Add support for building to the wasm32-wasi target

This change performs initial adaptations allowing the PHP interpreter
to be compiled for the wasm32-wasi target.

Challenges found:

- `setjmp`/`longjmp` not available in this platform: can be worked
around by using Asyncify; this is work in progress to be presented at
a later stage.

- Processes: WebAssembly has no concept of processes, so many
process-related operations had to be emulated, or are no-ops. `getpid`
and clocks are emulated. Signal handling has minimal emulation. `uid`,
`gid`, and related concepts that do not map well to a WebAssembly
sandbox are also removed.
  - `mmap` implementation is a minimal emulation that allows to
  allocate memory or map file contents to it.

- Filesystem: many filesystem operations cannot be mapped to WASI, and
so these operations are no-ops.
  - `S_IFSOCK` matches `S_IFIFO` in current `wasi-libc`; removed the
  `S_IFIFO` case, because `clang` will fail building due to a
  duplicated constant in `switch` statements.
  - File locks are also stubbed and always return success.

- Networking: given the WebAssembly sandbox and capability-based
security, it is not possible to open a socket as of today. Although
the host is able to open a socket and forward a file descriptor,
leaving the `accept()` syscall to the WebAssembly module, we have
removed networking syscalls at this time.

The only binary that builds with this change is `php-cgi` at this
time. It's possible to adapt the change so that `php` builds and is
able to run scripts and programs in the filesystem in a CLI
fashion. However, starting a PHP server to listen on a socket using
the CLI is a more contrived situation, given it's not possible to open
a listening socket from within the sandbox; however it would be
possible to create the socket on the host side, and forward the file
descriptor to the module.

Co-Authored-By: Asen Alexandrov <[email protected]>

Author: ereslibre

Zend/zend.c

@@ -834,7 +834,9 @@ static void executor_globals_ctor(zend_executor_globals *executor_globals) /* {{
 #endif
 	executor_globals->saved_fpu_cw_ptr = NULL;
 	executor_globals->active = 0;
+#ifndef __wasi__
 	executor_globals->bailout = NULL;
+#endif // __wasi__
 	executor_globals->error_handling  = EH_NORMAL;
 	executor_globals->exception_class = NULL;
 	executor_globals->exception = NULL;
@@ -1223,6 +1225,7 @@ ZEND_COLD void zenderror(const char *error) /* {{{ */
 ZEND_API ZEND_COLD ZEND_NORETURN void _zend_bailout(const char *filename, uint32_t lineno) /* {{{ */
 {
 
+#ifndef __wasi__
 	if (!EG(bailout)) {
 		zend_output_debug_string(1, "%s(%d) : Bailed out without a bailout address!", filename, lineno);
 		exit(-1);
@@ -1233,6 +1236,7 @@ ZEND_API ZEND_COLD ZEND_NORETURN void _zend_bailout(const char *filename, uint32
 	CG(in_compilation) = 0;
 	EG(current_execute_data) = NULL;
 	LONGJMP(*EG(bailout), FAILURE);
+#endif // __wasi__
 }
 /* }}} */
 

Zend/zend.h

@@ -256,6 +256,7 @@ typedef size_t (*zend_write_func_t)(const char *str, size_t str_length);
 
 #define zend_bailout()		_zend_bailout(__FILE__, __LINE__)
 
+#ifndef __wasi__
 #define zend_try												\
 	{															\
 		JMP_BUF *__orig_bailout = EG(bailout);					\
@@ -272,6 +273,19 @@ typedef size_t (*zend_write_func_t)(const char *str, size_t str_length);
 	}
 #define zend_first_try		EG(bailout)=NULL;	zend_try
 
+#else // __wasi__
+#define zend_try												\
+	{															\
+		if (1) {
+#define zend_catch												\
+		} else {
+#define zend_end_try()											\
+		}														\
+	}
+#define zend_first_try		zend_try
+#endif // __wasi__
+
+
 BEGIN_EXTERN_C()
 void zend_startup(zend_utility_functions *utility_functions);
 void zend_shutdown(void);

