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/pagemap.h>
56 #include "squashfs_fs.h"
57 #include "squashfs_fs_sb.h"
58 #include "squashfs_fs_i.h"
62 * Look-up block in cache, and increment usage count. If not in cache, read
63 * and decompress it from disk.
65 struct squashfs_cache_entry *squashfs_cache_get(struct super_block *sb,
66 struct squashfs_cache *cache, u64 block, int length)
69 struct squashfs_cache_entry *entry;
71 spin_lock(&cache->lock);
74 for (i = 0; i < cache->entries; i++)
75 if (cache->entry[i].block == block)
78 if (i == cache->entries) {
80 * Block not in cache, if all cache entries are used
81 * go to sleep waiting for one to become available.
83 if (cache->unused == 0) {
85 spin_unlock(&cache->lock);
86 wait_event(cache->wait_queue, cache->unused);
87 spin_lock(&cache->lock);
93 * At least one unused cache entry. A simple
94 * round-robin strategy is used to choose the entry to
95 * be evicted from the cache.
98 for (n = 0; n < cache->entries; n++) {
99 if (cache->entry[i].refcount == 0)
101 i = (i + 1) % cache->entries;
104 cache->next_blk = (i + 1) % cache->entries;
105 entry = &cache->entry[i];
108 * Initialise choosen cache entry, and fill it in from
112 entry->block = block;
115 entry->num_waiters = 0;
117 spin_unlock(&cache->lock);
119 entry->length = squashfs_read_data(sb, entry->data,
120 block, length, &entry->next_index,
121 cache->block_size, cache->pages);
123 spin_lock(&cache->lock);
125 if (entry->length < 0)
126 entry->error = entry->length;
131 * While filling this entry one or more other processes
132 * have looked it up in the cache, and have slept
133 * waiting for it to become available.
135 if (entry->num_waiters) {
136 spin_unlock(&cache->lock);
137 wake_up_all(&entry->wait_queue);
139 spin_unlock(&cache->lock);
145 * Block already in cache. Increment refcount so it doesn't
146 * get reused until we're finished with it, if it was
147 * previously unused there's one less cache entry available
150 entry = &cache->entry[i];
151 if (entry->refcount == 0)
156 * If the entry is currently being filled in by another process
157 * go to sleep waiting for it to become available.
159 if (entry->pending) {
160 entry->num_waiters++;
161 spin_unlock(&cache->lock);
162 wait_event(entry->wait_queue, !entry->pending);
164 spin_unlock(&cache->lock);
170 TRACE("Got %s %d, start block %lld, refcount %d, error %d\n",
171 cache->name, i, entry->block, entry->refcount, entry->error);
174 ERROR("Unable to read %s cache entry [%llx]\n", cache->name,
181 * Release cache entry, once usage count is zero it can be reused.
183 void squashfs_cache_put(struct squashfs_cache_entry *entry)
185 struct squashfs_cache *cache = entry->cache;
187 spin_lock(&cache->lock);
189 if (entry->refcount == 0) {
192 * If there's any processes waiting for a block to become
193 * available, wake one up.
195 if (cache->num_waiters) {
196 spin_unlock(&cache->lock);
197 wake_up(&cache->wait_queue);
201 spin_unlock(&cache->lock);
205 * Delete cache reclaiming all kmalloced buffers.
207 void squashfs_cache_delete(struct squashfs_cache *cache)
214 for (i = 0; i < cache->entries; i++) {
215 if (cache->entry[i].data) {
216 for (j = 0; j < cache->pages; j++)
217 kfree(cache->entry[i].data[j]);
218 kfree(cache->entry[i].data);
228 * Initialise cache allocating the specified number of entries, each of
229 * size block_size. To avoid vmalloc fragmentation issues each entry
230 * is allocated as a sequence of kmalloced PAGE_CACHE_SIZE buffers.
