diff --git a/guides/repository-flow.md b/guides/repository-flow.md new file mode 100644 index 00000000..f5070f78 --- /dev/null +++ b/guides/repository-flow.md @@ -0,0 +1,11 @@ +# Repository choice in dependency and safe-update flows + +On PVE 8/9, the shared helper checks `pvesubscription get` (only the `status` field, never logging the key) and reads parsed APT sources through Proxmox's `/nodes/localhost/apt/repositories` API. An **active** subscription preserves the configured Enterprise repository and all other sources; if no PVE channel is active, the flow stops for manual repair rather than adding one. An already active PVE no-subscription or test channel is likewise preserved. Repository files may use custom names, `.list` or deb822 `.sources`, and mirrors; presence of a filename alone does not establish an active channel. + +A fresh ISO may have Enterprise enabled but **no subscription** (`status: notfound`). Before changing anything, interactive flows ask: **“This host has no subscription, switch to the no-subscription repository?”** Acceptance disables active Enterprise PVE entries with Proxmox's per-entry API and adds Proxmox's standard no-subscription PVE repository. If an Enterprise Ceph source is active, it is disabled too, but **no Ceph replacement is chosen**: the appropriate Ceph release/channel must be selected separately in **Node > Updates > Repositories** if Ceph is used. Unrelated stanzas and Debian sources remain untouched. A mixed-component Enterprise stanza cannot be disabled safely, so the flow stops for manual splitting instead. A missing Debian base source also stops for manual configuration, rather than guessing a mirror. + +Declining or running without an interactive terminal/web dialog makes **no source changes**. Unrecognized, expired, invalid, suspended, malformed, or unavailable subscription status is **not** interpreted as no subscription; lookup and repository parser errors stop before changes. The apply step rechecks the subscription and repository inventory and uses the API digest on writes. A failure during a multi-entry API update can leave a partial switch; inspect the repository UI before retrying. General APT error-handling policy is unchanged in this focused patch; stricter index refresh and broader dependency validation remain separate work. The NVIDIA interactive driver installer checks repositories before downloading or removing the working driver, and the noninteractive reinstall propagates refusal. + +This policy does **not** change the legacy automated post-install repository updater, which separately advertises and asks for free repositories. No fresh-ISO live integration has been run; fixture tests exercise the policy without touching the host's APT configuration. + +Primary references: [Proxmox subscription CLI and status schema](https://github.com/proxmox/pve-manager/blob/master/PVE/CLI/pvesubscription.pm), [Proxmox APT repository API](https://github.com/proxmox/pve-manager/blob/master/PVE/API2/APT.pm), [Proxmox standard repository definitions](https://github.com/proxmox/proxmox-rs/blob/master/proxmox-apt/src/repositories/standard.rs), [Package Repositories](https://pve.proxmox.com/wiki/Package_Repositories). diff --git a/scripts/global/repository-functions.sh b/scripts/global/repository-functions.sh new file mode 100644 index 00000000..0c3d1425 --- /dev/null +++ b/scripts/global/repository-functions.sh @@ -0,0 +1,36 @@ +#!/bin/bash +# Shared repository policy for package-install and safe-update flows. +# Proxmox's own repository API parses .list/.sources and edits individual +# entries. No shell rewriting of operator-maintained APT files. +repository_policy() { + python3 "$(dirname "${BASH_SOURCE[0]}")/repository_policy.py" "$@" +} + +ensure_repositories() { + local version suite decision + version=$(pveversion 2>/dev/null | grep -oP 'pve-manager/\K[0-9]+' | head -1) + case "$version" in + 8) suite=bookworm ;; + 9) suite=trixie ;; + *) msg_error "$(translate 'Unsupported or unknown Proxmox version; no repository changed.')"; return 1 ;; + esac + # Do not hide diagnostics or open a spinner during user interaction. + decision=$(repository_policy plan "$suite") || return 1 + case "$decision" in + preserve) return 0 ;; + offer) ;; + *) msg_error "$(translate 'Repository policy returned an unexpected result.')"; return 1 ;; + esac + if ! declare -F hybrid_yesno >/dev/null || { [[ ! -t 0 ]] && ! { declare -F is_web_mode >/dev/null && is_web_mode; }; }; then + msg_error "$(translate 'No subscription and no usable PVE repository. Noninteractive mode cannot change APT sources; configure them in Node > Updates > Repositories.')" + return 1 + fi + if ! hybrid_yesno "$(translate 'Proxmox repository')" \ + "$(translate 'This host has no subscription, switch to the no-subscription repository? The inaccessible Enterprise PVE source will be disabled. Enterprise Ceph sources, if present, will be disabled without choosing a replacement Ceph channel; configure Ceph separately if needed.')" 16 90; then + msg_error "$(translate 'Repository switch declined; no APT source changed.')" + return 1 + fi + decision=$(repository_policy apply "$suite") || return 1 + [[ "$decision" == changed || "$decision" == preserve ]] || return 1 + return 0 +} diff --git a/scripts/global/repository_policy.py b/scripts/global/repository_policy.py new file mode 100644 index 00000000..4241ea46 --- /dev/null +++ b/scripts/global/repository_policy.py @@ -0,0 +1,142 @@ +#!