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 | /* #define FS_DIR_MODIFY 0x00080000 */ /* Deprecated (reserved) */ |
59 | |
60 | #define FS_PRE_ACCESS 0x00100000 /* Pre-content access hook */ |
61 | |
62 | #define FS_MNT_ATTACH 0x01000000 /* Mount was attached */ |
63 | #define FS_MNT_DETACH 0x02000000 /* Mount was detached */ |
64 | #define FS_MNT_MOVE (FS_MNT_ATTACH | FS_MNT_DETACH) |
65 | |
66 | /* |
67 | * Set on inode mark that cares about things that happen to its children. |
68 | * Always set for dnotify and inotify. |
69 | * Set on inode/sb/mount marks that care about parent/name info. |
70 | */ |
71 | #define FS_EVENT_ON_CHILD 0x08000000 |
72 | |
73 | #define FS_RENAME 0x10000000 /* File was renamed */ |
74 | #define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */ |
75 | #define FS_ISDIR 0x40000000 /* event occurred against dir */ |
76 | |
77 | #define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO) |
78 | |
79 | /* |
80 | * Directory entry modification events - reported only to directory |
81 | * where entry is modified and not to a watching parent. |
82 | * The watching parent may get an FS_ATTRIB|FS_EVENT_ON_CHILD event |
83 | * when a directory entry inside a child subdir changes. |
84 | */ |
85 | #define ALL_FSNOTIFY_DIRENT_EVENTS (FS_CREATE | FS_DELETE | FS_MOVE | FS_RENAME) |
86 | |
87 | /* Mount namespace events */ |
88 | #define FSNOTIFY_MNT_EVENTS (FS_MNT_ATTACH | FS_MNT_DETACH) |
89 | |
90 | /* Content events can be used to inspect file content */ |
91 | #define FSNOTIFY_CONTENT_PERM_EVENTS (FS_OPEN_PERM | FS_OPEN_EXEC_PERM | \ |
92 | FS_ACCESS_PERM) |
93 | /* Pre-content events can be used to fill file content */ |
94 | #define FSNOTIFY_PRE_CONTENT_EVENTS (FS_PRE_ACCESS) |
95 | |
96 | #define ALL_FSNOTIFY_PERM_EVENTS (FSNOTIFY_CONTENT_PERM_EVENTS | \ |
97 | FSNOTIFY_PRE_CONTENT_EVENTS) |
98 | |
99 | /* |
100 | * This is a list of all events that may get sent to a parent that is watching |
101 | * with flag FS_EVENT_ON_CHILD based on fs event on a child of that directory. |
102 | */ |
103 | #define FS_EVENTS_POSS_ON_CHILD (ALL_FSNOTIFY_PERM_EVENTS | \ |
104 | FS_ACCESS | FS_MODIFY | FS_ATTRIB | \ |
105 | FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | \ |
106 | FS_OPEN | FS_OPEN_EXEC) |
107 | |
108 | /* |
109 | * This is a list of all events that may get sent with the parent inode as the |
110 | * @to_tell argument of fsnotify(). |
111 | * It may include events that can be sent to an inode/sb/mount mark, but cannot |
112 | * be sent to a parent watching children. |
113 | */ |
114 | #define FS_EVENTS_POSS_TO_PARENT (FS_EVENTS_POSS_ON_CHILD) |
115 | |
116 | /* Events that can be reported to backends */ |
117 | #define ALL_FSNOTIFY_EVENTS (ALL_FSNOTIFY_DIRENT_EVENTS | \ |
118 | FSNOTIFY_MNT_EVENTS | \ |
119 | FS_EVENTS_POSS_ON_CHILD | \ |
120 | FS_DELETE_SELF | FS_MOVE_SELF | \ |
121 | FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \ |
122 | FS_ERROR) |
123 | |
124 | /* Extra flags that may be reported with event or control handling of events */ |
125 | #define ALL_FSNOTIFY_FLAGS (FS_ISDIR | FS_EVENT_ON_CHILD | FS_DN_MULTISHOT) |
126 | |
127 | #define ALL_FSNOTIFY_BITS (ALL_FSNOTIFY_EVENTS | ALL_FSNOTIFY_FLAGS) |
128 | |
129 | struct fsnotify_group; |
130 | struct fsnotify_event; |
131 | struct fsnotify_mark; |
132 | struct fsnotify_event_private_data; |
133 | struct fsnotify_fname; |
134 | struct fsnotify_iter_info; |
135 | |
136 | struct mem_cgroup; |
137 | |
138 | /* |
139 | * Each group much define these ops. The fsnotify infrastructure will call |
140 | * these operations for each relevant group. |
141 | * |
142 | * handle_event - main call for a group to handle an fs event |
143 | * @group: group to notify |
144 | * @mask: event type and flags |
145 | * @data: object that event happened on |
146 | * @data_type: type of object for fanotify_data_XXX() accessors |
147 | * @dir: optional directory associated with event - |
148 | * if @file_name is not NULL, this is the directory that |
149 | * @file_name is relative to |
150 | * @file_name: optional file name associated with event |
151 | * @cookie: inotify rename cookie |
152 | * @iter_info: array of marks from this group that are interested in the event |
153 | * |
154 | * handle_inode_event - simple variant of handle_event() for groups that only |
155 | * have inode marks and don't have ignore mask |
156 | * @mark: mark to notify |
157 | * @mask: event type and flags |
158 | * @inode: inode that event happened on |
159 | * @dir: optional directory associated with event - |
