buninu-linux 0.7.2 → 0.7.4

package/build-uki.sh

@@ -35,13 +35,33 @@ else
 fi
 printf '%s' "$consoles panic=0 PATH=/bin KERNEL_RELEASE=$kernel_release rdinit=/bin/bun -- -e import('/init.js')" > build/cmdline
 
+# Only this architecture's removable-media name may remain: firmware of the
+# other architecture ignores it, but run-qemu.sh reads the guest from it.
+for efi in vda/EFI/BOOT/*.EFI; do
+    [ "$efi" = "vda/EFI/BOOT/$efi_boot_name" ] || rm -f "$efi"
+done
+
+stub="$kernel_dir/$efi_stub"
+# BUNINU_JS_UKI=1 assembles the UKI with scripts/uki.js (bun or node) instead
+# of binutils; it lays the sections out by the same rules as below.
+if [ "${BUNINU_JS_UKI:-}" = 1 ]; then
+    js=$(command -v bun || command -v node) || {
+        echo "error: BUNINU_JS_UKI=1 needs bun or node on PATH" >&2
+        exit 1
+    }
+    "$js" scripts/uki.js "$stub" "vda/EFI/BOOT/$efi_boot_name" \
+        .osrel=build/os-release .cmdline=build/cmdline \
+        .linux="$kernel_image" .initrd=build/initramfs.cpio.gz
+    echo "Built $arch UKI with Alpine linux-$linux_flavor at vda/EFI/BOOT/$efi_boot_name (scripts/uki.js)"
+    exit 0
+fi
+
 if ! command -v "$uki_objcopy" >/dev/null || ! command -v "$uki_objdump" >/dev/null; then
     echo "error: build stage needs $uki_objcopy and $uki_objdump (Debian $uki_binutils_package)" >&2
     echo "run inside PRoot or with ./index.js --docker; see README Build environment" >&2
     exit 1
 fi
 
-stub="$kernel_dir/$efi_stub"
 pe_header() { "$uki_objdump" -p "$stub" | awk -v key="$1" '$1 == key { print $2; exit }'; }
 align_up() { echo $(( ($1 + $2 - 1) / $2 * $2 )); }
 max() { echo $(( $1 > $2 ? $1 : $2 )); }
@@ -65,12 +85,6 @@ cmdline_vma=$((image_base + cmdline_offset))
 linux_vma=$((image_base + linux_offset))
 initrd_vma=$((image_base + initrd_offset))
 
-# Only this architecture's removable-media name may remain: firmware of the
-# other architecture ignores it, but run-qemu.sh reads the guest from it.
-for efi in vda/EFI/BOOT/*.EFI; do
-    [ "$efi" = "vda/EFI/BOOT/$efi_boot_name" ] || rm -f "$efi"
-done
-
 "$uki_objcopy" \
     --add-section .osrel=build/os-release \
     --change-section-vma .osrel="$osrel_vma" \

