diff --git a/dulwich/hooks.py b/dulwich/hooks.py index def9c600..32f98844 100644 --- a/dulwich/hooks.py +++ b/dulwich/hooks.py @@ -1,162 +1,202 @@ # hooks.py -- for dealing with git hooks # Copyright (C) 2012-2013 Jelmer Vernooij and others. # # 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. # """Access to hooks.""" import os import subprocess import sys import tempfile from dulwich.errors import ( HookError, ) class Hook(object): """Generic hook object.""" def execute(self, *args): """Execute the hook with the given args Args: args: argument list to hook Raises: HookError: hook execution failure Returns: a hook may return a useful value """ raise NotImplementedError(self.execute) class ShellHook(Hook): """Hook by executable file Implements standard githooks(5) [0]: [0] http://www.kernel.org/pub/software/scm/git/docs/githooks.html """ def __init__(self, name, path, numparam, pre_exec_callback=None, post_exec_callback=None, cwd=None): """Setup shell hook definition Args: name: name of hook for error messages path: absolute path to executable file numparam: number of requirements parameters pre_exec_callback: closure for setup before execution Defaults to None. Takes in the variable argument list from the execute functions and returns a modified argument list for the shell hook. post_exec_callback: closure for cleanup after execution Defaults to None. Takes in a boolean for hook success and the modified argument list and returns the final hook return value if applicable cwd: working directory to switch to when executing the hook """ self.name = name self.filepath = path self.numparam = numparam self.pre_exec_callback = pre_exec_callback self.post_exec_callback = post_exec_callback self.cwd = cwd if sys.version_info[0] == 2 and sys.platform == 'win32': # Python 2 on windows does not support unicode file paths # http://bugs.python.org/issue1759845 self.filepath = self.filepath.encode(sys.getfilesystemencoding()) def execute(self, *args): """Execute the hook with given args""" if len(args) != self.numparam: raise HookError("Hook %s executed with wrong number of args. \ Expected %d. Saw %d. args: %s" % (self.name, self.numparam, len(args), args)) if (self.pre_exec_callback is not None): args = self.pre_exec_callback(*args) try: ret = subprocess.call([self.filepath] + list(args), cwd=self.cwd) if ret != 0: if (self.post_exec_callback is not None): self.post_exec_callback(0, *args) raise HookError("Hook %s exited with non-zero status" % (self.name)) if (self.post_exec_callback is not None): return self.post_exec_callback(1, *args) except OSError: # no file. silent failure. if (self.post_exec_callback is not None): self.post_exec_callback(0, *args) class PreCommitShellHook(ShellHook): """pre-commit shell hook""" def __init__(self, controldir): filepath = os.path.join(controldir, 'hooks', 'pre-commit') ShellHook.__init__(self, 'pre-commit', filepath, 0, cwd=controldir) class PostCommitShellHook(ShellHook): """post-commit shell hook""" def __init__(self, controldir): filepath = os.path.join(controldir, 'hooks', 'post-commit') ShellHook.__init__(self, 'post-commit', filepath, 0, cwd=controldir) class CommitMsgShellHook(ShellHook): """commit-msg shell hook Args: args[0]: commit message Returns: new commit message or None """ def __init__(self, controldir): filepath = os.path.join(controldir, 'hooks', 'commit-msg') def prepare_msg(*args): (fd, path) = tempfile.mkstemp() with os.fdopen(fd, 'wb') as f: f.write(args[0]) return (path,) def clean_msg(success, *args): if success: with open(args[0], 'rb') as f: new_msg = f.read() os.unlink(args[0]) return new_msg os.unlink(args[0]) ShellHook.__init__(self, 'commit-msg', filepath, 1, prepare_msg, clean_msg, controldir) + + +class PostReceiveShellHook(ShellHook): + """post-receive shell hook""" + + def __init__(self, controldir): + self.controldir = controldir + filepath = os.path.join(controldir, 'hooks', 'post-receive') + ShellHook.__init__(self, 'post-receive', filepath, 0) + + def execute(self, client_refs): + # do nothing if the script doesn't exist + if not os.path.exists(self.filepath): + return None + + try: + env = os.environ.copy() + env['GIT_DIR'] = self.controldir + + p = subprocess.Popen( + self.filepath, + stdin=subprocess.PIPE, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + env=env + ) + + # client_refs is a list of (oldsha, newsha, ref) + in_data = '\n'.join([' '.join(ref) for ref in client_refs]) + + out_data, err_data = p.communicate(in_data) + + if (p.returncode != 0) or err_data: + err_fmt = "post-receive exit code: %d\n" \ + + "stdout:\n%s\nstderr:\n%s" + err_msg = err_fmt % (p.returncode, out_data, err_data) + raise HookError(err_msg) + return out_data + except OSError as err: + raise HookError(repr(err)) diff --git a/dulwich/repo.py b/dulwich/repo.py index 1e1a4c42..187cac45 100644 --- a/dulwich/repo.py +++ b/dulwich/repo.py @@ -1,1480 +1,1482 @@ # repo.py -- For dealing with git repositories. # Copyright (C) 2007 James Westby # 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. # """Repository access. This module contains the base class for git repositories (BaseRepo) and an implementation which uses a repository on local disk (Repo). """ from io import BytesIO import errno import os import sys import stat import time from dulwich.errors import ( NoIndexPresent, NotBlobError, NotCommitError, NotGitRepository, NotTreeError, NotTagError, CommitError, RefFormatError, HookError, ) from dulwich.file import ( GitFile, ) from dulwich.object_store import ( DiskObjectStore, MemoryObjectStore, ObjectStoreGraphWalker, ) from dulwich.objects import ( check_hexsha, Blob, Commit, ShaFile, Tag, Tree, ) from dulwich.pack import ( pack_objects_to_data, ) from dulwich.hooks import ( PreCommitShellHook, PostCommitShellHook, CommitMsgShellHook, + PostReceiveShellHook, ) from