]>
Commit | Line | Data |
---|---|---|
1701aecb PL |
1 | /* |
2 | * Squashfs - a compressed read only filesystem for Linux | |
3 | * | |
4 | * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008 | |
d7f2ff67 | 5 | * Phillip Lougher <phillip@squashfs.org.uk> |
1701aecb PL |
6 | * |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version 2, | |
10 | * or (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
20 | * | |
21 | * file.c | |
22 | */ | |
23 | ||
24 | /* | |
25 | * This file contains code for handling regular files. A regular file | |
26 | * consists of a sequence of contiguous compressed blocks, and/or a | |
27 | * compressed fragment block (tail-end packed block). The compressed size | |
28 | * of each datablock is stored in a block list contained within the | |
29 | * file inode (itself stored in one or more compressed metadata blocks). | |
30 | * | |
31 | * To speed up access to datablocks when reading 'large' files (256 Mbytes or | |
32 | * larger), the code implements an index cache that caches the mapping from | |
33 | * block index to datablock location on disk. | |
34 | * | |
35 | * The index cache allows Squashfs to handle large files (up to 1.75 TiB) while | |
36 | * retaining a simple and space-efficient block list on disk. The cache | |
37 | * is split into slots, caching up to eight 224 GiB files (128 KiB blocks). | |
38 | * Larger files use multiple slots, with 1.75 TiB files using all 8 slots. | |
39 | * The index cache is designed to be memory efficient, and by default uses | |
40 | * 16 KiB. | |
41 | */ | |
42 | ||
43 | #include <linux/fs.h> | |
44 | #include <linux/vfs.h> | |
45 | #include <linux/kernel.h> | |
46 | #include <linux/slab.h> | |
47 | #include <linux/string.h> | |
48 | #include <linux/pagemap.h> | |
49 | #include <linux/mutex.h> | |
1701aecb PL |
50 | |
51 | #include "squashfs_fs.h" | |
52 | #include "squashfs_fs_sb.h" | |
53 | #include "squashfs_fs_i.h" | |
54 | #include "squashfs.h" | |
55 | ||
56 | /* | |
57 | * Locate cache slot in range [offset, index] for specified inode. If | |
58 | * there's more than one return the slot closest to index. | |
59 | */ | |
60 | static struct meta_index *locate_meta_index(struct inode *inode, int offset, | |
61 | int index) | |
62 | { | |
63 | struct meta_index *meta = NULL; | |
64 | struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info; | |
65 | int i; | |
66 | ||
67 | mutex_lock(&msblk->meta_index_mutex); | |
68 | ||
69 | TRACE("locate_meta_index: index %d, offset %d\n", index, offset); | |
70 | ||
71 | if (msblk->meta_index == NULL) | |
72 | goto not_allocated; | |
73 | ||
74 | for (i = 0; i < SQUASHFS_META_SLOTS; i++) { | |
75 | if (msblk->meta_index[i].inode_number == inode->i_ino && | |
76 | msblk->meta_index[i].offset >= offset && | |
77 | msblk->meta_index[i].offset <= index && | |
78 | msblk->meta_index[i].locked == 0) { | |
79 | TRACE("locate_meta_index: entry %d, offset %d\n", i, | |
80 | msblk->meta_index[i].offset); | |
81 | meta = &msblk->meta_index[i]; | |
82 | offset = meta->offset; | |
83 | } | |
84 | } | |
85 | ||
86 | if (meta) | |
87 | meta->locked = 1; | |
88 | ||
89 | not_allocated: | |
90 | mutex_unlock(&msblk->meta_index_mutex); | |
91 | ||
92 | return meta; | |
93 | } | |
94 | ||
95 | ||
96 | /* | |
97 | * Find and initialise an empty cache slot for index offset. | |
98 | */ | |
99 | static struct meta_index *empty_meta_index(struct inode *inode, int offset, | |
100 | int skip) | |
101 | { | |
102 | struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info; | |
103 | struct meta_index *meta = NULL; | |
104 | int i; | |
105 | ||
106 | mutex_lock(&msblk->meta_index_mutex); | |
107 | ||
108 | TRACE("empty_meta_index: offset %d, skip %d\n", offset, skip); | |
