1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Filesystem access notification for Linux
4  *
5  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <[email protected]>
6  */
7 
8 #ifndef __LINUX_FSNOTIFY_BACKEND_H
9 #define __LINUX_FSNOTIFY_BACKEND_H
10 
11 #ifdef __KERNEL__
12 
13 #include <linux/idr.h> /* inotify uses this */
14 #include <linux/fs.h> /* struct inode */
15 #include <linux/list.h>
16 #include <linux/path.h> /* struct path */
17 #include <linux/spinlock.h>
18 #include <linux/types.h>
19 #include <linux/atomic.h>
20 #include <linux/user_namespace.h>
21 #include <linux/refcount.h>
22 
23 /*
24  * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
25  * convert between them.  dnotify only needs conversion at watch creation
26  * so no perf loss there.  fanotify isn't defined yet, so it can use the
27  * wholes if it needs more events.
28  */
29 #define FS_ACCESS		0x00000001	/* File was accessed */
30 #define FS_MODIFY		0x00000002	/* File was modified */
31 #define FS_ATTRIB		0x00000004	/* Metadata changed */
32 #define FS_CLOSE_WRITE		0x00000008	/* Writtable file was closed */
33 #define FS_CLOSE_NOWRITE	0x00000010	/* Unwrittable file closed */
34 #define FS_OPEN			0x00000020	/* File was opened */
35 #define FS_MOVED_FROM		0x00000040	/* File was moved from X */
36 #define FS_MOVED_TO		0x00000080	/* File was moved to Y */
37 #define FS_CREATE		0x00000100	/* Subfile was created */
38 #define FS_DELETE		0x00000200	/* Subfile was deleted */
39 #define FS_DELETE_SELF		0x00000400	/* Self was deleted */
40 #define FS_MOVE_SELF		0x00000800	/* Self was moved */
41 
42 #define FS_UNMOUNT		0x00002000	/* inode on umount fs */
43 #define FS_Q_OVERFLOW		0x00004000	/* Event queued overflowed */
44 #define FS_IN_IGNORED		0x00008000	/* last inotify event here */
45 
46 #define FS_OPEN_PERM		0x00010000	/* open event in an permission hook */
47 #define FS_ACCESS_PERM		0x00020000	/* access event in a permissions hook */
48 
49 #define FS_EXCL_UNLINK		0x04000000	/* do not send events if object is unlinked */
50 #define FS_ISDIR		0x40000000	/* event occurred against dir */
51 #define FS_IN_ONESHOT		0x80000000	/* only send event once */
52 
53 #define FS_DN_RENAME		0x10000000	/* file renamed */
54 #define FS_DN_MULTISHOT		0x20000000	/* dnotify multishot */
55 
56 /* This inode cares about things that happen to its children.  Always set for
57  * dnotify and inotify. */
58 #define FS_EVENT_ON_CHILD	0x08000000
59 
60 /* This is a list of all events that may get sent to a parernt based on fs event
61  * happening to inodes inside that directory */
62 #define FS_EVENTS_POSS_ON_CHILD   (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
63 				   FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
64 				   FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
65 				   FS_DELETE | FS_OPEN_PERM | FS_ACCESS_PERM)
66 
67 #define FS_MOVE			(FS_MOVED_FROM | FS_MOVED_TO)
68 
69 #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM)
70 
71 #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
72 			     FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \
73 			     FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \
74 			     FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \
75 			     FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
76 			     FS_OPEN_PERM | FS_ACCESS_PERM | FS_EXCL_UNLINK | \
77 			     FS_ISDIR | FS_IN_ONESHOT | FS_DN_RENAME | \
78 			     FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
79 
80 struct fsnotify_group;
81 struct fsnotify_event;
82 struct fsnotify_mark;
83 struct fsnotify_event_private_data;
84 struct fsnotify_fname;
85 struct fsnotify_iter_info;
86 
87 /*
88  * Each group much define these ops.  The fsnotify infrastructure will call
89  * these operations for each relevant group.
90  *
91  * handle_event - main call for a group to handle an fs event
92  * free_group_priv - called when a group refcnt hits 0 to clean up the private union
93  * freeing_mark - called when a mark is being destroyed for some reason.  The group
94  * 		MUST be holding a reference on each mark and that reference must be
95  * 		dropped in this function.  inotify uses this function to send
96  * 		userspace messages that marks have been removed.
