discourse/spec/lib/middleware/request_tracker_spec.rb

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

765 lines
24 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
RSpec.describe Middleware::RequestTracker do
def env(opts = {})
path = opts.delete(:path) || "/path?bla=1"
create_request_env(path: path).merge(
"HTTP_HOST" => "http://test.com",
"HTTP_USER_AGENT" =>
"Mozilla/5.0 (Windows NT 6.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/41.0.2228.0 Safari/537.36",
"REQUEST_METHOD" => "GET",
"HTTP_ACCEPT" =>
"text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,image/apng,*/*;q=0.8",
"rack.input" => StringIO.new,
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
).merge(opts)
end
before do
ApplicationRequest.enable
CachedCounting.reset
CachedCounting.enable
end
after do
ApplicationRequest.disable
CachedCounting.disable
end
describe "full request" do
it "can handle rogue user agents" do
agent = (+"Evil Googlebot String \xc3\x28").force_encoding("Windows-1252")
middleware =
Middleware::RequestTracker.new(->(env) { ["200", { "Content-Type" => "text/html" }, [""]] })
middleware.call(env("HTTP_USER_AGENT" => agent))
CachedCounting.flush
expect(WebCrawlerRequest.where(user_agent: agent.encode("utf-8")).count).to eq(1)
end
end
describe "log_request" do
before do
freeze_time
ApplicationRequest.clear_cache!
end
def log_tracked_view(val)
data =
Middleware::RequestTracker.get_data(
env("HTTP_DISCOURSE_TRACK_VIEW" => val),
["200", { "Content-Type" => "text/html" }],
0.2,
)
Middleware::RequestTracker.log_request(data)
end
it "can exclude/include based on custom header" do
log_tracked_view("true")
log_tracked_view("1")
log_tracked_view("false")
log_tracked_view("0")
CachedCounting.flush
expect(ApplicationRequest.page_view_anon.first.count).to eq(2)
end
it "can log requests correctly" do
data =
Middleware::RequestTracker.get_data(
env("HTTP_USER_AGENT" => "AdsBot-Google (+http://www.google.com/adsbot.html)"),
["200", { "Content-Type" => "text/html" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
data =
Middleware::RequestTracker.get_data(
env("HTTP_DISCOURSE_TRACK_VIEW" => "1"),
["200", {}],
0.1,
)
Middleware::RequestTracker.log_request(data)
data =
Middleware::RequestTracker.get_data(
env(
"HTTP_USER_AGENT" =>
"Mozilla/5.0 (iPhone; CPU iPhone OS 8_1 like Mac OS X) AppleWebKit/600.1.4 (KHTML, like Gecko) Version/8.0 Mobile/12B410 Safari/600.1.4",
),
["200", { "Content-Type" => "text/html" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
# /srv/status is never a tracked view because content-type is text/plain
data =
Middleware::RequestTracker.get_data(
env("HTTP_USER_AGENT" => "kube-probe/1.18", "REQUEST_URI" => "/srv/status?shutdown_ok=1"),
["200", { "Content-Type" => "text/plain" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
CachedCounting.flush
expect(ApplicationRequest.http_total.first.count).to eq(4)
expect(ApplicationRequest.http_2xx.first.count).to eq(4)
expect(ApplicationRequest.page_view_anon.first.count).to eq(2)
expect(ApplicationRequest.page_view_crawler.first.count).to eq(1)
expect(ApplicationRequest.page_view_anon_mobile.first.count).to eq(1)
expect(ApplicationRequest.page_view_crawler.first.count).to eq(1)
end
it "logs API requests correctly" do
data =
Middleware::RequestTracker.get_data(
env("_DISCOURSE_API" => "1"),
["200", { "Content-Type" => "text/json" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
data =
Middleware::RequestTracker.get_data(
env("_DISCOURSE_API" => "1"),
["404", { "Content-Type" => "text/json" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
data =
Middleware::RequestTracker.get_data(env("_DISCOURSE_USER_API" => "1"), ["200", {}], 0.1)
Middleware::RequestTracker.log_request(data)
CachedCounting.flush
expect(ApplicationRequest.http_total.first.count).to eq(3)
expect(ApplicationRequest.http_2xx.first.count).to eq(2)
expect(ApplicationRequest.api.first.count).to eq(2)
expect(ApplicationRequest.user_api.first.count).to eq(1)
