From: Aneesh Kumar K.V on

The below set of patches implement open by handle support using exportfs
operations. This allows user space application to map a file name to file
handle and later open the file using handle. This should be usable
for userspace NFS [1] and 9P server [2]. XFS already support this with the ioctls


git repo for the patchset at:

git:// open-by-handle

Changes from V15:
a) Add mount id to file handle
b) Add support for optional uuid tag in /proc/<pid>/mountinfo
c) Added MAY_OPEN_LINK to support open on symlink instead of
adding a new may_open function. Also limited the open flag

Changes from V14:
a) Use fget_light instead of fget in the patches
b) Drop uuid from struct file_handle as per the last review

Changes from V13:
a) Add support for file descriptor to handle conversion. This is needed
so that we find the right file handle for newly created files.

Changes from V12:
a) Use CAP_DAC_READ_SEARCH instead of CAP_DAC_OVERRIDE in open_by_handle
b) Return -ENOTDIR if O_DIRECTORY flag is specified in open_by_handle with
handle for non directory

Changes from V11:
a) Add necessary documentation to different functions
b) Add null pathname support to faccessat and linkat similar to
c) compile fix on x86_64

Changes from V10:
a) Missed an stg refresh before sending out the patchset. Send
updated patchset.

Changes from V9:
a) Fix compile errors with CONFIG_EXPORTFS not defined
b) Return -EOPNOTSUPP if file system doesn't support fh_to_dentry exportfs callback.

Changes from V8:
a) exportfs_decode_fh now returns -ESTALE if export operations is not defined.
b) drop get_fsid super_operations. Instead use superblock to store uuid.

Changes from V7:
a) open_by_handle now use mountdirfd to identify the vfsmount.
b) We don't validate the UUID passed as a part of file handle in open_by_handle.
UUID is provided as a part of file handle as an easy way for userspace to
use the kernel returned handle as it is. It also helps in finding the 16 byte
filessytem UUID in userspace without using file system specific libraries to
read file system superblock. If a particular file system doesn't support UUID
or any form of unique id this field in the file handle will be zero filled.
c) drop freadlink syscall. Instead use readlinkat with NULL pathname to indicate
read the link target name of the link pointed by fd. This is similar to
d) Instead of opencoding all the open flag related check use helper functions.
Did finish_open_by_handle similar to finish_open.
c) Fix may_open to not return ELOOP for symlink when we are called from handle open.
open(2) still returns error as expected.

Changes from V6:
a) Add uuid to vfsmount lookup and drop uuid to superblock lookup
b) Return -EOPNOTSUPP in sys_name_to_handle if the file system returned uuid
doesn't give the same vfsmount on lookup. This ensure that we fail
sys_name_to_handle when we have multiple file system returning same UUID.

Changes from V5:
a) added sys_name_to_handle_at syscall which takes AT_SYMLINK_NOFOLLOW flag
instead of two syscalls sys_name_to_handle and sys_lname_to_handle.
b) addressed review comments from Niel Brown
c) rebased to b91ce4d14a21fc04d165be30319541e0f9204f15
d) Add compat_sys_open_by_handle

Chages from V4:
a) Changed the syscal arguments so that we don't need compat syscalls
as suggested by Christoph
c) Added two new syscall sys_lname_to_handle and sys_freadlink to work with
d) Changed open_by_handle to work with all file types
e) Add ext3 support

Changes from V3:
a) Code cleanup suggested by Andreas
b) x86_64 syscall support
c) add compat syscall

Chages from V2:
a) Support system wide unique handle.

Changes from v1:
a) handle size is now specified in bytes
b) returns -EOVERFLOW if the handle size is small
c) dropped open_handle syscall and added open_by_handle_at syscall
open_by_handle_at takes mount_fd as the directory fd of the mount point
containing the file
e) handle will only be unique in a given file system. So for an NFS server
exporting multiple file system, NFS server will have to internally track the
mount point to which a file handle belongs to. We should be able to do it much
easily than expecting kernel to give a system wide unique file handle. System
wide unique file handle would need much larger changes to the exportfs or VFS
interface and I was not sure whether we really need to do that in the kernel or
in the user space
f) open_handle_at now only check for DAC_OVERRIDE capability

Example program: (x86_32). (x86_64 would need a different syscall number)
cc <source.c>
#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>

#include <fcntl.h>
#include <unistd.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <string.h>

struct file_handle {
int mnt_id;
int handle_size;
int handle_type;
unsigned char handle[0];

#define AT_FDCWD -100
#define AT_SYMLINK_FOLLOW 0x400

static int name_to_handle(const char *name, struct file_handle *fh)
return syscall(338, AT_FDCWD, name, fh, AT_SYMLINK_FOLLOW);

static int lname_to_handle(const char *name, struct file_handle *fh)
return syscall(338, AT_FDCWD, name, fh, 0);

static int fd_to_handle(int fd, struct file_handle *fh)
return syscall(338, fd, NULL, fh, AT_SYMLINK_FOLLOW);

static int open_by_handle(int mountfd, struct file_handle *fh, int flags)
return syscall(339, mountfd, fh, flags);

#define BUFSZ 100
int main(int argc, char *argv[])
int fd;
int ret, done = 0;
int mountfd;
int handle_sz;
struct stat bufstat;
char buf[BUFSZ];
struct file_handle *fh = NULL;;
if (argc != 3 ) {
printf("Usage: %s <filename> <mount-dir-name>\n", argv[0]);
if (fh && fh->handle_size) {
handle_sz = fh->handle_size;
fh = malloc(sizeof(struct file_handle) + handle_sz);
fh->handle_size = handle_sz;
} else {
fh = malloc(sizeof(struct file_handle));
fh->handle_size = 0;
errno = 0;
ret = lname_to_handle(argv[1], fh);
if (ret && errno == EOVERFLOW) {
printf("Found the handle size needed to be %d\n", fh->handle_size);
goto again;
} else if (ret) {
printf("found mount_id %d\n", fh->mnt_id);
printf("Waiting for input");
mountfd = open(argv[2], O_RDONLY | O_DIRECTORY);
if (mountfd <= 0) {
fd = open_by_handle(mountfd, fh, O_RDONLY);
if (fd <= 0 ) {
printf("Reading the content now \n");
fstat(fd, &bufstat);
ret = S_ISLNK(bufstat.st_mode);
if (ret) {
memset(buf, 0 , BUFSZ);
readlinkat(fd, NULL, buf, BUFSZ);
printf("%s is a symlink pointing to %s\n", argv[1], buf);
memset(buf, 0 , BUFSZ);
while (1) {
ret = read(fd, buf, BUFSZ -1);
if (ret <= 0)
buf[ret] = '\0';
printf("%s", buf);
memset(buf, 0 , BUFSZ);
/* Now check for faccess */
if (faccessat(fd, NULL, W_OK, 0) == 0) {
printf("Got write permission on the file \n");
} else
perror("faccess error");
/* now try to create a hardlink */
if (linkat(fd, NULL, AT_FDCWD, "test", 0) == 0){
printf("created hardlink\n");
} else
perror("linkat error");
if (done)
printf("Map fd to handle \n");
ret = fd_to_handle(fd, fh);
if (ret) {
done = 1;
goto do_again;


To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo(a)
More majordomo info at

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)
More majordomo info at
Please read the FAQ at