Changeset - 64c194492aad
[Not reviewed]
beta
0 3 0
Marcin Kuzminski - 13 years ago 2013-02-25 17:16:45
marcin@python-works.com
--version command should be safe, and bare no modifications
- improved subprocess calls error detection
- fixed I/O read on closed file errors
3 files changed with 25 insertions and 4 deletions:
0 comments (0 inline, 0 general)
rhodecode/lib/subprocessio.py
Show inline comments
 
@@ -74,97 +74,101 @@ class StreamFeeder(Thread):
 
    @property
 
    def output(self):
 
        return self.readiface
 

	
 

	
 
class InputStreamChunker(Thread):
 
    def __init__(self, source, target, buffer_size, chunk_size):
 

	
 
        super(InputStreamChunker, self).__init__()
 

	
 
        self.daemon = True  # die die die.
 

	
 
        self.source = source
 
        self.target = target
 
        self.chunk_count_max = int(buffer_size / chunk_size) + 1
 
        self.chunk_size = chunk_size
 

	
 
        self.data_added = Event()
 
        self.data_added.clear()
 

	
 
        self.keep_reading = Event()
 
        self.keep_reading.set()
 

	
 
        self.EOF = Event()
 
        self.EOF.clear()
 

	
 
        self.go = Event()
 
        self.go.set()
 

	
 
    def stop(self):
 
        self.go.clear()
 
        self.EOF.set()
 
        try:
 
            # this is not proper, but is done to force the reader thread let
 
            # go of the input because, if successful, .close() will send EOF
 
            # down the pipe.
 
            self.source.close()
 
        except:
 
            pass
 

	
 
    def run(self):
 
        s = self.source
 
        t = self.target
 
        cs = self.chunk_size
 
        ccm = self.chunk_count_max
 
        kr = self.keep_reading
 
        da = self.data_added
 
        go = self.go
 

	
 
        try:
 
        b = s.read(cs)
 
        except ValueError:
 
            b = ''
 

	
 
        while b and go.is_set():
 
            if len(t) > ccm:
 
                kr.clear()
 
                kr.wait(2)
 
#                # this only works on 2.7.x and up
 
#                if not kr.wait(10):
 
#                    raise Exception("Timed out while waiting for input to be read.")
 
                # instead we'll use this
 
                if len(t) > ccm + 3:
 
                    raise IOError("Timed out while waiting for input from subprocess.")
 
            t.append(b)
 
            da.set()
 
            b = s.read(cs)
 
        self.EOF.set()
 
        da.set()  # for cases when done but there was no input.
 

	
 

	
 
class BufferedGenerator():
 
    '''
 
    Class behaves as a non-blocking, buffered pipe reader.
 
    Reads chunks of data (through a thread)
 
    from a blocking pipe, and attaches these to an array (Deque) of chunks.
 
    Reading is halted in the thread when max chunks is internally buffered.
 
    The .next() may operate in blocking or non-blocking fashion by yielding
 
    '' if no data is ready
 
    to be sent or by not returning until there is some data to send
 
    When we get EOF from underlying source pipe we raise the marker to raise
 
    StopIteration after the last chunk of data is yielded.
 
    '''
 

	
 
    def __init__(self, source, buffer_size=65536, chunk_size=4096,
 
                 starting_values=[], bottomless=False):
 

	
 
        if bottomless:
 
            maxlen = int(buffer_size / chunk_size)
 
        else:
 
            maxlen = None
 

	
 
        self.data = deque(starting_values, maxlen)
 

	
 
        self.worker = InputStreamChunker(source, self.data, buffer_size,
 
                                         chunk_size)
 
        if starting_values:
 
            self.worker.data_added.set()
 
        self.worker.start()
 

	
 
    ####################
 
@@ -327,83 +331,85 @@ class SubprocessIOChunker(object):
 
        Initializes SubprocessIOChunker
 

	
 
        :param cmd: A Subprocess.Popen style "cmd". Can be string or array of strings
 
        :param inputstream: (Default: None) A file-like, string, or file pointer.
 