/usr/bin/env python3 +"""Proxmox repository policy via its own parsed APT repository API. + +Only `notfound` permits an offered switch. Never print raw subscription or +repository API responses: they can contain subscription keys or credentials. +""" +import copy +import json +import re +import subprocess +import sys + +ENDPOINT = '/nodes/localhost/apt/repositories' +PVE_CHANNELS = {'pve-enterprise', 'pve-no-subscription', 'pve-test', 'pvetest'} + + +def run(args): + try: + return subprocess.run(args, check=True, capture_output=True, text=True).stdout + except (OSError, subprocess.CalledProcessError) as exc: + raise ValueError(f'Unable to query or change Proxmox repository state ({args[0]}). Check the service and retry.') from exc + + +def subscription_status(): + # pvesubscription's CLI get printer emits sorted "key: value" lines, + # including a secret key and server ID. Only parse the exact status line. + output = run(['pvesubscription', 'get']) + statuses = re.findall(r'^status: ([a-z]+)$', output, re.MULTILINE) + if len(statuses) != 1 or statuses[0] not in ('active', 'notfound'): + raise ValueError('Subscription status is not unambiguously active or notfound. Check pvesubscription get locally; no repository changed.') + return statuses[0] + + +def repository_state(suite): + try: + data = json.loads(run(['pvesh', 'get', ENDPOINT, '--output-format', 'json'])) + if not isinstance(data, dict) or not isinstance(data['files'], list) or not isinstance(data['errors'], list) or not isinstance(data['digest'], str) or not isinstance(data['standard-repos'], list): + raise ValueError() + if data['errors']: + raise ValueError() + entries = [] + for file in data['files']: + for index, row in enumerate(file['repositories']): + if not isinstance(row, dict): + raise ValueError() + # Flat repositories (e.g. suite './') legitimately omit this. + row.setdefault('Components', []) + if not isinstance(row['Enabled'], bool) or not all( + isinstance(row[k], list) and all(isinstance(value, str) for value in row[k]) + for k in ('Types', 'URIs', 'Suites', 'Components') + ): + raise ValueError() + entries.append((file['path'], index, row)) + return data, entries + except (KeyError, TypeError, ValueError, IndexError) as exc: + raise ValueError('Proxmox APT repository inventory is invalid or reports parse errors. Fix sources in Node > Updates > Repositories; no repository changed.') from exc + + +def evaluate(suite, apply=False): + if suite not in ('bookworm', 'trixie'): + raise ValueError('Unsupported Proxmox suite; no repository changed.') + status = subscription_status() # fail closed before inventory or any write + data, entries = repository_state(suite) + pve = [] + ceph = [] + debian = False + for path, index, row in entries: + if not row['Enabled'] or 'deb' not in row['Types']: + continue + components = set(row['Components']) + if components & PVE_CHANNELS: + if suite not in row['Suites']: + raise ValueError('PVE repository suite does not match the installed version; no repository changed.') + pve.append((path, index, row)) + if 'main' in components and suite in row['Suites'] and not components & PVE_CHANNELS: + debian = True + if 'enterprise' in components and any('/ceph-' in uri for uri in row['URIs']): + if suite not in row['Suites']: + raise ValueError('Enterprise Ceph repository suite does not match this PVE version; correct it in the Proxmox repository UI before switching.') + ceph.append((path, index, row)) + + if status == 'active': + if not pve: + raise ValueError('Host has an active subscription but no active PVE repository. Configure its Enterprise source in Node > Updates > Repositories; no repository changed.') + return 'preserve' + if any(set(row['Components']) & {'pve-no-subscription', 'pve-test', 'pvetest'} for _, _, row in pve): + return 'preserve' + if not debian: + raise ValueError('No active Debian base repository for this suite; configure it in the Proxmox repository UI before continuing.') + # No active PVE source or only inaccessible Enterprise. A mixed stanza + # cannot be disabled without also disabling an unrelated component. + disable = [item for item in pve if 'pve-enterprise' in item[2]['Components']] + ceph + for _, _, row in disable: + expected = {'pve-enterprise'} if 'pve-enterprise' in row['Components'] else {'enterprise'} + if set(row['Components']) != expected: + raise ValueError('Enterprise shares an APT stanza with other components. Split it in the Proxmox repository UI; no repository changed.') + handles = [r.get('handle') for r in data['standard-repos'] if r.get('handle') == 'no-subscription'] + if len(handles) != 1: + raise ValueError('Proxmox no-subscription standard repository handle unavailable; no repository changed.') + if not apply: + return 'offer' + # Adopt a refreshed digest only after checking the complete expected + # inventory: path/index targets are unsafe if another writer changed it. + expected = copy.deepcopy(data['files']) + try: + for path, index, original in disable: + next(file for file in expected if file['path'] == path)['repositories'][index]['Enabled'] = False + run(['pvesh', 'create', ENDPOINT, '--path', path, '--index', str(index), + '--enabled', '0', '--digest', data['digest']]) + refreshed, current = repository_state(suite) + # Per-file digests change when Proxmox serializes the disabled entry. + inventory = lambda files: sorted( + ({key: value for key, value in file.items() if key != 'digest'} for file in files), + key=lambda file: file['path']) + if inventory(refreshed['files']) != inventory(expected): + raise ValueError('Could not verify the expected repository inventory after Enterprise disable; possible concurrent edit. Inspect the Proxmox repository