Skip to contentSkip to conversation

bpf, bpftool: Support dumping kfunc prototypes from BTF

PATCH Superseded Daniel Xu 4 participants 12 messages Feb 4, 2024 – Mar 18, 2024

Subsystems (2)

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Daniel Xu

author
#0 lore raw

v2 cover bpf, bpftool: Support dumping kfunc prototypes from BTF

This patchset enables dumping kfunc prototypes from bpftool. This is useful b/c with this patchset, end users will no longer have to manually define kfunc prototypes. For the kernel tree, this also means we can drop kfunc prototypes from:

tools/testing/selftests/bpf/bpf_kfuncs.h tools/testing/selftests/bpf/bpf_experimental.h

Example usage:

        $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
        $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
        extern void cgroup_rstat_updated(struct cgroup *cgrp, int cpu) __weak __ksym;
        extern void cgroup_rstat_flush(struct cgroup *cgrp) __weak __ksym;
        extern struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) __weak __ksym;

Note that this patchset is only effective after the enabling pahole [0] change is merged and the resulting feature enabled with --btf_features=decl_tag_kfuncs.

[0]: https://lore.kernel.org/bpf/[email protected]/

=== Changelog ===

From v1:
* Add __weak annotation
* Use btf_dump for kfunc prototypes
* Update kernel bpf_rdonly_cast() signature

Daniel Xu (2):
bpf: Have bpf_rdonly_cast() take a const pointer
bpftool: Support dumping kfunc prototypes from BTF

 kernel/bpf/helpers.c    |  4 ++--
 tools/bpf/bpftool/btf.c | 45 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 47 insertions(+), 2 deletions(-)
Signature · 1 line
2.42.1
author
#1 lore raw

PATCH v2 1/2 bpf: Have bpf_rdonly_cast() take a const pointer

Since 20d59ee55172 ("libbpf: add bpf_core_cast() macro"), libbpf is now exporting a const arg version of bpf_rdonly_cast(). This causes the following conflicting type error when generating kfunc prototypes from BTF:

In file included from skeleton/pid_iter.bpf.c:5:
/home/dxu/dev/linux/tools/bpf/bpftool/bootstrap/libbpf/include/bpf/bpf_core_read.h:297:14: error: conflicting types for 'bpf_rdonly_cast'

extern void *bpf_rdonly_cast(const void *obj__ign, __u32 btf_id__k) __ksym __weak;
             ^

./vmlinux.h:135625:14: note: previous declaration is here extern void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) __weak __ksym;

This is b/c the kernel defines bpf_rdonly_cast() with non-const arg. Since const arg is more permissive and thus backwards compatible, we change the kernel definition as well to avoid conflicting type errors.

---
 kernel/bpf/helpers.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/bpf/helpers.c +2 -2
index 4db1c658254c..3503949b4c1b 100644
@@ -2484,9 +2484,9 @@ __bpf_kfunc void *bpf_cast_to_kern_ctx(void *obj)
 	return obj;
 }
 
-__bpf_kfunc void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k)
+__bpf_kfunc void *bpf_rdonly_cast(const void *obj__ign, u32 btf_id__k)
 {
-	return obj__ign;
+	return (void *)obj__ign;
 }
 
 __bpf_kfunc void bpf_rcu_read_lock(void)
Signature · 1 line
2.42.1
reviewer Acked-by on v2 1/2
#2 lore raw

Re: [PATCH bpf-next v2 1/2] bpf: Have bpf_rdonly_cast() take a const pointer

Quoted 9 lines · Daniel Xu

Since 20d59ee55172 ("libbpf: add bpf_core_cast() macro"), libbpf is now exporting a const arg version of bpf_rdonly_cast(). This causes the following conflicting type error when generating kfunc prototypes from BTF:

In file included from skeleton/pid_iter.bpf.c:5:
/home/dxu/dev/linux/tools/bpf/bpftool/bootstrap/libbpf/include/bpf/bpf_core_read.h:297:14: error: conflicting types for 'bpf_rdonly_cast'

extern void *bpf_rdonly_cast(const void *obj__ign, __u32 btf_id__k) __ksym __weak;
              ^
Quoted 8 lines

./vmlinux.h:135625:14: note: previous declaration is here extern void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) __weak __ksym;

This is b/c the kernel defines bpf_rdonly_cast() with non-const arg. Since const arg is more permissive and thus backwards compatible, we change the kernel definition as well to avoid conflicting type errors.

