sha1_file.con commit Add diff-option --ext-diff (72909be)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This handles basic git sha1 object files - packing, unpacking,
   7 * creation etc.
   8 */
   9#include "cache.h"
  10#include "delta.h"
  11#include "pack.h"
  12#include "blob.h"
  13#include "commit.h"
  14#include "tag.h"
  15#include "tree.h"
  16#include "refs.h"
  17
  18#ifndef O_NOATIME
  19#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  20#define O_NOATIME 01000000
  21#else
  22#define O_NOATIME 0
  23#endif
  24#endif
  25
  26#ifdef NO_C99_FORMAT
  27#define SZ_FMT "lu"
  28#else
  29#define SZ_FMT "zu"
  30#endif
  31
  32const unsigned char null_sha1[20];
  33
  34static unsigned int sha1_file_open_flag = O_NOATIME;
  35
  36const signed char hexval_table[256] = {
  37         -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
  38         -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
  39         -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
  40         -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
  41         -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
  42         -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
  43          0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
  44          8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
  45         -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
  46         -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
  47         -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
  48         -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
  49         -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
  50         -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
  51         -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
  52         -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
  53         -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
  54         -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
  55         -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
  56         -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
  57         -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
  58         -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
  59         -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
  60         -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
  61         -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
  62         -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
  63         -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
  64         -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
  65         -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
  66         -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
  67         -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
  68         -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
  69};
  70
  71int get_sha1_hex(const char *hex, unsigned char *sha1)
  72{
  73        int i;
  74        for (i = 0; i < 20; i++) {
  75                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  76                if (val & ~0xff)
  77                        return -1;
  78                *sha1++ = val;
  79                hex += 2;
  80        }
  81        return 0;
  82}
  83
  84int safe_create_leading_directories(char *path)
  85{
  86        char *pos = path;
  87        struct stat st;
  88
  89        if (*pos == '/')
  90                pos++;
  91
  92        while (pos) {
  93                pos = strchr(pos, '/');
  94                if (!pos)
  95                        break;
  96                *pos = 0;
  97                if (!stat(path, &st)) {
  98                        /* path exists */
  99                        if (!S_ISDIR(st.st_mode)) {
 100                                *pos = '/';
 101                                return -3;
 102                        }
 103                }
 104                else if (mkdir(path, 0777)) {
 105                        *pos = '/';
 106                        return -1;
 107                }
 108                else if (adjust_shared_perm(path)) {
 109                        *pos = '/';
 110                        return -2;
 111                }
 112                *pos++ = '/';
 113        }
 114        return 0;
 115}
 116
 117char * sha1_to_hex(const unsigned char *sha1)
 118{
 119        static int bufno;
 120        static char hexbuffer[4][50];
 121        static const char hex[] = "0123456789abcdef";
 122        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 123        int i;
 124
 125        for (i = 0; i < 20; i++) {
 126                unsigned int val = *sha1++;
 127                *buf++ = hex[val >> 4];
 128                *buf++ = hex[val & 0xf];
 129        }
 130        *buf = '\0';
 131
 132        return buffer;
 133}
 134
 135static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 136{
 137        int i;
 138        for (i = 0; i < 20; i++) {
 139                static char hex[] = "0123456789abcdef";
 140                unsigned int val = sha1[i];
 141                char *pos = pathbuf + i*2 + (i > 0);
 142                *pos++ = hex[val >> 4];
 143                *pos = hex[val & 0xf];
 144        }
 145}
 146
 147/*
 148 * NOTE! This returns a statically allocated buffer, so you have to be
 149 * careful about using it. Do a "xstrdup()" if you need to save the
 150 * filename.
 151 *
 152 * Also note that this returns the location for creating.  Reading
 153 * SHA1 file can happen from any alternate directory listed in the
 154 * DB_ENVIRONMENT environment variable if it is not found in
 155 * the primary object database.
 156 */
 157char *sha1_file_name(const unsigned char *sha1)
 158{
 159        static char *name, *base;
 160
 161        if (!base) {
 162                const char *sha1_file_directory = get_object_directory();
 163                int len = strlen(sha1_file_directory);
 164                base = xmalloc(len + 60);
 165                memcpy(base, sha1_file_directory, len);
 166                memset(base+len, 0, 60);
 167                base[len] = '/';
 168                base[len+3] = '/';
 169                name = base + len + 1;
 170        }
 171        fill_sha1_path(name, sha1);
 172        return base;
 173}
 174
 175char *sha1_pack_name(const unsigned char *sha1)
 176{
 177        static const char hex[] = "0123456789abcdef";
 178        static char *name, *base, *buf;
 179        int i;
 180
 181        if (!base) {
 182                const char *sha1_file_directory = get_object_directory();
 183                int len = strlen(sha1_file_directory);
 184                base = xmalloc(len + 60);
 185                sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
 186                name = base + len + 11;
 187        }
 188
 189        buf = name;
 190
 191        for (i = 0; i < 20; i++) {
 192                unsigned int val = *sha1++;
 193                *buf++ = hex[val >> 4];
 194                *buf++ = hex[val & 0xf];
 195        }
 196
 197        return base;
 198}
 199
 200char *sha1_pack_index_name(const unsigned char *sha1)
 201{
 202        static const char hex[] = "0123456789abcdef";
 203        static char *name, *base, *buf;
 204        int i;
 205
 206        if (!base) {
 207                const char *sha1_file_directory = get_object_directory();
 208                int len = strlen(sha1_file_directory);
 209                base = xmalloc(len + 60);
 210                sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
 211                name = base + len + 11;
 212        }
 213
 214        buf = name;
 215
 216        for (i = 0; i < 20; i++) {
 217                unsigned int val = *sha1++;
 218                *buf++ = hex[val >> 4];
 219                *buf++ = hex[val & 0xf];
 220        }
 221
 222        return base;
 223}
 224
 225struct alternate_object_database *alt_odb_list;
 226static struct alternate_object_database **alt_odb_tail;
 227
 228static void read_info_alternates(const char * alternates, int depth);
 229
 230/*
 231 * Prepare alternate object database registry.
 232 *
 233 * The variable alt_odb_list points at the list of struct
 234 * alternate_object_database.  The elements on this list come from
 235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 237 * whose contents is similar to that environment variable but can be
 238 * LF separated.  Its base points at a statically allocated buffer that
 239 * contains "/the/directory/corresponding/to/.git/objects/...", while
 240 * its name points just after the slash at the end of ".git/objects/"
 241 * in the example above, and has enough space to hold 40-byte hex
 242 * SHA1, an extra slash for the first level indirection, and the
 243 * terminating NUL.
 244 */
 245static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 246{
 247        struct stat st;
 248        const char *objdir = get_object_directory();
 249        struct alternate_object_database *ent;
 250        struct alternate_object_database *alt;
 251        /* 43 = 40-byte + 2 '/' + terminating NUL */
 252        int pfxlen = len;
 253        int entlen = pfxlen + 43;
 254        int base_len = -1;
 255
 256        if (*entry != '/' && relative_base) {
 257                /* Relative alt-odb */
 258                if (base_len < 0)
 259                        base_len = strlen(relative_base) + 1;
 260                entlen += base_len;
 261                pfxlen += base_len;
 262        }
 263        ent = xmalloc(sizeof(*ent) + entlen);
 264
 265        if (*entry != '/' && relative_base) {
 266                memcpy(ent->base, relative_base, base_len - 1);
 267                ent->base[base_len - 1] = '/';
 268                memcpy(ent->base + base_len, entry, len);
 269        }
 270        else
 271                memcpy(ent->base, entry, pfxlen);
 272
 273        ent->name = ent->base + pfxlen + 1;
 274        ent->base[pfxlen + 3] = '/';
 275        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 276
 277        /* Detect cases where alternate disappeared */
 278        if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
 279                error("object directory %s does not exist; "
 280                      "check .git/objects/info/alternates.",
 281                      ent->base);
 282                free(ent);
 283                return -1;
 284        }
 285
 286        /* Prevent the common mistake of listing the same
 287         * thing twice, or object directory itself.