160 | * if @file_name is not NULL, this is the directory that |
161 | * @file_name is relative to. |
162 | * Either @inode or @dir must be non-NULL. |
163 | * @file_name: optional file name associated with event |
164 | * @cookie: inotify rename cookie |
165 | * |
166 | * free_group_priv - called when a group refcnt hits 0 to clean up the private union |
167 | * freeing_mark - called when a mark is being destroyed for some reason. The group |
168 | * MUST be holding a reference on each mark and that reference must be |
169 | * dropped in this function. inotify uses this function to send |
170 | * userspace messages that marks have been removed. |
171 | */ |
172 | struct fsnotify_ops { |
173 | int (*handle_event)(struct fsnotify_group *group, u32 mask, |
174 | const void *data, int data_type, struct inode *dir, |
175 | const struct qstr *file_name, u32 cookie, |
176 | struct fsnotify_iter_info *iter_info); |
177 | int (*handle_inode_event)(struct fsnotify_mark *mark, u32 mask, |
178 | struct inode *inode, struct inode *dir, |
179 | const struct qstr *file_name, u32 cookie); |
180 | void (*free_group_priv)(struct fsnotify_group *group); |
181 | void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group); |
182 | void (*free_event)(struct fsnotify_group *group, struct fsnotify_event *event); |
183 | /* called on final put+free to free memory */ |
184 | void (*free_mark)(struct fsnotify_mark *mark); |
185 | }; |
186 | |
187 | /* |
188 | * all of the information about the original object we want to now send to |
189 | * a group. If you want to carry more info from the accessing task to the |
190 | * listener this structure is where you need to be adding fields. |
191 | */ |
192 | struct fsnotify_event { |
193 | struct list_head list; |
194 | }; |
195 | |
196 | /* |
197 | * fsnotify group priorities. |
198 | * Events are sent in order from highest priority to lowest priority. |
199 | */ |
200 | enum fsnotify_group_prio { |
201 | FSNOTIFY_PRIO_NORMAL = 0, /* normal notifiers, no permissions */ |
202 | FSNOTIFY_PRIO_CONTENT, /* fanotify permission events */ |
203 | FSNOTIFY_PRIO_PRE_CONTENT, /* fanotify pre-content events */ |
204 | __FSNOTIFY_PRIO_NUM |
205 | }; |
206 | |
207 | /* |
208 | * A group is a "thing" that wants to receive notification about filesystem |
209 | * events. The mask holds the subset of event types this group cares about. |
210 | * refcnt on a group is up to the implementor and at any moment if it goes 0 |
211 | * everything will be cleaned up. |
212 | */ |
213 | struct fsnotify_group { |
214 | const struct fsnotify_ops *ops; /* how this group handles things */ |
215 | |
216 | /* |
217 | * How the refcnt is used is up to each group. When the refcnt hits 0 |
218 | * fsnotify will clean up all of the resources associated with this group. |
219 | * As an example, the dnotify group will always have a refcnt=1 and that |
220 | * will never change. Inotify, on the other hand, has a group per |
221 | * inotify_init() and the refcnt will hit 0 only when that fd has been |
222 | * closed. |
223 | */ |
224 | refcount_t refcnt; /* things with interest in this group */ |
225 | |
226 | /* needed to send notification to userspace */ |
227 | spinlock_t notification_lock; /* protect the notification_list */ |
228 | struct list_head notification_list; /* list of event_holder this group needs to send to userspace */ |
229 | wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */ |
230 | unsigned int q_len; /* events on the queue */ |
231 | unsigned int max_events; /* maximum events allowed on the list */ |
232 | enum fsnotify_group_prio priority; /* priority for sending events */ |
233 | bool shutdown; /* group is being shut down, don't queue more events */ |
234 | |
235 | #define FSNOTIFY_GROUP_USER 0x01 /* user allocated group */ |
236 | #define FSNOTIFY_GROUP_DUPS 0x02 /* allow multiple marks per object */ |
237 | int flags; |
238 | unsigned int owner_flags; /* stored flags of mark_mutex owner */ |
239 | |
240 | /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */ |
241 | struct mutex mark_mutex; /* protect marks_list */ |
242 | atomic_t user_waits; /* Number of tasks waiting for user |
243 | * response */ |
244 | struct list_head marks_list; /* all inode marks for this group */ |
245 | |
246 | struct fasync_struct *fsn_fa; /* async notification */ |
247 | |
248 | struct fsnotify_event *overflow_event; /* Event we queue when the |
249 | * notification list is too |
250 | * full */ |
251 | |
252 | struct mem_cgroup *memcg; /* memcg to charge allocations */ |
