1/*
2 * Cleancache frontend
3 *
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.
7 *
8 * Copyright (C) 2009-2010 Oracle Corp. All rights reserved.
9 * Author: Dan Magenheimer
10 *
11 * This work is licensed under the terms of the GNU GPL, version 2.
12 */
13
14#include <linux/module.h>
15#include <linux/fs.h>
16#include <linux/exportfs.h>
17#include <linux/mm.h>
18#include <linux/debugfs.h>
19#include <linux/cleancache.h>
20
21/*
22 * cleancache_ops is set by cleancache_register_ops to contain the pointers
23 * to the cleancache "backend" implementation functions.
24 */
25static struct cleancache_ops *cleancache_ops __read_mostly;
26
27/*
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.
31 */
32static u64 cleancache_succ_gets;
33static u64 cleancache_failed_gets;
34static u64 cleancache_puts;
35static u64 cleancache_invalidates;
36
37static void cleancache_register_ops_sb(struct super_block *sb, void *unused)
38{
39	switch (sb->cleancache_poolid) {
40	case CLEANCACHE_NO_BACKEND:
41		__cleancache_init_fs(sb);
42		break;
43	case CLEANCACHE_NO_BACKEND_SHARED:
44		__cleancache_init_shared_fs(sb);
45		break;
46	}
47}
48
49/*
50 * Register operations for cleancache. Returns 0 on success.
51 */
52int cleancache_register_ops(struct cleancache_ops *ops)
53{
54	if (cmpxchg(&cleancache_ops, NULL, ops))
55		return -EBUSY;
56
57	/*
58	 * A cleancache backend can be built as a module and hence loaded after
59	 * a cleancache enabled filesystem has called cleancache_init_fs. To
60	 * handle such a scenario, here we call ->init_fs or ->init_shared_fs
61	 * for each active super block. To differentiate between local and
62	 * shared filesystems, we temporarily initialize sb->cleancache_poolid
63	 * to CLEANCACHE_NO_BACKEND or CLEANCACHE_NO_BACKEND_SHARED
64	 * respectively in case there is no backend registered at the time
65	 * cleancache_init_fs or cleancache_init_shared_fs is called.
66	 *
67	 * Since filesystems can be mounted concurrently with cleancache
68	 * backend registration, we have to be careful to guarantee that all
69	 * cleancache enabled filesystems that has been mounted by the time
70	 * cleancache_register_ops is called has got and all mounted later will
71	 * get cleancache_poolid. This is assured by the following statements
72	 * tied together:
73	 *
74	 * a) iterate_supers skips only those super blocks that has started
75	 *    ->kill_sb
76	 *
77	 * b) if iterate_supers encounters a super block that has not finished
78	 *    ->mount yet, it waits until it is finished
79	 *
80	 * c) cleancache_init_fs is called from ->mount and
81	 *    cleancache_invalidate_fs is called from ->kill_sb
82	 *
83	 * d) we call iterate_supers after cleancache_ops has been set
84	 *
85	 * From a) it follows that if iterate_supers skips a super block, then
86	 * either the super block is already dead, in which case we do not need
87	 * to bother initializing cleancache for it, or it was mounted after we
88	 * initiated iterate_supers. In the latter case, it must have seen
89	 * cleancache_ops set according to d) and initialized cleancache from
90	 * ->mount by itself according to c). This proves that we call
91	 * ->init_fs at least once for each active super block.
92	 *
93	 * From b) and c) it follows that if iterate_supers encounters a super
94	 * block that has already started ->init_fs, it will wait until ->mount
95	 * and hence ->init_fs has finished, then check cleancache_poolid, see
96	 * that it has already been set and therefore do nothing. This proves
97	 * that we call ->init_fs no more than once for each super block.
98	 *
99	 * Combined together, the last two paragraphs prove the function
100	 * correctness.
101	 *
102	 * Note that various cleancache callbacks may proceed before this
103	 * function is called or even concurrently with it, but since
104	 * CLEANCACHE_NO_BACKEND is negative, they will all result in a noop
105	 * until the corresponding ->init_fs has been actually called and
106	 * cleancache_ops has been set.
