fanotify - monitoring filesystem events

The fanotify API provides notification and interception of filesystem events. Use cases include virus scanning and hierarchical storage management. In the original fanotify API, only a limited set of events was supported. In particular, there was no support for create, delete, and move events. The support for those events was added in Linux 5.1. (See inotify(7) for details of an API that did notify those events pre Linux 5.1.)

Additional capabilities compared to the inotify(7) API include the ability to monitor all of the objects in a mounted filesystem, the ability to make access permission decisions, and the possibility to read or modify files before access by other applications.

The following system calls are used with this API: fanotify_init(2), fanotify_mark(2), read(2), write(2), and close(2).

fanotify_init(), fanotify_mark(), and notification groups

The fanotify_init(2) system call creates and initializes an fanotify notification group and returns a file descriptor referring to it.

An fanotify notification group is a kernel-internal object that holds a list of files, directories, filesystems, and mounts for which events shall be created.

For each entry in an fanotify notification group, two bit masks exist: the mark mask and the ignore mask. The mark mask defines file activities for which an event shall be created. The ignore mask defines activities for which no event shall be generated. Having these two types of masks permits a filesystem, mount, or directory to be marked for receiving events, while at the same time ignoring events for specific objects under a mount or directory.

The fanotify_mark(2) system call adds a file, directory, filesystem, or mount to a notification group and specifies which events shall be reported (or ignored), or removes or modifies such an entry.

A possible usage of the ignore mask is for a file cache. Events of interest for a file cache are modification of a file and closing of the same. Hence, the cached directory or mount is to be marked to receive these events. After receiving the first event informing that a file has been modified, the corresponding cache entry will be invalidated. No further modification events for this file are of interest until the file is closed. Hence, the modify event can be added to the ignore mask. Upon receiving the close event, the modify event can be removed from the ignore mask and the file cache entry can be updated.

The entries in the fanotify notification groups refer to files and directories via their inode number and to mounts via their mount ID. If files or directories are renamed or moved within the same mount, the respective entries survive. If files or directories are deleted or moved to another mount or if filesystems or mounts are unmounted, the corresponding entries are deleted.

The event queue

As events occur on the filesystem objects monitored by a notification group, the fanotify system generates events that are collected in a queue. These events can then be read (using read(2) or similar) from the fanotify file descriptor returned by fanotify_init(2).

Two types of events are generated: notification events and permission events. Notification events are merely informative and require no action to be taken by the receiving application with one exception: if a valid file descriptor is provided within a generic event, the file descriptor must be closed. Permission events are requests to the receiving application to decide whether permission for a file access shall be granted. For these events, the recipient must write a response which decides whether access is granted or not.

An event is removed from the event queue of the fanotify group when it has been read. Permission events that have been read are kept in an internal list of the fanotify group until either a permission decision has been taken by writing to the fanotify file descriptor or the fanotify file descriptor is closed.

Reading fanotify events

Calling read(2) for the file descriptor returned by fanotify_init(2) blocks (if the flag FAN_NONBLOCK is not specified in the call to fanotify_init(2)) until either a file event occurs or the call is interrupted by a signal (see signal(7)).

After a successful read(2), the read buffer contains one or more of the following structures:

struct fanotify_event_metadata {
    __u32 event_len;
    __u8 vers;
    __u8 reserved;
    __u16 metadata_len;
    __aligned_u64 mask;
    __s32 fd;
    __s32 pid;
};

