Merge branch 'main' into python-model-python2-specific-command-execution

python/ql/src/experimental/Security-new-dataflow/CWE-094/CodeInjection.ql

@@ -0,0 +1,35 @@
+/**
+ * @name Code injection
+ * @description Interpreting unsanitized user input as code allows a malicious user to perform arbitrary
+ *              code execution.
+ * @kind path-problem
+ * @problem.severity error
+ * @sub-severity high
+ * @precision high
+ * @id py/code-injection
+ * @tags security
+ *       external/owasp/owasp-a1
+ *       external/cwe/cwe-094
+ *       external/cwe/cwe-095
+ *       external/cwe/cwe-116
+ */
+
+import python
+import experimental.dataflow.DataFlow
+import experimental.dataflow.TaintTracking
+import experimental.semmle.python.Concepts
+import experimental.dataflow.RemoteFlowSources
+import DataFlow::PathGraph
+
+class CodeInjectionConfiguration extends TaintTracking::Configuration {
+  CodeInjectionConfiguration() { this = "CodeInjectionConfiguration" }
+
+  override predicate isSource(DataFlow::Node source) { source instanceof RemoteFlowSource }
+
+  override predicate isSink(DataFlow::Node sink) { sink = any(CodeExecution e).getCode() }
+}
+
+from CodeInjectionConfiguration config, DataFlow::PathNode source, DataFlow::PathNode sink
+where config.hasFlowPath(source, sink)
+select sink.getNode(), source, sink, "$@ flows to here and is interpreted as code.",
+  source.getNode(), "A user-provided value"

python/ql/src/experimental/dataflow/RemoteFlowSources.qll

@@ -2,6 +2,7 @@ private import python
 private import experimental.dataflow.DataFlow
 // Need to import since frameworks can extend `RemoteFlowSource::Range`
 private import experimental.semmle.python.Frameworks
+private import experimental.semmle.python.Concepts
 
 /**
  * A data flow source of remote user input.

python/ql/src/experimental/dataflow/TypeTracker.qll

@@ -71,7 +71,8 @@ module StepSummary {
 /** Holds if it's reasonable to expect the data flow step from `nodeFrom` to `nodeTo` to preserve types. */
 private predicate typePreservingStep(Node nodeFrom, Node nodeTo) {
   EssaFlow::essaFlowStep(nodeFrom, nodeTo) or
-  jumpStep(nodeFrom, nodeTo)
+  jumpStep(nodeFrom, nodeTo) or
+  nodeFrom = nodeTo.(PostUpdateNode).getPreUpdateNode()
 }
 
 /** Holds if `nodeFrom` steps to `nodeTo` by being passed as a parameter in a call. */

python/ql/src/experimental/dataflow/internal/DataFlowPrivate.qll

@@ -150,22 +150,6 @@ module EssaFlow {
     //   nodeTo is `y` on second line, cfg node
     useToNextUse(nodeFrom.asCfgNode(), nodeTo.asCfgNode())
     or
-    // Refinements
-    exists(EssaEdgeRefinement r |
-      nodeTo.(EssaNode).getVar() = r.getVariable() and
-      nodeFrom.(EssaNode).getVar() = r.getInput()
-    )
-    or
-    exists(EssaNodeRefinement r |
-      nodeTo.(EssaNode).getVar() = r.getVariable() and
-      nodeFrom.(EssaNode).getVar() = r.getInput()
-    )
-    or
-    exists(PhiFunction p |
-      nodeTo.(EssaNode).getVar() = p.getVariable() and
-      nodeFrom.(EssaNode).getVar() = p.getAnInput()
-    )
-    or
     // If expressions
     nodeFrom.asCfgNode() = nodeTo.asCfgNode().(IfExprNode).getAnOperand()
   }

python/ql/src/experimental/dataflow/internal/DataFlowPublic.qll

@@ -152,6 +152,9 @@ class ParameterNode extends EssaNode {
   }
 
   override DataFlowCallable getEnclosingCallable() { this.isParameterOf(result, _) }
+
+  /** Gets the `Parameter` this `ParameterNode` represents. */
+  Parameter getParameter() { result = var.(ParameterDefinition).getParameter() }
 }
 
