http-fetch.con commit Fix bunch of fd leaks in http-fetch (313c471)
   1#include "cache.h"
   2#include "commit.h"
   3#include "pack.h"
   4#include "fetch.h"
   5
   6#include <curl/curl.h>
   7#include <curl/easy.h>
   8
   9#if LIBCURL_VERSION_NUM >= 0x070908
  10#define USE_CURL_MULTI
  11#define DEFAULT_MAX_REQUESTS 5
  12#endif
  13
  14#if LIBCURL_VERSION_NUM < 0x070704
  15#define curl_global_cleanup() do { /* nothing */ } while(0)
  16#endif
  17#if LIBCURL_VERSION_NUM < 0x070800
  18#define curl_global_init(a) do { /* nothing */ } while(0)
  19#endif
  20
  21#if LIBCURL_VERSION_NUM < 0x070c04
  22#define NO_CURL_EASY_DUPHANDLE
  23#endif
  24
  25#define PREV_BUF_SIZE 4096
  26#define RANGE_HEADER_SIZE 30
  27
  28static int got_alternates = -1;
  29static int active_requests = 0;
  30static int data_received;
  31
  32#ifdef USE_CURL_MULTI
  33static int max_requests = -1;
  34static CURLM *curlm;
  35#endif
  36#ifndef NO_CURL_EASY_DUPHANDLE
  37static CURL *curl_default;
  38#endif
  39static struct curl_slist *pragma_header;
  40static struct curl_slist *no_pragma_header;
  41static struct curl_slist *no_range_header;
  42static char curl_errorstr[CURL_ERROR_SIZE];
  43
  44struct alt_base
  45{
  46        char *base;
  47        int got_indices;
  48        struct packed_git *packs;
  49        struct alt_base *next;
  50};
  51
  52static struct alt_base *alt = NULL;
  53
  54enum transfer_state {
  55        WAITING,
  56        ABORTED,
  57        ACTIVE,
  58        COMPLETE,
  59};
  60
  61struct transfer_request
  62{
  63        unsigned char sha1[20];
  64        struct alt_base *repo;
  65        char *url;
  66        char filename[PATH_MAX];
  67        char tmpfile[PATH_MAX];
  68        int local;
  69        enum transfer_state state;
  70        CURLcode curl_result;
  71        char errorstr[CURL_ERROR_SIZE];
  72        long http_code;
  73        unsigned char real_sha1[20];
  74        SHA_CTX c;
  75        z_stream stream;
  76        int zret;
  77        int rename;
  78        struct active_request_slot *slot;
  79        struct transfer_request *next;
  80};
  81
  82struct active_request_slot
  83{
  84        CURL *curl;
  85        FILE *local;
  86        int in_use;
  87        int done;
  88        CURLcode curl_result;
  89        long http_code;
  90        void *callback_data;
  91        void (*callback_func)(void *data);
  92        struct active_request_slot *next;
  93};
  94
  95struct alt_request {
  96        char *base;
  97        char *url;
  98        struct buffer *buffer;
  99        struct active_request_slot *slot;
 100        int http_specific;
 101};
 102
 103static struct transfer_request *request_queue_head = NULL;
 104static struct active_request_slot *active_queue_head = NULL;
 105
 106static int curl_ssl_verify = -1;
 107static char *ssl_cert = NULL;
 108#if LIBCURL_VERSION_NUM >= 0x070902
 109static char *ssl_key = NULL;
 110#endif
 111#if LIBCURL_VERSION_NUM >= 0x070908
 112static char *ssl_capath = NULL;
 113#endif
 114static char *ssl_cainfo = NULL;
 115static long curl_low_speed_limit = -1;
 116static long curl_low_speed_time = -1;
 117
 118struct buffer
 119{
 120        size_t posn;
 121        size_t size;
 122        void *buffer;
 123};
 124
 125static int http_options(const char *var, const char *value)
 126{
 127        if (!strcmp("http.sslverify", var)) {
 128                if (curl_ssl_verify == -1) {
 129                        curl_ssl_verify = git_config_bool(var, value);
 130                }
 131                return 0;
 132        }
 133
 134        if (!strcmp("http.sslcert", var)) {
 135                if (ssl_cert == NULL) {
 136                        ssl_cert = xmalloc(strlen(value)+1);
 137                        strcpy(ssl_cert, value);
 138                }
 139                return 0;
 140        }
 141#if LIBCURL_VERSION_NUM >= 0x070902
 142        if (!strcmp("http.sslkey", var)) {
 143                if (ssl_key == NULL) {
 144                        ssl_key = xmalloc(strlen(value)+1);
 145                        strcpy(ssl_key, value);
 146                }
 147                return 0;
 148        }
 149#endif
 150#if LIBCURL_VERSION_NUM >= 0x070908
 151        if (!strcmp("http.sslcapath", var)) {
 152                if (ssl_capath == NULL) {
 153                        ssl_capath = xmalloc(strlen(value)+1);
 154                        strcpy(ssl_capath, value);
 155                }
 156                return 0;
 157        }
 158#endif
 159        if (!strcmp("http.sslcainfo", var)) {
 160                if (ssl_cainfo == NULL) {
 161                        ssl_cainfo = xmalloc(strlen(value)+1);
 162                        strcpy(ssl_cainfo, value);
 163                }
 164                return 0;
 165        }
 166
 167#ifdef USE_CURL_MULTI   
 168        if (!strcmp("http.maxrequests", var)) {
 169                if (max_requests == -1)
 170                        max_requests = git_config_int(var, value);
 171                return 0;
 172        }
 173#endif
 174
 175        if (!strcmp("http.lowspeedlimit", var)) {
 176                if (curl_low_speed_limit == -1)
 177                        curl_low_speed_limit = (long)git_config_int(var, value);
 178                return 0;
 179        }
 180        if (!strcmp("http.lowspeedtime", var)) {
 181                if (curl_low_speed_time == -1)
 182                        curl_low_speed_time = (long)git_config_int(var, value);
 183                return 0;
 184        }
 185
 186        /* Fall back on the default ones */
 187        return git_default_config(var, value);
 188}
 189
 190static size_t fwrite_buffer(void *ptr, size_t eltsize, size_t nmemb,
 191                            struct buffer *buffer)
 192{
 193        size_t size = eltsize * nmemb;
 194        if (size > buffer->size - buffer->posn)
 195                size = buffer->size - buffer->posn;
 196        memcpy(buffer->buffer + buffer->posn, ptr, size);
 197        buffer->posn += size;
 198        data_received++;
 199        return size;
 200}
 201
 202static size_t fwrite_buffer_dynamic(const void *ptr, size_t eltsize,
 203                                    size_t nmemb, struct buffer *buffer)
 204{
 205        size_t size = eltsize * nmemb;
 206        if (size > buffer->size - buffer->posn) {
 207                buffer->size = buffer->size * 3 / 2;
 208                if (buffer->size < buffer->posn + size)
 209                        buffer->size = buffer->posn + size;
 210                buffer->buffer = xrealloc(buffer->buffer, buffer->size);
 211        }
 212        memcpy(buffer->buffer + buffer->posn, ptr, size);
