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/memblock.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 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, 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, 0);
  48         BUG_ON(ret);
  49         bounce_bs_setup = true;
  50 }
  51
  52 #if defined(CONFIG_HIGHMEM)
  53 static __init int init_emergency_pool(void)
  54 {
  55         int ret;
  56 #if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
  57         if (max_pfn <= max_low_pfn)
  58                 return 0;
  59 #endif
  60
  61         ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 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 #endif
  71
  72 #ifdef CONFIG_HIGHMEM
  73 /*
  74  * highmem version, map in to vec
  75  */
  76 static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
  77 {
  78         unsigned char *vto;
  79
  80         vto = kmap_atomic(to->bv_page);
  81         memcpy(vto + to->bv_offset, vfrom, to->bv_len);
  82         kunmap_atomic(vto);
  83 }
  84
  85 #else /* CONFIG_HIGHMEM */
  86
  87 #define bounce_copy_vec(to, vfrom)      \
  88         memcpy(page_address((to)->bv_page) + (to)->bv_offset, vfrom, (to)->bv_len)
  89
  90 #endif /* CONFIG_HIGHMEM */
  91
  92 /*
  93  * allocate pages in the DMA region for the ISA pool
  94  */
  95 static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
  96 {
  97         return mempool_alloc_pages(gfp_mask | GFP_DMA, data);
  98 }
  99
 100 static DEFINE_MUTEX(isa_mutex);
 101
 102 /*
 103  * gets called "every" time someone init's a queue with BLK_BOUNCE_ISA
 104  * as the max address, so check if the pool has already been created.
 105  */
 106 int init_emergency_isa_pool(void)
 107 {
 108         int ret;
 109
 110         mutex_lock(&isa_mutex);
 111
 112         if (mempool_initialized(&isa_page_pool)) {
 113                 mutex_unlock(&isa_mutex);
 114                 return 0;
 115         }
 116
 117         ret = mempool_init(&isa_page_pool, ISA_POOL_SIZE, mempool_alloc_pages_isa,
 118                            mempool_free_pages, (void *) 0);
 119         BUG_ON(ret);
 120
 121         pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
 122         init_bounce_bioset();
 123         mutex_unlock(&isa_mutex);
 124         return 0;
 125 }
 126
 127 /*
 128  * Simple bounce buffer support for highmem pages. Depending on the
 129  * queue gfp mask set, *to may or may not be a highmem page. kmap it
 130  * always, it will do the Right Thing
 131  */
 132 static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
 133 {
 134         unsigned char *vfrom;
 135         struct bio_vec tovec, fromvec;
 136         struct bvec_iter iter;
 137         /*
 138          * The bio of @from is created by bounce, so we can iterate
 139          * its bvec from start to end, but the @from->bi_iter can't be
 140          * trusted because it might be changed by splitting.
 141          */
 142         struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
 143
 144         bio_for_each_segment(tovec, to, iter) {
 145                 fromvec = bio_iter_iovec(from, from_iter);
 146                 if (tovec.bv_page != fromvec.bv_page) {
 147                         /*
 148                          * fromvec->bv_offset and fromvec->bv_len might have
 149                          * been modified by the block layer, so use the original
 150                          * copy, bounce_copy_vec already uses tovec->bv_len
 151                          */
 152                         vfrom = page_address(fromvec.bv_page) +
 153                                 tovec.bv_offset;
 154
 155                         bounce_copy_vec(&tovec, vfrom);
 156                         flush_dcache_page(tovec.bv_page);
 157                 }
 158                 bio_advance_iter(from, &from_iter, tovec.bv_len);
 159         }
 160 }
 161
 162 static void bounce_end_io(struct bio *bio, mempool_t *pool)
 163 {
 164         struct bio *bio_orig = bio->bi_private;
 165         struct bio_vec *bvec, orig_vec;
 166         struct bvec_iter orig_iter = bio_orig->bi_iter;
 167         struct bvec_iter_all iter_all;
 168
 169         /*
