_ _ _ | \ | | |_ ___ _ __ | \| | __/ _ \| '_ \ | |\ | || (_) | |_) | |_| \_|\__\___/| .__/ |_| Network Top -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- v2.1 (Jun 2002) B1. https:// & tcp wrappers If you compile ntop with tcp wrappers and attempt to access an https:// url, it will fail with "Jun 11 14:22:52 www ntop[22622]: warning: can't get client address: Bad file descriptor" in the log. hosts.allow: ntop: 192.168.0. hosts.deny: ntop: ALL No real clue why - the code that ntop uses: { struct request_info req; request_init(&req, RQ_DAEMON, DAEMONNAME, RQ_FILE, myGlobals.newSock, NULL); fromhost(&req); if(!hosts_access(&req)) { closelog(); /* just in case */ openlog(DAEMONNAME, LOG_PID, deny_severity); syslog(deny_severity, "refused connect from %s", eval_client(&req)); } else handleHTTPrequest(from.sin_addr); } is the same as the code used, for example, by sshd. It's the fromhost() call - without a value set there, there is no file descriptor set, and the hosts_access() call is then "properly" rejecting the request. ======================================== B2. https:// & Opera With fixes through 25Jun2002, this works, albeit VERY SLOWLY on the handshake. Sometimes as much as two minutes. After that, clicking around through the ntop web server works fine, but it is also subject to a noticeable delay bringing up the png images. However, some versions of Opera (6.03 under Win98), give an SSL error and no results: Jun 29 08:17:31 localhost ntop[17731]: SSL(read)ERROR [Thread 17731]: error:140EC0E5:SSL routines:SSL2_READ_INTERNAL:ssl handshake failure at s2_pkt.c(142) Sometimes with tcp wrappers enabled you get errors: Jun 25 12:17:37 tigger ntop[26664]: SSL(ssl_init_connection)ERROR [Thread 26664]: error:1407609C:SSL routines:SSL23_GET_CLIENT_HELLO:http request at s23_srvr.c(390) Jun 25 12:17:37 tigger ntop[26664]: warning: can't get client address: Bad file descriptor It's unknown - but suspected - that this last case is the same as #1, above. ======================================== B3. sflowPlugin.c will not compile in single threaded mode (it uses queuePacket, which is #ifdef ed out in pbuf.c) ================================================================================ ISSUES - not bugs per se, but worth a comment... ================================================================================ I1. https:// & Netscape 6.2.2 et al Hangs the ntop web server thread. It goes into the ssl_accept() routine and never comes back. ntop continues processing packets, but there is no way to retrieve data or softly shutdown ntop. According to openSSL, it's a problem with Netscape not handling a legal combination of flags on the handshake. openSSL 0.9.6c (or d?) is supposed to have a work-around for this. Both sides (client and server) are waiting for the other to finish the handshake. A partial fix is present in the final code, but disabled by default. This uses a watchdog thread to detect the hang and cancel out the connection. It means that the offending user gets nothing returned, but at least the ntop web server isn't hung. The fix is enabled either by the ./configure parameter --enable-sslwatchdog or the run-time parameter --ssl-watchdog. Testing seems to show this is not necessary for: Linux / konqueror 2.2.2 (under kde 2.2.2-2) Linux / mozilla 0.9.9-12.7.2 Linux / mozilla 1.0.1-4 Linux / Netscape (Navigator) 4.78-2 Win2K / Internet Explorer 5.5 Win2K / Netscape (Navigator) 4.79 Win2K / Netscape (Navigator) 4.61 (oh the things I do for Luca...) Win98 / Internet Explorer 5.5 It is necessary, but the fix "works" for: Linux / Galeon 1.2.0-5 Linux / Galeon 1.2.5-3 Linux / Opera 6.0-20020225.3 Win2K / Netscape 6.2.2 Win2K / Opera 6.03 Remember, saying it "works" means that the hangup doesn't kill ntop's web server for other users, not that the user of the offending browser will see anything meaningful. Galeon gets farther, but still hangs itself up. Recommended: If you aren't seeing the problem: don't worry - be happy. If you are, try running for a while with --ssl-watchdog If that seems to fix it, then compile with --enable-sslwatchdog ======================================== I2. SIGPIPE - primarilly under gdb in Win2K environment. The problem is that Internet Explorer (and other browsers) seem to close the connection when they receive an unknown certificate in response to an https:// request. This causes a SIGPIPE and kills the web handling thread - sometimes (for sure, the ONLY case I know of is if ntop is run under Linux gdb connected to from a Win2K browser). This code simply counts SIGPIPEs and ignores them. However, it's not that simple - under