inet_aton in Socket.pm core module is broken #11169
Comments
From quanah@zimbra.comCreated by quanah@zimbra.comThis is a bug report for perl from quanah@zimbra.com, ----------------------------------------------------------------- Perl Info |
From @smpetersOn Wed Mar 02 15:35:49 2011, quanah@zimbra.com wrote:
Can you provide some sample code along with what operating system this Thanks! |
|
The RT System itself - Status changed from 'new' to 'open' |
From quanah@zimbra.com--On Wednesday, March 02, 2011 3:35 PM -0800 perlbug-followup@perl.org
Here is an example. If /etc/hosts is ordered with IPv6 before IPv4, inet_aton fails to return zimbra@zqa-064:~/libexec$ ./qtest.pl "IO::Socket::INET h:" is the value of inet_aton. Example /etc/hosts used here: 127.0.0.1 localhost # The following lines are desirable for IPv6 capable hosts 10.137.245.250 w1-zbuild-ns120-01-qa-nfs.eng.vmware.com Even modifying it thusly doesn't change the behavior: # The following lines are desirable for IPv6 capable hosts 127.0.0.1 localhost 10.137.245.250 w1-zbuild-ns120-01-qa-nfs.eng.vmware.com Now, if I move the IPv4 addresses to be before the IPv6 configuration, zimbra@zqa-064:~/libexec$ ./qtest.pl Here is /etc/hosts for this run: 127.0.0.1 localhost 10.137.245.250 w1-zbuild-ns120-01-qa-nfs.eng.vmware.com # The following lines are desirable for IPv6 capable hosts It basically seems that Perl's inet_aton stops parsing /etc/hosts for IPv4 --Quanah -- Quanah Gibson-Mount Zimbra :: the leader in open source messaging and collaboration |
From quanah@zimbra.com--On Friday, March 04, 2011 11:37 AM -0800 Steve Peters via RT
I discovered the issue while using Net::LDAP. Here is a brief set of #!/usr/bin/perl my $ldap_master_url="ldap://zqa-064.eng.vmware.com:389"; my $ldap = Net::LDAP->new("$ldap_master_url") or die "$@"; $ldap->unbind(); Net::LDAP makes a call to IO::Socket::INET, which in turn calls inet_aton. I made the following modifications to IO::Socket::INET to see what was root@zqa-064:/opt/zimbra/zimbramon/lib/x86_64-linux-gnu-thread-multi/IO/Socket# Inline Patchdiff -u INET.pm.orig INET.pm
--- INET.pm.orig 2011-03-01 13:07:51.970458204 -1000
+++ INET.pm 2011-03-01 13:10:37.430467992 -1000
@@ -77,6 +77,8 @@
$port = $1
if(defined $addr && $addr =~ s,:([\w\(\)/]+)$,,);
+ print "IO::Socket::INET Address: $addr\n";
+
if(defined $proto && $proto =~ /\D/) {
my $num = _get_proto_number($proto);
unless (defined $num) {
@@ -127,9 +129,12 @@
my @addr;
if ($multi && $addr_str !~ /^\d+(?:\.\d+){3}$/) {
(undef, undef, undef, undef, @addr) = gethostbyname($addr_str);
+ print "IO::Socket::Inet addr_str: $addr_str\n";
} else {
+ print "IO::Socket::Inet addr_str: $addr_str\n";
my $h = inet_aton($addr_str);
push(@addr, $h) if defined $h;
+ print "IO::Socket::INET h: $h\n";
}
@addr;
}
This is perl, v5.10.1 (*) built for x86_64-linux-gnu-thread-multi The version of Socket shipped with it is $VERSION = "1.82"; This is on Ubuntu 10.04 LTS 64-bit --Quanah -- Quanah Gibson-Mount Zimbra :: the leader in open source messaging and collaboration |
From @smpetersOn Fri, Mar 4, 2011 at 2:43 PM, Quanah Gibson-Mount <quanah@zimbra.com>wrote:
The simple answer is that inet_aton() on a libc level does not handle IPv6 Onto the fun part... We are obviously still using inet_aton() in Steve Peters |
From quanah@zimbra.com--On Friday, March 04, 2011 11:57 PM -0600 Steve Peters
Well, there is IO::Socket::INET6, which works great with IPv4 and IPv6. And, as I noted, it is valid to define the IPv4 addresses after the IPv6 --Quanah -- Quanah Gibson-Mount Zimbra :: the leader in open source messaging and collaboration |
From @nwc10On Fri, Mar 04, 2011 at 10:42:28PM -0800, Quanah Gibson-Mount wrote:
That's not quite correct, given that Socket::inet_aton() isn't actually a pure void if ((*host != '\0') && inet_aton(host, &ip_address)) { phe = gethostbyname(host); XSRETURN_UNDEF;
