Windows NT/2000/XP下不用驱动的Ring0代码实现
WebCrazy(http://webcrazy.yeah.net/)
大家知道，Windows NT/2000为实现其可靠性，严格将系统划分为内核模式与用户模式，在i386系统中分别对应CPU的Ring0与Ring3级别。Ring0下，可以执行特权级指令，对任何I/O设备都有访问权等等。要实现从用户态进入核心态，即从Ring 3进入Ring 0必须借助CPU的某种门机制，如中断门、调用门等。而Windows NT/2000提供用户态执行系统服务(Ring 0例程)的此类机制即System Service的int 2eh中断服务等，严格的参数检查，只能严格的执行Windows NT/2000提供的服务，而如果想执行用户提供的Ring 0代码(指运行在Ring 0权限的代码)，常规方法似乎只有编写设备驱动程序。本文将介绍一种在用户态不借助任何驱动程序执行Ring0代码的方法。
Windows NT/2000将设备驱动程序调入内核区域(常见的位于地址0x80000000上)，由DPL为0的GDT项8，即cs为8时实现Ring 0权限。本文通过在系统中构造一个指向我们的代码的调用门(CallGate)，实现Ring0代码。基于这个思路，为实现这个目的主要是构造自己的CallGate。CallGate由系统中叫Global Descriptor Table(GDT)的全局表指定。GDT地址可由i386指令sgdt获得(sgdt不是特权级指令，普通Ring 3程序均可执行)。GDT地址在Windows NT/2000保存于KPCR(Processor Control Region)结构中(见《再谈Windows NT/2000环境切换》)。GDT中的CallGate是如下的格式：
typedef
struct
{
unsigned short offset_0_15;
unsigned short selector;
unsigned char param_count : 4;
unsigned char some_bits : 4;
unsigned char type : 4;
unsigned char app_system : 1;
unsigned char dpl : 2;
unsigned char present : 1;
unsigned short offset_16_31;
} CALLGATE_DESCRIPTOR;
GDT位于内核区域，一般用户态的程序是不可能对这段内存区域有直接的访问权。幸运的是Windows NT/2000提供了一个叫PhysicalMemory的Section内核对象位于\Device的路径下。顾名思义，通过这个Section对象可以对物理内存进行操作。用objdir.exe对这个对象分析如下：
C:\NTDDK\bin>objdir /D \Device
PhysicalMemory
Section
DACL -
Ace[ 0] - Grant - 0xf001f - NT AUTHORITY\SYSTEM
Inherit:
Access: 0x001F and ( D RCtl WOwn WDacl )
Ace[ 1] - Grant - 0x2000d - BUILTIN\Administrators
Inherit:
Access: 0x000D and ( RCtl )
从dump出的这个对象DACL的Ace可以看出默认情况下只有SYSTEM用户才有对这个对象的读写权限，即对物理内存有读写能力，而Administrator只有读权限，普通用户根本就没有权限。不过如果我们有Administrator权限就可以通过GetSecurityInfo、SetEntriesInAcl与SetSecurityInfo这些API来修改这个对象的ACE。这也是我提供的代码需要Administrator的原因。实现的代码如下：
VOID SetPhyscialMemorySectionCanBeWrited(HANDLE hSection)
{
PACL pDacl=NULL;
PACL pNewDacl=NULL;
PSECURITY_DESCRIPTOR pSD=NULL;
DWORD dwRes;
EXPLICIT_ACCESS ea;
if(dwRes=GetSecurityInfo(hSection,SE_KERNEL_OBJECT,DACL_SECURITY_INFORMATION,
NULL,NULL,&pDacl,NULL,&pSD)!=ERROR_SUCCESS)
{
printf( "GetSecurityInfo Error %u\n", dwRes );
goto CleanUp;
}
ZeroMemory(&ea, sizeof(EXPLICIT_ACCESS));
ea.grfAccessPermissions = SECTION_MAP_WRITE;
ea.grfAccessMode = GRANT_ACCESS;
ea.grfInheritance= NO_INHERITANCE;
ea.Trustee.TrusteeForm = TRUSTEE_IS_NAME;
ea.Trustee.TrusteeType = TRUSTEE_IS_USER;
ea.Trustee.ptstrName = "CURRENT_USER";
if(dwRes=SetEntriesInAcl(1,&ea,pDacl,&pNewDacl)!=ERROR_SUCCESS)
{
printf( "SetEntriesInAcl %u\n", dwRes );
goto CleanUp;
}
if(dwRes=SetSecurityInfo(hSection,SE_KERNEL_OBJECT,DACL_SECURITY_INFORMATION,NULL,NULL,pNewDacl,NULL)!=ERROR_SUCCESS)
{
printf("SetSecurityInfo %u\n",dwRes);
goto CleanUp;
}
CleanUp:
if(pSD)
LocalFree(pSD);
if(pNewDacl)
LocalFree(pSD);
}
