http-walker.con commit Merge branch 'jk/fix-remote-curl-url-wo-proto' into maint (35ec7fd)
   1#include "cache.h"
   2#include "commit.h"
   3#include "walker.h"
   4#include "http.h"
   5#include "list.h"
   6
   7struct alt_base {
   8        char *base;
   9        int got_indices;
  10        struct packed_git *packs;
  11        struct alt_base *next;
  12};
  13
  14enum object_request_state {
  15        WAITING,
  16        ABORTED,
  17        ACTIVE,
  18        COMPLETE
  19};
  20
  21struct object_request {
  22        struct walker *walker;
  23        unsigned char sha1[20];
  24        struct alt_base *repo;
  25        enum object_request_state state;
  26        struct http_object_request *req;
  27        struct list_head node;
  28};
  29
  30struct alternates_request {
  31        struct walker *walker;
  32        const char *base;
  33        struct strbuf *url;
  34        struct strbuf *buffer;
  35        struct active_request_slot *slot;
  36        int http_specific;
  37};
  38
  39struct walker_data {
  40        const char *url;
  41        int got_alternates;
  42        struct alt_base *alt;
  43};
  44
  45static LIST_HEAD(object_queue_head);
  46
  47static void fetch_alternates(struct walker *walker, const char *base);
  48
  49static void process_object_response(void *callback_data);
  50
  51static void start_object_request(struct walker *walker,
  52                                 struct object_request *obj_req)
  53{
  54        struct active_request_slot *slot;
  55        struct http_object_request *req;
  56
  57        req = new_http_object_request(obj_req->repo->base, obj_req->sha1);
  58        if (req == NULL) {
  59                obj_req->state = ABORTED;
  60                return;
  61        }
  62        obj_req->req = req;
  63
  64        slot = req->slot;
  65        slot->callback_func = process_object_response;
  66        slot->callback_data = obj_req;
  67
  68        /* Try to get the request started, abort the request on error */
  69        obj_req->state = ACTIVE;
  70        if (!start_active_slot(slot)) {
  71                obj_req->state = ABORTED;
  72                release_http_object_request(req);
  73                return;
  74        }
  75}
  76
  77static void finish_object_request(struct object_request *obj_req)
  78{
  79        if (finish_http_object_request(obj_req->req))
  80                return;
  81
  82        if (obj_req->req->rename == 0)
  83                walker_say(obj_req->walker, "got %s\n", sha1_to_hex(obj_req->sha1));
  84}
  85
  86static void process_object_response(void *callback_data)
  87{
  88        struct object_request *obj_req =
  89                (struct object_request *)callback_data;
  90        struct walker *walker = obj_req->walker;
  91        struct walker_data *data = walker->data;
  92        struct alt_base *alt = data->alt;
  93
  94        process_http_object_request(obj_req->req);
  95        obj_req->state = COMPLETE;
  96
  97        /* Use alternates if necessary */
  98        if (missing_target(obj_req->req)) {
  99                fetch_alternates(walker, alt->base);
 100                if (obj_req->repo->next != NULL) {
 101                        obj_req->repo =
 102                                obj_req->repo->next;
 103                        release_http_object_request(obj_req->req);
 104                        start_object_request(walker, obj_req);
 105                        return;
 106                }
 107        }
 108
 109        finish_object_request(obj_req);
 110}
 111
 112static void release_object_request(struct object_request *obj_req)
 113{
 114        if (obj_req->req !=NULL && obj_req->req->localfile != -1)
 115                error("fd leakage in release: %d", obj_req->req->localfile);
 116
 117        list_del(&obj_req->node);
 118        free(obj_req);
 119}
 120
 121#ifdef USE_CURL_MULTI
 122static int fill_active_slot(struct walker *walker)
 123{
 124        struct object_request *obj_req;
 125        struct list_head *pos, *tmp, *head = &object_queue_head;
 126
 127        list_for_each_safe(pos, tmp, head) {
 128                obj_req = list_entry(pos, struct object_request, node);
 129                if (obj_req->state == WAITING) {
 130                        if (has_sha1_file(obj_req->sha1))
 131                                obj_req->state = COMPLETE;
 132                        else {
 133                                start_object_request(walker, obj_req);
