1// Copyright 2012 The Go Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style
3// license that can be found in the LICENSE file.
4
5package vcs
6
7import (
8	"bytes"
9	"errors"
10	"fmt"
11	"internal/lazyregexp"
12	"internal/singleflight"
13	"io/fs"
14	"log"
15	urlpkg "net/url"
16	"os"
17	"os/exec"
18	"path/filepath"
19	"regexp"
20	"strconv"
21	"strings"
22	"sync"
23	"time"
24
25	"cmd/go/internal/base"
26	"cmd/go/internal/cfg"
27	"cmd/go/internal/search"
28	"cmd/go/internal/str"
29	"cmd/go/internal/web"
30
31	"golang.org/x/mod/module"
32)
33
34// A Cmd describes how to use a version control system
35// like Mercurial, Git, or Subversion.
36type Cmd struct {
37	Name      string
38	Cmd       string     // name of binary to invoke command
39	RootNames []rootName // filename and mode indicating the root of a checkout directory
40
41	CreateCmd   []string // commands to download a fresh copy of a repository
42	DownloadCmd []string // commands to download updates into an existing repository
43
44	TagCmd         []tagCmd // commands to list tags
45	TagLookupCmd   []tagCmd // commands to lookup tags before running tagSyncCmd
46	TagSyncCmd     []string // commands to sync to specific tag
47	TagSyncDefault []string // commands to sync to default tag
48
49	Scheme  []string
50	PingCmd string
51
52	RemoteRepo  func(v *Cmd, rootDir string) (remoteRepo string, err error)
53	ResolveRepo func(v *Cmd, rootDir, remoteRepo string) (realRepo string, err error)
54	Status      func(v *Cmd, rootDir string) (Status, error)
55}
56
57// Status is the current state of a local repository.
58type Status struct {
59	Revision    string    // Optional.
60	CommitTime  time.Time // Optional.
61	Uncommitted bool      // Required.
62}
63
64var (
65	// VCSTestRepoURL is the URL of the HTTP server that serves the repos for
66	// vcs-test.golang.org.
67	//
68	// In tests, this is set to the URL of an httptest.Server hosting a
69	// cmd/go/internal/vcweb.Server.
70	VCSTestRepoURL string
71
72	// VCSTestHosts is the set of hosts supported by the vcs-test server.
73	VCSTestHosts []string
74
75	// VCSTestIsLocalHost reports whether the given URL refers to a local
76	// (loopback) host, such as "localhost" or "127.0.0.1:8080".
77	VCSTestIsLocalHost func(*urlpkg.URL) bool
78)
79
80var defaultSecureScheme = map[string]bool{
81	"https":   true,
82	"git+ssh": true,
83	"bzr+ssh": true,
84	"svn+ssh": true,
85	"ssh":     true,
86}
87
88func (v *Cmd) IsSecure(repo string) bool {
89	u, err := urlpkg.Parse(repo)
90	if err != nil {
91		// If repo is not a URL, it's not secure.
92		return false
93	}
94	if VCSTestRepoURL != "" && web.IsLocalHost(u) {
95		// If the vcstest server is in use, it may redirect to other local ports for
96		// other protocols (such as svn). Assume that all loopback addresses are
97		// secure during testing.
98		return true
99	}
100	return v.isSecureScheme(u.Scheme)
101}
102
103func (v *Cmd) isSecureScheme(scheme string) bool {
104	switch v.Cmd {
105	case "git":
106		// GIT_ALLOW_PROTOCOL is an environment variable defined by Git. It is a
107		// colon-separated list of schemes that are allowed to be used with git
108		// fetch/clone. Any scheme not mentioned will be considered insecure.
109		if allow := os.Getenv("GIT_ALLOW_PROTOCOL"); allow != "" {
110			for _, s := range strings.Split(allow, ":") {
111				if s == scheme {
112					return true
113				}
114			}
115			return false
116		}
117	}
118	return defaultSecureScheme[scheme]
119}
120
121// A tagCmd describes a command to list available tags
122// that can be passed to tagSyncCmd.
123type tagCmd struct {
124	cmd     string // command to list tags
125	pattern string // regexp to extract tags from list
126}
127
128// vcsList lists the known version control systems
129var vcsList = []*Cmd{
130	vcsHg,
131	vcsGit,
132	vcsSvn,
133	vcsBzr,
134	vcsFossil,
135}
136
137// vcsMod is a stub for the "mod" scheme. It's returned by
138// repoRootForImportPathDynamic, but is otherwise not treated as a VCS command.
139var vcsMod = &Cmd{Name: "mod"}
140
141// vcsByCmd returns the version control system for the given
142// command name (hg, git, svn, bzr).
143func vcsByCmd(cmd string) *Cmd {
144	for _, vcs := range vcsList {
145		if vcs.Cmd == cmd {
146			return vcs
147		}
148	}
149	return nil
150}
151
152// vcsHg describes how to use Mercurial.
153var vcsHg = &Cmd{
154	Name: "Mercurial",
155	Cmd:  "hg",
156	RootNames: []rootName{
157		{filename: ".hg", isDir: true},
158	},
159
160	CreateCmd:   []string{"clone -U -- {repo} {dir}"},
161	DownloadCmd: []string{"pull"},
162
163	// We allow both tag and branch names as 'tags'
164	// for selecting a version. This lets people have
165	// a go.release.r60 branch and a go1 branch
166	// and make changes in both, without constantly
167	// editing .hgtags.
168	TagCmd: []tagCmd{
169		{"tags", `^(\S+)`},
170		{"branches", `^(\S+)`},
171	},
172	TagSyncCmd:     []string{"update -r {tag}"},
173	TagSyncDefault: []string{"update default"},
174
175	Scheme:     []string{"https", "http", "ssh"},
176	PingCmd:    "identify -- {scheme}://{repo}",
177	RemoteRepo: hgRemoteRepo,
178	Status:     hgStatus,
179}
180
181func hgRemoteRepo(vcsHg *Cmd, rootDir string) (remoteRepo string, err error) {
182	out, err := vcsHg.runOutput(rootDir, "paths default")
183	if err != nil {
184		return "", err
185	}
186	return strings.TrimSpace(string(out)), nil
187}
188
189func hgStatus(vcsHg *Cmd, rootDir string) (Status, error) {
190	// Output changeset ID and seconds since epoch.
191	out, err := vcsHg.runOutputVerboseOnly(rootDir, `log -l1 -T {node}:{date|hgdate}`)
192	if err != nil {
193		return Status{}, err
194	}
195
196	// Successful execution without output indicates an empty repo (no commits).
197	var rev string
198	var commitTime time.Time
199	if len(out) > 0 {
200		// Strip trailing timezone offset.
201		if i := bytes.IndexByte(out, ' '); i > 0 {
202			out = out[:i]
203		}
204		rev, commitTime, err = parseRevTime(out)
205		if err != nil {
206			return Status{}, err
207		}
208	}
209
210	// Also look for untracked files.
211	out, err = vcsHg.runOutputVerboseOnly(rootDir, "status")
212	if err != nil {
213		return Status{}, err
214	}
215	uncommitted := len(out) > 0
216
217	return Status{
218		Revision:    rev,
219		CommitTime:  commitTime,
220		Uncommitted: uncommitted,
221	}, nil
222}
223
224// parseRevTime parses commit details in "revision:seconds" format.
