Re: Windows Oday release

[ge () linuxbox org]

On 2007-06-13 02:58+0800, Thomas Lim wrote:
> dear all

Dear all, this is not a 0day, it is a public release of a responsibly
disclosed vulnerability.

Thank you for sharing your research,

        Gadi.

> SChannel Off-By-One Heap Corruption
> ===================================
>
> Discovery Date:
> 28th August 2006
>
> Date reported to Microsoft:
> 19th March 2007
>
> Summary:
> The Secure Channel (SChannel) library on WinXP-SP1/SP2 is vulnerable to
> a off-by-one heap buffer overwrite. The SChannel library implements
> PCT/TLS/SSL protocols exported via the Security Service Provider Interface
> (SSPI). It is one of several Security Service Providers loaded-in and 
> supported
> by the privileged Local Security Authority (Lsass.exe) process.
>
> In SChannel's implementation of the client-side SSLv3 handshake protocol,
> specifically in the processing of the server-key-exchange SSL handshake
> record, there is insufficient checks for malformed server-sent digital 
> signature,
> with its length-field set to 0. This results in a allocation of a 
> 0-length heap
> buffer (with a valid heap address). A reverse memory copy is then 
> performed to
> copy-in the digital signature, by decrementing the 0-length by 1. This 
> results in
> an integer-underflow, causing the heap-buffer pointer to decrement 
> before its start
> address, ultimately leading to an overwrite of exactly one-byte of 
> user-controlled
> value, into the heap control-block. Depending on the robustness of the 
> application
> in question, this may lead to an unrecoverable heap corruption 
> condition, causing
> the application to terminate. In the case of Lsass.exe on WinXP-SP2, we 
> can crash
> it locally after several iterations, from a less-privileged user, 
> causing a system
> reboot. Vulnerable code although also exists in WinXP-SP1 but it does 
> not cause an
> unrecoverable heap corruption in Lsass.exe.
>
> Vendor Affected:
> Microsoft
>
>
> Systems Affected:
> ========
> WinXP-SP2 (DOS/Reboot)
> WinXP-SP1 (minimal impact)
>
>
> Exploitation:
> =============
> 1) For local machine reboot via normal user account, on WinXP-SP2
>   OR
>   For remote machine reboot by enticing user to visit HTTPS site via IE,
>   on WinXP-SP2 (but over 2000 iterations required)
>     
>     
> POC (crash-test/reboot):
> ========================
> 1) Run sctest.exe from a normal user account, on client machine
>   running WinXP-SP2.
>  
> 2) sctest.exe will attempt to use SChannel's SSL implementation to
>   parse pre-generated malformed SSL handshake records, over
>   several iterations, causing multiple off-by-one overwrites with
>   0xFF byte, within the Lsass.exe process.
>
> 3) Attach Debugger to Lsass.exe to see crash. The system will notify
>   the user and perform a 60sec. reboot count-down, after detecting
>   the Lsass.exe crash.
>    
> ** Lsass.exe crash-test can also be done by forcing/enticing Internet 
> Explorer
>   to access a HTTPS site, serving out the same malformed SSL handshake
>   records (as shown in source code below). However, over 2000 iterations
>   are needed (IE needs to access HTTPS site over 2000 times), before 
> Lsass.exe
>   heap corruption occurs.
