add deploy script with live verification; cache file hashes by size and mtime

4 files changed, 295 insertions, 16 deletions

diff --git a/scripts/deploy.sh b/scripts/deploy.sh
new file mode 100755
index 0000000..99129ad
--- /dev/null
+++ b/scripts/deploy.sh
@@ -0,0 +1,136 @@
+#!/usr/bin/env bash
+# Upload web/ and verify every published artifact over the public URL.
+#
+# Usage:
+#   ./scripts/deploy.sh user@host:/srv/zagan.bet https://zagan.bet/
+#   ./scripts/deploy.sh /srv/staging http://127.0.0.1:8899/
+#   ./scripts/deploy.sh --verify-only user@host:/srv/zagan.bet https://zagan.bet/
+#
+# The target may be a local directory, which is handy for staging and for
+# testing the verification itself. Old versions are left in place on purpose;
+# prune them by hand when you want to drop them.
+#
+# Environment:
+#   DEPLOY_TARGET, DEPLOY_URL   same as the two arguments
+#   SSH_PORT                    ssh port for a remote target
+set -euo pipefail
+
+ROOT=$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)
+cd "$ROOT"
+
+die() { echo "deploy: $*" >&2; exit 1; }
+
+VERIFY_ONLY=0
+if [ "${1:-}" = "--verify-only" ]; then
+  VERIFY_ONLY=1
+  shift
+fi
+
+TARGET=${1:-${DEPLOY_TARGET:-}}
+URL=${2:-${DEPLOY_URL:-}}
+if [ -z "$TARGET" ] || [ -z "$URL" ]; then
+  cat >&2 <<'EOF'
+usage: scripts/deploy.sh [--verify-only] <target> <url>
+
+  target   user@host:/srv/zagan.bet   or a local directory
+  url      public URL of the page, e.g. https://zagan.bet/
+           --verify-only skips the upload and only checks the live page
+
+environment:
+  DEPLOY_TARGET, DEPLOY_URL   same as the two arguments
+  SSH_PORT                    ssh port for a remote target
+EOF
+  exit 2
+fi
+
+for tool in rsync curl sha256sum; do
+  command -v "$tool" >/dev/null || die "$tool is required"
+done
+
+[ -f web/index.php ] || die "web/index.php is missing"
+
+shopt -s nullglob
+artifacts=(web/bin/client/zagan-bet-* web/bin/server/zagan-bet-*)
+shopt -u nullglob
+[ ${#artifacts[@]} -gt 0 ] || die "no artifacts in web/bin; run scripts/publish.sh first"
+
+if [ "$VERIFY_ONLY" -eq 0 ]; then
+  echo "==> uploading web/ to $TARGET"
+
+  # --delay-updates writes to temp names and renames at the end, so a visitor
+  # never sees a half-written binary. Old versions are kept: no --delete.
+  RSYNC_FLAGS=(
+    -a --delay-updates --human-readable --chmod=D755,F644
+    --exclude=README.md --exclude=.gitignore
+  )
+
+  if [[ "$TARGET" == *:* ]]; then
+    host=${TARGET%%:*}
+    path=${TARGET#*:}
+    ssh_args=()
+    if [ -n "${SSH_PORT:-}" ]; then
+      ssh_args+=(-p "$SSH_PORT")
+    fi
+    ssh ${ssh_args[@]+"${ssh_args[@]}"} "$host" "mkdir -p $(printf '%q' "$path")"
+    rsync "${RSYNC_FLAGS[@]}" -e "ssh ${ssh_args[*]:-}" web/ "$host:$path/"
+  else
+    mkdir -p "$TARGET"
+    rsync "${RSYNC_FLAGS[@]}" web/ "$TARGET/"
+  fi
+fi
+
+echo "==> verifying $URL"
+tmp=$(mktemp -d)
+trap 'rm -rf "$tmp"' EXIT
+
+curl -fsSL --max-time 30 "$URL" -o "$tmp/page.html" \
+  || die "could not fetch $URL"
+
+origin=$(printf '%s' "$URL" | sed -E 's#^(https?://[^/]+).*#\1#')
+grep -oE '[0-9a-f]{64}  [A-Za-z0-9._-]+' "$tmp/page.html" > "$tmp/sums" || true
+[ -s "$tmp/sums" ] || die "no SHA-256 lines on the page (is index.php being served?)"
+
+failed=0
+verified=0
+for local in "${artifacts[@]}"; do
+  [ -f "$local" ] || continue
+
+  file=$(basename "$local")
+  kind=$(basename "$(dirname "$local")")
+  local_hash=$(sha256sum "$local" | cut -d' ' -f1)
+
+  href=$(grep -oE "href=\"[^\"]*/bin/$kind/$file\"" "$tmp/page.html" | head -1 \
+    | sed 's/^href="//;s/"$//' || true)
+  if [ -z "$href" ]; then
+    echo "  $file: not listed on the page"
+    failed=1
+    continue
+  fi
+
+  shown=$(awk -v name="$file" '$2 == name { print $1 }' "$tmp/sums" | head -1)
+  if [ "$shown" != "$local_hash" ]; then
+    echo "  $file: page shows ${shown:-nothing}, local is $local_hash"
+    failed=1
+    continue
+  fi
+
+  if ! curl -fsSL --max-time 120 "$origin$href" -o "$tmp/$file"; then
+    echo "  $file: download failed"
+    failed=1
+    continue
+  fi
+
+  remote_hash=$(sha256sum "$tmp/$file" | cut -d' ' -f1)
+  if [ "$remote_hash" != "$local_hash" ]; then
+    echo "  $file: downloaded bytes differ from local"
+    failed=1
+    continue
+  fi
+
+  echo "  $file: ok"
+  verified=$((verified + 1))
+done
+
+[ "$failed" -eq 0 ] || die "verification failed"
+
+echo "==> $verified artifact(s) verified at $URL"
diff --git a/scripts/publish.sh b/scripts/publish.sh
index d8cc77a..8240859 100755
--- a/scripts/publish.sh
+++ b/scripts/publish.sh
@@ -27,7 +27,7 @@ DEST=${DEST:-web/bin}
 publish() {
   local triple=$1 kind=$2 platform=$3 arch=$4
   local ext=""
-  if [ "$kind" = client ] && [ "$platform" = windows ]; then
+  if [ "$platform" = windows ]; then
     ext=".exe"
   fi
 
