[Security] Stop USB device names from being executed as Hyprland Lua (backport of #8129)

Backport of the input-device name fix (PR #8129 by @acrogenesis, merged to
quattro as 9285b19d) onto the v4-0-1 release branch.

Hyprland input-device and monitor names come from USB descriptors and hyprctl
output, so they are attacker-influenceable, yet the toggle and monitor commands
interpolated them straight into hyprctl eval and into generated Lua that
Hyprland re-executes on every reload. XF86TouchpadToggle is bound with
locked = true, so a malicious USB name reached Lua execution from the lock
screen as the logged-in user, and a persisted disable made it run on every
start. Publicly reported by Jorrit Jongma / Chainfire.

The disable is no longer executable Lua anywhere. The device name is stored as
plain-text data in a *-disabled-name sidecar and read back by a packaged module,
default/hypr/disabled-input-device.lua, on every reload; the live hyprctl eval
Lua-quotes the name and rejects control characters outright. The reload loader
excludes the two legacy filenames, so a leftover generated *-disabled.lua on a
not-yet-migrated install can never be sourced as code again, and a migration
recovers the device name from it and deletes it, sanitizing installs that ran
the vulnerable version. All four monitor scripts validate an output name against
a plain-connector-name pattern before writing it as Lua, closing the same latent
pattern in the siblings, and paths.lua treats a set-but-empty XDG_STATE_HOME as
unset to match the bash side.

Clean cherry-pick: all fourteen files are byte-identical to quattro, so merging
v4-0-1 into quattro resolves without a conflict. This branch ships no leftover
*-disabled.lua template of its own -- the tracked "disabled" files are the same
two quattro has -- so the migration is the only path that has to sanitize
anything here.

test/shell passes: 192 files, including the three this adds. The toggle suite's
public-PoC case passes here, as do the monitor scripts' accept/reject cases and
the XDG path cases. test/cli passes.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018DEMYa9UWtroz93DhMTtcV