 288         */
 289        for (alt = alt_odb_list; alt; alt = alt->next) {
 290                if (!memcmp(ent->base, alt->base, pfxlen)) {
 291                        free(ent);
 292                        return -1;
 293                }
 294        }
 295        if (!memcmp(ent->base, objdir, pfxlen)) {
 296                free(ent);
 297                return -1;
 298        }
 299
 300        /* add the alternate entry */
 301        *alt_odb_tail = ent;
 302        alt_odb_tail = &(ent->next);
 303        ent->next = NULL;
 304
 305        /* recursively add alternates */
 306        read_info_alternates(ent->base, depth + 1);
 307
 308        ent->base[pfxlen] = '/';
 309
 310        return 0;
 311}
 312
 313static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 314                                 const char *relative_base, int depth)
 315{
 316        const char *cp, *last;
 317
 318        if (depth > 5) {
 319                error("%s: ignoring alternate object stores, nesting too deep.",
 320                                relative_base);
 321                return;
 322        }
 323
 324        last = alt;
 325        while (last < ep) {
 326                cp = last;
 327                if (cp < ep && *cp == '#') {
 328                        while (cp < ep && *cp != sep)
 329                                cp++;
 330                        last = cp + 1;
 331                        continue;
 332                }
 333                while (cp < ep && *cp != sep)
 334                        cp++;
 335                if (last != cp) {
 336                        if ((*last != '/') && depth) {
 337                                error("%s: ignoring relative alternate object store %s",
 338                                                relative_base, last);
 339                        } else {
 340                                link_alt_odb_entry(last, cp - last,
 341                                                relative_base, depth);
 342                        }
 343                }
 344                while (cp < ep && *cp == sep)
 345                        cp++;
 346                last = cp;
 347        }
 348}
 349
 350static void read_info_alternates(const char * relative_base, int depth)
 351{
 352        char *map;
 353        size_t mapsz;
 354        struct stat st;
 355        char path[PATH_MAX];
 356        int fd;
 357
 358        sprintf(path, "%s/info/alternates", relative_base);
 359        fd = open(path, O_RDONLY);
 360        if (fd < 0)
 361                return;
 362        if (fstat(fd, &st) || (st.st_size == 0)) {
 363                close(fd);
 364                return;
 365        }
 366        mapsz = xsize_t(st.st_size);
 367        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 368        close(fd);
 369
 370        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 371
 372        munmap(map, mapsz);
 373}
 374
 375void prepare_alt_odb(void)
 376{
 377        const char *alt;
 378
 379        if (alt_odb_tail)
 380                return;
 381
 382        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 383        if (!alt) alt = "";
 384
 385        alt_odb_tail = &alt_odb_list;
 386        link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
 387
 388        read_info_alternates(get_object_directory(), 0);
 389}
 390
 391static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 392{
 393        char *name = sha1_file_name(sha1);
 394        struct alternate_object_database *alt;
 395
 396        if (!stat(name, st))
 397                return name;
 398        prepare_alt_odb();
 399        for (alt = alt_odb_list; alt; alt = alt->next) {
 400                name = alt->name;
 401                fill_sha1_path(name, sha1);
 402                if (!stat(alt->base, st))
 403                        return alt->base;
 404        }
 405        return NULL;
 406}
 407
 408static unsigned int pack_used_ctr;
 409static unsigned int pack_mmap_calls;
 410static unsigned int peak_pack_open_windows;
 411static unsigned int pack_open_windows;
 412static size_t peak_pack_mapped;
 413static size_t pack_mapped;
 414struct packed_git *packed_git;
 415
 416void pack_report(void)
 417{
 418        fprintf(stderr,
 419                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 420                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 421                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 422                (size_t) getpagesize(),
 423                packed_git_window_size,
 424                packed_git_limit);
 425        fprintf(stderr,
 426                "pack_report: pack_used_ctr            = %10u\n"
 427                "pack_report: pack_mmap_calls          = %10u\n"
 428                "pack_report: pack_open_windows        = %10u / %10u\n"
 429                "pack_report: pack_mapped              = "
 430                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 431                pack_used_ctr,
 432                pack_mmap_calls,
 433                pack_open_windows, peak_pack_open_windows,
 434                pack_mapped, peak_pack_mapped);
 435}
 436
 437static int check_packed_git_idx(const char *path,  struct packed_git *p)
 438{
 439        void *idx_map;
 440        struct pack_idx_header *hdr;
 441        size_t idx_size;
 442        uint32_t version, nr, i, *index;
 443        int fd = open(path, O_RDONLY);
 444        struct stat st;
 445
 446        if (fd < 0)
 447                return -1;
 448        if (fstat(fd, &st)) {
 449                close(fd);
 450                return -1;
 451        }
 452        idx_size = xsize_t(st.st_size);
 453        if (idx_size < 4 * 256 + 20 + 20) {
 454                close(fd);
 455                return error("index file %s is too small", path);
 456        }
 457        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 458        close(fd);
 459
 460        hdr = idx_map;
 461        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 462                version = ntohl(hdr->idx_version);
 463                if (version < 2 || version > 2) {
 464                        munmap(idx_map, idx_size);
 465                        return error("index file %s is version %d"
 466                                     " and is not supported by this binary"
 467                                     " (try upgrading GIT to a newer version)",
 468                                     path, version);
 469                }
 470        } else
 471                version = 1;
 472
 473        nr = 0;
 474        index = idx_map;
 475        if (version > 1)
 476                index += 2;  /* skip index header */
 477        for (i = 0; i < 256; i++) {
 478                uint32_t n = ntohl(index[i]);
 479                if (n < nr) {
 480                        munmap(idx_map, idx_size);
 481                        return error("non-monotonic index %s", path);
 482                }
 483                nr = n;
 484        }
 485
 486        if (version == 1) {
 487                /*
 488                 * Total size:
 489                 *  - 256 index entries 4 bytes each
 490                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 491                 *  - 20-byte SHA1 of the packfile
 492                 *  - 20-byte SHA1 file checksum
 493                 */
 494                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 495                        munmap(idx_map, idx_size);
 496                        return error("wrong index file size in %s", path);
 497                }
 498        } else if (version == 2) {
 499                /*
 500                 * Minimum size:
 501                 *  - 8 bytes of header
 502                 *  - 256 index entries 4 bytes each
 503                 *  - 20-byte sha1 entry * nr
 504                 *  - 4-byte crc entry * nr
 505                 *  - 4-byte offset entry * nr
 506                 *  - 20-byte SHA1 of the packfile
 507                 *  - 20-byte SHA1 file checksum
 508                 * And after the 4-byte offset table might be a
 509                 * variable sized table containing 8-byte entries
 510                 * for offsets larger than 2^31.
 511                 */
 512                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 513                unsigned long max_size = min_size;
 514                if (nr)
 515                        max_size += (nr - 1)*8;
 516                if (idx_size < min_size || idx_size > max_size) {
 517                        munmap(idx_map, idx_size);
 518                        return error("wrong index file size in %s", path);
 519                }
 520                if (idx_size != min_size) {
 521                        /* make sure we can deal with large pack offsets */
 522                        off_t x = 0x7fffffffUL, y = 0xffffffffUL;
 523                        if (x > (x + 1) || y > (y + 1)) {
 524                                munmap(idx_map, idx_size);
 525                                return error("pack too large for current definition of off_t in %s", path);
 526                        }
 527                }
 528        }
 529
 530        p->index_version = version;
 531        p->index_data = idx_map;
 532        p->index_size = idx_size;
 533        p->num_objects = nr;
 534        return 0;
 535}
 536
 537int open_pack_index(struct packed_git *p)
 538{
 539        char *idx_name;
 540        int ret;
 541
 542        if (p->index_data)
 543                return 0;
 544
 545        idx_name = xstrdup(p->pack_name);
 546        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 547        ret = check_packed_git_idx(idx_name, p);
 548        free(idx_name);
 549        return ret;
 550}
 551
 552static void scan_windows(struct packed_git *p,
 553        struct packed_git **lru_p,
 554        struct pack_window **lru_w,
 555        struct pack_window **lru_l)
 556{
 557        struct pack_window *w, *w_l;
 558
 559        for (w_l = NULL, w = p->windows; w; w = w->next) {
 560                if (!w->inuse_cnt) {
 561                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 562                                *lru_p = p;
 563                                *lru_w = w;
 564                                *lru_l = w_l;
 565                        }
 566                }
 567                w_l = w;
 568        }
 569}
 570
 571static int unuse_one_window(struct packed_git *current, int keep_fd)
 572{
 573        struct packed_git *p, *lru_p = NULL;
 574        struct pack_window *lru_w = NULL, *lru_l = NULL;
 575
 576        if (current)
 577                scan_windows(current, &lru_p, &lru_w, &lru_l);
 578        for (p = packed_git; p; p = p->next)
 579                scan_windows(p, &lru_p, &lru_w, &lru_l);
 580        if (lru_p) {
 581                munmap(lru_w->base, lru_w->len);
 582                pack_mapped -= lru_w->len;
 583                if (lru_l)
 584                        lru_l->next = lru_w->next;
 585                else {
 586                        lru_p->windows = lru_w->next;
 587                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 588                                close(lru_p->pack_fd);
 589                                lru_p->pack_fd = -1;
 590                        }
 591                }
 592                free(lru_w);
 593                pack_open_windows--;
 594                return 1;
 595        }
 596        return 0;
 597}
 598
 599void release_pack_memory(size_t need, int fd)
 600{
 601        size_t cur = pack_mapped;
 602        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 603                ; /* nothing */
 604}
 605
 606void unuse_pack(struct pack_window **w_cursor)
 607{
 608        struct pack_window *w = *w_cursor;
 609        if (w) {
 610                w->inuse_cnt--;
 611                *w_cursor = NULL;
 612        }
 613}
 614
 615/*
 616 * Do not call this directly as this leaks p->pack_fd on error return;
 617 * call open_packed_git() instead.
 618 */
 619static int open_packed_git_1(struct packed_git *p)
 620{
 621        struct stat st;
 622        struct pack_header hdr;
 623        unsigned char sha1[20];
 624        unsigned char *idx_sha1;
 625        long fd_flag;
 626
 627        if (!p->index_data && open_pack_index(p))
 628                return error("packfile %s index unavailable", p->pack_name);
 629
 630        p->pack_fd = open(p->pack_name, O_RDONLY);
 631        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 632                return -1;
 633
 634        /* If we created the struct before we had the pack we lack size. */
 635        if (!p->pack_size) {
 636                if (!S_ISREG(st.st_mode))
 637                        return error("packfile %s not a regular file", p->pack_name);
 638                p->pack_size = st.st_size;
 639        } else if (p->pack_size != st.st_size)
 640                return error("packfile %s size changed", p->pack_name);
 641
 642        /* We leave these file descriptors open with sliding mmap;
 643         * there is no point keeping them open across exec(), though.
