1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Filesystem access notification for Linux 4 * 5 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <[email protected]> 6 */ 7 8 #ifndef __LINUX_FSNOTIFY_BACKEND_H 9 #define __LINUX_FSNOTIFY_BACKEND_H 10 11 #ifdef __KERNEL__ 12 13 #include <linux/idr.h> /* inotify uses this */ 14 #include <linux/fs.h> /* struct inode */ 15 #include <linux/list.h> 16 #include <linux/path.h> /* struct path */ 17 #include <linux/spinlock.h> 18 #include <linux/types.h> 19 #include <linux/atomic.h> 20 #include <linux/user_namespace.h> 21 #include <linux/refcount.h> 22 23 /* 24 * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily 25 * convert between them. dnotify only needs conversion at watch creation 26 * so no perf loss there. fanotify isn't defined yet, so it can use the 27 * wholes if it needs more events. 28 */ 29 #define FS_ACCESS 0x00000001 /* File was accessed */ 30 #define FS_MODIFY 0x00000002 /* File was modified */ 31 #define FS_ATTRIB 0x00000004 /* Metadata changed */ 32 #define FS_CLOSE_WRITE 0x00000008 /* Writtable file was closed */ 33 #define FS_CLOSE_NOWRITE 0x00000010 /* Unwrittable file closed */ 34 #define FS_OPEN 0x00000020 /* File was opened */ 35 #define FS_MOVED_FROM 0x00000040 /* File was moved from X */ 36 #define FS_MOVED_TO 0x00000080 /* File was moved to Y */ 37 #define FS_CREATE 0x00000100 /* Subfile was created */ 38 #define FS_DELETE 0x00000200 /* Subfile was deleted */ 39 #define FS_DELETE_SELF 0x00000400 /* Self was deleted */ 40 #define FS_MOVE_SELF 0x00000800 /* Self was moved */ 41 42 #define FS_UNMOUNT 0x00002000 /* inode on umount fs */ 43 #define FS_Q_OVERFLOW 0x00004000 /* Event queued overflowed */ 44 #define FS_IN_IGNORED 0x00008000 /* last inotify event here */ 45 46 #define FS_OPEN_PERM 0x00010000 /* open event in an permission hook */ 47 #define FS_ACCESS_PERM 0x00020000 /* access event in a permissions hook */ 48 49 #define FS_EXCL_UNLINK 0x04000000 /* do not send events if object is unlinked */ 50 #define FS_ISDIR 0x40000000 /* event occurred against dir */ 51 #define FS_IN_ONESHOT 0x80000000 /* only send event once */ 52 53 #define FS_DN_RENAME 0x10000000 /* file renamed */ 54 #define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */ 55 56 /* This inode cares about things that happen to its children. Always set for 57 * dnotify and inotify. */ 58 #define FS_EVENT_ON_CHILD 0x08000000 59 60 /* This is a list of all events that may get sent to a parernt based on fs event 61 * happening to inodes inside that directory */ 62 #define FS_EVENTS_POSS_ON_CHILD (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\ 63 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\ 64 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\ 65 FS_DELETE | FS_OPEN_PERM | FS_ACCESS_PERM) 66 67 #define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO) 68 69 #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM) 70 71 #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \ 72 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \ 73 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \ 74 FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \ 75 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \ 76 FS_OPEN_PERM | FS_ACCESS_PERM | FS_EXCL_UNLINK | \ 77 FS_ISDIR | FS_IN_ONESHOT | FS_DN_RENAME | \ 78 FS_DN_MULTISHOT | FS_EVENT_ON_CHILD) 79 80 struct fsnotify_group; 81 struct fsnotify_event; 82 struct fsnotify_mark; 83 struct fsnotify_event_private_data; 84 struct fsnotify_fname; 85 struct fsnotify_iter_info; 86 87 struct mem_cgroup; 88 89 /* 90 * Each group much define these ops. The fsnotify infrastructure will call 91 * these operations for each relevant group. 92 * 93 * handle_event - main call for a group to handle an fs event 94 * free_group_priv - called when a group refcnt hits 0 to clean up the private union 95 * freeing_mark - called when a mark is being destroyed for some reason. The group 96 * MUST be holding a reference on each mark and that reference must be 97 * dropped in this function. inotify uses this function to send 98 * userspace messages that marks have been removed. 