253 | struct user_namespace *user_ns; /* user ns where group was created */ |
254 | |
255 | /* groups can define private fields here or use the void *private */ |
256 | union { |
257 | void *private; |
258 | #ifdef CONFIG_INOTIFY_USER |
259 | struct inotify_group_private_data { |
260 | spinlock_t idr_lock; |
261 | struct idr idr; |
262 | struct ucounts *ucounts; |
263 | } inotify_data; |
264 | #endif |
265 | #ifdef CONFIG_FANOTIFY |
266 | struct fanotify_group_private_data { |
267 | /* Hash table of events for merge */ |
268 | struct hlist_head *merge_hash; |
269 | /* allows a group to block waiting for a userspace response */ |
270 | struct list_head access_list; |
271 | wait_queue_head_t access_waitq; |
272 | int flags; /* flags from fanotify_init() */ |
273 | int f_flags; /* event_f_flags from fanotify_init() */ |
274 | struct ucounts *ucounts; |
275 | mempool_t error_events_pool; |
276 | } fanotify_data; |
277 | #endif /* CONFIG_FANOTIFY */ |
278 | }; |
279 | }; |
280 | |
281 | /* |
282 | * These helpers are used to prevent deadlock when reclaiming inodes with |
283 | * evictable marks of the same group that is allocating a new mark. |
284 | */ |
285 | static inline void fsnotify_group_lock(struct fsnotify_group *group) |
286 | { |
287 | mutex_lock(&group->mark_mutex); |
288 | group->owner_flags = memalloc_nofs_save(); |
289 | } |
290 | |
291 | static inline void fsnotify_group_unlock(struct fsnotify_group *group) |
292 | { |
293 | memalloc_nofs_restore(flags: group->owner_flags); |
294 | mutex_unlock(lock: &group->mark_mutex); |
295 | } |
296 | |
297 | static inline void fsnotify_group_assert_locked(struct fsnotify_group *group) |
298 | { |
299 | WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex)); |
300 | WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS)); |
301 | } |
302 | |
303 | /* When calling fsnotify tell it if the data is a path or inode */ |
304 | enum fsnotify_data_type { |
305 | FSNOTIFY_EVENT_NONE, |
306 | FSNOTIFY_EVENT_FILE_RANGE, |
307 | FSNOTIFY_EVENT_PATH, |
308 | FSNOTIFY_EVENT_INODE, |
309 | FSNOTIFY_EVENT_DENTRY, |
310 | FSNOTIFY_EVENT_MNT, |
311 | FSNOTIFY_EVENT_ERROR, |
312 | }; |
313 | |
314 | struct fs_error_report { |
315 | int error; |
316 | struct inode *inode; |
317 | struct super_block *sb; |
318 | }; |
319 | |
320 | struct file_range { |
321 | const struct path *path; |
322 | loff_t pos; |
323 | size_t count; |
324 | }; |
325 | |
326 | static inline const struct path *file_range_path(const struct file_range *range) |
327 | { |
328 | return range->path; |
329 | } |
330 | |
331 | struct fsnotify_mnt { |
332 | const struct mnt_namespace *ns; |
333 | u64 mnt_id; |
334 | }; |
335 | |
336 | static inline struct inode *fsnotify_data_inode(const void *data, int data_type) |
337 | { |
338 | switch (data_type) { |
339 | case FSNOTIFY_EVENT_INODE: |
340 | return (struct inode *)data; |
341 | case FSNOTIFY_EVENT_DENTRY: |
342 | return d_inode(dentry: data); |
343 | case FSNOTIFY_EVENT_PATH: |
344 | return d_inode(dentry: ((const struct path *)data)->dentry); |
345 | case FSNOTIFY_EVENT_FILE_RANGE: |
346 | return d_inode(dentry: file_range_path(range: data)->dentry); |
347 | case FSNOTIFY_EVENT_ERROR: |
348 | return ((struct fs_error_report *)data)->inode; |
349 | default: |
350 | return NULL; |
351 | } |
352 | } |
353 | |
354 | static inline struct dentry *fsnotify_data_dentry(const void *data, int data_type) |
355 | { |
356 | switch (data_type) { |
357 | case FSNOTIFY_EVENT_DENTRY: |
358 | /* Non const is needed for dget() */ |
359 | return (struct dentry *)data; |
360 | case FSNOTIFY_EVENT_PATH: |
361 | return ((const struct path *)data)->dentry; |
362 | case FSNOTIFY_EVENT_FILE_RANGE: |
363 | return file_range_path(range: data)->dentry; |
364 | default: |
365 | return NULL; |
366 | } |
367 | } |
368 | |
369 | static inline const struct path *fsnotify_data_path(const void *data, |
370 | int data_type) |
371 | { |
372 | switch (data_type) { |
373 | case FSNOTIFY_EVENT_PATH: |
374 | return data; |
375 | case FSNOTIFY_EVENT_FILE_RANGE: |
376 | return file_range_path(range: data); |
377 | default: |
378 | return NULL; |
379 | } |
380 | } |
381 | |
382 | static inline struct super_block *fsnotify_data_sb(const void *data, |
383 | int data_type) |
384 | { |
385 | switch (data_type) { |
386 | case FSNOTIFY_EVENT_INODE: |
387 | return ((struct inode *)data)->i_sb; |
388 | case FSNOTIFY_EVENT_DENTRY: |
389 | return ((struct dentry *)data)->d_sb; |
390 | case FSNOTIFY_EVENT_PATH: |
391 | return ((const struct path *)data)->dentry->d_sb; |
