linux-kernel - Re: [PATCH] tools/nolibc: completely remove optional environ support

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <ZK0AudpEIDHEdm74@1wt.eu>
Date:   Tue, 11 Jul 2023 09:11:53 +0200
From:   Willy Tarreau <w@....eu>
To:     Thomas Weißschuh <thomas@...ch.de>
Cc:     linux-kernel@...r.kernel.org, Zhangjin Wu <falcon@...ylab.org>
Subject: Re: [PATCH] tools/nolibc: completely remove optional environ support

On Mon, Jul 10, 2023 at 07:51:53PM +0200, Thomas Weißschuh wrote:
> On 2023-07-10 19:43:27+0200, Willy Tarreau wrote:
> > Hi Thomas,
> > 
> > On Mon, Jul 10, 2023 at 07:22:53PM +0200, Thomas Weißschuh wrote:
> > > In commit 52e423f5b93e ("tools/nolibc: export environ as a weak symbol on i386")
> > > and friends the asm startup logic was extended to directly populate the
> > > "environ" array.
> > > 
> > > This makes it impossible for "environ" to be dropped by the linker.
> > > Therefore also drop the other logic to handle non-present "environ".
> > 
> > Hmmm OK but at least I'd like that we continue to reference it from
> > nolibc-test to make sure it's still visible. Maybe we could just check
> > that it's always equal to envp ? If we drop its reference from there,
> > sooner or later someone will find it interesting to rename it and some
> > programs referencing it will break.
> 
> Easy enough to test for. I'll send a v2.

Thanks!

> > > Note:
> > > 
> > > Given that nowadays both _auxv and environ are mandatory symbols imposed
> > > by nolibc of pointer size does it make sense to keep the code to make
> > > int-sized errno optional?
> > 
> > While it indeed used to be related to having a data segment or not
> > initially, it still has an impact on our ability to completely drop
> > the errno setting code from all syscalls. Given the SET_ERRNO() macro
> > now I guess it's very cheap to keep it, don't you think ?
> 
> SET_ERRNO irks me a tiny bit :-)

To be honest, it's the same for me. but it's cheap. And when you rebuild
a binary without it you can observe significant savings that can be
important for those who are space-constrained.

> But it's easy enough to keep, let's do so.
> Just wanted to have brought it up.

Yes, you're totally right to raise this, thank you!

Willy