Watch
0
0
Fork
You've already forked hyperhive
0

bao: OIDC login to the browser UI via authelia, as a metadata-only viewer
Some checks were skipped
public bin cache / build + push to preem:grid (push) Has been skipped

The bao UI at bao-ui.<swarm> took a raw store token and nothing else.
It now offers an OIDC tab: authelia's `admins` group logs in and lands
on `swarm-operator-viewer`, which is list+read on `secret/metadata/*`
and nothing under `secret/data/` or `sys/`.

- authelia registers an interactive client `swarm-bao-ui`
  (glue-bao-ui-oidc-client.nix) with redirect
  `https://bao-ui.<swarm>/ui/vault/auth/oidc/oidc/callback`; the secret
  publisher carries its secret to
  `secret/swarm/services/swarm-bao-ui/oidc/client`.
- `swarm-bao-granter-role` (bootstrap token) enables the `oidc` auth
  mount with listing visibility `unauth`, asked before attempted like
  cert/approle; `bao-bootstrap-policy.hcl` gains `sys/auth/oidc`.
- The granter's policy gains `auth/oidc/config`, `auth/oidc/role/swarm-*`
  and read on that one secret leaf. It still holds no `sys/auth`.
- New granting unit `swarm-bao-operator-viewer-policy` writes the viewer
  policy, and once the granter may configure `auth/oidc/config` (checked
  through `sys/capabilities-self`), writes the mount's config from the
  published secret and the role binding `groups=admins` to the viewer.
  Before the bootstrap step re-runs it writes the policy, logs the step
  and exits 0.

Route (a) per mara on #4775: enabling the auth method stays a
bootstrap-token step, re-run once on the live store.

module-eval pins the viewer policy's single metadata stanza, that the
granter's policy has no sys/auth path, the oidc enable in the bootstrap
unit, the exit-0 path, the config/role contents, and the client
registration + publish.
This commit is contained in:
atlas 2026-09-28 19:56:38 +02:00
commit b14ff2796c
9 changed files with 413 additions and 18 deletions

View file

@ -27,6 +27,10 @@ path "sys/auth/approle" {
capabilities = ["create", "update", "sudo"]
}
path "sys/auth/oidc" {
capabilities = ["create", "update", "sudo"]
}
# The granter's own policy and role, and nothing it may write.
path "sys/policies/acl/bao-granter" {
capabilities = ["create", "update"]

View file

@ -27,6 +27,7 @@
./otel.nix
./glue-bao-readers-policy-order.nix
./glue-bao-tls.nix
./glue-bao-ui-oidc-client.nix
./glue-controller-bao-identity.nix
./glue-forge-oidc-client.nix
./glue-grafana-oidc-client.nix

View file

@ -0,0 +1,35 @@
# Glue: register the secret store's browser UI as an OIDC client wherever
# authelia runs.
#
# ONE PAIRING PER FILE — the store's `oidc` auth method ← authelia, and nothing
# else. Deleting this leaves a UI whose OIDC button sends the browser to a
# client authelia has never heard of, and nothing mints the secret
# `swarm-bao-operator-viewer-policy` waits for.
#
# ⚠️ Gated on authelia being HERE, and deliberately NOT on this host running
# the store, for the reason ./glue-grafana-oidc-client.nix gives: a client is a
# row in THIS host's provider config.
{
lib,
config,
...
}:
let
hyperhiveCfg = config.services.hyperhive;
deployCfg = hyperhiveCfg.deploy;
uiCfg = hyperhiveCfg.swarm.bao.ui;
in
{
config = lib.mkIf deployCfg.authelia.enable {
# `kind` is left at its `interactive` default: a person logs in here, and
# that kind is what permits the `profile` and `groups` scopes the store's
# role asks for.
services.hyperhive.swarm.authelia.oidc.clients = [
{
id = uiCfg.oidc.clientId;
description = "HyperHive secret store UI";
redirectUris = [ uiCfg.oidc.redirectUri ];
}
];
};
}

