New tap for tethereal: io statistics that provides frames/bytes counts for frames...

[obnox/wireshark/wip.git] / README.irix

$Id: README.irix,v 1.4 2001/01/10 09:54:44 guy Exp $

1. Some problems seen by one person who tried to build Ethereal on IRIX;
your mileage may vary.

To: ethereal-dev@zing.org
Subject: Re: [ethereal-dev] Ethereal on SGI
From: Randall Hopper <aa8vb@yahoo.com>
Date: Sat, 18 Sep 1999 12:19:31 -0400

 |>      Just tried to build ethereal on SGI IRIX.  configure failed because
 |> there is no net/bpf.h.

Thanks for all the help.  I pulled the latest today (0.7.3), and after a
few fix-ups, I got it built.  Here are the tweaks I needed to build it for
SGI IRIX 6.5 using gcc.

==========
PROBLEM #1
==========

   Ethereal doesn't use RPATH (see ld(1) for shared library linking on all
   platforms that support it (e.g. SGI IRIX, FreeBSD ELF, etc.), just
   Solaris it appears.  Instead, ethereal recommends use of the
   LD_LIBRARY_PATH hack which causes all sorts problems.

   On SGI IRIX this is -rpath to the linker, on Solaris -R, on FreeBSD ELF
   --rpath, etc.

   SOLUTION: Slip in my linker flags when configuring ethereal so it can
             find libgtk, libgmodule, libsnmp, etc. at run-time without
             LD_LIBRARY_PATH:

   env LDFLAGS="-rpath $HOME/software/gtk+-1.2.3/lib:$HOME/software/glib-1.2.3/lib:/usr/local/lib" configure --prefix=$HOME/software/ethereal-0.7.3 --with-gtk-prefix=$HOME/software/gtk+-1.2.3

   BETTER SOLUTION:  Modify configure to support RPATH for more platforms
                     than Solaris.

==========
PROBLEM #2
==========

   __P not defined -- packet-snmp.c fails to build
   (gcc version 2.95.1 19990816 (release), SGI IRIX 6.5.5f)

   SOLUTION:  Include gnuc.h -- it defines it

   --- ORIG/packet-snmp.c  Mon Aug 30 11:10:17 1999
   +++ packet-snmp.c       Sat Sep 18 11:35:45 1999
   @@ -25,4 +25,6 @@
     */

   +#include "gnuc.h"
   +
    #ifdef HAVE_CONFIG_H
    # include "config.h"

Randall

===============================================================================

Problem #1:

checking for GTK - version >= 1.2.0... 186062:./conftest: rld: Fatal Error: Cannot Successfully map soname 'libgtk-1.2.so.1' under any of the filenames /usr/lib32/libgtk-1.2.so.1:/usr/lib32/internal/libgtk-1.2.so.1:/lib32/libgtk-1.2.so.1:/opt/lib32/libgtk
no
*** Could not run GTK test program, checking why...
*** The test program compiled, but did not run. This usually means
*** that the run-time linker is not finding GTK or finding the wrong
*** version of GTK. If it is not finding GTK, you'll need to set your
*** LD_LIBRARY_PATH environment variable, or edit /etc/ld.so.conf to point
*** to the installed location  Also, make sure you have run ldconfig if that
*** is required on your system
***
*** If you have an old version installed, it is best to remove it, although
*** you may also be able to get things to work by modifying LD_LIBRARY_PATH
***
*** If you have a RedHat 5.0 system, you should remove the GTK package that
*** came with the system with the command
***
***    rpm --erase --nodeps gtk gtk-devel
configure: error: GTK+ distribution not found.

==============================================================================

Problem #2:

gcc -DHAVE_CONFIG_H -I. -I. -I.     -Wall -g -O2 -Iwiretap -I/home/rhh/software/gtk+-1.2.3/include -I/home/rhh/software/glib-1.2.3/lib/glib/include -I/home/rhh/software/glib-1.2.3/include -c packet-snmp.c
In file included from packet-snmp.c:62:
/usr/local/include/ucd-snmp/asn1.h:131: parse error before `__P'
/usr/local/include/ucd-snmp/asn1.h:132: parse error before `__P'
/usr/local/include/ucd-snmp/asn1.h:133: parse error before `__P'
/usr/local/include/ucd-snmp/asn1.h:134: parse error before `__P'
/usr/local/include/ucd-snmp/asn1.h:135: parse error before `__P'
...
packet-snmp.c: In function `dissect_snmp':
packet-snmp.c:344: warning: implicit declaration of function `asn_parse_header'
packet-snmp.c:358: warning: implicit declaration of function `snmp_comstr_parse'
packet-snmp.c:358: warning: assignment makes pointer from integer without a cast
packet-snmp.c:407: warning: assignment makes pointer from integer without a cast
...
packet-snmp.c: In function `proto_register_snmp':
packet-snmp.c:869: warning: implicit declaration of function `init_mib'
gmake[2]: *** [packet-snmp.o] Error 1
gmake[2]: Leaving directory `/home/rhh/software/PKGS/t/ethereal-0.7.3'
gmake[1]: *** [all-recursive] Error 1
gmake[1]: Leaving directory `/home/rhh/software/PKGS/t/ethereal-0.7.3'
gmake: *** [all-recursive-am] Error 2

-------------------------------------------------------------------------------

Problem #3 (same as Problem #1):

> ./ethereal 
189684:./ethereal: rld: Fatal Error: Cannot Successfully map soname 'libsnmp.so' under any of the filenames /home/rhh/software/gtk+-1.2.3/lib/libsnmp.so:/home/rhh/software/glib-1.2.3/lib/libsnmp.so:/usr/lib32/libsnmp.so:/usr/lib32/internal/libsnmp.so:/lib32/libsnmp.so:/opt/lib32/libsnmp.so: 

2. Patches to libpcap that may be necessary

On IRIX, the libpcap library uses, in "pcap_open_live()", the snapshot
length passed in, without reducing it to be less than or equal to the
MTU of the interface.  If the snapshot length is larger than the
interface MTU, IRIX will return an error; this means that, for example,
the Ethereal default snapshot length of 65535, chosen so that all of the
packet is captured, will not work.

