summary refs log tree commit diff
path: root/synapse/rest/media/v1
diff options
context:
space:
mode:
authorMatthew Hodgson <matthew@arasphere.net>2016-04-11 10:40:29 +0100
committerMatthew Hodgson <matthew@arasphere.net>2016-04-11 10:40:29 +0100
commit4bd3d252188e5a4630d35d8b4bf1f3c287e92944 (patch)
tree9d78df7b09d93b0d49805ff1516f0ec5250c55a1 /synapse/rest/media/v1
parentMerge pull request #677 from matrix-org/erikj/dns_cache (diff)
parentfix typos and needless try/except from PR review (diff)
downloadsynapse-4bd3d252188e5a4630d35d8b4bf1f3c287e92944.tar.xz
Merge pull request #688 from matrix-org/matthew/preview_urls
URL previewing support
Diffstat (limited to 'synapse/rest/media/v1')
-rw-r--r--synapse/rest/media/v1/base_resource.py1
-rw-r--r--synapse/rest/media/v1/media_repository.py7
-rw-r--r--synapse/rest/media/v1/preview_url_resource.py462
-rw-r--r--synapse/rest/media/v1/thumbnail_resource.py22
4 files changed, 492 insertions, 0 deletions
diff --git a/synapse/rest/media/v1/base_resource.py b/synapse/rest/media/v1/base_resource.py
index 58ef91c0b8..2b1938dc8e 100644
--- a/synapse/rest/media/v1/base_resource.py
+++ b/synapse/rest/media/v1/base_resource.py
@@ -72,6 +72,7 @@ class BaseMediaResource(Resource):
         self.store = hs.get_datastore()
         self.max_upload_size = hs.config.max_upload_size
         self.max_image_pixels = hs.config.max_image_pixels
+        self.max_spider_size = hs.config.max_spider_size
         self.filepaths = filepaths
         self.version_string = hs.version_string
         self.downloads = {}
diff --git a/synapse/rest/media/v1/media_repository.py b/synapse/rest/media/v1/media_repository.py
index 7dfb027dd1..97b7e84af9 100644
--- a/synapse/rest/media/v1/media_repository.py
+++ b/synapse/rest/media/v1/media_repository.py
@@ -17,6 +17,7 @@ from .upload_resource import UploadResource
 from .download_resource import DownloadResource
 from .thumbnail_resource import ThumbnailResource
 from .identicon_resource import IdenticonResource
+from .preview_url_resource import PreviewUrlResource
 from .filepath import MediaFilePaths
 
 from twisted.web.resource import Resource
@@ -78,3 +79,9 @@ class MediaRepositoryResource(Resource):
         self.putChild("download", DownloadResource(hs, filepaths))
         self.putChild("thumbnail", ThumbnailResource(hs, filepaths))
         self.putChild("identicon", IdenticonResource())
+        if hs.config.url_preview_enabled:
+            try:
+                self.putChild("preview_url", PreviewUrlResource(hs, filepaths))
+            except Exception as e:
+                logger.warn("Failed to mount preview_url")
+                logger.exception(e)
diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py
new file mode 100644
index 0000000000..8464fc0176
--- /dev/null
+++ b/synapse/rest/media/v1/preview_url_resource.py
@@ -0,0 +1,462 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# 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.
