]>
Commit | Line | Data |
---|---|---|
c82ee6d3 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
90586523 EP |
2 | /* |
3 | * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com> | |
90586523 EP |
4 | */ |
5 | ||
6 | #include <linux/dcache.h> | |
7 | #include <linux/fs.h> | |
5a0e3ad6 | 8 | #include <linux/gfp.h> |
90586523 EP |
9 | #include <linux/init.h> |
10 | #include <linux/module.h> | |
7131485a | 11 | #include <linux/mount.h> |
90586523 EP |
12 | #include <linux/srcu.h> |
13 | ||
14 | #include <linux/fsnotify_backend.h> | |
15 | #include "fsnotify.h" | |
16 | ||
3be25f49 EP |
17 | /* |
18 | * Clear all of the marks on an inode when it is being evicted from core | |
19 | */ | |
20 | void __fsnotify_inode_delete(struct inode *inode) | |
21 | { | |
22 | fsnotify_clear_marks_by_inode(inode); | |
23 | } | |
24 | EXPORT_SYMBOL_GPL(__fsnotify_inode_delete); | |
25 | ||
ca9c726e AG |
26 | void __fsnotify_vfsmount_delete(struct vfsmount *mnt) |
27 | { | |
28 | fsnotify_clear_marks_by_mount(mnt); | |
29 | } | |
30 | ||
ebb3b47e JK |
31 | /** |
32 | * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes. | |
33 | * @sb: superblock being unmounted. | |
34 | * | |
35 | * Called during unmount with no locks held, so needs to be safe against | |
36 | * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block. | |
37 | */ | |
1e6cb723 | 38 | static void fsnotify_unmount_inodes(struct super_block *sb) |
ebb3b47e JK |
39 | { |
40 | struct inode *inode, *iput_inode = NULL; | |
41 | ||
42 | spin_lock(&sb->s_inode_list_lock); | |
43 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
44 | /* | |
45 | * We cannot __iget() an inode in state I_FREEING, | |
46 | * I_WILL_FREE, or I_NEW which is fine because by that point | |
47 | * the inode cannot have any associated watches. | |
48 | */ | |
49 | spin_lock(&inode->i_lock); | |
50 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) { | |
51 | spin_unlock(&inode->i_lock); | |
52 | continue; | |
53 | } | |
54 | ||
55 | /* | |
56 | * If i_count is zero, the inode cannot have any watches and | |
1751e8a6 | 57 | * doing an __iget/iput with SB_ACTIVE clear would actually |
ebb3b47e JK |
58 | * evict all inodes with zero i_count from icache which is |
59 | * unnecessarily violent and may in fact be illegal to do. | |
60 | */ | |
61 | if (!atomic_read(&inode->i_count)) { | |
62 | spin_unlock(&inode->i_lock); | |
63 | continue; | |
64 | } | |
65 | ||
66 | __iget(inode); | |
67 | spin_unlock(&inode->i_lock); | |
68 | spin_unlock(&sb->s_inode_list_lock); | |
69 | ||
70 | if (iput_inode) | |
71 | iput(iput_inode); | |
72 | ||
73 | /* for each watch, send FS_UNMOUNT and then remove it */ | |
74 | fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0); | |
75 | ||
76 | fsnotify_inode_delete(inode); | |
77 | ||
78 | iput_inode = inode; | |
79 | ||
80 | spin_lock(&sb->s_inode_list_lock); | |
81 | } | |
82 | spin_unlock(&sb->s_inode_list_lock); | |
83 | ||
84 | if (iput_inode) | |
85 | iput(iput_inode); | |
721fb6fb JK |
86 | /* Wait for outstanding inode references from connectors */ |
87 | wait_var_event(&sb->s_fsnotify_inode_refs, | |
88 | !atomic_long_read(&sb->s_fsnotify_inode_refs)); | |
ebb3b47e JK |
89 | } |
90 | ||
1e6cb723 AG |
91 | void fsnotify_sb_delete(struct super_block *sb) |
92 | { | |
93 | fsnotify_unmount_inodes(sb); | |
94 | fsnotify_clear_marks_by_sb(sb); | |
95 | } | |
96 | ||
c28f7e56 EP |
97 | /* |
98 | * Given an inode, first check if we care what happens to our children. Inotify | |
99 | * and dnotify both tell their parents about events. If we care about any event | |
100 | * on a child we run all of our children and set a dentry flag saying that the | |
101 | * parent cares. Thus when an event happens on a child it can quickly tell if | |
102 | * if there is a need to find a parent and send the event to the parent. | |
103 | */ | |
104 | void __fsnotify_update_child_dentry_flags(struct inode *inode) | |
105 | { | |
106 | struct dentry *alias; | |
107 | int watched; | |
108 | ||
109 | if (!S_ISDIR(inode->i_mode)) | |
