• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Squashfs - a compressed read only filesystem for Linux
3  *
4  * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
5  * Phillip Lougher <phillip@lougher.demon.co.uk>
6  *
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.
11  *
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.
16  *
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.
20  *
21  * cache.c
22  */
23 
24 /*
25  * Blocks in Squashfs are compressed.  To avoid repeatedly decompressing
26  * recently accessed data Squashfs uses two small metadata and fragment caches.
27  *
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.
31  *
32  * To avoid out of memory and fragmentation isssues with vmalloc the cache
33  * uses sequences of kmalloced PAGE_CACHE_SIZE buffers.
34  *
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.
45  */
46 
47 #include <linux/fs.h>
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>
56 
57 #include "squashfs_fs.h"
58 #include "squashfs_fs_sb.h"
59 #include "squashfs_fs_i.h"
60 #include "squashfs.h"
61 
62 /*
63  * Look-up block in cache, and increment usage count.  If not in cache, read
64  * and decompress it from disk.
65  */
squashfs_cache_get(struct super_block * sb,struct squashfs_cache * cache,u64 block,int length)66 struct squashfs_cache_entry *squashfs_cache_get(struct super_block *sb,
67 	struct squashfs_cache *cache, u64 block, int length)
68 {
69 	int i, n;
70 	struct squashfs_cache_entry *entry;
71 
72 	spin_lock(&cache->lock);
73 
74 	while (1) {
75 		for (i = 0; i < cache->entries; i++)
76 			if (cache->entry[i].block == block)
77 				break;
78 
79 		if (i == cache->entries) {
80 			/*
81 			 * Block not in cache, if all cache entries are used
82 			 * go to sleep waiting for one to become available.
83 			 */
84 			if (cache->unused == 0) {
85 				cache->num_waiters++;
86 				spin_unlock(&cache->lock);
87 				wait_event(cache->wait_queue, cache->unused);
88 				spin_lock(&cache->lock);
89 				cache->num_waiters--;
90 				continue;
91 			}
92 
93 			/*
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.
97 			 */
98 			i = cache->next_blk;
99 			for (n = 0; n < cache->entries; n++) {
100 				if (cache->entry[i].refcount == 0)
101 					break;
102 				i = (i + 1) % cache->entries;
103 			}
104 
105 			cache->next_blk = (i + 1) % cache->entries;
106 			entry = &cache->entry[i];
107 
108 			/*
109 			 * Initialise choosen cache entry, and fill it in from
110 			 * disk.
111 			 */
112 			cache->unused--;
113 			entry->block = block;
114 			entry->refcount = 1;
115 			entry->pending = 1;
116 			entry->num_waiters = 0;
117 			entry->error = 0;
118 			spin_unlock(&cache->lock);
119 
120 			entry->length = squashfs_read_data(sb, entry->data,
121 				block, length, &entry->next_index,
122 				cache->block_size);
123 
124 			spin_lock(&cache->lock);
125 
126 			if (entry->length < 0)
127 				entry->error = entry->length;
128 
129 			entry->pending = 0;
130 
131 			/*
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.
135 			 */
136 			if (entry->num_waiters) {
137 				spin_unlock(&cache->lock);
138 				wake_up_all(&entry->wait_queue);
139 			} else
140 				spin_unlock(&cache->lock);
141 
142 			goto out;
143 		}
144 
145 		/*
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
149 		 * for reuse.
150 		 */
151 		entry = &cache->entry[i];
152 		if (entry->refcount == 0)
153 			cache->unused--;
154 		entry->refcount++;
155 
156 		/*
157 		 * If the entry is currently being filled in by another process
158 		 * go to sleep waiting for it to become available.
159 		 */
160 		if (entry->pending) {
161 			entry->num_waiters++;
162 			spin_unlock(&cache->lock);
163 			wait_event(entry->wait_queue, !entry->pending);
164 		} else
165 			spin_unlock(&cache->lock);
166 
167 		goto out;
168 	}
169 
170 out:
171 	TRACE("Got %s %d, start block %lld, refcount %d, error %d\n",
172 		cache->name, i, entry->block, entry->refcount, entry->error);
173 
174 	if (entry->error)
175 		ERROR("Unable to read %s cache entry [%llx]\n", cache->name,
176 							block);
177 	return entry;
178 }
179 
180 
181 /*
182  * Release cache entry, once usage count is zero it can be reused.
183  */
squashfs_cache_put(struct squashfs_cache_entry * entry)184 void squashfs_cache_put(struct squashfs_cache_entry *entry)
185 {
186 	struct squashfs_cache *cache = entry->cache;
187 
188 	spin_lock(&cache->lock);
189 	entry->refcount--;
190 	if (entry->refcount == 0) {
191 		cache->unused++;
192 		/*
193 		 * If there's any processes waiting for a block to become
194 		 * available, wake one up.
195 		 */
196 		if (cache->num_waiters) {
197 			spin_unlock(&cache->lock);
198 			wake_up(&cache->wait_queue);
199 			return;
200 		}
201 	}
202 	spin_unlock(&cache->lock);
203 }
204 
205 /*
206  * Delete cache reclaiming all kmalloced buffers.
207  */
squashfs_cache_delete(struct squashfs_cache * cache)208 void squashfs_cache_delete(struct squashfs_cache *cache)
209 {
210 	int i, j;
211 
212 	if (cache == NULL)
213 		return;
214 
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);
220 		}
221 	}
222 
223 	kfree(cache->entry);
224 	kfree(cache);
225 }
226 
227 
228 /*
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.