这段代码对给定HANDLE的对象增加了如下的ACE:
PhysicalMemory
Section
DACL -
Ace[ 0] - Grant - 0x2 - WEBCRAZY\Administrator
Inherit:
Access: 0x0002
//
SECTION_MAP_WRITE
这样我们在有Administrator权限的条件下就有了对物理内存的读写能力。但若要修改GDT表实现Ring 0代码。我们将面临着另一个难题，因为sgdt指令获得的GDT地址是虚拟地址(线性地址)，我们只有知道GDT表的物理地址后才能通过\Device\PhysicalMemory对象修改GDT表，这就牵涉到了线性地址转化成物理地址的问题。我们先来看一看Windows NT/2000是如何实现这个的：
kd> u nt!MmGetPhysicalAddress l 30
ntoskrnl!MmGetPhysicalAddress:
801374e0 56 push esi
801374e1 8b742408 mov esi,[esp+0x8]
801374e5 33d2 xor edx,edx
801374e7 81fe00000080 cmp esi,0x80000000
801374ed 722c jb ntoskrnl!MmGetPhysicalAddress+0x2b (8013751b)
801374ef 81fe000000a0 cmp esi,0xa0000000
801374f5 7324 jnb ntoskrnl!MmGetPhysicalAddress+0x2b (8013751b)
801374f7 39153ce71780 cmp [ntoskrnl!MmKseg2Frame (8017e73c)],edx
801374fd 741c jz ntoskrnl!MmGetPhysicalAddress+0x2b (8013751b)
801374ff 8bc6 mov eax,esi
80137501 c1e80c shr eax,0xc
80137504 25ffff0100 and eax,0x1ffff
80137509 6a0c push 0xc
8013750b 59 pop ecx
8013750c e8d3a7fcff call ntoskrnl!_allshl (80101ce4)
80137511 81e6ff0f0000 and esi,0xfff
80137517 03c6 add eax,esi
80137519 eb17 jmp ntoskrnl!MmGetPhysicalAddress+0x57 (80137532)
8013751b 8bc6 mov eax,esi
8013751d c1e80a shr eax,0xa
80137520 25fcff3f00 and eax,0x3ffffc
80137525 2d00000040 sub eax,0x40000000
8013752a 8b00 mov eax,[eax]
8013752c a801 test al,0x1
8013752e 7506 jnz ntoskrnl!MmGetPhysicalAddress+0x44 (80137536)
80137530 33c0 xor eax,eax
80137532 5e pop esi
80137533 c20400 ret 0x4
从这段汇编代码可看出如果线性地址在0x80000000与0xa0000000范围内，只是简单的进行移位操作(位于801374ff-80137519指令间)，并未查页表。我想Microsoft这样安排肯定是出于执行效率的考虑。这也为我们指明了一线曙光，因为GDT表在Windows NT/2000中一般情况下均位于这个区域(我不知道/3GB开关的Windows NT/2000是不是这种情况)。
经过这样的分析，我们就可以只通过用户态程序修改GDT表了。而增加一个CallGate就不是我可以介绍的了，找本Intel手册自己看一看了。具体实现代码如下：
typedef
struct gdtr
{
short Limit;
short BaseLow;
short BaseHigh;
} Gdtr_t, *PGdtr_t;
ULONG MiniMmGetPhysicalAddress(ULONG virtualaddress)
{
if(virtualaddress<0x80000000||virtualaddress>=0xA0000000)
return 0;
return virtualaddress&0x1FFFF000;
}
BOOL ExecRing0Proc(ULONG Entry,ULONG seglen)
{
Gdtr_t gdt;
__asm sgdt gdt;
ULONG mapAddr=MiniMmGetPhysicalAddress(gdt.BaseHigh<<16U|gdt.BaseLow);
if(!mapAddr) return 0;
HANDLE hSection=NULL;
NTSTATUS status;
OBJECT_ATTRIBUTES objectAttributes;
UNICODE_STRING objName;
CALLGATE_DESCRIPTOR *cg;
status = STATUS_SUCCESS;
RtlInitUnicodeString(&objName,L"\\Device\\PhysicalMemory");
InitializeObjectAttributes(&objectAttributes,
&objName,
OBJ_CASE_INSENSITIVE | OBJ_KERNEL_HANDLE,
NULL,
(PSECURITY_DESCRIPTOR) NULL);
status = ZwOpenSection(&hSection,SECTION_MAP_READ|SECTION_MAP_WRITE,&objectAttributes);
if(status == STATUS_ACCESS_DENIED){
status = ZwOpenSection(&hSection,READ_CONTROL|WRITE_DAC,&objectAttributes);
