diff --git a/dulwich/client.py b/dulwich/client.py index 3d864682..ed1a2d96 100644 --- a/dulwich/client.py +++ b/dulwich/client.py @@ -1,1750 +1,1751 @@ # client.py -- Implementation of the client side git protocols # Copyright (C) 2008-2013 Jelmer Vernooij # # Dulwich is dual-licensed under the Apache License, Version 2.0 and the GNU # General Public License as public by the Free Software Foundation; version 2.0 # or (at your option) any later version. You can redistribute it and/or # modify it under the terms of either of these two licenses. # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # You should have received a copy of the licenses; if not, see # for a copy of the GNU General Public License # and for a copy of the Apache # License, Version 2.0. # """Client side support for the Git protocol. The Dulwich client supports the following capabilities: * thin-pack * multi_ack_detailed * multi_ack * side-band-64k * ofs-delta * quiet * report-status * delete-refs Known capabilities that are not supported: * shallow * no-progress * include-tag """ from contextlib import closing from io import BytesIO, BufferedReader import select import socket import subprocess import sys try: from urllib import quote as urlquote from urllib import unquote as urlunquote except ImportError: from urllib.parse import quote as urlquote from urllib.parse import unquote as urlunquote try: import urlparse except ImportError: import urllib.parse as urlparse import dulwich from dulwich.errors import ( GitProtocolError, NotGitRepository, SendPackError, UpdateRefsError, ) from dulwich.protocol import ( HangupException, _RBUFSIZE, agent_string, capability_agent, extract_capability_names, CAPABILITY_AGENT, CAPABILITY_DELETE_REFS, CAPABILITY_MULTI_ACK, CAPABILITY_MULTI_ACK_DETAILED, CAPABILITY_OFS_DELTA, CAPABILITY_QUIET, CAPABILITY_REPORT_STATUS, CAPABILITY_SHALLOW, CAPABILITY_SYMREF, CAPABILITY_SIDE_BAND_64K, CAPABILITY_THIN_PACK, CAPABILITIES_REF, KNOWN_RECEIVE_CAPABILITIES, KNOWN_UPLOAD_CAPABILITIES, COMMAND_DEEPEN, COMMAND_SHALLOW, COMMAND_UNSHALLOW, COMMAND_DONE, COMMAND_HAVE, COMMAND_WANT, SIDE_BAND_CHANNEL_DATA, SIDE_BAND_CHANNEL_PROGRESS, SIDE_BAND_CHANNEL_FATAL, PktLineParser, Protocol, ProtocolFile, TCP_GIT_PORT, ZERO_SHA, extract_capabilities, parse_capability, ) from dulwich.pack import ( write_pack_data, write_pack_objects, ) from dulwich.refs import ( read_info_refs, ANNOTATED_TAG_SUFFIX, ) class InvalidWants(Exception): """Invalid wants.""" def __init__(self, wants): Exception.__init__( self, "requested wants not in server provided refs: %r" % wants) def _fileno_can_read(fileno): """Check if a file descriptor is readable.""" return len(select.select([fileno], [], [], 0)[0]) > 0 def _win32_peek_avail(handle): """Wrapper around PeekNamedPipe to check how many bytes are available.""" from ctypes import byref, wintypes, windll c_avail = wintypes.DWORD() c_message = wintypes.DWORD() success = windll.kernel32.PeekNamedPipe( handle, None, 0, None, byref(c_avail), byref(c_message)) if not success: raise OSError(wintypes.GetLastError()) return c_avail.value COMMON_CAPABILITIES = [CAPABILITY_OFS_DELTA, CAPABILITY_SIDE_BAND_64K] UPLOAD_CAPABILITIES = ([CAPABILITY_THIN_PACK, CAPABILITY_MULTI_ACK, CAPABILITY_MULTI_ACK_DETAILED, CAPABILITY_SHALLOW] + COMMON_CAPABILITIES) RECEIVE_CAPABILITIES = [CAPABILITY_REPORT_STATUS] + COMMON_CAPABILITIES class ReportStatusParser(object): """Handle status as reported by servers with 'report-status' capability. """ def __init__(self): self._done = False self._pack_status = None self._ref_status_ok = True self._ref_statuses = [] def check(self): """Check if there were any errors and, if so, raise exceptions. :raise SendPackError: Raised when the server could not unpack :raise UpdateRefsError: Raised when refs could not be updated """ if self._pack_status not in (b'unpack ok', None): raise SendPackError(self._pack_status) if not self._ref_status_ok: ref_status = {} ok = set() for status in self._ref_statuses: if b' ' not in status: # malformed response, move on to the next one continue status, ref = status.split(b' ', 1) if status == b'ng': if b' ' in ref: ref, status = ref.split(b' ', 1) else: ok.add(ref) ref_status[ref] = status # TODO(jelmer): don't assume encoding of refs is ascii. raise UpdateRefsError(', '.join([ refname.decode('ascii') for refname in ref_status if refname not in ok]) + ' failed to update', ref_status=ref_status) def handle_packet(self, pkt): """Handle a packet. :raise GitProtocolError: Raised when packets are received after a flush packet. """ if self._done: raise GitProtocolError("received more data after status report") if pkt is None: self._done = True return if self._pack_status is None: self._pack_status = pkt.strip() else: ref_status = pkt.strip() self._ref_statuses.append(ref_status) if not ref_status.startswith(b'ok '): self._ref_status_ok = False def read_pkt_refs(proto): server_capabilities = None refs = {} # Receive refs from server for pkt in proto.read_pkt_seq(): (sha, ref) = pkt.rstrip(b'\n').split(None, 1) if sha == b'ERR': raise GitProtocolError(ref.decode('utf-8', 'replace')) if server_capabilities is None: (ref, server_capabilities) = extract_capabilities(ref) refs[ref] = sha if len(refs) == 0: return {}, set([]) if refs == {CAPABILITIES_REF: ZERO_SHA}: refs = {} return refs, set(server_capabilities) class FetchPackResult(object): """Result of a fetch-pack operation. :var refs: Dictionary with all remote refs :var symrefs: Dictionary with remote symrefs :var agent: User agent string """ _FORWARDED_ATTRS = [ 'clear', 'copy', 'fromkeys', 'get', 'has_key', 'items', 'iteritems', 'iterkeys', 'itervalues', 'keys', 'pop', 'popitem', 'setdefault', 'update', 'values', 'viewitems', 'viewkeys', 'viewvalues'] def __init__(self, refs, symrefs, agent, new_shallow=None, new_unshallow=None): self.refs = refs self.symrefs = symrefs self.agent = agent self.new_shallow = new_shallow self.new_unshallow = new_unshallow def _warn_deprecated(self): import warnings warnings.warn( "Use FetchPackResult.refs instead.", DeprecationWarning, stacklevel=3) def __eq__(self, other): if isinstance(other, dict): self._warn_deprecated() return (self.refs == other) return (self.refs == other.refs and self.symrefs == other.symrefs and self.agent == other.agent) def __contains__(self, name): self._warn_deprecated() return name in self.refs def __getitem__(self, name): self._warn_deprecated() return self.refs[name] def __len__(self): self._warn_deprecated() return len(self.refs) def __iter__(self): self._warn_deprecated() return iter(self.refs) def __getattribute__(self, name): if name in type(self)._FORWARDED_ATTRS: self._warn_deprecated() return getattr(self.refs, name) return super(FetchPackResult, self).__getattribute__(name) def __repr__(self): return "%s(%r, %r, %r)" % ( self.__class__.__name__, self.refs, self.symrefs, self.agent) def _read_shallow_updates(proto): new_shallow = set() new_unshallow = set() for pkt in proto.read_pkt_seq(): cmd, sha = pkt.split(b' ', 1) if cmd == COMMAND_SHALLOW: new_shallow.add(sha.strip()) elif cmd == COMMAND_UNSHALLOW: new_unshallow.add(sha.strip()) else: raise GitProtocolError('unknown command %s' % pkt) return (new_shallow, new_unshallow) # TODO(durin42): this doesn't correctly degrade if the server doesn't # support some capabilities. This should work properly with servers # that don't support multi_ack. class GitClient(object): """Git smart server client. """ def __init__(self, thin_packs=True, report_activity=None, quiet=False): """Create a new GitClient instance. :param thin_packs: Whether or not thin packs should be retrieved :param report_activity: Optional callback for reporting transport activity. """ self._report_activity = report_activity self._report_status_parser = None self._fetch_capabilities = set(UPLOAD_CAPABILITIES) self._fetch_capabilities.add(capability_agent()) self._send_capabilities = set(RECEIVE_CAPABILITIES) self._send_capabilities.add(capability_agent()) if quiet: self._send_capabilities.add(CAPABILITY_QUIET) if not thin_packs: self._fetch_capabilities.remove(CAPABILITY_THIN_PACK) def get_url(self, path): """Retrieves full url to given path. :param path: Repository path (as string) :return: Url to path (as string) """ raise NotImplementedError(self.get_url) @classmethod def from_parsedurl(cls, parsedurl, **kwargs): """Create an instance of this client from a urlparse.parsed object. :param parsedurl: Result of urlparse.urlparse() :return: A `GitClient` object """ raise NotImplementedError(cls.from_parsedurl) def send_pack(self, path, update_refs, generate_pack_data, progress=None): """Upload a pack to a remote repository. :param path: Repository path (as bytestring) :param update_refs: Function to determine changes to remote refs. Receive dict with existing remote refs, returns dict with changed refs (name -> sha, where sha=ZERO_SHA for deletions) :param generate_pack_data: Function that can return a tuple with number of objects and list of pack data to include :param progress: Optional progress function :raises SendPackError: if server rejects the pack data :raises UpdateRefsError: if the server supports report-status and rejects ref updates :return: new_refs dictionary containing the changes that were made {refname: new_ref}, including deleted refs. """ raise NotImplementedError(self.send_pack) def fetch(self, path, target, determine_wants=None, progress=None, depth=None): """Fetch into a target repository. :param path: Path to fetch from (as bytestring) :param target: Target repository to fetch into :param determine_wants: Optional function to determine what refs to fetch. Receives dictionary of name->sha, should return list of shas to fetch. Defaults to all shas. :param progress: Optional progress function :param depth: Depth to fetch at :return: Dictionary with all remote refs (not just those fetched) """ if determine_wants is None: determine_wants = target.object_store.determine_wants_all if CAPABILITY_THIN_PACK in self._fetch_capabilities: # TODO(jelmer): Avoid reading entire file into memory and # only processing it after the whole file has been fetched. f = BytesIO() def commit(): if f.tell(): f.seek(0) target.object_store.add_thin_pack(f.read, None) def abort(): pass else: f, commit, abort = target.object_store.add_pack() try: result = self.fetch_pack( path, determine_wants, target.get_graph_walker(), f.write, progress=progress, depth=depth) except BaseException: abort() raise else: commit() target.update_shallow(result.new_shallow, result.new_unshallow) return result def fetch_pack(self, path, determine_wants, graph_walker, pack_data, progress=None, depth=None): """Retrieve a pack from a git smart server. :param path: Remote path to fetch from :param determine_wants: Function determine what refs to fetch. Receives dictionary of name->sha, should return list of shas to fetch. :param graph_walker: Object with next() and ack(). :param pack_data: Callback called for each bit of data in the pack :param progress: Callback for progress reports (strings) :param depth: Shallow fetch depth :return: FetchPackResult object """ raise NotImplementedError(self.fetch_pack) def get_refs(self, path): """Retrieve the current refs from a git smart server. :param path: Path to the repo to fetch from. (as bytestring) """ raise NotImplementedError(self.get_refs) def _parse_status_report(self, proto): unpack = proto.read_pkt_line().strip() if unpack != b'unpack ok': st = True # flush remaining error data while st is not None: st = proto.read_pkt_line() raise SendPackError(unpack) statuses = [] errs = False ref_status = proto.read_pkt_line() while ref_status: ref_status = ref_status.strip() statuses.append(ref_status) if not ref_status.startswith(b'ok '): errs = True ref_status = proto.read_pkt_line() if errs: ref_status = {} ok = set() for status in statuses: if b' ' not in status: # malformed response, move on to the next one continue status, ref = status.split(b' ', 1) if status == b'ng': if b' ' in ref: ref, status = ref.split(b' ', 1) else: ok.add(ref) ref_status[ref] = status raise UpdateRefsError(', '.join([ refname for refname in ref_status if refname not in ok]) + b' failed to update', ref_status=ref_status) def _read_side_band64k_data(self, proto, channel_callbacks): """Read per-channel data. This requires the side-band-64k capability. :param proto: Protocol object to read from :param channel_callbacks: Dictionary mapping channels to packet handlers to use. None for a callback discards channel data. """ for pkt in proto.read_pkt_seq(): channel = ord(pkt[:1]) pkt = pkt[1:] try: cb = channel_callbacks[channel] except KeyError: raise AssertionError('Invalid sideband channel %d' % channel) else: if cb is not None: cb(pkt) def _handle_receive_pack_head(self, proto, capabilities, old_refs, new_refs): """Handle the head of a 'git-receive-pack' request. :param proto: Protocol object to read from :param capabilities: List of negotiated capabilities :param old_refs: Old refs, as received from the server :param new_refs: Refs to change :return: (have, want) tuple """ want = [] have = [x for x in old_refs.values() if not x == ZERO_SHA] sent_capabilities = False for refname in new_refs: if not isinstance(refname, bytes): raise TypeError('refname is not a bytestring: %r' % refname) old_sha1 = old_refs.get(refname, ZERO_SHA) if not isinstance(old_sha1, bytes): raise TypeError('old sha1 for %s is not a bytestring: %r' % (refname, old_sha1)) new_sha1 = new_refs.get(refname, ZERO_SHA) if not isinstance(new_sha1, bytes): raise TypeError('old sha1 for %s is not a bytestring %r' % (refname, new_sha1)) if old_sha1 != new_sha1: if sent_capabilities: proto.write_pkt_line(old_sha1 + b' ' + new_sha1 + b' ' + refname) else: proto.write_pkt_line( old_sha1 + b' ' + new_sha1 + b' ' + refname + b'\0' + b' '.join(capabilities)) sent_capabilities = True if new_sha1 not in have and new_sha1 != ZERO_SHA: want.append(new_sha1) proto.write_pkt_line(None) return (have, want) def _negotiate_receive_pack_capabilities(self, server_capabilities): negotiated_capabilities = ( self._send_capabilities & server_capabilities) unknown_capabilities = ( # noqa: F841 extract_capability_names(server_capabilities) - KNOWN_RECEIVE_CAPABILITIES) # TODO(jelmer): warn about unknown capabilities return negotiated_capabilities def _handle_receive_pack_tail(self, proto, capabilities, progress=None): """Handle the tail of a 'git-receive-pack' request. :param proto: Protocol object to read from :param capabilities: List of negotiated capabilities :param progress: Optional progress reporting function """ if CAPABILITY_SIDE_BAND_64K in capabilities: if progress is None: def progress(x): pass channel_callbacks = {2: progress} if CAPABILITY_REPORT_STATUS in capabilities: channel_callbacks[1] = PktLineParser( self._report_status_parser.handle_packet).parse self._read_side_band64k_data(proto, channel_callbacks) else: if CAPABILITY_REPORT_STATUS in capabilities: for pkt in proto.read_pkt_seq(): self._report_status_parser.handle_packet(pkt) if self._report_status_parser is not None: self._report_status_parser.check() def _negotiate_upload_pack_capabilities(self, server_capabilities): unknown_capabilities = ( # noqa: F841 extract_capability_names(server_capabilities) - KNOWN_UPLOAD_CAPABILITIES) # TODO(jelmer): warn about unknown capabilities symrefs = {} agent = None for capability in server_capabilities: k, v = parse_capability(capability) if k == CAPABILITY_SYMREF: (src, dst) = v.split(b':', 1) symrefs[src] = dst if k == CAPABILITY_AGENT: agent = v negotiated_capabilities = ( self._fetch_capabilities & server_capabilities) return (negotiated_capabilities, symrefs, agent) def _handle_upload_pack_head(self, proto, capabilities, graph_walker, wants, can_read, depth): """Handle the head of a 'git-upload-pack' request. :param proto: Protocol object to read from :param capabilities: List of negotiated capabilities :param graph_walker: GraphWalker instance to call .ack() on :param wants: List of commits to fetch :param can_read: function that returns a boolean that indicates whether there is extra graph data to read on proto :param depth: Depth for request """ assert isinstance(wants, list) and isinstance(wants[0], bytes) proto.write_pkt_line(COMMAND_WANT + b' ' + wants[0] + b' ' + b' '.join(capabilities) + b'\n') for want in wants[1:]: proto.write_pkt_line(COMMAND_WANT + b' ' + want + b'\n') if depth not in (0, None) or getattr(graph_walker, 'shallow', None): if CAPABILITY_SHALLOW not in capabilities: raise GitProtocolError( "server does not support shallow capability required for " "depth") for sha in graph_walker.shallow: proto.write_pkt_line(COMMAND_SHALLOW + b' ' + sha + b'\n') - proto.write_pkt_line(b'%s %d\n' % (COMMAND_DEEPEN, depth)) + proto.write_pkt_line(COMMAND_DEEPEN + b' ' + + str(depth).encode('ascii') + b'\n') proto.write_pkt_line(None) if can_read is not None: (new_shallow, new_unshallow) = _read_shallow_updates(proto) else: new_shallow = new_unshallow = None else: new_shallow = new_unshallow = set() proto.write_pkt_line(None) have = next(graph_walker) while have: proto.write_pkt_line(COMMAND_HAVE + b' ' + have + b'\n') if can_read is not None and can_read(): pkt = proto.read_pkt_line() parts = pkt.rstrip(b'\n').split(b' ') if parts[0] == b'ACK': graph_walker.ack(parts[1]) if parts[2] in (b'continue', b'common'): pass elif parts[2] == b'ready': break else: raise AssertionError( "%s not in ('continue', 'ready', 'common)" % parts[2]) have = next(graph_walker) proto.write_pkt_line(COMMAND_DONE + b'\n') return (new_shallow, new_unshallow) def _handle_upload_pack_tail(self, proto, capabilities, graph_walker, pack_data, progress=None, rbufsize=_RBUFSIZE): """Handle the tail of a 'git-upload-pack' request. :param proto: Protocol object to read from :param capabilities: List of negotiated capabilities :param graph_walker: GraphWalker instance to call .ack() on :param pack_data: Function to call with pack data :param progress: Optional progress reporting function :param rbufsize: Read buffer size """ pkt = proto.read_pkt_line() while pkt: parts = pkt.rstrip(b'\n').split(b' ') if parts[0] == b'ACK': graph_walker.ack(parts[1]) if len(parts) < 3 or parts[2] not in ( b'ready', b'continue', b'common'): break pkt = proto.read_pkt_line() if CAPABILITY_SIDE_BAND_64K in capabilities: if progress is None: # Just ignore progress data def progress(x): pass self._read_side_band64k_data(proto, { SIDE_BAND_CHANNEL_DATA: pack_data, SIDE_BAND_CHANNEL_PROGRESS: progress} ) else: while True: data = proto.read(rbufsize) if data == b"": break pack_data(data) def check_wants(wants, refs): """Check that a set of wants is valid. :param wants: Set of object SHAs to fetch :param refs: Refs dictionary to check against """ missing = set(wants) - { v for (k, v) in refs.items() if not k.endswith(ANNOTATED_TAG_SUFFIX)} if missing: raise InvalidWants(missing) def remote_error_from_stderr(stderr): """Return an appropriate exception based on stderr output. """ if stderr is None: return HangupException() for l in stderr.readlines(): if l.startswith(b'ERROR: '): return GitProtocolError( l[len(b'ERROR: '):].decode('utf-8', 'replace')) return GitProtocolError(l.decode('utf-8', 'replace')) return HangupException() class TraditionalGitClient(GitClient): """Traditional Git client.""" DEFAULT_ENCODING = 'utf-8' def __init__(self, path_encoding=DEFAULT_ENCODING, **kwargs): self._remote_path_encoding = path_encoding super(TraditionalGitClient, self).__init__(**kwargs) def _connect(self, cmd, path): """Create a connection to the server. This method is abstract - concrete implementations should implement their own variant which connects to the server and returns an initialized Protocol object with the service ready for use and a can_read function which may be used to see if reads would block. :param cmd: The git service name to which we should connect. :param path: The path we should pass to the service. (as bytestirng) """ raise NotImplementedError() def send_pack(self, path, update_refs, generate_pack_data, progress=None): """Upload a pack to a remote repository. :param path: Repository path (as bytestring) :param update_refs: Function to determine changes to remote refs. Receive dict with existing remote refs, returns dict with changed refs (name -> sha, where sha=ZERO_SHA for deletions) :param generate_pack_data: Function that can return a tuple with number of objects and pack data to upload. :param progress: Optional callback called with progress updates :raises SendPackError: if server rejects the pack data :raises UpdateRefsError: if the server supports report-status and rejects ref updates :return: new_refs dictionary containing the changes that were made {refname: new_ref}, including deleted refs. """ proto, unused_can_read, stderr = self._connect(b'receive-pack', path) with proto: try: old_refs, server_capabilities = read_pkt_refs(proto) except HangupException: raise remote_error_from_stderr(stderr) negotiated_capabilities = \ self._negotiate_receive_pack_capabilities(server_capabilities) if CAPABILITY_REPORT_STATUS in negotiated_capabilities: self._report_status_parser = ReportStatusParser() report_status_parser = self._report_status_parser try: new_refs = orig_new_refs = update_refs(dict(old_refs)) except BaseException: proto.write_pkt_line(None) raise if CAPABILITY_DELETE_REFS not in server_capabilities: # Server does not support deletions. Fail later. new_refs = dict(orig_new_refs) for ref, sha in orig_new_refs.items(): if sha == ZERO_SHA: if CAPABILITY_REPORT_STATUS in negotiated_capabilities: report_status_parser._ref_statuses.append( b'ng ' + sha + b' remote does not support deleting refs') report_status_parser._ref_status_ok = False del new_refs[ref] if new_refs is None: proto.write_pkt_line(None) return old_refs if len(new_refs) == 0 and len(orig_new_refs): # NOOP - Original new refs filtered out by policy proto.write_pkt_line(None) if report_status_parser is not None: report_status_parser.check() return old_refs (have, want) = self._handle_receive_pack_head( proto, negotiated_capabilities, old_refs, new_refs) if (not want and set(new_refs.items()).issubset(set(old_refs.items()))): return new_refs pack_data_count, pack_data = generate_pack_data( have, want, ofs_delta=(CAPABILITY_OFS_DELTA in negotiated_capabilities)) dowrite = bool(pack_data_count) dowrite = dowrite or any(old_refs.get(ref) != sha for (ref, sha) in new_refs.items() if sha != ZERO_SHA) if dowrite: write_pack_data(proto.write_file(), pack_data_count, pack_data) self._handle_receive_pack_tail( proto, negotiated_capabilities, progress) return new_refs def fetch_pack(self, path, determine_wants, graph_walker, pack_data, progress=None, depth=None): """Retrieve a pack from a git smart server. :param path: Remote path to fetch from :param determine_wants: Function determine what refs to fetch. Receives dictionary of name->sha, should return list of shas to fetch. :param graph_walker: Object with next() and ack(). :param pack_data: Callback called for each bit of data in the pack :param progress: Callback for progress reports (strings) :param depth: Shallow fetch depth :return: FetchPackResult object """ proto, can_read, stderr = self._connect(b'upload-pack', path) with proto: try: refs, server_capabilities = read_pkt_refs(proto) except HangupException: raise remote_error_from_stderr(stderr) negotiated_capabilities, symrefs, agent = ( self._negotiate_upload_pack_capabilities( server_capabilities)) if refs is None: proto.write_pkt_line(None) return FetchPackResult(refs, symrefs, agent) try: wants = determine_wants(refs) except BaseException: proto.write_pkt_line(None) raise if wants is not None: wants = [cid for cid in wants if cid != ZERO_SHA] if not wants: proto.write_pkt_line(None) return FetchPackResult(refs, symrefs, agent) check_wants(wants, refs) (new_shallow, new_unshallow) = self._handle_upload_pack_head( proto, negotiated_capabilities, graph_walker, wants, can_read, depth=depth) self._handle_upload_pack_tail( proto, negotiated_capabilities, graph_walker, pack_data, progress) return FetchPackResult( refs, symrefs, agent, new_shallow, new_unshallow) def get_refs(self, path): """Retrieve the current refs from a git smart server.""" # stock `git ls-remote` uses upload-pack proto, _, stderr = self._connect(b'upload-pack', path) with proto: try: refs, _ = read_pkt_refs(proto) except HangupException: raise remote_error_from_stderr(stderr) proto.write_pkt_line(None) return refs def archive(self, path, committish, write_data, progress=None, write_error=None, format=None, subdirs=None, prefix=None): proto, can_read, stderr = self._connect(b'upload-archive', path) with proto: if format is not None: proto.write_pkt_line(b"argument --format=" + format) proto.write_pkt_line(b"argument " + committish) if subdirs is not None: for subdir in subdirs: proto.write_pkt_line(b"argument " + subdir) if prefix is not None: proto.write_pkt_line(b"argument --prefix=" + prefix) proto.write_pkt_line(None) try: pkt = proto.read_pkt_line() except HangupException: raise remote_error_from_stderr(stderr) if pkt == b"NACK\n": return elif pkt == b"ACK\n": pass elif pkt.startswith(b"ERR "): raise GitProtocolError( pkt[4:].rstrip(b"\n").decode('utf-8', 'replace')) else: raise AssertionError("invalid response %r" % pkt) ret = proto.read_pkt_line() if ret is not None: raise AssertionError("expected pkt tail") self._read_side_band64k_data(proto, { SIDE_BAND_CHANNEL_DATA: write_data, SIDE_BAND_CHANNEL_PROGRESS: progress, SIDE_BAND_CHANNEL_FATAL: write_error}) class TCPGitClient(TraditionalGitClient): """A Git Client that works over TCP directly (i.e. git://).""" def __init__(self, host, port=None, **kwargs): if port is None: port = TCP_GIT_PORT self._host = host self._port = port super(TCPGitClient, self).__init__(**kwargs) @classmethod def from_parsedurl(cls, parsedurl, **kwargs): return cls(parsedurl.hostname, port=parsedurl.port, **kwargs) def get_url(self, path): netloc = self._host if self._port is not None and self._port != TCP_GIT_PORT: netloc += ":%d" % self._port return urlparse.urlunsplit(("git", netloc, path, '', '')) def _connect(self, cmd, path): if not isinstance(cmd, bytes): raise TypeError(cmd) if not isinstance(path, bytes): path = path.encode(self._remote_path_encoding) sockaddrs = socket.getaddrinfo( self._host, self._port, socket.AF_UNSPEC, socket.SOCK_STREAM) s = None err = socket.error("no address found for %s" % self._host) for (family, socktype, proto, canonname, sockaddr) in sockaddrs: s = socket.socket(family, socktype, proto) s.