dulwich.line_ending import BlobNormalizer from dulwich.refs import ( # noqa: F401 ANNOTATED_TAG_SUFFIX, check_ref_format, RefsContainer, DictRefsContainer, InfoRefsContainer, DiskRefsContainer, read_packed_refs, read_packed_refs_with_peeled, write_packed_refs, SYMREF, ) import warnings CONTROLDIR = '.git' OBJECTDIR = 'objects' REFSDIR = 'refs' REFSDIR_TAGS = 'tags' REFSDIR_HEADS = 'heads' INDEX_FILENAME = "index" COMMONDIR = 'commondir' GITDIR = 'gitdir' WORKTREES = 'worktrees' BASE_DIRECTORIES = [ ["branches"], [REFSDIR], [REFSDIR, REFSDIR_TAGS], [REFSDIR, REFSDIR_HEADS], ["hooks"], ["info"] ] DEFAULT_REF = b'refs/heads/master' class InvalidUserIdentity(Exception): """User identity is not of the format 'user '""" def __init__(self, identity): self.identity = identity def _get_default_identity(): import getpass import socket username = getpass.getuser() try: import pwd except ImportError: fullname = None else: try: gecos = pwd.getpwnam(username).pw_gecos except KeyError: fullname = None else: fullname = gecos.split(',')[0] if not fullname: fullname = username email = os.environ.get('EMAIL') if email is None: email = "{}@{}".format(username, socket.gethostname()) return (fullname, email) def get_user_identity(config, kind=None): """Determine the identity to use for new commits. """ if kind: user = os.environ.get("GIT_" + kind + "_NAME") if user is not None: user = user.encode('utf-8') email = os.environ.get("GIT_" + kind + "_EMAIL") if email is not None: email = email.encode('utf-8') else: user = None email = None if user is None: try: user = config.get(("user", ), "name") except KeyError: user = None if email is None: try: email = config.get(("user", ), "email") except KeyError: email = None default_user, default_email = _get_default_identity() if user is None: user = default_user if not isinstance(user, bytes): user = user.encode('utf-8') if email is None: email = default_email if not isinstance(email, bytes): email = email.encode('utf-8') if email.startswith(b'<') and email.endswith(b'>'): email = email[1:-1] return (user + b" <" + email + b">") def check_user_identity(identity): """Verify that a user identity is formatted correctly. Args: identity: User identity bytestring Raises: InvalidUserIdentity: Raised when identity is invalid """ try: fst, snd = identity.split(b' <', 1) except ValueError: raise InvalidUserIdentity(identity) if b'>' not in snd: raise InvalidUserIdentity(identity) def parse_graftpoints(graftpoints): """Convert a list of graftpoints into a dict Args: graftpoints: Iterator of graftpoint lines Each line is formatted as: []* Resulting dictionary is: : [*] https://git.wiki.kernel.org/index.php/GraftPoint """ grafts = {} for l in graftpoints: raw_graft = l.split(None, 1) commit = raw_graft[0] if len(raw_graft) == 2: parents = raw_graft[1].split() else: parents = [] for sha in [commit] + parents: check_hexsha(sha, 'Invalid graftpoint') grafts[commit] = parents return grafts def serialize_graftpoints(graftpoints): """Convert a dictionary of grafts into string The graft dictionary is: : [*] Each line is formatted as: []* https://git.wiki.kernel.org/index.php/GraftPoint """ graft_lines = [] for commit, parents in graftpoints.items(): if parents: graft_lines.append(commit + b' ' + b' '.join(parents)) else: graft_lines.append(commit) return b'\n'.join(graft_lines) def _set_filesystem_hidden(path): """Mark path as to be hidden if supported by platform and filesystem. On win32 uses SetFileAttributesW api: """ if sys.platform == 'win32': import ctypes from ctypes.wintypes import BOOL, DWORD, LPCWSTR FILE_ATTRIBUTE_HIDDEN = 2 SetFileAttributesW = ctypes.WINFUNCTYPE(BOOL, LPCWSTR, DWORD)( ("SetFileAttributesW", ctypes.windll.kernel32)) if isinstance(path, bytes): path = path.decode(sys.getfilesystemencoding()) if not SetFileAttributesW(path, FILE_ATTRIBUTE_HIDDEN): pass # Could raise or log `ctypes.WinError()` here # Could implement other platform specific filesytem hiding here class BaseRepo(object): """Base class for a git repository. :ivar object_store: Dictionary-like object for accessing the objects :ivar refs: Dictionary-like object with the refs in this repository """ def __init__(self, object_store, refs): """Open a repository. This shouldn't be called directly, but rather through one of the base classes, such as MemoryRepo or Repo. Args: object_store: Object store to use refs: Refs container to use """ self.object_store = object_store self.refs = refs self._graftpoints = {} self.hooks = {} def _determine_file_mode(self): """Probe the file-system to determine whether permissions can be trusted. Returns: True if permissions can be trusted, False otherwise. """ raise NotImplementedError(self._determine_file_mode) def _init_files(self, bare): """Initialize a default set of named files.""" from dulwich.config import ConfigFile self._put_named_file('description', b"Unnamed repository") f = BytesIO() cf = ConfigFile() cf.set("core", "repositoryformatversion", "0") if self._determine_file_mode(): cf.set("core", "filemode", True) else: cf.set("core", "filemode", False) cf.set("core", "bare", bare) cf.set("core", "logallrefupdates", True) cf.write_to_file(f) self._put_named_file('config', f.getvalue()) self._put_named_file(os.path.join('info', 'exclude'), b'') def get_named_file(self, path): """Get a file from the control dir with a specific name. Although the filename should be interpreted as a filename relative to the control dir in a disk-based Repo, the object returned need not be pointing to a file in that location. Args: path: The path to the file, relative to the control dir. Returns: An open file object, or None if the file does not exist. """ raise NotImplementedError(self.get_named_file) def _put_named_file(self, path, contents): """Write a file to the control dir with the given name and contents. Args: path: The path