 644         */
 645        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 646        if (fd_flag < 0)
 647                return error("cannot determine file descriptor flags");
 648        fd_flag |= FD_CLOEXEC;
 649        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 650                return error("cannot set FD_CLOEXEC");
 651
 652        /* Verify we recognize this pack file format. */
 653        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 654                return error("file %s is far too short to be a packfile", p->pack_name);
 655        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 656                return error("file %s is not a GIT packfile", p->pack_name);
 657        if (!pack_version_ok(hdr.hdr_version))
 658                return error("packfile %s is version %u and not supported"
 659                        " (try upgrading GIT to a newer version)",
 660                        p->pack_name, ntohl(hdr.hdr_version));
 661
 662        /* Verify the pack matches its index. */
 663        if (p->num_objects != ntohl(hdr.hdr_entries))
 664                return error("packfile %s claims to have %u objects"
 665                             " while index indicates %u objects",
 666                             p->pack_name, ntohl(hdr.hdr_entries),
 667                             p->num_objects);
 668        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 669                return error("end of packfile %s is unavailable", p->pack_name);
 670        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 671                return error("packfile %s signature is unavailable", p->pack_name);
 672        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 673        if (hashcmp(sha1, idx_sha1))
 674                return error("packfile %s does not match index", p->pack_name);
 675        return 0;
 676}
 677
 678static int open_packed_git(struct packed_git *p)
 679{
 680        if (!open_packed_git_1(p))
 681                return 0;
 682        if (p->pack_fd != -1) {
 683                close(p->pack_fd);
 684                p->pack_fd = -1;
 685        }
 686        return -1;
 687}
 688
 689static int in_window(struct pack_window *win, off_t offset)
 690{
 691        /* We must promise at least 20 bytes (one hash) after the
 692         * offset is available from this window, otherwise the offset
 693         * is not actually in this window and a different window (which
 694         * has that one hash excess) must be used.  This is to support
 695         * the object header and delta base parsing routines below.
 696         */
 697        off_t win_off = win->offset;
 698        return win_off <= offset
 699                && (offset + 20) <= (win_off + win->len);
 700}
 701
 702unsigned char* use_pack(struct packed_git *p,
 703                struct pack_window **w_cursor,
 704                off_t offset,
 705                unsigned int *left)
 706{
 707        struct pack_window *win = *w_cursor;
 708
 709        if (p->pack_fd == -1 && open_packed_git(p))
 710                die("packfile %s cannot be accessed", p->pack_name);
 711
 712        /* Since packfiles end in a hash of their content and its
 713         * pointless to ask for an offset into the middle of that
 714         * hash, and the in_window function above wouldn't match
 715         * don't allow an offset too close to the end of the file.
 716         */
 717        if (offset > (p->pack_size - 20))
 718                die("offset beyond end of packfile (truncated pack?)");
 719
 720        if (!win || !in_window(win, offset)) {
 721                if (win)
 722                        win->inuse_cnt--;
 723                for (win = p->windows; win; win = win->next) {
 724                        if (in_window(win, offset))
 725                                break;
 726                }
 727                if (!win) {
 728                        size_t window_align = packed_git_window_size / 2;
 729                        off_t len;
 730                        win = xcalloc(1, sizeof(*win));
 731                        win->offset = (offset / window_align) * window_align;
 732                        len = p->pack_size - win->offset;
 733                        if (len > packed_git_window_size)
 734                                len = packed_git_window_size;
 735                        win->len = (size_t)len;
 736                        pack_mapped += win->len;
 737                        while (packed_git_limit < pack_mapped
 738                                && unuse_one_window(p, p->pack_fd))
 739                                ; /* nothing */
 740                        win->base = xmmap(NULL, win->len,
 741                                PROT_READ, MAP_PRIVATE,
 742                                p->pack_fd, win->offset);
 743                        if (win->base == MAP_FAILED)
 744                                die("packfile %s cannot be mapped: %s",
 745                                        p->pack_name,
 746                                        strerror(errno));
 747                        pack_mmap_calls++;
 748                        pack_open_windows++;
 749                        if (pack_mapped > peak_pack_mapped)
 750                                peak_pack_mapped = pack_mapped;
 751                        if (pack_open_windows > peak_pack_open_windows)
 752                                peak_pack_open_windows = pack_open_windows;
 753                        win->next = p->windows;
 754                        p->windows = win;
 755                }
 756        }
 757        if (win != *w_cursor) {
 758                win->last_used = pack_used_ctr++;
 759                win->inuse_cnt++;
 760                *w_cursor = win;
 761        }
 762        offset -= win->offset;
 763        if (left)
 764                *left = win->len - xsize_t(offset);
 765        return win->base + offset;
 766}
 767
 768struct packed_git *add_packed_git(const char *path, int path_len, int local)
 769{
 770        struct stat st;
 771        struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
 772
 773        /*
 774         * Make sure a corresponding .pack file exists and that
 775         * the index looks sane.
 776         */
 777        path_len -= strlen(".idx");
 778        if (path_len < 1)
 779                return NULL;
 780        memcpy(p->pack_name, path, path_len);
 781        strcpy(p->pack_name + path_len, ".pack");
 782        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 783                free(p);
 784                return NULL;
 785        }
 786
 787        /* ok, it looks sane as far as we can check without
 788         * actually mapping the pack file.
 789         */
 790        p->index_version = 0;
 791        p->index_data = NULL;
 792        p->index_size = 0;
 793        p->num_objects = 0;
 794        p->pack_size = st.st_size;
 795        p->next = NULL;
 796        p->windows = NULL;
 797        p->pack_fd = -1;
 798        p->pack_local = local;
 799        p->mtime = st.st_mtime;
 800        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 801                hashclr(p->sha1);
 802        return p;
 803}
 804
 805struct packed_git *parse_pack_index(unsigned char *sha1)
 806{
 807        char *path = sha1_pack_index_name(sha1);
 808        return parse_pack_index_file(sha1, path);
 809}
 810
 811struct packed_git *parse_pack_index_file(const unsigned char *sha1,
 812                                         const char *idx_path)
 813{
 814        const char *path = sha1_pack_name(sha1);
 815        struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
 816
 817        if (check_packed_git_idx(idx_path, p)) {
 818                free(p);
 819                return NULL;
 820        }
 821
 822        strcpy(p->pack_name, path);
 823        p->pack_size = 0;
 824        p->next = NULL;
 825        p->windows = NULL;
 826        p->pack_fd = -1;
 827        hashcpy(p->sha1, sha1);
 828        return p;
 829}
 830
 831void install_packed_git(struct packed_git *pack)
 832{
 833        pack->next = packed_git;
 834        packed_git = pack;
 835}
 836
 837static void prepare_packed_git_one(char *objdir, int local)
 838{
 839        char path[PATH_MAX];
 840        int len;
 841        DIR *dir;
 842        struct dirent *de;
 843
 844        sprintf(path, "%s/pack", objdir);
 845        len = strlen(path);
 846        dir = opendir(path);
 847        if (!dir) {
 848                if (errno != ENOENT)
 849                        error("unable to open object pack directory: %s: %s",
 850                              path, strerror(errno));
 851                return;
 852        }
 853        path[len++] = '/';
 854        while ((de = readdir(dir)) != NULL) {
 855                int namelen = strlen(de->d_name);
 856                struct packed_git *p;
 857
 858                if (!has_extension(de->d_name, ".idx"))
 859                        continue;
 860
 861                /* Don't reopen a pack we already have. */
 862                strcpy(path + len, de->d_name);
 863                for (p = packed_git; p; p = p->next) {
 864                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 865                                break;
 866                }
 867                if (p)
 868                        continue;
 869                /* See if it really is a valid .idx file with corresponding
 870                 * .pack file that we can map.
 871                 */
 872                p = add_packed_git(path, len + namelen, local);
 873                if (!p)
 874                        continue;
 875                install_packed_git(p);
 876        }
 877        closedir(dir);
 878}
 879
 880static int sort_pack(const void *a_, const void *b_)
 881{
 882        struct packed_git *a = *((struct packed_git **)a_);
 883        struct packed_git *b = *((struct packed_git **)b_);
 884        int st;
 885
 886        /*
 887         * Local packs tend to contain objects specific to our
 888         * variant of the project than remote ones.  In addition,
 889         * remote ones could be on a network mounted filesystem.
 890         * Favor local ones for these reasons.
 891         */
 892        st = a->pack_local - b->pack_local;
 893        if (st)
 894                return -st;
 895
 896        /*
 897         * Younger packs tend to contain more recent objects,
 898         * and more recent objects tend to get accessed more
 899         * often.
