mjolnir/test/integration/banListTest.ts

597 lines
38 KiB
TypeScript
Raw Permalink Normal View History

2021-11-19 11:44:48 -05:00
import { strict as assert } from "assert";
import { newTestUser } from "./clientHelper";
import { LogService, MatrixClient, Permalinks, UserID } from "matrix-bot-sdk";
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
import PolicyList, { ChangeType } from "../../src/models/PolicyList";
import { ServerAcl } from "../../src/models/ServerAcl";
import { getFirstReaction } from "./commands/commandUtils";
import { getMessagesByUserIn } from "../../src/utils";
import { Mjolnir } from "../../src/Mjolnir";
import { ALL_RULE_TYPES, Recommendation, RULE_SERVER, RULE_USER, SERVER_RULE_TYPES } from "../../src/models/ListRule";
import AccessControlUnit, { Access, EntityAccess } from "../../src/models/AccessControlUnit";
import { randomUUID } from "crypto";
import { MatrixSendClient } from "../../src/MatrixEmitter";
2021-11-19 11:44:48 -05:00
/**
* Create a policy rule in a policy room.
* @param client A matrix client that is logged in
* @param policyRoomId The room id to add the policy to.
* @param policyType The type of policy to add e.g. m.policy.rule.user. (Use RULE_USER though).
* @param entity The entity to ban e.g. @foo:example.org
* @param reason A reason for the rule e.g. 'Wouldn't stop posting spam links'
* @param template The template to use for the policy rule event.
2021-11-19 11:44:48 -05:00
* @returns The event id of the newly created policy rule.
*/
async function createPolicyRule(client: MatrixSendClient, policyRoomId: string, policyType: string, entity: string, reason: string, template = { recommendation: 'm.ban' }, stateKey = `rule:${entity}`) {
return await client.sendStateEvent(policyRoomId, policyType, stateKey, {
2021-11-19 11:44:48 -05:00
entity,
reason,
...template,
2021-11-19 11:44:48 -05:00
});
}
/**
* Remove a policy rule from a list.
* @param client A matrix client that is logged in
* @param policyRoomId The room id to add the policy to.
* @param policyType The type of policy to add e.g. m.policy.rule.user. (Use RULE_USER though).
* @param entity The entity to ban e.g. @foo:example.org
* @param stateKey The key for the rule.
* @returns The event id of the void rule that was created to override the old one.
*/
async function removePolicyRule(client: MatrixSendClient, policyRoomId: string, policyType: string, entity: string, stateKey = `rule:${entity}`) {
return await client.sendStateEvent(policyRoomId, policyType, stateKey, {});
}
describe("Test: Updating the PolicyList", function() {
it("Calculates what has changed correctly.", async function() {
2021-11-19 11:44:48 -05:00
this.timeout(10000);
const mjolnir: Mjolnir = this.mjolnir!
const moderator = await newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" } });
const banListId = await mjolnir.client.createRoom({ invite: [await moderator.getUserId()] });
const banList = new PolicyList(banListId, banListId, mjolnir.client);
await mjolnir.client.setUserPowerLevel(await moderator.getUserId(), banListId, 100);
2021-11-19 11:44:48 -05:00
assert.equal(banList.allRules.length, 0);
2021-11-19 11:44:48 -05:00
// Test adding a new rule
await createPolicyRule(mjolnir.client, banListId, RULE_USER, '@added:localhost:9999', '');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
let { changes } = await banList.updateList();
2021-11-19 11:44:48 -05:00
assert.equal(changes.length, 1, 'There should only be one change');
assert.equal(changes[0].changeType, ChangeType.Added);
assert.equal(changes[0].sender, await mjolnir.client.getUserId());
assert.equal(banList.userRules.length, 1);
assert.equal(banList.allRules.length, 1);
2021-11-19 11:44:48 -05:00
// Test modifiying a rule
let originalEventId = await createPolicyRule(mjolnir.client, banListId, RULE_USER, '@modified:localhost:9999', '');
2021-11-19 11:44:48 -05:00
await banList.updateList();
let modifyingEventId = await createPolicyRule(mjolnir.client, banListId, RULE_USER, '@modified:localhost:9999', 'modified reason');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
2021-11-19 11:44:48 -05:00
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Modified);
assert.equal(changes[0].previousState['event_id'], originalEventId, 'There should be a previous state event for a modified rule');
assert.equal(changes[0].event['event_id'], modifyingEventId);
let modifyingAgainEventId = await createPolicyRule(mjolnir.client, banListId, RULE_USER, '@modified:localhost:9999', 'modified again');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Modified);
