http-backend: cleanup writing to child process
[git] / http-backend.c
1 #include "cache.h"
2 #include "config.h"
3 #include "repository.h"
4 #include "refs.h"
5 #include "pkt-line.h"
6 #include "object.h"
7 #include "tag.h"
8 #include "exec-cmd.h"
9 #include "run-command.h"
10 #include "string-list.h"
11 #include "url.h"
12 #include "argv-array.h"
13 #include "packfile.h"
14 #include "object-store.h"
15 #include "protocol.h"
16
17 static const char content_type[] = "Content-Type";
18 static const char content_length[] = "Content-Length";
19 static const char last_modified[] = "Last-Modified";
20 static int getanyfile = 1;
21 static unsigned long max_request_buffer = 10 * 1024 * 1024;
22
23 static struct string_list *query_params;
24
25 struct rpc_service {
26         const char *name;
27         const char *config_name;
28         unsigned buffer_input : 1;
29         signed enabled : 2;
30 };
31
32 static struct rpc_service rpc_service[] = {
33         { "upload-pack", "uploadpack", 1, 1 },
34         { "receive-pack", "receivepack", 0, -1 },
35 };
36
37 static struct string_list *get_parameters(void)
38 {
39         if (!query_params) {
40                 const char *query = getenv("QUERY_STRING");
41
42                 query_params = xcalloc(1, sizeof(*query_params));
43                 while (query && *query) {
44                         char *name = url_decode_parameter_name(&query);
45                         char *value = url_decode_parameter_value(&query);
46                         struct string_list_item *i;
47
48                         i = string_list_lookup(query_params, name);
49                         if (!i)
50                                 i = string_list_insert(query_params, name);
51                         else
52                                 free(i->util);
53                         i->util = value;
54                 }
55         }
56         return query_params;
57 }
58
59 static const char *get_parameter(const char *name)
60 {
61         struct string_list_item *i;
62         i = string_list_lookup(get_parameters(), name);
63         return i ? i->util : NULL;
64 }
65
66 __attribute__((format (printf, 2, 3)))
67 static void format_write(int fd, const char *fmt, ...)
68 {
69         static char buffer[1024];
70
71         va_list args;
72         unsigned n;
73
74         va_start(args, fmt);
75         n = vsnprintf(buffer, sizeof(buffer), fmt, args);
76         va_end(args);
77         if (n >= sizeof(buffer))
78                 die("protocol error: impossibly long line");
79
80         write_or_die(fd, buffer, n);
81 }
82
83 static void http_status(struct strbuf *hdr, unsigned code, const char *msg)
84 {
85         strbuf_addf(hdr, "Status: %u %s\r\n", code, msg);
86 }
87
88 static void hdr_str(struct strbuf *hdr, const char *name, const char *value)
89 {
90         strbuf_addf(hdr, "%s: %s\r\n", name, value);
91 }
92
93 static void hdr_int(struct strbuf *hdr, const char *name, uintmax_t value)
94 {
95         strbuf_addf(hdr, "%s: %" PRIuMAX "\r\n", name, value);
96 }
97
98 static void hdr_date(struct strbuf *hdr, const char *name, timestamp_t when)
99 {
100         const char *value = show_date(when, 0, DATE_MODE(RFC2822));
101         hdr_str(hdr, name, value);
102 }
103
104 static void hdr_nocache(struct strbuf *hdr)
105 {
106         hdr_str(hdr, "Expires", "Fri, 01 Jan 1980 00:00:00 GMT");
107         hdr_str(hdr, "Pragma", "no-cache");
108         hdr_str(hdr, "Cache-Control", "no-cache, max-age=0, must-revalidate");
109 }
110
111 static void hdr_cache_forever(struct strbuf *hdr)
112 {
113         timestamp_t now = time(NULL);
114         hdr_date(hdr, "Date", now);
115         hdr_date(hdr, "Expires", now + 31536000);
116         hdr_str(hdr, "Cache-Control", "public, max-age=31536000");
117 }
118
119 static void end_headers(struct strbuf *hdr)
120 {
121         strbuf_add(hdr, "\r\n", 2);
122         write_or_die(1, hdr->buf, hdr->len);
123         strbuf_release(hdr);
124 }
125
126 __attribute__((format (printf, 2, 3)))
127 static NORETURN void not_found(struct strbuf *hdr, const char *err, ...)
