blame: use changed-path Bloom filters
[git.git] / http-backend.c
blobec3144b4447548cfd9cf23d22c1f7f5f8c7cde35
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"
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;
23 static struct string_list *query_params;
25 struct rpc_service {
26 const char *name;
27 const char *config_name;
28 unsigned buffer_input : 1;
29 signed enabled : 2;
32 static struct rpc_service rpc_service[] = {
33 { "upload-pack", "uploadpack", 1, 1 },
34 { "receive-pack", "receivepack", 0, -1 },
37 static struct string_list *get_parameters(void)
39 if (!query_params) {
40 const char *query = getenv("QUERY_STRING");
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;
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;
56 return query_params;
59 static const char *get_parameter(const char *name)
61 struct string_list_item *i;
62 i = string_list_lookup(get_parameters(), name);
63 return i ? i->util : NULL;
66 __attribute__((format (printf, 2, 3)))
67 static void format_write(int fd, const char *fmt, ...)
69 static char buffer[1024];
71 va_list args;
72 unsigned n;
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");
80 write_or_die(fd, buffer, n);
83 static void http_status(struct strbuf *hdr, unsigned code, const char *msg)
85 strbuf_addf(hdr, "Status: %u %s\r\n", code, msg);
88 static void hdr_str(struct strbuf *hdr, const char *name, const char *value)
90 strbuf_addf(hdr, "%s: %s\r\n", name, value);
93 static void hdr_int(struct strbuf *hdr, const char *name, uintmax_t value)
95 strbuf_addf(hdr, "%s: %" PRIuMAX "\r\n", name, value);
98 static void hdr_date(struct strbuf *hdr, const char *name, timestamp_t when)
100 const char *value = show_date(when, 0, DATE_MODE(RFC2822));
101 hdr_str(hdr, name, value);
104 static void hdr_nocache(struct strbuf *hdr)
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");
111 static void hdr_cache_forever(struct strbuf *hdr)
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");
119 static void end_headers(struct strbuf *hdr)
121 strbuf_add(hdr, "\r\n", 2);
122 write_or_die(1, hdr->buf, hdr->len);
123 strbuf_release(hdr);
126 __attribute__((format (printf, 2, 3)))
127 static NORETURN void not_found(struct strbuf *hdr, const char *err, ...)
129 va_list params;
131 http_status(hdr, 404, "Not Found");
132 hdr_nocache(hdr);
133 end_headers(hdr);
135 va_start(params, err);
136 if (err && *err)
137 vfprintf(stderr, err, params);
138 va_end(params);
139 exit(0);
142 __attribute__((format (printf, 2, 3)))
143 static NORETURN void forbidden(struct strbuf *hdr, const char *err, ...)
145 va_list params;
147 http_status(hdr, 403, "Forbidden");
148 hdr_nocache(hdr);
149 end_headers(hdr);
151 va_start(params, err);
152 if (err && *err)
153 vfprintf(stderr, err, params);
154 va_end(params);
155 exit(0);
158 static void select_getanyfile(struct strbuf *hdr)
160 if (!getanyfile)
161 forbidden(hdr, "Unsupported service: getanyfile");
164 static void send_strbuf(struct strbuf *hdr,
165 const char *type, struct strbuf *buf)
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);
173 static void send_local_file(struct strbuf *hdr, const char *the_type,
174 const char *name)
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;
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);
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);
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);
201 close(fd);
202 free(buf);
203 free(p);
206 static void get_text_file(struct strbuf *hdr, char *name)
208 select_getanyfile(hdr);
209 hdr_nocache(hdr);
210 send_local_file(hdr, "text/plain", name);
213 static void get_loose_object(struct strbuf *hdr, char *name)
215 select_getanyfile(hdr);
216 hdr_cache_forever(hdr);
217 send_local_file(hdr, "application/x-git-loose-object", name);
220 static void get_pack_file(struct strbuf *hdr, char *name)
222 select_getanyfile(hdr);
223 hdr_cache_forever(hdr);
224 send_local_file(hdr, "application/x-git-packed-objects", name);
227 static void get_idx_file(struct strbuf *hdr, char *name)
229 select_getanyfile(hdr);
230 hdr_cache_forever(hdr);
231 send_local_file(hdr, "application/x-git-packed-objects-toc", name);
234 static void http_config(void)
236 int i, value = 0;
237 struct strbuf var = STRBUF_INIT;
239 git_config_get_bool("http.getanyfile", &getanyfile);
240 git_config_get_ulong("http.maxrequestbuffer", &max_request_buffer);
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);
250 strbuf_release(&var);
253 static struct rpc_service *select_service(struct strbuf *hdr, const char *name)
255 const char *svc_name;
256 struct rpc_service *svc = NULL;
257 int i;
259 if (!skip_prefix(name, "git-", &svc_name))
260 forbidden(hdr, "Unsupported service: '%s'", name);
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;
270 if (!svc)
271 forbidden(hdr, "Unsupported service: '%s'", name);
273 if (svc->enabled < 0) {
274 const char *user = getenv("REMOTE_USER");
275 svc->enabled = (user && *user) ? 1 : 0;
277 if (!svc->enabled)
278 forbidden(hdr, "Service not enabled: '%s'", svc->name);
279 return svc;
282 static void write_to_child(int out, const unsigned char *buf, ssize_t len, const char *prog_name)
284 if (write_in_full(out, buf, len) < 0)
285 die("unable to write to '%s'", prog_name);
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).