assert.equal(changes[0].previousState['event_id'], modifyingEventId, 'There should be a previous state event for a modified rule');
assert.equal(changes[0].event['event_id'], modifyingAgainEventId);
assert.equal(banList.userRules.length, 2, 'There should be two rules, one for @modified:localhost:9999 and one for @added:localhost:9999');
2021-11-19 11:44:48 -05:00
// Test redacting a rule
const redactThis = await createPolicyRule(mjolnir.client, banListId, RULE_USER, '@redacted:localhost:9999', '');
2021-11-19 11:44:48 -05:00
await banList.updateList();
assert.equal(banList.userRules.filter(r => r.entity === '@redacted:localhost:9999').length, 1);
await mjolnir.client.redactEvent(banListId, redactThis);
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
2021-11-19 11:44:48 -05:00
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Removed);
assert.equal(changes[0].event['event_id'], redactThis, 'Should show the new version of the event with redacted content');
assert.equal(Object.keys(changes[0].event['content']).length, 0, 'Should show the new version of the event with redacted content');
assert.notEqual(Object.keys(changes[0].previousState['content']), 0, 'Should have a copy of the unredacted state');
assert.notEqual(changes[0].rule, undefined, 'The previous rule should be present');
assert.equal(banList.userRules.filter(r => r.entity === '@redacted:localhost:9999').length, 0, 'The rule should be removed.');
2021-11-19 11:44:48 -05:00
// Test soft redaction of a rule
const softRedactedEntity = '@softredacted:localhost:9999'
await createPolicyRule(mjolnir.client, banListId, RULE_USER, softRedactedEntity, '');
2021-11-19 11:44:48 -05:00
await banList.updateList();
assert.equal(banList.userRules.filter(r => r.entity === softRedactedEntity).length, 1);
await mjolnir.client.sendStateEvent(banListId, RULE_USER, `rule:${softRedactedEntity}`, {});
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
2021-11-19 11:44:48 -05:00
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Removed);
assert.equal(Object.keys(changes[0].event['content']).length, 0, 'Should show the new version of the event with redacted content');
assert.notEqual(Object.keys(changes[0].previousState['content']), 0, 'Should have a copy of the unredacted state');
assert.notEqual(changes[0].rule, undefined, 'The previous rule should be present');
assert.equal(banList.userRules.filter(r => r.entity === softRedactedEntity).length, 0, 'The rule should have been removed');
2021-11-19 11:44:48 -05:00
// Now test a double soft redaction just to make sure stuff doesn't explode
await mjolnir.client.sendStateEvent(banListId, RULE_USER, `rule:${softRedactedEntity}`, {});
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
2021-11-19 11:44:48 -05:00
assert.equal(changes.length, 0, "It shouldn't detect a double soft redaction as a change, it should be seen as adding an invalid rule.");
assert.equal(banList.userRules.filter(r => r.entity === softRedactedEntity).length, 0, 'The rule should have been removed');
// Test that different (old) rule types will be modelled as the latest event type.
originalEventId = await createPolicyRule(mjolnir.client, banListId, 'org.matrix.mjolnir.rule.user', '@old:localhost:9999', '');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Added);
assert.equal(banList.userRules.filter(r => r.entity === '@old:localhost:9999').length, 1);
modifyingEventId = await createPolicyRule(mjolnir.client, banListId, 'm.room.rule.user', '@old:localhost:9999', 'modified reason');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Modified);
assert.equal(changes[0].event['event_id'], modifyingEventId);
assert.equal(changes[0].previousState['event_id'], originalEventId, 'There should be a previous state event for a modified rule');
assert.equal(banList.userRules.filter(r => r.entity === '@old:localhost:9999').length, 1);
modifyingAgainEventId = await createPolicyRule(mjolnir.client, banListId, RULE_USER, '@old:localhost:9999', 'changes again');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Modified);
assert.equal(changes[0].event['event_id'], modifyingAgainEventId);
assert.equal(changes[0].previousState['event_id'], modifyingEventId, 'There should be a previous state event for a modified rule');
assert.equal(banList.userRules.filter(r => r.entity === '@old:localhost:9999').length, 1);
})
it("Will remove rules with old types when they are 'soft redacted' with a different but more recent event type.", async function() {
this.timeout(3000);
const mjolnir: Mjolnir = this.mjolnir!
