git-fetch-pack uses URLs like git-fetch
[git/gitweb.git] / http.c
blob3447945ee4f4e0e4b273d979ea1805b4c0f63b4c
1 #include "http.h"
2 #include "pack.h"
3 #include "sideband.h"
4 #include "run-command.h"
5 #include "url.h"
6 #include "credential.h"
7 #include "version.h"
8 #include "pkt-line.h"
10 int active_requests;
11 int http_is_verbose;
12 size_t http_post_buffer = 16 * LARGE_PACKET_MAX;
14 #if LIBCURL_VERSION_NUM >= 0x070a06
15 #define LIBCURL_CAN_HANDLE_AUTH_ANY
16 #endif
18 static int min_curl_sessions = 1;
19 static int curl_session_count;
20 #ifdef USE_CURL_MULTI
21 static int max_requests = -1;
22 static CURLM *curlm;
23 #endif
24 #ifndef NO_CURL_EASY_DUPHANDLE
25 static CURL *curl_default;
26 #endif
28 #define PREV_BUF_SIZE 4096
29 #define RANGE_HEADER_SIZE 30
31 char curl_errorstr[CURL_ERROR_SIZE];
33 static int curl_ssl_verify = -1;
34 static int curl_ssl_try;
35 static const char *ssl_cert;
36 #if LIBCURL_VERSION_NUM >= 0x070903
37 static const char *ssl_key;
38 #endif
39 #if LIBCURL_VERSION_NUM >= 0x070908
40 static const char *ssl_capath;
41 #endif
42 static const char *ssl_cainfo;
43 static long curl_low_speed_limit = -1;
44 static long curl_low_speed_time = -1;
45 static int curl_ftp_no_epsv;
46 static const char *curl_http_proxy;
47 static const char *curl_cookie_file;
48 struct credential http_auth = CREDENTIAL_INIT;
49 static int http_proactive_auth;
50 static const char *user_agent;
52 #if LIBCURL_VERSION_NUM >= 0x071700
53 /* Use CURLOPT_KEYPASSWD as is */
54 #elif LIBCURL_VERSION_NUM >= 0x070903
55 #define CURLOPT_KEYPASSWD CURLOPT_SSLKEYPASSWD
56 #else
57 #define CURLOPT_KEYPASSWD CURLOPT_SSLCERTPASSWD
58 #endif
60 static struct credential cert_auth = CREDENTIAL_INIT;
61 static int ssl_cert_password_required;
63 static struct curl_slist *pragma_header;
64 static struct curl_slist *no_pragma_header;
66 static struct active_request_slot *active_queue_head;
68 size_t fread_buffer(char *ptr, size_t eltsize, size_t nmemb, void *buffer_)
70 size_t size = eltsize * nmemb;
71 struct buffer *buffer = buffer_;
73 if (size > buffer->buf.len - buffer->posn)
74 size = buffer->buf.len - buffer->posn;
75 memcpy(ptr, buffer->buf.buf + buffer->posn, size);
76 buffer->posn += size;
78 return size;
81 #ifndef NO_CURL_IOCTL
82 curlioerr ioctl_buffer(CURL *handle, int cmd, void *clientp)
84 struct buffer *buffer = clientp;
86 switch (cmd) {
87 case CURLIOCMD_NOP:
88 return CURLIOE_OK;
90 case CURLIOCMD_RESTARTREAD:
91 buffer->posn = 0;
92 return CURLIOE_OK;
94 default:
95 return CURLIOE_UNKNOWNCMD;
98 #endif
100 size_t fwrite_buffer(char *ptr, size_t eltsize, size_t nmemb, void *buffer_)
102 size_t size = eltsize * nmemb;
103 struct strbuf *buffer = buffer_;
105 strbuf_add(buffer, ptr, size);
106 return size;
109 size_t fwrite_null(char *ptr, size_t eltsize, size_t nmemb, void *strbuf)
111 return eltsize * nmemb;
114 #ifdef USE_CURL_MULTI
115 static void process_curl_messages(void)
117 int num_messages;
118 struct active_request_slot *slot;
119 CURLMsg *curl_message = curl_multi_info_read(curlm, &num_messages);
121 while (curl_message != NULL) {
122 if (curl_message->msg == CURLMSG_DONE) {
123 int curl_result = curl_message->data.result;
124 slot = active_queue_head;
125 while (slot != NULL &&
126 slot->curl != curl_message->easy_handle)
127 slot = slot->next;
128 if (slot != NULL) {
129 curl_multi_remove_handle(curlm, slot->curl);
130 slot->curl_result = curl_result;
131 finish_active_slot(slot);
132 } else {
133 fprintf(stderr, "Received DONE message for unknown request!\n");
135 } else {
136 fprintf(stderr, "Unknown CURL message received: %d\n",
137 (int)curl_message->msg);
139 curl_message = curl_multi_info_read(curlm, &num_messages);
142 #endif
144 static int http_options(const char *var, const char *value, void *cb)
146 if (!strcmp("http.sslverify", var)) {
147 curl_ssl_verify = git_config_bool(var, value);
148 return 0;
150 if (!strcmp("http.sslcert", var))
151 return git_config_string(&ssl_cert, var, value);
152 #if LIBCURL_VERSION_NUM >= 0x070903
153 if (!strcmp("http.sslkey", var))
154 return git_config_string(&ssl_key, var, value);
155 #endif
156 #if LIBCURL_VERSION_NUM >= 0x070908
157 if (!strcmp("http.sslcapath", var))
158 return git_config_string(&ssl_capath, var, value);
159 #endif
160 if (!strcmp("http.sslcainfo", var))
161 return git_config_string(&ssl_cainfo, var, value);
162 if (!strcmp("http.sslcertpasswordprotected", var)) {
163 if (git_config_bool(var, value))
164 ssl_cert_password_required = 1;
165 return 0;
167 if (!strcmp("http.ssltry", var)) {
168 curl_ssl_try = git_config_bool(var, value);
169 return 0;
171 if (!strcmp("http.minsessions", var)) {
172 min_curl_sessions = git_config_int(var, value);
173 #ifndef USE_CURL_MULTI
174 if (min_curl_sessions > 1)
175 min_curl_sessions = 1;
176 #endif
177 return 0;
179 #ifdef USE_CURL_MULTI
180 if (!strcmp("http.maxrequests", var)) {
181 max_requests = git_config_int(var, value);