UI before retrying.') + data = refreshed + run(['pvesh', 'set', ENDPOINT, '--handle', 'no-subscription', '--digest', data['digest']]) + _, current = repository_state(suite) + if not any(row['Enabled'] and 'deb' in row['Types'] and suite in row['Suites'] + and 'pve-no-subscription' in row['Components'] for _, _, row in current) \ + or any(p == path and i == index and row['Enabled'] + for path, index, _ in disable for p, i, row in current): + raise ValueError('Could not verify the switched repositories; inspect the Proxmox repository UI before retrying.') + except ValueError as exc: + raise ValueError('Could not complete or verify the repository switch; changes may be partial and repository state is unknown. Inspect Node > Updates > Repositories before retrying; no automatic rollback was attempted.') from exc + return 'changed' + + +def main(): + try: + if len(sys.argv) != 3 or sys.argv[1] not in ('plan', 'apply'): + raise ValueError('Usage: repository_policy.py plan|apply bookworm|trixie') + print(evaluate(sys.argv[2], apply=sys.argv[1] == 'apply')) + except ValueError as exc: + print(str(exc), file=sys.stderr) + return 1 + return 0 + + +if __name__ == '__main__': + sys.exit(main()) diff --git a/scripts/global/share-common.func b/scripts/global/share-common.func index 02a9297e..6381dc0f 100644 --- a/scripts/global/share-common.func +++ b/scripts/global/share-common.func @@ -12,88 +12,8 @@ -# ========================================================== -# Ensure repositories are properly configured -# ========================================================== -ensure_repositories() { - local pve_version need_update=false - pve_version=$(pveversion 2>/dev/null | grep -oP 'pve-manager/\K[0-9]+' | head -1) - - if [[ -z "$pve_version" ]]; then - msg_error "$(translate 'Unable to detect Proxmox version.')" - return 1 - fi - - if (( pve_version >= 9 )); then - # ===== PVE 9 (Debian 13 - trixie) ===== - # proxmox.sources (no-subscription) - create if missing. - # chmod 0644 explicit on every new .sources file: under the default - # root umask 0027 the redirect would land at 0640, which the PVE 9 - # webgui's repository manager treats as unparseable and silently - # hides the source — issue #230. - if [[ ! -f /etc/apt/sources.list.d/proxmox.sources ]]; then - cat > /etc/apt/sources.list.d/proxmox.sources <<'EOF' -Enabled: true -Types: deb -URIs: http://download.proxmox.com/debian/pve -Suites: trixie -Components: pve-no-subscription -Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg -EOF - chmod 0644 /etc/apt/sources.list.d/proxmox.sources - need_update=true - fi - - # debian.sources - create if missing - if [[ ! -f /etc/apt/sources.list.d/debian.sources ]]; then - cat > /etc/apt/sources.list.d/debian.sources <<'EOF' -Types: deb -URIs: http://deb.debian.org/debian/ -Suites: trixie trixie-updates -Components: main contrib non-free-firmware -Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg - -Types: deb -URIs: http://security.debian.org/debian-security/ -Suites: trixie-security -Components: main contrib non-free-firmware -Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg -EOF - chmod 0644 /etc/apt/sources.list.d/debian.sources - need_update=true - fi - - else - # ===== PVE 8 (Debian 12 - bookworm) ===== - local sources_file="/etc/apt/sources.list" - - # Debian base (create or append minimal lines if missing) - if ! grep -qE 'deb .* bookworm .* main' "$sources_file" 2>/dev/null; then - { - echo "deb http://deb.debian.org/debian bookworm main contrib non-free non-free-firmware" - echo "deb http://deb.debian.org/debian bookworm-updates main contrib non-free non-free-firmware" - echo "deb http://security.debian.org/debian-security bookworm-security main contrib non-free non-free-firmware" - } >> "$sources_file" - need_update=true - fi - - # Proxmox no-subscription list (classic) if missing - if [[ ! -f /etc/apt/sources.list.d/pve-no-subscription.list ]]; then - echo "deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription" \ - > /etc/apt/sources.list.d/pve-no-subscription.list - need_update=true - fi - fi - - # apt-get update only if needed or lists are empty - if [[ "$need_update" == true ]] || [[ ! -d /var/lib/apt/lists || -z "$(ls -A /var/lib/apt/lists 2>/dev/null)" ]]; then - msg_info "$(translate 'Updating APT package lists...')" - apt-get update >/dev/null 2>&1 || apt-get update - msg_ok "$(translate 'APT package lists updated')" - fi - - return 0 -} +# Repository policy shared with utility installers. +source "$(dirname "${BASH_SOURCE[0]}")/repository-functions.sh" # ========================================================== diff --git a/scripts/global/update-pve-safe.sh b/scripts/global/update-pve-safe.sh index b53f8184..5fd99fdc 100644 --- a/scripts/global/update-pve-safe.sh +++ b/scripts/global/update-pve-safe.sh @@ -9,36 +9,31 @@ # Description: # Update path intended for a Proxmox host ALREADY in # production. Unlike scripts/global/update-pve8.sh and -# update-pve9_2.sh (invoked by post_install), this variant -# NEVER modifies the operator's own configuration: +# update-pve9_2.sh (invoked by post_install), this variant preserves +# operator-maintained sources unless an unsubscribed host explicitly chooses +# to switch. Otherwise, the operator's source configuration is preserved: # -# - Does NOT disable Enterprise / Ceph repositories -# - Does NOT delete legacy repo files +# - Does NOT silently disable Enterprise / Ceph repositories +# - Does NOT delete or deduplicate existing repo files # - Does NOT overwrite proxmox.sources / debian.sources -# when they already exist # - Does NOT purge alternative NTP services # - Does NOT force-install zfsutils / chrony / # proxmox-backup-restore-image # - Does NOT write no-firmware-warnings.conf # # What it DOES: -# 1. Sanity checks (disk space, network) -# 2. ensure_repositories() — only when repos are MISSING +# 1. Sanity checks (disk space) +# 2. ensure_repositories() — check subscription and request consent if needed # 3. apt-get update, with automatic GPG key import when apt # reports NO_PUBKEY (any repo, user's or ours) -# 4. cleanup_duplicate_repos() — exact URL+Suite+Component -# match against proxmox.sources / debian.sources; leaves -# unrelated custom `download.proxmox.com/*` and -# user-authored pve-*.list files alone; backs each file -# up before modifying -# 5. Detect pending upgrades + security count -# 6. Confirmation dialog -# 7. apt-get full-upgrade with --force-confdef / --force-confold +# 4. Detect pending upgrades + security count +# 5. Confirmation dialog +# 6. apt-get full-upgrade with --force-confdef / --force-confold # (never overwrites the operator's edited config files) -# 8. lvm_repair_check() — refreshes VG metadata when disks +# 7. lvm_repair_check() — refreshes VG metadata when disks # passed through to guest VMs (DSM, TrueNAS, …) come back # with old PV headers -# 9. apt-get autoremove + autoclean +# 8. apt-get autoremove + autoclean # # Reboot detection is handled by the caller (utilities/proxmox_update.sh). # ========================================================== @@ -66,6 +61,8 @@ source_install_functions() { local f="$LOCAL_SCRIPTS/global/utils-install-functions.sh" if [[ -f "$f" ]]; then source "$f" + else + return 1 fi } @@ -91,8 +88,11 @@ update_pve_safe() { local screen_capture="/tmp/proxmenux_screen_capture_$$.txt" : > "$screen_capture" - download_common_functions - source_install_functions + if ! download_common_functions || ! source_install_functions; then + msg_error "$(translate 'Required update helpers unavailable. Update stopped.')" + rm -f "$screen_capture" + return 1 + fi { msg_info2 "$(translate "Detected: Proxmox VE $pve_version — running safe update path")" @@ -110,39 +110,14 @@ update_pve_safe() { return 1 fi - # Reachability check: probe the public Proxmox repository over the - # transport apt is most likely to use. Many PVE installs use the - # official HTTP apt URI, while HTTPS may fail before apt ever runs - # if the CDN presents a certificate for another Proxmox hostname. - # Accept either transport and let apt-get update report repo-specific - # errors in the next step. - _repo_reachable() { - local url attempt - for url in "http://download.proxmox.com/" "https://download.proxmox.com/"; do - for attempt in 1 2; do - if curl -sfI --connect-timeout 5 --max-time 10 -o /dev/null "$url"; then - return 0 - fi - [[ $attempt -eq 1 ]] && sleep 1 - done - done - return 1 - } - if ! _repo_reachable; then - msg_error "$(translate "Cannot reach download.proxmox.com. Check network, proxy or DNS.")" - echo -e - msg_success "$(translate "Press Enter to return to menu...")" - read -r + # Enterprise hosts and custom mirrors need not reach the public CDN. + # The configured sources, not that hostname, are checked by APT below. + if ! declare -F ensure_repositories >/dev/null 2>&1 || ! ensure_repositories; then + msg_error "$(translate 'Repository check failed. Update stopped.')" rm -f "$screen_capture" return 1 fi - # ── 2. ensure_repositories: adds base Proxmox+Debian repos only if - # they don't already exist. On a configured host this is a no-op. ── - if declare -f ensure_repositories >/dev/null 2>&1; then - ensure_repositories - fi - # ── 3. apt-get update with automatic key recovery ── local update_output update_exit_code update_output=$(apt-get update 2>&1) @@ -191,11 +166,8 @@ update_pve_safe() { fi fi - # ── 4. Precise duplicate cleanup (exact URL+Suite+Component match, - # backs up files before modifying). Skipped if unavailable. ── - if declare -f cleanup_duplicate_repos >/dev/null 2>&1; then - cleanup_duplicate_repos - fi + # No duplicate-source rewrite here: even a seemingly duplicate entry may + # be operator-maintained. Only the consented switch above edits sources. # ── 5-6. Detect + confirm ── local current_pve_version available_pve_version upgradable security_updates diff --git a/scripts/global/utils-install-functions.sh b/scripts/global/utils-install-functions.sh index 7e9839a4..a1c702fa 100644 --- a/scripts/global/utils-install-functions.sh +++ b/scripts/global/utils-install-functions.sh @@ -39,95 +39,14 @@ PROXMENUX_UTILS=( ) -# Ensure APT repositories are configured for the current PVE version. -# Creates missing no-subscription repo entries for PVE8 (bookworm) or PVE9 (trixie). -# Shared journal helpers, so any script sourcing this file records what -# it installs without arranging for it. +# Shared journal helpers for utility installs. if [[ -f "${LOCAL_SCRIPTS:-/usr/local/share/proxmenux/scripts}/global/pmx_journal.sh" ]]; then source "${LOCAL_SCRIPTS:-/usr/local/share/proxmenux/scripts}/global/pmx_journal.sh" fi -ensure_repositories() { - local FUNC_VERSION="1.0" - pmx_journal_context "ensure_repositories" "$FUNC_VERSION" - local pve_version need_update=false - pve_version=$(pveversion 2>/dev/null | grep -oP 'pve-manager/\K[0-9]+' | head -1) - - if [[ -z "$pve_version" ]]; then - msg_error "Unable to detect Proxmox version." - return 1 - fi - - if (( pve_version >= 9 )); then - # ===== PVE 9 (Debian 13 - trixie) ===== - # Force 0644 (world-readable) on every .sources file we drop. - # Under the default root umask 0027 the redirect would land at - # 