Windows Kernel 64-bit stack memory disclosure in nt!KiDispatchException(CVE-2018-0897)

We have discovered a new Windows kernel memory disclosure vulnerability in the creation and copying of a EXCEPTION_RECORD structure to user-mode memory while passing execution to a user-mode exception handler. The vulnerability affects 64-bit versions of Windows 7 to 10. The leak was originally detected under the following stack trace (Windows 7): ``` kd> k # Child-SP RetAddr Call Site 00 fffff880`040b7e18 fffff800`026ca362 nt!memcpy+0x3 01 fffff880`040b7e20 fffff800`026db3bc nt!KiDispatchException+0x421 02 fffff880`040b84b0 fffff800`0268fafb nt!KiRaiseException+0x1b4 03 fffff880`040b8ae0 fffff800`0268d093 nt!NtRaiseException+0x7b 04 fffff880`040b8c20 00000000`74b5cb49 nt!KiSystemServiceCopyEnd+0x13 ``` and more specifically in the copying of the EXCEPTION_RECORD structure: ``` kd> dt _EXCEPTION_RECORD @rdx ntdll!_EXCEPTION_RECORD +0x000 ExceptionCode : 0n1722 +0x004 ExceptionFlags : 1 +0x008 ExceptionRecord : (null) +0x010 ExceptionAddress : 0x00000000`765fc54f Void +0x018 NumberParameters : 0 +0x020 ExceptionInformation : [15] 0xbbbbbbbb`bbbbbbbb ``` In that structure, the entire "ExceptionInformation" array consisting of 15*8=120 bytes is left uninitialized and provided this way to the ring-3 client. The overall EXCEPTION_RECORD structure (which contains the ExceptionInformation in question) is allocated in the stack frame of the nt!KiRaiseException function. Based on some cursory code analysis and manual experimentation, we believe that the kernel only fills as many ULONG_PTR's as the .NumberParameters field is set to (but not more than EXCEPTION_MAXIMUM_PARAMETERS), while the remaining entries of the array are never written to. As a result, running the attached proof-of-concept program reveals 120 bytes of kernel stack memory (set to the 0x41 marker with stack-spraying to illustrate the problem). An example output is as follows: ``` 00000000: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000010: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000020: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000030: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000040: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000050: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000060: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00000070: 41 41 41 41 41 41 41 41 ?? ?? ?? ?? ?? ?? ?? ?? AAAAAAAA........ ``` If we replace the stack-spraying function call in the code with a printf() call, we can immediately spot a number of kernel-mode addresses in the output dump: ``` 00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000010: a0 ce 1e 00 00 00 00 00 a0 ce 1e 00 00 00 00 00 ................ 00000020: 00 00 00 00 00 00 00 00 64 0b 00 00 00 40 00 00 ........d....@.. 00000030: b8 00 00 00 00 00 00 00 60 6e 83 5d 80 f9 ff ff ........`n.].... 00000040: 00 0d 78 60 80 f9 ff ff 00 00 00 00 80 f9 ff ff ..x`............ 00000050: 00 00 00 00 01 00 00 00 00 01 00 00 00 00 00 00 ................ 00000060: 10 01 00 00 00 00 00 00 00 70 f5 3f 01 00 00 00 .........p.?.... 00000070: 50 0b 10 60 80 f9 ff ff ?? ?? ?? ?? ?? ?? ?? ?? P..`............ ``` Repeatedly triggering the vulnerability could allow local authenticated attackers to defeat certain exploit mitigations (kernel ASLR) or read other secrets stored in the kernel address space.