182 return 0;
184 #endif
185 if (!strcmp("http.lowspeedlimit", var)) {
186 curl_low_speed_limit = (long)git_config_int(var, value);
187 return 0;
189 if (!strcmp("http.lowspeedtime", var)) {
190 curl_low_speed_time = (long)git_config_int(var, value);
191 return 0;
194 if (!strcmp("http.noepsv", var)) {
195 curl_ftp_no_epsv = git_config_bool(var, value);
196 return 0;
198 if (!strcmp("http.proxy", var))
199 return git_config_string(&curl_http_proxy, var, value);
201 if (!strcmp("http.cookiefile", var))
202 return git_config_string(&curl_cookie_file, var, value);
204 if (!strcmp("http.postbuffer", var)) {
205 http_post_buffer = git_config_int(var, value);
206 if (http_post_buffer < LARGE_PACKET_MAX)
207 http_post_buffer = LARGE_PACKET_MAX;
208 return 0;
211 if (!strcmp("http.useragent", var))
212 return git_config_string(&user_agent, var, value);
214 /* Fall back on the default ones */
215 return git_default_config(var, value, cb);
218 static void init_curl_http_auth(CURL *result)
220 if (!http_auth.username)
221 return;
223 credential_fill(&http_auth);
225 #if LIBCURL_VERSION_NUM >= 0x071301
226 curl_easy_setopt(result, CURLOPT_USERNAME, http_auth.username);
227 curl_easy_setopt(result, CURLOPT_PASSWORD, http_auth.password);
228 #else
230 static struct strbuf up = STRBUF_INIT;
232 * Note that we assume we only ever have a single set of
233 * credentials in a given program run, so we do not have
234 * to worry about updating this buffer, only setting its
235 * initial value.
237 if (!up.len)
238 strbuf_addf(&up, "%s:%s",
239 http_auth.username, http_auth.password);
240 curl_easy_setopt(result, CURLOPT_USERPWD, up.buf);
242 #endif
245 static int has_cert_password(void)
247 if (ssl_cert == NULL || ssl_cert_password_required != 1)
248 return 0;
249 if (!cert_auth.password) {
250 cert_auth.protocol = xstrdup("cert");
251 cert_auth.username = xstrdup("");
252 cert_auth.path = xstrdup(ssl_cert);
253 credential_fill(&cert_auth);
255 return 1;
258 static CURL *get_curl_handle(void)
260 CURL *result = curl_easy_init();
262 if (!curl_ssl_verify) {
263 curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 0);
264 curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 0);
265 } else {
266 /* Verify authenticity of the peer's certificate */
267 curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 1);
268 /* The name in the cert must match whom we tried to connect */
269 curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 2);
272 #if LIBCURL_VERSION_NUM >= 0x070907
273 curl_easy_setopt(result, CURLOPT_NETRC, CURL_NETRC_OPTIONAL);
274 #endif
275 #ifdef LIBCURL_CAN_HANDLE_AUTH_ANY
276 curl_easy_setopt(result, CURLOPT_HTTPAUTH, CURLAUTH_ANY);
277 #endif
279 if (http_proactive_auth)
280 init_curl_http_auth(result);
282 if (ssl_cert != NULL)
283 curl_easy_setopt(result, CURLOPT_SSLCERT, ssl_cert);
284 if (has_cert_password())
285 curl_easy_setopt(result, CURLOPT_KEYPASSWD, cert_auth.password);
286 #if LIBCURL_VERSION_NUM >= 0x070903
287 if (ssl_key != NULL)
288 curl_easy_setopt(result, CURLOPT_SSLKEY, ssl_key);
289 #endif
290 #if LIBCURL_VERSION_NUM >= 0x070908
291 if (ssl_capath != NULL)
292 curl_easy_setopt(result, CURLOPT_CAPATH, ssl_capath);
293 #endif
294 if (ssl_cainfo != NULL)
295 curl_easy_setopt(result, CURLOPT_CAINFO, ssl_cainfo);
297 if (curl_low_speed_limit > 0 && curl_low_speed_time > 0) {
298 curl_easy_setopt(result, CURLOPT_LOW_SPEED_LIMIT,
299 curl_low_speed_limit);
300 curl_easy_setopt(result, CURLOPT_LOW_SPEED_TIME,
301 curl_low_speed_time);
304 curl_easy_setopt(result, CURLOPT_FOLLOWLOCATION, 1);
305 #if LIBCURL_VERSION_NUM >= 0x071301
306 curl_easy_setopt(result, CURLOPT_POSTREDIR, CURL_REDIR_POST_ALL);
307 #elif LIBCURL_VERSION_NUM >= 0x071101
308 curl_easy_setopt(result, CURLOPT_POST301, 1);
309 #endif
311 if (getenv("GIT_CURL_VERBOSE"))
312 curl_easy_setopt(result, CURLOPT_VERBOSE, 1);
314 curl_easy_setopt(result, CURLOPT_USERAGENT,
315 user_agent ? user_agent : git_user_agent());
317 if (curl_ftp_no_epsv)
318 curl_easy_setopt(result, CURLOPT_FTP_USE_EPSV, 0);
320 #ifdef CURLOPT_USE_SSL
321 if (curl_ssl_try)
322 curl_easy_setopt(result, CURLOPT_USE_SSL, CURLUSESSL_TRY);
323 #endif
325 if (curl_http_proxy) {
326 curl_easy_setopt(result, CURLOPT_PROXY, curl_http_proxy);
327 curl_easy_setopt(result, CURLOPT_PROXYAUTH, CURLAUTH_ANY);
330 return result;
333 static void set_from_env(const char **var, const char *envname)
335 const char *val = getenv(envname);
336 if (val)
337 *var = val;
340 void http_init(struct remote *remote, const char *url, int proactive_auth)
342 char *low_speed_limit;
343 char *low_speed_time;
345 http_is_verbose = 0;
347 git_config(http_options, NULL);
349 curl_global_init(CURL_GLOBAL_ALL);
351 http_proactive_auth = proactive_auth;
353 if (remote && remote->http_proxy)
354 curl_http_proxy = xstrdup(remote->http_proxy);
356 pragma_header = curl_slist_append(pragma_header, "Pragma: no-cache");