end
it "can log Discourse user agent requests correctly" do
# log discourse api agents as crawlers for page view stats...
data =
Middleware::RequestTracker.get_data(
env("HTTP_USER_AGENT" => "DiscourseAPI Ruby Gem 0.19.0"),
["200", { "Content-Type" => "text/html" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
CachedCounting.flush
CachedCounting.reset
expect(ApplicationRequest.page_view_crawler.first.count).to eq(1)
# ...but count our mobile app user agents as regular visits
data =
Middleware::RequestTracker.get_data(
env("HTTP_USER_AGENT" => "Mozilla/5.0 AppleWebKit/605.1.15 Mobile/15E148 DiscourseHub)"),
["200", { "Content-Type" => "text/html" }],
0.1,
)
Middleware::RequestTracker.log_request(data)
CachedCounting.flush
expect(ApplicationRequest.page_view_crawler.first.count).to eq(1)
expect(ApplicationRequest.page_view_anon.first.count).to eq(1)
end
context "when ignoring anonymous page views" do
let(:anon_data) do
Middleware::RequestTracker.get_data(
env(
"HTTP_USER_AGENT" =>
"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/90.0.4430.72 Safari/537.36",
),
["200", { "Content-Type" => "text/html" }],
0.1,
)
end
let(:logged_in_data) do
user = Fabricate(:user, active: true)
token = UserAuthToken.generate!(user_id: user.id)
cookie =
create_auth_cookie(
token: token.unhashed_auth_token,
user_id: user.id,
trust_level: user.trust_level,
issued_at: 5.minutes.ago,
)
Middleware::RequestTracker.get_data(
env(
"HTTP_USER_AGENT" =>
"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/90.0.4430.72 Safari/537.36",
"HTTP_COOKIE" => "_t=#{cookie};",
),
["200", { "Content-Type" => "text/html" }],
0.1,
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
)
end
it "does not ignore anonymous requests for public sites" do
SiteSetting.login_required = false
Middleware::RequestTracker.log_request(anon_data)
Middleware::RequestTracker.log_request(logged_in_data)
CachedCounting.flush
expect(ApplicationRequest.http_total.first.count).to eq(2)
expect(ApplicationRequest.http_2xx.first.count).to eq(2)
expect(ApplicationRequest.page_view_logged_in.first.count).to eq(1)
expect(ApplicationRequest.page_view_anon.first.count).to eq(1)
end
it "ignores anonymous requests for private sites" do
SiteSetting.login_required = true
Middleware::RequestTracker.log_request(anon_data)
Middleware::RequestTracker.log_request(logged_in_data)
CachedCounting.flush
expect(ApplicationRequest.http_total.first.count).to eq(2)
expect(ApplicationRequest.http_2xx.first.count).to eq(2)
expect(ApplicationRequest.page_view_logged_in.first.count).to eq(1)
expect(ApplicationRequest.page_view_anon.first).to eq(nil)
end
end
end
describe "rate limiting" do
before do
RateLimiter.enable
RateLimiter.clear_all_global!
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
RateLimiter.clear_all!