 213        buffer->posn += size;
 214        data_received++;
 215        return size;
 216}
 217
 218static size_t fwrite_sha1_file(void *ptr, size_t eltsize, size_t nmemb,
 219                               void *data)
 220{
 221        unsigned char expn[4096];
 222        size_t size = eltsize * nmemb;
 223        int posn = 0;
 224        struct transfer_request *request = (struct transfer_request *)data;
 225        do {
 226                ssize_t retval = write(request->local,
 227                                       ptr + posn, size - posn);
 228                if (retval < 0)
 229                        return posn;
 230                posn += retval;
 231        } while (posn < size);
 232
 233        request->stream.avail_in = size;
 234        request->stream.next_in = ptr;
 235        do {
 236                request->stream.next_out = expn;
 237                request->stream.avail_out = sizeof(expn);
 238                request->zret = inflate(&request->stream, Z_SYNC_FLUSH);
 239                SHA1_Update(&request->c, expn,
 240                            sizeof(expn) - request->stream.avail_out);
 241        } while (request->stream.avail_in && request->zret == Z_OK);
 242        data_received++;
 243        return size;
 244}
 245
 246#ifdef USE_CURL_MULTI
 247static void process_curl_messages(void);
 248static void process_request_queue(void);
 249#endif
 250static void fetch_alternates(char *base);
 251
 252static CURL* get_curl_handle(void)
 253{
 254        CURL* result = curl_easy_init();
 255
 256        curl_easy_setopt(result, CURLOPT_SSL_VERIFYPEER, curl_ssl_verify);
 257#if LIBCURL_VERSION_NUM >= 0x070907
 258        curl_easy_setopt(result, CURLOPT_NETRC, CURL_NETRC_OPTIONAL);
 259#endif
 260
 261        if (ssl_cert != NULL)
 262                curl_easy_setopt(result, CURLOPT_SSLCERT, ssl_cert);
 263#if LIBCURL_VERSION_NUM >= 0x070902
 264        if (ssl_key != NULL)
 265                curl_easy_setopt(result, CURLOPT_SSLKEY, ssl_key);
 266#endif
 267#if LIBCURL_VERSION_NUM >= 0x070908
 268        if (ssl_capath != NULL)
 269                curl_easy_setopt(result, CURLOPT_CAPATH, ssl_capath);
 270#endif
 271        if (ssl_cainfo != NULL)
 272                curl_easy_setopt(result, CURLOPT_CAINFO, ssl_cainfo);
 273        curl_easy_setopt(result, CURLOPT_FAILONERROR, 1);
 274
 275        if (curl_low_speed_limit > 0 && curl_low_speed_time > 0) {
 276                curl_easy_setopt(result, CURLOPT_LOW_SPEED_LIMIT,
 277                                 curl_low_speed_limit);
 278                curl_easy_setopt(result, CURLOPT_LOW_SPEED_TIME,
 279                                 curl_low_speed_time);
 280        }
 281
 282        curl_easy_setopt(result, CURLOPT_FOLLOWLOCATION, 1);
 283
 284        return result;
 285}
 286
 287static struct active_request_slot *get_active_slot(void)
 288{
 289        struct active_request_slot *slot = active_queue_head;
 290        struct active_request_slot *newslot;
 291
 292#ifdef USE_CURL_MULTI
 293        int num_transfers;
 294
 295        /* Wait for a slot to open up if the queue is full */
 296        while (active_requests >= max_requests) {
 297                curl_multi_perform(curlm, &num_transfers);
 298                if (num_transfers < active_requests) {
 299                        process_curl_messages();
 300                }
 301        }
 302#endif
 303
 304        while (slot != NULL && slot->in_use) {
 305                slot = slot->next;
 306        }
 307        if (slot == NULL) {
 308                newslot = xmalloc(sizeof(*newslot));
 309                newslot->curl = NULL;
 310                newslot->in_use = 0;
 311                newslot->next = NULL;
 312
 313                slot = active_queue_head;
 314                if (slot == NULL) {
 315                        active_queue_head = newslot;
 316                } else {
 317                        while (slot->next != NULL) {
 318                                slot = slot->next;
 319                        }
 320                        slot->next = newslot;
 321                }
 322                slot = newslot;
 323        }
 324
 325        if (slot->curl == NULL) {
 326#ifdef NO_CURL_EASY_DUPHANDLE
 327                slot->curl = get_curl_handle();
 328#else
 329                slot->curl = curl_easy_duphandle(curl_default);
 330#endif
 331        }
 332
 333        active_requests++;
 334        slot->in_use = 1;
 335        slot->done = 0;
 336        slot->local = NULL;
 337        slot->callback_data = NULL;
 338        slot->callback_func = NULL;
 339        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, pragma_header);
 340        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_range_header);
 341        curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, curl_errorstr);
 342
 343        return slot;
 344}
 345
 346static int start_active_slot(struct active_request_slot *slot)
 347{
 348#ifdef USE_CURL_MULTI
 349        CURLMcode curlm_result = curl_multi_add_handle(curlm, slot->curl);
 350
 351        if (curlm_result != CURLM_OK &&
 352            curlm_result != CURLM_CALL_MULTI_PERFORM) {
 353                active_requests--;
 354                slot->in_use = 0;
 355                return 0;
 356        }
 357#endif
 358        return 1;
 359}
 360
 361static void run_active_slot(struct active_request_slot *slot)
 362{
 363#ifdef USE_CURL_MULTI
 364        int num_transfers;
 365        long last_pos = 0;
 366        long current_pos;
 367        fd_set readfds;
 368        fd_set writefds;
 369        fd_set excfds;
 370        int max_fd;
 371        struct timeval select_timeout;
 372        CURLMcode curlm_result;
 373
 374        while (!slot->done) {
 375                data_received = 0;
 376                do {
 377                        curlm_result = curl_multi_perform(curlm,
 378                                                          &num_transfers);
 379                } while (curlm_result == CURLM_CALL_MULTI_PERFORM);
 380                if (num_transfers < active_requests) {
 381                        process_curl_messages();
 382                        process_request_queue();
 383                }
 384
 385                if (!data_received && slot->local != NULL) {
 386                        current_pos = ftell(slot->local);
 387                        if (current_pos > last_pos)
 388                                data_received++;
 389                        last_pos = current_pos;
 390                }
 391
 392                if (!slot->done && !data_received) {
 393                        max_fd = 0;