232 struct squashfs_cache *squashfs_cache_init(char *name, int entries,
236 struct squashfs_cache *cache = kzalloc(sizeof(*cache), GFP_KERNEL);
239 ERROR("Failed to allocate %s cache\n", name);
243 cache->entry = kcalloc(entries, sizeof(*(cache->entry)), GFP_KERNEL);
244 if (cache->entry == NULL) {
245 ERROR("Failed to allocate %s cache\n", name);
250 cache->unused = entries;
251 cache->entries = entries;
252 cache->block_size = block_size;
253 cache->pages = block_size >> PAGE_CACHE_SHIFT;
254 cache->pages = cache->pages ? cache->pages : 1;
256 cache->num_waiters = 0;
257 spin_lock_init(&cache->lock);
258 init_waitqueue_head(&cache->wait_queue);
260 for (i = 0; i < entries; i++) {
261 struct squashfs_cache_entry *entry = &cache->entry[i];
263 init_waitqueue_head(&cache->entry[i].wait_queue);
264 entry->cache = cache;
265 entry->block = SQUASHFS_INVALID_BLK;
266 entry->data = kcalloc(cache->pages, sizeof(void *), GFP_KERNEL);
267 if (entry->data == NULL) {
268 ERROR("Failed to allocate %s cache entry\n", name);
272 for (j = 0; j < cache->pages; j++) {
273 entry->data[j] = kmalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
274 if (entry->data[j] == NULL) {
275 ERROR("Failed to allocate %s buffer\n", name);
284 squashfs_cache_delete(cache);
290 * Copy upto length bytes from cache entry to buffer starting at offset bytes
291 * into the cache entry. If there's not length bytes then copy the number of
292 * bytes available. In all cases return the number of bytes copied.
294 int squashfs_copy_data(void *buffer, struct squashfs_cache_entry *entry,
295 int offset, int length)
297 int remaining = length;
301 else if (buffer == NULL)
302 return min(length, entry->length - offset);
304 while (offset < entry->length) {
305 void *buff = entry->data[offset / PAGE_CACHE_SIZE]
306 + (offset % PAGE_CACHE_SIZE);
307 int bytes = min_t(int, entry->length - offset,
308 PAGE_CACHE_SIZE - (offset % PAGE_CACHE_SIZE));
310 if (bytes >= remaining) {
311 memcpy(buffer, buff, remaining);
316 memcpy(buffer, buff, bytes);
322 return length - remaining;
327 * Read length bytes from metadata position <block, offset> (block is the
328 * start of the compressed block on disk, and offset is the offset into
329 * the block once decompressed). Data is packed into consecutive blocks,
330 * and length bytes may require reading more than one block.
332 int squashfs_read_metadata(struct super_block *sb, void *buffer,
333 u64 *block, int *offset, int length)
335 struct squashfs_sb_info *msblk = sb->s_fs_info;
336 int bytes, copied = length;
337 struct squashfs_cache_entry *entry;
339 TRACE("Entered squashfs_read_metadata [%llx:%x]\n", *block, *offset);
342 entry = squashfs_cache_get(sb, msblk->block_cache, *block, 0);
345 else if (*offset >= entry->length)
348 bytes = squashfs_copy_data(buffer, entry, *offset, length);
354 if (*offset == entry->length) {
355 *block = entry->next_index;
359 squashfs_cache_put(entry);
367 * Look-up in the fragmment cache the fragment located at <start_block> in the
368 * filesystem. If necessary read and decompress it from disk.
370 struct squashfs_cache_entry *squashfs_get_fragment(struct super_block *sb,
371 u64 start_block, int length)
373 struct squashfs_sb_info *msblk = sb->s_fs_info;
375 return squashfs_cache_get(sb, msblk->fragment_cache, start_block,
381 * Read and decompress the datablock located at <start_block> in the
382 * filesystem. The cache is used here to avoid duplicating locking and
383 * read/decompress code.
385 struct squashfs_cache_entry *squashfs_get_datablock(struct super_block *sb,
386 u64 start_block, int length)
388 struct squashfs_sb_info *msblk = sb->s_fs_info;
390 return squashfs_cache_get(sb, msblk->read_page, start_block, length);
395 * Read a filesystem table (uncompressed sequence of bytes) from disk
397 int squashfs_read_table(struct super_block *sb, void *buffer, u64 block,
400 int pages = (length + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
402 void **data = kcalloc(pages, sizeof(void *), GFP_KERNEL);
406 for (i = 0; i < pages; i++, buffer += PAGE_CACHE_SIZE)
408 res = squashfs_read_data(sb, data, block, length |
409 SQUASHFS_COMPRESSED_BIT_BLOCK, NULL, length, pages);