fsnotify_backend.h

  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
129struct fsnotify_group;
130struct fsnotify_event;
131struct fsnotify_mark;
132struct fsnotify_event_private_data;
133struct fsnotify_fname;
134struct fsnotify_iter_info;
135
136struct 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 */
172struct 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 */
192struct 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 */
200enum 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 */
213struct 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			/* chained on perm_group_list */
277			struct list_head perm_grp_list;
278		} fanotify_data;
279#endif /* CONFIG_FANOTIFY */
280	};
281};
282
283/*
284 * These helpers are used to prevent deadlock when reclaiming inodes with
285 * evictable marks of the same group that is allocating a new mark.
286 */
287static inline void fsnotify_group_lock(struct fsnotify_group *group)
288{
289	mutex_lock(&group->mark_mutex);
290	group->owner_flags = memalloc_nofs_save();
291}
292
293static inline void fsnotify_group_unlock(struct fsnotify_group *group)
294{
295	memalloc_nofs_restore(group->owner_flags);
296	mutex_unlock(&group->mark_mutex);
297}
298
299static inline void fsnotify_group_assert_locked(struct fsnotify_group *group)
300{
301	WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
302	WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
303}
304
305/* When calling fsnotify tell it if the data is a path or inode */
306enum fsnotify_data_type {
307	FSNOTIFY_EVENT_NONE,
308	FSNOTIFY_EVENT_FILE_RANGE,
309	FSNOTIFY_EVENT_PATH,
310	FSNOTIFY_EVENT_INODE,
311	FSNOTIFY_EVENT_DENTRY,
312	FSNOTIFY_EVENT_MNT,
313	FSNOTIFY_EVENT_ERROR,
314};
315
316struct fs_error_report {
317	int error;
318	struct inode *inode;
319	struct super_block *sb;
320};
321
322struct file_range {
323	const struct path *path;
324	loff_t pos;
325	size_t count;
326};
327
328static inline const struct path *file_range_path(const struct file_range *range)
329{
330	return range->path;
331}
332
333struct fsnotify_mnt {
334	const struct mnt_namespace *ns;
335	u64 mnt_id;
336};
337
338static inline struct inode *fsnotify_data_inode(const void *data, int data_type)
339{
340	switch (data_type) {
341	case FSNOTIFY_EVENT_INODE:
342		return (struct inode *)data;
343	case FSNOTIFY_EVENT_DENTRY:
344		return d_inode(data);
345	case FSNOTIFY_EVENT_PATH:
346		return d_inode(((const struct path *)data)->dentry);
347	case FSNOTIFY_EVENT_FILE_RANGE:
348		return d_inode(file_range_path(data)->dentry);
349	case FSNOTIFY_EVENT_ERROR:
350		return ((struct fs_error_report *)data)->inode;
351	default:
352		return NULL;
353	}
354}
355
356static inline struct dentry *fsnotify_data_dentry(const void *data, int data_type)
357{
358	switch (data_type) {
359	case FSNOTIFY_EVENT_DENTRY:
360		/* Non const is needed for dget() */
361		return (struct dentry *)data;
362	case FSNOTIFY_EVENT_PATH:
363		return ((const struct path *)data)->dentry;
364	case FSNOTIFY_EVENT_FILE_RANGE:
365		return file_range_path(data)->dentry;
366	default:
367		return NULL;
368	}
369}
370
371static inline const struct path *fsnotify_data_path(const void *data,
372						    int data_type)
373{
374	switch (data_type) {
375	case FSNOTIFY_EVENT_PATH:
376		return data;
377	case FSNOTIFY_EVENT_FILE_RANGE:
378		return file_range_path(data);
379	default:
380		return NULL;
381	}
382}
383
384static inline struct super_block *fsnotify_data_sb(const void *data,
385						   int data_type)
386{
387	switch (data_type) {
388	case FSNOTIFY_EVENT_INODE:
389		return ((struct inode *)data)->i_sb;
390	case FSNOTIFY_EVENT_DENTRY:
391		return ((struct dentry *)data)->d_sb;
392	case FSNOTIFY_EVENT_PATH:
393		return ((const struct path *)data)->dentry->d_sb;
394	case FSNOTIFY_EVENT_FILE_RANGE:
395		return file_range_path(data)->dentry->d_sb;
396	case FSNOTIFY_EVENT_ERROR:
397		return ((struct fs_error_report *) data)->sb;
398	default:
399		return NULL;
400	}
401}
402
403static inline const struct fsnotify_mnt *fsnotify_data_mnt(const void *data,
404							   int data_type)