225func parseRevTime(out []byte) (string, time.Time, error) {
226	buf := string(bytes.TrimSpace(out))
227
228	i := strings.IndexByte(buf, ':')
229	if i < 1 {
230		return "", time.Time{}, errors.New("unrecognized VCS tool output")
231	}
232	rev := buf[:i]
233
234	secs, err := strconv.ParseInt(string(buf[i+1:]), 10, 64)
235	if err != nil {
236		return "", time.Time{}, fmt.Errorf("unrecognized VCS tool output: %v", err)
237	}
238
239	return rev, time.Unix(secs, 0), nil
240}
241
242// vcsGit describes how to use Git.
243var vcsGit = &Cmd{
244	Name: "Git",
245	Cmd:  "git",
246	RootNames: []rootName{
247		{filename: ".git", isDir: true},
248	},
249
250	CreateCmd:   []string{"clone -- {repo} {dir}", "-go-internal-cd {dir} submodule update --init --recursive"},
251	DownloadCmd: []string{"pull --ff-only", "submodule update --init --recursive"},
252
253	TagCmd: []tagCmd{
254		// tags/xxx matches a git tag named xxx
255		// origin/xxx matches a git branch named xxx on the default remote repository
256		{"show-ref", `(?:tags|origin)/(\S+)$`},
257	},
258	TagLookupCmd: []tagCmd{
259		{"show-ref tags/{tag} origin/{tag}", `((?:tags|origin)/\S+)$`},
260	},
261	TagSyncCmd: []string{"checkout {tag}", "submodule update --init --recursive"},
262	// both createCmd and downloadCmd update the working dir.
263	// No need to do more here. We used to 'checkout master'
264	// but that doesn't work if the default branch is not named master.
265	// DO NOT add 'checkout master' here.
266	// See golang.org/issue/9032.
267	TagSyncDefault: []string{"submodule update --init --recursive"},
268
269	Scheme: []string{"git", "https", "http", "git+ssh", "ssh"},
270
271	// Leave out the '--' separator in the ls-remote command: git 2.7.4 does not
272	// support such a separator for that command, and this use should be safe
273	// without it because the {scheme} value comes from the predefined list above.
274	// See golang.org/issue/33836.
275	PingCmd: "ls-remote {scheme}://{repo}",
276
277	RemoteRepo: gitRemoteRepo,
278	Status:     gitStatus,
279}
280
281// scpSyntaxRe matches the SCP-like addresses used by Git to access
282// repositories by SSH.
283var scpSyntaxRe = lazyregexp.New(`^(\w+)@([\w.-]+):(.*)$`)
284
285func gitRemoteRepo(vcsGit *Cmd, rootDir string) (remoteRepo string, err error) {
286	const cmd = "config remote.origin.url"
287	outb, err := vcsGit.run1(rootDir, cmd, nil, false)
288	if err != nil {
289		// if it doesn't output any message, it means the config argument is correct,
290		// but the config value itself doesn't exist
291		if outb != nil && len(outb) == 0 {
292			return "", errors.New("remote origin not found")
293		}
294		return "", err
295	}
296	out := strings.TrimSpace(string(outb))
297
298	var repoURL *urlpkg.URL
299	if m := scpSyntaxRe.FindStringSubmatch(out); m != nil {
300		// Match SCP-like syntax and convert it to a URL.
301		// Eg, "[email protected]:user/repo" becomes
302		// "ssh://[email protected]/user/repo".
303		repoURL = &urlpkg.URL{
304			Scheme: "ssh",
305			User:   urlpkg.User(m[1]),
306			Host:   m[2],
307			Path:   m[3],
308		}
309	} else {
310		repoURL, err = urlpkg.Parse(out)
311		if err != nil {
312			return "", err
313		}
314	}
315
316	// Iterate over insecure schemes too, because this function simply
317	// reports the state of the repo. If we can't see insecure schemes then
318	// we can't report the actual repo URL.
319	for _, s := range vcsGit.Scheme {
320		if repoURL.Scheme == s {
321			return repoURL.String(), nil
322		}
323	}
324	return "", errors.New("unable to parse output of git " + cmd)
325}
326
327func gitStatus(vcsGit *Cmd, rootDir string) (Status, error) {
328	out, err := vcsGit.runOutputVerboseOnly(rootDir, "status --porcelain")
329	if err != nil {
330		return Status{}, err
331	}
332	uncommitted := len(out) > 0
333
334	// "git status" works for empty repositories, but "git log" does not.
335	// Assume there are no commits in the repo when "git log" fails with
336	// uncommitted files and skip tagging revision / committime.
337	var rev string
338	var commitTime time.Time
339	out, err = vcsGit.runOutputVerboseOnly(rootDir, "-c log.showsignature=false log -1 --format=%H:%ct")
340	if err != nil && !uncommitted {
341		return Status{}, err
342	} else if err == nil {
343		rev, commitTime, err = parseRevTime(out)
344		if err != nil {
345			return Status{}, err
346		}
347	}
348
349	return Status{
350		Revision:    rev,
351		CommitTime:  commitTime,
352		Uncommitted: uncommitted,
353	}, nil
354}
355
356// vcsBzr describes how to use Bazaar.
357var vcsBzr = &Cmd{
358	Name: "Bazaar",
359	Cmd:  "bzr",
360	RootNames: []rootName{
361		{filename: ".bzr", isDir: true},
362	},
363
364	CreateCmd: []string{"branch -- {repo} {dir}"},
365
366	// Without --overwrite bzr will not pull tags that changed.
367	// Replace by --overwrite-tags after http://pad.lv/681792 goes in.
368	DownloadCmd: []string{"pull --overwrite"},
369
370	TagCmd:         []tagCmd{{"tags", `^(\S+)`}},
371	TagSyncCmd:     []string{"update -r {tag}"},
372	TagSyncDefault: []string{"update -r revno:-1"},
373
374	Scheme:      []string{"https", "http", "bzr", "bzr+ssh"},
375	PingCmd:     "info -- {scheme}://{repo}",
376	RemoteRepo:  bzrRemoteRepo,
377	ResolveRepo: bzrResolveRepo,
378	Status:      bzrStatus,
379}
380
381func bzrRemoteRepo(vcsBzr *Cmd, rootDir string) (remoteRepo string, err error) {
382	outb, err := vcsBzr.runOutput(rootDir, "config parent_location")
383	if err != nil {
384		return "", err
385	}
386	return strings.TrimSpace(string(outb)), nil
387}
388
389func bzrResolveRepo(vcsBzr *Cmd, rootDir, remoteRepo string) (realRepo string, err error) {
390	outb, err := vcsBzr.runOutput(rootDir, "info "+remoteRepo)
391	if err != nil {
392		return "", err
393	}
394	out := string(outb)
395
396	// Expect:
397	// ...
398	//   (branch root|repository branch): <URL>
399	// ...
400
401	found := false
402	for _, prefix := range []string{"\n  branch root: ", "\n  repository branch: "} {
403		i := strings.Index(out, prefix)
404		if i >= 0 {
405			out = out[i+len(prefix):]
406			found = true
407			break
408		}
409	}
410	if !found {
411		return "", fmt.Errorf("unable to parse output of bzr info")
412	}
413
414	i := strings.Index(out, "\n")
415	if i < 0 {
416		return "", fmt.Errorf("unable to parse output of bzr info")
417	}
418	out = out[:i]
419	return strings.TrimSpace(out), nil
420}
421
422func bzrStatus(vcsBzr *Cmd, rootDir string) (Status, error) {
423	outb, err := vcsBzr.runOutputVerboseOnly(rootDir, "version-info")
424	if err != nil {
425		return Status{}, err
426	}
427	out := string(outb)
428
429	// Expect (non-empty repositories only):
430	//
431	// revision-id: [email protected]
432	// date: 2021-09-21 12:00:00 +1000
433	// ...
