Merge pull request #16135 from sylwia-budzynska/gradio-model

Python: Add Gradio models

docs/codeql/reusables/supported-frameworks.rst

@@ -202,6 +202,7 @@ and the CodeQL library pack ``codeql/python-all`` (`changelog <https://github.co
    Flask-Admin, Web framework
    Tornado, Web framework
    Twisted, Web framework
+   Gradio, Web framework
    starlette, Asynchronous Server Gateway Interface (ASGI)
    ldap3, Lightweight Directory Access Protocol (LDAP)
    python-ldap, Lightweight Directory Access Protocol (LDAP)

python/ql/lib/semmle/python/Frameworks.qll

@@ -29,6 +29,7 @@ private import semmle.python.frameworks.FastApi
 private import semmle.python.frameworks.Flask
 private import semmle.python.frameworks.FlaskAdmin
 private import semmle.python.frameworks.FlaskSqlAlchemy
+private import semmle.python.frameworks.Gradio
 private import semmle.python.frameworks.Httpx
 private import semmle.python.frameworks.Idna
 private import semmle.python.frameworks.Invoke

python/ql/lib/semmle/python/frameworks/Gradio.qll

@@ -0,0 +1,123 @@
+/**
+ * Provides classes modeling security-relevant aspects of the `gradio` PyPI package.
+ * See https://pypi.org/project/gradio/.
+ */
+
+import python
+import semmle.python.dataflow.new.RemoteFlowSources
+import semmle.python.dataflow.new.TaintTracking
+import semmle.python.ApiGraphs
+
+/**
+ * Provides models for the `gradio` PyPI package.
+ * See https://pypi.org/project/gradio/.
+ */
+module Gradio {
+  /**
+   * The event handlers, Interface and gradio.ChatInterface classes, which take untrusted data.
+   */
+  private class GradioInput extends API::CallNode {
+    GradioInput() {
+      this =
+        API::moduleImport("gradio")
+            .getMember([
+                "Button", "Textbox", "UploadButton", "Slider", "JSON", "HTML", "Markdown", "File",
+                "AnnotatedImage", "Audio", "BarPlot", "Chatbot", "Checkbox", "CheckboxGroup",
+                "ClearButton", "Code", "ColorPicker", "Dataframe", "Dataset", "DownloadButton",
+                "Dropdown", "DuplicateButton", "FileExplorer", "Gallery", "HighlightedText",
+                "Image", "ImageEditor", "Label", "LinePlot", "LoginButton", "LogoutButton",
+                "Model3D", "Number", "ParamViewer", "Plot", "Radio", "ScatterPlot", "SimpleImage",
+                "State", "Video"
+              ])
+            .getReturn()
+            .getMember([
+                "change", "input", "click", "submit", "edit", "clear", "play", "pause", "stop",
+                "end", "start_recording", "pause_recording", "stop_recording", "focus", "blur",
+                "upload", "release", "select", "stream", "like", "load", "key_up",
+              ])
+            .getACall()
+      or
+      this = API::moduleImport("gradio").getMember(["Interface", "ChatInterface"]).getACall()
+    }
+  }
+
+  /**
+   * The `inputs` parameters in Gradio event handlers, that are lists and are sources of untrusted data.
+   * This model allows tracking each element list back to source, f.ex. `gr.Textbox(...)`.
+   */
+  private class GradioInputList extends RemoteFlowSource::Range {
+    GradioInputList() {
+      exists(GradioInput call |
+        // limit only to lists of parameters given to `inputs`.
+        (
+          (
+            call.getKeywordParameter("inputs").asSink().asCfgNode() instanceof ListNode
+            or
+            call.getParameter(1).asSink().asCfgNode() instanceof ListNode
+          ) and
+          (
+            this = call.getKeywordParameter("inputs").getASubscript().getAValueReachingSink()
+            or
+            this = call.getParameter(1).getASubscript().getAValueReachingSink()
+          )
+        )
+      )
+    }
+
+    override string getSourceType() { result = "Gradio untrusted input" }
+  }
+
+  /**
+   * The `inputs` parameters in Gradio event handlers, that are not lists and are sources of untrusted data.
+   */
+  private class GradioInputParameter extends RemoteFlowSource::Range {
+    GradioInputParameter() {
+      exists(GradioInput call |
+        this = call.getParameter(0, "fn").getParameter(_).asSource() and
+        // exclude lists of parameters given to `inputs`
+        not call.getKeywordParameter("inputs").asSink().asCfgNode() instanceof ListNode and
+        not call.getParameter(1).asSink().asCfgNode() instanceof ListNode
+      )
+    }
+
+    override string getSourceType() { result = "Gradio untrusted input" }
+  }
+
+  /**
+   * The `inputs` parameters in Gradio decorators to event handlers, that are sources of untrusted data.
+   */
+  private class GradioInputDecorator extends RemoteFlowSource::Range {
+    GradioInputDecorator() {
+      exists(GradioInput call |
+        this = call.getReturn().getACall().getParameter(0).getParameter(_).asSource()
+      )
+    }
+
+    override string getSourceType() { result = "Gradio untrusted input" }
+  }
+
+  /**
+   * Extra taint propagation for tracking `inputs` parameters in Gradio event handlers, that are lists.
+   */
+  private class ListTaintStep extends TaintTracking::AdditionalTaintStep {
+    override predicate step(DataFlow::Node nodeFrom, DataFlow::Node nodeTo) {
+      exists(GradioInput node |
+        // handle cases where there are multiple arguments passed as a list to `inputs`
+        (
+          (
+            node.getKeywordParameter("inputs").asSink().asCfgNode() instanceof ListNode
+            or
+            node.getParameter(1).asSink().asCfgNode() instanceof ListNode
+          ) and
+          exists(int i | nodeTo = node.getParameter(0, "fn").getParameter(i).asSource() |
+            nodeFrom.asCfgNode() =
+              node.getKeywordParameter("inputs").asSink().asCfgNode().(ListNode).getElement(i)
+            or
+            nodeFrom.asCfgNode() =
+              node.getParameter(1).asSink().asCfgNode().(ListNode).getElement(i)
+          )
+        )
+      )
+    }
+  }
+}

