1214152Sed// SPDX-License-Identifier: GPL-2.0-or-later
2214152Sed/* Manage high-level VFS aspects of a cache.
3214152Sed *
4214152Sed * Copyright (C) 2007, 2021 Red Hat, Inc. All Rights Reserved.
5222656Sed * Written by David Howells (dhowells@redhat.com)
6222656Sed */
7214152Sed
8214152Sed#include <linux/slab.h>
9214152Sed#include <linux/statfs.h>
10214152Sed#include <linux/namei.h>
11214152Sed#include "internal.h"
12214152Sed
13214152Sed/*
14214152Sed * Bring a cache online.
15263560Sdim */
16263560Sdimint cachefiles_add_cache(struct cachefiles_cache *cache)
17214152Sed{
18214152Sed	struct fscache_cache *cache_cookie;
19214152Sed	struct path path;
20214152Sed	struct kstatfs stats;
21214152Sed	struct dentry *graveyard, *cachedir, *root;
22214152Sed	const struct cred *saved_cred;
23214152Sed	int ret;
24214152Sed
25214152Sed	_enter("");
26214152Sed
27214152Sed	cache_cookie = fscache_acquire_cache(cache->tag);
28214152Sed	if (IS_ERR(cache_cookie))
29214152Sed		return PTR_ERR(cache_cookie);
30214152Sed
31214152Sed	/* we want to work under the module's security ID */
32214152Sed	ret = cachefiles_get_security_ID(cache);
33214152Sed	if (ret < 0)
34214152Sed		goto error_getsec;
35214152Sed
36214152Sed	cachefiles_begin_secure(cache, &saved_cred);
37214152Sed
38214152Sed	/* look up the directory at the root of the cache */
39214152Sed	ret = kern_path(cache->rootdirname, LOOKUP_DIRECTORY, &path);
40214152Sed	if (ret < 0)
41214152Sed		goto error_open_root;
42214152Sed
43214152Sed	cache->mnt = path.mnt;
44214152Sed	root = path.dentry;
45214152Sed
46	ret = -EINVAL;
47	if (is_idmapped_mnt(path.mnt)) {
48		pr_warn("File cache on idmapped mounts not supported");
49		goto error_unsupported;
50	}
51
52	/* Check features of the backing filesystem:
53	 * - Directories must support looking up and directory creation
54	 * - We create tmpfiles to handle invalidation
55	 * - We use xattrs to store metadata
56	 * - We need to be able to query the amount of space available
57	 * - We want to be able to sync the filesystem when stopping the cache
58	 * - We use DIO to/from pages, so the blocksize mustn't be too big.
59	 */
60	ret = -EOPNOTSUPP;
61	if (d_is_negative(root) ||
62	    !d_backing_inode(root)->i_op->lookup ||
63	    !d_backing_inode(root)->i_op->mkdir ||
64	    !d_backing_inode(root)->i_op->tmpfile ||
65	    !(d_backing_inode(root)->i_opflags & IOP_XATTR) ||
66	    !root->d_sb->s_op->statfs ||
67	    !root->d_sb->s_op->sync_fs ||
68	    root->d_sb->s_blocksize > PAGE_SIZE)
69		goto error_unsupported;
70
71	ret = -EROFS;
72	if (sb_rdonly(root->d_sb))
73		goto error_unsupported;
74
75	/* determine the security of the on-disk cache as this governs
76	 * security ID of files we create */
77	ret = cachefiles_determine_cache_security(cache, root, &saved_cred);
78	if (ret < 0)
79		goto error_unsupported;
80
81	/* get the cache size and blocksize */
82	ret = vfs_statfs(&path, &stats);
83	if (ret < 0)
84		goto error_unsupported;
85
86	ret = -ERANGE;
87	if (stats.f_bsize <= 0)
88		goto error_unsupported;
89
90	ret = -EOPNOTSUPP;
91	if (stats.f_bsize > PAGE_SIZE)
92		goto error_unsupported;
93
94	cache->bsize = stats.f_bsize;
95	cache->bshift = ilog2(stats.f_bsize);
96
97	_debug("blksize %u (shift %u)",
98	       cache->bsize, cache->bshift);
99
100	_debug("size %llu, avail %llu",
101	       (unsigned long long) stats.f_blocks,
102	       (unsigned long long) stats.f_bavail);
103
104	/* set up caching limits */
105	do_div(stats.f_files, 100);
106	cache->fstop = stats.f_files * cache->fstop_percent;
107	cache->fcull = stats.f_files * cache->fcull_percent;
108	cache->frun  = stats.f_files * cache->frun_percent;
109
110	_debug("limits {%llu,%llu,%llu} files",
111	       (unsigned long long) cache->frun,
112	       (unsigned long long) cache->fcull,
