Stop calling /zen before each installation token request. A short
network problem to GitHub failed deployments with a cURL timeout on
/zen. assertGithubClockInSync now takes the failed response and reads
its Date header, so the "System time is out of sync" error still shows
when GitHub rejects the JWT.
Call it on failed JWT requests in the webhook installation check, the
permission job, the source connection test, and the server transfer
checklist. Remove the /zen fakes from the tests.
The install callback swallowed every verification error (clock skew, network,
missing key, GitHub API errors) and returned an opaque 403. It also compared
the state's team with the session team, so switching teams during the GitHub
install returned 403.
- Return the verification reason, log it, and redirect to the GitHub App page
with an error toast (JSON requests still get 403 with the reason)
- Check membership of the state's team, and switch the session to it
- Show flashed errors on the GitHub App page
Fixes#12122
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Require deploy or root token access when creating managed mounts outside
resource directories or changing their file contents through the API.
Synchronize KeyDB and Dragonfly passwords with local and shared
REDIS_PASSWORD values while preserving secret manager resolution.
Allow validated host file and directory mounts outside resource directories
while protecting external paths from deletion.
Clean up cancelled deployment containers and clear tracked process IDs.
Preserve legacy KeyDB and Dragonfly password parsing, accept server backup
paths without extension restrictions, and allow Compose variable proxy ports.
- Manual webhooks (GitHub, GitLab, Gitea, Bitbucket) now reject all
deliveries for a locked-out scope before reading secrets or verifying
signatures, including correctly signed ones, until the window expires.
- ProvisionGithubRunnerJob checks whether the execution has already
finished before and after each provisioning step, and only sets
ready_at or Failed while it is still occupying/provisioning. A runner
that is cancelled, timed out or completed mid-provisioning is no longer
started, retried or overwritten.
- VolumeCloneJob creates bind-volume clones as plain local volumes
without the source driver options, so the clone never mounts the
source host directory.
- Image pulls for additional servers no longer ignore docker pull errors.
- Add tests for each change.
- Do not provision a runner for a queued pull request job when no matching
runner configuration allows pull request jobs (the run event is read from
GitHub; the job-started hook still refuses it if the lookup fails).
- Do not open an existing runner group to all or public repositories, so
restrictions an organization admin set on GitHub stay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The failed-authentication lockout for manual webhooks was checked before
the signature, and the bucket used the client IP from X-Forwarded-For.
An unauthenticated client could send failed attempts with the Git host's
address for a public repository and branch, and correctly signed
deliveries then got 429, so deployments were lost.
The lockout now applies only to deliveries that fail authentication; a
delivery with a valid secret for a matched application is always
processed. Failures are counted per connection address (REMOTE_ADDR, or
CF-Connecting-IP on cloud), so rotating X-Forwarded-For no longer gives
a new bucket.
GitHub, Gitea and Bitbucket manual webhooks accepted any signature when
APP_ENV=local. GitLab and the GitHub App webhook already did not.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Configuration diff: report all settings when an app has no earlier
snapshot, so older apps show "configuration changed" again (#11430).
- DigitalOcean/Vultr: do not repeat the "API error:" prefix.
- HasSecretManager: the deleting listener returns nothing, so later
deleting listeners still run.
- GitLab webhook: use new_public_id() like the other webhooks.
- Team danger zone: show the link and open the page only for the team owner.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Revert the app code changes of d666a0a95, 69eeb2f62, 1e66c9f9b and
a32082b03, so login, registration, OAuth, team danger zone, service
templates, and the other touched files are back to their earlier state.
Keep the removal of unused components and markup-only tests (d666a0a95
also contained those deletions by mistake).
StartupExecutionCleanupTest now fakes the template CDN with the current
file content, so app:init in the test no longer empties
templates/service-templates-latest.json.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- TrustHosts: validate the Host header also when the FQDN cache is cold.
- Settings email: authorize the SMTP and Resend toggles.
- Server navbar: show Sentinel status only to users who can view Sentinel.
- GitHub source: restore permission gates on Sync name and Delete, and the
Test connection button.
- Profile: make avatar upload work (missing import and file uploads trait).
- API tokens: keep the "Never" expiry choice.
- Environment variables: show validation errors instead of a generic error.
- Servers API: read the server uuid from the route in domains_by_server.
- Configuration diff: report all settings when no earlier snapshot exists.
- Cloud servers: restore Power On for stopped servers; do not repeat the
provider error prefix.
- HasSecretManager: do not stop later deleting listeners.
- GitLab webhook: use new_public_id() like the other webhooks.
- Dev: show the Lima VM start command on the localhost server page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Gitea handler read the pull request commit from a top-level
head.sha that Gitea does not send, so preview deployments always used
HEAD. It now reads pull_request.head.sha, with the top-level after of
newer Gitea versions as a fallback.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every webhook handler (GitHub, GitLab, Gitea, Bitbucket, GitHub App,
GitLab App) now reads the fields it uses through typed readers: refs,
titles, and actions must be strings, ids positive integers, commit
SHAs 7-64 hex characters, and URLs http(s). An invalid value returns
"Nothing to do. Invalid '<field>' in the request." without a
deployment. Signature checks and the failure lockout are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- GitLab, GitHub, and Gitea crashed with a 500 on push payloads without
a commit list, and other malformed payloads (missing repository,
project, ref, Bitbucket changes, non-string values, unsupported
Gitea events) also returned 500. They now get a clean response; a
push with an unknown file list still deploys, and a branch deletion
does not deploy.
