Merge branch 'tb/filter-alternate-refs'
[git] / connect.c
1 #include "git-compat-util.h"
2 #include "cache.h"
3 #include "config.h"
4 #include "pkt-line.h"
5 #include "quote.h"
6 #include "refs.h"
7 #include "run-command.h"
8 #include "remote.h"
9 #include "connect.h"
10 #include "url.h"
11 #include "string-list.h"
12 #include "sha1-array.h"
13 #include "transport.h"
14 #include "strbuf.h"
15 #include "version.h"
16 #include "protocol.h"
17 #include "alias.h"
18
19 static char *server_capabilities_v1;
20 static struct argv_array server_capabilities_v2 = ARGV_ARRAY_INIT;
21 static const char *parse_feature_value(const char *, const char *, int *);
22
23 static int check_ref(const char *name, unsigned int flags)
24 {
25         if (!flags)
26                 return 1;
27
28         if (!skip_prefix(name, "refs/", &name))
29                 return 0;
30
31         /* REF_NORMAL means that we don't want the magic fake tag refs */
32         if ((flags & REF_NORMAL) && check_refname_format(name, 0))
33                 return 0;
34
35         /* REF_HEADS means that we want regular branch heads */
36         if ((flags & REF_HEADS) && starts_with(name, "heads/"))
37                 return 1;
38
39         /* REF_TAGS means that we want tags */
40         if ((flags & REF_TAGS) && starts_with(name, "tags/"))
41                 return 1;
42
43         /* All type bits clear means that we are ok with anything */
44         return !(flags & ~REF_NORMAL);
45 }
46
47 int check_ref_type(const struct ref *ref, int flags)
48 {
49         return check_ref(ref->name, flags);
50 }
51
52 static NORETURN void die_initial_contact(int unexpected)
53 {
54         /*
55          * A hang-up after seeing some response from the other end
56          * means that it is unexpected, as we know the other end is
57          * willing to talk to us.  A hang-up before seeing any
58          * response does not necessarily mean an ACL problem, though.
59          */
60         if (unexpected)
61                 die(_("the remote end hung up upon initial contact"));
62         else
63                 die(_("Could not read from remote repository.\n\n"
64                       "Please make sure you have the correct access rights\n"
65                       "and the repository exists."));
66 }
67
68 /* Checks if the server supports the capability 'c' */
69 int server_supports_v2(const char *c, int die_on_error)
70 {
71         int i;
72
73         for (i = 0; i < server_capabilities_v2.argc; i++) {
74                 const char *out;
75                 if (skip_prefix(server_capabilities_v2.argv[i], c, &out) &&
76                     (!*out || *out == '='))
77                         return 1;
78         }
79
80         if (die_on_error)
81                 die(_("server doesn't support '%s'"), c);
82
83         return 0;
84 }
85
86 int server_supports_feature(const char *c, const char *feature,
87                             int die_on_error)
88 {
89         int i;
90
91         for (i = 0; i < server_capabilities_v2.argc; i++) {
92                 const char *out;
93                 if (skip_prefix(server_capabilities_v2.argv[i], c, &out) &&
94                     (!*out || *(out++) == '=')) {
95                         if (parse_feature_request(out, feature))
96                                 return 1;
97                         else
98                                 break;
99                 }
100         }
101
102         if (die_on_error)
103                 die(_("server doesn't support feature '%s'"), feature);
104
105         return 0;
106 }
107
108 static void process_capabilities_v2(struct packet_reader *reader)
109 {
110         while (packet_reader_read(reader) == PACKET_READ_NORMAL)
111                 argv_array_push(&server_capabilities_v2, reader->line);
112
113         if (reader->status != PACKET_READ_FLUSH)
114                 die(_("expected flush after capabilities"));
115 }
116
117 enum protocol_version discover_version(struct packet_reader *reader)
118 {
119         enum protocol_version version = protocol_unknown_version;
120
121         /*
122          * Peek the first line of the server's response to
123          * determine the protocol version the server is speaking.
