Repository navigation
fix(minvmd): bound the in-force baseline's host-alias reach to its fetch ports - #2444
Conversation
…tch ports The node-plane baseline's compiled set admits by subnet alone — it carries no port dimension — so once the baseline phase goes in force, the gate's node-plane arm admits frames to the switch's host alias at every port. The built-in registry and cache endpoints name that alias because it is where the helper's store surfaces answer, but the reach the daemon's own fetches need is the HTTPS fetch port, nothing wider. Bound the in-force arm in the gate: a node-plane frame to the alias at a port the daemon's fetches do not use is the infrastructure drop, the same refusal a box row's alias frames take, so in-force reach is no wider than the announced phase's. The announced arm and every box row's verdict are untouched. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Merge Risk: ⚪ Minimal · up to
|
… policy view The in-force arm compared only the destination port, which the summary reads from every protocol's bytes, so ICMP or GRE frames whose bytes read 443 reached the alias. Open only TCP at the fetch port, and refuse every other protocol as the infrastructure drop. State where 443 comes from: the planned port of the helper's store surface, with no listener yet. The change that flips the phase to in force must feed the list from that listener's stack. Drop the claim that in-force reach is no wider than announced, and note that the bound must also admit the configured host exposures once they exist. The policy text and JSON views now spell the alias endpoint with its bound, as "100.64.255.254/32 tcp/443".
|
This PR has merge conflicts with its base branch ( Conflicted files:
Reason it's not trivial: the only conflicted file is Rust source ( Recommended action: rebase or merge manually, then remove the
|
|
/revise Merge main into this branch and resolve the conflicts. Keep this PR's behaviour and take main's changes elsewhere; make no other changes. Use git's default merge subject ("Merge ..."). |
|
Round 2 of 10 is on this pull request. branch
Updated 2026-10-22T05:27:12.483Z. |
…force-baseline-alias-ports-1698b5b64d54dd6e # Conflicts: # crates/minimal/src/cmd/session.rs
|
Round 1 of 10 on a /revise note, pushed as Resolved conflicts in: No The branch now contains main. It differs from main only by this PR's change: the in-force baseline bound in egress_gate.rs and baseline.rs, the cli.rs tests, and the policy view that prints the bound (100.64.255.254/32 tcp/443) from crates/minimal/src/cmd/boxes/policy.rs. Notes from the agent: Only crates/minimal/src/cmd/session.rs conflicted. Main moved format_policy and write_policy_json_noting from session.rs to crates/minimal/src/cmd/boxes/policy.rs. I took main's session.rs as is, and it now matches origin/main. I applied this PR's two-line change (entry.endpoints() -> baseline.display_endpoints(entry)) at its new location in policy.rs, in the merge commit. The merge commit uses git's default subject. The clippy -D warnings errors are in files from main that this PR does not touch, and the local rustc 1.98.1 can be newer than the pinned toolchain. I made no other changes. |
|
/arch-review |
There was a problem hiding this comment.
arch-review: CONDITIONS · 2 conditions · the full ruling
The in-force gate now bounds the host alias to TCP 443 and refuses all other frames to the alias. This narrows reach and conforms to §7.1 line 244. Two gaps remain. The head defers host-exposure admission through this bound, but no issue owns that work (C1). The REGISTRY_CACHE_FETCH_PORTS doc says the list is empty with no listener, but the value is [443] (C2). Make both edits, and the change conforms.
Ruled at c38e023 against gominimal/arch@109541a48c31 by claude-opus-5-5.
|
/revise Apply /arch-review's conditions at c38e023. C1: where the head defers host-exposure admission through the in-force bound, add a comment naming the issues that own that work: gominimal/inbox#1144 and gominimal/inbox#1147. C2: fix the |
…force-baseline-alias-ports-1698b5b64d54dd6e
Name gominimal/inbox#1144 and gominimal/inbox#1147 where the in-force alias bound defers host-exposure admission. Make the REGISTRY_CACHE_FETCH_PORTS doc match its value [443].
