sha1_file.con commit git-archive: work in bare repos (ddff856)
   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#include "pack-revindex.h"
  18#include "sha1-lookup.h"
  19
  20#ifndef O_NOATIME
  21#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  22#define O_NOATIME 01000000
  23#else
  24#define O_NOATIME 0
  25#endif
  26#endif
  27
  28#ifdef NO_C99_FORMAT
  29#define SZ_FMT "lu"
  30static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
  31#else
  32#define SZ_FMT "zu"
  33static size_t sz_fmt(size_t s) { return s; }
  34#endif
  35
  36const unsigned char null_sha1[20];
  37
  38const signed char hexval_table[256] = {
  39         -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
  40         -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
  41         -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
  42         -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
  43         -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
  44         -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
  45          0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
  46          8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
  47         -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
  48         -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
  49         -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
  50         -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
  51         -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
  52         -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
  53         -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
  54         -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
  55         -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
  56         -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
  57         -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
  58         -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
  59         -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
  60         -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
  61         -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
  62         -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
  63         -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
  64         -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
  65         -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
  66         -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
  67         -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
  68         -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
  69         -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
  70         -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
  71};
  72
  73int get_sha1_hex(const char *hex, unsigned char *sha1)
  74{
  75        int i;
  76        for (i = 0; i < 20; i++) {
  77                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  78                if (val & ~0xff)
  79                        return -1;
  80                *sha1++ = val;
  81                hex += 2;
  82        }
  83        return 0;
  84}
  85
  86static inline int offset_1st_component(const char *path)
  87{
  88        if (has_dos_drive_prefix(path))
  89                return 2 + (path[2] == '/');
  90        return *path == '/';
  91}
  92
  93int safe_create_leading_directories(char *path)
  94{
  95        char *pos = path + offset_1st_component(path);
  96        struct stat st;
  97
  98        while (pos) {
  99                pos = strchr(pos, '/');
 100                if (!pos)
 101                        break;
 102                *pos = 0;
 103                if (!stat(path, &st)) {
 104                        /* path exists */
 105                        if (!S_ISDIR(st.st_mode)) {
 106                                *pos = '/';
 107                                return -3;
 108                        }
 109                }
 110                else if (mkdir(path, 0777)) {
 111                        *pos = '/';
 112                        return -1;
 113                }
 114                else if (adjust_shared_perm(path)) {
 115                        *pos = '/';
 116                        return -2;
 117                }
 118                *pos++ = '/';
 119        }
 120        return 0;
 121}
 122
 123int safe_create_leading_directories_const(const char *path)
 124{
 125        /* path points to cache entries, so xstrdup before messing with it */
 126        char *buf = xstrdup(path);
 127        int result = safe_create_leading_directories(buf);
 128        free(buf);
 129        return result;
 130}
 131
 132char *sha1_to_hex(const unsigned char *sha1)
 133{
 134        static int bufno;
 135        static char hexbuffer[4][50];
 136        static const char hex[] = "0123456789abcdef";
 137        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 138        int i;
 139
 140        for (i = 0; i < 20; i++) {
 141                unsigned int val = *sha1++;
 142                *buf++ = hex[val >> 4];
 143                *buf++ = hex[val & 0xf];
 144        }
 145        *buf = '\0';
 146
 147        return buffer;
 148}
 149
 150static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 151{
 152        int i;
 153        for (i = 0; i < 20; i++) {
 154                static char hex[] = "0123456789abcdef";
 155                unsigned int val = sha1[i];
 156                char *pos = pathbuf + i*2 + (i > 0);
 157                *pos++ = hex[val >> 4];
 158                *pos = hex[val & 0xf];
 159        }
 160}
 161
 162/*
 163 * NOTE! This returns a statically allocated buffer, so you have to be
 164 * careful about using it. Do an "xstrdup()" if you need to save the
 165 * filename.
 166 *
 167 * Also note that this returns the location for creating.  Reading
 168 * SHA1 file can happen from any alternate directory listed in the
 169 * DB_ENVIRONMENT environment variable if it is not found in
 170 * the primary object database.
 171 */
 172char *sha1_file_name(const unsigned char *sha1)
 173{
 174        static char *name, *base;
 175
 176        if (!base) {
 177                const char *sha1_file_directory = get_object_directory();
 178                int len = strlen(sha1_file_directory);
 179                base = xmalloc(len + 60);
 180                memcpy(base, sha1_file_directory, len);
 181                memset(base+len, 0, 60);
 182                base[len] = '/';
 183                base[len+3] = '/';
 184                name = base + len + 1;
 185        }
 186        fill_sha1_path(name, sha1);
 187        return base;
 188}
 189
 190static char *sha1_get_pack_name(const unsigned char *sha1,
 191                                char **name, char **base, const char *which)
 192{
 193        static const char hex[] = "0123456789abcdef";
 194        char *buf;
 195        int i;
 196
 197        if (!*base) {
 198                const char *sha1_file_directory = get_object_directory();
 199                int len = strlen(sha1_file_directory);
 200                *base = xmalloc(len + 60);
 201                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 202                        sha1_file_directory, which);
 203                *name = *base + len + 11;
 204        }
 205
 206        buf = *name;
 207
 208        for (i = 0; i < 20; i++) {
 209                unsigned int val = *sha1++;
 210                *buf++ = hex[val >> 4];
 211                *buf++ = hex[val & 0xf];
 212        }
 213
 214        return *base;
 215}
 216
 217char *sha1_pack_name(const unsigned char *sha1)
 218{
 219        static char *name, *base;
 220
 221        return sha1_get_pack_name(sha1, &name, &base, "pack");
 222}
 223
 224char *sha1_pack_index_name(const unsigned char *sha1)
 225{
 226        static char *name, *base;
 227
 228        return sha1_get_pack_name(sha1, &name, &base, "idx");
 229}
 230
 231struct alternate_object_database *alt_odb_list;
 232static struct alternate_object_database **alt_odb_tail;
 233
 234static void read_info_alternates(const char * alternates, int depth);
 235
 236/*
 237 * Prepare alternate object database registry.
 238 *
 239 * The variable alt_odb_list points at the list of struct
 240 * alternate_object_database.  The elements on this list come from
 241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 243 * whose contents is similar to that environment variable but can be
 244 * LF separated.  Its base points at a statically allocated buffer that
 245 * contains "/the/directory/corresponding/to/.git/objects/...", while
 246 * its name points just after the slash at the end of ".git/objects/"
 247 * in the example above, and has enough space to hold 40-byte hex
 248 * SHA1, an extra slash for the first level indirection, and the
 249 * terminating NUL.
 250 */
 251static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 252{
 253        struct stat st;
 254        const char *objdir = get_object_directory();
 255        struct alternate_object_database *ent;
 256        struct alternate_object_database *alt;
 257        /* 43 = 40-byte + 2 '/' + terminating NUL */
 258        int pfxlen = len;
 259        int entlen = pfxlen + 43;
 260        int base_len = -1;
 261
 262        if (!is_absolute_path(entry) && relative_base) {
 263                /* Relative alt-odb */
 264                if (base_len < 0)
 265                        base_len = strlen(relative_base) + 1;
 266                entlen += base_len;
 267                pfxlen += base_len;
 268        }
 269        ent = xmalloc(sizeof(*ent) + entlen);
 270
 271        if (!is_absolute_path(entry) && relative_base) {
 272                memcpy(ent->base, relative_base, base_len - 1);
 273                ent->base[base_len - 1] = '/';
 274                memcpy(ent->base + base_len, entry, len);
 275        }
 276        else
 277                memcpy(ent->base, entry, pfxlen);
 278
 279        ent->name = ent->base + pfxlen + 1;
 280        ent->base[pfxlen + 3] = '/';
 281        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 282
 283        /* Detect cases where alternate disappeared */
 284        if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
 285                error("object directory %s does not exist; "
 286                      "check .git/objects/info/alternates.",
 287                      ent->base);
 288                free(ent);
 289                return -1;
 290        }
 291
 292        /* Prevent the common mistake of listing the same
 293         * thing twice, or object directory itself.