124          */
125         switch (packet_reader_peek(reader)) {
126         case PACKET_READ_EOF:
127                 die_initial_contact(0);
128         case PACKET_READ_FLUSH:
129         case PACKET_READ_DELIM:
130                 version = protocol_v0;
131                 break;
132         case PACKET_READ_NORMAL:
133                 version = determine_protocol_version_client(reader->line);
134                 break;
135         }
136
137         switch (version) {
138         case protocol_v2:
139                 process_capabilities_v2(reader);
140                 break;
141         case protocol_v1:
142                 /* Read the peeked version line */
143                 packet_reader_read(reader);
144                 break;
145         case protocol_v0:
146                 break;
147         case protocol_unknown_version:
148                 BUG("unknown protocol version");
149         }
150
151         return version;
152 }
153
154 static void parse_one_symref_info(struct string_list *symref, const char *val, int len)
155 {
156         char *sym, *target;
157         struct string_list_item *item;
158
159         if (!len)
160                 return; /* just "symref" */
161         /* e.g. "symref=HEAD:refs/heads/master" */
162         sym = xmemdupz(val, len);
163         target = strchr(sym, ':');
164         if (!target)
165                 /* just "symref=something" */
166                 goto reject;
167         *(target++) = '\0';
168         if (check_refname_format(sym, REFNAME_ALLOW_ONELEVEL) ||
169             check_refname_format(target, REFNAME_ALLOW_ONELEVEL))
170                 /* "symref=bogus:pair */
171                 goto reject;
172         item = string_list_append_nodup(symref, sym);
173         item->util = target;
174         return;
175 reject:
176         free(sym);
177         return;
178 }
179
180 static void annotate_refs_with_symref_info(struct ref *ref)
181 {
182         struct string_list symref = STRING_LIST_INIT_DUP;
183         const char *feature_list = server_capabilities_v1;
184
185         while (feature_list) {
186                 int len;
187                 const char *val;
188
189                 val = parse_feature_value(feature_list, "symref", &len);
190                 if (!val)
191                         break;
192                 parse_one_symref_info(&symref, val, len);
193                 feature_list = val + 1;
194         }
195         string_list_sort(&symref);
196
197         for (; ref; ref = ref->next) {
198                 struct string_list_item *item;
199                 item = string_list_lookup(&symref, ref->name);
200                 if (!item)
201                         continue;
202                 ref->symref = xstrdup((char *)item->util);
203         }
204         string_list_clear(&symref, 0);
205 }
206
207 static void process_capabilities(const char *line, int *len)
208 {
209         int nul_location = strlen(line);
210         if (nul_location == *len)
211                 return;
212         server_capabilities_v1 = xstrdup(line + nul_location + 1);
213         *len = nul_location;
214 }
215
216 static int process_dummy_ref(const char *line)
217 {
218         struct object_id oid;
219         const char *name;
220
221         if (parse_oid_hex(line, &oid, &name))
222                 return 0;
223         if (*name != ' ')
224                 return 0;
225         name++;
226
227         return oideq(&null_oid, &oid) && !strcmp(name, "capabilities^{}");
228 }
229
230 static void check_no_capabilities(const char *line, int len)
231 {
232         if (strlen(line) != len)
233                 warning(_("ignoring capabilities after first line '%s'"),
234                         line + strlen(line));
235 }
236
237 static int process_ref(const char *line, int len, struct ref ***list,
238                        unsigned int flags, struct oid_array *extra_have)
239 {
240         struct object_id old_oid;
241         const char *name;
242
243         if (parse_oid_hex(line, &old_oid, &name))
244                 return 0;
245         if (*name != ' ')
246                 return 0;
247         name++;
248
249         if (extra_have && !strcmp(name, ".have")) {
250                 oid_array_append(extra_have, &old_oid);
251         } else if (!strcmp(name, "capabilities^{}")) {
252                 die(_("protocol error: unexpected capabilities^{}"));
253         } else if (check_ref(name, flags)) {
254                 struct ref *ref = alloc_ref(name);
255                 oidcpy(&ref->old_oid, &old_oid);
256                 **list = ref;
257                 *list = &ref->next;
258         }
259         check_no_capabilities(line, len);
260         return 1;
261 }
262
263 static int process_shallow(const char *line, int len,
264                            struct oid_array *shallow_points)
265 {
266         const char *arg;
267         struct object_id old_oid;
268
269         if (!skip_prefix(line, "shallow ", &arg))
270                 return 0;
271
272         if (get_oid_hex(arg, &old_oid))
273                 die(_("protocol error: expected shallow sha-1, got '%s'"), arg);
274         if (!shallow_points)
275                 die(_("repository on the other end cannot be shallow"));
276         oid_array_append(shallow_points, &old_oid);
277         check_no_capabilities(line, len);
278         return 1;
279 }
280
281 enum get_remote_heads_state {
282         EXPECTING_FIRST_REF = 0,
283         EXPECTING_REF,
284         EXPECTING_SHALLOW,
285         EXPECTING_DONE,
286 };
287
288 /*
289  * Read all the refs from the other end
290  */
291 struct ref **get_remote_heads(struct packet_reader *reader,
292                               struct ref **list, unsigned int flags,
293                               struct oid_array *extra_have,
294                               struct oid_array *shallow_points)
295 {
296         struct ref **orig_list = list;
297         int len = 0;
298         enum get_remote_heads_state state = EXPECTING_FIRST_REF;
299         const char *arg;
300
301         *list = NULL;
302
303         while (state != EXPECTING_DONE) {
304                 switch (packet_reader_read(reader)) {
305                 case PACKET_READ_EOF:
306                         die_initial_contact(1);
307                 case PACKET_READ_NORMAL:
308                         len = reader->pktlen;
309                         if (len > 4 && skip_prefix(reader->line, "ERR ", &arg))
310                                 die(_("remote error: %s"), arg);
311                         break;
312                 case PACKET_READ_FLUSH:
313                         state = EXPECTING_DONE;
314                         break;
315                 case PACKET_READ_DELIM:
316                         die(_("invalid packet"));
317                 }
318
319                 switch (state) {
320                 case EXPECTING_FIRST_REF:
321                         process_capabilities(reader->line, &len);
322                         if (process_dummy_ref(reader->line)) {
323                                 state = EXPECTING_SHALLOW;
324                                 break;
325                         }
326                         state = EXPECTING_REF;
327                         /* fallthrough */
328                 case EXPECTING_REF:
329                         if (process_ref(reader->line, len, &list, flags, extra_have))
330                                 break;
331                         state = EXPECTING_SHALLOW;
332                         /* fallthrough */
333                 case EXPECTING_SHALLOW:
334                         if (process_shallow(reader->line, len, shallow_points))
335                                 break;
336                         die(_("protocol error: unexpected '%s'"), reader->line);
337                 case EXPECTING_DONE:
338                         break;
339                 }
340         }
341
342         annotate_refs_with_symref_info(*orig_list);
343
344         return list;
345 }
346
347 /* Returns 1 when a valid ref has been added to `list`, 0 otherwise */
348 static int process_ref_v2(const char *line, struct ref ***list)
349 {
350         int ret = 1;
351         int i = 0;
352         struct object_id old_oid;
353         struct ref *ref;
354         struct string_list line_sections = STRING_LIST_INIT_DUP;
355         const char *end;
356
357         /*
358          * Ref lines have a number of fields which are space deliminated.  The
359          * first field is the OID of the ref.  The second field is the ref
360          * name.  Subsequent fields (symref-target and peeled) are optional and
361          * don't have a particular order.
