url: decode buffers that are not NUL-terminated
[git] / http.c
1 #include "http.h"
2 #include "pack.h"
3 #include "sideband.h"
4 #include "run-command.h"
5 #include "url.h"
6
7 int data_received;
8 int active_requests;
9 int http_is_verbose;
10 size_t http_post_buffer = 16 * LARGE_PACKET_MAX;
11
12 #if LIBCURL_VERSION_NUM >= 0x070a06
13 #define LIBCURL_CAN_HANDLE_AUTH_ANY
14 #endif
15
16 static int min_curl_sessions = 1;
17 static int curl_session_count;
18 #ifdef USE_CURL_MULTI
19 static int max_requests = -1;
20 static CURLM *curlm;
21 #endif
22 #ifndef NO_CURL_EASY_DUPHANDLE
23 static CURL *curl_default;
24 #endif
25
26 #define PREV_BUF_SIZE 4096
27 #define RANGE_HEADER_SIZE 30
28
29 char curl_errorstr[CURL_ERROR_SIZE];
30
31 static int curl_ssl_verify = -1;
32 static const char *ssl_cert;
33 #if LIBCURL_VERSION_NUM >= 0x070903
34 static const char *ssl_key;
35 #endif
36 #if LIBCURL_VERSION_NUM >= 0x070908
37 static const char *ssl_capath;
38 #endif
39 static const char *ssl_cainfo;
40 static long curl_low_speed_limit = -1;
41 static long curl_low_speed_time = -1;
42 static int curl_ftp_no_epsv;
43 static const char *curl_http_proxy;
44 static const char *curl_cookie_file;
45 static char *user_name, *user_pass;
46 static const char *user_agent;
47
48 #if LIBCURL_VERSION_NUM >= 0x071700
49 /* Use CURLOPT_KEYPASSWD as is */
50 #elif LIBCURL_VERSION_NUM >= 0x070903
51 #define CURLOPT_KEYPASSWD CURLOPT_SSLKEYPASSWD
52 #else
53 #define CURLOPT_KEYPASSWD CURLOPT_SSLCERTPASSWD
54 #endif
55
56 static char *ssl_cert_password;
57 static int ssl_cert_password_required;
58
59 static struct curl_slist *pragma_header;
60 static struct curl_slist *no_pragma_header;
61
62 static struct active_request_slot *active_queue_head;
63
64 size_t fread_buffer(char *ptr, size_t eltsize, size_t nmemb, void *buffer_)
65 {
66         size_t size = eltsize * nmemb;
67         struct buffer *buffer = buffer_;
68
69         if (size > buffer->buf.len - buffer->posn)
70                 size = buffer->buf.len - buffer->posn;
71         memcpy(ptr, buffer->buf.buf + buffer->posn, size);
72         buffer->posn += size;
73
74         return size;
75 }
76
77 #ifndef NO_CURL_IOCTL
78 curlioerr ioctl_buffer(CURL *handle, int cmd, void *clientp)
79 {
80         struct buffer *buffer = clientp;
81
82         switch (cmd) {
83         case CURLIOCMD_NOP:
84                 return CURLIOE_OK;
85
86         case CURLIOCMD_RESTARTREAD:
87                 buffer->posn = 0;
88                 return CURLIOE_OK;
89
90         default:
91                 return CURLIOE_UNKNOWNCMD;
92         }
93 }
94 #endif
95
96 size_t fwrite_buffer(char *ptr, size_t eltsize, size_t nmemb, void *buffer_)
97 {
98         size_t size = eltsize * nmemb;
99         struct strbuf *buffer = buffer_;
100
101         strbuf_add(buffer, ptr, size);
102         data_received++;
103         return size;
104 }
105
106 size_t fwrite_null(char *ptr, size_t eltsize, size_t nmemb, void *strbuf)
107 {
108         data_received++;
109         return eltsize * nmemb;
110 }
111
112 #ifdef USE_CURL_MULTI
113 static void process_curl_messages(void)
114 {
115         int num_messages;
116         struct active_request_slot *slot;
117         CURLMsg *curl_message = curl_multi_info_read(curlm, &num_messages);
118
119         while (curl_message != NULL) {
120                 if (curl_message->msg == CURLMSG_DONE) {
121                         int curl_result = curl_message->data.result;
122                         slot = active_queue_head;
123                         while (slot != NULL &&
124                                slot->curl != curl_message->easy_handle)
125                                 slot = slot->next;
126                         if (slot != NULL) {
127                                 curl_multi_remove_handle(curlm, slot->curl);
128                                 slot->curl_result = curl_result;
129                                 finish_active_slot(slot);
130                         } else {
131                                 fprintf(stderr, "Received DONE message for unknown request!\n");
132                         }
133                 } else {
134                         fprintf(stderr, "Unknown CURL message received: %d\n",
135                                 (int)curl_message->msg);
136                 }
137                 curl_message = curl_multi_info_read(curlm, &num_messages);
138         }
139 }
140 #endif
141
142 static int http_options(const char *var, const char *value, void *cb)
143 {
144         if (!strcmp("http.sslverify", var)) {
145                 curl_ssl_verify = git_config_bool(var, value);
146                 return 0;
147         }
148         if (!strcmp("http.sslcert", var))
149                 return git_config_string(&ssl_cert, var, value);
150 #if LIBCURL_VERSION_NUM >= 0x070903
151         if (!strcmp("http.sslkey", var))
152                 return git_config_string(&ssl_key, var, value);
153 #endif
154 #if LIBCURL_VERSION_NUM >= 0x070908
155         if (!strcmp("http.sslcapath", var))
156                 return git_config_string(&ssl_capath, var, value);
157 #endif
158         if (!strcmp("http.sslcainfo", var))
159                 return git_config_string(&ssl_cainfo, var, value);
160         if (!strcmp("http.sslcertpasswordprotected", var)) {
161                 if (git_config_bool(var, value))
162                         ssl_cert_password_required = 1;
163                 return 0;
164         }
165         if (!strcmp("http.minsessions", var)) {
166                 min_curl_sessions = git_config_int(var, value);
167 #ifndef USE_CURL_MULTI
168                 if (min_curl_sessions > 1)
169                         min_curl_sessions = 1;
170 #endif
171                 return 0;
