feat(svn): improvements to handle SVN protocol 1.4 features

- added support to actually allow HEAD type calls
- support HEAD / DELETE calls properly with new requets
- use request session for performance optimization on making a lot of calls.

rhodecode/lib/middleware/simplesvn.py

@@ -1,231 +1,239 b''
 
 # Copyright (C) 2010-2023 RhodeCode GmbH
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License, version 3
 # (only), as published by the Free Software Foundation.
 #
 # This program is distributed in the hope that it will be useful,
 # but WITHOUT ANY WARRANTY; without even the implied warranty of
 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 # GNU General Public License for more details.
 #
 # You should have received a copy of the GNU Affero General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 # This program is dual-licensed. If you wish to learn more about the
 # RhodeCode Enterprise Edition, including its added features, Support services,
 # and proprietary license terms, please see https://rhodecode.com/licenses/
 
 import base64
 import logging
 import urllib.request
 import urllib.parse
 import urllib.error
 import urllib.parse
 
 import requests
 from pyramid.httpexceptions import HTTPNotAcceptable
 
 from rhodecode.lib import rc_cache
 from rhodecode.lib.middleware import simplevcs
 from rhodecode.lib.middleware.utils import get_path_info
 from rhodecode.lib.utils import is_valid_repo
 from rhodecode.lib.str_utils import safe_str, safe_int, safe_bytes
 from rhodecode.lib.type_utils import str2bool
 from rhodecode.lib.ext_json import json
 from rhodecode.lib.hooks_daemon import store_txn_id_data
 
 
 log = logging.getLogger(__name__)
 
 
 class SimpleSvnApp(object):
     IGNORED_HEADERS = [
         'connection', 'keep-alive', 'content-encoding',
         'transfer-encoding', 'content-length']
     rc_extras = {}
 
     def __init__(self, config):
         self.config = config
+        self.session = requests.Session()
 
     def __call__(self, environ, start_response):
         request_headers = self._get_request_headers(environ)
         data_io = environ['wsgi.input']
         req_method: str = environ['REQUEST_METHOD']
         has_content_length = 'CONTENT_LENGTH' in environ
 
         path_info = self._get_url(
             self.config.get('subversion_http_server_url', ''), get_path_info(environ))
         transfer_encoding = environ.get('HTTP_TRANSFER_ENCODING', '')
         log.debug('Handling: %s method via `%s`', req_method, path_info)
 
         # stream control flag, based on request and content type...
         stream = False
 
         if req_method in ['MKCOL'] or has_content_length:
             data_processed = False
             # read chunk to check if we have txn-with-props
             initial_data: bytes = data_io.read(1024)
             if initial_data.startswith(b'(create-txn-with-props'):
                 data_io = initial_data + data_io.read()
                 # store on-the-fly our rc_extra using svn revision properties
                 # those can be read later on in hooks executed so we have a way
                 # to pass in the data into svn hooks
                 rc_data = base64.urlsafe_b64encode(json.dumps(self.rc_extras))
                 rc_data_len = str(len(rc_data))
                 # header defines data length, and serialized data
                 skel = b' rc-scm-extras %b %b' % (safe_bytes(rc_data_len), safe_bytes(rc_data))
                 data_io = data_io[:-2] + skel + b'))'
                 data_processed = True
 
             if not data_processed:
                 # NOTE(johbo): Avoid that we end up with sending the request in chunked
                 # transfer encoding (mainly on Gunicorn). If we know the content
                 # length, then we should transfer the payload in one request.
                 data_io = initial_data + data_io.read()
 
         if req_method in ['GET', 'PUT'] or transfer_encoding == 'chunked':
             # NOTE(marcink): when getting/uploading files, we want to STREAM content
             # back to the client/proxy instead of buffering it here...
             stream = True
 
         stream = stream
         log.debug('Calling SVN PROXY at `%s`, using method:%s. Stream: %s',
                   path_info, req_method, stream)
 
+        call_kwargs = dict(
+            data=data_io,
+            headers=request_headers,
+            stream=stream
+        )
+        if req_method in ['HEAD', 'DELETE']:
+            del call_kwargs['data']
+
         try:
-            response = requests.request(
-                req_method, path_info,
-                data=data_io, headers=request_headers, stream=stream)
+            response = self.session.request(
+                req_method, path_info, **call_kwargs)
         except requests.ConnectionError:
             log.exception('ConnectionError occurred for endpoint %s', path_info)
             raise
 
         if response.status_code not in [200, 401]:
             text = '\n{}'.format(safe_str(response.text)) if response.text else ''
             if response.status_code >= 500:
                 log.error('Got SVN response:%s with text:`%s`', response, text)
             else:
                 log.debug('Got SVN response:%s with text:`%s`', response, text)
         else:
             log.debug('got response code: %s', response.status_code)
 
         response_headers = self._get_response_headers(response.headers)
 
         if response.headers.get('SVN-Txn-name'):
             svn_tx_id = response.headers.get('SVN-Txn-name')
             txn_id = rc_cache.utils.compute_key_from_params(
                 self.config['repository'], svn_tx_id)
             port = safe_int(self.rc_extras['hooks_uri'].split(':')[-1])
             store_txn_id_data(txn_id, {'port': port})
 
         start_response(f'{response.status_code} {response.reason}', response_headers)
         return response.iter_content(chunk_size=1024)
 
     def _get_url(self, svn_http_server, path):
         svn_http_server_url = (svn_http_server or '').rstrip('/')
         url_path = urllib.parse.urljoin(svn_http_server_url + '/', (path or '').lstrip('/'))
         url_path = urllib.parse.quote(url_path, safe="/:=~+!$,;'")
         return url_path
 