Zend/zend_alloc.c

@@ -80,7 +80,7 @@
 #include <fcntl.h>
 #include <errno.h>
 
-#ifndef _WIN32
+#if !defined(_WIN32) && HAVE_MMAP
 # include <sys/mman.h>
 # ifndef MAP_ANON
 #  ifdef MAP_ANONYMOUS
@@ -445,6 +445,8 @@ static void zend_mm_munmap(void *addr, size_t size)
 #endif
 		}
 	}
+#elif ! HAVE_MMAP
+  free(addr);
 #else
 	if (munmap(addr, size) != 0) {
 #if ZEND_MM_ERROR
@@ -472,6 +474,8 @@ static void *zend_mm_mmap_fixed(void *addr, size_t size)
 	}
 	ZEND_ASSERT(ptr == addr);
 	return ptr;
+#elif ! HAVE_MMAP
+	return NULL;
 #else
 	int flags = MAP_PRIVATE | MAP_ANON;
 #if defined(MAP_EXCL)
@@ -508,6 +512,10 @@ static void *zend_mm_mmap(size_t size)
 		return NULL;
 	}
 	return ptr;
+#elif ! HAVE_MMAP
+	void* ptr = malloc(size);
+	memset(ptr, 0, size);
+	return ptr;
 #else
 	void *ptr;
 
@@ -717,6 +725,11 @@ static zend_always_inline void zend_mm_hugepage(void* ptr, size_t size)
 
 static void *zend_mm_chunk_alloc_int(size_t size, size_t alignment)
 {
+#if ! HAVE_MMAP
+	void* ptr = aligned_alloc(alignment, size);
+	memset(ptr, 0, size);
+	return ptr;
+#else
 	void *ptr = zend_mm_mmap(size);
 
 	if (ptr == NULL) {
@@ -767,6 +780,7 @@ static void *zend_mm_chunk_alloc_int(size_t size, size_t alignment)
 #endif
 		return ptr;
 	}
+#endif
 }
 
 static void *zend_mm_chunk_alloc(zend_mm_heap *heap, size_t size, size_t alignment)
@@ -2932,7 +2946,7 @@ ZEND_API void start_memory_manager(void)
 #else
 	alloc_globals_ctor(&alloc_globals);
 #endif
-#ifndef _WIN32
+#if !defined(_WIN32) && HAVE_MMAP
 #  if defined(_SC_PAGESIZE)
 	REAL_PAGE_SIZE = sysconf(_SC_PAGESIZE);
 #  elif defined(_SC_PAGE_SIZE)

Zend/zend_fibers.c

@@ -39,7 +39,7 @@
 # include <ucontext.h>
 #endif
 
-#ifndef ZEND_WIN32
+#if !defined(ZEND_WIN32) && HAVE_MMAP
 # include <unistd.h>
 # include <sys/mman.h>
 # include <limits.h>
@@ -108,7 +108,9 @@ typedef struct _zend_fiber_vm_state {
 	zend_execute_data *current_execute_data;
 	int error_reporting;
 	uint32_t jit_trace_num;
+#ifndef __wasi__
 	JMP_BUF *bailout;
+#endif // __wasi__
 	zend_fiber *active_fiber;
 #ifdef ZEND_CHECK_STACK_LIMIT
 	void *stack_base;
@@ -125,7 +127,9 @@ static zend_always_inline void zend_fiber_capture_vm_state(zend_fiber_vm_state *
 	state->current_execute_data = EG(current_execute_data);
 	state->error_reporting = EG(error_reporting);
 	state->jit_trace_num = EG(jit_trace_num);
+#ifndef __wasi__
 	state->bailout = EG(bailout);
+#endif // __wasi__
 	state->active_fiber = EG(active_fiber);
 #ifdef ZEND_CHECK_STACK_LIMIT
 	state->stack_base = EG(stack_base);
@@ -142,7 +146,9 @@ static zend_always_inline void zend_fiber_restore_vm_state(zend_fiber_vm_state *
 	EG(current_execute_data) = state->current_execute_data;
 	EG(error_reporting) = state->error_reporting;
 	EG(jit_trace_num) = state->jit_trace_num;
+#ifndef __wasi__
 	EG(bailout) = state->bailout;
+#endif // __wasi__
 	EG(active_fiber) = state->active_fiber;
 #ifdef ZEND_CHECK_STACK_LIMIT
 	EG(stack_base) = state->stack_base;
@@ -236,6 +242,8 @@ static zend_fiber_stack *zend_fiber_stack_allocate(size_t size)
 		return NULL;
 	}
 # endif
+#elif defined __wasi__
+	pointer = malloc(alloc_size);
 #else
 	pointer = mmap(NULL, alloc_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, -1, 0);
 
