apply: demonstrate a problem applying svn diffs
[git] / http-backend.c
1 #include "cache.h"
2 #include "config.h"
3 #include "refs.h"
4 #include "pkt-line.h"
5 #include "object.h"
6 #include "tag.h"
7 #include "exec_cmd.h"
8 #include "run-command.h"
9 #include "string-list.h"
10 #include "url.h"
11 #include "argv-array.h"
12
13 static const char content_type[] = "Content-Type";
14 static const char content_length[] = "Content-Length";
15 static const char last_modified[] = "Last-Modified";
16 static int getanyfile = 1;
17 static unsigned long max_request_buffer = 10 * 1024 * 1024;
18
19 static struct string_list *query_params;
20
21 struct rpc_service {
22         const char *name;
23         const char *config_name;
24         unsigned buffer_input : 1;
25         signed enabled : 2;
26 };
27
28 static struct rpc_service rpc_service[] = {
29         { "upload-pack", "uploadpack", 1, 1 },
30         { "receive-pack", "receivepack", 0, -1 },
31 };
32
33 static struct string_list *get_parameters(void)
34 {
35         if (!query_params) {
36                 const char *query = getenv("QUERY_STRING");
37
38                 query_params = xcalloc(1, sizeof(*query_params));
39                 while (query && *query) {
40                         char *name = url_decode_parameter_name(&query);
41                         char *value = url_decode_parameter_value(&query);
42                         struct string_list_item *i;
43
44                         i = string_list_lookup(query_params, name);
45                         if (!i)
46                                 i = string_list_insert(query_params, name);
47                         else
48                                 free(i->util);
49                         i->util = value;
50                 }
51         }
52         return query_params;
53 }
54
55 static const char *get_parameter(const char *name)
56 {
57         struct string_list_item *i;
58         i = string_list_lookup(get_parameters(), name);
59         return i ? i->util : NULL;
60 }
61
62 __attribute__((format (printf, 2, 3)))
63 static void format_write(int fd, const char *fmt, ...)
64 {
65         static char buffer[1024];
66
67         va_list args;
68         unsigned n;
69
70         va_start(args, fmt);
71         n = vsnprintf(buffer, sizeof(buffer), fmt, args);
72         va_end(args);
73         if (n >= sizeof(buffer))
74                 die("protocol error: impossibly long line");
75
76         write_or_die(fd, buffer, n);
77 }
78
79 static void http_status(struct strbuf *hdr, unsigned code, const char *msg)
80 {
81         strbuf_addf(hdr, "Status: %u %s\r\n", code, msg);
82 }
83
84 static void hdr_str(struct strbuf *hdr, const char *name, const char *value)
85 {
86         strbuf_addf(hdr, "%s: %s\r\n", name, value);
87 }
88
89 static void hdr_int(struct strbuf *hdr, const char *name, uintmax_t value)
90 {
91         strbuf_addf(hdr, "%s: %" PRIuMAX "\r\n", name, value);
92 }
93
94 static void hdr_date(struct strbuf *hdr, const char *name, timestamp_t when)
95 {
96         const char *value = show_date(when, 0, DATE_MODE(RFC2822));
97         hdr_str(hdr, name, value);
98 }
99
100 static void hdr_nocache(struct strbuf *hdr)
101 {
102         hdr_str(hdr, "Expires", "Fri, 01 Jan 1980 00:00:00 GMT");
103         hdr_str(hdr, "Pragma", "no-cache");
104         hdr_str(hdr, "Cache-Control", "no-cache, max-age=0, must-revalidate");
105 }
106
107 static void hdr_cache_forever(struct strbuf *hdr)
108 {
109         timestamp_t now = time(NULL);
110         hdr_date(hdr, "Date", now);
111         hdr_date(hdr, "Expires", now + 31536000);
112         hdr_str(hdr, "Cache-Control", "public, max-age=31536000");
113 }
114
115 static void end_headers(struct strbuf *hdr)
116 {
117         strbuf_add(hdr, "\r\n", 2);
118         write_or_die(1, hdr->buf, hdr->len);
119         strbuf_release(hdr);
120 }
121
122 __attribute__((format (printf, 2, 3)))
123 static NORETURN void not_found(struct strbuf *hdr, const char *err, ...)
