Do linear-time/space rename logic for exact renames
[git/dscho.git] / http-walker.c
blob444aebf5268c982255f73b151ff5c03bdffecdbe
1 #include "cache.h"
2 #include "commit.h"
3 #include "pack.h"
4 #include "walker.h"
5 #include "http.h"
7 #define PREV_BUF_SIZE 4096
8 #define RANGE_HEADER_SIZE 30
10 struct alt_base
12 char *base;
13 int got_indices;
14 struct packed_git *packs;
15 struct alt_base *next;
18 enum object_request_state {
19 WAITING,
20 ABORTED,
21 ACTIVE,
22 COMPLETE,
25 struct object_request
27 struct walker *walker;
28 unsigned char sha1[20];
29 struct alt_base *repo;
30 char *url;
31 char filename[PATH_MAX];
32 char tmpfile[PATH_MAX];
33 int local;
34 enum object_request_state state;
35 CURLcode curl_result;
36 char errorstr[CURL_ERROR_SIZE];
37 long http_code;
38 unsigned char real_sha1[20];
39 SHA_CTX c;
40 z_stream stream;
41 int zret;
42 int rename;
43 struct active_request_slot *slot;
44 struct object_request *next;
47 struct alternates_request {
48 struct walker *walker;
49 const char *base;
50 char *url;
51 struct buffer *buffer;
52 struct active_request_slot *slot;
53 int http_specific;
56 struct walker_data {
57 const char *url;
58 int got_alternates;
59 struct alt_base *alt;
60 struct curl_slist *no_pragma_header;
63 static struct object_request *object_queue_head;
65 static size_t fwrite_sha1_file(void *ptr, size_t eltsize, size_t nmemb,
66 void *data)
68 unsigned char expn[4096];
69 size_t size = eltsize * nmemb;
70 int posn = 0;
71 struct object_request *obj_req = (struct object_request *)data;
72 do {
73 ssize_t retval = xwrite(obj_req->local,
74 (char *) ptr + posn, size - posn);
75 if (retval < 0)
76 return posn;
77 posn += retval;
78 } while (posn < size);
80 obj_req->stream.avail_in = size;
81 obj_req->stream.next_in = ptr;
82 do {
83 obj_req->stream.next_out = expn;
84 obj_req->stream.avail_out = sizeof(expn);
85 obj_req->zret = inflate(&obj_req->stream, Z_SYNC_FLUSH);
86 SHA1_Update(&obj_req->c, expn,
87 sizeof(expn) - obj_req->stream.avail_out);
88 } while (obj_req->stream.avail_in && obj_req->zret == Z_OK);
89 data_received++;
90 return size;
93 static int missing__target(int code, int result)
95 return /* file:// URL -- do we ever use one??? */
96 (result == CURLE_FILE_COULDNT_READ_FILE) ||
97 /* http:// and https:// URL */
98 (code == 404 && result == CURLE_HTTP_RETURNED_ERROR) ||
99 /* ftp:// URL */
100 (code == 550 && result == CURLE_FTP_COULDNT_RETR_FILE)
104 #define missing_target(a) missing__target((a)->http_code, (a)->curl_result)
106 static void fetch_alternates(struct walker *walker, const char *base);
108 static void process_object_response(void *callback_data);
110 static void start_object_request(struct walker *walker,
111 struct object_request *obj_req)
113 char *hex = sha1_to_hex(obj_req->sha1);
114 char prevfile[PATH_MAX];
115 char *url;
116 char *posn;
117 int prevlocal;
118 unsigned char prev_buf[PREV_BUF_SIZE];
119 ssize_t prev_read = 0;
120 long prev_posn = 0;
121 char range[RANGE_HEADER_SIZE];
122 struct curl_slist *range_header = NULL;
123 struct active_request_slot *slot;
124 struct walker_data *data = walker->data;
126 snprintf(prevfile, sizeof(prevfile), "%s.prev", obj_req->filename);
127 unlink(prevfile);
128 rename(obj_req->tmpfile, prevfile);
129 unlink(obj_req->tmpfile);
131 if (obj_req->local != -1)
132 error("fd leakage in start: %d", obj_req->local);
133 obj_req->local = open(obj_req->tmpfile,
134 O_WRONLY | O_CREAT | O_EXCL, 0666);
135 /* This could have failed due to the "lazy directory creation";
136 * try to mkdir the last path component.
