fetch: add --prefetch option
[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 "oid-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 strvec server_capabilities_v2 = STRVEC_INIT;
21 static const char *next_server_feature_value(const char *feature, int *len, int *offset);
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.nr; i++) {
74                 const char *out;
75                 if (skip_prefix(server_capabilities_v2.v[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_feature_v2(const char *c, const char **v)
87 {
88         int i;
89
90         for (i = 0; i < server_capabilities_v2.nr; i++) {
91                 const char *out;
92                 if (skip_prefix(server_capabilities_v2.v[i], c, &out) &&
93                     (*out == '=')) {
94                         *v = out + 1;
95                         return 1;
96                 }
97         }
98         return 0;
99 }
100
101 int server_supports_feature(const char *c, const char *feature,
102                             int die_on_error)
103 {
104         int i;
105
106         for (i = 0; i < server_capabilities_v2.nr; i++) {
107                 const char *out;
108                 if (skip_prefix(server_capabilities_v2.v[i], c, &out) &&
109                     (!*out || *(out++) == '=')) {
110                         if (parse_feature_request(out, feature))
111                                 return 1;
112                         else
113                                 break;
114                 }
115         }
116
117         if (die_on_error)
118                 die(_("server doesn't support feature '%s'"), feature);
119
120         return 0;
121 }
122
123 static void process_capabilities_v2(struct packet_reader *reader)
124 {
125         while (packet_reader_read(reader) == PACKET_READ_NORMAL)
126                 strvec_push(&server_capabilities_v2, reader->line);
127
128         if (reader->status != PACKET_READ_FLUSH)
129                 die(_("expected flush after capabilities"));
130 }
131
132 enum protocol_version discover_version(struct packet_reader *reader)
133 {
134         enum protocol_version version = protocol_unknown_version;
135
136         /*
137          * Peek the first line of the server's response to
138          * determine the protocol version the server is speaking.
139          */
140         switch (packet_reader_peek(reader)) {
141         case PACKET_READ_EOF:
142                 die_initial_contact(0);
143         case PACKET_READ_FLUSH:
144         case PACKET_READ_DELIM:
145         case PACKET_READ_RESPONSE_END:
146                 version = protocol_v0;
147                 break;
148         case PACKET_READ_NORMAL:
149                 version = determine_protocol_version_client(reader->line);
150                 break;
151         }
152
153         switch (version) {
154         case protocol_v2:
155                 process_capabilities_v2(reader);
156                 break;
157         case protocol_v1:
158                 /* Read the peeked version line */
159                 packet_reader_read(reader);
160                 break;
161         case protocol_v0:
162                 break;
163         case protocol_unknown_version:
164                 BUG("unknown protocol version");
165         }
166
167         return version;
168 }
169
170 static void parse_one_symref_info(struct string_list *symref, const char *val, int len)
171 {
172         char *sym, *target;
173         struct string_list_item *item;
174
175         if (!len)
176                 return; /* just "symref" */
177         /* e.g. "symref=HEAD:refs/heads/master" */
178         sym = xmemdupz(val, len);
179         target = strchr(sym, ':');
180         if (!target)
181                 /* just "symref=something" */
182                 goto reject;
183         *(target++) = '\0';
184         if (check_refname_format(sym, REFNAME_ALLOW_ONELEVEL) ||
185             check_refname_format(target, REFNAME_ALLOW_ONELEVEL))
186                 /* "symref=bogus:pair */
187                 goto reject;
188         item = string_list_append_nodup(symref, sym);
189         item->util = target;
190         return;
191 reject:
192         free(sym);
193         return;
194 }
195
196 static void annotate_refs_with_symref_info(struct ref *ref)
197 {
198         struct string_list symref = STRING_LIST_INIT_DUP;
199         int offset = 0;
200
201         while (1) {
202                 int len;
203                 const char *val;
204
205                 val = next_server_feature_value("symref", &len, &offset);
206                 if (!val)
207                         break;
208                 parse_one_symref_info(&symref, val, len);
209         }
210         string_list_sort(&symref);
211
212         for (; ref; ref = ref->next) {
213                 struct string_list_item *item;
214                 item = string_list_lookup(&symref, ref->name);
215                 if (!item)
216                         continue;
217                 ref->symref = xstrdup((char *)item->util);
218         }
219         string_list_clear(&symref, 0);
220 }
221
222 static void process_capabilities(struct packet_reader *reader, int *linelen)
223 {
224         const char *feat_val;
225         int feat_len;
226         const char *line = reader->line;
227         int nul_location = strlen(line);
228         if (nul_location == *linelen)
229                 return;
230         server_capabilities_v1 = xstrdup(line + nul_location + 1);
231         *linelen = nul_location;
232
233         feat_val = server_feature_value("object-format", &feat_len);
234         if (feat_val) {
235                 char *hash_name = xstrndup(feat_val, feat_len);
236                 int hash_algo = hash_algo_by_name(hash_name);
237                 if (hash_algo != GIT_HASH_UNKNOWN)
238                         reader->hash_algo = &hash_algos[hash_algo];
239                 free(hash_name);
240         } else {
241                 reader->hash_algo = &hash_algos[GIT_HASH_SHA1];
242         }
243 }
244
245 static int process_dummy_ref(const struct packet_reader *reader)
246 {
247         const char *line = reader->line;
248         struct object_id oid;
249         const char *name;
250
251         if (parse_oid_hex_algop(line, &oid, &name, reader->hash_algo))
252                 return 0;
253         if (*name != ' ')
254                 return 0;
255         name++;
256
257         return oideq(&null_oid, &oid) && !strcmp(name, "capabilities^{}");
258 }
259
260 static void check_no_capabilities(const char *line, int len)
261 {
262         if (strlen(line) != len)
263                 warning(_("ignoring capabilities after first line '%s'"),
264                         line + strlen(line));
265 }
266
267 static int process_ref(const struct packet_reader *reader, int len,
268                        struct ref ***list, unsigned int flags,
269                        struct oid_array *extra_have)
270 {
271         const char *line = reader->line;
272         struct object_id old_oid;
273         const char *name;
274
275         if (parse_oid_hex_algop(line, &old_oid, &name, reader->hash_algo))
276                 return 0;
277         if (*name != ' ')
278                 return 0;
279         name++;
280
281         if (extra_have && !strcmp(name, ".have")) {
282                 oid_array_append(extra_have, &old_oid);
283         } else if (!strcmp(name, "capabilities^{}")) {
284                 die(_("protocol error: unexpected capabilities^{}"));
285         } else if (check_ref(name, flags)) {
286                 struct ref *ref = alloc_ref(name);
287                 oidcpy(&ref->old_oid, &old_oid);
288                 **list = ref;
289                 *list = &ref->next;
290         }
