Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
f64e02fe TT |
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 | * | |
6311f91f MWO |
10 | * The ext4_mpage_readpages() function here is intended to |
11 | * replace mpage_readahead() in the general case, not just for | |
f64e02fe TT |
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: | |
ea1754a0 | 27 | * the end-of-file on blocksize < PAGE_SIZE setups. |
f64e02fe TT |
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> | |
f64e02fe TT |
46 | |
47 | #include "ext4.h" | |
48 | ||
22cfe4b4 EB |
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, | |
68e45330 | 59 | STEP_MAX, |
22cfe4b4 EB |
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) | |
c9c7429c | 70 | { |
f2b229a8 | 71 | struct folio_iter fi; |
22cfe4b4 | 72 | |
f8174a11 MWO |
73 | bio_for_each_folio_all(fi, bio) |
74 | folio_end_read(fi.folio, bio->bi_status == 0); | |
22cfe4b4 EB |
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); | |
14db0b3c | 86 | struct bio *bio = ctx->bio; |
22cfe4b4 | 87 | |
14db0b3c EB |
88 | if (fscrypt_decrypt_bio(bio)) |
89 | bio_post_read_processing(ctx); | |
90 | else | |
91 | __read_end_io(bio); | |
22cfe4b4 EB |
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); | |
68e45330 | 98 | struct bio *bio = ctx->bio; |
22cfe4b4 | 99 | |
68e45330 | 100 | /* |
704528d8 | 101 | * fsverity_verify_bio() may call readahead() again, and although verity |
68e45330 EB |
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); | |
22cfe4b4 EB |
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++; | |
70d7ced2 | 131 | fallthrough; |
22cfe4b4 EB |
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++; | |
70d7ced2 | 139 | fallthrough; |
22cfe4b4 EB |
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; | |
c9c7429c MH |
148 | } |
149 | ||
f64e02fe TT |
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 | |
2c69e205 | 155 | * back to block_read_full_folio(). |
f64e02fe TT |
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 | */ | |
4246a0b6 | 162 | static void mpage_end_io(struct bio *bio) |
f64e02fe | 163 | { |
22cfe4b4 EB |
164 | if (bio_post_read_required(bio)) { |
165 | struct bio_post_read_ctx *ctx = bio->bi_private; | |
f64e02fe | 166 | |
22cfe4b4 EB |
167 | ctx->cur_step = STEP_INITIAL; |
168 | bio_post_read_processing(ctx); | |
169 | return; | |
c9c7429c | 170 | } |
22cfe4b4 EB |
171 | __read_end_io(bio); |
172 | } | |
f64e02fe | 173 | |
22cfe4b4 EB |
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 | ||
fd5fe253 EB |
180 | static void ext4_set_bio_post_read_ctx(struct bio *bio, |
181 | const struct inode *inode, | |
182 | pgoff_t first_idx) | |
22cfe4b4 EB |
183 | { |
184 | unsigned int post_read_steps = 0; | |
22cfe4b4 | 185 | |
4f74d15f | 186 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) |
22cfe4b4 EB |
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) { | |
fd5fe253 EB |
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 | ||
22cfe4b4 EB |
197 | ctx->bio = bio; |
198 | ctx->enabled_steps = post_read_steps; | |
199 | bio->bi_private = ctx; | |
f64e02fe | 200 | } |
22cfe4b4 | 201 | } |
f64e02fe | 202 | |
22cfe4b4 EB |
203 | static inline loff_t ext4_readpage_limit(struct inode *inode) |
204 | { | |
5e122148 | 205 | if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) |
22cfe4b4 EB |
206 | return inode->i_sb->s_maxbytes; |
207 | ||
208 | return i_size_read(inode); | |
f64e02fe TT |
209 | } |
210 | ||
a07f624b | 211 | int ext4_mpage_readpages(struct inode *inode, |
c0be8e6f | 212 | struct readahead_control *rac, struct folio *folio) |
f64e02fe TT |
213 | { |
214 | struct bio *bio = NULL; | |
