1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
6 #include <linux/dcache.h>
9 #include <linux/init.h>
10 #include <linux/module.h>
11 #include <linux/mount.h>
12 #include <linux/srcu.h>
14 #include <linux/fsnotify_backend.h>
18 * Clear all of the marks on an inode when it is being evicted from core
20 void __fsnotify_inode_delete(struct inode
*inode
)
22 fsnotify_clear_marks_by_inode(inode
);
24 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete
);
26 void __fsnotify_vfsmount_delete(struct vfsmount
*mnt
)
28 fsnotify_clear_marks_by_mount(mnt
);
32 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes.
33 * @sb: superblock being unmounted.
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.
38 static void fsnotify_unmount_inodes(struct super_block
*sb
)
40 struct inode
*inode
, *iput_inode
= NULL
;
42 spin_lock(&sb
->s_inode_list_lock
);
43 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
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.
49 spin_lock(&inode
->i_lock
);
50 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) {
51 spin_unlock(&inode
->i_lock
);
56 * If i_count is zero, the inode cannot have any watches and
57 * doing an __iget/iput with SB_ACTIVE clear would actually
58 * evict all inodes with zero i_count from icache which is
59 * unnecessarily violent and may in fact be illegal to do.
60 * However, we should have been called /after/ evict_inodes
61 * removed all zero refcount inodes, in any case. Test to
64 if (!atomic_read(&inode
->i_count
)) {
65 spin_unlock(&inode
->i_lock
);
70 spin_unlock(&inode
->i_lock
);
71 spin_unlock(&sb
->s_inode_list_lock
);
76 /* for each watch, send FS_UNMOUNT and then remove it */
77 fsnotify(inode
, FS_UNMOUNT
, inode
, FSNOTIFY_EVENT_INODE
, NULL
, 0);
79 fsnotify_inode_delete(inode
);
84 spin_lock(&sb
->s_inode_list_lock
);
86 spin_unlock(&sb
->s_inode_list_lock
);
90 /* Wait for outstanding inode references from connectors */
91 wait_var_event(&sb
->s_fsnotify_inode_refs
,
92 !atomic_long_read(&sb
->s_fsnotify_inode_refs
));
95 void fsnotify_sb_delete(struct super_block
*sb
)
97 fsnotify_unmount_inodes(sb
);
98 fsnotify_clear_marks_by_sb(sb
);
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.
108 void __fsnotify_update_child_dentry_flags(struct inode
*inode
)
110 struct dentry
*alias
;
113 if (!S_ISDIR(inode
->i_mode
))
116 /* determine if the children should tell inode about their events */
117 watched
= fsnotify_inode_watches_children(inode
);
119 spin_lock(&inode
->i_lock
);
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 */
122 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
123 struct dentry
*child
;
125 /* run all of the children of the original inode and fix their
126 * d_flags to indicate parental interest (their parent is the
128 spin_lock(&alias
->d_lock
);
129 list_for_each_entry(child
, &alias
->d_subdirs
, d_child
) {
133 spin_lock_nested(&child
->d_lock
, DENTRY_D_LOCK_NESTED
);
135 child
->d_flags
|= DCACHE_FSNOTIFY_PARENT_WATCHED
;
137 child
->d_flags
&= ~DCACHE_FSNOTIFY_PARENT_WATCHED
;
138 spin_unlock(&child
->d_lock
);
140 spin_unlock(&alias
->d_lock
);
142 spin_unlock(&inode
->i_lock
);
145 /* Notify this dentry's parent about a child's events. */
146 int fsnotify_parent(struct dentry
*dentry
, __u32 mask
, const void *data
,
149 struct dentry
*parent
;
150 struct inode
*p_inode
;
153 if (!(dentry
->d_flags
& DCACHE_FSNOTIFY_PARENT_WATCHED
))
156 parent
= dget_parent(dentry
);
157 p_inode
= parent
->d_inode
;
159 if (unlikely(!fsnotify_inode_watches_children(p_inode
))) {
160 __fsnotify_update_child_dentry_flags(p_inode
);
161 } else if (p_inode