138 if (obj_req->local < 0 && errno == ENOENT) {
139 char *dir = strrchr(obj_req->tmpfile, '/');
140 if (dir) {
141 *dir = 0;
142 mkdir(obj_req->tmpfile, 0777);
143 *dir = '/';
145 obj_req->local = open(obj_req->tmpfile,
146 O_WRONLY | O_CREAT | O_EXCL, 0666);
149 if (obj_req->local < 0) {
150 obj_req->state = ABORTED;
151 error("Couldn't create temporary file %s for %s: %s",
152 obj_req->tmpfile, obj_req->filename, strerror(errno));
153 return;
156 memset(&obj_req->stream, 0, sizeof(obj_req->stream));
158 inflateInit(&obj_req->stream);
160 SHA1_Init(&obj_req->c);
162 url = xmalloc(strlen(obj_req->repo->base) + 51);
163 obj_req->url = xmalloc(strlen(obj_req->repo->base) + 51);
164 strcpy(url, obj_req->repo->base);
165 posn = url + strlen(obj_req->repo->base);
166 strcpy(posn, "/objects/");
167 posn += 9;
168 memcpy(posn, hex, 2);
169 posn += 2;
170 *(posn++) = '/';
171 strcpy(posn, hex + 2);
172 strcpy(obj_req->url, url);
174 /* If a previous temp file is present, process what was already
175 fetched. */
176 prevlocal = open(prevfile, O_RDONLY);
177 if (prevlocal != -1) {
178 do {
179 prev_read = xread(prevlocal, prev_buf, PREV_BUF_SIZE);
180 if (prev_read>0) {
181 if (fwrite_sha1_file(prev_buf,
183 prev_read,
184 obj_req) == prev_read) {
185 prev_posn += prev_read;
186 } else {
187 prev_read = -1;
190 } while (prev_read > 0);
191 close(prevlocal);
193 unlink(prevfile);
195 /* Reset inflate/SHA1 if there was an error reading the previous temp
196 file; also rewind to the beginning of the local file. */
197 if (prev_read == -1) {
198 memset(&obj_req->stream, 0, sizeof(obj_req->stream));
199 inflateInit(&obj_req->stream);
200 SHA1_Init(&obj_req->c);
201 if (prev_posn>0) {
202 prev_posn = 0;
203 lseek(obj_req->local, 0, SEEK_SET);
204 ftruncate(obj_req->local, 0);
208 slot = get_active_slot();
209 slot->callback_func = process_object_response;
210 slot->callback_data = obj_req;
211 obj_req->slot = slot;
213 curl_easy_setopt(slot->curl, CURLOPT_FILE, obj_req);
214 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file);
215 curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, obj_req->errorstr);
216 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
217 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, data->no_pragma_header);
219 /* If we have successfully processed data from a previous fetch
220 attempt, only fetch the data we don't already have. */
221 if (prev_posn>0) {
222 if (walker->get_verbosely)
223 fprintf(stderr,
224 "Resuming fetch of object %s at byte %ld\n",
225 hex, prev_posn);
226 sprintf(range, "Range: bytes=%ld-", prev_posn);
227 range_header = curl_slist_append(range_header, range);
228 curl_easy_setopt(slot->curl,
229 CURLOPT_HTTPHEADER, range_header);
232 /* Try to get the request started, abort the request on error */
233 obj_req->state = ACTIVE;
234 if (!start_active_slot(slot)) {
235 obj_req->state = ABORTED;
236 obj_req->slot = NULL;
237 close(obj_req->local); obj_req->local = -1;
238 free(obj_req->url);
239 return;
243 static void finish_object_request(struct object_request *obj_req)
245 struct stat st;
247 fchmod(obj_req->local, 0444);
248 close(obj_req->local); obj_req->local = -1;
250 if (obj_req->http_code == 416) {
251 fprintf(stderr, "Warning: requested range invalid; we may already have all the data.\n");
252 } else if (obj_req->curl_result != CURLE_OK) {
253 if (stat(obj_req->tmpfile, &st) == 0)
254 if (st.st_size == 0)
255 unlink(obj_req->tmpfile);
256 return;
259 inflateEnd(&obj_req->stream);
260 SHA1_Final(obj_req->real_sha1, &obj_req->c);