172         }
173 #ifdef USE_CURL_MULTI
174         if (!strcmp("http.maxrequests", var)) {
175                 max_requests = git_config_int(var, value);
176                 return 0;
177         }
178 #endif
179         if (!strcmp("http.lowspeedlimit", var)) {
180                 curl_low_speed_limit = (long)git_config_int(var, value);
181                 return 0;
182         }
183         if (!strcmp("http.lowspeedtime", var)) {
184                 curl_low_speed_time = (long)git_config_int(var, value);
185                 return 0;
186         }
187
188         if (!strcmp("http.noepsv", var)) {
189                 curl_ftp_no_epsv = git_config_bool(var, value);
190                 return 0;
191         }
192         if (!strcmp("http.proxy", var))
193                 return git_config_string(&curl_http_proxy, var, value);
194
195         if (!strcmp("http.cookiefile", var))
196                 return git_config_string(&curl_cookie_file, var, value);
197
198         if (!strcmp("http.postbuffer", var)) {
199                 http_post_buffer = git_config_int(var, value);
200                 if (http_post_buffer < LARGE_PACKET_MAX)
201                         http_post_buffer = LARGE_PACKET_MAX;
202                 return 0;
203         }
204
205         if (!strcmp("http.useragent", var))
206                 return git_config_string(&user_agent, var, value);
207
208         /* Fall back on the default ones */
209         return git_default_config(var, value, cb);
210 }
211
212 static void init_curl_http_auth(CURL *result)
213 {
214         if (user_name) {
215                 struct strbuf up = STRBUF_INIT;
216                 if (!user_pass)
217                         user_pass = xstrdup(git_getpass("Password: "));
218                 strbuf_addf(&up, "%s:%s", user_name, user_pass);
219                 curl_easy_setopt(result, CURLOPT_USERPWD,
220                                  strbuf_detach(&up, NULL));
221         }
222 }
223
224 static int has_cert_password(void)
225 {
226         if (ssl_cert_password != NULL)
227                 return 1;
228         if (ssl_cert == NULL || ssl_cert_password_required != 1)
229                 return 0;
230         /* Only prompt the user once. */
231         ssl_cert_password_required = -1;
232         ssl_cert_password = git_getpass("Certificate Password: ");
233         if (ssl_cert_password != NULL) {
234                 ssl_cert_password = xstrdup(ssl_cert_password);
235                 return 1;
236         } else
237                 return 0;
238 }
239
240 static CURL *get_curl_handle(void)
241 {
242         CURL *result = curl_easy_init();
243
244         if (!curl_ssl_verify) {
245                 curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 0);
246                 curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 0);
247         } else {
248                 /* Verify authenticity of the peer's certificate */
249                 curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, 1);
250                 /* The name in the cert must match whom we tried to connect */
251                 curl_easy_setopt(result, CURLOPT_SSL_VERIFYHOST, 2);
252         }
253
254 #if LIBCURL_VERSION_NUM >= 0x070907
255         curl_easy_setopt(result, CURLOPT_NETRC, CURL_NETRC_OPTIONAL);
256 #endif
257 #ifdef LIBCURL_CAN_HANDLE_AUTH_ANY
258         curl_easy_setopt(result, CURLOPT_HTTPAUTH, CURLAUTH_ANY);
259 #endif
260
261         init_curl_http_auth(result);
262
263         if (ssl_cert != NULL)
264                 curl_easy_setopt(result, CURLOPT_SSLCERT, ssl_cert);
265         if (has_cert_password())
266                 curl_easy_setopt(result, CURLOPT_KEYPASSWD, ssl_cert_password);
267 #if LIBCURL_VERSION_NUM >= 0x070903
268         if (ssl_key != NULL)
269                 curl_easy_setopt(result, CURLOPT_SSLKEY, ssl_key);
270 #endif
271 #if LIBCURL_VERSION_NUM >= 0x070908
272         if (ssl_capath != NULL)
273                 curl_easy_setopt(result, CURLOPT_CAPATH, ssl_capath);
274 #endif
275         if (ssl_cainfo != NULL)
276                 curl_easy_setopt(result, CURLOPT_CAINFO, ssl_cainfo);
277         curl_easy_setopt(result, CURLOPT_FAILONERROR, 1);
278
279         if (curl_low_speed_limit > 0 && curl_low_speed_time > 0) {
280                 curl_easy_setopt(result, CURLOPT_LOW_SPEED_LIMIT,
281                                  curl_low_speed_limit);
282                 curl_easy_setopt(result, CURLOPT_LOW_SPEED_TIME,
283                                  curl_low_speed_time);
284         }
285
286         curl_easy_setopt(result, CURLOPT_FOLLOWLOCATION, 1);
287 #if LIBCURL_VERSION_NUM >= 0x071301
288         curl_easy_setopt(result, CURLOPT_POSTREDIR, CURL_REDIR_POST_ALL);
289 #elif LIBCURL_VERSION_NUM >= 0x071101
290         curl_easy_setopt(result, CURLOPT_POST301, 1);
291 #endif
292
293         if (getenv("GIT_CURL_VERBOSE"))
294                 curl_easy_setopt(result, CURLOPT_VERBOSE, 1);
295
296         curl_easy_setopt(result, CURLOPT_USERAGENT,
297                 user_agent ? user_agent : GIT_HTTP_USER_AGENT);
298
299         if (curl_ftp_no_epsv)
300                 curl_easy_setopt(result, CURLOPT_FTP_USE_EPSV, 0);
301
302         if (curl_http_proxy)
303                 curl_easy_setopt(result, CURLOPT_PROXY, curl_http_proxy);
304
305         return result;
306 }
307
308 static void http_auth_init(const char *url)
309 {
310         char *at, *colon, *cp, *slash;
311
312         cp = strstr(url, "://");
313         if (!cp)
314                 return;
315
316         /*
317          * Ok, the URL looks like "proto://something".  Which one?
318          * "proto://<user>:<pass>@<host>/...",
319          * "proto://<user>@<host>/...", or just
320          * "proto://<host>/..."?