291         check_no_capabilities(line, len);
292         return 1;
293 }
294
295 static int process_shallow(const struct packet_reader *reader, int len,
296                            struct oid_array *shallow_points)
297 {
298         const char *line = reader->line;
299         const char *arg;
300         struct object_id old_oid;
301
302         if (!skip_prefix(line, "shallow ", &arg))
303                 return 0;
304
305         if (get_oid_hex_algop(arg, &old_oid, reader->hash_algo))
306                 die(_("protocol error: expected shallow sha-1, got '%s'"), arg);
307         if (!shallow_points)
308                 die(_("repository on the other end cannot be shallow"));
309         oid_array_append(shallow_points, &old_oid);
310         check_no_capabilities(line, len);
311         return 1;
312 }
313
314 enum get_remote_heads_state {
315         EXPECTING_FIRST_REF = 0,
316         EXPECTING_REF,
317         EXPECTING_SHALLOW,
318         EXPECTING_DONE,
319 };
320
321 /*
322  * Read all the refs from the other end
323  */
324 struct ref **get_remote_heads(struct packet_reader *reader,
325                               struct ref **list, unsigned int flags,
326                               struct oid_array *extra_have,
327                               struct oid_array *shallow_points)
328 {
329         struct ref **orig_list = list;
330         int len = 0;
331         enum get_remote_heads_state state = EXPECTING_FIRST_REF;
332
333         *list = NULL;
334
335         while (state != EXPECTING_DONE) {
336                 switch (packet_reader_read(reader)) {
337                 case PACKET_READ_EOF:
338                         die_initial_contact(1);
339                 case PACKET_READ_NORMAL:
340                         len = reader->pktlen;
341                         break;
342                 case PACKET_READ_FLUSH:
343                         state = EXPECTING_DONE;
344                         break;
345                 case PACKET_READ_DELIM:
346                 case PACKET_READ_RESPONSE_END:
347                         die(_("invalid packet"));
348                 }
349
350                 switch (state) {
351                 case EXPECTING_FIRST_REF:
352                         process_capabilities(reader, &len);
353                         if (process_dummy_ref(reader)) {
354                                 state = EXPECTING_SHALLOW;
355                                 break;
356                         }
357                         state = EXPECTING_REF;
358                         /* fallthrough */
359                 case EXPECTING_REF:
360                         if (process_ref(reader, len, &list, flags, extra_have))
361                                 break;
362                         state = EXPECTING_SHALLOW;
363                         /* fallthrough */
364                 case EXPECTING_SHALLOW:
365                         if (process_shallow(reader, len, shallow_points))
366                                 break;
367                         die(_("protocol error: unexpected '%s'"), reader->line);
368                 case EXPECTING_DONE:
369                         break;
370                 }
371         }
372
373         annotate_refs_with_symref_info(*orig_list);
374
375         return list;
376 }
377
378 /* Returns 1 when a valid ref has been added to `list`, 0 otherwise */
379 static int process_ref_v2(struct packet_reader *reader, struct ref ***list,
380                           char **unborn_head_target)
381 {
382         int ret = 1;
383         int i = 0;
384         struct object_id old_oid;
385         struct ref *ref;
386         struct string_list line_sections = STRING_LIST_INIT_DUP;
387         const char *end;
388         const char *line = reader->line;
389
390         /*
391          * Ref lines have a number of fields which are space deliminated.  The
392          * first field is the OID of the ref.  The second field is the ref
393          * name.  Subsequent fields (symref-target and peeled) are optional and
394          * don't have a particular order.
395          */
396         if (string_list_split(&line_sections, line, ' ', -1) < 2) {
397                 ret = 0;
398                 goto out;
399         }
400
401         if (!strcmp("unborn", line_sections.items[i].string)) {
402                 i++;
403                 if (unborn_head_target &&
404                     !strcmp("HEAD", line_sections.items[i++].string)) {
405                         /*
406                          * Look for the symref target (if any). If found,
407                          * return it to the caller.
408                          */
409                         for (; i < line_sections.nr; i++) {
410                                 const char *arg = line_sections.items[i].string;
411
412                                 if (skip_prefix(arg, "symref-target:", &arg)) {
413                                         *unborn_head_target = xstrdup(arg);
414                                         break;
415                                 }
416                         }
417                 }
418                 goto out;
419         }
420         if (parse_oid_hex_algop(line_sections.items[i++].string, &old_oid, &end, reader->hash_algo) ||
421             *end) {
422                 ret = 0;
423                 goto out;
424         }
425
426         ref = alloc_ref(line_sections.items[i++].string);
427
428         memcpy(ref->old_oid.hash, old_oid.hash, reader->hash_algo->rawsz);
429         **list = ref;
430         *list = &ref->next;
431
432         for (; i < line_sections.nr; i++) {
433                 const char *arg = line_sections.items[i].string;
434                 if (skip_prefix(arg, "symref-target:", &arg))
435                         ref->symref = xstrdup(arg);
436
437                 if (skip_prefix(arg, "peeled:", &arg)) {
438                         struct object_id peeled_oid;
439                         char *peeled_name;
440                         struct ref *peeled;
441                         if (parse_oid_hex_algop(arg, &peeled_oid, &end,
442                                                 reader->hash_algo) || *end) {
443                                 ret = 0;
444                                 goto out;
445                         }
446
447                         peeled_name = xstrfmt("%s^{}", ref->name);
448                         peeled = alloc_ref(peeled_name);
449
450                         memcpy(peeled->old_oid.hash, peeled_oid.hash,
451                                reader->hash_algo->rawsz);
452                         **list = peeled;
453                         *list = &peeled->next;
454
455                         free(peeled_name);
456                 }
457         }
458
459 out:
460         string_list_clear(&line_sections, 0);
461         return ret;
462 }
463
464 void check_stateless_delimiter(int stateless_rpc,
465                               struct packet_reader *reader,
466                               const char *error)
467 {
468         if (!stateless_rpc)
469                 return; /* not in stateless mode, no delimiter expected */
470         if (packet_reader_read(reader) != PACKET_READ_RESPONSE_END)
471                 die("%s", error);
472 }
473
474 struct ref **get_remote_refs(int fd_out, struct packet_reader *reader,
475                              struct ref **list, int for_push,
476                              struct transport_ls_refs_options *transport_options,
477                              const struct string_list *server_options,
478                              int stateless_rpc)
479 {
480         int i;
481         const char *hash_name;
482         struct strvec *ref_prefixes = transport_options ?
