v0.28.0

scripts/deploy/deploy_core.py

@@ -1,10 +1,10 @@
 #!/usr/bin/env python3
 """Install/update a local release binary on Linux/systemd. No builds.
 
-First install downloads the PDDL user-country GeoIP CSV files from the
-sapics/ip-location-db GitHub releases and verifies publisher SHA-256 files.
-Updates compare the installed GeoIP files with current publisher checksums and
-stage only stale assets before service downtime. Updates stop the service before
+First install downloads the PDDL user-country GeoIP CSV files directly from the
+sapics/ip-location-db GitHub release and validates their structure before publish.
+Updates download/validate current files before downtime, compare content SHA-256
+locally and stage only changed assets. Updates stop the service before
 copying all private data, refresh the managed systemd unit, switch an immutable
 release symlink, check health/version and restore code, data and unit on failure.
 Use --plan to inspect operations without changing the host.
@@ -35,12 +35,10 @@ ROOT = Path(__file__).resolve().parents[2]
 UNIT_PATH = Path('/etc/systemd/system/rustproxy.service')
 SHELL_PATH_PROFILE = Path('/etc/profile.d/rustproxy-path.sh')
 SERVICE = 'rustproxy.service'
-GEOIP_ASSETS = (
-    'user-country-ipv4.csv',
-    'user-country-ipv6.csv',
-)
-GEOIP_RELEASE = 'https://github.com/sapics/ip-location-db/releases/download/latest/'
-GEOIP_CHECKSUM = 'https://github.com/sapics/ip-location-db/releases/download/checksum/'
+GEOIP_SOURCES = {
+    'user-country-ipv4.csv': 'https://github.com/sapics/ip-location-db/releases/download/latest/user-country-ipv4.csv',
+    'user-country-ipv6.csv': 'https://github.com/sapics/ip-location-db/releases/download/latest/user-country-ipv6.csv',
+}
 GEOIP_MAX_BYTES = 256 * 1024 * 1024
 
 
@@ -97,7 +95,7 @@ def digest(path: Path) -> str:
 def fetch_https(url: str, max_bytes: int) -> bytes:
     if urllib.parse.urlparse(url).scheme != 'https':
         raise ValueError('GeoIP downloads require HTTPS.')
-    request = urllib.request.Request(url, headers={'User-Agent': 'RustProxy-installer/0.27.3'})
+    request = urllib.request.Request(url, headers={'User-Agent': 'RustProxy-installer/0.28.0'})
     with urllib.request.urlopen(request, timeout=30) as response:
         if urllib.parse.urlparse(response.geturl()).scheme != 'https':
             raise ValueError('GeoIP download redirected away from HTTPS.')
@@ -115,15 +113,65 @@ def fetch_https(url: str, max_bytes: int) -> bytes:
         return bytes(result)
 