 294         */
 295        for (alt = alt_odb_list; alt; alt = alt->next) {
 296                if (!memcmp(ent->base, alt->base, pfxlen)) {
 297                        free(ent);
 298                        return -1;
 299                }
 300        }
 301        if (!memcmp(ent->base, objdir, pfxlen)) {
 302                free(ent);
 303                return -1;
 304        }
 305
 306        /* add the alternate entry */
 307        *alt_odb_tail = ent;
 308        alt_odb_tail = &(ent->next);
 309        ent->next = NULL;
 310
 311        /* recursively add alternates */
 312        read_info_alternates(ent->base, depth + 1);
 313
 314        ent->base[pfxlen] = '/';
 315
 316        return 0;
 317}
 318
 319static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 320                                 const char *relative_base, int depth)
 321{
 322        const char *cp, *last;
 323
 324        if (depth > 5) {
 325                error("%s: ignoring alternate object stores, nesting too deep.",
 326                                relative_base);
 327                return;
 328        }
 329
 330        last = alt;
 331        while (last < ep) {
 332                cp = last;
 333                if (cp < ep && *cp == '#') {
 334                        while (cp < ep && *cp != sep)
 335                                cp++;
 336                        last = cp + 1;
 337                        continue;
 338                }
 339                while (cp < ep && *cp != sep)
 340                        cp++;
 341                if (last != cp) {
 342                        if (!is_absolute_path(last) && depth) {
 343                                error("%s: ignoring relative alternate object store %s",
 344                                                relative_base, last);
 345                        } else {
 346                                link_alt_odb_entry(last, cp - last,
 347                                                relative_base, depth);
 348                        }
 349                }
 350                while (cp < ep && *cp == sep)
 351                        cp++;
 352                last = cp;
 353        }
 354}
 355
 356static void read_info_alternates(const char * relative_base, int depth)
 357{
 358        char *map;
 359        size_t mapsz;
 360        struct stat st;
 361        const char alt_file_name[] = "info/alternates";
 362        /* Given that relative_base is no longer than PATH_MAX,
 363           ensure that "path" has enough space to append "/", the
 364           file name, "info/alternates", and a trailing NUL.  */
 365        char path[PATH_MAX + 1 + sizeof alt_file_name];
 366        int fd;
 367
 368        sprintf(path, "%s/%s", relative_base, alt_file_name);
 369        fd = open(path, O_RDONLY);
 370        if (fd < 0)
 371                return;
 372        if (fstat(fd, &st) || (st.st_size == 0)) {
 373                close(fd);
 374                return;
 375        }
 376        mapsz = xsize_t(st.st_size);
 377        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 378        close(fd);
 379
 380        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 381
 382        munmap(map, mapsz);
 383}
 384
 385void add_to_alternates_file(const char *reference)
 386{
 387        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 388        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
 389        char *alt = mkpath("%s/objects\n", reference);
 390        write_or_die(fd, alt, strlen(alt));
 391        if (commit_lock_file(lock))
 392                die("could not close alternates file");
 393        if (alt_odb_tail)
 394                link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
 395}
 396
 397void prepare_alt_odb(void)
 398{
 399        const char *alt;
 400
 401        if (alt_odb_tail)
 402                return;
 403
 404        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 405        if (!alt) alt = "";
 406
 407        alt_odb_tail = &alt_odb_list;
 408        link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
 409
 410        read_info_alternates(get_object_directory(), 0);
 411}
 412
 413static int has_loose_object(const unsigned char *sha1)
 414{
 415        char *name = sha1_file_name(sha1);
 416        struct alternate_object_database *alt;
 417
 418        if (!access(name, F_OK))
 419                return 1;
 420        prepare_alt_odb();
 421        for (alt = alt_odb_list; alt; alt = alt->next) {
 422                name = alt->name;
 423                fill_sha1_path(name, sha1);
 424                if (!access(alt->base, F_OK))
 425                        return 1;
 426        }
 427        return 0;
 428}
 429
 430static unsigned int pack_used_ctr;
 431static unsigned int pack_mmap_calls;
 432static unsigned int peak_pack_open_windows;
 433static unsigned int pack_open_windows;
 434static size_t peak_pack_mapped;
 435static size_t pack_mapped;
 436struct packed_git *packed_git;
 437
 438void pack_report(void)
 439{
 440        fprintf(stderr,
 441                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 442                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 443                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 444                sz_fmt(getpagesize()),
 445                sz_fmt(packed_git_window_size),
 446                sz_fmt(packed_git_limit));
 447        fprintf(stderr,
 448                "pack_report: pack_used_ctr            = %10u\n"
 449                "pack_report: pack_mmap_calls          = %10u\n"
 450                "pack_report: pack_open_windows        = %10u / %10u\n"
 451                "pack_report: pack_mapped              = "
 452                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 453                pack_used_ctr,
 454                pack_mmap_calls,
 455                pack_open_windows, peak_pack_open_windows,
 456                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 457}
 458
 459static int check_packed_git_idx(const char *path,  struct packed_git *p)
 460{
 461        void *idx_map;
 462        struct pack_idx_header *hdr;
 463        size_t idx_size;
 464        uint32_t version, nr, i, *index;
 465        int fd = open(path, O_RDONLY);
 466        struct stat st;
 467
 468        if (fd < 0)
 469                return -1;
 470        if (fstat(fd, &st)) {
 471                close(fd);
 472                return -1;
 473        }
 474        idx_size = xsize_t(st.st_size);
 475        if (idx_size < 4 * 256 + 20 + 20) {
 476                close(fd);
 477                return error("index file %s is too small", path);
 478        }
 479        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 480        close(fd);
 481
 482        hdr = idx_map;
 483        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 484                version = ntohl(hdr->idx_version);
 485                if (version < 2 || version > 2) {
 486                        munmap(idx_map, idx_size);
 487                        return error("index file %s is version %"PRIu32
 488                                     " and is not supported by this binary"
 489                                     " (try upgrading GIT to a newer version)",
 490                                     path, version);
 491                }
 492        } else
 493                version = 1;
 494
 495        nr = 0;
 496        index = idx_map;
 497        if (version > 1)
 498                index += 2;  /* skip index header */
 499        for (i = 0; i < 256; i++) {
 500                uint32_t n = ntohl(index[i]);
 501                if (n < nr) {
 502                        munmap(idx_map, idx_size);
 503                        return error("non-monotonic index %s", path);
 504                }
 505                nr = n;
 506        }
 507
 508        if (version == 1) {
 509                /*
 510                 * Total size:
 511                 *  - 256 index entries 4 bytes each
 512                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 513                 *  - 20-byte SHA1 of the packfile
 514                 *  - 20-byte SHA1 file checksum
 515                 */
 516                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 517                        munmap(idx_map, idx_size);
 518                        return error("wrong index v1 file size in %s", path);
 519                }
 520        } else if (version == 2) {
 521                /*
 522                 * Minimum size:
 523                 *  - 8 bytes of header
 524                 *  - 256 index entries 4 bytes each
 525                 *  - 20-byte sha1 entry * nr
 526                 *  - 4-byte crc entry * nr
 527                 *  - 4-byte offset entry * nr
 528                 *  - 20-byte SHA1 of the packfile
 529                 *  - 20-byte SHA1 file checksum
 530                 * And after the 4-byte offset table might be a
 531                 * variable sized table containing 8-byte entries
 532                 * for offsets larger than 2^31.
 533                 */
 534                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 535                unsigned long max_size = min_size;
 536                if (nr)
 537                        max_size += (nr - 1)*8;
 538                if (idx_size < min_size || idx_size > max_size) {
 539                        munmap(idx_map, idx_size);
 540                        return error("wrong index v2 file size in %s", path);
 541                }
 542                if (idx_size != min_size &&
 543                    /*
 544                     * make sure we can deal with large pack offsets.
 545                     * 31-bit signed offset won't be enough, neither
 546                     * 32-bit unsigned one will be.
 547                     */
 548                    (sizeof(off_t) <= 4)) {
 549                        munmap(idx_map, idx_size);
 550                        return error("pack too large for current definition of off_t in %s", path);
 551                }
 552        }
 553
 554        p->index_version = version;
 555        p->index_data = idx_map;
 556        p->index_size = idx_size;
 557        p->num_objects = nr;
 558        return 0;
 559}
 560
 561int open_pack_index(struct packed_git *p)
 562{
 563        char *idx_name;
 564        int ret;
 565
 566        if (p->index_data)
 567                return 0;
 568
 569        idx_name = xstrdup(p->pack_name);
 570        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 571        ret = check_packed_git_idx(idx_name, p);
 572        free(idx_name);
 573        return ret;
 574}
 575
 576static void scan_windows(struct packed_git *p,
 577        struct packed_git **lru_p,
 578        struct pack_window **lru_w,
 579        struct pack_window **lru_l)
 580{
 581        struct pack_window *w, *w_l;
 582
 583        for (w_l = NULL, w = p->windows; w; w = w->next) {
 584                if (!w->inuse_cnt) {
 585                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 586                                *lru_p = p;
 587                                *lru_w = w;
 588                                *lru_l = w_l;
 589                        }
 590                }
 591                w_l = w;
 592        }
 593}
 594
 595static int unuse_one_window(struct packed_git *current, int keep_fd)
 596{
 597        struct packed_git *p, *lru_p = NULL;
 598        struct pack_window *lru_w = NULL, *lru_l = NULL;
 599
 600        if (current)
 601                scan_windows(current, &lru_p, &lru_w, &lru_l);
 602        for (p = packed_git; p; p = p->next)
 603                scan_windows(p, &lru_p, &lru_w, &lru_l);
 604        if (lru_p) {
 605                munmap(lru_w->base, lru_w->len);
 606                pack_mapped -= lru_w->len;
 607                if (lru_l)
 608                        lru_l->next = lru_w->next;
 609                else {
 610                        lru_p->windows = lru_w->next;
 611                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 612                                close(lru_p->pack_fd);
 613                                lru_p->pack_fd = -1;
 614                        }
 615                }
 616                free(lru_w);
 617                pack_open_windows--;
 618                return 1;
 619        }
 620        return 0;
 621}
 622
 623void release_pack_memory(size_t need, int fd)
 624{
 625        size_t cur = pack_mapped;
 626        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 627                ; /* nothing */
 628}
 629
 630void close_pack_windows(struct packed_git *p)
 631{
 632        while (p->windows) {
 633                struct pack_window *w = p->windows;
 634
 635                if (w->inuse_cnt)
 636                        die("pack '%s' still has open windows to it",
 637                            p->pack_name);
 638                munmap(w->base, w->len);
 639                pack_mapped -= w->len;
 640                pack_open_windows--;
 641                p->windows = w->next;
 642                free(w);
 643        }
 644}
 645
 646void unuse_pack(struct pack_window **w_cursor)
 647{
 648        struct pack_window *w = *w_cursor;
 649        if (w) {
 650                w->inuse_cnt--;
 651                *w_cursor = NULL;
 652        }
 653}
 654
 655/*
 656 * Do not call this directly as this leaks p->pack_fd on error return;
 657 * call open_packed_git() instead.