|
Round 2 of 10 on a /revise note, pushed as No Review threads the coder judged:
A reader of the in-force alias bound in egress_gate.rs now sees that gominimal/inbox#1144 and gominimal/inbox#1147 own the deferred host-exposure admission. The REGISTRY_CACHE_FETCH_PORTS doc comment now says that [443] is the planned port with no listener, that only the in-force tests exercise it while the phase is Announced, and that #1786 replaces it with the listener's port or an empty list. Notes from the agent: This round changes comments only. No code and no values change. The -D warnings clippy failure is a chunks_exact_to_as_chunks lint in minimald-rpc. That crate is outside this pull request, so the failure is probably from the toolchain on this box. cargo doc did not run because rustdoc is not on this box. The new intra-doc link [ |
|
/arch-review |
There was a problem hiding this comment.
arch-review: PASS · the full ruling
This head conforms to §5.1 line 109 and §7.1 lines 244 and 248. The in-force arm now refuses every frame from the node plane to the host alias, except TCP to port 443. The arm stays latent because NODE_BASELINE_PHASE is still Announced. The last commit names the owners of the deferred host-exposure work and makes the port-list doc match [443]. Before the phase flips, #1786 must show which line-244 opening admits a store surface on the alias. List gominimal/inbox#1144, gominimal/inbox#1147 and #1786 as residuals in the PR body.
Ruled at 0b7e341 against gominimal/arch@109541a48c31 by claude-opus-5-5.
|
/review |
|
Reviewed: 1 finding, none blocking at Address the comments on the diff. Clobber: none seen. |
norrietaylor
left a comment
There was a problem hiding this comment.
Arch PASS at 0b7e341, CI green, threads resolved; LOW moved to the follow-up digest.
…tch ports (#2444) Prerequisite for T102. The node-plane baseline's compiled set admits by subnet alone, so once the baseline phase goes in force the gate's node-plane arm would admit frames to the switch's host alias at every port. The in-force arm now opens the alias only to TCP at the daemon's fetch port and refuses every other port and protocol as the infrastructure drop. The announced arm and every box row's verdict are untouched. The policy text and JSON views spell the alias endpoint with its bound, as 100.64.255.254/32 tcp/443. Backport of #2444. Adapted to the line: the CLI's policy view lives in cmd/session.rs, not cmd/boxes/policy.rs, and boxes:: reads sessions::. (cherry picked from commit 6471908)
…tch ports (#2444) Prerequisite for T102. The node-plane baseline's compiled set admits by subnet alone, so once the baseline phase goes in force the gate's node-plane arm would admit frames to the switch's host alias at every port. The in-force arm now opens the alias only to TCP at the daemon's fetch port and refuses every other port and protocol as the infrastructure drop. The announced arm and every box row's verdict are untouched. The policy text and JSON views spell the alias endpoint with its bound, as 100.64.255.254/32 tcp/443. Backport of #2444. Adapted to the line: the CLI's policy view lives in cmd/session.rs, not cmd/boxes/policy.rs, and boxes:: reads sessions::. (cherry picked from commit 6471908)
Routing-Key: inbox-route/I_kwDOSUhdos8AAAABYbq6lA
The node-plane baseline's compiled set admits by subnet alone — no port
dimension — so once the baseline phase goes in force, the gate's
node-plane arm admits frames to the switch's host alias at every port.
The built-in registry and cache endpoints name that alias as the planned
home of the helper's store surface.
Bound the in-force arm in the gate: a node-plane frame to the alias is
admitted only as TCP to the planned store-surface port (443). Every other
protocol or port is the infrastructure drop, the same refusal a box row's
alias frames take. Once host exposures close by default (#2331), this is
wider than the announced phase's alias reach, and the bound must then also
admit the configured exposures. No listener serves 443 on the alias yet;
the PR that flips the phase must feed the port list from that listener.
The policy view shows the bound (
100.64.255.254/32 tcp/443). Announcedbehavior is untouched.
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 --manifest-path target/Cargo.toml -p minvmd— 576 passed; the 3 remaining failures (net::answerer×2,sock::tests::resolve_uds_path_lives_in_provider_dir) reproduce identically on the unmodified tree — pre-existing environment flakes, unrelated to this changein_force_baseline_bounds_the_host_alias_to_its_fetch_portsfailed before the fix (alias admitted at 18081) and passes afterSummary by CodeRabbit
Security Enhancements
CLI Improvements