1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* bounce buffer handling for block devices |
3 | * |
4 | * - Split from highmem.c |
5 | */ |
6 | |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
8 | |
9 | #include <linux/mm.h> |
10 | #include <linux/export.h> |
11 | #include <linux/swap.h> |
12 | #include <linux/gfp.h> |
13 | #include <linux/bio.h> |
14 | #include <linux/pagemap.h> |
15 | #include <linux/mempool.h> |
16 | #include <linux/blkdev.h> |
17 | #include <linux/backing-dev.h> |
18 | #include <linux/init.h> |
19 | #include <linux/hash.h> |
20 | #include <linux/highmem.h> |
21 | #include <linux/printk.h> |
22 | #include <asm/tlbflush.h> |
23 | |
24 | #include <trace/events/block.h> |
25 | #include "blk.h" |
26 | #include "blk-cgroup.h" |
27 | |
28 | #define POOL_SIZE 64 |
29 | #define ISA_POOL_SIZE 16 |
30 | |
31 | static struct bio_set bounce_bio_set, bounce_bio_split; |
32 | static mempool_t page_pool; |
33 | |
34 | static void init_bounce_bioset(void) |
35 | { |
36 | static bool bounce_bs_setup; |
37 | int ret; |
38 | |
39 | if (bounce_bs_setup) |
40 | return; |
41 | |
42 | ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, flags: BIOSET_NEED_BVECS); |
43 | BUG_ON(ret); |
44 | if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE)) |
45 | BUG_ON(1); |
46 | |
47 | ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, flags: 0); |
48 | BUG_ON(ret); |
49 | bounce_bs_setup = true; |
50 | } |
51 | |
52 | static __init int init_emergency_pool(void) |
53 | { |
54 | int ret; |
55 | |
56 | #ifndef CONFIG_MEMORY_HOTPLUG |
57 | if (max_pfn <= max_low_pfn) |
58 | return 0; |
59 | #endif |
60 | |
61 | ret = mempool_init_page_pool(pool: &page_pool, POOL_SIZE, order: 0); |
62 | BUG_ON(ret); |
63 | pr_info("pool size: %d pages\n" , POOL_SIZE); |
64 | |
65 | init_bounce_bioset(); |
66 | return 0; |
67 | } |
68 | |
69 | __initcall(init_emergency_pool); |
70 | |
71 | /* |
72 | * Simple bounce buffer support for highmem pages. Depending on the |
73 | * queue gfp mask set, *to may or may not be a highmem page. kmap it |
74 | * always, it will do the Right Thing |
75 | */ |
76 | static void copy_to_high_bio_irq(struct bio *to, struct bio *from) |
77 | { |
78 | struct bio_vec tovec, fromvec; |
79 | struct bvec_iter iter; |
80 | /* |
81 | * The bio of @from is created by bounce, so we can iterate |
82 | * its bvec from start to end, but the @from->bi_iter can't be |
83 | * trusted because it might be changed by splitting. |
84 | */ |
85 | struct bvec_iter from_iter = BVEC_ITER_ALL_INIT; |
86 | |
87 | bio_for_each_segment(tovec, to, iter) { |
88 | fromvec = bio_iter_iovec(from, from_iter); |
89 | if (tovec.bv_page != fromvec.bv_page) { |
90 | /* |
91 | * fromvec->bv_offset and fromvec->bv_len might have |
92 | * been modified by the block layer, so use the original |
93 | * copy, bounce_copy_vec already uses tovec->bv_len |
94 | */ |
95 | memcpy_to_bvec(bvec: &tovec, page_address(fromvec.bv_page) + |
96 | tovec.bv_offset); |
97 | } |
98 | bio_advance_iter(bio: from, iter: &from_iter, bytes: tovec.bv_len); |
99 | } |
100 | } |
101 | |
102 | static void bounce_end_io(struct bio *bio) |
103 | { |
104 | struct bio *bio_orig = bio->bi_private; |
105 | struct bio_vec *bvec, orig_vec; |
106 | struct bvec_iter orig_iter = bio_orig->bi_iter; |
107 | struct bvec_iter_all iter_all; |
108 | |
109 | /* |
110 | * free up bounce indirect pages used |
111 | */ |
112 | bio_for_each_segment_all(bvec, bio, iter_all) { |
113 | orig_vec = bio_iter_iovec(bio_orig, orig_iter); |
114 | if (bvec->bv_page != orig_vec.bv_page) { |
115 | dec_zone_page_state(bvec->bv_page, NR_BOUNCE); |
116 | mempool_free(element: bvec->bv_page, pool: &page_pool); |
117 | } |
118 | bio_advance_iter(bio: bio_orig, iter: &orig_iter, bytes: orig_vec.bv_len); |
119 | } |
120 | |
121 | bio_orig->bi_status = bio->bi_status; |
122 | bio_endio(bio_orig); |
123 | bio_put(bio); |
124 | } |
125 | |
126 | static void bounce_end_io_write(struct bio *bio) |
127 | { |
128 | bounce_end_io(bio); |
129 | } |
130 | |
131 | static void bounce_end_io_read(struct bio *bio) |
132 | { |
133 | struct bio *bio_orig = bio->bi_private; |
134 | |
135 | if (!bio->bi_status) |
136 | copy_to_high_bio_irq(to: bio_orig, from: bio); |
137 | |
138 | bounce_end_io(bio); |
139 | } |
140 | |
141 | static struct bio *bounce_clone_bio(struct bio *bio_src) |
142 | { |
143 | struct bvec_iter iter; |
144 | struct bio_vec bv; |
145 | struct bio *bio; |
