]> git.ipfire.org Git - thirdparty/kernel/stable.git/blame - block/bounce.c
blkcg: reassociate bios when make_request() is called recursively
[thirdparty/kernel/stable.git] / block / bounce.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
831058de
DH
2/* bounce buffer handling for block devices
3 *
4 * - Split from highmem.c
5 */
6
b1de0d13
MH
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
831058de 9#include <linux/mm.h>
b95f1b31 10#include <linux/export.h>
831058de 11#include <linux/swap.h>
5a0e3ad6 12#include <linux/gfp.h>
831058de
DH
13#include <linux/bio.h>
14#include <linux/pagemap.h>
15#include <linux/mempool.h>
16#include <linux/blkdev.h>
66114cad 17#include <linux/backing-dev.h>
831058de
DH
18#include <linux/init.h>
19#include <linux/hash.h>
20#include <linux/highmem.h>
3bcfeaf9 21#include <linux/bootmem.h>
b1de0d13 22#include <linux/printk.h>
831058de
DH
23#include <asm/tlbflush.h>
24
55782138 25#include <trace/events/block.h>
3bce016a 26#include "blk.h"
55782138 27
831058de
DH
28#define POOL_SIZE 64
29#define ISA_POOL_SIZE 16
30
338aa96d
KO
31static struct bio_set bounce_bio_set, bounce_bio_split;
32static mempool_t page_pool, isa_page_pool;
831058de 33
a687a533 34#if defined(CONFIG_HIGHMEM)
831058de
DH
35static __init int init_emergency_pool(void)
36{
338aa96d 37 int ret;
f1006257 38#if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
3bcfeaf9 39 if (max_pfn <= max_low_pfn)
831058de 40 return 0;
3bcfeaf9 41#endif
831058de 42
338aa96d
KO
43 ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
44 BUG_ON(ret);
b1de0d13 45 pr_info("pool size: %d pages\n", POOL_SIZE);
831058de 46
338aa96d
KO
47 ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
48 BUG_ON(ret);
04c4950d 49 if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
a8821f3f
N
50 BUG_ON(1);
51
338aa96d
KO
52 ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
53 BUG_ON(ret);
a8821f3f 54
831058de
DH
55 return 0;
56}
57
58__initcall(init_emergency_pool);
f1006257 59#endif
831058de 60
f1006257 61#ifdef CONFIG_HIGHMEM
831058de
DH
62/*
63 * highmem version, map in to vec
64 */
65static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
66{
831058de
DH
67 unsigned char *vto;
68
9b04c5fe 69 vto = kmap_atomic(to->bv_page);
831058de 70 memcpy(vto + to->bv_offset, vfrom, to->bv_len);
9b04c5fe 71 kunmap_atomic(vto);
831058de
DH
72}
73
74#else /* CONFIG_HIGHMEM */
75
76#define bounce_copy_vec(to, vfrom) \
77 memcpy(page_address((to)->bv_page) + (to)->bv_offset, vfrom, (to)->bv_len)
78
79#endif /* CONFIG_HIGHMEM */
80
81/*
82 * allocate pages in the DMA region for the ISA pool
83 */
84static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
85{
86 return mempool_alloc_pages(gfp_mask | GFP_DMA, data);
87}
88
89/*
90 * gets called "every" time someone init's a queue with BLK_BOUNCE_ISA
91 * as the max address, so check if the pool has already been created.
92 */
93int init_emergency_isa_pool(void)
94{
338aa96d
KO
95 int ret;
96
97 if (mempool_initialized(&isa_page_pool))
831058de
DH
98 return 0;
99
338aa96d
KO
100 ret = mempool_init(&isa_page_pool, ISA_POOL_SIZE, mempool_alloc_pages_isa,
101 mempool_free_pages, (void *) 0);
102 BUG_ON(ret);
831058de 103
b1de0d13 104 pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
831058de
DH
105 return 0;
106}
107
108/*
109 * Simple bounce buffer support for highmem pages. Depending on the
110 * queue gfp mask set, *to may or may not be a highmem page. kmap it
111 * always, it will do the Right Thing
112 */
113static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
114{
115 unsigned char *vfrom;
3c892a09 116 struct bio_vec tovec, fromvec;
7988613b 117 struct bvec_iter iter;
3c892a09
ML
118 /*
119 * The bio of @from is created by bounce, so we can iterate
120 * its bvec from start to end, but the @from->bi_iter can't be
121 * trusted because it might be changed by splitting.
