fsnotify: allow marks to not pin inodes in core
[deliverable/linux.git] / include / 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 <asm/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_IN_ISDIR 0x40000000 /* event occurred against dir */
45 #define FS_IN_ONESHOT 0x80000000 /* only send event once */
46
47 #define FS_DN_RENAME 0x10000000 /* file renamed */
48 #define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
49
50 /* This inode cares about things that happen to its children. Always set for
51 * dnotify and inotify. */
52 #define FS_EVENT_ON_CHILD 0x08000000
53
54 /* This is a list of all events that may get sent to a parernt based on fs event
55 * happening to inodes inside that directory */
56 #define FS_EVENTS_POSS_ON_CHILD (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
57 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
58 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
59 FS_DELETE)
60
61 #define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
62
63 struct fsnotify_group;
64 struct fsnotify_event;
65 struct fsnotify_mark;
66 struct fsnotify_event_private_data;
67
68 /*
69 * Each group much define these ops. The fsnotify infrastructure will call
70 * these operations for each relevant group.
71 *
72 * should_send_event - given a group, inode, and mask this function determines
73 * if the group is interested in this event.
74 * handle_event - main call for a group to handle an fs event
75 * free_group_priv - called when a group refcnt hits 0 to clean up the private union
76 * freeing-mark - this means that a mark has been flagged to die when everything
77 * finishes using it. The function is supplied with what must be a
78 * valid group and inode to use to clean up.
79 */
80 struct fsnotify_ops {
81 bool (*should_send_event)(struct fsnotify_group *group, struct inode *inode,
82 struct vfsmount *mnt, __u32 mask, void *data,
83 int data_type);
84 int (*handle_event)(struct fsnotify_group *group, struct fsnotify_event *event);
85 void (*free_group_priv)(struct fsnotify_group *group);
86 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
87 void (*free_event_priv)(struct fsnotify_event_private_data *priv);
88 };
89
90 /*
91 * A group is a "thing" that wants to receive notification about filesystem
92 * events. The mask holds the subset of event types this group cares about.
93 * refcnt on a group is up to the implementor and at any moment if it goes 0
94 * everything will be cleaned up.
95 */
96 struct fsnotify_group {
97 /*
98 * global list of all groups receiving events from fsnotify.
99 * anchored by fsnotify_inode_groups and protected by either fsnotify_grp_mutex
100 * or fsnotify_grp_srcu depending on write vs read.
101 */
102 struct list_head inode_group_list;
103 /*
104 * same as above except anchored by fsnotify_vfsmount_groups
105 */
106 struct list_head vfsmount_group_list;
107
108 /*
109 * Defines all of the event types in which this group is interested.
110 * This mask is a bitwise OR of the FS_* events from above. Each time
111 * this mask changes for a group (if it changes) the correct functions
112 * must be called to update the global structures which indicate global
113 * interest in event types.
114 */
115 __u32 mask;
116
117 /*
118 * How the refcnt is used is up to each group. When the refcnt hits 0
119 * fsnotify will clean up all of the resources associated with this group.
120 * As an example, the dnotify group will always have a refcnt=1 and that
121 * will never change. Inotify, on the other hand, has a group per
122 * inotify_init() and the refcnt will hit 0 only when that fd has been
123 * closed.
124 */
125 atomic_t refcnt; /* things with interest in this group */
126
127 const struct fsnotify_ops *ops; /* how this group handles things */
128
129 /* needed to send notification to userspace */
130 struct mutex notification_mutex; /* protect the notification_list */
131 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
132 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
133 unsigned int q_len; /* events on the queue */
134 unsigned int max_events; /* maximum events allowed on the list */
135
136 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
137 spinlock_t mark_lock; /* protect marks_list */
138 atomic_t num_marks; /* 1 for each mark and 1 for not being
139 * past the point of no return when freeing
140 * a group */
141 struct list_head marks_list; /* all inode marks for this group */
142
143 /* prevents double list_del of group_list. protected by global fsnotify_grp_mutex */
144 bool on_inode_group_list;
145 bool on_vfsmount_group_list;
146
147 /* groups can define private fields here or use the void *private */
148 union {
149 void *private;
150 #ifdef CONFIG_INOTIFY_USER
151 struct inotify_group_private_data {
152 spinlock_t idr_lock;
153 struct idr idr;
154 u32 last_wd;
155 struct fasync_struct *fa; /* async notification */
156 struct user_struct *user;
157 } inotify_data;
158 #endif
159 };
160 };
161
162 /*
163 * A single event can be queued in multiple group->notification_lists.
164 *
165 * each group->notification_list will point to an event_holder which in turns points
166 * to the actual event that needs to be sent to userspace.
