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 | ||
78 | /* PG_error was set if any post_read step failed */ | |
79 | if (bio->bi_status || PageError(page)) { | |
80 | ClearPageUptodate(page); | |
81 | /* will re-read again later */ | |
82 | ClearPageError(page); | |
83 | } else { | |
84 | SetPageUptodate(page); | |
85 | } | |
86 | unlock_page(page); | |
87 | } | |
88 | if (bio->bi_private) | |
89 | mempool_free(bio->bi_private, bio_post_read_ctx_pool); | |
90 | bio_put(bio); | |
91 | } | |
92 | ||
93 | static void bio_post_read_processing(struct bio_post_read_ctx *ctx); | |
94 | ||
95 | static void decrypt_work(struct work_struct *work) | |
96 | { | |
97 | struct bio_post_read_ctx *ctx = | |
98 | container_of(work, struct bio_post_read_ctx, work); | |
99 | ||
100 | fscrypt_decrypt_bio(ctx->bio); | |
101 | ||
102 | bio_post_read_processing(ctx); | |
103 | } | |
104 | ||
105 | static void verity_work(struct work_struct *work) | |
106 | { | |
107 | struct bio_post_read_ctx *ctx = | |
108 | container_of(work, struct bio_post_read_ctx, work); | |
68e45330 | 109 | struct bio *bio = ctx->bio; |
22cfe4b4 | 110 | |
68e45330 EB |
111 | /* |
112 | * fsverity_verify_bio() may call readpages() again, and although verity | |
113 | * will be disabled for that, decryption may still be needed, causing | |
114 | * another bio_post_read_ctx to be allocated. So to guarantee that | |
115 | * mempool_alloc() never deadlocks we must free the current ctx first. | |
116 | * This is safe because verity is the last post-read step. | |
117 | */ | |
118 | BUILD_BUG_ON(STEP_VERITY + 1 != STEP_MAX); | |
119 | mempool_free(ctx, bio_post_read_ctx_pool); | |
120 | bio->bi_private = NULL; | |
121 | ||
122 | fsverity_verify_bio(bio); | |
123 | ||
124 | __read_end_io(bio); | |
22cfe4b4 EB |
125 | } |
126 | ||
127 | static void bio_post_read_processing(struct bio_post_read_ctx *ctx) | |
128 | { | |
129 | /* | |
130 | * We use different work queues for decryption and for verity because | |
131 | * verity may require reading metadata pages that need decryption, and | |
132 | * we shouldn't recurse to the same workqueue. | |
133 | */ | |
134 | switch (++ctx->cur_step) { | |
135 | case STEP_DECRYPT: | |
136 | if (ctx->enabled_steps & (1 << STEP_DECRYPT)) { | |
137 | INIT_WORK(&ctx->work, decrypt_work); | |
138 | fscrypt_enqueue_decrypt_work(&ctx->work); | |
139 | return; | |
140 | } | |
141 | ctx->cur_step++; | |
70d7ced2 | 142 | fallthrough; |
22cfe4b4 EB |
143 | case STEP_VERITY: |
144 | if (ctx->enabled_steps & (1 << STEP_VERITY)) { | |
145 | INIT_WORK(&ctx->work, verity_work); | |
146 | fsverity_enqueue_verify_work(&ctx->work); | |
147 | return; | |
148 | } | |
149 | ctx->cur_step++; | |
70d7ced2 | 150 | fallthrough; |
22cfe4b4 EB |
151 | default: |
152 | __read_end_io(ctx->bio); | |
153 | } | |
154 | } | |
155 | ||
156 | static bool bio_post_read_required(struct bio *bio) | |
157 | { | |
158 | return bio->bi_private && !bio->bi_status; | |
c9c7429c MH |
159 | } |
160 | ||
f64e02fe TT |
161 | /* |
162 | * I/O completion handler for multipage BIOs. | |
163 | * | |
164 | * The mpage code never puts partial pages into a BIO (except for end-of-file). | |
165 | * If a page does not map to a contiguous run of blocks then it simply falls | |
