geth/README.md

299 lines
15 KiB
Markdown

# geth
`geth` is a personal, local-first mesh runtime for scripts, devices, databases,
documents, blobs, pipes, and future multi-user collaboration.
This project is not the Ethereum `geth` client. The project and executable are
still named `geth`.
## One Binary
There is one executable: `geth`.
It has daemon mode and control mode:
```sh
geth init
geth daemon run
geth daemon service install
geth status
geth node id
geth resource list
geth cas add ./file
```
The daemon owns local identity, the Iroh endpoint, trust state, resource
registry, module router, local metadata store, and synchronized data structures.
Most non-daemon commands talk to the daemon through a local Unix socket at
`$GETH_HOME/run/geth.sock`.
`geth keychain init --admin-key <public-key> --signing-key <private-key>` records
the keychain initialization/admin-key operations and signs their canonical
payloads through `ssh-keygen -Y sign` using the
`geth.keychain.v1@geth.local` namespace. This is the bootstrap path for
admin/YubiKey-rooted trust. `geth keychain status` reports the number of stored
keychain signatures plus how many currently verify with OpenSSH; rejecting
unsigned or invalid replicated keychain ops is still future work.
The daemon can also install itself as a user service:
```sh
geth daemon service install
geth daemon service status
geth daemon service uninstall
```
The bootstrap service managers are systemd user units on Linux, launchd user
agents on macOS, and per-user scheduled tasks on Windows. These are user-level
services, not system services.
## Transport And SSH
All remote node-to-node geth communication is designed to happen over Iroh only.
SSH is not a geth transport backend, and there is no SSH fallback transport.
The default node config uses Iroh's default relay policy for practical
connectivity; set `[iroh].relay_mode = "disabled"` for local-only/offline
development. Named custom relay maps can be selected with
`relay_mode = "custom"` and `relay_map = "<name>"`. Iroh local-network
discovery is enabled by default with `[iroh].local_discovery = true`.
SSH keys are used as admin trust anchors and ecosystem integration points.
OpenSSH, FIDO, and YubiKey-backed keys can sign geth trust objects through
canonical geth envelopes with explicit namespaces such as
`geth.keychain.v1@geth.local`. Future SSH proxying may carry SSH protocol bytes
over authorized Iroh streams, but the geth transport remains Iroh.
SSH certificate request and renewal flows are managed as geth metadata. A node
can create a certificate request, another machine can approve it and receive an
explicit `ssh-keygen -s ...` command suitable for a CA key or YubiKey-backed CA,
or pass `--sign` to run `ssh-keygen` immediately and import the resulting
`-cert.pub` for distribution. Certificate and key revocation entries are tracked
locally and can be exported as JSONL or as an OpenSSH KRL specification file or
a binary OpenSSH KRL generated through `ssh-keygen -k`. `geth ssh cert sync
<node-id>` and
`geth ssh revocation sync <node-id>` pull certificate-flow and revocation
metadata from an authorized peer over Iroh.
## MVP Features
The bootstrap implementation provides:
- `geth init`
- `geth daemon run`
- `geth daemon service install|uninstall|start|stop|status|print`
- `geth status`
- `geth node id`
- `geth peer export [--out <path>]`
- `geth peer import <path>`
- `geth peer list`
- `geth peer ping <node-id>`
- `geth peer auth-check <node-id> <resource> <capability>`
- `geth resource list`
- `geth resource create <kind> <name>`
- `geth keychain init [--admin-key <path>] [--signing-key <path>]`
- `geth keychain status`
- `geth secret status`
- `geth secret create <resource>`
- `geth secret rotate <resource>`
- `geth secret bearer create <resource> --capability <capability>`
- `geth secret bearer list`
- `geth secret bearer challenge <resource> --capability <capability>`
- `geth secret bearer prove <token> <resource> --nonce <nonce> --capability <capability>`
- `geth secret bearer verify <token> <resource> --nonce <nonce> --response <response> --capability <capability>`
- `geth secret bearer revoke <resource> <bearer-id>`
- `geth auth explain <subject> <resource> <capability>`
- `geth auth grant <subject> <resource> <capability> [--grant-id <id>]`
- `geth auth revoke <resource> <grant-id>`
- local filesystem CAS commands: `add`, `get`, `fetch`, `hash`, `has`, `pin`,
`unpin`, `cleanup`, `providers`, `list`; remote fetch accepts
`--bearer-secret <secret>`
- local CAS tree objects describe file trees and are stored as CAS blobs
- local file-root commands: `geth cas root add/list/scan/sync/apply`; root sync
pulls authorized remote tree metadata and CAS tree bytes into a peer-qualified
remote root, and apply materializes a tree without deleting files or
overwriting local edits
- local file conflict metadata commands:
`geth cas conflict record/list/resolve`
- local DB resource registration: `geth db add <name> <path>` and
`geth db status <name>` with schema and `crsql_changes` metadata; the DB