gdb under Linux (and perhaps other OSes), the thread mask on a child thread disables our ability to set a signal handler for SIGPIPE. However, gdb seems to trap the SIGPIPE and reflect it to the code, even if the code wouldn't see it without the debug! Hence the multi-step code. This code SHOULD be safe. Many other programs have had to so this. Because I'm not sure, I've put both a compile time (--enable-ignoresigpipe) and run-time --ignore-sigpipe option in place. Recommended: If you aren't seeing the "mysterious death of web server" problem: don't worry - be happy. If you are, try running for a while with --ignore-sigpipe If that seems to fix it, then compile with --enable-ignoressigpipe -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Older bugs (v1.3) - Unknown if these are still current -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- KNOWN_BUGS ========== O1. Linux Sockets Bug (Platform Linux/i386) [Alan.Cox@linux.org is aware of the above issue. Status: no reply] Below you can find the packetLogger code that can be used to reproduce the problem. Suppose to have host A (MAC Address 08:00:69:0B:6F:A1) and host B (MAC Address 00:20:AF:73:C6:2E). Host B is an i386 running Linux 2.X. No matter what OS runs on A. Now start "packetLogger 08:00:69:0B:6F:A1 00:20:AF:73:C6:2E" in order to filter packets flowing though A and B. I suppose there's no other traffic (e.g. telnet) between A and B. Now from A do 'ftp B' and transfer a file C (large, e.g. > 1 MB). Stop packetLogger, look at the # of packets and restart it. Now from B do 'ftp A' and transfer the very same file C [if A and B are Linux boxes you can start packetLogger on both hosts]. You will notice that in the second case you've lost many packets whereas in the first case everything works fine. ======================================== #include <sys/param.h> #include <sys/ioctl.h> #include <sys/socket.h> #include <sys/time.h> #include <net/if.h> #include <linux/if_ether.h> #include <netinet/in.h> #include <errno.h> #include <malloc.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <unistd.h> struct ethernet_header { u_char ether_dhost[6]; u_char ether_shost[6]; u_short ether_type; }; char* etheraddr_string(const u_char *ep) { u_int i, j; char *cp; struct enamemem *tp; static char buf[sizeof("00:00:00:00:00:00")]; char hex[] = "0123456789ABCDEF"; cp = buf; if ((j = *ep >> 4) != 0) *cp++ = hex[j]; else *cp++ = '0'; *cp++ = hex[*ep++ & 0xf]; for(i = 5; (int)--i >= 0;) { *cp++ = ':'; if ((j = *ep >> 4) != 0) *cp++ = hex[j]; else *cp++ = '0'; *cp++ = hex[*ep++ & 0xf]; } *cp = '\0'; return (buf); } int main(int argc, char* argv[]) { struct ifreq ifr; struct sockaddr sa; char *device = "eth0"; unsigned long packetNum=0, totLen=0; int fd; if(argc != 3) { printf("Usage: %s <MAC Addr. host A> <MAC Addr. host B>\n", argv[0]); printf("Example: %s 08:00:69:0B:6F:A1 00:20:AF:73:C6:2E\n", argv[0]); return(-1); } fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL)); if (fd < 0) { printf("Error creating socket.\n"); return(-1); } /* Bind to the interface name */ memset(&sa, 0, sizeof(sa)); sa.sa_family = AF_INET; (void)strncpy(sa.sa_data, device, sizeof(sa.sa_data)); if (bind(fd, &sa, sizeof(sa))) { printf("bind: error\n"); return(-1); } memset(&ifr, 0, sizeof(ifr)); strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0 ) { printf("SIOCGIFHWADDR: error\n"); return(-1); } /* Base the buffer size on the interface MTU */ memset(&ifr, 0, sizeof(ifr)); strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); if (ioctl(fd, SIOCGIFMTU, &ifr) < 0 ) { printf("SIOCGIFMTU: error\n"); return(-1); } while(1) { struct sockaddr from; int fromlen, cc, len=0; u_char bp[2048], srcHost[64], dstHost[64]; struct ethernet_header *ep; do { fromlen = sizeof(from); cc = recvfrom(fd, bp, 2048, 0, &from, &fromlen); len += cc; } while (strcmp(device, from.sa_data)); ep = (struct ethernet_header*)bp; strcpy(srcHost, etheraddr_string(ep->ether_shost)); strcpy(dstHost, etheraddr_string(ep->ether_dhost)); if(strcmp(srcHost, argv[1]) && strcmp(srcHost, argv[2])) continue; else if(strcmp(dstHost, argv[1]) && strcmp(dstHost, argv[2])) continue; else { totLen += len; printf("%5d\t%8u\t%s -> %s (len=%d)\n", ++packetNum, totLen, srcHost, dstHost, len); } } close(fd); return(0); } ======================================== O2. iPPP (Linux) [iPPP guys have been informed. Status: no reply] ntop works with PPP but it presents some problems with iPPP (ISDN PPP). Some packets cannot be decoded properly. Tools other than ntop (e.g. tcpdump, ethereal) can't handle such packets either. -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-