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 /* Events that can be reported to backends */ 72 #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \ 73 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \ 74 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \ 75 FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \ 76 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \ 77 FS_OPEN_PERM | FS_ACCESS_PERM | FS_DN_RENAME) 78 79 /* Extra flags that may be reported with event or control handling of events */ 80 #define ALL_FSNOTIFY_FLAGS (FS_EXCL_UNLINK | FS_ISDIR | FS_IN_ONESHOT | \ 81 FS_DN_MULTISHOT | FS_EVENT_ON_CHILD) 82 83 #define ALL_FSNOTIFY_BITS (ALL_FSNOTIFY_EVENTS | ALL_FSNOTIFY_FLAGS) 84 85 struct fsnotify_group; 86 struct fsnotify_event; 87 struct fsnotify_mark; 88 struct fsnotify_event_private_data; 89 struct fsnotify_fname; 90 struct fsnotify_iter_info; 91 92 struct mem_cgroup; 93 94 /* 95 * Each group much define these ops. The fsnotify infrastructure will call 96 * these operations for each relevant group. 97 * 98 * handle_event - main call for a group to handle an fs event 99 * free_group_priv - called when a group refcnt hits 0 to clean up the private union 100 * freeing_mark - called when a mark is being destroyed for some reason. The group 101 * MUST be holding a reference on each mark and that reference must be 102 * dropped in this function. inotify uses this function to send 103 * userspace messages that marks have been removed. 104 */ 105 struct fsnotify_ops { 106 int (*handle_event)(struct fsnotify_group *group, 107 struct inode *inode, 108 u32 mask, const void *data, int data_type, 109 const unsigned char *file_name, u32 cookie, 110 struct fsnotify_iter_info *iter_info); 111 void (*free_group_priv)(struct fsnotify_group *group); 112 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group); 113 void (*free_event)(struct fsnotify_event *event); 114 /* called on final put+free to free memory */ 115 void (*free_mark)(struct fsnotify_mark *mark); 116 }; 117 118 /* 119 * all of the information about the original object we want to now send to 120 * a group. If you want to carry more info from the accessing task to the 121 * listener this structure is where you need to be adding fields. 122 */ 123 struct fsnotify_event { 124 struct list_head list; 125 /* inode may ONLY be dereferenced during handle_event(). */ 126 struct inode *inode; /* either the inode the event happened to or its parent */ 127 u32 mask; /* the type of access, bitwise OR for FS_* event types */ 128 }; 129 130 /* 131 * A group is a "thing" that wants to receive notification about filesystem 132 * events. The mask holds the subset of event types this group cares about. 133 * refcnt on a group is up to the implementor and at any moment if it goes 0 134 * everything will be cleaned up. 135 */ 136 struct fsnotify_group { 137 const struct fsnotify_ops *ops; /* how this group handles things */ 138 139 /* 140 * How the refcnt is used is up to each group. When the refcnt hits 0 141 * fsnotify will clean up all of the resources associated with this group. 142 * As an example, the dnotify group will always have a refcnt=1 and that 143 * will never change. Inotify, on the other hand, has a group per 144 * inotify_init() and the refcnt will hit 0 only when that fd has been 145 * closed. 146 */ 147 refcount_t refcnt; /* things with interest in this group */ 148 149 /* needed to send notification to userspace */ 150 spinlock_t notification_lock; /* protect the notification_list */ 151 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */ 152 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */ 153 unsigned int q_len; /* events on the queue */ 154 unsigned int max_events; /* maximum events allowed on the list */ 155 /* 156 * Valid fsnotify group priorities. Events are send in order from highest 157 * priority to lowest priority. We default to the lowest priority. 