 /**

python/ql/src/experimental/dataflow/internal/DataFlowUtil.qll

@@ -68,5 +68,5 @@ Node importNode(string name) {
   // Because named imports are modelled as `AttrRead`s, the statement `from foo import bar as baz`
   // is interpreted as if it was an assignment `baz = foo.bar`, which means `baz` gets tracked as a
   // reference to `foo.bar`, as desired.
-  result.asCfgNode().getNode() = any(ImportExpr i | i.getAnImportedModuleName() = name)
+  result.asCfgNode().getNode() = any(ImportExpr i | i.getName() = name)
 }

python/ql/src/experimental/semmle/python/Concepts.qll

@@ -7,6 +7,7 @@
 import python
 private import experimental.dataflow.DataFlow
 private import experimental.semmle.python.Frameworks
+private import experimental.dataflow.RemoteFlowSources
 
 /**
  * A data-flow node that executes an operating system command,
@@ -16,12 +17,12 @@ private import experimental.semmle.python.Frameworks
  * extend `SystemCommandExecution::Range` instead.
  */
 class SystemCommandExecution extends DataFlow::Node {
-  SystemCommandExecution::Range self;
+  SystemCommandExecution::Range range;
 
-  SystemCommandExecution() { this = self }
+  SystemCommandExecution() { this = range }
 
   /** Gets the argument that specifies the command to be executed. */
-  DataFlow::Node getCommand() { result = self.getCommand() }
+  DataFlow::Node getCommand() { result = range.getCommand() }
 }
 
 /** Provides a class for modeling new system-command execution APIs. */
@@ -38,3 +39,91 @@ module SystemCommandExecution {
     abstract DataFlow::Node getCommand();
   }
 }
+
+/**
+ * A data-flow node that dynamically executes Python code.
+ *
+ * Extend this class to refine existing API models. If you want to model new APIs,
+ * extend `CodeExecution::Range` instead.
+ */
+class CodeExecution extends DataFlow::Node {
+  CodeExecution::Range range;
+
+  CodeExecution() { this = range }
+
+  /** Gets the argument that specifies the code to be executed. */
+  DataFlow::Node getCode() { result = range.getCode() }
+}
+
+/** Provides a class for modeling new dynamic code execution APIs. */
+module CodeExecution {
+  /**
+   * A data-flow node that dynamically executes Python code.
+   *
+   * Extend this class to model new APIs. If you want to refine existing API models,
+   * extend `CodeExecution` instead.
+   */
+  abstract class Range extends DataFlow::Node {
+    /** Gets the argument that specifies the code to be executed. */
+    abstract DataFlow::Node getCode();
+  }
+}
+
+/** Provides classes for modeling HTTP-related APIs. */
+module HTTP {
+  /** Provides classes for modeling HTTP servers. */
+  module Server {
+    /**
+     * An data-flow node that sets up a route on a server.
+     *
+     * Extend this class to refine existing API models. If you want to model new APIs,
+     * extend `RouteSetup::Range` instead.
+     */
+    class RouteSetup extends DataFlow::Node {
+      RouteSetup::Range range;
+
+      RouteSetup() { this = range }
+
+      /** Gets the URL pattern for this route, if it can be statically determined. */
+      string getUrlPattern() { result = range.getUrlPattern() }
+
+      /** Gets a function that will handle incoming requests for this route, if any. */
+      Function getARouteHandler() { result = range.getARouteHandler() }
+
+      /**
+       * Gets a parameter that will receive parts of the url when handling incoming
+       * requests for this route, if any. These automatically become a `RemoteFlowSource`.
+       */
+      Parameter getARoutedParameter() { result = range.getARoutedParameter() }
+    }
+
+    /** Provides a class for modeling new HTTP routing APIs. */
+    module RouteSetup {
+      /**
+       * An data-flow node that sets up a route on a server.
+       *
+       * Extend this class to model new APIs. If you want to refine existing API models,
+       * extend `RouteSetup` instead.
+       */
+      abstract class Range extends DataFlow::Node {
+        /** Gets the URL pattern for this route, if it can be statically determined. */
+        abstract string getUrlPattern();
+
+        /** Gets a function that will handle incoming requests for this route, if any. */
+        abstract Function getARouteHandler();
+
+        /**
+         * Gets a parameter that will receive parts of the url when handling incoming
+         * requests for this route, if any. These automatically become a `RemoteFlowSource`.
+         */
+        abstract Parameter getARoutedParameter();
+      }
+    }
+
+    private class RoutedParameter extends RemoteFlowSource::Range, DataFlow::ParameterNode {
+      RoutedParameter() { this.getParameter() = any(RouteSetup setup).getARoutedParameter() }
+
+      override string getSourceType() { result = "RoutedParameter" }
+    }
+  }
+}