405{
406	switch (data_type) {
407	case FSNOTIFY_EVENT_MNT:
408		return data;
409	default:
410		return NULL;
411	}
412}
413
414static inline u64 fsnotify_data_mnt_id(const void *data, int data_type)
415{
416	const struct fsnotify_mnt *mnt_data = fsnotify_data_mnt(data, data_type);
417
418	return mnt_data ? mnt_data->mnt_id : 0;
419}
420
421static inline struct fs_error_report *fsnotify_data_error_report(
422							const void *data,
423							int data_type)
424{
425	switch (data_type) {
426	case FSNOTIFY_EVENT_ERROR:
427		return (struct fs_error_report *) data;
428	default:
429		return NULL;
430	}
431}
432
433static inline const struct file_range *fsnotify_data_file_range(
434							const void *data,
435							int data_type)
436{
437	switch (data_type) {
438	case FSNOTIFY_EVENT_FILE_RANGE:
439		return (struct file_range *)data;
440	default:
441		return NULL;
442	}
443}
444
445/*
446 * Index to merged marks iterator array that correlates to a type of watch.
447 * The type of watched object can be deduced from the iterator type, but not
448 * the other way around, because an event can match different watched objects
449 * of the same object type.
450 * For example, both parent and child are watching an object of type inode.
451 */
452enum fsnotify_iter_type {
453	FSNOTIFY_ITER_TYPE_INODE,
454	FSNOTIFY_ITER_TYPE_VFSMOUNT,
455	FSNOTIFY_ITER_TYPE_SB,
456	FSNOTIFY_ITER_TYPE_PARENT,
457	FSNOTIFY_ITER_TYPE_INODE2,
458	FSNOTIFY_ITER_TYPE_MNTNS,
459	FSNOTIFY_ITER_TYPE_COUNT
460};
461
462/* The type of object that a mark is attached to */
463enum fsnotify_obj_type {
464	FSNOTIFY_OBJ_TYPE_ANY = -1,
465	FSNOTIFY_OBJ_TYPE_INODE,
466	FSNOTIFY_OBJ_TYPE_VFSMOUNT,
467	FSNOTIFY_OBJ_TYPE_SB,
468	FSNOTIFY_OBJ_TYPE_MNTNS,
469	FSNOTIFY_OBJ_TYPE_COUNT,
470	FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
471};
472
473static inline bool fsnotify_valid_obj_type(unsigned int obj_type)
474{
475	return (obj_type < FSNOTIFY_OBJ_TYPE_COUNT);
476}
477
478struct fsnotify_iter_info {
479	struct fsnotify_mark *marks[FSNOTIFY_ITER_TYPE_COUNT];
480	struct fsnotify_group *current_group;
481	unsigned int report_mask;
482	int srcu_idx;
483};
484
485static inline bool fsnotify_iter_should_report_type(
486		struct fsnotify_iter_info *iter_info, int iter_type)
487{
488	return (iter_info->report_mask & (1U << iter_type));
489}
490
491static inline void fsnotify_iter_set_report_type(
492		struct fsnotify_iter_info *iter_info, int iter_type)
493{
494	iter_info->report_mask |= (1U << iter_type);
495}
496
497static inline struct fsnotify_mark *fsnotify_iter_mark(
498		struct fsnotify_iter_info *iter_info, int iter_type)
499{
500	if (fsnotify_iter_should_report_type(iter_info, iter_type))
501		return iter_info->marks[iter_type];
502	return NULL;
503}
504
505static inline int fsnotify_iter_step(struct fsnotify_iter_info *iter, int type,
506				     struct fsnotify_mark **markp)
507{
508	while (type < FSNOTIFY_ITER_TYPE_COUNT) {
509		*markp = fsnotify_iter_mark(iter, type);
510		if (*markp)
511			break;
512		type++;
513	}
514	return type;
515}
516
517#define FSNOTIFY_ITER_FUNCS(name, NAME) \
518static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \
519		struct fsnotify_iter_info *iter_info) \
520{ \
521	return fsnotify_iter_mark(iter_info, FSNOTIFY_ITER_TYPE_##NAME); \
522}
523
524FSNOTIFY_ITER_FUNCS(inode, INODE)
525FSNOTIFY_ITER_FUNCS(parent, PARENT)
526FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT)
527FSNOTIFY_ITER_FUNCS(sb, SB)
528
529#define fsnotify_foreach_iter_type(type) \
530	for (type = 0; type < FSNOTIFY_ITER_TYPE_COUNT; type++)
531#define fsnotify_foreach_iter_mark_type(iter, mark, type) \
532	for (type = 0; \
533	     type = fsnotify_iter_step(iter, type, &mark), \
534	     type < FSNOTIFY_ITER_TYPE_COUNT; \
535	     type++)
536
537/*
538 * Inode/vfsmount/sb point to this structure which tracks all marks attached to
539 * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this
540 * structure. We destroy this structure when there are no more marks attached
541 * to it. The structure is protected by fsnotify_mark_srcu.
