include/linux/fsnotify_backend.h

   1 /*
   2  * Filesystem access notification for Linux
   3  *
   4  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
   5  */
   6
   7 #ifndef __LINUX_FSNOTIFY_BACKEND_H
   8 #define __LINUX_FSNOTIFY_BACKEND_H
   9
  10 #ifdef __KERNEL__
  11
  12 #include <linux/idr.h> /* inotify uses this */
  13 #include <linux/fs.h> /* struct inode */
  14 #include <linux/list.h>
  15 #include <linux/path.h> /* struct path */
  16 #include <linux/spinlock.h>
  17 #include <linux/types.h>
  18
  19 #include <linux/atomic.h>
  20
  21 /*
  22  * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
  23  * convert between them.  dnotify only needs conversion at watch creation
  24  * so no perf loss there.  fanotify isn't defined yet, so it can use the
  25  * wholes if it needs more events.
  26  */
  27 #define FS_ACCESS               0x00000001      /* File was accessed */
  28 #define FS_MODIFY               0x00000002      /* File was modified */
  29 #define FS_ATTRIB               0x00000004      /* Metadata changed */
  30 #define FS_CLOSE_WRITE          0x00000008      /* Writtable file was closed */
  31 #define FS_CLOSE_NOWRITE        0x00000010      /* Unwrittable file closed */
  32 #define FS_OPEN                 0x00000020      /* File was opened */
  33 #define FS_MOVED_FROM           0x00000040      /* File was moved from X */
  34 #define FS_MOVED_TO             0x00000080      /* File was moved to Y */
  35 #define FS_CREATE               0x00000100      /* Subfile was created */
  36 #define FS_DELETE               0x00000200      /* Subfile was deleted */
  37 #define FS_DELETE_SELF          0x00000400      /* Self was deleted */
  38 #define FS_MOVE_SELF            0x00000800      /* Self was moved */
  39
  40 #define FS_UNMOUNT              0x00002000      /* inode on umount fs */
  41 #define FS_Q_OVERFLOW           0x00004000      /* Event queued overflowed */
  42 #define FS_IN_IGNORED           0x00008000      /* last inotify event here */
  43
  44 #define FS_OPEN_PERM            0x00010000      /* open event in an permission hook */
  45 #define FS_ACCESS_PERM          0x00020000      /* access event in a permissions hook */
  46
  47 #define FS_EXCL_UNLINK          0x04000000      /* do not send events if object is unlinked */
  48 #define FS_ISDIR                0x40000000      /* event occurred against dir */
  49 #define FS_IN_ONESHOT           0x80000000      /* only send event once */
  50
  51 #define FS_DN_RENAME            0x10000000      /* file renamed */
  52 #define FS_DN_MULTISHOT         0x20000000      /* dnotify multishot */
  53
  54 /* This inode cares about things that happen to its children.  Always set for
  55  * dnotify and inotify. */
  56 #define FS_EVENT_ON_CHILD       0x08000000
  57
  58 /* This is a list of all events that may get sent to a parernt based on fs event
  59  * happening to inodes inside that directory */
  60 #define FS_EVENTS_POSS_ON_CHILD   (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
  61                                    FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
  62                                    FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
  63                                    FS_DELETE)
  64
  65 #define FS_MOVE                 (FS_MOVED_FROM | FS_MOVED_TO)
  66
  67 #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM)
  68
  69 #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
  70                              FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \
  71                              FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \
  72                              FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \
  73                              FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
  74                              FS_OPEN_PERM | FS_ACCESS_PERM | FS_EXCL_UNLINK | \
  75                              FS_ISDIR | FS_IN_ONESHOT | FS_DN_RENAME | \
  76                              FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
  77
  78 struct fsnotify_group;
  79 struct fsnotify_event;
  80 struct fsnotify_mark;
  81 struct fsnotify_event_private_data;
  82
  83 /*
  84  * Each group much define these ops.  The fsnotify infrastructure will call
  85  * these operations for each relevant group.
