4 * This code provides the generic "frontend" layer to call a matching
5 * "backend" driver implementation of cleancache. See
6 * Documentation/vm/cleancache.txt for more information.
8 * Copyright (C) 2009-2010 Oracle Corp. All rights reserved.
9 * Author: Dan Magenheimer
11 * This work is licensed under the terms of the GNU GPL, version 2.
14 #include <linux/module.h>
16 #include <linux/exportfs.h>
18 #include <linux/debugfs.h>
19 #include <linux/cleancache.h>
22 * cleancache_ops is set by cleancache_ops_register to contain the pointers
23 * to the cleancache "backend" implementation functions.
25 static struct cleancache_ops *cleancache_ops __read_mostly;
28 * Counters available via /sys/kernel/debug/cleancache (if debugfs is
29 * properly configured. These are for information only so are not protected
30 * against increment races.
32 static u64 cleancache_succ_gets;
33 static u64 cleancache_failed_gets;
34 static u64 cleancache_puts;
35 static u64 cleancache_invalidates;
38 * When no backend is registered all calls to init_fs and init_shared_fs
39 * are registered and fake poolids (FAKE_FS_POOLID_OFFSET or
40 * FAKE_SHARED_FS_POOLID_OFFSET, plus offset in the respective array
41 * [shared_|]fs_poolid_map) are given to the respective super block
42 * (sb->cleancache_poolid) and no tmem_pools are created. When a backend
43 * registers with cleancache the previous calls to init_fs and init_shared_fs
44 * are executed to create tmem_pools and set the respective poolids. While no
45 * backend is registered all "puts", "gets" and "flushes" are ignored or failed.
47 #define MAX_INITIALIZABLE_FS 32
48 #define FAKE_FS_POOLID_OFFSET 1000
49 #define FAKE_SHARED_FS_POOLID_OFFSET 2000
51 #define FS_NO_BACKEND (-1)
52 #define FS_UNKNOWN (-2)
53 static int fs_poolid_map[MAX_INITIALIZABLE_FS];
54 static int shared_fs_poolid_map[MAX_INITIALIZABLE_FS];
55 static char *uuids[MAX_INITIALIZABLE_FS];
57 * Mutex for the [shared_|]fs_poolid_map to guard against multiple threads
58 * invoking umount (and ending in __cleancache_invalidate_fs) and also multiple
59 * threads calling mount (and ending up in __cleancache_init_[shared|]fs).
61 static DEFINE_MUTEX(poolid_mutex);
63 * When set to false (default) all calls to the cleancache functions, except
64 * the __cleancache_invalidate_fs and __cleancache_init_[shared|]fs are guarded
65 * by the if (!cleancache_ops) return. This means multiple threads (from
66 * different filesystems) will be checking cleancache_ops. The usage of a
67 * bool instead of a atomic_t or a bool guarded by a spinlock is OK - we are
68 * OK if the time between the backend's have been initialized (and
69 * cleancache_ops has been set to not NULL) and when the filesystems start
70 * actually calling the backends. The inverse (when unloading) is obviously
71 * not good - but this shim does not do that (yet).
75 * The backends and filesystems work all asynchronously. This is b/c the
76 * backends can be built as modules.
77 * The usual sequence of events is:
78 * a) mount / -> __cleancache_init_fs is called. We set the
79 * [shared_|]fs_poolid_map and uuids for.
81 * b). user does I/Os -> we call the rest of __cleancache_* functions
82 * which return immediately as cleancache_ops is false.
84 * c). modprobe zcache -> cleancache_register_ops. We init the backend
85 * and set cleancache_ops to true, and for any fs_poolid_map
86 * (which is set by __cleancache_init_fs) we initialize the poolid.
88 * d). user does I/Os -> now that cleancache_ops is true all the
89 * __cleancache_* functions can call the backend. They all check
90 * that fs_poolid_map is valid and if so invoke the backend.
92 * e). umount / -> __cleancache_invalidate_fs, the fs_poolid_map is
93 * reset (which is the second check in the __cleancache_* ops
94 * to call the backend).
96 * The sequence of event could also be c), followed by a), and d). and e). The
97 * c) would not happen anymore. There is also the chance of c), and one thread
98 * doing a) + d), and another doing e). For that case we depend on the
99 * filesystem calling __cleancache_invalidate_fs in the proper sequence (so
100 * that it handles all I/Os before it invalidates the fs (which is last part
101 * of unmounting process).
103 * Note: The acute reader will notice that there is no "rmmod zcache" case.
104 * This is b/c the functionality for that is not yet implemented and when
105 * done, will require some extra locking not yet devised.
109 * Register operations for cleancache. Returns 0 on success.
