regression: file does not understand the -S option
Recently a commit was added to call 'file -zSb' instead of 'file -zb'. This causes a regression on Leap 15 (but not on Tumbleweed), because file 5.32 does not understand the -S option. How can this be fixed properly, to handle both cases either at runtime or at buildtime? Thanks, Olaf
On 9/20/23 23:42, Olaf Hering wrote:
Recently a commit was added to call 'file -zSb' instead of 'file -zb'.
This causes a regression on Leap 15 (but not on Tumbleweed), because file 5.32 does not understand the -S option.
How can this be fixed properly, to handle both cases either at runtime or at buildtime?
This is likely from commit 23986d3c4f4d ("file: Use -S option with -z", 2022-11-28). Does your error output contain file: invalid option -- 'S' ? If it does, then I think we could modify "daemon/file.ml" and "daemon/filearch.ml". Try "file" with the current options, and if there's a failure, and stderr contains the above string, retry without -S. Unfortunately, this is a bit messy. We'd probably want to cache the availability of -S. Also, because this logic is used from multiple places, we'd first have to factor out the current "file" invocation -- minimally, try to rebase the "file" invocation in "daemon/filearch.ml" to the interface exposed by "daemon/file.mli". Laszlo
Thanks, Olaf
_______________________________________________ Libguestfs mailing list Libguestfs@redhat.com https://listman.redhat.com/mailman/listinfo/libguestfs
Thu, 21 Sep 2023 10:21:41 +0200 Laszlo Ersek <lersek@redhat.com>:
Does your error output contain file: invalid option -- 'S' ?
Yes, it looks like this: # /usr/bin/env -i /usr/bin/file -S /usr/bin/file: invalid option -- 'S' Usage: file [-bcEhikLlNnprsvzZ0] [--apple] [--extension] [--mime-encoding] [--mime-type] [-e testname] [-F separator] [-f namefile] [-m magicfiles] file ... file -C [-m magicfiles] file [--help] Olaf
On Wed, Sep 20, 2023 at 11:42:55PM +0200, Olaf Hering wrote:
Recently a commit was added to call 'file -zSb' instead of 'file -zb'.
This causes a regression on Leap 15 (but not on Tumbleweed), because file 5.32 does not understand the -S option.
How can this be fixed properly, to handle both cases either at runtime or at buildtime?
The background to this was: https://github.com/libguestfs/libguestfs/issues/100 It took a while to work out what was going on in the original bug report, but it turned out that Arch (IIRC) enabled the seccomp feature in the 'file' command. This filters what system calls 'file' is allowed to make, which strengthens security as 'file' is often run on untrusted inputs. Unfortunately the seccomp rules for 'file' don't cope with running external programs (ie. 'file -z' which runs zcat). We filed a bug to try to get that fixed: https://bugzilla.redhat.com/show_bug.cgi?id=2148753 https://bugs.astron.com/view.php?id=406 but the fix to seccomp policy was rejected recently in both Fedora & upstream. The patch we added to libguestfs turns off seccomp sandboxing, both because it's broken (see above) and because we don't really need it as we run stuff in a virtual machine already: https://github.com/libguestfs/libguestfs/commit/23986d3c4f4d1f9cbac44cc743d3... I didn't realise there were distros that lack support for the 'file -S' option. So I guess the fix is to detect if 'file' has the -S option ... I think we can just grep 'file --help' for the -S / --no-sandbox option. Let me try for a patch now. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com Fedora Windows cross-compiler. Compile Windows programs, test, and build Windows installers. Over 100 libraries supported. http://fedoraproject.org/wiki/MinGW
On Thu, Sep 21, 2023 at 12:25:21PM +0100, Richard W.M. Jones wrote:
On Wed, Sep 20, 2023 at 11:42:55PM +0200, Olaf Hering wrote:
Recently a commit was added to call 'file -zSb' instead of 'file -zb'.
This causes a regression on Leap 15 (but not on Tumbleweed), because file 5.32 does not understand the -S option.
How can this be fixed properly, to handle both cases either at runtime or at buildtime?
The background to this was:
https://github.com/libguestfs/libguestfs/issues/100
It took a while to work out what was going on in the original bug report, but it turned out that Arch (IIRC) enabled the seccomp feature in the 'file' command. This filters what system calls 'file' is allowed to make, which strengthens security as 'file' is often run on untrusted inputs.
Unfortunately the seccomp rules for 'file' don't cope with running external programs (ie. 'file -z' which runs zcat). We filed a bug to try to get that fixed:
https://bugzilla.redhat.com/show_bug.cgi?id=2148753 https://bugs.astron.com/view.php?id=406
but the fix to seccomp policy was rejected recently in both Fedora & upstream.
Their rationale in that bug makes no sense. Not allowing 'clone+execve' etc is correct when '-z' is NOT specified by the user. No argument there. If '-z' is specified then adding clone+execve etc is the only way it can work. They should apply a different seccomp filter for '-z' only which includes clone+execve, etc. Telling people to turn off seccomp entirely in order to use '-z' is even worse for security than just allowing clone+execve. With regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
participants (4)
-
Daniel P. Berrangé -
Laszlo Ersek -
Olaf Hering -
Richard W.M. Jones