#3 lore raw

Re: [PATCH bpf-next v2 1/2] bpf: Have bpf_rdonly_cast() take a const pointer

In reply to Daniel Xu
Quoted 11 lines · Daniel Xu

Since 20d59ee55172 ("libbpf: add bpf_core_cast() macro"), libbpf is now exporting a const arg version of bpf_rdonly_cast(). This causes the following conflicting type error when generating kfunc prototypes from BTF:

In file included from skeleton/pid_iter.bpf.c:5:
/home/dxu/dev/linux/tools/bpf/bpftool/bootstrap/libbpf/include/bpf/bpf_core_read.h:297:14: error: conflicting types for 'bpf_rdonly_cast'

extern void *bpf_rdonly_cast(const void *obj__ign, __u32 btf_id__k) __ksym __weak;
             ^

./vmlinux.h:135625:14: note: previous declaration is here extern void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) __weak __ksym;

hi,
I'm hiting more of these when compiling bpf selftests (attached), it looks like some kfuncs declarations in bpf_kfuncs.h might be in conflict

jirka

---
CLNG-BPF [test_maps] connect_unix_prog.bpf.o
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:11:12: error: conflicting types for 'bpf_dynptr_from_skb' extern int bpf_dynptr_from_skb(struct __sk_buff *skb, __u64 flags,

           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164853:12: note: previous declaration is here
extern int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags, struct bpf_dynptr_kern *ptr__uninit) __weak __ksym;
           ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:19:12: error: conflicting types for 'bpf_dynptr_from_xdp'
extern int bpf_dynptr_from_xdp(struct xdp_md *xdp, __u64 flags,
           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164855:12: note: previous declaration is here
extern int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags, struct bpf_dynptr_kern *ptr__uninit) __weak __ksym;
           ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:28:14: error: conflicting types for 'bpf_dynptr_slice'
extern void *bpf_dynptr_slice(const struct bpf_dynptr *ptr, __u32 offset,
             ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164747:14: note: previous declaration is here
extern void *bpf_dynptr_slice(const struct bpf_dynptr_kern *ptr, u32 offset, void *buffer__opt, u32 buffer__szk) __weak __ksym;
             ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:37:14: error: conflicting types for 'bpf_dynptr_slice_rdwr'
extern void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr *ptr, __u32 offset,
             ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164749:14: note: previous declaration is here
extern void *bpf_dynptr_slice_rdwr(const struct bpf_dynptr_kern *ptr, u32 offset, void *buffer__opt, u32 buffer__szk) __weak __ksym;
             ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:40:12: error: conflicting types for 'bpf_dynptr_adjust'
extern int bpf_dynptr_adjust(const struct bpf_dynptr *ptr, __u32 start, __u32 end) __ksym __weak;
           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164781:12: note: previous declaration is here
extern int bpf_dynptr_adjust(struct bpf_dynptr_kern *ptr, u32 start, u32 end) __weak __ksym;
           ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:41:13: error: conflicting types for 'bpf_dynptr_is_null'
extern bool bpf_dynptr_is_null(const struct bpf_dynptr *ptr) __ksym __weak;
            ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164783:13: note: previous declaration is here
extern bool bpf_dynptr_is_null(struct bpf_dynptr_kern *ptr) __weak __ksym;
            ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:42:13: error: conflicting types for 'bpf_dynptr_is_rdonly'
extern bool bpf_dynptr_is_rdonly(const struct bpf_dynptr *ptr) __ksym __weak;
            ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164785:13: note: previous declaration is here
extern bool bpf_dynptr_is_rdonly(struct bpf_dynptr_kern *ptr) __weak __ksym;
            ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:43:14: error: conflicting types for 'bpf_dynptr_size'
extern __u32 bpf_dynptr_size(const struct bpf_dynptr *ptr) __ksym __weak;
             ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164787:14: note: previous declaration is here
extern __u32 bpf_dynptr_size(const struct bpf_dynptr_kern *ptr) __weak __ksym;
             ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:44:12: error: conflicting types for 'bpf_dynptr_clone'
extern int bpf_dynptr_clone(const struct bpf_dynptr *ptr, struct bpf_dynptr *clone__init) __ksym __weak;
           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164789:12: note: previous declaration is here
extern int bpf_dynptr_clone(struct bpf_dynptr_kern *ptr, struct bpf_dynptr_kern *clone__uninit) __weak __ksym;
           ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:61:12: error: conflicting types for 'bpf_sk_assign_tcp_reqsk'
extern int bpf_sk_assign_tcp_reqsk(struct __sk_buff *skb, struct sock *sk,
           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164859:12: note: previous declaration is here
extern int bpf_sk_assign_tcp_reqsk(struct sk_buff *skb, struct sock *sk, struct bpf_tcp_req_attrs *attrs, int attrs__sz) __weak __ksym;
           ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:68:12: error: conflicting types for 'bpf_get_file_xattr'
extern int bpf_get_file_xattr(struct file *file, const char *name,
           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164691:12: note: previous declaration is here
extern int bpf_get_file_xattr(struct file *file, const char *name__str, struct bpf_dynptr_kern *value_ptr) __weak __ksym;
           ^
In file included from progs/connect_unix_prog.c:9:
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/bpf_kfuncs.h:75:12: error: conflicting types for 'bpf_verify_pkcs7_signature'
extern int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_ptr,
           ^
/home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/tools/include/vmlinux.h:164689:12: note: previous declaration is here
extern int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr, struct bpf_dynptr_kern *sig_ptr, struct bpf_key *trusted_keyring) __weak __ksym;
           ^

