Merge branch 'jc/fsck-nul-in-commit' into maint
[git] / http-backend.c
1 #include "cache.h"
2 #include "refs.h"
3 #include "pkt-line.h"
4 #include "object.h"
5 #include "tag.h"
6 #include "exec_cmd.h"
7 #include "run-command.h"
8 #include "string-list.h"
9 #include "url.h"
10 #include "argv-array.h"
11
12 static const char content_type[] = "Content-Type";
13 static const char content_length[] = "Content-Length";
14 static const char last_modified[] = "Last-Modified";
15 static int getanyfile = 1;
16 static unsigned long max_request_buffer = 10 * 1024 * 1024;
17
18 static struct string_list *query_params;
19
20 struct rpc_service {
21         const char *name;
22         const char *config_name;
23         unsigned buffer_input : 1;
24         signed enabled : 2;
25 };
26
27 static struct rpc_service rpc_service[] = {
28         { "upload-pack", "uploadpack", 1, 1 },
29         { "receive-pack", "receivepack", 0, -1 },
30 };
31
32 static struct string_list *get_parameters(void)
33 {
34         if (!query_params) {
35                 const char *query = getenv("QUERY_STRING");
36
37                 query_params = xcalloc(1, sizeof(*query_params));
38                 while (query && *query) {
39                         char *name = url_decode_parameter_name(&query);
40                         char *value = url_decode_parameter_value(&query);
41                         struct string_list_item *i;
42
43                         i = string_list_lookup(query_params, name);
44                         if (!i)
45                                 i = string_list_insert(query_params, name);
46                         else
47                                 free(i->util);
48                         i->util = value;
49                 }
50         }
51         return query_params;
52 }
53
54 static const char *get_parameter(const char *name)
55 {
56         struct string_list_item *i;
57         i = string_list_lookup(get_parameters(), name);
58         return i ? i->util : NULL;
59 }
60
61 __attribute__((format (printf, 2, 3)))
62 static void format_write(int fd, const char *fmt, ...)
63 {
64         static char buffer[1024];
65
66         va_list args;
67         unsigned n;
68
69         va_start(args, fmt);
70         n = vsnprintf(buffer, sizeof(buffer), fmt, args);
71         va_end(args);
72         if (n >= sizeof(buffer))
73                 die("protocol error: impossibly long line");
74
75         write_or_die(fd, buffer, n);
76 }
77
78 static void http_status(unsigned code, const char *msg)
79 {
80         format_write(1, "Status: %u %s\r\n", code, msg);
81 }
82
83 static void hdr_str(const char *name, const char *value)
84 {
85         format_write(1, "%s: %s\r\n", name, value);
86 }
87
88 static void hdr_int(const char *name, uintmax_t value)
89 {
90         format_write(1, "%s: %" PRIuMAX "\r\n", name, value);
91 }
92
93 static void hdr_date(const char *name, unsigned long when)
94 {
95         const char *value = show_date(when, 0, DATE_MODE(RFC2822));
96         hdr_str(name, value);
97 }
98
99 static void hdr_nocache(void)
100 {
101         hdr_str("Expires", "Fri, 01 Jan 1980 00:00:00 GMT");
102         hdr_str("Pragma", "no-cache");
103         hdr_str("Cache-Control", "no-cache, max-age=0, must-revalidate");
104 }
105
106 static void hdr_cache_forever(void)
107 {
108         unsigned long now = time(NULL);
109         hdr_date("Date", now);
110         hdr_date("Expires", now + 31536000);
111         hdr_str("Cache-Control", "public, max-age=31536000");
112 }
113
114 static void end_headers(void)
115 {
116         write_or_die(1, "\r\n", 2);
117 }
118
119 __attribute__((format (printf, 1, 2)))
120 static NORETURN void not_found(const char *err, ...)
121 {
122         va_list params;
123
124         http_status(404, "Not Found");
125         hdr_nocache();
126         end_headers();
127
128         va_start(params, err);
129         if (err && *err)
130                 vfprintf(stderr, err, params);
131         va_end(params);
132         exit(0);
133 }
134
135 __attribute__((format (printf, 1, 2)))
136 static NORETURN void forbidden(const char *err, ...)