357 no_pragma_header = curl_slist_append(no_pragma_header, "Pragma:");
359 #ifdef USE_CURL_MULTI
361 char *http_max_requests = getenv("GIT_HTTP_MAX_REQUESTS");
362 if (http_max_requests != NULL)
363 max_requests = atoi(http_max_requests);
366 curlm = curl_multi_init();
367 if (curlm == NULL) {
368 fprintf(stderr, "Error creating curl multi handle.\n");
369 exit(1);
371 #endif
373 if (getenv("GIT_SSL_NO_VERIFY"))
374 curl_ssl_verify = 0;
376 set_from_env(&ssl_cert, "GIT_SSL_CERT");
377 #if LIBCURL_VERSION_NUM >= 0x070903
378 set_from_env(&ssl_key, "GIT_SSL_KEY");
379 #endif
380 #if LIBCURL_VERSION_NUM >= 0x070908
381 set_from_env(&ssl_capath, "GIT_SSL_CAPATH");
382 #endif
383 set_from_env(&ssl_cainfo, "GIT_SSL_CAINFO");
385 set_from_env(&user_agent, "GIT_HTTP_USER_AGENT");
387 low_speed_limit = getenv("GIT_HTTP_LOW_SPEED_LIMIT");
388 if (low_speed_limit != NULL)
389 curl_low_speed_limit = strtol(low_speed_limit, NULL, 10);
390 low_speed_time = getenv("GIT_HTTP_LOW_SPEED_TIME");
391 if (low_speed_time != NULL)
392 curl_low_speed_time = strtol(low_speed_time, NULL, 10);
394 if (curl_ssl_verify == -1)
395 curl_ssl_verify = 1;
397 curl_session_count = 0;
398 #ifdef USE_CURL_MULTI
399 if (max_requests < 1)
400 max_requests = DEFAULT_MAX_REQUESTS;
401 #endif
403 if (getenv("GIT_CURL_FTP_NO_EPSV"))
404 curl_ftp_no_epsv = 1;
406 if (url) {
407 credential_from_url(&http_auth, url);
408 if (!ssl_cert_password_required &&
409 getenv("GIT_SSL_CERT_PASSWORD_PROTECTED") &&
410 !prefixcmp(url, "https://"))
411 ssl_cert_password_required = 1;
414 #ifndef NO_CURL_EASY_DUPHANDLE
415 curl_default = get_curl_handle();
416 #endif
419 void http_cleanup(void)
421 struct active_request_slot *slot = active_queue_head;
423 while (slot != NULL) {
424 struct active_request_slot *next = slot->next;
425 if (slot->curl != NULL) {
426 #ifdef USE_CURL_MULTI
427 curl_multi_remove_handle(curlm, slot->curl);
428 #endif
429 curl_easy_cleanup(slot->curl);
431 free(slot);
432 slot = next;
434 active_queue_head = NULL;
436 #ifndef NO_CURL_EASY_DUPHANDLE
437 curl_easy_cleanup(curl_default);
438 #endif
440 #ifdef USE_CURL_MULTI
441 curl_multi_cleanup(curlm);
442 #endif
443 curl_global_cleanup();
445 curl_slist_free_all(pragma_header);
446 pragma_header = NULL;
448 curl_slist_free_all(no_pragma_header);
449 no_pragma_header = NULL;
451 if (curl_http_proxy) {
452 free((void *)curl_http_proxy);
453 curl_http_proxy = NULL;
456 if (cert_auth.password != NULL) {
457 memset(cert_auth.password, 0, strlen(cert_auth.password));
458 free(cert_auth.password);
459 cert_auth.password = NULL;
461 ssl_cert_password_required = 0;
464 struct active_request_slot *get_active_slot(void)
466 struct active_request_slot *slot = active_queue_head;
467 struct active_request_slot *newslot;
469 #ifdef USE_CURL_MULTI
470 int num_transfers;
472 /* Wait for a slot to open up if the queue is full */
473 while (active_requests >= max_requests) {
474 curl_multi_perform(curlm, &num_transfers);
475 if (num_transfers < active_requests)
476 process_curl_messages();
478 #endif
480 while (slot != NULL && slot->in_use)
481 slot = slot->next;
483 if (slot == NULL) {
484 newslot = xmalloc(sizeof(*newslot));
485 newslot->curl = NULL;
486 newslot->in_use = 0;
487 newslot->next = NULL;
489 slot = active_queue_head;
490 if (slot == NULL) {
491 active_queue_head = newslot;
492 } else {
493 while (slot->next != NULL)
494 slot = slot->next;
495 slot->next = newslot;
497 slot = newslot;
500 if (slot->curl == NULL) {
501 #ifdef NO_CURL_EASY_DUPHANDLE
502 slot->curl = get_curl_handle();
503 #else
504 slot->curl = curl_easy_duphandle(curl_default);
505 #endif
506 curl_session_count++;
509 active_requests++;
510 slot->in_use = 1;
511 slot->results = NULL;
512 slot->finished = NULL;
513 slot->callback_data = NULL;
514 slot->callback_func = NULL;
515 curl_easy_setopt(slot->curl, CURLOPT_COOKIEFILE, curl_cookie_file);
516 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, pragma_header);
517 curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, curl_errorstr);
518 curl_easy_setopt(slot->curl, CURLOPT_CUSTOMREQUEST, NULL);
519 curl_easy_setopt(slot->curl, CURLOPT_READFUNCTION, NULL);
520 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, NULL);
521 curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, NULL);
522 curl_easy_setopt(slot->curl, CURLOPT_UPLOAD, 0);
523 curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1);
524 curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 1);
525 if (http_auth.password)
526 init_curl_http_auth(slot->curl);
528 return slot;
531 int start_active_slot(struct active_request_slot *slot)
533 #ifdef USE_CURL_MULTI
534 CURLMcode curlm_result = curl_multi_add_handle(curlm, slot->curl);
535 int num_transfers;
537 if (curlm_result != CURLM_OK &&
538 curlm_result != CURLM_CALL_MULTI_PERFORM) {
539 active_requests--;
540 slot->in_use = 0;
541 return 0;
545 * We know there must be something to do, since we just added
546 * something.
