| 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
| ||
|
Date: Tue, 17 Mar 2020 09:24:05 -0700
From: Stanislav Fomichev <sdf@...ichev.me>
To: Fangrui Song <maskray@...gle.com>
Cc: Andrii Nakryiko <andrii.nakryiko@...il.com>,
bpf <bpf@...r.kernel.org>, Networking <netdev@...r.kernel.org>,
Daniel Borkmann <daniel@...earbox.net>,
Nick Desaulniers <ndesaulniers@...gle.com>,
Nathan Chancellor <natechancellor@...il.com>,
clang-built-linux@...glegroups.com,
Stanislav Fomichev <sdf@...gle.com>,
"David S. Miller" <davem@...emloft.net>,
Alexei Starovoitov <ast@...nel.org>
Subject: Re: [PATCH bpf v3] bpf: Support llvm-objcopy and llvm-objdump for
vmlinux BTF
On 03/16, Fangrui Song wrote:
> On 2020-03-16, Andrii Nakryiko wrote:
> > On Mon, Mar 16, 2020 at 10:21 PM Fangrui Song <maskray@...gle.com> wrote:
> > >
> > >
> > > On 2020-03-16, Andrii Nakryiko wrote:
> > > >On Mon, Mar 16, 2020 at 8:37 PM Fangrui Song <maskray@...gle.com> wrote:
> > > >>
> > > >> On 2020-03-16, Andrii Nakryiko wrote:
> > > >> >On Mon, Mar 16, 2020 at 6:17 PM Fangrui Song <maskray@...gle.com> wrote:
> > > >> >>
> > > >> >> Simplify gen_btf logic to make it work with llvm-objcopy and
> > > >> >> llvm-objdump. We just need to retain one section .BTF. To do so, we can
> > > >> >> use a simple objcopy --only-section=.BTF instead of jumping all the
> > > >> >> hoops via an architecture-less binary file.
> > > >> >>
> > > >> >> We use a dd comment to change the e_type field in the ELF header from
> > > >> >> ET_EXEC to ET_REL so that .btf.vmlinux.bin.o will be accepted by lld.
> > > >> >>
> > > >> >> Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
> > > >> >> Cc: Stanislav Fomichev <sdf@...gle.com>
> > > >> >> Cc: Nick Desaulniers <ndesaulniers@...gle.com>
> > > >> >> Tested-by: Nick Desaulniers <ndesaulniers@...gle.com>
> > > >> >> Reported-by: Nathan Chancellor <natechancellor@...il.com>
> > > >> >> Link: https://github.com/ClangBuiltLinux/linux/issues/871
> > > >> >> Signed-off-by: Fangrui Song <maskray@...gle.com>
> > > >> >> ---
> > > >> >> scripts/link-vmlinux.sh | 13 ++-----------
> > > >> >> 1 file changed, 2 insertions(+), 11 deletions(-)
> > > >> >>
> > > >> >> diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
> > > >> >> index dd484e92752e..84be8d7c361d 100755
> > > >> >> --- a/scripts/link-vmlinux.sh
> > > >> >> +++ b/scripts/link-vmlinux.sh
> > > >> >> @@ -120,18 +120,9 @@ gen_btf()
> > > >> >>
> > > >> >> info "BTF" ${2}
> > > >> >> vmlinux_link ${1}
> > > >> >> - LLVM_OBJCOPY=${OBJCOPY} ${PAHOLE} -J ${1}
> > > >> >
> > > >> >Is it really tested? Seems like you just dropped .BTF generation step
> > > >> >completely...
> > > >>
> > > >> Sorry, dropped the whole line:/
> > > >> I don't know how to test .BTF . I can only check readelf -S...
> > > >>
> > > >> Attached the new patch.
