GNU bug report logs - #43415
coreutils 8.32: install: fchmod fails with EBADF

Previous Next

To reply to this bug, email your comments to 43415 AT debbugs.gnu.org.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Cameron Nemo <cnemo <at> tutanota.com>
To: bug-coreutils <at> gnu.org
Subject: coreutils 8.32: install: fchmod fails with EBADF
Date: Tue, 15 Sep 2020 03:31:43 +0200 (CEST)

[Message part 1 (text/plain, inline)]

Hello,

I am experiencing some curious behavior when trying to set the file mode using
the install command.

Version info: install (GNU coreutils) 8.32

Summary of behavior: fchmod to 1777 fails with EBADF

The log from strace shows the following:

    ...
    geteuid()                               = 0
    umask(000)                              = 022
    mkdirat(AT_FDCWD, "tmp", 01755)         = -1 EEXIST (File exists)
    openat(AT_FDCWD, "tmp", O_RDONLY|O_NOCTTY|O_NONBLOCK|O_LARGEFILE|O_PATH|O_DIRECTORY) = 3
    fstat(3, {st_mode=S_IFDIR|S_ISVTX|0755, st_size=40, ...}) = 0
    fchmod(3, 01777)                        = -1 EBADF (Bad file descriptor)
    fcntl(3, F_GETFD)                       = 0
    fchmodat(AT_FDCWD, "/proc/self/fd/3", 01777) = 0
    close(3)                                = 0
    ...

For some odd reason, the call to fchmod is failing with EBADF.
This is causing the chmod() function to fall back to using an fd link in /proc.

Normally (i.e. when /proc is mounted) this poses no issues.
But I am trying to use the install command in a bootstrap/chroot environment,
and having to incorporate mount and unmount logic adds a significant amount of
non-idempodent complexity.

It seems like relying on the /proc link is not ideal,
and a bug is being hidden by such behavior.
Is there any chance that this can be resolved?

Regards,
--
Cameron Nemo

[Message part 2 (text/html, inline)]

Message #8 received at 43415 <at> debbugs.gnu.org (full text, mbox):

From: Paul Eggert <eggert <at> cs.ucla.edu>
To: Cameron Nemo <cnemo <at> tutanota.com>
Cc: 43415 <at> debbugs.gnu.org
Subject: Re: bug#43415: coreutils 8.32: install: fchmod fails with EBADF
Date: Tue, 15 Sep 2020 16:49:04 -0700

On 9/14/20 6:31 PM, Cameron Nemo via GNU coreutils Bug Reports wrote:
> It seems like relying on the /proc link is not ideal,
> and a bug is being hidden by such behavior.
> Is there any chance that this can be resolved?

It really should be fixed in the Linux kernel: it needs a proper way to 
implement POSIX fchmodat <https://pubs.opengroup.org/onlinepubs/9699919799/> 
with the AT_SYMLINK_NOFOLLOW flag, in order to plug some security holes 
involving symlink attacks. See:

https://bugzilla.redhat.com/show_bug.cgi?id=1810141
https://lkml.org/lkml/2020/6/9/548

In the meantime, mounting /proc may be your best bet. I vaguely recall there are 
other places in glibc that assume /proc.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.