fs/ext4/readpage.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * linux/fs/ext4/readpage.c
0004  *
0005  * Copyright (C) 2002, Linus Torvalds.
0006  * Copyright (C) 2015, Google, Inc.
0007  *
0008  * This was originally taken from fs/mpage.c
0009  *
0010  * The ext4_mpage_readpages() function here is intended to
0011  * replace mpage_readahead() in the general case, not just for
0012  * encrypted files.  It has some limitations (see below), where it
0013  * will fall back to read_block_full_page(), but these limitations
0014  * should only be hit when page_size != block_size.
0015  *
0016  * This will allow us to attach a callback function to support ext4
0017  * encryption.
0018  *
0019  * If anything unusual happens, such as:
0020  *
0021  * - encountering a page which has buffers
0022  * - encountering a page which has a non-hole after a hole
0023  * - encountering a page with non-contiguous blocks
0024  *
0025  * then this code just gives up and calls the buffer_head-based read function.
0026  * It does handle a page which has holes at the end - that is a common case:
0027  * the end-of-file on blocksize < PAGE_SIZE setups.
0028  *
0029  */
0030
0031 #include <linux/kernel.h>
0032 #include <linux/export.h>
0033 #include <linux/mm.h>
0034 #include <linux/kdev_t.h>
0035 #include <linux/gfp.h>
0036 #include <linux/bio.h>
0037 #include <linux/fs.h>
0038 #include <linux/buffer_head.h>
0039 #include <linux/blkdev.h>
0040 #include <linux/highmem.h>
0041 #include <linux/prefetch.h>
0042 #include <linux/mpage.h>
0043 #include <linux/writeback.h>
0044 #include <linux/backing-dev.h>
0045 #include <linux/pagevec.h>
0046
0047 #include "ext4.h"
0048
0049 #define NUM_PREALLOC_POST_READ_CTXS 128
0050
0051 static struct kmem_cache *bio_post_read_ctx_cache;
0052 static mempool_t *bio_post_read_ctx_pool;
0053
0054 /* postprocessing steps for read bios */
0055 enum bio_post_read_step {
0056     STEP_INITIAL = 0,
0057     STEP_DECRYPT,
0058     STEP_VERITY,
0059     STEP_MAX,
0060 };
0061
0062 struct bio_post_read_ctx {
0063     struct bio *bio;
0064     struct work_struct work;
0065     unsigned int cur_step;
0066     unsigned int enabled_steps;
0067 };
0068
0069 static void __read_end_io(struct bio *bio)
0070 {
0071     struct page *page;
0072     struct bio_vec *bv;
0073     struct bvec_iter_all iter_all;
0074
0075     bio_for_each_segment_all(bv, bio, iter_all) {
0076         page = bv->bv_page;
0077
0078         /* PG_error was set if any post_read step failed */
0079         if (bio->bi_status || PageError(page)) {
0080             ClearPageUptodate(page);
0081             /* will re-read again later */
0082             ClearPageError(page);
0083         } else {
0084             SetPageUptodate(page);
0085         }
0086         unlock_page(page);
0087     }
0088     if (bio->bi_private)
0089         mempool_free(bio->bi_private, bio_post_read_ctx_pool);
0090     bio_put(bio);
0091 }
0092
0093 static void bio_post_read_processing(struct bio_post_read_ctx *ctx);
0094
0095 static void decrypt_work(struct work_struct *work)
0096 {
0097     struct bio_post_read_ctx *ctx =
0098         container_of(work, struct bio_post_read_ctx, work);
0099
0100     fscrypt_decrypt_bio(ctx->bio);
0101
0102     bio_post_read_processing(ctx);
0103 }
0104
0105 static void verity_work(struct work_struct *work)
0106 {
0107     struct bio_post_read_ctx *ctx =
0108         container_of(work, struct bio_post_read_ctx, work);
0109     struct bio *bio = ctx->bio;
0110
0111     /*
0112      * fsverity_verify_bio() may call readahead() again, and although verity
0113      * will be disabled for that, decryption may still be needed, causing
0114      * another bio_post_read_ctx to be allocated.  So to guarantee that
0115      * mempool_alloc() never deadlocks we must free the current ctx first.