12 errors generated.
make: *** [Makefile:642: /home/jolsa/kernel/linux-qemu/tools/testing/selftests/bpf/connect_unix_prog.bpf.o] Error 1

author
#4 lore raw

Re: [PATCH bpf-next v2 1/2] bpf: Have bpf_rdonly_cast() take a const pointer

Hi Jiri,

Quoted 13 lines · Jiri Olsa
Quoted 11 lines · Daniel Xu

Since 20d59ee55172 ("libbpf: add bpf_core_cast() macro"), libbpf is now exporting a const arg version of bpf_rdonly_cast(). This causes the following conflicting type error when generating kfunc prototypes from BTF:

In file included from skeleton/pid_iter.bpf.c:5:
/home/dxu/dev/linux/tools/bpf/bpftool/bootstrap/libbpf/include/bpf/bpf_core_read.h:297:14: error: conflicting types for 'bpf_rdonly_cast'

extern void *bpf_rdonly_cast(const void *obj__ign, __u32 btf_id__k) __ksym __weak;
             ^

./vmlinux.h:135625:14: note: previous declaration is here extern void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) __weak __ksym;

Quoted 3 lines

hi,
I'm hiting more of these when compiling bpf selftests (attached), it looks like some kfuncs declarations in bpf_kfuncs.h might be in conflict

Yep, I was actually going to put that as an office hours topic on how we want to handle that for selftests. Marking kfuncs in bpf_kfuncs.h and bpf_experimental.h as __weak is an option. ifdef is another option. Final option I can think of is bumping required pahole version up and simply deleting all the kfunc definitions.

But given that pahole changes come with the feature flag, I don't see this as a pressing issue. So I was planning on getting to that after current outstanding patchsets (just so there's less stuff for me to juggle).

[...]