261 if (obj_req->zret != Z_STREAM_END) {
262 unlink(obj_req->tmpfile);
263 return;
265 if (hashcmp(obj_req->sha1, obj_req->real_sha1)) {
266 unlink(obj_req->tmpfile);
267 return;
269 obj_req->rename =
270 move_temp_to_file(obj_req->tmpfile, obj_req->filename);
272 if (obj_req->rename == 0)
273 walker_say(obj_req->walker, "got %s\n", sha1_to_hex(obj_req->sha1));
276 static void process_object_response(void *callback_data)
278 struct object_request *obj_req =
279 (struct object_request *)callback_data;
280 struct walker *walker = obj_req->walker;
281 struct walker_data *data = walker->data;
282 struct alt_base *alt = data->alt;
284 obj_req->curl_result = obj_req->slot->curl_result;
285 obj_req->http_code = obj_req->slot->http_code;
286 obj_req->slot = NULL;
287 obj_req->state = COMPLETE;
289 /* Use alternates if necessary */
290 if (missing_target(obj_req)) {
291 fetch_alternates(walker, alt->base);
292 if (obj_req->repo->next != NULL) {
293 obj_req->repo =
294 obj_req->repo->next;
295 close(obj_req->local);
296 obj_req->local = -1;
297 start_object_request(walker, obj_req);
298 return;
302 finish_object_request(obj_req);
305 static void release_object_request(struct object_request *obj_req)
307 struct object_request *entry = object_queue_head;
309 if (obj_req->local != -1)
310 error("fd leakage in release: %d", obj_req->local);
311 if (obj_req == object_queue_head) {
312 object_queue_head = obj_req->next;
313 } else {
314 while (entry->next != NULL && entry->next != obj_req)
315 entry = entry->next;
316 if (entry->next == obj_req)
317 entry->next = entry->next->next;
320 free(obj_req->url);
321 free(obj_req);
324 #ifdef USE_CURL_MULTI
325 static int fill_active_slot(struct walker *walker)
327 struct object_request *obj_req;
329 for (obj_req = object_queue_head; obj_req; obj_req = obj_req->next) {
330 if (obj_req->state == WAITING) {
331 if (has_sha1_file(obj_req->sha1))
332 obj_req->state = COMPLETE;
333 else {
334 start_object_request(walker, obj_req);
335 return 1;
339 return 0;
341 #endif
343 static void prefetch(struct walker *walker, unsigned char *sha1)
345 struct object_request *newreq;
346 struct object_request *tail;
347 struct walker_data *data = walker->data;
348 char *filename = sha1_file_name(sha1);
350 newreq = xmalloc(sizeof(*newreq));
351 newreq->walker = walker;
352 hashcpy(newreq->sha1, sha1);
353 newreq->repo = data->alt;
354 newreq->url = NULL;
355 newreq->local = -1;
356 newreq->state = WAITING;
357 snprintf(newreq->filename, sizeof(newreq->filename), "%s", filename);
358 snprintf(newreq->tmpfile, sizeof(newreq->tmpfile),
359 "%s.temp", filename);
360 newreq->slot = NULL;
361 newreq->next = NULL;
363 if (object_queue_head == NULL) {
364 object_queue_head = newreq;
365 } else {
366 tail = object_queue_head;
367 while (tail->next != NULL) {
368 tail = tail->next;
370 tail->next = newreq;
373 #ifdef USE_CURL_MULTI
374 fill_active_slots();
375 step_active_slots();
376 #endif
379 static int fetch_index(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
381 char *hex = sha1_to_hex(sha1);
382 char *filename;
383 char *url;
384 char tmpfile[PATH_MAX];
385 long prev_posn = 0;
386 char range[RANGE_HEADER_SIZE];
387 struct curl_slist *range_header = NULL;
388 struct walker_data *data = walker->data;
390 FILE *indexfile;
391 struct active_request_slot *slot;
392 struct slot_results results;
394 if (has_pack_index(sha1))
395 return 0;
397 if (walker->get_verbosely)
398 fprintf(stderr, "Getting index for pack %s\n", hex);
400 url = xmalloc(strlen(repo->base) + 64);
