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 #include <linux/mempool.h>
23 #include <linux/sched/mm.h>
24
25 /*
26 * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
27 * convert between them. dnotify only needs conversion at watch creation
28 * so no perf loss there. fanotify isn't defined yet, so it can use the
29 * wholes if it needs more events.
30 */
31 #define FS_ACCESS 0x00000001 /* File was accessed */
32 #define FS_MODIFY 0x00000002 /* File was modified */
33 #define FS_ATTRIB 0x00000004 /* Metadata changed */
34 #define FS_CLOSE_WRITE 0x00000008 /* Writable file was closed */
35 #define FS_CLOSE_NOWRITE 0x00000010 /* Unwritable file closed */
36 #define FS_OPEN 0x00000020 /* File was opened */
37 #define FS_MOVED_FROM 0x00000040 /* File was moved from X */
38 #define FS_MOVED_TO 0x00000080 /* File was moved to Y */
39 #define FS_CREATE 0x00000100 /* Subfile was created */
40 #define FS_DELETE 0x00000200 /* Subfile was deleted */
41 #define FS_DELETE_SELF 0x00000400 /* Self was deleted */
42 #define FS_MOVE_SELF 0x00000800 /* Self was moved */
43 #define FS_OPEN_EXEC 0x00001000 /* File was opened for exec */
44
45 #define FS_UNMOUNT 0x00002000 /* inode on umount fs */
46 #define FS_Q_OVERFLOW 0x00004000 /* Event queued overflowed */
47 #define FS_ERROR 0x00008000 /* Filesystem Error (fanotify) */
48
49 /*
50 * FS_IN_IGNORED overloads FS_ERROR. It is only used internally by inotify
51 * which does not support FS_ERROR.
52 */
53 #define FS_IN_IGNORED 0x00008000 /* last inotify event here */
54
55 #define FS_OPEN_PERM 0x00010000 /* open event in an permission hook */
56 #define FS_ACCESS_PERM 0x00020000 /* access event in a permissions hook */
57 #define FS_OPEN_EXEC_PERM 0x00040000 /* open/exec event in a permission hook */
58
59 /*
60 * Set on inode mark that cares about things that happen to its children.
61 * Always set for dnotify and inotify.
62 * Set on inode/sb/mount marks that care about parent/name info.
63 */
64 #define FS_EVENT_ON_CHILD 0x08000000
65
66 #define FS_RENAME 0x10000000 /* File was renamed */
67 #define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
68 #define FS_ISDIR 0x40000000 /* event occurred against dir */
69
70 #define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
71
72 /*
73 * Directory entry modification events - reported only to directory
74 * where entry is modified and not to a watching parent.
75 * The watching parent may get an FS_ATTRIB|FS_EVENT_ON_CHILD event
76 * when a directory entry inside a child subdir changes.
77 */
78 #define ALL_FSNOTIFY_DIRENT_EVENTS (FS_CREATE | FS_DELETE | FS_MOVE | FS_RENAME)
79
80 #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM | \
81 FS_OPEN_EXEC_PERM)
82
83 /*
84 * This is a list of all events that may get sent to a parent that is watching
85 * with flag FS_EVENT_ON_CHILD based on fs event on a child of that directory.
86 */
87 #define FS_EVENTS_POSS_ON_CHILD (ALL_FSNOTIFY_PERM_EVENTS | \
88 FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
89 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | \
90 FS_OPEN | FS_OPEN_EXEC)
91
92 /*
93 * This is a list of all events that may get sent with the parent inode as the
94 * @to_tell argument of fsnotify().
95 * It may include events that can be sent to an inode/sb/mount mark, but cannot
96 * be sent to a parent watching children.
97 */
98 #define FS_EVENTS_POSS_TO_PARENT (FS_EVENTS_POSS_ON_CHILD)
99
100 /* Events that can be reported to backends */
101 #define ALL_FSNOTIFY_EVENTS (ALL_FSNOTIFY_DIRENT_EVENTS | \
102 FS_EVENTS_POSS_ON_CHILD | \
103 FS_DELETE_SELF | FS_MOVE_SELF | \
104 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
105 FS_ERROR)
106
107 /* Extra flags that may be reported with event or control handling of events */
108 #define ALL_FSNOTIFY_FLAGS (FS_ISDIR | FS_EVENT_ON_CHILD | FS_DN_MULTISHOT)
109
110 #define ALL_FSNOTIFY_BITS (ALL_FSNOTIFY_EVENTS | ALL_FSNOTIFY_FLAGS)
111
112 struct fsnotify_group;
113 struct fsnotify_event;
114 struct fsnotify_mark;
115 struct fsnotify_event_private_data;
116 struct fsnotify_fname;
117 struct fsnotify_iter_info;
118
119 struct mem_cgroup;
120
121 /*
122 * Each group much define these ops. The fsnotify infrastructure will call
123 * these operations for each relevant group.