const moderator = await newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" }} );
const banListId = await mjolnir.client.createRoom({ invite: [await moderator.getUserId()] });
const banList = new PolicyList(banListId, banListId, mjolnir.client);
await mjolnir.client.setUserPowerLevel(await moderator.getUserId(), banListId, 100);
const entity = '@old:localhost:9999';
let originalEventId = await createPolicyRule(mjolnir.client, banListId, 'm.room.rule.user', entity, '');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
let { changes } = await banList.updateList();
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Added);
assert.equal(banList.userRules.filter(rule => rule.entity === entity).length, 1, 'There should be a rule stored that we just added...')
let softRedactingEventId = await mjolnir.client.sendStateEvent(banListId, RULE_USER, `rule:${entity}`, {});
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Removed);
assert.equal(changes[0].event['event_id'], softRedactingEventId);
assert.equal(changes[0].previousState['event_id'], originalEventId, 'There should be a previous state event for a modified rule');
assert.equal(banList.userRules.filter(rule => rule.entity === entity).length, 0, 'The rule should no longer be stored.');
})
it("A rule of the most recent type won't be deleted when an old rule is deleted for the same entity.", async function() {
const mjolnir: Mjolnir = this.mjolnir!
const moderator = await newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" } });
const banListId = await mjolnir.client.createRoom({ invite: [await moderator.getUserId()] });
const banList = new PolicyList(banListId, banListId, mjolnir.client);
await mjolnir.client.setUserPowerLevel(await moderator.getUserId(), banListId, 100);
const entity = '@old:localhost:9999';
let originalEventId = await createPolicyRule(mjolnir.client, banListId, 'm.room.rule.user', entity, '');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
let { changes } = await banList.updateList();
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Added);
assert.equal(banList.userRules.filter(rule => rule.entity === entity).length, 1, 'There should be a rule stored that we just added...')
let updatedEventId = await createPolicyRule(mjolnir.client, banListId, RULE_USER, entity, '');
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
// If in the future you change this and it fails, it's really subjective whether this constitutes a modification, since the only thing that has changed
// is the rule type. The actual content is identical.
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Modified);
assert.equal(changes[0].event['event_id'], updatedEventId);
assert.equal(changes[0].previousState['event_id'], originalEventId, 'There should be a previous state event for a modified rule');
assert.equal(banList.userRules.filter(rule => rule.entity === entity).length, 1, 'Only the latest version of the rule gets returned.');
// Now we delete the old version of the rule without consequence.
await mjolnir.client.sendStateEvent(banListId, 'm.room.rule.user', `rule:${entity}`, {});
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 0);
assert.equal(banList.userRules.filter(rule => rule.entity === entity).length, 1, 'The rule should still be active.');
// And we can still delete the new version of the rule.
let softRedactingEventId = await mjolnir.client.sendStateEvent(banListId, RULE_USER, `rule:${entity}`, {});
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
changes = (await banList.updateList()).changes;
assert.equal(changes.length, 1);
assert.equal(changes[0].changeType, ChangeType.Removed);
assert.equal(changes[0].event['event_id'], softRedactingEventId);
assert.equal(changes[0].previousState['event_id'], updatedEventId, 'There should be a previous state event for a modified rule');
assert.equal(banList.userRules.filter(rule => rule.entity === entity).length, 0, 'The rule should no longer be stored.');
2021-11-19 11:44:48 -05:00
})
it('Test: PolicyList Supports all entity types.', async function () {
const mjolnir: Mjolnir = this.mjolnir!
const banListId = await mjolnir.client.createRoom();
const banList = new PolicyList(banListId, banListId, mjolnir.client);
for (let i = 0; i < ALL_RULE_TYPES.length; i++) {
await createPolicyRule(mjolnir.client, banListId, ALL_RULE_TYPES[i], `*${i}*`, '');
}
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
let { changes } = await banList.updateList();
assert.equal(changes.length, ALL_RULE_TYPES.length);
assert.equal(banList.allRules.length, ALL_RULE_TYPES.length);
})
2021-11-19 11:44:48 -05:00
});
describe('Test: We will not be able to ban ourselves via ACL.', function() {
it('We do not ban ourselves when we put ourselves into the policy list.', async function() {
const mjolnir: Mjolnir = this.mjolnir
const serverName = new UserID(await mjolnir.client.getUserId()).domain;
const banListId = await mjolnir.client.createRoom();
const banList = new PolicyList(banListId, banListId, mjolnir.client);
const aclUnit = new AccessControlUnit([banList]);
await createPolicyRule(mjolnir.client, banListId, RULE_SERVER, serverName, '');
await createPolicyRule(mjolnir.client, banListId, RULE_SERVER, 'evil.com', '');
await createPolicyRule(mjolnir.client, banListId, RULE_SERVER, '*', '');
// We should still intern the matching rules rule.
