Fix "Do not ignore a detected patchfile brokenness."
[git/fastimport.git] / connect.c
blobc55a20a4aa31e7cf1bbf0dcec6b4ebccb655d850
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 match;
178 int patlen = strlen(pattern);
180 for (match = 0; refs; refs = refs->next) {
181 char *name = refs->name;
182 int namelen = strlen(name);
183 if (namelen < patlen ||
184 memcmp(name + namelen - patlen, pattern, patlen))
185 continue;
186 if (namelen != patlen && name[namelen - patlen - 1] != '/')
187 continue;
188 match++;
189 *matched_ref = refs;
191 return match;
194 static void link_dst_tail(struct ref *ref, struct ref ***tail)
196 **tail = ref;
197 *tail = &ref->next;
198 **tail = NULL;
201 static struct ref *try_explicit_object_name(const char *name)
203 unsigned char sha1[20];
204 struct ref *ref;
205 int len;
206 if (get_sha1(name, sha1))
207 return NULL;
208 len = strlen(name) + 1;
209 ref = xcalloc(1, sizeof(*ref) + len);
210 memcpy(ref->name, name, len);
211 hashcpy(ref->new_sha1, sha1);
212 return ref;
215 static int match_explicit_refs(struct ref *src, struct ref *dst,
216 struct ref ***dst_tail, struct refspec *rs)
218 int i, errs;
219 for (i = errs = 0; rs[i].src; i++) {
220 struct ref *matched_src, *matched_dst;
222 matched_src = matched_dst = NULL;
223 switch (count_refspec_match(rs[i].src, src, &matched_src)) {
224 case 1:
225 break;
226 case 0:
227 /* The source could be in the get_sha1() format
228 * not a reference name.
230 matched_src = try_explicit_object_name(rs[i].src);
231 if (matched_src)
232 break;
233 errs = 1;
234 error("src refspec %s does not match any.",
235 rs[i].src);
236 break;
237 default:
238 errs = 1;
239 error("src refspec %s matches more than one.",
240 rs[i].src);
241 break;
243 switch (count_refspec_match(rs[i].dst, dst, &matched_dst)) {
244 case 1:
245 break;
246 case 0:
247 if (!memcmp(rs[i].dst, "refs/", 5)) {
248 int len = strlen(rs[i].dst) + 1;
249 matched_dst = xcalloc(1, sizeof(*dst) + len);
250 memcpy(matched_dst->name, rs[i].dst, len);
251 link_dst_tail(matched_dst, dst_tail);
253 else if (!strcmp(rs[i].src, rs[i].dst) &&
254 matched_src) {
255 /* pushing "master:master" when
256 * remote does not have master yet.
258 int len = strlen(matched_src->name) + 1;
259 matched_dst = xcalloc(1, sizeof(*dst) + len);
260 memcpy(matched_dst->name, matched_src->name,
261 len);
262 link_dst_tail(matched_dst, dst_tail);
264 else {
265 errs = 1;
266 error("dst refspec %s does not match any "
267 "existing ref on the remote and does "
268 "not start with refs/.", rs[i].dst);
270 break;
271 default:
272 errs = 1;
273 error("dst refspec %s matches more than one.",
274 rs[i].dst);
275 break;
277 if (errs)
278 continue;
279 if (matched_dst->peer_ref) {
280 errs = 1;
281 error("dst ref %s receives from more than one src.",
282 matched_dst->name);
284 else {
285 matched_dst->peer_ref = matched_src;
286 matched_dst->force = rs[i].force;
289 return -errs;
292 static struct ref *find_ref_by_name(struct ref *list, const char *name)
294 for ( ; list; list = list->next)
295 if (!strcmp(list->name, name))
296 return list;
297 return NULL;
300 int match_refs(struct ref *src, struct ref *dst, struct ref ***dst_tail,
301 int nr_refspec, char **refspec, int all)
303 struct refspec *rs = parse_ref_spec(nr_refspec, refspec);
305 if (nr_refspec)
306 return match_explicit_refs(src, dst, dst_tail, rs);