97  */
98 struct fsnotify_ops {
99 	int (*handle_event)(struct fsnotify_group *group,
100 			    struct inode *inode,
101 			    struct fsnotify_mark *inode_mark,
102 			    struct fsnotify_mark *vfsmount_mark,
103 			    u32 mask, const void *data, int data_type,
104 			    const unsigned char *file_name, u32 cookie,
105 			    struct fsnotify_iter_info *iter_info);
106 	void (*free_group_priv)(struct fsnotify_group *group);
107 	void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
108 	void (*free_event)(struct fsnotify_event *event);
109 	/* called on final put+free to free memory */
110 	void (*free_mark)(struct fsnotify_mark *mark);
111 };
112 
113 /*
114  * all of the information about the original object we want to now send to
115  * a group.  If you want to carry more info from the accessing task to the
116  * listener this structure is where you need to be adding fields.
117  */
118 struct fsnotify_event {
119 	struct list_head list;
120 	/* inode may ONLY be dereferenced during handle_event(). */
121 	struct inode *inode;	/* either the inode the event happened to or its parent */
122 	u32 mask;		/* the type of access, bitwise OR for FS_* event types */
123 };
124 
125 /*
126  * A group is a "thing" that wants to receive notification about filesystem
127  * events.  The mask holds the subset of event types this group cares about.
128  * refcnt on a group is up to the implementor and at any moment if it goes 0
129  * everything will be cleaned up.
130  */
131 struct fsnotify_group {
132 	/*
133 	 * How the refcnt is used is up to each group.  When the refcnt hits 0
134 	 * fsnotify will clean up all of the resources associated with this group.
135 	 * As an example, the dnotify group will always have a refcnt=1 and that
136 	 * will never change.  Inotify, on the other hand, has a group per
137 	 * inotify_init() and the refcnt will hit 0 only when that fd has been
138 	 * closed.
139 	 */
140 	refcount_t refcnt;		/* things with interest in this group */
141 
142 	const struct fsnotify_ops *ops;	/* how this group handles things */
143 
144 	/* needed to send notification to userspace */
145 	spinlock_t notification_lock;		/* protect the notification_list */
146 	struct list_head notification_list;	/* list of event_holder this group needs to send to userspace */
147 	wait_queue_head_t notification_waitq;	/* read() on the notification file blocks on this waitq */
148 	unsigned int q_len;			/* events on the queue */
149 	unsigned int max_events;		/* maximum events allowed on the list */
150 	/*
151 	 * Valid fsnotify group priorities.  Events are send in order from highest
152 	 * priority to lowest priority.  We default to the lowest priority.
153 	 */
154 	#define FS_PRIO_0	0 /* normal notifiers, no permissions */
155 	#define FS_PRIO_1	1 /* fanotify content based access control */
156 	#define FS_PRIO_2	2 /* fanotify pre-content access */
157 	unsigned int priority;
158 	bool shutdown;		/* group is being shut down, don't queue more events */
159 
160 	/* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
161 	struct mutex mark_mutex;	/* protect marks_list */
162 	atomic_t num_marks;		/* 1 for each mark and 1 for not being
163 					 * past the point of no return when freeing
164 					 * a group */
165 	struct list_head marks_list;	/* all inode marks for this group */
166 
167 	struct fasync_struct *fsn_fa;    /* async notification */
168 
169 	struct fsnotify_event *overflow_event;	/* Event we queue when the
170 						 * notification list is too
171 						 * full */
172 	atomic_t user_waits;		/* Number of tasks waiting for user
173 					 * response */
174 
175 	/* groups can define private fields here or use the void *private */
176 	union {
177 		void *private;
178 #ifdef CONFIG_INOTIFY_USER
179 		struct inotify_group_private_data {
180 			spinlock_t	idr_lock;
181 			struct idr      idr;
182 			struct ucounts *ucounts;
183 		} inotify_data;
184 #endif
185 #ifdef CONFIG_FANOTIFY
186 		struct fanotify_group_private_data {
187 			/* allows a group to block waiting for a userspace response */
188 			struct list_head access_list;
189 			wait_queue_head_t access_waitq;
190 			int f_flags;
191 			unsigned int max_marks;
192 			struct user_struct *user;
193 			bool audit;
194 		} fanotify_data;
195 #endif /* CONFIG_FANOTIFY */
196 	};
197 };
198 
199 /* when calling fsnotify tell it if the data is a path or inode */
200 #define FSNOTIFY_EVENT_NONE	0
201 #define FSNOTIFY_EVENT_PATH	1
202 #define FSNOTIFY_EVENT_INODE	2
203 
204 /*
205  * Inode / vfsmount point to this structure which tracks all marks attached to
206  * the inode / vfsmount. The reference to inode / vfsmount is held by this
207  * structure. We destroy this structure when there are no more marks attached
208  * to it. The structure is protected by fsnotify_mark_srcu.