167 *
168 * Seemed cheaper to create a refcnt'd event and a small holder for every group
169 * than create a different event for every group
170 *
171 */
172 struct fsnotify_event_holder {
173 struct fsnotify_event *event;
174 struct list_head event_list;
175 };
176
177 /*
178 * Inotify needs to tack data onto an event. This struct lets us later find the
179 * correct private data of the correct group.
180 */
181 struct fsnotify_event_private_data {
182 struct fsnotify_group *group;
183 struct list_head event_list;
184 };
185
186 /*
187 * all of the information about the original object we want to now send to
188 * a group. If you want to carry more info from the accessing task to the
189 * listener this structure is where you need to be adding fields.
190 */
191 struct fsnotify_event {
192 /*
193 * If we create an event we are also likely going to need a holder
194 * to link to a group. So embed one holder in the event. Means only
195 * one allocation for the common case where we only have one group
196 */
197 struct fsnotify_event_holder holder;
198 spinlock_t lock; /* protection for the associated event_holder and private_list */
199 /* to_tell may ONLY be dereferenced during handle_event(). */
200 struct inode *to_tell; /* either the inode the event happened to or its parent */
201 /*
202 * depending on the event type we should have either a path or inode
203 * We hold a reference on path, but NOT on inode. Since we have the ref on
204 * the path, it may be dereferenced at any point during this object's
205 * lifetime. That reference is dropped when this object's refcnt hits
206 * 0. If this event contains an inode instead of a path, the inode may
207 * ONLY be used during handle_event().
208 */
209 union {
210 struct path path;
211 struct inode *inode;
212 };
213 /* when calling fsnotify tell it if the data is a path or inode */
214 #define FSNOTIFY_EVENT_NONE 0
215 #define FSNOTIFY_EVENT_PATH 1
216 #define FSNOTIFY_EVENT_INODE 2
217 int data_type; /* which of the above union we have */
218 atomic_t refcnt; /* how many groups still are using/need to send this event */
219 __u32 mask; /* the type of access, bitwise OR for FS_* event types */
220
221 u32 sync_cookie; /* used to corrolate events, namely inotify mv events */
222 char *file_name;
223 size_t name_len;
224 struct pid *tgid;
225
226 struct list_head private_data_list; /* groups can store private data here */
227 };
228
229 /*
230 * Inode specific fields in an fsnotify_mark
231 */
232 struct fsnotify_inode_mark {
233 struct inode *inode; /* inode this mark is associated with */
234 struct hlist_node i_list; /* list of marks by inode->i_fsnotify_marks */
235 struct list_head free_i_list; /* tmp list used when freeing this mark */
236 };
237
238 /*
239 * Mount point specific fields in an fsnotify_mark
240 */
241 struct fsnotify_vfsmount_mark {
242 struct vfsmount *mnt; /* vfsmount this mark is associated with */
243 struct hlist_node m_list; /* list of marks by inode->i_fsnotify_marks */
244 struct list_head free_m_list; /* tmp list used when freeing this mark */
245 };
246
247 /*
248 * a mark is simply an object attached to an in core inode which allows an
249 * fsnotify listener to indicate they are either no longer interested in events
250 * of a type matching mask or only interested in those events.
251 *
252 * these are flushed when an inode is evicted from core and may be flushed
253 * when the inode is modified (as seen by fsnotify_access). Some fsnotify users
254 * (such as dnotify) will flush these when the open fd is closed and not at
255 * inode eviction or modification.
256 */
257 struct fsnotify_mark {
258 __u32 mask; /* mask this mark is for */
259 /* we hold ref for each i_list and g_list. also one ref for each 'thing'
260 * in kernel that found and may be using this mark. */
261 atomic_t refcnt; /* active things looking at this mark */
262 struct fsnotify_group *group; /* group this mark is for */
263 struct list_head g_list; /* list of marks by group->i_fsnotify_marks */
264 spinlock_t lock; /* protect group and inode */
265 union {
266 struct fsnotify_inode_mark i;
267 struct fsnotify_vfsmount_mark m;
268 };
269 struct list_head free_g_list; /* tmp list used when freeing this mark */
270 #define FSNOTIFY_MARK_FLAG_INODE 0x01
271 #define FSNOTIFY_MARK_FLAG_VFSMOUNT 0x02
272 #define FSNOTIFY_MARK_FLAG_OBJECT_PINNED 0x04
273 unsigned int flags; /* vfsmount or inode mark? */
274 void (*free_mark)(struct fsnotify_mark *mark); /* called on final put+free */
275 };
276
277 #ifdef CONFIG_FSNOTIFY
278
279 /* called from the vfs helpers */
280
281 /* main fsnotify call to send events */
282 extern void fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
283 const char *name, u32 cookie);
284 extern void __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask);
285 extern void __fsnotify_inode_delete(struct inode *inode);
286 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
287 extern u32 fsnotify_get_cookie(void);
288
289 static inline int fsnotify_inode_watches_children(struct inode *inode)
290 {
291 /* FS_EVENT_ON_CHILD is set if the inode may care */
292 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
293 return 0;
294 /* this inode might care about child events, does it care about the
295 * specific set of events that can happen on a child? */
296 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
297 }
298
299 /*
300 * Update the dentry with a flag indicating the interest of its parent to receive
301 * filesystem events when those events happens to this dentry->d_inode.