124 *
125 * handle_event - main call for a group to handle an fs event
126 * @group: group to notify
127 * @mask: event type and flags
128 * @data: object that event happened on
129 * @data_type: type of object for fanotify_data_XXX() accessors
130 * @dir: optional directory associated with event -
131 * if @file_name is not NULL, this is the directory that
132 * @file_name is relative to
133 * @file_name: optional file name associated with event
134 * @cookie: inotify rename cookie
135 * @iter_info: array of marks from this group that are interested in the event
136 *
137 * handle_inode_event - simple variant of handle_event() for groups that only
138 * have inode marks and don't have ignore mask
139 * @mark: mark to notify
140 * @mask: event type and flags
141 * @inode: inode that event happened on
142 * @dir: optional directory associated with event -
143 * if @file_name is not NULL, this is the directory that
144 * @file_name is relative to.
145 * Either @inode or @dir must be non-NULL.
146 * @file_name: optional file name associated with event
147 * @cookie: inotify rename cookie
148 *
149 * free_group_priv - called when a group refcnt hits 0 to clean up the private union
150 * freeing_mark - called when a mark is being destroyed for some reason. The group
151 * MUST be holding a reference on each mark and that reference must be
152 * dropped in this function. inotify uses this function to send
153 * userspace messages that marks have been removed.
154 */
155 struct fsnotify_ops {
156 int (*handle_event)(struct fsnotify_group *group, u32 mask,
157 const void *data, int data_type, struct inode *dir,
158 const struct qstr *file_name, u32 cookie,
159 struct fsnotify_iter_info *iter_info);
160 int (*handle_inode_event)(struct fsnotify_mark *mark, u32 mask,
161 struct inode *inode, struct inode *dir,
162 const struct qstr *file_name, u32 cookie);
163 void (*free_group_priv)(struct fsnotify_group *group);
164 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
165 void (*free_event)(struct fsnotify_group *group, struct fsnotify_event *event);
166 /* called on final put+free to free memory */
167 void (*free_mark)(struct fsnotify_mark *mark);
168 };
169
170 /*
171 * all of the information about the original object we want to now send to
172 * a group. If you want to carry more info from the accessing task to the
173 * listener this structure is where you need to be adding fields.
174 */
175 struct fsnotify_event {
176 struct list_head list;
177 };
178
179 /*
180 * fsnotify group priorities.
181 * Events are sent in order from highest priority to lowest priority.
182 */
183 enum fsnotify_group_prio {
184 FSNOTIFY_PRIO_NORMAL = 0, /* normal notifiers, no permissions */
185 FSNOTIFY_PRIO_CONTENT, /* fanotify permission events */
186 FSNOTIFY_PRIO_PRE_CONTENT, /* fanotify pre-content events */
187 __FSNOTIFY_PRIO_NUM
188 };
189
190 /*
191 * A group is a "thing" that wants to receive notification about filesystem
192 * events. The mask holds the subset of event types this group cares about.
193 * refcnt on a group is up to the implementor and at any moment if it goes 0
194 * everything will be cleaned up.
195 */
196 struct fsnotify_group {
197 const struct fsnotify_ops *ops; /* how this group handles things */
198
199 /*
200 * How the refcnt is used is up to each group. When the refcnt hits 0
201 * fsnotify will clean up all of the resources associated with this group.
202 * As an example, the dnotify group will always have a refcnt=1 and that
203 * will never change. Inotify, on the other hand, has a group per
204 * inotify_init() and the refcnt will hit 0 only when that fd has been
205 * closed.