test/shell.d/toggle-input-device-test.sh

@@ -0,0 +1,286 @@
+#!/bin/bash
+
+set -euo pipefail
+
+source "$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)/base-test.sh"
+
+require_command lua
+
+tmpdir=$(mktemp -d)
+trap 'rm -rf "$tmpdir"' EXIT
+
+stub_dir="$tmpdir/bin"
+home_dir="$tmpdir/home"
+xdg_decoy="$tmpdir/xdg-decoy"
+log_file="$tmpdir/hyprctl.log"
+marker="$tmpdir/marker"
+mkdir -p "$stub_dir" "$home_dir" "$xdg_decoy"
+
+state_dir="$home_dir/.local/state/omarchy/toggles/hypr"
+name_file="$state_dir/touchpad-disabled-name"
+state_lua="$state_dir/touchpad-disabled.lua"
+
+cat >"$stub_dir/hyprctl" <<'EOF'
+#!/bin/bash
+case $1 in
+  eval) printf '%s\n' "$2" >>"$HYPRCTL_LOG" ;;
+  reload) printf 'reload\n' >>"$HYPRCTL_LOG" ;;
+esac
+EOF
+chmod +x "$stub_dir/hyprctl"
+
+cat >"$stub_dir/omarchy-osd" <<'EOF'
+#!/bin/bash
+:
+EOF
+chmod +x "$stub_dir/omarchy-osd"
+
+stub_device() {
+  local kind=$1
+  local name=$2
+  cat >"$stub_dir/omarchy-hw-$kind" <<EOF
+#!/bin/bash
+printf '%s\n' '$name'
+EOF
+  chmod +x "$stub_dir/omarchy-hw-$kind"
+}
+
+# XDG_STATE_HOME deliberately points away from HOME everywhere below: the
+# input-device state is hardcoded to ~/.local/state like the sibling toggle
+# tools and the pre-migration script, so nothing may read or write the XDG
+# directory.
+run_toggle() {
+  HOME="$home_dir" \
+    XDG_STATE_HOME="$xdg_decoy" \
+    HYPRCTL_LOG="$log_file" \
+    PATH="$stub_dir:$ROOT/bin:$PATH" \
+    "$ROOT/bin/omarchy-toggle-input-device" "$@"
+}
+
+assert_decoy_untouched() {
+  [[ -z $(find "$xdg_decoy" -mindepth 1 -print -quit 2>/dev/null) ]] ||
+    fail "input-device state must ignore XDG_STATE_HOME"
+}
+
+: >"$log_file"
+stub_device touchpad 'elan-touchpad'
+
+run_toggle touchpad off
+[[ $(<"$name_file") == "elan-touchpad" ]] || fail "touchpad disable stores the device name as data"
+[[ ! -e $state_lua ]] || fail "touchpad disable writes no generated Lua"
+grep -Fx 'hl.device({ name = "elan-touchpad", enabled = false })' "$log_file" >/dev/null ||
+  fail "touchpad disable applies a quoted Lua device name"
+assert_decoy_untouched
+pass "touchpad disable persists the device name as data"
+
+: >"$log_file"
+run_toggle touchpad on
+[[ ! -e $name_file ]] || fail "touchpad enable clears the persisted device name"
+grep -Fx 'hl.device({ name = "elan-touchpad", enabled = true })' "$log_file" >/dev/null ||
+  fail "touchpad enable applies a quoted Lua device name"
+pass "touchpad enable clears persisted disable state"
+
+run_toggle touchpad
+[[ -f $name_file ]] || fail "default toggle action disables an enabled touchpad"
+run_toggle touchpad
+[[ ! -e $name_file ]] || fail "default toggle action enables a disabled touchpad"
+pass "default toggle action flips the persisted state"
+
+: >"$log_file"
+stub_device touchscreen 'wacom-hid-52eb-finger'
+ts_name_file="$state_dir/touchscreen-disabled-name"
+
+run_toggle touchscreen off
+[[ $(<"$ts_name_file") == "wacom-hid-52eb-finger" ]] ||
+  fail "touchscreen disable stores the device name as data"
+grep -Fx 'hl.device({ name = "wacom-hid-52eb-finger", enabled = false })' "$log_file" >/dev/null ||
+  fail "touchscreen disable applies a quoted Lua device name"
+run_toggle touchscreen on
+[[ ! -e $ts_name_file ]] || fail "touchscreen enable clears the persisted device name"
+pass "touchscreen routes through the same persisted-name state"
+
+: >"$log_file"
+rm -f "$marker"
+stub_device touchpad 'touchpad"; touch '"$marker"'; echo "'
+
+run_toggle touchpad off
+[[ ! -e $marker ]] || fail "touchpad disable does not execute metacharacters in the device name"
+[[ $(<"$name_file") == 'touchpad"; touch '"$marker"'; echo "' ]] ||
