libbpf: only reset sec_def handler when necessary
Andrii Nakryiko
maintainerPATCH 1/1 libbpf: only reset sec_def handler when necessary
Don't reset recorded sec_def handler unconditionally on bpf_program__set_type(). There are two situations where this is wrong.
First, if the program type didn't actually change. In that case original SEC handler should work just fine.
Second, catch-all custom SEC handler is supposed to work with any BPF program type and SEC() annotation, so it also doesn't make sense to reset that.
This patch fixes both issues. This was reported recently in the context of breaking perf tool, which uses custom catch-all handler for fancy BPF prologue generation logic. This patch should fix the issue.
[0] https://lore.kernel.org/linux-perf-users/[email protected]/
--- tools/lib/bpf/libbpf.c | 27 +++++++++++++++++++-------- 1 file changed, 19 insertions(+), 8 deletions(-)
tools/lib/bpf/libbpf.c +19 -8
@@ -8562,13 +8562,31 @@ enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)return prog->type; } +static size_t custom_sec_def_cnt; +static struct bpf_sec_def *custom_sec_defs; +static struct bpf_sec_def custom_fallback_def; +static bool has_custom_fallback_def; +static int last_custom_sec_def_handler_id; +int bpf_program__set_type (struct bpf_program *prog,enum bpf_prog_type type) {if (prog->obj->loaded)return libbpf_err (-EBUSY ); +/* if type is not changed, do nothing */ +if (prog->type == type) +return 0 ; + prog->type = type; - prog->sec_def =NULL ; + +/* If a program type was changed, we need to reset associated SEC() +* handler, as it will be invalid now. The only exception is a generic +* fallback handler, which by definition is program type-agnostic and +* is a catch-all custom handler, optionally set by the application, +* so should be able to handle any type of BPF program. +*/ +if (prog->sec_def != &custom_fallback_def) + prog->sec_def =NULL ;return 0 ; }
@@ -8744,13 +8762,6 @@ static const struct bpf_sec_def section_defs[] = { SEC_DEF("netfilter", NETFILTER, BPF_NETFILTER, SEC_NONE), }; -static size_t custom_sec_def_cnt; -static struct bpf_sec_def *custom_sec_defs; -static struct bpf_sec_def custom_fallback_def; -static bool has_custom_fallback_def; - -static int last_custom_sec_def_handler_id; - int libbpf_register_prog_handler(const char *sec, enum bpf_prog_type prog_type, enum bpf_attach_type exp_attach_type,
Signature · 1 line
2.34.1
Quoted 14 lines · Andrii Nakryiko
Don't reset recorded
sec_defhandler unconditionally onbpf_program__set_type(). There are two situations where this is wrong.First, if the program type didn't actually change. In that case original SEC handler should work just fine.
Second, catch-all custom SEC handler is supposed to work with any BPF program type and
SEC()annotation, so it also doesn't make sense to reset that.This patch fixes both issues. This was reported recently in the context of breaking perf tool, which uses custom catch-all handler for fancy BPF prologue generation logic. This patch should fix the issue.
Quoted 5 lines
[0] https://lore.kernel.org/linux-perf-users/[email protected]/
Fixes d6e6286a12e7 ("libbpf: disassociate section handler on explicit bpf_program__set_type() call")Reported-by Ravi Bangoria <[email protected]>Signed-off-by Andrii Nakryiko <[email protected]>
This patch was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:
Hello:
This patch was applied to bpf/bpf-next.git (master) by Alexei Starovoitov <[email protected]>:
Quoted 11 lines · On Fri, 7 Jul 2023 16:11:56 -0700 you wrote:
Don't reset recorded
sec_defhandler unconditionally onbpf_program__set_type(). There are two situations where this is wrong.First, if the program type didn't actually change. In that case original SEC handler should work just fine.
Second, catch-all custom SEC handler is supposed to work with any BPF program type and
SEC()annotation, so it also doesn't make sense to reset that.[...]
Here is the summary with links:
- [bpf-next] libbpf: only reset sec_def handler when necessary
https://git.kernel.org/bpf/bpf-next/c/c628747cc880
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
Quoted 14 lines · Andrii Nakryiko
Don't reset recorded
sec_defhandler unconditionally onbpf_program__set_type(). There are two situations where this is wrong.First, if the program type didn't actually change. In that case original SEC handler should work just fine.
Second, catch-all custom SEC handler is supposed to work with any BPF program type and
SEC()annotation, so it also doesn't make sense to reset that.This patch fixes both issues. This was reported recently in the context of breaking perf tool, which uses custom catch-all handler for fancy BPF prologue generation logic. This patch should fix the issue.
Quoted 5 lines
[0] https://lore.kernel.org/linux-perf-users/[email protected]/
Fixes d6e6286a12e7 ("libbpf: disassociate section handler on explicit bpf_program__set_type() call")Reported-by Ravi Bangoria <[email protected]>Signed-off-by Andrii Nakryiko <[email protected]>
FWIW, the cited commit broke multibuffer xdp programs (the frags prog_flags wasn't passed from userspace), this commit fixed the issue.
Thanks Andrii!