sha1_file.con commit t7700: test that 'repack -a' packs alternate packed objects (3289b9d)
   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"), LOCK_DIE_ON_ERROR);
 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_local(const unsigned char *sha1)
 414{
 415        char *name = sha1_file_name(sha1);
 416        return !access(name, F_OK);
 417}
 418
 419int has_loose_object_nonlocal(const unsigned char *sha1)
 420{
 421        struct alternate_object_database *alt;
 422        prepare_alt_odb();
 423        for (alt = alt_odb_list; alt; alt = alt->next) {
 424                fill_sha1_path(alt->name, sha1);
 425                if (!access(alt->base, F_OK))
 426                        return 1;
 427        }
 428        return 0;
 429}
 430
 431static int has_loose_object(const unsigned char *sha1)
 432{
 433        return has_loose_object_local(sha1) ||
 434               has_loose_object_nonlocal(sha1);
 435}
 436
 437static unsigned int pack_used_ctr;
 438static unsigned int pack_mmap_calls;
 439static unsigned int peak_pack_open_windows;
 440static unsigned int pack_open_windows;
 441static size_t peak_pack_mapped;
 442static size_t pack_mapped;
 443struct packed_git *packed_git;
 444
 445void pack_report(void)
 446{
 447        fprintf(stderr,
 448                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 449                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 450                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 451                sz_fmt(getpagesize()),
 452                sz_fmt(packed_git_window_size),
 453                sz_fmt(packed_git_limit));
 454        fprintf(stderr,
 455                "pack_report: pack_used_ctr            = %10u\n"
 456                "pack_report: pack_mmap_calls          = %10u\n"
 457                "pack_report: pack_open_windows        = %10u / %10u\n"
 458                "pack_report: pack_mapped              = "
 459                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 460                pack_used_ctr,
 461                pack_mmap_calls,
 462                pack_open_windows, peak_pack_open_windows,
 463                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 464}
 465
 466static int check_packed_git_idx(const char *path,  struct packed_git *p)
 467{
 468        void *idx_map;
 469        struct pack_idx_header *hdr;
 470        size_t idx_size;
 471        uint32_t version, nr, i, *index;
 472        int fd = open(path, O_RDONLY);
 473        struct stat st;
 474
 475        if (fd < 0)
 476                return -1;
 477        if (fstat(fd, &st)) {
 478                close(fd);
 479                return -1;
 480        }
 481        idx_size = xsize_t(st.st_size);
 482        if (idx_size < 4 * 256 + 20 + 20) {
 483                close(fd);
 484                return error("index file %s is too small", path);
 485        }
 486        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 487        close(fd);
 488
 489        hdr = idx_map;
 490        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 491                version = ntohl(hdr->idx_version);
 492                if (version < 2 || version > 2) {
 493                        munmap(idx_map, idx_size);
 494                        return error("index file %s is version %"PRIu32
 495                                     " and is not supported by this binary"
 496                                     " (try upgrading GIT to a newer version)",
 497                                     path, version);
 498                }
 499        } else
 500                version = 1;
 501
 502        nr = 0;
 503        index = idx_map;
 504        if (version > 1)
 505                index += 2;  /* skip index header */
 506        for (i = 0; i < 256; i++) {
 507                uint32_t n = ntohl(index[i]);
 508                if (n < nr) {
 509                        munmap(idx_map, idx_size);
 510                        return error("non-monotonic index %s", path);
 511                }
 512                nr = n;
 513        }
 514
 515        if (version == 1) {
 516                /*
 517                 * Total size:
 518                 *  - 256 index entries 4 bytes each
 519                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 520                 *  - 20-byte SHA1 of the packfile
 521                 *  - 20-byte SHA1 file checksum
 522                 */
 523                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 524                        munmap(idx_map, idx_size);
 525                        return error("wrong index v1 file size in %s", path);
 526                }
 527        } else if (version == 2) {
 528                /*
 529                 * Minimum size:
 530                 *  - 8 bytes of header
 531                 *  - 256 index entries 4 bytes each
 532                 *  - 20-byte sha1 entry * nr
 533                 *  - 4-byte crc entry * nr
 534                 *  - 4-byte offset entry * nr
 535                 *  - 20-byte SHA1 of the packfile
 536                 *  - 20-byte SHA1 file checksum
 537                 * And after the 4-byte offset table might be a
 538                 * variable sized table containing 8-byte entries
 539                 * for offsets larger than 2^31.
 540                 */
 541                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 542                unsigned long max_size = min_size;
 543                if (nr)
 544                        max_size += (nr - 1)*8;
 545                if (idx_size < min_size || idx_size > max_size) {
 546                        munmap(idx_map, idx_size);
 547                        return error("wrong index v2 file size in %s", path);
 548                }
 549                if (idx_size != min_size &&
 550                    /*
 551                     * make sure we can deal with large pack offsets.
 552                     * 31-bit signed offset won't be enough, neither
 553                     * 32-bit unsigned one will be.
 554                     */
 555                    (sizeof(off_t) <= 4)) {
 556                        munmap(idx_map, idx_size);
 557                        return error("pack too large for current definition of off_t in %s", path);
 558                }
 559        }
 560
 561        p->index_version = version;
 562        p->index_data = idx_map;
 563        p->index_size = idx_size;
 564        p->num_objects = nr;
 565        return 0;
 566}
 567
 568int open_pack_index(struct packed_git *p)
 569{
 570        char *idx_name;
 571        int ret;
 572
 573        if (p->index_data)
 574                return 0;
 575
 576        idx_name = xstrdup(p->pack_name);
 577        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 578        ret = check_packed_git_idx(idx_name, p);
 579        free(idx_name);
 580        return ret;
 581}
 582
 583static void scan_windows(struct packed_git *p,
 584        struct packed_git **lru_p,
 585        struct pack_window **lru_w,
 586        struct pack_window **lru_l)
 587{
 588        struct pack_window *w, *w_l;
 589
 590        for (w_l = NULL, w = p->windows; w; w = w->next) {
 591                if (!w->inuse_cnt) {
 592                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 593                                *lru_p = p;
 594                                *lru_w = w;
 595                                *lru_l = w_l;
 596                        }
 597                }
 598                w_l = w;
 599        }
 600}
 601
 602static int unuse_one_window(struct packed_git *current, int keep_fd)
 603{
 604        struct packed_git *p, *lru_p = NULL;
 605        struct pack_window *lru_w = NULL, *lru_l = NULL;
 606
 607        if (current)
 608                scan_windows(current, &lru_p, &lru_w, &lru_l);
 609        for (p = packed_git; p; p = p->next)
 610                scan_windows(p, &lru_p, &lru_w, &lru_l);
 611        if (lru_p) {
 612                munmap(lru_w->base, lru_w->len);
 613                pack_mapped -= lru_w->len;
 614                if (lru_l)
 615                        lru_l->next = lru_w->next;
 616                else {
 617                        lru_p->windows = lru_w->next;
 618                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 619                                close(lru_p->pack_fd);
 620                                lru_p->pack_fd = -1;
 621                        }
 622                }
 623                free(lru_w);
 624                pack_open_windows--;
 625                return 1;
 626        }
 627        return 0;
 628}
 629
 630void release_pack_memory(size_t need, int fd)
 631{
 632        size_t cur = pack_mapped;
 633        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 634                ; /* nothing */
 635}
 636
 637void close_pack_windows(struct packed_git *p)
 638{
 639        while (p->windows) {
 640                struct pack_window *w = p->windows;
 641
 642                if (w->inuse_cnt)
 643                        die("pack '%s' still has open windows to it",
 644                            p->pack_name);
 645                munmap(w->base, w->len);
 646                pack_mapped -= w->len;
 647                pack_open_windows--;
 648                p->windows = w->next;
 649                free(w);
 650        }
 651}
 652
 653void unuse_pack(struct pack_window **w_cursor)
 654{
 655        struct pack_window *w = *w_cursor;
 656        if (w) {
 657                w->inuse_cnt--;
 658                *w_cursor = NULL;
 659        }
 660}
 661
 662/*
 663 * Do not call this directly as this leaks p->pack_fd on error return;
 664 * call open_packed_git() instead.