  86  *
  87  * should_send_event - given a group, inode, and mask this function determines
  88  *              if the group is interested in this event.
  89  * handle_event - main call for a group to handle an fs event
  90  * free_group_priv - called when a group refcnt hits 0 to clean up the private union
  91  * freeing-mark - this means that a mark has been flagged to die when everything
  92  *              finishes using it.  The function is supplied with what must be a
  93  *              valid group and inode to use to clean up.
  94  */
  95 struct fsnotify_ops {
  96         bool (*should_send_event)(struct fsnotify_group *group, struct inode *inode,
  97                                   struct fsnotify_mark *inode_mark,
  98                                   struct fsnotify_mark *vfsmount_mark,
  99                                   __u32 mask, void *data, int data_type);
 100         int (*handle_event)(struct fsnotify_group *group,
 101                             struct fsnotify_mark *inode_mark,
 102                             struct fsnotify_mark *vfsmount_mark,
 103                             struct fsnotify_event *event);
 104         void (*free_group_priv)(struct fsnotify_group *group);
 105         void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
 106         void (*free_event_priv)(struct fsnotify_event_private_data *priv);
 107 };
 108
 109 /*
 110  * A group is a "thing" that wants to receive notification about filesystem
 111  * events.  The mask holds the subset of event types this group cares about.
 112  * refcnt on a group is up to the implementor and at any moment if it goes 0
 113  * everything will be cleaned up.
 114  */
 115 struct fsnotify_group {
 116         /*
 117          * How the refcnt is used is up to each group.  When the refcnt hits 0
 118          * fsnotify will clean up all of the resources associated with this group.
 119          * As an example, the dnotify group will always have a refcnt=1 and that
 120          * will never change.  Inotify, on the other hand, has a group per
 121          * inotify_init() and the refcnt will hit 0 only when that fd has been
 122          * closed.
 123          */
 124         atomic_t refcnt;                /* things with interest in this group */
 125
 126         const struct fsnotify_ops *ops; /* how this group handles things */
 127
 128         /* needed to send notification to userspace */
 129         struct mutex notification_mutex;        /* protect the notification_list */
 130         struct list_head notification_list;     /* list of event_holder this group needs to send to userspace */
 131         wait_queue_head_t notification_waitq;   /* read() on the notification file blocks on this waitq */
 132         unsigned int q_len;                     /* events on the queue */
 133         unsigned int max_events;                /* maximum events allowed on the list */
 134         /*
 135          * Valid fsnotify group priorities.  Events are send in order from highest
 136          * priority to lowest priority.  We default to the lowest priority.
 137          */
 138         #define FS_PRIO_0       0 /* normal notifiers, no permissions */
 139         #define FS_PRIO_1       1 /* fanotify content based access control */
 140         #define FS_PRIO_2       2 /* fanotify pre-content access */
 141         unsigned int priority;
 142
 143         /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
 144         spinlock_t mark_lock;           /* protect marks_list */
 145         atomic_t num_marks;             /* 1 for each mark and 1 for not being
 146                                          * past the point of no return when freeing
 147                                          * a group */
 148         struct list_head marks_list;    /* all inode marks for this group */
 149
 150         /* groups can define private fields here or use the void *private */
 151         union {
 152                 void *private;
 153 #ifdef CONFIG_INOTIFY_USER
 154                 struct inotify_group_private_data {
 155                         spinlock_t      idr_lock;
 156                         struct idr      idr;
 157                         u32             last_wd;
 158                         struct fasync_struct    *fa;    /* async notification */
 159                         struct user_struct      *user;
 160                 } inotify_data;
 161 #endif
 162 #ifdef CONFIG_FANOTIFY
 163                 struct fanotify_group_private_data {
 164 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 165                         /* allows a group to block waiting for a userspace response */
 166                         struct mutex access_mutex;
 167                         struct list_head access_list;
 168                         wait_queue_head_t access_waitq;
 169                         atomic_t bypass_perm;
 170 #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
 171                         int f_flags;
 172                         unsigned int max_marks;
 173                         struct user_struct *user;
 174                 } fanotify_data;
 175 #endif /* CONFIG_FANOTIFY */
 176         };
 177 };
 178
 179 /*
 180  * A single event can be queued in multiple group->notification_lists.