434	var rev string
435	var commitTime time.Time
436
437	for _, line := range strings.Split(out, "\n") {
438		i := strings.IndexByte(line, ':')
439		if i < 0 {
440			continue
441		}
442		key := line[:i]
443		value := strings.TrimSpace(line[i+1:])
444
445		switch key {
446		case "revision-id":
447			rev = value
448		case "date":
449			var err error
450			commitTime, err = time.Parse("2006-01-02 15:04:05 -0700", value)
451			if err != nil {
452				return Status{}, errors.New("unable to parse output of bzr version-info")
453			}
454		}
455	}
456
457	outb, err = vcsBzr.runOutputVerboseOnly(rootDir, "status")
458	if err != nil {
459		return Status{}, err
460	}
461
462	// Skip warning when working directory is set to an older revision.
463	if bytes.HasPrefix(outb, []byte("working tree is out of date")) {
464		i := bytes.IndexByte(outb, '\n')
465		if i < 0 {
466			i = len(outb)
467		}
468		outb = outb[:i]
469	}
470	uncommitted := len(outb) > 0
471
472	return Status{
473		Revision:    rev,
474		CommitTime:  commitTime,
475		Uncommitted: uncommitted,
476	}, nil
477}
478
479// vcsSvn describes how to use Subversion.
480var vcsSvn = &Cmd{
481	Name: "Subversion",
482	Cmd:  "svn",
483	RootNames: []rootName{
484		{filename: ".svn", isDir: true},
485	},
486
487	CreateCmd:   []string{"checkout -- {repo} {dir}"},
488	DownloadCmd: []string{"update"},
489
490	// There is no tag command in subversion.
491	// The branch information is all in the path names.
492
493	Scheme:     []string{"https", "http", "svn", "svn+ssh"},
494	PingCmd:    "info -- {scheme}://{repo}",
495	RemoteRepo: svnRemoteRepo,
496}
497
498func svnRemoteRepo(vcsSvn *Cmd, rootDir string) (remoteRepo string, err error) {
499	outb, err := vcsSvn.runOutput(rootDir, "info")
500	if err != nil {
501		return "", err
502	}
503	out := string(outb)
504
505	// Expect:
506	//
507	//	 ...
508	// 	URL: <URL>
509	// 	...
510	//
511	// Note that we're not using the Repository Root line,
512	// because svn allows checking out subtrees.
513	// The URL will be the URL of the subtree (what we used with 'svn co')
514	// while the Repository Root may be a much higher parent.
515	i := strings.Index(out, "\nURL: ")
516	if i < 0 {
517		return "", fmt.Errorf("unable to parse output of svn info")
518	}
519	out = out[i+len("\nURL: "):]
520	i = strings.Index(out, "\n")
521	if i < 0 {
522		return "", fmt.Errorf("unable to parse output of svn info")
523	}
524	out = out[:i]
525	return strings.TrimSpace(out), nil
526}
527
528// fossilRepoName is the name go get associates with a fossil repository. In the
529// real world the file can be named anything.
530const fossilRepoName = ".fossil"
531
532// vcsFossil describes how to use Fossil (fossil-scm.org)
533var vcsFossil = &Cmd{
534	Name: "Fossil",
535	Cmd:  "fossil",
536	RootNames: []rootName{
537		{filename: ".fslckout", isDir: false},
538		{filename: "_FOSSIL_", isDir: false},
539	},
540
541	CreateCmd:   []string{"-go-internal-mkdir {dir} clone -- {repo} " + filepath.Join("{dir}", fossilRepoName), "-go-internal-cd {dir} open .fossil"},
542	DownloadCmd: []string{"up"},
543
544	TagCmd:         []tagCmd{{"tag ls", `(.*)`}},
545	TagSyncCmd:     []string{"up tag:{tag}"},
546	TagSyncDefault: []string{"up trunk"},
547
548	Scheme:     []string{"https", "http"},
549	RemoteRepo: fossilRemoteRepo,
550	Status:     fossilStatus,
551}
552
553func fossilRemoteRepo(vcsFossil *Cmd, rootDir string) (remoteRepo string, err error) {
554	out, err := vcsFossil.runOutput(rootDir, "remote-url")
555	if err != nil {
556		return "", err
557	}
558	return strings.TrimSpace(string(out)), nil
559}
560
561var errFossilInfo = errors.New("unable to parse output of fossil info")
562
563func fossilStatus(vcsFossil *Cmd, rootDir string) (Status, error) {
564	outb, err := vcsFossil.runOutputVerboseOnly(rootDir, "info")
565	if err != nil {
566		return Status{}, err
567	}
568	out := string(outb)
569
570	// Expect:
571	// ...
572	// checkout:     91ed71f22c77be0c3e250920f47bfd4e1f9024d2 2021-09-21 12:00:00 UTC
573	// ...
574
575	// Extract revision and commit time.
576	// Ensure line ends with UTC (known timezone offset).
577	const prefix = "\ncheckout:"
578	const suffix = " UTC"
579	i := strings.Index(out, prefix)
580	if i < 0 {
581		return Status{}, errFossilInfo
582	}
583	checkout := out[i+len(prefix):]
584	i = strings.Index(checkout, suffix)
585	if i < 0 {
586		return Status{}, errFossilInfo
587	}
588	checkout = strings.TrimSpace(checkout[:i])
589
590	i = strings.IndexByte(checkout, ' ')
591	if i < 0 {
592		return Status{}, errFossilInfo
593	}
594	rev := checkout[:i]
595
596	commitTime, err := time.ParseInLocation(time.DateTime, checkout[i+1:], time.UTC)
597	if err != nil {
598		return Status{}, fmt.Errorf("%v: %v", errFossilInfo, err)
599	}
600
601	// Also look for untracked changes.
602	outb, err = vcsFossil.runOutputVerboseOnly(rootDir, "changes --differ")
603	if err != nil {
604		return Status{}, err
605	}
606	uncommitted := len(outb) > 0
607
608	return Status{
609		Revision:    rev,
610		CommitTime:  commitTime,
611		Uncommitted: uncommitted,
612	}, nil
613}
614
615func (v *Cmd) String() string {
616	return v.Name
617}
618
619// run runs the command line cmd in the given directory.
620// keyval is a list of key, value pairs. run expands
621// instances of {key} in cmd into value, but only after
622// splitting cmd into individual arguments.
623// If an error occurs, run prints the command line and the
624// command's combined stdout+stderr to standard error.
625// Otherwise run discards the command's output.
626func (v *Cmd) run(dir string, cmd string, keyval ...string) error {
627	_, err := v.run1(dir, cmd, keyval, true)
628	return err
629}
630
631// runVerboseOnly is like run but only generates error output to standard error in verbose mode.
632func (v *Cmd) runVerboseOnly(dir string, cmd string, keyval ...string) error {
633	_, err := v.run1(dir, cmd, keyval, false)
634	return err
635}
636
637// runOutput is like run but returns the output of the command.
638func (v *Cmd) runOutput(dir string, cmd string, keyval ...string) ([]byte, error) {
639	return v.run1(dir, cmd, keyval, true)
640}
641
642// runOutputVerboseOnly is like runOutput but only generates error output to
643// standard error in verbose mode.
644func (v *Cmd) runOutputVerboseOnly(dir string, cmd string, keyval ...string) ([]byte, error) {
645	return v.run1(dir, cmd, keyval, false)
646}
647
648// run1 is the generalized implementation of run and runOutput.
