<?php
defined( 'ABSPATH' ) || exit;

/**
 * Asks the site for its own pages the way a visitor would, and reads which layer answered.
 * One request function, two users: the doctor (is the cache really serving, and if not, why) and
 * the warm-up after a purge (refill the most visited pages and measure them on the way).
 *
 * Transport: the site's own server first, by pinning the real host name to 127.0.0.1 (scheme, port,
 * TLS name and Host stay what a visitor sends, so the page is rendered exactly as for https visitors,
 * while DNS, CDN and firewall are skipped). When that cannot connect, the public address. Every answer
 * must prove it came from this plugin (our header, our cache note, or our engine header); otherwise
 * something in front answered, and that is reported, never counted as a success.
 */
class ZeroConfig_Probe {

	const UA_DESKTOP = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/126.0 Safari/537.36 ZeroConfigProbe/1.0';
	// Must match the drop-in's mobile pattern, or the "mobile" pass would fill the desktop copy.
	const UA_MOBILE  = 'Mozilla/5.0 (iPhone; CPU iPhone OS 17_5 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.5 Mobile/15E148 Safari/604.1 ZeroConfigProbe/1.0';
	const HEALTH     = 'zeroconfig_health';
	const MAX_URLS   = 10;

	private static $resolve = '';

	/* ---------- one request ---------- */

	/**
	 * GET $url as a visitor on $device ('d' or 'm'). Returns code, ms, layer (htaccess | dropin | wp |
	 * miss | foreign | unknown | error), via (loopback | public), foreign (cache headers seen) and error.
	 */
	public static function fetch( $url, $device = 'd', $loopback_only = false, $timeout = 8 ) {
		// $timeout is the budget for all attempts together, not per attempt.
		$end  = microtime( true ) + $timeout;
		$left = function () use ( $end ) {
			return max( 1, (int) ceil( $end - microtime( true ) ) );
		};
		$out = self::fetch_via( $url, $device, $left(), self::custom_loopback() ? 'custom' : '127.0.0.1' );
		if ( 'error' === $out['layer'] && ! self::custom_loopback() && ! self::timed_out( $out ) && microtime( true ) < $end ) {
			$out = self::fetch_via( $url, $device, $left(), '[::1]' ); // servers listening on IPv6 only
		}
		// A timeout means the server is there but slow: asking again over the public address only waits longer.
		if ( 'error' === $out['layer'] && ! $loopback_only && ! self::timed_out( $out ) && microtime( true ) < $end ) {
			$out = self::fetch_via( $url, $device, $left(), false );
		}
		return $out;
	}

	/**
	 * A static file from this server (no PHP involved), the same way as the ping. Tells "PHP is busy"
	 * apart from "the whole server is slow" when a page request times out.
	 */
	public static function fetch_static( $url, $timeout = 2 ) {
		$r = self::fetch_via( $url, 'd', $timeout, self::custom_loopback() ? 'custom' : '127.0.0.1' );
		if ( 'error' === $r['layer'] && ! self::custom_loopback() && ! self::timed_out( $r ) ) {
			$r = self::fetch_via( $url, 'd', $timeout, '[::1]' );
		}
		return $r;
	}

	public static function timed_out( array $r ) {
		return 'error' === $r['layer'] && (bool) preg_match( '/timed out|timeout/i', (string) $r['error'] );
	}

	/**
	 * Do the .htaccess rules run? /zc-ping.txt must answer with the token in the cache folder.
	 * Returns 'ok', 'no' (something else answered) or 'unknown' (this server cannot reach itself).
	 */
	public static function ping( $timeout = 3 ) {
		$token = ZeroConfig_Page_Cache::ping_token();
		$r     = self::fetch_raw( home_url( '/zc-ping.txt' ), 'd', self::custom_loopback() ? 'custom' : '127.0.0.1', $timeout );
		if ( 'error' === $r['layer'] && ! self::custom_loopback() && ! self::timed_out( $r ) ) {
			$r = self::fetch_raw( home_url( '/zc-ping.txt' ), 'd', '[::1]', $timeout );
		}
		if ( 'error' === $r['layer'] ) {
			return 'unknown';
		}
		return 200 === $r['code'] && trim( $r['body'] ) === $token ? 'ok' : 'no';
	}

	private static function fetch_raw( $url, $device, $loopback, $timeout ) {
		return self::fetch_via( $url, $device, $timeout, $loopback, true );
	}

