sha1_file.con commit Fix "git add -u" data corruption. (a4882c2)
   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
  36const signed 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        const char alt_file_name[] = "info/alternates";
 356        /* Given that relative_base is no longer than PATH_MAX,
 357           ensure that "path" has enough space to append "/", the
 358           file name, "info/alternates", and a trailing NUL.  */
 359        char path[PATH_MAX + 1 + sizeof alt_file_name];
 360        int fd;
 361
 362        sprintf(path, "%s/%s", relative_base, alt_file_name);
 363        fd = open(path, O_RDONLY);
 364        if (fd < 0)
 365                return;
 366        if (fstat(fd, &st) || (st.st_size == 0)) {
 367                close(fd);
 368                return;
 369        }
 370        mapsz = xsize_t(st.st_size);
 371        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 372        close(fd);
 373
 374        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 375
 376        munmap(map, mapsz);
 377}
 378
 379void prepare_alt_odb(void)
 380{
 381        const char *alt;
 382
 383        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 384        if (!alt) alt = "";
 385
 386        if (alt_odb_tail)
 387                return;
 388        alt_odb_tail = &alt_odb_list;
 389        link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
 390
 391        read_info_alternates(get_object_directory(), 0);
 392}
 393
 394static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 395{
 396        char *name = sha1_file_name(sha1);
 397        struct alternate_object_database *alt;
 398
 399        if (!stat(name, st))
 400                return name;
 401        prepare_alt_odb();
 402        for (alt = alt_odb_list; alt; alt = alt->next) {
 403                name = alt->name;
 404                fill_sha1_path(name, sha1);
 405                if (!stat(alt->base, st))
 406                        return alt->base;
 407        }
 408        return NULL;
 409}
 410
 411static unsigned int pack_used_ctr;
 412static unsigned int pack_mmap_calls;
 413static unsigned int peak_pack_open_windows;
 414static unsigned int pack_open_windows;
 415static size_t peak_pack_mapped;
 416static size_t pack_mapped;
 417struct packed_git *packed_git;
 418
 419void pack_report()
 420{
 421        fprintf(stderr,
 422                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 423                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 424                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 425                (size_t) getpagesize(),
 426                packed_git_window_size,
 427                packed_git_limit);
 428        fprintf(stderr,
 429                "pack_report: pack_used_ctr            = %10u\n"
 430                "pack_report: pack_mmap_calls          = %10u\n"
 431                "pack_report: pack_open_windows        = %10u / %10u\n"
 432                "pack_report: pack_mapped              = "
 433                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 434                pack_used_ctr,
 435                pack_mmap_calls,
 436                pack_open_windows, peak_pack_open_windows,
 437                pack_mapped, peak_pack_mapped);
 438}
 439
 440static int check_packed_git_idx(const char *path,  struct packed_git *p)
 441{
 442        void *idx_map;
 443        struct pack_idx_header *hdr;
 444        size_t idx_size;
 445        uint32_t version, nr, i, *index;
 446        int fd = open(path, O_RDONLY);
 447        struct stat st;
 448
 449        if (fd < 0)
 450                return -1;
 451        if (fstat(fd, &st)) {
 452                close(fd);
 453                return -1;
 454        }
 455        idx_size = xsize_t(st.st_size);
 456        if (idx_size < 4 * 256 + 20 + 20) {
 457                close(fd);
 458                return error("index file %s is too small", path);
 459        }
 460        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 461        close(fd);
 462
 463        hdr = idx_map;
 464        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 465                version = ntohl(hdr->idx_version);
 466                if (version < 2 || version > 2) {
 467                        munmap(idx_map, idx_size);
 468                        return error("index file %s is version %d"
 469                                     " and is not supported by this binary"
 470                                     " (try upgrading GIT to a newer version)",
 471                                     path, version);
 472                }
 473        } else
 474                version = 1;
 475
 476        nr = 0;
 477        index = idx_map;
 478        if (version > 1)
 479                index += 2;  /* skip index header */
 480        for (i = 0; i < 256; i++) {
 481                uint32_t n = ntohl(index[i]);
 482                if (n < nr) {
 483                        munmap(idx_map, idx_size);
 484                        return error("non-monotonic index %s", path);
 485                }
 486                nr = n;
 487        }
 488
 489        if (version == 1) {
 490                /*
 491                 * Total size:
 492                 *  - 256 index entries 4 bytes each
 493                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 494                 *  - 20-byte SHA1 of the packfile
 495                 *  - 20-byte SHA1 file checksum
 496                 */
 497                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 498                        munmap(idx_map, idx_size);
 499                        return error("wrong index file size in %s", path);
 500                }
 501        } else if (version == 2) {
 502                /*
 503                 * Minimum size:
 504                 *  - 8 bytes of header
 505                 *  - 256 index entries 4 bytes each
 506                 *  - 20-byte sha1 entry * nr
 507                 *  - 4-byte crc entry * nr
 508                 *  - 4-byte offset entry * nr
 509                 *  - 20-byte SHA1 of the packfile
 510                 *  - 20-byte SHA1 file checksum
 511                 * And after the 4-byte offset table might be a
 512                 * variable sized table containing 8-byte entries
 513                 * for offsets larger than 2^31.
