Loading...
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/*
53 * Set on inode mark that cares about things that happen to its children.
54 * Always set for dnotify and inotify.
55 * Set on inode/sb/mount marks that care about parent/name info.
56 */
57#define FS_EVENT_ON_CHILD 0x08000000
58
59#define FS_DN_RENAME 0x10000000 /* file renamed */
60#define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
61#define FS_ISDIR 0x40000000 /* event occurred against dir */
62#define FS_IN_ONESHOT 0x80000000 /* only send event once */
63
64#define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
65
66/*
67 * Directory entry modification events - reported only to directory
68 * where entry is modified and not to a watching parent.
69 * The watching parent may get an FS_ATTRIB|FS_EVENT_ON_CHILD event
70 * when a directory entry inside a child subdir changes.
71 */
72#define ALL_FSNOTIFY_DIRENT_EVENTS (FS_CREATE | FS_DELETE | FS_MOVE)
73
74#define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM | \
75 FS_OPEN_EXEC_PERM)
76
77/*
78 * This is a list of all events that may get sent to a parent that is watching
79 * with flag FS_EVENT_ON_CHILD based on fs event on a child of that directory.
80 */
81#define FS_EVENTS_POSS_ON_CHILD (ALL_FSNOTIFY_PERM_EVENTS | \
82 FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
83 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | \
84 FS_OPEN | FS_OPEN_EXEC)
85
86/*
87 * This is a list of all events that may get sent with the parent inode as the
88 * @to_tell argument of fsnotify().
89 * It may include events that can be sent to an inode/sb/mount mark, but cannot
90 * be sent to a parent watching children.
91 */
92#define FS_EVENTS_POSS_TO_PARENT (FS_EVENTS_POSS_ON_CHILD)
93
94/* Events that can be reported to backends */
95#define ALL_FSNOTIFY_EVENTS (ALL_FSNOTIFY_DIRENT_EVENTS | \
96 FS_EVENTS_POSS_ON_CHILD | \
97 FS_DELETE_SELF | FS_MOVE_SELF | FS_DN_RENAME | \
98 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED)
99
100/* Extra flags that may be reported with event or control handling of events */
101#define ALL_FSNOTIFY_FLAGS (FS_EXCL_UNLINK | FS_ISDIR | FS_IN_ONESHOT | \
102 FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
103
104#define ALL_FSNOTIFY_BITS (ALL_FSNOTIFY_EVENTS | ALL_FSNOTIFY_FLAGS)
105
106struct fsnotify_group;
107struct fsnotify_event;
108struct fsnotify_mark;
109struct fsnotify_event_private_data;
110struct fsnotify_fname;
111struct fsnotify_iter_info;
112
113struct mem_cgroup;
114
115/*
116 * Each group much define these ops. The fsnotify infrastructure will call
117 * these operations for each relevant group.
118 *
119 * handle_event - main call for a group to handle an fs event
120 * @group: group to notify
121 * @mask: event type and flags
122 * @data: object that event happened on
123 * @data_type: type of object for fanotify_data_XXX() accessors
124 * @dir: optional directory associated with event -
125 * if @file_name is not NULL, this is the directory that
126 * @file_name is relative to
127 * @file_name: optional file name associated with event
128 * @cookie: inotify rename cookie
129 * @iter_info: array of marks from this group that are interested in the event
130 *
131 * handle_inode_event - simple variant of handle_event() for groups that only
132 * have inode marks and don't have ignore mask
133 * @mark: mark to notify
134 * @mask: event type and flags
135 * @inode: inode that event happened on
136 * @dir: optional directory associated with event -
137 * if @file_name is not NULL, this is the directory that
138 * @file_name is relative to.
139 * @file_name: optional file name associated with event
140 *
141 * free_group_priv - called when a group refcnt hits 0 to clean up the private union
142 * freeing_mark - called when a mark is being destroyed for some reason. The group
143 * MUST be holding a reference on each mark and that reference must be
144 * dropped in this function. inotify uses this function to send
145 * userspace messages that marks have been removed.
146 */
147struct fsnotify_ops {
148 int (*handle_event)(struct fsnotify_group *group, u32 mask,
149 const void *data, int data_type, struct inode *dir,
150 const struct qstr *file_name, u32 cookie,
151 struct fsnotify_iter_info *iter_info);
152 int (*handle_inode_event)(struct fsnotify_mark *mark, u32 mask,
153 struct inode *inode, struct inode *dir,
154 const struct qstr *file_name);
155 void (*free_group_priv)(struct fsnotify_group *group);
156 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
157 void (*free_event)(struct fsnotify_event *event);
158 /* called on final put+free to free memory */
159 void (*free_mark)(struct fsnotify_mark *mark);
160};
161
162/*
163 * all of the information about the original object we want to now send to
164 * a group. If you want to carry more info from the accessing task to the
165 * listener this structure is where you need to be adding fields.