View file

@ -188,11 +188,13 @@ let
#
# The first three are the per-principal grants. The next eight are what
# `swarm-bao-controller-policy` does besides grants: the KV and pki mounts and
# the services root. The last six set up the agent PKI mount: the mount, its
# the services root. The next six set up the agent PKI mount: the mount, its
# root and the `swarm-*` role agent certificates are issued through. No
# `root` delete there: every agent's cert-auth role pins that root by value,
# so replacing it would lock every agent out. No `sys/auth`: the auth mounts
# are created with the bootstrap token by `swarm-bao-granter-role`.
# so replacing it would lock every agent out. The last three configure the
# `oidc` auth method: its config, its `swarm-*` roles, and a read on the one
# client secret that config carries. No `sys/auth`: the auth mounts are
# created with the bootstrap token by `swarm-bao-granter-role`.
#
# Piped as a shell-quoted argument like `controllerPolicyText`, so
# ../module-eval/bao-grants.nix can read it out of the unit script.
@ -264,6 +266,18 @@ let
path "${agentPkiMountPath}/roles/swarm-*" {
capabilities = ["create", "update"]
}
path "auth/oidc/config" {
capabilities = ["create", "update"]
}
path "auth/oidc/role/swarm-*" {
capabilities = ["create", "update"]
}
path "${credentialMountPath}/data/${baoUiClientLeaf}" {
capabilities = ["read"]
}
'';
# What a granting unit prints when the store refuses the granter: the
@ -504,6 +518,54 @@ let
# it; the controller reads it and holds no other copy.
controllerClientLeaf = "swarm/controller/swarm-controller/oidc/client";
# The UI's OIDC client secret. The publisher writes it; the granter reads it
# into the `oidc` auth method's config, which is the only copy bao uses.
baoUiClientLeaf = "swarm/services/${cfg.ui.oidc.clientId}/oidc/client";
# What an OIDC login through the UI gets: the key tree and each key's KV
# metadata, never a value — KV v2 serves values under `data/`, which no
# stanza here names. The UI needs no `sys/` grant on top:
# `sys/internal/ui/mounts` lists any mount the token holds a capability
# under, and the rest it calls is in the `default` policy.
operatorViewerPolicyName = "swarm-operator-viewer";
operatorViewerPolicyText = ''
path "${credentialMountPath}/metadata/*" {
capabilities = ["list", "read"]
}
'';
# authelia's operator group (`operatorGroup` in ./swarm-authelia.nix, the
# group its rule for the UI's vhost admits) → the viewer policy. authelia
# sends `groups` and `preferred_username` from its userinfo endpoint, and
# bao merges those into the ID token's claims before it checks
# `bound_claims`.
operatorViewerRole = builtins.toJSON {
role_type = "oidc";
user_claim = "preferred_username";
groups_claim = "groups";
oidc_scopes = [
"profile"
"groups"
];
bound_claims.groups = [ "admins" ];
allowed_redirect_uris = [ cfg.ui.oidc.redirectUri ];
token_policies = [ operatorViewerPolicyName ];
token_ttl = "1h";
token_max_ttl = "8h";
};
# The `oidc` auth method's config, as `bao write` arguments. The client
# secret is `-`, read from stdin. On a self-signed gateway authelia's
# certificate chains to the host's anchor bundle; bao takes every
# certificate in `oidc_discovery_ca_pem`, not only the first.
operatorViewerOidcConfig = [
"oidc_discovery_url=${toString autheliaCfg.url}"
"oidc_client_id=${cfg.ui.oidc.clientId}"
"oidc_client_secret=-"
"default_role=${operatorViewerPolicyName}"
]
++ lib.optional caTrust.useSelfSigned "oidc_discovery_ca_pem=@${deployCfg.hive-controller.tls.stateDir}/trust-bundle.pem";
# The KV v2 engine the controller writes agent credentials through. Named
# once because the grant above and the `secrets enable` in the bootstrap unit
# have to agree: a policy pointing at a mount nobody created is precisely the
@ -1959,6 +2021,37 @@ in
'';
};
ui.oidc.clientId = lib.mkOption {
type = lib.types.str;
readOnly = true;
default = "swarm-bao-ui";
description = ''
OAuth2 client id the store's `oidc` auth method logs browser users
in as, at authelia.
Swarm-wide and read-only because two hosts have to agree on it:
authelia registers the client (`glue-bao-ui-oidc-client.nix`) and
mints its secret, and the store's host reads that secret back out of
the store under a path composed from this id.
'';
};
ui.oidc.redirectUri = lib.mkOption {
type = lib.types.str;
readOnly = true;