392 | case FSNOTIFY_EVENT_FILE_RANGE: |
393 | return file_range_path(range: data)->dentry->d_sb; |
394 | case FSNOTIFY_EVENT_ERROR: |
395 | return ((struct fs_error_report *) data)->sb; |
396 | default: |
397 | return NULL; |
398 | } |
399 | } |
400 | |
401 | static inline const struct fsnotify_mnt *fsnotify_data_mnt(const void *data, |
402 | int data_type) |
403 | { |
404 | switch (data_type) { |
405 | case FSNOTIFY_EVENT_MNT: |
406 | return data; |
407 | default: |
408 | return NULL; |
409 | } |
410 | } |
411 | |
412 | static inline u64 fsnotify_data_mnt_id(const void *data, int data_type) |
413 | { |
414 | const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type); |
415 | |
416 | return mnt_data ? mnt_data->mnt_id : 0; |
417 | } |
418 | |
419 | static inline struct fs_error_report *fsnotify_data_error_report( |
420 | const void *data, |
421 | int data_type) |
422 | { |
423 | switch (data_type) { |
424 | case FSNOTIFY_EVENT_ERROR: |
425 | return (struct fs_error_report *) data; |
426 | default: |
427 | return NULL; |
428 | } |
429 | } |
430 | |
431 | static inline const struct file_range *fsnotify_data_file_range( |
432 | const void *data, |
433 | int data_type) |
434 | { |
435 | switch (data_type) { |
436 | case FSNOTIFY_EVENT_FILE_RANGE: |
437 | return (struct file_range *)data; |
438 | default: |
439 | return NULL; |
440 | } |
441 | } |
442 | |
443 | /* |
444 | * Index to merged marks iterator array that correlates to a type of watch. |
445 | * The type of watched object can be deduced from the iterator type, but not |
446 | * the other way around, because an event can match different watched objects |
447 | * of the same object type. |
448 | * For example, both parent and child are watching an object of type inode. |
449 | */ |
450 | enum fsnotify_iter_type { |
451 | FSNOTIFY_ITER_TYPE_INODE, |
452 | FSNOTIFY_ITER_TYPE_VFSMOUNT, |
453 | FSNOTIFY_ITER_TYPE_SB, |
454 | FSNOTIFY_ITER_TYPE_PARENT, |
455 | FSNOTIFY_ITER_TYPE_INODE2, |
456 | FSNOTIFY_ITER_TYPE_MNTNS, |
457 | FSNOTIFY_ITER_TYPE_COUNT |
458 | }; |
459 | |
460 | /* The type of object that a mark is attached to */ |
461 | enum fsnotify_obj_type { |
462 | FSNOTIFY_OBJ_TYPE_ANY = -1, |
463 | FSNOTIFY_OBJ_TYPE_INODE, |
464 | FSNOTIFY_OBJ_TYPE_VFSMOUNT, |
465 | FSNOTIFY_OBJ_TYPE_SB, |
466 | FSNOTIFY_OBJ_TYPE_MNTNS, |
467 | FSNOTIFY_OBJ_TYPE_COUNT, |
468 | FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT |
469 | }; |
470 | |
471 | static inline bool fsnotify_valid_obj_type(unsigned int obj_type) |
472 | { |
473 | return (obj_type < FSNOTIFY_OBJ_TYPE_COUNT); |
474 | } |
475 | |
476 | struct fsnotify_iter_info { |
477 | struct fsnotify_mark *marks[FSNOTIFY_ITER_TYPE_COUNT]; |
478 | struct fsnotify_group *current_group; |
479 | unsigned int report_mask; |
480 | int srcu_idx; |
481 | }; |
482 | |
483 | static inline bool fsnotify_iter_should_report_type( |
484 | struct fsnotify_iter_info *iter_info, int iter_type) |
485 | { |
486 | return (iter_info->report_mask & (1U << iter_type)); |
487 | } |
488 | |
489 | static inline void fsnotify_iter_set_report_type( |
490 | struct fsnotify_iter_info *iter_info, int iter_type) |
491 | { |
492 | iter_info->report_mask |= (1U << iter_type); |
493 | } |
494 | |
495 | static inline struct fsnotify_mark *fsnotify_iter_mark( |
496 | struct fsnotify_iter_info *iter_info, int iter_type) |
497 | { |
498 | if (fsnotify_iter_should_report_type(iter_info, iter_type)) |
499 | return iter_info->marks[iter_type]; |
500 | return NULL; |
501 | } |
502 | |
503 | static inline int fsnotify_iter_step(struct fsnotify_iter_info *iter, int type, |
504 | struct fsnotify_mark **markp) |
505 | { |
506 | while (type < FSNOTIFY_ITER_TYPE_COUNT) { |
507 | *markp = fsnotify_iter_mark(iter_info: iter, iter_type: type); |
508 | if (*markp) |
509 | break; |
510 | type++; |
511 | } |
512 | return type; |
513 | } |
514 | |
515 | #define FSNOTIFY_ITER_FUNCS(name, NAME) \ |
516 | static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \ |
517 | struct fsnotify_iter_info *iter_info) \ |
518 | { \ |
519 | return fsnotify_iter_mark(iter_info, FSNOTIFY_ITER_TYPE_##NAME); \ |
520 | } |
521 | |
522 | FSNOTIFY_ITER_FUNCS(inode, INODE) |
523 | FSNOTIFY_ITER_FUNCS(parent, PARENT) |
524 | FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT) |
525 | FSNOTIFY_ITER_FUNCS(sb, SB) |
526 | |
527 | #define fsnotify_foreach_iter_type(type) \ |
528 | for (type = 0; type < FSNOTIFY_ITER_TYPE_COUNT; type++) |
529 | #define fsnotify_foreach_iter_mark_type(iter, mark, type) \ |
530 | for (type = 0; \ |