 658 */
 659static int open_packed_git_1(struct packed_git *p)
 660{
 661        struct stat st;
 662        struct pack_header hdr;
 663        unsigned char sha1[20];
 664        unsigned char *idx_sha1;
 665        long fd_flag;
 666
 667        if (!p->index_data && open_pack_index(p))
 668                return error("packfile %s index unavailable", p->pack_name);
 669
 670        p->pack_fd = open(p->pack_name, O_RDONLY);
 671        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 672                return -1;
 673
 674        /* If we created the struct before we had the pack we lack size. */
 675        if (!p->pack_size) {
 676                if (!S_ISREG(st.st_mode))
 677                        return error("packfile %s not a regular file", p->pack_name);
 678                p->pack_size = st.st_size;
 679        } else if (p->pack_size != st.st_size)
 680                return error("packfile %s size changed", p->pack_name);
 681
 682        /* We leave these file descriptors open with sliding mmap;
 683         * there is no point keeping them open across exec(), though.
 684         */
 685        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 686        if (fd_flag < 0)
 687                return error("cannot determine file descriptor flags");
 688        fd_flag |= FD_CLOEXEC;
 689        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 690                return error("cannot set FD_CLOEXEC");
 691
 692        /* Verify we recognize this pack file format. */
 693        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 694                return error("file %s is far too short to be a packfile", p->pack_name);
 695        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 696                return error("file %s is not a GIT packfile", p->pack_name);
 697        if (!pack_version_ok(hdr.hdr_version))
 698                return error("packfile %s is version %"PRIu32" and not"
 699                        " supported (try upgrading GIT to a newer version)",
 700                        p->pack_name, ntohl(hdr.hdr_version));
 701
 702        /* Verify the pack matches its index. */
 703        if (p->num_objects != ntohl(hdr.hdr_entries))
 704                return error("packfile %s claims to have %"PRIu32" objects"
 705                             " while index indicates %"PRIu32" objects",
 706                             p->pack_name, ntohl(hdr.hdr_entries),
 707                             p->num_objects);
 708        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 709                return error("end of packfile %s is unavailable", p->pack_name);
 710        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 711                return error("packfile %s signature is unavailable", p->pack_name);
 712        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 713        if (hashcmp(sha1, idx_sha1))
 714                return error("packfile %s does not match index", p->pack_name);
 715        return 0;
 716}
 717
 718static int open_packed_git(struct packed_git *p)
 719{
 720        if (!open_packed_git_1(p))
 721                return 0;
 722        if (p->pack_fd != -1) {
 723                close(p->pack_fd);
 724                p->pack_fd = -1;
 725        }
 726        return -1;
 727}
 728
 729static int in_window(struct pack_window *win, off_t offset)
 730{
 731        /* We must promise at least 20 bytes (one hash) after the
 732         * offset is available from this window, otherwise the offset
 733         * is not actually in this window and a different window (which
 734         * has that one hash excess) must be used.  This is to support
 735         * the object header and delta base parsing routines below.
 736         */
 737        off_t win_off = win->offset;
 738        return win_off <= offset
 739                && (offset + 20) <= (win_off + win->len);
 740}
 741
 742unsigned char* use_pack(struct packed_git *p,
 743                struct pack_window **w_cursor,
 744                off_t offset,
 745                unsigned int *left)
 746{
 747        struct pack_window *win = *w_cursor;
 748
 749        if (p->pack_fd == -1 && open_packed_git(p))
 750                die("packfile %s cannot be accessed", p->pack_name);
 751
 752        /* Since packfiles end in a hash of their content and its
 753         * pointless to ask for an offset into the middle of that
 754         * hash, and the in_window function above wouldn't match
 755         * don't allow an offset too close to the end of the file.
 756         */
 757        if (offset > (p->pack_size - 20))
 758                die("offset beyond end of packfile (truncated pack?)");
 759
 760        if (!win || !in_window(win, offset)) {
 761                if (win)
 762                        win->inuse_cnt--;
 763                for (win = p->windows; win; win = win->next) {
 764                        if (in_window(win, offset))
 765                                break;
 766                }
 767                if (!win) {
 768                        size_t window_align = packed_git_window_size / 2;
 769                        off_t len;
 770                        win = xcalloc(1, sizeof(*win));
 771                        win->offset = (offset / window_align) * window_align;
 772                        len = p->pack_size - win->offset;
 773                        if (len > packed_git_window_size)
 774                                len = packed_git_window_size;
 775                        win->len = (size_t)len;
 776                        pack_mapped += win->len;
 777                        while (packed_git_limit < pack_mapped
 778                                && unuse_one_window(p, p->pack_fd))
 779                                ; /* nothing */
 780                        win->base = xmmap(NULL, win->len,
 781                                PROT_READ, MAP_PRIVATE,
 782                                p->pack_fd, win->offset);
 783                        if (win->base == MAP_FAILED)
 784                                die("packfile %s cannot be mapped: %s",
 785                                        p->pack_name,
 786                                        strerror(errno));
 787                        pack_mmap_calls++;
 788                        pack_open_windows++;
 789                        if (pack_mapped > peak_pack_mapped)
 790                                peak_pack_mapped = pack_mapped;
 791                        if (pack_open_windows > peak_pack_open_windows)
 792                                peak_pack_open_windows = pack_open_windows;
 793                        win->next = p->windows;
 794                        p->windows = win;
 795                }
 796        }
 797        if (win != *w_cursor) {
 798                win->last_used = pack_used_ctr++;
 799                win->inuse_cnt++;
 800                *w_cursor = win;
 801        }
 802        offset -= win->offset;
 803        if (left)
 804                *left = win->len - xsize_t(offset);
 805        return win->base + offset;
 806}
 807
 808static struct packed_git *alloc_packed_git(int extra)
 809{
 810        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 811        memset(p, 0, sizeof(*p));
 812        p->pack_fd = -1;
 813        return p;
 814}
 815
 816struct packed_git *add_packed_git(const char *path, int path_len, int local)
 817{
 818        struct stat st;
 819        struct packed_git *p = alloc_packed_git(path_len + 2);
 820
 821        /*
 822         * Make sure a corresponding .pack file exists and that
 823         * the index looks sane.
 824         */
 825        path_len -= strlen(".idx");
 826        if (path_len < 1) {
 827                free(p);
 828                return NULL;
 829        }
 830        memcpy(p->pack_name, path, path_len);
 831        strcpy(p->pack_name + path_len, ".pack");
 832        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 833                free(p);
 834                return NULL;
 835        }
 836
 837        /* ok, it looks sane as far as we can check without
 838         * actually mapping the pack file.
 839         */
 840        p->pack_size = st.st_size;
 841        p->pack_local = local;
 842        p->mtime = st.st_mtime;
 843        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 844                hashclr(p->sha1);
 845        return p;
 846}
 847
 848struct packed_git *parse_pack_index(unsigned char *sha1)
 849{
 850        const char *idx_path = sha1_pack_index_name(sha1);
 851        const char *path = sha1_pack_name(sha1);
 852        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 853
 854        strcpy(p->pack_name, path);
 855        hashcpy(p->sha1, sha1);
 856        if (check_packed_git_idx(idx_path, p)) {
 857                free(p);
 858                return NULL;
 859        }
 860
 861        return p;
 862}
 863
 864void install_packed_git(struct packed_git *pack)
 865{
 866        pack->next = packed_git;
 867        packed_git = pack;
 868}
 869
 870static void prepare_packed_git_one(char *objdir, int local)
 871{
 872        /* Ensure that this buffer is large enough so that we can
 873           append "/pack/" without clobbering the stack even if
 874           strlen(objdir) were PATH_MAX.  */
 875        char path[PATH_MAX + 1 + 4 + 1 + 1];
 876        int len;
 877        DIR *dir;
 878        struct dirent *de;
 879
 880        sprintf(path, "%s/pack", objdir);
 881        len = strlen(path);
 882        dir = opendir(path);
 883        if (!dir) {
 884                if (errno != ENOENT)
 885                        error("unable to open object pack directory: %s: %s",
 886                              path, strerror(errno));
 887                return;
 888        }
 889        path[len++] = '/';
 890        while ((de = readdir(dir)) != NULL) {
 891                int namelen = strlen(de->d_name);
 892                struct packed_git *p;
 893
 894                if (!has_extension(de->d_name, ".idx"))
 895                        continue;
 896
 897                if (len + namelen + 1 > sizeof(path))
 898                        continue;
 899
 900                /* Don't reopen a pack we already have. */
 901                strcpy(path + len, de->d_name);
 902                for (p = packed_git; p; p = p->next) {
 903                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 904                                break;
 905                }
 906                if (p)
 907                        continue;
 908                /* See if it really is a valid .idx file with corresponding
 909                 * .pack file that we can map.