166 */
167struct fsnotify_event {
168 struct list_head list;
169 unsigned long objectid; /* identifier for queue merges */
170};
171
172/*
173 * A group is a "thing" that wants to receive notification about filesystem
174 * events. The mask holds the subset of event types this group cares about.
175 * refcnt on a group is up to the implementor and at any moment if it goes 0
176 * everything will be cleaned up.
177 */
178struct fsnotify_group {
179 const struct fsnotify_ops *ops; /* how this group handles things */
180
181 /*
182 * How the refcnt is used is up to each group. When the refcnt hits 0
183 * fsnotify will clean up all of the resources associated with this group.
184 * As an example, the dnotify group will always have a refcnt=1 and that
185 * will never change. Inotify, on the other hand, has a group per
186 * inotify_init() and the refcnt will hit 0 only when that fd has been
187 * closed.
188 */
189 refcount_t refcnt; /* things with interest in this group */
190
191 /* needed to send notification to userspace */
192 spinlock_t notification_lock; /* protect the notification_list */
193 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
194 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
195 unsigned int q_len; /* events on the queue */
196 unsigned int max_events; /* maximum events allowed on the list */
197 /*
198 * Valid fsnotify group priorities. Events are send in order from highest
199 * priority to lowest priority. We default to the lowest priority.
200 */
201 #define FS_PRIO_0 0 /* normal notifiers, no permissions */
202 #define FS_PRIO_1 1 /* fanotify content based access control */
203 #define FS_PRIO_2 2 /* fanotify pre-content access */
204 unsigned int priority;
205 bool shutdown; /* group is being shut down, don't queue more events */
206
207 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
208 struct mutex mark_mutex; /* protect marks_list */
209 atomic_t num_marks; /* 1 for each mark and 1 for not being
210 * past the point of no return when freeing
211 * a group */
212 atomic_t user_waits; /* Number of tasks waiting for user
213 * response */
214 struct list_head marks_list; /* all inode marks for this group */
215
216 struct fasync_struct *fsn_fa; /* async notification */
217
218 struct fsnotify_event *overflow_event; /* Event we queue when the
219 * notification list is too
220 * full */
221
222 struct mem_cgroup *memcg; /* memcg to charge allocations */
223
224 /* groups can define private fields here or use the void *private */
225 union {
226 void *private;
227#ifdef CONFIG_INOTIFY_USER
228 struct inotify_group_private_data {
229 spinlock_t idr_lock;
230 struct idr idr;
231 struct ucounts *ucounts;
232 } inotify_data;
233#endif
234#ifdef CONFIG_FANOTIFY
235 struct fanotify_group_private_data {
236 /* allows a group to block waiting for a userspace response */
237 struct list_head access_list;
238 wait_queue_head_t access_waitq;
239 int flags; /* flags from fanotify_init() */
240 int f_flags; /* event_f_flags from fanotify_init() */
241 unsigned int max_marks;
242 struct user_struct *user;
243 } fanotify_data;
244#endif /* CONFIG_FANOTIFY */
245 };
246};
247
248/* When calling fsnotify tell it if the data is a path or inode */
249enum fsnotify_data_type {
250 FSNOTIFY_EVENT_NONE,
251 FSNOTIFY_EVENT_PATH,
252 FSNOTIFY_EVENT_INODE,
253};
254
255static inline struct inode *fsnotify_data_inode(const void *data, int data_type)
256{
257 switch (data_type) {
258 case FSNOTIFY_EVENT_INODE:
259 return (struct inode *)data;
260 case FSNOTIFY_EVENT_PATH:
261 return d_inode(((const struct path *)data)->dentry);
262 default:
263 return NULL;
264 }
265}
266
267static inline const struct path *fsnotify_data_path(const void *data,
268 int data_type)
269{
270 switch (data_type) {
271 case FSNOTIFY_EVENT_PATH:
272 return data;
273 default:
274 return NULL;
275 }
276}
277
278enum fsnotify_obj_type {
279 FSNOTIFY_OBJ_TYPE_INODE,
280 FSNOTIFY_OBJ_TYPE_CHILD,
281 FSNOTIFY_OBJ_TYPE_VFSMOUNT,
282 FSNOTIFY_OBJ_TYPE_SB,
283 FSNOTIFY_OBJ_TYPE_COUNT,
284 FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
285};
286
287#define FSNOTIFY_OBJ_TYPE_INODE_FL (1U << FSNOTIFY_OBJ_TYPE_INODE)
288#define FSNOTIFY_OBJ_TYPE_CHILD_FL (1U << FSNOTIFY_OBJ_TYPE_CHILD)
289#define FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL (1U << FSNOTIFY_OBJ_TYPE_VFSMOUNT)
290#define FSNOTIFY_OBJ_TYPE_SB_FL (1U << FSNOTIFY_OBJ_TYPE_SB)
291#define FSNOTIFY_OBJ_ALL_TYPES_MASK ((1U << FSNOTIFY_OBJ_TYPE_COUNT) - 1)
292
293static inline bool fsnotify_valid_obj_type(unsigned int type)
294{
295 return (type < FSNOTIFY_OBJ_TYPE_COUNT);
296}
297
298struct fsnotify_iter_info {
299 struct fsnotify_mark *marks[FSNOTIFY_OBJ_TYPE_COUNT];
300 unsigned int report_mask;
301 int srcu_idx;
302};
303
304static inline bool fsnotify_iter_should_report_type(
305 struct fsnotify_iter_info *iter_info, int type)
306{
307 return (iter_info->report_mask & (1U << type));