 134                                return 1;
 135                        }
 136                }
 137        }
 138        return 0;
 139}
 140#endif
 141
 142static void prefetch(struct walker *walker, unsigned char *sha1)
 143{
 144        struct object_request *newreq;
 145        struct walker_data *data = walker->data;
 146
 147        newreq = xmalloc(sizeof(*newreq));
 148        newreq->walker = walker;
 149        hashcpy(newreq->sha1, sha1);
 150        newreq->repo = data->alt;
 151        newreq->state = WAITING;
 152        newreq->req = NULL;
 153
 154        http_is_verbose = walker->get_verbosely;
 155        list_add_tail(&newreq->node, &object_queue_head);
 156
 157#ifdef USE_CURL_MULTI
 158        fill_active_slots();
 159        step_active_slots();
 160#endif
 161}
 162
 163static void process_alternates_response(void *callback_data)
 164{
 165        struct alternates_request *alt_req =
 166                (struct alternates_request *)callback_data;
 167        struct walker *walker = alt_req->walker;
 168        struct walker_data *cdata = walker->data;
 169        struct active_request_slot *slot = alt_req->slot;
 170        struct alt_base *tail = cdata->alt;
 171        const char *base = alt_req->base;
 172        const char null_byte = '\0';
 173        char *data;
 174        int i = 0;
 175
 176        if (alt_req->http_specific) {
 177                if (slot->curl_result != CURLE_OK ||
 178                    !alt_req->buffer->len) {
 179
 180                        /* Try reusing the slot to get non-http alternates */
 181                        alt_req->http_specific = 0;
 182                        strbuf_reset(alt_req->url);
 183                        strbuf_addf(alt_req->url, "%s/objects/info/alternates",
 184                                    base);
 185                        curl_easy_setopt(slot->curl, CURLOPT_URL,
 186                                         alt_req->url->buf);
 187                        active_requests++;
 188                        slot->in_use = 1;
 189                        if (slot->finished != NULL)
 190                                (*slot->finished) = 0;
 191                        if (!start_active_slot(slot)) {
 192                                cdata->got_alternates = -1;
 193                                slot->in_use = 0;
 194                                if (slot->finished != NULL)
 195                                        (*slot->finished) = 1;
 196                        }
 197                        return;
 198                }
 199        } else if (slot->curl_result != CURLE_OK) {
 200                if (!missing_target(slot)) {
 201                        cdata->got_alternates = -1;
 202                        return;
 203                }
 204        }
 205
 206        fwrite_buffer((char *)&null_byte, 1, 1, alt_req->buffer);
 207        alt_req->buffer->len--;
 208        data = alt_req->buffer->buf;
 209
 210        while (i < alt_req->buffer->len) {
 211                int posn = i;
 212                while (posn < alt_req->buffer->len && data[posn] != '\n')
 213                        posn++;
 214                if (data[posn] == '\n') {
 215                        int okay = 0;
 216                        int serverlen = 0;
 217                        struct alt_base *newalt;
 218                        if (data[i] == '/') {
 219                                /*
 220                                 * This counts
 221                                 * http://git.host/pub/scm/linux.git/
 222                                 * -----------here^
 223                                 * so memcpy(dst, base, serverlen) will
 224                                 * copy up to "...git.host".
 225                                 */
 226                                const char *colon_ss = strstr(base,"://");
 227                                if (colon_ss) {
 228                                        serverlen = (strchr(colon_ss + 3, '/')
 229                                                     - base);
 230                                        okay = 1;
 231                                }
 232                        } else if (!memcmp(data + i, "../", 3)) {
 233                                /*
 234                                 * Relative URL; chop the corresponding
 235                                 * number of subpath from base (and ../
 236                                 * from data), and concatenate the result.