python/ql/src/experimental/semmle/python/Frameworks.qll

@@ -2,6 +2,7 @@
  * Helper file that imports all framework modeling.
  */
 
-private import experimental.semmle.python.frameworks.Flask
 private import experimental.semmle.python.frameworks.Django
+private import experimental.semmle.python.frameworks.Flask
+private import experimental.semmle.python.frameworks.Invoke
 private import experimental.semmle.python.frameworks.Stdlib

python/ql/src/experimental/semmle/python/frameworks/Flask.qll

@@ -1,5 +1,6 @@
 /**
- * Provides classes modeling security-relevant aspects of the `flask` package.
+ * Provides classes modeling security-relevant aspects of the `flask` PyPI package.
+ * See https://flask.palletsprojects.com/en/1.1.x/.
  */
 
 private import python
@@ -11,6 +12,10 @@ private import experimental.semmle.python.frameworks.Werkzeug
 
 // for old improved impl see
 // https://github.com/github/codeql/blob/9f95212e103c68d0c1dfa4b6f30fb5d53954ccef/python/ql/src/semmle/python/web/flask/Request.qll
+/**
+ * Provides models for the `flask` PyPI package.
+ * See https://flask.palletsprojects.com/en/1.1.x/.
+ */
 private module Flask {
   /** Gets a reference to the `flask` module. */
   DataFlow::Node flask(DataFlow::TypeTracker t) {
@@ -23,6 +28,7 @@ private module Flask {
   /** Gets a reference to the `flask` module. */
   DataFlow::Node flask() { result = flask(DataFlow::TypeTracker::end()) }
 
+  /** Provides models for the `flask` module. */
   module flask {
     /** Gets a reference to the `flask.request` object. */
     DataFlow::Node request(DataFlow::TypeTracker t) {
@@ -32,13 +38,154 @@ private module Flask {
       t.startInAttr("request") and
       result = flask()
       or
-      exists(DataFlow::TypeTracker t2 | result = flask::request(t2).track(t2, t))
+      exists(DataFlow::TypeTracker t2 | result = request(t2).track(t2, t))
     }
 
     /** Gets a reference to the `flask.request` object. */
-    DataFlow::Node request() { result = flask::request(DataFlow::TypeTracker::end()) }
+    DataFlow::Node request() { result = request(DataFlow::TypeTracker::end()) }
+
+    /** Gets a reference to the `flask.Flask` class. */
+    private DataFlow::Node classFlask(DataFlow::TypeTracker t) {
+      t.start() and
+      result = DataFlow::importNode("flask.Flask")
+      or
+      t.startInAttr("Flask") and
+      result = flask()
+      or
+      exists(DataFlow::TypeTracker t2 | result = classFlask(t2).track(t2, t))
+    }
+
+    /** Gets a reference to the `flask.Flask` class. */
+    DataFlow::Node classFlask() { result = classFlask(DataFlow::TypeTracker::end()) }
+
+    /** Gets a reference to an instance of `flask.Flask` (a Flask application). */
+    private DataFlow::Node app(DataFlow::TypeTracker t) {
+      t.start() and
+      result.asCfgNode().(CallNode).getFunction() = flask::classFlask().asCfgNode()
+      or
+      exists(DataFlow::TypeTracker t2 | result = app(t2).track(t2, t))
+    }
+
+    /** Gets a reference to an instance of `flask.Flask` (a flask application). */
+    DataFlow::Node app() { result = app(DataFlow::TypeTracker::end()) }
   }
 