113	       (unsigned long long) cache->fstop);
114
115	do_div(stats.f_blocks, 100);
116	cache->bstop = stats.f_blocks * cache->bstop_percent;
117	cache->bcull = stats.f_blocks * cache->bcull_percent;
118	cache->brun  = stats.f_blocks * cache->brun_percent;
119
120	_debug("limits {%llu,%llu,%llu} blocks",
121	       (unsigned long long) cache->brun,
122	       (unsigned long long) cache->bcull,
123	       (unsigned long long) cache->bstop);
124
125	/* get the cache directory and check its type */
126	cachedir = cachefiles_get_directory(cache, root, "cache", NULL);
127	if (IS_ERR(cachedir)) {
128		ret = PTR_ERR(cachedir);
129		goto error_unsupported;
130	}
131
132	cache->store = cachedir;
133
134	/* get the graveyard directory */
135	graveyard = cachefiles_get_directory(cache, root, "graveyard", NULL);
136	if (IS_ERR(graveyard)) {
137		ret = PTR_ERR(graveyard);
138		goto error_unsupported;
139	}
140
141	cache->graveyard = graveyard;
142	cache->cache = cache_cookie;
143
144	ret = fscache_add_cache(cache_cookie, &cachefiles_cache_ops, cache);
145	if (ret < 0)
146		goto error_add_cache;
147
148	/* done */
149	set_bit(CACHEFILES_READY, &cache->flags);
150	dput(root);
151
152	pr_info("File cache on %s registered\n", cache_cookie->name);
153
154	/* check how much space the cache has */
155	cachefiles_has_space(cache, 0, 0, cachefiles_has_space_check);
156	cachefiles_end_secure(cache, saved_cred);
157	_leave(" = 0 [%px]", cache->cache);
158	return 0;
159
160error_add_cache:
161	cachefiles_put_directory(cache->graveyard);
162	cache->graveyard = NULL;
163error_unsupported:
164	cachefiles_put_directory(cache->store);
165	cache->store = NULL;
166	mntput(cache->mnt);
167	cache->mnt = NULL;
168	dput(root);
169error_open_root:
170	cachefiles_end_secure(cache, saved_cred);
171	put_cred(cache->cache_cred);
172	cache->cache_cred = NULL;
173error_getsec:
174	fscache_relinquish_cache(cache_cookie);
175	cache->cache = NULL;
176	pr_err("Failed to register: %d\n", ret);
177	return ret;
178}
179
180/*
181 * See if we have space for a number of pages and/or a number of files in the
182 * cache
183 */
184int cachefiles_has_space(struct cachefiles_cache *cache,
185			 unsigned fnr, unsigned bnr,
186			 enum cachefiles_has_space_for reason)
187{
188	struct kstatfs stats;
189	u64 b_avail, b_writing;
190	int ret;
191
192	struct path path = {
193		.mnt	= cache->mnt,
194		.dentry	= cache->mnt->mnt_root,
195	};
196
197	//_enter("{%llu,%llu,%llu,%llu,%llu,%llu},%u,%u",
198	//       (unsigned long long) cache->frun,
199	//       (unsigned long long) cache->fcull,
200	//       (unsigned long long) cache->fstop,
201	//       (unsigned long long) cache->brun,
202	//       (unsigned long long) cache->bcull,
203	//       (unsigned long long) cache->bstop,
204	//       fnr, bnr);
205
206	/* find out how many pages of blockdev are available */
207	memset(&stats, 0, sizeof(stats));
208
209	ret = vfs_statfs(&path, &stats);
210	if (ret < 0) {
211		trace_cachefiles_vfs_error(NULL, d_inode(path.dentry), ret,
212					   cachefiles_trace_statfs_error);
213		if (ret == -EIO)
214			cachefiles_io_error(cache, "statfs failed");
215		_leave(" = %d", ret);
216		return ret;
217	}
218
219	b_avail = stats.f_bavail;
220	b_writing = atomic_long_read(&cache->b_writing);
221	if (b_avail > b_writing)
222		b_avail -= b_writing;
223	else
224		b_avail = 0;
225
226	//_debug("avail %llu,%llu",
227	//       (unsigned long long)stats.f_ffree,
228	//       (unsigned long long)b_avail);
229
230	/* see if there is sufficient space */
231	if (stats.f_ffree > fnr)
232		stats.f_ffree -= fnr;
233	else
234		stats.f_ffree = 0;
235
236	if (b_avail > bnr)
237		b_avail -= bnr;
238	else
239		b_avail = 0;
240
241	ret = -ENOBUFS;
242	if (stats.f_ffree < cache->fstop ||
243	    b_avail < cache->bstop)
