sha1_file.con commit Optimized cvsexportcommit: calling 'cvs status' once instead of once per touched file. (c56f0d9)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This handles basic git sha1 object files - packing, unpacking,
   7 * creation etc.
   8 */
   9#include "cache.h"
  10#include "delta.h"
  11#include "pack.h"
  12#include "blob.h"
  13#include "commit.h"
  14#include "tag.h"
  15#include "tree.h"
  16#include "refs.h"
  17
  18#ifndef O_NOATIME
  19#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  20#define O_NOATIME 01000000
  21#else
  22#define O_NOATIME 0
  23#endif
  24#endif
  25
  26#ifdef NO_C99_FORMAT
  27#define SZ_FMT "lu"
  28#else
  29#define SZ_FMT "zu"
  30#endif
  31
  32const unsigned char null_sha1[20];
  33
  34static unsigned int sha1_file_open_flag = O_NOATIME;
  35
  36signed char hexval_table[256] = {
  37         -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
  38         -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
  39         -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
  40         -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
  41         -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
  42         -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
  43          0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
  44          8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
  45         -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
  46         -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
  47         -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
  48         -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
  49         -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
  50         -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
  51         -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
  52         -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
  53         -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
  54         -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
  55         -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
  56         -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
  57         -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
  58         -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
  59         -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
  60         -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
  61         -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
  62         -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
  63         -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
  64         -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
  65         -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
  66         -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
  67         -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
  68         -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
  69};
  70
  71int get_sha1_hex(const char *hex, unsigned char *sha1)
  72{
  73        int i;
  74        for (i = 0; i < 20; i++) {
  75                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  76                if (val & ~0xff)
  77                        return -1;
  78                *sha1++ = val;
  79                hex += 2;
  80        }
  81        return 0;
  82}
  83
  84int safe_create_leading_directories(char *path)
  85{
  86        char *pos = path;
  87        struct stat st;
  88
  89        if (*pos == '/')
  90                pos++;
  91
  92        while (pos) {
  93                pos = strchr(pos, '/');
  94                if (!pos)
  95                        break;
  96                *pos = 0;
  97                if (!stat(path, &st)) {
  98                        /* path exists */
  99                        if (!S_ISDIR(st.st_mode)) {
 100                                *pos = '/';
 101                                return -3;
 102                        }
 103                }
 104                else if (mkdir(path, 0777)) {
 105                        *pos = '/';
 106                        return -1;
 107                }
 108                else if (adjust_shared_perm(path)) {
 109                        *pos = '/';
 110                        return -2;
 111                }
 112                *pos++ = '/';
 113        }
 114        return 0;
 115}
 116
 117char * sha1_to_hex(const unsigned char *sha1)
 118{
 119        static int bufno;
 120        static char hexbuffer[4][50];
 121        static const char hex[] = "0123456789abcdef";
 122        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 123        int i;
 124
 125        for (i = 0; i < 20; i++) {
 126                unsigned int val = *sha1++;
 127                *buf++ = hex[val >> 4];
 128                *buf++ = hex[val & 0xf];
 129        }
 130        *buf = '\0';
 131
 132        return buffer;
 133}
 134
 135static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 136{
 137        int i;
 138        for (i = 0; i < 20; i++) {
 139                static char hex[] = "0123456789abcdef";
 140                unsigned int val = sha1[i];
 141                char *pos = pathbuf + i*2 + (i > 0);
 142                *pos++ = hex[val >> 4];
 143                *pos = hex[val & 0xf];
 144        }
 145}
 146
 147/*
 148 * NOTE! This returns a statically allocated buffer, so you have to be
 149 * careful about using it. Do a "xstrdup()" if you need to save the
 150 * filename.
 151 *
 152 * Also note that this returns the location for creating.  Reading
 153 * SHA1 file can happen from any alternate directory listed in the
 154 * DB_ENVIRONMENT environment variable if it is not found in
 155 * the primary object database.
 156 */
 157char *sha1_file_name(const unsigned char *sha1)
 158{
 159        static char *name, *base;
 160
 161        if (!base) {
 162                const char *sha1_file_directory = get_object_directory();
 163                int len = strlen(sha1_file_directory);
 164                base = xmalloc(len + 60);
 165                memcpy(base, sha1_file_directory, len);
 166                memset(base+len, 0, 60);
 167                base[len] = '/';
 168                base[len+3] = '/';
 169                name = base + len + 1;
 170        }
 171        fill_sha1_path(name, sha1);
 172        return base;
 173}
 174
 175char *sha1_pack_name(const unsigned char *sha1)
 176{
 177        static const char hex[] = "0123456789abcdef";
 178        static char *name, *base, *buf;
 179        int i;
 180
 181        if (!base) {
 182                const char *sha1_file_directory = get_object_directory();
 183                int len = strlen(sha1_file_directory);
 184                base = xmalloc(len + 60);
 185                sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
 186                name = base + len + 11;
 187        }
 188
 189        buf = name;
 190
 191        for (i = 0; i < 20; i++) {
 192                unsigned int val = *sha1++;
 193                *buf++ = hex[val >> 4];
 194                *buf++ = hex[val & 0xf];
 195        }
 196        
 197        return base;
 198}
 199
 200char *sha1_pack_index_name(const unsigned char *sha1)
 201{
 202        static const char hex[] = "0123456789abcdef";
 203        static char *name, *base, *buf;
 204        int i;
 205
 206        if (!base) {
 207                const char *sha1_file_directory = get_object_directory();
 208                int len = strlen(sha1_file_directory);
 209                base = xmalloc(len + 60);
 210                sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
 211                name = base + len + 11;
 212        }
 213
 214        buf = name;
 215
 216        for (i = 0; i < 20; i++) {
 217                unsigned int val = *sha1++;
 218                *buf++ = hex[val >> 4];
 219                *buf++ = hex[val & 0xf];
 220        }
 221        
 222        return base;
 223}
 224
 225struct alternate_object_database *alt_odb_list;
 226static struct alternate_object_database **alt_odb_tail;
 227
 228static void read_info_alternates(const char * alternates, int depth);
 229
 230/*
 231 * Prepare alternate object database registry.
 232 *
 233 * The variable alt_odb_list points at the list of struct
 234 * alternate_object_database.  The elements on this list come from
 235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 237 * whose contents is similar to that environment variable but can be
 238 * LF separated.  Its base points at a statically allocated buffer that
 239 * contains "/the/directory/corresponding/to/.git/objects/...", while
 240 * its name points just after the slash at the end of ".git/objects/"
 241 * in the example above, and has enough space to hold 40-byte hex
 242 * SHA1, an extra slash for the first level indirection, and the
 243 * terminating NUL.
 244 */
 245static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 246{
 247        struct stat st;
 248        const char *objdir = get_object_directory();
 249        struct alternate_object_database *ent;
 250        struct alternate_object_database *alt;
 251        /* 43 = 40-byte + 2 '/' + terminating NUL */
 252        int pfxlen = len;
 253        int entlen = pfxlen + 43;
 254        int base_len = -1;
 255
 256        if (*entry != '/' && relative_base) {
 257                /* Relative alt-odb */
 258                if (base_len < 0)
 259                        base_len = strlen(relative_base) + 1;
 260                entlen += base_len;
 261                pfxlen += base_len;
 262        }
 263        ent = xmalloc(sizeof(*ent) + entlen);
 264
 265        if (*entry != '/' && relative_base) {
 266                memcpy(ent->base, relative_base, base_len - 1);
 267                ent->base[base_len - 1] = '/';
 268                memcpy(ent->base + base_len, entry, len);
 269        }
 270        else
 271                memcpy(ent->base, entry, pfxlen);
 272
 273        ent->name = ent->base + pfxlen + 1;
 274        ent->base[pfxlen + 3] = '/';
 275        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 276
 277        /* Detect cases where alternate disappeared */
 278        if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
 279                error("object directory %s does not exist; "
 280                      "check .git/objects/info/alternates.",
 281                      ent->base);
 282                free(ent);
 283                return -1;
 284        }
 285
 286        /* Prevent the common mistake of listing the same
 287         * thing twice, or object directory itself.