99 */ 100 struct fsnotify_ops { 101 int (*handle_event)(struct fsnotify_group *group, 102 struct inode *inode, 103 u32 mask, const void *data, int data_type, 104 const unsigned char *file_name, u32 cookie, 105 struct fsnotify_iter_info *iter_info); 106 void (*free_group_priv)(struct fsnotify_group *group); 107 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group); 108 void (*free_event)(struct fsnotify_event *event); 109 /* called on final put+free to free memory */ 110 void (*free_mark)(struct fsnotify_mark *mark); 111 }; 112 113 /* 114 * all of the information about the original object we want to now send to 115 * a group. If you want to carry more info from the accessing task to the 116 * listener this structure is where you need to be adding fields. 117 */ 118 struct fsnotify_event { 119 struct list_head list; 120 /* inode may ONLY be dereferenced during handle_event(). */ 121 struct inode *inode; /* either the inode the event happened to or its parent */ 122 u32 mask; /* the type of access, bitwise OR for FS_* event types */ 123 }; 124 125 /* 126 * A group is a "thing" that wants to receive notification about filesystem 127 * events. The mask holds the subset of event types this group cares about. 128 * refcnt on a group is up to the implementor and at any moment if it goes 0 129 * everything will be cleaned up. 130 */ 131 struct fsnotify_group { 132 const struct fsnotify_ops *ops; /* how this group handles things */ 133 134 /* 135 * How the refcnt is used is up to each group. When the refcnt hits 0 136 * fsnotify will clean up all of the resources associated with this group. 137 * As an example, the dnotify group will always have a refcnt=1 and that 138 * will never change. Inotify, on the other hand, has a group per 139 * inotify_init() and the refcnt will hit 0 only when that fd has been 140 * closed. 141 */ 142 refcount_t refcnt; /* things with interest in this group */ 143 144 /* needed to send notification to userspace */ 145 spinlock_t notification_lock; /* protect the notification_list */ 146 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */ 147 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */ 148 unsigned int q_len; /* events on the queue */ 149 unsigned int max_events; /* maximum events allowed on the list */ 150 /* 151 * Valid fsnotify group priorities. Events are send in order from highest 152 * priority to lowest priority. We default to the lowest priority. 153 */ 154 #define FS_PRIO_0 0 /* normal notifiers, no permissions */ 155 #define FS_PRIO_1 1 /* fanotify content based access control */ 156 #define FS_PRIO_2 2 /* fanotify pre-content access */ 157 unsigned int priority; 158 bool shutdown; /* group is being shut down, don't queue more events */ 159 160 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */ 161 struct mutex mark_mutex; /* protect marks_list */ 162 atomic_t num_marks; /* 1 for each mark and 1 for not being 163 * past the point of no return when freeing 164 * a group */ 165 atomic_t user_waits; /* Number of tasks waiting for user 166 * response */ 167 struct list_head marks_list; /* all inode marks for this group */ 168 169 struct fasync_struct *fsn_fa; /* async notification */ 170 171 struct fsnotify_event *overflow_event; /* Event we queue when the 172 * notification list is too 173 * full */ 174 175 struct mem_cgroup *memcg; /* memcg to charge allocations */ 176 177 /* groups can define private fields here or use the void *private */ 178 union { 179 void *private; 180 #ifdef CONFIG_INOTIFY_USER 181 struct inotify_group_private_data { 182 spinlock_t idr_lock; 183 struct idr idr; 184 struct ucounts *ucounts; 185 } inotify_data; 186 #endif 187 #ifdef CONFIG_FANOTIFY 188 struct fanotify_group_private_data { 189 /* allows a group to block waiting for a userspace response */ 190 struct list_head access_list; 191 wait_queue_head_t access_waitq; 192 int