0640, which the PVE 9 webgui's repository manager treats as - # unparseable and silently hides the source — issue #230. - if [[ ! -f /etc/apt/sources.list.d/proxmox.sources ]]; then - pmx_write_file /etc/apt/sources.list.d/proxmox.sources <<'EOF' -Enabled: true -Types: deb -URIs: http://download.proxmox.com/debian/pve -Suites: trixie -Components: pve-no-subscription -Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg -EOF - chmod 0644 /etc/apt/sources.list.d/proxmox.sources - need_update=true - fi - - if [[ ! -f /etc/apt/sources.list.d/debian.sources ]]; then - pmx_write_file /etc/apt/sources.list.d/debian.sources <<'EOF' -Types: deb -URIs: http://deb.debian.org/debian/ -Suites: trixie trixie-updates -Components: main contrib non-free-firmware -Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg - -Types: deb -URIs: http://security.debian.org/debian-security/ -Suites: trixie-security -Components: main contrib non-free-firmware -Signed-By: /usr/share/keyrings/debian-archive-keyring.gpg -EOF - chmod 0644 /etc/apt/sources.list.d/debian.sources - need_update=true - fi - - else - # ===== PVE 8 (Debian 12 - bookworm) ===== - local sources_file="/etc/apt/sources.list" - - if ! grep -qE 'deb .* bookworm .* main' "$sources_file" 2>/dev/null; then - { - echo "deb http://deb.debian.org/debian bookworm main contrib non-free non-free-firmware" - echo "deb http://deb.debian.org/debian bookworm-updates main contrib non-free non-free-firmware" - echo "deb http://security.debian.org/debian-security bookworm-security main contrib non-free non-free-firmware" - } | pmx_append_file "$sources_file" - need_update=true - fi - - if [[ ! -f /etc/apt/sources.list.d/pve-no-subscription.list ]]; then - echo "deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription" \ - | pmx_write_file /etc/apt/sources.list.d/pve-no-subscription.list - need_update=true - fi - fi - - if [[ "$need_update" == true ]] || [[ ! -d /var/lib/apt/lists || -z "$(ls -A /var/lib/apt/lists 2>/dev/null)" ]]; then - msg_info "$(translate "Updating APT package lists...")" - apt-get update >/dev/null 2>&1 || apt-get update - # Spinner pair: msg_info must be closed before returning. - # Without this the next `msg_info` caller spawns a second - # spinner on top of ours and the original line never gets - # ✓'d — leaving a dangling progress char on screen. - msg_ok "$(translate "APT package lists updated")" - fi - - return 0 -} +# Shared subscription and consent policy for all utility callers. +source "$(dirname "${BASH_SOURCE[0]}")/repository-functions.sh" # Install a single package and verify the resulting command is available. diff --git a/scripts/gpu_tpu/nvidia_installer.sh b/scripts/gpu_tpu/nvidia_installer.sh index 8aa880e9..0a364451 100644 --- a/scripts/gpu_tpu/nvidia_installer.sh +++ b/scripts/gpu_tpu/nvidia_installer.sh @@ -525,21 +525,24 @@ offer_lxc_updates_if_any() { # ========================================================== ensure_repos_and_headers() { pmx_journal_context "ensure_repos_and_headers" "1.3" "nvidia_installer.sh" - # Bootstrap APT repos FIRST. On a fresh Proxmox install the - # pve-no-subscription / debian repos aren't configured by default - # → `pve-headers-$(uname -r)` and `build-essential` come back as - # "Unable to locate package" and the NVIDIA install bails out with - # "no cc found". We delegate to the shared helper (same one the - # post-install flow uses), which owns its own spinner pair — that's - # why this block has to run BEFORE we open our own msg_info. + # Check repositories before opening the headers spinner. A fresh ISO may + # have Enterprise enabled but no subscription; the shared policy asks the + # operator before switching. Refusal stops before driver removal. if ! declare -F ensure_repositories >/dev/null 2>&1; then local _utils_install="$LOCAL_SCRIPTS/global/utils-install-functions.sh" [[ ! -f "$_utils_install" ]] && _utils_install="/usr/local/share/proxmenux/scripts/global/utils-install-functions.sh" # shellcheck source=/dev/null [[ -f "$_utils_install" ]] && source "$_utils_install" fi - if declare -F ensure_repositories >/dev/null 2>&1; then - ensure_repositories >>"$LOG_FILE" 2>&1 || true + if ! declare -F ensure_repositories >/dev/null 2>&1; then + msg_error "$(translate 'Repository helper unavailable. NVIDIA installation stopped.')" + return 1 + fi + # Keep the consent dialog and diagnostic visible, not just in the log. + ensure_repositories 2>&1 | tee -a "$LOG_FILE" + if (( PIPESTATUS[0] != 0 )); then + msg_error "$(translate 'Repository check failed. NVIDIA installation stopped.')" + return 1 fi # Now own the spinner for the headers + build-tools check. @@ -548,7 +551,10 @@ ensure_repos_and_headers() { local kver kver=$(uname -r) - apt-get update -qq >>"$LOG_FILE" 2>&1 + if ! apt-get update -qq >>"$LOG_FILE" 2>&1; then + msg_error "$(translate 'APT update failed. Check repository access; NVIDIA installation stopped.')" + return 1 + fi if ! dpkg -s "pve-headers-$kver" >/dev/null 2>&1 && \ ! dpkg -s "proxmox-headers-$kver" >/dev/null 2>&1; then @@ -2184,7 +2190,7 @@ main() { # Headers before anything else: the build check below needs them, # and so does DKMS afterwards. - ensure_repos_and_headers + ensure_repos_and_headers || exit 1 installer=$(download_nvidia_installer "$DRIVER_VERSION") local download_result=$? @@ -2442,7 +2448,7 @@ auto_reinstall_from_state() { # dialogs and confirmations. echo "Reinstalling NVIDIA driver $DRIVER_VERSION non-interactively..." | tee -a "$LOG_FILE" ensure_workdir - ensure_repos_and_headers >>"$LOG_FILE" 2>&1 + ensure_repos_and_headers >>"$LOG_FILE" 2>&1 || return 2 blacklist_nouveau >>"$LOG_FILE" 2>&1 ensure_modules_config >>"$LOG_FILE" 2>&1 diff --git a/tests/test_repository_preservation.py b/tests/test_repository_preservation.py new file mode 100644 index 00000000..2dff9c5d --- /dev/null +++ b/tests/test_repository_preservation.py @@ -0,0 +1,381 @@ +"""Fixture-only policy checks: never call host pvesh or edit host APT sources.""" +import importlib.util +import json +import os +from pathlib import Path +import subprocess +import tempfile +import unittest +from unittest.mock import patch + +ROOT = Path(__file__).resolve().parents[1] +POLICY = ROOT / 'scripts/global/repository_policy.py' + + +def repo(components, *, uri='https://mirror.example/debian/pve', suite='trixie', enabled=True): + return {'Types': ['deb'], 'URIs': [uri], 'Suites': [suite], + 'Components': components.split(), 'Enabled': enabled} + + +def fixture(*entries): + files = {} + for path, entries_for_file in entries: + files[path] = {'path': path, 'repositories': entries_for_file} + return {'files': list(files.values()), 'errors': [], 'digest': 'fixture-digest', + 'standard-repos': [{'handle': 'no-subscription', 'name': 'PVE No-Subscription'}]} + + +class RepositoryPolicyTest(unittest.TestCase): + @classmethod + def setUpClass(cls): + spec = importlib.util.spec_from_file_location('repository_policy', POLICY) + cls.module = importlib.util.module_from_spec(spec) + spec.loader.exec_module(cls.module) + + def setUp(self): + self.calls = [] + self.mutate = True + self.after_write = lambda: None + self.writes = 0 + self.subscription = 'status: notfound\nmessage: There is no subscription key\n' + self.repositories = fixture( + ('/etc/apt/sources.list.d/custom.sources', [repo('pve-enterprise')]), + ('/etc/apt/sources.list.d/debian.sources', [repo('main', uri='https://deb.example/debian')]), + ) + + def run_policy(self, apply=False, suite='trixie'): + def run(args): + self.calls.append(args) + if args[:2] == ['pvesubscription', 'get']: + if isinstance(self.subscription, Exception): + raise ValueError('service unavailable') from self.subscription + return self.subscription + if args[:2] == ['pvesh', 'get']: + return json.dumps(self.repositories) + if args[1] in ('create', 'set'): + self.assertEqual(args[args.index('--digest') + 1], self.repositories['digest']) + self.writes += 1 + if self.mutate and args[:2] == ['pvesh', 'create']: + path = args[args.index('--path') + 1] + index = int(args[args.index('--index') + 1]) + next(f for f in self.repositories['files'] if f['path'] == path)['repositories'][index]['Enabled'] = False + if self.mutate and args[:2] == ['pvesh', 'set']: + self.repositories['files'].append({'path': '/etc/apt/sources.list.d/proxmox.sources', + 'repositories': [repo('pve-no-subscription')]}) + self.repositories['digest'] = f'write-{self.writes}' + self.after_write() + return '' + with patch.object(self.module, 'run', side_effect=run): + return self.module.evaluate(suite, apply=apply) + + def test_concurrent_inventory_edit_aborts_before_second_mutation(self): + for edit in ('replace', 'reorder', 'unrelated'): + with self.subTest(edit=edit): + self.setUp() + rows = self.repositories['files'][0]['repositories'] + rows.extend([repo('enterprise', uri='https://example/ceph-squid'), + repo('custom')]) + + def external_edit(): + if self.writes != 1: + return + if edit == 'replace': + rows[1] = repo('custom-replacement') + elif edit == 'reorder': + rows[1], rows[2] = rows[2], rows[1] + else: + self.repositories['files'][1]['repositories'][0]['Options'] = [ + {'Key': 'Signed-By', 'Values': ['/external/keyring.gpg']}] + self.repositories['digest'] = 'external-edit' + + self.after_write = external_edit + with self.assertRaises(ValueError): + self.run_policy(apply=True) + self.assertEqual(self.writes, 1) + self.assertTrue(rows[1]['Enabled']) + self.assertTrue(rows[2]['Enabled']) + + def test_failed_post_write_readback_reports_partial_or_unknown_state(self): + for after in (1, 2): + with self.subTest(after=after): + self.setUp() + + def break_readback(): + if self.writes == after: + self.repositories['errors'] = [{'error': 'external parse error'}] + + self.after_write = break_readback + with self.assertRaises(ValueError) as error: + self.run_policy(apply=True) + self.assertEqual(self.writes, after) + self.assertNotIn('no repository changed', str(error.exception)) + self.assertRegex(str(error.exception), 'partial|unknown') + + def test_inventory_comparison_accepts_api_serialization_and_new_digests(self): + self.repositories['files'][0]['repositories'].append( + repo('enterprise', uri='https://example/ceph-squid')) + self.repositories['files'].append({'path': '/etc/apt/sources.list.d/flat.list', + 'repositories': [repo('', uri='https://example/flat', suite='./')]}) + + def serialize(): + for file in self.repositories['files']: + file['digest'] = f'file-digest-{self.writes}' + for row in file['repositories']: + if row.get('Components') == []: + del row['Components'] + self.repositories['files'].reverse() + + self.after_write = serialize + self.assertEqual(self.run_policy(apply=True), 'changed') + self.assertEqual(self.writes, 3) + + def test_fresh_iso_requires_consent_before_any_write(self): + self.assertEqual(self.run_policy(), 'offer') + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_consented_switch_disables_enterprise_then_adds_public_pve(self): + self.assertEqual(self.run_policy(apply=True), 'changed') + writes = [c for c in self.calls if c[1] in ('create', 'set')] + self.assertEqual([c[1] for c in writes], ['create', 'set']) + self.assertEqual(writes[0][writes[0].index('--path') + 