1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/readpage.c
5 * Copyright (C) 2002, Linus Torvalds.
6 * Copyright (C) 2015, Google, Inc.
8 * This was originally taken from fs/mpage.c
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.
16 * This will allow us to attach a callback function to support ext4
19 * If anything unusual happens, such as:
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
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.
31 #include <linux/kernel.h>
32 #include <linux/export.h>
34 #include <linux/kdev_t.h>
35 #include <linux/gfp.h>
36 #include <linux/bio.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>
49 #define NUM_PREALLOC_POST_READ_CTXS 128
51 static struct kmem_cache
*bio_post_read_ctx_cache
;
52 static mempool_t
*bio_post_read_ctx_pool
;
54 /* postprocessing steps for read bios */
55 enum bio_post_read_step
{
62 struct bio_post_read_ctx
{
64 struct work_struct work
;
65 unsigned int cur_step
;
66 unsigned int enabled_steps
;
69 static void __read_end_io(struct bio
*bio
)
73 bio_for_each_folio_all(fi
, bio
)
74 folio_end_read(fi
.folio
, bio
->bi_status
== 0);
76 mempool_free(bio
->bi_private
, bio_post_read_ctx_pool
);
80 static void bio_post_read_processing(struct bio_post_read_ctx
*ctx
);
82 static void decrypt_work(struct work_struct
*work
)
84 struct bio_post_read_ctx
*ctx
=
85 container_of(work
, struct bio_post_read_ctx
, work
);
86 struct bio
*bio
= ctx
->bio
;
88 if (fscrypt_decrypt_bio(bio
))
89 bio_post_read_processing(ctx
);
94 static void verity_work(struct work_struct
*work
)
96 struct bio_post_read_ctx
*ctx
=
97 container_of(work
, struct bio_post_read_ctx
, work
);
98 struct bio
*bio
= ctx
->bio
;
101 * fsverity_verify_bio() may call readahead() again, and although verity
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.
107 BUILD_BUG_ON(STEP_VERITY
+ 1 != STEP_MAX
);
108 mempool_free(ctx
, bio_post_read_ctx_pool
);
109 bio
->bi_private
= NULL
;
111 fsverity_verify_bio(bio
);
116 static void bio_post_read_processing(struct bio_post_read_ctx
*ctx
)
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.
123 switch (++ctx
->cur_step
) {
125 if (ctx
->enabled_steps
& (1 << STEP_DECRYPT
)) {
126 INIT_WORK(&ctx
->work
, decrypt_work
);
127 fscrypt_enqueue_decrypt_work(&ctx
->work
);
133 if (ctx
->enabled_steps
& (1 << STEP_VERITY
)) {
134 INIT_WORK(&ctx
->work
, verity_work
);
135 fsverity_enqueue_verify_work(&ctx
->work
);
141 __read_end_io(ctx
->bio
);
145 static bool bio_post_read_required(struct bio
*bio
)
147 return bio
->bi_private
&& !bio
->bi_status
;
151 * I/O completion handler for multipage BIOs.
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
155 * back to block_read_full_folio().
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.
162 static void mpage_end_io(struct bio
*bio
)
164 if (bio_post_read_required(bio
)) {
165 struct bio_post_read_ctx
*ctx
= bio
->bi_private
;
167 ctx
->cur_step
= STEP_INITIAL
;
168 bio_post_read_processing(ctx
);
174 static inline bool ext4_need_verity(const struct inode
*inode
, pgoff_t idx
)
176 return fsverity_active(inode
) &&
177 idx
< DIV_ROUND_UP(inode
->i_size
, PAGE_SIZE
);
180 static void ext4_set_bio_post_read_ctx(struct bio
*bio
,
181 const struct inode
*inode
,
184 unsigned int post_read_steps
= 0;
186 if (fscrypt_inode_uses_fs_layer_crypto(inode
))
187 post_read_steps
|= 1 << STEP_DECRYPT
;
189 if (ext4_need_verity(inode
, first_idx
))
190 post_read_steps
|= 1 << STEP_VERITY
;
192 if (post_read_steps
) {
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
);
198 ctx
->enabled_steps
= post_read_steps
;
199 bio
->bi_private
= ctx
;
203 static inline loff_t
ext4_readpage_limit(struct inode
*inode
)
205 if (IS_ENABLED(CONFIG_FS_VERITY
) && IS_VERITY(inode
))
206 return inode
->i_sb
->s_maxbytes
;
208 return i_size_read(inode
);
211 int ext4_mpage_readpages(struct inode
*inode
,
212 struct readahead_control
*rac
, struct folio
*folio
)
214 struct bio
*bio
= NULL
;
215 sector_t last_block_in_bio
= 0;
217 const unsigned blkbits
= inode
->i_blkbits
;
218 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
219 const unsigned blocksize
= 1 << blkbits
;
221 sector_t block_in_file
;
223 sector_t last_block_in_file
;
224 sector_t first_block
;
226 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
228 unsigned relative_block
= 0;
229 struct ext4_map_blocks map
;
230 unsigned int nr_pages
= rac
? readahead_count(rac
) : 1;
237 for (; nr_pages
; nr_pages
--) {
238 int fully_mapped
= 1;
239 unsigned first_hole
= blocks_per_page
;
242 folio
= readahead_folio(rac
);
243 prefetchw(&folio
->flags
);
245 if (folio_buffers(folio
))
248 block_in_file
= next_block
=
249 (sector_t
)folio
->index
<< (PAGE_SHIFT
- blkbits
);
250 last_block
= block_in_file
+ nr_pages
* blocks_per_page
;
251 last_block_in_file
= (ext4_readpage_limit(inode
) +
252 blocksize
- 1) >> blkbits
;
253 if (last_block
> last_block_in_file
)
254 last_block
= last_block_in_file
;
258 * Map blocks using the previous result first.