401 sprintf(url, "%s/objects/pack/pack-%s.idx", repo->base, hex);
403 filename = sha1_pack_index_name(sha1);
404 snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
405 indexfile = fopen(tmpfile, "a");
406 if (!indexfile)
407 return error("Unable to open local file %s for pack index",
408 filename);
410 slot = get_active_slot();
411 slot->results = &results;
412 curl_easy_setopt(slot->curl, CURLOPT_FILE, indexfile);
413 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
414 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
415 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, data->no_pragma_header);
416 slot->local = indexfile;
418 /* If there is data present from a previous transfer attempt,
419 resume where it left off */
420 prev_posn = ftell(indexfile);
421 if (prev_posn>0) {
422 if (walker->get_verbosely)
423 fprintf(stderr,
424 "Resuming fetch of index for pack %s at byte %ld\n",
425 hex, prev_posn);
426 sprintf(range, "Range: bytes=%ld-", prev_posn);
427 range_header = curl_slist_append(range_header, range);
428 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
431 if (start_active_slot(slot)) {
432 run_active_slot(slot);
433 if (results.curl_result != CURLE_OK) {
434 fclose(indexfile);
435 return error("Unable to get pack index %s\n%s", url,
436 curl_errorstr);
438 } else {
439 fclose(indexfile);
440 return error("Unable to start request");
443 fclose(indexfile);
445 return move_temp_to_file(tmpfile, filename);
448 static int setup_index(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
450 struct packed_git *new_pack;
451 if (has_pack_file(sha1))
452 return 0; /* don't list this as something we can get */
454 if (fetch_index(walker, repo, sha1))
455 return -1;
457 new_pack = parse_pack_index(sha1);
458 new_pack->next = repo->packs;
459 repo->packs = new_pack;
460 return 0;
463 static void process_alternates_response(void *callback_data)
465 struct alternates_request *alt_req =
466 (struct alternates_request *)callback_data;
467 struct walker *walker = alt_req->walker;
468 struct walker_data *cdata = walker->data;
469 struct active_request_slot *slot = alt_req->slot;
470 struct alt_base *tail = cdata->alt;
471 const char *base = alt_req->base;
472 static const char null_byte = '\0';
473 char *data;
474 int i = 0;
476 if (alt_req->http_specific) {
477 if (slot->curl_result != CURLE_OK ||
478 !alt_req->buffer->posn) {
480 /* Try reusing the slot to get non-http alternates */
481 alt_req->http_specific = 0;
482 sprintf(alt_req->url, "%s/objects/info/alternates",
483 base);
484 curl_easy_setopt(slot->curl, CURLOPT_URL,
485 alt_req->url);
486 active_requests++;
487 slot->in_use = 1;
488 if (slot->finished != NULL)
489 (*slot->finished) = 0;
490 if (!start_active_slot(slot)) {
491 cdata->got_alternates = -1;
492 slot->in_use = 0;
493 if (slot->finished != NULL)
494 (*slot->finished) = 1;
496 return;
498 } else if (slot->curl_result != CURLE_OK) {
499 if (!missing_target(slot)) {
500 cdata->got_alternates = -1;
501 return;
505 fwrite_buffer(&null_byte, 1, 1, alt_req->buffer);
506 alt_req->buffer->posn--;
507 data = alt_req->buffer->buffer;
509 while (i < alt_req->buffer->posn) {
510 int posn = i;
511 while (posn < alt_req->buffer->posn && data[posn] != '\n')
512 posn++;
513 if (data[posn] == '\n') {
514 int okay = 0;
515 int serverlen = 0;
516 struct alt_base *newalt;
517 char *target = NULL;
518 if (data[i] == '/') {
519 /* This counts
520 * http://git.host/pub/scm/linux.git/
521 * -----------here^
522 * so memcpy(dst, base, serverlen) will
523 * copy up to "...git.host".