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1) try: s.connect(sockaddr) break except socket.error as e: err = e if s is not None: s.close() s = None if s is None: raise err # -1 means system default buffering rfile = s.makefile('rb', -1) # 0 means unbuffered wfile = s.makefile('wb', 0) def close(): rfile.close() wfile.close() s.close() proto = Protocol(rfile.read, wfile.write, close, report_activity=self._report_activity) if path.startswith(b"/~"): path = path[1:] # TODO(jelmer): Alternative to ascii? proto.send_cmd( b'git-' + cmd, path, b'host=' + self._host.encode('ascii')) return proto, lambda: _fileno_can_read(s), None class SubprocessWrapper(object): """A socket-like object that talks to a subprocess via pipes.""" def __init__(self, proc): self.proc = proc if sys.version_info[0] == 2: self.read = proc.stdout.read else: self.read = BufferedReader(proc.stdout).read self.write = proc.stdin.write @property def stderr(self): return self.proc.stderr def can_read(self): if sys.platform == 'win32': from msvcrt import get_osfhandle handle = get_osfhandle(self.proc.stdout.fileno()) return _win32_peek_avail(handle) != 0 else: return _fileno_can_read(self.proc.stdout.fileno()) def close(self): self.proc.stdin.close() self.proc.stdout.close() if self.proc.stderr: self.proc.stderr.close() self.proc.wait() def find_git_command(): """Find command to run for system Git (usually C Git). """ if sys.platform == 'win32': # support .exe, .bat and .cmd try: # to avoid overhead import win32api except ImportError: # run through cmd.exe with some overhead return ['cmd', '/c', 'git'] else: status, git = win32api.FindExecutable('git') return [git] else: return ['git'] class SubprocessGitClient(TraditionalGitClient): """Git client that talks to a server using a subprocess.""" @classmethod def from_parsedurl(cls, parsedurl, **kwargs): return cls(**kwargs) git_command = None def _connect(self, service, path): if not isinstance(service, bytes): raise TypeError(service) if isinstance(path, bytes): path = path.decode(self._remote_path_encoding) if self.git_command is None: git_command = find_git_command() argv = git_command + [service.decode('ascii'), path] p = subprocess.Popen(argv, bufsize=0, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE) pw = SubprocessWrapper(p) return (Protocol(pw.read, pw.write, pw.close, report_activity=self._report_activity), pw.can_read, p.stderr) class LocalGitClient(GitClient): """Git Client that just uses a local Repo.""" def __init__(self, thin_packs=True, report_activity=None, config=None): """Create a new LocalGitClient instance. :param thin_packs: Whether or not thin packs should be retrieved :param report_activity: Optional callback for reporting transport activity. """ self._report_activity = report_activity # Ignore the thin_packs argument def get_url(self, path): return urlparse.urlunsplit(('file', '', path, '', '')) @classmethod def from_parsedurl(cls, parsedurl, **kwargs): return cls(**kwargs) @classmethod def _open_repo(cls, path): from dulwich.repo import Repo if not isinstance(path, str): path = path.decode(sys.getfilesystemencoding()) return closing(Repo(path)) def send_pack(self, path, update_refs, generate_pack_data, progress=None): """Upload a pack to a remote repository. :param path: Repository path (as bytestring) :param update_refs: Function to determine changes to remote refs. Receive dict with existing remote refs, returns dict with changed refs (name -> sha, where sha=ZERO_SHA for deletions) :param generate_pack_data: Function that can return a tuple with number of items and pack data to upload. :param progress: Optional progress function :raises SendPackError: if server rejects the pack data :raises UpdateRefsError: if the server supports report-status and rejects ref updates :return: new_refs dictionary containing the changes that were made {refname: new_ref}, including deleted refs. """ if not progress: def progress(x): pass with self._open_repo(path) as target: old_refs = target.get_refs() new_refs = update_refs(dict(old_refs)) have = [sha1 for sha1 in old_refs.values() if sha1 != ZERO_SHA] want = [] for refname, new_sha1 in new_refs.items(): if (new_sha1 not in have and new_sha1 not in want and new_sha1 != ZERO_SHA): want.append(new_sha1) if (not want and set(new_refs.items()).issubset(set(old_refs.items()))): return new_refs target.object_store.add_pack_data( *generate_pack_data(have, want, ofs_delta=True)) for refname, new_sha1 in new_refs.items(): old_sha1 = old_refs.get(refname, ZERO_SHA) if new_sha1 != ZERO_SHA: if not target.refs.set_if_equals( refname, old_sha1, new_sha1): progress('unable to set %s to %s' % (refname, new_sha1)) else: if not target.refs.remove_if_equals(refname, old_sha1): progress('unable to remove %s' % refname) return new_refs def fetch(self, path, target, determine_wants=None, progress=None, depth=None): """Fetch into a target repository. :param path: Path to fetch from (as bytestring) :param target: Target repository to fetch into :param determine_wants: Optional function determine what refs to fetch. Receives dictionary of name->sha, should return list of shas to fetch. Defaults to all shas. :param progress: Optional progress function :param depth: Shallow fetch depth :return: FetchPackResult object """ with self._open_repo(path) as r: refs = r.fetch(target, determine_wants=determine_wants, progress=progress, depth=depth) return FetchPackResult(refs, r.refs.get_symrefs(), agent_string()) def fetch_pack(self, path, determine_wants, graph_walker, pack_data, progress=None, depth=None): """Retrieve a pack from a git smart server. :param path: Remote path to fetch from :param determine_wants: Function determine what refs to fetch. Receives dictionary of name->sha, should return list of shas to fetch. :param graph_walker: Object with next() and ack(). :param pack_data: Callback called for each bit of data in the pack :param progress: Callback for progress reports (strings) :param depth: Shallow fetch depth :return: FetchPackResult object """ with self._open_repo(path) as r: objects_iter = r.fetch_objects( determine_wants, graph_walker, progress=progress, depth=depth) symrefs = r.refs.get_symrefs() agent = agent_string() # Did the process short-circuit (e.g. in a stateless RPC call)? # Note that the client still expects a 0-object pack in most cases. if objects_iter is None: return FetchPackResult(None, symrefs, agent) protocol = ProtocolFile(None, pack_data) write_pack_objects(protocol, objects_iter) return FetchPackResult(r.get_refs(), symrefs, agent) def get_refs(self, path): """Retrieve the current refs from a git smart server.""" with self._open_repo(path) as target: return target.get_refs() # What Git client to use for local access default_local_git_client_cls = LocalGitClient class SSHVendor(object): """A client side SSH implementation.""" def connect_ssh(self, host, command, username=None, port=None, password=None, key_filename=None): # This function was deprecated in 0.9.1 import warnings warnings.warn( "SSHVendor.connect_ssh has been renamed to SSHVendor.run_command", DeprecationWarning) return self.run_command(host, command, username=username, port=port, password=password, key_filename=key_filename) def run_command(self, host, command, username=None, port=None, password=None, key_filename=None): """Connect to an SSH server. Run a command remotely and return a file-like object for interaction with the remote command. :param host: Host name :param command: Command to run (as argv array) :param username: Optional ame of user to log in as :param port: Optional SSH port to use :param password: Optional ssh password for login or private key :param key_filename: Optional path to private keyfile """ raise NotImplementedError(self.run_command) class StrangeHostname(Exception): """Refusing to connect to strange SSH hostname.""" def __init__(self, hostname): super(StrangeHostname, self).__init__(hostname) class SubprocessSSHVendor(SSHVendor): """SSH vendor that shells out to the local 'ssh' command.""" def run_command(self, host, command, username=None, port=None, password=None, key_filename=None): if password is not None: raise NotImplementedError( "Setting password not supported by SubprocessSSHVendor.") args = ['ssh', '-x'] if port: args.extend(['-p', str(port)]) if key_filename: args.extend(['-i', str(key_filename)]) if username: host = '%s@%s' % (username, host) if host.startswith('-'): raise StrangeHostname(hostname=host) args.append(host) proc = subprocess.Popen(args + [command], bufsize=0, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE) return SubprocessWrapper(proc) class PLinkSSHVendor(SSHVendor): """SSH vendor that shells out to the local 'plink' command.""" def run_command(self, host, command, username=None, port=None, password=None, key_filename=None): if sys.platform == 'win32': args = ['plink.exe', '-ssh'] else: args = ['plink', '-ssh'] if password is not None: import warnings warnings.warn( "Invoking PLink with a password exposes the password in the " "process list.") args.extend(['-pw', str(password)]) if port: args.extend(['-P', str(port)]) if key_filename: args.extend(['-i', str(key_filename)]) if username: host = '%s@%s' % (username, host) if host.startswith('-'): raise StrangeHostname(hostname=host) args.append(host) proc = subprocess.Popen(args + [command], bufsize=0, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE) return SubprocessWrapper(proc) def ParamikoSSHVendor(**kwargs): import warnings warnings.warn( "ParamikoSSHVendor has been moved to dulwich.contrib.paramiko_vendor.", DeprecationWarning) from dulwich.contrib.paramiko_vendor import ParamikoSSHVendor return ParamikoSSHVendor(**kwargs) # Can be overridden by users get_ssh_vendor = SubprocessSSHVendor class SSHGitClient(TraditionalGitClient): def __init__(self, host, port=None, username=None, vendor=None, config=None, password=None, key_filename=None, **kwargs): self.host = host self.port = port self.username = username self.password = password self.key_filename = key_filename super(SSHGitClient, self).__init__(**kwargs) self.alternative_paths = {} if vendor is not None: self.ssh_vendor = vendor else: self.ssh_vendor = get_ssh_vendor() def get_url(self, path): netloc = self.host if self.port is not None: netloc += ":%d" % self.port if self.username is not None: netloc = urlquote(self.username, '@/:') + "@" + netloc return urlparse.urlunsplit(('ssh', netloc, path, '', '')) @classmethod def from_parsedurl(cls, parsedurl, **kwargs): return cls(host=parsedurl.hostname, port=parsedurl.port, username=parsedurl.username, **kwargs) def _get_cmd_path(self, cmd): cmd = self.alternative_paths.get(cmd, b'git-' + cmd) assert isinstance(cmd, bytes) return cmd def _connect(self, cmd, path): if not isinstance(cmd, bytes): raise TypeError(cmd) if isinstance(path, bytes): path = path.decode(self._remote_path_encoding) if path.startswith("/~"): path = path[1:] argv = (self._get_cmd_path(cmd).decode(self._remote_path_encoding) + " '" + path + "'") kwargs = {} if self.password is not None: kwargs['password'] = self.password if self.key_filename is not None: kwargs['key_filename'] = self.key_filename con = self.ssh_vendor.run_command( self.host, argv, port=self.port, username=self.username, **kwargs) return (Protocol(con.read, con.write, con.close, report_activity=self._report_activity), con.can_read, getattr(con, 'stderr', None)) def default_user_agent_string(): # Start user agent with "git/", because GitHub requires this. :-( See # https://github.com/jelmer/dulwich/issues/562 for details. return "git/dulwich/%s" % ".".join([str(x) for x in dulwich.__version__]) def default_urllib3_manager(config, **override_kwargs): """Return `urllib3` connection pool manager. Honour detected proxy configurations. :param config: `dulwich.config.ConfigDict` instance with Git configuration. :param kwargs: Additional arguments for urllib3.ProxyManager :return: `urllib3.ProxyManager` instance for proxy configurations, `urllib3.PoolManager` otherwise. """ proxy_server = user_agent = None ca_certs = ssl_verify = None if config is not None: try: proxy_server = config.get(b"http", b"proxy") except KeyError: pass try: user_agent = config.get(b"http", b"useragent") except KeyError: pass # TODO(jelmer): Support per-host settings try: ssl_verify = config.get_boolean(b"http", b"sslVerify") except KeyError: ssl_verify = True try: ca_certs = config.get_boolean(b"http", b"sslCAInfo") except KeyError: ca_certs = None if user_agent is None: user_agent = default_user_agent_string() headers = {"User-agent": user_agent} kwargs = {} if ssl_verify is True: kwargs['cert_reqs'] = "CERT_REQUIRED" elif ssl_verify is False: kwargs['cert_reqs'] = 'CERT_NONE' else: # Default to SSL verification kwargs['cert_reqs'] = "CERT_REQUIRED" if ca_certs is not None: kwargs['ca_certs'] = ca_certs kwargs.update(override_kwargs) # Try really hard to find a SSL certificate path if 'ca_certs' not in kwargs and kwargs.get('cert_reqs') != 'CERT_NONE': try: import certifi except ImportError: pass else: kwargs['ca_certs'] = certifi.where() import urllib3 if proxy_server is not None: # `urllib3` requires a `str` object in both Python 2 and 3, while # `ConfigDict` coerces entries to `bytes` on Python 3. Compensate. if not isinstance(proxy_server, str): proxy_server = proxy_server.decode() manager = urllib3.ProxyManager(proxy_server, headers=headers, **kwargs) else: manager = urllib3.PoolManager(headers=headers, **kwargs) return manager class HttpGitClient(GitClient): def __init__(self, base_url, dumb=None, pool_manager=None, config=None, username=None, password=None, **kwargs): self._base_url = base_url.rstrip("/") + "/" self._username = username self._password = password self.dumb = dumb if pool_manager is None: self.pool_manager = default_urllib3_manager(config) else: self.pool_manager = pool_manager if username is not None: # No escaping needed: ":" is not allowed in username: # https://tools.ietf.org/html/rfc2617#section-2 credentials = "%s:%s" % (username, password) import urllib3.util basic_auth = urllib3.util.make_headers(basic_auth=credentials) self.pool_manager.headers.update(basic_auth) GitClient.