542 */
543struct fsnotify_mark_connector {
544	spinlock_t lock;
545	unsigned char type;	/* Type of object [lock] */
546	unsigned char prio;	/* Highest priority group */
547#define FSNOTIFY_CONN_FLAG_IS_WATCHED	0x01
548#define FSNOTIFY_CONN_FLAG_HAS_IREF	0x02
549	unsigned short flags;	/* flags [lock] */
550	union {
551		/* Object pointer [lock] */
552		void *obj;
553		/* Used listing heads to free after srcu period expires */
554		struct fsnotify_mark_connector *destroy_next;
555	};
556	struct hlist_head list;
557};
558
559/*
560 * Container for per-sb fsnotify state (sb marks and more).
561 * Attached lazily on first marked object on the sb and freed when killing sb.
562 */
563struct fsnotify_sb_info {
564	struct fsnotify_mark_connector __rcu *sb_marks;
565	/*
566	 * Number of inode/mount/sb objects that are being watched in this sb.
567	 * Note that inodes objects are currently double-accounted.
568	 *
569	 * The value in watched_objects[prio] is the number of objects that are
570	 * watched by groups of priority >= prio, so watched_objects[0] is the
571	 * total number of watched objects in this sb.
572	 */
573	atomic_long_t watched_objects[__FSNOTIFY_PRIO_NUM];
574};
575
576static inline struct fsnotify_sb_info *fsnotify_sb_info(struct super_block *sb)
577{
578#ifdef CONFIG_FSNOTIFY
579	return READ_ONCE(sb->s_fsnotify_info);
580#else
581	return NULL;
582#endif
583}
584
585static inline atomic_long_t *fsnotify_sb_watched_objects(struct super_block *sb)
586{
587	return &fsnotify_sb_info(sb)->watched_objects[0];
588}
589
590/*
591 * A mark is simply an object attached to an in core inode which allows an
592 * fsnotify listener to indicate they are either no longer interested in events
593 * of a type matching mask or only interested in those events.
594 *
595 * These are flushed when an inode is evicted from core and may be flushed
596 * when the inode is modified (as seen by fsnotify_access).  Some fsnotify
597 * users (such as dnotify) will flush these when the open fd is closed and not
598 * at inode eviction or modification.
599 *
600 * Text in brackets is showing the lock(s) protecting modifications of a
601 * particular entry. obj_lock means either inode->i_lock or
602 * mnt->mnt_root->d_lock depending on the mark type.
