Merge branch 'mh/refs-have-new'
[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, unsigned int flags)
17 {
18         if (!flags)
19                 return 1;
20
21         if (!skip_prefix(name, "refs/", &name))
22                 return 0;
23
24         /* REF_NORMAL means that we don't want the magic fake tag refs */
25         if ((flags & REF_NORMAL) && check_refname_format(name, 0))
26                 return 0;
27
28         /* REF_HEADS means that we want regular branch heads */
29         if ((flags & REF_HEADS) && starts_with(name, "heads/"))
30                 return 1;
31
32         /* REF_TAGS means that we want tags */
33         if ((flags & REF_TAGS) && starts_with(name, "tags/"))
34                 return 1;
35
36         /* All type bits clear means that we are ok with anything */
37         return !(flags & ~REF_NORMAL);
38 }
39
40 int check_ref_type(const struct ref *ref, int flags)
41 {
42         return check_ref(ref->name, flags);
43 }
44
45 static void die_initial_contact(int got_at_least_one_head)
46 {
47         if (got_at_least_one_head)
48                 die("The remote end hung up upon initial contact");
49         else
50                 die("Could not read from remote repository.\n\n"
51                     "Please make sure you have the correct access rights\n"
52                     "and the repository exists.");
53 }
54
55 static void parse_one_symref_info(struct string_list *symref, const char *val, int len)
56 {
57         char *sym, *target;
58         struct string_list_item *item;
59
60         if (!len)
61                 return; /* just "symref" */
62         /* e.g. "symref=HEAD:refs/heads/master" */
63         sym = xmemdupz(val, len);
64         target = strchr(sym, ':');
65         if (!target)
66                 /* just "symref=something" */
67                 goto reject;
68         *(target++) = '\0';
69         if (check_refname_format(sym, REFNAME_ALLOW_ONELEVEL) ||
70             check_refname_format(target, REFNAME_ALLOW_ONELEVEL))
71                 /* "symref=bogus:pair */
72                 goto reject;
73         item = string_list_append(symref, sym);
74         item->util = target;
75         return;
76 reject:
77         free(sym);
78         return;
79 }
80
81 static void annotate_refs_with_symref_info(struct ref *ref)
82 {
83         struct string_list symref = STRING_LIST_INIT_DUP;
84         const char *feature_list = server_capabilities;
85
86         while (feature_list) {
87                 int len;
88                 const char *val;
89
90                 val = parse_feature_value(feature_list, "symref", &len);
91                 if (!val)
92                         break;
93                 parse_one_symref_info(&symref, val, len);
94                 feature_list = val + 1;
95         }
96         string_list_sort(&symref);
97
98         for (; ref; ref = ref->next) {
99                 struct string_list_item *item;
100                 item = string_list_lookup(&symref, ref->name);
101                 if (!item)
102                         continue;
103                 ref->symref = xstrdup((char *)item->util);
104         }
105         string_list_clear(&symref, 0);
106 }
107
108 /*
109  * Read all the refs from the other end
110  */
111 struct ref **get_remote_heads(int in, char *src_buf, size_t src_len,
112                               struct ref **list, unsigned int flags,
113                               struct sha1_array *extra_have,
114                               struct sha1_array *shallow_points)
115 {
116         struct ref **orig_list = list;
117         int got_at_least_one_head = 0;
118
119         *list = NULL;
120         for (;;) {
121                 struct ref *ref;
122                 unsigned char old_sha1[20];
123                 char *name;
124                 int len, name_len;
125                 char *buffer = packet_buffer;
126                 const char *arg;
127
128                 len = packet_read(in, &src_buf, &src_len,
129                                   packet_buffer, sizeof(packet_buffer),
130                                   PACKET_READ_GENTLE_ON_EOF |
131                                   PACKET_READ_CHOMP_NEWLINE);
132                 if (len < 0)
133                         die_initial_contact(got_at_least_one_head);
134
135                 if (!len)
136                         break;
137