     def _get_request_headers(self, environ):
         headers = {}
 
         for key in environ:
             if not key.startswith('HTTP_'):
                 continue
             new_key = key.split('_')
             new_key = [k.capitalize() for k in new_key[1:]]
             new_key = '-'.join(new_key)
             headers[new_key] = environ[key]
 
         if 'CONTENT_TYPE' in environ:
             headers['Content-Type'] = environ['CONTENT_TYPE']
 
         if 'CONTENT_LENGTH' in environ:
             headers['Content-Length'] = environ['CONTENT_LENGTH']
 
         return headers
 
     def _get_response_headers(self, headers):
         headers = [
             (h, headers[h])
             for h in headers
             if h.lower() not in self.IGNORED_HEADERS
         ]
 
         return headers
 
 
 class DisabledSimpleSvnApp(object):
     def __init__(self, config):
         self.config = config
 
     def __call__(self, environ, start_response):
         reason = 'Cannot handle SVN call because: SVN HTTP Proxy is not enabled'
         log.warning(reason)
         return HTTPNotAcceptable(reason)(environ, start_response)
 
 
 class SimpleSvn(simplevcs.SimpleVCS):
 
     SCM = 'svn'
     READ_ONLY_COMMANDS = ('OPTIONS', 'PROPFIND', 'GET', 'REPORT')
     DEFAULT_HTTP_SERVER = 'http://localhost:8090'
 
     def _get_repository_name(self, environ):
         """
         Gets repository name out of PATH_INFO header
 
         :param environ: environ where PATH_INFO is stored
         """
         path = get_path_info(environ).split('!')
         repo_name = path[0].strip('/')
 
         # SVN includes the whole path in it's requests, including
         # subdirectories inside the repo. Therefore we have to search for
         # the repo root directory.
         if not is_valid_repo(
                 repo_name, self.base_path, explicit_scm=self.SCM):
             current_path = ''
             for component in repo_name.split('/'):
                 current_path += component
                 if is_valid_repo(
                         current_path, self.base_path, explicit_scm=self.SCM):
                     return current_path
                 current_path += '/'
 
         return repo_name
 
     def _get_action(self, environ):
         return (
             'pull'
             if environ['REQUEST_METHOD'] in self.READ_ONLY_COMMANDS
             else 'push')
 
     def _should_use_callback_daemon(self, extras, environ, action):
         # only MERGE command triggers hooks, so we don't want to start
         # hooks server too many times. POST however starts the svn transaction
         # so we also need to run the init of callback daemon of POST
         if environ['REQUEST_METHOD'] in ['MERGE', 'POST']:
             return True
         return False
 
     def _create_wsgi_app(self, repo_path, repo_name, config):
         if self._is_svn_enabled():
             return SimpleSvnApp(config)
         # we don't have http proxy enabled return dummy request handler
         return DisabledSimpleSvnApp(config)
 
     def _is_svn_enabled(self):
         conf = self.repo_vcs_config
         return str2bool(conf.get('vcs_svn_proxy', 'http_requests_enabled'))
 
     def _create_config(self, extras, repo_name, scheme='http'):
         conf = self.repo_vcs_config
         server_url = conf.get('vcs_svn_proxy', 'http_server_url')
         server_url = server_url or self.DEFAULT_HTTP_SERVER
 
         extras['subversion_http_server_url'] = server_url
         return extras

rhodecode/lib/middleware/simplevcs.py

@@ -1,700 +1,701 b''
 
 
 # Copyright (C) 2014-2023 RhodeCode GmbH
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License, version 3
 # (only), as published by the Free Software Foundation.
 #
 # This program is distributed in the hope that it will be useful,
 # but WITHOUT ANY WARRANTY; without even the implied warranty of
 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 # GNU General Public License for more details.
 #
 # You should have received a copy of the GNU Affero General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 # This program is dual-licensed. If you wish to learn more about the
 # RhodeCode Enterprise Edition, including its added features, Support services,
 # and proprietary license terms, please see https://rhodecode.com/licenses/
 
 """
 SimpleVCS middleware for handling protocol request (push/clone etc.)
 It's implemented with basic auth function
 """
 
 import os
 import re
 import io
 import logging
 import importlib
 from functools import wraps
 from lxml import etree
 
 import time
 from paste.httpheaders import REMOTE_USER, AUTH_TYPE
 
 from pyramid.httpexceptions import (
     HTTPNotFound, HTTPForbidden, HTTPNotAcceptable, HTTPInternalServerError)
 from zope.cachedescriptors.property import Lazy as LazyProperty
 
 import rhodecode
 from rhodecode.authentication.base import authenticate, VCS_TYPE, loadplugin
 from rhodecode.lib import rc_cache
 from rhodecode.lib.auth import AuthUser, HasPermissionAnyMiddleware
 from rhodecode.lib.base import (
     BasicAuth, get_ip_addr, get_user_agent, vcs_operation_context)
 from rhodecode.lib.exceptions import (UserCreationError, NotAllowedToCreateUserError)
 from rhodecode.lib.hooks_daemon import prepare_callback_daemon
 from rhodecode.lib.middleware import appenlight
 from rhodecode.lib.middleware.utils import scm_app_http
 from rhodecode.lib.str_utils import safe_bytes
 from rhodecode.lib.utils import is_valid_repo, SLUG_RE
 from rhodecode.lib.utils2 import safe_str, fix_PATH, str2bool
 from rhodecode.lib.vcs.conf import settings as vcs_settings
 from rhodecode.lib.vcs.backends import base
 
 from rhodecode.model import meta
 from rhodecode.model.db import User, Repository, PullRequest
 from rhodecode.model.scm import ScmModel
 from rhodecode.model.pull_request import PullRequestModel
 from rhodecode.model.settings import SettingsModel, VcsSettingsModel
 
 log = logging.getLogger(__name__)
 