 288         */
 289        for (alt = alt_odb_list; alt; alt = alt->next) {
 290                if (!memcmp(ent->base, alt->base, pfxlen)) {
 291                        free(ent);
 292                        return -1;
 293                }
 294        }
 295        if (!memcmp(ent->base, objdir, pfxlen)) {
 296                free(ent);
 297                return -1;
 298        }
 299
 300        /* add the alternate entry */
 301        *alt_odb_tail = ent;
 302        alt_odb_tail = &(ent->next);
 303        ent->next = NULL;
 304
 305        /* recursively add alternates */
 306        read_info_alternates(ent->base, depth + 1);
 307
 308        ent->base[pfxlen] = '/';
 309
 310        return 0;
 311}
 312
 313static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 314                                 const char *relative_base, int depth)
 315{
 316        const char *cp, *last;
 317
 318        if (depth > 5) {
 319                error("%s: ignoring alternate object stores, nesting too deep.",
 320                                relative_base);
 321                return;
 322        }
 323
 324        last = alt;
 325        while (last < ep) {
 326                cp = last;
 327                if (cp < ep && *cp == '#') {
 328                        while (cp < ep && *cp != sep)
 329                                cp++;
 330                        last = cp + 1;
 331                        continue;
 332                }
 333                while (cp < ep && *cp != sep)
 334                        cp++;
 335                if (last != cp) {
 336                        if ((*last != '/') && depth) {
 337                                error("%s: ignoring relative alternate object store %s",
 338                                                relative_base, last);
 339                        } else {
 340                                link_alt_odb_entry(last, cp - last,
 341                                                relative_base, depth);
 342                        }
 343                }
 344                while (cp < ep && *cp == sep)
 345                        cp++;
 346                last = cp;
 347        }
 348}
 349
 350static void read_info_alternates(const char * relative_base, int depth)
 351{
 352        char *map;
 353        size_t mapsz;
 354        struct stat st;
 355        char path[PATH_MAX];
 356        int fd;
 357
 358        sprintf(path, "%s/info/alternates", relative_base);
 359        fd = open(path, O_RDONLY);
 360        if (fd < 0)
 361                return;
 362        if (fstat(fd, &st) || (st.st_size == 0)) {
 363                close(fd);
 364                return;
 365        }
 366        mapsz = xsize_t(st.st_size);
 367        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 368        close(fd);
 369
 370        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 371
 372        munmap(map, mapsz);
 373}
 374
 375void prepare_alt_odb(void)
 376{
 377        const char *alt;
 378
 379        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 380        if (!alt) alt = "";
 381
 382        if (alt_odb_tail)
 383                return;
 384        alt_odb_tail = &alt_odb_list;
 385        link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
 386
 387        read_info_alternates(get_object_directory(), 0);
 388}
 389
 390static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 391{
 392        char *name = sha1_file_name(sha1);
 393        struct alternate_object_database *alt;
 394
 395        if (!stat(name, st))
 396                return name;
 397        prepare_alt_odb();
 398        for (alt = alt_odb_list; alt; alt = alt->next) {
 399                name = alt->name;
 400                fill_sha1_path(name, sha1);
 401                if (!stat(alt->base, st))
 402                        return alt->base;
 403        }
 404        return NULL;
 405}
 406
 407static unsigned int pack_used_ctr;
 408static unsigned int pack_mmap_calls;
 409static unsigned int peak_pack_open_windows;
 410static unsigned int pack_open_windows;
 411static size_t peak_pack_mapped;
 412static size_t pack_mapped;
 413struct packed_git *packed_git;
 414
 415void pack_report()
 416{
 417        fprintf(stderr,
 418                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 419                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 420                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 421                (size_t) getpagesize(),
 422                packed_git_window_size,
 423                packed_git_limit);
 424        fprintf(stderr,
 425                "pack_report: pack_used_ctr            = %10u\n"
 426                "pack_report: pack_mmap_calls          = %10u\n"
 427                "pack_report: pack_open_windows        = %10u / %10u\n"
 428                "pack_report: pack_mapped              = "
 429                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 430                pack_used_ctr,
 431                pack_mmap_calls,
 432                pack_open_windows, peak_pack_open_windows,
 433                pack_mapped, peak_pack_mapped);
 434}
 435
 436static int check_packed_git_idx(const char *path,  struct packed_git *p)
 437{
 438        void *idx_map;
 439        struct pack_idx_header *hdr;
 440        size_t idx_size;
 441        uint32_t version, nr, i, *index;
 442        int fd = open(path, O_RDONLY);
 443        struct stat st;
 444
 445        if (fd < 0)
 446                return -1;
 447        if (fstat(fd, &st)) {
 448                close(fd);
 449                return -1;
 450        }
 451        idx_size = xsize_t(st.st_size);
 452        if (idx_size < 4 * 256 + 20 + 20) {
 453                close(fd);
 454                return error("index file %s is too small", path);
 455        }
 456        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 457        close(fd);
 458
 459        hdr = idx_map;
 460        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 461                version = ntohl(hdr->idx_version);
 462                if (version < 2 || version > 2) {
 463                        munmap(idx_map, idx_size);
 464                        return error("index file %s is version %d"
 465                                     " and is not supported by this binary"
 466                                     " (try upgrading GIT to a newer version)",
 467                                     path, version);
 468                }
 469        } else
 470                version = 1;
 471
 472        nr = 0;
 473        index = idx_map;
 474        if (version > 1)
 475                index += 2;  /* skip index header */
 476        for (i = 0; i < 256; i++) {
 477                uint32_t n = ntohl(index[i]);
 478                if (n < nr) {
 479                        munmap(idx_map, idx_size);
 480                        return error("non-monotonic index %s", path);
 481                }
 482                nr = n;
 483        }
 484
 485        if (version == 1) {
 486                /*
 487                 * Total size:
 488                 *  - 256 index entries 4 bytes each
 489                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 490                 *  - 20-byte SHA1 of the packfile
 491                 *  - 20-byte SHA1 file checksum
 492                 */
 493                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 494                        munmap(idx_map, idx_size);
 495                        return error("wrong index file size in %s", path);
 496                }
 497        } else if (version == 2) {
 498                /*
 499                 * Minimum size:
 500                 *  - 8 bytes of header
 501                 *  - 256 index entries 4 bytes each
 502                 *  - 20-byte sha1 entry * nr
 503                 *  - 4-byte crc entry * nr
 504                 *  - 4-byte offset entry * nr
 505                 *  - 20-byte SHA1 of the packfile
 506                 *  - 20-byte SHA1 file checksum
 507                 * And after the 4-byte offset table might be a
 508                 * variable sized table containing 8-byte entries
 509                 * for offsets larger than 2^31.
 510                 */
 511                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 512                if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
 513                        munmap(idx_map, idx_size);
 514                        return error("wrong index file size in %s", path);
 515                }
 516                if (idx_size != min_size) {
 517                        /* make sure we can deal with large pack offsets */
 518                        off_t x = 0x7fffffffUL, y = 0xffffffffUL;
 519                        if (x > (x + 1) || y > (y + 1)) {
 520                                munmap(idx_map, idx_size);
 521                                return error("pack too large for current definition of off_t in %s", path);
 522                        }
 523                }
 524        }
 525
 526        p->index_version = version;
 527        p->index_data = idx_map;
 528        p->index_size = idx_size;
 529        p->num_objects = nr;
 530        return 0;
 531}
 532
 533static void scan_windows(struct packed_git *p,
 534        struct packed_git **lru_p,
 535        struct pack_window **lru_w,
 536        struct pack_window **lru_l)
 537{
 538        struct pack_window *w, *w_l;
 539
 540        for (w_l = NULL, w = p->windows; w; w = w->next) {
 541                if (!w->inuse_cnt) {
 542                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 543                                *lru_p = p;
 544                                *lru_w = w;
 545                                *lru_l = w_l;
 546                        }
 547                }
 548                w_l = w;
 549        }
 550}
 551
 552static int unuse_one_window(struct packed_git *current, int keep_fd)
 553{
 554        struct packed_git *p, *lru_p = NULL;
 555        struct pack_window *lru_w = NULL, *lru_l = NULL;
 556
 557        if (current)
 558                scan_windows(current, &lru_p, &lru_w, &lru_l);
 559        for (p = packed_git; p; p = p->next)
 560                scan_windows(p, &lru_p, &lru_w, &lru_l);
 561        if (lru_p) {
 562                munmap(lru_w->base, lru_w->len);
 563                pack_mapped -= lru_w->len;
 564                if (lru_l)
 565                        lru_l->next = lru_w->next;
 566                else {
 567                        lru_p->windows = lru_w->next;
 568                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 569                                close(lru_p->pack_fd);
 570                                lru_p->pack_fd = -1;
 571                        }
 572                }
 573                free(lru_w);
 574                pack_open_windows--;
 575                return 1;
 576        }
 577        return 0;
 578}
 579
 580void release_pack_memory(size_t need, int fd)
 581{
 582        size_t cur = pack_mapped;
 583        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 584                ; /* nothing */
 585}
 586
 587void unuse_pack(struct pack_window **w_cursor)
 588{
 589        struct pack_window *w = *w_cursor;
 590        if (w) {
 591                w->inuse_cnt--;
 592                *w_cursor = NULL;
 593        }
 594}
 595
 596/*
 597 * Do not call this directly as this leaks p->pack_fd on error return;
 598 * call open_packed_git() instead.
 599 */
 600static int open_packed_git_1(struct packed_git *p)
 601{
 602        struct stat st;
 603        struct pack_header hdr;
 604        unsigned char sha1[20];
 605        unsigned char *idx_sha1;
 606        long fd_flag;
 607
 608        p->pack_fd = open(p->pack_name, O_RDONLY);
 609        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 610                return -1;
 611
 612        /* If we created the struct before we had the pack we lack size. */
 613        if (!p->pack_size) {
 614                if (!S_ISREG(st.st_mode))
 615                        return error("packfile %s not a regular file", p->pack_name);
 616                p->pack_size = st.st_size;
 617        } else if (p->pack_size != st.st_size)
 618                return error("packfile %s size changed", p->pack_name);
 619
 620        /* We leave these file descriptors open with sliding mmap;
 621         * there is no point keeping them open across exec(), though.