649func (v *Cmd) run1(dir string, cmdline string, keyval []string, verbose bool) ([]byte, error) {
650	m := make(map[string]string)
651	for i := 0; i < len(keyval); i += 2 {
652		m[keyval[i]] = keyval[i+1]
653	}
654	args := strings.Fields(cmdline)
655	for i, arg := range args {
656		args[i] = expand(m, arg)
657	}
658
659	if len(args) >= 2 && args[0] == "-go-internal-mkdir" {
660		var err error
661		if filepath.IsAbs(args[1]) {
662			err = os.Mkdir(args[1], fs.ModePerm)
663		} else {
664			err = os.Mkdir(filepath.Join(dir, args[1]), fs.ModePerm)
665		}
666		if err != nil {
667			return nil, err
668		}
669		args = args[2:]
670	}
671
672	if len(args) >= 2 && args[0] == "-go-internal-cd" {
673		if filepath.IsAbs(args[1]) {
674			dir = args[1]
675		} else {
676			dir = filepath.Join(dir, args[1])
677		}
678		args = args[2:]
679	}
680
681	_, err := cfg.LookPath(v.Cmd)
682	if err != nil {
683		fmt.Fprintf(os.Stderr,
684			"go: missing %s command. See https://golang.org/s/gogetcmd\n",
685			v.Name)
686		return nil, err
687	}
688
689	cmd := exec.Command(v.Cmd, args...)
690	cmd.Dir = dir
691	if cfg.BuildX {
692		fmt.Fprintf(os.Stderr, "cd %s\n", dir)
693		fmt.Fprintf(os.Stderr, "%s %s\n", v.Cmd, strings.Join(args, " "))
694	}
695	out, err := cmd.Output()
696	if err != nil {
697		if verbose || cfg.BuildV {
698			fmt.Fprintf(os.Stderr, "# cd %s; %s %s\n", dir, v.Cmd, strings.Join(args, " "))
699			if ee, ok := err.(*exec.ExitError); ok && len(ee.Stderr) > 0 {
700				os.Stderr.Write(ee.Stderr)
701			} else {
702				fmt.Fprintln(os.Stderr, err.Error())
703			}
704		}
705	}
706	return out, err
707}
708
709// Ping pings to determine scheme to use.
710func (v *Cmd) Ping(scheme, repo string) error {
711	// Run the ping command in an arbitrary working directory,
712	// but don't let the current working directory pollute the results.
713	// In module mode, we expect GOMODCACHE to exist and be a safe place for
714	// commands; in GOPATH mode, we expect that to be true of GOPATH/src.
715	dir := cfg.GOMODCACHE
716	if !cfg.ModulesEnabled {
717		dir = filepath.Join(cfg.BuildContext.GOPATH, "src")
718	}
719	os.MkdirAll(dir, 0777) // Ignore errors — if unsuccessful, the command will likely fail.
720
721	release, err := base.AcquireNet()
722	if err != nil {
723		return err
724	}
725	defer release()
726
727	return v.runVerboseOnly(dir, v.PingCmd, "scheme", scheme, "repo", repo)
728}
729
730// Create creates a new copy of repo in dir.
731// The parent of dir must exist; dir must not.
732func (v *Cmd) Create(dir, repo string) error {
733	release, err := base.AcquireNet()
734	if err != nil {
735		return err
736	}
737	defer release()
738
739	for _, cmd := range v.CreateCmd {
740		if err := v.run(filepath.Dir(dir), cmd, "dir", dir, "repo", repo); err != nil {
741			return err
742		}
743	}
744	return nil
745}
746
747// Download downloads any new changes for the repo in dir.
748func (v *Cmd) Download(dir string) error {
749	release, err := base.AcquireNet()
750	if err != nil {
751		return err
752	}
753	defer release()
754
755	for _, cmd := range v.DownloadCmd {
756		if err := v.run(dir, cmd); err != nil {
757			return err
758		}
759	}
760	return nil
761}
762
763// Tags returns the list of available tags for the repo in dir.
764func (v *Cmd) Tags(dir string) ([]string, error) {
765	var tags []string
766	for _, tc := range v.TagCmd {
767		out, err := v.runOutput(dir, tc.cmd)
768		if err != nil {
769			return nil, err
770		}
771		re := regexp.MustCompile(`(?m-s)` + tc.pattern)
772		for _, m := range re.FindAllStringSubmatch(string(out), -1) {
773			tags = append(tags, m[1])
774		}
775	}
776	return tags, nil
777}
778
779// TagSync syncs the repo in dir to the named tag,
780// which either is a tag returned by tags or is v.tagDefault.
781func (v *Cmd) TagSync(dir, tag string) error {
782	if v.TagSyncCmd == nil {
783		return nil
784	}
785	if tag != "" {
786		for _, tc := range v.TagLookupCmd {
787			out, err := v.runOutput(dir, tc.cmd, "tag", tag)
788			if err != nil {
789				return err
790			}
791			re := regexp.MustCompile(`(?m-s)` + tc.pattern)
792			m := re.FindStringSubmatch(string(out))
793			if len(m) > 1 {
794				tag = m[1]
795				break
796			}
797		}
798	}
799
800	release, err := base.AcquireNet()
801	if err != nil {
802		return err
803	}
804	defer release()
805
806	if tag == "" && v.TagSyncDefault != nil {
807		for _, cmd := range v.TagSyncDefault {
808			if err := v.run(dir, cmd); err != nil {
809				return err
810			}
811		}
812		return nil
813	}
814
815	for _, cmd := range v.TagSyncCmd {
816		if err := v.run(dir, cmd, "tag", tag); err != nil {
817			return err
818		}
819	}
820	return nil
821}
822
823// A vcsPath describes how to convert an import path into a
824// version control system and repository name.
825type vcsPath struct {
826	pathPrefix     string                              // prefix this description applies to
827	regexp         *lazyregexp.Regexp                  // compiled pattern for import path
828	repo           string                              // repository to use (expand with match of re)
829	vcs            string                              // version control system to use (expand with match of re)
830	check          func(match map[string]string) error // additional checks
831	schemelessRepo bool                                // if true, the repo pattern lacks a scheme
832}
833
834// FromDir inspects dir and its parents to determine the
835// version control system and code repository to use.
836// If no repository is found, FromDir returns an error
837// equivalent to os.ErrNotExist.
838func FromDir(dir, srcRoot string, allowNesting bool) (repoDir string, vcsCmd *Cmd, err error) {
839	// Clean and double-check that dir is in (a subdirectory of) srcRoot.
840	dir = filepath.Clean(dir)
841	if srcRoot != "" {
842		srcRoot = filepath.Clean(srcRoot)
843		if len(dir) <= len(srcRoot) || dir[len(srcRoot)] != filepath.Separator {
844			return "", nil, fmt.Errorf("directory %q is outside source root %q", dir, srcRoot)
845		}
846	}
847
848	origDir := dir
849	for len(dir) > len(srcRoot) {
850		for _, vcs := range vcsList {
851			if isVCSRoot(dir, vcs.RootNames) {
852				// Record first VCS we find.
853				// If allowNesting is false (as it is in GOPATH), keep looking for
854				// repositories in parent directories and report an error if one is
855				// found to mitigate VCS injection attacks.
856				if vcsCmd == nil {
857					vcsCmd = vcs
858					repoDir = dir
859					if allowNesting {
860						return repoDir, vcsCmd, nil
861					}
862					continue
863				}
864				// Otherwise, we have one VCS inside a different VCS.
865				return "", nil, fmt.Errorf("directory %q uses %s, but parent %q uses %s",
866					repoDir, vcsCmd.Cmd, dir, vcs.Cmd)
867			}
868		}
869
870		// Move to parent.
