PHP.wasm: Explore WASMFS/OPFS

Trying to switch to WASMFS and lean on the native OPFS support

see emscripten-core/emscripten#15949

Work in progress

Author: adamziel

packages/php-wasm/compile/php/Dockerfile

@@ -332,29 +332,12 @@ RUN if [[ "${PHP_VERSION:0:1}" -le "7" && "${PHP_VERSION:2:1}" -le "3" ]]; then
 		echo '#define HAVE_POSIX_READDIR_R 1' >> /root/php-src/main/php_config.h; \
 	fi;
 
-# Rename the original php_pollfd_for() implementation so that we can link our own version.
-RUN /root/replace.sh 's/static inline int php_pollfd_for\(/int php_pollfd_for(php_socket_t fd, int events, struct timeval *timeouttv); static inline int __real_php_pollfd_for(/g' /root/php-src/main/php_network.h
-
-RUN echo 'extern ssize_t wasm_read(int fd, void *buf, size_t count);' >> /root/php-src/main/php.h;
-RUN /root/replace.sh 's/ret = read/ret = wasm_read/g' /root/php-src/main/streams/plain_wrapper.c
-
 # Provide a custom implementation of the php_exec() function that handles spawning
 # the process inside exec(), passthru(), system(), etc.
 # We effectively remove the php_exec() implementation from the build by renaming it
 # to an unused identifier "php_exec_old", and then we mark php_exec as extern.
 RUN /root/replace.sh 's/PHPAPI int php_exec(.+)$/PHPAPI extern int php_exec\1; int php_exec_old\1/g' /root/php-src/ext/standard/exec.c
 