f_flags; 193 unsigned int max_marks; 194 struct user_struct *user; 195 bool audit; 196 } fanotify_data; 197 #endif /* CONFIG_FANOTIFY */ 198 }; 199 }; 200 201 /* when calling fsnotify tell it if the data is a path or inode */ 202 #define FSNOTIFY_EVENT_NONE 0 203 #define FSNOTIFY_EVENT_PATH 1 204 #define FSNOTIFY_EVENT_INODE 2 205 206 enum fsnotify_obj_type { 207 FSNOTIFY_OBJ_TYPE_INODE, 208 FSNOTIFY_OBJ_TYPE_VFSMOUNT, 209 FSNOTIFY_OBJ_TYPE_COUNT, 210 FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT 211 }; 212 213 #define FSNOTIFY_OBJ_TYPE_INODE_FL (1U << FSNOTIFY_OBJ_TYPE_INODE) 214 #define FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL (1U << FSNOTIFY_OBJ_TYPE_VFSMOUNT) 215 #define FSNOTIFY_OBJ_ALL_TYPES_MASK ((1U << FSNOTIFY_OBJ_TYPE_COUNT) - 1) 216 217 static inline bool fsnotify_valid_obj_type(unsigned int type) 218 { 219 return (type < FSNOTIFY_OBJ_TYPE_COUNT); 220 } 221 222 struct fsnotify_iter_info { 223 struct fsnotify_mark *marks[FSNOTIFY_OBJ_TYPE_COUNT]; 224 unsigned int report_mask; 225 int srcu_idx; 226 }; 227 228 static inline bool fsnotify_iter_should_report_type( 229 struct fsnotify_iter_info *iter_info, int type) 230 { 231 return (iter_info->report_mask & (1U << type)); 232 } 233 234 static inline void fsnotify_iter_set_report_type( 235 struct fsnotify_iter_info *iter_info, int type) 236 { 237 iter_info->report_mask |= (1U << type); 238 } 239 240 static inline void fsnotify_iter_set_report_type_mark( 241 struct fsnotify_iter_info *iter_info, int type, 242 struct fsnotify_mark *mark) 243 { 244 iter_info->marks[type] = mark; 245 iter_info->report_mask |= (1U << type); 246 } 247 248 #define FSNOTIFY_ITER_FUNCS(name, NAME) \ 249 static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \ 250 struct fsnotify_iter_info *iter_info) \ 251 { \ 252 return (iter_info->report_mask & FSNOTIFY_OBJ_TYPE_##NAME##_FL) ? \ 253 iter_info->marks[FSNOTIFY_OBJ_TYPE_##NAME] : NULL; \ 254 } 255 256 FSNOTIFY_ITER_FUNCS(inode, INODE) 257 FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT) 258 259 #define fsnotify_foreach_obj_type(type) \ 260 for (type = 0; type < FSNOTIFY_OBJ_TYPE_COUNT; type++) 261 262 /* 263 * fsnotify_connp_t is what we embed in objects which connector can be attached 264 * to. fsnotify_connp_t * is how we refer from connector back to object. 265 */ 266 struct fsnotify_mark_connector; 267 typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t; 268 269 /* 270 * Inode / vfsmount point to this structure which tracks all marks attached to 271 * the inode / vfsmount. The reference to inode / vfsmount is held by this 272 * structure. We destroy this structure when there are no more marks attached 273 * to it. The structure is protected by fsnotify_mark_srcu. 274 */ 275 struct fsnotify_mark_connector { 276 spinlock_t lock; 277 unsigned int type; /* Type of object [lock] */ 278 union { 279 /* Object pointer [lock] */ 280 fsnotify_connp_t *obj; 281 /* Used listing heads to free after srcu period expires */ 282 struct fsnotify_mark_connector *destroy_next; 283 }; 284 struct hlist_head list; 285 }; 286 287 /* 288 * A mark is simply an object attached to an in core inode which allows an 289 * fsnotify listener to indicate they are either no longer interested in events 290 * of a type matching mask or only interested in those events. 291 * 292 * These are flushed when an inode is evicted from core and may be flushed 293 * when the inode is modified (as seen by fsnotify_access). Some fsnotify 294 * users (such as dnotify) will flush these when the open fd is closed and not 295 * at inode eviction or modification. 