Skip to content

Cloudflare Minimum TLS Version: what to set and what breaks when you raise it

The edge default still accepts TLS 1.0. Raising the floor to 1.2 is the right call for almost every zone — but it locks out a predictable list of old clients, and none of them will show up in your origin logs. What each setting drops, how to test, and when 1.3-only makes sense.

The setting lives under SSL/TLS → Edge Certificates → Minimum TLS Version and controls one thing: the oldest protocol version Cloudflare’s edge will accept from visitors. The default on new zones is TLS 1.0, which is a compatibility choice, not a security recommendation. Set 1.2 for general-purpose sites — it satisfies PCI DSS and drops only genuinely ancient clients. Reserve 1.3-only for API zones where you control every client. The origin hop is a separate conversation configured on your own server.

What’s happening

During the handshake, the client offers the versions it speaks and the edge picks the best mutually supported one; the minimum version setting refuses the handshake when the client’s best is below the floor. The setting applies zone-wide on Free, Pro, and Business; a per-hostname minimum requires Advanced Certificate Manager, a paid add-on. Enabling TLS 1.3 support is a separate toggle in the same section and should stay on regardless of the floor you choose.

Who still speaks only TLS 1.0/1.1 in practice: Android 4.3 and older, Internet Explorer 10 and older, Java 6 and 7 with default settings, anything on Windows XP, old embedded devices — payment terminals, printers, kiosk browsers, smart TVs — and legacy scripts built on OpenSSL 0.9.8. Who fails below a 1.3-only floor is a much bigger and more surprising list: clients on OpenSSL below 1.1.1, Java 8 before the 1.3 backport, plenty of corporate proxy appliances, and older monitoring probes.

The failures are invisible on your side by design: a client below the floor never completes the handshake, so nothing reaches the origin and nothing appears in origin logs. Broken old clients report vague connection errors, and you learn about it from support tickets unless you measure first.

What to check first

  • The current value under SSL/TLS → Edge Certificates → Minimum TLS Version.
  • What the zone serves: a public marketing site, an e-commerce checkout in PCI scope (1.2 is mandatory there), or an API with known consumers.
  • What share of traffic still arrives over old protocol versions — from the zone analytics TLS breakdown where your plan exposes it, or from your application’s own client telemetry.
  • Whether one hostname needs a different floor than the rest, which is the ACM per-hostname case.

How to verify

# Which versions does the edge accept today?
openssl s_client -connect example.com:443 -servername example.com -tls1_2 </dev/null 2>&1 | grep -E "Protocol|Cipher is"
openssl s_client -connect example.com:443 -servername example.com -tls1_3 </dev/null 2>&1 | grep -E "Protocol|Cipher is"

# Old protocols should FAIL after you raise the floor
# (many modern openssl builds no longer ship -tls1/-tls1_1; curl can cap instead)
curl -sv --tls-max 1.1 https://example.com/ -o /dev/null 2>&1 | tail -3

# Full protocol/cipher sweep if nmap is available
nmap --script ssl-enum-ciphers -p 443 example.com

A successful --tls-max 1.1 connection after you set a 1.2 floor means the change has not applied where you think — check that you are testing the right zone and hostname, not a grey-clouded record that bypasses the edge entirely.

The fix

  1. Measure before you move. Pull whatever TLS-version breakdown you have and identify the clients behind the old-protocol share. One percent of traffic on TLS 1.0 is a rounding error for a blog and a revenue event for a B2B API with two large customers on old JVMs. Announce the change to known API consumers with a date.
  2. Raise the floor to TLS 1.2 under SSL/TLS → Edge Certificates. The change is immediate and touches no certificates. This is the correct resting state for nearly every zone, and the same floor PCI DSS requires — the on-origin half of that retirement is covered in planning the TLS 1.0/1.1 shutdown.
  3. Test the fence from both sides. --tls-max 1.1 must fail; -tls1_2 and -tls1_3 must succeed. Test the real production hostnames, since a grey-cloud record never sees edge settings.
  4. Watch the human channels for a week. The affected clients cannot reach your origin, so watch support queues and partner escalations rather than server logs. Failures are deterministic — a locked-out client fails every time, so intermittent complaints are a different problem.
  5. Handle the legacy exception with ACM, not a zone-wide downgrade. If one endpoint must keep accepting an old fleet — a legacy.example.com for devices that will never update — buy Advanced Certificate Manager and set the per-hostname minimum there, keeping the rest of the zone at 1.2. Do not drag the whole zone down for one endpoint.
  6. Set the origin’s own floor separately. Cloudflare negotiates modern TLS to your origin, but direct connections do not pass through the edge: mirror the policy in Nginx or Apache (ssl_protocols TLSv1.2 TLSv1.3; or the SSLProtocol equivalent). Which suites to keep alongside is the subject of TLS configs: keep or retire, and Cloudflare’s own reference is at developers.cloudflare.com/ssl/.

When this fix is wrong

If a subdomain throws ERR_SSL_VERSION_OR_CIPHER_MISMATCH while the apex works, the cause is almost certainly certificate coverage — a multi-level subdomain outside Universal SSL’s reach — not the protocol floor; the error name misleads. If failures are intermittent for the same client, version negotiation is not your culprit either, because a floor violation fails every single time. And a 525 after tightening things is an origin-side handshake problem, unrelated to this edge setting.

Common mistakes

  • Setting 1.3-only zone-wide because newest sounds best, then spending a week discovering which partner integrations broke.
  • Testing from a current laptop, seeing everything work, and concluding nothing broke — modern clients were never at risk; the affected ones are exactly the ones not on your desk.
  • Waiting for blocked clients to appear in origin logs. They never arrive there.
  • Confusing the minimum-version floor with the TLS 1.3 enable toggle — one refuses old clients, the other permits new protocol use.

Prevention

Review the floor annually alongside the certificate inventory, and record it per zone in the runbook so audits stop being archaeology. For API products, publish a supported-TLS policy and a deprecation calendar so client upgrades happen on your schedule. Add a synthetic check pinned at your floor version (and one pinned just below, expected to fail) so drift in either direction pages someone before a customer does.

Getting end-to-end TLS right — edge, origin certificate, and the redirect chain — is standard work inside Managed Cloud Ops; book an infrastructure review if you want it verified end to end.