Re: [PATCH] bpf.h: fix clang compiler warning with unpriv_ebpf_notify()

[Luis Chamberlain]

On Wed, May 11, 2022 at 06:17:26PM +0200, Daniel Borkmann wrote:
> On 5/11/22 6:08 PM, Luis Chamberlain wrote:
> > On Wed, May 11, 2022 at 09:03:13AM -0700, Alexei Starovoitov wrote:
> > > On Wed, May 11, 2022 at 8:58 AM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> > > > On Mon, May 09, 2022 at 01:36:23PM -0700, Luis Chamberlain wrote:
> > > > > The recent commit "bpf: Move BPF sysctls from kernel/sysctl.c to BPF core"
> > > > > triggered 0-day to issue an email for what seems to have been an old
> > > > > clang warning. So this issue should have existed before as well, from
> > > > > what I can tell. The issue is that clang expects a forward declaration
> > > > > for routines declared as weak while gcc does not.
> > > > > 
> > > > > This can be reproduced with 0-day's x86_64-randconfig-c007
> > > > > https://download.01.org/0day-ci/archive/20220424/202204240008.JDntM9cU-lkp@xxxxxxxxx/config
> > > > > 
> > > > > And using:
> > > > > 
> > > > > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 ARCH=x86_64 SHELL=/bin/bash kernel/bpf/syscall.o
> > > > > Compiler will be installed in /home/mcgrof/0day
> > > > > make --keep-going HOSTCC=/home/mcgrof/0day/clang/bin/clang CC=/home/mcgrof/0day/clang/bin/clang LD=/home/mcgrof/0day/clang/bin/ld.lld HOSTLD=/home/mcgrof/0day/clang/bin/ld.lld AR=llvm-ar NM=llvm-nm STRIP=llvm-strip OBJCOPY=llvm-objcopy OBJDUMP=llvm-objdump OBJSIZE=llvm-size READELF=llvm-readelf HOSTCXX=clang++ HOSTAR=llvm-ar CROSS_COMPILE=x86_64-linux-gnu- --jobs=24 W=1 ARCH=x86_64 SHELL=/bin/bash kernel/bpf/syscall.o
> > > > >    DESCEND objtool
> > > > >    CALL    scripts/atomic/check-atomics.sh
> > > > >    CALL    scripts/checksyscalls.sh
> > > > >    CC      kernel/bpf/syscall.o
> > > > > kernel/bpf/syscall.c:4944:13: warning: no previous prototype for function 'unpriv_ebpf_notify' [-Wmissing-prototypes]
> > > > > void __weak unpriv_ebpf_notify(int new_state)
> > > > >              ^
> > > > > kernel/bpf/syscall.c:4944:1: note: declare 'static' if the function is not intended to be used outside of this translation unit
> > > > > void __weak unpriv_ebpf_notify(int new_state)
> > > > > ^
> > > > > static
> > > > > 
> > > > > Fixes: 2900005ea287 ("bpf: Move BPF sysctls from kernel/sysctl.c to BPF core")
> > > > > Signed-off-by: Luis Chamberlain <mcgrof@xxxxxxxxxx>
> > > > > ---
> > > > > 
> > > > > Daniel,
> > > > > 
> > > > > Given what we did fore 2900005ea287 ("bpf: Move BPF sysctls from
> > > > > kernel/sysctl.c to BPF core") where I had pulled pr/bpf-sysctl a
> > > > > while ago into sysctl-next and then merged the patch in question,
> > > > > should I just safely carry this patch onto sysctl-next? Let me know
> > > > > how you'd like to proceed.
> > > > > 
> > > > > Also, it wasn't clear if putting this forward declaration on
> > > > > bpf.h was your ideal preference.
> > > > 
> > > > After testing this on sysctl-testing without issues going to move this
> > > > to sysctl-next now.
> > > 
> > > Hmm. No.
> > > A similar patch should be in tip already. You have to wait
> > > for it to go through Linus's tree and back to whatever tree you use.
> > 
> > I'm a bit confused, the patch in question which my patch fixes should only
> > be in my sysctl-next tree at this point, not in Linus's tree.
> 
> Borislav was planning to route it via tip tree, maybe confusion was that the
> fix in the link below is from Josh:
> 
> https://lore.kernel.org/bpf/CAADnVQKjfQMG_zFf9F9P7m0UzqESs7XoRy=udqrDSodxa8yBpg@xxxxxxxxxxxxxx/

Ah, Josh posted a fix for the same compile warning.

> But I presume this is routed as fix to Linus, so should land in both sysctl
> and bpf tree at some point after re-sync.

It may be the case indeed that the code in question was triggering a
compile warning without the patch I have merged which moves the BPF
sysctls ("bpf: Move BPF sysctls from kernel/sysctl.c to BPF core").

So I'll just drop my fix.

  Luis