Recently, bpf_program__flags and bpf_program__set_extra_flags were
introduced to the libbpf API but they only allow adding load flags.
We have a use-case where we construct a skeleton with a sleepable
program and if it fails to load then we want to make it non-sleepable by
clearing BPF_F_SLEEPABLE.
Signed-off-by: Florent Revest <[email protected]>
---
tools/lib/bpf/libbpf.c | 9 +++++++++
tools/lib/bpf/libbpf.h | 1 +
tools/lib/bpf/libbpf.map | 1 +
3 files changed, 11 insertions(+)
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index de7e09a6b5ec..dcb7fced5fd2 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -8305,6 +8305,15 @@ int bpf_program__set_extra_flags(struct bpf_program *prog, __u32 extra_flags)
return 0;
}
+int bpf_program__clear_flags(struct bpf_program *prog, __u32 flags)
+{
+ if (prog->obj->loaded)
+ return libbpf_err(-EBUSY);
+
+ prog->prog_flags &= ~flags;
+ return 0;
+}
+
#define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
.sec = sec_pfx, \
.prog_type = BPF_PROG_TYPE_##ptype, \
diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
index 4ec69f224342..08f108e49841 100644
--- a/tools/lib/bpf/libbpf.h
+++ b/tools/lib/bpf/libbpf.h
@@ -495,6 +495,7 @@ bpf_program__set_expected_attach_type(struct bpf_program *prog,
LIBBPF_API __u32 bpf_program__flags(const struct bpf_program *prog);
LIBBPF_API int bpf_program__set_extra_flags(struct bpf_program *prog, __u32 extra_flags);
+LIBBPF_API int bpf_program__clear_flags(struct bpf_program *prog, __u32 flags);
LIBBPF_API int
bpf_program__set_attach_target(struct bpf_program *prog, int attach_prog_fd,
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
index 6a59514a48cf..eeff700240dc 100644
--- a/tools/lib/bpf/libbpf.map
+++ b/tools/lib/bpf/libbpf.map
@@ -401,6 +401,7 @@ LIBBPF_0.6.0 {
bpf_program__insn_cnt;
bpf_program__insns;
bpf_program__set_extra_flags;
+ bpf_program__clear_flags;
btf__add_btf;
btf__add_decl_tag;
btf__add_type_tag;
--
2.34.0.rc2.393.gf8c9666880-goog
On Thu, Nov 18, 2021 at 2:33 AM Florent Revest <[email protected]> wrote:
>
> Recently, bpf_program__flags and bpf_program__set_extra_flags were
> introduced to the libbpf API but they only allow adding load flags.
>
> We have a use-case where we construct a skeleton with a sleepable
> program and if it fails to load then we want to make it non-sleepable by
> clearing BPF_F_SLEEPABLE.
I'd say the better way to do this is to have two programs (that share
the logic, of course) and pick one or another at runtime:
static int whatever_logic(bool sleepable) { ... }
SEC("fentry.s/whatever")
int BPF_PROG(whatever_sleepable, ...)
{
return whatever_logic(true);
}
SEC("fentry/whatever")
int BPF_PROG(whatever_nonsleepable, ...)
{
return whatever_logic(false);
}
Then at runtime you can bpf_program__autoload(..., false) for a
variant you don't want to load.
This clear_flags business seems too low-level and too limited. Next
thing we'll be adding a few more bit manipulation variants (e.g, reset
flags). Let's see how far you can get with the use of existing
features. I'd set_extra_flags() to be almost never used, btw. And they
shouldn't, if can be avoided. So I'm hesitant to keep extending
operations around prog_flags.
But given we just added set_extra_flags() and it's already too
limiting, let's change set_extra flags to just set_flags() that will
override the flags with whatever user provides. Then with
bpf_program__flags() and bpf_program__set_flags() you can express
whatever you want without adding extra APIs. Care to fix that?
