Merge branch 'jk/output-prefix-cleanup'
[git/gitster.git] / http-walker.c
blobfb2d86d5e70f0e3b7a12ac74d910e032dff9a3a7
1 #define USE_THE_REPOSITORY_VARIABLE
3 #include "git-compat-util.h"
4 #include "repository.h"
5 #include "hex.h"
6 #include "walker.h"
7 #include "http.h"
8 #include "list.h"
9 #include "transport.h"
10 #include "packfile.h"
11 #include "object-store-ll.h"
13 struct alt_base {
14 char *base;
15 int got_indices;
16 struct packed_git *packs;
17 struct alt_base *next;
20 enum object_request_state {
21 WAITING,
22 ABORTED,
23 ACTIVE,
24 COMPLETE
27 struct object_request {
28 struct walker *walker;
29 struct object_id oid;
30 struct alt_base *repo;
31 enum object_request_state state;
32 struct http_object_request *req;
33 struct list_head node;
36 struct alternates_request {
37 struct walker *walker;
38 const char *base;
39 struct strbuf *url;
40 struct strbuf *buffer;
41 struct active_request_slot *slot;
42 int http_specific;
45 struct walker_data {
46 const char *url;
47 int got_alternates;
48 struct alt_base *alt;
51 static LIST_HEAD(object_queue_head);
53 static void fetch_alternates(struct walker *walker, const char *base);
55 static void process_object_response(void *callback_data);
57 static void start_object_request(struct object_request *obj_req)
59 struct active_request_slot *slot;
60 struct http_object_request *req;
62 req = new_http_object_request(obj_req->repo->base, &obj_req->oid);
63 if (!req) {
64 obj_req->state = ABORTED;
65 return;
67 obj_req->req = req;
69 slot = req->slot;
70 slot->callback_func = process_object_response;
71 slot->callback_data = obj_req;
73 /* Try to get the request started, abort the request on error */
74 obj_req->state = ACTIVE;
75 if (!start_active_slot(slot)) {
76 obj_req->state = ABORTED;
77 release_http_object_request(&req);
78 return;
82 static void finish_object_request(struct object_request *obj_req)
84 if (finish_http_object_request(obj_req->req))
85 return;
87 if (obj_req->req->rename == 0)
88 walker_say(obj_req->walker, "got %s\n", oid_to_hex(&obj_req->oid));
91 static void process_object_response(void *callback_data)
93 struct object_request *obj_req =
94 (struct object_request *)callback_data;
95 struct walker *walker = obj_req->walker;
96 struct walker_data *data = walker->data;
97 struct alt_base *alt = data->alt;
99 process_http_object_request(obj_req->req);
100 obj_req->state = COMPLETE;
102 normalize_curl_result(&obj_req->req->curl_result,
103 obj_req->req->http_code,
104 obj_req->req->errorstr,
105 sizeof(obj_req->req->errorstr));
107 /* Use alternates if necessary */
108 if (missing_target(obj_req->req)) {
109 fetch_alternates(walker, alt->base);
110 if (obj_req->repo->next) {
111 obj_req->repo =
112 obj_req->repo->next;
113 release_http_object_request(&obj_req->req);
114 start_object_request(obj_req);
115 return;
119 finish_object_request(obj_req);
122 static void release_object_request(struct object_request *obj_req)
124 if (obj_req->req !=NULL && obj_req->req->localfile != -1)
125 error("fd leakage in release: %d", obj_req->req->localfile);
127 list_del(&obj_req->node);
128 free(obj_req);
131 static int fill_active_slot(void *data UNUSED)
133 struct object_request *obj_req;
134 struct list_head *pos, *tmp, *head = &object_queue_head;
136 list_for_each_safe(pos, tmp, head) {
137 obj_req = list_entry(pos, struct object_request, node);
138 if (obj_req->state == WAITING) {
139 if (repo_has_object_file(the_repository, &obj_req->oid))
140 obj_req->state = COMPLETE;
141 else {
142 start_object_request(obj_req);
143 return 1;
147 return 0;
150 static void prefetch(struct walker *walker, unsigned char *sha1)
152 struct object_request *newreq;
