• v8.6.8 facfb91f16

    v8.6.8 Stable

    tech released this 2026-11-20 21:26:17 +00:00 | 6 commits to main since this release

    • Security: uploaded file content is now scanned by WAF PHP injection rules (crs-933xxx, phpi-*, eval-based rules), so webshells embedded inside files with benign names (e.g. "image.jpg") are detected even when a plugin moves the file directly without passing through the Upload Guard.
    • Security: removed User-Agent and Referer header exclusions from all custom WAF rules covering SQL injection, remote code execution, PHP injection, path traversal, XXE, and SSRF — these attack vectors are regularly delivered through request headers.
    • Security: WAF live-corpus scan window expanded to cover values up to 64 KB without a gap — previously values over 16 KB had an unscanned middle band that could be exploited by padding an attack payload with 16 KB of harmless content.
    • Security: the WAF no longer fails open. A value that cannot be fully evaluated within its CPU budget now raises a high-severity "evasion" signal and is blocked for untrusted users; verified authors with the unfiltered_html capability may still save large content. Padding, repeated units and oversized values can no longer be used to slip an attack past the rules.
    • Performance/Security: the slow-regex (ReDoS) rules in the CRS-derived set (crs-941160, the 9322xx command-injection family, 942190, 942230, 934200 and others) now run in linear time, with and without PCRE JIT. Worst-case CPU per request is bounded (about 2 s with JIT, 3.5 s without) including cookies and headers, which now have a size budget too.
    • Security: uploaded files are inspected by kind instead of as text. Images, PDFs, archives and office files are no longer blocked or banned by text rules; PHP code hidden in a media file, spoofed archives/executables and PHP-style or double extensions (including chunked uploads that name the destination file in a POST field) are still blocked. SVG uploads remain blocked on purpose.
    • Security: credentials that only look valid (forged Basic header or logged-in cookie) no longer open the REST "verified author" hold; application passwords are verified server-side before any exception applies.
    • Fix (multisite): network deactivation, per-site deactivation and uninstall now handle every site of the network and clean up all plugin data; deactivation stops enforcement.
    • Fix: the MFA browser-binding cookies work on sites with a login URL outside the home path, with visible redirects and clear failure notices.
    • Fix: forwarded headers from private addresses are never trusted automatically; a mod_remoteip diagnostic was added.
    • Privacy: GeoIP lookup is opt-in; the self-update check discloses what it sends.
    • Vulnerability baseline: the free baseline is now built from NVD data with provenance.
    Downloads