packed_ref_cache: add a backlink to the associated `packed_ref_store`
[git/debian.git] / http-walker.c
blob1ae8363de2bb618df534cc85c86eba3eac9e9c0a
1 #include "cache.h"
2 #include "commit.h"
3 #include "walker.h"
4 #include "http.h"
5 #include "list.h"
6 #include "transport.h"
7 #include "packfile.h"
9 struct alt_base {
10 char *base;
11 int got_indices;
12 struct packed_git *packs;
13 struct alt_base *next;
16 enum object_request_state {
17 WAITING,
18 ABORTED,
19 ACTIVE,
20 COMPLETE
23 struct object_request {
24 struct walker *walker;
25 unsigned char sha1[20];
26 struct alt_base *repo;
27 enum object_request_state state;
28 struct http_object_request *req;
29 struct list_head node;
32 struct alternates_request {
33 struct walker *walker;
34 const char *base;
35 struct strbuf *url;
36 struct strbuf *buffer;
37 struct active_request_slot *slot;
38 int http_specific;
41 struct walker_data {
42 const char *url;
43 int got_alternates;
44 struct alt_base *alt;
47 static LIST_HEAD(object_queue_head);
49 static void fetch_alternates(struct walker *walker, const char *base);
51 static void process_object_response(void *callback_data);
53 static void start_object_request(struct walker *walker,
54 struct object_request *obj_req)
56 struct active_request_slot *slot;
57 struct http_object_request *req;
59 req = new_http_object_request(obj_req->repo->base, obj_req->sha1);
60 if (req == NULL) {
61 obj_req->state = ABORTED;
62 return;
64 obj_req->req = req;
66 slot = req->slot;
67 slot->callback_func = process_object_response;
68 slot->callback_data = obj_req;
70 /* Try to get the request started, abort the request on error */
71 obj_req->state = ACTIVE;
72 if (!start_active_slot(slot)) {
73 obj_req->state = ABORTED;
74 release_http_object_request(req);
75 return;
79 static void finish_object_request(struct object_request *obj_req)
81 if (finish_http_object_request(obj_req->req))
82 return;
84 if (obj_req->req->rename == 0)
85 walker_say(obj_req->walker, "got %s\n", sha1_to_hex(obj_req->sha1));
88 static void process_object_response(void *callback_data)
90 struct object_request *obj_req =
91 (struct object_request *)callback_data;
92 struct walker *walker = obj_req->walker;
93 struct walker_data *data = walker->data;
94 struct alt_base *alt = data->alt;
96 process_http_object_request(obj_req->req);
97 obj_req->state = COMPLETE;
99 /* Use alternates if necessary */
100 if (missing_target(obj_req->req)) {
101 fetch_alternates(walker, alt->base);
102 if (obj_req->repo->next != NULL) {
103 obj_req->repo =
104 obj_req->repo->next;
105 release_http_object_request(obj_req->req);
106 start_object_request(walker, obj_req);
107 return;
111 finish_object_request(obj_req);
114 static void release_object_request(struct object_request *obj_req)
116 if (obj_req->req !=NULL && obj_req->req->localfile != -1)
117 error("fd leakage in release: %d", obj_req->req->localfile);
119 list_del(&obj_req->node);
120 free(obj_req);
123 #ifdef USE_CURL_MULTI
124 static int fill_active_slot(struct walker *walker)
126 struct object_request *obj_req;
127 struct list_head *pos, *tmp, *head = &object_queue_head;
129 list_for_each_safe(pos, tmp, head) {
130 obj_req = list_entry(pos, struct object_request, node);
131 if (obj_req->state == WAITING) {
132 if (has_sha1_file(obj_req->sha1))
133 obj_req->state = COMPLETE;
134 else {
135 start_object_request(walker, obj_req);
136 return 1;
140 return 0;
142 #endif
144 static void prefetch(struct walker *walker, unsigned char *sha1)
146 struct object_request *newreq;
147 struct walker_data *data = walker->data;