153 struct walker_data *data = walker->data;
155 newreq = xmalloc(sizeof(*newreq));
156 newreq->walker = walker;
157 oidread(&newreq->oid, sha1, the_repository->hash_algo);
158 newreq->repo = data->alt;
159 newreq->state = WAITING;
160 newreq->req = NULL;
162 http_is_verbose = walker->get_verbosely;
163 list_add_tail(&newreq->node, &object_queue_head);
165 fill_active_slots();
166 step_active_slots();
169 static int is_alternate_allowed(const char *url)
171 const char *protocols[] = {
172 "http", "https", "ftp", "ftps"
174 int i;
176 if (http_follow_config != HTTP_FOLLOW_ALWAYS) {
177 warning("alternate disabled by http.followRedirects: %s", url);
178 return 0;
181 for (i = 0; i < ARRAY_SIZE(protocols); i++) {
182 const char *end;
183 if (skip_prefix(url, protocols[i], &end) &&
184 starts_with(end, "://"))
185 break;
188 if (i >= ARRAY_SIZE(protocols)) {
189 warning("ignoring alternate with unknown protocol: %s", url);
190 return 0;
192 if (!is_transport_allowed(protocols[i], 0)) {
193 warning("ignoring alternate with restricted protocol: %s", url);
194 return 0;
197 return 1;
200 static void process_alternates_response(void *callback_data)
202 struct alternates_request *alt_req =
203 (struct alternates_request *)callback_data;
204 struct walker *walker = alt_req->walker;
205 struct walker_data *cdata = walker->data;
206 struct active_request_slot *slot = alt_req->slot;
207 struct alt_base *tail = cdata->alt;
208 const char *base = alt_req->base;
209 const char null_byte = '\0';
210 char *data;
211 int i = 0;
213 normalize_curl_result(&slot->curl_result, slot->http_code,
214 curl_errorstr, sizeof(curl_errorstr));
216 if (alt_req->http_specific) {
217 if (slot->curl_result != CURLE_OK ||
218 !alt_req->buffer->len) {
220 /* Try reusing the slot to get non-http alternates */
221 alt_req->http_specific = 0;
222 strbuf_reset(alt_req->url);
223 strbuf_addf(alt_req->url, "%s/objects/info/alternates",
224 base);
225 curl_easy_setopt(slot->curl, CURLOPT_URL,
226 alt_req->url->buf);
227 active_requests++;
228 slot->in_use = 1;
229 if (slot->finished)
230 (*slot->finished) = 0;
231 if (!start_active_slot(slot)) {
232 cdata->got_alternates = -1;
233 slot->in_use = 0;
234 if (slot->finished)
235 (*slot->finished) = 1;
237 return;
239 } else if (slot->curl_result != CURLE_OK) {
240 if (!missing_target(slot)) {
241 cdata->got_alternates = -1;
242 return;
246 fwrite_buffer((char *)&null_byte, 1, 1, alt_req->buffer);
247 alt_req->buffer->len--;
248 data = alt_req->buffer->buf;
250 while (i < alt_req->buffer->len) {
251 int posn = i;
252 while (posn < alt_req->buffer->len && data[posn] != '\n')
253 posn++;
254 if (data[posn] == '\n') {
255 int okay = 0;
256 int serverlen = 0;
257 struct alt_base *newalt;
258 if (data[i] == '/') {
260 * This counts
261 * http://git.host/pub/scm/linux.git/
262 * -----------here^
263 * so memcpy(dst, base, serverlen) will
264 * copy up to "...git.host".
266 const char *colon_ss = strstr(base,"://");
267 if (colon_ss) {
268 serverlen = (strchr(colon_ss + 3, '/')
269 - base);
270 okay = 1;
272 } else if (!memcmp(data + i, "../", 3)) {
274 * Relative URL; chop the corresponding
275 * number of subpath from base (and ../
276 * from data), and concatenate the result.
278 * The code first drops ../ from data, and
279 * then drops one ../ from data and one path
280 * from base. IOW, one extra ../ is dropped
281 * from data than path is dropped from base.
283 * This is not wrong. The alternate in
284 * http://git.host/pub/scm/linux.git/
285 * to borrow from
286 * http://git.host/pub/scm/linus.git/
287 * is ../../linus.git/objects/. You need
288 * two ../../ to borrow from your direct
289 * neighbour.
