For reading and verification only. You may not copy, use, modify or redistribute this code. Full notice on the main page.
tunnel/certreload.go
| 1 | package main |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/tls" |
| 6 | "crypto/x509" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "log/slog" |
| 10 | "os" |
| 11 | "sync" |
| 12 | "time" |
| 13 | ) |
| 14 | |
| 15 | // The node's own public-CA certificate (the one devices verify on the node's own name, ALPN all-tunnel/2) is renewed by the node's own ACME client |
| 16 | // (certbot, HTTP-01 through the webroot on :80) and replaced on disk. certReloader puts the new one in service without a restart and without a dropped |
| 17 | // handshake: it reads the files at start and again whenever they change, at a handshake or on its own timer, and a renewal that is not good (half |
| 18 | // written, a key that does not match, another name, already expired) never replaces the certificate that works. |
| 19 | |
| 20 | // fileSig is what the node looks at to know a file changed: its modification time and size (certbot's live files are symlinks into an archive, and Stat |
| 21 | // follows them, so a renewal that moves the link is a different file). |
| 22 | type fileSig struct{ mod, size int64 } |
| 23 | |
| 24 | func sigOf(path string) (fileSig, error) { |
| 25 | st, err := os.Stat(path) |
| 26 | if err != nil { |
| 27 | return fileSig{}, err |
| 28 | } |
| 29 | return fileSig{st.ModTime().UnixNano(), st.Size()}, nil |
| 30 | } |
| 31 | |
| 32 | const ( |
| 33 | certLookGap = time.Second // a handshake looks at the files at most this often (certReloader.lookGap) |
| 34 | certRetryGap = 30 * time.Second // a renewal that did not load is tried again after this even if its files did not change again |
| 35 | certWatchEvery = time.Minute // the node looks at the files by itself this often |
| 36 | certWarnDays = 5 // days before the end at which the log says it loudly |
| 37 | ) |
| 38 | |
| 39 | type certReloader struct { |
| 40 | certPath, keyPath string |
| 41 | host string // the node's own name the certificate must be valid for; "" skips the name check (tests) |
| 42 | clk clock |
| 43 | log *slog.Logger |
| 44 | lookGap time.Duration |
| 45 | |
| 46 | mu sync.Mutex |
| 47 | pair *tls.Certificate |
| 48 | notAfter time.Time |
| 49 | certSig fileSig // what the certificate in service was read from |
| 50 | keySig fileSig |
| 51 | lookedAt time.Time |
| 52 | failedAt time.Time |
| 53 | failedCrt fileSig // what the last refused renewal was read from |
| 54 | failedKey fileSig |
| 55 | lastErr string |
| 56 | loads int |
| 57 | } |
| 58 | |
| 59 | func newCertReloader(certPath, keyPath, host string, clk clock, log *slog.Logger) (*certReloader, error) { |
| 60 | if clk.now == nil { |
| 61 | clk = newClock(nil, nil) |
| 62 | } |
| 63 | if log == nil { |
| 64 | log = slog.New(slog.DiscardHandler) |
| 65 | } |
| 66 | r := &certReloader{certPath: certPath, keyPath: keyPath, host: host, clk: clk, log: log, lookGap: certLookGap} |
| 67 | if err := r.load(); err != nil { |
| 68 | return nil, fmt.Errorf("node certificate: %w", err) |
| 69 | } |
| 70 | return r, nil |
| 71 | } |
| 72 | |
| 73 | // load reads the pair, checks it and puts it in service. The caller holds no lock. |
| 74 | func (r *certReloader) load() error { |
| 75 | cs, cerr := sigOf(r.certPath) |
| 76 | ks, kerr := sigOf(r.keyPath) |
| 77 | if err := errors.Join(cerr, kerr); err != nil { |
| 78 | return err |
| 79 | } |
| 80 | pair, err := tls.LoadX509KeyPair(r.certPath, r.keyPath) |
| 81 | if err != nil { |
| 82 | return err |
| 83 | } |
| 84 | leaf, err := x509.ParseCertificate(pair.Certificate[0]) |
| 85 | if err != nil { |
| 86 | return err |
| 87 | } |
| 88 | now := r.clk.Now() |
| 89 | if r.host != "" { |
| 90 | if err := leaf.VerifyHostname(r.host); err != nil { |
| 91 | return fmt.Errorf("the certificate is not for %s", r.host) |
| 92 | } |
| 93 | } |
| 94 | if !now.Before(leaf.NotAfter) { |
| 95 | return fmt.Errorf("the certificate expired at %s", leaf.NotAfter.UTC().Format(time.RFC3339)) |
