1 .\" Copyright (C) 2013, Heinrich Schuchardt <xypron.glpk@gmx.de>
2 .\" and Copyright (C) 2014, Michael Kerrisk <mtk.manpages@gmail.com>
4 .\" %%%LICENSE_START(VERBATIM)
5 .\" Permission is granted to make and distribute verbatim copies of this
6 .\" manual provided the copyright notice and this permission notice are
7 .\" preserved on all copies.
9 .\" Permission is granted to copy and distribute modified versions of
10 .\" this manual under the conditions for verbatim copying, provided that
11 .\" the entire resulting derived work is distributed under the terms of
12 .\" a permission notice identical to this one.
14 .\" Since the Linux kernel and libraries are constantly changing, this
15 .\" manual page may be incorrect or out-of-date. The author(s) assume.
16 .\" no responsibility for errors or omissions, or for damages resulting.
17 .\" from the use of the information contained herein. The author(s) may.
18 .\" not have taken the same level of care in the production of this.
19 .\" manual, which is licensed free of charge, as they might when working.
22 .\" Formatted or processed versions of this manual, if unaccompanied by
23 .\" the source, must acknowledge the copyright and authors of this work.
25 .TH FANOTIFY 7 2014-04-24 "Linux" "Linux Programmer's Manual"
27 fanotify \- monitoring filesystem events
29 The fanotify API provides notification and interception of
31 Use cases include virus scanning and hierarchical storage management.
32 Currently, only a limited set of events is supported.
33 In particular, there is no support for create, delete, and move events.
36 for details of an API that does notify those events.)
38 Additional capabilities compared to the
40 API include the ability to monitor all of the objects
41 in a mounted filesystem,
42 the ability to make access permission decisions, and the
43 possibility to read or modify files before access by other applications.
45 The following system calls are used with this API:
46 .BR fanotify_init (2),
47 .BR fanotify_mark (2),
52 .SS fanotify_init(), fanotify_mark(), and notification groups
55 system call creates and initializes an fanotify notification group
56 and returns a file descriptor referring to it.
58 An fanotify notification group is a kernel-internal object that holds
59 a list of files, directories, and mount points for which events shall be
62 For each entry in an fanotify notification group, two bit masks exist: the
67 The mark mask defines file activities for which an event shall be created.
68 The ignore mask defines activities for which no event shall be generated.
69 Having these two types of masks permits a mount point or directory to be
70 marked for receiving events, while at the same time ignoring events for
71 specific objects under that mount point or directory.
75 system call adds a file, directory, or mount to a notification group
76 and specifies which events
77 shall be reported (or ignored), or removes or modifies such an entry.
79 A possible usage of the ignore mask is for a file cache.
80 Events of interest for a file cache are modification of a file and closing
82 Hence, the cached directory or mount point is to be marked to receive these
84 After receiving the first event informing that a file has been modified,
85 the corresponding cache entry will be invalidated.
86 No further modification events for this file are of interest until the file
88 Hence, the modify event can be added to the ignore mask.
89 Upon receiving the close event, the modify event can be removed from the
90 ignore mask and the file cache entry can be updated.
92 The entries in the fanotify notification groups refer to files and
93 directories via their inode number and to mounts via their mount ID.
94 If files or directories are renamed or moved,
95 the respective entries survive.
96 If files or directories are deleted or mounts are unmounted,
97 the corresponding entries are deleted.
99 As events occur on the filesystem objects monitored by a notification group,
100 the fanotify system generates events that are collected in a queue.
101 These events can then be read (using
104 from the fanotify file descriptor
106 .BR fanotify_init (2).
108 Two types of events are generated:
113 Notification events are merely informative
114 and require no action to be taken by
115 the receiving application except for closing the file descriptor passed
116 in the event (see below).
117 Permission events are requests to the receiving application to decide
118 whether permission for a file access shall be granted.
119 For these events, the recipient must write a response which decides whether
120 access is granted or not.
122 An event is removed from the event queue of the fanotify group
123 when it has been read.
