Merge branch 'svnup' of http://netz.smurf.noris.de/git/git
[git.git] / http-fetch.c
1 #include "cache.h"
2 #include "commit.h"
3 #include "pack.h"
4 #include "fetch.h"
5
6 #include <curl/curl.h>
7 #include <curl/easy.h>
8
9 #if LIBCURL_VERSION_NUM >= 0x070908
10 #define USE_CURL_MULTI
11 #define DEFAULT_MAX_REQUESTS 5
12 #endif
13
14 #if LIBCURL_VERSION_NUM < 0x070704
15 #define curl_global_cleanup() do { /* nothing */ } while(0)
16 #endif
17 #if LIBCURL_VERSION_NUM < 0x070800
18 #define curl_global_init(a) do { /* nothing */ } while(0)
19 #endif
20
21 #if LIBCURL_VERSION_NUM < 0x070c04
22 #define NO_CURL_EASY_DUPHANDLE
23 #endif
24
25 #define PREV_BUF_SIZE 4096
26 #define RANGE_HEADER_SIZE 30
27
28 static int got_alternates = -1;
29 static int active_requests = 0;
30 static int data_received;
31
32 #ifdef USE_CURL_MULTI
33 static int max_requests = -1;
34 static CURLM *curlm;
35 #endif
36 #ifndef NO_CURL_EASY_DUPHANDLE
37 static CURL *curl_default;
38 #endif
39 static struct curl_slist *pragma_header;
40 static struct curl_slist *no_pragma_header;
41 static struct curl_slist *no_range_header;
42 static char curl_errorstr[CURL_ERROR_SIZE];
43
44 struct alt_base
45 {
46         char *base;
47         int got_indices;
48         struct packed_git *packs;
49         struct alt_base *next;
50 };
51
52 static struct alt_base *alt = NULL;
53
54 enum transfer_state {
55         WAITING,
56         ABORTED,
57         ACTIVE,
58         COMPLETE,
59 };
60
61 struct transfer_request
62 {
63         unsigned char sha1[20];
64         struct alt_base *repo;
65         char *url;
66         char filename[PATH_MAX];
67         char tmpfile[PATH_MAX];
68         int local;
69         enum transfer_state state;
70         CURLcode curl_result;
71         char errorstr[CURL_ERROR_SIZE];
72         long http_code;
73         unsigned char real_sha1[20];
74         SHA_CTX c;
75         z_stream stream;
76         int zret;
77         int rename;
78         struct active_request_slot *slot;
79         struct transfer_request *next;
80 };
81
82 struct active_request_slot
83 {
84         CURL *curl;
85         FILE *local;
86         int in_use;
87         int done;
88         CURLcode curl_result;
89         long http_code;
90         void *callback_data;
91         void (*callback_func)(void *data);
92         struct active_request_slot *next;
93 };
94
95 struct alt_request {
96         char *base;
97         char *url;
98         struct buffer *buffer;
99         struct active_request_slot *slot;
100         int http_specific;
101 };
102
103 static struct transfer_request *request_queue_head = NULL;
104 static struct active_request_slot *active_queue_head = NULL;
105
106 static int curl_ssl_verify = -1;
107 static char *ssl_cert = NULL;
108 #if LIBCURL_VERSION_NUM >= 0x070902
109 static char *ssl_key = NULL;
110 #endif
111 #if LIBCURL_VERSION_NUM >= 0x070908
112 static char *ssl_capath = NULL;
113 #endif
114 static char *ssl_cainfo = NULL;
115 static long curl_low_speed_limit = -1;
116 static long curl_low_speed_time = -1;
117
118 struct buffer
119 {
120         size_t posn;
121         size_t size;
122         void *buffer;
123 };
124
125 static int http_options(const char *var, const char *value)
126 {
127         if (!strcmp("http.sslverify", var)) {
128                 if (curl_ssl_verify == -1) {
129                         curl_ssl_verify = git_config_bool(var, value);
130                 }
131                 return 0;
132         }
133
134         if (!strcmp("http.sslcert", var)) {
135                 if (ssl_cert == NULL) {
136                         ssl_cert = xmalloc(strlen(value)+1);
137                         strcpy(ssl_cert, value);
138                 }
139                 return 0;
140         }
141 #if LIBCURL_VERSION_NUM >= 0x070902
142         if (!strcmp("http.sslkey", var)) {
143                 if (ssl_key == NULL) {
144                         ssl_key = xmalloc(strlen(value)+1);
145                         strcpy(ssl_key, value);
146                 }
147                 return 0;
148         }
149 #endif
150 #if LIBCURL_VERSION_NUM >= 0x070908
151         if (!strcmp("http.sslcapath", var)) {
152                 if (ssl_capath == NULL) {
153                         ssl_capath = xmalloc(strlen(value)+1);
154                         strcpy(ssl_capath, value);
155                 }
156                 return 0;
157         }
158 #endif
159         if (!strcmp("http.sslcainfo", var)) {
160                 if (ssl_cainfo == NULL) {
161                         ssl_cainfo = xmalloc(strlen(value)+1);
162                         strcpy(ssl_cainfo, value);
163                 }
164                 return 0;
165         }
166
167 #ifdef USE_CURL_MULTI   
168         if (!strcmp("http.maxrequests", var)) {
169                 if (max_requests == -1)
170                         max_requests = git_config_int(var, value);
171                 return 0;
172         }
173 #endif
174
175         if (!strcmp("http.lowspeedlimit", var)) {
176                 if (curl_low_speed_limit == -1)
177                         curl_low_speed_limit = (long)git_config_int(var, value);
178                 return 0;
179         }
180         if (!strcmp("http.lowspeedtime", var)) {
181                 if (curl_low_speed_time == -1)
182                         curl_low_speed_time = (long)git_config_int(var, value);
183                 return 0;
184         }
185
186         /* Fall back on the default ones */
187         return git_default_config(var, value);
188 }
189
190 static size_t fwrite_buffer(void *ptr, size_t eltsize, size_t nmemb,
191                             struct buffer *buffer)
192 {
193         size_t size = eltsize * nmemb;
194         if (size > buffer->size - buffer->posn)
195                 size = buffer->size - buffer->posn;
196         memcpy(buffer->buffer + buffer->posn, ptr, size);
197         buffer->posn += size;
198         data_received++;
199         return size;
200 }
201
202 static size_t fwrite_buffer_dynamic(const void *ptr, size_t eltsize,
203                                     size_t nmemb, struct buffer *buffer)
204 {
205         size_t size = eltsize * nmemb;
206         if (size > buffer->size - buffer->posn) {
207                 buffer->size = buffer->size * 3 / 2;
208                 if (buffer->size < buffer->posn + size)
209                         buffer->size = buffer->posn + size;
210                 buffer->buffer = xrealloc(buffer->buffer, buffer->size);
211         }
212         memcpy(buffer->buffer + buffer->posn, ptr, size);
