2 * Squashfs - a compressed read only filesystem for Linux
4 * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
5 * Phillip Lougher <phillip@squashfs.org.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>
51 #include "squashfs_fs.h"
52 #include "squashfs_fs_sb.h"
53 #include "squashfs_fs_i.h"
57 * Locate cache slot in range [offset, index] for specified inode. If
58 * there's more than one return the slot closest to index.
60 static struct meta_index *locate_meta_index(struct inode *inode, int offset,
63 struct meta_index *meta = NULL;
64 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
67 mutex_lock(&msblk->meta_index_mutex);
69 TRACE("locate_meta_index: index %d, offset %d\n", index, offset);
71 if (msblk->meta_index == NULL)
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;
90 mutex_unlock(&msblk->meta_index_mutex);
97 * Find and initialise an empty cache slot for index offset.
99 static struct meta_index *empty_meta_index(struct inode *inode, int offset,
102 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
103 struct meta_index *meta = NULL;
106 mutex_lock(&msblk->meta_index_mutex);
108 TRACE("empty_meta_index: offset %d, skip %d\n", offset, skip);
110 if (msblk->meta_index == NULL) {
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.
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");
123 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
124 msblk->meta_index[i].inode_number = 0;
125 msblk->meta_index[i].locked = 0;
127 msblk->next_meta_index = 0;
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) %
136 TRACE("empty_meta_index: failed!\n");
140 TRACE("empty_meta_index: returned meta entry %d, %p\n",
141 msblk->next_meta_index,
142 &msblk->meta_index[msblk->next_meta_index]);
144 meta = &msblk->meta_index[msblk->next_meta_index];
145 msblk->next_meta_index = (msblk->next_meta_index + 1) %
148 meta->inode_number = inode->i_ino;
149 meta->offset = offset;
155 mutex_unlock(&msblk->meta_index_mutex);
160 static void release_meta_index(struct inode *inode, struct meta_index *meta)
162 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
163 mutex_lock(&msblk->meta_index_mutex);
165 mutex_unlock(&msblk->meta_index_mutex);
170 * Read the next n blocks from the block list, starting from
171 * metadata block <start_block, offset>.
173 static long long read_indexes(struct super_block *sb, int n,
174 u64 *start_block, int *offset)
178 __le32 *blist = kmalloc(PAGE_SIZE, GFP_KERNEL);
181 ERROR("read_indexes: Failed to allocate block_list\n");
186 int blocks = min_t(int, n, PAGE_SIZE >> 2);
188 err = squashfs_read_metadata(sb, blist, start_block,
189 offset, blocks << 2);
191 ERROR("read_indexes: reading block [%llx:%x]\n",
192 *start_block, *offset);
196 for (i = 0; i < blocks; i++) {
197 int size = squashfs_block_size(blist[i]);
202 block += SQUASHFS_COMPRESSED_SIZE_BLOCK(size);
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
227 static inline int calculate_skip(u64 blocks)
229 u64 skip = blocks / ((SQUASHFS_META_ENTRIES + 1)
230 * SQUASHFS_META_INDEXES);
231 return min((u64) SQUASHFS_CACHED_BLKS - 1, skip + 1);
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).
240 static int fill_meta_index(struct inode *inode, int index,
241 u64 *index_block, int *index_offset, u64 *data_block)
243 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
244 int skip = calculate_skip(i_size_read(inode) >> msblk->block_log);
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;
254 * Scale index to cache index (cache slot entry)
256 index /= SQUASHFS_META_INDEXES * skip;
258 while (offset < index) {
259 meta = locate_meta_index(inode, offset + 1, index);
262 meta = empty_meta_index(inode, offset + 1, skip);
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 +
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,
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);
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.
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);
293 if (meta->entries == 0)
295 * Don't leave an empty slot on read
296 * error allocated to this inode...
298 meta->inode_number = 0;
303 cur_data_block += res;
304 meta_entry = &meta->meta_entry[i - meta->offset];
305 meta_entry->index_block = cur_index_block -
307 meta_entry->offset = cur_offset;
308 meta_entry->data_block = cur_data_block;
313 TRACE("get_meta_index: meta->offset %d, meta->entries %d\n",
314 meta->offset, meta->entries);
316 release_meta_index(inode, meta);
320 *index_block = cur_index_block;
321 *index_offset = cur_offset;
322 *data_block = cur_data_block;
325 * Scale cache index (cache slot entry) to index
327 return offset * SQUASHFS_META_INDEXES * skip;
330 release_meta_index(inode, meta);
336 * Get the on-disk location and compressed size of the datablock
337 * specified by index. Fill_meta_index() does most of the work.