 237                                 *
 238                                 * The code first drops ../ from data, and
 239                                 * then drops one ../ from data and one path
 240                                 * from base.  IOW, one extra ../ is dropped
 241                                 * from data than path is dropped from base.
 242                                 *
 243                                 * This is not wrong.  The alternate in
 244                                 *     http://git.host/pub/scm/linux.git/
 245                                 * to borrow from
 246                                 *     http://git.host/pub/scm/linus.git/
 247                                 * is ../../linus.git/objects/.  You need
 248                                 * two ../../ to borrow from your direct
 249                                 * neighbour.
 250                                 */
 251                                i += 3;
 252                                serverlen = strlen(base);
 253                                while (i + 2 < posn &&
 254                                       !memcmp(data + i, "../", 3)) {
 255                                        do {
 256                                                serverlen--;
 257                                        } while (serverlen &&
 258                                                 base[serverlen - 1] != '/');
 259                                        i += 3;
 260                                }
 261                                /* If the server got removed, give up. */
 262                                okay = strchr(base, ':') - base + 3 <
 263                                       serverlen;
 264                        } else if (alt_req->http_specific) {
 265                                char *colon = strchr(data + i, ':');
 266                                char *slash = strchr(data + i, '/');
 267                                if (colon && slash && colon < data + posn &&
 268                                    slash < data + posn && colon < slash) {
 269                                        okay = 1;
 270                                }
 271                        }
 272                        /* skip "objects\n" at end */
 273                        if (okay) {
 274                                struct strbuf target = STRBUF_INIT;
 275                                strbuf_add(&target, base, serverlen);
 276                                strbuf_add(&target, data + i, posn - i - 7);
 277                                if (walker->get_verbosely)
 278                                        fprintf(stderr, "Also look at %s\n",
 279                                                target.buf);
 280                                newalt = xmalloc(sizeof(*newalt));
 281                                newalt->next = NULL;
 282                                newalt->base = strbuf_detach(&target, NULL);
 283                                newalt->got_indices = 0;
 284                                newalt->packs = NULL;
 285
 286                                while (tail->next != NULL)
 287                                        tail = tail->next;
 288                                tail->next = newalt;
 289                        }
 290                }
 291                i = posn + 1;
 292        }
 293
 294        cdata->got_alternates = 1;
 295}
 296
 297static void fetch_alternates(struct walker *walker, const char *base)
 298{
 299        struct strbuf buffer = STRBUF_INIT;
 300        struct strbuf url = STRBUF_INIT;
 301        struct active_request_slot *slot;
 302        struct alternates_request alt_req;
 303        struct walker_data *cdata = walker->data;
 304
 305        /*
 306         * If another request has already started fetching alternates,
 307         * wait for them to arrive and return to processing this request's
 308         * curl message
 309         */
 310#ifdef USE_CURL_MULTI
 311        while (cdata->got_alternates == 0) {
 312                step_active_slots();
 313        }
 314#endif
 315
 316        /* Nothing to do if they've already been fetched */
 317        if (cdata->got_alternates == 1)
 318                return;
 319
 320        /* Start the fetch */
 321        cdata->got_alternates = 0;
 322
 323        if (walker->get_verbosely)
 324                fprintf(stderr, "Getting alternates list for %s\n", base);
 325
 326        strbuf_addf(&url, "%s/objects/info/http-alternates", base);
 327
 328        /*
 329         * Use a callback to process the result, since another request
