Merge branch 'sb/object-store-replace'
[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 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 {
413         int i;
414         *list = NULL;
415
416         if (server_supports_v2("ls-refs", 1))
417                 packet_write_fmt(fd_out, "command=ls-refs\n");
418
419         if (server_supports_v2("agent", 0))
420                 packet_write_fmt(fd_out, "agent=%s", git_user_agent_sanitized());
421
422         packet_delim(fd_out);
423         /* When pushing we don't want to request the peeled tags */
424         if (!for_push)
425                 packet_write_fmt(fd_out, "peel\n");
426         packet_write_fmt(fd_out, "symrefs\n");
427         for (i = 0; ref_prefixes && i < ref_prefixes->argc; i++) {
428                 packet_write_fmt(fd_out, "ref-prefix %s\n",
429                                  ref_prefixes->argv[i]);
430         }
431         packet_flush(fd_out);
432
433         /* Process response from server */
434         while (packet_reader_read(reader) == PACKET_READ_NORMAL) {
435                 if (!process_ref_v2(reader->line, &list))
436                         die("invalid ls-refs response: %s", reader->line);
437         }
438
439         if (reader->status != PACKET_READ_FLUSH)
440                 die("expected flush after ref listing");
441
442         return list;
443 }
444
445 static const char *parse_feature_value(const char *feature_list, const char *feature, int *lenp)
446 {
447         int len;
448
449         if (!feature_list)
450                 return NULL;
451
452         len = strlen(feature);
453         while (*feature_list) {
454                 const char *found = strstr(feature_list, feature);
455                 if (!found)
456                         return NULL;
457                 if (feature_list == found || isspace(found[-1])) {
458                         const char *value = found + len;
459                         /* feature with no value (e.g., "thin-pack") */
460                         if (!*value || isspace(*value)) {
461                                 if (lenp)
462                                         *lenp = 0;
463                                 return value;
464                         }
465                         /* feature with a value (e.g., "agent=git/1.2.3") */
466                         else if (*value == '=') {
467                                 value++;
468                                 if (lenp)
469                                         *lenp = strcspn(value, " \t\n");
470                                 return value;
471                         }
472                         /*
473                          * otherwise we matched a substring of another feature;
474                          * keep looking
475                          */
476                 }
477                 feature_list = found + 1;
478         }
479         return NULL;
480 }
481
482 int parse_feature_request(const char *feature_list, const char *feature)
483 {
484         return !!parse_feature_value(feature_list, feature, NULL);
485 }
486
487 const char *server_feature_value(const char *feature, int *len)
488 {
489         return parse_feature_value(server_capabilities_v1, feature, len);
490 }
491
492 int server_supports(const char *feature)
493 {
494         return !!server_feature_value(feature, NULL);
495 }
496
497 enum protocol {
498         PROTO_LOCAL = 1,
499         PROTO_FILE,
500         PROTO_SSH,
501         PROTO_GIT
502 };
503
504 int url_is_local_not_ssh(const char *url)
505 {
506         const char *colon = strchr(url, ':');
507         const char *slash = strchr(url, '/');
508         return !colon || (slash && slash < colon) ||
509                 has_dos_drive_prefix(url);
510 }
511
512 static const char *prot_name(enum protocol protocol)
513 {
514         switch (protocol) {
515                 case PROTO_LOCAL:
516                 case PROTO_FILE:
517                         return "file";
518                 case PROTO_SSH:
519                         return "ssh";
520                 case PROTO_GIT:
521                         return "git";
522                 default:
523                         return "unknown protocol";
524         }
525 }
526
527 static enum protocol get_protocol(const char *name)
528 {
529         if (!strcmp(name, "ssh"))
530                 return PROTO_SSH;
531         if (!strcmp(name, "git"))
532                 return PROTO_GIT;
533         if (!strcmp(name, "git+ssh")) /* deprecated - do not use */
534                 return PROTO_SSH;
535         if (!strcmp(name, "ssh+git")) /* deprecated - do not use */
536                 return PROTO_SSH;
537         if (!strcmp(name, "file"))
