[s20071127] John Heffner (6): [iputils] tracepath: Add length flag to set initial MTU. [iputils] tracepath: Add documentation for the -l flag. [iputils] tracepath: Use PMTUDISC_PROBE mode if it exists. [iputils] tracepath: Document -n flag. [iputils] tracepath: Fix asymm messages. [iputils] tracepath: Re-probe at same TTL after MTU reduction. YOSHIFUJI Hideaki (8): [DOC]: Delete duplicated lines in RELNOTES. Fix white space errors. [CLOCKDIFF,PING,RDISC,TRACEROUTE6]: Support uClibc. [RARPD]: Fixed several signedness issues for char strings. [PING]: Use inet_pton() instead of sscan(). [PING6]: Use IN6_IS_ADDR_xxx() macro. [MAKEFILE]: Change authorized host to push snapshots. [MAKEFILE]: Use git-archive instead of git-tar-tree. [s20070202] Mike Frysinger (2): Use socklen_t in all the right places. [IPG]: handle pktgen setup in newer kernels. Mitsuru Chinen (2): [CLOCKDIFF]: Fix compilation errors about labels at end of compound statements. [PING6]: Use getaddrinfo() for the name resolution of intermediate nodes. YOSHIFUJI Hideaki (9): [MAKEFILE] Remove unused -I../include [TRACEPATH] Print usage if we met incorrect option. [PING6]: Fix compilation error with glibc-2.4 and later. [PING6]: Use getaddrinfo() to allow scoped addresses Merge branch 'miscommit-20060927' [PING6]: Ensure not to reverse-lookup if target is numeric address. [s20060512] YOSHIFUJI Hideaki: [BUILD] Build with standard headers. [ARPING,PING6] Build fix for some old systems. [s20060425] YOSHIFUJI Hideaki: [TRACEROUTE6] Fix ICMPv6 type printing with -v option [TRACEROUTE6] Mark ICMPv6 messages as known [DOC] Maintainer / Contact change [PING6,TRACEPATH6,TRACEROUTE6] Define SOL_IPV6,SOL_ICMPV6 where needed [TRACEROUTE6] Fix source/destination address with -v option [PING6,TRACEPATH6,TRACEROUTE6] Use new RFC3542 advanced API if available [RDISC] Use proper type for is_directly_connected() [PING,PING6] Use proper type for printf() [TRACEROUTE6] Fix inet_pton() error handling [TRACEROUTE6] Use minimum format if 0 is specified for datalen [TRACEROUTE6] Optimize datalen sanity checking code [TRACEPATH6] Use getaddrinfo() to allow scoped addresses [RDISC] Use strerror(errno) instead of sys_errlist[errno] [PING,PING6] Avoid using __constant_htons() if it is really needed [TRACEPATH6] Fix format for subseconds [ARPING,CLOCKDIFF,PING,PING6,TRACEROUTE6] Check return value from setuid(). [PING,PING6] ensure to initialize msg. [MAKEFILE] Make snapshot using git [020927] * arping.sgml, some options were forgotten. * send seqno in network byte order. Me. * Mads Martin Jørgensen Recursive citation: "On request of Mads Martin Jørgensen I've added manpages pregenerated from the Docbook sources. One could argue it is redundant when the Docbook sources are also there, but the argument of not having to install Docbook on a very small system to get the man pages was convinving enough to me. To quote Mads Martin: "How would a system be without a man page for ping?" As a chilidish revenge from my side enjoy with cyrillic date in these man pages. :-) * Ken Cox . Bogus definition of SOCK_DRGAM&SOCK_STREAM on mips. * Error returned from recvmsg() resulted in a bogus printout in traceroute6. Me. * Use IPV6_CHECKSUM on icmp socket in traceroute6. Me. * Noah L. Meyerhans Fix to doc. !* Noah L. Meyerhans What is the problem with "long" triptime? ! Reporter does not respond. _Malignantly_. * Thomas 'Dent' Mirlacher Ping did not exit sometimes! * Add option -W to override default 10 second linger timeout. Me. * Mads Martin Jørgensen : ping should not bind to autoselected source address, it used to work when routing changes. Return classic behaviour, option -B is added to enforce binding. * Pekka Savola Forgotten \n messing output of ping6. * Noah L. Meyerhans traceroute6 -q 1 did not work. * Pekka Savola various sizeof() cleanups in traceroute6.c * "Dmitry V. Levin" wrote: > ping (as well as other utilities) may open raw socket with descriptor <=2; > In case of suid-root, it can be used by malicious user to send data to > this raw socket. > > Yes, modern glibc and some kernels have workaround for it, but > IMHO iputils shouldn't rely on this feature. Taken into account, but no changes made. * "Tilman Heinrich" said some scripts are broken when word "packet" disappeared from "100% packet loss". Despite of the inarguable fact that such scripts are truly mad and deserve breaking (sigh... exit codes are too smart concept for script writers, I guess), I have to recognize removing this word carrying zero information was not enough motivated. Returned. * ping used to retry