137 {
138         va_list params;
139
140         http_status(403, "Forbidden");
141         hdr_nocache();
142         end_headers();
143
144         va_start(params, err);
145         if (err && *err)
146                 vfprintf(stderr, err, params);
147         va_end(params);
148         exit(0);
149 }
150
151 static void select_getanyfile(void)
152 {
153         if (!getanyfile)
154                 forbidden("Unsupported service: getanyfile");
155 }
156
157 static void send_strbuf(const char *type, struct strbuf *buf)
158 {
159         hdr_int(content_length, buf->len);
160         hdr_str(content_type, type);
161         end_headers();
162         write_or_die(1, buf->buf, buf->len);
163 }
164
165 static void send_local_file(const char *the_type, const char *name)
166 {
167         char *p = git_pathdup("%s", name);
168         size_t buf_alloc = 8192;
169         char *buf = xmalloc(buf_alloc);
170         int fd;
171         struct stat sb;
172
173         fd = open(p, O_RDONLY);
174         if (fd < 0)
175                 not_found("Cannot open '%s': %s", p, strerror(errno));
176         if (fstat(fd, &sb) < 0)
177                 die_errno("Cannot stat '%s'", p);
178
179         hdr_int(content_length, sb.st_size);
180         hdr_str(content_type, the_type);
181         hdr_date(last_modified, sb.st_mtime);
182         end_headers();
183
184         for (;;) {
185                 ssize_t n = xread(fd, buf, buf_alloc);
186                 if (n < 0)
187                         die_errno("Cannot read '%s'", p);
188                 if (!n)
189                         break;
190                 write_or_die(1, buf, n);
191         }
192         close(fd);
193         free(buf);
194         free(p);
195 }
196
197 static void get_text_file(char *name)
198 {
199         select_getanyfile();
200         hdr_nocache();
201         send_local_file("text/plain", name);
202 }
203
204 static void get_loose_object(char *name)
205 {
206         select_getanyfile();
207         hdr_cache_forever();
208         send_local_file("application/x-git-loose-object", name);
209 }
210
211 static void get_pack_file(char *name)
212 {
213         select_getanyfile();
214         hdr_cache_forever();
215         send_local_file("application/x-git-packed-objects", name);
216 }
217
218 static void get_idx_file(char *name)
219 {
220         select_getanyfile();
221         hdr_cache_forever();
222         send_local_file("application/x-git-packed-objects-toc", name);
223 }
224
225 static void http_config(void)
226 {
227         int i, value = 0;
228         struct strbuf var = STRBUF_INIT;
229
230         git_config_get_bool("http.getanyfile", &getanyfile);
231         git_config_get_ulong("http.maxrequestbuffer", &max_request_buffer);
232
233         for (i = 0; i < ARRAY_SIZE(rpc_service); i++) {
234                 struct rpc_service *svc = &rpc_service[i];
235                 strbuf_addf(&var, "http.%s", svc->config_name);
236                 if (!git_config_get_bool(var.buf, &value))
237                         svc->enabled = value;
238                 strbuf_reset(&var);
239         }
240
241         strbuf_release(&var);
242 }
243
244 static struct rpc_service *select_service(const char *name)
245 {
246         const char *svc_name;
247         struct rpc_service *svc = NULL;
248         int i;
249
250         if (!skip_prefix(name, "git-", &svc_name))
251                 forbidden("Unsupported service: '%s'", name);
252
253         for (i = 0; i < ARRAY_SIZE(rpc_service); i++) {
254                 struct rpc_service *s = &rpc_service[i];
255                 if (!strcmp(s->name, svc_name)) {
256                         svc = s;
257                         break;
258                 }
259         }
260
261         if (!svc)
262                 forbidden("Unsupported service: '%s'", name);
263
264         if (svc->enabled < 0) {
265                 const char *user = getenv("REMOTE_USER");
266                 svc->enabled = (user && *user) ? 1 : 0;
267         }
268         if (!svc->enabled)
269                 forbidden("Service not enabled: '%s'", svc->name);
270         return svc;
271 }
272
273 /*
274  * This is basically strbuf_read(), except that if we
275  * hit max_request_buffer we die (we'd rather reject a
276  * maliciously large request than chew up infinite memory).