 900         */
 901        if (a->mtime < b->mtime)
 902                return 1;
 903        else if (a->mtime == b->mtime)
 904                return 0;
 905        return -1;
 906}
 907
 908static void rearrange_packed_git(void)
 909{
 910        struct packed_git **ary, *p;
 911        int i, n;
 912
 913        for (n = 0, p = packed_git; p; p = p->next)
 914                n++;
 915        if (n < 2)
 916                return;
 917
 918        /* prepare an array of packed_git for easier sorting */
 919        ary = xcalloc(n, sizeof(struct packed_git *));
 920        for (n = 0, p = packed_git; p; p = p->next)
 921                ary[n++] = p;
 922
 923        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 924
 925        /* link them back again */
 926        for (i = 0; i < n - 1; i++)
 927                ary[i]->next = ary[i + 1];
 928        ary[n - 1]->next = NULL;
 929        packed_git = ary[0];
 930
 931        free(ary);
 932}
 933
 934static int prepare_packed_git_run_once = 0;
 935void prepare_packed_git(void)
 936{
 937        struct alternate_object_database *alt;
 938
 939        if (prepare_packed_git_run_once)
 940                return;
 941        prepare_packed_git_one(get_object_directory(), 1);
 942        prepare_alt_odb();
 943        for (alt = alt_odb_list; alt; alt = alt->next) {
 944                alt->name[-1] = 0;
 945                prepare_packed_git_one(alt->base, 0);
 946                alt->name[-1] = '/';
 947        }
 948        rearrange_packed_git();
 949        prepare_packed_git_run_once = 1;
 950}
 951
 952void reprepare_packed_git(void)
 953{
 954        prepare_packed_git_run_once = 0;
 955        prepare_packed_git();
 956}
 957
 958int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 959{
 960        unsigned char real_sha1[20];
 961        hash_sha1_file(map, size, type, real_sha1);
 962        return hashcmp(sha1, real_sha1) ? -1 : 0;
 963}
 964
 965static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 966{
 967        struct stat st;
 968        void *map;
 969        int fd;
 970        char *filename = find_sha1_file(sha1, &st);
 971
 972        if (!filename) {
 973                return NULL;
 974        }
 975
 976        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 977        if (fd < 0) {
 978                /* See if it works without O_NOATIME */
 979                switch (sha1_file_open_flag) {
 980                default:
 981                        fd = open(filename, O_RDONLY);
 982                        if (fd >= 0)
 983                                break;
 984                /* Fallthrough */
 985                case 0:
 986                        return NULL;
 987                }
 988
 989                /* If it failed once, it will probably fail again.
 990                 * Stop using O_NOATIME
 991                 */
 992                sha1_file_open_flag = 0;
 993        }
 994        *size = xsize_t(st.st_size);
 995        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 996        close(fd);
 997        return map;
 998}
 999
1000static int legacy_loose_object(unsigned char *map)
1001{
1002        unsigned int word;
1003
1004        /*
1005         * Is it a zlib-compressed buffer? If so, the first byte
1006         * must be 0x78 (15-bit window size, deflated), and the
1007         * first 16-bit word is evenly divisible by 31
1008         */
1009        word = (map[0] << 8) + map[1];
1010        if (map[0] == 0x78 && !(word % 31))
1011                return 1;
1012        else
1013                return 0;
1014}
1015
1016unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1017{
1018        unsigned shift;
1019        unsigned char c;
1020        unsigned long size;
1021        unsigned long used = 0;
1022
1023        c = buf[used++];
1024        *type = (c >> 4) & 7;
1025        size = c & 15;
1026        shift = 4;
1027        while (c & 0x80) {
1028                if (len <= used)
1029                        return 0;
1030                if (sizeof(long) * 8 <= shift)
1031                        return 0;
1032                c = buf[used++];
1033                size += (c & 0x7f) << shift;
1034                shift += 7;
1035        }
1036        *sizep = size;
1037        return used;
1038}
1039
1040static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1041{
1042        unsigned long size, used;
1043        static const char valid_loose_object_type[8] = {
1044                0, /* OBJ_EXT */
1045                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1046                0, /* "delta" and others are invalid in a loose object */
1047        };
1048        enum object_type type;
1049
1050        /* Get the data stream */
1051        memset(stream, 0, sizeof(*stream));
1052        stream->next_in = map;
1053        stream->avail_in = mapsize;
1054        stream->next_out = buffer;
1055        stream->avail_out = bufsiz;
1056
1057        if (legacy_loose_object(map)) {
1058                inflateInit(stream);
1059                return inflate(stream, 0);
1060        }
1061
1062
1063        /*
1064         * There used to be a second loose object header format which
1065         * was meant to mimic the in-pack format, allowing for direct
1066         * copy of the object data.  This format turned up not to be
1067         * really worth it and we don't write it any longer.  But we
1068         * can still read it.
1069         */
1070        used = unpack_object_header_gently(map, mapsize, &type, &size);
1071        if (!used || !valid_loose_object_type[type])
1072                return -1;
1073        map += used;
1074        mapsize -= used;
1075
1076        /* Set up the stream for the rest.. */
1077        stream->next_in = map;
1078        stream->avail_in = mapsize;
1079        inflateInit(stream);
1080
1081        /* And generate the fake traditional header */
1082        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1083                                         typename(type), size);
1084        return 0;
1085}
1086
1087static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1088{
1089        int bytes = strlen(buffer) + 1;
1090        unsigned char *buf = xmalloc(1+size);
1091        unsigned long n;
1092        int status = Z_OK;
1093
1094        n = stream->total_out - bytes;
1095        if (n > size)
1096                n = size;
1097        memcpy(buf, (char *) buffer + bytes, n);
1098        bytes = n;
1099        if (bytes <= size) {
1100                /*
1101                 * The above condition must be (bytes <= size), not
1102                 * (bytes < size).  In other words, even though we
1103                 * expect no more output and set avail_out to zer0,
1104                 * the input zlib stream may have bytes that express
1105                 * "this concludes the stream", and we *do* want to
1106                 * eat that input.
1107                 *
1108                 * Otherwise we would not be able to test that we
1109                 * consumed all the input to reach the expected size;
1110                 * we also want to check that zlib tells us that all
1111                 * went well with status == Z_STREAM_END at the end.
1112                 */
1113                stream->next_out = buf + bytes;
1114                stream->avail_out = size - bytes;
1115                while (status == Z_OK)
1116                        status = inflate(stream, Z_FINISH);
1117        }
1118        buf[size] = 0;
1119        if (status == Z_STREAM_END && !stream->avail_in) {
1120                inflateEnd(stream);
1121                return buf;
1122        }
1123
1124        if (status < 0)
1125                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1126        else if (stream->avail_in)
1127                error("garbage at end of loose object '%s'",
1128                      sha1_to_hex(sha1));
1129        free(buf);
1130        return NULL;
1131}
1132
1133/*
1134 * We used to just use "sscanf()", but that's actually way
1135 * too permissive for what we want to check. So do an anal
1136 * object header parse by hand.
1137 */
1138static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1139{
1140        char type[10];
1141        int i;
1142        unsigned long size;
1143
1144        /*
1145         * The type can be at most ten bytes (including the
1146         * terminating '\0' that we add), and is followed by
1147         * a space.
1148         */
1149        i = 0;
1150        for (;;) {
1151                char c = *hdr++;
1152                if (c == ' ')
1153                        break;
1154                type[i++] = c;
1155                if (i >= sizeof(type))
1156                        return -1;
1157        }
1158        type[i] = 0;
1159
1160        /*
1161         * The length must follow immediately, and be in canonical
1162         * decimal format (ie "010" is not valid).
1163         */
1164        size = *hdr++ - '0';
1165        if (size > 9)
1166                return -1;
1167        if (size) {
1168                for (;;) {
1169                        unsigned long c = *hdr - '0';
1170                        if (c > 9)
1171                                break;
1172                        hdr++;
1173                        size = size * 10 + c;
1174                }
1175        }
1176        *sizep = size;
1177
1178        /*
1179         * The length must be followed by a zero byte
1180         */
1181        return *hdr ? -1 : type_from_string(type);
1182}
1183
1184static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1185{
1186        int ret;
1187        z_stream stream;
1188        char hdr[8192];
1189
1190        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1191        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1192                return NULL;
1193
1194        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1195}
1196
1197unsigned long get_size_from_delta(struct packed_git *p,
1198                                  struct pack_window **w_curs,
1199                                  off_t curpos)
1200{
1201        const unsigned char *data;
1202        unsigned char delta_head[20], *in;
1203        z_stream stream;
1204        int st;
1205
1206        memset(&stream, 0, sizeof(stream));
1207        stream.next_out = delta_head;
1208        stream.avail_out = sizeof(delta_head);
1209
1210        inflateInit(&stream);
1211        do {
1212                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1213                stream.next_in = in;
1214                st = inflate(&stream, Z_FINISH);
1215                curpos += stream.next_in - in;
1216        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1217                 stream.total_out < sizeof(delta_head));
1218        inflateEnd(&stream);
1219        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1220                die("delta data unpack-initial failed");
1221
1222        /* Examine the initial part of the delta to figure out
1223         * the result size.
