blob: 297492e87af4568ce62a734d9c09b403ed268230 [file] [log] [blame]
Don Newton98fd8812019-09-23 15:15:02 -04001/*
2 *
3 * Copyright 2018 gRPC authors.
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 *
17 */
18
19// Package dns implements a dns resolver to be installed as the default resolver
20// in grpc.
21package dns
22
23import (
24 "context"
25 "encoding/json"
26 "errors"
27 "fmt"
28 "net"
29 "os"
30 "strconv"
31 "strings"
32 "sync"
33 "time"
34
35 "google.golang.org/grpc/grpclog"
36 "google.golang.org/grpc/internal/backoff"
37 "google.golang.org/grpc/internal/grpcrand"
38 "google.golang.org/grpc/resolver"
39)
40
41func init() {
42 resolver.Register(NewBuilder())
43}
44
45const (
46 defaultPort = "443"
47 defaultFreq = time.Minute * 30
48 defaultDNSSvrPort = "53"
49 golang = "GO"
50 // txtPrefix is the prefix string to be prepended to the host name for txt record lookup.
51 txtPrefix = "_grpc_config."
52 // In DNS, service config is encoded in a TXT record via the mechanism
53 // described in RFC-1464 using the attribute name grpc_config.
54 txtAttribute = "grpc_config="
55)
56
57var (
58 errMissingAddr = errors.New("dns resolver: missing address")
59
60 // Addresses ending with a colon that is supposed to be the separator
61 // between host and port is not allowed. E.g. "::" is a valid address as
62 // it is an IPv6 address (host only) and "[::]:" is invalid as it ends with
63 // a colon as the host and port separator
64 errEndsWithColon = errors.New("dns resolver: missing port after port-separator colon")
65)
66
67var (
68 defaultResolver netResolver = net.DefaultResolver
69 // To prevent excessive re-resolution, we enforce a rate limit on DNS
70 // resolution requests.
71 minDNSResRate = 30 * time.Second
72)
73
74var customAuthorityDialler = func(authority string) func(ctx context.Context, network, address string) (net.Conn, error) {
75 return func(ctx context.Context, network, address string) (net.Conn, error) {
76 var dialer net.Dialer
77 return dialer.DialContext(ctx, network, authority)
78 }
79}
80
81var customAuthorityResolver = func(authority string) (netResolver, error) {
82 host, port, err := parseTarget(authority, defaultDNSSvrPort)
83 if err != nil {
84 return nil, err
85 }
86
87 authorityWithPort := net.JoinHostPort(host, port)
88
89 return &net.Resolver{
90 PreferGo: true,
91 Dial: customAuthorityDialler(authorityWithPort),
92 }, nil
93}
94
95// NewBuilder creates a dnsBuilder which is used to factory DNS resolvers.
96func NewBuilder() resolver.Builder {
97 return &dnsBuilder{minFreq: defaultFreq}
98}
99
100type dnsBuilder struct {
101 // minimum frequency of polling the DNS server.
102 minFreq time.Duration
103}
104
105// Build creates and starts a DNS resolver that watches the name resolution of the target.
106func (b *dnsBuilder) Build(target resolver.Target, cc resolver.ClientConn, opts resolver.BuildOption) (resolver.Resolver, error) {
107 host, port, err := parseTarget(target.Endpoint, defaultPort)
108 if err != nil {
109 return nil, err
110 }
111
112 // IP address.
113 if net.ParseIP(host) != nil {
114 host, _ = formatIP(host)
115 addr := []resolver.Address{{Addr: host + ":" + port}}
116 i := &ipResolver{
117 cc: cc,
118 ip: addr,
119 rn: make(chan struct{}, 1),
120 q: make(chan struct{}),
121 }
122 cc.NewAddress(addr)
123 go i.watcher()
124 return i, nil
125 }
126
127 // DNS address (non-IP).
128 ctx, cancel := context.WithCancel(context.Background())
129 d := &dnsResolver{
130 freq: b.minFreq,
131 backoff: backoff.Exponential{MaxDelay: b.minFreq},
132 host: host,
133 port: port,
134 ctx: ctx,
135 cancel: cancel,
136 cc: cc,
137 t: time.NewTimer(0),
138 rn: make(chan struct{}, 1),
139 disableServiceConfig: opts.DisableServiceConfig,
140 }
141
142 if target.Authority == "" {
143 d.resolver = defaultResolver
144 } else {
145 d.resolver, err = customAuthorityResolver(target.Authority)
146 if err != nil {
147 return nil, err
148 }
149 }
150
151 d.wg.Add(1)
152 go d.watcher()
153 return d, nil
154}
155
156// Scheme returns the naming scheme of this resolver builder, which is "dns".
157func (b *dnsBuilder) Scheme() string {
158 return "dns"
159}
160
161type netResolver interface {
162 LookupHost(ctx context.Context, host string) (addrs []string, err error)
163 LookupSRV(ctx context.Context, service, proto, name string) (cname string, addrs []*net.SRV, err error)
164 LookupTXT(ctx context.Context, name string) (txts []string, err error)
165}
166
167// ipResolver watches for the name resolution update for an IP address.