548 curl_multi_perform(curlm, &num_transfers);
549 #endif
550 return 1;
553 #ifdef USE_CURL_MULTI
554 struct fill_chain {
555 void *data;
556 int (*fill)(void *);
557 struct fill_chain *next;
560 static struct fill_chain *fill_cfg;
562 void add_fill_function(void *data, int (*fill)(void *))
564 struct fill_chain *new = xmalloc(sizeof(*new));
565 struct fill_chain **linkp = &fill_cfg;
566 new->data = data;
567 new->fill = fill;
568 new->next = NULL;
569 while (*linkp)
570 linkp = &(*linkp)->next;
571 *linkp = new;
574 void fill_active_slots(void)
576 struct active_request_slot *slot = active_queue_head;
578 while (active_requests < max_requests) {
579 struct fill_chain *fill;
580 for (fill = fill_cfg; fill; fill = fill->next)
581 if (fill->fill(fill->data))
582 break;
584 if (!fill)
585 break;
588 while (slot != NULL) {
589 if (!slot->in_use && slot->curl != NULL
590 && curl_session_count > min_curl_sessions) {
591 curl_easy_cleanup(slot->curl);
592 slot->curl = NULL;
593 curl_session_count--;
595 slot = slot->next;
599 void step_active_slots(void)
601 int num_transfers;
602 CURLMcode curlm_result;
604 do {
605 curlm_result = curl_multi_perform(curlm, &num_transfers);
606 } while (curlm_result == CURLM_CALL_MULTI_PERFORM);
607 if (num_transfers < active_requests) {
608 process_curl_messages();
609 fill_active_slots();
612 #endif
614 void run_active_slot(struct active_request_slot *slot)
616 #ifdef USE_CURL_MULTI
617 fd_set readfds;
618 fd_set writefds;
619 fd_set excfds;
620 int max_fd;
621 struct timeval select_timeout;
622 int finished = 0;
624 slot->finished = &finished;
625 while (!finished) {
626 step_active_slots();
628 if (slot->in_use) {
629 #if LIBCURL_VERSION_NUM >= 0x070f04
630 long curl_timeout;
631 curl_multi_timeout(curlm, &curl_timeout);
632 if (curl_timeout == 0) {
633 continue;
634 } else if (curl_timeout == -1) {
635 select_timeout.tv_sec = 0;
636 select_timeout.tv_usec = 50000;
637 } else {
638 select_timeout.tv_sec = curl_timeout / 1000;
639 select_timeout.tv_usec = (curl_timeout % 1000) * 1000;
641 #else
642 select_timeout.tv_sec = 0;
643 select_timeout.tv_usec = 50000;
644 #endif
646 max_fd = -1;
647 FD_ZERO(&readfds);
648 FD_ZERO(&writefds);
649 FD_ZERO(&excfds);
650 curl_multi_fdset(curlm, &readfds, &writefds, &excfds, &max_fd);
653 * It can happen that curl_multi_timeout returns a pathologically
654 * long timeout when curl_multi_fdset returns no file descriptors
655 * to read. See commit message for more details.
657 if (max_fd < 0 &&
658 (select_timeout.tv_sec > 0 ||
659 select_timeout.tv_usec > 50000)) {
660 select_timeout.tv_sec = 0;
661 select_timeout.tv_usec = 50000;
664 select(max_fd+1, &readfds, &writefds, &excfds, &select_timeout);
667 #else
668 while (slot->in_use) {
669 slot->curl_result = curl_easy_perform(slot->curl);
670 finish_active_slot(slot);
672 #endif
675 static void closedown_active_slot(struct active_request_slot *slot)
677 active_requests--;
678 slot->in_use = 0;
681 static void release_active_slot(struct active_request_slot *slot)
683 closedown_active_slot(slot);
684 if (slot->curl && curl_session_count > min_curl_sessions) {
685 #ifdef USE_CURL_MULTI
686 curl_multi_remove_handle(curlm, slot->curl);
687 #endif
688 curl_easy_cleanup(slot->curl);
689 slot->curl = NULL;
690 curl_session_count--;
692 #ifdef USE_CURL_MULTI
693 fill_active_slots();
694 #endif
697 void finish_active_slot(struct active_request_slot *slot)
699 closedown_active_slot(slot);
700 curl_easy_getinfo(slot->curl, CURLINFO_HTTP_CODE, &slot->http_code);
702 if (slot->finished != NULL)
703 (*slot->finished) = 1;
705 /* Store slot results so they can be read after the slot is reused */
706 if (slot->results != NULL) {
707 slot->results->curl_result = slot->curl_result;
708 slot->results->http_code = slot->http_code;
711 /* Run callback if appropriate */
712 if (slot->callback_func != NULL)
713 slot->callback_func(slot->callback_data);
716 void finish_all_active_slots(void)
718 struct active_request_slot *slot = active_queue_head;
720 while (slot != NULL)
721 if (slot->in_use) {
722 run_active_slot(slot);
723 slot = active_queue_head;
724 } else {
725 slot = slot->next;
729 /* Helpers for modifying and creating URLs */
730 static inline int needs_quote(int ch)
732 if (((ch >= 'A') && (ch <= 'Z'))
733 || ((ch >= 'a') && (ch <= 'z'))
734 || ((ch >= '0') && (ch <= '9'))
735 || (ch == '/')
736 || (ch == '-')
737 || (ch == '.'))
