xref: /openbmc/linux/include/linux/fsnotify_backend.h (revision 791d3ef2e11100449837dc0b6fe884e60ca3a484)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Filesystem access notification for Linux
4  *
5  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
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 			    u32 mask, const void *data, int data_type,
102 			    const unsigned char *file_name, u32 cookie,
103 			    struct fsnotify_iter_info *iter_info);
104 	void (*free_group_priv)(struct fsnotify_group *group);
105 	void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
106 	void (*free_event)(struct fsnotify_event *event);
107 	/* called on final put+free to free memory */
108 	void (*free_mark)(struct fsnotify_mark *mark);
109 };
110 
111 /*
112  * all of the information about the original object we want to now send to
113  * a group.  If you want to carry more info from the accessing task to the
114  * listener this structure is where you need to be adding fields.
115  */
116 struct fsnotify_event {
117 	struct list_head list;
118 	/* inode may ONLY be dereferenced during handle_event(). */
119 	struct inode *inode;	/* either the inode the event happened to or its parent */
120 	u32 mask;		/* the type of access, bitwise OR for FS_* event types */
121 };
122 
123 /*
124  * A group is a "thing" that wants to receive notification about filesystem
125  * events.  The mask holds the subset of event types this group cares about.
126  * refcnt on a group is up to the implementor and at any moment if it goes 0
127  * everything will be cleaned up.
128  */
129 struct fsnotify_group {
130 	/*
131 	 * How the refcnt is used is up to each group.  When the refcnt hits 0
132 	 * fsnotify will clean up all of the resources associated with this group.
133 	 * As an example, the dnotify group will always have a refcnt=1 and that
134 	 * will never change.  Inotify, on the other hand, has a group per
135 	 * inotify_init() and the refcnt will hit 0 only when that fd has been
136 	 * closed.
137 	 */
138 	refcount_t refcnt;		/* things with interest in this group */
139 
140 	const struct fsnotify_ops *ops;	/* how this group handles things */
141 
142 	/* needed to send notification to userspace */
143 	spinlock_t notification_lock;		/* protect the notification_list */
144 	struct list_head notification_list;	/* list of event_holder this group needs to send to userspace */
145 	wait_queue_head_t notification_waitq;	/* read() on the notification file blocks on this waitq */
146 	unsigned int q_len;			/* events on the queue */
147 	unsigned int max_events;		/* maximum events allowed on the list */
148 	/*
149 	 * Valid fsnotify group priorities.  Events are send in order from highest
150 	 * priority to lowest priority.  We default to the lowest priority.
151 	 */
152 	#define FS_PRIO_0	0 /* normal notifiers, no permissions */
153 	#define FS_PRIO_1	1 /* fanotify content based access control */
154 	#define FS_PRIO_2	2 /* fanotify pre-content access */
155 	unsigned int priority;
156 	bool shutdown;		/* group is being shut down, don't queue more events */
157 
158 	/* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
159 	struct mutex mark_mutex;	/* protect marks_list */
160 	atomic_t num_marks;		/* 1 for each mark and 1 for not being
161 					 * past the point of no return when freeing
162 					 * a group */
163 	struct list_head marks_list;	/* all inode marks for this group */
164 
165 	struct fasync_struct *fsn_fa;    /* async notification */
166 
167 	struct fsnotify_event *overflow_event;	/* Event we queue when the
168 						 * notification list is too
169 						 * full */
170 	atomic_t user_waits;		/* Number of tasks waiting for user
171 					 * response */
172 
173 	/* groups can define private fields here or use the void *private */
174 	union {
175 		void *private;
176 #ifdef CONFIG_INOTIFY_USER
177 		struct inotify_group_private_data {
178 			spinlock_t	idr_lock;
179 			struct idr      idr;
180 			struct ucounts *ucounts;
181 		} inotify_data;
182 #endif
183 #ifdef CONFIG_FANOTIFY
184 		struct fanotify_group_private_data {
185 			/* allows a group to block waiting for a userspace response */
186 			struct list_head access_list;
187 			wait_queue_head_t access_waitq;
188 			int f_flags;
189 			unsigned int max_marks;
190 			struct user_struct *user;
191 			bool audit;
192 		} fanotify_data;
193 #endif /* CONFIG_FANOTIFY */
194 	};
195 };
196 
197 /* when calling fsnotify tell it if the data is a path or inode */
198 #define FSNOTIFY_EVENT_NONE	0
199 #define FSNOTIFY_EVENT_PATH	1
200 #define FSNOTIFY_EVENT_INODE	2
201 
202 enum fsnotify_obj_type {
203 	FSNOTIFY_OBJ_TYPE_INODE,
204 	FSNOTIFY_OBJ_TYPE_VFSMOUNT,
205 	FSNOTIFY_OBJ_TYPE_COUNT,
206 	FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
207 };
208 
209 #define FSNOTIFY_OBJ_TYPE_INODE_FL	(1U << FSNOTIFY_OBJ_TYPE_INODE)
210 #define FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL	(1U << FSNOTIFY_OBJ_TYPE_VFSMOUNT)
211 #define FSNOTIFY_OBJ_ALL_TYPES_MASK	((1U << FSNOTIFY_OBJ_TYPE_COUNT) - 1)
