sha1_file.con commit gitweb: quote commands properly when calling the shell (516381d)
   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
  86int safe_create_leading_directories(char *path)
  87{
  88        char *pos = path;
  89        struct stat st;
  90
  91        if (is_absolute_path(path))
  92                pos++;
  93
  94        while (pos) {
  95                pos = strchr(pos, '/');
  96                if (!pos)
  97                        break;
  98                *pos = 0;
  99                if (!stat(path, &st)) {
 100                        /* path exists */
 101                        if (!S_ISDIR(st.st_mode)) {
 102                                *pos = '/';
 103                                return -3;
 104                        }
 105                }
 106                else if (mkdir(path, 0777)) {
 107                        *pos = '/';
 108                        return -1;
 109                }
 110                else if (adjust_shared_perm(path)) {
 111                        *pos = '/';
 112                        return -2;
 113                }
 114                *pos++ = '/';
 115        }
 116        return 0;
 117}
 118
 119char *sha1_to_hex(const unsigned char *sha1)
 120{
 121        static int bufno;
 122        static char hexbuffer[4][50];
 123        static const char hex[] = "0123456789abcdef";
 124        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 125        int i;
 126
 127        for (i = 0; i < 20; i++) {
 128                unsigned int val = *sha1++;
 129                *buf++ = hex[val >> 4];
 130                *buf++ = hex[val & 0xf];
 131        }
 132        *buf = '\0';
 133
 134        return buffer;
 135}
 136
 137static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 138{
 139        int i;
 140        for (i = 0; i < 20; i++) {
 141                static char hex[] = "0123456789abcdef";
 142                unsigned int val = sha1[i];
 143                char *pos = pathbuf + i*2 + (i > 0);
 144                *pos++ = hex[val >> 4];
 145                *pos = hex[val & 0xf];
 146        }
 147}
 148
 149/*
 150 * NOTE! This returns a statically allocated buffer, so you have to be
 151 * careful about using it. Do an "xstrdup()" if you need to save the
 152 * filename.
 153 *
 154 * Also note that this returns the location for creating.  Reading
 155 * SHA1 file can happen from any alternate directory listed in the
 156 * DB_ENVIRONMENT environment variable if it is not found in
 157 * the primary object database.
 158 */
 159char *sha1_file_name(const unsigned char *sha1)
 160{
 161        static char *name, *base;
 162
 163        if (!base) {
 164                const char *sha1_file_directory = get_object_directory();
 165                int len = strlen(sha1_file_directory);
 166                base = xmalloc(len + 60);
 167                memcpy(base, sha1_file_directory, len);
 168                memset(base+len, 0, 60);
 169                base[len] = '/';
 170                base[len+3] = '/';
 171                name = base + len + 1;
 172        }
 173        fill_sha1_path(name, sha1);
 174        return base;
 175}
 176
 177static char *sha1_get_pack_name(const unsigned char *sha1,
 178                                char **name, char **base, const char *which)
 179{
 180        static const char hex[] = "0123456789abcdef";
 181        char *buf;
 182        int i;
 183
 184        if (!*base) {
 185                const char *sha1_file_directory = get_object_directory();
 186                int len = strlen(sha1_file_directory);
 187                *base = xmalloc(len + 60);
 188                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 189                        sha1_file_directory, which);
 190                *name = *base + len + 11;
 191        }
 192
 193        buf = *name;
 194
 195        for (i = 0; i < 20; i++) {
 196                unsigned int val = *sha1++;
 197                *buf++ = hex[val >> 4];
 198                *buf++ = hex[val & 0xf];
 199        }
 200
 201        return *base;
 202}
 203
 204char *sha1_pack_name(const unsigned char *sha1)
 205{
 206        static char *name, *base;
 207
 208        return sha1_get_pack_name(sha1, &name, &base, "pack");
 209}
 210
 211char *sha1_pack_index_name(const unsigned char *sha1)
 212{
 213        static char *name, *base;
 214
 215        return sha1_get_pack_name(sha1, &name, &base, "idx");
 216}
 217
 218struct alternate_object_database *alt_odb_list;
 219static struct alternate_object_database **alt_odb_tail;
 220
 221static void read_info_alternates(const char * alternates, int depth);
 222
 223/*
 224 * Prepare alternate object database registry.
 225 *
 226 * The variable alt_odb_list points at the list of struct
 227 * alternate_object_database.  The elements on this list come from
 228 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 229 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 230 * whose contents is similar to that environment variable but can be
 231 * LF separated.  Its base points at a statically allocated buffer that
 232 * contains "/the/directory/corresponding/to/.git/objects/...", while
 233 * its name points just after the slash at the end of ".git/objects/"
 234 * in the example above, and has enough space to hold 40-byte hex
 235 * SHA1, an extra slash for the first level indirection, and the
 236 * terminating NUL.
 237 */
 238static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 239{
 240        struct stat st;
 241        const char *objdir = get_object_directory();
 242        struct alternate_object_database *ent;
 243        struct alternate_object_database *alt;
 244        /* 43 = 40-byte + 2 '/' + terminating NUL */
 245        int pfxlen = len;
 246        int entlen = pfxlen + 43;
 247        int base_len = -1;
 248
 249        if (!is_absolute_path(entry) && relative_base) {
 250                /* Relative alt-odb */
 251                if (base_len < 0)
 252                        base_len = strlen(relative_base) + 1;
 253                entlen += base_len;
 254                pfxlen += base_len;
 255        }
 256        ent = xmalloc(sizeof(*ent) + entlen);
 257
 258        if (!is_absolute_path(entry) && relative_base) {
 259                memcpy(ent->base, relative_base, base_len - 1);
 260                ent->base[base_len - 1] = '/';
 261                memcpy(ent->base + base_len, entry, len);
 262        }
 263        else
 264                memcpy(ent->base, entry, pfxlen);
 265
 266        ent->name = ent->base + pfxlen + 1;
 267        ent->base[pfxlen + 3] = '/';
 268        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 269
 270        /* Detect cases where alternate disappeared */
 271        if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
 272                error("object directory %s does not exist; "
 273                      "check .git/objects/info/alternates.",
 274                      ent->base);
 275                free(ent);
 276                return -1;
 277        }
 278
 279        /* Prevent the common mistake of listing the same
 280         * thing twice, or object directory itself.