209  */
210 struct fsnotify_mark_connector {
211 	spinlock_t lock;
212 #define FSNOTIFY_OBJ_TYPE_INODE		0x01
213 #define FSNOTIFY_OBJ_TYPE_VFSMOUNT	0x02
214 #define FSNOTIFY_OBJ_ALL_TYPES		(FSNOTIFY_OBJ_TYPE_INODE | \
215 					 FSNOTIFY_OBJ_TYPE_VFSMOUNT)
216 	unsigned int flags;	/* Type of object [lock] */
217 	union {	/* Object pointer [lock] */
218 		struct inode *inode;
219 		struct vfsmount *mnt;
220 	};
221 	union {
222 		struct hlist_head list;
223 		/* Used listing heads to free after srcu period expires */
224 		struct fsnotify_mark_connector *destroy_next;
225 	};
226 };
227 
228 /*
229  * A mark is simply an object attached to an in core inode which allows an
230  * fsnotify listener to indicate they are either no longer interested in events
231  * of a type matching mask or only interested in those events.
232  *
233  * These are flushed when an inode is evicted from core and may be flushed
234  * when the inode is modified (as seen by fsnotify_access).  Some fsnotify
235  * users (such as dnotify) will flush these when the open fd is closed and not
236  * at inode eviction or modification.
237  *
238  * Text in brackets is showing the lock(s) protecting modifications of a
239  * particular entry. obj_lock means either inode->i_lock or
240  * mnt->mnt_root->d_lock depending on the mark type.
241  */
242 struct fsnotify_mark {
243 	/* Mask this mark is for [mark->lock, group->mark_mutex] */
244 	__u32 mask;
245 	/* We hold one for presence in g_list. Also one ref for each 'thing'
246 	 * in kernel that found and may be using this mark. */
247 	refcount_t refcnt;
248 	/* Group this mark is for. Set on mark creation, stable until last ref
249 	 * is dropped */
250 	struct fsnotify_group *group;
251 	/* List of marks by group->i_fsnotify_marks. Also reused for queueing
252 	 * mark into destroy_list when it's waiting for the end of SRCU period
253 	 * before it can be freed. [group->mark_mutex] */
254 	struct list_head g_list;
255 	/* Protects inode / mnt pointers, flags, masks */
256 	spinlock_t lock;
257 	/* List of marks for inode / vfsmount [connector->lock, mark ref] */
258 	struct hlist_node obj_list;
259 	/* Head of list of marks for an object [mark ref] */
260 	struct fsnotify_mark_connector *connector;
261 	/* Events types to ignore [mark->lock, group->mark_mutex] */
262 	__u32 ignored_mask;
263 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY	0x01
264 #define FSNOTIFY_MARK_FLAG_ALIVE		0x02
265 #define FSNOTIFY_MARK_FLAG_ATTACHED		0x04
266 	unsigned int flags;		/* flags [mark->lock] */
267 };
268 
269 #ifdef CONFIG_FSNOTIFY
270 
271 /* called from the vfs helpers */
272 
273 /* main fsnotify call to send events */
274 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
275 		    const unsigned char *name, u32 cookie);
276 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
277 extern void __fsnotify_inode_delete(struct inode *inode);
278 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
279 extern u32 fsnotify_get_cookie(void);
280 
281 static inline int fsnotify_inode_watches_children(struct inode *inode)
282 {
283 	/* FS_EVENT_ON_CHILD is set if the inode may care */
284 	if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
285 		return 0;
286 	/* this inode might care about child events, does it care about the
287 	 * specific set of events that can happen on a child? */
288 	return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
289 }
290 
291 /*
292  * Update the dentry with a flag indicating the interest of its parent to receive
293  * filesystem events when those events happens to this dentry->d_inode.
