7 #define PREV_BUF_SIZE 4096
8 #define RANGE_HEADER_SIZE 30
14 struct packed_git *packs;
15 struct alt_base *next;
18 enum object_request_state {
27 struct walker *walker;
28 unsigned char sha1[20];
29 struct alt_base *repo;
31 char filename[PATH_MAX];
32 char tmpfile[PATH_MAX];
34 enum object_request_state state;
36 char errorstr[CURL_ERROR_SIZE];
38 unsigned char real_sha1[20];
43 struct active_request_slot *slot;
44 struct object_request *next;
47 struct alternates_request {
48 struct walker *walker;
51 struct strbuf *buffer;
52 struct active_request_slot *slot;
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,
68 unsigned char expn[4096];
69 size_t size = eltsize * nmemb;
71 struct object_request *obj_req = (struct object_request *)data;
73 ssize_t retval = xwrite(obj_req->local,
74 (char *) ptr + posn, size - posn);
78 } while (posn < size);
80 obj_req->stream.avail_in = size;
81 obj_req->stream.next_in = ptr;
83 obj_req->stream.next_out = expn;
84 obj_req->stream.avail_out = sizeof(expn);
85 obj_req->zret = git_inflate(&obj_req->stream, Z_SYNC_FLUSH);
86 git_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);
93 static void fetch_alternates(struct walker *walker, const char *base);
95 static void process_object_response(void *callback_data);
97 static void start_object_request(struct walker *walker,
98 struct object_request *obj_req)
100 char *hex = sha1_to_hex(obj_req->sha1);
101 char prevfile[PATH_MAX];
105 unsigned char prev_buf[PREV_BUF_SIZE];
106 ssize_t prev_read = 0;
108 char range[RANGE_HEADER_SIZE];
109 struct curl_slist *range_header = NULL;
110 struct active_request_slot *slot;
111 struct walker_data *data = walker->data;
113 snprintf(prevfile, sizeof(prevfile), "%s.prev", obj_req->filename);
114 unlink_or_warn(prevfile);
115 rename(obj_req->tmpfile, prevfile);
116 unlink_or_warn(obj_req->tmpfile);
118 if (obj_req->local != -1)
119 error("fd leakage in start: %d", obj_req->local);
120 obj_req->local = open(obj_req->tmpfile,
121 O_WRONLY | O_CREAT | O_EXCL, 0666);
123 * This could have failed due to the "lazy directory creation";
124 * try to mkdir the last path component.
126 if (obj_req->local < 0 && errno == ENOENT) {
127 char *dir = strrchr(obj_req->tmpfile, '/');
130 mkdir(obj_req->tmpfile, 0777);
133 obj_req->local = open(obj_req->tmpfile,
134 O_WRONLY | O_CREAT | O_EXCL, 0666);
137 if (obj_req->local < 0) {
138 obj_req->state = ABORTED;
139 error("Couldn't create temporary file %s for %s: %s",
140 obj_req->tmpfile, obj_req->filename, strerror(errno));
144 memset(&obj_req->stream, 0, sizeof(obj_req->stream));
146 git_inflate_init(&obj_req->stream);
148 git_SHA1_Init(&obj_req->c);
150 url = xmalloc(strlen(obj_req->repo->base) + 51);
151 obj_req->url = xmalloc(strlen(obj_req->repo->base) + 51);
152 strcpy(url, obj_req->repo->base);
153 posn = url + strlen(obj_req->repo->base);
154 strcpy(posn, "/objects/");
156 memcpy(posn, hex, 2);
159 strcpy(posn, hex + 2);
160 strcpy(obj_req->url, url);
163 * If a previous temp file is present, process what was already
166 prevlocal = open(prevfile, O_RDONLY);
167 if (prevlocal != -1) {
169 prev_read = xread(prevlocal, prev_buf, PREV_BUF_SIZE);
171 if (fwrite_sha1_file(prev_buf,
174 obj_req) == prev_read)
175 prev_posn += prev_read;
179 } while (prev_read > 0);
182 unlink_or_warn(prevfile);
185 * Reset inflate/SHA1 if there was an error reading the previous temp
186 * file; also rewind to the beginning of the local file.