	/**
	 * Where the web server listens when that is not the site's own address, for sites behind a proxy
	 * that ends TLS (Docker, a load balancer): define( 'ZEROCONFIG_LOOPBACK', 'http://127.0.0.1:8080' );
	 * Requests then go there with the site's Host and X-Forwarded-Proto, as the proxy would send them.
	 */
	public static function custom_loopback() {
		return defined( 'ZEROCONFIG_LOOPBACK' ) && ZEROCONFIG_LOOPBACK ? untrailingslashit( (string) ZEROCONFIG_LOOPBACK ) : '';
	}

	/** $loopback: the local address to pin the host name to, 'custom' for ZEROCONFIG_LOOPBACK, or false for the public address. */
	private static function fetch_via( $url, $device, $timeout, $loopback, $with_body = false ) {
		$p     = wp_parse_url( $url );
		$extra = array();
		if ( 'custom' === $loopback ) {
			$extra = array( 'Host' => $p['host'] . ( isset( $p['port'] ) ? ':' . $p['port'] : '' ), 'X-Forwarded-Proto' => $p['scheme'] ?? 'http' );
			$url   = self::custom_loopback() . ( $p['path'] ?? '/' ) . ( isset( $p['query'] ) ? '?' . $p['query'] : '' );
			$p     = wp_parse_url( $url );
		}
		$port = isset( $p['port'] ) ? (int) $p['port'] : ( 'https' === ( $p['scheme'] ?? 'http' ) ? 443 : 80 );
		self::$resolve = $loopback && 'custom' !== $loopback && ! empty( $p['host'] ) ? $p['host'] . ':' . $port . ':' . $loopback : '';
		$pinned        = false;
		$pin           = function ( $handle ) use ( &$pinned ) {
			if ( '' !== self::$resolve && defined( 'CURLOPT_RESOLVE' ) ) {
				// Inside the HTTP API's own http_api_curl hook: pins this site's own host name to the loopback
				// address so the check reaches this server, not a CDN in front of it. TLS checks are unchanged.
				curl_setopt( $handle, CURLOPT_RESOLVE, array( self::$resolve ) ); // phpcs:ignore WordPress.WP.AlternativeFunctions.curl_curl_setopt
				$pinned = true;
			}
		};
		add_action( 'http_api_curl', $pin );
		$t   = microtime( true );
		$res = wp_remote_get(
			$url,
			array(
				'timeout'     => $timeout,
				'redirection' => 0,
				'user-agent'  => 'm' === $device ? self::UA_MOBILE : self::UA_DESKTOP,
				'headers'     => array_merge( array( 'Accept-Encoding' => 'gzip' ), $extra ), // the .htaccess rules serve the .gz copy only to gzip clients
				'cookies'     => array(),
			)
		);
		$ms = (int) round( ( microtime( true ) - $t ) * 1000 );
		remove_action( 'http_api_curl', $pin );
		self::$resolve = '';
		$via = $loopback ? 'loopback' : 'public';
		if ( is_wp_error( $res ) ) {
			return array( 'code' => 0, 'ms' => $ms, 'layer' => 'error', 'via' => $via, 'foreign' => array(), 'error' => $res->get_error_message() );
		}
		if ( $loopback && 'custom' !== $loopback && ! $pinned ) { // not curl: there was no way to pin, so this went out like a public request
			return array( 'code' => 0, 'ms' => $ms, 'layer' => 'error', 'via' => $via, 'foreign' => array(), 'error' => 'no curl' );
		}
		$code = (int) wp_remote_retrieve_response_code( $res );
		$h    = wp_remote_retrieve_headers( $res );
		$body = (string) wp_remote_retrieve_body( $res );
		$out  = array(
			'code'    => $code,
			'ms'      => $ms,
			'layer'   => 200 === $code ? self::layer( $h, $body ) : 'http_' . $code,
			'via'     => $via,
			'foreign' => self::foreign_cache( $h ),
			'headers' => isset( $h['x-zeroconfig-cache'] ) ? 'ours' : 'none',
			'server'  => substr( (string) ( $h['server'] ?? '' ), 0, 60 ),
			'browser' => isset( $h['cache-control'] ) || isset( $h['expires'] ) ? 'told' : 'none', // Cache-Control/Expires present?
			'error'   => '',
		);
		if ( $with_body ) {
			$out['body'] = substr( $body, 0, 200 );
		}
		return $out;
	}

