- bump sso-manager-node submodule gitlink v1.17.1 -> v1.17.2
(post-deploy fixes: auto-slug plugins, schedule dropdown, /profile
rendering, plugin-edit persistence, nmap in image, SMS/TOS on /conf,
sso-side /vault policy grants)
- setup.sh: add sso-admin list grant on secret/metadata (KV mount root)
so the /vault secrets list no longer 403s for admins
- setup.sh: ensure_policy now always (re)writes the policy so policy
edits apply on a re-run instead of stranding the old HCL
- CHANGELOG embeds the full sso v1.17.2 changelog
Co-Authored-By: Claude <noreply@anthropic.com>
Prerequisite for the SSO Manager plugin system (shipped in sso-manager-node
v1.17.0). Adds secret/data/plugins/* (CRUD+list) + secret/metadata/plugins/*
(list/read/delete) to the sso-broker policy HCL so the SSO can store per-instance
plugin secrets in OpenBao instead of sso-secrets.js. ensure_policy is idempotent,
so re-running ./setup.sh grants the existing SSO_VAULT_TOKEN live.
Docs: secrets.md (Plugin secrets section + policy row), architecture.md.
Co-Authored-By: Claude <noreply@anthropic.com>
Rename the project to theta-suite (it is now an integrated suite of four
apps around a shared OpenBao secrets store, not a two-project env).
- theta-env -> theta-suite across the superproject: _config.yml (title +
baseurl /theta-suite + repo URLs), README, setup.sh (incl. the
THETA_SUITE_REEXECED self-update sentinel), docker-compose.yml,
bootstrap.js, lint.yml, config.example/*, docs/robots.txt, all docs,
this changelog.
- architecture.md rewritten: real 4-service + ldap-client topology, OpenBao
secrets section, OpenBao-aware config flow; removed "two containers" /
"three repos" / LDAP-"legacy" framing.
- index.md: integrated-suite framing + secrets/OpenBao + ldap-client.
- standalone.md + README: standalone reframed as advanced opt-in.
- sso-manager-node submodule -> v1.16.1 (401 fix on /conf and /vault).
Co-authored-by: Claude <noreply@anthropic.com>
Two fresh-install fixes and promote the SSH jump host from opt-in to core.
setup.sh: fix silent abort after "Minting per-app OpenBao tokens". env_get's
grep|cut pipeline returns non-zero under set -euo pipefail when .env exists
(created by the root VAULT_TOKEN env_upsert) but an app-token key is absent
(the normal first-run state); the unguarded existing assignment from env_get
then tripped set -e and killed the script before minting any token. env_get
now always returns 0 (|| true). Reproduced + verified under the exact condition.
jump host is no longer optional:
- docker-compose.yml: drop profiles jump-host from the jump-host service
(always started); rename the opt-in test fixture profile jump-host to ldap-test.
- setup.sh: SUBMODULES always includes jump-host; build/start/register/summary
no longer guarded by JUMP_ENABLED; drop the COMPOSE_PROFILES export.
- bootstrap.js: jump provisioning + directory record run unconditionally.
- setup.env.example/docs: drop optional/CFG_JUMP_HOST_ENABLED wording.
Co-authored-by: Claude <noreply@anthropic.com>
Bumps submodules to their latest releases:
- jump-host v1.13.0: Title changed to 'SSO Manager'
- sso-manager-node v1.13.0: Directory page cleanup, users list key column fix
- proxy v1.11.0: Table-based list views, auto-refresh groups
Bumps the SSO to the release that closes the end-user half of the directory
(catalog, self-service access requests, admin access visibility) and adds
nested LDAP groups, and ldap-client to the release that makes SSSD resolve
that nesting on hosts pointed at a server without the nestgroup overlay.
Operational note: the SSO image now compiles OpenLDAP from a pinned master
commit, because nestgroup exists only on master -- no 2.6.x release ships it.
That makes the image slower to build, and master's LMDB 1.0.0 cannot read the
0.9.x on-disk format from 2.6.x (or vice versa), so moving an existing
/var/lib/ldap onto this image is a slapcat/slapadd reload rather than a
restart. There is a TODO to drop the from-source stage once nestgroup ships
in a release; the entrypoint already probes for it and the app keys off
app_ldap__nestedGroupsServerSide, so that swap needs no other changes.
jump-host and proxy pointers are deliberately unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Bump jump-host, ldap-client, proxy, sso-manager-node submodules to their
new tags (cross-app app_super_admin group, jump-host's app_jump_admin,
ldap-client's SSH host access for super admins).
- setup.sh: new env_upsert helper persists SSO_GIT_COMMIT/PROXY_GIT_COMMIT/
JUMP_GIT_COMMIT into ./.env (docker compose's auto-loaded env file) so an
ad-hoc single-service rebuild outside a full setup.sh run still bakes the
right commit hash instead of "unknown".
New optional service (jump-host compose profile) built from the new
theta42/ldap-client submodule: a genuinely SSSD + AuthorizedKeysCommand
LDAP-joined host, not a container with a manually-dropped public key in
authorized_keys. Lets jump-host's actual key-injection -> upstream-connect
flow be exercised end-to-end against the local dev stack's own LDAP.
