Merge branch 'cb/byte-order'
[git/git.git] / connect.c
1 #include "git-compat-util.h"
2 #include "cache.h"
3 #include "pkt-line.h"
4 #include "quote.h"
5 #include "refs.h"
6 #include "run-command.h"
7 #include "remote.h"
8 #include "connect.h"
9 #include "url.h"
10 #include "string-list.h"
11 #include "sha1-array.h"
12
13 static char *server_capabilities;
14 static const char *parse_feature_value(const char *, const char *, int *);
15
16 static int check_ref(const char *name, int len, unsigned int flags)
17 {
18 if (!flags)
19 return 1;
20
21 if (len < 5 || memcmp(name, "refs/", 5))
22 return 0;
23
24 /* Skip the "refs/" part */
25 name += 5;
26 len -= 5;
27
28 /* REF_NORMAL means that we don't want the magic fake tag refs */
29 if ((flags & REF_NORMAL) && check_refname_format(name, 0))
30 return 0;
31
32 /* REF_HEADS means that we want regular branch heads */
33 if ((flags & REF_HEADS) && !memcmp(name, "heads/", 6))
34 return 1;
35
36 /* REF_TAGS means that we want tags */
37 if ((flags & REF_TAGS) && !memcmp(name, "tags/", 5))
38 return 1;
39
40 /* All type bits clear means that we are ok with anything */
41 return !(flags & ~REF_NORMAL);
42 }
43
44 int check_ref_type(const struct ref *ref, int flags)
45 {
46 return check_ref(ref->name, strlen(ref->name), flags);
47 }
48
49 static void die_initial_contact(int got_at_least_one_head)
50 {
51 if (got_at_least_one_head)
52 die("The remote end hung up upon initial contact");
53 else
54 die("Could not read from remote repository.\n\n"
55 "Please make sure you have the correct access rights\n"
56 "and the repository exists.");
57 }
58
59 static void parse_one_symref_info(struct string_list *symref, const char *val, int len)
60 {
61 char *sym, *target;
62 struct string_list_item *item;
63
64 if (!len)
65 return; /* just "symref" */
66 /* e.g. "symref=HEAD:refs/heads/master" */
67 sym = xmalloc(len + 1);
68 memcpy(sym, val, len);
69 sym[len] = '\0';
70 target = strchr(sym, ':');
71 if (!target)
72 /* just "symref=something" */
73 goto reject;
74 *(target++) = '\0';
75 if (check_refname_format(sym, REFNAME_ALLOW_ONELEVEL) ||
76 check_refname_format(target, REFNAME_ALLOW_ONELEVEL))
77 /* "symref=bogus:pair */
78 goto reject;
79 item = string_list_append(symref, sym);
80 item->util = target;
81 return;
82 reject:
83 free(sym);
84 return;
85 }
86
87 static void annotate_refs_with_symref_info(struct ref *ref)
88 {
89 struct string_list symref = STRING_LIST_INIT_DUP;
90 const char *feature_list = server_capabilities;
91
92 while (feature_list) {
93 int len;
94 const char *val;
95
96 val = parse_feature_value(feature_list, "symref", &len);
97 if (!val)
98 break;
99 parse_one_symref_info(&symref, val, len);
100 feature_list = val + 1;
101 }
102 sort_string_list(&symref);
103
104 for (; ref; ref = ref->next) {
105 struct string_list_item *item;
106 item = string_list_lookup(&symref, ref->name);
107 if (!item)
108 continue;
109 ref->symref = xstrdup((char *)item->util);
110 }
111 string_list_clear(&symref, 0);
112 }
113
114 /*
115 * Read all the refs from the other end
116 */
117 struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
118 struct ref **list, unsigned int flags,
119 struct sha1_array *extra_have,
120 struct sha1_array *shallow_points)
121 {
122 struct ref **orig_list = list;
