fix(host,sync): publish lists readable by nginx, and push the sync over a deploy key (#5)

The first publish on phishing.pezkuwichain.io copied the stage directory's
0700 mode onto the web root with rsync -a, and every list answered 403 until
the modes were put back by hand, about three minutes later. The publish script
now sets modes (--chmod=D755,F644), and test-publish.sh checks the web root's
directories are 755 and the lists 644; without --chmod it fails.

master is now protected by a ruleset (pull request, "lists" required). GitHub
refuses the Actions app as a bypass actor, so the sync's github.token push
would have been rejected. The sync now pushes over a write deploy key, the
repository's only one, kept as a secret of a "sync" environment that runs for
master only; the ruleset exempts deploy keys. GitHub's host keys are pinned
in pezkuwi/github_host_keys, and the job's token drops to contents: read.

.github/workflows/sync-upstream.yml

@@ -22,12 +22,16 @@ jobs:
     name: sync
     runs-on: ubuntu-latest
     timeout-minutes: 15
+    # master is protected (pull request, "lists" required) and only a deploy key may bypass it.
+    # The key is a secret of the "sync" environment, which runs for master only.
+    environment:
+      name: sync
     permissions:
-      contents: write
+      contents: read
       issues: write
     steps:
       # No credentials in the checkout: the sync and the checks process upstream's data, and only
-      # the commit step, below, is handed the token it pushes with.
+      # the commit step, below, is handed the key it pushes with.
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           persist-credentials: false
@@ -54,7 +58,7 @@ jobs:
       - name: Commit to master
         if: steps.sync.outputs.rc == '0'
         env:
-          GH_TOKEN: ${{ github.token }}
+          DEPLOY_KEY: ${{ secrets.SYNC_DEPLOY_KEY }}
         run: |
           set -euo pipefail
           if [ -z "$(git status --porcelain -- all.json address.json known.json all meta pezkuwi/upstream.json)" ]; then
@@ -65,8 +69,14 @@ jobs:
           git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
           git add -A -- all.json address.json known.json all meta pezkuwi/upstream.json
           git commit -q -m "chore(lists): sync with polkadot-js/phishing" -m "$(grep -E '^(SYNC |unblocked: )' sync.out)"
-          gh auth setup-git
-          git push origin HEAD:master
+          install -d -m 700 ~/.ssh
+          umask 077
+          printf '%s\n' "$DEPLOY_KEY" > ~/.ssh/sync_deploy
+          # GitHub's own host keys, committed in pezkuwi/github_host_keys: nothing learned at run time.
+          sed 's|^|github.com |' pezkuwi/github_host_keys > ~/.ssh/known_hosts
+          GIT_SSH_COMMAND="ssh -i $HOME/.ssh/sync_deploy -o IdentitiesOnly=yes -o StrictHostKeyChecking=yes" \
+            git push git@github.com:pezkuwichain/phishing.git HEAD:master
+          rm -f ~/.ssh/sync_deploy
 
       - name: Ask a person
         if: steps.sync.outputs.rc != '0'

pezkuwi/github_host_keys

@@ -0,0 +1,3 @@
+github.com ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIOMqqnkVzrm0SdG6UOoqKLsabgH5C9okWi0dh2l9GKJl
+github.com ecdsa-sha2-nistp256 AAAAE2VjZHNhLXNoYTItbmlzdHAyNTYAAAAIbmlzdHAyNTYAAABBBEmKSENjQEezOmxkZMy7opKgwFB9nkt5YRrYMjNuG5N87uRgg6CLrbo5wAdT/y6v0mKV0U2w0WZ2YB/++Tpockg=
+github.com ssh-rsa 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

pezkuwi/host/phishing-publish.sh

@@ -136,7 +136,9 @@ for domain in protected:
 print(f"checked: {len(deny)} blocked sites, {sum(len(v) for v in address.values())} scam addresses")
 PY
 
-rsync -a --checksum --delete --delay-updates --chown=root:root --exclude index.html "$stage/" "$WEB_ROOT/"
+# Modes are set, not copied: the stage is a mktemp directory, 0700, and copying its mode made the
+# web root unreadable to nginx. Every list answered 403 until it was put back (2026-09-13).
+rsync -a --checksum --delete --delay-updates --chown=root:root --chmod=D755,F644 --exclude index.html "$stage/" "$WEB_ROOT/"
 printf '%s\n' "$new" >"$STATE_DIR/published.new"
 mv -f "$STATE_DIR/published.new" "$STATE_DIR/published"
 echo "published ${new:0:12}"

pezkuwi/host/test-publish.sh

@@ -75,6 +75,9 @@ check "all.json is master's" "$(sha256sum <"$LISTS/all.json")" "$(sha256sum <"$S
 check "a shard is master's" "$(sha256sum <"$LISTS/all/com/all.json")" "$(sha256sum <"$SB/www/all/com/all.json")"
 check "the stray all/all.json is gone" no "$([ -e "$SB/www/all/all.json" ] && echo yes || echo no)"
 check "index.html untouched" "$INDEX" "$(sha256sum "$SB/www/index.html")"
+# nginx reads as another user: a web root or list it cannot read answers 403.
+check "the web root and its directories are 755" 0 "$(find "$SB/www" -type d ! -perm 755 | wc -l)"
+check "the lists are 644" 0 "$(find "$SB/www" -type f ! -name index.html ! -perm 644 | wc -l)"
 check "the published commit is recorded" "$(git -C "$SB/work" rev-parse HEAD)" "$(cat "$SB/state/published")"
 check "nothing new: exit" 0 "$(publish)"
 check "nothing new: silent" "" "$(cat "$SB/out")"