362          */
363         if (string_list_split(&line_sections, line, ' ', -1) < 2) {
364                 ret = 0;
365                 goto out;
366         }
367
368         if (parse_oid_hex(line_sections.items[i++].string, &old_oid, &end) ||
369             *end) {
370                 ret = 0;
371                 goto out;
372         }
373
374         ref = alloc_ref(line_sections.items[i++].string);
375
376         oidcpy(&ref->old_oid, &old_oid);
377         **list = ref;
378         *list = &ref->next;
379
380         for (; i < line_sections.nr; i++) {
381                 const char *arg = line_sections.items[i].string;
382                 if (skip_prefix(arg, "symref-target:", &arg))
383                         ref->symref = xstrdup(arg);
384
385                 if (skip_prefix(arg, "peeled:", &arg)) {
386                         struct object_id peeled_oid;
387                         char *peeled_name;
388                         struct ref *peeled;
389                         if (parse_oid_hex(arg, &peeled_oid, &end) || *end) {
390                                 ret = 0;
391                                 goto out;
392                         }
393
394                         peeled_name = xstrfmt("%s^{}", ref->name);
395                         peeled = alloc_ref(peeled_name);
396
397                         oidcpy(&peeled->old_oid, &peeled_oid);
398                         **list = peeled;
399                         *list = &peeled->next;
400
401                         free(peeled_name);
402                 }
403         }
404
405 out:
406         string_list_clear(&line_sections, 0);
407         return ret;
408 }
409
410 struct ref **get_remote_refs(int fd_out, struct packet_reader *reader,
411                              struct ref **list, int for_push,
412                              const struct argv_array *ref_prefixes,
413                              const struct string_list *server_options)
414 {
415         int i;
416         *list = NULL;
417
418         if (server_supports_v2("ls-refs", 1))
419                 packet_write_fmt(fd_out, "command=ls-refs\n");
420
421         if (server_supports_v2("agent", 0))
422                 packet_write_fmt(fd_out, "agent=%s", git_user_agent_sanitized());
423
424         if (server_options && server_options->nr &&
425             server_supports_v2("server-option", 1))
426                 for (i = 0; i < server_options->nr; i++)
427                         packet_write_fmt(fd_out, "server-option=%s",
428                                          server_options->items[i].string);
429
430         packet_delim(fd_out);
431         /* When pushing we don't want to request the peeled tags */
432         if (!for_push)
433                 packet_write_fmt(fd_out, "peel\n");
434         packet_write_fmt(fd_out, "symrefs\n");
435         for (i = 0; ref_prefixes && i < ref_prefixes->argc; i++) {
436                 packet_write_fmt(fd_out, "ref-prefix %s\n",
437                                  ref_prefixes->argv[i]);
438         }
439         packet_flush(fd_out);
440
441         /* Process response from server */
442         while (packet_reader_read(reader) == PACKET_READ_NORMAL) {
443                 if (!process_ref_v2(reader->line, &list))
444                         die(_("invalid ls-refs response: %s"), reader->line);
445         }
446
447         if (reader->status != PACKET_READ_FLUSH)
448                 die(_("expected flush after ref listing"));
449
450         return list;
451 }
452
453 static const char *parse_feature_value(const char *feature_list, const char *feature, int *lenp)
454 {
455         int len;
456
457         if (!feature_list)
458                 return NULL;
459
460         len = strlen(feature);
461         while (*feature_list) {
462                 const char *found = strstr(feature_list, feature);
463                 if (!found)
464                         return NULL;
465                 if (feature_list == found || isspace(found[-1])) {
466                         const char *value = found + len;
467                         /* feature with no value (e.g., "thin-pack") */
468                         if (!*value || isspace(*value)) {
469                                 if (lenp)
470                                         *lenp = 0;
471                                 return value;
472                         }
473                         /* feature with a value (e.g., "agent=git/1.2.3") */
474                         else if (*value == '=') {
475                                 value++;
476                                 if (lenp)
477                                         *lenp = strcspn(value, " \t\n");
478                                 return value;
479                         }
480                         /*
481                          * otherwise we matched a substring of another feature;
482                          * keep looking
483                          */
484                 }
485                 feature_list = found + 1;
486         }
487         return NULL;
488 }
489
490 int parse_feature_request(const char *feature_list, const char *feature)
491 {
492         return !!parse_feature_value(feature_list, feature, NULL);
493 }
494
495 const char *server_feature_value(const char *feature, int *len)
496 {
497         return parse_feature_value(server_capabilities_v1, feature, len);
498 }
499
500 int server_supports(const char *feature)
501 {
502         return !!server_feature_value(feature, NULL);
503 }
504
505 enum protocol {
506         PROTO_LOCAL = 1,
507         PROTO_FILE,
508         PROTO_SSH,
509         PROTO_GIT
510 };
511
512 int url_is_local_not_ssh(const char *url)
513 {
514         const char *colon = strchr(url, ':');
515         const char *slash = strchr(url, '/');
516         return !colon || (slash && slash < colon) ||