__init__(self, **kwargs) def get_url(self, path): return self._get_url(path).rstrip("/") @classmethod def from_parsedurl(cls, parsedurl, **kwargs): password = parsedurl.password if password is not None: kwargs['password'] = urlunquote(password) username = parsedurl.username if username is not None: kwargs['username'] = urlunquote(username) # TODO(jelmer): This also strips the username parsedurl = parsedurl._replace(netloc=parsedurl.hostname) return cls(urlparse.urlunparse(parsedurl), **kwargs) def __repr__(self): return "%s(%r, dumb=%r)" % ( type(self).__name__, self._base_url, self.dumb) def _get_url(self, path): if not isinstance(path, str): # TODO(jelmer): this is unrelated to the local filesystem; # This is not necessarily the right encoding to decode the path # with. path = path.decode(sys.getfilesystemencoding()) return urlparse.urljoin(self._base_url, path).rstrip("/") + "/" def _http_request(self, url, headers=None, data=None, allow_compression=False): """Perform HTTP request. :param url: Request URL. :param headers: Optional custom headers to override defaults. :param data: Request data. :param allow_compression: Allow GZipped communication. :return: Tuple (`response`, `read`), where response is an `urllib3` response object with additional `content_type` and `redirect_location` properties, and `read` is a consumable read method for the response data. """ req_headers = self.pool_manager.headers.copy() if headers is not None: req_headers.update(headers) req_headers["Pragma"] = "no-cache" if allow_compression: req_headers["Accept-Encoding"] = "gzip" else: req_headers["Accept-Encoding"] = "identity" if data is None: resp = self.pool_manager.request("GET", url, headers=req_headers) else: resp = self.pool_manager.request("POST", url, headers=req_headers, body=data) if resp.status == 404: raise NotGitRepository() elif resp.status != 200: raise GitProtocolError("unexpected http resp %d for %s" % (resp.status, url)) # TODO: Optimization available by adding `preload_content=False` to the # request and just passing the `read` method on instead of going via # `BytesIO`, if we can guarantee that the entire response is consumed # before issuing the next to still allow for connection reuse from the # pool. read = BytesIO(resp.data).read resp.content_type = resp.getheader("Content-Type") resp.redirect_location = resp.get_redirect_location() return resp, read def _discover_references(self, service, base_url): assert base_url[-1] == "/" tail = "info/refs" headers = {"Accept": "*/*"} if self.dumb is not True: tail += "?service=%s" % service.decode('ascii') url = urlparse.urljoin(base_url, tail) resp, read = self._http_request(url, headers, allow_compression=True) if resp.redirect_location: # Something changed (redirect!), so let's update the base URL if not resp.redirect_location.endswith(tail): raise GitProtocolError( "Redirected from URL %s to URL %s without %s" % ( url, resp.redirect_location, tail)) base_url = resp.redirect_location[:-len(tail)] try: self.dumb = not resp.content_type.startswith("application/x-git-") if not self.dumb: proto = Protocol(read, None) # The first line should mention the service try: [pkt] = list(proto.read_pkt_seq()) except ValueError: raise GitProtocolError( "unexpected number of packets received") if pkt.rstrip(b'\n') != (b'# service=' + service): raise GitProtocolError( "unexpected first line %r from smart server" % pkt) return read_pkt_refs(proto) + (base_url, ) else: return read_info_refs(resp), set(), base_url finally: resp.close() def _smart_request(self, service, url, data): assert url[-1] == "/" url = urlparse.urljoin(url, service) result_content_type = "application/x-%s-result" % service headers = { "Content-Type": "application/x-%s-request" % service, "Accept": result_content_type, "Content-Length": str(len(data)), } resp, read = self._http_request(url, headers, data) if resp.content_type != result_content_type: raise GitProtocolError("Invalid content-type from server: %s" % resp.content_type) return resp, read def send_pack(self, path, update_refs, generate_pack_data, progress=None): """Upload a pack to a remote repository. :param path: Repository path (as bytestring) :param update_refs: Function to determine changes to remote refs. Receive dict with existing remote refs, returns dict with changed refs (name -> sha, where sha=ZERO_SHA for deletions) :param generate_pack_data: Function that can return a tuple with number of elements and pack data to upload. :param progress: Optional progress function :raises SendPackError: if server rejects the pack data :raises UpdateRefsError: if the server supports report-status and rejects ref updates :return: new_refs dictionary containing the changes that were made {refname: new_ref}, including deleted refs. """ url = self._get_url(path) old_refs, server_capabilities, url = self._discover_references( b"git-receive-pack", url) negotiated_capabilities = self._negotiate_receive_pack_capabilities( server_capabilities) negotiated_capabilities.add(capability_agent()) if CAPABILITY_REPORT_STATUS in negotiated_capabilities: self._report_status_parser = ReportStatusParser() new_refs = update_refs(dict(old_refs)) if new_refs is None: # Determine wants function is aborting the push. return old_refs if self.dumb: raise NotImplementedError(self.fetch_pack) req_data = BytesIO() req_proto = Protocol(None, req_data.write) (have, want) = self._handle_receive_pack_head( req_proto, negotiated_capabilities, old_refs, new_refs) if not want and set(new_refs.items()).issubset(set(old_refs.items())): return new_refs pack_data_count, pack_data = generate_pack_data( have, want, ofs_delta=(CAPABILITY_OFS_DELTA in negotiated_capabilities)) if pack_data_count: write_pack_data(req_proto.write_file(), pack_data_count, pack_data) resp, read = self._smart_request("git-receive-pack", url, data=req_data.getvalue()) try: resp_proto = Protocol(read, None) self._handle_receive_pack_tail( resp_proto, negotiated_capabilities, progress) return new_refs finally: resp.close() def fetch_pack(self, path, determine_wants, graph_walker, pack_data, progress=None, depth=None): """Retrieve a pack from a git smart server. :param determine_wants: Callback that returns list of commits to fetch :param graph_walker: Object with next() and ack(). :param pack_data: Callback called for each bit of data in the pack :param progress: Callback for progress reports (strings) :param depth: Depth for request :return: FetchPackResult object """ url = self._get_url(path) refs, server_capabilities, url = self._discover_references( b"git-upload-pack", url) negotiated_capabilities, symrefs, agent = ( self._negotiate_upload_pack_capabilities( server_capabilities)) wants = determine_wants(refs) if wants is not None: wants = [cid for cid in wants if cid != ZERO_SHA] if not wants: return FetchPackResult(refs, symrefs, agent) if self.dumb: raise NotImplementedError(self.send_pack) check_wants(wants, refs) req_data = BytesIO() req_proto = Protocol(None, req_data.write) (new_shallow, new_unshallow) = self._handle_upload_pack_head( req_proto, negotiated_capabilities, graph_walker, wants, can_read=None, depth=depth) resp, read = self._smart_request( "git-upload-pack", url, data=req_data.getvalue()) try: resp_proto = Protocol(read, None) if new_shallow is None and new_unshallow is None: (new_shallow, new_unshallow) = _read_shallow_updates( resp_proto) self._handle_upload_pack_tail( resp_proto, negotiated_capabilities, graph_walker, pack_data, progress) return FetchPackResult( refs, symrefs, agent, new_shallow, new_unshallow) finally: resp.close() def get_refs(self, path): """Retrieve the current refs from a git smart server.""" url = self._get_url(path) refs, _, _ = self._discover_references( b"git-upload-pack", url) return refs def get_transport_and_path_from_url(url, config=None, **kwargs): """Obtain a git client from a URL. :param url: URL to open (a unicode string) :param config: Optional config object :param thin_packs: Whether or not thin packs should be retrieved :param report_activity: Optional callback for reporting transport activity. :return: Tuple with client instance and relative path. """ parsed = urlparse.urlparse(url) if parsed.scheme == 'git': return (TCPGitClient.from_parsedurl(parsed, **kwargs), parsed.path) elif parsed.scheme in ('git+ssh', 'ssh'): return SSHGitClient.from_parsedurl(parsed, **kwargs), parsed.path elif parsed.scheme in ('http', 'https'): return HttpGitClient.from_parsedurl( parsed, config=config, **kwargs), parsed.path elif parsed.scheme == 'file': return default_local_git_client_cls.from_parsedurl( parsed, **kwargs), parsed.path raise ValueError("unknown scheme '%s'" % parsed.scheme) def parse_rsync_url(location): """Parse a rsync-style URL.""" if ':' in location and '@' not in location: # SSH with no user@, zero or one leading slash. (host, path) = location.split(':', 1) user = None elif ':' in location: # SSH with user@host:foo. user_host, path = location.split(':', 1) if '@' in user_host: user, host = user_host.rsplit('@', 1) else: user = None host = user_host else: raise ValueError('not a valid rsync-style URL') return (user, host, path) def get_transport_and_path(location, **kwargs): """Obtain a git client from a URL. :param location: URL or path (a string) :param config: Optional config object :param thin_packs: Whether or not thin packs should be retrieved :param report_activity: Optional callback for reporting transport activity. :return: Tuple with client instance and relative path. """ # First, try to parse it as a URL try: return get_transport_and_path_from_url(location, **kwargs) except ValueError: pass if (sys.platform == 'win32' and location[0].isalpha() and location[1:3] == ':\\'): # Windows local path return default_local_git_client_cls(**kwargs), location try: (username, hostname, path) = parse_rsync_url(location) except ValueError: # Otherwise, assume it's a local path. return default_local_git_client_cls(**kwargs), location else: return SSHGitClient(hostname, username=username, **kwargs), path diff --git a/dulwich/refs.py b/dulwich/refs.py index 89c4cfdb..7c44bd5f 100644 --- a/dulwich/refs.py +++ b/dulwich/refs.py @@ -1,922 +1,922 @@ # refs.py -- For dealing with git refs # Copyright (C) 2008-2013 Jelmer Vernooij # # Dulwich is dual-licensed under the Apache License, Version 2.0 and the GNU # General Public License as public by the Free Software Foundation; version 2.0 # or (at your option) any later version. You can redistribute it and/or # modify it under the terms of either of these two licenses. # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # You should have received a copy of the licenses; if not, see # for a copy of the GNU General Public License # and for a copy of the Apache # License, Version 2.0. # """Ref handling. """ import errno import os import sys from dulwich.errors import ( PackedRefsException, RefFormatError, ) from dulwich.objects import ( git_line, valid_hexsha, ZERO_SHA, ) from dulwich.file import ( GitFile, ensure_dir_exists, ) SYMREF = b'ref: ' LOCAL_BRANCH_PREFIX = b'refs/heads/' LOCAL_TAG_PREFIX = b'refs/tags/' BAD_REF_CHARS = set(b'\177 ~^:?