 665 */
 666static int open_packed_git_1(struct packed_git *p)
 667{
 668        struct stat st;
 669        struct pack_header hdr;
 670        unsigned char sha1[20];
 671        unsigned char *idx_sha1;
 672        long fd_flag;
 673
 674        if (!p->index_data && open_pack_index(p))
 675                return error("packfile %s index unavailable", p->pack_name);
 676
 677        p->pack_fd = open(p->pack_name, O_RDONLY);
 678        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 679                return -1;
 680
 681        /* If we created the struct before we had the pack we lack size. */
 682        if (!p->pack_size) {
 683                if (!S_ISREG(st.st_mode))
 684                        return error("packfile %s not a regular file", p->pack_name);
 685                p->pack_size = st.st_size;
 686        } else if (p->pack_size != st.st_size)
 687                return error("packfile %s size changed", p->pack_name);
 688
 689        /* We leave these file descriptors open with sliding mmap;
 690         * there is no point keeping them open across exec(), though.
 691         */
 692        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 693        if (fd_flag < 0)
 694                return error("cannot determine file descriptor flags");
 695        fd_flag |= FD_CLOEXEC;
 696        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 697                return error("cannot set FD_CLOEXEC");
 698
 699        /* Verify we recognize this pack file format. */
 700        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 701                return error("file %s is far too short to be a packfile", p->pack_name);
 702        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 703                return error("file %s is not a GIT packfile", p->pack_name);
 704        if (!pack_version_ok(hdr.hdr_version))
 705                return error("packfile %s is version %"PRIu32" and not"
 706                        " supported (try upgrading GIT to a newer version)",
 707                        p->pack_name, ntohl(hdr.hdr_version));
 708
 709        /* Verify the pack matches its index. */
 710        if (p->num_objects != ntohl(hdr.hdr_entries))
 711                return error("packfile %s claims to have %"PRIu32" objects"
 712                             " while index indicates %"PRIu32" objects",
 713                             p->pack_name, ntohl(hdr.hdr_entries),
 714                             p->num_objects);
 715        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 716                return error("end of packfile %s is unavailable", p->pack_name);
 717        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 718                return error("packfile %s signature is unavailable", p->pack_name);
 719        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 720        if (hashcmp(sha1, idx_sha1))
 721                return error("packfile %s does not match index", p->pack_name);
 722        return 0;
 723}
 724
 725static int open_packed_git(struct packed_git *p)
 726{
 727        if (!open_packed_git_1(p))
 728                return 0;
 729        if (p->pack_fd != -1) {
 730                close(p->pack_fd);
 731                p->pack_fd = -1;
 732        }
 733        return -1;
 734}
 735
 736static int in_window(struct pack_window *win, off_t offset)
 737{
 738        /* We must promise at least 20 bytes (one hash) after the
 739         * offset is available from this window, otherwise the offset
 740         * is not actually in this window and a different window (which
 741         * has that one hash excess) must be used.  This is to support
 742         * the object header and delta base parsing routines below.
 743         */
 744        off_t win_off = win->offset;
 745        return win_off <= offset
 746                && (offset + 20) <= (win_off + win->len);
 747}
 748
 749unsigned char* use_pack(struct packed_git *p,
 750                struct pack_window **w_cursor,
 751                off_t offset,
 752                unsigned int *left)
 753{
 754        struct pack_window *win = *w_cursor;
 755
 756        if (p->pack_fd == -1 && open_packed_git(p))
 757                die("packfile %s cannot be accessed", p->pack_name);
 758
 759        /* Since packfiles end in a hash of their content and its
 760         * pointless to ask for an offset into the middle of that
 761         * hash, and the in_window function above wouldn't match
 762         * don't allow an offset too close to the end of the file.
 763         */
 764        if (offset > (p->pack_size - 20))
 765                die("offset beyond end of packfile (truncated pack?)");
 766
 767        if (!win || !in_window(win, offset)) {
 768                if (win)
 769                        win->inuse_cnt--;
 770                for (win = p->windows; win; win = win->next) {
 771                        if (in_window(win, offset))
 772                                break;
 773                }
 774                if (!win) {
 775                        size_t window_align = packed_git_window_size / 2;
 776                        off_t len;
 777                        win = xcalloc(1, sizeof(*win));
 778                        win->offset = (offset / window_align) * window_align;
 779                        len = p->pack_size - win->offset;
 780                        if (len > packed_git_window_size)
 781                                len = packed_git_window_size;
 782                        win->len = (size_t)len;
 783                        pack_mapped += win->len;
 784                        while (packed_git_limit < pack_mapped
 785                                && unuse_one_window(p, p->pack_fd))
 786                                ; /* nothing */
 787                        win->base = xmmap(NULL, win->len,
 788                                PROT_READ, MAP_PRIVATE,
 789                                p->pack_fd, win->offset);
 790                        if (win->base == MAP_FAILED)
 791                                die("packfile %s cannot be mapped: %s",
 792                                        p->pack_name,
 793                                        strerror(errno));
 794                        pack_mmap_calls++;
 795                        pack_open_windows++;
 796                        if (pack_mapped > peak_pack_mapped)
 797                                peak_pack_mapped = pack_mapped;
 798                        if (pack_open_windows > peak_pack_open_windows)
 799                                peak_pack_open_windows = pack_open_windows;
 800                        win->next = p->windows;
 801                        p->windows = win;
 802                }
 803        }
 804        if (win != *w_cursor) {
 805                win->last_used = pack_used_ctr++;
 806                win->inuse_cnt++;
 807                *w_cursor = win;
 808        }
 809        offset -= win->offset;
 810        if (left)
 811                *left = win->len - xsize_t(offset);
 812        return win->base + offset;
 813}
 814
 815static struct packed_git *alloc_packed_git(int extra)
 816{
 817        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 818        memset(p, 0, sizeof(*p));
 819        p->pack_fd = -1;
 820        return p;
 821}
 822
 823struct packed_git *add_packed_git(const char *path, int path_len, int local)
 824{
 825        struct stat st;
 826        struct packed_git *p = alloc_packed_git(path_len + 2);
 827
 828        /*
 829         * Make sure a corresponding .pack file exists and that
 830         * the index looks sane.
 831         */
 832        path_len -= strlen(".idx");
 833        if (path_len < 1) {
 834                free(p);
 835                return NULL;
 836        }
 837        memcpy(p->pack_name, path, path_len);
 838
 839        strcpy(p->pack_name + path_len, ".keep");
 840        if (!access(p->pack_name, F_OK))
 841                p->pack_keep = 1;
 842
 843        strcpy(p->pack_name + path_len, ".pack");
 844        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 845                free(p);
 846                return NULL;
 847        }
 848
 849        /* ok, it looks sane as far as we can check without
 850         * actually mapping the pack file.
 851         */
 852        p->pack_size = st.st_size;
 853        p->pack_local = local;
 854        p->mtime = st.st_mtime;
 855        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 856                hashclr(p->sha1);
 857        return p;
 858}
 859
 860struct packed_git *parse_pack_index(unsigned char *sha1)
 861{
 862        const char *idx_path = sha1_pack_index_name(sha1);
 863        const char *path = sha1_pack_name(sha1);
 864        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 865
 866        strcpy(p->pack_name, path);
 867        hashcpy(p->sha1, sha1);
 868        if (check_packed_git_idx(idx_path, p)) {
 869                free(p);
 870                return NULL;
 871        }
 872
 873        return p;
 874}
 875
 876void install_packed_git(struct packed_git *pack)
 877{
 878        pack->next = packed_git;
 879        packed_git = pack;
 880}
 881
 882static void prepare_packed_git_one(char *objdir, int local)
 883{
 884        /* Ensure that this buffer is large enough so that we can
 885           append "/pack/" without clobbering the stack even if
 886           strlen(objdir) were PATH_MAX.  */
 887        char path[PATH_MAX + 1 + 4 + 1 + 1];
 888        int len;
 889        DIR *dir;
 890        struct dirent *de;
 891
 892        sprintf(path, "%s/pack", objdir);
 893        len = strlen(path);
 894        dir = opendir(path);
 895        if (!dir) {
 896                if (errno != ENOENT)
 897                        error("unable to open object pack directory: %s: %s",
 898                              path, strerror(errno));
 899                return;
 900        }
 901        path[len++] = '/';
 902        while ((de = readdir(dir)) != NULL) {
 903                int namelen = strlen(de->d_name);
 904                struct packed_git *p;
 905
 906                if (!has_extension(de->d_name, ".idx"))
 907                        continue;
 908
 909                if (len + namelen + 1 > sizeof(path))
 910                        continue;
 911
 912                /* Don't reopen a pack we already have. */
 913                strcpy(path + len, de->d_name);
 914                for (p = packed_git; p; p = p->next) {
 915                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 916                                break;
 917                }
 918                if (p)
 919                        continue;
 920                /* See if it really is a valid .idx file with corresponding
 921                 * .pack file that we can map.