Information records are supplemental pieces of information that may be provided alongside the generic fanotify_event_metadata structure. The flags passed to fanotify_init(2) have influence over the type of information records that may be returned for an event. For example, if a notification group is initialized with FAN_REPORT_FID or FAN_REPORT_DIR_FID, then event listeners should also expect to receive a fanotify_event_info_fid structure alongside the fanotify_event_metadata structure, whereby file handles are used to identify filesystem objects rather than file descriptors. Information records may also be stacked, meaning that using the various FAN_REPORT_* flags in conjunction with one another is supported. In such cases, multiple information records can be returned for an event alongside the generic fanotify_event_metadata structure. For example, if a notification group is initialized with FAN_REPORT_TARGET_FID and FAN_REPORT_PIDFD, then an event listener should expect to receive up to two fanotify_event_info_fid information records and one fanotify_event_info_pidfd information record alongside the generic fanotify_event_metadata structure. Importantly, fanotify provides no guarantee around the ordering of information records when a notification group is initialized with a stacked based configuration. Each information record has a nested structure of type fanotify_event_info_header. It is imperative for event listeners to inspect the info_type field of this structure in order to determine the type of information record that had been received for a given event.

In cases where an fanotify group identifies filesystem objects by file handles, event listeners should also expect to receive one or more of the below information record objects alongside the generic fanotify_event_metadata structure within the read buffer:

struct fanotify_event_info_fid {
    struct fanotify_event_info_header hdr;
    __kernel_fsid_t fsid;
    unsigned char handle[];
};

In cases where an fanotify group is initialized with FAN_REPORT_PIDFD, event listeners should expect to receive the below information record object alongside the generic fanotify_event_metadata structure within the read buffer:

struct fanotify_event_info_pidfd {
        struct fanotify_event_info_header hdr;
        __s32 pidfd;
};

In case of a FAN_FS_ERROR event, an additional information record describing the error that occurred is returned alongside the generic fanotify_event_metadata structure within the read buffer. This structure is defined as follows:

struct fanotify_event_info_error {
    struct fanotify_event_info_header hdr;
    __s32 error;
    __u32 error_count;
};

All information records contain a nested structure of type fanotify_event_info_header. This structure holds meta-information about the information record that may have been returned alongside the generic fanotify_event_metadata structure. This structure is defined as follows:

struct fanotify_event_info_header {
	__u8 info_type;
	__u8 pad;
	__u16 len;
};

For performance reasons, it is recommended to use a large buffer size (for example, 4096 bytes), so that multiple events can be retrieved by a single read(2).

The return value of read(2) is the number of bytes placed in the buffer, or -1 in case of an error (but see BUGS).

The fields of the fanotify_event_metadata structure are as follows:

event_len

This is the length of the data for the current event and the offset to the next event in the buffer. Unless the group identifies filesystem objects by file handles, the value of event_len is always FAN_EVENT_METADATA_LEN. For a group that identifies filesystem objects by file handles, event_len also includes the variable length file identifier records.

vers

This field holds a version number for the structure. It must be compared to FANOTIFY_METADATA_VERSION to verify that the structures returned at run time match the structures defined at compile time. In case of a mismatch, the application should abandon trying to use the fanotify file descriptor.

reserved

This field is not used.

metadata_len

This is the length of the structure. The field was introduced to facilitate the implementation of optional headers per event type. No such optional headers exist in the current implementation.

mask

This is a bit mask describing the event (see below).

fd

This is an open file descriptor for the object being accessed, or FAN_NOFD if a queue overflow occurred. With an fanotify group that identifies filesystem objects by file handles, applications should expect this value to be set to FAN_NOFD for each event that is received. The file descriptor can be used to access the contents of the monitored file or directory. The reading application is responsible for closing this file descriptor.

When calling fanotify_init(2), the caller may specify (via the event_f_flags argument) various file status flags that are to be set on the open file description that corresponds to this file descriptor. In addition, the (kernel-internal) FMODE_NONOTIFY file status flag is set on the open file description. This flag suppresses fanotify event generation. Hence, when the receiver of the fanotify event accesses the notified file or directory using this file descriptor, no additional events will be created.

pid

If flag FAN_REPORT_TID was set in fanotify_init(2), this is the TID of the thread that caused the event. Otherwise, this the PID of the process that caused the event.