*[') ANNOTATED_TAG_SUFFIX = b'^{}' def parse_symref_value(contents): """Parse a symref value. :param contents: Contents to parse :return: Destination """ if contents.startswith(SYMREF): return contents[len(SYMREF):].rstrip(b'\r\n') raise ValueError(contents) def check_ref_format(refname): """Check if a refname is correctly formatted. Implements all the same rules as git-check-ref-format[1]. [1] http://www.kernel.org/pub/software/scm/git/docs/git-check-ref-format.html :param refname: The refname to check :return: True if refname is valid, False otherwise """ # These could be combined into one big expression, but are listed # separately to parallel [1]. if b'/.' in refname or refname.startswith(b'.'): return False if b'/' not in refname: return False if b'..' in refname: return False for i, c in enumerate(refname): if ord(refname[i:i+1]) < 0o40 or c in BAD_REF_CHARS: return False if refname[-1] in b'/.': return False if refname.endswith(b'.lock'): return False if b'@{' in refname: return False if b'\\' in refname: return False return True class RefsContainer(object): """A container for refs.""" def __init__(self, logger=None): self._logger = logger def _log(self, ref, old_sha, new_sha, committer=None, timestamp=None, timezone=None, message=None): if self._logger is None: return if message is None: return self._logger(ref, old_sha, new_sha, committer, timestamp, timezone, message) def set_symbolic_ref(self, name, other, committer=None, timestamp=None, timezone=None, message=None): """Make a ref point at another ref. :param name: Name of the ref to set :param other: Name of the ref to point at :param message: Optional message """ raise NotImplementedError(self.set_symbolic_ref) def get_packed_refs(self): """Get contents of the packed-refs file. :return: Dictionary mapping ref names to SHA1s :note: Will return an empty dictionary when no packed-refs file is present. """ raise NotImplementedError(self.get_packed_refs) def get_peeled(self, name): """Return the cached peeled value of a ref, if available. :param name: Name of the ref to peel :return: The peeled value of the ref. If the ref is known not point to a tag, this will be the SHA the ref refers to. If the ref may point to a tag, but no cached information is available, None is returned. """ return None def import_refs(self, base, other, committer=None, timestamp=None, timezone=None, message=None): for name, value in other.items(): self.set_if_equals(b'/'.join((base, name)), None, value, message=message) def allkeys(self): """All refs present in this container.""" raise NotImplementedError(self.allkeys) def keys(self, base=None): """Refs present in this container. :param base: An optional base to return refs under. :return: An unsorted set of valid refs in this container, including packed refs. """ if base is not None: return self.subkeys(base) else: return self.allkeys() def subkeys(self, base): """Refs present in this container under a base. :param base: The base to return refs under. :return: A set of valid refs in this container under the base; the base prefix is stripped from the ref names returned. """ keys = set() base_len = len(base) + 1 for refname in self.allkeys(): if refname.startswith(base): keys.add(refname[base_len:]) return keys def as_dict(self, base=None): """Return the contents of this container as a dictionary. """ ret = {} keys = self.keys(base) if base is None: base = b'' else: base = base.rstrip(b'/') for key in keys: try: ret[key] = self[(base + b'/' + key).strip(b'/')] except KeyError: continue # Unable to resolve return ret def _check_refname(self, name): """Ensure a refname is valid and lives in refs or is HEAD. HEAD is not a valid refname according to git-check-ref-format, but this class needs to be able to touch HEAD. Also, check_ref_format expects refnames without the leading 'refs/', but this class requires that so it cannot touch anything outside the refs dir (or HEAD). :param name: The name of the reference. :raises KeyError: if a refname is not HEAD or is otherwise not valid. """ if name in (b'HEAD', b'refs/stash'): return if not name.startswith(b'refs/') or not check_ref_format(name[5:]): raise RefFormatError(name) def read_ref(self, refname): """Read a reference without following any references. :param refname: The name of the reference :return: The contents of the ref file, or None if it does not exist. """ contents = self.read_loose_ref(refname) if not contents: contents = self.get_packed_refs().get(refname, None) return contents def read_loose_ref(self, name): """Read a loose reference and return its contents. :param name: the refname to read :return: The contents of the ref file, or None if it does not exist. """ raise NotImplementedError(self.read_loose_ref) def follow(self, name): """Follow a reference name. :return: a tuple of (refnames, sha), wheres refnames are the names of references in the chain """ contents = SYMREF + name depth = 0 refnames = [] while contents.startswith(SYMREF): refname = contents[len(SYMREF):] refnames.append(refname) contents = self.read_ref(refname) if not contents: break depth += 1 if depth > 5: raise KeyError(name) return refnames, contents def _follow(self, name): import warnings warnings.warn( "RefsContainer._follow is deprecated. Use RefsContainer.follow " "instead.", DeprecationWarning) refnames, contents = self.follow(name) if not refnames: return (None, contents) return (refnames[-1], contents) def __contains__(self, refname): if self.read_ref(refname): return True return False def __getitem__(self, name): """Get the SHA1 for a reference name. This method follows all symbolic references. """ _, sha = self.follow(name) if sha is None: raise KeyError(name) return sha def set_if_equals(self, name, old_ref, new_ref, committer=None, timestamp=None, timezone=None, message=None): """Set a refname to new_ref only if it currently equals old_ref. This method follows all symbolic references if applicable for the subclass, and can be used to perform an atomic compare-and-swap operation. :param name: The refname to set. :param old_ref: The old sha the refname must refer to, or None to set unconditionally. :param new_ref: The new sha the refname will refer to. :param message: Message for reflog :return: True if the set was successful, False otherwise. """ raise NotImplementedError(self.set_if_equals) def add_if_new(self, name, ref): """Add a new reference only if it does not already exist. :param name: Ref name :param ref: Ref value :param message: Message for reflog """ raise NotImplementedError(self.add_if_new) def __setitem__(self, name, ref): """Set a reference name to point to the given SHA1. This method follows all symbolic references if applicable for the subclass. :note: This method unconditionally overwrites the contents of a reference. To update atomically only if the reference has not changed, use set_if_equals(). :param name: The refname to set. :param ref: The new sha the refname will refer to. """ self.set_if_equals(name, None, ref) def remove_if_equals(self, name, old_ref, committer=None, timestamp=None, timezone=None, message=None): """Remove a refname only if it currently equals old_ref. This method does not follow symbolic references, even if applicable for the subclass. It can be used to perform an atomic compare-and-delete operation. :param name: The refname to delete. :param old_ref: The old sha the refname must refer to, or None to delete unconditionally. :param message: Message for reflog :return: True if the delete was successful, False otherwise. """ raise NotImplementedError(self.remove_if_equals) def __delitem__(self, name): """Remove a refname. This method does not follow symbolic references, even if applicable for the subclass. :note: This method unconditionally deletes the contents of a reference. To delete atomically only if the reference has not changed, use remove_if_equals(). :param name: The refname to delete. """ self.remove_if_equals(name, None) def get_symrefs(self): """Get a dict with all symrefs in this container. :return: Dictionary mapping source ref to target ref """ ret = {} for src in self.allkeys(): try: dst = parse_symref_value(self.read_ref(src)) except ValueError: pass else: ret[src] = dst return ret class DictRefsContainer(RefsContainer): """RefsContainer backed by a simple dict. This container does not support symbolic or packed references and is not threadsafe. """ def __init__(self, refs, logger=None): super(DictRefsContainer, self).__init__(logger=logger) self._refs = refs self._peeled = {} def allkeys(self): return self._refs.keys() def read_loose_ref(self, name): return self._refs.get(name, None) def get_packed_refs(self): return {} def set_symbolic_ref(self, name, other, committer=None, timestamp=None, timezone=None, message=None): old = self.follow(name)[-1] self._refs[name] = SYMREF + other self._log(name, old, old, committer=committer, timestamp=timestamp, timezone=timezone, message=message) def set_if_equals(self, name, old_ref, new_ref, committer=None, timestamp=None, timezone=None, message=None): if old_ref is not None and self._refs.get(name, ZERO_SHA) != old_ref: return False realnames, _ = self.follow(name) for realname in realnames: self._check_refname(realname) old = self._refs.get(realname) self._refs[realname] = new_ref self._log(realname, old, new_ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def add_if_new(self, name, ref, committer=None, timestamp=None, timezone=None, message=None): if name in self._refs: return False self._refs[name] = ref self._log(name, None, ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def remove_if_equals(self, name, old_ref, committer=None, timestamp=None, timezone=None, message=None): if old_ref is not None and self._refs.get(name, ZERO_SHA) != old_ref: return False try: old = self._refs.pop(name) except KeyError: pass else: self._log(name, old, None, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def get_peeled(self, name): return self._peeled.get(name) def _update(self, refs): """Update multiple refs; intended only for testing.""" # TODO(dborowitz): replace this with a public function that uses # set_if_equal. self._refs.update(refs) def _update_peeled(self, peeled): """Update cached peeled refs; intended only for testing.""" self._peeled.update(peeled) class InfoRefsContainer(RefsContainer): """Refs container that reads refs from a info/refs file.""" def __init__(self, f): self._refs = {} self._peeled = {} for l in f.readlines(): sha, name = l.rstrip(b'\n').split(b'\t') if name.endswith(ANNOTATED_TAG_SUFFIX): name = name[:-3] if not check_ref_format(name): raise ValueError("invalid ref name %r" % name) self._peeled[name] = sha else: if not check_ref_format(name): raise ValueError("invalid ref name %r" % name) self._refs[name] = sha def allkeys(self): return self._refs.keys() def read_loose_ref(self, name): return self._refs.get(name, None) def get_packed_refs(self): return {} def get_peeled(self, name): try: return self._peeled[name] except KeyError: return self._refs[name] class DiskRefsContainer(RefsContainer): """Refs container that reads refs from disk.""" def __init__(self, path, worktree_path=None, logger=None): super(DiskRefsContainer, self).__init__(logger=logger) if getattr(path, 'encode', None) is not None: path = path.encode(sys.getfilesystemencoding()) self.path = path if worktree_path is None: worktree_path = path if getattr(worktree_path, 'encode', None) is not None: worktree_path = worktree_path.encode(sys.getfilesystemencoding()) self.worktree_path = worktree_path self._packed_refs = None self._peeled_refs = None def __repr__(self): return "%s(%r)" % (self.__class__.__name__, self.path) def subkeys(self, base): subkeys = set() path = self.refpath(base) for root, unused_dirs, files in os.walk(path): dir = root[len(path):] if os.path.sep != '/': dir = dir.replace(os.path.sep.encode( sys.getfilesystemencoding()), b"/") dir = dir.strip(b'/') for filename in files: refname = b"/".join(([dir] if dir else []) + [filename]) # check_ref_format requires at least one /, so we prepend the # base before calling it. if check_ref_format(base + b'/' + refname): subkeys.add(refname) for key in self.get_packed_refs(): if key.startswith(base): subkeys.add(key[len(base):].strip(b'/')) return subkeys def allkeys(self): allkeys = set() if os.path.exists(self.refpath(b'HEAD')): allkeys.add(b'HEAD') path = self.refpath(b'') refspath = self.refpath(b'refs') for root, unused_dirs, files in os.walk(refspath): dir = root[len(path):] if os.path.sep != '/': dir = dir.replace( os.path.sep.encode(sys.getfilesystemencoding()), b"/") for filename in files: refname = b"/".join([dir, filename]) if check_ref_format(refname): allkeys.add(refname) allkeys.update(self.get_packed_refs()) return allkeys def refpath(self, name): """Return the disk path of a ref. """ if os.path.sep != "/": name = name.replace( b"/", os.path.sep.encode(sys.getfilesystemencoding())) # TODO: as the 'HEAD' reference is working tree specific, it # should actually not be a part of RefsContainer if name == b'HEAD': return os.path.join(self.worktree_path, name) else: return os.path.join(self.path, name) def get_packed_refs(self): """Get contents of the packed-refs file. :return: Dictionary mapping ref names to SHA1s :note: Will return an empty dictionary when no packed-refs file is present. """ # TODO: invalidate the cache on repacking if self._packed_refs is None: # set both to empty because we want _peeled_refs to be # None if and only if _packed_refs is also None. self._packed_refs = {} self._peeled_refs = {} path = os.path.join(self.path, b'packed-refs') try: f = GitFile(path, 'rb') except IOError as e: if e.errno == errno.ENOENT: return {} raise with f: first_line = next(iter(f)).rstrip() if (first_line.startswith(b'# pack-refs') and b' peeled' in first_line): for sha, name, peeled in read_packed_refs_with_peeled(f): self._packed_refs[name] = sha if peeled: self._peeled_refs[name] = peeled else: f.seek(0) for sha, name in read_packed_refs(f): self._packed_refs[name] = sha return self._packed_refs def get_peeled(self, name): """Return the cached peeled value of a ref, if available. :param name: Name of the ref to peel :return: The peeled value of the ref. If the ref is known not point to a tag, this will be the SHA the ref refers to. If the ref may point to a tag, but no cached information is available, None is returned. """ self.get_packed_refs() if self._peeled_refs is None or name not in self._packed_refs: # No cache: no peeled refs were read, or this ref is loose return None if name in self._peeled_refs: return self._peeled_refs[name] else: # Known not peelable return self[name] def read_loose_ref(self, name): """Read a reference file and return its contents. If the reference file a symbolic reference, only read the first line of the file. Otherwise, only read the first 40 bytes. :param name: the refname to read, relative to refpath :return: The contents of the ref file, or None if the file does not exist. :raises IOError: if any other error occurs """ filename = self.refpath(name) try: with GitFile(filename, 'rb') as f: header = f.read(len(SYMREF)) if header == SYMREF: # Read only the