So I think that this is actually a bug in the OS gethostbyname() However, I'm hoping that it can be worked round with something like this: From f0a6093 Mon Sep 17 00:00:00 2001 This is a last resort, in case gethostbyname() is getting confused because it ext/Socket/Socket.xs | 20 +++++++++++++++++++- Inline Patchdiff --git a/ext/Socket/Socket.xs b/ext/Socket/Socket.xs
index 9214fc1..71f0c57 100644
--- a/ext/Socket/Socket.xs
+++ b/ext/Socket/Socket.xs
@@ -435,7 +435,25 @@ inet_aton(host)
XSRETURN(1);
}
- XSRETURN_UNDEF;
+ ST(0) = &PL_sv_undef;
+#ifdef HAS_GETADDRINFO
+ {
+ struct addrinfo hints = { 0 };
+ struct addrinfo *result;
+
+ hints.ai_family = AF_INET;
+ if (!getaddrinfo(host, NULL, &hints, &result) && result) {
+ /* These two checks are just paranoia. */
+ if (result->ai_family == AF_INET && result->ai_addr->sa_family == AF_INET) {
+ ST(0) = newSVpvn_flags((char *)&((const struct sockaddr_in *)result->ai_addr)->sin_addr,
+ sizeof(ip_address), SVs_TEMP);
+ }
+ freeaddrinfo(result);
+ }
+ }
+#endif
+
+ XSRETURN(1);
}
void
--
I can't test that for sure, as I've not found a machine where I can replicate That patch, as inlined, isn't going to apply to 5.10.1's Socket.xs, or even Nicholas Clark |
From @nwc10On Mon, Mar 07, 2011 at 02:09:26PM -0800, Quanah Gibson-Mount wrote: [hmm, I had hoped that that would work]
I think if you can get rewrite the test as a single standalone file of C, and If the C code works, whereas the XS version doesn't, then we need to scratch Nicholas Clark |
From quanah@zimbra.com--On Sunday, March 06, 2011 11:35 AM +0000 Nicholas Clark <nick@ccl4.org>
I downloaded the 5.10.1 Socket code, and modified Socket.xs thusly: root@zqa-064:/tmp/q# diff -u Socket.xs.orig Socket.xs Inline Patch--- Socket.xs.orig 2011-03-07 11:59:03.640485053 -1000
+++ Socket.xs 2011-03-07 12:00:13.840461063 -1000
@@ -231,17 +231,34 @@
{
struct in_addr ip_address;
struct hostent * phe;
- int ok = (*host != '\0') && inet_aton(host, &ip_address);
- if (!ok && (phe = gethostbyname(host)) &&
- phe->h_addrtype == AF_INET && phe->h_length == 4) {
- Copy( phe->h_addr, &ip_address, phe->h_length, char );
- ok = 1;
+ if ((*host != '\0') && inet_aton(host, &ip_address)) {
+ ST(0) = newSVpvn_flags((char *)&ip_address, sizeof
- ST(0) = sv_newmortal(); void I then rebuilt Socket, installing it into a different location I then moved the system socket aside (renamed Socket.pm and Socket.xs in Still fails: zimbra@zqa-064:~/libexec$ ./qtest.pl Maybe I should be bothering the core linux devs instead? --Quanah -- Quanah Gibson-Mount Zimbra :: the leader in open source messaging and collaboration |
From quanah@zimbra.com--On Monday, March 07, 2011 10:15 PM +0000 Nicholas Clark <nick@ccl4.org>
root@zqa-064:~# ./a.out zqa-064.eng.vmware.com So it correctly gets the IPv6 and IPv4 addresses, although it doesn't have I used <http://www.logix.cz/michal/devel/various/getaddrinfo.c.xp> for the --Quanah -- Quanah Gibson-Mount Zimbra :: the leader in open source messaging and collaboration |
From @ikegamiOn Sun, Mar 6, 2011 at 6:35 AM, Nicholas Clark <nick@ccl4.org> wrote:
Or its returning an IPv6 address? Maybe it would be more appropriate to use struct hostent * gethostbyname(const char *name); |
From @nwc10On Wed Mar 02 15:35:49 2011, quanah@zimbra.com wrote:
I'm curious. How many lines of "locally applied patches" did you edit I don't think that it makes a difference to the problem, as Debian Nicholas Clark |
From @nwc10On Mon, Mar 07, 2011 at 02:09:26PM -0800, Quanah Gibson-Mount wrote:
You typed Socket.xs What matters at load time is the shared object (which on Linux is a .so, If you didn't move that, then I think that *it* will still be loaded, and On Mon, Mar 07, 2011 at 02:45:50PM -0800, Quanah Gibson-Mount wrote:
If I compile that code on an Ubuntu box, and use your first /etc/hosts, I $ ./getaddrinfo zqa-064.eng.vmware.com If I change it to pass AF_INET in the hints to getaddrinfo(), and also make it $ ./85308 zqa-064.eng.vmware.com Please could you compile the attached variant, and run it? I'm not sure what Nicholas Clark |
From @nwc10/*
* getaddrinfo.c - Simple example of using getaddrinfo(3) function.
