fs/ext4/readpage.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * linux/fs/ext4/readpage.c
   4  *
   5  * Copyright (C) 2002, Linus Torvalds.
   6  * Copyright (C) 2015, Google, Inc.
   7  *
   8  * This was originally taken from fs/mpage.c
   9  *
  10  * The ext4_mpage_readpages() function here is intended to
  11  * replace mpage_readahead() in the general case, not just for
  12  * encrypted files.  It has some limitations (see below), where it
  13  * will fall back to read_block_full_page(), but these limitations
  14  * should only be hit when page_size != block_size.
  15  *
  16  * This will allow us to attach a callback function to support ext4
  17  * encryption.
  18  *
  19  * If anything unusual happens, such as:
  20  *
  21  * - encountering a page which has buffers
  22  * - encountering a page which has a non-hole after a hole
  23  * - encountering a page with non-contiguous blocks
  24  *
  25  * then this code just gives up and calls the buffer_head-based read function.
  26  * It does handle a page which has holes at the end - that is a common case:
  27  * the end-of-file on blocksize < PAGE_SIZE setups.
  28  *
  29  */
  30
  31 #include <linux/kernel.h>
  32 #include <linux/export.h>
  33 #include <linux/mm.h>
  34 #include <linux/kdev_t.h>
  35 #include <linux/gfp.h>
  36 #include <linux/bio.h>
  37 #include <linux/fs.h>
  38 #include <linux/buffer_head.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/highmem.h>
  41 #include <linux/prefetch.h>
  42 #include <linux/mpage.h>
  43 #include <linux/writeback.h>
  44 #include <linux/backing-dev.h>
  45 #include <linux/pagevec.h>
  46
  47 #include "ext4.h"
  48
  49 #define NUM_PREALLOC_POST_READ_CTXS     128
  50
  51 static struct kmem_cache *bio_post_read_ctx_cache;
  52 static mempool_t *bio_post_read_ctx_pool;
  53
  54 /* postprocessing steps for read bios */
  55 enum bio_post_read_step {
  56         STEP_INITIAL = 0,
  57         STEP_DECRYPT,
  58         STEP_VERITY,
  59         STEP_MAX,
  60 };
  61
  62 struct bio_post_read_ctx {
  63         struct bio *bio;
  64         struct work_struct work;
  65         unsigned int cur_step;
  66         unsigned int enabled_steps;
  67 };
  68
  69 static void __read_end_io(struct bio *bio)
  70 {
  71         struct folio_iter fi;
  72
  73         bio_for_each_folio_all(fi, bio)
  74                 folio_end_read(fi.folio, bio->bi_status == 0);
  75         if (bio->bi_private)
  76                 mempool_free(bio->bi_private, bio_post_read_ctx_pool);
  77         bio_put(bio);
  78 }
  79
  80 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
  81
  82 static void decrypt_work(struct work_struct *work)
  83 {
  84         struct bio_post_read_ctx *ctx =
  85                 container_of(work, struct bio_post_read_ctx, work);
  86         struct bio *bio = ctx->bio;
  87
  88         if (fscrypt_decrypt_bio(bio))
  89                 bio_post_read_processing(ctx);
  90         else
  91                 __read_end_io(bio);
  92 }
  93
  94 static void verity_work(struct work_struct *work)
  95 {
  96         struct bio_post_read_ctx *ctx =
  97                 container_of(work, struct bio_post_read_ctx, work);
  98         struct bio *bio = ctx->bio;
  99
 100         /*
 101          * fsverity_verify_bio() may call readahead() again, and although verity
 102          * will be disabled for that, decryption may still be needed, causing
 103          * another bio_post_read_ctx to be allocated.  So to guarantee that
 104          * mempool_alloc() never deadlocks we must free the current ctx first.
 105          * This is safe because verity is the last post-read step.
 106          */
 107         BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX);
 108         mempool_free(ctx, bio_post_read_ctx_pool);
 109         bio->bi_private = NULL;
 110
 111         fsverity_verify_bio(bio);
 112
 113         __read_end_io(bio);
 114 }
 115
 116 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
 117 {
 118         /*
 119          * We use different work queues for decryption and for verity because
 120          * verity may require reading metadata pages that need decryption, and
 121          * we shouldn't recurse to the same workqueue.
 122          */
 123         switch (++ctx->cur_step) {
 124         case STEP_DECRYPT:
 125                 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
 126                         INIT_WORK(&ctx->work, decrypt_work);
 127                         fscrypt_enqueue_decrypt_work(&ctx->work);
 128                         return;
 129                 }
 130                 ctx->cur_step++;
 131                 fallthrough;
 132         case STEP_VERITY:
 133                 if (ctx->enabled_steps & (1 << STEP_VERITY)) {
 134                         INIT_WORK(&ctx->work, verity_work);
 135                         fsverity_enqueue_verify_work(&ctx->work);
 136                         return;
 137                 }
 138                 ctx->cur_step++;
 139                 fallthrough;
 140         default:
 141                 __read_end_io(ctx->bio);
 142         }
 143 }
 144
 145 static bool bio_post_read_required(struct bio *bio)
 146 {
 147         return bio->bi_private && !bio->bi_status;
 148 }
 149
 150 /*
 151  * I/O completion handler for multipage BIOs.