to the file, relative to the control dir. contents: A string to write to the file. """ raise NotImplementedError(self._put_named_file) def _del_named_file(self, path): """Delete a file in the contrl directory with the given name.""" raise NotImplementedError(self._del_named_file) def open_index(self): """Open the index for this repository. Raises: NoIndexPresent: If no index is present Returns: The matching `Index` """ raise NotImplementedError(self.open_index) def fetch(self, target, determine_wants=None, progress=None, depth=None): """Fetch objects into another repository. Args: target: The target repository determine_wants: Optional function to determine what refs to fetch. progress: Optional progress function depth: Optional shallow fetch depth Returns: The local refs """ if determine_wants is None: determine_wants = target.object_store.determine_wants_all count, pack_data = self.fetch_pack_data( determine_wants, target.get_graph_walker(), progress=progress, depth=depth) target.object_store.add_pack_data(count, pack_data, progress) return self.get_refs() def fetch_pack_data(self, determine_wants, graph_walker, progress, get_tagged=None, depth=None): """Fetch the pack data required for a set of revisions. Args: determine_wants: Function that takes a dictionary with heads and returns the list of heads to fetch. graph_walker: Object that can iterate over the list of revisions to fetch and has an "ack" method that will be called to acknowledge that a revision is present. progress: Simple progress function that will be called with updated progress strings. get_tagged: Function that returns a dict of pointed-to sha -> tag sha for including tags. depth: Shallow fetch depth Returns: count and iterator over pack data """ # TODO(jelmer): Fetch pack data directly, don't create objects first. objects = self.fetch_objects(determine_wants, graph_walker, progress, get_tagged, depth=depth) return pack_objects_to_data(objects) def fetch_objects(self, determine_wants, graph_walker, progress, get_tagged=None, depth=None): """Fetch the missing objects required for a set of revisions. Args: determine_wants: Function that takes a dictionary with heads and returns the list of heads to fetch. graph_walker: Object that can iterate over the list of revisions to fetch and has an "ack" method that will be called to acknowledge that a revision is present. progress: Simple progress function that will be called with updated progress strings. get_tagged: Function that returns a dict of pointed-to sha -> tag sha for including tags. depth: Shallow fetch depth Returns: iterator over objects, with __len__ implemented """ if depth not in (None, 0): raise NotImplementedError("depth not supported yet") refs = {} for ref, sha in self.get_refs().items(): try: obj = self.object_store[sha] except KeyError: warnings.warn( 'ref %s points at non-present sha %s' % ( ref.decode('utf-8', 'replace'), sha.decode('ascii')), UserWarning) continue else: if isinstance(obj, Tag): refs[ref + ANNOTATED_TAG_SUFFIX] = obj.object[1] refs[ref] = sha wants = determine_wants(refs) if not isinstance(wants, list): raise TypeError("determine_wants() did not return a list") shallows = getattr(graph_walker, 'shallow', frozenset()) unshallows = getattr(graph_walker, 'unshallow', frozenset()) if wants == []: # TODO(dborowitz): find a way to short-circuit that doesn't change # this interface. if shallows or unshallows: # Do not send a pack in shallow short-circuit path return None return [] # If the graph walker is set up with an implementation that can # ACK/NAK to the wire, it will write data to the client through # this call as a side-effect. haves = self.object_store.find_common_revisions(graph_walker) # Deal with shallow requests separately because the haves do # not reflect what objects are missing if shallows or unshallows: # TODO: filter the haves commits from iter_shas. the specific # commits aren't missing. haves = [] def get_parents(commit): if commit.id in shallows: return [] return self.get_parents(commit.id, commit) return self.object_store.iter_shas( self.object_store.find_missing_objects( haves, wants, progress, get_tagged, get_parents=get_parents)) def get_graph_walker(self, heads=None): """Retrieve a graph walker. A graph walker is used by a remote repository (or proxy) to find out which objects are present in this repository. Args: heads: Repository heads to use (optional) Returns: A graph walker object """ if heads is None: heads = [ sha for sha in self.refs.as_dict(b'refs/heads').values() if sha in self.object_store] return ObjectStoreGraphWalker( heads, self.get_parents, shallow=self.get_shallow()) def get_refs(self): """Get dictionary with all refs. Returns: A ``dict`` mapping ref names to SHA1s """ return self.refs.as_dict() def head(self): """Return the SHA1 pointed at by HEAD.""" return self.refs[b'HEAD'] def _get_object(self, sha, cls): assert len(sha) in (20, 40) ret = self.get_object(sha) if not isinstance(ret, cls): if cls is Commit: raise NotCommitError(ret) elif cls is Blob: raise NotBlobError(ret) elif cls is Tree: raise NotTreeError(ret) elif cls is Tag: raise NotTagError(ret) else: raise Exception("Type invalid: %r != %r" % ( ret.type_name, cls.type_name)) return ret def get_object(self, sha): """Retrieve the object with the specified SHA. Args: sha: SHA to retrieve Returns: A ShaFile object Raises: KeyError: when the object can not be found """ return self.object_store[sha] def get_parents(self, sha, commit=None): """Retrieve the parents of a specific commit. If the specific commit is a graftpoint, the graft parents will be returned instead. Args: sha: SHA of the commit for which to retrieve the parents commit: Optional commit matching the sha Returns: List of parents """ try: return self._graftpoints[sha] except KeyError: if commit is None: commit = self[sha] return