aboutsummaryrefslogtreecommitdiffhomepage

NAME

dpibreak - circumventing deep packet inspection

SYNOPSIS

dpibreak [OPTIONS]

DESCRIPTION

DPIBreak is a tool for circumventing Deep Packet Inspection (DPI) on HTTPS connections. It is only applied locally and no other external connection is needed.

Mainly it fragments the TCP packet carrying the TLS ClientHello so that certain DPI devices cannot extract the Server Name Indication (SNI) field and identify the destination site. Other method can also be used along with it, and it is described in the OPTIONS section.

DPIBreak registers firewall rules to handle inbound and outbound packets. The rules are automatically added on startup and removed on exit, making it effective system-wide without manual intervention. Firewall rule cleanup relies on SIGTERM/SIGINT. If the process is killed with SIGKILL, cleanup will not occur. However, the registered nfqueue rules simply pass packets through when no process is consuming the queue, so this is not a concern in practice. In that case, restarting and gracefully stopping DPIBreak will clean up the leftover rules.

To register firewall rules and verdict packets, root privilege is required on Linux (nft(8) or iptables(8)/ip6tables(8) with xt_u32 must be available); administrator privilege is required on Windows (WinDivert64.sys and WinDivert.dll must be in the same directory as dpibreak.exe).

This only applies to TLS-based connections (HTTPS). UDP/QUIC (RFC 9000) is not affected.

OPTIONS

-d, --daemon
Run as a background daemon. Logs are written to /var/log/dpibreak.log. If a daemon is already running, it will fail with "unable to lock pid file". To stop it, run kill `cat /run/dpibreak.pid` as root.

On Windows, this option enters the service controller entry point. Example: sc create dpibreak binPath= "dpibreak.exe -d" start= auto; sc start dpibreak

--delay-ms \<u64>
Delay in milliseconds to apply between fragmented pieces of the ClientHello. Typical values are 0–1000; larger values may increase handshake latency. (Default: 0)

--fake
Enable fake ClientHello packet injection before sending each packet fragmented. TCP/IP header fields follow the original packet unless you override it using the --fake-* options described below. Packets are transmitted in an interleaved order: (fake 1), (orig 1), (fake 2), (orig 2), ...

-t, --fake-ttl \<u8>
Override ttl (IPv4) / hop_limit (IPv6) of fake packet. Implicitly enables --fake. (Default: 8)

-a, --fake-autottl
Automatically infer the hop count (TTL/Hop Limit) to the destination by analyzing the SYN/ACK packet. It assumes the server's initial TTL is either 64, 128, or 255. The inferred value is used for fake packet transmission to ensure they reach the censor but not the destination. By default, it uses a delta of 1. If the hop count cannot be determined, it falls back to the value specified by --fake-ttl. Implicitly enables --fake.

--fake-badsum
Corrupts the TCP checksum of fake packets. When enabled, fake packets cannot pass through most routers and will not behave as expected. It can be useful if your router/firewall provides an option to disable TCP checksum verification. Implicitly enables --fake.

-o, --segment-order \<u32,u32,...>
Specify the order in which TCP segments of the TLS ClientHello are transmitted. The argument is a comma-separated list of byte offsets, which must include 0. Offsets are sorted to define segment boundaries: for example, 0,1,3,5 defines segments [0,1), [1,3), [3,5), and [5,end). The transmission order follows the order given on the command line. Segments whose start offset exceeds the payload length are silently skipped. (Default: 0,1)

--queue-num \<u16>
NFQUEUE number to attach to. The same queue number is used for IPv4 and IPv6. (Default: 1)

--nft-command \<string>
Custom nftables command to be executed. (Default: nft)

--log-level \<debug|info|warning|error>
Set the logging level (Default: warning). Aliases: warn -> warning, err -> error.

--no-splash
Disable splash messages at startup.

-h, --help
Show usage information and exit.

EXIT STATUS

0
Successfully terminated.

non-zero
The program encountered an error during initialization or runtime, resulting in an abnormal exit.

EXAMPLES

Run with default options:

dpibreak

Run as daemon with fake ClientHello injection:

dpibreak -d --fake-autottl

Send the TLS ClientHello with the second byte first, followed by the remainder, then the first byte:

dpibreak --segment-order 1,2,0

This causes DPI equipment to receive an incomplete record header in the first observed segment, preventing SNI extraction, while the destination server reassembles the segments normally via TCP sequence numbers.

Send segments in a custom out-of-order sequence with a delay:

dpibreak --segment-order 5,1,3,0 --delay-ms 10

Run with a 10 ms delay between fragmented packets and verbose logging:

dpibreak --delay-ms 10 --log-level debug

Use a custom NFQUEUE on Linux:

dpibreak --queue-num 3

FILES

/run/dpibreak.pid
Whether running as a daemon or not, DPIBreak will fail through this file if another instance is already running on the system.

/var/log/dpibreak.log
Only for daemon. log goes here.

BUGS

There are three types of bugs:

a. DPIBreak does not work as described in the manual.\ b. It works as described but fails to bypass the DPI.\ c. A site becomes unreachable or misbehaves with DPIBreak running, but works when it is stopped.

Reporting all cases to the bug tracker helps improve the program. (a) indicates an implementation bug. (b) may motivate new circumvention techniques. (c) may be resolved with a site-specific workaround or a narrowing flag. For (b) and (c), please include your region, ISP, flags in use, and any other network software running (VPN, custom DNS, ad-blocker, firewall).

Any other minor improvements or suggestions are also welcome.

You can view the known bugs list and submit reports at:\ https://github.com/dilluti0n/dpibreak/issues

SEE ALSO

nft(8), iptables(8), ip6tables(8)

AUTHOR

Written by Dilluti0n \<[email protected]>.