Thanks,
Daniel

#5 lore raw

Re: [PATCH bpf-next v2 1/2] bpf: Have bpf_rdonly_cast() take a const pointer

Quoted 26 lines · Daniel Xu

Hi Jiri,

Quoted 13 lines · Jiri Olsa
Quoted 11 lines · Daniel Xu

Since 20d59ee55172 ("libbpf: add bpf_core_cast() macro"), libbpf is now exporting a const arg version of bpf_rdonly_cast(). This causes the following conflicting type error when generating kfunc prototypes from BTF:

In file included from skeleton/pid_iter.bpf.c:5:
/home/dxu/dev/linux/tools/bpf/bpftool/bootstrap/libbpf/include/bpf/bpf_core_read.h:297:14: error: conflicting types for 'bpf_rdonly_cast'

extern void *bpf_rdonly_cast(const void *obj__ign, __u32 btf_id__k) __ksym __weak;
             ^

./vmlinux.h:135625:14: note: previous declaration is here extern void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) __weak __ksym;

Quoted 3 lines

hi,
I'm hiting more of these when compiling bpf selftests (attached), it looks like some kfuncs declarations in bpf_kfuncs.h might be in conflict

Yep, I was actually going to put that as an office hours topic on how we want to handle that for selftests. Marking kfuncs in bpf_kfuncs.h and bpf_experimental.h as __weak is an option. ifdef is another option. Final option I can think of is bumping required pahole version up and simply deleting all the kfunc definitions.

Quoted 4 lines

But given that pahole changes come with the feature flag, I don't see this as a pressing issue. So I was planning on getting to that after current outstanding patchsets (just so there's less stuff for me to juggle).

ok, I guess if the fix goes in together with the scripts/Makefile.btf change then we're fine

jirka

#6 lore raw

Re: [PATCH bpf-next v2 1/2] bpf: Have bpf_rdonly_cast() take a const pointer

In reply to Daniel Xu
Quoted 21 lines · Daniel Xu

Hi Jiri,

Quoted 13 lines · Jiri Olsa
Quoted 11 lines · Daniel Xu

Since 20d59ee55172 ("libbpf: add bpf_core_cast() macro"), libbpf is now exporting a const arg version of bpf_rdonly_cast(). This causes the following conflicting type error when generating kfunc prototypes from BTF:

In file included from skeleton/pid_iter.bpf.c:5:
/home/dxu/dev/linux/tools/bpf/bpftool/bootstrap/libbpf/include/bpf/bpf_core_read.h:297:14: error: conflicting types for 'bpf_rdonly_cast'

extern void *bpf_rdonly_cast(const void *obj__ign, __u32 btf_id__k) __ksym __weak;
             ^

./vmlinux.h:135625:14: note: previous declaration is here extern void *bpf_rdonly_cast(void *obj__ign, u32 btf_id__k) __weak __ksym;

Quoted 3 lines

hi,
I'm hiting more of these when compiling bpf selftests (attached), it looks like some kfuncs declarations in bpf_kfuncs.h might be in conflict

Quoted 5 lines

Yep, I was actually going to put that as an office hours topic on how we want to handle that for selftests. Marking kfuncs in bpf_kfuncs.h and bpf_experimental.h as __weak is an option. ifdef is another option. Final option I can think of is bumping required pahole version up and simply deleting all the kfunc definitions.

I think we should mark all kfuncs in selftests as __weak. And then audit all of them and add consts where it makes sense and make sure that all definitions are compatible between vmlinux.h and manual declarations. I'd rather avoid #ifdef'ing anything, it should work as is.

Quoted 10 lines

But given that pahole changes come with the feature flag, I don't see this as a pressing issue. So I was planning on getting to that after current outstanding patchsets (just so there's less stuff for me to juggle).

[...]

Thanks,
Daniel

author
#7 lore raw

PATCH v2 2/2 bpftool: Support dumping kfunc prototypes from BTF

In reply to Daniel Xu

This patch enables dumping kfunc prototypes from bpftool. This is useful b/c with this patch, end users will no longer have to manually define kfunc prototypes. For the kernel tree, this also means we can drop kfunc prototypes from:

tools/testing/selftests/bpf/bpf_kfuncs.h tools/testing/selftests/bpf/bpf_experimental.h

Example usage:

        $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
        $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
        extern void cgroup_rstat_updated(struct cgroup *cgrp, int cpu) __weak __ksym;
        extern void cgroup_rstat_flush(struct cgroup *cgrp) __weak __ksym;
        extern struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) __weak __ksym;

Note that this patch is only effective after the enabling pahole [0] change is merged and the resulting feature enabled with --btf_features=decl_tag_kfuncs.