871		ndir := filepath.Dir(dir)
872		if len(ndir) >= len(dir) {
873			break
874		}
875		dir = ndir
876	}
877	if vcsCmd == nil {
878		return "", nil, &vcsNotFoundError{dir: origDir}
879	}
880	return repoDir, vcsCmd, nil
881}
882
883// isVCSRoot identifies a VCS root by checking whether the directory contains
884// any of the listed root names.
885func isVCSRoot(dir string, rootNames []rootName) bool {
886	for _, root := range rootNames {
887		fi, err := os.Stat(filepath.Join(dir, root.filename))
888		if err == nil && fi.IsDir() == root.isDir {
889			return true
890		}
891	}
892
893	return false
894}
895
896type rootName struct {
897	filename string
898	isDir    bool
899}
900
901type vcsNotFoundError struct {
902	dir string
903}
904
905func (e *vcsNotFoundError) Error() string {
906	return fmt.Sprintf("directory %q is not using a known version control system", e.dir)
907}
908
909func (e *vcsNotFoundError) Is(err error) bool {
910	return err == os.ErrNotExist
911}
912
913// A govcsRule is a single GOVCS rule like private:hg|svn.
914type govcsRule struct {
915	pattern string
916	allowed []string
917}
918
919// A govcsConfig is a full GOVCS configuration.
920type govcsConfig []govcsRule
921
922func parseGOVCS(s string) (govcsConfig, error) {
923	s = strings.TrimSpace(s)
924	if s == "" {
925		return nil, nil
926	}
927	var cfg govcsConfig
928	have := make(map[string]string)
929	for _, item := range strings.Split(s, ",") {
930		item = strings.TrimSpace(item)
931		if item == "" {
932			return nil, fmt.Errorf("empty entry in GOVCS")
933		}
934		pattern, list, found := strings.Cut(item, ":")
935		if !found {
936			return nil, fmt.Errorf("malformed entry in GOVCS (missing colon): %q", item)
937		}
938		pattern, list = strings.TrimSpace(pattern), strings.TrimSpace(list)
939		if pattern == "" {
940			return nil, fmt.Errorf("empty pattern in GOVCS: %q", item)
941		}
942		if list == "" {
943			return nil, fmt.Errorf("empty VCS list in GOVCS: %q", item)
944		}
945		if search.IsRelativePath(pattern) {
946			return nil, fmt.Errorf("relative pattern not allowed in GOVCS: %q", pattern)
947		}
948		if old := have[pattern]; old != "" {
949			return nil, fmt.Errorf("unreachable pattern in GOVCS: %q after %q", item, old)
950		}
951		have[pattern] = item
952		allowed := strings.Split(list, "|")
953		for i, a := range allowed {
954			a = strings.TrimSpace(a)
955			if a == "" {
956				return nil, fmt.Errorf("empty VCS name in GOVCS: %q", item)
957			}
958			allowed[i] = a
959		}
960		cfg = append(cfg, govcsRule{pattern, allowed})
961	}
962	return cfg, nil
963}
964
965func (c *govcsConfig) allow(path string, private bool, vcs string) bool {
966	for _, rule := range *c {
967		match := false
968		switch rule.pattern {
969		case "private":
970			match = private
971		case "public":
972			match = !private
973		default:
974			// Note: rule.pattern is known to be comma-free,
975			// so MatchPrefixPatterns is only matching a single pattern for us.
976			match = module.MatchPrefixPatterns(rule.pattern, path)
977		}
978		if !match {
979			continue
980		}
981		for _, allow := range rule.allowed {
982			if allow == vcs || allow == "all" {
983				return true
984			}
985		}
986		return false
987	}
988
989	// By default, nothing is allowed.
990	return false
991}
992
993var (
994	govcs     govcsConfig
995	govcsErr  error
996	govcsOnce sync.Once
997)
998
999// defaultGOVCS is the default setting for GOVCS.
1000// Setting GOVCS adds entries ahead of these but does not remove them.
1001// (They are appended to the parsed GOVCS setting.)
1002//
1003// The rationale behind allowing only Git and Mercurial is that
1004// these two systems have had the most attention to issues
1005// of being run as clients of untrusted servers. In contrast,
1006// Bazaar, Fossil, and Subversion have primarily been used
1007// in trusted, authenticated environments and are not as well
1008// scrutinized as attack surfaces.
1009//
1010// See golang.org/issue/41730 for details.
1011var defaultGOVCS = govcsConfig{
1012	{"private", []string{"all"}},
1013	{"public", []string{"git", "hg"}},
1014}
1015
1016// checkGOVCS checks whether the policy defined by the environment variable
1017// GOVCS allows the given vcs command to be used with the given repository
1018// root path. Note that root may not be a real package or module path; it's
1019// the same as the root path in the go-import meta tag.
1020func checkGOVCS(vcs *Cmd, root string) error {
1021	if vcs == vcsMod {
1022		// Direct module (proxy protocol) fetches don't
1023		// involve an external version control system
1024		// and are always allowed.
1025		return nil
1026	}
1027
1028	govcsOnce.Do(func() {
1029		govcs, govcsErr = parseGOVCS(os.Getenv("GOVCS"))
1030		govcs = append(govcs, defaultGOVCS...)
1031	})
1032	if govcsErr != nil {
1033		return govcsErr
1034	}
1035
1036	private := module.MatchPrefixPatterns(cfg.GOPRIVATE, root)
1037	if !govcs.allow(root, private, vcs.Cmd) {
1038		what := "public"
1039		if private {
1040			what = "private"
1041		}
1042		return fmt.Errorf("GOVCS disallows using %s for %s %s; see 'go help vcs'", vcs.Cmd, what, root)
1043	}
1044
1045	return nil
1046}
1047
1048// RepoRoot describes the repository root for a tree of source code.
1049type RepoRoot struct {
1050	Repo     string // repository URL, including scheme
1051	Root     string // import path corresponding to root of repo
1052	IsCustom bool   // defined by served <meta> tags (as opposed to hard-coded pattern)
1053	VCS      *Cmd
1054}
1055
1056func httpPrefix(s string) string {
1057	for _, prefix := range [...]string{"http:", "https:"} {
1058		if strings.HasPrefix(s, prefix) {
1059			return prefix
1060		}
1061	}
1062	return ""
1063}
1064
1065// ModuleMode specifies whether to prefer modules when looking up code sources.
1066type ModuleMode int
1067
1068const (
1069	IgnoreMod ModuleMode = iota
1070	PreferMod
1071)
1072
1073// RepoRootForImportPath analyzes importPath to determine the
1074// version control system, and code repository to use.
1075func RepoRootForImportPath(importPath string, mod ModuleMode, security web.SecurityMode) (*RepoRoot, error) {
1076	rr, err := repoRootFromVCSPaths(importPath, security, vcsPaths)
1077	if err == errUnknownSite {
1078		rr, err = repoRootForImportDynamic(importPath, mod, security)
1079		if err != nil {
1080			err = importErrorf(importPath, "unrecognized import path %q: %v", importPath, err)
1081		}
1082	}
1083	if err != nil {
1084		rr1, err1 := repoRootFromVCSPaths(importPath, security, vcsPathsAfterDynamic)
1085		if err1 == nil {
1086			rr = rr1
1087			err = nil
1088		}
1089	}
1090
1091	// Should have been taken care of above, but make sure.
1092	if err == nil && strings.Contains(importPath, "...") && strings.Contains(rr.Root, "...") {
1093		// Do not allow wildcards in the repo root.