+  // ---------------------------------------------------------------------------
+  // routing modeling
+  // ---------------------------------------------------------------------------
+  /**
+   * Gets a reference to the attribute `attr_name` of a flask application.
+   * WARNING: Only holds for a few predefined attributes.
+   */
+  private DataFlow::Node app_attr(DataFlow::TypeTracker t, string attr_name) {
+    attr_name in ["route", "add_url_rule"] and
+    t.startInAttr(attr_name) and
+    result = flask::app()
+    or
+    // Due to bad performance when using normal setup with `app_attr(t2, attr_name).track(t2, t)`
+    // we have inlined that code and forced a join
+    exists(DataFlow::TypeTracker t2 |
+      exists(DataFlow::StepSummary summary |
+        app_attr_first_join(t2, attr_name, result, summary) and
+        t = t2.append(summary)
+      )
+    )
+  }
+
+  pragma[nomagic]
+  private predicate app_attr_first_join(
+    DataFlow::TypeTracker t2, string attr_name, DataFlow::Node res, DataFlow::StepSummary summary
+  ) {
+    DataFlow::StepSummary::step(app_attr(t2, attr_name), res, summary)
+  }
+
+  /**
+   * Gets a reference to the attribute `attr_name` of a flask application.
+   * WARNING: Only holds for a few predefined attributes.
+   */
+  private DataFlow::Node app_attr(string attr_name) {
+    result = app_attr(DataFlow::TypeTracker::end(), attr_name)
+  }
+
+  private string werkzeug_rule_re() {
+    // since flask uses werkzeug internally, we are using its routing rules from
+    // https://github.com/pallets/werkzeug/blob/4dc8d6ab840d4b78cbd5789cef91b01e3bde01d5/src/werkzeug/routing.py#L138-L151
+    result =
+      "(?<static>[^<]*)<(?:(?<converter>[a-zA-Z_][a-zA-Z0-9_]*)(?:\\((?<args>.*?)\\))?\\:)?(?<variable>[a-zA-Z_][a-zA-Z0-9_]*)>"
+  }
+
+  /** A route setup made by flask (sharing handling of URL patterns). */
+  abstract private class FlaskRouteSetup extends HTTP::Server::RouteSetup::Range {
+    override Parameter getARoutedParameter() {
+      // If we don't know the URL pattern, we simply mark all parameters as a routed
+      // parameter. This should give us more RemoteFlowSources but could also lead to
+      // more FPs. If this turns out to be the wrong tradeoff, we can always change our mind.
+      not exists(this.getUrlPattern()) and
+      result = this.getARouteHandler().getArgByName(_)
+      or
+      exists(string name |
+        result = this.getARouteHandler().getArgByName(name) and
+        exists(string match |
+          match = this.getUrlPattern().regexpFind(werkzeug_rule_re(), _, _) and
+          name = match.regexpCapture(werkzeug_rule_re(), 4)
+        )
+      )
+    }
+
+    /** Gets the argument used to pass in the URL pattern. */
+    abstract DataFlow::Node getUrlPatternArg();
+
+    override string getUrlPattern() {
+      exists(StrConst str |
+        DataFlow::localFlow(DataFlow::exprNode(str), this.getUrlPatternArg()) and
+        result = str.getText()
+      )
+    }
+  }
+
+  /**
+   * A call to `flask.Flask.route`.
+   *
+   * See https://flask.palletsprojects.com/en/1.1.x/api/#flask.Flask.route
+   */
+  private class FlaskAppRouteCall extends FlaskRouteSetup, DataFlow::CfgNode {
+    override CallNode node;
+
+    FlaskAppRouteCall() { node.getFunction() = app_attr("route").asCfgNode() }
+
+    override DataFlow::Node getUrlPatternArg() {
+      result.asCfgNode() in [node.getArg(0), node.getArgByName("rule")]
+    }
+
+    override Function getARouteHandler() { result.getADecorator().getAFlowNode() = node }
+  }
+
+  /**
+   * A call to `flask.Flask.add_url_rule`.
+   *
+   * See https://flask.palletsprojects.com/en/1.1.x/api/#flask.Flask.add_url_rule
+   */
+  private class FlaskAppAddUrlRule extends FlaskRouteSetup, DataFlow::CfgNode {
+    override CallNode node;
+
+    FlaskAppAddUrlRule() { node.getFunction() = app_attr("add_url_rule").asCfgNode() }
+
+    override DataFlow::Node getUrlPatternArg() {
+      result.asCfgNode() in [node.getArg(0), node.getArgByName("rule")]
+    }
+
+    override Function getARouteHandler() {
+      exists(DataFlow::Node view_func_arg, DataFlow::Node func_src |
+        view_func_arg.asCfgNode() in [node.getArg(2), node.getArgByName("view_func")] and
+        DataFlow::localFlow(func_src, view_func_arg) and
+        func_src.asExpr().(CallableExpr) = result.getDefinition()
+      )
+    }
+  }
+
+  // ---------------------------------------------------------------------------
+  // flask.Request taint modeling
+  // ---------------------------------------------------------------------------
   // TODO: Do we even need this class? :|
   /**
    * A source of remote flow from a flask request.