1224         */
1225        data = delta_head;
1226
1227        /* ignore base size */
1228        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1229
1230        /* Read the result size */
1231        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1232}
1233
1234static off_t get_delta_base(struct packed_git *p,
1235                                    struct pack_window **w_curs,
1236                                    off_t *curpos,
1237                                    enum object_type type,
1238                                    off_t delta_obj_offset)
1239{
1240        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1241        off_t base_offset;
1242
1243        /* use_pack() assured us we have [base_info, base_info + 20)
1244         * as a range that we can look at without walking off the
1245         * end of the mapped window.  Its actually the hash size
1246         * that is assured.  An OFS_DELTA longer than the hash size
1247         * is stupid, as then a REF_DELTA would be smaller to store.
1248         */
1249        if (type == OBJ_OFS_DELTA) {
1250                unsigned used = 0;
1251                unsigned char c = base_info[used++];
1252                base_offset = c & 127;
1253                while (c & 128) {
1254                        base_offset += 1;
1255                        if (!base_offset || MSB(base_offset, 7))
1256                                die("offset value overflow for delta base object");
1257                        c = base_info[used++];
1258                        base_offset = (base_offset << 7) + (c & 127);
1259                }
1260                base_offset = delta_obj_offset - base_offset;
1261                if (base_offset >= delta_obj_offset)
1262                        die("delta base offset out of bound");
1263                *curpos += used;
1264        } else if (type == OBJ_REF_DELTA) {
1265                /* The base entry _must_ be in the same pack */
1266                base_offset = find_pack_entry_one(base_info, p);
1267                if (!base_offset)
1268                        die("failed to find delta-pack base object %s",
1269                                sha1_to_hex(base_info));
1270                *curpos += 20;
1271        } else
1272                die("I am totally screwed");
1273        return base_offset;
1274}
1275
1276/* forward declaration for a mutually recursive function */
1277static int packed_object_info(struct packed_git *p, off_t offset,
1278                              unsigned long *sizep);
1279
1280static int packed_delta_info(struct packed_git *p,
1281                             struct pack_window **w_curs,
1282                             off_t curpos,
1283                             enum object_type type,
1284                             off_t obj_offset,
1285                             unsigned long *sizep)
1286{
1287        off_t base_offset;
1288
1289        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1290        type = packed_object_info(p, base_offset, NULL);
1291
1292        /* We choose to only get the type of the base object and
1293         * ignore potentially corrupt pack file that expects the delta
1294         * based on a base with a wrong size.  This saves tons of
1295         * inflate() calls.
1296         */
1297        if (sizep)
1298                *sizep = get_size_from_delta(p, w_curs, curpos);
1299
1300        return type;
1301}
1302
1303static int unpack_object_header(struct packed_git *p,
1304                                struct pack_window **w_curs,
1305                                off_t *curpos,
1306                                unsigned long *sizep)
1307{
1308        unsigned char *base;
1309        unsigned int left;
1310        unsigned long used;
1311        enum object_type type;
1312
1313        /* use_pack() assures us we have [base, base + 20) available
1314         * as a range that we can look at at.  (Its actually the hash
1315         * size that is assured.)  With our object header encoding
1316         * the maximum deflated object size is 2^137, which is just
1317         * insane, so we know won't exceed what we have been given.
1318         */
1319        base = use_pack(p, w_curs, *curpos, &left);
1320        used = unpack_object_header_gently(base, left, &type, sizep);
1321        if (!used)
1322                die("object offset outside of pack file");
1323        *curpos += used;
1324
1325        return type;
1326}
1327
1328const char *packed_object_info_detail(struct packed_git *p,
1329                                      off_t obj_offset,
1330                                      unsigned long *size,
1331                                      unsigned long *store_size,
1332                                      unsigned int *delta_chain_length,
1333                                      unsigned char *base_sha1)
1334{
1335        struct pack_window *w_curs = NULL;
1336        off_t curpos;
1337        unsigned long dummy;
1338        unsigned char *next_sha1;
1339        enum object_type type;
1340
1341        *delta_chain_length = 0;
1342        curpos = obj_offset;
1343        type = unpack_object_header(p, &w_curs, &curpos, size);
1344
1345        for (;;) {
1346                switch (type) {
1347                default:
1348                        die("pack %s contains unknown object type %d",
1349                            p->pack_name, type);
1350                case OBJ_COMMIT:
1351                case OBJ_TREE:
1352                case OBJ_BLOB:
1353                case OBJ_TAG:
1354                        *store_size = 0; /* notyet */
1355                        unuse_pack(&w_curs);
1356                        return typename(type);
1357                case OBJ_OFS_DELTA:
1358                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1359                        if (*delta_chain_length == 0) {
1360                                /* TODO: find base_sha1 as pointed by curpos */
1361                                hashclr(base_sha1);
1362                        }
1363                        break;
1364                case OBJ_REF_DELTA:
1365                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1366                        if (*delta_chain_length == 0)
1367                                hashcpy(base_sha1, next_sha1);
1368                        obj_offset = find_pack_entry_one(next_sha1, p);
1369                        break;
1370                }
1371                (*delta_chain_length)++;
1372                curpos = obj_offset;
1373                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1374        }
1375}
1376
1377static int packed_object_info(struct packed_git *p, off_t obj_offset,
1378                              unsigned long *sizep)
1379{
1380        struct pack_window *w_curs = NULL;
1381        unsigned long size;
1382        off_t curpos = obj_offset;
1383        enum object_type type;
1384
1385        type = unpack_object_header(p, &w_curs, &curpos, &size);
1386
1387        switch (type) {
1388        case OBJ_OFS_DELTA:
1389        case OBJ_REF_DELTA:
1390                type = packed_delta_info(p, &w_curs, curpos,
1391                                         type, obj_offset, sizep);
1392                break;
1393        case OBJ_COMMIT:
1394        case OBJ_TREE:
1395        case OBJ_BLOB:
1396        case OBJ_TAG:
1397                if (sizep)
1398                        *sizep = size;
1399                break;
1400        default:
1401                die("pack %s contains unknown object type %d",
1402                    p->pack_name, type);
1403        }
1404        unuse_pack(&w_curs);
1405        return type;
1406}
1407
1408static void *unpack_compressed_entry(struct packed_git *p,
1409                                    struct pack_window **w_curs,
1410                                    off_t curpos,
1411                                    unsigned long size)
1412{
1413        int st;
1414        z_stream stream;
1415        unsigned char *buffer, *in;
1416
1417        buffer = xmalloc(size + 1);
1418        buffer[size] = 0;
1419        memset(&stream, 0, sizeof(stream));
1420        stream.next_out = buffer;
1421        stream.avail_out = size;
1422
1423        inflateInit(&stream);
1424        do {
1425                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1426                stream.next_in = in;
1427                st = inflate(&stream, Z_FINISH);
1428                curpos += stream.next_in - in;
1429        } while (st == Z_OK || st == Z_BUF_ERROR);
1430        inflateEnd(&stream);
1431        if ((st != Z_STREAM_END) || stream.total_out != size) {
1432                free(buffer);
1433                return NULL;
1434        }
1435
1436        return buffer;
1437}
1438
1439#define MAX_DELTA_CACHE (256)
1440
1441static size_t delta_base_cached;
1442
1443static struct delta_base_cache_lru_list {
1444        struct delta_base_cache_lru_list *prev;
1445        struct delta_base_cache_lru_list *next;
1446} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1447
1448static struct delta_base_cache_entry {
1449        struct delta_base_cache_lru_list lru;
1450        void *data;
1451        struct packed_git *p;
1452        off_t base_offset;
1453        unsigned long size;
1454        enum object_type type;
1455} delta_base_cache[MAX_DELTA_CACHE];
1456
1457static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1458{
1459        unsigned long hash;
1460
1461        hash = (unsigned long)p + (unsigned long)base_offset;
1462        hash += (hash >> 8) + (hash >> 16);
1463        return hash % MAX_DELTA_CACHE;
1464}
1465
1466static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1467        unsigned long *base_size, enum object_type *type, int keep_cache)
1468{
1469        void *ret;
1470        unsigned long hash = pack_entry_hash(p, base_offset);
1471        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1472
1473        ret = ent->data;
1474        if (ret && ent->p == p && ent->base_offset == base_offset)
1475                goto found_cache_entry;
1476        return unpack_entry(p, base_offset, type, base_size);
1477
1478found_cache_entry:
1479        if (!keep_cache) {
1480                ent->data = NULL;
1481                ent->lru.next->prev = ent->lru.prev;
1482                ent->lru.prev->next = ent->lru.next;
1483                delta_base_cached -= ent->size;
1484        }
1485        else {
1486                ret = xmalloc(ent->size + 1);
1487                memcpy(ret, ent->data, ent->size);
1488                ((char *)ret)[ent->size] = 0;
1489        }
1490        *type = ent->type;
1491        *base_size = ent->size;
1492        return ret;
1493}
1494
1495static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1496{
1497        if (ent->data) {
1498                free(ent->data);
1499                ent->data = NULL;
1500                ent->lru.next->prev = ent->lru.prev;
1501                ent->lru.prev->next = ent->lru.next;
1502                delta_base_cached -= ent->size;
1503        }
1504}
1505
1506static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1507        void *base, unsigned long base_size, enum object_type type)
