vinzenz/hyperhive
vinzenz/hyperhive
Watch
0
0
Fork
You've already forked hyperhive
 0
Code Activity

swarm-bao: agent certificates issued by a store-generated agent CA

Browse source 
An agent's store identity was signed in swarm-controller's memory by a CA
a controller-host unit generated on disk, and the listener never trusted
that CA. Agent leaves now come from the store itself: a `pki-agents` PKI
mount whose root openbao generates internally, so the agent CA's key
never exists outside the store.

- swarm-bao-agent-pki (new, store host, as the bao granter): enables and
  tunes the mount, generates the root once (guarded on an empty issuer
  list, no replace branch), upserts the `swarm-agent` role (client
  certificates named `hive-agent-*` only, 90 days), caches the CA at
  /var/lib/swarm-bao-tls/agent-ca.pem and composes the listener bundle.
- The listener's tls_client_ca_file is a new listener-client-ca.pem
  (client-ca.pem, then the agent CA). Host cert-auth roles still pin
  client-ca.pem, so an agent leaf satisfies no host role. swarm-bao-certs
  composes the same bundle before openbao starts.
- openbao reads tls_client_ca_file only at start, so when the bundle
  changed after openbao started, swarm-bao-agent-pki restarts
  openbao.service in the container; under `seal = "shamir"` it prints
  the step instead. Once swarm-bao-certs has a cached CA, later boots
  start openbao with it and do not restart.
- The controller policy gains exactly `update` on
  pki-agents/issue/swarm-agent. mint_and_verify now asks that role for
  the leaf (the store generates the key), writes the agent's cert-auth
  role pinning the issuing CA bao returned, and writes the agent's
  policy as render_agent alone: the hive-shared queue credential stanza
  is gone.
- deploy.bao.agentPkiRoleName (must start `swarm-`, asserted with the
  other pki role names); swarm-controller gets
  SWARM_CONTROLLER_AGENT_PKI_MOUNT/_ROLE from the deploy.bao options.

Deleted: swarm-controller-agent-ca and its options (agentCaFile,
agentCaKeyFile), env, LoadCredential entries and assertion;
agent_identity's Authority, rcgen signing and validity window; the
rcgen and time dependencies of swarm-controller (rcgen leaves the
workspace); policy::render_agent_with_queue and its tests. The CN-prefix
assertion policy.rs said was owed is not: agent and host roles pin
different CAs.

Migration is re-creating each agent after deploy; that overwrites the
stale role and policy.

Closes #4756
This commit is contained in:
atlas 2026-09-27 19:30:38 +02:00
commit 6170e74a31
16 changed files with 894 additions and 925 deletions
Download patch file Download diff file Expand all files Collapse all files