138                 if (len > 4 && skip_prefix(buffer, "ERR ", &arg))
139                         die("remote error: %s", arg);
140
141                 if (len == 48 && skip_prefix(buffer, "shallow ", &arg)) {
142                         if (get_sha1_hex(arg, old_sha1))
143                                 die("protocol error: expected shallow sha-1, got '%s'", arg);
144                         if (!shallow_points)
145                                 die("repository on the other end cannot be shallow");
146                         sha1_array_append(shallow_points, old_sha1);
147                         continue;
148                 }
149
150                 if (len < 42 || get_sha1_hex(buffer, old_sha1) || buffer[40] != ' ')
151                         die("protocol error: expected sha/ref, got '%s'", buffer);
152                 name = buffer + 41;
153
154                 name_len = strlen(name);
155                 if (len != name_len + 41) {
156                         free(server_capabilities);
157                         server_capabilities = xstrdup(name + name_len + 1);
158                 }
159
160                 if (extra_have &&
161                     name_len == 5 && !memcmp(".have", name, 5)) {
162                         sha1_array_append(extra_have, old_sha1);
163                         continue;
164                 }
165
166                 if (!check_ref(name, flags))
167                         continue;
168                 ref = alloc_ref(buffer + 41);
169                 hashcpy(ref->old_sha1, old_sha1);
170                 *list = ref;
171                 list = &ref->next;
172                 got_at_least_one_head = 1;
173         }
174
175         annotate_refs_with_symref_info(*orig_list);
176
177         return list;
178 }
179
180 static const char *parse_feature_value(const char *feature_list, const char *feature, int *lenp)
181 {
182         int len;
183
184         if (!feature_list)
185                 return NULL;
186
187         len = strlen(feature);
188         while (*feature_list) {
189                 const char *found = strstr(feature_list, feature);
190                 if (!found)
191                         return NULL;
192                 if (feature_list == found || isspace(found[-1])) {
193                         const char *value = found + len;
194                         /* feature with no value (e.g., "thin-pack") */
195                         if (!*value || isspace(*value)) {
196                                 if (lenp)
197                                         *lenp = 0;
198                                 return value;
199                         }
200                         /* feature with a value (e.g., "agent=git/1.2.3") */
201                         else if (*value == '=') {
202                                 value++;
203                                 if (lenp)
204                                         *lenp = strcspn(value, " \t\n");
205                                 return value;
206                         }
207                         /*
208                          * otherwise we matched a substring of another feature;
209                          * keep looking
210                          */
211                 }
212                 feature_list = found + 1;
213         }
214         return NULL;
215 }
216
217 int parse_feature_request(const char *feature_list, const char *feature)
218 {
219         return !!parse_feature_value(feature_list, feature, NULL);
220 }
221
222 const char *server_feature_value(const char *feature, int *len)
223 {
224         return parse_feature_value(server_capabilities, feature, len);
225 }
226
227 int server_supports(const char *feature)
228 {
229         return !!server_feature_value(feature, NULL);
230 }
231
232 enum protocol {
233         PROTO_LOCAL = 1,
234         PROTO_FILE,
235         PROTO_SSH,
236         PROTO_GIT
237 };
238
239 int url_is_local_not_ssh(const char *url)
240 {
241         const char *colon = strchr(url, ':');
242         const char *slash = strchr(url, '/');
243         return !colon || (slash && slash < colon) ||
244                 has_dos_drive_prefix(url);
245 }
246
247 static const char *prot_name(enum protocol protocol)
248 {
249         switch (protocol) {
250                 case PROTO_LOCAL:
251                 case PROTO_FILE:
252                         return "file";