206 */
207 refcount_t refcnt; /* things with interest in this group */
208
209 /* needed to send notification to userspace */
210 spinlock_t notification_lock; /* protect the notification_list */
211 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
212 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
213 unsigned int q_len; /* events on the queue */
214 unsigned int max_events; /* maximum events allowed on the list */
215 enum fsnotify_group_prio priority; /* priority for sending events */
216 bool shutdown; /* group is being shut down, don't queue more events */
217
218 #define FSNOTIFY_GROUP_USER 0x01 /* user allocated group */
219 #define FSNOTIFY_GROUP_DUPS 0x02 /* allow multiple marks per object */
220 #define FSNOTIFY_GROUP_NOFS 0x04 /* group lock is not direct reclaim safe */
221 int flags;
222 unsigned int owner_flags; /* stored flags of mark_mutex owner */
223
224 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
225 struct mutex mark_mutex; /* protect marks_list */
226 atomic_t user_waits; /* Number of tasks waiting for user
227 * response */
228 struct list_head marks_list; /* all inode marks for this group */
229
230 struct fasync_struct *fsn_fa; /* async notification */
231
232 struct fsnotify_event *overflow_event; /* Event we queue when the
233 * notification list is too
234 * full */
235
236 struct mem_cgroup *memcg; /* memcg to charge allocations */
237
238 /* groups can define private fields here or use the void *private */
239 union {
240 void *private;
241 #ifdef CONFIG_INOTIFY_USER
242 struct inotify_group_private_data {
243 spinlock_t idr_lock;
244 struct idr idr;
245 struct ucounts *ucounts;
246 } inotify_data;
247 #endif
248 #ifdef CONFIG_FANOTIFY
249 struct fanotify_group_private_data {
250 /* Hash table of events for merge */
251 struct hlist_head *merge_hash;
252 /* allows a group to block waiting for a userspace response */
253 struct list_head access_list;
254 wait_queue_head_t access_waitq;
255 int flags; /* flags from fanotify_init() */
256 int f_flags; /* event_f_flags from fanotify_init() */
257 struct ucounts *ucounts;
258 mempool_t error_events_pool;
259 } fanotify_data;
260 #endif /* CONFIG_FANOTIFY */
261 };
262 };
263
264 /*
265 * These helpers are used to prevent deadlock when reclaiming inodes with
266 * evictable marks of the same group that is allocating a new mark.
267 */
fsnotify_group_lock(struct fsnotify_group * group)268 static inline void fsnotify_group_lock(struct fsnotify_group *group)
269 {
270 mutex_lock(&group->mark_mutex);
271 if (group->flags & FSNOTIFY_GROUP_NOFS)
272 group->owner_flags = memalloc_nofs_save();
273 }
274
fsnotify_group_unlock(struct fsnotify_group * group)275 static inline void fsnotify_group_unlock(struct fsnotify_group *group)
276 {
277 if (group->flags & FSNOTIFY_GROUP_NOFS)
278 memalloc_nofs_restore(group->owner_flags);
279 mutex_unlock(&group->mark_mutex);
280 }
281
fsnotify_group_assert_locked(struct fsnotify_group * group)282 static inline void fsnotify_group_assert_locked(struct fsnotify_group *group)
283 {
284 WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
285 if (group->flags & FSNOTIFY_GROUP_NOFS)
286 WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
287 }
288
289 /* When calling fsnotify tell it if the data is a path or inode */
290 enum fsnotify_data_type {
291 FSNOTIFY_EVENT_NONE,
292 FSNOTIFY_EVENT_PATH,
293 FSNOTIFY_EVENT_INODE,
294 FSNOTIFY_EVENT_DENTRY,
295 FSNOTIFY_EVENT_ERROR,
296 };
297
298 struct fs_error_report {
299 int error;
300 struct inode *inode;
301 struct super_block *sb;
302 };
303
fsnotify_data_inode(const void * data,int data_type)304 static inline struct inode *fsnotify_data_inode(const void *data, int data_type)
305 {
306 switch (data_type) {
307 case FSNOTIFY_EVENT_INODE:
308 return (struct inode *)data;
309 case FSNOTIFY_EVENT_DENTRY:
310 return d_inode(data);
311 case FSNOTIFY_EVENT_PATH:
312 return d_inode(((const struct path *)data)->dentry);
313 case FSNOTIFY_EVENT_ERROR:
314 return ((struct fs_error_report *)data)->inode;
315 default:
316 return NULL;
317 }
318 }
319
fsnotify_data_dentry(const void * data,int data_type)320 static inline struct dentry *fsnotify_data_dentry(const void *data, int data_type)
321 {
322 switch (data_type) {
323 case FSNOTIFY_EVENT_DENTRY:
324 /* Non const is needed for dget() */
325 return (struct dentry *)data;
326 case FSNOTIFY_EVENT_PATH:
327 return ((const struct path *)data)->dentry;
328 default:
329 return NULL;
330 }
331 }
332
fsnotify_data_path(const void * data,int data_type)333 static inline const struct path *fsnotify_data_path(const void *data,
334 int data_type)
335 {
336 switch (data_type) {
337 case FSNOTIFY_EVENT_PATH:
338 return data;
339 default:
340 return NULL;
341 }
342 }
343
fsnotify_data_sb(const void * data,int data_type)344 static inline struct super_block *fsnotify_data_sb(const void *data,
345 int data_type)
346 {
347 switch (data_type) {
348 case FSNOTIFY_EVENT_INODE:
349 return ((struct inode *)data)->i_sb;
350 case FSNOTIFY_EVENT_DENTRY:
351 return ((struct dentry *)data)->d_sb;
352 case FSNOTIFY_EVENT_PATH:
353 return ((const struct path *)data)->dentry->d_sb;
354 case FSNOTIFY_EVENT_ERROR:
355 return ((struct fs_error_report *) data)->sb;
356 default:
357 return NULL;
358 }
359 }
360
fsnotify_data_error_report(const void * data,int data_type)361 static inline struct fs_error_report *fsnotify_data_error_report(
362 const void *data,
363 int data_type)
364 {
365 switch (data_type) {
366 case FSNOTIFY_EVENT_ERROR:
367 return (struct fs_error_report *) data;
368 default:
369 return NULL;
370 }
371 }
372
373 /*
374 * Index to merged marks iterator array that correlates to a type of watch.