 281         */
 282        for (alt = alt_odb_list; alt; alt = alt->next) {
 283                if (!memcmp(ent->base, alt->base, pfxlen)) {
 284                        free(ent);
 285                        return -1;
 286                }
 287        }
 288        if (!memcmp(ent->base, objdir, pfxlen)) {
 289                free(ent);
 290                return -1;
 291        }
 292
 293        /* add the alternate entry */
 294        *alt_odb_tail = ent;
 295        alt_odb_tail = &(ent->next);
 296        ent->next = NULL;
 297
 298        /* recursively add alternates */
 299        read_info_alternates(ent->base, depth + 1);
 300
 301        ent->base[pfxlen] = '/';
 302
 303        return 0;
 304}
 305
 306static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 307                                 const char *relative_base, int depth)
 308{
 309        const char *cp, *last;
 310
 311        if (depth > 5) {
 312                error("%s: ignoring alternate object stores, nesting too deep.",
 313                                relative_base);
 314                return;
 315        }
 316
 317        last = alt;
 318        while (last < ep) {
 319                cp = last;
 320                if (cp < ep && *cp == '#') {
 321                        while (cp < ep && *cp != sep)
 322                                cp++;
 323                        last = cp + 1;
 324                        continue;
 325                }
 326                while (cp < ep && *cp != sep)
 327                        cp++;
 328                if (last != cp) {
 329                        if (!is_absolute_path(last) && depth) {
 330                                error("%s: ignoring relative alternate object store %s",
 331                                                relative_base, last);
 332                        } else {
 333                                link_alt_odb_entry(last, cp - last,
 334                                                relative_base, depth);
 335                        }
 336                }
 337                while (cp < ep && *cp == sep)
 338                        cp++;
 339                last = cp;
 340        }
 341}
 342
 343static void read_info_alternates(const char * relative_base, int depth)
 344{
 345        char *map;
 346        size_t mapsz;
 347        struct stat st;
 348        const char alt_file_name[] = "info/alternates";
 349        /* Given that relative_base is no longer than PATH_MAX,
 350           ensure that "path" has enough space to append "/", the
 351           file name, "info/alternates", and a trailing NUL.  */
 352        char path[PATH_MAX + 1 + sizeof alt_file_name];
 353        int fd;
 354
 355        sprintf(path, "%s/%s", relative_base, alt_file_name);
 356        fd = open(path, O_RDONLY);
 357        if (fd < 0)
 358                return;
 359        if (fstat(fd, &st) || (st.st_size == 0)) {
 360                close(fd);
 361                return;
 362        }
 363        mapsz = xsize_t(st.st_size);
 364        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 365        close(fd);
 366
 367        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 368
 369        munmap(map, mapsz);
 370}
 371
 372void add_to_alternates_file(const char *reference)
 373{
 374        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 375        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
 376        char *alt = mkpath("%s/objects\n", reference);
 377        write_or_die(fd, alt, strlen(alt));
 378        if (commit_lock_file(lock))
 379                die("could not close alternates file");
 380        if (alt_odb_tail)
 381                link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
 382}
 383
 384void prepare_alt_odb(void)
 385{
 386        const char *alt;
 387
 388        if (alt_odb_tail)
 389                return;
 390
 391        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 392        if (!alt) alt = "";
 393
 394        alt_odb_tail = &alt_odb_list;
 395        link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
 396
 397        read_info_alternates(get_object_directory(), 0);
 398}
 399
 400static int has_loose_object(const unsigned char *sha1)
 401{
 402        char *name = sha1_file_name(sha1);
 403        struct alternate_object_database *alt;
 404
 405        if (!access(name, F_OK))
 406                return 1;
 407        prepare_alt_odb();
 408        for (alt = alt_odb_list; alt; alt = alt->next) {
 409                name = alt->name;
 410                fill_sha1_path(name, sha1);
 411                if (!access(alt->base, F_OK))
 412                        return 1;
 413        }
 414        return 0;
 415}
 416
 417static unsigned int pack_used_ctr;
 418static unsigned int pack_mmap_calls;
 419static unsigned int peak_pack_open_windows;
 420static unsigned int pack_open_windows;
 421static size_t peak_pack_mapped;
 422static size_t pack_mapped;
 423struct packed_git *packed_git;
 424
 425void pack_report(void)
 426{
 427        fprintf(stderr,
 428                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 429                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 430                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 431                sz_fmt(getpagesize()),
 432                sz_fmt(packed_git_window_size),
 433                sz_fmt(packed_git_limit));
 434        fprintf(stderr,
 435                "pack_report: pack_used_ctr            = %10u\n"
 436                "pack_report: pack_mmap_calls          = %10u\n"
 437                "pack_report: pack_open_windows        = %10u / %10u\n"
 438                "pack_report: pack_mapped              = "
 439                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 440                pack_used_ctr,
 441                pack_mmap_calls,
 442                pack_open_windows, peak_pack_open_windows,
 443                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 444}
 445
 446static int check_packed_git_idx(const char *path,  struct packed_git *p)
 447{
 448        void *idx_map;
 449        struct pack_idx_header *hdr;
 450        size_t idx_size;
 451        uint32_t version, nr, i, *index;
 452        int fd = open(path, O_RDONLY);
 453        struct stat st;
 454
 455        if (fd < 0)
 456                return -1;
 457        if (fstat(fd, &st)) {
 458                close(fd);
 459                return -1;
 460        }
 461        idx_size = xsize_t(st.st_size);
 462        if (idx_size < 4 * 256 + 20 + 20) {
 463                close(fd);
 464                return error("index file %s is too small", path);
 465        }
 466        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 467        close(fd);
 468
 469        hdr = idx_map;
 470        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 471                version = ntohl(hdr->idx_version);
 472                if (version < 2 || version > 2) {
 473                        munmap(idx_map, idx_size);
 474                        return error("index file %s is version %d"
 475                                     " and is not supported by this binary"
 476                                     " (try upgrading GIT to a newer version)",
 477                                     path, version);
 478                }
 479        } else
 480                version = 1;
 481
 482        nr = 0;
 483        index = idx_map;
 484        if (version > 1)
 485                index += 2;  /* skip index header */
 486        for (i = 0; i < 256; i++) {
 487                uint32_t n = ntohl(index[i]);
 488                if (n < nr) {
 489                        munmap(idx_map, idx_size);
 490                        return error("non-monotonic index %s", path);
 491                }
 492                nr = n;
 493        }
 494
 495        if (version == 1) {
 496                /*
 497                 * Total size:
 498                 *  - 256 index entries 4 bytes each
 499                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 500                 *  - 20-byte SHA1 of the packfile
 501                 *  - 20-byte SHA1 file checksum
 502                 */
 503                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 504                        munmap(idx_map, idx_size);
 505                        return error("wrong index v1 file size in %s", path);
 506                }
 507        } else if (version == 2) {
 508                /*
 509                 * Minimum size:
 510                 *  - 8 bytes of header
 511                 *  - 256 index entries 4 bytes each
 512                 *  - 20-byte sha1 entry * nr
 513                 *  - 4-byte crc entry * nr
 514                 *  - 4-byte offset entry * nr
 515                 *  - 20-byte SHA1 of the packfile
 516                 *  - 20-byte SHA1 file checksum
 517                 * And after the 4-byte offset table might be a
 518                 * variable sized table containing 8-byte entries
 519                 * for offsets larger than 2^31.
 520                 */
 521                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 522                unsigned long max_size = min_size;
 523                if (nr)
 524                        max_size += (nr - 1)*8;
 525                if (idx_size < min_size || idx_size > max_size) {
 526                        munmap(idx_map, idx_size);
 527                        return error("wrong index v2 file size in %s", path);
 528                }
 529                if (idx_size != min_size &&
 530                    /*
 531                     * make sure we can deal with large pack offsets.
 532                     * 31-bit signed offset won't be enough, neither
 533                     * 32-bit unsigned one will be.
 534                     */
 535                    (sizeof(off_t) <= 4)) {
 536                        munmap(idx_map, idx_size);
 537                        return error("pack too large for current definition of off_t in %s", path);
 538                }
 539        }
 540
 541        p->index_version = version;
 542        p->index_data = idx_map;
 543        p->index_size = idx_size;
 544        p->num_objects = nr;
 545        return 0;
 546}
 547
 548int open_pack_index(struct packed_git *p)
 549{
 550        char *idx_name;
 551        int ret;
 552
 553        if (p->index_data)
 554                return 0;
 555
 556        idx_name = xstrdup(p->pack_name);
 557        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 558        ret = check_packed_git_idx(idx_name, p);
 559        free(idx_name);
 560        return ret;
 561}
 562
 563static void scan_windows(struct packed_git *p,
 564        struct packed_git **lru_p,
 565        struct pack_window **lru_w,
 566        struct pack_window **lru_l)
 567{
 568        struct pack_window *w, *w_l;
 569
 570        for (w_l = NULL, w = p->windows; w; w = w->next) {
 571                if (!w->inuse_cnt) {
 572                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 573                                *lru_p = p;
 574                                *lru_w = w;
 575                                *lru_l = w_l;
 576                        }
 577                }
 578                w_l = w;
 579        }
 580}
 581
 582static int unuse_one_window(struct packed_git *current, int keep_fd)
 583{
 584        struct packed_git *p, *lru_p = NULL;
 585        struct pack_window *lru_w = NULL, *lru_l = NULL;
 586
 587        if (current)
 588                scan_windows(current, &lru_p, &lru_w, &lru_l);
 589        for (p = packed_git; p; p = p->next)
 590                scan_windows(p, &lru_p, &lru_w, &lru_l);
 591        if (lru_p) {
 592                munmap(lru_w->base, lru_w->len);
 593                pack_mapped -= lru_w->len;
 594                if (lru_l)
 595                        lru_l->next = lru_w->next;
 596                else {
 597                        lru_p->windows = lru_w->next;
 598                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 599                                close(lru_p->pack_fd);
 600                                lru_p->pack_fd = -1;
 601                        }
 602                }
 603                free(lru_w);
 604                pack_open_windows--;
 605                return 1;
 606        }
 607        return 0;
 608}
 609
 610void release_pack_memory(size_t need, int fd)
 611{
 612        size_t cur = pack_mapped;
 613        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 614                ; /* nothing */
 615}
 616
 617void close_pack_windows(struct packed_git *p)
 618{
 619        while (p->windows) {
 620                struct pack_window *w = p->windows;
 621
 622                if (w->inuse_cnt)
 623                        die("pack '%s' still has open windows to it",
 624                            p->pack_name);
 625                munmap(w->base, w->len);
 626                pack_mapped -= w->len;
 627                pack_open_windows--;
 628                p->windows = w->next;
 629                free(w);
 630        }
 631}
 632
 633void unuse_pack(struct pack_window **w_cursor)
 634{
 635        struct pack_window *w = *w_cursor;
 636        if (w) {
 637                w->inuse_cnt--;
 638                *w_cursor = NULL;
 639        }
 640}
 641
 642/*
 643 * Do not call this directly as this leaks p->pack_fd on error return;
 644 * call open_packed_git() instead.