        :param buffer_size: (Default: 65536) A size of total buffer per stream in bytes.
 
        :param chunk_size: (Default: 4096) A max size of a chunk. Actual chunk may be smaller.
 
        :param starting_values: (Default: []) An array of strings to put in front of output que.
 
        '''
 

	
 
        if inputstream:
 
            input_streamer = StreamFeeder(inputstream)
 
            input_streamer.start()
 
            inputstream = input_streamer.output
 

	
 
        if isinstance(cmd, (list, tuple)):
 
            cmd = ' '.join(cmd)
 

	
 
        _shell = kwargs.get('shell') or True
 
        kwargs['shell'] = _shell
 
        _p = subprocess.Popen(cmd,
 
            bufsize=-1,
 
            stdin=inputstream,
 
            stdout=subprocess.PIPE,
 
            stderr=subprocess.PIPE,
 
            **kwargs
 
            )
 

	
 
        bg_out = BufferedGenerator(_p.stdout, buffer_size, chunk_size, starting_values)
 
        bg_err = BufferedGenerator(_p.stderr, 16000, 1, bottomless=True)
 

	
 
        while not bg_out.done_reading and not bg_out.reading_paused and not bg_err.length:
 
            # doing this until we reach either end of file, or end of buffer.
 
            bg_out.data_added_event.wait(1)
 
            bg_out.data_added_event.clear()
 

	
 
        # at this point it's still ambiguous if we are done reading or just full buffer.
 
        # Either way, if error (returned by ended process, or implied based on
 
        # presence of stuff in stderr output) we error out.
 
        # Else, we are happy.
 
        _returncode = _p.poll()
 
        if _returncode or (_returncode == None and bg_err.length):
 
            try:
 
                _p.terminate()
 
            except:
 
                pass
 
            bg_out.stop()
 
            bg_err.stop()
 
            err = '%s' % ''.join(bg_err)
 
            if err:
 
            raise EnvironmentError("Subprocess exited due to an error:\n" + err)
 
            raise EnvironmentError("Subprocess exited with non 0 ret code:%s" % _returncode)
 

	
 
        self.process = _p
 
        self.output = bg_out
 
        self.error = bg_err
 

	
 
    def __iter__(self):
 
        return self
 

	
 
    def next(self):
 
        if self.process.poll():
 
            err = '%s' % ''.join(self.error)
 
            raise EnvironmentError("Subprocess exited due to an error:\n" + err)
 
        return self.output.next()
 

	
 
    def throw(self, type, value=None, traceback=None):
 
        if self.output.length or not self.output.done_reading:
 
            raise type(value)
 

	
 
    def close(self):
 
        try:
 
            self.process.terminate()
 
        except:
 
            pass
 
        try:
 
            self.output.close()
 
        except:
 
            pass
 
        try:
 
            self.error.close()
 
        except:
 
            pass
 

	
 
    def __del__(self):
 
        self.close()
rhodecode/lib/utils.py
Show inline comments
 
@@ -703,97 +703,98 @@ class BasePasterCommand(Command):
 
        self.min_args -= 1
 
        self.bootstrap_config(args[0])
 
        self.update_parser()
 
        return super(BasePasterCommand, self).run(args[1:])
 

	
 
    def update_parser(self):
 
        """
 
        Abstract method.  Allows for the class's parser to be updated
 
        before the superclass's `run` method is called.  Necessary to
 
        allow options/arguments to be passed through to the underlying
 
        celery command.
 
        """
 
        raise NotImplementedError("Abstract Method.")
 

	
 
    def bootstrap_config(self, conf):
 
        """
 
        Loads the pylons configuration.
 