->i_fsnotify_mask
& mask
& ALL_FSNOTIFY_EVENTS
) {
162 struct name_snapshot name
;
164 /* we are notifying a parent so come up with the new mask which
165 * specifies these are events which came from a child. */
166 mask
|= FS_EVENT_ON_CHILD
;
168 take_dentry_name_snapshot(&name
, dentry
);
169 ret
= fsnotify(p_inode
, mask
, data
, data_type
, &name
.name
, 0);
170 release_dentry_name_snapshot(&name
);
177 EXPORT_SYMBOL_GPL(fsnotify_parent
);
179 static int send_to_group(struct inode
*to_tell
,
180 __u32 mask
, const void *data
,
181 int data_is
, u32 cookie
,
182 const struct qstr
*file_name
,
183 struct fsnotify_iter_info
*iter_info
)
185 struct fsnotify_group
*group
= NULL
;
186 __u32 test_mask
= (mask
& ALL_FSNOTIFY_EVENTS
);
187 __u32 marks_mask
= 0;
188 __u32 marks_ignored_mask
= 0;
189 struct fsnotify_mark
*mark
;
192 if (WARN_ON(!iter_info
->report_mask
))
195 /* clear ignored on inode modification */
196 if (mask
& FS_MODIFY
) {
197 fsnotify_foreach_obj_type(type
) {
198 if (!fsnotify_iter_should_report_type(iter_info
, type
))
200 mark
= iter_info
->marks
[type
];
202 !(mark
->flags
& FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY
))
203 mark
->ignored_mask
= 0;
207 fsnotify_foreach_obj_type(type
) {
208 if (!fsnotify_iter_should_report_type(iter_info
, type
))
210 mark
= iter_info
->marks
[type
];
211 /* does the object mark tell us to do something? */
214 marks_mask
|= mark
->mask
;
215 marks_ignored_mask
|= mark
->ignored_mask
;
219 pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x"
220 " data=%p data_is=%d cookie=%d\n",
221 __func__
, group
, to_tell
, mask
, marks_mask
, marks_ignored_mask
,
222 data
, data_is
, cookie
);
224 if (!(test_mask
& marks_mask
& ~marks_ignored_mask
))
227 return group
->ops
->handle_event(group
, to_tell
, mask
, data
, data_is
,
228 file_name
, cookie
, iter_info
);
231 static struct fsnotify_mark
*fsnotify_first_mark(struct fsnotify_mark_connector
**connp
)
233 struct fsnotify_mark_connector
*conn
;
234 struct hlist_node
*node
= NULL
;
236 conn
= srcu_dereference(*connp
, &fsnotify_mark_srcu
);
238 node
= srcu_dereference(conn
->list
.first
, &fsnotify_mark_srcu
);
240 return hlist_entry_safe(node
, struct fsnotify_mark
, obj_list
);
243 static struct fsnotify_mark
*fsnotify_next_mark(struct fsnotify_mark
*mark
)
245 struct hlist_node
*node
= NULL
;
248 node
= srcu_dereference(mark
->obj_list
.next
,
249 &fsnotify_mark_srcu
);
251 return hlist_entry_safe(node
, struct fsnotify_mark
, obj_list
);
255 * iter_info is a multi head priority queue of marks.
256 * Pick a subset of marks from queue heads, all with the
257 * same group and set the report_mask for selected subset.
258 * Returns the report_mask of the selected subset.
260 static unsigned int fsnotify_iter_select_report_types(
261 struct fsnotify_iter_info
*iter_info
)
263 struct fsnotify_group
*max_prio_group
= NULL
;
264 struct fsnotify_mark
*mark
;
267 /* Choose max prio group among groups of all queue heads */
268 fsnotify_foreach_obj_type(type
) {
269 mark
= iter_info
->marks
[type
];
271 fsnotify_compare_groups(max_prio_group
, mark
->group
) > 0)
272 max_prio_group
= mark
->group
;
278 /* Set the report mask for marks from same group as max prio group */
279 iter_info
->report_mask
= 0;
280 fsnotify_foreach_obj_type(type
) {
281 mark
= iter_info
->marks
[type
];
283 fsnotify_compare_groups(max_prio_group
, mark
->group
) == 0)
284 fsnotify_iter_set_report_type(iter_info
, type
);
287 return iter_info
->report_mask
;
291 * Pop from iter_info multi head queue, the marks that were iterated in the
292 * current iteration step.