483                 &transport_options->ref_prefixes : NULL;
484         char **unborn_head_target = transport_options ?
485                 &transport_options->unborn_head_target : NULL;
486         *list = NULL;
487
488         if (server_supports_v2("ls-refs", 1))
489                 packet_write_fmt(fd_out, "command=ls-refs\n");
490
491         if (server_supports_v2("agent", 0))
492                 packet_write_fmt(fd_out, "agent=%s", git_user_agent_sanitized());
493
494         if (server_feature_v2("object-format", &hash_name)) {
495                 int hash_algo = hash_algo_by_name(hash_name);
496                 if (hash_algo == GIT_HASH_UNKNOWN)
497                         die(_("unknown object format '%s' specified by server"), hash_name);
498                 reader->hash_algo = &hash_algos[hash_algo];
499                 packet_write_fmt(fd_out, "object-format=%s", reader->hash_algo->name);
500         } else {
501                 reader->hash_algo = &hash_algos[GIT_HASH_SHA1];
502         }
503
504         if (server_options && server_options->nr &&
505             server_supports_v2("server-option", 1))
506                 for (i = 0; i < server_options->nr; i++)
507                         packet_write_fmt(fd_out, "server-option=%s",
508                                          server_options->items[i].string);
509
510         packet_delim(fd_out);
511         /* When pushing we don't want to request the peeled tags */
512         if (!for_push)
513                 packet_write_fmt(fd_out, "peel\n");
514         packet_write_fmt(fd_out, "symrefs\n");
515         if (server_supports_feature("ls-refs", "unborn", 0))
516                 packet_write_fmt(fd_out, "unborn\n");
517         for (i = 0; ref_prefixes && i < ref_prefixes->nr; i++) {
518                 packet_write_fmt(fd_out, "ref-prefix %s\n",
519                                  ref_prefixes->v[i]);
520         }
521         packet_flush(fd_out);
522
523         /* Process response from server */
524         while (packet_reader_read(reader) == PACKET_READ_NORMAL) {
525                 if (!process_ref_v2(reader, &list, unborn_head_target))
526                         die(_("invalid ls-refs response: %s"), reader->line);
527         }
528
529         if (reader->status != PACKET_READ_FLUSH)
530                 die(_("expected flush after ref listing"));
531
532         check_stateless_delimiter(stateless_rpc, reader,
533                                   _("expected response end packet after ref listing"));
534
535         return list;
536 }
537
538 const char *parse_feature_value(const char *feature_list, const char *feature, int *lenp, int *offset)
539 {
540         int len;
541
542         if (!feature_list)
543                 return NULL;
544
545         len = strlen(feature);
546         if (offset)
547                 feature_list += *offset;
548         while (*feature_list) {
549                 const char *found = strstr(feature_list, feature);
550                 if (!found)
551                         return NULL;
552                 if (feature_list == found || isspace(found[-1])) {
553                         const char *value = found + len;
554                         /* feature with no value (e.g., "thin-pack") */
555                         if (!*value || isspace(*value)) {
556                                 if (lenp)
557                                         *lenp = 0;
558                                 return value;
559                         }
560                         /* feature with a value (e.g., "agent=git/1.2.3") */
561                         else if (*value == '=') {
562                                 int end;
563
564                                 value++;
565                                 end = strcspn(value, " \t\n");
566                                 if (lenp)
567                                         *lenp = end;
568                                 if (offset)
569                                         *offset = value + end - feature_list;
570                                 return value;
571                         }
572                         /*
573                          * otherwise we matched a substring of another feature;
574                          * keep looking
575                          */
576                 }
577                 feature_list = found + 1;
578         }
579         return NULL;
580 }
581
582 int server_supports_hash(const char *desired, int *feature_supported)
583 {
584         int offset = 0;
585         int len;
586         const char *hash;
587
588         hash = next_server_feature_value("object-format", &len, &offset);
589         if (feature_supported)
590                 *feature_supported = !!hash;
591         if (!hash) {
592                 hash = hash_algos[GIT_HASH_SHA1].name;
593                 len = strlen(hash);
594         }
595         while (hash) {
596                 if (!xstrncmpz(desired, hash, len))
597                         return 1;
598
599                 hash = next_server_feature_value("object-format", &len, &offset);
600         }
601         return 0;
602 }
603
604 int parse_feature_request(const char *feature_list, const char *feature)
605 {
606         return !!parse_feature_value(feature_list, feature, NULL, NULL);
607 }
608
609 static const char *next_server_feature_value(const char *feature, int *len, int *offset)
610 {
611         return parse_feature_value(server_capabilities_v1, feature, len, offset);
612 }
613
614 const char *server_feature_value(const char *feature, int *len)
615 {
616         return parse_feature_value(server_capabilities_v1, feature, len, NULL);
617 }
618
619 int server_supports(const char *feature)
620 {
621         return !!server_feature_value(feature, NULL);
622 }
623
624 enum protocol {
625         PROTO_LOCAL = 1,
626         PROTO_FILE,
627         PROTO_SSH,
628         PROTO_GIT
629 };
630
631 int url_is_local_not_ssh(const char *url)
632 {
633         const char *colon = strchr(url, ':');
