Repository navigation
test(e2e): aim the escape-union cross arms at public halves of 198.18.0.0/15 - #2350
Merged
norrietaylor merged 3 commits intoOct 20, 2026
Conversation
The cross-box spoof arms of proof_escape_reaches_only_declared_union aimed at RFC 1918 addresses (192.168.77.7, 10.0.0.7), so the gate's infrastructure rule took them before the row's own rules could and the refusals landed on egress-infrastructure-destination. A union- bound regression would have stayed green: the verdict the case rides on never fired. Give the two boxes disjoint public halves of 198.18.0.0/15 instead (box A 198.18.0.0/16, box B 198.19.0.0/16) — RFC 2544 benchmark space, outside every infrastructure range — so a cross-box frame falls through to the row's own rules and is refused under egress-undeclared-subnet, the verdict that actually proves the union bound. Both of box A's undeclared-subnet arms (cross and far-side) wear the same address under the same rule, sharing one 60s rate-limit window (DropLimiter, keyed by source and rule only). The made-up arms dodge the limiter with a fresh source address; these two cannot, so the window is waited out between them instead. The in-crate twin needs no change: its cross arms were removed in T89 and the bound is pinned by the unit case on TEST-NET-1. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Contributor
bryan-minimal
approved these changes
Oct 19, 2026
bryan-minimal
left a comment
Member
There was a problem hiding this comment.
Right call: aiming the cross arms at RFC 1918 let egress-infrastructure-destination answer before the row's own rules, so the union bound was never the verdict under test. 198.18/16 and 198.19/16 sit outside every infrastructure range, and the sleep 60 between box A's two undeclared-subnet arms is forced by the limiter's (source, rule) key; the comment explains why the made-up arms can dodge it and these cannot. Approving.
…es-74686e4a0c5f1acb
…es-74686e4a0c5f1acb
norrietaylor
deleted the
inbox-patch/fix/e2e-escape-union-public-halves-74686e4a0c5f1acb
branch
October 20, 2026 18:27
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Routing-Key: inbox-route/I_kwDOSUhdos8AAAABU7fGdg
The cross-box spoof arms of the escape-union e2e proof
(
proof_escape_reaches_only_declared_unioninscripts/session-e2e.sh)aimed at RFC 1918 addresses (192.168.77.7, 10.0.0.7). The gate's
infrastructure rule therefore took those frames before the row's own
rules could, and the refusals landed on
egress-infrastructure-destinationinstead of
egress-undeclared-subnet— so a union-bound regression wouldhave stayed green: the verdict the case rides on never fired.
The two boxes now declare the disjoint public halves of 198.18.0.0/15
(box A 198.18.0.0/16, box B 198.19.0.0/16) — RFC 2544 benchmark space,
outside every infrastructure range — so a cross-box frame falls through
to the row's own rules and is refused under
egress-undeclared-subnet,the verdict that actually proves the union bound. Raised in the design
review of #1943.
Because both of box A's undeclared-subnet arms (cross and far-side) wear
the same address under the same rule, they share one 60s rate-limit
window (
DropLimiter, keyed by source and rule only, never destination).The made-up arms dodge the limiter by wearing a fresh source address;
these two cannot — the address each wears is box A's own, which is the
bound they prove — so the window is waited out between them (
sleep 60).The in-crate twin needs no change: its cross arms were removed upstream
(T89) and the bound is pinned by the unit case
spoofed_source_bounded_to_resident_union, already on TEST-NET-1.Verification
cargo fmt --all --check --manifest-path target/Cargo.toml— cleancargo clippy --workspace --all-targets --locked --manifest-path target/Cargo.toml -- -D warnings— cleancargo test --locked -p minvmd --manifest-path target/Cargo.toml— 575 passed; 3 failures are pre-existing sandbox flakes (tracing-subscriber global-state collisions and a/state/minimalpath this sandbox lacks), reproduced on the pristine baseline and passing in isolationcargo test -p minvmd egress_gate(with the change in place) — 123 passed, 0 failedscripts/**and run on this PR; not runnable in the fixer sandbox (no /dev/kvm)Summary by CodeRabbit