 645 */
 646static int open_packed_git_1(struct packed_git *p)
 647{
 648        struct stat st;
 649        struct pack_header hdr;
 650        unsigned char sha1[20];
 651        unsigned char *idx_sha1;
 652        long fd_flag;
 653
 654        if (!p->index_data && open_pack_index(p))
 655                return error("packfile %s index unavailable", p->pack_name);
 656
 657        p->pack_fd = open(p->pack_name, O_RDONLY);
 658        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 659                return -1;
 660
 661        /* If we created the struct before we had the pack we lack size. */
 662        if (!p->pack_size) {
 663                if (!S_ISREG(st.st_mode))
 664                        return error("packfile %s not a regular file", p->pack_name);
 665                p->pack_size = st.st_size;
 666        } else if (p->pack_size != st.st_size)
 667                return error("packfile %s size changed", p->pack_name);
 668
 669        /* We leave these file descriptors open with sliding mmap;
 670         * there is no point keeping them open across exec(), though.
 671         */
 672        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 673        if (fd_flag < 0)
 674                return error("cannot determine file descriptor flags");
 675        fd_flag |= FD_CLOEXEC;
 676        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 677                return error("cannot set FD_CLOEXEC");
 678
 679        /* Verify we recognize this pack file format. */
 680        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 681                return error("file %s is far too short to be a packfile", p->pack_name);
 682        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 683                return error("file %s is not a GIT packfile", p->pack_name);
 684        if (!pack_version_ok(hdr.hdr_version))
 685                return error("packfile %s is version %u and not supported"
 686                        " (try upgrading GIT to a newer version)",
 687                        p->pack_name, ntohl(hdr.hdr_version));
 688
 689        /* Verify the pack matches its index. */
 690        if (p->num_objects != ntohl(hdr.hdr_entries))
 691                return error("packfile %s claims to have %u objects"
 692                             " while index indicates %u objects",
 693                             p->pack_name, ntohl(hdr.hdr_entries),
 694                             p->num_objects);
 695        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 696                return error("end of packfile %s is unavailable", p->pack_name);
 697        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 698                return error("packfile %s signature is unavailable", p->pack_name);
 699        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 700        if (hashcmp(sha1, idx_sha1))
 701                return error("packfile %s does not match index", p->pack_name);
 702        return 0;
 703}
 704
 705static int open_packed_git(struct packed_git *p)
 706{
 707        if (!open_packed_git_1(p))
 708                return 0;
 709        if (p->pack_fd != -1) {
 710                close(p->pack_fd);
 711                p->pack_fd = -1;
 712        }
 713        return -1;
 714}
 715
 716static int in_window(struct pack_window *win, off_t offset)
 717{
 718        /* We must promise at least 20 bytes (one hash) after the
 719         * offset is available from this window, otherwise the offset
 720         * is not actually in this window and a different window (which
 721         * has that one hash excess) must be used.  This is to support
 722         * the object header and delta base parsing routines below.
 723         */
 724        off_t win_off = win->offset;
 725        return win_off <= offset
 726                && (offset + 20) <= (win_off + win->len);
 727}
 728
 729unsigned char* use_pack(struct packed_git *p,
 730                struct pack_window **w_cursor,
 731                off_t offset,
 732                unsigned int *left)
 733{
 734        struct pack_window *win = *w_cursor;
 735
 736        if (p->pack_fd == -1 && open_packed_git(p))
 737                die("packfile %s cannot be accessed", p->pack_name);
 738
 739        /* Since packfiles end in a hash of their content and its
 740         * pointless to ask for an offset into the middle of that
 741         * hash, and the in_window function above wouldn't match
 742         * don't allow an offset too close to the end of the file.
 743         */
 744        if (offset > (p->pack_size - 20))
 745                die("offset beyond end of packfile (truncated pack?)");
 746
 747        if (!win || !in_window(win, offset)) {
 748                if (win)
 749                        win->inuse_cnt--;
 750                for (win = p->windows; win; win = win->next) {
 751                        if (in_window(win, offset))
 752                                break;
 753                }
 754                if (!win) {
 755                        size_t window_align = packed_git_window_size / 2;
 756                        off_t len;
 757                        win = xcalloc(1, sizeof(*win));
 758                        win->offset = (offset / window_align) * window_align;
 759                        len = p->pack_size - win->offset;
 760                        if (len > packed_git_window_size)
 761                                len = packed_git_window_size;
 762                        win->len = (size_t)len;
 763                        pack_mapped += win->len;
 764                        while (packed_git_limit < pack_mapped
 765                                && unuse_one_window(p, p->pack_fd))
 766                                ; /* nothing */
 767                        win->base = xmmap(NULL, win->len,
 768                                PROT_READ, MAP_PRIVATE,
 769                                p->pack_fd, win->offset);
 770                        if (win->base == MAP_FAILED)
 771                                die("packfile %s cannot be mapped: %s",
 772                                        p->pack_name,
 773                                        strerror(errno));
 774                        pack_mmap_calls++;
 775                        pack_open_windows++;
 776                        if (pack_mapped > peak_pack_mapped)
 777                                peak_pack_mapped = pack_mapped;
 778                        if (pack_open_windows > peak_pack_open_windows)
 779                                peak_pack_open_windows = pack_open_windows;
 780                        win->next = p->windows;
 781                        p->windows = win;
 782                }
 783        }
 784        if (win != *w_cursor) {
 785                win->last_used = pack_used_ctr++;
 786                win->inuse_cnt++;
 787                *w_cursor = win;
 788        }
 789        offset -= win->offset;
 790        if (left)
 791                *left = win->len - xsize_t(offset);
 792        return win->base + offset;
 793}
 794
 795struct packed_git *add_packed_git(const char *path, int path_len, int local)
 796{
 797        struct stat st;
 798        struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
 799
 800        /*
 801         * Make sure a corresponding .pack file exists and that
 802         * the index looks sane.
 803         */
 804        path_len -= strlen(".idx");
 805        if (path_len < 1)
 806                return NULL;
 807        memcpy(p->pack_name, path, path_len);
 808        strcpy(p->pack_name + path_len, ".pack");
 809        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 810                free(p);
 811                return NULL;
 812        }
 813
 814        /* ok, it looks sane as far as we can check without
 815         * actually mapping the pack file.
 816         */
 817        p->index_version = 0;
 818        p->index_data = NULL;
 819        p->index_size = 0;
 820        p->num_objects = 0;
 821        p->pack_size = st.st_size;
 822        p->next = NULL;
 823        p->windows = NULL;
 824        p->pack_fd = -1;
 825        p->pack_local = local;
 826        p->mtime = st.st_mtime;
 827        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 828                hashclr(p->sha1);
 829        return p;
 830}
 831
 832struct packed_git *parse_pack_index(unsigned char *sha1)
 833{
 834        const char *idx_path = sha1_pack_index_name(sha1);
 835        const char *path = sha1_pack_name(sha1);
 836        struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
 837
 838        if (check_packed_git_idx(idx_path, p)) {
 839                free(p);
 840                return NULL;
 841        }
 842
 843        strcpy(p->pack_name, path);
 844        p->pack_size = 0;
 845        p->next = NULL;
 846        p->windows = NULL;
 847        p->pack_fd = -1;
 848        hashcpy(p->sha1, sha1);
 849        return p;
 850}
 851
 852void install_packed_git(struct packed_git *pack)
 853{
 854        pack->next = packed_git;
 855        packed_git = pack;
 856}
 857
 858static void prepare_packed_git_one(char *objdir, int local)
 859{
 860        /* Ensure that this buffer is large enough so that we can
 861           append "/pack/" without clobbering the stack even if
 862           strlen(objdir) were PATH_MAX.  */
 863        char path[PATH_MAX + 1 + 4 + 1 + 1];
 864        int len;
 865        DIR *dir;
 866        struct dirent *de;
 867
 868        sprintf(path, "%s/pack", objdir);
 869        len = strlen(path);
 870        dir = opendir(path);
 871        if (!dir) {
 872                if (errno != ENOENT)
 873                        error("unable to open object pack directory: %s: %s",
 874                              path, strerror(errno));
 875                return;
 876        }
 877        path[len++] = '/';
 878        while ((de = readdir(dir)) != NULL) {
 879                int namelen = strlen(de->d_name);
 880                struct packed_git *p;
 881
 882                if (!has_extension(de->d_name, ".idx"))
 883                        continue;
 884
 885                if (len + namelen + 1 > sizeof(path))
 886                        continue;
 887
 888                /* Don't reopen a pack we already have. */
 889                strcpy(path + len, de->d_name);
 890                for (p = packed_git; p; p = p->next) {
 891                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 892                                break;
 893                }
 894                if (p)
 895                        continue;
 896                /* See if it really is a valid .idx file with corresponding
 897                 * .pack file that we can map.