python/ql/src/experimental/semmle/python/frameworks/Invoke.qll

@@ -0,0 +1,149 @@
+/**
+ * Provides classes modeling security-relevant aspects of the `invoke` PyPI package.
+ * See https://www.pyinvoke.org/.
+ */
+
+private import python
+private import experimental.dataflow.DataFlow
+private import experimental.semmle.python.Concepts
+
+/**
+ * Provides models for the `invoke` PyPI package.
+ * See https://www.pyinvoke.org/.
+ */
+private module Invoke {
+  // ---------------------------------------------------------------------------
+  // invoke
+  // ---------------------------------------------------------------------------
+  /** Gets a reference to the `invoke` module. */
+  private DataFlow::Node invoke(DataFlow::TypeTracker t) {
+    t.start() and
+    result = DataFlow::importNode("invoke")
+    or
+    exists(DataFlow::TypeTracker t2 | result = invoke(t2).track(t2, t))
+  }
+
+  /** Gets a reference to the `invoke` module. */
+  DataFlow::Node invoke() { result = invoke(DataFlow::TypeTracker::end()) }
+
+  /**
+   * Gets a reference to the attribute `attr_name` of the `invoke` module.
+   * WARNING: Only holds for a few predefined attributes.
+   */
+  private DataFlow::Node invoke_attr(DataFlow::TypeTracker t, string attr_name) {
+    attr_name in ["run", "sudo", "context", "Context", "task"] and
+    (
+      t.start() and
+      result = DataFlow::importNode("invoke." + attr_name)
+      or
+      t.startInAttr(attr_name) and
+      result = DataFlow::importNode("invoke")
+    )
+    or
+    // Due to bad performance when using normal setup with `invoke_attr(t2, attr_name).track(t2, t)`
+    // we have inlined that code and forced a join
+    exists(DataFlow::TypeTracker t2 |
+      exists(DataFlow::StepSummary summary |
+        invoke_attr_first_join(t2, attr_name, result, summary) and
+        t = t2.append(summary)
+      )
+    )
+  }
+
+  pragma[nomagic]
+  private predicate invoke_attr_first_join(
+    DataFlow::TypeTracker t2, string attr_name, DataFlow::Node res, DataFlow::StepSummary summary
+  ) {
+    DataFlow::StepSummary::step(invoke_attr(t2, attr_name), res, summary)
+  }
+
+  /**
+   * Gets a reference to the attribute `attr_name` of the `invoke` module.
+   * WARNING: Only holds for a few predefined attributes.
+   */
+  private DataFlow::Node invoke_attr(string attr_name) {
+    result = invoke_attr(DataFlow::TypeTracker::end(), attr_name)
+  }
+
+  /** Provides models for the `invoke` module. */
+  module invoke {
+    /** Gets a reference to the `invoke.context` module. */
+    DataFlow::Node context() { result = invoke_attr("context") }
+
+    /** Provides models for the `invoke.context` module */
+    module context {
+      /** Provides models for the `invoke.context.Context` class */
+      module Context {
+        /** Gets a reference to the `invoke.context.Context` class. */