107	 */
108	iterate_supers(cleancache_register_ops_sb, NULL);
109	return 0;
110}
111EXPORT_SYMBOL(cleancache_register_ops);
112
113/* Called by a cleancache-enabled filesystem at time of mount */
114void __cleancache_init_fs(struct super_block *sb)
115{
116	int pool_id = CLEANCACHE_NO_BACKEND;
117
118	if (cleancache_ops) {
119		pool_id = cleancache_ops->init_fs(PAGE_SIZE);
120		if (pool_id < 0)
121			pool_id = CLEANCACHE_NO_POOL;
122	}
123	sb->cleancache_poolid = pool_id;
124}
125EXPORT_SYMBOL(__cleancache_init_fs);
126
127/* Called by a cleancache-enabled clustered filesystem at time of mount */
128void __cleancache_init_shared_fs(struct super_block *sb)
129{
130	int pool_id = CLEANCACHE_NO_BACKEND_SHARED;
131
132	if (cleancache_ops) {
133		pool_id = cleancache_ops->init_shared_fs(sb->s_uuid, PAGE_SIZE);
134		if (pool_id < 0)
135			pool_id = CLEANCACHE_NO_POOL;
136	}
137	sb->cleancache_poolid = pool_id;
138}
139EXPORT_SYMBOL(__cleancache_init_shared_fs);
140
141/*
142 * If the filesystem uses exportable filehandles, use the filehandle as
143 * the key, else use the inode number.
144 */
145static int cleancache_get_key(struct inode *inode,
146			      struct cleancache_filekey *key)
147{
148	int (*fhfn)(struct inode *, __u32 *fh, int *, struct inode *);
149	int len = 0, maxlen = CLEANCACHE_KEY_MAX;
150	struct super_block *sb = inode->i_sb;
151
152	key->u.ino = inode->i_ino;
153	if (sb->s_export_op != NULL) {
154		fhfn = sb->s_export_op->encode_fh;
155		if  (fhfn) {
156			len = (*fhfn)(inode, &key->u.fh[0], &maxlen, NULL);
157			if (len <= FILEID_ROOT || len == FILEID_INVALID)
158				return -1;
159			if (maxlen > CLEANCACHE_KEY_MAX)
160				return -1;
161		}
162	}
163	return 0;
164}
165
166/*
167 * "Get" data from cleancache associated with the poolid/inode/index
168 * that were specified when the data was put to cleanache and, if
169 * successful, use it to fill the specified page with data and return 0.
170 * The pageframe is unchanged and returns -1 if the get fails.
171 * Page must be locked by caller.
172 *
173 * The function has two checks before any action is taken - whether
174 * a backend is registered and whether the sb->cleancache_poolid
175 * is correct.
176 */
177int __cleancache_get_page(struct page *page)
178{
179	int ret = -1;
180	int pool_id;
181	struct cleancache_filekey key = { .u.key = { 0 } };
182
183	if (!cleancache_ops) {
184		cleancache_failed_gets++;
185		goto out;
186	}
187
188	VM_BUG_ON_PAGE(!PageLocked(page), page);
189	pool_id = page->mapping->host->i_sb->cleancache_poolid;
190	if (pool_id < 0)
191		goto out;
192
193	if (cleancache_get_key(page->mapping->host, &key) < 0)
194		goto out;
195
196	ret = cleancache_ops->get_page(pool_id, key, page->index, page);
197	if (ret == 0)
198		cleancache_succ_gets++;
199	else
200		cleancache_failed_gets++;
201out:
202	return ret;
203}
204EXPORT_SYMBOL(__cleancache_get_page);
205
206/*
207 * "Put" data from a page to cleancache and associate it with the
208 * (previously-obtained per-filesystem) poolid and the page's,
209 * inode and page index.  Page must be locked.  Note that a put_page
210 * always "succeeds", though a subsequent get_page may succeed or fail.