124 {
125         va_list params;
126
127         http_status(hdr, 404, "Not Found");
128         hdr_nocache(hdr);
129         end_headers(hdr);
130
131         va_start(params, err);
132         if (err && *err)
133                 vfprintf(stderr, err, params);
134         va_end(params);
135         exit(0);
136 }
137
138 __attribute__((format (printf, 2, 3)))
139 static NORETURN void forbidden(struct strbuf *hdr, const char *err, ...)
140 {
141         va_list params;
142
143         http_status(hdr, 403, "Forbidden");
144         hdr_nocache(hdr);
145         end_headers(hdr);
146
147         va_start(params, err);
148         if (err && *err)
149                 vfprintf(stderr, err, params);
150         va_end(params);
151         exit(0);
152 }
153
154 static void select_getanyfile(struct strbuf *hdr)
155 {
156         if (!getanyfile)
157                 forbidden(hdr, "Unsupported service: getanyfile");
158 }
159
160 static void send_strbuf(struct strbuf *hdr,
161                         const char *type, struct strbuf *buf)
162 {
163         hdr_int(hdr, content_length, buf->len);
164         hdr_str(hdr, content_type, type);
165         end_headers(hdr);
166         write_or_die(1, buf->buf, buf->len);
167 }
168
169 static void send_local_file(struct strbuf *hdr, const char *the_type,
170                                 const char *name)
171 {
172         char *p = git_pathdup("%s", name);
173         size_t buf_alloc = 8192;
174         char *buf = xmalloc(buf_alloc);
175         int fd;
176         struct stat sb;
177
178         fd = open(p, O_RDONLY);
179         if (fd < 0)
180                 not_found(hdr, "Cannot open '%s': %s", p, strerror(errno));
181         if (fstat(fd, &sb) < 0)
182                 die_errno("Cannot stat '%s'", p);
183
184         hdr_int(hdr, content_length, sb.st_size);
185         hdr_str(hdr, content_type, the_type);
186         hdr_date(hdr, last_modified, sb.st_mtime);
187         end_headers(hdr);
188
189         for (;;) {
190                 ssize_t n = xread(fd, buf, buf_alloc);
191                 if (n < 0)
192                         die_errno("Cannot read '%s'", p);
193                 if (!n)
194                         break;
195                 write_or_die(1, buf, n);
196         }
197         close(fd);
198         free(buf);
199         free(p);
200 }
201
202 static void get_text_file(struct strbuf *hdr, char *name)
203 {
204         select_getanyfile(hdr);
205         hdr_nocache(hdr);
206         send_local_file(hdr, "text/plain", name);
207 }
208
209 static void get_loose_object(struct strbuf *hdr, char *name)
210 {
211         select_getanyfile(hdr);
212         hdr_cache_forever(hdr);
213         send_local_file(hdr, "application/x-git-loose-object", name);
214 }
215
216 static void get_pack_file(struct strbuf *hdr, char *name)
217 {
218         select_getanyfile(hdr);
219         hdr_cache_forever(hdr);
220         send_local_file(hdr, "application/x-git-packed-objects", name);
221 }
222
223 static void get_idx_file(struct strbuf *hdr, char *name)
224 {
225         select_getanyfile(hdr);
226         hdr_cache_forever(hdr);
227         send_local_file(hdr, "application/x-git-packed-objects-toc", name);
228 }
229
230 static void http_config(void)
231 {
232         int i, value = 0;
233         struct strbuf var = STRBUF_INIT;
234
235         git_config_get_bool("http.getanyfile", &getanyfile);
236         git_config_get_ulong("http.maxrequestbuffer", &max_request_buffer);
237
238         for (i = 0; i < ARRAY_SIZE(rpc_service); i++) {
239                 struct rpc_service *svc = &rpc_service[i];
240                 strbuf_addf(&var, "http.%s", svc->config_name);
241                 if (!git_config_get_bool(var.buf, &value))
242                         svc->enabled = value;
243                 strbuf_reset(&var);
244         }
245
246         strbuf_release(&var);
247 }
248
249 static struct rpc_service *select_service(struct strbuf *hdr, const char *name)
250 {
251         const char *svc_name;
252         struct rpc_service *svc = NULL;
253         int i;
254
255         if (!skip_prefix(name, "git-", &svc_name))
