mirror of
https://akkoma.dev/AkkomaGang/akkoma.git
synced 2024-11-18 02:49:18 +00:00
3ff9c5e2a6
Fixes cycles in lib/pleroma/ecto_type/activity_pub/object_validators/safe_text.ex
94 lines
2.4 KiB
Elixir
94 lines
2.4 KiB
Elixir
# Pleroma: A lightweight social networking server
|
|
# Copyright © 2017-2021 Pleroma Authors <https://pleroma.social/>
|
|
# SPDX-License-Identifier: AGPL-3.0-only
|
|
|
|
defmodule Pleroma.HTML do
|
|
# Scrubbers are compiled on boot so they can be configured in OTP releases
|
|
# @on_load :compile_scrubbers
|
|
|
|
@cachex Pleroma.Config.get([:cachex, :provider], Cachex)
|
|
|
|
def compile_scrubbers do
|
|
dir = Path.join(:code.priv_dir(:pleroma), "scrubbers")
|
|
|
|
dir
|
|
|> Pleroma.Utils.compile_dir()
|
|
|> case do
|
|
{:error, _errors, _warnings} ->
|
|
raise "Compiling scrubbers failed"
|
|
|
|
{:ok, _modules, _warnings} ->
|
|
:ok
|
|
end
|
|
end
|
|
|
|
defp get_scrubbers(scrubber) when is_atom(scrubber), do: [scrubber]
|
|
defp get_scrubbers(scrubbers) when is_list(scrubbers), do: scrubbers
|
|
defp get_scrubbers(_), do: [Pleroma.HTML.Scrubber.Default]
|
|
|
|
def get_scrubbers do
|
|
Pleroma.Config.get([:markup, :scrub_policy])
|
|
|> get_scrubbers
|
|
end
|
|
|
|
def filter_tags(html, nil) do
|
|
filter_tags(html, get_scrubbers())
|
|
end
|
|
|
|
def filter_tags(html, scrubbers) when is_list(scrubbers) do
|
|
Enum.reduce(scrubbers, html, fn scrubber, html ->
|
|
filter_tags(html, scrubber)
|
|
end)
|
|
end
|
|
|
|
def filter_tags(html, scrubber) do
|
|
{:ok, content} = FastSanitize.Sanitizer.scrub(html, scrubber)
|
|
content
|
|
end
|
|
|
|
def filter_tags(html), do: filter_tags(html, nil)
|
|
def strip_tags(html), do: filter_tags(html, FastSanitize.Sanitizer.StripTags)
|
|
|
|
def ensure_scrubbed_html(
|
|
content,
|
|
scrubbers,
|
|
fake,
|
|
callback
|
|
) do
|
|
content =
|
|
content
|
|
|> filter_tags(scrubbers)
|
|
|> callback.()
|
|
|
|
if fake do
|
|
{:ignore, content}
|
|
else
|
|
{:commit, content}
|
|
end
|
|
end
|
|
|
|
def extract_first_external_url_from_object(%{data: %{"content" => content}} = object)
|
|
when is_binary(content) do
|
|
unless object.data["fake"] do
|
|
key = "URL|#{object.id}"
|
|
|
|
@cachex.fetch!(:scrubber_cache, key, fn _key ->
|
|
{:commit, {:ok, extract_first_external_url(content)}}
|
|
end)
|
|
else
|
|
{:ok, extract_first_external_url(content)}
|
|
end
|
|
end
|
|
|
|
def extract_first_external_url_from_object(_), do: {:error, :no_content}
|
|
|
|
def extract_first_external_url(content) do
|
|
content
|
|
|> Floki.parse_fragment!()
|
|
|> Floki.find("a:not(.mention,.hashtag,.attachment,[rel~=\"tag\"])")
|
|
|> Enum.take(1)
|
|
|> Floki.attribute("href")
|
|
|> Enum.at(0)
|
|
end
|
|
end
|