Actually make it worktls-cert-cache-internal-reload

1 files changed, 30 insertions, 16 deletions

diff --git a/modules/caddytls/tls.go b/modules/caddytls/tls.go
index 095571ae6..14965533e 100644
--- a/modules/caddytls/tls.go
+++ b/modules/caddytls/tls.go
@@ -22,7 +22,6 @@ import (
 	"log"
 	"net/http"
 	"runtime/debug"
-	"strings"
 	"sync"
 	"time"
 
@@ -92,7 +91,8 @@ type TLS struct {
 
 	// set of subjects with managed certificates,
 	// and hashes of manually-loaded certificates
-	managing, loaded map[string]struct{}
+	// (managing's value is an optional issuer key, for distinction)
+	managing, loaded map[string]string
 }
 
 // CaddyModule returns the Caddy module information.
@@ -113,7 +113,7 @@ func (t *TLS) Provision(ctx caddy.Context) error {
 	t.ctx = ctx
 	t.logger = ctx.Logger()
 	repl := caddy.NewReplacer()
-	t.managing, t.loaded = make(map[string]struct{}), make(map[string]struct{})
+	t.managing, t.loaded = make(map[string]string), make(map[string]string)
 
 	// set up a new certificate cache; this (re)loads all certificates
 	cacheOpts := certmagic.CacheOptions{
@@ -267,7 +267,7 @@ func (t *TLS) Provision(ctx caddy.Context) error {
 			if err != nil {
 				return fmt.Errorf("caching unmanaged certificate: %v", err)
 			}
-			t.loaded[hash] = struct{}{}
+			t.loaded[hash] = ""
 		}
 	}
 
@@ -361,14 +361,24 @@ func (t *TLS) Cleanup() error {
 		// new app instance (which just started), and remove them from the cache
 		var noLongerManaged []certmagic.SubjectIssuer
 		var reManage, noLongerLoaded []string
-		for subj := range t.managing {
-			if _, ok := nextTLSApp.managing[subj]; !ok {
-				name, issuerKey, _ := strings.Cut(subj, "~")
-				noLongerManaged = append(noLongerManaged, certmagic.SubjectIssuer{Subject: name, IssuerKey: issuerKey})
-				// if the new TLS app is managing a cert for the same subject we are evicting,
-				// make sure it obtains or loads the cert using that config afterwards
-				if _, ok := nextTLSApp.managing[name]; ok {
-					reManage = append(reManage, name)
+		for subj, currentIssuerKey := range t.managing {
+			// It's a bit nuanced: managed certs can sometimes be different enough that we have to
+			// swap them out for a different one, even if they are for the same subject/domain.
+			// We consider "private" certs (internal CA/locally-trusted/etc) to be significantly
+			// distinct from "public" certs (production CAs/globally-trusted/etc) because of the
+			// implications when it comes to actual deployments: switching between an internal CA
+			// and a production CA, for example, is quite significant. Switching from one public CA
+			// to another, however, is not, and for our purposes we consider those to be the same.
+			// Anyway, if the next TLS app does not manage a cert for this name at all, definitely
+			// remove it from the cache. But if it does, and it's not the same kind of issuer/CA
+			// as we have, also remove it, so that it can swap it out for the right one.
+			if nextIssuerKey, ok := nextTLSApp.managing[subj]; !ok || nextIssuerKey != currentIssuerKey {
+				// next app is not managing a cert for this domain at all or is using a different issuer, so remove it
+				noLongerManaged = append(noLongerManaged, certmagic.SubjectIssuer{Subject: subj, IssuerKey: currentIssuerKey})
+
+				// then, if the next app is managing a cert for this name, but with a different issuer, re-manage it
+				if ok && nextIssuerKey != currentIssuerKey {
+					reManage = append(reManage, subj)
 				}
 			}
 		}
@@ -378,6 +388,7 @@ func (t *TLS) Cleanup() error {
 			}
 		}
 
+		// remove the certs
 		certCacheMu.RLock()
 		certCache.RemoveManaged(noLongerManaged)
 		certCache.Remove(noLongerLoaded)
@@ -427,13 +438,16 @@ func (t *TLS) Manage(names []string) error {
 			// a special case: if you have an initial config that manages example.com
 			// using internal CA, then after testing it you switch to a production CA,
 			// you wouldn't want to keep using the same self-signed cert, obviously;
-			// so we differentiate these in the list
+			// so we differentiate these by associating the subject with its issuer key;
+			// we do this because CertMagic has no notion of "InternalIssuer" like we
+			// do, so we have to do this logic ourselves
+			var issuerKey string
 			if len(ap.Issuers) == 1 {
-				if _, ok := ap.Issuers[0].(*InternalIssuer); ok {
-					name += "~" + ap.Issuers[0].IssuerKey()
+				if intIss, ok := ap.Issuers[0].(*InternalIssuer); ok && intIss != nil {
+					issuerKey = intIss.IssuerKey()
 				}
 			}
-			t.managing[name] = struct{}{}
+			t.managing[name] = issuerKey
 		}
 	}