 898                 */
 899                p = add_packed_git(path, len + namelen, local);
 900                if (!p)
 901                        continue;
 902                install_packed_git(p);
 903        }
 904        closedir(dir);
 905}
 906
 907static int sort_pack(const void *a_, const void *b_)
 908{
 909        struct packed_git *a = *((struct packed_git **)a_);
 910        struct packed_git *b = *((struct packed_git **)b_);
 911        int st;
 912
 913        /*
 914         * Local packs tend to contain objects specific to our
 915         * variant of the project than remote ones.  In addition,
 916         * remote ones could be on a network mounted filesystem.
 917         * Favor local ones for these reasons.
 918         */
 919        st = a->pack_local - b->pack_local;
 920        if (st)
 921                return -st;
 922
 923        /*
 924         * Younger packs tend to contain more recent objects,
 925         * and more recent objects tend to get accessed more
 926         * often.
 927         */
 928        if (a->mtime < b->mtime)
 929                return 1;
 930        else if (a->mtime == b->mtime)
 931                return 0;
 932        return -1;
 933}
 934
 935static void rearrange_packed_git(void)
 936{
 937        struct packed_git **ary, *p;
 938        int i, n;
 939
 940        for (n = 0, p = packed_git; p; p = p->next)
 941                n++;
 942        if (n < 2)
 943                return;
 944
 945        /* prepare an array of packed_git for easier sorting */
 946        ary = xcalloc(n, sizeof(struct packed_git *));
 947        for (n = 0, p = packed_git; p; p = p->next)
 948                ary[n++] = p;
 949
 950        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 951
 952        /* link them back again */
 953        for (i = 0; i < n - 1; i++)
 954                ary[i]->next = ary[i + 1];
 955        ary[n - 1]->next = NULL;
 956        packed_git = ary[0];
 957
 958        free(ary);
 959}
 960
 961static int prepare_packed_git_run_once = 0;
 962void prepare_packed_git(void)
 963{
 964        struct alternate_object_database *alt;
 965
 966        if (prepare_packed_git_run_once)
 967                return;
 968        prepare_packed_git_one(get_object_directory(), 1);
 969        prepare_alt_odb();
 970        for (alt = alt_odb_list; alt; alt = alt->next) {
 971                alt->name[-1] = 0;
 972                prepare_packed_git_one(alt->base, 0);
 973                alt->name[-1] = '/';
 974        }
 975        rearrange_packed_git();
 976        prepare_packed_git_run_once = 1;
 977}
 978
 979void reprepare_packed_git(void)
 980{
 981        prepare_packed_git_run_once = 0;
 982        prepare_packed_git();
 983}
 984
 985int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 986{
 987        unsigned char real_sha1[20];
 988        hash_sha1_file(map, size, type, real_sha1);
 989        return hashcmp(sha1, real_sha1) ? -1 : 0;
 990}
 991
 992static int git_open_noatime(const char *name)
 993{
 994        static int sha1_file_open_flag = O_NOATIME;
 995        int fd = open(name, O_RDONLY | sha1_file_open_flag);
 996
 997        /* Might the failure be due to O_NOATIME? */
 998        if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
 999                fd = open(name, O_RDONLY);
1000                if (fd >= 0)
1001                        sha1_file_open_flag = 0;
1002        }
1003        return fd;
1004}
1005
1006static int open_sha1_file(const unsigned char *sha1)
1007{
1008        int fd;
1009        char *name = sha1_file_name(sha1);
1010        struct alternate_object_database *alt;
1011
1012        fd = git_open_noatime(name);
1013        if (fd >= 0)
1014                return fd;
1015
1016        prepare_alt_odb();
1017        errno = ENOENT;
1018        for (alt = alt_odb_list; alt; alt = alt->next) {
1019                name = alt->name;
1020                fill_sha1_path(name, sha1);
1021                fd = git_open_noatime(alt->base);
1022                if (fd >= 0)
1023                        return fd;
1024        }
1025        return -1;
1026}
1027
1028static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1029{
1030        void *map;
1031        int fd;
1032
1033        fd = open_sha1_file(sha1);
1034        map = NULL;
1035        if (fd >= 0) {
1036                struct stat st;
1037
1038                if (!fstat(fd, &st)) {
1039                        *size = xsize_t(st.st_size);
1040                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1041                }
1042                close(fd);
1043        }
1044        return map;
1045}
1046
1047static int legacy_loose_object(unsigned char *map)
1048{
1049        unsigned int word;
1050
1051        /*
1052         * Is it a zlib-compressed buffer? If so, the first byte
1053         * must be 0x78 (15-bit window size, deflated), and the
1054         * first 16-bit word is evenly divisible by 31
1055         */
1056        word = (map[0] << 8) + map[1];
1057        if (map[0] == 0x78 && !(word % 31))
1058                return 1;
1059        else
1060                return 0;
1061}
1062
1063unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1064{
1065        unsigned shift;
1066        unsigned char c;
1067        unsigned long size;
1068        unsigned long used = 0;
1069
1070        c = buf[used++];
1071        *type = (c >> 4) & 7;
1072        size = c & 15;
1073        shift = 4;
1074        while (c & 0x80) {
1075                if (len <= used)
1076                        return 0;
1077                if (sizeof(long) * 8 <= shift)
1078                        return 0;
1079                c = buf[used++];
1080                size += (c & 0x7f) << shift;
1081                shift += 7;
1082        }
1083        *sizep = size;
1084        return used;
1085}
1086
1087static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1088{
1089        unsigned long size, used;
1090        static const char valid_loose_object_type[8] = {
1091                0, /* OBJ_EXT */
1092                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1093                0, /* "delta" and others are invalid in a loose object */
1094        };
1095        enum object_type type;
1096
1097        /* Get the data stream */
1098        memset(stream, 0, sizeof(*stream));
1099        stream->next_in = map;
1100        stream->avail_in = mapsize;
1101        stream->next_out = buffer;
1102        stream->avail_out = bufsiz;
1103
1104        if (legacy_loose_object(map)) {
1105                inflateInit(stream);
1106                return inflate(stream, 0);
1107        }
1108
1109
1110        /*
1111         * There used to be a second loose object header format which
1112         * was meant to mimic the in-pack format, allowing for direct
1113         * copy of the object data.  This format turned up not to be
1114         * really worth it and we don't write it any longer.  But we
1115         * can still read it.
1116         */
1117        used = unpack_object_header_gently(map, mapsize, &type, &size);
1118        if (!used || !valid_loose_object_type[type])
1119                return -1;
1120        map += used;
1121        mapsize -= used;
1122
1123        /* Set up the stream for the rest.. */
1124        stream->next_in = map;
1125        stream->avail_in = mapsize;
1126        inflateInit(stream);
1127
1128        /* And generate the fake traditional header */
1129        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1130                                         typename(type), size);
1131        return 0;
1132}
1133
1134static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1135{
1136        int bytes = strlen(buffer) + 1;
1137        unsigned char *buf = xmalloc(1+size);
1138        unsigned long n;
1139        int status = Z_OK;
1140
1141        n = stream->total_out - bytes;
1142        if (n > size)
1143                n = size;
1144        memcpy(buf, (char *) buffer + bytes, n);
1145        bytes = n;
1146        if (bytes <= size) {
1147                /*
1148                 * The above condition must be (bytes <= size), not
1149                 * (bytes < size).  In other words, even though we
1150                 * expect no more output and set avail_out to zer0,
1151                 * the input zlib stream may have bytes that express
1152                 * "this concludes the stream", and we *do* want to
1153                 * eat that input.
1154                 *
1155                 * Otherwise we would not be able to test that we
1156                 * consumed all the input to reach the expected size;
1157                 * we also want to check that zlib tells us that all
1158                 * went well with status == Z_STREAM_END at the end.
1159                 */
1160                stream->next_out = buf + bytes;
1161                stream->avail_out = size - bytes;
1162                while (status == Z_OK)
1163                        status = inflate(stream, Z_FINISH);
1164        }
1165        buf[size] = 0;
1166        if (status == Z_STREAM_END && !stream->avail_in) {
1167                inflateEnd(stream);
1168                return buf;
1169        }
1170
1171        if (status < 0)
1172                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1173        else if (stream->avail_in)
1174                error("garbage at end of loose object '%s'",
1175                      sha1_to_hex(sha1));
1176        free(buf);
1177        return NULL;
1178}
1179
1180/*
1181 * We used to just use "sscanf()", but that's actually way
1182 * too permissive for what we want to check. So do an anal
1183 * object header parse by hand.
1184 */
1185static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1186{
1187        char type[10];
1188        int i;
1189        unsigned long size;
1190
1191        /*
1192         * The type can be at most ten bytes (including the
1193         * terminating '\0' that we add), and is followed by
1194         * a space.
1195         */
1196        i = 0;
1197        for (;;) {
1198                char c = *hdr++;
1199                if (c == ' ')
1200                        break;
1201                type[i++] = c;
1202                if (i >= sizeof(type))
1203                        return -1;
1204        }
1205        type[i] = 0;
1206
1207        /*
1208         * The length must follow immediately, and be in canonical
1209         * decimal format (ie "010" is not valid).
