block/bounce.c

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