294 static void fsnotify_iter_next(struct fsnotify_iter_info
*iter_info
)
298 fsnotify_foreach_obj_type(type
) {
299 if (fsnotify_iter_should_report_type(iter_info
, type
))
300 iter_info
->marks
[type
] =
301 fsnotify_next_mark(iter_info
->marks
[type
]);
306 * This is the main call to fsnotify. The VFS calls into hook specific functions
307 * in linux/fsnotify.h. Those functions then in turn call here. Here will call
308 * out to all of the registered fsnotify_group. Those groups can then use the
309 * notification event in whatever means they feel necessary.
311 int fsnotify(struct inode
*to_tell
, __u32 mask
, const void *data
, int data_is
,
312 const struct qstr
*file_name
, u32 cookie
)
314 const struct path
*path
= fsnotify_data_path(data
, data_is
);
315 struct fsnotify_iter_info iter_info
= {};
316 struct super_block
*sb
= to_tell
->i_sb
;
317 struct mount
*mnt
= NULL
;
318 __u32 mnt_or_sb_mask
= sb
->s_fsnotify_mask
;
320 __u32 test_mask
= (mask
& ALL_FSNOTIFY_EVENTS
);
323 mnt
= real_mount(path
->mnt
);
324 mnt_or_sb_mask
|= mnt
->mnt_fsnotify_mask
;
326 /* An event "on child" is not intended for a mount/sb mark */
327 if (mask
& FS_EVENT_ON_CHILD
)
331 * Optimization: srcu_read_lock() has a memory barrier which can
332 * be expensive. It protects walking the *_fsnotify_marks lists.
333 * However, if we do not walk the lists, we do not have to do
334 * SRCU because we have no references to any objects and do not
335 * need SRCU to keep them "alive".
337 if (!to_tell
->i_fsnotify_marks
&& !sb
->s_fsnotify_marks
&&
338 (!mnt
|| !mnt
->mnt_fsnotify_marks
))
341 * if this is a modify event we may need to clear the ignored masks
342 * otherwise return if neither the inode nor the vfsmount/sb care about
343 * this type of event.
345 if (!(mask
& FS_MODIFY
) &&
346 !(test_mask
& (to_tell
->i_fsnotify_mask
| mnt_or_sb_mask
)))
349 iter_info
.srcu_idx
= srcu_read_lock(&fsnotify_mark_srcu
);
351 iter_info
.marks
[FSNOTIFY_OBJ_TYPE_INODE
] =
352 fsnotify_first_mark(&to_tell
->i_fsnotify_marks
);
353 iter_info
.marks
[FSNOTIFY_OBJ_TYPE_SB
] =
354 fsnotify_first_mark(&sb
->s_fsnotify_marks
);
356 iter_info
.marks
[FSNOTIFY_OBJ_TYPE_VFSMOUNT
] =
357 fsnotify_first_mark(&mnt
->mnt_fsnotify_marks
);
361 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
362 * ignore masks are properly reflected for mount/sb mark notifications.
363 * That's why this traversal is so complicated...
365 while (fsnotify_iter_select_report_types(&iter_info
)) {
366 ret
= send_to_group(to_tell
, mask
, data
, data_is
, cookie
,
367 file_name
, &iter_info
);
369 if (ret
&& (mask
& ALL_FSNOTIFY_PERM_EVENTS
))
372 fsnotify_iter_next(&iter_info
);
376 srcu_read_unlock(&fsnotify_mark_srcu
, iter_info
.srcu_idx
);
380 EXPORT_SYMBOL_GPL(fsnotify
);
382 static __init
int fsnotify_init(void)
386 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS
) != 26);
388 ret
= init_srcu_struct(&fsnotify_mark_srcu
);
390 panic("initializing fsnotify_mark_srcu");
392 fsnotify_mark_connector_cachep
= KMEM_CACHE(fsnotify_mark_connector
,
397 core_initcall(fsnotify_init
);