1210         */
1211        size = *hdr++ - '0';
1212        if (size > 9)
1213                return -1;
1214        if (size) {
1215                for (;;) {
1216                        unsigned long c = *hdr - '0';
1217                        if (c > 9)
1218                                break;
1219                        hdr++;
1220                        size = size * 10 + c;
1221                }
1222        }
1223        *sizep = size;
1224
1225        /*
1226         * The length must be followed by a zero byte
1227         */
1228        return *hdr ? -1 : type_from_string(type);
1229}
1230
1231static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1232{
1233        int ret;
1234        z_stream stream;
1235        char hdr[8192];
1236
1237        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1238        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1239                return NULL;
1240
1241        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1242}
1243
1244unsigned long get_size_from_delta(struct packed_git *p,
1245                                  struct pack_window **w_curs,
1246                                  off_t curpos)
1247{
1248        const unsigned char *data;
1249        unsigned char delta_head[20], *in;
1250        z_stream stream;
1251        int st;
1252
1253        memset(&stream, 0, sizeof(stream));
1254        stream.next_out = delta_head;
1255        stream.avail_out = sizeof(delta_head);
1256
1257        inflateInit(&stream);
1258        do {
1259                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1260                stream.next_in = in;
1261                st = inflate(&stream, Z_FINISH);
1262                curpos += stream.next_in - in;
1263        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1264                 stream.total_out < sizeof(delta_head));
1265        inflateEnd(&stream);
1266        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1267                die("delta data unpack-initial failed");
1268
1269        /* Examine the initial part of the delta to figure out
1270         * the result size.
1271         */
1272        data = delta_head;
1273
1274        /* ignore base size */
1275        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1276
1277        /* Read the result size */
1278        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1279}
1280
1281static off_t get_delta_base(struct packed_git *p,
1282                                    struct pack_window **w_curs,
1283                                    off_t *curpos,
1284                                    enum object_type type,
1285                                    off_t delta_obj_offset)
1286{
1287        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1288        off_t base_offset;
1289
1290        /* use_pack() assured us we have [base_info, base_info + 20)
1291         * as a range that we can look at without walking off the
1292         * end of the mapped window.  Its actually the hash size
1293         * that is assured.  An OFS_DELTA longer than the hash size
1294         * is stupid, as then a REF_DELTA would be smaller to store.
1295         */
1296        if (type == OBJ_OFS_DELTA) {
1297                unsigned used = 0;
1298                unsigned char c = base_info[used++];
1299                base_offset = c & 127;
1300                while (c & 128) {
1301                        base_offset += 1;
1302                        if (!base_offset || MSB(base_offset, 7))
1303                                die("offset value overflow for delta base object");
1304                        c = base_info[used++];
1305                        base_offset = (base_offset << 7) + (c & 127);
1306                }
1307                base_offset = delta_obj_offset - base_offset;
1308                if (base_offset >= delta_obj_offset)
1309                        die("delta base offset out of bound");
1310                *curpos += used;
1311        } else if (type == OBJ_REF_DELTA) {
1312                /* The base entry _must_ be in the same pack */
1313                base_offset = find_pack_entry_one(base_info, p);
1314                if (!base_offset)
1315                        die("failed to find delta-pack base object %s",
1316                                sha1_to_hex(base_info));
1317                *curpos += 20;
1318        } else
1319                die("I am totally screwed");
1320        return base_offset;
1321}
1322
1323/* forward declaration for a mutually recursive function */
1324static int packed_object_info(struct packed_git *p, off_t offset,
1325                              unsigned long *sizep);
1326
1327static int packed_delta_info(struct packed_git *p,
1328                             struct pack_window **w_curs,
1329                             off_t curpos,
1330                             enum object_type type,
1331                             off_t obj_offset,
1332                             unsigned long *sizep)
1333{
1334        off_t base_offset;
1335
1336        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1337        type = packed_object_info(p, base_offset, NULL);
1338
1339        /* We choose to only get the type of the base object and
1340         * ignore potentially corrupt pack file that expects the delta
1341         * based on a base with a wrong size.  This saves tons of
1342         * inflate() calls.
1343         */
1344        if (sizep)
1345                *sizep = get_size_from_delta(p, w_curs, curpos);
1346
1347        return type;
1348}
1349
1350static int unpack_object_header(struct packed_git *p,
1351                                struct pack_window **w_curs,
1352                                off_t *curpos,
1353                                unsigned long *sizep)
1354{
1355        unsigned char *base;
1356        unsigned int left;
1357        unsigned long used;
1358        enum object_type type;
1359
1360        /* use_pack() assures us we have [base, base + 20) available
1361         * as a range that we can look at at.  (Its actually the hash
1362         * size that is assured.)  With our object header encoding
1363         * the maximum deflated object size is 2^137, which is just
1364         * insane, so we know won't exceed what we have been given.
1365         */
1366        base = use_pack(p, w_curs, *curpos, &left);
1367        used = unpack_object_header_gently(base, left, &type, sizep);
1368        if (!used)
1369                die("object offset outside of pack file");
1370        *curpos += used;
1371
1372        return type;
1373}
1374
1375const char *packed_object_info_detail(struct packed_git *p,
1376                                      off_t obj_offset,
1377                                      unsigned long *size,
1378                                      unsigned long *store_size,
1379                                      unsigned int *delta_chain_length,
1380                                      unsigned char *base_sha1)
1381{
1382        struct pack_window *w_curs = NULL;
1383        off_t curpos;
1384        unsigned long dummy;
1385        unsigned char *next_sha1;
1386        enum object_type type;
1387        struct revindex_entry *revidx;
1388
1389        *delta_chain_length = 0;
1390        curpos = obj_offset;
1391        type = unpack_object_header(p, &w_curs, &curpos, size);
1392
1393        revidx = find_pack_revindex(p, obj_offset);
1394        *store_size = revidx[1].offset - obj_offset;
1395
1396        for (;;) {
1397                switch (type) {
1398                default:
1399                        die("pack %s contains unknown object type %d",
1400                            p->pack_name, type);
1401                case OBJ_COMMIT:
1402                case OBJ_TREE:
1403                case OBJ_BLOB:
1404                case OBJ_TAG:
1405                        unuse_pack(&w_curs);
1406                        return typename(type);
1407                case OBJ_OFS_DELTA:
1408                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1409                        if (*delta_chain_length == 0) {
1410                                revidx = find_pack_revindex(p, obj_offset);
1411                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1412                        }
1413                        break;
1414                case OBJ_REF_DELTA:
1415                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1416                        if (*delta_chain_length == 0)
1417                                hashcpy(base_sha1, next_sha1);
1418                        obj_offset = find_pack_entry_one(next_sha1, p);
1419                        break;
1420                }
1421                (*delta_chain_length)++;
1422                curpos = obj_offset;
1423                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1424        }
1425}
1426
1427static int packed_object_info(struct packed_git *p, off_t obj_offset,
1428                              unsigned long *sizep)
1429{
1430        struct pack_window *w_curs = NULL;
1431        unsigned long size;
1432        off_t curpos = obj_offset;
1433        enum object_type type;
1434
1435        type = unpack_object_header(p, &w_curs, &curpos, &size);
1436
1437        switch (type) {
1438        case OBJ_OFS_DELTA:
1439        case OBJ_REF_DELTA:
1440                type = packed_delta_info(p, &w_curs, curpos,
1441                                         type, obj_offset, sizep);
1442                break;
1443        case OBJ_COMMIT:
1444        case OBJ_TREE:
1445        case OBJ_BLOB:
1446        case OBJ_TAG:
1447                if (sizep)
1448                        *sizep = size;
1449                break;
1450        default:
1451                die("pack %s contains unknown object type %d",
1452                    p->pack_name, type);
1453        }
1454        unuse_pack(&w_curs);
1455        return type;
1456}
1457
1458static void *unpack_compressed_entry(struct packed_git *p,
1459                                    struct pack_window **w_curs,
1460                                    off_t curpos,
1461                                    unsigned long size)
1462{
1463        int st;
1464        z_stream stream;
1465        unsigned char *buffer, *in;
1466
1467        buffer = xmalloc(size + 1);
1468        buffer[size] = 0;
1469        memset(&stream, 0, sizeof(stream));
1470        stream.next_out = buffer;
1471        stream.avail_out = size;
1472
1473        inflateInit(&stream);
1474        do {
1475                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1476                stream.next_in = in;
1477                st = inflate(&stream, Z_FINISH);
1478                curpos += stream.next_in - in;
1479        } while (st == Z_OK || st == Z_BUF_ERROR);
1480        inflateEnd(&stream);
1481        if ((st != Z_STREAM_END) || stream.total_out != size) {
1482                free(buffer);
1483                return NULL;
1484        }
1485
1486        return buffer;
1487}
1488
1489#define MAX_DELTA_CACHE (256)
1490
1491static size_t delta_base_cached;
1492
1493static struct delta_base_cache_lru_list {
1494        struct delta_base_cache_lru_list *prev;
1495        struct delta_base_cache_lru_list *next;
1496} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1497
1498static struct delta_base_cache_entry {