forever when seeing ENOBUFS/ENOMEM without explicitly given deadline. Being logically correct it is bad in practice f.e. when pinging buggy device which locked up with some packets in queue. So, retry for a finite time... let is be lingertime. Fair? Me. * Two "messages" are sent to rpm maintainers to make their wrong patches failed. * Fix from RH iputils-20001007-deadline.patch. It was lost in the latest rpms btw. * Dax Kelson : added _unsupported_ option to comppile rdisc_srv. [020124] * Michal Kochanowicz typos in tracepath.8 * Michael Wardle : undo silly change of ss000305 (printing rtt in some funny units). Michael noticed that "sec" is not standard abbreviation for time units (bullshit, of course), but real concern is that it is more difficult to interpret with a neglibible improvement to appearance. So, do this as expected: in "ms". * Documentation. Wow! I did it. man pages are disassembled to docbook, audited wrt real state, edited... and promised to be maintained in sync with the state of utilities. [011202] * Utz Bacher Bitops in ping6 were wrong on bigendian machines. Wow, luckily I forgot to acknowledge that patch of 010805 which has gotten rid of kernel bitops and did this so wrongly. * Michael Bakunin (:-)) found mud in tftpd.c, it will crash when directory supplied in argument is longer ~512 symbols. * Alexandr D. Kanevskiy : buffer overflow in clockdiff. Very stupid one, the overflowed buffer even was not used. :-) * Alexandr D. Kanevskiy : shit! Code recognizing kernels with broken IP_RECVERR for raw sockets depended on race and accused even good kernel of being buggy. :-) [011002] * Stepan Koltsov , tracepath/tracepth6 segfaulted when used without address. * Alexandr D. Kanevskiy : arping printed "permission denied" instead of showing help page to non-superuser. [010824] * Alexandr D. Kanevskiy : ping compiled for linux-2.4 forgot to send the second packet, when used with linux-2.2 * Chris Evans : buffer overflow in traceroute6. datalen was messed: counting header in half of places. Funny, looking into LBL traceroute, it is even worse :-) * Alexandr D. Kanevskiy : relayed patches by Solar_Diz. Only missing description of option -q is accepted. * ping6 printed wrong mtu. * Alexandr D. Kanevskiy : -Werror is removed. Newer gcc are buggy and generates some wrong warnings about uninitalized variables, which are evidently initialized. [010805] * Some news from Pekka Savola around setting tos bits. * arping: broadcast-only mode by Ard van Breemen * ping6/traceroute6: parse ICMP errors with extension headers (me) traceroute6 works with size > mtu now. Nice. * ping: Erik Quanstrom . Serious patch. ping interval timer was not very broken, but very unintelligible. Though I remade the code to use leaky bucket logic, which is the most transparent one. Anyway, contribution by Eric is the most important one since the previous release. Short theory of operation: option -i (interval) sets rate r=1/interval pps, option -l (preload) sets burst size of l packets. So, ping sends at most r*t+l packets for an arbitrary interval t. Default values: l=1 and for non-flood case: r=1pps, for flood r=infinity. Nice? Exact algorithm is: Let N(t) be l/r=l*i initially and N(t) grow continuously with time as: N(t+delta) = min{l*i, N(t) + delta} Packet can be transmitted only at the time t_* when 1/r=i <= N(t_*) and in this case N(t) jumps: N(t_* + 0) = N(t_* - 0) - i. When interval is zero, algo degenerates allowing to send any amount of messages. In this case we modify it using l as limit on amount of unanswered requests and waiting for 10msec, when something is not answered. Note that the last thing (10msec) is just to be compatible with BSD manual pages. BSD ping is simply not able to avoid delay technically, we are able now. In result we got some new facilities: * "-f -l 100" becomes very aggressive, in fact on good link it holds permanently 100 packets in flight, which is very different of earlier bevaviour (one packet in flight). * -f and -i are not incompatible more. In fact, "-f -i 1" is equivalent to plain ping, only output is different (dotted). Essentially, change of output format is the only effect. "ping -i 0" is flood printing output in normal format. Moved some parts of code to ping_common.c. Common part is not fully trivial now. :-) * ping: Ian Lynagh , larger and dynamic dup detector. Also, Ian submitted two large patches, one fixing formatting, another doing something with signedness/longness. Not now... Later note: found not working. x + 7 / 8 :-). Sorry... dubious, withdrawn. size of table increased