212 
213 struct fsnotify_iter_info {
214 	struct fsnotify_mark *marks[FSNOTIFY_OBJ_TYPE_COUNT];
215 	unsigned int report_mask;
216 	int srcu_idx;
217 };
218 
219 static inline bool fsnotify_iter_should_report_type(
220 		struct fsnotify_iter_info *iter_info, int type)
221 {
222 	return (iter_info->report_mask & (1U << type));
223 }
224 
225 static inline void fsnotify_iter_set_report_type(
226 		struct fsnotify_iter_info *iter_info, int type)
227 {
228 	iter_info->report_mask |= (1U << type);
229 }
230 
231 static inline void fsnotify_iter_set_report_type_mark(
232 		struct fsnotify_iter_info *iter_info, int type,
233 		struct fsnotify_mark *mark)
234 {
235 	iter_info->marks[type] = mark;
236 	iter_info->report_mask |= (1U << type);
237 }
238 
239 #define FSNOTIFY_ITER_FUNCS(name, NAME) \
240 static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \
241 		struct fsnotify_iter_info *iter_info) \
242 { \
243 	return (iter_info->report_mask & FSNOTIFY_OBJ_TYPE_##NAME##_FL) ? \
244 		iter_info->marks[FSNOTIFY_OBJ_TYPE_##NAME] : NULL; \
245 }
246 
247 FSNOTIFY_ITER_FUNCS(inode, INODE)
248 FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT)
249 
250 #define fsnotify_foreach_obj_type(type) \
251 	for (type = 0; type < FSNOTIFY_OBJ_TYPE_COUNT; type++)
252 
253 /*
254  * Inode / vfsmount point to this structure which tracks all marks attached to
255  * the inode / vfsmount. The reference to inode / vfsmount is held by this
256  * structure. We destroy this structure when there are no more marks attached
257  * to it. The structure is protected by fsnotify_mark_srcu.
258  */
259 struct fsnotify_mark_connector {
260 	spinlock_t lock;
261 	unsigned int type;	/* Type of object [lock] */
262 	union {	/* Object pointer [lock] */
263 		struct inode *inode;
264 		struct vfsmount *mnt;
265 		/* Used listing heads to free after srcu period expires */
266 		struct fsnotify_mark_connector *destroy_next;
267 	};
268 	struct hlist_head list;
269 };
270 
271 /*
272  * A mark is simply an object attached to an in core inode which allows an
273  * fsnotify listener to indicate they are either no longer interested in events
274  * of a type matching mask or only interested in those events.
275  *
276  * These are flushed when an inode is evicted from core and may be flushed
277  * when the inode is modified (as seen by fsnotify_access).  Some fsnotify
278  * users (such as dnotify) will flush these when the open fd is closed and not
279  * at inode eviction or modification.
280  *
281  * Text in brackets is showing the lock(s) protecting modifications of a
282  * particular entry. obj_lock means either inode->i_lock or
283  * mnt->mnt_root->d_lock depending on the mark type.
284  */
285 struct fsnotify_mark {
286 	/* Mask this mark is for [mark->lock, group->mark_mutex] */
287 	__u32 mask;
288 	/* We hold one for presence in g_list. Also one ref for each 'thing'
289 	 * in kernel that found and may be using this mark. */
290 	refcount_t refcnt;
291 	/* Group this mark is for. Set on mark creation, stable until last ref
292 	 * is dropped */
293 	struct fsnotify_group *group;
294 	/* List of marks by group->marks_list. Also reused for queueing
295 	 * mark into destroy_list when it's waiting for the end of SRCU period
296 	 * before it can be freed. [group->mark_mutex] */
297 	struct list_head g_list;
298 	/* Protects inode / mnt pointers, flags, masks */
299 	spinlock_t lock;
300 	/* List of marks for inode / vfsmount [connector->lock, mark ref] */
301 	struct hlist_node obj_list;
302 	/* Head of list of marks for an object [mark ref] */
303 	struct fsnotify_mark_connector *connector;
304 	/* Events types to ignore [mark->lock, group->mark_mutex] */
305 	__u32 ignored_mask;
306 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY	0x01
307 #define FSNOTIFY_MARK_FLAG_ALIVE		0x02
308 #define FSNOTIFY_MARK_FLAG_ATTACHED		0x04
309 	unsigned int flags;		/* flags [mark->lock] */
310 };
311 
312 #ifdef CONFIG_FSNOTIFY
313 
314 /* called from the vfs helpers */
315 
316 /* main fsnotify call to send events */
317 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
318 		    const unsigned char *name, u32 cookie);
319 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
320 extern void __fsnotify_inode_delete(struct inode *inode);
321 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
322 extern u32 fsnotify_get_cookie(void);
323 
324 static inline int fsnotify_inode_watches_children(struct inode *inode)
325 {
326 	/* FS_EVENT_ON_CHILD is set if the inode may care */
327 	if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
328 		return 0;
329 	/* this inode might care about child events, does it care about the
330 	 * specific set of events that can happen on a child? */
331 	return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
332 }
333 
334 /*
335  * Update the dentry with a flag indicating the interest of its parent to receive
336  * filesystem events when those events happens to this dentry->d_inode.
