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  * highmem version, map in to vec
  72  */
  73 static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
  74 {
  75         unsigned char *vto;
  76
  77         vto = kmap_atomic(to->bv_page);
  78         memcpy(vto + to->bv_offset, vfrom, to->bv_len);
  79         kunmap_atomic(vto);
  80 }
  81
  82 /*
  83  * Simple bounce buffer support for highmem pages. Depending on the
  84  * queue gfp mask set, *to may or may not be a highmem page. kmap it
  85  * always, it will do the Right Thing
  86  */
  87 static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
  88 {
  89         unsigned char *vfrom;
  90         struct bio_vec tovec, fromvec;
  91         struct bvec_iter iter;
  92         /*
  93          * The bio of @from is created by bounce, so we can iterate
  94          * its bvec from start to end, but the @from->bi_iter can't be
  95          * trusted because it might be changed by splitting.
  96          */
  97         struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
  98
  99         bio_for_each_segment(tovec, to, iter) {
 100                 fromvec = bio_iter_iovec(from, from_iter);
 101                 if (tovec.bv_page != fromvec.bv_page) {
 102                         /*
 103                          * fromvec->bv_offset and fromvec->bv_len might have
 104                          * been modified by the block layer, so use the original
 105                          * copy, bounce_copy_vec already uses tovec->bv_len
 106                          */
 107                         vfrom = page_address(fromvec.bv_page) +
 108                                 tovec.bv_offset;
 109
 110                         bounce_copy_vec(&tovec, vfrom);
 111                         flush_dcache_page(tovec.bv_page);
 112                 }
 113                 bio_advance_iter(from, &from_iter, tovec.bv_len);
 114         }
 115 }
 116
 117 static void bounce_end_io(struct bio *bio)
 118 {
 119         struct bio *bio_orig = bio->bi_private;
 120         struct bio_vec *bvec, orig_vec;
 121         struct bvec_iter orig_iter = bio_orig->bi_iter;
 122         struct bvec_iter_all iter_all;
 123
 124         /*
 125          * free up bounce indirect pages used
 126          */
 127         bio_for_each_segment_all(bvec, bio, iter_all) {
 128                 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
 129                 if (bvec->bv_page != orig_vec.bv_page) {
 130                         dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
 131                         mempool_free(bvec->bv_page, &page_pool);
 132                 }
 133                 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
 134         }
 135
 136         bio_orig->bi_status = bio->bi_status;
 137         bio_endio(bio_orig);
 138         bio_put(bio);
 139 }
 140
 141 static void bounce_end_io_write(struct bio *bio)
 142 {
 143         bounce_end_io(bio);
 144 }
 145
 146 static void bounce_end_io_read(struct bio *bio)
 147 {
 148         struct bio *bio_orig = bio->bi_private;
 149
 150         if (!bio->bi_status)
 151                 copy_to_high_bio_irq(bio_orig, bio);
 152
 153         bounce_end_io(bio);
 154 }
 155
 156 static struct bio *bounce_clone_bio(struct bio *bio_src)
 157 {
 158         struct bvec_iter iter;
 159         struct bio_vec bv;
 160         struct bio *bio;
 161
 162         /*
 163          * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
 164          * bio_src->bi_io_vec to bio->bi_io_vec.
 165          *
 166          * We can't do that anymore, because:
 167          *
 168          *  - The point of cloning the biovec is to produce a bio with a biovec
 169          *    the caller can modify: bi_idx and bi_bvec_done should be 0.
 170          *
 171          *  - The original bio could've had more than BIO_MAX_VECS biovecs; if
 172          *    we tried to clone the whole thing bio_alloc_bioset() would fail.
 173          *    But the clone should succeed as long as the number of biovecs we
 174          *    actually need to allocate is fewer than BIO_MAX_VECS.
 175          *
 176          *  - Lastly, bi_vcnt should not be looked at or relied upon by code
 177          *    that does not own the bio - reason being drivers don't use it for
 178          *    iterating over the biovec anymore, so expecting it to be kept up
 179          *    to date (i.e. for clones that share the parent biovec) is just
 180          *    asking for trouble and would force extra work on
 181          *    __bio_clone_fast() anyways.