213         buffer->posn += size;
214         data_received++;
215         return size;
216 }
217
218 static size_t fwrite_sha1_file(void *ptr, size_t eltsize, size_t nmemb,
219                                void *data)
220 {
221         unsigned char expn[4096];
222         size_t size = eltsize * nmemb;
223         int posn = 0;
224         struct transfer_request *request = (struct transfer_request *)data;
225         do {
226                 ssize_t retval = write(request->local,
227                                        ptr + posn, size - posn);
228                 if (retval < 0)
229                         return posn;
230                 posn += retval;
231         } while (posn < size);
232
233         request->stream.avail_in = size;
234         request->stream.next_in = ptr;
235         do {
236                 request->stream.next_out = expn;
237                 request->stream.avail_out = sizeof(expn);
238                 request->zret = inflate(&request->stream, Z_SYNC_FLUSH);
239                 SHA1_Update(&request->c, expn,
240                             sizeof(expn) - request->stream.avail_out);
241         } while (request->stream.avail_in && request->zret == Z_OK);
242         data_received++;
243         return size;
244 }
245
246 #ifdef USE_CURL_MULTI
247 static void process_curl_messages(void);
248 static void process_request_queue(void);
249 #endif
250 static void fetch_alternates(char *base);
251
252 static CURL* get_curl_handle(void)
253 {
254         CURL* result = curl_easy_init();
255
256         curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, curl_ssl_verify);
257 #if LIBCURL_VERSION_NUM >= 0x070907
258         curl_easy_setopt(result, CURLOPT_NETRC, CURL_NETRC_OPTIONAL);
259 #endif
260
261         if (ssl_cert != NULL)
262                 curl_easy_setopt(result, CURLOPT_SSLCERT, ssl_cert);
263 #if LIBCURL_VERSION_NUM >= 0x070902
264         if (ssl_key != NULL)
265                 curl_easy_setopt(result, CURLOPT_SSLKEY, ssl_key);
266 #endif
267 #if LIBCURL_VERSION_NUM >= 0x070908
268         if (ssl_capath != NULL)
269                 curl_easy_setopt(result, CURLOPT_CAPATH, ssl_capath);
270 #endif
271         if (ssl_cainfo != NULL)
272                 curl_easy_setopt(result, CURLOPT_CAINFO, ssl_cainfo);
273         curl_easy_setopt(result, CURLOPT_FAILONERROR, 1);
274
275         if (curl_low_speed_limit > 0 && curl_low_speed_time > 0) {
276                 curl_easy_setopt(result, CURLOPT_LOW_SPEED_LIMIT,
277                                  curl_low_speed_limit);
278                 curl_easy_setopt(result, CURLOPT_LOW_SPEED_TIME,
279                                  curl_low_speed_time);
280         }
281
282         curl_easy_setopt(result, CURLOPT_FOLLOWLOCATION, 1);
283
284         return result;
285 }
286
287 static struct active_request_slot *get_active_slot(void)
288 {
289         struct active_request_slot *slot = active_queue_head;
290         struct active_request_slot *newslot;
291
292 #ifdef USE_CURL_MULTI
293         int num_transfers;
294
295         /* Wait for a slot to open up if the queue is full */
296         while (active_requests >= max_requests) {
297                 curl_multi_perform(curlm, &num_transfers);
298                 if (num_transfers < active_requests) {
299                         process_curl_messages();
300                 }
301         }
302 #endif
303
304         while (slot != NULL && slot->in_use) {
305                 slot = slot->next;
306         }
307         if (slot == NULL) {
308                 newslot = xmalloc(sizeof(*newslot));
309                 newslot->curl = NULL;
310                 newslot->in_use = 0;
311                 newslot->next = NULL;
312
313                 slot = active_queue_head;
314                 if (slot == NULL) {
315                         active_queue_head = newslot;
316                 } else {
317                         while (slot->next != NULL) {
318                                 slot = slot->next;
319                         }
320                         slot->next = newslot;
321                 }
322                 slot = newslot;
323         }
324
325         if (slot->curl == NULL) {
326 #ifdef NO_CURL_EASY_DUPHANDLE
327                 slot->curl = get_curl_handle();
328 #else
329                 slot->curl = curl_easy_duphandle(curl_default);
330 #endif
331         }
332
333         active_requests++;
334         slot->in_use = 1;
335         slot->done = 0;
336         slot->local = NULL;
337         slot->callback_data = NULL;
338         slot->callback_func = NULL;
339         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, pragma_header);
340         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_range_header);
341         curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, curl_errorstr);
342
343         return slot;
344 }
345
346 static int start_active_slot(struct active_request_slot *slot)
347 {
348 #ifdef USE_CURL_MULTI
349         CURLMcode curlm_result = curl_multi_add_handle(curlm, slot->curl);
350
351         if (curlm_result != CURLM_OK &&
352             curlm_result != CURLM_CALL_MULTI_PERFORM) {
353                 active_requests--;
354                 slot->in_use = 0;
355                 return 0;
356         }
357 #endif
358         return 1;
359 }
360
361 static void run_active_slot(struct active_request_slot *slot)
362 {
363 #ifdef USE_CURL_MULTI
364         int num_transfers;
365         long last_pos = 0;
366         long current_pos;
367         fd_set readfds;
368         fd_set writefds;
369         fd_set excfds;
370         int max_fd;
371         struct timeval select_timeout;
372         CURLMcode curlm_result;
373
374         while (!slot->done) {
375                 data_received = 0;
376                 do {
377                         curlm_result = curl_multi_perform(curlm,
378                                                           &num_transfers);
379                 } while (curlm_result == CURLM_CALL_MULTI_PERFORM);
380                 if (num_transfers < active_requests) {
381                         process_curl_messages();
382                         process_request_queue();
383                 }
384
385                 if (!data_received && slot->local != NULL) {
386                         current_pos = ftell(slot->local);
387                         if (current_pos > last_pos)
388                                 data_received++;
389                         last_pos = current_pos;
390                 }