634         const char *slash = strchr(url, '/');
635         return !colon || (slash && slash < colon) ||
636                 (has_dos_drive_prefix(url) && is_valid_path(url));
637 }
638
639 static const char *prot_name(enum protocol protocol)
640 {
641         switch (protocol) {
642                 case PROTO_LOCAL:
643                 case PROTO_FILE:
644                         return "file";
645                 case PROTO_SSH:
646                         return "ssh";
647                 case PROTO_GIT:
648                         return "git";
649                 default:
650                         return "unknown protocol";
651         }
652 }
653
654 static enum protocol get_protocol(const char *name)
655 {
656         if (!strcmp(name, "ssh"))
657                 return PROTO_SSH;
658         if (!strcmp(name, "git"))
659                 return PROTO_GIT;
660         if (!strcmp(name, "git+ssh")) /* deprecated - do not use */
661                 return PROTO_SSH;
662         if (!strcmp(name, "ssh+git")) /* deprecated - do not use */
663                 return PROTO_SSH;
664         if (!strcmp(name, "file"))
665                 return PROTO_FILE;
666         die(_("protocol '%s' is not supported"), name);
667 }
668
669 static char *host_end(char **hoststart, int removebrackets)
670 {
671         char *host = *hoststart;
672         char *end;
673         char *start = strstr(host, "@[");
674         if (start)
675                 start++; /* Jump over '@' */
676         else
677                 start = host;
678         if (start[0] == '[') {
679                 end = strchr(start + 1, ']');
680                 if (end) {
681                         if (removebrackets) {
682                                 *end = 0;
683                                 memmove(start, start + 1, end - start);
684                                 end++;
685                         }
686                 } else
687                         end = host;
688         } else
689                 end = host;
690         return end;
691 }
692
693 #define STR_(s) # s
694 #define STR(s)  STR_(s)
695
696 static void get_host_and_port(char **host, const char **port)
697 {
698         char *colon, *end;
699         end = host_end(host, 1);
700         colon = strchr(end, ':');
701         if (colon) {
702                 long portnr = strtol(colon + 1, &end, 10);
703                 if (end != colon + 1 && *end == '\0' && 0 <= portnr && portnr < 65536) {
704                         *colon = 0;
705                         *port = colon + 1;
706                 } else if (!colon[1]) {
707                         *colon = 0;
708                 }
709         }
710 }
711
712 static void enable_keepalive(int sockfd)
713 {
714         int ka = 1;
715
716         if (setsockopt(sockfd, SOL_SOCKET, SO_KEEPALIVE, &ka, sizeof(ka)) < 0)
717                 error_errno(_("unable to set SO_KEEPALIVE on socket"));
718 }
719
720 #ifndef NO_IPV6
721
722 static const char *ai_name(const struct addrinfo *ai)
723 {
724         static char addr[NI_MAXHOST];
725         if (getnameinfo(ai->ai_addr, ai->ai_addrlen, addr, sizeof(addr), NULL, 0,
726                         NI_NUMERICHOST) != 0)
727                 xsnprintf(addr, sizeof(addr), "(unknown)");
728
729         return addr;
730 }
731
732 /*
733  * Returns a connected socket() fd, or else die()s.
734  */
735 static int git_tcp_connect_sock(char *host, int flags)
736 {
737         struct strbuf error_message = STRBUF_INIT;
738         int sockfd = -1;
739         const char *port = STR(DEFAULT_GIT_PORT);
740         struct addrinfo hints, *ai0, *ai;
741         int gai;
742         int cnt = 0;
743
744         get_host_and_port(&host, &port);
745         if (!*port)
746                 port = "<none>";
747
748         memset(&hints, 0, sizeof(hints));
749         if (flags & CONNECT_IPV4)
750                 hints.ai_family = AF_INET;
751         else if (flags & CONNECT_IPV6)
752                 hints.ai_family = AF_INET6;
753         hints.ai_socktype = SOCK_STREAM;
754         hints.ai_protocol = IPPROTO_TCP;
755
756         if (flags & CONNECT_VERBOSE)
757                 fprintf(stderr, _("Looking up %s ... "), host);
758
759         gai = getaddrinfo(host, port, &hints, &ai);
760         if (gai)
761                 die(_("unable to look up %s (port %s) (%s)"), host, port, gai_strerror(gai));
762
763         if (flags & CONNECT_VERBOSE)
764                 /* TRANSLATORS: this is the end of "Looking up %s ... " */
765                 fprintf(stderr, _("done.\nConnecting to %s (port %s) ... "), host, port);
766
767         for (ai0 = ai; ai; ai = ai->ai_next, cnt++) {
768                 sockfd = socket(ai->ai_family,
769                                 ai->ai_socktype, ai->ai_protocol);
770                 if ((sockfd < 0) ||
771                     (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0)) {
772                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
773                                     host, cnt, ai_name(ai), strerror(errno));
774                         if (0 <= sockfd)
775                                 close(sockfd);
776                         sockfd = -1;
777                         continue;
778                 }
779                 if (flags & CONNECT_VERBOSE)
780                         fprintf(stderr, "%s ", ai_name(ai));
781                 break;
782         }
783
784         freeaddrinfo(ai0);
785
786         if (sockfd < 0)
787                 die(_("unable to connect to %s:\n%s"), host, error_message.buf);
788
789         enable_keepalive(sockfd);
790
791         if (flags & CONNECT_VERBOSE)
792                 /* TRANSLATORS: this is the end of "Connecting to %s (port %s) ... " */
793                 fprintf_ln(stderr, _("done."));
794
795         strbuf_release(&error_message);
796
797         return sockfd;
798 }
799
800 #else /* NO_IPV6 */
801
802 /*
803  * Returns a connected socket() fd, or else die()s.