1499        struct delta_base_cache_lru_list lru;
1500        void *data;
1501        struct packed_git *p;
1502        off_t base_offset;
1503        unsigned long size;
1504        enum object_type type;
1505} delta_base_cache[MAX_DELTA_CACHE];
1506
1507static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1508{
1509        unsigned long hash;
1510
1511        hash = (unsigned long)p + (unsigned long)base_offset;
1512        hash += (hash >> 8) + (hash >> 16);
1513        return hash % MAX_DELTA_CACHE;
1514}
1515
1516static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1517        unsigned long *base_size, enum object_type *type, int keep_cache)
1518{
1519        void *ret;
1520        unsigned long hash = pack_entry_hash(p, base_offset);
1521        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1522
1523        ret = ent->data;
1524        if (ret && ent->p == p && ent->base_offset == base_offset)
1525                goto found_cache_entry;
1526        return unpack_entry(p, base_offset, type, base_size);
1527
1528found_cache_entry:
1529        if (!keep_cache) {
1530                ent->data = NULL;
1531                ent->lru.next->prev = ent->lru.prev;
1532                ent->lru.prev->next = ent->lru.next;
1533                delta_base_cached -= ent->size;
1534        } else {
1535                ret = xmemdupz(ent->data, ent->size);
1536        }
1537        *type = ent->type;
1538        *base_size = ent->size;
1539        return ret;
1540}
1541
1542static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1543{
1544        if (ent->data) {
1545                free(ent->data);
1546                ent->data = NULL;
1547                ent->lru.next->prev = ent->lru.prev;
1548                ent->lru.prev->next = ent->lru.next;
1549                delta_base_cached -= ent->size;
1550        }
1551}
1552
1553static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1554        void *base, unsigned long base_size, enum object_type type)
1555{
1556        unsigned long hash = pack_entry_hash(p, base_offset);
1557        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1558        struct delta_base_cache_lru_list *lru;
1559
1560        release_delta_base_cache(ent);
1561        delta_base_cached += base_size;
1562
1563        for (lru = delta_base_cache_lru.next;
1564             delta_base_cached > delta_base_cache_limit
1565             && lru != &delta_base_cache_lru;
1566             lru = lru->next) {
1567                struct delta_base_cache_entry *f = (void *)lru;
1568                if (f->type == OBJ_BLOB)
1569                        release_delta_base_cache(f);
1570        }
1571        for (lru = delta_base_cache_lru.next;
1572             delta_base_cached > delta_base_cache_limit
1573             && lru != &delta_base_cache_lru;
1574             lru = lru->next) {
1575                struct delta_base_cache_entry *f = (void *)lru;
1576                release_delta_base_cache(f);
1577        }
1578
1579        ent->p = p;
1580        ent->base_offset = base_offset;
1581        ent->type = type;
1582        ent->data = base;
1583        ent->size = base_size;
1584        ent->lru.next = &delta_base_cache_lru;
1585        ent->lru.prev = delta_base_cache_lru.prev;
1586        delta_base_cache_lru.prev->next = &ent->lru;
1587        delta_base_cache_lru.prev = &ent->lru;
1588}
1589
1590static void *unpack_delta_entry(struct packed_git *p,
1591                                struct pack_window **w_curs,
1592                                off_t curpos,
1593                                unsigned long delta_size,
1594                                off_t obj_offset,
1595                                enum object_type *type,
1596                                unsigned long *sizep)
1597{
1598        void *delta_data, *result, *base;
1599        unsigned long base_size;
1600        off_t base_offset;
1601
1602        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1603        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1604        if (!base)
1605                die("failed to read delta base object"
1606                    " at %"PRIuMAX" from %s",
1607                    (uintmax_t)base_offset, p->pack_name);
1608
1609        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1610        if (!delta_data)
1611                die("failed to unpack compressed delta"
1612                    " at %"PRIuMAX" from %s",
1613                    (uintmax_t)curpos, p->pack_name);
1614        result = patch_delta(base, base_size,
1615                             delta_data, delta_size,
1616                             sizep);
1617        if (!result)
1618                die("failed to apply delta");
1619        free(delta_data);
1620        add_delta_base_cache(p, base_offset, base, base_size, *type);
1621        return result;
1622}
1623
1624void *unpack_entry(struct packed_git *p, off_t obj_offset,
1625                   enum object_type *type, unsigned long *sizep)
1626{
1627        struct pack_window *w_curs = NULL;
1628        off_t curpos = obj_offset;
1629        void *data;
1630
1631        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1632        switch (*type) {
1633        case OBJ_OFS_DELTA:
1634        case OBJ_REF_DELTA:
1635                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1636                                          obj_offset, type, sizep);
1637                break;
1638        case OBJ_COMMIT:
1639        case OBJ_TREE:
1640        case OBJ_BLOB:
1641        case OBJ_TAG:
1642                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1643                break;
1644        default:
1645                die("unknown object type %i in %s", *type, p->pack_name);
1646        }
1647        unuse_pack(&w_curs);
1648        return data;
1649}
1650
1651const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1652                                            uint32_t n)
1653{
1654        const unsigned char *index = p->index_data;
1655        if (!index) {
1656                if (open_pack_index(p))
1657                        return NULL;
1658                index = p->index_data;
1659        }
1660        if (n >= p->num_objects)
1661                return NULL;
1662        index += 4 * 256;
1663        if (p->index_version == 1) {
1664                return index + 24 * n + 4;
1665        } else {
1666                index += 8;
1667                return index + 20 * n;
1668        }
1669}
1670
1671static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1672{
1673        const unsigned char *index = p->index_data;
1674        index += 4 * 256;
1675        if (p->index_version == 1) {
1676                return ntohl(*((uint32_t *)(index + 24 * n)));
1677        } else {
1678                uint32_t off;
1679                index += 8 + p->num_objects * (20 + 4);
1680                off = ntohl(*((uint32_t *)(index + 4 * n)));
1681                if (!(off & 0x80000000))
1682                        return off;
1683                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1684                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1685                                   ntohl(*((uint32_t *)(index + 4)));
1686        }
1687}
1688
1689off_t find_pack_entry_one(const unsigned char *sha1,
1690                                  struct packed_git *p)
1691{
1692        const uint32_t *level1_ofs = p->index_data;
1693        const unsigned char *index = p->index_data;
1694        unsigned hi, lo, stride;
1695        static int use_lookup = -1;
1696        static int debug_lookup = -1;
1697
1698        if (debug_lookup < 0)
1699                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1700
1701        if (!index) {
1702                if (open_pack_index(p))
1703                        return 0;
1704                level1_ofs = p->index_data;
1705                index = p->index_data;
1706        }
1707        if (p->index_version > 1) {
1708                level1_ofs += 2;
1709                index += 8;
1710        }
1711        index += 4 * 256;
1712        hi = ntohl(level1_ofs[*sha1]);
1713        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1714        if (p->index_version > 1) {
1715                stride = 20;
1716        } else {
1717                stride = 24;
1718                index += 4;
1719        }
1720
1721        if (debug_lookup)
1722                printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1723                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1724
1725        if (use_lookup < 0)
1726                use_lookup = !!getenv("GIT_USE_LOOKUP");
1727        if (use_lookup) {
1728                int pos = sha1_entry_pos(index, stride, 0,
1729                                         lo, hi, p->num_objects, sha1);
1730                if (pos < 0)
1731                        return 0;
1732                return nth_packed_object_offset(p, pos);
1733        }
1734
1735        do {
1736                unsigned mi = (lo + hi) / 2;
1737                int cmp = hashcmp(index + mi * stride, sha1);
1738
1739                if (debug_lookup)
1740                        printf("lo %u hi %u rg %u mi %u\n",
1741                               lo, hi, hi - lo, mi);
1742                if (!cmp)
1743                        return nth_packed_object_offset(p, mi);
1744                if (cmp > 0)
1745                        hi = mi;
1746                else
1747                        lo = mi+1;
1748        } while (lo < hi);
1749        return 0;
1750}
1751
1752int matches_pack_name(struct packed_git *p, const char *name)
1753{
1754        const char *last_c, *c;
1755
1756        if (!strcmp(p->pack_name, name))
1757                return 1;
1758
1759        for (c = p->pack_name, last_c = c; *c;)
1760                if (*c == '/')
1761                        last_c = ++c;
1762                else
1763                        ++c;
1764        if (!strcmp(last_c, name))
1765                return 1;
1766
1767        return 0;
1768}
1769
1770static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1771{
1772        static struct packed_git *last_found = (void *)1;
1773        struct packed_git *p;
1774        off_t offset;
1775
1776        prepare_packed_git();
1777        if (!packed_git)
1778                return 0;
1779        p = (last_found == (void *)1) ? packed_git : last_found;
1780
1781        do {
1782                if (ignore_packed) {
1783                        const char **ig;
1784                        for (ig = ignore_packed; *ig; ig++)
1785                                if (matches_pack_name(p, *ig))
1786                                        break;
1787                        if (*ig)
1788                                goto next;
1789                }
1790
1791                offset = find_pack_entry_one(sha1, p);
1792                if (offset) {
1793                        /*
1794                         * We are about to tell the caller where they can
1795                         * locate the requested object.  We better make
1796                         * sure the packfile is still here and can be
1797                         * accessed before supplying that answer, as
1798                         * it may have been deleted since the index
1799                         * was loaded!