 181  *
 182  * each group->notification_list will point to an event_holder which in turns points
 183  * to the actual event that needs to be sent to userspace.
 184  *
 185  * Seemed cheaper to create a refcnt'd event and a small holder for every group
 186  * than create a different event for every group
 187  *
 188  */
 189 struct fsnotify_event_holder {
 190         struct fsnotify_event *event;
 191         struct list_head event_list;
 192 };
 193
 194 /*
 195  * Inotify needs to tack data onto an event.  This struct lets us later find the
 196  * correct private data of the correct group.
 197  */
 198 struct fsnotify_event_private_data {
 199         struct fsnotify_group *group;
 200         struct list_head event_list;
 201 };
 202
 203 /*
 204  * all of the information about the original object we want to now send to
 205  * a group.  If you want to carry more info from the accessing task to the
 206  * listener this structure is where you need to be adding fields.
 207  */
 208 struct fsnotify_event {
 209         /*
 210          * If we create an event we are also likely going to need a holder
 211          * to link to a group.  So embed one holder in the event.  Means only
 212          * one allocation for the common case where we only have one group
 213          */
 214         struct fsnotify_event_holder holder;
 215         spinlock_t lock;        /* protection for the associated event_holder and private_list */
 216         /* to_tell may ONLY be dereferenced during handle_event(). */
 217         struct inode *to_tell;  /* either the inode the event happened to or its parent */
 218         /*
 219          * depending on the event type we should have either a path or inode
 220          * We hold a reference on path, but NOT on inode.  Since we have the ref on
 221          * the path, it may be dereferenced at any point during this object's
 222          * lifetime.  That reference is dropped when this object's refcnt hits
 223          * 0.  If this event contains an inode instead of a path, the inode may
 224          * ONLY be used during handle_event().
 225          */
 226         union {
 227                 struct path path;
 228                 struct inode *inode;
 229         };
 230 /* when calling fsnotify tell it if the data is a path or inode */
 231 #define FSNOTIFY_EVENT_NONE     0
 232 #define FSNOTIFY_EVENT_PATH     1
 233 #define FSNOTIFY_EVENT_INODE    2
 234         int data_type;          /* which of the above union we have */
 235         atomic_t refcnt;        /* how many groups still are using/need to send this event */
 236         __u32 mask;             /* the type of access, bitwise OR for FS_* event types */
 237
 238         u32 sync_cookie;        /* used to corrolate events, namely inotify mv events */
 239         const unsigned char *file_name;
 240         size_t name_len;
 241         struct pid *tgid;
 242
 243 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 244         __u32 response; /* userspace answer to question */
 245 #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
 246
 247         struct list_head private_data_list;     /* groups can store private data here */
 248 };
 249
 250 /*
 251  * Inode specific fields in an fsnotify_mark
 252  */
 253 struct fsnotify_inode_mark {
 254         struct inode *inode;            /* inode this mark is associated with */
 255         struct hlist_node i_list;       /* list of marks by inode->i_fsnotify_marks */
 256         struct list_head free_i_list;   /* tmp list used when freeing this mark */
 257 };
 258
 259 /*
 260  * Mount point specific fields in an fsnotify_mark
 261  */
 262 struct fsnotify_vfsmount_mark {
 263         struct vfsmount *mnt;           /* vfsmount this mark is associated with */
 264         struct hlist_node m_list;       /* list of marks by inode->i_fsnotify_marks */
 265         struct list_head free_m_list;   /* tmp list used when freeing this mark */
 266 };
 267
 268 /*
 269  * a mark is simply an object attached to an in core inode which allows an
 270  * fsnotify listener to indicate they are either no longer interested in events
 271  * of a type matching mask or only interested in those events.
