:58 -04001966 * @bs: pool to initialize
Kent Overstreet917a38c2018-05-08 21:33:51 -04001967 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1968 * @front_pad: Number of bytes to allocate in front of the returned bio
1969 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1970 * and %BIOSET_NEED_RESCUER
1971 *
Kent Overstreetdad08522018-05-20 18:25:58 -04001972 * Description:
1973 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1974 * to ask for a number of bytes to be allocated in front of the bio.
1975 * Front pad allocation is useful for embedding the bio inside
1976 * another structure, to avoid allocating extra data to go with the bio.
1977 * Note that the bio must be embedded at the END of that structure always,
1978 * or things will break badly.
1979 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1980 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1981 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1982 * dispatch queued requests when the mempool runs out of space.
1983 *
Kent Overstreet917a38c2018-05-08 21:33:51 -04001984 */
1985int bioset_init(struct bio_set *bs,
1986 unsigned int pool_size,
1987 unsigned int front_pad,
1988 int flags)
1989{
1990 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1991
1992 bs->front_pad = front_pad;
1993
1994 spin_lock_init(&bs->rescue_lock);
1995 bio_list_init(&bs->rescue_list);
1996 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1997
1998 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1999 if (!bs->bio_slab)
2000 return -ENOMEM;
2001
2002 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
2003 goto bad;
2004
2005 if ((flags & BIOSET_NEED_BVECS) &&
2006 biovec_init_pool(&bs->bvec_pool, pool_size))
2007 goto bad;
2008
2009 if (!(flags & BIOSET_NEED_RESCUER))
2010 return 0;
2011
2012 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
2013 if (!bs->rescue_workqueue)
2014 goto bad;
2015
2016 return 0;
2017bad:
2018 bioset_exit(bs);
2019 return -ENOMEM;
2020}
2021EXPORT_SYMBOL(bioset_init);
2022
Jens Axboe28e89fd92018-06-07 14:42:05 -06002023/*
2024 * Initialize and setup a new bio_set, based on the settings from
2025 * another bio_set.
2026 */
2027int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
2028{
2029 int flags;
2030
2031 flags = 0;
2032 if (src->bvec_pool.min_nr)
2033 flags |= BIOSET_NEED_BVECS;
2034 if (src->rescue_workqueue)
2035 flags |= BIOSET_NEED_RESCUER;
2036
2037 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
2038}
2039EXPORT_SYMBOL(bioset_init_from_src);
2040
Tejun Heo852c7882012-03-05 13:15:27 -08002041#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04002042
2043/**
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002044 * bio_disassociate_blkg - puts back the blkg reference if associated
Dennis Zhoub5f29542018-11-01 17:24:10 -04002045 * @bio: target bio
Dennis Zhoub5f29542018-11-01 17:24:10 -04002046 *
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002047 * Helper to disassociate the blkg from @bio if a blkg is associated.
Dennis Zhoub5f29542018-11-01 17:24:10 -04002048 */
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002049void bio_disassociate_blkg(struct bio *bio)
Dennis Zhoub5f29542018-11-01 17:24:10 -04002050{
Josef Bacik08e18ea2018-07-03 11:14:50 -04002051 if (bio->bi_blkg) {
2052 blkg_put(bio->bi_blkg);
2053 bio->bi_blkg = NULL;
2054 }
Tejun Heo852c7882012-03-05 13:15:27 -08002055}
Dennis Zhou892ad712018-12-05 12:10:30 -05002056EXPORT_SYMBOL_GPL(bio_disassociate_blkg);
Tejun Heo852c7882012-03-05 13:15:27 -08002057
Paolo Valente20bd7232016-07-27 07:22:05 +02002058/**
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002059 * __bio_associate_blkg - associate a bio with the a blkg
Tejun Heo852c7882012-03-05 13:15:27 -08002060 * @bio: target bio
2061 * @blkg: the blkg to associate
2062 *
Dennis Zhoubeea9da2018-12-05 12:10:28 -05002063 * This tries to associate @bio with the specified @blkg. Association failure
2064 * is handled by walking up the blkg tree. Therefore, the blkg associated can
2065 * be anything between @blkg and the root_blkg. This situation only happens
2066 * when a cgroup is dying and then the remaining bios will spill to the closest
2067 * alive blkg.
2068 *
2069 * A reference will be taken on the @blkg and will be released when @bio is
2070 * freed.
Tejun Heo852c7882012-03-05 13:15:27 -08002071 */
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002072static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
Tejun Heo852c7882012-03-05 13:15:27 -08002073{
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002074 bio_disassociate_blkg(bio);
2075
Dennis Zhou7754f662018-12-05 12:10:39 -05002076 bio->bi_blkg = blkg_tryget_closest(blkg);
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002077}
2078
Dennis Zhoufd42df32018-12-05 12:10:34 -05002079/**
2080 * bio_associate_blkg_from_css - associate a bio with a specified css
2081 * @bio: target bio
2082 * @css: target css
2083 *
2084 * Associate @bio with the blkg found by combining the css's blkg and the
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002085 * request_queue of the @bio. This falls back to the queue's root_blkg if
2086 * the association fails with the css.