f64e02fe TT |
215 | sector_t last_block_in_bio = 0; |
216 | ||
f64e02fe | 217 | const unsigned blkbits = inode->i_blkbits; |
09cbfeaf | 218 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
f64e02fe | 219 | const unsigned blocksize = 1 << blkbits; |
4f74d15f | 220 | sector_t next_block; |
f64e02fe TT |
221 | sector_t block_in_file; |
222 | sector_t last_block; | |
223 | sector_t last_block_in_file; | |
e37c9e17 | 224 | sector_t first_block; |
f64e02fe TT |
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; | |
6311f91f | 230 | unsigned int nr_pages = rac ? readahead_count(rac) : 1; |
f64e02fe TT |
231 | |
232 | map.m_pblk = 0; | |
233 | map.m_lblk = 0; | |
234 | map.m_len = 0; | |
235 | map.m_flags = 0; | |
236 | ||
de9e9181 | 237 | for (; nr_pages; nr_pages--) { |
f64e02fe TT |
238 | int fully_mapped = 1; |
239 | unsigned first_hole = blocks_per_page; | |
240 | ||
c0be8e6f MW |
241 | if (rac) |
242 | folio = readahead_folio(rac); | |
243 | prefetchw(&folio->flags); | |
f64e02fe | 244 | |
c0be8e6f | 245 | if (folio_buffers(folio)) |
f64e02fe TT |
246 | goto confused; |
247 | ||
4f74d15f | 248 | block_in_file = next_block = |
c0be8e6f | 249 | (sector_t)folio->index << (PAGE_SHIFT - blkbits); |
f64e02fe | 250 | last_block = block_in_file + nr_pages * blocks_per_page; |
22cfe4b4 EB |
251 | last_block_in_file = (ext4_readpage_limit(inode) + |
252 | blocksize - 1) >> blkbits; | |
f64e02fe TT |
253 | if (last_block > last_block_in_file) |
254 | last_block = last_block_in_file; | |
255 | page_block = 0; | |
256 | ||
257 | /* | |
258 | * Map blocks using the previous result first. | |
259 | */ | |
260 | if ((map.m_flags & EXT4_MAP_MAPPED) && | |
261 | block_in_file > map.m_lblk && | |
262 | block_in_file < (map.m_lblk + map.m_len)) { | |
263 | unsigned map_offset = block_in_file - map.m_lblk; | |
264 | unsigned last = map.m_len - map_offset; | |
265 | ||
e37c9e17 | 266 | first_block = map.m_pblk + map_offset; |
f64e02fe TT |
267 | for (relative_block = 0; ; relative_block++) { |
268 | if (relative_block == last) { | |
269 | /* needed? */ | |
270 | map.m_flags &= ~EXT4_MAP_MAPPED; | |
271 | break; | |
272 | } | |
273 | if (page_block == blocks_per_page) | |
274 | break; | |
f64e02fe TT |
275 | page_block++; |
276 | block_in_file++; | |
277 | } | |
278 | } | |
279 | ||
280 | /* | |
281 | * Then do more ext4_map_blocks() calls until we are | |
c0be8e6f | 282 | * done with this folio. |
f64e02fe TT |
283 | */ |
284 | while (page_block < blocks_per_page) { | |
285 | if (block_in_file < last_block) { | |
286 | map.m_lblk = block_in_file; | |
287 | map.m_len = last_block - block_in_file; | |
288 | ||
289 | if (ext4_map_blocks(NULL, inode, &map, 0) < 0) { | |
290 | set_error_page: | |
c0be8e6f MW |
291 | folio_zero_segment(folio, 0, |
292 | folio_size(folio)); | |
293 | folio_unlock(folio); | |
f64e02fe TT |
294 | goto next_page; |
295 | } | |
296 | } | |
297 | if ((map.m_flags & EXT4_MAP_MAPPED) == 0) { | |
298 | fully_mapped = 0; | |
299 | if (first_hole == blocks_per_page) | |
300 | first_hole = page_block; | |
301 | page_block++; | |
302 | block_in_file++; | |
303 | continue; | |
304 | } | |
305 | if (first_hole != blocks_per_page) | |
306 | goto confused; /* hole -> non-hole */ | |
307 | ||
308 | /* Contiguous blocks? */ | |
e37c9e17 MWO |
309 | if (!page_block) |
310 | first_block = map.m_pblk; | |
311 | else if (first_block + page_block != map.m_pblk) | |
f64e02fe TT |
312 | goto confused; |
313 | for (relative_block = 0; ; relative_block++) { | |
314 | if (relative_block == map.m_len) { | |
315 | /* needed? */ | |
316 | map.m_flags &= ~EXT4_MAP_MAPPED; | |
317 | break; | |
