* [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
@ 2020-03-17 21:16 Fangrui Song
2020-03-17 21:51 ` Stanislav Fomichev
2020-03-17 21:55 ` Kees Cook
0 siblings, 2 replies; 8+ messages in thread
From: Fangrui Song @ 2020-03-17 21:16 UTC (permalink / raw)
To: bpf, Networking
Cc: Andrii Nakryiko, Daniel Borkmann, Nick Desaulniers,
Nathan Chancellor, clang-built-linux, Stanislav Fomichev,
David S. Miller, Alexei Starovoitov, Kees Cook
Simplify gen_btf logic to make it work with llvm-objcopy and llvm-objdump.
The existing 'file format' and 'architecture' parsing logic is brittle
and does not work with llvm-objcopy/llvm-objdump.
.BTF in .tmp_vmlinux.btf is non-SHF_ALLOC. Add the SHF_ALLOC flag and
rename .BTF to BTF so that C code can reference the section via linker
synthesized __start_BTF and __stop_BTF. This fixes a small problem that
previous .BTF had the SHF_WRITE flag. Additionally, `objcopy -I binary`
synthesized symbols _binary__btf_vmlinux_bin_start and
_binary__btf_vmlinux_bin_start (not used elsewhere) are replaced with
more common __start_BTF and __stop_BTF.
Add 2>/dev/null because GNU objcopy (but not llvm-objcopy) warns
"empty loadable segment detected at vaddr=0xffffffff81000000, is this intentional?"
We use a dd command to change the e_type field in the ELF header from
ET_EXEC to ET_REL so that lld will accept .btf.vmlinux.bin.o. Accepting
ET_EXEC as an input file is an extremely rare GNU ld feature that lld
does not intend to support, because this is error-prone.
Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
Cc: Stanislav Fomichev <sdf@google.com>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Reported-by: Nathan Chancellor <natechancellor@gmail.com>
Link: https://github.com/ClangBuiltLinux/linux/issues/871
Signed-off-by: Fangrui Song <maskray@google.com>
---
kernel/bpf/btf.c | 9 ++++-----
kernel/bpf/sysfs_btf.c | 11 +++++------
scripts/link-vmlinux.sh | 17 ++++++-----------
3 files changed, 15 insertions(+), 22 deletions(-)
diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index 50080add2ab9..6f397c4da05e 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -3477,8 +3477,8 @@ static struct btf *btf_parse(void __user *btf_data, u32 btf_data_size,
return ERR_PTR(err);
}
-extern char __weak _binary__btf_vmlinux_bin_start[];
-extern char __weak _binary__btf_vmlinux_bin_end[];
+extern char __weak __start_BTF[];
+extern char __weak __stop_BTF[];
extern struct btf *btf_vmlinux;
#define BPF_MAP_TYPE(_id, _ops)
@@ -3605,9 +3605,8 @@ struct btf *btf_parse_vmlinux(void)
}
env->btf = btf;
- btf->data = _binary__btf_vmlinux_bin_start;
- btf->data_size = _binary__btf_vmlinux_bin_end -
- _binary__btf_vmlinux_bin_start;
+ btf->data = __start_BTF;
+ btf->data_size = __stop_BTF - __start_BTF;
err = btf_parse_hdr(env);
if (err)
diff --git a/kernel/bpf/sysfs_btf.c b/kernel/bpf/sysfs_btf.c
index 7ae5dddd1fe6..3b495773de5a 100644
--- a/kernel/bpf/sysfs_btf.c
+++ b/kernel/bpf/sysfs_btf.c
@@ -9,15 +9,15 @@
#include <linux/sysfs.h>
/* See scripts/link-vmlinux.sh, gen_btf() func for details */
-extern char __weak _binary__btf_vmlinux_bin_start[];
-extern char __weak _binary__btf_vmlinux_bin_end[];
+extern char __weak __start_BTF[];
+extern char __weak __stop_BTF[];
static ssize_t
btf_vmlinux_read(struct file *file, struct kobject *kobj,
struct bin_attribute *bin_attr,
char *buf, loff_t off, size_t len)
{
- memcpy(buf, _binary__btf_vmlinux_bin_start + off, len);
+ memcpy(buf, __start_BTF + off, len);
return len;
}
@@ -30,15 +30,14 @@ static struct kobject *btf_kobj;
static int __init btf_vmlinux_init(void)
{
- if (!_binary__btf_vmlinux_bin_start)
+ if (!__start_BTF)
return 0;
btf_kobj = kobject_create_and_add("btf", kernel_kobj);
if (!btf_kobj)
return -ENOMEM;
- bin_attr_btf_vmlinux.size = _binary__btf_vmlinux_bin_end -
- _binary__btf_vmlinux_bin_start;
+ bin_attr_btf_vmlinux.size = __stop_BTF - __start_BTF;
return sysfs_create_bin_file(btf_kobj, &bin_attr_btf_vmlinux);
}
diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
index ac569e197bfa..ae2048625f1e 100755
--- a/scripts/link-vmlinux.sh
+++ b/scripts/link-vmlinux.sh
@@ -133,17 +133,12 @@ gen_btf()
info "BTF" ${2}
LLVM_OBJCOPY=${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}')
- bin_file=.btf.vmlinux.bin
- ${OBJCOPY} --change-section-address .BTF=0 \
- --set-section-flags .BTF=alloc -O binary \
- --only-section=.BTF ${1} $bin_file
- ${OBJCOPY} -I binary -O ${bin_format} -B ${bin_arch} \
- --rename-section .data=.BTF $bin_file ${2}
+ # Extract .BTF, add SHF_ALLOC, rename to BTF so that we can reference
+ # it via linker synthesized __start_BTF and __stop_BTF. Change e_type
+ # to ET_REL so that it can be used to link final vmlinux.