738 return 0;
739 return 1;
742 static char *quote_ref_url(const char *base, const char *ref)
744 struct strbuf buf = STRBUF_INIT;
745 const char *cp;
746 int ch;
748 end_url_with_slash(&buf, base);
750 for (cp = ref; (ch = *cp) != 0; cp++)
751 if (needs_quote(ch))
752 strbuf_addf(&buf, "%%%02x", ch);
753 else
754 strbuf_addch(&buf, *cp);
756 return strbuf_detach(&buf, NULL);
759 void append_remote_object_url(struct strbuf *buf, const char *url,
760 const char *hex,
761 int only_two_digit_prefix)
763 end_url_with_slash(buf, url);
765 strbuf_addf(buf, "objects/%.*s/", 2, hex);
766 if (!only_two_digit_prefix)
767 strbuf_addf(buf, "%s", hex+2);
770 char *get_remote_object_url(const char *url, const char *hex,
771 int only_two_digit_prefix)
773 struct strbuf buf = STRBUF_INIT;
774 append_remote_object_url(&buf, url, hex, only_two_digit_prefix);
775 return strbuf_detach(&buf, NULL);
778 int handle_curl_result(struct slot_results *results)
781 * If we see a failing http code with CURLE_OK, we have turned off
782 * FAILONERROR (to keep the server's custom error response), and should
783 * translate the code into failure here.
785 if (results->curl_result == CURLE_OK &&
786 results->http_code >= 400) {
787 results->curl_result = CURLE_HTTP_RETURNED_ERROR;
789 * Normally curl will already have put the "reason phrase"
790 * from the server into curl_errorstr; unfortunately without
791 * FAILONERROR it is lost, so we can give only the numeric
792 * status code.
794 snprintf(curl_errorstr, sizeof(curl_errorstr),
795 "The requested URL returned error: %ld",
796 results->http_code);
799 if (results->curl_result == CURLE_OK) {
800 credential_approve(&http_auth);
801 return HTTP_OK;
802 } else if (missing_target(results))
803 return HTTP_MISSING_TARGET;
804 else if (results->http_code == 401) {
805 if (http_auth.username && http_auth.password) {
806 credential_reject(&http_auth);
807 return HTTP_NOAUTH;
808 } else {
809 return HTTP_REAUTH;
811 } else {
812 #if LIBCURL_VERSION_NUM >= 0x070c00
813 if (!curl_errorstr[0])
814 strlcpy(curl_errorstr,
815 curl_easy_strerror(results->curl_result),
816 sizeof(curl_errorstr));
817 #endif
818 return HTTP_ERROR;
822 static CURLcode curlinfo_strbuf(CURL *curl, CURLINFO info, struct strbuf *buf)
824 char *ptr;
825 CURLcode ret;
827 strbuf_reset(buf);
828 ret = curl_easy_getinfo(curl, info, &ptr);
829 if (!ret && ptr)
830 strbuf_addstr(buf, ptr);
831 return ret;
834 /* http_request() targets */
835 #define HTTP_REQUEST_STRBUF 0
836 #define HTTP_REQUEST_FILE 1
838 static int http_request(const char *url,
839 void *result, int target,
840 const struct http_get_options *options)
842 struct active_request_slot *slot;
843 struct slot_results results;
844 struct curl_slist *headers = NULL;
845 struct strbuf buf = STRBUF_INIT;
846 int ret;
848 slot = get_active_slot();
849 slot->results = &results;
850 curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1);
852 if (result == NULL) {
853 curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 1);
854 } else {
855 curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0);
856 curl_easy_setopt(slot->curl, CURLOPT_FILE, result);
858 if (target == HTTP_REQUEST_FILE) {
859 long posn = ftell(result);
860 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
861 fwrite);
862 if (posn > 0) {
863 strbuf_addf(&buf, "Range: bytes=%ld-", posn);
864 headers = curl_slist_append(headers, buf.buf);
865 strbuf_reset(&buf);
867 } else
868 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
869 fwrite_buffer);
872 strbuf_addstr(&buf, "Pragma:");
873 if (options && options->no_cache)
874 strbuf_addstr(&buf, " no-cache");
875 if (options && options->keep_error)
876 curl_easy_setopt(slot->curl, CURLOPT_FAILONERROR, 0);
878 headers = curl_slist_append(headers, buf.buf);
880 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
881 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, headers);
882 curl_easy_setopt(slot->curl, CURLOPT_ENCODING, "gzip");
884 if (start_active_slot(slot)) {
885 run_active_slot(slot);
886 ret = handle_curl_result(&results);
887 } else {
888 snprintf(curl_errorstr, sizeof(curl_errorstr),
889 "failed to start HTTP request");
890 ret = HTTP_START_FAILED;
893 if (options && options->content_type)
894 curlinfo_strbuf(slot->curl, CURLINFO_CONTENT_TYPE,
895 options->content_type);
897 if (options && options->effective_url)
898 curlinfo_strbuf(slot->curl, CURLINFO_EFFECTIVE_URL,
899 options->effective_url);
901 curl_slist_free_all(headers);
902 strbuf_release(&buf);
904 return ret;
908 * Update the "base" url to a more appropriate value, as deduced by
909 * redirects seen when requesting a URL starting with "url".
911 * The "asked" parameter is a URL that we asked curl to access, and must begin
912 * with "base".
914 * The "got" parameter is the URL that curl reported to us as where we ended
915 * up.
917 * Returns 1 if we updated the base url, 0 otherwise.
919 * Our basic strategy is to compare "base" and "asked" to find the bits
920 * specific to our request. We then strip those bits off of "got" to yield the
921 * new base. So for example, if our base is "http://example.com/foo.git",
922 * and we ask for "http://example.com/foo.git/info/refs", we might end up
923 * with "https://other.example.com/foo.git/info/refs". We would want the
924 * new URL to become "https://other.example.com/foo.git".
926 * Note that this assumes a sane redirect scheme. It's entirely possible
927 * in the example above to end up at a URL that does not even end in
928 * "info/refs". In such a case we simply punt, as there is not much we can
929 * do (and such a scheme is unlikely to represent a real git repository,
930 * which means we are likely about to abort anyway).
