path: root/src/leap/soledad/server/_blobs/__init__.py

# -*- coding: utf-8 -*-
# _blobs/__init__.py
# Copyright (C) 2017 LEAP
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# 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 General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.

"""
Blobs Server implementation.

This is a very simplistic implementation for the time being.
Clients should be able to opt-in util the feature is complete.

A more performant BlobsBackend can (and should) be implemented for production
environments.
"""
import os
import base64
import json
import re
import time

from twisted.web import resource
from twisted.web.client import FileBodyProducer
from twisted.web.server import NOT_DONE_YET
from twisted.internet import utils, defer

from collections import defaultdict
from zope.interface import implementer

from leap.common.files import mkdir_p
from leap.soledad.common.log import getLogger
from leap.soledad.server import interfaces
from leap.soledad.common.blobs import ACCEPTED_FLAGS
from leap.soledad.common.blobs import InvalidFlag


__all__ = ['BlobsResource']


logger = getLogger(__name__)

# Used for sanitizers, we accept only letters, numbers, '-' and '_'
VALID_STRINGS = re.compile('^[a-zA-Z0-9_-]+$')


# for the future:
# [ ] isolate user avatar in a safer way
# [ ] catch timeout in the server (and delete incomplete upload)
# [ ] chunking (should we do it on the client or on the server?)


class BlobNotFound(Exception):
    """
    Raised when a blob is not found in data storage backend.
    """


class BlobExists(Exception):
    """
    Raised when a blob already exists in data storage backend.
    """


class QuotaExceeded(Exception):
    """
    Raised when the quota would be exceeded if an operation would be held.
    """


@implementer(interfaces.IBlobsBackend)
class FilesystemBlobsBackend(object):

    USAGE_TIMEOUT = 30

    def __init__(self, blobs_path='/tmp/blobs/', quota=200 * 1024,
                 concurrent_writes=50):
        self.quota = quota
        self.semaphore = defer.DeferredSemaphore(concurrent_writes)
        if not os.path.isdir(blobs_path):
            os.makedirs(blobs_path)
        self.path = blobs_path
        self.usage = defaultdict(lambda: (None, None))
        self.usage_locks = defaultdict(defer.DeferredLock)

    def __touch(self, path):
        open(path, 'a')

    @defer.inlineCallbacks
    def read_blob(self, user, blob_id, consumer, namespace=''):
        logger.info('reading blob: %s - %s@%s' % (user, blob_id, namespace))
        path = self._get_path(user, blob_id, namespace)
        logger.debug('blob path: %s' % path)
        with open(path) as fd:
            producer = FileBodyProducer(fd)
            yield producer.startProducing(consumer)

    def get_flags(self, user, blob_id, namespace=''):
        try:
            path = self._get_path(user, blob_id, namespace)
        except Exception as e:
            return defer.fail(e)
        if not os.path.isfile(path):
            return defer.fail(BlobNotFound())
        if not os.path.isfile(path + '.flags'):
            return defer.succeed([])
        with open(path + '.flags', 'r') as flags_file:
            flags = json.loads(flags_file.read())
            return defer.succeed(flags)

    def set_flags(self, user, blob_id, flags, namespace=''):
        try:
            path = self._get_path(user, blob_id, namespace)
        except Exception as e:
            return defer.fail(e)
        if not os.path.isfile(path):
            return defer.fail(BlobNotFound())
        for flag in flags:
            if flag not in ACCEPTED_FLAGS:
                return defer.fail(InvalidFlag(flag))
        with open(path + '.flags', 'w') as flags_file:
            raw_flags = json.dumps(flags)
            flags_file.write(raw_flags)
        return defer.succeed(None)

    @defer.inlineCallbacks
    def write_blob(self, user, blob_id, producer, namespace=''):
        # limit the number of concurrent writes to disk
        yield self.semaphore.acquire()
        try:
            path = self._get_path(user, blob_id, namespace)
            try:
                mkdir_p(os.path.split(path)[0])
            except OSError as e:
                logger.warn("Got exception trying to create directory: %r" % e)
            if os.path.isfile(path):
                raise BlobExists
            used = yield self.get_total_storage(user)
            length = producer.length / 1024.0  # original length is in bytes
            if used + length > self.quota:
                raise QuotaExceeded
            logger.info('writing blob: %s - %s' % (user, blob_id))
            with open(path, 'wb') as blobfile:
                yield producer.startProducing(blobfile)
            used += length
            yield self._update_usage(user, used)
        finally:
            self.semaphore.release()