 394                        FD_ZERO(&readfds);
 395                        FD_ZERO(&writefds);
 396                        FD_ZERO(&excfds);
 397                        select_timeout.tv_sec = 0;
 398                        select_timeout.tv_usec = 50000;
 399                        select(max_fd, &readfds, &writefds,
 400                               &excfds, &select_timeout);
 401                }
 402        }
 403#else
 404        slot->curl_result = curl_easy_perform(slot->curl);
 405        active_requests--;
 406#endif
 407}
 408
 409static void start_request(struct transfer_request *request)
 410{
 411        char *hex = sha1_to_hex(request->sha1);
 412        char prevfile[PATH_MAX];
 413        char *url;
 414        char *posn;
 415        int prevlocal;
 416        unsigned char prev_buf[PREV_BUF_SIZE];
 417        ssize_t prev_read = 0;
 418        long prev_posn = 0;
 419        char range[RANGE_HEADER_SIZE];
 420        struct curl_slist *range_header = NULL;
 421        struct active_request_slot *slot;
 422
 423        snprintf(prevfile, sizeof(prevfile), "%s.prev", request->filename);
 424        unlink(prevfile);
 425        rename(request->tmpfile, prevfile);
 426        unlink(request->tmpfile);
 427
 428        if (request->local != -1)
 429                error("fd leakage in start: %d", request->local);
 430        request->local = open(request->tmpfile,
 431                              O_WRONLY | O_CREAT | O_EXCL, 0666);
 432        /* This could have failed due to the "lazy directory creation";
 433         * try to mkdir the last path component.
 434         */
 435        if (request->local < 0 && errno == ENOENT) {
 436                char *dir = strrchr(request->tmpfile, '/');
 437                if (dir) {
 438                        *dir = 0;
 439                        mkdir(request->tmpfile, 0777);
 440                        *dir = '/';
 441                }
 442                request->local = open(request->tmpfile,
 443                                      O_WRONLY | O_CREAT | O_EXCL, 0666);
 444        }
 445
 446        if (request->local < 0) {
 447                request->state = ABORTED;
 448                error("Couldn't create temporary file %s for %s: %s\n",
 449                      request->tmpfile, request->filename, strerror(errno));
 450                return;
 451        }
 452
 453        memset(&request->stream, 0, sizeof(request->stream));
 454
 455        inflateInit(&request->stream);
 456
 457        SHA1_Init(&request->c);
 458
 459        url = xmalloc(strlen(request->repo->base) + 50);
 460        request->url = xmalloc(strlen(request->repo->base) + 50);
 461        strcpy(url, request->repo->base);
 462        posn = url + strlen(request->repo->base);
 463        strcpy(posn, "objects/");
 464        posn += 8;
 465        memcpy(posn, hex, 2);
 466        posn += 2;
 467        *(posn++) = '/';
 468        strcpy(posn, hex + 2);
 469        strcpy(request->url, url);
 470
 471        /* If a previous temp file is present, process what was already
 472           fetched. */
 473        prevlocal = open(prevfile, O_RDONLY);
 474        if (prevlocal != -1) {
 475                do {
 476                        prev_read = read(prevlocal, prev_buf, PREV_BUF_SIZE);
 477                        if (prev_read>0) {
 478                                if (fwrite_sha1_file(prev_buf,
 479                                                     1,
 480                                                     prev_read,
 481                                                     request) == prev_read) {
 482                                        prev_posn += prev_read;
 483                                } else {
 484                                        prev_read = -1;
 485                                }
 486                        }
 487                } while (prev_read > 0);
 488                close(prevlocal);
 489        }
 490        unlink(prevfile);
 491
 492        /* Reset inflate/SHA1 if there was an error reading the previous temp
 493           file; also rewind to the beginning of the local file. */
 494        if (prev_read == -1) {
 495                memset(&request->stream, 0, sizeof(request->stream));
 496                inflateInit(&request->stream);
 497                SHA1_Init(&request->c);
 498                if (prev_posn>0) {
 499                        prev_posn = 0;
 500                        lseek(request->local, SEEK_SET, 0);
 501                        ftruncate(request->local, 0);
 502                }
 503        }
 504
 505        slot = get_active_slot();
 506        curl_easy_setopt(slot->curl, CURLOPT_FILE, request);
 507        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_sha1_file);
 508        curl_easy_setopt(slot->curl, CURLOPT_ERRORBUFFER, request->errorstr);
 509        curl_easy_setopt(slot->curl, CURLOPT_URL, url);
 510        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
 511
 512        /* If we have successfully processed data from a previous fetch
 513           attempt, only fetch the data we don't already have. */
 514        if (prev_posn>0) {
 515                if (get_verbosely)
 516                        fprintf(stderr,
 517                                "Resuming fetch of object %s at byte %ld\n",
 518                                hex, prev_posn);
 519                sprintf(range, "Range: bytes=%ld-", prev_posn);
 520                range_header = curl_slist_append(range_header, range);
 521                curl_easy_setopt(slot->curl,
 522                                 CURLOPT_HTTPHEADER, range_header);
 523        }
 524
 525        /* Try to get the request started, abort the request on error */
 526        if (!start_active_slot(slot)) {
 527                request->state = ABORTED;
 528                close(request->local); request->local = -1;
 529                free(request->url);
 530                return;
 531        }
 532        
 533        request->slot = slot;
 534        request->state = ACTIVE;
 535}
 536
 537static void finish_request(struct transfer_request *request)
 538{
 539        struct stat st;
 540
 541        fchmod(request->local, 0444);
 542        close(request->local); request->local = -1;
 543
 544        if (request->http_code == 416) {
 545                fprintf(stderr, "Warning: requested range invalid; we may already have all the data.\n");
 546        } else if (request->curl_result != CURLE_OK) {
 547                if (stat(request->tmpfile, &st) == 0)
 548                        if (st.st_size == 0)
 549                                unlink(request->tmpfile);
 550                return;
