From 4d0935ba5e6900d87e8420ac9d7de89e5440f24d Mon Sep 17 00:00:00 2001
From: epriestley <git@epriestley.com>
Date: Tue, 8 Apr 2014 18:36:21 -0700
Subject: [PATCH] Rate limit requests by IP

Summary:
Fixes T3923. On `secure.phabricator.com`, we occasionally get slowed to a crawl when someone runs a security scanner against us, or 5 search bots decide to simultaneously index every line of every file in Diffusion.

Every time a user makes a request, give their IP address some points. If they get too many points in 5 minutes, start blocking their requests automatically for a while.

We give fewer points for logged in requests. We could futher refine this (more points for a 404, more points for a really slow page, etc.) but let's start simply.

Also, provide a mechanism for configuring this, and configuring the LB environment stuff at the same time (this comes up rarely, but we don't have a good answer right now).

Test Plan: Used `ab` and reloading over and over again to hit rate limits. Read documentation.

Reviewers: btrahan

Reviewed By: btrahan

Subscribers: chad, epriestley

Maniphest Tasks: T3923

Differential Revision: https://secure.phabricator.com/D8713
---
 .gitignore                                    |   3 +
 .../configuration/configuration_guide.diviner |   3 +
 .../configuring_preamble.diviner              | 114 ++++++++
 support/PhabricatorStartup.php                | 252 ++++++++++++++++++
 webroot/index.php                             |  27 +-
 5 files changed, 398 insertions(+), 1 deletion(-)
 create mode 100644 src/docs/user/configuration/configuring_preamble.diviner

diff --git a/.gitignore b/.gitignore
index b3c0a8afdd..a84df0e9b2 100644
--- a/.gitignore
+++ b/.gitignore
@@ -33,6 +33,9 @@
 # User-accessible hook for adhoc debugging scripts
 /support/debug.php
 
+# User-accessible hook for adhoc startup code
+/support/preamble.php
+
 # Users can link binaries here
 /support/bin/*
 