A program listening to fanotify events can compare this PID to the PID returned by getpid(2), to determine whether the event is caused by the listener itself, or is due to a file access by another process.

The bit mask in mask indicates which events have occurred for a single filesystem object. Multiple bits may be set in this mask, if more than one event occurred for the monitored filesystem object. In particular, consecutive events for the same filesystem object and originating from the same process may be merged into a single event, with the exception that two permission events are never merged into one queue entry.

The bits that may appear in mask are as follows:

FAN_ACCESS

A file or a directory (but see BUGS) was accessed (read).

FAN_OPEN

A file or a directory was opened.

FAN_OPEN_EXEC

A file was opened with the intent to be executed. See NOTES in fanotify_mark(2) for additional details.

FAN_ATTRIB

A file or directory metadata was changed.

FAN_CREATE

A child file or directory was created in a watched parent.

FAN_DELETE

A child file or directory was deleted in a watched parent.

FAN_DELETE_SELF

A watched file or directory was deleted.

FAN_FS_ERROR

A filesystem error was detected.

FAN_RENAME

A file or directory has been moved to or from a watched parent directory.

FAN_MOVED_FROM

A file or directory has been moved from a watched parent directory.

FAN_MOVED_TO

A file or directory has been moved to a watched parent directory.

FAN_MOVE_SELF

A watched file or directory was moved.

FAN_MODIFY

A file was modified.

FAN_CLOSE_WRITE

A file that was opened for writing (O_WRONLY or O_RDWR) was closed.

FAN_CLOSE_NOWRITE

A file or directory that was opened read-only (O_RDONLY) was closed.

FAN_Q_OVERFLOW

The event queue exceeded the limit on number of events. This limit can be overridden by specifying the FAN_UNLIMITED_QUEUE flag when calling fanotify_init(2).

FAN_ACCESS_PERM

An application wants to read a file or directory, for example using read(2) or readdir(2). The reader must write a response (as described below) that determines whether the permission to access the filesystem object shall be granted.

FAN_OPEN_PERM

An application wants to open a file or directory. The reader must write a response that determines whether the permission to open the filesystem object shall be granted.

FAN_OPEN_EXEC_PERM

An application wants to open a file for execution. The reader must write a response that determines whether the permission to open the filesystem object for execution shall be granted. See NOTES in fanotify_mark(2) for additional details.

To check for any close event, the following bit mask may be used:

FAN_CLOSE

A file was closed. This is a synonym for:

FAN_CLOSE_WRITE | FAN_CLOSE_NOWRITE

To check for any move event, the following bit mask may be used:

FAN_MOVE

A file or directory was moved. This is a synonym for:

FAN_MOVED_FROM | FAN_MOVED_TO

The following bits may appear in mask only in conjunction with other event type bits:

FAN_ONDIR

The events described in the mask have occurred on a directory object. Reporting events on directories requires setting this flag in the mark mask. See fanotify_mark(2) for additional details. The FAN_ONDIR flag is reported in an event mask only if the fanotify group identifies filesystem objects by file handles.

Information records that are supplied alongside the generic fanotify_event_metadata structure will always contain a nested structure of type fanotify_event_info_header. The fields of the fanotify_event_info_header are as follows:

info_type

A unique integer value representing the type of information record object received for an event. The value of this field can be set to one of the following: FAN_EVENT_INFO_TYPE_FID, FAN_EVENT_INFO_TYPE_DFID, FAN_EVENT_INFO_TYPE_DFID_NAME, or FAN_EVENT_INFO_TYPE_PIDFD. The value set for this field is dependent on the flags that have been supplied to fanotify_init(2). Refer to the field details of each information record object type below to understand the different cases in which the info_type values can be set.

pad

This field is currently not used by any information record object type and therefore is set to zero.

len

The value of len is set to the size of the information record object, including the fanotify_event_info_header. The total size of all additional information records is not expected to be larger than (event_len - metadata_len).

