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 * This global enablement flag may be read thousands of times per second
23 * by cleancache_get/put/invalidate even on systems where cleancache_ops
24 * is not claimed (e.g. cleancache is config'ed on but remains
25 * disabled), so is preferred to the slower alternative: a function
26 * call that checks a non-global.
28 int cleancache_enabled __read_mostly;
29 EXPORT_SYMBOL(cleancache_enabled);
32 * cleancache_ops is set by cleancache_ops_register to contain the pointers
33 * to the cleancache "backend" implementation functions.
35 static struct cleancache_ops *cleancache_ops __read_mostly;
38 * Counters available via /sys/kernel/debug/frontswap (if debugfs is
39 * properly configured. These are for information only so are not protected
40 * against increment races.
42 static u64 cleancache_succ_gets;
43 static u64 cleancache_failed_gets;
44 static u64 cleancache_puts;
45 static u64 cleancache_invalidates;
48 * When no backend is registered all calls to init_fs and init_shared_fs
49 * are registered and fake poolids (FAKE_FS_POOLID_OFFSET or
50 * FAKE_SHARED_FS_POOLID_OFFSET, plus offset in the respective array
51 * [shared_|]fs_poolid_map) are given to the respective super block
52 * (sb->cleancache_poolid) and no tmem_pools are created. When a backend
53 * registers with cleancache the previous calls to init_fs and init_shared_fs
54 * are executed to create tmem_pools and set the respective poolids. While no
55 * backend is registered all "puts", "gets" and "flushes" are ignored or failed.
57 #define MAX_INITIALIZABLE_FS 32
58 #define FAKE_FS_POOLID_OFFSET 1000
59 #define FAKE_SHARED_FS_POOLID_OFFSET 2000
61 #define FS_NO_BACKEND (-1)
62 #define FS_UNKNOWN (-2)
63 static int fs_poolid_map[MAX_INITIALIZABLE_FS];
64 static int shared_fs_poolid_map[MAX_INITIALIZABLE_FS];
65 static char *uuids[MAX_INITIALIZABLE_FS];
67 * Mutex for the [shared_|]fs_poolid_map to guard against multiple threads
68 * invoking umount (and ending in __cleancache_invalidate_fs) and also multiple
69 * threads calling mount (and ending up in __cleancache_init_[shared|]fs).
71 static DEFINE_MUTEX(poolid_mutex);
73 * When set to false (default) all calls to the cleancache functions, except
74 * the __cleancache_invalidate_fs and __cleancache_init_[shared|]fs are guarded
75 * by the if (!cleancache_ops) return. This means multiple threads (from
76 * different filesystems) will be checking cleancache_ops. The usage of a
77 * bool instead of a atomic_t or a bool guarded by a spinlock is OK - we are
78 * OK if the time between the backend's have been initialized (and
79 * cleancache_ops has been set to not NULL) and when the filesystems start
80 * actually calling the backends. The inverse (when unloading) is obviously
81 * not good - but this shim does not do that (yet).
85 * The backends and filesystems work all asynchronously. This is b/c the
86 * backends can be built as modules.
87 * The usual sequence of events is:
88 * a) mount / -> __cleancache_init_fs is called. We set the
89 * [shared_|]fs_poolid_map and uuids for.
91 * b). user does I/Os -> we call the rest of __cleancache_* functions
92 * which return immediately as cleancache_ops is false.
94 * c). modprobe zcache -> cleancache_register_ops. We init the backend
95 * and set cleancache_ops to true, and for any fs_poolid_map
96 * (which is set by __cleancache_init_fs) we initialize the poolid.
98 * d). user does I/Os -> now that cleancache_ops is true all the
99 * __cleancache_* functions can call the backend. They all check
100 * that fs_poolid_map is valid and if so invoke the backend.
102 * e). umount / -> __cleancache_invalidate_fs, the fs_poolid_map is
103 * reset (which is the second check in the __cleancache_* ops
104 * to call the backend).
106 * The sequence of event could also be c), followed by a), and d). and e). The
107 * c) would not happen anymore. There is also the chance of c), and one thread
108 * doing a) + d), and another doing e). For that case we depend on the
109 * filesystem calling __cleancache_invalidate_fs in the proper sequence (so
110 * that it handles all I/Os before it invalidates the fs (which is last part
111 * of unmounting process).
113 * Note: The acute reader will notice that there is no "rmmod zcache" case.
114 * This is b/c the functionality for that is not yet implemented and when
115 * done, will require some extra locking not yet devised.
119 * Register operations for cleancache, returning previous thus allowing
120 * detection of multiple backends and possible nesting.