123 int got_at_least_one_head = 0;
124
125 *list = NULL;
126 for (;;) {
127 struct ref *ref;
128 unsigned char old_sha1[20];
129 char *name;
130 int len, name_len;
131 char *buffer = packet_buffer;
132
133 len = packet_read(in, &src_buf, &src_len,
134 packet_buffer, sizeof(packet_buffer),
135 PACKET_READ_GENTLE_ON_EOF |
136 PACKET_READ_CHOMP_NEWLINE);
137 if (len < 0)
138 die_initial_contact(got_at_least_one_head);
139
140 if (!len)
141 break;
142
143 if (len > 4 && starts_with(buffer, "ERR "))
144 die("remote error: %s", buffer + 4);
145
146 if (len == 48 && starts_with(buffer, "shallow ")) {
147 if (get_sha1_hex(buffer + 8, old_sha1))
148 die("protocol error: expected shallow sha-1, got '%s'", buffer + 8);
149 if (!shallow_points)
150 die("repository on the other end cannot be shallow");
151 sha1_array_append(shallow_points, old_sha1);
152 continue;
153 }
154
155 if (len < 42 || get_sha1_hex(buffer, old_sha1) || buffer[40] != ' ')
156 die("protocol error: expected sha/ref, got '%s'", buffer);
157 name = buffer + 41;
158
159 name_len = strlen(name);
160 if (len != name_len + 41) {
161 free(server_capabilities);
162 server_capabilities = xstrdup(name + name_len + 1);
163 }
164
165 if (extra_have &&
166 name_len == 5 && !memcmp(".have", name, 5)) {
167 sha1_array_append(extra_have, old_sha1);
168 continue;
169 }
170
171 if (!check_ref(name, name_len, flags))
172 continue;
173 ref = alloc_ref(buffer + 41);
174 hashcpy(ref->old_sha1, old_sha1);
175 *list = ref;
176 list = &ref->next;
177 got_at_least_one_head = 1;
178 }
179
180 annotate_refs_with_symref_info(*orig_list);
181
182 return list;
183 }
184
185 static const char *parse_feature_value(const char *feature_list, const char *feature, int *lenp)
186 {
187 int len;
188
189 if (!feature_list)
190 return NULL;
191
192 len = strlen(feature);
193 while (*feature_list) {
194 const char *found = strstr(feature_list, feature);
195 if (!found)
196 return NULL;
197 if (feature_list == found || isspace(found[-1])) {
198 const char *value = found + len;
199 /* feature with no value (e.g., "thin-pack") */
200 if (!*value || isspace(*value)) {
201 if (lenp)
202 *lenp = 0;
203 return value;
204 }
205 /* feature with a value (e.g., "agent=git/1.2.3") */
206 else if (*value == '=') {
207 value++;
208 if (lenp)
209 *lenp = strcspn(value, " \t\n");
210 return value;
211 }
212 /*
213 * otherwise we matched a substring of another feature;
214 * keep looking
215 */
216 }
217 feature_list = found + 1;
218 }
219 return NULL;
220 }
221
222 int parse_feature_request(const char *feature_list, const char *feature)
223 {
224 return !!parse_feature_value(feature_list, feature, NULL);
225 }
226
227 const char *server_feature_value(const char *feature, int *len)
228 {
229 return parse_feature_value(server_capabilities, feature, len);
230 }
231
232 int server_supports(const char *feature)
233 {
234 return !!server_feature_value(feature, NULL);
235 }
236
237 enum protocol {
238 PROTO_LOCAL = 1,
239 PROTO_FILE,
240 PROTO_SSH,
241 PROTO_GIT
242 };
243
244 int url_is_local_not_ssh(const char *url)
245 {
246 const char *colon = strchr(url, ':');
247 const char *slash = strchr(url, '/');
248 return !colon || (slash && slash < colon) ||
249 has_dos_drive_prefix(url);
250 }
251
252 static const char *prot_name(enum protocol protocol)
253 {
