1// SPDX-License-Identifier: AGPL-3.0-or-later
2
3package web
4
5import (
6 "context"
7 "errors"
8 "fmt"
9 "io"
10 "net/http"
11 "net/url"
12 "strconv"
13 "strings"
14 "sync"
15 "time"
16
17 "github.com/bnema/zerowrap"
18
19 "bnema.dev/ermite/internal/clientlimit"
20 "bnema.dev/ermite/internal/domain"
21 "bnema.dev/ermite/internal/ports"
22)
23
24// Live updates. A page that can change under the eyes of its reader opens
25// GET /-/events (the Live component of the hint badge). The handler follows
26// EventService.Watch as the principal of the request and turns each event into
27// the HTML fragments the page shows, sent as Server-Sent Events:
28//
29// GET /-/events?topic=home
30// GET /-/events?topic=owner&view=inbox[&repo=<name>] (the signed-in owner only)
31// GET /-/events?topic=repo:<name>&view=summary|refs
32// GET /-/events?topic=repo:<name>&view=commits&ref=<branch>&head=<sha>
33// GET /-/events?topic=repo:<name>&view=mrs
34// GET /-/events?topic=repo:<name>&view=mr&n=<number>&ref=<source branch>
35// GET /-/events?topic=repo:<name>&view=issues
36// GET /-/events?topic=repo:<name>&view=issue&n=<number>
37// GET /-/events?topic=repo:<name>&view=actions
38// GET /-/events?topic=repo:<name>&view=run&n=<run id>
39// GET /-/events?topic=repo:<name>&view=job&job=<job id>
40//
41// topic is the Watch topic. view says which fragments of a repository the
42// page shows, so that the server renders only those (summary when absent).
43// On the commits page, ref is the branch it lists and head the commit it
44// lists first: the banner counts the commits between that head and the new
45// tip, so it stays right however many pushes came in and however often the
46// stream reconnected. A Last-Event-ID header resumes the stream
47// (resume_after).
48//
49// Events, each with the id of the Watch event it comes from:
50//
51// home-list data: <li id="repo-<name>">…</li>, or <li id="repo-<name>" data-gone></li>
52// for a repository that is deleted or no longer visible
53// Activity rows carry data-activity: push, MR lifecycle/comment,
54// or issue lifecycle/comment (never pending moderation or CI).
55// repo-last-push data: <dd id="last-push-<name>">…</dd>
56// refs-list data: <tbody id="ref-rows">…</tbody>, and again with
57// <dl id="refs-kv">…</dl>, the facts of the rail
58// commits-head data: <div id="commits-head">N new commits — reload</div>
59// mr-notice data: <div id="mr-notice">…</div>, on the merge request and issue pages:
60// something changed, with the link that reloads the page
61// ci-run data: <tr id="run-<id>">…</tr> (the list of runs),
62// <section id="recent-actions">…</section> (the rail of the summary) or
63// <div id="run-body">…</div> (one run)
64// ci-job data: <h2 id="job-head">…</h2> and <ol id="job-steps">…</ol>; the log
65// itself is never sent in this stream: the page follows it on its own
66// feed, /<repo>/actions/runs/<id>/job/<job>/stream?from=<line>
67// gap data: gap; events were lost, or what the page shows is gone or
68// changed: the page fetches itself again (ermiteLive in ermite.js)
69//
70// The data of an event is one fragment, a top-level element with an id, one
71// "data:" line per line of HTML; the page swaps the element that has that id
72// (ermiteLive.swap). The slots of the banners (commits-head, mr-notice) are
73// status regions that stay in the page: only their children are replaced, so
74// that a screen reader announces the banner.
75// An event that changes nothing on the page is still sent as an id alone, so
76// that a reconnection resumes after it.
77
78// DefaultHeartbeat is how often an idle stream gets a comment line, so that
79// proxies keep it open and a dead peer is noticed.
80const DefaultHeartbeat = 25 * time.Second
81
82const (
83 // writeWindow bounds one write to the visitor, whatever the write timeout
84 // of the server is.
85 writeWindow = 30 * time.Second
86 // retryAfter is the delay asked from a visitor refused for lack of room.
87 retryAfter = 10
88)
89
90// ownerTopic is the stream of the inbox: the merge requests and issues of every
91// repository, for the owner.