391
392                 if (!slot->done && !data_received) {
393                         max_fd = 0;
394                         FD_ZERO(&readfds);
395                         FD_ZERO(&writefds);
396                         FD_ZERO(&excfds);
397                         select_timeout.tv_sec = 0;
398                         select_timeout.tv_usec = 50000;
399                         select(max_fd, &readfds, &writefds,
400                                &excfds, &select_timeout);
401                 }
402         }
403 #else
404         slot->curl_result = curl_easy_perform(slot->curl);
405         active_requests--;
406 #endif
407 }
408
409 static void start_request(struct transfer_request *request)
410 {
411         char *hex = sha1_to_hex(request->sha1);
412         char prevfile[PATH_MAX];
413         char *url;
414         char *posn;
415         int prevlocal;
416         unsigned char prev_buf[PREV_BUF_SIZE];
417         ssize_t prev_read = 0;
418         long prev_posn = 0;
419         char range[RANGE_HEADER_SIZE];
420         struct curl_slist *range_header = NULL;
421         struct active_request_slot *slot;
422
423         snprintf(prevfile, sizeof(prevfile), "%s.prev", request->filename);
424         unlink(prevfile);
425         rename(request->tmpfile, prevfile);
426         unlink(request->tmpfile);
427
428         request->local = open(request->tmpfile,
429                               O_WRONLY | O_CREAT | O_EXCL, 0666);
430         /* This could have failed due to the "lazy directory creation";
431          * try to mkdir the last path component.
432          */
433         if (request->local < 0 && errno == ENOENT) {
434                 char *dir = strrchr(request->tmpfile, '/');
435                 if (dir) {
436                         *dir = 0;
437                         mkdir(request->tmpfile, 0777);
438                         *dir = '/';
439                 }
440                 request->local = open(request->tmpfile,
441                                       O_WRONLY | O_CREAT | O_EXCL, 0666);
442         }
443
444         if (request->local < 0) {
445                 request->state = ABORTED;
446                 error("Couldn't create temporary file %s for %s: %s\n",
447                       request->tmpfile, request->filename, strerror(errno));
448                 return;
449         }
450
451         memset(&request->stream, 0, sizeof(request->stream));
452
453         inflateInit(&request->stream);
454
455         SHA1_Init(&request->c);
456
457         url = xmalloc(strlen(request->repo->base) + 50);
458         request->url = xmalloc(strlen(request->repo->base) + 50);
459         strcpy(url, request->repo->base);
460         posn = url + strlen(request->repo->base);
461         strcpy(posn, "objects/");
462         posn += 8;
463         memcpy(posn, hex, 2);
464         posn += 2;
465         *(posn++) = '/';
466         strcpy(posn, hex + 2);
467         strcpy(request->url, url);
468
469         /* If a previous temp file is present, process what was already
470            fetched. */
471         prevlocal = open(prevfile, O_RDONLY);
472         if (prevlocal != -1) {
473                 do {
474                         prev_read = read(prevlocal, prev_buf, PREV_BUF_SIZE);
475                         if (prev_read>0) {
476                                 if (fwrite_sha1_file(prev_buf,
477                                                      1,
478                                                      prev_read,
479                                                      request) == prev_read) {
480                                         prev_posn += prev_read;
481                                 } else {
482                                         prev_read = -1;
483                                 }
484                         }
485                 } while (prev_read > 0);
486                 close(prevlocal);
487         }
488         unlink(prevfile);
489
490         /* Reset inflate/SHA1 if there was an error reading the previous temp
491            file; also rewind to the beginning of the local file. */
492         if (prev_read == -1) {
493                 memset(&request->stream, 0, sizeof(request->stream));
494                 inflateInit(&request->stream);
495                 SHA1_Init(&request->c);
496                 if (prev_posn>0) {
497                         prev_posn = 0;
498                         lseek(request->local, SEEK_SET, 0);
499                         ftruncate(request->local, 0);
500                 }
501         }
502
503         slot = get_active_slot();
504         curl_easy_setopt(slot->curl, CURLOPT_FILE, request);
505         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file);
506         curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, request->errorstr);
507         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
508         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
509
510         /* If we have successfully processed data from a previous fetch
511            attempt, only fetch the data we don't already have. */
512         if (prev_posn>0) {
513                 if (get_verbosely)
514                         fprintf(stderr,
515                                 "Resuming fetch of object %s at byte %ld\n",
516                                 hex, prev_posn);
517                 sprintf(range, "Range: bytes=%ld-", prev_posn);
518                 range_header = curl_slist_append(range_header, range);
519                 curl_easy_setopt(slot->curl,
520                                  CURLOPT_HTTPHEADER, range_header);
521         }
522
523         /* Try to get the request started, abort the request on error */
524         if (!start_active_slot(slot)) {
525                 request->state = ABORTED;
526                 close(request->local);
527                 free(request->url);
528                 return;
529         }
530         
531         request->slot = slot;
532         request->state = ACTIVE;
533 }
534
535 static void finish_request(struct transfer_request *request)
536 {
537         struct stat st;
538
539         fchmod(request->local, 0444);
540         close(request->local);
541
542         if (request->http_code == 416) {
543                 fprintf(stderr, "Warning: requested range invalid; we may already have all the data.\n");
544         } else if (request->curl_result != CURLE_OK) {
545                 if (stat(request->tmpfile, &st) == 0)
546                         if (st.st_size == 0)