253                 case PROTO_SSH:
254                         return "ssh";
255                 case PROTO_GIT:
256                         return "git";
257                 default:
258                         return "unkown protocol";
259         }
260 }
261
262 static enum protocol get_protocol(const char *name)
263 {
264         if (!strcmp(name, "ssh"))
265                 return PROTO_SSH;
266         if (!strcmp(name, "git"))
267                 return PROTO_GIT;
268         if (!strcmp(name, "git+ssh"))
269                 return PROTO_SSH;
270         if (!strcmp(name, "ssh+git"))
271                 return PROTO_SSH;
272         if (!strcmp(name, "file"))
273                 return PROTO_FILE;
274         die("I don't handle protocol '%s'", name);
275 }
276
277 #define STR_(s) # s
278 #define STR(s)  STR_(s)
279
280 static void get_host_and_port(char **host, const char **port)
281 {
282         char *colon, *end;
283
284         if (*host[0] == '[') {
285                 end = strchr(*host + 1, ']');
286                 if (end) {
287                         *end = 0;
288                         end++;
289                         (*host)++;
290                 } else
291                         end = *host;
292         } else
293                 end = *host;
294         colon = strchr(end, ':');
295
296         if (colon) {
297                 *colon = 0;
298                 *port = colon + 1;
299         }
300 }
301
302 static void enable_keepalive(int sockfd)
303 {
304         int ka = 1;
305
306         if (setsockopt(sockfd, SOL_SOCKET, SO_KEEPALIVE, &ka, sizeof(ka)) < 0)
307                 fprintf(stderr, "unable to set SO_KEEPALIVE on socket: %s\n",
308                         strerror(errno));
309 }
310
311 #ifndef NO_IPV6
312
313 static const char *ai_name(const struct addrinfo *ai)
314 {
315         static char addr[NI_MAXHOST];
316         if (getnameinfo(ai->ai_addr, ai->ai_addrlen, addr, sizeof(addr), NULL, 0,
317                         NI_NUMERICHOST) != 0)
318                 strcpy(addr, "(unknown)");
319
320         return addr;
321 }
322
323 /*
324  * Returns a connected socket() fd, or else die()s.
325  */
326 static int git_tcp_connect_sock(char *host, int flags)
327 {
328         struct strbuf error_message = STRBUF_INIT;
329         int sockfd = -1;
330         const char *port = STR(DEFAULT_GIT_PORT);
331         struct addrinfo hints, *ai0, *ai;
332         int gai;
333         int cnt = 0;
334
335         get_host_and_port(&host, &port);
336         if (!*port)
337                 port = "<none>";
338
339         memset(&hints, 0, sizeof(hints));
340         hints.ai_socktype = SOCK_STREAM;
341         hints.ai_protocol = IPPROTO_TCP;
342
343         if (flags & CONNECT_VERBOSE)
344                 fprintf(stderr, "Looking up %s ... ", host);
345
346         gai = getaddrinfo(host, port, &hints, &ai);
347         if (gai)
348                 die("Unable to look up %s (port %s) (%s)", host, port, gai_strerror(gai));
349
350         if (flags & CONNECT_VERBOSE)
351                 fprintf(stderr, "done.\nConnecting to %s (port %s) ... ", host, port);
352
353         for (ai0 = ai; ai; ai = ai->ai_next, cnt++) {
354                 sockfd = socket(ai->ai_family,
355                                 ai->ai_socktype, ai->ai_protocol);
356                 if ((sockfd < 0) ||
357                     (connect(sockfd, ai->ai_addr, ai->ai_addrlen) < 0)) {
358                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
359                                     host, cnt, ai_name(ai), strerror(errno));
360                         if (0 <= sockfd)
361                                 close(sockfd);
362                         sockfd = -1;
363                         continue;
364                 }
365                 if (flags & CONNECT_VERBOSE)
366                         fprintf(stderr, "%s ", ai_name(ai));
367                 break;
368         }
369
370         freeaddrinfo(ai0);
371
372         if (sockfd < 0)
373                 die("unable to connect to %s:\n%s", host, error_message.buf);
374
375         enable_keepalive(sockfd);
376
377         if (flags & CONNECT_VERBOSE)
378                 fprintf(stderr, "done.\n");
379
380         strbuf_release(&error_message);
381
382         return sockfd;
383 }
384
385 #else /* NO_IPV6 */
386
387 /*
388  * Returns a connected socket() fd, or else die()s.