149 newreq = xmalloc(sizeof(*newreq));
150 newreq->walker = walker;
151 hashcpy(newreq->sha1, sha1);
152 newreq->repo = data->alt;
153 newreq->state = WAITING;
154 newreq->req = NULL;
156 http_is_verbose = walker->get_verbosely;
157 list_add_tail(&newreq->node, &object_queue_head);
159 #ifdef USE_CURL_MULTI
160 fill_active_slots();
161 step_active_slots();
162 #endif
165 static int is_alternate_allowed(const char *url)
167 const char *protocols[] = {
168 "http", "https", "ftp", "ftps"
170 int i;
172 if (http_follow_config != HTTP_FOLLOW_ALWAYS) {
173 warning("alternate disabled by http.followRedirects: %s", url);
174 return 0;
177 for (i = 0; i < ARRAY_SIZE(protocols); i++) {
178 const char *end;
179 if (skip_prefix(url, protocols[i], &end) &&
180 starts_with(end, "://"))
181 break;
184 if (i >= ARRAY_SIZE(protocols)) {
185 warning("ignoring alternate with unknown protocol: %s", url);
186 return 0;
188 if (!is_transport_allowed(protocols[i], 0)) {
189 warning("ignoring alternate with restricted protocol: %s", url);
190 return 0;
193 return 1;
196 static void process_alternates_response(void *callback_data)
198 struct alternates_request *alt_req =
199 (struct alternates_request *)callback_data;
200 struct walker *walker = alt_req->walker;
201 struct walker_data *cdata = walker->data;
202 struct active_request_slot *slot = alt_req->slot;
203 struct alt_base *tail = cdata->alt;
204 const char *base = alt_req->base;
205 const char null_byte = '\0';
206 char *data;
207 int i = 0;
209 if (alt_req->http_specific) {
210 if (slot->curl_result != CURLE_OK ||
211 !alt_req->buffer->len) {
213 /* Try reusing the slot to get non-http alternates */
214 alt_req->http_specific = 0;
215 strbuf_reset(alt_req->url);
216 strbuf_addf(alt_req->url, "%s/objects/info/alternates",
217 base);
218 curl_easy_setopt(slot->curl, CURLOPT_URL,
219 alt_req->url->buf);
220 active_requests++;
221 slot->in_use = 1;
222 if (slot->finished != NULL)
223 (*slot->finished) = 0;
224 if (!start_active_slot(slot)) {
225 cdata->got_alternates = -1;
226 slot->in_use = 0;
227 if (slot->finished != NULL)
228 (*slot->finished) = 1;
230 return;
232 } else if (slot->curl_result != CURLE_OK) {
233 if (!missing_target(slot)) {
234 cdata->got_alternates = -1;
235 return;
239 fwrite_buffer((char *)&null_byte, 1, 1, alt_req->buffer);
240 alt_req->buffer->len--;
241 data = alt_req->buffer->buf;
243 while (i < alt_req->buffer->len) {
244 int posn = i;
245 while (posn < alt_req->buffer->len && data[posn] != '\n')
246 posn++;
247 if (data[posn] == '\n') {
248 int okay = 0;
249 int serverlen = 0;
250 struct alt_base *newalt;
251 if (data[i] == '/') {
253 * This counts
254 * http://git.host/pub/scm/linux.git/
255 * -----------here^
256 * so memcpy(dst, base, serverlen) will
257 * copy up to "...git.host".
259 const char *colon_ss = strstr(base,"://");
260 if (colon_ss) {
261 serverlen = (strchr(colon_ss + 3, '/')
262 - base);
263 okay = 1;
265 } else if (!memcmp(data + i, "../", 3)) {
267 * Relative URL; chop the corresponding
268 * number of subpath from base (and ../
269 * from data), and concatenate the result.
271 * The code first drops ../ from data, and
272 * then drops one ../ from data and one path
273 * from base. IOW, one extra ../ is dropped
274 * from data than path is dropped from base.
276 * This is not wrong. The alternate in
277 * http://git.host/pub/scm/linux.git/
278 * to borrow from
279 * http://git.host/pub/scm/linus.git/
280 * is ../../linus.git/objects/. You need
281 * two ../../ to borrow from your direct
282 * neighbour.