 622         */
 623        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 624        if (fd_flag < 0)
 625                return error("cannot determine file descriptor flags");
 626        fd_flag |= FD_CLOEXEC;
 627        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 628                return error("cannot set FD_CLOEXEC");
 629
 630        /* Verify we recognize this pack file format. */
 631        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 632                return error("file %s is far too short to be a packfile", p->pack_name);
 633        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 634                return error("file %s is not a GIT packfile", p->pack_name);
 635        if (!pack_version_ok(hdr.hdr_version))
 636                return error("packfile %s is version %u and not supported"
 637                        " (try upgrading GIT to a newer version)",
 638                        p->pack_name, ntohl(hdr.hdr_version));
 639
 640        /* Verify the pack matches its index. */
 641        if (p->num_objects != ntohl(hdr.hdr_entries))
 642                return error("packfile %s claims to have %u objects"
 643                             " while index indicates %u objects",
 644                             p->pack_name, ntohl(hdr.hdr_entries),
 645                             p->num_objects);
 646        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 647                return error("end of packfile %s is unavailable", p->pack_name);
 648        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 649                return error("packfile %s signature is unavailable", p->pack_name);
 650        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 651        if (hashcmp(sha1, idx_sha1))
 652                return error("packfile %s does not match index", p->pack_name);
 653        return 0;
 654}
 655
 656static int open_packed_git(struct packed_git *p)
 657{
 658        if (!open_packed_git_1(p))
 659                return 0;
 660        if (p->pack_fd != -1) {
 661                close(p->pack_fd);
 662                p->pack_fd = -1;
 663        }
 664        return -1;
 665}
 666
 667static int in_window(struct pack_window *win, off_t offset)
 668{
 669        /* We must promise at least 20 bytes (one hash) after the
 670         * offset is available from this window, otherwise the offset
 671         * is not actually in this window and a different window (which
 672         * has that one hash excess) must be used.  This is to support
 673         * the object header and delta base parsing routines below.
 674         */
 675        off_t win_off = win->offset;
 676        return win_off <= offset
 677                && (offset + 20) <= (win_off + win->len);
 678}
 679
 680unsigned char* use_pack(struct packed_git *p,
 681                struct pack_window **w_cursor,
 682                off_t offset,
 683                unsigned int *left)
 684{
 685        struct pack_window *win = *w_cursor;
 686
 687        if (p->pack_fd == -1 && open_packed_git(p))
 688                die("packfile %s cannot be accessed", p->pack_name);
 689
 690        /* Since packfiles end in a hash of their content and its
 691         * pointless to ask for an offset into the middle of that
 692         * hash, and the in_window function above wouldn't match
 693         * don't allow an offset too close to the end of the file.
 694         */
 695        if (offset > (p->pack_size - 20))
 696                die("offset beyond end of packfile (truncated pack?)");
 697
 698        if (!win || !in_window(win, offset)) {
 699                if (win)
 700                        win->inuse_cnt--;
 701                for (win = p->windows; win; win = win->next) {
 702                        if (in_window(win, offset))
 703                                break;
 704                }
 705                if (!win) {
 706                        size_t window_align = packed_git_window_size / 2;
 707                        off_t len;
 708                        win = xcalloc(1, sizeof(*win));
 709                        win->offset = (offset / window_align) * window_align;
 710                        len = p->pack_size - win->offset;
 711                        if (len > packed_git_window_size)
 712                                len = packed_git_window_size;
 713                        win->len = (size_t)len;
 714                        pack_mapped += win->len;
 715                        while (packed_git_limit < pack_mapped
 716                                && unuse_one_window(p, p->pack_fd))
 717                                ; /* nothing */
 718                        win->base = xmmap(NULL, win->len,
 719                                PROT_READ, MAP_PRIVATE,
 720                                p->pack_fd, win->offset);
 721                        if (win->base == MAP_FAILED)
 722                                die("packfile %s cannot be mapped: %s",
 723                                        p->pack_name,
 724                                        strerror(errno));
 725                        pack_mmap_calls++;
 726                        pack_open_windows++;
 727                        if (pack_mapped > peak_pack_mapped)
 728                                peak_pack_mapped = pack_mapped;
 729                        if (pack_open_windows > peak_pack_open_windows)
 730                                peak_pack_open_windows = pack_open_windows;
 731                        win->next = p->windows;
 732                        p->windows = win;
 733                }
 734        }
 735        if (win != *w_cursor) {
 736                win->last_used = pack_used_ctr++;
 737                win->inuse_cnt++;
 738                *w_cursor = win;
 739        }
 740        offset -= win->offset;
 741        if (left)
 742                *left = win->len - xsize_t(offset);
 743        return win->base + offset;
 744}
 745
 746struct packed_git *add_packed_git(const char *path, int path_len, int local)
 747{
 748        struct stat st;
 749        struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
 750
 751        /*
 752         * Make sure a corresponding .pack file exists and that
 753         * the index looks sane.
 754         */
 755        path_len -= strlen(".idx");
 756        if (path_len < 1)
 757                return NULL;
 758        memcpy(p->pack_name, path, path_len);
 759        strcpy(p->pack_name + path_len, ".pack");
 760        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
 761            check_packed_git_idx(path, p)) {
 762                free(p);
 763                return NULL;
 764        }
 765
 766        /* ok, it looks sane as far as we can check without
 767         * actually mapping the pack file.
 768         */
 769        p->pack_size = st.st_size;
 770        p->next = NULL;
 771        p->windows = NULL;
 772        p->pack_fd = -1;
 773        p->pack_local = local;
 774        p->mtime = st.st_mtime;
 775        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 776                hashclr(p->sha1);
 777        return p;
 778}
 779
 780struct packed_git *parse_pack_index(unsigned char *sha1)
 781{
 782        char *path = sha1_pack_index_name(sha1);
 783        return parse_pack_index_file(sha1, path);
 784}
 785
 786struct packed_git *parse_pack_index_file(const unsigned char *sha1,
 787                                         const char *idx_path)
 788{
 789        const char *path = sha1_pack_name(sha1);
 790        struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
 791
 792        if (check_packed_git_idx(idx_path, p)) {
 793                free(p);
 794                return NULL;
 795        }
 796
 797        strcpy(p->pack_name, path);
 798        p->pack_size = 0;
 799        p->next = NULL;
 800        p->windows = NULL;
 801        p->pack_fd = -1;
 802        hashcpy(p->sha1, sha1);
 803        return p;
 804}
 805
 806void install_packed_git(struct packed_git *pack)
 807{
 808        pack->next = packed_git;
 809        packed_git = pack;
 810}
 811
 812static void prepare_packed_git_one(char *objdir, int local)
 813{
 814        char path[PATH_MAX];
 815        int len;
 816        DIR *dir;
 817        struct dirent *de;
 818
 819        sprintf(path, "%s/pack", objdir);
 820        len = strlen(path);
 821        dir = opendir(path);
 822        if (!dir) {
 823                if (errno != ENOENT)
 824                        error("unable to open object pack directory: %s: %s",
 825                              path, strerror(errno));
 826                return;
 827        }
 828        path[len++] = '/';
 829        while ((de = readdir(dir)) != NULL) {
 830                int namelen = strlen(de->d_name);
 831                struct packed_git *p;
 832
 833                if (!has_extension(de->d_name, ".idx"))
 834                        continue;
 835
 836                /* Don't reopen a pack we already have. */
 837                strcpy(path + len, de->d_name);
 838                for (p = packed_git; p; p = p->next) {
 839                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 840                                break;
 841                }
 842                if (p)
 843                        continue;
 844                /* See if it really is a valid .idx file with corresponding
 845                 * .pack file that we can map.
 846                 */
 847                p = add_packed_git(path, len + namelen, local);
 848                if (!p)
 849                        continue;
 850                install_packed_git(p);
 851        }
 852        closedir(dir);
 853}
 854
 855static int sort_pack(const void *a_, const void *b_)
 856{
 857        struct packed_git *a = *((struct packed_git **)a_);
 858        struct packed_git *b = *((struct packed_git **)b_);
 859        int st;
 860
 861        /*
 862         * Local packs tend to contain objects specific to our
 863         * variant of the project than remote ones.  In addition,
 864         * remote ones could be on a network mounted filesystem.
 865         * Favor local ones for these reasons.
 866         */
 867        st = a->pack_local - b->pack_local;
 868        if (st)
 869                return -st;
 870
 871        /*
 872         * Younger packs tend to contain more recent objects,
 873         * and more recent objects tend to get accessed more
 874         * often.