525 const char *colon_ss = strstr(base,"://");
526 if (colon_ss) {
527 serverlen = (strchr(colon_ss + 3, '/')
528 - base);
529 okay = 1;
531 } else if (!memcmp(data + i, "../", 3)) {
532 /* Relative URL; chop the corresponding
533 * number of subpath from base (and ../
534 * from data), and concatenate the result.
536 * The code first drops ../ from data, and
537 * then drops one ../ from data and one path
538 * from base. IOW, one extra ../ is dropped
539 * from data than path is dropped from base.
541 * This is not wrong. The alternate in
542 * http://git.host/pub/scm/linux.git/
543 * to borrow from
544 * http://git.host/pub/scm/linus.git/
545 * is ../../linus.git/objects/. You need
546 * two ../../ to borrow from your direct
547 * neighbour.
549 i += 3;
550 serverlen = strlen(base);
551 while (i + 2 < posn &&
552 !memcmp(data + i, "../", 3)) {
553 do {
554 serverlen--;
555 } while (serverlen &&
556 base[serverlen - 1] != '/');
557 i += 3;
559 /* If the server got removed, give up. */
560 okay = strchr(base, ':') - base + 3 <
561 serverlen;
562 } else if (alt_req->http_specific) {
563 char *colon = strchr(data + i, ':');
564 char *slash = strchr(data + i, '/');
565 if (colon && slash && colon < data + posn &&
566 slash < data + posn && colon < slash) {
567 okay = 1;
570 /* skip "objects\n" at end */
571 if (okay) {
572 target = xmalloc(serverlen + posn - i - 6);
573 memcpy(target, base, serverlen);
574 memcpy(target + serverlen, data + i,
575 posn - i - 7);
576 target[serverlen + posn - i - 7] = 0;
577 if (walker->get_verbosely)
578 fprintf(stderr,
579 "Also look at %s\n", target);
580 newalt = xmalloc(sizeof(*newalt));
581 newalt->next = NULL;
582 newalt->base = target;
583 newalt->got_indices = 0;
584 newalt->packs = NULL;
586 while (tail->next != NULL)
587 tail = tail->next;
588 tail->next = newalt;
591 i = posn + 1;
594 cdata->got_alternates = 1;
597 static void fetch_alternates(struct walker *walker, const char *base)
599 struct buffer buffer;
600 char *url;
601 char *data;
602 struct active_request_slot *slot;
603 struct alternates_request alt_req;
604 struct walker_data *cdata = walker->data;
606 /* If another request has already started fetching alternates,
607 wait for them to arrive and return to processing this request's
608 curl message */
609 #ifdef USE_CURL_MULTI
610 while (cdata->got_alternates == 0) {
611 step_active_slots();
613 #endif
615 /* Nothing to do if they've already been fetched */
616 if (cdata->got_alternates == 1)
617 return;
619 /* Start the fetch */
620 cdata->got_alternates = 0;
622 data = xmalloc(4096);
623 buffer.size = 4096;
624 buffer.posn = 0;
625 buffer.buffer = data;
627 if (walker->get_verbosely)
628 fprintf(stderr, "Getting alternates list for %s\n", base);
630 url = xmalloc(strlen(base) + 31);
631 sprintf(url, "%s/objects/info/http-alternates", base);
633 /* Use a callback to process the result, since another request
634 may fail and need to have alternates loaded before continuing */
635 slot = get_active_slot();
636 slot->callback_func = process_alternates_response;
637 alt_req.walker = walker;
638 slot->callback_data = &alt_req;
640 curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
641 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
642 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
644 alt_req.base = base;
645 alt_req.url = url;
646 alt_req.buffer = &buffer;
647 alt_req.http_specific = 1;
648 alt_req.slot = slot;
650 if (start_active_slot(slot))