default = "https://${cfg.ui.domain}/ui/vault/auth/oidc/oidc/callback";
defaultText = lib.literalExpression ''"https://''${services.hyperhive.swarm.bao.ui.domain}/ui/vault/auth/oidc/oidc/callback"'';
description = ''
Where authelia sends the browser back to after an OIDC login, and the
URI both authelia and the store's `oidc` role match **exactly**.
The format is the OpenBao UI's own route,
`/ui/vault/auth/<mount>/oidc/callback`, with the mount `oidc`. The
UI composes it from the page's origin, so it only matches when the
gateway serves the UI on port 443.
'';
};
port = lib.mkOption {
type = lib.types.port;
default = 8200;
@ -2196,6 +2289,7 @@ in
"swarm-bao-nats-tls-policy"
"swarm-bao-agent-pki"
"swarm-bao-nats-auth-policy"
"swarm-bao-operator-viewer-policy"
];
# 🚫 No `swarm.otel.scrapeTargets.bao` entry any more, and its absence is
@ -2598,6 +2692,14 @@ in
*) bao auth enable approle ;;
esac
# The UI's OIDC login; `swarm-bao-operator-viewer-policy` writes its
# config and role. Listed on the UI's login page, which shows a
# method as a tab only when it is listed.
case "$mounted" in
*'"oidc/"'*) ;;
*) bao auth enable -listing-visibility=unauth oidc ;;
esac
printf '%s' ${lib.escapeShellArg granterPolicyText} |
bao policy write ${lib.escapeShellArg granterPolicyName} -
@ -3236,6 +3338,84 @@ in
);
};
# The UI's OIDC login: the viewer policy, then the `oidc` auth method's
# config and the role that hands `admins` that policy. The mount itself,
# and the granter's grants on it, exist only once `swarm-bao-granter-role`
# has run with a bootstrap token that carries `sys/auth/oidc`. Until then
# this writes the policy and stops with exit 0: nothing is broken, the UI
# still takes a token, and a day of retries would change nothing.
systemd.services.swarm-bao-operator-viewer-policy = lib.mkIf haveGranter {
description = "write the bao UI's OIDC login: the operator viewer policy, the oidc config and its role";
after = [
"container@${cfg.machine}.service"
"swarm-bao-controller-policy.service"
]
++ granterAfter;
requires = [ "swarm-bao-pki.service" ];
wantedBy = [ "multi-user.target" ];
path = [
baoCli
pkgs.coreutils
];
environment = granterEnv;
# Same unseal wait as its siblings above, for the reason stated there.
startLimitBurst = 2880;
startLimitIntervalSec = 90000;
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
Restart = "on-failure";
RestartSec = 30;
};
script = ''
set -euo pipefail
${granterLogin}
printf '%s' ${lib.escapeShellArg operatorViewerPolicyText} |
bao policy write ${lib.escapeShellArg operatorViewerPolicyName} -
# Asks the granter's own token, through the `default` policy's
# `sys/capabilities-self`, rather than probing the mount: the granter
# holds no `sys/auth` to list mounts with.
caps="$(bao token capabilities auth/oidc/config)"
case "$caps" in
*update*) ;;
*)
echo "the granter may not configure auth/oidc yet (capabilities: $caps), so the UI's OIDC login stays off and token login is unchanged." >&2
echo "one-time step, as root on this host (docs/getting-started/setup.md):" >&2
${lib.concatMapStringsSep "\n" (l: "echo ${lib.escapeShellArg " ${l}"} >&2") (
granterSetupSteps ++ [ "systemctl restart swarm-bao-operator-viewer-policy" ]
)}
exit 0
;;
esac
# Published by ./swarm-secret-publisher.nix on authelia's host,
# which this boot does not wait on, so absence is retried.
if ! secret="$(bao kv get -field=value ${lib.escapeShellArg "${credentialMountPath}/${baoUiClientLeaf}"} 2>"$err")"; then
echo ${lib.escapeShellArg "the store did not return ${credentialMountPath}/${baoUiClientLeaf}: the UI's OIDC client secret is not published yet."} >&2
cat "$err" >&2
exit 1
fi
if [ -z "$secret" ]; then
echo ${lib.escapeShellArg "the store returned an empty ${credentialMountPath}/${baoUiClientLeaf}."} >&2
exit 1
fi
# bao fetches authelia's discovery document on this write, so an
# unreachable or untrusted authelia fails it and the unit retries.
printf '%s' "$secret" |
bao write auth/oidc/config ${
lib.concatMapStringsSep " " lib.escapeShellArg operatorViewerOidcConfig
}
# A JSON body on stdin, because `bound_claims` is a map.
printf '%s' ${lib.escapeShellArg operatorViewerRole} |
bao write auth/oidc/role/${lib.escapeShellArg operatorViewerPolicyName} -
'';
};
# The CA bind source is written at runtime by a host unit, so the
# container has to start after it — otherwise nspawn sets up a mount
# over a file that does not exist yet.