1800                         */
1801                        if (p->pack_fd == -1 && open_packed_git(p)) {
1802                                error("packfile %s cannot be accessed", p->pack_name);
1803                                goto next;
1804                        }
1805                        e->offset = offset;
1806                        e->p = p;
1807                        hashcpy(e->sha1, sha1);
1808                        last_found = p;
1809                        return 1;
1810                }
1811
1812                next:
1813                if (p == last_found)
1814                        p = packed_git;
1815                else
1816                        p = p->next;
1817                if (p == last_found)
1818                        p = p->next;
1819        } while (p);
1820        return 0;
1821}
1822
1823struct packed_git *find_sha1_pack(const unsigned char *sha1,
1824                                  struct packed_git *packs)
1825{
1826        struct packed_git *p;
1827
1828        for (p = packs; p; p = p->next) {
1829                if (find_pack_entry_one(sha1, p))
1830                        return p;
1831        }
1832        return NULL;
1833
1834}
1835
1836static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1837{
1838        int status;
1839        unsigned long mapsize, size;
1840        void *map;
1841        z_stream stream;
1842        char hdr[32];
1843
1844        map = map_sha1_file(sha1, &mapsize);
1845        if (!map)
1846                return error("unable to find %s", sha1_to_hex(sha1));
1847        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1848                status = error("unable to unpack %s header",
1849                               sha1_to_hex(sha1));
1850        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1851                status = error("unable to parse %s header", sha1_to_hex(sha1));
1852        else if (sizep)
1853                *sizep = size;
1854        inflateEnd(&stream);
1855        munmap(map, mapsize);
1856        return status;
1857}
1858
1859int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1860{
1861        struct pack_entry e;
1862
1863        if (!find_pack_entry(sha1, &e, NULL)) {
1864                reprepare_packed_git();
1865                if (!find_pack_entry(sha1, &e, NULL))
1866                        return sha1_loose_object_info(sha1, sizep);
1867        }
1868        return packed_object_info(e.p, e.offset, sizep);
1869}
1870
1871static void *read_packed_sha1(const unsigned char *sha1,
1872                              enum object_type *type, unsigned long *size)
1873{
1874        struct pack_entry e;
1875
1876        if (!find_pack_entry(sha1, &e, NULL))
1877                return NULL;
1878        else
1879                return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1880}
1881
1882/*
1883 * This is meant to hold a *small* number of objects that you would
1884 * want read_sha1_file() to be able to return, but yet you do not want
1885 * to write them into the object store (e.g. a browse-only
1886 * application).
1887 */
1888static struct cached_object {
1889        unsigned char sha1[20];
1890        enum object_type type;
1891        void *buf;
1892        unsigned long size;
1893} *cached_objects;
1894static int cached_object_nr, cached_object_alloc;
1895
1896static struct cached_object empty_tree = {
1897        /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1898        "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1899        "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1900        OBJ_TREE,
1901        "",
1902        0
1903};
1904
1905static struct cached_object *find_cached_object(const unsigned char *sha1)
1906{
1907        int i;
1908        struct cached_object *co = cached_objects;
1909
1910        for (i = 0; i < cached_object_nr; i++, co++) {
1911                if (!hashcmp(co->sha1, sha1))
1912                        return co;
1913        }
1914        if (!hashcmp(sha1, empty_tree.sha1))
1915                return &empty_tree;
1916        return NULL;
1917}
1918
1919int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1920                      unsigned char *sha1)
1921{
1922        struct cached_object *co;
1923
1924        hash_sha1_file(buf, len, typename(type), sha1);
1925        if (has_sha1_file(sha1) || find_cached_object(sha1))
1926                return 0;
1927        if (cached_object_alloc <= cached_object_nr) {
1928                cached_object_alloc = alloc_nr(cached_object_alloc);
1929                cached_objects = xrealloc(cached_objects,
1930                                          sizeof(*cached_objects) *
1931                                          cached_object_alloc);
1932        }
1933        co = &cached_objects[cached_object_nr++];
1934        co->size = len;
1935        co->type = type;
1936        co->buf = xmalloc(len);
1937        memcpy(co->buf, buf, len);
1938        hashcpy(co->sha1, sha1);
1939        return 0;
1940}
1941
1942void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1943                     unsigned long *size)
1944{
1945        unsigned long mapsize;
1946        void *map, *buf;
1947        struct cached_object *co;
1948
1949        co = find_cached_object(sha1);
1950        if (co) {
1951                *type = co->type;
1952                *size = co->size;
1953                return xmemdupz(co->buf, co->size);
1954        }
1955
1956        buf = read_packed_sha1(sha1, type, size);
1957        if (buf)
1958                return buf;
1959        map = map_sha1_file(sha1, &mapsize);
1960        if (map) {
1961                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1962                munmap(map, mapsize);
1963                return buf;
1964        }
1965        reprepare_packed_git();
1966        return read_packed_sha1(sha1, type, size);
1967}
1968
1969void *read_object_with_reference(const unsigned char *sha1,
1970                                 const char *required_type_name,
1971                                 unsigned long *size,
1972                                 unsigned char *actual_sha1_return)
1973{
1974        enum object_type type, required_type;
1975        void *buffer;
1976        unsigned long isize;
1977        unsigned char actual_sha1[20];
1978
1979        required_type = type_from_string(required_type_name);
1980        hashcpy(actual_sha1, sha1);
1981        while (1) {
1982                int ref_length = -1;
1983                const char *ref_type = NULL;
1984
1985                buffer = read_sha1_file(actual_sha1, &type, &isize);
1986                if (!buffer)
1987                        return NULL;
1988                if (type == required_type) {
1989                        *size = isize;
1990                        if (actual_sha1_return)
1991                                hashcpy(actual_sha1_return, actual_sha1);
1992                        return buffer;
1993                }
1994                /* Handle references */
1995                else if (type == OBJ_COMMIT)
1996                        ref_type = "tree ";
1997                else if (type == OBJ_TAG)
1998                        ref_type = "object ";
1999                else {
2000                        free(buffer);
2001                        return NULL;
2002                }
2003                ref_length = strlen(ref_type);
2004
2005                if (ref_length + 40 > isize ||
2006                    memcmp(buffer, ref_type, ref_length) ||
2007                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2008                        free(buffer);
2009                        return NULL;
2010                }
2011                free(buffer);
2012                /* Now we have the ID of the referred-to object in
2013                 * actual_sha1.  Check again. */
2014        }
2015}
2016
2017static void write_sha1_file_prepare(const void *buf, unsigned long len,
2018                                    const char *type, unsigned char *sha1,
2019                                    char *hdr, int *hdrlen)
2020{
2021        SHA_CTX c;
2022
2023        /* Generate the header */
2024        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2025
2026        /* Sha1.. */
2027        SHA1_Init(&c);
2028        SHA1_Update(&c, hdr, *hdrlen);
2029        SHA1_Update(&c, buf, len);
2030        SHA1_Final(sha1, &c);
2031}
2032
2033/*
2034 * Move the just written object into its final resting place
2035 */
2036int move_temp_to_file(const char *tmpfile, const char *filename)
2037{
2038        int ret = link(tmpfile, filename);
2039
2040        /*
2041         * Coda hack - coda doesn't like cross-directory links,
2042         * so we fall back to a rename, which will mean that it
2043         * won't be able to check collisions, but that's not a
2044         * big deal.
2045         *
2046         * The same holds for FAT formatted media.
2047         *
2048         * When this succeeds, we just return 0. We have nothing
2049         * left to unlink.