+  fail "a hostile device name is stored only as data"
+[[ ! -e $state_lua ]] || fail "a hostile device name is not written as Lua"
+grep -F 'hl.device({ name = "touchpad\"' "$log_file" >/dev/null ||
+  fail "hyprctl eval Lua-quotes quotes in the device name" "$(<"$log_file")"
+pass "touchpad disable treats USB device names as data"
+
+HOME="$home_dir" XDG_STATE_HOME="$xdg_decoy" OMARCHY_PATH="$ROOT" MARKER="$marker" lua - <<'LUA'
+local seen = {}
+hl = {
+  device = function(opts)
+    table.insert(seen, opts)
+  end,
+}
+
+dofile(os.getenv("OMARCHY_PATH") .. "/default/hypr/bootstrap.lua")
+require("default.hypr.toggles")
+assert(#seen == 1, "reload disables one device")
+assert(seen[1].enabled == false)
+assert(seen[1].name == 'touchpad"; touch ' .. os.getenv("MARKER") .. '; echo "', "device name is passed as a string")
+LUA
+pass "Hyprland reload loads the device name as a string"
+
+# Public PoC device name: USB iProduct is interpolated into hl.device({ name = "..." }).
+# os.execute is stubbed so the string is only checked as data.
+poc_name='trackpad"})os.execute("~/calc&")--'
+stub_device touchpad "$poc_name"
+
+run_toggle touchpad on
+: >"$log_file"
+run_toggle touchpad off
+[[ $(<"$name_file") == "$poc_name" ]] || fail "PoC device name is stored only as data"
+[[ ! -e $state_lua ]] || fail "PoC device name is not written as Lua"
+
+HOME="$home_dir" XDG_STATE_HOME="$xdg_decoy" OMARCHY_PATH="$ROOT" \
+  POC_NAME="$poc_name" EVAL_SNIPPET="$(<"$log_file")" lua - <<'LUA'
+local poc = os.getenv("POC_NAME")
+local snippet = os.getenv("EVAL_SNIPPET")
+local seen, executed = {}, false
+
+hl = {
+  device = function(opts)
+    table.insert(seen, opts)
+  end,
+}
+os.execute = function()
+  executed = true
+end
+
+assert(load(snippet, "eval", "t"))()
+assert(executed == false, "quoted hyprctl eval must not run os.execute")
+assert(#seen == 1)
+assert(seen[1].name == poc)
+assert(seen[1].enabled == false)
+
+seen, executed = {}, false
+assert(load('hl.device({ name = "' .. poc .. '", enabled = false })', "unquoted", "t"))()
+assert(executed == true, "unquoted interpolation is the Lua injection")
+
+seen, executed = {}, false
+dofile(os.getenv("OMARCHY_PATH") .. "/default/hypr/bootstrap.lua")
+require("default.hypr.toggles")
+assert(executed == false, "reload must not run os.execute")
+assert(#seen == 1)
+assert(seen[1].name == poc)
+LUA
+pass "PoC device name cannot execute via eval or reload"
+
+cat >"$stub_dir/omarchy-hw-touchpad" <<'EOF'
+#!/bin/bash
+printf 'evil\nname\n'
+EOF
+chmod +x "$stub_dir/omarchy-hw-touchpad"
+
+rm -f "$name_file"
+set +e
+run_toggle touchpad off >/dev/null 2>&1
+status=$?
+set -e
+(( status != 0 )) || fail "disable rejects a device name with a newline"
+[[ ! -e $name_file ]] || fail "a rejected device name is not persisted"
+pass "disable rejects control characters in a device name"
+
+printf 'elan-touchpad\n' >"$name_file"
+set +e
+run_toggle touchpad on >/dev/null 2>&1
+status=$?
+set -e
+(( status != 0 )) || fail "enable still reports an invalid device name"
+[[ ! -e $name_file ]] || fail "enable clears persisted state even with an invalid device name"
+pass "a bad device name cannot wedge the persisted disable"
+
+cat >"$stub_dir/omarchy-hw-touchpad" <<'EOF'
+#!/bin/bash
+:
+EOF
+chmod +x "$stub_dir/omarchy-hw-touchpad"
+
+set +e
+run_toggle touchpad off >/dev/null 2>&1
+status=$?
+set -e
+(( status != 0 )) || fail "disable errors when no device is found"
+[[ ! -e $name_file ]] || fail "no state is written when no device is found"
+pass "disable errors when no device is found"
+
+# The migration runs with the same XDG decoy: legacy files were written to
