SAPP

SAPP stands for Static Analysis Post Processor. SAPP takes the raw results of Pysa and Mariana Trench, and makes them explorable both through a command-line interface and a web UI.

SAPP is also available on GitHub Marketplace as a GitHub Action

Installation

To run SAPP, you will need Python 3.8 or later. SAPP can be installed through PyPI with pip install fb-sapp.

Getting Started

This guide assumes that you have results from a Pysa run saved in an ~/example directory. If you are new to Pysa, you can follow this tutorial to get started.

Processing the Results

The postprocessing will translate the raw output containing models for every analyzed function into a format that is more suitable for exploration.

[~/example]$ sapp --database-name sapp.db analyze taint-output.json

After the results have been processed we can now explore them through the UI and a command-line interface. We will briefly look at both of those methods here.

Web Interface

Start the web interface with

[~/example]$ sapp --database-name sapp.db server --source-directory=<WHERE YOUR CODE LIVES>

and visit http://localhost:13337 in your browser (note: the URL displayed in the code output currently will not work). You will be presented with a list of issues that provide access to example traces.

Command-Line Interface

The same information can be accessed through the command-line interface:

[~/example]$ sapp --database-name sapp.db explore

This will launch a custom IPython interface that is connected to the sqlite file. In this mode, you can dig into the issues that Pyre surfaces. Following is an example of how to use the various commands.

Start by listing all known issues:

==========================================================
Interactive issue exploration. Type 'help' for help.
==========================================================

[ run 1 ]
>>> issues
Issue 1
    Code: 5001
 Message: Possible shell injection Data from [UserControlled] source(s) may reach [RemoteCodeExecution] sink(s)
Callable: source.convert
 Sources: input
   Sinks: os.system
Location: source.py:9|22|32
Found 1 issues with run_id 1.

As expected, we have 1 issue. To select it:

[ run 1 ]
>>> issue 1
Set issue to 1.

Issue 1
    Code: 5001
 Message: Possible shell injection Data from [UserControlled] source(s) may reach [RemoteCodeExecution] sink(s)
Callable: source.convert
 Sources: input
   Sinks: os.system
Location: source.py:9|22|32

View how the data flows from source to sink:

[ run 1 > issue 1 > source.convert ]
>>> trace
     # ⎇  [callable]       [port]      [location]
     1    leaf             source      source.py:8|17|22
 --> 2    source.convert   root        source.py:9|22|32
     3    source.get_image formal(url) source.py:9|22|32
     4    leaf             sink        source.py:5|21|28

Move to the next callable:

[ run 1 > issue 1 > source.convert ]
>>> n
     # ⎇  [callable]       [port]      [location]
     1    leaf             source      source.py:8|17|22
     2    source.convert   root        source.py:9|22|32
 --> 3    source.get_image formal(url) source.py:9|22|32
     4    leaf             sink        source.py:5|21|28

Show the source code at that callable:

[ run 1 > issue 1 > source.get_image ]
>>> list
In source.convert [source.py:9|22|32]
     4      command = "wget -q https:{}".format(url)
     5      return os.system(command)
     6
     7  def convert() -> None:
     8      image_link = input("image link: ")
 --> 9      image = get_image(image_link)
                              ^^^^^^^^^^

Move to the next callable and show source code:

[ run 1 > issue 1 > source.get_image ]
>>> n
     # ⎇  [callable]       [port]      [location]
     1    leaf             source      source.py:8|17|22
     2    source.convert   root        source.py:9|22|32
     3    source.get_image formal(url) source.py:9|22|32
 --> 4    leaf             sink        source.py:5|21|28

[ run 1 > issue 1 > leaf ]
>>> list
In source.get_image [source.py:5|21|28]
     1  import os
     2
     3  def get_image(url: str) -> int:
     4      command = "wget -q https:{}".format(url)
 --> 5      return os.system(command)
                             ^^^^^^^
     6
     7  def convert() -> None:
     8      image_link = input("image link: ")
     9      image = get_image(image_link)