146 | |
147 | /* |
148 | * Pre immutable biovecs, __bio_clone() used to just do a memcpy from |
149 | * bio_src->bi_io_vec to bio->bi_io_vec. |
150 | * |
151 | * We can't do that anymore, because: |
152 | * |
153 | * - The point of cloning the biovec is to produce a bio with a biovec |
154 | * the caller can modify: bi_idx and bi_bvec_done should be 0. |
155 | * |
156 | * - The original bio could've had more than BIO_MAX_VECS biovecs; if |
157 | * we tried to clone the whole thing bio_alloc_bioset() would fail. |
158 | * But the clone should succeed as long as the number of biovecs we |
159 | * actually need to allocate is fewer than BIO_MAX_VECS. |
160 | * |
161 | * - Lastly, bi_vcnt should not be looked at or relied upon by code |
162 | * that does not own the bio - reason being drivers don't use it for |
163 | * iterating over the biovec anymore, so expecting it to be kept up |
164 | * to date (i.e. for clones that share the parent biovec) is just |
165 | * asking for trouble and would force extra work. |
166 | */ |
167 | bio = bio_alloc_bioset(bdev: bio_src->bi_bdev, nr_vecs: bio_segments(bio: bio_src), |
168 | opf: bio_src->bi_opf, GFP_NOIO, bs: &bounce_bio_set); |
169 | if (bio_flagged(bio: bio_src, bit: BIO_REMAPPED)) |
170 | bio_set_flag(bio, bit: BIO_REMAPPED); |
171 | bio->bi_ioprio = bio_src->bi_ioprio; |
172 | bio->bi_write_hint = bio_src->bi_write_hint; |
173 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; |
174 | bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; |
175 | |
176 | switch (bio_op(bio)) { |
177 | case REQ_OP_DISCARD: |
178 | case REQ_OP_SECURE_ERASE: |
179 | case REQ_OP_WRITE_ZEROES: |
180 | break; |
181 | default: |
182 | bio_for_each_segment(bv, bio_src, iter) |
183 | bio->bi_io_vec[bio->bi_vcnt++] = bv; |
184 | break; |
185 | } |
186 | |
187 | if (bio_crypt_clone(dst: bio, src: bio_src, GFP_NOIO) < 0) |
188 | goto err_put; |
189 | |
190 | if (bio_integrity(bio: bio_src) && |
191 | bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0) |
192 | goto err_put; |
193 | |
194 | bio_clone_blkg_association(dst: bio, src: bio_src); |
195 | |
196 | return bio; |
197 | |
198 | err_put: |
199 | bio_put(bio); |
200 | return NULL; |
201 | } |
202 | |
203 | struct bio *__blk_queue_bounce(struct bio *bio_orig, struct request_queue *q) |
204 | { |
205 | struct bio *bio; |
206 | int rw = bio_data_dir(bio_orig); |
207 | struct bio_vec *to, from; |
208 | struct bvec_iter iter; |
209 | unsigned i = 0, bytes = 0; |
210 | bool bounce = false; |
211 | int sectors; |
212 | |
213 | bio_for_each_segment(from, bio_orig, iter) { |
214 | if (i++ < BIO_MAX_VECS) |
215 | bytes += from.bv_len; |
216 | if (PageHighMem(page: from.bv_page)) |
217 | bounce = true; |
218 | } |
219 | if (!bounce) |
220 | return bio_orig; |
221 | |
222 | /* |
223 | * Individual bvecs might not be logical block aligned. Round down |
224 | * the split size so that each bio is properly block size aligned, |
225 | * even if we do not use the full hardware limits. |
226 | */ |
227 | sectors = ALIGN_DOWN(bytes, queue_logical_block_size(q)) >> |
228 | SECTOR_SHIFT; |
229 | if (sectors < bio_sectors(bio_orig)) { |
230 | bio = bio_split(bio: bio_orig, sectors, GFP_NOIO, bs: &bounce_bio_split); |
231 | bio_chain(bio, bio_orig); |
232 | submit_bio_noacct(bio: bio_orig); |
233 | bio_orig = bio; |
234 | } |
235 | bio = bounce_clone_bio(bio_src: bio_orig); |
236 | |
237 | /* |
238 | * Bvec table can't be updated by bio_for_each_segment_all(), |
239 | * so retrieve bvec from the table directly. This way is safe |
240 | * because the 'bio' is single-page bvec. |
241 | */ |
242 | for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) { |
243 | struct page *bounce_page; |
244 | |
245 | if (!PageHighMem(page: to->bv_page)) |
246 | continue; |
247 | |
248 | bounce_page = mempool_alloc(pool: &page_pool, GFP_NOIO); |
249 | inc_zone_page_state(bounce_page, NR_BOUNCE); |
250 | |
251 | if (rw == WRITE) { |
252 | flush_dcache_page(page: to->bv_page); |
253 | memcpy_from_bvec(page_address(bounce_page), bvec: to); |
254 | } |
255 | to->bv_page = bounce_page; |
256 | } |
257 | |
258 | trace_block_bio_bounce(bio: bio_orig); |
259 | |
260 | bio->bi_flags |= (1 << BIO_BOUNCED); |
261 | |
262 | if (rw == READ) |
263 | bio->bi_end_io = bounce_end_io_read; |
264 | else |
265 | bio->bi_end_io = bounce_end_io_write; |
266 | |
267 | bio->bi_private = bio_orig; |
268 | return bio; |
269 | } |
270 | |