PHP: Support blocking pipes

Shims read(2) functionallity by providing an alternative read()
function called wasm_read() that gracefully handles blocking
pipes. This is neecessary because Emscripten does not support
blocking pipes and instead returns EWOULDBLOCK.

See:

* #951
* emscripten-core/emscripten#13214

 ## Testing instructions

Confirm the `proc_open()` tests pass on CI. This PR adjusts a few
of them to make sure the output is read without the tricky sleep()
call.

packages/php-wasm/cli/src/main.ts

@@ -1,7 +1,7 @@
 /**
  * A CLI script that runs PHP CLI via the WebAssembly build.
  */
-import { writeFileSync, existsSync, mkdtempSync } from 'fs';
+import { writeFileSync, existsSync, mkdtempSync, rmSync, rmdirSync } from 'fs';
 import { rootCertificates } from 'tls';
 
 import {
@@ -66,11 +66,17 @@ async function run() {
 	`
 		);
 
-		return spawn('sh', [tempScriptPath], {
-			shell: true,
-			stdio: ['pipe', 'pipe', 'pipe'],
-			timeout: 100,
-		});
+		try {
+			return spawn(updatedCommand, [], {
+				shell: true,
+				stdio: ['pipe', 'pipe', 'pipe'],
+				timeout: 5000,
+			});
+		} finally {
+			// Remove the temporary directory
+			rmSync(tempScriptPath);
+			rmdirSync(tempDir);
+		}
 	});
 
 	const hasMinusCOption = args.some((arg) => arg.startsWith('-c'));
@@ -87,7 +93,11 @@ async function run() {
 			throw result;
 		})
 		.finally(() => {
-			process.exit(0);
+			setTimeout(() => {
+				process.exit(0);
+				// 100 is an arbitrary number. It's there to give any child processes
+				// a chance to pass their output to JS before the main process exits.
+			}, 100);
 		});
 }
 

packages/php-wasm/compile/php/Dockerfile

@@ -247,9 +247,6 @@ RUN echo '#define ZEND_MM_ERROR 0' >> /root/php-src/main/php_config.h;
 RUN /root/replace.sh 's/define php_sleep sleep/define php_sleep wasm_sleep/g' /root/php-src/main/php.h
 RUN echo 'extern unsigned int wasm_sleep(unsigned int time);' >> /root/php-src/main/php.h;
 
-RUN /root/replace.sh 's/#define php_stream_flush\(stream\)	_php_stream_flush/extern int wasm_php_stream_flush(php_stream *stream, int closing);\n#define php_stream_flush(stream)	wasm_php_stream_flush/g' /root/php-src/main/php_streams.h
-RUN /root/replace.sh 's/_php_stream_flush\(stream/wasm_php_stream_flush(stream/g' /root/php-src/main/streams/streams.c
-
 RUN /root/replace.sh 's/define HAVE_UNISTD_H 1/define HAVE_UNISTD_H 0/g' /root/php-src/main/php_config.h
 
 # PHP <= 7.3 is not very good at detecting the presence of the POSIX readdir_r function
@@ -262,6 +259,9 @@ RUN if [[ "${PHP_VERSION:0:1}" -le "7" && "${PHP_VERSION:2:1}" -le "3" ]]; then
 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 /root/replace.sh 's/static int php_cli_server_poller_poll/extern int wasm_select(int, fd_set * __restrict, fd_set * __restrict, fd_set * __restrict, struct timeval * __restrict); static int php_cli_server_poller_poll/g' /root/php-src/sapi/cli/php_cli_server.c
 
+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_select() function.
 RUN /root/replace.sh 's/return php_select\(/return wasm_select(/g' /root/php-src/sapi/cli/php_cli_server.c
 