804  */
805 static int git_tcp_connect_sock(char *host, int flags)
806 {
807         struct strbuf error_message = STRBUF_INIT;
808         int sockfd = -1;
809         const char *port = STR(DEFAULT_GIT_PORT);
810         char *ep;
811         struct hostent *he;
812         struct sockaddr_in sa;
813         char **ap;
814         unsigned int nport;
815         int cnt;
816
817         get_host_and_port(&host, &port);
818
819         if (flags & CONNECT_VERBOSE)
820                 fprintf(stderr, _("Looking up %s ... "), host);
821
822         he = gethostbyname(host);
823         if (!he)
824                 die(_("unable to look up %s (%s)"), host, hstrerror(h_errno));
825         nport = strtoul(port, &ep, 10);
826         if ( ep == port || *ep ) {
827                 /* Not numeric */
828                 struct servent *se = getservbyname(port,"tcp");
829                 if ( !se )
830                         die(_("unknown port %s"), port);
831                 nport = se->s_port;
832         }
833
834         if (flags & CONNECT_VERBOSE)
835                 /* TRANSLATORS: this is the end of "Looking up %s ... " */
836                 fprintf(stderr, _("done.\nConnecting to %s (port %s) ... "), host, port);
837
838         for (cnt = 0, ap = he->h_addr_list; *ap; ap++, cnt++) {
839                 memset(&sa, 0, sizeof sa);
840                 sa.sin_family = he->h_addrtype;
841                 sa.sin_port = htons(nport);
842                 memcpy(&sa.sin_addr, *ap, he->h_length);
843
844                 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
845                 if ((sockfd < 0) ||
846                     connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
847                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
848                                 host,
849                                 cnt,
850                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr),
851                                 strerror(errno));
852                         if (0 <= sockfd)
853                                 close(sockfd);
854                         sockfd = -1;
855                         continue;
856                 }
857                 if (flags & CONNECT_VERBOSE)
858                         fprintf(stderr, "%s ",
859                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr));
860                 break;
861         }
862
863         if (sockfd < 0)
864                 die(_("unable to connect to %s:\n%s"), host, error_message.buf);
865
866         enable_keepalive(sockfd);
867
868         if (flags & CONNECT_VERBOSE)
869                 /* TRANSLATORS: this is the end of "Connecting to %s (port %s) ... " */
870                 fprintf_ln(stderr, _("done."));
871
872         return sockfd;
873 }
874
875 #endif /* NO_IPV6 */
876
877
878 /*
879  * Dummy child_process returned by git_connect() if the transport protocol
880  * does not need fork(2).
881  */
882 static struct child_process no_fork = CHILD_PROCESS_INIT;
883
884 int git_connection_is_socket(struct child_process *conn)
885 {
886         return conn == &no_fork;
887 }
888
889 static struct child_process *git_tcp_connect(int fd[2], char *host, int flags)
890 {
891         int sockfd = git_tcp_connect_sock(host, flags);
892
893         fd[0] = sockfd;
894         fd[1] = dup(sockfd);
895
896         return &no_fork;
897 }
898
899
900 static char *git_proxy_command;
901
902 static int git_proxy_command_options(const char *var, const char *value,
903                 void *cb)
904 {
905         if (!strcmp(var, "core.gitproxy")) {
906                 const char *for_pos;
907                 int matchlen = -1;
908                 int hostlen;
909                 const char *rhost_name = cb;
910                 int rhost_len = strlen(rhost_name);
911
912                 if (git_proxy_command)
913                         return 0;
914                 if (!value)
915                         return config_error_nonbool(var);
916                 /* [core]
917                  * ;# matches www.kernel.org as well
918                  * gitproxy = netcatter-1 for kernel.org
919                  * gitproxy = netcatter-2 for sample.xz
920                  * gitproxy = netcatter-default
921                  */
922                 for_pos = strstr(value, " for ");
923                 if (!for_pos)
924                         /* matches everybody */
925                         matchlen = strlen(value);
926                 else {
927                         hostlen = strlen(for_pos + 5);
928                         if (rhost_len < hostlen)
929                                 matchlen = -1;
930                         else if (!strncmp(for_pos + 5,
931                                           rhost_name + rhost_len - hostlen,
932                                           hostlen) &&
933                                  ((rhost_len == hostlen) ||
934                                   rhost_name[rhost_len - hostlen -1] == '.'))
935                                 matchlen = for_pos - value;
936                         else
937                                 matchlen = -1;
938                 }
939                 if (0 <= matchlen) {
940                         /* core.gitproxy = none for kernel.org */
941                         if (matchlen == 4 &&
942                             !memcmp(value, "none", 4))
943                                 matchlen = 0;
944                         git_proxy_command = xmemdupz(value, matchlen);
945                 }
946                 return 0;
947         }
948
949         return git_default_config(var, value, cb);
950 }
951
952 static int git_use_proxy(const char *host)
953 {
954         git_proxy_command = getenv("GIT_PROXY_COMMAND");
955         git_config(git_proxy_command_options, (void*)host);
956         return (git_proxy_command && *git_proxy_command);
957 }
958
959 static struct child_process *git_proxy_connect(int fd[2], char *host)
960 {
961         const char *port = STR(DEFAULT_GIT_PORT);
962         struct child_process *proxy;
963
964         get_host_and_port(&host, &port);
965
966         if (looks_like_command_line_option(host))
967                 die(_("strange hostname '%s' blocked"), host);
968         if (looks_like_command_line_option(port))
969                 die(_("strange port '%s' blocked"), port);
970
971         proxy = xmalloc(sizeof(*proxy));
972         child_process_init(proxy);
973         strvec_push(&proxy->args, git_proxy_command);
974         strvec_push(&proxy->args, host);
975         strvec_push(&proxy->args, port);
976         proxy->in = -1;
977         proxy->out = -1;
978         if (start_command(proxy))
979                 die(_("cannot start proxy %s"), git_proxy_command);
980         fd[0] = proxy->out; /* read from proxy stdout */
981         fd[1] = proxy->in;  /* write to proxy stdin */
982         return proxy;
983 }
984
985 static char *get_port(char *host)
986 {
987         char *end;
988         char *p = strchr(host, ':');
989
990         if (p) {
991                 long port = strtol(p + 1, &end, 10);
992                 if (end != p + 1 && *end == '\0' && 0 <= port && port < 65536) {
993                         *p = '\0';
994                         return p+1;
995                 }
996         }
997
998         return NULL;
999 }
1000
1001 /*
1002  * Extract protocol and relevant parts from the specified connection URL.
