summaryrefslogtreecommitdiff
path: root/synapse/rest/media
diff options
context:
space:
mode:
Diffstat (limited to 'synapse/rest/media')
-rw-r--r--synapse/rest/media/v1/__init__.py38
-rw-r--r--synapse/rest/media/v1/oembed.py28
-rw-r--r--synapse/rest/media/v1/preview_url_resource.py186
-rw-r--r--synapse/rest/media/v1/thumbnailer.py21
4 files changed, 174 insertions, 99 deletions
diff --git a/synapse/rest/media/v1/__init__.py b/synapse/rest/media/v1/__init__.py
index 3dd16d4b..d5b74cdd 100644
--- a/synapse/rest/media/v1/__init__.py
+++ b/synapse/rest/media/v1/__init__.py
@@ -12,33 +12,21 @@
# See the License for the specific language governing permissions and
# limitations under the License.
-import PIL.Image
+from PIL.features import check_codec
# check for JPEG support.
-try:
- PIL.Image._getdecoder("rgb", "jpeg", None)
-except OSError as e:
- if str(e).startswith("decoder jpeg not available"):
- raise Exception(
- "FATAL: jpeg codec not supported. Install pillow correctly! "
- " 'sudo apt-get install libjpeg-dev' then 'pip uninstall pillow &&"
- " pip install pillow --user'"
- )
-except Exception:
- # any other exception is fine
- pass
+if not check_codec("jpg"):
+ raise Exception(
+ "FATAL: jpeg codec not supported. Install pillow correctly! "
+ " 'sudo apt-get install libjpeg-dev' then 'pip uninstall pillow &&"
+ " pip install pillow --user'"
+ )
# check for PNG support.
-try:
- PIL.Image._getdecoder("rgb", "zip", None)
-except OSError as e:
- if str(e).startswith("decoder zip not available"):
- raise Exception(
- "FATAL: zip codec not supported. Install pillow correctly! "
- " 'sudo apt-get install libjpeg-dev' then 'pip uninstall pillow &&"
- " pip install pillow --user'"
- )
-except Exception:
- # any other exception is fine
- pass
+if not check_codec("zlib"):
+ raise Exception(
+ "FATAL: zip codec not supported. Install pillow correctly! "
+ " 'sudo apt-get install libjpeg-dev' then 'pip uninstall pillow &&"
+ " pip install pillow --user'"
+ )
diff --git a/synapse/rest/media/v1/oembed.py b/synapse/rest/media/v1/oembed.py
index e04671fb..78b1603f 100644
--- a/synapse/rest/media/v1/oembed.py
+++ b/synapse/rest/media/v1/oembed.py
@@ -96,6 +96,32 @@ class OEmbedProvider:
# No match.
return None
+ def autodiscover_from_html(self, tree: "etree.Element") -> Optional[str]:
+ """
+ Search an HTML document for oEmbed autodiscovery information.
+
+ Args:
+ tree: The parsed HTML body.
+
+ Returns:
+ The URL to use for oEmbed information, or None if no URL was found.
+ """
+ # Search for link elements with the proper rel and type attributes.
+ for tag in tree.xpath(
+ "//link[@rel='alternate'][@type='application/json+oembed']"
+ ):
+ if "href" in tag.attrib:
+ return tag.attrib["href"]
+
+ # Some providers (e.g. Flickr) use alternative instead of alternate.
+ for tag in tree.xpath(
+ "//link[@rel='alternative'][@type='application/json+oembed']"
+ ):
+ if "href" in tag.attrib:
+ return tag.attrib["href"]
+
+ return None
+
def parse_oembed_response(self, url: str, raw_body: bytes) -> OEmbedResult:
"""
Parse the oEmbed response into an Open Graph response.