first line return header + next(iter(f)).rstrip(b'\r\n') else: # Read only the first 40 bytes return header + f.read(40 - len(SYMREF)) except IOError as e: if e.errno in (errno.ENOENT, errno.EISDIR): return None raise def _remove_packed_ref(self, name): if self._packed_refs is None: return filename = os.path.join(self.path, b'packed-refs') # reread cached refs from disk, while holding the lock f = GitFile(filename, 'wb') try: self._packed_refs = None self.get_packed_refs() if name not in self._packed_refs: return del self._packed_refs[name] if name in self._peeled_refs: del self._peeled_refs[name] write_packed_refs(f, self._packed_refs, self._peeled_refs) f.close() finally: f.abort() def set_symbolic_ref(self, name, other, committer=None, timestamp=None, timezone=None, message=None): """Make a ref point at another ref. :param name: Name of the ref to set :param other: Name of the ref to point at :param message: Optional message to describe the change """ self._check_refname(name) self._check_refname(other) filename = self.refpath(name) f = GitFile(filename, 'wb') try: f.write(SYMREF + other + b'\n') sha = self.follow(name)[-1] self._log(name, sha, sha, committer=committer, timestamp=timestamp, timezone=timezone, message=message) except BaseException: f.abort() raise else: f.close() def set_if_equals(self, name, old_ref, new_ref, committer=None, timestamp=None, timezone=None, message=None): """Set a refname to new_ref only if it currently equals old_ref. This method follows all symbolic references, and can be used to perform an atomic compare-and-swap operation. :param name: The refname to set. :param old_ref: The old sha the refname must refer to, or None to set unconditionally. :param new_ref: The new sha the refname will refer to. :param message: Set message for reflog :return: True if the set was successful, False otherwise. """ self._check_refname(name) try: realnames, _ = self.follow(name) realname = realnames[-1] except (KeyError, IndexError): realname = name filename = self.refpath(realname) ensure_dir_exists(os.path.dirname(filename)) with GitFile(filename, 'wb') as f: if old_ref is not None: try: # read again while holding the lock orig_ref = self.read_loose_ref(realname) if orig_ref is None: orig_ref = self.get_packed_refs().get( realname, ZERO_SHA) if orig_ref != old_ref: f.abort() return False except (OSError, IOError): f.abort() raise try: f.write(new_ref + b'\n') except (OSError, IOError): f.abort() raise self._log(realname, old_ref, new_ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def add_if_new(self, name, ref, committer=None, timestamp=None, timezone=None, message=None): """Add a new reference only if it does not already exist. This method follows symrefs, and only ensures that the last ref in the chain does not exist. :param name: The refname to set. :param ref: The new sha the refname will refer to. :param message: Optional message for reflog :return: True if the add was successful, False otherwise. """ try: realnames, contents = self.follow(name) if contents is not None: return False realname = realnames[-1] except (KeyError, IndexError): realname = name self._check_refname(realname) filename = self.refpath(realname) ensure_dir_exists(os.path.dirname(filename)) with GitFile(filename, 'wb') as f: if os.path.exists(filename) or name in self.get_packed_refs(): f.abort() return False try: f.write(ref + b'\n') except (OSError, IOError): f.abort() raise else: self._log(name, None, ref, committer=committer, timestamp=timestamp, timezone=timezone, message=message) return True def remove_if_equals(self, name, old_ref, committer=None, timestamp=None, timezone=None, message=None): """Remove a refname only if it currently equals old_ref. This method does not follow symbolic references. It can be used to perform an atomic compare-and-delete operation. :param name: The refname to delete. :param old_ref: The old sha the refname must refer to, or None to delete unconditionally. :param message: Optional message :return: True if the delete was successful, False otherwise. """ self._check_refname(name) filename = self.refpath(name) ensure_dir_exists(os.path.dirname(filename)) f = GitFile(filename, 'wb') try: if old_ref is not None: orig_ref = self.read_loose_ref(name) if orig_ref is None: orig_ref = self.get_packed_refs().get(name, ZERO_SHA) if orig_ref != old_ref: return False # remove the reference file itself try: os.remove(filename) except OSError as e: if e.errno != errno.ENOENT: # may only be packed raise self._remove_packed_ref(name) self._log(name, old_ref, None, committer=committer, timestamp=timestamp, timezone=timezone, message=message) finally: # never write, we just wanted the lock f.abort() # outside of the lock, clean-up any parent directory that might now # be empty. this ensures that re-creating a reference of the same # name of what was previously a directory works as expected parent = name while True: try: parent, _ = parent.rsplit(b'/', 1) except ValueError: break parent_filename = self.refpath(parent) try: os.rmdir(parent_filename) - except OSError as e: + except OSError: # this can be caused by the parent directory being # removed by another process, being not empty, etc. # in any case, this is non fatal because we already # removed the reference, just ignore it break return True def _split_ref_line(line): """Split a single ref line into a tuple of SHA1 and name.""" fields = line.rstrip(b'\n\r').split(b' ') if len(fields) != 2: raise PackedRefsException("invalid ref line %r" % line) sha, name = fields if not valid_hexsha(sha): raise PackedRefsException("Invalid hex sha %r" % sha) if not check_ref_format(name): raise PackedRefsException("invalid ref name %r" % name) return (sha, name) def read_packed_refs(f): """Read a packed refs file. :param f: file-like object to read from :return: Iterator over tuples with SHA1s and ref names. """ for l in f: if l.startswith(b'#'): # Comment continue if l.startswith(b'^'): raise PackedRefsException( "found peeled ref in packed-refs without peeled") yield _split_ref_line(l) def read_packed_refs_with_peeled(f): """Read a packed refs file including peeled refs. Assumes the "# pack-refs with: peeled" line was already read. Yields tuples with ref names, SHA1s, and peeled SHA1s (or None). :param f: file-like object to read from, seek'ed to the second line """ last = None for line in f: if line[0] == b'#': continue line = line.rstrip(b'\r\n') if line.startswith(b'^'): if not last: raise PackedRefsException("unexpected peeled ref line") if not valid_hexsha(line[1:]): raise PackedRefsException("Invalid hex sha %r" % line[1:]) sha, name = _split_ref_line(last) last = None yield (sha, name, line[1:]) else: if last: sha, name = _split_ref_line(last) yield (sha, name, None) last = line if last: sha, name = _split_ref_line(last) yield (sha, name, None) def write_packed_refs(f, packed_refs, peeled_refs=None): """Write a packed refs file. :param f: empty file-like object to write to :param packed_refs: dict of refname to sha of packed refs to write :param peeled_refs: dict of refname to peeled value of sha """ if peeled_refs is None: peeled_refs = {} else: f.write(b'# pack-refs with: peeled\n') for refname in sorted(packed_refs.keys()): f.write(git_line(packed_refs[refname], refname)) if refname in peeled_refs: f.write(b'^' + peeled_refs[refname] + b'\n') def read_info_refs(f): ret = {} for l in f.readlines(): (sha, name) = l.rstrip(b"\r\n").split(b"\t", 1) ret[name] = sha return ret def write_info_refs(refs, store): """Generate info refs.""" for name, sha in sorted(refs.items()): # get_refs() includes HEAD as a special case, but we don't want to # advertise it if name == b'HEAD': continue try: o = store[sha] except KeyError: continue peeled = store.peel_sha(sha) yield o.id + b'\t' + name + b'\n' if o.id != peeled.id: yield peeled.id + b'\t' + name + ANNOTATED_TAG_SUFFIX + b'\n' def is_local_branch(x): return x.startswith(LOCAL_BRANCH_PREFIX) def strip_peeled_refs(refs): """Remove all peeled refs""" return {ref: sha for (ref, sha) in refs.items() if not ref.endswith(ANNOTATED_TAG_SUFFIX)} diff --git a/dulwich/server.py b/dulwich/server.py index ddd01c91..6b77575e 100644 --- a/dulwich/server.py +++ b/dulwich/server.py @@ -1,1147 +1,1147 @@ # server.py -- Implementation of the server side git protocols # Copyright (C) 2008 John Carr # Coprygith (C) 2011-2012 Jelmer Vernooij # # Dulwich is dual-licensed under the Apache License, Version 2.0 and the GNU # General Public License as public by the Free Software Foundation; version 2.0 # or (at your option) any later version. You can redistribute it and/or # modify it under the terms of either of these two licenses. # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # You should have received a copy of the licenses; if not, see # for a copy of the GNU General Public License # and for a copy of the Apache # License, Version 2.0. # """Git smart network protocol server implementation. For more detailed implementation on the network protocol, see the Documentation/technical directory in the cgit distribution, and in particular: * Documentation/technical/protocol-capabilities.txt * Documentation/technical/pack-protocol.txt Currently supported capabilities: * include-tag * thin-pack * multi_ack_detailed * multi_ack * side-band-64k * ofs-delta * no-progress * report-status * delete-refs * shallow * symref """ import collections import os import socket import sys import zlib try: import SocketServer except ImportError: import socketserver as SocketServer from dulwich.errors import ( ApplyDeltaError, ChecksumMismatch, GitProtocolError, NotGitRepository, UnexpectedCommandError, ObjectFormatException, ) from dulwich import log_utils from dulwich.objects import ( Commit, valid_hexsha, ) from dulwich.pack import ( write_pack_objects, ) from dulwich.protocol import ( # noqa: F401 BufferedPktLineWriter, capability_agent, CAPABILITIES_REF, CAPABILITY_DELETE_REFS, CAPABILITY_INCLUDE_TAG, CAPABILITY_MULTI_ACK_DETAILED, CAPABILITY_MULTI_ACK, CAPABILITY_NO_DONE, CAPABILITY_NO_PROGRESS, CAPABILITY_OFS_DELTA, CAPABILITY_QUIET, CAPABILITY_REPORT_STATUS, CAPABILITY_SHALLOW, CAPABILITY_SIDE_BAND_64K, CAPABILITY_THIN_PACK, COMMAND_DEEPEN, COMMAND_DONE, COMMAND_HAVE, COMMAND_SHALLOW, COMMAND_UNSHALLOW, COMMAND_WANT, MULTI_ACK, MULTI_ACK_DETAILED, Protocol, ProtocolFile, ReceivableProtocol, SIDE_BAND_CHANNEL_DATA, SIDE_BAND_CHANNEL_PROGRESS, SIDE_BAND_CHANNEL_FATAL, SINGLE_ACK, TCP_GIT_PORT, ZERO_SHA, ack_type, extract_capabilities, extract_want_line_capabilities, symref_capabilities, ) from dulwich.refs import ( ANNOTATED_TAG_SUFFIX, write_info_refs, ) from dulwich.repo import ( Repo, ) logger = log_utils.getLogger(__name__) class Backend(object): """A backend for the Git smart server implementation.""" def open_repository(self, path): """Open the repository at a path. :param path: Path to the repository :raise NotGitRepository: no git repository was found at path :return: Instance of BackendRepo """ raise NotImplementedError(self.open_repository) class BackendRepo(object): """Repository abstraction used by the Git server. The methods required here are a subset of those provided by dulwich.repo.Repo. """ object_store = None refs = None def get_refs(self): """ Get all the refs in the repository :return: dict of name -> sha """ raise NotImplementedError def get_peeled(self, name): """Return the cached peeled value of a ref, if available. :param name: Name of the ref to peel :return: The peeled value of the ref. If the ref is known not point to a tag, this will be the SHA the ref refers to. If no cached information about a tag is available, this method may return None, but it should attempt to peel the tag if possible. """ return None def fetch_objects(self, determine_wants, graph_walker, progress, get_tagged=None): """ Yield the objects required for a list of commits. :param progress: is a callback to send progress messages to the client :param get_tagged: Function that returns a dict of pointed-to sha -> tag sha for including tags. """ raise NotImplementedError class DictBackend(Backend): """Trivial backend that looks up Git repositories in a dictionary.""" def __init__(self, repos): self.repos = repos def open_repository(self, path): logger.debug('Opening repository at %s', path) try: return self.repos[path] except KeyError: raise NotGitRepository( "No git repository was found at %(path)s" % dict(path=path) ) class FileSystemBackend(Backend): """Simple backend looking up Git repositories in the local file system.""" def __init__(self, root=os.sep): super(FileSystemBackend, self).__init__() self.root = (os.path.abspath(root) + os.sep).replace( os.sep * 2, os.sep) def open_repository(self, path): logger.debug('opening repository at %s', path) abspath = os.path.abspath(os.path.join(self.root, path)) + os.sep normcase_abspath = os.path.normcase(abspath) normcase_root = os.path.normcase(self.root) if not normcase_abspath.startswith(normcase_root): raise NotGitRepository( "Path %r not inside root %r" % (path, self.root)) return Repo(abspath) class Handler(object): """Smart protocol command handler base class.""" def __init__(self, backend, proto, http_req=None): self.backend = backend self.proto = proto self.http_req = http_req def handle(self): raise NotImplementedError(self.handle) class PackHandler(Handler): """Protocol handler for packs.""" def __init__(self, backend, proto, http_req=None): super(PackHandler, self).