Improved remote thread fetching (#10106)

* Fetch up to 5 replies when discovering a new remote status

This is used for resolving threads downwards. The originating
server must add a “replies” attributes with such replies for it to
be useful.

* Add some tests for ActivityPub::FetchRepliesWorker

* Add specs for ActivityPub::FetchRepliesService

* Serialize up to 5 public self-replies for ActivityPub notes

* Add specs for ActivityPub::NoteSerializer

* Move exponential backoff logic to a worker concern

* Fetch first page of paginated collections when fetching thread replies

* Add specs for paginated collections in replies

* Move Note replies serialization to a first CollectionPage

The collection isn't actually paginable yet as it has no id nor
a `next` field. This may come in another PR.

* Use pluck(:uri) instead of map(&:uri) to improve performances

* Fix fetching replies when they are in a CollectionPage

1 files changed, 60 insertions, 0 deletions

diff --git a/app/services/activitypub/fetch_replies_service.rb b/app/services/activitypub/fetch_replies_service.rb
new file mode 100644
index 000000000..95c486a43
--- /dev/null
+++ b/app/services/activitypub/fetch_replies_service.rb
@@ -0,0 +1,60 @@
+# frozen_string_literal: true
+
+class ActivityPub::FetchRepliesService < BaseService
+  include JsonLdHelper
+
+  def call(parent_status, collection_or_uri, allow_synchronous_requests = true)
+    @account = parent_status.account
+    @allow_synchronous_requests = allow_synchronous_requests
+
+    @items = collection_items(collection_or_uri)
+    return if @items.nil?
+
+    FetchReplyWorker.push_bulk(filtered_replies)
+
+    @items
+  end
+
+  private
+
+  def collection_items(collection_or_uri)
+    collection = fetch_collection(collection_or_uri)
+    return unless collection.is_a?(Hash)
+
+    collection = fetch_collection(collection['first']) if collection['first'].present?
+    return unless collection.is_a?(Hash)
+
+    case collection['type']
+    when 'Collection', 'CollectionPage'
+      collection['items']
+    when 'OrderedCollection', 'OrderedCollectionPage'
+      collection['orderedItems']
+    end
+  end
+
+  def fetch_collection(collection_or_uri)
+    return collection_or_uri if collection_or_uri.is_a?(Hash)
+    return unless @allow_synchronous_requests
+    return if invalid_origin?(collection_or_uri)
+    collection = fetch_resource_without_id_validation(collection_or_uri)
+    raise Mastodon::UnexpectedResponseError if collection.nil?
+    collection
+  end
+
+  def filtered_replies
+    # Only fetch replies to the same server as the original status to avoid
+    # amplification attacks.
+
+    # Also limit to 5 fetched replies to limit potential for DoS.
+    @items.map { |item| value_or_id(item) }.reject { |uri| invalid_origin?(uri) }.take(5)
+  end
+
+  def invalid_origin?(url)
+    return true if unsupported_uri_scheme?(url)
+
+    needle   = Addressable::URI.parse(url).host
+    haystack = Addressable::URI.parse(@account.uri).host
+
+    !haystack.casecmp(needle).zero?
+  end
+end