 272  *
 273  * these are flushed when an inode is evicted from core and may be flushed
 274  * when the inode is modified (as seen by fsnotify_access).  Some fsnotify users
 275  * (such as dnotify) will flush these when the open fd is closed and not at
 276  * inode eviction or modification.
 277  */
 278 struct fsnotify_mark {
 279         __u32 mask;                     /* mask this mark is for */
 280         /* we hold ref for each i_list and g_list.  also one ref for each 'thing'
 281          * in kernel that found and may be using this mark. */
 282         atomic_t refcnt;                /* active things looking at this mark */
 283         struct fsnotify_group *group;   /* group this mark is for */
 284         struct list_head g_list;        /* list of marks by group->i_fsnotify_marks */
 285         spinlock_t lock;                /* protect group and inode */
 286         union {
 287                 struct fsnotify_inode_mark i;
 288                 struct fsnotify_vfsmount_mark m;
 289         };
 290         struct list_head free_g_list;   /* tmp list used when freeing this mark */
 291         __u32 ignored_mask;             /* events types to ignore */
 292 #define FSNOTIFY_MARK_FLAG_INODE                0x01
 293 #define FSNOTIFY_MARK_FLAG_VFSMOUNT             0x02
 294 #define FSNOTIFY_MARK_FLAG_OBJECT_PINNED        0x04
 295 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY  0x08
 296 #define FSNOTIFY_MARK_FLAG_ALIVE                0x10
 297         unsigned int flags;             /* vfsmount or inode mark? */
 298         struct list_head destroy_list;
 299         void (*free_mark)(struct fsnotify_mark *mark); /* called on final put+free */
 300 };
 301
 302 #ifdef CONFIG_FSNOTIFY
 303
 304 /* called from the vfs helpers */
 305
 306 /* main fsnotify call to send events */
 307 extern int fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
 308                     const unsigned char *name, u32 cookie);
 309 extern int __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask);
 310 extern void __fsnotify_inode_delete(struct inode *inode);
 311 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
 312 extern u32 fsnotify_get_cookie(void);
 313
 314 static inline int fsnotify_inode_watches_children(struct inode *inode)
 315 {
 316         /* FS_EVENT_ON_CHILD is set if the inode may care */
 317         if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
 318                 return 0;
 319         /* this inode might care about child events, does it care about the
 320          * specific set of events that can happen on a child? */
 321         return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
 322 }
 323
 324 /*
 325  * Update the dentry with a flag indicating the interest of its parent to receive
 326  * filesystem events when those events happens to this dentry->d_inode.
 327  */
 328 static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
 329 {
 330         struct dentry *parent;
 331
 332         assert_spin_locked(&dentry->d_lock);
 333
 334         /*
 335          * Serialisation of setting PARENT_WATCHED on the dentries is provided
 336          * by d_lock. If inotify_inode_watched changes after we have taken
 337          * d_lock, the following __fsnotify_update_child_dentry_flags call will
 338          * find our entry, so it will spin until we complete here, and update
 339          * us with the new state.
