2 * Squashfs - a compressed read only filesystem for Linux
4 * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
5 * Phillip Lougher <phillip@lougher.demon.co.uk>
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.
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.
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.
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).
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.
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
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>
50 #include <linux/zlib.h>
52 #include "squashfs_fs.h"
53 #include "squashfs_fs_sb.h"
54 #include "squashfs_fs_i.h"
58 * Locate cache slot in range [offset, index] for specified inode. If
59 * there's more than one return the slot closest to index.
61 static struct meta_index *locate_meta_index(struct inode *inode, int offset,
64 struct meta_index *meta = NULL;
65 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
68 mutex_lock(&msblk->meta_index_mutex);
70 TRACE("locate_meta_index: index %d, offset %d\n", index, offset);
72 if (msblk->meta_index == NULL)
75 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
76 if (msblk->meta_index[i].inode_number == inode->i_ino &&
77 msblk->meta_index[i].offset >= offset &&
78 msblk->meta_index[i].offset <= index &&
79 msblk->meta_index[i].locked == 0) {
80 TRACE("locate_meta_index: entry %d, offset %d\n", i,
81 msblk->meta_index[i].offset);
82 meta = &msblk->meta_index[i];
83 offset = meta->offset;
91 mutex_unlock(&msblk->meta_index_mutex);
98 * Find and initialise an empty cache slot for index offset.
100 static struct meta_index *empty_meta_index(struct inode *inode, int offset,
103 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
104 struct meta_index *meta = NULL;
107 mutex_lock(&msblk->meta_index_mutex);
109 TRACE("empty_meta_index: offset %d, skip %d\n", offset, skip);
111 if (msblk->meta_index == NULL) {
113 * First time cache index has been used, allocate and
114 * initialise. The cache index could be allocated at
115 * mount time but doing it here means it is allocated only
116 * if a 'large' file is read.
118 msblk->meta_index = kcalloc(SQUASHFS_META_SLOTS,
119 sizeof(*(msblk->meta_index)), GFP_KERNEL);
120 if (msblk->meta_index == NULL) {
121 ERROR("Failed to allocate meta_index\n");
124 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
125 msblk->meta_index[i].inode_number = 0;
126 msblk->meta_index[i].locked = 0;
128 msblk->next_meta_index = 0;
131 for (i = SQUASHFS_META_SLOTS; i &&
132 msblk->meta_index[msblk->next_meta_index].locked; i--)
133 msblk->next_meta_index = (msblk->next_meta_index + 1) %
137 TRACE("empty_meta_index: failed!\n");
141 TRACE("empty_meta_index: returned meta entry %d, %p\n",
142 msblk->next_meta_index,
143 &msblk->meta_index[msblk->next_meta_index]);
145 meta = &msblk->meta_index[msblk->next_meta_index];
146 msblk->next_meta_index = (msblk->next_meta_index + 1) %
149 meta->inode_number = inode->i_ino;
150 meta->offset = offset;
156 mutex_unlock(&msblk->meta_index_mutex);
161 static void release_meta_index(struct inode *inode, struct meta_index *meta)
163 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
164 mutex_lock(&msblk->meta_index_mutex);
166 mutex_unlock(&msblk->meta_index_mutex);
171 * Read the next n blocks from the block list, starting from
172 * metadata block <start_block, offset>.
174 static long long read_indexes(struct super_block *sb, int n,
175 u64 *start_block, int *offset)
179 __le32 *blist = kmalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
182 ERROR("read_indexes: Failed to allocate block_list\n");
187 int blocks = min_t(int, n, PAGE_CACHE_SIZE >> 2);
189 err = squashfs_read_metadata(sb, blist, start_block,
190 offset, blocks << 2);
192 ERROR("read_indexes: reading block [%llx:%x]\n",
193 *start_block, *offset);
197 for (i = 0; i < blocks; i++) {
198 int size = le32_to_cpu(blist[i]);
199 block += SQUASHFS_COMPRESSED_SIZE_BLOCK(size);
214 * Each cache index slot has SQUASHFS_META_ENTRIES, each of which
215 * can cache one index -> datablock/blocklist-block mapping. We wish
216 * to distribute these over the length of the file, entry[0] maps index x,
217 * entry[1] maps index x + skip, entry[2] maps index x + 2 * skip, and so on.
