Merge branch 'jc/numstat'
[git.git] / connect.c
blobb9666cc0d826890b5e84a58c7f7ed52ff56c3e79
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 <sys/wait.h>
7 #include <sys/socket.h>
8 #include <netinet/in.h>
9 #include <arpa/inet.h>
10 #include <netdb.h>
11 #include <signal.h>
13 static char *server_capabilities;
15 static int check_ref(const char *name, int len, unsigned int flags)
17 if (!flags)
18 return 1;
20 if (len < 5 || memcmp(name, "refs/", 5))
21 return 0;
23 /* Skip the "refs/" part */
24 name += 5;
25 len -= 5;
27 /* REF_NORMAL means that we don't want the magic fake tag refs */
28 if ((flags & REF_NORMAL) && check_ref_format(name) < 0)
29 return 0;
31 /* REF_HEADS means that we want regular branch heads */
32 if ((flags & REF_HEADS) && !memcmp(name, "heads/", 6))
33 return 1;
35 /* REF_TAGS means that we want tags */
36 if ((flags & REF_TAGS) && !memcmp(name, "tags/", 5))
37 return 1;
39 /* All type bits clear means that we are ok with anything */
40 return !(flags & ~REF_NORMAL);
44 * Read all the refs from the other end
46 struct ref **get_remote_heads(int in, struct ref **list,
47 int nr_match, char **match,
48 unsigned int flags)
50 *list = NULL;
51 for (;;) {
52 struct ref *ref;
53 unsigned char old_sha1[20];
54 static char buffer[1000];
55 char *name;
56 int len, name_len;
58 len = packet_read_line(in, buffer, sizeof(buffer));
59 if (!len)
60 break;
61 if (buffer[len-1] == '\n')
62 buffer[--len] = 0;
64 if (len < 42 || get_sha1_hex(buffer, old_sha1) || buffer[40] != ' ')
65 die("protocol error: expected sha/ref, got '%s'", buffer);
66 name = buffer + 41;
68 name_len = strlen(name);
69 if (len != name_len + 41) {
70 if (server_capabilities)
71 free(server_capabilities);
72 server_capabilities = xstrdup(name + name_len + 1);
75 if (!check_ref(name, name_len, flags))
76 continue;
77 if (nr_match && !path_match(name, nr_match, match))
78 continue;
79 ref = xcalloc(1, sizeof(*ref) + len - 40);
80 hashcpy(ref->old_sha1, old_sha1);
81 memcpy(ref->name, buffer + 41, len - 40);
82 *list = ref;
83 list = &ref->next;
85 return list;
88 int server_supports(const char *feature)
90 return server_capabilities &&
91 strstr(server_capabilities, feature) != NULL;
94 int get_ack(int fd, unsigned char *result_sha1)
96 static char line[1000];
97 int len = packet_read_line(fd, line, sizeof(line));
99 if (!len)
100 die("git-fetch-pack: expected ACK/NAK, got EOF");
101 if (line[len-1] == '\n')
102 line[--len] = 0;
103 if (!strcmp(line, "NAK"))
104 return 0;
105 if (!strncmp(line, "ACK ", 4)) {
106 if (!get_sha1_hex(line+4, result_sha1)) {
107 if (strstr(line+45, "continue"))
108 return 2;
109 return 1;
112 die("git-fetch_pack: expected ACK/NAK, got '%s'", line);
115 int path_match(const char *path, int nr, char **match)
117 int i;
118 int pathlen = strlen(path);
120 for (i = 0; i < nr; i++) {
121 char *s = match[i];
122 int len = strlen(s);
124 if (!len || len > pathlen)
125 continue;
126 if (memcmp(path + pathlen - len, s, len))
127 continue;
128 if (pathlen > len && path[pathlen - len - 1] != '/')
129 continue;
130 *s = 0;
131 return (i + 1);
133 return 0;
136 struct refspec {
137 char *src;
138 char *dst;
139 char force;
143 * A:B means fast forward remote B with local A.
144 * +A:B means overwrite remote B with local A.
145 * +A is a shorthand for +A:A.
146 * A is a shorthand for A:A.