293 static ssize_t read_request_eof(int fd, unsigned char **out)
295 size_t len = 0, alloc = 8192;
296 unsigned char *buf = xmalloc(alloc);
298 if (max_request_buffer < alloc)
299 max_request_buffer = alloc;
301 while (1) {
302 ssize_t cnt;
304 cnt = read_in_full(fd, buf + len, alloc - len);
305 if (cnt < 0) {
306 free(buf);
307 return -1;
310 /* partial read from read_in_full means we hit EOF */
311 len += cnt;
312 if (len < alloc) {
313 *out = buf;
314 return len;
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);
323 alloc = alloc_nr(alloc);
324 if (alloc > max_request_buffer)
325 alloc = max_request_buffer;
326 REALLOC_ARRAY(buf, alloc);
330 static ssize_t read_request_fixed_len(int fd, ssize_t req_len, unsigned char **out)
332 unsigned char *buf = NULL;
333 ssize_t cnt = 0;
335 if (max_request_buffer < req_len) {
336 die("request was larger than our maximum size (%lu): "
337 "%" PRIuMAX "; try setting GIT_HTTP_MAX_REQUEST_BUFFER",
338 max_request_buffer, (uintmax_t)req_len);
341 buf = xmalloc(req_len);
342 cnt = read_in_full(fd, buf, req_len);
343 if (cnt < 0) {
344 free(buf);
345 return -1;
347 *out = buf;
348 return cnt;
351 static ssize_t get_content_length(void)
353 ssize_t val = -1;
354 const char *str = getenv("CONTENT_LENGTH");
356 if (str && *str && !git_parse_ssize_t(str, &val))
357 die("failed to parse CONTENT_LENGTH: %s", str);
358 return val;
361 static ssize_t read_request(int fd, unsigned char **out, ssize_t req_len)
363 if (req_len < 0)
364 return read_request_eof(fd, out);
365 else
366 return read_request_fixed_len(fd, req_len, out);
369 static void inflate_request(const char *prog_name, int out, int buffer_input, ssize_t req_len)
371 git_zstream stream;
372 unsigned char *full_request = NULL;
373 unsigned char in_buf[8192];
374 unsigned char out_buf[8192];
375 unsigned long cnt = 0;
376 int req_len_defined = req_len >= 0;
377 size_t req_remaining_len = req_len;
379 memset(&stream, 0, sizeof(stream));
380 git_inflate_init_gzip_only(&stream);
382 while (1) {
383 ssize_t n;
385 if (buffer_input) {
386 if (full_request)
387 n = 0; /* nothing left to read */
388 else
389 n = read_request(0, &full_request, req_len);
390 stream.next_in = full_request;
391 } else {
392 ssize_t buffer_len;
393 if (req_len_defined && req_remaining_len <= sizeof(in_buf))
394 buffer_len = req_remaining_len;
395 else
396 buffer_len = sizeof(in_buf);
397 n = xread(0, in_buf, buffer_len);
398 stream.next_in = in_buf;
399 if (req_len_defined && n > 0)
400 req_remaining_len -= n;
403 if (n <= 0)
404 die("request ended in the middle of the gzip stream");
405 stream.avail_in = n;
407 while (0 < stream.avail_in) {
408 int ret;
410 stream.next_out = out_buf;
411 stream.avail_out = sizeof(out_buf);
413 ret = git_inflate(&stream, Z_NO_FLUSH);
414 if (ret != Z_OK && ret != Z_STREAM_END)
415 die("zlib error inflating request, result %d", ret);
417 n = stream.total_out - cnt;
418 write_to_child(out, out_buf, stream.total_out - cnt, prog_name);
419 cnt = stream.total_out;
421 if (ret == Z_STREAM_END)
422 goto done;
426 done:
427 git_inflate_end(&stream);
428 close(out);
429 free(full_request);
432 static void copy_request(const char *prog_name, int out, ssize_t req_len)