crate and daemon can extract typed local `crsql_changes` batches through
`geth db changes <name>` and exchange authorized remote batches with
`geth db sync <node-id> <name>`
- local SQLite-backed KV commands: `geth kv create/set/get`; `kv set` accepts
`--subject <principal>` to exercise local capability checks for non-local
callers; `geth kv sync <node-id> <name> [--bearer-secret <secret>]` pulls
authorized remote updates
- local JSON document commands: `geth document create/status/set/get`; `geth
document sync <node-id> <name> [--bearer-secret <secret>]` pulls authorized
remote JSON state
- local daemon-lifetime pubsub snapshots: `geth pubsub pub/sub`; `geth pubsub
pub <topic> <message> --node <node-id>` publishes to an authorized peer;
`geth pubsub sub <topic> --node <node-id>` reads an authorized peer snapshot
- SSH certificate flow metadata:
- `geth ssh cert request --public-key <path> --principal <name> [--subject <principal>]`
- `geth ssh cert requests [--subject <principal>]`
- `geth ssh cert approve <request-id> --ca-key <path> [--sign] [--subject <principal>]`
- `geth ssh cert import <request-id> --cert <path> [--subject <principal>]`
- `geth ssh cert list [--subject <principal>]`
- `geth ssh cert sync <node-id> [--bearer-secret <secret>]`
- `geth ssh revocation add <kind> <target> [--subject <principal>]`
- `geth ssh revocation list [--subject <principal>]`
- `geth ssh revocation export --out <path> [--format jsonl|openssh-krl-spec|openssh-krl] [--subject <principal>]`
- `geth ssh revocation import <path> [--format jsonl|openssh-krl-spec] [--subject <principal>]`
- `geth ssh revocation sync <node-id> [--bearer-secret <secret>]`
- SSH proxy authorization probe: `geth ssh proxy <node-id> [--bearer-secret <secret>]`
- pipe registry/connect commands:
`geth pipe listen <name> [--node <node-id>] [--bearer-secret <secret>]` and
`geth pipe connect <name> [--node <node-id>] [--bearer-secret <secret>]`
`geth peer export/import/list` is for untrusted peer-card exchange. Peer cards
include the Iroh EndpointID plus currently known relay/direct addresses.
`geth peer ping <node-id>` uses the local daemon's Iroh endpoint to dial an
imported peer card and exchange a signed candidate-only peer-card ping.
`geth peer auth-check <node-id> <resource> <capability>` sends a protected
Iroh control request: the remote daemon verifies that the caller's signed peer
card binds the actual Iroh EndpointID before reducing resource-local auth ops.
`geth cas fetch <node-id> <hash>` uses the same protected Iroh control path to
request a blob from a peer. The remote daemon only returns bytes when the caller
has `cas.fetch` on `resource:cas:local`, and the caller verifies that the bytes
hash to the requested BLAKE3 CAS hash before storing them locally. Successful
fetches record the serving peer as a local provider, visible with
`geth cas providers <hash>`. This is the bootstrap transfer path; future work
will move provider/fetch behavior to `iroh-blobs`.
Remote resource commands that accept `--bearer-secret` can also authorize with a
resource-scoped bearer proof generated from the private bearer token returned at
creation time. The persisted auth log stores a public bearer id and token
verifier, not the private token. This does not enroll the caller as a trusted
node; it only unlocks the requested capability on that one resource.
`geth ssh cert sync <node-id>` requires `ssh_cert.sync` on `resource:ssh:certs`
at the peer. `geth ssh revocation sync <node-id>` requires
`ssh_revocation.sync` on `resource:ssh:revocations`. Both commands merge
authorized peer metadata into the local store for offline listing and later
approval/signing workflows. While the daemon is running, it also performs a
background live-sync tick for known peers every 30 seconds. Live-sync stores
per-peer high-water cursors in local metadata so repeated ticks request only
newer SSH certificate-flow and revocation records. Sync import preserves local
metadata by rejecting conflicting records with ids that already exist locally.
Before probing individual modules, the daemon asks the peer for an authorized
sync-status summary over Iroh. The peer only returns stream watermarks for
resources where the caller already has the matching capability, letting the
local daemon skip unchanged or unauthorized streams.
File roots advertise `cas-tree:<name>` watermarks when the caller has
`cas.fetch`; background live-sync imports updated tree metadata and CAS tree
bytes into peer-qualified remote roots without writing files.
`geth cas root apply <root> --to <path>` can then materialize that tree locally:
it creates missing directories/files, never deletes extra files, never
overwrites differing local files, and records conflicts for manual resolution.
Named KV stores participate in the same live-sync loop once they exist locally:
manual `geth kv sync <node-id> <name>` and background ticks require `kv.read`
on the remote `resource:kv:<name>` and import only remote entries that are not
older than the local value.
