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 issues 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/pagemap.h>
56 #include "squashfs_fs.h"
57 #include "squashfs_fs_sb.h"
61 * Look-up block in cache, and increment usage count. If not in cache, read
62 * and decompress it from disk.
64 struct squashfs_cache_entry *squashfs_cache_get(struct super_block *sb,
65 struct squashfs_cache *cache, u64 block, int length)
68 struct squashfs_cache_entry *entry;
70 spin_lock(&cache->lock);
73 for (i = 0; i < cache->entries; i++)
74 if (cache->entry[i].block == block)
77 if (i == cache->entries) {
79 * Block not in cache, if all cache entries are used
80 * go to sleep waiting for one to become available.
82 if (cache->unused == 0) {
84 spin_unlock(&cache->lock);
85 wait_event(cache->wait_queue, cache->unused);
86 spin_lock(&cache->lock);
92 * At least one unused cache entry. A simple
93 * round-robin strategy is used to choose the entry to
94 * be evicted from the cache.
97 for (n = 0; n < cache->entries; n++) {
98 if (cache->entry[i].refcount == 0)
100 i = (i + 1) % cache->entries;
103 cache->next_blk = (i + 1) % cache->entries;
104 entry = &cache->entry[i];
107 * Initialise chosen cache entry, and fill it in from
111 entry->block = block;
114 entry->num_waiters = 0;
116 spin_unlock(&cache->lock);
118 entry->length = squashfs_read_data(sb, entry->data,
119 block, length, &entry->next_index,
120 cache->block_size, cache->pages);
122 spin_lock(&cache->lock);
124 if (entry->length < 0)
125 entry->error = entry->length;
130 * While filling this entry one or more other processes
131 * have looked it up in the cache, and have slept
132 * waiting for it to become available.
134 if (entry->num_waiters) {
135 spin_unlock(&cache->lock);
136 wake_up_all(&entry->wait_queue);
138 spin_unlock(&cache->lock);
144 * Block already in cache. Increment refcount so it doesn't
145 * get reused until we're finished with it, if it was
146 * previously unused there's one less cache entry available
149 entry = &cache->entry[i];
150 if (entry->refcount == 0)
155 * If the entry is currently being filled in by another process
156 * go to sleep waiting for it to become available.
158 if (entry->pending) {
159 entry->num_waiters++;
160 spin_unlock(&cache->lock);
161 wait_event(entry->wait_queue, !entry->pending);
163 spin_unlock(&cache->lock);
169 TRACE("Got %s %d, start block %lld, refcount %d, error %d\n",
170 cache->name, i, entry->block, entry->refcount, entry->error);
173 ERROR("Unable to read %s cache entry [%llx]\n", cache->name,
180 * Release cache entry, once usage count is zero it can be reused.
182 void squashfs_cache_put(struct squashfs_cache_entry *entry)
184 struct squashfs_cache *cache = entry->cache;
186 spin_lock(&cache->lock);
188 if (entry->refcount == 0) {
191 * If there's any processes waiting for a block to become
192 * available, wake one up.
194 if (cache->num_waiters) {
195 spin_unlock(&cache->lock);
196 wake_up(&cache->wait_queue);
200 spin_unlock(&cache->lock);
204 * Delete cache reclaiming all kmalloced buffers.
206 void squashfs_cache_delete(struct squashfs_cache *cache)
213 for (i = 0; i < cache->entries; i++) {
214 if (cache->entry[i].data) {
215 for (j = 0; j < cache->pages; j++)
216 kfree(cache->entry[i].data[j]);
217 kfree(cache->entry[i].data);
227 * Initialise cache allocating the specified number of entries, each of
228 * size block_size. To avoid vmalloc fragmentation issues each entry
229 * is allocated as a sequence of kmalloced PAGE_CACHE_SIZE buffers.