434 unsigned char *buf;
435 ssize_t n = read_request(0, &buf, req_len);
436 if (n < 0)
437 die_errno("error reading request body");
438 write_to_child(out, buf, n, prog_name);
439 close(out);
440 free(buf);
443 static void pipe_fixed_length(const char *prog_name, int out, size_t req_len)
445 unsigned char buf[8192];
446 size_t remaining_len = req_len;
448 while (remaining_len > 0) {
449 size_t chunk_length = remaining_len > sizeof(buf) ? sizeof(buf) : remaining_len;
450 ssize_t n = xread(0, buf, chunk_length);
451 if (n < 0)
452 die_errno("Reading request failed");
453 write_to_child(out, buf, n, prog_name);
454 remaining_len -= n;
457 close(out);
460 static void run_service(const char **argv, int buffer_input)
462 const char *encoding = getenv("HTTP_CONTENT_ENCODING");
463 const char *user = getenv("REMOTE_USER");
464 const char *host = getenv("REMOTE_ADDR");
465 int gzipped_request = 0;
466 struct child_process cld = CHILD_PROCESS_INIT;
467 ssize_t req_len = get_content_length();
469 if (encoding && !strcmp(encoding, "gzip"))
470 gzipped_request = 1;
471 else if (encoding && !strcmp(encoding, "x-gzip"))
472 gzipped_request = 1;
474 if (!user || !*user)
475 user = "anonymous";
476 if (!host || !*host)
477 host = "(none)";
479 if (!getenv("GIT_COMMITTER_NAME"))
480 argv_array_pushf(&cld.env_array, "GIT_COMMITTER_NAME=%s", user);
481 if (!getenv("GIT_COMMITTER_EMAIL"))
482 argv_array_pushf(&cld.env_array,
483 "GIT_COMMITTER_EMAIL=%s@http.%s", user, host);
485 cld.argv = argv;
486 if (buffer_input || gzipped_request || req_len >= 0)
487 cld.in = -1;
488 cld.git_cmd = 1;
489 cld.clean_on_exit = 1;
490 cld.wait_after_clean = 1;
491 if (start_command(&cld))
492 exit(1);
494 close(1);
495 if (gzipped_request)
496 inflate_request(argv[0], cld.in, buffer_input, req_len);
497 else if (buffer_input)
498 copy_request(argv[0], cld.in, req_len);
499 else if (req_len >= 0)
500 pipe_fixed_length(argv[0], cld.in, req_len);
501 else
502 close(0);
504 if (finish_command(&cld))
505 exit(1);
508 static int show_text_ref(const char *name, const struct object_id *oid,
509 int flag, void *cb_data)
511 const char *name_nons = strip_namespace(name);
512 struct strbuf *buf = cb_data;
513 struct object *o = parse_object(the_repository, oid);
514 if (!o)
515 return 0;
517 strbuf_addf(buf, "%s\t%s\n", oid_to_hex(oid), name_nons);
518 if (o->type == OBJ_TAG) {
519 o = deref_tag(the_repository, o, name, 0);
520 if (!o)
521 return 0;
522 strbuf_addf(buf, "%s\t%s^{}\n", oid_to_hex(&o->oid),
523 name_nons);
525 return 0;
528 static void get_info_refs(struct strbuf *hdr, char *arg)
530 const char *service_name = get_parameter("service");
531 struct strbuf buf = STRBUF_INIT;
533 hdr_nocache(hdr);
535 if (service_name) {
536 const char *argv[] = {NULL /* service name */,
537 "--stateless-rpc", "--advertise-refs",
538 ".", NULL};
539 struct rpc_service *svc = select_service(hdr, service_name);
541 strbuf_addf(&buf, "application/x-git-%s-advertisement",
542 svc->name);
543 hdr_str(hdr, content_type, buf.buf);
544 end_headers(hdr);
547 if (determine_protocol_version_server() != protocol_v2) {
548 packet_write_fmt(1, "# service=git-%s\n", svc->name);
549 packet_flush(1);
552 argv[0] = svc->name;