302 */
303 static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
304 {
305 struct dentry *parent;
306
307 assert_spin_locked(&dcache_lock);
308 assert_spin_locked(&dentry->d_lock);
309
310 parent = dentry->d_parent;
311 if (parent->d_inode && fsnotify_inode_watches_children(parent->d_inode))
312 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
313 else
314 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
315 }
316
317 /*
318 * fsnotify_d_instantiate - instantiate a dentry for inode
319 * Called with dcache_lock held.
320 */
321 static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
322 {
323 if (!inode)
324 return;
325
326 assert_spin_locked(&dcache_lock);
327
328 spin_lock(&dentry->d_lock);
329 __fsnotify_update_dcache_flags(dentry);
330 spin_unlock(&dentry->d_lock);
331 }
332
333 /* called from fsnotify listeners, such as fanotify or dnotify */
334
335 /* must call when a group changes its ->mask */
336 extern void fsnotify_recalc_global_mask(void);
337 /* get a reference to an existing or create a new group */
338 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
339 /* run all marks associated with this group and update group->mask */
340 extern void fsnotify_recalc_group_mask(struct fsnotify_group *group);
341 /* drop reference on a group from fsnotify_alloc_group */
342 extern void fsnotify_put_group(struct fsnotify_group *group);
343
344 /* take a reference to an event */
345 extern void fsnotify_get_event(struct fsnotify_event *event);
346 extern void fsnotify_put_event(struct fsnotify_event *event);
347 /* find private data previously attached to an event and unlink it */
348 extern struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group,
349 struct fsnotify_event *event);
350
351 /* attach the event to the group notification queue */
352 extern int fsnotify_add_notify_event(struct fsnotify_group *group,
353 struct fsnotify_event *event,
354 struct fsnotify_event_private_data *priv,
355 int (*merge)(struct list_head *, struct fsnotify_event *));
356 /* true if the group notification queue is empty */
357 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
358 /* return, but do not dequeue the first event on the notification queue */
359 extern struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group);
360 /* return AND dequeue the first event on the notification queue */
361 extern struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group);
362
363 /* functions used to manipulate the marks attached to inodes */
364
365 /* run all marks associated with a vfsmount and update mnt->mnt_fsnotify_mask */
366 extern void fsnotify_recalc_vfsmount_mask(struct vfsmount *mnt);
367 /* run all marks associated with an inode and update inode->i_fsnotify_mask */
368 extern void fsnotify_recalc_inode_mask(struct inode *inode);
369 extern void fsnotify_init_mark(struct fsnotify_mark *mark, void (*free_mark)(struct fsnotify_mark *mark));
370 /* find (and take a reference) to a mark associated with group and inode */
371 extern struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group, struct inode *inode);
372 /* find (and take a reference) to a mark associated with group and vfsmount */
373 extern struct fsnotify_mark *fsnotify_find_vfsmount_mark(struct fsnotify_group *group, struct vfsmount *mnt);
374 /* copy the values from old into new */
375 extern void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old);
376 /* set the mask of a mark (might pin the object into memory */
377 extern void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask);
378 /* attach the mark to both the group and the inode */
379 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
380 struct inode *inode, struct vfsmount *mnt, int allow_dups);
381 /* given a mark, flag it to be freed when all references are dropped */
382 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark);
383 /* run all the marks in a group, and flag them to be freed */
384 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group);
385 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
386 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
387 extern void fsnotify_unmount_inodes(struct list_head *list);
388
389 /* put here because inotify does some weird stuff when destroying watches */
390 extern struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask,
391 void *data, int data_is, const char *name,
392 u32 cookie, gfp_t gfp);
393
394 /* fanotify likes to change events after they are on lists... */
395 extern struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event);
396 extern int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
397 struct fsnotify_event *new_event);
398
399 #else
400
401 static inline void fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
402 const char *name, u32 cookie)
403 {}
404
405 static inline void __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask)
406 {}
407
408 static inline void __fsnotify_inode_delete(struct inode *inode)
409 {}
410
411 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
412 {}
413
414 static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
415 {}
416
417 static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
418 {}
419
420 static inline u32 fsnotify_get_cookie(void)
421 {
422 return 0;
423 }
424
425 static inline void fsnotify_unmount_inodes(struct list_head *list)
426 {}
427
428 #endif /* CONFIG_FSNOTIFY */
429
430 #endif /* __KERNEL __ */
431
432 #endif /* __LINUX_FSNOTIFY_BACKEND_H */
This page took 0.050105 seconds and 5 git commands to generate.