@@ -51,6 +51,7 @@ publish x86_64-unknown-linux-gnu client linux x86_64
 publish i686-unknown-linux-gnu   client linux i686
 publish x86_64-pc-windows-gnu    client windows x86_64
 publish x86_64-unknown-linux-gnu server linux x86_64
+publish x86_64-pc-windows-gnu    server windows x86_64
 
 echo
 echo "published $VERSION to $DEST"
diff --git a/web/README.md b/web/README.md
index 5de180d..a82c492 100644
--- a/web/README.md
+++ b/web/README.md
@@ -58,10 +58,34 @@ php -S 127.0.0.1:8899 -t web
 
 Then open <http://127.0.0.1:8899/>.
 
+## Deploying
+
+```sh
+./scripts/publish.sh
+./scripts/deploy.sh user@host:/srv/zagan.bet https://zagan.bet/
+```
+
+`deploy.sh` rsyncs `web/` with `--delay-updates`, so a visitor never sees a
+half-written binary, then fetches the live page and every artifact it lists and
+compares the downloaded bytes with the local SHA-256. `--verify-only` skips the
+upload and only checks the live site. Old versions are left on the server;
+prune them by hand.
+
+A local directory works as a target, which is how the verification itself is
+tested:
+
+```sh
+php -S 127.0.0.1:8899 -t /srv/staging
+./scripts/deploy.sh /srv/staging http://127.0.0.1:8899/
+```
+
 ## Notes
 
-- Hashes are computed per request with `hash_file()`. That is fine for a
-  handful of builds; add an mtime-keyed cache if the directory grows.
+- Hashes are cached in a JSON file keyed on size and mtime, so a file is only
+  read once until it changes. The cache lives in the system temp directory by
+  default because the web root is often not writable by the PHP user; set
+  `ZAGAN_HASH_CACHE` to put it elsewhere. If the cache cannot be written, the
+  page simply hashes every request.
 - The page lists every build it finds, newest version first, so old releases
   stay downloadable until you remove them from `bin/`.
 - The checksum block is in `sha256sum` format, so visitors can save it as
diff --git a/web/index.php b/web/index.php
index 361fdcf..fd0a396 100644
--- a/web/index.php
+++ b/web/index.php
@@ -19,9 +19,10 @@ declare(strict_types=1);
  *   bin/client/zagan-bet-client-3.1.1-linux-x86_64
  *   bin/server/zagan-bet-server-3.1.1-linux-x86_64
  *
- * The hashes are computed per request with hash_file(). That is a few tens of
- * milliseconds for a handful of builds; add a cache keyed on mtime if the
- * directory ever grows.
+ * SHA-256 values are cached in a JSON file keyed on size and mtime, so a file
+ * is only re-read when it changes. The cache lives in the system temp
+ * directory by default (the web root is often not writable by the PHP user);
+ * set ZAGAN_HASH_CACHE to put it somewhere else.
  */
 
