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 page *page; 72 struct bio_vec *bv; 73 struct bvec_iter_all iter_all; 74 75 bio_for_each_segment_all(bv, bio, iter_all) { 76 page = bv->bv_page; 77 78 if (bio->bi_status) 79 ClearPageUptodate(page); 80 else 81 SetPageUptodate(page); 82 unlock_page(page); 83 } 84 if (bio->bi_private) 85 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 86 bio_put(bio); 87 } 88 89 static void bio_post_read_processing(struct bio_post_read_ctx *ctx); 90 91 static void decrypt_work(struct work_struct *work) 92 { 93 struct bio_post_read_ctx *ctx = 94 container_of(work, struct bio_post_read_ctx, work); 95 struct bio *bio = ctx->bio; 96 97 if (fscrypt_decrypt_bio(bio)) 98 bio_post_read_processing(ctx); 99 else 100 __read_end_io(bio); 101 } 102 103 static void verity_work(struct work_struct *work) 104 { 105 struct bio_post_read_ctx *ctx = 106 container_of(work, struct bio_post_read_ctx, work); 107 struct bio *bio = ctx->bio; 108 109 /* 110 * fsverity_verify_bio() may call readahead() again, and although verity 111 * will be disabled for that, decryption may still be needed, causing 112 * another bio_post_read_ctx to be allocated. So to guarantee that 113 * mempool_alloc() never deadlocks we must free the current ctx first. 114 * This is safe because verity is the last post-read step. 115 */ 116 BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX); 117 mempool_free(ctx, bio_post_read_ctx_pool); 118 bio->bi_private = NULL; 119 120 fsverity_verify_bio(bio); 121 122 __read_end_io(bio); 123 } 124 125 static void bio_post_read_processing(struct bio_post_read_ctx *ctx) 126 { 127 /* 128 * We use different work queues for decryption and for verity because 129 * verity may require reading metadata pages that need decryption, and 130 * we shouldn't recurse to the same workqueue. 131 */ 132 switch (++ctx->cur_step) { 133 case STEP_DECRYPT: 134 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) { 135 INIT_WORK(&ctx->work, decrypt_work); 136 fscrypt_enqueue_decrypt_work(&ctx->work); 137 return; 138 } 139 ctx->cur_step++; 140 fallthrough; 141 case STEP_VERITY: 142 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 143 INIT_WORK(&ctx->work, verity_work); 144 fsverity_enqueue_verify_work(&ctx->work); 145 return; 146 } 147 ctx->cur_step++; 148 fallthrough; 149 default: 150 __read_end_io(ctx->bio); 151 } 152 } 153 154 static bool bio_post_read_required(struct bio *bio) 155 { 156 return bio->bi_private && !bio->bi_status; 157 } 158 159 /* 160 * I/O completion handler for multipage BIOs. 161 * 162 * The mpage code never puts partial pages into a BIO (except for end-of-file). 163 * If a page does not map to a contiguous run of blocks then it simply falls 164 * back to block_read_full_folio(). 165 * 166 * Why is this? If a page's completion depends on a number of different BIOs 167 * which can complete in any order (or at the same time) then determining the 168 * status of that page is hard. See end_buffer_async_read() for the details. 169 * There is no point in duplicating all that complexity. 