277  */
278 static ssize_t read_request(int fd, unsigned char **out)
279 {
280         size_t len = 0, alloc = 8192;
281         unsigned char *buf = xmalloc(alloc);
282
283         if (max_request_buffer < alloc)
284                 max_request_buffer = alloc;
285
286         while (1) {
287                 ssize_t cnt;
288
289                 cnt = read_in_full(fd, buf + len, alloc - len);
290                 if (cnt < 0) {
291                         free(buf);
292                         return -1;
293                 }
294
295                 /* partial read from read_in_full means we hit EOF */
296                 len += cnt;
297                 if (len < alloc) {
298                         *out = buf;
299                         return len;
300                 }
301
302                 /* otherwise, grow and try again (if we can) */
303                 if (alloc == max_request_buffer)
304                         die("request was larger than our maximum size (%lu);"
305                             " try setting GIT_HTTP_MAX_REQUEST_BUFFER",
306                             max_request_buffer);
307
308                 alloc = alloc_nr(alloc);
309                 if (alloc > max_request_buffer)
310                         alloc = max_request_buffer;
311                 REALLOC_ARRAY(buf, alloc);
312         }
313 }
314
315 static void inflate_request(const char *prog_name, int out, int buffer_input)
316 {
317         git_zstream stream;
318         unsigned char *full_request = NULL;
319         unsigned char in_buf[8192];
320         unsigned char out_buf[8192];
321         unsigned long cnt = 0;
322
323         memset(&stream, 0, sizeof(stream));
324         git_inflate_init_gzip_only(&stream);
325
326         while (1) {
327                 ssize_t n;
328
329                 if (buffer_input) {
330                         if (full_request)
331                                 n = 0; /* nothing left to read */
332                         else
333                                 n = read_request(0, &full_request);
334                         stream.next_in = full_request;
335                 } else {
336                         n = xread(0, in_buf, sizeof(in_buf));
337                         stream.next_in = in_buf;
338                 }
339
340                 if (n <= 0)
341                         die("request ended in the middle of the gzip stream");
342                 stream.avail_in = n;
343
344                 while (0 < stream.avail_in) {
345                         int ret;
346
347                         stream.next_out = out_buf;
348                         stream.avail_out = sizeof(out_buf);
349
350                         ret = git_inflate(&stream, Z_NO_FLUSH);
351                         if (ret != Z_OK && ret != Z_STREAM_END)
352                                 die("zlib error inflating request, result %d", ret);
353
354                         n = stream.total_out - cnt;
355                         if (write_in_full(out, out_buf, n) != n)
356                                 die("%s aborted reading request", prog_name);
357                         cnt += n;
358
359                         if (ret == Z_STREAM_END)
360                                 goto done;
361                 }
362         }
363
364 done:
365         git_inflate_end(&stream);
366         close(out);
367         free(full_request);
368 }
369
370 static void copy_request(const char *prog_name, int out)
371 {
372         unsigned char *buf;
373         ssize_t n = read_request(0, &buf);
374         if (n < 0)
375                 die_errno("error reading request body");
376         if (write_in_full(out, buf, n) != n)
377                 die("%s aborted reading request", prog_name);
378         close(out);
379         free(buf);
380 }
381
382 static void run_service(const char **argv, int buffer_input)
383 {
384         const char *encoding = getenv("HTTP_CONTENT_ENCODING");
385         const char *user = getenv("REMOTE_USER");
386         const char *host = getenv("REMOTE_ADDR");
387         int gzipped_request = 0;
388         struct child_process cld = CHILD_PROCESS_INIT;
389
390         if (encoding && !strcmp(encoding, "gzip"))
391                 gzipped_request = 1;
392         else if (encoding && !strcmp(encoding, "x-gzip"))
393                 gzipped_request = 1;
394
395         if (!user || !*user)
396                 user = "anonymous";
397         if (!host || !*host)
398                 host = "(none)";
399
400         if (!getenv("GIT_COMMITTER_NAME"))
401                 argv_array_pushf(&cld.env_array, "GIT_COMMITTER_NAME=%s", user);
402         if (!getenv("GIT_COMMITTER_EMAIL"))
403                 argv_array_pushf(&cld.env_array,
404                                  "GIT_COMMITTER_EMAIL=%s@http.%s", user, host);
405
406         cld.argv = argv;
407         if (buffer_input || gzipped_request)
408                 cld.in = -1;
409         cld.git_cmd = 1;
410         if (start_command(&cld))
411                 exit(1);
412
413         close(1);
414         if (gzipped_request)
415                 inflate_request(argv[0], cld.in, buffer_input);
416         else if (buffer_input)