 910                 */
 911                p = add_packed_git(path, len + namelen, local);
 912                if (!p)
 913                        continue;
 914                install_packed_git(p);
 915        }
 916        closedir(dir);
 917}
 918
 919static int sort_pack(const void *a_, const void *b_)
 920{
 921        struct packed_git *a = *((struct packed_git **)a_);
 922        struct packed_git *b = *((struct packed_git **)b_);
 923        int st;
 924
 925        /*
 926         * Local packs tend to contain objects specific to our
 927         * variant of the project than remote ones.  In addition,
 928         * remote ones could be on a network mounted filesystem.
 929         * Favor local ones for these reasons.
 930         */
 931        st = a->pack_local - b->pack_local;
 932        if (st)
 933                return -st;
 934
 935        /*
 936         * Younger packs tend to contain more recent objects,
 937         * and more recent objects tend to get accessed more
 938         * often.
 939         */
 940        if (a->mtime < b->mtime)
 941                return 1;
 942        else if (a->mtime == b->mtime)
 943                return 0;
 944        return -1;
 945}
 946
 947static void rearrange_packed_git(void)
 948{
 949        struct packed_git **ary, *p;
 950        int i, n;
 951
 952        for (n = 0, p = packed_git; p; p = p->next)
 953                n++;
 954        if (n < 2)
 955                return;
 956
 957        /* prepare an array of packed_git for easier sorting */
 958        ary = xcalloc(n, sizeof(struct packed_git *));
 959        for (n = 0, p = packed_git; p; p = p->next)
 960                ary[n++] = p;
 961
 962        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 963
 964        /* link them back again */
 965        for (i = 0; i < n - 1; i++)
 966                ary[i]->next = ary[i + 1];
 967        ary[n - 1]->next = NULL;
 968        packed_git = ary[0];
 969
 970        free(ary);
 971}
 972
 973static int prepare_packed_git_run_once = 0;
 974void prepare_packed_git(void)
 975{
 976        struct alternate_object_database *alt;
 977
 978        if (prepare_packed_git_run_once)
 979                return;
 980        prepare_packed_git_one(get_object_directory(), 1);
 981        prepare_alt_odb();
 982        for (alt = alt_odb_list; alt; alt = alt->next) {
 983                alt->name[-1] = 0;
 984                prepare_packed_git_one(alt->base, 0);
 985                alt->name[-1] = '/';
 986        }
 987        rearrange_packed_git();
 988        prepare_packed_git_run_once = 1;
 989}
 990
 991void reprepare_packed_git(void)
 992{
 993        discard_revindex();
 994        prepare_packed_git_run_once = 0;
 995        prepare_packed_git();
 996}
 997
 998static void mark_bad_packed_object(struct packed_git *p,
 999                                   const unsigned char *sha1)
1000{
1001        unsigned i;
1002        for (i = 0; i < p->num_bad_objects; i++)
1003                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1004                        return;
1005        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1006        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1007        p->num_bad_objects++;
1008}
1009
1010static int has_packed_and_bad(const unsigned char *sha1)
1011{
1012        struct packed_git *p;
1013        unsigned i;
1014
1015        for (p = packed_git; p; p = p->next)
1016                for (i = 0; i < p->num_bad_objects; i++)
1017                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1018                                return 1;
1019        return 0;
1020}
1021
1022int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1023{
1024        unsigned char real_sha1[20];
1025        hash_sha1_file(map, size, type, real_sha1);
1026        return hashcmp(sha1, real_sha1) ? -1 : 0;
1027}
1028
1029static int git_open_noatime(const char *name)
1030{
1031        static int sha1_file_open_flag = O_NOATIME;
1032        int fd = open(name, O_RDONLY | sha1_file_open_flag);
1033
1034        /* Might the failure be due to O_NOATIME? */
1035        if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1036                fd = open(name, O_RDONLY);
1037                if (fd >= 0)
1038                        sha1_file_open_flag = 0;
1039        }
1040        return fd;
1041}
1042
1043static int open_sha1_file(const unsigned char *sha1)
1044{
1045        int fd;
1046        char *name = sha1_file_name(sha1);
1047        struct alternate_object_database *alt;
1048
1049        fd = git_open_noatime(name);
1050        if (fd >= 0)
1051                return fd;
1052
1053        prepare_alt_odb();
1054        errno = ENOENT;
1055        for (alt = alt_odb_list; alt; alt = alt->next) {
1056                name = alt->name;
1057                fill_sha1_path(name, sha1);
1058                fd = git_open_noatime(alt->base);
1059                if (fd >= 0)
1060                        return fd;
1061        }
1062        return -1;
1063}
1064
1065static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1066{
1067        void *map;
1068        int fd;
1069
1070        fd = open_sha1_file(sha1);
1071        map = NULL;
1072        if (fd >= 0) {
1073                struct stat st;
1074
1075                if (!fstat(fd, &st)) {
1076                        *size = xsize_t(st.st_size);
1077                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1078                }
1079                close(fd);
1080        }
1081        return map;
1082}
1083
1084static int legacy_loose_object(unsigned char *map)
1085{
1086        unsigned int word;
1087
1088        /*
1089         * Is it a zlib-compressed buffer? If so, the first byte
1090         * must be 0x78 (15-bit window size, deflated), and the
1091         * first 16-bit word is evenly divisible by 31
1092         */
1093        word = (map[0] << 8) + map[1];
1094        if (map[0] == 0x78 && !(word % 31))
1095                return 1;
1096        else
1097                return 0;
1098}
1099
1100unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1101{
1102        unsigned shift;
1103        unsigned char c;
1104        unsigned long size;
1105        unsigned long used = 0;
1106
1107        c = buf[used++];
1108        *type = (c >> 4) & 7;
1109        size = c & 15;
1110        shift = 4;
1111        while (c & 0x80) {
1112                if (len <= used)
1113                        return 0;
1114                if (sizeof(long) * 8 <= shift)
1115                        return 0;
1116                c = buf[used++];
1117                size += (c & 0x7f) << shift;
1118                shift += 7;
1119        }
1120        *sizep = size;
1121        return used;
1122}
1123
1124static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1125{
1126        unsigned long size, used;
1127        static const char valid_loose_object_type[8] = {
1128                0, /* OBJ_EXT */
1129                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1130                0, /* "delta" and others are invalid in a loose object */
1131        };
1132        enum object_type type;
1133
1134        /* Get the data stream */
1135        memset(stream, 0, sizeof(*stream));
1136        stream->next_in = map;
1137        stream->avail_in = mapsize;
1138        stream->next_out = buffer;
1139        stream->avail_out = bufsiz;
1140
1141        if (legacy_loose_object(map)) {
1142                inflateInit(stream);
1143                return inflate(stream, 0);
1144        }
1145
1146
1147        /*
1148         * There used to be a second loose object header format which
1149         * was meant to mimic the in-pack format, allowing for direct
1150         * copy of the object data.  This format turned up not to be
1151         * really worth it and we don't write it any longer.  But we
1152         * can still read it.
1153         */
1154        used = unpack_object_header_gently(map, mapsize, &type, &size);
1155        if (!used || !valid_loose_object_type[type])
1156                return -1;
1157        map += used;
1158        mapsize -= used;
1159
1160        /* Set up the stream for the rest.. */
1161        stream->next_in = map;
1162        stream->avail_in = mapsize;
1163        inflateInit(stream);
1164
1165        /* And generate the fake traditional header */
1166        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1167                                         typename(type), size);
1168        return 0;
1169}
1170
1171static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1172{
1173        int bytes = strlen(buffer) + 1;
1174        unsigned char *buf = xmalloc(1+size);
1175        unsigned long n;
1176        int status = Z_OK;
1177
1178        n = stream->total_out - bytes;
1179        if (n > size)
1180                n = size;
1181        memcpy(buf, (char *) buffer + bytes, n);
1182        bytes = n;
1183        if (bytes <= size) {
1184                /*
1185                 * The above condition must be (bytes <= size), not
1186                 * (bytes < size).  In other words, even though we
1187                 * expect no more output and set avail_out to zer0,
1188                 * the input zlib stream may have bytes that express
1189                 * "this concludes the stream", and we *do* want to
1190                 * eat that input.
1191                 *
1192                 * Otherwise we would not be able to test that we
1193                 * consumed all the input to reach the expected size;
1194                 * we also want to check that zlib tells us that all
1195                 * went well with status == Z_STREAM_END at the end.
1196                 */
1197                stream->next_out = buf + bytes;
1198                stream->avail_out = size - bytes;
1199                while (status == Z_OK)
1200                        status = inflate(stream, Z_FINISH);
1201        }
1202        buf[size] = 0;
1203        if (status == Z_STREAM_END && !stream->avail_in) {
1204                inflateEnd(stream);
1205                return buf;
1206        }
1207
1208        if (status < 0)
1209                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1210        else if (stream->avail_in)
1211                error("garbage at end of loose object '%s'",
1212                      sha1_to_hex(sha1));
1213        free(buf);
1214        return NULL;
1215}
1216
1217/*
1218 * We used to just use "sscanf()", but that's actually way
1219 * too permissive for what we want to check. So do an anal
1220 * object header parse by hand.
1221 */
1222static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1223{
1224        char type[10];
1225        int i;
1226        unsigned long size;
1227
1228        /*
1229         * The type can be at most ten bytes (including the
1230         * terminating '\0' that we add), and is followed by
1231         * a space.
1232         */
1233        i = 0;
1234        for (;;) {
1235                char c = *hdr++;
1236                if (c == ' ')
1237                        break;
1238                type[i++] = c;
1239                if (i >= sizeof(type))
1240                        return -1;
1241        }
1242        type[i] = 0;
1243
1244        /*
1245         * The length must follow immediately, and be in canonical
1246         * decimal format (ie "010" is not valid).