1094		rr = nil
1095		err = importErrorf(importPath, "cannot expand ... in %q", importPath)
1096	}
1097	return rr, err
1098}
1099
1100var errUnknownSite = errors.New("dynamic lookup required to find mapping")
1101
1102// repoRootFromVCSPaths attempts to map importPath to a repoRoot
1103// using the mappings defined in vcsPaths.
1104func repoRootFromVCSPaths(importPath string, security web.SecurityMode, vcsPaths []*vcsPath) (*RepoRoot, error) {
1105	if str.HasPathPrefix(importPath, "example.net") {
1106		// TODO(rsc): This should not be necessary, but it's required to keep
1107		// tests like ../../testdata/script/mod_get_extra.txt from using the network.
1108		// That script has everything it needs in the replacement set, but it is still
1109		// doing network calls.
1110		return nil, fmt.Errorf("no modules on example.net")
1111	}
1112	if importPath == "rsc.io" {
1113		// This special case allows tests like ../../testdata/script/govcs.txt
1114		// to avoid making any network calls. The module lookup for a path
1115		// like rsc.io/nonexist.svn/foo needs to not make a network call for
1116		// a lookup on rsc.io.
1117		return nil, fmt.Errorf("rsc.io is not a module")
1118	}
1119	// A common error is to use https://packagepath because that's what
1120	// hg and git require. Diagnose this helpfully.
1121	if prefix := httpPrefix(importPath); prefix != "" {
1122		// The importPath has been cleaned, so has only one slash. The pattern
1123		// ignores the slashes; the error message puts them back on the RHS at least.
1124		return nil, fmt.Errorf("%q not allowed in import path", prefix+"//")
1125	}
1126	for _, srv := range vcsPaths {
1127		if !str.HasPathPrefix(importPath, srv.pathPrefix) {
1128			continue
1129		}
1130		m := srv.regexp.FindStringSubmatch(importPath)
1131		if m == nil {
1132			if srv.pathPrefix != "" {
1133				return nil, importErrorf(importPath, "invalid %s import path %q", srv.pathPrefix, importPath)
1134			}
1135			continue
1136		}
1137
1138		// Build map of named subexpression matches for expand.
1139		match := map[string]string{
1140			"prefix": srv.pathPrefix + "/",
1141			"import": importPath,
1142		}
1143		for i, name := range srv.regexp.SubexpNames() {
1144			if name != "" && match[name] == "" {
1145				match[name] = m[i]
1146			}
1147		}
1148		if srv.vcs != "" {
1149			match["vcs"] = expand(match, srv.vcs)
1150		}
1151		if srv.repo != "" {
1152			match["repo"] = expand(match, srv.repo)
1153		}
1154		if srv.check != nil {
1155			if err := srv.check(match); err != nil {
1156				return nil, err
1157			}
1158		}
1159		vcs := vcsByCmd(match["vcs"])
1160		if vcs == nil {
1161			return nil, fmt.Errorf("unknown version control system %q", match["vcs"])
1162		}
1163		if err := checkGOVCS(vcs, match["root"]); err != nil {
1164			return nil, err
1165		}
1166		var repoURL string
1167		if !srv.schemelessRepo {
1168			repoURL = match["repo"]
1169		} else {
1170			repo := match["repo"]
1171			var ok bool
1172			repoURL, ok = interceptVCSTest(repo, vcs, security)
1173			if !ok {
1174				scheme, err := func() (string, error) {
1175					for _, s := range vcs.Scheme {
1176						if security == web.SecureOnly && !vcs.isSecureScheme(s) {
1177							continue
1178						}
1179
1180						// If we know how to ping URL schemes for this VCS,
1181						// check that this repo works.
1182						// Otherwise, default to the first scheme
1183						// that meets the requested security level.
1184						if vcs.PingCmd == "" {
1185							return s, nil
1186						}
1187						if err := vcs.Ping(s, repo); err == nil {
1188							return s, nil
1189						}
1190					}
1191					securityFrag := ""
1192					if security == web.SecureOnly {
1193						securityFrag = "secure "
1194					}
1195					return "", fmt.Errorf("no %sprotocol found for repository", securityFrag)
1196				}()
1197				if err != nil {
1198					return nil, err
1199				}
1200				repoURL = scheme + "://" + repo
1201			}
1202		}
1203		rr := &RepoRoot{
1204			Repo: repoURL,
1205			Root: match["root"],
1206			VCS:  vcs,
1207		}
1208		return rr, nil
1209	}
1210	return nil, errUnknownSite
1211}
1212
1213func interceptVCSTest(repo string, vcs *Cmd, security web.SecurityMode) (repoURL string, ok bool) {
1214	if VCSTestRepoURL == "" {
1215		return "", false
1216	}
1217	if vcs == vcsMod {
1218		// Since the "mod" protocol is implemented internally,
1219		// requests will be intercepted at a lower level (in cmd/go/internal/web).
1220		return "", false
1221	}
1222
1223	if scheme, path, ok := strings.Cut(repo, "://"); ok {
1224		if security == web.SecureOnly && !vcs.isSecureScheme(scheme) {
1225			return "", false // Let the caller reject the original URL.
1226		}
1227		repo = path // Remove leading URL scheme if present.
1228	}
1229	for _, host := range VCSTestHosts {
1230		if !str.HasPathPrefix(repo, host) {
1231			continue
1232		}
1233
1234		httpURL := VCSTestRepoURL + strings.TrimPrefix(repo, host)
1235
1236		if vcs == vcsSvn {
1237			// Ping the vcweb HTTP server to tell it to initialize the SVN repository
1238			// and get the SVN server URL.
1239			u, err := urlpkg.Parse(httpURL + "?vcwebsvn=1")
1240			if err != nil {
1241				panic(fmt.Sprintf("invalid vcs-test repo URL: %v", err))
1242			}
1243			svnURL, err := web.GetBytes(u)
1244			svnURL = bytes.TrimSpace(svnURL)
1245			if err == nil && len(svnURL) > 0 {
1246				return string(svnURL) + strings.TrimPrefix(repo, host), true
1247			}
1248
1249			// vcs-test doesn't have a svn handler for the given path,
1250			// so resolve the repo to HTTPS instead.
1251		}
1252
1253		return httpURL, true
1254	}
1255	return "", false
1256}
1257
1258// urlForImportPath returns a partially-populated URL for the given Go import path.
1259//
1260// The URL leaves the Scheme field blank so that web.Get will try any scheme
1261// allowed by the selected security mode.
1262func urlForImportPath(importPath string) (*urlpkg.URL, error) {
1263	slash := strings.Index(importPath, "/")
1264	if slash < 0 {
1265		slash = len(importPath)
1266	}
1267	host, path := importPath[:slash], importPath[slash:]
1268	if !strings.Contains(host, ".") {
1269		return nil, errors.New("import path does not begin with hostname")
1270	}
1271	if len(path) == 0 {
1272		path = "/"
1273	}
1274	return &urlpkg.URL{Host: host, Path: path, RawQuery: "go-get=1"}, nil
1275}
1276
1277// repoRootForImportDynamic finds a *RepoRoot for a custom domain that's not
1278// statically known by repoRootFromVCSPaths.
1279//
1280// This handles custom import paths like "name.tld/pkg/foo" or just "name.tld".