 const KINDS = [
@@ -68,10 +69,99 @@ function human_size(int $bytes): string
 }
 
 /**
- * @return array{version:string,platform:string,arch:string,file:string,size:int,mtime:int,sha256:string}|null
+ * Hash to use for a published file, reusing the cached value when the file has
+ * not changed since it was hashed.
+ *
+ * A cached hash is valid only when the size and mtime still match and the hash
+ * was taken at or after the file's mtime, so a restored or touched file is
+ * re-read.
+ *
+ * @param array{files:array<string,array{size:int,mtime:int,sha256:string,hashed_at:int}>} $cache
+ */
+function cached_hash(
+    string $key,
+    string $path,
+    int $size,
+    int $mtime,
+    array &$cache,
+    bool &$dirty
+): ?string {
+    $entry = $cache['files'][$key] ?? null;
+    if (
+        is_array($entry)
+        && ($entry['size'] ?? null) === $size
+        && ($entry['mtime'] ?? null) === $mtime
+        && (int) ($entry['hashed_at'] ?? 0) >= $mtime
+        && isset($entry['sha256'])
+        && is_string($entry['sha256'])
+        && preg_match('/\A[0-9a-f]{64}\z/', $entry['sha256']) === 1
+    ) {
+        return $entry['sha256'];
+    }
+
+    $hash = hash_file('sha256', $path);
+    if ($hash === false) {
+        return null;
+    }
+
+    $cache['files'][$key] = [
+        'size' => $size,
+        'mtime' => $mtime,
+        'sha256' => $hash,
+        'hashed_at' => time(),
+    ];
+    $dirty = true;
+
+    return $hash;
+}
+
+/** @return array{files:array<string,array{size:int,mtime:int,sha256:string,hashed_at:int}>} */
+function load_hash_cache(string $path): array
+{
+    $raw = @file_get_contents($path);
+    if ($raw === false || $raw === '') {
+        return ['files' => []];
+    }
+    $decoded = json_decode($raw, true);
+    if (!is_array($decoded) || !isset($decoded['files']) || !is_array($decoded['files'])) {
+        return ['files' => []];
+    }
+
+    return ['files' => $decoded['files']];
+}
+
+/**
+ * Best effort: a read-only web root, a full disk or a racing writer must never
+ * break the page, so failures are ignored and the next request recomputes.
+ *
+ * @param array{files:array<string,array{size:int,mtime:int,sha256:string,hashed_at:int}>} $cache
  */
-function read_release(string $kind, string $directory, string $entry): ?array
+function save_hash_cache(string $path, array $cache): void
 {
+    $json = json_encode($cache, JSON_UNESCAPED_SLASHES);
+    if ($json === false) {
+        return;
+    }
+    $directory = dirname($path);
+    $temporary = @tempnam($directory, '.zagan-hashes-');
+    if ($temporary === false) {
+        return;
+    }
+    if (@file_put_contents($temporary, $json, LOCK_EX) === false || !@rename($temporary, $path)) {
+        @unlink($temporary);
+    }
+}
+
+/**
+ * @return array{version:string,platform:string,arch:string,file:string,size:int,mtime:int,sha256:string}|null
+ */
+function read_release(
+    string $kind,
+    string $directory,
+    string $entry,
+    array &$cache,
+    bool &$dirty
+): ?array {
     $pattern = '/^zagan-bet-' . preg_quote($kind, '/')
         . '-v?(\d+\.\d+\.\d+)-([a-z]+)-([a-z0-9_]+)(\.exe)?$/i';
     if (preg_match($pattern, $entry, $matches) !== 1) {
@@ -92,8 +182,13 @@ function read_release(string $kind, string $directory, string $entry): ?array
     }
 
     $size = filesize($real);
-    $hash = hash_file('sha256', $real);
-    if ($size === false || $hash === false) {
+    $mtime = filemtime($real);
+    if ($size === false || $mtime === false) {
+        return null;
+    }
+
+    $hash = cached_hash($kind . '/' . $entry, $real, $size, $mtime, $cache, $dirty);
+    if ($hash === null) {
         return null;
     }
 
@@ -103,7 +198,7 @@ function read_release(string $kind, string $directory, string $entry): ?array
         'arch' => strtolower($matches[3]),
         'file' => $entry,
         'size' => $size,
-        'mtime' => filemtime($real) ?: 0,
+        'mtime' => $mtime,
         'sha256' => $hash,
     ];
 }
@@ -113,6 +208,16 @@ if ($base === '.' || $base === '/') {
     $base = '';
 }
 
