Home

Awesome

kfile-over-icmp

Note: this LKM was developed using linux kernel version 4.4.10 and 4.98.13. Should you have problems compiling this module, please check for your kernel version first.

kfile-over-icmp is a loadable kernel module for stealth sending of files over ICMP communication.

Features

Installation

$ git clone https://github.com/spiderpig1297/kfile-over-icmp.git
$ cd kfile-over-icmp
$ make
$ make install

To uninstall the module, run:

$ make remove

Usage

Once installed, the module registers a character device from which it reads the paths of the files to send. Run dmesg to retrieve the major number of the device, and then create a node to it using mknod:

$ dmesg | tail

[18850.002913] kfile-over-icmp: stopped payload generator thread
[18850.002914] kfile-over-icmp: LKM unloaded successfully
[18852.059661] kfile-over-icmp: LKM loaded
[18852.059713] kfile-over-icmp: registered netfilter hook
[18852.059715] kfile-over-icmp: successfully registered character device. major: 243

$ mknod /dev/readfile c <MAJOR_GOES_HERE> 0

In order to send a file over ICMP, write its (absolute!) path to our newly-created device:

$ echo "/file/to/send/ > /dev/readfile

From now on, once an ICMP-request (ping) packet will reach the machine, the module will inject the file's data on the ICMP-reply packet.

For more information, see How It Works.

Example

How It Works


Why an LKM?

You probably ask yourselves why to implement this complicated logic as a kernel module. It is harder to implement and it can certainly be implemented in user-mode by using netfilter's user-mode functionality.

Well, there are few reasons why I chose to implement it as an LKM:

Mangling outgoing packets

The module uses netfilter to intercept every outgoing packet in the system.

From netfilter's documentation:

netfilter is a set of hooks inside the Linux kernel that allows kernel modules to register callback functions with the network stack. A registered callback function is then called back for every packet that traverses the respective hook within the network stack.

netfilter offers a number of different places where a user can place his hooks:

--->[1]--->[ROUTE]--->[3]--->[4]--->
             |            ^
             |            |
             |         [ROUTE]
             v            |
            [2]          [5]
             |            ^
             |            |
             v            |

[1] NF_IP_PER_ROUNTING — triggered by any incoming traffic very soon after entering the network 
                         stack. This hook is processed before any routing decisions have been made 
                         regarding where to send the packet.
[2] NF_IP_LOCAL_IN —     triggered after an incoming packet has been routed if the packet is destined 
                         for the local system.
[3] NF_IP_FORWARD —      triggered after an incoming packet has been routed if the packet is to be 
                         forwarded to another host.
[4] NF_IP_POST_ROUTING — triggered by any outgoing or forwarded traffic after routing has taken place 
                         and just before "hitting the wire".
[5] NF_IP_LOCAL_OUT —    triggered by any locally created outbound traffic as soon it hits the network 
                         stack.

The module places an hook on NF_IP_POST_ROUTING in order to intercept every outgoing packets, meaning that the callback will be called for every packet sent from the machine, with the packet as its argument.

Then the module checks whether the packet's protocol is ICMP and its type is ICMP-reply. Every packet that doesn't meet the requirements is ignored and left untouched. Every packet that meets the requirements will later be injected with the file's data.

All the above logic is implemented in net/netfilter.c.

Interrupt context

When a packet is received or about to be sent via the network card, the kernel invokes the interrupt handler routines of its matching driver. It means that when our netfilter hook is being called, we are already in interrupt context. Running inside an interrupt context means that the rules of that context are also applied here, among them:

Any attempt to do one of the above inside the netfilter's hook can result in kernel panic.

Injecting data to ICMP packets

This is the time to admit that this title is a bit misleading, as we are injecting data on top of ICMP packets (spoiler!), and not into ICMP packets.