308}
309
310static inline void fsnotify_iter_set_report_type(
311 struct fsnotify_iter_info *iter_info, int type)
312{
313 iter_info->report_mask |= (1U << type);
314}
315
316static inline void fsnotify_iter_set_report_type_mark(
317 struct fsnotify_iter_info *iter_info, int type,
318 struct fsnotify_mark *mark)
319{
320 iter_info->marks[type] = mark;
321 iter_info->report_mask |= (1U << type);
322}
323
324#define FSNOTIFY_ITER_FUNCS(name, NAME) \
325static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \
326 struct fsnotify_iter_info *iter_info) \
327{ \
328 return (iter_info->report_mask & FSNOTIFY_OBJ_TYPE_##NAME##_FL) ? \
329 iter_info->marks[FSNOTIFY_OBJ_TYPE_##NAME] : NULL; \
330}
331
332FSNOTIFY_ITER_FUNCS(inode, INODE)
333FSNOTIFY_ITER_FUNCS(child, CHILD)
334FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT)
335FSNOTIFY_ITER_FUNCS(sb, SB)
336
337#define fsnotify_foreach_obj_type(type) \
338 for (type = 0; type < FSNOTIFY_OBJ_TYPE_COUNT; type++)
339
340/*
341 * fsnotify_connp_t is what we embed in objects which connector can be attached
342 * to. fsnotify_connp_t * is how we refer from connector back to object.
343 */
344struct fsnotify_mark_connector;
345typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t;
346
347/*
348 * Inode/vfsmount/sb point to this structure which tracks all marks attached to
349 * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this
350 * structure. We destroy this structure when there are no more marks attached
351 * to it. The structure is protected by fsnotify_mark_srcu.
352 */
353struct fsnotify_mark_connector {
354 spinlock_t lock;
355 unsigned short type; /* Type of object [lock] */
356#define FSNOTIFY_CONN_FLAG_HAS_FSID 0x01
357 unsigned short flags; /* flags [lock] */
358 __kernel_fsid_t fsid; /* fsid of filesystem containing object */
359 union {
360 /* Object pointer [lock] */
361 fsnotify_connp_t *obj;
362 /* Used listing heads to free after srcu period expires */
363 struct fsnotify_mark_connector *destroy_next;
364 };
365 struct hlist_head list;
366};
367
368/*
369 * A mark is simply an object attached to an in core inode which allows an
370 * fsnotify listener to indicate they are either no longer interested in events
371 * of a type matching mask or only interested in those events.
372 *
373 * These are flushed when an inode is evicted from core and may be flushed
374 * when the inode is modified (as seen by fsnotify_access). Some fsnotify
375 * users (such as dnotify) will flush these when the open fd is closed and not
376 * at inode eviction or modification.
377 *
378 * Text in brackets is showing the lock(s) protecting modifications of a
379 * particular entry. obj_lock means either inode->i_lock or
380 * mnt->mnt_root->d_lock depending on the mark type.
381 */
382struct fsnotify_mark {
383 /* Mask this mark is for [mark->lock, group->mark_mutex] */
384 __u32 mask;
385 /* We hold one for presence in g_list. Also one ref for each 'thing'
386 * in kernel that found and may be using this mark. */
387 refcount_t refcnt;
388 /* Group this mark is for. Set on mark creation, stable until last ref
389 * is dropped */
390 struct fsnotify_group *group;
391 /* List of marks by group->marks_list. Also reused for queueing
392 * mark into destroy_list when it's waiting for the end of SRCU period
393 * before it can be freed. [group->mark_mutex] */
394 struct list_head g_list;
395 /* Protects inode / mnt pointers, flags, masks */
396 spinlock_t lock;
397 /* List of marks for inode / vfsmount [connector->lock, mark ref] */
398 struct hlist_node obj_list;
399 /* Head of list of marks for an object [mark ref] */
400 struct fsnotify_mark_connector *connector;
401 /* Events types to ignore [mark->lock, group->mark_mutex] */
402 __u32 ignored_mask;
403#define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x01
404#define FSNOTIFY_MARK_FLAG_ALIVE 0x02
405#define FSNOTIFY_MARK_FLAG_ATTACHED 0x04
406 unsigned int flags; /* flags [mark->lock] */
407};
408
409#ifdef CONFIG_FSNOTIFY
410
411/* called from the vfs helpers */
412
413/* main fsnotify call to send events */
414extern int fsnotify(__u32 mask, const void *data, int data_type,
415 struct inode *dir, const struct qstr *name,
416 struct inode *inode, u32 cookie);
417extern int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
418 int data_type);
419extern void __fsnotify_inode_delete(struct inode *inode);
420extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
421extern void fsnotify_sb_delete(struct super_block *sb);
422extern u32 fsnotify_get_cookie(void);
423
424static inline __u32 fsnotify_parent_needed_mask(__u32 mask)
425{
426 /* FS_EVENT_ON_CHILD is set on marks that want parent/name info */
427 if (!(mask & FS_EVENT_ON_CHILD))
428 return 0;
429 /*
430 * This object might be watched by a mark that cares about parent/name
431 * info, does it care about the specific set of events that can be
432 * reported with parent/name info?