 152  *
 153  * The mpage code never puts partial pages into a BIO (except for end-of-file).
 154  * If a page does not map to a contiguous run of blocks then it simply falls
 155  * back to block_read_full_folio().
 156  *
 157  * Why is this?  If a page's completion depends on a number of different BIOs
 158  * which can complete in any order (or at the same time) then determining the
 159  * status of that page is hard.  See end_buffer_async_read() for the details.
 160  * There is no point in duplicating all that complexity.
 161  */
 162 static void mpage_end_io(struct bio *bio)
 163 {
 164         if (bio_post_read_required(bio)) {
 165                 struct bio_post_read_ctx *ctx = bio->bi_private;
 166
 167                 ctx->cur_step = STEP_INITIAL;
 168                 bio_post_read_processing(ctx);
 169                 return;
 170         }
 171         __read_end_io(bio);
 172 }
 173
 174 static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx)
 175 {
 176         return fsverity_active(inode) &&
 177                idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
 178 }
 179
 180 static void ext4_set_bio_post_read_ctx(struct bio *bio,
 181                                        const struct inode *inode,
 182                                        pgoff_t first_idx)
 183 {
 184         unsigned int post_read_steps = 0;
 185
 186         if (fscrypt_inode_uses_fs_layer_crypto(inode))
 187                 post_read_steps |= 1 << STEP_DECRYPT;
 188
 189         if (ext4_need_verity(inode, first_idx))
 190                 post_read_steps |= 1 << STEP_VERITY;
 191
 192         if (post_read_steps) {
 193                 /* Due to the mempool, this never fails. */
 194                 struct bio_post_read_ctx *ctx =
 195                         mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
 196
 197                 ctx->bio = bio;
 198                 ctx->enabled_steps = post_read_steps;
 199                 bio->bi_private = ctx;
 200         }
 201 }
 202
 203 static inline loff_t ext4_readpage_limit(struct inode *inode)
 204 {
 205         if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
 206                 return inode->i_sb->s_maxbytes;
 207
 208         return i_size_read(inode);
 209 }
 210
 211 int ext4_mpage_readpages(struct inode *inode,
 212                 struct readahead_control *rac, struct folio *folio)
 213 {
 214         struct bio *bio = NULL;
 215         sector_t last_block_in_bio = 0;
 216
 217         const unsigned blkbits = inode->i_blkbits;
 218         const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
 219         const unsigned blocksize = 1 << blkbits;
 220         sector_t next_block;
 221         sector_t block_in_file;
 222         sector_t last_block;
 223         sector_t last_block_in_file;
 224         sector_t first_block;
 225         unsigned page_block;
 226         struct block_device *bdev = inode->i_sb->s_bdev;
 227         int length;
 228         unsigned relative_block = 0;
 229         struct ext4_map_blocks map;
 230         unsigned int nr_pages, folio_pages;
 231
 232         map.m_pblk = 0;
 233         map.m_lblk = 0;
 234         map.m_len = 0;
 235         map.m_flags = 0;
 236
 237         nr_pages = rac ? readahead_count(rac) : folio_nr_pages(folio);
 238         for (; nr_pages; nr_pages -= folio_pages) {
 239                 int fully_mapped = 1;
 240                 unsigned int first_hole;
 241                 unsigned int blocks_per_folio;
 242
 243                 if (rac)
 244                         folio = readahead_folio(rac);
 245
 246                 folio_pages = folio_nr_pages(folio);
 247                 prefetchw(&folio->flags);
 248
 249                 if (folio_buffers(folio))
 250                         goto confused;
 251
 252                 blocks_per_folio = folio_size(folio) >> blkbits;
 253                 first_hole = blocks_per_folio;
 254                 block_in_file = next_block =
 255                         (sector_t)folio->index << (PAGE_SHIFT - blkbits);
 256                 last_block = block_in_file + nr_pages * blocks_per_page;
 257                 last_block_in_file = (ext4_readpage_limit(inode) +
 258                                       blocksize - 1) >> blkbits;
 259                 if (last_block > last_block_in_file)
 260                         last_block = last_block_in_file;
 261                 page_block = 0;
 262
 263                 /*
 264                  * Map blocks using the previous result first.
 265                  */
 266                 if ((map.m_flags & EXT4_MAP_MAPPED) &&
 267                     block_in_file > map.m_lblk &&
 268                     block_in_file < (map.m_lblk + map.m_len)) {
 269                         unsigned map_offset = block_in_file - map.m_lblk;
 270                         unsigned last = map.m_len - map_offset;
 271
 272                         first_block = map.m_pblk + map_offset;
 273                         for (relative_block = 0; ; relative_block++) {
 274                                 if (relative_block == last) {
 275                                         /* needed? */
 276                                         map.m_flags &= ~EXT4_MAP_MAPPED;
 277                                         break;
 278                                 }
 279                                 if (page_block == blocks_per_folio)
 280                                         break;
 281                                 page_block++;
 282                                 block_in_file++;
 283                         }
 284                 }
 285
 286                 /*
 287                  * Then do more ext4_map_blocks() calls until we are
 288                  * done with this folio.