View file

@ -78,6 +78,9 @@ let
# Missing from here, authelia mints the value and nothing carries it, so
# the reader waits on a path that is never written.
hyperhiveCfg.swarm.bao.otel.clientId
# The store's browser UI, whose `oidc` auth method the store's host
# configures with this secret.
hyperhiveCfg.swarm.bao.ui.oidc.clientId
];
# The swarm controller's OIDC client, which it reads back from the store

View file

@ -151,7 +151,7 @@ let
_: u: (u.environment.BAO_CLIENT_CERT or null) == granterCertFile
) baoGrantWithConsumers.systemd.services;
# The twelve units that write a `swarm-*` grant, by name, for the discovery
# The thirteen units that write a `swarm-*` grant, by name, for the discovery
# control below.
grantingUnitNames = [
"swarm-bao-controller-policy"
@ -166,6 +166,7 @@ let
"swarm-bao-nats-tls-policy"
"swarm-bao-agent-pki"
"swarm-bao-nats-auth-policy"
"swarm-bao-operator-viewer-policy"
];
# Comment lines dropped first: both the HCL and the scripts explain
@ -217,7 +218,9 @@ let
];
in
# A login and a seal-status check are unauthenticated: no policy grants them.
if a 0 == "login" || a 0 == "status" then
# A token's own capabilities are `sys/capabilities-self`, which the
# `default` policy grants every token.
if a 0 == "login" || a 0 == "status" || (a 0 == "token" && a 1 == "capabilities") then
null
else if a 0 == "policy" && a 1 == "write" then
need "sys/policies/acl/${a 2}" cu
@ -235,6 +238,18 @@ let
need (a 1) cu
else if a 0 == "read" then
need (a 1) [ "read" ]
# KV v2 inserts `data/` after the mount, the first segment.
else if a 0 == "kv" && a 1 == "get" then
let
segs = lib.splitString "/" (a 2);
in
need (lib.concatStringsSep "/" (
[
(lib.head segs)
"data"
]
++ lib.tail segs
)) [ "read" ]
else if a 0 == "list" then
need (a 1) [ "list" ]
else if a 0 == "delete" then
@ -780,24 +795,24 @@ let
}
{
# A store host without the granter's pair writes its grants some other
# way, so none of the twelve units may exist. Without this arm
# way, so none of the thirteen units may exist. Without this arm
# `lib.mkIf haveGranter` could be dropped from any of them and every other
# case here would still pass.
name = "without the granter's pair none of the twelve granting units render";
name = "without the granter's pair none of the thirteen granting units render";
ok =
let
s = baoGranterOptOut.systemd.services;
in
lib.all (unit: !(s ? ${unit})) (grantingUnitNames ++ [ "swarm-bao-granter-role" ])
# The control: the same store with the pair renders all twelve.
# The control: the same store with the pair renders all thirteen.
&& lib.all (unit: baoGrantHere.systemd.services ? ${unit}) grantingUnitNames;
}
{
# 🩸 What replaced the silent skip. With no bootstrap token the twelve still
# 🩸 What replaced the silent skip. With no bootstrap token the thirteen still
# render, and a refused granter fails them with the step that fixes it.
# A store host that never named a token is told to name one, since the
# unit that sets the granter up renders only where it has.
name = "a store host without a bootstrap token renders the twelve, each failing loudly with the one-time step";
name = "a store host without a bootstrap token renders the thirteen, each failing loudly with the one-time step";
ok =
let
s = baoGranterNoToken.systemd.services;
@ -876,7 +891,7 @@ let
{