-# Provide a custom implementation of the VCWD_POPEN() function that handles spawning
-# the process inside PHP_FUNCTION(popen).
-RUN /root/replace.sh 's/#define VCWD_POPEN.+/#define VCWD_POPEN(command, type) wasm_popen(command,type)/g' /root/php-src/Zend/zend_virtual_cwd.h
-RUN echo 'extern FILE *wasm_popen(const char *cmd, const char *mode);' >> /root/php-src/Zend/zend_virtual_cwd.h
-
-# Provide a custom implementation of the shutdown() function.
-RUN perl -pi.bak -e $'s/(\s+)shutdown\(/$1 wasm_shutdown(/g' /root/php-src/sapi/cli/php_cli_server.c
-RUN perl -pi.bak -e $'s/(\s+)closesocket\(/$1 wasm_close(/g' /root/php-src/sapi/cli/php_cli_server.c
-RUN echo 'extern int wasm_shutdown(int fd, int how);' >> /root/php-src/main/php_config.h;
-RUN echo 'extern int wasm_close(int fd);' >> /root/php-src/main/php_config.h;
-
 # Don't ship PHP_FUNCTION(proc_open) with the PHP build
 # so that we can ship a patched version with php_wasm.c
 RUN echo '' > /root/php-src/ext/standard/proc_open.h;
@@ -363,7 +346,7 @@ RUN echo '' > /root/php-src/ext/standard/proc_open.c;
 RUN source /root/emsdk/emsdk_env.sh && \
 	# We're compiling PHP as emscripten's side module...
 	export JSPI_FLAGS=$(if [ "$WITH_JSPI" = "yes" ]; then echo "-sSUPPORT_LONGJMP=wasm -fwasm-exceptions"; else echo ""; fi) && \
-	EMCC_FLAGS=" -sSIDE_MODULE -Dsetsockopt=wasm_setsockopt -Dphp_exec=wasm_php_exec $JSPI_FLAGS " \
+	EMCC_FLAGS=" -sSIDE_MODULE -Dphp_exec=wasm_php_exec $JSPI_FLAGS " \
 	# ...which means we must skip all the libraries - they will be provided in the final linking step.
 	EMCC_SKIP="-lz -ledit -ldl -lncurses -lzip -lpng16 -lssl -lcrypto -lxml2 -lc -lm -lsqlite3 /root/lib/lib/libxml2.a /root/lib/lib/libsqlite3.so /root/lib/lib/libsqlite3.a /root/lib/lib/libsqlite3.a /root/lib/lib/libpng16.so /root/lib/lib/libwebp.a /root/lib/lib/libjpeg.a" \
 	emmake make -j1
@@ -927,7 +910,7 @@ RUN set -euxo pipefail; \
 	mkdir -p /build/output; \
 	source /root/emsdk/emsdk_env.sh; \
 	if [ "$WITH_JSPI" = "yes" ]; then \
-		export ASYNCIFY_FLAGS=" -s ASYNCIFY=2 -sSUPPORT_LONGJMP=wasm -fwasm-exceptions -sJSPI_IMPORTS=js_open_process,js_waitpid,js_process_status,js_create_input_device,wasm_setsockopt,wasm_shutdown,wasm_close -sJSPI_EXPORTS=wasm_sleep,wasm_read,emscripten_sleep,wasm_sapi_handle_request,wasm_sapi_request_shutdown,wasm_poll_socket,wrap_select,__wrap_select,select,php_pollfd_for,fflush,wasm_popen,wasm_read,wasm_php_exec,run_cli -s EXTRA_EXPORTED_RUNTIME_METHODS=ccall,PROXYFS,wasmExports,_malloc "; \
+		export ASYNCIFY_FLAGS=" -sWASMFS -DWASMFS_SETUP -lopfs.js -s ASYNCIFY=2 -sSUPPORT_LONGJMP=wasm -fwasm-exceptions -sJSPI_IMPORTS=js_open_process,js_waitpid,js_process_status,js_create_input_device,wasm_setsockopt,wasm_shutdown,wasm_close -sJSPI_EXPORTS=php_wasm_init,wasm_sleep,wasm_read,emscripten_sleep,wasm_sapi_handle_request,wasm_sapi_request_shutdown,wasm_poll_socket,wrap_select,__wrap_select,select,php_pollfd_for,fflush,wasm_popen,wasm_read,wasm_php_exec,run_cli -s EXTRA_EXPORTED_RUNTIME_METHODS=ccall,wasmExports,_malloc "; \
 		echo '#define PLAYGROUND_JSPI 1' > /root/php_wasm_asyncify.h; \
 	else \
 		export ASYNCIFY_FLAGS=" -s ASYNCIFY=1 -s ASYNCIFY_IGNORE_INDIRECT=1 -s EXPORTED_RUNTIME_METHODS=ccall,PROXYFS,wasmExports $(cat /root/.emcc-php-asyncify-flags)  "; \
@@ -939,6 +922,7 @@ RUN set -euxo pipefail; \
 "lengthBytesUTF8", \n\
 "FS", \n\
 "___wrap_select", \n\
+"_emscripten_stack_get_current", \n\
 "_wasm_set_sapi_name", \n\
 "_php_wasm_init", \n\
 "_emscripten_sleep", \n\
@@ -975,7 +959,6 @@ RUN set -euxo pipefail; \
 	-I TSRM/ \
 	-I /root/lib/include \
 	-L/root/lib -L/root/lib/lib/ \
-	-lproxyfs.js \
 	$ASYNCIFY_FLAGS \
 	$(cat /root/.emcc-php-wasm-flags) \
 	-s EXPORTED_FUNCTIONS="$EXPORTED_FUNCTIONS" \
@@ -1034,21 +1017,21 @@ RUN set -euxo pipefail; \
 		# Emscripten produces an if that checks a stream.stream_ops.poll property. However,
 		# stream.stream_ops is sometimes undefined and the check fails. Let's adjust it to
 		# tolerate a null stream.stream_ops value.