110 | return; | |
111 | ||
112 | /* determine if the children should tell inode about their events */ | |
113 | watched = fsnotify_inode_watches_children(inode); | |
114 | ||
873feea0 | 115 | spin_lock(&inode->i_lock); |
c28f7e56 EP |
116 | /* run all of the dentries associated with this inode. Since this is a |
117 | * directory, there damn well better only be one item on this list */ | |
946e51f2 | 118 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
c28f7e56 EP |
119 | struct dentry *child; |
120 | ||
121 | /* run all of the children of the original inode and fix their | |
122 | * d_flags to indicate parental interest (their parent is the | |
123 | * original inode) */ | |
2fd6b7f5 | 124 | spin_lock(&alias->d_lock); |
946e51f2 | 125 | list_for_each_entry(child, &alias->d_subdirs, d_child) { |
c28f7e56 EP |
126 | if (!child->d_inode) |
127 | continue; | |
128 | ||
2fd6b7f5 | 129 | spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED); |
c28f7e56 EP |
130 | if (watched) |
131 | child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; | |
132 | else | |
133 | child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; | |
134 | spin_unlock(&child->d_lock); | |
135 | } | |
2fd6b7f5 | 136 | spin_unlock(&alias->d_lock); |
c28f7e56 | 137 | } |
873feea0 | 138 | spin_unlock(&inode->i_lock); |
c28f7e56 EP |
139 | } |
140 | ||
141 | /* Notify this dentry's parent about a child's events. */ | |
12c7f9dc | 142 | int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask) |
c28f7e56 EP |
143 | { |
144 | struct dentry *parent; | |
145 | struct inode *p_inode; | |
52420392 | 146 | int ret = 0; |
c28f7e56 | 147 | |
72acc854 | 148 | if (!dentry) |
2069601b | 149 | dentry = path->dentry; |
28c60e37 | 150 | |
c28f7e56 | 151 | if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED)) |
52420392 | 152 | return 0; |
c28f7e56 | 153 | |
4d4eb366 | 154 | parent = dget_parent(dentry); |
c28f7e56 EP |
155 | p_inode = parent->d_inode; |
156 | ||
b469e7e4 | 157 | if (unlikely(!fsnotify_inode_watches_children(p_inode))) { |
4d4eb366 | 158 | __fsnotify_update_child_dentry_flags(p_inode); |
b469e7e4 | 159 | } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) { |
49d31c2f AV |
160 | struct name_snapshot name; |
161 | ||
c28f7e56 EP |
162 | /* we are notifying a parent so come up with the new mask which |
163 | * specifies these are events which came from a child. */ | |
164 | mask |= FS_EVENT_ON_CHILD; | |
165 | ||
49d31c2f | 166 | take_dentry_name_snapshot(&name, dentry); |
2069601b | 167 | if (path) |
52420392 | 168 | ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH, |
25b229df | 169 | &name.name, 0); |
28c60e37 | 170 | else |
52420392 | 171 | ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE, |
25b229df | 172 | &name.name, 0); |
49d31c2f | 173 | release_dentry_name_snapshot(&name); |
c28f7e56 EP |
174 | } |
175 | ||
4d4eb366 | 176 | dput(parent); |
52420392 EP |
177 | |
178 | return ret; | |
c28f7e56 EP |
179 | } |
180 | EXPORT_SYMBOL_GPL(__fsnotify_parent); | |
181 | ||
fd657170 | 182 | static int send_to_group(struct inode *to_tell, |
e637835e | 183 | __u32 mask, const void *data, |
613a807f | 184 | int data_is, u32 cookie, |
e43e9c33 | 185 | const struct qstr *file_name, |
9385a84d | 186 | struct fsnotify_iter_info *iter_info) |
7131485a | 187 | { |
faa9560a | 188 | struct fsnotify_group *group = NULL; |
007d1e83 | 189 | __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS); |
92183a42 AG |
190 | __u32 marks_mask = 0; |
191 | __u32 marks_ignored_mask = 0; | |
3dca1a74 AG |
192 | struct fsnotify_mark *mark; |
193 | int type; | |
613a807f | 194 | |
5b0457ad | 195 | if (WARN_ON(!iter_info->report_mask)) |
faa9560a | 196 | return 0; |
ce8f76fb EP |
197 | |
198 | /* clear ignored on inode modification */ | |
199 | if (mask & FS_MODIFY) { | |
3dca1a74 AG |
200 | fsnotify_foreach_obj_type(type) { |
201 | if (!fsnotify_iter_should_report_type(iter_info, type)) | |
202 | continue; | |
203 | mark = iter_info->marks[type]; | |
204 | if (mark && | |
205 | !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) | |