# The granter's grants, whole. Pinned as the full list, because an added
# path or capability is exactly what a presence check misses.
name = "the granter's policy is exactly these seventeen stanzas";
name = "the granter's policy is exactly these twenty stanzas";
ok =
let
cu = [
@ -956,6 +971,18 @@ let
path = "pki-agents/roles/swarm-*";
caps = cu;
}
{
path = "auth/oidc/config";
caps = cu;
}
{
path = "auth/oidc/role/swarm-*";
caps = cu;
}
{
path = "secret/data/swarm/services/swarm-bao-ui/oidc/client";
caps = [ "read" ];
}
];
}
{
@ -979,7 +1006,11 @@ let
"auth/cert/certs/hive-x"
"sys/auth"
"sys/auth/cert"
"sys/auth/oidc"
"sys/auth/oidc/tune"
"sys/auth/x"
"auth/oidc/role/x"
"secret/data/swarm/services/x/oidc/client"
"auth/token/create"
"auth/token/create-orphan"
"secret/data/x"
@ -1185,8 +1216,8 @@ let
}
{
# What makes the case above mean something: discovery by the granter's
# certificate reaches all twelve units, and each yields calls.
name = "the granter-policy check sees all twelve granting units, and parses calls from each";
# certificate reaches all thirteen units, and each yields calls.
name = "the granter-policy check sees all thirteen granting units, and parses calls from each";
ok =
lib.sort lib.lessThan (lib.attrNames granterUnits) == lib.sort lib.lessThan grantingUnitNames
&& lib.all (u: baoCalls u.script != [ ]) (lib.attrValues granterUnits)
@ -1249,6 +1280,7 @@ let
"sys/auth"
"sys/auth/cert"
"sys/auth/approle"
"sys/auth/oidc"
"sys/policies/acl/bao-granter"
"auth/cert/certs/bao-granter"
]
@ -1344,6 +1376,99 @@ let
lib.any (a: !a.assertion && names a) baoGranterOddAgentRole.assertions
&& !(lib.any (a: !a.assertion && names a) baoGrantHere.assertions);
}
# ── the UI's OIDC login ─────────────────────────────────────────────────
{
# What an `admins` login through the UI holds: the key tree and KV
# metadata. Any `data/` path would hand a browser session every secret
# in the store, and any `sys/` one more than the UI needs.
name = "the operator viewer policy is list and read on KV metadata, and nothing under data/ or sys/";
ok =
let
viewer = grantsIn baoGrantHere.systemd.services.swarm-bao-operator-viewer-policy.script;
in
viewer == [
{
path = "secret/metadata/*";
caps = [
"list"
"read"
];
}
]
&& grantFor viewer "secret/data/swarm/agents/x/queue" == null
&& grantFor viewer "secret/metadata/swarm/agents/x/queue" != null
&& !(lib.any (g: lib.hasPrefix "secret/data" g.path || lib.hasPrefix "sys/" g.path) viewer);
}
{
# Route (a): enabling an auth method stays a bootstrap-token act. The
# granter configures the `oidc` mount; it never creates or tunes one.
name = "the granter's policy has no sys/auth path, and the bootstrap policy holds sys/auth/oidc";
ok =
!(lib.any (g: lib.hasPrefix "sys/auth" g.path) granterGrants)
&& grantFor bootstrapGrants "sys/auth/oidc" != null;
}
{
# Asked before attempted, like the cert and approle mounts, so re-running
# the step on a store that has the mount is a no-op; listed, or the UI's
# login page shows no OIDC tab.
name = "the granter's role unit enables the oidc mount once, listed on the UI's login page";
ok =
let
g = baoGrantHere.systemd.services.swarm-bao-granter-role.script;
in
lib.hasInfix "*'\"oidc/\"'*) ;;" g
&& lib.hasInfix "bao auth enable -listing-visibility=unauth oidc" g;
}
{