517                 has_dos_drive_prefix(url);
518 }
519
520 static const char *prot_name(enum protocol protocol)
521 {
522         switch (protocol) {
523                 case PROTO_LOCAL:
524                 case PROTO_FILE:
525                         return "file";
526                 case PROTO_SSH:
527                         return "ssh";
528                 case PROTO_GIT:
529                         return "git";
530                 default:
531                         return "unknown protocol";
532         }
533 }
534
535 static enum protocol get_protocol(const char *name)
536 {
537         if (!strcmp(name, "ssh"))
538                 return PROTO_SSH;
539         if (!strcmp(name, "git"))
540                 return PROTO_GIT;
541         if (!strcmp(name, "git+ssh")) /* deprecated - do not use */
542                 return PROTO_SSH;
543         if (!strcmp(name, "ssh+git")) /* deprecated - do not use */
544                 return PROTO_SSH;
545         if (!strcmp(name, "file"))
546                 return PROTO_FILE;
547         die(_("protocol '%s' is not supported"), name);
548 }
549
550 static char *host_end(char **hoststart, int removebrackets)
551 {
552         char *host = *hoststart;
553         char *end;
554         char *start = strstr(host, "@[");
555         if (start)
556                 start++; /* Jump over '@' */
557         else
558                 start = host;
559         if (start[0] == '[') {
560                 end = strchr(start + 1, ']');
561                 if (end) {
562                         if (removebrackets) {
563                                 *end = 0;
564                                 memmove(start, start + 1, end - start);
565                                 end++;
566                         }
567                 } else
568                         end = host;
569         } else
570                 end = host;
571         return end;
572 }
573
574 #define STR_(s) # s
575 #define STR(s)  STR_(s)
576
577 static void get_host_and_port(char **host, const char **port)
578 {
579         char *colon, *end;
580         end = host_end(host, 1);
581         colon = strchr(end, ':');
582         if (colon) {
583                 long portnr = strtol(colon + 1, &end, 10);
584                 if (end != colon + 1 && *end == '\0' && 0 <= portnr && portnr < 65536) {
585                         *colon = 0;
586                         *port = colon + 1;
587                 } else if (!colon[1]) {
588                         *colon = 0;
589                 }
590         }
591 }
592
593 static void enable_keepalive(int sockfd)
594 {
595         int ka = 1;
596
597         if (setsockopt(sockfd, SOL_SOCKET, SO_KEEPALIVE, &ka, sizeof(ka)) < 0)
598                 error_errno(_("unable to set SO_KEEPALIVE on socket"));
599 }
600
601 #ifndef NO_IPV6
602
603 static const char *ai_name(const struct addrinfo *ai)
604 {
605         static char addr[NI_MAXHOST];
606         if (getnameinfo(ai->ai_addr, ai->ai_addrlen, addr, sizeof(addr), NULL, 0,
607                         NI_NUMERICHOST) != 0)
608                 xsnprintf(addr, sizeof(addr), "(unknown)");
609
610         return addr;
611 }
612
613 /*
614  * Returns a connected socket() fd, or else die()s.
615  */
616 static int git_tcp_connect_sock(char *host, int flags)
617 {
618         struct strbuf error_message = STRBUF_INIT;
619         int sockfd = -1;
620         const char *port = STR(DEFAULT_GIT_PORT);
621         struct addrinfo hints, *ai0, *ai;
622         int gai;
623         int cnt = 0;
624
625         get_host_and_port(&host, &port);
626         if (!*port)
627                 port = "<none>";
628
629         memset(&hints, 0, sizeof(hints));
630         if (flags & CONNECT_IPV4)
631                 hints.ai_family = AF_INET;
632         else if (flags & CONNECT_IPV6)
633                 hints.ai_family = AF_INET6;
634         hints.ai_socktype = SOCK_STREAM;
635         hints.ai_protocol = IPPROTO_TCP;
636
637         if (flags & CONNECT_VERBOSE)
638                 fprintf(stderr, _("Looking up %s ... "), host);
639
640         gai = getaddrinfo(host, port, &hints, &ai);
641         if (gai)
642                 die(_("unable to look up %s (port %s) (%s)"), host, port, gai_strerror(gai));
643
644         if (flags & CONNECT_VERBOSE)
645                 /* TRANSLATORS: this is the end of "Looking up %s ... " */
646                 fprintf(stderr, _("done.\nConnecting to %s (port %s) ... "), host, port);
647
648         for (ai0 = ai; ai; ai = ai->ai_next, cnt++) {
649                 sockfd = socket(ai->ai_family,
650                                 ai->ai_socktype, ai->ai_protocol);
651                 if ((sockfd < 0) ||
652                     (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0)) {
653                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
654                                     host, cnt, ai_name(ai), strerror(errno));
655                         if (0 <= sockfd)
656                                 close(sockfd);
657                         sockfd = -1;
658                         continue;
659                 }
660                 if (flags & CONNECT_VERBOSE)
661                         fprintf(stderr, "%s ", ai_name(ai));
662                 break;
663         }
664
665         freeaddrinfo(ai0);
666
667         if (sockfd < 0)
668                 die(_("unable to connect to %s:\n%s"), host, error_message.buf);
669
670         enable_keepalive(sockfd);
671
672         if (flags & CONNECT_VERBOSE)
673                 /* TRANSLATORS: this is the end of "Connecting to %s (port %s) ... " */
674                 fprintf_ln(stderr, _("done."));
675
676         strbuf_release(&error_message);
677
678         return sockfd;
679 }
680
681 #else /* NO_IPV6 */
682
683 /*
684  * Returns a connected socket() fd, or else die()s.