package/check_build.js

@@ -0,0 +1,217 @@
+#!/usr/bin/env bun
+// Content checksums of the build outputs that ignore what changes from build
+// to build without changing what boots: file mtimes and inode numbers in the
+// initramfs, the PE timestamp and checksum of the UKI, and the GPT GUIDs,
+// FAT serial and FAT timestamps of the disk image. Two builds of the same
+// inputs print the same sums.
+//
+//   ./check_build.js                 build/initramfs.cpio.gz, the UKI(s) under vda/EFI/BOOT, vda.img
+//   ./check_build.js FILE...         any of: .cpio(.gz), PE/EFI, GPT disk image or bare FAT image
+//
+// Written against node: APIs only, so it runs under node as well as bun.
+
+import { createHash } from "node:crypto";
+import { readdirSync, readFileSync, existsSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { gunzipSync } from "node:zlib";
+
+const sha256 = (...parts) => {
+  const hasher = createHash("sha256");
+  for (const part of parts) hasher.update(part);
+  return hasher.digest("hex");
+};
+
+const u16 = (b, o) => b[o] | (b[o + 1] << 8);
+const u32 = (b, o) => (b[o] | (b[o + 1] << 8) | (b[o + 2] << 16) | (b[o + 3] << 24)) >>> 0;
+const u64 = (b, o) => u32(b, o) + u32(b, o + 4) * 2 ** 32;
+const ascii = (b, o, n) => String.fromCharCode(...b.subarray(o, o + n));
+
+// --- initramfs: newc cpio, optionally gzipped -------------------------------
+// Kept per entry: name, mode, uid, gid, device numbers of special files, size
+// and data. Dropped: inode, mtime, link count, the containing device, and
+// the order of the entries (`sort` follows the locale, and the unpacked tree
+// is the same either way): entries count sorted by name, byte by byte.
+function cpioSum(bytes) {
+  if (bytes[0] === 0x1f && bytes[1] === 0x8b) bytes = gunzipSync(bytes);
+  const entries = [];
+  let offset = 0;
+  while (offset + 110 <= bytes.length) {
+    const magic = ascii(bytes, offset, 6);
+    if (magic !== "070701" && magic !== "070702") throw new Error(`not a newc cpio entry at ${offset}`);
+    const field = (index) => parseInt(ascii(bytes, offset + 6 + index * 8, 8), 16);
+    const [mode, uid, gid, size, rdevMajor, rdevMinor, nameSize] =
+      [field(1), field(2), field(3), field(6), field(9), field(10), field(11)];
+    const name = Buffer.from(bytes.subarray(offset + 110, offset + 110 + nameSize - 1));
+    const data = (offset + 110 + nameSize + 3) & ~3;
+    const next = (data + size + 3) & ~3;
+    if (name.toString("latin1") === "TRAILER!!!") break;
+    entries.push({ name, header: `\0${mode.toString(8)} ${uid} ${gid} ${rdevMajor}:${rdevMinor} ${size}\n`, data: bytes.subarray(data, data + size) });
+    offset = next;
+  }
+  entries.sort((a, b) => Buffer.compare(a.name, b.name));
+  const hasher = createHash("sha256");
+  for (const { name, header, data } of entries) {
+    hasher.update(name);
+    hasher.update(header);
+    hasher.update(data);
+  }
+  return { sum: hasher.digest("hex"), detail: `${entries.length} entries` };
+}
+
+// --- UKI: PE image ----------------------------------------------------------
+// The whole file with the COFF TimeDateStamp and optional-header CheckSum
+// zeroed, except that the .initrd section counts by its cpio content sum.
+// The gzip output's size varies with the mtimes inside, so everything that
+// follows it is left out too: the .initrd sizes in the section table, the
+// file length (its raw data is cut out rather than zeroed), SizeOfImage and
+// SizeOfInitializedData.
+function peSum(bytes) {
+  const pe = u32(bytes, 0x3c);
+  if (ascii(bytes, pe, 4) !== "PE\0\0") throw new Error("not a PE image");
+  const coff = pe + 4;
+  const sections = u16(bytes, coff + 2);
+  const optional = coff + 20;
+  const table = optional + u16(bytes, coff + 16);
+  const copy = Uint8Array.from(bytes);
+  copy.fill(0, coff + 4, coff + 8); // TimeDateStamp
+  copy.fill(0, optional + 64, optional + 68); // CheckSum
+  let initrd = null;
+  let cut = null;
+  for (let index = 0; index < sections; index++) {
+    const header = table + index * 40;
+    const name = ascii(bytes, header, 8).replace(/\0+$/, "");
+    if (name !== ".initrd") continue;
+    const rawSize = u32(bytes, header + 16);
+    const start = u32(bytes, header + 20);
+    initrd = cpioSum(bytes.subarray(start, start + (u32(bytes, header + 8) || rawSize)));
+    cut = [start, start + rawSize];
+    copy.fill(0, header + 8, header + 12); // VirtualSize
+    copy.fill(0, header + 16, header + 20); // SizeOfRawData
+    copy.fill(0, optional + 8, optional + 12); // SizeOfInitializedData
+    copy.fill(0, optional + 56, optional + 60); // SizeOfImage
+    // Sections stored after it move with its size.
+    for (let other = 0; other < sections; other++) {
+      const otherHeader = table + other * 40;
+      if (u32(bytes, otherHeader + 20) > start) copy.fill(0, otherHeader + 20, otherHeader + 24);
+    }
+  }
+  const parts = cut ? [copy.subarray(0, cut[0]), copy.subarray(cut[1])] : [copy];
+  const sum = sha256(...parts, initrd?.sum ?? "");
+  return { sum, detail: initrd ? `.initrd ${initrd.sum.slice(0, 16)} (${initrd.detail})` : "no .initrd" };
+}
+
+// --- disk image: GPT with a FAT ESP, or a bare FAT filesystem ----------------