 
-def extract_svn_txn_id(acl_repo_name, data):
+def extract_svn_txn_id(acl_repo_name, data: bytes):
     """
     Helper method for extraction of svn txn_id from submitted XML data during
     POST operations
     """
+
     try:
         root = etree.fromstring(data)
         pat = re.compile(r'/txn/(?P<txn_id>.*)')
         for el in root:
             if el.tag == '{DAV:}source':
                 for sub_el in el:
                     if sub_el.tag == '{DAV:}href':
                         match = pat.search(sub_el.text)
                         if match:
                             svn_tx_id = match.groupdict()['txn_id']
                             txn_id = rc_cache.utils.compute_key_from_params(
                                 acl_repo_name, svn_tx_id)
                             return txn_id
     except Exception:
         log.exception('Failed to extract txn_id')
 
 
 def initialize_generator(factory):
     """
     Initializes the returned generator by draining its first element.
 
     This can be used to give a generator an initializer, which is the code
     up to the first yield statement. This decorator enforces that the first
     produced element has the value ``"__init__"`` to make its special
     purpose very explicit in the using code.
     """
 
     @wraps(factory)
     def wrapper(*args, **kwargs):
         gen = factory(*args, **kwargs)
         try:
             init = next(gen)
         except StopIteration:
             raise ValueError('Generator must yield at least one element.')
         if init != "__init__":
             raise ValueError('First yielded element must be "__init__".')
         return gen
     return wrapper
 
 
 class SimpleVCS(object):
     """Common functionality for SCM HTTP handlers."""
 
     SCM = 'unknown'
 
     acl_repo_name = None
     url_repo_name = None
     vcs_repo_name = None
     rc_extras = {}
 
     # We have to handle requests to shadow repositories different than requests
     # to normal repositories. Therefore we have to distinguish them. To do this
     # we use this regex which will match only on URLs pointing to shadow
     # repositories.
     shadow_repo_re = re.compile(
         '(?P<groups>(?:{slug_pat}/)*)'   # repo groups
         '(?P<target>{slug_pat})/'        # target repo
         'pull-request/(?P<pr_id>\\d+)/'  # pull request
         'repository$'                    # shadow repo
         .format(slug_pat=SLUG_RE.pattern))
 
     def __init__(self, config, registry):
         self.registry = registry
         self.config = config
         # re-populated by specialized middleware
         self.repo_vcs_config = base.Config()
 
         rc_settings = SettingsModel().get_all_settings(cache=True, from_request=False)
         realm = rc_settings.get('rhodecode_realm') or 'RhodeCode AUTH'
 
         # authenticate this VCS request using authfunc
         auth_ret_code_detection = \
             str2bool(self.config.get('auth_ret_code_detection', False))
         self.authenticate = BasicAuth(
             '', authenticate, registry, config.get('auth_ret_code'),
             auth_ret_code_detection, rc_realm=realm)
         self.ip_addr = '0.0.0.0'
 
     @LazyProperty
     def global_vcs_config(self):
         try:
             return VcsSettingsModel().get_ui_settings_as_config_obj()
         except Exception:
             return base.Config()
 
     @property
     def base_path(self):
         settings_path = self.repo_vcs_config.get(*VcsSettingsModel.PATH_SETTING)
 
         if not settings_path:
             settings_path = self.global_vcs_config.get(*VcsSettingsModel.PATH_SETTING)
 
         if not settings_path:
             # try, maybe we passed in explicitly as config option
             settings_path = self.config.get('base_path')
 
         if not settings_path:
             raise ValueError('FATAL: base_path is empty')
         return settings_path
 
     def set_repo_names(self, environ):
         """
         This will populate the attributes acl_repo_name, url_repo_name,
         vcs_repo_name and is_shadow_repo. In case of requests to normal (non
         shadow) repositories all names are equal. In case of requests to a
         shadow repository the acl-name points to the target repo of the pull
         request and the vcs-name points to the shadow repo file system path.
         The url-name is always the URL used by the vcs client program.
 
         Example in case of a shadow repo:
             acl_repo_name = RepoGroup/MyRepo
             url_repo_name = RepoGroup/MyRepo/pull-request/3/repository
             vcs_repo_name = /repo/base/path/RepoGroup/.__shadow_MyRepo_pr-3'
         """
         # First we set the repo name from URL for all attributes. This is the
         # default if handling normal (non shadow) repo requests.
         self.url_repo_name = self._get_repository_name(environ)
         self.acl_repo_name = self.vcs_repo_name = self.url_repo_name
         self.is_shadow_repo = False
 
         # Check if this is a request to a shadow repository.
         match = self.shadow_repo_re.match(self.url_repo_name)
         if match:
             match_dict = match.groupdict()
 
             # Build acl repo name from regex match.
             acl_repo_name = safe_str('{groups}{target}'.format(
                 groups=match_dict['groups'] or '',
                 target=match_dict['target']))
 
             # Retrieve pull request instance by ID from regex match.
             pull_request = PullRequest.get(match_dict['pr_id'])
 
             # Only proceed if we got a pull request and if acl repo name from
             # URL equals the target repo name of the pull request.
             if pull_request and (acl_repo_name == pull_request.target_repo.repo_name):
 
                 # Get file system path to shadow repository.
                 workspace_id = PullRequestModel()._workspace_id(pull_request)
                 vcs_repo_name = pull_request.target_repo.get_shadow_repository_path(workspace_id)
 
