release: 1.4.7 — failover на Keenetic, cron, таймаут pull

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

CHANGELOG.md

@@ -19,6 +19,12 @@ phobos-update beta      # или: phobos-update dev
 
 ---
 
+## 1.4.7 — роутеры Keenetic сами переключаются между серверами
+- Исправлено: после переключения на другой сервер туннель на роутере переставал работать — пир прописывался на порт обфускатора 13255, а он мог слушать другой. Порт теперь берётся из `wg-obfuscator.conf`.
+- Исправлено: проверка и обновление конфига по расписанию на Keenetic не запускались — задачи писались в `/opt/etc/crontab`, который демон не читает. Теперь busybox `crond` и crontab пользователя root.
+- Исправлено: `phobos-pull` навсегда зависал на `wget`, если туннель не работает (таймаут 8 с).
+- Роутерам: переустановить клиента или обновить скрипты из панели; серверу — `phobos-update stable`.
+
 ## 1.4.6 — роутеры снова получают смену адреса
 - Исправлено: роутер получал 403 при pull конфига, если клиент создан через API/бота (`/api/client/ensure`) — смена адреса/сервера до него не доходила. Теперь принимается и общий ключ из пакета, и личный токен клиента. Роутеры переустанавливать не нужно — достаточно обновить панель.
 

VERSION

@@ -1 +1 @@
-1.4.6
+1.4.7

client/phobos-health.sh

@@ -1,28 +1,41 @@
 #!/bin/sh
 # ============================================================
-#  Phobos Health Monitor — Auto-heal + Failover for routers
-#  Runs via cron every 60 seconds on Keenetic/Entware
-#
-#  Install: crontab -e → */1 * * * * /opt/etc/Phobos/phobos-health.sh
+#  Phobos Health Monitor v2 — Keenetic Edition
+#  Uses ndmc/RCI instead of wg-tools. Connectivity-based failover.
+#  Runs via cron every 60 seconds
 # ============================================================
 
 PHOBOS_DIR="/opt/etc/Phobos"
 CONF="$PHOBOS_DIR/failover.conf"
 STATE="$PHOBOS_DIR/state"
 LOG="$PHOBOS_DIR/health.log"
-WG_CONF="$PHOBOS_DIR/wg0.conf"
 OBF_CONF="$PHOBOS_DIR/wg-obfuscator.conf"
 LOCKFILE="/tmp/phobos-health.lock"
+WG_IF="Wireguard3"
 
 MAX_LOG_LINES=200
-HANDSHAKE_WARN=180
-HANDSHAKE_RESTART_OBF=300
-HANDSHAKE_RESTART_WG=600
-HANDSHAKE_FAILOVER=900
+# Thresholds (seconds)
+HANDSHAKE_WARN=150
+HANDSHAKE_PORT_HOP=300
+HANDSHAKE_SERVER_SWITCH=600
+PRIMARY_CHECK_INTERVAL=300
+# Primary-alive probe: servers block ICMP and busybox `nc -z` is unreliable,
+# so reachability is tested with curl against the panel's obfuscator-health
+# endpoint on the primary VPS. It returns http 200 ONLY when all
+# wg-obfuscator-* services are active — the bare panel port stays up even
+# when the tunnel path is dead, so we must NOT switch back on panel liveness
+# alone (that caused premature switchback to a dead primary).
+PRIMARY_PROBE_SCHEME="${PRIMARY_PROBE_SCHEME:-$(cat "$PHOBOS_DIR/panel_scheme" 2>/dev/null)}"
+PRIMARY_PROBE_SCHEME="${PRIMARY_PROBE_SCHEME:-http}"
+PRIMARY_PROBE_PORT="${PRIMARY_PROBE_PORT:-$(cat "$PHOBOS_DIR/panel_port" 2>/dev/null)}"
+PRIMARY_PROBE_PORT="${PRIMARY_PROBE_PORT:-8443}"
+PRIMARY_PROBE_PATH="/api/obf-health"
+# Connectivity check targets (logged for context only)
+CHECK_HOST_1="8.8.8.8"
+CHECK_HOST_2="1.1.1.1"
 
 log() {
     echo "$(date '+%H:%M:%S') $1" >> "$LOG"
-    # Trim log
     if [ "$(wc -l < "$LOG" 2>/dev/null)" -gt "$MAX_LOG_LINES" ]; then
         tail -n 100 "$LOG" > "$LOG.tmp" && mv "$LOG.tmp" "$LOG"
     fi
@@ -38,33 +51,35 @@ fi
 echo $$ > "$LOCKFILE"
 trap 'rm -f "$LOCKFILE"' EXIT
 