-		/root/replace.sh "s/if\s*\(stream\.stream_ops\.poll\)/if (stream.stream_ops?.poll)/g" /root/output/php.js; \
+		# /root/replace.sh "s/if\s*\(stream\.stream_ops\.poll\)/if (stream.stream_ops?.poll)/g" /root/output/php.js; \
 	# Make Emscripten websockets configurable
 		# Emscripten makes the Websocket proxy connect to a fixed URL.
 		# This assumes the traffic is always forwarded to the same target.
 		# However, we want to support arbitrary targets, so we need to
 		# replace the hardcoded websocket target URL with a dynamic callback.
-		/root/replace.sh $'s/if\s*\(\s*["\']string["\']\s*===\s*typeof Module\[["\']websocket["\']\]\[["\']url["\']\]\s*\)/if("function"===typeof Module["websocket"]["url"]) {\nurl = Module["websocket"]["url"](...arguments);\n}else if ("string" === typeof Module["websocket"]["url"])/g' \
-			/root/output/php.js; \
+		# /root/replace.sh $'s/if\s*\(\s*["\']string["\']\s*===\s*typeof Module\[["\']websocket["\']\]\[["\']url["\']\]\s*\)/if("function"===typeof Module["websocket"]["url"]) {\nurl = Module["websocket"]["url"](...arguments);\n}else if ("string" === typeof Module["websocket"]["url"])/g' \
+		# 	/root/output/php.js; \
 		# Enable custom WebSocket constructors to support socket options.
-		/root/replace.sh "s/ws\s*=\s*new WebSocketConstructor/if (Module['websocket']['decorator']) {WebSocketConstructor = Module['websocket']['decorator'](WebSocketConstructor);}ws = new WebSocketConstructor/g" /root/output/php.js && \
-		if [ "$EMSCRIPTEN_ENVIRONMENT" = "node" ]; then \
-		if [ "$WITH_WS_NETWORKING_PROXY" = "yes" ]; then \
-			/root/replace.sh "s/sock\.server\s*=\s*new WebSocketServer/if (Module['websocket']['serverDecorator']) {WebSocketServer = Module['websocket']['serverDecorator'](WebSocketServer);}sock.server = new WebSocketServer/g" /root/output/php.js; \
-		fi; \
-		fi; \
+		# /root/replace.sh "s/ws\s*=\s*new WebSocketConstructor/if (Module['websocket']['decorator']) {WebSocketConstructor = Module['websocket']['decorator'](WebSocketConstructor);}ws = new WebSocketConstructor/g" /root/output/php.js && \
+		# if [ "$EMSCRIPTEN_ENVIRONMENT" = "node" ]; then \
+		# if [ "$WITH_WS_NETWORKING_PROXY" = "yes" ]; then \
+		# 	/root/replace.sh "s/sock\.server\s*=\s*new WebSocketServer/if (Module['websocket']['serverDecorator']) {WebSocketServer = Module['websocket']['serverDecorator'](WebSocketServer);}sock.server = new WebSocketServer/g" /root/output/php.js; \
+		# fi; \
+		# fi; \
 	# Add MSG_PEEK flag support in recvfrom
 		#
 		# Emscripten ignores the flags argument to ___syscall_recvfrom.
@@ -1060,11 +1043,11 @@ RUN set -euxo pipefail; \
 		# reading the remaining "TTP/1.1 200 OK" and not recognizing it as a valid
 		# status line.
 		# We need to patch the syscall to support the MSG_PEEK flag.
-		if [ "$WITH_WS_NETWORKING_PROXY" = "yes" ]; then \
-			/root/replace.sh 's/sock\.sock_ops\.recvmsg\(sock,\s*len\);/sock.sock_ops.recvmsg(sock, len, typeof flags !== "undefined" ? flags : 0);/g' /root/output/php.js; \
-			/root/replace.sh 's/recvmsg\(sock,\s*length\)\s*{/recvmsg(sock, length, flags) {/g' /root/output/php.js; \
-			/root/replace.sh 's/if\s*\(sock\.type\s*===\s*1\s*&&\s*bytesRead\s*<\s*queuedLength\)/if (flags&2) {bytesRead = 0;} if (sock.type === 1 && bytesRead < queuedLength)/g' /root/output/php.js; \
-		fi ; \
+		# if [ "$WITH_WS_NETWORKING_PROXY" = "yes" ]; then \
+		# 	/root/replace.sh 's/sock\.sock_ops\.recvmsg\(sock,\s*len\);/sock.sock_ops.recvmsg(sock, len, typeof flags !== "undefined" ? flags : 0);/g' /root/output/php.js; \
+		# 	/root/replace.sh 's/recvmsg\(sock,\s*length\)\s*{/recvmsg(sock, length, flags) {/g' /root/output/php.js; \
+		# 	/root/replace.sh 's/if\s*\(sock\.type\s*===\s*1\s*&&\s*bytesRead\s*<\s*queuedLength\)/if (flags&2) {bytesRead = 0;} if (sock.type === 1 && bytesRead < queuedLength)/g' /root/output/php.js; \
+		# fi ; \
 	# Replace the hardcoded ENVIRONMENT variable with a dynamic computation
 		#
 		# The JavaScript code of the web loader and web worker loader is identical,