122 struct cleancache_ops *cleancache_register_ops(struct cleancache_ops *ops)
124 struct cleancache_ops *old = cleancache_ops;
127 mutex_lock(&poolid_mutex);
128 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
129 if (fs_poolid_map[i] == FS_NO_BACKEND)
130 fs_poolid_map[i] = ops->init_fs(PAGE_SIZE);
131 if (shared_fs_poolid_map[i] == FS_NO_BACKEND)
132 shared_fs_poolid_map[i] = ops->init_shared_fs
133 (uuids[i], PAGE_SIZE);
136 * We MUST set cleancache_ops _after_ we have called the backends
137 * init_fs or init_shared_fs functions. Otherwise the compiler might
138 * re-order where cleancache_ops is set in this function.
141 cleancache_ops = ops;
142 mutex_unlock(&poolid_mutex);
145 EXPORT_SYMBOL(cleancache_register_ops);
147 /* Called by a cleancache-enabled filesystem at time of mount */
148 void __cleancache_init_fs(struct super_block *sb)
152 mutex_lock(&poolid_mutex);
153 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
154 if (fs_poolid_map[i] == FS_UNKNOWN) {
155 sb->cleancache_poolid = i + FAKE_FS_POOLID_OFFSET;
157 fs_poolid_map[i] = cleancache_ops->init_fs(PAGE_SIZE);
159 fs_poolid_map[i] = FS_NO_BACKEND;
163 mutex_unlock(&poolid_mutex);
165 EXPORT_SYMBOL(__cleancache_init_fs);
167 /* Called by a cleancache-enabled clustered filesystem at time of mount */
168 void __cleancache_init_shared_fs(char *uuid, struct super_block *sb)
172 mutex_lock(&poolid_mutex);
173 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
174 if (shared_fs_poolid_map[i] == FS_UNKNOWN) {
175 sb->cleancache_poolid = i + FAKE_SHARED_FS_POOLID_OFFSET;
178 shared_fs_poolid_map[i] = cleancache_ops->init_shared_fs
181 shared_fs_poolid_map[i] = FS_NO_BACKEND;
185 mutex_unlock(&poolid_mutex);
187 EXPORT_SYMBOL(__cleancache_init_shared_fs);
190 * If the filesystem uses exportable filehandles, use the filehandle as
191 * the key, else use the inode number.
193 static int cleancache_get_key(struct inode *inode,
194 struct cleancache_filekey *key)
196 int (*fhfn)(struct inode *, __u32 *fh, int *, struct inode *);
197 int len = 0, maxlen = CLEANCACHE_KEY_MAX;
198 struct super_block *sb = inode->i_sb;
200 key->u.ino = inode->i_ino;
201 if (sb->s_export_op != NULL) {
202 fhfn = sb->s_export_op->encode_fh;
204 len = (*fhfn)(inode, &key->u.fh[0], &maxlen, NULL);
205 if (len <= FILEID_ROOT || len == FILEID_INVALID)
207 if (maxlen > CLEANCACHE_KEY_MAX)
215 * Returns a pool_id that is associated with a given fake poolid.
217 static int get_poolid_from_fake(int fake_pool_id)
219 if (fake_pool_id >= FAKE_SHARED_FS_POOLID_OFFSET)
220 return shared_fs_poolid_map[fake_pool_id -
221 FAKE_SHARED_FS_POOLID_OFFSET];
222 else if (fake_pool_id >= FAKE_FS_POOLID_OFFSET)
223 return fs_poolid_map[fake_pool_id - FAKE_FS_POOLID_OFFSET];
224 return FS_NO_BACKEND;
228 * "Get" data from cleancache associated with the poolid/inode/index
229 * that were specified when the data was put to cleanache and, if
230 * successful, use it to fill the specified page with data and return 0.
231 * The pageframe is unchanged and returns -1 if the get fails.
232 * Page must be locked by caller.
234 * The function has two checks before any action is taken - whether
235 * a backend is registered and whether the sb->cleancache_poolid
238 int __cleancache_get_page(struct page *page)
243 struct cleancache_filekey key = { .u.key = { 0 } };
245 if (!cleancache_ops) {
246 cleancache_failed_gets++;
250 VM_BUG_ON(!PageLocked(page));
251 fake_pool_id = page->mapping->host->i_sb->cleancache_poolid;
252 if (fake_pool_id < 0)
254 pool_id = get_poolid_from_fake(fake_pool_id);
256 if (cleancache_get_key(page->mapping->host, &key) < 0)
260 ret = cleancache_ops->get_page(pool_id,
261 key, page->index, page);
263 cleancache_succ_gets++;
265 cleancache_failed_gets++;
269 EXPORT_SYMBOL(__cleancache_get_page);
272 * "Put" data from a page to cleancache and associate it with the
273 * (previously-obtained per-filesystem) poolid and the page's,
274 * inode and page index. Page must be locked. Note that a put_page
275 * always "succeeds", though a subsequent get_page may succeed or fail.