0116      * This is safe because verity is the last post-read step.
0117      */
0118     BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX);
0119     mempool_free(ctx, bio_post_read_ctx_pool);
0120     bio->bi_private = NULL;
0121
0122     fsverity_verify_bio(bio);
0123
0124     __read_end_io(bio);
0125 }
0126
0127 static void bio_post_read_processing(struct bio_post_read_ctx *ctx)
0128 {
0129     /*
0130      * We use different work queues for decryption and for verity because
0131      * verity may require reading metadata pages that need decryption, and
0132      * we shouldn't recurse to the same workqueue.
0133      */
0134     switch (++ctx->cur_step) {
0135     case STEP_DECRYPT:
0136         if (ctx->enabled_steps & (1 << STEP_DECRYPT)) {
0137             INIT_WORK(&ctx->work, decrypt_work);
0138             fscrypt_enqueue_decrypt_work(&ctx->work);
0139             return;
0140         }
0141         ctx->cur_step++;
0142         fallthrough;
0143     case STEP_VERITY:
0144         if (ctx->enabled_steps & (1 << STEP_VERITY)) {
0145             INIT_WORK(&ctx->work, verity_work);
0146             fsverity_enqueue_verify_work(&ctx->work);
0147             return;
0148         }
0149         ctx->cur_step++;
0150         fallthrough;
0151     default:
0152         __read_end_io(ctx->bio);
0153     }
0154 }
0155
0156 static bool bio_post_read_required(struct bio *bio)
0157 {
0158     return bio->bi_private && !bio->bi_status;
0159 }
0160
0161 /*
0162  * I/O completion handler for multipage BIOs.
0163  *
0164  * The mpage code never puts partial pages into a BIO (except for end-of-file).
0165  * If a page does not map to a contiguous run of blocks then it simply falls
0166  * back to block_read_full_folio().
0167  *
0168  * Why is this?  If a page's completion depends on a number of different BIOs
0169  * which can complete in any order (or at the same time) then determining the
0170  * status of that page is hard.  See end_buffer_async_read() for the details.
0171  * There is no point in duplicating all that complexity.
0172  */
0173 static void mpage_end_io(struct bio *bio)
0174 {
0175     if (bio_post_read_required(bio)) {
0176         struct bio_post_read_ctx *ctx = bio->bi_private;
0177
0178         ctx->cur_step = STEP_INITIAL;
0179         bio_post_read_processing(ctx);
0180         return;
0181     }
0182     __read_end_io(bio);
0183 }
0184
0185 static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx)
0186 {
0187     return fsverity_active(inode) &&
0188            idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
0189 }
0190
0191 static void ext4_set_bio_post_read_ctx(struct bio *bio,
0192                        const struct inode *inode,
0193                        pgoff_t first_idx)
0194 {
0195     unsigned int post_read_steps = 0;
0196
0197     if (fscrypt_inode_uses_fs_layer_crypto(inode))
0198         post_read_steps |= 1 << STEP_DECRYPT;
0199
0200     if (ext4_need_verity(inode, first_idx))
0201         post_read_steps |= 1 << STEP_VERITY;
0202
0203     if (post_read_steps) {
0204         /* Due to the mempool, this never fails. */
0205         struct bio_post_read_ctx *ctx =
0206             mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
