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 | { |
22cfe4b4 EB |
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 | ||
98dc08ba | 78 | if (bio->bi_status) |
22cfe4b4 | 79 | ClearPageUptodate(page); |
98dc08ba | 80 | else |
22cfe4b4 | 81 | SetPageUptodate(page); |
22cfe4b4 EB |
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); | |
14db0b3c | 95 | struct bio *bio = ctx->bio; |
22cfe4b4 | 96 | |
14db0b3c EB |
97 | if (fscrypt_decrypt_bio(bio)) |
98 | bio_post_read_processing(ctx); | |
99 | else | |
100 | __read_end_io(bio); | |
22cfe4b4 EB |
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); | |
68e45330 | 107 | struct bio *bio = ctx->bio; |
22cfe4b4 | 108 | |
68e45330 | 109 | /* |
704528d8 | 110 | * fsverity_verify_bio() may call readahead() again, and although verity |
68e45330 EB |
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); | |
22cfe4b4 EB |
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++; | |
70d7ced2 | 140 | fallthrough; |
22cfe4b4 EB |
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++; | |
70d7ced2 | 148 | fallthrough; |
22cfe4b4 EB |
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; | |
c9c7429c MH |
157 | } |
158 | ||
f64e02fe TT |
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 | |
2c69e205 | 164 | * back to block_read_full_folio(). |
f64e02fe TT |
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 | */ | |
4246a0b6 | 171 | static void mpage_end_io(struct bio *bio) |
f64e02fe | 172 | { |
22cfe4b4 EB |
173 | if (bio_post_read_required(bio)) { |
174 | struct bio_post_read_ctx *ctx = bio->bi_private; | |
f64e02fe | 175 | |
22cfe4b4 EB |
176 | ctx->cur_step = STEP_INITIAL; |
177 | bio_post_read_processing(ctx); | |
178 | return; | |
c9c7429c | 179 | } |
22cfe4b4 EB |
180 | __read_end_io(bio); |
181 | } | |
f64e02fe | 182 | |
22cfe4b4 EB |
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 | ||
fd5fe253 EB |
189 | static void ext4_set_bio_post_read_ctx(struct bio *bio, |
190 | const struct inode *inode, | |
191 | pgoff_t first_idx) | |
22cfe4b4 EB |
192 | { |
193 | unsigned int post_read_steps = 0; | |
22cfe4b4 | 194 | |
4f74d15f | 195 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) |
22cfe4b4 EB |
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) { | |
fd5fe253 EB |
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 | ||
22cfe4b4 EB |
206 | ctx->bio = bio; |
207 | ctx->enabled_steps = post_read_steps; | |
208 | bio->bi_private = ctx; | |
f64e02fe | 209 | } |
22cfe4b4 | 210 | } |
f64e02fe | 211 | |
22cfe4b4 EB |
212 | static inline loff_t ext4_readpage_limit(struct inode *inode) |
213 | { | |
214 | if (IS_ENABLED(CONFIG_FS_VERITY) && | |
215 | (IS_VERITY(inode) || ext4_verity_in_progress(inode))) | |
216 | return inode->i_sb->s_maxbytes; | |
217 | ||
218 | return i_size_read(inode); | |
f64e02fe TT |
219 | } |
220 | ||
a07f624b | 221 | int ext4_mpage_readpages(struct inode *inode, |
6311f91f | 222 | struct readahead_control *rac, struct page *page) |
f64e02fe TT |
223 | { |
224 | struct bio *bio = NULL; | |
f64e02fe TT |
225 | sector_t last_block_in_bio = 0; |
226 | ||
f64e02fe | 227 | const unsigned blkbits = inode->i_blkbits; |
09cbfeaf | 228 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
f64e02fe | 229 | const unsigned blocksize = 1 << blkbits; |
4f74d15f | 230 | sector_t next_block; |
f64e02fe TT |
231 | sector_t block_in_file; |
232 | sector_t last_block; | |
233 | sector_t last_block_in_file; | |
