17 KiB

Raw Blame History

FediBlockHole

A tool for keeping a Mastodon instance blocklist synchronised with remote lists.

The broad design goal for FediBlockHole is to support pulling in a list of blocklists from a set of trusted sources, merge them into a combined blocklist, and then push that merged list to a set of managed instances.

Mastodon admins can choose who they think maintain quality lists and subscribe to them, helping to distribute the load for maintaining blocklists among a community of people. Control ultimately rests with the admins themselves so they can outsource as much, or as little, of the effort to others as they deem appropriate.

Inspired by the way PiHole works for maintaining a set of blocklists of adtech domains. Builds on the work of @CaribenxMarciaX@scholar.social and @gingerrroot@kitty.town who started the #Fediblock hashtag and did a lot of advocacy around it, often at great personal cost.

Features

Blocklist Sources

Read domain block lists from other instances via the Mastodon API.
Supports both public lists (no auth required) and 'admin' lists requiring authentication to an instance.
Read domain block lists from arbitrary URLs, including local files.
Supports CSV and JSON format blocklists
Supports RapidBlock CSV and JSON format blocklists

Blocklist Export/Push

Push a merged blocklist to a set of Mastodon instances.
Export per-source, unmerged block lists to local files, in CSV format.
Export merged blocklists to local files, in CSV format.
Read block lists from multiple remote instances
Read block lists from multiple URLs, including local files
Write a unified block list to a local CSV file
Push unified blocklist updates to multiple remote instances
Control import and export fields

Flexible Configuration

Provides (hopefully) sensible defaults to minimise first-time setup.
Global and fine-grained configuration options available for those complex situations that crop up sometimes.
Allowlists to override blocks in blocklists to ensure you never block instances you want to keep.
Blocklist thresholds if you want to only block when an instance shows up in multiple blocklists.

Installing

Installable using pip.

python3 -m pip install fediblockhole

Install from source by cloning the repo, cd fediblockhole and run:

python3 -m pip install .

Installation adds a commandline tool: fediblock-sync

Instance admins who want to use this tool for their instance will need to add an Application at https://<instance-domain>/settings/applications/ so they can authorize the tool to create and update domain blocks with an OAuth token.

Reading remote instance blocklists

If a remote instance makes its domain blocks public, you don't need a token to read them.

If a remote instance only shows its domain blocks to local accounts you'll need to have a token with read:blocks authorization set up. If you have an account on that instance, you can get a token by setting up a new Application at https://<instance-domain>/settings/applications/.

To read admin blocks from a remote instance, you'll need to ask the instance admin to add a new Application at https://<instance-domain>/settings/applications/ and then tell you the access token.

The application needs the admin:read:domain_blocks OAuth scope. You can allow full admin:read access, but be aware that this authorizes someone to read all the data in the instance. That's asking a lot of a remote instance admin who just wants to share domain_blocks with you.

The admin:read:domain_blocks scope is available as of Mastodon v4.1.0, but for earlier versions admins will need to use the manual method described below.

You can update the scope for your application in the database directly like this:

UPDATE oauth_applications as app
  SET scopes = 'admin:read:domain_blocks'
  FROM oauth_access_tokens as tok
  WHERE app.id = tok.application_id
  AND app.name = '<the_app_name>'
;

When that's done, regenerate the token (so it has the new scopes) in the application screen in the instance GUI. FediBlockHole should then able to use the app token to read domain blocks via the API, but nothing else.

Alternately, you could ask the remote instance admin to set up FediBlockHole and use it to dump out a CSV blocklist from their instance and then put it somewhere trusted parties can read it. Then you can define the blocklist as a URL source, as explained below.

Writing instance blocklists

To write domain blocks into an instance requires both the admin:read and admin:write:domain_blocks OAuth scopes.

The tool needs admin:read:domain_blocks scope to read the current list of domain blocks so we update ones that already exist, rather than trying to add all new ones and clutter up the instance.

admin:read access is needed to check if the instance has any accounts that follow accounts on a domain that is about to get suspended and automatically drop the block severity to silence level so people have time to migrate accounts before a full defederation takes effect. Unfortunately, the statistics measure used to learn this information requires admin:read scope.

You can add admin:read scope in the application admin screen. Please be aware that this grants full read access to all information in the instance to the application token, so make sure you keep it a secret. At least remove world-readable permission to any config file you put it in, e.g.:

chmod o-r <configfile>

You can also grant full admin:write scope to the application, but if you'd prefer to keep things more tightly secured, limit the scope to admin:read:domain_blocks.

