ext/curl: Add CURLOPT_OPENSOCKETFUNCTION

NEWS

@@ -15,6 +15,8 @@ PHP                                                                        NEWS
   . Uploading multiple files with the same field name is now possible by
     passing an array of CURLFile or CURLStringFile objects to
     CURLOPT_POSTFIELDS. (Sjoerd Langkemper)
+  . Added CURLOPT_OPENSOCKETFUNCTION to allow or refuse each address libcurl
+    is about to connect to. (Xavier Leune, Nicolas Grekas)
 
 - FPM:
   . Use recommended TYPE, HELP order for OpenMetrics metadata.

UPGRADING

@@ -76,6 +76,13 @@ PHP 8.7 UPGRADE NOTES
     array of CURLFile or CURLStringFile objects to CURLOPT_POSTFIELDS.
   . Support for upstream options CURLOPT_ECH, CURLOPT_HAPROXY_CLIENT_IP, and
     CURLOPT_SERVER_RESPONSE_TIMEOUT_MS.
+  . Added CURLOPT_OPENSOCKETFUNCTION to register a callback that allows or
+    refuses each address libcurl is about to connect to, after DNS resolution
+    and before connect(). The callback receives the CurlHandle, the address
+    and the port. It must return true to let libcurl create the socket, or
+    false to make libcurl try the next address. The address is an IP address,
+    or the path of a UNIX domain socket with port 0. The callback isn't called
+    when a connection is reused, which CURLOPT_PREREQFUNCTION can check.
 
 - Intl:
   . Added IntlRelativeDateTimeFormatter to format relative date and time offsets
@@ -172,6 +179,7 @@ PHP 8.7 UPGRADE NOTES
     CURLOPT_SERVER_RESPONSE_TIMEOUT_MS.
   . CURLOPT_MAIL_RCPT_ALLOWFAILS is the correctly spelled alternative to
     CURLOPT_MAIL_RCPT_ALLLOWFAILS.
+  . CURLOPT_OPENSOCKETFUNCTION.
 
 ========================================
 11. Changes to INI File Handling

ext/curl/curl.stub.php

@@ -257,6 +257,11 @@
  * @cvalue CURLOPT_NOSIGNAL
  */
 const CURLOPT_NOSIGNAL = UNKNOWN;
+/**
+ * @var int
+ * @cvalue CURLOPT_OPENSOCKETFUNCTION
+ */
+const CURLOPT_OPENSOCKETFUNCTION = UNKNOWN;
 /**
  * @var int
  * @cvalue CURLOPT_PORT

ext/curl/curl_private.h

@@ -83,6 +83,7 @@ typedef struct {
 	zend_fcall_info_cache xferinfo;
 	zend_fcall_info_cache fnmatch;
 	zend_fcall_info_cache debug;
+	zend_fcall_info_cache opensocket;
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 	zend_fcall_info_cache prereq;
 #endif

ext/curl/interface.c

@@ -26,7 +26,15 @@
 
 #ifdef PHP_WIN32
 #include <winsock2.h>
+#include <ws2tcpip.h>
 #include <sys/types.h>
+#else
+#include <sys/socket.h>
+#include <netinet/in.h>
+# ifdef AF_UNIX
+# include <sys/un.h>
+# define PHP_CURL_HAVE_AF_UNIX 1
+# endif
 #endif
 
 #include <curl/curl.h>
@@ -36,6 +44,8 @@
 #include "ext/standard/info.h"
 #include "ext/standard/file.h"
 #include "ext/standard/url.h"
+#include "ext/standard/php_net.h"
+#include "php_network.h"
 #include "curl_private.h"
 
 #ifdef __GNUC__
@@ -499,6 +509,10 @@ static HashTable *curl_get_gc(zend_object *object, zval **table, int *n)
 		zend_get_gc_buffer_add_fcc(gc_buffer, &curl->handlers.debug);
 	}
 