375 * The type of watched object can be deduced from the iterator type, but not
376 * the other way around, because an event can match different watched objects
377 * of the same object type.
378 * For example, both parent and child are watching an object of type inode.
379 */
380 enum fsnotify_iter_type {
381 FSNOTIFY_ITER_TYPE_INODE,
382 FSNOTIFY_ITER_TYPE_VFSMOUNT,
383 FSNOTIFY_ITER_TYPE_SB,
384 FSNOTIFY_ITER_TYPE_PARENT,
385 FSNOTIFY_ITER_TYPE_INODE2,
386 FSNOTIFY_ITER_TYPE_COUNT
387 };
388
389 /* The type of object that a mark is attached to */
390 enum fsnotify_obj_type {
391 FSNOTIFY_OBJ_TYPE_ANY = -1,
392 FSNOTIFY_OBJ_TYPE_INODE,
393 FSNOTIFY_OBJ_TYPE_VFSMOUNT,
394 FSNOTIFY_OBJ_TYPE_SB,
395 FSNOTIFY_OBJ_TYPE_COUNT,
396 FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
397 };
398
fsnotify_valid_obj_type(unsigned int obj_type)399 static inline bool fsnotify_valid_obj_type(unsigned int obj_type)
400 {
401 return (obj_type < FSNOTIFY_OBJ_TYPE_COUNT);
402 }
403
404 struct fsnotify_iter_info {
405 struct fsnotify_mark *marks[FSNOTIFY_ITER_TYPE_COUNT];
406 struct fsnotify_group *current_group;
407 unsigned int report_mask;
408 int srcu_idx;
409 };
410
fsnotify_iter_should_report_type(struct fsnotify_iter_info * iter_info,int iter_type)411 static inline bool fsnotify_iter_should_report_type(
412 struct fsnotify_iter_info *iter_info, int iter_type)
413 {
414 return (iter_info->report_mask & (1U << iter_type));
415 }
416
fsnotify_iter_set_report_type(struct fsnotify_iter_info * iter_info,int iter_type)417 static inline void fsnotify_iter_set_report_type(
418 struct fsnotify_iter_info *iter_info, int iter_type)
419 {
420 iter_info->report_mask |= (1U << iter_type);
421 }
422
fsnotify_iter_mark(struct fsnotify_iter_info * iter_info,int iter_type)423 static inline struct fsnotify_mark *fsnotify_iter_mark(
424 struct fsnotify_iter_info *iter_info, int iter_type)
425 {
426 if (fsnotify_iter_should_report_type(iter_info, iter_type))
427 return iter_info->marks[iter_type];
428 return NULL;
429 }
430
fsnotify_iter_step(struct fsnotify_iter_info * iter,int type,struct fsnotify_mark ** markp)431 static inline int fsnotify_iter_step(struct fsnotify_iter_info *iter, int type,
432 struct fsnotify_mark **markp)
433 {
434 while (type < FSNOTIFY_ITER_TYPE_COUNT) {
435 *markp = fsnotify_iter_mark(iter, type);
436 if (*markp)
437 break;
438 type++;
439 }
440 return type;
441 }
442
443 #define FSNOTIFY_ITER_FUNCS(name, NAME) \
444 static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \
445 struct fsnotify_iter_info *iter_info) \
446 { \
447 return fsnotify_iter_mark(iter_info, FSNOTIFY_ITER_TYPE_##NAME); \
448 }
449
450 FSNOTIFY_ITER_FUNCS(inode, INODE)
451 FSNOTIFY_ITER_FUNCS(parent, PARENT)
452 FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT)
453 FSNOTIFY_ITER_FUNCS(sb, SB)
454
455 #define fsnotify_foreach_iter_type(type) \
456 for (type = 0; type < FSNOTIFY_ITER_TYPE_COUNT; type++)
457 #define fsnotify_foreach_iter_mark_type(iter, mark, type) \
458 for (type = 0; \
459 type = fsnotify_iter_step(iter, type, &mark), \
460 type < FSNOTIFY_ITER_TYPE_COUNT; \
461 type++)
462
463 /*
464 * Inode/vfsmount/sb point to this structure which tracks all marks attached to
465 * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this
466 * structure. We destroy this structure when there are no more marks attached
467 * to it. The structure is protected by fsnotify_mark_srcu.