111 int cleancache_register_ops(struct cleancache_ops *ops)
115 mutex_lock(&poolid_mutex);
116 if (cleancache_ops) {
117 mutex_unlock(&poolid_mutex);
120 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
121 if (fs_poolid_map[i] == FS_NO_BACKEND)
122 fs_poolid_map[i] = ops->init_fs(PAGE_SIZE);
123 if (shared_fs_poolid_map[i] == FS_NO_BACKEND)
124 shared_fs_poolid_map[i] = ops->init_shared_fs
125 (uuids[i], PAGE_SIZE);
128 * We MUST set cleancache_ops _after_ we have called the backends
129 * init_fs or init_shared_fs functions. Otherwise the compiler might
130 * re-order where cleancache_ops is set in this function.
133 cleancache_ops = ops;
134 mutex_unlock(&poolid_mutex);
137 EXPORT_SYMBOL(cleancache_register_ops);
139 /* Called by a cleancache-enabled filesystem at time of mount */
140 void __cleancache_init_fs(struct super_block *sb)
144 mutex_lock(&poolid_mutex);
145 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
146 if (fs_poolid_map[i] == FS_UNKNOWN) {
147 sb->cleancache_poolid = i + FAKE_FS_POOLID_OFFSET;
149 fs_poolid_map[i] = cleancache_ops->init_fs(PAGE_SIZE);
151 fs_poolid_map[i] = FS_NO_BACKEND;
155 mutex_unlock(&poolid_mutex);
157 EXPORT_SYMBOL(__cleancache_init_fs);
159 /* Called by a cleancache-enabled clustered filesystem at time of mount */
160 void __cleancache_init_shared_fs(struct super_block *sb)
164 mutex_lock(&poolid_mutex);
165 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
166 if (shared_fs_poolid_map[i] == FS_UNKNOWN) {
167 sb->cleancache_poolid = i + FAKE_SHARED_FS_POOLID_OFFSET;
168 uuids[i] = sb->s_uuid;
170 shared_fs_poolid_map[i] = cleancache_ops->init_shared_fs
171 (sb->s_uuid, PAGE_SIZE);
173 shared_fs_poolid_map[i] = FS_NO_BACKEND;
177 mutex_unlock(&poolid_mutex);
179 EXPORT_SYMBOL(__cleancache_init_shared_fs);
182 * If the filesystem uses exportable filehandles, use the filehandle as
183 * the key, else use the inode number.
185 static int cleancache_get_key(struct inode *inode,
186 struct cleancache_filekey *key)
188 int (*fhfn)(struct inode *, __u32 *fh, int *, struct inode *);
189 int len = 0, maxlen = CLEANCACHE_KEY_MAX;
190 struct super_block *sb = inode->i_sb;
192 key->u.ino = inode->i_ino;
193 if (sb->s_export_op != NULL) {
194 fhfn = sb->s_export_op->encode_fh;
196 len = (*fhfn)(inode, &key->u.fh[0], &maxlen, NULL);
197 if (len <= FILEID_ROOT || len == FILEID_INVALID)
199 if (maxlen > CLEANCACHE_KEY_MAX)
207 * Returns a pool_id that is associated with a given fake poolid.
209 static int get_poolid_from_fake(int fake_pool_id)
211 if (fake_pool_id >= FAKE_SHARED_FS_POOLID_OFFSET)
212 return shared_fs_poolid_map[fake_pool_id -
213 FAKE_SHARED_FS_POOLID_OFFSET];
214 else if (fake_pool_id >= FAKE_FS_POOLID_OFFSET)
215 return fs_poolid_map[fake_pool_id - FAKE_FS_POOLID_OFFSET];
216 return FS_NO_BACKEND;
220 * "Get" data from cleancache associated with the poolid/inode/index
221 * that were specified when the data was put to cleanache and, if
222 * successful, use it to fill the specified page with data and return 0.
223 * The pageframe is unchanged and returns -1 if the get fails.
224 * Page must be locked by caller.
226 * The function has two checks before any action is taken - whether
227 * a backend is registered and whether the sb->cleancache_poolid
230 int __cleancache_get_page(struct page *page)
235 struct cleancache_filekey key = { .u.key = { 0 } };
237 if (!cleancache_ops) {
238 cleancache_failed_gets++;
242 VM_BUG_ON_PAGE(!PageLocked(page), page);
243 fake_pool_id = page->mapping->host->i_sb->cleancache_poolid;
244 if (fake_pool_id < 0)
246 pool_id = get_poolid_from_fake(fake_pool_id);
248 if (cleancache_get_key(page->mapping->host, &key) < 0)
252 ret = cleancache_ops->get_page(pool_id,
253 key, page->index, page);
255 cleancache_succ_gets++;
257 cleancache_failed_gets++;
261 EXPORT_SYMBOL(__cleancache_get_page);
264 * "Put" data from a page to cleancache and associate it with the
265 * (previously-obtained per-filesystem) poolid and the page's,
266 * inode and page index. Page must be locked. Note that a put_page
267 * always "succeeds", though a subsequent get_page may succeed or fail.