> > > >>
> > > >>
> > > >> From 02afb9417d4f0f8d2175c94fc3797a94a95cc248 Mon Sep 17 00:00:00 2001
> > > >> From: Fangrui Song <maskray@...gle.com>
> > > >> Date: Mon, 16 Mar 2020 18:02:31 -0700
> > > >> Subject: [PATCH bpf v2] bpf: Support llvm-objcopy and llvm-objdump for
> > > >> vmlinux BTF
> > > >>
> > > >> Simplify gen_btf logic to make it work with llvm-objcopy and llvm-objdump.
> > > >> We use a dd comment to change the e_type field in the ELF header from
> > > >> ET_EXEC to ET_REL so that .btf.vmlinux.bin.o can be accepted by lld.
> > > >>
> > > >> Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
> > > >> Cc: Stanislav Fomichev <sdf@...gle.com>
> > > >> Cc: Nick Desaulniers <ndesaulniers@...gle.com>
> > > >> Reported-by: Nathan Chancellor <natechancellor@...il.com>
> > > >> Link: https://github.com/ClangBuiltLinux/linux/issues/871
> > > >> Signed-off-by: Fangrui Song <maskray@...gle.com>
> > > >> ---
> > > >> scripts/link-vmlinux.sh | 14 +++-----------
> > > >> 1 file changed, 3 insertions(+), 11 deletions(-)
> > > >>
> > > >> diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
> > > >> index dd484e92752e..b23313944c89 100755
> > > >> --- a/scripts/link-vmlinux.sh
> > > >> +++ b/scripts/link-vmlinux.sh
> > > >> @@ -120,18 +120,10 @@ gen_btf()
> > > >>
> > > >> info "BTF" ${2}
> > > >> vmlinux_link ${1}
> > > >> - LLVM_OBJCOPY=${OBJCOPY} ${PAHOLE} -J ${1}
> > > >> + ${PAHOLE} -J ${1}
> > > >
> > > >I'm not sure why you are touching this line at all. LLVM_OBJCOPY part
> > > >is necessary, pahole assumes llvm-objcopy by default, but that can
> > > >(and should for objcopy) be overridden with LLVM_OBJCOPY.
> > >
> > > Why is LLVM_OBJCOPY assumed? What if llvm-objcopy is not available?
> >
> > It's pahole assumption that we have to live with. pahole assumes
> > llvm-objcopy internally, unless it is overriden with LLVM_OBJCOPY env
> > var. So please revert this line otherwise you are breaking it for GCC
> > objcopy case.
>
> Acknowledged. Uploaded v3.
>
> I added back 2>/dev/null which was removed by a previous change, to
> suppress GNU objcopy warnings. The warnings could be annoying in V=1
> output.
>
> > > This is confusing that one tool assumes llvm-objcopy while the block
> > > below immediately uses GNU objcopy (without this patch).
> > >
> > > e83b9f55448afce3fe1abcd1d10db9584f8042a6 "kbuild: add ability to
> > > generate BTF type info for vmlinux" does not say why LLVM_OBJCOPY is
> > > set.
> > >
> > > >>
> > > >> - # dump .BTF section into raw binary file to link with final vmlinux
> > > >> - bin_arch=$(LANG=C ${OBJDUMP} -f ${1} | grep architecture | \
> > > >> - cut -d, -f1 | cut -d' ' -f2)
> > > >> - bin_format=$(LANG=C ${OBJDUMP} -f ${1} | grep 'file format' | \
> > > >> - awk '{print $4}')
> > > >> - ${OBJCOPY} --change-section-address .BTF=0 \
> > > >> - --set-section-flags .BTF=alloc -O binary \
> > > >> - --only-section=.BTF ${1} .btf.vmlinux.bin
> > > >> - ${OBJCOPY} -I binary -O ${bin_format} -B ${bin_arch} \
> > > >> - --rename-section .data=.BTF .btf.vmlinux.bin ${2}
> > > >> + # Extract .BTF section, change e_type to ET_REL, to link with final vmlinux
> > > >> + ${OBJCOPY} --only-section=.BTF ${1} ${2} && printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16
> > > >> }
> > > >>
> > > >> # Create ${2} .o file with all symbols from the ${1} object file
> > > >> --
> > > >> 2.25.1.481.gfbce0eb801-goog
> > > >>
>
> From ca3597477542453e9f63185c27c162da081a4baf Mon Sep 17 00:00:00 2001
> From: Fangrui Song <maskray@...gle.com>
> Date: Mon, 16 Mar 2020 22:38:23 -0700
> Subject: [PATCH bpf v3] bpf: Support llvm-objcopy and llvm-objdump for
> vmlinux BTF
>
> Simplify gen_btf logic to make it work with llvm-objcopy and llvm-objdump.