1247         */
1248        size = *hdr++ - '0';
1249        if (size > 9)
1250                return -1;
1251        if (size) {
1252                for (;;) {
1253                        unsigned long c = *hdr - '0';
1254                        if (c > 9)
1255                                break;
1256                        hdr++;
1257                        size = size * 10 + c;
1258                }
1259        }
1260        *sizep = size;
1261
1262        /*
1263         * The length must be followed by a zero byte
1264         */
1265        return *hdr ? -1 : type_from_string(type);
1266}
1267
1268static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1269{
1270        int ret;
1271        z_stream stream;
1272        char hdr[8192];
1273
1274        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1275        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1276                return NULL;
1277
1278        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1279}
1280
1281unsigned long get_size_from_delta(struct packed_git *p,
1282                                  struct pack_window **w_curs,
1283                                  off_t curpos)
1284{
1285        const unsigned char *data;
1286        unsigned char delta_head[20], *in;
1287        z_stream stream;
1288        int st;
1289
1290        memset(&stream, 0, sizeof(stream));
1291        stream.next_out = delta_head;
1292        stream.avail_out = sizeof(delta_head);
1293
1294        inflateInit(&stream);
1295        do {
1296                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1297                stream.next_in = in;
1298                st = inflate(&stream, Z_FINISH);
1299                curpos += stream.next_in - in;
1300        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1301                 stream.total_out < sizeof(delta_head));
1302        inflateEnd(&stream);
1303        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1304                die("delta data unpack-initial failed");
1305
1306        /* Examine the initial part of the delta to figure out
1307         * the result size.
1308         */
1309        data = delta_head;
1310
1311        /* ignore base size */
1312        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1313
1314        /* Read the result size */
1315        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1316}
1317
1318static off_t get_delta_base(struct packed_git *p,
1319                                    struct pack_window **w_curs,
1320                                    off_t *curpos,
1321                                    enum object_type type,
1322                                    off_t delta_obj_offset)
1323{
1324        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1325        off_t base_offset;
1326
1327        /* use_pack() assured us we have [base_info, base_info + 20)
1328         * as a range that we can look at without walking off the
1329         * end of the mapped window.  Its actually the hash size
1330         * that is assured.  An OFS_DELTA longer than the hash size
1331         * is stupid, as then a REF_DELTA would be smaller to store.
1332         */
1333        if (type == OBJ_OFS_DELTA) {
1334                unsigned used = 0;
1335                unsigned char c = base_info[used++];
1336                base_offset = c & 127;
1337                while (c & 128) {
1338                        base_offset += 1;
1339                        if (!base_offset || MSB(base_offset, 7))
1340                                return 0;  /* overflow */
1341                        c = base_info[used++];
1342                        base_offset = (base_offset << 7) + (c & 127);
1343                }
1344                base_offset = delta_obj_offset - base_offset;
1345                if (base_offset >= delta_obj_offset)
1346                        return 0;  /* out of bound */
1347                *curpos += used;
1348        } else if (type == OBJ_REF_DELTA) {
1349                /* The base entry _must_ be in the same pack */
1350                base_offset = find_pack_entry_one(base_info, p);
1351                *curpos += 20;
1352        } else
1353                die("I am totally screwed");
1354        return base_offset;
1355}
1356
1357/* forward declaration for a mutually recursive function */
1358static int packed_object_info(struct packed_git *p, off_t offset,
1359                              unsigned long *sizep);
1360
1361static int packed_delta_info(struct packed_git *p,
1362                             struct pack_window **w_curs,
1363                             off_t curpos,
1364                             enum object_type type,
1365                             off_t obj_offset,
1366                             unsigned long *sizep)
1367{
1368        off_t base_offset;
1369
1370        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1371        type = packed_object_info(p, base_offset, NULL);
1372
1373        /* We choose to only get the type of the base object and
1374         * ignore potentially corrupt pack file that expects the delta
1375         * based on a base with a wrong size.  This saves tons of
1376         * inflate() calls.
1377         */
1378        if (sizep)
1379                *sizep = get_size_from_delta(p, w_curs, curpos);
1380
1381        return type;
1382}
1383
1384static int unpack_object_header(struct packed_git *p,
1385                                struct pack_window **w_curs,
1386                                off_t *curpos,
1387                                unsigned long *sizep)
1388{
1389        unsigned char *base;
1390        unsigned int left;
1391        unsigned long used;
1392        enum object_type type;
1393
1394        /* use_pack() assures us we have [base, base + 20) available
1395         * as a range that we can look at at.  (Its actually the hash
1396         * size that is assured.)  With our object header encoding
1397         * the maximum deflated object size is 2^137, which is just
1398         * insane, so we know won't exceed what we have been given.
1399         */
1400        base = use_pack(p, w_curs, *curpos, &left);
1401        used = unpack_object_header_gently(base, left, &type, sizep);
1402        if (!used)
1403                die("object offset outside of pack file");
1404        *curpos += used;
1405
1406        return type;
1407}
1408
1409const char *packed_object_info_detail(struct packed_git *p,
1410                                      off_t obj_offset,
1411                                      unsigned long *size,
1412                                      unsigned long *store_size,
1413                                      unsigned int *delta_chain_length,
1414                                      unsigned char *base_sha1)
1415{
1416        struct pack_window *w_curs = NULL;
1417        off_t curpos;
1418        unsigned long dummy;
1419        unsigned char *next_sha1;
1420        enum object_type type;
1421        struct revindex_entry *revidx;
1422
1423        *delta_chain_length = 0;
1424        curpos = obj_offset;
1425        type = unpack_object_header(p, &w_curs, &curpos, size);
1426
1427        revidx = find_pack_revindex(p, obj_offset);
1428        *store_size = revidx[1].offset - obj_offset;
1429
1430        for (;;) {
1431                switch (type) {
1432                default:
1433                        die("pack %s contains unknown object type %d",
1434                            p->pack_name, type);
1435                case OBJ_COMMIT:
1436                case OBJ_TREE:
1437                case OBJ_BLOB:
1438                case OBJ_TAG:
1439                        unuse_pack(&w_curs);
1440                        return typename(type);
1441                case OBJ_OFS_DELTA:
1442                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1443                        if (!obj_offset)
1444                                die("pack %s contains bad delta base reference of type %s",
1445                                    p->pack_name, typename(type));
1446                        if (*delta_chain_length == 0) {
1447                                revidx = find_pack_revindex(p, obj_offset);
1448                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1449                        }
1450                        break;
1451                case OBJ_REF_DELTA:
1452                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1453                        if (*delta_chain_length == 0)
1454                                hashcpy(base_sha1, next_sha1);
1455                        obj_offset = find_pack_entry_one(next_sha1, p);
1456                        break;
1457                }
1458                (*delta_chain_length)++;
1459                curpos = obj_offset;
1460                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1461        }
1462}
1463
1464static int packed_object_info(struct packed_git *p, off_t obj_offset,
1465                              unsigned long *sizep)
1466{
1467        struct pack_window *w_curs = NULL;
1468        unsigned long size;
1469        off_t curpos = obj_offset;
1470        enum object_type type;
1471
1472        type = unpack_object_header(p, &w_curs, &curpos, &size);
1473
1474        switch (type) {
1475        case OBJ_OFS_DELTA:
1476        case OBJ_REF_DELTA:
1477                type = packed_delta_info(p, &w_curs, curpos,
1478                                         type, obj_offset, sizep);
1479                break;
1480        case OBJ_COMMIT:
1481        case OBJ_TREE:
1482        case OBJ_BLOB:
1483        case OBJ_TAG:
1484                if (sizep)
1485                        *sizep = size;
1486                break;
1487        default:
1488                die("pack %s contains unknown object type %d",
1489                    p->pack_name, type);
1490        }
1491        unuse_pack(&w_curs);
1492        return type;
1493}
1494
1495static void *unpack_compressed_entry(struct packed_git *p,
1496                                    struct pack_window **w_curs,
1497                                    off_t curpos,
1498                                    unsigned long size)
1499{
1500        int st;
1501        z_stream stream;
1502        unsigned char *buffer, *in;
1503
1504        buffer = xmalloc(size + 1);
1505        buffer[size] = 0;
1506        memset(&stream, 0, sizeof(stream));
1507        stream.next_out = buffer;
1508        stream.avail_out = size;
1509
1510        inflateInit(&stream);
1511        do {
1512                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1513                stream.next_in = in;
1514                st = inflate(&stream, Z_FINISH);
1515                curpos += stream.next_in - in;
1516        } while (st == Z_OK || st == Z_BUF_ERROR);
1517        inflateEnd(&stream);
1518        if ((st != Z_STREAM_END) || stream.total_out != size) {
1519                free(buffer);
1520                return NULL;
1521        }
1522
1523        return buffer;
1524}
1525
1526#define MAX_DELTA_CACHE (256)
1527
1528static size_t delta_base_cached;
1529
1530static struct delta_base_cache_lru_list {
1531        struct delta_base_cache_lru_list *prev;
1532        struct delta_base_cache_lru_list *next;
1533} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1534
1535static struct delta_base_cache_entry {
1536        struct delta_base_cache_lru_list lru;
1537        void *data;
1538        struct packed_git *p;
1539        off_t base_offset;
1540        unsigned long size;
1541        enum object_type type;
1542} delta_base_cache[MAX_DELTA_CACHE];
1543
1544static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1545{
1546        unsigned long hash;
1547
1548        hash = (unsigned long)p + (unsigned long)base_offset;
1549        hash += (hash >> 8) + (hash >> 16);
1550        return hash % MAX_DELTA_CACHE;
1551}
1552
1553static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1554        unsigned long *base_size, enum object_type *type, int keep_cache)
1555{
1556        void *ret;
1557        unsigned long hash = pack_entry_hash(p, base_offset);
1558        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1559
