From 64f06bbbbc785fb88588ed27d632f329c3b31fa0 Mon Sep 17 00:00:00 2001 From: bchanot Date: Mon, 12 Oct 2026 01:14:40 +0200 Subject: [PATCH 1/3] fix(install): sshd hardening no longer aborts when ssh.service is inactive harden_sshd ended with an unconditional 'systemctl reload ssh'. On WSL (and any host where sshd was never started) the unit is inactive, systemctl exits with 'ssh.service is not active, cannot reload.' and set -e killed the whole install right after the drop-in was validated. - reload only when 'systemctl is-active --quiet ssh'; else say the drop-in is in place and applies at the next sshd start - try-reload-or-restart instead of reload: exits 0 if sshd stopped in between - create /run/sshd before 'sshd -t': on Ubuntu 24.04 the privsep dir only exists while ssh.service runs, so the validation failed on an inactive host and the valid drop-in was removed as 'rejected' --- install.sh | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/install.sh b/install.sh index 3e8e473..ac23fc2 100755 --- a/install.sh +++ b/install.sh @@ -220,19 +220,25 @@ install_unattended_upgrades() { # sshd hardening drop-in (PermitRootLogin, MaxAuthTries, LoginGraceTime): only # settings that cannot lock anyone out; authentication methods stay untouched. -# Validated with sshd -t before the reload. A rejected file is removed rather than -# left in place, so sshd keeps starting on the next boot; the install goes on and -# the warning tells you. +# Validated with sshd -t (/run/sshd is created first so the check works without a +# running sshd); the reload happens only when sshd is running. A rejected file is +# removed rather than left in place, so sshd keeps starting on the next boot; the +# install goes on and the warning tells you. harden_sshd() { echo "Deploying sshd hardening drop-in" sudo install -D -m 0644 "$SCRIPT_DIR/etc/ssh/sshd_config.d/20-hardening.conf" \ /etc/ssh/sshd_config.d/20-hardening.conf + sudo install -d -m 0755 /run/sshd if ! sudo sshd -t; then sudo rm -f /etc/ssh/sshd_config.d/20-hardening.conf echo "sshd rejected 20-hardening.conf — removed, sshd config unchanged" >&2 return 0 fi - sudo systemctl reload ssh + if systemctl is-active --quiet ssh; then + sudo systemctl try-reload-or-restart ssh + else + echo "ssh.service not running — drop-in in place, applies at the next sshd start" + fi } # Yes/no prompt for the optional system changes. $2 is the default taken on Enter: From 9d8347f58fd8faa8315d0dc55a95360bcadadc6d Mon Sep 17 00:00:00 2001 From: bchanot Date: Mon, 12 Oct 2026 21:23:59 +0200 Subject: [PATCH 2/3] =?UTF-8?q?docs:=20README=20step=2013=20sshd=20reload?= =?UTF-8?q?=20guard=20+=20CHANGELOG=20Fixed=20=E2=80=94=20hotfix=20sshd-re?= =?UTF-8?q?load-inactive?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- CHANGELOG.md | 1 + README.md | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index fd8c57f..b284f51 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -21,6 +21,7 @@ Format: [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). Versioning: se ### Fixed - `~/.gitconfig` was deployed with the literal `@USER@` / `@EMAIL@` placeholders: the installer read the identity from the repo bashrc template instead of the answers. - An `&` in the name no longer corrupts the rendered identity on bash 5.2 and later (`patsub_replacement` turned off). +- `install.sh`: the sshd hardening step no longer aborts the install when `ssh.service` is inactive (WSL, sshd never started): `/run/sshd` is created before `sshd -t`, so a valid drop-in is no longer removed as rejected, and the reload (`try-reload-or-restart`) runs only when sshd is active. ## [1.0.0] — 2026-10-06 diff --git a/README.md b/README.md index 0574349..861c32f 100644 --- a/README.md +++ b/README.md @@ -75,7 +75,7 @@ What it does: 10. On Linux, installs **code-server** (VS Code in the browser) via its vendor script — skipped if already present — and enables the `code-server@` systemd service (login from `id -un`: the bashrc overrides `$USER`). 