# Before the step the granter lacks `auth/oidc/*`: the unit writes the
# policy, says what to run, and exits 0 rather than retrying for a day.
# The policy write comes first so it lands either way; the capability
# check comes before the secret read and the config write it guards.
name = "the viewer unit writes its policy, then stops with exit 0 while the granter may not configure oidc";
ok =
let
s = baoGrantHere.systemd.services.swarm-bao-operator-viewer-policy.script;
at = needle: lib.stringLength (lib.head (lib.splitString needle s));
probe = "caps=\"$(bao token capabilities auth/oidc/config)\"";
in
lib.hasInfix probe s
&& at "bao policy write swarm-operator-viewer -" < at probe
&& at probe < at "exit 0"
&& at "exit 0" < at "bao kv get"
&& at "bao kv get" < at "bao write auth/oidc/config"
&& lib.hasInfix "systemctl restart swarm-bao-operator-viewer-policy" s;
}
{
# The client secret is on stdin, never an argument in /proc; the rest is
# the swarm's authelia and the UI's own client id.
name = "the oidc config names authelia and the UI's client, with the secret on stdin";
ok =
let
s = baoGrantHere.systemd.services.swarm-bao-operator-viewer-policy.script;
in
lib.hasInfix "printf '%s' \"$secret\" |\n bao write auth/oidc/config" s
&& lib.hasInfix "'oidc_client_secret=-'" s
&& lib.hasInfix "'oidc_discovery_url=https://auth.t.local'" s
&& lib.hasInfix "'oidc_client_id=swarm-bao-ui'" s
&& lib.hasInfix "'default_role=swarm-operator-viewer'" s
&& lib.hasInfix "bao kv get -field=value secret/swarm/services/swarm-bao-ui/oidc/client" s
&&
(lib.hasInfix "oidc_discovery_ca_pem=@" s) == baoGrantHere.services.hyperhive.gateway.useSelfSigned;
}
{
# `admins` → the viewer policy, and only at the UI's own callback.
name = "the oidc role binds authelia's admins group to the viewer policy at the UI's callback";
ok =
let
s = baoGrantHere.systemd.services.swarm-bao-operator-viewer-policy.script;
in
lib.hasInfix "bao write auth/oidc/role/swarm-operator-viewer -" s
&& lib.hasInfix ''"bound_claims":{"groups":["admins"]}'' s
&& lib.hasInfix ''"groups_claim":"groups"'' s
&& lib.hasInfix ''"token_policies":["swarm-operator-viewer"]'' s
&& lib.hasInfix ''"allowed_redirect_uris":["https://bao-ui.t.local/ui/vault/auth/oidc/oidc/callback"]'' s;
}
];
in
runGroup "bao-grants" cases

View file

@ -157,6 +157,28 @@ let
secretPublisherHere.systemd.services.swarm-secret-publish.script
);
}
{
# The store's UI login: authelia registers the client with the callback
# the store's `oidc` role allows, and the publisher carries its secret
# to the path `swarm-bao-operator-viewer-policy` reads. Not registered
# where authelia is not.
name = "the store UI's client is registered interactive at its callback, and the publisher carries its secret";
ok =
let
ui = lib.findFirst (
c: c.id == "swarm-bao-ui"
) null secretPublisherHere.services.hyperhive.swarm.authelia.oidc.clients;
in
ui != null
&& ui.kind == "interactive"
&& ui.redirectUris == [ "https://bao-ui.t.local/ui/vault/auth/oidc/oidc/callback" ]
&& lib.hasInfix "secret/swarm/services/swarm-bao-ui/oidc/client" (
secretPublisherHere.systemd.services.swarm-secret-publish.script
)
&& !(lib.any (
c: c.id == "swarm-bao-ui"
) grafanaRemoteAuthelia.services.hyperhive.swarm.authelia.oidc.clients);
}
{
# The same hole the controller's case above names, open a second time: the
# PKI script grew a third leaf and no case read it.