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 51static struct kmem_cache *bio_post_read_ctx_cache; 52static mempool_t *bio_post_read_ctx_pool; 53 54/* postprocessing steps for read bios */ 55enum bio_post_read_step { 56 STEP_INITIAL = 0, 57 STEP_DECRYPT, 58 STEP_VERITY, 59 STEP_MAX, 60}; 61 62struct 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 69static void __read_end_io(struct bio *bio) 70{ 71 struct folio_iter fi; 72 73 bio_for_each_folio_all(fi, bio) { 74 struct folio *folio = fi.folio; 75 76 if (bio->bi_status) 77 folio_clear_uptodate(folio); 78 else 79 folio_mark_uptodate(folio); 80 folio_unlock(folio); 81 } 82 if (bio->bi_private) 83 mempool_free(bio->bi_private, bio_post_read_ctx_pool); 84 bio_put(bio); 85} 86 87static void bio_post_read_processing(struct bio_post_read_ctx *ctx); 88 89static void decrypt_work(struct work_struct *work) 90{ 91 struct bio_post_read_ctx *ctx = 92 container_of(work, struct bio_post_read_ctx, work); 93 struct bio *bio = ctx->bio; 94 95 if (fscrypt_decrypt_bio(bio)) 96 bio_post_read_processing(ctx); 97 else 98 __read_end_io(bio); 99} 100 101static void verity_work(struct work_struct *work) 102{ 103 struct bio_post_read_ctx *ctx = 104 container_of(work, struct bio_post_read_ctx, work); 105 struct bio *bio = ctx->bio; 106 107 /* 108 * fsverity_verify_bio() may call readahead() again, and although verity 109 * will be disabled for that, decryption may still be needed, causing 110 * another bio_post_read_ctx to be allocated. So to guarantee that 111 * mempool_alloc() never deadlocks we must free the current ctx first. 112 * This is safe because verity is the last post-read step. 113 */ 114 BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX); 115 mempool_free(ctx, bio_post_read_ctx_pool); 116 bio->bi_private = NULL; 117 118 fsverity_verify_bio(bio); 119 120 __read_end_io(bio); 121} 122 123static void bio_post_read_processing(struct bio_post_read_ctx *ctx) 124{ 125 /* 126 * We use different work queues for decryption and for verity because 127 * verity may require reading metadata pages that need decryption, and 128 * we shouldn't recurse to the same workqueue. 129 */ 130 switch (++ctx->cur_step) { 131 case STEP_DECRYPT: 132 if (ctx->enabled_steps & (1 << STEP_DECRYPT)) { 133 INIT_WORK(&ctx->work, decrypt_work); 134 fscrypt_enqueue_decrypt_work(&ctx->work); 135 return; 136 } 137 ctx->cur_step++; 138 fallthrough; 139 case STEP_VERITY: 140 if (ctx->enabled_steps & (1 << STEP_VERITY)) { 141 INIT_WORK(&ctx->work, verity_work); 142 fsverity_enqueue_verify_work(&ctx->work); 143 return; 144 } 145 ctx->cur_step++; 146 fallthrough; 147 default: 148 __read_end_io(ctx->bio); 149 } 150} 151 152static bool bio_post_read_required(struct bio *bio) 153{ 154 return bio->bi_private && !bio->bi_status; 155} 156 157/* 158 * I/O completion handler for multipage BIOs. 159 * 160 * The mpage code never puts partial pages into a BIO (except for end-of-file). 161 * If a page does not map to a contiguous run of blocks then it simply falls 162 * back to block_read_full_folio(). 163 * 164 * Why is this? If a page's completion depends on a number of different BIOs 165 * which can complete in any order (or at the same time) then determining the 166 * status of that page is hard. See end_buffer_async_read() for the details. 167 * There is no point in duplicating all that complexity. 