 551        }
 552
 553        inflateEnd(&request->stream);
 554        SHA1_Final(request->real_sha1, &request->c);
 555        if (request->zret != Z_STREAM_END) {
 556                unlink(request->tmpfile);
 557                return;
 558        }
 559        if (memcmp(request->sha1, request->real_sha1, 20)) {
 560                unlink(request->tmpfile);
 561                return;
 562        }
 563        request->rename =
 564                move_temp_to_file(request->tmpfile, request->filename);
 565
 566        if (request->rename == 0)
 567                pull_say("got %s\n", sha1_to_hex(request->sha1));
 568}
 569
 570static void release_request(struct transfer_request *request)
 571{
 572        struct transfer_request *entry = request_queue_head;
 573
 574        if (request->local != -1)
 575                error("fd leakage in release: %d", request->local);
 576        if (request == request_queue_head) {
 577                request_queue_head = request->next;
 578        } else {
 579                while (entry->next != NULL && entry->next != request)
 580                        entry = entry->next;
 581                if (entry->next == request)
 582                        entry->next = entry->next->next;
 583        }
 584
 585        free(request->url);
 586        free(request);
 587}
 588
 589#ifdef USE_CURL_MULTI
 590static void process_curl_messages(void)
 591{
 592        int num_messages;
 593        struct active_request_slot *slot;
 594        struct transfer_request *request = NULL;
 595        CURLMsg *curl_message = curl_multi_info_read(curlm, &num_messages);
 596
 597        while (curl_message != NULL) {
 598                if (curl_message->msg == CURLMSG_DONE) {
 599                        int curl_result = curl_message->data.result;
 600                        slot = active_queue_head;
 601                        while (slot != NULL &&
 602                               slot->curl != curl_message->easy_handle)
 603                                slot = slot->next;
 604                        if (slot != NULL) {
 605                                curl_multi_remove_handle(curlm, slot->curl);
 606                                active_requests--;
 607                                slot->done = 1;
 608                                slot->in_use = 0;
 609                                slot->curl_result = curl_result;
 610                                curl_easy_getinfo(slot->curl,
 611                                                  CURLINFO_HTTP_CODE,
 612                                                  &slot->http_code);
 613                                request = request_queue_head;
 614                                while (request != NULL &&
 615                                       request->slot != slot)
 616                                        request = request->next;
 617                        } else {
 618                                fprintf(stderr, "Received DONE message for unknown request!\n");
 619                        }
 620
 621                        /* Process slot callback if appropriate */
 622                        if (slot->callback_func != NULL) {
 623                                slot->callback_func(slot->callback_data);
 624                        }
 625
 626                        if (request != NULL) {
 627                                request->curl_result = curl_result;
 628                                request->http_code = slot->http_code;
 629                                request->slot = NULL;
 630                                request->state = COMPLETE;
 631
 632                                /* Use alternates if necessary */
 633                                if (request->http_code == 404) {
 634                                        fetch_alternates(alt->base);
 635                                        if (request->repo->next != NULL) {
 636                                                request->repo =
 637                                                        request->repo->next;
 638                                                close(request->local);
 639                                                        request->local = -1;
 640                                                start_request(request);
 641                                        } else {
 642                                                finish_request(request);
 643                                        }
 644                                } else {
 645                                        finish_request(request);
 646                                }
 647                        }
 648                } else {
 649                        fprintf(stderr, "Unknown CURL message received: %d\n",
 650                                (int)curl_message->msg);
 651                }
 652                curl_message = curl_multi_info_read(curlm, &num_messages);
 653        }
 654}
 655
 656static void process_request_queue(void)
 657{
 658        struct transfer_request *request = request_queue_head;
 659        struct active_request_slot *slot = active_queue_head;
 660        int num_transfers;
 661
 662        while (active_requests < max_requests && request != NULL) {
 663                if (request->state == WAITING) {
 664                        if (has_sha1_file(request->sha1))
 665                                release_request(request);
 666                        else
 667                                start_request(request);
 668                        curl_multi_perform(curlm, &num_transfers);
 669                }
 670                request = request->next;
 671        }
 672
 673        while (slot != NULL) {
 674                if (!slot->in_use && slot->curl != NULL) {
 675                        curl_easy_cleanup(slot->curl);
 676                        slot->curl = NULL;
 677                }
 678                slot = slot->next;
 679        }                               
 680}
 681#endif
 682
 683void prefetch(unsigned char *sha1)
 684{
 685        struct transfer_request *newreq;
 686        struct transfer_request *tail;
 687        char *filename = sha1_file_name(sha1);
 688
 689        newreq = xmalloc(sizeof(*newreq));
 690        memcpy(newreq->sha1, sha1, 20);
 691        newreq->repo = alt;
 692        newreq->url = NULL;
 693        newreq->local = -1;
 694        newreq->state = WAITING;
 695        snprintf(newreq->filename, sizeof(newreq->filename), "%s", filename);
 696        snprintf(newreq->tmpfile, sizeof(newreq->tmpfile),
 697                 "%s.temp", filename);
 698        newreq->next = NULL;
 699
 700        if (request_queue_head == NULL) {