@orig_logger = Rails.logger
Rails.logger = @fake_logger = FakeLogger.new
# rate limiter tests depend on checks for retry-after
# they can be sensitive to clock skew during test runs
freeze_time DateTime.parse("2021-01-01 01:00")
end
after { Rails.logger = @orig_logger }
let :middleware do
app = lambda { |env| [200, {}, ["OK"]] }
Middleware::RequestTracker.new(app)
end
it "does nothing if configured to do nothing" do
global_setting :max_reqs_per_ip_mode, "none"
global_setting :max_reqs_per_ip_per_10_seconds, 1
status, _ = middleware.call(env)
status, _ = middleware.call(env)
expect(status).to eq(200)
end
it "blocks private IPs if not skipped" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "warn+block"
global_setting :max_reqs_rate_limit_on_private, true
addresses = %w[
127.1.2.3
127.0.0.2
192.168.1.2
10.0.1.2
172.16.9.8
172.19.1.2
172.20.9.8
172.29.1.2
172.30.9.8
172.31.1.2
]
warn_count = 1
addresses.each do |addr|
env1 = env("REMOTE_ADDR" => addr)
status, _ = middleware.call(env1)
status, _ = middleware.call(env1)
expect(
@fake_logger.warnings.count { |w| w.include?("Global IP rate limit exceeded") },
).to eq(warn_count)
expect(status).to eq(429)
warn_count += 1
end
end
it "blocks if the ip isn't static skipped" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "block"
env1 = env("REMOTE_ADDR" => "1.1.1.1")
status, _ = middleware.call(env1)
status, _ = middleware.call(env1)
expect(status).to eq(429)
end
it "doesn't block if rate limiter is enabled but IP is on the static exception list" do
stub_const(
Middleware::RequestTracker,
"STATIC_IP_SKIPPER",
"177.33.14.73 191.209.88.192/30"&.split&.map { |ip| IPAddr.new(ip) },
) do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "block"
env1 = env("REMOTE_ADDR" => "177.33.14.73")
env2 = env("REMOTE_ADDR" => "191.209.88.194")
status, _ = middleware.call(env1)
expect(status).to eq(200)
status, _ = middleware.call(env1)
expect(status).to eq(200)
status, _ = middleware.call(env2)
expect(status).to eq(200)
status, _ = middleware.call(env2)
expect(status).to eq(200)
end
end
describe "register_ip_skipper" do
before do
Middleware::RequestTracker.register_ip_skipper { |ip| ip == "1.1.1.2" }
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "block"
end
after { Middleware::RequestTracker.unregister_ip_skipper }
it "won't block if the ip is skipped" do
env1 = env("REMOTE_ADDR" => "1.1.1.2")
status, _ = middleware.call(env1)
status, _ = middleware.call(env1)
expect(status).to eq(200)
end
it "blocks if the ip isn't skipped" do
env1 = env("REMOTE_ADDR" => "1.1.1.1")
status, _ = middleware.call(env1)
status, _ = middleware.call(env1)
expect(status).to eq(429)
end
end
it "does nothing for private IPs if skipped" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "warn+block"
global_setting :max_reqs_rate_limit_on_private, false
addresses = %w[
127.1.2.3
127.0.3.1
192.168.1.2
10.0.1.2
172.16.9.8
172.19.1.2
172.20.9.8
172.29.1.2
172.30.9.8
172.31.1.2
]
addresses.each do |addr|
env1 = env("REMOTE_ADDR" => addr)
status, _ = middleware.call(env1)
status, _ = middleware.call(env1)
expect(
@fake_logger.warnings.count { |w| w.include?("Global IP rate limit exceeded") },
).to eq(0)
expect(status).to eq(200)
end
end
it "does warn if rate limiter is enabled via warn+block" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "warn+block"
status, _ = middleware.call(env)
status, headers = middleware.call(env)
expect(@fake_logger.warnings.count { |w| w.include?("Global IP rate limit exceeded") }).to eq(
1,
)
expect(status).to eq(429)
expect(headers["Retry-After"]).to eq("10")
end
it "does warn if rate limiter is enabled" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "warn"
status, _ = middleware.call(env)
status, _ = middleware.call(env)
expect(@fake_logger.warnings.count { |w| w.include?("Global IP rate limit exceeded") }).to eq(
1,
)
expect(status).to eq(200)
end
it "allows assets for more requests" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "block"
global_setting :max_asset_reqs_per_ip_per_10_seconds, 3
env1 = env("REMOTE_ADDR" => "1.1.1.1", "DISCOURSE_IS_ASSET_PATH" => 1)
status, _ = middleware.call(env1)
expect(status).to eq(200)
status, _ = middleware.call(env1)
expect(status).to eq(200)
