software/theia: Fix resilient signature checking

software/theia/buildout.hash.cfg

@@ -35,15 +35,15 @@ md5sum = d78a9f885bdebf6720197209e0c21aa0
 
 [theia-common]
 _update_hash_filename_ = theia_common.py
-md5sum = 38eba0fb605953677b5a2a2e686a66a2
+md5sum = 6a25c6a7f1beb27232a3c9acd8a76500
 
 [theia-export]
 _update_hash_filename_ = theia_export.py
-md5sum = d338b2d3ba1dcd7a919c0baf51dd11da
+md5sum = e2f6c483cce09f87ab1e63ae8be0daf4
 
 [theia-import]
 _update_hash_filename_ = theia_import.py
-md5sum = 3580f1eec1099bebd550ba1eacd9c116
+md5sum = 5dea99b0106cccba65f8ae90d110e111
 
 [yarn.lock]
 _update_hash_filename_ = yarn.lock

software/theia/theia_common.py

@@ -81,7 +81,6 @@ def remove(path):
 
 def parse_installed(partition):
   paths = []
-  custom_script = os.path.join(partition, 'srv', '.backup_identity_script')
   for cfg in glob.glob(os.path.join(partition, '.installed*.cfg')):
     try:
       with open(cfg) as f:
@@ -93,7 +92,7 @@ def parse_installed(partition):
       for section in six.itervalues(installed_cfg):
         for p in section.get('__buildout_installed__', '').splitlines():
           p = p.strip()
-          if p and p != custom_script:
+          if p:
             paths.append(p)
   return paths
 
@@ -108,31 +107,44 @@ def sha256sum(file_path, chunk_size=1024 * 1024):
   return sha256.hexdigest()
 
 
-def hashwalk(backup_dir, mirror_partitions):
-  scripts = {}
-  for p in mirror_partitions:
-    script_path = os.path.join(p, 'srv', '.backup_identity_script')
-    if os.path.exists(script_path):
-      scripts[os.path.abspath(p)] = script_path
-  for dirpath, dirnames, filenames in os.walk(backup_dir):
-    filenames.sort()
+def fast_hashwalk(root_dir):
+  for dirpath, dirnames, filenames in os.walk(root_dir):
     for f in filenames:
       filepath = os.path.join(dirpath, f)
       if os.path.isfile(filepath):
-        displaypath = os.path.relpath(filepath, start=backup_dir)
+        displaypath = os.path.relpath(filepath, start=root_dir)
         yield '%s %s' % (sha256sum(filepath), displaypath)
-    remaining_dirnames = []
-    for subdir in dirnames:
-      subdirpath = os.path.abspath(os.path.join(dirpath, subdir))
-      custom_hashscript = scripts.get(subdirpath)
-      if custom_hashscript:
-        print('Using custom signature script %s' % custom_hashscript)
-        for s in hashcustom(subdirpath, backup_dir, custom_hashscript):
-          yield s
-      else:
-        remaining_dirnames.append(subdir)
-    remaining_dirnames.sort()
-    dirnames[:] = remaining_dirnames
+
+
+def exclude_hashwalk(root_dir, instance_dir):
+  root_dir = os.path.abspath(root_dir)
+  instance_dir = os.path.abspath(instance_dir)
+  for dirpath, dirnames, filenames in os.walk(root_dir):
+    for f in filenames:
+      filepath = os.path.join(dirpath, f)
+      if os.path.isfile(filepath):
+        displaypath = os.path.relpath(filepath, start=root_dir)
+        yield '%s %s' % (sha256sum(filepath), displaypath)
+    if dirpath == instance_dir:
+      remaining_dirs = []
+      for d in dirnames:
+        if not d.startswith('slappart'):
+          remaining_dirs.append(d)
+      dirnames[:] = remaining_dirs
+
+
+def hashwalk(root_dir, instance_dir=None):
+  if instance_dir and not os.path.relpath(
+      instance_dir, start=root_dir).startswith(os.pardir):
+    return exclude_hashwalk(root_dir, instance_dir)
+  return fast_hashwalk(root_dir)
+
+
+def hashscript(partition):
+  script = os.path.join(partition, 'srv', '.backup_identity_script')
+  if os.path.exists(script):
+    return script
+  return None
 
 
 @contextlib.contextmanager
@@ -145,10 +157,11 @@ def cwd(path):
     os.chdir(old_path)
 
