2026-08-23 22:40:55 +00:00
1 changed files with 118 additions and 15 deletions
@@ -17,22 +17,57 @@ everything in the lab hangs off, so it is worth getting small.
 It states what a hosting provider and a home router would provide, and nothing the mesh is
 responsible for ([ADR 0031](../../02-DECISIONS/0031-the-lab-provides-the-underlay.md)).
 ## What NAT does, and why the design turns on it
 A household or office has **one** address the outside world can see, and **many** machines
 behind it. Network address translation is what reconciles those.
 When a machine inside dials out, the gateway rewrites the packet's source from the private
 address to the public one, **remembers the mapping**, and rewrites the replies on the way back.
 Four consequences follow, and every one of them shapes this design:
 1. **Outbound works; inbound does not.** A mapping exists only because something inside started
   a conversation. Nothing outside can start one — there is no mapping to look up, and no way
   to know which internal machine was meant.
 2. **A forwarded port is a permanent mapping made by hand**, in the inbound direction:
   *anything arriving at the public address on 443 goes to this machine.* That is the only way
   a machine behind NAT becomes reachable, and it requires control of the gateway.
 3. **Mappings expire.** A gateway forgets one that goes unused. This is why anything holding a
   connection through NAT sends keepalives, and why a mesh that does not is fine until it is
   idle.
 4. **From outside, every machine behind the gateway looks like one address.** Identity and
   address stop corresponding.
 This is why the mesh dials outward and never inward
 ([ADR 0001](../../02-DECISIONS/0001-nodes-communicate-over-a-broker.md)), why a hub exists at
 all, and why a node's endpoint is something a peer **learns** from arriving packets rather than
 something anyone configures.
 **Carrier-grade NAT** is the same mechanism applied by an ISP: your own gateway gets a private
 address too, and the public one is shared with strangers. Nothing can be forwarded, because the
 rule would have to live on equipment you do not own. Common on mobile connections and
 increasingly on fixed ones.
 ## Three positions a machine can be in
 The underlay's whole job is to reproduce **where a machine sits relative to the internet**,
 because that is what the mesh has to cope with and what only production currently exercises.
 There are three positions, and they are genuinely different:
-| Position | Reachable from outside | Address | Example |
+| Position | Reachable from outside | Apparent address | Example |
 |---|---|---|---|
-| **Directly attached** | yes, at its own address | fixed, its own | a hosted server |
+| **Attached** | yes, at its own address | its own | a hosted server |
-| **Behind a gateway you control** | only through a forwarded port, at the *gateway's* address | private, plus the gateway's public one | a machine at home |
+| **Behind a forwardable gateway** | only through a forwarded port, at the *gateway's* address | the gateway's | a machine at home |
-| **Behind a gateway you don't control** | **no** | private, and it changes | a laptop on someone else's network |
+| **Behind an unforwardable gateway** | **no** | someone else's, and it changes | a laptop on a café network; anything behind carrier-grade NAT |
-The third is the hard one and the reason this matters. A machine there can dial out and nothing
+The axis is **forwardability, not ownership** — which is worth stating because the obvious
-more: it cannot be published, its apparent address belongs to somebody else's router, and that
+framing gets it wrong. Carrier-grade NAT is *your* connection and is still unforwardable, so it
-address changes when it moves. Every assumption a mesh makes about reachability breaks there
+belongs in the third row alongside the café. What the mesh has to cope with is whether an
-first.
+inbound mapping can be made, not who owns the equipment.
 The third position is the hard one. A machine there can dial out and nothing more: it cannot be
 published, its apparent address belongs to a router it does not control, and that address
 changes when it moves. Every assumption a mesh makes about reachability breaks there first.
 A declaration has to be able to say all three, and to move a machine between them.
@@ -43,19 +78,24 @@ scenario: roaming-and-published
 segments:
  internet:
-    cidr: 203.0.113.0/24          # the simulated public internet, RFC 5737
+    kind: public                  # stands in for the internet — RFC 5737 addresses
    cidr: 203.0.113.0/24
  home:
    kind: private
    cidr: 192.168.1.0/24
    gateway:
      to: internet
      address: 203.0.113.50       # what the world sees this network as
      nat: true
      forwardable: true           # we control it, so ports can be opened
  elsewhere:                      # a network we do not control
    kind: private
    cidr: 198.51.100.0/24
    gateway:
      to: internet
      address: 203.0.113.80
      nat: true
      forwardable: false          # café wifi, or carrier-grade NAT
 machines:
  anchor:
@@ -81,8 +121,14 @@ snapshot: raised
 ## What each part means, precisely
-**`segments`** — a broadcast domain with an address range. A segment with no `gateway:` *is*
+**`segments`** — a broadcast domain with an address range, and a `kind:`.
-the internet as far as the scenario is concerned. A segment with one sits behind it.
+
 `kind: public` marks the segment that stands in for the internet. `kind: private` is everything
 else. This is stated rather than inferred, and the earlier version inferred it — *a segment with
 no gateway is the internet* — which made an **isolated network inexpressible**: a LAN with no
 route out is a private segment with no gateway, and would have been read as the internet and
 forced to use documentation addresses. A mesh spanning a site with no internet access is a real
 topology, and the model has to be able to say it.