 922                 */
 923                p = add_packed_git(path, len + namelen, local);
 924                if (!p)
 925                        continue;
 926                install_packed_git(p);
 927        }
 928        closedir(dir);
 929}
 930
 931static int sort_pack(const void *a_, const void *b_)
 932{
 933        struct packed_git *a = *((struct packed_git **)a_);
 934        struct packed_git *b = *((struct packed_git **)b_);
 935        int st;
 936
 937        /*
 938         * Local packs tend to contain objects specific to our
 939         * variant of the project than remote ones.  In addition,
 940         * remote ones could be on a network mounted filesystem.
 941         * Favor local ones for these reasons.
 942         */
 943        st = a->pack_local - b->pack_local;
 944        if (st)
 945                return -st;
 946
 947        /*
 948         * Younger packs tend to contain more recent objects,
 949         * and more recent objects tend to get accessed more
 950         * often.
 951         */
 952        if (a->mtime < b->mtime)
 953                return 1;
 954        else if (a->mtime == b->mtime)
 955                return 0;
 956        return -1;
 957}
 958
 959static void rearrange_packed_git(void)
 960{
 961        struct packed_git **ary, *p;
 962        int i, n;
 963
 964        for (n = 0, p = packed_git; p; p = p->next)
 965                n++;
 966        if (n < 2)
 967                return;
 968
 969        /* prepare an array of packed_git for easier sorting */
 970        ary = xcalloc(n, sizeof(struct packed_git *));
 971        for (n = 0, p = packed_git; p; p = p->next)
 972                ary[n++] = p;
 973
 974        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 975
 976        /* link them back again */
 977        for (i = 0; i < n - 1; i++)
 978                ary[i]->next = ary[i + 1];
 979        ary[n - 1]->next = NULL;
 980        packed_git = ary[0];
 981
 982        free(ary);
 983}
 984
 985static int prepare_packed_git_run_once = 0;
 986void prepare_packed_git(void)
 987{
 988        struct alternate_object_database *alt;
 989
 990        if (prepare_packed_git_run_once)
 991                return;
 992        prepare_packed_git_one(get_object_directory(), 1);
 993        prepare_alt_odb();
 994        for (alt = alt_odb_list; alt; alt = alt->next) {
 995                alt->name[-1] = 0;
 996                prepare_packed_git_one(alt->base, 0);
 997                alt->name[-1] = '/';
 998        }
 999        rearrange_packed_git();
1000        prepare_packed_git_run_once = 1;
1001}
1002
1003void reprepare_packed_git(void)
1004{
1005        discard_revindex();
1006        prepare_packed_git_run_once = 0;
1007        prepare_packed_git();
1008}
1009
1010static void mark_bad_packed_object(struct packed_git *p,
1011                                   const unsigned char *sha1)
1012{
1013        unsigned i;
1014        for (i = 0; i < p->num_bad_objects; i++)
1015                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1016                        return;
1017        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1018        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1019        p->num_bad_objects++;
1020}
1021
1022static int has_packed_and_bad(const unsigned char *sha1)
1023{
1024        struct packed_git *p;
1025        unsigned i;
1026
1027        for (p = packed_git; p; p = p->next)
1028                for (i = 0; i < p->num_bad_objects; i++)
1029                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1030                                return 1;
1031        return 0;
1032}
1033
1034int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1035{
1036        unsigned char real_sha1[20];
1037        hash_sha1_file(map, size, type, real_sha1);
1038        return hashcmp(sha1, real_sha1) ? -1 : 0;
1039}
1040
1041static int git_open_noatime(const char *name)
1042{
1043        static int sha1_file_open_flag = O_NOATIME;
1044        int fd = open(name, O_RDONLY | sha1_file_open_flag);
1045
1046        /* Might the failure be due to O_NOATIME? */
1047        if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1048                fd = open(name, O_RDONLY);
1049                if (fd >= 0)
1050                        sha1_file_open_flag = 0;
1051        }
1052        return fd;
1053}
1054
1055static int open_sha1_file(const unsigned char *sha1)
1056{
1057        int fd;
1058        char *name = sha1_file_name(sha1);
1059        struct alternate_object_database *alt;
1060
1061        fd = git_open_noatime(name);
1062        if (fd >= 0)
1063                return fd;
1064
1065        prepare_alt_odb();
1066        errno = ENOENT;
1067        for (alt = alt_odb_list; alt; alt = alt->next) {
1068                name = alt->name;
1069                fill_sha1_path(name, sha1);
1070                fd = git_open_noatime(alt->base);
1071                if (fd >= 0)
1072                        return fd;
1073        }
1074        return -1;
1075}
1076
1077static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1078{
1079        void *map;
1080        int fd;
1081
1082        fd = open_sha1_file(sha1);
1083        map = NULL;
1084        if (fd >= 0) {
1085                struct stat st;
1086
1087                if (!fstat(fd, &st)) {
1088                        *size = xsize_t(st.st_size);
1089                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1090                }
1091                close(fd);
1092        }
1093        return map;
1094}
1095
1096static int legacy_loose_object(unsigned char *map)
1097{
1098        unsigned int word;
1099
1100        /*
1101         * Is it a zlib-compressed buffer? If so, the first byte
1102         * must be 0x78 (15-bit window size, deflated), and the
1103         * first 16-bit word is evenly divisible by 31
1104         */
1105        word = (map[0] << 8) + map[1];
1106        if (map[0] == 0x78 && !(word % 31))
1107                return 1;
1108        else
1109                return 0;
1110}
1111
1112unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1113{
1114        unsigned shift;
1115        unsigned char c;
1116        unsigned long size;
1117        unsigned long used = 0;
1118
1119        c = buf[used++];
1120        *type = (c >> 4) & 7;
1121        size = c & 15;
1122        shift = 4;
1123        while (c & 0x80) {
1124                if (len <= used)
1125                        return 0;
1126                if (sizeof(long) * 8 <= shift)
1127                        return 0;
1128                c = buf[used++];
1129                size += (c & 0x7f) << shift;
1130                shift += 7;
1131        }
1132        *sizep = size;
1133        return used;
1134}
1135
1136static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1137{
1138        unsigned long size, used;
1139        static const char valid_loose_object_type[8] = {
1140                0, /* OBJ_EXT */
1141                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1142                0, /* "delta" and others are invalid in a loose object */
1143        };
1144        enum object_type type;
1145
1146        /* Get the data stream */
1147        memset(stream, 0, sizeof(*stream));
1148        stream->next_in = map;
1149        stream->avail_in = mapsize;
1150        stream->next_out = buffer;
1151        stream->avail_out = bufsiz;
1152
1153        if (legacy_loose_object(map)) {
1154                inflateInit(stream);
1155                return inflate(stream, 0);
1156        }
1157
1158
1159        /*
1160         * There used to be a second loose object header format which
1161         * was meant to mimic the in-pack format, allowing for direct
1162         * copy of the object data.  This format turned up not to be
1163         * really worth it and we don't write it any longer.  But we
1164         * can still read it.