321          */
322         cp += 3;
323         at = strchr(cp, '@');
324         colon = strchr(cp, ':');
325         slash = strchrnul(cp, '/');
326         if (!at || slash <= at)
327                 return; /* No credentials */
328         if (!colon || at <= colon) {
329                 /* Only username */
330                 user_name = url_decode_mem(cp, at - cp);
331                 user_pass = NULL;
332         } else {
333                 user_name = url_decode_mem(cp, colon - cp);
334                 user_pass = url_decode_mem(colon + 1, at - (colon + 1));
335         }
336 }
337
338 static void set_from_env(const char **var, const char *envname)
339 {
340         const char *val = getenv(envname);
341         if (val)
342                 *var = val;
343 }
344
345 void http_init(struct remote *remote)
346 {
347         char *low_speed_limit;
348         char *low_speed_time;
349
350         http_is_verbose = 0;
351
352         git_config(http_options, NULL);
353
354         curl_global_init(CURL_GLOBAL_ALL);
355
356         if (remote && remote->http_proxy)
357                 curl_http_proxy = xstrdup(remote->http_proxy);
358
359         pragma_header = curl_slist_append(pragma_header, "Pragma: no-cache");
360         no_pragma_header = curl_slist_append(no_pragma_header, "Pragma:");
361
362 #ifdef USE_CURL_MULTI
363         {
364                 char *http_max_requests = getenv("GIT_HTTP_MAX_REQUESTS");
365                 if (http_max_requests != NULL)
366                         max_requests = atoi(http_max_requests);
367         }
368
369         curlm = curl_multi_init();
370         if (curlm == NULL) {
371                 fprintf(stderr, "Error creating curl multi handle.\n");
372                 exit(1);
373         }
374 #endif
375
376         if (getenv("GIT_SSL_NO_VERIFY"))
377                 curl_ssl_verify = 0;
378
379         set_from_env(&ssl_cert, "GIT_SSL_CERT");
380 #if LIBCURL_VERSION_NUM >= 0x070903
381         set_from_env(&ssl_key, "GIT_SSL_KEY");
382 #endif
383 #if LIBCURL_VERSION_NUM >= 0x070908
384         set_from_env(&ssl_capath, "GIT_SSL_CAPATH");
385 #endif
386         set_from_env(&ssl_cainfo, "GIT_SSL_CAINFO");
387
388         set_from_env(&user_agent, "GIT_HTTP_USER_AGENT");
389
390         low_speed_limit = getenv("GIT_HTTP_LOW_SPEED_LIMIT");
391         if (low_speed_limit != NULL)
392                 curl_low_speed_limit = strtol(low_speed_limit, NULL, 10);
393         low_speed_time = getenv("GIT_HTTP_LOW_SPEED_TIME");
394         if (low_speed_time != NULL)
395                 curl_low_speed_time = strtol(low_speed_time, NULL, 10);
396
397         if (curl_ssl_verify == -1)
398                 curl_ssl_verify = 1;
399
400         curl_session_count = 0;
401 #ifdef USE_CURL_MULTI
402         if (max_requests < 1)
403                 max_requests = DEFAULT_MAX_REQUESTS;
404 #endif
405
406         if (getenv("GIT_CURL_FTP_NO_EPSV"))
407                 curl_ftp_no_epsv = 1;
408
409         if (remote && remote->url && remote->url[0]) {
410                 http_auth_init(remote->url[0]);
411                 if (!ssl_cert_password_required &&
412                     getenv("GIT_SSL_CERT_PASSWORD_PROTECTED") &&
413                     !prefixcmp(remote->url[0], "https://"))
414                         ssl_cert_password_required = 1;
415         }
416
417 #ifndef NO_CURL_EASY_DUPHANDLE
418         curl_default = get_curl_handle();
419 #endif
420 }
421
422 void http_cleanup(void)
423 {
424         struct active_request_slot *slot = active_queue_head;
425
426         while (slot != NULL) {
427                 struct active_request_slot *next = slot->next;
428                 if (slot->curl != NULL) {
429 #ifdef USE_CURL_MULTI
430                         curl_multi_remove_handle(curlm, slot->curl);
431 #endif
432                         curl_easy_cleanup(slot->curl);
433                 }
434                 free(slot);
435                 slot = next;
436         }
437         active_queue_head = NULL;
438
439 #ifndef NO_CURL_EASY_DUPHANDLE
440         curl_easy_cleanup(curl_default);
441 #endif
442
443 #ifdef USE_CURL_MULTI
444         curl_multi_cleanup(curlm);
445 #endif
446         curl_global_cleanup();
447
448         curl_slist_free_all(pragma_header);
449         pragma_header = NULL;
450
451         curl_slist_free_all(no_pragma_header);
452         no_pragma_header = NULL;
453
454         if (curl_http_proxy) {
455                 free((void *)curl_http_proxy);
456                 curl_http_proxy = NULL;
457         }
458
459         if (ssl_cert_password != NULL) {
460                 memset(ssl_cert_password, 0, strlen(ssl_cert_password));
461                 free(ssl_cert_password);
462                 ssl_cert_password = NULL;
463         }
464         ssl_cert_password_required = 0;
465 }
466
467 struct active_request_slot *get_active_slot(void)
468 {
469         struct active_request_slot *slot = active_queue_head;
470         struct active_request_slot *newslot;
471
472 #ifdef USE_CURL_MULTI
473         int num_transfers;
474
475         /* Wait for a slot to open up if the queue is full */
476         while (active_requests >= max_requests) {
477                 curl_multi_perform(curlm, &num_transfers);
478                 if (num_transfers < active_requests)
479                         process_curl_messages();
480         }
481 #endif
482
483         while (slot != NULL && slot->in_use)
484                 slot = slot->next;
485
486         if (slot == NULL) {
487                 newslot = xmalloc(sizeof(*newslot));
488                 newslot->curl = NULL;
489                 newslot->in_use = 0;
490                 newslot->next = NULL;
491
492                 slot = active_queue_head;
493                 if (slot == NULL) {
494                         active_queue_head = newslot;
495                 } else {
496                         while (slot->next != NULL)
497                                 slot = slot->next;
498                         slot->next = newslot;
499                 }
500                 slot = newslot;
501         }
502
503         if (slot->curl == NULL) {
504 #ifdef NO_CURL_EASY_DUPHANDLE
505                 slot->curl = get_curl_handle();
506 #else
507                 slot->curl = curl_easy_duphandle(curl_default);
508 #endif
509                 curl_session_count++;
510         }
511
512         active_requests++;
513         slot->in_use = 1;
514         slot->local = NULL;
515         slot->results = NULL;
516         slot->finished = NULL;
517         slot->callback_data = NULL;
518         slot->callback_func = NULL;
519         curl_easy_setopt(slot->curl, CURLOPT_COOKIEFILE, curl_cookie_file);
520         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, pragma_header);
521         curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, curl_errorstr);
522         curl_easy_setopt(slot->curl, CURLOPT_CUSTOMREQUEST, NULL);
523         curl_easy_setopt(slot->curl, CURLOPT_READFUNCTION, NULL);
524         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, NULL);
525         curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, NULL);
526         curl_easy_setopt(slot->curl, CURLOPT_UPLOAD, 0);
527         curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1);
528
529         return slot;
530 }
531
532 int start_active_slot(struct active_request_slot *slot)
533 {
534 #ifdef USE_CURL_MULTI
535         CURLMcode curlm_result = curl_multi_add_handle(curlm, slot->curl);
536         int num_transfers;
537
538         if (curlm_result != CURLM_OK &&
539             curlm_result != CURLM_CALL_MULTI_PERFORM) {
540                 active_requests--;
541                 slot->in_use = 0;
542                 return 0;
543         }
544
545         /*
546          * We know there must be something to do, since we just added
547          * something.