433 */
434 return mask & FS_EVENTS_POSS_TO_PARENT;
435}
436
437static inline int fsnotify_inode_watches_children(struct inode *inode)
438{
439 /* FS_EVENT_ON_CHILD is set if the inode may care */
440 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
441 return 0;
442 /* this inode might care about child events, does it care about the
443 * specific set of events that can happen on a child? */
444 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
445}
446
447/*
448 * Update the dentry with a flag indicating the interest of its parent to receive
449 * filesystem events when those events happens to this dentry->d_inode.
450 */
451static inline void fsnotify_update_flags(struct dentry *dentry)
452{
453 assert_spin_locked(&dentry->d_lock);
454
455 /*
456 * Serialisation of setting PARENT_WATCHED on the dentries is provided
457 * by d_lock. If inotify_inode_watched changes after we have taken
458 * d_lock, the following __fsnotify_update_child_dentry_flags call will
459 * find our entry, so it will spin until we complete here, and update
460 * us with the new state.
461 */
462 if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
463 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
464 else
465 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
466}
467
468/* called from fsnotify listeners, such as fanotify or dnotify */
469
470/* create a new group */
471extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
472/* get reference to a group */
473extern void fsnotify_get_group(struct fsnotify_group *group);
474/* drop reference on a group from fsnotify_alloc_group */
475extern void fsnotify_put_group(struct fsnotify_group *group);
476/* group destruction begins, stop queuing new events */
477extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
478/* destroy group */
479extern void fsnotify_destroy_group(struct fsnotify_group *group);
480/* fasync handler function */
481extern int fsnotify_fasync(int fd, struct file *file, int on);
482/* Free event from memory */
483extern void fsnotify_destroy_event(struct fsnotify_group *group,
484 struct fsnotify_event *event);
485/* attach the event to the group notification queue */
486extern int fsnotify_add_event(struct fsnotify_group *group,
487 struct fsnotify_event *event,
488 int (*merge)(struct list_head *,
489 struct fsnotify_event *));
490/* Queue overflow event to a notification group */
491static inline void fsnotify_queue_overflow(struct fsnotify_group *group)
492{
493 fsnotify_add_event(group, group->overflow_event, NULL);
494}
495
496/* true if the group notification queue is empty */
497extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
498/* return, but do not dequeue the first event on the notification queue */
499extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
500/* return AND dequeue the first event on the notification queue */
501extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
502/* Remove event queued in the notification list */
503extern void fsnotify_remove_queued_event(struct fsnotify_group *group,
504 struct fsnotify_event *event);
505
506/* functions used to manipulate the marks attached to inodes */
507
508/* Get mask of events for a list of marks */
509extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn);
510/* Calculate mask of events for a list of marks */
511extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
512extern void fsnotify_init_mark(struct fsnotify_mark *mark,
513 struct fsnotify_group *group);
514/* Find mark belonging to given group in the list of marks */
515extern struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
516 struct fsnotify_group *group);
517/* Get cached fsid of filesystem containing object */
518extern int fsnotify_get_conn_fsid(const struct fsnotify_mark_connector *conn,
519 __kernel_fsid_t *fsid);
520/* attach the mark to the object */
521extern int fsnotify_add_mark(struct fsnotify_mark *mark,
522 fsnotify_connp_t *connp, unsigned int type,
523 int allow_dups, __kernel_fsid_t *fsid);
524extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
525 fsnotify_connp_t *connp,
526 unsigned int type, int allow_dups,
527 __kernel_fsid_t *fsid);
528
529/* attach the mark to the inode */
530static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
531 struct inode *inode,
532 int allow_dups)
533{
534 return fsnotify_add_mark(mark, &inode->i_fsnotify_marks,
535 FSNOTIFY_OBJ_TYPE_INODE, allow_dups, NULL);
536}
537static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
538 struct inode *inode,
539 int allow_dups)
540{
541 return fsnotify_add_mark_locked(mark, &inode->i_fsnotify_marks,
542 FSNOTIFY_OBJ_TYPE_INODE, allow_dups,
543 NULL);
544}
545
546/* given a group and a mark, flag mark to be freed when all references are dropped */
547extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
548 struct fsnotify_group *group);
549/* detach mark from inode / mount list, group list, drop inode reference */
550extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
551/* free mark */
552extern void fsnotify_free_mark(struct fsnotify_mark *mark);
553/* Wait until all marks queued for destruction are destroyed */
554extern void fsnotify_wait_marks_destroyed(void);
555/* run all the marks in a group, and clear all of the marks attached to given object type */
556extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
