Commit | Line | Data |
---|---|---|
90586523 EP |
1 | /* |
2 | * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com> | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2, or (at your option) | |
7 | * any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; see the file COPYING. If not, write to | |
16 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/dcache.h> | |
20 | #include <linux/fs.h> | |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
90586523 EP |
22 | #include <linux/init.h> |
23 | #include <linux/module.h> | |
7131485a | 24 | #include <linux/mount.h> |
90586523 EP |
25 | #include <linux/srcu.h> |
26 | ||
27 | #include <linux/fsnotify_backend.h> | |
28 | #include "fsnotify.h" | |
29 | ||
3be25f49 EP |
30 | /* |
31 | * Clear all of the marks on an inode when it is being evicted from core | |
32 | */ | |
33 | void __fsnotify_inode_delete(struct inode *inode) | |
34 | { | |
35 | fsnotify_clear_marks_by_inode(inode); | |
36 | } | |
37 | EXPORT_SYMBOL_GPL(__fsnotify_inode_delete); | |
38 | ||
ca9c726e AG |
39 | void __fsnotify_vfsmount_delete(struct vfsmount *mnt) |
40 | { | |
41 | fsnotify_clear_marks_by_mount(mnt); | |
42 | } | |
43 | ||
ebb3b47e JK |
44 | /** |
45 | * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes. | |
46 | * @sb: superblock being unmounted. | |
47 | * | |
48 | * Called during unmount with no locks held, so needs to be safe against | |
49 | * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block. | |
50 | */ | |
51 | void fsnotify_unmount_inodes(struct super_block *sb) | |
52 | { | |
53 | struct inode *inode, *iput_inode = NULL; | |
54 | ||
55 | spin_lock(&sb->s_inode_list_lock); | |
56 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
57 | /* | |
58 | * We cannot __iget() an inode in state I_FREEING, | |
59 | * I_WILL_FREE, or I_NEW which is fine because by that point | |
60 | * the inode cannot have any associated watches. | |
61 | */ | |
62 | spin_lock(&inode->i_lock); | |
63 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) { | |
64 | spin_unlock(&inode->i_lock); | |
65 | continue; | |
66 | } | |
67 | ||
68 | /* | |
69 | * If i_count is zero, the inode cannot have any watches and | |
1751e8a6 | 70 | * doing an __iget/iput with SB_ACTIVE clear would actually |
ebb3b47e JK |
71 | * evict all inodes with zero i_count from icache which is |
72 | * unnecessarily violent and may in fact be illegal to do. | |
73 | */ | |
74 | if (!atomic_read(&inode->i_count)) { | |
75 | spin_unlock(&inode->i_lock); | |
76 | continue; | |
77 | } | |
78 | ||
79 | __iget(inode); | |
80 | spin_unlock(&inode->i_lock); | |
81 | spin_unlock(&sb->s_inode_list_lock); | |
82 | ||
83 | if (iput_inode) | |
84 | iput(iput_inode); | |
85 | ||
86 | /* for each watch, send FS_UNMOUNT and then remove it */ | |
87 | fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0); | |
88 | ||
89 | fsnotify_inode_delete(inode); | |
90 | ||
91 | iput_inode = inode; | |
92 | ||
93 | spin_lock(&sb->s_inode_list_lock); | |
94 | } | |
95 | spin_unlock(&sb->s_inode_list_lock); | |
96 | ||
97 | if (iput_inode) | |
98 | iput(iput_inode); | |
99 | } | |
100 | ||
c28f7e56 EP |
101 | /* |
102 | * Given an inode, first check if we care what happens to our children. Inotify | |
103 | * and dnotify both tell their parents about events. If we care about any event | |
104 | * on a child we run all of our children and set a dentry flag saying that the | |
105 | * parent cares. Thus when an event happens on a child it can quickly tell if | |
106 | * if there is a need to find a parent and send the event to the parent. | |
107 | */ | |
108 | void __fsnotify_update_child_dentry_flags(struct inode *inode) | |
109 | { | |
110 | struct dentry *alias; | |
111 | int watched; | |
