]>
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 | */ | |
1e6cb723 | 51 | static void fsnotify_unmount_inodes(struct super_block *sb) |
ebb3b47e JK |
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); | |
721fb6fb JK |
99 | /* Wait for outstanding inode references from connectors */ |
100 | wait_var_event(&sb->s_fsnotify_inode_refs, | |
101 | !atomic_long_read(&sb->s_fsnotify_inode_refs)); | |
ebb3b47e JK |
102 | } |
103 | ||
1e6cb723 AG |
104 | void fsnotify_sb_delete(struct super_block *sb) |
105 | { | |
106 | fsnotify_unmount_inodes(sb); | |
107 | fsnotify_clear_marks_by_sb(sb); | |
108 | } | |
109 | ||
4d8e7055 AG |
110 | /* |
111 | * fsnotify_nameremove - a filename was removed from a directory | |
112 | * | |
113 | * This is mostly called under parent vfs inode lock so name and | |
114 | * dentry->d_parent should be stable. However there are some corner cases where | |
115 | * inode lock is not held. So to be on the safe side and be reselient to future | |
116 | * callers and out of tree users of d_delete(), we do not assume that d_parent | |
117 | * and d_name are stable and we use dget_parent() and | |
118 | * take_dentry_name_snapshot() to grab stable references. | |
119 | */ | |
120 | void fsnotify_nameremove(struct dentry *dentry, int isdir) | |
121 | { | |
122 | struct dentry *parent; | |
123 | struct name_snapshot name; | |
124 | __u32 mask = FS_DELETE; | |
125 | ||
126 | /* d_delete() of pseudo inode? (e.g. __ns_get_path() playing tricks) */ | |
127 | if (IS_ROOT(dentry)) | |
128 | return; | |
129 | ||
130 | if (isdir) | |
131 | mask |= FS_ISDIR; | |
132 | ||
133 | parent = dget_parent(dentry); | |
134 | /* Avoid unneeded take_dentry_name_snapshot() */ | |
135 | if (!(d_inode(parent)->i_fsnotify_mask & FS_DELETE) && | |
136 | !(dentry->d_sb->s_fsnotify_mask & FS_DELETE)) | |
137 | goto out_dput; | |
138 | ||
139 | take_dentry_name_snapshot(&name, dentry); | |
140 | ||
141 | fsnotify(d_inode(parent), mask, d_inode(dentry), FSNOTIFY_EVENT_INODE, | |
d4c60811 | 142 | &name.name, 0); |
4d8e7055 AG |
143 | |
144 | release_dentry_name_snapshot(&name); | |
145 | ||
146 | out_dput: | |
147 | dput(parent); | |
148 | } | |
149 | EXPORT_SYMBOL(fsnotify_nameremove); | |
150 | ||
c28f7e56 EP |
151 | /* |
152 | * Given an inode, first check if we care what happens to our children. Inotify | |
153 | * and dnotify both tell their parents about events. If we care about any event | |
154 | * on a child we run all of our children and set a dentry flag saying that the | |
155 | * parent cares. Thus when an event happens on a child it can quickly tell if | |
156 | * if there is a need to find a parent and send the event to the parent. | |
157 | */ | |
158 | void __fsnotify_update_child_dentry_flags(struct inode *inode) | |
159 | { | |
160 | struct dentry *alias; | |
161 | int watched; | |
162 | ||
163 | if (!S_ISDIR(inode->i_mode)) | |
164 | return; | |
165 | ||
166 | /* determine if the children should tell inode about their events */ | |
167 | watched = fsnotify_inode_watches_children(inode); | |
168 | ||
873feea0 | 169 | spin_lock(&inode->i_lock); |
c28f7e56 EP |
170 | /* run all of the dentries associated with this inode. Since this is a |
171 | * directory, there damn well better only be one item on this list */ | |
946e51f2 | 172 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
c28f7e56 EP |
173 | struct dentry *child; |
174 | ||
175 | /* run all of the children of the original inode and fix their | |
176 | * d_flags to indicate parental interest (their parent is the | |
177 | * original inode) */ | |
2fd6b7f5 | 178 | spin_lock(&alias->d_lock); |
946e51f2 | 179 | list_for_each_entry(child, &alias->d_subdirs, d_child) { |
c28f7e56 EP |
180 | if (!child->d_inode) |
181 | continue; | |
182 | ||