commit.parents def get_config(self): """Retrieve the config object. Returns: `ConfigFile` object for the ``.git/config`` file. """ raise NotImplementedError(self.get_config) def get_description(self): """Retrieve the description for this repository. Returns: String with the description of the repository as set by the user. """ raise NotImplementedError(self.get_description) def set_description(self, description): """Set the description for this repository. Args: description: Text to set as description for this repository. """ raise NotImplementedError(self.set_description) def get_config_stack(self): """Return a config stack for this repository. This stack accesses the configuration for both this repository itself (.git/config) and the global configuration, which usually lives in ~/.gitconfig. Returns: `Config` instance for this repository """ from dulwich.config import StackedConfig backends = [self.get_config()] + StackedConfig.default_backends() return StackedConfig(backends, writable=backends[0]) def get_shallow(self): """Get the set of shallow commits. Returns: Set of shallow commits. """ f = self.get_named_file('shallow') if f is None: return set() with f: return set(l.strip() for l in f) def update_shallow(self, new_shallow, new_unshallow): """Update the list of shallow objects. Args: new_shallow: Newly shallow objects new_unshallow: Newly no longer shallow objects """ shallow = self.get_shallow() if new_shallow: shallow.update(new_shallow) if new_unshallow: shallow.difference_update(new_unshallow) self._put_named_file( 'shallow', b''.join([sha + b'\n' for sha in shallow])) def get_peeled(self, ref): """Get the peeled value of a ref. Args: ref: The refname to peel. Returns: The fully-peeled SHA1 of a tag object, after peeling all intermediate tags; if the original ref does not point to a tag, this will equal the original SHA1. """ cached = self.refs.get_peeled(ref) if cached is not None: return cached return self.object_store.peel_sha(self.refs[ref]).id def get_walker(self, include=None, *args, **kwargs): """Obtain a walker for this repository. Args: include: Iterable of SHAs of commits to include along with their ancestors. Defaults to [HEAD] exclude: Iterable of SHAs of commits to exclude along with their ancestors, overriding includes. order: ORDER_* constant specifying the order of results. Anything other than ORDER_DATE may result in O(n) memory usage. reverse: If True, reverse the order of output, requiring O(n) memory. max_entries: The maximum number of entries to yield, or None for no limit. paths: Iterable of file or subtree paths to show entries for. rename_detector: diff.RenameDetector object for detecting renames. follow: If True, follow path across renames/copies. Forces a default rename_detector. since: Timestamp to list commits after. until: Timestamp to list commits before. queue_cls: A class to use for a queue of commits, supporting the iterator protocol. The constructor takes a single argument, the Walker. Returns: A `Walker` object """ from dulwich.walk import Walker if include is None: include = [self.head()] if isinstance(include, str): include = [include] kwargs['get_parents'] = lambda commit: self.get_parents( commit.id, commit) return Walker(self.object_store, include, *args, **kwargs) def __getitem__(self, name): """Retrieve a Git object by SHA1 or ref. Args: name: A Git object SHA1 or a ref name Returns: A `ShaFile` object, such as a Commit or Blob Raises: KeyError: when the specified ref or object does not exist """ if not isinstance(name, bytes): raise TypeError("'name' must be bytestring, not %.80s" % type(name).__name__) if len(name) in (20, 40): try: return self.object_store[name] except (KeyError, ValueError): pass try: return self.object_store[self.refs[name]] except RefFormatError: raise KeyError(name) def __contains__(self, name): """Check if a specific Git object or ref is present. Args: name: Git object SHA1 or ref name """ if len(name) in (20, 40): return name in self.object_store or name in self.refs else: return name in self.refs def __setitem__(self, name, value): """Set a ref. Args: name: ref name value: Ref value - either a ShaFile object, or a hex sha """ if name.startswith(b"refs/") or name == b'HEAD': if isinstance(value, ShaFile): self.refs[name] = value.id elif isinstance(value, bytes): self.refs[name] = value else: raise TypeError(value) else: raise ValueError(name) def __delitem__(self, name): """Remove a ref. Args: name: Name of the ref to remove """ if name.startswith(b"refs/") or name == b"HEAD": del self.refs[name] else: raise ValueError(name) def _get_user_identity(self, config, kind=None): """Determine the identity to use for new commits. """ # TODO(jelmer): Deprecate this function in favor of get_user_identity return get_user_identity(config) def _add_graftpoints(self, updated_graftpoints): """Add or modify graftpoints Args: updated_graftpoints: Dict of commit shas to list of parent shas """ # Simple validation for commit, parents in updated_graftpoints.items(): for sha in [commit] + parents: check_hexsha(sha, 'Invalid graftpoint') self._graftpoints.update(updated_graftpoints) def _remove_graftpoints(self, to_remove=[]): """Remove graftpoints Args: to_remove: List of commit shas """ for sha in to_remove: del self._graftpoints[sha] def _read_heads(self, name): f = self.get_named_file(name) if f is None: return [] with f: return [l.strip() for l in f.readlines() if l.strip()] def do_commit(self, message=None, committer=None, author=None, commit_timestamp=None, commit_timezone=None, author_timestamp=None, author_timezone=None, tree=None, encoding=None, ref=b'HEAD', merge_heads=None): """Create a new commit. Args: message: Commit message committer: Committer fullname author: Author fullname (defaults to committer) commit_timestamp: Commit timestamp (defaults to