Re: [PATCH v2 1/2] exec: add a flag for "reasonable" execveat() comm

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Sep 27, 2024 at 10:45:58AM -0500, Eric W. Biederman wrote:
> Tycho Andersen <tycho@tycho.pizza> writes:
> 
> > From: Tycho Andersen <tandersen@xxxxxxxxxxx>
> >
> > Zbigniew mentioned at Linux Plumber's that systemd is interested in
> > switching to execveat() for service execution, but can't, because the
> > contents of /proc/pid/comm are the file descriptor which was used,
> > instead of the path to the binary. This makes the output of tools like
> > top and ps useless, especially in a world where most fds are opened
> > CLOEXEC so the number is truly meaningless.
> >
> > Change exec path to fix up /proc/pid/comm in the case where we have
> > allocated one of these synthetic paths in bprm_init(). This way the actual
> > exec machinery is unchanged, but cosmetically the comm looks reasonable to
> > admins investigating things.
> 
> Perhaps change the subject to match the code.
> 
> > Signed-off-by: Tycho Andersen <tandersen@xxxxxxxxxxx>
> > Suggested-by: Zbigniew Jędrzejewski-Szmek <zbyszek@xxxxxxxxx>
> > CC: Aleksa Sarai <cyphar@xxxxxxxxxx>
> > Link: https://github.com/uapi-group/kernel-features#set-comm-field-before-exec
> > ---
> > v2: * drop the flag, everyone :)
> >     * change the rendered value to f_path.dentry->d_name.name instead of
> >       argv[0], Eric
> > ---
> >  fs/exec.c | 13 ++++++++++++-
> >  1 file changed, 12 insertions(+), 1 deletion(-)
> >
> > diff --git a/fs/exec.c b/fs/exec.c
> > index dad402d55681..9520359a8dcc 100644
> > --- a/fs/exec.c
> > +++ b/fs/exec.c
> > @@ -1416,7 +1416,18 @@ int begin_new_exec(struct linux_binprm * bprm)
> >  		set_dumpable(current->mm, SUID_DUMP_USER);
> >  
> >  	perf_event_exec();
> > -	__set_task_comm(me, kbasename(bprm->filename), true);
> > +
> > +	/*
> > +	 * If fdpath was set, execveat() made up a path that will
> > +	 * probably not be useful to admins running ps or similar.
> > +	 * Let's fix it up to be something reasonable.
> > +	 */
> > +	if (bprm->fdpath) {
> > +		BUILD_BUG_ON(TASK_COMM_LEN > DNAME_INLINE_LEN);
> > +		__set_task_comm(me, bprm->file->f_path.dentry->d_name.name, true);
> 
> We can just do this regardless of bprm->fdpath.
> 
> It will be a change of behavior on when executing symlinks and possibly
> mount points but I don't think we care.  If we do then we can add make
> it conditional with "if (bprm->fdpath)"
> 
> At the very least using the above version unconditionally ought to flush
> out any bugs.

I'm not super comfortable doing this regardless of bprm->fdpath; that
seems like too many cases getting changed. Can we just leave it as
depending on bprm->fdpath?

Also, is d_name.name always going to be set? e.g. what about memfd, etc?

-- 
Kees Cook




[Index of Archives]     [Linux Wireless]     [Linux Kernel]     [ATH6KL]     [Linux Bluetooth]     [Linux Netdev]     [Kernel Newbies]     [Share Photos]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Samba]     [Device Mapper]

  Powered by Linux