+	if (ZEND_FCC_INITIALIZED(curl->handlers.opensocket)) {
+		zend_get_gc_buffer_add_fcc(gc_buffer, &curl->handlers.opensocket);
+	}
+
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 	if (ZEND_FCC_INITIALIZED(curl->handlers.prereq)) {
 		zend_get_gc_buffer_add_fcc(gc_buffer, &curl->handlers.prereq);
@@ -837,6 +851,100 @@ static int curl_ssh_hostkeyfunction(void *clientp, int keytype, const char *key,
 }
 #endif
 
+/* {{{ curl_opensocket */
+static curl_socket_t curl_opensocket(void *clientp, curlsocktype purpose, struct curl_sockaddr *address)
+{
+	php_curl *ch = (php_curl *) clientp;
+	curl_socket_t sock = CURL_SOCKET_BAD;
+
+#if PHP_CURL_DEBUG
+	fprintf(stderr, "curl_opensocket() called\n");
+	fprintf(stderr, "clientp = %p, purpose = %d, family = %d, socktype = %d, protocol = %d\n", clientp, purpose, address->family, address->socktype, address->protocol);
+#endif
+
+	/* Refuse what can't be passed to the callback. Once the callback threw, refuse
+	 * the remaining addresses too. */
+	if (UNEXPECTED(purpose != CURLSOCKTYPE_IPCXN || !ZEND_FCC_INITIALIZED(ch->handlers.opensocket) || EG(exception))) {
+		return sock;
+	}
+
+	zval args[3];
+	zend_long port = 0;
+	bool is_ip = true;
+
+	switch (address->family) {
+		case AF_INET:
+			port = ntohs(((const struct sockaddr_in *) &address->addr)->sin_port);
+			break;
+#ifdef AF_INET6
+		case AF_INET6:
+			port = ntohs(((const struct sockaddr_in6 *) &address->addr)->sin6_port);
+			break;
+#endif
+#ifdef PHP_CURL_HAVE_AF_UNIX
+		case AF_UNIX: {
+			/* The path of the socket, with its leading NUL byte for an abstract one */
+			const struct sockaddr_un *sa_un = (const struct sockaddr_un *) &address->addr;
+			size_t len = 0;
+			if (address->addrlen > offsetof(struct sockaddr_un, sun_path)) {
+				len = MIN(address->addrlen - offsetof(struct sockaddr_un, sun_path), sizeof(sa_un->sun_path));
+				if (len && sa_un->sun_path[0] != '\0') {
+					const char *nul = memchr(sa_un->sun_path, '\0', len);
+					if (nul) {
+						len = nul - sa_un->sun_path;
+					}
+				}
+			}
+			ZVAL_STRINGL(&args[1], sa_un->sun_path, len);
+			is_ip = false;
+			break;
+		}
+#endif
+		default:
+			return sock;
+	}
+
+	if (is_ip) {
+		zend_string *ip = php_inet_ntop(&address->addr);
+		if (UNEXPECTED(!ip)) {
+			return sock;
+		}
+		ZVAL_STR(&args[1], ip);
+	}
+
+	zval retval;
+
+	GC_ADDREF(&ch->std);
+	ZVAL_OBJ(&args[0], &ch->std);
+	ZVAL_LONG(&args[2], port);
+
+	php_curl_call_callback(ch, &ch->handlers.opensocket, &retval, /* argc */ 3, args);
+
+	if (!Z_ISUNDEF(retval)) {
+		_php_curl_verify_handlers(ch, /* reporterror */ true);
+		if (Z_TYPE(retval) == IS_TRUE) {
+			if (!EG(exception)) {
+				/* What libcurl does when no callback is set */
+				sock = socket(address->family, address->socktype, address->protocol);
+				/* libcurl doesn't make the socket non-blocking if the callback removed itself */
+				if (sock != CURL_SOCKET_BAD && php_set_sock_blocking(sock, false) != SUCCESS) {
+					closesocket(sock);
+					sock = CURL_SOCKET_BAD;
+				}
+			}
+		} else if (Z_TYPE(retval) != IS_FALSE) {
+			zend_type_error("The CURLOPT_OPENSOCKETFUNCTION callback must return a bool");
+			zval_ptr_dtor(&retval);
+		}
+	}
+
+	zval_ptr_dtor(&args[0]);
+	zval_ptr_dtor(&args[1]);
+
+	return sock;
+}
+/* }}} */
+
 /* {{{ curl_read */
 static size_t curl_read(char *data, size_t size, size_t nmemb, void *ctx)
 {
@@ -1155,6 +1263,7 @@ void init_curl_handle(php_curl *ch)
 	ch->handlers.xferinfo = empty_fcall_info_cache;
 	ch->handlers.fnmatch = empty_fcall_info_cache;
 	ch->handlers.debug = empty_fcall_info_cache;
+	ch->handlers.opensocket = empty_fcall_info_cache;
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 	ch->handlers.prereq = empty_fcall_info_cache;
 #endif
@@ -1332,6 +1441,7 @@ void _php_setup_easy_copy_handlers(php_curl *ch, php_curl *source)
 	php_curl_copy_fcc_with_option(ch, CURLOPT_XFERINFODATA, &ch->handlers.xferinfo, &source->handlers.xferinfo);
 	php_curl_copy_fcc_with_option(ch, CURLOPT_FNMATCH_DATA, &ch->handlers.fnmatch, &source->handlers.fnmatch);
 	php_curl_copy_fcc_with_option(ch, CURLOPT_DEBUGDATA, &ch->handlers.debug, &source->handlers.debug);
+	php_curl_copy_fcc_with_option(ch, CURLOPT_OPENSOCKETDATA, &ch->handlers.opensocket, &source->handlers.opensocket);
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 	php_curl_copy_fcc_with_option(ch, CURLOPT_PREREQDATA, &ch->handlers.prereq, &source->handlers.prereq);
 #endif
@@ -1727,6 +1837,21 @@ static zend_result _php_curl_setopt(php_curl *ch, zend_long option, zval *zvalue
 		HANDLE_CURL_OPTION_CALLABLE(ch, CURLOPT_FNMATCH_, handlers.fnmatch, curl_fnmatch);
 		HANDLE_CURL_OPTION_CALLABLE(ch, CURLOPT_DEBUG, handlers.debug, curl_debug);
 