 **`gateway:`** — how a segment reaches its parent, and this is where the previous version was
 too thin. It carries three facts, and all three are load-bearing:
@@ -95,6 +141,10 @@ too thin. It carries three facts, and all three are load-bearing:
 - `nat:` — whether addresses are translated. `true` gives the ordinary household case: many
  private machines behind one public address. `false` describes a routed range, where machines
  keep their own addresses and the gateway only forwards.
 - `forwardable:` — whether an inbound mapping can be created. Independent of `nat:`, and the
  field that separates a home gateway from carrier-grade NAT. Publishing through a gateway with
  `forwardable: false` is a declaration error, because that is exactly the constraint being
  reproduced.
 The lab materialises a machine to be the gateway. That is the one implicit machine in an
 otherwise explicit declaration, and it exists because NAT has to run somewhere.
@@ -199,6 +249,61 @@ not first.
 - **Steps.** A scenario is a desired state. Anything expressed as an ordered list of actions
  belongs in the lifecycle, not the declaration.
 ## Is this general? — the axes a setup can vary along
 The question that matters is not *does this cover our mesh*, but **can it express any mesh**.
 Audited against the axes a real deployment varies along, the answer is *most, deliberately not
 all, and three genuine gaps*.
 The standard applied is not "every property a network has". It is **every property that changes
 how the mesh behaves**. Bandwidth does not change correctness; MTU does.
 | Axis | Values | Expressible | |
 |---|---|---|---|
 | **Reachability** | attached · forwardable gateway · unforwardable gateway · isolated | yes | the core of the model |
 | **Address stability** | static · dynamic · changes mid-run | **partly** | a machine can be *moved*, but an address that changes under it cannot be stated |
 | **Gateway depth** | direct · one gateway · nested gateways | **partly** | `to:` chains, so nesting exists; `published:` names one gateway, so forwarding through two does not |
 | **Address family** | IPv4 · IPv6 · dual-stack | **no** | `cidr:` is implicitly v4. A v6-only node is a real topology and cannot be written |
 | **Interfaces per machine** | one · several | **no** | `at:` is singular. A multi-homed node — on a LAN and a WAN at once — is inexpressible |
 | **Path properties** | MTU · latency · loss | **no** | MTU matters: tunnels fragment, and a lower-MTU path is a classic silent failure |
 | **Reachability policy** | symmetric · asymmetric | **no** | a firewall dropping inbound while outbound works is different from NAT and behaves differently |
 | **Gateway state** | permanent · expiring mappings | **no** | mappings time out; whether keepalives work is untestable without it |
 | **Overlapping ranges** | distinct · two sites both on `192.168.1.0/24` | yes | two segments may carry the same range — common, and it breaks routing |
 | **Segment count** | one · many · isolated island | yes | after the `kind:` fix above |
 ### What this says
 **Three gaps are real and should be closed**, in this order:
 1. **Address family.** A v6-only or dual-stack node is not exotic, and a mesh that assumes v4
   fails there completely rather than partially. This is the largest gap.
 2. **Expiring NAT mappings.** Without it, keepalive behaviour is hoped for rather than tested —
   and for a mesh where most nodes sit behind NAT, that is the failure mode most likely to
   appear only after everything has been idle overnight.
 3. **MTU.** Tunnels fragment. A path with a smaller MTU produces a connection that establishes
   and then silently drops large packets, which is exactly the shape of fault this whole effort
   exists to stop shipping.
 **Two are deliberately out of scope** unless something argues otherwise: latency and loss.
 They change performance, not correctness, and a scenario that models them is a network
 simulator rather than a fixture.
 **Two are partial and probably fine for now:** nested forwarding and mid-run address change.
 Both are expressible with small extensions when something needs them, and neither blocks the
 bootstrap scenario.
 ### The honest summary
 The model covers **where a machine sits**, which is what the mesh's reachability logic turns
 on, and it now covers it completely. It does not yet cover **what the path between machines is
 like**, and one of those — address family — is not a refinement but a second world the mesh
 would have to work in.
 None of this blocks phase 0. A bootstrap scenario is one machine and a pinned bundle, and needs
 none of it. But the gaps should be closed before the lab is trusted to say a mesh *works*,
 because today it could only say it works over IPv4, on an unconstrained path, against gateways
 that never forget.
 ## Open
 - **`user` and `edge` profiles have no scenario.** A lab machine is always privileged, so the
@@ -212,7 +317,5 @@ not first.
  writes addresses absolutely. Whether a scenario carries literal addresses or a template the
  lab allocates from decides whether two can run side by side — and there are only three
  documentation ranges to go round.
- **Gateway behaviour beyond forwarding.** A real household gateway also has a NAT table with
+- **The three gaps from the audit above** — address family, expiring NAT mappings, MTU — in
-  timeouts, and connection tracking that drops idle flows. Whether a scenario can express *"the
+  that order. The first is the one that is a second world rather than a refinement.
  gateway forgets a mapping after N seconds"* decides whether keepalive behaviour is testable
  or merely hoped for.