206 | mark->ignored_mask = 0; | |
207 | } | |
ce8f76fb | 208 | } |
613a807f | 209 | |
3dca1a74 AG |
210 | fsnotify_foreach_obj_type(type) { |
211 | if (!fsnotify_iter_should_report_type(iter_info, type)) | |
212 | continue; | |
213 | mark = iter_info->marks[type]; | |
214 | /* does the object mark tell us to do something? */ | |
215 | if (mark) { | |
216 | group = mark->group; | |
217 | marks_mask |= mark->mask; | |
218 | marks_ignored_mask |= mark->ignored_mask; | |
219 | } | |
ce8f76fb EP |
220 | } |
221 | ||
3dca1a74 | 222 | pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x" |
7053aee2 | 223 | " data=%p data_is=%d cookie=%d\n", |
3dca1a74 AG |
224 | __func__, group, to_tell, mask, marks_mask, marks_ignored_mask, |
225 | data, data_is, cookie); | |
faa9560a | 226 | |
92183a42 | 227 | if (!(test_mask & marks_mask & ~marks_ignored_mask)) |
613a807f EP |
228 | return 0; |
229 | ||
5b0457ad | 230 | return group->ops->handle_event(group, to_tell, mask, data, data_is, |
9385a84d | 231 | file_name, cookie, iter_info); |
7131485a EP |
232 | } |
233 | ||
3427ce71 MS |
234 | static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp) |
235 | { | |
236 | struct fsnotify_mark_connector *conn; | |
237 | struct hlist_node *node = NULL; | |
238 | ||
239 | conn = srcu_dereference(*connp, &fsnotify_mark_srcu); | |
240 | if (conn) | |
241 | node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu); | |
242 | ||
243 | return hlist_entry_safe(node, struct fsnotify_mark, obj_list); | |
244 | } | |
245 | ||
246 | static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark) | |
247 | { | |
248 | struct hlist_node *node = NULL; | |
249 | ||
250 | if (mark) | |
251 | node = srcu_dereference(mark->obj_list.next, | |
252 | &fsnotify_mark_srcu); | |
253 | ||
254 | return hlist_entry_safe(node, struct fsnotify_mark, obj_list); | |
255 | } | |
256 | ||
d9a6f30b AG |
257 | /* |
258 | * iter_info is a multi head priority queue of marks. | |
259 | * Pick a subset of marks from queue heads, all with the | |
260 | * same group and set the report_mask for selected subset. | |
261 | * Returns the report_mask of the selected subset. | |
262 | */ | |
263 | static unsigned int fsnotify_iter_select_report_types( | |
264 | struct fsnotify_iter_info *iter_info) | |
265 | { | |
47d9c7cc AG |
266 | struct fsnotify_group *max_prio_group = NULL; |
267 | struct fsnotify_mark *mark; | |
268 | int type; | |
269 | ||
270 | /* Choose max prio group among groups of all queue heads */ | |
271 | fsnotify_foreach_obj_type(type) { | |
272 | mark = iter_info->marks[type]; | |
273 | if (mark && | |
274 | fsnotify_compare_groups(max_prio_group, mark->group) > 0) | |
275 | max_prio_group = mark->group; | |
276 | } | |
d9a6f30b | 277 | |
47d9c7cc | 278 | if (!max_prio_group) |
d9a6f30b AG |
279 | return 0; |
280 | ||
47d9c7cc | 281 | /* Set the report mask for marks from same group as max prio group */ |
d9a6f30b | 282 | iter_info->report_mask = 0; |
47d9c7cc AG |
283 | fsnotify_foreach_obj_type(type) { |
284 | mark = iter_info->marks[type]; | |
285 | if (mark && | |
286 | fsnotify_compare_groups(max_prio_group, mark->group) == 0) | |
287 | fsnotify_iter_set_report_type(iter_info, type); | |
288 | } | |
d9a6f30b AG |
289 | |
290 | return iter_info->report_mask; | |
291 | } | |
292 | ||
293 | /* | |
294 | * Pop from iter_info multi head queue, the marks that were iterated in the | |
295 | * current iteration step. | |
296 | */ | |
297 | static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info) | |
298 | { | |
47d9c7cc | 299 | int type; |
d9a6f30b | 300 | |
47d9c7cc AG |
301 | fsnotify_foreach_obj_type(type) { |
302 | if (fsnotify_iter_should_report_type(iter_info, type)) | |
303 | iter_info->marks[type] = | |
304 | fsnotify_next_mark(iter_info->marks[type]); | |
305 | } | |
d9a6f30b AG |
306 | } |
307 | ||
90586523 EP |
308 | /* |
309 | * This is the main call to fsnotify. The VFS calls into hook specific functions | |
310 | * in linux/fsnotify.h. Those functions then in turn call here. Here will call | |