128 {
129         va_list params;
130
131         http_status(hdr, 404, "Not Found");
132         hdr_nocache(hdr);
133         end_headers(hdr);
134
135         va_start(params, err);
136         if (err && *err)
137                 vfprintf(stderr, err, params);
138         va_end(params);
139         exit(0);
140 }
141
142 __attribute__((format (printf, 2, 3)))
143 static NORETURN void forbidden(struct strbuf *hdr, const char *err, ...)
144 {
145         va_list params;
146
147         http_status(hdr, 403, "Forbidden");
148         hdr_nocache(hdr);
149         end_headers(hdr);
150
151         va_start(params, err);
152         if (err && *err)
153                 vfprintf(stderr, err, params);
154         va_end(params);
155         exit(0);
156 }
157
158 static void select_getanyfile(struct strbuf *hdr)
159 {
160         if (!getanyfile)
161                 forbidden(hdr, "Unsupported service: getanyfile");
162 }
163
164 static void send_strbuf(struct strbuf *hdr,
165                         const char *type, struct strbuf *buf)
166 {
167         hdr_int(hdr, content_length, buf->len);
168         hdr_str(hdr, content_type, type);
169         end_headers(hdr);
170         write_or_die(1, buf->buf, buf->len);
171 }
172
173 static void send_local_file(struct strbuf *hdr, const char *the_type,
174                                 const char *name)
175 {
176         char *p = git_pathdup("%s", name);
177         size_t buf_alloc = 8192;
178         char *buf = xmalloc(buf_alloc);
179         int fd;
180         struct stat sb;
181
182         fd = open(p, O_RDONLY);
183         if (fd < 0)
184                 not_found(hdr, "Cannot open '%s': %s", p, strerror(errno));
185         if (fstat(fd, &sb) < 0)
186                 die_errno("Cannot stat '%s'", p);
187
188         hdr_int(hdr, content_length, sb.st_size);
189         hdr_str(hdr, content_type, the_type);
190         hdr_date(hdr, last_modified, sb.st_mtime);
191         end_headers(hdr);
192
193         for (;;) {
194                 ssize_t n = xread(fd, buf, buf_alloc);
195                 if (n < 0)
196                         die_errno("Cannot read '%s'", p);
197                 if (!n)
198                         break;
199                 write_or_die(1, buf, n);
200         }
201         close(fd);
202         free(buf);
203         free(p);
204 }
205
206 static void get_text_file(struct strbuf *hdr, char *name)
207 {
208         select_getanyfile(hdr);
209         hdr_nocache(hdr);
210         send_local_file(hdr, "text/plain", name);
211 }
212
213 static void get_loose_object(struct strbuf *hdr, char *name)
214 {
215         select_getanyfile(hdr);
216         hdr_cache_forever(hdr);
217         send_local_file(hdr, "application/x-git-loose-object", name);
218 }
219
220 static void get_pack_file(struct strbuf *hdr, char *name)
221 {
222         select_getanyfile(hdr);
223         hdr_cache_forever(hdr);
224         send_local_file(hdr, "application/x-git-packed-objects", name);
225 }
226
227 static void get_idx_file(struct strbuf *hdr, char *name)
228 {
229         select_getanyfile(hdr);
230         hdr_cache_forever(hdr);
231         send_local_file(hdr, "application/x-git-packed-objects-toc", name);
232 }
233
234 static void http_config(void)
235 {
236         int i, value = 0;
237         struct strbuf var = STRBUF_INIT;
238
239         git_config_get_bool("http.getanyfile", &getanyfile);
240         git_config_get_ulong("http.maxrequestbuffer", &max_request_buffer);
241
242         for (i = 0; i < ARRAY_SIZE(rpc_service); i++) {
243                 struct rpc_service *svc = &rpc_service[i];
244                 strbuf_addf(&var, "http.%s", svc->config_name);
245                 if (!git_config_get_bool(var.buf, &value))
246                         svc->enabled = value;
247                 strbuf_reset(&var);
248         }
249
250         strbuf_release(&var);
251 }
252
253 static struct rpc_service *select_service(struct strbuf *hdr, const char *name)
254 {
255         const char *svc_name;
256         struct rpc_service *svc = NULL;
257         int i;
258
259         if (!skip_prefix(name, "git-", &svc_name))
260                 forbidden(hdr, "Unsupported service: '%s'", name);