548          */
549         curl_multi_perform(curlm, &num_transfers);
550 #endif
551         return 1;
552 }
553
554 #ifdef USE_CURL_MULTI
555 struct fill_chain {
556         void *data;
557         int (*fill)(void *);
558         struct fill_chain *next;
559 };
560
561 static struct fill_chain *fill_cfg;
562
563 void add_fill_function(void *data, int (*fill)(void *))
564 {
565         struct fill_chain *new = xmalloc(sizeof(*new));
566         struct fill_chain **linkp = &fill_cfg;
567         new->data = data;
568         new->fill = fill;
569         new->next = NULL;
570         while (*linkp)
571                 linkp = &(*linkp)->next;
572         *linkp = new;
573 }
574
575 void fill_active_slots(void)
576 {
577         struct active_request_slot *slot = active_queue_head;
578
579         while (active_requests < max_requests) {
580                 struct fill_chain *fill;
581                 for (fill = fill_cfg; fill; fill = fill->next)
582                         if (fill->fill(fill->data))
583                                 break;
584
585                 if (!fill)
586                         break;
587         }
588
589         while (slot != NULL) {
590                 if (!slot->in_use && slot->curl != NULL
591                         && curl_session_count > min_curl_sessions) {
592                         curl_easy_cleanup(slot->curl);
593                         slot->curl = NULL;
594                         curl_session_count--;
595                 }
596                 slot = slot->next;
597         }
598 }
599
600 void step_active_slots(void)
601 {
602         int num_transfers;
603         CURLMcode curlm_result;
604
605         do {
606                 curlm_result = curl_multi_perform(curlm, &num_transfers);
607         } while (curlm_result == CURLM_CALL_MULTI_PERFORM);
608         if (num_transfers < active_requests) {
609                 process_curl_messages();
610                 fill_active_slots();
611         }
612 }
613 #endif
614
615 void run_active_slot(struct active_request_slot *slot)
616 {
617 #ifdef USE_CURL_MULTI
618         long last_pos = 0;
619         long current_pos;
620         fd_set readfds;
621         fd_set writefds;
622         fd_set excfds;
623         int max_fd;
624         struct timeval select_timeout;
625         int finished = 0;
626
627         slot->finished = &finished;
628         while (!finished) {
629                 data_received = 0;
630                 step_active_slots();
631
632                 if (!data_received && slot->local != NULL) {
633                         current_pos = ftell(slot->local);
634                         if (current_pos > last_pos)
635                                 data_received++;
636                         last_pos = current_pos;
637                 }
638
639                 if (slot->in_use && !data_received) {
640                         max_fd = 0;
641                         FD_ZERO(&readfds);
642                         FD_ZERO(&writefds);
643                         FD_ZERO(&excfds);
644                         select_timeout.tv_sec = 0;
645                         select_timeout.tv_usec = 50000;
646                         select(max_fd, &readfds, &writefds,
647                                &excfds, &select_timeout);
648                 }
649         }
650 #else
651         while (slot->in_use) {
652                 slot->curl_result = curl_easy_perform(slot->curl);
653                 finish_active_slot(slot);
654         }
655 #endif
656 }
657
658 static void closedown_active_slot(struct active_request_slot *slot)
659 {
660         active_requests--;
661         slot->in_use = 0;
662 }
663
664 static void release_active_slot(struct active_request_slot *slot)
665 {
666         closedown_active_slot(slot);
667         if (slot->curl && curl_session_count > min_curl_sessions) {
668 #ifdef USE_CURL_MULTI
669                 curl_multi_remove_handle(curlm, slot->curl);
670 #endif
671                 curl_easy_cleanup(slot->curl);
672                 slot->curl = NULL;
673                 curl_session_count--;
674         }
675 #ifdef USE_CURL_MULTI
676         fill_active_slots();
677 #endif
678 }
679
680 void finish_active_slot(struct active_request_slot *slot)
681 {
682         closedown_active_slot(slot);
683         curl_easy_getinfo(slot->curl, CURLINFO_HTTP_CODE, &slot->http_code);
684
685         if (slot->finished != NULL)
686                 (*slot->finished) = 1;
687
688         /* Store slot results so they can be read after the slot is reused */
689         if (slot->results != NULL) {
690                 slot->results->curl_result = slot->curl_result;
691                 slot->results->http_code = slot->http_code;
692         }
693
694         /* Run callback if appropriate */
695         if (slot->callback_func != NULL)
696                 slot->callback_func(slot->callback_data);
697 }
698
699 void finish_all_active_slots(void)
700 {
701         struct active_request_slot *slot = active_queue_head;
702
703         while (slot != NULL)
704                 if (slot->in_use) {
705                         run_active_slot(slot);
706                         slot = active_queue_head;
707                 } else {
708                         slot = slot->next;
709                 }
710 }
711
712 /* Helpers for modifying and creating URLs */
713 static inline int needs_quote(int ch)
714 {
715         if (((ch >= 'A') && (ch <= 'Z'))
716                         || ((ch >= 'a') && (ch <= 'z'))
717                         || ((ch >= '0') && (ch <= '9'))
718                         || (ch == '/')
719                         || (ch == '-')
720                         || (ch == '.'))