to maximal value instead (8K of memory, not a big deal). * tftpd: an old misprint. left@sbor.spb.su (Igor A. Lefterov) * clockdiff: do not fail, if reversed resolution failed. Tommy Lacroix * ping: audible ping by Patrik Schilt Patrick's option renamed to -a to align to freebsd. * ping: react to device queue overflows using IP_RECVERR. me. * ping: option -S allows to change sndbuf * rarpd is moved from separate package here (people asked) * ping6: kernel style bitops are not used more. * Option -A to adapt to network rtt. * Use BPF, when multiple pings are detected. [001110] * ping is able to select TOS. By Pekka Savola * tracepath* DNS names. By Pawel Krawczyk and Arkadiusz Miskiewicz * ping6 is expected to be compiled with linux-2.2. [001011] * RH bugid#16677: segfault, when ping is used by root and size is large enough. Fix is to allow oversize by root (it is necessary to check kernel side), but clamp it at some safe value. [001010] * More bug fixes from Chris Evans - do not trust h_length returned by system resolver. This value is meaningless in any case. - ping: buffer overflow in fill()!!! Disgraceful bug. * ping: allow not-priviledged users to use broadcasts. It was paranoia. Multicasts were allowed. 8) * ping: but force broadcasts&multicasts not to fragment. BSD does not allow to do this to anyone, we still allow this for superuser. * Option -M to control path mtu discovery. [001007] * By Pekka Savola - SIOCGSTAMP/SO_TIMESTAMP are sensitive to bug in kernel. When get_fast_time != gettimeofday (f.e. timestampless x86), returned stamp can be out of sync with gettimeofday. Workaround is not to use SIOCGSTAMP/SO_TIMESTAMP on such systems. - fixes in man pages - compiles under rh-7.0 * Chris Evans - ping: possible buffer overflow in pr_addr(). [000928] * Sorry. I have lost all the CVS with changes made since 000418. If someone sent me a patch after this date, please, resubmit. Restored from the last backup and mailboxes: * ping*, SO_TIMESTAMP support. * ping*, allow zero data length (reported by Damjan Lango ) * iputils man and help updates. Pekka Savola * ping.8, fix to ping man page. By Dadid Eisner * ping prints addresses in numeric, if destination is numeric. Proposed by Tim Waugh New: * ping: strncpy bug * arping: improvements by Charles Howes - a feature to arping: quit as soon as a reply is received. - default to eth0. - spelling [000418] * llsqrt() was buggy again! (noticed by Sam Farin ) [000404] * tracepath*, "NURDUnet-gw" bug workaround. (noticed by Vitaly E.Lavrov ) * tracepath*, handle case of routers initializing rtt to 128. Vitaly E.Lavrov * shadowed icmp_sock in ping6. James Morris * Bug in ping -f, introduced with SO_RCVTIMEO. me. * llsqrt() (ping, ping6) was wrong yet. me. [000310] * Print mean deviation of RTT in ping/ping6. * Use SIOCGSTAMP in ping/ping6. Old behaviour calculating true user-to-user latency is restored with option -U. Reason for this stupid change is mainly political; people wonder why freebsd has twice less latency on loopback. If to follow along this line, we have to print rtt equal to 0. 8) [ LATER NOTE: actually, the change is _right_ without any doubts. Ping has another bug: nameresolver is blocking, so that when it dies not respond, ping shows evenly increasing by 1 sec RTT. It is very confusing (look through linux-kernel maillists to count number of people, who were cheated by misconfigured dns). ] * Use SO_RCVTIMEO instead of poll() with ping/ping6 -f. * Added -V option to arping/ping/ping6/traceroute6/rdisc to print snapshot number. [000305] * rdisc: ugly bug in getting interface list. me. * ping/ping6: ping -i N, N>=3 did not work. Jeff Jonson * ping/ping6: microsecond rtt measurements. me. [000120] * ping/ping6: non-zero exit code even without -w. [991024] * Option "-i" to ping/ping6 takes fractional time now, so that "ping -i 0.3 xxx" pings each 300 msec. The idea is by Marc Boucher * alpha/glibc-2.1 alignment problems in ping are fixed (struct timeval was wrongly aligned). [990915] * ping/ping6 worked only with kernels 2.3.15+ in 990824. [990824] * tftpd is added. It uses MSG_CONFIRM to confirm arp entries. * ping6: workaround for bug in some egcs versions. [990610] * ping: output buffer was too small for full sized ping. * ping: silly restriction on ping size is removed. [990530] * short man pages (Oleg M. Shumsky ) * ping6: get and print hop limit of reply packets (ME) * rdisc deletes routes before exit with -TERM * ping/ping6: option -w TIMEOUT * arping: exit with error, if received no replies in normal (not DAD and not unsilicited ARP) mode.