__init__(backend, proto, http_req) self._client_capabilities = None # Flags needed for the no-done capability self._done_received = False @classmethod def capability_line(cls, capabilities): logger.info('Sending capabilities: %s', capabilities) return b"".join([b" " + c for c in capabilities]) @classmethod def capabilities(cls): raise NotImplementedError(cls.capabilities) @classmethod def innocuous_capabilities(cls): return [CAPABILITY_INCLUDE_TAG, CAPABILITY_THIN_PACK, CAPABILITY_NO_PROGRESS, CAPABILITY_OFS_DELTA, capability_agent()] @classmethod def required_capabilities(cls): """Return a list of capabilities that we require the client to have.""" return [] def set_client_capabilities(self, caps): allowable_caps = set(self.innocuous_capabilities()) allowable_caps.update(self.capabilities()) for cap in caps: if cap not in allowable_caps: raise GitProtocolError('Client asked for capability %s that ' 'was not advertised.' % cap) for cap in self.required_capabilities(): if cap not in caps: raise GitProtocolError('Client does not support required ' 'capability %s.' % cap) self._client_capabilities = set(caps) logger.info('Client capabilities: %s', caps) def has_capability(self, cap): if self._client_capabilities is None: raise GitProtocolError('Server attempted to access capability %s ' 'before asking client' % cap) return cap in self._client_capabilities def notify_done(self): self._done_received = True class UploadPackHandler(PackHandler): """Protocol handler for uploading a pack to the client.""" def __init__(self, backend, args, proto, http_req=None, advertise_refs=False): super(UploadPackHandler, self).__init__( backend, proto, http_req=http_req) self.repo = backend.open_repository(args[0]) self._graph_walker = None self.advertise_refs = advertise_refs # A state variable for denoting that the have list is still # being processed, and the client is not accepting any other # data (such as side-band, see the progress method here). self._processing_have_lines = False @classmethod def capabilities(cls): return [CAPABILITY_MULTI_ACK_DETAILED, CAPABILITY_MULTI_ACK, CAPABILITY_SIDE_BAND_64K, CAPABILITY_THIN_PACK, CAPABILITY_OFS_DELTA, CAPABILITY_NO_PROGRESS, CAPABILITY_INCLUDE_TAG, CAPABILITY_SHALLOW, CAPABILITY_NO_DONE] @classmethod def required_capabilities(cls): return (CAPABILITY_SIDE_BAND_64K, CAPABILITY_THIN_PACK, CAPABILITY_OFS_DELTA) def progress(self, message): if (self.has_capability(CAPABILITY_NO_PROGRESS) or self._processing_have_lines): return self.proto.write_sideband(SIDE_BAND_CHANNEL_PROGRESS, message) def get_tagged(self, refs=None, repo=None): """Get a dict of peeled values of tags to their original tag shas. :param refs: dict of refname -> sha of possible tags; defaults to all of the backend's refs. :param repo: optional Repo instance for getting peeled refs; defaults to the backend's repo, if available :return: dict of peeled_sha -> tag_sha, where tag_sha is the sha of a tag whose peeled value is peeled_sha. """ if not self.has_capability(CAPABILITY_INCLUDE_TAG): return {} if refs is None: refs = self.repo.get_refs() if repo is None: repo = getattr(self.repo, "repo", None) if repo is None: # Bail if we don't have a Repo available; this is ok since # clients must be able to handle if the server doesn't include # all relevant tags. # TODO: fix behavior when missing return {} tagged = {} for name, sha in refs.items(): peeled_sha = repo.get_peeled(name) if peeled_sha != sha: tagged[peeled_sha] = sha return tagged def handle(self): def write(x): return self.proto.write_sideband(SIDE_BAND_CHANNEL_DATA, x) graph_walker = _ProtocolGraphWalker( self, self.repo.object_store, self.repo.get_peeled, self.repo.refs.get_symrefs) objects_iter = self.repo.fetch_objects( graph_walker.determine_wants, graph_walker, self.progress, get_tagged=self.get_tagged) # Note the fact that client is only processing responses related # to the have lines it sent, and any other data (including side- # band) will be be considered a fatal error. self._processing_have_lines = True # Did the process short-circuit (e.g. in a stateless RPC call)? Note # that the client still expects a 0-object pack in most cases. # Also, if it also happens that the object_iter is instantiated # with a graph walker with an implementation that talks over the # wire (which is this instance of this class) this will actually # iterate through everything and write things out to the wire. if len(objects_iter) == 0: return # The provided haves are processed, and it is safe to send side- # band data now. self._processing_have_lines = False if not graph_walker.handle_done( not self.has_capability(CAPABILITY_NO_DONE), self._done_received): return self.progress( ("counting objects: %d, done.\n" % len(objects_iter)).encode( 'ascii')) write_pack_objects(ProtocolFile(None, write), objects_iter) # we are done self.proto.write_pkt_line(None) def _split_proto_line(line, allowed): """Split a line read from the wire. :param line: The line read from the wire. :param allowed: An iterable of command names that should be allowed. Command names not listed below as possible return values will be ignored. If None, any commands from the possible return values are allowed. :return: a tuple having one of the following forms: ('want', obj_id) ('have', obj_id) ('done', None) (None, None) (for a flush-pkt) :raise UnexpectedCommandError: if the line cannot be parsed into one of the allowed return values. """ if not line: fields = [None] else: fields = line.rstrip(b'\n').split(b' ', 1) command = fields[0] if allowed is not None and command not in allowed: raise UnexpectedCommandError(command) if len(fields) == 1 and command in (COMMAND_DONE, None): return (command, None) elif len(fields) == 2: if command in (COMMAND_WANT, COMMAND_HAVE, COMMAND_SHALLOW, COMMAND_UNSHALLOW): if not valid_hexsha(fields[1]): raise GitProtocolError("Invalid sha") return tuple(fields) elif command == COMMAND_DEEPEN: return command, int(fields[1]) raise GitProtocolError('Received invalid line from client: %r' % line) def _find_shallow(store, heads, depth): """Find shallow commits according to a given depth. :param store: An ObjectStore for looking up objects. :param heads: Iterable of head SHAs to start walking from. :param depth: The depth of ancestors to include. A depth of one includes only the heads themselves. :return: A tuple of (shallow, not_shallow), sets of SHAs that should be considered shallow and unshallow according to the arguments. Note that these sets may overlap if a commit is reachable along multiple paths. """ parents = {} def get_parents(sha): result = parents.get(sha, None) if not result: result = store[sha].parents parents[sha] = result return result todo = [] # stack of (sha, depth) for head_sha in heads: obj = store.peel_sha(head_sha) if isinstance(obj, Commit): todo.append((obj.id, 1)) not_shallow = set() shallow = set() while todo: sha, cur_depth = todo.pop() if cur_depth < depth: not_shallow.add(sha) new_depth = cur_depth + 1 todo.extend((p, new_depth) for p in get_parents(sha)) else: shallow.add(sha) return shallow, not_shallow def _want_satisfied(store, haves, want, earliest): o = store[want] pending = collections.deque([o]) known = set([want]) while pending: commit = pending.popleft() if commit.id in haves: return True if commit.type_name != b"commit": # non-commit wants are assumed to be satisfied continue for parent in commit.parents: if parent in known: continue known.add(parent) parent_obj = store[parent] # TODO: handle parents with later commit times than children if parent_obj.commit_time >= earliest: pending.append(parent_obj) return False def _all_wants_satisfied(store, haves, wants): """Check whether all the current wants are satisfied by a set of haves. :param store: Object store to retrieve objects from :param haves: A set of commits we know the client has. :param wants: A set of commits the client wants :note: Wants are specified with set_wants rather than passed in since in the current interface they are determined outside this class. """ haves = set(haves) if haves: earliest = min([store[h].commit_time for h in haves]) else: earliest = 0 for want in wants: if not _want_satisfied(store, haves, want, earliest): return False return True class _ProtocolGraphWalker(object): """A graph walker that knows the git protocol. As a graph walker, this class implements ack(), next(), and reset(). It also contains some base methods for interacting with the wire and walking the commit tree. The work of determining which acks to send is passed on to the implementation instance stored in _impl. The reason for this is that we do not know at object creation time what ack level the protocol requires. A call to set_ack_type() is required to set up the implementation, before any calls to next() or ack() are made. """ def __init__(self, handler, object_store, get_peeled, get_symrefs): self.handler = handler self.store = object_store self.get_peeled = get_peeled self.get_symrefs = get_symrefs self.proto = handler.proto self.http_req = handler.http_req self.advertise_refs = handler.advertise_refs self._wants = [] self.shallow = set() self.client_shallow = set() self.unshallow = set() self._cached = False self._cache = [] self._cache_index = 0 self._impl = None def determine_wants(self, heads): """Determine the wants for a set of heads. The given heads are advertised to the client, who then specifies which refs he wants using 'want' lines. This portion of the protocol is the same regardless of ack type, and in fact is used to set the ack type of the ProtocolGraphWalker. If the client has the 'shallow' capability, this method also reads and responds to the 'shallow' and 'deepen' lines from the client. These are not part of the wants per se, but they set up necessary state for walking the graph. Additionally, later code depends on this method consuming everything up to the first 'have' line. :param heads: a dict of refname->SHA1 to advertise :return: a list of SHA1s requested by the client """ symrefs = self.get_symrefs() values = set(heads.values()) if self.advertise_refs or not self.http_req: for i, (ref, sha) in enumerate(sorted(heads.items())): line = sha + b' ' + ref if not i: line += (b'\x00' + self.handler.capability_line( self.handler.capabilities() + symref_capabilities(symrefs.items()))) self.proto.write_pkt_line(line + b'\n') peeled_sha = self.get_peeled(ref) if peeled_sha != sha: self.proto.write_pkt_line( peeled_sha + b' ' + ref + ANNOTATED_TAG_SUFFIX + b'\n') # i'm done.. self.proto.write_pkt_line(None) if self.advertise_refs: return [] # Now client will sending want want want commands want = self.proto.read_pkt_line() if not want: return [] line, caps = extract_want_line_capabilities(want) self.handler.set_client_capabilities(caps) self.set_ack_type(ack_type(caps)) allowed = (COMMAND_WANT, COMMAND_SHALLOW, COMMAND_DEEPEN, None) command, sha = _split_proto_line(line, allowed) want_revs = [] while command == COMMAND_WANT: if sha not in values: raise GitProtocolError( 'Client wants invalid object %s' % sha) want_revs.append(sha) command, sha = self.read_proto_line(allowed) self.set_wants(want_revs) if command in (COMMAND_SHALLOW, COMMAND_DEEPEN): self.unread_proto_line(command, sha) self._handle_shallow_request(want_revs) if self.http_req and self.proto.eof(): # The client may close the socket at this point, expecting a # flush-pkt from the server. We might be ready to send a packfile # at this point, so we need to explicitly short-circuit in this # case. return [] return want_revs def unread_proto_line(self, command, value): if isinstance(value, int): value = str(value).encode('ascii') self.proto.unread_pkt_line(command + b' ' + value) def ack(self, have_ref): if len(have_ref) != 40: raise ValueError("invalid sha %r" % have_ref) return self._impl.ack(have_ref) def reset(self): self._cached = True self._cache_index = 0 def next(self): if not self._cached: if not self._impl and self.http_req: return None return next(self._impl) self._cache_index += 1 if self._cache_index > len(self._cache): return None return self._cache[self._cache_index] __next__ = next def read_proto_line(self, allowed): """Read a line from the wire. :param allowed: An iterable of command names that should be allowed. :return: A tuple of (command, value); see _split_proto_line. :raise UnexpectedCommandError: If an error occurred reading the line. """ return _split_proto_line(self.proto.read_pkt_line(), allowed) def _handle_shallow_request(self, wants): while True: command, val = self.read_proto_line( (COMMAND_DEEPEN, COMMAND_SHALLOW)) if command == COMMAND_DEEPEN: depth = val break self.client_shallow.add(val) self.read_proto_line((None,)) # consume client's flush-pkt shallow, not_shallow = _find_shallow(self.store, wants, depth) # Update self.shallow instead of reassigning it since we passed a # reference to it before this method was called. self.shallow.update(shallow - not_shallow) new_shallow = self.shallow - self.client_shallow unshallow = self.unshallow = not_shallow & self.client_shallow for sha in sorted(new_shallow): self.proto.write_pkt_line(COMMAND_SHALLOW + b' ' + sha) for sha in sorted(unshallow): self.proto.write_pkt_line(COMMAND_UNSHALLOW + b' ' + sha) self.proto.write_pkt_line(None) def notify_done(self): # relay the message down to the handler. self.handler.notify_done() def send_ack(self, sha, ack_type=b''): if