547                                 unlink(request->tmpfile);
548                 return;
549         }
550
551         inflateEnd(&request->stream);
552         SHA1_Final(request->real_sha1, &request->c);
553         if (request->zret != Z_STREAM_END) {
554                 unlink(request->tmpfile);
555                 return;
556         }
557         if (memcmp(request->sha1, request->real_sha1, 20)) {
558                 unlink(request->tmpfile);
559                 return;
560         }
561         request->rename =
562                 move_temp_to_file(request->tmpfile, request->filename);
563
564         if (request->rename == 0)
565                 pull_say("got %s\n", sha1_to_hex(request->sha1));
566 }
567
568 static void release_request(struct transfer_request *request)
569 {
570         struct transfer_request *entry = request_queue_head;
571
572         if (request == request_queue_head) {
573                 request_queue_head = request->next;
574         } else {
575                 while (entry->next != NULL && entry->next != request)
576                         entry = entry->next;
577                 if (entry->next == request)
578                         entry->next = entry->next->next;
579         }
580
581         free(request->url);
582         free(request);
583 }
584
585 #ifdef USE_CURL_MULTI
586 static void process_curl_messages(void)
587 {
588         int num_messages;
589         struct active_request_slot *slot;
590         struct transfer_request *request = NULL;
591         CURLMsg *curl_message = curl_multi_info_read(curlm, &num_messages);
592
593         while (curl_message != NULL) {
594                 if (curl_message->msg == CURLMSG_DONE) {
595                         int curl_result = curl_message->data.result;
596                         slot = active_queue_head;
597                         while (slot != NULL &&
598                                slot->curl != curl_message->easy_handle)
599                                 slot = slot->next;
600                         if (slot != NULL) {
601                                 curl_multi_remove_handle(curlm, slot->curl);
602                                 active_requests--;
603                                 slot->done = 1;
604                                 slot->in_use = 0;
605                                 slot->curl_result = curl_result;
606                                 curl_easy_getinfo(slot->curl,
607                                                   CURLINFO_HTTP_CODE,
608                                                   &slot->http_code);
609                                 request = request_queue_head;
610                                 while (request != NULL &&
611                                        request->slot != slot)
612                                         request = request->next;
613                         } else {
614                                 fprintf(stderr, "Received DONE message for unknown request!\n");
615                         }
616
617                         /* Process slot callback if appropriate */
618                         if (slot->callback_func != NULL) {
619                                 slot->callback_func(slot->callback_data);
620                         }
621
622                         if (request != NULL) {
623                                 request->curl_result = curl_result;
624                                 request->http_code = slot->http_code;
625                                 request->slot = NULL;
626                                 request->state = COMPLETE;
627
628                                 /* Use alternates if necessary */
629                                 if (request->http_code == 404) {
630                                         fetch_alternates(alt->base);
631                                         if (request->repo->next != NULL) {
632                                                 request->repo =
633                                                         request->repo->next;
634                                                 start_request(request);
635                                         } else {
636                                                 finish_request(request);
637                                         }
638                                 } else {
639                                         finish_request(request);
640                                 }
641                         }
642                 } else {
643                         fprintf(stderr, "Unknown CURL message received: %d\n",
644                                 (int)curl_message->msg);
645                 }
646                 curl_message = curl_multi_info_read(curlm, &num_messages);
647         }
648 }
649
650 static void process_request_queue(void)
651 {
652         struct transfer_request *request = request_queue_head;
653         struct active_request_slot *slot = active_queue_head;
654         int num_transfers;
655
656         while (active_requests < max_requests && request != NULL) {
657                 if (request->state == WAITING) {
658                         if (has_sha1_file(request->sha1))
659                                 release_request(request);
660                         else
661                                 start_request(request);
662                         curl_multi_perform(curlm, &num_transfers);
663                 }
664                 request = request->next;
665         }
666
667         while (slot != NULL) {
668                 if (!slot->in_use && slot->curl != NULL) {
669                         curl_easy_cleanup(slot->curl);
670                         slot->curl = NULL;
671                 }
672                 slot = slot->next;
673         }                               
674 }
675 #endif
676
677 void prefetch(unsigned char *sha1)
678 {
679         struct transfer_request *newreq;
680         struct transfer_request *tail;
681         char *filename = sha1_file_name(sha1);
682
683         newreq = xmalloc(sizeof(*newreq));
684         memcpy(newreq->sha1, sha1, 20);
685         newreq->repo = alt;
686         newreq->url = NULL;
687         newreq->local = -1;
688         newreq->state = WAITING;
689         snprintf(newreq->filename, sizeof(newreq->filename), "%s", filename);
690         snprintf(newreq->tmpfile, sizeof(newreq->tmpfile),
691                  "%s.temp", filename);
692         newreq->next = NULL;
693
694         if (request_queue_head == NULL) {
695                 request_queue_head = newreq;
696         } else {