261
262         for (i = 0; i < ARRAY_SIZE(rpc_service); i++) {
263                 struct rpc_service *s = &rpc_service[i];
264                 if (!strcmp(s->name, svc_name)) {
265                         svc = s;
266                         break;
267                 }
268         }
269
270         if (!svc)
271                 forbidden(hdr, "Unsupported service: '%s'", name);
272
273         if (svc->enabled < 0) {
274                 const char *user = getenv("REMOTE_USER");
275                 svc->enabled = (user && *user) ? 1 : 0;
276         }
277         if (!svc->enabled)
278                 forbidden(hdr, "Service not enabled: '%s'", svc->name);
279         return svc;
280 }
281
282 static void write_to_child(int out, const unsigned char *buf, ssize_t len, const char *prog_name)
283 {
284         if (write_in_full(out, buf, len) < 0)
285                 die("unable to write to '%s'", prog_name);
286 }
287
288 /*
289  * This is basically strbuf_read(), except that if we
290  * hit max_request_buffer we die (we'd rather reject a
291  * maliciously large request than chew up infinite memory).
292  */
293 static ssize_t read_request(int fd, unsigned char **out)
294 {
295         size_t len = 0, alloc = 8192;
296         unsigned char *buf = xmalloc(alloc);
297
298         if (max_request_buffer < alloc)
299                 max_request_buffer = alloc;
300
301         while (1) {
302                 ssize_t cnt;
303
304                 cnt = read_in_full(fd, buf + len, alloc - len);
305                 if (cnt < 0) {
306                         free(buf);
307                         return -1;
308                 }
309
310                 /* partial read from read_in_full means we hit EOF */
311                 len += cnt;
312                 if (len < alloc) {
313                         *out = buf;
314                         return len;
315                 }
316
317                 /* otherwise, grow and try again (if we can) */
318                 if (alloc == max_request_buffer)
319                         die("request was larger than our maximum size (%lu);"
320                             " try setting GIT_HTTP_MAX_REQUEST_BUFFER",
321                             max_request_buffer);
322
323                 alloc = alloc_nr(alloc);
324                 if (alloc > max_request_buffer)
325                         alloc = max_request_buffer;
326                 REALLOC_ARRAY(buf, alloc);
327         }
328 }
329
330 static void inflate_request(const char *prog_name, int out, int buffer_input)
331 {
332         git_zstream stream;
333         unsigned char *full_request = NULL;
334         unsigned char in_buf[8192];
335         unsigned char out_buf[8192];
336         unsigned long cnt = 0;
337
338         memset(&stream, 0, sizeof(stream));
339         git_inflate_init_gzip_only(&stream);
340
341         while (1) {
342                 ssize_t n;
343
344                 if (buffer_input) {
345                         if (full_request)
346                                 n = 0; /* nothing left to read */
347                         else
348                                 n = read_request(0, &full_request);
349                         stream.next_in = full_request;
350                 } else {
351                         n = xread(0, in_buf, sizeof(in_buf));
352                         stream.next_in = in_buf;
353                 }
354
355                 if (n <= 0)
356                         die("request ended in the middle of the gzip stream");
357                 stream.avail_in = n;
358
359                 while (0 < stream.avail_in) {
360                         int ret;
361
362                         stream.next_out = out_buf;
363                         stream.avail_out = sizeof(out_buf);
364
365                         ret = git_inflate(&stream, Z_NO_FLUSH);
366                         if (ret != Z_OK && ret != Z_STREAM_END)
367                                 die("zlib error inflating request, result %d", ret);
368
369                         n = stream.total_out - cnt;
370                         write_to_child(out, out_buf, stream.total_out - cnt, prog_name);