1508{
1509        unsigned long hash = pack_entry_hash(p, base_offset);
1510        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1511        struct delta_base_cache_lru_list *lru;
1512
1513        release_delta_base_cache(ent);
1514        delta_base_cached += base_size;
1515
1516        for (lru = delta_base_cache_lru.next;
1517             delta_base_cached > delta_base_cache_limit
1518             && lru != &delta_base_cache_lru;
1519             lru = lru->next) {
1520                struct delta_base_cache_entry *f = (void *)lru;
1521                if (f->type == OBJ_BLOB)
1522                        release_delta_base_cache(f);
1523        }
1524        for (lru = delta_base_cache_lru.next;
1525             delta_base_cached > delta_base_cache_limit
1526             && lru != &delta_base_cache_lru;
1527             lru = lru->next) {
1528                struct delta_base_cache_entry *f = (void *)lru;
1529                release_delta_base_cache(f);
1530        }
1531
1532        ent->p = p;
1533        ent->base_offset = base_offset;
1534        ent->type = type;
1535        ent->data = base;
1536        ent->size = base_size;
1537        ent->lru.next = &delta_base_cache_lru;
1538        ent->lru.prev = delta_base_cache_lru.prev;
1539        delta_base_cache_lru.prev->next = &ent->lru;
1540        delta_base_cache_lru.prev = &ent->lru;
1541}
1542
1543static void *unpack_delta_entry(struct packed_git *p,
1544                                struct pack_window **w_curs,
1545                                off_t curpos,
1546                                unsigned long delta_size,
1547                                off_t obj_offset,
1548                                enum object_type *type,
1549                                unsigned long *sizep)
1550{
1551        void *delta_data, *result, *base;
1552        unsigned long base_size;
1553        off_t base_offset;
1554
1555        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1556        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1557        if (!base)
1558                die("failed to read delta base object"
1559                    " at %"PRIuMAX" from %s",
1560                    (uintmax_t)base_offset, p->pack_name);
1561
1562        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1563        result = patch_delta(base, base_size,
1564                             delta_data, delta_size,
1565                             sizep);
1566        if (!result)
1567                die("failed to apply delta");
1568        free(delta_data);
1569        add_delta_base_cache(p, base_offset, base, base_size, *type);
1570        return result;
1571}
1572
1573void *unpack_entry(struct packed_git *p, off_t obj_offset,
1574                   enum object_type *type, unsigned long *sizep)
1575{
1576        struct pack_window *w_curs = NULL;
1577        off_t curpos = obj_offset;
1578        void *data;
1579
1580        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1581        switch (*type) {
1582        case OBJ_OFS_DELTA:
1583        case OBJ_REF_DELTA:
1584                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1585                                          obj_offset, type, sizep);
1586                break;
1587        case OBJ_COMMIT:
1588        case OBJ_TREE:
1589        case OBJ_BLOB:
1590        case OBJ_TAG:
1591                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1592                break;
1593        default:
1594                die("unknown object type %i in %s", *type, p->pack_name);
1595        }
1596        unuse_pack(&w_curs);
1597        return data;
1598}
1599
1600const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1601                                            uint32_t n)
1602{
1603        const unsigned char *index = p->index_data;
1604        if (!index) {
1605                if (open_pack_index(p))
1606                        return NULL;
1607                index = p->index_data;
1608        }
1609        if (n >= p->num_objects)
1610                return NULL;
1611        index += 4 * 256;
1612        if (p->index_version == 1) {
1613                return index + 24 * n + 4;
1614        } else {
1615                index += 8;
1616                return index + 20 * n;
1617        }
1618}
1619
1620static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1621{
1622        const unsigned char *index = p->index_data;
1623        index += 4 * 256;
1624        if (p->index_version == 1) {
1625                return ntohl(*((uint32_t *)(index + 24 * n)));
1626        } else {
1627                uint32_t off;
1628                index += 8 + p->num_objects * (20 + 4);
1629                off = ntohl(*((uint32_t *)(index + 4 * n)));
1630                if (!(off & 0x80000000))
1631                        return off;
1632                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1633                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1634                                   ntohl(*((uint32_t *)(index + 4)));
1635        }
1636}
1637
1638off_t find_pack_entry_one(const unsigned char *sha1,
1639                                  struct packed_git *p)
1640{
1641        const uint32_t *level1_ofs = p->index_data;
1642        const unsigned char *index = p->index_data;
1643        unsigned hi, lo;
1644
1645        if (!index) {
1646                if (open_pack_index(p))
1647                        return 0;
1648                level1_ofs = p->index_data;
1649                index = p->index_data;
1650        }
1651        if (p->index_version > 1) {
1652                level1_ofs += 2;
1653                index += 8;
1654        }
1655        index += 4 * 256;
1656        hi = ntohl(level1_ofs[*sha1]);
1657        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1658
1659        do {
1660                unsigned mi = (lo + hi) / 2;
1661                unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1662                int cmp = hashcmp(index + x, sha1);
1663                if (!cmp)
1664                        return nth_packed_object_offset(p, mi);
1665                if (cmp > 0)
1666                        hi = mi;
1667                else
1668                        lo = mi+1;
1669        } while (lo < hi);
1670        return 0;
1671}
1672
1673static int matches_pack_name(struct packed_git *p, const char *ig)
1674{
1675        const char *last_c, *c;
1676
1677        if (!strcmp(p->pack_name, ig))
1678                return 0;
1679
1680        for (c = p->pack_name, last_c = c; *c;)
1681                if (*c == '/')
1682                        last_c = ++c;
1683                else
1684                        ++c;
1685        if (!strcmp(last_c, ig))
1686                return 0;
1687
1688        return 1;
1689}
1690
1691static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1692{
1693        static struct packed_git *last_found = (void *)1;
1694        struct packed_git *p;
1695        off_t offset;
1696
1697        prepare_packed_git();
1698        if (!packed_git)
1699                return 0;
1700        p = (last_found == (void *)1) ? packed_git : last_found;
1701
1702        do {
1703                if (ignore_packed) {
1704                        const char **ig;
1705                        for (ig = ignore_packed; *ig; ig++)
1706                                if (!matches_pack_name(p, *ig))
1707                                        break;
1708                        if (*ig)
1709                                goto next;
1710                }
1711
1712                offset = find_pack_entry_one(sha1, p);
1713                if (offset) {
1714                        /*
1715                         * We are about to tell the caller where they can
1716                         * locate the requested object.  We better make
1717                         * sure the packfile is still here and can be
1718                         * accessed before supplying that answer, as
1719                         * it may have been deleted since the index
1720                         * was loaded!
1721                         */
1722                        if (p->pack_fd == -1 && open_packed_git(p)) {
1723                                error("packfile %s cannot be accessed", p->pack_name);
1724                                goto next;
1725                        }
1726                        e->offset = offset;
1727                        e->p = p;
1728                        hashcpy(e->sha1, sha1);
1729                        last_found = p;
1730                        return 1;
1731                }
1732
1733                next:
1734                if (p == last_found)
1735                        p = packed_git;
1736                else
1737                        p = p->next;
1738                if (p == last_found)
1739                        p = p->next;
1740        } while (p);
1741        return 0;
1742}
1743
1744struct packed_git *find_sha1_pack(const unsigned char *sha1,
1745                                  struct packed_git *packs)
1746{
1747        struct packed_git *p;
1748
1749        for (p = packs; p; p = p->next) {
1750                if (find_pack_entry_one(sha1, p))
1751                        return p;
1752        }
1753        return NULL;
1754
1755}
1756
1757static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1758{
1759        int status;
1760        unsigned long mapsize, size;
1761        void *map;
1762        z_stream stream;
1763        char hdr[32];
1764
1765        map = map_sha1_file(sha1, &mapsize);
1766        if (!map)
1767                return error("unable to find %s", sha1_to_hex(sha1));
1768        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1769                status = error("unable to unpack %s header",
1770                               sha1_to_hex(sha1));
1771        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1772                status = error("unable to parse %s header", sha1_to_hex(sha1));
1773        else if (sizep)
1774                *sizep = size;
1775        inflateEnd(&stream);
1776        munmap(map, mapsize);
1777        return status;
1778}
1779
1780int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1781{
1782        struct pack_entry e;
1783
1784        if (!find_pack_entry(sha1, &e, NULL)) {
1785                reprepare_packed_git();
1786                if (!find_pack_entry(sha1, &e, NULL))
1787                        return sha1_loose_object_info(sha1, sizep);
1788        }
1789        return packed_object_info(e.p, e.offset, sizep);
1790}
1791
1792static void *read_packed_sha1(const unsigned char *sha1,
1793                              enum object_type *type, unsigned long *size)
1794{
1795        struct pack_entry e;
1796
1797        if (!find_pack_entry(sha1, &e, NULL))
1798                return NULL;
1799        else
1800                return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1801}
1802
1803/*
1804 * This is meant to hold a *small* number of objects that you would
1805 * want read_sha1_file() to be able to return, but yet you do not want
1806 * to write them into the object store (e.g. a browse-only
1807 * application).