277 * The function has two checks before any action is taken - whether
278 * a backend is registered and whether the sb->cleancache_poolid
281 void __cleancache_put_page(struct page *page)
285 struct cleancache_filekey key = { .u.key = { 0 } };
287 if (!cleancache_ops) {
292 VM_BUG_ON(!PageLocked(page));
293 fake_pool_id = page->mapping->host->i_sb->cleancache_poolid;
294 if (fake_pool_id < 0)
297 pool_id = get_poolid_from_fake(fake_pool_id);
300 cleancache_get_key(page->mapping->host, &key) >= 0) {
301 cleancache_ops->put_page(pool_id, key, page->index, page);
305 EXPORT_SYMBOL(__cleancache_put_page);
308 * Invalidate any data from cleancache associated with the poolid and the
309 * page's inode and page index so that a subsequent "get" will fail.
311 * The function has two checks before any action is taken - whether
312 * a backend is registered and whether the sb->cleancache_poolid
315 void __cleancache_invalidate_page(struct address_space *mapping,
318 /* careful... page->mapping is NULL sometimes when this is called */
320 int fake_pool_id = mapping->host->i_sb->cleancache_poolid;
321 struct cleancache_filekey key = { .u.key = { 0 } };
326 if (fake_pool_id >= 0) {
327 pool_id = get_poolid_from_fake(fake_pool_id);
331 VM_BUG_ON(!PageLocked(page));
332 if (cleancache_get_key(mapping->host, &key) >= 0) {
333 cleancache_ops->invalidate_page(pool_id,
335 cleancache_invalidates++;
339 EXPORT_SYMBOL(__cleancache_invalidate_page);
342 * Invalidate all data from cleancache associated with the poolid and the
343 * mappings's inode so that all subsequent gets to this poolid/inode
346 * The function has two checks before any action is taken - whether
347 * a backend is registered and whether the sb->cleancache_poolid
350 void __cleancache_invalidate_inode(struct address_space *mapping)
353 int fake_pool_id = mapping->host->i_sb->cleancache_poolid;
354 struct cleancache_filekey key = { .u.key = { 0 } };
359 if (fake_pool_id < 0)
362 pool_id = get_poolid_from_fake(fake_pool_id);
364 if (pool_id >= 0 && cleancache_get_key(mapping->host, &key) >= 0)
365 cleancache_ops->invalidate_inode(pool_id, key);
367 EXPORT_SYMBOL(__cleancache_invalidate_inode);
370 * Called by any cleancache-enabled filesystem at time of unmount;
371 * note that pool_id is surrendered and may be returned by a subsequent
372 * cleancache_init_fs or cleancache_init_shared_fs.
374 void __cleancache_invalidate_fs(struct super_block *sb)
377 int fake_pool_id = sb->cleancache_poolid;
378 int old_poolid = fake_pool_id;
380 mutex_lock(&poolid_mutex);
381 if (fake_pool_id >= FAKE_SHARED_FS_POOLID_OFFSET) {
382 index = fake_pool_id - FAKE_SHARED_FS_POOLID_OFFSET;
383 old_poolid = shared_fs_poolid_map[index];
384 shared_fs_poolid_map[index] = FS_UNKNOWN;
386 } else if (fake_pool_id >= FAKE_FS_POOLID_OFFSET) {
387 index = fake_pool_id - FAKE_FS_POOLID_OFFSET;
388 old_poolid = fs_poolid_map[index];
389 fs_poolid_map[index] = FS_UNKNOWN;
391 sb->cleancache_poolid = -1;
393 cleancache_ops->invalidate_fs(old_poolid);
394 mutex_unlock(&poolid_mutex);
396 EXPORT_SYMBOL(__cleancache_invalidate_fs);
398 static int __init init_cleancache(void)
402 #ifdef CONFIG_DEBUG_FS
403 struct dentry *root = debugfs_create_dir("cleancache", NULL);
406 debugfs_create_u64("succ_gets", S_IRUGO, root, &cleancache_succ_gets);
407 debugfs_create_u64("failed_gets", S_IRUGO,
408 root, &cleancache_failed_gets);
409 debugfs_create_u64("puts", S_IRUGO, root, &cleancache_puts);
410 debugfs_create_u64("invalidates", S_IRUGO,
411 root, &cleancache_invalidates);
413 for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
414 fs_poolid_map[i] = FS_UNKNOWN;
415 shared_fs_poolid_map[i] = FS_UNKNOWN;
417 cleancache_enabled = 1;
420 module_init(init_cleancache)