371                         cnt = stream.total_out;
372
373                         if (ret == Z_STREAM_END)
374                                 goto done;
375                 }
376         }
377
378 done:
379         git_inflate_end(&stream);
380         close(out);
381         free(full_request);
382 }
383
384 static void copy_request(const char *prog_name, int out)
385 {
386         unsigned char *buf;
387         ssize_t n = read_request(0, &buf);
388         if (n < 0)
389                 die_errno("error reading request body");
390         write_to_child(out, buf, n, prog_name);
391         close(out);
392         free(buf);
393 }
394
395 static void run_service(const char **argv, int buffer_input)
396 {
397         const char *encoding = getenv("HTTP_CONTENT_ENCODING");
398         const char *user = getenv("REMOTE_USER");
399         const char *host = getenv("REMOTE_ADDR");
400         int gzipped_request = 0;
401         struct child_process cld = CHILD_PROCESS_INIT;
402
403         if (encoding && !strcmp(encoding, "gzip"))
404                 gzipped_request = 1;
405         else if (encoding && !strcmp(encoding, "x-gzip"))
406                 gzipped_request = 1;
407
408         if (!user || !*user)
409                 user = "anonymous";
410         if (!host || !*host)
411                 host = "(none)";
412
413         if (!getenv("GIT_COMMITTER_NAME"))
414                 argv_array_pushf(&cld.env_array, "GIT_COMMITTER_NAME=%s", user);
415         if (!getenv("GIT_COMMITTER_EMAIL"))
416                 argv_array_pushf(&cld.env_array,
417                                  "GIT_COMMITTER_EMAIL=%s@http.%s", user, host);
418
419         cld.argv = argv;
420         if (buffer_input || gzipped_request)
421                 cld.in = -1;
422         cld.git_cmd = 1;
423         if (start_command(&cld))
424                 exit(1);
425
426         close(1);
427         if (gzipped_request)
428                 inflate_request(argv[0], cld.in, buffer_input);
429         else if (buffer_input)
430                 copy_request(argv[0], cld.in);
431         else
432                 close(0);
433
434         if (finish_command(&cld))
435                 exit(1);
436 }
437
438 static int show_text_ref(const char *name, const struct object_id *oid,
439                          int flag, void *cb_data)
440 {
441         const char *name_nons = strip_namespace(name);
442         struct strbuf *buf = cb_data;
443         struct object *o = parse_object(oid);
444         if (!o)
445                 return 0;
446
447         strbuf_addf(buf, "%s\t%s\n", oid_to_hex(oid), name_nons);
448         if (o->type == OBJ_TAG) {
449                 o = deref_tag(o, name, 0);
450                 if (!o)
451                         return 0;
452                 strbuf_addf(buf, "%s\t%s^{}\n", oid_to_hex(&o->oid),
453                             name_nons);
454         }
455         return 0;
456 }
457
458 static void get_info_refs(struct strbuf *hdr, char *arg)
459 {
460         const char *service_name = get_parameter("service");
461         struct strbuf buf = STRBUF_INIT;
462
463         hdr_nocache(hdr);
464
465         if (service_name) {
466                 const char *argv[] = {NULL /* service name */,
467                         "--stateless-rpc", "--advertise-refs",
468                         ".", NULL};
469                 struct rpc_service *svc = select_service(hdr, service_name);
470
471                 strbuf_addf(&buf, "application/x-git-%s-advertisement",
472                         svc->name);
473                 hdr_str(hdr, content_type, buf.buf);
474                 end_headers(hdr);
475
476
477                 if (determine_protocol_version_server() != protocol_v2) {
478                         packet_write_fmt(1, "# service=git-%s\n", svc->name);
479                         packet_flush(1);
480                 }
481
482                 argv[0] = svc->name;
483                 run_service(argv, 0);
484
485         } else {
486                 select_getanyfile(hdr);
487                 for_each_namespaced_ref(show_text_ref, &buf);
488                 send_strbuf(hdr, "text/plain", &buf);
489         }
490         strbuf_release(&buf);
491 }
492