+
+from .base_resource import BaseMediaResource
+
+from twisted.web.server import NOT_DONE_YET
+from twisted.internet import defer
+from urlparse import urlparse, urlsplit, urlunparse
+
+from synapse.api.errors import (
+    SynapseError, Codes,
+)
+from synapse.util.stringutils import random_string
+from synapse.util.caches.expiringcache import ExpiringCache
+from synapse.http.client import SpiderHttpClient
+from synapse.http.server import (
+    request_handler, respond_with_json_bytes
+)
+from synapse.util.async import ObservableDeferred
+from synapse.util.stringutils import is_ascii
+
+import os
+import re
+import fnmatch
+import cgi
+import ujson as json
+
+import logging
+logger = logging.getLogger(__name__)
+
+try:
+    from lxml import html
+except ImportError:
+    pass
+
+
+class PreviewUrlResource(BaseMediaResource):
+    isLeaf = True
+
+    def __init__(self, hs, filepaths):
+        try:
+            if html:
+                pass
+        except:
+            raise RuntimeError("Disabling PreviewUrlResource as lxml not available")
+
+        if not hasattr(hs.config, "url_preview_ip_range_blacklist"):
+            logger.warn(
+                "For security, you must specify an explicit target IP address "
+                "blacklist in url_preview_ip_range_blacklist for url previewing "
+                "to work"
+            )
+            raise RuntimeError(
+                "Disabling PreviewUrlResource as "
+                "url_preview_ip_range_blacklist not specified"
+            )
+
+        BaseMediaResource.__init__(self, hs, filepaths)
+        self.client = SpiderHttpClient(hs)
+        if hasattr(hs.config, "url_preview_url_blacklist"):
+            self.url_preview_url_blacklist = hs.config.url_preview_url_blacklist
+
+        # simple memory cache mapping urls to OG metadata
+        self.cache = ExpiringCache(
+            cache_name="url_previews",
+            clock=self.clock,
+            # don't spider URLs more often than once an hour
+            expiry_ms=60 * 60 * 1000,
+        )
+        self.cache.start()
+
+        self.downloads = {}
+
+    def render_GET(self, request):
+        self._async_render_GET(request)
+        return NOT_DONE_YET
+
+    @request_handler
+    @defer.inlineCallbacks
+    def _async_render_GET(self, request):
+
+        # XXX: if get_user_by_req fails, what should we do in an async render?
+        requester = yield self.auth.get_user_by_req(request)
+        url = request.args.get("url")[0]
+        if "ts" in request.args:
+            ts = int(request.args.get("ts")[0])
+        else:
+            ts = self.clock.time_msec()
+
+        # impose the URL pattern blacklist
+        if hasattr(self, "url_preview_url_blacklist"):
+            url_tuple = urlsplit(url)
+            for entry in self.url_preview_url_blacklist:
+                match = True
+                for attrib in entry:
+                    pattern = entry[attrib]
+                    value = getattr(url_tuple, attrib)
+                    logger.debug((
+                        "Matching attrib '%s' with value '%s' against"
+                        " pattern '%s'"
+                    ) % (attrib, value, pattern))
+
+                    if value is None:
+                        match = False
+                        continue
+
+                    if pattern.startswith('^'):
+                        if not re.match(pattern, getattr(url_tuple, attrib)):
+                            match = False
+                            continue
+                    else:
+                        if not fnmatch.fnmatch(getattr(url_tuple, attrib), pattern):
+                            match = False
+                            continue
+                if match:
+                    logger.warn(
+                        "URL %s blocked by url_blacklist entry %s", url, entry
+                    )
+                    raise SynapseError(
+                        403, "URL blocked by url pattern blacklist entry",
+                        Codes.UNKNOWN
+                    )
+
+        # first check the memory cache - good to handle all the clients on this
+        # HS thundering away to preview the same URL at the same time.
+        og = self.cache.get(url)
+        if og:
+            respond_with_json_bytes(request, 200, json.dumps(og), send_cors=True)
+            return
+
+        # then check the URL cache in the DB (which will also provide us with
+        # historical previews, if we have any)
+        cache_result = yield self.store.get_url_cache(url, ts)
+        if (
+            cache_result and
+            cache_result["download_ts"] + cache_result["expires"] > ts and
+            cache_result["response_code"] / 100 == 2
+        ):
+            respond_with_json_bytes(
+                request, 200, cache_result["og"].encode('utf-8'),
+                send_cors=True
+            )
+            return
+
+        # Ensure only one download for a given URL is active at a time
+        download = self.downloads.get(url)
+        if download is None:
+            download = self._download_url(url, requester.user)
+            download = ObservableDeferred(
+                download,
+                consumeErrors=True
+            )
+            self.downloads[url] = download
+
+            @download.addBoth
+            def callback(media_info):
+                del self.downloads[url]
+                return media_info
+        media_info = yield download.observe()
+
+        # FIXME: we should probably update our cache now anyway, so that
+        # even if the OG calculation raises, we don't keep hammering on the
+        # remote server.  For now, leave it uncached to aid debugging OG
+        # calculation problems
+
+        logger.debug("got media_info of '%s'" % media_info)
+
+        if self._is_media(media_info['media_type']):
+            dims = yield self._generate_local_thumbnails(
+                media_info['filesystem_id'], media_info
+            )
+
+            og = {
+                "og:description": media_info['download_name'],
+                "og:image": "mxc://%s/%s" % (
+                    self.server_name, media_info['filesystem_id']
+                ),
+                "og:image:type": media_info['media_type'],
+                "matrix:image:size": media_info['media_length'],
+            }
+
+            if dims:
+                og["og:image:width"] = dims['width']
+                og["og:image:height"] = dims['height']
+            else:
+                logger.warn("Couldn't get dims for %s" % url)
+
+            # define our OG response for this media
+        elif self._is_html(media_info['media_type']):
+            # TODO: somehow stop a big HTML tree from exploding synapse's RAM
+
+            try:
+                tree = html.parse(media_info['filename'])
+                og = yield self._calc_og(tree, media_info, requester)
+            except UnicodeDecodeError:
+                # XXX: evil evil bodge
+                # Empirically, sites like google.com mix Latin-1 and utf-8
+                # encodings in the same page.  The rogue Latin-1 characters
+                # cause lxml to choke with a UnicodeDecodeError, so if we
+                # see this we go and do a manual decode of the HTML before
+                # handing it to lxml as utf-8 encoding, counter-intuitively,
+                # which seems to make it happier...