685  */
686 static int git_tcp_connect_sock(char *host, int flags)
687 {
688         struct strbuf error_message = STRBUF_INIT;
689         int sockfd = -1;
690         const char *port = STR(DEFAULT_GIT_PORT);
691         char *ep;
692         struct hostent *he;
693         struct sockaddr_in sa;
694         char **ap;
695         unsigned int nport;
696         int cnt;
697
698         get_host_and_port(&host, &port);
699
700         if (flags & CONNECT_VERBOSE)
701                 fprintf(stderr, _("Looking up %s ... "), host);
702
703         he = gethostbyname(host);
704         if (!he)
705                 die(_("unable to look up %s (%s)"), host, hstrerror(h_errno));
706         nport = strtoul(port, &ep, 10);
707         if ( ep == port || *ep ) {
708                 /* Not numeric */
709                 struct servent *se = getservbyname(port,"tcp");
710                 if ( !se )
711                         die(_("unknown port %s"), port);
712                 nport = se->s_port;
713         }
714
715         if (flags & CONNECT_VERBOSE)
716                 /* TRANSLATORS: this is the end of "Looking up %s ... " */
717                 fprintf(stderr, _("done.\nConnecting to %s (port %s) ... "), host, port);
718
719         for (cnt = 0, ap = he->h_addr_list; *ap; ap++, cnt++) {
720                 memset(&sa, 0, sizeof sa);
721                 sa.sin_family = he->h_addrtype;
722                 sa.sin_port = htons(nport);
723                 memcpy(&sa.sin_addr, *ap, he->h_length);
724
725                 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
726                 if ((sockfd < 0) ||
727                     connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
728                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
729                                 host,
730                                 cnt,
731                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr),
732                                 strerror(errno));
733                         if (0 <= sockfd)
734                                 close(sockfd);
735                         sockfd = -1;
736                         continue;
737                 }
738                 if (flags & CONNECT_VERBOSE)
739                         fprintf(stderr, "%s ",
740                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr));
741                 break;
742         }
743
744         if (sockfd < 0)
745                 die(_("unable to connect to %s:\n%s"), host, error_message.buf);
746
747         enable_keepalive(sockfd);
748
749         if (flags & CONNECT_VERBOSE)
750                 /* TRANSLATORS: this is the end of "Connecting to %s (port %s) ... " */
751                 fprintf_ln(stderr, _("done."));
752
753         return sockfd;
754 }
755
756 #endif /* NO_IPV6 */
757
758
759 /*
760  * Dummy child_process returned by git_connect() if the transport protocol
761  * does not need fork(2).
762  */
763 static struct child_process no_fork = CHILD_PROCESS_INIT;
764
765 int git_connection_is_socket(struct child_process *conn)
766 {
767         return conn == &no_fork;
768 }
769
770 static struct child_process *git_tcp_connect(int fd[2], char *host, int flags)
771 {
772         int sockfd = git_tcp_connect_sock(host, flags);
773
774         fd[0] = sockfd;
775         fd[1] = dup(sockfd);
776
777         return &no_fork;
778 }
779
780
781 static char *git_proxy_command;
782
783 static int git_proxy_command_options(const char *var, const char *value,
784                 void *cb)
785 {
786         if (!strcmp(var, "core.gitproxy")) {
787                 const char *for_pos;
788                 int matchlen = -1;
789                 int hostlen;
790                 const char *rhost_name = cb;
791                 int rhost_len = strlen(rhost_name);
792
793                 if (git_proxy_command)
794                         return 0;
795                 if (!value)
796                         return config_error_nonbool(var);
797                 /* [core]
798                  * ;# matches www.kernel.org as well
799                  * gitproxy = netcatter-1 for kernel.org
800                  * gitproxy = netcatter-2 for sample.xz
801                  * gitproxy = netcatter-default
802                  */
803                 for_pos = strstr(value, " for ");
804                 if (!for_pos)
805                         /* matches everybody */
806                         matchlen = strlen(value);
807                 else {
808                         hostlen = strlen(for_pos + 5);
809                         if (rhost_len < hostlen)
810                                 matchlen = -1;
811                         else if (!strncmp(for_pos + 5,
812                                           rhost_name + rhost_len - hostlen,
813                                           hostlen) &&
814                                  ((rhost_len == hostlen) ||
815                                   rhost_name[rhost_len - hostlen -1] == '.'))
816                                 matchlen = for_pos - value;
817                         else
818                                 matchlen = -1;
819                 }
820                 if (0 <= matchlen) {
821                         /* core.gitproxy = none for kernel.org */
822                         if (matchlen == 4 &&
823                             !memcmp(value, "none", 4))
824                                 matchlen = 0;
825                         git_proxy_command = xmemdupz(value, matchlen);
826                 }
827                 return 0;
828         }
829
830         return git_default_config(var, value, cb);
831 }
832
833 static int git_use_proxy(const char *host)
834 {
835         git_proxy_command = getenv("GIT_PROXY_COMMAND");
836         git_config(git_proxy_command_options, (void*)host);
837         return (git_proxy_command && *git_proxy_command);
838 }
839
840 static struct child_process *git_proxy_connect(int fd[2], char *host)
841 {
842         const char *port = STR(DEFAULT_GIT_PORT);
843         struct child_process *proxy;
844
845         get_host_and_port(&host, &port);
846
847         if (looks_like_command_line_option(host))
848                 die(_("strange hostname '%s' blocked"), host);
849         if (looks_like_command_line_option(port))
850                 die(_("strange port '%s' blocked"), port);
851
852         proxy = xmalloc(sizeof(*proxy));
853         child_process_init(proxy);
854         argv_array_push(&proxy->args, git_proxy_command);
855         argv_array_push(&proxy->args, host);
856         argv_array_push(&proxy->args, port);
857         proxy->in = -1;
858         proxy->out = -1;
859         if (start_command(proxy))
860                 die(_("cannot start proxy %s"), git_proxy_command);
861         fd[0] = proxy->out; /* read from proxy stdout */
862         fd[1] = proxy->in;  /* write to proxy stdin */
863         return proxy;
864 }
865
866 static char *get_port(char *host)
867 {
868         char *end;
869         char *p = strchr(host, ':');
870
871         if (p) {
872                 long port = strtol(p + 1, &end, 10);
873                 if (end != p + 1 && *end == '\0' && 0 <= port && port < 65536) {
874                         *p = '\0';
875                         return p+1;
876                 }
877         }
878
879         return NULL;
880 }
881
882 /*
883  * Extract protocol and relevant parts from the specified connection URL.