The fields of the fanotify_event_info_fid structure are as follows:

hdr

This is a structure of type fanotify_event_info_header. For example, when an fanotify file descriptor is created using FAN_REPORT_FID, a single information record is expected to be attached to the event with info_type field value of FAN_EVENT_INFO_TYPE_FID. When an fanotify file descriptor is created using the combination of FAN_REPORT_FID and FAN_REPORT_DIR_FID, there may be two information records attached to the event: one with info_type field value of FAN_EVENT_INFO_TYPE_DFID, identifying a parent directory object, and one with info_type field value of FAN_EVENT_INFO_TYPE_FID, identifying a child object. Note that for the directory entry modification events FAN_CREATE, FAN_DELETE, FAN_MOVE, and FAN_RENAME, an information record identifying the created/deleted/moved child object is reported only if an fanotify group was initialized with the flag FAN_REPORT_TARGET_FID.

fsid

This is a unique identifier of the filesystem containing the object associated with the event. It is a structure of type __kernel_fsid_t and contains the same value as f_fsid when calling statfs(2).

handle

This field contains a variable-length structure of type struct file_handle. It is an opaque handle that corresponds to a specified object on a filesystem as returned by name_to_handle_at(2). It can be used to uniquely identify a file on a filesystem and can be passed as an argument to open_by_handle_at(2). If the value of info_type field is FAN_EVENT_INFO_TYPE_DFID_NAME, the file handle is followed by a null terminated string that identifies the created/deleted/moved directory entry name. For other events such as FAN_OPEN, FAN_ATTRIB, FAN_DELETE_SELF, and FAN_MOVE_SELF, if the value of info_type field is FAN_EVENT_INFO_TYPE_FID, the handle identifies the object correlated to the event. If the value of info_type field is FAN_EVENT_INFO_TYPE_DFID, the handle identifies the directory object correlated to the event or the parent directory of a non-directory object correlated to the event. If the value of info_type field is FAN_EVENT_INFO_TYPE_DFID_NAME, the handle identifies the same directory object that would be reported with FAN_EVENT_INFO_TYPE_DFID and the file handle is followed by a null terminated string that identifies the name of a directory entry in that directory, or '.' to identify the directory object itself.

The fields of the fanotify_event_info_pidfd structure are as follows:

hdr

This is a structure of type fanotify_event_info_header. When an fanotify group is initialized using FAN_REPORT_PIDFD, the info_type field value of the fanotify_event_info_header is set to FAN_EVENT_INFO_TYPE_PIDFD.

pidfd

This is a process file descriptor that refers to the process responsible for generating the event. The returned process file descriptor is no different from one which could be obtained manually if pidfd_open(2) were to be called on fanotify_event_metadata.pid. In the instance that an error is encountered during pidfd creation, one of two possible error types represented by a negative integer value may be returned in this pidfd field. In cases where the process responsible for generating the event has terminated prior to the event listener being able to read events from the notification queue, FAN_NOPIDFD is returned. The pidfd creation for an event is only performed at the time the events are read from the notification queue. All other possible pidfd creation failures are represented by FAN_EPIDFD. Once the event listener has dealt with an event and the pidfd is no longer required, the pidfd should be closed via close(2).

The fields of the fanotify_event_info_error structure are as follows:

hdr

This is a structure of type fanotify_event_info_header. The info_type field is set to FAN_EVENT_INFO_TYPE_ERROR.

error

Identifies the type of error that occurred.

error_count

This is a counter of the number of errors suppressed since the last error was read.

The following macros are provided to iterate over a buffer containing fanotify event metadata returned by a read(2) from an fanotify file descriptor:

FAN_EVENT_OK(meta, len)

This macro checks the remaining length len of the buffer meta against the length of the metadata structure and the event_len field of the first metadata structure in the buffer.