1560        ret = ent->data;
1561        if (ret && ent->p == p && ent->base_offset == base_offset)
1562                goto found_cache_entry;
1563        return unpack_entry(p, base_offset, type, base_size);
1564
1565found_cache_entry:
1566        if (!keep_cache) {
1567                ent->data = NULL;
1568                ent->lru.next->prev = ent->lru.prev;
1569                ent->lru.prev->next = ent->lru.next;
1570                delta_base_cached -= ent->size;
1571        } else {
1572                ret = xmemdupz(ent->data, ent->size);
1573        }
1574        *type = ent->type;
1575        *base_size = ent->size;
1576        return ret;
1577}
1578
1579static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1580{
1581        if (ent->data) {
1582                free(ent->data);
1583                ent->data = NULL;
1584                ent->lru.next->prev = ent->lru.prev;
1585                ent->lru.prev->next = ent->lru.next;
1586                delta_base_cached -= ent->size;
1587        }
1588}
1589
1590static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1591        void *base, unsigned long base_size, enum object_type type)
1592{
1593        unsigned long hash = pack_entry_hash(p, base_offset);
1594        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1595        struct delta_base_cache_lru_list *lru;
1596
1597        release_delta_base_cache(ent);
1598        delta_base_cached += base_size;
1599
1600        for (lru = delta_base_cache_lru.next;
1601             delta_base_cached > delta_base_cache_limit
1602             && lru != &delta_base_cache_lru;
1603             lru = lru->next) {
1604                struct delta_base_cache_entry *f = (void *)lru;
1605                if (f->type == OBJ_BLOB)
1606                        release_delta_base_cache(f);
1607        }
1608        for (lru = delta_base_cache_lru.next;
1609             delta_base_cached > delta_base_cache_limit
1610             && lru != &delta_base_cache_lru;
1611             lru = lru->next) {
1612                struct delta_base_cache_entry *f = (void *)lru;
1613                release_delta_base_cache(f);
1614        }
1615
1616        ent->p = p;
1617        ent->base_offset = base_offset;
1618        ent->type = type;
1619        ent->data = base;
1620        ent->size = base_size;
1621        ent->lru.next = &delta_base_cache_lru;
1622        ent->lru.prev = delta_base_cache_lru.prev;
1623        delta_base_cache_lru.prev->next = &ent->lru;
1624        delta_base_cache_lru.prev = &ent->lru;
1625}
1626
1627static void *unpack_delta_entry(struct packed_git *p,
1628                                struct pack_window **w_curs,
1629                                off_t curpos,
1630                                unsigned long delta_size,
1631                                off_t obj_offset,
1632                                enum object_type *type,
1633                                unsigned long *sizep)
1634{
1635        void *delta_data, *result, *base;
1636        unsigned long base_size;
1637        off_t base_offset;
1638
1639        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1640        if (!base_offset) {
1641                error("failed to validate delta base reference "
1642                      "at offset %"PRIuMAX" from %s",
1643                      (uintmax_t)curpos, p->pack_name);
1644                return NULL;
1645        }
1646        unuse_pack(w_curs);
1647        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1648        if (!base) {
1649                /*
1650                 * We're probably in deep shit, but let's try to fetch
1651                 * the required base anyway from another pack or loose.
1652                 * This is costly but should happen only in the presence
1653                 * of a corrupted pack, and is better than failing outright.
1654                 */
1655                struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1656                const unsigned char *base_sha1 =
1657                                        nth_packed_object_sha1(p, revidx->nr);
1658                error("failed to read delta base object %s"
1659                      " at offset %"PRIuMAX" from %s",
1660                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1661                      p->pack_name);
1662                mark_bad_packed_object(p, base_sha1);
1663                base = read_object(base_sha1, type, &base_size);
1664                if (!base)
1665                        return NULL;
1666        }
1667
1668        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1669        if (!delta_data) {
1670                error("failed to unpack compressed delta "
1671                      "at offset %"PRIuMAX" from %s",
1672                      (uintmax_t)curpos, p->pack_name);
1673                free(base);
1674                return NULL;
1675        }
1676        result = patch_delta(base, base_size,
1677                             delta_data, delta_size,
1678                             sizep);
1679        if (!result)
1680                die("failed to apply delta");
1681        free(delta_data);
1682        add_delta_base_cache(p, base_offset, base, base_size, *type);
1683        return result;
1684}
1685
1686void *unpack_entry(struct packed_git *p, off_t obj_offset,
1687                   enum object_type *type, unsigned long *sizep)
1688{
1689        struct pack_window *w_curs = NULL;
1690        off_t curpos = obj_offset;
1691        void *data;
1692
1693        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1694        switch (*type) {
1695        case OBJ_OFS_DELTA:
1696        case OBJ_REF_DELTA:
1697                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1698                                          obj_offset, type, sizep);
1699                break;
1700        case OBJ_COMMIT:
1701        case OBJ_TREE:
1702        case OBJ_BLOB:
1703        case OBJ_TAG:
1704                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1705                break;
1706        default:
1707                data = NULL;
1708                error("unknown object type %i at offset %"PRIuMAX" in %s",
1709                      *type, (uintmax_t)obj_offset, p->pack_name);
1710        }
1711        unuse_pack(&w_curs);
1712        return data;
1713}
1714
1715const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1716                                            uint32_t n)
1717{
1718        const unsigned char *index = p->index_data;
1719        if (!index) {
1720                if (open_pack_index(p))
1721                        return NULL;
1722                index = p->index_data;
1723        }
1724        if (n >= p->num_objects)
1725                return NULL;
1726        index += 4 * 256;
1727        if (p->index_version == 1) {
1728                return index + 24 * n + 4;
1729        } else {
1730                index += 8;
1731                return index + 20 * n;
1732        }
1733}
1734
1735off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1736{
1737        const unsigned char *index = p->index_data;
1738        index += 4 * 256;
1739        if (p->index_version == 1) {
1740                return ntohl(*((uint32_t *)(index + 24 * n)));
1741        } else {
1742                uint32_t off;
1743                index += 8 + p->num_objects * (20 + 4);
1744                off = ntohl(*((uint32_t *)(index + 4 * n)));
1745                if (!(off & 0x80000000))
1746                        return off;
1747                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1748                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1749                                   ntohl(*((uint32_t *)(index + 4)));
1750        }
1751}
1752
1753off_t find_pack_entry_one(const unsigned char *sha1,
1754                                  struct packed_git *p)
1755{
1756        const uint32_t *level1_ofs = p->index_data;
1757        const unsigned char *index = p->index_data;
1758        unsigned hi, lo, stride;
1759        static int use_lookup = -1;
1760        static int debug_lookup = -1;
1761
1762        if (debug_lookup < 0)
1763                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1764
1765        if (!index) {
1766                if (open_pack_index(p))
1767                        return 0;
1768                level1_ofs = p->index_data;
1769                index = p->index_data;
1770        }
1771        if (p->index_version > 1) {
1772                level1_ofs += 2;
1773                index += 8;
1774        }
1775        index += 4 * 256;
1776        hi = ntohl(level1_ofs[*sha1]);
1777        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1778        if (p->index_version > 1) {
1779                stride = 20;
1780        } else {
1781                stride = 24;
1782                index += 4;
1783        }
1784
1785        if (debug_lookup)
1786                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1787                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1788
1789        if (use_lookup < 0)
1790                use_lookup = !!getenv("GIT_USE_LOOKUP");
1791        if (use_lookup) {
1792                int pos = sha1_entry_pos(index, stride, 0,
1793                                         lo, hi, p->num_objects, sha1);
1794                if (pos < 0)
1795                        return 0;
1796                return nth_packed_object_offset(p, pos);
1797        }
1798
1799        do {
1800                unsigned mi = (lo + hi) / 2;
1801                int cmp = hashcmp(index + mi * stride, sha1);
1802
1803                if (debug_lookup)
1804                        printf("lo %u hi %u rg %u mi %u\n",
1805                               lo, hi, hi - lo, mi);
1806                if (!cmp)
1807                        return nth_packed_object_offset(p, mi);
1808                if (cmp > 0)
1809                        hi = mi;
1810                else
1811                        lo = mi+1;
1812        } while (lo < hi);
1813        return 0;
1814}
1815
1816int matches_pack_name(struct packed_git *p, const char *name)
1817{
1818        const char *last_c, *c;
1819
1820        if (!strcmp(p->pack_name, name))
1821                return 1;
1822
1823        for (c = p->pack_name, last_c = c; *c;)
1824                if (*c == '/')
1825                        last_c = ++c;
1826                else
1827                        ++c;
1828        if (!strcmp(last_c, name))
1829                return 1;
1830
1831        return 0;
1832}
1833
1834static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1835{
1836        static struct packed_git *last_found = (void *)1;
1837        struct packed_git *p;
1838        off_t offset;
1839
1840        prepare_packed_git();
1841        if (!packed_git)
1842                return 0;
1843        p = (last_found == (void *)1) ? packed_git : last_found;
1844
1845        do {
1846                if (ignore_packed) {
1847                        const char **ig;
1848                        for (ig = ignore_packed; *ig; ig++)
1849                                if (matches_pack_name(p, *ig))
1850                                        break;
1851                        if (*ig)
1852                                goto next;
1853                }
1854
1855                if (p->num_bad_objects) {
1856                        unsigned i;
1857                        for (i = 0; i < p->num_bad_objects; i++)
1858                                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1859                                        goto next;
1860                }
1861
1862                offset = find_pack_entry_one(sha1, p);
1863                if (offset) {
1864                        /*
1865                         * We are about to tell the caller where they can
1866                         * locate the requested object.  We better make
1867                         * sure the packfile is still here and can be
1868                         * accessed before supplying that answer, as
1869                         * it may have been deleted since the index
1870                         * was loaded!