 
-def hashcustom(mirrordir, backup_dir, custom_hashscript):
-  workingdir = os.path.join(mirrordir, os.pardir, os.pardir, os.pardir)
+def hashcustom(partition, script):
+  workingdir = os.path.join(partition, os.pardir, os.pardir, os.pardir)
   with cwd(os.path.abspath(workingdir)):
-    for dirpath, _, filenames in os.walk(mirrordir):
+    for dirpath, dirnames, filenames in os.walk(partition):
+      dirnames.sort()
       filepaths = []
       for f in filenames:
         path = os.path.join(dirpath, f)
@@ -157,16 +170,16 @@ def hashcustom(mirrordir, backup_dir, custom_hashscript):
       if not filepaths:
         continue
       hashprocess = sp.Popen(
-        custom_hashscript, stdin=sp.PIPE, stdout=sp.PIPE, stderr=sp.PIPE)
+        script, stdin=sp.PIPE, stdout=sp.PIPE, stderr=sp.PIPE)
       out, err = hashprocess.communicate(str2bytes('\0'.join(filepaths)))
       if hashprocess.returncode != 0:
         template = "Custom signature script %s failed on inputs:\n%s"
-        msg = template % (custom_hashscript, '\n'.join(filepaths))
+        msg = template % (script, '\n'.join(filepaths))
         msg += "\nwith stdout:\n%s" % bytes2str(out)
         msg += "\nand stderr:\n%s" % bytes2str(err)
         raise Exception(msg)
       signatures = bytes2str(out).strip('\n').split('\n')
       signatures.sort()
-      displaypath = os.path.relpath(dirpath, start=backup_dir)
+      displaypath = os.path.relpath(dirpath, start=partition)
       for s in signatures:
-        yield '%s %s/ (custom)' % (s, displaypath)
+        yield '%s %s' % (s, displaypath)

software/theia/theia_export.py

@@ -55,49 +55,74 @@ class TheiaExport(object):
     self.copytree_partitions_args = {}
     self.logs = []
 
-  def mirrorpath(self, src):
+  def mirror_path(self, src):
     return os.path.abspath(os.path.join(
       self.backup_dir, os.path.relpath(src, start=self.root_dir)))
 
-  def backuptree(self, src, exclude=(), extrargs=(), verbosity='-v'):
-    dst = self.mirrorpath(src)
-    return copytree(self.rsync_bin, src, dst, exclude, extrargs, verbosity)
+  def backup_tree(self, src):
+    return copytree(self.rsync_bin, src, self.mirror_path(src))
 
-  def backupfile(self, src):
-    dst = self.mirrorpath(src)
-    return copyfile(src, dst)
+  def backup_file(self, src):
+    return copyfile(src, self.mirror_path(src))
 
-  def backupdb(self):
-    copydb(self.sqlite3_bin, self.proxy_db, self.mirrorpath(self.proxy_db))
+  def backup_db(self):
+    copydb(self.sqlite3_bin, self.proxy_db, self.mirror_path(self.proxy_db))
 
-  def backuppartition(self, partition):
+  def backup_partition(self, partition):
     installed = parse_installed(partition)
     rules = os.path.join(partition, 'srv', 'exporter.exclude')
     extrargs = ('--filter=.-/ ' + rules,) if os.path.exists(rules) else ()
-    self.backuptree(partition, exclude=installed, extrargs=extrargs)
-    self.copytree_partitions_args[partition] = (installed, extrargs)
+    dst = self.mirror_path(partition)
+    copytree(self.rsync_bin, partition, dst, installed, extrargs)
+    self.copytree_partitions_args[partition] = (dst, installed, extrargs)
 
-  def sign(self, signaturefile):
+  def sign(self, signaturefile, signatures):
     remove(signaturefile)
     pardir = os.path.abspath(os.path.join(self.backup_dir, os.pardir))
-    tmpfile = os.path.join(pardir, 'backup.signature.tmp')
-    mirror_partitions = [self.mirrorpath(p) for p in self.partition_dirs]
+    tmpfile = os.path.join(pardir, os.path.basename(signaturefile) + '.tmp')
     with open(tmpfile, 'w') as f:
-      for s in hashwalk(self.backup_dir, mirror_partitions):
+      for s in signatures:
         f.write(s + '\n')
     os.rename(tmpfile, signaturefile)
 