ICMP packets are very narrowed in places where we can inject our data. To refresh our memory, here is the ICMP header layout taken from its RFC:

Echo or Echo Reply Message

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|     Type      |     Code      |          Checksum             |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|           Identifier          |        Sequence Number        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|     Data ...
+-+-+-+-+-

As we can see, the ICMP header is very small. If its not enough, its data must be the same between the reply and request packets, otherwise each side won't treat the packet as valid (hence won't print the famous 64 from 192.168.1.1 ... output).

Since injecting our data on one of the above fields is not possible without causing an anomaly in both sides we have to find another way.

What if we will inject our data after the end of the ICMP layer?

A quick research shows that this is possible.

As you can see in the picture, our data is inserted on top of the ICMP layer, without creating any anomaly in the packet. Also, wireshark parses that as a padding layer (not seen in the picture).

So, whenever the module identifies an outgoing ICMP-reply packet, it shoves the desired payload after the ICMP layer of the packet. The method of injecting payload as a padding layer has several advantages:

Each time that netfilter's hook is called, it provides the relevant sk_buff. sk_buff, or "socket buffer" is the most fundamental data structure in the Linux networking code. Every packet sent or received is handled using this data structure. is the represntation of a socket used widely across the kernel. The sk_buff provides all kinds of information on the packet, including its headers and data.

By using the function skb_data_put one can inject its own data into the skb, which will later be sent over the network.

All the logic of reading (and splitting) files and providing the data to the netfilter hook is implemented in net/generate_payload.c.

Modifying file data before sending

The module supports modifying each file's data before sending it. It can be good in cases where we want to compress or encrypt the data before sending it to increase performance / avoid being detected by network sniffers.

The core is responsible to add all to modifiers (according to the configuration found in config.h) to the payload_generator thread. The latter then invokes each modifier when a file is read. For example:

    original file data ---> g_modifiers.compress() ---> g_modifiers.encrypt() ---> modified data
    

Reading user-space files

Dealing with files from inside the kernel is considered by many as a bad practice (and for some by a very bad one).

The main reason for that consensus is that the kernel is the one reponsible for tranlsating logical I/O (e.g user-space files) to physical I/O (e.g physical location on the disk), and performing logical I/O operations from within the kernel is weird and involves many edge-cases, and may even result in a kernel panic.

Most of the times - this is true. If a kernel module should get information from the user-space, the common way is to register a device and use ioctl to give it information. But as our goal is to secretly read files from the user, it is inevitable for us to perform file I/O operations from within the kernel.

So, how it is done?

As dealing with files from the kernel is a bad practice, there is very little information on the internet regarding to that topic. The main source I used while writing this module is the kernel's source-code, with the help of fractions of stackoverflow threads.

The module uses the functions filp_open and vfs_read to open and read files respectively. It then splits the files to chunks with a fixed size (defined in the code), which will later be sent on top of our ICMP packets.

The main issue with vfs_read is that it expects to save the output in a user-space buffer. In order to overcome this requirement, one should change the FS the KERNEL_DS, telling the kernel's memory-checking mechanism to expect a kernel-space buffer. Newer kernel versions have introduced the functions kernel_read and kernel_write, which does that fs dance within them and saving if from the caller. Generally, it is perferred to use these function, but it didn't work in this case :(

I suspect that kernel_read had failed due to the fact that modern filesystems (such as ext4 which I use) define filp->f_op->read to be NULL. Although it does supports filp->f_op->read_iter, and although it should be abstract to the caller which one of this functions is implemented, kernel_read fails de facto from an unknown reason (actually it fails with -EINVAL but I have no idea why).

Testing

Emulation: it is important to mention that this module may very likely not work on emulations (such as QEMU) or virtualization (such as VMWare) system. The reason is that most of the time these systems are doing some black magic in the backend to support networking to the emulated or virtualized machines, mostly invlove reading the packets' data and re-packaging them, which may cause our data to be lost. I assume that with the right configuration, you can run it on QEMU of VMWare, but don't be surprised if it doesn't work at first.

Future features

Reading material