 701                request_queue_head = newreq;
 702        } else {
 703                tail = request_queue_head;
 704                while (tail->next != NULL) {
 705                        tail = tail->next;
 706                }
 707                tail->next = newreq;
 708        }
 709#ifdef USE_CURL_MULTI
 710        process_request_queue();
 711        process_curl_messages();
 712#endif
 713}
 714
 715static int fetch_index(struct alt_base *repo, unsigned char *sha1)
 716{
 717        char *hex = sha1_to_hex(sha1);
 718        char *filename;
 719        char *url;
 720        char tmpfile[PATH_MAX];
 721        long prev_posn = 0;
 722        char range[RANGE_HEADER_SIZE];
 723        struct curl_slist *range_header = NULL;
 724
 725        FILE *indexfile;
 726        struct active_request_slot *slot;
 727
 728        if (has_pack_index(sha1))
 729                return 0;
 730
 731        if (get_verbosely)
 732                fprintf(stderr, "Getting index for pack %s\n", hex);
 733        
 734        url = xmalloc(strlen(repo->base) + 64);
 735        sprintf(url, "%s/objects/pack/pack-%s.idx", repo->base, hex);
 736        
 737        filename = sha1_pack_index_name(sha1);
 738        snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
 739        indexfile = fopen(tmpfile, "a");
 740        if (!indexfile)
 741                return error("Unable to open local file %s for pack index",
 742                             filename);
 743
 744        slot = get_active_slot();
 745        curl_easy_setopt(slot->curl, CURLOPT_FILE, indexfile);
 746        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
 747        curl_easy_setopt(slot->curl, CURLOPT_URL, url);
 748        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
 749        slot->local = indexfile;
 750
 751        /* If there is data present from a previous transfer attempt,
 752           resume where it left off */
 753        prev_posn = ftell(indexfile);
 754        if (prev_posn>0) {
 755                if (get_verbosely)
 756                        fprintf(stderr,
 757                                "Resuming fetch of index for pack %s at byte %ld\n",
 758                                hex, prev_posn);
 759                sprintf(range, "Range: bytes=%ld-", prev_posn);
 760                range_header = curl_slist_append(range_header, range);
 761                curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
 762        }
 763
 764        if (start_active_slot(slot)) {
 765                run_active_slot(slot);
 766                if (slot->curl_result != CURLE_OK) {
 767                        fclose(indexfile);
 768                        return error("Unable to get pack index %s\n%s", url,
 769                                     curl_errorstr);
 770                }
 771        } else {
 772                fclose(indexfile);
 773                return error("Unable to start request");
 774        }
 775
 776        fclose(indexfile);
 777
 778        return move_temp_to_file(tmpfile, filename);
 779}
 780
 781static int setup_index(struct alt_base *repo, unsigned char *sha1)
 782{
 783        struct packed_git *new_pack;
 784        if (has_pack_file(sha1))
 785                return 0; // don't list this as something we can get
 786
 787        if (fetch_index(repo, sha1))
 788                return -1;
 789
 790        new_pack = parse_pack_index(sha1);
 791        new_pack->next = repo->packs;
 792        repo->packs = new_pack;
 793        return 0;
 794}
 795
 796static void process_alternates(void *callback_data)
 797{
 798        struct alt_request *alt_req = (struct alt_request *)callback_data;
 799        struct active_request_slot *slot = alt_req->slot;
 800        struct alt_base *tail = alt;
 801        char *base = alt_req->base;
 802        static const char null_byte = '\0';
 803        char *data;
 804        int i = 0;
 805
 806        if (alt_req->http_specific) {
 807                if (slot->curl_result != CURLE_OK ||
 808                    !alt_req->buffer->posn) {
 809
 810                        /* Try reusing the slot to get non-http alternates */
 811                        alt_req->http_specific = 0;
 812                        sprintf(alt_req->url, "%s/objects/info/alternates",
 813                                base);
 814                        curl_easy_setopt(slot->curl, CURLOPT_URL,
 815                                         alt_req->url);
 816                        active_requests++;
 817                        slot->in_use = 1;
 818                        slot->done = 0;
 819                        if (start_active_slot(slot)) {
 820                                return;
 821                        } else {
 822                                got_alternates = -1;
 823                                slot->done = 1;
 824                                return;
 825                        }
 826                }
 827        } else if (slot->curl_result != CURLE_OK) {
 828                if (slot->http_code != 404) {
 829                        got_alternates = -1;
 830                        return;
 831                }
 832        }
 833
 834        fwrite_buffer_dynamic(&null_byte, 1, 1, alt_req->buffer);
 835        alt_req->buffer->posn--;
 836        data = alt_req->buffer->buffer;
 837
 838        while (i < alt_req->buffer->posn) {
 839                int posn = i;
 840                while (posn < alt_req->buffer->posn && data[posn] != '\n')
 841                        posn++;
 842                if (data[posn] == '\n') {
 843                        int okay = 0;
 844                        int serverlen = 0;
 845                        struct alt_base *newalt;
 846                        char *target = NULL;
 847                        if (data[i] == '/') {
 848                                serverlen = strchr(base + 8, '/') - base;
 849                                okay = 1;
 850                        } else if (!memcmp(data + i, "../", 3)) {
 851                                i += 3;
 852                                serverlen = strlen(base);
 853                                while (i + 2 < posn && 
 854                                       !memcmp(data + i, "../", 3)) {
 855                                        do {
 856                                                serverlen--;
 857                                        } while (serverlen &&
 858                                                 base[serverlen - 1] != '/');
 859                                        i += 3;
 860                                }
 861                                // If the server got removed, give up.