-# Create state dir
 mkdir -p "$STATE"
 
+# ── Detect WG interface ──
+detect_wg_interface() {
+    for i in 0 1 2 3 4 5 6 7 8 9; do
+        desc=$(ndmc -c "show interface Wireguard${i}" 2>/dev/null | grep description | head -1)
+        if echo "$desc" | grep -qi phobos; then
+            WG_IF="Wireguard${i}"
+            return 0
+        fi
+    done
+    return 1
+}
+
 # ── Read failover config ──
 if [ ! -f "$CONF" ]; then
     log "ERROR: no failover.conf"
     exit 1
 fi
 
-# Parse servers: SERVER_1=ip:port1,port2,port3
-# Parse current state
-CURRENT_SERVER=$(cat "$STATE/current_server" 2>/dev/null || echo "1")
-CURRENT_PORT_IDX=$(cat "$STATE/current_port_idx" 2>/dev/null || echo "0")
-RESTART_COUNT=$(cat "$STATE/restart_count" 2>/dev/null || echo "0")
-PRIMARY_CHECK_TS=$(cat "$STATE/primary_check_ts" 2>/dev/null || echo "0")
-
-# Load server list
 SERVER_COUNT=0
 idx=1
 while true; do
     val=$(grep "^SERVER_${idx}=" "$CONF" 2>/dev/null | cut -d= -f2-)
-    if [ -z "$val" ]; then
-        break
-    fi
+    [ -z "$val" ] && break
     eval "SERVER_${idx}_HOST=$(echo "$val" | cut -d: -f1)"
     eval "SERVER_${idx}_PORTS=$(echo "$val" | cut -d: -f2-)"
     eval "SERVER_${idx}_KEY=$(grep "^KEY_${idx}=" "$CONF" 2>/dev/null | cut -d= -f2-)"
+    eval "SERVER_${idx}_WGKEY=$(grep "^WGKEY_${idx}=" "$CONF" 2>/dev/null | cut -d= -f2-)"
     SERVER_COUNT=$idx
     idx=$((idx + 1))
 done
@@ -74,22 +89,43 @@ if [ "$SERVER_COUNT" -eq 0 ]; then
     exit 1
 fi
 
-# ── Get current handshake age ──
-get_handshake_age() {
-    hs=$(wg show wg0 latest-handshakes 2>/dev/null | awk '{print $2}')
-    if [ -z "$hs" ] || [ "$hs" = "0" ]; then
-        echo "9999"
-        return
-    fi
-    now=$(date +%s)
-    age=$((now - hs))
-    echo "$age"
+# Read state
+CURRENT_SERVER=$(cat "$STATE/current_server" 2>/dev/null || echo "1")
+CURRENT_PORT_IDX=$(cat "$STATE/current_port_idx" 2>/dev/null || echo "0")
+FAIL_COUNT=$(cat "$STATE/fail_count" 2>/dev/null || echo "0")
+PRIMARY_CHECK_TS=$(cat "$STATE/primary_check_ts" 2>/dev/null || echo "0")
+
+# ── Detect Phobos WG interface (non-fatal, fallback to Wireguard3) ──
+detect_wg_interface || {
+    log "WARN: detect failed, using default $WG_IF"
 }
 
-# ── Get port by index from comma-separated list ──
+# ── Get handshake age via ndmc ──
+# A peer switch can leave the removed peer's stale handshake listed in ndmc
+# output (often a huge sentinel like 2147483647). Take the FRESHEST (minimum
+# positive, sane) handshake across all peers — that is the live tunnel.
+get_handshake_age() {
+    hs=$(ndmc -c "show interface $WG_IF" 2>/dev/null \
+        | grep "last-handshake" \
+        | awk '{v=$2} v>0 && v<86400 {print v}' \
+        | sort -n | head -1)
+    if [ -z "$hs" ]; then
+        echo "9999"
+    else
+        echo "$hs"
+    fi
+}
+
+# ── Check real connectivity ──
+check_connectivity() {
+    ping -c 1 -W 3 "$CHECK_HOST_1" >/dev/null 2>&1 && return 0
+    ping -c 1 -W 3 "$CHECK_HOST_2" >/dev/null 2>&1 && return 0
+    return 1
+}
+
+# ── Get port by index ──
 get_port() {
-    server_idx=$1
-    port_idx=$2
+    server_idx=$1; port_idx=$2
     eval "ports=\$SERVER_${server_idx}_PORTS"
     echo "$ports" | tr ',' '\n' | sed -n "$((port_idx + 1))p"
 }
@@ -100,6 +136,78 @@ get_port_count() {
     echo "$ports" | tr ',' '\n' | wc -l
 }
 