603 */
604struct fsnotify_mark {
605	/* Mask this mark is for [mark->lock, group->mark_mutex] */
606	__u32 mask;
607	/* We hold one for presence in g_list. Also one ref for each 'thing'
608	 * in kernel that found and may be using this mark. */
609	refcount_t refcnt;
610	/* Group this mark is for. Set on mark creation, stable until last ref
611	 * is dropped */
612	struct fsnotify_group *group;
613	/* List of marks by group->marks_list. Also reused for queueing
614	 * mark into destroy_list when it's waiting for the end of SRCU period
615	 * before it can be freed. [group->mark_mutex] */
616	struct list_head g_list;
617	/* Protects inode / mnt pointers, flags, masks */
618	spinlock_t lock;
619	/* List of marks for inode / vfsmount [connector->lock, mark ref] */
620	struct hlist_node obj_list;
621	/* Head of list of marks for an object [mark ref] */
622	struct fsnotify_mark_connector *connector;
623	/* Events types and flags to ignore [mark->lock, group->mark_mutex] */
624	__u32 ignore_mask;
625	/* General fsnotify mark flags */
626#define FSNOTIFY_MARK_FLAG_ALIVE		0x0001
627#define FSNOTIFY_MARK_FLAG_ATTACHED		0x0002
628	/* inotify mark flags */
629#define FSNOTIFY_MARK_FLAG_EXCL_UNLINK		0x0010
630#define FSNOTIFY_MARK_FLAG_IN_ONESHOT		0x0020
631	/* fanotify mark flags */
632#define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY	0x0100
633#define FSNOTIFY_MARK_FLAG_NO_IREF		0x0200
634#define FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS	0x0400
635#define FSNOTIFY_MARK_FLAG_HAS_FSID		0x0800
636#define FSNOTIFY_MARK_FLAG_WEAK_FSID		0x1000
637	unsigned int flags;		/* flags [mark->lock] */
638};
639
640#ifdef CONFIG_FSNOTIFY
641
642/* called from the vfs helpers */
643
644/* main fsnotify call to send events */
645extern int fsnotify(__u32 mask, const void *data, int data_type,
646		    struct inode *dir, const struct qstr *name,
647		    struct inode *inode, u32 cookie);
648extern int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data,
649			   int data_type);
650extern void __fsnotify_inode_delete(struct inode *inode);
651extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
652extern void fsnotify_sb_delete(struct super_block *sb);
653extern void __fsnotify_mntns_delete(struct mnt_namespace *mntns);
654extern void fsnotify_sb_free(struct super_block *sb);
655extern u32 fsnotify_get_cookie(void);
656extern void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt);
657
658static inline __u32 fsnotify_parent_needed_mask(__u32 mask)
659{
660	/* FS_EVENT_ON_CHILD is set on marks that want parent/name info */
661	if (!(mask & FS_EVENT_ON_CHILD))
662		return 0;
663	/*
664	 * This object might be watched by a mark that cares about parent/name
665	 * info, does it care about the specific set of events that can be
666	 * reported with parent/name info?
667	 */
668	return mask & FS_EVENTS_POSS_TO_PARENT;
669}
670
671static inline int fsnotify_inode_watches_children(struct inode *inode)
672{
673	__u32 parent_mask = READ_ONCE(inode->i_fsnotify_mask);
674
675	/* FS_EVENT_ON_CHILD is set if the inode may care */
676	if (!(parent_mask & FS_EVENT_ON_CHILD))
677		return 0;
678	/* this inode might care about child events, does it care about the
679	 * specific set of events that can happen on a child? */
680	return parent_mask & FS_EVENTS_POSS_ON_CHILD;
681}
682
683/*
684 * Update the dentry with a flag indicating the interest of its parent to receive
685 * filesystem events when those events happens to this dentry->d_inode.
686 */
687static inline void fsnotify_update_flags(struct dentry *dentry)
688{
689	assert_spin_locked(&dentry->d_lock);
690
691	/*
692	 * Serialisation of setting PARENT_WATCHED on the dentries is provided
693	 * by d_lock. If inotify_inode_watched changes after we have taken
694	 * d_lock, the following fsnotify_set_children_dentry_flags call will
695	 * find our entry, so it will spin until we complete here, and update
696	 * us with the new state.