>
>          
> Vuln Analysis:
> ==============
> (Based on schannel.dll/v5.1.2600.2180/WinXP-SP2)
> The vulnerability exists in schannel.dll component, that implements the 
> SSPI-compliant
> PCT/TLS/SSL protocol handling implementation. For more information on 
> SSPI and
> how it relates to LSA, refer to 
> 1) 
> http://msdn.microsoft.com/library/default.asp?url=/library/en-us/secauthn/security/authentication_packages.asp
> 2) 
> http://msdn.microsoft.com/library/default.asp?url=/library/en-us/secauthn/security/sspi.asp
>
> Essentially, in the case of SSPI authentication libraries like schannel, 
> kerberos, msv1_0 (ntlm),
> data is exchanged between less-privileged user applications requring 
> authentication, and Lsass.exe.
> With LSA providing the authentication back-end support. Both LSA and the 
> less-privileged application
> communicate indirectly via the SSPI interface. Specifically, in SSL 
> authentication, untrusted SSL
> record packets are passed from the less-privileged application to the 
> privileged LSA. While extensive
> efforts are made in LSA to validate the SSL records, on WinXP's version 
> of schannel, an off-by-one
> vulnerability exists in the parsing of the less-common and less-used SSL 
> server-key-exchange record.
> The vulnerability can hence be triggered via less-privileged client 
> applications utilizing the
> schannel's client-side SSL protocol implementation. This includes 
> Internet Explorer, whenever the user
> uses IE to browse a HTTPS site.
>
> The vulnerable code exists in the _ReverseMemCopy() function and is 
> reachable from Ssl3ParseServerKeyExchange():
> (via SPProcessHandshake()->PkcsGenerateClientExchangeValue())
>
> ; On WinXP-SP1, the code below is located at 0x767FF976 (no symbols 
> available)
> Ssl3ParseServerKeyExchange()
> ...                                                           
> .text:767FFFC8                 movzx   ebx, byte ptr [esi]      ; 
> MSB-byte of malformed signature length field               
> .text:767FFFCB                 movzx   eax, byte ptr [esi+1]    ; 
> LSB-byte of malformed signature length field
> .text:767FFFCF                 shl     ebx, 8
> .text:767FFFD2                 add     ebx, eax
> .text:767FFFD4                 push    ebx                      ; size=0
> .text:767FFFD5                 call    _SPExternalAlloc@4       ; 
> HeapAlloc will return a valid 0-length heap buffer address
> .text:767FFFDA                 test    eax, eax
> .text:767FFFDC                 mov     [ebp+pbSignature], eax
> .text:767FFFDF                 jz      loc_768000B9
> .text:767FFFE5                 push    ebx                      ; size=0
> .text:767FFFE6                 lea     ecx, [esi+2]             ; 
> address of the signature data in our malformed record
>                                                                ; 
> containing 0xFF,0x41,0x41...
> .text:767FFFE9                 push    ecx
> .text:767FFFEA                 push    eax                      ; 
> 0-length heap buffer
> .text:767FFFEB                 call    _ReverseMemCopy@12
>        _ReverseMemCopy()
>        .text:767FF46F                 mov     edi, edi
>        .text:767FF471                 push    ebp