538                 return PROTO_FILE;
539         die("I don't handle protocol '%s'", name);
540 }
541
542 static char *host_end(char **hoststart, int removebrackets)
543 {
544         char *host = *hoststart;
545         char *end;
546         char *start = strstr(host, "@[");
547         if (start)
548                 start++; /* Jump over '@' */
549         else
550                 start = host;
551         if (start[0] == '[') {
552                 end = strchr(start + 1, ']');
553                 if (end) {
554                         if (removebrackets) {
555                                 *end = 0;
556                                 memmove(start, start + 1, end - start);
557                                 end++;
558                         }
559                 } else
560                         end = host;
561         } else
562                 end = host;
563         return end;
564 }
565
566 #define STR_(s) # s
567 #define STR(s)  STR_(s)
568
569 static void get_host_and_port(char **host, const char **port)
570 {
571         char *colon, *end;
572         end = host_end(host, 1);
573         colon = strchr(end, ':');
574         if (colon) {
575                 long portnr = strtol(colon + 1, &end, 10);
576                 if (end != colon + 1 && *end == '\0' && 0 <= portnr && portnr < 65536) {
577                         *colon = 0;
578                         *port = colon + 1;
579                 } else if (!colon[1]) {
580                         *colon = 0;
581                 }
582         }
583 }
584
585 static void enable_keepalive(int sockfd)
586 {
587         int ka = 1;
588
589         if (setsockopt(sockfd, SOL_SOCKET, SO_KEEPALIVE, &ka, sizeof(ka)) < 0)
590                 fprintf(stderr, "unable to set SO_KEEPALIVE on socket: %s\n",
591                         strerror(errno));
592 }
593
594 #ifndef NO_IPV6
595
596 static const char *ai_name(const struct addrinfo *ai)
597 {
598         static char addr[NI_MAXHOST];
599         if (getnameinfo(ai->ai_addr, ai->ai_addrlen, addr, sizeof(addr), NULL, 0,
600                         NI_NUMERICHOST) != 0)
601                 xsnprintf(addr, sizeof(addr), "(unknown)");
602
603         return addr;
604 }
605
606 /*
607  * Returns a connected socket() fd, or else die()s.
608  */
609 static int git_tcp_connect_sock(char *host, int flags)
610 {
611         struct strbuf error_message = STRBUF_INIT;
612         int sockfd = -1;
613         const char *port = STR(DEFAULT_GIT_PORT);
614         struct addrinfo hints, *ai0, *ai;
615         int gai;
616         int cnt = 0;
617
618         get_host_and_port(&host, &port);
619         if (!*port)
620                 port = "<none>";
621
622         memset(&hints, 0, sizeof(hints));
623         if (flags & CONNECT_IPV4)
624                 hints.ai_family = AF_INET;
625         else if (flags & CONNECT_IPV6)
626                 hints.ai_family = AF_INET6;
627         hints.ai_socktype = SOCK_STREAM;
628         hints.ai_protocol = IPPROTO_TCP;
629
630         if (flags & CONNECT_VERBOSE)
631                 fprintf(stderr, "Looking up %s ... ", host);
632
633         gai = getaddrinfo(host, port, &hints, &ai);
634         if (gai)
635                 die("Unable to look up %s (port %s) (%s)", host, port, gai_strerror(gai));
636
637         if (flags & CONNECT_VERBOSE)
638                 fprintf(stderr, "done.\nConnecting to %s (port %s) ... ", host, port);
639
640         for (ai0 = ai; ai; ai = ai->ai_next, cnt++) {
641                 sockfd = socket(ai->ai_family,
642                                 ai->ai_socktype, ai->ai_protocol);
643                 if ((sockfd < 0) ||
644                     (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0)) {
645                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
646                                     host, cnt, ai_name(ai), strerror(errno));
647                         if (0 <= sockfd)
648                                 close(sockfd);
649                         sockfd = -1;
650                         continue;
651                 }
652                 if (flags & CONNECT_VERBOSE)
653                         fprintf(stderr, "%s ", ai_name(ai));
654                 break;
655         }
656
657         freeaddrinfo(ai0);
658
659         if (sockfd < 0)
660                 die("unable to connect to %s:\n%s", host, error_message.buf);
661
662         enable_keepalive(sockfd);
663
664         if (flags & CONNECT_VERBOSE)
665                 fprintf(stderr, "done.\n");
666
667         strbuf_release(&error_message);
668
669         return sockfd;
670 }
671
672 #else /* NO_IPV6 */
673
674 /*
675  * Returns a connected socket() fd, or else die()s.