697	 */
698	if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
699		dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
700	else
701		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
702}
703
704/* called from fsnotify listeners, such as fanotify or dnotify */
705
706/* create a new group */
707extern struct fsnotify_group *fsnotify_alloc_group(
708				const struct fsnotify_ops *ops,
709				int flags);
710/* get reference to a group */
711extern void fsnotify_get_group(struct fsnotify_group *group);
712/* drop reference on a group from fsnotify_alloc_group */
713extern void fsnotify_put_group(struct fsnotify_group *group);
714/* group destruction begins, stop queuing new events */
715extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
716/* destroy group */
717extern void fsnotify_destroy_group(struct fsnotify_group *group);
718/* fasync handler function */
719extern int fsnotify_fasync(int fd, struct file *file, int on);
720/* Free event from memory */
721extern void fsnotify_destroy_event(struct fsnotify_group *group,
722				   struct fsnotify_event *event);
723/* attach the event to the group notification queue */
724extern int fsnotify_insert_event(struct fsnotify_group *group,
725				 struct fsnotify_event *event,
726				 int (*merge)(struct fsnotify_group *,
727					      struct fsnotify_event *),
728				 void (*insert)(struct fsnotify_group *,
729						struct fsnotify_event *));
730
731static inline int fsnotify_add_event(struct fsnotify_group *group,
732				     struct fsnotify_event *event,
733				     int (*merge)(struct fsnotify_group *,
734						  struct fsnotify_event *))
735{
736	return fsnotify_insert_event(group, event, merge, NULL);
737}
738
739/* Queue overflow event to a notification group */
740static inline void fsnotify_queue_overflow(struct fsnotify_group *group)
741{
742	fsnotify_add_event(group, group->overflow_event, NULL);
743}
744
745static inline bool fsnotify_is_overflow_event(u32 mask)
746{
747	return mask & FS_Q_OVERFLOW;
748}
749
750static inline bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group)
751{
752	assert_spin_locked(&group->notification_lock);
753
754	return list_empty(&group->notification_list);
755}
756
757extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
758/* return, but do not dequeue the first event on the notification queue */
759extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
760/* return AND dequeue the first event on the notification queue */
761extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
762/* Remove event queued in the notification list */
763extern void fsnotify_remove_queued_event(struct fsnotify_group *group,
764					 struct fsnotify_event *event);
765
766/* functions used to manipulate the marks attached to inodes */
767
768/*
769 * Canonical "ignore mask" including event flags.
770 *
771 * Note the subtle semantic difference from the legacy ->ignored_mask.
772 * ->ignored_mask traditionally only meant which events should be ignored,
773 * while ->ignore_mask also includes flags regarding the type of objects on
774 * which events should be ignored.
775 */
776static inline __u32 fsnotify_ignore_mask(struct fsnotify_mark *mark)
777{
778	__u32 ignore_mask = mark->ignore_mask;
779
780	/* The event flags in ignore mask take effect */
781	if (mark->flags & FSNOTIFY_MARK_FLAG_HAS_IGNORE_FLAGS)
782		return ignore_mask;
783
784	/*
785	 * Legacy behavior:
786	 * - Always ignore events on dir
787	 * - Ignore events on child if parent is watching children
788	 */
789	ignore_mask |= FS_ISDIR;
790	ignore_mask &= ~FS_EVENT_ON_CHILD;
791	ignore_mask |= mark->mask & FS_EVENT_ON_CHILD;
792
793	return ignore_mask;
794}
795
796/* Legacy ignored_mask - only event types to ignore */
797static inline __u32 fsnotify_ignored_events(struct fsnotify_mark *mark)
798{
799	return mark->ignore_mask & ALL_FSNOTIFY_EVENTS;
800}
801
802/*
803 * Check if mask (or ignore mask) should be applied depending if victim is a
804 * directory and whether it is reported to a watching parent.
805 */
806static inline bool fsnotify_mask_applicable(__u32 mask, bool is_dir,
807					    int iter_type)
808{
809	/* Should mask be applied to a directory? */
810	if (is_dir && !(mask & FS_ISDIR))
811		return false;
812
813	/* Should mask be applied to a child? */
814	if (iter_type == FSNOTIFY_ITER_TYPE_PARENT &&
815	    !(mask & FS_EVENT_ON_CHILD))
816		return false;
817
818	return true;
819}
820
821/*
822 * Effective ignore mask taking into account if event victim is a
823 * directory and whether it is reported to a watching parent.
824 */
825static inline __u32 fsnotify_effective_ignore_mask(struct fsnotify_mark *mark,
826						   bool is_dir, int iter_type)
827{
828	__u32 ignore_mask = fsnotify_ignored_events(mark);
829
830	if (!ignore_mask)
831		return 0;
832
833	/* For non-dir and non-child, no need to consult the event flags */
834	if (!is_dir && iter_type != FSNOTIFY_ITER_TYPE_PARENT)
835		return ignore_mask;
836
837	ignore_mask = fsnotify_ignore_mask(mark);
838	if (!fsnotify_mask_applicable(ignore_mask, is_dir, iter_type))
839		return 0;
840
841	return ignore_mask & ALL_FSNOTIFY_EVENTS;
842}
843
844/* Get mask for calculating object interest taking ignore mask into account */
845static inline __u32 fsnotify_calc_mask(struct fsnotify_mark *mark)
846{
847	__u32 mask = mark->mask;
848
849	if (!fsnotify_ignored_events(mark))
850		return mask;
851
852	/* Interest in FS_MODIFY may be needed for clearing ignore mask */
853	if (!(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
854		mask |= FS_MODIFY;
855
856	/*
857	 * If mark is interested in ignoring events on children, the object must
858	 * show interest in those events for fsnotify_parent() to notice it.