112 | ||
113 | if (!S_ISDIR(inode->i_mode)) | |
114 | return; | |
115 | ||
116 | /* determine if the children should tell inode about their events */ | |
117 | watched = fsnotify_inode_watches_children(inode); | |
118 | ||
873feea0 | 119 | spin_lock(&inode->i_lock); |
c28f7e56 EP |
120 | /* run all of the dentries associated with this inode. Since this is a |
121 | * directory, there damn well better only be one item on this list */ | |
946e51f2 | 122 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
c28f7e56 EP |
123 | struct dentry *child; |
124 | ||
125 | /* run all of the children of the original inode and fix their | |
126 | * d_flags to indicate parental interest (their parent is the | |
127 | * original inode) */ | |
2fd6b7f5 | 128 | spin_lock(&alias->d_lock); |
946e51f2 | 129 | list_for_each_entry(child, &alias->d_subdirs, d_child) { |
c28f7e56 EP |
130 | if (!child->d_inode) |
131 | continue; | |
132 | ||
2fd6b7f5 | 133 | spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED); |
c28f7e56 EP |
134 | if (watched) |
135 | child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; | |
136 | else | |
137 | child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; | |
138 | spin_unlock(&child->d_lock); | |
139 | } | |
2fd6b7f5 | 140 | spin_unlock(&alias->d_lock); |
c28f7e56 | 141 | } |
873feea0 | 142 | spin_unlock(&inode->i_lock); |
c28f7e56 EP |
143 | } |
144 | ||
145 | /* Notify this dentry's parent about a child's events. */ | |
12c7f9dc | 146 | int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask) |
c28f7e56 EP |
147 | { |
148 | struct dentry *parent; | |
149 | struct inode *p_inode; | |
52420392 | 150 | int ret = 0; |
c28f7e56 | 151 | |
72acc854 | 152 | if (!dentry) |
2069601b | 153 | dentry = path->dentry; |
28c60e37 | 154 | |
c28f7e56 | 155 | if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED)) |
52420392 | 156 | return 0; |
c28f7e56 | 157 | |
4d4eb366 | 158 | parent = dget_parent(dentry); |
c28f7e56 EP |
159 | p_inode = parent->d_inode; |
160 | ||
4d4eb366 CH |
161 | if (unlikely(!fsnotify_inode_watches_children(p_inode))) |
162 | __fsnotify_update_child_dentry_flags(p_inode); | |
163 | else if (p_inode->i_fsnotify_mask & mask) { | |
49d31c2f AV |
164 | struct name_snapshot name; |
165 | ||
c28f7e56 EP |
166 | /* we are notifying a parent so come up with the new mask which |
167 | * specifies these are events which came from a child. */ | |
168 | mask |= FS_EVENT_ON_CHILD; | |
169 | ||
49d31c2f | 170 | take_dentry_name_snapshot(&name, dentry); |
2069601b | 171 | if (path) |
52420392 | 172 | ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH, |
49d31c2f | 173 | name.name, 0); |
28c60e37 | 174 | else |
52420392 | 175 | ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE, |
49d31c2f AV |
176 | name.name, 0); |
177 | release_dentry_name_snapshot(&name); | |
c28f7e56 EP |
178 | } |
179 | ||
4d4eb366 | 180 | dput(parent); |
52420392 EP |
181 | |
182 | return ret; | |
c28f7e56 EP |
183 | } |
184 | EXPORT_SYMBOL_GPL(__fsnotify_parent); | |
185 | ||
fd657170 | 186 | static int send_to_group(struct inode *to_tell, |
ce8f76fb EP |
187 | struct fsnotify_mark *inode_mark, |
188 | struct fsnotify_mark *vfsmount_mark, | |
e637835e | 189 | __u32 mask, const void *data, |
613a807f | 190 | int data_is, u32 cookie, |
9385a84d JK |
191 | const unsigned char *file_name, |
192 | struct fsnotify_iter_info *iter_info) | |
7131485a | 193 | { |
faa9560a | 194 | struct fsnotify_group *group = NULL; |
92183a42 AG |
195 | __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD); |
196 | __u32 marks_mask = 0; | |
197 | __u32 marks_ignored_mask = 0; | |
613a807f | 198 | |
faa9560a EP |
199 | if (unlikely(!inode_mark && !vfsmount_mark)) { |
