libbpf-tools: add statsnoop

Signed-off-by: Hengqi Chen <chenhengqi@outlook.com>

libbpf-tools/.gitignore

@@ -23,6 +23,7 @@
 /runqlen
 /runqslower
 /softirqs
+/statsnoop
 /syscount
 /tcpconnect
 /tcpconnlat

libbpf-tools/Makefile

@@ -40,6 +40,7 @@ APPS = \
 	runqlen \
 	runqslower \
 	softirqs \
+	statsnoop \
 	syscount \
 	tcpconnect \
 	tcpconnlat \

libbpf-tools/statsnoop.bpf.c

+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2021 Hengqi Chen
+#include <vmlinux.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include "statsnoop.h"
+
+#define MAX_ENTRIES 10240
+
+const volatile pid_t target_pid = 0;
+const volatile bool  trace_failed_only = false;
+
+struct {
+	__uint(type, BPF_MAP_TYPE_HASH);
+	__uint(max_entries, MAX_ENTRIES);
+	__type(key, __u32);
+	__type(value, const char *);
+} values SEC(".maps");
+
+struct {
+	__uint(type, BPF_MAP_TYPE_PERF_EVENT_ARRAY);
+	__uint(key_size, sizeof(__u32));
+	__uint(value_size, sizeof(__u32));
+} events SEC(".maps");
+
+static int probe_entry(void *ctx, const char *pathname)
+{
+	__u64 id = bpf_get_current_pid_tgid();
+	__u32 pid = id >> 32;
+	__u32 tid = (__u32)id;
+
+	if (!pathname)
+		return 0;
+
+	if (target_pid && target_pid != pid)
+		return 0;
+
+	bpf_map_update_elem(&values, &tid, &pathname, BPF_ANY);
+	return 0;
+};
+
+static int probe_return(void *ctx, int ret)
+{
+	__u64 id = bpf_get_current_pid_tgid();
+	__u32 pid = id >> 32;
+	__u32 tid = (__u32)id;
+	const char **pathname;
+	struct event event = {};
+
+	pathname = bpf_map_lookup_elem(&values, &tid);
+	if (!pathname)
+		return 0;
+
+	if (trace_failed_only && ret >= 0) {
+		bpf_map_delete_elem(&values, &tid);
+		return 0;
+	}
+
+	event.pid = pid;
+	event.ts_ns = bpf_ktime_get_ns();
+	event.ret = ret;
+	bpf_get_current_comm(&event.comm, sizeof(event.comm));
+	bpf_probe_read_user_str(event.pathname, sizeof(event.pathname), *pathname);
+
+	bpf_perf_event_output(ctx, &events, BPF_F_CURRENT_CPU, &event, sizeof(event));
+	bpf_map_delete_elem(&values, &tid);
+	return 0;
+}
+
+SEC("tracepoint/syscalls/sys_enter_statfs")
+int handle_statfs_entry(struct trace_event_raw_sys_enter *ctx)
+{
+	return probe_entry(ctx, (const char *)ctx->args[0]);
+}
+
+SEC("tracepoint/syscalls/sys_exit_statfs")
+int handle_statfs_return(struct trace_event_raw_sys_exit *ctx)
+{
+	return probe_return(ctx, (int)ctx->ret);
+}
+
+SEC("tracepoint/syscalls/sys_enter_newstat")
+int handle_newstat_entry(struct trace_event_raw_sys_enter *ctx)
+{
+	return probe_entry(ctx, (const char *)ctx->args[0]);
+}
+
+SEC("tracepoint/syscalls/sys_exit_newstat")
+int handle_newstat_return(struct trace_event_raw_sys_exit *ctx)
+{
+	return probe_return(ctx, (int)ctx->ret);
+}
+
+char LICENSE[] SEC("license") = "GPL";