168type ipResolver struct {
169 cc resolver.ClientConn
170 ip []resolver.Address
171 // rn channel is used by ResolveNow() to force an immediate resolution of the target.
172 rn chan struct{}
173 q chan struct{}
174}
175
176// ResolveNow resend the address it stores, no resolution is needed.
177func (i *ipResolver) ResolveNow(opt resolver.ResolveNowOption) {
178 select {
179 case i.rn <- struct{}{}:
180 default:
181 }
182}
183
184// Close closes the ipResolver.
185func (i *ipResolver) Close() {
186 close(i.q)
187}
188
189func (i *ipResolver) watcher() {
190 for {
191 select {
192 case <-i.rn:
193 i.cc.NewAddress(i.ip)
194 case <-i.q:
195 return
196 }
197 }
198}
199
200// dnsResolver watches for the name resolution update for a non-IP target.
201type dnsResolver struct {
202 freq time.Duration
203 backoff backoff.Exponential
204 retryCount int
205 host string
206 port string
207 resolver netResolver
208 ctx context.Context
209 cancel context.CancelFunc
210 cc resolver.ClientConn
211 // rn channel is used by ResolveNow() to force an immediate resolution of the target.
212 rn chan struct{}
213 t *time.Timer
214 // wg is used to enforce Close() to return after the watcher() goroutine has finished.
215 // Otherwise, data race will be possible. [Race Example] in dns_resolver_test we
216 // replace the real lookup functions with mocked ones to facilitate testing.
217 // If Close() doesn't wait for watcher() goroutine finishes, race detector sometimes
218 // will warns lookup (READ the lookup function pointers) inside watcher() goroutine
219 // has data race with replaceNetFunc (WRITE the lookup function pointers).
220 wg sync.WaitGroup
221 disableServiceConfig bool
222}
223
224// ResolveNow invoke an immediate resolution of the target that this dnsResolver watches.
225func (d *dnsResolver) ResolveNow(opt resolver.ResolveNowOption) {
226 select {
227 case d.rn <- struct{}{}:
228 default:
229 }
230}
231
232// Close closes the dnsResolver.
233func (d *dnsResolver) Close() {
234 d.cancel()
235 d.wg.Wait()
236 d.t.Stop()
237}
238
239func (d *dnsResolver) watcher() {
240 defer d.wg.Done()
241 for {
242 select {
243 case <-d.ctx.Done():
244 return
245 case <-d.t.C:
246 case <-d.rn:
247 if !d.t.Stop() {
248 // Before resetting a timer, it should be stopped to prevent racing with
249 // reads on it's channel.
250 <-d.t.C
251 }
252 }
253
254 result, sc := d.lookup()
255 // Next lookup should happen within an interval defined by d.freq. It may be
256 // more often due to exponential retry on empty address list.
257 if len(result) == 0 {
258 d.retryCount++
259 d.t.Reset(d.backoff.Backoff(d.retryCount))
260 } else {
261 d.retryCount = 0
262 d.t.Reset(d.freq)
263 }
264 d.cc.NewServiceConfig(sc)
265 d.cc.NewAddress(result)
266
267 // Sleep to prevent excessive re-resolutions. Incoming resolution requests
268 // will be queued in d.rn.
269 t := time.NewTimer(minDNSResRate)
270 select {
271 case <-t.C:
272 case <-d.ctx.Done():
273 t.Stop()
274 return
275 }
276 }
277}
278
279func (d *dnsResolver) lookupSRV() []resolver.Address {
280 var newAddrs []resolver.Address
281 _, srvs, err := d.resolver.LookupSRV(d.ctx, "grpclb", "tcp", d.host)
282 if err != nil {
283 grpclog.Infof("grpc: failed dns SRV record lookup due to %v.\n", err)
284 return nil
285 }
286 for _, s := range srvs {
287 lbAddrs, err := d.resolver.LookupHost(d.ctx, s.Target)
288 if err != nil {
289 grpclog.Infof("grpc: failed load balancer address dns lookup due to %v.\n", err)
290 continue
291 }
292 for _, a := range lbAddrs {
293 a, ok := formatIP(a)
294 if !ok {
295 grpclog.Errorf("grpc: failed IP parsing due to %v.\n", err)
296 continue
297 }
298 addr := a + ":" + strconv.Itoa(int(s.Port))
299 newAddrs = append(newAddrs, resolver.Address{Addr: addr, Type: resolver.GRPCLB, ServerName: s.Target})
300 }
301 }
302 return newAddrs
303}
304
305func (d *dnsResolver) lookupTXT() string {
306 ss, err := d.resolver.LookupTXT(d.ctx, txtPrefix+d.host)
307 if err != nil {
308 grpclog.Infof("grpc: failed dns TXT record lookup due to %v.\n", err)
309 return ""
310 }
311 var res string
312 for _, s := range ss {
313 res += s
314 }
315
316 // TXT record must have "grpc_config=" attribute in order to be used as service config.