158 */ 159 #define FS_PRIO_0 0 /* normal notifiers, no permissions */ 160 #define FS_PRIO_1 1 /* fanotify content based access control */ 161 #define FS_PRIO_2 2 /* fanotify pre-content access */ 162 unsigned int priority; 163 bool shutdown; /* group is being shut down, don't queue more events */ 164 165 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */ 166 struct mutex mark_mutex; /* protect marks_list */ 167 atomic_t num_marks; /* 1 for each mark and 1 for not being 168 * past the point of no return when freeing 169 * a group */ 170 atomic_t user_waits; /* Number of tasks waiting for user 171 * response */ 172 struct list_head marks_list; /* all inode marks for this group */ 173 174 struct fasync_struct *fsn_fa; /* async notification */ 175 176 struct fsnotify_event *overflow_event; /* Event we queue when the 177 * notification list is too 178 * full */ 179 180 struct mem_cgroup *memcg; /* memcg to charge allocations */ 181 182 /* groups can define private fields here or use the void *private */ 183 union { 184 void *private; 185 #ifdef CONFIG_INOTIFY_USER 186 struct inotify_group_private_data { 187 spinlock_t idr_lock; 188 struct idr idr; 189 struct ucounts *ucounts; 190 } inotify_data; 191 #endif 192 #ifdef CONFIG_FANOTIFY 193 struct fanotify_group_private_data { 194 /* allows a group to block waiting for a userspace response */ 195 struct list_head access_list; 196 wait_queue_head_t access_waitq; 197 int flags; /* flags from fanotify_init() */ 198 int f_flags; /* event_f_flags from fanotify_init() */ 199 unsigned int max_marks; 200 struct user_struct *user; 201 } fanotify_data; 202 #endif /* CONFIG_FANOTIFY */ 203 }; 204 }; 205 206 /* when calling fsnotify tell it if the data is a path or inode */ 207 #define FSNOTIFY_EVENT_NONE 0 208 #define FSNOTIFY_EVENT_PATH 1 209 #define FSNOTIFY_EVENT_INODE 2 210 211 enum fsnotify_obj_type { 212 FSNOTIFY_OBJ_TYPE_INODE, 213 FSNOTIFY_OBJ_TYPE_VFSMOUNT, 214 FSNOTIFY_OBJ_TYPE_SB, 215 FSNOTIFY_OBJ_TYPE_COUNT, 216 FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT 217 }; 218 219 #define FSNOTIFY_OBJ_TYPE_INODE_FL (1U << FSNOTIFY_OBJ_TYPE_INODE) 220 #define FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL (1U << FSNOTIFY_OBJ_TYPE_VFSMOUNT) 221 #define FSNOTIFY_OBJ_TYPE_SB_FL (1U << FSNOTIFY_OBJ_TYPE_SB) 222 #define FSNOTIFY_OBJ_ALL_TYPES_MASK ((1U << FSNOTIFY_OBJ_TYPE_COUNT) - 1) 223 224 static inline bool fsnotify_valid_obj_type(unsigned int type) 225 { 226 return (type < FSNOTIFY_OBJ_TYPE_COUNT); 227 } 228 229 struct fsnotify_iter_info { 230 struct fsnotify_mark *marks[FSNOTIFY_OBJ_TYPE_COUNT]; 231 unsigned int report_mask; 232 int srcu_idx; 233 }; 234 235 static inline bool fsnotify_iter_should_report_type( 236 struct fsnotify_iter_info *iter_info, int type) 237 { 238 return (iter_info->report_mask & (1U << type)); 239 } 240 241 static inline void fsnotify_iter_set_report_type( 242 struct fsnotify_iter_info *iter_info, int type) 243 { 244 iter_info->report_mask |= (1U << type); 245 } 246 247 static inline void fsnotify_iter_set_report_type_mark( 248 struct fsnotify_iter_info *iter_info, int type, 249 struct fsnotify_mark *mark) 250 { 251 iter_info->marks[type] = mark; 252 iter_info->report_mask |= (1U << type); 253 } 254 255 #define FSNOTIFY_ITER_FUNCS(name, NAME) \ 256 static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \ 257 struct fsnotify_iter_info *iter_info) \ 258 { \ 259 return (iter_info->report_mask & FSNOTIFY_OBJ_TYPE_##NAME##_FL) ? \ 260 iter_info->marks[FSNOTIFY_OBJ_TYPE_##NAME] : NULL; \ 261 } 262 263 FSNOTIFY_ITER_FUNCS(inode, INODE) 264 FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT) 265 FSNOTIFY_ITER_FUNCS(sb, SB) 266 267 #define fsnotify_foreach_obj_type(type) \ 268 for (type = 0; type < FSNOTIFY_OBJ_TYPE_COUNT; type++) 269 270 /* 271 * fsnotify_connp_t is what we embed in objects which connector can be attached 272 * to. fsnotify_connp_t * is how we refer from connector back to object. 273 */ 274 struct fsnotify_mark_connector; 275 typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t; 276 277 /* 278 * Inode/vfsmount/sb point to this structure which tracks all marks attached to 279 * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this 280 * structure. We destroy this structure when there are no more marks attached 281 * to it. The structure is protected by fsnotify_mark_srcu. 282 */ 283 struct fsnotify_mark_connector { 284 spinlock_t lock; 285 unsigned int type; /* Type of object [lock] */ 286 union { 287 /* Object pointer [lock] */ 288 fsnotify_connp_t *obj; 289 /* Used listing heads to free after srcu period expires */ 290 struct fsnotify_mark_connector *destroy_next; 291 }; 292 struct hlist_head list; 293 }; 294 295 /* 296 * A mark is simply an object attached to an in core inode which allows an 297 * fsnotify listener to indicate they are either no longer interested in events 298 * of a type matching mask or only interested in those events. 