269 * The function has two checks before any action is taken - whether
270 * a backend is registered and whether the sb->cleancache_poolid
273 void __cleancache_put_page(struct page *page)
277 struct cleancache_filekey key = { .u.key = { 0 } };
279 if (!cleancache_ops) {
284 VM_BUG_ON_PAGE(!PageLocked(page), page);
285 fake_pool_id = page->mapping->host->i_sb->cleancache_poolid;
286 if (fake_pool_id < 0)
289 pool_id = get_poolid_from_fake(fake_pool_id);
292 cleancache_get_key(page->mapping->host, &key) >= 0) {
293 cleancache_ops->put_page(pool_id, key, page->index, page);
297 EXPORT_SYMBOL(__cleancache_put_page);
300 * Invalidate any data from cleancache associated with the poolid and the
301 * page's inode and page index so that a subsequent "get" will fail.
303 * The function has two checks before any action is taken - whether
304 * a backend is registered and whether the sb->cleancache_poolid
307 void __cleancache_invalidate_page(struct address_space *mapping,
310 /* careful... page->mapping is NULL sometimes when this is called */
312 int fake_pool_id = mapping->host->i_sb->cleancache_poolid;
313 struct cleancache_filekey key = { .u.key = { 0 } };
318 if (fake_pool_id >= 0) {
319 pool_id = get_poolid_from_fake(fake_pool_id);
323 VM_BUG_ON_PAGE(!PageLocked(page), page);
324 if (cleancache_get_key(mapping->host, &key) >= 0) {
325 cleancache_ops->invalidate_page(pool_id,
327 cleancache_invalidates++;
331 EXPORT_SYMBOL(__cleancache_invalidate_page);
334 * Invalidate all data from cleancache associated with the poolid and the
335 * mappings's inode so that all subsequent gets to this poolid/inode
338 * The function has two checks before any action is taken - whether
339 * a backend is registered and whether the sb->cleancache_poolid
342 void __cleancache_invalidate_inode(struct address_space *mapping)
345 int fake_pool_id = mapping->host->i_sb->cleancache_poolid;
346 struct cleancache_filekey key = { .u.key = { 0 } };
351 if (fake_pool_id < 0)
354 pool_id = get_poolid_from_fake(fake_pool_id);
356 if (pool_id >= 0 && cleancache_get_key(mapping->host, &key) >= 0)
357 cleancache_ops->invalidate_inode(pool_id, key);
359 EXPORT_SYMBOL(__cleancache_invalidate_inode);
362 * Called by any cleancache-enabled filesystem at time of unmount;
363 * note that pool_id is surrendered and may be returned by a subsequent
364 * cleancache_init_fs or cleancache_init_shared_fs.
366 void __cleancache_invalidate_fs(struct super_block *sb)
369 int fake_pool_id = sb->cleancache_poolid;
370 int old_poolid = fake_pool_id;
372 mutex_lock(&poolid_mutex);
373 if (fake_pool_id >= FAKE_SHARED_FS_POOLID_OFFSET) {
374 index = fake_pool_id - FAKE_SHARED_FS_POOLID_OFFSET;
375 old_poolid = shared_fs_poolid_map[index];
376 shared_fs_poolid_map[index] = FS_UNKNOWN;
378 } else if (fake_pool_id >= FAKE_FS_POOLID_OFFSET) {
379 index = fake_pool_id - FAKE_FS_POOLID_OFFSET;
380 old_poolid = fs_poolid_map[index];
381 fs_poolid_map[index] = FS_UNKNOWN;
383 sb->cleancache_poolid = -1;
385 cleancache_ops->invalidate_fs(old_poolid);
386 mutex_unlock(&poolid_mutex);
388 EXPORT_SYMBOL(__cleancache_invalidate_fs);
390 static int __init init_cleancache(void)
394 #ifdef CONFIG_DEBUG_FS
395 struct dentry *root = debugfs_create_dir("cleancache", NULL);
398 debugfs_create_u64("succ_gets", S_IRUGO, root, &cleancache_succ_gets);
399 debugfs_create_u64("failed_gets", S_IRUGO,
400 root, &cleancache_failed_gets);
401 debugfs_create_u64("puts", S_IRUGO, root, &cleancache_puts);
402 debugfs_create_u64("invalidates", S_IRUGO,
403 root, &cleancache_invalidates);
405 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
406 fs_poolid_map[i] = FS_UNKNOWN;
407 shared_fs_poolid_map[i] = FS_UNKNOWN;
411 module_init(init_cleancache)