*
* Michal Ludvig <michal@logix.cz> (c) 2002, 2003
* http://www.logix.cz/michal/devel/
*
* License: public domain.
*/
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <netdb.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <arpa/inet.h>
int
lookup_host (const char *host)
{
struct addrinfo hints, *res;
int errcode;
char addrstr[100];
void *ptr;
memset (&hints, 0, sizeof (hints));
hints.ai_family = PF_INET;
hints.ai_socktype = SOCK_STREAM;
hints.ai_flags |= AI_CANONNAME;
errcode = getaddrinfo (host, NULL, &hints, &res);
if (errcode != 0)
{
perror ("getaddrinfo");
return -1;
}
printf ("Host: %s\n", host);
while (res)
{
inet_ntop (res->ai_family, res->ai_addr->sa_data, addrstr, 100);
switch (res->ai_family)
{
case AF_INET:
ptr = &((struct sockaddr_in *) res->ai_addr)->sin_addr;
break;
case AF_INET6:
ptr = &((struct sockaddr_in6 *) res->ai_addr)->sin6_addr;
break;
}
inet_ntop (res->ai_family, ptr, addrstr, 100);
printf ("IPv%d address: %s (%s)\n", res->ai_family == PF_INET6 ? 6 : 4,
addrstr, res->ai_canonname);
res = res->ai_next;
}
{
struct hostent * phe = gethostbyname(host);
char **addr;
puts("gethostbyname");
if (!phe) {
perror ("gethostbyname");
return -1;
}
addr = phe->h_addr_list;
while (*addr) {
inet_ntop (phe->h_addrtype, *addr, addrstr, 100);
printf ("IPv%d address: %s (%s)\n", phe->h_addrtype == AF_INET6 ? 6 : 4,
addrstr, phe->h_name);
++addr;
}
}
return 0;
}
int
main (int argc, char *argv[])
{
if (argc < 2)
exit (1);
return lookup_host (argv[1]);
} |
From quanah@zimbra.com--On March 8, 2011 12:51:08 PM -0800 Nicholas Clark via RT
I didn't edit any out. I sent the full report that came from perlbug. I --Quanah -- Zimbra :: the leader in open source messaging and collaboration |
From quanah@zimbra.com--On March 8, 2011 9:38:02 PM +0000 Nicholas Clark <nick@ccl4.org> wrote:
Yes, Socket.so, sorry.
Yes, I'm quite aware of that. ;)
It is what was moved, to explicitly avoid it being loaded. I will build the attached variant and let you know how it goes. --Quanah -- Zimbra :: the leader in open source messaging and collaboration |
From @ntyniOn Tue, Mar 08, 2011 at 12:51:08PM -0800, Nicholas Clark via RT wrote:
I only added proper patchlevel.h handling a year ago in Debian package It looks like at least one supported Ubuntu version ("Lucid Lynx") has |
From quanah@zimbra.com--On March 8, 2011 9:38:02 PM +0000 Nicholas Clark <nick@ccl4.org> wrote:
root@zqa-064:~# ./inettest2 zqa-064.eng.vmware.com --Quanah -- Zimbra :: the leader in open source messaging and collaboration |
From @jmcabandaraOperations logged to /tmp/install.log.22392 |
From [Unknown Contact. See original ticket]Operations logged to /tmp/install.log.22392 |
From @tonycozOn Thu May 14 10:06:09 2015, jmcabandara@gmail.com wrote:
This doesn't appear to have anything to do with the original problem in Are you trying to report a new problem? If you're having problems installing pre-built packages supplied by Tony |
From mishikal@yahoo.comOn Thu May 14 10:06:09 2015, jmcabandara@gmail.com wrote: This indicates you installed a version of ZCS that doesn't match your operating system. I.e., you installed a RHEL6 build on RHEL7, or an Ubuntu12 build on Ubuntu14, etc. |
Migrated from rt.perl.org#85308 (status was 'open')
Searchable as RT85308$
The text was updated successfully, but these errors were encountered: