sha1_file.con commit Remove redundant code, eliminate one static variable (633f43e)
   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
  38static unsigned int sha1_file_open_flag = O_NOATIME;
  39
  40const signed char hexval_table[256] = {
  41         -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
  42         -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
  43         -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
  44         -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
  45         -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
  46         -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
  47          0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
  48          8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
  49         -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
  50         -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
  51         -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
  52         -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
  53         -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
  54         -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
  55         -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
  56         -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
  57         -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
  58         -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
  59         -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
  60         -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
  61         -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
  62         -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
  63         -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
  64         -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
  65         -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
  66         -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
  67         -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
  68         -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
  69         -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
  70         -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
  71         -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
  72         -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
  73};
  74
  75int get_sha1_hex(const char *hex, unsigned char *sha1)
  76{
  77        int i;
  78        for (i = 0; i < 20; i++) {
  79                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  80                if (val & ~0xff)
  81                        return -1;
  82                *sha1++ = val;
  83                hex += 2;
  84        }
  85        return 0;
  86}
  87
  88int safe_create_leading_directories(char *path)
  89{
  90        char *pos = path;
  91        struct stat st;
  92
  93        if (is_absolute_path(path))
  94                pos++;
  95
  96        while (pos) {
  97                pos = strchr(pos, '/');
  98                if (!pos)
  99                        break;
 100                *pos = 0;
 101                if (!stat(path, &st)) {
 102                        /* path exists */
 103                        if (!S_ISDIR(st.st_mode)) {
 104                                *pos = '/';
 105                                return -3;
 106                        }
 107                }
 108                else if (mkdir(path, 0777)) {
 109                        *pos = '/';
 110                        return -1;
 111                }
 112                else if (adjust_shared_perm(path)) {
 113                        *pos = '/';
 114                        return -2;
 115                }
 116                *pos++ = '/';
 117        }
 118        return 0;
 119}
 120
 121char * sha1_to_hex(const unsigned char *sha1)
 122{
 123        static int bufno;
 124        static char hexbuffer[4][50];
 125        static const char hex[] = "0123456789abcdef";
 126        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 127        int i;
 128
 129        for (i = 0; i < 20; i++) {
 130                unsigned int val = *sha1++;
 131                *buf++ = hex[val >> 4];
 132                *buf++ = hex[val & 0xf];
 133        }
 134        *buf = '\0';
 135
 136        return buffer;
 137}
 138
 139static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 140{
 141        int i;
 142        for (i = 0; i < 20; i++) {
 143                static char hex[] = "0123456789abcdef";
 144                unsigned int val = sha1[i];
 145                char *pos = pathbuf + i*2 + (i > 0);
 146                *pos++ = hex[val >> 4];
 147                *pos = hex[val & 0xf];
 148        }
 149}
 150
 151/*
 152 * NOTE! This returns a statically allocated buffer, so you have to be
 153 * careful about using it. Do an "xstrdup()" if you need to save the
 154 * filename.
 155 *
 156 * Also note that this returns the location for creating.  Reading
 157 * SHA1 file can happen from any alternate directory listed in the
 158 * DB_ENVIRONMENT environment variable if it is not found in
 159 * the primary object database.
 160 */
 161char *sha1_file_name(const unsigned char *sha1)
 162{
 163        static char *name, *base;
 164
 165        if (!base) {
 166                const char *sha1_file_directory = get_object_directory();
 167                int len = strlen(sha1_file_directory);
 168                base = xmalloc(len + 60);
 169                memcpy(base, sha1_file_directory, len);
 170                memset(base+len, 0, 60);
 171                base[len] = '/';
 172                base[len+3] = '/';
 173                name = base + len + 1;
 174        }
 175        fill_sha1_path(name, sha1);
 176        return base;
 177}
 178
 179static char *sha1_get_pack_name(const unsigned char *sha1,
 180                                char **name, char **base)
 181{
 182        static const char hex[] = "0123456789abcdef";
 183        char *buf;
 184        int i;
 185
 186        if (!*base) {
 187                const char *sha1_file_directory = get_object_directory();
 188                int len = strlen(sha1_file_directory);
 189                *base = xmalloc(len + 60);
 190                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
 191                *name = *base + len + 11;
 192        }
 193
 194        buf = *name;
 195
 196        for (i = 0; i < 20; i++) {
 197                unsigned int val = *sha1++;
 198                *buf++ = hex[val >> 4];
 199                *buf++ = hex[val & 0xf];
 200        }
 201
 202        return *base;
 203}
 204
 205char *sha1_pack_name(const unsigned char *sha1)
 206{
 207        static char *name, *base;
 208
 209        return sha1_get_pack_name(sha1, &name, &base);
 210}
 211
 212char *sha1_pack_index_name(const unsigned char *sha1)
 213{
 214        static char *name, *base;
 215
 216        return sha1_get_pack_name(sha1, &name, &base);
 217}
 218
 219struct alternate_object_database *alt_odb_list;
 220static struct alternate_object_database **alt_odb_tail;
 221
 222static void read_info_alternates(const char * alternates, int depth);
 223
 224/*
 225 * Prepare alternate object database registry.
 226 *
 227 * The variable alt_odb_list points at the list of struct
 228 * alternate_object_database.  The elements on this list come from
 229 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 230 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 231 * whose contents is similar to that environment variable but can be
 232 * LF separated.  Its base points at a statically allocated buffer that
 233 * contains "/the/directory/corresponding/to/.git/objects/...", while
 234 * its name points just after the slash at the end of ".git/objects/"
 235 * in the example above, and has enough space to hold 40-byte hex
 236 * SHA1, an extra slash for the first level indirection, and the
 237 * terminating NUL.
 238 */
 239static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 240{
 241        struct stat st;
 242        const char *objdir = get_object_directory();
 243        struct alternate_object_database *ent;
 244        struct alternate_object_database *alt;
 245        /* 43 = 40-byte + 2 '/' + terminating NUL */
 246        int pfxlen = len;
 247        int entlen = pfxlen + 43;
 248        int base_len = -1;
 249
 250        if (!is_absolute_path(entry) && relative_base) {
 251                /* Relative alt-odb */
 252                if (base_len < 0)
 253                        base_len = strlen(relative_base) + 1;
 254                entlen += base_len;
 255                pfxlen += base_len;
 256        }
 257        ent = xmalloc(sizeof(*ent) + entlen);
 258
 259        if (!is_absolute_path(entry) && relative_base) {
 260                memcpy(ent->base, relative_base, base_len - 1);
 261                ent->base[base_len - 1] = '/';
 262                memcpy(ent->base + base_len, entry, len);
 263        }
 264        else
 265                memcpy(ent->base, entry, pfxlen);
 266
 267        ent->name = ent->base + pfxlen + 1;
 268        ent->base[pfxlen + 3] = '/';
 269        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 270
 271        /* Detect cases where alternate disappeared */
 272        if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
 273                error("object directory %s does not exist; "
 274                      "check .git/objects/info/alternates.",
 275                      ent->base);
 276                free(ent);
 277                return -1;
 278        }
 279
 280        /* Prevent the common mistake of listing the same
 281         * thing twice, or object directory itself.