[0]: https://lore.kernel.org/bpf/[email protected]/

---
 tools/bpf/bpftool/btf.c | 45 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)
tools/bpf/bpftool/btf.c +45 -0
index 91fcb75babe3..0fd78a476286 100644
@@ -20,6 +20,8 @@
 #include "json_writer.h"
 #include "main.h"
 
+#define KFUNC_DECL_TAG		"bpf_kfunc"
+
 static const char * const btf_kind_str[NR_BTF_KINDS] = {
 	[BTF_KIND_UNKN]		= "UNKNOWN",
 	[BTF_KIND_INT]		= "INT",
@@ -454,6 +456,39 @@ static int dump_btf_raw(const struct btf *btf,
 	return 0;
 }
 
+static int dump_btf_kfuncs(struct btf_dump *d, const struct btf *btf)
+{
+	DECLARE_LIBBPF_OPTS(btf_dump_emit_type_decl_opts, opts);
+	int cnt = btf__type_cnt(btf);
+	int i;
+
+	for (i = 1; i < cnt; i++) {
+		const struct btf_type *t = btf__type_by_id(btf, i);
+		const struct btf_type *kft;
+		const char *name;
+		int err;
+
+		if (!btf_is_decl_tag(t))
+			continue;
+
+		name = btf__name_by_offset(btf, t->name_off);
+		if (strncmp(name, KFUNC_DECL_TAG, sizeof(KFUNC_DECL_TAG)))
+			continue;
+
+		printf("extern ");
+
+		kft = btf__type_by_id(btf, t->type);
+		opts.field_name = btf__name_by_offset(btf, kft->name_off);
+		err = btf_dump__emit_type_decl(d, kft->type, &opts);
+		if (err)
+			return err;
+
+		printf(" __weak __ksym;\n\n");
+	}
+
+	return 0;
+}
+
 static void __printf(2, 0) btf_dump_printf(void *ctx,
 					   const char *fmt, va_list args)
 {
@@ -476,6 +511,12 @@ static int dump_btf_c(const struct btf *btf,
 	printf("#ifndef BPF_NO_PRESERVE_ACCESS_INDEX\n");
 	printf("#pragma clang attribute push (__attribute__((preserve_access_index)), apply_to = record)\n");
 	printf("#endif\n\n");
+	printf("#ifndef __ksym\n");
+	printf("#define __ksym __attribute__((section(\".ksyms\")))\n");
+	printf("#endif\n\n");
+	printf("#ifndef __weak\n");
+	printf("#define __weak __attribute__((weak))\n");
+	printf("#endif\n\n");
 
 	if (root_type_cnt) {
 		for (i = 0; i < root_type_cnt; i++) {
@@ -491,6 +532,10 @@ static int dump_btf_c(const struct btf *btf,
 			if (err)
 				goto done;
 		}
+
+		err = dump_btf_kfuncs(d, btf);
+		if (err)
+			goto done;
 	}
 
 	printf("#ifndef BPF_NO_PRESERVE_ACCESS_INDEX\n");
Signature · 1 line
2.42.1
#8 lore raw

Re: [PATCH bpf-next v2 2/2] bpftool: Support dumping kfunc prototypes from BTF

Quoted 43 lines · Daniel Xu

This patch enables dumping kfunc prototypes from bpftool. This is useful b/c with this patch, end users will no longer have to manually define kfunc prototypes. For the kernel tree, this also means we can drop kfunc prototypes from:

tools/testing/selftests/bpf/bpf_kfuncs.h tools/testing/selftests/bpf/bpf_experimental.h

Example usage:

        $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
        $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
        extern void cgroup_rstat_updated(struct cgroup *cgrp, int cpu) __weak __ksym;
        extern void cgroup_rstat_flush(struct cgroup *cgrp) __weak __ksym;
        extern struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) __weak __ksym;

Note that this patch is only effective after the enabling pahole [0] change is merged and the resulting feature enabled with --btf_features=decl_tag_kfuncs.