Remote pubsub publish uses the protected Iroh control path too. The remote peer
requires `pubsub.publish` on `resource:pubsub:<topic>` before recording the
message in its local daemon-lifetime ring buffer. Pubsub remains lossy and is
not durable storage. Remote pubsub subscribe uses the same protected path and
requires `pubsub.subscribe` on `resource:pubsub:<topic>` before returning the
peer's current daemon-lifetime snapshot for that topic.
Remote pipe connect uses the same protected Iroh control path and requires
`pipe.connect` on `resource:pipe:<name>`. The current prototype records a remote
connection attempt and whether a listener exists; byte streaming and forwarding
are still future work.
Remote pipe listen uses the same protected path:
`geth pipe listen <name> --node <node-id>` requires `pipe.listen` on
`resource:pipe:<name>` before registering a daemon-lifetime listener on the
peer.
`geth ssh proxy <node-id>` also uses the protected Iroh control path. The remote
peer validates the caller's endpoint/card binding and requires
`ssh_proxy.connect` on `resource:ssh-proxy:local` before returning proxy
connection metadata. The current prototype does not carry SSH bytes or connect
to remote sshd yet; it only proves the authorization gate.
Document sync is a bootstrap JSON last-writer-wins path before Automerge:
manual `geth document sync <node-id> <name>` and background live-sync require
`document.read` on `resource:document:<name>` and import only state that is not
older than the local document timestamp.
DB sync is a staged cr-sqlite path: manual `geth db sync <node-id> <name>` and
background live-sync require `db.sync` on `resource:db:<name>`, exchange typed
`crsql_changes` batches over the protected Iroh control path, and check remote
schema metadata against the local DB before applying and advancing the per-peer
cursor. Compatible batches are inserted into the local `crsql_changes` table or
view; for real cr-sqlite databases, loading/configuring cr-sqlite remains the
database owner's responsibility.
Importing or pinging a peer card never grants capabilities by itself.
When `[iroh].local_discovery = true`, the daemon also advertises and discovers
signed peer cards on LAN using a geth-specific mDNS TXT payload. That payload is
candidate metadata only; all geth node-to-node requests still run over Iroh.
## Resource Modules
Everything meaningful is modeled as a resource. Planned resource kinds are:
- `db`: SQLite/cr-sqlite synchronization
- `kv`: Iroh Documents backed key-value stores
- `pipe`: dumbpipe-like byte streams over Iroh; the bootstrap has a local
daemon registry only
- `document`: Automerge documents over Iroh streams
- `pubsub`: lossy notifications, not authoritative storage; the bootstrap
keeps only an in-memory daemon-lifetime ring buffer
- `cas`: content-addressed blob storage and distribution
- `ssh-proxy`: authorized SSH proxy/admin access over Iroh
Authorization is resource-scoped and capability-based. Bearer secrets may grant
specific resource capabilities but do not create trusted node identity. The auth
evaluator supports scoped KV write grants such as `kv.write_prefix:apps/foo/`
for `kv.write_key:apps/foo/config` explain checks. `geth kv set --subject
<principal>` enforces those local grants for test callers; the local node/agent
still has owner access for local administration. SSH certificate and revocation
commands also accept `--subject <principal>` on local metadata operations to
exercise the same capability checks: certificate requests/read/approval/import
use `ssh_cert.request`, `ssh_cert.read`, `ssh_cert.approve`, and
`ssh_cert.import` on `resource:ssh:certs`, while revocation publish/read/import
use `ssh_revocation.publish`, `ssh_revocation.read`, and
`ssh_revocation.import` on `resource:ssh:revocations`.
## Local State
If `GETH_HOME` is set, geth uses it. Otherwise it uses an OS-specific data
directory. The bootstrap layout is:
```text
$GETH_HOME/
geth.sqlite
config.toml
identity/agent.ed25519
identity/iroh.ed25519
cas/blobs/
run/geth.sock
```
## Quick Start
In one shell:
```sh
export GETH_HOME="$(mktemp -d)"
cargo run -p geth -- init
cargo run -p geth -- daemon run
```
In another shell:
```sh
export GETH_HOME="<same dir>"
cargo run -p geth -- status
cargo run -p geth -- node id
echo "hello geth" > /tmp/hello-geth.txt
cargo run -p geth -- cas add /tmp/hello-geth.txt
cargo run -p geth -- cas list
```
## Authorization Direction
The MVP defines the split between:
- keychain: SSH-rooted users, devices, nodes, agents, and endpoint bindings
- auth: resource-local signed authorization operations and capability grants
- secrets: resource master secrets, epochs, envelopes, and bearer access
The current code does not implement Keyhive, BeeKEM, strong forward secrecy, or
post-compromise security. It leaves room for future local-first, replicated auth
logs and BeeKEM/CGKA-style group key evolution.