299 * 300 * These are flushed when an inode is evicted from core and may be flushed 301 * when the inode is modified (as seen by fsnotify_access). Some fsnotify 302 * users (such as dnotify) will flush these when the open fd is closed and not 303 * at inode eviction or modification. 304 * 305 * Text in brackets is showing the lock(s) protecting modifications of a 306 * particular entry. obj_lock means either inode->i_lock or 307 * mnt->mnt_root->d_lock depending on the mark type. 308 */ 309 struct fsnotify_mark { 310 /* Mask this mark is for [mark->lock, group->mark_mutex] */ 311 __u32 mask; 312 /* We hold one for presence in g_list. Also one ref for each 'thing' 313 * in kernel that found and may be using this mark. */ 314 refcount_t refcnt; 315 /* Group this mark is for. Set on mark creation, stable until last ref 316 * is dropped */ 317 struct fsnotify_group *group; 318 /* List of marks by group->marks_list. Also reused for queueing 319 * mark into destroy_list when it's waiting for the end of SRCU period 320 * before it can be freed. [group->mark_mutex] */ 321 struct list_head g_list; 322 /* Protects inode / mnt pointers, flags, masks */ 323 spinlock_t lock; 324 /* List of marks for inode / vfsmount [connector->lock, mark ref] */ 325 struct hlist_node obj_list; 326 /* Head of list of marks for an object [mark ref] */ 327 struct fsnotify_mark_connector *connector; 328 /* Events types to ignore [mark->lock, group->mark_mutex] */ 329 __u32 ignored_mask; 330 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x01 331 #define FSNOTIFY_MARK_FLAG_ALIVE 0x02 332 #define FSNOTIFY_MARK_FLAG_ATTACHED 0x04 333 unsigned int flags; /* flags [mark->lock] */ 334 }; 335 336 #ifdef CONFIG_FSNOTIFY 337 338 /* called from the vfs helpers */ 339 340 /* main fsnotify call to send events */ 341 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, 342 const unsigned char *name, u32 cookie); 343 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask); 344 extern void __fsnotify_inode_delete(struct inode *inode); 345 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt); 346 extern void fsnotify_sb_delete(struct super_block *sb); 347 extern u32 fsnotify_get_cookie(void); 348 349 static inline int fsnotify_inode_watches_children(struct inode *inode) 350 { 351 /* FS_EVENT_ON_CHILD is set if the inode may care */ 352 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD)) 353 return 0; 354 /* this inode might care about child events, does it care about the 355 * specific set of events that can happen on a child? */ 356 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD; 357 } 358 359 /* 360 * Update the dentry with a flag indicating the interest of its parent to receive 361 * filesystem events when those events happens to this dentry->d_inode. 362 */ 363 static inline void fsnotify_update_flags(struct dentry *dentry) 364 { 365 assert_spin_locked(&dentry->d_lock); 366 367 /* 368 * Serialisation of setting PARENT_WATCHED on the dentries is provided 369 * by d_lock. If inotify_inode_watched changes after we have taken 370 * d_lock, the following __fsnotify_update_child_dentry_flags call will 371 * find our entry, so it will spin until we complete here, and update 372 * us with the new state. 373 */ 374 if (fsnotify_inode_watches_children(dentry->d_parent->d_inode)) 375 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; 376 else 377 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; 378 } 379 380 /* called from fsnotify listeners, such as fanotify or dnotify */ 381 382 /* create a new group */ 383 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops); 384 /* get reference to a group */ 385 extern void fsnotify_get_group(struct fsnotify_group *group); 386 /* drop reference on a group from fsnotify_alloc_group */ 387 extern void fsnotify_put_group(struct fsnotify_group *group); 388 /* group destruction begins, stop queuing new events */ 389 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group); 390 /* destroy group */ 391 extern void fsnotify_destroy_group(struct fsnotify_group *group); 392 /* fasync handler function */ 393 extern int fsnotify_fasync(int fd, struct file *file, int on); 394 /* Free event from memory */ 395 extern void fsnotify_destroy_event(struct fsnotify_group *group, 396 struct fsnotify_event *event); 397 /* attach the event to the group notification queue */ 398 extern int fsnotify_add_event(struct fsnotify_group *group, 399 struct fsnotify_event *event, 400 int (*merge)(struct list_head *, 401 struct fsnotify_event *)); 402 /* Queue overflow