@@ -302,6 +310,8 @@ static void zend_fiber_stack_free(zend_fiber_stack *stack)
 
 #ifdef ZEND_WIN32
 	VirtualFree(pointer, 0, MEM_RELEASE);
+#elif defined __wasi__
+	free(pointer);
 #else
 	munmap(pointer, stack->size + ZEND_FIBER_GUARD_PAGES * page_size);
 #endif
@@ -415,7 +425,7 @@ ZEND_API bool zend_fiber_init_context(zend_fiber_context *context, void *kind, z
 	makecontext(handle, (void (*)(void)) zend_fiber_trampoline, 0);
 
 	context->handle = handle;
-#else
+#elif !defined(__wasi__)
 	// Stack grows down, calculate the top of the stack. make_fcontext then shifts pointer to lower 16-byte boundary.
 	void *stack = (void *) ((uintptr_t) context->stack->pointer + context->stack->size);
 
@@ -428,6 +438,8 @@ ZEND_API bool zend_fiber_init_context(zend_fiber_context *context, void *kind, z
 
 	context->handle = make_fcontext(stack, context->stack->size, zend_fiber_trampoline);
 	ZEND_ASSERT(context->handle != NULL && "make_fcontext() never returns NULL");
+#else
+	return false;
 #endif
 
 	context->kind = kind;
@@ -502,16 +514,18 @@ ZEND_API void zend_fiber_switch_context(zend_fiber_transfer *transfer)
 
 	/* Copy transfer struct because it might live on the other fiber's stack that will eventually be destroyed. */
 	*transfer = *transfer_data;
-#else
+#elif !defined(__wasi__)
 	boost_context_data data = jump_fcontext(to->handle, transfer);
 
 	/* Copy transfer struct because it might live on the other fiber's stack that will eventually be destroyed. */
 	*transfer = *data.transfer;
+#else
+	return;
 #endif
 
 	to = transfer->context;
 
-#ifndef ZEND_FIBER_UCONTEXT
+#if !defined(ZEND_FIBER_UCONTEXT) && !defined(__wasi__)
 	/* Get the context that resumed us and update its handle to allow for symmetric coroutines. */
 	to->handle = data.handle;
 #endif

Zend/zend_globals.h

@@ -21,7 +21,9 @@
 #define ZEND_GLOBALS_H
 
 
+#ifndef __wasi__
 #include <setjmp.h>
+#endif
 
 #include "zend_globals_macros.h"
 
@@ -162,7 +164,9 @@ struct _zend_executor_globals {
 
 	HashTable included_files;	/* files already included */
 
+#ifndef __wasi__
 	JMP_BUF *bailout;
+#endif
 
 	int error_reporting;
 	int exit_status;

Zend/zend_types.h

@@ -1453,5 +1453,15 @@ static zend_always_inline uint32_t zval_delref_p(zval* pz) {
 #define ZVAL_COPY_OR_DUP_PROP(z, v) \
 	do { ZVAL_COPY_OR_DUP(z, v); Z_PROP_FLAG_P(z) = Z_PROP_FLAG_P(v); } while (0)
 
