CVE-2021-28678

Vulnerability

In Pillow versions prior to 8.2.0, the BlpImagePlugin does not properly validate that reads performed after seeking to file offsets return actual data. This flaw exists in the BLP image decoder, which was introduced in Pillow 5.1.0 [1][3]. The plugin uses self.fd.read() without checking for end-of-file or empty return values, allowing crafted BLP files to cause the decoder to run many times on zero-length data [2][4].

Exploitation

An attacker can exploit this vulnerability by providing a specially crafted BLP image file that triggers invalid file offsets. When Pillow processes such a file (e.g., via Image.open()), the decoder loops over seemingly valid data blocks that are in fact empty, leading to excessive CPU consumption [1][3]. No authentication or special privileges are required; the attack vector is network-based if the file is loaded from an untrusted source [4].

Impact

Successful exploitation results in a denial-of-service (DoS) condition. The Pillow process may become unresponsive or consume excessive CPU resources, effectively preventing legitimate image processing [1][3]. The vulnerability does not lead to code execution or information disclosure; it primarily affects availability [4].

Mitigation

Pillow version 8.2.0, released on 2021-04-01, fixes the issue by introducing a _safe_read() helper that verifies reads return the expected amount of data, and applying it to all file read operations in the BLP plugin [2][3]. Users should upgrade to Pillow 8.2.0 or later. If upgrading is not possible, avoid processing BLP files from untrusted sources [1].