+                file = open(media_info['filename'])
+                body = file.read()
+                file.close()
+                tree = html.fromstring(body.decode('utf-8', 'ignore'))
+                og = yield self._calc_og(tree, media_info, requester)
+
+        else:
+            logger.warn("Failed to find any OG data in %s", url)
+            og = {}
+
+        logger.debug("Calculated OG for %s as %s" % (url, og))
+
+        # store OG in ephemeral in-memory cache
+        self.cache[url] = og
+
+        # store OG in history-aware DB cache
+        yield self.store.store_url_cache(
+            url,
+            media_info["response_code"],
+            media_info["etag"],
+            media_info["expires"],
+            json.dumps(og),
+            media_info["filesystem_id"],
+            media_info["created_ts"],
+        )
+
+        respond_with_json_bytes(request, 200, json.dumps(og), send_cors=True)
+
+
+    @defer.inlineCallbacks
+    def _calc_og(self, tree, media_info, requester):
+        # suck our tree into lxml and define our OG response.
+
+        # if we see any image URLs in the OG response, then spider them
+        # (although the client could choose to do this by asking for previews of those
+        # URLs to avoid DoSing the server)
+
+        # "og:type"         : "video",
+        # "og:url"          : "https://www.youtube.com/watch?v=LXDBoHyjmtw",
+        # "og:site_name"    : "YouTube",
+        # "og:video:type"   : "application/x-shockwave-flash",
+        # "og:description"  : "Fun stuff happening here",
+        # "og:title"        : "RemoteJam - Matrix team hack for Disrupt Europe Hackathon",
+        # "og:image"        : "https://i.ytimg.com/vi/LXDBoHyjmtw/maxresdefault.jpg",
+        # "og:video:url"    : "http://www.youtube.com/v/LXDBoHyjmtw?version=3&autohide=1",
+        # "og:video:width"  : "1280"
+        # "og:video:height" : "720",
+        # "og:video:secure_url": "https://www.youtube.com/v/LXDBoHyjmtw?version=3",
+
+        og = {}
+        for tag in tree.xpath("//*/meta[starts-with(@property, 'og:')]"):
+            og[tag.attrib['property']] = tag.attrib['content']
+
+        # TODO: grab article: meta tags too, e.g.:
+
+        # "article:publisher" : "https://www.facebook.com/thethudonline" />
+        # "article:author" content="https://www.facebook.com/thethudonline" />
+        # "article:tag" content="baby" />
+        # "article:section" content="Breaking News" />
+        # "article:published_time" content="2016-03-31T19:58:24+00:00" />
+        # "article:modified_time" content="2016-04-01T18:31:53+00:00" />
+
+        if 'og:title' not in og:
+            # do some basic spidering of the HTML
+            title = tree.xpath("(//title)[1] | (//h1)[1] | (//h2)[1] | (//h3)[1]")
+            og['og:title'] = title[0].text.strip() if title else None
+
+        if 'og:image' not in og:
+            # TODO: extract a favicon failing all else
+            meta_image = tree.xpath(
+                "//*/meta[translate(@itemprop, 'IMAGE', 'image')='image']/@content"
+            )
+            if meta_image:
+                og['og:image'] = self._rebase_url(meta_image[0], media_info['uri'])
+            else:
+                # TODO: consider inlined CSS styles as well as width & height attribs
+                images = tree.xpath("//img[@src][number(@width)>10][number(@height)>10]")
+                images = sorted(images, key=lambda i: (
+                    -1 * int(i.attrib['width']) * int(i.attrib['height'])
+                ))
+                if not images:
+                    images = tree.xpath("//img[@src]")
+                if images:
+                    og['og:image'] = images[0].attrib['src']
+
+        # pre-cache the image for posterity
+        # FIXME: it might be cleaner to use the same flow as the main /preview_url request
+        # itself and benefit from the same caching etc.  But for now we just rely on the
+        # caching on the master request to speed things up.
