1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Squashfs - a compressed read only filesystem for Linux
4 *
5 * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
6 * Phillip Lougher <phillip@squashfs.org.uk>
7 *
8 * block.c
9 */
10
11 /*
12 * This file implements the low-level routines to read and decompress
13 * datablocks and metadata blocks.
14 */
15
16 #include <linux/blkdev.h>
17 #include <linux/fs.h>
18 #include <linux/vfs.h>
19 #include <linux/slab.h>
20 #include <linux/pagemap.h>
21 #include <linux/string.h>
22 #include <linux/bio.h>
23
24 #include "squashfs_fs.h"
25 #include "squashfs_fs_sb.h"
26 #include "squashfs.h"
27 #include "decompressor.h"
28 #include "page_actor.h"
29
30 /*
31 * Returns the amount of bytes copied to the page actor.
32 */
33 static int copy_bio_to_actor(struct bio *bio,
34 struct squashfs_page_actor *actor,
35 int offset, int req_length)
36 {
37 void *actor_addr;
38 struct bvec_iter_all iter_all = {};
39 struct bio_vec *bvec = bvec_init_iter_all(&iter_all);
40 int copied_bytes = 0;
41 int actor_offset = 0;
42
43 squashfs_actor_nobuff(actor);
44 actor_addr = squashfs_first_page(actor);
45
46 if (WARN_ON_ONCE(!bio_next_segment(bio, &iter_all)))
47 return 0;
48
49 while (copied_bytes < req_length) {
50 int bytes_to_copy = min_t(int, bvec->bv_len - offset,
51 PAGE_SIZE - actor_offset);
52
53 bytes_to_copy = min_t(int, bytes_to_copy,
54 req_length - copied_bytes);
55 if (!IS_ERR(actor_addr))
56 memcpy(actor_addr + actor_offset, bvec_virt(bvec) +
57 offset, bytes_to_copy);
58
59 actor_offset += bytes_to_copy;
60 copied_bytes += bytes_to_copy;
61 offset += bytes_to_copy;
62
63 if (actor_offset >= PAGE_SIZE) {
64 actor_addr = squashfs_next_page(actor);
65 if (!actor_addr)
66 break;
67 actor_offset = 0;
68 }
69 if (offset >= bvec->bv_len) {
70 if (!bio_next_segment(bio, &iter_all))
71 break;
72 offset = 0;
73 }
74 }
75 squashfs_finish_page(actor);
76 return copied_bytes;
77 }
78
79 static int squashfs_bio_read_cached(struct bio *fullbio,
80 struct address_space *cache_mapping, u64 index, int length,
81 u64 read_start, u64 read_end, int page_count)
82 {
83 struct page *head_to_cache = NULL, *tail_to_cache = NULL;
84 struct block_device *bdev = fullbio->bi_bdev;
85 int start_idx = 0, end_idx = 0;
86 struct bvec_iter_all iter_all;
87 struct bio *bio = NULL;
88 struct bio_vec *bv;
89 int idx = 0;
90 int err = 0;
91
92 bio_for_each_segment_all(bv, fullbio, iter_all) {
93 struct page *page = bv->bv_page;
94
95 if (page->mapping == cache_mapping) {
96 idx++;
97 continue;
98 }
99
100 /*
101 * We only use this when the device block size is the same as
102 * the page size, so read_start and read_end cover full pages.
103 *
104 * Compare these to the original required index and length to
105 * only cache pages which were requested partially, since these
106 * are the ones which are likely to be needed when reading
107 * adjacent blocks.
108 */
109 if (idx == 0 && index != read_start)
110 head_to_cache = page;
111 else if (idx == page_count - 1 && index + length != read_end)
112 tail_to_cache = page;
113
114 if (!bio || idx != end_idx) {
115 struct bio *new = bio_alloc_clone(bdev, fullbio,
116 GFP_NOIO, &fs_bio_set);
117
118 if (bio) {
119 bio_trim(bio, start_idx * PAGE_SECTORS,
120 (end_idx - start_idx) * PAGE_SECTORS);
121 bio_chain(bio, new);
122 submit_bio(bio);
123 }
124
125 bio = new;
126 start_idx = idx;
127 }
128
129 idx++;
130 end_idx = idx;
131 }
132
133 if (bio) {
134 bio_trim(bio, start_idx * PAGE_SECTORS,
135 (end_idx - start_idx) * PAGE_SECTORS);
136 err = submit_bio_wait(bio);
137 bio_put(bio);
138 }
139
140 if (err)
141 return err;
142
143 if (head_to_cache) {
144 int ret = add_to_page_cache_lru(head_to_cache, cache_mapping,
145 read_start >> PAGE_SHIFT,
146 GFP_NOIO);
147
148 if (!ret) {
149 SetPageUptodate(head_to_cache);
150 unlock_page(head_to_cache);
151 }
152
153 }
154
155 if (tail_to_cache) {
156 int ret = add_to_page_cache_lru(tail_to_cache, cache_mapping,
157 (read_end >> PAGE_SHIFT) - 1,
158 GFP_NOIO);
159
160 if (!ret) {
161 SetPageUptodate(tail_to_cache);
162 unlock_page(tail_to_cache);
163 }
164 }
165
166 return 0;
167 }
168
169 static struct page *squashfs_get_cache_page(struct address_space *mapping,
170 pgoff_t index)
171 {
172 struct page *page;
173
174 if (!mapping)
175 return NULL;
176
177 page = find_get_page(mapping, index);
178 if (!page)
179 return NULL;
180
181 if (!PageUptodate(page)) {
182 put_page(page);
183 return NULL;
184 }
185
186 return page;
187 }
188
189 static int squashfs_bio_read(struct super_block *sb, u64 index, int length,
190 struct bio **biop, int *block_offset)
191 {
192 struct squashfs_sb_info *msblk = sb->s_fs_info;
193 struct address_space *cache_mapping = msblk->cache_mapping;
194 const u64 read_start = round_down(index, msblk->devblksize);
195 const sector_t block = read_start >> msblk->devblksize_log2;
196 const u64 read_end = round_up(index + length, msblk->devblksize);
197 const sector_t block_end = read_end >> msblk->devblksize_log2;
198 int offset = read_start - round_down(index, PAGE_SIZE);
199 int total_len = (block_end - block) << msblk->devblksize_log2;
200 const int page_count = DIV_ROUND_UP(total_len + offset, PAGE_SIZE);
201 int error, i;
202 struct bio *bio;
203
204 bio = bio_kmalloc(page_count, GFP_NOIO);
205 if (!bio)
206 return -ENOMEM;
207 bio_init(bio, sb->s_bdev, bio->bi_inline_vecs, page_count, REQ_OP_READ);
208 bio->bi_iter.bi_sector = block * (msblk->devblksize >> SECTOR_SHIFT);
209
210 for (i = 0; i < page_count; ++i) {
211 unsigned int len =
212 min_t(unsigned int, PAGE_SIZE - offset, total_len);
213 pgoff_t index = (read_start >> PAGE_SHIFT) + i;
214 struct page *page;
215
216 page = squashfs_get_cache_page(cache_mapping, index);
217 if (!page)
218 page = alloc_page(GFP_NOIO);
219
220 if (!page) {
221 error = -ENOMEM;
222 goto out_free_bio;
223 }
224
225 /*
226 * Use the __ version to avoid merging since we need each page
227 * to be separate when we check for and avoid cached pages.