256                 forbidden(hdr, "Unsupported service: '%s'", name);
257
258         for (i = 0; i < ARRAY_SIZE(rpc_service); i++) {
259                 struct rpc_service *s = &rpc_service[i];
260                 if (!strcmp(s->name, svc_name)) {
261                         svc = s;
262                         break;
263                 }
264         }
265
266         if (!svc)
267                 forbidden(hdr, "Unsupported service: '%s'", name);
268
269         if (svc->enabled < 0) {
270                 const char *user = getenv("REMOTE_USER");
271                 svc->enabled = (user && *user) ? 1 : 0;
272         }
273         if (!svc->enabled)
274                 forbidden(hdr, "Service not enabled: '%s'", svc->name);
275         return svc;
276 }
277
278 /*
279  * This is basically strbuf_read(), except that if we
280  * hit max_request_buffer we die (we'd rather reject a
281  * maliciously large request than chew up infinite memory).
282  */
283 static ssize_t read_request(int fd, unsigned char **out)
284 {
285         size_t len = 0, alloc = 8192;
286         unsigned char *buf = xmalloc(alloc);
287
288         if (max_request_buffer < alloc)
289                 max_request_buffer = alloc;
290
291         while (1) {
292                 ssize_t cnt;
293
294                 cnt = read_in_full(fd, buf + len, alloc - len);
295                 if (cnt < 0) {
296                         free(buf);
297                         return -1;
298                 }
299
300                 /* partial read from read_in_full means we hit EOF */
301                 len += cnt;
302                 if (len < alloc) {
303                         *out = buf;
304                         return len;
305                 }
306
307                 /* otherwise, grow and try again (if we can) */
308                 if (alloc == max_request_buffer)
309                         die("request was larger than our maximum size (%lu);"
310                             " try setting GIT_HTTP_MAX_REQUEST_BUFFER",
311                             max_request_buffer);
312
313                 alloc = alloc_nr(alloc);
314                 if (alloc > max_request_buffer)
315                         alloc = max_request_buffer;
316                 REALLOC_ARRAY(buf, alloc);
317         }
318 }
319
320 static void inflate_request(const char *prog_name, int out, int buffer_input)
321 {
322         git_zstream stream;
323         unsigned char *full_request = NULL;
324         unsigned char in_buf[8192];
325         unsigned char out_buf[8192];
326         unsigned long cnt = 0;
327
328         memset(&stream, 0, sizeof(stream));
329         git_inflate_init_gzip_only(&stream);
330
331         while (1) {
332                 ssize_t n;
333
334                 if (buffer_input) {
335                         if (full_request)
336                                 n = 0; /* nothing left to read */
337                         else
338                                 n = read_request(0, &full_request);
339                         stream.next_in = full_request;
340                 } else {
341                         n = xread(0, in_buf, sizeof(in_buf));
342                         stream.next_in = in_buf;
343                 }
344
345                 if (n <= 0)
346                         die("request ended in the middle of the gzip stream");
347                 stream.avail_in = n;
348
349                 while (0 < stream.avail_in) {
350                         int ret;
351
352                         stream.next_out = out_buf;
353                         stream.avail_out = sizeof(out_buf);
354
355                         ret = git_inflate(&stream, Z_NO_FLUSH);
356                         if (ret != Z_OK && ret != Z_STREAM_END)
357                                 die("zlib error inflating request, result %d", ret);
358
359                         n = stream.total_out - cnt;
360                         if (write_in_full(out, out_buf, n) < 0)
361                                 die("%s aborted reading request", prog_name);
362                         cnt += n;
363
364                         if (ret == Z_STREAM_END)
365                                 goto done;