92const ownerTopic = "owner"
93
94// parseTopic reads "home", "owner" or "repo:<name>"; repo is "" for home and
95// owner.
96func parseTopic(topic string) (repo string, ok bool) {
97 if topic == "home" || topic == ownerTopic {
98 return "", true
99 }
100 if name, found := strings.CutPrefix(topic, "repo:"); found && name != "" {
101 return name, true
102 }
103 return "", false
104}
105
106// ownerStreamOK tells whether the request may follow the owner topic: the
107// signed-in owner, with the inbox view and a valid repository filter, on a
108// server that has the inbox.
109func (s *Server) ownerStreamOK(ctx context.Context, q url.Values) bool {
110 if s.cfg.Inbox == nil || !SignedIn(ctx) || q.Get("view") != "inbox" {
111 return false
112 }
113 if f := q.Get("repo"); f != "" {
114 if _, err := domain.ParseRepoName(f); err != nil {
115 return false
116 }
117 }
118 return true
119}
120
121// events is GET /-/events.
122func (s *Server) events(w http.ResponseWriter, r *http.Request) {
123 if s.cfg.Events == nil {
124 http.NotFound(w, r)
125 return
126 }
127 q := r.URL.Query()
128 topic := q.Get("topic")
129 repo, ok := parseTopic(topic)
130 if topic == ownerTopic {
131 // Anyone but the signed-in owner gets the answer of an unknown topic.
132 ok = s.ownerStreamOK(r.Context(), q)
133 repo = ""
134 }
135 if !ok {
136 denyStream(w, http.StatusBadRequest, "topic must be home or repo:<name>")
137 return
138 }
139 ctx, cancel := context.WithCancel(r.Context())
140 defer cancel()
141 log := zerowrap.FromCtx(ctx)
142
143 if repo != "" {
144 // A repository the visitor cannot see is a missing one (AC2).
145 cctx, ccancel := context.WithTimeout(ctx, requestTimeout)
146 err := s.canSee(cctx, repo)
147 if err == nil {
148 if v := q.Get("view"); v == "issues" || v == "issue" {
149 err = s.issuesOn(cctx, repo)
150 }
151 }
152 ccancel()
153 if err != nil {
154 s.denyEvents(ctx, w, err)
155 return
156 }
157 }
158
159 // The commit of the run of a job page is read once here: the annotations of
160 // the live job fragments link to files at it, and it never changes, so a
161 // job update does not read the run again. A run that is not there or not
162 // readable refuses the stream like a missing repository.
163 var sha string
164 if repo != "" && q.Get("view") == "job" && s.cfg.Runs != nil {
165 cctx, ccancel := context.WithTimeout(ctx, requestTimeout)
166 var err error
167 sha, err = s.jobRunSha(cctx, repo, q.Get("job"))
168 ccancel()
169 if err != nil {
170 s.denyEvents(ctx, w, err)
171 return
172 }
173 }
174
175 // The watch runs in a goroutine and hands each event to this one through
176 // an unbuffered channel: a visitor that reads slowly holds the watch in
177 // its send, falls behind the buffer of the use case, and gets a gap.
178 msgs := make(chan recvd)
179 started := make(chan struct{})
180 var startOnce sync.Once
181 start := func() { startOnce.Do(func() { close(started) }) }
182 finished := make(chan struct{})
183 params := ports.WatchParams{
184 Topics: []string{topic}, ResumeAfter: r.Header.Get("Last-Event-ID"),
185 // The visitor is the client of the watch, for the limit per client.
186 Client: clientlimit.Key(clientlimit.ClientAddr(r, s.cfg.TrustedProxies)),
187 Ready: start,
188 }
189 go func() {
190 defer close(finished)
191 defer close(msgs)
192 // net/http only recovers the handler goroutine: a panic in the watch
193 // would end the process. It ends this stream as an error instead.