188 if (prev_read == -1) {
189 memset(&obj_req->stream, 0, sizeof(obj_req->stream));
190 git_inflate_init(&obj_req->stream);
191 git_SHA1_Init(&obj_req->c);
194 lseek(obj_req->local, 0, SEEK_SET);
195 ftruncate(obj_req->local, 0);
199 slot = get_active_slot();
200 slot->callback_func = process_object_response;
201 slot->callback_data = obj_req;
202 obj_req->slot = slot;
204 curl_easy_setopt(slot->curl, CURLOPT_FILE, obj_req);
205 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file);
206 curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, obj_req->errorstr);
207 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
208 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, data->no_pragma_header);
211 * If we have successfully processed data from a previous fetch
212 * attempt, only fetch the data we don't already have.
215 if (walker->get_verbosely)
217 "Resuming fetch of object %s at byte %ld\n",
219 sprintf(range, "Range: bytes=%ld-", prev_posn);
220 range_header = curl_slist_append(range_header, range);
221 curl_easy_setopt(slot->curl,
222 CURLOPT_HTTPHEADER, range_header);
225 /* Try to get the request started, abort the request on error */
226 obj_req->state = ACTIVE;
227 if (!start_active_slot(slot)) {
228 obj_req->state = ABORTED;
229 obj_req->slot = NULL;
230 close(obj_req->local);
237 static void finish_object_request(struct object_request *obj_req)
241 close(obj_req->local);
244 if (obj_req->http_code == 416) {
245 fprintf(stderr, "Warning: requested range invalid; we may already have all the data.\n");
246 } else if (obj_req->curl_result != CURLE_OK) {
247 if (stat(obj_req->tmpfile, &st) == 0)
249 unlink_or_warn(obj_req->tmpfile);
253 git_inflate_end(&obj_req->stream);
254 git_SHA1_Final(obj_req->real_sha1, &obj_req->c);
255 if (obj_req->zret != Z_STREAM_END) {
256 unlink_or_warn(obj_req->tmpfile);
259 if (hashcmp(obj_req->sha1, obj_req->real_sha1)) {
260 unlink_or_warn(obj_req->tmpfile);
264 move_temp_to_file(obj_req->tmpfile, obj_req->filename);
266 if (obj_req->rename == 0)
267 walker_say(obj_req->walker, "got %s\n", sha1_to_hex(obj_req->sha1));
270 static void process_object_response(void *callback_data)
272 struct object_request *obj_req =
273 (struct object_request *)callback_data;
274 struct walker *walker = obj_req->walker;
275 struct walker_data *data = walker->data;
276 struct alt_base *alt = data->alt;
278 obj_req->curl_result = obj_req->slot->curl_result;
279 obj_req->http_code = obj_req->slot->http_code;
280 obj_req->slot = NULL;
281 obj_req->state = COMPLETE;
283 /* Use alternates if necessary */
284 if (missing_target(obj_req)) {
285 fetch_alternates(walker, alt->base);
286 if (obj_req->repo->next != NULL) {
289 close(obj_req->local);
291 start_object_request(walker, obj_req);
296 finish_object_request(obj_req);
299 static void release_object_request(struct object_request *obj_req)
301 struct object_request *entry = object_queue_head;
303 if (obj_req->local != -1)
304 error("fd leakage in release: %d", obj_req->local);
305 if (obj_req == object_queue_head) {
306 object_queue_head = obj_req->next;
308 while (entry->next != NULL && entry->next != obj_req)
310 if (entry->next == obj_req)
311 entry->next = entry->next->next;
318 #ifdef USE_CURL_MULTI
319 static int fill_active_slot(struct walker *walker)
321 struct object_request *obj_req;
323 for (obj_req = object_queue_head; obj_req; obj_req = obj_req->next) {
324 if (obj_req->state == WAITING) {
325 if (has_sha1_file(obj_req->sha1))
326 obj_req->state = COMPLETE;
328 start_object_request(walker, obj_req);
337 static void prefetch(struct walker *walker, unsigned char *sha1)
339 struct object_request *newreq;
340 struct object_request *tail;
341 struct walker_data *data = walker->data;
342 char *filename = sha1_file_name(sha1);
344 newreq = xmalloc(sizeof(*newreq));
345 newreq->walker = walker;
346 hashcpy(newreq->sha1, sha1);
347 newreq->repo = data->alt;
350 newreq->state = WAITING;