 170          * free up bounce indirect pages used
 171          */
 172         bio_for_each_segment_all(bvec, bio, iter_all) {
 173                 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
 174                 if (bvec->bv_page != orig_vec.bv_page) {
 175                         dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
 176                         mempool_free(bvec->bv_page, pool);
 177                 }
 178                 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
 179         }
 180
 181         bio_orig->bi_status = bio->bi_status;
 182         bio_endio(bio_orig);
 183         bio_put(bio);
 184 }
 185
 186 static void bounce_end_io_write(struct bio *bio)
 187 {
 188         bounce_end_io(bio, &page_pool);
 189 }
 190
 191 static void bounce_end_io_write_isa(struct bio *bio)
 192 {
 193
 194         bounce_end_io(bio, &isa_page_pool);
 195 }
 196
 197 static void __bounce_end_io_read(struct bio *bio, mempool_t *pool)
 198 {
 199         struct bio *bio_orig = bio->bi_private;
 200
 201         if (!bio->bi_status)
 202                 copy_to_high_bio_irq(bio_orig, bio);
 203
 204         bounce_end_io(bio, pool);
 205 }
 206
 207 static void bounce_end_io_read(struct bio *bio)
 208 {
 209         __bounce_end_io_read(bio, &page_pool);
 210 }
 211
 212 static void bounce_end_io_read_isa(struct bio *bio)
 213 {
 214         __bounce_end_io_read(bio, &isa_page_pool);
 215 }
 216
 217 static struct bio *bounce_clone_bio(struct bio *bio_src)
 218 {
 219         struct bvec_iter iter;
 220         struct bio_vec bv;
 221         struct bio *bio;
 222
 223         /*
 224          * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
 225          * bio_src->bi_io_vec to bio->bi_io_vec.
 226          *
 227          * We can't do that anymore, because:
 228          *
 229          *  - The point of cloning the biovec is to produce a bio with a biovec
 230          *    the caller can modify: bi_idx and bi_bvec_done should be 0.
 231          *
 232          *  - The original bio could've had more than BIO_MAX_PAGES biovecs; if
 233          *    we tried to clone the whole thing bio_alloc_bioset() would fail.
 234          *    But the clone should succeed as long as the number of biovecs we
 235          *    actually need to allocate is fewer than BIO_MAX_PAGES.
 236          *
 237          *  - Lastly, bi_vcnt should not be looked at or relied upon by code
 238          *    that does not own the bio - reason being drivers don't use it for
 239          *    iterating over the biovec anymore, so expecting it to be kept up
 240          *    to date (i.e. for clones that share the parent biovec) is just
 241          *    asking for trouble and would force extra work on
 242          *    __bio_clone_fast() anyways.
 243          */
 244         if (bio_is_passthrough(bio_src))
 245                 bio = bio_kmalloc(GFP_NOIO | __GFP_NOFAIL,
 246                                   bio_segments(bio_src));
 247         else
 248                 bio = bio_alloc_bioset(GFP_NOIO, bio_segments(bio_src),
 249                                        &bounce_bio_set);
 250         bio->bi_bdev            = bio_src->bi_bdev;
 251         if (bio_flagged(bio_src, BIO_REMAPPED))
 252                 bio_set_flag(bio, BIO_REMAPPED);
 253         bio->bi_opf             = bio_src->bi_opf;
 254         bio->bi_ioprio          = bio_src->bi_ioprio;
 255         bio->bi_write_hint      = bio_src->bi_write_hint;
 256         bio->bi_iter.bi_sector  = bio_src->bi_iter.bi_sector;
 257         bio->bi_iter.bi_size    = bio_src->bi_iter.bi_size;
 258
 259         switch (bio_op(bio)) {
 260         case REQ_OP_DISCARD:
 261         case REQ_OP_SECURE_ERASE:
 262         case REQ_OP_WRITE_ZEROES:
 263                 break;
 264         case REQ_OP_WRITE_SAME:
 265                 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
 266                 break;
 267         default:
 268                 bio_for_each_segment(bv, bio_src, iter)
 269                         bio->bi_io_vec[bio->bi_vcnt++] = bv;
 270                 break;
 271         }
 272
 273         if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0)
