Pillow: PDF unbounded recursion in page tree and trailer parsing (CVE-2026-112895)

Pillow, a popular Python imaging library, contains vulnerabilities that allow an attacker to trigger unbounded recursion during the parsing of malformed PDF files. When an application utilizes Image.save with append=True, the library invokes PdfParser to process the existing PDF structure. Two specific areas of the parser lack necessary bounds: linearize_page_tree follows /Kids entries without tracking visited nodes or enforcing a depth limit, and read_prev_trailer permits long chains of unique /Prev offsets.

These vulnerabilities serve as an incomplete fix for CVE-2026-42310. An attacker can craft malicious PDF files containing cyclic page trees or extensively deep /Prev chains to drive the parsing process into high CPU and memory utilization, leading to application crashes or denial of service. The vulnerability affects all Pillow versions from 5.1.0 up to, but not including, 13.0.0. Defenders should note this impacts any service that processes user-supplied PDFs for operations such as stamping, watermarking, or merging.

Impact

A successful exploit causes denial of service by exhausting stack space, CPU, and memory resources. Any service performing automated processing of user-provided PDFs - common in document management, e-signature, or image processing pipelines - is susceptible. The impact is significant for services where availability is critical and where the processing component shares memory or CPU resources with other application functions.

Recommendation

• Upgrade the Pillow library to version 13.0.0 or later immediately to address CVE-2026-112895.
• Audit application code to identify endpoints that invoke Image.save(..., format="PDF", append=True) on untrusted user input.
• Implement input validation or limit file sizes and complexity for incoming PDF documents before passing them to the Pillow library.
• If upgrading is not immediately possible, implement a file-parsing timeout for any process handling PDF manipulation to mitigate the impact of CPU and memory exhaustion.