296 * 297 * Text in brackets is showing the lock(s) protecting modifications of a 298 * particular entry. obj_lock means either inode->i_lock or 299 * mnt->mnt_root->d_lock depending on the mark type. 300 */ 301 struct fsnotify_mark { 302 /* Mask this mark is for [mark->lock, group->mark_mutex] */ 303 __u32 mask; 304 /* We hold one for presence in g_list. Also one ref for each 'thing' 305 * in kernel that found and may be using this mark. */ 306 refcount_t refcnt; 307 /* Group this mark is for. Set on mark creation, stable until last ref 308 * is dropped */ 309 struct fsnotify_group *group; 310 /* List of marks by group->marks_list. Also reused for queueing 311 * mark into destroy_list when it's waiting for the end of SRCU period 312 * before it can be freed. [group->mark_mutex] */ 313 struct list_head g_list; 314 /* Protects inode / mnt pointers, flags, masks */ 315 spinlock_t lock; 316 /* List of marks for inode / vfsmount [connector->lock, mark ref] */ 317 struct hlist_node obj_list; 318 /* Head of list of marks for an object [mark ref] */ 319 struct fsnotify_mark_connector *connector; 320 /* Events types to ignore [mark->lock, group->mark_mutex] */ 321 __u32 ignored_mask; 322 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x01 323 #define FSNOTIFY_MARK_FLAG_ALIVE 0x02 324 #define FSNOTIFY_MARK_FLAG_ATTACHED 0x04 325 unsigned int flags; /* flags [mark->lock] */ 326 }; 327 328 #ifdef CONFIG_FSNOTIFY 329 330 /* called from the vfs helpers */ 331 332 /* main fsnotify call to send events */ 333 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, 334 const unsigned char *name, u32 cookie); 335 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask); 336 extern void __fsnotify_inode_delete(struct inode *inode); 337 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt); 338 extern u32 fsnotify_get_cookie(void); 339 340 static inline int fsnotify_inode_watches_children(struct inode *inode) 341 { 342 /* FS_EVENT_ON_CHILD is set if the inode may care */ 343 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD)) 344 return 0; 345 /* this inode might care about child events, does it care about the 346 * specific set of events that can happen on a child? */ 347 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD; 348 } 349 350 /* 351 * Update the dentry with a flag indicating the interest of its parent to receive 352 * filesystem events when those events happens to this dentry->d_inode. 353 */ 354 static inline void fsnotify_update_flags(struct dentry *dentry) 355 { 356 assert_spin_locked(&dentry->d_lock); 357 358 /* 359 * Serialisation of setting PARENT_WATCHED on the dentries is provided 360 * by d_lock. If inotify_inode_watched changes after we have taken 361 * d_lock, the following __fsnotify_update_child_dentry_flags call will 362 * find our entry, so it will spin until we complete here, and update 363 * us with the new state. 364 */ 365 if (fsnotify_inode_watches_children(dentry->d_parent->d_inode)) 366 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; 367 else 368 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; 369 } 370 371 /* called from fsnotify listeners, such as fanotify or dnotify */ 372 373 /* create a new group */ 374 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops); 375 /* get reference to a group */ 376 extern void fsnotify_get_group(struct fsnotify_group *group); 377 /* drop reference on a group from fsnotify_alloc_group */ 378 extern void fsnotify_put_group(struct fsnotify_group *group); 379 /* group destruction begins, stop queuing new events */ 380 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group); 381 /* destroy group */ 382 extern void fsnotify_destroy_group(struct fsnotify_group *group); 383 /* fasync handler function */ 384 extern int fsnotify_fasync(int fd, struct file *file, int on); 385 /* Free event from memory */ 386 extern void fsnotify_destroy_event(struct fsnotify_group *group, 387 struct fsnotify_event *event); 388 /* attach the event to the group notification queue */ 389 extern int fsnotify_add_event(struct fsnotify_group *group, 390 struct fsnotify_event *event, 391 int (*merge)(struct list_head *, 392 struct fsnotify_event *)); 393 /* Queue overflow event to a notification group */ 394 static inline void fsnotify_queue_overflow(struct fsnotify_group *group) 395 { 396 fsnotify_add_event(group, group->overflow_event, NULL); 397 } 398 399 /* true if the group