366                 }
367         }
368
369 done:
370         git_inflate_end(&stream);
371         close(out);
372         free(full_request);
373 }
374
375 static void copy_request(const char *prog_name, int out)
376 {
377         unsigned char *buf;
378         ssize_t n = read_request(0, &buf);
379         if (n < 0)
380                 die_errno("error reading request body");
381         if (write_in_full(out, buf, n) < 0)
382                 die("%s aborted reading request", prog_name);
383         close(out);
384         free(buf);
385 }
386
387 static void run_service(const char **argv, int buffer_input)
388 {
389         const char *encoding = getenv("HTTP_CONTENT_ENCODING");
390         const char *user = getenv("REMOTE_USER");
391         const char *host = getenv("REMOTE_ADDR");
392         int gzipped_request = 0;
393         struct child_process cld = CHILD_PROCESS_INIT;
394
395         if (encoding && !strcmp(encoding, "gzip"))
396                 gzipped_request = 1;
397         else if (encoding && !strcmp(encoding, "x-gzip"))
398                 gzipped_request = 1;
399
400         if (!user || !*user)
401                 user = "anonymous";
402         if (!host || !*host)
403                 host = "(none)";
404
405         if (!getenv("GIT_COMMITTER_NAME"))
406                 argv_array_pushf(&cld.env_array, "GIT_COMMITTER_NAME=%s", user);
407         if (!getenv("GIT_COMMITTER_EMAIL"))
408                 argv_array_pushf(&cld.env_array,
409                                  "GIT_COMMITTER_EMAIL=%s@http.%s", user, host);
410
411         cld.argv = argv;
412         if (buffer_input || gzipped_request)
413                 cld.in = -1;
414         cld.git_cmd = 1;
415         if (start_command(&cld))
416                 exit(1);
417
418         close(1);
419         if (gzipped_request)
420                 inflate_request(argv[0], cld.in, buffer_input);
421         else if (buffer_input)
422                 copy_request(argv[0], cld.in);
423         else
424                 close(0);
425
426         if (finish_command(&cld))
427                 exit(1);
428 }
429
430 static int show_text_ref(const char *name, const struct object_id *oid,
431                          int flag, void *cb_data)
432 {
433         const char *name_nons = strip_namespace(name);
434         struct strbuf *buf = cb_data;
435         struct object *o = parse_object(oid);
436         if (!o)
437                 return 0;
438
439         strbuf_addf(buf, "%s\t%s\n", oid_to_hex(oid), name_nons);
440         if (o->type == OBJ_TAG) {
441                 o = deref_tag(o, name, 0);
442                 if (!o)
443                         return 0;
444                 strbuf_addf(buf, "%s\t%s^{}\n", oid_to_hex(&o->oid),
445                             name_nons);
446         }
447         return 0;
448 }
449
450 static void get_info_refs(struct strbuf *hdr, char *arg)
451 {
452         const char *service_name = get_parameter("service");
453         struct strbuf buf = STRBUF_INIT;
454
455         hdr_nocache(hdr);
456
457         if (service_name) {
458                 const char *argv[] = {NULL /* service name */,
459                         "--stateless-rpc", "--advertise-refs",
460                         ".", NULL};
461                 struct rpc_service *svc = select_service(hdr, service_name);
462
463                 strbuf_addf(&buf, "application/x-git-%s-advertisement",
464                         svc->name);
465                 hdr_str(hdr, content_type, buf.buf);
466                 end_headers(hdr);
467
468                 packet_write_fmt(1, "# service=git-%s\n", svc->name);
469                 packet_flush(1);
470
471                 argv[0] = svc->name;
472                 run_service(argv, 0);
473
474         } else {
475                 select_getanyfile(hdr);
476                 for_each_namespaced_ref(show_text_ref, &buf);
477                 send_strbuf(hdr, "text/plain", &buf);
478         }
479         strbuf_release(&buf);
480 }
481
482 static int show_head_ref(const char *refname, const struct object_id *oid,
483                          int flag, void *cb_data)