 862                                okay = strchr(base, ':') - base + 3 < 
 863                                        serverlen;
 864                        } else if (alt_req->http_specific) {
 865                                char *colon = strchr(data + i, ':');
 866                                char *slash = strchr(data + i, '/');
 867                                if (colon && slash && colon < data + posn &&
 868                                    slash < data + posn && colon < slash) {
 869                                        okay = 1;
 870                                }
 871                        }
 872                        // skip 'objects' at end
 873                        if (okay) {
 874                                target = xmalloc(serverlen + posn - i - 6);
 875                                strncpy(target, base, serverlen);
 876                                strncpy(target + serverlen, data + i,
 877                                        posn - i - 7);
 878                                target[serverlen + posn - i - 7] = '\0';
 879                                if (get_verbosely)
 880                                        fprintf(stderr, 
 881                                                "Also look at %s\n", target);
 882                                newalt = xmalloc(sizeof(*newalt));
 883                                newalt->next = NULL;
 884                                newalt->base = target;
 885                                newalt->got_indices = 0;
 886                                newalt->packs = NULL;
 887                                while (tail->next != NULL)
 888                                        tail = tail->next;
 889                                tail->next = newalt;
 890                        }
 891                }
 892                i = posn + 1;
 893        }
 894
 895        got_alternates = 1;
 896}
 897
 898static void fetch_alternates(char *base)
 899{
 900        struct buffer buffer;
 901        char *url;
 902        char *data;
 903        struct active_request_slot *slot;
 904        static struct alt_request alt_req;
 905        int num_transfers;
 906
 907        /* If another request has already started fetching alternates,
 908           wait for them to arrive and return to processing this request's
 909           curl message */
 910        while (got_alternates == 0) {
 911                curl_multi_perform(curlm, &num_transfers);
 912                process_curl_messages();
 913                process_request_queue();
 914        }
 915
 916        /* Nothing to do if they've already been fetched */
 917        if (got_alternates == 1)
 918                return;
 919
 920        /* Start the fetch */
 921        got_alternates = 0;
 922
 923        data = xmalloc(4096);
 924        buffer.size = 4096;
 925        buffer.posn = 0;
 926        buffer.buffer = data;
 927
 928        if (get_verbosely)
 929                fprintf(stderr, "Getting alternates list for %s\n", base);
 930        
 931        url = xmalloc(strlen(base) + 31);
 932        sprintf(url, "%s/objects/info/http-alternates", base);
 933
 934        /* Use a callback to process the result, since another request
 935           may fail and need to have alternates loaded before continuing */
 936        slot = get_active_slot();
 937        slot->callback_func = process_alternates;
 938        slot->callback_data = &alt_req;
 939
 940        curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
 941        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
 942                         fwrite_buffer_dynamic);
 943        curl_easy_setopt(slot->curl, CURLOPT_URL, url);
 944
 945        alt_req.base = base;
 946        alt_req.url = url;
 947        alt_req.buffer = &buffer;
 948        alt_req.http_specific = 1;
 949        alt_req.slot = slot;
 950
 951        if (start_active_slot(slot))
 952                run_active_slot(slot);
 953        else
 954                got_alternates = -1;
 955
 956        free(data);
 957        free(url);
 958}
 959
 960static int fetch_indices(struct alt_base *repo)
 961{
 962        unsigned char sha1[20];
 963        char *url;
 964        struct buffer buffer;
 965        char *data;
 966        int i = 0;
 967
 968        struct active_request_slot *slot;
 969
 970        if (repo->got_indices)
 971                return 0;
 972
 973        data = xmalloc(4096);
 974        buffer.size = 4096;
 975        buffer.posn = 0;
 976        buffer.buffer = data;
 977
 978        if (get_verbosely)
 979                fprintf(stderr, "Getting pack list for %s\n", repo->base);
 980        
 981        url = xmalloc(strlen(repo->base) + 21);
 982        sprintf(url, "%s/objects/info/packs", repo->base);
 983
 984        slot = get_active_slot();
 985        curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
 986        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION,
 987                         fwrite_buffer_dynamic);
 988        curl_easy_setopt(slot->curl, CURLOPT_URL, url);
 989        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
 990        if (start_active_slot(slot)) {
 991                run_active_slot(slot);
 992                if (slot->curl_result != CURLE_OK) {
 993                        free(buffer.buffer);
 994                        return error("%s", curl_errorstr);
 995                }
 996        } else {
 997                free(buffer.buffer);
 998                return error("Unable to start request");
 999        }
1000
1001        data = buffer.buffer;
1002        while (i < buffer.posn) {
1003                switch (data[i]) {
1004                case 'P':
1005                        i++;
1006                        if (i + 52 < buffer.posn &&
1007                            !strncmp(data + i, " pack-", 6) &&
1008                            !strncmp(data + i + 46, ".pack\n", 6)) {
1009                                get_sha1_hex(data + i + 6, sha1);
1010                                setup_index(repo, sha1);
1011                                i += 51;
1012                                break;
1013                        }
1014                default:
1015                        while (data[i] != '\n')
1016                                i++;
1017                }
1018                i++;
1019        }
1020
1021        free(buffer.buffer);
1022        repo->got_indices = 1;
1023        return 0;
1024}
1025
1026static int fetch_pack(struct alt_base *repo, unsigned char *sha1)
1027{
1028        char *url;
1029        struct packed_git *target;
1030        struct packed_git **lst;
1031        FILE *packfile;
1032        char *filename;
1033        char tmpfile[PATH_MAX];
1034        int ret;
1035        long prev_posn = 0;
1036        char range[RANGE_HEADER_SIZE];
1037        struct curl_slist *range_header = NULL;
1038
1039        struct active_request_slot *slot;
1040
1041        if (fetch_indices(repo))
1042                return -1;
1043        target = find_sha1_pack(sha1, repo->packs);
1044        if (!target)
