[Secure-testing-team] Bug#478680: [gnome-peercast] remote stack-based buffer overflow in HTTP::getAuthUserPass function

[Nico Golde]

Package: gnome-peercast
Version: 0.5.4-1.1
Severity: grave
Tags: security
X-Debbugs-CC: secure-testing-team at lists.alioth.debian.org

Hi,
I found a security issue in the peercast server which also 
affects gnome-peercast. I contacted the upstream author of 
peercast 6 days ago without any reply so far.
Here are the details:
From core/common/http.cpp:

105 void HTTP::getAuthUserPass(char *user, char *pass)
106 {
107         if (arg)
108         {   
109                 char *s = stristr(arg,"Basic");
110                 if (s) 
111                 {   
112                         while (*s)
113                                 if (*s++ == ' ')
114                                         break;
115                         String str;
116                         str.set(s,String::T_BASE64);
117                         str.convertTo(String::T_ASCII);
118                         s = strstr(str.cstr(),":");
119                         if (s) 
120                         {   
121                                 *s = 0;
122                                 if (user)
123                                         strcpy(user,str.cstr());
124                                 if (pass)
125                                         strcpy(pass,s+1);

This function is used if authentication to the gnome-peercast server is done by basic http auth
which is the case in the standard configuration of gnome-peercast.
In line 116 the base64 encoded string is copied into str.
Note the set method is peercasts/gnome-peercasts own implementation of set
since it reimplements the String class. set looks like this:

From core/common/sys.h:
38                 MAX_LEN = 256 
...
62         void set(const char *p, TYPE t=T_ASCII)
63         {   
64                 strncpy(data,p,MAX_LEN-1);
65                 data[MAX_LEN-1] = 0;
66                 type = t;
67         }   

In line 117 the string gets decoded and in line 118 and 
following the part before ':' in the decoded string gets copied
into user and the part after it into pass.

From core/common/servhs.cpp:
558 bool Servent::handshakeAuth(HTTP &http,const char *args,bool local)
559 {
560         char user[64],pass[64];
561         user[0] = pass[0] = 0;
...
580     while (http.nextHeader())
581         {   
582                 char *arg = http.getArgStr();
583                 if (!arg)
584                         continue;
585
586                 switch (servMgr->authType)
587                 {   
588                         case ServMgr::AUTH_HTTPBASIC:
589                                 if (http.isHeader("Authorization"))
590                                         http.getAuthUserPass(user,pass);
591                                 break;

user and pass are only declared to have 64 bytes (line 558) while the buffer used for 
copy can store up to MAX_LEN (256) bytes (ok minus the : here). Servent::handshakeAuth calls then
the getAuthUserPass function triggering a buffer overflow.
It's thus possible to crash the server and execute arbitrary code if the server
allows http-basic authentication.

I already requested a CVE id for this.

PoC attached.
Kind regards
Nico

-- 
Nico Golde - http://www.ngolde.de - nion at jabber.ccc.de - GPG: 0x73647CFF
For security reasons, all text in this mail is double-rot13 encrypted.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: peercast.py
Type: text/x-python
Size: 435 bytes
Desc: not available
Url : http://lists.alioth.debian.org/pipermail/secure-testing-team/attachments/20080430/8558f434/attachment.py 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.alioth.debian.org/pipermail/secure-testing-team/attachments/20080430/8558f434/attachment.pgp