SetPhyscialMemorySectionCanBeWrited(hSection);
ZwClose(hSection);
status =ZwOpenSection(&hSection,SECTION_MAP_WRITE|SECTION_MAP_WRITE,&objectAttributes);
}
if(status != STATUS_SUCCESS)
{
printf("Error Open PhysicalMemory Section Object,Status:%08X\n",status);
return 0;
}
PVOID BaseAddress;
BaseAddress=MapViewOfFile(hSection,
FILE_MAP_READ|FILE_MAP_WRITE,
0,
mapAddr, //low part
(gdt.Limit+1));
if(!BaseAddress)
{
printf("Error MapViewOfFile:");
PrintWin32Error(GetLastError());
return 0;
}
BOOL setcg=FALSE;
for(cg=(CALLGATE_DESCRIPTOR *)((ULONG)BaseAddress+(gdt.Limit&0xFFF8));(ULONG)cg>(ULONG)BaseAddress;cg--)
if(cg->type == 0){
cg->offset_0_15 = LOWORD(Entry);
cg->selector = 8;
cg->param_count = 0;
cg->some_bits = 0;
cg->type = 0xC; // 386 call gate
cg->app_system = 0; // A system descriptor
cg->dpl = 3; // Ring 3 code can call
cg->present = 1;
cg->offset_16_31 = HIWORD(Entry);
setcg=TRUE;
break;
}
if(!setcg){
ZwClose(hSection);
return 0;
}
short farcall[3];
farcall[2]=((short)((ULONG)cg-(ULONG)BaseAddress))|3; //Ring 3 callgate;
if(!VirtualLock((PVOID)Entry,seglen))
{
printf("Error VirtualLock:");
PrintWin32Error(GetLastError());
return 0;
}
SetThreadPriority(GetCurrentThread(),THREAD_PRIORITY_TIME_CRITICAL);
Sleep(0);
_asm call fword ptr [farcall]
SetThreadPriority(GetCurrentThread(),THREAD_PRIORITY_NORMAL);
VirtualUnlock((PVOID)Entry,seglen);
//Clear callgate
*(ULONG *)cg=0;
*((ULONG *)cg+1)=0;
ZwClose(hSection);
return TRUE;
}
我在提供的代码中演示了对Control Register与I/O端口的操作。CIH病毒在Windows 9X中就是因为获得Ring 0权限才有了一定的危害，但Windows NT/2000毕竟不是Windows 9X，她已经有了比较多的安全审核机制，本文提供的代码也要求具有Administrator权限，但如果系统存在某种漏洞，如缓冲区溢出等等，还是有可能获得这种权限的，所以我不对本文提供的方法负有任何的责任，所有讨论只是一个技术热爱者在讨论技术而已。谢谢！
参考资料：
1.Intel Corp<<Intel Architecture Software Developer's Manual,Volume 3>>
WebCrazy(http://webcrazy.yeah.net/)
大家知道，Windows NT/2000为实现其可靠性，严格将系统划分为内核模式与用户模式，在i386系统中分别对应CPU的Ring0与Ring3级别。Ring0下，可以执行特权级指令，对任何I/O设备都有访问权等等。要实现从用户态进入核心态，即从Ring 3进入Ring 0必须借助CPU的某种门机制，如中断门、调用门等。而Windows NT/2000提供用户态执行系统服务(Ring 0例程)的此类机制即System Service的int 2eh中断服务等，严格的参数检查，只能严格的执行Windows NT/2000提供的服务，而如果想执行用户提供的Ring 0代码(指运行在Ring 0权限的代码)，常规方法似乎只有编写设备驱动程序。本文将介绍一种在用户态不借助任何驱动程序执行Ring0代码的方法。
Windows NT/2000将设备驱动程序调入内核区域(常见的位于地址0x80000000上)，由DPL为0的GDT项8，即cs为8时实现Ring 0权限。本文通过在系统中构造一个指向我们的代码的调用门(CallGate)，实现Ring0代码。基于这个思路，为实现这个目的主要是构造自己的CallGate。CallGate由系统中叫Global Descriptor Table(GDT)的全局表指定。GDT地址可由i386指令sgdt获得(sgdt不是特权级指令，普通Ring 3程序均可执行)。GDT地址在Windows NT/2000保存于KPCR(Processor Control Region)结构中(见《再谈Windows NT/2000环境切换》)。GDT中的CallGate是如下的格式：
typedef
struct
{ unsigned short offset_0_15; unsigned short selector; unsigned char param_count : 4; unsigned char some_bits : 4; unsigned char type : 4; unsigned char app_system : 1; unsigned char dpl : 2; unsigned char present : 1; unsigned short offset_16_31; } CALLGATE_DESCRIPTOR;