676  */
677 static int git_tcp_connect_sock(char *host, int flags)
678 {
679         struct strbuf error_message = STRBUF_INIT;
680         int sockfd = -1;
681         const char *port = STR(DEFAULT_GIT_PORT);
682         char *ep;
683         struct hostent *he;
684         struct sockaddr_in sa;
685         char **ap;
686         unsigned int nport;
687         int cnt;
688
689         get_host_and_port(&host, &port);
690
691         if (flags & CONNECT_VERBOSE)
692                 fprintf(stderr, "Looking up %s ... ", host);
693
694         he = gethostbyname(host);
695         if (!he)
696                 die("Unable to look up %s (%s)", host, hstrerror(h_errno));
697         nport = strtoul(port, &ep, 10);
698         if ( ep == port || *ep ) {
699                 /* Not numeric */
700                 struct servent *se = getservbyname(port,"tcp");
701                 if ( !se )
702                         die("Unknown port %s", port);
703                 nport = se->s_port;
704         }
705
706         if (flags & CONNECT_VERBOSE)
707                 fprintf(stderr, "done.\nConnecting to %s (port %s) ... ", host, port);
708
709         for (cnt = 0, ap = he->h_addr_list; *ap; ap++, cnt++) {
710                 memset(&sa, 0, sizeof sa);
711                 sa.sin_family = he->h_addrtype;
712                 sa.sin_port = htons(nport);
713                 memcpy(&sa.sin_addr, *ap, he->h_length);
714
715                 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
716                 if ((sockfd < 0) ||
717                     connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
718                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
719                                 host,
720                                 cnt,
721                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr),
722                                 strerror(errno));
723                         if (0 <= sockfd)
724                                 close(sockfd);
725                         sockfd = -1;
726                         continue;
727                 }
728                 if (flags & CONNECT_VERBOSE)
729                         fprintf(stderr, "%s ",
730                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr));
731                 break;
732         }
733
734         if (sockfd < 0)
735                 die("unable to connect to %s:\n%s", host, error_message.buf);
736
737         enable_keepalive(sockfd);
738
739         if (flags & CONNECT_VERBOSE)
740                 fprintf(stderr, "done.\n");
741
742         return sockfd;
743 }
744
745 #endif /* NO_IPV6 */
746
747
748 /*
749  * Dummy child_process returned by git_connect() if the transport protocol
750  * does not need fork(2).