651 run_active_slot(slot);
652 else
653 cdata->got_alternates = -1;
655 free(data);
656 free(url);
659 static int fetch_indices(struct walker *walker, struct alt_base *repo)
661 unsigned char sha1[20];
662 char *url;
663 struct buffer buffer;
664 char *data;
665 int i = 0;
667 struct active_request_slot *slot;
668 struct slot_results results;
670 if (repo->got_indices)
671 return 0;
673 data = xmalloc(4096);
674 buffer.size = 4096;
675 buffer.posn = 0;
676 buffer.buffer = data;
678 if (walker->get_verbosely)
679 fprintf(stderr, "Getting pack list for %s\n", repo->base);
681 url = xmalloc(strlen(repo->base) + 21);
682 sprintf(url, "%s/objects/info/packs", repo->base);
684 slot = get_active_slot();
685 slot->results = &results;
686 curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
687 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
688 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
689 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
690 if (start_active_slot(slot)) {
691 run_active_slot(slot);
692 if (results.curl_result != CURLE_OK) {
693 if (missing_target(&results)) {
694 repo->got_indices = 1;
695 free(buffer.buffer);
696 return 0;
697 } else {
698 repo->got_indices = 0;
699 free(buffer.buffer);
700 return error("%s", curl_errorstr);
703 } else {
704 repo->got_indices = 0;
705 free(buffer.buffer);
706 return error("Unable to start request");
709 data = buffer.buffer;
710 while (i < buffer.posn) {
711 switch (data[i]) {
712 case 'P':
713 i++;
714 if (i + 52 <= buffer.posn &&
715 !prefixcmp(data + i, " pack-") &&
716 !prefixcmp(data + i + 46, ".pack\n")) {
717 get_sha1_hex(data + i + 6, sha1);
718 setup_index(walker, repo, sha1);
719 i += 51;
720 break;
722 default:
723 while (i < buffer.posn && data[i] != '\n')
724 i++;
726 i++;
729 free(buffer.buffer);
730 repo->got_indices = 1;
731 return 0;
734 static int fetch_pack(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
736 char *url;
737 struct packed_git *target;
738 struct packed_git **lst;
739 FILE *packfile;
740 char *filename;
741 char tmpfile[PATH_MAX];
742 int ret;
743 long prev_posn = 0;
744 char range[RANGE_HEADER_SIZE];
745 struct curl_slist *range_header = NULL;
746 struct walker_data *data = walker->data;
748 struct active_request_slot *slot;
749 struct slot_results results;
751 if (fetch_indices(walker, repo))
752 return -1;
753 target = find_sha1_pack(sha1, repo->packs);
754 if (!target)
755 return -1;
757 if (walker->get_verbosely) {
758 fprintf(stderr, "Getting pack %s\n",
759 sha1_to_hex(target->sha1));
760 fprintf(stderr, " which contains %s\n",
761 sha1_to_hex(sha1));
764 url = xmalloc(strlen(repo->base) + 65);
765 sprintf(url, "%s/objects/pack/pack-%s.pack",
766 repo->base, sha1_to_hex(target->sha1));
768 filename = sha1_pack_name(target->sha1);
769 snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
770 packfile = fopen(tmpfile, "a");
771 if (!packfile)
772 return error("Unable to open local file %s for pack",
773 filename);
775 slot = get_active_slot();
776 slot->results = &results;
777 curl_easy_setopt(slot->curl, CURLOPT_FILE, packfile);
778 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
779 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
780 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, data->no_pragma_header);
781 slot->local = packfile;
783 /* If there is data present from a previous transfer attempt,
784 resume where it left off */
785 prev_posn = ftell(packfile);
786 if (prev_posn>0) {
