x86(wasi): support mapdir

Signed-off-by: Kohei Tokunaga <ktokunaga.mail@gmail.com>

Dockerfile

@@ -5,7 +5,7 @@ ARG WASI_SDK_VERSION=19
 ARG WASI_SDK_VERSION_FULL=${WASI_SDK_VERSION}.0
 ARG WASI_VFS_VERSION=v0.3.0
 ARG WIZER_VERSION=04e49c989542f2bf3a112d60fbf88a62cce2d0d0
-ARG EMSDK_VERSION=3.1.42
+ARG EMSDK_VERSION=3.1.40 # TODO: support recent version
 ARG BINARYEN_VERSION=113
 ARG BUSYBOX_VERSION=1_36_0
 ARG RUNC_VERSION=v1.1.8

README.md

@@ -40,7 +40,13 @@ bin   dev  home  lib32	libx32	mnt  proc  run	 srv  tmp  var
 boot  etc  lib	 lib64	media	opt  root  sbin  sys  usr
 ```
 
-> NOTE: Directory mapping is available on non-x86_64 containers (TinyEMU-emulated ones) as of now. Other WASI features untested. Future version will support more WASI features.
+Directories mapped to the WASM program is accessible on the container as well.
+
+```
+$ mkdir -p /tmp/share/ && echo hi > /tmp/share/from-host
+$ wasmtime --mapdir /mnt/share::/tmp/share out.wasm cat /mnt/share/from-host
+hi
+```
 
 ### Container on Browser
 
@@ -184,8 +190,6 @@ $ wasmtime --mapdir /test/dir/share::/tmp/share /app/out.wasm ls /test/dir/share
 hi
 ```
 
-> NOTE: Directory mapping is available on non-x86_64 containers (TinyEMU-emulated) as of now. This should be available on all containers in the future.
-
 ## Motivation
 
 Though more and more programming languages start to support WASM, it's not easy to run the existing programs on WASM.
@@ -201,7 +205,7 @@ The following shows the techniqual details:
 - Builder: [BuildKit](https://github.com/moby/buildkit) runs the conversion steps written in Dockerfile.
 - Emulator: [Bochs](https://bochs.sourceforge.io/) emulates x86_64 CPU on WASM. [TinyEMU](https://bellard.org/tinyemu/) emulates RISC-V CPU on WASM. They're compiled to WASM using [wasi-sdk](https://github.com/WebAssembly/wasi-sdk) (for WASI and on-browser) and [emscripten](https://github.com/emscripten-core/emscripten) (for on-browser).
 - Guest OS: Linux runs on the emulated CPU. [runc](https://github.com/opencontainers/runc) starts the container. Non-x86 and non-RISC-V containers runs with additional emulation by QEMU installed via [`tonistiigi/binfmt`](https://github.com/tonistiigi/binfmt).
-- Directory Mapping: WASI filesystem API makes host directories visible to the emulator. TinyEMU mounts them to the guest linux via virtio-9p. Unsupported on Bochs (for x86_64 emulation) as of now.
+- Directory Mapping: WASI filesystem API makes host directories visible to the emulator. Emulators mount them to the guest linux via virtio-9p.
 - Packaging: [wasi-vfs](https://github.com/kateinoigakukun/wasi-vfs) (for WASI and on-browser) and emscripten (for on-browser) are used for packaging the dependencies. The kernel is pre-booted during the build using [wizer](https://github.com/bytecodealliance/wizer/) to minimize the startup latency (for WASI only as of now).
 - Security: The converted container runs in the sandboxed WASM (WASI) VM with the limited access to the host system.
 
@@ -216,11 +220,11 @@ The following shows the techniqual details:
 
 |runtime |stdio|mapdir|note|
 |---|---|---|---|
-|wasmtime|:heavy_check_mark:|:construction:||
-|wamr(wasm-micro-runtime)|:heavy_check_mark:|:construction:||
-|wazero|:heavy_check_mark:|:construction:||
-|wasmer|:construction: (stdin unsupported)|:construction:|non-blocking stdin doesn't seem to work|
-|wasmedge|:construction: (stdin unsupported)|:construction:|non-blocking stdin doesn't seem to work|
+|wasmtime|:heavy_check_mark:|:heavy_check_mark:||
+|wamr(wasm-micro-runtime)|:heavy_check_mark:|:heavy_check_mark:||
+|wazero|:heavy_check_mark:|:heavy_check_mark:||
+|wasmer|:construction: (stdin unsupported)|:heavy_check_mark:|non-blocking stdin doesn't seem to work|
+|wasmedge|:construction: (stdin unsupported)|:heavy_check_mark:|non-blocking stdin doesn't seem to work|
 
 ### risc-v and other architecutre's containers
 
@@ -232,14 +236,6 @@ The following shows the techniqual details:
 |wasmer|:construction: (stdin unsupported)|:heavy_check_mark:|non-blocking stdin doesn't seem to work|
 |wasmedge|:construction: (stdin unsupported)|:heavy_check_mark:|non-blocking stdin doesn't seem to work|
 