484 {
485         struct strbuf *buf = cb_data;
486
487         if (flag & REF_ISSYMREF) {
488                 struct object_id unused;
489                 const char *target = resolve_ref_unsafe(refname,
490                                                         RESOLVE_REF_READING,
491                                                         unused.hash, NULL);
492
493                 if (target)
494                         strbuf_addf(buf, "ref: %s\n", strip_namespace(target));
495         } else {
496                 strbuf_addf(buf, "%s\n", oid_to_hex(oid));
497         }
498
499         return 0;
500 }
501
502 static void get_head(struct strbuf *hdr, char *arg)
503 {
504         struct strbuf buf = STRBUF_INIT;
505
506         select_getanyfile(hdr);
507         head_ref_namespaced(show_head_ref, &buf);
508         send_strbuf(hdr, "text/plain", &buf);
509         strbuf_release(&buf);
510 }
511
512 static void get_info_packs(struct strbuf *hdr, char *arg)
513 {
514         size_t objdirlen = strlen(get_object_directory());
515         struct strbuf buf = STRBUF_INIT;
516         struct packed_git *p;
517         size_t cnt = 0;
518
519         select_getanyfile(hdr);
520         prepare_packed_git();
521         for (p = packed_git; p; p = p->next) {
522                 if (p->pack_local)
523                         cnt++;
524         }
525
526         strbuf_grow(&buf, cnt * 53 + 2);
527         for (p = packed_git; p; p = p->next) {
528                 if (p->pack_local)
529                         strbuf_addf(&buf, "P %s\n", p->pack_name + objdirlen + 6);
530         }
531         strbuf_addch(&buf, '\n');
532
533         hdr_nocache(hdr);
534         send_strbuf(hdr, "text/plain; charset=utf-8", &buf);
535         strbuf_release(&buf);
536 }
537
538 static void check_content_type(struct strbuf *hdr, const char *accepted_type)
539 {
540         const char *actual_type = getenv("CONTENT_TYPE");
541
542         if (!actual_type)
543                 actual_type = "";
544
545         if (strcmp(actual_type, accepted_type)) {
546                 http_status(hdr, 415, "Unsupported Media Type");
547                 hdr_nocache(hdr);
548                 end_headers(hdr);
549                 format_write(1,
550                         "Expected POST with Content-Type '%s',"
551                         " but received '%s' instead.\n",
552                         accepted_type, actual_type);
553                 exit(0);
554         }
555 }
556
557 static void service_rpc(struct strbuf *hdr, char *service_name)
558 {
559         const char *argv[] = {NULL, "--stateless-rpc", ".", NULL};
560         struct rpc_service *svc = select_service(hdr, service_name);
561         struct strbuf buf = STRBUF_INIT;
562
563         strbuf_reset(&buf);
564         strbuf_addf(&buf, "application/x-git-%s-request", svc->name);
565         check_content_type(hdr, buf.buf);
566
567         hdr_nocache(hdr);
568
569         strbuf_reset(&buf);
570         strbuf_addf(&buf, "application/x-git-%s-result", svc->name);
571         hdr_str(hdr, content_type, buf.buf);
572
573         end_headers(hdr);
574
575         argv[0] = svc->name;
576         run_service(argv, svc->buffer_input);
577         strbuf_release(&buf);
578 }
579
580 static int dead;
581 static NORETURN void die_webcgi(const char *err, va_list params)
582 {
583         if (dead <= 1) {
584                 struct strbuf hdr = STRBUF_INIT;
585
586                 vreportf("fatal: ", err, params);
587
588                 http_status(&hdr, 500, "Internal Server Error");
589                 hdr_nocache(&hdr);
590                 end_headers(&hdr);
591         }
592         exit(0); /* we successfully reported a failure ;-) */
593 }
594
595 static int die_webcgi_recursing(void)
596 {
597         return dead++ > 1;
598 }
599
600 static char* getdir(void)
601 {
602         struct strbuf buf = STRBUF_INIT;
603         char *pathinfo = getenv("PATH_INFO");
604         char *root = getenv("GIT_PROJECT_ROOT");
605         char *path = getenv("PATH_TRANSLATED");
606
607         if (root && *root) {
608                 if (!pathinfo || !*pathinfo)