1003  * The caller must free() the returned strings.
1004  */
1005 static enum protocol parse_connect_url(const char *url_orig, char **ret_host,
1006                                        char **ret_path)
1007 {
1008         char *url;
1009         char *host, *path;
1010         char *end;
1011         int separator = '/';
1012         enum protocol protocol = PROTO_LOCAL;
1013
1014         if (is_url(url_orig))
1015                 url = url_decode(url_orig);
1016         else
1017                 url = xstrdup(url_orig);
1018
1019         host = strstr(url, "://");
1020         if (host) {
1021                 *host = '\0';
1022                 protocol = get_protocol(url);
1023                 host += 3;
1024         } else {
1025                 host = url;
1026                 if (!url_is_local_not_ssh(url)) {
1027                         protocol = PROTO_SSH;
1028                         separator = ':';
1029                 }
1030         }
1031
1032         /*
1033          * Don't do destructive transforms as protocol code does
1034          * '[]' unwrapping in get_host_and_port()
1035          */
1036         end = host_end(&host, 0);
1037
1038         if (protocol == PROTO_LOCAL)
1039                 path = end;
1040         else if (protocol == PROTO_FILE && *host != '/' &&
1041                  !has_dos_drive_prefix(host) &&
1042                  offset_1st_component(host - 2) > 1)
1043                 path = host - 2; /* include the leading "//" */
1044         else if (protocol == PROTO_FILE && has_dos_drive_prefix(end))
1045                 path = end; /* "file://$(pwd)" may be "file://C:/projects/repo" */
1046         else
1047                 path = strchr(end, separator);
1048
1049         if (!path || !*path)
1050                 die(_("no path specified; see 'git help pull' for valid url syntax"));
1051
1052         /*
1053          * null-terminate hostname and point path to ~ for URL's like this:
1054          *    ssh://host.xz/~user/repo
1055          */
1056
1057         end = path; /* Need to \0 terminate host here */
1058         if (separator == ':')
1059                 path++; /* path starts after ':' */
1060         if (protocol == PROTO_GIT || protocol == PROTO_SSH) {
1061                 if (path[1] == '~')
1062                         path++;
1063         }
1064
1065         path = xstrdup(path);
1066         *end = '\0';
1067
1068         *ret_host = xstrdup(host);
1069         *ret_path = path;
1070         free(url);
1071         return protocol;
1072 }
1073
1074 static const char *get_ssh_command(void)
1075 {
1076         const char *ssh;
1077
1078         if ((ssh = getenv("GIT_SSH_COMMAND")))
1079                 return ssh;
1080
1081         if (!git_config_get_string_tmp("core.sshcommand", &ssh))
1082                 return ssh;
1083
1084         return NULL;
1085 }
1086
1087 enum ssh_variant {
1088         VARIANT_AUTO,
1089         VARIANT_SIMPLE,
1090         VARIANT_SSH,
1091         VARIANT_PLINK,
1092         VARIANT_PUTTY,
1093         VARIANT_TORTOISEPLINK,
1094 };
1095
1096 static void override_ssh_variant(enum ssh_variant *ssh_variant)
1097 {
1098         const char *variant = getenv("GIT_SSH_VARIANT");
1099
1100         if (!variant && git_config_get_string_tmp("ssh.variant", &variant))
1101                 return;
1102
1103         if (!strcmp(variant, "auto"))
1104                 *ssh_variant = VARIANT_AUTO;
1105         else if (!strcmp(variant, "plink"))
1106                 *ssh_variant = VARIANT_PLINK;
1107         else if (!strcmp(variant, "putty"))
1108                 *ssh_variant = VARIANT_PUTTY;
1109         else if (!strcmp(variant, "tortoiseplink"))
1110                 *ssh_variant = VARIANT_TORTOISEPLINK;
1111         else if (!strcmp(variant, "simple"))
1112                 *ssh_variant = VARIANT_SIMPLE;
1113         else
1114                 *ssh_variant = VARIANT_SSH;
1115 }
1116
1117 static enum ssh_variant determine_ssh_variant(const char *ssh_command,
1118                                               int is_cmdline)
1119 {
1120         enum ssh_variant ssh_variant = VARIANT_AUTO;
1121         const char *variant;
1122         char *p = NULL;
1123
1124         override_ssh_variant(&ssh_variant);
1125
1126         if (ssh_variant != VARIANT_AUTO)
1127                 return ssh_variant;
1128
1129         if (!is_cmdline) {
1130                 p = xstrdup(ssh_command);
1131                 variant = basename(p);
1132         } else {
1133                 const char **ssh_argv;
1134
1135                 p = xstrdup(ssh_command);
1136                 if (split_cmdline(p, &ssh_argv) > 0) {
1137                         variant = basename((char *)ssh_argv[0]);
1138                         /*
1139                          * At this point, variant points into the buffer
1140                          * referenced by p, hence we do not need ssh_argv
1141                          * any longer.
1142                          */
1143                         free(ssh_argv);
1144                 } else {
1145                         free(p);
1146                         return ssh_variant;
1147                 }
1148         }
1149
1150         if (!strcasecmp(variant, "ssh") ||
1151             !strcasecmp(variant, "ssh.exe"))
1152                 ssh_variant = VARIANT_SSH;
1153         else if (!strcasecmp(variant, "plink") ||
1154                  !strcasecmp(variant, "plink.exe"))
1155                 ssh_variant = VARIANT_PLINK;
1156         else if (!strcasecmp(variant, "tortoiseplink") ||
1157                  !strcasecmp(variant, "tortoiseplink.exe"))
1158                 ssh_variant = VARIANT_TORTOISEPLINK;
1159
1160         free(p);
1161         return ssh_variant;
1162 }
1163
1164 /*
1165  * Open a connection using Git's native protocol.
1166  *
1167  * The caller is responsible for freeing hostandport, but this function may
1168  * modify it (for example, to truncate it to remove the port part).
1169  */
1170 static struct child_process *git_connect_git(int fd[2], char *hostandport,
1171                                              const char *path, const char *prog,
1172                                              enum protocol_version version,
1173                                              int flags)
1174 {
1175         struct child_process *conn;
1176         struct strbuf request = STRBUF_INIT;
1177         /*
1178          * Set up virtual host information based on where we will
1179          * connect, unless the user has overridden us in
1180          * the environment.
1181          */
1182         char *target_host = getenv("GIT_OVERRIDE_VIRTUAL_HOST");
1183         if (target_host)
1184                 target_host = xstrdup(target_host);
1185         else
1186                 target_host = xstrdup(hostandport);
1187
1188         transport_check_allowed("git");
1189         if (strchr(target_host, '\n') || strchr(path, '\n'))
1190                 die(_("newline is forbidden in git:// hosts and repo paths"));
1191
1192         /*
1193          * These underlying connection commands die() if they
1194          * cannot connect.
1195          */
1196         if (git_use_proxy(hostandport))
1197                 conn = git_proxy_connect(fd, hostandport);
1198         else
1199                 conn = git_tcp_connect(fd, hostandport, flags);
1200         /*
1201          * Separate original protocol components prog and path
1202          * from extended host header with a NUL byte.
1203          *
1204          * Note: Do not add any other headers here!  Doing so
1205          * will cause older git-daemon servers to crash.
1206          */
1207         strbuf_addf(&request,
1208                     "%s %s%chost=%s%c",
1209                     prog, path, 0,
1210                     target_host, 0);
1211
1212         /* If using a new version put that stuff here after a second null byte */
1213         if (version > 0) {
1214                 strbuf_addch(&request, '\0');
1215                 strbuf_addf(&request, "version=%d%c",
1216                             version, '\0');
1217         }
1218
1219         packet_write(fd[1], request.buf, request.len);
1220
1221         free(target_host);
1222         strbuf_release(&request);
1223         return conn;
1224 }
1225
1226 /*
1227  * Append the appropriate environment variables to `env` and options to
1228  * `args` for running ssh in Git's SSH-tunneled transport.
1229  */
1230 static void push_ssh_options(struct strvec *args, struct strvec *env,
1231                              enum ssh_variant variant, const char *port,
1232                              enum protocol_version version, int flags)
1233 {
1234         if (variant == VARIANT_SSH &&
1235             version > 0) {
1236                 strvec_push(args, "-o");
1237                 strvec_push(args, "SendEnv=" GIT_PROTOCOL_ENVIRONMENT);
1238                 strvec_pushf(env, GIT_PROTOCOL_ENVIRONMENT "=version=%d",
1239                              version);
1240         }
1241
1242         if (flags & CONNECT_IPV4) {
1243                 switch (variant) {
1244                 case VARIANT_AUTO:
1245                         BUG("VARIANT_AUTO passed to push_ssh_options");
1246                 case VARIANT_SIMPLE:
1247                         die(_("ssh variant 'simple' does not support -4"));
1248                 case VARIANT_SSH:
1249                 case VARIANT_PLINK:
1250                 case VARIANT_PUTTY:
1251                 case VARIANT_TORTOISEPLINK:
1252                         strvec_push(args, "-4");
1253                 }
1254         } else if (flags & CONNECT_IPV6) {
1255                 switch (variant) {
1256                 case VARIANT_AUTO:
1257                         BUG("VARIANT_AUTO passed to push_ssh_options");
1258                 case VARIANT_SIMPLE:
1259                         die(_("ssh variant 'simple' does not support -6"));
1260                 case VARIANT_SSH:
1261                 case VARIANT_PLINK:
1262                 case VARIANT_PUTTY:
1263                 case VARIANT_TORTOISEPLINK:
1264                         strvec_push(args, "-6");
1265                 }
1266         }
1267
1268         if (variant == VARIANT_TORTOISEPLINK)
1269                 strvec_push(args, "-batch");
1270
1271         if (port) {
1272                 switch (variant) {
1273                 case VARIANT_AUTO:
1274                         BUG("VARIANT_AUTO passed to push_ssh_options");
1275                 case VARIANT_SIMPLE:
1276                         die(_("ssh variant 'simple' does not support setting port"));
1277                 case VARIANT_SSH:
1278                         strvec_push(args, "-p");
1279                         break;
1280                 case VARIANT_PLINK:
1281                 case VARIANT_PUTTY:
1282                 case VARIANT_TORTOISEPLINK:
1283                         strvec_push(args, "-P");
1284                 }
1285
1286                 strvec_push(args, port);
1287         }
1288 }
1289
1290 /* Prepare a child_process for use by Git's SSH-tunneled transport. */
1291 static void fill_ssh_args(struct child_process *conn, const char *ssh_host,
1292                           const char *port, enum protocol_version version,
1293                           int flags)
1294 {
1295         const char *ssh;
1296         enum ssh_variant variant;
1297
1298         if (looks_like_command_line_option(ssh_host))
1299                 die(_("strange hostname '%s' blocked"), ssh_host);
1300
1301         ssh = get_ssh_command();
1302         if (ssh) {
1303                 variant = determine_ssh_variant(ssh, 1);
1304         } else {
1305                 /*
1306                  * GIT_SSH is the no-shell version of
1307                  * GIT_SSH_COMMAND (and must remain so for
1308                  * historical compatibility).