+# ~/.local/state, so that is where it must look no matter what XDG says.
+run_migration() {
+  HOME="$home_dir" XDG_STATE_HOME="$xdg_decoy" HYPRCTL_LOG="$log_file" \
+    PATH="$stub_dir:$ROOT/bin:$PATH" \
+    bash -euo pipefail "$ROOT/migrations/1787618700.sh" >/dev/null
+}
+
+mkdir -p "$state_dir"
+rm -f "$state_dir"/*-disabled-name
+printf 'hl.device({ name = "synps/2-synaptics-touchpad", enabled = false })\n' >"$state_lua"
+printf 'hl.device({ name = "hostile\\"")", enabled = false })\n' >"$state_dir/touchscreen-disabled.lua"
+
+: >"$log_file"
+run_migration
+[[ $(<"$name_file") == "synps/2-synaptics-touchpad" ]] ||
+  fail "migration recovers a device name containing a slash"
+[[ ! -e $state_lua ]] || fail "migration deletes the generated touchpad Lua"
+[[ ! -e $state_dir/touchscreen-disabled-name ]] ||
+  fail "migration does not copy a hostile name out of generated Lua"
+[[ ! -e $state_dir/touchscreen-disabled.lua ]] ||
+  fail "migration deletes hostile generated Lua even when no name is recovered"
+assert_decoy_untouched
+# The package hook reloads Hyprland before migrations run, so the disable was
+# already dropped for this session; the migration has to put it back.
+grep -Fx 'reload' "$log_file" >/dev/null ||
+  fail "migration reloads so the recovered disable applies to this session"
+pass "migration recovers plain names and discards hostile generated Lua"
+
+printf 'kept-name\n' >"$name_file"
+printf 'hl.device({ name = "other-touchpad", enabled = false })\n' >"$state_lua"
+run_migration
+[[ $(<"$name_file") == "kept-name" ]] || fail "migration keeps an existing device-name file"
+[[ ! -e $state_lua ]] || fail "migration still deletes the generated Lua"
+pass "migration is idempotent over an existing device-name file"
+
+rm -f "$name_file"
+printf 'garbage\n' >"$state_lua"
+chmod 000 "$state_lua"
+run_migration
+[[ ! -e $state_lua ]] || fail "migration removes an unreadable generated Lua"
+[[ ! -e $name_file ]] || fail "no name is recovered from an unreadable file"
+pass "an unreadable state file does not wedge the migration"
+
+: >"$log_file"
+run_migration
+[[ ! -s $log_file ]] || fail "migration with nothing to migrate does not reload"
+pass "migration no-ops with nothing left to migrate"
+
+# A compromised install carries a leftover generated touchpad-disabled.lua whose
+# device name broke out into os.execute. Until the migration deletes it, a reload
+# must not source it. toggles.lua excludes those two names from require_all, so the
+# payload never runs, while a current name-file disable still applies.
+reload_home="$tmpdir/reload-home"
+reload_state="$reload_home/.local/state/omarchy/toggles/hypr"
+mkdir -p "$reload_state"
+reload_marker="$tmpdir/reload-executed"
+rm -f "$reload_marker"
+printf 'hl.device({ name = "trackpad"})os.execute("touch %s")--", enabled = false })\n' "$reload_marker" \
+  >"$reload_state/touchpad-disabled.lua"
+printf 'elan-touchpad\n' >"$reload_state/touchpad-disabled-name"
+
+HOME="$reload_home" XDG_STATE_HOME="$reload_home/.local/state" OMARCHY_PATH="$ROOT" lua - <<'LUA'
+local disabled = {}
+hl = { device = function(opts) table.insert(disabled, opts) end }
+dofile(os.getenv("OMARCHY_PATH") .. "/default/hypr/bootstrap.lua")
+require("default.hypr.toggles")
+assert(#disabled == 1, "only the current name-file disable is applied")
+assert(disabled[1].name == "elan-touchpad", "disable uses the stored device name")
+assert(disabled[1].enabled == false)
+LUA
+[[ ! -e $reload_marker ]] || fail "a leftover legacy generated toggle Lua must not execute on reload"
+pass "reload excludes leftover legacy toggle Lua while applying the data disable"