ack_type: ack_type = b' ' + ack_type self.proto.write_pkt_line(b'ACK ' + sha + ack_type + b'\n') def send_nak(self): self.proto.write_pkt_line(b'NAK\n') def handle_done(self, done_required, done_received): # Delegate this to the implementation. return self._impl.handle_done(done_required, done_received) def set_wants(self, wants): self._wants = wants def all_wants_satisfied(self, haves): """Check whether all the current wants are satisfied by a set of haves. :param haves: A set of commits we know the client has. :note: Wants are specified with set_wants rather than passed in since in the current interface they are determined outside this class. """ return _all_wants_satisfied(self.store, haves, self._wants) def set_ack_type(self, ack_type): impl_classes = { MULTI_ACK: MultiAckGraphWalkerImpl, MULTI_ACK_DETAILED: MultiAckDetailedGraphWalkerImpl, SINGLE_ACK: SingleAckGraphWalkerImpl, } self._impl = impl_classes[ack_type](self) _GRAPH_WALKER_COMMANDS = (COMMAND_HAVE, COMMAND_DONE, None) class SingleAckGraphWalkerImpl(object): """Graph walker implementation that speaks the single-ack protocol.""" def __init__(self, walker): self.walker = walker self._common = [] def ack(self, have_ref): if not self._common: self.walker.send_ack(have_ref) self._common.append(have_ref) def next(self): command, sha = self.walker.read_proto_line(_GRAPH_WALKER_COMMANDS) if command in (None, COMMAND_DONE): # defer the handling of done self.walker.notify_done() return None elif command == COMMAND_HAVE: return sha __next__ = next def handle_done(self, done_required, done_received): if not self._common: self.walker.send_nak() if done_required and not done_received: # we are not done, especially when done is required; skip # the pack for this request and especially do not handle # the done. return False if not done_received and not self._common: # Okay we are not actually done then since the walker picked # up no haves. This is usually triggered when client attempts # to pull from a source that has no common base_commit. # See: test_server.MultiAckDetailedGraphWalkerImplTestCase.\ # test_multi_ack_stateless_nodone return False return True class MultiAckGraphWalkerImpl(object): """Graph walker implementation that speaks the multi-ack protocol.""" def __init__(self, walker): self.walker = walker self._found_base = False self._common = [] def ack(self, have_ref): self._common.append(have_ref) if not self._found_base: self.walker.send_ack(have_ref, b'continue') if self.walker.all_wants_satisfied(self._common): self._found_base = True # else we blind ack within next def next(self): while True: command, sha = self.walker.read_proto_line(_GRAPH_WALKER_COMMANDS) if command is None: self.walker.send_nak() # in multi-ack mode, a flush-pkt indicates the client wants to # flush but more have lines are still coming continue elif command == COMMAND_DONE: self.walker.notify_done() return None elif command == COMMAND_HAVE: if self._found_base: # blind ack self.walker.send_ack(sha, b'continue') return sha __next__ = next def handle_done(self, done_required, done_received): if done_required and not done_received: # we are not done, especially when done is required; skip # the pack for this request and especially do not handle # the done. return False if not done_received and not self._common: # Okay we are not actually done then since the walker picked # up no haves. This is usually triggered when client attempts # to pull from a source that has no common base_commit. # See: test_server.MultiAckDetailedGraphWalkerImplTestCase.\ # test_multi_ack_stateless_nodone return False # don't nak unless no common commits were found, even if not # everything is satisfied if self._common: self.walker.send_ack(self._common[-1]) else: self.walker.send_nak() return True class MultiAckDetailedGraphWalkerImpl(object): """Graph walker implementation speaking the multi-ack-detailed protocol.""" def __init__(self, walker): self.walker = walker self._common = [] def ack(self, have_ref): # Should only be called iff have_ref is common self._common.append(have_ref) self.walker.send_ack(have_ref, b'common') def next(self): while True: command, sha = self.walker.read_proto_line(_GRAPH_WALKER_COMMANDS) if command is None: if self.walker.all_wants_satisfied(self._common): self.walker.send_ack(self._common[-1], b'ready') self.walker.send_nak() if self.walker.http_req: # The HTTP version of this request a flush-pkt always # signifies an end of request, so we also return # nothing here as if we are done (but not really, as # it depends on whether no-done capability was # specified and that's handled in handle_done which # may or may not call post_nodone_check depending on # that). return None elif command == COMMAND_DONE: # Let the walker know that we got a done. self.walker.notify_done() break elif command == COMMAND_HAVE: # return the sha and let the caller ACK it with the # above ack method. return sha # don't nak unless no common commits were found, even if not # everything is satisfied __next__ = next def handle_done(self, done_required, done_received): if done_required and not done_received: # we are not done, especially when done is required; skip # the pack for this request and especially do not handle # the done. return False if not done_received and not self._common: # Okay we are not actually done then since the walker picked # up no haves. This is usually triggered when client attempts # to pull from a source that has no common base_commit. # See: test_server.MultiAckDetailedGraphWalkerImplTestCase.\ # test_multi_ack_stateless_nodone return False # don't nak unless no common commits were found, even if not # everything is satisfied if self._common: self.walker.send_ack(self._common[-1]) else: self.walker.send_nak() return True class ReceivePackHandler(PackHandler): """Protocol handler for downloading a pack from the client.""" def __init__(self, backend, args, proto, http_req=None, advertise_refs=False): super(ReceivePackHandler, self).__init__( backend, proto, http_req=http_req) self.repo = backend.open_repository(args[0]) self.advertise_refs = advertise_refs @classmethod def capabilities(cls): return [CAPABILITY_REPORT_STATUS, CAPABILITY_DELETE_REFS, CAPABILITY_QUIET, CAPABILITY_OFS_DELTA, CAPABILITY_SIDE_BAND_64K, CAPABILITY_NO_DONE] def _apply_pack(self, refs): all_exceptions = (IOError, OSError, ChecksumMismatch, ApplyDeltaError, AssertionError, socket.error, zlib.error, ObjectFormatException) status = [] will_send_pack = False for command in refs: if command[1] != ZERO_SHA: will_send_pack = True if will_send_pack: # TODO: more informative error messages than just the exception # string try: recv = getattr(self.proto, "recv", None) self.repo.object_store.add_thin_pack(self.proto.read, recv) status.append((b'unpack', b'ok')) except all_exceptions as e: status.append((b'unpack', str(e).replace('\n', ''))) # The pack may still have been moved in, but it may contain # broken objects. We trust a later GC to clean it up. else: # The git protocol want to find a status entry related to unpack # process even if no pack data has been sent. status.append((b'unpack', b'ok')) for oldsha, sha, ref in refs: ref_status = b'ok' try: if sha == ZERO_SHA: if CAPABILITY_DELETE_REFS not in self.capabilities(): raise GitProtocolError( 'Attempted to delete refs without delete-refs ' 'capability.') try: self.repo.refs.remove_if_equals(ref, oldsha) except all_exceptions: ref_status = b'failed to delete' else: try: self.repo.refs.set_if_equals(ref, oldsha, sha) except all_exceptions: ref_status = b'failed to write' - except KeyError as e: + except KeyError: ref_status = b'bad ref' status.append((ref, ref_status)) return status def _report_status(self, status): if self.has_capability(CAPABILITY_SIDE_BAND_64K): writer = BufferedPktLineWriter( lambda d: self.proto.write_sideband(SIDE_BAND_CHANNEL_DATA, d)) write = writer.write def flush(): writer.flush() self.proto.write_pkt_line(None) else: write = self.proto.write_pkt_line def flush(): pass for name, msg in status: if name == b'unpack': write(b'unpack ' + msg + b'\n') elif msg == b'ok': write(b'ok ' + name + b'\n') else: write(b'ng ' + name + b' ' + msg + b'\n') write(None) flush() def handle(self): if self.advertise_refs or not self.http_req: refs = sorted(self.repo.get_refs().items()) symrefs = sorted(self.repo.refs.get_symrefs().items()) if not refs: refs = [(CAPABILITIES_REF, ZERO_SHA)] self.proto.write_pkt_line( refs[0][1] + b' ' + refs[0][0] + b'\0' + self.capability_line( self.capabilities() + symref_capabilities(symrefs)) + b'\n') for i in range(1, len(refs)): ref = refs[i] self.proto.write_pkt_line(ref[1] + b' ' + ref[0] + b'\n') self.proto.write_pkt_line(None) if self.advertise_refs: return client_refs = [] ref = self.proto.read_pkt_line() # if ref is none then client doesnt want to send us anything.. if ref is None: return ref, caps = extract_capabilities(ref) self.set_client_capabilities(caps) # client will now send us a list of (oldsha, newsha, ref) while ref: client_refs.append(ref.split()) ref = self.proto.read_pkt_line() # backend can now deal with this refs and read a pack using self.read status = self._apply_pack(client_refs) # when we have read all the pack from the client, send a status report # if the client asked for it if self.has_capability(CAPABILITY_REPORT_STATUS): self._report_status(status) class UploadArchiveHandler(Handler): def __init__(self, backend, proto, http_req=None): super(UploadArchiveHandler, self).__init__(backend, proto, http_req) def handle(self): # TODO(jelmer) raise NotImplementedError(self.handle) # Default handler classes for git services. DEFAULT_HANDLERS = { b'git-upload-pack': UploadPackHandler, b'git-receive-pack': ReceivePackHandler, # b'git-upload-archive': UploadArchiveHandler, } class TCPGitRequestHandler(SocketServer.StreamRequestHandler): def __init__(self, handlers, *args, **kwargs): self.handlers = handlers SocketServer.StreamRequestHandler.__init__(self, *args, **kwargs) def handle(self): proto = ReceivableProtocol(self.connection.recv, self.wfile.write) command, args = proto.read_cmd() logger.info('Handling %s request, args=%s', command, args) cls = self.handlers.get(command, None) if not callable(cls): raise GitProtocolError('Invalid service %s' % command) h = cls(self.server.backend, args, proto) h.handle() class TCPGitServer(SocketServer.TCPServer): allow_reuse_address = True serve = SocketServer.TCPServer.serve_forever def _make_handler(self, *args, **kwargs): return TCPGitRequestHandler(self.handlers, *args, **kwargs) def __init__(self, backend, listen_addr, port=TCP_GIT_PORT, handlers=None): self.handlers = dict(DEFAULT_HANDLERS) if handlers is not None: self.handlers.update(handlers) self.backend = backend logger.info('Listening for TCP connections on %s:%d', listen_addr, port) SocketServer.TCPServer.__init__(self, (listen_addr, port), self._make_handler) def verify_request(self, request, client_address): logger.info('Handling request from %s', client_address) return True def handle_error(self, request, client_address): logger.exception('Exception happened during processing of request ' 'from %s', client_address) def main(argv=sys.argv): """Entry point for starting a TCP git server.""" import optparse parser = optparse.OptionParser() parser.add_option("-l", "--listen_address", dest="listen_address", default="localhost", help="Binding IP address.") parser.add_option("-p", "--port", dest="port", type=int, default=TCP_GIT_PORT, help="Binding TCP port.") options, args = parser.parse_args(argv) log_utils.default_logging_config() if len(args) > 1: gitdir = args[1] else: gitdir = '.' # TODO(jelmer): Support git-daemon-export-ok and --export-all. backend = FileSystemBackend(gitdir) server = TCPGitServer(backend, options.listen_address, options.port) server.serve_forever() def serve_command(handler_cls, argv=sys.argv, backend=None, inf=sys.stdin, outf=sys.stdout): """Serve a single command. This is mostly useful for the implementation of commands used by e.g. git+ssh. :param handler_cls: `Handler` class to use for the request :param argv: execv-style command-line arguments. Defaults to sys.argv. :param backend: `Backend` to use :param inf: File-like object to read from, defaults to standard input. :param outf: File-like object to write to, defaults to standard output. :return: Exit code for use with sys.exit. 0 on success, 1 on failure. """ if backend is None: backend = FileSystemBackend() def send_fn(data): outf.write(data) outf.flush() proto = Protocol(inf.read, send_fn) handler = handler_cls(backend, argv[1:], proto) # FIXME: Catch exceptions and write a single-line summary to outf. handler.handle() return 0 def generate_info_refs(repo): """Generate an info refs file.""" refs = repo.get_refs() return write_info_refs(refs, repo.object_store) def generate_objects_info_packs(repo): """Generate an index for for packs.""" for pack in repo.object_store.packs: yield ( b'P ' + pack.data.filename.encode(sys.getfilesystemencoding()) + b'\n') def update_server_info(repo): """Generate server info for dumb file access. This generates info/refs and objects/info/packs, similar to "git update-server-info". """ repo._put_named_file( os.path.join('info', 'refs'), b"".join(generate_info_refs(repo))) repo._put_named_file( os.path.join('objects', 'info', 'packs'), b"".join(generate_objects_info_packs(repo))) if __name__ == '__main__': main()