1281func repoRootForImportDynamic(importPath string, mod ModuleMode, security web.SecurityMode) (*RepoRoot, error) {
1282	url, err := urlForImportPath(importPath)
1283	if err != nil {
1284		return nil, err
1285	}
1286	resp, err := web.Get(security, url)
1287	if err != nil {
1288		msg := "https fetch: %v"
1289		if security == web.Insecure {
1290			msg = "http/" + msg
1291		}
1292		return nil, fmt.Errorf(msg, err)
1293	}
1294	body := resp.Body
1295	defer body.Close()
1296	imports, err := parseMetaGoImports(body, mod)
1297	if len(imports) == 0 {
1298		if respErr := resp.Err(); respErr != nil {
1299			// If the server's status was not OK, prefer to report that instead of
1300			// an XML parse error.
1301			return nil, respErr
1302		}
1303	}
1304	if err != nil {
1305		return nil, fmt.Errorf("parsing %s: %v", importPath, err)
1306	}
1307	// Find the matched meta import.
1308	mmi, err := matchGoImport(imports, importPath)
1309	if err != nil {
1310		if _, ok := err.(ImportMismatchError); !ok {
1311			return nil, fmt.Errorf("parse %s: %v", url, err)
1312		}
1313		return nil, fmt.Errorf("parse %s: no go-import meta tags (%s)", resp.URL, err)
1314	}
1315	if cfg.BuildV {
1316		log.Printf("get %q: found meta tag %#v at %s", importPath, mmi, url)
1317	}
1318	// If the import was "uni.edu/bob/project", which said the
1319	// prefix was "uni.edu" and the RepoRoot was "evilroot.com",
1320	// make sure we don't trust Bob and check out evilroot.com to
1321	// "uni.edu" yet (possibly overwriting/preempting another
1322	// non-evil student). Instead, first verify the root and see
1323	// if it matches Bob's claim.
1324	if mmi.Prefix != importPath {
1325		if cfg.BuildV {
1326			log.Printf("get %q: verifying non-authoritative meta tag", importPath)
1327		}
1328		var imports []metaImport
1329		url, imports, err = metaImportsForPrefix(mmi.Prefix, mod, security)
1330		if err != nil {
1331			return nil, err
1332		}
1333		metaImport2, err := matchGoImport(imports, importPath)
1334		if err != nil || mmi != metaImport2 {
1335			return nil, fmt.Errorf("%s and %s disagree about go-import for %s", resp.URL, url, mmi.Prefix)
1336		}
1337	}
1338
1339	if err := validateRepoRoot(mmi.RepoRoot); err != nil {
1340		return nil, fmt.Errorf("%s: invalid repo root %q: %v", resp.URL, mmi.RepoRoot, err)
1341	}
1342	var vcs *Cmd
1343	if mmi.VCS == "mod" {
1344		vcs = vcsMod
1345	} else {
1346		vcs = vcsByCmd(mmi.VCS)
1347		if vcs == nil {
1348			return nil, fmt.Errorf("%s: unknown vcs %q", resp.URL, mmi.VCS)
1349		}
1350	}
1351
1352	if err := checkGOVCS(vcs, mmi.Prefix); err != nil {
1353		return nil, err
1354	}
1355
1356	repoURL, ok := interceptVCSTest(mmi.RepoRoot, vcs, security)
1357	if !ok {
1358		repoURL = mmi.RepoRoot
1359	}
1360	rr := &RepoRoot{
1361		Repo:     repoURL,
1362		Root:     mmi.Prefix,
1363		IsCustom: true,
1364		VCS:      vcs,
1365	}
1366	return rr, nil
1367}
1368
1369// validateRepoRoot returns an error if repoRoot does not seem to be
1370// a valid URL with scheme.
1371func validateRepoRoot(repoRoot string) error {
1372	url, err := urlpkg.Parse(repoRoot)
1373	if err != nil {
1374		return err
1375	}
1376	if url.Scheme == "" {
1377		return errors.New("no scheme")
1378	}
1379	if url.Scheme == "file" {
1380		return errors.New("file scheme disallowed")
1381	}
1382	return nil
1383}
1384
1385var fetchGroup singleflight.Group
1386var (
1387	fetchCacheMu sync.Mutex
1388	fetchCache   = map[string]fetchResult{} // key is metaImportsForPrefix's importPrefix
1389)
1390
1391// metaImportsForPrefix takes a package's root import path as declared in a <meta> tag
1392// and returns its HTML discovery URL and the parsed metaImport lines
1393// found on the page.
1394//
1395// The importPath is of the form "golang.org/x/tools".
1396// It is an error if no imports are found.
1397// url will still be valid if err != nil.
1398// The returned url will be of the form "https://golang.org/x/tools?go-get=1"
1399func metaImportsForPrefix(importPrefix string, mod ModuleMode, security web.SecurityMode) (*urlpkg.URL, []metaImport, error) {
1400	setCache := func(res fetchResult) (fetchResult, error) {
1401		fetchCacheMu.Lock()
1402		defer fetchCacheMu.Unlock()
1403		fetchCache[importPrefix] = res
1404		return res, nil
1405	}
1406
1407	resi, _, _ := fetchGroup.Do(importPrefix, func() (resi any, err error) {
1408		fetchCacheMu.Lock()
1409		if res, ok := fetchCache[importPrefix]; ok {
1410			fetchCacheMu.Unlock()
1411			return res, nil
1412		}
1413		fetchCacheMu.Unlock()
1414
1415		url, err := urlForImportPath(importPrefix)
1416		if err != nil {
1417			return setCache(fetchResult{err: err})
1418		}
1419		resp, err := web.Get(security, url)
1420		if err != nil {
1421			return setCache(fetchResult{url: url, err: fmt.Errorf("fetching %s: %v", importPrefix, err)})
1422		}
1423		body := resp.Body
1424		defer body.Close()
1425		imports, err := parseMetaGoImports(body, mod)
1426		if len(imports) == 0 {
1427			if respErr := resp.Err(); respErr != nil {
1428				// If the server's status was not OK, prefer to report that instead of
1429				// an XML parse error.
1430				return setCache(fetchResult{url: url, err: respErr})
1431			}
1432		}
1433		if err != nil {
1434			return setCache(fetchResult{url: url, err: fmt.Errorf("parsing %s: %v", resp.URL, err)})
1435		}
1436		if len(imports) == 0 {
1437			err = fmt.Errorf("fetching %s: no go-import meta tag found in %s", importPrefix, resp.URL)
1438		}
1439		return setCache(fetchResult{url: url, imports: imports, err: err})
1440	})
1441	res := resi.(fetchResult)
1442	return res.url, res.imports, res.err
1443}
1444
1445type fetchResult struct {
1446	url     *urlpkg.URL
1447	imports []metaImport
1448	err     error
1449}
1450
1451// metaImport represents the parsed <meta name="go-import"
1452// content="prefix vcs reporoot" /> tags from HTML files.
1453type metaImport struct {
1454	Prefix, VCS, RepoRoot string
1455}
1456
1457// An ImportMismatchError is returned where metaImport/s are present
1458// but none match our import path.
1459type ImportMismatchError struct {
1460	importPath string
1461	mismatches []string // the meta imports that were discarded for not matching our importPath
1462}
1463
1464func (m ImportMismatchError) Error() string {
1465	formattedStrings := make([]string, len(m.mismatches))
1466	for i, pre := range m.mismatches {
1467		formattedStrings[i] = fmt.Sprintf("meta tag %s did not match import path %s", pre, m.importPath)
1468	}
1469	return strings.Join(formattedStrings, ", ")
1470}
1471
1472// matchGoImport returns the metaImport from imports matching importPath.
1473// An error is returned if there are multiple matches.
1474// An ImportMismatchError is returned if none match.
