Fluentd supports buffer compression (e.g., <compress gzip> or <compress zstd>) to reduce the on-disk footprint of buffered data. It was discovered that the previous mitigation for decompression bombs (CVE-2026-44160) is incomplete.

While the decompression_size_limit introduced in the previous fix is properly enforced for string-based data reads, the streaming IO decompression path (specifically Buffer::Chunk#open and #write_to) bypasses this limit entirely. If buffer compression is enabled, an attacker can send highly compressible payloads. These payloads bypass buffer chunk limits because they are evaluated based on their tiny compressed size. However, when the chunk is flushed to an output IO (such as a file, socket, or standard output) or triggers the automatic error-backup path (backup_chunk), it expands to an unbounded size without any limitation.

Impact

This vulnerability allows for an unauthenticated remote Denial of Service (DoS). By injecting highly repetitive, highly compressible data into a Fluentd instance with buffer compression enabled, an attacker can cause massive data amplification during the buffer flush phase. This results in severe memory exhaustion (OOM) or total disk space depletion, crashing the Fluentd worker and disrupting log routing.