389  */
390 static int git_tcp_connect_sock(char *host, int flags)
391 {
392         struct strbuf error_message = STRBUF_INIT;
393         int sockfd = -1;
394         const char *port = STR(DEFAULT_GIT_PORT);
395         char *ep;
396         struct hostent *he;
397         struct sockaddr_in sa;
398         char **ap;
399         unsigned int nport;
400         int cnt;
401
402         get_host_and_port(&host, &port);
403
404         if (flags & CONNECT_VERBOSE)
405                 fprintf(stderr, "Looking up %s ... ", host);
406
407         he = gethostbyname(host);
408         if (!he)
409                 die("Unable to look up %s (%s)", host, hstrerror(h_errno));
410         nport = strtoul(port, &ep, 10);
411         if ( ep == port || *ep ) {
412                 /* Not numeric */
413                 struct servent *se = getservbyname(port,"tcp");
414                 if ( !se )
415                         die("Unknown port %s", port);
416                 nport = se->s_port;
417         }
418
419         if (flags & CONNECT_VERBOSE)
420                 fprintf(stderr, "done.\nConnecting to %s (port %s) ... ", host, port);
421
422         for (cnt = 0, ap = he->h_addr_list; *ap; ap++, cnt++) {
423                 memset(&sa, 0, sizeof sa);
424                 sa.sin_family = he->h_addrtype;
425                 sa.sin_port = htons(nport);
426                 memcpy(&sa.sin_addr, *ap, he->h_length);
427
428                 sockfd = socket(he->h_addrtype, SOCK_STREAM, 0);
429                 if ((sockfd < 0) ||
430                     connect(sockfd, (struct sockaddr *)&sa, sizeof sa) < 0) {
431                         strbuf_addf(&error_message, "%s[%d: %s]: errno=%s\n",
432                                 host,
433                                 cnt,
434                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr),
435                                 strerror(errno));
436                         if (0 <= sockfd)
437                                 close(sockfd);
438                         sockfd = -1;
439                         continue;
440                 }
441                 if (flags & CONNECT_VERBOSE)
442                         fprintf(stderr, "%s ",
443                                 inet_ntoa(*(struct in_addr *)&sa.sin_addr));
444                 break;
445         }
446
447         if (sockfd < 0)
448                 die("unable to connect to %s:\n%s", host, error_message.buf);
449
450         enable_keepalive(sockfd);
451
452         if (flags & CONNECT_VERBOSE)
453                 fprintf(stderr, "done.\n");
454
455         return sockfd;
456 }
457
458 #endif /* NO_IPV6 */
459
460
461 static void git_tcp_connect(int fd[2], char *host, int flags)
462 {
463         int sockfd = git_tcp_connect_sock(host, flags);
464
465         fd[0] = sockfd;
466         fd[1] = dup(sockfd);
467 }
468
469
470 static char *git_proxy_command;
471
472 static int git_proxy_command_options(const char *var, const char *value,
473                 void *cb)
474 {
475         if (!strcmp(var, "core.gitproxy")) {
476                 const char *for_pos;
477                 int matchlen = -1;
478                 int hostlen;
479                 const char *rhost_name = cb;
480                 int rhost_len = strlen(rhost_name);
481
482                 if (git_proxy_command)
483                         return 0;
484                 if (!value)
485                         return config_error_nonbool(var);
486                 /* [core]
487                  * ;# matches www.kernel.org as well
488                  * gitproxy = netcatter-1 for kernel.org
489                  * gitproxy = netcatter-2 for sample.xz
490                  * gitproxy = netcatter-default
491                  */
492                 for_pos = strstr(value, " for ");
493                 if (!for_pos)
494                         /* matches everybody */
495                         matchlen = strlen(value);
496                 else {
497                         hostlen = strlen(for_pos + 5);
498                         if (rhost_len < hostlen)
499                                 matchlen = -1;
500                         else if (!strncmp(for_pos + 5,
501                                           rhost_name + rhost_len - hostlen,
502                                           hostlen) &&
503                                  ((rhost_len == hostlen) ||
504                                   rhost_name[rhost_len - hostlen -1] == '.'))