[0]: https://lore.kernel.org/bpf/[email protected]/

---
 tools/bpf/bpftool/btf.c | 45 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)
tools/bpf/bpftool/btf.c +2 -0
index 91fcb75babe3..0fd78a476286 100644
@@ -20,6 +20,8 @@
 #include "json_writer.h"
 #include "main.h"
 
+#define KFUNC_DECL_TAG         "bpf_kfunc"
+
 static const char * const btf_kind_str[NR_BTF_KINDS] = {
        [BTF_KIND_UNKN]         = "UNKNOWN",
        [BTF_KIND_INT]          = "INT",

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 461

Quoted 6 lines
tools/bpf/bpftool/btf.c +3 -0
@@ -454,6 +456,39 @@ static int dump_btf_raw(const struct btf *btf,
        return 0;
 }
 
+static int dump_btf_kfuncs(struct btf_dump *d, const struct btf *btf)
+{
+       DECLARE_LIBBPF_OPTS(btf_dump_emit_type_decl_opts, opts);

nit: use shorter LIBBPF_OPTS, DECLARE_LIBBPF_OPTS is a "deprecated" macro name I hid, but didn't remove

Quoted 5 lines
tools/bpf/bpftool/btf.c +5 -0
+       int cnt = btf__type_cnt(btf);
+       int i;
+
+       for (i = 1; i < cnt; i++) {
+               const struct btf_type *t = btf__type_by_id(btf, i);

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 476

Quoted 10 lines
tools/bpf/bpftool/btf.c +10 -0
+               const struct btf_type *kft;
+               const char *name;
+               int err;
+
+               if (!btf_is_decl_tag(t))
+                       continue;
+
+               name = btf__name_by_offset(btf, t->name_off);
+               if (strncmp(name, KFUNC_DECL_TAG, sizeof(KFUNC_DECL_TAG)))
+                       continue;

should we do a bit more sanity checking here? Check that component_idx = -1 (entire func) and pointee type is FUNC?

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 480

Quoted 4 lines
tools/bpf/bpftool/btf.c +4 -0
+
+               printf("extern ");
+
+               kft = btf__type_by_id(btf, t->type);

nit: reuse t?

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 486

Quoted 6 lines
tools/bpf/bpftool/btf.c +6 -0
+               opts.field_name = btf__name_by_offset(btf, kft->name_off);
+               err = btf_dump__emit_type_decl(d, kft->type, &opts);
+               if (err)
+                       return err;
+
+               printf(" __weak __ksym;\n\n");

why extra endline?

though I'd ensure two empty lines before the first kfunc declaration to visually separate it from other type. Maybe even add a comment like /* BPF kfuncs */ or something like that?

Quoted 35 lines
tools/bpf/bpftool/btf.c +15 -1
+       }
+
+       return 0;
+}
+
 static void __printf(2, 0) btf_dump_printf(void *ctx,
                                           const char *fmt, va_list args)
 {
@@ -476,6 +511,12 @@ static int dump_btf_c(const struct btf *btf,
        printf("#ifndef BPF_NO_PRESERVE_ACCESS_INDEX\n");
        printf("#pragma clang attribute push (__attribute__((preserve_access_index)), apply_to = record)\n");
        printf("#endif\n\n");
+       printf("#ifndef __ksym\n");
+       printf("#define __ksym __attribute__((section(\".ksyms\")))\n");
+       printf("#endif\n\n");
+       printf("#ifndef __weak\n");
+       printf("#define __weak __attribute__((weak))\n");
+       printf("#endif\n\n");
 
        if (root_type_cnt) {
                for (i = 0; i < root_type_cnt; i++) {
@@ -491,6 +532,10 @@ static int dump_btf_c(const struct btf *btf,
                        if (err)
                                goto done;
                }
+
+               err = dump_btf_kfuncs(d, btf);
+               if (err)
+                       goto done;
        }
 
        printf("#ifndef BPF_NO_PRESERVE_ACCESS_INDEX\n");
--

2.42.1

author
#9 lore raw

Re: [PATCH bpf-next v2 2/2] bpftool: Support dumping kfunc prototypes from BTF

Hi Andrii,

Quoted 47 lines · Andrii Nakryiko
Quoted 46 lines · Daniel Xu

This patch enables dumping kfunc prototypes from bpftool. This is useful b/c with this patch, end users will no longer have to manually define kfunc prototypes. For the kernel tree, this also means we can drop kfunc prototypes from:

tools/testing/selftests/bpf/bpf_kfuncs.h tools/testing/selftests/bpf/bpf_experimental.h