721                 return 0;
722         return 1;
723 }
724
725 static inline int hex(int v)
726 {
727         if (v < 10)
728                 return '0' + v;
729         else
730                 return 'A' + v - 10;
731 }
732
733 static char *quote_ref_url(const char *base, const char *ref)
734 {
735         struct strbuf buf = STRBUF_INIT;
736         const char *cp;
737         int ch;
738
739         end_url_with_slash(&buf, base);
740
741         for (cp = ref; (ch = *cp) != 0; cp++)
742                 if (needs_quote(ch))
743                         strbuf_addf(&buf, "%%%02x", ch);
744                 else
745                         strbuf_addch(&buf, *cp);
746
747         return strbuf_detach(&buf, NULL);
748 }
749
750 void append_remote_object_url(struct strbuf *buf, const char *url,
751                               const char *hex,
752                               int only_two_digit_prefix)
753 {
754         end_url_with_slash(buf, url);
755
756         strbuf_addf(buf, "objects/%.*s/", 2, hex);
757         if (!only_two_digit_prefix)
758                 strbuf_addf(buf, "%s", hex+2);
759 }
760
761 char *get_remote_object_url(const char *url, const char *hex,
762                             int only_two_digit_prefix)
763 {
764         struct strbuf buf = STRBUF_INIT;
765         append_remote_object_url(&buf, url, hex, only_two_digit_prefix);
766         return strbuf_detach(&buf, NULL);
767 }
768
769 /* http_request() targets */
770 #define HTTP_REQUEST_STRBUF     0
771 #define HTTP_REQUEST_FILE       1
772
773 static int http_request(const char *url, void *result, int target, int options)
774 {
775         struct active_request_slot *slot;
776         struct slot_results results;
777         struct curl_slist *headers = NULL;
778         struct strbuf buf = STRBUF_INIT;
779         int ret;
780
781         slot = get_active_slot();
782         slot->results = &results;
783         curl_easy_setopt(slot->curl, CURLOPT_HTTPGET, 1);
784
785         if (result == NULL) {
786                 curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 1);
787         } else {
788                 curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0);
789                 curl_easy_setopt(slot->curl, CURLOPT_FILE, result);
790
791                 if (target == HTTP_REQUEST_FILE) {
792                         long posn = ftell(result);
793                         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
794                                          fwrite);
795                         if (posn > 0) {
796                                 strbuf_addf(&buf, "Range: bytes=%ld-", posn);
797                                 headers = curl_slist_append(headers, buf.buf);
798                                 strbuf_reset(&buf);
799                         }
800                         slot->local = result;
801                 } else
802                         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
803                                          fwrite_buffer);
804         }
805
806         strbuf_addstr(&buf, "Pragma:");
807         if (options & HTTP_NO_CACHE)
808                 strbuf_addstr(&buf, " no-cache");
809
810         headers = curl_slist_append(headers, buf.buf);
811
812         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
813         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, headers);
814
815         if (start_active_slot(slot)) {
816                 run_active_slot(slot);
817                 if (results.curl_result == CURLE_OK)
818                         ret = HTTP_OK;
819                 else if (missing_target(&results))
820                         ret = HTTP_MISSING_TARGET;
821                 else if (results.http_code == 401) {
822                         if (user_name) {
823                                 ret = HTTP_NOAUTH;
824                         } else {
825                                 /*
826                                  * git_getpass is needed here because its very likely stdin/stdout are
827                                  * pipes to our parent process.  So we instead need to use /dev/tty,
828                                  * but that is non-portable.  Using git_getpass() can at least be stubbed
829                                  * on other platforms with a different implementation if/when necessary.
830                                  */
831                                 user_name = xstrdup(git_getpass("Username: "));
832                                 init_curl_http_auth(slot->curl);
833                                 ret = HTTP_REAUTH;
834                         }
835                 } else
836                         ret = HTTP_ERROR;
837         } else {
838                 error("Unable to start HTTP request for %s", url);
839                 ret = HTTP_START_FAILED;
840         }
841
842         slot->local = NULL;
843         curl_slist_free_all(headers);
844         strbuf_release(&buf);
845
846         return ret;
847 }
848
849 int http_get_strbuf(const char *url, struct strbuf *result, int options)
850 {
851         int http_ret = http_request(url, result, HTTP_REQUEST_STRBUF, options);
852         if (http_ret == HTTP_REAUTH) {
853                 http_ret = http_request(url, result, HTTP_REQUEST_STRBUF, options);
854         }
855         return http_ret;
856 }
857
858 /*
859  * Downloads an url and stores the result in the given file.
860  *
861  * If a previous interrupted download is detected (i.e. a previous temporary
862  * file is still around) the download is resumed.