751  */
752 static struct child_process no_fork = CHILD_PROCESS_INIT;
753
754 int git_connection_is_socket(struct child_process *conn)
755 {
756         return conn == &no_fork;
757 }
758
759 static struct child_process *git_tcp_connect(int fd[2], char *host, int flags)
760 {
761         int sockfd = git_tcp_connect_sock(host, flags);
762
763         fd[0] = sockfd;
764         fd[1] = dup(sockfd);
765
766         return &no_fork;
767 }
768
769
770 static char *git_proxy_command;
771
772 static int git_proxy_command_options(const char *var, const char *value,
773                 void *cb)
774 {
775         if (!strcmp(var, "core.gitproxy")) {
776                 const char *for_pos;
777                 int matchlen = -1;
778                 int hostlen;
779                 const char *rhost_name = cb;
780                 int rhost_len = strlen(rhost_name);
781
782                 if (git_proxy_command)
783                         return 0;
784                 if (!value)
785                         return config_error_nonbool(var);
786                 /* [core]
787                  * ;# matches www.kernel.org as well
788                  * gitproxy = netcatter-1 for kernel.org
789                  * gitproxy = netcatter-2 for sample.xz
790                  * gitproxy = netcatter-default
791                  */
792                 for_pos = strstr(value, " for ");
793                 if (!for_pos)
794                         /* matches everybody */
795                         matchlen = strlen(value);
796                 else {
797                         hostlen = strlen(for_pos + 5);
798                         if (rhost_len < hostlen)
799                                 matchlen = -1;
800                         else if (!strncmp(for_pos + 5,
801                                           rhost_name + rhost_len - hostlen,
802                                           hostlen) &&
803                                  ((rhost_len == hostlen) ||
804                                   rhost_name[rhost_len - hostlen -1] == '.'))
805                                 matchlen = for_pos - value;
806                         else
807                                 matchlen = -1;
808                 }
809                 if (0 <= matchlen) {
810                         /* core.gitproxy = none for kernel.org */
811                         if (matchlen == 4 &&
812                             !memcmp(value, "none", 4))
813                                 matchlen = 0;
814                         git_proxy_command = xmemdupz(value, matchlen);
815                 }
816                 return 0;
817         }
818
819         return git_default_config(var, value, cb);
820 }
821
822 static int git_use_proxy(const char *host)
823 {
824         git_proxy_command = getenv("GIT_PROXY_COMMAND");
825         git_config(git_proxy_command_options, (void*)host);
826         return (git_proxy_command && *git_proxy_command);
827 }
828
829 static struct child_process *git_proxy_connect(int fd[2], char *host)
830 {
831         const char *port = STR(DEFAULT_GIT_PORT);
832         struct child_process *proxy;
833
834         get_host_and_port(&host, &port);
835
836         if (looks_like_command_line_option(host))
837                 die("strange hostname '%s' blocked", host);
838         if (looks_like_command_line_option(port))
839                 die("strange port '%s' blocked", port);
840
841         proxy = xmalloc(sizeof(*proxy));
842         child_process_init(proxy);
843         argv_array_push(&proxy->args, git_proxy_command);
844         argv_array_push(&proxy->args, host);
845         argv_array_push(&proxy->args, port);
846         proxy->in = -1;
847         proxy->out = -1;
848         if (start_command(proxy))
849                 die("cannot start proxy %s", git_proxy_command);
850         fd[0] = proxy->out; /* read from proxy stdout */
851         fd[1] = proxy->in;  /* write to proxy stdin */
852         return proxy;
853 }
854
855 static char *get_port(char *host)
856 {
857         char *end;
858         char *p = strchr(host, ':');
859
860         if (p) {
861                 long port = strtol(p + 1, &end, 10);
862                 if (end != p + 1 && *end == '\0' && 0 <= port && port < 65536) {
863                         *p = '\0';
864                         return p+1;
865                 }
866         }
867
868         return NULL;
869 }
870
871 /*
872  * Extract protocol and relevant parts from the specified connection URL.
873  * The caller must free() the returned strings.