 514                 */
 515                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 516                unsigned long max_size = min_size;
 517                if (nr)
 518                        max_size += (nr - 1)*8;
 519                if (idx_size < min_size || idx_size > max_size) {
 520                        munmap(idx_map, idx_size);
 521                        return error("wrong index file size in %s", path);
 522                }
 523                if (idx_size != min_size) {
 524                        /* make sure we can deal with large pack offsets */
 525                        off_t x = 0x7fffffffUL, y = 0xffffffffUL;
 526                        if (x > (x + 1) || y > (y + 1)) {
 527                                munmap(idx_map, idx_size);
 528                                return error("pack too large for current definition of off_t in %s", path);
 529                        }
 530                }
 531        }
 532
 533        p->index_version = version;
 534        p->index_data = idx_map;
 535        p->index_size = idx_size;
 536        p->num_objects = nr;
 537        return 0;
 538}
 539
 540static void scan_windows(struct packed_git *p,
 541        struct packed_git **lru_p,
 542        struct pack_window **lru_w,
 543        struct pack_window **lru_l)
 544{
 545        struct pack_window *w, *w_l;
 546
 547        for (w_l = NULL, w = p->windows; w; w = w->next) {
 548                if (!w->inuse_cnt) {
 549                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 550                                *lru_p = p;
 551                                *lru_w = w;
 552                                *lru_l = w_l;
 553                        }
 554                }
 555                w_l = w;
 556        }
 557}
 558
 559static int unuse_one_window(struct packed_git *current, int keep_fd)
 560{
 561        struct packed_git *p, *lru_p = NULL;
 562        struct pack_window *lru_w = NULL, *lru_l = NULL;
 563
 564        if (current)
 565                scan_windows(current, &lru_p, &lru_w, &lru_l);
 566        for (p = packed_git; p; p = p->next)
 567                scan_windows(p, &lru_p, &lru_w, &lru_l);
 568        if (lru_p) {
 569                munmap(lru_w->base, lru_w->len);
 570                pack_mapped -= lru_w->len;
 571                if (lru_l)
 572                        lru_l->next = lru_w->next;
 573                else {
 574                        lru_p->windows = lru_w->next;
 575                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 576                                close(lru_p->pack_fd);
 577                                lru_p->pack_fd = -1;
 578                        }
 579                }
 580                free(lru_w);
 581                pack_open_windows--;
 582                return 1;
 583        }
 584        return 0;
 585}
 586
 587void release_pack_memory(size_t need, int fd)
 588{
 589        size_t cur = pack_mapped;
 590        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 591                ; /* nothing */
 592}
 593
 594void unuse_pack(struct pack_window **w_cursor)
 595{
 596        struct pack_window *w = *w_cursor;
 597        if (w) {
 598                w->inuse_cnt--;
 599                *w_cursor = NULL;
 600        }
 601}
 602
 603/*
 604 * Do not call this directly as this leaks p->pack_fd on error return;
 605 * call open_packed_git() instead.
 606 */
 607static int open_packed_git_1(struct packed_git *p)
 608{
 609        struct stat st;
 610        struct pack_header hdr;
 611        unsigned char sha1[20];
 612        unsigned char *idx_sha1;
 613        long fd_flag;
 614
 615        p->pack_fd = open(p->pack_name, O_RDONLY);
 616        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 617                return -1;
 618
 619        /* If we created the struct before we had the pack we lack size. */
 620        if (!p->pack_size) {
 621                if (!S_ISREG(st.st_mode))
 622                        return error("packfile %s not a regular file", p->pack_name);
 623                p->pack_size = st.st_size;
 624        } else if (p->pack_size != st.st_size)
 625                return error("packfile %s size changed", p->pack_name);
 626
 627        /* We leave these file descriptors open with sliding mmap;
 628         * there is no point keeping them open across exec(), though.