148 static struct refspec *parse_ref_spec(int nr_refspec, char **refspec)
150 int i;
151 struct refspec *rs = xcalloc(sizeof(*rs), (nr_refspec + 1));
152 for (i = 0; i < nr_refspec; i++) {
153 char *sp, *dp, *ep;
154 sp = refspec[i];
155 if (*sp == '+') {
156 rs[i].force = 1;
157 sp++;
159 ep = strchr(sp, ':');
160 if (ep) {
161 dp = ep + 1;
162 *ep = 0;
164 else
165 dp = sp;
166 rs[i].src = sp;
167 rs[i].dst = dp;
169 rs[nr_refspec].src = rs[nr_refspec].dst = NULL;
170 return rs;
173 static int count_refspec_match(const char *pattern,
174 struct ref *refs,
175 struct ref **matched_ref)
177 int patlen = strlen(pattern);
178 struct ref *matched_weak = NULL;
179 struct ref *matched = NULL;
180 int weak_match = 0;
181 int match = 0;
183 for (weak_match = match = 0; refs; refs = refs->next) {
184 char *name = refs->name;
185 int namelen = strlen(name);
186 int weak_match;
188 if (namelen < patlen ||
189 memcmp(name + namelen - patlen, pattern, patlen))
190 continue;
191 if (namelen != patlen && name[namelen - patlen - 1] != '/')
192 continue;
194 /* A match is "weak" if it is with refs outside
195 * heads or tags, and did not specify the pattern
196 * in full (e.g. "refs/remotes/origin/master") or at
197 * least from the toplevel (e.g. "remotes/origin/master");
198 * otherwise "git push $URL master" would result in
199 * ambiguity between remotes/origin/master and heads/master
200 * at the remote site.
202 if (namelen != patlen &&
203 patlen != namelen - 5 &&
204 strncmp(name, "refs/heads/", 11) &&
205 strncmp(name, "refs/tags/", 10)) {
206 /* We want to catch the case where only weak
207 * matches are found and there are multiple
208 * matches, and where more than one strong
209 * matches are found, as ambiguous. One
210 * strong match with zero or more weak matches
211 * are acceptable as a unique match.
213 matched_weak = refs;
214 weak_match++;
216 else {
217 matched = refs;
218 match++;
221 if (!matched) {
222 *matched_ref = matched_weak;
223 return weak_match;
225 else {
226 *matched_ref = matched;
227 return match;
231 static void link_dst_tail(struct ref *ref, struct ref ***tail)
233 **tail = ref;
234 *tail = &ref->next;
235 **tail = NULL;
238 static struct ref *try_explicit_object_name(const char *name)
240 unsigned char sha1[20];
241 struct ref *ref;
242 int len;
243 if (get_sha1(name, sha1))
244 return NULL;
245 len = strlen(name) + 1;
246 ref = xcalloc(1, sizeof(*ref) + len);
247 memcpy(ref->name, name, len);
248 hashcpy(ref->new_sha1, sha1);
249 return ref;
252 static int match_explicit_refs(struct ref *src, struct ref *dst,
253 struct ref ***dst_tail, struct refspec *rs)
255 int i, errs;
256 for (i = errs = 0; rs[i].src; i++) {
257 struct ref *matched_src, *matched_dst;
259 matched_src = matched_dst = NULL;
260 switch (count_refspec_match(rs[i].src, src, &matched_src)) {
261 case 1:
262 break;
263 case 0:
264 /* The source could be in the get_sha1() format
265 * not a reference name.
267 matched_src = try_explicit_object_name(rs[i].src);
268 if (matched_src)
269 break;
270 errs = 1;
271 error("src refspec %s does not match any.",
272 rs[i].src);
273 break;
274 default:
275 errs = 1;
276 error("src refspec %s matches more than one.",
277 rs[i].src);
278 break;
280 switch (count_refspec_match(rs[i].dst, dst, &matched_dst)) {
281 case 1:
282 break;
283 case 0:
284 if (!memcmp(rs[i].dst, "refs/", 5)) {
285 int len = strlen(rs[i].dst) + 1;
286 matched_dst = xcalloc(1, sizeof(*dst) + len);
287 memcpy(matched_dst->name, rs[i].dst, len);
288 link_dst_tail(matched_dst, dst_tail);
290 else if (!strcmp(rs[i].src, rs[i].dst) &&
291 matched_src) {
292 /* pushing "master:master" when
293 * remote does not have master yet.