932 static int update_url_from_redirect(struct strbuf *base,
933 const char *asked,
934 const struct strbuf *got)
936 const char *tail;
937 size_t tail_len;
939 if (!strcmp(asked, got->buf))
940 return 0;
942 if (prefixcmp(asked, base->buf))
943 die("BUG: update_url_from_redirect: %s is not a superset of %s",
944 asked, base->buf);
946 tail = asked + base->len;
947 tail_len = strlen(tail);
949 if (got->len < tail_len ||
950 strcmp(tail, got->buf + got->len - tail_len))
951 return 0; /* insane redirect scheme */
953 strbuf_reset(base);
954 strbuf_add(base, got->buf, got->len - tail_len);
955 return 1;
958 static int http_request_reauth(const char *url,
959 void *result, int target,
960 struct http_get_options *options)
962 int ret = http_request(url, result, target, options);
964 if (options && options->effective_url && options->base_url) {
965 if (update_url_from_redirect(options->base_url,
966 url, options->effective_url)) {
967 credential_from_url(&http_auth, options->base_url->buf);
968 url = options->effective_url->buf;
972 if (ret != HTTP_REAUTH)
973 return ret;
976 * If we are using KEEP_ERROR, the previous request may have
977 * put cruft into our output stream; we should clear it out before
978 * making our next request. We only know how to do this for
979 * the strbuf case, but that is enough to satisfy current callers.
981 if (options && options->keep_error) {
982 switch (target) {
983 case HTTP_REQUEST_STRBUF:
984 strbuf_reset(result);
985 break;
986 default:
987 die("BUG: HTTP_KEEP_ERROR is only supported with strbufs");
991 credential_fill(&http_auth);
993 return http_request(url, result, target, options);
996 int http_get_strbuf(const char *url,
997 struct strbuf *result,
998 struct http_get_options *options)
1000 return http_request_reauth(url, result, HTTP_REQUEST_STRBUF, options);
1004 * Downloads a URL and stores the result in the given file.
1006 * If a previous interrupted download is detected (i.e. a previous temporary
1007 * file is still around) the download is resumed.
1009 static int http_get_file(const char *url, const char *filename,
1010 struct http_get_options *options)
1012 int ret;
1013 struct strbuf tmpfile = STRBUF_INIT;
1014 FILE *result;
1016 strbuf_addf(&tmpfile, "%s.temp", filename);
1017 result = fopen(tmpfile.buf, "a");
1018 if (!result) {
1019 error("Unable to open local file %s", tmpfile.buf);
1020 ret = HTTP_ERROR;
1021 goto cleanup;
1024 ret = http_request_reauth(url, result, HTTP_REQUEST_FILE, options);
1025 fclose(result);
1027 if (ret == HTTP_OK && move_temp_to_file(tmpfile.buf, filename))
1028 ret = HTTP_ERROR;
1029 cleanup:
1030 strbuf_release(&tmpfile);
1031 return ret;
1034 int http_fetch_ref(const char *base, struct ref *ref)
1036 struct http_get_options options = {0};
1037 char *url;
1038 struct strbuf buffer = STRBUF_INIT;
1039 int ret = -1;
1041 options.no_cache = 1;
1043 url = quote_ref_url(base, ref->name);
1044 if (http_get_strbuf(url, &buffer, &options) == HTTP_OK) {
1045 strbuf_rtrim(&buffer);
1046 if (buffer.len == 40)
1047 ret = get_sha1_hex(buffer.buf, ref->old_sha1);
1048 else if (!prefixcmp(buffer.buf, "ref: ")) {
1049 ref->symref = xstrdup(buffer.buf + 5);
1050 ret = 0;
1054 strbuf_release(&buffer);
1055 free(url);
1056 return ret;
1059 /* Helpers for fetching packs */
1060 static char *fetch_pack_index(unsigned char *sha1, const char *base_url)
1062 char *url, *tmp;
1063 struct strbuf buf = STRBUF_INIT;
1065 if (http_is_verbose)
1066 fprintf(stderr, "Getting index for pack %s\n", sha1_to_hex(sha1));
1068 end_url_with_slash(&buf, base_url);
1069 strbuf_addf(&buf, "objects/pack/pack-%s.idx", sha1_to_hex(sha1));
1070 url = strbuf_detach(&buf, NULL);
1072 strbuf_addf(&buf, "%s.temp", sha1_pack_index_name(sha1));
1073 tmp = strbuf_detach(&buf, NULL);
1075 if (http_get_file(url, tmp, NULL) != HTTP_OK) {
1076 error("Unable to get pack index %s", url);
1077 free(tmp);
1078 tmp = NULL;
1081 free(url);
1082 return tmp;
1085 static int fetch_and_setup_pack_index(struct packed_git **packs_head,
1086 unsigned char *sha1, const char *base_url)
1088 struct packed_git *new_pack;
1089 char *tmp_idx = NULL;
1090 int ret;
1092 if (has_pack_index(sha1)) {
1093 new_pack = parse_pack_index(sha1, NULL);
1094 if (!new_pack)
1095 return -1; /* parse_pack_index() already issued error message */
1096 goto add_pack;
1099 tmp_idx = fetch_pack_index(sha1, base_url);
1100 if (!tmp_idx)
1101 return -1;
1103 new_pack = parse_pack_index(sha1, tmp_idx);
1104 if (!new_pack) {
1105 unlink(tmp_idx);
1106 free(tmp_idx);
1108 return -1; /* parse_pack_index() already issued error message */
1111 ret = verify_pack_index(new_pack);
1112 if (!ret) {
1113 close_pack_index(new_pack);
1114 ret = move_temp_to_file(tmp_idx, sha1_pack_index_name(sha1));
1116 free(tmp_idx);
1117 if (ret)
1118 return -1;
1120 add_pack:
1121 new_pack->next = *packs_head;
1122 *packs_head = new_pack;
1123 return 0;
1126 int http_get_info_packs(const char *base_url, struct packed_git **packs_head)
1128 struct http_get_options options = {0};
1129 int ret = 0, i = 0;
1130 char *url, *data;
1131 struct strbuf buf = STRBUF_INIT;
1132 unsigned char sha1[20];
1134 end_url_with_slash(&buf, base_url);
1135 strbuf_addstr(&buf, "objects/info/packs");
1136 url = strbuf_detach(&buf, NULL);
1138 options.no_cache = 1;
1139 ret = http_get_strbuf(url, &buf, &options);
1140 if (ret != HTTP_OK)
1141 goto cleanup;
1143 data = buf.buf;
1144 while (i < buf.len) {
1145 switch (data[i]) {
1146 case 'P':
1147 i++;