now) commit_timezone: Commit timestamp timezone (defaults to GMT) author_timestamp: Author timestamp (defaults to commit timestamp) author_timezone: Author timestamp timezone (defaults to commit timestamp timezone) tree: SHA1 of the tree root to use (if not specified the current index will be committed). encoding: Encoding ref: Optional ref to commit to (defaults to current branch) merge_heads: Merge heads (defaults to .git/MERGE_HEADS) Returns: New commit SHA1 """ import time c = Commit() if tree is None: index = self.open_index() c.tree = index.commit(self.object_store) else: if len(tree) != 40: raise ValueError("tree must be a 40-byte hex sha string") c.tree = tree try: self.hooks['pre-commit'].execute() except HookError as e: raise CommitError(e) except KeyError: # no hook defined, silent fallthrough pass config = self.get_config_stack() if merge_heads is None: merge_heads = self._read_heads('MERGE_HEADS') if committer is None: committer = get_user_identity(config, kind='COMMITTER') check_user_identity(committer) c.committer = committer if commit_timestamp is None: # FIXME: Support GIT_COMMITTER_DATE environment variable commit_timestamp = time.time() c.commit_time = int(commit_timestamp) if commit_timezone is None: # FIXME: Use current user timezone rather than UTC commit_timezone = 0 c.commit_timezone = commit_timezone if author is None: author = get_user_identity(config, kind='AUTHOR') c.author = author check_user_identity(author) if author_timestamp is None: # FIXME: Support GIT_AUTHOR_DATE environment variable author_timestamp = commit_timestamp c.author_time = int(author_timestamp) if author_timezone is None: author_timezone = commit_timezone c.author_timezone = author_timezone if encoding is None: try: encoding = config.get(('i18n', ), 'commitEncoding') except KeyError: pass # No dice if encoding is not None: c.encoding = encoding if message is None: # FIXME: Try to read commit message from .git/MERGE_MSG raise ValueError("No commit message specified") try: c.message = self.hooks['commit-msg'].execute(message) if c.message is None: c.message = message except HookError as e: raise CommitError(e) except KeyError: # no hook defined, message not modified c.message = message if ref is None: # Create a dangling commit c.parents = merge_heads self.object_store.add_object(c) else: try: old_head = self.refs[ref] c.parents = [old_head] + merge_heads self.object_store.add_object(c) ok = self.refs.set_if_equals( ref, old_head, c.id, message=b"commit: " + message, committer=committer, timestamp=commit_timestamp, timezone=commit_timezone) except KeyError: c.parents = merge_heads self.object_store.add_object(c) ok = self.refs.add_if_new( ref, c.id, message=b"commit: " + message, committer=committer, timestamp=commit_timestamp, timezone=commit_timezone) if not ok: # Fail if the atomic compare-and-swap failed, leaving the # commit and all its objects as garbage. raise CommitError("%s changed during commit" % (ref,)) self._del_named_file('MERGE_HEADS') try: self.hooks['post-commit'].execute() except HookError as e: # silent failure warnings.warn("post-commit hook failed: %s" % e, UserWarning) except KeyError: # no hook defined, silent fallthrough pass return c.id def read_gitfile(f): """Read a ``.git`` file. The first line of the file should start with "gitdir: " Args: f: File-like object to read from Returns: A path """ cs = f.read() if not cs.startswith("gitdir: "): raise ValueError("Expected file to start with 'gitdir: '") return cs[len("gitdir: "):].rstrip("\n") class Repo(BaseRepo): """A git repository backed by local disk. To open an existing repository, call the contructor with the path of the repository. To create a new repository, use the Repo.init class method. """ def __init__(self, root): hidden_path = os.path.join(root, CONTROLDIR) if os.path.isdir(os.path.join(hidden_path, OBJECTDIR)): self.bare = False self._controldir = hidden_path elif (os.path.isdir(os.path.join(root, OBJECTDIR)) and os.path.isdir(os.path.join(root, REFSDIR))): self.bare = True self._controldir = root elif os.path.isfile(hidden_path): self.bare = False with open(hidden_path, 'r') as f: path = read_gitfile(f) self.bare = False self._controldir = os.path.join(root, path) else: raise NotGitRepository( "No git repository was found at %(path)s" % dict(path=root) ) commondir = self.get_named_file(COMMONDIR) if commondir is not None: with commondir: self._commondir = os.path.join( self.controldir(), commondir.read().rstrip(b"\r\n").decode( sys.getfilesystemencoding())) else: self._commondir = self._controldir self.path = root object_store = DiskObjectStore( os.path.join(self.commondir(), OBJECTDIR)) refs = DiskRefsContainer(self.commondir(), self._controldir, logger=self._write_reflog) BaseRepo.__init__(self, object_store, refs) self._graftpoints = {} graft_file = self.get_named_file(os.path.join("info", "grafts"), basedir=self.commondir()) if graft_file: with graft_file: self._graftpoints.update(parse_graftpoints(graft_file)) graft_file = self.get_named_file("shallow", basedir=self.commondir()) if graft_file: with graft_file: self._graftpoints.update(parse_graftpoints(graft_file)) self.hooks['pre-commit'] = PreCommitShellHook(self.controldir()) self.hooks['commit-msg'] = CommitMsgShellHook(self.controldir()) self.hooks['post-commit'] = PostCommitShellHook(self.controldir()) + self.hooks['post-receive'] = PostReceiveShellHook(self.controldir()) def _write_reflog(self, ref, old_sha, new_sha, committer, timestamp, timezone, message): from .reflog import format_reflog_line path = os.path.join( self.controldir(), 'logs', ref.decode(sys.getfilesystemencoding())) try: os.makedirs(os.path.dirname(path)) except OSError as e: if e.errno != errno.EEXIST: raise if committer is None: config = self.get_config_stack() committer = self._get_user_identity(config) check_user_identity(committer) if