perf trace: Add BPF augmenter to perf_event_open()'s 'struct perf_event_attr' arg

Using BPF for that, doing a cleverish reuse of perf_event_attr__fprintf(),
that really needs to be turned into __snprintf(), etc.

But since the plan is to go the BTF way probably use libbpf's
btf_dump__dump_type_data().

Example:

[root@quaco ~]# perf trace -e ~acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.c,perf_event_open --max-events 10 perf stat --quiet sleep 0.001
fg
     0.000 perf_event_open(attr_uptr: { type: 1, size: 128, config: 0x1, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 3
     0.067 perf_event_open(attr_uptr: { type: 1, size: 128, config: 0x3, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
     0.120 perf_event_open(attr_uptr: { type: 1, size: 128, config: 0x4, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 5
     0.172 perf_event_open(attr_uptr: { type: 1, size: 128, config: 0x2, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 7
     0.190 perf_event_open(attr_uptr: { size: 128, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 8
     0.199 perf_event_open(attr_uptr: { size: 128, config: 0x1, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 9
     0.204 perf_event_open(attr_uptr: { size: 128, config: 0x4, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 10
     0.210 perf_event_open(attr_uptr: { size: 128, config: 0x5, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258859 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 11
[root@quaco ~]#
Suggested-by: Ian Rogers <irogers@google.com>
Tested-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/Y2V2Tpu+2vzJyon2@kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/perf/builtin-trace.c

@@ -1053,7 +1053,8 @@ static struct syscall_fmt syscall_fmts[] = {
 	  .arg = { [0] = { .scnprintf = SCA_FDAT,	/* dfd */ },
 		   [2] = { .scnprintf = SCA_OPEN_FLAGS, /* flags */ }, }, },
 	{ .name	    = "perf_event_open",
-	  .arg = { [2] = { .scnprintf = SCA_INT,	/* cpu */ },
+	  .arg = { [0] = { .scnprintf = SCA_PERF_ATTR,  /* attr */ },
+		   [2] = { .scnprintf = SCA_INT,	/* cpu */ },
 		   [3] = { .scnprintf = SCA_FD,		/* group_fd */ },
 		   [4] = { .scnprintf = SCA_PERF_FLAGS, /* flags */ }, }, },
 	{ .name	    = "pipe2",

tools/perf/examples/bpf/augmented_raw_syscalls.c

@@ -129,6 +129,7 @@ struct augmented_args_payload {
 			struct augmented_arg arg, arg2;
 		};
 		struct sockaddr_storage saddr;
+		char   __data[sizeof(struct augmented_arg)];
 	};
 };
 
@@ -293,6 +294,49 @@ int sys_enter_renameat(struct syscall_enter_args *args)
 	return augmented__output(args, augmented_args, len);
 }
 
+#define PERF_ATTR_SIZE_VER0     64      /* sizeof first published struct */
+
+// we need just the start, get the size to then copy it
+struct perf_event_attr_size {
+        __u32                   type;
+        /*
+         * Size of the attr structure, for fwd/bwd compat.
+         */
+        __u32                   size;
+};
+
+SEC("!syscalls:sys_enter_perf_event_open")
+int sys_enter_perf_event_open(struct syscall_enter_args *args)
+{
+	struct augmented_args_payload *augmented_args = augmented_args_payload();
+	const struct perf_event_attr_size *attr = (const struct perf_event_attr_size *)args->args[0], *attr_read;
+	unsigned int len = sizeof(augmented_args->args);
+
+        if (augmented_args == NULL)
+		goto failure;
+
+	if (bpf_probe_read(&augmented_args->__data, sizeof(*attr), attr) < 0)
+		goto failure;
+
+	attr_read = (const struct perf_event_attr_size *)augmented_args->__data;
+
+	__u32 size = attr_read->size;
+
+	if (!size)
+		size = PERF_ATTR_SIZE_VER0;
+
+	if (size > sizeof(augmented_args->__data))
+                goto failure;
+
+	// Now that we read attr->size and tested it against the size limits, read it completely
+	if (bpf_probe_read(&augmented_args->__data, size, attr) < 0)
+		goto failure;
+
+	return augmented__output(args, augmented_args, len + size);
+failure:
+	return 1; /* Failure: don't filter */
+}
+
 static pid_t getpid(void)
 {
 	return bpf_get_current_pid_tgid();

tools/perf/trace/beauty/perf_event_open.c

@@ -44,3 +44,47 @@ static size_t syscall_arg__scnprintf_perf_flags(char *bf, size_t size,
 }
 
 #define SCA_PERF_FLAGS syscall_arg__scnprintf_perf_flags
+
+struct attr_fprintf_args {
+	size_t size, printed;
+	char *bf;
+	bool first;
+};
+
+static int attr__fprintf(FILE *fp __maybe_unused, const char *name, const char *val, void *priv)
+{
+	struct attr_fprintf_args *args = priv;
+	size_t printed = scnprintf(args->bf + args->printed , args->size - args->printed, "%s%s: %s", args->first ? "" : ", ", name, val);
+
+	args->first = false;
+	args->printed += printed;
+	return printed;
+}
+
+static size_t perf_event_attr___scnprintf(struct perf_event_attr *attr, char *bf, size_t size, bool show_zeros __maybe_unused)
+{
+	struct attr_fprintf_args args = {
+		.printed = scnprintf(bf, size, "{ "),
+		.size    = size,
+		.first   = true,
+		.bf	 = bf,
+	};
+
+	perf_event_attr__fprintf(stdout, attr, attr__fprintf, &args);
+	return args.printed + scnprintf(bf + args.printed, size - args.printed, " }");
+}
+
+static size_t syscall_arg__scnprintf_augmented_perf_event_attr(struct syscall_arg *arg, char *bf, size_t size)
+{
+	return perf_event_attr___scnprintf((void *)arg->augmented.args, bf, size, arg->trace->show_zeros);
+}
+
+static size_t syscall_arg__scnprintf_perf_event_attr(char *bf, size_t size, struct syscall_arg *arg)
+{
+	if (arg->augmented.args)
+		return syscall_arg__scnprintf_augmented_perf_event_attr(arg, bf, size);
+
+	return scnprintf(bf, size, "%#lx", arg->val);
+}
+
+#define SCA_PERF_ATTR syscall_arg__scnprintf_perf_event_attr