-Example of mapdir with wasmtime:
-
-```console
-$ mkdir -p /tmp/share/ && echo hi > /tmp/share/hi
-$ wasmtime --mapdir /test/dir/share::/tmp/share -- out.wasm --entrypoint=cat -- /test/dir/share/hi
-hi
-```
-
 ## Similar projects
 
 There are several container runtimes support running WASM applications, but they don't run containers on WASM.

cmd/init/main.go

@@ -104,22 +104,25 @@ func doInit() error {
 	}
 
 	wg.Wait()
-	if infoSourceRemoteAddr == "" {
 
-		// WASI-related filesystems
-		for _, tag := range []string{rootFSTag, packFSTag} {
-			dst := filepath.Join("/mnt", tag)
-			if err := os.Mkdir(dst, 0777); err != nil {
-				return err
-			}
-			log.Printf("mounting %q to %q\n", tag, dst)
-			if err := syscall.Mount(tag, dst, "9p", 0, "trans=virtio,version=9p2000.L,msize=8192"); err != nil {
-				log.Printf("failed mounting %q: %v\n", tag, err)
-				break
-			}
-		}
+	initMounts := []string{rootFSTag, packFSTag}
+	if infoSourceRemoteAddr != "" {
+		initMounts = []string{rootFSTag}
+	}
 
+	// WASI-related filesystems
+	for _, tag := range initMounts {
+		dst := filepath.Join("/mnt", tag)
+		if err := os.Mkdir(dst, 0777); err != nil {
+			return err
+		}
+		log.Printf("mounting %q to %q\n", tag, dst)
+		if err := syscall.Mount(tag, dst, "9p", 0, "trans=virtio,version=9p2000.L,msize=8192"); err != nil {
+			log.Printf("failed mounting %q: %v\n", tag, err)
+			break
+		}
 	}
+
 	specD, err := os.ReadFile(cfg.Container.RuntimeConfigPath)
 	if err != nil {
 		return err

patches/bochs/Bochs/bochs/Makefile.in

@@ -143,6 +143,7 @@ NONINLINE_OBJS = \
 	plugin.o \
 	crc.o \
 	bxthread.o \
+	wasm.o \
 	@EXTRA_BX_OBJS@
 
 EXTERN_ENVIRONMENT_OBJS = \
@@ -819,3 +820,7 @@ plugin.o: plugin.@CPP_SUFFIX@ bochs.h config.h osdep.h gui/paramtree.h logio.h \
  plugin.h extplugin.h param_names.h pc_system.h bx_debug/debug.h config.h \
  osdep.h memory/memory-bochs.h gui/siminterface.h gui/paramtree.h \
  gui/gui.h plugin.h
+wasm.o: wasm.@CPP_SUFFIX@ iodev/iodev.h bochs.h config.h osdep.h gui/paramtree.h \
+ logio.h misc/bswap.h plugin.h extplugin.h param_names.h pc_system.h \
+ bx_debug/debug.h config.h osdep.h memory/memory-bochs.h \
+ gui/siminterface.h gui/gui.h iodev/pci.h wasm.h

patches/bochs/Bochs/bochs/iodev/devices.cc

@@ -359,6 +359,9 @@ void bx_devices_c::init(BX_MEM_C *newmem)
   bulkIOQuantumsRequested = 0;
   bulkIOQuantumsTransferred = 0;
 
+#if defined(EMSCRIPTEN) || defined(WASI)
+  PLUG_load_plugin(mapdirVirtio9p, PLUGTYPE_STANDARD);
+#endif
   bx_init_plugins();
 
   /* now perform checksum of CMOS memory */

patches/bochs/Bochs/bochs/main.cc

@@ -34,6 +34,10 @@
 #include "iodev/usb/usb_common.h"
 #endif
 
+#if defined(EMSCRIPTEN) || defined(WASI)
+#include "wasm.h"
+#endif
+
 #ifdef WASI
 #include <wizer.h>
 extern "C" {
@@ -315,34 +319,6 @@ void print_statistics_tree(bx_param_c *node, int level)
 
 bool vm_init_done = false;
 
-typedef struct {
-  char *contents;
-  int len;
-  int lim;
-} FSVirtFile;
-
-int write_info(FSVirtFile *f, int pos, int len, const char *str)
-{
-  if ((pos + len) > f->lim) {
-    printf("too many write (%d > %d)", pos + len, f->lim);
-    return -1;
-  }
-  for (int i = 0; i < len; i++) {
-    f->contents[pos + i] = str[i];
-  }
-  return len;
-}
-
-int putchar_info(FSVirtFile *f, int pos, char c)
-{
-  if ((pos + 1) > f->lim) {
-    printf("too many write (%d > %d)", pos + 1, f->lim);
-    return -1;
-  }
-  f->contents[pos] = c;
-  return 1;
-}
-
 int write_entrypoint(FSVirtFile *f, int pos1, const char *entrypoint)
 {
   int p, pos = pos1;
@@ -490,6 +466,9 @@ int init_wasi_info(int argc, char **argv, FSVirtFile *info)
     }
 
     info->len = pos;
+#ifdef WASI
+    info->len += write_preopen_info(info, pos);
+#endif
     return 0;
 }
 