 629         */
 630        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 631        if (fd_flag < 0)
 632                return error("cannot determine file descriptor flags");
 633        fd_flag |= FD_CLOEXEC;
 634        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 635                return error("cannot set FD_CLOEXEC");
 636
 637        /* Verify we recognize this pack file format. */
 638        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 639                return error("file %s is far too short to be a packfile", p->pack_name);
 640        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 641                return error("file %s is not a GIT packfile", p->pack_name);
 642        if (!pack_version_ok(hdr.hdr_version))
 643                return error("packfile %s is version %u and not supported"
 644                        " (try upgrading GIT to a newer version)",
 645                        p->pack_name, ntohl(hdr.hdr_version));
 646
 647        /* Verify the pack matches its index. */
 648        if (p->num_objects != ntohl(hdr.hdr_entries))
 649                return error("packfile %s claims to have %u objects"
 650                             " while index indicates %u objects",
 651                             p->pack_name, ntohl(hdr.hdr_entries),
 652                             p->num_objects);
 653        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 654                return error("end of packfile %s is unavailable", p->pack_name);
 655        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 656                return error("packfile %s signature is unavailable", p->pack_name);
 657        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 658        if (hashcmp(sha1, idx_sha1))
 659                return error("packfile %s does not match index", p->pack_name);
 660        return 0;
 661}
 662
 663static int open_packed_git(struct packed_git *p)
 664{
 665        if (!open_packed_git_1(p))
 666                return 0;
 667        if (p->pack_fd != -1) {
 668                close(p->pack_fd);
 669                p->pack_fd = -1;
 670        }
 671        return -1;
 672}
 673
 674static int in_window(struct pack_window *win, off_t offset)
 675{
 676        /* We must promise at least 20 bytes (one hash) after the
 677         * offset is available from this window, otherwise the offset
 678         * is not actually in this window and a different window (which
 679         * has that one hash excess) must be used.  This is to support
 680         * the object header and delta base parsing routines below.
 681         */
 682        off_t win_off = win->offset;
 683        return win_off <= offset
 684                && (offset + 20) <= (win_off + win->len);
 685}
 686
 687unsigned char* use_pack(struct packed_git *p,
 688                struct pack_window **w_cursor,
 689                off_t offset,
 690                unsigned int *left)
 691{
 692        struct pack_window *win = *w_cursor;
 693
 694        if (p->pack_fd == -1 && open_packed_git(p))
 695                die("packfile %s cannot be accessed", p->pack_name);
 696
 697        /* Since packfiles end in a hash of their content and its
 698         * pointless to ask for an offset into the middle of that
 699         * hash, and the in_window function above wouldn't match
 700         * don't allow an offset too close to the end of the file.
 701         */
 702        if (offset > (p->pack_size - 20))
 703                die("offset beyond end of packfile (truncated pack?)");
 704
 705        if (!win || !in_window(win, offset)) {
 706                if (win)
 707                        win->inuse_cnt--;
 708                for (win = p->windows; win; win = win->next) {
 709                        if (in_window(win, offset))
 710                                break;
 711                }
 712                if (!win) {
 713                        size_t window_align = packed_git_window_size / 2;
 714                        off_t len;
 715                        win = xcalloc(1, sizeof(*win));
 716                        win->offset = (offset / window_align) * window_align;
 717                        len = p->pack_size - win->offset;
 718                        if (len > packed_git_window_size)
 719                                len = packed_git_window_size;
 720                        win->len = (size_t)len;
 721                        pack_mapped += win->len;
 722                        while (packed_git_limit < pack_mapped
 723                                && unuse_one_window(p, p->pack_fd))
 724                                ; /* nothing */
 725                        win->base = xmmap(NULL, win->len,
 726                                PROT_READ, MAP_PRIVATE,
 727                                p->pack_fd, win->offset);
 728                        if (win->base == MAP_FAILED)
 729                                die("packfile %s cannot be mapped: %s",
 730                                        p->pack_name,
 731                                        strerror(errno));
 732                        pack_mmap_calls++;
 733                        pack_open_windows++;
 734                        if (pack_mapped > peak_pack_mapped)
 735                                peak_pack_mapped = pack_mapped;
 736                        if (pack_open_windows > peak_pack_open_windows)
 737                                peak_pack_open_windows = pack_open_windows;
 738                        win->next = p->windows;
 739                        p->windows = win;
 740                }
 741        }
 742        if (win != *w_cursor) {
 743                win->last_used = pack_used_ctr++;
 744                win->inuse_cnt++;
 745                *w_cursor = win;
 746        }
 747        offset -= win->offset;
 748        if (left)
 749                *left = win->len - xsize_t(offset);
 750        return win->base + offset;
 751}
 752
 753struct packed_git *add_packed_git(const char *path, int path_len, int local)
 754{
 755        struct stat st;
 756        struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
 757
 758        /*
 759         * Make sure a corresponding .pack file exists and that
 760         * the index looks sane.
