mlfi_eoh: emit dkim=permerror for malformed DKIM-Signature (fixes #194)

[thegushi]

Summary

• When dkim_header() returns DKIM_STAT_SYNTAX for a malformed DKIM-Signature, mlfi_eoh() was setting ms and returning early, bypassing mlfi_eom() entirely. With AlwaysAddARHeader or SoftwareHeader enabled, the message was accepted with no Authentication-Results header — making it indistinguishable from mail that was never processed by OpenDKIM.
• RFC 8601 §2.7.1 defines dkim=permerror for exactly this case.
• The fix routes syntax errors through the existing mctx_headeronly / DKIMF_STATUS_BADFORMAT path that mlfi_eom() already handles correctly, rather than short-circuiting past it.

How it works

When dkim_header() returns DKIM_STAT_SYNTAX, instead of setting ms (which causes mlfi_eoh() to return early), the fix sets mctx_headeronly, mctx_addheader, and mctx_status=DKIMF_STATUS_BADFORMAT. mlfi_eoh() returns SMFIS_CONTINUE; mlfi_body() skips the body via the existing mctx_headeronly check; mlfi_eom() reaches its own mctx_headeronly branch which emits dkim=permerror and accepts.

A guard is also added to the DKIM_STAT_NOSIG branch of the dkim_eoh() result switch so it does not overwrite a BADFORMAT status already set from the header loop. (A handle with a bad DKIM-Signature produces NOSIG from dkim_eoh_verify() once the bad sig's set_bad flag is set.)

Test

t-test208 confirms the library behaviour the fix relies on: dkim_header() returns DKIM_STAT_SYNTAX for a malformed DKIM-Signature, the handle remains usable, and dkim_eoh() returns DKIM_STAT_NOSIG. Verified on FreeBSD.

Test plan

• Build and run make check — t-test208 should pass
• Manual test: send a message with DKIM-Signature: eeee through a milter instance configured with AlwaysAddARHeader yes; confirm Authentication-Results header contains dkim=permerror