318 | } else if (page_block == blocks_per_page) | |
319 | break; | |
f64e02fe TT |
320 | page_block++; |
321 | block_in_file++; | |
322 | } | |
323 | } | |
324 | if (first_hole != blocks_per_page) { | |
c0be8e6f MW |
325 | folio_zero_segment(folio, first_hole << blkbits, |
326 | folio_size(folio)); | |
f64e02fe | 327 | if (first_hole == 0) { |
c0be8e6f | 328 | if (ext4_need_verity(inode, folio->index) && |
0dea40aa | 329 | !fsverity_verify_folio(folio)) |
22cfe4b4 | 330 | goto set_error_page; |
f8174a11 | 331 | folio_end_read(folio, true); |
c0be8e6f | 332 | continue; |
f64e02fe TT |
333 | } |
334 | } else if (fully_mapped) { | |
c0be8e6f | 335 | folio_set_mappedtodisk(folio); |
f64e02fe | 336 | } |
f64e02fe TT |
337 | |
338 | /* | |
c0be8e6f | 339 | * This folio will go to BIO. Do we need to send this |
f64e02fe TT |
340 | * BIO off first? |
341 | */ | |
e37c9e17 | 342 | if (bio && (last_block_in_bio != first_block - 1 || |
4f74d15f | 343 | !fscrypt_mergeable_bio(bio, inode, next_block))) { |
f64e02fe | 344 | submit_and_realloc: |
4e49ea4a | 345 | submit_bio(bio); |
f64e02fe TT |
346 | bio = NULL; |
347 | } | |
348 | if (bio == NULL) { | |
5500221e GX |
349 | /* |
350 | * bio_alloc will _always_ be able to allocate a bio if | |
351 | * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset(). | |
352 | */ | |
07888c66 CH |
353 | bio = bio_alloc(bdev, bio_max_segs(nr_pages), |
354 | REQ_OP_READ, GFP_KERNEL); | |
4f74d15f EB |
355 | fscrypt_set_bio_crypt_ctx(bio, inode, next_block, |
356 | GFP_KERNEL); | |
c0be8e6f | 357 | ext4_set_bio_post_read_ctx(bio, inode, folio->index); |
e37c9e17 | 358 | bio->bi_iter.bi_sector = first_block << (blkbits - 9); |
f64e02fe | 359 | bio->bi_end_io = mpage_end_io; |
07888c66 CH |
360 | if (rac) |
361 | bio->bi_opf |= REQ_RAHEAD; | |
f64e02fe TT |
362 | } |
363 | ||
364 | length = first_hole << blkbits; | |
c0be8e6f | 365 | if (!bio_add_folio(bio, folio, length, 0)) |
f64e02fe TT |
366 | goto submit_and_realloc; |
367 | ||
368 | if (((map.m_flags & EXT4_MAP_BOUNDARY) && | |
369 | (relative_block == map.m_len)) || | |
370 | (first_hole != blocks_per_page)) { | |
4e49ea4a | 371 | submit_bio(bio); |
f64e02fe TT |
372 | bio = NULL; |
373 | } else | |
e37c9e17 | 374 | last_block_in_bio = first_block + blocks_per_page - 1; |
c0be8e6f | 375 | continue; |
f64e02fe TT |
376 | confused: |
377 | if (bio) { | |
4e49ea4a | 378 | submit_bio(bio); |
f64e02fe TT |
379 | bio = NULL; |
380 | } | |
c0be8e6f MW |
381 | if (!folio_test_uptodate(folio)) |
382 | block_read_full_folio(folio, ext4_get_block); | |
f64e02fe | 383 | else |
c0be8e6f MW |
384 | folio_unlock(folio); |
385 | next_page: | |
386 | ; /* A label shall be followed by a statement until C23 */ | |
f64e02fe | 387 | } |
f64e02fe | 388 | if (bio) |
4e49ea4a | 389 | submit_bio(bio); |
f64e02fe TT |
390 | return 0; |
391 | } | |
22cfe4b4 EB |
392 | |
393 | int __init ext4_init_post_read_processing(void) | |
394 | { | |
060f7739 JS |
395 | bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT); |
396 | ||
22cfe4b4 EB |
397 | if (!bio_post_read_ctx_cache) |
398 | goto fail; | |
399 | bio_post_read_ctx_pool = | |
400 | mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, | |
401 | bio_post_read_ctx_cache); | |
402 | if (!bio_post_read_ctx_pool) | |
403 | goto fail_free_cache; | |
404 | return 0; | |
405 | ||
406 | fail_free_cache: | |
407 | kmem_cache_destroy(bio_post_read_ctx_cache); | |
408 | fail: | |
409 | return -ENOMEM; | |
410 | } | |
411 | ||
412 | void ext4_exit_post_read_processing(void) | |
413 | { | |
414 | mempool_destroy(bio_post_read_ctx_pool); | |
415 | kmem_cache_destroy(bio_post_read_ctx_cache); | |
416 | } |