1475func matchGoImport(imports []metaImport, importPath string) (metaImport, error) {
1476	match := -1
1477
1478	errImportMismatch := ImportMismatchError{importPath: importPath}
1479	for i, im := range imports {
1480		if !str.HasPathPrefix(importPath, im.Prefix) {
1481			errImportMismatch.mismatches = append(errImportMismatch.mismatches, im.Prefix)
1482			continue
1483		}
1484
1485		if match >= 0 {
1486			if imports[match].VCS == "mod" && im.VCS != "mod" {
1487				// All the mod entries precede all the non-mod entries.
1488				// We have a mod entry and don't care about the rest,
1489				// matching or not.
1490				break
1491			}
1492			return metaImport{}, fmt.Errorf("multiple meta tags match import path %q", importPath)
1493		}
1494		match = i
1495	}
1496
1497	if match == -1 {
1498		return metaImport{}, errImportMismatch
1499	}
1500	return imports[match], nil
1501}
1502
1503// expand rewrites s to replace {k} with match[k] for each key k in match.
1504func expand(match map[string]string, s string) string {
1505	// We want to replace each match exactly once, and the result of expansion
1506	// must not depend on the iteration order through the map.
1507	// A strings.Replacer has exactly the properties we're looking for.
1508	oldNew := make([]string, 0, 2*len(match))
1509	for k, v := range match {
1510		oldNew = append(oldNew, "{"+k+"}", v)
1511	}
1512	return strings.NewReplacer(oldNew...).Replace(s)
1513}
1514
1515// vcsPaths defines the meaning of import paths referring to
1516// commonly-used VCS hosting sites (github.com/user/dir)
1517// and import paths referring to a fully-qualified importPath
1518// containing a VCS type (foo.com/repo.git/dir)
1519var vcsPaths = []*vcsPath{
1520	// GitHub
1521	{
1522		pathPrefix: "github.com",
1523		regexp:     lazyregexp.New(`^(?P<root>github\.com/[\w.\-]+/[\w.\-]+)(/[\w.\-]+)*$`),
1524		vcs:        "git",
1525		repo:       "https://{root}",
1526		check:      noVCSSuffix,
1527	},
1528
1529	// Bitbucket
1530	{
1531		pathPrefix: "bitbucket.org",
1532		regexp:     lazyregexp.New(`^(?P<root>bitbucket\.org/(?P<bitname>[\w.\-]+/[\w.\-]+))(/[\w.\-]+)*$`),
1533		vcs:        "git",
1534		repo:       "https://{root}",
1535		check:      noVCSSuffix,
1536	},
1537
1538	// IBM DevOps Services (JazzHub)
1539	{
1540		pathPrefix: "hub.jazz.net/git",
1541		regexp:     lazyregexp.New(`^(?P<root>hub\.jazz\.net/git/[a-z0-9]+/[\w.\-]+)(/[\w.\-]+)*$`),
1542		vcs:        "git",
1543		repo:       "https://{root}",
1544		check:      noVCSSuffix,
1545	},
1546
1547	// Git at Apache
1548	{
1549		pathPrefix: "git.apache.org",
1550		regexp:     lazyregexp.New(`^(?P<root>git\.apache\.org/[a-z0-9_.\-]+\.git)(/[\w.\-]+)*$`),
1551		vcs:        "git",
1552		repo:       "https://{root}",
1553	},
1554
1555	// Git at OpenStack
1556	{
1557		pathPrefix: "git.openstack.org",
1558		regexp:     lazyregexp.New(`^(?P<root>git\.openstack\.org/[\w.\-]+/[\w.\-]+)(\.git)?(/[\w.\-]+)*$`),
1559		vcs:        "git",
1560		repo:       "https://{root}",
1561	},
1562
1563	// chiselapp.com for fossil
1564	{
1565		pathPrefix: "chiselapp.com",
1566		regexp:     lazyregexp.New(`^(?P<root>chiselapp\.com/user/[A-Za-z0-9]+/repository/[\w.\-]+)$`),
1567		vcs:        "fossil",
1568		repo:       "https://{root}",
1569	},
1570
1571	// General syntax for any server.
1572	// Must be last.
1573	{
1574		regexp:         lazyregexp.New(`(?P<root>(?P<repo>([a-z0-9.\-]+\.)+[a-z0-9.\-]+(:[0-9]+)?(/~?[\w.\-]+)+?)\.(?P<vcs>bzr|fossil|git|hg|svn))(/~?[\w.\-]+)*$`),
1575		schemelessRepo: true,
1576	},
1577}
1578
1579// vcsPathsAfterDynamic gives additional vcsPaths entries
1580// to try after the dynamic HTML check.
1581// This gives those sites a chance to introduce <meta> tags
1582// as part of a graceful transition away from the hard-coded logic.
1583var vcsPathsAfterDynamic = []*vcsPath{
1584	// Launchpad. See golang.org/issue/11436.
1585	{
1586		pathPrefix: "launchpad.net",
1587		regexp:     lazyregexp.New(`^(?P<root>launchpad\.net/((?P<project>[\w.\-]+)(?P<series>/[\w.\-]+)?|~[\w.\-]+/(\+junk|[\w.\-]+)/[\w.\-]+))(/[\w.\-]+)*$`),
1588		vcs:        "bzr",
1589		repo:       "https://{root}",
1590		check:      launchpadVCS,
1591	},
1592}
1593
1594// noVCSSuffix checks that the repository name does not
1595// end in .foo for any version control system foo.
1596// The usual culprit is ".git".
1597func noVCSSuffix(match map[string]string) error {
1598	repo := match["repo"]
1599	for _, vcs := range vcsList {
1600		if strings.HasSuffix(repo, "."+vcs.Cmd) {
1601			return fmt.Errorf("invalid version control suffix in %s path", match["prefix"])
1602		}
1603	}
1604	return nil
1605}
1606
1607// launchpadVCS solves the ambiguity for "lp.net/project/foo". In this case,
1608// "foo" could be a series name registered in Launchpad with its own branch,
1609// and it could also be the name of a directory within the main project
1610// branch one level up.
1611func launchpadVCS(match map[string]string) error {
1612	if match["project"] == "" || match["series"] == "" {
1613		return nil
1614	}
1615	url := &urlpkg.URL{
1616		Scheme: "https",
1617		Host:   "code.launchpad.net",
1618		Path:   expand(match, "/{project}{series}/.bzr/branch-format"),
1619	}
1620	_, err := web.GetBytes(url)
1621	if err != nil {
1622		match["root"] = expand(match, "launchpad.net/{project}")
1623		match["repo"] = expand(match, "https://{root}")
1624	}
1625	return nil
1626}
1627
1628// importError is a copy of load.importError, made to avoid a dependency cycle
1629// on cmd/go/internal/load. It just needs to satisfy load.ImportPathError.
1630type importError struct {
1631	importPath string
1632	err        error
1633}
1634
1635func importErrorf(path, format string, args ...any) error {
1636	err := &importError{importPath: path, err: fmt.Errorf(format, args...)}
1637	if errStr := err.Error(); !strings.Contains(errStr, path) {
1638		panic(fmt.Sprintf("path %q not in error %q", path, errStr))
1639	}
1640	return err
1641}
1642
1643func (e *importError) Error() string {
1644	return e.err.Error()
1645}
1646
1647func (e *importError) Unwrap() error {
1648	// Don't return e.err directly, since we're only wrapping an error if %w
1649	// was passed to ImportErrorf.
1650	return errors.Unwrap(e.err)
1651}
1652
1653func (e *importError) ImportPath() string {
1654	return e.importPath
1655}
1656