    @defer.inlineCallbacks
    def _update_usage(self, user, used):
        lock = self.usage_locks[user]
        yield lock.acquire()
        try:
            _, timestamp = self.usage[user]
            self.usage[user] = (used, timestamp)
        finally:
            lock.release()

    def delete_blob(self, user, blob_id, namespace=''):
        try:
            blob_path = self._get_path(user, blob_id, namespace)
        except Exception as e:
            return defer.fail(e)
        if not os.path.isfile(blob_path):
            return defer.fail(BlobNotFound())
        self.__touch(blob_path + '.deleted')
        os.unlink(blob_path)
        try:
            os.unlink(blob_path + '.flags')
        except Exception:
            pass
        return defer.succeed(None)

    def get_blob_size(self, user, blob_id, namespace=''):
        try:
            blob_path = self._get_path(user, blob_id, namespace)
        except Exception as e:
            return defer.fail(e)
        size = os.stat(blob_path).st_size
        return defer.succeed(size)

    def count(self, user, namespace=''):
        try:
            base_path = self._get_path(user, namespace=namespace)
        except Exception as e:
            return defer.fail(e)
        count = 0
        for _, _, filenames in os.walk(base_path):
            count += len(filter(lambda i: not i.endswith('.flags'), filenames))
        return defer.succeed(count)

    def list_blobs(self, user, namespace='', order_by=None, deleted=False,
                   filter_flag=False):
        namespace = namespace or 'default'
        blob_ids = []
        try:
            base_path = self._get_path(user, namespace=namespace)
        except Exception as e:
            return defer.fail(e)

        def match(name):
            if deleted:
                return name.endswith('.deleted')
            return VALID_STRINGS.match(name)
        for root, dirs, filenames in os.walk(base_path):
            blob_ids += [os.path.join(root, name) for name in filenames
                         if match(name)]
        if order_by in ['date', '+date']:
            blob_ids.sort(key=lambda x: os.path.getmtime(x))
        elif order_by == '-date':
            blob_ids.sort(key=lambda x: os.path.getmtime(x), reverse=True)
        elif order_by:
            exc = Exception("Unsupported order_by parameter: %s" % order_by)
            return defer.fail(exc)
        if filter_flag:
            blob_ids = list(self._filter_flag(blob_ids, filter_flag))
        blob_ids = [os.path.basename(path).replace('.deleted', '')
                    for path in blob_ids]
        return defer.succeed(blob_ids)

    def _filter_flag(self, blob_paths, flag):
        for blob_path in blob_paths:
            flag_path = blob_path + '.flags'
            if not os.path.isfile(flag_path):
                continue
            with open(flag_path, 'r') as flags_file:
                blob_flags = json.loads(flags_file.read())
            if flag in blob_flags:
                yield blob_path

    @defer.inlineCallbacks
    def get_total_storage(self, user):
        lock = self.usage_locks[user]
        yield lock.acquire()
        try:
            used, timestamp = self.usage[user]
            if used is None or time.time() > timestamp + self.USAGE_TIMEOUT:
                path = self._get_path(user)
                used = yield self._get_disk_usage(path)
                self.usage[user] = (used, time.time())
            defer.returnValue(used)
        finally:
            lock.release()

    def get_tag(self, user, blob_id, namespace=''):
        try:
            blob_path = self._get_path(user, blob_id, namespace)
        except Exception as e:
            return defer.fail(e)
        if not os.path.isfile(blob_path):
            return defer.fail(BlobNotFound())
        with open(blob_path) as doc_file:
            doc_file.seek(-16, 2)
            tag = base64.urlsafe_b64encode(doc_file.read())
            return defer.succeed(tag)

    @defer.inlineCallbacks
    def _get_disk_usage(self, start_path):
        if not os.path.isdir(start_path):
            defer.returnValue(0)
        cmd = ['/usr/bin/du', '-s', '-c', start_path]
        output = yield utils.getProcessOutput(cmd[0], cmd[1:])
        size = output.split()[0]
        defer.returnValue(int(size))

    def _validate_path(self, desired_path, user, blob_id):
        if not VALID_STRINGS.match(user):
            raise Exception("Invalid characters on user: %s" % user)
        if blob_id and not VALID_STRINGS.match(blob_id):
            raise Exception("Invalid characters on blob_id: %s" % blob_id)
        desired_path = os.path.realpath(desired_path)  # expand path references
        root = os.path.realpath(self.path)
        if not desired_path.startswith(root + os.sep + user):
            err = "User %s tried accessing a invalid path: %s" % (user,
                                                                  desired_path)
            raise Exception(err)
        return desired_path