505                                 matchlen = for_pos - value;
506                         else
507                                 matchlen = -1;
508                 }
509                 if (0 <= matchlen) {
510                         /* core.gitproxy = none for kernel.org */
511                         if (matchlen == 4 &&
512                             !memcmp(value, "none", 4))
513                                 matchlen = 0;
514                         git_proxy_command = xmemdupz(value, matchlen);
515                 }
516                 return 0;
517         }
518
519         return git_default_config(var, value, cb);
520 }
521
522 static int git_use_proxy(const char *host)
523 {
524         git_proxy_command = getenv("GIT_PROXY_COMMAND");
525         git_config(git_proxy_command_options, (void*)host);
526         return (git_proxy_command && *git_proxy_command);
527 }
528
529 static struct child_process *git_proxy_connect(int fd[2], char *host)
530 {
531         const char *port = STR(DEFAULT_GIT_PORT);
532         struct child_process *proxy;
533
534         get_host_and_port(&host, &port);
535
536         proxy = xmalloc(sizeof(*proxy));
537         child_process_init(proxy);
538         argv_array_push(&proxy->args, git_proxy_command);
539         argv_array_push(&proxy->args, host);
540         argv_array_push(&proxy->args, port);
541         proxy->in = -1;
542         proxy->out = -1;
543         if (start_command(proxy))
544                 die("cannot start proxy %s", git_proxy_command);
545         fd[0] = proxy->out; /* read from proxy stdout */
546         fd[1] = proxy->in;  /* write to proxy stdin */
547         return proxy;
548 }
549
550 static const char *get_port_numeric(const char *p)
551 {
552         char *end;
553         if (p) {
554                 long port = strtol(p + 1, &end, 10);
555                 if (end != p + 1 && *end == '\0' && 0 <= port && port < 65536) {
556                         return p;
557                 }
558         }
559
560         return NULL;
561 }
562
563 /*
564  * Extract protocol and relevant parts from the specified connection URL.
565  * The caller must free() the returned strings.
566  */
567 static enum protocol parse_connect_url(const char *url_orig, char **ret_host,
568                                        char **ret_path)
569 {
570         char *url;
571         char *host, *path;
572         char *end;
573         int separator = '/';
574         enum protocol protocol = PROTO_LOCAL;
575
576         if (is_url(url_orig))
577                 url = url_decode(url_orig);
578         else
579                 url = xstrdup(url_orig);
580
581         host = strstr(url, "://");
582         if (host) {
583                 *host = '\0';
584                 protocol = get_protocol(url);
585                 host += 3;
586         } else {
587                 host = url;
588                 if (!url_is_local_not_ssh(url)) {
589                         protocol = PROTO_SSH;
590                         separator = ':';
591                 }
592         }
593
594         /*
595          * Don't do destructive transforms as protocol code does
596          * '[]' unwrapping in get_host_and_port()
597          */
598         if (host[0] == '[') {
599                 end = strchr(host + 1, ']');
600                 if (end) {
601                         end++;
602                 } else
603                         end = host;
604         } else
605                 end = host;
606
607         if (protocol == PROTO_LOCAL)
608                 path = end;
609         else if (protocol == PROTO_FILE && has_dos_drive_prefix(end))
610                 path = end; /* "file://$(pwd)" may be "file://C:/projects/repo" */
611         else
612                 path = strchr(end, separator);
613
614         if (!path || !*path)
615                 die("No path specified. See 'man git-pull' for valid url syntax");
616
617         /*
618          * null-terminate hostname and point path to ~ for URL's like this:
619          *    ssh://host.xz/~user/repo
620          */
621
622         end = path; /* Need to \0 terminate host here */
623         if (separator == ':')
624                 path++; /* path starts after ':' */
625         if (protocol == PROTO_GIT || protocol == PROTO_SSH) {
626                 if (path[1] == '~')
627                         path++;
628         }
629
630         path = xstrdup(path);
631         *end = '\0';
632
633         *ret_host = xstrdup(host);
634         *ret_path = path;
635         free(url);
636         return protocol;
637 }
638
639 static struct child_process no_fork = CHILD_PROCESS_INIT;
640
641 /*
642  * This returns a dummy child_process if the transport protocol does not
643  * need fork(2), or a struct child_process object if it does.  Once done,
644  * finish the connection with finish_connect() with the value returned from
645  * this function (it is safe to call finish_connect() with NULL to support
646  * the former case).
647  *
648  * If it returns, the connect is successful; it just dies on errors (this
649  * will hopefully be changed in a libification effort, to return NULL when
650  * the connection failed).