Jump to the first callable and show source code:

[ run 1 > issue 1 > leaf ]
>>> jump 1
     # ⎇  [callable]       [port]      [location]
 --> 1    leaf             source      source.py:8|17|22
     2    source.convert   root        source.py:9|22|32
     3    source.get_image formal(url) source.py:9|22|32
     4    leaf             sink        source.py:5|21|28

[ run 1 > issue 1 > leaf ]
>>> list
In source.convert [source.py:8|17|22]
     3  def get_image(url: str) -> int:
     4      command = "wget -q https:{}".format(url)
     5      return os.system(command)
     6
     7  def convert() -> None:
 --> 8      image_link = input("image link: ")
                         ^^^^^
     9      image = get_image(image_link)

Help

You can refer to the help command to get more information about available commands in the command-line interface.

$ sapp --help
Usage: sapp [OPTIONS] COMMAND [ARGS]...

Options:
  -v, --verbosity LVL             Either CRITICAL, ERROR, WARNING, INFO or
                                  DEBUG
  -r, --repository DIRECTORY      Root of the repository (regardless of the
                                  directory analyzed)
  --database-name, --dbname FILE
  --database-engine, --database [sqlite|memory]
                                  database engine to use
  --tool [pysa|mariana-trench]    tool the data is coming from
  -h, --help                      Show this message and exit.

Commands:
  analyze  parse static analysis output and save to disk
  explore  interactive exploration of issues
  filter
  lint     Output DB models in a lint-friendly format
  server   backend flask server for exploration of issues
  update

Terminology

A single SAPP database can keep track of more than just a single run. This opens up the possibility of reasoning about newly introduced issues in a codebase.

Every invocation of

[~/example]$ sapp --database-name sapp.db analyze taint-output.json

will add a single run to the database. An issue can exist over multiple runs (we typically call the issue in a single run an instance). You can select a run from the web UI and look at all the instances of that run. You can also choose to only show the instances of issues that are newly introduced in this run in the filter menu.

Each instance consists of a data flow from a particular source kind (e.g. user-controlled input) into a callable (i.e. a function or method), and a data flow from that callable into a particular sink kind (e.g. RCE).

Note: the data can come from different sources of the same kind and flow into different sinks of the same kind. The traces view of a single instance represents a multitude of traces, not just a single trace.

Filters

SAPP filters are used to include/exclude which issues are shown to you by the issue properties you choose. Filters are useful to remove noise from the output from your static analysis tool, so you can focus on the particular properties of issues you care about.

SAPP functionality can be accessed through the web interface or a subcommand of sapp filter.

File Format

A filter is required to have a name and at least one other key, excluding description. Filters can be stored as JSON in the following format:

{
  "name": "Name of filter",
  "description": "Description for the filter",
  "features": [
    {
      "mode": "all of",
      "features": ["via:feature1", "feature2"]
    },
    {
      "mode": "any of",
      "features": ["always-via:feature3"]
    },
    {
      "mode": "none of",
      "features": ["type:feature5"]
    }
  ],
  "codes": [5005],
  "paths": ["filename.py"],
  "callables": ["main.function_name"],
  "traceLengthFromSources": [0, 3],
  "traceLengthToSinks": [0, 5],
  "is_new_issue": false
}

You can find some example filters to reference in the pyre-check repo

Importing filters

You can import a filter from a file by running:

[~/example]$ sapp --database-name sapp.db filter import filter-filename.json

You can also import all filters within a directory by running:

[~/example]$ sapp --database-name sapp.db filter import path/to/list_of_filters

Exporting filters

You can view a filter in a SAPP DB by running:

[~/example]$ sapp --database-name sapp.db filter export "filter name"

You can export a filter from a SAPP DB to a file by running:

[~/example]$ sapp --database-name sapp.db filter export "filter name" --output-path /path/to/filename.json