531 | type = fsnotify_iter_step(iter, type, &mark), \ |
532 | type < FSNOTIFY_ITER_TYPE_COUNT; \ |
533 | type++) |
534 | |
535 | /* |
536 | * Inode/vfsmount/sb point to this structure which tracks all marks attached to |
537 | * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this |
538 | * structure. We destroy this structure when there are no more marks attached |
539 | * to it. The structure is protected by fsnotify_mark_srcu. |
540 | */ |
541 | struct fsnotify_mark_connector { |
542 | spinlock_t lock; |
543 | unsigned char type; /* Type of object [lock] */ |
544 | unsigned char prio; /* Highest priority group */ |
545 | #define FSNOTIFY_CONN_FLAG_IS_WATCHED 0x01 |
546 | #define FSNOTIFY_CONN_FLAG_HAS_IREF 0x02 |
547 | unsigned short flags; /* flags [lock] */ |
548 | union { |
549 | /* Object pointer [lock] */ |
550 | void *obj; |
551 | /* Used listing heads to free after srcu period expires */ |
552 | struct fsnotify_mark_connector *destroy_next; |
553 | }; |
554 | struct hlist_head list; |
555 | }; |
556 | |
557 | /* |
558 | * Container for per-sb fsnotify state (sb marks and more). |
559 | * Attached lazily on first marked object on the sb and freed when killing sb. |
560 | */ |
561 | struct fsnotify_sb_info { |
562 | struct fsnotify_mark_connector __rcu *sb_marks; |
563 | /* |
564 | * Number of inode/mount/sb objects that are being watched in this sb. |
565 | * Note that inodes objects are currently double-accounted. |
566 | * |
567 | * The value in watched_objects[prio] is the number of objects that are |
568 | * watched by groups of priority >= prio, so watched_objects[0] is the |
569 | * total number of watched objects in this sb. |
570 | */ |
571 | atomic_long_t watched_objects[__FSNOTIFY_PRIO_NUM]; |
572 | }; |
573 | |
574 | static inline struct fsnotify_sb_info *fsnotify_sb_info(struct super_block *sb) |
575 | { |
576 | #ifdef CONFIG_FSNOTIFY |
577 | return READ_ONCE(sb->s_fsnotify_info); |
578 | #else |
579 | return NULL; |
580 | #endif |
581 | } |
582 | |
583 | static inline atomic_long_t *fsnotify_sb_watched_objects(struct super_block *sb) |
584 | { |
585 | return &fsnotify_sb_info(sb)->watched_objects[0]; |
586 | } |
587 | |
588 | /* |
589 | * A mark is simply an object attached to an in core inode which allows an |
590 | * fsnotify listener to indicate they are either no longer interested in events |
591 | * of a type matching mask or only interested in those events. |
592 | * |
593 | * These are flushed when an inode is evicted from core and may be flushed |
594 | * when the inode is modified (as seen by fsnotify_access). Some fsnotify |
595 | * users (such as dnotify) will flush these when the open fd is closed and not |
596 | * at inode eviction or modification. |
597 | * |
598 | * Text in brackets is showing the lock(s) protecting modifications of a |
599 | * particular entry. obj_lock means either inode->i_lock or |
600 | * mnt->mnt_root->d_lock depending on the mark type. |
601 | */ |
602 | struct fsnotify_mark { |
603 | /* Mask this mark is for [mark->lock, group->mark_mutex] */ |
604 | __u32 mask; |
605 | /* We hold one for presence in g_list. Also one ref for each 'thing' |
606 | * in kernel that found and may be using this mark. */ |
607 | refcount_t refcnt; |
608 | /* Group this mark is for. Set on mark creation, stable until last ref |
609 | * is dropped */ |
610 | struct fsnotify_group *group; |
611 | /* List of marks by group->marks_list. Also reused for queueing |
612 | * mark into destroy_list when it's waiting for the end of SRCU period |
613 | * before it can be freed. [group->mark_mutex] */ |
614 | struct list_head g_list; |
615 | /* Protects inode / mnt pointers, flags, masks */ |
616 | spinlock_t lock; |
617 | /* List of marks for inode / vfsmount [connector->lock, mark ref] */ |
618 | struct hlist_node obj_list; |
619 | /* Head of list of marks for an object [mark ref] */ |
620 | struct fsnotify_mark_connector *connector; |
621 | /* Events types and flags to ignore [mark->lock, group->mark_mutex] */ |
622 | __u32 ignore_mask; |
623 | /* General fsnotify mark flags */ |
624 | #define FSNOTIFY_MARK_FLAG_ALIVE 0x0001 |
625 | #define FSNOTIFY_MARK_FLAG_ATTACHED 0x0002 |
626 | /* inotify mark flags */ |
627 | #define FSNOTIFY_MARK_FLAG_EXCL_UNLINK 0x0010 |
628 | #define FSNOTIFY_MARK_FLAG_IN_ONESHOT 0x0020 |
629 | /* fanotify mark flags */ |
630 | #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x0100 |
631 | #define FSNOTIFY_MARK_FLAG_NO_IREF 0x0200 |
632 | #define FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS 0x0400 |