1148 if (i + 52 <= buf.len &&
1149 !prefixcmp(data + i, " pack-") &&
1150 !prefixcmp(data + i + 46, ".pack\n")) {
1151 get_sha1_hex(data + i + 6, sha1);
1152 fetch_and_setup_pack_index(packs_head, sha1,
1153 base_url);
1154 i += 51;
1155 break;
1157 default:
1158 while (i < buf.len && data[i] != '\n')
1159 i++;
1161 i++;
1164 cleanup:
1165 free(url);
1166 return ret;
1169 void release_http_pack_request(struct http_pack_request *preq)
1171 if (preq->packfile != NULL) {
1172 fclose(preq->packfile);
1173 preq->packfile = NULL;
1175 if (preq->range_header != NULL) {
1176 curl_slist_free_all(preq->range_header);
1177 preq->range_header = NULL;
1179 preq->slot = NULL;
1180 free(preq->url);
1183 int finish_http_pack_request(struct http_pack_request *preq)
1185 struct packed_git **lst;
1186 struct packed_git *p = preq->target;
1187 char *tmp_idx;
1188 struct child_process ip;
1189 const char *ip_argv[8];
1191 close_pack_index(p);
1193 fclose(preq->packfile);
1194 preq->packfile = NULL;
1196 lst = preq->lst;
1197 while (*lst != p)
1198 lst = &((*lst)->next);
1199 *lst = (*lst)->next;
1201 tmp_idx = xstrdup(preq->tmpfile);
1202 strcpy(tmp_idx + strlen(tmp_idx) - strlen(".pack.temp"),
1203 ".idx.temp");
1205 ip_argv[0] = "index-pack";
1206 ip_argv[1] = "-o";
1207 ip_argv[2] = tmp_idx;
1208 ip_argv[3] = preq->tmpfile;
1209 ip_argv[4] = NULL;
1211 memset(&ip, 0, sizeof(ip));
1212 ip.argv = ip_argv;
1213 ip.git_cmd = 1;
1214 ip.no_stdin = 1;
1215 ip.no_stdout = 1;
1217 if (run_command(&ip)) {
1218 unlink(preq->tmpfile);
1219 unlink(tmp_idx);
1220 free(tmp_idx);
1221 return -1;
1224 unlink(sha1_pack_index_name(p->sha1));
1226 if (move_temp_to_file(preq->tmpfile, sha1_pack_name(p->sha1))
1227 || move_temp_to_file(tmp_idx, sha1_pack_index_name(p->sha1))) {
1228 free(tmp_idx);
1229 return -1;
1232 install_packed_git(p);
1233 free(tmp_idx);
1234 return 0;
1237 struct http_pack_request *new_http_pack_request(
1238 struct packed_git *target, const char *base_url)
1240 long prev_posn = 0;
1241 char range[RANGE_HEADER_SIZE];
1242 struct strbuf buf = STRBUF_INIT;
1243 struct http_pack_request *preq;
1245 preq = xcalloc(1, sizeof(*preq));
1246 preq->target = target;
1248 end_url_with_slash(&buf, base_url);
1249 strbuf_addf(&buf, "objects/pack/pack-%s.pack",
1250 sha1_to_hex(target->sha1));
1251 preq->url = strbuf_detach(&buf, NULL);
1253 snprintf(preq->tmpfile, sizeof(preq->tmpfile), "%s.temp",
1254 sha1_pack_name(target->sha1));
1255 preq->packfile = fopen(preq->tmpfile, "a");
1256 if (!preq->packfile) {
1257 error("Unable to open local file %s for pack",
1258 preq->tmpfile);
1259 goto abort;
1262 preq->slot = get_active_slot();
1263 curl_easy_setopt(preq->slot->curl, CURLOPT_FILE, preq->packfile);
1264 curl_easy_setopt(preq->slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
1265 curl_easy_setopt(preq->slot->curl, CURLOPT_URL, preq->url);
1266 curl_easy_setopt(preq->slot->curl, CURLOPT_HTTPHEADER,
1267 no_pragma_header);
1270 * If there is data present from a previous transfer attempt,
1271 * resume where it left off
1273 prev_posn = ftell(preq->packfile);
1274 if (prev_posn>0) {
1275 if (http_is_verbose)
1276 fprintf(stderr,
1277 "Resuming fetch of pack %s at byte %ld\n",
1278 sha1_to_hex(target->sha1), prev_posn);
1279 sprintf(range, "Range: bytes=%ld-", prev_posn);
1280 preq->range_header = curl_slist_append(NULL, range);
1281 curl_easy_setopt(preq->slot->curl, CURLOPT_HTTPHEADER,
1282 preq->range_header);
1285 return preq;
1287 abort:
1288 free(preq->url);
1289 free(preq);
1290 return NULL;
1293 /* Helpers for fetching objects (loose) */
1294 static size_t fwrite_sha1_file(char *ptr, size_t eltsize, size_t nmemb,
1295 void *data)
1297 unsigned char expn[4096];
1298 size_t size = eltsize * nmemb;
1299 int posn = 0;
1300 struct http_object_request *freq =
1301 (struct http_object_request *)data;
1302 do {
1303 ssize_t retval = xwrite(freq->localfile,
1304 (char *) ptr + posn, size - posn);
1305 if (retval < 0)
1306 return posn;
1307 posn += retval;
1308 } while (posn < size);
1310 freq->stream.avail_in = size;
1311 freq->stream.next_in = (void *)ptr;
1312 do {
1313 freq->stream.next_out = expn;
1314 freq->stream.avail_out = sizeof(expn);
1315 freq->zret = git_inflate(&freq->stream, Z_SYNC_FLUSH);
1316 git_SHA1_Update(&freq->c, expn,
1317 sizeof(expn) - freq->stream.avail_out);
1318 } while (freq->stream.avail_in && freq->zret == Z_OK);
1319 return size;
1322 struct http_object_request *new_http_object_request(const char *base_url,
1323 unsigned char *sha1)
1325 char *hex = sha1_to_hex(sha1);
1326 char *filename;
1327 char prevfile[PATH_MAX];
1328 int prevlocal;
1329 char prev_buf[PREV_BUF_SIZE];
1330 ssize_t prev_read = 0;
1331 long prev_posn = 0;
1332 char range[RANGE_HEADER_SIZE];
1333 struct curl_slist *range_header = NULL;
1334 struct http_object_request *freq;
1336 freq = xcalloc(1, sizeof(*freq));
1337 hashcpy(freq->sha1, sha1);
1338 freq->localfile = -1;
1340 filename = sha1_file_name(sha1);
1341 snprintf(freq->tmpfile, sizeof(freq->tmpfile),
1342 "%s.temp", filename);
1344 snprintf(prevfile, sizeof(prevfile), "%s.prev", filename);
1345 unlink_or_warn(prevfile);
1346 rename(freq->tmpfile, prevfile);
1347 unlink_or_warn(freq->tmpfile);
1349 if (freq->localfile != -1)
1350 error("fd leakage in start: %d", freq->localfile);
1351 freq->localfile = open(freq->tmpfile,
1352 O_WRONLY | O_CREAT | O_EXCL, 0666);
1354 * This could have failed due to the "lazy directory creation";
1355 * try to mkdir the last path component.