status, _ = middleware.call(env1)
expect(status).to eq(200)
status, headers = middleware.call(env1)
expect(status).to eq(429)
expect(headers["Retry-After"]).to eq("10")
env2 = env("REMOTE_ADDR" => "1.1.1.1")
status, headers = middleware.call(env2)
expect(status).to eq(429)
expect(headers["Retry-After"]).to eq("10")
end
it "does block if rate limiter is enabled" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
global_setting :max_reqs_per_ip_mode, "block"
env1 = env("REMOTE_ADDR" => "1.1.1.1")
env2 = env("REMOTE_ADDR" => "1.1.1.2")
status, _ = middleware.call(env1)
expect(status).to eq(200)
status, headers = middleware.call(env1)
expect(status).to eq(429)
expect(headers["Retry-After"]).to eq("10")
status, _ = middleware.call(env2)
expect(status).to eq(200)
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
describe "diagnostic information" do
it "is included when the requests-per-10-seconds limit is reached" do
global_setting :max_reqs_per_ip_per_10_seconds, 1
called = 0
app =
lambda do |_|
called += 1
[200, {}, ["OK"]]
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
env = env("REMOTE_ADDR" => "1.1.1.1")
middleware = Middleware::RequestTracker.new(app)
status, = middleware.call(env)
expect(status).to eq(200)
expect(called).to eq(1)
env = env("REMOTE_ADDR" => "1.1.1.1")
middleware = Middleware::RequestTracker.new(app)
status, headers, response = middleware.call(env)
expect(status).to eq(429)
expect(called).to eq(1)
expect(headers["Discourse-Rate-Limit-Error-Code"]).to eq("ip_10_secs_limit")
expect(response.first).to include("Error code: ip_10_secs_limit.")
end
it "is included when the requests-per-minute limit is reached" do
global_setting :max_reqs_per_ip_per_minute, 1
called = 0
app =
lambda do |_|
called += 1
[200, {}, ["OK"]]
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
env = env("REMOTE_ADDR" => "1.1.1.1")
middleware = Middleware::RequestTracker.new(app)
status, = middleware.call(env)
expect(status).to eq(200)
expect(called).to eq(1)
env = env("REMOTE_ADDR" => "1.1.1.1")
middleware = Middleware::RequestTracker.new(app)
status, headers, response = middleware.call(env)
expect(status).to eq(429)
expect(called).to eq(1)
expect(headers["Discourse-Rate-Limit-Error-Code"]).to eq("ip_60_secs_limit")
expect(response.first).to include("Error code: ip_60_secs_limit.")
end
it "is included when the assets-requests-per-10-seconds limit is reached" do
global_setting :max_asset_reqs_per_ip_per_10_seconds, 1
called = 0
app =
lambda do |env|
called += 1
env["DISCOURSE_IS_ASSET_PATH"] = true
[200, {}, ["OK"]]
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
env = env("REMOTE_ADDR" => "1.1.1.1")
middleware = Middleware::RequestTracker.new(app)
status, = middleware.call(env)
expect(status).to eq(200)
expect(called).to eq(1)
env = env("REMOTE_ADDR" => "1.1.1.1")
middleware = Middleware::RequestTracker.new(app)
status, headers, response = middleware.call(env)
expect(status).to eq(429)
expect(called).to eq(1)
expect(headers["Discourse-Rate-Limit-Error-Code"]).to eq("ip_assets_10_secs_limit")
expect(response.first).to include("Error code: ip_assets_10_secs_limit.")
end
end
it "users with high enough trust level are not rate limited per ip" do
global_setting :max_reqs_per_ip_per_minute, 1
global_setting :skip_per_ip_rate_limit_trust_level, 3
envs =
3.times.map do |n|
user = Fabricate(:user, trust_level: 3)
token = UserAuthToken.generate!(user_id: user.id)
cookie =
create_auth_cookie(
token: token.unhashed_auth_token,
user_id: user.id,
trust_level: user.trust_level,
issued_at: 5.minutes.ago,
)
env("HTTP_COOKIE" => "_t=#{cookie}", "REMOTE_ADDR" => "1.1.1.1")
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
called = 0
app =
lambda do |env|
called += 1
[200, {}, ["OK"]]
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
envs.each do |env|
middleware = Middleware::RequestTracker.new(app)
status, = middleware.call(env)
expect(status).to eq(200)
end
expect(called).to eq(3)
envs.each do |env|
middleware = Middleware::RequestTracker.new(app)
status, headers, response = middleware.call(env)
expect(status).to eq(429)
expect(headers["Discourse-Rate-Limit-Error-Code"]).to eq("id_60_secs_limit")
expect(response.first).to include("Error code: id_60_secs_limit.")