    def exists(self, user, blob_id, namespace):
        try:
            path = self._get_path(user, blob_id=blob_id, namespace=namespace)
        except Exception as e:
            return defer.fail(e)
        return os.path.isfile(path)

    def _get_path(self, user, blob_id='', namespace=''):
        parts = [user]
        if blob_id:
            namespace = namespace or 'default'
            parts += self._get_path_parts(blob_id, namespace)
        elif namespace and not blob_id:
            parts += [namespace]  # namespace path
        else:
            pass  # root path
        path = os.path.join(self.path, *parts)
        return self._validate_path(path, user, blob_id)

    def _get_path_parts(self, blob_id, custom):
        if custom and not blob_id:
            return [custom]
        return [custom] + [blob_id[0], blob_id[0:3], blob_id[0:6]] + [blob_id]


class ImproperlyConfiguredException(Exception):
    pass


def _catchBlobNotFound(failure, request, user, blob_id):
    failure.trap(BlobNotFound)
    logger.error("Error 404: Blob %s does not exist for user %s"
                 % (blob_id, user))
    request.setResponseCode(404)
    request.write("Blob doesn't exists: %s" % blob_id)
    request.finish()


def _catchBlobExists(failure, request, user, blob_id):
    failure.trap(BlobExists)
    logger.error("Error 409: Blob %s already exists for user %s"
                 % (blob_id, user))
    request.setResponseCode(409)
    request.write("Blob already exists: %s" % blob_id)
    request.finish()


def _catchQuotaExceeded(failure, request, user):
    failure.trap(QuotaExceeded)
    logger.error("Error 507: Quota exceeded for user: %s" % user)
    request.setResponseCode(507)
    request.write('Quota Exceeded!')
    request.finish()


def _catchInvalidFlag(failure, request, user, blob_id):
    failure.trap(InvalidFlag)
    flag = failure.value.message
    logger.error("Error 406: Attempted to set invalid flag %s for blob %s "
                 "for user %s" % (flag, blob_id, user))
    request.setResponseCode(406)
    request.write("Invalid flag: %s" % str(flag))
    request.finish()


def _catchAllErrors(self, e, request):
    logger.error('Error processing request: %s' % e.getErrorMessage())
    request.setResponseCode(500)
    request.finish()


class BlobsResource(resource.Resource):

    isLeaf = True

    # Allowed backend classes are defined here
    handlers = {"filesystem": FilesystemBlobsBackend}

    def __init__(self, backend, blobs_path, **backend_kwargs):
        resource.Resource.__init__(self)
        self._blobs_path = blobs_path
        backend_kwargs.update({'blobs_path': blobs_path})
        if backend not in self.handlers:
            raise ImproperlyConfiguredException("No such backend: %s", backend)
        self._handler = self.handlers[backend](**backend_kwargs)
        assert interfaces.IBlobsBackend.providedBy(self._handler)

    # TODO double check credentials, we can have then
    # under request.

    def _only_count(self, request, user, namespace):
        d = self._handler.count(user, namespace)
        d.addCallback(lambda count: json.dumps({"count": count}))
        d.addCallback(lambda count: request.write(count))
        d.addCallback(lambda _: request.finish())
        return NOT_DONE_YET

    def _list(self, request, user, namespace):
        order = request.args.get('order_by', [None])[0]
        filter_flag = request.args.get('filter_flag', [False])[0]
        deleted = request.args.get('deleted', [False])[0]
        d = self._handler.list_blobs(user, namespace,
                                     order_by=order, deleted=deleted,
                                     filter_flag=filter_flag)
        d.addCallback(lambda blobs: json.dumps(blobs))
        d.addCallback(lambda blobs: request.write(blobs))
        d.addCallback(lambda _: request.finish())
        return NOT_DONE_YET

    def _only_flags(self, request, user, blob_id, namespace):
        d = self._handler.get_flags(user, blob_id, namespace)
        d.addCallback(lambda flags: json.dumps(flags))
        d.addCallback(lambda flags: request.write(flags))
        d.addCallback(lambda _: request.finish())
        d.addErrback(_catchBlobNotFound, request, user, blob_id)
        d.addErrback(_catchAllErrors, request)
        return NOT_DONE_YET

    def _get_blob(self, request, user, blob_id, namespace):

        def _set_tag_header(tag):
            request.responseHeaders.setRawHeaders('Tag', [tag])