697                 tail = request_queue_head;
698                 while (tail->next != NULL) {
699                         tail = tail->next;
700                 }
701                 tail->next = newreq;
702         }
703 #ifdef USE_CURL_MULTI
704         process_request_queue();
705         process_curl_messages();
706 #endif
707 }
708
709 static int fetch_index(struct alt_base *repo, unsigned char *sha1)
710 {
711         char *hex = sha1_to_hex(sha1);
712         char *filename;
713         char *url;
714         char tmpfile[PATH_MAX];
715         long prev_posn = 0;
716         char range[RANGE_HEADER_SIZE];
717         struct curl_slist *range_header = NULL;
718
719         FILE *indexfile;
720         struct active_request_slot *slot;
721
722         if (has_pack_index(sha1))
723                 return 0;
724
725         if (get_verbosely)
726                 fprintf(stderr, "Getting index for pack %s\n", hex);
727         
728         url = xmalloc(strlen(repo->base) + 64);
729         sprintf(url, "%s/objects/pack/pack-%s.idx", repo->base, hex);
730         
731         filename = sha1_pack_index_name(sha1);
732         snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
733         indexfile = fopen(tmpfile, "a");
734         if (!indexfile)
735                 return error("Unable to open local file %s for pack index",
736                              filename);
737
738         slot = get_active_slot();
739         curl_easy_setopt(slot->curl, CURLOPT_FILE, indexfile);
740         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
741         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
742         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
743         slot->local = indexfile;
744
745         /* If there is data present from a previous transfer attempt,
746            resume where it left off */
747         prev_posn = ftell(indexfile);
748         if (prev_posn>0) {
749                 if (get_verbosely)
750                         fprintf(stderr,
751                                 "Resuming fetch of index for pack %s at byte %ld\n",
752                                 hex, prev_posn);
753                 sprintf(range, "Range: bytes=%ld-", prev_posn);
754                 range_header = curl_slist_append(range_header, range);
755                 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
756         }
757
758         if (start_active_slot(slot)) {
759                 run_active_slot(slot);
760                 if (slot->curl_result != CURLE_OK) {
761                         fclose(indexfile);
762                         return error("Unable to get pack index %s\n%s", url,
763                                      curl_errorstr);
764                 }
765         } else {
766                 return error("Unable to start request");
767         }
768
769         fclose(indexfile);
770
771         return move_temp_to_file(tmpfile, filename);
772 }
773
774 static int setup_index(struct alt_base *repo, unsigned char *sha1)
775 {
776         struct packed_git *new_pack;
777         if (has_pack_file(sha1))
778                 return 0; // don't list this as something we can get
779
780         if (fetch_index(repo, sha1))
781                 return -1;
782
783         new_pack = parse_pack_index(sha1);
784         new_pack->next = repo->packs;
785         repo->packs = new_pack;
786         return 0;
787 }
788
789 static void process_alternates(void *callback_data)
790 {
791         struct alt_request *alt_req = (struct alt_request *)callback_data;
792         struct active_request_slot *slot = alt_req->slot;
793         struct alt_base *tail = alt;
794         char *base = alt_req->base;
795         static const char null_byte = '\0';
796         char *data;
797         int i = 0;
798
799         if (alt_req->http_specific) {
800                 if (slot->curl_result != CURLE_OK ||
801                     !alt_req->buffer->posn) {
802
803                         /* Try reusing the slot to get non-http alternates */
804                         alt_req->http_specific = 0;
805                         sprintf(alt_req->url, "%s/objects/info/alternates",
806                                 base);
807                         curl_easy_setopt(slot->curl, CURLOPT_URL,
808                                          alt_req->url);
809                         active_requests++;
810                         slot->in_use = 1;
811                         slot->done = 0;
812                         if (start_active_slot(slot)) {
813                                 return;
814                         } else {
815                                 got_alternates = -1;
816                                 slot->done = 1;
817                                 return;
818                         }
819                 }
820         } else if (slot->curl_result != CURLE_OK) {
821                 if (slot->http_code != 404) {
822                         got_alternates = -1;
823                         return;
824                 }
825         }
826
827         fwrite_buffer_dynamic(&null_byte, 1, 1, alt_req->buffer);
828         alt_req->buffer->posn--;
829         data = alt_req->buffer->buffer;
830
831         while (i < alt_req->buffer->posn) {
832                 int posn = i;
833                 while (posn < alt_req->buffer->posn && data[posn] != '\n')
834                         posn++;
835                 if (data[posn] == '\n') {
836                         int okay = 0;
837                         int serverlen = 0;
838                         struct alt_base *newalt;
839                         char *target = NULL;
840                         if (data[i] == '/') {
841                                 serverlen = strchr(base + 8, '/') - base;
842                                 okay = 1;
843                         } else if (!memcmp(data + i, "../", 3)) {
844                                 i += 3;
845                                 serverlen = strlen(base);
846                                 while (i + 2 < posn && 
847                                        !memcmp(data + i, "../", 3)) {
848                                         do {
849                                                 serverlen--;
850                                         } while (serverlen &&
851                                                  base[serverlen - 1] != '/');
852                                         i += 3;
853                                 }
854                                 // If the server got removed, give up.
