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 * 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_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"
59 #include "page_actor.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 = cache->curr_blk, n = 0; n < cache->entries; n++) {
75 if (cache->entry[i].block == block) {
79 i = (i + 1) % cache->entries;
82 if (n == cache->entries) {
84 * Block not in cache, if all cache entries are used
85 * go to sleep waiting for one to become available.
87 if (cache->unused == 0) {
89 spin_unlock(&cache->lock);
90 wait_event(cache->wait_queue, cache->unused);
91 spin_lock(&cache->lock);
97 * At least one unused cache entry. A simple
98 * round-robin strategy is used to choose the entry to
99 * be evicted from the cache.
102 for (n = 0; n < cache->entries; n++) {
103 if (cache->entry[i].refcount == 0)
105 i = (i + 1) % cache->entries;
108 cache->next_blk = (i + 1) % cache->entries;
109 entry = &cache->entry[i];
112 * Initialise chosen cache entry, and fill it in from
116 entry->block = block;
119 entry->num_waiters = 0;
121 spin_unlock(&cache->lock);
123 entry->length = squashfs_read_data(sb, block, length,
124 &entry->next_index, entry->actor);
126 spin_lock(&cache->lock);
128 if (entry->length < 0)
129 entry->error = entry->length;
134 * While filling this entry one or more other processes
135 * have looked it up in the cache, and have slept
136 * waiting for it to become available.
138 if (entry->num_waiters) {
139 spin_unlock(&cache->lock);
140 wake_up_all(&entry->wait_queue);
142 spin_unlock(&cache->lock);
148 * Block already in cache. Increment refcount so it doesn't
149 * get reused until we're finished with it, if it was
150 * previously unused there's one less cache entry available
153 entry = &cache->entry[i];
154 if (entry->refcount == 0)
159 * If the entry is currently being filled in by another process
160 * go to sleep waiting for it to become available.
162 if (entry->pending) {
163 entry->num_waiters++;
164 spin_unlock(&cache->lock);
165 wait_event(entry->wait_queue, !entry->pending);
167 spin_unlock(&cache->lock);
173 TRACE("Got %s %d, start block %lld, refcount %d, error %d\n",
174 cache->name, i, entry->block, entry->refcount, entry->error);
177 ERROR("Unable to read %s cache entry [%llx]\n", cache->name,
184 * Release cache entry, once usage count is zero it can be reused.
186 void squashfs_cache_put(struct squashfs_cache_entry *entry)
188 struct squashfs_cache *cache = entry->cache;
190 spin_lock(&cache->lock);
192 if (entry->refcount == 0) {
195 * If there's any processes waiting for a block to become
196 * available, wake one up.
198 if (cache->num_waiters) {
199 spin_unlock(&cache->lock);
200 wake_up(&cache->wait_queue);
204 spin_unlock(&cache->lock);
208 * Delete cache reclaiming all kmalloced buffers.
210 void squashfs_cache_delete(struct squashfs_cache *cache)
217 for (i = 0; i < cache->entries; i++) {
218 if (cache->entry[i].data) {
219 for (j = 0; j < cache->pages; j++)
220 kfree(cache->entry[i].data[j]);
221 kfree(cache->entry[i].data);
223 kfree(cache->entry[i].actor);
232 * Initialise cache allocating the specified number of entries, each of
233 * size block_size. To avoid vmalloc fragmentation issues each entry
234 * is allocated as a sequence of kmalloced PAGE_SIZE buffers.