 
-def publisher_geoip_checksums() -> dict[str, str]:
-    expected: dict[str, str] = {}
-    for name in GEOIP_ASSETS:
-        checksum = fetch_https(GEOIP_CHECKSUM + name + '.sha256', 4096).decode('ascii', 'strict').strip()
-        matched = re.search(r'\b([0-9a-fA-F]{64})\b', checksum)
-        if not matched:
-            raise ValueError(f'Invalid publisher checksum for {name}.')
-        expected[name] = matched.group(1).lower()
-    return expected
+def acquire_geoip_update_lock(data: Path, uid: int | None = None, gid: int | None = None) -> int:
+    """Serialize installer/manual GeoIP refresh with the in-process worker."""
+    path = data / '.geoip-update.lock'
+    flags = os.O_RDWR | os.O_CREAT
+    if hasattr(os, 'O_NOFOLLOW'):
+        flags |= os.O_NOFOLLOW
+    fd = os.open(path, flags, 0o600)
+    try:
+        os.fchmod(fd, 0o600)
+        if uid is not None and gid is not None:
+            os.fchown(fd, uid, gid)
+        fcntl.flock(fd, fcntl.LOCK_EX)
+        return fd
+    except BaseException:
+        os.close(fd)
+        raise
+
+
+def release_geoip_update_lock(fd: int | None) -> None:
+    if fd is None:
+        return
+    try:
+        fcntl.flock(fd, fcntl.LOCK_UN)
+    finally:
+        os.close(fd)
+
+
+def validate_geoip_csv(name: str, payload: bytes) -> None:
+    """Reject truncated/error payloads before a GeoIP CSV is published."""
+    family = 4 if name == 'user-country-ipv4.csv' else 6
+    try:
+        text = payload.decode('utf-8', 'strict')
+    except UnicodeDecodeError as error:
+        raise ValueError(f'GeoIP {name} is not valid UTF-8.') from error
+    previous_end: int | None = None
+    rows = 0
+    for number, raw in enumerate(text.splitlines(), 1):
+        line = raw.strip()
+        if not line or line.startswith('#'):
+            continue
+        fields = [value.strip().strip('"') for value in line.split(',')]
+        if len(fields) != 3:
+            raise ValueError(f'GeoIP {name} row {number} must contain exactly 3 columns.')
+        try:
+            start = ipaddress.ip_address(fields[0])
+            end = ipaddress.ip_address(fields[1])
+        except ValueError as error:
+            raise ValueError(f'GeoIP {name} row {number} contains an invalid IP address.') from error
+        if start.version != family or end.version != family or int(start) > int(end):
+            raise ValueError(f'GeoIP {name} row {number} contains an invalid address range.')
+        if previous_end is not None and int(start) <= previous_end:
+            raise ValueError(f'GeoIP {name} ranges overlap or are unsorted at row {number}.')
+        country = fields[2]
+        if country and not re.fullmatch(r'[A-Za-z]{2}', country):
+            raise ValueError(f'GeoIP {name} row {number} contains an invalid country code.')
+        previous_end = int(end)
+        rows += 1
+    if rows == 0:
+        raise ValueError(f'GeoIP {name} does not contain any address ranges.')
 
 
 def geoip_file_digest(path: Path) -> str | None:
@@ -137,29 +185,31 @@ def geoip_file_digest(path: Path) -> str | None:
 
 
 def prepare_geoip_update(data: Path, prefix: Path) -> tuple[Path | None, list[str]]:
-    """Verify local GeoIP files against current publisher checksums.
+    """Download and validate current GeoIP CSVs before service downtime.
 
-    Network/download work happens before the service is stopped. Changed files are
-    staged under the code prefix and are copied into the private data transaction
-    only after its backup exists.
+    Publisher checksums are intentionally not required here: the upstream checksum
+    release can move independently from the `latest` data release. Local SHA-256 is
+    used only to decide whether the validated payload changes an installed file.
     """
-    expected = publisher_geoip_checksums()
     directory = data / 'geoip'
     if directory.exists() and directory.is_symlink():
         raise ValueError('GeoIP directory cannot be a symlink.')
-    stale = [name for name in GEOIP_ASSETS if geoip_file_digest(directory / name) != expected[name]]
+    payloads: dict[str, bytes] = {}
+    stale: list[str] = []
+    for name, url in GEOIP_SOURCES.items():
+        payload = fetch_https(url, GEOIP_MAX_BYTES)
+        validate_geoip_csv(name, payload)
+        payloads[name] = payload
+        if geoip_file_digest(directory / name) != hashlib.sha256(payload).hexdigest():
+            stale.append(name)
     if not stale:
-        print('GeoIP database is current (publisher SHA-256 matches).')
+        print('GeoIP database is current (validated content is unchanged).')
         return None, []
     staging = prefix / ('.geoip-stage-' + uuid.uuid4().hex)
     staging.mkdir(mode=0o700)
     try:
         for name in stale:
-            payload = fetch_https(GEOIP_RELEASE + name, GEOIP_MAX_BYTES)
-            actual = hashlib.sha256(payload).hexdigest()
-            if actual != expected[name]:
-                raise ValueError(f'GeoIP SHA-256 mismatch for {name}.')
-            write_private(staging / name, payload, 0o600)
+            write_private(staging / name, payloads[name], 0o600)
         print('GeoIP update prepared: ' + ', '.join(stale))
         return staging, stale
     except BaseException:
@@ -204,20 +254,17 @@ def apply_geoip_update(data: Path, staging: Path | None, names: list[str], uid:
 