+ ${OBJCOPY} --only-section=.BTF --set-section-flags .BTF=alloc,readonly \
+ --rename-section .BTF=BTF ${1} ${2} 2>/dev/null && \
+ printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16 status=none
}
# Create ${2} .o file with all symbols from the ${1} object file
--
2.25.1.481.gfbce0eb801-goog
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 21:16 [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF Fangrui Song
@ 2020-03-17 21:51 ` Stanislav Fomichev
2020-03-17 22:01 ` Fangrui Song
2020-03-17 21:55 ` Kees Cook
1 sibling, 1 reply; 8+ messages in thread
From: Stanislav Fomichev @ 2020-03-17 21:51 UTC (permalink / raw)
To: Fangrui Song
Cc: bpf, Networking, Andrii Nakryiko, Daniel Borkmann,
Nick Desaulniers, Nathan Chancellor, clang-built-linux,
Stanislav Fomichev, David S. Miller, Alexei Starovoitov,
Kees Cook
On 03/17, Fangrui Song wrote:
> Simplify gen_btf logic to make it work with llvm-objcopy and llvm-objdump.
> The existing 'file format' and 'architecture' parsing logic is brittle
> and does not work with llvm-objcopy/llvm-objdump.
Thanks, it all makes sense and looks much better/portable (too much
dependence on binutils :-).
I've left a bunch of questions/nits below.
>
> .BTF in .tmp_vmlinux.btf is non-SHF_ALLOC. Add the SHF_ALLOC flag and
> rename .BTF to BTF so that C code can reference the section via linker
> synthesized __start_BTF and __stop_BTF. This fixes a small problem that
> previous .BTF had the SHF_WRITE flag. Additionally, `objcopy -I binary`
> synthesized symbols _binary__btf_vmlinux_bin_start and
> _binary__btf_vmlinux_bin_start (not used elsewhere) are replaced with
> more common __start_BTF and __stop_BTF.
>
> Add 2>/dev/null because GNU objcopy (but not llvm-objcopy) warns
> "empty loadable segment detected at vaddr=0xffffffff81000000, is this intentional?"
>
> We use a dd command to change the e_type field in the ELF header from
> ET_EXEC to ET_REL so that lld will accept .btf.vmlinux.bin.o. Accepting
> ET_EXEC as an input file is an extremely rare GNU ld feature that lld
> does not intend to support, because this is error-prone.
Please keep small changelog here, for example:
v5:
* rebased on top of bpfnext
Btw, I tried to pull and test it and failed:
$ curl -LO https://lore.kernel.org/bpf/20200317211649.o4fzaxrzy6qxvz4f@google.com/raw
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 9627 100 9627 0 0 47191 0 --:--:-- --:--:-- --:--:-- 47191
$ git am raw
warning: Patch sent with format=flowed; space at the end of lines might
be lost.
Applying: bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
error: corrupt patch at line 20
Patch failed at 0001 bpf: Support llvm-objcopy and llvm-objdump for
vmlinux BTF
hint: Use 'git am --show-current-patch' to see the failed patch
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".
$ git describe
v5.6-rc5-1621-g230021539e8c
Are you sure it's on top of the bpf-next? Or am I doing something wrong?
> Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
> Cc: Stanislav Fomichev <sdf@google.com>
> Cc: Nick Desaulniers <ndesaulniers@google.com>
> Reported-by: Nathan Chancellor <natechancellor@gmail.com>
> Link: https://github.com/ClangBuiltLinux/linux/issues/871
> Signed-off-by: Fangrui Song <maskray@google.com>
> ---
> kernel/bpf/btf.c | 9 ++++-----
> kernel/bpf/sysfs_btf.c | 11 +++++------
> scripts/link-vmlinux.sh | 17 ++++++-----------
> 3 files changed, 15 insertions(+), 22 deletions(-)
>
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index 50080add2ab9..6f397c4da05e 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -3477,8 +3477,8 @@ static struct btf *btf_parse(void __user *btf_data, u32 btf_data_size,
> return ERR_PTR(err);
> }
> -extern char __weak _binary__btf_vmlinux_bin_start[];
> -extern char __weak _binary__btf_vmlinux_bin_end[];
> +extern char __weak __start_BTF[];
> +extern char __weak __stop_BTF[];
> extern struct btf *btf_vmlinux;
> #define BPF_MAP_TYPE(_id, _ops)
> @@ -3605,9 +3605,8 @@ struct btf *btf_parse_vmlinux(void)
> }
> env->btf = btf;
> - btf->data = _binary__btf_vmlinux_bin_start;
> - btf->data_size = _binary__btf_vmlinux_bin_end -
> - _binary__btf_vmlinux_bin_start;
> + btf->data = __start_BTF;
> + btf->data_size = __stop_BTF - __start_BTF;
> err = btf_parse_hdr(env);
> if (err)
> diff --git a/kernel/bpf/sysfs_btf.c b/kernel/bpf/sysfs_btf.c
> index 7ae5dddd1fe6..3b495773de5a 100644
> --- a/kernel/bpf/sysfs_btf.c
> +++ b/kernel/bpf/sysfs_btf.c
> @@ -9,15 +9,15 @@
> #include <linux/sysfs.h>
> /* See scripts/link-vmlinux.sh, gen_btf() func for details */
> -extern char __weak _binary__btf_vmlinux_bin_start[];
> -extern char __weak _binary__btf_vmlinux_bin_end[];
> +extern char __weak __start_BTF[];
> +extern char __weak __stop_BTF[];
> static ssize_t
> btf_vmlinux_read(struct file *file, struct kobject *kobj,
> struct bin_attribute *bin_attr,
> char *buf, loff_t off, size_t len)
> {
> - memcpy(buf, _binary__btf_vmlinux_bin_start + off, len);
> + memcpy(buf, __start_BTF + off, len);
> return len;
> }
> @@ -30,15 +30,14 @@ static struct kobject *btf_kobj;
> static int __init btf_vmlinux_init(void)
> {
> - if (!_binary__btf_vmlinux_bin_start)
> + if (!__start_BTF)
> return 0;
> btf_kobj = kobject_create_and_add("btf", kernel_kobj);
> if (!btf_kobj)
> return -ENOMEM;
> - bin_attr_btf_vmlinux.size = _binary__btf_vmlinux_bin_end -
> - _binary__btf_vmlinux_bin_start;
> + bin_attr_btf_vmlinux.size = __stop_BTF - __start_BTF;
> return sysfs_create_bin_file(btf_kobj, &bin_attr_btf_vmlinux);
> }
> diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
> index ac569e197bfa..ae2048625f1e 100755
> --- a/scripts/link-vmlinux.sh
> +++ b/scripts/link-vmlinux.sh
> @@ -133,17 +133,12 @@ gen_btf()
> info "BTF" ${2}
> LLVM_OBJCOPY=${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}')
> - bin_file=.btf.vmlinux.bin
We still have the following in gen_btf that you need to remove:
local bin_arch
local bin_format
local bin_file
> - ${OBJCOPY} --change-section-address .BTF=0 \
> - --set-section-flags .BTF=alloc -O binary \
> - --only-section=.BTF ${1} $bin_file
> - ${OBJCOPY} -I binary -O ${bin_format} -B ${bin_arch} \
> - --rename-section .data=.BTF $bin_file ${2}
> + # Extract .BTF, add SHF_ALLOC, rename to BTF so that we can reference
> + # it via linker synthesized __start_BTF and __stop_BTF. Change e_type
> + # to ET_REL so that it can be used to link final vmlinux.
> + ${OBJCOPY} --only-section=.BTF --set-section-flags .BTF=alloc,readonly \
> + --rename-section .BTF=BTF ${1} ${2} 2>/dev/null && \
> + printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16 status=none
We have 'set -e' so && can be omitted. Maybe it will be a bit
more clear:
# Extract .BTF, add SHF_ALLOC, rename to BTF so that we can reference
# it via linker synthesized __start_BTF and __stop_BTF.
${OBJCOPY} --only-section=.BTF --set-section-flags .BTF=alloc,readonly \
--rename-section .BTF=BTF ${1} ${2} 2>/dev/null
# Change e_type to ET_REL (0x01) so that it can be used to link final
# vmlinux.
printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16 status=none
^^^ maybe also point out that this is required for llvm's ldd.
I know you point it out in the commit message, but still nice
to have as a comment here.