339 static int read_blocklist(struct inode *inode, int index, u64 *block)
345 int res = fill_meta_index(inode, index, &start, &offset, block);
347 TRACE("read_blocklist: res %d, index %d, start 0x%llx, offset"
348 " 0x%x, block 0x%llx\n", res, index, start, offset,
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.
361 blks = read_indexes(inode->i_sb, index - res, &start, &offset);
368 * Read length of block specified by index.
370 res = squashfs_read_metadata(inode->i_sb, &size, &start, &offset,
374 return squashfs_block_size(size);
377 void squashfs_fill_page(struct page *page, struct squashfs_cache_entry *buffer, int offset, int avail)
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);
387 flush_dcache_page(page);
389 SetPageUptodate(page);
394 /* Copy data into page cache */
395 void squashfs_copy_cache(struct page *page, struct squashfs_cache_entry *buffer,
396 int bytes, int offset)
398 struct inode *inode = page->mapping->host;
399 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
400 int i, mask = (1 << (msblk->block_log - PAGE_SHIFT)) - 1;
401 int start_index = page->index & ~mask, end_index = start_index | mask;
404 * Loop copying datablock into pages. As the datablock likely covers
405 * many PAGE_SIZE pages (default block size is 128 KiB) explicitly
406 * grab the pages from the page cache, except for the page that we've
407 * been called to fill.
409 for (i = start_index; i <= end_index && bytes > 0; i++,
410 bytes -= PAGE_SIZE, offset += PAGE_SIZE) {
411 struct page *push_page;
412 int avail = buffer ? min_t(int, bytes, PAGE_SIZE) : 0;
414 TRACE("bytes %d, i %d, available_bytes %d\n", bytes, i, avail);
416 push_page = (i == page->index) ? page :
417 grab_cache_page_nowait(page->mapping, i);
422 if (PageUptodate(push_page))
425 squashfs_fill_page(push_page, buffer, offset, avail);
427 unlock_page(push_page);
428 if (i != page->index)
433 /* Read datablock stored packed inside a fragment (tail-end packed block) */
434 static int squashfs_readpage_fragment(struct page *page, int expected)
436 struct inode *inode = page->mapping->host;
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;
443 ERROR("Unable to read page, block %llx, size %x\n",
444 squashfs_i(inode)->fragment_block,
445 squashfs_i(inode)->fragment_size);
447 squashfs_copy_cache(page, buffer, expected,
448 squashfs_i(inode)->fragment_offset);
450 squashfs_cache_put(buffer);
454 static int squashfs_readpage_sparse(struct page *page, int expected)
456 squashfs_copy_cache(page, NULL, expected, 0);
460 static int squashfs_readpage(struct file *file, struct page *page)
462 struct inode *inode = page->mapping->host;
463 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
464 int index = page->index >> (msblk->block_log - PAGE_SHIFT);
465 int file_end = i_size_read(inode) >> msblk->block_log;
466 int expected = index == file_end ?
467 (i_size_read(inode) & (msblk->block_size - 1)) :
472 TRACE("Entered squashfs_readpage, page index %lx, start block %llx\n",
473 page->index, squashfs_i(inode)->start);
475 if (page->index >= ((i_size_read(inode) + PAGE_SIZE - 1) >>
479 if (index < file_end || squashfs_i(inode)->fragment_block ==
480 SQUASHFS_INVALID_BLK) {
482 int bsize = read_blocklist(inode, index, &block);
487 res = squashfs_readpage_sparse(page, expected);
489 res = squashfs_readpage_block(page, block, bsize, expected);
491 res = squashfs_readpage_fragment(page, expected);
499 pageaddr = kmap_atomic(page);
500 memset(pageaddr, 0, PAGE_SIZE);
501 kunmap_atomic(pageaddr);
502 flush_dcache_page(page);
503 if (!PageError(page))
504 SetPageUptodate(page);
511 const struct address_space_operations squashfs_aops = {
512 .readpage = squashfs_readpage