Fixes and new features from a couple months of daily driving (v2.7)

CHANGELOG.md

@@ -1,22 +1,22 @@
-1. Fix multiq total_bands detection.
-2. Update service.sh timings to prevent some race conditions. [OnePlus devices]
-3. Create post-fs-data.d directory if it doesn't exist.
-4. Add support for local tc binary which supports all qdics.
-5. Fix qdisc reset in following cases:
-   - Toggle Airplane Mode.
-   - Toggle WiFi / Cellular.
-6. Add qdisc monitor to prevent netd from changing qdisc.
-7. Simplify fq_codel leaf queue for htb.
-8. Add bbr3 detection in post-fs stage.
-9. Enable full screen mode to prevent navigation bar from interfering.
-10. Add pfifo_head_drop, pie, fq_pie and cake full support.
-11. Optimize sysctl parameters for low latency, high throughput and stability.
-12. Add newly introduced PLB sysctl param support for optimisations.
-13. Reduce initrwnd max value to 20 based on RFC 6928.
-14. Set tcp_mem dynamically.
-
-Note:
-1. More qdisc tuning will be added in next version when I get more information from many users.
-2. qdisc fine-tuning currently doesn't support custom mode. Might be added in future.
-3. If your kernel supports bbr3, then please use fq / cake / fq_pie for best results.
-4. Some kernels convert bbrv1 to bbrv3 and keep the name as bbr. Even in that case use fq / cake / fq_pie instead of other qdisc.
+1. New Tools tab: Auto-Optimize (benchmarks every available congestion control, not a fixed whitelist), CC Duel A/B test, live BBR telemetry graph and a multi-server bufferbloat test.
+2. New Results tab: every bufferbloat run is saved (connection type, cc, qdisc, signal, download, idle/loaded ping, grade) and survives upgrades.
+3. Optional "Advanced TCP buffers" toggle: raise-only (never lowers ROM-tuned min/default values), bumps rmem/wmem ceilings to 16 MiB.
+4. initcwnd/initrwnd and per-route congctl are now also applied to the per-interface route table — Android keeps the default route there, so the old main-table-only write never touched internet traffic.
+5. Watchdog extended to a full drift guard: global cc, per-route congctl/initcwnd and buffer ceilings are guarded, not just the qdisc; re-asserts quietly (no ss -K) and logs what drifted. Root-anchored qdisc match — a foreign root with a matching leaf no longer passes.
+6. Drift guard pauses while WebUI tests run, so Auto-Optimize / CC Duel probes are no longer reverted mid-measurement (pause is crash-safe, capped at 10 minutes).
+7. Boot sysctl block made raise-only — no longer clobbers ROM-tuned tcp_rmem/tcp_wmem min/default; tcp_mtu_probing enabled.
+8. Fix qdisc matching to use word boundaries (fq no longer false-matches fq_codel in apply checks and the watchdog).
+9. Fix per-route congctl bbr/bbr3 prefix collision that left stale bbr3 pins on routes after switching to bbr.
+10. Fix VoWiFi detection on OxygenOS/ColorOS (OPlus SystemUI pattern + AOSP fallback; previously every Wi-Fi join ate the full 20 s timeout).
+11. Fix cellular signal reading: real dBm (RSRP/ssRsrp) from the active-data SIM's primary RAT instead of the dead CDMA level= that pinned dual-SIM devices at "poor".
+12. Fix Manual Settings not applying on Wi-Fi (force-apply was gated behind the VoWiFi connect timer).
+13. VPN-aware interface resolution: tun/ppp/wg map to the underlying wlan/rmnet interface.
+14. Upgrades preserve all user state: markers, Advanced-buffers toggle, cake hints, active profile and test history.
+15. Mount debugfs when absent (Android 11+) so bbr3 socket telemetry works.
+16. Drop inert system.prop (set Android props nothing consumes; could contradict the live sysctls).
+17. Fix Logs tab re-rendering every poll (visible flicker / scroll jump).
+18. WebUI polish: themed dropdowns, animated nav pill, per-series lanes + 60 fps glide on the telemetry graph.
+
+Note:
+1. If your kernel supports bbr3, then please use fq / cake / fq_pie for best results.
+2. Some kernels convert bbrv1 to bbrv3 and keep the name as bbr. Even in that case use fq / cake / fq_pie instead of other qdisc.