end
expect(called).to eq(3)
end
it "falls back to IP rate limiting if the cookie is too old" do
unfreeze_time
global_setting :max_reqs_per_ip_per_minute, 1
global_setting :skip_per_ip_rate_limit_trust_level, 3
user = Fabricate(:user, trust_level: 3)
token = UserAuthToken.generate!(user_id: user.id)
cookie =
create_auth_cookie(
token: token.unhashed_auth_token,
user_id: user.id,
trust_level: user.trust_level,
issued_at: 5.minutes.ago,
)
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
env = env("HTTP_COOKIE" => "_t=#{cookie}", "REMOTE_ADDR" => "1.1.1.1")
called = 0
app =
lambda do |_|
called += 1
[200, {}, ["OK"]]
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
freeze_time(12.minutes.from_now) do
middleware = Middleware::RequestTracker.new(app)
status, = middleware.call(env)
expect(status).to eq(200)
middleware = Middleware::RequestTracker.new(app)
status, headers, response = middleware.call(env)
expect(status).to eq(429)
expect(headers["Discourse-Rate-Limit-Error-Code"]).to eq("ip_60_secs_limit")
expect(response.first).to include("Error code: ip_60_secs_limit.")
end
end
it "falls back to IP rate limiting if the cookie is tampered with" do
unfreeze_time
global_setting :max_reqs_per_ip_per_minute, 1
global_setting :skip_per_ip_rate_limit_trust_level, 3
user = Fabricate(:user, trust_level: 3)
token = UserAuthToken.generate!(user_id: user.id)
cookie =
create_auth_cookie(
token: token.unhashed_auth_token,
user_id: user.id,
trust_level: user.trust_level,
issued_at: Time.zone.now,
)
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
cookie = swap_2_different_characters(cookie)
env = env("HTTP_COOKIE" => "_t=#{cookie}", "REMOTE_ADDR" => "1.1.1.1")
called = 0
app =
lambda do |_|
called += 1
[200, {}, ["OK"]]
end
FEATURE: Apply rate limits per user instead of IP for trusted users (#14706) Currently, Discourse rate limits all incoming requests by the IP address they originate from regardless of the user making the request. This can be frustrating if there are multiple users using Discourse simultaneously while sharing the same IP address (e.g. employees in an office). This commit implements a new feature to make Discourse apply rate limits by user id rather than IP address for users at or higher than the configured trust level (1 is the default). For example, let's say a Discourse instance is configured to allow 200 requests per minute per IP address, and we have 10 users at trust level 4 using Discourse simultaneously from the same IP address. Before this feature, the 10 users could only make a total of 200 requests per minute before they got rate limited. But with the new feature, each user is allowed to make 200 requests per minute because the rate limits are applied on user id rather than the IP address. The minimum trust level for applying user-id-based rate limits can be configured by the `skip_per_ip_rate_limit_trust_level` global setting. The default is 1, but it can be changed by either adding the `DISCOURSE_SKIP_PER_IP_RATE_LIMIT_TRUST_LEVEL` environment variable with the desired value to your `app.yml`, or changing the setting's value in the `discourse.conf` file. Requests made with API keys are still rate limited by IP address and the relevant global settings that control API keys rate limits. Before this commit, Discourse's auth cookie (`_t`) was simply a 32 characters string that Discourse used to lookup the current user from the database and the cookie contained no additional information about the user. However, we had to change the cookie content in this commit so we could identify the user from the cookie without making a database query before the rate limits logic and avoid introducing a bottleneck on busy sites. Besides the 32 characters auth token, the cookie now includes the user id, trust level and the cookie's generation date, and we encrypt/sign the cookie to prevent tampering. Internal ticket number: t54739.