855                                 okay = strchr(base, ':') - base + 3 < 
856                                         serverlen;
857                         } else if (alt_req->http_specific) {
858                                 char *colon = strchr(data + i, ':');
859                                 char *slash = strchr(data + i, '/');
860                                 if (colon && slash && colon < data + posn &&
861                                     slash < data + posn && colon < slash) {
862                                         okay = 1;
863                                 }
864                         }
865                         // skip 'objects' at end
866                         if (okay) {
867                                 target = xmalloc(serverlen + posn - i - 6);
868                                 strncpy(target, base, serverlen);
869                                 strncpy(target + serverlen, data + i,
870                                         posn - i - 7);
871                                 target[serverlen + posn - i - 7] = '\0';
872                                 if (get_verbosely)
873                                         fprintf(stderr, 
874                                                 "Also look at %s\n", target);
875                                 newalt = xmalloc(sizeof(*newalt));
876                                 newalt->next = NULL;
877                                 newalt->base = target;
878                                 newalt->got_indices = 0;
879                                 newalt->packs = NULL;
880                                 while (tail->next != NULL)
881                                         tail = tail->next;
882                                 tail->next = newalt;
883                         }
884                 }
885                 i = posn + 1;
886         }
887
888         got_alternates = 1;
889 }
890
891 static void fetch_alternates(char *base)
892 {
893         struct buffer buffer;
894         char *url;
895         char *data;
896         struct active_request_slot *slot;
897         static struct alt_request alt_req;
898         int num_transfers;
899
900         /* If another request has already started fetching alternates,
901            wait for them to arrive and return to processing this request's
902            curl message */
903         while (got_alternates == 0) {
904                 curl_multi_perform(curlm, &num_transfers);
905                 process_curl_messages();
906                 process_request_queue();
907         }
908
909         /* Nothing to do if they've already been fetched */
910         if (got_alternates == 1)
911                 return;
912
913         /* Start the fetch */
914         got_alternates = 0;
915
916         data = xmalloc(4096);
917         buffer.size = 4096;
918         buffer.posn = 0;
919         buffer.buffer = data;
920
921         if (get_verbosely)
922                 fprintf(stderr, "Getting alternates list for %s\n", base);
923         
924         url = xmalloc(strlen(base) + 31);
925         sprintf(url, "%s/objects/info/http-alternates", base);
926
927         /* Use a callback to process the result, since another request
928            may fail and need to have alternates loaded before continuing */
929         slot = get_active_slot();
930         slot->callback_func = process_alternates;
931         slot->callback_data = &alt_req;
932
933         curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
934         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
935                          fwrite_buffer_dynamic);
936         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
937
938         alt_req.base = base;
939         alt_req.url = url;
940         alt_req.buffer = &buffer;
941         alt_req.http_specific = 1;
942         alt_req.slot = slot;
943
944         if (start_active_slot(slot))
945                 run_active_slot(slot);
946         else
947                 got_alternates = -1;
948
949         free(data);
950         free(url);
951 }
952
953 static int fetch_indices(struct alt_base *repo)
954 {
955         unsigned char sha1[20];
956         char *url;
957         struct buffer buffer;
958         char *data;
959         int i = 0;
960
961         struct active_request_slot *slot;
962
963         if (repo->got_indices)
964                 return 0;
965
966         data = xmalloc(4096);
967         buffer.size = 4096;
968         buffer.posn = 0;
969         buffer.buffer = data;
970
971         if (get_verbosely)
972                 fprintf(stderr, "Getting pack list for %s\n", repo->base);
973         
974         url = xmalloc(strlen(repo->base) + 21);
975         sprintf(url, "%s/objects/info/packs", repo->base);
976
977         slot = get_active_slot();
978         curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
979         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
980                          fwrite_buffer_dynamic);
981         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
982         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
983         if (start_active_slot(slot)) {
984                 run_active_slot(slot);
985                 if (slot->curl_result != CURLE_OK) {
986                         free(buffer.buffer);
987                         return error("%s", curl_errorstr);
988                 }
989         } else {
990                 free(buffer.buffer);
991                 return error("Unable to start request");
992         }
993
994         data = buffer.buffer;
995         while (i < buffer.posn) {
996                 switch (data[i]) {
997                 case 'P':
998                         i++;
999                         if (i + 52 < buffer.posn &&
1000                             !strncmp(data + i, " pack-", 6) &&
1001                             !strncmp(data + i + 46, ".pack\n", 6)) {
1002                                 get_sha1_hex(data + i + 6, sha1);
1003                                 setup_index(repo, sha1);
1004                                 i += 51;
1005                                 break;
1006                         }
1007                 default:
1008                         while (data[i] != '\n')
1009                                 i++;
1010                 }
1011                 i++;
1012         }
1013
1014         free(buffer.buffer);
1015         repo->got_indices = 1;
1016         return 0;
1017 }
1018
1019 static int fetch_pack(struct alt_base *repo, unsigned char *sha1)
1020 {
1021         char *url;
1022         struct packed_git *target;
1023         struct packed_git **lst;
1024         FILE *packfile;
1025         char *filename;
1026         char tmpfile[PATH_MAX];
1027         int ret;
1028         long prev_posn = 0;
1029         char range[RANGE_HEADER_SIZE];
1030         struct curl_slist *range_header = NULL;
1031
1032         struct active_request_slot *slot;
1033
1034         if (fetch_indices(repo))
1035                 return -1;
1036         target = find_sha1_pack(sha1, repo->packs);
1037         if (!target)