557/* run all the marks in a group, and clear all of the vfsmount marks */
558static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
559{
560 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL);
561}
562/* run all the marks in a group, and clear all of the inode marks */
563static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
564{
565 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE_FL);
566}
567/* run all the marks in a group, and clear all of the sn marks */
568static inline void fsnotify_clear_sb_marks_by_group(struct fsnotify_group *group)
569{
570 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_SB_FL);
571}
572extern void fsnotify_get_mark(struct fsnotify_mark *mark);
573extern void fsnotify_put_mark(struct fsnotify_mark *mark);
574extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
575extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
576
577static inline void fsnotify_init_event(struct fsnotify_event *event,
578 unsigned long objectid)
579{
580 INIT_LIST_HEAD(&event->list);
581 event->objectid = objectid;
582}
583
584#else
585
586static inline int fsnotify(__u32 mask, const void *data, int data_type,
587 struct inode *dir, const struct qstr *name,
588 struct inode *inode, u32 cookie)
589{
590 return 0;
591}
592
593static inline int __fsnotify_parent(struct dentry *dentry, __u32 mask,
594 const void *data, int data_type)
595{
596 return 0;
597}
598
599static inline void __fsnotify_inode_delete(struct inode *inode)
600{}
601
602static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
603{}
604
605static inline void fsnotify_sb_delete(struct super_block *sb)
606{}
607
608static inline void fsnotify_update_flags(struct dentry *dentry)
609{}
610
611static inline u32 fsnotify_get_cookie(void)
612{
613 return 0;
614}
615
616static inline void fsnotify_unmount_inodes(struct super_block *sb)
617{}
618
619#endif /* CONFIG_FSNOTIFY */
620
621#endif /* __KERNEL __ */
622
623#endif /* __LINUX_FSNOTIFY_BACKEND_H */
1/*
2 * Filesystem access notification for Linux
3 *
4 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
5 */
6
7#ifndef __LINUX_FSNOTIFY_BACKEND_H
8#define __LINUX_FSNOTIFY_BACKEND_H
9
10#ifdef __KERNEL__
11
12#include <linux/idr.h> /* inotify uses this */
13#include <linux/fs.h> /* struct inode */
14#include <linux/list.h>
15#include <linux/path.h> /* struct path */
16#include <linux/spinlock.h>
17#include <linux/types.h>
18
19#include <linux/atomic.h>
20
21/*
22 * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
23 * convert between them. dnotify only needs conversion at watch creation
24 * so no perf loss there. fanotify isn't defined yet, so it can use the
25 * wholes if it needs more events.
26 */
27#define FS_ACCESS 0x00000001 /* File was accessed */
28#define FS_MODIFY 0x00000002 /* File was modified */
29#define FS_ATTRIB 0x00000004 /* Metadata changed */
30#define FS_CLOSE_WRITE 0x00000008 /* Writtable file was closed */
31#define FS_CLOSE_NOWRITE 0x00000010 /* Unwrittable file closed */
32#define FS_OPEN 0x00000020 /* File was opened */
33#define FS_MOVED_FROM 0x00000040 /* File was moved from X */
34#define FS_MOVED_TO 0x00000080 /* File was moved to Y */
35#define FS_CREATE 0x00000100 /* Subfile was created */
36#define FS_DELETE 0x00000200 /* Subfile was deleted */
37#define FS_DELETE_SELF 0x00000400 /* Self was deleted */
38#define FS_MOVE_SELF 0x00000800 /* Self was moved */
39
40#define FS_UNMOUNT 0x00002000 /* inode on umount fs */
41#define FS_Q_OVERFLOW 0x00004000 /* Event queued overflowed */
42#define FS_IN_IGNORED 0x00008000 /* last inotify event here */
43
44#define FS_OPEN_PERM 0x00010000 /* open event in an permission hook */
45#define FS_ACCESS_PERM 0x00020000 /* access event in a permissions hook */
46
47#define FS_EXCL_UNLINK 0x04000000 /* do not send events if object is unlinked */
48#define FS_ISDIR 0x40000000 /* event occurred against dir */
49#define FS_IN_ONESHOT 0x80000000 /* only send event once */
50
51#define FS_DN_RENAME 0x10000000 /* file renamed */
52#define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
53
54/* This inode cares about things that happen to its children. Always set for
55 * dnotify and inotify. */
56#define FS_EVENT_ON_CHILD 0x08000000
57
58/* This is a list of all events that may get sent to a parernt based on fs event
59 * happening to inodes inside that directory */
60#define FS_EVENTS_POSS_ON_CHILD (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
61 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
62 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
63 FS_DELETE)
64
65#define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
66
67#define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM)
68
69#define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
70 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \
71 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \
72 FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \
73 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
74 FS_OPEN_PERM | FS_ACCESS_PERM | FS_EXCL_UNLINK | \
75 FS_ISDIR | FS_IN_ONESHOT | FS_DN_RENAME | \
76 FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
77
78struct fsnotify_group;
79struct fsnotify_event;
80struct fsnotify_mark;
81struct fsnotify_event_private_data;
82
83/*
84 * Each group much define these ops. The fsnotify infrastructure will call
85 * these operations for each relevant group.