122 */
123 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
7988613b
KO
124
125 bio_for_each_segment(tovec, to, iter) {
3c892a09
ML
126 fromvec = bio_iter_iovec(from, from_iter);
127 if (tovec.bv_page != fromvec.bv_page) {
7988613b
KO
128 /*
129 * fromvec->bv_offset and fromvec->bv_len might have
130 * been modified by the block layer, so use the original
131 * copy, bounce_copy_vec already uses tovec->bv_len
132 */
3c892a09 133 vfrom = page_address(fromvec.bv_page) +
7988613b
KO
134 tovec.bv_offset;
135
136 bounce_copy_vec(&tovec, vfrom);
137 flush_dcache_page(tovec.bv_page);
138 }
3c892a09 139 bio_advance_iter(from, &from_iter, tovec.bv_len);
831058de
DH
140 }
141}
142
4246a0b6 143static void bounce_end_io(struct bio *bio, mempool_t *pool)
831058de
DH
144{
145 struct bio *bio_orig = bio->bi_private;
7891f05c 146 struct bio_vec *bvec, orig_vec;
831058de 147 int i;
7891f05c 148 struct bvec_iter orig_iter = bio_orig->bi_iter;
831058de 149
831058de
DH
150 /*
151 * free up bounce indirect pages used
152 */
d74c6d51 153 bio_for_each_segment_all(bvec, bio, i) {
7891f05c
ML
154 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
155 if (bvec->bv_page != orig_vec.bv_page) {
156 dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
157 mempool_free(bvec->bv_page, pool);
158 }
159 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
831058de
DH
160 }
161
4e4cbee9 162 bio_orig->bi_status = bio->bi_status;
4246a0b6 163 bio_endio(bio_orig);
831058de
DH
164 bio_put(bio);
165}
166
4246a0b6 167static void bounce_end_io_write(struct bio *bio)
831058de 168{
338aa96d 169 bounce_end_io(bio, &page_pool);
831058de
DH
170}
171
4246a0b6 172static void bounce_end_io_write_isa(struct bio *bio)
831058de 173{
831058de 174
338aa96d 175 bounce_end_io(bio, &isa_page_pool);
831058de
DH
176}
177
4246a0b6 178static void __bounce_end_io_read(struct bio *bio, mempool_t *pool)
831058de
DH
179{
180 struct bio *bio_orig = bio->bi_private;
181
4e4cbee9 182 if (!bio->bi_status)
831058de
DH
183 copy_to_high_bio_irq(bio_orig, bio);
184
4246a0b6 185 bounce_end_io(bio, pool);
831058de
DH
186}
187
4246a0b6 188static void bounce_end_io_read(struct bio *bio)
831058de 189{
338aa96d 190 __bounce_end_io_read(bio, &page_pool);
831058de
DH
191}
192
4246a0b6 193static void bounce_end_io_read_isa(struct bio *bio)
831058de 194{
338aa96d 195 __bounce_end_io_read(bio, &isa_page_pool);
831058de
DH
196}
197
c55183c9
CH
198static struct bio *bounce_clone_bio(struct bio *bio_src, gfp_t gfp_mask,
199 struct bio_set *bs)
200{
201 struct bvec_iter iter;
202 struct bio_vec bv;
203 struct bio *bio;
204
205 /*
206 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
207 * bio_src->bi_io_vec to bio->bi_io_vec.
208 *
209 * We can't do that anymore, because:
210 *
211 * - The point of cloning the biovec is to produce a bio with a biovec
212 * the caller can modify: bi_idx and bi_bvec_done should be 0.
213 *
214 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
215 * we tried to clone the whole thing bio_alloc_bioset() would fail.
216 * But the clone should succeed as long as the number of biovecs we
217 * actually need to allocate is fewer than BIO_MAX_PAGES.
218 *
219 * - Lastly, bi_vcnt should not be looked at or relied upon by code
220 * that does not own the bio - reason being drivers don't use it for
221 * iterating over the biovec anymore, so expecting it to be kept up
222 * to date (i.e. for clones that share the parent biovec) is just
223 * asking for trouble and would force extra work on
224 * __bio_clone_fast() anyways.