884  * The caller must free() the returned strings.
885  */
886 static enum protocol parse_connect_url(const char *url_orig, char **ret_host,
887                                        char **ret_path)
888 {
889         char *url;
890         char *host, *path;
891         char *end;
892         int separator = '/';
893         enum protocol protocol = PROTO_LOCAL;
894
895         if (is_url(url_orig))
896                 url = url_decode(url_orig);
897         else
898                 url = xstrdup(url_orig);
899
900         host = strstr(url, "://");
901         if (host) {
902                 *host = '\0';
903                 protocol = get_protocol(url);
904                 host += 3;
905         } else {
906                 host = url;
907                 if (!url_is_local_not_ssh(url)) {
908                         protocol = PROTO_SSH;
909                         separator = ':';
910                 }
911         }
912
913         /*
914          * Don't do destructive transforms as protocol code does
915          * '[]' unwrapping in get_host_and_port()
916          */
917         end = host_end(&host, 0);
918
919         if (protocol == PROTO_LOCAL)
920                 path = end;
921         else if (protocol == PROTO_FILE && has_dos_drive_prefix(end))
922                 path = end; /* "file://$(pwd)" may be "file://C:/projects/repo" */
923         else
924                 path = strchr(end, separator);
925
926         if (!path || !*path)
927                 die(_("no path specified; see 'git help pull' for valid url syntax"));
928
929         /*
930          * null-terminate hostname and point path to ~ for URL's like this:
931          *    ssh://host.xz/~user/repo
932          */
933
934         end = path; /* Need to \0 terminate host here */
935         if (separator == ':')
936                 path++; /* path starts after ':' */
937         if (protocol == PROTO_GIT || protocol == PROTO_SSH) {
938                 if (path[1] == '~')
939                         path++;
940         }
941
942         path = xstrdup(path);
943         *end = '\0';
944
945         *ret_host = xstrdup(host);
946         *ret_path = path;
947         free(url);
948         return protocol;
949 }
950
951 static const char *get_ssh_command(void)
952 {
953         const char *ssh;
954
955         if ((ssh = getenv("GIT_SSH_COMMAND")))
956                 return ssh;
957
958         if (!git_config_get_string_const("core.sshcommand", &ssh))
959                 return ssh;
960
961         return NULL;
962 }
963
964 enum ssh_variant {
965         VARIANT_AUTO,
966         VARIANT_SIMPLE,
967         VARIANT_SSH,
968         VARIANT_PLINK,
969         VARIANT_PUTTY,
970         VARIANT_TORTOISEPLINK,
971 };
972
973 static void override_ssh_variant(enum ssh_variant *ssh_variant)
974 {
975         const char *variant = getenv("GIT_SSH_VARIANT");
976
977         if (!variant && git_config_get_string_const("ssh.variant", &variant))
978                 return;
979
980         if (!strcmp(variant, "auto"))
981                 *ssh_variant = VARIANT_AUTO;
982         else if (!strcmp(variant, "plink"))
983                 *ssh_variant = VARIANT_PLINK;
984         else if (!strcmp(variant, "putty"))
985                 *ssh_variant = VARIANT_PUTTY;
986         else if (!strcmp(variant, "tortoiseplink"))
987                 *ssh_variant = VARIANT_TORTOISEPLINK;
988         else if (!strcmp(variant, "simple"))
989                 *ssh_variant = VARIANT_SIMPLE;
990         else
991                 *ssh_variant = VARIANT_SSH;
992 }
993
994 static enum ssh_variant determine_ssh_variant(const char *ssh_command,
995                                               int is_cmdline)
996 {
997         enum ssh_variant ssh_variant = VARIANT_AUTO;
998         const char *variant;
999         char *p = NULL;
1000
1001         override_ssh_variant(&ssh_variant);
1002
1003         if (ssh_variant != VARIANT_AUTO)
1004                 return ssh_variant;
1005
1006         if (!is_cmdline) {
1007                 p = xstrdup(ssh_command);
1008                 variant = basename(p);
1009         } else {
1010                 const char **ssh_argv;
1011
1012                 p = xstrdup(ssh_command);
1013                 if (split_cmdline(p, &ssh_argv) > 0) {
1014                         variant = basename((char *)ssh_argv[0]);
1015                         /*
1016                          * At this point, variant points into the buffer
1017                          * referenced by p, hence we do not need ssh_argv
1018                          * any longer.
1019                          */
1020                         free(ssh_argv);
1021                 } else {
1022                         free(p);
1023                         return ssh_variant;
1024                 }
1025         }
1026
1027         if (!strcasecmp(variant, "ssh") ||
1028             !strcasecmp(variant, "ssh.exe"))
1029                 ssh_variant = VARIANT_SSH;
1030         else if (!strcasecmp(variant, "plink") ||
1031                  !strcasecmp(variant, "plink.exe"))
1032                 ssh_variant = VARIANT_PLINK;
1033         else if (!strcasecmp(variant, "tortoiseplink") ||
1034                  !strcasecmp(variant, "tortoiseplink.exe"))
1035                 ssh_variant = VARIANT_TORTOISEPLINK;
1036
1037         free(p);
1038         return ssh_variant;
1039 }
1040
1041 /*
1042  * Open a connection using Git's native protocol.