1808 */
1809static struct cached_object {
1810        unsigned char sha1[20];
1811        enum object_type type;
1812        void *buf;
1813        unsigned long size;
1814} *cached_objects;
1815static int cached_object_nr, cached_object_alloc;
1816
1817static struct cached_object *find_cached_object(const unsigned char *sha1)
1818{
1819        int i;
1820        struct cached_object *co = cached_objects;
1821
1822        for (i = 0; i < cached_object_nr; i++, co++) {
1823                if (!hashcmp(co->sha1, sha1))
1824                        return co;
1825        }
1826        return NULL;
1827}
1828
1829int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1830                      unsigned char *sha1)
1831{
1832        struct cached_object *co;
1833
1834        hash_sha1_file(buf, len, typename(type), sha1);
1835        if (has_sha1_file(sha1) || find_cached_object(sha1))
1836                return 0;
1837        if (cached_object_alloc <= cached_object_nr) {
1838                cached_object_alloc = alloc_nr(cached_object_alloc);
1839                cached_objects = xrealloc(cached_objects,
1840                                          sizeof(*cached_objects) *
1841                                          cached_object_alloc);
1842        }
1843        co = &cached_objects[cached_object_nr++];
1844        co->size = len;
1845        co->type = type;
1846        co->buf = xmalloc(len);
1847        memcpy(co->buf, buf, len);
1848        hashcpy(co->sha1, sha1);
1849        return 0;
1850}
1851
1852void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1853                     unsigned long *size)
1854{
1855        unsigned long mapsize;
1856        void *map, *buf;
1857        struct cached_object *co;
1858
1859        co = find_cached_object(sha1);
1860        if (co) {
1861                buf = xmalloc(co->size + 1);
1862                memcpy(buf, co->buf, co->size);
1863                ((char*)buf)[co->size] = 0;
1864                *type = co->type;
1865                *size = co->size;
1866                return buf;
1867        }
1868
1869        buf = read_packed_sha1(sha1, type, size);
1870        if (buf)
1871                return buf;
1872        map = map_sha1_file(sha1, &mapsize);
1873        if (map) {
1874                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1875                munmap(map, mapsize);
1876                return buf;
1877        }
1878        reprepare_packed_git();
1879        return read_packed_sha1(sha1, type, size);
1880}
1881
1882void *read_object_with_reference(const unsigned char *sha1,
1883                                 const char *required_type_name,
1884                                 unsigned long *size,
1885                                 unsigned char *actual_sha1_return)
1886{
1887        enum object_type type, required_type;
1888        void *buffer;
1889        unsigned long isize;
1890        unsigned char actual_sha1[20];
1891
1892        required_type = type_from_string(required_type_name);
1893        hashcpy(actual_sha1, sha1);
1894        while (1) {
1895                int ref_length = -1;
1896                const char *ref_type = NULL;
1897
1898                buffer = read_sha1_file(actual_sha1, &type, &isize);
1899                if (!buffer)
1900                        return NULL;
1901                if (type == required_type) {
1902                        *size = isize;
1903                        if (actual_sha1_return)
1904                                hashcpy(actual_sha1_return, actual_sha1);
1905                        return buffer;
1906                }
1907                /* Handle references */
1908                else if (type == OBJ_COMMIT)
1909                        ref_type = "tree ";
1910                else if (type == OBJ_TAG)
1911                        ref_type = "object ";
1912                else {
1913                        free(buffer);
1914                        return NULL;
1915                }
1916                ref_length = strlen(ref_type);
1917
1918                if (memcmp(buffer, ref_type, ref_length) ||
1919                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1920                        free(buffer);
1921                        return NULL;
1922                }
1923                free(buffer);
1924                /* Now we have the ID of the referred-to object in
1925                 * actual_sha1.  Check again. */
1926        }
1927}
1928
1929static void write_sha1_file_prepare(const void *buf, unsigned long len,
1930                                    const char *type, unsigned char *sha1,
1931                                    char *hdr, int *hdrlen)
1932{
1933        SHA_CTX c;
1934
1935        /* Generate the header */
1936        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1937
1938        /* Sha1.. */
1939        SHA1_Init(&c);
1940        SHA1_Update(&c, hdr, *hdrlen);
1941        SHA1_Update(&c, buf, len);
1942        SHA1_Final(sha1, &c);
1943}
1944
1945/*
1946 * Link the tempfile to the final place, possibly creating the
1947 * last directory level as you do so.
1948 *
1949 * Returns the errno on failure, 0 on success.
1950 */
1951static int link_temp_to_file(const char *tmpfile, const char *filename)
1952{
1953        int ret;
1954        char *dir;
1955
1956        if (!link(tmpfile, filename))
1957                return 0;
1958
1959        /*
1960         * Try to mkdir the last path component if that failed.
1961         *
1962         * Re-try the "link()" regardless of whether the mkdir
1963         * succeeds, since a race might mean that somebody
1964         * else succeeded.
1965         */
1966        ret = errno;
1967        dir = strrchr(filename, '/');
1968        if (dir) {
1969                *dir = 0;
1970                if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1971                        *dir = '/';
1972                        return -2;
1973                }
1974                *dir = '/';
1975                if (!link(tmpfile, filename))
1976                        return 0;
1977                ret = errno;
1978        }
1979        return ret;
1980}
1981
1982/*
1983 * Move the just written object into its final resting place
1984 */
1985int move_temp_to_file(const char *tmpfile, const char *filename)
1986{
1987        int ret = link_temp_to_file(tmpfile, filename);
1988
1989        /*
1990         * Coda hack - coda doesn't like cross-directory links,
1991         * so we fall back to a rename, which will mean that it
1992         * won't be able to check collisions, but that's not a
1993         * big deal.
1994         *
1995         * The same holds for FAT formatted media.
1996         *
1997         * When this succeeds, we just return 0. We have nothing
1998         * left to unlink.
1999         */
2000        if (ret && ret != EEXIST) {
2001                if (!rename(tmpfile, filename))
2002                        return 0;
2003                ret = errno;
2004        }
2005        unlink(tmpfile);
2006        if (ret) {
2007                if (ret != EEXIST) {
2008                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2009                }
2010                /* FIXME!!! Collision check here ? */
2011        }
2012
2013        return 0;
2014}
2015
2016static int write_buffer(int fd, const void *buf, size_t len)
2017{
2018        if (write_in_full(fd, buf, len) < 0)
2019                return error("file write error (%s)", strerror(errno));
2020        return 0;
2021}
2022
2023int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2024                   unsigned char *sha1)
2025{
2026        char hdr[32];
2027        int hdrlen;
2028        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2029        return 0;
2030}
2031
2032int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2033{
2034        int size, ret;
2035        unsigned char *compressed;
2036        z_stream stream;
2037        unsigned char sha1[20];
2038        char *filename;
2039        static char tmpfile[PATH_MAX];
2040        char hdr[32];
2041        int fd, hdrlen;
2042
2043        /* Normally if we have it in the pack then we do not bother writing
2044         * it out into .git/objects/??/?{38} file.
2045         */
2046        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2047        filename = sha1_file_name(sha1);
2048        if (returnsha1)
2049                hashcpy(returnsha1, sha1);
2050        if (has_sha1_file(sha1))
2051                return 0;
2052        fd = open(filename, O_RDONLY);
2053        if (fd >= 0) {
2054                /*
2055                 * FIXME!!! We might do collision checking here, but we'd
2056                 * need to uncompress the old file and check it. Later.
2057                 */
2058                close(fd);
2059                return 0;
2060        }
2061
2062        if (errno != ENOENT) {
2063                return error("sha1 file %s: %s\n", filename, strerror(errno));
2064        }
2065
2066        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2067
2068        fd = mkstemp(tmpfile);
2069        if (fd < 0) {
2070                if (errno == EPERM)
2071                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2072                else
2073                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2074        }
2075
2076        /* Set it up */
2077        memset(&stream, 0, sizeof(stream));
2078        deflateInit(&stream, zlib_compression_level);
2079        size = 8 + deflateBound(&stream, len+hdrlen);
2080        compressed = xmalloc(size);
2081
2082        /* Compress it */
2083        stream.next_out = compressed;
2084        stream.avail_out = size;
2085
2086        /* First header.. */
2087        stream.next_in = (unsigned char *)hdr;
2088        stream.avail_in = hdrlen;
2089        while (deflate(&stream, 0) == Z_OK)
2090                /* nothing */;
2091
2092        /* Then the data itself.. */
2093        stream.next_in = buf;
2094        stream.avail_in = len;
2095        ret = deflate(&stream, Z_FINISH);
2096        if (ret != Z_STREAM_END)
2097                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2098
2099        ret = deflateEnd(&stream);
2100        if (ret != Z_OK)
2101                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2102
2103        size = stream.total_out;
2104
2105        if (write_buffer(fd, compressed, size) < 0)
2106                die("unable to write sha1 file");
2107        fchmod(fd, 0444);
2108        if (close(fd))
2109                die("unable to write sha1 file");
2110        free(compressed);
2111
2112        return move_temp_to_file(tmpfile, filename);
2113}
2114
2115/*
2116 * We need to unpack and recompress the object for writing
2117 * it out to a different file.