234 | sector_t blocks[MAX_BUF_PER_PAGE]; | |
235 | unsigned page_block; | |
236 | struct block_device *bdev = inode->i_sb->s_bdev; | |
237 | int length; | |
238 | unsigned relative_block = 0; | |
239 | struct ext4_map_blocks map; | |
6311f91f | 240 | unsigned int nr_pages = rac ? readahead_count(rac) : 1; |
f64e02fe TT |
241 | |
242 | map.m_pblk = 0; | |
243 | map.m_lblk = 0; | |
244 | map.m_len = 0; | |
245 | map.m_flags = 0; | |
246 | ||
de9e9181 | 247 | for (; nr_pages; nr_pages--) { |
f64e02fe TT |
248 | int fully_mapped = 1; |
249 | unsigned first_hole = blocks_per_page; | |
250 | ||
6311f91f MWO |
251 | if (rac) { |
252 | page = readahead_page(rac); | |
d454a273 | 253 | prefetchw(&page->flags); |
f64e02fe TT |
254 | } |
255 | ||
256 | if (page_has_buffers(page)) | |
257 | goto confused; | |
258 | ||
4f74d15f EB |
259 | block_in_file = next_block = |
260 | (sector_t)page->index << (PAGE_SHIFT - blkbits); | |
f64e02fe | 261 | last_block = block_in_file + nr_pages * blocks_per_page; |
22cfe4b4 EB |
262 | last_block_in_file = (ext4_readpage_limit(inode) + |
263 | blocksize - 1) >> blkbits; | |
f64e02fe TT |
264 | if (last_block > last_block_in_file) |
265 | last_block = last_block_in_file; | |
266 | page_block = 0; | |
267 | ||
268 | /* | |
269 | * Map blocks using the previous result first. | |
270 | */ | |
271 | if ((map.m_flags & EXT4_MAP_MAPPED) && | |
272 | block_in_file > map.m_lblk && | |
273 | block_in_file < (map.m_lblk + map.m_len)) { | |
274 | unsigned map_offset = block_in_file - map.m_lblk; | |
275 | unsigned last = map.m_len - map_offset; | |
276 | ||
277 | for (relative_block = 0; ; relative_block++) { | |
278 | if (relative_block == last) { | |
279 | /* needed? */ | |
280 | map.m_flags &= ~EXT4_MAP_MAPPED; | |
281 | break; | |
282 | } | |
283 | if (page_block == blocks_per_page) | |
284 | break; | |
285 | blocks[page_block] = map.m_pblk + map_offset + | |
286 | relative_block; | |
287 | page_block++; | |
288 | block_in_file++; | |
289 | } | |
290 | } | |
291 | ||
292 | /* | |
293 | * Then do more ext4_map_blocks() calls until we are | |
294 | * done with this page. | |
295 | */ | |
296 | while (page_block < blocks_per_page) { | |
297 | if (block_in_file < last_block) { | |
298 | map.m_lblk = block_in_file; | |
299 | map.m_len = last_block - block_in_file; | |
300 | ||
301 | if (ext4_map_blocks(NULL, inode, &map, 0) < 0) { | |
302 | set_error_page: | |
303 | SetPageError(page); | |
304 | zero_user_segment(page, 0, | |
09cbfeaf | 305 | PAGE_SIZE); |
f64e02fe TT |
306 | unlock_page(page); |
307 | goto next_page; | |
308 | } | |
309 | } | |
310 | if ((map.m_flags & EXT4_MAP_MAPPED) == 0) { | |
311 | fully_mapped = 0; | |
312 | if (first_hole == blocks_per_page) | |
313 | first_hole = page_block; | |
314 | page_block++; | |
315 | block_in_file++; | |
316 | continue; | |
317 | } | |
318 | if (first_hole != blocks_per_page) | |
319 | goto confused; /* hole -> non-hole */ | |
320 | ||
321 | /* Contiguous blocks? */ | |
322 | if (page_block && blocks[page_block-1] != map.m_pblk-1) | |
323 | goto confused; | |
324 | for (relative_block = 0; ; relative_block++) { | |
325 | if (relative_block == map.m_len) { | |
326 | /* needed? */ | |
327 | map.m_flags &= ~EXT4_MAP_MAPPED; | |
328 | break; | |
329 | } else if (page_block == blocks_per_page) | |
330 | break; | |
331 | blocks[page_block] = map.m_pblk+relative_block; | |
332 | page_block++; | |