1165         */
1166        used = unpack_object_header_gently(map, mapsize, &type, &size);
1167        if (!used || !valid_loose_object_type[type])
1168                return -1;
1169        map += used;
1170        mapsize -= used;
1171
1172        /* Set up the stream for the rest.. */
1173        stream->next_in = map;
1174        stream->avail_in = mapsize;
1175        inflateInit(stream);
1176
1177        /* And generate the fake traditional header */
1178        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1179                                         typename(type), size);
1180        return 0;
1181}
1182
1183static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1184{
1185        int bytes = strlen(buffer) + 1;
1186        unsigned char *buf = xmalloc(1+size);
1187        unsigned long n;
1188        int status = Z_OK;
1189
1190        n = stream->total_out - bytes;
1191        if (n > size)
1192                n = size;
1193        memcpy(buf, (char *) buffer + bytes, n);
1194        bytes = n;
1195        if (bytes <= size) {
1196                /*
1197                 * The above condition must be (bytes <= size), not
1198                 * (bytes < size).  In other words, even though we
1199                 * expect no more output and set avail_out to zer0,
1200                 * the input zlib stream may have bytes that express
1201                 * "this concludes the stream", and we *do* want to
1202                 * eat that input.
1203                 *
1204                 * Otherwise we would not be able to test that we
1205                 * consumed all the input to reach the expected size;
1206                 * we also want to check that zlib tells us that all
1207                 * went well with status == Z_STREAM_END at the end.
1208                 */
1209                stream->next_out = buf + bytes;
1210                stream->avail_out = size - bytes;
1211                while (status == Z_OK)
1212                        status = inflate(stream, Z_FINISH);
1213        }
1214        buf[size] = 0;
1215        if (status == Z_STREAM_END && !stream->avail_in) {
1216                inflateEnd(stream);
1217                return buf;
1218        }
1219
1220        if (status < 0)
1221                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1222        else if (stream->avail_in)
1223                error("garbage at end of loose object '%s'",
1224                      sha1_to_hex(sha1));
1225        free(buf);
1226        return NULL;
1227}
1228
1229/*
1230 * We used to just use "sscanf()", but that's actually way
1231 * too permissive for what we want to check. So do an anal
1232 * object header parse by hand.
1233 */
1234static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1235{
1236        char type[10];
1237        int i;
1238        unsigned long size;
1239
1240        /*
1241         * The type can be at most ten bytes (including the
1242         * terminating '\0' that we add), and is followed by
1243         * a space.
1244         */
1245        i = 0;
1246        for (;;) {
1247                char c = *hdr++;
1248                if (c == ' ')
1249                        break;
1250                type[i++] = c;
1251                if (i >= sizeof(type))
1252                        return -1;
1253        }
1254        type[i] = 0;
1255
1256        /*
1257         * The length must follow immediately, and be in canonical
1258         * decimal format (ie "010" is not valid).
1259         */
1260        size = *hdr++ - '0';
1261        if (size > 9)
1262                return -1;
1263        if (size) {
1264                for (;;) {
1265                        unsigned long c = *hdr - '0';
1266                        if (c > 9)
1267                                break;
1268                        hdr++;
1269                        size = size * 10 + c;
1270                }
1271        }
1272        *sizep = size;
1273
1274        /*
1275         * The length must be followed by a zero byte
1276         */
1277        return *hdr ? -1 : type_from_string(type);
1278}
1279
1280static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1281{
1282        int ret;
1283        z_stream stream;
1284        char hdr[8192];
1285
1286        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1287        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1288                return NULL;
1289
1290        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1291}
1292
1293unsigned long get_size_from_delta(struct packed_git *p,
1294                                  struct pack_window **w_curs,
1295                                  off_t curpos)
1296{
1297        const unsigned char *data;
1298        unsigned char delta_head[20], *in;
1299        z_stream stream;
1300        int st;
1301
1302        memset(&stream, 0, sizeof(stream));
1303        stream.next_out = delta_head;
1304        stream.avail_out = sizeof(delta_head);
1305
1306        inflateInit(&stream);
1307        do {
1308                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1309                stream.next_in = in;
1310                st = inflate(&stream, Z_FINISH);
1311                curpos += stream.next_in - in;
1312        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1313                 stream.total_out < sizeof(delta_head));
1314        inflateEnd(&stream);
1315        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1316                die("delta data unpack-initial failed");
1317
1318        /* Examine the initial part of the delta to figure out
1319         * the result size.
1320         */
1321        data = delta_head;
1322
1323        /* ignore base size */
1324        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1325
1326        /* Read the result size */
1327        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1328}
1329
1330static off_t get_delta_base(struct packed_git *p,
1331                                    struct pack_window **w_curs,
1332                                    off_t *curpos,
1333                                    enum object_type type,
1334                                    off_t delta_obj_offset)
1335{
1336        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1337        off_t base_offset;
1338
1339        /* use_pack() assured us we have [base_info, base_info + 20)
1340         * as a range that we can look at without walking off the
1341         * end of the mapped window.  Its actually the hash size
1342         * that is assured.  An OFS_DELTA longer than the hash size
1343         * is stupid, as then a REF_DELTA would be smaller to store.
1344         */
1345        if (type == OBJ_OFS_DELTA) {
1346                unsigned used = 0;
1347                unsigned char c = base_info[used++];
1348                base_offset = c & 127;
1349                while (c & 128) {
1350                        base_offset += 1;
1351                        if (!base_offset || MSB(base_offset, 7))
1352                                return 0;  /* overflow */
1353                        c = base_info[used++];
1354                        base_offset = (base_offset << 7) + (c & 127);
1355                }
1356                base_offset = delta_obj_offset - base_offset;
1357                if (base_offset >= delta_obj_offset)
1358                        return 0;  /* out of bound */
1359                *curpos += used;
1360        } else if (type == OBJ_REF_DELTA) {
1361                /* The base entry _must_ be in the same pack */
1362                base_offset = find_pack_entry_one(base_info, p);
1363                *curpos += 20;
1364        } else
1365                die("I am totally screwed");
1366        return base_offset;
1367}
1368
1369/* forward declaration for a mutually recursive function */
1370static int packed_object_info(struct packed_git *p, off_t offset,
1371                              unsigned long *sizep);
1372
1373static int packed_delta_info(struct packed_git *p,
1374                             struct pack_window **w_curs,
1375                             off_t curpos,
1376                             enum object_type type,
1377                             off_t obj_offset,
1378                             unsigned long *sizep)
1379{
1380        off_t base_offset;
1381
1382        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1383        type = packed_object_info(p, base_offset, NULL);
1384
1385        /* We choose to only get the type of the base object and
1386         * ignore potentially corrupt pack file that expects the delta
1387         * based on a base with a wrong size.  This saves tons of
1388         * inflate() calls.
1389         */
1390        if (sizep)
1391                *sizep = get_size_from_delta(p, w_curs, curpos);
1392
1393        return type;
1394}
1395
1396static int unpack_object_header(struct packed_git *p,
1397                                struct pack_window **w_curs,
1398                                off_t *curpos,
1399                                unsigned long *sizep)
1400{
1401        unsigned char *base;
1402        unsigned int left;
1403        unsigned long used;
1404        enum object_type type;
1405
1406        /* use_pack() assures us we have [base, base + 20) available
1407         * as a range that we can look at at.  (Its actually the hash
1408         * size that is assured.)  With our object header encoding
1409         * the maximum deflated object size is 2^137, which is just
1410         * insane, so we know won't exceed what we have been given.