notification queue is empty */ 400 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group); 401 /* return, but do not dequeue the first event on the notification queue */ 402 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group); 403 /* return AND dequeue the first event on the notification queue */ 404 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group); 405 406 /* functions used to manipulate the marks attached to inodes */ 407 408 /* Get mask of events for a list of marks */ 409 extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn); 410 /* Calculate mask of events for a list of marks */ 411 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn); 412 extern void fsnotify_init_mark(struct fsnotify_mark *mark, 413 struct fsnotify_group *group); 414 /* Find mark belonging to given group in the list of marks */ 415 extern struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp, 416 struct fsnotify_group *group); 417 /* attach the mark to the object */ 418 extern int fsnotify_add_mark(struct fsnotify_mark *mark, 419 fsnotify_connp_t *connp, unsigned int type, 420 int allow_dups); 421 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark, 422 fsnotify_connp_t *connp, unsigned int type, 423 int allow_dups); 424 /* attach the mark to the inode */ 425 static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark, 426 struct inode *inode, 427 int allow_dups) 428 { 429 return fsnotify_add_mark(mark, &inode->i_fsnotify_marks, 430 FSNOTIFY_OBJ_TYPE_INODE, allow_dups); 431 } 432 static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark, 433 struct inode *inode, 434 int allow_dups) 435 { 436 return fsnotify_add_mark_locked(mark, &inode->i_fsnotify_marks, 437 FSNOTIFY_OBJ_TYPE_INODE, allow_dups); 438 } 439 /* given a group and a mark, flag mark to be freed when all references are dropped */ 440 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark, 441 struct fsnotify_group *group); 442 /* detach mark from inode / mount list, group list, drop inode reference */ 443 extern void fsnotify_detach_mark(struct fsnotify_mark *mark); 444 /* free mark */ 445 extern void fsnotify_free_mark(struct fsnotify_mark *mark); 446 /* run all the marks in a group, and clear all of the marks attached to given object type */ 447 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type); 448 /* run all the marks in a group, and clear all of the vfsmount marks */ 449 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group) 450 { 451 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL); 452 } 453 /* run all the marks in a group, and clear all of the inode marks */ 454 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group) 455 { 456 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE_FL); 457 } 458 extern void fsnotify_get_mark(struct fsnotify_mark *mark); 459 extern void fsnotify_put_mark(struct fsnotify_mark *mark); 460 extern void fsnotify_unmount_inodes(struct super_block *sb); 461 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info); 462 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info); 463 464 /* put here because inotify does some weird stuff when destroying watches */ 465 extern void fsnotify_init_event(struct fsnotify_event *event, 466 struct inode *to_tell, u32 mask); 467 468 #else 469 470 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, 471 const unsigned char *name, u32 cookie) 472 { 473 return 0; 474 } 475 476 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask) 477 { 478 return 0; 479 } 480 481 static inline void __fsnotify_inode_delete(struct inode *inode) 482 {} 483 484 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt) 485 {} 486 487 static inline void fsnotify_update_flags(struct dentry *dentry) 488 {} 489 490 static inline u32 fsnotify_get_cookie(void) 491 { 492 return 0; 493 } 494 495 static inline void fsnotify_unmount_inodes(struct super_block *sb) 496 {} 497 498 #endif /* CONFIG_FSNOTIFY */ 499 500 #endif /* __KERNEL __ */ 501 502 #endif /* __LINUX_FSNOTIFY_BACKEND_H */ 503