 761         */
 762        path_len -= strlen(".idx");
 763        if (path_len < 1)
 764                return NULL;
 765        memcpy(p->pack_name, path, path_len);
 766        strcpy(p->pack_name + path_len, ".pack");
 767        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
 768            check_packed_git_idx(path, p)) {
 769                free(p);
 770                return NULL;
 771        }
 772
 773        /* ok, it looks sane as far as we can check without
 774         * actually mapping the pack file.
 775         */
 776        p->pack_size = st.st_size;
 777        p->next = NULL;
 778        p->windows = NULL;
 779        p->pack_fd = -1;
 780        p->pack_local = local;
 781        p->mtime = st.st_mtime;
 782        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 783                hashclr(p->sha1);
 784        return p;
 785}
 786
 787struct packed_git *parse_pack_index(unsigned char *sha1)
 788{
 789        char *path = sha1_pack_index_name(sha1);
 790        return parse_pack_index_file(sha1, path);
 791}
 792
 793struct packed_git *parse_pack_index_file(const unsigned char *sha1,
 794                                         const char *idx_path)
 795{
 796        const char *path = sha1_pack_name(sha1);
 797        struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
 798
 799        if (check_packed_git_idx(idx_path, p)) {
 800                free(p);
 801                return NULL;
 802        }
 803
 804        strcpy(p->pack_name, path);
 805        p->pack_size = 0;
 806        p->next = NULL;
 807        p->windows = NULL;
 808        p->pack_fd = -1;
 809        hashcpy(p->sha1, sha1);
 810        return p;
 811}
 812
 813void install_packed_git(struct packed_git *pack)
 814{
 815        pack->next = packed_git;
 816        packed_git = pack;
 817}
 818
 819static void prepare_packed_git_one(char *objdir, int local)
 820{
 821        /* Ensure that this buffer is large enough so that we can
 822           append "/pack/" without clobbering the stack even if
 823           strlen(objdir) were PATH_MAX.  */
 824        char path[PATH_MAX + 1 + 4 + 1 + 1];
 825        int len;
 826        DIR *dir;
 827        struct dirent *de;
 828
 829        sprintf(path, "%s/pack", objdir);
 830        len = strlen(path);
 831        dir = opendir(path);
 832        if (!dir) {
 833                if (errno != ENOENT)
 834                        error("unable to open object pack directory: %s: %s",
 835                              path, strerror(errno));
 836                return;
 837        }
 838        path[len++] = '/';
 839        while ((de = readdir(dir)) != NULL) {
 840                int namelen = strlen(de->d_name);
 841                struct packed_git *p;
 842
 843                if (!has_extension(de->d_name, ".idx"))
 844                        continue;
 845
 846                if (len + namelen + 1 > sizeof(path))
 847                        continue;
 848
 849                /* Don't reopen a pack we already have. */
 850                strcpy(path + len, de->d_name);
 851                for (p = packed_git; p; p = p->next) {
 852                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 853                                break;
 854                }
 855                if (p)
 856                        continue;
 857                /* See if it really is a valid .idx file with corresponding
 858                 * .pack file that we can map.
 859                 */
 860                p = add_packed_git(path, len + namelen, local);
 861                if (!p)
 862                        continue;
 863                install_packed_git(p);
 864        }
 865        closedir(dir);
 866}
 867
 868static int sort_pack(const void *a_, const void *b_)
 869{
 870        struct packed_git *a = *((struct packed_git **)a_);
 871        struct packed_git *b = *((struct packed_git **)b_);
 872        int st;
 873
 874        /*
 875         * Local packs tend to contain objects specific to our
 876         * variant of the project than remote ones.  In addition,
 877         * remote ones could be on a network mounted filesystem.
 878         * Favor local ones for these reasons.
 879         */
 880        st = a->pack_local - b->pack_local;
 881        if (st)
 882                return -st;
 883
 884        /*
 885         * Younger packs tend to contain more recent objects,
 886         * and more recent objects tend to get accessed more
 887         * often.
