2023-02-13 Dmitry Butskoy - 2.1.2 * Fix unprivileged ICMP tracerouting with Linux kernel >= 6.1 (Eric Dumazet, SF bug #14) 2022-12-27 Dmitry Butskoy - 2.1.1 * Interpret ipv4-mapped ipv6 addresses (::ffff:A.B.C.D) as true ipv4. There are no ipv4-mapped addresses in the real network which we operate on, so use just ipv4 in such cases, but allow users to specify it this way for convenience. * Return back more robast poll(2) loop handling. 2016-03-08 Dmitry Butskoy - 2.1.0 * Improve the main loop for better interactivity. Instead of waiting silently for maximum expiration time of probes in progress, use timeout of the first probe (which will be printed first from now) only. * Speedup wait mechanism. Traditional traceroute implementation always waited the whole timeout for any probe. But if we already have some replies from the same hop, or even from some next hop, we can use the round trip time of such a reply as a hint to determine the actual reasonable amount of time to wait. Now the `-w' option has a form of three (in general) float values separated by a comma (or a slash): `-w MAX_SECS,HERE,NEAR' . (last two are optional). MAX_SECS specifies the maximum time (in seconds) to wait, in any case. The optional HERE specifies a factor to multiply the round trip time of an already received response from the same hop. The resulting value is used as a timeout for the probe, instead of (but no more than) MAX_SECS. The optional NEAR specifies a similar factor for a response from some next hop. The time of the first found result is used in both cases. First, we look for the same hop (of the probe which will be printed first from now). If nothing found, then look for some next hop. If nothing found, use MAX_SECS. If HERE and/or NEAR have zero values, the corresponding computation is skipped. HERE and NEAR are always set to zero if only MAX_SECS is specified (which provides compatibility with previous versions). Thus, if your scripts use `-w SECS', then nothing changed for you, since the lonely SECS implies `-w SECS,0,0' . Defaults are 5.0 seconds for MAX_SECS, 3.0 times for HERE and 10.0 times for NEAR. Certainly, the new algorithm can lead to premature expiry (especially when response times differ at times) and printing "*" instead of a time. Anyway, you can always switch this algorithm off, just by specifying `-w' with the desired timeout only (fe. `-w 5'). We continue to wait whole MAX_SECS when one probe per time must be sent (`--sport', `-P proto'), because it seems more harmful rather than helpful to try to wait less in such cases. To provide compatibility with 2.0.x versions, use: traceroute -w 5 (or any other desired `-w' value). * Hint people to use the system traceroute(8) instead of tcptraceroute wrapper (by providing a stderr header). The using of this wrapper is a little bit harmful, since it has less possibilities and a little different set of options. For those who are used to use tcptraceroute in cmdline, just create a link with that name to the system traceroute. When invoked as "tcp*", it then behaves as `traceroute -T'. (The simple manual page added for this case in the wrapper subdir). The original tcptraceroute had some options differ ("lpNSAE"), but they was rare used. Most common "dnFifmqwst" was just the same. Therefore it should be painless to use the system binary directly, instead of the limited wrapper (which is still provided indeed). 2016-02-15 Dmitry Butskoy - 2.0.22 * Some portability fixing and improvements (Felix Janda) * Require clear numbers for options and arguments (Sergey Salnikov) * Drop compilation date from the version string (Debian #774365) * New tcp module option `reuse', which utilize SO_REUSEADDR to reuse local port numbers for the huge workloads (Richard Sheehan) * Avoid poll(2) call with spurious zero timeout in some rare cases by rounding the value properly using ceil(3) 2014-11-12 Dmitry Butskoy - 2.0.21 * Fix `--mtu' and `-F' working on kernels >= 3.13 * Some manual page improving (Christopher Mann) 2014-06-14 Dmitry Butskoy - 2.0.20 * Describe all complementary long options in the man page (Jan Synacek) * Use correct service name for AS lookups (Frederic Mangano) * Avoid some rare case null dereference (geogriffin@jsgriff.com) * Improve expiration check for simultaneous probes 2012-11-19 Dmitry Butskoy - 2.0.19 * DCCP protocol support (rfc4340), by Samuel Jero Use "-D" option for it (the protocol-specific options are available too). * Update COPYING and COPYING.LIB license files to the latest published ones (due to FSF address