@@ -165,7 +191,7 @@ class OEmbedProvider:
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}")
+ logger.warning("Error parsing oEmbed metadata from %s: %r", url, e)
open_graph_response = {}
cache_age = None
diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py
index 79a42b24..1fe0fc8a 100644
--- a/synapse/rest/media/v1/preview_url_resource.py
+++ b/synapse/rest/media/v1/preview_url_resource.py
@@ -22,7 +22,7 @@ import re
import shutil
import sys
import traceback
-from typing import TYPE_CHECKING, Dict, Generator, Iterable, Optional, Union
+from typing import TYPE_CHECKING, Dict, Generator, Iterable, Optional, Tuple, Union
from urllib import parse as urlparse
import attr
@@ -73,6 +73,7 @@ OG_TAG_VALUE_MAXLEN = 1000
ONE_HOUR = 60 * 60 * 1000
ONE_DAY = 24 * ONE_HOUR
+IMAGE_CACHE_EXPIRY_MS = 2 * ONE_DAY
@attr.s(slots=True, frozen=True, auto_attribs=True)
@@ -295,22 +296,32 @@ class PreviewUrlResource(DirectServeJsonResource):
body = file.read()
encoding = get_html_media_encoding(body, media_info.media_type)
- og = decode_and_calc_og(body, media_info.uri, encoding)
-
- 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(url, 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
+ tree = decode_body(body, encoding)
+ if tree is not None:
+ # Check if this HTML document points to oEmbed information and
+ # defer to that.
+ oembed_url = self._oembed.autodiscover_from_html(tree)
+ og = {}
+ if oembed_url:
+ oembed_info = await self._download_url(oembed_url, user)
+ og, expiration_ms = await self._handle_oembed_response(
+ url, oembed_info, expiration_ms
+ )
+
+ # If there was no oEmbed URL (or oEmbed parsing failed), attempt
+ # to generate the Open Graph information from the HTML.
+ if not oembed_url or not og:
+ og = _calc_og(tree, media_info.uri)
+
+ await self._precache_image_url(user, media_info, og)
+ else:
+ og = {}
+ elif oembed_url:
+ # Handle the oEmbed information.
+ og, expiration_ms = await self._handle_oembed_response(
+ url, media_info, expiration_ms
+ )
await self._precache_image_url(user, media_info, og)
else:
@@ -478,6 +489,39 @@ class PreviewUrlResource(DirectServeJsonResource):
else:
del og["og:image"]
+ async def _handle_oembed_response(
+ self, url: str, media_info: MediaInfo, expiration_ms: int
+ ) -> Tuple[JsonDict, int]:
+ """
+ Parse the downloaded oEmbed info.
+
+ Args:
+ url: The URL which is being previewed (not the one which was
+ requested).
+ media_info: The media being previewed.
+ expiration_ms: The length of time, in milliseconds, the media is valid for.
+
+ Returns:
+ A tuple of:
+ The Open Graph dictionary, if the oEmbed info can be parsed.
+ The (possibly updated) length of time, in milliseconds, the media is valid for.
+ """
+ # If JSON was not returned, there's nothing to do.
+ if not _is_json(media_info.media_type):
+ return {}, expiration_ms
+
+ with open(media_info.filename, "rb") as file:
+ body = file.read()
+
+ oembed_response = self._oembed.parse_oembed_response(url, body)
+ open_graph_result = oembed_response.open_graph_result
+
+ # Use the cache age from the oEmbed result, if one was given.
+ if open_graph_result and oembed_response.cache_age is not None:
+ expiration_ms = oembed_response.cache_age
+
+ return open_graph_result, expiration_ms
+
def _start_expire_url_cache_data(self) -> Deferred:
return run_as_background_process(
"expire_url_cache_data", self._expire_url_cache_data
@@ -496,6 +540,27 @@ class PreviewUrlResource(DirectServeJsonResource):
logger.info("Still running DB updates; skipping expiry")
return
+ def try_remove_parent_dirs(dirs: Iterable[str]) -> None:
+ """Attempt to remove the given chain of parent directories
+
+ Args:
+ dirs: The list of directory paths to delete, with children appearing
+ before their parents.