 
 def download_geoip(data: Path, uid: int, gid: int) -> None:
-    # First install has no code-prefix staging transaction, so download and verify
-    # each asset directly before publishing it atomically into the private data dir.
-    expected = publisher_geoip_checksums()
+    # First install has no code-prefix staging transaction. Validate each direct
+    # payload before publishing it atomically into the private data directory.
     directory = data / 'geoip'
     if directory.exists() and directory.is_symlink():
         raise ValueError('GeoIP directory cannot be a symlink.')
     directory.mkdir(mode=0o700, exist_ok=True)
     os.chmod(directory, 0o700)
     os.chown(directory, uid, gid)
-    for name in GEOIP_ASSETS:
-        payload = fetch_https(GEOIP_RELEASE + name, GEOIP_MAX_BYTES)
-        actual = hashlib.sha256(payload).hexdigest()
-        if actual != expected[name]:
-            raise ValueError(f'GeoIP SHA-256 mismatch for {name}.')
+    for name, url in GEOIP_SOURCES.items():
+        payload = fetch_https(url, GEOIP_MAX_BYTES)
+        validate_geoip_csv(name, payload)
         target = directory / name
         write_private(target, payload, 0o600)
         os.chown(target, uid, gid)

scripts/deploy/deploy_runtime.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 from deploy_core import *  # re-exported internal deployment primitives
 
 ACME_RUNTIME_NAMES = frozenset({'acme-python', 'acme-python.installing', 'acme-python.virtualenv-bootstrap'})
+RUNTIME_STATE_NAMES = frozenset({'.geoip-update.lock'})
 CERTBOT_PACKAGES = ('certbot==5.8.0', 'certbot-dns-cloudflare==5.8.0', 'certbot-dns-ovh==5.8.0', 'certbot-dns-rfc2136==5.8.0')
 
 def _under_acme_runtime(source: Path, path: Path) -> bool:
@@ -30,7 +31,9 @@ def copy_data(source: Path, target: Path) -> None:
     checked_tree(source)
     root = source.resolve(strict=True)
     def ignore(directory: str, names: list[str]) -> set[str]:
-        return set(names) & ACME_RUNTIME_NAMES if Path(directory).resolve(strict=True) == root else set()
+        if Path(directory).resolve(strict=True) != root:
+            return set()
+        return set(names) & (ACME_RUNTIME_NAMES | RUNTIME_STATE_NAMES)
     shutil.copytree(source, target, symlinks=True, ignore=ignore)
     os.chmod(target, 0o700)
 
@@ -353,64 +356,68 @@ def stop_for_managed_update(data: Path) -> None:
 