FAN_EVENT_NEXT(meta, len)

This macro uses the length indicated in the event_len field of the metadata structure pointed to by meta to calculate the address of the next metadata structure that follows meta. len is the number of bytes of metadata that currently remain in the buffer. The macro returns a pointer to the next metadata structure that follows meta, and reduces len by the number of bytes in the metadata structure that has been skipped over (i.e., it subtracts meta->event_len from len).

In addition, there is:

FAN_EVENT_METADATA_LEN

This macro returns the size (in bytes) of the structure fanotify_event_metadata. This is the minimum size (and currently the only size) of any event metadata.

Monitoring an fanotify file descriptor for events

When an fanotify event occurs, the fanotify file descriptor indicates as readable when passed to epoll(7), poll(2), or select(2).

Dealing with permission events

For permission events, the application must write(2) a structure of the following form to the fanotify file descriptor:

struct fanotify_response {
    __s32 fd;
    __u32 response;
};

The fields of this structure are as follows:

fd

This is the file descriptor from the structure fanotify_event_metadata.

response

This field indicates whether or not the permission is to be granted. Its value must be either FAN_ALLOW to allow the file operation or FAN_DENY to deny the file operation.

If access is denied, the requesting application call will receive an EPERM error. Additionally, if the notification group has been created with the FAN_ENABLE_AUDIT flag, then the FAN_AUDIT flag can be set in the response field. In that case, the audit subsystem will log information about the access decision to the audit logs.

Monitoring filesystems for errors

A single FAN_FS_ERROR event is stored per filesystem at once. Extra error messages are suppressed and accounted for in the error_count field of the existing FAN_FS_ERROR event record, but details about the errors are lost.

Errors reported by FAN_FS_ERROR are generic errno values, but not all kinds of error types are reported by all filesystems.

Errors not directly related to a file (i.e. super block corruption) are reported with an invalid handle. For these errors, the handle will have the field handle_type set to FILEID_INVALID, and the handle buffer size set to 0.

Closing the fanotify file descriptor

When all file descriptors referring to the fanotify notification group are closed, the fanotify group is released and its resources are freed for reuse by the kernel. Upon close(2), outstanding permission events will be set to allowed.

/proc interfaces

The file /proc/pid/fdinfo/fd contains information about fanotify marks for file descriptor fd of process pid. See proc(5) for details.

Since Linux 5.13, the following interfaces can be used to control the amount of kernel resources consumed by fanotify:

/proc/sys/fs/fanotify/max_queued_events

The value in this file is used when an application calls fanotify_init(2) to set an upper limit on the number of events that can be queued to the corresponding fanotify group. Events in excess of this limit are dropped, but an FAN_Q_OVERFLOW event is always generated. Prior to Linux kernel 5.13, the hardcoded limit was 16384 events.

/proc/sys/fs/fanotify/max_user_group

This specifies an upper limit on the number of fanotify groups that can be created per real user ID. Prior to Linux kernel 5.13, the hardcoded limit was 128 groups per user.

/proc/sys/fs/fanotify/max_user_marks

This specifies an upper limit on the number of fanotify marks that can be created per real user ID. Prior to Linux kernel 5.13, the hardcoded limit was 8192 marks per group (not per user).

The two example programs below demonstrate the usage of the fanotify API.

Example program: fanotify_example.c

The first program is an example of fanotify being used with its event object information passed in the form of a file descriptor. The program marks the mount passed as a command-line argument and waits for events of type FAN_OPEN_PERM and FAN_CLOSE_WRITE. When a permission event occurs, a FAN_ALLOW response is given.

The following shell session shows an example of running this program. This session involved editing the file /home/user/temp/notes. Before the file was opened, a FAN_OPEN_PERM event occurred. After the file was closed, a FAN_CLOSE_WRITE event occurred. Execution of the program ends when the user presses the ENTER key.