787 if (walker->get_verbosely)
788 fprintf(stderr,
789 "Resuming fetch of pack %s at byte %ld\n",
790 sha1_to_hex(target->sha1), prev_posn);
791 sprintf(range, "Range: bytes=%ld-", prev_posn);
792 range_header = curl_slist_append(range_header, range);
793 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
796 if (start_active_slot(slot)) {
797 run_active_slot(slot);
798 if (results.curl_result != CURLE_OK) {
799 fclose(packfile);
800 return error("Unable to get pack file %s\n%s", url,
801 curl_errorstr);
803 } else {
804 fclose(packfile);
805 return error("Unable to start request");
808 target->pack_size = ftell(packfile);
809 fclose(packfile);
811 ret = move_temp_to_file(tmpfile, filename);
812 if (ret)
813 return ret;
815 lst = &repo->packs;
816 while (*lst != target)
817 lst = &((*lst)->next);
818 *lst = (*lst)->next;
820 if (verify_pack(target, 0))
821 return -1;
822 install_packed_git(target);
824 return 0;
827 static void abort_object_request(struct object_request *obj_req)
829 if (obj_req->local >= 0) {
830 close(obj_req->local);
831 obj_req->local = -1;
833 unlink(obj_req->tmpfile);
834 if (obj_req->slot) {
835 release_active_slot(obj_req->slot);
836 obj_req->slot = NULL;
838 release_object_request(obj_req);
841 static int fetch_object(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
843 char *hex = sha1_to_hex(sha1);
844 int ret = 0;
845 struct object_request *obj_req = object_queue_head;
847 while (obj_req != NULL && hashcmp(obj_req->sha1, sha1))
848 obj_req = obj_req->next;
849 if (obj_req == NULL)
850 return error("Couldn't find request for %s in the queue", hex);
852 if (has_sha1_file(obj_req->sha1)) {
853 abort_object_request(obj_req);
854 return 0;
857 #ifdef USE_CURL_MULTI
858 while (obj_req->state == WAITING) {
859 step_active_slots();
861 #else
862 start_object_request(walker, obj_req);
863 #endif
865 while (obj_req->state == ACTIVE) {
866 run_active_slot(obj_req->slot);
868 if (obj_req->local != -1) {
869 close(obj_req->local); obj_req->local = -1;
872 if (obj_req->state == ABORTED) {
873 ret = error("Request for %s aborted", hex);
874 } else if (obj_req->curl_result != CURLE_OK &&
875 obj_req->http_code != 416) {
876 if (missing_target(obj_req))
877 ret = -1; /* Be silent, it is probably in a pack. */
878 else
879 ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
880 obj_req->errorstr, obj_req->curl_result,
881 obj_req->http_code, hex);
882 } else if (obj_req->zret != Z_STREAM_END) {
883 walker->corrupt_object_found++;
884 ret = error("File %s (%s) corrupt", hex, obj_req->url);
885 } else if (hashcmp(obj_req->sha1, obj_req->real_sha1)) {
886 ret = error("File %s has bad hash", hex);
887 } else if (obj_req->rename < 0) {
888 ret = error("unable to write sha1 filename %s",
889 obj_req->filename);
892 release_object_request(obj_req);
893 return ret;
896 static int fetch(struct walker *walker, unsigned char *sha1)
898 struct walker_data *data = walker->data;
899 struct alt_base *altbase = data->alt;
901 if (!fetch_object(walker, altbase, sha1))
902 return 0;
903 while (altbase) {
904 if (!fetch_pack(walker, altbase, sha1))
905 return 0;
906 fetch_alternates(walker, data->alt->base);
907 altbase = altbase->next;
909 return error("Unable to find %s under %s", sha1_to_hex(sha1),
910 data->alt->base);
913 static inline int needs_quote(int ch)
915 if (((ch >= 'A') && (ch <= 'Z'))
916 || ((ch >= 'a') && (ch <= 'z'))
917 || ((ch >= '0') && (ch <= '9'))
918 || (ch == '/')
919 || (ch == '-')
920 || (ch == '.'))