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
let { changes } = await banList.updateList();
assert.equal(banList.serverRules.length, 3);
// But when we construct an ACL, we should be safe.
const acl = new ServerAcl(serverName)
changes.forEach(change => acl.denyServer(change.rule.entity));
assert.equal(acl.safeAclContent().deny.length, 1);
assert.equal(acl.literalAclContent().deny.length, 3);
const aclUnitAcl = aclUnit.compileServerAcl(serverName);
assert.equal(aclUnitAcl.literalAclContent().deny.length, 1);
})
})
describe('Test: ACL updates will batch when rules are added in succession.', function() {
it('Will batch ACL updates if we spam rules into a PolicyList', async function() {
const mjolnir: Mjolnir = this.mjolnir!
const serverName: string = new UserID(await mjolnir.client.getUserId()).domain
const moderator = await newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" } });
await moderator.joinRoom(mjolnir.managementRoomId);
const mjolnirId = await mjolnir.client.getUserId();
// Setup some protected rooms so we can check their ACL state later.
const protectedRooms: string[] = [];
for (let i = 0; i < 5; i++) {
const room = await moderator.createRoom({ invite: [mjolnirId] });
await mjolnir.client.joinRoom(room);
await moderator.setUserPowerLevel(mjolnirId, room, 100);
await mjolnir.addProtectedRoom(room);
protectedRooms.push(room);
}
// If a previous test hasn't cleaned up properly, these rooms will be populated by bogus ACLs at this point.
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
await Promise.all(protectedRooms.map(async room => {
// We're going to need timeline pagination I'm afraid.
const roomAcl = await mjolnir.client.getRoomStateEvent(room, "m.room.server_acl", "");
assert.equal(roomAcl?.deny?.length ?? 0, 0, 'There should be no entries in the deny ACL.');
}));
// Flood the watched list with banned servers, which should prompt Mjolnir to update server ACL in protected rooms.
const banListId = await moderator.createRoom({ invite: [mjolnirId] });
await mjolnir.client.joinRoom(banListId);
await mjolnir.policyListManager.watchList(Permalinks.forRoom(banListId));
const acl = new ServerAcl(serverName).denyIpAddresses().allowServer("*");
const evilServerCount = 200;
for (let i = 0; i < evilServerCount; i++) {
const badServer = `${i}.evil.com`;
acl.denyServer(badServer);
await createPolicyRule(moderator, banListId, RULE_SERVER, badServer, `Rule #${i}`);
// Give them a bit of a spread over time.
await new Promise(resolve => setTimeout(resolve, 5));
}
// We do this because it should force us to wait until all the ACL events have been applied.
// Even if that does mean the last few events will not go through batching...
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
// At this point we check that the state within Mjolnir is internally consistent, this is just because debugging the following
// is a pita.
const list: PolicyList = this.mjolnir.policyListManager.lists[0]!;
assert.equal(list.serverRules.length, evilServerCount, `There should be ${evilServerCount} rules in here`);
// Check each of the protected rooms for ACL events and make sure they were batched and are correct.
await Promise.all(protectedRooms.map(async room => {
const roomAcl = await mjolnir.client.getRoomStateEvent(room, "m.room.server_acl", "");
if (!acl.matches(roomAcl)) {
assert.fail(`Room ${room} doesn't have the correct ACL: ${JSON.stringify(roomAcl, null, 2)}`)
}
let aclEventCount = 0;
await getMessagesByUserIn(mjolnir.client, mjolnirId, room, 100, events => {
events.forEach(event => event.type === 'm.room.server_acl' ? aclEventCount += 1 : null);
});
LogService.debug('PolicyListTest', `aclEventCount: ${aclEventCount}`);
// If there's less than two then it means the ACL was updated by this test calling `this.mjolnir!.syncLists()`
// and not the listener that detects changes to ban lists (that we want to test!).