124 Permission events that have been read are kept in an internal list of the
125 fanotify group until either a permission decision has been taken by
126 writing to the fanotify file descriptor or the fanotify file descriptor
128 .SS Reading fanotify events
131 for the file descriptor returned by
132 .BR fanotify_init (2)
135 is not specified in the call to
136 .BR fanotify_init (2))
137 until either a file event occurs or the call is interrupted by a signal
143 the read buffer contains one or more of the following structures:
147 struct fanotify_event_metadata {
159 For performance reasons, it is recommended to use a large
160 buffer size (for example, 4096 bytes),
161 so that multiple events can be retrieved by a single
166 is the number of bytes placed in the buffer,
167 or \-1 in case of an error (but see BUGS).
170 .I fanotify_event_metadata
171 structure are as follows:
174 This is the length of the data for the current event and the offset
175 to the next event in the buffer.
176 In the current implementation, the value of
179 .BR FAN_EVENT_METADATA_LEN .
180 However, the API is designed to allow
181 variable-length structures to be returned in the future.
184 This field holds a version number for the structure.
185 It must be compared to
186 .B FANOTIFY_METADATA_VERSION
187 to verify that the structures returned at runtime match
188 the structures defined at compile time.
189 In case of a mismatch, the application should abandon trying to use the
190 fanotify file descriptor.
193 This field is not used.
196 This is the length of the structure.
197 The field was introduced to facilitate the implementation of
198 optional headers per event type.
199 No such optional headers exist in the current implementation.
202 This is a bit mask describing the event (see below).
205 This is an open file descriptor for the object being accessed, or
207 if a queue overflow occurred.
208 The file descriptor can be used to access the contents
209 of the monitored file or directory.
210 The reading application is responsible for closing this file descriptor.
213 .BR fanotify_init (2),
214 the caller may specify (via the
216 argument) various file status flags that are to be set
217 on the open file description that corresponds to this file descriptor.
218 In addition, the (kernel-internal)
220 file status flag is set on the open file description.
221 This flag suppresses fanotify event generation.
222 Hence, when the receiver of the fanotify event accesses the notified file or
223 directory using this file descriptor, no additional events will be created.
226 This is the ID of the process that caused the event.
227 A program listening to fanotify events can compare this PID
228 to the PID returned by
230 to determine whether the event is caused by the listener itself,
231 or is due to a file access by another process.
235 indicates which events have occurred for a single filesystem object.
236 Multiple bits may be set in this mask,
237 if more than one event occurred for the monitored filesystem object.
239 consecutive events for the same filesystem object and originating from the
240 same process may be merged into a single event, with the exception that two
241 permission events are never merged into one queue entry.
243 The bits that may appear in
248 A file or a directory (but see BUGS) was accessed (read).
251 A file or a directory was opened.
257 A file that was opened for writing
264 A file or directory that was opened read-only
269 The event queue exceeded the limit of 16384 entries.
270 This limit can be overridden by specifying the
271 .BR FAN_UNLIMITED_QUEUE
273 .BR fanotify_init (2).
276 An application wants to read a file or directory, for example using
280 The reader must write a response (as described below)
281 that determines whether the permission to
282 access the filesystem object shall be granted.
285 An application wants to open a file or directory.
286 The reader must write a response that determines whether the permission to
287 open the filesystem object shall be granted.
289 To check for any close event, the following bit mask may be used:
293 This is a synonym for:
295 FAN_CLOSE_WRITE | FAN_CLOSE_NOWRITE
297 The following macros are provided to iterate over a buffer containing
298 fanotify event metadata returned by a
300 from an fanotify file descriptor:
302 .B FAN_EVENT_OK(meta, len)
303 This macro checks the remaining length
307 against the length of the metadata structure and the
309 field of the first metadata structure in the buffer.
