add deploy-stack skill; falco: fix hook-tamper condition, add three more post-incident rules

git-hook-tamper.yaml's condition used proc.exepath, which resolves to the
script interpreter's path (e.g. /bin/busybox) for shebang scripts, not the
script's own path - switched to proc.cmdline, which retains the originally
invoked path. Confirmed via live-testing both ways.

ssh-persistence/cloud-metadata-probe/db-spawned-process round out the
post-incident hardening pass with a few more incubating-ruleset adaptations.
This commit is contained in:
2026-08-16 14:33:56 +00:00
parent 9bf72f8e19
commit 2ed0565486
6 changed files with 172 additions and 1 deletions
+79
View File
@@ -0,0 +1,79 @@
---
name: deploy-stack
description: Deploy a docker-infrastructure change - push to Gitea, then redeploy the stack via Portainer. Use whenever a service's docker-compose.yaml (or other tracked file) changed and needs to go live.
---
# Deploy a stack (Gitea → Portainer)
This repo's services go live via GitOps: commit → push to Gitea (self-hosted, SSH
remote) → tell Portainer to pull + redeploy. Always use `./portainer.sh` for the
Portainer side; never raw `curl` unless `portainer.sh` can't do it.
## Steps
1. **Stage and commit** only the files that changed for this task (never `git add -A`).
Imperative mood, first line < 72 chars, explain *why* not *what*.
2. **Sync with remote before pushing.** `git push origin main` fails non-fast-forward
if anyone (or the Gitea web UI) committed since your last pull. Prefer:
```
git pull --no-edit origin main # fetch + merge in one step
git push origin main
```
The remote is SSH (`git@gitea.kolpacksoftware.com:...`). `GITEA_TOKEN` in
`.credentials` is NOT valid for HTTPS push (403) - don't try HTTPS for push.
3. **If push/fetch fails with a low-level git protocol error** (`bad pack header`,
`unable to fork git-pack-objects`, `cannot exec '.../hooks/...'`) - this is NOT
a normal merge conflict, it means Gitea's git service itself is broken
(corrupted repo, poisoned global gitconfig, etc). Do not just retry blindly.
Stop and investigate server-side (`docker logs gitea`, check
`/data/gitea/home/.gitconfig` inside the container) before continuing - this
exact failure mode was an active RCE backdoor once (see
`gitea-log-poisoning-attack-2026-08.md` memory). HTTPS fetch with `GITEA_TOKEN`
can help bisect whether it's SSH-specific or instance-wide:
```
source .credentials
git fetch "https://${GITEA_USER}:${GITEA_TOKEN}@gitea.kolpacksoftware.com/homelab/docker-infrastructure.git" main
```
4. **Find the stack name/ID** if you don't already know it:
```
source .credentials && ./portainer.sh list
```
Stack names in Portainer usually match the service directory name, but not
always (verify with `list`, don't assume).
5. **Redeploy** (pulls latest git commit + recreates containers):
```
source .credentials && ./portainer.sh redeploy <stack-name>
```
This only works for stacks that are git-linked in Portainer. If it's not
git-linked, changes to `.env` values must go through
`./portainer.sh set-env <stack-name> KEY=VALUE` instead - a `.env` file at
the repo path is gitignored and NOT read by git-linked Portainer deploys.
6. **Verify**: `docker ps --filter name=<container>` for status, `docker logs
<container> --tail 50` to confirm it actually came up clean, not just "Up".
## Gotchas
- **`.env` files are gitignored and invisible to git-linked stacks.** Portainer
reads env vars it has stored for the stack (set via UI or `portainer.sh
set-env`), not the local `.env` file. Check current values first with
`./portainer.sh get-env <stack-name>` before adding new ones, and set any new
var explicitly - editing the local `.env` alone does nothing for a deployed
git-linked stack.
- **Not every service is a Portainer stack.** Some (e.g. `falco`) are run via
plain `docker run`/manual `docker compose`, bind-mounting config directly from
this repo checkout. For those, `portainer.sh redeploy` will fail with "stack
not found" - check `docker inspect <container> --format '{{json .Mounts}}'`
to see if it's bind-mounted from this repo (if so, a `git push` alone is
enough to update the *source* files; the container itself needs a manual
`docker restart` or recreate to pick up new volume mounts).
- **No `docker compose` CLI on this host.** `docker compose up -d` fails
outright - always go through Portainer, or `docker run`/`docker restart`
directly for manually-managed services.
- Portainer's `GET /api/stacks/<id>` can return Unauthorized for some stacks -
`portainer.sh` already works around this by listing + filtering; don't call
the raw API directly.
