From ba7a91aea5fd624bf048f0fda0dca80da7a1945e Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Tue, 21 Sep 2021 12:09:57 -0400 Subject: Refactor oEmbed previews (#10814) The major change is moving the decision of whether to use oEmbed further up the call-stack. This reverts the _download_url method to being a "dumb" functionwhich takes a single URL and downloads it (as it was before #7920). This also makes more minor refactorings: * Renames internal variables for clarity. * Factors out shared code between the HTML and rich oEmbed previews. * Fixes tests to preview an oEmbed image. --- synapse/rest/media/v1/oembed.py | 145 +++++++----- synapse/rest/media/v1/preview_url_resource.py | 326 +++++++++++++++----------- 2 files changed, 272 insertions(+), 199 deletions(-) (limited to 'synapse/rest/media') diff --git a/synapse/rest/media/v1/oembed.py b/synapse/rest/media/v1/oembed.py index 2e6706dbfa..8b74e72655 100644 --- a/synapse/rest/media/v1/oembed.py +++ b/synapse/rest/media/v1/oembed.py @@ -12,11 +12,14 @@ # See the License for the specific language governing permissions and # limitations under the License. import logging +import urllib.parse from typing import TYPE_CHECKING, Optional import attr from synapse.http.client import SimpleHttpClient +from synapse.types import JsonDict +from synapse.util import json_decoder if TYPE_CHECKING: from synapse.server import HomeServer @@ -24,18 +27,15 @@ if TYPE_CHECKING: logger = logging.getLogger(__name__) -@attr.s(slots=True, auto_attribs=True) +@attr.s(slots=True, frozen=True, auto_attribs=True) class OEmbedResult: - # Either HTML content or URL must be provided. - html: Optional[str] - url: Optional[str] - title: Optional[str] - # Number of seconds to cache the content. - cache_age: int - - -class OEmbedError(Exception): - """An error occurred processing the oEmbed object.""" + # The Open Graph result (converted from the oEmbed result). + open_graph_result: JsonDict + # Number of seconds to cache the content, according to the oEmbed response. + # + # This will be None if no cache-age is provided in the oEmbed response (or + # if the oEmbed response cannot be turned into an Open Graph response). + cache_age: Optional[int] class OEmbedProvider: @@ -81,75 +81,106 @@ class OEmbedProvider: """ for url_pattern, endpoint in self._oembed_patterns.items(): if url_pattern.fullmatch(url): - return endpoint + # TODO Specify max height / width. + + # Note that only the JSON format is supported, some endpoints want + # this in the URL, others want it as an argument. + endpoint = endpoint.replace("{format}", "json") + + args = {"url": url, "format": "json"} + query_str = urllib.parse.urlencode(args, True) + return f"{endpoint}?{query_str}" # No match. return None - async def get_oembed_content(self, endpoint: str, url: str) -> OEmbedResult: + def parse_oembed_response(self, url: str, raw_body: bytes) -> OEmbedResult: """ - Request content from an oEmbed endpoint. + Parse the oEmbed response into an Open Graph response. Args: - endpoint: The oEmbed API endpoint. - url: The URL to pass to the API. + url: The URL which is being previewed (not the one which was + requested). + raw_body: The oEmbed response as JSON encoded as bytes. Returns: - An object representing the metadata returned. - - Raises: - OEmbedError if fetching or parsing of the oEmbed information fails. + json-encoded Open Graph data """ - try: - logger.debug("Trying to get oEmbed content for url '%s'", url) - # Note that only the JSON format is supported, some endpoints want - # this in the URL, others want it as an argument. - endpoint = endpoint.replace("{format}", "json") - - result = await self._client.get_json( - endpoint, - # TODO Specify max height / width. - args={"url": url, "format": "json"}, - ) + try: + # oEmbed responses *must* be UTF-8 according to the spec. + oembed = json_decoder.decode(raw_body.decode("utf-8")) # Ensure there's a version of 1.0. - if result.get("version") != "1.0": - raise OEmbedError("Invalid version: %s" % (result.get("version"),)) - - oembed_type = result.get("type") + oembed_version = oembed["version"] + if oembed_version != "1.0": + raise RuntimeError(f"Invalid version: {oembed_version}") # Ensure the cache age is None or an int. - cache_age = result.get("cache_age") + cache_age = oembed.get("cache_age") if cache_age: cache_age = int(cache_age) - oembed_result = OEmbedResult(None, None, result.get("title"), cache_age) + # The