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#define FS_OPEN_EXEC 0x00001000 /* File was opened for exec */
42
43#define FS_UNMOUNT 0x00002000 /* inode on umount fs */
44#define FS_Q_OVERFLOW 0x00004000 /* Event queued overflowed */
45#define FS_IN_IGNORED 0x00008000 /* last inotify event here */
46
47#define FS_OPEN_PERM 0x00010000 /* open event in an permission hook */
48#define FS_ACCESS_PERM 0x00020000 /* access event in a permissions hook */
49#define FS_OPEN_EXEC_PERM 0x00040000 /* open/exec event in a permission hook */
50
51#define FS_EXCL_UNLINK 0x04000000 /* do not send events if object is unlinked */
52#define FS_ISDIR 0x40000000 /* event occurred against dir */
53#define FS_IN_ONESHOT 0x80000000 /* only send event once */
54
55#define FS_DN_RENAME 0x10000000 /* file renamed */
56#define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
57
58/* This inode cares about things that happen to its children. Always set for
59 * dnotify and inotify. */
60#define FS_EVENT_ON_CHILD 0x08000000
61
62#define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
63
64/*
65 * Directory entry modification events - reported only to directory
66 * where entry is modified and not to a watching parent.
67 * The watching parent may get an FS_ATTRIB|FS_EVENT_ON_CHILD event
68 * when a directory entry inside a child subdir changes.
69 */
70#define ALL_FSNOTIFY_DIRENT_EVENTS (FS_CREATE | FS_DELETE | FS_MOVE)
71
72#define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM | \
73 FS_OPEN_EXEC_PERM)
74
75/*
76 * This is a list of all events that may get sent to a parent based on fs event
77 * happening to inodes inside that directory.
78 */
79#define FS_EVENTS_POSS_ON_CHILD (ALL_FSNOTIFY_PERM_EVENTS | \
80 FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
81 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | \
82 FS_OPEN | FS_OPEN_EXEC)
83
84/* Events that can be reported to backends */
85#define ALL_FSNOTIFY_EVENTS (ALL_FSNOTIFY_DIRENT_EVENTS | \
86 FS_EVENTS_POSS_ON_CHILD | \
87 FS_DELETE_SELF | FS_MOVE_SELF | FS_DN_RENAME | \
88 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED)
89
90/* Extra flags that may be reported with event or control handling of events */
91#define ALL_FSNOTIFY_FLAGS (FS_EXCL_UNLINK | FS_ISDIR | FS_IN_ONESHOT | \
92 FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
93
94#define ALL_FSNOTIFY_BITS (ALL_FSNOTIFY_EVENTS | ALL_FSNOTIFY_FLAGS)
95
96struct fsnotify_group;
97struct fsnotify_event;
98struct fsnotify_mark;
99struct fsnotify_event_private_data;
100struct fsnotify_fname;
101struct fsnotify_iter_info;
102
103struct mem_cgroup;
104
105/*
106 * Each group much define these ops. The fsnotify infrastructure will call
107 * these operations for each relevant group.
108 *
109 * handle_event - main call for a group to handle an fs event
110 * free_group_priv - called when a group refcnt hits 0 to clean up the private union
111 * freeing_mark - called when a mark is being destroyed for some reason. The group
112 * MUST be holding a reference on each mark and that reference must be
113 * dropped in this function. inotify uses this function to send
114 * userspace messages that marks have been removed.
115 */
116struct fsnotify_ops {
117 int (*handle_event)(struct fsnotify_group *group,
118 struct inode *inode,
119 u32 mask, const void *data, int data_type,
120 const unsigned char *file_name, u32 cookie,
121 struct fsnotify_iter_info *iter_info);
122 void (*free_group_priv)(struct fsnotify_group *group);
123 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
124 void (*free_event)(struct fsnotify_event *event);
125 /* called on final put+free to free memory */
126 void (*free_mark)(struct fsnotify_mark *mark);
127};
128
129/*
130 * all of the information about the original object we want to now send to
131 * a group. If you want to carry more info from the accessing task to the
132 * listener this structure is where you need to be adding fields.
133 */
134struct fsnotify_event {
135 struct list_head list;
136 /* inode may ONLY be dereferenced during handle_event(). */
137 struct inode *inode; /* either the inode the event happened to or its parent */
138};
139
140/*
141 * A group is a "thing" that wants to receive notification about filesystem
142 * events. The mask holds the subset of event types this group cares about.
143 * refcnt on a group is up to the implementor and at any moment if it goes 0
144 * everything will be cleaned up.
145 */
146struct fsnotify_group {
147 const struct fsnotify_ops *ops; /* how this group handles things */
148
149 /*
150 * How the refcnt is used is up to each group. When the refcnt hits 0
151 * fsnotify will clean up all of the resources associated with this group.