timestamp is None: timestamp = int(time.time()) if timezone is None: timezone = 0 # FIXME with open(path, 'ab') as f: f.write(format_reflog_line(old_sha, new_sha, committer, timestamp, timezone, message) + b'\n') @classmethod def discover(cls, start='.'): """Iterate parent directories to discover a repository Return a Repo object for the first parent directory that looks like a Git repository. Args: start: The directory to start discovery from (defaults to '.') """ remaining = True path = os.path.abspath(start) while remaining: try: return cls(path) except NotGitRepository: path, remaining = os.path.split(path) raise NotGitRepository( "No git repository was found at %(path)s" % dict(path=start) ) def controldir(self): """Return the path of the control directory.""" return self._controldir def commondir(self): """Return the path of the common directory. For a main working tree, it is identical to controldir(). For a linked working tree, it is the control directory of the main working tree.""" return self._commondir def _determine_file_mode(self): """Probe the file-system to determine whether permissions can be trusted. Returns: True if permissions can be trusted, False otherwise. """ fname = os.path.join(self.path, '.probe-permissions') with open(fname, 'w') as f: f.write('') st1 = os.lstat(fname) try: os.chmod(fname, st1.st_mode ^ stat.S_IXUSR) except EnvironmentError as e: if e.errno == errno.EPERM: return False raise st2 = os.lstat(fname) os.unlink(fname) mode_differs = st1.st_mode != st2.st_mode st2_has_exec = (st2.st_mode & stat.S_IXUSR) != 0 return mode_differs and st2_has_exec def _put_named_file(self, path, contents): """Write a file to the control dir with the given name and contents. Args: path: The path to the file, relative to the control dir. contents: A string to write to the file. """ path = path.lstrip(os.path.sep) with GitFile(os.path.join(self.controldir(), path), 'wb') as f: f.write(contents) def _del_named_file(self, path): try: os.unlink(os.path.join(self.controldir(), path)) except (IOError, OSError) as e: if e.errno == errno.ENOENT: return raise def get_named_file(self, path, basedir=None): """Get a file from the control dir with a specific name. Although the filename should be interpreted as a filename relative to the control dir in a disk-based Repo, the object returned need not be pointing to a file in that location. Args: path: The path to the file, relative to the control dir. basedir: Optional argument that specifies an alternative to the control dir. Returns: An open file object, or None if the file does not exist. """ # TODO(dborowitz): sanitize filenames, since this is used directly by # the dumb web serving code. if basedir is None: basedir = self.controldir() path = path.lstrip(os.path.sep) try: return open(os.path.join(basedir, path), 'rb') except (IOError, OSError) as e: if e.errno == errno.ENOENT: return None raise def index_path(self): """Return path to the index file.""" return os.path.join(self.controldir(), INDEX_FILENAME) def open_index(self): """Open the index for this repository. Raises: NoIndexPresent: If no index is present Returns: The matching `Index` """ from dulwich.index import Index if not self.has_index(): raise NoIndexPresent() return Index(self.index_path()) def has_index(self): """Check if an index is present.""" # Bare repos must never have index files; non-bare repos may have a # missing index file, which is treated as empty. return not self.bare def stage(self, fs_paths): """Stage a set of paths. Args: fs_paths: List of paths, relative to the repository path """ root_path_bytes = self.path.encode(sys.getfilesystemencoding()) if not isinstance(fs_paths, list): fs_paths = [fs_paths] from dulwich.index import ( blob_from_path_and_stat, index_entry_from_stat, _fs_to_tree_path, ) index = self.open_index() blob_normalizer = self.get_blob_normalizer() for fs_path in fs_paths: if not isinstance(fs_path, bytes): fs_path = fs_path.encode(sys.getfilesystemencoding()) if os.path.isabs(fs_path): raise ValueError( "path %r should be relative to " "repository root, not absolute" % fs_path) tree_path = _fs_to_tree_path(fs_path) full_path = os.path.join(root_path_bytes, fs_path) try: st = os.lstat(full_path) except OSError: # File no longer exists try: del index[tree_path] except KeyError: pass # already removed else: if not stat.S_ISDIR(st.st_mode): blob = blob_from_path_and_stat(full_path, st) blob = blob_normalizer.checkin_normalize(blob, fs_path) self.object_store.add_object(blob) index[tree_path] = index_entry_from_stat(st, blob.id, 0) else: try: del index[tree_path] except KeyError: pass index.write() def clone(self, target_path, mkdir=True, bare=False, origin=b"origin", checkout=None): """Clone this repository. Args: target_path: Target path mkdir: Create the target directory bare: Whether to create a bare repository origin: Base name for refs in target repository cloned from this repository Returns: Created repository as `Repo` """ if not bare: target = self.init(target_path, mkdir=mkdir) else: if checkout: raise ValueError("checkout and bare are incompatible") target = self.init_bare(target_path, mkdir=mkdir) self.fetch(target) encoded_path = self.path if not isinstance(encoded_path, bytes): encoded_path = encoded_path.encode(sys.getfilesystemencoding()) ref_message = b"clone: from " + encoded_path target.refs.import_refs( b'refs/remotes/' + origin, self.refs.as_dict(b'refs/heads'), message=ref_message) target.refs.import_refs( b'refs/tags', self.refs.as_dict(b'refs/tags'), message=ref_message) try: target.refs.add_if_new( DEFAULT_REF, self.refs[DEFAULT_REF], message=ref_message) except KeyError: pass target_config = target.get_config() target_config.set(('remote', 'origin'), 'url', encoded_path) target_config.set(('remote', 'origin'), 