200 | BUG(); | |
201 | return 0; | |
202 | } | |
ce8f76fb EP |
203 | |
204 | /* clear ignored on inode modification */ | |
205 | if (mask & FS_MODIFY) { | |
206 | if (inode_mark && | |
207 | !(inode_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) | |
208 | inode_mark->ignored_mask = 0; | |
209 | if (vfsmount_mark && | |
210 | !(vfsmount_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) | |
211 | vfsmount_mark->ignored_mask = 0; | |
212 | } | |
5ba08e2e | 213 | |
ce8f76fb EP |
214 | /* does the inode mark tell us to do something? */ |
215 | if (inode_mark) { | |
faa9560a | 216 | group = inode_mark->group; |
92183a42 AG |
217 | marks_mask |= inode_mark->mask; |
218 | marks_ignored_mask |= inode_mark->ignored_mask; | |
ce8f76fb | 219 | } |
613a807f | 220 | |
ce8f76fb EP |
221 | /* does the vfsmount_mark tell us to do something? */ |
222 | if (vfsmount_mark) { | |
faa9560a | 223 | group = vfsmount_mark->group; |
92183a42 AG |
224 | marks_mask |= vfsmount_mark->mask; |
225 | marks_ignored_mask |= vfsmount_mark->ignored_mask; | |
ce8f76fb EP |
226 | } |
227 | ||
fd657170 | 228 | pr_debug("%s: group=%p to_tell=%p mask=%x inode_mark=%p" |
92183a42 | 229 | " vfsmount_mark=%p marks_mask=%x marks_ignored_mask=%x" |
7053aee2 | 230 | " data=%p data_is=%d cookie=%d\n", |
92183a42 AG |
231 | __func__, group, to_tell, mask, inode_mark, vfsmount_mark, |
232 | marks_mask, marks_ignored_mask, data, | |
7053aee2 | 233 | data_is, cookie); |
faa9560a | 234 | |
92183a42 | 235 | if (!(test_mask & marks_mask & ~marks_ignored_mask)) |
613a807f EP |
236 | return 0; |
237 | ||
7053aee2 JK |
238 | return group->ops->handle_event(group, to_tell, inode_mark, |
239 | vfsmount_mark, mask, data, data_is, | |
9385a84d | 240 | file_name, cookie, iter_info); |
7131485a EP |
241 | } |
242 | ||
3427ce71 MS |
243 | static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp) |
244 | { | |
245 | struct fsnotify_mark_connector *conn; | |
246 | struct hlist_node *node = NULL; | |
247 | ||
248 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
249 | if (conn) | |
250 | node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu); | |
251 | ||
252 | return hlist_entry_safe(node, struct fsnotify_mark, obj_list); | |
253 | } | |
254 | ||
255 | static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark) | |
256 | { | |
257 | struct hlist_node *node = NULL; | |
258 | ||
259 | if (mark) | |
260 | node = srcu_dereference(mark->obj_list.next, | |
261 | &fsnotify_mark_srcu); | |
262 | ||
263 | return hlist_entry_safe(node, struct fsnotify_mark, obj_list); | |
264 | } | |
265 | ||
90586523 EP |
266 | /* |
267 | * This is the main call to fsnotify. The VFS calls into hook specific functions | |
268 | * in linux/fsnotify.h. Those functions then in turn call here. Here will call | |
269 | * out to all of the registered fsnotify_group. Those groups can then use the | |
270 | * notification event in whatever means they feel necessary. | |
271 | */ | |
e637835e | 272 | int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, |
c4ec54b4 | 273 | const unsigned char *file_name, u32 cookie) |
90586523 | 274 | { |
3427ce71 | 275 | struct fsnotify_iter_info iter_info = {}; |
c63181e6 | 276 | struct mount *mnt; |
9385a84d | 277 | int ret = 0; |
e42e2773 EP |
278 | /* global tests shouldn't care about events on child only the specific event */ |
279 | __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD); | |
90586523 | 280 | |
2069601b | 281 | if (data_is == FSNOTIFY_EVENT_PATH) |
e637835e | 282 | mnt = real_mount(((const struct path *)data)->mnt); |
613a807f EP |
283 | else |
284 | mnt = NULL; | |
285 | ||
7c49b861 DH |
286 | /* |
287 | * Optimization: srcu_read_lock() has a memory barrier which can | |