// It used to be 10, but it was too low, 30 seems better for CI.
assert.equal(aclEventCount < 30 && aclEventCount > 2, true, 'We should have sent less than 30 ACL events to each room because they should be batched')
}));
})
})
describe('Test: unbaning entities via the PolicyList.', function() {
afterEach(function() { this.moderator?.stop(); });
it('Will remove rules that have legacy types', async function() {
const mjolnir: Mjolnir = this.mjolnir!
const serverName: string = new UserID(await mjolnir.client.getUserId()).domain
const moderator: MatrixClient = await newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" } });
this.moderator = moderator;
await moderator.joinRoom(mjolnir.managementRoomId);
const mjolnirId = await mjolnir.client.getUserId();
// We'll make 1 protected room to test ACLs in.
const protectedRoom = await moderator.createRoom({ invite: [mjolnirId] });
await mjolnir.client.joinRoom(protectedRoom);
await moderator.setUserPowerLevel(mjolnirId, protectedRoom, 100);
await mjolnir.addProtectedRoom(protectedRoom);
// If a previous test hasn't cleaned up properly, these rooms will be populated by bogus ACLs at this point.
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
// If this is not present, then it means the room isn't being protected, which is really bad.
const roomAcl = await mjolnir.client.getRoomStateEvent(protectedRoom, "m.room.server_acl", "");
assert.equal(roomAcl?.deny?.length ?? 0, 0, 'There should be no entries in the deny ACL.');
// Create some legacy rules on a PolicyList.
const banListId = await moderator.createRoom({ invite: [mjolnirId] });
await moderator.setUserPowerLevel(await mjolnir.client.getUserId(), banListId, 100);
await moderator.sendStateEvent(banListId, 'org.matrix.mjolnir.shortcode', '', { shortcode: "unban-test" });
await mjolnir.client.joinRoom(banListId);
await mjolnir.policyListManager.watchList(Permalinks.forRoom(banListId));
// we use this to compare changes.
const banList = new PolicyList(banListId, banListId, moderator);
// we need two because we need to test the case where an entity has all rule types in the list
// and another one that only has one (so that we would hit 404 while looking up state)
const olderBadServer = "old.evil.example"
const newerBadServer = "new.evil.example"
await Promise.all(SERVER_RULE_TYPES.map(type => createPolicyRule(moderator, banListId, type, olderBadServer, 'gregg rulz ok')));
await createPolicyRule(moderator, banListId, RULE_SERVER, newerBadServer, 'this is bad sort it out.');
await createPolicyRule(moderator, banListId, RULE_SERVER, newerBadServer, 'hidden with a non-standard state key', undefined, "rule_1");
// Wait for the ACL event to be applied to our protected room.
await mjolnir.protectedRoomsTracker.syncLists();
await banList.updateList();
// rules are normalized by rule type, that's why there should only be 3.
assert.equal(banList.allRules.length, 3);
// Check that we have setup our test properly and therefore evil.example is banned.
const acl = new ServerAcl(serverName).denyIpAddresses().allowServer("*").denyServer(olderBadServer).denyServer(newerBadServer);
const protectedAcl = await mjolnir.client.getRoomStateEvent(protectedRoom, "m.room.server_acl", "");
if (!acl.matches(protectedAcl)) {
assert.fail(`Room ${protectedRoom} doesn't have the correct ACL: ${JSON.stringify(roomAcl, null, 2)}`);
}
// Now unban the servers, we will go via the unban command for completeness sake.
try {
await moderator.start();
for (const server of [olderBadServer, newerBadServer]) {
await getFirstReaction(moderator, mjolnir.managementRoomId, '✅', async () => {
return await moderator.sendMessage(mjolnir.managementRoomId, { msgtype: 'm.text', body: `!mjolnir unban unban-test server ${server}` });
});
}
} finally {
moderator.stop();
}
// Wait for mjolnir to sync protected rooms to update ACL.
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
// Confirm that the server is unbanned.
await banList.updateList();
assert.equal(banList.allRules.length, 0);
const aclAfter = await mjolnir.client.getRoomStateEvent(protectedRoom, "m.room.server_acl", "");
assert.equal(aclAfter.deny.length, 0, 'Should be no servers denied anymore');
})
})
describe('Test: should apply bans to the most recently active rooms first', function() {
it('Applies bans to the most recently active rooms first', async function() {
this.timeout(180000)
const mjolnir: Mjolnir = this.mjolnir!
const serverName: string = new UserID(await mjolnir.client.getUserId()).domain
const moderator = await newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" } });
await moderator.joinRoom(mjolnir.managementRoomId);
const mjolnirId = await mjolnir.client.getUserId();
// Setup some protected rooms so we can check their ACL state later.
const protectedRooms: string[] = [];
for (let i = 0; i < 10; i++) {
const room = await moderator.createRoom({ invite: [mjolnirId] });
await mjolnir.client.joinRoom(room);
await moderator.setUserPowerLevel(mjolnirId, room, 100);
await mjolnir.addProtectedRoom(room);
protectedRooms.push(room);
}
// If a previous test hasn't cleaned up properly, these rooms will be populated by bogus ACLs at this point.