1411         */
1412        base = use_pack(p, w_curs, *curpos, &left);
1413        used = unpack_object_header_gently(base, left, &type, sizep);
1414        if (!used)
1415                die("object offset outside of pack file");
1416        *curpos += used;
1417
1418        return type;
1419}
1420
1421const char *packed_object_info_detail(struct packed_git *p,
1422                                      off_t obj_offset,
1423                                      unsigned long *size,
1424                                      unsigned long *store_size,
1425                                      unsigned int *delta_chain_length,
1426                                      unsigned char *base_sha1)
1427{
1428        struct pack_window *w_curs = NULL;
1429        off_t curpos;
1430        unsigned long dummy;
1431        unsigned char *next_sha1;
1432        enum object_type type;
1433        struct revindex_entry *revidx;
1434
1435        *delta_chain_length = 0;
1436        curpos = obj_offset;
1437        type = unpack_object_header(p, &w_curs, &curpos, size);
1438
1439        revidx = find_pack_revindex(p, obj_offset);
1440        *store_size = revidx[1].offset - obj_offset;
1441
1442        for (;;) {
1443                switch (type) {
1444                default:
1445                        die("pack %s contains unknown object type %d",
1446                            p->pack_name, type);
1447                case OBJ_COMMIT:
1448                case OBJ_TREE:
1449                case OBJ_BLOB:
1450                case OBJ_TAG:
1451                        unuse_pack(&w_curs);
1452                        return typename(type);
1453                case OBJ_OFS_DELTA:
1454                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1455                        if (!obj_offset)
1456                                die("pack %s contains bad delta base reference of type %s",
1457                                    p->pack_name, typename(type));
1458                        if (*delta_chain_length == 0) {
1459                                revidx = find_pack_revindex(p, obj_offset);
1460                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1461                        }
1462                        break;
1463                case OBJ_REF_DELTA:
1464                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1465                        if (*delta_chain_length == 0)
1466                                hashcpy(base_sha1, next_sha1);
1467                        obj_offset = find_pack_entry_one(next_sha1, p);
1468                        break;
1469                }
1470                (*delta_chain_length)++;
1471                curpos = obj_offset;
1472                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1473        }
1474}
1475
1476static int packed_object_info(struct packed_git *p, off_t obj_offset,
1477                              unsigned long *sizep)
1478{
1479        struct pack_window *w_curs = NULL;
1480        unsigned long size;
1481        off_t curpos = obj_offset;
1482        enum object_type type;
1483
1484        type = unpack_object_header(p, &w_curs, &curpos, &size);
1485
1486        switch (type) {
1487        case OBJ_OFS_DELTA:
1488        case OBJ_REF_DELTA:
1489                type = packed_delta_info(p, &w_curs, curpos,
1490                                         type, obj_offset, sizep);
1491                break;
1492        case OBJ_COMMIT:
1493        case OBJ_TREE:
1494        case OBJ_BLOB:
1495        case OBJ_TAG:
1496                if (sizep)
1497                        *sizep = size;
1498                break;
1499        default:
1500                die("pack %s contains unknown object type %d",
1501                    p->pack_name, type);
1502        }
1503        unuse_pack(&w_curs);
1504        return type;
1505}
1506
1507static void *unpack_compressed_entry(struct packed_git *p,
1508                                    struct pack_window **w_curs,
1509                                    off_t curpos,
1510                                    unsigned long size)
1511{
1512        int st;
1513        z_stream stream;
1514        unsigned char *buffer, *in;
1515
1516        buffer = xmalloc(size + 1);
1517        buffer[size] = 0;
1518        memset(&stream, 0, sizeof(stream));
1519        stream.next_out = buffer;
1520        stream.avail_out = size;
1521
1522        inflateInit(&stream);
1523        do {
1524                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1525                stream.next_in = in;
1526                st = inflate(&stream, Z_FINISH);
1527                curpos += stream.next_in - in;
1528        } while (st == Z_OK || st == Z_BUF_ERROR);
1529        inflateEnd(&stream);
1530        if ((st != Z_STREAM_END) || stream.total_out != size) {
1531                free(buffer);
1532                return NULL;
1533        }
1534
1535        return buffer;
1536}
1537
1538#define MAX_DELTA_CACHE (256)
1539
1540static size_t delta_base_cached;
1541
1542static struct delta_base_cache_lru_list {
1543        struct delta_base_cache_lru_list *prev;
1544        struct delta_base_cache_lru_list *next;
1545} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1546
1547static struct delta_base_cache_entry {
1548        struct delta_base_cache_lru_list lru;
1549        void *data;
1550        struct packed_git *p;
1551        off_t base_offset;
1552        unsigned long size;
1553        enum object_type type;
1554} delta_base_cache[MAX_DELTA_CACHE];
1555
1556static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1557{
1558        unsigned long hash;
1559
1560        hash = (unsigned long)p + (unsigned long)base_offset;
1561        hash += (hash >> 8) + (hash >> 16);
1562        return hash % MAX_DELTA_CACHE;
1563}
1564
1565static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1566        unsigned long *base_size, enum object_type *type, int keep_cache)
1567{
1568        void *ret;
1569        unsigned long hash = pack_entry_hash(p, base_offset);
1570        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1571
1572        ret = ent->data;
1573        if (ret && ent->p == p && ent->base_offset == base_offset)
1574                goto found_cache_entry;
1575        return unpack_entry(p, base_offset, type, base_size);
1576
1577found_cache_entry:
1578        if (!keep_cache) {
1579                ent->data = NULL;
1580                ent->lru.next->prev = ent->lru.prev;
1581                ent->lru.prev->next = ent->lru.next;
1582                delta_base_cached -= ent->size;
1583        } else {
1584                ret = xmemdupz(ent->data, ent->size);
1585        }
1586        *type = ent->type;
1587        *base_size = ent->size;
1588        return ret;
1589}
1590
1591static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1592{
1593        if (ent->data) {
1594                free(ent->data);
1595                ent->data = NULL;
1596                ent->lru.next->prev = ent->lru.prev;
1597                ent->lru.prev->next = ent->lru.next;
1598                delta_base_cached -= ent->size;
1599        }
1600}
1601
1602static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1603        void *base, unsigned long base_size, enum object_type type)
1604{
1605        unsigned long hash = pack_entry_hash(p, base_offset);
1606        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1607        struct delta_base_cache_lru_list *lru;
1608
1609        release_delta_base_cache(ent);
1610        delta_base_cached += base_size;
1611
1612        for (lru = delta_base_cache_lru.next;
1613             delta_base_cached > delta_base_cache_limit
1614             && lru != &delta_base_cache_lru;
1615             lru = lru->next) {
1616                struct delta_base_cache_entry *f = (void *)lru;
1617                if (f->type == OBJ_BLOB)
1618                        release_delta_base_cache(f);
1619        }
1620        for (lru = delta_base_cache_lru.next;
1621             delta_base_cached > delta_base_cache_limit
1622             && lru != &delta_base_cache_lru;
1623             lru = lru->next) {
1624                struct delta_base_cache_entry *f = (void *)lru;
1625                release_delta_base_cache(f);
1626        }
1627
1628        ent->p = p;
1629        ent->base_offset = base_offset;
1630        ent->type = type;
1631        ent->data = base;
1632        ent->size = base_size;
1633        ent->lru.next = &delta_base_cache_lru;
1634        ent->lru.prev = delta_base_cache_lru.prev;
1635        delta_base_cache_lru.prev->next = &ent->lru;
1636        delta_base_cache_lru.prev = &ent->lru;
1637}
1638
1639static void *unpack_delta_entry(struct packed_git *p,
1640                                struct pack_window **w_curs,
1641                                off_t curpos,
1642                                unsigned long delta_size,
1643                                off_t obj_offset,
1644                                enum object_type *type,
1645                                unsigned long *sizep)
1646{
1647        void *delta_data, *result, *base;
1648        unsigned long base_size;
1649        off_t base_offset;
1650
1651        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1652        if (!base_offset) {
1653                error("failed to validate delta base reference "
1654                      "at offset %"PRIuMAX" from %s",
1655                      (uintmax_t)curpos, p->pack_name);
1656                return NULL;
1657        }
1658        unuse_pack(w_curs);
1659        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1660        if (!base) {
1661                /*
1662                 * We're probably in deep shit, but let's try to fetch
1663                 * the required base anyway from another pack or loose.
1664                 * This is costly but should happen only in the presence
1665                 * of a corrupted pack, and is better than failing outright.