228 */
229 __bio_add_page(bio, page, len, offset);
230 offset = 0;
231 total_len -= len;
232 }
233
234 if (cache_mapping)
235 error = squashfs_bio_read_cached(bio, cache_mapping, index,
236 length, read_start, read_end,
237 page_count);
238 else
239 error = submit_bio_wait(bio);
240 if (error)
241 goto out_free_bio;
242
243 *biop = bio;
244 *block_offset = index & ((1 << msblk->devblksize_log2) - 1);
245 return 0;
246
247 out_free_bio:
248 bio_free_pages(bio);
249 bio_uninit(bio);
250 kfree(bio);
251 return error;
252 }
253
254 /*
255 * Read and decompress a metadata block or datablock. Length is non-zero
256 * if a datablock is being read (the size is stored elsewhere in the
257 * filesystem), otherwise the length is obtained from the first two bytes of
258 * the metadata block. A bit in the length field indicates if the block
259 * is stored uncompressed in the filesystem (usually because compression
260 * generated a larger block - this does occasionally happen with compression
261 * algorithms).
262 */
263 int squashfs_read_data(struct super_block *sb, u64 index, int length,
264 u64 *next_index, struct squashfs_page_actor *output)
265 {
266 struct squashfs_sb_info *msblk = sb->s_fs_info;
267 struct bio *bio = NULL;
268 int compressed;
269 int res;
270 int offset;
271
272 if (length) {
273 /*
274 * Datablock.
275 */
276 compressed = SQUASHFS_COMPRESSED_BLOCK(length);
277 length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
278 TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
279 index, compressed ? "" : "un", length, output->length);
280 } else {
281 /*
282 * Metadata block.
283 */
284 const u8 *data;
285 struct bvec_iter_all iter_all = {};
286 struct bio_vec *bvec = bvec_init_iter_all(&iter_all);
287
288 if (index + 2 > msblk->bytes_used) {
289 res = -EIO;
290 goto out;
291 }
292 res = squashfs_bio_read(sb, index, 2, &bio, &offset);
293 if (res)
294 goto out;
295
296 if (WARN_ON_ONCE(!bio_next_segment(bio, &iter_all))) {
297 res = -EIO;
298 goto out_free_bio;
299 }
300 /* Extract the length of the metadata block */
301 data = bvec_virt(bvec);
302 length = data[offset];
303 if (offset < bvec->bv_len - 1) {
304 length |= data[offset + 1] << 8;
305 } else {
306 if (WARN_ON_ONCE(!bio_next_segment(bio, &iter_all))) {
307 res = -EIO;
308 goto out_free_bio;
309 }
310 data = bvec_virt(bvec);
311 length |= data[0] << 8;
312 }
313 bio_free_pages(bio);
314 bio_uninit(bio);
315 kfree(bio);
316
317 compressed = SQUASHFS_COMPRESSED(length);
318 length = SQUASHFS_COMPRESSED_SIZE(length);
319 index += 2;
320
321 TRACE("Block @ 0x%llx, %scompressed size %d\n", index - 2,
322 compressed ? "" : "un", length);
323 }
324 if (length <= 0 || length > output->length ||
325 (index + length) > msblk->bytes_used) {
326 res = -EIO;
327 goto out;
328 }
329
330 if (next_index)
331 *next_index = index + length;
332
333 res = squashfs_bio_read(sb, index, length, &bio, &offset);
334 if (res)
335 goto out;
336
337 if (compressed) {
338 if (!msblk->stream) {
339 res = -EIO;
340 goto out_free_bio;
341 }
342 res = msblk->thread_ops->decompress(msblk, bio, offset, length, output);
343 } else {
344 res = copy_bio_to_actor(bio, output, offset, length);
345 }
346
347 out_free_bio:
348 bio_free_pages(bio);
349 bio_uninit(bio);
350 kfree(bio);
351 out:
352 if (res < 0) {
353 ERROR("Failed to read block 0x%llx: %d\n", index, res);
354 if (msblk->panic_on_errors)
355 panic("squashfs read failed");
356 }
357
358 return res;
359 }
360
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.