308 /* pick the remainder */
309 for ( ; src; src = src->next) {
310 struct ref *dst_peer;
311 if (src->peer_ref)
312 continue;
313 dst_peer = find_ref_by_name(dst, src->name);
314 if ((dst_peer && dst_peer->peer_ref) || (!dst_peer && !all))
315 continue;
316 if (!dst_peer) {
317 /* Create a new one and link it */
318 int len = strlen(src->name) + 1;
319 dst_peer = xcalloc(1, sizeof(*dst_peer) + len);
320 memcpy(dst_peer->name, src->name, len);
321 hashcpy(dst_peer->new_sha1, src->new_sha1);
322 link_dst_tail(dst_peer, dst_tail);
324 dst_peer->peer_ref = src;
326 return 0;
329 enum protocol {
330 PROTO_LOCAL = 1,
331 PROTO_SSH,
332 PROTO_GIT,
335 static enum protocol get_protocol(const char *name)
337 if (!strcmp(name, "ssh"))
338 return PROTO_SSH;
339 if (!strcmp(name, "git"))
340 return PROTO_GIT;
341 if (!strcmp(name, "git+ssh"))
342 return PROTO_SSH;
343 if (!strcmp(name, "ssh+git"))
344 return PROTO_SSH;
345 die("I don't handle protocol '%s'", name);
348 #define STR_(s) # s
349 #define STR(s) STR_(s)
351 #ifndef NO_IPV6
354 * Returns a connected socket() fd, or else die()s.
356 static int git_tcp_connect_sock(char *host)
358 int sockfd = -1, saved_errno = 0;
359 char *colon, *end;
360 const char *port = STR(DEFAULT_GIT_PORT);
361 struct addrinfo hints, *ai0, *ai;
362 int gai;
364 if (host[0] == '[') {
365 end = strchr(host + 1, ']');
366 if (end) {
367 *end = 0;
368 end++;
369 host++;
370 } else
371 end = host;
372 } else
373 end = host;
374 colon = strchr(end, ':');
376 if (colon) {
377 *colon = 0;
378 port = colon + 1;
381 memset(&hints, 0, sizeof(hints));
382 hints.ai_socktype = SOCK_STREAM;
383 hints.ai_protocol = IPPROTO_TCP;
385 gai = getaddrinfo(host, port, &hints, &ai);
386 if (gai)
387 die("Unable to look up %s (%s)", host, gai_strerror(gai));
389 for (ai0 = ai; ai; ai = ai->ai_next) {
390 sockfd = socket(ai->ai_family,
391 ai->ai_socktype, ai->ai_protocol);
392 if (sockfd < 0) {
393 saved_errno = errno;
394 continue;
396 if (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0) {
397 saved_errno = errno;
398 close(sockfd);
399 sockfd = -1;
400 continue;
402 break;
405 freeaddrinfo(ai0);
407 if (sockfd < 0)
408 die("unable to connect a socket (%s)", strerror(saved_errno));
410 return sockfd;
413 #else /* NO_IPV6 */
416 * Returns a connected socket() fd, or else die()s.
418 static int git_tcp_connect_sock(char *host)
420 int sockfd = -1, saved_errno = 0;
421 char *colon, *end;
422 char *port = STR(DEFAULT_GIT_PORT), *ep;
423 struct hostent *he;
424 struct sockaddr_in sa;
425 char **ap;
426 unsigned int nport;
428 if (host[0] == '[') {
429 end = strchr(host + 1, ']');
430 if (end) {
431 *end = 0;
432 end++;
433 host++;
434 } else
435 end = host;
436 } else
437 end = host;
438 colon = strchr(end, ':');
440 if (colon) {
441 *colon = 0;
442 port = colon + 1;
445 he = gethostbyname(host);
446 if (!he)
447 die("Unable to look up %s (%s)", host, hstrerror(h_errno));
448 nport = strtoul(port, &ep, 10);
449 if ( ep == port || *ep ) {
450 /* Not numeric */
451 struct servent *se = getservbyname(port,"tcp");
452 if ( !se )
453 die("Unknown port %s\n", port);
454 nport = se->s_port;
457 for (ap = he->h_addr_list; *ap; ap++) {
458 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
459 if (sockfd < 0) {
460 saved_errno = errno;
461 continue;
464 memset(&sa, 0, sizeof sa);
465 sa.sin_family = he->h_addrtype;