 def update(record: dict, binary: Path, sha: str, start: bool, refresh_geoip: bool = False, reset_config: bool = False) -> dict:
     prefix = Path(record['prefix']); data = Path(record['data_dir']); account = pwd.getpwnam(record['user'])
-    # Updates deliberately do not rewrite shell environment. Check the installer-
-    # managed PATH profile early and report the exact remediation target instead.
-    check_shell_path(prefix)
-    # Do this before staging/network work and, critically, before stopping the
-    # currently healthy service. It also repairs stale root:root/0700 metadata
-    # left by manual maintenance while preserving private 0700 semantics.
-    ensure_service_data_access(data, account)
-    release, version = stage(binary, prefix, record['user'], sha)
+    geoip_lock_fd: int | None = None
     geoip_stage: Path | None = None
-    geoip_names: list[str] = []
-    if refresh_geoip:
-        geoip_stage, geoip_names = prepare_geoip_update(data, prefix)
-    active = subprocess.run(['systemctl', 'is-active', '--quiet', SERVICE], check=False, stdout=subprocess.DEVNULL).returncode == 0
-    stop_for_managed_update(data)
     try:
-        backup = backup_stopped(record, prefix)
-    except BaseException:
-        if active: run(['systemctl', 'start', SERVICE])
-        if geoip_stage is not None:
-            shutil.rmtree(geoip_stage, ignore_errors=True)
-        raise
-    replacement = dict(record, release=release, version=version, sha256=sha)
-    try:
-        if reset_config:
-            reset_saved_config(data)
-        apply_geoip_update(data, geoip_stage, geoip_names, account.pw_uid, account.pw_gid)
-        prepare_managed_acme_runtime(data, account)
-        switch(prefix, release)
-        write_private(UNIT_PATH, service_unit(replacement).encode(), 0o644)
-        run(['systemctl', 'daemon-reload'])
-        if start:
-            run(['systemctl', 'start', SERVICE])
-            if not healthy(replacement, version):
-                raise RuntimeError('New release failed the local health/version check.')
-        write_private(prefix / 'install.json', json.dumps(replacement, indent=2).encode())
-    except BaseException as error:
-        details = service_failure_details()
-        if details:
-            print('New release diagnostics:\n' + details, file=sys.stderr)
+        # Updates deliberately do not rewrite shell environment. Check the installer-
+        # managed PATH profile early and report the exact remediation target instead.
+        check_shell_path(prefix)
+        # Repair stale root:root/0700 metadata before staging/network work and before
+        # stopping the currently healthy service.
+        ensure_service_data_access(data, account)
+        release, version = stage(binary, prefix, record['user'], sha)
+        geoip_names: list[str] = []
+        if refresh_geoip:
+            # Keep the same lock for checksum/download staging and publication. The
+            # runtime worker uses this file too, so manual upgrades cannot race it.
+            geoip_lock_fd = acquire_geoip_update_lock(data, account.pw_uid, account.pw_gid)
+            geoip_stage, geoip_names = prepare_geoip_update(data, prefix)
+        active = subprocess.run(['systemctl', 'is-active', '--quiet', SERVICE], check=False, stdout=subprocess.DEVNULL).returncode == 0
         stop_for_managed_update(data)
-        failed = data.with_name(data.name + '.failed-' + uuid.uuid4().hex[:12])
-        data.rename(failed)  # keep the failed generation for diagnosis; do not delete it
-        copy_data(backup / 'data', data)
-        restore_managed_acme_runtime(failed, data)
-        chown_tree(data, account.pw_uid, account.pw_gid)
-        switch(prefix, record['release'])
-        restore_service_unit(backup)
-        write_private(prefix / 'install.json', json.dumps(record, indent=2).encode())
-        if active:
-            run(['systemctl', 'start', SERVICE])
-            if not healthy(record, record['version']):
-                raise RuntimeError(f'Update failed: {error}. Rollback restored code/data/unit but previous service health also failed. Backup: {backup}; failed state: {failed}') from error
-        raise RuntimeError(f'Update failed: {error}. Previous code, private data and service unit were restored. Installation is still present; do not run install.sh. Fix the cause and rerun update.sh. Backup: {backup}; failed state: {failed}') from error
+        try:
+            backup = backup_stopped(record, prefix)
+        except BaseException:
+            if active:
+                run(['systemctl', 'start', SERVICE])
+            raise
+        replacement = dict(record, release=release, version=version, sha256=sha)
+        try:
+            if reset_config:
+                reset_saved_config(data)
+            apply_geoip_update(data, geoip_stage, geoip_names, account.pw_uid, account.pw_gid)
+            prepare_managed_acme_runtime(data, account)
+            switch(prefix, release)
+            write_private(UNIT_PATH, service_unit(replacement).encode(), 0o644)
+            run(['systemctl', 'daemon-reload'])
+            if start:
+                run(['systemctl', 'start', SERVICE])
+                if not healthy(replacement, version):
+                    raise RuntimeError('New release failed the local health/version check.')
+            write_private(prefix / 'install.json', json.dumps(replacement, indent=2).encode())
+        except BaseException as error:
+            details = service_failure_details()
+            if details:
+                print('New release diagnostics:\n' + details, file=sys.stderr)
+            stop_for_managed_update(data)
+            failed = data.with_name(data.name + '.failed-' + uuid.uuid4().hex[:12])
+            data.rename(failed)  # keep the failed generation for diagnosis; do not delete it
+            copy_data(backup / 'data', data)
+            restore_managed_acme_runtime(failed, data)
+            chown_tree(data, account.pw_uid, account.pw_gid)
+            switch(prefix, record['release'])
+            restore_service_unit(backup)
+            write_private(prefix / 'install.json', json.dumps(record, indent=2).encode())
+            if active:
+                run(['systemctl', 'start', SERVICE])
+                if not healthy(record, record['version']):
+                    raise RuntimeError(f'Update failed: {error}. Rollback restored code/data/unit but previous service health also failed. Backup: {backup}; failed state: {failed}') from error
+            raise RuntimeError(f'Update failed: {error}. Previous code, private data and service unit were restored. Installation is still present; do not run install.sh. Fix the cause and rerun update.sh. Backup: {backup}; failed state: {failed}') from error
+        print(f'Backup (contains private keys): {backup}')
+        return replacement
     finally:
         if geoip_stage is not None:
             shutil.rmtree(geoip_stage, ignore_errors=True)
-    print(f'Backup (contains private keys): {backup}')
-    return replacement
+        release_geoip_update_lock(geoip_lock_fd)
 