152 * As an example, the dnotify group will always have a refcnt=1 and that
153 * will never change. Inotify, on the other hand, has a group per
154 * inotify_init() and the refcnt will hit 0 only when that fd has been
155 * closed.
156 */
157 refcount_t refcnt; /* things with interest in this group */
158
159 /* needed to send notification to userspace */
160 spinlock_t notification_lock; /* protect the notification_list */
161 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
162 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
163 unsigned int q_len; /* events on the queue */
164 unsigned int max_events; /* maximum events allowed on the list */
165 /*
166 * Valid fsnotify group priorities. Events are send in order from highest
167 * priority to lowest priority. We default to the lowest priority.
168 */
169 #define FS_PRIO_0 0 /* normal notifiers, no permissions */
170 #define FS_PRIO_1 1 /* fanotify content based access control */
171 #define FS_PRIO_2 2 /* fanotify pre-content access */
172 unsigned int priority;
173 bool shutdown; /* group is being shut down, don't queue more events */
174
175 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
176 struct mutex mark_mutex; /* protect marks_list */
177 atomic_t num_marks; /* 1 for each mark and 1 for not being
178 * past the point of no return when freeing
179 * a group */
180 atomic_t user_waits; /* Number of tasks waiting for user
181 * response */
182 struct list_head marks_list; /* all inode marks for this group */
183
184 struct fasync_struct *fsn_fa; /* async notification */
185
186 struct fsnotify_event *overflow_event; /* Event we queue when the
187 * notification list is too
188 * full */
189
190 struct mem_cgroup *memcg; /* memcg to charge allocations */
191
192 /* groups can define private fields here or use the void *private */
193 union {
194 void *private;
195#ifdef CONFIG_INOTIFY_USER
196 struct inotify_group_private_data {
197 spinlock_t idr_lock;
198 struct idr idr;
199 struct ucounts *ucounts;
200 } inotify_data;
201#endif
202#ifdef CONFIG_FANOTIFY
203 struct fanotify_group_private_data {
204 /* allows a group to block waiting for a userspace response */
205 struct list_head access_list;
206 wait_queue_head_t access_waitq;
207 int flags; /* flags from fanotify_init() */
208 int f_flags; /* event_f_flags from fanotify_init() */
209 unsigned int max_marks;
210 struct user_struct *user;
211 } fanotify_data;
212#endif /* CONFIG_FANOTIFY */
213 };
214};
215
216/* when calling fsnotify tell it if the data is a path or inode */
217#define FSNOTIFY_EVENT_NONE 0
218#define FSNOTIFY_EVENT_PATH 1
219#define FSNOTIFY_EVENT_INODE 2
220
221enum fsnotify_obj_type {
222 FSNOTIFY_OBJ_TYPE_INODE,
223 FSNOTIFY_OBJ_TYPE_VFSMOUNT,
224 FSNOTIFY_OBJ_TYPE_SB,
225 FSNOTIFY_OBJ_TYPE_COUNT,
226 FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
227};
228
229#define FSNOTIFY_OBJ_TYPE_INODE_FL (1U << FSNOTIFY_OBJ_TYPE_INODE)
230#define FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL (1U << FSNOTIFY_OBJ_TYPE_VFSMOUNT)
231#define FSNOTIFY_OBJ_TYPE_SB_FL (1U << FSNOTIFY_OBJ_TYPE_SB)
232#define FSNOTIFY_OBJ_ALL_TYPES_MASK ((1U << FSNOTIFY_OBJ_TYPE_COUNT) - 1)
233
234static inline bool fsnotify_valid_obj_type(unsigned int type)
235{
236 return (type < FSNOTIFY_OBJ_TYPE_COUNT);
237}
238
239struct fsnotify_iter_info {
240 struct fsnotify_mark *marks[FSNOTIFY_OBJ_TYPE_COUNT];
241 unsigned int report_mask;
242 int srcu_idx;
243};
244
245static inline bool fsnotify_iter_should_report_type(
246 struct fsnotify_iter_info *iter_info, int type)
247{
248 return (iter_info->report_mask & (1U << type));
249}
250
251static inline void fsnotify_iter_set_report_type(
252 struct fsnotify_iter_info *iter_info, int type)
253{
254 iter_info->report_mask |= (1U << type);
255}
256
257static inline void fsnotify_iter_set_report_type_mark(
258 struct fsnotify_iter_info *iter_info, int type,
259 struct fsnotify_mark *mark)
260{
261 iter_info->marks[type] = mark;
262 iter_info->report_mask |= (1U << type);
263}
264
265#define FSNOTIFY_ITER_FUNCS(name, NAME) \
266static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \
267 struct fsnotify_iter_info *iter_info) \
268{ \
269 return (iter_info->report_mask & FSNOTIFY_OBJ_TYPE_##NAME##_FL) ? \
270 iter_info->marks[FSNOTIFY_OBJ_TYPE_##NAME] : NULL; \
271}
272
273FSNOTIFY_ITER_FUNCS(inode, INODE)
274FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT)
275FSNOTIFY_ITER_FUNCS(sb, SB)
276
277#define fsnotify_foreach_obj_type(type) \
278 for (type = 0; type < FSNOTIFY_OBJ_TYPE_COUNT; type++)
279
280/*
281 * fsnotify_connp_t is what we embed in objects which connector can be attached
282 * to. fsnotify_connp_t * is how we refer from connector back to object.