351 snprintf(newreq->filename, sizeof(newreq->filename), "%s", filename);
352 snprintf(newreq->tmpfile, sizeof(newreq->tmpfile),
353 "%s.temp", filename);
357 if (object_queue_head == NULL) {
358 object_queue_head = newreq;
360 tail = object_queue_head;
361 while (tail->next != NULL)
366 #ifdef USE_CURL_MULTI
372 static int fetch_index(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
374 char *hex = sha1_to_hex(sha1);
377 char tmpfile[PATH_MAX];
379 char range[RANGE_HEADER_SIZE];
380 struct curl_slist *range_header = NULL;
381 struct walker_data *data = walker->data;
384 struct active_request_slot *slot;
385 struct slot_results results;
387 if (has_pack_index(sha1))
390 if (walker->get_verbosely)
391 fprintf(stderr, "Getting index for pack %s\n", hex);
393 url = xmalloc(strlen(repo->base) + 64);
394 sprintf(url, "%s/objects/pack/pack-%s.idx", repo->base, hex);
396 filename = sha1_pack_index_name(sha1);
397 snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
398 indexfile = fopen(tmpfile, "a");
400 return error("Unable to open local file %s for pack index",
403 slot = get_active_slot();
404 slot->results = &results;
405 curl_easy_setopt(slot->curl, CURLOPT_FILE, indexfile);
406 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
407 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
408 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, data->no_pragma_header);
409 slot->local = indexfile;
412 * If there is data present from a previous transfer attempt,
413 * resume where it left off
415 prev_posn = ftell(indexfile);
417 if (walker->get_verbosely)
419 "Resuming fetch of index for pack %s at byte %ld\n",
421 sprintf(range, "Range: bytes=%ld-", prev_posn);
422 range_header = curl_slist_append(range_header, range);
423 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
426 if (start_active_slot(slot)) {
427 run_active_slot(slot);
428 if (results.curl_result != CURLE_OK) {
431 return error("Unable to get pack index %s\n%s", url,
437 return error("Unable to start request");
443 return move_temp_to_file(tmpfile, filename);
446 static int setup_index(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
448 struct packed_git *new_pack;
449 if (has_pack_file(sha1))
450 return 0; /* don't list this as something we can get */
452 if (fetch_index(walker, repo, sha1))
455 new_pack = parse_pack_index(sha1);
457 return -1; /* parse_pack_index() already issued error message */
458 new_pack->next = repo->packs;
459 repo->packs = new_pack;
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';
476 if (alt_req->http_specific) {
477 if (slot->curl_result != CURLE_OK ||
478 !alt_req->buffer->len) {
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",
484 curl_easy_setopt(slot->curl, CURLOPT_URL,
488 if (slot->finished != NULL)
489 (*slot->finished) = 0;
490 if (!start_active_slot(slot)) {
491 cdata->got_alternates = -1;
493 if (slot->finished != NULL)
494 (*slot->finished) = 1;
498 } else if (slot->curl_result != CURLE_OK) {
499 if (!missing_target(slot)) {
500 cdata->got_alternates = -1;
505 fwrite_buffer(&null_byte, 1, 1, alt_req->buffer);
506 alt_req->buffer->len--;
507 data = alt_req->buffer->buf;
509 while (i < alt_req->buffer->len) {
511 while (posn < alt_req->buffer->len && data[posn] != '\n')
513 if (data[posn] == '\n') {
516 struct alt_base *newalt;
518 if (data[i] == '/') {
521 * http://git.host/pub/scm/linux.git/
523 * so memcpy(dst, base, serverlen) will
524 * copy up to "...git.host".
526 const char *colon_ss = strstr(base,"://");
528 serverlen = (strchr(colon_ss + 3, '/')
532 } else if (!memcmp(data + i, "../", 3)) {
534 * Relative URL; chop the corresponding
535 * number of subpath from base (and ../
536 * from data), and concatenate the result.
538 * The code first drops ../ from data, and
539 * then drops one ../ from data and one path
540 * from base. IOW, one extra ../ is dropped
541 * from data than path is dropped from base.