 289                  */
 290                 while (page_block < blocks_per_folio) {
 291                         if (block_in_file < last_block) {
 292                                 map.m_lblk = block_in_file;
 293                                 map.m_len = last_block - block_in_file;
 294
 295                                 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
 296                                 set_error_page:
 297                                         folio_zero_segment(folio, 0,
 298                                                           folio_size(folio));
 299                                         folio_unlock(folio);
 300                                         goto next_page;
 301                                 }
 302                         }
 303                         if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
 304                                 fully_mapped = 0;
 305                                 if (first_hole == blocks_per_folio)
 306                                         first_hole = page_block;
 307                                 page_block++;
 308                                 block_in_file++;
 309                                 continue;
 310                         }
 311                         if (first_hole != blocks_per_folio)
 312                                 goto confused;          /* hole -> non-hole */
 313
 314                         /* Contiguous blocks? */
 315                         if (!page_block)
 316                                 first_block = map.m_pblk;
 317                         else if (first_block + page_block != map.m_pblk)
 318                                 goto confused;
 319                         for (relative_block = 0; ; relative_block++) {
 320                                 if (relative_block == map.m_len) {
 321                                         /* needed? */
 322                                         map.m_flags &= ~EXT4_MAP_MAPPED;
 323                                         break;
 324                                 } else if (page_block == blocks_per_folio)
 325                                         break;
 326                                 page_block++;
 327                                 block_in_file++;
 328                         }
 329                 }
 330                 if (first_hole != blocks_per_folio) {
 331                         folio_zero_segment(folio, first_hole << blkbits,
 332                                           folio_size(folio));
 333                         if (first_hole == 0) {
 334                                 if (ext4_need_verity(inode, folio->index) &&
 335                                     !fsverity_verify_folio(folio))
 336                                         goto set_error_page;
 337                                 folio_end_read(folio, true);
 338                                 continue;
 339                         }
 340                 } else if (fully_mapped) {
 341                         folio_set_mappedtodisk(folio);
 342                 }
 343
 344                 /*
 345                  * This folio will go to BIO.  Do we need to send this
 346                  * BIO off first?
 347                  */
 348                 if (bio && (last_block_in_bio != first_block - 1 ||
 349                             !fscrypt_mergeable_bio(bio, inode, next_block))) {
 350                 submit_and_realloc:
 351                         submit_bio(bio);
 352                         bio = NULL;
 353                 }
 354                 if (bio == NULL) {
 355                         /*
 356                          * bio_alloc will _always_ be able to allocate a bio if
 357                          * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset().
 358                          */
 359                         bio = bio_alloc(bdev, bio_max_segs(nr_pages),
 360                                         REQ_OP_READ, GFP_KERNEL);
 361                         fscrypt_set_bio_crypt_ctx(bio, inode, next_block,
 362                                                   GFP_KERNEL);
 363                         ext4_set_bio_post_read_ctx(bio, inode, folio->index);
 364                         bio->bi_iter.bi_sector = first_block << (blkbits - 9);
 365                         bio->bi_end_io = mpage_end_io;
 366                         if (rac)
 367                                 bio->bi_opf |= REQ_RAHEAD;
 368                 }
 369
 370                 length = first_hole << blkbits;
 371                 if (!bio_add_folio(bio, folio, length, 0))
 372                         goto submit_and_realloc;
 373
 374                 if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
 375                      (relative_block == map.m_len)) ||
 376                     (first_hole != blocks_per_folio)) {
 377                         submit_bio(bio);
 378                         bio = NULL;
 379                 } else
 380                         last_block_in_bio = first_block + blocks_per_folio - 1;
 381                 continue;
 382         confused:
 383                 if (bio) {
 384                         submit_bio(bio);
 385                         bio = NULL;
 386                 }
 387                 if (!folio_test_uptodate(folio))
 388                         block_read_full_folio(folio, ext4_get_block);
 389                 else
 390                         folio_unlock(folio);
 391 next_page:
 392                 ; /* A label shall be followed by a statement until C23 */
 393         }
 394         if (bio)
 395                 submit_bio(bio);
 396         return 0;
 397 }
 398
 399 int __init ext4_init_post_read_processing(void)
 400 {
 401         bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT);
 402
 403         if (!bio_post_read_ctx_cache)
 404                 goto fail;
 405         bio_post_read_ctx_pool =
 406                 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
 407                                          bio_post_read_ctx_cache);
 408         if (!bio_post_read_ctx_pool)
 409                 goto fail_free_cache;
 410         return 0;
 411
 412 fail_free_cache:
 413         kmem_cache_destroy(bio_post_read_ctx_cache);
 414 fail:
 415         return -ENOMEM;
 416 }
 417
 418 void ext4_exit_post_read_processing(void)
 419 {
 420         mempool_destroy(bio_post_read_ctx_pool);
 421         kmem_cache_destroy(bio_post_read_ctx_cache);
 422 }