libbpf: Add support for prog_tracing
authorAlexei Starovoitov <ast@kernel.org>
Wed, 30 Oct 2019 22:32:12 +0000 (15:32 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Thu, 31 Oct 2019 14:16:59 +0000 (15:16 +0100)
Cleanup libbpf from expected_attach_type == attach_btf_id hack
and introduce BPF_PROG_TYPE_TRACING.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20191030223212.953010-3-ast@kernel.org
tools/include/uapi/linux/bpf.h
tools/lib/bpf/bpf.c
tools/lib/bpf/bpf.h
tools/lib/bpf/libbpf.c
tools/lib/bpf/libbpf.h
tools/lib/bpf/libbpf.map
tools/lib/bpf/libbpf_probes.c

index 4af8b08..a6bf19d 100644 (file)
@@ -173,6 +173,7 @@ enum bpf_prog_type {
        BPF_PROG_TYPE_CGROUP_SYSCTL,
        BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE,
        BPF_PROG_TYPE_CGROUP_SOCKOPT,
+       BPF_PROG_TYPE_TRACING,
 };
 
 enum bpf_attach_type {
@@ -199,6 +200,7 @@ enum bpf_attach_type {
        BPF_CGROUP_UDP6_RECVMSG,
        BPF_CGROUP_GETSOCKOPT,
        BPF_CGROUP_SETSOCKOPT,
+       BPF_TRACE_RAW_TP,
        __MAX_BPF_ATTACH_TYPE
 };
 
index 7904606..ca0d635 100644 (file)
@@ -228,9 +228,10 @@ int bpf_load_program_xattr(const struct bpf_load_program_attr *load_attr,
        memset(&attr, 0, sizeof(attr));
        attr.prog_type = load_attr->prog_type;
        attr.expected_attach_type = load_attr->expected_attach_type;
-       if (attr.prog_type == BPF_PROG_TYPE_RAW_TRACEPOINT)
-               /* expected_attach_type is ignored for tracing progs */
-               attr.attach_btf_id = attr.expected_attach_type;
+       if (attr.prog_type == BPF_PROG_TYPE_TRACING)
+               attr.attach_btf_id = load_attr->attach_btf_id;
+       else
+               attr.prog_ifindex = load_attr->prog_ifindex;
        attr.insn_cnt = (__u32)load_attr->insns_cnt;
        attr.insns = ptr_to_u64(load_attr->insns);
        attr.license = ptr_to_u64(load_attr->license);
@@ -245,7 +246,6 @@ int bpf_load_program_xattr(const struct bpf_load_program_attr *load_attr,
        }
 
        attr.kern_version = load_attr->kern_version;
-       attr.prog_ifindex = load_attr->prog_ifindex;
        attr.prog_btf_fd = load_attr->prog_btf_fd;
        attr.func_info_rec_size = load_attr->func_info_rec_size;
        attr.func_info_cnt = load_attr->func_info_cnt;
index 0db0133..1c53bc5 100644 (file)
@@ -78,7 +78,10 @@ struct bpf_load_program_attr {
        size_t insns_cnt;
        const char *license;
        __u32 kern_version;
-       __u32 prog_ifindex;
+       union {
+               __u32 prog_ifindex;
+               __u32 attach_btf_id;
+       };
        __u32 prog_btf_fd;
        __u32 func_info_rec_size;
        const void *func_info;
index 5d15cc4..c80f316 100644 (file)
@@ -188,6 +188,7 @@ struct bpf_program {
        bpf_program_clear_priv_t clear_priv;
 
        enum bpf_attach_type expected_attach_type;
+       __u32 attach_btf_id;
        void *func_info;
        __u32 func_info_rec_size;
        __u32 func_info_cnt;
@@ -3446,6 +3447,7 @@ load_program(struct bpf_program *prog, struct bpf_insn *insns, int insns_cnt,
        load_attr.line_info_cnt = prog->line_info_cnt;
        load_attr.log_level = prog->log_level;
        load_attr.prog_flags = prog->prog_flags;
+       load_attr.attach_btf_id = prog->attach_btf_id;
 
 retry_load:
        log_buf = malloc(log_buf_size);
@@ -3607,6 +3609,8 @@ bpf_object__load_progs(struct bpf_object *obj, int log_level)
        return 0;
 }
 