1045                return -1;
1046
1047        if (get_verbosely) {
1048                fprintf(stderr, "Getting pack %s\n",
1049                        sha1_to_hex(target->sha1));
1050                fprintf(stderr, " which contains %s\n",
1051                        sha1_to_hex(sha1));
1052        }
1053
1054        url = xmalloc(strlen(repo->base) + 65);
1055        sprintf(url, "%s/objects/pack/pack-%s.pack",
1056                repo->base, sha1_to_hex(target->sha1));
1057
1058        filename = sha1_pack_name(target->sha1);
1059        snprintf(tmpfile, sizeof(tmpfile), "%s.temp", filename);
1060        packfile = fopen(tmpfile, "a");
1061        if (!packfile)
1062                return error("Unable to open local file %s for pack",
1063                             filename);
1064
1065        slot = get_active_slot();
1066        curl_easy_setopt(slot->curl, CURLOPT_FILE, packfile);
1067        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite);
1068        curl_easy_setopt(slot->curl, CURLOPT_URL, url);
1069        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, no_pragma_header);
1070        slot->local = packfile;
1071
1072        /* If there is data present from a previous transfer attempt,
1073           resume where it left off */
1074        prev_posn = ftell(packfile);
1075        if (prev_posn>0) {
1076                if (get_verbosely)
1077                        fprintf(stderr,
1078                                "Resuming fetch of pack %s at byte %ld\n",
1079                                sha1_to_hex(target->sha1), prev_posn);
1080                sprintf(range, "Range: bytes=%ld-", prev_posn);
1081                range_header = curl_slist_append(range_header, range);
1082                curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, range_header);
1083        }
1084
1085        if (start_active_slot(slot)) {
1086                run_active_slot(slot);
1087                if (slot->curl_result != CURLE_OK) {
1088                        fclose(packfile);
1089                        return error("Unable to get pack file %s\n%s", url,
1090                                     curl_errorstr);
1091                }
1092        } else {
1093                fclose(packfile);
1094                return error("Unable to start request");
1095        }
1096
1097        fclose(packfile);
1098
1099        ret = move_temp_to_file(tmpfile, filename);
1100        if (ret)
1101                return ret;
1102
1103        lst = &repo->packs;
1104        while (*lst != target)
1105                lst = &((*lst)->next);
1106        *lst = (*lst)->next;
1107
1108        if (verify_pack(target, 0))
1109                return -1;
1110        install_packed_git(target);
1111
1112        return 0;
1113}
1114
1115static int fetch_object(struct alt_base *repo, unsigned char *sha1)
1116{
1117        char *hex = sha1_to_hex(sha1);
1118        int ret;
1119        struct transfer_request *request = request_queue_head;
1120
1121        while (request != NULL && memcmp(request->sha1, sha1, 20))
1122                request = request->next;
1123        if (request == NULL)
1124                return error("Couldn't find request for %s in the queue", hex);
1125
1126        if (has_sha1_file(request->sha1)) {
1127                release_request(request);
1128                return 0;
1129        }
1130
1131#ifdef USE_CURL_MULTI
1132        while (request->state == WAITING) {
1133                int num_transfers;
1134                curl_multi_perform(curlm, &num_transfers);
1135                if (num_transfers < active_requests) {
1136                        process_curl_messages();
1137                        process_request_queue();
1138                }
1139        }
1140#else
1141        start_request(request);
1142#endif
1143
1144        while (request->state == ACTIVE) {
1145                run_active_slot(request->slot);
1146#ifndef USE_CURL_MULTI
1147                request->curl_result = request->slot->curl_result;
1148                request->http_code = request->slot->http_code;
1149                request->slot = NULL;
1150
1151                /* Use alternates if necessary */
1152                if (request->http_code == 404) {
1153                        fetch_alternates(alt->base);
1154                        if (request->repo->next != NULL) {
1155                                request->repo = request->repo->next;
1156                                close(request->local); request->local = -1;
1157                                start_request(request);
1158                        }
1159                } else {
1160                        finish_request(request);
1161                        request->state = COMPLETE;
1162                }
1163#endif
1164        }
1165        if (request->local != -1) {
1166                close(request->local); request->local = -1;
1167        }
1168
1169        if (request->state == ABORTED) {
1170                release_request(request);
1171                return error("Request for %s aborted", hex);
1172        }
1173
1174        if (request->curl_result != CURLE_OK && request->http_code != 416) {
1175                if (request->http_code == 404)
1176                        ret = -1; /* Be silent, it is probably in a pack. */
1177                else
1178                        ret = error("%s (curl_result = %d, http_code = %ld, sha1 = %s)",
1179                                    request->errorstr, request->curl_result,
1180                                    request->http_code, hex);
1181                release_request(request);
1182                return ret;
1183        }
1184
1185        if (request->zret != Z_STREAM_END) {
1186                ret = error("File %s (%s) corrupt\n", hex, request->url);
1187                release_request(request);
1188                return ret;
1189        }
1190
1191        if (memcmp(request->sha1, request->real_sha1, 20)) {
1192                release_request(request);
1193                return error("File %s has bad hash\n", hex);
1194        }
1195
1196        if (request->rename < 0) {
1197                ret = error("unable to write sha1 filename %s: %s",
1198                            request->filename,
1199                            strerror(request->rename));
1200                release_request(request);
1201                return ret;
1202        }
1203
1204        release_request(request);
1205        return 0;
1206}
1207
1208int fetch(unsigned char *sha1)
1209{
1210        struct alt_base *altbase = alt;
1211
1212        if (!fetch_object(altbase, sha1))
1213                return 0;
1214        while (altbase) {
1215                if (!fetch_pack(altbase, sha1))
1216                        return 0;
1217                fetch_alternates(alt->base);
1218                altbase = altbase->next;
1219        }
1220        return error("Unable to find %s under %s\n", sha1_to_hex(sha1), 
1221                     alt->base);
1222}
1223
1224static inline int needs_quote(int ch)
1225{
1226        switch (ch) {
1227        case '/': case '-': case '.':