 875         */
 876        if (a->mtime < b->mtime)
 877                return 1;
 878        else if (a->mtime == b->mtime)
 879                return 0;
 880        return -1;
 881}
 882
 883static void rearrange_packed_git(void)
 884{
 885        struct packed_git **ary, *p;
 886        int i, n;
 887
 888        for (n = 0, p = packed_git; p; p = p->next)
 889                n++;
 890        if (n < 2)
 891                return;
 892
 893        /* prepare an array of packed_git for easier sorting */
 894        ary = xcalloc(n, sizeof(struct packed_git *));
 895        for (n = 0, p = packed_git; p; p = p->next)
 896                ary[n++] = p;
 897
 898        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 899
 900        /* link them back again */
 901        for (i = 0; i < n - 1; i++)
 902                ary[i]->next = ary[i + 1];
 903        ary[n - 1]->next = NULL;
 904        packed_git = ary[0];
 905
 906        free(ary);
 907}
 908
 909static int prepare_packed_git_run_once = 0;
 910void prepare_packed_git(void)
 911{
 912        struct alternate_object_database *alt;
 913
 914        if (prepare_packed_git_run_once)
 915                return;
 916        prepare_packed_git_one(get_object_directory(), 1);
 917        prepare_alt_odb();
 918        for (alt = alt_odb_list; alt; alt = alt->next) {
 919                alt->name[-1] = 0;
 920                prepare_packed_git_one(alt->base, 0);
 921                alt->name[-1] = '/';
 922        }
 923        rearrange_packed_git();
 924        prepare_packed_git_run_once = 1;
 925}
 926
 927void reprepare_packed_git(void)
 928{
 929        prepare_packed_git_run_once = 0;
 930        prepare_packed_git();
 931}
 932
 933int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 934{
 935        unsigned char real_sha1[20];
 936        hash_sha1_file(map, size, type, real_sha1);
 937        return hashcmp(sha1, real_sha1) ? -1 : 0;
 938}
 939
 940void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 941{
 942        struct stat st;
 943        void *map;
 944        int fd;
 945        char *filename = find_sha1_file(sha1, &st);
 946
 947        if (!filename) {
 948                return NULL;
 949        }
 950
 951        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 952        if (fd < 0) {
 953                /* See if it works without O_NOATIME */
 954                switch (sha1_file_open_flag) {
 955                default:
 956                        fd = open(filename, O_RDONLY);
 957                        if (fd >= 0)
 958                                break;
 959                /* Fallthrough */
 960                case 0:
 961                        return NULL;
 962                }
 963
 964                /* If it failed once, it will probably fail again.
 965                 * Stop using O_NOATIME
 966                 */
 967                sha1_file_open_flag = 0;
 968        }
 969        *size = xsize_t(st.st_size);
 970        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 971        close(fd);
 972        return map;
 973}
 974
 975int legacy_loose_object(unsigned char *map)
 976{
 977        unsigned int word;
 978
 979        /*
 980         * Is it a zlib-compressed buffer? If so, the first byte
 981         * must be 0x78 (15-bit window size, deflated), and the
 982         * first 16-bit word is evenly divisible by 31
 983         */
 984        word = (map[0] << 8) + map[1];
 985        if (map[0] == 0x78 && !(word % 31))
 986                return 1;
 987        else
 988                return 0;
 989}
 990
 991unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
 992{
 993        unsigned shift;
 994        unsigned char c;
 995        unsigned long size;
 996        unsigned long used = 0;
 997
 998        c = buf[used++];
 999        *type = (c >> 4) & 7;
1000        size = c & 15;
1001        shift = 4;
1002        while (c & 0x80) {
1003                if (len <= used)
1004                        return 0;
1005                if (sizeof(long) * 8 <= shift)
1006                        return 0;
1007                c = buf[used++];
1008                size += (c & 0x7f) << shift;
1009                shift += 7;
1010        }
1011        *sizep = size;
1012        return used;
1013}
1014
1015static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1016{
1017        unsigned long size, used;
1018        static const char valid_loose_object_type[8] = {
1019                0, /* OBJ_EXT */
1020                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1021                0, /* "delta" and others are invalid in a loose object */
1022        };
1023        enum object_type type;
1024
1025        /* Get the data stream */
1026        memset(stream, 0, sizeof(*stream));
1027        stream->next_in = map;
1028        stream->avail_in = mapsize;
1029        stream->next_out = buffer;
1030        stream->avail_out = bufsiz;
1031
1032        if (legacy_loose_object(map)) {
1033                inflateInit(stream);
1034                return inflate(stream, 0);
1035        }
1036
1037        used = unpack_object_header_gently(map, mapsize, &type, &size);
1038        if (!used || !valid_loose_object_type[type])
1039                return -1;
1040        map += used;
1041        mapsize -= used;
1042
1043        /* Set up the stream for the rest.. */
1044        stream->next_in = map;
1045        stream->avail_in = mapsize;
1046        inflateInit(stream);
1047
1048        /* And generate the fake traditional header */
1049        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1050                                         typename(type), size);
1051        return 0;
1052}
1053
1054static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1055{
1056        int bytes = strlen(buffer) + 1;
1057        unsigned char *buf = xmalloc(1+size);
1058        unsigned long n;
1059        int status = Z_OK;
1060
1061        n = stream->total_out - bytes;
1062        if (n > size)
1063                n = size;
1064        memcpy(buf, (char *) buffer + bytes, n);
1065        bytes = n;
1066        if (bytes <= size) {
1067                /*
1068                 * The above condition must be (bytes <= size), not
1069                 * (bytes < size).  In other words, even though we
1070                 * expect no more output and set avail_out to zer0,
1071                 * the input zlib stream may have bytes that express
1072                 * "this concludes the stream", and we *do* want to
1073                 * eat that input.
1074                 *
1075                 * Otherwise we would not be able to test that we
1076                 * consumed all the input to reach the expected size;
1077                 * we also want to check that zlib tells us that all
1078                 * went well with status == Z_STREAM_END at the end.
1079                 */
1080                stream->next_out = buf + bytes;
1081                stream->avail_out = size - bytes;
1082                while (status == Z_OK)
1083                        status = inflate(stream, Z_FINISH);
1084        }
1085        buf[size] = 0;
1086        if (status == Z_STREAM_END && !stream->avail_in) {
1087                inflateEnd(stream);
1088                return buf;
1089        }
1090
1091        if (status < 0)
1092                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1093        else if (stream->avail_in)
1094                error("garbage at end of loose object '%s'",
1095                      sha1_to_hex(sha1));
1096        free(buf);
1097        return NULL;
1098}
1099
1100/*
1101 * We used to just use "sscanf()", but that's actually way
1102 * too permissive for what we want to check. So do an anal
1103 * object header parse by hand.
1104 */
1105static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1106{
1107        char type[10];
1108        int i;
1109        unsigned long size;
1110
1111        /*
1112         * The type can be at most ten bytes (including the 
1113         * terminating '\0' that we add), and is followed by
1114         * a space.
1115         */
1116        i = 0;
1117        for (;;) {
1118                char c = *hdr++;
1119                if (c == ' ')
1120                        break;
1121                type[i++] = c;
1122                if (i >= sizeof(type))
1123                        return -1;
1124        }
1125        type[i] = 0;
1126
1127        /*
1128         * The length must follow immediately, and be in canonical
1129         * decimal format (ie "010" is not valid).
1130         */
1131        size = *hdr++ - '0';
1132        if (size > 9)
1133                return -1;
1134        if (size) {
1135                for (;;) {
1136                        unsigned long c = *hdr - '0';
1137                        if (c > 9)
1138                                break;
1139                        hdr++;
1140                        size = size * 10 + c;
1141                }
1142        }
1143        *sizep = size;
1144
1145        /*
1146         * The length must be followed by a zero byte
1147         */
1148        return *hdr ? -1 : type_from_string(type);
1149}
1150
1151static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1152{
1153        int ret;
1154        z_stream stream;
1155        char hdr[8192];
1156
1157        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1158        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1159                return NULL;
1160
1161        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1162}
1163
1164unsigned long get_size_from_delta(struct packed_git *p,
1165                                  struct pack_window **w_curs,
1166                                  off_t curpos)
1167{
1168        const unsigned char *data;
1169        unsigned char delta_head[20], *in;
1170        z_stream stream;
1171        int st;
1172
1173        memset(&stream, 0, sizeof(stream));
1174        stream.next_out = delta_head;
1175        stream.avail_out = sizeof(delta_head);
1176
1177        inflateInit(&stream);
1178        do {
1179                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1180                stream.next_in = in;
1181                st = inflate(&stream, Z_FINISH);
1182                curpos += stream.next_in - in;
1183        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1184                 stream.total_out < sizeof(delta_head));
1185        inflateEnd(&stream);
1186        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1187                die("delta data unpack-initial failed");
1188
1189        /* Examine the initial part of the delta to figure out
1190         * the result size.
1191         */
1192        data = delta_head;
1193
1194        /* ignore base size */
1195        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1196
1197        /* Read the result size */
1198        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1199}
1200
1201static off_t get_delta_base(struct packed_git *p,
1202                                    struct pack_window **w_curs,
1203                                    off_t *curpos,
1204                                    enum object_type type,
1205                                    off_t delta_obj_offset)
1206{
1207        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1208        off_t base_offset;
1209
1210        /* use_pack() assured us we have [base_info, base_info + 20)
1211         * as a range that we can look at without walking off the
1212         * end of the mapped window.  Its actually the hash size
1213         * that is assured.  An OFS_DELTA longer than the hash size
1214         * is stupid, as then a REF_DELTA would be smaller to store.