+static int libbpf_attach_btf_id_by_name(const char *name, __u32 *btf_id);
+
 static struct bpf_object *
 __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
                   struct bpf_object_open_opts *opts)
@@ -3656,6 +3660,7 @@ __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
        bpf_object__for_each_program(prog, obj) {
                enum bpf_prog_type prog_type;
                enum bpf_attach_type attach_type;
+               __u32 btf_id;
 
                err = libbpf_prog_type_by_name(prog->section_name, &prog_type,
                                               &attach_type);
@@ -3667,6 +3672,12 @@ __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
 
                bpf_program__set_type(prog, prog_type);
                bpf_program__set_expected_attach_type(prog, attach_type);
+               if (prog_type == BPF_PROG_TYPE_TRACING) {
+                       err = libbpf_attach_btf_id_by_name(prog->section_name, &btf_id);
+                       if (err)
+                               goto out;
+                       prog->attach_btf_id = btf_id;
+               }
        }
 
        return obj;
@@ -4518,6 +4529,7 @@ BPF_PROG_TYPE_FNS(tracepoint, BPF_PROG_TYPE_TRACEPOINT);
 BPF_PROG_TYPE_FNS(raw_tracepoint, BPF_PROG_TYPE_RAW_TRACEPOINT);
 BPF_PROG_TYPE_FNS(xdp, BPF_PROG_TYPE_XDP);
 BPF_PROG_TYPE_FNS(perf_event, BPF_PROG_TYPE_PERF_EVENT);
+BPF_PROG_TYPE_FNS(tracing, BPF_PROG_TYPE_TRACING);
 
 enum bpf_attach_type
 bpf_program__get_expected_attach_type(struct bpf_program *prog)
@@ -4546,7 +4558,8 @@ void bpf_program__set_expected_attach_type(struct bpf_program *prog,
        BPF_PROG_SEC_IMPL(string, ptype, eatype, 1, 0, eatype)
 
 /* Programs that use BTF to identify attach point */
-#define BPF_PROG_BTF(string, ptype) BPF_PROG_SEC_IMPL(string, ptype, 0, 0, 1, 0)
+#define BPF_PROG_BTF(string, ptype, eatype) \
+       BPF_PROG_SEC_IMPL(string, ptype, eatype, 0, 1, 0)
 
 /* Programs that can be attached but attach type can't be identified by section
  * name. Kept for backward compatibility.
@@ -4573,7 +4586,8 @@ static const struct {
        BPF_PROG_SEC("tp/",                     BPF_PROG_TYPE_TRACEPOINT),
        BPF_PROG_SEC("raw_tracepoint/",         BPF_PROG_TYPE_RAW_TRACEPOINT),
        BPF_PROG_SEC("raw_tp/",                 BPF_PROG_TYPE_RAW_TRACEPOINT),
-       BPF_PROG_BTF("tp_btf/",                 BPF_PROG_TYPE_RAW_TRACEPOINT),
+       BPF_PROG_BTF("tp_btf/",                 BPF_PROG_TYPE_TRACING,
+                                               BPF_TRACE_RAW_TP),
        BPF_PROG_SEC("xdp",                     BPF_PROG_TYPE_XDP),
        BPF_PROG_SEC("perf_event",              BPF_PROG_TYPE_PERF_EVENT),
        BPF_PROG_SEC("lwt_in",                  BPF_PROG_TYPE_LWT_IN),
@@ -4678,27 +4692,6 @@ int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
                        continue;
                *prog_type = section_names[i].prog_type;
                *expected_attach_type = section_names[i].expected_attach_type;
-               if (section_names[i].is_attach_btf) {
-                       struct btf *btf = bpf_core_find_kernel_btf();
-                       char raw_tp_btf_name[128] = "btf_trace_";
-                       char *dst = raw_tp_btf_name + sizeof("btf_trace_") - 1;
-                       int ret;
-
-                       if (IS_ERR(btf)) {
-                               pr_warn("vmlinux BTF is not found\n");
-                               return -EINVAL;
-                       }
-                       /* prepend "btf_trace_" prefix per kernel convention */
-                       strncat(dst, name + section_names[i].len,
-                               sizeof(raw_tp_btf_name) - sizeof("btf_trace_"));
-                       ret = btf__find_by_name(btf, raw_tp_btf_name);
-                       btf__free(btf);
-                       if (ret <= 0) {
-                               pr_warn("%s is not found in vmlinux BTF\n", dst);
-                               return -EINVAL;
-                       }
-                       *expected_attach_type = ret;
-               }
                return 0;
        }
        pr_warn("failed to guess program type based on ELF section name '%s'\n", name);
@@ -4711,6 +4704,46 @@ int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
        return -ESRCH;
 }
 