> }
> # Create ${2} .o file with all symbols from the ${1} object file
> --
> 2.25.1.481.gfbce0eb801-goog
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 21:16 [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF Fangrui Song
2020-03-17 21:51 ` Stanislav Fomichev
@ 2020-03-17 21:55 ` Kees Cook
1 sibling, 0 replies; 8+ messages in thread
From: Kees Cook @ 2020-03-17 21:55 UTC (permalink / raw)
To: Fangrui Song
Cc: bpf, Networking, Andrii Nakryiko, Daniel Borkmann,
Nick Desaulniers, Nathan Chancellor, clang-built-linux,
Stanislav Fomichev, David S. Miller, Alexei Starovoitov
On Tue, Mar 17, 2020 at 02:16:49PM -0700, Fangrui Song wrote:
> Simplify gen_btf logic to make it work with llvm-objcopy and llvm-objdump.
> The existing 'file format' and 'architecture' parsing logic is brittle
> and does not work with llvm-objcopy/llvm-objdump.
>
> .BTF in .tmp_vmlinux.btf is non-SHF_ALLOC. Add the SHF_ALLOC flag and
> rename .BTF to BTF so that C code can reference the section via linker
> synthesized __start_BTF and __stop_BTF. This fixes a small problem that
> previous .BTF had the SHF_WRITE flag. Additionally, `objcopy -I binary`
> synthesized symbols _binary__btf_vmlinux_bin_start and
> _binary__btf_vmlinux_bin_start (not used elsewhere) are replaced with
> more common __start_BTF and __stop_BTF.
I'm glad to see the name change benefit here. Just reducing the number
of execs in this path is only worth this change, IMO. Going from 2
objdump and 2 objcopy calls to a single objcopy is very nice. :)
>
> Add 2>/dev/null because GNU objcopy (but not llvm-objcopy) warns
> "empty loadable segment detected at vaddr=0xffffffff81000000, is this intentional?"
>
> We use a dd command to change the e_type field in the ELF header from
> ET_EXEC to ET_REL so that lld will accept .btf.vmlinux.bin.o. Accepting
> ET_EXEC as an input file is an extremely rare GNU ld feature that lld
> does not intend to support, because this is error-prone.
Nice ELF trick. :)
>
> Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
> Cc: Stanislav Fomichev <sdf@google.com>
> Cc: Nick Desaulniers <ndesaulniers@google.com>
> Reported-by: Nathan Chancellor <natechancellor@gmail.com>
> Link: https://github.com/ClangBuiltLinux/linux/issues/871
> Signed-off-by: Fangrui Song <maskray@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
-Kees
> ---
> kernel/bpf/btf.c | 9 ++++-----
> kernel/bpf/sysfs_btf.c | 11 +++++------
> scripts/link-vmlinux.sh | 17 ++++++-----------
> 3 files changed, 15 insertions(+), 22 deletions(-)
>
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index 50080add2ab9..6f397c4da05e 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -3477,8 +3477,8 @@ static struct btf *btf_parse(void __user *btf_data, u32 btf_data_size,
> return ERR_PTR(err);
> }
> -extern char __weak _binary__btf_vmlinux_bin_start[];
> -extern char __weak _binary__btf_vmlinux_bin_end[];
> +extern char __weak __start_BTF[];
> +extern char __weak __stop_BTF[];
> extern struct btf *btf_vmlinux;
> #define BPF_MAP_TYPE(_id, _ops)
> @@ -3605,9 +3605,8 @@ struct btf *btf_parse_vmlinux(void)
> }
> env->btf = btf;
> - btf->data = _binary__btf_vmlinux_bin_start;
> - btf->data_size = _binary__btf_vmlinux_bin_end -
> - _binary__btf_vmlinux_bin_start;
> + btf->data = __start_BTF;
> + btf->data_size = __stop_BTF - __start_BTF;
> err = btf_parse_hdr(env);
> if (err)
> diff --git a/kernel/bpf/sysfs_btf.c b/kernel/bpf/sysfs_btf.c
> index 7ae5dddd1fe6..3b495773de5a 100644
> --- a/kernel/bpf/sysfs_btf.c
> +++ b/kernel/bpf/sysfs_btf.c
> @@ -9,15 +9,15 @@
> #include <linux/sysfs.h>
> /* See scripts/link-vmlinux.sh, gen_btf() func for details */
> -extern char __weak _binary__btf_vmlinux_bin_start[];
> -extern char __weak _binary__btf_vmlinux_bin_end[];
> +extern char __weak __start_BTF[];
> +extern char __weak __stop_BTF[];
> static ssize_t
> btf_vmlinux_read(struct file *file, struct kobject *kobj,
> struct bin_attribute *bin_attr,
> char *buf, loff_t off, size_t len)
> {
> - memcpy(buf, _binary__btf_vmlinux_bin_start + off, len);
> + memcpy(buf, __start_BTF + off, len);
> return len;
> }
> @@ -30,15 +30,14 @@ static struct kobject *btf_kobj;