        """
 
        from pylons import config as pylonsconfig
 

	
 
        self.path_to_ini_file = os.path.realpath(conf)
 
        conf = paste.deploy.appconfig('config:' + self.path_to_ini_file)
 
        pylonsconfig.init_app(conf.global_conf, conf.local_conf)
 

	
 
    def _init_session(self):
 
        """
 
        Inits SqlAlchemy Session
 
        """
 
        logging.config.fileConfig(self.path_to_ini_file)
 
        from pylons import config
 
        from rhodecode.model import init_model
 
        from rhodecode.lib.utils2 import engine_from_config
 

	
 
        #get to remove repos !!
 
        add_cache(config)
 
        engine = engine_from_config(config, 'sqlalchemy.db1.')
 
        init_model(engine)
 

	
 

	
 
def check_git_version():
 
    """
 
    Checks what version of git is installed in system, and issues a warning
 
    if it's too old for RhodeCode to properly work.
 
    """
 
    from rhodecode import BACKENDS
 
    from rhodecode.lib.vcs.backends.git.repository import GitRepository
 
    from distutils.version import StrictVersion
 

	
 
    stdout, stderr = GitRepository._run_git_command('--version')
 
    stdout, stderr = GitRepository._run_git_command('--version', _bare=True,
 
                                                    _safe=True)
 

	
 
    ver = (stdout.split(' ')[-1] or '').strip() or '0.0.0'
 
    if len(ver.split('.')) > 3:
 
        #StrictVersion needs to be only 3 element type
 
        ver = '.'.join(ver.split('.')[:3])
 
    try:
 
        _ver = StrictVersion(ver)
 
    except:
 
        _ver = StrictVersion('0.0.0')
 
        stderr = traceback.format_exc()
 

	
 
    req_ver = '1.7.4'
 
    to_old_git = False
 
    if  _ver < StrictVersion(req_ver):
 
        to_old_git = True
 

	
 
    if 'git' in BACKENDS:
 
        log.debug('GIT version detected: %s' % stdout)
 
        if stderr:
 
            log.warning('Unable to detect git version org error was:%r' % stderr)
 
        elif to_old_git:
 
            log.warning('RhodeCode detected git version %s, which is too old '
 
                        'for the system to function properly. Make sure '
 
                        'its version is at least %s' % (ver, req_ver))
 
    return _ver
 

	
 

	
 
@decorator.decorator
 
def jsonify(func, *args, **kwargs):
 
    """Action decorator that formats output for JSON
 

	
 
    Given a function that will return content, this decorator will turn
 
    the result into JSON, with a content-type of 'application/json' and
 
    output it.
 

	
 
    """
 
    from pylons.decorators.util import get_pylons
 
    from rhodecode.lib.ext_json import json
 
    pylons = get_pylons(args)
 
    pylons.response.headers['Content-Type'] = 'application/json; charset=utf-8'
 
    data = func(*args, **kwargs)
 
    if isinstance(data, (list, tuple)):
 
        msg = "JSON responses with Array envelopes are susceptible to " \
 
              "cross-site data leak attacks, see " \
 
              "http://wiki.pylonshq.com/display/pylonsfaq/Warnings"
 
        warnings.warn(msg, Warning, 2)
 
        log.warning(msg)
 
    log.debug("Returning JSON wrapped action output")
rhodecode/lib/vcs/backends/git/repository.py
Show inline comments
 
@@ -57,123 +57,137 @@ class GitRepository(BaseRepository):
 
        self.path = abspath(repo_path)
 
        repo = self._get_repo(create, src_url, update_after_clone, bare)
 
        self.bare = repo.bare
 

	
 
        self._config_files = [
 
            bare and abspath(self.path, 'config')
 
                     or abspath(self.path, '.git', 'config'),
 
            abspath(get_user_home(), '.gitconfig'),
 
        ]
 

	
 
    @ThreadLocalLazyProperty
 
    def _repo(self):
 
        repo = Repo(self.path)
 
        #temporary set that to now at later we will move it to constructor
 
        baseui = None
 
        if baseui is None:
 
            from mercurial.ui import ui
 
            baseui = ui()
 
        # patch the instance of GitRepo with an "FAKE" ui object to add
 
        # compatibility layer with Mercurial
 
        setattr(repo, 'ui', baseui)
 
        return repo
 

	
 
    @property
 
    def head(self):
 
        try:
 
            return self._repo.head()
 
        except KeyError:
 
            return None
 

	
 
    @LazyProperty
 
    def revisions(self):
 
        """
 
        Returns list of revisions' ids, in ascending order.  Being lazy
 
        attribute allows external tools to inject shas from cache.
 
