Drupal security release¶
Now: PSA-2026-11-16, core release on 2026-11-18, 17:00-21:00 UTC (18:00-22:00 CET)
PSA-2026-11-16, published 2026-11-16: a Drupal core security release for all supported branches.
- Risk: Critical, 18/25 (AC:Complex/A:None/CI:All/II:All/E:Proof/TD:Uncommon). At least one vulnerability can be exploited by anonymous users. More than one configuration is affected: do not assume a site is safe.
- Affected: 11.4.x, 11.3.x, 10.6.x, also 12.0.0-beta3 and 11.5.0-beta1. Everything before 10.6.x, and Drupal 11 before 11.3.x, is end-of-life: those sites only get the interim mitigation steps of the advisory.
- Exploits might be developed within hours or days.
- Drupal Steward mitigates the known attack vectors for sites on Steward.
- Recommendation: update every site to the latest patch release of its branch before 2026-11-18.
- No details before the release; ignore unsolicited emails about it.
- Announcements: drupal.org/security
(RSS), Drupal Slack
#security-team, the security team on Mastodon, Bluesky, X and LinkedIn, and its email list.
Principle¶
The fix is the core update, deployed by each Drupal project's own pipeline.
The toolkit does not deploy Drupal sites (gitlab-ci-cd/drupal with
docker4drupal, see
Contract with docker4drupal).
The emergency measures bridge the time until a site is patched, from least
to most disruptive: waf update → block or waf rule → shutdown.
Before the window¶
Today and tomorrow, before 17:00 UTC:
-
Patch releases. Every Drupal project on the latest patch release of its branch. Flag end-of-life sites (before 10.6, and 11.0 to 11.2): they need the advisory's interim mitigations or a
shutdown. The core version of every running site in one go, read-only (it runs the command for real, in every php container of docker4drupal; needs toolkit 0.38.0):1lda-fleet -- exec --reason "Drupal core versions" --image '*drupal-php*' --cmd "drush status --field=drupal-version" -
Hosts ready, read-only (one pipeline per client;
statusnever changes anything). Expect no active measure and no problem lines:1lda-fleet -- status -
Preview the Drupal shutdown on every host; it writes nothing:
1lda-fleet -- shutdown --id drupal-sa-off --reason "Drupal core security release" --image '*drupal-php*' --previewdocker4drupal stacks run the image
wodby/drupal-php:<version>(optionally behind a registry prefix) in the compose projectsdrupal_<project>_<ref>. The glob*drupal-php*selects exactly those projects with all their routers (web server, adminer, mailpit, …); the broader*php*would also take other PHP apps. Compare with--stack 'drupal_*'. Check the routers in each host's job log and adjust the selection now, not on the evening. The same for a drushexec, which lists the containers and runs nothing:1lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' --cmd "drush -y pm:uninstall MODULE" --preview -
Operator workstation: personal access token (scope
api, Developer or higher in the client projects), Docker, the imageregistry.lakedrops.com/ansible/playbooks/toolkit/ee:0.38.0,tmux(setup). This proves the token and lists the clients:1lda-fleet --dry-run -- status -
Notice text agreed: a title, a one-paragraph message, optionally an https link. The page adds its fixed texts in English and German itself.
- On duty 17:00-21:00 UTC, and who may decide a shutdown.
- Hosts without the tool (client
tineon: thxr1, thdbm1, thdbo1, thwv1): if they serve Drupal, plan the manual path (Hosts without the tool).
In the window: decide¶
When the advisory (SA-CORE-…) is out, read it: the affected versions and configurations, the attack vector if it is described, the interim mitigations. Then:
flowchart TD
A["Advisory SA-CORE-… is out"] --> B{"Patch deployable within the hour,<br>no exploitation seen?"}
B -- yes --> C["Deploy the core update<br>+ waf-update as a cheap extra"]
B -- no --> D{"What does the advisory give you?"}
D -- "pattern in path, query,<br>method or headers" --> E["block → patch → lift"]
D -- "pattern only in the body" --> F["waf rule → patch → lift"]
D -- "hub ships vpatch-CVE-…" --> G["waf-update"]
D -- "Drupal setting or module change" --> H{"Module attacked<br>through its URLs?"}
H -- yes --> J["block → lasting change in the project → lift"]
H -- no --> K["exec drush on all sites now<br>+ lasting change in each project"]
D -- "nothing, exploitation likely" --> I["shutdown → patch or upgrade → lift"]
| Situation | Action |
|---|---|
| Patch deployable within the hour, no exploitation seen | Deploy the core update by each project's pipeline; lda-fleet -- waf-update as a cheap extra. Nothing to lift. |
| The advisory or mitigation names a request pattern in the path, query, method or headers | block on all clients with an id, then patch, then lift |
| The pattern is only in the body (a POST of a form or JSON that real clients also use) | waf rule (AppSec, on by default on every Traefik host), then patch, then lift |
The CrowdSec hub ships crowdsecurity/vpatch-CVE-… for it |
lda-fleet -- waf-update; look for new: crowdsecurity/vpatch-CVE-… in the output |
| The interim mitigation is a Drupal setting or module change | Module attacked through its URLs: prefer block. Otherwise exec with drush on all sites now, and the lasting change in each project (below) |
| No mitigation, exploitation likely, the site cannot be patched now (e.g. end-of-life) | shutdown the affected stacks with the notice, patch or upgrade, lift |
Ready-to-adapt commands for all clients (the host and client forms are on Measures):
1 2 3 4 5 6 7 8 9 10 11 12 13 | |
Run each one first with --preview, then again without it. lda-fleet
lists the clients and the ASCR_COMMAND it will run, and asks for
confirmation.