218 * The larger the file, the greater the skip factor. The skip factor is
219 * limited to the size of the metadata cache (SQUASHFS_CACHED_BLKS) to ensure
220 * the number of metadata blocks that need to be read fits into the cache.
221 * If the skip factor is limited in this way then the file will use multiple
224 static inline int calculate_skip(int blocks)
226 int skip = blocks / ((SQUASHFS_META_ENTRIES + 1)
227 * SQUASHFS_META_INDEXES);
228 return min(SQUASHFS_CACHED_BLKS - 1, skip + 1);
233 * Search and grow the index cache for the specified inode, returning the
234 * on-disk locations of the datablock and block list metadata block
235 * <index_block, index_offset> for index (scaled to nearest cache index).
237 static int fill_meta_index(struct inode *inode, int index,
238 u64 *index_block, int *index_offset, u64 *data_block)
240 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
241 int skip = calculate_skip(i_size_read(inode) >> msblk->block_log);
243 struct meta_index *meta;
244 struct meta_entry *meta_entry;
245 u64 cur_index_block = squashfs_i(inode)->block_list_start;
246 int cur_offset = squashfs_i(inode)->offset;
247 u64 cur_data_block = squashfs_i(inode)->start;
251 * Scale index to cache index (cache slot entry)
253 index /= SQUASHFS_META_INDEXES * skip;
255 while (offset < index) {
256 meta = locate_meta_index(inode, offset + 1, index);
259 meta = empty_meta_index(inode, offset + 1, skip);
263 offset = index < meta->offset + meta->entries ? index :
264 meta->offset + meta->entries - 1;
265 meta_entry = &meta->meta_entry[offset - meta->offset];
266 cur_index_block = meta_entry->index_block +
268 cur_offset = meta_entry->offset;
269 cur_data_block = meta_entry->data_block;
270 TRACE("get_meta_index: offset %d, meta->offset %d, "
271 "meta->entries %d\n", offset, meta->offset,
273 TRACE("get_meta_index: index_block 0x%llx, offset 0x%x"
274 " data_block 0x%llx\n", cur_index_block,
275 cur_offset, cur_data_block);
279 * If necessary grow cache slot by reading block list. Cache
280 * slot is extended up to index or to the end of the slot, in
281 * which case further slots will be used.
283 for (i = meta->offset + meta->entries; i <= index &&
284 i < meta->offset + SQUASHFS_META_ENTRIES; i++) {
285 int blocks = skip * SQUASHFS_META_INDEXES;
286 long long res = read_indexes(inode->i_sb, blocks,
287 &cur_index_block, &cur_offset);
290 if (meta->entries == 0)
292 * Don't leave an empty slot on read
293 * error allocated to this inode...
295 meta->inode_number = 0;
300 cur_data_block += res;
301 meta_entry = &meta->meta_entry[i - meta->offset];
302 meta_entry->index_block = cur_index_block -
304 meta_entry->offset = cur_offset;
305 meta_entry->data_block = cur_data_block;
310 TRACE("get_meta_index: meta->offset %d, meta->entries %d\n",
311 meta->offset, meta->entries);
313 release_meta_index(inode, meta);
317 *index_block = cur_index_block;
318 *index_offset = cur_offset;
319 *data_block = cur_data_block;
322 * Scale cache index (cache slot entry) to index
324 return offset * SQUASHFS_META_INDEXES * skip;
327 release_meta_index(inode, meta);
333 * Get the on-disk location and compressed size of the datablock
334 * specified by index. Fill_meta_index() does most of the work.
336 static int read_blocklist(struct inode *inode, int index, u64 *block)
342 int res = fill_meta_index(inode, index, &start, &offset, block);
344 TRACE("read_blocklist: res %d, index %d, start 0x%llx, offset"
345 " 0x%x, block 0x%llx\n", res, index, start, offset,
352 * res contains the index of the mapping returned by fill_meta_index(),
353 * this will likely be less than the desired index (because the
354 * meta_index cache works at a higher granularity). Read any
355 * extra block indexes needed.
