Re: [PATCH] selftests/bpf: Fix bind{4,6} tcp/socket header type conflict

From: James Hilliard
Date: Fri Aug 26 2022 - 13:15:49 EST


On Fri, Aug 26, 2022 at 10:33 AM Jose E. Marchesi
<jose.marchesi@xxxxxxxxxx> wrote:
>
>
> > On Fri, Aug 26, 2022 at 7:19 AM Jose E. Marchesi
> > <jose.marchesi@xxxxxxxxxx> wrote:
> >>
> >>
> >> > On Thu, Aug 25, 2022 at 11:49 PM Martin KaFai Lau <kafai@xxxxxx> wrote:
> >> >>
> >> >> On Thu, Aug 25, 2022 at 11:31:15PM -0600, James Hilliard wrote:
> >> >> > On Thu, Aug 25, 2022 at 11:16 PM Martin KaFai Lau <kafai@xxxxxx> wrote:
> >> >> > >
> >> >> > > On Thu, Aug 25, 2022 at 04:17:49PM -0600, James Hilliard wrote:
> >> >> > > > There is a potential for us to hit a type conflict when including
> >> >> > > > netinet/tcp.h with sys/socket.h, we can replace both of these includes
> >> >> > > > with linux/tcp.h to avoid this conflict.
> >> >> > > >
> >> >> > > > Fixes errors like:
> >> >> > > > In file included from /usr/include/netinet/tcp.h:91,
> >> >> > > > from progs/bind4_prog.c:10:
> >> >> > > > /home/buildroot/opt/cross/lib/gcc/bpf/13.0.0/include/stdint.h:34:23: error: conflicting types for 'int8_t'; have 'char'
> >> >> > > > 34 | typedef __INT8_TYPE__ int8_t;
> >> >> > > > | ^~~~~~
> >> >> > > > In file included from /usr/include/x86_64-linux-gnu/sys/types.h:155,
> >> >> > > > from /usr/include/x86_64-linux-gnu/bits/socket.h:29,
> >> >> > > > from /usr/include/x86_64-linux-gnu/sys/socket.h:33,
> >> >> > > > from progs/bind4_prog.c:9:
> >> >> > > > /usr/include/x86_64-linux-gnu/bits/stdint-intn.h:24:18: note: previous declaration of 'int8_t' with type 'int8_t' {aka 'signed char'}
> >> >> > > > 24 | typedef __int8_t int8_t;
> >> >> > > > | ^~~~~~
> >> >> > > > /home/buildroot/opt/cross/lib/gcc/bpf/13.0.0/include/stdint.h:43:24:
> >> >> > > > error: conflicting types for 'int64_t'; have 'long int'
> >> >> > > > 43 | typedef __INT64_TYPE__ int64_t;
> >> >> > > > | ^~~~~~~
> >> >> > > > /usr/include/x86_64-linux-gnu/bits/stdint-intn.h:27:19: note:
> >> >> > > > previous declaration of 'int64_t' with type 'int64_t' {aka
> >> >> > > > 'long long int'}
> >> >> > > > 27 | typedef __int64_t int64_t;
> >> >> > > > | ^~~~~~~
> >> >> > > > make: *** [Makefile:537:
> >> >> > > > /home/buildroot/bpf-next/tools/testing/selftests/bpf/bpf_gcc/bind4_prog.o]
> >> >> > > > Error 1
> >> >> > > >
> >> >> > > > Signed-off-by: James Hilliard <james.hilliard1@xxxxxxxxx>
> >> >> > > > ---
> >> >> > > > tools/testing/selftests/bpf/progs/bind4_prog.c | 3 +--
> >> >> > > > tools/testing/selftests/bpf/progs/bind6_prog.c | 3 +--
> >> >> > > > 2 files changed, 2 insertions(+), 4 deletions(-)
> >> >> > > >
> >> >> > > > diff --git a/tools/testing/selftests/bpf/progs/bind4_prog.c
> >> >> > > > b/tools/testing/selftests/bpf/progs/bind4_prog.c
> >> >> > > > index 474c6a62078a..6bd20042fd53 100644
> >> >> > > > --- a/tools/testing/selftests/bpf/progs/bind4_prog.c
> >> >> > > > +++ b/tools/testing/selftests/bpf/progs/bind4_prog.c
> >> >> > > > @@ -6,8 +6,7 @@
> >> >> > > > #include <linux/bpf.h>
> >> >> > > > #include <linux/in.h>
> >> >> > > > #include <linux/in6.h>
> >> >> > > > -#include <sys/socket.h>
> >> >> > > > -#include <netinet/tcp.h>
> >> >> > > These includes look normal to me. What environment is hitting this.
> >> >> >
> >> >> > I was hitting this error with GCC 13(GCC master branch).
> >> >> These two includes (<sys/socket.h> and <netinet/tcp.h>) are normal,
> >> >> so does it mean all existing programs need to change to use gcc 13 ?
> >> >
> >> > Well I think it's mostly just an issue getting hit with GCC-BPF as it
> >> > looks to me like a cross compilation host/target header conflict.
> >>
> >> This is an interesting issue.
> >>
> >> Right now the BPF GCC target is a sort of a bare-metal target. As such,
> >> it provides a set of header files that implement ISO C types and other
> >> machinery (i.e. it doesn't rely on a C library to provide them):
> >>
> >> iso646.h
> >> stdalign.h
> >> stdarg.h
> >> stdatomic.h
> >> stdbool.h
> >> stddef.h
> >> stdfix.h
> >> stdint.h
> >> stdnoreturn.h
> >> tgmath.h
> >> unwind.h
> >> varargs.h
> >>
> >> This is because we were expecting this to be used like:
> >>
> >> <compiler-provided std C headers>
> >> | |
> >> v |
> >> <kernel headers> |
> >> | |
> >> v v
> >> <BPF C program>
> >>
> >> However, if it is expected/intended for C BPF programs to include libc
> >> headers, such as sys/socket.h, this can quickly go sour as you have
> >> found with that conflict.
> >>
> >> So this leads to the question: should we turn the BPF target into a
> >> target that assumes a libc? This basically means we will be assuming
> >> BPF programs are always compiled in an environment that provides a
> >> standard stdint.h, stdbool.h and friends.
> >
> > Well for a normal GCC BPF setup we're basically cross compiling for the
> > BPF bare metal target while sharing headers with the build host(for libbpf
> > and any other libc headers that get included).
> >
> > On the other hand when using GCC BPF as part of a full cross toolchain
> > we actually end up sharing headers with our real cross target architecture
> > sysroot(which would provide a libc), essentially in that case BPF is a bare
> > metal cross target which shares headers with the real cross target(which
> > is not a bare metal target). For this libbpf is installed to the real
> > cross target
> > sysroot which is used by both GCC BPF(for bpf progs) and the real cross
> > target GCC compiler(for userspace side). From my understanding with this
> > setup GCC BPF will pick up the real cross target libc headers as a fallback
> > which may sometimes have conflict/compatibility issues with the kernel
> > headers.
> >
> > I think it's probably best to avoid depending on libc headers as things may
> > otherwise get even more complex. You would essentially have 2 libc's
> > in a normal GCC BPF setup and 3 libc's in a full cross toolchain setup(you'd
> > have one for the build host, one for the real cross target arch and one for
> > the BPF target arch).
> >
> > Cross build systems will typically allow a libc choice as
> > well(glibc/musl/uclibc)
> > and we don't really want the bpf programs to have to care about the specific
> > libc being used as they are bare metal programs which shouldn't depend on
> > a libc.
> >
>
> I don't understand what do you mean with "real cross target".