packages/php-wasm/compile/php/php_wasm.c

@@ -29,6 +29,7 @@
 
 // Created by Dockerfile:
 #include "php_wasm_asyncify.h"
+#include <emscripten/wasmfs.h>
 
 unsigned int wasm_sleep(unsigned int time)
 {
@@ -118,7 +119,6 @@ EM_JS(char*, js_popen_to_file, (const char *command, const char *mode, uint8_t *
 	});
 });
 
-
 /**
  * Shims poll(2) functionallity for asynchronous websockets:
  * https://man7.org/linux/man-pages/man2/poll.2.html
@@ -247,112 +247,7 @@ EM_JS(int, wasm_poll_socket, (php_socket_t socketd, int events, int timeout), {
     });
 });
 
-/**
- * Shims read(2) functionallity.
- * Enables reading from blocking pipes. By default, Emscripten
- * will throw an EWOULDBLOCK error when trying to read from a
- * blocking pipe. This function overrides that behavior and
- * instead waits for the pipe to become readable.
- *
- * @see https://github.com/WordPress/wordpress-playground/issues/951
- * @see https://github.com/emscripten-core/emscripten/issues/13214
- */
-#ifdef PLAYGROUND_JSPI
-EM_ASYNC_JS(__wasi_errno_t, js_fd_read, (__wasi_fd_t fd, const __wasi_iovec_t *iov, size_t iovcnt, __wasi_size_t *pnum), {
-	const returnCallback = (resolver) => new Promise(resolver);
-#else
-EM_JS(__wasi_errno_t, js_fd_read, (__wasi_fd_t fd, const __wasi_iovec_t *iov, size_t iovcnt, __wasi_size_t *pnum), {
-	const returnCallback = (resolver) => Asyncify.handleSleep(resolver);
-#endif
-	if (Asyncify?.State?.Normal === undefined || Asyncify?.state === Asyncify?.State?.Normal) {
-		var returnCode;
-		var stream;
-		let num = 0;
-		try
-		{
-			stream = SYSCALLS.getStreamFromFD(fd);
-			const num = doReadv(stream, iov, iovcnt);
-			HEAPU32[pnum >> 2] = num;
-			return 0;
-		}
-		catch (e)
-		{
-			// Rethrow any unexpected non-filesystem errors.
-			if (typeof FS == "undefined" || !(e.name === "ErrnoError"))
-			{
-				throw e;
-			}
-			// Only return synchronously if this isn't an asynchronous pipe.
-			// Error code 6 indicates EWOULDBLOCK – this is our signal to wait.
-			// We also need to distinguish between a process pipe and a file pipe, otherwise
-			// reading from an empty file would block until the timeout.
-			if (e.errno !== 6 || !(stream?.fd in PHPWASM.child_proc_by_fd))
-			{
-				// On failure, yield 0 bytes read to indicate EOF.
-				HEAPU32[pnum >> 2] = 0;
-				return returnCode
-			}
-		}
-	}
-
-    // At this point we know we have to poll.
-    // You might wonder why we duplicate the code here instead of always using
-    // Asyncify.handleSleep(). The reason is performance. Most of the time,
-    // the read operation will work synchronously and won't require yielding
-    // back to JS. In these cases we don't want to pay the Asyncify overhead,
-    // save the stack, yield back to JS, restore the stack etc.
-    return returnCallback((wakeUp) => {
-        var retries = 0;
-        var interval = 50;
-        var timeout = 5000;
-        // We poll for data and give up after a timeout.
-        // We can't simply rely on PHP timeout here because we don't want
-        // to, say, block the entire PHPUnit test suite without any visible
-        // feedback.
-        var maxRetries = timeout / interval;
-        function poll() {
-            var returnCode;
-            var stream;
-            let num;
-            try {
-                stream = SYSCALLS.getStreamFromFD(fd);
-                num = doReadv(stream, iov, iovcnt);
-                returnCode = 0;
-            } catch (e) {
-                if (
-                    typeof FS == 'undefined' ||
-                    !(e.name === 'ErrnoError')
-                ) {
-                    console.error(e);
-                    throw e;
-                }
-                returnCode = e.errno;
-            }
 