>        .text:767FF472                 mov     ebp, esp
>        .text:767FF474                 mov     eax, [ebp+arg_8]
>        .text:767FF477                 mov     ecx, [ebp+arg_4]
>        .text:767FF47A                 push    esi
>        .text:767FF47B                 mov     esi, [ebp+arg_0]
>        .text:767FF47E                 lea     eax, [esi+eax-1]         
> ; EAX=0, ESI which points to 0-length heap buffer
>                                                                        
> ; is decremented to, before start of heap buffer
>        .text:767FF482                 mov     dl, [ecx]
>        .text:767FF484                 mov     [eax], dl                
> ; Off-by-one overwrite with 0xFF from our signature data
>        .text:767FF486                 dec     eax
>        .text:767FF487                 inc     ecx
>        .text:767FF488                 cmp     eax, esi
>        .text:767FF48A                 jnb     short loc_767FF482       
> ; Just one-byte overwrite!     
>        .text:767FF48C                 pop     esi
>        .text:767FF48D                 pop     ebp
>        .text:767FF48E                 retn    0Ch
>
>
> Discovered by:
> Steven
> Security Researcher
> Vulnerability Research Lab
> COSEINC
>
> --
> Thank you
> Thomas Lim
> COSEINC Private Limited
>
> -----BEGIN PGP PUBLIC KEY BLOCK-----
> Version: PGP 8.1 - not licensed for commercial use: www.pgp.com
>
> mQGiBEQM9cARBADvlIe8Ck5/u2EtX3ikd/eKjI7uZKyIFHNLxEYBB1AaHmEvYCPi
> VpvNr7ArKjbqlEpdsl6c9gQUY8vir5Lfk/p6siCD1aIYfCdPa64gKJQ66UVIUy7a
> hIlE8sJ86mcbvVGzA4f1LjwPUPwymeKEQeDJyRLlRnPkxWzaoiZqHuEa/QCg/2/t
> IAlQdVT7Q+ss51/NcL87RoEEANcf+ChnlH6vhXLSwnH1iXUMBbGA6t2F0/q29ROR
> lsMoUQW5hvjuOw+4yDzGzmBDQUYbN0GI7pNOBs7UwerGOInTGCFD6nan0JpONT51
> bp5sfF93PNH12I1qVFf+h/qdX4me2mhyKfSNvc0qQMydwfsCJ3vBbEWTF7CqWZFO
> VadVA/9uJTKjJ7ZnN1enBBGUhLl6bA9estqH6lyP69B6Y1tGahDSqVyDe9Q9zs0T
> XDcM6aS+PRnybzX9gfgPfSYtDzX3AU6C7N2XgSK5DnjVZVr2Tdd/2ttM7ApvzaeV
> +ifO/nLGIQ38ik7mKlul5vlXsISShzHpUIdswuQtMp0R2sa+6bQfVGhvbWFzIExp
> bSA8dGhvbWFzQGNvc2VpbmMuY29tPokAXQQQEQIAHQUCRAz1wAcLCQgHAwIKAhkB
> BRsDAAAABR4BAAAAAAoJELxffA89J0fkz+cAn3cklzVq/VYiD9wgH0J2ULsuTbMl
> AJ9NMdYJHBlunYjbPJIcRgGwhAkY4LkCDQREDPXAEAgA9kJXtwh/CBdyorrWqULz
> Bej5UxE5T7bxbrlLOCDaAadWoxTpj0BV89AHxstDqZSt90xkhkn4DIO9ZekX1KHT
> UPj1WV/cdlJPPT2N286Z4VeSWc39uK50T8X8dryDxUcwYc58yWb/Ffm7/ZFexwGq
> 01uejaClcjrUGvC/RgBYK+X0iP1YTknbzSC0neSRBzZrM2w4DUUdD3yIsxx8Wy2O
> 9vPJI8BD8KVbGI2Ou1WMuF040zT9fBdXQ6MdGGzeMyEstSr/POGxKUAYEY18hKcK
> ctaGxAMZyAcpesqVDNmWn6vQClCbAkbTCD1mpF1Bn5x8vYlLIhkmuquiXsNV6TIL
> OwACAggAwTip4JFx4LCDazFSyTG7qzIlZonEf3QTHNH4jP15CLvVFxjaHE8g2EgL
> pt2+E6XDg7IGuZ2iXS9gwHkyLKzGR4bwpanAHyMZZbcQOglPHUkxuJZW+AjfcfOD
> 5jB+cUOtxk97ca/z9Fz+2qS8Q3sz2QSkHcZgBBxTS07cvd2P60ecVECBKG+dgxGw
> X13e5hgw2tzFRMqnty66lKXYEIUj4ZSl70UPAmy5xUaU4EahLURN29f3zM+EPy72
> 374v28ud28yO59iyRqoUGiHr3c87wumrDtjwm8qKIkWHsi+7AiES29nCqtm4zN45
> 3yXkalvQ/O97ZJSinzZb851ToowyFIkATAQYEQIADAUCRAz1wAUbDAAAAAAKCRC8
> X3wPPSdH5MLbAKCIYtkvUPIoxm15I4UlvCIZjT8hdACdEWiQKWdlwZCJTePk4CF9
> swkS3cQ=
> =Q3SR
> -----END PGP PUBLIC KEY BLOCK-----

--
--
"beepbeep it, i leave work, stop reading sec lists and im still hearing
gadi"
- HD Moore to Gadi Evron on IM, on Gadi's interview on npr, March 2007.