553 run_service(argv, 0);
555 } else {
556 select_getanyfile(hdr);
557 for_each_namespaced_ref(show_text_ref, &buf);
558 send_strbuf(hdr, "text/plain", &buf);
560 strbuf_release(&buf);
563 static int show_head_ref(const char *refname, const struct object_id *oid,
564 int flag, void *cb_data)
566 struct strbuf *buf = cb_data;
568 if (flag & REF_ISSYMREF) {
569 const char *target = resolve_ref_unsafe(refname,
570 RESOLVE_REF_READING,
571 NULL, NULL);
573 if (target)
574 strbuf_addf(buf, "ref: %s\n", strip_namespace(target));
575 } else {
576 strbuf_addf(buf, "%s\n", oid_to_hex(oid));
579 return 0;
582 static void get_head(struct strbuf *hdr, char *arg)
584 struct strbuf buf = STRBUF_INIT;
586 select_getanyfile(hdr);
587 head_ref_namespaced(show_head_ref, &buf);
588 send_strbuf(hdr, "text/plain", &buf);
589 strbuf_release(&buf);
592 static void get_info_packs(struct strbuf *hdr, char *arg)
594 size_t objdirlen = strlen(get_object_directory());
595 struct strbuf buf = STRBUF_INIT;
596 struct packed_git *p;
597 size_t cnt = 0;
599 select_getanyfile(hdr);
600 for (p = get_all_packs(the_repository); p; p = p->next) {
601 if (p->pack_local)
602 cnt++;
605 strbuf_grow(&buf, cnt * 53 + 2);
606 for (p = get_all_packs(the_repository); p; p = p->next) {
607 if (p->pack_local)
608 strbuf_addf(&buf, "P %s\n", p->pack_name + objdirlen + 6);
610 strbuf_addch(&buf, '\n');
612 hdr_nocache(hdr);
613 send_strbuf(hdr, "text/plain; charset=utf-8", &buf);
614 strbuf_release(&buf);
617 static void check_content_type(struct strbuf *hdr, const char *accepted_type)
619 const char *actual_type = getenv("CONTENT_TYPE");
621 if (!actual_type)
622 actual_type = "";
624 if (strcmp(actual_type, accepted_type)) {
625 http_status(hdr, 415, "Unsupported Media Type");
626 hdr_nocache(hdr);
627 end_headers(hdr);
628 format_write(1,
629 "Expected POST with Content-Type '%s',"
630 " but received '%s' instead.\n",
631 accepted_type, actual_type);
632 exit(0);
636 static void service_rpc(struct strbuf *hdr, char *service_name)
638 const char *argv[] = {NULL, "--stateless-rpc", ".", NULL};
639 struct rpc_service *svc = select_service(hdr, service_name);
640 struct strbuf buf = STRBUF_INIT;
642 strbuf_reset(&buf);
643 strbuf_addf(&buf, "application/x-git-%s-request", svc->name);
644 check_content_type(hdr, buf.buf);
646 hdr_nocache(hdr);
648 strbuf_reset(&buf);
649 strbuf_addf(&buf, "application/x-git-%s-result", svc->name);
650 hdr_str(hdr, content_type, buf.buf);
652 end_headers(hdr);
654 argv[0] = svc->name;
655 run_service(argv, svc->buffer_input);
656 strbuf_release(&buf);
659 static int dead;
660 static NORETURN void die_webcgi(const char *err, va_list params)
662 if (dead <= 1) {
663 struct strbuf hdr = STRBUF_INIT;
665 vreportf("fatal: ", err, params);
667 http_status(&hdr, 500, "Internal Server Error");
668 hdr_nocache(&hdr);
669 end_headers(&hdr);
671 exit(0); /* we successfully reported a failure ;-) */
674 static int die_webcgi_recursing(void)
676 return dead++ > 1;
679 static char* getdir(void)
681 struct strbuf buf = STRBUF_INIT;
682 char *pathinfo = getenv("PATH_INFO");
683 char *root = getenv("GIT_PROJECT_ROOT");
684 char *path = getenv("PATH_TRANSLATED");
686 if (root && *root) {
