# Copyright (C) 2015-2021 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information """Loader in charge of injecting either new or existing svn mirrors to swh-storage. """ from datetime import datetime from mmap import ACCESS_WRITE, mmap import os import pty import re import shutil from subprocess import Popen import tempfile from typing import Any, Dict, Iterator, List, Optional, Sequence, Tuple from subvertpy import SubversionException from swh.loader.core.loader import BaseLoader from swh.loader.core.utils import clean_dangling_folders from swh.loader.exception import NotFound from swh.loader.svn.svn import SvnRepo from swh.model import from_disk, hashutil from swh.model.model import ( Content, Directory, Revision, SkippedContent, Snapshot, SnapshotBranch, TargetType, ) from swh.storage.algos.snapshot import snapshot_get_latest from swh.storage.interface import StorageInterface from . import converters from .exception import SvnLoaderHistoryAltered, SvnLoaderUneventful from .utils import ( OutputStream, init_svn_repo_from_archive_dump, init_svn_repo_from_dump, ) DEFAULT_BRANCH = b"HEAD" TEMPORARY_DIR_PREFIX_PATTERN = "swh.loader.svn." SUBVERSION_ERROR = re.compile(r".*(E[0-9]{6}):.*") SUBVERSION_NOT_FOUND = "E170013" class SvnLoader(BaseLoader): """SVN loader. The repository is either remote or local. The loader deals with update on an already previously loaded repository. """ visit_type = "svn" def __init__( self, storage: StorageInterface, url: str, origin_url: Optional[str] = None, visit_date: Optional[datetime] = None, incremental: bool = True, temp_directory: str = "/tmp", debug: bool = False, check_revision: int = 0, **kwargs: Any, ): """Load a svn repository (either remote or local). Args: url: The default origin url origin_url: Optional original url override to use as origin reference in the archive. If not provided, "url" is used as origin. visit_date: Optional date to override the visit date incremental: If True, the default, starts from the last snapshot (if any). Otherwise, starts from the initial commit of the repository. temp_directory: The temporary directory to use as root directory for working directory computations debug: If true, run the loader in debug mode. At the end of the loading, the temporary working directory is not cleaned up to ease inspection. Defaults to false. check_revision: The number of svn commits between checks for hash divergence """ # technical svn uri to act on svn repository self.svn_url = url # origin url as unique identifier for origin in swh archive origin_url = origin_url or self.svn_url super().__init__(storage=storage, origin_url=origin_url, **kwargs) self.debug = debug self.temp_directory = temp_directory self.done = False self.svnrepo = None self.skip_post_load = False # Revision check is configurable self.check_revision = check_revision # internal state used to store swh objects self._contents: List[Content] = [] self._skipped_contents: List[SkippedContent] = [] self._directories: List[Directory] = [] self._revisions: List[Revision] = [] self._snapshot: Optional[Snapshot] = None # internal state, current visit self._last_revision = None self._visit_status = "full" self._load_status = "uneventful" self.visit_date = visit_date or self.visit_date self.incremental = incremental self.snapshot: Optional[Snapshot] = None # state from previous visit self.latest_snapshot = None self.latest_revision: Optional[Revision] = None self.from_dump = False def pre_cleanup(self): """Cleanup potential dangling files from prior runs (e.g. OOM killed tasks) """ clean_dangling_folders( self.temp_directory, pattern_check=TEMPORARY_DIR_PREFIX_PATTERN, log=self.log, ) def cleanup(self): """Clean up the svn repository's working representation on disk.""" if not self.svnrepo: # could happen if `prepare` fails return if self.debug: self.log.error( """NOT FOR PRODUCTION - debug flag activated Local repository not cleaned up for investigation: %s""", self.svnrepo.local_url.decode("utf-8"), ) return self.svnrepo.clean_fs() def swh_revision_hash_tree_at_svn_revision(self, revision: int) -> bytes: """Compute and return the hash tree at a given svn revision. Args: rev: the svn revision we want to check Returns: The hash tree directory as bytes. """ assert self.svnrepo is not None local_dirname, local_url = self.svnrepo.export_temporary(revision) root_dir = from_disk.Directory.from_disk(path=local_url) self.svnrepo.clean_fs(local_dirname) return root_dir.hash def _latest_snapshot_revision( self, origin_url: str, ) -> Optional[Tuple[Snapshot, Revision]]: """Look for latest snapshot