651  */
652 struct child_process *git_connect(int fd[2], const char *url,
653                                   const char *prog, int flags)
654 {
655         char *hostandport, *path;
656         struct child_process *conn = &no_fork;
657         enum protocol protocol;
658         struct strbuf cmd = STRBUF_INIT;
659
660         /* Without this we cannot rely on waitpid() to tell
661          * what happened to our children.
662          */
663         signal(SIGCHLD, SIG_DFL);
664
665         protocol = parse_connect_url(url, &hostandport, &path);
666         if (flags & CONNECT_DIAG_URL) {
667                 printf("Diag: url=%s\n", url ? url : "NULL");
668                 printf("Diag: protocol=%s\n", prot_name(protocol));
669                 printf("Diag: hostandport=%s\n", hostandport ? hostandport : "NULL");
670                 printf("Diag: path=%s\n", path ? path : "NULL");
671                 conn = NULL;
672         } else if (protocol == PROTO_GIT) {
673                 /*
674                  * Set up virtual host information based on where we will
675                  * connect, unless the user has overridden us in
676                  * the environment.
677                  */
678                 char *target_host = getenv("GIT_OVERRIDE_VIRTUAL_HOST");
679                 if (target_host)
680                         target_host = xstrdup(target_host);
681                 else
682                         target_host = xstrdup(hostandport);
683
684                 /* These underlying connection commands die() if they
685                  * cannot connect.
686                  */
687                 if (git_use_proxy(hostandport))
688                         conn = git_proxy_connect(fd, hostandport);
689                 else
690                         git_tcp_connect(fd, hostandport, flags);
691                 /*
692                  * Separate original protocol components prog and path
693                  * from extended host header with a NUL byte.
694                  *
695                  * Note: Do not add any other headers here!  Doing so
696                  * will cause older git-daemon servers to crash.
697                  */
698                 packet_write(fd[1],
699                              "%s %s%chost=%s%c",
700                              prog, path, 0,
701                              target_host, 0);
702                 free(target_host);
703         } else {
704                 conn = xmalloc(sizeof(*conn));
705                 child_process_init(conn);
706
707                 strbuf_addstr(&cmd, prog);
708                 strbuf_addch(&cmd, ' ');
709                 sq_quote_buf(&cmd, path);
710
711                 conn->in = conn->out = -1;
712                 if (protocol == PROTO_SSH) {
713                         const char *ssh;
714                         int putty;
715                         char *ssh_host = hostandport;
716                         const char *port = NULL;
717                         get_host_and_port(&ssh_host, &port);
718                         port = get_port_numeric(port);
719
720                         ssh = getenv("GIT_SSH_COMMAND");
721                         if (ssh) {
722                                 conn->use_shell = 1;
723                                 putty = 0;
724                         } else {
725                                 ssh = getenv("GIT_SSH");
726                                 if (!ssh)
727                                         ssh = "ssh";
728                                 putty = !!strcasestr(ssh, "plink");
729                         }
730
731                         argv_array_push(&conn->args, ssh);
732                         if (putty && !strcasestr(ssh, "tortoiseplink"))
733                                 argv_array_push(&conn->args, "-batch");
734                         if (port) {
735                                 /* P is for PuTTY, p is for OpenSSH */
736                                 argv_array_push(&conn->args, putty ? "-P" : "-p");
737                                 argv_array_push(&conn->args, port);
738                         }
739                         argv_array_push(&conn->args, ssh_host);
740                 } else {
741                         /* remove repo-local variables from the environment */
742                         conn->env = local_repo_env;
743                         conn->use_shell = 1;
744                 }
745                 argv_array_push(&conn->args, cmd.buf);
746
747                 if (start_command(conn))
748                         die("unable to fork");
749
750                 fd[0] = conn->out; /* read from child's stdout */
751                 fd[1] = conn->in;  /* write to child's stdin */
752                 strbuf_release(&cmd);
753         }
754         free(hostandport);
755         free(path);
756         return conn;
757 }
758
759 int git_connection_is_socket(struct child_process *conn)
760 {
761         return conn == &no_fork;
762 }
763
764 int finish_connect(struct child_process *conn)
765 {
766         int code;
767         if (!conn || git_connection_is_socket(conn))
768                 return 0;
769
770         code = finish_command(conn);
771         free(conn);
772         return code;
773 }