Example usage:

        $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
        $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
        extern void cgroup_rstat_updated(struct cgroup *cgrp, int cpu) __weak __ksym;
        extern void cgroup_rstat_flush(struct cgroup *cgrp) __weak __ksym;
        extern struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) __weak __ksym;

Note that this patch is only effective after the enabling pahole [0] change is merged and the resulting feature enabled with --btf_features=decl_tag_kfuncs.

[0]: https://lore.kernel.org/bpf/[email protected]/

---
 tools/bpf/bpftool/btf.c | 45 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)
tools/bpf/bpftool/btf.c +2 -0
index 91fcb75babe3..0fd78a476286 100644
@@ -20,6 +20,8 @@
 #include "json_writer.h"
 #include "main.h"
 
+#define KFUNC_DECL_TAG         "bpf_kfunc"
+
 static const char * const btf_kind_str[NR_BTF_KINDS] = {
        [BTF_KIND_UNKN]         = "UNKNOWN",
        [BTF_KIND_INT]          = "INT",
@@ -454,6 +456,39 @@ static int dump_btf_raw(const struct btf *btf,
        return 0;
 }
 
Quoted 6 lines

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 461

Quoted 3 lines
tools/bpf/bpftool/btf.c +3 -0
+static int dump_btf_kfuncs(struct btf_dump *d, const struct btf *btf)
+{
+       DECLARE_LIBBPF_OPTS(btf_dump_emit_type_decl_opts, opts);

nit: use shorter LIBBPF_OPTS, DECLARE_LIBBPF_OPTS is a "deprecated" macro name I hid, but didn't remove

Ack.

Quoted 11 lines
Quoted 11 lines
tools/bpf/bpftool/btf.c +10 -0
+       int cnt = btf__type_cnt(btf);
+       int i;
+
+       for (i = 1; i < cnt; i++) {
+               const struct btf_type *t = btf__type_by_id(btf, i);
+               const struct btf_type *kft;
+               const char *name;
+               int err;
+
+               if (!btf_is_decl_tag(t))
Quoted 8 lines

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 476

Quoted 5 lines
tools/bpf/bpftool/btf.c +5 -0
+                       continue;
+
+               name = btf__name_by_offset(btf, t->name_off);
+               if (strncmp(name, KFUNC_DECL_TAG, sizeof(KFUNC_DECL_TAG)))
+                       continue;

should we do a bit more sanity checking here? Check that component_idx = -1 (entire func) and pointee type is FUNC?

Makes sense. Will add.

Quoted 8 lines

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 480

Quoted 5 lines
tools/bpf/bpftool/btf.c +4 -0
+
+               printf("extern ");
+
+               kft = btf__type_by_id(btf, t->type);

nit: reuse t?

Quoted 8 lines

tools/bpf/bpftool/btf.c static int dump_btf_raw(const struct btf *btf, · line 486

Quoted 6 lines
tools/bpf/bpftool/btf.c +6 -0
+               opts.field_name = btf__name_by_offset(btf, kft->name_off);
+               err = btf_dump__emit_type_decl(d, kft->type, &opts);
+               if (err)
+                       return err;
+
+               printf(" __weak __ksym;\n\n");

why extra endline?

Was thinking b/c other entities (structs) are double newline separated that we should keep it up for kfunc prototypes. No opinion -- I'll change it.

Quoted 4 lines

though I'd ensure two empty lines before the first kfunc declaration to visually separate it from other type. Maybe even add a comment like /* BPF kfuncs */ or something like that?

Ack.