168 */ 169static void mpage_end_io(struct bio *bio) 170{ 171 if (bio_post_read_required(bio)) { 172 struct bio_post_read_ctx *ctx = bio->bi_private; 173 174 ctx->cur_step = STEP_INITIAL; 175 bio_post_read_processing(ctx); 176 return; 177 } 178 __read_end_io(bio); 179} 180 181static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx) 182{ 183 return fsverity_active(inode) && 184 idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); 185} 186 187static void ext4_set_bio_post_read_ctx(struct bio *bio, 188 const struct inode *inode, 189 pgoff_t first_idx) 190{ 191 unsigned int post_read_steps = 0; 192 193 if (fscrypt_inode_uses_fs_layer_crypto(inode)) 194 post_read_steps |= 1 << STEP_DECRYPT; 195 196 if (ext4_need_verity(inode, first_idx)) 197 post_read_steps |= 1 << STEP_VERITY; 198 199 if (post_read_steps) { 200 /* Due to the mempool, this never fails. */ 201 struct bio_post_read_ctx *ctx = 202 mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); 203 204 ctx->bio = bio; 205 ctx->enabled_steps = post_read_steps; 206 bio->bi_private = ctx; 207 } 208} 209 210static inline loff_t ext4_readpage_limit(struct inode *inode) 211{ 212 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) 213 return inode->i_sb->s_maxbytes; 214 215 return i_size_read(inode); 216} 217 218int ext4_mpage_readpages(struct inode *inode, 219 struct readahead_control *rac, struct folio *folio) 220{ 221 struct bio *bio = NULL; 222 sector_t last_block_in_bio = 0; 223 224 const unsigned blkbits = inode->i_blkbits; 225 const unsigned blocks_per_page = PAGE_SIZE >> blkbits; 226 const unsigned blocksize = 1 << blkbits; 227 sector_t next_block; 228 sector_t block_in_file; 229 sector_t last_block; 230 sector_t last_block_in_file; 231 sector_t blocks[MAX_BUF_PER_PAGE]; 232 unsigned page_block; 233 struct block_device *bdev = inode->i_sb->s_bdev; 234 int length; 235 unsigned relative_block = 0; 236 struct ext4_map_blocks map; 237 unsigned int nr_pages = rac ? readahead_count(rac) : 1; 238 239 map.m_pblk = 0; 240 map.m_lblk = 0; 241 map.m_len = 0; 242 map.m_flags = 0; 243 244 for (; nr_pages; nr_pages--) { 245 int fully_mapped = 1; 246 unsigned first_hole = blocks_per_page; 247 248 if (rac) 249 folio = readahead_folio(rac); 250 prefetchw(&folio->flags); 251 252 if (folio_buffers(folio)) 253 goto confused; 254 255 block_in_file = next_block = 256 (sector_t)folio->index << (PAGE_SHIFT - blkbits); 257 last_block = block_in_file + nr_pages * blocks_per_page; 258 last_block_in_file = (ext4_readpage_limit(inode) + 259 blocksize - 1) >> blkbits; 260 if (last_block > last_block_in_file) 261 last_block = last_block_in_file; 262 page_block = 0; 263 264 /* 265 * Map blocks using the previous result first. 266 */ 267 if ((map.m_flags & EXT4_MAP_MAPPED) && 268 block_in_file > map.m_lblk && 269 block_in_file < (map.m_lblk + map.m_len)) { 270 unsigned map_offset = block_in_file - map.m_lblk; 271 unsigned last = map.m_len - map_offset; 272 273 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_page) 280 break; 281 blocks[page_block] = map.m_pblk + map_offset + 282 relative_block; 283 page_block++; 284 block_in_file++; 285 } 286 } 287 288 /* 289 * Then do more ext4_map_blocks() calls until we are 290 * done with this folio. 