 274                 goto err_put;
 275
 276         if (bio_integrity(bio_src) &&
 277             bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0)
 278                 goto err_put;
 279
 280         bio_clone_blkg_association(bio, bio_src);
 281         blkcg_bio_issue_init(bio);
 282
 283         return bio;
 284
 285 err_put:
 286         bio_put(bio);
 287         return NULL;
 288 }
 289
 290 static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
 291                                mempool_t *pool)
 292 {
 293         struct bio *bio;
 294         int rw = bio_data_dir(*bio_orig);
 295         struct bio_vec *to, from;
 296         struct bvec_iter iter;
 297         unsigned i = 0;
 298         bool bounce = false;
 299         int sectors = 0;
 300
 301         bio_for_each_segment(from, *bio_orig, iter) {
 302                 if (i++ < BIO_MAX_PAGES)
 303                         sectors += from.bv_len >> 9;
 304                 if (page_to_pfn(from.bv_page) > q->limits.bounce_pfn)
 305                         bounce = true;
 306         }
 307         if (!bounce)
 308                 return;
 309
 310         if (!bio_is_passthrough(*bio_orig) &&
 311             sectors < bio_sectors(*bio_orig)) {
 312                 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
 313                 bio_chain(bio, *bio_orig);
 314                 submit_bio_noacct(*bio_orig);
 315                 *bio_orig = bio;
 316         }
 317         bio = bounce_clone_bio(*bio_orig);
 318
 319         /*
 320          * Bvec table can't be updated by bio_for_each_segment_all(),
 321          * so retrieve bvec from the table directly. This way is safe
 322          * because the 'bio' is single-page bvec.
 323          */
 324         for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
 325                 struct page *page = to->bv_page;
 326
 327                 if (page_to_pfn(page) <= q->limits.bounce_pfn)
 328                         continue;
 329
 330                 to->bv_page = mempool_alloc(pool, q->bounce_gfp);
 331                 inc_zone_page_state(to->bv_page, NR_BOUNCE);
 332
 333                 if (rw == WRITE) {
 334                         char *vto, *vfrom;
 335
 336                         flush_dcache_page(page);
 337
 338                         vto = page_address(to->bv_page) + to->bv_offset;
 339                         vfrom = kmap_atomic(page) + to->bv_offset;
 340                         memcpy(vto, vfrom, to->bv_len);
 341                         kunmap_atomic(vfrom);
 342                 }
 343         }
 344
 345         trace_block_bio_bounce(*bio_orig);
 346
 347         bio->bi_flags |= (1 << BIO_BOUNCED);
 348
 349         if (pool == &page_pool) {
 350                 bio->bi_end_io = bounce_end_io_write;
 351                 if (rw == READ)
 352                         bio->bi_end_io = bounce_end_io_read;
 353         } else {
 354                 bio->bi_end_io = bounce_end_io_write_isa;
 355                 if (rw == READ)
 356                         bio->bi_end_io = bounce_end_io_read_isa;
 357         }
 358
 359         bio->bi_private = *bio_orig;
 360         *bio_orig = bio;
 361 }
 362
 363 void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
 364 {
 365         mempool_t *pool;
 366
 367         /*
 368          * Data-less bio, nothing to bounce
 369          */
 370         if (!bio_has_data(*bio_orig))
 371                 return;
 372
 373         /*
 374          * for non-isa bounce case, just check if the bounce pfn is equal
 375          * to or bigger than the highest pfn in the system -- in that case,
 376          * don't waste time iterating over bio segments
 377          */
 378         if (!(q->bounce_gfp & GFP_DMA)) {
 379                 if (q->limits.bounce_pfn >= blk_max_pfn)
 380                         return;
 381                 pool = &page_pool;
 382         } else {
 383                 BUG_ON(!mempool_initialized(&isa_page_pool));
 384                 pool = &isa_page_pool;
 385         }
 386
 387         /*
 388          * slow path
 389          */
 390         __blk_queue_bounce(q, bio_orig, pool);
 391 }