284 i += 3;
285 serverlen = strlen(base);
286 while (i + 2 < posn &&
287 !memcmp(data + i, "../", 3)) {
288 do {
289 serverlen--;
290 } while (serverlen &&
291 base[serverlen - 1] != '/');
292 i += 3;
294 /* If the server got removed, give up. */
295 okay = strchr(base, ':') - base + 3 <
296 serverlen;
297 } else if (alt_req->http_specific) {
298 char *colon = strchr(data + i, ':');
299 char *slash = strchr(data + i, '/');
300 if (colon && slash && colon < data + posn &&
301 slash < data + posn && colon < slash) {
302 okay = 1;
305 if (okay) {
306 struct strbuf target = STRBUF_INIT;
307 strbuf_add(&target, base, serverlen);
308 strbuf_add(&target, data + i, posn - i);
309 if (!strbuf_strip_suffix(&target, "objects")) {
310 warning("ignoring alternate that does"
311 " not end in 'objects': %s",
312 target.buf);
313 strbuf_release(&target);
314 } else if (is_alternate_allowed(target.buf)) {
315 warning("adding alternate object store: %s",
316 target.buf);
317 newalt = xmalloc(sizeof(*newalt));
318 newalt->next = NULL;
319 newalt->base = strbuf_detach(&target, NULL);
320 newalt->got_indices = 0;
321 newalt->packs = NULL;
323 while (tail->next != NULL)
324 tail = tail->next;
325 tail->next = newalt;
326 } else {
327 strbuf_release(&target);
331 i = posn + 1;
334 cdata->got_alternates = 1;
337 static void fetch_alternates(struct walker *walker, const char *base)
339 struct strbuf buffer = STRBUF_INIT;
340 struct strbuf url = STRBUF_INIT;
341 struct active_request_slot *slot;
342 struct alternates_request alt_req;
343 struct walker_data *cdata = walker->data;
346 * If another request has already started fetching alternates,
347 * wait for them to arrive and return to processing this request's
348 * curl message
350 #ifdef USE_CURL_MULTI
351 while (cdata->got_alternates == 0) {
352 step_active_slots();
354 #endif
356 /* Nothing to do if they've already been fetched */
357 if (cdata->got_alternates == 1)
358 return;
360 /* Start the fetch */
361 cdata->got_alternates = 0;
363 if (walker->get_verbosely)
364 fprintf(stderr, "Getting alternates list for %s\n", base);
366 strbuf_addf(&url, "%s/objects/info/http-alternates", base);
369 * Use a callback to process the result, since another request
370 * may fail and need to have alternates loaded before continuing
372 slot = get_active_slot();
373 slot->callback_func = process_alternates_response;
374 alt_req.walker = walker;
375 slot->callback_data = &alt_req;
377 curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
378 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
379 curl_easy_setopt(slot->curl, CURLOPT_URL, url.buf);
381 alt_req.base = base;
382 alt_req.url = &url;
383 alt_req.buffer = &buffer;
384 alt_req.http_specific = 1;
385 alt_req.slot = slot;
387 if (start_active_slot(slot))
388 run_active_slot(slot);
389 else
390 cdata->got_alternates = -1;
392 strbuf_release(&buffer);
393 strbuf_release(&url);
396 static int fetch_indices(struct walker *walker, struct alt_base *repo)
398 int ret;
400 if (repo->got_indices)
401 return 0;
403 if (walker->get_verbosely)
404 fprintf(stderr, "Getting pack list for %s\n", repo->base);
406 switch (http_get_info_packs(repo->base, &repo->packs)) {
407 case HTTP_OK:
408 case HTTP_MISSING_TARGET:
409 repo->got_indices = 1;
410 ret = 0;
411 break;
412 default:
413 repo->got_indices = 0;
414 ret = -1;
417 return ret;
420 static int http_fetch_pack(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
422 struct packed_git *target;
423 int ret;
424 struct slot_results results;
425 struct http_pack_request *preq;
427 if (fetch_indices(walker, repo))
428 return -1;
429 target = find_sha1_pack(sha1, repo->packs);
430 if (!target)
431 return -1;
433 if (walker->get_verbosely) {
434 fprintf(stderr, "Getting pack %s\n",
435 sha1_to_hex(target->sha1));
436 fprintf(stderr, " which contains %s\n",
437 sha1_to_hex(sha1));
440 preq = new_http_pack_request(target, repo->base);
441 if (preq == NULL)
442 goto abort;
443 preq->lst = &repo->packs;
444 preq->slot->results = &results;
446 if (start_active_slot(preq->slot)) {
447 run_active_slot(preq->slot);
448 if (results.curl_result != CURLE_OK) {
449 error("Unable to get pack file %s\n%s", preq->url,
450 curl_errorstr);
451 goto abort;
453 } else {
454 error("Unable to start request");
455 goto abort;
458 ret = finish_http_pack_request(preq);
459 release_http_pack_request(preq);
460 if (ret)
461 return ret;
463 return 0;
465 abort:
466 return -1;
469 static void abort_object_request(struct object_request *obj_req)
471 release_object_request(obj_req);
474 static int fetch_object(struct walker *walker, unsigned char *sha1)
476 char *hex = sha1_to_hex(sha1);
477 int ret = 0;
478 struct object_request *obj_req = NULL;
479 struct http_object_request *req;
480 struct list_head *pos, *head = &object_queue_head;
482 list_for_each(pos, head) {
483 obj_req = list_entry(pos, struct object_request, node);
484 if (!hashcmp(obj_req->sha1, sha1))
485 break;
487 if (obj_req == NULL)
488 return error("Couldn't find request for %s in the queue", hex);
490 if (has_sha1_file(obj_req->sha1)) {
491 if (obj_req->req != NULL)
492 abort_http_object_request(obj_req->req);
493 abort_object_request(obj_req);
494 return 0;
497 #ifdef USE_CURL_MULTI
498 while (obj_req->state == WAITING)
499 step_active_slots();
500 #else
501 start_object_request(walker, obj_req);
502 #endif
505 * obj_req->req might change when fetching alternates in the callback
506 * process_object_response; therefore, the "shortcut" variable, req,
507 * is used only after we're done with slots.