2021-11-18 04:27:30 +08:00
middleware = Middleware::RequestTracker.new(app)
status, = middleware.call(env)
expect(status).to eq(200)
middleware = Middleware::RequestTracker.new(app)
status, headers, response = middleware.call(env)
expect(status).to eq(429)
expect(headers["Discourse-Rate-Limit-Error-Code"]).to eq("ip_60_secs_limit")
expect(response.first).to include("Error code: ip_60_secs_limit.")
end
end
describe "callbacks" do
def app(result, sql_calls: 0, redis_calls: 0)
lambda do |env|
sql_calls.times { User.where(id: -100).pluck(:id) }
redis_calls.times { Discourse.redis.get("x") }
result
end
end
let(:logger) do
->(env, data) {
@env = env
@data = data
}
end
before { Middleware::RequestTracker.register_detailed_request_logger(logger) }
after { Middleware::RequestTracker.unregister_detailed_request_logger(logger) }
it "can report data from anon cache" do
cache = Middleware::AnonymousCache.new(app([200, {}, ["i am a thing"]]))
tracker = Middleware::RequestTracker.new(cache)
uri = "/path?#{SecureRandom.hex}"
request_params = { "a" => "b", "action" => "bob", "controller" => "jane" }
tracker.call(
env(
"REQUEST_URI" => uri,
"ANON_CACHE_DURATION" => 60,
"action_dispatch.request.parameters" => request_params,
),
)
expect(@data[:cache]).to eq("skip")
tracker.call(
env(
"REQUEST_URI" => uri,
"ANON_CACHE_DURATION" => 60,
"action_dispatch.request.parameters" => request_params,
),
)
expect(@data[:cache]).to eq("store")
tracker.call(env("REQUEST_URI" => uri, "ANON_CACHE_DURATION" => 60))
expect(@data[:cache]).to eq("true")
# not allowlisted
request_params.delete("a")
expect(@env["action_dispatch.request.parameters"]).to eq(request_params)
end
it "can correctly log detailed data" do
global_setting :enable_performance_http_headers, true
2018-03-27 14:57:19 +08:00
# ensure pg is warmed up with the select 1 query
User.where(id: -100).pluck(:id)
freeze_time
start = Time.now.to_f
freeze_time 1.minute.from_now
tracker = Middleware::RequestTracker.new(app([200, {}, []], sql_calls: 2, redis_calls: 2))
_, headers, _ = tracker.call(env("HTTP_X_REQUEST_START" => "t=#{start}"))
expect(@data[:queue_seconds]).to eq(60)
timing = @data[:timing]
expect(timing[:total_duration]).to be > 0
expect(timing[:sql][:duration]).to be > 0
expect(timing[:sql][:calls]).to eq 2
expect(timing[:redis][:duration]).to be > 0
expect(timing[:redis][:calls]).to eq 2
expect(headers["X-Queue-Time"]).to eq("60.000000")
expect(headers["X-Redis-Calls"]).to eq("2")
expect(headers["X-Redis-Time"].to_f).to be > 0
expect(headers["X-Sql-Calls"]).to eq("2")
expect(headers["X-Sql-Time"].to_f).to be > 0
expect(headers["X-Runtime"].to_f).to be > 0
end
it "can correctly log messagebus request types" do
tracker = Middleware::RequestTracker.new(app([200, {}, []]))
tracker.call(env(path: "/message-bus/abcde/poll"))
expect(@data[:is_background]).to eq(true)
expect(@data[:background_type]).to eq("message-bus")
tracker.call(env(path: "/message-bus/abcde/poll?dlp=t"))
expect(@data[:is_background]).to eq(true)
expect(@data[:background_type]).to eq("message-bus-dlp")
tracker.call(env("HTTP_DONT_CHUNK" => "True", :path => "/message-bus/abcde/poll"))
expect(@data[:is_background]).to eq(true)
expect(@data[:background_type]).to eq("message-bus-dontchunk")
end
end
end