468 */
469 struct fsnotify_mark_connector {
470 spinlock_t lock;
471 unsigned char type; /* Type of object [lock] */
472 unsigned char prio; /* Highest priority group */
473 #define FSNOTIFY_CONN_FLAG_IS_WATCHED 0x01
474 #define FSNOTIFY_CONN_FLAG_HAS_IREF 0x02
475 unsigned short flags; /* flags [lock] */
476 union {
477 /* Object pointer [lock] */
478 void *obj;
479 /* Used listing heads to free after srcu period expires */
480 struct fsnotify_mark_connector *destroy_next;
481 };
482 struct hlist_head list;
483 };
484
485 /*
486 * Container for per-sb fsnotify state (sb marks and more).
487 * Attached lazily on first marked object on the sb and freed when killing sb.
488 */
489 struct fsnotify_sb_info {
490 struct fsnotify_mark_connector __rcu *sb_marks;
491 /*
492 * Number of inode/mount/sb objects that are being watched in this sb.
493 * Note that inodes objects are currently double-accounted.
494 *
495 * The value in watched_objects[prio] is the number of objects that are
496 * watched by groups of priority >= prio, so watched_objects[0] is the
497 * total number of watched objects in this sb.
498 */
499 atomic_long_t watched_objects[__FSNOTIFY_PRIO_NUM];
500 };
501
fsnotify_sb_info(struct super_block * sb)502 static inline struct fsnotify_sb_info *fsnotify_sb_info(struct super_block *sb)
503 {
504 #ifdef CONFIG_FSNOTIFY
505 return READ_ONCE(sb->s_fsnotify_info);
506 #else
507 return NULL;
508 #endif
509 }
510
fsnotify_sb_watched_objects(struct super_block * sb)511 static inline atomic_long_t *fsnotify_sb_watched_objects(struct super_block *sb)
512 {
513 return &fsnotify_sb_info(sb)->watched_objects[0];
514 }
515
516 /*
517 * A mark is simply an object attached to an in core inode which allows an
518 * fsnotify listener to indicate they are either no longer interested in events
519 * of a type matching mask or only interested in those events.
520 *
521 * These are flushed when an inode is evicted from core and may be flushed
522 * when the inode is modified (as seen by fsnotify_access). Some fsnotify
523 * users (such as dnotify) will flush these when the open fd is closed and not
524 * at inode eviction or modification.
525 *
526 * Text in brackets is showing the lock(s) protecting modifications of a
527 * particular entry. obj_lock means either inode->i_lock or
528 * mnt->mnt_root->d_lock depending on the mark type.
529 */
530 struct fsnotify_mark {
531 /* Mask this mark is for [mark->lock, group->mark_mutex] */
532 __u32 mask;
533 /* We hold one for presence in g_list. Also one ref for each 'thing'
534 * in kernel that found and may be using this mark. */
535 refcount_t refcnt;
536 /* Group this mark is for. Set on mark creation, stable until last ref
537 * is dropped */
538 struct fsnotify_group *group;
539 /* List of marks by group->marks_list. Also reused for queueing
540 * mark into destroy_list when it's waiting for the end of SRCU period
541 * before it can be freed. [group->mark_mutex] */
542 struct list_head g_list;
543 /* Protects inode / mnt pointers, flags, masks */
544 spinlock_t lock;
545 /* List of marks for inode / vfsmount [connector->lock, mark ref] */
546 struct hlist_node obj_list;
547 /* Head of list of marks for an object [mark ref] */
548 struct fsnotify_mark_connector *connector;
549 /* Events types and flags to ignore [mark->lock, group->mark_mutex] */
550 __u32 ignore_mask;
551 /* General fsnotify mark flags */
552 #define FSNOTIFY_MARK_FLAG_ALIVE 0x0001
553 #define FSNOTIFY_MARK_FLAG_ATTACHED 0x0002
554 /* inotify mark flags */
555 #define FSNOTIFY_MARK_FLAG_EXCL_UNLINK 0x0010
556 #define FSNOTIFY_MARK_FLAG_IN_ONESHOT 0x0020
557 /* fanotify mark flags */
558 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x0100
559 #define FSNOTIFY_MARK_FLAG_NO_IREF 0x0200
560 #define FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS 0x0400
561 #define FSNOTIFY_MARK_FLAG_HAS_FSID 0x0800
562 #define FSNOTIFY_MARK_FLAG_WEAK_FSID 0x1000
563 unsigned int flags; /* flags [mark->lock] */
564 };
565
566 #ifdef CONFIG_FSNOTIFY
567
568 /* called from the vfs helpers */
569
570 /* main fsnotify call to send events */
571 extern int fsnotify(__u32 mask, const void *data, int data_type,
572 struct inode *dir, const struct qstr *name,
573 struct inode *inode, u32 cookie);
574 extern int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
575 int data_type);
576 extern void __fsnotify_inode_delete(struct inode *inode);
577 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
578 extern void fsnotify_sb_delete(struct super_block *sb);
579 extern void fsnotify_sb_free(struct super_block *sb);
580 extern u32 fsnotify_get_cookie(void);
581
fsnotify_parent_needed_mask(__u32 mask)582 static inline __u32 fsnotify_parent_needed_mask(__u32 mask)
583 {
584 /* FS_EVENT_ON_CHILD is set on marks that want parent/name info */
585 if (!(mask & FS_EVENT_ON_CHILD))
586 return 0;
587 /*
588 * This object might be watched by a mark that cares about parent/name
589 * info, does it care about the specific set of events that can be
590 * reported with parent/name info?