                 # Store names for later usage.
                 self.vcs_repo_name = vcs_repo_name
                 self.acl_repo_name = acl_repo_name
                 self.is_shadow_repo = True
 
         log.debug('Setting all VCS repository names: %s', {
             'acl_repo_name': self.acl_repo_name,
             'url_repo_name': self.url_repo_name,
             'vcs_repo_name': self.vcs_repo_name,
         })
 
     @property
     def scm_app(self):
         custom_implementation = self.config['vcs.scm_app_implementation']
         if custom_implementation == 'http':
             log.debug('Using HTTP implementation of scm app.')
             scm_app_impl = scm_app_http
         else:
             log.debug('Using custom implementation of scm_app: "{}"'.format(
                 custom_implementation))
             scm_app_impl = importlib.import_module(custom_implementation)
         return scm_app_impl
 
     def _get_by_id(self, repo_name):
         """
         Gets a special pattern _<ID> from clone url and tries to replace it
         with a repository_name for support of _<ID> non changeable urls
         """
 
         data = repo_name.split('/')
         if len(data) >= 2:
             from rhodecode.model.repo import RepoModel
             by_id_match = RepoModel().get_repo_by_id(repo_name)
             if by_id_match:
                 data[1] = by_id_match.repo_name
 
         # Because PEP-3333-WSGI uses bytes-tunneled-in-latin-1 as PATH_INFO
         # and we use this data
         maybe_new_path = '/'.join(data)
         return safe_bytes(maybe_new_path).decode('latin1')
 
     def _invalidate_cache(self, repo_name):
         """
         Set's cache for this repository for invalidation on next access
 
         :param repo_name: full repo name, also a cache key
         """
         ScmModel().mark_for_invalidation(repo_name)
 
     def is_valid_and_existing_repo(self, repo_name, base_path, scm_type):
         db_repo = Repository.get_by_repo_name(repo_name)
         if not db_repo:
             log.debug('Repository `%s` not found inside the database.',
                       repo_name)
             return False
 
         if db_repo.repo_type != scm_type:
             log.warning(
                 'Repository `%s` have incorrect scm_type, expected %s got %s',
                 repo_name, db_repo.repo_type, scm_type)
             return False
 
         config = db_repo._config
         config.set('extensions', 'largefiles', '')
         return is_valid_repo(
             repo_name, base_path,
             explicit_scm=scm_type, expect_scm=scm_type, config=config)
 
     def valid_and_active_user(self, user):
         """
         Checks if that user is not empty, and if it's actually object it checks
         if he's active.
 
         :param user: user object or None
         :return: boolean
         """
         if user is None:
             return False
 
         elif user.active:
             return True
 
         return False
 
     @property
     def is_shadow_repo_dir(self):
         return os.path.isdir(self.vcs_repo_name)
 
     def _check_permission(self, action, user, auth_user, repo_name, ip_addr=None,
                           plugin_id='', plugin_cache_active=False, cache_ttl=0):
         """
         Checks permissions using action (push/pull) user and repository
         name. If plugin_cache and ttl is set it will use the plugin which
         authenticated the user to store the cached permissions result for N
         amount of seconds as in cache_ttl
 
         :param action: push or pull action
         :param user: user instance
         :param repo_name: repository name
         """
 
         log.debug('AUTH_CACHE_TTL for permissions `%s` active: %s (TTL: %s)',
                   plugin_id, plugin_cache_active, cache_ttl)
 
         user_id = user.user_id
         cache_namespace_uid = f'cache_user_auth.{rc_cache.PERMISSIONS_CACHE_VER}.{user_id}'
         region = rc_cache.get_or_create_region('cache_perms', cache_namespace_uid)
 
         @region.conditional_cache_on_arguments(namespace=cache_namespace_uid,
                                                expiration_time=cache_ttl,
                                                condition=plugin_cache_active)
         def compute_perm_vcs(
                 cache_name, plugin_id, action, user_id, repo_name, ip_addr):
 
             log.debug('auth: calculating permission access now...')
             # check IP
             inherit = user.inherit_default_permissions
             ip_allowed = AuthUser.check_ip_allowed(
                 user_id, ip_addr, inherit_from_default=inherit)
             if ip_allowed:
                 log.info('Access for IP:%s allowed', ip_addr)
             else:
                 return False
 
             if action == 'push':
                 perms = ('repository.write', 'repository.admin')
                 if not HasPermissionAnyMiddleware(*perms)(auth_user, repo_name):
                     return False
 
             else:
                 # any other action need at least read permission
                 perms = (
                     'repository.read', 'repository.write', 'repository.admin')
                 if not HasPermissionAnyMiddleware(*perms)(auth_user, repo_name):
                     return False
 
             return True
 
         start = time.time()
         log.debug('Running plugin `%s` permissions check', plugin_id)
 