863  */
864 static int http_get_file(const char *url, const char *filename, int options)
865 {
866         int ret;
867         struct strbuf tmpfile = STRBUF_INIT;
868         FILE *result;
869
870         strbuf_addf(&tmpfile, "%s.temp", filename);
871         result = fopen(tmpfile.buf, "a");
872         if (! result) {
873                 error("Unable to open local file %s", tmpfile.buf);
874                 ret = HTTP_ERROR;
875                 goto cleanup;
876         }
877
878         ret = http_request(url, result, HTTP_REQUEST_FILE, options);
879         fclose(result);
880
881         if ((ret == HTTP_OK) && move_temp_to_file(tmpfile.buf, filename))
882                 ret = HTTP_ERROR;
883 cleanup:
884         strbuf_release(&tmpfile);
885         return ret;
886 }
887
888 int http_error(const char *url, int ret)
889 {
890         /* http_request has already handled HTTP_START_FAILED. */
891         if (ret != HTTP_START_FAILED)
892                 error("%s while accessing %s\n", curl_errorstr, url);
893
894         return ret;
895 }
896
897 int http_fetch_ref(const char *base, struct ref *ref)
898 {
899         char *url;
900         struct strbuf buffer = STRBUF_INIT;
901         int ret = -1;
902
903         url = quote_ref_url(base, ref->name);
904         if (http_get_strbuf(url, &buffer, HTTP_NO_CACHE) == HTTP_OK) {
905                 strbuf_rtrim(&buffer);
906                 if (buffer.len == 40)
907                         ret = get_sha1_hex(buffer.buf, ref->old_sha1);
908                 else if (!prefixcmp(buffer.buf, "ref: ")) {
909                         ref->symref = xstrdup(buffer.buf + 5);
910                         ret = 0;
911                 }
912         }
913
914         strbuf_release(&buffer);
915         free(url);
916         return ret;
917 }
918
919 /* Helpers for fetching packs */
920 static char *fetch_pack_index(unsigned char *sha1, const char *base_url)
921 {
922         char *url, *tmp;
923         struct strbuf buf = STRBUF_INIT;
924
925         if (http_is_verbose)
926                 fprintf(stderr, "Getting index for pack %s\n", sha1_to_hex(sha1));
927
928         end_url_with_slash(&buf, base_url);
929         strbuf_addf(&buf, "objects/pack/pack-%s.idx", sha1_to_hex(sha1));
930         url = strbuf_detach(&buf, NULL);
931
932         strbuf_addf(&buf, "%s.temp", sha1_pack_index_name(sha1));
933         tmp = strbuf_detach(&buf, NULL);
934
935         if (http_get_file(url, tmp, 0) != HTTP_OK) {
936                 error("Unable to get pack index %s\n", url);
937                 free(tmp);
938                 tmp = NULL;
939         }
940
941         free(url);
942         return tmp;
943 }
944
945 static int fetch_and_setup_pack_index(struct packed_git **packs_head,
946         unsigned char *sha1, const char *base_url)
947 {
948         struct packed_git *new_pack;
949         char *tmp_idx = NULL;
950         int ret;
951
952         if (has_pack_index(sha1)) {
953                 new_pack = parse_pack_index(sha1, NULL);
954                 if (!new_pack)
955                         return -1; /* parse_pack_index() already issued error message */
956                 goto add_pack;
957         }
958
959         tmp_idx = fetch_pack_index(sha1, base_url);
960         if (!tmp_idx)
961                 return -1;
962
963         new_pack = parse_pack_index(sha1, tmp_idx);
964         if (!new_pack) {
965                 unlink(tmp_idx);
966                 free(tmp_idx);
967
968                 return -1; /* parse_pack_index() already issued error message */
969         }
970
971         ret = verify_pack_index(new_pack);
972         if (!ret) {
973                 close_pack_index(new_pack);
974                 ret = move_temp_to_file(tmp_idx, sha1_pack_index_name(sha1));
975         }
976         free(tmp_idx);
977         if (ret)
978                 return -1;
979
980 add_pack:
981         new_pack->next = *packs_head;
982         *packs_head = new_pack;
983         return 0;
984 }
985
986 int http_get_info_packs(const char *base_url, struct packed_git **packs_head)
987 {
988         int ret = 0, i = 0;
989         char *url, *data;
990         struct strbuf buf = STRBUF_INIT;
991         unsigned char sha1[20];
992
993         end_url_with_slash(&buf, base_url);
994         strbuf_addstr(&buf, "objects/info/packs");
995         url = strbuf_detach(&buf, NULL);
996
997         ret = http_get_strbuf(url, &buf, HTTP_NO_CACHE);
998         if (ret != HTTP_OK)
999                 goto cleanup;
1000
1001         data = buf.buf;
1002         while (i < buf.len) {
1003                 switch (data[i]) {
1004                 case 'P':
1005                         i++;
1006                         if (i + 52 <= buf.len &&
1007                             !prefixcmp(data + i, " pack-") &&
1008                             !prefixcmp(data + i + 46, ".pack\n")) {
1009                                 get_sha1_hex(data + i + 6, sha1);
1010                                 fetch_and_setup_pack_index(packs_head, sha1,
1011                                                       base_url);
1012                                 i += 51;
1013                                 break;
1014                         }
1015                 default:
1016                         while (i < buf.len && data[i] != '\n')
1017                                 i++;
1018                 }
1019                 i++;
1020         }
1021
1022 cleanup:
1023         free(url);
1024         return ret;
1025 }
1026
1027 void release_http_pack_request(struct http_pack_request *preq)
1028 {
1029         if (preq->packfile != NULL) {
1030                 fclose(preq->packfile);
1031                 preq->packfile = NULL;
1032                 preq->slot->local = NULL;
1033         }
1034         if (preq->range_header != NULL) {
1035                 curl_slist_free_all(preq->range_header);
1036                 preq->range_header = NULL;
1037         }
1038         preq->slot = NULL;
1039         free(preq->url);
1040 }
1041
1042 int finish_http_pack_request(struct http_pack_request *preq)
1043 {
1044         struct packed_git **lst;
1045         struct packed_git *p = preq->target;
1046         char *tmp_idx;
1047         struct child_process ip;
1048         const char *ip_argv[8];
1049
1050         close_pack_index(p);
1051
1052         fclose(preq->packfile);