86 *
87 * should_send_event - given a group, inode, and mask this function determines
88 * if the group is interested in this event.
89 * handle_event - main call for a group to handle an fs event
90 * free_group_priv - called when a group refcnt hits 0 to clean up the private union
91 * freeing-mark - this means that a mark has been flagged to die when everything
92 * finishes using it. The function is supplied with what must be a
93 * valid group and inode to use to clean up.
94 */
95struct fsnotify_ops {
96 bool (*should_send_event)(struct fsnotify_group *group, struct inode *inode,
97 struct fsnotify_mark *inode_mark,
98 struct fsnotify_mark *vfsmount_mark,
99 __u32 mask, void *data, int data_type);
100 int (*handle_event)(struct fsnotify_group *group,
101 struct fsnotify_mark *inode_mark,
102 struct fsnotify_mark *vfsmount_mark,
103 struct fsnotify_event *event);
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_priv)(struct fsnotify_event_private_data *priv);
107};
108
109/*
110 * A group is a "thing" that wants to receive notification about filesystem
111 * events. The mask holds the subset of event types this group cares about.
112 * refcnt on a group is up to the implementor and at any moment if it goes 0
113 * everything will be cleaned up.
114 */
115struct fsnotify_group {
116 /*
117 * How the refcnt is used is up to each group. When the refcnt hits 0
118 * fsnotify will clean up all of the resources associated with this group.
119 * As an example, the dnotify group will always have a refcnt=1 and that
120 * will never change. Inotify, on the other hand, has a group per
121 * inotify_init() and the refcnt will hit 0 only when that fd has been
122 * closed.
123 */
124 atomic_t refcnt; /* things with interest in this group */
125
126 const struct fsnotify_ops *ops; /* how this group handles things */
127
128 /* needed to send notification to userspace */
129 struct mutex notification_mutex; /* protect the notification_list */
130 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
131 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
132 unsigned int q_len; /* events on the queue */
133 unsigned int max_events; /* maximum events allowed on the list */
134 /*
135 * Valid fsnotify group priorities. Events are send in order from highest
136 * priority to lowest priority. We default to the lowest priority.
137 */
138 #define FS_PRIO_0 0 /* normal notifiers, no permissions */
139 #define FS_PRIO_1 1 /* fanotify content based access control */
140 #define FS_PRIO_2 2 /* fanotify pre-content access */
141 unsigned int priority;
142
143 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
144 spinlock_t mark_lock; /* protect marks_list */
145 atomic_t num_marks; /* 1 for each mark and 1 for not being
146 * past the point of no return when freeing
147 * a group */
148 struct list_head marks_list; /* all inode marks for this group */
149
150 /* groups can define private fields here or use the void *private */
151 union {
152 void *private;
153#ifdef CONFIG_INOTIFY_USER
154 struct inotify_group_private_data {
155 spinlock_t idr_lock;
156 struct idr idr;
157 u32 last_wd;
158 struct fasync_struct *fa; /* async notification */
159 struct user_struct *user;
160 } inotify_data;
161#endif
162#ifdef CONFIG_FANOTIFY
163 struct fanotify_group_private_data {
164#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
165 /* allows a group to block waiting for a userspace response */
166 struct mutex access_mutex;
167 struct list_head access_list;
168 wait_queue_head_t access_waitq;
169 atomic_t bypass_perm;
170#endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
171 int f_flags;
172 unsigned int max_marks;
173 struct user_struct *user;
174 } fanotify_data;
175#endif /* CONFIG_FANOTIFY */
176 };
177};
178
179/*
180 * A single event can be queued in multiple group->notification_lists.
181 *
182 * each group->notification_list will point to an event_holder which in turns points
183 * to the actual event that needs to be sent to userspace.
184 *
185 * Seemed cheaper to create a refcnt'd event and a small holder for every group
186 * than create a different event for every group
187 *
188 */
189struct fsnotify_event_holder {
190 struct fsnotify_event *event;
191 struct list_head event_list;
192};
193
194/*
195 * Inotify needs to tack data onto an event. This struct lets us later find the
196 * correct private data of the correct group.
197 */
198struct fsnotify_event_private_data {
199 struct fsnotify_group *group;
200 struct list_head event_list;
201};
202
203/*
204 * all of the information about the original object we want to now send to
205 * a group. If you want to carry more info from the accessing task to the
206 * listener this structure is where you need to be adding fields.
