Free trial The cheapest email marketing platform: try Moosend free for 30 days, no credit card. Don't miss your seat. The cheapest email marketing platform Start free trial →

Rspamd 4.3.0 Fixes Worker Crashes and Closes ARC Tricks That Softened DMARC Rejects

The open-source spam filter's 4.3.0 release fixes crash bugs, a milter proxy use-after-free and forged ARC and Authentication-Results paths to dmarc=pass, and it changes SPF and DKIM scoring on purpose.

Alaa Touil
Email infrastructure specialist, 8+ years in SMTP and deliverability
Event Nov 13, 2026 6 min read Nov 16, 2026 56 views
Independent email news, no paywall.Add as preferred sourceBuy us a coffeeRSS
Rspamd 4.3.0 Fixes Worker Crashes and Closes ARC Tricks That Softened DMARC Rejects (SMTPedia News)

Rspamd, the open-source spam filter that sits in front of many Postfix, Exim and other mail servers, has shipped version 4.3.0 with a long list of security fixes. According to the project's official changelog, released on 13 November 2026, the update closes bugs that let an unauthenticated client crash Rspamd workers, fixes a use-after-free in the milter proxy that has been present since 4.0.0, and stops forged ARC and Authentication-Results headers from turning a DMARC reject into a pass.

The release also changes how SPF, DKIM and phishing URLs are scored on purpose. Mail that passed or failed before can get a different verdict after the upgrade, so administrators are asked to review local scores and rules before rolling it out. The project recommends the upgrade for all users, and says milter deployments on 4.0.0 or later should prioritise it.

Key facts

  • Release: Rspamd 4.3.0, a major release dated 13 November 2026; the previous version was 4.2.3 (21 October 2026)
  • Worker crashes: a single unauthenticated HTTP request with a header line without a colon could abort a worker; an empty Deliver-To value could crash workers again and again
  • Milter proxy: a use-after-free affects 4.0.0 up to and including 4.2.3; version 3.14 is not affected
  • DMARC bypass paths closed: appended ARC sets could soften a DMARC reject, and unescaped MAIL FROM or HELO text could inject a fake dmarc=pass into Authentication-Results
  • Verdicts may change: SPF terms are now evaluated in record order, RSA DKIM keys under 1024 bits fail, and DKIM d= domains are lowercased
  • CVE identifiers: none listed in the changelog

The security fixes

Workers that one request could stop

The bundled HTTP parser accepted a line break as the end of a header name. A header line with no colon and no value could make a length counter wrap around and abort the worker, and the changelog says one unauthenticated request was enough. Rspamd now rejects such lines, as nginx and the upstream http-parser do. Separately, an empty Deliver-To header or query value triggered an assertion that aborted the worker, which meant any client able to reach the worker could keep it crashing. An empty value now keeps the previous one.

Other fixes in the same area cover recipient indexes in the X-Rspamd-Rcpt-Esmtp-Args header, which were read without range checks (negative values wrote before an array, large ones grew it to billions of slots), DKIM keys whose real type did not match the declared k= tag, which aborted the worker instead of failing, and MIME parts made of a single line break, which stopped parsing of the whole message.

The milter proxy use-after-free

The milter proxy is the part of Rspamd that connects it to the MTA. From 4.0.0 onward, when the proxy could not pick an upstream for a retry, it released the session of the MTA connection one time too many. The changelog says this always happens with a single upstream, which is the default setup (hosts = "localhost") with the default token bucket rotation. The worker then either crashed with a segmentation fault or stopped accepting connections; in the second case the listen backlog filled up and, because heartbeats_loss_max defaults to 0, the worker could stay stuck indefinitely. In 4.3.0 a failed message gets exactly one TEMPFAIL and a fresh session. A second end-of-body command sent before the first reply, which caused another use-after-free, is now treated as a protocol error.

ARC and DMARC trust

This is the part that matters most for deliverability. Rspamd can relax a DMARC failure when a trusted forwarder, such as a mailing list, vouches for the original result through ARC. Until now, the seal of a whitelisted forwarder was trusted at any position in the chain, which let an attacker append their own ARC set and soften a DMARC reject. In 4.3.0 a forwarder is trusted only if every later ARC instance is also whitelisted and verified, the newest ARC-Message-Signature must verify on its own, and the DMARC adjustment requires the forwarder's own dmarc=pass with a matching header.from. ARC chains with more than 50 sets now fail early.

Two parsing flaws are also fixed. An encoded-word hidden in a comment could add a dmarc=pass when arc.lua decoded the header before parsing it; results are now read from the raw header, and ambiguous DMARC results are not trusted. And MAIL FROM and HELO values were written unescaped into a comment of the Authentication-Results header, so a sender could close the comment and insert a result such as dmarc=pass. Comments are now escaped.

HTTP client fixes

The Lua HTTP client and its connection pool also get fixes: the pool key now includes TLS verification and the peer address, a plaintext reply to an encrypted request is no longer accepted, and the forbid_local option now blocks 0.0.0.0, :: and more special-purpose IPv4 and IPv6 ranges.

Changes that can move verdicts

Beyond the fixes, 4.3.0 deliberately changes results. The changelog lists, among others:

  • SPF: terms are evaluated in record order and a redirect applies after all mechanisms. A record such as -ip4:1.2.3.4 +ip4:1.2.3.0/24 -all no longer passes 1.2.3.4, and a/24/64 is no longer valid syntax.
  • DKIM: signatures fail when a known tag appears twice, when l=0 is used on a message with a body, when the RSA key is shorter than 1024 bits, or when the key type does not match k=. The d= domain is lowercased, so maps with mixed-case signing domains need lowercase entries.
  • Phishing: mismatches between a link's displayed text and its real destination are now scored separately from phishing.
  • Symbol options: the asn, reputation, rbl and dmarc modules now emit each value as a separate option, which can break scripts that parse the old comma-joined values.
  • SMTPUTF8: new address checks follow the IETF draft on SMTPUTF8 syntax; the project notes the new ASCII address symbols are weak evidence.

What it means for senders

What changes

Rspamd 4.3.0 closes crash bugs and DMARC bypass paths, and it scores SPF and DKIM more strictly, so mail with weak DKIM keys or badly ordered SPF records can lose points at receivers that upgrade.

Who is affected

Administrators running Rspamd, especially with the milter proxy on 4.0.0 to 4.2.3, and any sender whose mail reaches mailbox hosts or companies that filter with Rspamd.

When

Released on 13 November 2026. The project recommends upgrading now, with priority for milter deployments.

Action checklist

  • Upgrade to 4.3.0 first on servers that use the milter proxy on 4.0.0 or later.
  • Restrict access to worker ports so that only your MTA and trusted hosts can reach them.
  • Test local rules before rollout: review custom phishing scores, scripts that parse symbol options and maps that hold mixed-case DKIM d= values.
  • Rotate DKIM keys shorter than 1024 bits to 2048-bit RSA keys, since those signatures now fail.
  • Audit your SPF record order so that the first matching term gives the result you intend.
  • Check your ARC forwarder whitelist and keep only mailing lists and forwarders you actually rely on.

Sources

  1. Rspamd changelog: 4.3.0, 13 November 2026 (primary source)
  2. Rspamd changelog index

SMTPedia News follows our editorial standards. Spotted an error? See how we correct it.

Get SMTPedia news first on Google

Add us as a preferred source to see our stories in Top Stories and Discover.

Add as preferred source

Useful story?

SMTPedia news is free and independent. A coffee helps us keep covering what changes for senders.

Buy SMTPedia a coffee

Newsletter

Get the next rule change before it bounces your mail