1357 if (freq->localfile < 0 && errno == ENOENT) {
1358 char *dir = strrchr(freq->tmpfile, '/');
1359 if (dir) {
1360 *dir = 0;
1361 mkdir(freq->tmpfile, 0777);
1362 *dir = '/';
1364 freq->localfile = open(freq->tmpfile,
1365 O_WRONLY | O_CREAT | O_EXCL, 0666);
1368 if (freq->localfile < 0) {
1369 error("Couldn't create temporary file %s: %s",
1370 freq->tmpfile, strerror(errno));
1371 goto abort;
1374 git_inflate_init(&freq->stream);
1376 git_SHA1_Init(&freq->c);
1378 freq->url = get_remote_object_url(base_url, hex, 0);
1381 * If a previous temp file is present, process what was already
1382 * fetched.
1384 prevlocal = open(prevfile, O_RDONLY);
1385 if (prevlocal != -1) {
1386 do {
1387 prev_read = xread(prevlocal, prev_buf, PREV_BUF_SIZE);
1388 if (prev_read>0) {
1389 if (fwrite_sha1_file(prev_buf,
1391 prev_read,
1392 freq) == prev_read) {
1393 prev_posn += prev_read;
1394 } else {
1395 prev_read = -1;
1398 } while (prev_read > 0);
1399 close(prevlocal);
1401 unlink_or_warn(prevfile);
1404 * Reset inflate/SHA1 if there was an error reading the previous temp
1405 * file; also rewind to the beginning of the local file.
1407 if (prev_read == -1) {
1408 memset(&freq->stream, 0, sizeof(freq->stream));
1409 git_inflate_init(&freq->stream);
1410 git_SHA1_Init(&freq->c);
1411 if (prev_posn>0) {
1412 prev_posn = 0;
1413 lseek(freq->localfile, 0, SEEK_SET);
1414 if (ftruncate(freq->localfile, 0) < 0) {
1415 error("Couldn't truncate temporary file %s: %s",
1416 freq->tmpfile, strerror(errno));
1417 goto abort;
1422 freq->slot = get_active_slot();
1424 curl_easy_setopt(freq->slot->curl, CURLOPT_FILE, freq);
1425 curl_easy_setopt(freq->slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file);
1426 curl_easy_setopt(freq->slot->curl, CURLOPT_ERRORBUFFER, freq->errorstr);
1427 curl_easy_setopt(freq->slot->curl, CURLOPT_URL, freq->url);
1428 curl_easy_setopt(freq->slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
1431 * If we have successfully processed data from a previous fetch
1432 * attempt, only fetch the data we don't already have.
1434 if (prev_posn>0) {
1435 if (http_is_verbose)
1436 fprintf(stderr,
1437 "Resuming fetch of object %s at byte %ld\n",
1438 hex, prev_posn);
1439 sprintf(range, "Range: bytes=%ld-", prev_posn);
1440 range_header = curl_slist_append(range_header, range);
1441 curl_easy_setopt(freq->slot->curl,
1442 CURLOPT_HTTPHEADER, range_header);
1445 return freq;
1447 abort:
1448 free(freq->url);
1449 free(freq);
1450 return NULL;
1453 void process_http_object_request(struct http_object_request *freq)
1455 if (freq->slot == NULL)
1456 return;
1457 freq->curl_result = freq->slot->curl_result;
1458 freq->http_code = freq->slot->http_code;
1459 freq->slot = NULL;
1462 int finish_http_object_request(struct http_object_request *freq)
1464 struct stat st;
1466 close(freq->localfile);
1467 freq->localfile = -1;
1469 process_http_object_request(freq);
1471 if (freq->http_code == 416) {
1472 warning("requested range invalid; we may already have all the data.");
1473 } else if (freq->curl_result != CURLE_OK) {
1474 if (stat(freq->tmpfile, &st) == 0)
1475 if (st.st_size == 0)
1476 unlink_or_warn(freq->tmpfile);
1477 return -1;
1480 git_inflate_end(&freq->stream);
1481 git_SHA1_Final(freq->real_sha1, &freq->c);
1482 if (freq->zret != Z_STREAM_END) {
1483 unlink_or_warn(freq->tmpfile);
1484 return -1;
1486 if (hashcmp(freq->sha1, freq->real_sha1)) {
1487 unlink_or_warn(freq->tmpfile);
1488 return -1;
1490 freq->rename =
1491 move_temp_to_file(freq->tmpfile, sha1_file_name(freq->sha1));
1493 return freq->rename;
1496 void abort_http_object_request(struct http_object_request *freq)
1498 unlink_or_warn(freq->tmpfile);
1500 release_http_object_request(freq);
1503 void release_http_object_request(struct http_object_request *freq)
1505 if (freq->localfile != -1) {
1506 close(freq->localfile);
1507 freq->localfile = -1;
1509 if (freq->url != NULL) {
1510 free(freq->url);
1511 freq->url = NULL;
1513 if (freq->slot != NULL) {
1514 freq->slot->callback_func = NULL;
1515 freq->slot->callback_data = NULL;
1516 release_active_slot(freq->slot);
1517 freq->slot = NULL;