1], '/etc/apt/sources.list.d/custom.sources') + self.assertEqual(writes[0][writes[0].index('--enabled') + 1], '0') + self.assertEqual(writes[1][writes[1].index('--handle') + 1], 'no-subscription') + + def test_api_noop_after_write_does_not_report_success(self): + self.mutate = False + with self.assertRaisesRegex(ValueError, 'verify'): + self.run_policy(apply=True) + + def test_flat_repository_without_components_is_preserved(self): + self.subscription = 'status: active\n' + flat = repo('', uri='https://vendor.example/flat', suite='./') + del flat['Components'] + self.repositories['files'].append({'path': '/etc/apt/sources.list.d/vendor.list', + 'repositories': [flat]}) + before = json.dumps(self.repositories) + self.assertEqual(self.run_policy(apply=True), 'preserve') + self.assertEqual(json.dumps(self.repositories), before) + self.assertEqual(self.writes, 0) + + def test_malformed_components_fail_closed(self): + self.subscription = 'status: active\n' + for components in (None, 'main', {}, [42]): + with self.subTest(components=components): + self.repositories['files'][1]['repositories'][0]['Components'] = components + with self.assertRaises(ValueError): + self.run_policy(apply=True) + self.assertEqual(self.writes, 0) + + def test_active_subscription_preserves_enterprise(self): + self.subscription = 'key: pve1x-SECRET\nstatus: active\nserverid: SECRET\n' + self.assertEqual(self.run_policy(apply=True), 'preserve') + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_active_subscription_without_pve_source_stops_without_add(self): + self.subscription = 'status: active\n' + self.repositories['files'][0]['repositories'][0]['Enabled'] = False + with self.assertRaisesRegex(ValueError, 'active subscription'): + self.run_policy(apply=True) + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_existing_public_or_test_preserved_without_switch(self): + for channel in ('pve-no-subscription', 'pve-test'): + with self.subTest(channel=channel): + self.repositories['files'][0]['repositories'] = [repo(channel)] + self.assertEqual(self.run_policy(), 'preserve') + + def test_legacy_pvetest_list_on_pve8_is_preserved(self): + self.repositories = fixture( + ('/etc/apt/sources.list.d/operator.list', [repo('pvetest', suite='bookworm')]), + ('/etc/apt/sources.list', [repo('main', uri='https://deb.example/debian', suite='bookworm')]), + ) + self.assertEqual(self.run_policy(suite='bookworm'), 'preserve') + + def test_disabled_enterprise_stanza_is_not_disabled_again(self): + self.repositories['files'][0]['repositories'][0]['Enabled'] = False + self.assertEqual(self.run_policy(apply=True), 'changed') + self.assertFalse(any(c[1] == 'create' for c in self.calls)) + + def test_unrecognized_and_failed_subscription_cannot_mutate(self): + for response in ('status: unknown\n', 'status: invalid\n', 'status: active\nstatus: notfound\n', + 'key: secret\n', RuntimeError('service unavailable')): + with self.subTest(response=response): + self.subscription = response + self.calls = [] + with self.assertRaises(ValueError): + self.run_policy(apply=True) + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_multiple_stanzas_disables_only_pve_enterprise_and_ceph_enterprise(self): + self.repositories = fixture( + ('/etc/apt/sources.list.d/anything.sources', [ + repo('main', uri='https://debian.example/debian'), + repo('pve-enterprise'), + repo('enterprise', uri='https://enterprise.proxmox.com/debian/ceph-squid'), + repo('pve-enterprise', enabled=False), + ]), + ) + self.assertEqual(self.run_policy(apply=True), 'changed') + writes = [c for c in self.calls if c[1] == 'create'] + self.assertEqual([int(c[c.index('--index') + 1]) for c in writes], [1, 2]) + + def test_mixed_components_block_switch_without_partial_write(self): + self.repositories['files'][0]['repositories'] = [repo('pve-enterprise custom')] + with self.assertRaises(ValueError): + self.run_policy(apply=True) + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_wrong_suite_and_parse_errors_block_before_writes(self): + for mode in ('suite', 'error'): + with self.subTest(mode=mode): + self.repositories = fixture( + ('/etc/apt/sources.list.d/custom.list', [repo('pve-enterprise', suite='bookworm')]), + ('/etc/apt/sources.list.d/debian.sources', [repo('main', uri='https://deb.example/debian')]), + ) + if mode == 'error': + self.repositories['errors'] = [{'path': 'custom.list', 'error': 'invalid'}] + with self.assertRaises(ValueError): + self.run_policy(apply=True) + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_ceph_enterprise_mirror_is_disabled_without_choosing_ceph_channel(self): + self.repositories['files'].append({'path': '/etc/apt/sources.list.d/storage.list', + 'repositories': [repo('enterprise', uri='https://mirror.example/ceph-squid')]}) + self.assertEqual(self.run_policy(apply=True), 'changed') + writes = [c for c in self.calls if c[1] == 'create'] + self.assertEqual(len(writes), 2) + self.assertFalse(any(c[1] == 'set' and 'ceph' in ' '.join(c) for c in self.calls)) + + def test_wrong_suite_ceph_stops_before_switch(self): + self.repositories['files'].append({'path': '/etc/apt/sources.list.d/ceph.sources', + 'repositories': [repo('enterprise', uri='https://enterprise.proxmox.com/debian/ceph-reef', suite='bookworm')]}) + with self.assertRaisesRegex(ValueError, 'suite'): + self.run_policy(apply=True) + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + def test_missing_debian_base_blocks_switch(self): + self.repositories['files'].pop() + with self.assertRaises(ValueError): + self.run_policy(apply=True) + self.assertFalse(any(c[1] in ('create', 'set') for