170 */ 171 static void mpage_end_io(struct bio *bio) 172 { 173 if (bio_post_read_required(bio)) { 174 struct bio_post_read_ctx *ctx = bio->bi_private; 175 176 ctx->cur_step = STEP_INITIAL; 177 bio_post_read_processing(ctx); 178 return; 179 } 180 __read_end_io(bio); 181 } 182 183 static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx) 184 { 185 return fsverity_active(inode) && 186 idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 187 } 188 189 static void ext4_set_bio_post_read_ctx(struct bio *bio, 190 const struct inode *inode, 191 pgoff_t first_idx) 192 { 193 unsigned int post_read_steps = 0; 194 195 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 196 post_read_steps |= 1 << STEP_DECRYPT; 197 198 if (ext4_need_verity(inode, first_idx)) 199 post_read_steps |= 1 << STEP_VERITY; 200 201 if (post_read_steps) { 202 /* Due to the mempool, this never fails. */ 203 struct bio_post_read_ctx *ctx = 204 mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 205 206 ctx->bio = bio; 207 ctx->enabled_steps = post_read_steps; 208 bio->bi_private = ctx; 209 } 210 } 211 212 static inline loff_t ext4_readpage_limit(struct inode *inode) 213 { 214 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 215 return inode->i_sb->s_maxbytes; 216 217 return i_size_read(inode); 218 } 219 220 int ext4_mpage_readpages(struct inode *inode, 221 struct readahead_control *rac, struct folio *folio) 222 { 223 struct bio *bio = NULL; 224 sector_t last_block_in_bio = 0; 225 226 const unsigned blkbits = inode->i_blkbits; 227 const unsigned blocks_per_page = PAGE_SIZE >> blkbits; 228 const unsigned blocksize = 1 << blkbits; 229 sector_t next_block; 230 sector_t block_in_file; 231 sector_t last_block; 232 sector_t last_block_in_file; 233 sector_t blocks[MAX_BUF_PER_PAGE]; 234 unsigned page_block; 235 struct block_device *bdev = inode->i_sb->s_bdev; 236 int length; 237 unsigned relative_block = 0; 238 struct ext4_map_blocks map; 239 unsigned int nr_pages = rac ? readahead_count(rac) : 1; 240 241 map.m_pblk = 0; 242 map.m_lblk = 0; 243 map.m_len = 0; 244 map.m_flags = 0; 245 246 for (; nr_pages; nr_pages--) { 247 int fully_mapped = 1; 248 unsigned first_hole = blocks_per_page; 249 250 if (rac) 251 folio = readahead_folio(rac); 252 prefetchw(&folio->flags); 253 254 if (folio_buffers(folio)) 255 goto confused; 256 257 block_in_file = next_block = 258 (sector_t)folio->index << (PAGE_SHIFT - blkbits); 259 last_block = block_in_file + nr_pages * blocks_per_page; 260 last_block_in_file = (ext4_readpage_limit(inode) + 261 blocksize - 1) >> blkbits; 262 if (last_block > last_block_in_file) 263 last_block = last_block_in_file; 264 page_block = 0; 265 266 /* 267 * Map blocks using the previous result first. 268 */ 269 if ((map.m_flags & EXT4_MAP_MAPPED) && 270 block_in_file > map.m_lblk && 271 block_in_file < (map.m_lblk + map.m_len)) { 272 unsigned map_offset = block_in_file - map.m_lblk; 273 unsigned last = map.m_len - map_offset; 274 275 for (relative_block = 0; ; relative_block++) { 276 if (relative_block == last) { 277 /* needed? */ 278 map.m_flags &= ~EXT4_MAP_MAPPED; 279 break; 280 } 281 if (page_block == blocks_per_page) 282 break; 283 blocks[page_block] = map.m_pblk + map_offset + 284 relative_block; 285 page_block++; 286 block_in_file++; 287 } 288 } 289 290 /* 291 * Then do more ext4_map_blocks() calls until we are 292 * done with this folio. 