#!/usr/bin/env bpftrace
/*
 * tcplife - Trace TCP session lifespans with connection details.
 *
 * See BPF Performance Tools, Chapter 10, for an explanation of this tool.
 *
 * Copyright (c) 2019 Brendan Gregg.
 * Licensed under the Apache License, Version 2.0 (the "License").
 * This was originally created for the BPF Performance Tools book
 * published by Addison Wesley. ISBN-13: 9780136554820
 * When copying or porting, include this comment.
 *
 * 17-Apr-2019  Brendan Gregg   Created this.
 */

#ifndef BPFTRACE_HAVE_BTF
#include <net/tcp_states.h>
#include <net/sock.h>
#include <linux/socket.h>
#include <linux/tcp.h>
#else
#include <sys/socket.h>
#endif

BEGIN
{
	printf("%-5s %-10s %-15s %-5s %-15s %-5s ", "PID", "COMM",
	    "LADDR", "LPORT", "RADDR", "RPORT");
	printf("%5s %5s %s\n", "TX_KB", "RX_KB", "MS");
}

kprobe:tcp_set_state
{
	$sk = (struct sock *)arg0;
	$newstate = arg1;

	/*
	 * This tool includes PID and comm context. From TCP this is best
	 * effort, and may be wrong in some situations. It does this:
	 * - record timestamp on any state < TCP_FIN_WAIT1
	 *	note some state transitions may not be present via this kprobe
	 * - cache task context on:
	 *	TCP_SYN_SENT: tracing from client
	 *	TCP_LAST_ACK: client-closed from server
	 * - do output on TCP_CLOSE:
	 *	fetch task context if cached, or use current task
	 */

	// record first timestamp seen for this socket
	if ($newstate < TCP_FIN_WAIT1 && @birth[$sk] == 0) {
		@birth[$sk] = nsecs;
	}

	// record PID & comm on SYN_SENT
	if ($newstate == TCP_SYN_SENT || $newstate == TCP_LAST_ACK) {
		@skpid[$sk] = pid;
		@skcomm[$sk] = comm;
	}

	// session ended: calculate lifespan and print
	if ($newstate == TCP_CLOSE && @birth[$sk]) {
		$delta_ms = (nsecs - @birth[$sk]) / 1e6;
		$lport = $sk->__sk_common.skc_num;
		$dport = $sk->__sk_common.skc_dport;
		$dport = bswap($dport);
		$tp = (struct tcp_sock *)$sk;
		$pid = @skpid[$sk];
		$comm = @skcomm[$sk];
		if ($comm == "") {
			// not cached, use current task
			$pid = pid;
			$comm = comm;
		}

		$family = $sk->__sk_common.skc_family;
		$saddr = ntop(0);
		$daddr = ntop(0);
		if ($family == AF_INET) {
			$saddr = ntop(AF_INET, $sk->__sk_common.skc_rcv_saddr);
			$daddr = ntop(AF_INET, $sk->__sk_common.skc_daddr);
		} else {
			// AF_INET6
			$saddr = ntop(AF_INET6,
			    $sk->__sk_common.skc_v6_rcv_saddr.in6_u.u6_addr8);
			$daddr = ntop(AF_INET6,
			    $sk->__sk_common.skc_v6_daddr.in6_u.u6_addr8);
		}
		printf("%-5d %-10.10s %-15s %-5d %-15s %-6d ", $pid,
		    $comm, $saddr, $lport, $daddr, $dport);
		printf("%5d %5d %d\n", $tp->bytes_acked / 1024,
		    $tp->bytes_received / 1024, $delta_ms);

		delete(@birth[$sk]);
		delete(@skpid[$sk]);
		delete(@skcomm[$sk]);
	}
}

END
{
	clear(@birth); clear(@skpid); clear(@skcomm);
}