11. On Linux, **offers** (`[Y/n]`, Enter = yes) to install **`ubuntu-desktop-minimal`** (GDM + GNOME Shell, ~1.5 GB): the RDP remote login below hands out a GNOME session, which a bare server install does not have. If accepted, then **offers** (`[Y/n]`) **RDP remote login** via `gnome-remote-desktop` (Wayland-native); no desktop means no RDP question. On **WSL** (`WSL_DISTRO_NAME` set or `microsoft` in `/proc/version`) both default to **no** (`[y/N]`): Windows owns the display, and the desktop drags in NetworkManager + systemd-resolved, which rewrite `/etc/resolv.conf` and break DNS for the rest of the install. With no terminal attached the default applies. The RDP setup installs the daemon + `openssl`, generates a self-signed TLS cert once, and prompts interactively for shared "gate" credentials (skipped when no terminal is attached, or already set). Disables `xrdp` if present; opens UFW port `3389` only when UFW is already active. Finally, when `lspci` sees an NVIDIA GPU, runs `ubuntu-drivers install` to put on the driver the distro recommends for the card (no version pinned; loads at the next reboot). Skipped on machines without an NVIDIA GPU. 12. On Linux, at the very end, **offers** (`[y/N]`) to install the **`cloudpex`** NAS mount helper to `/usr/local/bin` via `cloudpex/install.sh`, which prompts for the NAS host, share name, SMB user, mount point and SMB version and writes them to `/etc/cloudpex.conf` (root, `0600`; an existing config is shown and kept unless you say `n`; skipped when no terminal is attached). Nothing is mounted, no password stored, see [`cloudpex/README.md`](cloudpex/README.md). -13. On Linux, installs the **security baseline**, always, no prompt: **fail2ban** (+ `nftables`) with `etc/fail2ban/jail.d/local.conf` (sshd jail reading the journal, bans the offending IP on every port so the SSH port does not matter, 5 failures in 10 min → 1 h ban, loopback and private LAN ranges never banned); **unattended-upgrades** enabled through `etc/apt/apt.conf.d/20auto-upgrades`; and the **sshd drop-in** `etc/ssh/sshd_config.d/20-hardening.conf` (`PermitRootLogin no`, `MaxAuthTries 3`, `LoginGraceTime 20`), checked with `sshd -t` and removed again if sshd rejects it, then `reload ssh`. Authentication methods, port and user lists are left as they are. +13. On Linux, installs the **security baseline**, always, no prompt: **fail2ban** (+ `nftables`) with `etc/fail2ban/jail.d/local.conf` (sshd jail reading the journal, bans the offending IP on every port so the SSH port does not matter, 5 failures in 10 min → 1 h ban, loopback and private LAN ranges never banned); **unattended-upgrades** enabled through `etc/apt/apt.conf.d/20auto-upgrades`; and the **sshd drop-in** `etc/ssh/sshd_config.d/20-hardening.conf` (`PermitRootLogin no`, `MaxAuthTries 3`, `LoginGraceTime 20`), checked with `sshd -t` (`/run/sshd` is created first, so the check also works when sshd is not running) and removed again if sshd rejects it, then reloaded with `systemctl try-reload-or-restart ssh` when `ssh.service` is active; otherwise (WSL, sshd never started) the drop-in stays in place and applies at the next sshd start. Authentication methods, port and user lists are left as they are. 14. On Linux, at the very end, **offers** (`[y/N]`, skipped when no terminal is attached) to move **`/tmp` to disk**: Ubuntu mounts `/tmp` as a RAM-backed tmpfs capped at 50% of RAM, which agent runs fill, halving the RAM and breaking every shell with "No space left on device". Accepting masks `tmp.mount` and installs `etc/tmpfiles.d/tmp.conf` (wipe at boot, 10-day purge). Effective at the next reboot. 15. On Linux, at the very end, **offers** to keep **SSH reachable under memory pressure**: installs the `ssh.service` drop-in (`OOMScoreAdjust=-1000`, `MemoryMin=256M`) and `earlyoom` with `etc/default/earlyoom` (kills the largest process, `node`/`java` first and never `sshd`, once free RAM and swap both drop under 10%). Restarting `ssh` keeps open sessions. Note: `MemoryMin` protects the sshd daemon only; login sessions live in `user.slice`, so no setting can reserve RAM for a future shell. earlyoom acting in time is the real protection. 16. On Linux, at the very end, **offers** to set up **rootless Docker**: installs `uidmap dbus-user-session slirp4netns docker-ce-rootless-extras`, runs `dockerd-rootless-setuptool.sh install` as you (the apt package puts it in `/usr/bin`, so no `PATH` line is needed, unlike the `~/bin` install from `get.docker.com/rootless`), then `loginctl enable-linger` so the user daemon starts at boot. The tool creates `~/.config/systemd/user/docker.service` and switches the CLI to the `rootless` context; the deployed bashrc exports `DOCKER_HOST=unix://$XDG_RUNTIME_DIR/docker.sock` whenever that socket exists, for tools that ignore contexts. The rootful daemon is disabled first (`systemctl disable --now docker.service docker.socket`, socket removed), as the docs recommend: `docker` then only talks to your user daemon. Needs `/etc/subuid` and `/etc/subgid` entries for the user (Ubuntu adds them at user creation). From c9d47ce91d5b2bd0416d5bec8b14e4d57fd63f62 Mon Sep 17 00:00:00 2001 From: bchanot Date: Mon, 12 Oct 2026 21:24:09 +0200 Subject: [PATCH 3/3] =?UTF-8?q?chore(memory):=20LRN-018=20+=20journal=20?