337  */
338 static inline void fsnotify_update_flags(struct dentry *dentry)
339 {
340 	assert_spin_locked(&dentry->d_lock);
341 
342 	/*
343 	 * Serialisation of setting PARENT_WATCHED on the dentries is provided
344 	 * by d_lock. If inotify_inode_watched changes after we have taken
345 	 * d_lock, the following __fsnotify_update_child_dentry_flags call will
346 	 * find our entry, so it will spin until we complete here, and update
347 	 * us with the new state.
348 	 */
349 	if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
350 		dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
351 	else
352 		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
353 }
354 
355 /* called from fsnotify listeners, such as fanotify or dnotify */
356 
357 /* create a new group */
358 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
359 /* get reference to a group */
360 extern void fsnotify_get_group(struct fsnotify_group *group);
361 /* drop reference on a group from fsnotify_alloc_group */
362 extern void fsnotify_put_group(struct fsnotify_group *group);
363 /* group destruction begins, stop queuing new events */
364 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
365 /* destroy group */
366 extern void fsnotify_destroy_group(struct fsnotify_group *group);
367 /* fasync handler function */
368 extern int fsnotify_fasync(int fd, struct file *file, int on);
369 /* Free event from memory */
370 extern void fsnotify_destroy_event(struct fsnotify_group *group,
371 				   struct fsnotify_event *event);
372 /* attach the event to the group notification queue */
373 extern int fsnotify_add_event(struct fsnotify_group *group,
374 			      struct fsnotify_event *event,
375 			      int (*merge)(struct list_head *,
376 					   struct fsnotify_event *));
377 /* Queue overflow event to a notification group */
378 static inline void fsnotify_queue_overflow(struct fsnotify_group *group)
379 {
380 	fsnotify_add_event(group, group->overflow_event, NULL);
381 }
382 
383 /* true if the group notification queue is empty */
384 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
385 /* return, but do not dequeue the first event on the notification queue */
386 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
387 /* return AND dequeue the first event on the notification queue */
388 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
389 
390 /* functions used to manipulate the marks attached to inodes */
391 
392 /* Calculate mask of events for a list of marks */
393 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
394 extern void fsnotify_init_mark(struct fsnotify_mark *mark,
395 			       struct fsnotify_group *group);
396 /* Find mark belonging to given group in the list of marks */
397 extern struct fsnotify_mark *fsnotify_find_mark(
398 				struct fsnotify_mark_connector __rcu **connp,
399 				struct fsnotify_group *group);
400 /* attach the mark to the inode or vfsmount */
401 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct inode *inode,
402 			     struct vfsmount *mnt, int allow_dups);
403 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
404 				    struct inode *inode, struct vfsmount *mnt,
405 				    int allow_dups);
406 /* attach the mark to the inode */
407 static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
408 					  struct inode *inode,
409 					  int allow_dups)
410 {
411 	return fsnotify_add_mark(mark, inode, NULL, allow_dups);
412 }
413 static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
414 						 struct inode *inode,
415 						 int allow_dups)
416 {
417 	return fsnotify_add_mark_locked(mark, inode, NULL, allow_dups);
418 }
419 /* given a group and a mark, flag mark to be freed when all references are dropped */
420 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
421 				  struct fsnotify_group *group);
422 /* detach mark from inode / mount list, group list, drop inode reference */
423 extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
424 /* free mark */
425 extern void fsnotify_free_mark(struct fsnotify_mark *mark);
426 /* run all the marks in a group, and clear all of the marks attached to given object type */
427 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
428 /* run all the marks in a group, and clear all of the vfsmount marks */
429 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
430 {
431 	fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL);
432 }
433 /* run all the marks in a group, and clear all of the inode marks */
434 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
435 {
436 	fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE_FL);
437 }
438 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
439 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
440 extern void fsnotify_unmount_inodes(struct super_block *sb);
441 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
442 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
443 
444 /* put here because inotify does some weird stuff when destroying watches */
445 extern void fsnotify_init_event(struct fsnotify_event *event,
446 				struct inode *to_tell, u32 mask);
447 
448 #else
449 
450 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
451 			   const unsigned char *name, u32 cookie)
452 {
453 	return 0;
454 }
455 
456 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
457 {
458 	return 0;
459 }
460 
461 static inline void __fsnotify_inode_delete(struct inode *inode)
462 {}
463 
464 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
465 {}
466 
467 static inline void fsnotify_update_flags(struct dentry *dentry)
468 {}
469 
470 static inline u32 fsnotify_get_cookie(void)
471 {
472 	return 0;
473 }
474 
475 static inline void fsnotify_unmount_inodes(struct super_block *sb)
476 {}
477 
478 #endif	/* CONFIG_FSNOTIFY */
479 
480 #endif	/* __KERNEL __ */
481 
482 #endif	/* __LINUX_FSNOTIFY_BACKEND_H */
483