0207
0208         ctx->bio = bio;
0209         ctx->enabled_steps = post_read_steps;
0210         bio->bi_private = ctx;
0211     }
0212 }
0213
0214 static inline loff_t ext4_readpage_limit(struct inode *inode)
0215 {
0216     if (IS_ENABLED(CONFIG_FS_VERITY) &&
0217         (IS_VERITY(inode) || ext4_verity_in_progress(inode)))
0218         return inode->i_sb->s_maxbytes;
0219
0220     return i_size_read(inode);
0221 }
0222
0223 int ext4_mpage_readpages(struct inode *inode,
0224         struct readahead_control *rac, struct page *page)
0225 {
0226     struct bio *bio = NULL;
0227     sector_t last_block_in_bio = 0;
0228
0229     const unsigned blkbits = inode->i_blkbits;
0230     const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
0231     const unsigned blocksize = 1 << blkbits;
0232     sector_t next_block;
0233     sector_t block_in_file;
0234     sector_t last_block;
0235     sector_t last_block_in_file;
0236     sector_t blocks[MAX_BUF_PER_PAGE];
0237     unsigned page_block;
0238     struct block_device *bdev = inode->i_sb->s_bdev;
0239     int length;
0240     unsigned relative_block = 0;
0241     struct ext4_map_blocks map;
0242     unsigned int nr_pages = rac ? readahead_count(rac) : 1;
0243
0244     map.m_pblk = 0;
0245     map.m_lblk = 0;
0246     map.m_len = 0;
0247     map.m_flags = 0;
0248
0249     for (; nr_pages; nr_pages--) {
0250         int fully_mapped = 1;
0251         unsigned first_hole = blocks_per_page;
0252
0253         if (rac) {
0254             page = readahead_page(rac);
0255             prefetchw(&page->flags);
0256         }
0257
0258         if (page_has_buffers(page))
0259             goto confused;
0260
0261         block_in_file = next_block =
0262             (sector_t)page->index << (PAGE_SHIFT - blkbits);
0263         last_block = block_in_file + nr_pages * blocks_per_page;
0264         last_block_in_file = (ext4_readpage_limit(inode) +
0265                       blocksize - 1) >> blkbits;
0266         if (last_block > last_block_in_file)
0267             last_block = last_block_in_file;
0268         page_block = 0;
0269
0270         /*
0271          * Map blocks using the previous result first.
0272          */
0273         if ((map.m_flags & EXT4_MAP_MAPPED) &&
0274             block_in_file > map.m_lblk &&
0275             block_in_file < (map.m_lblk + map.m_len)) {
0276             unsigned map_offset = block_in_file - map.m_lblk;
0277             unsigned last = map.m_len - map_offset;
0278
0279             for (relative_block = 0; ; relative_block++) {
0280                 if (relative_block == last) {
0281                     /* needed? */
0282                     map.m_flags &= ~EXT4_MAP_MAPPED;
0283                     break;
0284                 }
0285                 if (page_block == blocks_per_page)
0286                     break;
0287                 blocks[page_block] = map.m_pblk + map_offset +
0288                     relative_block;
0289                 page_block++;
0290                 block_in_file++;
0291             }
0292         }
0293
0294         /*
0295          * Then do more ext4_map_blocks() calls until we are
0296          * done with this page.