+#ifdef __wasi__
+#define	LOG_EMERG	0	/* system is unusable */
+#define	LOG_ALERT	1	/* action must be taken immediately */
+#define	LOG_CRIT	2	/* critical conditions */
+#define	LOG_ERR		3	/* error conditions */
+#define	LOG_WARNING	4	/* warning conditions */
+#define	LOG_NOTICE	5	/* normal but significant condition */
+#define	LOG_INFO	6	/* informational */
+#define	LOG_DEBUG	7	/* debug-level messages */
+#endif // __wasi__
 
 #endif /* ZEND_TYPES_H */

Zend/zend_virtual_cwd.c

@@ -1400,6 +1400,8 @@ CWD_API int virtual_chmod(const char *filename, mode_t mode) /* {{{ */
 		}
 		ret = php_win32_ioutil_chmod(new_state.cwd, mode);
 	}
+#elif defined __wasi__
+	ret = 0;
 #else
 	ret = chmod(new_state.cwd, mode);
 #endif
@@ -1428,7 +1430,11 @@ CWD_API int virtual_chown(const char *filename, uid_t owner, gid_t group, int li
 		ret = -1;
 #endif
 	} else {
+#ifndef __wasi__
 		ret = chown(new_state.cwd, owner, group);
+#else
+		ret = 0;
+#endif // __wasi__
 	}
 
 	CWD_STATE_FREE_ERR(&new_state);
@@ -1706,8 +1712,11 @@ CWD_API FILE *virtual_popen(const char *command, const char *type) /* {{{ */
 	*ptr++ = ' ';
 
 	memcpy(ptr, command, command_length+1);
+#ifndef __wasi__
 	retval = popen(command_line, type);
-
+#else
+	retval = 0;
+#endif // __wasi__
 	efree(command_line);
 	return retval;
 }

configure.ac

@@ -199,6 +199,9 @@ label2:
 fi
 PHP_SUBST(RE2C_FLAGS)
 
+dnl Check if __wasi__ is defined by the compiler
+AC_CHECK_DECLS([__wasi__])
+
 dnl Platform-specific compile settings.
 dnl ----------------------------------------------------------------------------
 
@@ -1334,11 +1337,13 @@ else
   if test "$fiber_os" = 'mac'; then
     AC_DEFINE([_XOPEN_SOURCE], 1, [ ])
   fi
-  AC_CHECK_HEADER(ucontext.h, [
-    AC_DEFINE([ZEND_FIBER_UCONTEXT], 1, [ ])
-  ], [
-       AC_MSG_ERROR([fibers not available on this platform])
-  ])
+  if test "$ac_cv_have_decl___wasi__" != "yes"; then
+    AC_CHECK_HEADER(ucontext.h, [
+      AC_DEFINE([ZEND_FIBER_UCONTEXT], 1, [ ])
+    ], [
+        AC_MSG_ERROR([fibers not available on this platform])
+    ])
+  fi
 fi
 
 LIBZEND_BASIC_CHECKS

ext/fileinfo/libmagic/fsmagic.c

@@ -167,6 +167,7 @@ file_fsmagic(struct magic_set *ms, const char *fn, zend_stat_t *sb)
 # endif
 #endif
 
+#ifndef __wasi__
 #ifdef	S_IFIFO
 	case S_IFIFO:
 		if((ms->flags & MAGIC_DEVICES) != 0)
@@ -179,6 +180,7 @@ file_fsmagic(struct magic_set *ms, const char *fn, zend_stat_t *sb)
 			return -1;
 		break;
 #endif
+#endif // __wasi__
 #ifdef	S_IFDOOR
 	case S_IFDOOR:
 		if (mime) {

ext/pdo_sqlite/sqlite_statement.c

@@ -32,7 +32,9 @@ static int pdo_sqlite_stmt_dtor(pdo_stmt_t *stmt)
 	pdo_sqlite_stmt *S = (pdo_sqlite_stmt*)stmt->driver_data;
 
 	if (S->stmt) {
+#ifndef __wasi__
 		sqlite3_finalize(S->stmt);
+#endif
 		S->stmt = NULL;
 	}
 	efree(S);