+ """
+ for dir in dirs:
+ try:
+ os.rmdir(dir)
+ except FileNotFoundError:
+ # Already deleted, continue with deleting the rest
+ pass
+ except OSError as e:
+ # Failed, skip deleting the rest of the parent dirs
+ if e.errno != errno.ENOTEMPTY:
+ logger.warning(
+ "Failed to remove media directory: %r: %s", dir, e
+ )
+ break
+
# First we delete expired url cache entries
media_ids = await self.store.get_expired_url_cache(now)
@@ -504,20 +569,16 @@ class PreviewUrlResource(DirectServeJsonResource):
fname = self.filepaths.url_cache_filepath(media_id)
try:
os.remove(fname)
+ except FileNotFoundError:
+ pass # If the path doesn't exist, meh
except OSError as e:
- # If the path doesn't exist, meh
- if e.errno != errno.ENOENT:
- logger.warning("Failed to remove media: %r: %s", media_id, e)
- continue
+ logger.warning("Failed to remove media: %r: %s", media_id, e)
+ continue
removed_media.append(media_id)
- try:
- dirs = self.filepaths.url_cache_filepath_dirs_to_delete(media_id)
- for dir in dirs:
- os.rmdir(dir)
- except Exception:
- pass
+ dirs = self.filepaths.url_cache_filepath_dirs_to_delete(media_id)
+ try_remove_parent_dirs(dirs)
await self.store.delete_url_cache(removed_media)
@@ -530,7 +591,7 @@ 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 * ONE_DAY
+ expire_before = now - IMAGE_CACHE_EXPIRY_MS
media_ids = await self.store.get_url_cache_media_before(expire_before)
removed_media = []
@@ -538,36 +599,30 @@ class PreviewUrlResource(DirectServeJsonResource):
fname = self.filepaths.url_cache_filepath(media_id)
try:
os.remove(fname)
+ except FileNotFoundError:
+ pass # If the path doesn't exist, meh
except OSError as e:
- # If the path doesn't exist, meh
- if e.errno != errno.ENOENT:
- logger.warning("Failed to remove media: %r: %s", media_id, e)
- continue
+ logger.warning("Failed to remove media: %r: %s", media_id, e)
+ continue
- try:
- dirs = self.filepaths.url_cache_filepath_dirs_to_delete(media_id)
- for dir in dirs:
- os.rmdir(dir)
- except Exception:
- pass
+ dirs = self.filepaths.url_cache_filepath_dirs_to_delete(media_id)
+ try_remove_parent_dirs(dirs)
thumbnail_dir = self.filepaths.url_cache_thumbnail_directory(media_id)
try:
shutil.rmtree(thumbnail_dir)
+ except FileNotFoundError:
+ pass # If the path doesn't exist, meh
except OSError as e:
- # If the path doesn't exist, meh
- if e.errno != errno.ENOENT:
- logger.warning("Failed to remove media: %r: %s", media_id, e)
- continue
+ logger.warning("Failed to remove media: %r: %s", media_id, e)
+ continue
removed_media.append(media_id)
- try:
- dirs = self.filepaths.url_cache_thumbnail_dirs_to_delete(media_id)
- for dir in dirs:
- os.rmdir(dir)
- except Exception:
- pass
+ dirs = self.filepaths.url_cache_thumbnail_dirs_to_delete(media_id)
+ # Note that one of the directories to be deleted has already been
+ # removed by the `rmtree` above.
+ try_remove_parent_dirs(dirs)
await self.store.delete_url_cache_media(removed_media)
@@ -619,26 +674,22 @@ def get_html_media_encoding(body: bytes, content_type: str) -> str:
return "utf-8"
-def decode_and_calc_og(
- body: bytes, media_uri: str, request_encoding: Optional[str] = None
-) -> JsonDict:
+def decode_body(
+ body: bytes, request_encoding: Optional[str] = None
+) -> Optional["etree.Element"]:
"""
- Calculate metadata for an HTML document.
-
- This uses lxml to parse the HTML document into the OG response. If errors
- occur during processing of the document, an empty response is returned.
+ This uses lxml to parse the HTML document.
Args:
body: The HTML document, as bytes.
- media_url: The URI used to download the body.
request_encoding: The character encoding of the body, as a string.
Returns:
- The OG response as a dictionary.
+ The parsed HTML body, or None if an error occurred during processed.