311 | * out to all of the registered fsnotify_group. Those groups can then use the | |
312 | * notification event in whatever means they feel necessary. | |
313 | */ | |
e637835e | 314 | int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, |
25b229df | 315 | const struct qstr *file_name, u32 cookie) |
90586523 | 316 | { |
3427ce71 | 317 | struct fsnotify_iter_info iter_info = {}; |
45a9fb37 | 318 | struct super_block *sb = to_tell->i_sb; |
60f7ed8c | 319 | struct mount *mnt = NULL; |
45a9fb37 | 320 | __u32 mnt_or_sb_mask = sb->s_fsnotify_mask; |
9385a84d | 321 | int ret = 0; |
007d1e83 | 322 | __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS); |
90586523 | 323 | |
60f7ed8c | 324 | if (data_is == FSNOTIFY_EVENT_PATH) { |
e637835e | 325 | mnt = real_mount(((const struct path *)data)->mnt); |
45a9fb37 | 326 | mnt_or_sb_mask |= mnt->mnt_fsnotify_mask; |
60f7ed8c | 327 | } |
b469e7e4 AG |
328 | /* An event "on child" is not intended for a mount/sb mark */ |
329 | if (mask & FS_EVENT_ON_CHILD) | |
330 | mnt_or_sb_mask = 0; | |
613a807f | 331 | |
7c49b861 DH |
332 | /* |
333 | * Optimization: srcu_read_lock() has a memory barrier which can | |
334 | * be expensive. It protects walking the *_fsnotify_marks lists. | |
335 | * However, if we do not walk the lists, we do not have to do | |
336 | * SRCU because we have no references to any objects and do not | |
337 | * need SRCU to keep them "alive". | |
338 | */ | |
45a9fb37 AG |
339 | if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks && |
340 | (!mnt || !mnt->mnt_fsnotify_marks)) | |
7c49b861 | 341 | return 0; |
613a807f EP |
342 | /* |
343 | * if this is a modify event we may need to clear the ignored masks | |
60f7ed8c | 344 | * otherwise return if neither the inode nor the vfsmount/sb care about |
613a807f EP |
345 | * this type of event. |
346 | */ | |
347 | if (!(mask & FS_MODIFY) && | |
60f7ed8c | 348 | !(test_mask & (to_tell->i_fsnotify_mask | mnt_or_sb_mask))) |
613a807f | 349 | return 0; |
3a9fb89f | 350 | |
9385a84d | 351 | iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); |
7131485a | 352 | |
9bdda4e9 AG |
353 | iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] = |
354 | fsnotify_first_mark(&to_tell->i_fsnotify_marks); | |
45a9fb37 AG |
355 | iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] = |
356 | fsnotify_first_mark(&sb->s_fsnotify_marks); | |
9bdda4e9 | 357 | if (mnt) { |
47d9c7cc | 358 | iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] = |
3427ce71 | 359 | fsnotify_first_mark(&mnt->mnt_fsnotify_marks); |
90586523 | 360 | } |
75c1be48 | 361 | |
8edc6e16 | 362 | /* |
60f7ed8c AG |
363 | * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark |
364 | * ignore masks are properly reflected for mount/sb mark notifications. | |
8edc6e16 JK |
365 | * That's why this traversal is so complicated... |
366 | */ | |
d9a6f30b | 367 | while (fsnotify_iter_select_report_types(&iter_info)) { |
5b0457ad | 368 | ret = send_to_group(to_tell, mask, data, data_is, cookie, |
e43e9c33 | 369 | file_name, &iter_info); |
613a807f | 370 | |
ff8bcbd0 EP |
371 | if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS)) |
372 | goto out; | |
373 | ||
d9a6f30b | 374 | fsnotify_iter_next(&iter_info); |
7131485a | 375 | } |
ff8bcbd0 EP |
376 | ret = 0; |
377 | out: | |
9385a84d | 378 | srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx); |
c4ec54b4 | 379 | |
98b5c10d | 380 | return ret; |
90586523 EP |
381 | } |
382 | EXPORT_SYMBOL_GPL(fsnotify); | |
383 | ||
384 | static __init int fsnotify_init(void) | |
385 | { | |
75c1be48 EP |
386 | int ret; |
387 | ||
66917a31 | 388 | BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25); |
20dee624 | 389 | |
75c1be48 EP |
390 | ret = init_srcu_struct(&fsnotify_mark_srcu); |
391 | if (ret) | |
392 | panic("initializing fsnotify_mark_srcu"); | |
393 | ||
9dd813c1 JK |
394 | fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector, |
395 | SLAB_PANIC); | |
396 | ||
75c1be48 | 397 | return 0; |
90586523 | 398 | } |
75c1be48 | 399 | core_initcall(fsnotify_init); |