109 | ||
110 | if (msblk->meta_index == NULL) { | |
111 | /* | |
112 | * First time cache index has been used, allocate and | |
113 | * initialise. The cache index could be allocated at | |
114 | * mount time but doing it here means it is allocated only | |
115 | * if a 'large' file is read. | |
116 | */ | |
117 | msblk->meta_index = kcalloc(SQUASHFS_META_SLOTS, | |
118 | sizeof(*(msblk->meta_index)), GFP_KERNEL); | |
119 | if (msblk->meta_index == NULL) { | |
120 | ERROR("Failed to allocate meta_index\n"); | |
121 | goto failed; | |
122 | } | |
123 | for (i = 0; i < SQUASHFS_META_SLOTS; i++) { | |
124 | msblk->meta_index[i].inode_number = 0; | |
125 | msblk->meta_index[i].locked = 0; | |
126 | } | |
127 | msblk->next_meta_index = 0; | |
128 | } | |
129 | ||
130 | for (i = SQUASHFS_META_SLOTS; i && | |
131 | msblk->meta_index[msblk->next_meta_index].locked; i--) | |
132 | msblk->next_meta_index = (msblk->next_meta_index + 1) % | |
133 | SQUASHFS_META_SLOTS; | |
134 | ||
135 | if (i == 0) { | |
136 | TRACE("empty_meta_index: failed!\n"); | |
137 | goto failed; | |
138 | } | |
139 | ||
140 | TRACE("empty_meta_index: returned meta entry %d, %p\n", | |
141 | msblk->next_meta_index, | |
142 | &msblk->meta_index[msblk->next_meta_index]); | |
143 | ||
144 | meta = &msblk->meta_index[msblk->next_meta_index]; | |
145 | msblk->next_meta_index = (msblk->next_meta_index + 1) % | |
146 | SQUASHFS_META_SLOTS; | |
147 | ||
148 | meta->inode_number = inode->i_ino; | |
149 | meta->offset = offset; | |
150 | meta->skip = skip; | |
151 | meta->entries = 0; | |
152 | meta->locked = 1; | |
153 | ||
154 | failed: | |
155 | mutex_unlock(&msblk->meta_index_mutex); | |
156 | return meta; | |
157 | } | |
158 | ||
159 | ||
160 | static void release_meta_index(struct inode *inode, struct meta_index *meta) | |
161 | { | |
162 | struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info; | |
163 | mutex_lock(&msblk->meta_index_mutex); | |
164 | meta->locked = 0; | |
165 | mutex_unlock(&msblk->meta_index_mutex); | |
166 | } | |
167 | ||
168 | ||
169 | /* | |
170 | * Read the next n blocks from the block list, starting from | |
171 | * metadata block <start_block, offset>. | |
172 | */ | |
173 | static long long read_indexes(struct super_block *sb, int n, | |
174 | u64 *start_block, int *offset) | |
175 | { | |
176 | int err, i; | |
177 | long long block = 0; | |
09cbfeaf | 178 | __le32 *blist = kmalloc(PAGE_SIZE, GFP_KERNEL); |
1701aecb PL |
179 | |
180 | if (blist == NULL) { | |
181 | ERROR("read_indexes: Failed to allocate block_list\n"); | |
182 | return -ENOMEM; | |
183 | } | |
184 | ||
185 | while (n) { | |
09cbfeaf | 186 | int blocks = min_t(int, n, PAGE_SIZE >> 2); |
1701aecb PL |
187 | |
188 | err = squashfs_read_metadata(sb, blist, start_block, | |
189 | offset, blocks << 2); | |
190 | if (err < 0) { | |
191 | ERROR("read_indexes: reading block [%llx:%x]\n", | |
192 | *start_block, *offset); | |
193 | goto failure; | |
194 | } | |
195 | ||
196 | for (i = 0; i < blocks; i++) { | |
01cfb793 LT |
197 | int size = squashfs_block_size(blist[i]); |
198 | if (size < 0) { | |
199 | err = size; | |
200 | goto failure; | |
201 | } | |
1701aecb PL |
202 | block += SQUASHFS_COMPRESSED_SIZE_BLOCK(size); |
203 | } | |
204 | n -= blocks; | |
205 | } | |
206 | ||
207 | kfree(blist); | |
208 | return block; | |
209 | ||
210 | failure: | |
211 | kfree(blist); | |
212 | return err; | |
213 | } | |
214 | ||
215 | ||
216 | /* | |
217 | * Each cache index slot has SQUASHFS_META_ENTRIES, each of which | |
218 | * can cache one index -> datablock/blocklist-block mapping. We wish | |
219 | * to distribute these over the length of the file, entry[0] maps index x, | |