633 | #define FSNOTIFY_MARK_FLAG_HAS_FSID 0x0800 |
634 | #define FSNOTIFY_MARK_FLAG_WEAK_FSID 0x1000 |
635 | unsigned int flags; /* flags [mark->lock] */ |
636 | }; |
637 | |
638 | #ifdef CONFIG_FSNOTIFY |
639 | |
640 | /* called from the vfs helpers */ |
641 | |
642 | /* main fsnotify call to send events */ |
643 | extern int fsnotify(__u32 mask, const void *data, int data_type, |
644 | struct inode *dir, const struct qstr *name, |
645 | struct inode *inode, u32 cookie); |
646 | extern int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data, |
647 | int data_type); |
648 | extern void __fsnotify_inode_delete(struct inode *inode); |
649 | extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt); |
650 | extern void fsnotify_sb_delete(struct super_block *sb); |
651 | extern void __fsnotify_mntns_delete(struct mnt_namespace *mntns); |
652 | extern void fsnotify_sb_free(struct super_block *sb); |
653 | extern u32 fsnotify_get_cookie(void); |
654 | extern void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt); |
655 | |
656 | static inline __u32 fsnotify_parent_needed_mask(__u32 mask) |
657 | { |
658 | /* FS_EVENT_ON_CHILD is set on marks that want parent/name info */ |
659 | if (!(mask & FS_EVENT_ON_CHILD)) |
660 | return 0; |
661 | /* |
662 | * This object might be watched by a mark that cares about parent/name |
663 | * info, does it care about the specific set of events that can be |
664 | * reported with parent/name info? |
665 | */ |
666 | return mask & FS_EVENTS_POSS_TO_PARENT; |
667 | } |
668 | |
669 | static inline int fsnotify_inode_watches_children(struct inode *inode) |
670 | { |
671 | __u32 parent_mask = READ_ONCE(inode->i_fsnotify_mask); |
672 | |
673 | /* FS_EVENT_ON_CHILD is set if the inode may care */ |
674 | if (!(parent_mask & FS_EVENT_ON_CHILD)) |
675 | return 0; |
676 | /* this inode might care about child events, does it care about the |
677 | * specific set of events that can happen on a child? */ |
678 | return parent_mask & FS_EVENTS_POSS_ON_CHILD; |
679 | } |
680 | |
681 | /* |
682 | * Update the dentry with a flag indicating the interest of its parent to receive |
683 | * filesystem events when those events happens to this dentry->d_inode. |
684 | */ |
685 | static inline void fsnotify_update_flags(struct dentry *dentry) |
686 | { |
687 | assert_spin_locked(&dentry->d_lock); |
688 | |
689 | /* |
690 | * Serialisation of setting PARENT_WATCHED on the dentries is provided |
691 | * by d_lock. If inotify_inode_watched changes after we have taken |
692 | * d_lock, the following fsnotify_set_children_dentry_flags call will |
693 | * find our entry, so it will spin until we complete here, and update |
694 | * us with the new state. |
695 | */ |
696 | if (fsnotify_inode_watches_children(inode: dentry->d_parent->d_inode)) |
697 | dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; |
698 | else |
699 | dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; |
700 | } |
701 | |
702 | /* called from fsnotify listeners, such as fanotify or dnotify */ |
703 | |
704 | /* create a new group */ |
705 | extern struct fsnotify_group *fsnotify_alloc_group( |
706 | const struct fsnotify_ops *ops, |
707 | int flags); |
708 | /* get reference to a group */ |
709 | extern void fsnotify_get_group(struct fsnotify_group *group); |
710 | /* drop reference on a group from fsnotify_alloc_group */ |
711 | extern void fsnotify_put_group(struct fsnotify_group *group); |
712 | /* group destruction begins, stop queuing new events */ |
713 | extern void fsnotify_group_stop_queueing(struct fsnotify_group *group); |
714 | /* destroy group */ |
715 | extern void fsnotify_destroy_group(struct fsnotify_group *group); |
716 | /* fasync handler function */ |
717 | extern int fsnotify_fasync(int fd, struct file *file, int on); |
718 | /* Free event from memory */ |
719 | extern void fsnotify_destroy_event(struct fsnotify_group *group, |
720 | struct fsnotify_event *event); |
721 | /* attach the event to the group notification queue */ |
722 | extern int fsnotify_insert_event(struct fsnotify_group *group, |
723 | struct fsnotify_event *event, |
724 | int (*merge)(struct fsnotify_group *, |
725 | struct fsnotify_event *), |
726 | void (*insert)(struct fsnotify_group *, |
727 | struct fsnotify_event *)); |
728 | |
729 | static inline int fsnotify_add_event(struct fsnotify_group *group, |
730 | struct fsnotify_event *event, |
731 | int (*merge)(struct fsnotify_group *, |
732 | struct fsnotify_event *)) |
733 | { |
734 | return fsnotify_insert_event(group, event, merge, NULL); |