+// GPT: each partition's start, size and type, but not the disk or partition
+// GUIDs. FAT: the file tree (names, sizes, contents), not timestamps, serial
+// or free-space contents. .EFI files inside count by their PE sum, not their length.
+function diskSum(bytes) {
+  if (ascii(bytes, 512, 8) !== "EFI PART") return fatSum(bytes);
+  const entriesLba = u64(bytes, 512 + 72);
+  const count = u32(bytes, 512 + 80);
+  const entrySize = u32(bytes, 512 + 84);
+  const parts = [];
+  for (let index = 0; index < count; index++) {
+    const entry = entriesLba * 512 + index * entrySize;
+    const type = bytes.subarray(entry, entry + 16);
+    if (type.every((byte) => byte === 0)) continue;
+    const first = u64(bytes, entry + 32);
+    const last = u64(bytes, entry + 40);
+    const fs = fatSum(bytes.subarray(first * 512, (last + 1) * 512));
+    parts.push(`${Buffer.from(type).toString("hex")} ${first} ${last} ${fs.sum}\n`);
+  }
+  const lines = parts.map((line) => line.split(" ")[3].trim().slice(0, 16));
+  return { sum: sha256(...parts), detail: `${parts.length} partition(s), files ${lines.join(", ")}` };
+}
+
+function fatSum(bytes) {
+  const bytesPerSector = u16(bytes, 11);
+  const sectorsPerCluster = bytes[13];
+  const reserved = u16(bytes, 14);
+  const fats = bytes[16];
+  const rootEntries = u16(bytes, 17);
+  const fatSize = u16(bytes, 22) || u32(bytes, 36);
+  const totalSectors = u16(bytes, 19) || u32(bytes, 32);
+  if (bytes[510] !== 0x55 || bytes[511] !== 0xaa || !bytesPerSector) throw new Error("no FAT filesystem");
+  const rootSectors = Math.ceil((rootEntries * 32) / bytesPerSector);
+  const dataStart = (reserved + fats * fatSize + rootSectors) * bytesPerSector;
+  const clusterBytes = sectorsPerCluster * bytesPerSector;
+  const clusters = (totalSectors * bytesPerSector - dataStart) / clusterBytes;
+  const fat32 = clusters >= 65525;
+  const fat12 = clusters < 4085;
+  const fatOffset = reserved * bytesPerSector;
+  const nextCluster = (cluster) => {
+    if (fat32) return u32(bytes, fatOffset + cluster * 4) & 0x0fffffff;
+    if (!fat12) return u16(bytes, fatOffset + cluster * 2);
+    const value = u16(bytes, fatOffset + Math.floor(cluster * 1.5));
+    return cluster & 1 ? value >> 4 : value & 0xfff;
+  };
+  const end = fat32 ? 0x0ffffff8 : fat12 ? 0xff8 : 0xfff8;
+  const chain = (cluster) => {
+    const pieces = [];
+    for (; cluster >= 2 && cluster < end; cluster = nextCluster(cluster)) {
+      const start = dataStart + (cluster - 2) * clusterBytes;
+      pieces.push(bytes.subarray(start, start + clusterBytes));
+    }
+    return Buffer.concat(pieces);
+  };
+  const rootStart = (reserved + fats * fatSize) * bytesPerSector;
+  const root = fat32 ? chain(u32(bytes, 44)) : bytes.subarray(rootStart, rootStart + rootEntries * 32);
+  const files = [];
+  const walk = (directory, path) => {
+    for (let offset = 0; offset + 32 <= directory.length; offset += 32) {
+      const first = directory[offset];
+      if (first === 0) break;
+      const attributes = directory[offset + 11];
+      if (first === 0xe5 || attributes === 0x0f || attributes & 0x08) continue; // deleted, long name, label
+      const base = ascii(directory, offset, 8).trimEnd();
+      const extension = ascii(directory, offset + 8, 3).trimEnd();
+      if (base === "." || base === "..") continue;
+      const name = path + base + (extension ? "." + extension : "");
+      const cluster = (u16(directory, offset + 20) << 16) | u16(directory, offset + 26);
+      if (attributes & 0x10) {
+        files.push(`${name}/\n`);
+        walk(chain(cluster), name + "/");
+      } else {
+        const data = chain(cluster).subarray(0, u32(directory, offset + 28));
+        // A UKI's length follows its .initrd's gzip size; its PE sum covers the rest.
+        const pe = isPE(data);
+        files.push(`${name} ${pe ? "pe" : data.length} ${pe ? peSum(data).sum : sha256(data)}\n`);
+      }
+    }
+  };
+  walk(root, "/");
+  return { sum: sha256(...files), detail: files.map((line) => line.split(" ")[0].trim()).join(" ") };
+}
+
+const isPE = (bytes) => bytes[0] === 0x4d && bytes[1] === 0x5a && bytes.length > 0x40 &&
+  ascii(bytes, u32(bytes, 0x3c), 4) === "PE\0\0";
+
+function check(path) {
+  const bytes = new Uint8Array(readFileSync(path));
+  if ((bytes[0] === 0x1f && bytes[1] === 0x8b) || ascii(bytes, 0, 5) === "07070") return cpioSum(bytes);
+  if (isPE(bytes)) return peSum(bytes);
+  return diskSum(bytes);
+}
+
+const root = dirname(fileURLToPath(import.meta.url));
+let paths = process.argv.slice(2);
+if (paths.length === 0) {
+  const boot = join(root, "vda/EFI/BOOT");
+  paths = [
+    join(root, "build/initramfs.cpio.gz"),
+    ...(existsSync(boot) ? readdirSync(boot).filter((name) => /\.efi$/i.test(name)).sort().map((name) => join(boot, name)) : []),
+    join(root, "vda.img"),
+  ].filter((path) => existsSync(path));
+}
+let failed = false;
+for (const path of paths) {
+  try {
+    const { sum, detail } = check(path);
+    console.log(`${sum}  ${path}  (${detail})`);
+  } catch (error) {
+    console.error(`${path}: ${error.message}`);
+    failed = true;
+  }
+}
+process.exit(failed ? 1 : 0);