591 */
592 return mask & FS_EVENTS_POSS_TO_PARENT;
593 }
594
fsnotify_inode_watches_children(struct inode * inode)595 static inline int fsnotify_inode_watches_children(struct inode *inode)
596 {
597 __u32 parent_mask = READ_ONCE(inode->i_fsnotify_mask);
598
599 /* FS_EVENT_ON_CHILD is set if the inode may care */
600 if (!(parent_mask & FS_EVENT_ON_CHILD))
601 return 0;
602 /* this inode might care about child events, does it care about the
603 * specific set of events that can happen on a child? */
604 return parent_mask & FS_EVENTS_POSS_ON_CHILD;
605 }
606
607 /*
608 * Update the dentry with a flag indicating the interest of its parent to receive
609 * filesystem events when those events happens to this dentry->d_inode.
610 */
fsnotify_update_flags(struct dentry * dentry)611 static inline void fsnotify_update_flags(struct dentry *dentry)
612 {
613 assert_spin_locked(&dentry->d_lock);
614
615 /*
616 * Serialisation of setting PARENT_WATCHED on the dentries is provided
617 * by d_lock. If inotify_inode_watched changes after we have taken
618 * d_lock, the following fsnotify_set_children_dentry_flags call will
619 * find our entry, so it will spin until we complete here, and update
620 * us with the new state.
621 */
622 if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
623 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
624 else
625 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
626 }
627
628 /* called from fsnotify listeners, such as fanotify or dnotify */
629
630 /* create a new group */
631 extern struct fsnotify_group *fsnotify_alloc_group(
632 const struct fsnotify_ops *ops,
633 int flags);
634 /* get reference to a group */
635 extern void fsnotify_get_group(struct fsnotify_group *group);
636 /* drop reference on a group from fsnotify_alloc_group */
637 extern void fsnotify_put_group(struct fsnotify_group *group);
638 /* group destruction begins, stop queuing new events */
639 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
640 /* destroy group */
641 extern void fsnotify_destroy_group(struct fsnotify_group *group);
642 /* fasync handler function */
643 extern int fsnotify_fasync(int fd, struct file *file, int on);
644 /* Free event from memory */
645 extern void fsnotify_destroy_event(struct fsnotify_group *group,
646 struct fsnotify_event *event);
647 /* attach the event to the group notification queue */
648 extern int fsnotify_insert_event(struct fsnotify_group *group,
649 struct fsnotify_event *event,
650 int (*merge)(struct fsnotify_group *,
651 struct fsnotify_event *),
652 void (*insert)(struct fsnotify_group *,
653 struct fsnotify_event *));
654
fsnotify_add_event(struct fsnotify_group * group,struct fsnotify_event * event,int (* merge)(struct fsnotify_group *,struct fsnotify_event *))655 static inline int fsnotify_add_event(struct fsnotify_group *group,
656 struct fsnotify_event *event,
657 int (*merge)(struct fsnotify_group *,
658 struct fsnotify_event *))
659 {
660 return fsnotify_insert_event(group, event, merge, NULL);
661 }
662
663 /* Queue overflow event to a notification group */
fsnotify_queue_overflow(struct fsnotify_group * group)664 static inline void fsnotify_queue_overflow(struct fsnotify_group *group)
665 {
666 fsnotify_add_event(group, group->overflow_event, NULL);
667 }
668
fsnotify_is_overflow_event(u32 mask)669 static inline bool fsnotify_is_overflow_event(u32 mask)
670 {
671 return mask & FS_Q_OVERFLOW;
672 }
673
fsnotify_notify_queue_is_empty(struct fsnotify_group * group)674 static inline bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group)
675 {
676 assert_spin_locked(&group->notification_lock);
677
678 return list_empty(&group->notification_list);
679 }
680
681 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
682 /* return, but do not dequeue the first event on the notification queue */
683 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
684 /* return AND dequeue the first event on the notification queue */
685 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
686 /* Remove event queued in the notification list */
687 extern void fsnotify_remove_queued_event(struct fsnotify_group *group,
688 struct fsnotify_event *event);
689
690 /* functions used to manipulate the marks attached to inodes */
691
692 /*
693 * Canonical "ignore mask" including event flags.