2fd6b7f5 | 183 | spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED); |
c28f7e56 EP |
184 | if (watched) |
185 | child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; | |
186 | else | |
187 | child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; | |
188 | spin_unlock(&child->d_lock); | |
189 | } | |
2fd6b7f5 | 190 | spin_unlock(&alias->d_lock); |
c28f7e56 | 191 | } |
873feea0 | 192 | spin_unlock(&inode->i_lock); |
c28f7e56 EP |
193 | } |
194 | ||
195 | /* Notify this dentry's parent about a child's events. */ | |
12c7f9dc | 196 | int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask) |
c28f7e56 EP |
197 | { |
198 | struct dentry *parent; | |
199 | struct inode *p_inode; | |
52420392 | 200 | int ret = 0; |
c28f7e56 | 201 | |
72acc854 | 202 | if (!dentry) |
2069601b | 203 | dentry = path->dentry; |
28c60e37 | 204 | |
c28f7e56 | 205 | if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED)) |
52420392 | 206 | return 0; |
c28f7e56 | 207 | |
4d4eb366 | 208 | parent = dget_parent(dentry); |
c28f7e56 EP |
209 | p_inode = parent->d_inode; |
210 | ||
b469e7e4 | 211 | if (unlikely(!fsnotify_inode_watches_children(p_inode))) { |
4d4eb366 | 212 | __fsnotify_update_child_dentry_flags(p_inode); |
b469e7e4 | 213 | } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) { |
49d31c2f AV |
214 | struct name_snapshot name; |
215 | ||
c28f7e56 EP |
216 | /* we are notifying a parent so come up with the new mask which |
217 | * specifies these are events which came from a child. */ | |
218 | mask |= FS_EVENT_ON_CHILD; | |
219 | ||
49d31c2f | 220 | take_dentry_name_snapshot(&name, dentry); |
2069601b | 221 | if (path) |
52420392 | 222 | ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH, |
25b229df | 223 | &name.name, 0); |
28c60e37 | 224 | else |
52420392 | 225 | ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE, |
25b229df | 226 | &name.name, 0); |
49d31c2f | 227 | release_dentry_name_snapshot(&name); |
c28f7e56 EP |
228 | } |
229 | ||
4d4eb366 | 230 | dput(parent); |
52420392 EP |
231 | |
232 | return ret; | |
c28f7e56 EP |
233 | } |
234 | EXPORT_SYMBOL_GPL(__fsnotify_parent); | |
235 | ||
fd657170 | 236 | static int send_to_group(struct inode *to_tell, |
e637835e | 237 | __u32 mask, const void *data, |
613a807f | 238 | int data_is, u32 cookie, |
e43e9c33 | 239 | const struct qstr *file_name, |
9385a84d | 240 | struct fsnotify_iter_info *iter_info) |
7131485a | 241 | { |
faa9560a | 242 | struct fsnotify_group *group = NULL; |
007d1e83 | 243 | __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS); |
92183a42 AG |
244 | __u32 marks_mask = 0; |
245 | __u32 marks_ignored_mask = 0; | |
3dca1a74 AG |
246 | struct fsnotify_mark *mark; |
247 | int type; | |
613a807f | 248 | |
5b0457ad | 249 | if (WARN_ON(!iter_info->report_mask)) |
faa9560a | 250 | return 0; |
ce8f76fb EP |
251 | |
252 | /* clear ignored on inode modification */ | |
253 | if (mask & FS_MODIFY) { | |
3dca1a74 AG |
254 | fsnotify_foreach_obj_type(type) { |
255 | if (!fsnotify_iter_should_report_type(iter_info, type)) | |
256 | continue; | |
257 | mark = iter_info->marks[type]; | |
258 | if (mark && | |
259 | !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) | |
260 | mark->ignored_mask = 0; | |
261 | } | |
ce8f76fb | 262 | } |
613a807f | 263 | |
3dca1a74 AG |
264 | fsnotify_foreach_obj_type(type) { |
265 | if (!fsnotify_iter_should_report_type(iter_info, type)) | |
266 | continue; | |
267 | mark = iter_info->marks[type]; | |
268 | /* does the object mark tell us to do something? */ | |
269 | if (mark) { | |
270 | group = mark->group; | |
271 | marks_mask |= mark->mask; | |
272 | marks_ignored_mask |= mark->ignored_mask; | |
273 | } | |
ce8f76fb EP |
274 | } |
275 | ||
3dca1a74 | 276 | pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x" |