1053         preq->packfile = NULL;
1054         preq->slot->local = NULL;
1055
1056         lst = preq->lst;
1057         while (*lst != p)
1058                 lst = &((*lst)->next);
1059         *lst = (*lst)->next;
1060
1061         tmp_idx = xstrdup(preq->tmpfile);
1062         strcpy(tmp_idx + strlen(tmp_idx) - strlen(".pack.temp"),
1063                ".idx.temp");
1064
1065         ip_argv[0] = "index-pack";
1066         ip_argv[1] = "-o";
1067         ip_argv[2] = tmp_idx;
1068         ip_argv[3] = preq->tmpfile;
1069         ip_argv[4] = NULL;
1070
1071         memset(&ip, 0, sizeof(ip));
1072         ip.argv = ip_argv;
1073         ip.git_cmd = 1;
1074         ip.no_stdin = 1;
1075         ip.no_stdout = 1;
1076
1077         if (run_command(&ip)) {
1078                 unlink(preq->tmpfile);
1079                 unlink(tmp_idx);
1080                 free(tmp_idx);
1081                 return -1;
1082         }
1083
1084         unlink(sha1_pack_index_name(p->sha1));
1085
1086         if (move_temp_to_file(preq->tmpfile, sha1_pack_name(p->sha1))
1087          || move_temp_to_file(tmp_idx, sha1_pack_index_name(p->sha1))) {
1088                 free(tmp_idx);
1089                 return -1;
1090         }
1091
1092         install_packed_git(p);
1093         free(tmp_idx);
1094         return 0;
1095 }
1096
1097 struct http_pack_request *new_http_pack_request(
1098         struct packed_git *target, const char *base_url)
1099 {
1100         long prev_posn = 0;
1101         char range[RANGE_HEADER_SIZE];
1102         struct strbuf buf = STRBUF_INIT;
1103         struct http_pack_request *preq;
1104
1105         preq = xmalloc(sizeof(*preq));
1106         preq->target = target;
1107         preq->range_header = NULL;
1108
1109         end_url_with_slash(&buf, base_url);
1110         strbuf_addf(&buf, "objects/pack/pack-%s.pack",
1111                 sha1_to_hex(target->sha1));
1112         preq->url = strbuf_detach(&buf, NULL);
1113
1114         snprintf(preq->tmpfile, sizeof(preq->tmpfile), "%s.temp",
1115                 sha1_pack_name(target->sha1));
1116         preq->packfile = fopen(preq->tmpfile, "a");
1117         if (!preq->packfile) {
1118                 error("Unable to open local file %s for pack",
1119                       preq->tmpfile);
1120                 goto abort;
1121         }
1122
1123         preq->slot = get_active_slot();
1124         preq->slot->local = preq->packfile;
1125         curl_easy_setopt(preq->slot->curl, CURLOPT_FILE, preq->packfile);
1126         curl_easy_setopt(preq->slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
1127         curl_easy_setopt(preq->slot->curl, CURLOPT_URL, preq->url);
1128         curl_easy_setopt(preq->slot->curl, CURLOPT_HTTPHEADER,
1129                 no_pragma_header);
1130
1131         /*
1132          * If there is data present from a previous transfer attempt,
1133          * resume where it left off
1134          */
1135         prev_posn = ftell(preq->packfile);
1136         if (prev_posn>0) {
1137                 if (http_is_verbose)
1138                         fprintf(stderr,
1139                                 "Resuming fetch of pack %s at byte %ld\n",
1140                                 sha1_to_hex(target->sha1), prev_posn);
1141                 sprintf(range, "Range: bytes=%ld-", prev_posn);
1142                 preq->range_header = curl_slist_append(NULL, range);
1143                 curl_easy_setopt(preq->slot->curl, CURLOPT_HTTPHEADER,
1144                         preq->range_header);
1145         }
1146
1147         return preq;
1148
1149 abort:
1150         free(preq->url);
1151         free(preq);
1152         return NULL;
1153 }
1154
1155 /* Helpers for fetching objects (loose) */
1156 static size_t fwrite_sha1_file(char *ptr, size_t eltsize, size_t nmemb,
1157                                void *data)
1158 {
1159         unsigned char expn[4096];
1160         size_t size = eltsize * nmemb;
1161         int posn = 0;
1162         struct http_object_request *freq =
1163                 (struct http_object_request *)data;
1164         do {
1165                 ssize_t retval = xwrite(freq->localfile,
1166                                         (char *) ptr + posn, size - posn);
1167                 if (retval < 0)
1168                         return posn;
1169                 posn += retval;
1170         } while (posn < size);
1171
1172         freq->stream.avail_in = size;
1173         freq->stream.next_in = (void *)ptr;
1174         do {
1175                 freq->stream.next_out = expn;
1176                 freq->stream.avail_out = sizeof(expn);
1177                 freq->zret = git_inflate(&freq->stream, Z_SYNC_FLUSH);
1178                 git_SHA1_Update(&freq->c, expn,
1179                                 sizeof(expn) - freq->stream.avail_out);
1180         } while (freq->stream.avail_in && freq->zret == Z_OK);
1181         data_received++;
1182         return size;
1183 }
1184
1185 struct http_object_request *new_http_object_request(const char *base_url,
1186         unsigned char *sha1)
1187 {
1188         char *hex = sha1_to_hex(sha1);
1189         char *filename;
1190         char prevfile[PATH_MAX];
1191         int prevlocal;
1192         char prev_buf[PREV_BUF_SIZE];
1193         ssize_t prev_read = 0;
1194         long prev_posn = 0;
1195         char range[RANGE_HEADER_SIZE];
1196         struct curl_slist *range_header = NULL;
1197         struct http_object_request *freq;
1198
1199         freq = xmalloc(sizeof(*freq));
1200         hashcpy(freq->sha1, sha1);
1201         freq->localfile = -1;
1202
1203         filename = sha1_file_name(sha1);
1204         snprintf(freq->tmpfile, sizeof(freq->tmpfile),
1205                  "%s.temp", filename);
1206
1207         snprintf(prevfile, sizeof(prevfile), "%s.prev", filename);
1208         unlink_or_warn(prevfile);
1209         rename(freq->tmpfile, prevfile);
1210         unlink_or_warn(freq->tmpfile);
1211
1212         if (freq->localfile != -1)
1213                 error("fd leakage in start: %d", freq->localfile);
1214         freq->localfile = open(freq->tmpfile,
1215                                O_WRONLY | O_CREAT | O_EXCL, 0666);
1216         /*
1217          * This could have failed due to the "lazy directory creation";
1218          * try to mkdir the last path component.