Again, this scope is only available in the application config screen as of Mastodon v4.1.0. If your instance is on an earlier version, you'll need to use SQL to set the scopes in the database and then regenerate the token:

UPDATE oauth_applications as app
  SET scopes = 'admin:read admin:write:domain_blocks'
  FROM oauth_access_tokens as tok
  WHERE app.id = tok.application_id
  AND app.name = '<the_app_name>'
;

When that's done, FediBlockHole should be able to use its token to authorise adding or updating domain blocks via the API.

Using the tool

Run the tool like this:

fediblock-sync -c <configfile_path>

If you put the config file in /etc/default/fediblockhole.conf.toml you don't need to pass in the config file path.

For a list of possible configuration options, check the --help.

You can also read the heavily commented sample configuration file in the repo at etc/sample.fediblockhole.conf.toml.

Configuring

Once you have your applications and tokens and scopes set up, create a configuration file for FediBlockHole to use. You can put it anywhere and use the -c <configfile> commandline parameter to tell FediBlockHole where it is.

Or you can use the default location of /etc/default/fediblockhole.conf.toml.

As the filename suggests, FediBlockHole uses TOML syntax.

There are 4 key sections:

blocklist_urls_sources: A list of URLs to read blocklists from
blocklist_instance_sources: A list of Mastodon instances to read blocklists from via API
blocklist_instance_destinations: A list of Mastodon instances to write blocklists to via API
allowlist_url_sources: A list of URLs to read allowlists from

More detail on configuring the tool is provided below.

URL sources

The URL sources is a list of URLs to fetch blocklists from.

Supported formats are currently:

Comma-Separated Values (CSV)
JSON
Mastodon v4.1 flavoured CSV
RapidBlock CSV
RapidBlock JSON

Blocklists must provide a domain field, and should provide a severity field.

domain is the domain name of the instance to be blocked/limited.

severity is the severity level of the block/limit. Supported values are: noop, silence, and suspend.

Optional fields that the tool understands are public_comment, private_comment, reject_media, reject_reports, and obfuscate.

CSV format

A CSV format blocklist must contain a header row with at least a domain and severity field.

Optional fields, as listed about, may also be included.

Mastodon v4.1 CSV format

As of v4.1.0, Mastodon can export domain blocks as a CSV file. However, in their infinite wisdom, the Mastodon devs decided that field names should begin with a # character in the header, unlike the field names in the JSON output via the API… or in pretty much any other CSV file anywhere else.

Setting the format to mastodon_csv will strip off the # character when parsing and FediBlockHole can then use Mastodon v4.1 CSV blocklists like any other CSV formatted blocklist.

JSON format

JSON is also supported. It uses the same format as the JSON returned from the Mastodon API.

This is a list of dictionaries, with at minimum a domain field, and preferably a severity field. The other optional fields are, well, optional.

RapidBlock CSV format

The RapidBlock CSV format has no header and a single field, so it's not strictly a CSV file as there are no commas separating values. It is basically just a list of domains to block, separated by '\r\n'.

When using this format, the tool assumes the severity level is suspend.

RapidBlock JSON format

The RapidBlock JSON format provides more detailed information about domain blocks, but is still somewhat limited.

It has a single isBlocked flag indicating if a domain should be blocked or not. There is no support for the 'silence' block level.

There is no support for 'reject_media' or 'reject_reports' or 'obfuscate'.

All comments are public, by virtue of the public nature of RapidBlock.

Instance sources

The tool can also read domain_blocks from instances directly.

The configuration is a list of dictionaries of the form:

{ domain = '<domain_name>', token = '<BearerToken>', admin = false }

The domain is the fully-qualified domain name of the API host for an instance you want to read domain blocks from.

The token is an optional OAuth token for the application that's configured in the instance to allow you to read domain blocks, as discussed above.

admin is an optional field that tells the tool to use the more detailed admin API endpoint for domain_blocks, rather than the more public API endpoint that doesn't provide as much detail. You will need a token that's been configured to permit access to the admin domain_blocks scope, as detailed above.

Instance destinations

The tool supports pushing a unified blocklist to multiple instances.

Configure the list of instances you want to push your blocklist to in the blocklist_instance_detinations list. Each entry is of the form:

{ domain = '<domain_name>', token = '<BearerToken>', import_fields = ['public_comment'], max_severity = 'suspend', max_followed_severity = 'suspend' }

The fields domain and token are required.

The fields max_followed_severity and import_fields are optional.

The domain is the hostname of the instance you want to push to. The token is an application token with both admin:read:domain_blocks and admin:write:domain_blocks authorization.