 282         */
 283        for (alt = alt_odb_list; alt; alt = alt->next) {
 284                if (!memcmp(ent->base, alt->base, pfxlen)) {
 285                        free(ent);
 286                        return -1;
 287                }
 288        }
 289        if (!memcmp(ent->base, objdir, pfxlen)) {
 290                free(ent);
 291                return -1;
 292        }
 293
 294        /* add the alternate entry */
 295        *alt_odb_tail = ent;
 296        alt_odb_tail = &(ent->next);
 297        ent->next = NULL;
 298
 299        /* recursively add alternates */
 300        read_info_alternates(ent->base, depth + 1);
 301
 302        ent->base[pfxlen] = '/';
 303
 304        return 0;
 305}
 306
 307static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 308                                 const char *relative_base, int depth)
 309{
 310        const char *cp, *last;
 311
 312        if (depth > 5) {
 313                error("%s: ignoring alternate object stores, nesting too deep.",
 314                                relative_base);
 315                return;
 316        }
 317
 318        last = alt;
 319        while (last < ep) {
 320                cp = last;
 321                if (cp < ep && *cp == '#') {
 322                        while (cp < ep && *cp != sep)
 323                                cp++;
 324                        last = cp + 1;
 325                        continue;
 326                }
 327                while (cp < ep && *cp != sep)
 328                        cp++;
 329                if (last != cp) {
 330                        if (!is_absolute_path(last) && depth) {
 331                                error("%s: ignoring relative alternate object store %s",
 332                                                relative_base, last);
 333                        } else {
 334                                link_alt_odb_entry(last, cp - last,
 335                                                relative_base, depth);
 336                        }
 337                }
 338                while (cp < ep && *cp == sep)
 339                        cp++;
 340                last = cp;
 341        }
 342}
 343
 344static void read_info_alternates(const char * relative_base, int depth)
 345{
 346        char *map;
 347        size_t mapsz;
 348        struct stat st;
 349        const char alt_file_name[] = "info/alternates";
 350        /* Given that relative_base is no longer than PATH_MAX,
 351           ensure that "path" has enough space to append "/", the
 352           file name, "info/alternates", and a trailing NUL.  */
 353        char path[PATH_MAX + 1 + sizeof alt_file_name];
 354        int fd;
 355
 356        sprintf(path, "%s/%s", relative_base, alt_file_name);
 357        fd = open(path, O_RDONLY);
 358        if (fd < 0)
 359                return;
 360        if (fstat(fd, &st) || (st.st_size == 0)) {
 361                close(fd);
 362                return;
 363        }
 364        mapsz = xsize_t(st.st_size);
 365        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 366        close(fd);
 367
 368        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 369
 370        munmap(map, mapsz);
 371}
 372
 373void prepare_alt_odb(void)
 374{
 375        const char *alt;
 376
 377        if (alt_odb_tail)
 378                return;
 379
 380        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 381        if (!alt) alt = "";
 382
 383        alt_odb_tail = &alt_odb_list;
 384        link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
 385
 386        read_info_alternates(get_object_directory(), 0);
 387}
 388
 389static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 390{
 391        char *name = sha1_file_name(sha1);
 392        struct alternate_object_database *alt;
 393
 394        if (!stat(name, st))
 395                return name;
 396        prepare_alt_odb();
 397        for (alt = alt_odb_list; alt; alt = alt->next) {
 398                name = alt->name;
 399                fill_sha1_path(name, sha1);
 400                if (!stat(alt->base, st))
 401                        return alt->base;
 402        }
 403        return NULL;
 404}
 405
 406static unsigned int pack_used_ctr;
 407static unsigned int pack_mmap_calls;
 408static unsigned int peak_pack_open_windows;
 409static unsigned int pack_open_windows;
 410static size_t peak_pack_mapped;
 411static size_t pack_mapped;
 412struct packed_git *packed_git;
 413
 414void pack_report(void)
 415{
 416        fprintf(stderr,
 417                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 418                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 419                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 420                sz_fmt(getpagesize()),
 421                sz_fmt(packed_git_window_size),
 422                sz_fmt(packed_git_limit));
 423        fprintf(stderr,
 424                "pack_report: pack_used_ctr            = %10u\n"
 425                "pack_report: pack_mmap_calls          = %10u\n"
 426                "pack_report: pack_open_windows        = %10u / %10u\n"
 427                "pack_report: pack_mapped              = "
 428                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 429                pack_used_ctr,
 430                pack_mmap_calls,
 431                pack_open_windows, peak_pack_open_windows,
 432                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 433}
 434
 435static int check_packed_git_idx(const char *path,  struct packed_git *p)
 436{
 437        void *idx_map;
 438        struct pack_idx_header *hdr;
 439        size_t idx_size;
 440        uint32_t version, nr, i, *index;
 441        int fd = open(path, O_RDONLY);
 442        struct stat st;
 443
 444        if (fd < 0)
 445                return -1;
 446        if (fstat(fd, &st)) {
 447                close(fd);
 448                return -1;
 449        }
 450        idx_size = xsize_t(st.st_size);
 451        if (idx_size < 4 * 256 + 20 + 20) {
 452                close(fd);
 453                return error("index file %s is too small", path);
 454        }
 455        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 456        close(fd);
 457
 458        hdr = idx_map;
 459        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 460                version = ntohl(hdr->idx_version);
 461                if (version < 2 || version > 2) {
 462                        munmap(idx_map, idx_size);
 463                        return error("index file %s is version %d"
 464                                     " and is not supported by this binary"
 465                                     " (try upgrading GIT to a newer version)",
 466                                     path, version);
 467                }
 468        } else
 469                version = 1;
 470
 471        nr = 0;
 472        index = idx_map;
 473        if (version > 1)
 474                index += 2;  /* skip index header */
 475        for (i = 0; i < 256; i++) {
 476                uint32_t n = ntohl(index[i]);
 477                if (n < nr) {
 478                        munmap(idx_map, idx_size);
 479                        return error("non-monotonic index %s", path);
 480                }
 481                nr = n;
 482        }
 483
 484        if (version == 1) {
 485                /*
 486                 * Total size:
 487                 *  - 256 index entries 4 bytes each
 488                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 489                 *  - 20-byte SHA1 of the packfile
 490                 *  - 20-byte SHA1 file checksum
 491                 */
 492                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 493                        munmap(idx_map, idx_size);
 494                        return error("wrong index v1 file size in %s", path);
 495                }
 496        } else if (version == 2) {
 497                /*
 498                 * Minimum size:
 499                 *  - 8 bytes of header
 500                 *  - 256 index entries 4 bytes each
 501                 *  - 20-byte sha1 entry * nr
 502                 *  - 4-byte crc entry * nr
 503                 *  - 4-byte offset entry * nr
 504                 *  - 20-byte SHA1 of the packfile
 505                 *  - 20-byte SHA1 file checksum
 506                 * And after the 4-byte offset table might be a
 507                 * variable sized table containing 8-byte entries
 508                 * for offsets larger than 2^31.
 509                 */
 510                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 511                unsigned long max_size = min_size;
 512                if (nr)
 513                        max_size += (nr - 1)*8;
 514                if (idx_size < min_size || idx_size > max_size) {
 515                        munmap(idx_map, idx_size);
 516                        return error("wrong index v2 file size in %s", path);
 517                }
 518                if (idx_size != min_size &&
 519                    /*
 520                     * make sure we can deal with large pack offsets.
 521                     * 31-bit signed offset won't be enough, neither
 522                     * 32-bit unsigned one will be.
 523                     */
 524                    (sizeof(off_t) <= 4)) {
 525                        munmap(idx_map, idx_size);
 526                        return error("pack too large for current definition of off_t in %s", path);
 527                }
 528        }
 529
 530        p->index_version = version;
 531        p->index_data = idx_map;
 532        p->index_size = idx_size;
 533        p->num_objects = nr;
 534        return 0;
 535}
 536
 537int open_pack_index(struct packed_git *p)
 538{
 539        char *idx_name;
 540        int ret;
 541
 542        if (p->index_data)
 543                return 0;
 544
 545        idx_name = xstrdup(p->pack_name);
 546        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 547        ret = check_packed_git_idx(idx_name, p);
 548        free(idx_name);
 549        return ret;
 550}
 551
 552static void scan_windows(struct packed_git *p,
 553        struct packed_git **lru_p,
 554        struct pack_window **lru_w,
 555        struct pack_window **lru_l)
 556{
 557        struct pack_window *w, *w_l;
 558
 559        for (w_l = NULL, w = p->windows; w; w = w->next) {
 560                if (!w->inuse_cnt) {
 561                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 562                                *lru_p = p;
 563                                *lru_w = w;
 564                                *lru_l = w_l;
 565                        }
 566                }
 567                w_l = w;
 568        }
 569}
 570
 571static int unuse_one_window(struct packed_git *current, int keep_fd)
 572{
 573        struct packed_git *p, *lru_p = NULL;
 574        struct pack_window *lru_w = NULL, *lru_l = NULL;
 575
 576        if (current)
 577                scan_windows(current, &lru_p, &lru_w, &lru_l);
 578        for (p = packed_git; p; p = p->next)
 579                scan_windows(p, &lru_p, &lru_w, &lru_l);
 580        if (lru_p) {
 581                munmap(lru_w->base, lru_w->len);
 582                pack_mapped -= lru_w->len;
 583                if (lru_l)
 584                        lru_l->next = lru_w->next;
 585                else {
 586                        lru_p->windows = lru_w->next;
 587                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 588                                close(lru_p->pack_fd);
 589                                lru_p->pack_fd = -1;
 590                        }
 591                }
 592                free(lru_w);
 593                pack_open_windows--;
 594                return 1;
 595        }
 596        return 0;
 597}
 598
 599void release_pack_memory(size_t need, int fd)
 600{
 601        size_t cur = pack_mapped;
 602        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 603                ; /* nothing */
 604}
 605
 606void close_pack_windows(struct packed_git *p)
 607{
 608        while (p->windows) {
 609                struct pack_window *w = p->windows;
 610
 611                if (w->inuse_cnt)
 612                        die("pack '%s' still has open windows to it",
 613                            p->pack_name);
 614                munmap(w->base, w->len);
 615                pack_mapped -= w->len;
 616                pack_open_windows--;
 617                p->windows = w->next;
 618                free(w);
 619        }
 620}
 621
 622void unuse_pack(struct pack_window **w_cursor)
 623{
 624        struct pack_window *w = *w_cursor;
 625        if (w) {
 626                w->inuse_cnt--;
 627                *w_cursor = NULL;
 628        }
 629}
 630
 631/*
 632 * Do not call this directly as this leaks p->pack_fd on error return;
 633 * call open_packed_git() instead.
 634 */
 635static int open_packed_git_1(struct packed_git *p)
 636{
 637        struct stat st;
 638        struct pack_header hdr;
 639        unsigned char sha1[20];
 640        unsigned char *idx_sha1;
 641        long fd_flag;
 642
 643        if (!p->index_data && open_pack_index(p))
 644                return error("packfile %s index unavailable", p->pack_name);
 645
 646        p->pack_fd = open(p->pack_name, O_RDONLY);
 647        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 648                return -1;
 649
 650        /* If we created the struct before we had the pack we lack size. */
 651        if (!p->pack_size) {
 652                if (!S_ISREG(st.st_mode))
 653                        return error("packfile %s not a regular file", p->pack_name);
 654                p->pack_size = st.st_size;
 655        } else if (p->pack_size != st.st_size)
 656                return error("packfile %s size changed", p->pack_name);
 657
 658        /* We leave these file descriptors open with sliding mmap;
 659         * there is no point keeping them open across exec(), though.
 660         */
 661        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 662        if (fd_flag < 0)
 663                return error("cannot determine file descriptor flags");
 664        fd_flag |= FD_CLOEXEC;
 665        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 666                return error("cannot set FD_CLOEXEC");
 667
 668        /* Verify we recognize this pack file format. */
 669        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 670                return error("file %s is far too short to be a packfile", p->pack_name);
 671        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 672                return error("file %s is not a GIT packfile", p->pack_name);
 673        if (!pack_version_ok(hdr.hdr_version))
 674                return error("packfile %s is version %u and not supported"
 675                        " (try upgrading GIT to a newer version)",
 676                        p->pack_name, ntohl(hdr.hdr_version));
 677
 678        /* Verify the pack matches its index. */
 679        if (p->num_objects != ntohl(hdr.hdr_entries))
 680                return error("packfile %s claims to have %u objects"
 681                             " while index indicates %u objects",
 682                             p->pack_name, ntohl(hdr.hdr_entries),
 683                             p->num_objects);
 684        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 685                return error("end of packfile %s is unavailable", p->pack_name);
 686        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 687                return error("packfile %s signature is unavailable", p->pack_name);
 688        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 689        if (hashcmp(sha1, idx_sha1))
 690                return error("packfile %s does not match index", p->pack_name);
 691        return 0;
 692}
 693
 694static int open_packed_git(struct packed_git *p)
 695{
 696        if (!open_packed_git_1(p))
 697                return 0;
 698        if (p->pack_fd != -1) {
 699                close(p->pack_fd);
 700                p->pack_fd = -1;
 701        }
 702        return -1;
 703}
 704
 705static int in_window(struct pack_window *win, off_t offset)
 706{
 707        /* We must promise at least 20 bytes (one hash) after the
 708         * offset is available from this window, otherwise the offset
 709         * is not actually in this window and a different window (which
 710         * has that one hash excess) must be used.  This is to support
 711         * the object header and delta base parsing routines below.
 712         */
 713        off_t win_off = win->offset;
 714        return win_off <= offset
 715                && (offset + 20) <= (win_off + win->len);
 716}
 717
 718unsigned char* use_pack(struct packed_git *p,
 719                struct pack_window **w_cursor,
 720                off_t offset,
 721                unsigned int *left)
 722{
 723        struct pack_window *win = *w_cursor;
 724
 725        if (p->pack_fd == -1 && open_packed_git(p))
 726                die("packfile %s cannot be accessed", p->pack_name);
 727
 728        /* Since packfiles end in a hash of their content and its
 729         * pointless to ask for an offset into the middle of that
 730         * hash, and the in_window function above wouldn't match
 731         * don't allow an offset too close to the end of the file.
 732         */
 733        if (offset > (p->pack_size - 20))
 734                die("offset beyond end of packfile (truncated pack?)");
 735
 736        if (!win || !in_window(win, offset)) {
 737                if (win)
 738                        win->inuse_cnt--;
 739                for (win = p->windows; win; win = win->next) {
 740                        if (in_window(win, offset))
 741                                break;
 742                }
 743                if (!win) {
 744                        size_t window_align = packed_git_window_size / 2;
 745                        off_t len;
 746                        win = xcalloc(1, sizeof(*win));
 747                        win->offset = (offset / window_align) * window_align;
 748                        len = p->pack_size - win->offset;
 749                        if (len > packed_git_window_size)
 750                                len = packed_git_window_size;
 751                        win->len = (size_t)len;
 752                        pack_mapped += win->len;
 753                        while (packed_git_limit < pack_mapped
 754                                && unuse_one_window(p, p->pack_fd))
 755                                ; /* nothing */
 756                        win->base = xmmap(NULL, win->len,
 757                                PROT_READ, MAP_PRIVATE,
 758                                p->pack_fd, win->offset);
 759                        if (win->base == MAP_FAILED)
 760                                die("packfile %s cannot be mapped: %s",
 761                                        p->pack_name,
 762                                        strerror(errno));
 763                        pack_mmap_calls++;
 764                        pack_open_windows++;
 765                        if (pack_mapped > peak_pack_mapped)
 766                                peak_pack_mapped = pack_mapped;
 767                        if (pack_open_windows > peak_pack_open_windows)
 768                                peak_pack_open_windows = pack_open_windows;
 769                        win->next = p->windows;
 770                        p->windows = win;
 771                }
 772        }
 773        if (win != *w_cursor) {
 774                win->last_used = pack_used_ctr++;
 775                win->inuse_cnt++;
 776                *w_cursor = win;
 777        }
 778        offset -= win->offset;
 779        if (left)
 780                *left = win->len - xsize_t(offset);
 781        return win->base + offset;
 782}
 783
 784struct packed_git *add_packed_git(const char *path, int path_len, int local)
 785{
 786        struct stat st;
 787        struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
 788
 789        /*
 790         * Make sure a corresponding .pack file exists and that
 791         * the index looks sane.
 792         */
 793        path_len -= strlen(".idx");
 794        if (path_len < 1)
 795                return NULL;
 796        memcpy(p->pack_name, path, path_len);
 797        strcpy(p->pack_name + path_len, ".pack");
 798        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 799                free(p);
 800                return NULL;
 801        }
 802
 803        /* ok, it looks sane as far as we can check without
 804         * actually mapping the pack file.
 805         */
 806        p->index_version = 0;
 807        p->index_data = NULL;
 808        p->index_size = 0;
 809        p->num_objects = 0;
 810        p->pack_size = st.st_size;
 811        p->next = NULL;
 812        p->windows = NULL;
 813        p->pack_fd = -1;
 814        p->pack_local = local;
 815        p->mtime = st.st_mtime;
 816        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 817                hashclr(p->sha1);
 818        return p;
 819}
 820
 821struct packed_git *parse_pack_index(unsigned char *sha1)
 822{
 823        char *path = sha1_pack_index_name(sha1);
 824        return parse_pack_index_file(sha1, path);
 825}
 826
 827struct packed_git *parse_pack_index_file(const unsigned char *sha1,
 828                                         const char *idx_path)
 829{
 830        const char *path = sha1_pack_name(sha1);
 831        struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
 832
 833        if (check_packed_git_idx(idx_path, p)) {
 834                free(p);
 835                return NULL;
 836        }
 837
 838        strcpy(p->pack_name, path);
 839        p->pack_size = 0;
 840        p->next = NULL;
 841        p->windows = NULL;
 842        p->pack_fd = -1;
 843        hashcpy(p->sha1, sha1);
 844        return p;
 845}
 846
 847void install_packed_git(struct packed_git *pack)
 848{
 849        pack->next = packed_git;
 850        packed_git = pack;
 851}
 852
 853static void prepare_packed_git_one(char *objdir, int local)
 854{
 855        /* Ensure that this buffer is large enough so that we can
 856           append "/pack/" without clobbering the stack even if
 857           strlen(objdir) were PATH_MAX.  */
 858        char path[PATH_MAX + 1 + 4 + 1 + 1];
 859        int len;
 860        DIR *dir;
 861        struct dirent *de;
 862
 863        sprintf(path, "%s/pack", objdir);
 864        len = strlen(path);
 865        dir = opendir(path);
 866        if (!dir) {
 867                if (errno != ENOENT)
 868                        error("unable to open object pack directory: %s: %s",
 869                              path, strerror(errno));
 870                return;
 871        }
 872        path[len++] = '/';
 873        while ((de = readdir(dir)) != NULL) {
 874                int namelen = strlen(de->d_name);
 875                struct packed_git *p;
 876
 877                if (!has_extension(de->d_name, ".idx"))
 878                        continue;
 879
 880                if (len + namelen + 1 > sizeof(path))
 881                        continue;
 882
 883                /* Don't reopen a pack we already have. */
 884                strcpy(path + len, de->d_name);
 885                for (p = packed_git; p; p = p->next) {
 886                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 887                                break;
 888                }
 889                if (p)
 890                        continue;
 891                /* See if it really is a valid .idx file with corresponding
 892                 * .pack file that we can map.
 893                 */
 894                p = add_packed_git(path, len + namelen, local);
 895                if (!p)
 896                        continue;
 897                install_packed_git(p);
 898        }
 899        closedir(dir);
 900}
 901
 902static int sort_pack(const void *a_, const void *b_)
 903{
 904        struct packed_git *a = *((struct packed_git **)a_);
 905        struct packed_git *b = *((struct packed_git **)b_);
 906        int st;
 907
 908        /*
 909         * Local packs tend to contain objects specific to our
 910         * variant of the project than remote ones.  In addition,
 911         * remote ones could be on a network mounted filesystem.
 912         * Favor local ones for these reasons.
 913         */
 914        st = a->pack_local - b->pack_local;
 915        if (st)
 916                return -st;
 917
 918        /*
 919         * Younger packs tend to contain more recent objects,
 920         * and more recent objects tend to get accessed more
 921         * often.