7053aee2 | 277 | " data=%p data_is=%d cookie=%d\n", |
3dca1a74 AG |
278 | __func__, group, to_tell, mask, marks_mask, marks_ignored_mask, |
279 | data, data_is, cookie); | |
faa9560a | 280 | |
92183a42 | 281 | if (!(test_mask & marks_mask & ~marks_ignored_mask)) |
613a807f EP |
282 | return 0; |
283 | ||
5b0457ad | 284 | return group->ops->handle_event(group, to_tell, mask, data, data_is, |
9385a84d | 285 | file_name, cookie, iter_info); |
7131485a EP |
286 | } |
287 | ||
3427ce71 MS |
288 | static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp) |
289 | { | |
290 | struct fsnotify_mark_connector *conn; | |
291 | struct hlist_node *node = NULL; | |
292 | ||
293 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
294 | if (conn) | |
295 | node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu); | |
296 | ||
297 | return hlist_entry_safe(node, struct fsnotify_mark, obj_list); | |
298 | } | |
299 | ||
300 | static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark) | |
301 | { | |
302 | struct hlist_node *node = NULL; | |
303 | ||
304 | if (mark) | |
305 | node = srcu_dereference(mark->obj_list.next, | |
306 | &fsnotify_mark_srcu); | |
307 | ||
308 | return hlist_entry_safe(node, struct fsnotify_mark, obj_list); | |
309 | } | |
310 | ||
d9a6f30b AG |
311 | /* |
312 | * iter_info is a multi head priority queue of marks. | |
313 | * Pick a subset of marks from queue heads, all with the | |
314 | * same group and set the report_mask for selected subset. | |
315 | * Returns the report_mask of the selected subset. | |
316 | */ | |
317 | static unsigned int fsnotify_iter_select_report_types( | |
318 | struct fsnotify_iter_info *iter_info) | |
319 | { | |
47d9c7cc AG |
320 | struct fsnotify_group *max_prio_group = NULL; |
321 | struct fsnotify_mark *mark; | |
322 | int type; | |
323 | ||
324 | /* Choose max prio group among groups of all queue heads */ | |
325 | fsnotify_foreach_obj_type(type) { | |
326 | mark = iter_info->marks[type]; | |
327 | if (mark && | |
328 | fsnotify_compare_groups(max_prio_group, mark->group) > 0) | |
329 | max_prio_group = mark->group; | |
330 | } | |
d9a6f30b | 331 | |
47d9c7cc | 332 | if (!max_prio_group) |
d9a6f30b AG |
333 | return 0; |
334 | ||
47d9c7cc | 335 | /* Set the report mask for marks from same group as max prio group */ |
d9a6f30b | 336 | iter_info->report_mask = 0; |
47d9c7cc AG |
337 | fsnotify_foreach_obj_type(type) { |
338 | mark = iter_info->marks[type]; | |
339 | if (mark && | |
340 | fsnotify_compare_groups(max_prio_group, mark->group) == 0) | |
341 | fsnotify_iter_set_report_type(iter_info, type); | |
342 | } | |
d9a6f30b AG |
343 | |
344 | return iter_info->report_mask; | |
345 | } | |
346 | ||
347 | /* | |
348 | * Pop from iter_info multi head queue, the marks that were iterated in the | |
349 | * current iteration step. | |
350 | */ | |
351 | static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info) | |
352 | { | |
47d9c7cc | 353 | int type; |
d9a6f30b | 354 | |
47d9c7cc AG |
355 | fsnotify_foreach_obj_type(type) { |
356 | if (fsnotify_iter_should_report_type(iter_info, type)) | |
357 | iter_info->marks[type] = | |
358 | fsnotify_next_mark(iter_info->marks[type]); | |
359 | } | |
d9a6f30b AG |
360 | } |
361 | ||
90586523 EP |
362 | /* |
363 | * This is the main call to fsnotify. The VFS calls into hook specific functions | |
364 | * in linux/fsnotify.h. Those functions then in turn call here. Here will call | |
365 | * out to all of the registered fsnotify_group. Those groups can then use the | |
366 | * notification event in whatever means they feel necessary. | |
367 | */ | |
e637835e | 368 | int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, |
25b229df | 369 | const struct qstr *file_name, u32 cookie) |
90586523 | 370 | { |
3427ce71 | 371 | struct fsnotify_iter_info iter_info = {}; |