+# ── Keys currently on the interface (includes the interface's OWN key) ──
+# WG public keys are 43 base64 chars + '='. The interface's own public-key is
+# also matched here — callers must only act on KNOWN server WGKEYs so the
+# local key is never touched.
+list_iface_keys() {
+    ndmc -c "show interface $WG_IF" 2>/dev/null \
+        | grep -oE '[A-Za-z0-9+/]{43}=' | sort -u
+}
+
+# Count how many KNOWN server WGKEYs are currently attached as peers.
+count_server_peers() {
+    present=$(list_iface_keys)
+    n=0
+    i=1
+    while [ "$i" -le "$SERVER_COUNT" ]; do
+        eval "wk=\$SERVER_${i}_WGKEY"
+        if [ -n "$wk" ] && echo "$present" | grep -q "^${wk}$"; then
+            n=$((n + 1))
+        fi
+        i=$((i + 1))
+    done
+    echo "$n"
+}
+
+# ── Switch WG peer so EXACTLY ONE server peer (new_key) remains ──
+# Failover adds a new peer; the RCI "remove" is unreliable on Keenetic and
+# leaves dead peers behind. Multiple peers each with allow-ips 0.0.0.0/0 make
+# egress routing ambiguous. We hard-purge every OTHER known server key via the
+# ndmc CLI (reliable) — never the interface's own key — then ensure new_key.
+switch_wg_peer() {
+    new_key=$1
+
+    if [ -z "$new_key" ]; then
+        log "WARN: no WGKEY for target server, skip peer switch"
+        return 1
+    fi
+
+    present=$(list_iface_keys)
+
+    # Purge any OTHER known server key that is attached
+    i=1
+    while [ "$i" -le "$SERVER_COUNT" ]; do
+        eval "wk=\$SERVER_${i}_WGKEY"
+        if [ -n "$wk" ] && [ "$wk" != "$new_key" ] && echo "$present" | grep -q "^${wk}$"; then
+            log "WG PEER purge: $wk"
+            ndmc -c "interface $WG_IF no wireguard peer $wk" >/dev/null 2>&1
+        fi
+        i=$((i + 1))
+    done
+
+    # Add the target peer if it is not already present
+    if ! echo "$present" | grep -q "^${new_key}$"; then
+        log "WG PEER add: $new_key"
+        # 01.11.2026: порт — из конфига обфускатора (было зашито 13255, а установщик
+        # берёт свободный, на части роутеров 51820 → после переключения туннель мёртв).
+        # Через ndmc по шагам: JSON-RCI с allow-ips "::"/0 молча не ставил allow-ips.
+        lport=$(grep 'source-lport' "$PHOBOS_DIR/wg-obfuscator.conf" 2>/dev/null | tr -d ' ' | cut -d'=' -f2)
+        lport=${lport:-13255}
+        ndmc -c "interface $WG_IF wireguard peer $new_key" >/dev/null 2>&1
+        ndmc -c "interface $WG_IF wireguard peer $new_key endpoint 127.0.0.1:$lport" >/dev/null 2>&1
+        ndmc -c "interface $WG_IF wireguard peer $new_key keepalive-interval 25" >/dev/null 2>&1
+        ndmc -c "interface $WG_IF wireguard peer $new_key allow-ips 0.0.0.0 0.0.0.0" >/dev/null 2>&1
+        # ядро не всегда подхватывает allow-ips нового пира без перезапуска интерфейса
+        ndmc -c "interface $WG_IF down" >/dev/null 2>&1; sleep 1
+        ndmc -c "interface $WG_IF up" >/dev/null 2>&1
+    fi
+
+    # Persist config
+    ndmc -c "system configuration save" >/dev/null 2>&1
+    return 0
+}
+
 # ── Switch to specific server:port ──
 switch_endpoint() {
     server_idx=$1
@@ -107,6 +215,7 @@ switch_endpoint() {
 
     eval "host=\$SERVER_${server_idx}_HOST"
     eval "obf_key=\$SERVER_${server_idx}_KEY"
+    eval "wg_key=\$SERVER_${server_idx}_WGKEY"
     port=$(get_port "$server_idx" "$port_idx")
 
     if [ -z "$host" ] || [ -z "$port" ]; then
@@ -116,7 +225,10 @@ switch_endpoint() {
 
     log "SWITCH → server $server_idx ($host:$port)"
 