+#define BTF_PREFIX "btf_trace_"
+static int libbpf_attach_btf_id_by_name(const char *name, __u32 *btf_id)
+{
+       struct btf *btf = bpf_core_find_kernel_btf();
+       char raw_tp_btf_name[128] = BTF_PREFIX;
+       char *dst = raw_tp_btf_name + sizeof(BTF_PREFIX) - 1;
+       int ret, i, err = -EINVAL;
+
+       if (IS_ERR(btf)) {
+               pr_warn("vmlinux BTF is not found\n");
+               return -EINVAL;
+       }
+
+       if (!name)
+               goto out;
+
+       for (i = 0; i < ARRAY_SIZE(section_names); i++) {
+               if (!section_names[i].is_attach_btf)
+                       continue;
+               if (strncmp(name, section_names[i].sec, section_names[i].len))
+                       continue;
+               /* prepend "btf_trace_" prefix per kernel convention */
+               strncat(dst, name + section_names[i].len,
+                       sizeof(raw_tp_btf_name) - sizeof(BTF_PREFIX));
+               ret = btf__find_by_name(btf, raw_tp_btf_name);
+               if (ret <= 0) {
+                       pr_warn("%s is not found in vmlinux BTF\n", dst);
+                       goto out;
+               }
+               *btf_id = ret;
+               err = 0;
+               goto out;
+       }
+       pr_warn("failed to identify btf_id based on ELF section name '%s'\n", name);
+       err = -ESRCH;
+out:
+       btf__free(btf);
+       return err;
+}
+
 int libbpf_attach_type_by_name(const char *name,
                               enum bpf_attach_type *attach_type)
 {
index c63e2ff..2b126ee 100644 (file)
@@ -307,6 +307,7 @@ LIBBPF_API int bpf_program__set_sched_cls(struct bpf_program *prog);
 LIBBPF_API int bpf_program__set_sched_act(struct bpf_program *prog);
 LIBBPF_API int bpf_program__set_xdp(struct bpf_program *prog);
 LIBBPF_API int bpf_program__set_perf_event(struct bpf_program *prog);
+LIBBPF_API int bpf_program__set_tracing(struct bpf_program *prog);
 
 LIBBPF_API enum bpf_prog_type bpf_program__get_type(struct bpf_program *prog);
 LIBBPF_API void bpf_program__set_type(struct bpf_program *prog,
@@ -326,6 +327,7 @@ LIBBPF_API bool bpf_program__is_sched_cls(const struct bpf_program *prog);
 LIBBPF_API bool bpf_program__is_sched_act(const struct bpf_program *prog);
 LIBBPF_API bool bpf_program__is_xdp(const struct bpf_program *prog);
 LIBBPF_API bool bpf_program__is_perf_event(const struct bpf_program *prog);
+LIBBPF_API bool bpf_program__is_tracing(const struct bpf_program *prog);
 
 /*
  * No need for __attribute__((packed)), all members of 'bpf_map_def'
index d1473ea..69dded5 100644 (file)
@@ -197,4 +197,6 @@ LIBBPF_0.0.6 {
                bpf_object__open_mem;
                bpf_program__get_expected_attach_type;
                bpf_program__get_type;
+               bpf_program__is_tracing;
+               bpf_program__set_tracing;
 } LIBBPF_0.0.5;
index 4b0b036..a9eb8b3 100644 (file)
@@ -102,6 +102,7 @@ probe_load(enum bpf_prog_type prog_type, const struct bpf_insn *insns,
        case BPF_PROG_TYPE_FLOW_DISSECTOR:
        case BPF_PROG_TYPE_CGROUP_SYSCTL:
        case BPF_PROG_TYPE_CGROUP_SOCKOPT:
+       case BPF_PROG_TYPE_TRACING:
        default:
                break;
        }