 340          */
 341         parent = dentry->d_parent;
 342         if (parent->d_inode && fsnotify_inode_watches_children(parent->d_inode))
 343                 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
 344         else
 345                 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
 346 }
 347
 348 /*
 349  * fsnotify_d_instantiate - instantiate a dentry for inode
 350  */
 351 static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
 352 {
 353         if (!inode)
 354                 return;
 355
 356         spin_lock(&dentry->d_lock);
 357         __fsnotify_update_dcache_flags(dentry);
 358         spin_unlock(&dentry->d_lock);
 359 }
 360
 361 /* called from fsnotify listeners, such as fanotify or dnotify */
 362
 363 /* get a reference to an existing or create a new group */
 364 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
 365 /* drop reference on a group from fsnotify_alloc_group */
 366 extern void fsnotify_put_group(struct fsnotify_group *group);
 367
 368 /* take a reference to an event */
 369 extern void fsnotify_get_event(struct fsnotify_event *event);
 370 extern void fsnotify_put_event(struct fsnotify_event *event);
 371 /* find private data previously attached to an event and unlink it */
 372 extern struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group,
 373                                                                            struct fsnotify_event *event);
 374
 375 /* attach the event to the group notification queue */
 376 extern struct fsnotify_event *fsnotify_add_notify_event(struct fsnotify_group *group,
 377                                                         struct fsnotify_event *event,
 378                                                         struct fsnotify_event_private_data *priv,
 379                                                         struct fsnotify_event *(*merge)(struct list_head *,
 380                                                                                         struct fsnotify_event *));
 381 /* true if the group notification queue is empty */
 382 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
 383 /* return, but do not dequeue the first event on the notification queue */
 384 extern struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group);
 385 /* return AND dequeue the first event on the notification queue */
 386 extern struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group);
 387
 388 /* functions used to manipulate the marks attached to inodes */
 389
 390 /* run all marks associated with a vfsmount and update mnt->mnt_fsnotify_mask */
 391 extern void fsnotify_recalc_vfsmount_mask(struct vfsmount *mnt);
 392 /* run all marks associated with an inode and update inode->i_fsnotify_mask */
 393 extern void fsnotify_recalc_inode_mask(struct inode *inode);
 394 extern void fsnotify_init_mark(struct fsnotify_mark *mark, void (*free_mark)(struct fsnotify_mark *mark));
 395 /* find (and take a reference) to a mark associated with group and inode */
 396 extern struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group, struct inode *inode);
 397 /* find (and take a reference) to a mark associated with group and vfsmount */
 398 extern struct fsnotify_mark *fsnotify_find_vfsmount_mark(struct fsnotify_group *group, struct vfsmount *mnt);
 399 /* copy the values from old into new */
 400 extern void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old);
 401 /* set the ignored_mask of a mark */
 402 extern void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask);
 403 /* set the mask of a mark (might pin the object into memory */
 404 extern void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask);
 405 /* attach the mark to both the group and the inode */
 406 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
 407                              struct inode *inode, struct vfsmount *mnt, int allow_dups);
 408 /* given a mark, flag it to be freed when all references are dropped */
 409 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark);
 410 /* run all the marks in a group, and clear all of the vfsmount marks */
 411 extern void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group);
 412 /* run all the marks in a group, and clear all of the inode marks */
 413 extern void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group);
 414 /* run all the marks in a group, and clear all of the marks where mark->flags & flags is true*/
 415 extern void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group, unsigned int flags);
 416 /* run all the marks in a group, and flag them to be freed */
 417 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group);
 418 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
 419 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
 420 extern void fsnotify_unmount_inodes(struct list_head *list);
 421
 422 /* put here because inotify does some weird stuff when destroying watches */
 423 extern struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask,
 424                                                     void *data, int data_is,
 425                                                     const unsigned char *name,
 426                                                     u32 cookie, gfp_t gfp);
 427
 428 /* fanotify likes to change events after they are on lists... */
 429 extern struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event);
 430 extern int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
 431                                   struct fsnotify_event *new_event);
 432
 433 #else
 434
 435 static inline int fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
 436                            const unsigned char *name, u32 cookie)
 437 {
 438         return 0;
 439 }
 440
 441 static inline int __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask)
 442 {
 443         return 0;
 444 }
 445
 446 static inline void __fsnotify_inode_delete(struct inode *inode)
 447 {}
 448
 449 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
 450 {}
 451
 452 static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
 453 {}
 454
 455 static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
 456 {}
 457
 458 static inline u32 fsnotify_get_cookie(void)
 459 {
 460         return 0;
 461 }
 462
 463 static inline void fsnotify_unmount_inodes(struct list_head *list)
 464 {}
 465
 466 #endif  /* CONFIG_FSNOTIFY */
 467
 468 #endif  /* __KERNEL __ */
 469
 470 #endif  /* __LINUX_FSNOTIFY_BACKEND_H */