-    # Update obfuscator config target
+    # 1. Switch WG peer key if different server
+    switch_wg_peer "$wg_key"
+
+    # 2. Update obfuscator config
     if [ -f "$OBF_CONF" ]; then
         sed -i "s|^target = .*|target = ${host}:${port}|" "$OBF_CONF"
         if [ -n "$obf_key" ]; then
@@ -124,25 +236,26 @@ switch_endpoint() {
         fi
     fi
 
-    # Restart obfuscator
+    # 3. Restart obfuscator
+    if [ -f /opt/etc/init.d/S49wg-obfuscator ]; then
+        /opt/etc/init.d/S49wg-obfuscator restart >/dev/null 2>&1
+    else
         killall wg-obfuscator 2>/dev/null
         sleep 1
         wg-obfuscator --config "$OBF_CONF" &
+    fi
 
-    # Save state
+    # 4. Save state (fail_count is managed by the caller, NOT reset here —
+    #    resetting on a port-hop would prevent escalation to server failover)
     echo "$server_idx" > "$STATE/current_server"
     echo "$port_idx" > "$STATE/current_port_idx"
-    echo "0" > "$STATE/restart_count"
 }
 
 # ── Try next port on current server ──
 try_next_port() {
     port_count=$(get_port_count "$CURRENT_SERVER")
     next_idx=$(( (CURRENT_PORT_IDX + 1) % port_count ))
-
-    if [ "$next_idx" -eq 0 ]; then
-        return 1
-    fi
+    [ "$next_idx" -eq 0 ] && return 1
 
     log "PORT HOP → port idx $next_idx on server $CURRENT_SERVER"
     switch_endpoint "$CURRENT_SERVER" "$next_idx"
@@ -152,96 +265,203 @@ try_next_port() {
 # ── Try next server ──
 try_next_server() {
     next=$((CURRENT_SERVER + 1))
-    if [ "$next" -gt "$SERVER_COUNT" ]; then
-        next=1
-    fi
-
-    if [ "$next" -eq "$CURRENT_SERVER" ]; then
-        log "WARN: only one server, can't failover"
-        return 1
-    fi
+    [ "$next" -gt "$SERVER_COUNT" ] && next=1
+    [ "$next" -eq "$CURRENT_SERVER" ] && return 1
 
     log "FAILOVER → server $next"
     switch_endpoint "$next" "0"
     return 0
 }
 
-# ── Check if primary is back (every 5 min when on secondary) ──
+# ── Check if primary is back ──
 check_primary() {
-    if [ "$CURRENT_SERVER" -eq 1 ]; then
-        return
-    fi
+    [ "$CURRENT_SERVER" -eq 1 ] && return
 
     now=$(date +%s)
     elapsed=$((now - PRIMARY_CHECK_TS))
-    if [ "$elapsed" -lt 300 ]; then
-        return
-    fi
+    [ "$elapsed" -lt "$PRIMARY_CHECK_INTERVAL" ] && return
     echo "$now" > "$STATE/primary_check_ts"
 
     eval "host=\$SERVER_1_HOST"
-    port=$(get_port 1 0)
-
-    # Quick UDP probe — send a byte, see if port responds
-    if ping -c 1 -W 2 "$host" >/dev/null 2>&1; then
-        log "PRIMARY alive, switching back"
+    # Probe the obfuscator-health endpoint: http 200 ONLY when the primary's
+    # obfuscator path is actually up. Anything else (000 no-response, 503
+    # obf-down, redirects) means the tunnel path is NOT viable → stay put.
+    ok_url=""
+    last_code=""
+    for url in "${PRIMARY_PROBE_SCHEME}://${host}:${PRIMARY_PROBE_PORT}${PRIMARY_PROBE_PATH}" "http://${host}:8444${PRIMARY_PROBE_PATH}"; do
+        code=$(curl -k -s -m 4 -o /dev/null -w '%{http_code}' "$url" 2>/dev/null)
+        last_code="$code"
+        if [ "$code" = "200" ]; then
+            ok_url="$url"
+            break
+        fi
+    done
+    if [ -n "$ok_url" ]; then
+        log "PRIMARY ($ok_url http=200) alive, switching back"
         switch_endpoint 1 0
+        echo "0" > "$STATE/fail_count"
+    else
+        log "PRIMARY still down (obf-health http=${last_code:-none}), staying on server $CURRENT_SERVER"
     fi
 }
 