"""
# If there's no body, nothing useful is going to be found.
if not body:
- return {}
+ return None
from lxml import etree
@@ -650,25 +701,22 @@ def decode_and_calc_og(
parser = etree.HTMLParser(recover=True, encoding="utf-8")
except Exception as e:
logger.warning("Unable to create HTML parser: %s" % (e,))
- return {}
-
- def _attempt_calc_og(body_attempt: Union[bytes, str]) -> Dict[str, Optional[str]]:
- # Attempt to parse the body. If this fails, log and return no metadata.
- tree = etree.fromstring(body_attempt, parser)
-
- # The data was successfully parsed, but no tree was found.
- if tree is None:
- return {}
+ return None
- return _calc_og(tree, media_uri)
+ def _attempt_decode_body(
+ body_attempt: Union[bytes, str]
+ ) -> Optional["etree.Element"]:
+ # Attempt to parse the body. Returns None if the body was successfully
+ # parsed, but no tree was found.
+ return etree.fromstring(body_attempt, parser)
# Attempt to parse the body. If this fails, log and return no metadata.
try:
- return _attempt_calc_og(body)
+ return _attempt_decode_body(body)
except UnicodeDecodeError:
# blindly try decoding the body as utf-8, which seems to fix
# the charset mismatches on https://google.com
- return _attempt_calc_og(body.decode("utf-8", "ignore"))
+ return _attempt_decode_body(body.decode("utf-8", "ignore"))
def _calc_og(tree: "etree.Element", media_uri: str) -> Dict[str, Optional[str]]:
diff --git a/synapse/rest/media/v1/thumbnailer.py b/synapse/rest/media/v1/thumbnailer.py
index df54a406..46701a8b 100644
--- a/synapse/rest/media/v1/thumbnailer.py
+++ b/synapse/rest/media/v1/thumbnailer.py
@@ -61,9 +61,19 @@ class Thumbnailer:
self.transpose_method = None
try:
# We don't use ImageOps.exif_transpose since it crashes with big EXIF
- image_exif = self.image._getexif()
+ #
+ # Ignore safety: Pillow seems to acknowledge that this method is
+ # "private, experimental, but generally widely used". Pillow 6
+ # includes a public getexif() method (no underscore) that we might
+ # consider using instead when we can bump that dependency.
+ #
+ # At the time of writing, Debian buster (currently oldstable)
+ # provides version 5.4.1. It's expected to EOL in mid-2022, see
+ # https://wiki.debian.org/DebianReleases#Production_Releases
+ image_exif = self.image._getexif() # type: ignore
if image_exif is not None:
image_orientation = image_exif.get(EXIF_ORIENTATION_TAG)
+ assert isinstance(image_orientation, int)
self.transpose_method = EXIF_TRANSPOSE_MAPPINGS.get(image_orientation)
except Exception as e:
# A lot of parsing errors can happen when parsing EXIF
@@ -76,7 +86,10 @@ class Thumbnailer:
A tuple containing the new image size in pixels as (width, height).
"""
if self.transpose_method is not None:
- self.image = self.image.transpose(self.transpose_method)
+ # Safety: `transpose` takes an int rather than e.g. an IntEnum.
+ # self.transpose_method is set above to be a value in
+ # EXIF_TRANSPOSE_MAPPINGS, and that only contains correct values.
+ self.image = self.image.transpose(self.transpose_method) # type: ignore[arg-type]
self.width, self.height = self.image.size
self.transpose_method = None
# We don't need EXIF any more
@@ -101,7 +114,7 @@ class Thumbnailer:
else:
return (max_height * self.width) // self.height, max_height
- def _resize(self, width: int, height: int) -> Image:
+ def _resize(self, width: int, height: int) -> Image.Image:
# 1-bit or 8-bit color palette images need converting to RGB
# otherwise they will be scaled using nearest neighbour which
# looks awful.
@@ -151,7 +164,7 @@ class Thumbnailer:
cropped = scaled_image.crop((crop_left, 0, crop_right, height))
return self._encode_image(cropped, output_type)
- def _encode_image(self, output_image: Image, output_type: str) -> BytesIO:
+ def _encode_image(self, output_image: Image.Image, output_type: str) -> BytesIO:
output_bytes_io = BytesIO()
fmt = self.FORMATS[output_type]
if fmt == "JPEG":
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-03 13:39:44 +0000