921 return 0;
922 return 1;
925 static inline int hex(int v)
927 if (v < 10) return '0' + v;
928 else return 'A' + v - 10;
931 static char *quote_ref_url(const char *base, const char *ref)
933 const char *cp;
934 char *dp, *qref;
935 int len, baselen, ch;
937 baselen = strlen(base);
938 len = baselen + 7; /* "/refs/" + NUL */
939 for (cp = ref; (ch = *cp) != 0; cp++, len++)
940 if (needs_quote(ch))
941 len += 2; /* extra two hex plus replacement % */
942 qref = xmalloc(len);
943 memcpy(qref, base, baselen);
944 memcpy(qref + baselen, "/refs/", 6);
945 for (cp = ref, dp = qref + baselen + 6; (ch = *cp) != 0; cp++) {
946 if (needs_quote(ch)) {
947 *dp++ = '%';
948 *dp++ = hex((ch >> 4) & 0xF);
949 *dp++ = hex(ch & 0xF);
951 else
952 *dp++ = ch;
954 *dp = 0;
956 return qref;
959 static int fetch_ref(struct walker *walker, char *ref, unsigned char *sha1)
961 char *url;
962 char hex[42];
963 struct buffer buffer;
964 struct walker_data *data = walker->data;
965 const char *base = data->alt->base;
966 struct active_request_slot *slot;
967 struct slot_results results;
968 buffer.size = 41;
969 buffer.posn = 0;
970 buffer.buffer = hex;
971 hex[41] = '\0';
973 url = quote_ref_url(base, ref);
974 slot = get_active_slot();
975 slot->results = &results;
976 curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
977 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
978 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
979 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
980 if (start_active_slot(slot)) {
981 run_active_slot(slot);
982 if (results.curl_result != CURLE_OK)
983 return error("Couldn't get %s for %s\n%s",
984 url, ref, curl_errorstr);
985 } else {
986 return error("Unable to start request");
989 hex[40] = '\0';
990 get_sha1_hex(hex, sha1);
991 return 0;
994 static void cleanup(struct walker *walker)
996 struct walker_data *data = walker->data;
997 http_cleanup();
999 curl_slist_free_all(data->no_pragma_header);
1002 struct walker *get_http_walker(const char *url)
1004 char *s;
1005 struct walker_data *data = xmalloc(sizeof(struct walker_data));
1006 struct walker *walker = xmalloc(sizeof(struct walker));
1008 http_init();
1010 data->no_pragma_header = curl_slist_append(NULL, "Pragma:");
1012 data->alt = xmalloc(sizeof(*data->alt));
1013 data->alt->base = xmalloc(strlen(url) + 1);
1014 strcpy(data->alt->base, url);
1015 for (s = data->alt->base + strlen(data->alt->base) - 1; *s == '/'; --s)
1016 *s = 0;
1018 data->alt->got_indices = 0;
1019 data->alt->packs = NULL;
1020 data->alt->next = NULL;
1021 data->got_alternates = -1;
1023 walker->corrupt_object_found = 0;
1024 walker->fetch = fetch;
1025 walker->fetch_ref = fetch_ref;
1026 walker->prefetch = prefetch;
1027 walker->cleanup = cleanup;
1028 walker->data = data;
1030 #ifdef USE_CURL_MULTI
1031 add_fill_function(walker, (int (*)(void *)) fill_active_slot);
1032 #endif
1034 return walker;