-            const success = returnCode === 0;
-            const failure = (
-                ++retries > maxRetries ||
-                !(fd in PHPWASM.child_proc_by_fd) ||
-                PHPWASM.child_proc_by_fd[fd]?.exited ||
-                FS.isClosed(stream)
-            );
-
-            if (success) {
-                HEAPU32[pnum >> 2] = num;
-                wakeUp(0);
-            } else if (failure) {
-                // On failure, yield 0 bytes read to indicate EOF.
-                HEAPU32[pnum >> 2] = 0;
-                // If the failure is due to a timeout, return 0 to indicate that we
-                // reached EOF. Otherwise, propagate the error code.
-                wakeUp(returnCode === 6 ? 0 : returnCode);
-            } else {
-                setTimeout(poll, interval);
-            }
-        }
-        poll();
-    })
-});
 extern int __wasi_syscall_ret(__wasi_errno_t code);
 
 // Exit code of the last exited child process call.
@@ -1825,6 +1720,23 @@ static void wasm_sapi_log_message(char *message TSRMLS_DC)
  */
 int php_wasm_init()
 {
+	int err;
+	// backend_t memory = wasmfs_create_memory_backend();
+	// The /internal directory is required by the C module. It's where the
+	// stdout, stderr, and headers information are written for the JavaScript
+	// code to read later on.
+	// err = wasmfs_create_directory("/internal", 0777, memory);
+	// err = wasmfs_create_directory("/wordpress", 0777, memory);
+	// The files from the shared directory are shared between all the
+	// PHP processes managed by PHPProcessManager.
+	// FS.mkdir('/internal/shared');
+	// The files from the preload directory are preloaded using the
+	// auto_prepend_file php.ini directive.
+	// FS.mkdir('/internal/shared/preload');
+
+	backend_t opfs = wasmfs_create_opfs_backend();
+	err = wasmfs_create_directory("/internal", 0777, opfs);
+
 	wasm_server_context = malloc(sizeof(wasm_server_context_t));
 	wasm_init_server_context();
 

packages/php-wasm/compile/php/phpwasm-emscripten-library.js

@@ -16,17 +16,6 @@ const LibraryExample = {
 	// JavaScript library under the PHPWASM object:
 	$PHPWASM: {
 		init: function () {
-			// The /internal directory is required by the C module. It's where the
-			// stdout, stderr, and headers information are written for the JavaScript
-			// code to read later on.
-			FS.mkdir('/internal');
-			// The files from the shared directory are shared between all the
-			// PHP processes managed by PHPProcessManager.
-			FS.mkdir('/internal/shared');
-			// The files from the preload directory are preloaded using the
-			// auto_prepend_file php.ini directive.
-			FS.mkdir('/internal/shared/preload');
-
 			PHPWASM.EventEmitter = ENVIRONMENT_IS_NODE
 				? require('events').EventEmitter
 				: class EventEmitter {