Hi,

the new --numeric-ports option makes netstat act as if the
--numeric-hosts has been passed. That is, just like -n.


~$ netstat -Ainet --numeric-hosts
tcp        0      0 10.75.100.76:4411     10.175.254.9:ssh ES...
~$ netstat -Ainet --numeric-ports
tcp        0      0 10.75.100.76:4411     10.175.254.9:22  ES...

and the man pages says :

Hi,

I'm amazed nobody replied to this bug for 3.5 years already!
I discovered myself the problem and I could get a behaviour closer
to the expectations by applying those changes:
--- inet.c.orig 2006-12-28 19:00:45.000000000 +0100
+++ inet.c      2006-12-28 18:57:37.000000000 +0100
@@ -159,16 +159,11 @@
 #ifdef DEBUG
     fprintf (stderr, "rresolve: %08lx, mask %08x, num %08x \n", ad, netmask, numeric);
 #endif
-    if (ad == INADDR_ANY) {
-       if ((numeric & 0x0FFF) == 0) {
-           if (numeric & 0x8000)
-               safe_strncpy(name, "default", len);
-           else
-               safe_strncpy(name, "*", len);
-           return (0);
-       }
+    if (ad == INADDR_ANY && ! (numeric & 0x0004)) {
+       safe_strncpy(name, "*", len);
+       return (0);
     }
-    if (numeric & 0x0FFF) {
+    if (numeric & 0x0004) {
         safe_strncpy(name, inet_ntoa(sin->sin_addr), len);
        return (0);
     }

At least when calling netstat --numeric-ports we now get
tcp        0      0 my.machine.com:139             *:*                     LISTEN
and netstat --numeric-hosts:
tcp        0      0 127.0.0.1:imap2         0.0.0.0:*               LISTEN

But
- I don't know what is the supposed usage of "default" vs "*" for anycasts
- When I see how wrong the interpretation of the "numeric" flags
is in this part of the code I'm wondering for the whole stuff...

Phil

I think no matter what flags are given, the remote address of a LISTENING
socket should be *:*.

(at least if ad == INADDR_ANY, however I guess there are no other cases)

Gruss
Bernd

Bernd Eckenfels wrote:
Maybe, but in the fct we don't know in which case we are
(listening or not? local or remote addr?)
so we've to choose 0.0.0.0 or * (or default) for all INADDR_ANY

or make a lot of changes in the code just for cosmetics...

Personally I'm fine with * in any cases but today netstat -ln
gives 0.0.0.0:* so if you want to fix the --numeric-ports option
while keeping the current behaviour of -l and -ln it's better
to keep my initial proposal.

Just my 2 cents.

Phil

Hi,

The following is a cut-down version of Philippe Teuwen's patch.

I have taken out the changes he made to the special handling of
INADDR_ANY, and just corrected the mask for reading the numeric
parameter.  The presence of FLAG_NUM_PORT or FLAG_NUM_USER should
certainly be ignored in INET_rresolve(), shouldn't it?

Hopefully you will be able to apply this patch.  This is a very simple
bug, but it has still managed to stay around for more than five
years :-(.

--- lib/inet.c.orig 2008-08-02 13:07:24.000000000 +0100
+++ lib/inet.c      2008-08-02 14:28:37.000000000 +0100
@@ -161,5 +161,5 @@
 #endif
     if (ad == INADDR_ANY) {
-	if ((numeric & 0x0FFF) == 0) {
+	if (( numeric & FLAG_NUM_HOST ) == 0) {
 	    if (numeric & 0x8000)
 		safe_strncpy(name, "default", len);
@@ -169,5 +169,5 @@
 	}
     }
-    if (numeric & 0x0FFF) {
+    if (numeric & FLAG_NUM_HOST) {
         safe_strncpy(name, inet_ntoa(sin->sin_addr), len);
 	return (0);


Thanks,
Tom Vajzovic