> static int __init btf_vmlinux_init(void)
> {
> - if (!_binary__btf_vmlinux_bin_start)
> + if (!__start_BTF)
> return 0;
> btf_kobj = kobject_create_and_add("btf", kernel_kobj);
> if (!btf_kobj)
> return -ENOMEM;
> - bin_attr_btf_vmlinux.size = _binary__btf_vmlinux_bin_end -
> - _binary__btf_vmlinux_bin_start;
> + bin_attr_btf_vmlinux.size = __stop_BTF - __start_BTF;
> return sysfs_create_bin_file(btf_kobj, &bin_attr_btf_vmlinux);
> }
> diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
> index ac569e197bfa..ae2048625f1e 100755
> --- a/scripts/link-vmlinux.sh
> +++ b/scripts/link-vmlinux.sh
> @@ -133,17 +133,12 @@ gen_btf()
> info "BTF" ${2}
> LLVM_OBJCOPY=${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}')
> - bin_file=.btf.vmlinux.bin
> - ${OBJCOPY} --change-section-address .BTF=0 \
> - --set-section-flags .BTF=alloc -O binary \
> - --only-section=.BTF ${1} $bin_file
> - ${OBJCOPY} -I binary -O ${bin_format} -B ${bin_arch} \
> - --rename-section .data=.BTF $bin_file ${2}
> + # Extract .BTF, add SHF_ALLOC, rename to BTF so that we can reference
> + # it via linker synthesized __start_BTF and __stop_BTF. Change e_type
> + # to ET_REL so that it can be used to link final vmlinux.
> + ${OBJCOPY} --only-section=.BTF --set-section-flags .BTF=alloc,readonly \
> + --rename-section .BTF=BTF ${1} ${2} 2>/dev/null && \
> + printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16 status=none
> }
> # Create ${2} .o file with all symbols from the ${1} object file
> --
> 2.25.1.481.gfbce0eb801-goog
--
Kees Cook
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 21:51 ` Stanislav Fomichev
@ 2020-03-17 22:01 ` Fangrui Song
2020-03-17 22:08 ` Nick Desaulniers
0 siblings, 1 reply; 8+ messages in thread
From: Fangrui Song @ 2020-03-17 22:01 UTC (permalink / raw)
To: Stanislav Fomichev
Cc: bpf, Networking, Andrii Nakryiko, Daniel Borkmann,
Nick Desaulniers, Nathan Chancellor, clang-built-linux,
Stanislav Fomichev, David S. Miller, Alexei Starovoitov,
Kees Cook
On 2020-03-17, Stanislav Fomichev wrote:
>On 03/17, Fangrui Song wrote:
>> Simplify gen_btf logic to make it work with llvm-objcopy and llvm-objdump.
>> The existing 'file format' and 'architecture' parsing logic is brittle
>> and does not work with llvm-objcopy/llvm-objdump.
>Thanks, it all makes sense and looks much better/portable (too much
>dependence on binutils :-).
>I've left a bunch of questions/nits below.
>
>>
>> .BTF in .tmp_vmlinux.btf is non-SHF_ALLOC. Add the SHF_ALLOC flag and
>> rename .BTF to BTF so that C code can reference the section via linker
>> synthesized __start_BTF and __stop_BTF. This fixes a small problem that
>> previous .BTF had the SHF_WRITE flag. Additionally, `objcopy -I binary`
>> synthesized symbols _binary__btf_vmlinux_bin_start and
>> _binary__btf_vmlinux_bin_start (not used elsewhere) are replaced with
>> more common __start_BTF and __stop_BTF.
>>
>> Add 2>/dev/null because GNU objcopy (but not llvm-objcopy) warns
>> "empty loadable segment detected at vaddr=0xffffffff81000000, is this intentional?"
>>
>> We use a dd command to change the e_type field in the ELF header from
>> ET_EXEC to ET_REL so that lld will accept .btf.vmlinux.bin.o. Accepting
>> ET_EXEC as an input file is an extremely rare GNU ld feature that lld
>> does not intend to support, because this is error-prone.
>Please keep small changelog here, for example:
>
>v5:
>* rebased on top of bpfnext
Thanks for the tip. Add them at the bottom?
>Btw, I tried to pull and test it and failed:
>$ curl -LO https://lore.kernel.org/bpf/20200317211649.o4fzaxrzy6qxvz4f@google.com/raw
> % Total % Received % Xferd Average Speed Time Time Time Current
> Dload Upload Total Spent Left Speed
>100 9627 100 9627 0 0 47191 0 --:--:-- --:--:-- --:--:-- 47191
>$ git am raw
>warning: Patch sent with format=flowed; space at the end of lines might
>be lost.
>Applying: bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
>error: corrupt patch at line 20
>Patch failed at 0001 bpf: Support llvm-objcopy and llvm-objdump for
>vmlinux BTF
>hint: Use 'git am --show-current-patch' to see the failed patch
>When you have resolved this problem, run "git am --continue".