 182          */
 183         bio = bio_alloc_bioset(GFP_NOIO, bio_segments(bio_src),
 184                                &bounce_bio_set);
 185         bio->bi_bdev            = bio_src->bi_bdev;
 186         if (bio_flagged(bio_src, BIO_REMAPPED))
 187                 bio_set_flag(bio, BIO_REMAPPED);
 188         bio->bi_opf             = bio_src->bi_opf;
 189         bio->bi_ioprio          = bio_src->bi_ioprio;
 190         bio->bi_write_hint      = bio_src->bi_write_hint;
 191         bio->bi_iter.bi_sector  = bio_src->bi_iter.bi_sector;
 192         bio->bi_iter.bi_size    = bio_src->bi_iter.bi_size;
 193
 194         switch (bio_op(bio)) {
 195         case REQ_OP_DISCARD:
 196         case REQ_OP_SECURE_ERASE:
 197         case REQ_OP_WRITE_ZEROES:
 198                 break;
 199         case REQ_OP_WRITE_SAME:
 200                 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
 201                 break;
 202         default:
 203                 bio_for_each_segment(bv, bio_src, iter)
 204                         bio->bi_io_vec[bio->bi_vcnt++] = bv;
 205                 break;
 206         }
 207
 208         if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0)
 209                 goto err_put;
 210
 211         if (bio_integrity(bio_src) &&
 212             bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0)
 213                 goto err_put;
 214
 215         bio_clone_blkg_association(bio, bio_src);
 216         blkcg_bio_issue_init(bio);
 217
 218         return bio;
 219
 220 err_put:
 221         bio_put(bio);
 222         return NULL;
 223 }
 224
 225 void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
 226 {
 227         struct bio *bio;
 228         int rw = bio_data_dir(*bio_orig);
 229         struct bio_vec *to, from;
 230         struct bvec_iter iter;
 231         unsigned i = 0;
 232         bool bounce = false;
 233         int sectors = 0;
 234
 235         bio_for_each_segment(from, *bio_orig, iter) {
 236                 if (i++ < BIO_MAX_VECS)
 237                         sectors += from.bv_len >> 9;
 238                 if (PageHighMem(from.bv_page))
 239                         bounce = true;
 240         }
 241         if (!bounce)
 242                 return;
 243
 244         if (sectors < bio_sectors(*bio_orig)) {
 245                 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
 246                 bio_chain(bio, *bio_orig);
 247                 submit_bio_noacct(*bio_orig);
 248                 *bio_orig = bio;
 249         }
 250         bio = bounce_clone_bio(*bio_orig);
 251
 252         /*
 253          * Bvec table can't be updated by bio_for_each_segment_all(),
 254          * so retrieve bvec from the table directly. This way is safe
 255          * because the 'bio' is single-page bvec.
 256          */
 257         for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
 258                 struct page *page = to->bv_page;
 259
 260                 if (!PageHighMem(page))
 261                         continue;
 262
 263                 to->bv_page = mempool_alloc(&page_pool, GFP_NOIO);
 264                 inc_zone_page_state(to->bv_page, NR_BOUNCE);
 265
 266                 if (rw == WRITE) {
 267                         char *vto, *vfrom;
 268
 269                         flush_dcache_page(page);
 270
 271                         vto = page_address(to->bv_page) + to->bv_offset;
 272                         vfrom = kmap_atomic(page) + to->bv_offset;
 273                         memcpy(vto, vfrom, to->bv_len);
 274                         kunmap_atomic(vfrom);
 275                 }
 276         }
 277
 278         trace_block_bio_bounce(*bio_orig);
 279
 280         bio->bi_flags |= (1 << BIO_BOUNCED);
 281
 282         if (rw == READ)
 283                 bio->bi_end_io = bounce_end_io_read;
 284         else
 285                 bio->bi_end_io = bounce_end_io_write;
 286
 287         bio->bi_private = *bio_orig;
 288         *bio_orig = bio;
 289 }