1666                 */
1667                struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1668                const unsigned char *base_sha1 =
1669                                        nth_packed_object_sha1(p, revidx->nr);
1670                error("failed to read delta base object %s"
1671                      " at offset %"PRIuMAX" from %s",
1672                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1673                      p->pack_name);
1674                mark_bad_packed_object(p, base_sha1);
1675                base = read_object(base_sha1, type, &base_size);
1676                if (!base)
1677                        return NULL;
1678        }
1679
1680        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1681        if (!delta_data) {
1682                error("failed to unpack compressed delta "
1683                      "at offset %"PRIuMAX" from %s",
1684                      (uintmax_t)curpos, p->pack_name);
1685                free(base);
1686                return NULL;
1687        }
1688        result = patch_delta(base, base_size,
1689                             delta_data, delta_size,
1690                             sizep);
1691        if (!result)
1692                die("failed to apply delta");
1693        free(delta_data);
1694        add_delta_base_cache(p, base_offset, base, base_size, *type);
1695        return result;
1696}
1697
1698void *unpack_entry(struct packed_git *p, off_t obj_offset,
1699                   enum object_type *type, unsigned long *sizep)
1700{
1701        struct pack_window *w_curs = NULL;
1702        off_t curpos = obj_offset;
1703        void *data;
1704
1705        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1706        switch (*type) {
1707        case OBJ_OFS_DELTA:
1708        case OBJ_REF_DELTA:
1709                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1710                                          obj_offset, type, sizep);
1711                break;
1712        case OBJ_COMMIT:
1713        case OBJ_TREE:
1714        case OBJ_BLOB:
1715        case OBJ_TAG:
1716                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1717                break;
1718        default:
1719                data = NULL;
1720                error("unknown object type %i at offset %"PRIuMAX" in %s",
1721                      *type, (uintmax_t)obj_offset, p->pack_name);
1722        }
1723        unuse_pack(&w_curs);
1724        return data;
1725}
1726
1727const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1728                                            uint32_t n)
1729{
1730        const unsigned char *index = p->index_data;
1731        if (!index) {
1732                if (open_pack_index(p))
1733                        return NULL;
1734                index = p->index_data;
1735        }
1736        if (n >= p->num_objects)
1737                return NULL;
1738        index += 4 * 256;
1739        if (p->index_version == 1) {
1740                return index + 24 * n + 4;
1741        } else {
1742                index += 8;
1743                return index + 20 * n;
1744        }
1745}
1746
1747off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1748{
1749        const unsigned char *index = p->index_data;
1750        index += 4 * 256;
1751        if (p->index_version == 1) {
1752                return ntohl(*((uint32_t *)(index + 24 * n)));
1753        } else {
1754                uint32_t off;
1755                index += 8 + p->num_objects * (20 + 4);
1756                off = ntohl(*((uint32_t *)(index + 4 * n)));
1757                if (!(off & 0x80000000))
1758                        return off;
1759                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1760                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1761                                   ntohl(*((uint32_t *)(index + 4)));
1762        }
1763}
1764
1765off_t find_pack_entry_one(const unsigned char *sha1,
1766                                  struct packed_git *p)
1767{
1768        const uint32_t *level1_ofs = p->index_data;
1769        const unsigned char *index = p->index_data;
1770        unsigned hi, lo, stride;
1771        static int use_lookup = -1;
1772        static int debug_lookup = -1;
1773
1774        if (debug_lookup < 0)
1775                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1776
1777        if (!index) {
1778                if (open_pack_index(p))
1779                        return 0;
1780                level1_ofs = p->index_data;
1781                index = p->index_data;
1782        }
1783        if (p->index_version > 1) {
1784                level1_ofs += 2;
1785                index += 8;
1786        }
1787        index += 4 * 256;
1788        hi = ntohl(level1_ofs[*sha1]);
1789        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1790        if (p->index_version > 1) {
1791                stride = 20;
1792        } else {
1793                stride = 24;
1794                index += 4;
1795        }
1796
1797        if (debug_lookup)
1798                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1799                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1800
1801        if (use_lookup < 0)
1802                use_lookup = !!getenv("GIT_USE_LOOKUP");
1803        if (use_lookup) {
1804                int pos = sha1_entry_pos(index, stride, 0,
1805                                         lo, hi, p->num_objects, sha1);
1806                if (pos < 0)
1807                        return 0;
1808                return nth_packed_object_offset(p, pos);
1809        }
1810
1811        do {
1812                unsigned mi = (lo + hi) / 2;
1813                int cmp = hashcmp(index + mi * stride, sha1);
1814
1815                if (debug_lookup)
1816                        printf("lo %u hi %u rg %u mi %u\n",
1817                               lo, hi, hi - lo, mi);
1818                if (!cmp)
1819                        return nth_packed_object_offset(p, mi);
1820                if (cmp > 0)
1821                        hi = mi;
1822                else
1823                        lo = mi+1;
1824        } while (lo < hi);
1825        return 0;
1826}
1827
1828int matches_pack_name(struct packed_git *p, const char *name)
1829{
1830        const char *last_c, *c;
1831
1832        if (!strcmp(p->pack_name, name))
1833                return 1;
1834
1835        for (c = p->pack_name, last_c = c; *c;)
1836                if (*c == '/')
1837                        last_c = ++c;
1838                else
1839                        ++c;
1840        if (!strcmp(last_c, name))
1841                return 1;
1842
1843        return 0;
1844}
1845
1846static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1847{
1848        static struct packed_git *last_found = (void *)1;
1849        struct packed_git *p;
1850        off_t offset;
1851
1852        prepare_packed_git();
1853        if (!packed_git)
1854                return 0;
1855        p = (last_found == (void *)1) ? packed_git : last_found;
1856
1857        do {
1858                if (ignore_packed) {
1859                        const char **ig;
1860                        for (ig = ignore_packed; *ig; ig++)
1861                                if (matches_pack_name(p, *ig))
1862                                        break;
1863                        if (*ig)
1864                                goto next;
1865                }
1866
1867                if (p->num_bad_objects) {
1868                        unsigned i;
1869                        for (i = 0; i < p->num_bad_objects; i++)
1870                                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1871                                        goto next;
1872                }
1873
1874                offset = find_pack_entry_one(sha1, p);
1875                if (offset) {
1876                        /*
1877                         * We are about to tell the caller where they can
1878                         * locate the requested object.  We better make
1879                         * sure the packfile is still here and can be
1880                         * accessed before supplying that answer, as
1881                         * it may have been deleted since the index
1882                         * was loaded!
1883                         */
1884                        if (p->pack_fd == -1 && open_packed_git(p)) {
1885                                error("packfile %s cannot be accessed", p->pack_name);
1886                                goto next;
1887                        }
1888                        e->offset = offset;
1889                        e->p = p;
1890                        hashcpy(e->sha1, sha1);
1891                        last_found = p;
1892                        return 1;
1893                }
1894
1895                next:
1896                if (p == last_found)
1897                        p = packed_git;
1898                else
1899                        p = p->next;
1900                if (p == last_found)
1901                        p = p->next;
1902        } while (p);
1903        return 0;
1904}
1905
1906struct packed_git *find_sha1_pack(const unsigned char *sha1,
1907                                  struct packed_git *packs)
1908{
1909        struct packed_git *p;
1910
1911        for (p = packs; p; p = p->next) {
1912                if (find_pack_entry_one(sha1, p))
1913                        return p;
1914        }
1915        return NULL;
1916
1917}
1918
1919static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1920{
1921        int status;
1922        unsigned long mapsize, size;
1923        void *map;
1924        z_stream stream;
1925        char hdr[32];
1926
1927        map = map_sha1_file(sha1, &mapsize);
1928        if (!map)
1929                return error("unable to find %s", sha1_to_hex(sha1));
1930        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1931                status = error("unable to unpack %s header",
1932                               sha1_to_hex(sha1));
1933        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1934                status = error("unable to parse %s header", sha1_to_hex(sha1));
1935        else if (sizep)
1936                *sizep = size;
1937        inflateEnd(&stream);
1938        munmap(map, mapsize);
1939        return status;
1940}
1941
1942int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1943{
1944        struct pack_entry e;
1945        int status;
1946
1947        if (!find_pack_entry(sha1, &e, NULL)) {
1948                /* Most likely it's a loose object. */
1949                status = sha1_loose_object_info(sha1, sizep);
1950                if (status >= 0)
1951                        return status;
1952
1953                /* Not a loose object; someone else may have just packed it. */
1954                reprepare_packed_git();
1955                if (!find_pack_entry(sha1, &e, NULL))
1956                        return status;
1957        }
1958        return packed_object_info(e.p, e.offset, sizep);
1959}
1960
1961static void *read_packed_sha1(const unsigned char *sha1,
1962                              enum object_type *type, unsigned long *size)
1963{
1964        struct pack_entry e;
1965        void *data;
1966
1967        if (!find_pack_entry(sha1, &e, NULL))
1968                return NULL;
1969        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1970        if (!data) {
1971                /*
1972                 * We're probably in deep shit, but let's try to fetch
1973                 * the required object anyway from another pack or loose.