| 96 | } |
| 97 | if now.Add(time.Hour).Before(leaf.NotBefore) { |
| 98 | return fmt.Errorf("the certificate is not valid before %s", leaf.NotBefore.UTC().Format(time.RFC3339)) |
| 99 | } |
| 100 | pair.Leaf = leaf |
| 101 | r.mu.Lock() |
| 102 | r.pair, r.notAfter, r.certSig, r.keySig = &pair, leaf.NotAfter, cs, ks |
| 103 | r.lastErr, r.failedAt, r.failedCrt, r.failedKey = "", time.Time{}, fileSig{}, fileSig{} |
| 104 | r.loads++ |
| 105 | r.mu.Unlock() |
| 106 | return nil |
| 107 | } |
| 108 | |
| 109 | // refresh looks at the files (at most every certLookGap unless force) and loads them when they are not the ones in service. A refused renewal is |
| 110 | // remembered by its files, so it is parsed again only when they change or certRetryGap passed. |
| 111 | func (r *certReloader) refresh(force bool) { |
| 112 | now := r.clk.Now() |
| 113 | r.mu.Lock() |
| 114 | if !force && now.Sub(r.lookedAt) < r.lookGap { |
| 115 | r.mu.Unlock() |
| 116 | return |
| 117 | } |
| 118 | r.lookedAt = now |
| 119 | cur, curKey, failedCrt, failedKey, failedAt := r.certSig, r.keySig, r.failedCrt, r.failedKey, r.failedAt |
| 120 | r.mu.Unlock() |
| 121 | |
| 122 | cs, cerr := sigOf(r.certPath) |
| 123 | ks, kerr := sigOf(r.keyPath) |
| 124 | if cerr != nil || kerr != nil { |
| 125 | return // a renewal in the middle of replacing its files: the certificate in service stays |
| 126 | } |
| 127 | if cs == cur && ks == curKey { |
| 128 | return |
| 129 | } |
| 130 | if !force && cs == failedCrt && ks == failedKey && !failedAt.IsZero() && now.Sub(failedAt) < certRetryGap { |
| 131 | return |
| 132 | } |
| 133 | if err := r.load(); err != nil { |
| 134 | r.mu.Lock() |
| 135 | r.failedCrt, r.failedKey, r.failedAt, r.lastErr = cs, ks, now, err.Error() |
| 136 | r.mu.Unlock() |
| 137 | r.log.Warn("node_cert_reload_refused", "reason", err.Error()) |
| 138 | return |
| 139 | } |
| 140 | r.log.Info("node_cert_reloaded", "not_after", r.NotAfter().UTC().Format(time.RFC3339)) |
| 141 | } |
| 142 | |
| 143 | // Reload loads the files now, whatever they look like (SIGHUP): it returns why a renewal was refused. |
| 144 | func (r *certReloader) Reload() error { |
| 145 | if err := r.load(); err != nil { |
| 146 | r.mu.Lock() |
| 147 | r.lastErr = err.Error() |
| 148 | r.mu.Unlock() |
| 149 | return err |
| 150 | } |
| 151 | return nil |
| 152 | } |
| 153 | |
| 154 | // get is the Config.CertSource: the certificate in service, after a look at the files. |
| 155 | func (r *certReloader) get(*tls.ClientHelloInfo) (*tls.Certificate, error) { |
| 156 | r.refresh(false) |
| 157 | r.mu.Lock() |
| 158 | defer r.mu.Unlock() |
| 159 | return r.pair, nil |
| 160 | } |
| 161 | |
| 162 | // NotAfter is the end of the certificate in service (the monitor alerts under 14 days and under 5). |
| 163 | func (r *certReloader) NotAfter() time.Time { |
| 164 | r.mu.Lock() |
| 165 | defer r.mu.Unlock() |
| 166 | return r.notAfter |
| 167 | } |
| 168 | |
| 169 | // LastError is why the last renewal was refused ("" when the certificate in service is the newest on disk). |
| 170 | func (r *certReloader) LastError() string { |
| 171 | r.mu.Lock() |
| 172 | defer r.mu.Unlock() |
| 173 | return r.lastErr |
| 174 | } |
| 175 | |
| 176 | // Run looks at the files on its own timer, so a renewal is in service even when no device is connecting, and says in the log how long the certificate |
| 177 | // has left (the provider's expiry e-mails no longer exist, the node's own log and the site's monitor are what warn). |
| 178 | func (r *certReloader) Run(ctx context.Context) { |
| 179 | for { |
| 180 | r.refresh(true) |
| 181 | left := r.NotAfter().Sub(r.clk.Now()) |
| 182 | switch { |
| 183 | case left <= 0: |
| 184 | r.log.Error("node_cert_expired") |
| 185 | case left < certWarnDays*24*time.Hour: |
| 186 | r.log.Warn("node_cert_expires_soon", "days_left", int(left/(24*time.Hour))) |
| 187 | } |
| 188 | if !r.clk.sleep(ctx, certWatchEvery) { |
| 189 | return |
| 190 | } |
| 191 | } |
| 192 | } |