735 | } |
736 | |
737 | /* Queue overflow event to a notification group */ |
738 | static inline void fsnotify_queue_overflow(struct fsnotify_group *group) |
739 | { |
740 | fsnotify_add_event(group, event: group->overflow_event, NULL); |
741 | } |
742 | |
743 | static inline bool fsnotify_is_overflow_event(u32 mask) |
744 | { |
745 | return mask & FS_Q_OVERFLOW; |
746 | } |
747 | |
748 | static inline bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group) |
749 | { |
750 | assert_spin_locked(&group->notification_lock); |
751 | |
752 | return list_empty(head: &group->notification_list); |
753 | } |
754 | |
755 | extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group); |
756 | /* return, but do not dequeue the first event on the notification queue */ |
757 | extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group); |
758 | /* return AND dequeue the first event on the notification queue */ |
759 | extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group); |
760 | /* Remove event queued in the notification list */ |
761 | extern void fsnotify_remove_queued_event(struct fsnotify_group *group, |
762 | struct fsnotify_event *event); |
763 | |
764 | /* functions used to manipulate the marks attached to inodes */ |
765 | |
766 | /* |
767 | * Canonical "ignore mask" including event flags. |
768 | * |
769 | * Note the subtle semantic difference from the legacy ->ignored_mask. |
770 | * ->ignored_mask traditionally only meant which events should be ignored, |
771 | * while ->ignore_mask also includes flags regarding the type of objects on |
772 | * which events should be ignored. |
773 | */ |
774 | static inline __u32 fsnotify_ignore_mask(struct fsnotify_mark *mark) |
775 | { |
776 | __u32 ignore_mask = mark->ignore_mask; |
777 | |
778 | /* The event flags in ignore mask take effect */ |
779 | if (mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS) |
780 | return ignore_mask; |
781 | |
782 | /* |
783 | * Legacy behavior: |
784 | * - Always ignore events on dir |
785 | * - Ignore events on child if parent is watching children |
786 | */ |
787 | ignore_mask |= FS_ISDIR; |
788 | ignore_mask &= ~FS_EVENT_ON_CHILD; |
789 | ignore_mask |= mark->mask & FS_EVENT_ON_CHILD; |
790 | |
791 | return ignore_mask; |
792 | } |
793 | |
794 | /* Legacy ignored_mask - only event types to ignore */ |
795 | static inline __u32 fsnotify_ignored_events(struct fsnotify_mark *mark) |
796 | { |
797 | return mark->ignore_mask & ALL_FSNOTIFY_EVENTS; |
798 | } |
799 | |
800 | /* |
801 | * Check if mask (or ignore mask) should be applied depending if victim is a |
802 | * directory and whether it is reported to a watching parent. |
803 | */ |
804 | static inline bool fsnotify_mask_applicable(__u32 mask, bool is_dir, |
805 | int iter_type) |
806 | { |
807 | /* Should mask be applied to a directory? */ |
808 | if (is_dir && !(mask & FS_ISDIR)) |
809 | return false; |
810 | |
811 | /* Should mask be applied to a child? */ |
812 | if (iter_type == FSNOTIFY_ITER_TYPE_PARENT && |
813 | !(mask & FS_EVENT_ON_CHILD)) |
814 | return false; |
815 | |
816 | return true; |
817 | } |
818 | |
819 | /* |
820 | * Effective ignore mask taking into account if event victim is a |
821 | * directory and whether it is reported to a watching parent. |
822 | */ |
823 | static inline __u32 fsnotify_effective_ignore_mask(struct fsnotify_mark *mark, |
824 | bool is_dir, int iter_type) |
825 | { |
826 | __u32 ignore_mask = fsnotify_ignored_events(mark); |
827 | |
828 | if (!ignore_mask) |
829 | return 0; |
830 | |
831 | /* For non-dir and non-child, no need to consult the event flags */ |
832 | if (!is_dir && iter_type != FSNOTIFY_ITER_TYPE_PARENT) |
833 | return ignore_mask; |
834 | |
835 | ignore_mask = fsnotify_ignore_mask(mark); |
836 | if (!fsnotify_mask_applicable(mask: ignore_mask, is_dir, iter_type)) |
837 | return 0; |
838 | |
839 | return ignore_mask & ALL_FSNOTIFY_EVENTS; |
840 | } |
841 | |
842 | /* Get mask for calculating object interest taking ignore mask into account */ |
843 | static inline __u32 fsnotify_calc_mask(struct fsnotify_mark *mark) |
844 | { |
845 | __u32 mask = mark->mask; |
846 | |
847 | if (!fsnotify_ignored_events(mark)) |
848 | return mask; |
849 | |
850 | /* Interest in FS_MODIFY may be needed for clearing ignore mask */ |
851 | if (!(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) |
852 | mask |= FS_MODIFY; |
853 | |
854 | /* |
855 | * If mark is interested in ignoring events on children, the object must |
856 | * show interest in those events for fsnotify_parent() to notice it. |