event to a notification group */ 403 static inline void fsnotify_queue_overflow(struct fsnotify_group *group) 404 { 405 fsnotify_add_event(group, group->overflow_event, NULL); 406 } 407 408 /* true if the group notification queue is empty */ 409 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group); 410 /* return, but do not dequeue the first event on the notification queue */ 411 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group); 412 /* return AND dequeue the first event on the notification queue */ 413 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group); 414 415 /* functions used to manipulate the marks attached to inodes */ 416 417 /* Get mask of events for a list of marks */ 418 extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn); 419 /* Calculate mask of events for a list of marks */ 420 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn); 421 extern void fsnotify_init_mark(struct fsnotify_mark *mark, 422 struct fsnotify_group *group); 423 /* Find mark belonging to given group in the list of marks */ 424 extern struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp, 425 struct fsnotify_group *group); 426 /* attach the mark to the object */ 427 extern int fsnotify_add_mark(struct fsnotify_mark *mark, 428 fsnotify_connp_t *connp, unsigned int type, 429 int allow_dups); 430 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark, 431 fsnotify_connp_t *connp, unsigned int type, 432 int allow_dups); 433 /* attach the mark to the inode */ 434 static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark, 435 struct inode *inode, 436 int allow_dups) 437 { 438 return fsnotify_add_mark(mark, &inode->i_fsnotify_marks, 439 FSNOTIFY_OBJ_TYPE_INODE, allow_dups); 440 } 441 static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark, 442 struct inode *inode, 443 int allow_dups) 444 { 445 return fsnotify_add_mark_locked(mark, &inode->i_fsnotify_marks, 446 FSNOTIFY_OBJ_TYPE_INODE, allow_dups); 447 } 448 /* given a group and a mark, flag mark to be freed when all references are dropped */ 449 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark, 450 struct fsnotify_group *group); 451 /* detach mark from inode / mount list, group list, drop inode reference */ 452 extern void fsnotify_detach_mark(struct fsnotify_mark *mark); 453 /* free mark */ 454 extern void fsnotify_free_mark(struct fsnotify_mark *mark); 455 /* run all the marks in a group, and clear all of the marks attached to given object type */ 456 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type); 457 /* run all the marks in a group, and clear all of the vfsmount marks */ 458 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group) 459 { 460 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL); 461 } 462 /* run all the marks in a group, and clear all of the inode marks */ 463 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group) 464 { 465 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE_FL); 466 } 467 /* run all the marks in a group, and clear all of the sn marks */ 468 static inline void fsnotify_clear_sb_marks_by_group(struct fsnotify_group *group) 469 { 470 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_SB_FL); 471 } 472 extern void fsnotify_get_mark(struct fsnotify_mark *mark); 473 extern void fsnotify_put_mark(struct fsnotify_mark *mark); 474 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info); 475 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info); 476 477 /* put here because inotify does some weird stuff when destroying watches */ 478 extern void fsnotify_init_event(struct fsnotify_event *event, 479 struct inode *to_tell, u32 mask); 480 481 #else 482 483 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is, 484 const unsigned char *name, u32 cookie) 485 { 486 return 0; 487 } 488 489 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask) 490 { 491 return 0; 492 } 493 494 static inline void __fsnotify_inode_delete(struct inode *inode) 495 {} 496 497 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt) 498 {} 499 500 static inline void fsnotify_sb_delete(struct super_block *sb) 501 {} 502 503 static inline void fsnotify_update_flags(struct dentry *dentry) 504 {} 505 506 static inline u32 fsnotify_get_cookie(void) 507 { 508 return 0; 509 } 510 511 static inline void fsnotify_unmount_inodes(struct super_block *sb) 512 {} 513 514 #endif /* CONFIG_FSNOTIFY */ 515 516 #endif /* __KERNEL __ */ 517 518 #endif /* __LINUX_FSNOTIFY_BACKEND_H */ 519