288 | * be expensive. It protects walking the *_fsnotify_marks lists. | |
289 | * However, if we do not walk the lists, we do not have to do | |
290 | * SRCU because we have no references to any objects and do not | |
291 | * need SRCU to keep them "alive". | |
292 | */ | |
9dd813c1 JK |
293 | if (!to_tell->i_fsnotify_marks && |
294 | (!mnt || !mnt->mnt_fsnotify_marks)) | |
7c49b861 | 295 | return 0; |
613a807f EP |
296 | /* |
297 | * if this is a modify event we may need to clear the ignored masks | |
298 | * otherwise return if neither the inode nor the vfsmount care about | |
299 | * this type of event. | |
300 | */ | |
301 | if (!(mask & FS_MODIFY) && | |
302 | !(test_mask & to_tell->i_fsnotify_mask) && | |
303 | !(mnt && test_mask & mnt->mnt_fsnotify_mask)) | |
304 | return 0; | |
3a9fb89f | 305 | |
9385a84d | 306 | iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
7131485a | 307 | |
613a807f | 308 | if ((mask & FS_MODIFY) || |
9dd813c1 | 309 | (test_mask & to_tell->i_fsnotify_mask)) { |
3427ce71 MS |
310 | iter_info.inode_mark = |
311 | fsnotify_first_mark(&to_tell->i_fsnotify_marks); | |
9dd813c1 | 312 | } |
613a807f | 313 | |
84e1ab4d EP |
314 | if (mnt && ((mask & FS_MODIFY) || |
315 | (test_mask & mnt->mnt_fsnotify_mask))) { | |
3427ce71 MS |
316 | iter_info.inode_mark = |
317 | fsnotify_first_mark(&to_tell->i_fsnotify_marks); | |
318 | iter_info.vfsmount_mark = | |
319 | fsnotify_first_mark(&mnt->mnt_fsnotify_marks); | |
90586523 | 320 | } |
75c1be48 | 321 | |
8edc6e16 JK |
322 | /* |
323 | * We need to merge inode & vfsmount mark lists so that inode mark | |
324 | * ignore masks are properly reflected for mount mark notifications. | |
325 | * That's why this traversal is so complicated... | |
326 | */ | |
3427ce71 MS |
327 | while (iter_info.inode_mark || iter_info.vfsmount_mark) { |
328 | struct fsnotify_mark *inode_mark = iter_info.inode_mark; | |
329 | struct fsnotify_mark *vfsmount_mark = iter_info.vfsmount_mark; | |
330 | ||
331 | if (inode_mark && vfsmount_mark) { | |
332 | int cmp = fsnotify_compare_groups(inode_mark->group, | |
333 | vfsmount_mark->group); | |
334 | if (cmp > 0) | |
8edc6e16 | 335 | inode_mark = NULL; |
3427ce71 | 336 | else if (cmp < 0) |
8edc6e16 | 337 | vfsmount_mark = NULL; |
7131485a | 338 | } |
9385a84d | 339 | |
8edc6e16 | 340 | ret = send_to_group(to_tell, inode_mark, vfsmount_mark, mask, |
9385a84d JK |
341 | data, data_is, cookie, file_name, |
342 | &iter_info); | |
613a807f | 343 | |
ff8bcbd0 EP |
344 | if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS)) |
345 | goto out; | |
346 | ||
3427ce71 MS |
347 | if (inode_mark) |
348 | iter_info.inode_mark = | |
349 | fsnotify_next_mark(iter_info.inode_mark); | |
350 | if (vfsmount_mark) | |
351 | iter_info.vfsmount_mark = | |
352 | fsnotify_next_mark(iter_info.vfsmount_mark); | |
7131485a | 353 | } |
ff8bcbd0 EP |
354 | ret = 0; |
355 | out: | |
9385a84d | 356 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx); |
c4ec54b4 | 357 | |
98b5c10d | 358 | return ret; |
90586523 EP |
359 | } |
360 | EXPORT_SYMBOL_GPL(fsnotify); | |
361 | ||
9dd813c1 JK |
362 | extern struct kmem_cache *fsnotify_mark_connector_cachep; |
363 | ||
90586523 EP |
364 | static __init int fsnotify_init(void) |
365 | { | |
75c1be48 EP |
366 | int ret; |
367 | ||
20dee624 EP |
368 | BUG_ON(hweight32(ALL_FSNOTIFY_EVENTS) != 23); |
369 | ||
75c1be48 EP |
370 | ret = init_srcu_struct(&fsnotify_mark_srcu); |
371 | if (ret) | |
372 | panic("initializing fsnotify_mark_srcu"); | |
373 | ||
9dd813c1 JK |
374 | fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector, |
375 | SLAB_PANIC); | |
376 | ||
75c1be48 | 377 | return 0; |
90586523 | 378 | } |
75c1be48 | 379 | core_initcall(fsnotify_init); |