1038                 return -1;
1039
1040         if (get_verbosely) {
1041                 fprintf(stderr, "Getting pack %s\n",
1042                         sha1_to_hex(target->sha1));
1043                 fprintf(stderr, " which contains %s\n",
1044                         sha1_to_hex(sha1));
1045         }
1046
1047         url = xmalloc(strlen(repo->base) + 65);
1048         sprintf(url, "%s/objects/pack/pack-%s.pack",
1049                 repo->base, sha1_to_hex(target->sha1));
1050
1051         filename = sha1_pack_name(target->sha1);
1052         snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
1053         packfile = fopen(tmpfile, "a");
1054         if (!packfile)
1055                 return error("Unable to open local file %s for pack",
1056                              filename);
1057
1058         slot = get_active_slot();
1059         curl_easy_setopt(slot->curl, CURLOPT_FILE, packfile);
1060         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
1061         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
1062         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
1063         slot->local = packfile;
1064
1065         /* If there is data present from a previous transfer attempt,
1066            resume where it left off */
1067         prev_posn = ftell(packfile);
1068         if (prev_posn>0) {
1069                 if (get_verbosely)
1070                         fprintf(stderr,
1071                                 "Resuming fetch of pack %s at byte %ld\n",
1072                                 sha1_to_hex(target->sha1), prev_posn);
1073                 sprintf(range, "Range: bytes=%ld-", prev_posn);
1074                 range_header = curl_slist_append(range_header, range);
1075                 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
1076         }
1077
1078         if (start_active_slot(slot)) {
1079                 run_active_slot(slot);
1080                 if (slot->curl_result != CURLE_OK) {
1081                         fclose(packfile);
1082                         return error("Unable to get pack file %s\n%s", url,
1083                                      curl_errorstr);
1084                 }
1085         } else {
1086                 return error("Unable to start request");
1087         }
1088
1089         fclose(packfile);
1090
1091         ret = move_temp_to_file(tmpfile, filename);
1092         if (ret)
1093                 return ret;
1094
1095         lst = &repo->packs;
1096         while (*lst != target)
1097                 lst = &((*lst)->next);
1098         *lst = (*lst)->next;
1099
1100         if (verify_pack(target, 0))
1101                 return -1;
1102         install_packed_git(target);
1103
1104         return 0;
1105 }
1106
1107 static int fetch_object(struct alt_base *repo, unsigned char *sha1)
1108 {
1109         char *hex = sha1_to_hex(sha1);
1110         int ret;
1111         struct transfer_request *request = request_queue_head;
1112
1113         while (request != NULL && memcmp(request->sha1, sha1, 20))
1114                 request = request->next;
1115         if (request == NULL)
1116                 return error("Couldn't find request for %s in the queue", hex);
1117
1118         if (has_sha1_file(request->sha1)) {
1119                 release_request(request);
1120                 return 0;
1121         }
1122
1123 #ifdef USE_CURL_MULTI
1124         while (request->state == WAITING) {
1125                 int num_transfers;
1126                 curl_multi_perform(curlm, &num_transfers);
1127                 if (num_transfers < active_requests) {
1128                         process_curl_messages();
1129                         process_request_queue();
1130                 }
1131         }
1132 #else
1133         start_request(request);
1134 #endif
1135
1136         while (request->state == ACTIVE) {
1137                 run_active_slot(request->slot);
1138 #ifndef USE_CURL_MULTI
1139                 request->curl_result = request->slot->curl_result;
1140                 request->http_code = request->slot->http_code;
1141                 request->slot = NULL;
1142
1143                 /* Use alternates if necessary */
1144                 if (request->http_code == 404) {
1145                         fetch_alternates(alt->base);
1146                         if (request->repo->next != NULL) {
1147                                 request->repo = request->repo->next;
1148                                 start_request(request);
1149                         }
1150                 } else {
1151                         finish_request(request);
1152                         request->state = COMPLETE;
1153                 }
1154 #endif
1155         }
1156
1157         if (request->state == ABORTED) {
1158                 release_request(request);
1159                 return error("Request for %s aborted", hex);
1160         }
1161
1162         if (request->curl_result != CURLE_OK && request->http_code != 416) {
1163                 if (request->http_code == 404)
1164                         ret = -1; /* Be silent, it is probably in a pack. */
1165                 else
1166                         ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
1167                                     request->errorstr, request->curl_result,
1168                                     request->http_code, hex);
1169                 release_request(request);
1170                 return ret;
1171         }
1172
1173         if (request->zret != Z_STREAM_END) {
1174                 ret = error("File %s (%s) corrupt\n", hex, request->url);
1175                 release_request(request);
1176                 return ret;
1177         }
1178
1179         if (memcmp(request->sha1, request->real_sha1, 20)) {
1180                 release_request(request);
1181                 return error("File %s has bad hash\n", hex);
1182         }
1183
1184         if (request->rename < 0) {
1185                 ret = error("unable to write sha1 filename %s: %s",
1186                             request->filename,
1187                             strerror(request->rename));
1188                 release_request(request);
1189                 return ret;
1190         }
1191
1192         release_request(request);
1193         return 0;
1194 }
1195
1196 int fetch(unsigned char *sha1)
1197 {
1198         struct alt_base *altbase = alt;
1199
1200         if (!fetch_object(altbase, sha1))
1201                 return 0;
1202         while (altbase) {
1203                 if (!fetch_pack(altbase, sha1))
1204                         return 0;
1205                 fetch_alternates(alt->base);
1206                 altbase = altbase->next;
1207         }
1208         return error("Unable to find %s under %s\n", sha1_to_hex(sha1), 
1209                      alt->base);
1210 }
1211
1212 static inline int needs_quote(int ch)
1213 {
1214         switch (ch) {
1215         case '/': case '-': case '.':
1216         case 'A'...'Z': case 'a'...'z': case '0'...'9':
1217                 return 0;
1218         default:
1219                 return 1;