609                         die("GIT_PROJECT_ROOT is set but PATH_INFO is not");
610                 if (daemon_avoid_alias(pathinfo))
611                         die("'%s': aliased", pathinfo);
612                 end_url_with_slash(&buf, root);
613                 if (pathinfo[0] == '/')
614                         pathinfo++;
615                 strbuf_addstr(&buf, pathinfo);
616                 return strbuf_detach(&buf, NULL);
617         } else if (path && *path) {
618                 return xstrdup(path);
619         } else
620                 die("No GIT_PROJECT_ROOT or PATH_TRANSLATED from server");
621         return NULL;
622 }
623
624 static struct service_cmd {
625         const char *method;
626         const char *pattern;
627         void (*imp)(struct strbuf *, char *);
628 } services[] = {
629         {"GET", "/HEAD$", get_head},
630         {"GET", "/info/refs$", get_info_refs},
631         {"GET", "/objects/info/alternates$", get_text_file},
632         {"GET", "/objects/info/http-alternates$", get_text_file},
633         {"GET", "/objects/info/packs$", get_info_packs},
634         {"GET", "/objects/[0-9a-f]{2}/[0-9a-f]{38}$", get_loose_object},
635         {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.pack$", get_pack_file},
636         {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.idx$", get_idx_file},
637
638         {"POST", "/git-upload-pack$", service_rpc},
639         {"POST", "/git-receive-pack$", service_rpc}
640 };
641
642 static int bad_request(struct strbuf *hdr, const struct service_cmd *c)
643 {
644         const char *proto = getenv("SERVER_PROTOCOL");
645
646         if (proto && !strcmp(proto, "HTTP/1.1")) {
647                 http_status(hdr, 405, "Method Not Allowed");
648                 hdr_str(hdr, "Allow",
649                         !strcmp(c->method, "GET") ? "GET, HEAD" : c->method);
650         } else
651                 http_status(hdr, 400, "Bad Request");
652         hdr_nocache(hdr);
653         end_headers(hdr);
654         return 0;
655 }
656
657 int cmd_main(int argc, const char **argv)
658 {
659         char *method = getenv("REQUEST_METHOD");
660         char *dir;
661         struct service_cmd *cmd = NULL;
662         char *cmd_arg = NULL;
663         int i;
664         struct strbuf hdr = STRBUF_INIT;
665
666         set_die_routine(die_webcgi);
667         set_die_is_recursing_routine(die_webcgi_recursing);
668
669         if (!method)
670                 die("No REQUEST_METHOD from server");
671         if (!strcmp(method, "HEAD"))
672                 method = "GET";
673         dir = getdir();
674
675         for (i = 0; i < ARRAY_SIZE(services); i++) {
676                 struct service_cmd *c = &services[i];
677                 regex_t re;
678                 regmatch_t out[1];
679
680                 if (regcomp(&re, c->pattern, REG_EXTENDED))
681                         die("Bogus regex in service table: %s", c->pattern);
682                 if (!regexec(&re, dir, 1, out, 0)) {
683                         size_t n;
684
685                         if (strcmp(method, c->method))
686                                 return bad_request(&hdr, c);
687
688                         cmd = c;
689                         n = out[0].rm_eo - out[0].rm_so;
690                         cmd_arg = xmemdupz(dir + out[0].rm_so + 1, n - 1);
691                         dir[out[0].rm_so] = 0;
692                         break;
693                 }
694                 regfree(&re);
695         }
696
697         if (!cmd)
698                 not_found(&hdr, "Request not supported: '%s'", dir);
699
700         setup_path();
701         if (!enter_repo(dir, 0))
702                 not_found(&hdr, "Not a git repository: '%s'", dir);
703         if (!getenv("GIT_HTTP_EXPORT_ALL") &&
704             access("git-daemon-export-ok", F_OK) )
705                 not_found(&hdr, "Repository not exported: '%s'", dir);
706
707         http_config();
708         max_request_buffer = git_env_ulong("GIT_HTTP_MAX_REQUEST_BUFFER",
709                                            max_request_buffer);
710
711         cmd->imp(&hdr, cmd_arg);
712         return 0;
713 }