Deleting filters

You can delete filters by name with:

[~/example]$ sapp --database-name sapp.db filter delete "filter name 1" "filter name 2" "filter name 3"

Filtering list of issues

You can apply a filter to a list of issues by run number. For example, the following command will show you a list of issues after applying example-filter to run 1:

[~/example]$ sapp --database-name sapp.db filter issues 1 example-filter.json

You can also apply a list of filters to a single list of issues by run number. SAPP will apply each filter individually from the directory you specify to the list of issues and merge results into a single list of issues to show you. For example, the following command will show you a list of issues after applying every filter in list_of_filters to run 1:

[~/example]$ sapp --database-name sapp.db filter issues 1 path/to/list_of_filters

SARIF Output

You can get the output of a filtered run in SARIF by first storing warning codes information from the static analysis tool in SAPP:

sapp --database-name sapp.db update warning-codes taint-metadata.json

Then running sapp filter issues with --output-format=sarif:

sapp --database-name sapp.db filter issues 1 path/to/list_of_filters --output-format sarif

Development Environment Setup

Start by cloning the repo and setting up a virtual environment:

$ git clone git@github.com:facebook/sapp.git && cd sapp
$ python3 -m venv ~/.venvs/sapp
$ source ~/.venvs/sapp/bin/activate
(sapp) $ pip3 install -r requirements.txt

Run the flask server in debug mode:

(sapp) $ python3 -m sapp.cli server --debug

Parse static analysis output and save to disk:

(sapp) $ python3 -m sapp.cli analyze taint-output.json

Installing dependencies for frontend:

(sapp) $ cd sapp/ui/frontend && npm install

To run SAPP with hot reloading of the Web UI, you need have the frontend and backend running simultaneously. In a production environment, the frontend application is compiled and served directly by the backend exposed on port 13337. But in a development environment, the frontend runs in port 3000 by default if the PORT environment variable is not set and the backend runs in port 13337. You can indicate to SAPP to run in the development environment with the debug flag.

Run the flask server and react app in development mode:

(sapp) $ python3 -m sapp.cli server --debug
(sapp) $ cd sapp/ui/frontend && npm run-script start

Then visit http://localhost:3000 (or http://<HOST>:<PORT> if you have set the HOST and/or PORT environment variable).

Partial Flows

SAPP supports the concept of partial flows, which helps with triaging the issues more easily. In particular, supporting partial flows allows to filter out issues from the generic rules if they exist in the more specific one. When we have two rules, one of the form (SourceA -> SinkB), and a second one of the form (SourceA -> TransformB -> SinkC), we can associate SinkB and TransformB, and add breadcrumbs (such as some-feature) to all instances of the SourceA -> SinkB rule when a corresponding SourceA -> TransformB -> SinkC rule exists. These breadcrumbs allow us to filter away these issues, improving triage SNR.

SAPP supports partial flows by modifying the static analysis outputs (based on the configurations from those tools) before writing them into the database. To leverage this capability in SAPP, you can provide the following configurations in the metadata file from the static analysis outputs:

"partial_flows": [
  {
    "full_issue_code": 4101,
    "partial_issue_code": 4100,
    "full_issue_transform": "TransformB",
    "is_prefix_flow": true,
    "feature": "some_feature"
  }
]

Now we explain the meaning of the above configurations:

• full_issue_code: The rule code of the full issue, such as SourceA -> TransformB -> SinkC from the above example.
• partial_issue_code: The rule code of the partial issue, such as SourceA -> SinkB from the above example.
• full_issue_transform: The transform of the full issue, such as TransformB from the above example.
• is_prefix_flow: Whether the partial flow is a prefix flow or a suffix flow. In the above example, since the partial issue is a prefix of the full issue, we set this field to true.
• feature: The feature or breadcrumb to add to an issue, such as some-feature from the above example.

License

SAPP is licensed under the MIT license.