1220         }
1221 }
1222
1223 static inline int hex(int v)
1224 {
1225         if (v < 10) return '0' + v;
1226         else return 'A' + v - 10;
1227 }
1228
1229 static char *quote_ref_url(const char *base, const char *ref)
1230 {
1231         const char *cp;
1232         char *dp, *qref;
1233         int len, baselen, ch;
1234
1235         baselen = strlen(base);
1236         len = baselen + 6; /* "refs/" + NUL */
1237         for (cp = ref; (ch = *cp) != 0; cp++, len++)
1238                 if (needs_quote(ch))
1239                         len += 2; /* extra two hex plus replacement % */
1240         qref = xmalloc(len);
1241         memcpy(qref, base, baselen);
1242         memcpy(qref + baselen, "refs/", 5);
1243         for (cp = ref, dp = qref + baselen + 5; (ch = *cp) != 0; cp++) {
1244                 if (needs_quote(ch)) {
1245                         *dp++ = '%';
1246                         *dp++ = hex((ch >> 4) & 0xF);
1247                         *dp++ = hex(ch & 0xF);
1248                 }
1249                 else
1250                         *dp++ = ch;
1251         }
1252         *dp = 0;
1253
1254         return qref;
1255 }
1256
1257 int fetch_ref(char *ref, unsigned char *sha1)
1258 {
1259         char *url;
1260         char hex[42];
1261         struct buffer buffer;
1262         char *base = alt->base;
1263         struct active_request_slot *slot;
1264         buffer.size = 41;
1265         buffer.posn = 0;
1266         buffer.buffer = hex;
1267         hex[41] = '\0';
1268         
1269         url = quote_ref_url(base, ref);
1270         slot = get_active_slot();
1271         curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
1272         curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
1273         curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
1274         curl_easy_setopt(slot->curl, CURLOPT_URL, url);
1275         if (start_active_slot(slot)) {
1276                 run_active_slot(slot);
1277                 if (slot->curl_result != CURLE_OK)
1278                         return error("Couldn't get %s for %s\n%s",
1279                                      url, ref, curl_errorstr);
1280         } else {
1281                 return error("Unable to start request");
1282         }
1283
1284         hex[40] = '\0';
1285         get_sha1_hex(hex, sha1);
1286         return 0;
1287 }
1288
1289 int main(int argc, char **argv)
1290 {
1291         char *commit_id;
1292         char *url;
1293         int arg = 1;
1294         struct active_request_slot *slot;
1295         char *low_speed_limit;
1296         char *low_speed_time;
1297         char *wait_url;
1298         int rc = 0;
1299
1300         while (arg < argc && argv[arg][0] == '-') {
1301                 if (argv[arg][1] == 't') {
1302                         get_tree = 1;
1303                 } else if (argv[arg][1] == 'c') {
1304                         get_history = 1;
1305                 } else if (argv[arg][1] == 'a') {
1306                         get_all = 1;
1307                         get_tree = 1;
1308                         get_history = 1;
1309                 } else if (argv[arg][1] == 'v') {
1310                         get_verbosely = 1;
1311                 } else if (argv[arg][1] == 'w') {
1312                         write_ref = argv[arg + 1];
1313                         arg++;
1314                 } else if (!strcmp(argv[arg], "--recover")) {
1315                         get_recover = 1;
1316                 }
1317                 arg++;
1318         }
1319         if (argc < arg + 2) {
1320                 usage("git-http-fetch [-c] [-t] [-a] [-d] [-v] [--recover] [-w ref] commit-id url");
1321                 return 1;
1322         }
1323         commit_id = argv[arg];
1324         url = argv[arg + 1];
1325
1326         curl_global_init(CURL_GLOBAL_ALL);
1327
1328 #ifdef USE_CURL_MULTI
1329         {
1330                 char *http_max_requests = getenv("GIT_HTTP_MAX_REQUESTS");
1331                 if (http_max_requests != NULL)
1332                         max_requests = atoi(http_max_requests);
1333         }
1334
1335         curlm = curl_multi_init();
1336         if (curlm == NULL) {
1337                 fprintf(stderr, "Error creating curl multi handle.\n");
1338                 return 1;
1339         }
1340 #endif
1341
1342         if (getenv("GIT_SSL_NO_VERIFY"))
1343                 curl_ssl_verify = 0;
1344
1345         ssl_cert = getenv("GIT_SSL_CERT");
1346 #if LIBCURL_VERSION_NUM >= 0x070902
1347         ssl_key = getenv("GIT_SSL_KEY");
1348 #endif
1349 #if LIBCURL_VERSION_NUM >= 0x070908
1350         ssl_capath = getenv("GIT_SSL_CAPATH");
1351 #endif
1352         ssl_cainfo = getenv("GIT_SSL_CAINFO");
1353
1354         low_speed_limit = getenv("GIT_HTTP_LOW_SPEED_LIMIT");
1355         if (low_speed_limit != NULL)
1356                 curl_low_speed_limit = strtol(low_speed_limit, NULL, 10);
1357         low_speed_time = getenv("GIT_HTTP_LOW_SPEED_TIME");
1358         if (low_speed_time != NULL)
1359                 curl_low_speed_time = strtol(low_speed_time, NULL, 10);
1360
1361         git_config(http_options);
1362
1363         if (curl_ssl_verify == -1)
1364                 curl_ssl_verify = 1;
1365
1366 #ifdef USE_CURL_MULTI
1367         if (max_requests < 1)
1368                 max_requests = DEFAULT_MAX_REQUESTS;
1369 #endif
1370
1371         pragma_header = curl_slist_append(pragma_header, "Pragma: no-cache");
1372         no_pragma_header = curl_slist_append(no_pragma_header, "Pragma:");
1373         no_range_header = curl_slist_append(no_range_header, "Range:");
1374
1375 #ifndef NO_CURL_EASY_DUPHANDLE
1376         curl_default = get_curl_handle();
1377 #endif
1378
1379         alt = xmalloc(sizeof(*alt));
1380         alt->base = url;
1381         alt->got_indices = 0;
1382         alt->packs = NULL;
1383         alt->next = NULL;
1384
1385         if (pull(commit_id))
1386                 rc = 1;
1387
1388         curl_slist_free_all(pragma_header);
1389         curl_slist_free_all(no_pragma_header);
1390         curl_slist_free_all(no_range_header);
1391 #ifndef NO_CURL_EASY_DUPHANDLE
1392         curl_easy_cleanup(curl_default);
1393 #endif
1394         slot = active_queue_head;
1395         while (slot != NULL) {
1396                 if (slot->in_use) {
1397                         if (get_verbosely) {
1398                                 curl_easy_getinfo(slot->curl,
1399                                                   CURLINFO_EFFECTIVE_URL,
1400                                                   &wait_url);
1401                                 fprintf(stderr, "Waiting for %s\n", wait_url);
1402                         }
1403                         run_active_slot(slot);
1404                 }
1405                 if (slot->curl != NULL)
1406                         curl_easy_cleanup(slot->curl);
1407                 slot = slot->next;
1408         }
1409 #ifdef USE_CURL_MULTI
1410         curl_multi_cleanup(curlm);
1411 #endif
1412         curl_global_cleanup();
1413         return rc;
1414 }