509 while (obj_req->state == ACTIVE)
510 run_active_slot(obj_req->req->slot);
512 req = obj_req->req;
514 if (req->localfile != -1) {
515 close(req->localfile);
516 req->localfile = -1;
520 * we turned off CURLOPT_FAILONERROR to avoid losing a
521 * persistent connection and got CURLE_OK.
523 if (req->http_code >= 300 && req->curl_result == CURLE_OK &&
524 (starts_with(req->url, "http://") ||
525 starts_with(req->url, "https://"))) {
526 req->curl_result = CURLE_HTTP_RETURNED_ERROR;
527 xsnprintf(req->errorstr, sizeof(req->errorstr),
528 "HTTP request failed");
531 if (obj_req->state == ABORTED) {
532 ret = error("Request for %s aborted", hex);
533 } else if (req->curl_result != CURLE_OK &&
534 req->http_code != 416) {
535 if (missing_target(req))
536 ret = -1; /* Be silent, it is probably in a pack. */
537 else
538 ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
539 req->errorstr, req->curl_result,
540 req->http_code, hex);
541 } else if (req->zret != Z_STREAM_END) {
542 walker->corrupt_object_found++;
543 ret = error("File %s (%s) corrupt", hex, req->url);
544 } else if (hashcmp(obj_req->sha1, req->real_sha1)) {
545 ret = error("File %s has bad hash", hex);
546 } else if (req->rename < 0) {
547 ret = error("unable to write sha1 filename %s",
548 sha1_file_name(req->sha1));
551 release_http_object_request(req);
552 release_object_request(obj_req);
553 return ret;
556 static int fetch(struct walker *walker, unsigned char *sha1)
558 struct walker_data *data = walker->data;
559 struct alt_base *altbase = data->alt;
561 if (!fetch_object(walker, sha1))
562 return 0;
563 while (altbase) {
564 if (!http_fetch_pack(walker, altbase, sha1))
565 return 0;
566 fetch_alternates(walker, data->alt->base);
567 altbase = altbase->next;
569 return error("Unable to find %s under %s", sha1_to_hex(sha1),
570 data->alt->base);
573 static int fetch_ref(struct walker *walker, struct ref *ref)
575 struct walker_data *data = walker->data;
576 return http_fetch_ref(data->alt->base, ref);
579 static void cleanup(struct walker *walker)
581 struct walker_data *data = walker->data;
582 struct alt_base *alt, *alt_next;
584 if (data) {
585 alt = data->alt;
586 while (alt) {
587 alt_next = alt->next;
589 free(alt->base);
590 free(alt);
592 alt = alt_next;
594 free(data);
595 walker->data = NULL;
599 struct walker *get_http_walker(const char *url)
601 char *s;
602 struct walker_data *data = xmalloc(sizeof(struct walker_data));
603 struct walker *walker = xmalloc(sizeof(struct walker));
605 data->alt = xmalloc(sizeof(*data->alt));
606 data->alt->base = xstrdup(url);
607 for (s = data->alt->base + strlen(data->alt->base) - 1; *s == '/'; --s)
608 *s = 0;
610 data->alt->got_indices = 0;
611 data->alt->packs = NULL;
612 data->alt->next = NULL;
613 data->got_alternates = -1;
615 walker->corrupt_object_found = 0;
616 walker->fetch = fetch;
617 walker->fetch_ref = fetch_ref;
618 walker->prefetch = prefetch;
619 walker->cleanup = cleanup;
620 walker->data = data;
622 #ifdef USE_CURL_MULTI
623 add_fill_function(walker, (int (*)(void *)) fill_active_slot);
624 #endif
626 return walker;