1309                  */
1310                 conn->use_shell = 0;
1311
1312                 ssh = getenv("GIT_SSH");
1313                 if (!ssh)
1314                         ssh = "ssh";
1315                 variant = determine_ssh_variant(ssh, 0);
1316         }
1317
1318         if (variant == VARIANT_AUTO) {
1319                 struct child_process detect = CHILD_PROCESS_INIT;
1320
1321                 detect.use_shell = conn->use_shell;
1322                 detect.no_stdin = detect.no_stdout = detect.no_stderr = 1;
1323
1324                 strvec_push(&detect.args, ssh);
1325                 strvec_push(&detect.args, "-G");
1326                 push_ssh_options(&detect.args, &detect.env_array,
1327                                  VARIANT_SSH, port, version, flags);
1328                 strvec_push(&detect.args, ssh_host);
1329
1330                 variant = run_command(&detect) ? VARIANT_SIMPLE : VARIANT_SSH;
1331         }
1332
1333         strvec_push(&conn->args, ssh);
1334         push_ssh_options(&conn->args, &conn->env_array, variant, port, version, flags);
1335         strvec_push(&conn->args, ssh_host);
1336 }
1337
1338 /*
1339  * This returns the dummy child_process `no_fork` if the transport protocol
1340  * does not need fork(2), or a struct child_process object if it does.  Once
1341  * done, finish the connection with finish_connect() with the value returned
1342  * from this function (it is safe to call finish_connect() with NULL to
1343  * support the former case).
1344  *
1345  * If it returns, the connect is successful; it just dies on errors (this
1346  * will hopefully be changed in a libification effort, to return NULL when
1347  * the connection failed).
1348  */
1349 struct child_process *git_connect(int fd[2], const char *url,
1350                                   const char *prog, int flags)
1351 {
1352         char *hostandport, *path;
1353         struct child_process *conn;
1354         enum protocol protocol;
1355         enum protocol_version version = get_protocol_version_config();
1356
1357         /*
1358          * NEEDSWORK: If we are trying to use protocol v2 and we are planning
1359          * to perform a push, then fallback to v0 since the client doesn't know
1360          * how to push yet using v2.
1361          */
1362         if (version == protocol_v2 && !strcmp("git-receive-pack", prog))
1363                 version = protocol_v0;
1364
1365         /* Without this we cannot rely on waitpid() to tell
1366          * what happened to our children.
1367          */
1368         signal(SIGCHLD, SIG_DFL);
1369
1370         protocol = parse_connect_url(url, &hostandport, &path);
1371         if ((flags & CONNECT_DIAG_URL) && (protocol != PROTO_SSH)) {
1372                 printf("Diag: url=%s\n", url ? url : "NULL");
1373                 printf("Diag: protocol=%s\n", prot_name(protocol));
1374                 printf("Diag: hostandport=%s\n", hostandport ? hostandport : "NULL");
1375                 printf("Diag: path=%s\n", path ? path : "NULL");
1376                 conn = NULL;
1377         } else if (protocol == PROTO_GIT) {
1378                 conn = git_connect_git(fd, hostandport, path, prog, version, flags);
1379                 conn->trace2_child_class = "transport/git";
1380         } else {
1381                 struct strbuf cmd = STRBUF_INIT;
1382                 const char *const *var;
1383
1384                 conn = xmalloc(sizeof(*conn));
1385                 child_process_init(conn);
1386
1387                 if (looks_like_command_line_option(path))
1388                         die(_("strange pathname '%s' blocked"), path);
1389
1390                 strbuf_addstr(&cmd, prog);
1391                 strbuf_addch(&cmd, ' ');
1392                 sq_quote_buf(&cmd, path);
1393
1394                 /* remove repo-local variables from the environment */
1395                 for (var = local_repo_env; *var; var++)
1396                         strvec_push(&conn->env_array, *var);
1397
1398                 conn->use_shell = 1;
1399                 conn->in = conn->out = -1;
1400                 if (protocol == PROTO_SSH) {
1401                         char *ssh_host = hostandport;
1402                         const char *port = NULL;
1403                         transport_check_allowed("ssh");
1404                         get_host_and_port(&ssh_host, &port);
1405
1406                         if (!port)
1407                                 port = get_port(ssh_host);
1408
1409                         if (flags & CONNECT_DIAG_URL) {
1410                                 printf("Diag: url=%s\n", url ? url : "NULL");
1411                                 printf("Diag: protocol=%s\n", prot_name(protocol));
1412                                 printf("Diag: userandhost=%s\n", ssh_host ? ssh_host : "NULL");
1413                                 printf("Diag: port=%s\n", port ? port : "NONE");
1414                                 printf("Diag: path=%s\n", path ? path : "NULL");
1415
1416                                 free(hostandport);
1417                                 free(path);
1418                                 free(conn);
1419                                 strbuf_release(&cmd);
1420                                 return NULL;
1421                         }
1422                         conn->trace2_child_class = "transport/ssh";
1423                         fill_ssh_args(conn, ssh_host, port, version, flags);
1424                 } else {
1425                         transport_check_allowed("file");
1426                         conn->trace2_child_class = "transport/file";
1427                         if (version > 0) {
1428                                 strvec_pushf(&conn->env_array,
1429                                              GIT_PROTOCOL_ENVIRONMENT "=version=%d",
1430                                              version);
1431                         }
1432                 }
1433                 strvec_push(&conn->args, cmd.buf);
1434
1435                 if (start_command(conn))
1436                         die(_("unable to fork"));
1437
1438                 fd[0] = conn->out; /* read from child's stdout */
1439                 fd[1] = conn->in;  /* write to child's stdin */
1440                 strbuf_release(&cmd);
1441         }
1442         free(hostandport);
1443         free(path);
1444         return conn;
1445 }
1446
1447 int finish_connect(struct child_process *conn)
1448 {
1449         int code;
1450         if (!conn || git_connection_is_socket(conn))
1451                 return 0;
1452
1453         code = finish_command(conn);
1454         free(conn);
1455         return code;
1456 }