'fetch', '+refs/heads/*:refs/remotes/origin/*') target_config.write_to_path() # Update target head head_chain, head_sha = self.refs.follow(b'HEAD') if head_chain and head_sha is not None: target.refs.set_symbolic_ref(b'HEAD', head_chain[-1], message=ref_message) target[b'HEAD'] = head_sha if checkout is None: checkout = (not bare) if checkout: # Checkout HEAD to target dir target.reset_index() return target def reset_index(self, tree=None): """Reset the index back to a specific tree. Args: tree: Tree SHA to reset to, None for current HEAD tree. """ from dulwich.index import ( build_index_from_tree, validate_path_element_default, validate_path_element_ntfs, ) if tree is None: tree = self[b'HEAD'].tree config = self.get_config() honor_filemode = config.get_boolean( b'core', b'filemode', os.name != "nt") if config.get_boolean(b'core', b'core.protectNTFS', os.name == "nt"): validate_path_element = validate_path_element_ntfs else: validate_path_element = validate_path_element_default return build_index_from_tree( self.path, self.index_path(), self.object_store, tree, honor_filemode=honor_filemode, validate_path_element=validate_path_element) def get_config(self): """Retrieve the config object. Returns: `ConfigFile` object for the ``.git/config`` file. """ from dulwich.config import ConfigFile path = os.path.join(self._controldir, 'config') try: return ConfigFile.from_path(path) except (IOError, OSError) as e: if e.errno != errno.ENOENT: raise ret = ConfigFile() ret.path = path return ret def get_description(self): """Retrieve the description of this repository. Returns: A string describing the repository or None. """ path = os.path.join(self._controldir, 'description') try: with GitFile(path, 'rb') as f: return f.read() except (IOError, OSError) as e: if e.errno != errno.ENOENT: raise return None def __repr__(self): return "" % self.path def set_description(self, description): """Set the description for this repository. Args: description: Text to set as description for this repository. """ self._put_named_file('description', description) @classmethod def _init_maybe_bare(cls, path, bare): for d in BASE_DIRECTORIES: os.mkdir(os.path.join(path, *d)) DiskObjectStore.init(os.path.join(path, OBJECTDIR)) ret = cls(path) ret.refs.set_symbolic_ref(b'HEAD', DEFAULT_REF) ret._init_files(bare) return ret @classmethod def init(cls, path, mkdir=False): """Create a new repository. Args: path: Path in which to create the repository mkdir: Whether to create the directory Returns: `Repo` instance """ if mkdir: os.mkdir(path) controldir = os.path.join(path, CONTROLDIR) os.mkdir(controldir) _set_filesystem_hidden(controldir) cls._init_maybe_bare(controldir, False) return cls(path) @classmethod def _init_new_working_directory(cls, path, main_repo, identifier=None, mkdir=False): """Create a new working directory linked to a repository. Args: path: Path in which to create the working tree. main_repo: Main repository to reference identifier: Worktree identifier mkdir: Whether to create the directory Returns: `Repo` instance """ if mkdir: os.mkdir(path) if identifier is None: identifier = os.path.basename(path) main_worktreesdir = os.path.join(main_repo.controldir(), WORKTREES) worktree_controldir = os.path.join(main_worktreesdir, identifier) gitdirfile = os.path.join(path, CONTROLDIR) with open(gitdirfile, 'wb') as f: f.write(b'gitdir: ' + worktree_controldir.encode(sys.getfilesystemencoding()) + b'\n') try: os.mkdir(main_worktreesdir) except OSError as e: if e.errno != errno.EEXIST: raise try: os.mkdir(worktree_controldir) except OSError as e: if e.errno != errno.EEXIST: raise with open(os.path.join(worktree_controldir, GITDIR), 'wb') as f: f.write(gitdirfile.encode(sys.getfilesystemencoding()) + b'\n') with open(os.path.join(worktree_controldir, COMMONDIR), 'wb') as f: f.write(b'../..\n') with open(os.path.join(worktree_controldir, 'HEAD'), 'wb') as f: f.write(main_repo.head() + b'\n') r = cls(path) r.reset_index() return r @classmethod def init_bare(cls, path, mkdir=False): """Create a new bare repository. ``path`` should already exist and be an empty directory. Args: path: Path to create bare repository in Returns: a `Repo` instance """ if mkdir: os.mkdir(path) return cls._init_maybe_bare(path, True) create = init_bare def close(self): """Close any files opened by this repository.""" self.object_store.close() def __enter__(self): return self def __exit__(self, exc_type, exc_val, exc_tb): self.close() def get_blob_normalizer(self): """ Return a BlobNormalizer object """ # TODO Parse the git attributes files git_attributes = {} return BlobNormalizer( self.get_config_stack(), git_attributes ) class MemoryRepo(BaseRepo): """Repo that stores refs, objects, and named files in memory. MemoryRepos are always bare: they have no working tree and no index, since those have a stronger dependency on the filesystem. """ def __init__(self): from dulwich.config import ConfigFile self._reflog = [] refs_container = DictRefsContainer({}, logger=self._append_reflog) BaseRepo.