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
await Promise.all(protectedRooms.map(async room => {
const roomAcl = await mjolnir.client.getRoomStateEvent(room, "m.room.server_acl", "").catch(e => e.statusCode === 404 ? { deny: [] } : Promise.reject(e));
assert.equal(roomAcl?.deny?.length ?? 0, 0, 'There should be no entries in the deny ACL.');
}));
// Flood the watched list with banned servers, which should prompt Mjolnir to update server ACL in protected rooms.
const banListId = await moderator.createRoom({ invite: [mjolnirId] });
await mjolnir.client.joinRoom(banListId);
await mjolnir.policyListManager.watchList(Permalinks.forRoom(banListId));
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
// shuffle protected rooms https://stackoverflow.com/a/12646864, we do this so we can create activity "randomly" in them.
for (let i = protectedRooms.length - 1; i > 0; i--) {
const j = Math.floor(Math.random() * (i + 1));
[protectedRooms[i], protectedRooms[j]] = [protectedRooms[j], protectedRooms[i]];
}
// create some activity in the same order.
for (const roomId of protectedRooms.slice().reverse()) {
await moderator.sendMessage(roomId, { body: `activity`, msgtype: 'm.text' });
await new Promise(resolve => setTimeout(resolve, 100));
}
// check the rooms are in the expected order
for (let i = 0; i < protectedRooms.length; i++) {
assert.equal(mjolnir.protectedRoomsTracker.protectedRoomsByActivity()[i], protectedRooms[i]);
}
// just ban one server
const badServer = `evil.com`;
const acl = new ServerAcl(serverName).denyIpAddresses().allowServer("*").denyServer(badServer);
// collect all the rooms that received an ACL event.
const aclRooms: any[] = await new Promise(async resolve => {
const rooms: any[] = [];
this.mjolnir.client.on('room.event', (room: string, event: any) => {
if (protectedRooms.includes(room)) {
rooms.push(room);
}
if (rooms.length === protectedRooms.length) {
resolve(rooms)
}
});
// create the rule that will ban the server.
await createPolicyRule(moderator, banListId, RULE_SERVER, badServer, `Rule ${badServer}`);
})
// Wait until all the ACL events have been applied.
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
await mjolnir.protectedRoomsTracker.syncLists();
for (let i = 0; i < protectedRooms.length; i++) {
assert.equal(aclRooms[i], protectedRooms[i], "The ACL should have been applied to the active rooms first.");
}
// Check that the most recently active rooms got the ACL update first.
let last_event_ts = 0;
for (const roomId of protectedRooms) {
let roomAclEvent: null | any;
// Can't be the best way to get the whole event, but ok.
await getMessagesByUserIn(mjolnir.client, mjolnirId, roomId, 1, events => roomAclEvent = events[0]);
const roomAcl = roomAclEvent!.content;
if (!acl.matches(roomAcl)) {
assert.fail(`Room ${roomId} doesn't have the correct ACL: ${JSON.stringify(roomAcl, null, 2)}`)
}
assert.equal(roomAclEvent.origin_server_ts > last_event_ts, true, `This room was more recently active so should have the more recent timestamp`);
last_event_ts = roomAclEvent.origin_server_ts;
}
})
})
/**
* Assert that the AccessUnitOutcome entity test has the right access.
* @param expected The Access we expect the entity to have, See Access.
* @param query The result of a test on AccessControlUnit e.g. `unit.getAccessForUser'@meow:localhost')`
* @param message A message for the console if the entity doesn't have the expected access.
*/
function assertAccess(expected: Access, query: EntityAccess, message?: string) {
assert.equal(query.outcome, expected, message);
}
describe('Test: AccessControlUnit interaction with policy lists.', function() {
it('The AccessControlUnit correctly reflects the policies that have been set in its watched lists.', async function() {
const mjolnir: Mjolnir = this.mjolnir!
const policyListId = await mjolnir.client.createRoom();
const policyList = new PolicyList(policyListId, Permalinks.forRoom(policyListId), mjolnir.client);
const aclUnit = new AccessControlUnit([policyList]);
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@anyone:anywhere.example.com', "CHECK_SERVER"), 'Empty lists should implicitly allow.');
await createPolicyRule(mjolnir.client, policyListId, RULE_SERVER, 'matrix.org', '', { recommendation: Recommendation.Allow });
// we want to imagine that the banned server was never taken off the allow after being banned.
