Skip to content

BIP70 payment requests `?r=` field supports `file://` URIs, allowing attacker to trick victim machine to `open()` arbitrary file

Moderate
SomberNight published GHSA-4fh4-hx35-r355 Jun 7, 2022

Package

electrum (-)

Affected versions

2.1<=x<4.2.2

Patched versions

4.2.2

Description

Impact

In BIP70 payment requests, Electrum allows the ?r= field to contain file:// URIs (besides http(s)).
The ?r= field can contain an arbitrary file URI, chosen by an attacker.

A malicious merchant can provide a BIP70 payment request in the form of a QR code or text, which the victim user would then scan or copy-paste, as part of the payment flow. Electrum would then see the file URI, and try to open the file in read mode and read it. If the read succeeds, the data is parsed using protobuf.

Specifically regarding the QR code vector, note that Electrum starts the BIP70 flow as soon as a QR code is scanned, without giving a chance to the user to review the content of the decoded QR code.

The file URI support was originally added for local dev testing, with the implicit assumption that it is safe to open files on the local filesystem in read-only mode. This assumption is incorrect.

On Linux/macOS, e.g. trying to read /dev/zero results in a DOS attack, where the application would run out-of-memory and get killed.

On Windows, paths can be crafted that correspond to network requests, for example initiating an SMB connection. In particular, it seems that it might be possible for an attacker located in the same "trusted" Local Area Network as the victim, after getting the victim to scan a malicious QR code, to have the victim's computer initiate a same-LAN SMB connection to the attacker's computer, and to capture an authentication token. That authentication token could later be used to initiate an offline brute-force attack against the user's Windows account password.

Patches

We have removed the file URI support in commit b247aa5ffef0f9ef000772fcf9cd9c7141abded8.
Electrum version 4.2.2 contains the fix.

Credits

We thank the Unciphered team, and specifically Frank Davidson <[email protected]> for responsibly disclosing this issue to us.

Severity

Moderate

CVE ID

No known CVE

Weaknesses

No CWEs