 922         */
 923        if (a->mtime < b->mtime)
 924                return 1;
 925        else if (a->mtime == b->mtime)
 926                return 0;
 927        return -1;
 928}
 929
 930static void rearrange_packed_git(void)
 931{
 932        struct packed_git **ary, *p;
 933        int i, n;
 934
 935        for (n = 0, p = packed_git; p; p = p->next)
 936                n++;
 937        if (n < 2)
 938                return;
 939
 940        /* prepare an array of packed_git for easier sorting */
 941        ary = xcalloc(n, sizeof(struct packed_git *));
 942        for (n = 0, p = packed_git; p; p = p->next)
 943                ary[n++] = p;
 944
 945        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 946
 947        /* link them back again */
 948        for (i = 0; i < n - 1; i++)
 949                ary[i]->next = ary[i + 1];
 950        ary[n - 1]->next = NULL;
 951        packed_git = ary[0];
 952
 953        free(ary);
 954}
 955
 956static int prepare_packed_git_run_once = 0;
 957void prepare_packed_git(void)
 958{
 959        struct alternate_object_database *alt;
 960
 961        if (prepare_packed_git_run_once)
 962                return;
 963        prepare_packed_git_one(get_object_directory(), 1);
 964        prepare_alt_odb();
 965        for (alt = alt_odb_list; alt; alt = alt->next) {
 966                alt->name[-1] = 0;
 967                prepare_packed_git_one(alt->base, 0);
 968                alt->name[-1] = '/';
 969        }
 970        rearrange_packed_git();
 971        prepare_packed_git_run_once = 1;
 972}
 973
 974void reprepare_packed_git(void)
 975{
 976        prepare_packed_git_run_once = 0;
 977        prepare_packed_git();
 978}
 979
 980int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 981{
 982        unsigned char real_sha1[20];
 983        hash_sha1_file(map, size, type, real_sha1);
 984        return hashcmp(sha1, real_sha1) ? -1 : 0;
 985}
 986
 987static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 988{
 989        struct stat st;
 990        void *map;
 991        int fd;
 992        char *filename = find_sha1_file(sha1, &st);
 993
 994        if (!filename) {
 995                return NULL;
 996        }
 997
 998        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 999        if (fd < 0) {
1000                /* See if it works without O_NOATIME */
1001                switch (sha1_file_open_flag) {
1002                default:
1003                        fd = open(filename, O_RDONLY);
1004                        if (fd >= 0)
1005                                break;
1006                /* Fallthrough */
1007                case 0:
1008                        return NULL;
1009                }
1010
1011                /* If it failed once, it will probably fail again.
1012                 * Stop using O_NOATIME
1013                 */
1014                sha1_file_open_flag = 0;
1015        }
1016        *size = xsize_t(st.st_size);
1017        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1018        close(fd);
1019        return map;
1020}
1021
1022static int legacy_loose_object(unsigned char *map)
1023{
1024        unsigned int word;
1025
1026        /*
1027         * Is it a zlib-compressed buffer? If so, the first byte
1028         * must be 0x78 (15-bit window size, deflated), and the
1029         * first 16-bit word is evenly divisible by 31
1030         */
1031        word = (map[0] << 8) + map[1];
1032        if (map[0] == 0x78 && !(word % 31))
1033                return 1;
1034        else
1035                return 0;
1036}
1037
1038unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1039{
1040        unsigned shift;
1041        unsigned char c;
1042        unsigned long size;
1043        unsigned long used = 0;
1044
1045        c = buf[used++];
1046        *type = (c >> 4) & 7;
1047        size = c & 15;
1048        shift = 4;
1049        while (c & 0x80) {
1050                if (len <= used)
1051                        return 0;
1052                if (sizeof(long) * 8 <= shift)
1053                        return 0;
1054                c = buf[used++];
1055                size += (c & 0x7f) << shift;
1056                shift += 7;
1057        }
1058        *sizep = size;
1059        return used;
1060}
1061
1062static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1063{
1064        unsigned long size, used;
1065        static const char valid_loose_object_type[8] = {
1066                0, /* OBJ_EXT */
1067                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1068                0, /* "delta" and others are invalid in a loose object */
1069        };
1070        enum object_type type;
1071
1072        /* Get the data stream */
1073        memset(stream, 0, sizeof(*stream));
1074        stream->next_in = map;
1075        stream->avail_in = mapsize;
1076        stream->next_out = buffer;
1077        stream->avail_out = bufsiz;
1078
1079        if (legacy_loose_object(map)) {
1080                inflateInit(stream);
1081                return inflate(stream, 0);
1082        }
1083
1084
1085        /*
1086         * There used to be a second loose object header format which
1087         * was meant to mimic the in-pack format, allowing for direct
1088         * copy of the object data.  This format turned up not to be
1089         * really worth it and we don't write it any longer.  But we
1090         * can still read it.
1091         */
1092        used = unpack_object_header_gently(map, mapsize, &type, &size);
1093        if (!used || !valid_loose_object_type[type])
1094                return -1;
1095        map += used;
1096        mapsize -= used;
1097
1098        /* Set up the stream for the rest.. */
1099        stream->next_in = map;
1100        stream->avail_in = mapsize;
1101        inflateInit(stream);
1102
1103        /* And generate the fake traditional header */
1104        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1105                                         typename(type), size);
1106        return 0;
1107}
1108
1109static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1110{
1111        int bytes = strlen(buffer) + 1;
1112        unsigned char *buf = xmalloc(1+size);
1113        unsigned long n;
1114        int status = Z_OK;
1115
1116        n = stream->total_out - bytes;
1117        if (n > size)
1118                n = size;
1119        memcpy(buf, (char *) buffer + bytes, n);
1120        bytes = n;
1121        if (bytes <= size) {
1122                /*
1123                 * The above condition must be (bytes <= size), not
1124                 * (bytes < size).  In other words, even though we
1125                 * expect no more output and set avail_out to zer0,
1126                 * the input zlib stream may have bytes that express
1127                 * "this concludes the stream", and we *do* want to
1128                 * eat that input.
1129                 *
1130                 * Otherwise we would not be able to test that we
1131                 * consumed all the input to reach the expected size;
1132                 * we also want to check that zlib tells us that all
1133                 * went well with status == Z_STREAM_END at the end.
1134                 */
1135                stream->next_out = buf + bytes;
1136                stream->avail_out = size - bytes;
1137                while (status == Z_OK)
1138                        status = inflate(stream, Z_FINISH);
1139        }
1140        buf[size] = 0;
1141        if (status == Z_STREAM_END && !stream->avail_in) {
1142                inflateEnd(stream);
1143                return buf;
1144        }
1145
1146        if (status < 0)
1147                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1148        else if (stream->avail_in)
1149                error("garbage at end of loose object '%s'",
1150                      sha1_to_hex(sha1));
1151        free(buf);
1152        return NULL;
1153}
1154
1155/*
1156 * We used to just use "sscanf()", but that's actually way
1157 * too permissive for what we want to check. So do an anal
1158 * object header parse by hand.
1159 */
1160static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1161{
1162        char type[10];
1163        int i;
1164        unsigned long size;
1165
1166        /*
1167         * The type can be at most ten bytes (including the
1168         * terminating '\0' that we add), and is followed by
1169         * a space.
1170         */
1171        i = 0;
1172        for (;;) {
1173                char c = *hdr++;
1174                if (c == ' ')
1175                        break;
1176                type[i++] = c;
1177                if (i >= sizeof(type))
1178                        return -1;
1179        }
1180        type[i] = 0;
1181
1182        /*
1183         * The length must follow immediately, and be in canonical
1184         * decimal format (ie "010" is not valid).
1185         */
1186        size = *hdr++ - '0';
1187        if (size > 9)
1188                return -1;
1189        if (size) {
1190                for (;;) {
1191                        unsigned long c = *hdr - '0';
1192                        if (c > 9)
1193                                break;
1194                        hdr++;
1195                        size = size * 10 + c;
1196                }
1197        }
1198        *sizep = size;
1199
1200        /*
1201         * The length must be followed by a zero byte
1202         */
1203        return *hdr ? -1 : type_from_string(type);
1204}
1205
1206static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1207{
1208        int ret;
1209        z_stream stream;
1210        char hdr[8192];
1211
1212        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1213        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1214                return NULL;
1215
1216        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1217}
1218
1219unsigned long get_size_from_delta(struct packed_git *p,
1220                                  struct pack_window **w_curs,
1221                                  off_t curpos)
1222{
1223        const unsigned char *data;
1224        unsigned char delta_head[20], *in;
1225        z_stream stream;
1226        int st;
1227
1228        memset(&stream, 0, sizeof(stream));
1229        stream.next_out = delta_head;
1230        stream.avail_out = sizeof(delta_head);
1231
1232        inflateInit(&stream);
1233        do {
1234                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1235                stream.next_in = in;
1236                st = inflate(&stream, Z_FINISH);
1237                curpos += stream.next_in - in;
1238        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1239                 stream.total_out < sizeof(delta_head));
1240        inflateEnd(&stream);
1241        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1242                die("delta data unpack-initial failed");
1243
1244        /* Examine the initial part of the delta to figure out
1245         * the result size.
1246         */
1247        data = delta_head;
1248
1249        /* ignore base size */
1250        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1251
1252        /* Read the result size */
1253        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1254}
1255
1256static off_t get_delta_base(struct packed_git *p,
1257                                    struct pack_window **w_curs,
1258                                    off_t *curpos,
1259                                    enum object_type type,
1260                                    off_t delta_obj_offset)
1261{
1262        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1263        off_t base_offset;
1264
1265        /* use_pack() assured us we have [base_info, base_info + 20)
1266         * as a range that we can look at without walking off the
1267         * end of the mapped window.  Its actually the hash size
1268         * that is assured.  An OFS_DELTA longer than the hash size
1269         * is stupid, as then a REF_DELTA would be smaller to store.