-# ── Main logic ──
-AGE=$(get_handshake_age)
+# ── LAN routing self-heal ──────────────────────
+# A router REBOOT silently drops the per-device `ip hotspot host <mac> policy`
+# binding and can reset the WG interface security-level, so LAN clients leak to
+# WAN (no VPN). This:
+#   1) ensures WG_IF security-level = public (needed for masquerade),
+#   2) LEARNS current host->policy bindings into state/lan-hosts (append-only),
+#   3) RE-APPLIES any learned binding that is currently missing.
+# Append-only learning means a reboot (which clears live bindings) never erases
+# the record, so the next run restores them. ndmc show running-config is heavy,
+# so the caller gates this to run only every few minutes.
+heal_lan_routing() {
+    LANHOSTS="$STATE/lan-hosts"
+    rc=$(ndmc -c "show running-config" 2>/dev/null)
+    [ -z "$rc" ] && return 0
 
+    # 1) WG interface must be public
+    sl=$(echo "$rc" | awk -v ifc="interface $WG_IF" '$0~ifc{f=1} f&&/security-level/{print $2; exit}')
+    if [ -n "$sl" ] && [ "$sl" != "public" ]; then
+        ndmc -c "interface $WG_IF security-level public" >/dev/null 2>&1
+        ndmc -c "system configuration save" >/dev/null 2>&1
+        log "HEAL: $WG_IF security-level -> public"
+    fi
+
+    # 2) learn live bindings (append-only) into lan-hosts
+    touch "$LANHOSTS"
+    echo "$rc" | grep -oE "host [0-9a-f:]+ policy [A-Za-z0-9_]+" | while read -r _h mac _p pol; do
+        grep -q "^$mac " "$LANHOSTS" 2>/dev/null || echo "$mac $pol" >> "$LANHOSTS"
+    done
+
+    # 3) re-apply any learned binding that is missing live
+    changed=0
+    while read -r mac pol; do
+        [ -z "$mac" ] && continue
+        case "$mac" in \#*) continue;; esac
+        if ! echo "$rc" | grep -q "host $mac policy $pol"; then
+            ndmc -c "ip hotspot host $mac policy $pol" >/dev/null 2>&1
+            log "HEAL: re-bound host $mac -> $pol"
+            changed=1
+        fi
+    done < "$LANHOSTS"
+    [ "$changed" = 1 ] && ndmc -c "system configuration save" >/dev/null 2>&1
+}
+
+# ══════════════════════════════════════════════
+#  Explicit apply hook (used by phobos-pull.sh after a config change).
+#  `phobos-health.sh apply-server N` re-points the tunnel to SERVER_N
+#  immediately, skipping the failure-escalation logic. Reuses
+#  switch_endpoint so the WG-peer/obfuscator/state changes stay identical
+#  to a normal failover. Resets fail_count so the new server starts clean.
+# ══════════════════════════════════════════════
+if [ "$1" = "apply-server" ]; then
+    idx="${2:-1}"
+    eval "ahost=\$SERVER_${idx}_HOST"
+    if [ -z "$ahost" ]; then
+        log "APPLY: server $idx not in conf, ignore"
+        exit 1
+    fi
+    log "APPLY: pull requested server $idx ($ahost)"
+    switch_endpoint "$idx" "0"
+    echo "0" > "$STATE/fail_count"
+    echo "$idx" > "$STATE/current_server"
+    exit 0
+fi
+
+# ══════════════════════════════════════════════
+#  Pre-check: fix desync (obfuscator targeting wrong server)
+# ══════════════════════════════════════════════
+if [ -f "$OBF_CONF" ]; then
+    cur_target=$(grep "^target = " "$OBF_CONF" 2>/dev/null | sed 's/target = //')
+    eval "expected_host=\$SERVER_${CURRENT_SERVER}_HOST"
+    if [ -n "$cur_target" ] && [ -n "$expected_host" ]; then
+        echo "$cur_target" | grep -q "$expected_host" || {
+            log "DESYNC: obf=$cur_target state=server${CURRENT_SERVER}($expected_host). Resync."
+            switch_endpoint "$CURRENT_SERVER" "$CURRENT_PORT_IDX"
+            sleep 5
+        }
+    fi
+fi
+
+# ── LAN routing self-heal (gated ~5 min; a reboot drops host->policy bindings) ──
+HEAL_TS=$(cat "$STATE/heal_ts" 2>/dev/null || echo 0)
+now_heal=$(date +%s)
+if [ $((now_heal - HEAL_TS)) -ge 300 ]; then
+    echo "$now_heal" > "$STATE/heal_ts"
+    heal_lan_routing
+fi
+
+# ── Peer hygiene: keep exactly ONE WG peer = current server's key ──
+# Failovers can leave dead/duplicate peers; multiple 0.0.0.0/0 peers cause
+# ambiguous egress routing. Self-heal here every run (cheap when already clean).
+eval "cur_wgkey=\$SERVER_${CURRENT_SERVER}_WGKEY"
+if [ -n "$cur_wgkey" ]; then
+    peer_n=$(count_server_peers)
+    if [ "$peer_n" -gt 1 ]; then
+        log "HYGIENE: $peer_n server peers present, purging to server${CURRENT_SERVER}"
+        switch_wg_peer "$cur_wgkey"
+    fi
+fi
+
+# ══════════════════════════════════════════════
+#  Main logic
+#  Tunnel health = WireGuard handshake age (authoritative: a fresh
+#  handshake only happens through the full obfuscator→server→WG path).
+#  WAN ping is logged for context ONLY — it does NOT gate decisions,
+#  because the router's default route is not the tunnel, so WAN can be
+#  up while the tunnel is dead (and a WAN blip must not cause failover).
+# ══════════════════════════════════════════════
+AGE=$(get_handshake_age)
+WAN=$(check_connectivity && echo "yes" || echo "no")
+
+# Tunnel healthy = fresh handshake
 if [ "$AGE" -lt "$HANDSHAKE_WARN" ]; then
-    # All good, reset counters
-    if [ "$RESTART_COUNT" -gt 0 ]; then
-        log "OK: connection restored (age=${AGE}s)"
-        echo "0" > "$STATE/restart_count"
+    if [ "$FAIL_COUNT" -gt 0 ]; then
+        log "OK: tunnel up (handshake=${AGE}s, server=$CURRENT_SERVER, wan=$WAN)"
+        echo "0" > "$STATE/fail_count"
     fi
     check_primary
     exit 0
 fi
 
