linux-kernel - Re: [RFC bpf-next v2 0/1] libbpf: add compile-time OOB warning to bpf_tail_call

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: <CAK7-dKZc8JOyjwiAG0rj=F+gfe_M1rsH0AqaB6Nq5iGz==vmAg@mail.gmail.com>
Date: Mon, 8 Sep 2025 13:56:07 +0900
From: Hoyeon Lee <hoyeon.lee@...e.com>
To: Andrii Nakryiko <andrii.nakryiko@...il.com>
Cc: Shung-Hsi Yu <shung-hsi.yu@...e.com>, bpf@...r.kernel.org, netdev@...r.kernel.org, 
	Andrii Nakryiko <andrii@...nel.org>, Eduard Zingerman <eddyz87@...il.com>, 
	Alexei Starovoitov <ast@...nel.org>, Daniel Borkmann <daniel@...earbox.net>, 
	Martin KaFai Lau <martin.lau@...ux.dev>, Song Liu <song@...nel.org>, 
	Yonghong Song <yonghong.song@...ux.dev>, John Fastabend <john.fastabend@...il.com>, 
	KP Singh <kpsingh@...nel.org>, Stanislav Fomichev <sdf@...ichev.me>, Hao Luo <haoluo@...gle.com>, 
	Jiri Olsa <jolsa@...nel.org>, Nathan Chancellor <nathan@...nel.org>, 
	Nick Desaulniers <nick.desaulniers+lkml@...il.com>, Bill Wendling <morbo@...gle.com>, 
	Justin Stitt <justinstitt@...gle.com>, open list <linux-kernel@...r.kernel.org>, 
	"open list:CLANG/LLVM BUILD SUPPORT:Keyword:b(?i:clang|llvm)b" <llvm@...ts.linux.dev>
Subject: Re: [RFC bpf-next v2 0/1] libbpf: add compile-time OOB warning to bpf_tail_call_static

On Sat, Sep 6, 2025 at 3:57 AM Andrii Nakryiko
<andrii.nakryiko@...il.com> wrote:
>
> On Fri, Sep 5, 2025 at 1:42 AM Shung-Hsi Yu <shung-hsi.yu@...e.com> wrote:
> >
> > Adding some context that I think was miss per off-list discussion with
> > Hoyeon.
> >
> > On Fri, Sep 05, 2025 at 02:18:11PM +0900, Hoyeon Lee wrote:
> > > This RFC adds a compile-time check to bpf_tail_call_static() to warn
> > > when a constant slot(index) is >= map->max_entries. This uses a small
> > > BPF_MAP_ENTRIES() macro together with Clang's diagnose_if attribute.
> >
> > This is an attempt to see if it is possible to warn user of out-of-bound
> > tail calls, with the assumption being that with bpf_tail_call_static()
> > users would not be intentionally calling with an index that is superior
> > to the number of entries.
> >
> > However, there concerns with the current implementation, so this is
> > being sent as RFC to gather feedback, and to see if it can be better
> > done. Currently the concerns are:
> > - use macro to override bpf_tail_call_static()
> > - only works for Clang and not GCC
> > - uncertain whether this fit into libbpf conventions
>
> - map definition's max_entries can be set from user space at runtime
> making this check actively wrong
>
>
> This diagnose_if attribute seems very useful, but I'm not sure we
> should do this for anything map-related because statically provided
> map attributes are all overridable from user space when loading BPF
> object.
>

Thanks for pointing this out.

I hadn’t thought about map->max_entries being changed from user
space. That does make a compile-time check misleading, so I don’t
plan to take this RFC further. I’ve left some of the build workarounds
in the other thread in case they are useful.

Thanks to everyone for the time and thoughtful feedback, always
much appreciated.

> >
> > > Clang front-end keeps the map type with a '(*max_entries)[N]' field,
> > > so the expression
> > >
> > >     sizeof(*(m)->max_entries) / sizeof(**(m)->max_entries)
> > >
> > > is resolved to N entirely at compile time. This allows diagnose_if()
> > > to emit a warning when a constant slot index is out of range.
> > >
> > > Example:
> > >
> > >     struct { /* BPF_MAP_TYPE_PROG_ARRAY = 3 */
> > >         __uint(type, 3);             // int (*type)[3];
> > >         __uint(max_entries, 100);    // int (*max_entries)[100];
> > >         __type(key, __u32);          // typeof(__u32) *key;
> > >         __type(value, __u32);        // typeof(__u32) *value;
> > >     } progs SEC(".maps");
> > >
> > >     bpf_tail_call_static(ctx, &progs, 111);
> > >
> > > produces:
> > >
> > >     bound.bpf.c:26:9: warning: bpf_tail_call: slot >= max_entries [-Wuser-defined-warnings]
> > >        26 |         bpf_tail_call_static(ctx, &progs, 111);
> > >           |         ^
> > >     /usr/local/include/bpf/bpf_helpers.h:190:54: note: expanded from macro 'bpf_tail_call_static'
> > >       190 |          __bpf_tail_call_warn(__slot >= BPF_MAP_ENTRIES(map));                  \
> > >           |                                                             ^
> > >     /usr/local/include/bpf/bpf_helpers.h:183:20: note: from 'diagnose_if' attribute on '__bpf_tail_call_warn':
> > >       183 |     __attribute__((diagnose_if(oob, "bpf_tail_call: slot >= max_entries", "warning")));
> > >           |                    ^           ~~~
> > >
> > > Out-of-bounds tail call checkup is no-ops at runtime. Emitting a
> > > compile-time warning can help developers detect mistakes earlier. The
> > > check is currently limited to Clang (due to diagnose_if) and constant
> > > indices, but should catch common errors.
> > ...