1228        case 'A'...'Z': case 'a'...'z': case '0'...'9':
1229                return 0;
1230        default:
1231                return 1;
1232        }
1233}
1234
1235static inline int hex(int v)
1236{
1237        if (v < 10) return '0' + v;
1238        else return 'A' + v - 10;
1239}
1240
1241static char *quote_ref_url(const char *base, const char *ref)
1242{
1243        const char *cp;
1244        char *dp, *qref;
1245        int len, baselen, ch;
1246
1247        baselen = strlen(base);
1248        len = baselen + 6; /* "refs/" + NUL */
1249        for (cp = ref; (ch = *cp) != 0; cp++, len++)
1250                if (needs_quote(ch))
1251                        len += 2; /* extra two hex plus replacement % */
1252        qref = xmalloc(len);
1253        memcpy(qref, base, baselen);
1254        memcpy(qref + baselen, "refs/", 5);
1255        for (cp = ref, dp = qref + baselen + 5; (ch = *cp) != 0; cp++) {
1256                if (needs_quote(ch)) {
1257                        *dp++ = '%';
1258                        *dp++ = hex((ch >> 4) & 0xF);
1259                        *dp++ = hex(ch & 0xF);
1260                }
1261                else
1262                        *dp++ = ch;
1263        }
1264        *dp = 0;
1265
1266        return qref;
1267}
1268
1269int fetch_ref(char *ref, unsigned char *sha1)
1270{
1271        char *url;
1272        char hex[42];
1273        struct buffer buffer;
1274        char *base = alt->base;
1275        struct active_request_slot *slot;
1276        buffer.size = 41;
1277        buffer.posn = 0;
1278        buffer.buffer = hex;
1279        hex[41] = '\0';
1280        
1281        url = quote_ref_url(base, ref);
1282        slot = get_active_slot();
1283        curl_easy_setopt(slot->curl, CURLOPT_FILE, &buffer);
1284        curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, fwrite_buffer);
1285        curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, NULL);
1286        curl_easy_setopt(slot->curl, CURLOPT_URL, url);
1287        if (start_active_slot(slot)) {
1288                run_active_slot(slot);
1289                if (slot->curl_result != CURLE_OK)
1290                        return error("Couldn't get %s for %s\n%s",
1291                                     url, ref, curl_errorstr);
1292        } else {
1293                return error("Unable to start request");
1294        }
1295
1296        hex[40] = '\0';
1297        get_sha1_hex(hex, sha1);
1298        return 0;
1299}
1300
1301int main(int argc, char **argv)
1302{
1303        char *commit_id;
1304        char *url;
1305        int arg = 1;
1306        struct active_request_slot *slot;
1307        char *low_speed_limit;
1308        char *low_speed_time;
1309        char *wait_url;
1310        int rc = 0;
1311
1312        while (arg < argc && argv[arg][0] == '-') {
1313                if (argv[arg][1] == 't') {
1314                        get_tree = 1;
1315                } else if (argv[arg][1] == 'c') {
1316                        get_history = 1;
1317                } else if (argv[arg][1] == 'a') {
1318                        get_all = 1;
1319                        get_tree = 1;
1320                        get_history = 1;
1321                } else if (argv[arg][1] == 'v') {
1322                        get_verbosely = 1;
1323                } else if (argv[arg][1] == 'w') {
1324                        write_ref = argv[arg + 1];
1325                        arg++;
1326                } else if (!strcmp(argv[arg], "--recover")) {
1327                        get_recover = 1;
1328                }
1329                arg++;
1330        }
1331        if (argc < arg + 2) {
1332                usage("git-http-fetch [-c] [-t] [-a] [-d] [-v] [--recover] [-w ref] commit-id url");
1333                return 1;
1334        }
1335        commit_id = argv[arg];
1336        url = argv[arg + 1];
1337
1338        curl_global_init(CURL_GLOBAL_ALL);
1339
1340#ifdef USE_CURL_MULTI
1341        {
1342                char *http_max_requests = getenv("GIT_HTTP_MAX_REQUESTS");
1343                if (http_max_requests != NULL)
1344                        max_requests = atoi(http_max_requests);
1345        }
1346
1347        curlm = curl_multi_init();
1348        if (curlm == NULL) {
1349                fprintf(stderr, "Error creating curl multi handle.\n");
1350                return 1;
1351        }
1352#endif
1353
1354        if (getenv("GIT_SSL_NO_VERIFY"))
1355                curl_ssl_verify = 0;
1356
1357        ssl_cert = getenv("GIT_SSL_CERT");
1358#if LIBCURL_VERSION_NUM >= 0x070902
1359        ssl_key = getenv("GIT_SSL_KEY");
1360#endif
1361#if LIBCURL_VERSION_NUM >= 0x070908
1362        ssl_capath = getenv("GIT_SSL_CAPATH");
1363#endif
1364        ssl_cainfo = getenv("GIT_SSL_CAINFO");
1365
1366        low_speed_limit = getenv("GIT_HTTP_LOW_SPEED_LIMIT");
1367        if (low_speed_limit != NULL)
1368                curl_low_speed_limit = strtol(low_speed_limit, NULL, 10);
1369        low_speed_time = getenv("GIT_HTTP_LOW_SPEED_TIME");
1370        if (low_speed_time != NULL)
1371                curl_low_speed_time = strtol(low_speed_time, NULL, 10);
1372
1373        git_config(http_options);
1374
1375        if (curl_ssl_verify == -1)
1376                curl_ssl_verify = 1;
1377
1378#ifdef USE_CURL_MULTI
1379        if (max_requests < 1)
1380                max_requests = DEFAULT_MAX_REQUESTS;
1381#endif
1382
1383        pragma_header = curl_slist_append(pragma_header, "Pragma: no-cache");
1384        no_pragma_header = curl_slist_append(no_pragma_header, "Pragma:");
1385        no_range_header = curl_slist_append(no_range_header, "Range:");
1386
1387#ifndef NO_CURL_EASY_DUPHANDLE
1388        curl_default = get_curl_handle();
1389#endif
1390
1391        alt = xmalloc(sizeof(*alt));
1392        alt->base = url;
1393        alt->got_indices = 0;
1394        alt->packs = NULL;
1395        alt->next = NULL;
1396
1397        if (pull(commit_id))
1398                rc = 1;
1399
1400        curl_slist_free_all(pragma_header);
1401        curl_slist_free_all(no_pragma_header);
1402        curl_slist_free_all(no_range_header);
1403#ifndef NO_CURL_EASY_DUPHANDLE
1404        curl_easy_cleanup(curl_default);
1405#endif
1406        slot = active_queue_head;
1407        while (slot != NULL) {
1408                if (slot->in_use) {
1409                        if (get_verbosely) {
1410                                curl_easy_getinfo(slot->curl,
1411                                                  CURLINFO_EFFECTIVE_URL,
1412                                                  &wait_url);
1413                                fprintf(stderr, "Waiting for %s\n", wait_url);
1414                        }
1415                        run_active_slot(slot);
1416                }
1417                if (slot->curl != NULL)
1418                        curl_easy_cleanup(slot->curl);
1419                slot = slot->next;
1420        }
1421#ifdef USE_CURL_MULTI
1422        curl_multi_cleanup(curlm);
1423#endif
1424        curl_global_cleanup();
1425        return rc;
1426}