466 sa.sin_port = htons(nport);
467 memcpy(&sa.sin_addr, *ap, he->h_length);
469 if (connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
470 saved_errno = errno;
471 close(sockfd);
472 sockfd = -1;
473 continue;
475 break;
478 if (sockfd < 0)
479 die("unable to connect a socket (%s)", strerror(saved_errno));
481 return sockfd;
484 #endif /* NO_IPV6 */
487 static void git_tcp_connect(int fd[2], char *host)
489 int sockfd = git_tcp_connect_sock(host);
491 fd[0] = sockfd;
492 fd[1] = sockfd;
496 static char *git_proxy_command;
497 static const char *rhost_name;
498 static int rhost_len;
500 static int git_proxy_command_options(const char *var, const char *value)
502 if (!strcmp(var, "core.gitproxy")) {
503 const char *for_pos;
504 int matchlen = -1;
505 int hostlen;
507 if (git_proxy_command)
508 return 0;
509 /* [core]
510 * ;# matches www.kernel.org as well
511 * gitproxy = netcatter-1 for kernel.org
512 * gitproxy = netcatter-2 for sample.xz
513 * gitproxy = netcatter-default
515 for_pos = strstr(value, " for ");
516 if (!for_pos)
517 /* matches everybody */
518 matchlen = strlen(value);
519 else {
520 hostlen = strlen(for_pos + 5);
521 if (rhost_len < hostlen)
522 matchlen = -1;
523 else if (!strncmp(for_pos + 5,
524 rhost_name + rhost_len - hostlen,
525 hostlen) &&
526 ((rhost_len == hostlen) ||
527 rhost_name[rhost_len - hostlen -1] == '.'))
528 matchlen = for_pos - value;
529 else
530 matchlen = -1;
532 if (0 <= matchlen) {
533 /* core.gitproxy = none for kernel.org */
534 if (matchlen == 4 &&
535 !memcmp(value, "none", 4))
536 matchlen = 0;
537 git_proxy_command = xmalloc(matchlen + 1);
538 memcpy(git_proxy_command, value, matchlen);
539 git_proxy_command[matchlen] = 0;
541 return 0;
544 return git_default_config(var, value);
547 static int git_use_proxy(const char *host)
549 rhost_name = host;
550 rhost_len = strlen(host);
551 git_proxy_command = getenv("GIT_PROXY_COMMAND");
552 git_config(git_proxy_command_options);
553 rhost_name = NULL;
554 return (git_proxy_command && *git_proxy_command);
557 static void git_proxy_connect(int fd[2], char *host)
559 const char *port = STR(DEFAULT_GIT_PORT);
560 char *colon, *end;
561 int pipefd[2][2];
562 pid_t pid;
564 if (host[0] == '[') {
565 end = strchr(host + 1, ']');
566 if (end) {
567 *end = 0;
568 end++;
569 host++;
570 } else
571 end = host;
572 } else
573 end = host;
574 colon = strchr(end, ':');
576 if (colon) {
577 *colon = 0;
578 port = colon + 1;
581 if (pipe(pipefd[0]) < 0 || pipe(pipefd[1]) < 0)
582 die("unable to create pipe pair for communication");
583 pid = fork();
584 if (!pid) {
585 dup2(pipefd[1][0], 0);
586 dup2(pipefd[0][1], 1);
587 close(pipefd[0][0]);
588 close(pipefd[0][1]);
589 close(pipefd[1][0]);
590 close(pipefd[1][1]);
591 execlp(git_proxy_command, git_proxy_command, host, port, NULL);
592 die("exec failed");
594 if (pid < 0)
595 die("fork failed");
596 fd[0] = pipefd[0][0];
597 fd[1] = pipefd[1][1];
598 close(pipefd[0][1]);
599 close(pipefd[1][0]);
602 #define MAX_CMD_LEN 1024
605 * This returns 0 if the transport protocol does not need fork(2),
606 * or a process id if it does. Once done, finish the connection
607 * with finish_connect() with the value returned from this function
608 * (it is safe to call finish_connect() with 0 to support the former
609 * case).
611 * Does not return a negative value on error; it just dies.