294  */
295 static inline void fsnotify_update_flags(struct dentry *dentry)
296 {
297 	assert_spin_locked(&dentry->d_lock);
298 
299 	/*
300 	 * Serialisation of setting PARENT_WATCHED on the dentries is provided
301 	 * by d_lock. If inotify_inode_watched changes after we have taken
302 	 * d_lock, the following __fsnotify_update_child_dentry_flags call will
303 	 * find our entry, so it will spin until we complete here, and update
304 	 * us with the new state.
305 	 */
306 	if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
307 		dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
308 	else
309 		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
310 }
311 
312 /* called from fsnotify listeners, such as fanotify or dnotify */
313 
314 /* create a new group */
315 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
316 /* get reference to a group */
317 extern void fsnotify_get_group(struct fsnotify_group *group);
318 /* drop reference on a group from fsnotify_alloc_group */
319 extern void fsnotify_put_group(struct fsnotify_group *group);
320 /* group destruction begins, stop queuing new events */
321 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
322 /* destroy group */
323 extern void fsnotify_destroy_group(struct fsnotify_group *group);
324 /* fasync handler function */
325 extern int fsnotify_fasync(int fd, struct file *file, int on);
326 /* Free event from memory */
327 extern void fsnotify_destroy_event(struct fsnotify_group *group,
328 				   struct fsnotify_event *event);
329 /* attach the event to the group notification queue */
330 extern int fsnotify_add_event(struct fsnotify_group *group,
331 			      struct fsnotify_event *event,
332 			      int (*merge)(struct list_head *,
333 					   struct fsnotify_event *));
334 /* true if the group notification queue is empty */
335 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
336 /* return, but do not dequeue the first event on the notification queue */
337 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
338 /* return AND dequeue the first event on the notification queue */
339 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
340 
341 /* functions used to manipulate the marks attached to inodes */
342 
343 /* Calculate mask of events for a list of marks */
344 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
345 extern void fsnotify_init_mark(struct fsnotify_mark *mark,
346 			       struct fsnotify_group *group);
347 /* Find mark belonging to given group in the list of marks */
348 extern struct fsnotify_mark *fsnotify_find_mark(
349 				struct fsnotify_mark_connector __rcu **connp,
350 				struct fsnotify_group *group);
351 /* attach the mark to the inode or vfsmount */
352 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct inode *inode,
353 			     struct vfsmount *mnt, int allow_dups);
354 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
355 				    struct inode *inode, struct vfsmount *mnt, int allow_dups);
356 /* given a group and a mark, flag mark to be freed when all references are dropped */
357 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
358 				  struct fsnotify_group *group);
359 /* detach mark from inode / mount list, group list, drop inode reference */
360 extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
361 /* free mark */
362 extern void fsnotify_free_mark(struct fsnotify_mark *mark);
363 /* run all the marks in a group, and clear all of the marks attached to given object type */
364 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
365 /* run all the marks in a group, and clear all of the vfsmount marks */
366 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
367 {
368 	fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT);
369 }
370 /* run all the marks in a group, and clear all of the inode marks */
371 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
372 {
373 	fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE);
374 }
375 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
376 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
377 extern void fsnotify_unmount_inodes(struct super_block *sb);
378 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
379 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
380 
381 /* put here because inotify does some weird stuff when destroying watches */
382 extern void fsnotify_init_event(struct fsnotify_event *event,
383 				struct inode *to_tell, u32 mask);
384 
385 #else
386 
387 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
388 			   const unsigned char *name, u32 cookie)
389 {
390 	return 0;
391 }
392 
393 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
394 {
395 	return 0;
396 }
397 
398 static inline void __fsnotify_inode_delete(struct inode *inode)
399 {}
400 
401 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
402 {}
403 
404 static inline void fsnotify_update_flags(struct dentry *dentry)
405 {}
406 
407 static inline u32 fsnotify_get_cookie(void)
408 {
409 	return 0;
410 }
411 
412 static inline void fsnotify_unmount_inodes(struct super_block *sb)
413 {}
414 
415 #endif	/* CONFIG_FSNOTIFY */
416 
417 #endif	/* __KERNEL __ */
418 
419 #endif	/* __LINUX_FSNOTIFY_BACKEND_H */
420