Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 20 Oct 2013 17:30:01 +0200
From: Paul Schutte <sjpschutte@...il.com>
To: musl@...ts.openwall.com
Subject: Re: inet_pton problem

Hi,

My previous attempt still left the door open for the ":192.168.1.1" case to
sneak through.

The following handled everything I could dream up correctly:

--- a/musl/src/network/inet_pton.c
+++ b/musl/src/network/inet_pton.c
@@ -14,11 +14,11 @@
        return -1;
 }

-int inet_pton(int af, const char *restrict s, void *restrict a0)
+int inet_pton(int af, const char *restrict s0, void *restrict a0)
 {
        uint16_t ip[8];
        unsigned char *a = a0;
-       const char *z;
+       const char *z,*s = s0;
        unsigned long x;
        int i, j, v, d, brk=-1, need_v4=0;

@@ -73,6 +73,10 @@
                *a++ = ip[j]>>8;
                *a++ = ip[j];
        }
+
+       /* IPv4 dotted-quad should have valid IPv6 in front*/
+       if ((s-s0) <2) return 0;
+
        if (need_v4 && inet_pton(AF_INET, (void *)s, a-4) <= 0) return 0;
        return 1;
 }



On Sun, Oct 20, 2013 at 2:26 PM, Paul Schutte <sjpschutte@...il.com> wrote:

> Hi Rich,
>
> Unfortunately this is not the complete fix.
>
> Haproxy still complains about invalid networks.
>
> The following seems to fix the problem without adding too much bloat:
>
> --- a/musl/src/network/inet_pton.c
> +++ b/musl/src/network/inet_pton.c
> @@ -14,11 +14,11 @@
>         return -1;
>  }
>
> -int inet_pton(int af, const char *restrict s, void *restrict a0)
> +int inet_pton(int af, const char *restrict s0, void *restrict a0)
>  {
>         uint16_t ip[8];
>         unsigned char *a = a0;
> -       const char *z;
> +       const char *z,*s = s0;
>         unsigned long x;
>         int i, j, v, d, brk=-1, need_v4=0;
>
> @@ -73,6 +73,10 @@
>                 *a++ = ip[j]>>8;
>                 *a++ = ip[j];
>         }
> +
> +       /* There must have been valid IPv6 preceding IPv4 dotted-quad */
> +       if (s==s0) return 0;
> +
>         if (need_v4 && inet_pton(AF_INET, (void *)s, a-4) <= 0) return 0;
>         return 1;
>  }
>
>
>
> Regards
> Paul
>
>
>
> On Sun, Oct 20, 2013 at 11:50 AM, Paul Schutte <sjpschutte@...il.com>wrote:
>
>> Hi Rich,
>>
>> I agree with you, especially about the bloat part.
>>
>> They (haproxy) actually use this function to determine whether the
>> address they have is a valid IPv6 address.
>> They pass in either a valid IPv4 or IPv6 address and then rely on this
>> function to determine which they have (assuming a return value of 0).
>>
>> After reading the spec more carefully I realise that -1 should be
>> returned only when the address family is not AF_INET or AF_INET6.
>>
>> By changing the return value in the IPv6 code to 0 instead of -1, we
>> could get the correct behaviour without any extra code.
>>
>> Here is a patch to try and save you a bit of work:
>>
>> --- a/musl/src/network/inet_pton.c
>> +++ b/musl/src/network/inet_pton.c
>> @@ -46,24 +46,24 @@
>>                         if (!s[1]) break;
>>                         continue;
>>                 }
>> -               if (hexval(s[0])<0) return -1;
>> +               if (hexval(s[0])<0) return 0;
>>                 while (s[0]=='0' && s[1]=='0') s++;
>>                 for (v=j=0; j<5 && (d=hexval(s[j]))>=0; j++)
>>                         v=16*v+d;
>> -               if (v > 65535) return -1;
>> +               if (v > 65535) return 0;
>>                 ip[i] = v;
>>                 if (!s[j]) {
>> -                       if (brk<0 && i!=7) return -1;
>> +                       if (brk<0 && i!=7) return 0;
>>                         break;
>>                 }
>>                 if (i<7) {
>>                         if (s[j]==':') continue;
>> -                       if (s[j]!='.') return -1;
>> +                       if (s[j]!='.') return 0;
>>                         need_v4=1;
>>                         i++;
>>                         break;
>>                 }
>> -               return -1;
>> +               return 0;
>>         }
>>         if (brk>=0) {
>>                 memmove(ip+brk+7-i, ip+brk, 2*(i+1-brk));
>> @@ -73,6 +73,6 @@
>>                 *a++ = ip[j]>>8;
>>                 *a++ = ip[j];
>>         }
>> -       if (need_v4 &&inet_pton(AF_INET, (void *)s, a-4) <= 0) return -1;
>> +       if (need_v4 &&inet_pton(AF_INET, (void *)s, a-4) <= 0) return 0;
>>         return 1;
>>  }
>>
>> Regards
>> Paul
>>
>>
>>
>> On Sun, Oct 20, 2013 at 4:22 AM, Rich Felker <dalias@...ifal.cx> wrote:
>>
>>> On Sat, Oct 19, 2013 at 10:57:00PM +0200, Paul Schutte wrote:
>>> > Hi,
>>> >
>>> > I came across this and believe it is a bug.
>>> >
>>> > I have found that when you set str to an IPv4 addr of the from
>>> > "xxx.xxx.xxx.xxx' while the address family is AF_INET6, then instead of
>>> > returning a 0 to indicate an invalid IPv6 string, it is converted to
>>> > gibberish.
>>>
>>> From what I can tell, it's not converted to gibberish; instead, it's
>>> wrongly returning an error (-1) instead of a result indicating an
>>> invalid input string (0). One could argue that it's a programming
>>> error not to check this, but inet_pton should not have any reason to
>>> return -1 if the first argument (af) is valid, so one could also argue
>>> that such checks would be extraneous bloat.
>>>
>>> > inet_pton(AF_INET6, "192.168.1.1', &sa) should return 0 if I
>>> understand the
>>> > specification correctly.
>>>
>>> Agreed.
>>>
>>> Rich
>>>
>>
>>
>

Content of type "text/html" skipped

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.