874  */
875 static enum protocol parse_connect_url(const char *url_orig, char **ret_host,
876                                        char **ret_path)
877 {
878         char *url;
879         char *host, *path;
880         char *end;
881         int separator = '/';
882         enum protocol protocol = PROTO_LOCAL;
883
884         if (is_url(url_orig))
885                 url = url_decode(url_orig);
886         else
887                 url = xstrdup(url_orig);
888
889         host = strstr(url, "://");
890         if (host) {
891                 *host = '\0';
892                 protocol = get_protocol(url);
893                 host += 3;
894         } else {
895                 host = url;
896                 if (!url_is_local_not_ssh(url)) {
897                         protocol = PROTO_SSH;
898                         separator = ':';
899                 }
900         }
901
902         /*
903          * Don't do destructive transforms as protocol code does
904          * '[]' unwrapping in get_host_and_port()
905          */
906         end = host_end(&host, 0);
907
908         if (protocol == PROTO_LOCAL)
909                 path = end;
910         else if (protocol == PROTO_FILE && has_dos_drive_prefix(end))
911                 path = end; /* "file://$(pwd)" may be "file://C:/projects/repo" */
912         else
913                 path = strchr(end, separator);
914
915         if (!path || !*path)
916                 die("No path specified. See 'man git-pull' for valid url syntax");
917
918         /*
919          * null-terminate hostname and point path to ~ for URL's like this:
920          *    ssh://host.xz/~user/repo
921          */
922
923         end = path; /* Need to \0 terminate host here */
924         if (separator == ':')
925                 path++; /* path starts after ':' */
926         if (protocol == PROTO_GIT || protocol == PROTO_SSH) {
927                 if (path[1] == '~')
928                         path++;
929         }
930
931         path = xstrdup(path);
932         *end = '\0';
933
934         *ret_host = xstrdup(host);
935         *ret_path = path;
936         free(url);
937         return protocol;
938 }
939
940 static const char *get_ssh_command(void)
941 {
942         const char *ssh;
943
944         if ((ssh = getenv("GIT_SSH_COMMAND")))
945                 return ssh;
946
947         if (!git_config_get_string_const("core.sshcommand", &ssh))
948                 return ssh;
949
950         return NULL;
951 }
952
953 enum ssh_variant {
954         VARIANT_AUTO,
955         VARIANT_SIMPLE,
956         VARIANT_SSH,
957         VARIANT_PLINK,
958         VARIANT_PUTTY,
959         VARIANT_TORTOISEPLINK,
960 };
961
962 static void override_ssh_variant(enum ssh_variant *ssh_variant)
963 {
964         const char *variant = getenv("GIT_SSH_VARIANT");
965
966         if (!variant && git_config_get_string_const("ssh.variant", &variant))
967                 return;
968
969         if (!strcmp(variant, "auto"))
970                 *ssh_variant = VARIANT_AUTO;
971         else if (!strcmp(variant, "plink"))
972                 *ssh_variant = VARIANT_PLINK;
973         else if (!strcmp(variant, "putty"))
974                 *ssh_variant = VARIANT_PUTTY;
975         else if (!strcmp(variant, "tortoiseplink"))
976                 *ssh_variant = VARIANT_TORTOISEPLINK;
977         else if (!strcmp(variant, "simple"))
978                 *ssh_variant = VARIANT_SIMPLE;
979         else
980                 *ssh_variant = VARIANT_SSH;
981 }
982
983 static enum ssh_variant determine_ssh_variant(const char *ssh_command,
984                                               int is_cmdline)
985 {
986         enum ssh_variant ssh_variant = VARIANT_AUTO;
987         const char *variant;
988         char *p = NULL;
989
990         override_ssh_variant(&ssh_variant);
991
992         if (ssh_variant != VARIANT_AUTO)
993                 return ssh_variant;
994
995         if (!is_cmdline) {
996                 p = xstrdup(ssh_command);
997                 variant = basename(p);
998         } else {
999                 const char **ssh_argv;
1000
1001                 p = xstrdup(ssh_command);
1002                 if (split_cmdline(p, &ssh_argv) > 0) {
1003                         variant = basename((char *)ssh_argv[0]);
1004                         /*
1005                          * At this point, variant points into the buffer
1006                          * referenced by p, hence we do not need ssh_argv
1007                          * any longer.
