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