211 *
212 * The function has two checks before any action is taken - whether
213 * a backend is registered and whether the sb->cleancache_poolid
214 * is correct.
215 */
216void __cleancache_put_page(struct page *page)
217{
218	int pool_id;
219	struct cleancache_filekey key = { .u.key = { 0 } };
220
221	if (!cleancache_ops) {
222		cleancache_puts++;
223		return;
224	}
225
226	VM_BUG_ON_PAGE(!PageLocked(page), page);
227	pool_id = page->mapping->host->i_sb->cleancache_poolid;
228	if (pool_id >= 0 &&
229		cleancache_get_key(page->mapping->host, &key) >= 0) {
230		cleancache_ops->put_page(pool_id, key, page->index, page);
231		cleancache_puts++;
232	}
233}
234EXPORT_SYMBOL(__cleancache_put_page);
235
236/*
237 * Invalidate any data from cleancache associated with the poolid and the
238 * page's inode and page index so that a subsequent "get" will fail.
239 *
240 * The function has two checks before any action is taken - whether
241 * a backend is registered and whether the sb->cleancache_poolid
242 * is correct.
243 */
244void __cleancache_invalidate_page(struct address_space *mapping,
245					struct page *page)
246{
247	/* careful... page->mapping is NULL sometimes when this is called */
248	int pool_id = mapping->host->i_sb->cleancache_poolid;
249	struct cleancache_filekey key = { .u.key = { 0 } };
250
251	if (!cleancache_ops)
252		return;
253
254	if (pool_id >= 0) {
255		VM_BUG_ON_PAGE(!PageLocked(page), page);
256		if (cleancache_get_key(mapping->host, &key) >= 0) {
257			cleancache_ops->invalidate_page(pool_id,
258					key, page->index);
259			cleancache_invalidates++;
260		}
261	}
262}
263EXPORT_SYMBOL(__cleancache_invalidate_page);
264
265/*
266 * Invalidate all data from cleancache associated with the poolid and the
267 * mappings's inode so that all subsequent gets to this poolid/inode
268 * will fail.
269 *
270 * The function has two checks before any action is taken - whether
271 * a backend is registered and whether the sb->cleancache_poolid
272 * is correct.
273 */
274void __cleancache_invalidate_inode(struct address_space *mapping)
275{
276	int pool_id = mapping->host->i_sb->cleancache_poolid;
277	struct cleancache_filekey key = { .u.key = { 0 } };
278
279	if (!cleancache_ops)
280		return;
281
282	if (pool_id >= 0 && cleancache_get_key(mapping->host, &key) >= 0)
283		cleancache_ops->invalidate_inode(pool_id, key);
284}
285EXPORT_SYMBOL(__cleancache_invalidate_inode);
286
287/*
288 * Called by any cleancache-enabled filesystem at time of unmount;
289 * note that pool_id is surrendered and may be returned by a subsequent
290 * cleancache_init_fs or cleancache_init_shared_fs.
291 */
292void __cleancache_invalidate_fs(struct super_block *sb)
293{
294	int pool_id;
295
296	pool_id = sb->cleancache_poolid;
297	sb->cleancache_poolid = CLEANCACHE_NO_POOL;
298
299	if (cleancache_ops && pool_id >= 0)
300		cleancache_ops->invalidate_fs(pool_id);
301}
302EXPORT_SYMBOL(__cleancache_invalidate_fs);
303
304static int __init init_cleancache(void)
305{
306#ifdef CONFIG_DEBUG_FS
307	struct dentry *root = debugfs_create_dir("cleancache", NULL);
308	if (root == NULL)
309		return -ENXIO;
310	debugfs_create_u64("succ_gets", S_IRUGO, root, &cleancache_succ_gets);
311	debugfs_create_u64("failed_gets", S_IRUGO,
312				root, &cleancache_failed_gets);
313	debugfs_create_u64("puts", S_IRUGO, root, &cleancache_puts);
314	debugfs_create_u64("invalidates", S_IRUGO,
315				root, &cleancache_invalidates);
316#endif
317	return 0;
318}
319module_init(init_cleancache)
320