>
> Signed-off-by: Florent Revest <[email protected]>
> ---
> tools/lib/bpf/libbpf.c | 9 +++++++++
> tools/lib/bpf/libbpf.h | 1 +
> tools/lib/bpf/libbpf.map | 1 +
> 3 files changed, 11 insertions(+)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index de7e09a6b5ec..dcb7fced5fd2 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -8305,6 +8305,15 @@ int bpf_program__set_extra_flags(struct bpf_program *prog, __u32 extra_flags)
> return 0;
> }
>
> +int bpf_program__clear_flags(struct bpf_program *prog, __u32 flags)
> +{
> + if (prog->obj->loaded)
> + return libbpf_err(-EBUSY);
> +
> + prog->prog_flags &= ~flags;
> + return 0;
> +}
> +
> #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
> .sec = sec_pfx, \
> .prog_type = BPF_PROG_TYPE_##ptype, \
> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> index 4ec69f224342..08f108e49841 100644
> --- a/tools/lib/bpf/libbpf.h
> +++ b/tools/lib/bpf/libbpf.h
> @@ -495,6 +495,7 @@ bpf_program__set_expected_attach_type(struct bpf_program *prog,
>
> LIBBPF_API __u32 bpf_program__flags(const struct bpf_program *prog);
> LIBBPF_API int bpf_program__set_extra_flags(struct bpf_program *prog, __u32 extra_flags);
> +LIBBPF_API int bpf_program__clear_flags(struct bpf_program *prog, __u32 flags);
>
> LIBBPF_API int
> bpf_program__set_attach_target(struct bpf_program *prog, int attach_prog_fd,
> diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> index 6a59514a48cf..eeff700240dc 100644
> --- a/tools/lib/bpf/libbpf.map
> +++ b/tools/lib/bpf/libbpf.map
> @@ -401,6 +401,7 @@ LIBBPF_0.6.0 {
> bpf_program__insn_cnt;
> bpf_program__insns;
> bpf_program__set_extra_flags;
> + bpf_program__clear_flags;
> btf__add_btf;
> btf__add_decl_tag;
> btf__add_type_tag;
> --
> 2.34.0.rc2.393.gf8c9666880-goog
>
On Thu, Nov 18, 2021 at 6:49 PM Andrii Nakryiko
<[email protected]> wrote:
>
> On Thu, Nov 18, 2021 at 2:33 AM Florent Revest <[email protected]> wrote:
> >
> > We have a use-case where we construct a skeleton with a sleepable
> > program and if it fails to load then we want to make it non-sleepable by
> > clearing BPF_F_SLEEPABLE.
>
> I'd say the better way to do this is to have two programs (that share
> the logic, of course) and pick one or another at runtime:
>
> static int whatever_logic(bool sleepable) { ... }
>
> SEC("fentry.s/whatever")
> int BPF_PROG(whatever_sleepable, ...)
> {
> return whatever_logic(true);
> }
>
> SEC("fentry/whatever")
> int BPF_PROG(whatever_nonsleepable, ...)
> {
> return whatever_logic(false);
> }
>
>
> Then at runtime you can bpf_program__autoload(..., false) for a
> variant you don't want to load.
Ah cool, thanks! That's a good idea :) it will also look cleaner.
> This clear_flags business seems too low-level and too limited. Next
> thing we'll be adding a few more bit manipulation variants (e.g, reset
> flags). Let's see how far you can get with the use of existing
> features. I'd set_extra_flags() to be almost never used, btw. And they
> shouldn't, if can be avoided. So I'm hesitant to keep extending
> operations around prog_flags.
I agree
> But given we just added set_extra_flags() and it's already too
> limiting, let's change set_extra flags to just set_flags() that will
> override the flags with whatever user provides. Then with
> bpf_program__flags() and bpf_program__set_flags() you can express
> whatever you want without adding extra APIs. Care to fix that?
Sure, I'll send a patch for this! :)