166 | * back to block_read_full_page(). | |
167 | * | |
168 | * Why is this? If a page's completion depends on a number of different BIOs | |
169 | * which can complete in any order (or at the same time) then determining the | |
170 | * status of that page is hard. See end_buffer_async_read() for the details. | |
171 | * There is no point in duplicating all that complexity. | |
172 | */ | |
4246a0b6 | 173 | static void mpage_end_io(struct bio *bio) |
f64e02fe | 174 | { |
22cfe4b4 EB |
175 | if (bio_post_read_required(bio)) { |
176 | struct bio_post_read_ctx *ctx = bio->bi_private; | |
f64e02fe | 177 | |
22cfe4b4 EB |
178 | ctx->cur_step = STEP_INITIAL; |
179 | bio_post_read_processing(ctx); | |
180 | return; | |
c9c7429c | 181 | } |
22cfe4b4 EB |
182 | __read_end_io(bio); |
183 | } | |
f64e02fe | 184 | |
22cfe4b4 EB |
185 | static inline bool ext4_need_verity(const struct inode *inode, pgoff_t idx) |
186 | { | |
187 | return fsverity_active(inode) && | |
188 | idx < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); | |
189 | } | |
190 | ||
fd5fe253 EB |
191 | static void ext4_set_bio_post_read_ctx(struct bio *bio, |
192 | const struct inode *inode, | |
193 | pgoff_t first_idx) | |
22cfe4b4 EB |
194 | { |
195 | unsigned int post_read_steps = 0; | |
22cfe4b4 | 196 | |
4f74d15f | 197 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) |
22cfe4b4 EB |
198 | post_read_steps |= 1 << STEP_DECRYPT; |
199 | ||
200 | if (ext4_need_verity(inode, first_idx)) | |
201 | post_read_steps |= 1 << STEP_VERITY; | |
202 | ||
203 | if (post_read_steps) { | |
fd5fe253 EB |
204 | /* Due to the mempool, this never fails. */ |
205 | struct bio_post_read_ctx *ctx = | |
206 | mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS); | |
207 | ||
22cfe4b4 EB |
208 | ctx->bio = bio; |
209 | ctx->enabled_steps = post_read_steps; | |
210 | bio->bi_private = ctx; | |
f64e02fe | 211 | } |
22cfe4b4 | 212 | } |
f64e02fe | 213 | |
22cfe4b4 EB |
214 | static inline loff_t ext4_readpage_limit(struct inode *inode) |
215 | { | |
216 | if (IS_ENABLED(CONFIG_FS_VERITY) && | |
217 | (IS_VERITY(inode) || ext4_verity_in_progress(inode))) | |
218 | return inode->i_sb->s_maxbytes; | |
219 | ||
220 | return i_size_read(inode); | |
f64e02fe TT |
221 | } |
222 | ||
a07f624b | 223 | int ext4_mpage_readpages(struct inode *inode, |
6311f91f | 224 | struct readahead_control *rac, struct page *page) |
f64e02fe TT |
225 | { |
226 | struct bio *bio = NULL; | |
f64e02fe TT |
227 | sector_t last_block_in_bio = 0; |
228 | ||
f64e02fe | 229 | const unsigned blkbits = inode->i_blkbits; |
09cbfeaf | 230 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
f64e02fe | 231 | const unsigned blocksize = 1 << blkbits; |
4f74d15f | 232 | sector_t next_block; |
f64e02fe TT |
233 | sector_t block_in_file; |
234 | sector_t last_block; | |
235 | sector_t last_block_in_file; | |
236 | sector_t blocks[MAX_BUF_PER_PAGE]; | |
237 | unsigned page_block; | |
238 | struct block_device *bdev = inode->i_sb->s_bdev; | |
239 | int length; | |
240 | unsigned relative_block = 0; | |
241 | struct ext4_map_blocks map; | |
6311f91f | 242 | unsigned int nr_pages = rac ? readahead_count(rac) : 1; |
f64e02fe TT |
243 | |
244 | map.m_pblk = 0; | |
245 | map.m_lblk = 0; | |