694 *
695 * Note the subtle semantic difference from the legacy ->ignored_mask.
696 * ->ignored_mask traditionally only meant which events should be ignored,
697 * while ->ignore_mask also includes flags regarding the type of objects on
698 * which events should be ignored.
699 */
fsnotify_ignore_mask(struct fsnotify_mark * mark)700 static inline __u32 fsnotify_ignore_mask(struct fsnotify_mark *mark)
701 {
702 __u32 ignore_mask = mark->ignore_mask;
703
704 /* The event flags in ignore mask take effect */
705 if (mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
706 return ignore_mask;
707
708 /*
709 * Legacy behavior:
710 * - Always ignore events on dir
711 * - Ignore events on child if parent is watching children
712 */
713 ignore_mask |= FS_ISDIR;
714 ignore_mask &= ~FS_EVENT_ON_CHILD;
715 ignore_mask |= mark->mask & FS_EVENT_ON_CHILD;
716
717 return ignore_mask;
718 }
719
720 /* Legacy ignored_mask - only event types to ignore */
fsnotify_ignored_events(struct fsnotify_mark * mark)721 static inline __u32 fsnotify_ignored_events(struct fsnotify_mark *mark)
722 {
723 return mark->ignore_mask & ALL_FSNOTIFY_EVENTS;
724 }
725
726 /*
727 * Check if mask (or ignore mask) should be applied depending if victim is a
728 * directory and whether it is reported to a watching parent.
729 */
fsnotify_mask_applicable(__u32 mask,bool is_dir,int iter_type)730 static inline bool fsnotify_mask_applicable(__u32 mask, bool is_dir,
731 int iter_type)
732 {
733 /* Should mask be applied to a directory? */
734 if (is_dir && !(mask & FS_ISDIR))
735 return false;
736
737 /* Should mask be applied to a child? */
738 if (iter_type == FSNOTIFY_ITER_TYPE_PARENT &&
739 !(mask & FS_EVENT_ON_CHILD))
740 return false;
741
742 return true;
743 }
744
745 /*
746 * Effective ignore mask taking into account if event victim is a
747 * directory and whether it is reported to a watching parent.
748 */
fsnotify_effective_ignore_mask(struct fsnotify_mark * mark,bool is_dir,int iter_type)749 static inline __u32 fsnotify_effective_ignore_mask(struct fsnotify_mark *mark,
750 bool is_dir, int iter_type)
751 {
752 __u32 ignore_mask = fsnotify_ignored_events(mark);
753
754 if (!ignore_mask)
755 return 0;
756
757 /* For non-dir and non-child, no need to consult the event flags */
758 if (!is_dir && iter_type != FSNOTIFY_ITER_TYPE_PARENT)
759 return ignore_mask;
760
761 ignore_mask = fsnotify_ignore_mask(mark);
762 if (!fsnotify_mask_applicable(ignore_mask, is_dir, iter_type))
763 return 0;
764
765 return ignore_mask & ALL_FSNOTIFY_EVENTS;
766 }
767
768 /* Get mask for calculating object interest taking ignore mask into account */
fsnotify_calc_mask(struct fsnotify_mark * mark)769 static inline __u32 fsnotify_calc_mask(struct fsnotify_mark *mark)
770 {
771 __u32 mask = mark->mask;
772
773 if (!fsnotify_ignored_events(mark))
774 return mask;
775
776 /* Interest in FS_MODIFY may be needed for clearing ignore mask */
777 if (!(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
778 mask |= FS_MODIFY;
779
780 /*
781 * If mark is interested in ignoring events on children, the object must
782 * show interest in those events for fsnotify_parent() to notice it.