543 * This is not wrong. The alternate in
544 * http://git.host/pub/scm/linux.git/
546 * http://git.host/pub/scm/linus.git/
547 * is ../../linus.git/objects/. You need
548 * two ../../ to borrow from your direct
552 serverlen = strlen(base);
553 while (i + 2 < posn &&
554 !memcmp(data + i, "../", 3)) {
557 } while (serverlen &&
558 base[serverlen - 1] != '/');
561 /* If the server got removed, give up. */
562 okay = strchr(base, ':') - base + 3 <
564 } else if (alt_req->http_specific) {
565 char *colon = strchr(data + i, ':');
566 char *slash = strchr(data + i, '/');
567 if (colon && slash && colon < data + posn &&
568 slash < data + posn && colon < slash) {
572 /* skip "objects\n" at end */
574 target = xmalloc(serverlen + posn - i - 6);
575 memcpy(target, base, serverlen);
576 memcpy(target + serverlen, data + i,
578 target[serverlen + posn - i - 7] = 0;
579 if (walker->get_verbosely)
581 "Also look at %s\n", target);
582 newalt = xmalloc(sizeof(*newalt));
584 newalt->base = target;
585 newalt->got_indices = 0;
586 newalt->packs = NULL;
588 while (tail->next != NULL)
596 cdata->got_alternates = 1;
599 static void fetch_alternates(struct walker *walker, const char *base)
601 struct strbuf buffer = STRBUF_INIT;
603 struct active_request_slot *slot;
604 struct alternates_request alt_req;
605 struct walker_data *cdata = walker->data;
608 * If another request has already started fetching alternates,
609 * wait for them to arrive and return to processing this request's
612 #ifdef USE_CURL_MULTI
613 while (cdata->got_alternates == 0) {
618 /* Nothing to do if they've already been fetched */
619 if (cdata->got_alternates == 1)
622 /* Start the fetch */
623 cdata->got_alternates = 0;
625 if (walker->get_verbosely)
626 fprintf(stderr, "Getting alternates list for %s\n", base);
628 url = xmalloc(strlen(base) + 31);
629 sprintf(url, "%s/objects/info/http-alternates", base);
632 * Use a callback to process the result, since another request
633 * 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);
646 alt_req.buffer = &buffer;
647 alt_req.http_specific = 1;
650 if (start_active_slot(slot))
651 run_active_slot(slot);
653 cdata->got_alternates = -1;
655 strbuf_release(&buffer);
659 static int fetch_indices(struct walker *walker, struct alt_base *repo)
661 unsigned char sha1[20];
663 struct strbuf buffer = STRBUF_INIT;
668 struct active_request_slot *slot;
669 struct slot_results results;
671 if (repo->got_indices)
674 if (walker->get_verbosely)
675 fprintf(stderr, "Getting pack list for %s\n", repo->base);
677 url = xmalloc(strlen(repo->base) + 21);
678 sprintf(url, "%s/objects/info/packs", repo->base);
680 slot = get_active_slot();
681 slot->results = &results;
682 curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
683 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
684 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
685 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
686 if (start_active_slot(slot)) {
687 run_active_slot(slot);
688 if (results.curl_result != CURLE_OK) {
689 if (missing_target(&results)) {
690 repo->got_indices = 1;
693 repo->got_indices = 0;
694 ret = error("%s", curl_errorstr);
699 repo->got_indices = 0;
700 ret = error("Unable to start request");
705 while (i < buffer.len) {
709 if (i + 52 <= buffer.len &&
710 !prefixcmp(data + i, " pack-") &&
711 !prefixcmp(data + i + 46, ".pack\n")) {
712 get_sha1_hex(data + i + 6, sha1);
713 setup_index(walker, repo, sha1);
718 while (i < buffer.len && data[i] != '\n')
724 repo->got_indices = 1;
726 strbuf_release(&buffer);
731 static int fetch_pack(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
734 struct packed_git *target;
735 struct packed_git **lst;
738 char tmpfile[PATH_MAX];
741 char range[RANGE_HEADER_SIZE];
742 struct curl_slist *range_header = NULL;
743 struct walker_data *data = walker->data;
745 struct active_request_slot *slot;
746 struct slot_results results;
748 if (fetch_indices(walker, repo))
750 target = find_sha1_pack(sha1, repo->packs);
754 if (walker->get_verbosely) {
755 fprintf(stderr, "Getting pack %s\n",
756 sha1_to_hex(target->sha1));
757 fprintf(stderr, " which contains %s\n",
761 url = xmalloc(strlen(repo->base) + 65);
762 sprintf(url, "%s/objects/pack/pack-%s.pack",
763 repo->base, sha1_to_hex(target->sha1));
765 filename = sha1_pack_name(target->sha1);
766 snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
767 packfile = fopen(tmpfile, "a");