@@ -394,8 +394,9 @@ RUN echo -n ' -s ASYNCIFY=1 -s ASYNCIFY_IGNORE_INDIRECT=1 ' >> /root/.emcc-php-w
 "invoke_viiiiiiiii",\n\
 "js_open_process",\n\
 "js_popen_to_file",\n\
-"wasm_poll_socket",\n\
+"js_fd_read",\n\
 "js_module_onMessage",\n\
+"wasm_poll_socket",\n\
 "wasm_shutdown"]'; \
 	echo -n " -s ASYNCIFY_IMPORTS=$ASYNCIFY_IMPORTS " | tr -d "\n" >> /root/.emcc-php-wasm-flags; \
 	export ASYNCIFY_ONLY_UNPREFIXED=$'"dynCall_dd",\
@@ -430,8 +431,14 @@ RUN echo -n ' -s ASYNCIFY=1 -s ASYNCIFY_IGNORE_INDIRECT=1 ' >> /root/.emcc-php-w
 "dynCall_viiiiiii",\
 "dynCall_viiiiiiii",'; \
 	export ASYNCIFY_ONLY=$'"__fwritex",\
+"read",\
 "zif_sleep",\
 "zif_stream_get_contents",\
+"_php_stream_fill_read_buffer",\
+"_php_stream_read",\
+"php_stream_read_to_str",\
+"zif_fread",\
+"wasm_read",\
 "php_stdiop_read",\
 "fwrite",\
 "zif_fwrite",\
@@ -448,6 +455,8 @@ RUN echo -n ' -s ASYNCIFY=1 -s ASYNCIFY_IGNORE_INDIRECT=1 ' >> /root/.emcc-php-w
 "php_stream_temp_flush",\
 "_php_stream_flush",\
 "php_stream_flush",\
+"_php_stream_write",\
+"php_stream_write",\
 "_php_stream_free",\
 "_php_stream_free_enclosed",\
 "stream_resource_persistent_dtor",\
@@ -505,7 +514,6 @@ RUN echo -n ' -s ASYNCIFY=1 -s ASYNCIFY_IGNORE_INDIRECT=1 ' >> /root/.emcc-php-w
 "wasm_sleep",\
 "wasm_php_stream_flush",\
 "wasm_php_stream_read",\
-"_php_stream_read",\
 "php_crc32_stream_bulk_update",\
 "phar_parse_zipfile",\
 "get_http_body",\
@@ -768,7 +776,7 @@ RUN set -euxo pipefail; \
 "_wasm_set_phpini_path", \n\
 "_wasm_set_phpini_entries", \n\
 "_wasm_add_SERVER_entry", \n\
-"_wasm_php_stream_flush", \n\
+"_wasm_read", \n\
 "_wasm_add_uploaded_file", \n\
 "_wasm_sapi_handle_request", \n\
 "_wasm_set_content_length", \n\

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

@@ -33,6 +33,13 @@ unsigned int wasm_sleep(unsigned int time) {
 
 extern int *wasm_setsockopt(int sockfd, int level, int optname, intptr_t optval, size_t optlen, int dummy);
 extern char *js_popen_to_file(const char *cmd, const char *mode, uint8_t *exit_code_ptr);
+extern int __wasi_syscall_ret(__wasi_errno_t code);
+extern __wasi_errno_t js_fd_read(
+	__wasi_fd_t fd,
+	const __wasi_iovec_t* iovs,
+	size_t iovs_len,
+	__wasi_size_t* nread
+);
 
 /**
  * Passes a message to the JavaScript module and writes the response
@@ -152,6 +159,28 @@ static size_t handle_line(int type, zval *array, char *buf, size_t bufl) {
 	return bufl;
 }
 
+/**
+ * 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
+ */
+EMSCRIPTEN_KEEPALIVE ssize_t wasm_read(int fd, void *buf, size_t count) {
+	struct __wasi_iovec_t iov = {
+		.buf = buf,
+		.buf_len = count
+	};
+	size_t num;
+	if (__wasi_syscall_ret(js_fd_read(fd, &iov, 1, &num))) {
+		return -1;
+	}
+	return num;
+}
+
 /*
  * If type==0, only last line of output is returned (exec)
  * If type==1, all lines will be printed and last lined returned (system)
@@ -250,44 +279,6 @@ EMSCRIPTEN_KEEPALIVE int wasm_php_exec(int type, const char *cmd, zval *array, z
 