+        private DataFlow::Node classRef(DataFlow::TypeTracker t) {
+          t.start() and
+          result = DataFlow::importNode("invoke.context.Context")
+          or
+          t.startInAttr("Context") and
+          result = invoke::context()
+          or
+          // handle invoke.Context alias
+          t.start() and
+          result = invoke_attr("Context")
+          or
+          exists(DataFlow::TypeTracker t2 | result = classRef(t2).track(t2, t))
+        }
+
+        /** Gets a reference to the `invoke.context.Context` class. */
+        DataFlow::Node classRef() { result = classRef(DataFlow::TypeTracker::end()) }
+
+        /** Gets a reference to an instance of `invoke.context.Context`. */
+        private DataFlow::Node instance(DataFlow::TypeTracker t) {
+          t.start() and
+          result.asCfgNode().(CallNode).getFunction() =
+            invoke::context::Context::classRef().asCfgNode()
+          or
+          t.start() and
+          exists(Function func |
+            func.getADecorator() = invoke_attr("task").asExpr() and
+            result.(DataFlow::ParameterNode).getParameter() = func.getArg(0)
+          )
+          or
+          exists(DataFlow::TypeTracker t2 | result = instance(t2).track(t2, t))
+        }
+
+        /** Gets a reference to an instance of `invoke.context.Context`. */
+        DataFlow::Node instance() { result = instance(DataFlow::TypeTracker::end()) }
+
+        /** Gets a reference to the `run` or `sudo` methods on a `invoke.context.Context` instance. */
+        private DataFlow::Node instanceRunMethods(DataFlow::TypeTracker t) {
+          t.startInAttr(["run", "sudo"]) and
+          result = invoke::context::Context::instance()
+          or
+          exists(DataFlow::TypeTracker t2 | result = instanceRunMethods(t2).track(t2, t))
+        }
+
+        /** Gets a reference to the `run` or `sudo` methods on a `invoke.context.Context` instance. */
+        DataFlow::Node instanceRunMethods() {
+          result = instanceRunMethods(DataFlow::TypeTracker::end())
+        }
+      }
+    }
+  }
+
+  /**
+   * A call to either
+   * - `invoke.run` or `invoke.sudo` functions (http://docs.pyinvoke.org/en/stable/api/__init__.html)
+   * - `run` or `sudo` methods on a `invoke.context.Context` instance (http://docs.pyinvoke.org/en/stable/api/context.html#invoke.context.Context.run)
+   */
+  private class InvokeRunCommandCall extends SystemCommandExecution::Range, DataFlow::CfgNode {
+    override CallNode node;
+
+    InvokeRunCommandCall() {
+      exists(DataFlow::Node callFunction | node.getFunction() = callFunction.asCfgNode() |
+        callFunction = invoke_attr(["run", "sudo"])
+        or
+        callFunction = invoke::context::Context::instanceRunMethods()
+      )
+    }
+
+    override DataFlow::Node getCommand() {
+      result.asCfgNode() in [node.getArg(0), node.getArgByName("command")]
+    }
+  }
+}