	/** Which layer answered. Apache serves our files without PHP, and without mod_headers without our header. */
	public static function layer( $headers, $body ) {
		if ( self::foreign_cache( $headers ) ) {
			return 'foreign'; // a CDN or host cache answered; whatever it replays (our header, our marker) is old news
		}
		$hit = strtoupper( (string) ( $headers['x-zeroconfig-cache'] ?? '' ) );
		if ( 'HIT-HTACCESS' === $hit ) {
			return 'htaccess';
		}
		if ( 'HIT' === $hit ) {
			return 'dropin';
		}
		if ( 'HIT-WP' === $hit ) {
			return 'wp';
		}
		if ( isset( $headers['x-zc-render'] ) ) {
			return 'miss'; // WordPress built it for this probe (the engine marks probe renders): not from cache
		}
		// Our stored copy without our header: Apache straight from disk on a host without mod_headers,
		// or a cache in front that kept our copy. The doctor tells them apart with the ping rule. A weak
		// signal only: white label and the footer filter can remove the note, so its absence means nothing.
		if ( preg_match( '/<!-- Cached by [^@>]{1,60} @ \d{4}-\d\d-\d\dT/', substr( $body, -300 ) ) ) {
			return 'noted';
		}
		return 'unknown';
	}

	/** Cache headers of hosts and CDNs in front of WordPress that answered with their own copy. */
	public static function foreign_cache( $headers ) {
		$seen = array();
		foreach ( array( 'cf-cache-status', 'x-cache', 'x-kinsta-cache', 'x-litespeed-cache', 'x-proxy-cache', 'x-sg-cache', 'x-wpe-cached', 'x-fastcgi-cache', 'x-nginx-cache', 'x-srcache-fetch-status' ) as $k ) {
			$v = (string) ( $headers[ $k ] ?? '' );
			if ( '' !== $v && preg_match( '/\bhit\b/i', $v ) ) {
				$seen[] = $k . ': ' . $v;
			}
		}
		return $seen;
	}

	/* ---------- which pages ---------- */

	/** True when the plugin would store and serve this URL from cache (same rules as the cache key). */
	public static function cacheable( $url ) {
		$p = wp_parse_url( $url );
		if ( empty( $p['host'] ) || ! empty( $p['query'] ) ) {
			return false;
		}
		$host = preg_replace( '/[^a-z0-9.\-]/', '', strtolower( $p['host'] . ( isset( $p['port'] ) ? ':' . $p['port'] : '' ) ) );
		if ( ! in_array( $host, ZeroConfig_Page_Cache::hosts( true ), true ) ) {
			return false;
		}
		$path = isset( $p['path'] ) ? $p['path'] : '/';
		if ( preg_match( '#^/(wp-admin|wp-login|wp-json|xmlrpc|wp-cron|wp-comments|feed)|\.(php|xml|txt)$|/feed/?$#i', $path ) || false !== strpos( $path, '..' ) ) {
			return false;
		}
		foreach ( (array) ZeroConfig_Page_Cache::config()['exclude'] as $ex ) {
			if ( '' !== $ex && false !== stripos( $path, $ex ) ) {
				return false;
			}
		}
		if ( function_exists( 'wc_get_page_permalink' ) ) {
			foreach ( array( 'cart', 'checkout', 'myaccount' ) as $wc ) {
				$wp = (string) wp_parse_url( wc_get_page_permalink( $wc ), PHP_URL_PATH );
				if ( '' !== trim( $wp, '/' ) && 0 === strpos( trailingslashit( $path ), trailingslashit( $wp ) ) ) {
					return false;
				}
			}
		}
		return true;
	}