1215         */
1216        if (type == OBJ_OFS_DELTA) {
1217                unsigned used = 0;
1218                unsigned char c = base_info[used++];
1219                base_offset = c & 127;
1220                while (c & 128) {
1221                        base_offset += 1;
1222                        if (!base_offset || MSB(base_offset, 7))
1223                                die("offset value overflow for delta base object");
1224                        c = base_info[used++];
1225                        base_offset = (base_offset << 7) + (c & 127);
1226                }
1227                base_offset = delta_obj_offset - base_offset;
1228                if (base_offset >= delta_obj_offset)
1229                        die("delta base offset out of bound");
1230                *curpos += used;
1231        } else if (type == OBJ_REF_DELTA) {
1232                /* The base entry _must_ be in the same pack */
1233                base_offset = find_pack_entry_one(base_info, p);
1234                if (!base_offset)
1235                        die("failed to find delta-pack base object %s",
1236                                sha1_to_hex(base_info));
1237                *curpos += 20;
1238        } else
1239                die("I am totally screwed");
1240        return base_offset;
1241}
1242
1243/* forward declaration for a mutually recursive function */
1244static int packed_object_info(struct packed_git *p, off_t offset,
1245                              unsigned long *sizep);
1246
1247static int packed_delta_info(struct packed_git *p,
1248                             struct pack_window **w_curs,
1249                             off_t curpos,
1250                             enum object_type type,
1251                             off_t obj_offset,
1252                             unsigned long *sizep)
1253{
1254        off_t base_offset;
1255
1256        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1257        type = packed_object_info(p, base_offset, NULL);
1258
1259        /* We choose to only get the type of the base object and
1260         * ignore potentially corrupt pack file that expects the delta
1261         * based on a base with a wrong size.  This saves tons of
1262         * inflate() calls.
1263         */
1264        if (sizep)
1265                *sizep = get_size_from_delta(p, w_curs, curpos);
1266
1267        return type;
1268}
1269
1270static int unpack_object_header(struct packed_git *p,
1271                                struct pack_window **w_curs,
1272                                off_t *curpos,
1273                                unsigned long *sizep)
1274{
1275        unsigned char *base;
1276        unsigned int left;
1277        unsigned long used;
1278        enum object_type type;
1279
1280        /* use_pack() assures us we have [base, base + 20) available
1281         * as a range that we can look at at.  (Its actually the hash
1282         * size that is assured.)  With our object header encoding
1283         * the maximum deflated object size is 2^137, which is just
1284         * insane, so we know won't exceed what we have been given.
1285         */
1286        base = use_pack(p, w_curs, *curpos, &left);
1287        used = unpack_object_header_gently(base, left, &type, sizep);
1288        if (!used)
1289                die("object offset outside of pack file");
1290        *curpos += used;
1291
1292        return type;
1293}
1294
1295const char *packed_object_info_detail(struct packed_git *p,
1296                                      off_t obj_offset,
1297                                      unsigned long *size,
1298                                      unsigned long *store_size,
1299                                      unsigned int *delta_chain_length,
1300                                      unsigned char *base_sha1)
1301{
1302        struct pack_window *w_curs = NULL;
1303        off_t curpos;
1304        unsigned long dummy;
1305        unsigned char *next_sha1;
1306        enum object_type type;
1307
1308        *delta_chain_length = 0;
1309        curpos = obj_offset;
1310        type = unpack_object_header(p, &w_curs, &curpos, size);
1311
1312        for (;;) {
1313                switch (type) {
1314                default:
1315                        die("pack %s contains unknown object type %d",
1316                            p->pack_name, type);
1317                case OBJ_COMMIT:
1318                case OBJ_TREE:
1319                case OBJ_BLOB:
1320                case OBJ_TAG:
1321                        *store_size = 0; /* notyet */
1322                        unuse_pack(&w_curs);
1323                        return typename(type);
1324                case OBJ_OFS_DELTA:
1325                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1326                        if (*delta_chain_length == 0) {
1327                                /* TODO: find base_sha1 as pointed by curpos */
1328                                hashclr(base_sha1);
1329                        }
1330                        break;
1331                case OBJ_REF_DELTA:
1332                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1333                        if (*delta_chain_length == 0)
1334                                hashcpy(base_sha1, next_sha1);
1335                        obj_offset = find_pack_entry_one(next_sha1, p);
1336                        break;
1337                }
1338                (*delta_chain_length)++;
1339                curpos = obj_offset;
1340                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1341        }
1342}
1343
1344static int packed_object_info(struct packed_git *p, off_t obj_offset,
1345                              unsigned long *sizep)
1346{
1347        struct pack_window *w_curs = NULL;
1348        unsigned long size;
1349        off_t curpos = obj_offset;
1350        enum object_type type;
1351
1352        type = unpack_object_header(p, &w_curs, &curpos, &size);
1353
1354        switch (type) {
1355        case OBJ_OFS_DELTA:
1356        case OBJ_REF_DELTA:
1357                type = packed_delta_info(p, &w_curs, curpos,
1358                                         type, obj_offset, sizep);
1359                break;
1360        case OBJ_COMMIT:
1361        case OBJ_TREE:
1362        case OBJ_BLOB:
1363        case OBJ_TAG:
1364                if (sizep)
1365                        *sizep = size;
1366                break;
1367        default:
1368                die("pack %s contains unknown object type %d",
1369                    p->pack_name, type);
1370        }
1371        unuse_pack(&w_curs);
1372        return type;
1373}
1374
1375static void *unpack_compressed_entry(struct packed_git *p,
1376                                    struct pack_window **w_curs,
1377                                    off_t curpos,
1378                                    unsigned long size)
1379{
1380        int st;
1381        z_stream stream;
1382        unsigned char *buffer, *in;
1383
1384        buffer = xmalloc(size + 1);
1385        buffer[size] = 0;
1386        memset(&stream, 0, sizeof(stream));
1387        stream.next_out = buffer;
1388        stream.avail_out = size;
1389
1390        inflateInit(&stream);
1391        do {
1392                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1393                stream.next_in = in;
1394                st = inflate(&stream, Z_FINISH);
1395                curpos += stream.next_in - in;
1396        } while (st == Z_OK || st == Z_BUF_ERROR);
1397        inflateEnd(&stream);
1398        if ((st != Z_STREAM_END) || stream.total_out != size) {
1399                free(buffer);
1400                return NULL;
1401        }
1402
1403        return buffer;
1404}
1405
1406#define MAX_DELTA_CACHE (256)
1407
1408static size_t delta_base_cached;
1409
1410static struct delta_base_cache_lru_list {
1411        struct delta_base_cache_lru_list *prev;
1412        struct delta_base_cache_lru_list *next;
1413} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1414
1415static struct delta_base_cache_entry {
1416        struct delta_base_cache_lru_list lru;
1417        void *data;
1418        struct packed_git *p;
1419        off_t base_offset;
1420        unsigned long size;
1421        enum object_type type;
1422} delta_base_cache[MAX_DELTA_CACHE];
1423
1424static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1425{
1426        unsigned long hash;
1427
1428        hash = (unsigned long)p + (unsigned long)base_offset;
1429        hash += (hash >> 8) + (hash >> 16);
1430        return hash % MAX_DELTA_CACHE;
1431}
1432
1433static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1434        unsigned long *base_size, enum object_type *type, int keep_cache)
1435{
1436        void *ret;
1437        unsigned long hash = pack_entry_hash(p, base_offset);
1438        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1439
1440        ret = ent->data;
1441        if (ret && ent->p == p && ent->base_offset == base_offset)
1442                goto found_cache_entry;
1443        return unpack_entry(p, base_offset, type, base_size);
1444
1445found_cache_entry:
1446        if (!keep_cache) {
1447                ent->data = NULL;
1448                ent->lru.next->prev = ent->lru.prev;
1449                ent->lru.prev->next = ent->lru.next;
1450                delta_base_cached -= ent->size;
1451        }
1452        else {
1453                ret = xmalloc(ent->size + 1);
1454                memcpy(ret, ent->data, ent->size);
1455                ((char *)ret)[ent->size] = 0;
1456        }
1457        *type = ent->type;
1458        *base_size = ent->size;
1459        return ret;
1460}
1461
1462static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1463{
1464        if (ent->data) {
1465                free(ent->data);
1466                ent->data = NULL;
1467                ent->lru.next->prev = ent->lru.prev;
1468                ent->lru.prev->next = ent->lru.next;
1469                delta_base_cached -= ent->size;
1470        }
1471}
1472
1473static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1474        void *base, unsigned long base_size, enum object_type type)
1475{
1476        unsigned long hash = pack_entry_hash(p, base_offset);