C:\NTDDK\bin>objdir /D \Device
PhysicalMemory
Section
DACL -
Ace[ 0] - Grant - 0xf001f - NT AUTHORITY\SYSTEM
Inherit:
Access: 0x001F and ( D RCtl WOwn WDacl )
Ace[ 1] - Grant - 0x2000d - BUILTIN\Administrators
Inherit:
Access: 0x000D and ( RCtl )
VOID SetPhyscialMemorySectionCanBeWrited(HANDLE hSection)
{ PACL pDacl=NULL; PACL pNewDacl=NULL; PSECURITY_DESCRIPTOR pSD=NULL; DWORD dwRes; EXPLICIT_ACCESS ea; if(dwRes=GetSecurityInfo(hSection,SE_KERNEL_OBJECT,DACL_SECURITY_INFORMATION, NULL,NULL,&pDacl,NULL,&pSD)!=ERROR_SUCCESS) { printf( "GetSecurityInfo Error %u\n", dwRes ); goto CleanUp; } ZeroMemory(&ea, sizeof(EXPLICIT_ACCESS)); ea.grfAccessPermissions = SECTION_MAP_WRITE; ea.grfAccessMode = GRANT_ACCESS; ea.grfInheritance= NO_INHERITANCE; ea.Trustee.TrusteeForm = TRUSTEE_IS_NAME; ea.Trustee.TrusteeType = TRUSTEE_IS_USER; ea.Trustee.ptstrName = "CURRENT_USER"; if(dwRes=SetEntriesInAcl(1,&ea,pDacl,&pNewDacl)!=ERROR_SUCCESS) { printf( "SetEntriesInAcl %u\n", dwRes ); goto CleanUp; } if(dwRes=SetSecurityInfo(hSection,SE_KERNEL_OBJECT,DACL_SECURITY_INFORMATION,NULL,NULL,pNewDacl,NULL)!=ERROR_SUCCESS) { printf("SetSecurityInfo %u\n",dwRes); goto CleanUp; } CleanUp: if(pSD) LocalFree(pSD); if(pNewDacl) LocalFree(pSD); }
这段代码对给定HANDLE的对象增加了如下的ACE:
PhysicalMemory
Section
DACL -
Ace[ 0] - Grant - 0x2 - WEBCRAZY\Administrator
Inherit:
Access: 0x0002
//
SECTION_MAP_WRITE
这样我们在有Administrator权限的条件下就有了对物理内存的读写能力。但若要修改GDT表实现Ring 0代码。我们将面临着另一个难题，因为sgdt指令获得的GDT地址是虚拟地址(线性地址)，我们只有知道GDT表的物理地址后才能通过\Device\PhysicalMemory对象修改GDT表，这就牵涉到了线性地址转化成物理地址的问题。我们先来看一看Windows NT/2000是如何实现这个的：
kd> u nt!MmGetPhysicalAddress l 30
ntoskrnl!MmGetPhysicalAddress:
801374e0 56 push esi
801374e1 8b742408 mov esi,[esp+0x8]
801374e5 33d2 xor edx,edx
801374e7 81fe00000080 cmp esi,0x80000000
801374ed 722c jb ntoskrnl!MmGetPhysicalAddress+0x2b (8013751b)
801374ef 81fe000000a0 cmp esi,0xa0000000
801374f5 7324 jnb ntoskrnl!MmGetPhysicalAddress+0x2b (8013751b)
801374f7 39153ce71780 cmp [ntoskrnl!MmKseg2Frame (8017e73c)],edx
801374fd 741c jz ntoskrnl!MmGetPhysicalAddress+0x2b (8013751b)
801374ff 8bc6 mov eax,esi
80137501 c1e80c shr eax,0xc
80137504 25ffff0100 and eax,0x1ffff
80137509 6a0c push 0xc
8013750b 59 pop ecx
8013750c e8d3a7fcff call ntoskrnl!_allshl (80101ce4)
80137511 81e6ff0f0000 and esi,0xfff
80137517 03c6 add eax,esi
80137519 eb17 jmp ntoskrnl!MmGetPhysicalAddress+0x57 (80137532)
8013751b 8bc6 mov eax,esi
8013751d c1e80a shr eax,0xa
80137520 25fcff3f00 and eax,0x3ffffc
80137525 2d00000040 sub eax,0x40000000
8013752a 8b00 mov eax,[eax]
8013752c a801 test al,0x1