311 .B FAN_EVENT_NEXT(meta, len)
312 This macro sets the pointer
314 to the next metadata structure using the length indicated in the
316 field of the metadata structure and reduces the remaining length of the
319 .SS Monitoring an fanotify file descriptor for events
320 When an fanotify event occurs, the fanotify file descriptor indicates as
321 readable when passed to
326 .SS Dealing with permission events
327 For permission events, the application must
329 a structure of the following form to the
330 fanotify file descriptor:
334 struct fanotify_response {
341 The fields of this structure are as follows:
344 This is the file descriptor from the structure
345 .IR fanotify_event_metadata .
348 This field indicates whether or not the permission is to be granted.
349 Its value must be either
351 to allow the file operation or
353 to deny the file operation.
355 If access is denied, the requesting application call will receive an
358 .SS Closing the fanotify file descriptor
360 When all file descriptors referring to the fanotify notification group are
361 closed, the fanotify group is released and its resources
362 are freed for reuse by the kernel.
365 outstanding permission events will be set to allowed.
366 .SS /proc/[pid]/fdinfo
368 .I /proc/[pid]/fdinfo/[fd]
369 contains information about fanotify marks for file descriptor
373 See the kernel source file
374 .I Documentation/filesystems/proc.txt
377 In addition to the usual errors for
379 the following errors can occur when reading from the
380 fanotify file descriptor:
383 The buffer is too small to hold the event.
386 The per-process limit on the number of open files has been reached.
387 See the description of
393 The system-wide limit on the number of open files has been reached.
395 .I /proc/sys/fs/file-max
400 This error is returned by
408 argument when calling
409 .BR fanotify_init (2)
410 and an event occurred for a monitored file that is currently being executed.
412 In addition to the usual errors for
414 the following errors can occur when writing to the fanotify file descriptor:
417 Fanotify access permissions are not enabled in the kernel configuration
420 in the response structure is not valid.
425 in the response structure is not valid.
426 This may occur when a response for the permission event has already been
429 The fanotify API was introduced in version 2.6.36 of the Linux kernel and
430 enabled in version 2.6.37.
431 Fdinfo support was added in version 3.8.
433 The fanotify API is Linux-specific.
435 The fanotify API is available only if the kernel was built with the
437 configuration option enabled.
438 In addition, fanotify permission handling is available only if the
439 .B CONFIG_FANOTIFY_ACCESS_PERMISSIONS
440 configuration option is enabled.
441 .SS Limitations and caveats
442 Fanotify reports only events that a user-space program triggers through the
445 it does not catch remote events that occur on network filesystems.
447 The fanotify API does not report file accesses and modifications that
454 Events for directories are created only if the directory itself is opened,
456 Adding, removing, or changing children of a marked directory does not create
457 events for the monitored directory itself.
459 Fanotify monitoring of directories is not recursive:
460 to monitor subdirectories under a directory,
461 additional marks must be created.
462 (But note that the fanotify API provides no way of detecting when a
463 subdirectory has been created under a marked directory,
464 which makes recursive monitoring difficult.)
465 Monitoring mounts offers the capability to monitor a whole directory tree.
467 The event queue can overflow.
468 In this case, events are lost.
471 the following bugs exists:
473 .\" FIXME: A patch was proposed.
474 When an event is generated,
475 no check is made to see whether the user ID of the
476 receiving process has authorization to read or write the file
477 before passing a file descriptor for that file.
478 This poses a security risk, when the
480 capability is set for programs executed by unprivileged users.
484 processes multiple events from the fanotify queue and an error occurs, the
485 return value will be the cumulated length of the events successfully copied to
486 the user space buffer before the error occured.
487 The return value will not be \-1.
490 Thus, the user program has no way to detect the error.
492 The following program demonstrates the usage of the fanotify API.
493 It marks the mount point passed as a command-line argument
494 and waits for events of type
497 .BR FAN_CLOSE_WRITE .
498 When a permission event occurs, a
502 The following output was recorded while editing the file
503 .IR /home/user/temp/notes .