revision and returns it if any. Args: origin_url: Origin identifier previous_swh_revision: possible previous swh revision (either a dict or revision identifier) Returns: Tuple of the latest Snapshot from the previous visit and its targeted revision if any or None otherwise. """ storage = self.storage latest_snapshot = snapshot_get_latest(storage, origin_url) if not latest_snapshot: return None branches = latest_snapshot.branches if not branches: return None branch = branches.get(DEFAULT_BRANCH) if not branch: return None if branch.target_type != TargetType.REVISION: return None swh_id = branch.target revision = storage.revision_get([swh_id])[0] if not revision: return None return latest_snapshot, revision def build_swh_revision( self, rev: int, commit: Dict, dir_id: bytes, parents: Sequence[bytes] ) -> Revision: """Build the swh revision dictionary. This adds: - the `'synthetic`' flag to true - the '`extra_headers`' containing the repository's uuid and the svn revision number. Args: rev: the svn revision number commit: the commit data: revision id, date, author, and message dir_id: the upper tree's hash identifier parents: the parents' identifiers Returns: The swh revision corresponding to the svn revision. """ assert self.svnrepo is not None return converters.build_swh_revision( rev, commit, self.svnrepo.uuid, dir_id, parents ) def check_history_not_altered(self, revision_start: int, swh_rev: Revision) -> bool: """Given a svn repository, check if the history was modified in between visits.""" revision_id = swh_rev.id parents = swh_rev.parents assert self.svnrepo is not None commit, root_dir = self.svnrepo.swh_hash_data_at_revision(revision_start) dir_id = root_dir.hash swh_revision = self.build_swh_revision(revision_start, commit, dir_id, parents) swh_revision_id = swh_revision.id return swh_revision_id == revision_id def start_from(self) -> Tuple[int, int]: """Determine from where to start the loading. Returns: tuple (revision_start, revision_end) Raises: SvnLoaderHistoryAltered: When a hash divergence has been detected (should not happen) SvnLoaderUneventful: Nothing changed since last visit """ assert self.svnrepo is not None, "svnrepo initialized in the `prepare` method" revision_head = self.svnrepo.head_revision() if revision_head == 0: # empty repository case revision_start = 0 revision_end = 0 else: # default configuration revision_start = self.svnrepo.initial_revision() revision_end = revision_head # start from a previous revision if any if self.incremental and self.latest_revision is not None: extra_headers = dict(self.latest_revision.extra_headers) revision_start = int(extra_headers[b"svn_revision"]) if not self.check_history_not_altered(revision_start, self.latest_revision): self.log.debug( ( "History of svn %s@%s altered. " "A complete reloading of the repository will be performed." ), self.svnrepo.remote_url, revision_start, ) revision_start = 0 # now we know history is ok, we start at next revision revision_start = revision_start + 1 if revision_start > revision_end: msg = "%s@%s already injected." % (self.svnrepo.remote_url, revision_end) raise SvnLoaderUneventful(msg) self.log.info( "Processing revisions [%s-%s] for %s", revision_start, revision_end, self.svnrepo, ) return revision_start, revision_end def _check_revision_divergence(self, rev: int, dir_id: bytes) -> None: """Check for hash revision computation divergence. The Rationale behind this is that svn can trigger unknown edge cases (mixed CRLF, svn properties, etc...). Those are not always easy to spot. Adding a regular check will help spotting potential missing edge cases. Args: rev: The actual revision we are computing from dir_id: The actual directory for the given revision Raises ValueError if a hash divergence is detected """ self.log.debug("Checking hash computations on revision %s...", rev) checked_dir_id = self.swh_revision_hash_tree_at_svn_revision(rev) if checked_dir_id != dir_id: err = ( "Hash tree computation divergence detected " "(%s != %s), stopping!" % ( hashutil.hash_to_hex(dir_id), hashutil.hash_to_hex(checked_dir_id), ) ) raise ValueError(err) def process_svn_revisions( self, svnrepo, revision_start, revision_end ) -> Iterator[ Tuple[List[Content], List[SkippedContent], List[Directory], Revision] ]: """Process svn revisions from revision_start to revision_end. At each svn revision, apply new diffs and simultaneously compute swh hashes. This yields those computed swh hashes as a tuple (contents, directories, revision). Note that at every `self.check_revision`, a supplementary check takes place to check for hash-tree