333 | block_in_file++; | |
334 | } | |
335 | } | |
336 | if (first_hole != blocks_per_page) { | |
337 | zero_user_segment(page, first_hole << blkbits, | |
09cbfeaf | 338 | PAGE_SIZE); |
f64e02fe | 339 | if (first_hole == 0) { |
22cfe4b4 EB |
340 | if (ext4_need_verity(inode, page->index) && |
341 | !fsverity_verify_page(page)) | |
342 | goto set_error_page; | |
f64e02fe TT |
343 | SetPageUptodate(page); |
344 | unlock_page(page); | |
345 | goto next_page; | |
346 | } | |
347 | } else if (fully_mapped) { | |
348 | SetPageMappedToDisk(page); | |
349 | } | |
f64e02fe TT |
350 | |
351 | /* | |
352 | * This page will go to BIO. Do we need to send this | |
353 | * BIO off first? | |
354 | */ | |
4f74d15f EB |
355 | if (bio && (last_block_in_bio != blocks[0] - 1 || |
356 | !fscrypt_mergeable_bio(bio, inode, next_block))) { | |
f64e02fe | 357 | submit_and_realloc: |
4e49ea4a | 358 | submit_bio(bio); |
f64e02fe TT |
359 | bio = NULL; |
360 | } | |
361 | if (bio == NULL) { | |
5500221e GX |
362 | /* |
363 | * bio_alloc will _always_ be able to allocate a bio if | |
364 | * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset(). | |
365 | */ | |
07888c66 CH |
366 | bio = bio_alloc(bdev, bio_max_segs(nr_pages), |
367 | REQ_OP_READ, GFP_KERNEL); | |
4f74d15f EB |
368 | fscrypt_set_bio_crypt_ctx(bio, inode, next_block, |
369 | GFP_KERNEL); | |
fd5fe253 | 370 | ext4_set_bio_post_read_ctx(bio, inode, page->index); |
f64e02fe TT |
371 | bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); |
372 | bio->bi_end_io = mpage_end_io; | |
07888c66 CH |
373 | if (rac) |
374 | bio->bi_opf |= REQ_RAHEAD; | |
f64e02fe TT |
375 | } |
376 | ||
377 | length = first_hole << blkbits; | |
378 | if (bio_add_page(bio, page, length, 0) < length) | |
379 | goto submit_and_realloc; | |
380 | ||
381 | if (((map.m_flags & EXT4_MAP_BOUNDARY) && | |
382 | (relative_block == map.m_len)) || | |
383 | (first_hole != blocks_per_page)) { | |
4e49ea4a | 384 | submit_bio(bio); |
f64e02fe TT |
385 | bio = NULL; |
386 | } else | |
387 | last_block_in_bio = blocks[blocks_per_page - 1]; | |
388 | goto next_page; | |
389 | confused: | |
390 | if (bio) { | |
4e49ea4a | 391 | submit_bio(bio); |
f64e02fe TT |
392 | bio = NULL; |
393 | } | |
394 | if (!PageUptodate(page)) | |
2c69e205 | 395 | block_read_full_folio(page_folio(page), ext4_get_block); |
f64e02fe TT |
396 | else |
397 | unlock_page(page); | |
398 | next_page: | |
6311f91f | 399 | if (rac) |
09cbfeaf | 400 | put_page(page); |
f64e02fe | 401 | } |
f64e02fe | 402 | if (bio) |
4e49ea4a | 403 | submit_bio(bio); |
f64e02fe TT |
404 | return 0; |
405 | } | |
22cfe4b4 EB |
406 | |
407 | int __init ext4_init_post_read_processing(void) | |
408 | { | |
060f7739 JS |
409 | bio_post_read_ctx_cache = KMEM_CACHE(bio_post_read_ctx, SLAB_RECLAIM_ACCOUNT); |
410 | ||
22cfe4b4 EB |
411 | if (!bio_post_read_ctx_cache) |
412 | goto fail; | |
413 | bio_post_read_ctx_pool = | |
414 | mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, | |
415 | bio_post_read_ctx_cache); | |
416 | if (!bio_post_read_ctx_pool) | |
417 | goto fail_free_cache; | |
418 | return 0; | |
419 | ||
420 | fail_free_cache: | |
421 | kmem_cache_destroy(bio_post_read_ctx_cache); | |
422 | fail: | |
423 | return -ENOMEM; | |
424 | } | |
425 | ||
426 | void ext4_exit_post_read_processing(void) | |
427 | { | |
428 | mempool_destroy(bio_post_read_ctx_pool); | |
429 | kmem_cache_destroy(bio_post_read_ctx_cache); | |
430 | } |