 
 def install(record: dict, binary: Path, sha: str, start: bool, install_geoip: bool = False) -> dict:

scripts/deploy/upgrade.py

@@ -46,7 +46,7 @@ def upgrade(source: Path, target: Path, *, stopped: bool, data_dir: Path | None
         for name in MANAGED:
             if (target / name).is_symlink():
                 raise ValueError(f"Managed target {name} is a symlink; refusing to replace it.")
-        backup = target / ".upgrade-backup" / ("before-0.27.3-" + uuid.uuid4().hex[:12])
+        backup = target / ".upgrade-backup" / ("before-0.28.0-" + uuid.uuid4().hex[:12])
         backup.mkdir(parents=True, mode=0o700)
         installed: list[str] = []
         moved: list[str] = []
@@ -84,7 +84,7 @@ def upgrade(source: Path, target: Path, *, stopped: bool, data_dir: Path | None
             remove_legacy = legacy.exists()
             if remove_legacy and (legacy.is_symlink() or not legacy.is_file()):
                 raise ValueError("Obsolete renewal observation sidecar must be a regular file.")
-            (backup / "upgrade.json").write_text(json.dumps({"from_directory":str(target), "to_version":"0.27.3", "managed_paths":MANAGED, "removed_obsolete_runtime_state": remove_legacy}, indent=2)+"\n", encoding="utf-8")
+            (backup / "upgrade.json").write_text(json.dumps({"from_directory":str(target), "to_version":"0.28.0", "managed_paths":MANAGED, "removed_obsolete_runtime_state": remove_legacy}, indent=2)+"\n", encoding="utf-8")
             if remove_legacy:
                 legacy.unlink()
         except BaseException:
@@ -115,7 +115,7 @@ def main() -> None:
         backup = upgrade(Path(__file__).resolve().parents[2], args.target, stopped=args.stopped, data_dir=data_dir)
     except (OSError, ValueError) as error:
         parser.exit(2, f"Upgrade refused or rolled back: {error}\n")
-    print(f"Installed RustProxy 0.27.3. Previous managed source files: {backup}")
+    print(f"Installed RustProxy 0.28.0. Previous managed source files: {backup}")
     print("Preserved data, .env, .git, target and other unmanaged files. No database migration was run.")
     print(f"Next: cd {args.target}; ./dev.sh test; ./dev.sh")