await createPolicyRule(mjolnir.client, policyListId, RULE_SERVER, 'bad.example.com', '', { recommendation: Recommendation.Allow }, 'something-else');
await createPolicyRule(mjolnir.client, policyListId, RULE_SERVER, 'bad.example.com', '', { recommendation: Recommendation.Ban });
await createPolicyRule(mjolnir.client, policyListId, RULE_SERVER, '*.ddns.example.com', '', { recommendation: Recommendation.Ban });
await policyList.updateList();
assertAccess(Access.Allowed, aclUnit.getAccessForServer('matrix.org'));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@someone:matrix.org', "CHECK_SERVER"));
assertAccess(Access.NotAllowed, aclUnit.getAccessForServer('anywhere.else.example.com'));
assertAccess(Access.NotAllowed, aclUnit.getAccessForUser('@anyone:anywhere.else.example.com', "CHECK_SERVER"));
assertAccess(Access.Banned, aclUnit.getAccessForServer('bad.example.com'));
assertAccess(Access.Banned, aclUnit.getAccessForUser('@anyone:bad.example.com', "CHECK_SERVER"));
// They're not allowed in the first place, never mind that they are also banned.
assertAccess(Access.NotAllowed, aclUnit.getAccessForServer('meow.ddns.example.com'));
assertAccess(Access.NotAllowed, aclUnit.getAccessForUser('@anyone:meow.ddns.example.com', "CHECK_SERVER"));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@spam:matrix.org', "CHECK_SERVER"));
await createPolicyRule(mjolnir.client, policyListId, RULE_USER, '@spam:matrix.org', '', { recommendation: Recommendation.Ban });
await policyList.updateList();
assertAccess(Access.Banned, aclUnit.getAccessForUser('@spam:matrix.org', "CHECK_SERVER"));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@someone:matrix.org', "CHECK_SERVER"));
// protect a room and check that only bad.example.com, *.ddns.example.com are in the deny ACL and not matrix.org
await mjolnir.policyListManager.watchList(policyList.roomRef);
const protectedRoom = await mjolnir.client.createRoom();
await mjolnir.protectedRoomsTracker.addProtectedRoom(protectedRoom);
await mjolnir.protectedRoomsTracker.syncLists();
const roomAcl = await mjolnir.client.getRoomStateEvent(protectedRoom, "m.room.server_acl", "");
assert.equal(roomAcl?.deny?.length ?? 0, 2, 'There should be two entries in the deny ACL.');
for (const serverGlob of ["*.ddns.example.com", "bad.example.com"]) {
assert.equal((roomAcl?.deny ?? []).includes(serverGlob), true);
}
assert.equal(roomAcl.deny.includes("matrix.org"), false);
assert.equal(roomAcl.allow.includes("matrix.org"), true);
// Now we remove the rules and hope that everything functions noramally.
await removePolicyRule(mjolnir.client, policyListId, RULE_SERVER, 'matrix.org');
await removePolicyRule(mjolnir.client, policyListId, RULE_SERVER, 'bad.example.com', 'something-else');
await removePolicyRule(mjolnir.client, policyListId, RULE_SERVER, 'bad.example.com');
await removePolicyRule(mjolnir.client, policyListId, RULE_SERVER, '*.ddns.example.com');
await removePolicyRule(mjolnir.client, policyListId, RULE_USER, "@spam:matrix.org");
A look at `PolicyList.update` (#454) This started out as just a way to find out why mjolnir was syncing with lists several times for each update to a policy list. The main changes are - Verbosity was irrelevant to the sync command but for some reason was an option. Unfortunately all this did was suppress whether to tell you when it had finished, meaning it wouldn't when verbose logging was disabled. Historically this was probably a parameter that got passed through to applyServerAcl/applyUserBans, which can be horribly verbose, but they access the config directly. - Stop emitting `'PolicyList.update'` when there are no changes. - Include a revision ID for the `'PolicyList.update'`method and event. - Use the revision ID in the `ProtectedRoomsSet` so that we don't unnecessarily resynchronize all rooms when the `'PolicyList.update'` event is received. Though not when the `sync` command is used. Since this is supposed to `sync` in the case when there is a state reset or otherwise or the user has changed some room settings. - insert an await lock around the `PolicyList.update` method to avoid a race condition where a call can be started and finished within the extent of an existing call (via another task, this can happen if the server is slow with handling one request). `PolicyList.udpate` now has a helper that is synchronous to be called directly after requesting the room state. The reason for this is to enforce that no one `await`s while updating the policy list's cache of rules. Which is important because it is one of the biggest methods that I tolerate and visually checking for `await` is impossible. - The revision ID uses a ULID, but this is unnecessary and could have just been a "dumb counter". closes https://github.com/matrix-org/mjolnir/issues/447