         # for environ based auth, password can be empty, but then the validation is
         # on the server that fills in the env data needed for authentication
         perm_result = compute_perm_vcs(
             'vcs_permissions', plugin_id, action, user.user_id, repo_name, ip_addr)
 
         auth_time = time.time() - start
         log.debug('Permissions for plugin `%s` completed in %.4fs, '
                   'expiration time of fetched cache %.1fs.',
                   plugin_id, auth_time, cache_ttl)
 
         return perm_result
 
     def _get_http_scheme(self, environ):
         try:
             return environ['wsgi.url_scheme']
         except Exception:
             log.exception('Failed to read http scheme')
             return 'http'
 
     def _check_ssl(self, environ, start_response):
         """
         Checks the SSL check flag and returns False if SSL is not present
         and required True otherwise
         """
         org_proto = environ['wsgi._org_proto']
         # check if we have SSL required  ! if not it's a bad request !
         require_ssl = str2bool(self.repo_vcs_config.get('web', 'push_ssl'))
         if require_ssl and org_proto == 'http':
             log.debug(
                 'Bad request: detected protocol is `%s` and '
                 'SSL/HTTPS is required.', org_proto)
             return False
         return True
 
     def _get_default_cache_ttl(self):
         # take AUTH_CACHE_TTL from the `rhodecode` auth plugin
         plugin = loadplugin('egg:rhodecode-enterprise-ce#rhodecode')
         plugin_settings = plugin.get_settings()
         plugin_cache_active, cache_ttl = plugin.get_ttl_cache(
             plugin_settings) or (False, 0)
         return plugin_cache_active, cache_ttl
 
     def __call__(self, environ, start_response):
         try:
             return self._handle_request(environ, start_response)
         except Exception:
             log.exception("Exception while handling request")
             appenlight.track_exception(environ)
             return HTTPInternalServerError()(environ, start_response)
         finally:
             meta.Session.remove()
 
     def _handle_request(self, environ, start_response):
         if not self._check_ssl(environ, start_response):
             reason = ('SSL required, while RhodeCode was unable '
                       'to detect this as SSL request')
             log.debug('User not allowed to proceed, %s', reason)
             return HTTPNotAcceptable(reason)(environ, start_response)
 
         if not self.url_repo_name:
             log.warning('Repository name is empty: %s', self.url_repo_name)
             # failed to get repo name, we fail now
             return HTTPNotFound()(environ, start_response)
         log.debug('Extracted repo name is %s', self.url_repo_name)
 
         ip_addr = get_ip_addr(environ)
         user_agent = get_user_agent(environ)
         username = None
 
         # skip passing error to error controller
         environ['pylons.status_code_redirect'] = True
 
         # ======================================================================
         # GET ACTION PULL or PUSH
         # ======================================================================
         action = self._get_action(environ)
 
         # ======================================================================
         # Check if this is a request to a shadow repository of a pull request.
         # In this case only pull action is allowed.
         # ======================================================================
         if self.is_shadow_repo and action != 'pull':
             reason = 'Only pull action is allowed for shadow repositories.'
             log.debug('User not allowed to proceed, %s', reason)
             return HTTPNotAcceptable(reason)(environ, start_response)
 
         # Check if the shadow repo actually exists, in case someone refers
         # to it, and it has been deleted because of successful merge.
         if self.is_shadow_repo and not self.is_shadow_repo_dir:
             log.debug(
                 'Shadow repo detected, and shadow repo dir `%s` is missing',
                 self.is_shadow_repo_dir)
             return HTTPNotFound()(environ, start_response)
 
         # ======================================================================
         # CHECK ANONYMOUS PERMISSION
         # ======================================================================
         detect_force_push = False
         check_branch_perms = False
         if action in ['pull', 'push']:
             user_obj = anonymous_user = User.get_default_user()
             auth_user = user_obj.AuthUser()
             username = anonymous_user.username
             if anonymous_user.active:
                 plugin_cache_active, cache_ttl = self._get_default_cache_ttl()
                 # ONLY check permissions if the user is activated
                 anonymous_perm = self._check_permission(
                     action, anonymous_user, auth_user, self.acl_repo_name, ip_addr,
                     plugin_id='anonymous_access',
                     plugin_cache_active=plugin_cache_active,
                     cache_ttl=cache_ttl,
                 )
             else:
                 anonymous_perm = False
 
             if not anonymous_user.active or not anonymous_perm:
                 if not anonymous_user.active:
                     log.debug('Anonymous access is disabled, running '
                               'authentication')
 
                 if not anonymous_perm:
                     log.debug('Not enough credentials to access repo: `%s` '
                               'repository as anonymous user', self.acl_repo_name)
 
 
                 username = None
                 # ==============================================================
                 # DEFAULT PERM FAILED OR ANONYMOUS ACCESS IS DISABLED SO WE
                 # NEED TO AUTHENTICATE AND ASK FOR AUTH USER PERMISSIONS
                 # ==============================================================
 
                 # try to auth based on environ, container auth methods
                 log.debug('Running PRE-AUTH for container|headers based authentication')
 
                 # headers auth, by just reading special headers and bypass the auth with user/passwd
                 pre_auth = authenticate(
                     '', '', environ, VCS_TYPE, registry=self.registry,
                     acl_repo_name=self.acl_repo_name)
 
                 if pre_auth and pre_auth.get('username'):
                     username = pre_auth['username']
                 log.debug('PRE-AUTH got `%s` as username', username)
                 if pre_auth:
                     log.debug('PRE-AUTH successful from %s',
                               pre_auth.get('auth_data', {}).get('_plugin'))
 
                 # If not authenticated by the container, running basic auth
                 # before inject the calling repo_name for special scope checks
                 self.authenticate.acl_repo_name = self.acl_repo_name
 
                 plugin_cache_active, cache_ttl = False, 0
                 plugin = None
 
                 # regular auth chain
                 if not username:
                     self.authenticate.realm = self.authenticate.get_rc_realm()
 
                     try:
                         auth_result = self.authenticate(environ)
                     except (UserCreationError, NotAllowedToCreateUserError) as e:
                         log.error(e)
                         reason = safe_str(e)
                         return HTTPNotAcceptable(reason)(environ, start_response)
 
                     if isinstance(auth_result, dict):
                         AUTH_TYPE.update(environ, 'basic')
                         REMOTE_USER.update(environ, auth_result['username'])
                         username = auth_result['username']
                         plugin = auth_result.get('auth_data', {}).get('_plugin')
                         log.info(
                             'MAIN-AUTH successful for user `%s` from %s plugin',
                             username, plugin)
 