291 i += 3;
292 serverlen = strlen(base);
293 while (i + 2 < posn &&
294 !memcmp(data + i, "../", 3)) {
295 do {
296 serverlen--;
297 } while (serverlen &&
298 base[serverlen - 1] != '/');
299 i += 3;
301 /* If the server got removed, give up. */
302 okay = strchr(base, ':') - base + 3 <
303 serverlen;
304 } else if (alt_req->http_specific) {
305 char *colon = strchr(data + i, ':');
306 char *slash = strchr(data + i, '/');
307 if (colon && slash && colon < data + posn &&
308 slash < data + posn && colon < slash) {
309 okay = 1;
312 if (okay) {
313 struct strbuf target = STRBUF_INIT;
314 strbuf_add(&target, base, serverlen);
315 strbuf_add(&target, data + i, posn - i);
316 if (!strbuf_strip_suffix(&target, "objects")) {
317 warning("ignoring alternate that does"
318 " not end in 'objects': %s",
319 target.buf);
320 strbuf_release(&target);
321 } else if (is_alternate_allowed(target.buf)) {
322 warning("adding alternate object store: %s",
323 target.buf);
324 newalt = xmalloc(sizeof(*newalt));
325 newalt->next = NULL;
326 newalt->base = strbuf_detach(&target, NULL);
327 newalt->got_indices = 0;
328 newalt->packs = NULL;
330 while (tail->next != NULL)
331 tail = tail->next;
332 tail->next = newalt;
333 } else {
334 strbuf_release(&target);
338 i = posn + 1;
341 cdata->got_alternates = 1;
344 static void fetch_alternates(struct walker *walker, const char *base)
346 struct strbuf buffer = STRBUF_INIT;
347 struct strbuf url = STRBUF_INIT;
348 struct active_request_slot *slot;
349 struct alternates_request alt_req;
350 struct walker_data *cdata = walker->data;
353 * If another request has already started fetching alternates,
354 * wait for them to arrive and return to processing this request's
355 * curl message
357 while (cdata->got_alternates == 0) {
358 step_active_slots();
361 /* Nothing to do if they've already been fetched */
362 if (cdata->got_alternates == 1)
363 return;
365 /* Start the fetch */
366 cdata->got_alternates = 0;
368 if (walker->get_verbosely)
369 fprintf(stderr, "Getting alternates list for %s\n", base);
371 strbuf_addf(&url, "%s/objects/info/http-alternates", base);
374 * Use a callback to process the result, since another request
375 * may fail and need to have alternates loaded before continuing
377 slot = get_active_slot();
378 slot->callback_func = process_alternates_response;
379 alt_req.walker = walker;
380 slot->callback_data = &alt_req;
382 curl_easy_setopt(slot->curl, CURLOPT_WRITEDATA, &buffer);
383 curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
384 curl_easy_setopt(slot->curl, CURLOPT_URL, url.buf);
386 alt_req.base = base;
387 alt_req.url = &url;
388 alt_req.buffer = &buffer;
389 alt_req.http_specific = 1;
390 alt_req.slot = slot;
392 if (start_active_slot(slot))
393 run_active_slot(slot);
394 else
395 cdata->got_alternates = -1;
397 strbuf_release(&buffer);
398 strbuf_release(&url);
401 static int fetch_indices(struct walker *walker, struct alt_base *repo)
403 int ret;
405 if (repo->got_indices)
406 return 0;
408 if (walker->get_verbosely)
409 fprintf(stderr, "Getting pack list for %s\n", repo->base);
411 switch (http_get_info_packs(repo->base, &repo->packs)) {
412 case HTTP_OK:
413 case HTTP_MISSING_TARGET:
414 repo->got_indices = 1;
415 ret = 0;
416 break;
417 default:
418 repo->got_indices = 0;
419 ret = -1;
422 return ret;
425 static int http_fetch_pack(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
427 struct packed_git *target;
428 int ret;
429 struct slot_results results;
430 struct http_pack_request *preq;
432 if (fetch_indices(walker, repo))
433 return -1;
434 target = find_sha1_pack(sha1, repo->packs);
435 if (!target)
436 return -1;
437 close_pack_index(target);
439 if (walker->get_verbosely) {
440 fprintf(stderr, "Getting pack %s\n",
441 hash_to_hex(target->hash));
442 fprintf(stderr, " which contains %s\n",
443 hash_to_hex(sha1));
446 preq = new_http_pack_request(target->hash, repo->base);
447 if (!preq)
448 goto abort;
449 preq->slot->results = &results;
451 if (start_active_slot(preq->slot)) {
452 run_active_slot(preq->slot);
453 if (results.curl_result != CURLE_OK) {
454 error("Unable to get pack file %s\n%s", preq->url,
455 curl_errorstr);
456 goto abort;
458 } else {
459 error("Unable to start request");
460 goto abort;
463 ret = finish_http_pack_request(preq);
464 release_http_pack_request(preq);
465 if (ret)
466 return ret;
467 http_install_packfile(target, &repo->packs);
469 return 0;
471 abort:
472 return -1;
475 static void abort_object_request(struct object_request *obj_req)
477 release_object_request(obj_req);
480 static int fetch_object(struct walker *walker, unsigned char *hash)
482 char *hex = hash_to_hex(hash);
483 int ret = 0;
484 struct object_request *obj_req = NULL;
485 struct http_object_request *req;
486 struct list_head *pos, *head = &object_queue_head;
488 list_for_each(pos, head) {
489 obj_req = list_entry(pos, struct object_request, node);
490 if (hasheq(obj_req->oid.hash, hash, the_repository->hash_algo))
491 break;
493 if (!obj_req)
494 return error("Couldn't find request for %s in the queue", hex);
496 if (repo_has_object_file(the_repository, &obj_req->oid)) {
497 if (obj_req->req)
498 abort_http_object_request(&obj_req->req);
499 abort_object_request(obj_req);
500 return 0;
503 while (obj_req->state == WAITING)
504 step_active_slots();
507 * obj_req->req might change when fetching alternates in the callback
508 * process_object_response; therefore, the "shortcut" variable, req,
509 * is used only after we're done with slots.