1477        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1478        struct delta_base_cache_lru_list *lru;
1479
1480        release_delta_base_cache(ent);
1481        delta_base_cached += base_size;
1482
1483        for (lru = delta_base_cache_lru.next;
1484             delta_base_cached > delta_base_cache_limit
1485             && lru != &delta_base_cache_lru;
1486             lru = lru->next) {
1487                struct delta_base_cache_entry *f = (void *)lru;
1488                if (f->type == OBJ_BLOB)
1489                        release_delta_base_cache(f);
1490        }
1491        for (lru = delta_base_cache_lru.next;
1492             delta_base_cached > delta_base_cache_limit
1493             && lru != &delta_base_cache_lru;
1494             lru = lru->next) {
1495                struct delta_base_cache_entry *f = (void *)lru;
1496                release_delta_base_cache(f);
1497        }
1498
1499        ent->p = p;
1500        ent->base_offset = base_offset;
1501        ent->type = type;
1502        ent->data = base;
1503        ent->size = base_size;
1504        ent->lru.next = &delta_base_cache_lru;
1505        ent->lru.prev = delta_base_cache_lru.prev;
1506        delta_base_cache_lru.prev->next = &ent->lru;
1507        delta_base_cache_lru.prev = &ent->lru;
1508}
1509
1510static void *unpack_delta_entry(struct packed_git *p,
1511                                struct pack_window **w_curs,
1512                                off_t curpos,
1513                                unsigned long delta_size,
1514                                off_t obj_offset,
1515                                enum object_type *type,
1516                                unsigned long *sizep)
1517{
1518        void *delta_data, *result, *base;
1519        unsigned long base_size;
1520        off_t base_offset;
1521
1522        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1523        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1524        if (!base)
1525                die("failed to read delta base object"
1526                    " at %"PRIuMAX" from %s",
1527                    (uintmax_t)base_offset, p->pack_name);
1528
1529        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1530        result = patch_delta(base, base_size,
1531                             delta_data, delta_size,
1532                             sizep);
1533        if (!result)
1534                die("failed to apply delta");
1535        free(delta_data);
1536        add_delta_base_cache(p, base_offset, base, base_size, *type);
1537        return result;
1538}
1539
1540void *unpack_entry(struct packed_git *p, off_t obj_offset,
1541                   enum object_type *type, unsigned long *sizep)
1542{
1543        struct pack_window *w_curs = NULL;
1544        off_t curpos = obj_offset;
1545        void *data;
1546
1547        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1548        switch (*type) {
1549        case OBJ_OFS_DELTA:
1550        case OBJ_REF_DELTA:
1551                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1552                                          obj_offset, type, sizep);
1553                break;
1554        case OBJ_COMMIT:
1555        case OBJ_TREE:
1556        case OBJ_BLOB:
1557        case OBJ_TAG:
1558                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1559                break;
1560        default:
1561                die("unknown object type %i in %s", *type, p->pack_name);
1562        }
1563        unuse_pack(&w_curs);
1564        return data;
1565}
1566
1567const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1568                                            uint32_t n)
1569{
1570        const unsigned char *index = p->index_data;
1571        if (n >= p->num_objects)
1572                return NULL;
1573        index += 4 * 256;
1574        if (p->index_version == 1) {
1575                return index + 24 * n + 4;
1576        } else {
1577                index += 8;
1578                return index + 20 * n;
1579        }
1580}
1581
1582static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1583{
1584        const unsigned char *index = p->index_data;
1585        index += 4 * 256;
1586        if (p->index_version == 1) {
1587                return ntohl(*((uint32_t *)(index + 24 * n)));
1588        } else {
1589                uint32_t off;
1590                index += 8 + p->num_objects * (20 + 4);
1591                off = ntohl(*((uint32_t *)(index + 4 * n)));
1592                if (!(off & 0x80000000))
1593                        return off;
1594                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1595                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1596                                   ntohl(*((uint32_t *)(index + 4)));
1597        }
1598}
1599
1600off_t find_pack_entry_one(const unsigned char *sha1,
1601                                  struct packed_git *p)
1602{
1603        const uint32_t *level1_ofs = p->index_data;
1604        const unsigned char *index = p->index_data;
1605        unsigned hi, lo;
1606
1607        if (p->index_version > 1) {
1608                level1_ofs += 2;
1609                index += 8;
1610        }
1611        index += 4 * 256;
1612        hi = ntohl(level1_ofs[*sha1]);
1613        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1614
1615        do {
1616                unsigned mi = (lo + hi) / 2;
1617                unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1618                int cmp = hashcmp(index + x, sha1);
1619                if (!cmp)
1620                        return nth_packed_object_offset(p, mi);
1621                if (cmp > 0)
1622                        hi = mi;
1623                else
1624                        lo = mi+1;
1625        } while (lo < hi);
1626        return 0;
1627}
1628
1629static int matches_pack_name(struct packed_git *p, const char *ig)
1630{
1631        const char *last_c, *c;
1632
1633        if (!strcmp(p->pack_name, ig))
1634                return 0;
1635
1636        for (c = p->pack_name, last_c = c; *c;)
1637                if (*c == '/')
1638                        last_c = ++c;
1639                else
1640                        ++c;
1641        if (!strcmp(last_c, ig))
1642                return 0;
1643
1644        return 1;
1645}
1646
1647static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1648{
1649        struct packed_git *p;
1650        off_t offset;
1651
1652        prepare_packed_git();
1653
1654        for (p = packed_git; p; p = p->next) {
1655                if (ignore_packed) {
1656                        const char **ig;
1657                        for (ig = ignore_packed; *ig; ig++)
1658                                if (!matches_pack_name(p, *ig))
1659                                        break;
1660                        if (*ig)
1661                                continue;
1662                }
1663                offset = find_pack_entry_one(sha1, p);
1664                if (offset) {
1665                        /*
1666                         * We are about to tell the caller where they can
1667                         * locate the requested object.  We better make
1668                         * sure the packfile is still here and can be
1669                         * accessed before supplying that answer, as
1670                         * it may have been deleted since the index
1671                         * was loaded!
1672                         */
1673                        if (p->pack_fd == -1 && open_packed_git(p)) {
1674                                error("packfile %s cannot be accessed", p->pack_name);
1675                                continue;
1676                        }
1677                        e->offset = offset;
1678                        e->p = p;
1679                        hashcpy(e->sha1, sha1);
1680                        return 1;
1681                }
1682        }
1683        return 0;
1684}
1685
1686struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1687                                  struct packed_git *packs)
1688{
1689        struct packed_git *p;
1690
1691        for (p = packs; p; p = p->next) {
1692                if (find_pack_entry_one(sha1, p))
1693                        return p;
1694        }
1695        return NULL;
1696
1697}
1698
1699static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1700{
1701        int status;
1702        unsigned long mapsize, size;
1703        void *map;
1704        z_stream stream;
1705        char hdr[32];
1706
1707        map = map_sha1_file(sha1, &mapsize);
1708        if (!map)
1709                return error("unable to find %s", sha1_to_hex(sha1));
1710        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1711                status = error("unable to unpack %s header",
1712                               sha1_to_hex(sha1));
1713        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1714                status = error("unable to parse %s header", sha1_to_hex(sha1));
1715        else if (sizep)
1716                *sizep = size;
1717        inflateEnd(&stream);
1718        munmap(map, mapsize);
1719        return status;
1720}
1721
1722int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1723{
1724        struct pack_entry e;
1725
1726        if (!find_pack_entry(sha1, &e, NULL)) {
1727                reprepare_packed_git();
1728                if (!find_pack_entry(sha1, &e, NULL))
1729                        return sha1_loose_object_info(sha1, sizep);
1730        }
1731        return packed_object_info(e.p, e.offset, sizep);
1732}
1733
1734static void *read_packed_sha1(const unsigned char *sha1,
1735                              enum object_type *type, unsigned long *size)
1736{
1737        struct pack_entry e;
1738
1739        if (!find_pack_entry(sha1, &e, NULL))
1740                return NULL;
1741        else
1742                return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1743}
1744
1745/*
1746 * This is meant to hold a *small* number of objects that you would
1747 * want read_sha1_file() to be able to return, but yet you do not want
1748 * to write them into the object store (e.g. a browse-only
1749 * application).