@@ -735,90 +714,6 @@ extern "C" {
 extern void __wasi_vfs_rt_init(void);
 }
 
-#include <wasi/libc.h>
-
-// Populating preopen in the same way as wasi-libc does as well as updating
-// the preopens list managed by wasi-libc.
-// https://github.com/WebAssembly/wasi-libc/blob/wasi-sdk-19/libc-bottom-half/sources/preopens.c
-// We use our list of preopenes for mounting them to the container.
-
-typedef struct preopen {
-    /// The path prefix associated with the file descriptor.
-    const char *prefix;
-
-    /// The file descriptor.
-    __wasi_fd_t fd;
-} preopen;
-
-preopen *preopens;
-size_t num_preopens = 0;
-size_t preopen_capacity = 0;
-
-static int populate_preopens() {
-  for (__wasi_fd_t fd = 3; fd != 0; ++fd) {
-    __wasi_prestat_t prestat;
-    __wasi_errno_t ret = __wasi_fd_prestat_get(fd, &prestat);
-    if (ret == __WASI_ERRNO_BADF) {
-      break;
-    }
-    if (ret != __WASI_ERRNO_SUCCESS)
-      return -1;
-
-    switch (prestat.tag) {
-    case __WASI_PREOPENTYPE_DIR: {
-      char *prefix = (char *) malloc(prestat.u.dir.pr_name_len + 1);
-      if (prefix == NULL)
-        return -1;
-      if (__wasi_fd_prestat_dir_name(fd, (uint8_t *)prefix, prestat.u.dir.pr_name_len) != __WASI_ERRNO_SUCCESS)
-        return -1;
-      prefix[prestat.u.dir.pr_name_len] = '\0';
-
-      int off = 0;
-      while (1) {
-        if (prefix[off] == '/') {
-          off++;
-        } else if (prefix[off] == '.' && prefix[off + 1] == '/') {
-          off += 2;
-        } else if (prefix[off] == '.' && prefix[off + 1] == 0) {
-          off++;
-        } else {
-          break;
-        }
-      }
-
-      char *p = &(prefix[off]);
-      if (__wasilibc_register_preopened_fd(fd, strdup(p)) != 0)
-        return -1;
-
-      if (num_preopens == preopen_capacity) {
-        size_t new_capacity = preopen_capacity == 0 ? 4 : preopen_capacity * 2;
-        preopen *new_preopens = (preopen *)calloc(sizeof(preopen), new_capacity);
-        if (new_preopens == NULL) {
-          return -1;
-        }
-        memcpy(new_preopens, preopens, num_preopens * sizeof(preopen));
-        free(preopens);
-        preopens = new_preopens;
-        preopen_capacity = new_capacity;
-      }
-
-      char *prefix2 = strdup(p);
-      if (prefix2 == NULL)
-        return -1;
-
-      preopens[num_preopens++] = (preopen) { prefix2, fd, };
-      free(prefix);
-
-      break;
-    }
-    default:
-      break;
-    }
-  }
-
-  return 0;
-}
-
 int wasm_start(int (main)()) {
   int result;
   void *asyncify_buf;

patches/bochs/Bochs/bochs/plugin.cc

@@ -1062,6 +1062,7 @@ plugin_t bx_builtin_plugins[] = {
 #if BX_SUPPORT_USB_XHCI
   BUILTIN_OPTPCI_PLUGIN_ENTRY(usb_xhci),
 #endif
+  BUILTIN_OPTPCI_PLUGIN_ENTRY(mapdirVirtio9p),
 #if BX_SUPPORT_SOUNDLOW
   BUILTIN_SND_PLUGIN_ENTRY(dummy),
   BUILTIN_SND_PLUGIN_ENTRY(file),

patches/bochs/Bochs/bochs/plugin.h

@@ -76,6 +76,8 @@ extern "C" {
 #define BX_PLUGIN_HPET      "hpet"
 #define BX_PLUGIN_VOODOO    "voodoo"
 
+#define BX_PLUGIN_MAPDIR_VIRTIO_9P "mapdirVirtio9p"
+
 
 #define BX_REGISTER_DEVICE_DEVMODEL(a,b,c,d) pluginRegisterDeviceDevmodel(a,b,c,d)
 #define BX_UNREGISTER_DEVICE_DEVMODEL(a,b) pluginUnregisterDeviceDevmodel(a,b)
@@ -489,6 +491,7 @@ PLUGIN_ENTRY_FOR_IMG_MODULE(vbox);
 PLUGIN_ENTRY_FOR_IMG_MODULE(vpc);
 PLUGIN_ENTRY_FOR_IMG_MODULE(vvfat);
 
+PLUGIN_ENTRY_FOR_MODULE(mapdirVirtio9p);
 #endif
 
 #ifdef __cplusplus