# ./fanotify_example /home
Press enter key to terminate.
Listening for events.
FAN_OPEN_PERM: File /home/user/temp/notes
FAN_CLOSE_WRITE: File /home/user/temp/notes
Listening for events stopped.

Program source: fanotify_example.c

#define _GNU_SOURCE     /* Needed to get O_LARGEFILE definition */
#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <poll.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/fanotify.h>
#include <unistd.h>
/* Read all available fanotify events from the file descriptor 'fd'. */
static void
handle_events(int fd)
{
    const struct fanotify_event_metadata *metadata;
    struct fanotify_event_metadata buf[200];
    ssize_t len;
    char path[PATH_MAX];
    ssize_t path_len;
    char procfd_path[PATH_MAX];
    struct fanotify_response response;
    /* Loop while events can be read from fanotify file descriptor. */
    for (;;) {
        /* Read some events. */
        len = read(fd, buf, sizeof(buf));
        if (len == -1 && errno != EAGAIN) {
            perror("read");
            exit(EXIT_FAILURE);
        }
        /* Check if end of available data reached. */
        if (len <= 0)
            break;
        /* Point to the first event in the buffer. */
        metadata = buf;
        /* Loop over all events in the buffer. */
        while (FAN_EVENT_OK(metadata, len)) {
            /* Check that run-time and compile-time structures match. */
            if (metadata->vers != FANOTIFY_METADATA_VERSION) {
                fprintf(stderr,
                        "Mismatch of fanotify metadata version.\n");
                exit(EXIT_FAILURE);
            }
            /* metadata->fd contains either FAN_NOFD, indicating a
               queue overflow, or a file descriptor (a nonnegative
               integer). Here, we simply ignore queue overflow. */
            if (metadata->fd >= 0) {
                /* Handle open permission event. */
                if (metadata->mask & FAN_OPEN_PERM) {
                    printf("FAN_OPEN_PERM: ");
                    /* Allow file to be opened. */
                    response.fd = metadata->fd;
                    response.response = FAN_ALLOW;
                    write(fd, &response, sizeof(response));
                }
                /* Handle closing of writable file event. */
                if (metadata->mask & FAN_CLOSE_WRITE)
                    printf("FAN_CLOSE_WRITE: ");
                /* Retrieve and print pathname of the accessed file. */
                snprintf(procfd_path, sizeof(procfd_path),
                         "/proc/self/fd/%d", metadata->fd);
                path_len = readlink(procfd_path, path,
                                    sizeof(path) - 1);
                if (path_len == -1) {
                    perror("readlink");
                    exit(EXIT_FAILURE);
                }
                path[path_len] = '\0';
                printf("File %s\n", path);
                /* Close the file descriptor of the event. */
                close(metadata->fd);
            }
            /* Advance to next event. */
            metadata = FAN_EVENT_NEXT(metadata, len);
        }
    }
}
int
main(int argc, char *argv[])
{
    char buf;
    int fd, poll_num;
    nfds_t nfds;
    struct pollfd fds[2];
    /* Check mount point is supplied. */
    if (argc != 2) {
        fprintf(stderr, "Usage: %s MOUNT\n", argv[0]);
        exit(EXIT_FAILURE);
    }
    printf("Press enter key to terminate.\n");
    /* Create the file descriptor for accessing the fanotify API. */
    fd = fanotify_init(FAN_CLOEXEC | FAN_CLASS_CONTENT | FAN_NONBLOCK,
                       O_RDONLY | O_LARGEFILE);
    if (fd == -1) {
        perror("fanotify_init");
        exit(EXIT_FAILURE);
    }
    /* Mark the mount for:
       - permission events before opening files
       - notification events after closing a write-enabled
         file descriptor. */
    if (fanotify_mark(fd, FAN_MARK_ADD | FAN_MARK_MOUNT,
                      FAN_OPEN_PERM | FAN_CLOSE_WRITE, AT_FDCWD,
                      argv[1]) == -1) {
        perror("fanotify_mark");
        exit(EXIT_FAILURE);
    }
    /* Prepare for polling. */
    nfds = 2;
    fds[0].fd = STDIN_FILENO;       /* Console input */
    fds[0].events = POLLIN;
    fds[1].fd = fd;                 /* Fanotify input */
    fds[1].events = POLLIN;
    /* This is the loop to wait for incoming events. */
    printf("Listening for events.\n");
    while (1) {
        poll_num = poll(fds, nfds, -1);
        if (poll_num == -1) {
            if (errno == EINTR)     /* Interrupted by a signal */
                continue;           /* Restart poll() */
            perror("poll");         /* Unexpected error */
            exit(EXIT_FAILURE);
        }
        if (poll_num > 0) {
            if (fds[0].revents & POLLIN) {
                /* Console input is available: empty stdin and quit. */
                while (read(STDIN_FILENO, &buf, 1) > 0 && buf != '\n')
                    continue;
                break;
            }
            if (fds[1].revents & POLLIN) {
                /* Fanotify events are available. */
                handle_events(fd);
            }
        }
    }
    printf("Listening for events stopped.\n");
    exit(EXIT_SUCCESS);
}