2118 */
2119static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2120{
2121        size_t size;
2122        z_stream stream;
2123        unsigned char *unpacked;
2124        unsigned long len;
2125        enum object_type type;
2126        char hdr[32];
2127        int hdrlen;
2128        void *buf;
2129
2130        /* need to unpack and recompress it by itself */
2131        unpacked = read_packed_sha1(sha1, &type, &len);
2132        if (!unpacked)
2133                error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2134
2135        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2136
2137        /* Set it up */
2138        memset(&stream, 0, sizeof(stream));
2139        deflateInit(&stream, zlib_compression_level);
2140        size = deflateBound(&stream, len + hdrlen);
2141        buf = xmalloc(size);
2142
2143        /* Compress it */
2144        stream.next_out = buf;
2145        stream.avail_out = size;
2146
2147        /* First header.. */
2148        stream.next_in = (void *)hdr;
2149        stream.avail_in = hdrlen;
2150        while (deflate(&stream, 0) == Z_OK)
2151                /* nothing */;
2152
2153        /* Then the data itself.. */
2154        stream.next_in = unpacked;
2155        stream.avail_in = len;
2156        while (deflate(&stream, Z_FINISH) == Z_OK)
2157                /* nothing */;
2158        deflateEnd(&stream);
2159        free(unpacked);
2160
2161        *objsize = stream.total_out;
2162        return buf;
2163}
2164
2165int write_sha1_to_fd(int fd, const unsigned char *sha1)
2166{
2167        int retval;
2168        unsigned long objsize;
2169        void *buf = map_sha1_file(sha1, &objsize);
2170
2171        if (buf) {
2172                retval = write_buffer(fd, buf, objsize);
2173                munmap(buf, objsize);
2174                return retval;
2175        }
2176
2177        buf = repack_object(sha1, &objsize);
2178        retval = write_buffer(fd, buf, objsize);
2179        free(buf);
2180        return retval;
2181}
2182
2183int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2184                       size_t bufsize, size_t *bufposn)
2185{
2186        char tmpfile[PATH_MAX];
2187        int local;
2188        z_stream stream;
2189        unsigned char real_sha1[20];
2190        unsigned char discard[4096];
2191        int ret;
2192        SHA_CTX c;
2193
2194        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2195
2196        local = mkstemp(tmpfile);
2197        if (local < 0) {
2198                if (errno == EPERM)
2199                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2200                else
2201                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2202        }
2203
2204        memset(&stream, 0, sizeof(stream));
2205
2206        inflateInit(&stream);
2207
2208        SHA1_Init(&c);
2209
2210        do {
2211                ssize_t size;
2212                if (*bufposn) {
2213                        stream.avail_in = *bufposn;
2214                        stream.next_in = (unsigned char *) buffer;
2215                        do {
2216                                stream.next_out = discard;
2217                                stream.avail_out = sizeof(discard);
2218                                ret = inflate(&stream, Z_SYNC_FLUSH);
2219                                SHA1_Update(&c, discard, sizeof(discard) -
2220                                            stream.avail_out);
2221                        } while (stream.avail_in && ret == Z_OK);
2222                        if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2223                                die("unable to write sha1 file");
2224                        memmove(buffer, buffer + *bufposn - stream.avail_in,
2225                                stream.avail_in);
2226                        *bufposn = stream.avail_in;
2227                        if (ret != Z_OK)
2228                                break;
2229                }
2230                size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2231                if (size <= 0) {
2232                        close(local);
2233                        unlink(tmpfile);
2234                        if (!size)
2235                                return error("Connection closed?");
2236                        perror("Reading from connection");
2237                        return -1;
2238                }
2239                *bufposn += size;
2240        } while (1);
2241        inflateEnd(&stream);
2242
2243        fchmod(local, 0444);
2244        if (close(local) != 0)
2245                die("unable to write sha1 file");
2246        SHA1_Final(real_sha1, &c);
2247        if (ret != Z_STREAM_END) {
2248                unlink(tmpfile);
2249                return error("File %s corrupted", sha1_to_hex(sha1));
2250        }
2251        if (hashcmp(sha1, real_sha1)) {
2252                unlink(tmpfile);
2253                return error("File %s has bad hash", sha1_to_hex(sha1));
2254        }
2255
2256        return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2257}
2258
2259int has_pack_index(const unsigned char *sha1)
2260{
2261        struct stat st;
2262        if (stat(sha1_pack_index_name(sha1), &st))
2263                return 0;
2264        return 1;
2265}
2266
2267int has_pack_file(const unsigned char *sha1)
2268{
2269        struct stat st;
2270        if (stat(sha1_pack_name(sha1), &st))
2271                return 0;
2272        return 1;
2273}
2274
2275int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2276{
2277        struct pack_entry e;
2278        return find_pack_entry(sha1, &e, ignore_packed);
2279}
2280
2281int has_sha1_file(const unsigned char *sha1)
2282{
2283        struct stat st;
2284        struct pack_entry e;
2285
2286        if (find_pack_entry(sha1, &e, NULL))
2287                return 1;
2288        return find_sha1_file(sha1, &st) ? 1 : 0;
2289}
2290
2291/*
2292 * reads from fd as long as possible into a supplied buffer of size bytes.
2293 * If necessary the buffer's size is increased using realloc()
2294 *
2295 * returns 0 if anything went fine and -1 otherwise
2296 *
2297 * NOTE: both buf and size may change, but even when -1 is returned
2298 * you still have to free() it yourself.
2299 */
2300int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2301{
2302        char* buf = *return_buf;
2303        unsigned long size = *return_size;
2304        ssize_t iret;
2305        unsigned long off = 0;
2306
2307        do {
2308                iret = xread(fd, buf + off, size - off);
2309                if (iret > 0) {
2310                        off += iret;
2311                        if (off == size) {
2312                                size *= 2;
2313                                buf = xrealloc(buf, size);
2314                        }
2315                }
2316        } while (iret > 0);
2317
2318        *return_buf = buf;
2319        *return_size = off;
2320
2321        if (iret < 0)
2322                return -1;
2323        return 0;
2324}
2325
2326int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2327{
2328        unsigned long size = 4096;
2329        char *buf = xmalloc(size);
2330        int ret;
2331
2332        if (read_pipe(fd, &buf, &size)) {
2333                free(buf);
2334                return -1;
2335        }
2336
2337        if (!type)
2338                type = blob_type;
2339        if (write_object)
2340                ret = write_sha1_file(buf, size, type, sha1);
2341        else
2342                ret = hash_sha1_file(buf, size, type, sha1);
2343        free(buf);
2344        return ret;
2345}
2346
2347int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2348             enum object_type type, const char *path)
2349{
2350        size_t size = xsize_t(st->st_size);
2351        void *buf = NULL;
2352        int ret, re_allocated = 0;
2353
2354        if (size)
2355                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2356        close(fd);
2357
2358        if (!type)
2359                type = OBJ_BLOB;
2360
2361        /*
2362         * Convert blobs to git internal format
2363         */
2364        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2365                unsigned long nsize = size;
2366                char *nbuf = convert_to_git(path, buf, &nsize);
2367                if (nbuf) {
2368                        munmap(buf, size);
2369                        size = nsize;
2370                        buf = nbuf;
2371                        re_allocated = 1;
2372                }
2373        }
2374
2375        if (write_object)
2376                ret = write_sha1_file(buf, size, typename(type), sha1);
2377        else
2378                ret = hash_sha1_file(buf, size, typename(type), sha1);
2379        if (re_allocated) {
2380                free(buf);
2381                return ret;
2382        }
2383        if (size)
2384                munmap(buf, size);
2385        return ret;
2386}
2387
2388int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2389{
2390        int fd;
2391        char *target;
2392        size_t len;
2393
2394        switch (st->st_mode & S_IFMT) {
2395        case S_IFREG:
2396                fd = open(path, O_RDONLY);
2397                if (fd < 0)
2398                        return error("open(\"%s\"): %s", path,
2399                                     strerror(errno));
2400                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2401                        return error("%s: failed to insert into database",
2402                                     path);
2403                break;
2404        case S_IFLNK:
2405                len = xsize_t(st->st_size);
2406                target = xmalloc(len + 1);
2407                if (readlink(path, target, len + 1) != st->st_size) {
2408                        char *errstr = strerror(errno);
2409                        free(target);
2410                        return error("readlink(\"%s\"): %s", path,
2411                                     errstr);
2412                }
2413                if (!write_object)
2414                        hash_sha1_file(target, len, blob_type, sha1);
2415                else if (write_sha1_file(target, len, blob_type, sha1))
2416                        return error("%s: failed to insert into database",
2417                                     path);
2418                free(target);
2419                break;
2420        case S_IFDIR:
2421                return resolve_gitlink_ref(path, "HEAD", sha1);
2422        default:
2423                return error("%s: unsupported file type", path);
2424        }
2425        return 0;
2426}
2427
2428int read_pack_header(int fd, struct pack_header *header)
2429{
2430        char *c = (char*)header;
2431        ssize_t remaining = sizeof(struct pack_header);
2432        do {
2433                ssize_t r = xread(fd, c, remaining);
2434                if (r <= 0)
2435                        /* "eof before pack header was fully read" */
2436                        return PH_ERROR_EOF;
2437                remaining -= r;
2438                c += r;
2439        } while (remaining > 0);
2440        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2441                /* "protocol error (pack signature mismatch detected)" */
2442                return PH_ERROR_PACK_SIGNATURE;
2443        if (!pack_version_ok(header->hdr_version))
2444                /* "protocol error (pack version unsupported)" */
2445                return PH_ERROR_PROTOCOL;
2446        return 0;
2447}