1 // SPDX-License-Identifier: GPL-2.0
2 /* bounce buffer handling for block devices
4 * - Split from highmem.c
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
24 #include <trace/events/block.h>
26 #include "blk-cgroup.h"
29 #define ISA_POOL_SIZE 16
31 static struct bio_set bounce_bio_set, bounce_bio_split;
32 static mempool_t page_pool;
34 static void init_bounce_bioset(void)
36 static bool bounce_bs_setup;
42 ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
44 if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
47 ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
49 bounce_bs_setup = true;
52 static __init int init_emergency_pool(void)
56 #ifndef CONFIG_MEMORY_HOTPLUG
57 if (max_pfn <= max_low_pfn)
61 ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
63 pr_info("pool size: %d pages\n", POOL_SIZE);
69 __initcall(init_emergency_pool);
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
76 static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
78 struct bio_vec tovec, fromvec;
79 struct bvec_iter iter;
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.
85 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
87 bio_for_each_segment(tovec, to, iter) {
88 fromvec = bio_iter_iovec(from, from_iter);
89 if (tovec.bv_page != fromvec.bv_page) {
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
95 memcpy_to_bvec(&tovec, page_address(fromvec.bv_page) +
98 bio_advance_iter(from, &from_iter, tovec.bv_len);
102 static void bounce_end_io(struct bio *bio)
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;
110 * free up bounce indirect pages used
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(bvec->bv_page, &page_pool);
118 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
121 bio_orig->bi_status = bio->bi_status;
126 static void bounce_end_io_write(struct bio *bio)
131 static void bounce_end_io_read(struct bio *bio)
133 struct bio *bio_orig = bio->bi_private;
136 copy_to_high_bio_irq(bio_orig, bio);
141 static struct bio *bounce_clone_bio(struct bio *bio_src)
143 struct bvec_iter iter;
148 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
149 * bio_src->bi_io_vec to bio->bi_io_vec.
151 * We can't do that anymore, because:
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.
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.
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.
167 bio = bio_alloc_bioset(bio_src->bi_bdev, bio_segments(bio_src),
168 bio_src->bi_opf, GFP_NOIO, &bounce_bio_set);
169 if (bio_flagged(bio_src, BIO_REMAPPED))
170 bio_set_flag(bio, BIO_REMAPPED);
171 bio->bi_ioprio = bio_src->bi_ioprio;
172 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
173 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
175 switch (bio_op(bio)) {
177 case REQ_OP_SECURE_ERASE:
178 case REQ_OP_WRITE_ZEROES:
181 bio_for_each_segment(bv, bio_src, iter)
182 bio->bi_io_vec[bio->bi_vcnt++] = bv;
186 if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0)
189 if (bio_integrity(bio_src) &&
190 bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0)
193 bio_clone_blkg_association(bio, bio_src);
202 struct bio *__blk_queue_bounce(struct bio *bio_orig, struct request_queue *q)
205 int rw = bio_data_dir(bio_orig);
206 struct bio_vec *to, from;
207 struct bvec_iter iter;
208 unsigned i = 0, bytes = 0;
212 bio_for_each_segment(from, bio_orig, iter) {
213 if (i++ < BIO_MAX_VECS)
214 bytes += from.bv_len;
215 if (PageHighMem(from.bv_page))
222 * Individual bvecs might not be logical block aligned. Round down
223 * the split size so that each bio is properly block size aligned,
224 * even if we do not use the full hardware limits.
226 sectors = ALIGN_DOWN(bytes, queue_logical_block_size(q)) >>
228 if (sectors < bio_sectors(bio_orig)) {
229 bio = bio_split(bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
230 bio_chain(bio, bio_orig);
231 submit_bio_noacct(bio_orig);
234 bio = bounce_clone_bio(bio_orig);
237 * Bvec table can't be updated by bio_for_each_segment_all(),
238 * so retrieve bvec from the table directly. This way is safe
239 * because the 'bio' is single-page bvec.
241 for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
242 struct page *bounce_page;
244 if (!PageHighMem(to->bv_page))
247 bounce_page = mempool_alloc(&page_pool, GFP_NOIO);
248 inc_zone_page_state(bounce_page, NR_BOUNCE);
251 flush_dcache_page(to->bv_page);
252 memcpy_from_bvec(page_address(bounce_page), to);
254 to->bv_page = bounce_page;
257 trace_block_bio_bounce(bio_orig);
259 bio->bi_flags |= (1 << BIO_BOUNCED);
262 bio->bi_end_io = bounce_end_io_read;
264 bio->bi_end_io = bounce_end_io_write;
266 bio->bi_private = bio_orig;