244		goto stop_and_begin_cull;
245
246	ret = 0;
247	if (stats.f_ffree < cache->fcull ||
248	    b_avail < cache->bcull)
249		goto begin_cull;
250
251	if (test_bit(CACHEFILES_CULLING, &cache->flags) &&
252	    stats.f_ffree >= cache->frun &&
253	    b_avail >= cache->brun &&
254	    test_and_clear_bit(CACHEFILES_CULLING, &cache->flags)
255	    ) {
256		_debug("cease culling");
257		cachefiles_state_changed(cache);
258	}
259
260	//_leave(" = 0");
261	return 0;
262
263stop_and_begin_cull:
264	switch (reason) {
265	case cachefiles_has_space_for_write:
266		fscache_count_no_write_space();
267		break;
268	case cachefiles_has_space_for_create:
269		fscache_count_no_create_space();
270		break;
271	default:
272		break;
273	}
274begin_cull:
275	if (!test_and_set_bit(CACHEFILES_CULLING, &cache->flags)) {
276		_debug("### CULL CACHE ###");
277		cachefiles_state_changed(cache);
278	}
279
280	_leave(" = %d", ret);
281	return ret;
282}
283
284/*
285 * Mark all the objects as being out of service and queue them all for cleanup.
286 */
287static void cachefiles_withdraw_objects(struct cachefiles_cache *cache)
288{
289	struct cachefiles_object *object;
290	unsigned int count = 0;
291
292	_enter("");
293
294	spin_lock(&cache->object_list_lock);
295
296	while (!list_empty(&cache->object_list)) {
297		object = list_first_entry(&cache->object_list,
298					  struct cachefiles_object, cache_link);
299		cachefiles_see_object(object, cachefiles_obj_see_withdrawal);
300		list_del_init(&object->cache_link);
301		fscache_withdraw_cookie(object->cookie);
302		count++;
303		if ((count & 63) == 0) {
304			spin_unlock(&cache->object_list_lock);
305			cond_resched();
306			spin_lock(&cache->object_list_lock);
307		}
308	}
309
310	spin_unlock(&cache->object_list_lock);
311	_leave(" [%u objs]", count);
312}
313
314/*
315 * Withdraw volumes.
316 */
317static void cachefiles_withdraw_volumes(struct cachefiles_cache *cache)
318{
319	_enter("");
320
321	for (;;) {
322		struct cachefiles_volume *volume = NULL;
323
324		spin_lock(&cache->object_list_lock);
325		if (!list_empty(&cache->volumes)) {
326			volume = list_first_entry(&cache->volumes,
327						  struct cachefiles_volume, cache_link);
328			list_del_init(&volume->cache_link);
329		}
330		spin_unlock(&cache->object_list_lock);
331		if (!volume)
332			break;
333
334		cachefiles_withdraw_volume(volume);
335	}
336
337	_leave("");
338}
339
340/*
341 * Sync a cache to backing disk.
342 */
343static void cachefiles_sync_cache(struct cachefiles_cache *cache)
344{
345	const struct cred *saved_cred;
346	int ret;
347
348	_enter("%s", cache->cache->name);
349
350	/* make sure all pages pinned by operations on behalf of the netfs are
351	 * written to disc */
352	cachefiles_begin_secure(cache, &saved_cred);
353	down_read(&cache->mnt->mnt_sb->s_umount);
354	ret = sync_filesystem(cache->mnt->mnt_sb);
355	up_read(&cache->mnt->mnt_sb->s_umount);
356	cachefiles_end_secure(cache, saved_cred);
357
358	if (ret == -EIO)
359		cachefiles_io_error(cache,
360				    "Attempt to sync backing fs superblock returned error %d",
361				    ret);
362}
363
364/*
365 * Withdraw cache objects.
366 */
367void cachefiles_withdraw_cache(struct cachefiles_cache *cache)
368{
369	struct fscache_cache *fscache = cache->cache;
370
371	pr_info("File cache on %s unregistering\n", fscache->name);
372
373	fscache_withdraw_cache(fscache);
374
375	/* we now have to destroy all the active objects pertaining to this
376	 * cache - which we do by passing them off to thread pool to be
377	 * disposed of */
378	cachefiles_withdraw_objects(cache);
379	fscache_wait_for_objects(fscache);
380
381	cachefiles_withdraw_volumes(cache);
382	cachefiles_sync_cache(cache);
383	cache->cache = NULL;
384	fscache_relinquish_cache(fscache);
385}
386