I mean the real cross target architecture as in the real hardware target
architecture, for example aarch64 when cross compiling from a x86_64
build host.

>
> From the toolchain perspective, the compiler is targetted to just one
> platform: bpf-unknown-none. As is usual for bare-metal targets, the
> compiler provides headers to implement the C standard with things like
> floating-point types and standard integer types, `bool', etc.

Yeah, I mean gcc doesn't have proper multi-arch support like llvm does
so a complete gcc cross toolchain(one which is sufficient to build
kernel/userspace needed for say an aarch64 cross target along with bpf
programs) is effectively two gcc toolchains bundled together. In some ways
it gets used more like a separate language than a separate target.

I have a pending series for buildroot adding gcc-bpf support if you're
curious what this currently looks like:
https://lore.kernel.org/buildroot/20220809094109.2279598-1-james.hilliard1@xxxxxxxxx/

>
> If you then -I directories in order to "share headers with the build
> host" or with that "real cross target", or to use any other header that
> may implement the same types (typically a libc) then well, thats when
> the problem arises.

Well I'm using -idirafter for including those build host/real cross target
header directories with lowest priority, since those directories have least
priority the conflicts would otherwise be missing header errors AFAIU if
I didn't include them.

>From my understanding we need to include these directories as they
provide the kernel headers required by many bpf programs.

>
> I don't know how much sense does it makes to include glibc headers like
> sys/socket.h in BPF C programs: I'm no BPF programmer. But if it is
> something to be supported, we will have to change the compiler to not
> provide the standard headers.

I think it's best to just avoid libc headers in BPF programs.

>
> >> Thoughts?
> >>
> >> >> > > I don't prefer the selftest writers need to remember this rule.
> >> >> > >
> >> >> > > Beside, afaict, tcp.h should be removed because
> >> >> > > I don't see this test needs it. I tried removing it
> >> >> > > and it works fine. It should be removed instead of replacing it
> >> >> > > with another unnecessary tcp.h.
> >> >> >
> >> >> > Oh, that does also appear to work, thought I had tried that already but I guess
> >> >> > I hadn't, sent a v2 with them removed:
> >> >> > https://lore.kernel.org/bpf/20220826052925.980431-1-james.hilliard1@xxxxxxxxx/T/#u
> >> >> >
> >> >> > >
> >> >> > > > +#include <linux/tcp.h>
> >> >> > > > #include <linux/if.h>
> >> >> > > > #include <errno.h>
> >> >> > > >
> >> >> > > > diff --git a/tools/testing/selftests/bpf/progs/bind6_prog.c
> >> >> > > > b/tools/testing/selftests/bpf/progs/bind6_prog.c
> >> >> > > > index c19cfa869f30..f37617b35a55 100644
> >> >> > > > --- a/tools/testing/selftests/bpf/progs/bind6_prog.c
> >> >> > > > +++ b/tools/testing/selftests/bpf/progs/bind6_prog.c
> >> >> > > > @@ -6,8 +6,7 @@
> >> >> > > > #include <linux/bpf.h>
> >> >> > > > #include <linux/in.h>
> >> >> > > > #include <linux/in6.h>
> >> >> > > > -#include <sys/socket.h>
> >> >> > > > -#include <netinet/tcp.h>
> >> >> > > > +#include <linux/tcp.h>
> >> >> > > > #include <linux/if.h>
> >> >> > > > #include <errno.h>
> >> >> > > >
> >> >> > > > --
> >> >> > > > 2.34.1
> >> >> > > >