1750 */
1751static struct cached_object {
1752        unsigned char sha1[20];
1753        enum object_type type;
1754        void *buf;
1755        unsigned long size;
1756} *cached_objects;
1757static int cached_object_nr, cached_object_alloc;
1758
1759static struct cached_object *find_cached_object(const unsigned char *sha1)
1760{
1761        int i;
1762        struct cached_object *co = cached_objects;
1763
1764        for (i = 0; i < cached_object_nr; i++, co++) {
1765                if (!hashcmp(co->sha1, sha1))
1766                        return co;
1767        }
1768        return NULL;
1769}
1770
1771int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1772                      unsigned char *sha1)
1773{
1774        struct cached_object *co;
1775
1776        hash_sha1_file(buf, len, typename(type), sha1);
1777        if (has_sha1_file(sha1) || find_cached_object(sha1))
1778                return 0;
1779        if (cached_object_alloc <= cached_object_nr) {
1780                cached_object_alloc = alloc_nr(cached_object_alloc);
1781                cached_objects = xrealloc(cached_objects,
1782                                          sizeof(*cached_objects) *
1783                                          cached_object_alloc);
1784        }
1785        co = &cached_objects[cached_object_nr++];
1786        co->size = len;
1787        co->type = type;
1788        co->buf = xmalloc(len);
1789        memcpy(co->buf, buf, len);
1790        hashcpy(co->sha1, sha1);
1791        return 0;
1792}
1793
1794void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1795                     unsigned long *size)
1796{
1797        unsigned long mapsize;
1798        void *map, *buf;
1799        struct cached_object *co;
1800
1801        co = find_cached_object(sha1);
1802        if (co) {
1803                buf = xmalloc(co->size + 1);
1804                memcpy(buf, co->buf, co->size);
1805                ((char*)buf)[co->size] = 0;
1806                *type = co->type;
1807                *size = co->size;
1808                return buf;
1809        }
1810
1811        buf = read_packed_sha1(sha1, type, size);
1812        if (buf)
1813                return buf;
1814        map = map_sha1_file(sha1, &mapsize);
1815        if (map) {
1816                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1817                munmap(map, mapsize);
1818                return buf;
1819        }
1820        reprepare_packed_git();
1821        return read_packed_sha1(sha1, type, size);
1822}
1823
1824void *read_object_with_reference(const unsigned char *sha1,
1825                                 const char *required_type_name,
1826                                 unsigned long *size,
1827                                 unsigned char *actual_sha1_return)
1828{
1829        enum object_type type, required_type;
1830        void *buffer;
1831        unsigned long isize;
1832        unsigned char actual_sha1[20];
1833
1834        required_type = type_from_string(required_type_name);
1835        hashcpy(actual_sha1, sha1);
1836        while (1) {
1837                int ref_length = -1;
1838                const char *ref_type = NULL;
1839
1840                buffer = read_sha1_file(actual_sha1, &type, &isize);
1841                if (!buffer)
1842                        return NULL;
1843                if (type == required_type) {
1844                        *size = isize;
1845                        if (actual_sha1_return)
1846                                hashcpy(actual_sha1_return, actual_sha1);
1847                        return buffer;
1848                }
1849                /* Handle references */
1850                else if (type == OBJ_COMMIT)
1851                        ref_type = "tree ";
1852                else if (type == OBJ_TAG)
1853                        ref_type = "object ";
1854                else {
1855                        free(buffer);
1856                        return NULL;
1857                }
1858                ref_length = strlen(ref_type);
1859
1860                if (memcmp(buffer, ref_type, ref_length) ||
1861                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1862                        free(buffer);
1863                        return NULL;
1864                }
1865                free(buffer);
1866                /* Now we have the ID of the referred-to object in
1867                 * actual_sha1.  Check again. */
1868        }
1869}
1870
1871static void write_sha1_file_prepare(const void *buf, unsigned long len,
1872                                    const char *type, unsigned char *sha1,
1873                                    char *hdr, int *hdrlen)
1874{
1875        SHA_CTX c;
1876
1877        /* Generate the header */
1878        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1879
1880        /* Sha1.. */
1881        SHA1_Init(&c);
1882        SHA1_Update(&c, hdr, *hdrlen);
1883        SHA1_Update(&c, buf, len);
1884        SHA1_Final(sha1, &c);
1885}
1886
1887/*
1888 * Link the tempfile to the final place, possibly creating the
1889 * last directory level as you do so.
1890 *
1891 * Returns the errno on failure, 0 on success.
1892 */
1893static int link_temp_to_file(const char *tmpfile, const char *filename)
1894{
1895        int ret;
1896        char *dir;
1897
1898        if (!link(tmpfile, filename))
1899                return 0;
1900
1901        /*
1902         * Try to mkdir the last path component if that failed.
1903         *
1904         * Re-try the "link()" regardless of whether the mkdir
1905         * succeeds, since a race might mean that somebody
1906         * else succeeded.
1907         */
1908        ret = errno;
1909        dir = strrchr(filename, '/');
1910        if (dir) {
1911                *dir = 0;
1912                if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1913                        *dir = '/';
1914                        return -2;
1915                }
1916                *dir = '/';
1917                if (!link(tmpfile, filename))
1918                        return 0;
1919                ret = errno;
1920        }
1921        return ret;
1922}
1923
1924/*
1925 * Move the just written object into its final resting place
1926 */
1927int move_temp_to_file(const char *tmpfile, const char *filename)
1928{
1929        int ret = link_temp_to_file(tmpfile, filename);
1930
1931        /*
1932         * Coda hack - coda doesn't like cross-directory links,
1933         * so we fall back to a rename, which will mean that it
1934         * won't be able to check collisions, but that's not a
1935         * big deal.
1936         *
1937         * The same holds for FAT formatted media.
1938         *
1939         * When this succeeds, we just return 0. We have nothing
1940         * left to unlink.
1941         */
1942        if (ret && ret != EEXIST) {
1943                if (!rename(tmpfile, filename))
1944                        return 0;
1945                ret = errno;
1946        }
1947        unlink(tmpfile);
1948        if (ret) {
1949                if (ret != EEXIST) {
1950                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1951                }
1952                /* FIXME!!! Collision check here ? */
1953        }
1954
1955        return 0;
1956}
1957
1958static int write_buffer(int fd, const void *buf, size_t len)
1959{
1960        if (write_in_full(fd, buf, len) < 0)
1961                return error("file write error (%s)", strerror(errno));
1962        return 0;
1963}
1964
1965static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1966{
1967        int hdr_len;
1968        unsigned char c;
1969
1970        c = (type << 4) | (len & 15);
1971        len >>= 4;
1972        hdr_len = 1;
1973        while (len) {
1974                *hdr++ = c | 0x80;
1975                hdr_len++;
1976                c = (len & 0x7f);
1977                len >>= 7;
1978        }
1979        *hdr = c;
1980        return hdr_len;
1981}
1982
1983static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1984{
1985        int obj_type, hdrlen;
1986
1987        if (use_legacy_headers) {
1988                while (deflate(stream, 0) == Z_OK)
1989                        /* nothing */;
1990                return;
1991        }
1992        obj_type = type_from_string(type);
1993        hdrlen = write_binary_header(stream->next_out, obj_type, len);
1994        stream->total_out = hdrlen;
1995        stream->next_out += hdrlen;
1996        stream->avail_out -= hdrlen;
1997}
1998
1999int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2000                   unsigned char *sha1)
2001{
2002        char hdr[32];
2003        int hdrlen;
2004        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2005        return 0;
2006}
2007
2008int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2009{
2010        int size, ret;
2011        unsigned char *compressed;
2012        z_stream stream;
2013        unsigned char sha1[20];
2014        char *filename;
2015        static char tmpfile[PATH_MAX];
2016        char hdr[32];
2017        int fd, hdrlen;
2018
2019        /* Normally if we have it in the pack then we do not bother writing
2020         * it out into .git/objects/??/?{38} file.
2021         */
2022        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2023        filename = sha1_file_name(sha1);
2024        if (returnsha1)
2025                hashcpy(returnsha1, sha1);
2026        if (has_sha1_file(sha1))
2027                return 0;
2028        fd = open(filename, O_RDONLY);
2029        if (fd >= 0) {
2030                /*
2031                 * FIXME!!! We might do collision checking here, but we'd
2032                 * need to uncompress the old file and check it. Later.
2033                 */
2034                close(fd);
2035                return 0;
2036        }
2037
2038        if (errno != ENOENT) {
2039                return error("sha1 file %s: %s\n", filename, strerror(errno));
2040        }
2041
2042        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2043
2044        fd = mkstemp(tmpfile);
2045        if (fd < 0) {
2046                if (errno == EPERM)
2047                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2048                else
2049                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2050        }
2051
2052        /* Set it up */
2053        memset(&stream, 0, sizeof(stream));
2054        deflateInit(&stream, zlib_compression_level);
2055        size = 8 + deflateBound(&stream, len+hdrlen);
2056        compressed = xmalloc(size);
2057
2058        /* Compress it */
2059        stream.next_out = compressed;
2060        stream.avail_out = size;
2061
2062        /* First header.. */
2063        stream.next_in = (unsigned char *)hdr;
2064        stream.avail_in = hdrlen;
2065        setup_object_header(&stream, type, len);
2066
2067        /* Then the data itself.. */
2068        stream.next_in = buf;
2069        stream.avail_in = len;
2070        ret = deflate(&stream, Z_FINISH);
2071        if (ret != Z_STREAM_END)
2072                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2073
2074        ret = deflateEnd(&stream);
2075        if (ret != Z_OK)
2076                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2077
2078        size = stream.total_out;
2079
2080        if (write_buffer(fd, compressed, size) < 0)
2081                die("unable to write sha1 file");
2082        fchmod(fd, 0444);
2083        if (close(fd))
2084                die("unable to write sha1 file");
2085        free(compressed);
2086
2087        return move_temp_to_file(tmpfile, filename);
2088}
2089
2090/*
2091 * We need to unpack and recompress the object for writing
2092 * it out to a different file.