-log "STALE handshake: ${AGE}s (server=$CURRENT_SERVER port_idx=$CURRENT_PORT_IDX restarts=$RESTART_COUNT)"
+# Stale handshake = tunnel down → escalate.
+# fail_count is MONOTONIC: it climbs across stages and is only reset by the
+# OK branch (real recovery) or after a successful failover to a NEW server
+# (so the new server gets a fresh restart→port-hop→failover cycle).
+FAIL_COUNT=$((FAIL_COUNT + 1))
+echo "$FAIL_COUNT" > "$STATE/fail_count"
+log "STALE: handshake=${AGE}s wan=${WAN} server=$CURRENT_SERVER port=$CURRENT_PORT_IDX fails=$FAIL_COUNT"
 
-if [ "$AGE" -lt "$HANDSHAKE_RESTART_OBF" ]; then
-    # Restart obfuscator only
-    RESTART_COUNT=$((RESTART_COUNT + 1))
-    echo "$RESTART_COUNT" > "$STATE/restart_count"
-    log "ACTION: restart obfuscator (attempt $RESTART_COUNT)"
+# Stage 1 (fail 1): Restart obfuscator
+if [ "$FAIL_COUNT" -eq 1 ]; then
+    log "ACTION: restart obfuscator"
+    if [ -f /opt/etc/init.d/S49wg-obfuscator ]; then
+        /opt/etc/init.d/S49wg-obfuscator restart >/dev/null 2>&1
+    else
         killall wg-obfuscator 2>/dev/null
         sleep 1
         wg-obfuscator --config "$OBF_CONF" &
-    exit 0
-fi
-
-if [ "$AGE" -lt "$HANDSHAKE_RESTART_WG" ]; then
-    # Restart WireGuard + obfuscator
-    RESTART_COUNT=$((RESTART_COUNT + 1))
-    echo "$RESTART_COUNT" > "$STATE/restart_count"
-    log "ACTION: restart WG + obfuscator (attempt $RESTART_COUNT)"
-    wg-quick down wg0 2>/dev/null
-    killall wg-obfuscator 2>/dev/null
-    sleep 2
-    wg-quick up wg0 2>/dev/null
-    sleep 1
-    wg-obfuscator --config "$OBF_CONF" &
-    exit 0
-fi
-
-if [ "$AGE" -lt "$HANDSHAKE_FAILOVER" ]; then
-    # Try next port
-    log "ACTION: try port hop"
-    if ! try_next_port; then
-        log "ACTION: all ports exhausted, try next server"
-        try_next_server
     fi
     exit 0
 fi
 