> Add 2>/dev/null to suppress GNU objcopy (but not llvm-objcopy) warnings
> "empty loadable segment detected at vaddr=0xffffffff81000000, is this intentional?"
> Our use of --only-section drops many SHF_ALLOC sections which will essentially nullify
> program headers. When used as linker input, program headers are simply
> ignored.
>
> We use a dd command to change the e_type field in the ELF header from
> ET_EXEC to ET_REL so that .btf.vmlinux.bin.o can be accepted by lld.
> Accepting ET_EXEC as an input file is an extremely rare GNU ld feature
> that lld does not intend to support, because this is very error-prone.
>
> Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
> Cc: Stanislav Fomichev <sdf@...gle.com>
> Cc: Nick Desaulniers <ndesaulniers@...gle.com>
> Reported-by: Nathan Chancellor <natechancellor@...il.com>
> Link: https://github.com/ClangBuiltLinux/linux/issues/871
> Signed-off-by: Fangrui Song <maskray@...gle.com>
> ---
> scripts/link-vmlinux.sh | 12 ++----------
> 1 file changed, 2 insertions(+), 10 deletions(-)
>
> diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
> index dd484e92752e..c3e808a89d4a 100755
> --- a/scripts/link-vmlinux.sh
> +++ b/scripts/link-vmlinux.sh
> @@ -122,16 +122,8 @@ gen_btf()
> vmlinux_link ${1}
> LLVM_OBJCOPY=${OBJCOPY} ${PAHOLE} -J ${1}
> - # dump .BTF section into raw binary file to link with final vmlinux
> - bin_arch=$(LANG=C ${OBJDUMP} -f ${1} | grep architecture | \
> - cut -d, -f1 | cut -d' ' -f2)
> - bin_format=$(LANG=C ${OBJDUMP} -f ${1} | grep 'file format' | \
> - awk '{print $4}')
> - ${OBJCOPY} --change-section-address .BTF=0 \
> - --set-section-flags .BTF=alloc -O binary \
> - --only-section=.BTF ${1} .btf.vmlinux.bin
> - ${OBJCOPY} -I binary -O ${bin_format} -B ${bin_arch} \
> - --rename-section .data=.BTF .btf.vmlinux.bin ${2}
> + # Extract .BTF section, change e_type to ET_REL, to link with final vmlinux
> + ${OBJCOPY} --only-section=.BTF ${1} ${2} 2> /dev/null && printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16
No, it doesn't work unfortunately, I get "in-kernel BTF is malformed"
from the kernel.
I think that's because -O binary adds the following:
$ nm .btf.vmxlinux.bin
00000000002f7bc9 D _binary__btf_vmlinux_bin_end
00000000002f7bc9 A _binary__btf_vmlinux_bin_size
0000000000000000 D _binary__btf_vmlinux_bin_start
While non-binary mode doesn't:
$ nm .btf.vmxlinux.bin
We don't add them manually in the linker map and expect objcopy to add
them, see:
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/tree/kernel/bpf/btf.c#n3480
> }
> # Create ${2} .o file with all symbols from the ${1} object file
> --
> 2.25.1.481.gfbce0eb801-goog
>
Powered by blists - more mailing lists