504 Before the file was opened, a
507 After the file was closed, a
510 Execution of the program ends when the user presses the ENTER key.
514 # ./fanotify_example /home
515 Press enter key to terminate.
516 Listening for events.
517 FAN_OPEN_PERM: File /home/user/temp/notes
518 FAN_CLOSE_WRITE: File /home/user/temp/notes
520 Listening for events stopped.
525 #define _GNU_SOURCE /* Needed to get O_LARGEFILE definition */
532 #include <sys/fanotify.h>
535 /* Read all available fanotify events from the file descriptor 'fd' */
538 handle_events(int fd)
540 const struct fanotify_event_metadata *metadata;
545 char procfd_path[PATH_MAX];
546 struct fanotify_response response;
548 /* Loop while events can be read from fanotify file descriptor */
552 /* Read some events */
554 len = read(fd, (void *) &buf, sizeof(buf));
555 if (len == \-1 && errno != EAGAIN) {
560 /* Check if end of available data reached */
565 /* Point to the first event in the buffer */
567 metadata = (struct fanotify_event_metadata *) buf;
569 /* Loop over all events in the buffer */
571 while (FAN_EVENT_OK(metadata, len)) {
573 /* Check that run\-time and compile\-time structures match */
575 if (metadata\->vers != FANOTIFY_METADATA_VERSION) {
577 "Mismatch of fanotify metadata version.\\n");
581 /* metadata\->fd contains either FAN_NOFD, indicating a
582 queue overflow, or a file descriptor (a nonnegative
583 integer). Here, we simply ignore queue overflow. */
585 if (metadata\->fd >= 0) {
587 /* Handle open permission event */
589 if (metadata\->mask & FAN_OPEN_PERM) {
590 printf("FAN_OPEN_PERM: ");
592 /* Allow file to be opened */
594 response.fd = metadata\->fd;
595 response.response = FAN_ALLOW;
597 sizeof(struct fanotify_response));
600 /* Handle closing of writable file event */
602 if (metadata\->mask & FAN_CLOSE_WRITE)
603 printf("FAN_CLOSE_WRITE: ");
605 /* Retrieve and print pathname of the accessed file */
607 snprintf(procfd_path, sizeof(procfd_path),
608 "/proc/self/fd/%d", metadata\->fd);
609 path_len = readlink(procfd_path, path,
611 if (path_len == \-1) {
616 path[path_len] = '\\0';
617 printf("File %s\\n", path);
619 /* Close the file descriptor of the event */
621 close(metadata\->fd);
624 /* Advance to next event */
626 metadata = FAN_EVENT_NEXT(metadata, len);
632 main(int argc, char *argv[])
637 struct pollfd fds[2];
639 /* Check mount point is supplied */
642 fprintf(stderr, "Usage: %s MOUNT\\n", argv[0]);
646 printf("Press enter key to terminate.\\n");
648 /* Create the file descriptor for accessing the fanotify API */
650 fd = fanotify_init(FAN_CLOEXEC | FAN_CLASS_CONTENT | FAN_NONBLOCK,
651 O_RDONLY | O_LARGEFILE);
653 perror("fanotify_init");
657 /* Mark the mount for:
658 \- permission events before opening files
659 \- notification events after closing a write\-enabled
662 if (fanotify_mark(fd, FAN_MARK_ADD | FAN_MARK_MOUNT,
663 FAN_OPEN_PERM | FAN_CLOSE_WRITE, \-1,
665 perror("fanotify_mark");
669 /* Prepare for polling */
675 fds[0].fd = STDIN_FILENO;
676 fds[0].events = POLLIN;
681 fds[1].events = POLLIN;
683 /* This is the loop to wait for incoming events */
685 printf("Listening for events.\\n");
688 poll_num = poll(fds, nfds, \-1);
689 if (poll_num == \-1) {
690 if (errno == EINTR) /* Interrupted by a signal */
691 continue; /* Restart poll() */
693 perror("poll"); /* Unexpected error */
698 if (fds[0].revents & POLLIN) {
700 /* Console input is available: empty stdin and quit */
702 while (read(STDIN_FILENO, &buf, 1) > 0 && buf != '\\n')
707 if (fds[1].revents & POLLIN) {
709 /* Fanotify events are available */
716 printf("Listening for events stopped.\\n");
722 .BR fanotify_init (2),
723 .BR fanotify_mark (2),