2093 */
2094static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2095{
2096        size_t size;
2097        z_stream stream;
2098        unsigned char *unpacked;
2099        unsigned long len;
2100        enum object_type type;
2101        char hdr[32];
2102        int hdrlen;
2103        void *buf;
2104
2105        /* need to unpack and recompress it by itself */
2106        unpacked = read_packed_sha1(sha1, &type, &len);
2107        if (!unpacked)
2108                error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2109
2110        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2111
2112        /* Set it up */
2113        memset(&stream, 0, sizeof(stream));
2114        deflateInit(&stream, zlib_compression_level);
2115        size = deflateBound(&stream, len + hdrlen);
2116        buf = xmalloc(size);
2117
2118        /* Compress it */
2119        stream.next_out = buf;
2120        stream.avail_out = size;
2121
2122        /* First header.. */
2123        stream.next_in = (void *)hdr;
2124        stream.avail_in = hdrlen;
2125        while (deflate(&stream, 0) == Z_OK)
2126                /* nothing */;
2127
2128        /* Then the data itself.. */
2129        stream.next_in = unpacked;
2130        stream.avail_in = len;
2131        while (deflate(&stream, Z_FINISH) == Z_OK)
2132                /* nothing */;
2133        deflateEnd(&stream);
2134        free(unpacked);
2135
2136        *objsize = stream.total_out;
2137        return buf;
2138}
2139
2140int write_sha1_to_fd(int fd, const unsigned char *sha1)
2141{
2142        int retval;
2143        unsigned long objsize;
2144        void *buf = map_sha1_file(sha1, &objsize);
2145
2146        if (buf) {
2147                retval = write_buffer(fd, buf, objsize);
2148                munmap(buf, objsize);
2149                return retval;
2150        }
2151
2152        buf = repack_object(sha1, &objsize);
2153        retval = write_buffer(fd, buf, objsize);
2154        free(buf);
2155        return retval;
2156}
2157
2158int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2159                       size_t bufsize, size_t *bufposn)
2160{
2161        char tmpfile[PATH_MAX];
2162        int local;
2163        z_stream stream;
2164        unsigned char real_sha1[20];
2165        unsigned char discard[4096];
2166        int ret;
2167        SHA_CTX c;
2168
2169        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2170
2171        local = mkstemp(tmpfile);
2172        if (local < 0) {
2173                if (errno == EPERM)
2174                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2175                else
2176                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2177        }
2178
2179        memset(&stream, 0, sizeof(stream));
2180
2181        inflateInit(&stream);
2182
2183        SHA1_Init(&c);
2184
2185        do {
2186                ssize_t size;
2187                if (*bufposn) {
2188                        stream.avail_in = *bufposn;
2189                        stream.next_in = (unsigned char *) buffer;
2190                        do {
2191                                stream.next_out = discard;
2192                                stream.avail_out = sizeof(discard);
2193                                ret = inflate(&stream, Z_SYNC_FLUSH);
2194                                SHA1_Update(&c, discard, sizeof(discard) -
2195                                            stream.avail_out);
2196                        } while (stream.avail_in && ret == Z_OK);
2197                        if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2198                                die("unable to write sha1 file");
2199                        memmove(buffer, buffer + *bufposn - stream.avail_in,
2200                                stream.avail_in);
2201                        *bufposn = stream.avail_in;
2202                        if (ret != Z_OK)
2203                                break;
2204                }
2205                size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2206                if (size <= 0) {
2207                        close(local);
2208                        unlink(tmpfile);
2209                        if (!size)
2210                                return error("Connection closed?");
2211                        perror("Reading from connection");
2212                        return -1;
2213                }
2214                *bufposn += size;
2215        } while (1);
2216        inflateEnd(&stream);
2217
2218        fchmod(local, 0444);
2219        if (close(local) != 0)
2220                die("unable to write sha1 file");
2221        SHA1_Final(real_sha1, &c);
2222        if (ret != Z_STREAM_END) {
2223                unlink(tmpfile);
2224                return error("File %s corrupted", sha1_to_hex(sha1));
2225        }
2226        if (hashcmp(sha1, real_sha1)) {
2227                unlink(tmpfile);
2228                return error("File %s has bad hash", sha1_to_hex(sha1));
2229        }
2230
2231        return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2232}
2233
2234int has_pack_index(const unsigned char *sha1)
2235{
2236        struct stat st;
2237        if (stat(sha1_pack_index_name(sha1), &st))
2238                return 0;
2239        return 1;
2240}
2241
2242int has_pack_file(const unsigned char *sha1)
2243{
2244        struct stat st;
2245        if (stat(sha1_pack_name(sha1), &st))
2246                return 0;
2247        return 1;
2248}
2249
2250int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2251{
2252        struct pack_entry e;
2253        return find_pack_entry(sha1, &e, ignore_packed);
2254}
2255
2256int has_sha1_file(const unsigned char *sha1)
2257{
2258        struct stat st;
2259        struct pack_entry e;
2260
2261        if (find_pack_entry(sha1, &e, NULL))
2262                return 1;
2263        return find_sha1_file(sha1, &st) ? 1 : 0;
2264}
2265
2266/*
2267 * reads from fd as long as possible into a supplied buffer of size bytes.
2268 * If necessary the buffer's size is increased using realloc()
2269 *
2270 * returns 0 if anything went fine and -1 otherwise
2271 *
2272 * NOTE: both buf and size may change, but even when -1 is returned
2273 * you still have to free() it yourself.
2274 */
2275int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2276{
2277        char* buf = *return_buf;
2278        unsigned long size = *return_size;
2279        int iret;
2280        unsigned long off = 0;
2281
2282        do {
2283                iret = xread(fd, buf + off, size - off);
2284                if (iret > 0) {
2285                        off += iret;
2286                        if (off == size) {
2287                                size *= 2;
2288                                buf = xrealloc(buf, size);
2289                        }
2290                }
2291        } while (iret > 0);
2292
2293        *return_buf = buf;
2294        *return_size = off;
2295
2296        if (iret < 0)
2297                return -1;
2298        return 0;
2299}
2300
2301int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2302{
2303        unsigned long size = 4096;
2304        char *buf = xmalloc(size);
2305        int ret;
2306
2307        if (read_pipe(fd, &buf, &size)) {
2308                free(buf);
2309                return -1;
2310        }
2311
2312        if (!type)
2313                type = blob_type;
2314        if (write_object)
2315                ret = write_sha1_file(buf, size, type, sha1);
2316        else
2317                ret = hash_sha1_file(buf, size, type, sha1);
2318        free(buf);
2319        return ret;
2320}
2321
2322int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2323             enum object_type type, const char *path)
2324{
2325        size_t size = xsize_t(st->st_size);
2326        void *buf = NULL;
2327        int ret, re_allocated = 0;
2328
2329        if (size)
2330                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2331        close(fd);
2332
2333        if (!type)
2334                type = OBJ_BLOB;
2335
2336        /*
2337         * Convert blobs to git internal format
2338         */
2339        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2340                unsigned long nsize = size;
2341                char *nbuf = convert_to_git(path, buf, &nsize);
2342                if (nbuf) {
2343                        munmap(buf, size);
2344                        size = nsize;
2345                        buf = nbuf;
2346                        re_allocated = 1;
2347                }
2348        }
2349
2350        if (write_object)
2351                ret = write_sha1_file(buf, size, typename(type), sha1);
2352        else
2353                ret = hash_sha1_file(buf, size, typename(type), sha1);
2354        if (re_allocated) {
2355                free(buf);
2356                return ret;
2357        }
2358        if (size)
2359                munmap(buf, size);
2360        return ret;
2361}
2362
2363int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2364{
2365        int fd;
2366        char *target;
2367        size_t len;
2368
2369        switch (st->st_mode & S_IFMT) {
2370        case S_IFREG:
2371                fd = open(path, O_RDONLY);
2372                if (fd < 0)
2373                        return error("open(\"%s\"): %s", path,
2374                                     strerror(errno));
2375                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2376                        return error("%s: failed to insert into database",
2377                                     path);
2378                break;
2379        case S_IFLNK:
2380                len = xsize_t(st->st_size);
2381                target = xmalloc(len + 1);
2382                if (readlink(path, target, len + 1) != st->st_size) {
2383                        char *errstr = strerror(errno);
2384                        free(target);
2385                        return error("readlink(\"%s\"): %s", path,
2386                                     errstr);
2387                }
2388                if (!write_object)
2389                        hash_sha1_file(target, len, blob_type, sha1);
2390                else if (write_sha1_file(target, len, blob_type, sha1))
2391                        return error("%s: failed to insert into database",
2392                                     path);
2393                free(target);
2394                break;
2395        case S_IFDIR:
2396                return resolve_gitlink_ref(path, "HEAD", sha1);
2397        default:
2398                return error("%s: unsupported file type", path);
2399        }
2400        return 0;
2401}
2402
2403int read_pack_header(int fd, struct pack_header *header)
2404{
2405        char *c = (char*)header;
2406        ssize_t remaining = sizeof(struct pack_header);
2407        do {
2408                ssize_t r = xread(fd, c, remaining);
2409                if (r <= 0)
2410                        /* "eof before pack header was fully read" */
2411                        return PH_ERROR_EOF;
2412                remaining -= r;
2413                c += r;
2414        } while (remaining > 0);
2415        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2416                /* "protocol error (pack signature mismatch detected)" */
2417                return PH_ERROR_PACK_SIGNATURE;
2418        if (!pack_version_ok(header->hdr_version))
2419                /* "protocol error (pack version unsupported)" */
2420                return PH_ERROR_PROTOCOL;
2421        return 0;
2422}