246 | map.m_len = 0; | |
247 | map.m_flags = 0; | |
248 | ||
de9e9181 | 249 | for (; nr_pages; nr_pages--) { |
f64e02fe TT |
250 | int fully_mapped = 1; |
251 | unsigned first_hole = blocks_per_page; | |
252 | ||
6311f91f MWO |
253 | if (rac) { |
254 | page = readahead_page(rac); | |
d454a273 | 255 | prefetchw(&page->flags); |
f64e02fe TT |
256 | } |
257 | ||
258 | if (page_has_buffers(page)) | |
259 | goto confused; | |
260 | ||
4f74d15f EB |
261 | block_in_file = next_block = |
262 | (sector_t)page->index << (PAGE_SHIFT - blkbits); | |
f64e02fe | 263 | last_block = block_in_file + nr_pages * blocks_per_page; |
22cfe4b4 EB |
264 | last_block_in_file = (ext4_readpage_limit(inode) + |
265 | blocksize - 1) >> blkbits; | |
f64e02fe TT |
266 | if (last_block > last_block_in_file) |
267 | last_block = last_block_in_file; | |
268 | page_block = 0; | |
269 | ||
270 | /* | |
271 | * Map blocks using the previous result first. | |
272 | */ | |
273 | if ((map.m_flags & EXT4_MAP_MAPPED) && | |
274 | block_in_file > map.m_lblk && | |
275 | block_in_file < (map.m_lblk + map.m_len)) { | |
276 | unsigned map_offset = block_in_file - map.m_lblk; | |
277 | unsigned last = map.m_len - map_offset; | |
278 | ||
279 | for (relative_block = 0; ; relative_block++) { | |
280 | if (relative_block == last) { | |
281 | /* needed? */ | |
282 | map.m_flags &= ~EXT4_MAP_MAPPED; | |
283 | break; | |
284 | } | |
285 | if (page_block == blocks_per_page) | |
286 | break; | |
287 | blocks[page_block] = map.m_pblk + map_offset + | |
288 | relative_block; | |
289 | page_block++; | |
290 | block_in_file++; | |
291 | } | |
292 | } | |
293 | ||
294 | /* | |
295 | * Then do more ext4_map_blocks() calls until we are | |
296 | * done with this page. | |
297 | */ | |
298 | while (page_block < blocks_per_page) { | |
299 | if (block_in_file < last_block) { | |
300 | map.m_lblk = block_in_file; | |
301 | map.m_len = last_block - block_in_file; | |
302 | ||
303 | if (ext4_map_blocks(NULL, inode, &map, 0) < 0) { | |
304 | set_error_page: | |
305 | SetPageError(page); | |
306 | zero_user_segment(page, 0, | |
09cbfeaf | 307 | PAGE_SIZE); |
f64e02fe TT |
308 | unlock_page(page); |
309 | goto next_page; | |
310 | } | |
311 | } | |
312 | if ((map.m_flags & EXT4_MAP_MAPPED) == 0) { | |
313 | fully_mapped = 0; | |
314 | if (first_hole == blocks_per_page) | |
315 | first_hole = page_block; | |
316 | page_block++; | |
317 | block_in_file++; | |
318 | continue; | |
319 | } | |
320 | if (first_hole != blocks_per_page) | |
321 | goto confused; /* hole -> non-hole */ | |
322 | ||
323 | /* Contiguous blocks? */ | |
324 | if (page_block && blocks[page_block-1] != map.m_pblk-1) | |
325 | goto confused; | |
326 | for (relative_block = 0; ; relative_block++) { | |
327 | if (relative_block == map.m_len) { | |
328 | /* needed? */ | |
329 | map.m_flags &= ~EXT4_MAP_MAPPED; | |
330 | break; | |
331 | } else if (page_block == blocks_per_page) | |
332 | break; | |
333 | blocks[page_block] = map.m_pblk+relative_block; | |
334 | page_block++; | |
335 | block_in_file++; | |
336 | } | |
337 | } | |
338 | if (first_hole != blocks_per_page) { | |
339 | zero_user_segment(page, first_hole << blkbits, | |
09cbfeaf | 340 | PAGE_SIZE); |
f64e02fe | 341 | if (first_hole == 0) { |
