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 | FS_OPEN_PERM | FS_ACCESS_PERM)
  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 - called when a mark is being destroyed for some reason.  The group
  92  *              MUST be holding a reference on each mark and that reference must be
  93  *              dropped in this function.  inotify uses this function to send
  94  *              userspace messages that marks have been removed.
  95  */
  96 struct fsnotify_ops {
  97         bool (*should_send_event)(struct fsnotify_group *group, struct inode *inode,
  98                                   struct fsnotify_mark *inode_mark,
  99                                   struct fsnotify_mark *vfsmount_mark,
 100                                   __u32 mask, void *data, int data_type);
 101         int (*handle_event)(struct fsnotify_group *group,
 102                             struct fsnotify_mark *inode_mark,
 103                             struct fsnotify_mark *vfsmount_mark,
 104                             struct fsnotify_event *event);
 105         void (*free_group_priv)(struct fsnotify_group *group);
 106         void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
 107         void (*free_event_priv)(struct fsnotify_event_private_data *priv);
 108 };
 109
 110 /*
 111  * A group is a "thing" that wants to receive notification about filesystem
 112  * events.  The mask holds the subset of event types this group cares about.
 113  * refcnt on a group is up to the implementor and at any moment if it goes 0
 114  * everything will be cleaned up.
 115  */
 116 struct fsnotify_group {
 117         /*
 118          * How the refcnt is used is up to each group.  When the refcnt hits 0
 119          * fsnotify will clean up all of the resources associated with this group.
 120          * As an example, the dnotify group will always have a refcnt=1 and that
 121          * will never change.  Inotify, on the other hand, has a group per
 122          * inotify_init() and the refcnt will hit 0 only when that fd has been
 123          * closed.
 124          */
 125         atomic_t refcnt;                /* things with interest in this group */
 126
 127         const struct fsnotify_ops *ops; /* how this group handles things */
 128
 129         /* needed to send notification to userspace */
 130         struct mutex notification_mutex;        /* protect the notification_list */
 131         struct list_head notification_list;     /* list of event_holder this group needs to send to userspace */
 132         wait_queue_head_t notification_waitq;   /* read() on the notification file blocks on this waitq */
 133         unsigned int q_len;                     /* events on the queue */
 134         unsigned int max_events;                /* maximum events allowed on the list */
 135         /*
 136          * Valid fsnotify group priorities.  Events are send in order from highest
 137          * priority to lowest priority.  We default to the lowest priority.
 138          */
 139         #define FS_PRIO_0       0 /* normal notifiers, no permissions */
 140         #define FS_PRIO_1       1 /* fanotify content based access control */
 141         #define FS_PRIO_2       2 /* fanotify pre-content access */
 142         unsigned int priority;
 143
 144         /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
 145         struct mutex mark_mutex;        /* protect marks_list */
 146         atomic_t num_marks;             /* 1 for each mark and 1 for not being
 147                                          * past the point of no return when freeing
 148                                          * a group */
 149         struct list_head marks_list;    /* all inode marks for this group */
 150
 151         struct fasync_struct    *fsn_fa;    /* async notification */
 152
 153         /* groups can define private fields here or use the void *private */
 154         union {
 155                 void *private;
 156 #ifdef CONFIG_INOTIFY_USER
 157                 struct inotify_group_private_data {
 158                         spinlock_t      idr_lock;
 159                         struct idr      idr;
 160                         struct user_struct      *user;
 161                 } inotify_data;
 162 #endif
 163 #ifdef CONFIG_FANOTIFY
 164                 struct fanotify_group_private_data {
 165 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 166                         /* allows a group to block waiting for a userspace response */
 167                         struct mutex access_mutex;
 168                         struct list_head access_list;
 169                         wait_queue_head_t access_waitq;
 170                         atomic_t bypass_perm;
 171 #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
 172                         int f_flags;
 173                         unsigned int max_marks;
 174                         struct user_struct *user;
 175                 } fanotify_data;
 176 #endif /* CONFIG_FANOTIFY */
 177         };
 178 };
 179
 180 /*
 181  * A single event can be queued in multiple group->notification_lists.
 182  *
 183  * each group->notification_list will point to an event_holder which in turns points
 184  * to the actual event that needs to be sent to userspace.
 185  *
 186  * Seemed cheaper to create a refcnt'd event and a small holder for every group
 187  * than create a different event for every group
 188  *
 189  */
 190 struct fsnotify_event_holder {
 191         struct fsnotify_event *event;
 192         struct list_head event_list;
 193 };
 194
 195 /*
 196  * Inotify needs to tack data onto an event.  This struct lets us later find the
 197  * correct private data of the correct group.