- The manual webhook lockout counts only distinct failed attempts: the
same wrong GitLab token, or an identical HMAC redelivery, counts
once, so a misconfigured hook no longer locks out a valid one. Every
new guess still counts (30 per scope and minute). Attempts are
stored only as HMAC hashes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Failed manual webhook deliveries were counted per provider and source
IP. On Coolify Cloud many customers share the Git provider IP, so one
repository with a wrong secret locked out valid deliveries of all
other repositories for 60 seconds.
The failure key now also contains the repository and branch. Guessing
the secret of one application stays limited to 30 tries per minute,
and a lockout rejects all deliveries in that scope, also correct ones.
A GitLab request without a token no longer counts as a failure.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Proxy: list and delete Traefik ACME certificates from the server proxy
page via new TraefikAcmeService and Get/DeleteTraefikCertificate actions
- DNS: track ownership and cross-resource references for managed DNS
records so records are only deleted when no longer referenced; release
records asynchronously on resource deletion via ReleaseManagedDnsRecordsJob
and ManagedDnsRecordCleanup; harden Cloudflare provider deletion results
- Databases: fail closed on start when prerequisites or the CA certificate
are missing (DatabaseStartException, Server::ensureCaCertificate) and
clean up stale start activities via ResourceStartActivity
- Webhooks: throttle repeated manual webhook signature failures for
GitHub, GitLab, Gitea and Bitbucket
- Deployments: improve compose build-context handling and compose file
load error reporting
- Install scripts: rework terminal UI output in install.sh (stable and
nightly)
- Misc: settings sidebar accordion fixes, log drain toggle rollback,
add Serverside to README sponsors
- Add migrations and tests covering the above
Centralize scp-style Git URL parsing so user@host:path (including custom
usernames and embedded ports) is accepted and converted to HTTPS for
public clones, API create, webhooks, validation, and commit/branch links.
Webhook secret was stored and shown as plaintext. Use a password field,
encrypt at rest (with legacy plaintext read support), and look up tokens
via findByWebhookToken so encrypted values still authenticate webhooks.
GitlabAppPolicy previously allowed any authenticated user to update,
delete, and create GitLab sources. Align it with GithubAppPolicy, require
Application create authorization on the private-repo wizard, and reject
OAuth callbacks from non-admins so members cannot escalate privileges.
Adds self-hosted GitLab OAuth sources so Coolify can connect to a self-managed GitLab instance, list private repositories, clone over an OAuth token, and deploy (the GitLab counterpart to GitHub Apps).
Hardening: authenticated, one-time team-bound OAuth callback state; token redaction in deploy logs; custom host port/path kept in clone and ls-remote URLs; submodule OAuth auth; system-wide source selection. Covered by unit and feature tests.
cosigned by OpenAI Codex at M1 Max
Strip leading port segments from scp-style GitLab repository URLs so manual webhook matching compares the repository path consistently. Cover both ported and unported SSH URL forms.
- resolve the GitHub App by a stable identifier during installation
callbacks so installing and re-installing keeps working over the
full lifetime of the App
- verify the installation id received from the callback against the
GitHub API before persisting it
- support re-installing an already configured GitHub App instead of
blocking it
- require an authenticated session and rate limit the setup callback
routes
- extend manifest setup state validity to match GitHub's manifest
code lifetime
Adds feature coverage for the GitHub App setup and installation
callbacks.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Git hosts treat owner/repo names case-insensitively, but the exact
repository match used a case-sensitive comparison, so a payload whose
casing differed from the stored git remote would fail to match and
skip a legitimate deployment.
Lowercase both canonical repository paths before comparing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The manual webhook handlers selected target applications with a
`git_repository LIKE %full_name%` substring query, so a payload
repository name could match unintended applications when repository
names overlap.
Add a `MatchesManualWebhookApplications` trait that validates the
incoming `owner/repo` value and matches `Application.git_repository`
by exact normalized path. Github, Gitlab, Gitea and Bitbucket manual
handlers now use it, reject invalid repository input early, and return
a consistent generic webhook failure payload.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add DetectsSkipDeployCommits trait with two strategies: shouldSkipDeploy
(all commits must contain the marker) for push events, and
shouldSkipDeployAny (any single marker triggers skip) for PR/MR titles
and latest-commit signals.
Apply trait to Bitbucket, Gitea, GitHub, GitLab webhook controllers and
ProcessGithubPullRequestWebhook job. PRs pass pullRequestTitle through
to the job constructor for evaluation.
Florian Pfitzer
Andrew Matia
Mike Chong
ShadowArcanist