Example program: fanotify_fid.c

The second program is an example of fanotify being used with a group that identifies objects by file handles. The program marks the filesystem object that is passed as a command-line argument and waits until an event of type FAN_CREATE has occurred. The event mask indicates which type of filesystem object—either a file or a directory—was created. Once all events have been read from the buffer and processed accordingly, the program simply terminates.

The following shell sessions show two different invocations of this program, with different actions performed on a watched object.

The first session shows a mark being placed on /home/user. This is followed by the creation of a regular file, /home/user/testfile.txt. This results in a FAN_CREATE event being generated and reported against the file's parent watched directory object and with the created file name. Program execution ends once all events captured within the buffer have been processed.

# ./fanotify_fid /home/user
Listening for events.
FAN_CREATE (file created):
        Directory /home/user has been modified.
        Entry 'testfile.txt' is not a subdirectory.
All events processed successfully. Program exiting.
$ touch /home/user/testfile.txt              # In another terminal

The second session shows a mark being placed on /home/user. This is followed by the creation of a directory, /home/user/testdir. This specific action results in a FAN_CREATE event being generated and is reported with the FAN_ONDIR flag set and with the created directory name.

# ./fanotify_fid /home/user
Listening for events.
FAN_CREATE | FAN_ONDIR (subdirectory created):
        Directory /home/user has been modified.
        Entry 'testdir' is a subdirectory.
All events processed successfully. Program exiting.
$ mkdir -p /home/user/testdir          # In another terminal