+3
View File
@@ -34,3 +34,6 @@ services:
- ./rules/tune-noise.yaml:/etc/falco/rules.d/tune-noise.yaml:ro
- ./rules/git-hook-tamper.yaml:/etc/falco/rules.d/git-hook-tamper.yaml:ro
- ./rules/unexpected-child-of-git.yaml:/etc/falco/rules.d/unexpected-child-of-git.yaml:ro
- ./rules/ssh-persistence.yaml:/etc/falco/rules.d/ssh-persistence.yaml:ro
- ./rules/cloud-metadata-probe.yaml:/etc/falco/rules.d/cloud-metadata-probe.yaml:ro
- ./rules/db-spawned-process.yaml:/etc/falco/rules.d/db-spawned-process.yaml:ro
+25
View File
@@ -0,0 +1,25 @@
# Custom rule, adapted from falcosecurity/rules' Incubating ruleset (not
# shipped in this Falco image). This host has no cloud provider, so
# 169.254.169.254 (the AWS/GCP/Azure instance-metadata IP) has zero
# legitimate traffic ever - any outbound connection to it is a strong
# signal, either SSRF probing or a container image/script that assumes a
# cloud environment. Near-zero false-positive risk, so this stays a plain
# CRITICAL with no scoped exception - if something legitimate ever needs
# it, add a proc/container-scoped exception here rather than disabling.
- rule: Contact cloud metadata service from container
desc: >
Detects attempts to communicate with a cloud instance metadata service
(169.254.169.254) from a container. This host has no cloud provider,
so this endpoint should never see legitimate traffic - treat any hit
as SSRF probing or malware/scripts written for a cloud environment.
condition: >
outbound
and container
and fd.sip="169.254.169.254"
output: >
Outbound connection to cloud instance metadata service
(user=%user.name command=%proc.cmdline connection=%fd.name
container=%container.name image=%container.image.repository pid=%proc.pid)
priority: CRITICAL
tags: [network, credential_access, mitre_credential_access]
+30
View File
@@ -0,0 +1,30 @@
# Custom rule, adapted from falcosecurity/rules' Incubating ruleset (not
# shipped in this Falco image). Generalizes the lesson from the gitea
# log-poisoning RCE (a service's own process forking something unexpected)
# to every other DB-backed service on this host - immich, firefly-iii,
# romm, inbox-zero-db, etc. A database server forking a child process other
# than itself is not normal and often follows a SQL injection attack.
- list: db_server_binaries
items: [mysqld, postgres, sqlplus]
- macro: user_known_db_spawned_processes
condition: (never_true)
- rule: DB program spawned process
desc: >
A program related to a database server created an unexpected child
process (other than itself). This is not supposed to happen and often
follows SQL injection attacks - could indicate unauthorized data
extraction or tampering.
condition: >
spawned_process
and proc.pname in (db_server_binaries)
and not proc.name in (db_server_binaries)
and not user_known_db_spawned_processes
output: >
Database-related program spawned unexpected process
(user=%user.name command=%proc.cmdline parent=%proc.pname
container=%container.name image=%container.image.repository pid=%proc.pid)
priority: WARNING
tags: [database, process, mitre_execution]
+1 -1
View File
@@ -19,7 +19,7 @@
condition: >
spawned_process
and container.name = "gitea"
and proc.exepath contains "/hooks/"
and proc.cmdline contains "/hooks/"
and not proc.name in (gitea_managed_hook_names)
output: >
Unexpected git hook executed
+34
View File
@@ -0,0 +1,34 @@
# Custom rule, adapted from falcosecurity/rules' Incubating ruleset (not
# shipped in this Falco image - only the Stable ruleset ships by default).
# Added after research into what else might be useful post-incident: neither
# the gitea/xmrig nor gitea log-poisoning incidents involved SSH-key
# persistence, but it's the textbook next move after any RCE, and gitea
# itself has SSH access (port 222) making this directly relevant here.
- list: ssh_binaries
items: [
sshd, sftp-server, ssh-agent,
ssh, scp, sftp,
ssh-keygen, ssh-keysign, ssh-keyscan, ssh-add
]
- macro: user_ssh_directory
condition: (fd.name contains '/.ssh/' and fd.name glob '/home/*/.ssh/*')
- rule: Adding ssh keys to authorized_keys
desc: >
After gaining access, attackers can modify the authorized_keys file to
maintain persistence on a victim host. Detects any write to an
authorized_keys file under a user's .ssh directory or /root/.ssh,
by a process that isn't one of ssh's own binaries.
condition: >
open_write
and (user_ssh_directory or fd.name startswith /root/.ssh)
and fd.name endswith authorized_keys
and not proc.name in (ssh_binaries)
output: >
Adding ssh keys to authorized_keys
(user=%user.name file=%fd.name command=%proc.cmdline
container=%container.name image=%container.image.repository pid=%proc.pid)
priority: WARNING
tags: [ssh, persistence, mitre_persistence]