        """
 
        return self._get_all_revisions()
 

	
 
    @classmethod
 
    def _run_git_command(cls, cmd, **opts):
 
        """
 
        Runs given ``cmd`` as git command and returns tuple
 
        (stdout, stderr).
 

	
 
        :param cmd: git command to be executed
 
        :param opts: env options to pass into Subprocess command
 
        """
 

	
 
        if '_bare' in opts:
 
            _copts = []
 
            del opts['_bare']
 
        else:
 
        _copts = ['-c', 'core.quotepath=false', ]
 
        safe_call = False
 
        if '_safe' in opts:
 
            #no exc on failure
 
            del opts['_safe']
 
            safe_call = True
 

	
 
        _str_cmd = False
 
        if isinstance(cmd, basestring):
 
            cmd = [cmd]
 
            _str_cmd = True
 

	
 
        gitenv = os.environ
 
        # need to clean fix GIT_DIR !
 
        if 'GIT_DIR' in gitenv:
 
            del gitenv['GIT_DIR']
 
        gitenv['GIT_CONFIG_NOGLOBAL'] = '1'
 

	
 
        _git_path = rhodecode.CONFIG.get('git_path', 'git')
 
        cmd = [_git_path] + _copts + cmd
 
        if _str_cmd:
 
            cmd = ' '.join(cmd)
 
        try:
 
            _opts = dict(
 
                env=gitenv,
 
                shell=False,
 
            )
 
            _opts.update(opts)
 
            p = subprocessio.SubprocessIOChunker(cmd, **_opts)
 
        except (EnvironmentError, OSError), err:
 
            log.error(traceback.format_exc())
 
            raise RepositoryError("Couldn't run git command (%s).\n"
 
                                  "Original error was:%s" % (cmd, err))
 
            tb_err = ("Couldn't run git command (%s).\n"
 
                      "Original error was:%s\n" % (cmd, err))
 
            log.error(tb_err)
 
            if safe_call:
 
                return '', err
 
            else:
 
                raise RepositoryError(tb_err)
 

	
 
        return ''.join(p.output), ''.join(p.error)
 

	
 
    def run_git_command(self, cmd):
 
        opts = {}
 
        if os.path.isdir(self.path):
 
            opts['cwd'] = self.path
 
        return self._run_git_command(cmd, **opts)
 

	
 
    @classmethod
 
    def _check_url(cls, url):
 
        """
 
        Functon will check given url and try to verify if it's a valid
 
        link. Sometimes it may happened that mercurial will issue basic
 
        auth request that can cause whole API to hang when used from python
 
        or other external calls.
 

	
 
        On failures it'll raise urllib2.HTTPError
 
        """
 
        from mercurial.util import url as Url
 

	
 
        # those authnadlers are patched for python 2.6.5 bug an
 
        # infinit looping when given invalid resources
 
        from mercurial.url import httpbasicauthhandler, httpdigestauthhandler
 

	
 
        # check first if it's not an local url
 
        if os.path.isdir(url) or url.startswith('file:'):
 
            return True
 

	
 
        if('+' in url[:url.find('://')]):
 
            url = url[url.find('+') + 1:]
 

	
 
        handlers = []
 
        test_uri, authinfo = Url(url).authinfo()
 
        if not test_uri.endswith('info/refs'):
 
            test_uri = test_uri.rstrip('/') + '/info/refs'
 
        if authinfo:
 
            #create a password manager
 
            passmgr = urllib2.HTTPPasswordMgrWithDefaultRealm()
 
            passmgr.add_password(*authinfo)
 

	
 
            handlers.extend((httpbasicauthhandler(passmgr),
 
                             httpdigestauthhandler(passmgr)))
 

	
 
        o = urllib2.build_opener(*handlers)
 
        o.addheaders = [('User-Agent', 'git/1.7.8.0')]  # fake some git
 

	
 
        q = {"service": 'git-upload-pack'}
0 comments (0 inline, 0 general)