1974                 * This should happen only in the presence of a corrupted
1975                 * pack, and is better than failing outright.
1976                 */
1977                error("failed to read object %s at offset %"PRIuMAX" from %s",
1978                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1979                mark_bad_packed_object(e.p, sha1);
1980                data = read_object(sha1, type, size);
1981        }
1982        return data;
1983}
1984
1985/*
1986 * This is meant to hold a *small* number of objects that you would
1987 * want read_sha1_file() to be able to return, but yet you do not want
1988 * to write them into the object store (e.g. a browse-only
1989 * application).
1990 */
1991static struct cached_object {
1992        unsigned char sha1[20];
1993        enum object_type type;
1994        void *buf;
1995        unsigned long size;
1996} *cached_objects;
1997static int cached_object_nr, cached_object_alloc;
1998
1999static struct cached_object empty_tree = {
2000        /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2001        "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2002        "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2003        OBJ_TREE,
2004        "",
2005        0
2006};
2007
2008static struct cached_object *find_cached_object(const unsigned char *sha1)
2009{
2010        int i;
2011        struct cached_object *co = cached_objects;
2012
2013        for (i = 0; i < cached_object_nr; i++, co++) {
2014                if (!hashcmp(co->sha1, sha1))
2015                        return co;
2016        }
2017        if (!hashcmp(sha1, empty_tree.sha1))
2018                return &empty_tree;
2019        return NULL;
2020}
2021
2022int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2023                      unsigned char *sha1)
2024{
2025        struct cached_object *co;
2026
2027        hash_sha1_file(buf, len, typename(type), sha1);
2028        if (has_sha1_file(sha1) || find_cached_object(sha1))
2029                return 0;
2030        if (cached_object_alloc <= cached_object_nr) {
2031                cached_object_alloc = alloc_nr(cached_object_alloc);
2032                cached_objects = xrealloc(cached_objects,
2033                                          sizeof(*cached_objects) *
2034                                          cached_object_alloc);
2035        }
2036        co = &cached_objects[cached_object_nr++];
2037        co->size = len;
2038        co->type = type;
2039        co->buf = xmalloc(len);
2040        memcpy(co->buf, buf, len);
2041        hashcpy(co->sha1, sha1);
2042        return 0;
2043}
2044
2045void *read_object(const unsigned char *sha1, enum object_type *type,
2046                  unsigned long *size)
2047{
2048        unsigned long mapsize;
2049        void *map, *buf;
2050        struct cached_object *co;
2051
2052        co = find_cached_object(sha1);
2053        if (co) {
2054                *type = co->type;
2055                *size = co->size;
2056                return xmemdupz(co->buf, co->size);
2057        }
2058
2059        buf = read_packed_sha1(sha1, type, size);
2060        if (buf)
2061                return buf;
2062        map = map_sha1_file(sha1, &mapsize);
2063        if (map) {
2064                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2065                munmap(map, mapsize);
2066                return buf;
2067        }
2068        reprepare_packed_git();
2069        return read_packed_sha1(sha1, type, size);
2070}
2071
2072void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2073                     unsigned long *size)
2074{
2075        void *data = read_object(sha1, type, size);
2076        /* legacy behavior is to die on corrupted objects */
2077        if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2078                die("object %s is corrupted", sha1_to_hex(sha1));
2079        return data;
2080}
2081
2082void *read_object_with_reference(const unsigned char *sha1,
2083                                 const char *required_type_name,
2084                                 unsigned long *size,
2085                                 unsigned char *actual_sha1_return)
2086{
2087        enum object_type type, required_type;
2088        void *buffer;
2089        unsigned long isize;
2090        unsigned char actual_sha1[20];
2091
2092        required_type = type_from_string(required_type_name);
2093        hashcpy(actual_sha1, sha1);
2094        while (1) {
2095                int ref_length = -1;
2096                const char *ref_type = NULL;
2097
2098                buffer = read_sha1_file(actual_sha1, &type, &isize);
2099                if (!buffer)
2100                        return NULL;
2101                if (type == required_type) {
2102                        *size = isize;
2103                        if (actual_sha1_return)
2104                                hashcpy(actual_sha1_return, actual_sha1);
2105                        return buffer;
2106                }
2107                /* Handle references */
2108                else if (type == OBJ_COMMIT)
2109                        ref_type = "tree ";
2110                else if (type == OBJ_TAG)
2111                        ref_type = "object ";
2112                else {
2113                        free(buffer);
2114                        return NULL;
2115                }
2116                ref_length = strlen(ref_type);
2117
2118                if (ref_length + 40 > isize ||
2119                    memcmp(buffer, ref_type, ref_length) ||
2120                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2121                        free(buffer);
2122                        return NULL;
2123                }
2124                free(buffer);
2125                /* Now we have the ID of the referred-to object in
2126                 * actual_sha1.  Check again. */
2127        }
2128}
2129
2130static void write_sha1_file_prepare(const void *buf, unsigned long len,
2131                                    const char *type, unsigned char *sha1,
2132                                    char *hdr, int *hdrlen)
2133{
2134        SHA_CTX c;
2135
2136        /* Generate the header */
2137        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2138
2139        /* Sha1.. */
2140        SHA1_Init(&c);
2141        SHA1_Update(&c, hdr, *hdrlen);
2142        SHA1_Update(&c, buf, len);
2143        SHA1_Final(sha1, &c);
2144}
2145
2146/*
2147 * Move the just written object into its final resting place
2148 */
2149int move_temp_to_file(const char *tmpfile, const char *filename)
2150{
2151        int ret = 0;
2152        if (link(tmpfile, filename))
2153                ret = errno;
2154
2155        /*
2156         * Coda hack - coda doesn't like cross-directory links,
2157         * so we fall back to a rename, which will mean that it
2158         * won't be able to check collisions, but that's not a
2159         * big deal.
2160         *
2161         * The same holds for FAT formatted media.
2162         *
2163         * When this succeeds, we just return 0. We have nothing
2164         * left to unlink.
2165         */
2166        if (ret && ret != EEXIST) {
2167                if (!rename(tmpfile, filename))
2168                        return 0;
2169                ret = errno;
2170        }
2171        unlink(tmpfile);
2172        if (ret) {
2173                if (ret != EEXIST) {
2174                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2175                }
2176                /* FIXME!!! Collision check here ? */
2177        }
2178
2179        return 0;
2180}
2181
2182static int write_buffer(int fd, const void *buf, size_t len)
2183{
2184        if (write_in_full(fd, buf, len) < 0)
2185                return error("file write error (%s)", strerror(errno));
2186        return 0;
2187}
2188
2189int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2190                   unsigned char *sha1)
2191{
2192        char hdr[32];
2193        int hdrlen;
2194        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2195        return 0;
2196}
2197
2198/* Finalize a file on disk, and close it. */
2199static void close_sha1_file(int fd)
2200{
2201        if (fsync_object_files)
2202                fsync_or_die(fd, "sha1 file");
2203        fchmod(fd, 0444);
2204        if (close(fd) != 0)
2205                die("unable to write sha1 file");
2206}
2207
2208/* Size of directory component, including the ending '/' */
2209static inline int directory_size(const char *filename)
2210{
2211        const char *s = strrchr(filename, '/');
2212        if (!s)
2213                return 0;
2214        return s - filename + 1;
2215}
2216
2217/*
2218 * This creates a temporary file in the same directory as the final
2219 * 'filename'
2220 *
2221 * We want to avoid cross-directory filename renames, because those
2222 * can have problems on various filesystems (FAT, NFS, Coda).