2022-12-08 11:09:55 -05:00
const { changes } = await policyList.updateList()
await mjolnir.protectedRoomsTracker.syncLists();
assert.equal(changes.length, 5, "The rules should have correctly been removed");
assertAccess(Access.Allowed, aclUnit.getAccessForServer('matrix.org'));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@someone:matrix.org', "CHECK_SERVER"));
assertAccess(Access.Allowed, aclUnit.getAccessForServer('anywhere.else.example.com'));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@anyone:anywhere.else.example.com', "CHECK_SERVER"));
assertAccess(Access.Allowed, aclUnit.getAccessForServer('bad.example.com'));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@anyone:bad.example.com', "CHECK_SERVER"));
assertAccess(Access.Allowed, aclUnit.getAccessForServer('meow.ddns.example.com'));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@anyone:meow.ddns.example.com', "CHECK_SERVER"));
assertAccess(Access.Allowed, aclUnit.getAccessForUser('@spam:matrix.org', "CHECK_SERVER"));
const roomAclAfter = await mjolnir.client.getRoomStateEvent(protectedRoom, "m.room.server_acl", "");
assert.equal(roomAclAfter.deny?.length ?? 0, 0, 'There should be no entries in the deny ACL.');
assert.equal(roomAclAfter.allow?.length ?? 0, 1, 'There should be 1 entry in the allow ACL.');
assert.equal(roomAclAfter.allow.includes("*"), true);
})
it('removing a rule from a different list will not clobber anything.', async function() {
const mjolnir: Mjolnir = this.mjolnir!
const policyLists = await Promise.all([...Array(2).keys()].map(async _ => {
const policyListId = await mjolnir.client.createRoom();
return new PolicyList(policyListId, Permalinks.forRoom(policyListId), mjolnir.client);
}));
const banMeServer = 'banme.example.com';
const aclUnit = new AccessControlUnit(policyLists);
await Promise.all(policyLists.map(policyList => {
return createPolicyRule(mjolnir.client, policyList.roomId, RULE_SERVER, banMeServer, '', { recommendation: Recommendation.Ban })
}));
await Promise.all(policyLists.map(list => list.updateList()));
assertAccess(Access.Banned, aclUnit.getAccessForServer(banMeServer));
// remove the rule that bans `banme.example.com` from just one of the lists.
await removePolicyRule(mjolnir.client, policyLists[0].roomId, RULE_SERVER, banMeServer);
await Promise.all(policyLists.map(list => list.updateList()));
assertAccess(Access.Banned, aclUnit.getAccessForServer(banMeServer), "Should still be banned at this point.");
await removePolicyRule(mjolnir.client, policyLists[1].roomId, RULE_SERVER, banMeServer);
await Promise.all(policyLists.map(list => list.updateList()));
assertAccess(Access.Allowed, aclUnit.getAccessForServer(banMeServer), "Should not longer be any rules");
})
})
describe('Test: Creating policy lists.', function() {
it('Will automatically invite and op users from invites', async function() {
const mjolnir: Mjolnir = this.mjolnir;
const testUsers = await Promise.all([...Array(2)].map(_ => newTestUser(this.config.homeserverUrl, { name: { contains: "moderator" } })))
const invite = await Promise.all(testUsers.map(client => client.getUserId()));
const policyListId = await PolicyList.createList(
mjolnir.client,
randomUUID(),
invite
);
// Check power levels are right.
const powerLevelEvent = await mjolnir.client.getRoomStateEvent(policyListId, "m.room.power_levels", "");
assert.equal(Object.keys(powerLevelEvent.users ?? {}).length, invite.length + 1);
// Check create event for MSC3784 support.
const createEvent = await mjolnir.client.getRoomStateEvent(policyListId, "m.room.create", "");
assert.equal(createEvent.type, PolicyList.ROOM_TYPE);
// We can't create rooms without forgetting the type.
await assert.rejects(
async () => {
await PolicyList.createList(mjolnir.client, randomUUID(), [], {
creation_content: {}
})
},
TypeError
);
})
})