220 | * entry[1] maps index x + skip, entry[2] maps index x + 2 * skip, and so on. | |
221 | * The larger the file, the greater the skip factor. The skip factor is | |
222 | * limited to the size of the metadata cache (SQUASHFS_CACHED_BLKS) to ensure | |
223 | * the number of metadata blocks that need to be read fits into the cache. | |
224 | * If the skip factor is limited in this way then the file will use multiple | |
225 | * slots. | |
226 | */ | |
227 | static inline int calculate_skip(int blocks) | |
228 | { | |
229 | int skip = blocks / ((SQUASHFS_META_ENTRIES + 1) | |
230 | * SQUASHFS_META_INDEXES); | |
231 | return min(SQUASHFS_CACHED_BLKS - 1, skip + 1); | |
232 | } | |
233 | ||
234 | ||
235 | /* | |
236 | * Search and grow the index cache for the specified inode, returning the | |
237 | * on-disk locations of the datablock and block list metadata block | |
238 | * <index_block, index_offset> for index (scaled to nearest cache index). | |
239 | */ | |
240 | static int fill_meta_index(struct inode *inode, int index, | |
241 | u64 *index_block, int *index_offset, u64 *data_block) | |
242 | { | |
243 | struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info; | |
244 | int skip = calculate_skip(i_size_read(inode) >> msblk->block_log); | |
245 | int offset = 0; | |
246 | struct meta_index *meta; | |
247 | struct meta_entry *meta_entry; | |
248 | u64 cur_index_block = squashfs_i(inode)->block_list_start; | |
249 | int cur_offset = squashfs_i(inode)->offset; | |
250 | u64 cur_data_block = squashfs_i(inode)->start; | |
251 | int err, i; | |
252 | ||
253 | /* | |
254 | * Scale index to cache index (cache slot entry) | |
255 | */ | |
256 | index /= SQUASHFS_META_INDEXES * skip; | |
257 | ||
258 | while (offset < index) { | |
259 | meta = locate_meta_index(inode, offset + 1, index); | |
260 | ||
261 | if (meta == NULL) { | |
262 | meta = empty_meta_index(inode, offset + 1, skip); | |
263 | if (meta == NULL) | |
264 | goto all_done; | |
265 | } else { | |
266 | offset = index < meta->offset + meta->entries ? index : | |
267 | meta->offset + meta->entries - 1; | |
268 | meta_entry = &meta->meta_entry[offset - meta->offset]; | |
269 | cur_index_block = meta_entry->index_block + | |
270 | msblk->inode_table; | |
271 | cur_offset = meta_entry->offset; | |
272 | cur_data_block = meta_entry->data_block; | |
273 | TRACE("get_meta_index: offset %d, meta->offset %d, " | |
274 | "meta->entries %d\n", offset, meta->offset, | |
275 | meta->entries); | |
276 | TRACE("get_meta_index: index_block 0x%llx, offset 0x%x" | |
277 | " data_block 0x%llx\n", cur_index_block, | |
278 | cur_offset, cur_data_block); | |
279 | } | |
280 | ||
281 | /* | |
282 | * If necessary grow cache slot by reading block list. Cache | |
283 | * slot is extended up to index or to the end of the slot, in | |
284 | * which case further slots will be used. | |
285 | */ | |
286 | for (i = meta->offset + meta->entries; i <= index && | |
287 | i < meta->offset + SQUASHFS_META_ENTRIES; i++) { | |
288 | int blocks = skip * SQUASHFS_META_INDEXES; | |
289 | long long res = read_indexes(inode->i_sb, blocks, | |
290 | &cur_index_block, &cur_offset); | |
291 | ||
292 | if (res < 0) { | |
293 | if (meta->entries == 0) | |
294 | /* | |
295 | * Don't leave an empty slot on read | |
296 | * error allocated to this inode... | |
297 | */ | |
298 | meta->inode_number = 0; | |
299 | err = res; | |
300 | goto failed; | |
301 | } | |
302 | ||
303 | cur_data_block += res; | |
304 | meta_entry = &meta->meta_entry[i - meta->offset]; | |
305 | meta_entry->index_block = cur_index_block - | |
306 | msblk->inode_table; | |
307 | meta_entry->offset = cur_offset; | |
308 | meta_entry->data_block = cur_data_block; | |
309 | meta->entries++; | |
310 | offset++; | |
311 | } | |