changes etc.) (Jan Synacek) * Add mention of "-l" option to manual (Filip Holec) 2011-08-16 Dmitry Butskoy - 2.0.18 * Handle new dgram icmp sockets ("echo ping sockets"), appeared in kernel 3.0 . Now unprivileged users may perform ICMP tracerouting without any special rights of the executable (neither setuid bits nor cap_net_raw settings). It is allowed if any group of a user matches sysctl range of "net/ipv4/ping_group_range". The support for dgram icmp way (and whether it is allowed) is auto-detected at runtime. First, the traditional raw socket is tried (for full compatibility reasons), then new dgram socket as a fallback. The icmp module now has two additional options "raw" and "dgram", which cause to try one particular way only. Note, that there is no IPv6 implementation for dgram icmp sockets in kernels 3.0 yet, but new traceroute is ready for it anyway. * New tcp module option `info' ("-T -O info"), which prints all tcp flags of tcp reply from the reached target host. The flags are shown comma-separated in the same place where icmp extensions is printed (ie. in `<>' brackets) This feature is utilized by tcptraceroute wrapper now, and allow it to be completely functional replacement of the original tcptraceroute. * Fix determination of system-wide ECN setings for tcp module. Since the kernel 2.6.31 the default sysctl net/ipv4/tcp_ecn was changed from zero to '2', whereas the actual value for ecn to be set is still '1' * Allow different packet sizes for `--mtu'. Suport `-l' option for tracepath wrapper. * Some code and manual cleanups 2010-12-14 Dmitry Butskoy - 2.0.17 * Adapt code to make possible the use of Linux capabilities (for raw sockets etc.) instead of superuser privileges only. On modern systems the capabilities can be stored as file attributes, ie.: "setcap cap_net_raw=pe /usr/bin/traceroute" 2010-09-13 Dmitry Butskoy - 2.0.16 * A little work-around in the build system for the new (buggy?) make 3.82 * Add `--fwmark=num' option for firewall mark (for kernel >= 2.6.25). Idea comes from an anonymous SF patch #3042539 2010-07-14 Dmitry Butskoy - 2.0.15 * Use string routines more safely (fix SF bug #3029216) * Provide help for lft wrapper 2010-04-21 Dmitry Butskoy - 2.0.14 * Fix support for IPv6's flow_labels and tclass. Thanks to Peter Bieringer for testing * Use route header "type 2" instead of deprecated "type 0" for `-g' option for IPv6. The default value can be changed by specifying a number in the place of the first `-g' address. 2009-11-02 Dmitry Butskoy - 2.0.13 * Check for first_hop is not zero value (vladz@devzero.fr) * Always fill unresolved IP address by its numeric interpretation, even if getnameinfo(3) leaves it untouched (as it does for ipv6 in some glibc versions, whereas always fills for ipv4) * Cosmetic changes for man page (Andreas Mohr) 2008-09-15 Dmitry Butskoy - 2.0.12 * Use common recv_reply() routine for all modules which do recvmsg(2) call. Method-specific things go to callbacks. Pass to init methods pointer to datalen instead of the value. * Implement ICMP Extension support (rfc4884), `-e' option. Parse MPLS info (rfc4950) to be more readable (Kaj Niemi) * Implement Path MTU Discovery (similar to tracepath(1)), with `--mtu' option. Changed mtu is printed once in a form of `F=NUM' at the first probe of a hop which requires such mtu to be reached. (Actually, the correspond "frag needed" icmp message is normally sent by the previous hop). * Print the number of backward hops when it differs with forward, by `--back' option. The backward hops is guessed by a technique similar to tracepath(1), there is no reliable way to obtain such info though. * The optional second argument (packet_len) now is the full length of the packet, including IP headers. (It is obvious enough due to the nature of this feature, and this is the behaviour of the original traceroute). Particular trace methods can ignore this (fe. tcp), or increase it up to the minimal value (udp, icmp). The actual packet's size is alvays reported in the output header. * Add tracepath(1)/tracepath6(1) shell wrapper. * Allow DEF_AF to be redefined at cmdline (Teran McKinney) * Do not check the correctness of `sim_probes' value -- it is unneeded at all. This also fixes a bug when a value of sim_probes appears to be more than the total number of probes. Reported by Milos Malik. * Allow default UDP method to cross zero port boundary (Milos Malik). It is a strange corner case, but traditional traceroute behaves exactly so. 2008-04-25 Dmitry Butskoy - 2.0.11 * Use new pmtudisc value "probe" instead of "do" for `-F' option (available since the kernel 2.6.22). For kernels before 2.6.22, the `-F' (dontfragment) option seems completely useless for IPv6 and partially useful for IPv4 (when a user can flush routing caches some way). * Fix installation in build system (Mike Frysinger) * Don't compute checksum for ipv6 icmp packets ourselves, the kernel overwrites it anyway by the proper values. * Don't use explicit path to traceroute in wrapper scripts 2008-04-17 Dmitry Butskoy - 2.0.10 * raw_can_connect(): ipv6 connected raw sockets receive MSG_ERRQUEUE properly only for kernels >= 2.6.25 * remove useless "host" parameter for init methods * add probe_by_seq() and probe_by_sk() routines, don't pass whole probes' pointer to recv_probe method * collect all sends in do_send() routine * Interpret ENOBUFS errors for send(2) as "can retry later". Slow devices (like ppp) with small tx_queue_len can reject the sending of too many packets simultaneously. To handle this, do_send() now returns a negate value in a case of ENOBUFS and similar (instead of program exit). The send_probe method clears the probe and returns immediately in such cases. Then, if there is an amount of time to wait for something, the failed probe will be attempted again after that time expired. If nothing to wait more, the program is exited. 2007-09-26 Dmitry Butskoy - 2.0.9 * Complete manual page. * Edit manual page to sound more English, thanks to Chris Ward 2007-09-04 Dmitry Butskoy - 2.0.8 * Move all wrappers to special "wrappers/" dir. Add lft(8) shell wrapper. Add traceproto(8) shell wrapper. Add traceroute-nanog(8) shell wrapper. * Interpret first_hop as number, not index * Build system is re-worked to match more the modern requirements (Thanks to Mike Frysinger for testing). * Check for kernel version >= 2.6.22.2 in raw_can_connect() * Add generic "raw" method, "-P protonum" option. New "one_per_time" flag for tr_module. 2007-07-31 Dmitry Butskoy - 2.0.7 * Fix revents checking typo * Expect normal data reply from udp too. * Implement udp to port (-U) and udplite (-UL) methods. Both available for unprivileged users. Add "coverage" option for udplite. * Allow non-digit service names for `-p' and `--sport' * Drop period at the end of "SEE ALSO" section, and avoid specific distro names in the manual (Mike Frysinger) * Explicitly mention that this program is licensed as "GPL version 2 or any later version". (Similar for libsupp subdir: LGPL version 2.1 or any later). * Always check whether the dest and source port match in received packets. Can decrease an amount of (hypothetical) garbage received just after the bind() but before connect() 2007-07-19 Dmitry Butskoy - 2.0.6 * Rename tr_ops to tr_module * Implement module-specific options (-O opt,...) * Add TCP specific options (all the tcp header flags, ecn, sack, timestamps, window_scaling, mss, sysctl) Build tcp probe packet depending on them. * Add "--sport" option for explicit source port selection. Always cause "-N 1" when it is set. * Add new routine bind_socket(). Always (auto)bind sockets in tune_socket(). * Add tcptraceroute(8) shell wrapper 2007-07-16 Dmitry Butskoy - 2.0.5 * Use MSG_ERRQUEUE for raw sockets too. * raw_can_connect () work-around for kernel bug #8747 * random.c, csum.c: new separate files * New implementation of tcp method ("-T"), using half-open technique. The old implementation module renamed to "tcpconn" ("-M tcpconn"). * Common parse_cmsg() routine * put ee_info for parse_icmp_res() too, handle ICMP6_PACKET_TOO_BIG for IPv6, report "!F-num" when "frag needed" (legacy compatibility) 2007-07-11 Dmitry Butskoy - 2.0.4 * clear includes of unneeded headers * move poll stuff to separate poll.c * add module stuff (module.c), options etc. Adapt udp/icmp/tcp for this. * Add common routines use_recverr() and set_ttl() 2007-02-28 Dmitry Butskoy * fix variable type for getsockname (Mike Frysinger) 2007-01-09 Dmitry Butskoy - 2.0.3 * version 2.0.3 * allow option args without separator (add CLIF_MAY_JOIN_ARG flag), for compatibility (Benjamin LaHaise) * no more "tcptraceroute" symlink for rpm packages, because it conflicts with the same-name old package anyway (James Ralston) * fix compilation on glibc < 2.4 (Andy Shevchenko) 2006-10-30 Dmitry Butskoy - 2.0.2 * version 2.0.2 * More accurate check_expired() routine. * Some minor fixes. * Add NOTES section to manual 2006-10-20 Dmitry Butskoy - 2.0.1 * version 2.0.1 * Now ops methods write send_time (as well as recv_time) * Use SO_TIMESTAMP to obtain msecs precisely * Complete manual