-  def checkpartition(self, partition, pattern='/srv/backup/'):
-    installed, extrargs = self.copytree_partitions_args[partition]
-    output = self.backuptree(
+  def sign_root(self):
+    signaturefile = os.path.join(self.backup_dir, 'backup.signature')
+    signatures = hashwalk(self.backup_dir, self.mirror_path(self.instance_dir))
+    self.sign(signaturefile, signatures)
+
+  def sign_partition(self, partition):
+    dst = self.mirror_path(partition)
+    filename = os.path.basename(partition) + '.backup.signature'
+    signaturefile = os.path.join(self.backup_dir, filename)
+    script = hashscript(partition)
+    if script:
+      signaturefile += '.custom'
+      self.sign(signaturefile, hashcustom(dst, script))
+    else:
+      self.sign(signaturefile, hashwalk(dst))
+
+  def remove_signatures(self):
+    pattern = os.path.join(self.backup_dir, '*backup.signature*')
+    signature_files = glob.glob(pattern)
+    for f in signature_files:
+      try:
+        os.remove(f)
+      except OSError:
+        pass
+
+  def check_partition(self, partition, pattern='/srv/backup/'):
+    dst, installed, extrargs = self.copytree_partitions_args[partition]
+    output = copytree(
+      self.rsync_bin,
       partition,
+      dst,
       exclude=installed,
       extrargs=extrargs + ('--dry-run', '--update'),
       verbosity='--out-format=%n',
     )
     return [path for path in output.splitlines() if pattern in path]
 
-  def loginfo(self, msg):
+  def log(self, msg):
     print(msg)
     self.logs.append(msg)
 
@@ -126,36 +151,42 @@ class TheiaExport(object):
     with open(timestamp, 'w') as f:
       f.write(str(export_start_date))
 
-    self.loginfo('Backup resilient timestamp ' + timestamp)
-    self.backupfile(timestamp)
+    self.remove_signatures()
+
+    self.log('Backup resilient timestamp ' + timestamp)
+    self.backup_file(timestamp)
 
     for d in self.dirs:
-      self.loginfo('Backup directory ' + d)
-      self.backuptree(d)
+      self.log('Backup directory ' + d)
+      self.backup_tree(d)
 
-    self.loginfo('Backup slapproxy database')
-    self.backupdb()
+    self.log('Backup slapproxy database')
+    self.backup_db()
 
-    self.loginfo('Backup partitions')
+    self.log('Backup partitions')
     for p in self.partition_dirs:
-      self.backuppartition(p)
+      self.backup_partition(p)
 
-    self.loginfo('Compute backup signature')
-    self.sign(os.path.join(self.backup_dir, 'backup.signature'))
+    self.log('Compute root backup signature')
+    self.sign_root()
+
+    self.log('Compute partitions backup signatures')
+    for p in self.partition_dirs:
+      self.sign_partition(p)
 
     time.sleep(10)
-    self.loginfo('Check partitions')
+    self.log('Check partitions')
     modified = list(itertools.chain.from_iterable(
-      self.checkpartition(p) for p in self.partition_dirs))
+      self.check_partition(p) for p in self.partition_dirs))
     if modified:
       msg = 'Some files have been modified since the backup started'
-      self.loginfo(msg + ':')
-      self.loginfo('\n'.join(modified))
-      self.loginfo("Let's wait %d minutes and try again" % BACKUP_WAIT)
+      self.log(msg + ':')
+      self.log('\n'.join(modified))
+      self.log("Let's wait %d minutes and try again" % BACKUP_WAIT)
       time.sleep(BACKUP_WAIT * 60)
       raise Exception(msg)
 
-    self.loginfo('Done')
+    self.log('Done')
 
 
 if __name__ == '__main__':

software/theia/theia_import.py

@@ -57,32 +57,32 @@ class TheiaImport(object):
     configp.read(cfg)
     self.proxy_db = configp.get('slapproxy', 'database_uri')
     self.instance_dir = configp.get('slapos', 'instance_root')
-    mirror_dir = self.mirrorpath(self.instance_dir)
+    mirror_dir = self.mirror_path(self.instance_dir)
     partitions = glob.glob(os.path.join(mirror_dir, 'slappart*'))
     self.mirror_partition_dirs = [p for p in partitions if os.path.isdir(p)]
     self.logs = []
 