2223 */
2224static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2225{
2226        int fd, dirlen = directory_size(filename);
2227
2228        if (dirlen + 20 > bufsiz) {
2229                errno = ENAMETOOLONG;
2230                return -1;
2231        }
2232        memcpy(buffer, filename, dirlen);
2233        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2234        fd = mkstemp(buffer);
2235        if (fd < 0 && dirlen) {
2236                /* Make sure the directory exists */
2237                memcpy(buffer, filename, dirlen);
2238                buffer[dirlen-1] = 0;
2239                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2240                        return -1;
2241
2242                /* Try again */
2243                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2244                fd = mkstemp(buffer);
2245        }
2246        return fd;
2247}
2248
2249static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2250                              void *buf, unsigned long len, time_t mtime)
2251{
2252        int fd, size, ret;
2253        unsigned char *compressed;
2254        z_stream stream;
2255        char *filename;
2256        static char tmpfile[PATH_MAX];
2257
2258        filename = sha1_file_name(sha1);
2259        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2260        if (fd < 0) {
2261                if (errno == EPERM)
2262                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2263                else
2264                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2265        }
2266
2267        /* Set it up */
2268        memset(&stream, 0, sizeof(stream));
2269        deflateInit(&stream, zlib_compression_level);
2270        size = 8 + deflateBound(&stream, len+hdrlen);
2271        compressed = xmalloc(size);
2272
2273        /* Compress it */
2274        stream.next_out = compressed;
2275        stream.avail_out = size;
2276
2277        /* First header.. */
2278        stream.next_in = (unsigned char *)hdr;
2279        stream.avail_in = hdrlen;
2280        while (deflate(&stream, 0) == Z_OK)
2281                /* nothing */;
2282
2283        /* Then the data itself.. */
2284        stream.next_in = buf;
2285        stream.avail_in = len;
2286        ret = deflate(&stream, Z_FINISH);
2287        if (ret != Z_STREAM_END)
2288                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2289
2290        ret = deflateEnd(&stream);
2291        if (ret != Z_OK)
2292                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2293
2294        size = stream.total_out;
2295
2296        if (write_buffer(fd, compressed, size) < 0)
2297                die("unable to write sha1 file");
2298        close_sha1_file(fd);
2299        free(compressed);
2300
2301        if (mtime) {
2302                struct utimbuf utb;
2303                utb.actime = mtime;
2304                utb.modtime = mtime;
2305                if (utime(tmpfile, &utb) < 0)
2306                        warning("failed utime() on %s: %s",
2307                                tmpfile, strerror(errno));
2308        }
2309
2310        return move_temp_to_file(tmpfile, filename);
2311}
2312
2313int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2314{
2315        unsigned char sha1[20];
2316        char hdr[32];
2317        int hdrlen;
2318
2319        /* Normally if we have it in the pack then we do not bother writing
2320         * it out into .git/objects/??/?{38} file.
2321         */
2322        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2323        if (returnsha1)
2324                hashcpy(returnsha1, sha1);
2325        if (has_sha1_file(sha1))
2326                return 0;
2327        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2328}
2329
2330int force_object_loose(const unsigned char *sha1, time_t mtime)
2331{
2332        void *buf;
2333        unsigned long len;
2334        enum object_type type;
2335        char hdr[32];
2336        int hdrlen;
2337        int ret;
2338
2339        if (has_loose_object(sha1))
2340                return 0;
2341        buf = read_packed_sha1(sha1, &type, &len);
2342        if (!buf)
2343                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2344        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2345        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2346        free(buf);
2347
2348        return ret;
2349}
2350
2351int has_pack_index(const unsigned char *sha1)
2352{
2353        struct stat st;
2354        if (stat(sha1_pack_index_name(sha1), &st))
2355                return 0;
2356        return 1;
2357}
2358
2359int has_pack_file(const unsigned char *sha1)
2360{
2361        struct stat st;
2362        if (stat(sha1_pack_name(sha1), &st))
2363                return 0;
2364        return 1;
2365}
2366
2367int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2368{
2369        struct pack_entry e;
2370        return find_pack_entry(sha1, &e, ignore_packed);
2371}
2372
2373int has_sha1_file(const unsigned char *sha1)
2374{
2375        struct pack_entry e;
2376
2377        if (find_pack_entry(sha1, &e, NULL))
2378                return 1;
2379        return has_loose_object(sha1);
2380}
2381
2382int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2383{
2384        struct strbuf buf;
2385        int ret;
2386
2387        strbuf_init(&buf, 0);
2388        if (strbuf_read(&buf, fd, 4096) < 0) {
2389                strbuf_release(&buf);
2390                return -1;
2391        }
2392
2393        if (!type)
2394                type = blob_type;
2395        if (write_object)
2396                ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2397        else
2398                ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2399        strbuf_release(&buf);
2400
2401        return ret;
2402}
2403
2404int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2405             enum object_type type, const char *path)
2406{
2407        size_t size = xsize_t(st->st_size);
2408        void *buf = NULL;
2409        int ret, re_allocated = 0;
2410
2411        if (size)
2412                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2413        close(fd);
2414
2415        if (!type)
2416                type = OBJ_BLOB;
2417
2418        /*
2419         * Convert blobs to git internal format
2420         */
2421        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2422                struct strbuf nbuf;
2423                strbuf_init(&nbuf, 0);
2424                if (convert_to_git(path, buf, size, &nbuf,
2425                                   write_object ? safe_crlf : 0)) {
2426                        munmap(buf, size);
2427                        buf = strbuf_detach(&nbuf, &size);
2428                        re_allocated = 1;
2429                }
2430        }
2431
2432        if (write_object)
2433                ret = write_sha1_file(buf, size, typename(type), sha1);
2434        else
2435                ret = hash_sha1_file(buf, size, typename(type), sha1);
2436        if (re_allocated) {
2437                free(buf);
2438                return ret;
2439        }
2440        if (size)
2441                munmap(buf, size);
2442        return ret;
2443}
2444
2445int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2446{
2447        int fd;
2448        char *target;
2449        size_t len;
2450
2451        switch (st->st_mode & S_IFMT) {
2452        case S_IFREG:
2453                fd = open(path, O_RDONLY);
2454                if (fd < 0)
2455                        return error("open(\"%s\"): %s", path,
2456                                     strerror(errno));
2457                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2458                        return error("%s: failed to insert into database",
2459                                     path);
2460                break;
2461        case S_IFLNK:
2462                len = xsize_t(st->st_size);
2463                target = xmalloc(len + 1);
2464                if (readlink(path, target, len + 1) != st->st_size) {
2465                        char *errstr = strerror(errno);
2466                        free(target);
2467                        return error("readlink(\"%s\"): %s", path,
2468                                     errstr);
2469                }
2470                if (!write_object)
2471                        hash_sha1_file(target, len, blob_type, sha1);
2472                else if (write_sha1_file(target, len, blob_type, sha1))
2473                        return error("%s: failed to insert into database",
2474                                     path);
2475                free(target);
2476                break;
2477        case S_IFDIR:
2478                return resolve_gitlink_ref(path, "HEAD", sha1);
2479        default:
2480                return error("%s: unsupported file type", path);
2481        }
2482        return 0;
2483}
2484
2485int read_pack_header(int fd, struct pack_header *header)
2486{
2487        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2488                /* "eof before pack header was fully read" */
2489                return PH_ERROR_EOF;
2490
2491        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2492                /* "protocol error (pack signature mismatch detected)" */
2493                return PH_ERROR_PACK_SIGNATURE;
2494        if (!pack_version_ok(header->hdr_version))
2495                /* "protocol error (pack version unsupported)" */
2496                return PH_ERROR_PROTOCOL;
2497        return 0;
2498}