295 int len = strlen(matched_src->name) + 1;
296 matched_dst = xcalloc(1, sizeof(*dst) + len);
297 memcpy(matched_dst->name, matched_src->name,
298 len);
299 link_dst_tail(matched_dst, dst_tail);
301 else {
302 errs = 1;
303 error("dst refspec %s does not match any "
304 "existing ref on the remote and does "
305 "not start with refs/.", rs[i].dst);
307 break;
308 default:
309 errs = 1;
310 error("dst refspec %s matches more than one.",
311 rs[i].dst);
312 break;
314 if (errs)
315 continue;
316 if (matched_dst->peer_ref) {
317 errs = 1;
318 error("dst ref %s receives from more than one src.",
319 matched_dst->name);
321 else {
322 matched_dst->peer_ref = matched_src;
323 matched_dst->force = rs[i].force;
326 return -errs;
329 static struct ref *find_ref_by_name(struct ref *list, const char *name)
331 for ( ; list; list = list->next)
332 if (!strcmp(list->name, name))
333 return list;
334 return NULL;
337 int match_refs(struct ref *src, struct ref *dst, struct ref ***dst_tail,
338 int nr_refspec, char **refspec, int all)
340 struct refspec *rs = parse_ref_spec(nr_refspec, refspec);
342 if (nr_refspec)
343 return match_explicit_refs(src, dst, dst_tail, rs);
345 /* pick the remainder */
346 for ( ; src; src = src->next) {
347 struct ref *dst_peer;
348 if (src->peer_ref)
349 continue;
350 dst_peer = find_ref_by_name(dst, src->name);
351 if ((dst_peer && dst_peer->peer_ref) || (!dst_peer && !all))
352 continue;
353 if (!dst_peer) {
354 /* Create a new one and link it */
355 int len = strlen(src->name) + 1;
356 dst_peer = xcalloc(1, sizeof(*dst_peer) + len);
357 memcpy(dst_peer->name, src->name, len);
358 hashcpy(dst_peer->new_sha1, src->new_sha1);
359 link_dst_tail(dst_peer, dst_tail);
361 dst_peer->peer_ref = src;
363 return 0;
366 enum protocol {
367 PROTO_LOCAL = 1,
368 PROTO_SSH,
369 PROTO_GIT,
372 static enum protocol get_protocol(const char *name)
374 if (!strcmp(name, "ssh"))
375 return PROTO_SSH;
376 if (!strcmp(name, "git"))
377 return PROTO_GIT;
378 if (!strcmp(name, "git+ssh"))
379 return PROTO_SSH;
380 if (!strcmp(name, "ssh+git"))
381 return PROTO_SSH;
382 die("I don't handle protocol '%s'", name);
385 #define STR_(s) # s
386 #define STR(s) STR_(s)
388 #ifndef NO_IPV6
391 * Returns a connected socket() fd, or else die()s.
393 static int git_tcp_connect_sock(char *host)
395 int sockfd = -1, saved_errno = 0;
396 char *colon, *end;
397 const char *port = STR(DEFAULT_GIT_PORT);
398 struct addrinfo hints, *ai0, *ai;
399 int gai;
401 if (host[0] == '[') {
402 end = strchr(host + 1, ']');
403 if (end) {
404 *end = 0;
405 end++;
406 host++;
407 } else
408 end = host;
409 } else
410 end = host;
411 colon = strchr(end, ':');
413 if (colon) {
414 *colon = 0;
415 port = colon + 1;
418 memset(&hints, 0, sizeof(hints));
419 hints.ai_socktype = SOCK_STREAM;
420 hints.ai_protocol = IPPROTO_TCP;
422 gai = getaddrinfo(host, port, &hints, &ai);
423 if (gai)
424 die("Unable to look up %s (%s)", host, gai_strerror(gai));
426 for (ai0 = ai; ai; ai = ai->ai_next) {
427 sockfd = socket(ai->ai_family,
428 ai->ai_socktype, ai->ai_protocol);
429 if (sockfd < 0) {
430 saved_errno = errno;
431 continue;
433 if (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0) {
434 saved_errno = errno;
435 close(sockfd);
436 sockfd = -1;
437 continue;
439 break;
442 freeaddrinfo(ai0);
444 if (sockfd < 0)
445 die("unable to connect a socket (%s)", strerror(saved_errno));
447 return sockfd;
450 #else /* NO_IPV6 */
453 * Returns a connected socket() fd, or else die()s.