+		case CURLOPT_OPENSOCKETFUNCTION: {
+			bool result = php_curl_set_callable_handler(&ch->handlers.opensocket, zvalue, is_array_config, "CURLOPT_OPENSOCKETFUNCTION");
+			/* Without a callback, libcurl creates the sockets itself */
+			if (ZEND_FCC_INITIALIZED(ch->handlers.opensocket)) {
+				curl_easy_setopt(ch->cp, CURLOPT_OPENSOCKETFUNCTION, curl_opensocket);
+				curl_easy_setopt(ch->cp, CURLOPT_OPENSOCKETDATA, ch);
+			} else {
+				curl_easy_setopt(ch->cp, CURLOPT_OPENSOCKETFUNCTION, NULL);
+			}
+			if (!result) {
+				return FAILURE;
+			}
+			break;
+		}
+
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 		HANDLE_CURL_OPTION_CALLABLE(ch, CURLOPT_PREREQ, handlers.prereq, curl_prereqfunction);
 #endif
@@ -2925,6 +3050,9 @@ static void curl_free_obj(zend_object *object)
 	if (ZEND_FCC_INITIALIZED(ch->handlers.debug)) {
 		zend_fcc_dtor(&ch->handlers.debug);
 	}
+	if (ZEND_FCC_INITIALIZED(ch->handlers.opensocket)) {
+		zend_fcc_dtor(&ch->handlers.opensocket);
+	}
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 	if (ZEND_FCC_INITIALIZED(ch->handlers.prereq)) {
 		zend_fcc_dtor(&ch->handlers.prereq);
@@ -3017,6 +3145,10 @@ static void _php_curl_reset_handlers(php_curl *ch)
 		zend_fcc_dtor(&ch->handlers.debug);
 	}
 