        def _read_blob(_):
            handler = self._handler
            consumer = request
            d = handler.read_blob(user, blob_id, consumer, namespace=namespace)
            return d

        d = self._handler.get_tag(user, blob_id, namespace)
        d.addCallback(_set_tag_header)
        d.addCallback(_read_blob)
        d.addCallback(lambda _: request.finish())
        d.addErrback(_catchBlobNotFound, request, user, blob_id)
        d.addErrback(_catchAllErrors, request, finishRequest=True)

        return NOT_DONE_YET

    def render_GET(self, request):
        logger.info("http get: %s" % request.path)
        user, blob_id, namespace = self._validate(request)
        only_flags = request.args.get('only_flags', [False])[0]

        if not blob_id and request.args.get('only_count', [False])[0]:
            return self._only_count(request, user, namespace)

        if not blob_id:
            return self._list(request, user, namespace)

        if only_flags:
            return self._only_flags(request, user, blob_id, namespace)

        return self._get_blob(request, user, blob_id, namespace)

    def render_DELETE(self, request):
        logger.info("http put: %s" % request.path)
        user, blob_id, namespace = self._validate(request)
        d = self._handler.delete_blob(user, blob_id, namespace=namespace)
        d.addCallback(lambda _: request.finish())
        d.addErrback(_catchBlobNotFound, request, user, blob_id)
        d.addErrback(_catchAllErrors, request)
        return NOT_DONE_YET

    def render_PUT(self, request):
        logger.info("http put: %s" % request.path)
        user, blob_id, namespace = self._validate(request)
        producer = FileBodyProducer(request.content)
        handler = self._handler
        d = handler.write_blob(user, blob_id, producer, namespace=namespace)
        d.addCallback(lambda _: request.finish())
        d.addErrback(_catchBlobExists, request, user, blob_id)
        d.addErrback(_catchQuotaExceeded, request, user)
        d.addErrback(_catchAllErrors, request)
        return NOT_DONE_YET

    def render_POST(self, request):
        logger.info("http post: %s" % request.path)
        user, blob_id, namespace = self._validate(request)
        raw_flags = request.content.read()
        flags = json.loads(raw_flags)
        d = self._handler.set_flags(user, blob_id, flags, namespace=namespace)
        d.addCallback(lambda _: request.write(''))
        d.addCallback(lambda _: request.finish())
        d.addErrback(_catchBlobNotFound, request, user, blob_id)
        d.addErrback(_catchInvalidFlag, request, user, blob_id)
        d.addErrback(_catchAllErrors, request)
        return NOT_DONE_YET

    def _validate(self, request):
        for arg in request.postpath:
            if arg and not VALID_STRINGS.match(arg):
                raise Exception('Invalid blob resource argument: %s' % arg)
        namespace = request.args.get('namespace', ['default'])[0]
        if namespace and not VALID_STRINGS.match(namespace):
            raise Exception('Invalid blob namespace: %s' % namespace)
        return request.postpath + [namespace]


if __name__ == '__main__':
    # A dummy blob server
    # curl -X PUT --data-binary @/tmp/book.pdf localhost:9000/user/someid
    # curl -X GET -o /dev/null localhost:9000/user/somerandomstring
    from twisted.python import log
    import sys
    log.startLogging(sys.stdout)

    from twisted.web.server import Site
    from twisted.internet import reactor

    # parse command line arguments
    import argparse

    parser = argparse.ArgumentParser()
    parser.add_argument('--port', default=9000, type=int)
    parser.add_argument('--path', default='/tmp/blobs/user')
    args = parser.parse_args()

    root = BlobsResource("filesystem", args.path)
    # I picture somethink like
    # BlobsResource(backend="filesystem", backend_opts={'path': '/tmp/blobs'})

    factory = Site(root)
    reactor.listenTCP(args.port, factory)
    reactor.run()


class BlobsServerState(object):
    """
    Given a backend name, it gives a instance of IBlobsBackend
    """
    # Allowed backend classes are defined here
    handlers = {"filesystem": FilesystemBlobsBackend}

    def __init__(self, backend, **backend_kwargs):
        if backend not in self.handlers:
            raise ImproperlyConfiguredException("No such backend: %s", backend)
        self.backend = self.handlers[backend](**backend_kwargs)

    def open_database(self, user_id):
        """
        That method is just for compatibility with CouchServerState, so
        IncomingAPI can change backends.
        """
        # TODO: deprecate/refactor it as it's here for compatibility.
        return self.backend