diff --git a/src/docs/user/configuration/configuration_guide.diviner b/src/docs/user/configuration/configuration_guide.diviner
index 6666dd0ca2..bd0aa922cb 100644
--- a/src/docs/user/configuration/configuration_guide.diviner
+++ b/src/docs/user/configuration/configuration_guide.diviner
@@ -197,6 +197,9 @@ Continue by:
     @{article:Configuring Accounts and Registration}; or
   - understanding advanced configuration topics with
     @{article:Configuration User Guide: Advanced Configuration}; or
+  - configuring a preamble script to set up the environment properly behind a
+    load balancer, or adjust rate limiting with
+    @{article:Configuring a Preamble Script}; or
   - configuring where uploaded files and attachments will be stored with
     @{article:Configuring File Storage}; or
   - configuring Phabricator so it can send mail with
diff --git a/src/docs/user/configuration/configuring_preamble.diviner b/src/docs/user/configuration/configuring_preamble.diviner
new file mode 100644
index 0000000000..a8710598bf
--- /dev/null
+++ b/src/docs/user/configuration/configuring_preamble.diviner
@@ -0,0 +1,114 @@
+@title Configuring a Preamble Script
+@group config
+
+Adjust environmental settings (SSL, remote IP, rate limiting) using a preamble
+script.
+
+= Overview =
+
+If Phabricator is deployed in an environment where HTTP headers behave oddly
+(usually, because it is behind a load balancer), it may not be able to detect
+some environmental features (like the client's IP, or the presence of SSL)
+correctly.
+
+You can use a special preamble script to make arbitrary adjustments to the
+environment and some parts of Phabricator's configuration in order to fix these
+problems and set up the environment which Phabricator expects.
+
+NOTE: This is an advanced feature. Most installs should not need to configure
+a preamble script.
+
+= Creating a Preamble Script =
+
+To create a preamble script, write a file to:
+
+  phabricator/support/preamble.php
+
+(This file is in Phabricator's `.gitignore`, so you do not need to worry about
+colliding with `git` or interacting with updates.)
+
+This file should be a valid PHP script. If you aren't very familiar with PHP,
+you can check for syntax errors with `php -l`:
+
+  phabricator/ $ php -l support/preamble.php
+  No syntax errors detected in support/preamble.php
+
+If present, this script will be executed at the very beginning of each web
+request, allowing you to adjust the environment. For common adjustments and
+examples, see the next sections.
+
+= Adjusting Client IPs =
+
+If your install is behind a load balancer, Phabricator may incorrectly detect
+all requests as originating from the load balancer, rather than from the correct
+client IPs. If this is the case and some other header (like `X-Forwarded-For`)
+is known to be trustworthy, you can overwrite the `REMOTE_ADDR` setting so
+Phabricator can figure out the client IP correctly:
+
+```
+name=Overwrite REMOTE_ADDR with X-Forwarded-For
+<?php
+
+$_SERVER['REMOTE_ADDR'] = $_SERVER['HTTP_X_FORWARDED_FOR'];
+```
+
+You should do this //only// if the `X-Forwarded-For` header is always
+trustworthy. In particular, if users can make requests to the web server
+directly, they can provide an arbitrary `X-Forwarded-For` header, and thereby
+spoof an arbitrary client IP.
+
+= Adjusting SSL =
+
+If your install is behind an SSL terminating load balancer, Phabricator may
+detect requests as HTTP when the client sees them as HTTPS. This can cause
+Phabricator to generate links with the wrong protocol, issue cookies without
+the SSL-only flag, or reject requests outright.
+
+To fix this, you can set `$_SERVER['HTTPS']` explicitly:
+
+```
+name=Explicitly Configure SSL Availability
+<?php
+
+$_SERVER['HTTPS'] = true;
+```
+
+You can also set this value to `false` to explicitly tell Phabricator that a
+request is not an SSL request.
+
+= Adjusting Rate Limiting =
+
+Phabricator performs coarse, IP-based rate limiting by default. In most
+situations the default settings should be reasonable: they are set fairly high,
+and intended to prevent only significantly abusive behavior.
+
+However, if legitimate traffic is being rate limited (or you want to make the
+limits more strict) you can adjust the limits in the preamble script.
+
+```
+name=Adjust Rate Limiting Behavior
+<?php
+
+// The default is 1000, so a value of 2000 increases the limit by a factor
+// of 2: users will be able to make twice as many requests before being
+// rate limited.
+
+// You can set the limit to 0 to disable rate limiting.
+
+PhabricatorStartup::setMaximumRate(2000);
+```
+
+By examining `$_SERVER['REMOTE_ADDR']` or similar parameters, you could also
+adjust the rate limit dynamically: for example, remove it for requests from an
+internal network, but impose a strict limit for external requests.
+
+Rate limiting needs to be configured in this way in order to make it as cheap as
+possible to activate after a client is rate limited. The limiting checks execute
+before any libraries or configuration are loaded, and can emit a response within
+a few milliseconds.
+
+= Next Steps =
+
+Continue by:
+
+  - returning to the @{article:Configuration Guide}.
diff --git a/support/PhabricatorStartup.php b/support/PhabricatorStartup.php
index 47b902ee84..2f2bf245b9 100644
--- a/support/PhabricatorStartup.php
+++ b/support/PhabricatorStartup.php
@@ -8,10 +8,32 @@
  * NOTE: This class MUST NOT have any dependencies. It runs before libraries
  * load.
  *
+ * Rate Limiting
+ * =============
+ *
+ * Phabricator limits the rate at which clients can request pages, and issues
+ * HTTP 429 "Too Many Requests" responses if clients request too many pages too
+ * quickly. Although this is not a complete defense against high-volume attacks,
+ * it can  protect an install against aggressive crawlers, security scanners,
+ * and some types of malicious activity.
+ *
+ * To perform rate limiting, each page increments a score counter for the
+ * requesting user's IP. The page can give the IP more points for an expensive
+ * request, or fewer for an authetnicated request.
+ *
+ * Score counters are kept in buckets, and writes move to a new bucket every
+ * minute. After a few minutes (defined by @{method:getRateLimitBucketCount}),
+ * the oldest bucket is discarded. This provides a simple mechanism for keeping
+ * track of scores without needing to store, access, or read very much data.
+ *
+ * Users are allowed to accumulate up to 1000 points per minute, averaged across
+ * all of the tracked buckets.
+ *
  * @task info         Accessing Request Information
  * @task hook         Startup Hooks
  * @task apocalypse   In Case Of Apocalypse
  * @task validation   Validation
+ * @task ratelimit    Rate Limiting
  */
 final class PhabricatorStartup {
 
@@ -19,6 +41,7 @@ final class PhabricatorStartup {
   private static $globals = array();
   private static $capturingOutput;
   private static $rawInput;
+  private static $maximumRate = 1000;
 