1270         */
1271        if (type == OBJ_OFS_DELTA) {
1272                unsigned used = 0;
1273                unsigned char c = base_info[used++];
1274                base_offset = c & 127;
1275                while (c & 128) {
1276                        base_offset += 1;
1277                        if (!base_offset || MSB(base_offset, 7))
1278                                die("offset value overflow for delta base object");
1279                        c = base_info[used++];
1280                        base_offset = (base_offset << 7) + (c & 127);
1281                }
1282                base_offset = delta_obj_offset - base_offset;
1283                if (base_offset >= delta_obj_offset)
1284                        die("delta base offset out of bound");
1285                *curpos += used;
1286        } else if (type == OBJ_REF_DELTA) {
1287                /* The base entry _must_ be in the same pack */
1288                base_offset = find_pack_entry_one(base_info, p);
1289                if (!base_offset)
1290                        die("failed to find delta-pack base object %s",
1291                                sha1_to_hex(base_info));
1292                *curpos += 20;
1293        } else
1294                die("I am totally screwed");
1295        return base_offset;
1296}
1297
1298/* forward declaration for a mutually recursive function */
1299static int packed_object_info(struct packed_git *p, off_t offset,
1300                              unsigned long *sizep);
1301
1302static int packed_delta_info(struct packed_git *p,
1303                             struct pack_window **w_curs,
1304                             off_t curpos,
1305                             enum object_type type,
1306                             off_t obj_offset,
1307                             unsigned long *sizep)
1308{
1309        off_t base_offset;
1310
1311        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1312        type = packed_object_info(p, base_offset, NULL);
1313
1314        /* We choose to only get the type of the base object and
1315         * ignore potentially corrupt pack file that expects the delta
1316         * based on a base with a wrong size.  This saves tons of
1317         * inflate() calls.
1318         */
1319        if (sizep)
1320                *sizep = get_size_from_delta(p, w_curs, curpos);
1321
1322        return type;
1323}
1324
1325static int unpack_object_header(struct packed_git *p,
1326                                struct pack_window **w_curs,
1327                                off_t *curpos,
1328                                unsigned long *sizep)
1329{
1330        unsigned char *base;
1331        unsigned int left;
1332        unsigned long used;
1333        enum object_type type;
1334
1335        /* use_pack() assures us we have [base, base + 20) available
1336         * as a range that we can look at at.  (Its actually the hash
1337         * size that is assured.)  With our object header encoding
1338         * the maximum deflated object size is 2^137, which is just
1339         * insane, so we know won't exceed what we have been given.
1340         */
1341        base = use_pack(p, w_curs, *curpos, &left);
1342        used = unpack_object_header_gently(base, left, &type, sizep);
1343        if (!used)
1344                die("object offset outside of pack file");
1345        *curpos += used;
1346
1347        return type;
1348}
1349
1350const char *packed_object_info_detail(struct packed_git *p,
1351                                      off_t obj_offset,
1352                                      unsigned long *size,
1353                                      unsigned long *store_size,
1354                                      unsigned int *delta_chain_length,
1355                                      unsigned char *base_sha1)
1356{
1357        struct pack_window *w_curs = NULL;
1358        off_t curpos;
1359        unsigned long dummy;
1360        unsigned char *next_sha1;
1361        enum object_type type;
1362        struct revindex_entry *revidx;
1363
1364        *delta_chain_length = 0;
1365        curpos = obj_offset;
1366        type = unpack_object_header(p, &w_curs, &curpos, size);
1367
1368        revidx = find_pack_revindex(p, obj_offset);
1369        *store_size = revidx[1].offset - obj_offset;
1370
1371        for (;;) {
1372                switch (type) {
1373                default:
1374                        die("pack %s contains unknown object type %d",
1375                            p->pack_name, type);
1376                case OBJ_COMMIT:
1377                case OBJ_TREE:
1378                case OBJ_BLOB:
1379                case OBJ_TAG:
1380                        unuse_pack(&w_curs);
1381                        return typename(type);
1382                case OBJ_OFS_DELTA:
1383                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1384                        if (*delta_chain_length == 0) {
1385                                revidx = find_pack_revindex(p, obj_offset);
1386                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1387                        }
1388                        break;
1389                case OBJ_REF_DELTA:
1390                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1391                        if (*delta_chain_length == 0)
1392                                hashcpy(base_sha1, next_sha1);
1393                        obj_offset = find_pack_entry_one(next_sha1, p);
1394                        break;
1395                }
1396                (*delta_chain_length)++;
1397                curpos = obj_offset;
1398                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1399        }
1400}
1401
1402static int packed_object_info(struct packed_git *p, off_t obj_offset,
1403                              unsigned long *sizep)
1404{
1405        struct pack_window *w_curs = NULL;
1406        unsigned long size;
1407        off_t curpos = obj_offset;
1408        enum object_type type;
1409
1410        type = unpack_object_header(p, &w_curs, &curpos, &size);
1411
1412        switch (type) {
1413        case OBJ_OFS_DELTA:
1414        case OBJ_REF_DELTA:
1415                type = packed_delta_info(p, &w_curs, curpos,
1416                                         type, obj_offset, sizep);
1417                break;
1418        case OBJ_COMMIT:
1419        case OBJ_TREE:
1420        case OBJ_BLOB:
1421        case OBJ_TAG:
1422                if (sizep)
1423                        *sizep = size;
1424                break;
1425        default:
1426                die("pack %s contains unknown object type %d",
1427                    p->pack_name, type);
1428        }
1429        unuse_pack(&w_curs);
1430        return type;
1431}
1432
1433static void *unpack_compressed_entry(struct packed_git *p,
1434                                    struct pack_window **w_curs,
1435                                    off_t curpos,
1436                                    unsigned long size)
1437{
1438        int st;
1439        z_stream stream;
1440        unsigned char *buffer, *in;
1441
1442        buffer = xmalloc(size + 1);
1443        buffer[size] = 0;
1444        memset(&stream, 0, sizeof(stream));
1445        stream.next_out = buffer;
1446        stream.avail_out = size;
1447
1448        inflateInit(&stream);
1449        do {
1450                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1451                stream.next_in = in;
1452                st = inflate(&stream, Z_FINISH);
1453                curpos += stream.next_in - in;
1454        } while (st == Z_OK || st == Z_BUF_ERROR);
1455        inflateEnd(&stream);
1456        if ((st != Z_STREAM_END) || stream.total_out != size) {
1457                free(buffer);
1458                return NULL;
1459        }
1460
1461        return buffer;
1462}
1463
1464#define MAX_DELTA_CACHE (256)
1465
1466static size_t delta_base_cached;
1467
1468static struct delta_base_cache_lru_list {
1469        struct delta_base_cache_lru_list *prev;
1470        struct delta_base_cache_lru_list *next;
1471} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1472
1473static struct delta_base_cache_entry {
1474        struct delta_base_cache_lru_list lru;
1475        void *data;
1476        struct packed_git *p;
1477        off_t base_offset;
1478        unsigned long size;
1479        enum object_type type;
1480} delta_base_cache[MAX_DELTA_CACHE];
1481
1482static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1483{
1484        unsigned long hash;
1485
1486        hash = (unsigned long)p + (unsigned long)base_offset;
1487        hash += (hash >> 8) + (hash >> 16);
1488        return hash % MAX_DELTA_CACHE;
1489}
1490
1491static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1492        unsigned long *base_size, enum object_type *type, int keep_cache)
1493{
1494        void *ret;
1495        unsigned long hash = pack_entry_hash(p, base_offset);
1496        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1497
1498        ret = ent->data;
1499        if (ret && ent->p == p && ent->base_offset == base_offset)
1500                goto found_cache_entry;
1501        return unpack_entry(p, base_offset, type, base_size);
1502
1503found_cache_entry:
1504        if (!keep_cache) {
1505                ent->data = NULL;
1506                ent->lru.next->prev = ent->lru.prev;
1507                ent->lru.prev->next = ent->lru.next;
1508                delta_base_cached -= ent->size;
1509        } else {
1510                ret = xmemdupz(ent->data, ent->size);
1511        }
1512        *type = ent->type;
1513        *base_size = ent->size;
1514        return ret;
1515}
1516
1517static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1518{
1519        if (ent->data) {
1520                free(ent->data);
1521                ent->data = NULL;
1522                ent->lru.next->prev = ent->lru.prev;
1523                ent->lru.prev->next = ent->lru.next;
1524                delta_base_cached -= ent->size;
1525        }
1526}
1527
1528static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1529        void *base, unsigned long base_size, enum object_type type)
1530{
1531        unsigned long hash = pack_entry_hash(p, base_offset);
1532        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1533        struct delta_base_cache_lru_list *lru;
1534
1535        release_delta_base_cache(ent);
1536        delta_base_cached += base_size;
1537
1538        for (lru = delta_base_cache_lru.next;
1539             delta_base_cached > delta_base_cache_limit
1540             && lru != &delta_base_cache_lru;
1541             lru = lru->next) {
1542                struct delta_base_cache_entry *f = (void *)lru;
1543                if (f->type == OBJ_BLOB)
1544                        release_delta_base_cache(f);
1545        }
1546        for (lru = delta_base_cache_lru.next;
1547             delta_base_cached > delta_base_cache_limit
1548             && lru != &delta_base_cache_lru;
1549             lru = lru->next) {
1550                struct delta_base_cache_entry *f = (void *)lru;
1551                release_delta_base_cache(f);
1552        }
1553
1554        ent->p = p;
1555        ent->base_offset = base_offset;
1556        ent->type = type;
1557        ent->data = base;