283 */
284struct fsnotify_mark_connector;
285typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t;
286
287/*
288 * Inode/vfsmount/sb point to this structure which tracks all marks attached to
289 * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this
290 * structure. We destroy this structure when there are no more marks attached
291 * to it. The structure is protected by fsnotify_mark_srcu.
292 */
293struct fsnotify_mark_connector {
294 spinlock_t lock;
295 unsigned int type; /* Type of object [lock] */
296 __kernel_fsid_t fsid; /* fsid of filesystem containing object */
297 union {
298 /* Object pointer [lock] */
299 fsnotify_connp_t *obj;
300 /* Used listing heads to free after srcu period expires */
301 struct fsnotify_mark_connector *destroy_next;
302 };
303 struct hlist_head list;
304};
305
306/*
307 * A mark is simply an object attached to an in core inode which allows an
308 * fsnotify listener to indicate they are either no longer interested in events
309 * of a type matching mask or only interested in those events.
310 *
311 * These are flushed when an inode is evicted from core and may be flushed
312 * when the inode is modified (as seen by fsnotify_access). Some fsnotify
313 * users (such as dnotify) will flush these when the open fd is closed and not
314 * at inode eviction or modification.
315 *
316 * Text in brackets is showing the lock(s) protecting modifications of a
317 * particular entry. obj_lock means either inode->i_lock or
318 * mnt->mnt_root->d_lock depending on the mark type.
319 */
320struct fsnotify_mark {
321 /* Mask this mark is for [mark->lock, group->mark_mutex] */
322 __u32 mask;
323 /* We hold one for presence in g_list. Also one ref for each 'thing'
324 * in kernel that found and may be using this mark. */
325 refcount_t refcnt;
326 /* Group this mark is for. Set on mark creation, stable until last ref
327 * is dropped */
328 struct fsnotify_group *group;
329 /* List of marks by group->marks_list. Also reused for queueing
330 * mark into destroy_list when it's waiting for the end of SRCU period
331 * before it can be freed. [group->mark_mutex] */
332 struct list_head g_list;
333 /* Protects inode / mnt pointers, flags, masks */
334 spinlock_t lock;
335 /* List of marks for inode / vfsmount [connector->lock, mark ref] */
336 struct hlist_node obj_list;
337 /* Head of list of marks for an object [mark ref] */
338 struct fsnotify_mark_connector *connector;
339 /* Events types to ignore [mark->lock, group->mark_mutex] */
340 __u32 ignored_mask;
341#define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x01
342#define FSNOTIFY_MARK_FLAG_ALIVE 0x02
343#define FSNOTIFY_MARK_FLAG_ATTACHED 0x04
344 unsigned int flags; /* flags [mark->lock] */
345};
346
347#ifdef CONFIG_FSNOTIFY
348
349/* called from the vfs helpers */
350
351/* main fsnotify call to send events */
352extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
353 const unsigned char *name, u32 cookie);
354extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
355extern void __fsnotify_inode_delete(struct inode *inode);
356extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
357extern void fsnotify_sb_delete(struct super_block *sb);
358extern u32 fsnotify_get_cookie(void);
359
360static inline int fsnotify_inode_watches_children(struct inode *inode)
361{
362 /* FS_EVENT_ON_CHILD is set if the inode may care */
363 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
364 return 0;
365 /* this inode might care about child events, does it care about the
366 * specific set of events that can happen on a child? */
367 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
368}
369
370/*
371 * Update the dentry with a flag indicating the interest of its parent to receive
372 * filesystem events when those events happens to this dentry->d_inode.
373 */
374static inline void fsnotify_update_flags(struct dentry *dentry)
375{
376 assert_spin_locked(&dentry->d_lock);
377
378 /*
379 * Serialisation of setting PARENT_WATCHED on the dentries is provided
380 * by d_lock. If inotify_inode_watched changes after we have taken
381 * d_lock, the following __fsnotify_update_child_dentry_flags call will
382 * find our entry, so it will spin until we complete here, and update
383 * us with the new state.