1008                          */
1009                         free(ssh_argv);
1010                 } else {
1011                         free(p);
1012                         return ssh_variant;
1013                 }
1014         }
1015
1016         if (!strcasecmp(variant, "ssh") ||
1017             !strcasecmp(variant, "ssh.exe"))
1018                 ssh_variant = VARIANT_SSH;
1019         else if (!strcasecmp(variant, "plink") ||
1020                  !strcasecmp(variant, "plink.exe"))
1021                 ssh_variant = VARIANT_PLINK;
1022         else if (!strcasecmp(variant, "tortoiseplink") ||
1023                  !strcasecmp(variant, "tortoiseplink.exe"))
1024                 ssh_variant = VARIANT_TORTOISEPLINK;
1025
1026         free(p);
1027         return ssh_variant;
1028 }
1029
1030 /*
1031  * Open a connection using Git's native protocol.
1032  *
1033  * The caller is responsible for freeing hostandport, but this function may
1034  * modify it (for example, to truncate it to remove the port part).
1035  */
1036 static struct child_process *git_connect_git(int fd[2], char *hostandport,
1037                                              const char *path, const char *prog,
1038                                              enum protocol_version version,
1039                                              int flags)
1040 {
1041         struct child_process *conn;
1042         struct strbuf request = STRBUF_INIT;
1043         /*
1044          * Set up virtual host information based on where we will
1045          * connect, unless the user has overridden us in
1046          * the environment.
1047          */
1048         char *target_host = getenv("GIT_OVERRIDE_VIRTUAL_HOST");
1049         if (target_host)
1050                 target_host = xstrdup(target_host);
1051         else
1052                 target_host = xstrdup(hostandport);
1053
1054         transport_check_allowed("git");
1055
1056         /*
1057          * These underlying connection commands die() if they
1058          * cannot connect.
1059          */
1060         if (git_use_proxy(hostandport))
1061                 conn = git_proxy_connect(fd, hostandport);
1062         else
1063                 conn = git_tcp_connect(fd, hostandport, flags);
1064         /*
1065          * Separate original protocol components prog and path
1066          * from extended host header with a NUL byte.
1067          *
1068          * Note: Do not add any other headers here!  Doing so
1069          * will cause older git-daemon servers to crash.
1070          */
1071         strbuf_addf(&request,
1072                     "%s %s%chost=%s%c",
1073                     prog, path, 0,
1074                     target_host, 0);
1075
1076         /* If using a new version put that stuff here after a second null byte */
1077         if (version > 0) {
1078                 strbuf_addch(&request, '\0');
1079                 strbuf_addf(&request, "version=%d%c",
1080                             version, '\0');
1081         }
1082
1083         packet_write(fd[1], request.buf, request.len);
1084
1085         free(target_host);
1086         strbuf_release(&request);
1087         return conn;
1088 }
1089
1090 /*
1091  * Append the appropriate environment variables to `env` and options to
1092  * `args` for running ssh in Git's SSH-tunneled transport.
1093  */
1094 static void push_ssh_options(struct argv_array *args, struct argv_array *env,
1095                              enum ssh_variant variant, const char *port,
1096                              enum protocol_version version, int flags)
1097 {
1098         if (variant == VARIANT_SSH &&
1099             version > 0) {
1100                 argv_array_push(args, "-o");
1101                 argv_array_push(args, "SendEnv=" GIT_PROTOCOL_ENVIRONMENT);
1102                 argv_array_pushf(env, GIT_PROTOCOL_ENVIRONMENT "=version=%d",
1103                                  version);
1104         }
1105
1106         if (flags & CONNECT_IPV4) {
1107                 switch (variant) {
1108                 case VARIANT_AUTO:
1109                         BUG("VARIANT_AUTO passed to push_ssh_options");
1110                 case VARIANT_SIMPLE:
1111                         die("ssh variant 'simple' does not support -4");
1112                 case VARIANT_SSH:
1113                 case VARIANT_PLINK:
1114                 case VARIANT_PUTTY:
1115                 case VARIANT_TORTOISEPLINK:
1116                         argv_array_push(args, "-4");
1117                 }
1118         } else if (flags & CONNECT_IPV6) {