232  */
squashfs_cache_init(char * name,int entries,int block_size)233 struct squashfs_cache *squashfs_cache_init(char *name, int entries,
234 	int block_size)
235 {
236 	int i, j;
237 	struct squashfs_cache *cache = kzalloc(sizeof(*cache), GFP_KERNEL);
238 
239 	if (cache == NULL) {
240 		ERROR("Failed to allocate %s cache\n", name);
241 		return NULL;
242 	}
243 
244 	cache->entry = kcalloc(entries, sizeof(*(cache->entry)), GFP_KERNEL);
245 	if (cache->entry == NULL) {
246 		ERROR("Failed to allocate %s cache\n", name);
247 		goto cleanup;
248 	}
249 
250 	cache->next_blk = 0;
251 	cache->unused = entries;
252 	cache->entries = entries;
253 	cache->block_size = block_size;
254 	cache->pages = block_size >> PAGE_CACHE_SHIFT;
255 	cache->name = name;
256 	cache->num_waiters = 0;
257 	spin_lock_init(&cache->lock);
258 	init_waitqueue_head(&cache->wait_queue);
259 
260 	for (i = 0; i < entries; i++) {
261 		struct squashfs_cache_entry *entry = &cache->entry[i];
262 
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);
269 			goto cleanup;
270 		}
271 
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);
276 				goto cleanup;
277 			}
278 		}
279 	}
280 
281 	return cache;
282 
283 cleanup:
284 	squashfs_cache_delete(cache);
285 	return NULL;
286 }
287 
288 
289 /*
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.
293  */
squashfs_copy_data(void * buffer,struct squashfs_cache_entry * entry,int offset,int length)294 int squashfs_copy_data(void *buffer, struct squashfs_cache_entry *entry,
295 		int offset, int length)
296 {
297 	int remaining = length;
298 
299 	if (length == 0)
300 		return 0;
301 	else if (buffer == NULL)
302 		return min(length, entry->length - offset);
303 
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));
309 
310 		if (bytes >= remaining) {
311 			memcpy(buffer, buff, remaining);
312 			remaining = 0;
313 			break;
314 		}
315 
316 		memcpy(buffer, buff, bytes);
317 		buffer += bytes;
318 		remaining -= bytes;
319 		offset += bytes;
320 	}
321 
322 	return length - remaining;
323 }
324 
325 
326 /*
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.
331  */
squashfs_read_metadata(struct super_block * sb,void * buffer,u64 * block,int * offset,int length)332 int squashfs_read_metadata(struct super_block *sb, void *buffer,
333 		u64 *block, int *offset, int length)
334 {
335 	struct squashfs_sb_info *msblk = sb->s_fs_info;
336 	int bytes, copied = length;
337 	struct squashfs_cache_entry *entry;
338 
339 	TRACE("Entered squashfs_read_metadata [%llx:%x]\n", *block, *offset);
340 
341 	while (length) {
342 		entry = squashfs_cache_get(sb, msblk->block_cache, *block, 0);
343 		if (entry->error)
344 			return entry->error;
345 		else if (*offset >= entry->length)
346 			return -EIO;
347 
348 		bytes = squashfs_copy_data(buffer, entry, *offset, length);
349 		if (buffer)
350 			buffer += bytes;
351 		length -= bytes;
352 		*offset += bytes;
353 
354 		if (*offset == entry->length) {
355 			*block = entry->next_index;
356 			*offset = 0;
357 		}
358 
359 		squashfs_cache_put(entry);
360 	}
361 
362 	return copied;
363 }
364 
365 
366 /*
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.
369  */
squashfs_get_fragment(struct super_block * sb,u64 start_block,int length)370 struct squashfs_cache_entry *squashfs_get_fragment(struct super_block *sb,
371 				u64 start_block, int length)
372 {
373 	struct squashfs_sb_info *msblk = sb->s_fs_info;
374 
375 	return squashfs_cache_get(sb, msblk->fragment_cache, start_block,
376 		length);
377 }
378 
379 
380 /*
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.
384  */
squashfs_get_datablock(struct super_block * sb,u64 start_block,int length)385 struct squashfs_cache_entry *squashfs_get_datablock(struct super_block *sb,
386 				u64 start_block, int length)
387 {
388 	struct squashfs_sb_info *msblk = sb->s_fs_info;
389 
390 	return squashfs_cache_get(sb, msblk->read_page, start_block, length);
391 }
392 
393 
394 /*
395  * Read a filesystem table (uncompressed sequence of bytes) from disk
396  */
squashfs_read_table(struct super_block * sb,void * buffer,u64 block,int length)397 int squashfs_read_table(struct super_block *sb, void *buffer, u64 block,
398 	int length)
399 {
400 	int pages = (length + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
401 	int i, res;
402 	void **data = kcalloc(pages, sizeof(void *), GFP_KERNEL);
403 	if (data == NULL)
404 		return -ENOMEM;
405 
406 	for (i = 0; i < pages; i++, buffer += PAGE_CACHE_SIZE)
407 		data[i] = buffer;
408 	res = squashfs_read_data(sb, data, block, length |
409 		SQUASHFS_COMPRESSED_BIT_BLOCK, NULL, length);
410 	kfree(data);
411 	return res;
412 }
413