225 */
226
227 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
228 if (!bio)
229 return NULL;
230 bio->bi_disk = bio_src->bi_disk;
231 bio->bi_opf = bio_src->bi_opf;
232 bio->bi_write_hint = bio_src->bi_write_hint;
233 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
234 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
235
236 switch (bio_op(bio)) {
237 case REQ_OP_DISCARD:
238 case REQ_OP_SECURE_ERASE:
239 case REQ_OP_WRITE_ZEROES:
240 break;
241 case REQ_OP_WRITE_SAME:
242 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
243 break;
244 default:
245 bio_for_each_segment(bv, bio_src, iter)
246 bio->bi_io_vec[bio->bi_vcnt++] = bv;
247 break;
248 }
249
250 if (bio_integrity(bio_src)) {
251 int ret;
252
253 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
254 if (ret < 0) {
255 bio_put(bio);
256 return NULL;
257 }
258 }
259
c839e7a0 260 bio_clone_blkg_association(bio, bio_src);
c55183c9 261
5bf9a1f3
DZF
262 blkcg_bio_issue_init(bio);
263
c55183c9
CH
264 return bio;
265}
266
165125e1 267static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
a3ad0a9d 268 mempool_t *pool)
831058de 269{
6bc454d1
KO
270 struct bio *bio;
271 int rw = bio_data_dir(*bio_orig);
7988613b
KO
272 struct bio_vec *to, from;
273 struct bvec_iter iter;
a8821f3f
N
274 unsigned i = 0;
275 bool bounce = false;
276 int sectors = 0;
14cb0dc6 277 bool passthrough = bio_is_passthrough(*bio_orig);
831058de 278
a8821f3f
N
279 bio_for_each_segment(from, *bio_orig, iter) {
280 if (i++ < BIO_MAX_PAGES)
281 sectors += from.bv_len >> 9;
1c4bc3ab 282 if (page_to_pfn(from.bv_page) > q->limits.bounce_pfn)
a8821f3f
N
283 bounce = true;
284 }
285 if (!bounce)
286 return;
287
14cb0dc6 288 if (!passthrough && sectors < bio_sectors(*bio_orig)) {
338aa96d 289 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
a8821f3f
N
290 bio_chain(bio, *bio_orig);
291 generic_make_request(*bio_orig);
292 *bio_orig = bio;
293 }
c55183c9 294 bio = bounce_clone_bio(*bio_orig, GFP_NOIO, passthrough ? NULL :
338aa96d 295 &bounce_bio_set);
831058de 296
cb34e057 297 bio_for_each_segment_all(to, bio, i) {
6bc454d1 298 struct page *page = to->bv_page;
f735b5ee 299
1c4bc3ab 300 if (page_to_pfn(page) <= q->limits.bounce_pfn)
6bc454d1 301 continue;
831058de 302
6bc454d1 303 to->bv_page = mempool_alloc(pool, q->bounce_gfp);
393a3397 304 inc_zone_page_state(to->bv_page, NR_BOUNCE);
831058de
DH
305
306 if (rw == WRITE) {
307 char *vto, *vfrom;
308
6bc454d1
KO
309 flush_dcache_page(page);
310
831058de 311 vto = page_address(to->bv_page) + to->bv_offset;
6bc454d1 312 vfrom = kmap_atomic(page) + to->bv_offset;
831058de 313 memcpy(vto, vfrom, to->bv_len);
6bc454d1 314 kunmap_atomic(vfrom);
831058de
DH
315 }
316 }
317
5f3ea37c 318 trace_block_bio_bounce(q, *bio_orig);
c43a5082 319
831058de 320 bio->bi_flags |= (1 << BIO_BOUNCED);
831058de 321
338aa96d 322 if (pool == &page_pool) {
831058de
DH
323 bio->bi_end_io = bounce_end_io_write;
324 if (rw == READ)
325 bio->bi_end_io = bounce_end_io_read;
326 } else {
327 bio->bi_end_io = bounce_end_io_write_isa;
328 if (rw == READ)
329 bio->bi_end_io = bounce_end_io_read_isa;
330 }
331
332 bio->bi_private = *bio_orig;
333 *bio_orig = bio;
334}
335
165125e1 336void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
831058de
DH
337{
338 mempool_t *pool;
339
bf2de6f5
JA
340 /*
341 * Data-less bio, nothing to bounce
342 */
36144077 343 if (!bio_has_data(*bio_orig))
bf2de6f5
JA
344 return;
345
831058de
DH
346 /*
347 * for non-isa bounce case, just check if the bounce pfn is equal
348 * to or bigger than the highest pfn in the system -- in that case,
349 * don't waste time iterating over bio segments
350 */
351 if (!(q->bounce_gfp & GFP_DMA)) {
1c4bc3ab 352 if (q->limits.bounce_pfn >= blk_max_pfn)
831058de 353 return;
338aa96d 354 pool = &page_pool;
831058de 355 } else {
338aa96d
KO
356 BUG_ON(!mempool_initialized(&isa_page_pool));
357 pool = &isa_page_pool;
831058de
DH
358 }
359
831058de
DH
360 /*
361 * slow path
362 */
a3ad0a9d 363 __blk_queue_bounce(q, bio_orig, pool);
831058de 364}