687 if (!pathinfo || !*pathinfo)
688 die("GIT_PROJECT_ROOT is set but PATH_INFO is not");
689 if (daemon_avoid_alias(pathinfo))
690 die("'%s': aliased", pathinfo);
691 end_url_with_slash(&buf, root);
692 if (pathinfo[0] == '/')
693 pathinfo++;
694 strbuf_addstr(&buf, pathinfo);
695 return strbuf_detach(&buf, NULL);
696 } else if (path && *path) {
697 return xstrdup(path);
698 } else
699 die("No GIT_PROJECT_ROOT or PATH_TRANSLATED from server");
700 return NULL;
703 static struct service_cmd {
704 const char *method;
705 const char *pattern;
706 void (*imp)(struct strbuf *, char *);
707 } services[] = {
708 {"GET", "/HEAD$", get_head},
709 {"GET", "/info/refs$", get_info_refs},
710 {"GET", "/objects/info/alternates$", get_text_file},
711 {"GET", "/objects/info/http-alternates$", get_text_file},
712 {"GET", "/objects/info/packs$", get_info_packs},
713 {"GET", "/objects/[0-9a-f]{2}/[0-9a-f]{38}$", get_loose_object},
714 {"GET", "/objects/[0-9a-f]{2}/[0-9a-f]{62}$", get_loose_object},
715 {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.pack$", get_pack_file},
716 {"GET", "/objects/pack/pack-[0-9a-f]{64}\\.pack$", get_pack_file},
717 {"GET", "/objects/pack/pack-[0-9a-f]{40}\\.idx$", get_idx_file},
718 {"GET", "/objects/pack/pack-[0-9a-f]{64}\\.idx$", get_idx_file},
720 {"POST", "/git-upload-pack$", service_rpc},
721 {"POST", "/git-receive-pack$", service_rpc}
724 static int bad_request(struct strbuf *hdr, const struct service_cmd *c)
726 const char *proto = getenv("SERVER_PROTOCOL");
728 if (proto && !strcmp(proto, "HTTP/1.1")) {
729 http_status(hdr, 405, "Method Not Allowed");
730 hdr_str(hdr, "Allow",
731 !strcmp(c->method, "GET") ? "GET, HEAD" : c->method);
732 } else
733 http_status(hdr, 400, "Bad Request");
734 hdr_nocache(hdr);
735 end_headers(hdr);
736 return 0;
739 int cmd_main(int argc, const char **argv)
741 char *method = getenv("REQUEST_METHOD");
742 char *dir;
743 struct service_cmd *cmd = NULL;
744 char *cmd_arg = NULL;
745 int i;
746 struct strbuf hdr = STRBUF_INIT;
748 set_die_routine(die_webcgi);
749 set_die_is_recursing_routine(die_webcgi_recursing);
751 if (!method)
752 die("No REQUEST_METHOD from server");
753 if (!strcmp(method, "HEAD"))
754 method = "GET";
755 dir = getdir();
757 for (i = 0; i < ARRAY_SIZE(services); i++) {
758 struct service_cmd *c = &services[i];
759 regex_t re;
760 regmatch_t out[1];
762 if (regcomp(&re, c->pattern, REG_EXTENDED))
763 die("Bogus regex in service table: %s", c->pattern);
764 if (!regexec(&re, dir, 1, out, 0)) {
765 size_t n;
767 if (strcmp(method, c->method))
768 return bad_request(&hdr, c);
770 cmd = c;
771 n = out[0].rm_eo - out[0].rm_so;
772 cmd_arg = xmemdupz(dir + out[0].rm_so + 1, n - 1);
773 dir[out[0].rm_so] = 0;
774 break;
776 regfree(&re);
779 if (!cmd)
780 not_found(&hdr, "Request not supported: '%s'", dir);
782 setup_path();
783 if (!enter_repo(dir, 0))
784 not_found(&hdr, "Not a git repository: '%s'", dir);
785 if (!getenv("GIT_HTTP_EXPORT_ALL") &&
786 access("git-daemon-export-ok", F_OK) )
787 not_found(&hdr, "Repository not exported: '%s'", dir);
789 http_config();
790 max_request_buffer = git_env_ulong("GIT_HTTP_MAX_REQUEST_BUFFER",
791 max_request_buffer);
793 cmd->imp(&hdr, cmd_arg);
794 return 0;