module/customize.sh

@@ -122,6 +122,18 @@ if check_exists_anywhere "initcwnd"; then
     fi
 fi
 
+# Preserve the remaining per-user state across upgrades. The wlan_*/rmnet_data_*/
+# kill_connections/initcwnd_initrwnd markers are carried over above, but the
+# newer 2.7 state files were not, so every upgrade silently reset the Advanced
+# TCP buffers toggle, the Smart-cake hints, the active profile label and wiped
+# the whole Results-tab test history.
+for m in adv_buffers cake_hint_wlan cake_hint_rmnet active_profile bench_results.jsonl; do
+    if [ -f "$MODULE_PATH/$m" ] && [ ! -f "$MODPATH/$m" ]; then
+        cp "$MODULE_PATH/$m" "$MODPATH/$m"
+        ui_print " [+] Preserved user state: $m"
+    fi
+done
+
 cp "$MODPATH/module.prop" "$MODPATH/module.prop.bak"
 
 chmod +x "$MODPATH/bin/tc"

module/module.prop

@@ -1,7 +1,7 @@
 id=tcp_optimiser
 name=TCP Optimiser 
-version=2.5
-versionCode=17
+version=2.7
+versionCode=72
 author=fatalcoder524
 description=TCP Optimisations & update tcp_cong_algo based on interface
 updateJson=https://raw.githubusercontent.com/fatalcoder524/TCP_Optimiser_Module/main/update.json

module/post-fs-data.sh

@@ -2,6 +2,11 @@
 
 sleep 2
 
+# Mount debugfs (Android 11+ user builds leave it unmounted) so the WebUI
+# telemetry card can read /sys/kernel/debug/tcp_bbr3/sockets. Idempotent.
+mountpoint -q /sys/kernel/debug 2>/dev/null || \
+	mount -t debugfs debugfs /sys/kernel/debug 2>/dev/null
+
 AVAIL_CC="$(cat /proc/sys/net/ipv4/tcp_available_congestion_control)"
 # Check if BBR3 is available
 if echo "$AVAIL_CC" | grep -qw bbr3; then
@@ -26,18 +31,31 @@ echo "fq_codel" > /proc/sys/net/core/default_qdisc 2>/dev/null
 echo 120 > /proc/sys/net/ipv4/tcp_pacing_ca_ratio 2>/dev/null
 echo 180 > /proc/sys/net/ipv4/tcp_pacing_ss_ratio 2>/dev/null
 echo 1 > /proc/sys/net/ipv4/tcp_window_scaling 2>/dev/null
-echo "4096 2097152 16777216" > /proc/sys/net/ipv4/tcp_rmem 2>/dev/null
-echo "4096 2097152 16777216" > /proc/sys/net/ipv4/tcp_wmem 2>/dev/null
-echo 16777216 > /proc/sys/net/core/rmem_max 2>/dev/null
-echo 16777216 > /proc/sys/net/core/wmem_max 2>/dev/null
+
+# Socket buffers: RAISE ONLY. The old code wrote a fixed "4096 2097152 16777216"
+# over tcp_rmem/tcp_wmem, which replaces the ROM's tuned min & default (OxygenOS
+# ships 524288 1048576 16777216) with a 2 MiB default on every socket. That is
+# exactly what apply_tcp_buffers() in service.sh was written to avoid, so use the
+# same raise-only rule here and let the "Advanced TCP buffers" toggle stay
+# meaningful instead of being pre-empted by this block.
+for f in /proc/sys/net/core/rmem_max /proc/sys/net/core/wmem_max; do
+	cur=$(cat "$f" 2>/dev/null)
+	[ -n "$cur" ] && [ "$cur" -lt 16777216 ] && echo 16777216 > "$f" 2>/dev/null
+done
+for f in /proc/sys/net/ipv4/tcp_rmem /proc/sys/net/ipv4/tcp_wmem \
+         /proc/sys/net/ipv6/tcp_rmem /proc/sys/net/ipv6/tcp_wmem; do
+	set -- $(cat "$f" 2>/dev/null)
+	[ -n "$3" ] && [ "$3" -lt 16777216 ] && echo "$1 $2 16777216" > "$f" 2>/dev/null
+done
+
 echo 4096 > /proc/sys/net/ipv4/tcp_max_syn_backlog 2>/dev/null
-echo 0 > /proc/sys/net/ipv4/tcp_mtu_probing 2>/dev/null
+# PMTU probing: OxygenOS already ships 1; writing 0 here downgraded the ROM on
+# every boot (and fought apply_tcp_buffers, which sets 1).
+echo 1 > /proc/sys/net/ipv4/tcp_mtu_probing 2>/dev/null
 echo 16384 > /proc/sys/net/ipv4/tcp_notsent_lowat 2>/dev/null
 
 # IPv6 TCP tuning
 echo 1 > /proc/sys/net/ipv6/tcp_ecn 2>/dev/null
-echo "4096 2097152 16777216" > /proc/sys/net/ipv6/tcp_rmem 2>/dev/null
-echo "4096 2097152 16777216" > /proc/sys/net/ipv6/tcp_wmem 2>/dev/null
 
 # Extra settings for optimal stability
 echo 100000 > /proc/sys/net/core/netdev_max_backlog 2>/dev/null