2050         */
2051        if (ret && ret != EEXIST) {
2052                if (!rename(tmpfile, filename))
2053                        return 0;
2054                ret = errno;
2055        }
2056        unlink(tmpfile);
2057        if (ret) {
2058                if (ret != EEXIST) {
2059                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2060                }
2061                /* FIXME!!! Collision check here ? */
2062        }
2063
2064        return 0;
2065}
2066
2067static int write_buffer(int fd, const void *buf, size_t len)
2068{
2069        if (write_in_full(fd, buf, len) < 0)
2070                return error("file write error (%s)", strerror(errno));
2071        return 0;
2072}
2073
2074int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2075                   unsigned char *sha1)
2076{
2077        char hdr[32];
2078        int hdrlen;
2079        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2080        return 0;
2081}
2082
2083/* Finalize a file on disk, and close it. */
2084static void close_sha1_file(int fd)
2085{
2086        /* For safe-mode, we could fsync_or_die(fd, "sha1 file") here */
2087        fchmod(fd, 0444);
2088        if (close(fd) != 0)
2089                die("unable to write sha1 file");
2090}
2091
2092/* Size of directory component, including the ending '/' */
2093static inline int directory_size(const char *filename)
2094{
2095        const char *s = strrchr(filename, '/');
2096        if (!s)
2097                return 0;
2098        return s - filename + 1;
2099}
2100
2101/*
2102 * This creates a temporary file in the same directory as the final
2103 * 'filename'
2104 *
2105 * We want to avoid cross-directory filename renames, because those
2106 * can have problems on various filesystems (FAT, NFS, Coda).
2107 */
2108static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2109{
2110        int fd, dirlen = directory_size(filename);
2111
2112        if (dirlen + 20 > bufsiz) {
2113                errno = ENAMETOOLONG;
2114                return -1;
2115        }
2116        memcpy(buffer, filename, dirlen);
2117        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2118        fd = mkstemp(buffer);
2119        if (fd < 0 && dirlen) {
2120                /* Make sure the directory exists */
2121                buffer[dirlen-1] = 0;
2122                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2123                        return -1;
2124
2125                /* Try again */
2126                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2127                fd = mkstemp(buffer);
2128        }
2129        return fd;
2130}
2131
2132static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2133                              void *buf, unsigned long len, time_t mtime)
2134{
2135        int fd, size, ret;
2136        unsigned char *compressed;
2137        z_stream stream;
2138        char *filename;
2139        static char tmpfile[PATH_MAX];
2140
2141        filename = sha1_file_name(sha1);
2142        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2143        if (fd < 0) {
2144                if (errno == EPERM)
2145                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2146                else
2147                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2148        }
2149
2150        /* Set it up */
2151        memset(&stream, 0, sizeof(stream));
2152        deflateInit(&stream, zlib_compression_level);
2153        size = 8 + deflateBound(&stream, len+hdrlen);
2154        compressed = xmalloc(size);
2155
2156        /* Compress it */
2157        stream.next_out = compressed;
2158        stream.avail_out = size;
2159
2160        /* First header.. */
2161        stream.next_in = (unsigned char *)hdr;
2162        stream.avail_in = hdrlen;
2163        while (deflate(&stream, 0) == Z_OK)
2164                /* nothing */;
2165
2166        /* Then the data itself.. */
2167        stream.next_in = buf;
2168        stream.avail_in = len;
2169        ret = deflate(&stream, Z_FINISH);
2170        if (ret != Z_STREAM_END)
2171                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2172
2173        ret = deflateEnd(&stream);
2174        if (ret != Z_OK)
2175                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2176
2177        size = stream.total_out;
2178
2179        if (write_buffer(fd, compressed, size) < 0)
2180                die("unable to write sha1 file");
2181        close_sha1_file(fd);
2182        free(compressed);
2183
2184        if (mtime) {
2185                struct utimbuf utb;
2186                utb.actime = mtime;
2187                utb.modtime = mtime;
2188                if (utime(tmpfile, &utb) < 0)
2189                        warning("failed utime() on %s: %s",
2190                                tmpfile, strerror(errno));
2191        }
2192
2193        return move_temp_to_file(tmpfile, filename);
2194}
2195
2196int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2197{
2198        unsigned char sha1[20];
2199        char hdr[32];
2200        int hdrlen;
2201
2202        /* Normally if we have it in the pack then we do not bother writing
2203         * it out into .git/objects/??/?{38} file.
2204         */
2205        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2206        if (returnsha1)
2207                hashcpy(returnsha1, sha1);
2208        if (has_sha1_file(sha1))
2209                return 0;
2210        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2211}
2212
2213int force_object_loose(const unsigned char *sha1, time_t mtime)
2214{
2215        void *buf;
2216        unsigned long len;
2217        enum object_type type;
2218        char hdr[32];
2219        int hdrlen;
2220
2221        if (has_loose_object(sha1))
2222                return 0;
2223        buf = read_packed_sha1(sha1, &type, &len);
2224        if (!buf)
2225                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2226        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2227        return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2228}
2229
2230int has_pack_index(const unsigned char *sha1)
2231{
2232        struct stat st;
2233        if (stat(sha1_pack_index_name(sha1), &st))
2234                return 0;
2235        return 1;
2236}
2237
2238int has_pack_file(const unsigned char *sha1)
2239{
2240        struct stat st;
2241        if (stat(sha1_pack_name(sha1), &st))
2242                return 0;
2243        return 1;
2244}
2245
2246int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2247{
2248        struct pack_entry e;
2249        return find_pack_entry(sha1, &e, ignore_packed);
2250}
2251
2252int has_sha1_file(const unsigned char *sha1)
2253{
2254        struct pack_entry e;
2255
2256        if (find_pack_entry(sha1, &e, NULL))
2257                return 1;
2258        return has_loose_object(sha1);
2259}
2260
2261int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2262{
2263        struct strbuf buf;
2264        int ret;
2265
2266        strbuf_init(&buf, 0);
2267        if (strbuf_read(&buf, fd, 4096) < 0) {
2268                strbuf_release(&buf);
2269                return -1;
2270        }
2271
2272        if (!type)
2273                type = blob_type;
2274        if (write_object)
2275                ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2276        else
2277                ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2278        strbuf_release(&buf);
2279
2280        return ret;
2281}
2282
2283int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2284             enum object_type type, const char *path)
2285{
2286        size_t size = xsize_t(st->st_size);
2287        void *buf = NULL;
2288        int ret, re_allocated = 0;
2289
2290        if (size)
2291                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2292        close(fd);
2293
2294        if (!type)
2295                type = OBJ_BLOB;
2296
2297        /*
2298         * Convert blobs to git internal format
2299         */
2300        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2301                struct strbuf nbuf;
2302                strbuf_init(&nbuf, 0);
2303                if (convert_to_git(path, buf, size, &nbuf,
2304                                   write_object ? safe_crlf : 0)) {
2305                        munmap(buf, size);
2306                        buf = strbuf_detach(&nbuf, &size);
2307                        re_allocated = 1;
2308                }
2309        }
2310
2311        if (write_object)
2312                ret = write_sha1_file(buf, size, typename(type), sha1);
2313        else
2314                ret = hash_sha1_file(buf, size, typename(type), sha1);
2315        if (re_allocated) {
2316                free(buf);
2317                return ret;
2318        }
2319        if (size)
2320                munmap(buf, size);
2321        return ret;
2322}
2323
2324int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2325{
2326        int fd;
2327        char *target;
2328        size_t len;
2329
2330        switch (st->st_mode & S_IFMT) {
2331        case S_IFREG:
2332                fd = open(path, O_RDONLY);
2333                if (fd < 0)
2334                        return error("open(\"%s\"): %s", path,
2335                                     strerror(errno));
2336                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2337                        return error("%s: failed to insert into database",
2338                                     path);
2339                break;
2340        case S_IFLNK:
2341                len = xsize_t(st->st_size);
2342                target = xmalloc(len + 1);
2343                if (readlink(path, target, len + 1) != st->st_size) {
2344                        char *errstr = strerror(errno);
2345                        free(target);
2346                        return error("readlink(\"%s\"): %s", path,
2347                                     errstr);
2348                }
2349                if (!write_object)
2350                        hash_sha1_file(target, len, blob_type, sha1);
2351                else if (write_sha1_file(target, len, blob_type, sha1))
2352                        return error("%s: failed to insert into database",
2353                                     path);
2354                free(target);
2355                break;
2356        case S_IFDIR:
2357                return resolve_gitlink_ref(path, "HEAD", sha1);
2358        default:
2359                return error("%s: unsupported file type", path);
2360        }
2361        return 0;
2362}
2363
2364int read_pack_header(int fd, struct pack_header *header)
2365{
2366        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2367                /* "eof before pack header was fully read" */
2368                return PH_ERROR_EOF;
2369
2370        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2371                /* "protocol error (pack signature mismatch detected)" */
2372                return PH_ERROR_PACK_SIGNATURE;
2373        if (!pack_version_ok(header->hdr_version))
2374                /* "protocol error (pack version unsupported)" */
2375                return PH_ERROR_PROTOCOL;
2376        return 0;
2377}