                         plugin_cache_active, cache_ttl = auth_result.get(
                             'auth_data', {}).get('_ttl_cache') or (False, 0)
                     else:
                         return auth_result.wsgi_application(environ, start_response)
 
                 # ==============================================================
                 # CHECK PERMISSIONS FOR THIS REQUEST USING GIVEN USERNAME
                 # ==============================================================
                 user = User.get_by_username(username)
                 if not self.valid_and_active_user(user):
                     return HTTPForbidden()(environ, start_response)
                 username = user.username
                 user_id = user.user_id
 
                 # check user attributes for password change flag
                 user_obj = user
                 auth_user = user_obj.AuthUser()
                 if user_obj and user_obj.username != User.DEFAULT_USER and \
                         user_obj.user_data.get('force_password_change'):
                     reason = 'password change required'
                     log.debug('User not allowed to authenticate, %s', reason)
                     return HTTPNotAcceptable(reason)(environ, start_response)
 
                 # check permissions for this repository
                 perm = self._check_permission(
                     action, user, auth_user, self.acl_repo_name, ip_addr,
                     plugin, plugin_cache_active, cache_ttl)
                 if not perm:
                     return HTTPForbidden()(environ, start_response)
                 environ['rc_auth_user_id'] = str(user_id)
 
             if action == 'push':
                 perms = auth_user.get_branch_permissions(self.acl_repo_name)
                 if perms:
                     check_branch_perms = True
                     detect_force_push = True
 
         # extras are injected into UI object and later available
         # in hooks executed by RhodeCode
         check_locking = _should_check_locking(environ.get('QUERY_STRING'))
 
         extras = vcs_operation_context(
             environ, repo_name=self.acl_repo_name, username=username,
             action=action, scm=self.SCM, check_locking=check_locking,
             is_shadow_repo=self.is_shadow_repo, check_branch_perms=check_branch_perms,
             detect_force_push=detect_force_push
         )
 
         # ======================================================================
         # REQUEST HANDLING
         # ======================================================================
         repo_path = os.path.join(
             safe_str(self.base_path), safe_str(self.vcs_repo_name))
         log.debug('Repository path is %s', repo_path)
 
         fix_PATH()
 
         log.info(
             '%s action on %s repo "%s" by "%s" from %s %s',
             action, self.SCM, safe_str(self.url_repo_name),
             safe_str(username), ip_addr, user_agent)
 
         return self._generate_vcs_response(
             environ, start_response, repo_path, extras, action)
 
     @initialize_generator
     def _generate_vcs_response(
             self, environ, start_response, repo_path, extras, action):
         """
         Returns a generator for the response content.
 
         This method is implemented as a generator, so that it can trigger
         the cache validation after all content sent back to the client. It
         also handles the locking exceptions which will be triggered when
         the first chunk is produced by the underlying WSGI application.
         """
 
         txn_id = ''
         if 'CONTENT_LENGTH' in environ and environ['REQUEST_METHOD'] == 'MERGE':
             # case for SVN, we want to re-use the callback daemon port
             # so we use the txn_id, for this we peek the body, and still save
             # it as wsgi.input
 
             stream = environ['wsgi.input']
 
             if isinstance(stream, io.BytesIO):
-                data: str = safe_str(stream.getvalue())
+                data: bytes = stream.getvalue()
             elif hasattr(stream, 'buf'):  # most likely gunicorn.http.body.Body
-                data: str = safe_str(stream.buf.getvalue())
+                data: bytes = stream.buf.getvalue()
             else:
                 # fallback to the crudest way, copy the iterator
-                data = safe_str(stream.read())
-                environ['wsgi.input'] = io.BytesIO(safe_bytes(data))
+                data = safe_bytes(stream.read())
+                environ['wsgi.input'] = io.BytesIO(data)
 
             txn_id = extract_svn_txn_id(self.acl_repo_name, data)
 
         callback_daemon, extras = self._prepare_callback_daemon(
             extras, environ, action, txn_id=txn_id)
         log.debug('HOOKS extras is %s', extras)
 
         http_scheme = self._get_http_scheme(environ)
 
         config = self._create_config(extras, self.acl_repo_name, scheme=http_scheme)
         app = self._create_wsgi_app(repo_path, self.url_repo_name, config)
         with callback_daemon:
             app.rc_extras = extras
 
             try:
                 response = app(environ, start_response)
             finally:
                 # This statement works together with the decorator
                 # "initialize_generator" above. The decorator ensures that
                 # we hit the first yield statement before the generator is
                 # returned back to the WSGI server. This is needed to
                 # ensure that the call to "app" above triggers the
                 # needed callback to "start_response" before the
                 # generator is actually used.
                 yield "__init__"
 
             # iter content
             for chunk in response:
                 yield chunk
 
             try:
                 # invalidate cache on push
                 if action == 'push':
                     self._invalidate_cache(self.url_repo_name)
             finally:
                 meta.Session.remove()
 
     def _get_repository_name(self, environ):
         """Get repository name out of the environmnent
 
         :param environ: WSGI environment
         """
         raise NotImplementedError()
 
     def _get_action(self, environ):
         """Map request commands into a pull or push command.
 