312 | ||
313 | TRACE("get_meta_index: meta->offset %d, meta->entries %d\n", | |
314 | meta->offset, meta->entries); | |
315 | ||
316 | release_meta_index(inode, meta); | |
317 | } | |
318 | ||
319 | all_done: | |
320 | *index_block = cur_index_block; | |
321 | *index_offset = cur_offset; | |
322 | *data_block = cur_data_block; | |
323 | ||
324 | /* | |
325 | * Scale cache index (cache slot entry) to index | |
326 | */ | |
327 | return offset * SQUASHFS_META_INDEXES * skip; | |
328 | ||
329 | failed: | |
330 | release_meta_index(inode, meta); | |
331 | return err; | |
332 | } | |
333 | ||
334 | ||
335 | /* | |
336 | * Get the on-disk location and compressed size of the datablock | |
337 | * specified by index. Fill_meta_index() does most of the work. | |
338 | */ | |
339 | static int read_blocklist(struct inode *inode, int index, u64 *block) | |
340 | { | |
341 | u64 start; | |
342 | long long blks; | |
343 | int offset; | |
344 | __le32 size; | |
345 | int res = fill_meta_index(inode, index, &start, &offset, block); | |
346 | ||
347 | TRACE("read_blocklist: res %d, index %d, start 0x%llx, offset" | |
348 | " 0x%x, block 0x%llx\n", res, index, start, offset, | |
349 | *block); | |
350 | ||
351 | if (res < 0) | |
352 | return res; | |
353 | ||
354 | /* | |
355 | * res contains the index of the mapping returned by fill_meta_index(), | |
356 | * this will likely be less than the desired index (because the | |
357 | * meta_index cache works at a higher granularity). Read any | |
358 | * extra block indexes needed. | |
359 | */ | |
360 | if (res < index) { | |
361 | blks = read_indexes(inode->i_sb, index - res, &start, &offset); | |
362 | if (blks < 0) | |
363 | return (int) blks; | |
364 | *block += blks; | |
365 | } | |
366 | ||
367 | /* | |
368 | * Read length of block specified by index. | |
369 | */ | |
370 | res = squashfs_read_metadata(inode->i_sb, &size, &start, &offset, | |
371 | sizeof(size)); | |
372 | if (res < 0) | |
373 | return res; | |
01cfb793 | 374 | return squashfs_block_size(size); |
1701aecb PL |
375 | } |
376 | ||
cdbb65c4 LT |
377 | void squashfs_fill_page(struct page *page, struct squashfs_cache_entry *buffer, int offset, int avail) |
378 | { | |
379 | int copied; | |
380 | void *pageaddr; | |
381 | ||
382 | pageaddr = kmap_atomic(page); | |
383 | copied = squashfs_copy_data(pageaddr, buffer, offset, avail); | |
384 | memset(pageaddr + copied, 0, PAGE_SIZE - copied); | |
385 | kunmap_atomic(pageaddr); | |
386 | ||
387 | flush_dcache_page(page); | |
388 | if (copied == avail) | |
389 | SetPageUptodate(page); | |
390 | else | |
391 | SetPageError(page); | |
392 | } | |
393 | ||
5f55dbc0 PL |
394 | /* Copy data into page cache */ |
395 | void squashfs_copy_cache(struct page *page, struct squashfs_cache_entry *buffer, | |
396 | int bytes, int offset) | |
1701aecb PL |
397 | { |
398 | struct inode *inode = page->mapping->host; | |
399 | struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info; | |
09cbfeaf | 400 | int i, mask = (1 << (msblk->block_log - PAGE_SHIFT)) - 1; |
5f55dbc0 | 401 | int start_index = page->index & ~mask, end_index = start_index | mask; |
1701aecb PL |
402 | |
403 | /* | |
404 | * Loop copying datablock into pages. As the datablock likely covers | |
ea1754a0 | 405 | * many PAGE_SIZE pages (default block size is 128 KiB) explicitly |
1701aecb PL |
406 | * grab the pages from the page cache, except for the page that we've |
407 | * been called to fill. | |
408 | */ | |
409 | for (i = start_index; i <= end_index && bytes > 0; i++, | |
09cbfeaf | 410 | bytes -= PAGE_SIZE, offset += PAGE_SIZE) { |
1701aecb | 411 | struct page *push_page; |
09cbfeaf | 412 | int avail = buffer ? min_t(int, bytes, PAGE_SIZE) : 0; |
1701aecb PL |
413 | |