254 switch (protocol) {
255 case PROTO_LOCAL:
256 case PROTO_FILE:
257 return "file";
258 case PROTO_SSH:
259 return "ssh";
260 case PROTO_GIT:
261 return "git";
262 default:
263 return "unkown protocol";
264 }
265 }
266
267 static enum protocol get_protocol(const char *name)
268 {
269 if (!strcmp(name, "ssh"))
270 return PROTO_SSH;
271 if (!strcmp(name, "git"))
272 return PROTO_GIT;
273 if (!strcmp(name, "git+ssh"))
274 return PROTO_SSH;
275 if (!strcmp(name, "ssh+git"))
276 return PROTO_SSH;
277 if (!strcmp(name, "file"))
278 return PROTO_FILE;
279 die("I don't handle protocol '%s'", name);
280 }
281
282 #define STR_(s) # s
283 #define STR(s) STR_(s)
284
285 static void get_host_and_port(char **host, const char **port)
286 {
287 char *colon, *end;
288
289 if (*host[0] == '[') {
290 end = strchr(*host + 1, ']');
291 if (end) {
292 *end = 0;
293 end++;
294 (*host)++;
295 } else
296 end = *host;
297 } else
298 end = *host;
299 colon = strchr(end, ':');
300
301 if (colon) {
302 *colon = 0;
303 *port = colon + 1;
304 }
305 }
306
307 static void enable_keepalive(int sockfd)
308 {
309 int ka = 1;
310
311 if (setsockopt(sockfd, SOL_SOCKET, SO_KEEPALIVE, &ka, sizeof(ka)) < 0)
312 fprintf(stderr, "unable to set SO_KEEPALIVE on socket: %s\n",
313 strerror(errno));
314 }
315
316 #ifndef NO_IPV6
317
318 static const char *ai_name(const struct addrinfo *ai)
319 {
320 static char addr[NI_MAXHOST];
321 if (getnameinfo(ai->ai_addr, ai->ai_addrlen, addr, sizeof(addr), NULL, 0,
322 NI_NUMERICHOST) != 0)
323 strcpy(addr, "(unknown)");
324
325 return addr;
326 }
327
328 /*
329 * Returns a connected socket() fd, or else die()s.
330 */
331 static int git_tcp_connect_sock(char *host, int flags)
332 {
333 struct strbuf error_message = STRBUF_INIT;
334 int sockfd = -1;
335 const char *port = STR(DEFAULT_GIT_PORT);
336 struct addrinfo hints, *ai0, *ai;
337 int gai;
338 int cnt = 0;
339
340 get_host_and_port(&host, &port);
341 if (!*port)
342 port = "<none>";
343
344 memset(&hints, 0, sizeof(hints));
345 hints.ai_socktype = SOCK_STREAM;
346 hints.ai_protocol = IPPROTO_TCP;
347
348 if (flags & CONNECT_VERBOSE)
349 fprintf(stderr, "Looking up %s ... ", host);
350
351 gai = getaddrinfo(host, port, &hints, &ai);
352 if (gai)
353 die("Unable to look up %s (port %s) (%s)", host, port, gai_strerror(gai));
354
355 if (flags & CONNECT_VERBOSE)
356 fprintf(stderr, "done.\nConnecting to %s (port %s) ... ", host, port);
357
358 for (ai0 = ai; ai; ai = ai->ai_next, cnt++) {
359 sockfd = socket(ai->ai_family,
360 ai->ai_socktype, ai->ai_protocol);
361 if ((sockfd < 0) ||
362 (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0)) {
363 strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
364 host, cnt, ai_name(ai), strerror(errno));
365 if (0 <= sockfd)
366 close(sockfd);
367 sockfd = -1;
368 continue;
369 }
370 if (flags & CONNECT_VERBOSE)
371 fprintf(stderr, "%s ", ai_name(ai));
372 break;
373 }
374
375 freeaddrinfo(ai0);
376
377 if (sockfd < 0)
378 die("unable to connect to %s:\n%s", host, error_message.buf);
379
380 enable_keepalive(sockfd);
381
382 if (flags & CONNECT_VERBOSE)
383 fprintf(stderr, "done.\n");
384
385 strbuf_release(&error_message);
386
387 return sockfd;
388 }
389
390 #else /* NO_IPV6 */
391
392 /*
393 * Returns a connected socket() fd, or else die()s.