1871                         */
1872                        if (p->pack_fd == -1 && open_packed_git(p)) {
1873                                error("packfile %s cannot be accessed", p->pack_name);
1874                                goto next;
1875                        }
1876                        e->offset = offset;
1877                        e->p = p;
1878                        hashcpy(e->sha1, sha1);
1879                        last_found = p;
1880                        return 1;
1881                }
1882
1883                next:
1884                if (p == last_found)
1885                        p = packed_git;
1886                else
1887                        p = p->next;
1888                if (p == last_found)
1889                        p = p->next;
1890        } while (p);
1891        return 0;
1892}
1893
1894struct packed_git *find_sha1_pack(const unsigned char *sha1,
1895                                  struct packed_git *packs)
1896{
1897        struct packed_git *p;
1898
1899        for (p = packs; p; p = p->next) {
1900                if (find_pack_entry_one(sha1, p))
1901                        return p;
1902        }
1903        return NULL;
1904
1905}
1906
1907static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1908{
1909        int status;
1910        unsigned long mapsize, size;
1911        void *map;
1912        z_stream stream;
1913        char hdr[32];
1914
1915        map = map_sha1_file(sha1, &mapsize);
1916        if (!map)
1917                return error("unable to find %s", sha1_to_hex(sha1));
1918        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1919                status = error("unable to unpack %s header",
1920                               sha1_to_hex(sha1));
1921        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1922                status = error("unable to parse %s header", sha1_to_hex(sha1));
1923        else if (sizep)
1924                *sizep = size;
1925        inflateEnd(&stream);
1926        munmap(map, mapsize);
1927        return status;
1928}
1929
1930int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1931{
1932        struct pack_entry e;
1933        int status;
1934
1935        if (!find_pack_entry(sha1, &e, NULL)) {
1936                /* Most likely it's a loose object. */
1937                status = sha1_loose_object_info(sha1, sizep);
1938                if (status >= 0)
1939                        return status;
1940
1941                /* Not a loose object; someone else may have just packed it. */
1942                reprepare_packed_git();
1943                if (!find_pack_entry(sha1, &e, NULL))
1944                        return status;
1945        }
1946        return packed_object_info(e.p, e.offset, sizep);
1947}
1948
1949static void *read_packed_sha1(const unsigned char *sha1,
1950                              enum object_type *type, unsigned long *size)
1951{
1952        struct pack_entry e;
1953        void *data;
1954
1955        if (!find_pack_entry(sha1, &e, NULL))
1956                return NULL;
1957        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1958        if (!data) {
1959                /*
1960                 * We're probably in deep shit, but let's try to fetch
1961                 * the required object anyway from another pack or loose.
1962                 * This should happen only in the presence of a corrupted
1963                 * pack, and is better than failing outright.
1964                 */
1965                error("failed to read object %s at offset %"PRIuMAX" from %s",
1966                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1967                mark_bad_packed_object(e.p, sha1);
1968                data = read_object(sha1, type, size);
1969        }
1970        return data;
1971}
1972
1973/*
1974 * This is meant to hold a *small* number of objects that you would
1975 * want read_sha1_file() to be able to return, but yet you do not want
1976 * to write them into the object store (e.g. a browse-only
1977 * application).
1978 */
1979static struct cached_object {
1980        unsigned char sha1[20];
1981        enum object_type type;
1982        void *buf;
1983        unsigned long size;
1984} *cached_objects;
1985static int cached_object_nr, cached_object_alloc;
1986
1987static struct cached_object empty_tree = {
1988        /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1989        "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1990        "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1991        OBJ_TREE,
1992        "",
1993        0
1994};
1995
1996static struct cached_object *find_cached_object(const unsigned char *sha1)
1997{
1998        int i;
1999        struct cached_object *co = cached_objects;
2000
2001        for (i = 0; i < cached_object_nr; i++, co++) {
2002                if (!hashcmp(co->sha1, sha1))
2003                        return co;
2004        }
2005        if (!hashcmp(sha1, empty_tree.sha1))
2006                return &empty_tree;
2007        return NULL;
2008}
2009
2010int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2011                      unsigned char *sha1)
2012{
2013        struct cached_object *co;
2014
2015        hash_sha1_file(buf, len, typename(type), sha1);
2016        if (has_sha1_file(sha1) || find_cached_object(sha1))
2017                return 0;
2018        if (cached_object_alloc <= cached_object_nr) {
2019                cached_object_alloc = alloc_nr(cached_object_alloc);
2020                cached_objects = xrealloc(cached_objects,
2021                                          sizeof(*cached_objects) *
2022                                          cached_object_alloc);
2023        }
2024        co = &cached_objects[cached_object_nr++];
2025        co->size = len;
2026        co->type = type;
2027        co->buf = xmalloc(len);
2028        memcpy(co->buf, buf, len);
2029        hashcpy(co->sha1, sha1);
2030        return 0;
2031}
2032
2033void *read_object(const unsigned char *sha1, enum object_type *type,
2034                  unsigned long *size)
2035{
2036        unsigned long mapsize;
2037        void *map, *buf;
2038        struct cached_object *co;
2039
2040        co = find_cached_object(sha1);
2041        if (co) {
2042                *type = co->type;
2043                *size = co->size;
2044                return xmemdupz(co->buf, co->size);
2045        }
2046
2047        buf = read_packed_sha1(sha1, type, size);
2048        if (buf)
2049                return buf;
2050        map = map_sha1_file(sha1, &mapsize);
2051        if (map) {
2052                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2053                munmap(map, mapsize);
2054                return buf;
2055        }
2056        reprepare_packed_git();
2057        return read_packed_sha1(sha1, type, size);
2058}
2059
2060void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2061                     unsigned long *size)
2062{
2063        void *data = read_object(sha1, type, size);
2064        /* legacy behavior is to die on corrupted objects */
2065        if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2066                die("object %s is corrupted", sha1_to_hex(sha1));
2067        return data;
2068}
2069
2070void *read_object_with_reference(const unsigned char *sha1,
2071                                 const char *required_type_name,
2072                                 unsigned long *size,
2073                                 unsigned char *actual_sha1_return)
2074{
2075        enum object_type type, required_type;
2076        void *buffer;
2077        unsigned long isize;
2078        unsigned char actual_sha1[20];
2079
2080        required_type = type_from_string(required_type_name);
2081        hashcpy(actual_sha1, sha1);
2082        while (1) {
2083                int ref_length = -1;
2084                const char *ref_type = NULL;
2085
2086                buffer = read_sha1_file(actual_sha1, &type, &isize);
2087                if (!buffer)
2088                        return NULL;
2089                if (type == required_type) {
2090                        *size = isize;
2091                        if (actual_sha1_return)
2092                                hashcpy(actual_sha1_return, actual_sha1);
2093                        return buffer;
2094                }
2095                /* Handle references */
2096                else if (type == OBJ_COMMIT)
2097                        ref_type = "tree ";
2098                else if (type == OBJ_TAG)
2099                        ref_type = "object ";
2100                else {
2101                        free(buffer);
2102                        return NULL;
2103                }
2104                ref_length = strlen(ref_type);
2105
2106                if (ref_length + 40 > isize ||
2107                    memcmp(buffer, ref_type, ref_length) ||
2108                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2109                        free(buffer);
2110                        return NULL;
2111                }
2112                free(buffer);
2113                /* Now we have the ID of the referred-to object in
2114                 * actual_sha1.  Check again. */
2115        }
2116}
2117
2118static void write_sha1_file_prepare(const void *buf, unsigned long len,
2119                                    const char *type, unsigned char *sha1,
2120                                    char *hdr, int *hdrlen)
2121{
2122        SHA_CTX c;
2123
2124        /* Generate the header */
2125        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2126
2127        /* Sha1.. */
2128        SHA1_Init(&c);
2129        SHA1_Update(&c, hdr, *hdrlen);
2130        SHA1_Update(&c, buf, len);
2131        SHA1_Final(sha1, &c);
2132}
2133
2134/*
2135 * Move the just written object into its final resting place
2136 */
2137int move_temp_to_file(const char *tmpfile, const char *filename)
2138{
2139        int ret = 0;
2140        if (link(tmpfile, filename))
2141                ret = errno;
2142
2143        /*
2144         * Coda hack - coda doesn't like cross-directory links,
2145         * so we fall back to a rename, which will mean that it
2146         * won't be able to check collisions, but that's not a
2147         * big deal.