358 blks = read_indexes(inode->i_sb, index - res, &start, &offset);
365 * Read length of block specified by index.
367 res = squashfs_read_metadata(inode->i_sb, &size, &start, &offset,
371 return le32_to_cpu(size);
375 static int squashfs_readpage(struct file *file, struct page *page)
377 struct inode *inode = page->mapping->host;
378 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
379 int bytes, i, offset = 0, sparse = 0;
380 struct squashfs_cache_entry *buffer = NULL;
383 int mask = (1 << (msblk->block_log - PAGE_CACHE_SHIFT)) - 1;
384 int index = page->index >> (msblk->block_log - PAGE_CACHE_SHIFT);
385 int start_index = page->index & ~mask;
386 int end_index = start_index | mask;
387 int file_end = i_size_read(inode) >> msblk->block_log;
389 TRACE("Entered squashfs_readpage, page index %lx, start block %llx\n",
390 page->index, squashfs_i(inode)->start);
392 if (page->index >= ((i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
396 if (index < file_end || squashfs_i(inode)->fragment_block ==
397 SQUASHFS_INVALID_BLK) {
399 * Reading a datablock from disk. Need to read block list
400 * to get location and block size.
403 int bsize = read_blocklist(inode, index, &block);
407 if (bsize == 0) { /* hole */
408 bytes = index == file_end ?
409 (i_size_read(inode) & (msblk->block_size - 1)) :
414 * Read and decompress datablock.
416 buffer = squashfs_get_datablock(inode->i_sb,
419 ERROR("Unable to read page, block %llx, size %x"
421 squashfs_cache_put(buffer);
424 bytes = buffer->length;
428 * Datablock is stored inside a fragment (tail-end packed
431 buffer = squashfs_get_fragment(inode->i_sb,
432 squashfs_i(inode)->fragment_block,
433 squashfs_i(inode)->fragment_size);
436 ERROR("Unable to read page, block %llx, size %x\n",
437 squashfs_i(inode)->fragment_block,
438 squashfs_i(inode)->fragment_size);
439 squashfs_cache_put(buffer);
442 bytes = i_size_read(inode) & (msblk->block_size - 1);
443 offset = squashfs_i(inode)->fragment_offset;
447 * Loop copying datablock into pages. As the datablock likely covers
448 * many PAGE_CACHE_SIZE pages (default block size is 128 KiB) explicitly
449 * grab the pages from the page cache, except for the page that we've
450 * been called to fill.
452 for (i = start_index; i <= end_index && bytes > 0; i++,
453 bytes -= PAGE_CACHE_SIZE, offset += PAGE_CACHE_SIZE) {
454 struct page *push_page;
455 int avail = sparse ? 0 : min_t(int, bytes, PAGE_CACHE_SIZE);
457 TRACE("bytes %d, i %d, available_bytes %d\n", bytes, i, avail);
459 push_page = (i == page->index) ? page :
460 grab_cache_page_nowait(page->mapping, i);
465 if (PageUptodate(push_page))
468 pageaddr = kmap_atomic(push_page, KM_USER0);
469 squashfs_copy_data(pageaddr, buffer, offset, avail);
470 memset(pageaddr + avail, 0, PAGE_CACHE_SIZE - avail);
471 kunmap_atomic(pageaddr, KM_USER0);
472 flush_dcache_page(push_page);
473 SetPageUptodate(push_page);
475 unlock_page(push_page);
476 if (i != page->index)
477 page_cache_release(push_page);
481 squashfs_cache_put(buffer);
488 pageaddr = kmap_atomic(page, KM_USER0);
489 memset(pageaddr, 0, PAGE_CACHE_SIZE);
490 kunmap_atomic(pageaddr, KM_USER0);
491 flush_dcache_page(page);
492 if (!PageError(page))
493 SetPageUptodate(page);
500 const struct address_space_operations squashfs_aops = {
501 .readpage = squashfs_readpage