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
;
266 first_block
= map
.m_pblk
+ map_offset
;
267 for (relative_block
= 0; ; relative_block
++) {
268 if (relative_block
== last
) {
270 map
.m_flags
&= ~EXT4_MAP_MAPPED
;
273 if (page_block
== blocks_per_page
)
281 * Then do more ext4_map_blocks() calls until we are
282 * done with this folio.
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
;
289 if (ext4_map_blocks(NULL
, inode
, &map
, 0) < 0) {
291 folio_zero_segment(folio
, 0,
297 if ((map
.m_flags
& EXT4_MAP_MAPPED
) == 0) {
299 if (first_hole
== blocks_per_page
)
300 first_hole
= page_block
;
305 if (first_hole
!= blocks_per_page
)
306 goto confused
; /* hole -> non-hole */
308 /* Contiguous blocks? */
310 first_block
= map
.m_pblk
;
311 else if (first_block
+ page_block
!= map
.m_pblk
)
313 for (relative_block
= 0; ; relative_block
++) {
314 if (relative_block
== map
.m_len
) {
316 map
.m_flags
&= ~EXT4_MAP_MAPPED
;
318 } else if (page_block
== blocks_per_page
)
324 if (first_hole
!= blocks_per_page
) {
325 folio_zero_segment(folio
, first_hole
<< blkbits
,
327 if (first_hole
== 0) {
328 if (ext4_need_verity(inode
, folio
->index
) &&
329 !fsverity_verify_folio(folio
))
331 folio_end_read(folio
, true);
334 } else if (fully_mapped
) {
335 folio_set_mappedtodisk(folio
);
339 * This folio will go to BIO. Do we need to send this
342 if (bio
&& (last_block_in_bio
!= first_block
- 1 ||
343 !fscrypt_mergeable_bio(bio
, inode
, next_block
))) {
350 * bio_alloc will _always_ be able to allocate a bio if
351 * __GFP_DIRECT_RECLAIM is set, see bio_alloc_bioset().
353 bio
= bio_alloc(bdev
, bio_max_segs(nr_pages
),
354 REQ_OP_READ
, GFP_KERNEL
);
355 fscrypt_set_bio_crypt_ctx(bio
, inode
, next_block
,
357 ext4_set_bio_post_read_ctx(bio
, inode
, folio
->index
);
358 bio
->bi_iter
.bi_sector
= first_block
<< (blkbits
- 9);
359 bio
->bi_end_io
= mpage_end_io
;
361 bio
->bi_opf
|= REQ_RAHEAD
;
364 length
= first_hole
<< blkbits
;
365 if (!bio_add_folio(bio
, folio
, length
, 0))
366 goto submit_and_realloc
;
368 if (((map
.m_flags
& EXT4_MAP_BOUNDARY
) &&
369 (relative_block
== map
.m_len
)) ||
370 (first_hole
!= blocks_per_page
)) {
374 last_block_in_bio
= first_block
+ blocks_per_page
- 1;
381 if (!folio_test_uptodate(folio
))
382 block_read_full_folio(folio
, ext4_get_block
);
386 ; /* A label shall be followed by a statement until C23 */
393 int __init
ext4_init_post_read_processing(void)
395 bio_post_read_ctx_cache
= KMEM_CACHE(bio_post_read_ctx
, SLAB_RECLAIM_ACCOUNT
);
397 if (!bio_post_read_ctx_cache
)
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
;
407 kmem_cache_destroy(bio_post_read_ctx_cache
);
412 void ext4_exit_post_read_processing(void)
414 mempool_destroy(bio_post_read_ctx_pool
);
415 kmem_cache_destroy(bio_post_read_ctx_cache
);