 198  */
 199 struct fsnotify_event_private_data {
 200         struct fsnotify_group *group;
 201         struct list_head event_list;
 202 };
 203
 204 /*
 205  * all of the information about the original object we want to now send to
 206  * a group.  If you want to carry more info from the accessing task to the
 207  * listener this structure is where you need to be adding fields.
 208  */
 209 struct fsnotify_event {
 210         /*
 211          * If we create an event we are also likely going to need a holder
 212          * to link to a group.  So embed one holder in the event.  Means only
 213          * one allocation for the common case where we only have one group
 214          */
 215         struct fsnotify_event_holder holder;
 216         spinlock_t lock;        /* protection for the associated event_holder and private_list */
 217         /* to_tell may ONLY be dereferenced during handle_event(). */
 218         struct inode *to_tell;  /* either the inode the event happened to or its parent */
 219         /*
 220          * depending on the event type we should have either a path or inode
 221          * We hold a reference on path, but NOT on inode.  Since we have the ref on
 222          * the path, it may be dereferenced at any point during this object's
 223          * lifetime.  That reference is dropped when this object's refcnt hits
 224          * 0.  If this event contains an inode instead of a path, the inode may
 225          * ONLY be used during handle_event().
 226          */
 227         union {
 228                 struct path path;
 229                 struct inode *inode;
 230         };
 231 /* when calling fsnotify tell it if the data is a path or inode */
 232 #define FSNOTIFY_EVENT_NONE     0
 233 #define FSNOTIFY_EVENT_PATH     1
 234 #define FSNOTIFY_EVENT_INODE    2
 235         int data_type;          /* which of the above union we have */
 236         atomic_t refcnt;        /* how many groups still are using/need to send this event */
 237         __u32 mask;             /* the type of access, bitwise OR for FS_* event types */
 238
 239         u32 sync_cookie;        /* used to corrolate events, namely inotify mv events */
 240         const unsigned char *file_name;
 241         size_t name_len;
 242         struct pid *tgid;
 243
 244 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 245         __u32 response; /* userspace answer to question */
 246 #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
 247
 248         struct list_head private_data_list;     /* groups can store private data here */
 249 };
 250
 251 /*
 252  * Inode specific fields in an fsnotify_mark
 253  */
 254 struct fsnotify_inode_mark {
 255         struct inode *inode;            /* inode this mark is associated with */
 256         struct hlist_node i_list;       /* list of marks by inode->i_fsnotify_marks */
 257         struct list_head free_i_list;   /* tmp list used when freeing this mark */
 258 };
 259
 260 /*
 261  * Mount point specific fields in an fsnotify_mark
 262  */
 263 struct fsnotify_vfsmount_mark {
 264         struct vfsmount *mnt;           /* vfsmount this mark is associated with */
 265         struct hlist_node m_list;       /* list of marks by inode->i_fsnotify_marks */
 266         struct list_head free_m_list;   /* tmp list used when freeing this mark */
 267 };
 268
 269 /*
 270  * a mark is simply an object attached to an in core inode which allows an
 271  * fsnotify listener to indicate they are either no longer interested in events
 272  * of a type matching mask or only interested in those events.
 273  *
 274  * these are flushed when an inode is evicted from core and may be flushed
 275  * when the inode is modified (as seen by fsnotify_access).  Some fsnotify users
 276  * (such as dnotify) will flush these when the open fd is closed and not at
 277  * inode eviction or modification.