207 */
208struct fsnotify_event {
209 /*
210 * If we create an event we are also likely going to need a holder
211 * to link to a group. So embed one holder in the event. Means only
212 * one allocation for the common case where we only have one group
213 */
214 struct fsnotify_event_holder holder;
215 spinlock_t lock; /* protection for the associated event_holder and private_list */
216 /* to_tell may ONLY be dereferenced during handle_event(). */
217 struct inode *to_tell; /* either the inode the event happened to or its parent */
218 /*
219 * depending on the event type we should have either a path or inode
220 * We hold a reference on path, but NOT on inode. Since we have the ref on
221 * the path, it may be dereferenced at any point during this object's
222 * lifetime. That reference is dropped when this object's refcnt hits
223 * 0. If this event contains an inode instead of a path, the inode may
224 * ONLY be used during handle_event().
225 */
226 union {
227 struct path path;
228 struct inode *inode;
229 };
230/* when calling fsnotify tell it if the data is a path or inode */
231#define FSNOTIFY_EVENT_NONE 0
232#define FSNOTIFY_EVENT_PATH 1
233#define FSNOTIFY_EVENT_INODE 2
234 int data_type; /* which of the above union we have */
235 atomic_t refcnt; /* how many groups still are using/need to send this event */
236 __u32 mask; /* the type of access, bitwise OR for FS_* event types */
237
238 u32 sync_cookie; /* used to corrolate events, namely inotify mv events */
239 const unsigned char *file_name;
240 size_t name_len;
241 struct pid *tgid;
242
243#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
244 __u32 response; /* userspace answer to question */
245#endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
246
247 struct list_head private_data_list; /* groups can store private data here */
248};
249
250/*
251 * Inode specific fields in an fsnotify_mark
252 */
253struct fsnotify_inode_mark {
254 struct inode *inode; /* inode this mark is associated with */
255 struct hlist_node i_list; /* list of marks by inode->i_fsnotify_marks */
256 struct list_head free_i_list; /* tmp list used when freeing this mark */
257};
258
259/*
260 * Mount point specific fields in an fsnotify_mark
261 */
262struct fsnotify_vfsmount_mark {
263 struct vfsmount *mnt; /* vfsmount this mark is associated with */
264 struct hlist_node m_list; /* list of marks by inode->i_fsnotify_marks */
265 struct list_head free_m_list; /* tmp list used when freeing this mark */
266};
267
268/*
269 * a mark is simply an object attached to an in core inode which allows an
270 * fsnotify listener to indicate they are either no longer interested in events
271 * of a type matching mask or only interested in those events.
272 *
273 * these are flushed when an inode is evicted from core and may be flushed
274 * when the inode is modified (as seen by fsnotify_access). Some fsnotify users
275 * (such as dnotify) will flush these when the open fd is closed and not at
276 * inode eviction or modification.
277 */
278struct fsnotify_mark {
279 __u32 mask; /* mask this mark is for */
280 /* we hold ref for each i_list and g_list. also one ref for each 'thing'
281 * in kernel that found and may be using this mark. */
282 atomic_t refcnt; /* active things looking at this mark */
283 struct fsnotify_group *group; /* group this mark is for */
284 struct list_head g_list; /* list of marks by group->i_fsnotify_marks */
285 spinlock_t lock; /* protect group and inode */
286 union {
287 struct fsnotify_inode_mark i;
288 struct fsnotify_vfsmount_mark m;
289 };
290 struct list_head free_g_list; /* tmp list used when freeing this mark */
291 __u32 ignored_mask; /* events types to ignore */
292#define FSNOTIFY_MARK_FLAG_INODE 0x01
293#define FSNOTIFY_MARK_FLAG_VFSMOUNT 0x02
294#define FSNOTIFY_MARK_FLAG_OBJECT_PINNED 0x04
295#define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x08
296#define FSNOTIFY_MARK_FLAG_ALIVE 0x10
297 unsigned int flags; /* vfsmount or inode mark? */
298 struct list_head destroy_list;
299 void (*free_mark)(struct fsnotify_mark *mark); /* called on final put+free */
300};
301
302#ifdef CONFIG_FSNOTIFY
303
304/* called from the vfs helpers */
305
306/* main fsnotify call to send events */
307extern int fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
308 const unsigned char *name, u32 cookie);
309extern int __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask);
310extern void __fsnotify_inode_delete(struct inode *inode);
311extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
312extern u32 fsnotify_get_cookie(void);
313
314static inline int fsnotify_inode_watches_children(struct inode *inode)
315{
316 /* FS_EVENT_ON_CHILD is set if the inode may care */
317 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
318 return 0;
319 /* this inode might care about child events, does it care about the
320 * specific set of events that can happen on a child? */
321 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
322}
323
324/*
325 * Update the dentry with a flag indicating the interest of its parent to receive
326 * filesystem events when those events happens to this dentry->d_inode.