231 struct squashfs_cache *squashfs_cache_init(char *name, int entries,
235 struct squashfs_cache *cache = kzalloc(sizeof(*cache), GFP_KERNEL);
238 ERROR("Failed to allocate %s cache\n", name);
242 cache->entry = kcalloc(entries, sizeof(*(cache->entry)), GFP_KERNEL);
243 if (cache->entry == NULL) {
244 ERROR("Failed to allocate %s cache\n", name);
249 cache->unused = entries;
250 cache->entries = entries;
251 cache->block_size = block_size;
252 cache->pages = block_size >> PAGE_CACHE_SHIFT;
253 cache->pages = cache->pages ? cache->pages : 1;
255 cache->num_waiters = 0;
256 spin_lock_init(&cache->lock);
257 init_waitqueue_head(&cache->wait_queue);
259 for (i = 0; i < entries; i++) {
260 struct squashfs_cache_entry *entry = &cache->entry[i];
262 init_waitqueue_head(&cache->entry[i].wait_queue);
263 entry->cache = cache;
264 entry->block = SQUASHFS_INVALID_BLK;
265 entry->data = kcalloc(cache->pages, sizeof(void *), GFP_KERNEL);
266 if (entry->data == NULL) {
267 ERROR("Failed to allocate %s cache entry\n", name);
271 for (j = 0; j < cache->pages; j++) {
272 entry->data[j] = kmalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
273 if (entry->data[j] == NULL) {
274 ERROR("Failed to allocate %s buffer\n", name);
283 squashfs_cache_delete(cache);
289 * Copy up to length bytes from cache entry to buffer starting at offset bytes
290 * into the cache entry. If there's not length bytes then copy the number of
291 * bytes available. In all cases return the number of bytes copied.
293 int squashfs_copy_data(void *buffer, struct squashfs_cache_entry *entry,
294 int offset, int length)
296 int remaining = length;
300 else if (buffer == NULL)
301 return min(length, entry->length - offset);
303 while (offset < entry->length) {
304 void *buff = entry->data[offset / PAGE_CACHE_SIZE]
305 + (offset % PAGE_CACHE_SIZE);
306 int bytes = min_t(int, entry->length - offset,
307 PAGE_CACHE_SIZE - (offset % PAGE_CACHE_SIZE));
309 if (bytes >= remaining) {
310 memcpy(buffer, buff, remaining);
315 memcpy(buffer, buff, bytes);
321 return length - remaining;
326 * Read length bytes from metadata position <block, offset> (block is the
327 * start of the compressed block on disk, and offset is the offset into
328 * the block once decompressed). Data is packed into consecutive blocks,
329 * and length bytes may require reading more than one block.
331 int squashfs_read_metadata(struct super_block *sb, void *buffer,
332 u64 *block, int *offset, int length)
334 struct squashfs_sb_info *msblk = sb->s_fs_info;
335 int bytes, copied = length;
336 struct squashfs_cache_entry *entry;
338 TRACE("Entered squashfs_read_metadata [%llx:%x]\n", *block, *offset);
341 entry = squashfs_cache_get(sb, msblk->block_cache, *block, 0);
344 else if (*offset >= entry->length)
347 bytes = squashfs_copy_data(buffer, entry, *offset, length);
353 if (*offset == entry->length) {
354 *block = entry->next_index;
358 squashfs_cache_put(entry);
366 * Look-up in the fragmment cache the fragment located at <start_block> in the
367 * filesystem. If necessary read and decompress it from disk.
369 struct squashfs_cache_entry *squashfs_get_fragment(struct super_block *sb,
370 u64 start_block, int length)
372 struct squashfs_sb_info *msblk = sb->s_fs_info;
374 return squashfs_cache_get(sb, msblk->fragment_cache, start_block,
380 * Read and decompress the datablock located at <start_block> in the
381 * filesystem. The cache is used here to avoid duplicating locking and
382 * read/decompress code.
384 struct squashfs_cache_entry *squashfs_get_datablock(struct super_block *sb,
385 u64 start_block, int length)
387 struct squashfs_sb_info *msblk = sb->s_fs_info;
389 return squashfs_cache_get(sb, msblk->read_page, start_block, length);
394 * Read a filesystem table (uncompressed sequence of bytes) from disk
396 int squashfs_read_table(struct super_block *sb, void *buffer, u64 block,
399 int pages = (length + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
401 void **data = kcalloc(pages, sizeof(void *), GFP_KERNEL);
405 for (i = 0; i < pages; i++, buffer += PAGE_CACHE_SIZE)
407 res = squashfs_read_data(sb, data, block, length |
408 SQUASHFS_COMPRESSED_BIT_BLOCK, NULL, length, pages);