         :param environ: WSGI environment
         """
         raise NotImplementedError()
 
     def _create_wsgi_app(self, repo_path, repo_name, config):
         """Return the WSGI app that will finally handle the request."""
         raise NotImplementedError()
 
     def _create_config(self, extras, repo_name, scheme='http'):
         """Create a safe config representation."""
         raise NotImplementedError()
 
     def _should_use_callback_daemon(self, extras, environ, action):
         if extras.get('is_shadow_repo'):
             # we don't want to execute hooks, and callback daemon for shadow repos
             return False
         return True
 
     def _prepare_callback_daemon(self, extras, environ, action, txn_id=None):
         direct_calls = vcs_settings.HOOKS_DIRECT_CALLS
         if not self._should_use_callback_daemon(extras, environ, action):
             # disable callback daemon for actions that don't require it
             direct_calls = True
 
         return prepare_callback_daemon(
             extras, protocol=vcs_settings.HOOKS_PROTOCOL,
             host=vcs_settings.HOOKS_HOST, use_direct_calls=direct_calls, txn_id=txn_id)
 
 
 def _should_check_locking(query_string):
     # this is kind of hacky, but due to how mercurial handles client-server
     # server see all operation on commit; bookmarks, phases and
     # obsolescence marker in different transaction, we don't want to check
     # locking on those
     return query_string not in ['cmd=listkeys']

rhodecode/lib/middleware/vcs.py

@@ -1,298 +1,302 b''
 
 # Copyright (C) 2010-2023 RhodeCode GmbH
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU Affero General Public License, version 3
 # (only), as published by the Free Software Foundation.
 #
 # This program is distributed in the hope that it will be useful,
 # but WITHOUT ANY WARRANTY; without even the implied warranty of
 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 # GNU General Public License for more details.
 #
 # You should have received a copy of the GNU Affero General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 # This program is dual-licensed. If you wish to learn more about the
 # RhodeCode Enterprise Edition, including its added features, Support services,
 # and proprietary license terms, please see https://rhodecode.com/licenses/
 
 import gzip
 import shutil
 import logging
 import tempfile
 import urllib.parse
 
 from webob.exc import HTTPNotFound
 
 import rhodecode
 from rhodecode.lib.middleware.utils import get_path_info
 from rhodecode.lib.middleware.appenlight import wrap_in_appenlight_if_enabled
 from rhodecode.lib.middleware.simplegit import SimpleGit, GIT_PROTO_PAT
 from rhodecode.lib.middleware.simplehg import SimpleHg
 from rhodecode.lib.middleware.simplesvn import SimpleSvn
 from rhodecode.model.settings import VcsSettingsModel
 
 
 log = logging.getLogger(__name__)
 
 VCS_TYPE_KEY = '_rc_vcs_type'
 VCS_TYPE_SKIP = '_rc_vcs_skip'
 
 
 def is_git(environ):
     """
     Returns True if requests should be handled by GIT wsgi middleware
     """
     path_info = get_path_info(environ)
     is_git_path = GIT_PROTO_PAT.match(path_info)
     log.debug(
         'request path: `%s` detected as GIT PROTOCOL %s', path_info,
         is_git_path is not None)
 
     return is_git_path
 
 
 def is_hg(environ):
     """
     Returns True if requests target is mercurial server - header
     ``HTTP_ACCEPT`` of such request would start with ``application/mercurial``.
     """
     is_hg_path = False
 
     http_accept = environ.get('HTTP_ACCEPT')
 
     if http_accept and http_accept.startswith('application/mercurial'):
         query = urllib.parse.parse_qs(environ['QUERY_STRING'])
         if 'cmd' in query:
             is_hg_path = True
 
     path_info = get_path_info(environ)
     log.debug(
         'request path: `%s` detected as HG PROTOCOL %s', path_info,
         is_hg_path)
 
     return is_hg_path
 
 
 def is_svn(environ):
     """
     Returns True if requests target is Subversion server
     """
 
     http_dav = environ.get('HTTP_DAV', '')
     magic_path_segment = rhodecode.CONFIG.get(
         'rhodecode_subversion_magic_path', '/!svn')
     path_info = get_path_info(environ)
+    req_method = environ['REQUEST_METHOD']
+
     is_svn_path = (
         'subversion' in http_dav or
         magic_path_segment in path_info
-        or environ['REQUEST_METHOD'] in ['PROPFIND', 'PROPPATCH']
+        or req_method in ['PROPFIND', 'PROPPATCH', 'HEAD']
     )
     log.debug(
         'request path: `%s` detected as SVN PROTOCOL %s', path_info,
         is_svn_path)
 
     return is_svn_path
 
 
 class GunzipMiddleware(object):
     """
     WSGI middleware that unzips gzip-encoded requests before
     passing on to the underlying application.
     """
 
     def __init__(self, application):
         self.app = application
 
     def __call__(self, environ, start_response):
         accepts_encoding_header = environ.get('HTTP_CONTENT_ENCODING', b'')
 
         if b'gzip' in accepts_encoding_header:
             log.debug('gzip detected, now running gunzip wrapper')
             wsgi_input = environ['wsgi.input']
 
             if not hasattr(environ['wsgi.input'], 'seek'):
                 # The gzip implementation in the standard library of Python 2.x
                 # requires the '.seek()' and '.tell()' methods to be available
                 # on the input stream.  Read the data into a temporary file to
                 # work around this limitation.
 
                 wsgi_input = tempfile.SpooledTemporaryFile(64 * 1024 * 1024)
                 shutil.copyfileobj(environ['wsgi.input'], wsgi_input)
                 wsgi_input.seek(0)
 
             environ['wsgi.input'] = gzip.GzipFile(fileobj=wsgi_input, mode='r')
             # since we "Ungzipped" the content we say now it's no longer gzip
             # content encoding
             del environ['HTTP_CONTENT_ENCODING']
 
