Repository navigation
fix(corekit): non-zero padding octets are zeroed on rebuild #1223
Description
Describe the bug
PaddingField never stores the octets it read, so any non-zero padding comes back as zeros.
To Reproduce
- TCP options
020405b4 00aabbccrebuild as…00000000. - SCTP DATA padding
ffffffrebuilds as000000. - An SCTP parameter pad
eerebuilds as00.
Expected behavior
Padding octets round-trip as read. This may overlap the corekit round-trip audit, which is still running.
Found by the transport round-trip audit (#1202). The repro is from the audit report and I have not re-run it yet.
Activity
JarryShaw commentedon Oct 7, 2026
Blocked until the corekit round-trip audit reports, so the fix lands with whatever else it finds in
PaddingField.JarryShaw commentedon Oct 7, 2026
Twelve more HIP sites for this issue, listed by the #1342 cross-review in
schema/internet/hip.py: R1_COUNTER:563, NAT_TRAVERSAL_MODE:818, ENCRYPTED:849, NOTIFICATION:1056, REG_FROM:1165, ESP_TRANSFORM:1207, PAYLOAD_MIC:1256, ROUTE_DST:1304, RELAY_FROM:1426, RELAY_TO:1443, OVERLAY_TTL:1458and ROUTE_VIA:1476. Eleven were confirmed by writing0xffand rebuilding; ENCRYPTED is read from the schema only. The 13th HIP site,Locator.flags, is aBitFieldand is filed separately as #1365.JarryShaw commentedon Oct 8, 2026
More sites with the same root cause (
PaddingFieldpacks zeros, so non-zero octets parse silently and come back zeroed). ThePaddingFieldsweep found these and I re-checked them onmainat79949ec3d. The issue does not yet list any of them:schema/internet/hopopt.py:427andipv6_opts.py:427:3b00 0104 aabbccddbecomes3b00 0104 00000000.hopopt.py:687andipv6_opts.py:692:… 80 aa …becomes… 80 00 ….ipv6_frag.py:38; SR0 / Type 2 reserved(4),ipv6_route.py:162,:178.httpv2.py:240,:289,:387(RFC 7540 §6.1 lets a receiver reject non-zero pad); OSPF crypto reserved,ospf.py:47.Checked and fine, either rejected at parse or unreachable: IPv4 LSR/RR/SSR remainders, IPv6 MPL pad, IPv4 header padding. IPv4 TS is #1392 (fixed in #1399). The CALIPSO length loss is a separate defect, now #1403. Still unprobed: PCAP-NG alignment pads, IPv6-Route RPL pad and L2TP offset pad.
JarryShaw commentedon Oct 8, 2026
One more site, found while checking #1235: the Quick-Start report "not used" octet is a
PaddingField(length=1), so3b01260680ff12345679…rebuilds as…800012345679….JarryShaw commentedon Oct 8, 2026
One more site, from the #1420 cross-review: the Simple Packet Block
paddingis aPaddingField(schema/misc/pcapng.py:1549), so withoriginal_len=5and dataabcdefghthe 3 non-zero pad octets rebuild as zeros (24 → 24 octets, not exact).JarryShaw commentedon Oct 8, 2026
Unblocked: the corekit round-trip audit has filed its findings (#1288–#1300, filed 2026-10-07), and #1288 closed with #1416. A worker is on it now, on branch
fix/1223-padding-kept-as-captured. The worker also covers #1329's MH PadN remainder. The IPv4 Quick-Start report site is held for a follow-up, because the #1209 worker ownsipv4.pyin this wave.fix(corekit)!: non-zero padding octets are zeroed on rebuild (#1223)
fix(corekit)!: non-zero padding octets are zeroed on rebuild (#1223) (#…
docs(protocols): second pass over the IPv6 extension headers (#1103) (#…