 278  */
 279 struct fsnotify_mark {
 280         __u32 mask;                     /* mask this mark is for */
 281         /* we hold ref for each i_list and g_list.  also one ref for each 'thing'
 282          * in kernel that found and may be using this mark. */
 283         atomic_t refcnt;                /* active things looking at this mark */
 284         struct fsnotify_group *group;   /* group this mark is for */
 285         struct list_head g_list;        /* list of marks by group->i_fsnotify_marks */
 286         spinlock_t lock;                /* protect group and inode */
 287         union {
 288                 struct fsnotify_inode_mark i;
 289                 struct fsnotify_vfsmount_mark m;
 290         };
 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 /* create a new group */
 364 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
 365 /* get reference to a group */
 366 extern void fsnotify_get_group(struct fsnotify_group *group);
 367 /* drop reference on a group from fsnotify_alloc_group */
 368 extern void fsnotify_put_group(struct fsnotify_group *group);
 369 /* destroy group */
 370 extern void fsnotify_destroy_group(struct fsnotify_group *group);
 371 /* fasync handler function */
 372 extern int fsnotify_fasync(int fd, struct file *file, int on);
 373 /* take a reference to an event */
 374 extern void fsnotify_get_event(struct fsnotify_event *event);
 375 extern void fsnotify_put_event(struct fsnotify_event *event);
 376 /* find private data previously attached to an event and unlink it */
 377 extern struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnotify_group *group,
 378                                                                            struct fsnotify_event *event);
 379
 380 /* attach the event to the group notification queue */
 381 extern struct fsnotify_event *fsnotify_add_notify_event(struct fsnotify_group *group,
 382                                                         struct fsnotify_event *event,
 383                                                         struct fsnotify_event_private_data *priv,
 384                                                         struct fsnotify_event *(*merge)(struct list_head *,
 385                                                                                         struct fsnotify_event *));
 386 /* true if the group notification queue is empty */
 387 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
 388 /* return, but do not dequeue the first event on the notification queue */
 389 extern struct fsnotify_event *fsnotify_peek_notify_event(struct fsnotify_group *group);
 390 /* return AND dequeue the first event on the notification queue */
 391 extern struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group);
 392
 393 /* functions used to manipulate the marks attached to inodes */
 394
 395 /* run all marks associated with a vfsmount and update mnt->mnt_fsnotify_mask */
 396 extern void fsnotify_recalc_vfsmount_mask(struct vfsmount *mnt);
 397 /* run all marks associated with an inode and update inode->i_fsnotify_mask */
 398 extern void fsnotify_recalc_inode_mask(struct inode *inode);
 399 extern void fsnotify_init_mark(struct fsnotify_mark *mark, void (*free_mark)(struct fsnotify_mark *mark));
 400 /* find (and take a reference) to a mark associated with group and inode */
 401 extern struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group, struct inode *inode);
 402 /* find (and take a reference) to a mark associated with group and vfsmount */
 403 extern struct fsnotify_mark *fsnotify_find_vfsmount_mark(struct fsnotify_group *group, struct vfsmount *mnt);
 404 /* copy the values from old into new */
 405 extern void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old);
 406 /* set the ignored_mask of a mark */
 407 extern void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask);
 408 /* set the mask of a mark (might pin the object into memory */
 409 extern void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask);
 410 /* attach the mark to both the group and the inode */
 411 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
 412                              struct inode *inode, struct vfsmount *mnt, int allow_dups);
 413 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark, struct fsnotify_group *group,
 414                                     struct inode *inode, struct vfsmount *mnt, int allow_dups);
 415 /* given a group and a mark, flag mark to be freed when all references are dropped */
 416 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
 417                                   struct fsnotify_group *group);
 418 extern void fsnotify_destroy_mark_locked(struct fsnotify_mark *mark,
 419                                          struct fsnotify_group *group);
 420 /* run all the marks in a group, and clear all of the vfsmount marks */
 421 extern void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group);
 422 /* run all the marks in a group, and clear all of the inode marks */
 423 extern void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group);
 424 /* run all the marks in a group, and clear all of the marks where mark->flags & flags is true*/
 425 extern void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group, unsigned int flags);
 426 /* run all the marks in a group, and flag them to be freed */
 427 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group);
 428 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
 429 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
 430 extern void fsnotify_unmount_inodes(struct list_head *list);
 431
 432 /* put here because inotify does some weird stuff when destroying watches */
 433 extern struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask,
 434                                                     void *data, int data_is,
 435                                                     const unsigned char *name,
 436                                                     u32 cookie, gfp_t gfp);
 437
 438 /* fanotify likes to change events after they are on lists... */
 439 extern struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event);
 440 extern int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
 441                                   struct fsnotify_event *new_event);
 442
 443 #else
 444
 445 static inline int fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
 446                            const unsigned char *name, u32 cookie)
 447 {
 448         return 0;
 449 }
 450
 451 static inline int __fsnotify_parent(struct path *path, struct dentry *dentry, __u32 mask)
 452 {
 453         return 0;
 454 }
 455
 456 static inline void __fsnotify_inode_delete(struct inode *inode)
 457 {}
 458
 459 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
 460 {}
 461
 462 static inline void __fsnotify_update_dcache_flags(struct dentry *dentry)
 463 {}
 464
 465 static inline void __fsnotify_d_instantiate(struct dentry *dentry, struct inode *inode)
 466 {}
 467
 468 static inline u32 fsnotify_get_cookie(void)
 469 {
 470         return 0;
 471 }
 472
 473 static inline void fsnotify_unmount_inodes(struct list_head *list)
 474 {}
 475
 476 #endif  /* CONFIG_FSNOTIFY */
 477
 478 #endif  /* __KERNEL __ */
 479
 480 #endif  /* __LINUX_FSNOTIFY_BACKEND_H */