1043  *
1044  * The caller is responsible for freeing hostandport, but this function may
1045  * modify it (for example, to truncate it to remove the port part).
1046  */
1047 static struct child_process *git_connect_git(int fd[2], char *hostandport,
1048                                              const char *path, const char *prog,
1049                                              enum protocol_version version,
1050                                              int flags)
1051 {
1052         struct child_process *conn;
1053         struct strbuf request = STRBUF_INIT;
1054         /*
1055          * Set up virtual host information based on where we will
1056          * connect, unless the user has overridden us in
1057          * the environment.
1058          */
1059         char *target_host = getenv("GIT_OVERRIDE_VIRTUAL_HOST");
1060         if (target_host)
1061                 target_host = xstrdup(target_host);
1062         else
1063                 target_host = xstrdup(hostandport);
1064
1065         transport_check_allowed("git");
1066
1067         /*
1068          * These underlying connection commands die() if they
1069          * cannot connect.
1070          */
1071         if (git_use_proxy(hostandport))
1072                 conn = git_proxy_connect(fd, hostandport);
1073         else
1074                 conn = git_tcp_connect(fd, hostandport, flags);
1075         /*
1076          * Separate original protocol components prog and path
1077          * from extended host header with a NUL byte.
1078          *
1079          * Note: Do not add any other headers here!  Doing so
1080          * will cause older git-daemon servers to crash.
1081          */
1082         strbuf_addf(&request,
1083                     "%s %s%chost=%s%c",
1084                     prog, path, 0,
1085                     target_host, 0);
1086
1087         /* If using a new version put that stuff here after a second null byte */
1088         if (version > 0) {
1089                 strbuf_addch(&request, '\0');
1090                 strbuf_addf(&request, "version=%d%c",
1091                             version, '\0');
1092         }
1093
1094         packet_write(fd[1], request.buf, request.len);
1095
1096         free(target_host);
1097         strbuf_release(&request);
1098         return conn;
1099 }
1100
1101 /*
1102  * Append the appropriate environment variables to `env` and options to
1103  * `args` for running ssh in Git's SSH-tunneled transport.
1104  */
1105 static void push_ssh_options(struct argv_array *args, struct argv_array *env,
1106                              enum ssh_variant variant, const char *port,
1107                              enum protocol_version version, int flags)
1108 {
1109         if (variant == VARIANT_SSH &&
1110             version > 0) {
1111                 argv_array_push(args, "-o");
1112                 argv_array_push(args, "SendEnv=" GIT_PROTOCOL_ENVIRONMENT);
1113                 argv_array_pushf(env, GIT_PROTOCOL_ENVIRONMENT "=version=%d",
1114                                  version);
1115         }
1116
1117         if (flags & CONNECT_IPV4) {
1118                 switch (variant) {
1119                 case VARIANT_AUTO:
1120                         BUG("VARIANT_AUTO passed to push_ssh_options");
1121                 case VARIANT_SIMPLE:
1122                         die(_("ssh variant 'simple' does not support -4"));
1123                 case VARIANT_SSH:
1124                 case VARIANT_PLINK:
1125                 case VARIANT_PUTTY:
1126                 case VARIANT_TORTOISEPLINK:
1127                         argv_array_push(args, "-4");
1128                 }
1129         } else if (flags & CONNECT_IPV6) {
1130                 switch (variant) {
1131                 case VARIANT_AUTO:
1132                         BUG("VARIANT_AUTO passed to push_ssh_options");
1133                 case VARIANT_SIMPLE:
1134                         die(_("ssh variant 'simple' does not support -6"));
1135                 case VARIANT_SSH:
1136                 case VARIANT_PLINK:
1137                 case VARIANT_PUTTY:
1138                 case VARIANT_TORTOISEPLINK:
1139                         argv_array_push(args, "-6");
1140                 }
1141         }
1142
1143         if (variant == VARIANT_TORTOISEPLINK)
1144                 argv_array_push(args, "-batch");
1145
1146         if (port) {
1147                 switch (variant) {
1148                 case VARIANT_AUTO:
1149                         BUG("VARIANT_AUTO passed to push_ssh_options");
1150                 case VARIANT_SIMPLE:
1151                         die(_("ssh variant 'simple' does not support setting port"));
1152                 case VARIANT_SSH:
1153                         argv_array_push(args, "-p");
1154                         break;
1155                 case VARIANT_PLINK:
1156                 case VARIANT_PUTTY:
1157                 case VARIANT_TORTOISEPLINK:
1158                         argv_array_push(args, "-P");
1159                 }
1160
1161                 argv_array_push(args, port);
1162         }
1163 }
1164
1165 /* Prepare a child_process for use by Git's SSH-tunneled transport. */
1166 static void fill_ssh_args(struct child_process *conn, const char *ssh_host,
1167                           const char *port, enum protocol_version version,
1168                           int flags)
1169 {
1170         const char *ssh;
1171         enum ssh_variant variant;
1172
1173         if (looks_like_command_line_option(ssh_host))
1174                 die(_("strange hostname '%s' blocked"), ssh_host);
1175
1176         ssh = get_ssh_command();
1177         if (ssh) {
1178                 variant = determine_ssh_variant(ssh, 1);
1179         } else {
1180                 /*
1181                  * GIT_SSH is the no-shell version of
1182                  * GIT_SSH_COMMAND (and must remain so for
1183                  * historical compatibility).