1119                 switch (variant) {
1120                 case VARIANT_AUTO:
1121                         BUG("VARIANT_AUTO passed to push_ssh_options");
1122                 case VARIANT_SIMPLE:
1123                         die("ssh variant 'simple' does not support -6");
1124                 case VARIANT_SSH:
1125                 case VARIANT_PLINK:
1126                 case VARIANT_PUTTY:
1127                 case VARIANT_TORTOISEPLINK:
1128                         argv_array_push(args, "-6");
1129                 }
1130         }
1131
1132         if (variant == VARIANT_TORTOISEPLINK)
1133                 argv_array_push(args, "-batch");
1134
1135         if (port) {
1136                 switch (variant) {
1137                 case VARIANT_AUTO:
1138                         BUG("VARIANT_AUTO passed to push_ssh_options");
1139                 case VARIANT_SIMPLE:
1140                         die("ssh variant 'simple' does not support setting port");
1141                 case VARIANT_SSH:
1142                         argv_array_push(args, "-p");
1143                         break;
1144                 case VARIANT_PLINK:
1145                 case VARIANT_PUTTY:
1146                 case VARIANT_TORTOISEPLINK:
1147                         argv_array_push(args, "-P");
1148                 }
1149
1150                 argv_array_push(args, port);
1151         }
1152 }
1153
1154 /* Prepare a child_process for use by Git's SSH-tunneled transport. */
1155 static void fill_ssh_args(struct child_process *conn, const char *ssh_host,
1156                           const char *port, enum protocol_version version,
1157                           int flags)
1158 {
1159         const char *ssh;
1160         enum ssh_variant variant;
1161
1162         if (looks_like_command_line_option(ssh_host))
1163                 die("strange hostname '%s' blocked", ssh_host);
1164
1165         ssh = get_ssh_command();
1166         if (ssh) {
1167                 variant = determine_ssh_variant(ssh, 1);
1168         } else {
1169                 /*
1170                  * GIT_SSH is the no-shell version of
1171                  * GIT_SSH_COMMAND (and must remain so for
1172                  * historical compatibility).
1173                  */
1174                 conn->use_shell = 0;
1175
1176                 ssh = getenv("GIT_SSH");
1177                 if (!ssh)
1178                         ssh = "ssh";
1179                 variant = determine_ssh_variant(ssh, 0);
1180         }
1181
1182         if (variant == VARIANT_AUTO) {
1183                 struct child_process detect = CHILD_PROCESS_INIT;
1184
1185                 detect.use_shell = conn->use_shell;
1186                 detect.no_stdin = detect.no_stdout = detect.no_stderr = 1;
1187
1188                 argv_array_push(&detect.args, ssh);
1189                 argv_array_push(&detect.args, "-G");
1190                 push_ssh_options(&detect.args, &detect.env_array,
1191                                  VARIANT_SSH, port, version, flags);
1192                 argv_array_push(&detect.args, ssh_host);
1193
1194                 variant = run_command(&detect) ? VARIANT_SIMPLE : VARIANT_SSH;
1195         }
1196
1197         argv_array_push(&conn->args, ssh);
1198         push_ssh_options(&conn->args, &conn->env_array, variant, port, version, flags);
1199         argv_array_push(&conn->args, ssh_host);
1200 }
1201
1202 /*
1203  * This returns the dummy child_process `no_fork` if the transport protocol
1204  * does not need fork(2), or a struct child_process object if it does.  Once
1205  * done, finish the connection with finish_connect() with the value returned
1206  * from this function (it is safe to call finish_connect() with NULL to
1207  * support the former case).
1208  *
1209  * If it returns, the connect is successful; it just dies on errors (this
1210  * will hopefully be changed in a libification effort, to return NULL when
1211  * the connection failed).
1212  */
1213 struct child_process *git_connect(int fd[2], const char *url,
1214                                   const char *prog, int flags)
1215 {
1216         char *hostandport, *path;
1217         struct child_process *conn;
1218         enum protocol protocol;
1219         enum protocol_version version = get_protocol_version_config();
1220
1221         /*
1222          * NEEDSWORK: If we are trying to use protocol v2 and we are planning
1223          * to perform a push, then fallback to v0 since the client doesn't know
1224          * how to push yet using v2.