+        if 'og:image' in og and og['og:image']:
+            image_info = yield self._download_url(
+                self._rebase_url(og['og:image'], media_info['uri']), requester.user
+            )
+
+            if self._is_media(image_info['media_type']):
+                # TODO: make sure we don't choke on white-on-transparent images
+                dims = yield self._generate_local_thumbnails(
+                    image_info['filesystem_id'], image_info
+                )
+                if dims:
+                    og["og:image:width"] = dims['width']
+                    og["og:image:height"] = dims['height']
+                else:
+                    logger.warn("Couldn't get dims for %s" % og["og:image"])
+
+                og["og:image"] = "mxc://%s/%s" % (
+                    self.server_name, image_info['filesystem_id']
+                )
+                og["og:image:type"] = image_info['media_type']
+                og["matrix:image:size"] = image_info['media_length']
+            else:
+                del og["og:image"]
+
+        if 'og:description' not in og:
+            meta_description = tree.xpath(
+                "//*/meta"
+                "[translate(@name, 'DESCRIPTION', 'description')='description']"
+                "/@content")
+            if meta_description:
+                og['og:description'] = meta_description[0]
+            else:
+                # grab any text nodes which are inside the <body/> tag...
+                # unless they are within an HTML5 semantic markup tag...
+                # <header/>, <nav/>, <aside/>, <footer/>
+                # ...or if they are within a <script/> or <style/> tag.
+                # This is a very very very coarse approximation to a plain text
+                # render of the page.
+                text_nodes = tree.xpath("//text()[not(ancestor::header | ancestor::nav | "
+                                        "ancestor::aside | ancestor::footer | "
+                                        "ancestor::script | ancestor::style)]" +
+                                        "[ancestor::body]")
+                text = ''
+                for text_node in text_nodes:
+                    if len(text) < 500:
+                        text += text_node + ' '
+                    else:
+                        break
+                text = re.sub(r'[\t ]+', ' ', text)
+                text = re.sub(r'[\t \r\n]*[\r\n]+', '\n', text)
+                text = text.strip()[:500]
+                og['og:description'] = text if text else None
+
+        # TODO: delete the url downloads to stop diskfilling,
+        # as we only ever cared about its OG
+        defer.returnValue(og)
+
+    def _rebase_url(self, url, base):
+        base = list(urlparse(base))
+        url = list(urlparse(url))
+        if not url[0]:  # fix up schema
+            url[0] = base[0] or "http"
+        if not url[1]:  # fix up hostname
+            url[1] = base[1]
+            if not url[2].startswith('/'):
+                url[2] = re.sub(r'/[^/]+$', '/', base[2]) + url[2]
+        return urlunparse(url)
+
+    @defer.inlineCallbacks
+    def _download_url(self, url, user):
+        # TODO: we should probably honour robots.txt... except in practice
+        # we're most likely being explicitly triggered by a human rather than a
+        # bot, so are we really a robot?
+
+        # XXX: horrible duplication with base_resource's _download_remote_file()
+        file_id = random_string(24)
+
+        fname = self.filepaths.local_media_filepath(file_id)
+        self._makedirs(fname)
+
+        try:
+            with open(fname, "wb") as f:
+                logger.debug("Trying to get url '%s'" % url)
+                length, headers, uri, code = yield self.client.get_file(
+                    url, output_stream=f, max_size=self.max_spider_size,
+                )
+                # FIXME: pass through 404s and other error messages nicely
+
+            media_type = headers["Content-Type"][0]
+            time_now_ms = self.clock.time_msec()
+
+            content_disposition = headers.get("Content-Disposition", None)
+            if content_disposition:
+                _, params = cgi.parse_header(content_disposition[0],)
+                download_name = None
+
+                # First check if there is a valid UTF-8 filename
+                download_name_utf8 = params.get("filename*", None)
+                if download_name_utf8:
+                    if download_name_utf8.lower().startswith("utf-8''"):
+                        download_name = download_name_utf8[7:]
+
+                # If there isn't check for an ascii name.
