Feature: CAN XL Support by RB-FriedrichWiemer · Pull Request #5152 · secdev/scapy

RB-FriedrichWiemer · 2026-09-07T12:57:53Z

Description

This is my first pull request to scapy.
It adds CAN XL support to the existing CAN layer, making use of the CAN XL implementation available on Linux' SocketCAN.

polybassa · 2026-09-08T12:37:24Z

Thanks for this PR. Please resolve the conflicts with the master.

codecov · 2026-09-08T13:01:33Z

Codecov Report

❌ Patch coverage is 82.35294% with 18 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.13%. Comparing base (d1bbdaf) to head (50989fa).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
scapy/contrib/cansocket_native.py	55.55%	12 Missing ⚠️
scapy/layers/can.py	92.95%	5 Missing ⚠️
scapy/contrib/cansocket_python_can.py	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5152      +/-   ##
==========================================
+ Coverage   80.84%   81.13%   +0.29%     
==========================================
  Files         393      393              
  Lines       98325    98414      +89     
==========================================
+ Hits        79488    79853     +365     
+ Misses      18837    18561     -276

Files with missing lines	Coverage Δ
scapy/contrib/cansocket_python_can.py	`84.02% <75.00%> (-0.36%)`	⬇️
scapy/layers/can.py	`93.04% <92.95%> (-0.07%)`	⬇️
scapy/contrib/cansocket_native.py	`70.58% <55.55%> (-7.99%)`	⬇️

... and 25 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

pfasante · 2026-09-22T10:01:17Z

the byte-swap for wireshark pcaps I still need to check

polybassa

Looks good. Let's check what wireshark does, then we can merge

polybassa · 2026-09-25T19:10:36Z

Please fix the failing job.

polybassa · 2026-09-28T19:58:33Z

sorry, there is still the AI-Trailer job failing. You may need to to a rebase.

Introduce the CANXL class in scapy.layers.can, following the Linux struct canxl_frame wire format and inheriting from the CAN class. Provides additional ISO 11898-1:2024 property accessors and show(style='11898-1') for ISO field-name rendering. AI-Assisted: yes (Claude Opus and Sonnet)

AI-Assisted: yes (Claude Opus and Sonnet)

Tests pass on Linux. On Windows, tests appear to timeout -- needs further investigation. AI-Assisted: yes (Claude Opus and Sonnet)

Covers quick start, field naming (Linux vs ISO), byte-order handling, NativeCANSocket usage, can-utils interop, and known limitations. AI-Assisted: yes (Claude Opus and Sonnet)

AI-Assisted: yes (Claude Opus and Sonnet)

AI-Assisted: no

Declare the CAN XL header as little endian via BitField's tot_size / end_tot_size, LEShortField and XLEIntField, instead of swapping the prio, length and af regions by hand. Removes CANXL.inv_endianness() and the swap in pre_dissect() and post_build(). AI-Assisted: yes (Claude Opus and Sonnet)

post_dissect duplicated CAN.post_dissect, and the default guess_payload_class already returns conf.raw_layer when no payload class is bound. AI-Assisted: yes (Claude Opus and Sonnet)

Keep Packet.show()'s signature unchanged and expose the ISO 11898-1 rendering as its own method. AI-Assisted: yes (Claude Opus and Sonnet)

Use Packet.sprintf() for the fields it can resolve instead of formatting each one against the color theme by hand. Format, FTYPE, SEC and DLC are properties rather than fields, so they stay resolved in Python. AI-Assisted: yes (Claude Opus and Sonnet)

Both recv_raw() and send() open-coded the same four-byte swap that CAN.inv_endianness() already performs. Also drops _is_canxl(), a single-use wrapper around CANXL.is_canxl_frame(). AI-Assisted: yes (Claude Opus and Sonnet)

AI-Assisted: yes (Gemini 3.8 Flash)

gpotter2 · 2026-09-29T21:52:28Z

@polybassa

sorry, there is still the AI-Trailer job failing. You may need to to a rebase.

finally managed to fix that need for rebasing... will merge it ASAP. sorry about that 78c41b5

polybassa · 2026-10-07T21:02:09Z

Could you please have a look why some can related tests are failing

CAN XL support changed send() to choose the padding size from the packet type, so a classic CAN frame sent on a socket opened with fd=True was transmitted as a 16 byte frame instead of a CAN FD frame. AI-Assisted: yes (Claude Opus and Sonnet)

Python 3.7 rejects the backslash continuation when UTScapy compiles the test, with "SyntaxError: trailing comma not allowed without surrounding parentheses". AI-Assisted: yes (Claude Opus and Sonnet)

polybassa · 2026-10-09T02:56:59Z

+            # CAN/CANFD: pad to the frame size the socket was opened with, so
+            # that a classic CAN frame sent on an fd socket stays a CAN FD frame
+            mtu = CAN_FD_MTU if self.fd else CAN_MTU
+            bs = bs + b"\x00" * (mtu - len(bs))


Is there a reason why mtu is not self.MTU anymore?

See update in 32b78d3

I think its correct now?

polybassa · 2026-10-09T03:02:28Z

+            log_runtime.warning(
+                "CAN XL payload length %d exceeds the ISO 11898-1 "
+                "maximum of %d (11-bit field)", length, CANXL_MAX_DLEN)
+        pkt = pkt[:6] + struct.pack('<H', length) + pkt[8:]


You should only compute length if the field is None.
Scapys intention is to do the right thing by default, but if you specify a incorrect value yourself, this value should be sent to the wire

good point, fixed that.

Also required some test updates that where encoding this "fixing" behavior

polybassa · 2026-10-09T03:03:16Z

+                "CAN XL frame has IDE set; clearing it "
+                "(IDE is always 0 for CAN XL per ISO 11898-1)")
+        flags = (pkt[4] | CANXL_XLF | CANXL_FDF) & ~CANXL_IDE
+        pkt = pkt[:4] + bytes([flags]) + pkt[5:]


polybassa · 2026-10-09T03:07:04Z

+        # dropped rather than preserved as a Padding layer.  CAN XL
+        # frames from a native socket have exact-length data; any
+        # trailing garbage is safely discarded.
+        return p[:data_len], None


Please respect the remove_padding config here

polybassa · 2026-10-09T03:09:39Z

+        # but Linux struct canxl_frame uses a full 16-bit field.
+        # For kernel compatibility we use a 16-bit field; post_build warns
+        # if the computed length falls outside the valid range.
+        LEShortField('length', 0),


Ideally, change the default to None

self.MTU is CANXL_MTU on a socket opened with xl=True, which is not a valid frame size for a classic CAN or CAN FD frame, so those are padded to CAN_MTU there. AI-Assisted: yes (Claude Opus and Sonnet)

Default the length field to None and compute it from the payload in post_build only in that case, so that a value set by the user reaches the wire unchanged. AI-Assisted: yes (Claude Opus and Sonnet)

post_build forced XLF and FDF on and IDE off, overriding a value set by the user. The field default still follows ISO 11898-1. AI-Assisted: yes (Claude Opus and Sonnet)

Trailing bytes beyond the stated length were always discarded. They are now kept as Padding when remove-padding is disabled, as for CAN and CAN FD. AI-Assisted: yes (Claude Opus and Sonnet)