python/ql/src/change-notes/2024-04-05-gradio-models.md

@@ -0,0 +1,4 @@
+---
+category: minorAnalysis
+---
+* Added models of `gradio` PyPI package.

python/ql/test/library-tests/frameworks/gradio/source_test.expected

@@ -0,0 +1,2 @@
+testFailures
+failures

python/ql/test/library-tests/frameworks/gradio/source_test.py

@@ -0,0 +1,33 @@
+import gradio as gr
+
+
+with gr.Blocks() as demo:
+    name = gr.Textbox(label="Name")
+    output = gr.Textbox(label="Output Box")
+    # static block - not used as a source
+    static_block = gr.HTML("""
+    <div style='height: 100px; width: 800px; background-color: pink;'></div>
+    """)
+    greet_btn = gr.Button("Hello")
+
+	# decorator
+    @greet_btn.click(inputs=name, outputs=output)
+    def greet(name): # $ source=name
+        return "Hello " + name + "!"
+
+    # `click` event handler with keyword arguments
+    def greet1(name): # $ source=name
+        return "Hello " + name + "!"
+
+    greet1_btn = gr.Button("Hello")
+    greet1_btn.click(fn=greet1, inputs=name, outputs=output, api_name="greet")
+
+    # `click` event handler with positional arguments
+    def greet2(name): # $ source=name
+        return "Hello " + name + "!"
+
+    greet2_btn = gr.Button("Hello")
+    greet2_btn.click(fn=greet2, inputs=name, outputs=output, api_name="greet")
+
+
+demo.launch()

python/ql/test/library-tests/frameworks/gradio/source_test.ql

@@ -0,0 +1,20 @@
+import python
+import semmle.python.dataflow.new.RemoteFlowSources
+import TestUtilities.InlineExpectationsTest
+private import semmle.python.dataflow.new.internal.PrintNode
+
+module SourceTest implements TestSig {
+  string getARelevantTag() { result = "source" }
+
+  predicate hasActualResult(Location location, string element, string tag, string value) {
+    exists(location.getFile().getRelativePath()) and
+    exists(RemoteFlowSource rfs |
+      location = rfs.getLocation() and
+      element = rfs.toString() and
+      value = prettyNode(rfs) and
+      tag = "source"
+    )
+  }
+}
+
+import MakeTest<SourceTest>