859	 */
860	return mask | mark->ignore_mask;
861}
862
863/* Get mask of events for a list of marks */
864extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn);
865/* Calculate mask of events for a list of marks */
866extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
867extern void fsnotify_init_mark(struct fsnotify_mark *mark,
868			       struct fsnotify_group *group);
869/* Find mark belonging to given group in the list of marks */
870struct fsnotify_mark *fsnotify_find_mark(void *obj, unsigned int obj_type,
871					 struct fsnotify_group *group);
872/* attach the mark to the object */
873int fsnotify_add_mark(struct fsnotify_mark *mark, void *obj,
874		      unsigned int obj_type, int add_flags);
875int fsnotify_add_mark_locked(struct fsnotify_mark *mark, void *obj,
876			     unsigned int obj_type, int add_flags);
877
878/* attach the mark to the inode */
879static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
880					  struct inode *inode,
881					  int add_flags)
882{
883	return fsnotify_add_mark(mark, inode, FSNOTIFY_OBJ_TYPE_INODE,
884				 add_flags);
885}
886static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
887						 struct inode *inode,
888						 int add_flags)
889{
890	return fsnotify_add_mark_locked(mark, inode, FSNOTIFY_OBJ_TYPE_INODE,
891					add_flags);
892}
893
894static inline struct fsnotify_mark *fsnotify_find_inode_mark(
895						struct inode *inode,
896						struct fsnotify_group *group)
897{
898	return fsnotify_find_mark(inode, FSNOTIFY_OBJ_TYPE_INODE, group);
899}
900
901/* given a group and a mark, flag mark to be freed when all references are dropped */
902extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
903				  struct fsnotify_group *group);
904/* detach mark from inode / mount list, group list, drop inode reference */
905extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
906/* free mark */
907extern void fsnotify_free_mark(struct fsnotify_mark *mark);
908/* Wait until all marks queued for destruction are destroyed */
909extern void fsnotify_wait_marks_destroyed(void);
910/* Clear all of the marks of a group attached to a given object type */
911extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group,
912					  unsigned int obj_type);
913extern void fsnotify_get_mark(struct fsnotify_mark *mark);
914extern void fsnotify_put_mark(struct fsnotify_mark *mark);
915extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
916extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
917
918static inline void fsnotify_init_event(struct fsnotify_event *event)
919{
920	INIT_LIST_HEAD(&event->list);
921}
922int fsnotify_pre_content(const struct path *path, const loff_t *ppos,
923			 size_t count);
924
925#else
926
927static inline int fsnotify_pre_content(const struct path *path,
928				       const loff_t *ppos, size_t count)
929{
930	return 0;
931}
932
933static inline int fsnotify(__u32 mask, const void *data, int data_type,
934			   struct inode *dir, const struct qstr *name,
935			   struct inode *inode, u32 cookie)
936{
937	return 0;
938}
939
940static inline int __fsnotify_parent(struct dentry *dentry, __u32 mask,
941				  const void *data, int data_type)
942{
943	return 0;
944}
945
946static inline void __fsnotify_inode_delete(struct inode *inode)
947{}
948
949static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
950{}
951
952static inline void fsnotify_sb_delete(struct super_block *sb)
953{}
954
955static inline void __fsnotify_mntns_delete(struct mnt_namespace *mntns)
956{}
957
958static inline void fsnotify_sb_free(struct super_block *sb)
959{}
960
961static inline void fsnotify_update_flags(struct dentry *dentry)
962{}
963
964static inline u32 fsnotify_get_cookie(void)
965{
966	return 0;
967}
968
969static inline void fsnotify_unmount_inodes(struct super_block *sb)
970{}
971
972static inline void fsnotify_mnt(__u32 mask, struct mnt_namespace *ns, struct vfsmount *mnt)
973{}
974
975#endif	/* CONFIG_FSNOTIFY */
976
977#endif	/* __KERNEL __ */
978
979#endif	/* __LINUX_FSNOTIFY_BACKEND_H */