22cfe4b4 EB |
342 | if (ext4_need_verity(inode, page->index) && |
343 | !fsverity_verify_page(page)) | |
344 | goto set_error_page; | |
f64e02fe TT |
345 | SetPageUptodate(page); |
346 | unlock_page(page); | |
347 | goto next_page; | |
348 | } | |
349 | } else if (fully_mapped) { | |
350 | SetPageMappedToDisk(page); | |
351 | } | |
f64e02fe TT |
352 | |
353 | /* | |
354 | * This page will go to BIO. Do we need to send this | |
355 | * BIO off first? | |
356 | */ | |
4f74d15f EB |
357 | if (bio && (last_block_in_bio != blocks[0] - 1 || |
358 | !fscrypt_mergeable_bio(bio, inode, next_block))) { | |
f64e02fe | 359 | submit_and_realloc: |
4e49ea4a | 360 | submit_bio(bio); |
f64e02fe TT |
361 | bio = NULL; |
362 | } | |
363 | if (bio == NULL) { | |
5500221e GX |
364 | /* |
365 | * bio_alloc will _always_ be able to allocate a bio if | |
366 | * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset(). | |
367 | */ | |
5f7136db | 368 | bio = bio_alloc(GFP_KERNEL, bio_max_segs(nr_pages)); |
4f74d15f EB |
369 | fscrypt_set_bio_crypt_ctx(bio, inode, next_block, |
370 | GFP_KERNEL); | |
fd5fe253 | 371 | ext4_set_bio_post_read_ctx(bio, inode, page->index); |
74d46992 | 372 | bio_set_dev(bio, bdev); |
f64e02fe TT |
373 | bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); |
374 | bio->bi_end_io = mpage_end_io; | |
ac22b46a | 375 | bio_set_op_attrs(bio, REQ_OP_READ, |
6311f91f | 376 | rac ? REQ_RAHEAD : 0); |
f64e02fe TT |
377 | } |
378 | ||
379 | length = first_hole << blkbits; | |
380 | if (bio_add_page(bio, page, length, 0) < length) | |
381 | goto submit_and_realloc; | |
382 | ||
383 | if (((map.m_flags & EXT4_MAP_BOUNDARY) && | |
384 | (relative_block == map.m_len)) || | |
385 | (first_hole != blocks_per_page)) { | |
4e49ea4a | 386 | submit_bio(bio); |
f64e02fe TT |
387 | bio = NULL; |
388 | } else | |
389 | last_block_in_bio = blocks[blocks_per_page - 1]; | |
390 | goto next_page; | |
391 | confused: | |
392 | if (bio) { | |
4e49ea4a | 393 | submit_bio(bio); |
f64e02fe TT |
394 | bio = NULL; |
395 | } | |
396 | if (!PageUptodate(page)) | |
397 | block_read_full_page(page, ext4_get_block); | |
398 | else | |
399 | unlock_page(page); | |
400 | next_page: | |
6311f91f | 401 | if (rac) |
09cbfeaf | 402 | put_page(page); |
f64e02fe | 403 | } |
f64e02fe | 404 | if (bio) |
4e49ea4a | 405 | submit_bio(bio); |
f64e02fe TT |
406 | return 0; |
407 | } | |
22cfe4b4 EB |
408 | |
409 | int __init ext4_init_post_read_processing(void) | |
410 | { | |
411 | bio_post_read_ctx_cache = | |
412 | kmem_cache_create("ext4_bio_post_read_ctx", | |
413 | sizeof(struct bio_post_read_ctx), 0, 0, NULL); | |
414 | if (!bio_post_read_ctx_cache) | |
415 | goto fail; | |
416 | bio_post_read_ctx_pool = | |
417 | mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS, | |
418 | bio_post_read_ctx_cache); | |
419 | if (!bio_post_read_ctx_pool) | |
420 | goto fail_free_cache; | |
421 | return 0; | |
422 | ||
423 | fail_free_cache: | |
424 | kmem_cache_destroy(bio_post_read_ctx_cache); | |
425 | fail: | |
426 | return -ENOMEM; | |
427 | } | |
428 | ||
429 | void ext4_exit_post_read_processing(void) | |
430 | { | |
431 | mempool_destroy(bio_post_read_ctx_pool); | |
432 | kmem_cache_destroy(bio_post_read_ctx_cache); | |
433 | } |