 330         * may fail and need to have alternates loaded before continuing
 331         */
 332        slot = get_active_slot();
 333        slot->callback_func = process_alternates_response;
 334        alt_req.walker = walker;
 335        slot->callback_data = &alt_req;
 336
 337        curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
 338        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
 339        curl_easy_setopt(slot->curl, CURLOPT_URL, url.buf);
 340
 341        alt_req.base = base;
 342        alt_req.url = &url;
 343        alt_req.buffer = &buffer;
 344        alt_req.http_specific = 1;
 345        alt_req.slot = slot;
 346
 347        if (start_active_slot(slot))
 348                run_active_slot(slot);
 349        else
 350                cdata->got_alternates = -1;
 351
 352        strbuf_release(&buffer);
 353        strbuf_release(&url);
 354}
 355
 356static int fetch_indices(struct walker *walker, struct alt_base *repo)
 357{
 358        int ret;
 359
 360        if (repo->got_indices)
 361                return 0;
 362
 363        if (walker->get_verbosely)
 364                fprintf(stderr, "Getting pack list for %s\n", repo->base);
 365
 366        switch (http_get_info_packs(repo->base, &repo->packs)) {
 367        case HTTP_OK:
 368        case HTTP_MISSING_TARGET:
 369                repo->got_indices = 1;
 370                ret = 0;
 371                break;
 372        default:
 373                repo->got_indices = 0;
 374                ret = -1;
 375        }
 376
 377        return ret;
 378}
 379
 380static int http_fetch_pack(struct walker *walker, struct alt_base *repo, unsigned char *sha1)
 381{
 382        struct packed_git *target;
 383        int ret;
 384        struct slot_results results;
 385        struct http_pack_request *preq;
 386
 387        if (fetch_indices(walker, repo))
 388                return -1;
 389        target = find_sha1_pack(sha1, repo->packs);
 390        if (!target)
 391                return -1;
 392
 393        if (walker->get_verbosely) {
 394                fprintf(stderr, "Getting pack %s\n",
 395                        sha1_to_hex(target->sha1));
 396                fprintf(stderr, " which contains %s\n",
 397                        sha1_to_hex(sha1));
 398        }
 399
 400        preq = new_http_pack_request(target, repo->base);
 401        if (preq == NULL)
 402                goto abort;
 403        preq->lst = &repo->packs;
 404        preq->slot->results = &results;
 405
 406        if (start_active_slot(preq->slot)) {
 407                run_active_slot(preq->slot);
 408                if (results.curl_result != CURLE_OK) {
 409                        error("Unable to get pack file %s\n%s", preq->url,
 410                              curl_errorstr);
 411                        goto abort;
 412                }
 413        } else {
 414                error("Unable to start request");
 415                goto abort;
 416        }
 417
 418        ret = finish_http_pack_request(preq);
 419        release_http_pack_request(preq);
 420        if (ret)
 421                return ret;
 422
 423        return 0;
 424
 425abort:
 426        return -1;
 427}
 428
 429static void abort_object_request(struct object_request *obj_req)
 430{
 431        release_object_request(obj_req);
 432}
 433
 434static int fetch_object(struct walker *walker, unsigned char *sha1)
 435{
 436        char *hex = sha1_to_hex(sha1);
 437        int ret = 0;
 438        struct object_request *obj_req = NULL;
 439        struct http_object_request *req;
 440        struct list_head *pos, *head = &object_queue_head;
 441
 442        list_for_each(pos, head) {
 443                obj_req = list_entry(pos, struct object_request, node);
 444                if (!hashcmp(obj_req->sha1, sha1))
 445                        break;
 446        }
 447        if (obj_req == NULL)
 448                return error("Couldn't find request for %s in the queue", hex);
 449
 450        if (has_sha1_file(obj_req->sha1)) {
 451                if (obj_req->req != NULL)
 452                        abort_http_object_request(obj_req->req);
 453                abort_object_request(obj_req);
 454                return 0;
 455        }
 456
 457#ifdef USE_CURL_MULTI
 458        while (obj_req->state == WAITING)
 459                step_active_slots();
 460#else
 461        start_object_request(walker, obj_req);
 462#endif
 463
 464        /*
 465         * obj_req->req might change when fetching alternates in the callback
 466         * process_object_response; therefore, the "shortcut" variable, req,
 467         * is used only after we're done with slots.