327 */
328static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
329{
330 struct dentry *parent;
331
332 assert_spin_locked(&dentry->d_lock);
333
334 /*
335 * Serialisation of setting PARENT_WATCHED on the dentries is provided
336 * by d_lock. If inotify_inode_watched changes after we have taken
337 * d_lock, the following __fsnotify_update_child_dentry_flags call will
338 * find our entry, so it will spin until we complete here, and update
339 * us with the new state.
340 */
341 parent = dentry->d_parent;
342 if (parent->d_inode && fsnotify_inode_watches_children(parent->d_inode))
343 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
344 else
345 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
346}
347
348/*
349 * fsnotify_d_instantiate - instantiate a dentry for inode
350 */
351static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
352{
353 if (!inode)
354 return;
355
356 spin_lock(&dentry->d_lock);
357 __fsnotify_update_dcache_flags(dentry);
358 spin_unlock(&dentry->d_lock);
359}
360
361/* called from fsnotify listeners, such as fanotify or dnotify */
362
363/* get a reference to an existing or create a new group */
364extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
365/* drop reference on a group from fsnotify_alloc_group */
366extern void fsnotify_put_group(struct fsnotify_group *group);
367
368/* take a reference to an event */
369extern void fsnotify_get_event(struct fsnotify_event *event);
370extern void fsnotify_put_event(struct fsnotify_event *event);
371/* find private data previously attached to an event and unlink it */
372extern struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group,
373 struct fsnotify_event *event);
374
375/* attach the event to the group notification queue */
376extern struct fsnotify_event *fsnotify_add_notify_event(struct fsnotify_group *group,
377 struct fsnotify_event *event,
378 struct fsnotify_event_private_data *priv,
379 struct fsnotify_event *(*merge)(struct list_head *,
380 struct fsnotify_event *));
381/* true if the group notification queue is empty */
382extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
383/* return, but do not dequeue the first event on the notification queue */
384extern struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group);
385/* return AND dequeue the first event on the notification queue */
386extern struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group);
387
388/* functions used to manipulate the marks attached to inodes */
389
390/* run all marks associated with a vfsmount and update mnt->mnt_fsnotify_mask */
391extern void fsnotify_recalc_vfsmount_mask(struct vfsmount *mnt);
392/* run all marks associated with an inode and update inode->i_fsnotify_mask */
393extern void fsnotify_recalc_inode_mask(struct inode *inode);
394extern void fsnotify_init_mark(struct fsnotify_mark *mark, void (*free_mark)(struct fsnotify_mark *mark));
395/* find (and take a reference) to a mark associated with group and inode */
396extern struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group, struct inode *inode);
397/* find (and take a reference) to a mark associated with group and vfsmount */
398extern struct fsnotify_mark *fsnotify_find_vfsmount_mark(struct fsnotify_group *group, struct vfsmount *mnt);
399/* copy the values from old into new */
400extern void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old);
401/* set the ignored_mask of a mark */
402extern void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask);
403/* set the mask of a mark (might pin the object into memory */
404extern void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask);
405/* attach the mark to both the group and the inode */
406extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
407 struct inode *inode, struct vfsmount *mnt, int allow_dups);
408/* given a mark, flag it to be freed when all references are dropped */
409extern void fsnotify_destroy_mark(struct fsnotify_mark *mark);
410/* run all the marks in a group, and clear all of the vfsmount marks */
411extern void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group);
412/* run all the marks in a group, and clear all of the inode marks */
413extern void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group);
414/* run all the marks in a group, and clear all of the marks where mark->flags & flags is true*/
415extern void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group, unsigned int flags);
416/* run all the marks in a group, and flag them to be freed */
417extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group);
418extern void fsnotify_get_mark(struct fsnotify_mark *mark);
419extern void fsnotify_put_mark(struct fsnotify_mark *mark);
420extern void fsnotify_unmount_inodes(struct list_head *list);
421
422/* put here because inotify does some weird stuff when destroying watches */
423extern struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask,
424 void *data, int data_is,
425 const unsigned char *name,
426 u32 cookie, gfp_t gfp);
427
428/* fanotify likes to change events after they are on lists... */
429extern struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event);
430extern int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
431 struct fsnotify_event *new_event);
432
433#else
434
435static inline int fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
436 const unsigned char *name, u32 cookie)
437{
438 return 0;
439}
440
441static inline int __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask)
442{
443 return 0;
444}
445
446static inline void __fsnotify_inode_delete(struct inode *inode)
447{}
448
449static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
450{}
451
452static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
453{}
454
455static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
456{}
457
458static inline u32 fsnotify_get_cookie(void)
459{
460 return 0;
461}
462
463static inline void fsnotify_unmount_inodes(struct list_head *list)
464{}
465
466#endif /* CONFIG_FSNOTIFY */
467
468#endif /* __KERNEL __ */
469
470#endif /* __LINUX_FSNOTIFY_BACKEND_H */