613 pid_t git_connect(int fd[2], char *url, const char *prog)
615 char *host, *path = url;
616 char *end;
617 int c;
618 int pipefd[2][2];
619 pid_t pid;
620 enum protocol protocol = PROTO_LOCAL;
621 int free_path = 0;
623 /* Without this we cannot rely on waitpid() to tell
624 * what happened to our children.
626 signal(SIGCHLD, SIG_DFL);
628 host = strstr(url, "://");
629 if(host) {
630 *host = '\0';
631 protocol = get_protocol(url);
632 host += 3;
633 c = '/';
634 } else {
635 host = url;
636 c = ':';
639 if (host[0] == '[') {
640 end = strchr(host + 1, ']');
641 if (end) {
642 *end = 0;
643 end++;
644 host++;
645 } else
646 end = host;
647 } else
648 end = host;
650 path = strchr(end, c);
651 if (c == ':') {
652 if (path) {
653 protocol = PROTO_SSH;
654 *path++ = '\0';
655 } else
656 path = host;
659 if (!path || !*path)
660 die("No path specified. See 'man git-pull' for valid url syntax");
663 * null-terminate hostname and point path to ~ for URL's like this:
664 * ssh://host.xz/~user/repo
666 if (protocol != PROTO_LOCAL && host != url) {
667 char *ptr = path;
668 if (path[1] == '~')
669 path++;
670 else {
671 path = xstrdup(ptr);
672 free_path = 1;
675 *ptr = '\0';
678 if (protocol == PROTO_GIT) {
679 /* These underlying connection commands die() if they
680 * cannot connect.
682 char *target_host = xstrdup(host);
683 if (git_use_proxy(host))
684 git_proxy_connect(fd, host);
685 else
686 git_tcp_connect(fd, host);
688 * Separate original protocol components prog and path
689 * from extended components with a NUL byte.
691 packet_write(fd[1],
692 "%s %s%chost=%s%c",
693 prog, path, 0,
694 target_host, 0);
695 free(target_host);
696 if (free_path)
697 free(path);
698 return 0;
701 if (pipe(pipefd[0]) < 0 || pipe(pipefd[1]) < 0)
702 die("unable to create pipe pair for communication");
703 pid = fork();
704 if (pid < 0)
705 die("unable to fork");
706 if (!pid) {
707 char command[MAX_CMD_LEN];
708 char *posn = command;
709 int size = MAX_CMD_LEN;
710 int of = 0;
712 of |= add_to_string(&posn, &size, prog, 0);
713 of |= add_to_string(&posn, &size, " ", 0);
714 of |= add_to_string(&posn, &size, path, 1);
716 if (of)
717 die("command line too long");
719 dup2(pipefd[1][0], 0);
720 dup2(pipefd[0][1], 1);
721 close(pipefd[0][0]);
722 close(pipefd[0][1]);
723 close(pipefd[1][0]);
724 close(pipefd[1][1]);
725 if (protocol == PROTO_SSH) {
726 const char *ssh, *ssh_basename;
727 ssh = getenv("GIT_SSH");
728 if (!ssh) ssh = "ssh";
729 ssh_basename = strrchr(ssh, '/');
730 if (!ssh_basename)
731 ssh_basename = ssh;
732 else
733 ssh_basename++;
734 execlp(ssh, ssh_basename, host, command, NULL);
736 else {
737 unsetenv(ALTERNATE_DB_ENVIRONMENT);
738 unsetenv(DB_ENVIRONMENT);
739 unsetenv(GIT_DIR_ENVIRONMENT);
740 unsetenv(GRAFT_ENVIRONMENT);
741 unsetenv(INDEX_ENVIRONMENT);
742 execlp("sh", "sh", "-c", command, NULL);
744 die("exec failed");
746 fd[0] = pipefd[0][0];
747 fd[1] = pipefd[1][1];
748 close(pipefd[0][1]);
749 close(pipefd[1][0]);
750 if (free_path)
751 free(path);
752 return pid;
755 int finish_connect(pid_t pid)
757 if (pid == 0)
758 return 0;
760 while (waitpid(pid, NULL, 0) < 0) {
761 if (errno != EINTR)
762 return -1;
764 return 0;