-# Full failover
-log "ACTION: failover (handshake ${AGE}s)"
-if ! try_next_server; then
-    # Only one server, cycle ports
-    try_next_port
+# Stage 2 (fail 2): Port hop to an alternate port on the SAME server
+if [ "$FAIL_COUNT" -eq 2 ]; then
+    if try_next_port; then
+        exit 0
+    fi
+    # only one port → fall through to server failover
+    log "single port, escalate to failover"
+fi
+
+# Stage 3 (fail 3+): Server failover. Reset fail_count so the new server
+# gets its own restart→port-hop→failover cycle next ticks.
+log "ACTION: server failover (fail $FAIL_COUNT)"
+if try_next_server; then
+    echo "0" > "$STATE/fail_count"
+else
+    # Only one server — cycle back to port 0 and restart escalation
+    echo "0" > "$STATE/fail_count"
+    switch_endpoint "$CURRENT_SERVER" "0"
 fi

overlay/install-router.sh.template

@@ -456,27 +456,20 @@ deploy_health_monitor() {
   [ -f "$PHOBOS_DIR/phobos-pull.sh" ] || pull_line=""
 
   if [ "$ROUTER_PLATFORM" = "keenetic" ]; then
-    # Install cron if not present (Entware)
-    if ! command -v crontab >/dev/null 2>&1 && command -v opkg >/dev/null 2>&1; then
-      log "  Установка cron через opkg..."
-      opkg update >/dev/null 2>&1
-      opkg install cron >/dev/null 2>&1
-      /opt/etc/init.d/S10cron start >/dev/null 2>&1 || true
-    fi
-    local cron_file="/opt/etc/crontab"
-    # Create crontab file if missing
-    if [ ! -f "$cron_file" ]; then
-      mkdir -p /opt/etc
-      touch "$cron_file"
-    fi
-    if ! grep -q "phobos-health" "$cron_file" 2>/dev/null; then
-      echo "$cron_line" >> "$cron_file"
-      log "  Cron (health) добавлен в $cron_file"
-    fi
-    if [ -n "$pull_line" ] && ! grep -q "phobos-pull" "$cron_file" 2>/dev/null; then
-      echo "$pull_line" >> "$cron_file"
-      log "  Cron (pull) добавлен в $cron_file"
+    # Демон — busybox crond (есть в Entware как /opt/sbin/crond). Entware-пакет cron
+    # (vixie, PROCS=cron) на Keenetic задачи не выполнял — заменяем (01.11.2026)
+    local I=/opt/etc/init.d/S10cron
+    if [ ! -x $I ] || grep -q "^PROCS=cron$" $I; then
+      $I stop >/dev/null 2>&1
+      printf '#!/bin/sh\n\nENABLED=yes\nPROCS=crond\nARGS="-c /opt/var/spool/cron/crontabs -b"\nPREARGS=""\nDESC=$PROCS\nPATH=/opt/sbin:/opt/bin:/usr/sbin:/usr/bin:/sbin:/bin\n\n. /opt/etc/init.d/rc.func\n' > $I
+      chmod +x $I
     fi
+    mkdir -p /opt/var/spool/cron/crontabs
+    # 01.11.2026: демоны cron на Keenetic (busybox crond, Entware cron) читают crontab
+    # пользователя root, а не /opt/etc/crontab — туда писать бесполезно
+    sed -i "/phobos-health\|phobos-pull/d" /opt/etc/crontab 2>/dev/null
+    (crontab -l 2>/dev/null | grep -v "phobos-health" | grep -v "phobos-pull"; echo "$cron_line"; [ -n "$pull_line" ] && echo "$pull_line") | crontab -
+    log "  Cron добавлен через crontab (health + pull)"
     /opt/etc/init.d/S10cron restart >/dev/null 2>&1 || true
   elif [ "$ROUTER_PLATFORM" = "openwrt" ]; then
     (crontab -l 2>/dev/null | grep -v "phobos-health" | grep -v "phobos-pull"; echo "$cron_line"; [ -n "$pull_line" ] && echo "$pull_line") | crontab -

overlay/phobos-client.sh

@@ -220,7 +220,7 @@ action_package() {
   cp "$dir/wg-obfuscator.conf" "$pkg_root/wg-obfuscator.conf"
 