= =?UTF-8?q?=E2=80=94=20hotfix=20sshd-reload-inactive?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- .claude/memory/journal.md | 3 + .claude/memory/learnings.md | 12 ++++ .../2026-10-12-sshd-reload-inactive-0103.md | 19 +++++++ .../2026-10-12-sshd-reload-inactive-0103.md | 55 +++++++++++++++++++ 4 files changed, 89 insertions(+) create mode 100644 .claude/tasks/contracts/2026-10-12-sshd-reload-inactive-0103.md create mode 100644 .claude/tasks/plans/2026-10-12-sshd-reload-inactive-0103.md diff --git a/.claude/memory/journal.md b/.claude/memory/journal.md index f136dcb..1354800 100644 --- a/.claude/memory/journal.md +++ b/.claude/memory/journal.md @@ -144,3 +144,6 @@ Done: `bind -n C-M-Up/Down/Left/Right select-pane` (root + copy-mode-vi, no vim ## 2026-10-11 — desktop + RDP offered, WSL default no Done: `offer_desktop()` (desktop [Y/n] → RDP [Y/n], WSL → [y/N] + hint), `confirm` default arg, `is_wsl()`. Trigger: WSL install lost DNS mid-run after ubuntu-desktop-minimal (hypothesis: NetworkManager/systemd-resolved rewrote /etc/resolv.conf), fail2ban/whois fetch failed. 8-case stub harness OK, shellcheck clean. Branch feature/desktop-rdp-prompt, commit 5d60269, pushed, not merged. + +## 2026-10-12 — WSL DNS confirmed + sshd hardening abort fixed +Done: WSL DNS root cause confirmed (ubuntu-desktop-minimal → systemd-resolved/NetworkManager rewrote resolv.conf); repair runbook (disable both, rm resolv.conf, `wsl --shutdown`) worked. Then install aborted at harden_sshd: `systemctl reload ssh` on inactive unit + `set -e`. Fix 64f06bb on bugfix/sshd-reload-inactive: `is-active` guard + `try-reload-or-restart`, `/run/sshd` created before `sshd -t` (Ubuntu 24.04: privsep dir exists only while ssh.service runs — challenge correctness MAJOR). Stub smoke both branches, shellcheck, semgrep PASS. Not merged. diff --git a/.claude/memory/learnings.md b/.claude/memory/learnings.md index 462fe10..db9a8cf 100644 --- a/.claude/memory/learnings.md +++ b/.claude/memory/learnings.md @@ -135,3 +135,15 @@ triggering pane as target: `set -pu @opt` needs no `-t`; `#{hook_pane}` quoting testable headless on a scratch pane (`new-window -d -P -F '#{pane_id}'`, assert `#{pane_in_mode}`); address by pane id, window-name plugin renames `-n` names. (5) `\;` inside tmux.conf '...' stays literal, then splits a bind's command list when the string is parsed by if-shell: works, no need to convert to `{ }` blocks. + +## LRN-018 — systemctl under `set -e`: guard reload, create /run/sshd before `sshd -t` +2026-10-12. Found fixing install.sh `harden_sshd` on WSL (commit 64f06bb). (1) Bare `systemctl reload ` +exits non-zero when unit inactive (" is not active, cannot reload.") → `set -e` kills installer. Guard: +`if systemctl is-active --quiet ; then sudo systemctl try-reload-or-restart ; else echo ...; fi`. +`try-reload-or-restart` exits 0 if unit stopped between check and call (no TOCTOU abort); `is-active` needs no +sudo, inside `if` never trips `set -e`. `restart` on inactive unit starts it, different semantic. (2) Ubuntu 24.04 +(systemd 255): `/run/sshd` created only by ssh.service `RuntimeDirectory=sshd`, no tmpfiles rule (systemd ≥ 256 +adds `20-systemd-ssh-generator.conf`). sshd never started → `sudo sshd -t` fails "Missing privilege separation +directory" → looks like config rejection. `sudo install -d -m 0755 /run/sshd` before `sshd -t`, idempotent. +(3) WSL: no sshd running by default, unit socket-activated or never started → any installer "validate + reload" +path must survive inactive unit. Stub `systemctl`/`sudo`/`sshd` on PATH in scratchpad to smoke both branches. diff --git a/.claude/tasks/contracts/2026-10-12-sshd-reload-inactive-0103.md b/.claude/tasks/contracts/2026-10-12-sshd-reload-inactive-0103.md new file mode 100644 index 0000000..1a17b94 --- /dev/null +++ b/.claude/tasks/contracts/2026-10-12-sshd-reload-inactive-0103.md @@ -0,0 +1,19 @@ +# CONTRACT — sshd-reload-inactive +- date: 2026-10-12 | flow: hotfix | branch: bugfix/sshd-reload-inactive +- status: active + +## REQUEST (verbatim — IMMUTABLE) +install.sh harden_sshd: `sudo systemctl reload ssh` fails under set -e when ssh.service is not active (WSL, socket-activated or never started) → "ssh.service is not active, cannot reload." aborts the whole install. Fix: reload only if `systemctl is-active --quiet ssh`, otherwise print that the drop-in is in place and applies at the next sshd start. Keep shellcheck clean, tabs. + +User message: "ok ca a fonctionne, par contre maintenant je bloque a Deploying sshd hardening drop-in +ssh.service is not active, cannot reload.. je sais pas si c'es tpour installer ssh-server ? Si oui demande aussi pour l'installer, si non il bloque pour quoi ?" + +## CLARIFICATIONS +none — request complete (openssh-server is already in the apt list; the block is the unconditional reload) + +## ACCEPTANCE CRITERIA +1. Symptom gone: harden_sshd no longer aborts the install when ssh.service is inactive (reload guarded by `systemctl is-active --quiet ssh`, else an informative echo). +2. Build/tests green: `bash -n install.sh` and `shellcheck install.sh` pass; indentation stays tabs. + +## FILE SCOPE +install.sh (harden_sshd only) diff --git a/.claude/tasks/plans/2026-10-12-sshd-reload-inactive-0103.md b/.claude/tasks/plans/2026-10-12-sshd-reload-inactive-0103.md new file mode 100644 index 0000000..3aaeaec --- /dev/null +++ b/.claude/tasks/plans/2026-10-12-sshd-reload-inactive-0103.md @@ -0,0 +1,55 @@ +# PLAN — sshd-reload-inactive (hotfix) + +## Root cause +`install.sh` runs under `set -euo pipefail`. `harden_sshd()` (lines ~226-236) ends with an +unconditional `sudo systemctl reload ssh`. On WSL (and any host where sshd is socket-activated +or simply not started) ssh.service is inactive, systemctl prints +"ssh.service is not active, cannot reload." and exits non-zero, so `set -e` aborts the whole +install at "Deploying sshd hardening drop-in". openssh-server itself IS installed (apt list, +line ~730); the drop-in has already been installed and validated with `sshd -t` at that point. + +## Exact edit (install.sh, harden_sshd, tabs) — revised after challenge +1. Before `sudo sshd -t`, add one line so the validation also works when ssh.service is + inactive (privsep dir normally created by the unit's RuntimeDirectory; absent on Ubuntu + 24.04 when sshd never ran — challenge correctness MAJOR + robustness MINOR 2): +``` + sudo install -d -m 0755 /run/sshd +``` +2. Replace the unconditional reload +``` + sudo systemctl reload ssh +} +``` +with +``` + if systemctl is-active --quiet ssh; then + sudo systemctl try-reload-or-restart ssh + else + echo "ssh.service not running — drop-in in place, applies at the next sshd start" + fi +} +``` +`try-reload-or-restart` exits 0 if the unit stopped between the check and the call +(robustness MINOR 1); the `is-active` guard stays for the informative echo. +3. Update the function's header comment (one or two lines): /run/sshd is created so `sshd -t` + works without a running sshd; the reload happens only when sshd is running. + +## Smoke (robustness MINOR 3 — exercise the else branch for real) +In the scratchpad, extract `harden_sshd` from install.sh into a temp script with +`SCRIPT_DIR` pointing at the repo, prepend a stub dir to PATH where `systemctl` exits 3 for +`is-active` (and 0 otherwise), `sudo` runs its arguments without privilege, `sshd` exits 0; +run under `set -euo pipefail` and assert exit 0 + the "drop-in in place" echo. Then flip the +stub (`is-active` exits 0) and assert `try-reload-or-restart ssh` is what gets called. +Plus `bash -n install.sh` and `shellcheck install.sh`. + +## Constraints +- No other change; `offer_ssh_memory_guard` uses `systemctl restart ssh`, which starts an inactive + unit and is out of scope. +- shellcheck clean, `bash -n` clean, tabs. + +## Challenge summary +- correctness MAJOR (/run/sshd missing → sshd -t false reject) → closed by edit 1. +- robustness MINOR 1 (check-then-act race) → closed by `try-reload-or-restart`. +- robustness MINOR 2 (same /run/sshd risk) → closed by edit 1. +- robustness MINOR 3 (gate never runs the else path) → closed by the Smoke section. +- simplicity: SOLID, no findings.