1225          */
1226         if (version == protocol_v2 && !strcmp("git-receive-pack", prog))
1227                 version = protocol_v0;
1228
1229         /* Without this we cannot rely on waitpid() to tell
1230          * what happened to our children.
1231          */
1232         signal(SIGCHLD, SIG_DFL);
1233
1234         protocol = parse_connect_url(url, &hostandport, &path);
1235         if ((flags & CONNECT_DIAG_URL) && (protocol != PROTO_SSH)) {
1236                 printf("Diag: url=%s\n", url ? url : "NULL");
1237                 printf("Diag: protocol=%s\n", prot_name(protocol));
1238                 printf("Diag: hostandport=%s\n", hostandport ? hostandport : "NULL");
1239                 printf("Diag: path=%s\n", path ? path : "NULL");
1240                 conn = NULL;
1241         } else if (protocol == PROTO_GIT) {
1242                 conn = git_connect_git(fd, hostandport, path, prog, version, flags);
1243         } else {
1244                 struct strbuf cmd = STRBUF_INIT;
1245                 const char *const *var;
1246
1247                 conn = xmalloc(sizeof(*conn));
1248                 child_process_init(conn);
1249
1250                 if (looks_like_command_line_option(path))
1251                         die("strange pathname '%s' blocked", path);
1252
1253                 strbuf_addstr(&cmd, prog);
1254                 strbuf_addch(&cmd, ' ');
1255                 sq_quote_buf(&cmd, path);
1256
1257                 /* remove repo-local variables from the environment */
1258                 for (var = local_repo_env; *var; var++)
1259                         argv_array_push(&conn->env_array, *var);
1260
1261                 conn->use_shell = 1;
1262                 conn->in = conn->out = -1;
1263                 if (protocol == PROTO_SSH) {
1264                         char *ssh_host = hostandport;
1265                         const char *port = NULL;
1266                         transport_check_allowed("ssh");
1267                         get_host_and_port(&ssh_host, &port);
1268
1269                         if (!port)
1270                                 port = get_port(ssh_host);
1271
1272                         if (flags & CONNECT_DIAG_URL) {
1273                                 printf("Diag: url=%s\n", url ? url : "NULL");
1274                                 printf("Diag: protocol=%s\n", prot_name(protocol));
1275                                 printf("Diag: userandhost=%s\n", ssh_host ? ssh_host : "NULL");
1276                                 printf("Diag: port=%s\n", port ? port : "NONE");
1277                                 printf("Diag: path=%s\n", path ? path : "NULL");
1278
1279                                 free(hostandport);
1280                                 free(path);
1281                                 free(conn);
1282                                 strbuf_release(&cmd);
1283                                 return NULL;
1284                         }
1285                         fill_ssh_args(conn, ssh_host, port, version, flags);
1286                 } else {
1287                         transport_check_allowed("file");
1288                         if (version > 0) {
1289                                 argv_array_pushf(&conn->env_array, GIT_PROTOCOL_ENVIRONMENT "=version=%d",
1290                                                  version);
1291                         }
1292                 }
1293                 argv_array_push(&conn->args, cmd.buf);
1294
1295                 if (start_command(conn))
1296                         die("unable to fork");
1297
1298                 fd[0] = conn->out; /* read from child's stdout */
1299                 fd[1] = conn->in;  /* write to child's stdin */
1300                 strbuf_release(&cmd);
1301         }
1302         free(hostandport);
1303         free(path);
1304         return conn;
1305 }
1306
1307 int finish_connect(struct child_process *conn)
1308 {
1309         int code;
1310         if (!conn || git_connection_is_socket(conn))
1311                 return 0;
1312
1313         code = finish_command(conn);
1314         free(conn);
1315         return code;
1316 }