162
swarm-secret-client/src/policy.rs
View file

@ -70,13 +70,12 @@ pub fn hive_object_name(hive: &str) -> Result<String, Error> {
/// client ids; this is a second identifier family leaning on it, which is why
/// the fragment list is what to read before renaming either.
///
/// ⚠️ The controller's and publisher's subjects are *not* covered by that: their
/// policy names are literals outside `hive-`, but their **common names** are
/// operator-set options (`deploy.bao.controllerCommonName`,
/// `secretPublisherCommonName`) that nothing here can see, and an operator may
/// spell one `hive-agent-atlas`. Whichever change first mints an agent leaf
/// owes the assertion that neither starts with this prefix — `swarm.nix`'s
/// `certAuthCns` is where the mirror-image check for hive names lives.
/// Host principals' **common names** are operator-set options
/// (`deploy.bao.controllerCommonName`, `secretPublisherCommonName`, …) that
/// may spell this prefix, and that is harmless: an agent's cert-auth role pins
/// the store's agent CA (`deploy.bao.agentPkiMountPath`), which signs no host
/// leaf, and every host role pins `deploy.bao.clientCaFile`, which signs no
/// agent leaf. A CN match alone logs nobody in.
pub const AGENT_PREFIX: &str = "hive-agent-";
/// The policy and cert-auth role name for `agent`, and the common name of the
@ -203,43 +202,6 @@ pub fn render_agent(agent: &str) -> Result<String, Error> {
)))
}
/// Render `agent`'s policy document: read on that one agent's credentials and
/// on the hive's shared queue credential.
///
/// Extends [`render_agent`] with a second stanza granting read on
/// `swarm/hives/<hive>/queue/agent`. The queue credential is **hive-shared,
/// not per-agent** — every agent in a hive authenticates to the queue with the
/// same client secret (`queue.rs:1-8`), so a policy scoped strictly to
/// `agents/<agent>/*` cannot read it and an in-container pull would fail. That
/// hive-shared credential is already handed to every agent container on that
/// hive by the host today, so this grant adds no new authority — it merely
/// makes the existing capability reachable through the agent's own token
/// instead of requiring the credential to be delivered out of band.
///
/// ⚠️ **Every agent in a hive can read that hive's queue credential.** This is
/// not new authority (the host already provides this exact value to all agents
/// on the hive), but it is a documented property: an agent policy grants read
/// on a path shared across every agent on its hive, not on a path unique to
/// that agent alone.
///
/// Read-only, for the same reason [`render_agent`]'s is: an agent that could
/// write credentials could hand itself an identity it was never issued.
///
/// # Errors
/// [`Error::PathSegment`] when `agent` or `hive` holds anything but
/// `[A-Za-z0-9_-]` — both are interpolated into policy paths, so a name that
/// could close a stanza could grant itself anything.
pub fn render_agent_with_queue(agent: &str, hive: &str) -> Result<String, Error> {
checked_segment("agent", agent)?;
let agent_stanza = read_stanza(&format!(
"{MOUNT}/data/{ROOT}/{}/{agent}/*",
<&str>::from(Kind::Agent)
));
let queue_path = crate::queue::agent_client_path(hive)?;
let queue_stanza = read_stanza(&format!("{MOUNT}/data/{queue_path}"));
Ok(format!("{agent_stanza}{queue_stanza}"))
}
#[cfg(test)]
mod tests {
use super::*;
@ -430,6 +392,22 @@ mod tests {
);
}
#[test]
fn an_agents_document_is_exactly_its_own_read_stanza() {
// Pinned byte for byte: an added stanza (a hive's queue credential, a
// second agent) is exactly what a presence check misses.
assert_eq!(
render_agent("atlas").expect("legal"),
"path \"secret/data/swarm/agents/atlas/*\" {\n capabilities = [\"read\"]\n}\n"
);
// The control: the pin discriminates between agents.
assert!(
!render_agent("other")
.expect("legal")
.contains("agents/atlas/")
);
}
#[test]
fn an_agents_grant_is_read_only() {
// An agent that could write its own credentials could hand itself an
@ -517,102 +495,6 @@ mod tests {
assert!(hive.contains("path \"secret/data/swarm/agents/*\""));
}
#[test]
fn an_agents_document_with_queue_grants_both_paths() {
// The happy path: the document grants read on the agent's own namespace
// and on the hive's queue credential.
let p = render_agent_with_queue("atlas", "pr1ma").expect("legal");
assert!(
p.contains("path \"secret/data/swarm/agents/atlas/*\""),
"must grant the agent's own path: {p}"
);
let expected_queue_path = format!(
"path \"{MOUNT}/data/{}\"",
crate::queue::agent_client_path("pr1ma").expect("legal")
);
assert!(
p.contains(&expected_queue_path),
"must grant the hive's queue credential: {p}"
);
assert_eq!(
p.matches("path \"").count(),
2,
"two stanzas, one for the agent and one for the queue: {p}"
);
}
#[test]
fn an_agents_document_with_queue_is_read_only() {
// An agent that could write the queue credential could hand every agent
// on its hive an identity they were never issued.
let p = render_agent_with_queue("atlas", "pr1ma").expect("legal");
for capability in ["create", "update", "delete", "list", "sudo", "patch"] {
assert!(!p.contains(capability), "must not grant {capability}: {p}");
}
assert!(p.contains("capabilities = [\"read\"]"));
}
#[test]
fn an_agent_name_with_traversal_in_the_queue_variant_is_refused() {
// The agent parameter is an injection surface in both renderers, so
// refusing a traversal here proves the new one validates it.
assert!(
render_agent_with_queue("atlas/*\" { capabilities = [\"root\"] }", "pr1ma").is_err()
);
assert!(render_agent_with_queue("", "pr1ma").is_err());
// The control: legal names still work.
assert!(render_agent_with_queue("a-b_C9", "pr1ma").is_ok());
}
#[test]
fn a_hive_name_with_traversal_in_the_queue_variant_is_refused() {
// The hive parameter is a second injection surface that only the queue
// variant introduces, so this test proves that new parameter is
// validated. A name that could close the stanza could grant the agent
// anything.
assert!(
render_agent_with_queue("atlas", "pr1ma/*\" { capabilities = [\"root\"] }").is_err()
);
assert!(render_agent_with_queue("atlas", "").is_err());
assert!(
render_agent_with_queue("atlas", "../services/swarm-grafana").is_err(),
"a path traversal that could reach a different kind"
);
// The control: legal names still work.
assert!(render_agent_with_queue("atlas", "a-b_C9").is_ok());
}
#[test]
fn the_queue_variant_does_not_widen_the_agent_stanza() {
// The queue grant must not cause the agent stanza to widen from
// `agents/<agent>/*` to `agents/*` — that would give every agent every
// other agent's credentials.
let p = render_agent_with_queue("atlas", "pr1ma").expect("legal");
assert!(
!p.contains("swarm/agents/*"),
"must not grant the whole agent prefix: {p}"
);
assert!(p.contains("swarm/agents/atlas/*"));
}
#[test]
fn the_queue_variant_does_not_grant_the_whole_hive_prefix() {
// The queue stanza must grant only the queue credential path, not
// `hives/<hive>/*` — the latter would give the agent read on every
// secret of the hive that hosts it.
let p = render_agent_with_queue("atlas", "pr1ma").expect("legal");
assert!(
!p.contains("swarm/hives/*"),
"must not grant the whole hive prefix: {p}"
);
assert!(
!p.contains("swarm/hives/pr1ma/*"),
"must not grant the hive's whole path: {p}"
);
let expected_queue_path = crate::queue::agent_client_path("pr1ma").expect("legal");
assert!(p.contains(&expected_queue_path));
}
#[test]
fn only_agent_roles_come_back_and_without_their_prefix() {
let roles: Vec<String> = [