divergence (related T570). Yields: tuple (contents, directories, revision) of dict as a dictionary with keys, sha1_git, sha1, etc... Raises: ValueError in case of a hash divergence detection """ gen_revs = svnrepo.swh_hash_data_per_revision(revision_start, revision_end) parents = (self.latest_revision.id,) if self.latest_revision is not None else () count = 0 for rev, commit, new_objects, root_directory in gen_revs: count += 1 # Send the associated contents/directories _contents, _skipped_contents, _directories = new_objects # compute the fs tree's checksums dir_id = root_directory.hash swh_revision = self.build_swh_revision(rev, commit, dir_id, parents) self.log.debug( "rev: %s, swhrev: %s, dir: %s", rev, hashutil.hash_to_hex(swh_revision.id), hashutil.hash_to_hex(dir_id), ) if ( self.check_revision and self.check_revision != 0 and count % self.check_revision == 0 ): self._check_revision_divergence(rev, dir_id) parents = (swh_revision.id,) yield _contents, _skipped_contents, _directories, swh_revision if not self.debug and self.svnrepo: # clean directory where revisions were replayed to gain some disk space # before the post_load operation self.svnrepo.clean_fs(self.svnrepo.local_url) def svn_repo(self, *args, **kwargs): """Wraps the creation of SvnRepo object and handles not found repository errors.""" try: return SvnRepo(*args, **kwargs) except SubversionException as e: error_msgs = [ "Unable to connect to a repository at URL", "Unknown URL type", "is not a working copy", ] for msg in error_msgs: if msg in e.args[0]: self._load_status = "uneventful" raise NotFound(e) raise def prepare(self): if self.incremental: latest_snapshot_revision = self._latest_snapshot_revision(self.origin.url) if latest_snapshot_revision: self.latest_snapshot, self.latest_revision = latest_snapshot_revision self._snapshot = self.latest_snapshot self._last_revision = self.latest_revision local_dirname = self._create_tmp_dir(self.temp_directory) self.svnrepo = self.svn_repo( self.svn_url, self.origin.url, local_dirname, self.max_content_size, self.from_dump, ) try: revision_start, revision_end = self.start_from() self.swh_revision_gen = self.process_svn_revisions( self.svnrepo, revision_start, revision_end ) except SvnLoaderUneventful as e: self.log.warning(e) self.done = True self._load_status = "uneventful" except SvnLoaderHistoryAltered as e: self.log.error(e) self.done = True self._visit_status = "partial" def fetch_data(self): """Fetching svn revision information. This will apply svn revision as patch on disk, and at the same time, compute the swh hashes. In effect, fetch_data fetches those data and compute the necessary swh objects. It's then stored in the internal state instance variables (initialized in `_prepare_state`). This is up to `store_data` to actually discuss with the storage to store those objects. Returns: bool: True to continue fetching data (next svn revision), False to stop. """ if self.done: return False try: data = next(self.swh_revision_gen) self._load_status = "eventful" except StopIteration: self.done = True # Stopping iteration self._visit_status = "full" except Exception as e: # svn:external, hash divergence, i/o error... self.log.exception(e) self.done = True # Stopping iteration self._visit_status = "partial" else: self._contents, self._skipped_contents, self._directories, rev = data if rev: self._last_revision = rev self._revisions.append(rev) return not self.done def store_data(self): """We store the data accumulated in internal instance variable. If the iteration over the svn revisions is done, we create the snapshot and flush to storage the data. This also resets the internal instance variable state. """ self.storage.skipped_content_add(self._skipped_contents) self.storage.content_add(self._contents) self.storage.directory_add(self._directories) self.storage.revision_add(self._revisions) if self.done: # finish line, snapshot! self.snapshot = self.generate_and_load_snapshot( revision=self._last_revision, snapshot=self._snapshot ) self.flush() self.loaded_snapshot_id = self.snapshot.id # reset internal state for next iteration self._revisions = [] def generate_and_load_snapshot( self, revision: Optional[Revision] = None, snapshot: Optional[Snapshot] = None ) -> Snapshot: """Create the snapshot either from existing revision or snapshot. Revision (supposedly new) has priority over the snapshot (supposedly existing one). Args: revision (dict): Last revision seen if any (None by default) snapshot (dict): Snapshot to use if any (None by default) Returns: Optional[Snapshot] The newly created snapshot """ if revision: # Priority to the revision snap = Snapshot( branches={ DEFAULT_BRANCH: SnapshotBranch( target=revision.id, target_type=TargetType.REVISION ) } ) elif snapshot: # Fallback to prior snapshot snap = snapshot else: raise ValueError( "generate_and_load_snapshot called with null revision and snapshot!" ) self.log.debug("snapshot: %s", snap) self.storage.snapshot_add([snap]) return snap def load_status(self): return { "status": self._load_status, } def visit_status(self): return self._visit_status def post_load(self, success: bool = True) -> None: if self.skip_post_load: return if success and self._last_revision is not None: # check if the reconstructed filesystem for the last loaded revision is # consistent with the one obtained with a svn export operation. If it is not # the case, an exception will be raised to report the issue and mark the # visit as partial self._check_revision_divergence( int(dict(self._last_revision.extra_headers)[b"svn_revision"]), self._last_revision.directory, ) def _create_tmp_dir(self, root_tmp_dir: str) -> str: return tempfile.mkdtemp( dir=root_tmp_dir, prefix=TEMPORARY_DIR_PREFIX_PATTERN, suffix="-%s" % os.getpid(), ) class SvnLoaderFromDumpArchive(SvnLoader): """Uncompress an archive containing an svn dump, mount the svn dump as a local svn repository and load that repository. """ def __init__( self, storage: StorageInterface, url: str, archive_path: str, origin_url: Optional[str] = None, incremental: bool = False, visit_date: Optional[datetime] = None, temp_directory: str = "/tmp", debug: bool = False, check_revision: int = 0, **kwargs: Any, ): super().__init__( storage=storage, url=url, origin_url=origin_url, incremental=incremental, visit_date=visit_date, temp_directory=temp_directory, debug=debug, check_revision=check_revision, **kwargs, ) self.archive_path = archive_path self.temp_dir = None self.repo_path = None self.from_dump = True def prepare(self): self.log.info("Archive to mount and load %s", self.archive_path) self.temp_dir, self.repo_path = init_svn_repo_from_archive_dump( self.archive_path, prefix=TEMPORARY_DIR_PREFIX_PATTERN, suffix="-%s" % os.getpid(), root_dir=self.temp_directory, ) self.svn_url = f"file://{self.repo_path}" super().prepare() def cleanup(self): super().cleanup() if self.temp_dir and os.path.exists(self.temp_dir): self.log.debug( "Clean up temporary directory dump %s for project %s", self.temp_dir, os.path.basename(self.repo_path), ) shutil.rmtree(self.temp_dir) class SvnLoaderFromRemoteDump(SvnLoader): """Create a subversion repository dump out of a remote svn repository (using the svnrdump utility). Then, mount the repository locally and load that repository. """ def __init__( self, storage: StorageInterface, url: str, origin_url: Optional[str] = None, incremental: bool = True, visit_date: Optional[datetime] = None, temp_directory: str = "/tmp", debug: bool = False, check_revision: int = 0, **kwargs: Any, ): super().__init__( storage=storage, url=url, origin_url=origin_url, incremental=incremental, visit_date=visit_date, temp_directory=temp_directory, debug=debug, check_revision=check_revision, **kwargs, ) self.from_dump = True self.temp_dir = self._create_tmp_dir(self.temp_directory) self.repo_path = None self.truncated_dump = False def get_last_loaded_svn_rev(self, svn_url: str) -> int: """Check if the svn repository has already been visited and return the last loaded svn revision number or -1 otherwise. """ origin = list(self.storage.origin_get([svn_url]))[0] if not origin: return -1 svn_revision = -1 try: latest_snapshot_revision = self._latest_snapshot_revision(origin.url) if latest_snapshot_revision: _, latest_revision = latest_snapshot_revision latest_revision_headers = dict(latest_revision.extra_headers) svn_revision = int(latest_revision_headers[b"svn_revision"]) except Exception: pass return svn_revision def dump_svn_revisions(self, svn_url: str, last_loaded_svn_rev: int = -1) -> str: """Generate a subversion dump file using the svnrdump tool. If the svnrdump command failed somehow, the produced dump file is analyzed to determine if a partial loading is still feasible. Raises: NotFound when the repository is no longer found at url Returns: The dump_path of the repository mounted """ # Build the svnrdump command line svnrdump_cmd = ["svnrdump", "dump", svn_url] assert self.svnrepo is not None if self.svnrepo.username: svnrdump_cmd += [ "--username", self.svnrepo.username, "--password", self.svnrepo.password, ] # Launch the svnrdump command while capturing stderr as # successfully dumped revision numbers are printed to it