Verified: real SSH login through jump-host's uid_-_target grammar, live key
injection into LDAP, and ldap-test-host serving that key via
AuthorizedKeysCommand -- both via the SSH CLI and WinSCP.
config.example/ldap-test-host.vars.example documents the one field an
operator fills in (the ldapclient service account's bind password, already
generated by bootstrap.js as serviceAccountPass) -- the rest points at this
stack's own local LDAP by default.
sso-manager's Directory data had no persistent volume at all: @simpleworkjs/orm
falls back to './config/inventory.sqlite' (relative to /app) when conf.orm isn't
set, which lives in the container's ephemeral layer -- every container recreate
silently wiped every site/host/service/oauth-client resource and their
relationships. setup.sh's generated sso-secrets.js now sets
orm.storage=/data/inventory.sqlite, co-located with the already-persisted
sso-data volume (same one Redis uses). Verified live: migrated the running
stack's existing Directory data to /data, force-recreated the container, and
confirmed all 8 resources survived.
Also bumps sso-manager-node to v1.8.0 (resource modal standardization: tabs,
footer with created/updated-by metadata, linkable /directory/:slug URLs, a new
Children tab, and site-slug-prefixed auto-created LDAP groups).
Companion to theta42/jump-host#18: that PR turns on Redis persistence
(AOF + RDB) at /data, but without a volume mount, persistence-within-the-
container is pointless -- docker rm -f (which every real rebuild uses,
per this repo's own documented docker-compose-v1 recreate quirk) throws
the container's whole filesystem away regardless. Adds jump-redis-data:/data,
matching proxy-data's existing role for the same package (lua-resty-auto-ssl
also relies on Redis persistence, via its own volume).
Verified: minted a jump-host API token, force-recreated the container,
confirmed the token still worked afterward.
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
seedDirectory() only ever linked ONE OAuth client -- whatever id was
passed in, which was always the proxy's (resolvedClientId). jump-host's
own OAuth client (minted by provisionJumpHost) was never passed through,
so it was created but never got a ResourceEdge to the "SSH Jump Host"
service resource -- it just showed up in the Directory with no parent.
provisionJumpHost now returns the jump client's id (looking it up even
on the "already configured" early-return path, so existing deployments
self-heal on the next setup.sh run instead of needing this fixed only for
fresh installs), and seedDirectory takes it as a third argument, linking
it under the jump-host service the same way the proxy's client is linked
under the proxy service.
Also fixed live on the affected deployment via the directory-admin API
(created the missing edge directly) rather than waiting for a rebuild.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Regression guard for bootstrap.js's jump-secrets.js template: its ldap
block must use ldaps:// (implicit TLS, :636), not ldap:// (:389), as long
as tlsOptions is set alongside it. ldapts treats a non-empty tlsOptions as
"use implicit TLS" regardless of URL scheme, and jump-host's LDAP client
always sets tlsOptions -- so this exact combination broke every SSH login
to jump-host (any account, any password) before being root-caused
against a real deployment.
Static (parses bootstrap.js as text), not a require()+exec of it --
bootstrap.js is a self-running provisioning script with real side effects
(LDAP writes, live API calls), not a library, so there's nothing safe to
import and call in CI.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
jump-secrets.js's ldap.url was 'ldap://sso-manager:389' with tlsOptions
set. ldapts treats a non-empty tlsOptions as "use implicit TLS" regardless
of URL scheme, so every LDAP connection from jump-host opened a raw TLS
handshake against sso-manager's plaintext-LDAP port — slapd dropped the
connection before any LDAP message parsed (visible in slapd.log as
"connection lost" right after ACCEPT, no BIND ever logged). Every SSH
login failed with a generic "Permission denied" for any account, any
password — indistinguishable from a wrong credential.
Root-caused by building a local theta-env stack, restarting jump-host
with edited config, and calling userLdap.getUser/checkPassword directly
inside the container: got "Client network socket disconnected before
secure TLS connection was established" instead of a vague auth failure.
Switching to ldaps://sso-manager:636 (already exposed by the same
container, already what tlsOptions was meant for) fixes it — verified
getUser/checkPassword succeed and a real SSH login authenticates.
Companion defensive fix: simpleworkjs/ldap#1 (rejects this exact
ldap://+tlsOptions combination going forward, for proxy/sso too).
Existing deployments must edit their own ./config/jump-secrets.js (this
template only affects fresh bootstraps) — see the PR description.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CFG_HTTP_PROXY / CFG_HTTPS_PROXY / CFG_NO_PROXY in setup.env (all
optional, unset by default) get wired into every service's docker build
(npm/apt) and running container (SMTP, ACME/Let's Encrypt, DNS provider
calls, the jump-host directory API client) as HTTP_PROXY/HTTPS_PROXY/
NO_PROXY. Useful for isolated/offline/corporate-network test hosts that
only reach the internet through an upstream proxy — distinct from the
theta42 "proxy" app itself. CFG_NO_PROXY defaults to the stack's own
internal service names so container-to-container traffic never routes
through the proxy.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
wmantly