-  def mirrorpath(self, dst):
+  def mirror_path(self, dst):
     return os.path.abspath(os.path.join(
       self.backup_dir, os.path.relpath(dst, start=self.root_dir)))
 
-  def dstpath(self, src):
+  def dst_path(self, src):
     return os.path.abspath(os.path.join(
       self.root_dir, os.path.relpath(src, start=self.backup_dir)))
 
-  def restoretree(self, dst, exclude=(), extrargs=(), verbosity='-v'):
-    src = self.mirrorpath(dst)
+  def restore_tree(self, dst, exclude=(), extrargs=(), verbosity='-v'):
+    src = self.mirror_path(dst)
     return copytree(self.rsync_bin, src, dst, exclude, extrargs, verbosity)
 
-  def restorefile(self, dst):
-    src = self.mirrorpath(dst)
+  def restore_file(self, dst):
+    src = self.mirror_path(dst)
     return copyfile(src, dst)
 
-  def restoredb(self):
-    copydb(self.sqlite3_bin, self.mirrorpath(self.proxy_db), self.proxy_db)
+  def restore_db(self):
+    copydb(self.sqlite3_bin, self.mirror_path(self.proxy_db), self.proxy_db)
 
-  def restorepartition(self, mirror_partition):
-    p = self.dstpath(mirror_partition)
+  def restore_partition(self, mirror_partition):
+    p = self.dst_path(mirror_partition)
     installed = parse_installed(p) if os.path.exists(p) else []
     copytree(self.rsync_bin, mirror_partition, p, exclude=installed)
 
@@ -97,31 +97,60 @@ class TheiaImport(object):
     print(' '.join(command))
     sp.check_call(command)
 
-  def verify(self, signaturefile):
-    pardir = os.path.abspath(os.path.join(self.backup_dir, os.pardir))
-    moved = os.path.join(pardir, 'backup.signature.moved')
-    proof = os.path.join(pardir, 'backup.signature.proof')
-    if os.path.exists(signaturefile):
-      os.rename(signaturefile, moved)
-    if not os.path.exists(moved):
-      msg = 'ERROR the backup signature file is missing'
-      print(msg)
+  def sign(self, signaturefile, root_dir):
+    with open(signaturefile, 'r') as f:
+      for line in f:
+        try:
+          _, relpath = line.strip().split(None, 1)
+        except ValueError:
+          yield 'Could not parse: %s' % line
+          continue
+        filepath = os.path.join(root_dir, relpath)
+        try:
+          signature = sha256sum(filepath)
+        except IOError:
+          yield 'Could not read: %s' % filepath
+          continue
+        yield '%s %s' % (signature, relpath)
+
+  def sign_custom(self, root_dir):
+    partition = self.dst_path(root_dir)
+    script = hashscript(partition)
+    if not script:
+      msg = 'ERROR: missing custom signature script for partition ' + partition
       raise Exception(msg)
+    return hashcustom(root_dir, script)
+
+  def find_signature_file(self, partition):
+    filename = os.path.basename(partition) + '.backup.signature'
+    signaturefile = os.path.join(self.backup_dir, filename)
+    if os.path.exists(signaturefile):
+      return signaturefile, False
+    signaturefile += '.custom'
+    if os.path.exists(signaturefile):
+      return signaturefile, True
+    raise Exception('ERROR: missing signature file for partition ' + partition)
+
+  def verify(self, signaturefile, root_dir, custom=False):
+    proof = signaturefile + '.proof'
+    if custom:
+      signatures = self.sign_custom(root_dir)
+    else:
+      signatures = self.sign(signaturefile, root_dir)
     with open(proof, 'w') as f:
-      for s in hashwalk(self.backup_dir, self.mirror_partition_dirs):
+      for s in signatures:
         f.write(s + '\n')
-    diffcommand = ('diff', moved, proof)
-    print(' '.join(diffcommand))
+    diffcommand = ('diff', signaturefile, proof)
     try:
       sp.check_output(
         diffcommand, stderr=sp.STDOUT, universal_newlines=True)
     except sp.CalledProcessError as e:
-      template = 'ERROR the backup signatures do not match\n\n%s'
-      msg = template % e.output
+      template = 'ERROR the backup signatures do not match\n\n%s\n%s'
+      msg = template % (' '.join(diffcommand), e.output)
       print(msg)
       raise Exception(msg)
 