493 static int show_head_ref(const char *refname, const struct object_id *oid,
494                          int flag, void *cb_data)
495 {
496         struct strbuf *buf = cb_data;
497
498         if (flag & REF_ISSYMREF) {
499                 const char *target = resolve_ref_unsafe(refname,
500                                                         RESOLVE_REF_READING,
501                                                         NULL, NULL);
502
503                 if (target)
504                         strbuf_addf(buf, "ref: %s\n", strip_namespace(target));
505         } else {
506                 strbuf_addf(buf, "%s\n", oid_to_hex(oid));
507         }
508
509         return 0;
510 }
511
512 static void get_head(struct strbuf *hdr, char *arg)
513 {
514         struct strbuf buf = STRBUF_INIT;
515
516         select_getanyfile(hdr);
517         head_ref_namespaced(show_head_ref, &buf);
518         send_strbuf(hdr, "text/plain", &buf);
519         strbuf_release(&buf);
520 }
521
522 static void get_info_packs(struct strbuf *hdr, char *arg)
523 {
524         size_t objdirlen = strlen(get_object_directory());
525         struct strbuf buf = STRBUF_INIT;
526         struct packed_git *p;
527         size_t cnt = 0;
528
529         select_getanyfile(hdr);
530         for (p = get_packed_git(the_repository); p; p = p->next) {
531                 if (p->pack_local)
532                         cnt++;
533         }
534
535         strbuf_grow(&buf, cnt * 53 + 2);
536         for (p = get_packed_git(the_repository); p; p = p->next) {
537                 if (p->pack_local)
538                         strbuf_addf(&buf, "P %s\n", p->pack_name + objdirlen + 6);
539         }
540         strbuf_addch(&buf, '\n');
541
542         hdr_nocache(hdr);
543         send_strbuf(hdr, "text/plain; charset=utf-8", &buf);
544         strbuf_release(&buf);
545 }
546
547 static void check_content_type(struct strbuf *hdr, const char *accepted_type)
548 {
549         const char *actual_type = getenv("CONTENT_TYPE");
550
551         if (!actual_type)
552                 actual_type = "";
553
554         if (strcmp(actual_type, accepted_type)) {
555                 http_status(hdr, 415, "Unsupported Media Type");
556                 hdr_nocache(hdr);
557                 end_headers(hdr);
558                 format_write(1,
559                         "Expected POST with Content-Type '%s',"
560                         " but received '%s' instead.\n",
561                         accepted_type, actual_type);
562                 exit(0);
563         }
564 }
565
566 static void service_rpc(struct strbuf *hdr, char *service_name)
567 {
568         const char *argv[] = {NULL, "--stateless-rpc", ".", NULL};
569         struct rpc_service *svc = select_service(hdr, service_name);
570         struct strbuf buf = STRBUF_INIT;
571
572         strbuf_reset(&buf);
573         strbuf_addf(&buf, "application/x-git-%s-request", svc->name);
574         check_content_type(hdr, buf.buf);
575
576         hdr_nocache(hdr);
577
578         strbuf_reset(&buf);
579         strbuf_addf(&buf, "application/x-git-%s-result", svc->name);
580         hdr_str(hdr, content_type, buf.buf);
581
582         end_headers(hdr);
583
584         argv[0] = svc->name;
585         run_service(argv, svc->buffer_input);
586         strbuf_release(&buf);
587 }
588
589 static int dead;
590 static NORETURN void die_webcgi(const char *err, va_list params)
591 {
592         if (dead <= 1) {
593                 struct strbuf hdr = STRBUF_INIT;
594
595                 vreportf("fatal: ", err, params);
596
597                 http_status(&hdr, 500, "Internal Server Error");
598                 hdr_nocache(&hdr);
599                 end_headers(&hdr);
600         }
601         exit(0); /* we successfully reported a failure ;-) */
602 }
603
604 static int die_webcgi_recursing(void)
605 {
606         return dead++ > 1;
607 }
608
609 static char* getdir(void)
610 {
611         struct strbuf buf = STRBUF_INIT;
612         char *pathinfo = getenv("PATH_INFO");
613         char *root = getenv("GIT_PROJECT_ROOT");
614         char *path = getenv("PATH_TRANSLATED");
615
616         if (root && *root) {