python/ql/src/experimental/semmle/python/frameworks/Stdlib.qll

@@ -469,4 +469,115 @@ private module Stdlib {
       result.asCfgNode() in [node.getArg(0), node.getArgByName("cmd")]
     }
   }
+
+  // ---------------------------------------------------------------------------
+  // builtins
+  // ---------------------------------------------------------------------------
+  /** Gets a reference to the `builtins` module (called `__builtin__` in Python 2). */
+  private DataFlow::Node builtins(DataFlow::TypeTracker t) {
+    t.start() and
+    result = DataFlow::importNode(["builtins", "__builtin__"])
+    or
+    exists(DataFlow::TypeTracker t2 | result = builtins(t2).track(t2, t))
+  }
+
+  /** Gets a reference to the `builtins` module. */
+  DataFlow::Node builtins() { result = builtins(DataFlow::TypeTracker::end()) }
+
+  /**
+   * Gets a reference to the attribute `attr_name` of the `builtins` module.
+   * WARNING: Only holds for a few predefined attributes.
+   */
+  private DataFlow::Node builtins_attr(DataFlow::TypeTracker t, string attr_name) {
+    attr_name in ["exec", "eval", "compile"] and
+    (
+      t.start() and
+      result = DataFlow::importNode(["builtins", "__builtin__"] + "." + attr_name)
+      or
+      t.startInAttr(attr_name) and
+      result = DataFlow::importNode(["builtins", "__builtin__"])
+      or
+      // special handling of builtins, that are in scope without any imports
+      // TODO: Take care of overrides, either `def eval: ...`, `eval = ...`, or `builtins.eval = ...`
+      t.start() and
+      exists(NameNode ref | result.asCfgNode() = ref |
+        ref.isGlobal() and
+        ref.getId() = attr_name and
+        ref.isLoad()
+      )
+    )
+    or
+    // Due to bad performance when using normal setup with `builtins_attr(t2, attr_name).track(t2, t)`
+    // we have inlined that code and forced a join
+    exists(DataFlow::TypeTracker t2 |
+      exists(DataFlow::StepSummary summary |
+        builtins_attr_first_join(t2, attr_name, result, summary) and
+        t = t2.append(summary)
+      )
+    )
+  }
+
+  pragma[nomagic]
+  private predicate builtins_attr_first_join(
+    DataFlow::TypeTracker t2, string attr_name, DataFlow::Node res, DataFlow::StepSummary summary
+  ) {
+    DataFlow::StepSummary::step(builtins_attr(t2, attr_name), res, summary)
+  }
+
+  /**
+   * Gets a reference to the attribute `attr_name` of the `builtins` module.
+   * WARNING: Only holds for a few predefined attributes.
+   */
+  private DataFlow::Node builtins_attr(string attr_name) {
+    result = builtins_attr(DataFlow::TypeTracker::end(), attr_name)
+  }
+
+  /**
+   * A call to the builtin `exec` function.
+   * See https://docs.python.org/3/library/functions.html#exec
+   */
+  private class BuiltinsExecCall extends CodeExecution::Range, DataFlow::CfgNode {
+    override CallNode node;
+
+    BuiltinsExecCall() { node.getFunction() = builtins_attr("exec").asCfgNode() }
+
+    override DataFlow::Node getCode() { result.asCfgNode() = node.getArg(0) }
+  }
+
+  /**
+   * A call to the builtin `eval` function.
+   * See https://docs.python.org/3/library/functions.html#eval
+   */
+  private class BuiltinsEvalCall extends CodeExecution::Range, DataFlow::CfgNode {
+    override CallNode node;
+
+    BuiltinsEvalCall() { node.getFunction() = builtins_attr("eval").asCfgNode() }
+
+    override DataFlow::Node getCode() { result.asCfgNode() = node.getArg(0) }
+  }
+
+  /** An additional taint step for calls to the builtin function `compile` */
+  private class BuiltinsCompileCallAdditionalTaintStep extends TaintTracking::AdditionalTaintStep {
+    override predicate step(DataFlow::Node nodeFrom, DataFlow::Node nodeTo) {
+      exists(CallNode call |
+        nodeTo.asCfgNode() = call and
+        call.getFunction() = builtins_attr("compile").asCfgNode() and
+        nodeFrom.asCfgNode() in [call.getArg(0), call.getArgByName("source")]
+      )
+    }
+  }
+}
+
+/**
+ * An exec statement (only Python 2).
+ * Se ehttps://docs.python.org/2/reference/simple_stmts.html#the-exec-statement.
+ */
+private class ExecStatement extends CodeExecution::Range {
+  ExecStatement() {
+    // since there are no DataFlow::Nodes for a Statement, we can't do anything like
+    // `this = any(Exec exec)`
+    this.asExpr() = any(Exec exec).getBody()
+  }
+
+  override DataFlow::Node getCode() { result = this }
 }