python/ql/test/library-tests/frameworks/gradio/taint_step_test.expected

@@ -0,0 +1,26 @@
+edges
+| taint_step_test.py:5:5:5:8 | ControlFlowNode for path | taint_step_test.py:19:43:19:46 | ControlFlowNode for path | provenance |  |
+| taint_step_test.py:5:12:5:35 | ControlFlowNode for Attribute() | taint_step_test.py:5:5:5:8 | ControlFlowNode for path | provenance |  |
+| taint_step_test.py:6:5:6:8 | ControlFlowNode for file | taint_step_test.py:19:48:19:51 | ControlFlowNode for file | provenance |  |
+| taint_step_test.py:6:12:6:35 | ControlFlowNode for Attribute() | taint_step_test.py:6:5:6:8 | ControlFlowNode for file | provenance |  |
+| taint_step_test.py:11:18:11:21 | ControlFlowNode for path | taint_step_test.py:12:9:12:16 | ControlFlowNode for filepath | provenance |  |
+| taint_step_test.py:11:18:11:21 | ControlFlowNode for path | taint_step_test.py:12:9:12:16 | ControlFlowNode for filepath | provenance | AdditionalTaintStep |
+| taint_step_test.py:11:24:11:27 | ControlFlowNode for file | taint_step_test.py:12:9:12:16 | ControlFlowNode for filepath | provenance | AdditionalTaintStep |
+| taint_step_test.py:12:9:12:16 | ControlFlowNode for filepath | taint_step_test.py:13:19:13:26 | ControlFlowNode for filepath | provenance |  |
+| taint_step_test.py:19:43:19:46 | ControlFlowNode for path | taint_step_test.py:11:18:11:21 | ControlFlowNode for path | provenance | AdditionalTaintStep |
+| taint_step_test.py:19:48:19:51 | ControlFlowNode for file | taint_step_test.py:11:24:11:27 | ControlFlowNode for file | provenance | AdditionalTaintStep |
+nodes
+| taint_step_test.py:5:5:5:8 | ControlFlowNode for path | semmle.label | ControlFlowNode for path |
+| taint_step_test.py:5:12:5:35 | ControlFlowNode for Attribute() | semmle.label | ControlFlowNode for Attribute() |
+| taint_step_test.py:6:5:6:8 | ControlFlowNode for file | semmle.label | ControlFlowNode for file |
+| taint_step_test.py:6:12:6:35 | ControlFlowNode for Attribute() | semmle.label | ControlFlowNode for Attribute() |
+| taint_step_test.py:11:18:11:21 | ControlFlowNode for path | semmle.label | ControlFlowNode for path |
+| taint_step_test.py:11:24:11:27 | ControlFlowNode for file | semmle.label | ControlFlowNode for file |
+| taint_step_test.py:12:9:12:16 | ControlFlowNode for filepath | semmle.label | ControlFlowNode for filepath |
+| taint_step_test.py:13:19:13:26 | ControlFlowNode for filepath | semmle.label | ControlFlowNode for filepath |
+| taint_step_test.py:19:43:19:46 | ControlFlowNode for path | semmle.label | ControlFlowNode for path |
+| taint_step_test.py:19:48:19:51 | ControlFlowNode for file | semmle.label | ControlFlowNode for file |
+subpaths
+#select
+| taint_step_test.py:13:19:13:26 | ControlFlowNode for filepath | taint_step_test.py:5:12:5:35 | ControlFlowNode for Attribute() | taint_step_test.py:13:19:13:26 | ControlFlowNode for filepath | This path depends on a $@. | taint_step_test.py:5:12:5:35 | ControlFlowNode for Attribute() | user-provided value |
+| taint_step_test.py:13:19:13:26 | ControlFlowNode for filepath | taint_step_test.py:6:12:6:35 | ControlFlowNode for Attribute() | taint_step_test.py:13:19:13:26 | ControlFlowNode for filepath | This path depends on a $@. | taint_step_test.py:6:12:6:35 | ControlFlowNode for Attribute() | user-provided value |

python/ql/test/library-tests/frameworks/gradio/taint_step_test.py

@@ -0,0 +1,22 @@
+import gradio as gr
+import os
+
+with gr.Blocks() as demo:
+    path = gr.Textbox(label="Path") # $ source=gr.Textbox(..)
+    file = gr.Textbox(label="File") # $ source=gr.Textbox(..)
+    output = gr.Textbox(label="Output Box")
+
+
+	# path injection sink
+    def fileread(path, file):
+        filepath = os.path.join(path, file)
+        with open(filepath, "r") as f:
+                return f.read()
+
+
+    # `click` event handler with `inputs` containing a list
+    greet1_btn = gr.Button("Path for the file to display")
+    greet1_btn.click(fn=fileread, inputs=[path,file], outputs=output, api_name="fileread")
+
+
+demo.launch()

python/ql/test/library-tests/frameworks/gradio/taint_step_test.qlref

@@ -0,0 +1 @@
+Security/CWE-022/PathInjection.ql