455 static int git_tcp_connect_sock(char *host)
457 int sockfd = -1, saved_errno = 0;
458 char *colon, *end;
459 char *port = STR(DEFAULT_GIT_PORT), *ep;
460 struct hostent *he;
461 struct sockaddr_in sa;
462 char **ap;
463 unsigned int nport;
465 if (host[0] == '[') {
466 end = strchr(host + 1, ']');
467 if (end) {
468 *end = 0;
469 end++;
470 host++;
471 } else
472 end = host;
473 } else
474 end = host;
475 colon = strchr(end, ':');
477 if (colon) {
478 *colon = 0;
479 port = colon + 1;
482 he = gethostbyname(host);
483 if (!he)
484 die("Unable to look up %s (%s)", host, hstrerror(h_errno));
485 nport = strtoul(port, &ep, 10);
486 if ( ep == port || *ep ) {
487 /* Not numeric */
488 struct servent *se = getservbyname(port,"tcp");
489 if ( !se )
490 die("Unknown port %s\n", port);
491 nport = se->s_port;
494 for (ap = he->h_addr_list; *ap; ap++) {
495 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
496 if (sockfd < 0) {
497 saved_errno = errno;
498 continue;
501 memset(&sa, 0, sizeof sa);
502 sa.sin_family = he->h_addrtype;
503 sa.sin_port = htons(nport);
504 memcpy(&sa.sin_addr, *ap, he->h_length);
506 if (connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
507 saved_errno = errno;
508 close(sockfd);
509 sockfd = -1;
510 continue;
512 break;
515 if (sockfd < 0)
516 die("unable to connect a socket (%s)", strerror(saved_errno));
518 return sockfd;
521 #endif /* NO_IPV6 */
524 static void git_tcp_connect(int fd[2], char *host)
526 int sockfd = git_tcp_connect_sock(host);
528 fd[0] = sockfd;
529 fd[1] = sockfd;
533 static char *git_proxy_command;
534 static const char *rhost_name;
535 static int rhost_len;
537 static int git_proxy_command_options(const char *var, const char *value)
539 if (!strcmp(var, "core.gitproxy")) {
540 const char *for_pos;
541 int matchlen = -1;
542 int hostlen;
544 if (git_proxy_command)
545 return 0;
546 /* [core]
547 * ;# matches www.kernel.org as well
548 * gitproxy = netcatter-1 for kernel.org
549 * gitproxy = netcatter-2 for sample.xz
550 * gitproxy = netcatter-default
552 for_pos = strstr(value, " for ");
553 if (!for_pos)
554 /* matches everybody */
555 matchlen = strlen(value);
556 else {
557 hostlen = strlen(for_pos + 5);
558 if (rhost_len < hostlen)
559 matchlen = -1;
560 else if (!strncmp(for_pos + 5,
561 rhost_name + rhost_len - hostlen,
562 hostlen) &&
563 ((rhost_len == hostlen) ||
564 rhost_name[rhost_len - hostlen -1] == '.'))
565 matchlen = for_pos - value;
566 else
567 matchlen = -1;
569 if (0 <= matchlen) {
570 /* core.gitproxy = none for kernel.org */
571 if (matchlen == 4 &&
572 !memcmp(value, "none", 4))
573 matchlen = 0;
574 git_proxy_command = xmalloc(matchlen + 1);
575 memcpy(git_proxy_command, value, matchlen);
576 git_proxy_command[matchlen] = 0;
578 return 0;
581 return git_default_config(var, value);
584 static int git_use_proxy(const char *host)
586 rhost_name = host;
587 rhost_len = strlen(host);
588 git_proxy_command = getenv("GIT_PROXY_COMMAND");
589 git_config(git_proxy_command_options);
590 rhost_name = NULL;
591 return (git_proxy_command && *git_proxy_command);
594 static void git_proxy_connect(int fd[2], char *host)
596 const char *port = STR(DEFAULT_GIT_PORT);
597 char *colon, *end;
598 int pipefd[2][2];
599 pid_t pid;
601 if (host[0] == '[') {
602 end = strchr(host + 1, ']');
603 if (end) {
604 *end = 0;
605 end++;
606 host++;
607 } else
608 end = host;
609 } else
610 end = host;
611 colon = strchr(end, ':');
613 if (colon) {
614 *colon = 0;
615 port = colon + 1;
618 if (pipe(pipefd[0]) < 0 || pipe(pipefd[1]) < 0)
619 die("unable to create pipe pair for communication");
620 pid = fork();
621 if (!pid) {
622 dup2(pipefd[1][0], 0);
623 dup2(pipefd[0][1], 1);
624 close(pipefd[0][0]);
625 close(pipefd[0][1]);
626 close(pipefd[1][0]);
627 close(pipefd[1][1]);
628 execlp(git_proxy_command, git_proxy_command, host, port, NULL);
629 die("exec failed");
631 if (pid < 0)
632 die("fork failed");
633 fd[0] = pipefd[0][0];
634 fd[1] = pipefd[1][1];
635 close(pipefd[0][1]);
636 close(pipefd[1][0]);
639 #define MAX_CMD_LEN 1024
642 * This returns 0 if the transport protocol does not need fork(2),
643 * or a process id if it does. Once done, finish the connection
644 * with finish_connect() with the value returned from this function
645 * (it is safe to call finish_connect() with 0 to support the former
646 * case).