8013752e 7506 jnz ntoskrnl!MmGetPhysicalAddress+0x44 (80137536)
80137530 33c0 xor eax,eax
80137532 5e pop esi
80137533 c20400 ret 0x4
经过这样的分析，我们就可以只通过用户态程序修改GDT表了。而增加一个CallGate就不是我可以介绍的了，找本Intel手册自己看一看了。具体实现代码如下：
typedef
struct gdtr
{ short Limit; short BaseLow; short BaseHigh; } Gdtr_t, *PGdtr_t;
ULONG MiniMmGetPhysicalAddress(ULONG virtualaddress)
{ if(virtualaddress<0x80000000||virtualaddress>=0xA0000000) return 0; return virtualaddress&0x1FFFF000; }
BOOL ExecRing0Proc(ULONG Entry,ULONG seglen)
{ Gdtr_t gdt; __asm sgdt gdt; ULONG mapAddr=MiniMmGetPhysicalAddress(gdt.BaseHigh<<16U|gdt.BaseLow); if(!mapAddr) return 0; HANDLE hSection=NULL; NTSTATUS status; OBJECT_ATTRIBUTES objectAttributes; UNICODE_STRING objName; CALLGATE_DESCRIPTOR *cg; status = STATUS_SUCCESS; RtlInitUnicodeString(&objName,L"\\Device\\PhysicalMemory"); InitializeObjectAttributes(&objectAttributes, &objName, OBJ_CASE_INSENSITIVE | OBJ_KERNEL_HANDLE, NULL, (PSECURITY_DESCRIPTOR) NULL); status = ZwOpenSection(&hSection,SECTION_MAP_READ|SECTION_MAP_WRITE,&objectAttributes); if(status == STATUS_ACCESS_DENIED){ status = ZwOpenSection(&hSection,READ_CONTROL|WRITE_DAC,&objectAttributes); SetPhyscialMemorySectionCanBeWrited(hSection); ZwClose(hSection); status =ZwOpenSection(&hSection,SECTION_MAP_WRITE|SECTION_MAP_WRITE,&objectAttributes); } if(status != STATUS_SUCCESS) { printf("Error Open PhysicalMemory Section Object,Status:%08X\n",status); return 0; } PVOID BaseAddress; BaseAddress=MapViewOfFile(hSection, FILE_MAP_READ|FILE_MAP_WRITE, 0, mapAddr, //low part (gdt.Limit+1)); if(!BaseAddress) { printf("Error MapViewOfFile:"); PrintWin32Error(GetLastError()); return 0; } BOOL setcg=FALSE; for(cg=(CALLGATE_DESCRIPTOR *)((ULONG)BaseAddress+(gdt.Limit&0xFFF8));(ULONG)cg>(ULONG)BaseAddress;cg--) if(cg->type == 0){ cg->offset_0_15 = LOWORD(Entry); cg->selector = 8; cg->param_count = 0; cg->some_bits = 0; cg->type = 0xC; // 386 call gate cg->app_system = 0; // A system descriptor cg->dpl = 3; // Ring 3 code can call cg->present = 1; cg->offset_16_31 = HIWORD(Entry); setcg=TRUE; break; } if(!setcg){ ZwClose(hSection); return 0; } short farcall[3]; farcall[2]=((short)((ULONG)cg-(ULONG)BaseAddress))|3; //Ring 3 callgate; if(!VirtualLock((PVOID)Entry,seglen)) { printf("Error VirtualLock:"); PrintWin32Error(GetLastError()); return 0; } SetThreadPriority(GetCurrentThread(),THREAD_PRIORITY_TIME_CRITICAL); Sleep(0); _asm call fword ptr [farcall] SetThreadPriority(GetCurrentThread(),THREAD_PRIORITY_NORMAL); VirtualUnlock((PVOID)Entry,seglen); //Clear callgate *(ULONG *)cg=0; *((ULONG *)cg+1)=0; ZwClose(hSection); return TRUE; }
参考资料：
1.Intel Corp<<Intel Architecture Software Developer's Manual,Volume 3>>
