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/bootmem.h>
  22 #include <linux/printk.h>
  23 #include <asm/tlbflush.h>
  24
  25 #include <trace/events/block.h>
  26 #include "blk.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, isa_page_pool;
  33
  34 #if defined(CONFIG_HIGHMEM)
  35 static __init int init_emergency_pool(void)
  36 {
  37         int ret;
  38 #if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
  39         if (max_pfn <= max_low_pfn)
  40                 return 0;
  41 #endif
  42
  43         ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
  44         BUG_ON(ret);
  45         pr_info("pool size: %d pages\n", POOL_SIZE);
  46
  47         ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
  48         BUG_ON(ret);
  49         if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
  50                 BUG_ON(1);
  51
  52         ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
  53         BUG_ON(ret);
  54
  55         return 0;
  56 }
  57
  58 __initcall(init_emergency_pool);
  59 #endif
  60
  61 #ifdef CONFIG_HIGHMEM
  62 /*
  63  * highmem version, map in to vec
  64  */
  65 static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
  66 {
  67         unsigned char *vto;
  68
  69         vto = kmap_atomic(to->bv_page);
  70         memcpy(vto + to->bv_offset, vfrom, to->bv_len);
  71         kunmap_atomic(vto);
  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  */
  84 static 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  */
  93 int init_emergency_isa_pool(void)
  94 {
  95         int ret;
  96
  97         if (mempool_initialized(&isa_page_pool))
  98                 return 0;
  99
 100         ret = mempool_init(&isa_page_pool, ISA_POOL_SIZE, mempool_alloc_pages_isa,
 101                            mempool_free_pages, (void *) 0);
 102         BUG_ON(ret);
 103
 104         pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
 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  */
 113 static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
 114 {
 115         unsigned char *vfrom;
 116         struct bio_vec tovec, fromvec;
 117         struct bvec_iter iter;
 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;
 124
 125         bio_for_each_segment(tovec, to, iter) {
 126                 fromvec = bio_iter_iovec(from, from_iter);
 127                 if (tovec.bv_page != fromvec.bv_page) {
 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                          */
 133                         vfrom = page_address(fromvec.bv_page) +
 134                                 tovec.bv_offset;
 135
 136                         bounce_copy_vec(&tovec, vfrom);
 137                         flush_dcache_page(tovec.bv_page);
 138                 }
 139                 bio_advance_iter(from, &from_iter, tovec.bv_len);
 140         }
 141 }
 142
 143 static void bounce_end_io(struct bio *bio, mempool_t *pool)
 144 {
 145         struct bio *bio_orig = bio->bi_private;
 146         struct bio_vec *bvec, orig_vec;
 147         int i;
 148         struct bvec_iter orig_iter = bio_orig->bi_iter;
 149
 150         /*
 151          * free up bounce indirect pages used
 152          */
 153         bio_for_each_segment_all(bvec, bio, i) {
 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);
 160         }
 161
 162         bio_orig->bi_status = bio->bi_status;
 163         bio_endio(bio_orig);
 164         bio_put(bio);
 165 }
 166
 167 static void bounce_end_io_write(struct bio *bio)
 168 {
 169         bounce_end_io(bio, &page_pool);
 170 }
 171
 172 static void bounce_end_io_write_isa(struct bio *bio)
 173 {
 174
 175         bounce_end_io(bio, &isa_page_pool);
 176 }
 177
 178 static void __bounce_end_io_read(struct bio *bio, mempool_t *pool)
 179 {
 180         struct bio *bio_orig = bio->bi_private;
 181
 182         if (!bio->bi_status)
 183                 copy_to_high_bio_irq(bio_orig, bio);
 184
 185         bounce_end_io(bio, pool);
 186 }
 187
 188 static void bounce_end_io_read(struct bio *bio)
 189 {
 190         __bounce_end_io_read(bio, &page_pool);
 191 }
 192
 193 static void bounce_end_io_read_isa(struct bio *bio)
 194 {
 195         __bounce_end_io_read(bio, &isa_page_pool);
 196 }
 197
 198 static 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
 260         bio_clone_blkcg_association(bio, bio_src);
 261
 262         return bio;
 263 }
 264
 265 static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
 266                                mempool_t *pool)
 267 {
 268         struct bio *bio;
 269         int rw = bio_data_dir(*bio_orig);
 270         struct bio_vec *to, from;
 271         struct bvec_iter iter;
 272         unsigned i = 0;
 273         bool bounce = false;
 274         int sectors = 0;
 275         bool passthrough = bio_is_passthrough(*bio_orig);
 276
 277         bio_for_each_segment(from, *bio_orig, iter) {
 278                 if (i++ < BIO_MAX_PAGES)
 279                         sectors += from.bv_len >> 9;
 280                 if (page_to_pfn(from.bv_page) > q->limits.bounce_pfn)
 281                         bounce = true;
 282         }
 283         if (!bounce)
 284                 return;
 285
 286         if (!passthrough && sectors < bio_sectors(*bio_orig)) {
 287                 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
 288                 bio_chain(bio, *bio_orig);
 289                 generic_make_request(*bio_orig);
 290                 *bio_orig = bio;
 291         }
 292         bio = bounce_clone_bio(*bio_orig, GFP_NOIO, passthrough ? NULL :
 293                         &bounce_bio_set);
 294
 295         bio_for_each_segment_all(to, bio, i) {
 296                 struct page *page = to->bv_page;
 297
 298                 if (page_to_pfn(page) <= q->limits.bounce_pfn)
 299                         continue;
 300
 301                 to->bv_page = mempool_alloc(pool, q->bounce_gfp);
 302                 inc_zone_page_state(to->bv_page, NR_BOUNCE);
 303
 304                 if (rw == WRITE) {
 305                         char *vto, *vfrom;
 306
 307                         flush_dcache_page(page);
 308
 309                         vto = page_address(to->bv_page) + to->bv_offset;
 310                         vfrom = kmap_atomic(page) + to->bv_offset;
 311                         memcpy(vto, vfrom, to->bv_len);
 312                         kunmap_atomic(vfrom);
 313                 }
 314         }
 315
 316         trace_block_bio_bounce(q, *bio_orig);
 317
 318         bio->bi_flags |= (1 << BIO_BOUNCED);
 319
 320         if (pool == &page_pool) {
 321                 bio->bi_end_io = bounce_end_io_write;
 322                 if (rw == READ)
 323                         bio->bi_end_io = bounce_end_io_read;
 324         } else {
 325                 bio->bi_end_io = bounce_end_io_write_isa;
 326                 if (rw == READ)
 327                         bio->bi_end_io = bounce_end_io_read_isa;
 328         }
 329
 330         bio->bi_private = *bio_orig;
 331         *bio_orig = bio;
 332 }
 333
 334 void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
 335 {
 336         mempool_t *pool;
 337
 338         /*
 339          * Data-less bio, nothing to bounce
 340          */
 341         if (!bio_has_data(*bio_orig))
 342                 return;
 343
 344         /*
 345          * for non-isa bounce case, just check if the bounce pfn is equal
 346          * to or bigger than the highest pfn in the system -- in that case,
 347          * don't waste time iterating over bio segments
 348          */
 349         if (!(q->bounce_gfp & GFP_DMA)) {
 350                 if (q->limits.bounce_pfn >= blk_max_pfn)
 351                         return;
 352                 pool = &page_pool;
 353         } else {
 354                 BUG_ON(!mempool_initialized(&isa_page_pool));
 355                 pool = &isa_page_pool;
 356         }
 357
 358         /*
 359          * slow path
 360          */
 361         __blk_queue_bounce(q, bio_orig, pool);
 362 }