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 * Blocks in Squashfs are compressed. To avoid repeatedly decompressing
26 * recently accessed data Squashfs uses two small metadata and fragment caches.
28 * This file implements a generic cache implementation used for both caches,
29 * plus functions layered ontop of the generic cache implementation to
30 * access the metadata and fragment caches.
32 * To avoid out of memory and fragmentation isssues with vmalloc the cache
33 * uses sequences of kmalloced PAGE_CACHE_SIZE buffers.
35 * It should be noted that the cache is not used for file datablocks, these
36 * are decompressed and cached in the page-cache in the normal way. The
37 * cache is only used to temporarily cache fragment and metadata blocks
38 * which have been read as as a result of a metadata (i.e. inode or
39 * directory) or fragment access. Because metadata and fragments are packed
40 * together into blocks (to gain greater compression) the read of a particular
41 * piece of metadata or fragment will retrieve other metadata/fragments which
42 * have been packed with it, these because of locality-of-reference may be read
43 * in the near future. Temporarily caching them ensures they are available for
44 * near future access without requiring an additional read and decompress.
48 #include <linux/vfs.h>
49 #include <linux/slab.h>
50 #include <linux/vmalloc.h>
51 #include <linux/sched.h>
52 #include <linux/spinlock.h>
53 #include <linux/wait.h>
54 #include <linux/zlib.h>
55 #include <linux/pagemap.h>
57 #include "squashfs_fs.h"
58 #include "squashfs_fs_sb.h"
59 #include "squashfs_fs_i.h"
63 * Look-up block in cache, and increment usage count. If not in cache, read
64 * and decompress it from disk.
66 struct squashfs_cache_entry *squashfs_cache_get(struct super_block *sb,
67 struct squashfs_cache *cache, u64 block, int length)
70 struct squashfs_cache_entry *entry;
72 spin_lock(&cache->lock);
75 for (i = 0; i < cache->entries; i++)
76 if (cache->entry[i].block == block)
79 if (i == cache->entries) {
81 * Block not in cache, if all cache entries are used
82 * go to sleep waiting for one to become available.
84 if (cache->unused == 0) {
86 spin_unlock(&cache->lock);
87 wait_event(cache->wait_queue, cache->unused);
88 spin_lock(&cache->lock);
94 * At least one unused cache entry. A simple
95 * round-robin strategy is used to choose the entry to
96 * be evicted from the cache.
99 for (n = 0; n < cache->entries; n++) {
100 if (cache->entry[i].refcount == 0)
102 i = (i + 1) % cache->entries;
105 cache->next_blk = (i + 1) % cache->entries;
106 entry = &cache->entry[i];
109 * Initialise choosen cache entry, and fill it in from
113 entry->block = block;
116 entry->num_waiters = 0;
118 spin_unlock(&cache->lock);
120 entry->length = squashfs_read_data(sb, entry->data,
121 block, length, &entry->next_index,
122 cache->block_size, cache->pages);
124 spin_lock(&cache->lock);
126 if (entry->length < 0)
127 entry->error = entry->length;
132 * While filling this entry one or more other processes
133 * have looked it up in the cache, and have slept
134 * waiting for it to become available.
136 if (entry->num_waiters) {
137 spin_unlock(&cache->lock);
138 wake_up_all(&entry->wait_queue);
140 spin_unlock(&cache->lock);
146 * Block already in cache. Increment refcount so it doesn't
147 * get reused until we're finished with it, if it was
148 * previously unused there's one less cache entry available
151 entry = &cache->entry[i];
152 if (entry->refcount == 0)
157 * If the entry is currently being filled in by another process
158 * go to sleep waiting for it to become available.
160 if (entry->pending) {
161 entry->num_waiters++;
162 spin_unlock(&cache->lock);
163 wait_event(entry->wait_queue, !entry->pending);
165 spin_unlock(&cache->lock);
171 TRACE("Got %s %d, start block %lld, refcount %d, error %d\n",
172 cache->name, i, entry->block, entry->refcount, entry->error);
175 ERROR("Unable to read %s cache entry [%llx]\n", cache->name,
182 * Release cache entry, once usage count is zero it can be reused.
184 void squashfs_cache_put(struct squashfs_cache_entry *entry)
186 struct squashfs_cache *cache = entry->cache;
188 spin_lock(&cache->lock);
190 if (entry->refcount == 0) {
193 * If there's any processes waiting for a block to become
194 * available, wake one up.
196 if (cache->num_waiters) {
197 spin_unlock(&cache->lock);
198 wake_up(&cache->wait_queue);
202 spin_unlock(&cache->lock);
206 * Delete cache reclaiming all kmalloced buffers.
208 void squashfs_cache_delete(struct squashfs_cache *cache)
215 for (i = 0; i < cache->entries; i++) {
216 if (cache->entry[i].data) {
217 for (j = 0; j < cache->pages; j++)
218 kfree(cache->entry[i].data[j]);
219 kfree(cache->entry[i].data);
229 * Initialise cache allocating the specified number of entries, each of
230 * size block_size. To avoid vmalloc fragmentation issues each entry
231 * is allocated as a sequence of kmalloced PAGE_CACHE_SIZE buffers.