291 */ 292 while (page_block < blocks_per_page) { 293 if (block_in_file < last_block) { 294 map.m_lblk = block_in_file; 295 map.m_len = last_block - block_in_file; 296 297 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) { 298 set_error_page: 299 folio_set_error(folio); 300 folio_zero_segment(folio, 0, 301 folio_size(folio)); 302 folio_unlock(folio); 303 goto next_page; 304 } 305 } 306 if ((map.m_flags & EXT4_MAP_MAPPED) == 0) { 307 fully_mapped = 0; 308 if (first_hole == blocks_per_page) 309 first_hole = page_block; 310 page_block++; 311 block_in_file++; 312 continue; 313 } 314 if (first_hole != blocks_per_page) 315 goto confused; /* hole -> non-hole */ 316 317 /* Contiguous blocks? */ 318 if (page_block && blocks[page_block-1] != map.m_pblk-1) 319 goto confused; 320 for (relative_block = 0; ; relative_block++) { 321 if (relative_block == map.m_len) { 322 /* needed? */ 323 map.m_flags &= ~EXT4_MAP_MAPPED; 324 break; 325 } else if (page_block == blocks_per_page) 326 break; 327 blocks[page_block] = map.m_pblk+relative_block; 328 page_block++; 329 block_in_file++; 330 } 331 } 332 if (first_hole != blocks_per_page) { 333 folio_zero_segment(folio, first_hole << blkbits, 334 folio_size(folio)); 335 if (first_hole == 0) { 336 if (ext4_need_verity(inode, folio->index) && 337 !fsverity_verify_folio(folio)) 338 goto set_error_page; 339 folio_mark_uptodate(folio); 340 folio_unlock(folio); 341 continue; 342 } 343 } else if (fully_mapped) { 344 folio_set_mappedtodisk(folio); 345 } 346 347 /* 348 * This folio will go to BIO. Do we need to send this 349 * BIO off first? 350 */ 351 if (bio && (last_block_in_bio != blocks[0] - 1 || 352 !fscrypt_mergeable_bio(bio, inode, next_block))) { 353 submit_and_realloc: 354 submit_bio(bio); 355 bio = NULL; 356 } 357 if (bio == NULL) { 358 /* 359 * bio_alloc will _always_ be able to allocate a bio if 360 * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset(). 361 */ 362 bio = bio_alloc(bdev, bio_max_segs(nr_pages), 363 REQ_OP_READ, GFP_KERNEL); 364 fscrypt_set_bio_crypt_ctx(bio, inode, next_block, 365 GFP_KERNEL); 366 ext4_set_bio_post_read_ctx(bio, inode, folio->index); 367 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); 368 bio->bi_end_io = mpage_end_io; 369 if (rac) 370 bio->bi_opf |= REQ_RAHEAD; 371 } 372 373 length = first_hole << blkbits; 374 if (!bio_add_folio(bio, folio, length, 0)) 375 goto submit_and_realloc; 376 377 if (((map.m_flags & EXT4_MAP_BOUNDARY) && 378 (relative_block == map.m_len)) || 379 (first_hole != blocks_per_page)) { 380 submit_bio(bio); 381 bio = NULL; 382 } else 383 last_block_in_bio = blocks[blocks_per_page - 1]; 384 continue; 385 confused: 386 if (bio) { 387 submit_bio(bio); 388 bio = NULL; 389 } 390 if (!folio_test_uptodate(folio)) 391 block_read_full_folio(folio, ext4_get_block); 392 else 393 folio_unlock(folio); 394next_page: 395 ; /* A label shall be followed by a statement until C23 */ 396 } 397 if (bio) 398 submit_bio(bio); 399 return 0; 400} 401 402int __init ext4_init_post_read_processing(void) 403{ 404 bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT); 405 406 if (!bio_post_read_ctx_cache) 407 goto fail; 408 bio_post_read_ctx_pool = 409 mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, 410 bio_post_read_ctx_cache); 411 if (!bio_post_read_ctx_pool) 412 goto fail_free_cache; 413 return 0; 414 415fail_free_cache: 416 kmem_cache_destroy(bio_post_read_ctx_cache); 417fail: 418 return -ENOMEM; 419} 420 421void ext4_exit_post_read_processing(void) 422{ 423 mempool_destroy(bio_post_read_ctx_pool); 424 kmem_cache_destroy(bio_post_read_ctx_cache); 425} 426