1558        ent->size = base_size;
1559        ent->lru.next = &delta_base_cache_lru;
1560        ent->lru.prev = delta_base_cache_lru.prev;
1561        delta_base_cache_lru.prev->next = &ent->lru;
1562        delta_base_cache_lru.prev = &ent->lru;
1563}
1564
1565static void *unpack_delta_entry(struct packed_git *p,
1566                                struct pack_window **w_curs,
1567                                off_t curpos,
1568                                unsigned long delta_size,
1569                                off_t obj_offset,
1570                                enum object_type *type,
1571                                unsigned long *sizep)
1572{
1573        void *delta_data, *result, *base;
1574        unsigned long base_size;
1575        off_t base_offset;
1576
1577        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1578        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1579        if (!base)
1580                die("failed to read delta base object"
1581                    " at %"PRIuMAX" from %s",
1582                    (uintmax_t)base_offset, p->pack_name);
1583
1584        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1585        if (!delta_data)
1586                die("failed to unpack compressed delta"
1587                    " at %"PRIuMAX" from %s",
1588                    (uintmax_t)curpos, p->pack_name);
1589        result = patch_delta(base, base_size,
1590                             delta_data, delta_size,
1591                             sizep);
1592        if (!result)
1593                die("failed to apply delta");
1594        free(delta_data);
1595        add_delta_base_cache(p, base_offset, base, base_size, *type);
1596        return result;
1597}
1598
1599void *unpack_entry(struct packed_git *p, off_t obj_offset,
1600                   enum object_type *type, unsigned long *sizep)
1601{
1602        struct pack_window *w_curs = NULL;
1603        off_t curpos = obj_offset;
1604        void *data;
1605
1606        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1607        switch (*type) {
1608        case OBJ_OFS_DELTA:
1609        case OBJ_REF_DELTA:
1610                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1611                                          obj_offset, type, sizep);
1612                break;
1613        case OBJ_COMMIT:
1614        case OBJ_TREE:
1615        case OBJ_BLOB:
1616        case OBJ_TAG:
1617                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1618                break;
1619        default:
1620                die("unknown object type %i in %s", *type, p->pack_name);
1621        }
1622        unuse_pack(&w_curs);
1623        return data;
1624}
1625
1626const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1627                                            uint32_t n)
1628{
1629        const unsigned char *index = p->index_data;
1630        if (!index) {
1631                if (open_pack_index(p))
1632                        return NULL;
1633                index = p->index_data;
1634        }
1635        if (n >= p->num_objects)
1636                return NULL;
1637        index += 4 * 256;
1638        if (p->index_version == 1) {
1639                return index + 24 * n + 4;
1640        } else {
1641                index += 8;
1642                return index + 20 * n;
1643        }
1644}
1645
1646static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1647{
1648        const unsigned char *index = p->index_data;
1649        index += 4 * 256;
1650        if (p->index_version == 1) {
1651                return ntohl(*((uint32_t *)(index + 24 * n)));
1652        } else {
1653                uint32_t off;
1654                index += 8 + p->num_objects * (20 + 4);
1655                off = ntohl(*((uint32_t *)(index + 4 * n)));
1656                if (!(off & 0x80000000))
1657                        return off;
1658                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1659                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1660                                   ntohl(*((uint32_t *)(index + 4)));
1661        }
1662}
1663
1664off_t find_pack_entry_one(const unsigned char *sha1,
1665                                  struct packed_git *p)
1666{
1667        const uint32_t *level1_ofs = p->index_data;
1668        const unsigned char *index = p->index_data;
1669        unsigned hi, lo, stride;
1670        static int use_lookup = -1;
1671        static int debug_lookup = -1;
1672
1673        if (debug_lookup < 0)
1674                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1675
1676        if (!index) {
1677                if (open_pack_index(p))
1678                        return 0;
1679                level1_ofs = p->index_data;
1680                index = p->index_data;
1681        }
1682        if (p->index_version > 1) {
1683                level1_ofs += 2;
1684                index += 8;
1685        }
1686        index += 4 * 256;
1687        hi = ntohl(level1_ofs[*sha1]);
1688        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1689        if (p->index_version > 1) {
1690                stride = 20;
1691        } else {
1692                stride = 24;
1693                index += 4;
1694        }
1695
1696        if (debug_lookup)
1697                printf("%02x%02x%02x... lo %u hi %u nr %u\n",
1698                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1699
1700        if (use_lookup < 0)
1701                use_lookup = !!getenv("GIT_USE_LOOKUP");
1702        if (use_lookup) {
1703                int pos = sha1_entry_pos(index, stride, 0,
1704                                         lo, hi, p->num_objects, sha1);
1705                if (pos < 0)
1706                        return 0;
1707                return nth_packed_object_offset(p, pos);
1708        }
1709
1710        do {
1711                unsigned mi = (lo + hi) / 2;
1712                int cmp = hashcmp(index + mi * stride, sha1);
1713
1714                if (debug_lookup)
1715                        printf("lo %u hi %u rg %u mi %u\n",
1716                               lo, hi, hi - lo, mi);
1717                if (!cmp)
1718                        return nth_packed_object_offset(p, mi);
1719                if (cmp > 0)
1720                        hi = mi;
1721                else
1722                        lo = mi+1;
1723        } while (lo < hi);
1724        return 0;
1725}
1726
1727int matches_pack_name(struct packed_git *p, const char *name)
1728{
1729        const char *last_c, *c;
1730
1731        if (!strcmp(p->pack_name, name))
1732                return 1;
1733
1734        for (c = p->pack_name, last_c = c; *c;)
1735                if (*c == '/')
1736                        last_c = ++c;
1737                else
1738                        ++c;
1739        if (!strcmp(last_c, name))
1740                return 1;
1741
1742        return 0;
1743}
1744
1745static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1746{
1747        static struct packed_git *last_found = (void *)1;
1748        struct packed_git *p;
1749        off_t offset;
1750
1751        prepare_packed_git();
1752        if (!packed_git)
1753                return 0;
1754        p = (last_found == (void *)1) ? packed_git : last_found;
1755
1756        do {
1757                if (ignore_packed) {
1758                        const char **ig;
1759                        for (ig = ignore_packed; *ig; ig++)
1760                                if (matches_pack_name(p, *ig))
1761                                        break;
1762                        if (*ig)
1763                                goto next;
1764                }
1765
1766                offset = find_pack_entry_one(sha1, p);
1767                if (offset) {
1768                        /*
1769                         * We are about to tell the caller where they can
1770                         * locate the requested object.  We better make
1771                         * sure the packfile is still here and can be
1772                         * accessed before supplying that answer, as
1773                         * it may have been deleted since the index
1774                         * was loaded!
1775                         */
1776                        if (p->pack_fd == -1 && open_packed_git(p)) {
1777                                error("packfile %s cannot be accessed", p->pack_name);
1778                                goto next;
1779                        }
1780                        e->offset = offset;
1781                        e->p = p;
1782                        hashcpy(e->sha1, sha1);
1783                        last_found = p;
1784                        return 1;
1785                }
1786
1787                next:
1788                if (p == last_found)
1789                        p = packed_git;
1790                else
1791                        p = p->next;
1792                if (p == last_found)
1793                        p = p->next;
1794        } while (p);
1795        return 0;
1796}
1797
1798struct packed_git *find_sha1_pack(const unsigned char *sha1,
1799                                  struct packed_git *packs)
1800{
1801        struct packed_git *p;
1802
1803        for (p = packs; p; p = p->next) {
1804                if (find_pack_entry_one(sha1, p))
1805                        return p;
1806        }
1807        return NULL;
1808
1809}
1810
1811static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1812{
1813        int status;
1814        unsigned long mapsize, size;
1815        void *map;
1816        z_stream stream;
1817        char hdr[32];
1818
1819        map = map_sha1_file(sha1, &mapsize);
1820        if (!map)
1821                return error("unable to find %s", sha1_to_hex(sha1));
1822        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1823                status = error("unable to unpack %s header",
1824                               sha1_to_hex(sha1));
1825        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1826                status = error("unable to parse %s header", sha1_to_hex(sha1));
1827        else if (sizep)
1828                *sizep = size;
1829        inflateEnd(&stream);
1830        munmap(map, mapsize);
1831        return status;
1832}
1833
1834int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1835{
1836        struct pack_entry e;
1837
1838        if (!find_pack_entry(sha1, &e, NULL)) {
1839                reprepare_packed_git();
1840                if (!find_pack_entry(sha1, &e, NULL))
1841                        return sha1_loose_object_info(sha1, sizep);
1842        }
1843        return packed_object_info(e.p, e.offset, sizep);
1844}
1845
1846static void *read_packed_sha1(const unsigned char *sha1,
1847                              enum object_type *type, unsigned long *size)
1848{
1849        struct pack_entry e;
1850
1851        if (!find_pack_entry(sha1, &e, NULL))
1852                return NULL;
1853        else
1854                return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1855}
1856
1857/*
1858 * This is meant to hold a *small* number of objects that you would
1859 * want read_sha1_file() to be able to return, but yet you do not want
1860 * to write them into the object store (e.g. a browse-only
1861 * application).