 // -----------------------------------------------------------
 
-/**
- * Replaces the default php_stream_read() implementation with our own that yields back to the JavaScript event
- * loop to give JS a chance to actually populate the stream with any data that may already be buffered.
- * This is the magic sauce that makes the stream_get_contents here work:
- * 
- * $descriptorspec = array(
- *     0 => array("pipe", "r"),  // stdin
- *     1 => array("pipe", "w"),  // stdout
- *     2 => array("pipe", "w")   // stderr
- * );
- * $process = proc_open("less", $descriptorspec, $pipes);
- * fwrite($pipes[0], "Hello world");
- * fclose($pipes[0]);
- * stream_get_contents($pipes[1]);
- *
- * Without the yield, stream_get_contents() would return an empty string because the data is not yet available.
- * JavaScript populates the stream via asynchronous event handlers, so we need to yield back to give them a chance
- * to run.
- */
-int wasm_php_stream_flush(php_stream *stream, int closing)
-{
-	int retval = _php_stream_flush(stream, closing);
-	// Yield to JS event loop. 500 is an arbitrary value that
-	// seems to work well in practice when writing to stdin
-	// via proc_open() as the underlying process catches up
-	// with the input. It is quite a long time, though, so
-	// this should be very much revisited.
-	//
-	// Unfortunately it's not a bulletproof solution and will
-	// likely break in some cases. A proper fix would be to
-	// identify which low level libc function behaves differently
-	// in emscripten and natively, and then patch it to yield back
-	// to the JS event loop – much like is being done with `select()`
-	// below.
-	emscripten_sleep(500);
-	return retval;
-}
-
 int wasm_socket_has_data(php_socket_t fd);
 int wasm_poll_socket(php_socket_t fd, int events, int timeoutms);
 

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

@@ -598,6 +598,61 @@ const LibraryExample = {
 		return 0;
 	},
 
+	/**
+	 * 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
+	 */
+	js_fd_read: function (fd, iov, iovcnt, pnum) {
+		var returnCode;
+		try {
+			var stream = SYSCALLS.getStreamFromFD(fd);
+			var num = doReadv(stream, iov, iovcnt);
+			HEAPU32[pnum >> 2] = num;
+			returnCode = 0;
+		} catch (e) {
+			if (typeof FS == 'undefined' || !(e.name === 'ErrnoError')) throw e;
+			returnCode = e.errno;
+		}
+
+		if (returnCode === 6 /*EWOULDBLOCK*/) {
+			return Asyncify.handleSleep(function (wakeUp) {
+				var timeout = 10000; // @TODO Do not hardcode this
+				var interval = 50;
+				var retries = 0;
+				var maxRetries = timeout / interval;
+				var pollHandle = setInterval(function poll() {
+					var returnCode;
+					try {
+						var stream = SYSCALLS.getStreamFromFD(fd);
+						var num = doReadv(stream, iov, iovcnt);
+						HEAPU32[pnum >> 2] = num;
+						returnCode = 0;
+					} catch (e) {
+						if (
+							typeof FS == 'undefined' ||
+							!(e.name === 'ErrnoError')
+						) {
+							console.error(e);
+							throw e;
+						}
+						returnCode = e.errno;
+					}
+					if (returnCode !== 6 || ++retries > maxRetries) {
+						clearInterval(pollHandle);
+						wakeUp(returnCode);
+					}
+				}, interval);
+			});
+		}
+		return returnCode;
+	},
+
 	/**
 	 * Shims popen(3) functionallity:
 	 * https://man7.org/linux/man-pages/man3/popen.3.html