Habr•August 14, 2026•🇷🇺Translated from Russian

Nginx Leaves Dynamic Tables Unused in HTTP/2 and HTTP/3 Responses

Nginx applies header compression asymmetrically in both HTTP/2 and HTTP/3. While clients compress repeated headers after the first transmission, the server always emits full literals, keeping response header sizes constant regardless of repetition count.

Measurements on a minimal stand with four identical requests show server responses locked at 131 bytes under HTTP/3 and 129 bytes under HTTP/2. Client requests, by contrast, drop from 246 bytes to 8 bytes once the dynamic table is populated.

Inspection of release-1.31.3 reveals the root cause inside ngx_http_v3_filter_module.c: all twenty-two invocations of ngx_http_v3_encode_field_ri and ngx_http_v3_encode_field_lri pass a zero value for the dynamic flag, forcing every field to be encoded as a literal. The field-section prefix is likewise emitted with insert_count and delta_base set to zero. No server-side encoder stream of type 0x02 is ever opened.

The identical limitation exists in the HPACK implementation. The HTTP/2 filter module contains zero references to ngx_http_v2_hpack_t and explicitly sends a Dynamic Table Size Update of zero on the first HEADERS frame, disabling client-side table usage for server responses.

The receiving path, however, is fully functional. ngx_http_v3_table.c implements insertion, eviction, blocked-stream tracking and the default 4096-byte table capacity announced in the SETTINGS frame. Debug logs confirm that client-supplied headers are correctly inserted and later referenced by index.

The asymmetry is explicitly allowed by RFC 9204 §4.2, yet it eliminates the compression gains that motivated the introduction of QPACK and HPACK. The unused encoder code path also contributed to a use-after-free vulnerability (CVSS 9.2) in the QPACK decoder earlier this year.

Related articles

Habr•Vulnerabilities & Exploits

Unemployed AppSec Engineer Releases fe-zk Library to Catch Soundness Bugs in Zero-Knowledge Verifiers

A Russian developer identifying as 0xItsss, currently unemployed after months of rejections for AppSec, Rust, and backend roles, has published fe-zk, a verifier library and generator for the Fe smart-contract language. The project implements Groth16, PLONK, halo2-KZG, KZG openings, and FRI/STARK verifiers for Plonky3, all generated from existing verifying keys with built-in tests. The author highlights systemic soundness failures in on-chain verifiers, including missing public-input range checks, curve subgroup validation, Fiat-Shamir transcript coverage, and precompile return-length checks, citing the Frozen Heart class of transcript attacks. By leveraging Fe’s traits, const generics, static_assert, and private fields, fe-zk turns many of these rules into compile-time errors rather than auditor findings. The library ships with verified integrations for Semaphore v4, Tornado Cash mainnet proofs, and a zk Game of Life demo, plus BLS12-381 precompile wrappers from EIP-2537.

Habr•Vulnerabilities & Exploits

Chrome Researcher Edits Own VRP Reward to $1.3M in Bug Tracker Prank, Triggering AI Misinformation Spread

Security researcher Lyra Rebane, who has disclosed 14 CVEs in Chrome, reported a logical flaw in the Background Fetch API in 2022 that could enable browser-based botnets. Google awarded $1,000 after the issue was fixed in February 2026, but the researcher later received verifier rights in the Buganizer tracker. She jokingly set the vrp-reward field to $1,333,337, the largest amount in Google VRP history. The tweet about the edit went viral and was picked up by AI news generators and Google search, which began listing her as a millionaire. Community Notes initially flagged the post before an experimental AI note incorrectly claimed the screenshot was edited. The incident highlights how social media amplification and AI systems can turn a harmless joke into persistent misinformation across search engines including Kagi, DuckDuckGo, and Bing.

Securitylab•Vulnerabilities & Exploits

Java4Shell (CVE-2026-114938) Vulnerability Allows Remote Memory Corruption and Code Execution in OpenJDK

Researchers at Calif disclosed Java4Shell (CVE-2026-114938) on 17 November 2026, a critical flaw in the HotSpot JVM that enables memory corruption via a malicious Base64 payload. The issue affects multiple OpenJDK distributions including Oracle Java, Amazon Corretto, Eclipse Temurin and Azul Zulu when running on modern Intel Xeon and AMD EPYC processors supporting AVX-512 VBMI instructions. Rated 9.2 on CVSS 4.0, the vulnerability can lead to data theft and arbitrary code execution without authentication in several widely used products such as Keycloak, Jenkins and Atlassian Jira. Public exploit code and reconnaissance activity were observed within days, although no confirmed successful breaches have been reported yet. Patches are available for supported branches, while older intermediate versions require migration to maintained releases.

Habr•Vulnerabilities & Exploits

F6 Investigates First In-the-Wild Exploitation of Critical CVE-2026-76466 in Cisco Expressway

Russian cybersecurity firm F6 has documented the first confirmed exploitation of the critical CVE-2026-76466 vulnerability (CVSS 9.8) affecting Cisco Expressway devices. Attackers targeted a Russian company and quickly adapted a public proof-of-concept exploit to gain root-level remote command execution on version X15.5.3. The flaw resides in the Erlang distribution authentication mechanism on TCP port 4372, allowing unauthenticated command execution via a hardcoded cookie. After initial access, the threat actors deployed over 100 custom Python scripts for network reconnaissance, lateral movement, and data collection, including tools that abused WinRM, SMB, LDAP, and SQL Server protocols. F6 observed modifications to the original PoC that enabled arbitrary command execution and result retrieval, along with auxiliary scripts for TCP relaying and repeated connection attempts. The incident highlights how quickly public exploits can be weaponized against perimeter devices that bridge external and internal networks.