>If you prefer to skip this patch, run "git am --skip" instead.
>To restore the original branch and stop patching, run "git am --abort".
>$ git describe
>v5.6-rc5-1621-g230021539e8c
>
>Are you sure it's on top of the bpf-next? Or am I doing something wrong?
It can be my fault when messing with various patches..
Will be more cautious.
>> Fixes: df786c9b9476 ("bpf: Force .BTF section start to zero when dumping from vmlinux")
>> Cc: Stanislav Fomichev <sdf@google.com>
>> Cc: Nick Desaulniers <ndesaulniers@google.com>
>> Reported-by: Nathan Chancellor <natechancellor@gmail.com>
>> Link: https://github.com/ClangBuiltLinux/linux/issues/871
>> Signed-off-by: Fangrui Song <maskray@google.com>
>> ---
>> kernel/bpf/btf.c | 9 ++++-----
>> kernel/bpf/sysfs_btf.c | 11 +++++------
>> scripts/link-vmlinux.sh | 17 ++++++-----------
>> 3 files changed, 15 insertions(+), 22 deletions(-)
>>
>> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
>> index 50080add2ab9..6f397c4da05e 100644
>> --- a/kernel/bpf/btf.c
>> +++ b/kernel/bpf/btf.c
>> @@ -3477,8 +3477,8 @@ static struct btf *btf_parse(void __user *btf_data, u32 btf_data_size,
>> return ERR_PTR(err);
>> }
>> -extern char __weak _binary__btf_vmlinux_bin_start[];
>> -extern char __weak _binary__btf_vmlinux_bin_end[];
>> +extern char __weak __start_BTF[];
>> +extern char __weak __stop_BTF[];
>> extern struct btf *btf_vmlinux;
>> #define BPF_MAP_TYPE(_id, _ops)
>> @@ -3605,9 +3605,8 @@ struct btf *btf_parse_vmlinux(void)
>> }
>> env->btf = btf;
>> - btf->data = _binary__btf_vmlinux_bin_start;
>> - btf->data_size = _binary__btf_vmlinux_bin_end -
>> - _binary__btf_vmlinux_bin_start;
>> + btf->data = __start_BTF;
>> + btf->data_size = __stop_BTF - __start_BTF;
>> err = btf_parse_hdr(env);
>> if (err)
>> diff --git a/kernel/bpf/sysfs_btf.c b/kernel/bpf/sysfs_btf.c
>> index 7ae5dddd1fe6..3b495773de5a 100644
>> --- a/kernel/bpf/sysfs_btf.c
>> +++ b/kernel/bpf/sysfs_btf.c
>> @@ -9,15 +9,15 @@
>> #include <linux/sysfs.h>
>> /* See scripts/link-vmlinux.sh, gen_btf() func for details */
>> -extern char __weak _binary__btf_vmlinux_bin_start[];
>> -extern char __weak _binary__btf_vmlinux_bin_end[];
>> +extern char __weak __start_BTF[];
>> +extern char __weak __stop_BTF[];
>> static ssize_t
>> btf_vmlinux_read(struct file *file, struct kobject *kobj,
>> struct bin_attribute *bin_attr,
>> char *buf, loff_t off, size_t len)
>> {
>> - memcpy(buf, _binary__btf_vmlinux_bin_start + off, len);
>> + memcpy(buf, __start_BTF + off, len);
>> return len;
>> }
>> @@ -30,15 +30,14 @@ static struct kobject *btf_kobj;
>> static int __init btf_vmlinux_init(void)
>> {
>> - if (!_binary__btf_vmlinux_bin_start)
>> + if (!__start_BTF)
>> return 0;
>> btf_kobj = kobject_create_and_add("btf", kernel_kobj);
>> if (!btf_kobj)
>> return -ENOMEM;
>> - bin_attr_btf_vmlinux.size = _binary__btf_vmlinux_bin_end -
>> - _binary__btf_vmlinux_bin_start;
>> + bin_attr_btf_vmlinux.size = __stop_BTF - __start_BTF;
>> return sysfs_create_bin_file(btf_kobj, &bin_attr_btf_vmlinux);
>> }
>> diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
>> index ac569e197bfa..ae2048625f1e 100755
>> --- a/scripts/link-vmlinux.sh
>> +++ b/scripts/link-vmlinux.sh
>> @@ -133,17 +133,12 @@ gen_btf()
>> info "BTF" ${2}
>> LLVM_OBJCOPY=${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}')
>> - bin_file=.btf.vmlinux.bin
>We still have the following in gen_btf that you need to remove:
> local bin_arch
> local bin_format
> local bin_file
Thanks. Will delete them.