769 return error("Unable to open local file %s for pack",
772 slot = get_active_slot();
773 slot->results = &results;
774 curl_easy_setopt(slot->curl, CURLOPT_FILE, packfile);
775 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
776 curl_easy_setopt(slot->curl, CURLOPT_URL, url);
777 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, data->no_pragma_header);
778 slot->local = packfile;
781 * If there is data present from a previous transfer attempt,
782 * resume where it left off
784 prev_posn = ftell(packfile);
786 if (walker->get_verbosely)
788 "Resuming fetch of pack %s at byte %ld\n",
789 sha1_to_hex(target->sha1), prev_posn);
790 sprintf(range, "Range: bytes=%ld-", prev_posn);
791 range_header = curl_slist_append(range_header, range);
792 curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
795 if (start_active_slot(slot)) {
796 run_active_slot(slot);
797 if (results.curl_result != CURLE_OK) {
800 return error("Unable to get pack file %s\n%s", url,
806 return error("Unable to start request");
809 target->pack_size = ftell(packfile);
813 ret = move_temp_to_file(tmpfile, filename);
818 while (*lst != target)
819 lst = &((*lst)->next);
822 if (verify_pack(target))
824 install_packed_git(target);
829 static void abort_object_request(struct object_request *obj_req)
831 if (obj_req->local >= 0) {
832 close(obj_req->local);
835 unlink_or_warn(obj_req->tmpfile);
837 release_active_slot(obj_req->slot);
838 obj_req->slot = NULL;
840 release_object_request(obj_req);
843 static int fetch_object(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
845 char *hex = sha1_to_hex(sha1);
847 struct object_request *obj_req = object_queue_head;
849 while (obj_req != NULL && hashcmp(obj_req->sha1, sha1))
850 obj_req = obj_req->next;
852 return error("Couldn't find request for %s in the queue", hex);
854 if (has_sha1_file(obj_req->sha1)) {
855 abort_object_request(obj_req);
859 #ifdef USE_CURL_MULTI
860 while (obj_req->state == WAITING)
863 start_object_request(walker, obj_req);
866 while (obj_req->state == ACTIVE)
867 run_active_slot(obj_req->slot);
869 if (obj_req->local != -1) {
870 close(obj_req->local);
874 if (obj_req->state == ABORTED) {
875 ret = error("Request for %s aborted", hex);
876 } else if (obj_req->curl_result != CURLE_OK &&
877 obj_req->http_code != 416) {
878 if (missing_target(obj_req))
879 ret = -1; /* Be silent, it is probably in a pack. */
881 ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
882 obj_req->errorstr, obj_req->curl_result,
883 obj_req->http_code, hex);
884 } else if (obj_req->zret != Z_STREAM_END) {
885 walker->corrupt_object_found++;
886 ret = error("File %s (%s) corrupt", hex, obj_req->url);
887 } else if (hashcmp(obj_req->sha1, obj_req->real_sha1)) {
888 ret = error("File %s has bad hash", hex);
889 } else if (obj_req->rename < 0) {
890 ret = error("unable to write sha1 filename %s",
894 release_object_request(obj_req);
898 static int fetch(struct walker *walker, unsigned char *sha1)
900 struct walker_data *data = walker->data;
901 struct alt_base *altbase = data->alt;
903 if (!fetch_object(walker, altbase, sha1))
906 if (!fetch_pack(walker, altbase, sha1))
908 fetch_alternates(walker, data->alt->base);
909 altbase = altbase->next;
911 return error("Unable to find %s under %s", sha1_to_hex(sha1),
915 static int fetch_ref(struct walker *walker, struct ref *ref)
917 struct walker_data *data = walker->data;
918 return http_fetch_ref(data->alt->base, ref);
921 static void cleanup(struct walker *walker)
923 struct walker_data *data = walker->data;
926 curl_slist_free_all(data->no_pragma_header);
929 struct walker *get_http_walker(const char *url, struct remote *remote)
932 struct walker_data *data = xmalloc(sizeof(struct walker_data));
933 struct walker *walker = xmalloc(sizeof(struct walker));
937 data->no_pragma_header = curl_slist_append(NULL, "Pragma:");
939 data->alt = xmalloc(sizeof(*data->alt));
940 data->alt->base = xmalloc(strlen(url) + 1);
941 strcpy(data->alt->base, url);
942 for (s = data->alt->base + strlen(data->alt->base) - 1; *s == '/'; --s)
945 data->alt->got_indices = 0;
946 data->alt->packs = NULL;
947 data->alt->next = NULL;
948 data->got_alternates = -1;
950 walker->corrupt_object_found = 0;
951 walker->fetch = fetch;
952 walker->fetch_ref = fetch_ref;
953 walker->prefetch = prefetch;
954 walker->cleanup = cleanup;
957 #ifdef USE_CURL_MULTI
958 add_fill_function(walker, (int (*)(void *)) fill_active_slot);