783 */
784 return mask | mark->ignore_mask;
785 }
786
787 /* Get mask of events for a list of marks */
788 extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn);
789 /* Calculate mask of events for a list of marks */
790 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
791 extern void fsnotify_init_mark(struct fsnotify_mark *mark,
792 struct fsnotify_group *group);
793 /* Find mark belonging to given group in the list of marks */
794 struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type,
795 struct fsnotify_group *group);
796 /* attach the mark to the object */
797 int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj,
798 unsigned int obj_type, int add_flags);
799 int fsnotify_add_mark_locked(struct fsnotify_mark *mark, void *obj,
800 unsigned int obj_type, int add_flags);
801
802 /* attach the mark to the inode */
fsnotify_add_inode_mark(struct fsnotify_mark * mark,struct inode * inode,int add_flags)803 static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
804 struct inode *inode,
805 int add_flags)
806 {
807 return fsnotify_add_mark(mark, inode, FSNOTIFY_OBJ_TYPE_INODE,
808 add_flags);
809 }
fsnotify_add_inode_mark_locked(struct fsnotify_mark * mark,struct inode * inode,int add_flags)810 static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
811 struct inode *inode,
812 int add_flags)
813 {
814 return fsnotify_add_mark_locked(mark, inode, FSNOTIFY_OBJ_TYPE_INODE,
815 add_flags);
816 }
817
fsnotify_find_inode_mark(struct inode * inode,struct fsnotify_group * group)818 static inline struct fsnotify_mark *fsnotify_find_inode_mark(
819 struct inode *inode,
820 struct fsnotify_group *group)
821 {
822 return fsnotify_find_mark(inode, FSNOTIFY_OBJ_TYPE_INODE, group);
823 }
824
825 /* given a group and a mark, flag mark to be freed when all references are dropped */
826 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
827 struct fsnotify_group *group);
828 /* detach mark from inode / mount list, group list, drop inode reference */
829 extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
830 /* free mark */
831 extern void fsnotify_free_mark(struct fsnotify_mark *mark);
832 /* Wait until all marks queued for destruction are destroyed */
833 extern void fsnotify_wait_marks_destroyed(void);
834 /* Clear all of the marks of a group attached to a given object type */
835 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
836 unsigned int obj_type);
837 /* run all the marks in a group, and clear all of the vfsmount marks */
fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group * group)838 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
839 {
840 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT);
841 }
842 /* run all the marks in a group, and clear all of the inode marks */
fsnotify_clear_inode_marks_by_group(struct fsnotify_group * group)843 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
844 {
845 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE);
846 }
847 /* run all the marks in a group, and clear all of the sn marks */
fsnotify_clear_sb_marks_by_group(struct fsnotify_group * group)848 static inline void fsnotify_clear_sb_marks_by_group(struct fsnotify_group *group)
849 {
850 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_SB);
851 }
852 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
853 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
854 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
855 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
856
fsnotify_init_event(struct fsnotify_event * event)857 static inline void fsnotify_init_event(struct fsnotify_event *event)
858 {
859 INIT_LIST_HEAD(&event->list);
860 }
861
862 #else
863
fsnotify(__u32 mask,const void * data,int data_type,struct inode * dir,const struct qstr * name,struct inode * inode,u32 cookie)864 static inline int fsnotify(__u32 mask, const void *data, int data_type,
865 struct inode *dir, const struct qstr *name,
866 struct inode *inode, u32 cookie)
867 {
868 return 0;
869 }
870
__fsnotify_parent(struct dentry * dentry,__u32 mask,const void * data,int data_type)871 static inline int __fsnotify_parent(struct dentry *dentry, __u32 mask,
872 const void *data, int data_type)
873 {
874 return 0;
875 }
876
__fsnotify_inode_delete(struct inode * inode)877 static inline void __fsnotify_inode_delete(struct inode *inode)
878 {}
879
__fsnotify_vfsmount_delete(struct vfsmount * mnt)880 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
881 {}
882
fsnotify_sb_delete(struct super_block * sb)883 static inline void fsnotify_sb_delete(struct super_block *sb)
884 {}
885
fsnotify_sb_free(struct super_block * sb)886 static inline void fsnotify_sb_free(struct super_block *sb)
887 {}
888
fsnotify_update_flags(struct dentry * dentry)889 static inline void fsnotify_update_flags(struct dentry *dentry)
890 {}
891
fsnotify_get_cookie(void)892 static inline u32 fsnotify_get_cookie(void)
893 {
894 return 0;
895 }
896
fsnotify_unmount_inodes(struct super_block * sb)897 static inline void fsnotify_unmount_inodes(struct super_block *sb)
898 {}
899
900 #endif /* CONFIG_FSNOTIFY */
901
902 #endif /* __KERNEL __ */
903
904 #endif /* __LINUX_FSNOTIFY_BACKEND_H */
905