293 */ 294 while (page_block < blocks_per_page) { 295 if (block_in_file < last_block) { 296 map.m_lblk = block_in_file; 297 map.m_len = last_block - block_in_file; 298 299 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) { 300 set_error_page: 301 folio_set_error(folio); 302 folio_zero_segment(folio, 0, 303 folio_size(folio)); 304 folio_unlock(folio); 305 goto next_page; 306 } 307 } 308 if ((map.m_flags & EXT4_MAP_MAPPED) == 0) { 309 fully_mapped = 0; 310 if (first_hole == blocks_per_page) 311 first_hole = page_block; 312 page_block++; 313 block_in_file++; 314 continue; 315 } 316 if (first_hole != blocks_per_page) 317 goto confused; /* hole -> non-hole */ 318 319 /* Contiguous blocks? */ 320 if (page_block && blocks[page_block-1] != map.m_pblk-1) 321 goto confused; 322 for (relative_block = 0; ; relative_block++) { 323 if (relative_block == map.m_len) { 324 /* needed? */ 325 map.m_flags &= ~EXT4_MAP_MAPPED; 326 break; 327 } else if (page_block == blocks_per_page) 328 break; 329 blocks[page_block] = map.m_pblk+relative_block; 330 page_block++; 331 block_in_file++; 332 } 333 } 334 if (first_hole != blocks_per_page) { 335 folio_zero_segment(folio, first_hole << blkbits, 336 folio_size(folio)); 337 if (first_hole == 0) { 338 if (ext4_need_verity(inode, folio->index) && 339 !fsverity_verify_page(&folio->page)) 340 goto set_error_page; 341 folio_mark_uptodate(folio); 342 folio_unlock(folio); 343 continue; 344 } 345 } else if (fully_mapped) { 346 folio_set_mappedtodisk(folio); 347 } 348 349 /* 350 * This folio will go to BIO. Do we need to send this 351 * BIO off first? 352 */ 353 if (bio && (last_block_in_bio != blocks[0] - 1 || 354 !fscrypt_mergeable_bio(bio, inode, next_block))) { 355 submit_and_realloc: 356 submit_bio(bio); 357 bio = NULL; 358 } 359 if (bio == NULL) { 360 /* 361 * bio_alloc will _always_ be able to allocate a bio if 362 * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset(). 363 */ 364 bio = bio_alloc(bdev, bio_max_segs(nr_pages), 365 REQ_OP_READ, GFP_KERNEL); 366 fscrypt_set_bio_crypt_ctx(bio, inode, next_block, 367 GFP_KERNEL); 368 ext4_set_bio_post_read_ctx(bio, inode, folio->index); 369 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); 370 bio->bi_end_io = mpage_end_io; 371 if (rac) 372 bio->bi_opf |= REQ_RAHEAD; 373 } 374 375 length = first_hole << blkbits; 376 if (!bio_add_folio(bio, folio, length, 0)) 377 goto submit_and_realloc; 378 379 if (((map.m_flags & EXT4_MAP_BOUNDARY) && 380 (relative_block == map.m_len)) || 381 (first_hole != blocks_per_page)) { 382 submit_bio(bio); 383 bio = NULL; 384 } else 385 last_block_in_bio = blocks[blocks_per_page - 1]; 386 continue; 387 confused: 388 if (bio) { 389 submit_bio(bio); 390 bio = NULL; 391 } 392 if (!folio_test_uptodate(folio)) 393 block_read_full_folio(folio, ext4_get_block); 394 else 395 folio_unlock(folio); 396 next_page: 397 ; /* A label shall be followed by a statement until C23 */ 398 } 399 if (bio) 400 submit_bio(bio); 401 return 0; 402 } 403 404 int __init ext4_init_post_read_processing(void) 405 { 406 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT); 407 408 if (!bio_post_read_ctx_cache) 409 goto fail; 410 bio_post_read_ctx_pool = 411 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 412 bio_post_read_ctx_cache); 413 if (!bio_post_read_ctx_pool) 414 goto fail_free_cache; 415 return 0; 416 417 fail_free_cache: 418 kmem_cache_destroy(bio_post_read_ctx_cache); 419 fail: 420 return -ENOMEM; 421 } 422 423 void ext4_exit_post_read_processing(void) 424 { 425 mempool_destroy(bio_post_read_ctx_pool); 426 kmem_cache_destroy(bio_post_read_ctx_cache); 427 } 428