 
 /* -(  Accessing Request Information  )-------------------------------------- */
@@ -93,6 +116,10 @@ final class PhabricatorStartup {
     self::setupPHP();
     self::verifyPHP();
 
+    if (isset($_SERVER['REMOTE_ADDR'])) {
+      self::rateLimitRequest($_SERVER['REMOTE_ADDR']);
+    }
+
     self::normalizeInput();
 
     self::verifyRewriteRules();
@@ -521,4 +548,229 @@ final class PhabricatorStartup {
       "'post_max_size' is set to '{$config}'.");
   }
 
+
+/* -(  Rate Limiting  )------------------------------------------------------ */
+
+
+  /**
+   * Adjust the permissible rate limit score.
+   *
+   * By default, the limit is `1000`. You can use this method to set it to
+   * a larger or smaller value. If you set it to `2000`, users may make twice
+   * as many requests before rate limiting.
+   *
+   * @param int Maximum score before rate limiting.
+   * @return void
+   * @task ratelimit
+   */
+  public static function setMaximumRate($rate) {
+    self::$maximumRate = $rate;
+  }
+
+
+  /**
+   * Check if the user (identified by `$user_identity`) has issued too many
+   * requests recently. If they have, end the request with a 429 error code.
+   *
+   * The key just needs to identify the user. Phabricator uses both user PHIDs
+   * and user IPs as keys, tracking logged-in and logged-out users separately
+   * and enforcing different limits.
+   *
+   * @param   string  Some key which identifies the user making the request.
+   * @return  void    If the user has exceeded the rate limit, this method
+   *                  does not return.
+   * @task ratelimit
+   */
+  public static function rateLimitRequest($user_identity) {
+    if (!self::canRateLimit()) {
+      return;
+    }
+
+    $score = self::getRateLimitScore($user_identity);
+    if ($score > (self::$maximumRate * self::getRateLimitBucketCount())) {
+      // Give the user some bonus points for getting rate limited. This keeps
+      // bad actors who keep slamming the 429 page locked out completely,
+      // instead of letting them get a burst of requests through every minute
+      // after a bucket expires.
+      self::addRateLimitScore($user_identity, 50);
+      self::didRateLimit($user_identity);
+    }
+  }
+
+
+  /**
+   * Add points to the rate limit score for some user.
+   *
+   * If users have earned more than 1000 points per minute across all the
+   * buckets they'll be locked out of the application, so awarding 1 point per
+   * request roughly corresponds to allowing 1000 requests per second, while
+   * awarding 50 points roughly corresponds to allowing 20 requests per second.
+   *
+   * @param string  Some key which identifies the user making the request.
+   * @param float   The cost for this request; more points pushes them toward
+   *                the limit faster.
+   * @return void
+   * @task ratelimit
+   */
+  public static function addRateLimitScore($user_identity, $score) {
+    if (!self::canRateLimit()) {
+      return;
+    }
+
+    $current = self::getRateLimitBucket();
+
+    // There's a bit of a race here, if a second process reads the bucket before
+    // this one writes it, but it's fine if we occasionally fail to record a
+    // user's score. If they're making requests fast enough to hit rate
+    // limiting, we'll get them soon.
+
+    $bucket_key = self::getRateLimitBucketKey($current);
+    $bucket = apc_fetch($bucket_key);
+    if (!is_array($bucket)) {
+      $bucket = array();
+    }
+
+    if (empty($bucket[$user_identity])) {
+      $bucket[$user_identity] = 0;
+    }
+
+    $bucket[$user_identity] += $score;
+    apc_store($bucket_key, $bucket);
+  }
+
+
+  /**
+   * Determine if rate limiting is available.
+   *
+   * Rate limiting depends on APC, and isn't available unless the APC user
+   * cache is available.
+   *
+   * @return bool True if rate limiting is available.
+   * @task ratelimit
+   */
+  private static function canRateLimit() {
+    if (!self::$maximumRate) {
+      return false;
+    }
+
+    if (!function_exists('apc_fetch')) {
+      return false;
+    }
+
+    return true;
+  }
+
+
+  /**
+   * Get the current bucket for storing rate limit scores.
+   *
+   * @return int The current bucket.
+   * @task ratelimit
+   */
+  private static function getRateLimitBucket() {
+    return (int)(time() / 60);
+  }
+
+
+  /**
+   * Get the total number of rate limit buckets to retain.
+   *
+   * @return int Total number of rate limit buckets to retain.
+   * @task ratelimit
+   */
+  private static function getRateLimitBucketCount() {
+    return 5;
+  }
+
+
+  /**
+   * Get the APC key for a given bucket.
+   *
+   * @param int Bucket to get the key for.
+   * @return string APC key for the bucket.
+   * @task ratelimit
+   */
+  private static function getRateLimitBucketKey($bucket) {
+    return 'rate:bucket:'.$bucket;
+  }
+
+
+  /**
+   * Get the APC key for the smallest stored bucket.
+   *
+   * @return string APC key for the smallest stored bucket.
+   * @task ratelimit
+   */
+  private static function getRateLimitMinKey() {
+    return 'rate:min';
+  }
+
+
+  /**
+   * Get the current rate limit score for a given user.
+   *
+   * @param string Unique key identifying the user.
+   * @return float The user's current score.
+   * @task ratelimit
+   */
+  private static function getRateLimitScore($user_identity) {
+    $min_key = self::getRateLimitMinKey();
+
+    // Identify the oldest bucket stored in APC.
+    $cur = self::getRateLimitBucket();
+    $min = apc_fetch($min_key);
+
+    // If we don't have any buckets stored yet, store the current bucket as
+    // the oldest bucket.
+    if (!$min) {
+      apc_store($min_key, $cur);
+      $min = $cur;
+    }
+
+    // Destroy any buckets that are older than the minimum bucket we're keeping
+    // track of. Under load this normally shouldn't do anything, but will clean
+    // up an old bucket once per minute.
+    $count = self::getRateLimitBucketCount();
+    for ($cursor = $min; $cursor < ($cur - $count); $cursor++) {
+      apc_delete(self::getRateLimitBucketKey($cursor));
+      apc_store($min_key, $cursor + 1);
+    }
+
+    // Now, sum up the user's scores in all of the active buckets.
+    $score = 0;
+    for (; $cursor <= $cur; $cursor++) {
+      $bucket = apc_fetch(self::getRateLimitBucketKey($cursor));
+      if (isset($bucket[$user_identity])) {
+        $score += $bucket[$user_identity];
+      }
+    }
+
+    return $score;
+  }
+
+
+  /**
+   * Emit an HTTP 429 "Too Many Requests" response (indicating that the user
+   * has exceeded application rate limits) and exit.
+   *
+   * @return exit This method **does not return**.
+   * @task ratelimit
+   */
+  private static function didRateLimit() {
+    $message =
+      "TOO MANY REQUESTS\n".
+      "You are issuing too many requests too quickly.\n".
+      "To adjust limits, see \"Configuring a Preamble Script\" in the ".
+      "documentation.";
+
+    header(
+      'Content-Type: text/plain; charset=utf-8',
+      $replace = true,
+      $http_error = 429);
+
+    echo $message;
+
+    exit(1);
+  }
+
 }
diff --git a/webroot/index.php b/webroot/index.php
index 7072727e7f..0adaf7322f 100644
--- a/webroot/index.php
+++ b/webroot/index.php
@@ -1,6 +1,14 @@
 <?php
 