results. + open_graph_response = {"og:title": oembed.get("title")} - # HTML content. + # If a thumbnail exists, use it. Note that dimensions will be calculated later. + if "thumbnail_url" in oembed: + open_graph_response["og:image"] = oembed["thumbnail_url"] + + # Process each type separately. + oembed_type = oembed["type"] if oembed_type == "rich": - oembed_result.html = result.get("html") - return oembed_result + calc_description_and_urls(open_graph_response, oembed["html"]) - if oembed_type == "photo": - oembed_result.url = result.get("url") - return oembed_result + elif oembed_type == "photo": + # If this is a photo, use the full image, not the thumbnail. + open_graph_response["og:image"] = oembed["url"] - # TODO Handle link and video types. + else: + raise RuntimeError(f"Unknown oEmbed type: {oembed_type}") - if "thumbnail_url" in result: - oembed_result.url = result.get("thumbnail_url") - return oembed_result + except Exception as e: + # Trap any exception and let the code follow as usual. + logger.warning(f"Error parsing oEmbed metadata from {url}: {e:r}") + open_graph_response = {} + cache_age = None - raise OEmbedError("Incompatible oEmbed information.") + return OEmbedResult(open_graph_response, cache_age) - except OEmbedError as e: - # Trap OEmbedErrors first so we can directly re-raise them. - logger.warning("Error parsing oEmbed metadata from %s: %r", url, e) - raise - except Exception as e: - # Trap any exception and let the code follow as usual. - # FIXME: pass through 404s and other error messages nicely - logger.warning("Error downloading oEmbed metadata from %s: %r", url, e) - raise OEmbedError() from e +def calc_description_and_urls(open_graph_response: JsonDict, html_body: str) -> None: + """ + Calculate description for an HTML document. + + This uses lxml to convert the HTML document into plaintext. If errors + occur during processing of the document, an empty response is returned. + + Args: + open_graph_response: The current Open Graph summary. This is updated with additional fields. + html_body: The HTML document, as bytes. + + Returns: + The summary + """ + # If there's no body, nothing useful is going to be found. + if not html_body: + return + + from lxml import etree + + # Create an HTML parser. If this fails, log and return no metadata. + parser = etree.HTMLParser(recover=True, encoding="utf-8") + + # Attempt to parse the body. If this fails, log and return no metadata. + tree = etree.fromstring(html_body, parser) + + # The data was successfully parsed, but no tree was found. + if tree is None: + return + + from synapse.rest.media.v1.preview_url_resource import _calc_description + + description = _calc_description(tree) + if description: + open_graph_response["og:description"] = description diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py index fe0627d9b0..0a0b476d2b 100644 --- a/synapse/rest/media/v1/preview_url_resource.py +++ b/synapse/rest/media/v1/preview_url_resource.py @@ -44,7 +44,7 @@ from synapse.logging.context import make_deferred_yieldable, run_in_background from synapse.metrics.background_process_metrics import run_as_background_process from synapse.rest.media.v1._base import get_filename_from_headers from synapse.rest.media.v1.media_storage import MediaStorage -from synapse.rest.media.v1.oembed import OEmbedError, OEmbedProvider +from synapse.rest.media.v1.oembed import OEmbedProvider from synapse.types import JsonDict from synapse.util import json_encoder from synapse.util.async_helpers import ObservableDeferred @@ -73,6 +73,7 @@ OG_TAG_NAME_MAXLEN = 50 OG_TAG_VALUE_MAXLEN = 1000 ONE_HOUR = 60 * 60 * 1000 +ONE_DAY = 24 * ONE_HOUR @attr.s(slots=True, frozen=True, auto_attribs=True) @@ -255,10 +256,19 @@ class PreviewUrlResource(DirectServeJsonResource): og = og.encode("utf8") return og - media_info = await self._download_url(url, user) + # If this URL can be accessed via oEmbed, use that instead. + url_to_download = url + oembed_url = self._oembed.get_oembed_url(url) + if oembed_url: + url_to_download = oembed_url + + media_info = await self._download_url(url_to_download, user) logger.debug("got media_info of '%s'", media_info) + # The number of milliseconds that the response should be considered valid. + expiration_ms = media_info.expires + if _is_media(media_info.media_type): file_id = media_info.filesystem_id dims = await self.media_repo._generate_thumbnails( @@ -288,34 +298,22 @@ class PreviewUrlResource(DirectServeJsonResource): encoding = get_html_media_encoding(body, media_info.media_type) og = decode_and_calc_og(body, media_info.uri, encoding) - # 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 = await self._download_url( - _rebase_url(og["og:image"], media_info.uri), user - ) + await self._precache_image_url(user, media_info, og) + + elif oembed_url and _is_json(media_info.media_type): + # Handle an oEmbed response. + with open(media_info.filename, "rb") as file: + body = file.read() + + oembed_response = self._oembed.parse_oembed_response(media_info.uri, body) + og = oembed_response.open_graph_result + + # Use the cache age from the oEmbed result, instead of the HTTP response. + if oembed_response.cache_age is not None: + expiration_ms = oembed_response.cache_age + + await self._precache_image_url(user, media_info, og) - if _is_media(image_info.media_type): - # TODO: make sure we don't choke on white-on-transparent images - file_id = image_info.filesystem_id - dims = await self.media_repo._generate_thumbnails( - None, file_id, file_id, image_info.media_type, url_cache=True - ) - if dims: - og["og:image:width"] = dims["width"] - og["og:image:height"] = dims["height"] - else: - logger.warning("Couldn't get dims for %s", og["og:image"]) - - og[ - "og:image" - ] = f"mxc://{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"] else: logger.warning("Failed to find any OG data in %s", url) og = {} @@ -336,12 +334,15 @@ class PreviewUrlResource(DirectServeJsonResource): jsonog = json_encoder.encode(og) + # Cap the amount of time to consider a response valid. + expiration_ms = min(expiration_ms, ONE_DAY) + # store OG in history-aware DB cache await self.store.store_url_cache( url, media_info.response_code, media_info.etag, - media_info.expires + media_info.created_ts_ms, + media_info.created_ts_ms + expiration_ms, jsonog, media_info.filesystem_id, media_info.created_ts_ms, @@ -358,88 +359,52 @@ class PreviewUrlResource(DirectServeJsonResource): file_info = FileInfo(server_name=None, file_id=file_id, url_cache=True) - # If this URL can be accessed via oEmbed, use that instead. - url_to_download: Optional[str] = url - oembed_url = self._oembed.get_oembed_url(url) - if oembed_url: - # The result might be a new URL to download, or it might be HTML content. + with self.media_storage.store_into_file(file_info) as (f, fname, finish): try: - oembed_result = await self._oembed.get_oembed_content(oembed_url, url) - if oembed_result.url: - url_to_download = oembed_result.url - elif oembed_result.html: - url_to_download = None - except OEmbedError: - # If an error occurs, try doing a normal preview. - pass + logger.debug("Trying to get preview for url '%s'", url) + length, headers, uri, code = await self.client.get_file( + url, + output_stream=f, + max_size=self.max_spider_size, + headers={"Accept-Language": self.url_preview_accept_language}, + ) + except SynapseError: + # Pass SynapseErrors through directly, so that the servlet + # handler will return a SynapseError to the client instead of + # blank data or a 500. + raise + except DNSLookupError: + # DNS lookup returned no results + # Note: This will also be the case if one of the resolved IP + # addresses is blacklisted + raise SynapseError( + 502, + "DNS resolution failure during URL preview generation", + Codes.UNKNOWN, + ) + except Exception as e: + # FIXME: pass through 404s and other error messages nicely + logger.warning("Error downloading %s: %r", url, e) - if url_to_download: - with self.media_storage.store_into_file(file_info) as (f, fname, finish): - try: - logger.debug("Trying to get preview for url '%s'", url_to_download) - length, headers, uri, code = await self.client.get_file( - url_to_download, - output_stream=f, - max_size=self.max_spider_size, - headers={"Accept-Language": self.url_preview_accept_language}, - ) - except SynapseError: - # Pass SynapseErrors through directly, so that the servlet - # handler will return a SynapseError to the client instead of - # blank data or a 500. - raise - except DNSLookupError: - # DNS lookup returned no results - # Note: This will also be the case if one of the resolved IP - # addresses is blacklisted - raise SynapseError( - 502, - "DNS resolution failure during URL preview generation", - Codes.UNKNOWN, - ) - except Exception as e: - # FIXME: pass through 404s and other error messages nicely - logger.warning("Error downloading %s: %r", url_to_download, e) - - raise SynapseError( - 500, - "Failed to download content: %s" - % (traceback.format_exception_only(sys.exc_info()[0], e),), - Codes.UNKNOWN, - ) - await finish() - - if b"Content-Type" in headers: - media_type = headers[b"Content-Type"][0].decode("ascii") - else: - media_type = "application/octet-stream" + raise SynapseError( + 500, + "Failed to download content: %s" + % (traceback.format_exception_only(sys.exc_info()[0], e),), + Codes.UNKNOWN, + ) + await finish() - download_name = get_filename_from_headers(headers) + if b"Content-Type" in headers: + media_type = headers[b"Content-Type"][0].decode("ascii") + else: + media_type = "application/octet-stream" - # FIXME: we should calculate a proper expiration based on the - # Cache-Control and Expire headers. But for now, assume 1 hour. - expires = ONE_HOUR - etag = ( - headers[b"ETag"][0].decode("ascii") if b"ETag" in headers else None - ) - else: - # we can only get here if we did an oembed request and have an oembed_result.html - assert oembed_result.html is not None - assert oembed_url is not None - - html_bytes = oembed_result.html.encode("utf-8") - with self.media_storage.store_into_file(file_info) as (f, fname, finish): - f.write(html_bytes) - await finish() - - media_type = "text/html" - download_name = oembed_result.title - length = len(html_bytes) - # If a specific cache age was not given, assume 1 hour. - expires = oembed_result.cache_age or ONE_HOUR - uri = oembed_url - code = 200 - etag = None + download_name = get_filename_from_headers(headers) + + # FIXME: we should calculate a proper expiration based on the + # Cache-Control and Expire headers. But for now, assume 1 hour. + expires = ONE_HOUR + etag = headers[b"ETag"][0].decode("ascii") if b"ETag" in headers else None try: time_now_ms = self.clock.time_msec() @@ -474,6 +439,46 @@ class PreviewUrlResource(DirectServeJsonResource): etag=etag, ) + async def _precache_image_url( + self, user: str, media_info: MediaInfo, og: JsonDict + ) -> None: + """ + Pre-cache the image (if one exists) for posterity + + Args: + user: The user requesting the preview. + media_info: The media being previewed. + og: The Open Graph dictionary. This is modified with image information. + """ + # If there's no image or it is blank, there's nothing to do. + if "og:image" not in og or not og["og:image"]: + return + + # 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. + image_info = await self._download_url( + _rebase_url(og["og:image"], media_info.uri), user + ) + + if _is_media(image_info.media_type): + # TODO: make sure we don't choke on white-on-transparent images + file_id = image_info.filesystem_id + dims = await self.media_repo._generate_thumbnails( + None, file_id, file_id, image_info.media_type, url_cache=True + ) + if dims: + og["og:image:width"] = dims["width"] + og["og:image:height"] = dims["height"] + else: + logger.warning("Couldn't get dims for %s", og["og:image"]) + + og["og:image"] = f"mxc://{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"] + def _start_expire_url_cache_data(self) -> Deferred: return run_as_background_process( "expire_url_cache_data", self._expire_url_cache_data @@ -527,7 +532,7 @@ class PreviewUrlResource(DirectServeJsonResource): # These may be cached for a bit on the client (i.e., they # may have a room open with a preview url thing open). # So we wait a couple of days before deleting, just in case. - expire_before = now - 2 * 24 * ONE_HOUR + expire_before = now - 2 * ONE_DAY media_ids = await self.store.get_url_cache_media_before(expire_before) removed_media = [] @@ -669,7 +674,18 @@ def decode_and_calc_og( def _calc_og(tree: "etree.Element", media_uri: str) -> Dict[str, Optional[str]]: - # suck our tree into lxml and define our OG response. + """ + Calculate metadata for an HTML document. + + This uses lxml to search the HTML document for Open Graph data. + + Args: + tree: The parsed HTML document. + media_url: The URI used to download the body. + + Returns: + The Open Graph response as a dictionary. + """ # 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 @@ -743,35 +759,7 @@ def _calc_og(tree: "etree.Element", media_uri: str) -> Dict[str, Optional[str]]: if meta_description: og["og:description"] = meta_description[0] else: - # grab any text nodes which are inside the tag... - # unless they are within an HTML5 semantic markup tag... - #
,