394 */
395 static int git_tcp_connect_sock(char *host, int flags)
396 {
397 struct strbuf error_message = STRBUF_INIT;
398 int sockfd = -1;
399 const char *port = STR(DEFAULT_GIT_PORT);
400 char *ep;
401 struct hostent *he;
402 struct sockaddr_in sa;
403 char **ap;
404 unsigned int nport;
405 int cnt;
406
407 get_host_and_port(&host, &port);
408
409 if (flags & CONNECT_VERBOSE)
410 fprintf(stderr, "Looking up %s ... ", host);
411
412 he = gethostbyname(host);
413 if (!he)
414 die("Unable to look up %s (%s)", host, hstrerror(h_errno));
415 nport = strtoul(port, &ep, 10);
416 if ( ep == port || *ep ) {
417 /* Not numeric */
418 struct servent *se = getservbyname(port,"tcp");
419 if ( !se )
420 die("Unknown port %s", port);
421 nport = se->s_port;
422 }
423
424 if (flags & CONNECT_VERBOSE)
425 fprintf(stderr, "done.\nConnecting to %s (port %s) ... ", host, port);
426
427 for (cnt = 0, ap = he->h_addr_list; *ap; ap++, cnt++) {
428 memset(&sa, 0, sizeof sa);
429 sa.sin_family = he->h_addrtype;
430 sa.sin_port = htons(nport);
431 memcpy(&sa.sin_addr, *ap, he->h_length);
432
433 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
434 if ((sockfd < 0) ||
435 connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
436 strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
437 host,
438 cnt,
439 inet_ntoa(*(struct in_addr *)&sa.sin_addr),
440 strerror(errno));
441 if (0 <= sockfd)
442 close(sockfd);
443 sockfd = -1;
444 continue;
445 }
446 if (flags & CONNECT_VERBOSE)
447 fprintf(stderr, "%s ",
448 inet_ntoa(*(struct in_addr *)&sa.sin_addr));
449 break;
450 }
451
452 if (sockfd < 0)
453 die("unable to connect to %s:\n%s", host, error_message.buf);
454
455 enable_keepalive(sockfd);
456
457 if (flags & CONNECT_VERBOSE)
458 fprintf(stderr, "done.\n");
459
460 return sockfd;
461 }
462
463 #endif /* NO_IPV6 */
464
465
466 static void git_tcp_connect(int fd[2], char *host, int flags)
467 {
468 int sockfd = git_tcp_connect_sock(host, flags);
469
470 fd[0] = sockfd;
471 fd[1] = dup(sockfd);
472 }
473
474
475 static char *git_proxy_command;
476
477 static int git_proxy_command_options(const char *var, const char *value,
478 void *cb)
479 {
480 if (!strcmp(var, "core.gitproxy")) {
481 const char *for_pos;
482 int matchlen = -1;
483 int hostlen;
484 const char *rhost_name = cb;
485 int rhost_len = strlen(rhost_name);
486
487 if (git_proxy_command)
488 return 0;
489 if (!value)
490 return config_error_nonbool(var);
491 /* [core]
492 * ;# matches www.kernel.org as well
493 * gitproxy = netcatter-1 for kernel.org
494 * gitproxy = netcatter-2 for sample.xz
495 * gitproxy = netcatter-default
496 */
497 for_pos = strstr(value, " for ");
498 if (!for_pos)
499 /* matches everybody */
500 matchlen = strlen(value);
501 else {
502 hostlen = strlen(for_pos + 5);
503 if (rhost_len < hostlen)
504 matchlen = -1;
505 else if (!strncmp(for_pos + 5,
506 rhost_name + rhost_len - hostlen,
507 hostlen) &&
508 ((rhost_len == hostlen) ||
509 rhost_name[rhost_len - hostlen -1] == '.'))