1219          */
1220         if (freq->localfile < 0 && errno == ENOENT) {
1221                 char *dir = strrchr(freq->tmpfile, '/');
1222                 if (dir) {
1223                         *dir = 0;
1224                         mkdir(freq->tmpfile, 0777);
1225                         *dir = '/';
1226                 }
1227                 freq->localfile = open(freq->tmpfile,
1228                                        O_WRONLY | O_CREAT | O_EXCL, 0666);
1229         }
1230
1231         if (freq->localfile < 0) {
1232                 error("Couldn't create temporary file %s: %s",
1233                       freq->tmpfile, strerror(errno));
1234                 goto abort;
1235         }
1236
1237         memset(&freq->stream, 0, sizeof(freq->stream));
1238
1239         git_inflate_init(&freq->stream);
1240
1241         git_SHA1_Init(&freq->c);
1242
1243         freq->url = get_remote_object_url(base_url, hex, 0);
1244
1245         /*
1246          * If a previous temp file is present, process what was already
1247          * fetched.
1248          */
1249         prevlocal = open(prevfile, O_RDONLY);
1250         if (prevlocal != -1) {
1251                 do {
1252                         prev_read = xread(prevlocal, prev_buf, PREV_BUF_SIZE);
1253                         if (prev_read>0) {
1254                                 if (fwrite_sha1_file(prev_buf,
1255                                                      1,
1256                                                      prev_read,
1257                                                      freq) == prev_read) {
1258                                         prev_posn += prev_read;
1259                                 } else {
1260                                         prev_read = -1;
1261                                 }
1262                         }
1263                 } while (prev_read > 0);
1264                 close(prevlocal);
1265         }
1266         unlink_or_warn(prevfile);
1267
1268         /*
1269          * Reset inflate/SHA1 if there was an error reading the previous temp
1270          * file; also rewind to the beginning of the local file.
1271          */
1272         if (prev_read == -1) {
1273                 memset(&freq->stream, 0, sizeof(freq->stream));
1274                 git_inflate_init(&freq->stream);
1275                 git_SHA1_Init(&freq->c);
1276                 if (prev_posn>0) {
1277                         prev_posn = 0;
1278                         lseek(freq->localfile, 0, SEEK_SET);
1279                         if (ftruncate(freq->localfile, 0) < 0) {
1280                                 error("Couldn't truncate temporary file %s: %s",
1281                                           freq->tmpfile, strerror(errno));
1282                                 goto abort;
1283                         }
1284                 }
1285         }
1286
1287         freq->slot = get_active_slot();
1288
1289         curl_easy_setopt(freq->slot->curl, CURLOPT_FILE, freq);
1290         curl_easy_setopt(freq->slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file);
1291         curl_easy_setopt(freq->slot->curl, CURLOPT_ERRORBUFFER, freq->errorstr);
1292         curl_easy_setopt(freq->slot->curl, CURLOPT_URL, freq->url);
1293         curl_easy_setopt(freq->slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
1294
1295         /*
1296          * If we have successfully processed data from a previous fetch
1297          * attempt, only fetch the data we don't already have.
1298          */
1299         if (prev_posn>0) {
1300                 if (http_is_verbose)
1301                         fprintf(stderr,
1302                                 "Resuming fetch of object %s at byte %ld\n",
1303                                 hex, prev_posn);
1304                 sprintf(range, "Range: bytes=%ld-", prev_posn);
1305                 range_header = curl_slist_append(range_header, range);
1306                 curl_easy_setopt(freq->slot->curl,
1307                                  CURLOPT_HTTPHEADER, range_header);
1308         }
1309
1310         return freq;
1311
1312 abort:
1313         free(filename);
1314         free(freq->url);
1315         free(freq);
1316         return NULL;
1317 }
1318
1319 void process_http_object_request(struct http_object_request *freq)
1320 {
1321         if (freq->slot == NULL)
1322                 return;
1323         freq->curl_result = freq->slot->curl_result;
1324         freq->http_code = freq->slot->http_code;
1325         freq->slot = NULL;
1326 }
1327
1328 int finish_http_object_request(struct http_object_request *freq)
1329 {
1330         struct stat st;
1331
1332         close(freq->localfile);
1333         freq->localfile = -1;
1334
1335         process_http_object_request(freq);
1336
1337         if (freq->http_code == 416) {
1338                 warning("requested range invalid; we may already have all the data.");
1339         } else if (freq->curl_result != CURLE_OK) {
1340                 if (stat(freq->tmpfile, &st) == 0)
1341                         if (st.st_size == 0)
1342                                 unlink_or_warn(freq->tmpfile);
1343                 return -1;
1344         }
1345
1346         git_inflate_end(&freq->stream);
1347         git_SHA1_Final(freq->real_sha1, &freq->c);
1348         if (freq->zret != Z_STREAM_END) {
1349                 unlink_or_warn(freq->tmpfile);
1350                 return -1;
1351         }
1352         if (hashcmp(freq->sha1, freq->real_sha1)) {
1353                 unlink_or_warn(freq->tmpfile);
1354                 return -1;
1355         }
1356         freq->rename =
1357                 move_temp_to_file(freq->tmpfile, sha1_file_name(freq->sha1));
1358
1359         return freq->rename;
1360 }
1361
1362 void abort_http_object_request(struct http_object_request *freq)
1363 {
1364         unlink_or_warn(freq->tmpfile);
1365
1366         release_http_object_request(freq);
1367 }
1368
1369 void release_http_object_request(struct http_object_request *freq)
1370 {
1371         if (freq->localfile != -1) {
1372                 close(freq->localfile);
1373                 freq->localfile = -1;
1374         }
1375         if (freq->url != NULL) {
1376                 free(freq->url);
1377                 freq->url = NULL;
1378         }
1379         if (freq->slot != NULL) {
1380                 freq->slot->callback_func = NULL;
1381                 freq->slot->callback_data = NULL;
1382                 release_active_slot(freq->slot);
1383                 freq->slot = NULL;
1384         }
1385 }