dump_temp_dir = tempfile.mkdtemp(dir=self.temp_dir) dump_name = "".join(c for c in svn_url if c.isalnum()) dump_path = "%s/%s.svndump" % (dump_temp_dir, dump_name) stderr_lines = [] self.log.debug("Executing %s", " ".join(svnrdump_cmd)) with open(dump_path, "wb") as dump_file: stderr_r, stderr_w = pty.openpty() svnrdump = Popen(svnrdump_cmd, stdout=dump_file, stderr=stderr_w) os.close(stderr_w) stderr_stream = OutputStream(stderr_r) readable = True error_codes: List[str] = [] error_messages: List[str] = [] while readable: lines, readable = stderr_stream.read_lines() stderr_lines += lines for line in lines: self.log.debug(line) match = SUBVERSION_ERROR.search(line) if match: error_codes.append(match.group(1)) error_messages.append(line) svnrdump.wait() os.close(stderr_r) if svnrdump.returncode == 0: return dump_path # There was an error but it does not mean that no revisions # can be loaded. # Get the stderr line with latest dumped revision last_dumped_rev = None for stderr_line in reversed(stderr_lines): if stderr_line.startswith("* Dumped revision"): last_dumped_rev = stderr_line break if last_dumped_rev: # Get the latest dumped revision number matched_rev = re.search(".*revision ([0-9]+)", last_dumped_rev) last_dumped_rev = int(matched_rev.group(1)) if matched_rev else -1 # Check if revisions inside the dump file can be loaded anyway if last_dumped_rev > last_loaded_svn_rev: self.log.debug( ( "svnrdump did not dump all expected revisions " "but revisions range %s:%s are available in " "the generated dump file and will be loaded " "into the archive." ), last_loaded_svn_rev + 1, last_dumped_rev, ) # Truncate the dump file after the last successfully dumped # revision to avoid the loading of corrupted data self.log.debug( ( "Truncating dump file after the last " "successfully dumped revision (%s) to avoid " "the loading of corrupted data" ), last_dumped_rev, ) with open(dump_path, "r+b") as f: with mmap(f.fileno(), 0, access=ACCESS_WRITE) as s: pattern = ( "Revision-number: %s" % (last_dumped_rev + 1) ).encode() n = s.rfind(pattern) if n != -1: s.resize(n) self.truncated_dump = True return dump_path elif last_dumped_rev != -1 and last_dumped_rev < last_loaded_svn_rev: raise Exception( ( "Last dumped subversion revision (%s) is " "lesser than the last one loaded into the " "archive (%s)." ) % (last_dumped_rev, last_loaded_svn_rev) ) if SUBVERSION_NOT_FOUND in error_codes: raise NotFound( f"{SUBVERSION_NOT_FOUND}: Repository never existed or disappeared" ) raise Exception( "An error occurred when running svnrdump and " "no exploitable dump file has been generated.\n" + "\n".join(error_messages) ) def prepare(self): # First, check if previous revisions have been loaded for the # subversion origin and get the number of the last one last_loaded_svn_rev = self.get_last_loaded_svn_rev(self.origin.url) self.svnrepo = self.svn_repo( self.origin.url, self.origin.url, self.temp_dir, self.max_content_size ) # Ensure to use remote URL retrieved by SvnRepo as origin URL might redirect # and svnrdump does not handle URL redirection self.svn_url = self.svnrepo.remote_url # Then for stale repository, check if the last loaded revision in the archive # is different from the last revision on the remote subversion server. # Skip the dump of all revisions and the loading process if they are identical # to save some disk space and processing time. last_loaded_snp_and_rev = self._latest_snapshot_revision(self.origin.url) if last_loaded_snp_and_rev is not None: last_loaded_snp, last_loaded_rev = last_loaded_snp_and_rev stale_repository = self.svnrepo.head_revision() == last_loaded_svn_rev if stale_repository and self.check_history_not_altered( last_loaded_svn_rev, last_loaded_rev ): self._snapshot = last_loaded_snp self._last_revision = last_loaded_rev self.done = True self.skip_post_load = True return # Then try to generate a dump file containing relevant svn revisions # to load, an exception will be thrown if something wrong happened dump_path = self.dump_svn_revisions(self.svn_url, last_loaded_svn_rev) # Finally, mount the dump and load the repository self.log.debug('Mounting dump file with "svnadmin load".') _, self.repo_path = init_svn_repo_from_dump( dump_path, prefix=TEMPORARY_DIR_PREFIX_PATTERN, suffix="-%s" % os.getpid(), root_dir=self.temp_dir, ) self.svn_url = "file://%s" % self.repo_path super().prepare() def cleanup(self): super().cleanup() if self.temp_dir and os.path.exists(self.temp_dir): shutil.rmtree(self.temp_dir) def visit_status(self): if self.truncated_dump: return "partial" else: return super().visit_status()