+$cachePath = getenv('ZAGAN_HASH_CACHE');
+if (!is_string($cachePath) || $cachePath === '') {
+    $cachePath = rtrim(sys_get_temp_dir(), DIRECTORY_SEPARATOR)
+        . DIRECTORY_SEPARATOR . 'zagan-bet-hashes-'
+        . substr(hash('sha256', __DIR__), 0, 16) . '.json';
+}
+$cache = load_hash_cache($cachePath);
+$cacheDirty = false;
+$seen = [];
+
 $releases = [];
 $newest = 0;
 
@@ -127,10 +232,11 @@ foreach (KINDS as $kind => $meta) {
                 if ($entry === '.' || $entry === '..') {
                     continue;
                 }
-                $release = read_release($kind, $directory, $entry);
+                $release = read_release($kind, $directory, $entry, $cache, $cacheDirty);
                 if ($release === null) {
                     continue;
                 }
+                $seen[$kind . '/' . $entry] = true;
                 $release['href'] = $base . '/bin/' . $kind . '/' . rawurlencode($release['file']);
                 $found[] = $release;
                 $newest = max($newest, $release['mtime']);
@@ -156,19 +262,30 @@ foreach (KINDS as $kind => $meta) {
 
     $releases[$kind] = $found;
 }
+
+// Drop cache entries for files that are no longer published.
+foreach (array_keys($cache['files']) as $key) {
+    if (!isset($seen[$key])) {
+        unset($cache['files'][$key]);
+        $cacheDirty = true;
+    }
+}
+if ($cacheDirty) {
+    save_hash_cache($cachePath, $cache);
+}
 ?>
 <!doctype html>
 <html lang="en">
 <head>
 <meta charset="utf-8">
 <meta name="viewport" content="width=device-width, initial-scale=1">
-<title>Download · zagan.bet</title>
+<title>żagań.bet</title>
 <meta name="description" content="Official zagan.bet client and server downloads.">
 <meta http-equiv="Content-Security-Policy" content="default-src 'none'">
 </head>
 <body>
 
-<h1>zagan.bet</h1>
+<h1>żagań.bet</h1>
 <p>Official client and server builds.</p>
 <?php if ($newest > 0): ?>
 <p>Builds published <?= e(gmdate('j M Y H:i', $newest)) ?> UTC.</p>
@@ -181,7 +298,7 @@ foreach (KINDS as $kind => $meta) {
 <?php else: ?>
 <ul>
 <?php foreach ($releases['client'] as $release): ?>
-<li><a href="<?= e($release['href']) ?>" download><?= e(PLATFORMS[$release['platform']] ?? ucfirst($release['platform'])) ?> <?= e($release['arch']) ?><?php if (isset(ARCHES[$release['arch']])): ?> (<?= e(ARCHES[$release['arch']]) ?>)<?php endif; ?> — version <?= e($release['version']) ?> — <?= e(human_size($release['size'])) ?></a></li>
+<li><a href="<?= e($release['href']) ?>" download><?= e(PLATFORMS[$release['platform']] ?? ucfirst($release['platform'])) ?> <?= e($release['arch']) ?><?php if (isset(ARCHES[$release['arch']])): ?> (<?= e(ARCHES[$release['arch']]) ?>)<?php endif; ?> v<?= e($release['version']) ?> — <?= e(human_size($release['size'])) ?></a></li>
 <?php endforeach; ?>
 </ul>
 <?php endif; ?>
@@ -193,7 +310,7 @@ foreach (KINDS as $kind => $meta) {
 <?php else: ?>
 <ul>
 <?php foreach ($releases['server'] as $release): ?>
-<li><a href="<?= e($release['href']) ?>" download><?= e(PLATFORMS[$release['platform']] ?? ucfirst($release['platform'])) ?> <?= e($release['arch']) ?><?php if (isset(ARCHES[$release['arch']])): ?> (<?= e(ARCHES[$release['arch']]) ?>)<?php endif; ?> — version <?= e($release['version']) ?> — <?= e(human_size($release['size'])) ?></a></li>
+<li><a href="<?= e($release['href']) ?>" download><?= e(PLATFORMS[$release['platform']] ?? ucfirst($release['platform'])) ?> <?= e($release['arch']) ?><?php if (isset(ARCHES[$release['arch']])): ?> (<?= e(ARCHES[$release['arch']]) ?>)<?php endif; ?> v<?= e($release['version']) ?> — <?= e(human_size($release['size'])) ?></a></li>
 <?php endforeach; ?>
 </ul>
 <?php endif; ?>
@@ -222,5 +339,6 @@ echo e($sums);
 
 <p>Windows: run the <code>.exe</code> from Windows Terminal or another real terminal.</p>
 
+<p>When lost, execute <code>help</code> command inside zagan.bet client. Server binary supports <code>--help</code> flag</p>
 </body>
 </html>