1184                  */
1185                 conn->use_shell = 0;
1186
1187                 ssh = getenv("GIT_SSH");
1188                 if (!ssh)
1189                         ssh = "ssh";
1190                 variant = determine_ssh_variant(ssh, 0);
1191         }
1192
1193         if (variant == VARIANT_AUTO) {
1194                 struct child_process detect = CHILD_PROCESS_INIT;
1195
1196                 detect.use_shell = conn->use_shell;
1197                 detect.no_stdin = detect.no_stdout = detect.no_stderr = 1;
1198
1199                 argv_array_push(&detect.args, ssh);
1200                 argv_array_push(&detect.args, "-G");
1201                 push_ssh_options(&detect.args, &detect.env_array,
1202                                  VARIANT_SSH, port, version, flags);
1203                 argv_array_push(&detect.args, ssh_host);
1204
1205                 variant = run_command(&detect) ? VARIANT_SIMPLE : VARIANT_SSH;
1206         }
1207
1208         argv_array_push(&conn->args, ssh);
1209         push_ssh_options(&conn->args, &conn->env_array, variant, port, version, flags);
1210         argv_array_push(&conn->args, ssh_host);
1211 }
1212
1213 /*
1214  * This returns the dummy child_process `no_fork` if the transport protocol
1215  * does not need fork(2), or a struct child_process object if it does.  Once
1216  * done, finish the connection with finish_connect() with the value returned
1217  * from this function (it is safe to call finish_connect() with NULL to
1218  * support the former case).
1219  *
1220  * If it returns, the connect is successful; it just dies on errors (this
1221  * will hopefully be changed in a libification effort, to return NULL when
1222  * the connection failed).
1223  */
1224 struct child_process *git_connect(int fd[2], const char *url,
1225                                   const char *prog, int flags)
1226 {
1227         char *hostandport, *path;
1228         struct child_process *conn;
1229         enum protocol protocol;
1230         enum protocol_version version = get_protocol_version_config();
1231
1232         /*
1233          * NEEDSWORK: If we are trying to use protocol v2 and we are planning
1234          * to perform a push, then fallback to v0 since the client doesn't know
1235          * how to push yet using v2.
1236          */
1237         if (version == protocol_v2 && !strcmp("git-receive-pack", prog))
1238                 version = protocol_v0;
1239
1240         /* Without this we cannot rely on waitpid() to tell
1241          * what happened to our children.
1242          */
1243         signal(SIGCHLD, SIG_DFL);
1244
1245         protocol = parse_connect_url(url, &hostandport, &path);
1246         if ((flags & CONNECT_DIAG_URL) && (protocol != PROTO_SSH)) {
1247                 printf("Diag: url=%s\n", url ? url : "NULL");
1248                 printf("Diag: protocol=%s\n", prot_name(protocol));
1249                 printf("Diag: hostandport=%s\n", hostandport ? hostandport : "NULL");
1250                 printf("Diag: path=%s\n", path ? path : "NULL");
1251                 conn = NULL;
1252         } else if (protocol == PROTO_GIT) {
1253                 conn = git_connect_git(fd, hostandport, path, prog, version, flags);
1254         } else {
1255                 struct strbuf cmd = STRBUF_INIT;
1256                 const char *const *var;
1257
1258                 conn = xmalloc(sizeof(*conn));
1259                 child_process_init(conn);
1260
1261                 if (looks_like_command_line_option(path))
1262                         die(_("strange pathname '%s' blocked"), path);
1263
1264                 strbuf_addstr(&cmd, prog);
1265                 strbuf_addch(&cmd, ' ');
1266                 sq_quote_buf(&cmd, path);
1267
1268                 /* remove repo-local variables from the environment */
1269                 for (var = local_repo_env; *var; var++)
1270                         argv_array_push(&conn->env_array, *var);
1271
1272                 conn->use_shell = 1;
1273                 conn->in = conn->out = -1;
1274                 if (protocol == PROTO_SSH) {
1275                         char *ssh_host = hostandport;
1276                         const char *port = NULL;
1277                         transport_check_allowed("ssh");
1278                         get_host_and_port(&ssh_host, &port);
1279
1280                         if (!port)
1281                                 port = get_port(ssh_host);
1282
1283                         if (flags & CONNECT_DIAG_URL) {
1284                                 printf("Diag: url=%s\n", url ? url : "NULL");
1285                                 printf("Diag: protocol=%s\n", prot_name(protocol));
1286                                 printf("Diag: userandhost=%s\n", ssh_host ? ssh_host : "NULL");
1287                                 printf("Diag: port=%s\n", port ? port : "NONE");
1288                                 printf("Diag: path=%s\n", path ? path : "NULL");
1289
1290                                 free(hostandport);
1291                                 free(path);
1292                                 free(conn);
1293                                 strbuf_release(&cmd);
1294                                 return NULL;
1295                         }
1296                         fill_ssh_args(conn, ssh_host, port, version, flags);
1297                 } else {
1298                         transport_check_allowed("file");
1299                         if (version > 0) {
1300                                 argv_array_pushf(&conn->env_array, GIT_PROTOCOL_ENVIRONMENT "=version=%d",
1301                                                  version);
1302                         }
1303                 }
1304                 argv_array_push(&conn->args, cmd.buf);
1305
1306                 if (start_command(conn))
1307                         die(_("unable to fork"));
1308
1309                 fd[0] = conn->out; /* read from child's stdout */
1310                 fd[1] = conn->in;  /* write to child's stdin */
1311                 strbuf_release(&cmd);
1312         }
1313         free(hostandport);
1314         free(path);
1315         return conn;
1316 }
1317
1318 int finish_connect(struct child_process *conn)
1319 {
1320         int code;
1321         if (!conn || git_connection_is_socket(conn))
1322                 return 0;
1323
1324         code = finish_command(conn);
1325         free(conn);
1326         return code;
1327 }