1862 */
1863static struct cached_object {
1864        unsigned char sha1[20];
1865        enum object_type type;
1866        void *buf;
1867        unsigned long size;
1868} *cached_objects;
1869static int cached_object_nr, cached_object_alloc;
1870
1871static struct cached_object empty_tree = {
1872        /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1873        "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1874        "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
1875        OBJ_TREE,
1876        "",
1877        0
1878};
1879
1880static struct cached_object *find_cached_object(const unsigned char *sha1)
1881{
1882        int i;
1883        struct cached_object *co = cached_objects;
1884
1885        for (i = 0; i < cached_object_nr; i++, co++) {
1886                if (!hashcmp(co->sha1, sha1))
1887                        return co;
1888        }
1889        if (!hashcmp(sha1, empty_tree.sha1))
1890                return &empty_tree;
1891        return NULL;
1892}
1893
1894int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1895                      unsigned char *sha1)
1896{
1897        struct cached_object *co;
1898
1899        hash_sha1_file(buf, len, typename(type), sha1);
1900        if (has_sha1_file(sha1) || find_cached_object(sha1))
1901                return 0;
1902        if (cached_object_alloc <= cached_object_nr) {
1903                cached_object_alloc = alloc_nr(cached_object_alloc);
1904                cached_objects = xrealloc(cached_objects,
1905                                          sizeof(*cached_objects) *
1906                                          cached_object_alloc);
1907        }
1908        co = &cached_objects[cached_object_nr++];
1909        co->size = len;
1910        co->type = type;
1911        co->buf = xmalloc(len);
1912        memcpy(co->buf, buf, len);
1913        hashcpy(co->sha1, sha1);
1914        return 0;
1915}
1916
1917void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1918                     unsigned long *size)
1919{
1920        unsigned long mapsize;
1921        void *map, *buf;
1922        struct cached_object *co;
1923
1924        co = find_cached_object(sha1);
1925        if (co) {
1926                *type = co->type;
1927                *size = co->size;
1928                return xmemdupz(co->buf, co->size);
1929        }
1930
1931        buf = read_packed_sha1(sha1, type, size);
1932        if (buf)
1933                return buf;
1934        map = map_sha1_file(sha1, &mapsize);
1935        if (map) {
1936                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1937                munmap(map, mapsize);
1938                return buf;
1939        }
1940        reprepare_packed_git();
1941        return read_packed_sha1(sha1, type, size);
1942}
1943
1944void *read_object_with_reference(const unsigned char *sha1,
1945                                 const char *required_type_name,
1946                                 unsigned long *size,
1947                                 unsigned char *actual_sha1_return)
1948{
1949        enum object_type type, required_type;
1950        void *buffer;
1951        unsigned long isize;
1952        unsigned char actual_sha1[20];
1953
1954        required_type = type_from_string(required_type_name);
1955        hashcpy(actual_sha1, sha1);
1956        while (1) {
1957                int ref_length = -1;
1958                const char *ref_type = NULL;
1959
1960                buffer = read_sha1_file(actual_sha1, &type, &isize);
1961                if (!buffer)
1962                        return NULL;
1963                if (type == required_type) {
1964                        *size = isize;
1965                        if (actual_sha1_return)
1966                                hashcpy(actual_sha1_return, actual_sha1);
1967                        return buffer;
1968                }
1969                /* Handle references */
1970                else if (type == OBJ_COMMIT)
1971                        ref_type = "tree ";
1972                else if (type == OBJ_TAG)
1973                        ref_type = "object ";
1974                else {
1975                        free(buffer);
1976                        return NULL;
1977                }
1978                ref_length = strlen(ref_type);
1979
1980                if (ref_length + 40 > isize ||
1981                    memcmp(buffer, ref_type, ref_length) ||
1982                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1983                        free(buffer);
1984                        return NULL;
1985                }
1986                free(buffer);
1987                /* Now we have the ID of the referred-to object in
1988                 * actual_sha1.  Check again. */
1989        }
1990}
1991
1992static void write_sha1_file_prepare(const void *buf, unsigned long len,
1993                                    const char *type, unsigned char *sha1,
1994                                    char *hdr, int *hdrlen)
1995{
1996        SHA_CTX c;
1997
1998        /* Generate the header */
1999        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2000
2001        /* Sha1.. */
2002        SHA1_Init(&c);
2003        SHA1_Update(&c, hdr, *hdrlen);
2004        SHA1_Update(&c, buf, len);
2005        SHA1_Final(sha1, &c);
2006}
2007
2008/*
2009 * Link the tempfile to the final place, possibly creating the
2010 * last directory level as you do so.
2011 *
2012 * Returns the errno on failure, 0 on success.
2013 */
2014static int link_temp_to_file(const char *tmpfile, const char *filename)
2015{
2016        int ret;
2017        char *dir;
2018
2019        if (!link(tmpfile, filename))
2020                return 0;
2021
2022        /*
2023         * Try to mkdir the last path component if that failed.
2024         *
2025         * Re-try the "link()" regardless of whether the mkdir
2026         * succeeds, since a race might mean that somebody
2027         * else succeeded.
2028         */
2029        ret = errno;
2030        dir = strrchr(filename, '/');
2031        if (dir) {
2032                *dir = 0;
2033                if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
2034                        *dir = '/';
2035                        return -2;
2036                }
2037                *dir = '/';
2038                if (!link(tmpfile, filename))
2039                        return 0;
2040                ret = errno;
2041        }
2042        return ret;
2043}
2044
2045/*
2046 * Move the just written object into its final resting place
2047 */
2048int move_temp_to_file(const char *tmpfile, const char *filename)
2049{
2050        int ret = link_temp_to_file(tmpfile, filename);
2051
2052        /*
2053         * Coda hack - coda doesn't like cross-directory links,
2054         * so we fall back to a rename, which will mean that it
2055         * won't be able to check collisions, but that's not a
2056         * big deal.
2057         *
2058         * The same holds for FAT formatted media.
2059         *
2060         * When this succeeds, we just return 0. We have nothing
2061         * left to unlink.
2062         */
2063        if (ret && ret != EEXIST) {
2064                if (!rename(tmpfile, filename))
2065                        return 0;
2066                ret = errno;
2067        }
2068        unlink(tmpfile);
2069        if (ret) {
2070                if (ret != EEXIST) {
2071                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2072                }
2073                /* FIXME!!! Collision check here ? */
2074        }
2075
2076        return 0;
2077}
2078
2079static int write_buffer(int fd, const void *buf, size_t len)
2080{
2081        if (write_in_full(fd, buf, len) < 0)
2082                return error("file write error (%s)", strerror(errno));
2083        return 0;
2084}
2085
2086int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2087                   unsigned char *sha1)
2088{
2089        char hdr[32];
2090        int hdrlen;
2091        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2092        return 0;
2093}
2094
2095static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2096                              void *buf, unsigned long len, time_t mtime)
2097{
2098        int fd, size, ret;
2099        unsigned char *compressed;
2100        z_stream stream;
2101        char *filename;
2102        static char tmpfile[PATH_MAX];
2103
2104        filename = sha1_file_name(sha1);
2105        fd = open(filename, O_RDONLY);
2106        if (fd >= 0) {
2107                /*
2108                 * FIXME!!! We might do collision checking here, but we'd
2109                 * need to uncompress the old file and check it. Later.
2110                 */
2111                close(fd);
2112                return 0;
2113        }
2114
2115        if (errno != ENOENT) {
2116                return error("sha1 file %s: %s\n", filename, strerror(errno));
2117        }
2118
2119        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2120
2121        fd = mkstemp(tmpfile);
2122        if (fd < 0) {
2123                if (errno == EPERM)
2124                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2125                else
2126                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2127        }
2128
2129        /* Set it up */
2130        memset(&stream, 0, sizeof(stream));
2131        deflateInit(&stream, zlib_compression_level);
2132        size = 8 + deflateBound(&stream, len+hdrlen);
2133        compressed = xmalloc(size);
2134
2135        /* Compress it */
2136        stream.next_out = compressed;
2137        stream.avail_out = size;
2138
2139        /* First header.. */
2140        stream.next_in = (unsigned char *)hdr;
2141        stream.avail_in = hdrlen;
2142        while (deflate(&stream, 0) == Z_OK)
2143                /* nothing */;
2144
2145        /* Then the data itself.. */
2146        stream.next_in = buf;
2147        stream.avail_in = len;
2148        ret = deflate(&stream, Z_FINISH);
2149        if (ret != Z_STREAM_END)
2150                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2151
2152        ret = deflateEnd(&stream);
2153        if (ret != Z_OK)
2154                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2155
2156        size = stream.total_out;
2157
2158        if (write_buffer(fd, compressed, size) < 0)
2159                die("unable to write sha1 file");
2160        fchmod(fd, 0444);
2161        if (close(fd))
2162                die("unable to write sha1 file");
2163        free(compressed);
2164
2165        if (mtime) {
2166                struct utimbuf utb;
2167                utb.actime = mtime;
2168                utb.modtime = mtime;
2169                if (utime(tmpfile, &utb) < 0)
2170                        warning("failed utime() on %s: %s",
2171                                tmpfile, strerror(errno));
2172        }
2173
2174        return move_temp_to_file(tmpfile, filename);
2175}
2176
2177int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2178{
2179        unsigned char sha1[20];
2180        char hdr[32];
2181        int hdrlen;
2182
2183        /* Normally if we have it in the pack then we do not bother writing
2184         * it out into .git/objects/??/?{38} file.