c in self.calls)) + + +class CallerPropagationTest(unittest.TestCase): + def test_safe_update_refusal_blocks_apt_update(self): + source = (ROOT / 'scripts/global/update-pve-safe.sh').read_text() + body = source.split('update_pve_safe() {', 1)[1].split( + '\nif [[ "${BASH_SOURCE[0]}"', 1)[0] + with tempfile.TemporaryDirectory(dir=os.environ.get('TMPDIR')) as d: + body = body.replace('/var/log/', f'{d}/').replace( + '/tmp/proxmenux_screen_capture_', f'{d}/proxmenux_screen_capture_') + script = ''' + pveversion() { printf 'pve-manager/9.0.0\\n'; } + translate() { printf '%s' "$1"; } + msg_info2() { :; } + msg_error() { :; } + df() { printf 'Filesystem 1K-blocks Used Available Use%% Mounted on\\n'; + printf 'fixture 9999999 1 9999999 1%% /\\n'; } + download_common_functions() { :; } + source_install_functions() { :; } + ensure_repositories() { return 1; } + apt-get() { touch "$TEST_ROOT/apt_called"; } + update_pve_safe() { + ''' + body + '\nupdate_pve_safe\n' + result = subprocess.run(['bash', '-c', script], env=dict(os.environ, TEST_ROOT=d), + capture_output=True, text=True, stdin=subprocess.DEVNULL, + timeout=15) + self.assertNotEqual(result.returncode, 0) + self.assertFalse((Path(d) / 'apt_called').exists()) + + def test_nvidia_refusal_preserves_working_driver_before_uninstall(self): + source = (ROOT / 'scripts/gpu_tpu/nvidia_installer.sh').read_text() + main = 'main() {' + source.split('main() {', 1)[1].split( + '\n# ==========================================================\n# Non-interactive', 1)[0] + with tempfile.TemporaryDirectory(dir=os.environ.get('TMPDIR')) as d: + script = ''' + LOG_FILE="$TEST_ROOT/log"; screen_capture="$TEST_ROOT/screen" + NVIDIA_GPU_PRESENT=true; CURRENT_DRIVER_INSTALLED=true + CURRENT_DRIVER_VERSION=580.1; DRIVER_VERSION=580.2; ACTION=install + for fn in detect_nvidia_gpus detect_driver_status check_gpu_not_in_vm_passthrough \\ + check_stale_vfio_config_for_nvidia show_action_menu_if_installed \\ + show_install_overview get_system_info show_version_menu hybrid_yesno \\ + show_proxmenux_logo msg_title msg_info2 sleep; do + eval "$fn() { :; }" + done + translate() { printf '%s' "$1"; } + ensure_repos_and_headers() { return 1; } + download_nvidia_installer() { touch "$TEST_ROOT/downloaded"; return 1; } + complete_nvidia_uninstall() { touch "$TEST_ROOT/uninstalled"; } + ''' + main + '\nmain\n' + result = subprocess.run(['bash', '-c', script], + env=dict(os.environ, TEST_ROOT=d), + capture_output=True, text=True, timeout=15) + self.assertNotEqual(result.returncode, 0) + self.assertFalse((Path(d) / 'downloaded').exists()) + self.assertFalse((Path(d) / 'uninstalled').exists()) + + def test_nvidia_repository_failure_blocks_header_install(self): + source = (ROOT / 'scripts/gpu_tpu/nvidia_installer.sh').read_text() + fn = source.split('ensure_repos_and_headers() {', 1)[1].split( + '\n_nouveau_legacy_file_is_proxmenux_shape()', 1)[0] + with tempfile.TemporaryDirectory(dir=os.environ.get('TMPDIR')) as d: + script = ''' + LOG_FILE="$TEST_ROOT/log"; screen_capture="$TEST_ROOT/screen" + translate() { printf %s "$1"; } + msg_error() { :; } + ensure_repositories() { return 1; } + apt-get() { touch "$TEST_ROOT/apt_called"; } + ensure_repos_and_headers() { + ''' + fn + '\nensure_repos_and_headers\n' + result = subprocess.run(['bash', '-c', script], env=dict(os.environ, TEST_ROOT=d), + capture_output=True, text=True, timeout=15) + self.assertNotEqual(result.returncode, 0) + self.assertFalse((Path(d) / 'apt_called').exists()) + + +class ShellFlowTest(unittest.TestCase): + def test_web_consent_applies_only_after_prompt(self): + helper = ROOT / 'scripts/global/repository-functions.sh' + with tempfile.TemporaryDirectory(dir=os.environ.get('TMPDIR')) as d: + log = Path(d) / 'calls' + script = f'''source "{helper}" + pveversion() {{ echo pve-manager/9.0; }} + translate() {{ printf %s "$1"; }} + msg_error() {{ :; }} + is_web_mode() {{ return 0; }} + hybrid_yesno() {{ printf 'prompt\\n' >> "{log}"; [[ "$2" == *'This host has no subscription, switch to the no-subscription repository?'* ]]; }} + repository_policy() {{ printf '%s\\n' "$1" >> "{log}"; [[ "$1" == plan ]] && echo offer || echo changed; }} + ensure_repositories + ''' + result = subprocess.run(['bash', '-c', script], input='', text=True, capture_output=True) + self.assertEqual(result.returncode, 0, result.stderr) + self.assertEqual(log.read_text().splitlines(), ['plan', 'prompt', 'apply']) + + def test_refusal_and_noninteractive_never_apply(self): + helper = ROOT / 'scripts/global/repository-functions.sh' + with tempfile.TemporaryDirectory(dir=os.environ.get('TMPDIR')) as d: + mock = Path(d) / 'policy.py' + mock.write_text('import sys\nfrom pathlib import Path\n' + 'Path(sys.argv[1]).open("a").write(sys.argv[2]+"\\n")\n' + 'print("offer" if sys.argv[2] == "plan" else "changed")\n') + for web, confirm in ((False, True), (True, False)): + with self.subTest(web=web, confirm=confirm): + log = Path(d) / 'calls' + log.unlink(missing_ok=True) + script = f'''source "{helper}" + pveversion() {{ echo pve-manager/9.0; }} + translate() {{ printf %s "$1"; }} + msg_error() {{ :; }} + msg_info2() {{ :; }} + is_web_mode() {{ {'return 0' if web else 'return 1'}; }} + hybrid_yesno() {{ {'return 0' if confirm else 'return 1'}; }} + repository_policy() {{ python3 "{mock}" "{log}" "$1"; }} + ensure_repositories + ''' + result = subprocess.run(['bash', '-c', script], input='', text=True, capture_output=True) + self.assertNotEqual(result.returncode, 0) + self.assertEqual(log.read_text(), 'plan\n') + + +if __name__ == '__main__': + unittest.main()