+                if not download_name:
+                    download_name_ascii = params.get("filename", None)
+                    if download_name_ascii and is_ascii(download_name_ascii):
+                        download_name = download_name_ascii
+
+                if download_name:
+                    download_name = urlparse.unquote(download_name)
+                    try:
+                        download_name = download_name.decode("utf-8")
+                    except UnicodeDecodeError:
+                        download_name = None
+            else:
+                download_name = None
+
+            yield self.store.store_local_media(
+                media_id=file_id,
+                media_type=media_type,
+                time_now_ms=self.clock.time_msec(),
+                upload_name=download_name,
+                media_length=length,
+                user_id=user,
+            )
+
+        except Exception as e:
+            os.remove(fname)
+            raise SynapseError(
+                500, ("Failed to download content: %s" % e),
+                Codes.UNKNOWN
+            )
+
+        defer.returnValue({
+            "media_type": media_type,
+            "media_length": length,
+            "download_name": download_name,
+            "created_ts": time_now_ms,
+            "filesystem_id": file_id,
+            "filename": fname,
+            "uri": uri,
+            "response_code": code,
+            # FIXME: we should calculate a proper expiration based on the
+            # Cache-Control and Expire headers.  But for now, assume 1 hour.
+            "expires": 60 * 60 * 1000,
+            "etag": headers["ETag"][0] if "ETag" in headers else None,
+        })
+
+    def _is_media(self, content_type):
+        if content_type.lower().startswith("image/"):
+            return True
+
+    def _is_html(self, content_type):
+        content_type = content_type.lower()
+        if (
+            content_type.startswith("text/html") or
+            content_type.startswith("application/xhtml")
+        ):
+            return True
diff --git a/synapse/rest/media/v1/thumbnail_resource.py b/synapse/rest/media/v1/thumbnail_resource.py
index ab52499785..513b445688 100644
--- a/synapse/rest/media/v1/thumbnail_resource.py
+++ b/synapse/rest/media/v1/thumbnail_resource.py
@@ -72,6 +72,11 @@ class ThumbnailResource(BaseMediaResource):
             self._respond_404(request)
             return
 
+        if media_info["media_type"] == "image/svg+xml":
+            file_path = self.filepaths.local_media_filepath(media_id)
+            yield self._respond_with_file(request, media_info["media_type"], file_path)
+            return
+
         thumbnail_infos = yield self.store.get_local_media_thumbnails(media_id)
 
         if thumbnail_infos:
@@ -103,6 +108,11 @@ class ThumbnailResource(BaseMediaResource):
             self._respond_404(request)
             return
 
+        if media_info["media_type"] == "image/svg+xml":
+            file_path = self.filepaths.local_media_filepath(media_id)
+            yield self._respond_with_file(request, media_info["media_type"], file_path)
+            return
+
         thumbnail_infos = yield self.store.get_local_media_thumbnails(media_id)
         for info in thumbnail_infos:
             t_w = info["thumbnail_width"] == desired_width
@@ -138,6 +148,11 @@ class ThumbnailResource(BaseMediaResource):
                                              desired_method, desired_type):
         media_info = yield self._get_remote_media(server_name, media_id)
 
+        if media_info["media_type"] == "image/svg+xml":
+            file_path = self.filepaths.remote_media_filepath(server_name, media_id)
+            yield self._respond_with_file(request, media_info["media_type"], file_path)
+            return
+
         thumbnail_infos = yield self.store.get_remote_media_thumbnails(
             server_name, media_id,
         )
@@ -181,6 +196,11 @@ class ThumbnailResource(BaseMediaResource):
         # We should proxy the thumbnail from the remote server instead.
         media_info = yield self._get_remote_media(server_name, media_id)
 
+        if media_info["media_type"] == "image/svg+xml":
+            file_path = self.filepaths.remote_media_filepath(server_name, media_id)
+            yield self._respond_with_file(request, media_info["media_type"], file_path)
+            return
+
         thumbnail_infos = yield self.store.get_remote_media_thumbnails(
             server_name, media_id,
         )
@@ -208,6 +228,8 @@ class ThumbnailResource(BaseMediaResource):
     @defer.inlineCallbacks
     def _respond_default_thumbnail(self, request, media_info, width, height,
                                    method, m_type):
+        # XXX: how is this meant to work? store.get_default_thumbnails
+        # appears to always return [] so won't this always 404?
         media_type = media_info["media_type"]
         top_level_type = media_type.split("/")[0]
         sub_type = media_type.split("/")[-1].split(";")[0]