414 | TRACE("bytes %d, i %d, available_bytes %d\n", bytes, i, avail); | |
415 | ||
416 | push_page = (i == page->index) ? page : | |
417 | grab_cache_page_nowait(page->mapping, i); | |
418 | ||
419 | if (!push_page) | |
420 | continue; | |
421 | ||
422 | if (PageUptodate(push_page)) | |
423 | goto skip_page; | |
424 | ||
cdbb65c4 | 425 | squashfs_fill_page(push_page, buffer, offset, avail); |
1701aecb PL |
426 | skip_page: |
427 | unlock_page(push_page); | |
428 | if (i != page->index) | |
09cbfeaf | 429 | put_page(push_page); |
1701aecb | 430 | } |
5f55dbc0 PL |
431 | } |
432 | ||
433 | /* Read datablock stored packed inside a fragment (tail-end packed block) */ | |
a3f94cb9 | 434 | static int squashfs_readpage_fragment(struct page *page, int expected) |
5f55dbc0 PL |
435 | { |
436 | struct inode *inode = page->mapping->host; | |
5f55dbc0 PL |
437 | struct squashfs_cache_entry *buffer = squashfs_get_fragment(inode->i_sb, |
438 | squashfs_i(inode)->fragment_block, | |
439 | squashfs_i(inode)->fragment_size); | |
440 | int res = buffer->error; | |
441 | ||
442 | if (res) | |
443 | ERROR("Unable to read page, block %llx, size %x\n", | |
444 | squashfs_i(inode)->fragment_block, | |
445 | squashfs_i(inode)->fragment_size); | |
446 | else | |
a3f94cb9 | 447 | squashfs_copy_cache(page, buffer, expected, |
5f55dbc0 PL |
448 | squashfs_i(inode)->fragment_offset); |
449 | ||
450 | squashfs_cache_put(buffer); | |
451 | return res; | |
452 | } | |
1701aecb | 453 | |
a3f94cb9 | 454 | static int squashfs_readpage_sparse(struct page *page, int expected) |
5f55dbc0 | 455 | { |
a3f94cb9 | 456 | squashfs_copy_cache(page, NULL, expected, 0); |
1701aecb | 457 | return 0; |
5f55dbc0 PL |
458 | } |
459 | ||
460 | static int squashfs_readpage(struct file *file, struct page *page) | |
461 | { | |
462 | struct inode *inode = page->mapping->host; | |
463 | struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info; | |
09cbfeaf | 464 | int index = page->index >> (msblk->block_log - PAGE_SHIFT); |
5f55dbc0 | 465 | int file_end = i_size_read(inode) >> msblk->block_log; |
a3f94cb9 PL |
466 | int expected = index == file_end ? |
467 | (i_size_read(inode) & (msblk->block_size - 1)) : | |
468 | msblk->block_size; | |
5f55dbc0 PL |
469 | int res; |
470 | void *pageaddr; | |
471 | ||
472 | TRACE("Entered squashfs_readpage, page index %lx, start block %llx\n", | |
473 | page->index, squashfs_i(inode)->start); | |
474 | ||
09cbfeaf KS |
475 | if (page->index >= ((i_size_read(inode) + PAGE_SIZE - 1) >> |
476 | PAGE_SHIFT)) | |
5f55dbc0 PL |
477 | goto out; |
478 | ||
479 | if (index < file_end || squashfs_i(inode)->fragment_block == | |
480 | SQUASHFS_INVALID_BLK) { | |
481 | u64 block = 0; | |
482 | int bsize = read_blocklist(inode, index, &block); | |
483 | if (bsize < 0) | |
484 | goto error_out; | |
485 | ||
486 | if (bsize == 0) | |
a3f94cb9 | 487 | res = squashfs_readpage_sparse(page, expected); |
5f55dbc0 | 488 | else |
a3f94cb9 | 489 | res = squashfs_readpage_block(page, block, bsize, expected); |
5f55dbc0 | 490 | } else |
a3f94cb9 | 491 | res = squashfs_readpage_fragment(page, expected); |
5f55dbc0 PL |
492 | |
493 | if (!res) | |
494 | return 0; | |
1701aecb PL |
495 | |
496 | error_out: | |
497 | SetPageError(page); | |
498 | out: | |
53b55e55 | 499 | pageaddr = kmap_atomic(page); |
09cbfeaf | 500 | memset(pageaddr, 0, PAGE_SIZE); |
53b55e55 | 501 | kunmap_atomic(pageaddr); |
1701aecb PL |
502 | flush_dcache_page(page); |
503 | if (!PageError(page)) | |
504 | SetPageUptodate(page); | |
505 | unlock_page(page); | |
506 | ||
507 | return 0; | |
508 | } | |
509 | ||
510 | ||
511 | const struct address_space_operations squashfs_aops = { | |
512 | .readpage = squashfs_readpage | |
513 | }; |