package/index.js

@@ -51,8 +51,12 @@ const stages = [
     short: "b",
     scripts: ["build-uki.sh", "build-image.sh"],
     tools: (arch) => [
-      "fakeroot", "cpio", "gzip",
-      `${architectures[arch].binutils}-objcopy`, `${architectures[arch].binutils}-objdump`,
+      // BUNINU_JS_CPIO=1: pack-initramfs.sh writes the archive with scripts/cpio.js.
+      ...(process.env.BUNINU_JS_CPIO === "1" ? [] : ["fakeroot", "cpio", "gzip"]),
+      // BUNINU_JS_UKI=1: build-uki.sh assembles the UKI with scripts/uki.js.
+      ...(process.env.BUNINU_JS_UKI === "1"
+        ? []
+        : [`${architectures[arch].binutils}-objcopy`, `${architectures[arch].binutils}-objdump`]),
       "parted", "mkfs.fat", "mmd", "mcopy",
     ],
   },
@@ -75,6 +79,7 @@ function usage() {
 Usage:
   ${pkg.name} [-f] [-b] [-r] [--arch ARCH] [--docker] [--export] [--linux-lts]
                [--real] [-- qemu arguments]
+  ${pkg.name} --check [FILE...]
 
 Stages (always run in this order, whichever you pick):
   -f, --fetch   download and verify the pinned kernel, musl, GCC runtime,
@@ -100,6 +105,12 @@ Stages (always run in this order, whichever you pick):
                 primary console, and include USB xHCI/HID keyboard modules
                 (x86_64 only)
 
+  --check [FILE...]
+                print content checksums of the build outputs that ignore
+                mtimes, timestamps and GUIDs, and exit (check_build.js; the
+                arguments after it are its own: .img, .efi or .cpio files,
+                by default the initramfs, the UKI and vda.img)
+
   -h, --help    show this
   -V, --version show name, version, runtime and platform
   --readme      render README.md in the terminal and exit
@@ -149,6 +160,8 @@ function parse(argv) {
     } else if (argument === "--readme") {
       printReadme();
       process.exit(0);
+    } else if (argument === "--check") {
+      runCheck(argv.slice(index + 1));
     } else if (argument === "--linux-lts") {
       linuxLts = true;
     } else if (argument === "--real") {
@@ -217,6 +230,14 @@ function printReadme() {
   process.stdout.write(rendered.endsWith("\n") ? rendered : rendered + "\n");
 }
 