+	if (ZEND_FCC_INITIALIZED(ch->handlers.opensocket)) {
+		zend_fcc_dtor(&ch->handlers.opensocket);
+	}
+
 #if LIBCURL_VERSION_NUM >= 0x075000 /* Available since 7.80.0 */
 	if (ZEND_FCC_INITIALIZED(ch->handlers.prereq)) {
 		zend_fcc_dtor(&ch->handlers.prereq);

ext/curl/tests/curl_opensocketfunction_abstract_unix_socket.phpt

@@ -0,0 +1,30 @@
+--TEST--
+CURLOPT_OPENSOCKETFUNCTION: abstract UNIX domain socket
+--EXTENSIONS--
+curl
+--SKIPIF--
+<?php
+if (PHP_OS_FAMILY !== 'Linux') die('skip Linux only');
+if (!(curl_version()['features'] & CURL_VERSION_UNIX_SOCKETS)) die('skip libcurl without UNIX domain sockets');
+?>
+--FILE--
+<?php
+$ch = curl_init('http://localhost/');
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+curl_setopt($ch, CURLOPT_ABSTRACT_UNIX_SOCKET, 'php-curl-does-not-exist');
+curl_setopt($ch, CURLOPT_OPENSOCKETFUNCTION, function (CurlHandle $handle, string $address, int $port): bool {
+    /* The address keeps the leading NUL byte of abstract sockets */
+    var_dump($address === "\0php-curl-does-not-exist", $port);
+    return false;
+});
+var_dump(curl_exec($ch));
+var_dump(curl_errno($ch) === CURLE_COULDNT_CONNECT);
+
+echo "Done";
+?>
+--EXPECT--
+bool(true)
+int(0)
+bool(false)
+bool(true)
+Done

ext/curl/tests/curl_opensocketfunction_address.phpt

@@ -0,0 +1,37 @@
+--TEST--
+CURLOPT_OPENSOCKETFUNCTION: IPv6 addresses
+--EXTENSIONS--
+curl
+--SKIPIF--
+<?php
+$s = @stream_socket_server('tcp://[::1]:0', $errno, $errstr);
+if ($s === false) die('skip no IPv6 loopback available');
+fclose($s);
+?>
+--FILE--
+<?php
+/* No listener is needed: the callback runs before connect() */
+function probe(string $url): void {
+    $ch = curl_init($url);
+    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+    curl_setopt($ch, CURLOPT_OPENSOCKETFUNCTION, function (CurlHandle $handle, string $address, int $port): bool {
+        var_dump($address, $port);
+        return false;
+    });
+    curl_exec($ch);
+}
+
+probe('http://[::1]:1/');
+probe('http://[::ffff:127.0.0.1]:2/');
+probe('http://[::ffff:7f00:1]:3/');
+
+echo "Done";
+?>
+--EXPECT--
+string(3) "::1"
+int(1)
+string(16) "::ffff:127.0.0.1"
+int(2)
+string(16) "::ffff:127.0.0.1"
+int(3)
+Done

ext/curl/tests/curl_opensocketfunction_file_scheme.phpt

@@ -0,0 +1,45 @@
+--TEST--
+CURLOPT_OPENSOCKETFUNCTION is not called for file://, which creates no socket
+--EXTENSIONS--
+curl
+--SKIPIF--
+<?php
+if (!in_array('file', curl_version()['protocols'], true)) die('skip file:// protocol not supported');
+?>
+--FILE--
+<?php
+$file = __DIR__ . '/curl_opensocketfunction_file_scheme.txt';
+file_put_contents($file, "not a socket\n");
+
+$ch = curl_init();
+curl_setopt($ch, CURLOPT_URL, 'file://' . $file);
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+
+$calls = 0;
+curl_setopt($ch, CURLOPT_OPENSOCKETFUNCTION, function () use (&$calls): bool {
+    $calls++;
+    return false;
+});
+
+/* The callback refuses everything, yet the file is read: CURLOPT_PROTOCOLS_STR
+ * should be used too. */
+var_dump(curl_exec($ch));
+var_dump($calls);
+
+curl_setopt($ch, CURLOPT_PROTOCOLS_STR, 'http,https');
+var_dump(curl_exec($ch));
+var_dump(curl_errno($ch) === CURLE_UNSUPPORTED_PROTOCOL);
+
+echo "Done";
+?>
+--CLEAN--
+<?php
+@unlink(__DIR__ . '/curl_opensocketfunction_file_scheme.txt');
+?>
+--EXPECT--
+string(13) "not a socket
+"
+int(0)
+bool(false)
+bool(true)
+Done