417                 copy_request(argv[0], cld.in);
418         else
419                 close(0);
420
421         if (finish_command(&cld))
422                 exit(1);
423 }
424
425 static int show_text_ref(const char *name, const struct object_id *oid,
426                          int flag, void *cb_data)
427 {
428         const char *name_nons = strip_namespace(name);
429         struct strbuf *buf = cb_data;
430         struct object *o = parse_object(oid->hash);
431         if (!o)
432                 return 0;
433
434         strbuf_addf(buf, "%s\t%s\n", oid_to_hex(oid), name_nons);
435         if (o->type == OBJ_TAG) {
436                 o = deref_tag(o, name, 0);
437                 if (!o)
438                         return 0;
439                 strbuf_addf(buf, "%s\t%s^{}\n", oid_to_hex(&o->oid),
440                             name_nons);
441         }
442         return 0;
443 }
444
445 static void get_info_refs(char *arg)
446 {
447         const char *service_name = get_parameter("service");
448         struct strbuf buf = STRBUF_INIT;
449
450         hdr_nocache();
451
452         if (service_name) {
453                 const char *argv[] = {NULL /* service name */,
454                         "--stateless-rpc", "--advertise-refs",
455                         ".", NULL};
456                 struct rpc_service *svc = select_service(service_name);
457
458                 strbuf_addf(&buf, "application/x-git-%s-advertisement",
459                         svc->name);
460                 hdr_str(content_type, buf.buf);
461                 end_headers();
462
463                 packet_write(1, "# service=git-%s\n", svc->name);
464                 packet_flush(1);
465
466                 argv[0] = svc->name;
467                 run_service(argv, 0);
468
469         } else {
470                 select_getanyfile();
471                 for_each_namespaced_ref(show_text_ref, &buf);
472                 send_strbuf("text/plain", &buf);
473         }
474         strbuf_release(&buf);
475 }
476
477 static int show_head_ref(const char *refname, const struct object_id *oid,
478                          int flag, void *cb_data)
479 {
480         struct strbuf *buf = cb_data;
481
482         if (flag & REF_ISSYMREF) {
483                 struct object_id unused;
484                 const char *target = resolve_ref_unsafe(refname,
485                                                         RESOLVE_REF_READING,
486                                                         unused.hash, NULL);
487                 const char *target_nons = strip_namespace(target);
488
489                 strbuf_addf(buf, "ref: %s\n", target_nons);
490         } else {
491                 strbuf_addf(buf, "%s\n", oid_to_hex(oid));
492         }
493
494         return 0;
495 }
496
497 static void get_head(char *arg)
498 {
499         struct strbuf buf = STRBUF_INIT;
500
501         select_getanyfile();
502         head_ref_namespaced(show_head_ref, &buf);
503         send_strbuf("text/plain", &buf);
504         strbuf_release(&buf);
505 }
506
507 static void get_info_packs(char *arg)
508 {
509         size_t objdirlen = strlen(get_object_directory());
510         struct strbuf buf = STRBUF_INIT;
511         struct packed_git *p;
512         size_t cnt = 0;
513
514         select_getanyfile();
515         prepare_packed_git();
516         for (p = packed_git; p; p = p->next) {
517                 if (p->pack_local)
518                         cnt++;
519         }
520
521         strbuf_grow(&buf, cnt * 53 + 2);
522         for (p = packed_git; p; p = p->next) {
523                 if (p->pack_local)
524                         strbuf_addf(&buf, "P %s\n", p->pack_name + objdirlen + 6);
525         }
526         strbuf_addch(&buf, '\n');
527
528         hdr_nocache();
529         send_strbuf("text/plain; charset=utf-8", &buf);
530         strbuf_release(&buf);
531 }
532
533 static void check_content_type(const char *accepted_type)
534 {
535         const char *actual_type = getenv("CONTENT_TYPE");
536
537         if (!actual_type)
538                 actual_type = "";
539
540         if (strcmp(actual_type, accepted_type)) {
541                 http_status(415, "Unsupported Media Type");
542                 hdr_nocache();
543                 end_headers();
544                 format_write(1,
545                         "Expected POST with Content-Type '%s',"
546                         " but received '%s' instead.\n",
547                         accepted_type, actual_type);
548                 exit(0);
549         }
550 }
551
552 static void service_rpc(char *service_name)
553 {
554         const char *argv[] = {NULL, "--stateless-rpc", ".", NULL};
555         struct rpc_service *svc = select_service(service_name);
556         struct strbuf buf = STRBUF_INIT;
557
558         strbuf_reset(&buf);
559         strbuf_addf(&buf, "application/x-git-%s-request", svc->name);
560         check_content_type(buf.buf);
561
562         hdr_nocache();
563
564         strbuf_reset(&buf);
565         strbuf_addf(&buf, "application/x-git-%s-result", svc->name);
566         hdr_str(content_type, buf.buf);
567
568         end_headers();
569