-  def loginfo(self, msg):
+  def log(self, msg):
     print(msg)
     self.logs.append(msg)
 
@@ -144,44 +173,54 @@ class TheiaImport(object):
       sys.exit(exitcode)
 
   def restore(self):
-    self.loginfo('Verify backup signature')
-    self.verify(os.path.join(self.backup_dir, 'backup.signature'))
+    self.log('Verify main backup signature')
+    signaturefile = os.path.join(self.backup_dir, 'backup.signature')
+    self.verify(signaturefile, self.backup_dir)
+
+    custom_partition_signatures = []
+    for m in self.mirror_partition_dirs:
+      signaturefile, custom = self.find_signature_file(m)
+      if custom:
+        custom_partition_signatures.append((signaturefile, m))
+      else:
+        self.log('Verify backup signature for ' + m)
+        self.verify(signaturefile, m)
 
-    self.loginfo('Stop slapproxy')
+    self.log('Stop slapproxy')
     self.supervisorctl('stop', 'slapos-proxy')
 
-    self.loginfo('Restore partitions')
+    self.log('Restore partitions')
     for m in self.mirror_partition_dirs:
-      self.restorepartition(m)
+      self.restore_partition(m)
 
     for d in self.dirs:
-      self.loginfo('Restore directory ' + d)
-      self.restoretree(d)
+      self.log('Restore directory ' + d)
+      self.restore_tree(d)
 
-    self.loginfo('Restore slapproxy database')
-    self.restoredb()
+    self.log('Restore slapproxy database')
+    self.restore_db()
 
     timestamp = os.path.join(self.root_dir, 'etc', '.resilient_timestamp')
-    self.loginfo('Restore resilient timestamp ' + timestamp)
-    self.restorefile(timestamp)
+    self.log('Restore resilient timestamp ' + timestamp)
+    self.restore_file(timestamp)
 
     custom_script = os.path.join(self.root_dir, 'srv', 'runner-import-restore')
     if os.path.exists(custom_script):
-      self.loginfo('Run custom restore script %s' % custom_script)
+      self.log('Run custom restore script %s' % custom_script)
       sp.check_call(custom_script)
 
-    self.loginfo('Start slapproxy again')
+    self.log('Start slapproxy again')
     self.supervisorctl('start', 'slapos-proxy')
 
-    self.loginfo('Reformat partitions')
+    self.log('Reformat partitions')
     self.slapos('node', 'format', '--now')
 
-    self.loginfo('Remove old supervisord configuration files')
+    self.log('Remove old supervisord configuration files')
     conf_dir = os.path.join(self.instance_dir, 'etc', 'supervisor.conf.d')
     for f in glob.glob(os.path.join(conf_dir, '*')):
       os.remove(f)
 
-    self.loginfo('Build Software Releases')
+    self.log('Build Software Releases')
     for i in range(3):
       try:
         self.slapos('node', 'software', '--all', '--logfile', self.sr_log)
@@ -191,18 +230,18 @@ class TheiaImport(object):
       else:
         break
 
-    self.loginfo('Remove old custom instance scripts')
+    self.log('Remove old custom instance scripts')
     partitions_glob = os.path.join(self.instance_dir, 'slappart*')
     scripts = os.path.join(partitions_glob, 'srv', 'runner-import-restore')
     for f in glob.glob(scripts):
       remove(f)
 
-    self.loginfo('Remove partition timestamps')
+    self.log('Remove partition timestamps')
     timestamps = os.path.join(partitions_glob, '.timestamp')
     for f in glob.glob(timestamps):
       remove(f)
 
-    self.loginfo('Build Instances')
+    self.log('Build Instances')
     cp_log = self.cp_log
     for i in range(3):
       try:
@@ -213,11 +252,15 @@ class TheiaImport(object):
       else:
         break
 
+    self.log('Verify custom backup signatures')
+    for signaturefile, m in custom_partition_signatures:
+      self.verify(signaturefile, m, True)
+
     for custom_script in glob.glob(scripts):
-      self.loginfo('Running custom instance script %s' % custom_script)
+      self.log('Running custom instance script %s' % custom_script)
       sp.check_call(custom_script)
 
-    self.loginfo('Done')
+    self.log('Done')
 
 
 if __name__ == '__main__':