384 */
385 if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
386 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
387 else
388 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
389}
390
391/* called from fsnotify listeners, such as fanotify or dnotify */
392
393/* create a new group */
394extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
395/* get reference to a group */
396extern void fsnotify_get_group(struct fsnotify_group *group);
397/* drop reference on a group from fsnotify_alloc_group */
398extern void fsnotify_put_group(struct fsnotify_group *group);
399/* group destruction begins, stop queuing new events */
400extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
401/* destroy group */
402extern void fsnotify_destroy_group(struct fsnotify_group *group);
403/* fasync handler function */
404extern int fsnotify_fasync(int fd, struct file *file, int on);
405/* Free event from memory */
406extern void fsnotify_destroy_event(struct fsnotify_group *group,
407 struct fsnotify_event *event);
408/* attach the event to the group notification queue */
409extern int fsnotify_add_event(struct fsnotify_group *group,
410 struct fsnotify_event *event,
411 int (*merge)(struct list_head *,
412 struct fsnotify_event *));
413/* Queue overflow event to a notification group */
414static inline void fsnotify_queue_overflow(struct fsnotify_group *group)
415{
416 fsnotify_add_event(group, group->overflow_event, NULL);
417}
418
419/* true if the group notification queue is empty */
420extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
421/* return, but do not dequeue the first event on the notification queue */
422extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
423/* return AND dequeue the first event on the notification queue */
424extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
425/* Remove event queued in the notification list */
426extern void fsnotify_remove_queued_event(struct fsnotify_group *group,
427 struct fsnotify_event *event);
428
429/* functions used to manipulate the marks attached to inodes */
430
431/* Get mask of events for a list of marks */
432extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn);
433/* Calculate mask of events for a list of marks */
434extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
435extern void fsnotify_init_mark(struct fsnotify_mark *mark,
436 struct fsnotify_group *group);
437/* Find mark belonging to given group in the list of marks */
438extern struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
439 struct fsnotify_group *group);
440/* Get cached fsid of filesystem containing object */
441extern int fsnotify_get_conn_fsid(const struct fsnotify_mark_connector *conn,
442 __kernel_fsid_t *fsid);
443/* attach the mark to the object */
444extern int fsnotify_add_mark(struct fsnotify_mark *mark,
445 fsnotify_connp_t *connp, unsigned int type,
446 int allow_dups, __kernel_fsid_t *fsid);
447extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
448 fsnotify_connp_t *connp,
449 unsigned int type, int allow_dups,
450 __kernel_fsid_t *fsid);
451
452/* attach the mark to the inode */
453static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
454 struct inode *inode,
455 int allow_dups)
456{
457 return fsnotify_add_mark(mark, &inode->i_fsnotify_marks,
458 FSNOTIFY_OBJ_TYPE_INODE, allow_dups, NULL);
459}
460static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
461 struct inode *inode,
462 int allow_dups)
463{
464 return fsnotify_add_mark_locked(mark, &inode->i_fsnotify_marks,
465 FSNOTIFY_OBJ_TYPE_INODE, allow_dups,
466 NULL);
467}
468
469/* given a group and a mark, flag mark to be freed when all references are dropped */
470extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
471 struct fsnotify_group *group);
472/* detach mark from inode / mount list, group list, drop inode reference */
473extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
474/* free mark */
475extern void fsnotify_free_mark(struct fsnotify_mark *mark);
476/* run all the marks in a group, and clear all of the marks attached to given object type */
477extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
478/* run all the marks in a group, and clear all of the vfsmount marks */
479static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
480{
481 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL);
482}
483/* run all the marks in a group, and clear all of the inode marks */
484static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
485{
486 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE_FL);
487}
488/* run all the marks in a group, and clear all of the sn marks */
489static inline void fsnotify_clear_sb_marks_by_group(struct fsnotify_group *group)
490{
491 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_SB_FL);
492}
493extern void fsnotify_get_mark(struct fsnotify_mark *mark);
494extern void fsnotify_put_mark(struct fsnotify_mark *mark);
495extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
496extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
497
498static inline void fsnotify_init_event(struct fsnotify_event *event,
499 struct inode *inode)
500{
501 INIT_LIST_HEAD(&event->list);
502 event->inode = inode;
503}
504
505#else
506
507static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
508 const unsigned char *name, u32 cookie)
509{
510 return 0;
511}
512
513static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
514{
515 return 0;
516}
517
518static inline void __fsnotify_inode_delete(struct inode *inode)
519{}
520
521static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
522{}
523
524static inline void fsnotify_sb_delete(struct super_block *sb)
525{}
526
527static inline void fsnotify_update_flags(struct dentry *dentry)
528{}
529
530static inline u32 fsnotify_get_cookie(void)
531{
532 return 0;
533}
534
535static inline void fsnotify_unmount_inodes(struct super_block *sb)
536{}
537
538#endif /* CONFIG_FSNOTIFY */
539
540#endif /* __KERNEL __ */
541
542#endif /* __LINUX_FSNOTIFY_BACKEND_H */
543