 468         */
 469        while (obj_req->state == ACTIVE)
 470                run_active_slot(obj_req->req->slot);
 471
 472        req = obj_req->req;
 473
 474        if (req->localfile != -1) {
 475                close(req->localfile);
 476                req->localfile = -1;
 477        }
 478
 479        /*
 480         * we turned off CURLOPT_FAILONERROR to avoid losing a
 481         * persistent connection and got CURLE_OK.
 482         */
 483        if (req->http_code == 404 && req->curl_result == CURLE_OK &&
 484                        (starts_with(req->url, "http://") ||
 485                         starts_with(req->url, "https://")))
 486                req->curl_result = CURLE_HTTP_RETURNED_ERROR;
 487
 488        if (obj_req->state == ABORTED) {
 489                ret = error("Request for %s aborted", hex);
 490        } else if (req->curl_result != CURLE_OK &&
 491                   req->http_code != 416) {
 492                if (missing_target(req))
 493                        ret = -1; /* Be silent, it is probably in a pack. */
 494                else
 495                        ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
 496                                    req->errorstr, req->curl_result,
 497                                    req->http_code, hex);
 498        } else if (req->zret != Z_STREAM_END) {
 499                walker->corrupt_object_found++;
 500                ret = error("File %s (%s) corrupt", hex, req->url);
 501        } else if (hashcmp(obj_req->sha1, req->real_sha1)) {
 502                ret = error("File %s has bad hash", hex);
 503        } else if (req->rename < 0) {
 504                ret = error("unable to write sha1 filename %s",
 505                            sha1_file_name(req->sha1));
 506        }
 507
 508        release_http_object_request(req);
 509        release_object_request(obj_req);
 510        return ret;
 511}
 512
 513static int fetch(struct walker *walker, unsigned char *sha1)
 514{
 515        struct walker_data *data = walker->data;
 516        struct alt_base *altbase = data->alt;
 517
 518        if (!fetch_object(walker, sha1))
 519                return 0;
 520        while (altbase) {
 521                if (!http_fetch_pack(walker, altbase, sha1))
 522                        return 0;
 523                fetch_alternates(walker, data->alt->base);
 524                altbase = altbase->next;
 525        }
 526        return error("Unable to find %s under %s", sha1_to_hex(sha1),
 527                     data->alt->base);
 528}
 529
 530static int fetch_ref(struct walker *walker, struct ref *ref)
 531{
 532        struct walker_data *data = walker->data;
 533        return http_fetch_ref(data->alt->base, ref);
 534}
 535
 536static void cleanup(struct walker *walker)
 537{
 538        struct walker_data *data = walker->data;
 539        struct alt_base *alt, *alt_next;
 540
 541        if (data) {
 542                alt = data->alt;
 543                while (alt) {
 544                        alt_next = alt->next;
 545
 546                        free(alt->base);
 547                        free(alt);
 548
 549                        alt = alt_next;
 550                }
 551                free(data);
 552                walker->data = NULL;
 553        }
 554}
 555
 556struct walker *get_http_walker(const char *url)
 557{
 558        char *s;
 559        struct walker_data *data = xmalloc(sizeof(struct walker_data));
 560        struct walker *walker = xmalloc(sizeof(struct walker));
 561
 562        data->alt = xmalloc(sizeof(*data->alt));
 563        data->alt->base = xstrdup(url);
 564        for (s = data->alt->base + strlen(data->alt->base) - 1; *s == '/'; --s)
 565                *s = 0;
 566
 567        data->alt->got_indices = 0;
 568        data->alt->packs = NULL;
 569        data->alt->next = NULL;
 570        data->got_alternates = -1;
 571
 572        walker->corrupt_object_found = 0;
 573        walker->fetch = fetch;
 574        walker->fetch_ref = fetch_ref;
 575        walker->prefetch = prefetch;
 576        walker->cleanup = cleanup;
 577        walker->data = data;
 578
 579#ifdef USE_CURL_MULTI
 580        add_fill_function(walker, (int (*)(void *)) fill_active_slot);
 581#endif
 582
 583        return walker;
 584}