510 matchlen = for_pos - value;
511 else
512 matchlen = -1;
513 }
514 if (0 <= matchlen) {
515 /* core.gitproxy = none for kernel.org */
516 if (matchlen == 4 &&
517 !memcmp(value, "none", 4))
518 matchlen = 0;
519 git_proxy_command = xmemdupz(value, matchlen);
520 }
521 return 0;
522 }
523
524 return git_default_config(var, value, cb);
525 }
526
527 static int git_use_proxy(const char *host)
528 {
529 git_proxy_command = getenv("GIT_PROXY_COMMAND");
530 git_config(git_proxy_command_options, (void*)host);
531 return (git_proxy_command && *git_proxy_command);
532 }
533
534 static struct child_process *git_proxy_connect(int fd[2], char *host)
535 {
536 const char *port = STR(DEFAULT_GIT_PORT);
537 struct child_process *proxy;
538
539 get_host_and_port(&host, &port);
540
541 proxy = xcalloc(1, sizeof(*proxy));
542 argv_array_push(&proxy->args, git_proxy_command);
543 argv_array_push(&proxy->args, host);
544 argv_array_push(&proxy->args, port);
545 proxy->in = -1;
546 proxy->out = -1;
547 if (start_command(proxy))
548 die("cannot start proxy %s", git_proxy_command);
549 fd[0] = proxy->out; /* read from proxy stdout */
550 fd[1] = proxy->in; /* write to proxy stdin */
551 return proxy;
552 }
553
554 static const char *get_port_numeric(const char *p)
555 {
556 char *end;
557 if (p) {
558 long port = strtol(p + 1, &end, 10);
559 if (end != p + 1 && *end == '\0' && 0 <= port && port < 65536) {
560 return p;
561 }
562 }
563
564 return NULL;
565 }
566
567 /*
568 * Extract protocol and relevant parts from the specified connection URL.
569 * The caller must free() the returned strings.
570 */
571 static enum protocol parse_connect_url(const char *url_orig, char **ret_host,
572 char **ret_path)
573 {
574 char *url;
575 char *host, *path;
576 char *end;
577 int separator = '/';
578 enum protocol protocol = PROTO_LOCAL;
579
580 if (is_url(url_orig))
581 url = url_decode(url_orig);
582 else
583 url = xstrdup(url_orig);
584
585 host = strstr(url, "://");
586 if (host) {
587 *host = '\0';
588 protocol = get_protocol(url);
589 host += 3;
590 } else {
591 host = url;
592 if (!url_is_local_not_ssh(url)) {
593 protocol = PROTO_SSH;
594 separator = ':';
595 }
596 }
597
598 /*
599 * Don't do destructive transforms as protocol code does
600 * '[]' unwrapping in get_host_and_port()
601 */
602 if (host[0] == '[') {
603 end = strchr(host + 1, ']');
604 if (end) {
605 end++;
606 } else
607 end = host;
608 } else
609 end = host;
610
611 if (protocol == PROTO_LOCAL)
612 path = end;
613 else if (protocol == PROTO_FILE && has_dos_drive_prefix(end))
614 path = end; /* "file://$(pwd)" may be "file://C:/projects/repo" */
615 else
616 path = strchr(end, separator);
617
618 if (!path || !*path)
619 die("No path specified. See 'man git-pull' for valid url syntax");
620
621 /*
622 * null-terminate hostname and point path to ~ for URL's like this:
623 * ssh://host.xz/~user/repo
624 */
625
626 end = path; /* Need to \0 terminate host here */
627 if (separator == ':')
628 path++; /* path starts after ':' */
629 if (protocol == PROTO_GIT || protocol == PROTO_SSH) {
630 if (path[1] == '~')
631 path++;
632 }
633
634 path = xstrdup(path);
635 *end = '\0';
636
637 *ret_host = xstrdup(host);
638 *ret_path = path;
639 free(url);
640 return protocol;
641 }
642
643 static struct child_process no_fork;
644
645 /*
646 * This returns a dummy child_process if the transport protocol does not
647 * need fork(2), or a struct child_process object if it does. Once done,
648 * finish the connection with finish_connect() with the value returned from
649 * this function (it is safe to call finish_connect() with NULL to support
650 * the former case).