317 if !strings.HasPrefix(res, txtAttribute) {
318 grpclog.Warningf("grpc: TXT record %v missing %v attribute", res, txtAttribute)
319 return ""
320 }
321 return strings.TrimPrefix(res, txtAttribute)
322}
323
324func (d *dnsResolver) lookupHost() []resolver.Address {
325 var newAddrs []resolver.Address
326 addrs, err := d.resolver.LookupHost(d.ctx, d.host)
327 if err != nil {
328 grpclog.Warningf("grpc: failed dns A record lookup due to %v.\n", err)
329 return nil
330 }
331 for _, a := range addrs {
332 a, ok := formatIP(a)
333 if !ok {
334 grpclog.Errorf("grpc: failed IP parsing due to %v.\n", err)
335 continue
336 }
337 addr := a + ":" + d.port
338 newAddrs = append(newAddrs, resolver.Address{Addr: addr})
339 }
340 return newAddrs
341}
342
343func (d *dnsResolver) lookup() ([]resolver.Address, string) {
344 newAddrs := d.lookupSRV()
345 // Support fallback to non-balancer address.
346 newAddrs = append(newAddrs, d.lookupHost()...)
347 if d.disableServiceConfig {
348 return newAddrs, ""
349 }
350 sc := d.lookupTXT()
351 return newAddrs, canaryingSC(sc)
352}
353
354// formatIP returns ok = false if addr is not a valid textual representation of an IP address.
355// If addr is an IPv4 address, return the addr and ok = true.
356// If addr is an IPv6 address, return the addr enclosed in square brackets and ok = true.
357func formatIP(addr string) (addrIP string, ok bool) {
358 ip := net.ParseIP(addr)
359 if ip == nil {
360 return "", false
361 }
362 if ip.To4() != nil {
363 return addr, true
364 }
365 return "[" + addr + "]", true
366}
367
368// parseTarget takes the user input target string and default port, returns formatted host and port info.
369// If target doesn't specify a port, set the port to be the defaultPort.
370// If target is in IPv6 format and host-name is enclosed in square brackets, brackets
371// are stripped when setting the host.
372// examples:
373// target: "www.google.com" defaultPort: "443" returns host: "www.google.com", port: "443"
374// target: "ipv4-host:80" defaultPort: "443" returns host: "ipv4-host", port: "80"
375// target: "[ipv6-host]" defaultPort: "443" returns host: "ipv6-host", port: "443"
376// target: ":80" defaultPort: "443" returns host: "localhost", port: "80"
377func parseTarget(target, defaultPort string) (host, port string, err error) {
378 if target == "" {
379 return "", "", errMissingAddr
380 }
381 if ip := net.ParseIP(target); ip != nil {
382 // target is an IPv4 or IPv6(without brackets) address
383 return target, defaultPort, nil
384 }
385 if host, port, err = net.SplitHostPort(target); err == nil {
386 if port == "" {
387 // If the port field is empty (target ends with colon), e.g. "[::1]:", this is an error.
388 return "", "", errEndsWithColon
389 }
390 // target has port, i.e ipv4-host:port, [ipv6-host]:port, host-name:port
391 if host == "" {
392 // Keep consistent with net.Dial(): If the host is empty, as in ":80", the local system is assumed.
393 host = "localhost"
394 }
395 return host, port, nil
396 }
397 if host, port, err = net.SplitHostPort(target + ":" + defaultPort); err == nil {
398 // target doesn't have port
399 return host, port, nil
400 }
401 return "", "", fmt.Errorf("invalid target address %v, error info: %v", target, err)
402}
403
404type rawChoice struct {
405 ClientLanguage *[]string `json:"clientLanguage,omitempty"`
406 Percentage *int `json:"percentage,omitempty"`
407 ClientHostName *[]string `json:"clientHostName,omitempty"`
408 ServiceConfig *json.RawMessage `json:"serviceConfig,omitempty"`
409}
410
411func containsString(a *[]string, b string) bool {
412 if a == nil {
413 return true
414 }
415 for _, c := range *a {
416 if c == b {
417 return true
418 }
419 }
420 return false
421}
422
423func chosenByPercentage(a *int) bool {
424 if a == nil {
425 return true
426 }
427 return grpcrand.Intn(100)+1 <= *a
428}
429
430func canaryingSC(js string) string {
431 if js == "" {
432 return ""
433 }
434 var rcs []rawChoice
435 err := json.Unmarshal([]byte(js), &rcs)
436 if err != nil {
437 grpclog.Warningf("grpc: failed to parse service config json string due to %v.\n", err)
438 return ""
439 }
440 cliHostname, err := os.Hostname()
441 if err != nil {
442 grpclog.Warningf("grpc: failed to get client hostname due to %v.\n", err)
443 return ""
444 }
445 var sc string
446 for _, c := range rcs {
447 if !containsString(c.ClientLanguage, golang) ||
448 !chosenByPercentage(c.Percentage) ||
449 !containsString(c.ClientHostName, cliHostname) ||
450 c.ServiceConfig == nil {
451 continue
452 }
453 sc = string(*c.ServiceConfig)
454 break
455 }
456 return sc
457}