Dennis Zhoufd42df32018-12-05 12:10:34 -05002087 */
2088void bio_associate_blkg_from_css(struct bio *bio,
2089 struct cgroup_subsys_state *css)
2090{
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002091 struct request_queue *q = bio->bi_disk->queue;
2092 struct blkcg_gq *blkg;
2093
2094 rcu_read_lock();
2095
2096 if (!css || !css->parent)
2097 blkg = q->root_blkg;
2098 else
2099 blkg = blkg_lookup_create(css_to_blkcg(css), q);
2100
2101 __bio_associate_blkg(bio, blkg);
2102
2103 rcu_read_unlock();
Dennis Zhoufd42df32018-12-05 12:10:34 -05002104}
2105EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css);
2106
Dennis Zhou6a7f6d82018-12-05 12:10:33 -05002107#ifdef CONFIG_MEMCG
2108/**
2109 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2110 * @bio: target bio
2111 * @page: the page to lookup the blkcg from
2112 *
2113 * Associate @bio with the blkg from @page's owning memcg and the respective
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002114 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
2115 * root_blkg.
Dennis Zhou6a7f6d82018-12-05 12:10:33 -05002116 */
2117void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
2118{
2119 struct cgroup_subsys_state *css;
2120
Dennis Zhou6a7f6d82018-12-05 12:10:33 -05002121 if (!page->mem_cgroup)
2122 return;
2123
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002124 rcu_read_lock();
2125
2126 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
2127 bio_associate_blkg_from_css(bio, css);
2128
2129 rcu_read_unlock();
Dennis Zhou6a7f6d82018-12-05 12:10:33 -05002130}
2131#endif /* CONFIG_MEMCG */
2132
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002133/**
2134 * bio_associate_blkg - associate a bio with a blkg
2135 * @bio: target bio
2136 *
2137 * Associate @bio with the blkg found from the bio's css and request_queue.
2138 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
2139 * already associated, the css is reused and association redone as the
2140 * request_queue may have changed.
2141 */
2142void bio_associate_blkg(struct bio *bio)
2143{
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002144 struct cgroup_subsys_state *css;
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002145
2146 rcu_read_lock();
2147
Dennis Zhoudb6638d2018-12-05 12:10:35 -05002148 if (bio->bi_blkg)
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002149 css = &bio_blkcg(bio)->css;
Dennis Zhoudb6638d2018-12-05 12:10:35 -05002150 else
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002151 css = blkcg_css();
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002152
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002153 bio_associate_blkg_from_css(bio, css);
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002154
2155 rcu_read_unlock();
Tejun Heo852c7882012-03-05 13:15:27 -08002156}
Dennis Zhou5cdf2e32018-12-05 12:10:31 -05002157EXPORT_SYMBOL_GPL(bio_associate_blkg);
Tejun Heo852c7882012-03-05 13:15:27 -08002158
Paolo Valente20bd7232016-07-27 07:22:05 +02002159/**
Dennis Zhoudb6638d2018-12-05 12:10:35 -05002160 * bio_clone_blkg_association - clone blkg association from src to dst bio
Paolo Valente20bd7232016-07-27 07:22:05 +02002161 * @dst: destination bio
2162 * @src: source bio
2163 */
Dennis Zhoudb6638d2018-12-05 12:10:35 -05002164void bio_clone_blkg_association(struct bio *dst, struct bio *src)
Paolo Valente20bd7232016-07-27 07:22:05 +02002165{
Dennis Zhou6ab21872018-12-19 16:43:21 -06002166 rcu_read_lock();
2167
Dennis Zhoufc5a8282018-12-05 12:10:36 -05002168 if (src->bi_blkg)
Dennis Zhou2268c0f2018-12-05 12:10:29 -05002169 __bio_associate_blkg(dst, src->bi_blkg);
Dennis Zhou6ab21872018-12-19 16:43:21 -06002170
2171 rcu_read_unlock();
Paolo Valente20bd7232016-07-27 07:22:05 +02002172}
Dennis Zhoudb6638d2018-12-05 12:10:35 -05002173EXPORT_SYMBOL_GPL(bio_clone_blkg_association);
Tejun Heo852c7882012-03-05 13:15:27 -08002174#endif /* CONFIG_BLK_CGROUP */
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176static void __init biovec_init_slabs(void)
2177{
2178 int i;
2179
Christoph Hellwiged996a52016-07-19 11:28:42 +02002180 for (i = 0; i < BVEC_POOL_NR; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 int size;
2182 struct biovec_slab *bvs = bvec_slabs + i;
2183
Jens Axboea7fcd372008-12-05 16:10:29 +01002184 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2185 bvs->slab = NULL;
2186 continue;
2187 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 size = bvs->nr_vecs * sizeof(struct bio_vec);
2190 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002191 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 }
2193}
2194
2195static int __init init_bio(void)
2196{
Jens Axboebb799ca2008-12-10 15:35:05 +01002197 bio_slab_max = 2;
2198 bio_slab_nr = 0;
Kees Cook6396bb22018-06-12 14:03:40 -07002199 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
2200 GFP_KERNEL);
Johannes Thumshirn2b24e6f2019-04-03 11:15:19 +02002201
2202 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET);
2203
Jens Axboebb799ca2008-12-10 15:35:05 +01002204 if (!bio_slabs)
2205 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002207 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 biovec_init_slabs();
2209
Kent Overstreetf4f81542018-05-08 21:33:52 -04002210 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 panic("bio: can't allocate bios\n");
2212
Kent Overstreetf4f81542018-05-08 21:33:52 -04002213 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
Martin K. Petersena91a2782011-03-17 11:11:05 +01002214 panic("bio: can't create integrity pool\n");
2215
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 return 0;
2217}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218subsys_initcall(init_bio);