233 struct squashfs_cache *squashfs_cache_init(char *name, int entries,
237 struct squashfs_cache *cache = kzalloc(sizeof(*cache), GFP_KERNEL);
240 ERROR("Failed to allocate %s cache\n", name);
244 cache->entry = kcalloc(entries, sizeof(*(cache->entry)), GFP_KERNEL);
245 if (cache->entry == NULL) {
246 ERROR("Failed to allocate %s cache\n", name);
251 cache->unused = entries;
252 cache->entries = entries;
253 cache->block_size = block_size;
254 cache->pages = block_size >> PAGE_CACHE_SHIFT;
255 cache->pages = cache->pages ? cache->pages : 1;
257 cache->num_waiters = 0;
258 spin_lock_init(&cache->lock);
259 init_waitqueue_head(&cache->wait_queue);
261 for (i = 0; i < entries; i++) {
262 struct squashfs_cache_entry *entry = &cache->entry[i];
264 init_waitqueue_head(&cache->entry[i].wait_queue);
265 entry->cache = cache;
266 entry->block = SQUASHFS_INVALID_BLK;
267 entry->data = kcalloc(cache->pages, sizeof(void *), GFP_KERNEL);
268 if (entry->data == NULL) {
269 ERROR("Failed to allocate %s cache entry\n", name);
273 for (j = 0; j < cache->pages; j++) {
274 entry->data[j] = kmalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
275 if (entry->data[j] == NULL) {
276 ERROR("Failed to allocate %s buffer\n", name);
285 squashfs_cache_delete(cache);
291 * Copy upto length bytes from cache entry to buffer starting at offset bytes
292 * into the cache entry. If there's not length bytes then copy the number of
293 * bytes available. In all cases return the number of bytes copied.
295 int squashfs_copy_data(void *buffer, struct squashfs_cache_entry *entry,
296 int offset, int length)
298 int remaining = length;
302 else if (buffer == NULL)
303 return min(length, entry->length - offset);
305 while (offset < entry->length) {
306 void *buff = entry->data[offset / PAGE_CACHE_SIZE]
307 + (offset % PAGE_CACHE_SIZE);
308 int bytes = min_t(int, entry->length - offset,
309 PAGE_CACHE_SIZE - (offset % PAGE_CACHE_SIZE));
311 if (bytes >= remaining) {
312 memcpy(buffer, buff, remaining);
317 memcpy(buffer, buff, bytes);
323 return length - remaining;
328 * Read length bytes from metadata position <block, offset> (block is the
329 * start of the compressed block on disk, and offset is the offset into
330 * the block once decompressed). Data is packed into consecutive blocks,
331 * and length bytes may require reading more than one block.
333 int squashfs_read_metadata(struct super_block *sb, void *buffer,
334 u64 *block, int *offset, int length)
336 struct squashfs_sb_info *msblk = sb->s_fs_info;
337 int bytes, copied = length;
338 struct squashfs_cache_entry *entry;
340 TRACE("Entered squashfs_read_metadata [%llx:%x]\n", *block, *offset);
343 entry = squashfs_cache_get(sb, msblk->block_cache, *block, 0);
346 else if (*offset >= entry->length)
349 bytes = squashfs_copy_data(buffer, entry, *offset, length);
355 if (*offset == entry->length) {
356 *block = entry->next_index;
360 squashfs_cache_put(entry);
368 * Look-up in the fragmment cache the fragment located at <start_block> in the
369 * filesystem. If necessary read and decompress it from disk.
371 struct squashfs_cache_entry *squashfs_get_fragment(struct super_block *sb,
372 u64 start_block, int length)
374 struct squashfs_sb_info *msblk = sb->s_fs_info;
376 return squashfs_cache_get(sb, msblk->fragment_cache, start_block,
382 * Read and decompress the datablock located at <start_block> in the
383 * filesystem. The cache is used here to avoid duplicating locking and
384 * read/decompress code.
386 struct squashfs_cache_entry *squashfs_get_datablock(struct super_block *sb,
387 u64 start_block, int length)
389 struct squashfs_sb_info *msblk = sb->s_fs_info;
391 return squashfs_cache_get(sb, msblk->read_page, start_block, length);
396 * Read a filesystem table (uncompressed sequence of bytes) from disk
398 int squashfs_read_table(struct super_block *sb, void *buffer, u64 block,
401 int pages = (length + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
403 void **data = kcalloc(pages, sizeof(void *), GFP_KERNEL);
407 for (i = 0; i < pages; i++, buffer += PAGE_CACHE_SIZE)
409 res = squashfs_read_data(sb, data, block, length |
410 SQUASHFS_COMPRESSED_BIT_BLOCK, NULL, length, pages);