617                 if (!pathinfo || !*pathinfo)
618                         die("GIT_PROJECT_ROOT is set but PATH_INFO is not");
619                 if (daemon_avoid_alias(pathinfo))
620                         die("'%s': aliased", pathinfo);
621                 end_url_with_slash(&buf, root);
622                 if (pathinfo[0] == '/')
623                         pathinfo++;
624                 strbuf_addstr(&buf, pathinfo);
625                 return strbuf_detach(&buf, NULL);
626         } else if (path && *path) {
627                 return xstrdup(path);
628         } else
629                 die("No GIT_PROJECT_ROOT or PATH_TRANSLATED from server");
630         return NULL;
631 }
632
633 static struct service_cmd {
634         const char *method;
635         const char *pattern;
636         void (*imp)(struct strbuf *, char *);
637 } services[] = {
638         {"GET", "/HEAD$", get_head},
639         {"GET", "/info/refs$", get_info_refs},
640         {"GET", "/objects/info/alternates$", get_text_file},
641         {"GET", "/objects/info/http-alternates$", get_text_file},
642         {"GET", "/objects/info/packs$", get_info_packs},
643         {"GET", "/objects/[0-9a-f]{2}/[0-9a-f]{38}$", get_loose_object},
644         {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.pack$", get_pack_file},
645         {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.idx$", get_idx_file},
646
647         {"POST", "/git-upload-pack$", service_rpc},
648         {"POST", "/git-receive-pack$", service_rpc}
649 };
650
651 static int bad_request(struct strbuf *hdr, const struct service_cmd *c)
652 {
653         const char *proto = getenv("SERVER_PROTOCOL");
654
655         if (proto && !strcmp(proto, "HTTP/1.1")) {
656                 http_status(hdr, 405, "Method Not Allowed");
657                 hdr_str(hdr, "Allow",
658                         !strcmp(c->method, "GET") ? "GET, HEAD" : c->method);
659         } else
660                 http_status(hdr, 400, "Bad Request");
661         hdr_nocache(hdr);
662         end_headers(hdr);
663         return 0;
664 }
665
666 int cmd_main(int argc, const char **argv)
667 {
668         char *method = getenv("REQUEST_METHOD");
669         char *dir;
670         struct service_cmd *cmd = NULL;
671         char *cmd_arg = NULL;
672         int i;
673         struct strbuf hdr = STRBUF_INIT;
674
675         set_die_routine(die_webcgi);
676         set_die_is_recursing_routine(die_webcgi_recursing);
677
678         if (!method)
679                 die("No REQUEST_METHOD from server");
680         if (!strcmp(method, "HEAD"))
681                 method = "GET";
682         dir = getdir();
683
684         for (i = 0; i < ARRAY_SIZE(services); i++) {
685                 struct service_cmd *c = &services[i];
686                 regex_t re;
687                 regmatch_t out[1];
688
689                 if (regcomp(&re, c->pattern, REG_EXTENDED))
690                         die("Bogus regex in service table: %s", c->pattern);
691                 if (!regexec(&re, dir, 1, out, 0)) {
692                         size_t n;
693
694                         if (strcmp(method, c->method))
695                                 return bad_request(&hdr, c);
696
697                         cmd = c;
698                         n = out[0].rm_eo - out[0].rm_so;
699                         cmd_arg = xmemdupz(dir + out[0].rm_so + 1, n - 1);
700                         dir[out[0].rm_so] = 0;
701                         break;
702                 }
703                 regfree(&re);
704         }
705
706         if (!cmd)
707                 not_found(&hdr, "Request not supported: '%s'", dir);
708
709         setup_path();
710         if (!enter_repo(dir, 0))
711                 not_found(&hdr, "Not a git repository: '%s'", dir);
712         if (!getenv("GIT_HTTP_EXPORT_ALL") &&
713             access("git-daemon-export-ok", F_OK) )
714                 not_found(&hdr, "Repository not exported: '%s'", dir);
715
716         http_config();
717         max_request_buffer = git_env_ulong("GIT_HTTP_MAX_REQUEST_BUFFER",
718                                            max_request_buffer);
719
720         cmd->imp(&hdr, cmd_arg);
721         return 0;
722 }