             # content length has changes ? or i'm not sure
             if 'CONTENT_LENGTH' in environ:
                 del environ['CONTENT_LENGTH']
         else:
             log.debug('content not gzipped, gzipMiddleware passing '
                       'request further')
         return self.app(environ, start_response)
 
 
 def is_vcs_call(environ):
     if VCS_TYPE_KEY in environ:
         raw_type = environ[VCS_TYPE_KEY]
         return raw_type and raw_type != VCS_TYPE_SKIP
     return False
 
 
 def detect_vcs_request(environ, backends):
     checks = {
         'hg': (is_hg, SimpleHg),
         'git': (is_git, SimpleGit),
         'svn': (is_svn, SimpleSvn),
     }
     handler = None
     # List of path views first chunk we don't do any checks
     white_list = [
         # favicon often requested by browsers
         'favicon.ico',
 
         # e.g /_file_store/download
         '_file_store++',
 
         # login
         "_admin/login",
 
         # _admin/api is safe too
         '_admin/api',
 
         # _admin/gist is safe too
         '_admin/gists++',
 
         # _admin/my_account is safe too
         '_admin/my_account++',
 
         # static files no detection
         '_static++',
 
         # debug-toolbar
         '_debug_toolbar++',
 
         # skip ops ping, status
         '_admin/ops/ping',
         '_admin/ops/status',
 
         # full channelstream connect should be VCS skipped
         '_admin/channelstream/connect',
 
         '++/repo_creating_check'
     ]
     path_info = get_path_info(environ)
     path_url = path_info.lstrip('/')
+    req_method = environ.get('REQUEST_METHOD')
 
     for item in white_list:
         if item.endswith('++') and path_url.startswith(item[:-2]):
             log.debug('path `%s` in whitelist (match:%s), skipping...', path_url, item)
             return handler
         if item.startswith('++') and path_url.endswith(item[2:]):
             log.debug('path `%s` in whitelist (match:%s), skipping...', path_url, item)
             return handler
         if item == path_url:
             log.debug('path `%s` in whitelist (match:%s), skipping...', path_url, item)
             return handler
 
     if VCS_TYPE_KEY in environ:
         raw_type = environ[VCS_TYPE_KEY]
         if raw_type == VCS_TYPE_SKIP:
             log.debug('got `skip` marker for vcs detection, skipping...')
             return handler
 
         _check, handler = checks.get(raw_type) or [None, None]
         if handler:
             log.debug('got handler:%s from environ', handler)
 
     if not handler:
-        log.debug('request start: checking if request for `%s` is of VCS type in order: %s', path_url, backends)
+        log.debug('request start: checking if request for `%s:%s` is of VCS type in order: %s',
+                  req_method, path_url, backends)
         for vcs_type in backends:
             vcs_check, _handler = checks[vcs_type]
             if vcs_check(environ):
                 log.debug('vcs handler found %s', _handler)
                 handler = _handler
                 break
 
     return handler
 
 
 class VCSMiddleware(object):
 
     def __init__(self, app, registry, config, appenlight_client):
         self.application = app
         self.registry = registry
         self.config = config
         self.appenlight_client = appenlight_client
         self.use_gzip = True
         # order in which we check the middlewares, based on vcs.backends config
         self.check_middlewares = config['vcs.backends']
 
     def vcs_config(self, repo_name=None):
         """
         returns serialized VcsSettings
         """
         try:
             return VcsSettingsModel(
                 repo=repo_name).get_ui_settings_as_config_obj()
         except Exception:
             pass
 
     def wrap_in_gzip_if_enabled(self, app, config):
         if self.use_gzip:
             app = GunzipMiddleware(app)
         return app
 
     def _get_handler_app(self, environ):
         app = None
         log.debug('VCSMiddleware: detecting vcs type.')
         handler = detect_vcs_request(environ, self.check_middlewares)
         if handler:
             app = handler(self.config, self.registry)
 
         return app
 
     def __call__(self, environ, start_response):
         # check if we handle one of interesting protocols, optionally extract
         # specific vcsSettings and allow changes of how things are wrapped
         vcs_handler = self._get_handler_app(environ)
         if vcs_handler:
             # translate the _REPO_ID into real repo NAME for usage
             # in middleware
 
             path_info = get_path_info(environ)
             environ['PATH_INFO'] = vcs_handler._get_by_id(path_info)
 
             # Set acl, url and vcs repo names.
             vcs_handler.set_repo_names(environ)
 
             # register repo config back to the handler
             vcs_conf = self.vcs_config(vcs_handler.acl_repo_name)
             # maybe damaged/non existent settings. We still want to
             # pass that point to validate on is_valid_and_existing_repo
             # and return proper HTTP Code back to client
             if vcs_conf:
                 vcs_handler.repo_vcs_config = vcs_conf
 
             # check for type, presence in database and on filesystem
             if not vcs_handler.is_valid_and_existing_repo(
                     vcs_handler.acl_repo_name,
                     vcs_handler.base_path,
                     vcs_handler.SCM):
                 return HTTPNotFound()(environ, start_response)
 
             environ['REPO_NAME'] = vcs_handler.url_repo_name
 
             # Wrap handler in middlewares if they are enabled.
             vcs_handler = self.wrap_in_gzip_if_enabled(
                 vcs_handler, self.config)
             vcs_handler, _ = wrap_in_appenlight_if_enabled(
                 vcs_handler, self.config, self.appenlight_client)
 
             return vcs_handler(environ, start_response)
 
         return self.application(environ, start_response)