   # Fetch missing router-arch binaries from ClusterM on-demand
-  local _tag; _tag=$(curl -fsSL -m10 https://api.github.com/repos/ClusterM/wg-obfuscator/releases/latest 2>/dev/null | grep '"tag_name"' | head -1 | sed 's/.*"tag_name": *"\(.*\)".*/\1/')
+  local _tag; _tag=$(curl -fsSL -m10 https://api.github.com/repos/ClusterM/wg-obfuscator/releases/latest 2>/dev/null | grep '"tag_name"' | head -1 | sed 's/.*"tag_name": *"\(.*\)".*/\1/') || _tag=""  # ponytail: GitHub API rate limit/недоступность не должны ронять сборку пакета — берём уже скачанные бинарники
   local _cb="https://github.com/ClusterM/wg-obfuscator/releases/download/${_tag}"
   declare -A _am=([aarch64]="linux-arm64" [mipsel]="linux-mipsel-mips32" [mips]="linux-mips-mips32" [armv7]="linux-armv7-hf" [x86_64]="linux-x64")
   for _arch in aarch64 mipsel mips armv7 x86_64; do

overlay/phobos-pull.sh

@@ -71,7 +71,7 @@ fetch_config() {
     if command -v curl >/dev/null; then
         curl -k -s -m 8 -o "$TMP" "$url" 2>/dev/null && return 0
     fi
-    wget --no-check-certificate -q -O "$TMP" "$url" 2>/dev/null || wget -q -O "$TMP" "$url" 2>/dev/null
+    wget -T 8 --no-check-certificate -q -O "$TMP" "$url" 2>/dev/null || wget -T 8 -q -O "$TMP" "$url" 2>/dev/null
 }
 
 # One fetch + compare + apply pass. Returns 0 always (best-effort).

server/phobos-health.sh

@@ -189,9 +189,18 @@ switch_wg_peer() {
     # Add the target peer if it is not already present
     if ! echo "$present" | grep -q "^${new_key}$"; then
         log "WG PEER add: $new_key"
-        curl -s -X POST "http://localhost:79/rci/" \
-            -H "Content-Type: application/json" \
-            -d "{\"interface\":{\"${WG_IF}\":{\"wireguard\":{\"peer\":{\"key\":\"${new_key}\",\"comment\":\"Phobos VPS Server\",\"endpoint\":{\"address\":\"127.0.0.1:13255\"},\"keepalive-interval\":{\"interval\":25},\"allow-ips\":[{\"address\":\"0.0.0.0\",\"mask\":\"0.0.0.0\"},{\"address\":\"::\",\"mask\":\"0\"}]}}}}}" >/dev/null 2>&1
+        # 01.11.2026: порт — из конфига обфускатора (было зашито 13255, а установщик
+        # берёт свободный, на части роутеров 51820 → после переключения туннель мёртв).
+        # Через ndmc по шагам: JSON-RCI с allow-ips "::"/0 молча не ставил allow-ips.
+        lport=$(grep 'source-lport' "$PHOBOS_DIR/wg-obfuscator.conf" 2>/dev/null | tr -d ' ' | cut -d'=' -f2)
+        lport=${lport:-13255}
+        ndmc -c "interface $WG_IF wireguard peer $new_key" >/dev/null 2>&1
+        ndmc -c "interface $WG_IF wireguard peer $new_key endpoint 127.0.0.1:$lport" >/dev/null 2>&1
+        ndmc -c "interface $WG_IF wireguard peer $new_key keepalive-interval 25" >/dev/null 2>&1
+        ndmc -c "interface $WG_IF wireguard peer $new_key allow-ips 0.0.0.0 0.0.0.0" >/dev/null 2>&1
+        # ядро не всегда подхватывает allow-ips нового пира без перезапуска интерфейса
+        ndmc -c "interface $WG_IF down" >/dev/null 2>&1; sleep 1
+        ndmc -c "interface $WG_IF up" >/dev/null 2>&1
     fi
 
     # Persist config

server/phobos-pull.sh

@@ -71,7 +71,7 @@ fetch_config() {
     if command -v curl >/dev/null; then
         curl -k -s -m 8 -o "$TMP" "$url" 2>/dev/null && return 0
     fi
-    wget --no-check-certificate -q -O "$TMP" "$url" 2>/dev/null || wget -q -O "$TMP" "$url" 2>/dev/null
+    wget -T 8 --no-check-certificate -q -O "$TMP" "$url" 2>/dev/null || wget -T 8 -q -O "$TMP" "$url" 2>/dev/null
 }
 
 # One fetch + compare + apply pass. Returns 0 always (best-effort).