Quoted 32 lines
Quoted 32 lines
tools/bpf/bpftool/btf.c +15 -0
+       }
+
+       return 0;
+}
+
 static void __printf(2, 0) btf_dump_printf(void *ctx,
                                           const char *fmt, va_list args)
 {
@@ -476,6 +511,12 @@ static int dump_btf_c(const struct btf *btf,
        printf("#ifndef BPF_NO_PRESERVE_ACCESS_INDEX\n");
        printf("#pragma clang attribute push (__attribute__((preserve_access_index)), apply_to = record)\n");
        printf("#endif\n\n");
+       printf("#ifndef __ksym\n");
+       printf("#define __ksym __attribute__((section(\".ksyms\")))\n");
+       printf("#endif\n\n");
+       printf("#ifndef __weak\n");
+       printf("#define __weak __attribute__((weak))\n");
+       printf("#endif\n\n");
 
        if (root_type_cnt) {
                for (i = 0; i < root_type_cnt; i++) {
@@ -491,6 +532,10 @@ static int dump_btf_c(const struct btf *btf,
                        if (err)
                                goto done;
                }
+
+               err = dump_btf_kfuncs(d, btf);
+               if (err)
+                       goto done;
        }
 
Quoted 4 lines
Quoted 4 lines
tools/bpf/bpftool/btf.c +0 -1
        printf("#ifndef BPF_NO_PRESERVE_ACCESS_INDEX\n");
--

2.42.1

I'll send the next rev after the pahole changes get merged.

Thanks,
Daniel

#10 lore raw
In reply to Daniel Xu
Quoted 26 lines · Daniel Xu

This patchset enables dumping kfunc prototypes from bpftool. This is useful b/c with this patchset, end users will no longer have to manually define kfunc prototypes. For the kernel tree, this also means we can drop kfunc prototypes from:

tools/testing/selftests/bpf/bpf_kfuncs.h tools/testing/selftests/bpf/bpf_experimental.h

Example usage:

        $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
        $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
        extern void cgroup_rstat_updated(struct cgroup *cgrp, int cpu) __weak __ksym;
        extern void cgroup_rstat_flush(struct cgroup *cgrp) __weak __ksym;
        extern struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) __weak __ksym;

Note that this patchset is only effective after the enabling pahole [0] change is merged and the resulting feature enabled with --btf_features=decl_tag_kfuncs.

[0]: https://lore.kernel.org/bpf/[email protected]/

=== Changelog ===

Quoted 8 lines

From v1:
* Add __weak annotation
* Use btf_dump for kfunc prototypes
* Update kernel bpf_rdonly_cast() signature

Daniel Xu (2):
bpf: Have bpf_rdonly_cast() take a const pointer
bpftool: Support dumping kfunc prototypes from BTF

I've applied patch #1 as it's a good change regardless. Please send v2 for patch #2.

Quoted 8 lines
 kernel/bpf/helpers.c    |  4 ++--
 tools/bpf/bpftool/btf.c | 45 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 47 insertions(+), 2 deletions(-)

--
2.42.1

author
#11 lore raw
Quoted 32 lines · Andrii Nakryiko
Quoted 31 lines · Daniel Xu

This patchset enables dumping kfunc prototypes from bpftool. This is useful b/c with this patchset, end users will no longer have to manually define kfunc prototypes. For the kernel tree, this also means we can drop kfunc prototypes from:

tools/testing/selftests/bpf/bpf_kfuncs.h tools/testing/selftests/bpf/bpf_experimental.h

Example usage:

        $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
        $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
        extern void cgroup_rstat_updated(struct cgroup *cgrp, int cpu) __weak __ksym;
        extern void cgroup_rstat_flush(struct cgroup *cgrp) __weak __ksym;
        extern struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) __weak __ksym;

Note that this patchset is only effective after the enabling pahole [0] change is merged and the resulting feature enabled with --btf_features=decl_tag_kfuncs.

[0]: https://lore.kernel.org/bpf/[email protected]/

=== Changelog ===

From v1:
* Add __weak annotation
* Use btf_dump for kfunc prototypes
* Update kernel bpf_rdonly_cast() signature

Quoted 6 lines
Quoted 3 lines

Daniel Xu (2):
bpf: Have bpf_rdonly_cast() take a const pointer
bpftool: Support dumping kfunc prototypes from BTF

I've applied patch #1 as it's a good change regardless. Please send v2 for patch #2.

Ack. Been a bit busy recently - will probably have time to work on this again next weekend.