+// --check hands everything after it to check_build.js, under the same
+// runtime, and exits with its status.
+function runCheck(args) {
+  const result = spawnSync(process.execPath, [resolve(rootDir, "check_build.js"), ...args], { stdio: "inherit" });
+  if (result.error) fail(`could not start check_build.js: ${result.error.message}`);
+  process.exit(result.status ?? 1);
+}
+
 // `command -v` without a shell: look for an executable on PATH.
 function hasTool(name) {
   for (const directory of (process.env.PATH ?? "").split(delimiter)) {
@@ -284,7 +305,7 @@ function dockerImage() {
 // into the bind-mounted checkout stays owned by that user; fakeroot provides
 // the root ownership the initramfs records.
 function dockerRun(script, env) {
-  const passEnv = ["BUNINU_ARCH", "LINUX_FLAVOR", "REAL_MACHINE", "ALPINE_MIRROR"]
+  const passEnv = ["BUNINU_ARCH", "LINUX_FLAVOR", "REAL_MACHINE", "ALPINE_MIRROR", "BUNINU_JS_UKI", "BUNINU_JS_CPIO"]
     .filter((name) => env[name] !== undefined)
     .flatMap((name) => ["--env", `${name}=${env[name]}`]);
   const user = typeof process.getuid === "function"

package/initramfs/buninu/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog
 
+## 0.4.18 - 2026-09-29
+
+- Sync jsgotty to 1.1.14: Kitty for Windows
+- Sync bunmsh to 0.3.10: Fix several posix/gnu/external builtin commands
+
 ## 0.4.17 - 2026-09-26
 
 - readme: Add homepage URL

package/initramfs/buninu/apps/bunmsh/CHANGELOG.md

@@ -2,6 +2,21 @@
 
 All notable user-visible changes to bunmsh are documented here.
 
+## [0.3.10] - 2026-09-29
+
+### Fixed
+
+- Match system utilities when handling multiple operands: `printf` repeats its
+  format for remaining arguments; `ln` links multiple sources into a directory;
+  `head`, `tail`, `wc`, and `cut` process each file separately; and `md5sum` and
+  `sha256sum` print one labelled digest per file. `wc` also prints a total for
+  multiple files. Reference tests compare these results with system commands.
+
+- Make `grep -r PATTERN` without an explicit path search the current directory,
+  as system `grep` does, instead of reading a locked standard-input stream.
+  Reference tests cover recursive matches, line numbers, no matches, and quiet
+  mode.
+
 ## [0.3.9] - 2026-09-22
 
 ### Fixed

package/initramfs/buninu/apps/bunmsh/PORTING.md

@@ -32,8 +32,12 @@ pseudo-terminal tests.
 
 ### Grammar and execution
 
-- Background jobs (`&`), job control, coprocesses, and asynchronous command
-  lists are not implemented.
+- A final, standalone, unquoted `&` starts a single external command in the
+  background (trailing comments and newlines are allowed). Stdin is ignored;
+  stdout/stderr are inherited, and the process is exposed in `Bun.sha.procs`.
+  Background pipelines, builtins, functions, and all redirections are rejected
+  before files are opened. General asynchronous command lists, job control,
+  and coprocesses are not implemented.
 - Brace command groups, `select`, arithmetic commands, and `[[ ... ]]` are
   not implemented.
 - Functions and subshells cover practical cases but do not yet reproduce every

package/initramfs/buninu/apps/bunmsh/help/README.md

@@ -959,7 +959,8 @@ Create hard or symbolic links.
 ### Usage
 
 ```sh
-ln [-sfT] TARGET LINK
+ln [-sf] TARGET LINK
+ln [-sf] TARGET ... DIRECTORY
 ```
 
 ### Options and forms
@@ -967,6 +968,7 @@ ln [-sfT] TARGET LINK
 - `-s`: Create a symbolic link.
 - `-f`: Replace an existing destination.
 - `-T`: Treat the destination as a normal path, not a directory.
+- Multiple targets may be linked into an existing directory.
 
 ### Example
 

package/initramfs/buninu/apps/bunmsh/help/ln.md

@@ -5,7 +5,8 @@ Create hard or symbolic links.
 ### Usage
 
 ```sh
-ln [-sfT] TARGET LINK
+ln [-sf] TARGET LINK
+ln [-sf] TARGET ... DIRECTORY
 ```
 
 ### Options and forms
@@ -13,6 +14,7 @@ ln [-sfT] TARGET LINK
 - `-s`: Create a symbolic link.
 - `-f`: Replace an existing destination.
 - `-T`: Treat the destination as a normal path, not a directory.
+- Multiple targets may be linked into an existing directory.
 
 ### Example
 

package/initramfs/buninu/apps/bunmsh/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bunmsh",
-  "version": "0.3.9",
+  "version": "0.3.10",
   "description": "Cross-platform command shell for Bun JavaScript",
   "license": "MIT",
   "type": "module",