651 *
652 * If it returns, the connect is successful; it just dies on errors (this
653 * will hopefully be changed in a libification effort, to return NULL when
654 * the connection failed).
655 */
656 struct child_process *git_connect(int fd[2], const char *url,
657 const char *prog, int flags)
658 {
659 char *hostandport, *path;
660 struct child_process *conn = &no_fork;
661 enum protocol protocol;
662 struct strbuf cmd = STRBUF_INIT;
663
664 /* Without this we cannot rely on waitpid() to tell
665 * what happened to our children.
666 */
667 signal(SIGCHLD, SIG_DFL);
668
669 protocol = parse_connect_url(url, &hostandport, &path);
670 if (flags & CONNECT_DIAG_URL) {
671 printf("Diag: url=%s\n", url ? url : "NULL");
672 printf("Diag: protocol=%s\n", prot_name(protocol));
673 printf("Diag: hostandport=%s\n", hostandport ? hostandport : "NULL");
674 printf("Diag: path=%s\n", path ? path : "NULL");
675 conn = NULL;
676 } else if (protocol == PROTO_GIT) {
677 /* These underlying connection commands die() if they
678 * cannot connect.
679 */
680 char *target_host = xstrdup(hostandport);
681 if (git_use_proxy(hostandport))
682 conn = git_proxy_connect(fd, hostandport);
683 else
684 git_tcp_connect(fd, hostandport, flags);
685 /*
686 * Separate original protocol components prog and path
687 * from extended host header with a NUL byte.
688 *
689 * Note: Do not add any other headers here! Doing so
690 * will cause older git-daemon servers to crash.
691 */
692 packet_write(fd[1],
693 "%s %s%chost=%s%c",
694 prog, path, 0,
695 target_host, 0);
696 free(target_host);
697 } else {
698 conn = xcalloc(1, sizeof(*conn));
699
700 strbuf_addstr(&cmd, prog);
701 strbuf_addch(&cmd, ' ');
702 sq_quote_buf(&cmd, path);
703
704 conn->in = conn->out = -1;
705 if (protocol == PROTO_SSH) {
706 const char *ssh = getenv("GIT_SSH");
707 int putty = ssh && strcasestr(ssh, "plink");
708 char *ssh_host = hostandport;
709 const char *port = NULL;
710 get_host_and_port(&ssh_host, &port);
711 port = get_port_numeric(port);
712
713 if (!ssh) ssh = "ssh";
714
715 argv_array_push(&conn->args, ssh);
716 if (putty && !strcasestr(ssh, "tortoiseplink"))
717 argv_array_push(&conn->args, "-batch");
718 if (port) {
719 /* P is for PuTTY, p is for OpenSSH */
720 argv_array_push(&conn->args, putty ? "-P" : "-p");
721 argv_array_push(&conn->args, port);
722 }
723 argv_array_push(&conn->args, ssh_host);
724 } else {
725 /* remove repo-local variables from the environment */
726 conn->env = local_repo_env;
727 conn->use_shell = 1;
728 }
729 argv_array_push(&conn->args, cmd.buf);
730
731 if (start_command(conn))
732 die("unable to fork");
733
734 fd[0] = conn->out; /* read from child's stdout */
735 fd[1] = conn->in; /* write to child's stdin */
736 strbuf_release(&cmd);
737 }
738 free(hostandport);
739 free(path);
740 return conn;
741 }
742
743 int git_connection_is_socket(struct child_process *conn)
744 {
745 return conn == &no_fork;
746 }
747
748 int finish_connect(struct child_process *conn)
749 {
750 int code;
751 if (!conn || git_connection_is_socket(conn))
752 return 0;
753
754 code = finish_command(conn);
755 free(conn);
756 return code;
757 }