570         argv[0] = svc->name;
571         run_service(argv, svc->buffer_input);
572         strbuf_release(&buf);
573 }
574
575 static int dead;
576 static NORETURN void die_webcgi(const char *err, va_list params)
577 {
578         if (dead <= 1) {
579                 vreportf("fatal: ", err, params);
580
581                 http_status(500, "Internal Server Error");
582                 hdr_nocache();
583                 end_headers();
584         }
585         exit(0); /* we successfully reported a failure ;-) */
586 }
587
588 static int die_webcgi_recursing(void)
589 {
590         return dead++ > 1;
591 }
592
593 static char* getdir(void)
594 {
595         struct strbuf buf = STRBUF_INIT;
596         char *pathinfo = getenv("PATH_INFO");
597         char *root = getenv("GIT_PROJECT_ROOT");
598         char *path = getenv("PATH_TRANSLATED");
599
600         if (root && *root) {
601                 if (!pathinfo || !*pathinfo)
602                         die("GIT_PROJECT_ROOT is set but PATH_INFO is not");
603                 if (daemon_avoid_alias(pathinfo))
604                         die("'%s': aliased", pathinfo);
605                 end_url_with_slash(&buf, root);
606                 if (pathinfo[0] == '/')
607                         pathinfo++;
608                 strbuf_addstr(&buf, pathinfo);
609                 return strbuf_detach(&buf, NULL);
610         } else if (path && *path) {
611                 return xstrdup(path);
612         } else
613                 die("No GIT_PROJECT_ROOT or PATH_TRANSLATED from server");
614         return NULL;
615 }
616
617 static struct service_cmd {
618         const char *method;
619         const char *pattern;
620         void (*imp)(char *);
621 } services[] = {
622         {"GET", "/HEAD$", get_head},
623         {"GET", "/info/refs$", get_info_refs},
624         {"GET", "/objects/info/alternates$", get_text_file},
625         {"GET", "/objects/info/http-alternates$", get_text_file},
626         {"GET", "/objects/info/packs$", get_info_packs},
627         {"GET", "/objects/[0-9a-f]{2}/[0-9a-f]{38}$", get_loose_object},
628         {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.pack$", get_pack_file},
629         {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.idx$", get_idx_file},
630
631         {"POST", "/git-upload-pack$", service_rpc},
632         {"POST", "/git-receive-pack$", service_rpc}
633 };
634
635 int main(int argc, char **argv)
636 {
637         char *method = getenv("REQUEST_METHOD");
638         char *dir;
639         struct service_cmd *cmd = NULL;
640         char *cmd_arg = NULL;
641         int i;
642
643         git_setup_gettext();
644
645         git_extract_argv0_path(argv[0]);
646         set_die_routine(die_webcgi);
647         set_die_is_recursing_routine(die_webcgi_recursing);
648
649         if (!method)
650                 die("No REQUEST_METHOD from server");
651         if (!strcmp(method, "HEAD"))
652                 method = "GET";
653         dir = getdir();
654
655         for (i = 0; i < ARRAY_SIZE(services); i++) {
656                 struct service_cmd *c = &services[i];
657                 regex_t re;
658                 regmatch_t out[1];
659
660                 if (regcomp(&re, c->pattern, REG_EXTENDED))
661                         die("Bogus regex in service table: %s", c->pattern);
662                 if (!regexec(&re, dir, 1, out, 0)) {
663                         size_t n;
664
665                         if (strcmp(method, c->method)) {
666                                 const char *proto = getenv("SERVER_PROTOCOL");
667                                 if (proto && !strcmp(proto, "HTTP/1.1")) {
668                                         http_status(405, "Method Not Allowed");
669                                         hdr_str("Allow", !strcmp(c->method, "GET") ?
670                                                 "GET, HEAD" : c->method);
671                                 } else
672                                         http_status(400, "Bad Request");
673                                 hdr_nocache();
674                                 end_headers();
675                                 return 0;
676                         }
677
678                         cmd = c;
679                         n = out[0].rm_eo - out[0].rm_so;
680                         cmd_arg = xmemdupz(dir + out[0].rm_so + 1, n - 1);
681                         dir[out[0].rm_so] = 0;
682                         break;
683                 }
684                 regfree(&re);
685         }
686
687         if (!cmd)
688                 not_found("Request not supported: '%s'", dir);
689
690         setup_path();
691         if (!enter_repo(dir, 0))
692                 not_found("Not a git repository: '%s'", dir);
693         if (!getenv("GIT_HTTP_EXPORT_ALL") &&
694             access("git-daemon-export-ok", F_OK) )
695                 not_found("Repository not exported: '%s'", dir);
696
697         http_config();
698         max_request_buffer = git_env_ulong("GIT_HTTP_MAX_REQUEST_BUFFER",
699                                            max_request_buffer);
700
701         cmd->imp(cmd_arg);
702         return 0;
703 }