194 defer func() {
195 if v := recover(); v != nil {
196 log.Error().Interface("panic", v).Msg("event watch panicked")
197 select {
198 case msgs <- recvd{err: fmt.Errorf("watch panicked: %v", v)}:
199 case <-ctx.Done():
200 }
201 }
202 }()
203 err := s.cfg.Events.Watch(ctx, principal(ctx), params, func(we ports.WatchEvent) error {
204 start()
205 select {
206 case msgs <- recvd{ev: we}:
207 return nil
208 case <-ctx.Done():
209 return ctx.Err()
210 }
211 })
212 if err != nil && ctx.Err() == nil {
213 select {
214 case msgs <- recvd{err: err}:
215 case <-ctx.Done():
216 }
217 }
218 }()
219 defer func() {
220 cancel()
221 <-finished
222 }()
223
224 // The watch is registered (or refused, which ends it before): only then
225 // does the answer commit to 200.
226 var first *recvd
227 select {
228 case <-started:
229 case m, open := <-msgs:
230 if !open {
231 return // the server is stopping
232 }
233 if m.err != nil {
234 s.denyEvents(ctx, w, m.err)
235 return
236 }
237 first = &m
238 case <-ctx.Done():
239 return
240 }
241
242 rc := http.NewResponseController(w)
243 hdr := w.Header()
244 hdr.Set("Content-Type", "text/event-stream")
245 cacheHeaders(ctx, hdr, "no-cache, no-transform", sessionCache+", no-transform")
246 if hdr.Get("Cache-Control") == noStore { // an invalid cookie was cleared
247 hdr.Set("Cache-Control", noStore+", no-transform")
248 }
249 hdr.Set("X-Accel-Buffering", "no")
250 _ = rc.SetWriteDeadline(time.Now().Add(writeWindow))
251 w.WriteHeader(http.StatusOK)
252 out := &sseWriter{w: w, rc: rc}
253 if err := out.raw(": connected\n\n"); err != nil {
254 return
255 }
256
257 st := &liveStream{s: s, repo: repo, view: q.Get("view"), ref: q.Get("ref"), head: q.Get("head"), n: q.Get("n"), job: q.Get("job"), sha: sha}
258 if topic == ownerTopic {
259 st.filter = q.Get("repo")
260 }
261 handle := func(m recvd) bool {
262 if m.err != nil {
263 if ctx.Err() == nil && !errors.Is(m.err, context.Canceled) {
264 log.Debug().Err(m.err).Str("topic", topic).Msg("event stream ended")
265 }
266 return false
267 }
268 if !liveEvent(m.ev) {
269 return true // nothing to show, not even an id to resume after
270 }
271 // An owner's stream ends when the session does, before it sends
272 // anything more: an event can carry what only the owner may see, and
273 // waiting for the next heartbeat would leave a window.
274 if SignedIn(ctx) && !s.sessionStillValid(ctx, r) {
275 return false
276 }
277 rctx, rcancel := context.WithTimeout(ctx, requestTimeout)
278 frags, err := st.render(rctx, m.ev)
279 rcancel()
280 if err != nil {
281 if ctx.Err() != nil {
282 return false
283 }
284 // What the page shows can no longer be told from the events:
285 // have it fetch itself again.
286 log.Warn().Err(err).Str("topic", topic).Msg("render live fragment")
287 frags = gap
288 }
289 return out.send(m.ev.ID, frags) == nil
290 }
291 if first != nil && !handle(*first) {
292 return
293 }
294
295 beat := s.cfg.Heartbeat
296 if beat <= 0 {
297 beat = DefaultHeartbeat
298 }
299 tick := time.NewTicker(beat)
300 defer tick.Stop()
301 for {
302 select {
303 case m, open := <-msgs:
304 if !open || !handle(m) {
305 return
306 }
307 case <-tick.C:
308 // A stream opened by the owner ends when the session does
309 // (sign out everywhere, expiry): the page reconnects as it can.
310 if SignedIn(ctx) && !s.sessionStillValid(ctx, r) {
311 return
312 }
313 if out.raw(": heartbeat\n\n") != nil {
314 return
315 }
316 case <-ctx.Done():
317 return
318 }
319 }
320}
321
322// recvd is a message of the Watch stream, or how it failed.
323type recvd struct {
324 ev ports.WatchEvent
325 err error
326}
327
328// canSee checks that the caller of the request may see the repository named in
329// a topic: one it may not is as missing as one that does not exist.