2185         */
2186        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2187        if (returnsha1)
2188                hashcpy(returnsha1, sha1);
2189        if (has_sha1_file(sha1))
2190                return 0;
2191        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2192}
2193
2194int force_object_loose(const unsigned char *sha1, time_t mtime)
2195{
2196        struct stat st;
2197        void *buf;
2198        unsigned long len;
2199        enum object_type type;
2200        char hdr[32];
2201        int hdrlen;
2202
2203        if (find_sha1_file(sha1, &st))
2204                return 0;
2205        buf = read_packed_sha1(sha1, &type, &len);
2206        if (!buf)
2207                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2208        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2209        return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2210}
2211
2212/*
2213 * We need to unpack and recompress the object for writing
2214 * it out to a different file.
2215 */
2216static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2217{
2218        size_t size;
2219        z_stream stream;
2220        unsigned char *unpacked;
2221        unsigned long len;
2222        enum object_type type;
2223        char hdr[32];
2224        int hdrlen;
2225        void *buf;
2226
2227        /* need to unpack and recompress it by itself */
2228        unpacked = read_packed_sha1(sha1, &type, &len);
2229        if (!unpacked)
2230                error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2231
2232        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2233
2234        /* Set it up */
2235        memset(&stream, 0, sizeof(stream));
2236        deflateInit(&stream, zlib_compression_level);
2237        size = deflateBound(&stream, len + hdrlen);
2238        buf = xmalloc(size);
2239
2240        /* Compress it */
2241        stream.next_out = buf;
2242        stream.avail_out = size;
2243
2244        /* First header.. */
2245        stream.next_in = (void *)hdr;
2246        stream.avail_in = hdrlen;
2247        while (deflate(&stream, 0) == Z_OK)
2248                /* nothing */;
2249
2250        /* Then the data itself.. */
2251        stream.next_in = unpacked;
2252        stream.avail_in = len;
2253        while (deflate(&stream, Z_FINISH) == Z_OK)
2254                /* nothing */;
2255        deflateEnd(&stream);
2256        free(unpacked);
2257
2258        *objsize = stream.total_out;
2259        return buf;
2260}
2261
2262int write_sha1_to_fd(int fd, const unsigned char *sha1)
2263{
2264        int retval;
2265        unsigned long objsize;
2266        void *buf = map_sha1_file(sha1, &objsize);
2267
2268        if (buf) {
2269                retval = write_buffer(fd, buf, objsize);
2270                munmap(buf, objsize);
2271                return retval;
2272        }
2273
2274        buf = repack_object(sha1, &objsize);
2275        retval = write_buffer(fd, buf, objsize);
2276        free(buf);
2277        return retval;
2278}
2279
2280int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2281                       size_t bufsize, size_t *bufposn)
2282{
2283        char tmpfile[PATH_MAX];
2284        int local;
2285        z_stream stream;
2286        unsigned char real_sha1[20];
2287        unsigned char discard[4096];
2288        int ret;
2289        SHA_CTX c;
2290
2291        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2292
2293        local = mkstemp(tmpfile);
2294        if (local < 0) {
2295                if (errno == EPERM)
2296                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2297                else
2298                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2299        }
2300
2301        memset(&stream, 0, sizeof(stream));
2302
2303        inflateInit(&stream);
2304
2305        SHA1_Init(&c);
2306
2307        do {
2308                ssize_t size;
2309                if (*bufposn) {
2310                        stream.avail_in = *bufposn;
2311                        stream.next_in = (unsigned char *) buffer;
2312                        do {
2313                                stream.next_out = discard;
2314                                stream.avail_out = sizeof(discard);
2315                                ret = inflate(&stream, Z_SYNC_FLUSH);
2316                                SHA1_Update(&c, discard, sizeof(discard) -
2317                                            stream.avail_out);
2318                        } while (stream.avail_in && ret == Z_OK);
2319                        if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2320                                die("unable to write sha1 file");
2321                        memmove(buffer, buffer + *bufposn - stream.avail_in,
2322                                stream.avail_in);
2323                        *bufposn = stream.avail_in;
2324                        if (ret != Z_OK)
2325                                break;
2326                }
2327                size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2328                if (size <= 0) {
2329                        close(local);
2330                        unlink(tmpfile);
2331                        if (!size)
2332                                return error("Connection closed?");
2333                        perror("Reading from connection");
2334                        return -1;
2335                }
2336                *bufposn += size;
2337        } while (1);
2338        inflateEnd(&stream);
2339
2340        fchmod(local, 0444);
2341        if (close(local) != 0)
2342                die("unable to write sha1 file");
2343        SHA1_Final(real_sha1, &c);
2344        if (ret != Z_STREAM_END) {
2345                unlink(tmpfile);
2346                return error("File %s corrupted", sha1_to_hex(sha1));
2347        }
2348        if (hashcmp(sha1, real_sha1)) {
2349                unlink(tmpfile);
2350                return error("File %s has bad hash", sha1_to_hex(sha1));
2351        }
2352
2353        return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2354}
2355
2356int has_pack_index(const unsigned char *sha1)
2357{
2358        struct stat st;
2359        if (stat(sha1_pack_index_name(sha1), &st))
2360                return 0;
2361        return 1;
2362}
2363
2364int has_pack_file(const unsigned char *sha1)
2365{
2366        struct stat st;
2367        if (stat(sha1_pack_name(sha1), &st))
2368                return 0;
2369        return 1;
2370}
2371
2372int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2373{
2374        struct pack_entry e;
2375        return find_pack_entry(sha1, &e, ignore_packed);
2376}
2377
2378int has_sha1_file(const unsigned char *sha1)
2379{
2380        struct stat st;
2381        struct pack_entry e;
2382
2383        if (find_pack_entry(sha1, &e, NULL))
2384                return 1;
2385        return find_sha1_file(sha1, &st) ? 1 : 0;
2386}
2387
2388int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2389{
2390        struct strbuf buf;
2391        int ret;
2392
2393        strbuf_init(&buf, 0);
2394        if (strbuf_read(&buf, fd, 4096) < 0) {
2395                strbuf_release(&buf);
2396                return -1;
2397        }
2398
2399        if (!type)
2400                type = blob_type;
2401        if (write_object)
2402                ret = write_sha1_file(buf.buf, buf.len, type, sha1);
2403        else
2404                ret = hash_sha1_file(buf.buf, buf.len, type, sha1);
2405        strbuf_release(&buf);
2406
2407        return ret;
2408}
2409
2410int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2411             enum object_type type, const char *path)
2412{
2413        size_t size = xsize_t(st->st_size);
2414        void *buf = NULL;
2415        int ret, re_allocated = 0;
2416
2417        if (size)
2418                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2419        close(fd);
2420
2421        if (!type)
2422                type = OBJ_BLOB;
2423
2424        /*
2425         * Convert blobs to git internal format
2426         */
2427        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2428                struct strbuf nbuf;
2429                strbuf_init(&nbuf, 0);
2430                if (convert_to_git(path, buf, size, &nbuf,
2431                                   write_object ? safe_crlf : 0)) {
2432                        munmap(buf, size);
2433                        buf = strbuf_detach(&nbuf, &size);
2434                        re_allocated = 1;
2435                }
2436        }
2437
2438        if (write_object)
2439                ret = write_sha1_file(buf, size, typename(type), sha1);
2440        else
2441                ret = hash_sha1_file(buf, size, typename(type), sha1);
2442        if (re_allocated) {
2443                free(buf);
2444                return ret;
2445        }
2446        if (size)
2447                munmap(buf, size);
2448        return ret;
2449}
2450
2451int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2452{
2453        int fd;
2454        char *target;
2455        size_t len;
2456
2457        switch (st->st_mode & S_IFMT) {
2458        case S_IFREG:
2459                fd = open(path, O_RDONLY);
2460                if (fd < 0)
2461                        return error("open(\"%s\"): %s", path,
2462                                     strerror(errno));
2463                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2464                        return error("%s: failed to insert into database",
2465                                     path);
2466                break;
2467        case S_IFLNK:
2468                len = xsize_t(st->st_size);
2469                target = xmalloc(len + 1);
2470                if (readlink(path, target, len + 1) != st->st_size) {
2471                        char *errstr = strerror(errno);
2472                        free(target);
2473                        return error("readlink(\"%s\"): %s", path,
2474                                     errstr);
2475                }
2476                if (!write_object)
2477                        hash_sha1_file(target, len, blob_type, sha1);
2478                else if (write_sha1_file(target, len, blob_type, sha1))
2479                        return error("%s: failed to insert into database",
2480                                     path);
2481                free(target);
2482                break;
2483        case S_IFDIR:
2484                return resolve_gitlink_ref(path, "HEAD", sha1);
2485        default:
2486                return error("%s: unsupported file type", path);
2487        }
2488        return 0;
2489}
2490
2491int read_pack_header(int fd, struct pack_header *header)
2492{
2493        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2494                /* "eof before pack header was fully read" */
2495                return PH_ERROR_EOF;
2496
2497        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2498                /* "protocol error (pack signature mismatch detected)" */
2499                return PH_ERROR_PACK_SIGNATURE;
2500        if (!pack_version_ok(header->hdr_version))
2501                /* "protocol error (pack version unsupported)" */
2502                return PH_ERROR_PROTOCOL;
2503        return 0;
2504}