0297          */
0298         while (page_block < blocks_per_page) {
0299             if (block_in_file < last_block) {
0300                 map.m_lblk = block_in_file;
0301                 map.m_len = last_block - block_in_file;
0302
0303                 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
0304                 set_error_page:
0305                     SetPageError(page);
0306                     zero_user_segment(page, 0,
0307                               PAGE_SIZE);
0308                     unlock_page(page);
0309                     goto next_page;
0310                 }
0311             }
0312             if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
0313                 fully_mapped = 0;
0314                 if (first_hole == blocks_per_page)
0315                     first_hole = page_block;
0316                 page_block++;
0317                 block_in_file++;
0318                 continue;
0319             }
0320             if (first_hole != blocks_per_page)
0321                 goto confused;      /* hole -> non-hole */
0322
0323             /* Contiguous blocks? */
0324             if (page_block && blocks[page_block-1] != map.m_pblk-1)
0325                 goto confused;
0326             for (relative_block = 0; ; relative_block++) {
0327                 if (relative_block == map.m_len) {
0328                     /* needed? */
0329                     map.m_flags &= ~EXT4_MAP_MAPPED;
0330                     break;
0331                 } else if (page_block == blocks_per_page)
0332                     break;
0333                 blocks[page_block] = map.m_pblk+relative_block;
0334                 page_block++;
0335                 block_in_file++;
0336             }
0337         }
0338         if (first_hole != blocks_per_page) {
0339             zero_user_segment(page, first_hole << blkbits,
0340                       PAGE_SIZE);
0341             if (first_hole == 0) {
0342                 if (ext4_need_verity(inode, page->index) &&
0343                     !fsverity_verify_page(page))
0344                     goto set_error_page;
0345                 SetPageUptodate(page);
0346                 unlock_page(page);
0347                 goto next_page;
0348             }
0349         } else if (fully_mapped) {
0350             SetPageMappedToDisk(page);
0351         }
0352
0353         /*
0354          * This page will go to BIO.  Do we need to send this
0355          * BIO off first?
0356          */
0357         if (bio && (last_block_in_bio != blocks[0] - 1 ||
0358                 !fscrypt_mergeable_bio(bio, inode, next_block))) {
0359         submit_and_realloc:
0360             submit_bio(bio);
0361             bio = NULL;
0362         }
0363         if (bio == NULL) {
0364             /*
0365              * bio_alloc will _always_ be able to allocate a bio if
0366              * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset().
0367              */
0368             bio = bio_alloc(bdev, bio_max_segs(nr_pages),
0369                     REQ_OP_READ, GFP_KERNEL);
0370             fscrypt_set_bio_crypt_ctx(bio, inode, next_block,
0371                           GFP_KERNEL);
0372             ext4_set_bio_post_read_ctx(bio, inode, page->index);
0373             bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
0374             bio->bi_end_io = mpage_end_io;
0375             if (rac)
0376                 bio->bi_opf |= REQ_RAHEAD;
0377         }
0378
0379         length = first_hole << blkbits;
0380         if (bio_add_page(bio, page, length, 0) < length)
0381             goto submit_and_realloc;
0382
0383         if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
0384              (relative_block == map.m_len)) ||
0385             (first_hole != blocks_per_page)) {
0386             submit_bio(bio);
0387             bio = NULL;
0388         } else
0389             last_block_in_bio = blocks[blocks_per_page - 1];
0390         goto next_page;
0391     confused:
0392         if (bio) {
0393             submit_bio(bio);
0394             bio = NULL;
0395         }
0396         if (!PageUptodate(page))
0397             block_read_full_folio(page_folio(page), ext4_get_block);
0398         else
0399             unlock_page(page);
0400     next_page:
0401         if (rac)
0402             put_page(page);
0403     }
0404     if (bio)
0405         submit_bio(bio);
0406     return 0;
0407 }
0408
0409 int __init ext4_init_post_read_processing(void)
0410 {
0411     bio_post_read_ctx_cache =
0412         kmem_cache_create("ext4_bio_post_read_ctx",
0413                   sizeof(struct bio_post_read_ctx), 0, 0, NULL);
0414     if (!bio_post_read_ctx_cache)
0415         goto fail;
0416     bio_post_read_ctx_pool =
0417         mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
0418                      bio_post_read_ctx_cache);
0419     if (!bio_post_read_ctx_pool)
0420         goto fail_free_cache;
0421     return 0;
0422
0423 fail_free_cache:
0424     kmem_cache_destroy(bio_post_read_ctx_cache);
0425 fail:
0426     return -ENOMEM;
0427 }
0428
0429 void ext4_exit_post_read_processing(void)
0430 {
0431     mempool_destroy(bio_post_read_ctx_pool);
0432     kmem_cache_destroy(bio_post_read_ctx_cache);
0433 }