linux-kernel - Re: [PATCH 5.8 35/99] tools/libbpf: Avoid counting local symbols in ABI check

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] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAFxkdAqNwLB6xs1c+0PveuWtj64BQVfsqkunufo0Kf8MNANMYQ@mail.gmail.com>
Date:   Wed, 30 Sep 2020 10:40:33 -0500
From:   Justin Forbes <jmforbes@...uxtx.org>
To:     Tony Ambardar <tony.ambardar@...il.com>
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Stable <stable@...r.kernel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Andrii Nakryiko <andriin@...com>,
        Sasha Levin <sashal@...nel.org>,
        Michael Ellerman <mpe@...erman.id.au>
Subject: Re: [PATCH 5.8 35/99] tools/libbpf: Avoid counting local symbols in
 ABI check

On Wed, Sep 30, 2020 at 12:02 AM Tony Ambardar <tony.ambardar@...il.com> wrote:
>
> [adding Michael Ellerman, linux-ppc maintainer]
>
> Hello Justin,
>
> On Tue, 29 Sep 2020 at 14:54, Justin Forbes <jmforbes@...uxtx.org> wrote:
> >
> > On Tue, Sep 29, 2020 at 6:53 AM Greg Kroah-Hartman
> > <gregkh@...uxfoundation.org> wrote:
> > >
> > > From: Tony Ambardar <tony.ambardar@...il.com>
> > >
> > > [ Upstream commit 746f534a4809e07f427f7d13d10f3a6a9641e5c3 ]
> > >
> > > Encountered the following failure building libbpf from kernel 5.8.5 sources
> > > with GCC 8.4.0 and binutils 2.34: (long paths shortened)
> > >
> > >   Warning: Num of global symbols in sharedobjs/libbpf-in.o (234) does NOT
> > >   match with num of versioned symbols in libbpf.so (236). Please make sure
> > >   all LIBBPF_API symbols are versioned in libbpf.map.
> > > #  --- libbpf_global_syms.tmp    2020-09-02 07:30:58.920084380 +0000
> > > #  +++ libbpf_versioned_syms.tmp 2020-09-02 07:30:58.924084388 +0000
> > >   @@ -1,3 +1,5 @@
> > >   +_fini
> > >   +_init
> > >    bpf_btf_get_fd_by_id
> > >    bpf_btf_get_next_id
> > >    bpf_create_map
> > >   make[4]: *** [Makefile:210: check_abi] Error 1
> > >
> > > Investigation shows _fini and _init are actually local symbols counted
> > > amongst global ones:
> > >
> > >   $ readelf --dyn-syms --wide libbpf.so|head -10
> > >
> > >   Symbol table '.dynsym' contains 343 entries:
> > >      Num:    Value  Size Type    Bind   Vis      Ndx Name
> > >        0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
> > >        1: 00004098     0 SECTION LOCAL  DEFAULT   11
> > >        2: 00004098     8 FUNC    LOCAL  DEFAULT   11 _init@@LIBBPF_0.0.1
> > >        3: 00023040     8 FUNC    LOCAL  DEFAULT   14 _fini@@LIBBPF_0.0.1
> > >        4: 00000000     0 OBJECT  GLOBAL DEFAULT  ABS LIBBPF_0.0.4
> > >        5: 00000000     0 OBJECT  GLOBAL DEFAULT  ABS LIBBPF_0.0.1
> > >        6: 0000ffa4     8 FUNC    GLOBAL DEFAULT   12 bpf_object__find_map_by_offset@@LIBBPF_0.0.1
> > >
> > > A previous commit filtered global symbols in sharedobjs/libbpf-in.o. Do the
> > > same with the libbpf.so DSO for consistent comparison.
> > >
> > > Fixes: 306b267cb3c4 ("libbpf: Verify versioned symbols")
> > > Signed-off-by: Tony Ambardar <Tony.Ambardar@...il.com>
> > > Signed-off-by: Alexei Starovoitov <ast@...nel.org>
> > > Acked-by: Andrii Nakryiko <andriin@...com>
> > > Link: https://lore.kernel.org/bpf/20200905214831.1565465-1-Tony.Ambardar@gmail.com
> > > Signed-off-by: Sasha Levin <sashal@...nel.org>
> >
> > This seems to work everywhere else, but breaks PPC64LE.
> >
>
> I also ran into a PPC build error while working on some bpf problems,
> but it seemed
> like a pre-existing PPC issue. I did submit an upstream fix, which is
> marked for stable
> and being reviewed by Michael. See here for discussion and the patch:
> https://lkml.org/lkml/2020/9/17/668.
>
> Is that the same problem you encountered? Does that patch address your issue?

It is not, the issue I see is:
Warning: Num of global symbols in sharedobjs/libbpf-in.o (259) does
NOT match with num of versioned symbols in libbpf.so (50). Please make
sure all LIBBPF_API symbols are versioned in libbpf.map.

I only see it on ppc64le with this patch, all other arch that Fedora
builds are fine (x86_64, i686, aarch64, armv7, s390).  If I revert
this patch, all builds succeed.  We are using gcc 10.2.1 though.

Justin

>
> Thanks,
> Tony
>
> > Justin
> >
> > > ---
> > >  tools/lib/bpf/Makefile |    2 ++
> > >  1 file changed, 2 insertions(+)
> > >
> > > --- a/tools/lib/bpf/Makefile
> > > +++ b/tools/lib/bpf/Makefile
> > > @@ -152,6 +152,7 @@ GLOBAL_SYM_COUNT = $(shell readelf -s --
> > >                            awk '/GLOBAL/ && /DEFAULT/ && !/UND/ {print $$NF}' | \
> > >                            sort -u | wc -l)
> > >  VERSIONED_SYM_COUNT = $(shell readelf --dyn-syms --wide $(OUTPUT)libbpf.so | \
> > > +                             awk '/GLOBAL/ && /DEFAULT/ && !/UND/ {print $$NF}' | \
> > >                               grep -Eo '[^ ]+@...BPF_' | cut -d@ -f1 | sort -u | wc -l)
> > >
> > >  CMD_TARGETS = $(LIB_TARGET) $(PC_FILE)
> > > @@ -219,6 +220,7 @@ check_abi: $(OUTPUT)libbpf.so
> > >                     awk '/GLOBAL/ && /DEFAULT/ && !/UND/ {print $$NF}'|  \
> > >                     sort -u > $(OUTPUT)libbpf_global_syms.tmp;           \
> > >                 readelf --dyn-syms --wide $(OUTPUT)libbpf.so |           \
> > > +                   awk '/GLOBAL/ && /DEFAULT/ && !/UND/ {print $$NF}'|  \
> > >                     grep -Eo '[^ ]+@...BPF_' | cut -d@ -f1 |             \
> > >                     sort -u > $(OUTPUT)libbpf_versioned_syms.tmp;        \
> > >                 diff -u $(OUTPUT)libbpf_global_syms.tmp                  \
> > >
> > >