857 | */ |
858 | return mask | mark->ignore_mask; |
859 | } |
860 | |
861 | /* Get mask of events for a list of marks */ |
862 | extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn); |
863 | /* Calculate mask of events for a list of marks */ |
864 | extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn); |
865 | extern void fsnotify_init_mark(struct fsnotify_mark *mark, |
866 | struct fsnotify_group *group); |
867 | /* Find mark belonging to given group in the list of marks */ |
868 | struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type, |
869 | struct fsnotify_group *group); |
870 | /* attach the mark to the object */ |
871 | int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj, |
872 | unsigned int obj_type, int add_flags); |
873 | int fsnotify_add_mark_locked(struct fsnotify_mark *mark, void *obj, |
874 | unsigned int obj_type, int add_flags); |
875 | |
876 | /* attach the mark to the inode */ |
877 | static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark, |
878 | struct inode *inode, |
879 | int add_flags) |
880 | { |
881 | return fsnotify_add_mark(mark, obj: inode, obj_type: FSNOTIFY_OBJ_TYPE_INODE, |
882 | add_flags); |
883 | } |
884 | static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark, |
885 | struct inode *inode, |
886 | int add_flags) |
887 | { |
888 | return fsnotify_add_mark_locked(mark, obj: inode, obj_type: FSNOTIFY_OBJ_TYPE_INODE, |
889 | add_flags); |
890 | } |
891 | |
892 | static inline struct fsnotify_mark *fsnotify_find_inode_mark( |
893 | struct inode *inode, |
894 | struct fsnotify_group *group) |
895 | { |
896 | return fsnotify_find_mark(obj: inode, obj_type: FSNOTIFY_OBJ_TYPE_INODE, group); |
897 | } |
898 | |
899 | /* given a group and a mark, flag mark to be freed when all references are dropped */ |
900 | extern void fsnotify_destroy_mark(struct fsnotify_mark *mark, |
901 | struct fsnotify_group *group); |
902 | /* detach mark from inode / mount list, group list, drop inode reference */ |
903 | extern void fsnotify_detach_mark(struct fsnotify_mark *mark); |
904 | /* free mark */ |
905 | extern void fsnotify_free_mark(struct fsnotify_mark *mark); |
906 | /* Wait until all marks queued for destruction are destroyed */ |
907 | extern void fsnotify_wait_marks_destroyed(void); |
908 | /* Clear all of the marks of a group attached to a given object type */ |
909 | extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, |
910 | unsigned int obj_type); |
911 | extern void fsnotify_get_mark(struct fsnotify_mark *mark); |
912 | extern void fsnotify_put_mark(struct fsnotify_mark *mark); |
913 | extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info); |
914 | extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info); |
915 | |
916 | static inline void fsnotify_init_event(struct fsnotify_event *event) |
917 | { |
918 | INIT_LIST_HEAD(list: &event->list); |
919 | } |
920 | int fsnotify_pre_content(const struct path *path, const loff_t *ppos, |
921 | size_t count); |
922 | |
923 | #else |
924 | |
925 | static inline int fsnotify_pre_content(const struct path *path, |
926 | const loff_t *ppos, size_t count) |
927 | { |
928 | return 0; |
929 | } |
930 | |
931 | static inline int fsnotify(__u32 mask, const void *data, int data_type, |
932 | struct inode *dir, const struct qstr *name, |
933 | struct inode *inode, u32 cookie) |
934 | { |
935 | return 0; |
936 | } |
937 | |
938 | static inline int __fsnotify_parent(struct dentry *dentry, __u32 mask, |
939 | const void *data, int data_type) |
940 | { |
941 | return 0; |
942 | } |
943 | |
944 | static inline void __fsnotify_inode_delete(struct inode *inode) |
945 | {} |
946 | |
947 | static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt) |
948 | {} |
949 | |
950 | static inline void fsnotify_sb_delete(struct super_block *sb) |
951 | {} |
952 | |
953 | static inline void __fsnotify_mntns_delete(struct mnt_namespace *mntns) |
954 | {} |
955 | |
956 | static inline void fsnotify_sb_free(struct super_block *sb) |
957 | {} |
958 | |
959 | static inline void fsnotify_update_flags(struct dentry *dentry) |
960 | {} |
961 | |
962 | static inline u32 fsnotify_get_cookie(void) |
963 | { |
964 | return 0; |
965 | } |
966 | |
967 | static inline void fsnotify_unmount_inodes(struct super_block *sb) |
968 | {} |
969 | |
970 | static inline void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt) |
971 | {} |
972 | |
973 | #endif /* CONFIG_FSNOTIFY */ |
974 | |
975 | #endif /* __KERNEL __ */ |
976 | |
977 | #endif /* __LINUX_FSNOTIFY_BACKEND_H */ |
978 | |