libbpf-tools/statsnoop.c

+// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
+// Copyright (c) 2021 Hengqi Chen
+//
+// Based on statsnoop(8) from BCC by Brendan Gregg.
+// 09-May-2021   Hengqi Chen   Created this.
+#include <argp.h>
+#include <errno.h>
+#include <signal.h>
+#include <time.h>
+
+#include <bpf/libbpf.h>
+#include <bpf/bpf.h>
+#include "statsnoop.h"
+#include "statsnoop.skel.h"
+#include "trace_helpers.h"
+
+#define PERF_BUFFER_PAGES       16
+#define PERF_POLL_TIMEOUT_MS    100
+#define warn(...) fprintf(stderr, __VA_ARGS__)
+
+static volatile sig_atomic_t exiting = 0;
+
+static pid_t target_pid = 0;
+static bool trace_failed_only = false;
+static bool emit_timestamp = false;
+
+const char *argp_program_version = "statsnoop 0.1";
+const char *argp_program_bug_address =
+	"https://github.com/iovisor/bcc/tree/master/libbpf-tools";
+const char argp_program_doc[] =
+"Trace stat syscalls.\n"
+"\n"
+"USAGE: statsnoop [-h] [-t] [-x] [-p PID]\n"
+"\n"
+"EXAMPLES:\n"
+"    statsnoop             # trace all stat syscalls\n"
+"    statsnoop -t          # include timestamps\n"
+"    statsnoop -x          # only show failed stats\n"
+"    statsnoop -p 1216     # only trace PID 1216\n";
+
+static const struct argp_option opts[] = {
+	{"pid", 'p', "PID", 0, "Process ID to trace"},
+	{"failed", 'x', NULL, 0, "Only show failed stats"},
+	{"timestamp", 't', NULL, 0, "Include timestamp on output"},
+	{NULL, 'h', NULL, OPTION_HIDDEN, "Show the full help"},
+	{},
+};
+
+static error_t parse_arg(int key, char *arg, struct argp_state *state)
+{
+	long pid;
+
+	switch (key) {
+	case 'p':
+		errno = 0;
+		pid = strtol(arg, NULL, 10);
+		if (errno || pid <= 0) {
+			warn("Invalid PID: %s\n", arg);
+			argp_usage(state);
+		}
+		target_pid = pid;
+		break;
+	case 'x':
+		trace_failed_only = true;
+		break;
+	case 't':
+		emit_timestamp = true;
+		break;
+	case 'h':
+		argp_state_help(state, stderr, ARGP_HELP_STD_HELP);
+		break;
+	default:
+		return ARGP_ERR_UNKNOWN;
+	}
+	return 0;
+}
+
+static void sig_int(int signo)
+{
+	exiting = 1;
+}
+
+static void handle_event(void *ctx, int cpu, void *data, __u32 data_sz)
+{
+	static __u64 start_timestamp = 0;
+	const struct event *e = data;
+	int fd, err;
+	double ts = 0.0;
+
+	if (e->ret >= 0) {
+		fd = e->ret;
+		err = 0;
+	} else {
+		fd = -1;
+		err = -e->ret;
+	}
+	if (!start_timestamp)
+		start_timestamp = e->ts_ns;
+	if (emit_timestamp) {
+		ts = (double)(e->ts_ns - start_timestamp) / 1000000000;
+		printf("%-14.9f ", ts);
+	}
+	printf("%-7d %-20s %-4d %-4d %-s\n", e->pid, e->comm, fd, err, e->pathname);
+}
+
+static void handle_lost_events(void *ctx, int cpu, __u64 lost_cnt)
+{
+	warn("lost %llu events on CPU #%d\n", lost_cnt, cpu);
+}
+
+int main(int argc, char **argv)
+{
+	static const struct argp argp = {
+		.options = opts,
+		.parser = parse_arg,
+		.doc = argp_program_doc,
+	};
+	struct perf_buffer_opts pb_opts;
+	struct perf_buffer *pb = NULL;
+	struct statsnoop_bpf *obj;
+	int err;
+
+	err = argp_parse(&argp, argc, argv, 0, NULL, NULL);
+	if (err)
+		return err;
+
+	err = bump_memlock_rlimit();
+	if (err) {
+		warn("failed to increase rlimit: %d\n", err);
+		return 1;
+	}
+
+	obj = statsnoop_bpf__open();
+	if (!obj) {
+		warn("failed to open BPF object\n");
+		return 1;
+	}
+
+	obj->rodata->target_pid = target_pid;
+	obj->rodata->trace_failed_only = trace_failed_only;
+
+	err = statsnoop_bpf__load(obj);
+	if (err) {
+		warn("failed to load BPF object: %d\n", err);
+		goto cleanup;
+	}
+
+	err = statsnoop_bpf__attach(obj);
+	if (err) {
+		warn("failed to attach BPF programs: %d\n", err);
+		goto cleanup;
+	}
+
+	pb_opts.sample_cb = handle_event;
+	pb_opts.lost_cb = handle_lost_events;
+	pb = perf_buffer__new(bpf_map__fd(obj->maps.events), PERF_BUFFER_PAGES,
+			      &pb_opts);
+	err = libbpf_get_error(pb);
+	if (err) {
+		warn("failed to open perf buffer: %d\n", err);
+		goto cleanup;
+	}
+
+	if (signal(SIGINT, sig_int) == SIG_ERR) {
+		warn("can't set signal handler: %s\n", strerror(-errno));
+		goto cleanup;
+	}
+
+	if (emit_timestamp)
+		printf("%-14s ", "TIME(s)");
+	printf("%-7s %-20s %-4s %-4s %-s\n",
+	       "PID", "COMM", "RET", "ERR", "PATH");
+
+	while (1) {
+		if ((err = perf_buffer__poll(pb, PERF_POLL_TIMEOUT_MS)) < 0)
+			break;
+		if (exiting)
+			goto cleanup;
+	}
+	warn("error polling perf buffer: %d\n", err);
+
+cleanup:
+	statsnoop_bpf__destroy(obj);
+
+	return err != 0;
+}

libbpf-tools/statsnoop.h

+/* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
+#ifndef __STATSNOOP_H
+#define __STATSNOOP_H
+
+#define TASK_COMM_LEN	16
+#define NAME_MAX	255
+
+struct event {
+	__u64 ts_ns;
+	__u32 pid;
+	int ret;
+	char comm[TASK_COMM_LEN];
+	char pathname[NAME_MAX];
+};
+
+#endif /* __STATSNOOP_H */