__init__(self, MemoryObjectStore(), refs_container) self._named_files = {} self.bare = True self._config = ConfigFile() self._description = None def _append_reflog(self, *args): self._reflog.append(args) def set_description(self, description): self._description = description def get_description(self): return self._description def _determine_file_mode(self): """Probe the file-system to determine whether permissions can be trusted. Returns: True if permissions can be trusted, False otherwise. """ return sys.platform != 'win32' def _put_named_file(self, path, contents): """Write a file to the control dir with the given name and contents. Args: path: The path to the file, relative to the control dir. contents: A string to write to the file. """ self._named_files[path] = contents def _del_named_file(self, path): try: del self._named_files[path] except KeyError: pass def get_named_file(self, path, basedir=None): """Get a file from the control dir with a specific name. Although the filename should be interpreted as a filename relative to the control dir in a disk-baked Repo, the object returned need not be pointing to a file in that location. Args: path: The path to the file, relative to the control dir. Returns: An open file object, or None if the file does not exist. """ contents = self._named_files.get(path, None) if contents is None: return None return BytesIO(contents) def open_index(self): """Fail to open index for this repo, since it is bare. Raises: NoIndexPresent: Raised when no index is present """ raise NoIndexPresent() def get_config(self): """Retrieve the config object. Returns: `ConfigFile` object. """ return self._config @classmethod def init_bare(cls, objects, refs): """Create a new bare repository in memory. Args: objects: Objects for the new repository, as iterable refs: Refs as dictionary, mapping names to object SHA1s """ ret = cls() for obj in objects: ret.object_store.add_object(obj) for refname, sha in refs.items(): ret.refs.add_if_new(refname, sha) ret._init_files(bare=True) return ret diff --git a/dulwich/server.py b/dulwich/server.py index f5af537e..2aae5b6d 100644 --- a/dulwich/server.py +++ b/dulwich/server.py @@ -1,1200 +1,1214 @@ # 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 time import zlib try: import SocketServer except ImportError: import socketserver as SocketServer from dulwich.archive import tar_stream from dulwich.errors import ( ApplyDeltaError, ChecksumMismatch, GitProtocolError, + HookError, 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. Args: path: Path to the repository Raises: NotGitRepository: no git repository was found at path Returns: 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 Returns: dict of name -> sha """ raise NotImplementedError def get_peeled(self, name): """Return the cached peeled value of a ref, if available. Args: name: Name of the ref to peel Returns: 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. Args: progress: is a callback to send progress messages to the client 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. Args: refs: dict of refname -> sha of possible tags; defaults to all of the backend's refs. repo: optional Repo instance for getting peeled refs; defaults to the backend's repo, if available Returns: 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 {} # TODO(jelmer): Integrate this with the refs logic in # Repo.fetch_objects 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. Args: line: The line read from the wire. 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. Returns: a tuple having one of the following forms: ('want', obj_id) ('have', obj_id) ('done', None) (None, None) (for a flush-pkt) Raises: 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. Args: store: An ObjectStore for looking up objects. heads: Iterable of head SHAs to start walking from. depth: The depth of ancestors to include. A depth of one includes only the heads themselves. Returns: 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. Args: store: Object store to retrieve objects from haves: A set of commits we know the client has. 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. Args: heads: a dict of refname->SHA1 to advertise Returns: 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())): try: peeled_sha = self.get_peeled(ref) except KeyError: # Skip refs that are inaccessible # TODO(jelmer): Integrate with Repo.fetch_objects refs # logic. continue 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') 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. Args: allowed: An iterable of command names that should be allowed. Returns: A tuple of (command, value); see _split_proto_line. Raises: 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. Args: 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: 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 _on_post_receive(self, client_refs): + hook = self.repo.hooks.get('post-receive', None) + if not hook: + return + try: + output = hook.execute(client_refs) + if output: + self.proto.write_sideband(SIDE_BAND_CHANNEL_PROGRESS, output) + except HookError as err: + self.proto.write_sideband(SIDE_BAND_CHANNEL_FATAL, repr(err)) + 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) + self._on_post_receive(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, args, proto, http_req=None): super(UploadArchiveHandler, self).__init__(backend, proto, http_req) self.repo = backend.open_repository(args[0]) def handle(self): def write(x): return self.proto.write_sideband(SIDE_BAND_CHANNEL_DATA, x) arguments = [] for pkt in self.proto.read_pkt_seq(): (key, value) = pkt.split(b' ', 1) if key != b'argument': raise GitProtocolError('unknown command %s' % key) arguments.append(value.rstrip(b'\n')) prefix = b'' format = 'tar' i = 0 store = self.repo.object_store while i < len(arguments): argument = arguments[i] if argument == b'--prefix': i += 1 prefix = arguments[i] elif argument == b'--format': i += 1 format = arguments[i].decode('ascii') else: commit_sha = self.repo.refs[argument] tree = store[store[commit_sha].tree] i += 1 self.proto.write_pkt_line(b'ACK\n') self.proto.write_pkt_line(None) for chunk in tar_stream( store, tree, mtime=time.time(), prefix=prefix, format=format): write(chunk) self.proto.write_pkt_line(None) # 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. Args: handler_cls: `Handler` class to use for the request argv: execv-style command-line arguments. Defaults to sys.argv. backend: `Backend` to use inf: File-like object to read from, defaults to standard input. outf: File-like object to write to, defaults to standard output. Returns: 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()