Program source: fanotify_fid.c

#define _GNU_SOURCE
#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/fanotify.h>
#include <unistd.h>
#define BUF_SIZE 256
int
main(int argc, char *argv[])
{
    int fd, ret, event_fd, mount_fd;
    ssize_t len, path_len;
    char path[PATH_MAX];
    char procfd_path[PATH_MAX];
    char events_buf[BUF_SIZE];
    struct file_handle *file_handle;
    struct fanotify_event_metadata *metadata;
    struct fanotify_event_info_fid *fid;
    const char *file_name;
    struct stat sb;
    if (argc != 2) {
        fprintf(stderr, "Invalid number of command line arguments.\n");
        exit(EXIT_FAILURE);
    }
    mount_fd = open(argv[1], O_DIRECTORY | O_RDONLY);
    if (mount_fd == -1) {
        perror(argv[1]);
        exit(EXIT_FAILURE);
    }
    /* Create an fanotify file descriptor with FAN_REPORT_DFID_NAME as
       a flag so that program can receive fid events with directory
       entry name. */
    fd = fanotify_init(FAN_CLASS_NOTIF | FAN_REPORT_DFID_NAME, 0);
    if (fd == -1) {
        perror("fanotify_init");
        exit(EXIT_FAILURE);
    }
    /* Place a mark on the filesystem object supplied in argv[1]. */
    ret = fanotify_mark(fd, FAN_MARK_ADD | FAN_MARK_ONLYDIR,
                        FAN_CREATE | FAN_ONDIR,
                        AT_FDCWD, argv[1]);
    if (ret == -1) {
        perror("fanotify_mark");
        exit(EXIT_FAILURE);
    }
    printf("Listening for events.\n");
    /* Read events from the event queue into a buffer. */
    len = read(fd, events_buf, sizeof(events_buf));
    if (len == -1 && errno != EAGAIN) {
        perror("read");
        exit(EXIT_FAILURE);
    }
    /* Process all events within the buffer. */
    for (metadata = (struct fanotify_event_metadata *) events_buf;
            FAN_EVENT_OK(metadata, len);
            metadata = FAN_EVENT_NEXT(metadata, len)) {
        fid = (struct fanotify_event_info_fid *) (metadata + 1);
        file_handle = (struct file_handle *) fid->handle;
        /* Ensure that the event info is of the correct type. */
        if (fid->hdr.info_type == FAN_EVENT_INFO_TYPE_FID ||
            fid->hdr.info_type == FAN_EVENT_INFO_TYPE_DFID) {
            file_name = NULL;
        } else if (fid->hdr.info_type == FAN_EVENT_INFO_TYPE_DFID_NAME) {
            file_name = file_handle->f_handle +
                        file_handle->handle_bytes;
        } else {
            fprintf(stderr, "Received unexpected event info type.\n");
            exit(EXIT_FAILURE);
        }
        if (metadata->mask == FAN_CREATE)
            printf("FAN_CREATE (file created):\n");
        if (metadata->mask == (FAN_CREATE | FAN_ONDIR))
            printf("FAN_CREATE | FAN_ONDIR (subdirectory created):\n");
	/* metadata->fd is set to FAN_NOFD when the group identifies
	   objects by file handles.  To obtain a file descriptor for
	   the file object corresponding to an event you can use the
	   struct file_handle that's provided within the
	   fanotify_event_info_fid in conjunction with the
	   open_by_handle_at(2) system call.  A check for ESTALE is
	   done to accommodate for the situation where the file handle
	   for the object was deleted prior to this system call. */
        event_fd = open_by_handle_at(mount_fd, file_handle, O_RDONLY);
        if (event_fd == -1) {
            if (errno == ESTALE) {
                printf("File handle is no longer valid. "
                        "File has been deleted\n");
                continue;
            } else {
                perror("open_by_handle_at");
                exit(EXIT_FAILURE);
            }
        }
        snprintf(procfd_path, sizeof(procfd_path), "/proc/self/fd/%d",
                event_fd);
        /* Retrieve and print the path of the modified dentry. */
        path_len = readlink(procfd_path, path, sizeof(path) - 1);
        if (path_len == -1) {
            perror("readlink");
            exit(EXIT_FAILURE);
        }
        path[path_len] = '\0';
        printf("\tDirectory '%s' has been modified.\n", path);
        if (file_name) {
            ret = fstatat(event_fd, file_name, &sb, 0);
            if (ret == -1) {
                if (errno != ENOENT) {
                    perror("fstatat");
                    exit(EXIT_FAILURE);
                }
                printf("\tEntry '%s' does not exist.\n", file_name);
            } else if ((sb.st_mode & S_IFMT) == S_IFDIR) {
                printf("\tEntry '%s' is a subdirectory.\n", file_name);
            } else {
                printf("\tEntry '%s' is not a subdirectory.\n",
                        file_name);
            }
        }
        /* Close associated file descriptor for this event. */
        close(event_fd);
    }
    printf("All events processed successfully. Program exiting.\n");
    exit(EXIT_SUCCESS);
}

fanotify_init(2), fanotify_mark(2), inotify(7)