2148         *
2149         * The same holds for FAT formatted media.
2150         *
2151         * When this succeeds, we just return 0. We have nothing
2152         * left to unlink.
2153         */
2154        if (ret && ret != EEXIST) {
2155                if (!rename(tmpfile, filename))
2156                        return 0;
2157                ret = errno;
2158        }
2159        unlink(tmpfile);
2160        if (ret) {
2161                if (ret != EEXIST) {
2162                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2163                }
2164                /* FIXME!!! Collision check here ? */
2165        }
2166
2167        return 0;
2168}
2169
2170static int write_buffer(int fd, const void *buf, size_t len)
2171{
2172        if (write_in_full(fd, buf, len) < 0)
2173                return error("file write error (%s)", strerror(errno));
2174        return 0;
2175}
2176
2177int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2178                   unsigned char *sha1)
2179{
2180        char hdr[32];
2181        int hdrlen;
2182        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2183        return 0;
2184}
2185
2186/* Finalize a file on disk, and close it. */
2187static void close_sha1_file(int fd)
2188{
2189        if (fsync_object_files)
2190                fsync_or_die(fd, "sha1 file");
2191        fchmod(fd, 0444);
2192        if (close(fd) != 0)
2193                die("unable to write sha1 file");
2194}
2195
2196/* Size of directory component, including the ending '/' */
2197static inline int directory_size(const char *filename)
2198{
2199        const char *s = strrchr(filename, '/');
2200        if (!s)
2201                return 0;
2202        return s - filename + 1;
2203}
2204
2205/*
2206 * This creates a temporary file in the same directory as the final
2207 * 'filename'
2208 *
2209 * We want to avoid cross-directory filename renames, because those
2210 * can have problems on various filesystems (FAT, NFS, Coda).
2211 */
2212static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2213{
2214        int fd, dirlen = directory_size(filename);
2215
2216        if (dirlen + 20 > bufsiz) {
2217                errno = ENAMETOOLONG;
2218                return -1;
2219        }
2220        memcpy(buffer, filename, dirlen);
2221        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2222        fd = mkstemp(buffer);
2223        if (fd < 0 && dirlen) {
2224                /* Make sure the directory exists */
2225                memcpy(buffer, filename, dirlen);
2226                buffer[dirlen-1] = 0;
2227                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2228                        return -1;
2229
2230                /* Try again */
2231                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2232                fd = mkstemp(buffer);
2233        }
2234        return fd;
2235}
2236
2237static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2238                              void *buf, unsigned long len, time_t mtime)
2239{
2240        int fd, size, ret;
2241        unsigned char *compressed;
2242        z_stream stream;
2243        char *filename;
2244        static char tmpfile[PATH_MAX];
2245
2246        filename = sha1_file_name(sha1);
2247        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2248        if (fd < 0) {
2249                if (errno == EPERM)
2250                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2251                else
2252                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2253        }
2254
2255        /* Set it up */
2256        memset(&stream, 0, sizeof(stream));
2257        deflateInit(&stream, zlib_compression_level);
2258        size = 8 + deflateBound(&stream, len+hdrlen);
2259        compressed = xmalloc(size);
2260
2261        /* Compress it */
2262        stream.next_out = compressed;
2263        stream.avail_out = size;
2264
2265        /* First header.. */
2266        stream.next_in = (unsigned char *)hdr;
2267        stream.avail_in = hdrlen;
2268        while (deflate(&stream, 0) == Z_OK)
2269                /* nothing */;
2270
2271        /* Then the data itself.. */
2272        stream.next_in = buf;
2273        stream.avail_in = len;
2274        ret = deflate(&stream, Z_FINISH);
2275        if (ret != Z_STREAM_END)
2276                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2277
2278        ret = deflateEnd(&stream);
2279        if (ret != Z_OK)
2280                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2281
2282        size = stream.total_out;
2283
2284        if (write_buffer(fd, compressed, size) < 0)
2285                die("unable to write sha1 file");
2286        close_sha1_file(fd);
2287        free(compressed);
2288
2289        if (mtime) {
2290                struct utimbuf utb;
2291                utb.actime = mtime;
2292                utb.modtime = mtime;
2293                if (utime(tmpfile, &utb) < 0)
2294                        warning("failed utime() on %s: %s",
2295                                tmpfile, strerror(errno));
2296        }
2297
2298        return move_temp_to_file(tmpfile, filename);
2299}
2300
2301int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2302{
2303        unsigned char sha1[20];
2304        char hdr[32];
2305        int hdrlen;
2306
2307        /* Normally if we have it in the pack then we do not bother writing
2308         * it out into .git/objects/??/?{38} file.
2309         */
2310        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2311        if (returnsha1)
2312                hashcpy(returnsha1, sha1);
2313        if (has_sha1_file(sha1))
2314                return 0;
2315        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2316}
2317
2318int force_object_loose(const unsigned char *sha1, time_t mtime)
2319{
2320        void *buf;
2321        unsigned long len;
2322        enum object_type type;
2323        char hdr[32];
2324        int hdrlen;
2325        int ret;
2326
2327        if (has_loose_object(sha1))
2328                return 0;
2329        buf = read_packed_sha1(sha1, &type, &len);
2330        if (!buf)
2331                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2332        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2333        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2334        free(buf);
2335
2336        return ret;
2337}
2338
2339int has_pack_index(const unsigned char *sha1)
2340{
2341        struct stat st;
2342        if (stat(sha1_pack_index_name(sha1), &st))
2343                return 0;
2344        return 1;
2345}
2346
2347int has_pack_file(const unsigned char *sha1)
2348{
2349        struct stat st;
2350        if (stat(sha1_pack_name(sha1), &st))
2351                return 0;
2352        return 1;
2353}
2354
2355int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2356{
2357        struct pack_entry e;
2358        return find_pack_entry(sha1, &e, ignore_packed);
2359}
2360
2361int has_sha1_file(const unsigned char *sha1)
2362{
2363        struct pack_entry e;
2364
2365        if (find_pack_entry(sha1, &e, NULL))
2366                return 1;
2367        return has_loose_object(sha1);
2368}
2369
2370int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2371{
2372        struct strbuf buf;
2373        int ret;
2374
2375        strbuf_init(&buf, 0);
2376        if (strbuf_read(&buf, fd, 4096) < 0) {
2377                strbuf_release(&buf);
2378                return -1;
2379        }
2380
2381        if (!type)
2382                type = blob_type;
2383        if (write_object)
2384                ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2385        else
2386                ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2387        strbuf_release(&buf);
2388
2389        return ret;
2390}
2391
2392int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2393             enum object_type type, const char *path)
2394{
2395        size_t size = xsize_t(st->st_size);
2396        void *buf = NULL;
2397        int ret, re_allocated = 0;
2398
2399        if (size)
2400                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2401        close(fd);
2402
2403        if (!type)
2404                type = OBJ_BLOB;
2405
2406        /*
2407         * Convert blobs to git internal format
2408         */
2409        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2410                struct strbuf nbuf;
2411                strbuf_init(&nbuf, 0);
2412                if (convert_to_git(path, buf, size, &nbuf,
2413                                   write_object ? safe_crlf : 0)) {
2414                        munmap(buf, size);
2415                        buf = strbuf_detach(&nbuf, &size);
2416                        re_allocated = 1;
2417                }
2418        }
2419
2420        if (write_object)
2421                ret = write_sha1_file(buf, size, typename(type), sha1);
2422        else
2423                ret = hash_sha1_file(buf, size, typename(type), sha1);
2424        if (re_allocated) {
2425                free(buf);
2426                return ret;
2427        }
2428        if (size)
2429                munmap(buf, size);
2430        return ret;
2431}
2432
2433int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2434{
2435        int fd;
2436        char *target;
2437        size_t len;
2438
2439        switch (st->st_mode & S_IFMT) {
2440        case S_IFREG:
2441                fd = open(path, O_RDONLY);
2442                if (fd < 0)
2443                        return error("open(\"%s\"): %s", path,
2444                                     strerror(errno));
2445                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2446                        return error("%s: failed to insert into database",
2447                                     path);
2448                break;
2449        case S_IFLNK:
2450                len = xsize_t(st->st_size);
2451                target = xmalloc(len + 1);
2452                if (readlink(path, target, len + 1) != st->st_size) {
2453                        char *errstr = strerror(errno);
2454                        free(target);
2455                        return error("readlink(\"%s\"): %s", path,
2456                                     errstr);
2457                }
2458                if (!write_object)
2459                        hash_sha1_file(target, len, blob_type, sha1);
2460                else if (write_sha1_file(target, len, blob_type, sha1))
2461                        return error("%s: failed to insert into database",
2462                                     path);
2463                free(target);
2464                break;
2465        case S_IFDIR:
2466                return resolve_gitlink_ref(path, "HEAD", sha1);
2467        default:
2468                return error("%s: unsupported file type", path);
2469        }
2470        return 0;
2471}
2472
2473int read_pack_header(int fd, struct pack_header *header)
2474{
2475        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2476                /* "eof before pack header was fully read" */
2477                return PH_ERROR_EOF;
2478
2479        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2480                /* "protocol error (pack signature mismatch detected)" */
2481                return PH_ERROR_PACK_SIGNATURE;
2482        if (!pack_version_ok(header->hdr_version))
2483                /* "protocol error (pack version unsupported)" */
2484                return PH_ERROR_PROTOCOL;
2485        return 0;
2486}