45a9fb37 | 372 | struct super_block *sb = to_tell->i_sb; |
60f7ed8c | 373 | struct mount *mnt = NULL; |
45a9fb37 | 374 | __u32 mnt_or_sb_mask = sb->s_fsnotify_mask; |
9385a84d | 375 | int ret = 0; |
007d1e83 | 376 | __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS); |
90586523 | 377 | |
60f7ed8c | 378 | if (data_is == FSNOTIFY_EVENT_PATH) { |
e637835e | 379 | mnt = real_mount(((const struct path *)data)->mnt); |
45a9fb37 | 380 | mnt_or_sb_mask |= mnt->mnt_fsnotify_mask; |
60f7ed8c | 381 | } |
b469e7e4 AG |
382 | /* An event "on child" is not intended for a mount/sb mark */ |
383 | if (mask & FS_EVENT_ON_CHILD) | |
384 | mnt_or_sb_mask = 0; | |
613a807f | 385 | |
7c49b861 DH |
386 | /* |
387 | * Optimization: srcu_read_lock() has a memory barrier which can | |
388 | * be expensive. It protects walking the *_fsnotify_marks lists. | |
389 | * However, if we do not walk the lists, we do not have to do | |
390 | * SRCU because we have no references to any objects and do not | |
391 | * need SRCU to keep them "alive". | |
392 | */ | |
45a9fb37 AG |
393 | if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks && |
394 | (!mnt || !mnt->mnt_fsnotify_marks)) | |
7c49b861 | 395 | return 0; |
613a807f EP |
396 | /* |
397 | * if this is a modify event we may need to clear the ignored masks | |
60f7ed8c | 398 | * otherwise return if neither the inode nor the vfsmount/sb care about |
613a807f EP |
399 | * this type of event. |
400 | */ | |
401 | if (!(mask & FS_MODIFY) && | |
60f7ed8c | 402 | !(test_mask & (to_tell->i_fsnotify_mask | mnt_or_sb_mask))) |
613a807f | 403 | return 0; |
3a9fb89f | 404 | |
9385a84d | 405 | iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
7131485a | 406 | |
9bdda4e9 AG |
407 | iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] = |
408 | fsnotify_first_mark(&to_tell->i_fsnotify_marks); | |
45a9fb37 AG |
409 | iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] = |
410 | fsnotify_first_mark(&sb->s_fsnotify_marks); | |
9bdda4e9 | 411 | if (mnt) { |
47d9c7cc | 412 | iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] = |
3427ce71 | 413 | fsnotify_first_mark(&mnt->mnt_fsnotify_marks); |
90586523 | 414 | } |
75c1be48 | 415 | |
8edc6e16 | 416 | /* |
60f7ed8c AG |
417 | * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark |
418 | * ignore masks are properly reflected for mount/sb mark notifications. | |
8edc6e16 JK |
419 | * That's why this traversal is so complicated... |
420 | */ | |
d9a6f30b | 421 | while (fsnotify_iter_select_report_types(&iter_info)) { |
5b0457ad | 422 | ret = send_to_group(to_tell, mask, data, data_is, cookie, |
e43e9c33 | 423 | file_name, &iter_info); |
613a807f | 424 | |
ff8bcbd0 EP |
425 | if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS)) |
426 | goto out; | |
427 | ||
d9a6f30b | 428 | fsnotify_iter_next(&iter_info); |
7131485a | 429 | } |
ff8bcbd0 EP |
430 | ret = 0; |
431 | out: | |
9385a84d | 432 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx); |
c4ec54b4 | 433 | |
98b5c10d | 434 | return ret; |
90586523 EP |
435 | } |
436 | EXPORT_SYMBOL_GPL(fsnotify); | |
437 | ||
9dd813c1 JK |
438 | extern struct kmem_cache *fsnotify_mark_connector_cachep; |
439 | ||
90586523 EP |
440 | static __init int fsnotify_init(void) |
441 | { | |
75c1be48 EP |
442 | int ret; |
443 | ||
66917a31 | 444 | BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25); |
20dee624 | 445 | |
75c1be48 EP |
446 | ret = init_srcu_struct(&fsnotify_mark_srcu); |
447 | if (ret) | |
448 | panic("initializing fsnotify_mark_srcu"); | |
449 | ||
9dd813c1 JK |
450 | fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector, |
451 | SLAB_PANIC); | |
452 | ||
75c1be48 | 453 | return 0; |
90586523 | 454 | } |
75c1be48 | 455 | core_initcall(fsnotify_init); |