Picture the report. A Firefox user says your site shows a certificate warning. You open it in your browser and it is fine. You check the server and nothing has changed in months, except that the certificate renewed itself last week, which it does every few months without anybody looking.
Last week Mozilla shipped, and then reversed, a change that makes this report possible for real sites. The interesting part is not the incident. It is the mechanism, because it means a certificate can be refused for its birthday.
What happened
DigiCert's transition plan asked for its Trusted Root G4 to stop being trusted for websites. Mozilla made the change in NSS, the certificate library Firefox uses, with the note "Changes made in NSS 3.134 and FF 162". It then received "preliminary indications of TLS connection failures" on the site of a German bank, DKB, and on 13 November it changed course.
The commit that went in instead is short. Here is the relevant part of the diff to certdata.txt:
-CKA_NSS_SERVER_DISTRUST_AFTER CK_BBOOL CK_FALSE
+# For Server Distrust After: Mon Sep 14 23:59:59 2026
+CKA_NSS_SERVER_DISTRUST_AFTER MULTILINE_OCTAL
+\062\066\060\071\061\064\062\063\065\071\065\071\132
+END
-CKA_TRUST_SERVER_AUTH CK_TRUST CKT_NSS_MUST_VERIFY_TRUST
+CKA_TRUST_SERVER_AUTH CK_TRUST CKT_NSS_TRUSTED_DELEGATOR
The root is trusted for websites again, with a date attached. Mozilla's reviewer counted 148 unexpired certificates still chaining to that root, which is why a hard cut broke things and a date did not. Firefox 162 is two releases ahead of the 160 that shipped last week, so release users never saw the broken version.
The line that decides
A distrust-after date does not mean "stop trusting this root after this day". It means something narrower, and the narrowness is the hazard. In Firefox's NSSCertDBTrustDomain.cpp the check ends like this:
Time distrustAfterTime =
mozilla::pkix::TimeFromEpochInSeconds(distrustAfter / PR_USEC_PER_SEC);
distrustAfterTimeOut.emplace(distrustAfterTime);
if (endEntityNotBefore <= distrustAfterTime) {
isDistrusted = false;
}
endEntityNotBefore is the issue date of your certificate, the leaf, not the root's or the intermediate's, and not today's date. A certificate issued before the cutoff stays trusted until it expires. A certificate issued after it, from the same CA, through the same intermediate, for the same domain, is refused.
So on the day your renewal runs, nothing about your server changes except one date in one field, and Firefox's answer flips.
It is not easy to talk your way round either. In MultiLogCTVerifier.cpp, the Certificate Transparency timestamps a certificate carries are checked against the same date:
// If the root has a distrustAfter time, ensure that the SCT's timestamp is
// not after that time.
if (distrustAfterTime.isSome() && sctTime > distrustAfterTime.value()) {
result.sctsWithDistrustedTimestamps++;
return Success;
}
The logs' own clock is consulted, not just the date the CA wrote on the certificate.
Why this is invisible from your desk
Every check you are likely to run first says the site is fine.
Your own browser may not be Firefox, and each browser keeps its own list of trusted roots with its own rules attached. A monitoring probe or a script runs on whatever trust store its own runtime ships with. Your previous certificate, the one you tested when you set this up, was issued before the cutoff. None of those can see a rule that depends on the issue date of a certificate you have not looked at since it renewed.
And the report itself will not help. "Firefox says the connection is not secure" is the same sentence for an expired certificate, a wrong clock on their laptop, a corporate proxy, and this. The difference is in which root the chain ends at and when the leaf was issued, and the person reporting it has no reason to know either.
Checking your own chain
Two questions, and both have one-line answers:
# Which root does the chain end at? Read the last issuer.
openssl s_client -connect example.com:443 -servername example.com -showcerts </dev/null 2>/dev/null \
| grep -E '^ *[0-9] s:|i:'
# When was the certificate you are serving issued?
openssl s_client -connect example.com:443 -servername example.com </dev/null 2>/dev/null \
| openssl x509 -noout -startdate
If the chain ends at a root with a distrust-after date and notBefore is later than it, Firefox will refuse it once that rule is in the release a user is running, and nothing on your side will have changed. The fix is not on the server. It is a certificate issued under a different root.
What to take from it
A certificate warning in one browser only is a question about that browser's root store before it is a question about your server. Ask which browser and version, look at which root your chain ends at, and look at when the certificate was issued. Renewal automation is the thing most likely to move that last date without anybody noticing, and a distrust-after rule is the thing that makes the date matter.
Top comments (0)