>> - ${OBJCOPY} --change-section-address .BTF=0 \
>> - --set-section-flags .BTF=alloc -O binary \
>> - --only-section=.BTF ${1} $bin_file
>> - ${OBJCOPY} -I binary -O ${bin_format} -B ${bin_arch} \
>> - --rename-section .data=.BTF $bin_file ${2}
>> + # Extract .BTF, add SHF_ALLOC, rename to BTF so that we can reference
>> + # it via linker synthesized __start_BTF and __stop_BTF. Change e_type
>> + # to ET_REL so that it can be used to link final vmlinux.
>> + ${OBJCOPY} --only-section=.BTF --set-section-flags .BTF=alloc,readonly \
>> + --rename-section .BTF=BTF ${1} ${2} 2>/dev/null && \
>> + printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16 status=none
> We have 'set -e' so && can be omitted. Maybe it will be a bit
> more clear:
Agree.
> # Extract .BTF, add SHF_ALLOC, rename to BTF so that we can reference
> # it via linker synthesized __start_BTF and __stop_BTF.
> ${OBJCOPY} --only-section=.BTF --set-section-flags .BTF=alloc,readonly \
> --rename-section .BTF=BTF ${1} ${2} 2>/dev/null
>
> # Change e_type to ET_REL (0x01) so that it can be used to link final
> # vmlinux.
> printf '\1' | dd of=${2} conv=notrunc bs=1 seek=16 status=none
>
> ^^^ maybe also point out that this is required for llvm's ldd.
> I know you point it out in the commit message, but still nice
> to have as a comment here.
Will mention lld in v6.
>> }
>> # Create ${2} .o file with all symbols from the ${1} object file
>> --
>> 2.25.1.481.gfbce0eb801-goog
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 22:01 ` Fangrui Song
@ 2020-03-17 22:08 ` Nick Desaulniers
2020-03-17 22:12 ` Stanislav Fomichev
0 siblings, 1 reply; 8+ messages in thread
From: Nick Desaulniers @ 2020-03-17 22:08 UTC (permalink / raw)
To: Fangrui Song
Cc: Stanislav Fomichev, bpf, Networking, Andrii Nakryiko,
Daniel Borkmann, Nathan Chancellor, clang-built-linux,
Stanislav Fomichev, David S. Miller, Alexei Starovoitov,
Kees Cook
On Tue, Mar 17, 2020 at 3:01 PM Fangrui Song <maskray@google.com> wrote:
>
> On 2020-03-17, Stanislav Fomichev wrote:
> >Please keep small changelog here, for example:
> >
> >v5:
> >* rebased on top of bpfnext
>
> Thanks for the tip. Add them at the bottom?
"Below the fold" see this patch I just sent out:
https://lore.kernel.org/lkml/20200317215515.226917-1-ndesaulniers@google.com/T/#u
grep "Changes"
$ git format-patch -v2 HEAD~
$ vim 0001-...patch
<manually add changelog "below the fold">
--
Thanks,
~Nick Desaulniers
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 22:08 ` Nick Desaulniers
@ 2020-03-17 22:12 ` Stanislav Fomichev
2020-03-17 23:14 ` Nick Desaulniers
0 siblings, 1 reply; 8+ messages in thread
From: Stanislav Fomichev @ 2020-03-17 22:12 UTC (permalink / raw)
To: Nick Desaulniers
Cc: Fangrui Song, Stanislav Fomichev, bpf, Networking,
Andrii Nakryiko, Daniel Borkmann, Nathan Chancellor,
clang-built-linux, David S. Miller, Alexei Starovoitov,
Kees Cook
On Tue, Mar 17, 2020 at 3:08 PM Nick Desaulniers
<ndesaulniers@google.com> wrote:
>
> On Tue, Mar 17, 2020 at 3:01 PM Fangrui Song <maskray@google.com> wrote:
> >
> > On 2020-03-17, Stanislav Fomichev wrote:
> > >Please keep small changelog here, for example:
> > >
> > >v5:
> > >* rebased on top of bpfnext
> >
> > Thanks for the tip. Add them at the bottom?
>
> "Below the fold" see this patch I just sent out:
> https://lore.kernel.org/lkml/20200317215515.226917-1-ndesaulniers@google.com/T/#u
> grep "Changes"
>
> $ git format-patch -v2 HEAD~
> $ vim 0001-...patch
> <manually add changelog "below the fold">
BPF subtree prefers the changelog in the commit body, not the comments
(iow, before ---).