648 * Does not return a negative value on error; it just dies.
650 pid_t git_connect(int fd[2], char *url, const char *prog)
652 char *host, *path = url;
653 char *end;
654 int c;
655 int pipefd[2][2];
656 pid_t pid;
657 enum protocol protocol = PROTO_LOCAL;
658 int free_path = 0;
660 /* Without this we cannot rely on waitpid() to tell
661 * what happened to our children.
663 signal(SIGCHLD, SIG_DFL);
665 host = strstr(url, "://");
666 if(host) {
667 *host = '\0';
668 protocol = get_protocol(url);
669 host += 3;
670 c = '/';
671 } else {
672 host = url;
673 c = ':';
676 if (host[0] == '[') {
677 end = strchr(host + 1, ']');
678 if (end) {
679 *end = 0;
680 end++;
681 host++;
682 } else
683 end = host;
684 } else
685 end = host;
687 path = strchr(end, c);
688 if (c == ':') {
689 if (path) {
690 protocol = PROTO_SSH;
691 *path++ = '\0';
692 } else
693 path = host;
696 if (!path || !*path)
697 die("No path specified. See 'man git-pull' for valid url syntax");
700 * null-terminate hostname and point path to ~ for URL's like this:
701 * ssh://host.xz/~user/repo
703 if (protocol != PROTO_LOCAL && host != url) {
704 char *ptr = path;
705 if (path[1] == '~')
706 path++;
707 else {
708 path = xstrdup(ptr);
709 free_path = 1;
712 *ptr = '\0';
715 if (protocol == PROTO_GIT) {
716 /* These underlying connection commands die() if they
717 * cannot connect.
719 char *target_host = xstrdup(host);
720 if (git_use_proxy(host))
721 git_proxy_connect(fd, host);
722 else
723 git_tcp_connect(fd, host);
725 * Separate original protocol components prog and path
726 * from extended components with a NUL byte.
728 packet_write(fd[1],
729 "%s %s%chost=%s%c",
730 prog, path, 0,
731 target_host, 0);
732 free(target_host);
733 if (free_path)
734 free(path);
735 return 0;
738 if (pipe(pipefd[0]) < 0 || pipe(pipefd[1]) < 0)
739 die("unable to create pipe pair for communication");
740 pid = fork();
741 if (pid < 0)
742 die("unable to fork");
743 if (!pid) {
744 char command[MAX_CMD_LEN];
745 char *posn = command;
746 int size = MAX_CMD_LEN;
747 int of = 0;
749 of |= add_to_string(&posn, &size, prog, 0);
750 of |= add_to_string(&posn, &size, " ", 0);
751 of |= add_to_string(&posn, &size, path, 1);
753 if (of)
754 die("command line too long");
756 dup2(pipefd[1][0], 0);
757 dup2(pipefd[0][1], 1);
758 close(pipefd[0][0]);
759 close(pipefd[0][1]);
760 close(pipefd[1][0]);
761 close(pipefd[1][1]);
762 if (protocol == PROTO_SSH) {
763 const char *ssh, *ssh_basename;
764 ssh = getenv("GIT_SSH");
765 if (!ssh) ssh = "ssh";
766 ssh_basename = strrchr(ssh, '/');
767 if (!ssh_basename)
768 ssh_basename = ssh;
769 else
770 ssh_basename++;
771 execlp(ssh, ssh_basename, host, command, NULL);
773 else {
774 unsetenv(ALTERNATE_DB_ENVIRONMENT);
775 unsetenv(DB_ENVIRONMENT);
776 unsetenv(GIT_DIR_ENVIRONMENT);
777 unsetenv(GRAFT_ENVIRONMENT);
778 unsetenv(INDEX_ENVIRONMENT);
779 execlp("sh", "sh", "-c", command, NULL);
781 die("exec failed");
783 fd[0] = pipefd[0][0];
784 fd[1] = pipefd[1][1];
785 close(pipefd[0][1]);
786 close(pipefd[1][0]);
787 if (free_path)
788 free(path);
789 return pid;
792 int finish_connect(pid_t pid)
794 if (pid == 0)
795 return 0;
797 while (waitpid(pid, NULL, 0) < 0) {
798 if (errno != EINTR)
799 return -1;
801 return 0;