Project

General

Profile

Actions

Bug #13879

closed

flock(3C) on NFS files should fail with EOPNOTSUPP

Added by Joshua M. Clulow almost 2 years ago. Updated 7 months ago.

Status:
Closed
Priority:
Normal
Assignee:
Category:
filesystems (not ZFS)
Start date:
Due date:
% Done:

100%

Estimated time:
Difficulty:
Medium
Tags:
Gerrit CR:
External Bug:

Description

Our flock(3C) support does not presently support locking files on remote file systems like NFS (see F_REMOTELOCK, and comments in fs_frlock()). In the event that one tries to lock an NFS-backed file, flock() presently fails with EINVAL. This is probably not quite right; e.g., in the manual page:

       EINVAL
                           The operation argument does not contain one of
                           LOCK_SH, LOCK_EX, or LOCK_UN; or the operation
                           argument contains LOCK_UN and LOCK_NB; or the
                           operation argument contains any bits other than
                           those set by LOCK_SH, LOCK_EX, LOCK_NB, and
                           LOCK_UN.
...
       EOPNOTSUPP
                           The locking of files of the type indicated by
                           the fildes argument is not supported.

The latter seems more appropriate here.

As an aside, a brief survey of the documentation of other platforms shows:

  • FreeBSD and OpenBSD appear to use EOPNOTSUPP for this condition
  • Mac OS X appears to use ENOTSUP
  • Linux does not appear to document the possibility of a failure of this kind, but explicitly says EINVAL is for an invalid operation argument

It does not seem that flock() has been standardised.

Actions #1

Updated by Electric Monk over 1 year ago

  • Gerrit CR set to 1845
Actions #2

Updated by Joshua M. Clulow 12 months ago

  • Assignee set to Luqman Aden
Actions #3

Updated by Luqman Aden 7 months ago

#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/types.h>

int main() {
    int fd;
    int res;
    struct flock lockd = {};

    fd = open("/tmp/nfs-flock-test/test.lock", O_CREAT | O_WRONLY);
    if (fd < 0) {
        perror("Failed to open lock file");
        exit(1);
    }

    res = flock(fd, LOCK_EX);
    if (res != 0) {
        perror("[flock] Failed to grab lock");
    } else {
        flock(fd, LOCK_UN);
    }

    lockd.l_type = F_WRLCK;
    res = fcntl(fd, F_OFD_GETLK, &lockd);
    if (res != 0) {
        perror("[fcntl F_OFD_GETLK] Failed to grab lock");
    } else {
        lockd.l_type = F_UNLCK;
        fcntl(fd, F_OFD_SETLK, &lockd);
    }

    return 0;
}

Using the above program where /tmp/nfs-flock-test is an NFS mountpoint:

$ ./flock
[flock] Failed to grab lock: Invalid argument
[fcntl F_OFD_GETLK] Failed to grab lock: Invalid argument

After changes:

$ ./flock
[flock] Failed to grab lock: Operation not supported on transport endpoint
[fcntl F_OFD_GETLK] Failed to grab lock: Operation not supported on transport endpoint
Actions #4

Updated by Electric Monk 7 months ago

  • Status changed from New to Closed
  • % Done changed from 0 to 100

git commit b8af4a8966ef2150997d7664836f5c360b849005

commit  b8af4a8966ef2150997d7664836f5c360b849005
Author: Luqman Aden <luqman@oxide.computer>
Date:   2022-11-21T17:06:50.000Z

    13879 flock(3C) on NFS files should fail with EOPNOTSUPP
    Reviewed by: Andy Stormont <andyjstormont@gmail.com>
    Approved by: Dan McDonald <danmcd@mnx.io>

Actions

Also available in: Atom PDF