Mitigation inside Drupal: exec¶
Run the advisory's mitigation with drush in every docker4drupal php container, from the least to the most destructive:
-
A permission or a setting:
1 2 3 4 5 6 7 8 9
lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' \ --cmd "drush role:perm:remove anonymous 'PERMISSION'" --preview lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' \ --cmd "drush role:perm:remove anonymous 'PERMISSION'" lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' \ --cmd "drush -y cset CONFIG KEY VALUE" --preview lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' \ --cmd "drush -y cset CONFIG KEY VALUE" -
Uninstall the module, last: in Drupal 8 and later, "disable" means uninstall, which deletes the module's configuration and data, and Drupal refuses it while content uses the module.
1 2 3 4
lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' \ --cmd "drush -y pm:uninstall MODULE" --preview lda-fleet -- exec --reason "SA-CORE-2026-0XX mitigation" --image '*drupal-php*' \ --cmd "drush -y pm:uninstall MODULE"
The next deployment undoes it
The project pipeline's deployment (gitlab-ci-cd/drupal,
test-and-deploy.yml) runs drush config-import: it installs the
module again from core.extension.yml and restores the configuration
of the project. Make the lasting change in each project too, and deploy
it.
A site where the command fails (e.g. the module is not installed) reports
failed NAME: exit N; the others go on. Each client's job then fails, so
read the per-container result lines.
waf rule on all clients¶
The client's ascr job resolves --file in its checkout, so the rule file
must be in each client repository:
- Write the rule (template) and push it as
rules/drupal-sa-waf.yamltomainof the clients with Traefik hosts:v2/lakedrops,v2/local,v2/mobimo(v2/speicherhas none). The push builds the client image as usual; theascrjob uses the checkout. -
Preview, then run again without
--preview:1 2
lda-fleet --only v2/lakedrops,v2/local,v2/mobimo -- waf-rule --id drupal-sa-waf \ --reason "SA-CORE-2026-0XX body" --file rules/drupal-sa-waf.yaml --preview
While measures are active¶
A patched site stays behind the notice
A shutdown by --image or --stack is resolved again every 30
seconds: a site redeployed with the patch still matches and stays
behind the notice until you change the measure.
- Release patched sites one by one: run the same
shutdown --id drupal-sa-offagain with a narrower selection (the same type and id replaces it), e.g.--stack 'drupal_a_main,drupal_b_main'for the stacks that are still unpatched. Orliftone host:ascr emergency lift --id drupal-sa-off --limit HOSTin the client, orsudo lakedrops-emergency lift --id drupal-sa-offon the host. - Check a patched site while it is shut down:
--bypass CIDRlets the office's TCP peers through. Not behind a CDN: Traefik'sClientIPsees the CDN's address. -
Watch:
statusper host (Check: status), the journaljournalctl -t lakedrops-emergency(also shipped to OpenObserve), and the CrowdSec alerts on the host:1docker compose --project-directory /opt/traefik exec crowdsec cscli alerts list
After¶
- All sites patched: check the core version of every site, e.g. with the
read-only
execof Before the window. -
Lift every id you used. A
lifton a host without the id succeeds and changes nothing:1 2 3
lda-fleet -- lift --id drupal-sa-off lda-fleet -- lift --id drupal-sa-block lda-fleet -- lift --id drupal-sa-waf -
lda-fleet -- status: no measure left. The sanity issues close with the nextascr sanity check.statusdoes not show anexec: a mitigation byexecstays until you run its inverse or the project's next deployment imports its configuration. - Write a short incident note: what was used, the timings, what was missing.