511 while (obj_req->state == ACTIVE)
512 run_active_slot(obj_req->req->slot);
514 req = obj_req->req;
516 if (req->localfile != -1) {
517 close(req->localfile);
518 req->localfile = -1;
521 normalize_curl_result(&req->curl_result, req->http_code,
522 req->errorstr, sizeof(req->errorstr));
524 if (obj_req->state == ABORTED) {
525 ret = error("Request for %s aborted", hex);
526 } else if (req->curl_result != CURLE_OK &&
527 req->http_code != 416) {
528 if (missing_target(req))
529 ret = -1; /* Be silent, it is probably in a pack. */
530 else
531 ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
532 req->errorstr, req->curl_result,
533 req->http_code, hex);
534 } else if (req->zret != Z_STREAM_END) {
535 walker->corrupt_object_found++;
536 ret = error("File %s (%s) corrupt", hex, req->url);
537 } else if (!oideq(&obj_req->oid, &req->real_oid)) {
538 ret = error("File %s has bad hash", hex);
539 } else if (req->rename < 0) {
540 struct strbuf buf = STRBUF_INIT;
541 loose_object_path(the_repository, &buf, &req->oid);
542 ret = error("unable to write sha1 filename %s", buf.buf);
543 strbuf_release(&buf);
546 release_http_object_request(&obj_req->req);
547 release_object_request(obj_req);
548 return ret;
551 static int fetch(struct walker *walker, unsigned char *hash)
553 struct walker_data *data = walker->data;
554 struct alt_base *altbase = data->alt;
556 if (!fetch_object(walker, hash))
557 return 0;
558 while (altbase) {
559 if (!http_fetch_pack(walker, altbase, hash))
560 return 0;
561 fetch_alternates(walker, data->alt->base);
562 altbase = altbase->next;
564 return error("Unable to find %s under %s", hash_to_hex(hash),
565 data->alt->base);
568 static int fetch_ref(struct walker *walker, struct ref *ref)
570 struct walker_data *data = walker->data;
571 return http_fetch_ref(data->alt->base, ref);
574 static void cleanup(struct walker *walker)
576 struct walker_data *data = walker->data;
577 struct alt_base *alt, *alt_next;
579 if (data) {
580 alt = data->alt;
581 while (alt) {
582 struct packed_git *pack;
584 alt_next = alt->next;
586 pack = alt->packs;
587 while (pack) {
588 struct packed_git *pack_next = pack->next;
589 close_pack(pack);
590 free(pack);
591 pack = pack_next;
594 free(alt->base);
595 free(alt);
597 alt = alt_next;
599 free(data);
600 walker->data = NULL;
604 struct walker *get_http_walker(const char *url)
606 char *s;
607 struct walker_data *data = xmalloc(sizeof(struct walker_data));
608 struct walker *walker = xmalloc(sizeof(struct walker));
610 data->alt = xmalloc(sizeof(*data->alt));
611 data->alt->base = xstrdup(url);
612 for (s = data->alt->base + strlen(data->alt->base) - 1; *s == '/'; --s)
613 *s = 0;
615 data->alt->got_indices = 0;
616 data->alt->packs = NULL;
617 data->alt->next = NULL;
618 data->got_alternates = -1;
620 walker->corrupt_object_found = 0;
621 walker->fetch = fetch;
622 walker->fetch_ref = fetch_ref;
623 walker->prefetch = prefetch;
624 walker->cleanup = cleanup;
625 walker->data = data;
627 add_fill_function(NULL, fill_active_slot);
629 return walker;