Add them at the end of you message, see, for example:
https://lore.kernel.org/bpf/a428fb88-9b53-27dd-a195-497755944921@iogearbox.net/T/
> --
> Thanks,
> ~Nick Desaulniers
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 22:12 ` Stanislav Fomichev
@ 2020-03-17 23:14 ` Nick Desaulniers
2020-03-17 23:44 ` Daniel Borkmann
0 siblings, 1 reply; 8+ messages in thread
From: Nick Desaulniers @ 2020-03-17 23:14 UTC (permalink / raw)
To: Stanislav Fomichev
Cc: Fangrui Song, Stanislav Fomichev, bpf, Networking,
Andrii Nakryiko, Daniel Borkmann, Nathan Chancellor,
clang-built-linux, David S. Miller, Alexei Starovoitov,
Kees Cook
On Tue, Mar 17, 2020 at 3:13 PM Stanislav Fomichev <sdf@google.com> wrote:
>
> On Tue, Mar 17, 2020 at 3:08 PM Nick Desaulniers
> <ndesaulniers@google.com> wrote:
> >
> > On Tue, Mar 17, 2020 at 3:01 PM Fangrui Song <maskray@google.com> wrote:
> > >
> > > On 2020-03-17, Stanislav Fomichev wrote:
> > > >Please keep small changelog here, for example:
> > > >
> > > >v5:
> > > >* rebased on top of bpfnext
> > >
> > > Thanks for the tip. Add them at the bottom?
> >
> > "Below the fold" see this patch I just sent out:
> > https://lore.kernel.org/lkml/20200317215515.226917-1-ndesaulniers@google.com/T/#u
> > grep "Changes"
> >
> > $ git format-patch -v2 HEAD~
> > $ vim 0001-...patch
> > <manually add changelog "below the fold">
> BPF subtree prefers the changelog in the commit body, not the comments
> (iow, before ---).
> Add them at the end of you message, see, for example:
> https://lore.kernel.org/bpf/a428fb88-9b53-27dd-a195-497755944921@iogearbox.net/T/
Sigh, every maintainer is a special snowflake. In our tree, you're
only allowed to commit on Thursdays under a blood moon. /s
But thanks for the note.
--
Thanks,
~Nick Desaulniers
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF
2020-03-17 23:14 ` Nick Desaulniers
@ 2020-03-17 23:44 ` Daniel Borkmann
0 siblings, 0 replies; 8+ messages in thread
From: Daniel Borkmann @ 2020-03-17 23:44 UTC (permalink / raw)
To: Nick Desaulniers, Stanislav Fomichev
Cc: Fangrui Song, Stanislav Fomichev, bpf, Networking,
Andrii Nakryiko, Nathan Chancellor, clang-built-linux,
David S. Miller, Alexei Starovoitov, Kees Cook
On 3/18/20 12:14 AM, Nick Desaulniers wrote:
> On Tue, Mar 17, 2020 at 3:13 PM Stanislav Fomichev <sdf@google.com> wrote:
>> On Tue, Mar 17, 2020 at 3:08 PM Nick Desaulniers
>> <ndesaulniers@google.com> wrote:
>>> On Tue, Mar 17, 2020 at 3:01 PM Fangrui Song <maskray@google.com> wrote:
>>>> On 2020-03-17, Stanislav Fomichev wrote:
>>>>> Please keep small changelog here, for example:
>>>>>
>>>>> v5:
>>>>> * rebased on top of bpfnext
>>>>
>>>> Thanks for the tip. Add them at the bottom?
>>>
>>> "Below the fold" see this patch I just sent out:
>>> https://lore.kernel.org/lkml/20200317215515.226917-1-ndesaulniers@google.com/T/#u
>>> grep "Changes"
>>>
>>> $ git format-patch -v2 HEAD~
>>> $ vim 0001-...patch
>>> <manually add changelog "below the fold">
>> BPF subtree prefers the changelog in the commit body, not the comments
>> (iow, before ---).
>> Add them at the end of you message, see, for example:
>> https://lore.kernel.org/bpf/a428fb88-9b53-27dd-a195-497755944921@iogearbox.net/T/
>
> Sigh, every maintainer is a special snowflake. In our tree, you're
> only allowed to commit on Thursdays under a blood moon. /s
>
> But thanks for the note.
For every commit into bpf/bpf-next we automatically put a 'Link:' tag pointing
to the lore.kernel.org/bpf long-term archive, so there is always an easy way to
retrieve additional information or discussions ([0] as one example). So whether
you put it above or below the '---' line I personally don't mind either way. All
crucial information that helps a reader later on to understand /why/ we ended up
exactly with the current change should be part of the main commit log though.
[0] https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/commit/?id=483d7a30f538e2f8addd32aa9a3d2e94ae55fa65
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2020-03-17 23:44 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-03-17 21:16 [PATCH bpf-next v5] bpf: Support llvm-objcopy and llvm-objdump for vmlinux BTF Fangrui Song
2020-03-17 21:51 ` Stanislav Fomichev
2020-03-17 22:01 ` Fangrui Song
2020-03-17 22:08 ` Nick Desaulniers
2020-03-17 22:12 ` Stanislav Fomichev
2020-03-17 23:14 ` Nick Desaulniers
2020-03-17 23:44 ` Daniel Borkmann
2020-03-17 21:55 ` Kees Cook
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).