Here is a patch to libpcap 0.4 source that should fix that problem.  If
it doesn't fix the problem, or if it reduces the snapshot length below
the MTU (so that a snapshot length of 65535 doesn't capture all the data
in the packets), please report this to ethereal-dev@ethereal.com, so
that we know that it didn't work - we'll probably send you debugging
patches in the hopes of being able to make it work.

diff -c ../libpcap-0.4/pcap-snoop.c ./pcap-snoop.c
*** ../libpcap-0.4/pcap-snoop.c Tue Apr  8 21:07:01 1997
--- ./pcap-snoop.c      Tue Jan 18 00:16:18 2000
***************
*** 126,132 ****
--- 126,135 ----
        struct sockaddr_raw sr;
        struct snoopfilter sf;
        u_int v;
+       int ll_hdrlen;
+       int snooplen;
        pcap_t *p;
+       struct ifreq ifr;
  
        p = (pcap_t *)malloc(sizeof(*p));
        if (p == NULL) {
***************
*** 154,169 ****
        }
        v = 64 * 1024;
        (void)setsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char *)&v, sizeof(v));
-       if (ioctl(fd, SIOCSNOOPLEN, &snaplen) < 0) {
-               sprintf(ebuf, "SIOCSNOOPLEN: %s", pcap_strerror(errno));
-               goto bad;
-       }
-       p->snapshot = snaplen;
-       v = 1;
-       if (ioctl(fd, SIOCSNOOPING, &v) < 0) {
-               sprintf(ebuf, "SIOCSNOOPING: %s", pcap_strerror(errno));
-               goto bad;
-       }
        /*
         * XXX hack - map device name to link layer type
         */
--- 157,162 ----
***************
*** 179,195 ****
--- 172,242 ----
            strncmp("qaa", device, 3) == 0) {
                p->linktype = DLT_EN10MB;
                p->offset = RAW_HDRPAD(sizeof(struct ether_header));
+               ll_hdrlen = sizeof(struct ether_header);
        } else if (strncmp("ipg", device, 3) == 0 ||
                   strncmp("rns", device, 3) == 0 ||    /* O2/200/2000 FDDI */
                   strncmp("xpi", device, 3) == 0) {
                p->linktype = DLT_FDDI;
                p->offset = 3;                          /* XXX yeah? */
+               ll_hdrlen = 13;
        } else if (strncmp("ppp", device, 3) == 0) {
                p->linktype = DLT_RAW;
+               ll_hdrlen = 0;  /* DLT_RAW meaning "no PPP header, just the IP packet"? */
        } else if (strncmp("lo", device, 2) == 0) {
                p->linktype = DLT_NULL;
+               ll_hdrlen = 4;  /* is this just like BSD's loopback device? */
        } else {
                sprintf(ebuf, "snoop: unknown physical layer type");
+               goto bad;
+       }
+ #ifdef SIOCGIFMTU
+       /*
+        * XXX - IRIX appears to give you an error if you try to set the
+        * capture length to be greater than the MTU, so let's try to get
+        * the MTU first and, if that succeeds, trim the snap length
+        * to be no greater than the MTU.
+        */
+       (void)strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
+       if (ioctl(fd, SIOCGIFMTU, (char *)&ifr) < 0) {
+               sprintf(ebuf, "SIOCGIFMTU: %s", pcap_strerror(errno));
+               goto bad;
+       }
+       /*
+        * OK, we got it.
+        * XXX - some versions of IRIX 6.5 define "ifr_mtu" and have an
+        * "ifru_metric" member of the "ifr_ifru" union in an "ifreq"
+        * structure, others don't.
+        *
+        * I've no idea what's going on, so, if "ifr_mtu" isn't defined,
+        * we define it as "ifr_metric", as using that field appears to
+        * work on the versions that lack "ifr_mtu" (and, on those that
+        * don't lack it, "ifru_metric" and "ifru_mtu" are both "int"
+        * members of the "ifr_ifru" union, which suggests that they
+        * may be interchangeable in this case).
+        */
+ #ifndef ifr_mtu
+ #define ifr_mtu       ifr_metric
+ #endif
+       if (snaplen > ifr.ifr_mtu)
+               snaplen = ifr.ifr_mtu;
+ #endif
+ 
+       /*
+        * The argument to SIOCSNOOPLEN is the number of link-layer
+        * payload bytes to capture - it doesn't count link-layer
+        * header bytes.
+        */
+       snooplen = snaplen - ll_hdrlen;
+       if (snooplen < 0)
+               snooplen = 0;
+       if (ioctl(fd, SIOCSNOOPLEN, &snooplen) < 0) {
+               sprintf(ebuf, "SIOCSNOOPLEN: %s", pcap_strerror(errno));
+               goto bad;
+       }
+       p->snapshot = snaplen;
+       v = 1;
+       if (ioctl(fd, SIOCSNOOPING, &v) < 0) {
+               sprintf(ebuf, "SIOCSNOOPING: %s", pcap_strerror(errno));
                goto bad;
        }