330func (s *Server) canSee(ctx context.Context, rawName string) error {
331 name, err := domain.ParseRepoName(rawName)
332 if err != nil {
333 return err
334 }
335 _, err = s.cfg.Repos.Get(ctx, principal(ctx), name)
336 return err
337}
338
339// issuesOn is domain.ErrNotFound unless the repository has issues on and
340// s.cfg.Issues is set: the same answer as for a repository the visitor cannot see.
341func (s *Server) issuesOn(ctx context.Context, rawName string) error {
342 if s.cfg.Issues == nil {
343 return domain.ErrNotFound
344 }
345 name, err := domain.ParseRepoName(rawName)
346 if err != nil {
347 return err
348 }
349 v, err := s.cfg.Repos.Get(ctx, principal(ctx), name)
350 if err != nil {
351 return err
352 }
353 if v.Settings.Issues.OrDefault() == domain.IssuesOff {
354 return domain.ErrNotFound
355 }
356 return nil
357}
358
359// denyStream answers a request that gets no stream.
360func denyStream(w http.ResponseWriter, status int, msg string) {
361 w.Header().Set("Cache-Control", "no-store")
362 http.Error(w, msg, status)
363}
364
365// denyEvents maps an error of a use case, before the stream started, to a
366// status. Not found and invalid are 404 whatever the reason: a private
367// repository is a missing one.
368func (s *Server) denyEvents(ctx context.Context, w http.ResponseWriter, err error) {
369 switch {
370 case errors.Is(err, context.Canceled):
371 // The visitor is gone.
372 case errors.Is(err, domain.ErrNotFound), domain.IsInvalid(err):
373 denyStream(w, http.StatusNotFound, "not found")
374 case errors.Is(err, domain.ErrResourceExhausted):
375 w.Header().Set("Retry-After", strconv.Itoa(retryAfter))
376 denyStream(w, http.StatusServiceUnavailable, "too many event streams, retry later")
377 case errors.Is(err, domain.ErrUnauthenticated):
378 denyStream(w, http.StatusUnauthorized, "unauthorized")
379 default:
380 log := zerowrap.FromCtx(ctx)
381 log.Error().Err(err).Msg("open event stream")
382 denyStream(w, http.StatusInternalServerError, "internal error")
383 }
384}
385
386// fragment is one SSE event to send: the name and the HTML of the fragment.
387type fragment struct{ event, html string }
388
389// sseWriter writes events to a visitor, flushing each, under a deadline of
390// its own so that the write timeout of the server cannot end the stream.
391type sseWriter struct {
392 w http.ResponseWriter
393 rc *http.ResponseController
394}
395
396func (o *sseWriter) raw(s string) error {
397 _ = o.rc.SetWriteDeadline(time.Now().Add(writeWindow))
398 if _, err := io.WriteString(o.w, s); err != nil {
399 return err
400 }
401 return o.rc.Flush()
402}
403
404// send writes the fragments of one Watch event under its id. An event that
405// has no fragment still carries its id, so that a reconnection resumes after
406// it.
407func (o *sseWriter) send(id string, frags []fragment) error {
408 if len(frags) == 0 {
409 if id == "" {
410 return nil
411 }
412 return o.raw(formatEvent(id, "", "", false))
413 }
414 for _, f := range frags {
415 if err := o.raw(formatEvent(id, f.event, f.html, true)); err != nil {
416 return err
417 }
418 }
419 return nil
420}
421
422// formatEvent is one SSE event. Each line of data is a "data:" field, and a
423// browser joins them back with newlines.
424func formatEvent(id, event, data string, hasData bool) string {
425 var b strings.Builder
426 if event != "" {
427 b.WriteString("event: " + event + "\n")
428 }
429 if id != "" {
430 b.WriteString("id: " + id + "\n")
431 }
432 if hasData {
433 data = strings.ReplaceAll(strings.ReplaceAll(data, "\r\n", "\n"), "\r", "\n")
434 for line := range strings.SplitSeq(data, "\n") {
435 b.WriteString("data: " + line + "\n")
436 }
437 }
438 b.WriteString("\n")
439 return b.String()
440}
441
442// liveURL is the stream a page follows: the topic and, for a repository
443// page, what it shows (see the comment on events).
444func liveURL(topic, query string) string {
445 u := "/-/events?topic=" + url.QueryEscape(topic)
446 if query != "" {
447 u += "&" + query
448 }
449 return u
450}