-require_once dirname(dirname(__FILE__)).'/support/PhabricatorStartup.php';
+$phabricator_root = dirname(dirname(__FILE__));
+require_once $phabricator_root.'/support/PhabricatorStartup.php';
+
+// If the preamble script exists, load it.
+$preamble_path = $phabricator_root.'/support/preamble.php';
+if (file_exists($preamble_path)) {
+  require_once $preamble_path;
+}
+
 PhabricatorStartup::didStartup();
 
 $show_unexpected_traces = false;
@@ -142,6 +150,23 @@ try {
     ));
 
   DarkConsoleXHProfPluginAPI::saveProfilerSample($access_log);
+
+  // Add points to the rate limits for this request.
+  if (isset($_SERVER['REMOTE_ADDR'])) {
+    $user_ip = $_SERVER['REMOTE_ADDR'];
+
+    // The base score for a request allows users to make 30 requests per
+    // minute.
+    $score = (1000 / 30);
+
+    // If the user was logged in, let them make more requests.
+    if ($request->getUser() && $request->getUser()->getPHID()) {
+      $score = $score / 5;
+    }
+
+    PhabricatorStartup::addRateLimitScore($user_ip, $score);
+  }
+
 } catch (Exception $ex) {
   PhabricatorStartup::didEncounterFatalException(
     'Core Exception',