The optional import_fields setting allows you to restrict which fields are imported from each instance. If you want to import the reject_reports settings from one instance, but no others, you can use the import_fields setting to do it. Note: The domain and severity fields are always imported.

The optional max_severity setting limits the maximum severity you will allow a remote blocklist to set. This helps you import a list from a remote instance but only at the silence level, even if that remote instance has a block at suspend level. If not set, defaults to suspend.

The optional max_followed_severity setting sets a per-instance limit on the severity of a domain_block if there are accounts on the instance that follow accounts on the domain to be blocked. If max_followed_severity isn't set, it defaults to silence.

This setting exists to give people time to move off an instance that is about to be defederated and bring their followers from your instance with them. Without it, if a new suspend block appears in any of the blocklists you subscribe to (or a block level increases from silence to suspend) and you're using the default max mergeplan, the tool would immediately suspend the instance, cutting everyone on the blocked instance off from their existing followers on your instance, even if they move to a new instance. If you actually want that outcome, you can set max_followed_severity = 'suspend' and use the max mergeplan.

Once the follow count drops to 0 on your instance, the tool will automatically use the highest severity it finds again (if you're using the max mergeplan).

Allowlists

Sometimes you might want to completely ignore the blocklist definitions for certain domains. That's what allowlists are for.

Allowlists remove any domain in the list from the merged list of blocks before the merged list is saved out to a file or pushed to any instance.

Allowlists can be in any format supported by blocklist_urls_sources but ignore all fields that aren't domain.

You can also allow domains on the commandline by using the -A or --allow flag and providing the domain name to allow. You can use the flag multiple times to allow multiple domains.

It is probably wise to include your own instance domain in an allowlist so you don't accidentally defederate from yourself.

More advanced configuration

For a list of possible configuration options, check the --help and read the sample configuration file in etc/sample.fediblockhole.conf.toml.

save_intermediate

This option tells the tool to save the unmerged blocklists it fetches from remote instances and URLs into separate files. This is handy for debugging, or just to have a non-unified set of blocklist files.

Works with the savedir setting to control where to save the files.

These are parsed blocklists, not the raw data, and so will be affected by import_fields.

The filename is based on the URL or domain used so you can tell where each list came from.

savedir

Sets where to save intermediate blocklist files. Defaults to /tmp.

blocklist_auditfile

If provided, will save an audit file of counts and percentages by domain. Useful for debugging thresholds. Defaults to None.

no_push_instance

Defaults to False.

When set, the tool won't actually try to push the unified blocklist to any configured instances.

If you want to see what the tool would try to do, but not actually apply any updates, use --dryrun.

no_fetch_url

Skip the fetching of blocklists from any URLs that are configured.

no_fetch_instance

Skip the fetching of blocklists from any remote instances that are configured.

override_private_comment

Defaults to None.

Stamp all new blocks pushed to a remote server with this comment or code. Helps to identify blocks you've created on a server via Fediblockhole versus ones that already existed.

mergeplan

If two (or more) blocklists define blocks for the same domain, but they're different, mergeplan tells the tool how to resolve the conflict.

max is the default. It uses the highest severity block it finds as the one that should be used in the unified blocklist.

min does the opposite. It uses the lowest severity block it finds as the one to use in the unified blocklist.

A full discussion of severities is beyond the scope of this README, but here is a quick overview of how it works for this tool.

The severities are:

noop, level 0: This is essentially an 'unblock' but you can include a comment.
silence, level 1: A silence adds friction to federation with an instance.
suspend, level 2: A full defederation with the instance.

With mergeplan set to max, silence would take precedence over noop, and suspend would take precedence over both.

With mergeplan set to min, silence would take precedence over suspend, and noop would take precedence over both.

You would want to use max to ensure that you always block with whichever your harshest fellow admin thinks should happen.

You would want to use min to ensure that your blocks do what your most lenient fellow admin thinks should happen.

import_fields

import_fields controls which fields will be imported from remote instances and URL blocklists, and which fields are pushed to instances from the unified blocklist.

The fields domain and severity are always included, so only define extra fields, if you want them.

You can't export fields you haven't imported, so export_fields should be a subset of import_fields, but you can run the tool multiple times. You could, for example, include lots of fields for an initial import to build up a comprehensive list for export, combined with the --no-push-instances option so you don't actually apply the full list to anywhere.

Then you could use a different set of options when importing so you have all the detail in a file, but only push public_comment to instances.

export_fields

export_fields controls which fields will get saved to the unified blocklist file, if you export one.