Re: [PATCH] seccomp.sh: Quote tr character ranges to prevent glob expansion

[Stefano Brivio <sbrivio@redhat.com>]

On Wed, 5 Nov 2025 12:22:38 +1100
David Gibson <david@gibson.dropbear.id.au> wrote:

> On Tue, Nov 04, 2025 at 06:01:49AM +0100, Stefano Brivio wrote:
> > On Mon,  3 Nov 2025 13:08:34 +0100
> > Laurent Vivier <lvivier@redhat.com> wrote:
> >   
> > > we use [a-z] and [A-Z] patterns with 'tr', but
> > > if there are files with names matching these patterns they will be
> > > replaced by the name of the file and seccomp.h will not be generated
> > > correctly:
> > > $ rm seccomp.h
> > > $ touch a b
> > > $ make
> > > tr: extra operand '[A-Z]'
> > > Try 'tr --help' for more information.
> > > seccomp profile passt allows:  accept accept4 bind clock_gettime close connect epoll_ctl epoll_pwait epoll_wait exit_group
> > >    fallocate fcntl fsync ftruncate getsockname getsockopt listen lseek read recvfrom recvmmsg recvmsg sendmmsg sendmsg sendto
> > > ...
> > > cc -Wall -Wextra -Wno-format-zero-length -Wformat-security -pedantic -std=c11 -D_XOPEN_SOURCE=700 -D_GNU_SOURCE -D_FORTIFY_SOURCE=2 -O2 -pie -fPIE -DPAGE_SIZE=4096 -DVERSION="2025_09_19.623dbf6-54-gf6b6118fcabd" -DDUAL_STACK_SOCKETS=1 -DHAS_GETRANDOM -fstack-protector-strong   arch.c arp.c checksum.c conf.c dhcp.c dhcpv6.c epoll_ctl.c flow.c fwd.c icmp.c igmp.c inany.c iov.c ip.c isolation.c lineread.c log.c mld.c ndp.c netlink.c migrate.c packet.c passt.c pasta.c pcap.c pif.c repair.c tap.c tcp.c tcp_buf.c tcp_splice.c tcp_vu.c udp.c udp_flow.c udp_vu.c util.c vhost_user.c virtio.c vu_common.c -o passt
> > > In file included from isolation.c:83:
> > > seccomp.h:11:45: error: 'AUDIT_ARCH_' undeclared here (not in a function); did you mean 'AUDIT_ARCH'?
> > >    11 |         BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, AUDIT_ARCH_, 0, 80),
> > >       |                                             ^~~~~~~~~~~
> > > 
> > > Signed-off-by: Laurent Vivier <lvivier@redhat.com>
> > > ---
> > >  seccomp.sh | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/seccomp.sh b/seccomp.sh
> > > index a7bc417b9f6b..ba92b29d9a29 100755
> > > --- a/seccomp.sh
> > > +++ b/seccomp.sh
> > > @@ -22,7 +22,7 @@ IN="$@"
> > >  [ -z "${ARCH}" ] && ARCH="$(uname -m)"
> > >  [ -z "${CC}" ] && CC="cc"
> > >  
> > > -AUDIT_ARCH="AUDIT_ARCH_$(echo ${ARCH} | tr [a-z] [A-Z]             \
> > > +AUDIT_ARCH="AUDIT_ARCH_$(echo ${ARCH} | tr '[a-z]' '[A-Z]'         \  
> > 
> > Oops.
> > 
> > I wonder if this is a complete fix though, because in general I didn't
> > care about possible expansions and I just assumed I set -f on the whole
> > script, which I didn't for some reason. That is, it should be:
> > 
> > #!/bin/sh -euf
> > 
> > and if you run 'shellcheck seccomp.sh', you'll find many other places
> > where I didn't care, so perhaps we really need that -f, but I didn't
> > look into all those shellcheck reports.
> > 
> > And by the way of shellcheck and compatibility, this is still on my
> > to-do list:
> > 
> >   https://github.com/chimera-linux/cports/pull/1483#issuecomment-2079007408
> > 
> > All in all, I can apply this, it fixes a bit and surely doesn't hurt.
> > 
> > Or we can (also?) add -f, but we need to make sure we don't rely on
> > expansions. We should perhaps check / fix reasonable shellcheck reports
> > and compatibility issues too.  
> 
> I don't love that idea.  I hadn't even realised -f existed until right
> now, so having an obscure global flag change behaviour everywhere
> doesn't ideal for readability.

I don't think it's *that* obscure actually, I use it quite commonly
(unless the script is playing with files), I have a few occurrences of
it in my current /usr/lib, and it even predates POSIX and SUS.

From page 108 of AT&T's System V Interface Definition, Issue 2 Volume
II, Chapter 4 (Commands and Utilities), SH(BU_CMD):

  -f (New in System V Release 2.) Disable file name generation

  https://bitsavers.org/pdf/att/unix/SVID/System_V_Interface_Definition_Issue_2_Volume_2_1986.pdf

> Plus, disabling globs removes the need
> for _some_ escaping, but not all, so it just means there's now two
> different sets of rules you'd need to apply about what must be
> escaped.

Well, I think we should escape everything anyway, and make sure we
do by making it shellcheck(1)-clean, eventually.

But '[a-z]' in 'tr [a-z]' expanding to 'a' is the very madness that -f
is supposed to protect us from. That is, I see it as something needed
for defensive/robust programming rather than something hiding issues.

-- 
Stefano