	/** The pages worth refilling first: home, shop, blog index, the main menu, then the newest content. Frozen per run. */
	public static function candidates( $max = self::MAX_URLS ) {
		$urls = array( home_url( '/' ) );
		if ( function_exists( 'wc_get_page_permalink' ) ) {
			$urls[] = wc_get_page_permalink( 'shop' );
		}
		if ( get_option( 'page_for_posts' ) ) {
			$urls[] = get_permalink( (int) get_option( 'page_for_posts' ) );
		}
		foreach ( (array) get_nav_menu_locations() as $menu_id ) {
			foreach ( (array) wp_get_nav_menu_items( $menu_id ) as $item ) {
				if ( isset( $item->menu_item_parent, $item->url ) && 0 === (int) $item->menu_item_parent ) {
					$urls[] = $item->url;
				}
			}
			break; // the first menu location is the main one in practically every theme
		}
		foreach ( array( 'post', 'product', 'page' ) as $type ) {
			foreach ( (array) get_posts( array( 'post_type' => $type, 'post_status' => 'publish', 'numberposts' => 3, 'fields' => 'ids' ) ) as $id ) {
				$urls[] = get_permalink( $id );
			}
		}
		$out = array();
		foreach ( $urls as $u ) {
			$u = self::normalize( (string) $u );
			if ( '' !== $u && self::cacheable( $u ) && ! in_array( $u, $out, true ) ) {
				$out[] = $u;
			}
			if ( count( $out ) >= $max ) {
				break;
			}
		}
		return $out;
	}

	/** The address as WordPress serves it without a redirect: the permalink trailing slash, when it uses one. */
	public static function normalize( $url ) {
		if ( '' !== $url && '/' === $url[0] && ( ! isset( $url[1] ) || '/' !== $url[1] ) ) {
			$url = home_url( $url ); // a relative menu link ("/contact"): an address of this site
		}
		$p    = wp_parse_url( $url );
		$path = isset( $p['path'] ) ? $p['path'] : '/';
		if ( '' === $url || empty( $p['host'] ) || ! empty( $p['query'] ) || preg_match( '/\.[a-z0-9]{2,5}$/i', $path ) ) {
			return $url; // cacheable() decides what to do with anything left over
		}
		$slashed = user_trailingslashit( untrailingslashit( $path ) );
		if ( '' === $slashed ) {
			$slashed = '/';
		}
		return $slashed === $path ? $url : str_replace( '//' . $p['host'] . ( isset( $p['port'] ) ? ':' . $p['port'] : '' ) . $path, '//' . $p['host'] . ( isset( $p['port'] ) ? ':' . $p['port'] : '' ) . $slashed, $url );
	}

	/* ---------- health record ---------- */

	public static function health() {
		$h = get_option( self::HEALTH, array() );
		return is_array( $h ) ? $h : array();
	}

	public static function save_health( array $patch ) {
		wp_cache_delete( self::HEALTH, 'options' );
		update_option( self::HEALTH, array_merge( self::health(), $patch ), false );
	}

	/** Cache file for a URL and device, in the folder the cache key uses. */
	public static function file_for( $url, $device = 'd' ) {
		$p    = wp_parse_url( $url );
		$host = preg_replace( '/[^a-z0-9.\-]/', '', strtolower( ( $p['host'] ?? '' ) . ( isset( $p['port'] ) ? ':' . $p['port'] : '' ) ) );
		$rel  = trim( (string) ( $p['path'] ?? '' ), '/' );
		return ZEROCONFIG_CACHE_DIR . '/' . $host . ( '' !== $rel ? '/' . $rel : '' ) . '/index-' . $device . '.html';
	}
}