236 struct squashfs_cache *squashfs_cache_init(char *name, int entries,
240 struct squashfs_cache *cache = kzalloc(sizeof(*cache), GFP_KERNEL);
243 ERROR("Failed to allocate %s cache\n", name);
247 cache->entry = kcalloc(entries, sizeof(*(cache->entry)), GFP_KERNEL);
248 if (cache->entry == NULL) {
249 ERROR("Failed to allocate %s cache\n", name);
255 cache->unused = entries;
256 cache->entries = entries;
257 cache->block_size = block_size;
258 cache->pages = block_size >> PAGE_SHIFT;
259 cache->pages = cache->pages ? cache->pages : 1;
261 cache->num_waiters = 0;
262 spin_lock_init(&cache->lock);
263 init_waitqueue_head(&cache->wait_queue);
265 for (i = 0; i < entries; i++) {
266 struct squashfs_cache_entry *entry = &cache->entry[i];
268 init_waitqueue_head(&cache->entry[i].wait_queue);
269 entry->cache = cache;
270 entry->block = SQUASHFS_INVALID_BLK;
271 entry->data = kcalloc(cache->pages, sizeof(void *), GFP_KERNEL);
272 if (entry->data == NULL) {
273 ERROR("Failed to allocate %s cache entry\n", name);
277 for (j = 0; j < cache->pages; j++) {
278 entry->data[j] = kmalloc(PAGE_SIZE, GFP_KERNEL);
279 if (entry->data[j] == NULL) {
280 ERROR("Failed to allocate %s buffer\n", name);
285 entry->actor = squashfs_page_actor_init(entry->data,
287 if (entry->actor == NULL) {
288 ERROR("Failed to allocate %s cache entry\n", name);
296 squashfs_cache_delete(cache);
302 * Copy up to length bytes from cache entry to buffer starting at offset bytes
303 * into the cache entry. If there's not length bytes then copy the number of
304 * bytes available. In all cases return the number of bytes copied.
306 int squashfs_copy_data(void *buffer, struct squashfs_cache_entry *entry,
307 int offset, int length)
309 int remaining = length;
313 else if (buffer == NULL)
314 return min(length, entry->length - offset);
316 while (offset < entry->length) {
317 void *buff = entry->data[offset / PAGE_SIZE]
318 + (offset % PAGE_SIZE);
319 int bytes = min_t(int, entry->length - offset,
320 PAGE_SIZE - (offset % PAGE_SIZE));
322 if (bytes >= remaining) {
323 memcpy(buffer, buff, remaining);
328 memcpy(buffer, buff, bytes);
334 return length - remaining;
339 * Read length bytes from metadata position <block, offset> (block is the
340 * start of the compressed block on disk, and offset is the offset into
341 * the block once decompressed). Data is packed into consecutive blocks,
342 * and length bytes may require reading more than one block.
344 int squashfs_read_metadata(struct super_block *sb, void *buffer,
345 u64 *block, int *offset, int length)
347 struct squashfs_sb_info *msblk = sb->s_fs_info;
348 int bytes, res = length;
349 struct squashfs_cache_entry *entry;
351 TRACE("Entered squashfs_read_metadata [%llx:%x]\n", *block, *offset);
353 if (unlikely(length < 0))
357 entry = squashfs_cache_get(sb, msblk->block_cache, *block, 0);
361 } else if (*offset >= entry->length) {
366 bytes = squashfs_copy_data(buffer, entry, *offset, length);
372 if (*offset == entry->length) {
373 *block = entry->next_index;
377 squashfs_cache_put(entry);
383 squashfs_cache_put(entry);
389 * Look-up in the fragmment cache the fragment located at <start_block> in the
390 * filesystem. If necessary read and decompress it from disk.
392 struct squashfs_cache_entry *squashfs_get_fragment(struct super_block *sb,
393 u64 start_block, int length)
395 struct squashfs_sb_info *msblk = sb->s_fs_info;
397 return squashfs_cache_get(sb, msblk->fragment_cache, start_block,
403 * Read and decompress the datablock located at <start_block> in the
404 * filesystem. The cache is used here to avoid duplicating locking and
405 * read/decompress code.
407 struct squashfs_cache_entry *squashfs_get_datablock(struct super_block *sb,
408 u64 start_block, int length)
410 struct squashfs_sb_info *msblk = sb->s_fs_info;
412 return squashfs_cache_get(sb, msblk->read_page, start_block, length);
417 * Read a filesystem table (uncompressed sequence of bytes) from disk
419 void *squashfs_read_table(struct super_block *sb, u64 block, int length)
421 int pages = (length + PAGE_SIZE - 1) >> PAGE_SHIFT;
423 void *table, *buffer, **data;
424 struct squashfs_page_actor *actor;
426 table = buffer = kmalloc(length, GFP_KERNEL);
428 return ERR_PTR(-ENOMEM);
430 data = kcalloc(pages, sizeof(void *), GFP_KERNEL);
436 actor = squashfs_page_actor_init(data, pages, length);
442 for (i = 0; i < pages; i++, buffer += PAGE_SIZE)
445 res = squashfs_read_data(sb, block, length |
446 SQUASHFS_COMPRESSED_BIT_BLOCK, NULL, actor);