 888         */
 889        if (a->mtime < b->mtime)
 890                return 1;
 891        else if (a->mtime == b->mtime)
 892                return 0;
 893        return -1;
 894}
 895
 896static void rearrange_packed_git(void)
 897{
 898        struct packed_git **ary, *p;
 899        int i, n;
 900
 901        for (n = 0, p = packed_git; p; p = p->next)
 902                n++;
 903        if (n < 2)
 904                return;
 905
 906        /* prepare an array of packed_git for easier sorting */
 907        ary = xcalloc(n, sizeof(struct packed_git *));
 908        for (n = 0, p = packed_git; p; p = p->next)
 909                ary[n++] = p;
 910
 911        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 912
 913        /* link them back again */
 914        for (i = 0; i < n - 1; i++)
 915                ary[i]->next = ary[i + 1];
 916        ary[n - 1]->next = NULL;
 917        packed_git = ary[0];
 918
 919        free(ary);
 920}
 921
 922static int prepare_packed_git_run_once = 0;
 923void prepare_packed_git(void)
 924{
 925        struct alternate_object_database *alt;
 926
 927        if (prepare_packed_git_run_once)
 928                return;
 929        prepare_packed_git_one(get_object_directory(), 1);
 930        prepare_alt_odb();
 931        for (alt = alt_odb_list; alt; alt = alt->next) {
 932                alt->name[-1] = 0;
 933                prepare_packed_git_one(alt->base, 0);
 934                alt->name[-1] = '/';
 935        }
 936        rearrange_packed_git();
 937        prepare_packed_git_run_once = 1;
 938}
 939
 940void reprepare_packed_git(void)
 941{
 942        prepare_packed_git_run_once = 0;
 943        prepare_packed_git();
 944}
 945
 946int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 947{
 948        unsigned char real_sha1[20];
 949        hash_sha1_file(map, size, type, real_sha1);
 950        return hashcmp(sha1, real_sha1) ? -1 : 0;
 951}
 952
 953void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 954{
 955        struct stat st;
 956        void *map;
 957        int fd;
 958        char *filename = find_sha1_file(sha1, &st);
 959
 960        if (!filename) {
 961                return NULL;
 962        }
 963
 964        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 965        if (fd < 0) {
 966                /* See if it works without O_NOATIME */
 967                switch (sha1_file_open_flag) {
 968                default:
 969                        fd = open(filename, O_RDONLY);
 970                        if (fd >= 0)
 971                                break;
 972                /* Fallthrough */
 973                case 0:
 974                        return NULL;
 975                }
 976
 977                /* If it failed once, it will probably fail again.
 978                 * Stop using O_NOATIME
 979                 */
 980                sha1_file_open_flag = 0;
 981        }
 982        *size = xsize_t(st.st_size);
 983        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 984        close(fd);
 985        return map;
 986}
 987
 988int legacy_loose_object(unsigned char *map)
 989{
 990        unsigned int word;
 991
 992        /*
 993         * Is it a zlib-compressed buffer? If so, the first byte
 994         * must be 0x78 (15-bit window size, deflated), and the
 995         * first 16-bit word is evenly divisible by 31
 996         */
 997        word = (map[0] << 8) + map[1];
 998        if (map[0] == 0x78 && !(word % 31))
 999                return 1;
1000        else
1001                return 0;
1002}
1003
1004unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1005{
1006        unsigned shift;
1007        unsigned char c;
1008        unsigned long size;
1009        unsigned long used = 0;
1010
1011        c = buf[used++];
1012        *type = (c >> 4) & 7;
1013        size = c & 15;
1014        shift = 4;
1015        while (c & 0x80) {
1016                if (len <= used)
1017                        return 0;
1018                if (sizeof(long) * 8 <= shift)
1019                        return 0;
1020                c = buf[used++];
1021                size += (c & 0x7f) << shift;
1022                shift += 7;
1023        }
1024        *sizep = size;
1025        return used;
1026}
1027
1028static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1029{
1030        unsigned long size, used;
1031        static const char valid_loose_object_type[8] = {
1032                0, /* OBJ_EXT */
1033                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1034                0, /* "delta" and others are invalid in a loose object */
1035        };
1036        enum object_type type;
1037
1038        /* Get the data stream */
1039        memset(stream, 0, sizeof(*stream));
1040        stream->next_in = map;
1041        stream->avail_in = mapsize;
1042        stream->next_out = buffer;
1043        stream->avail_out = bufsiz;
1044
1045        if (legacy_loose_object(map)) {
1046                inflateInit(stream);
1047                return inflate(stream, 0);
1048        }
1049
1050        used = unpack_object_header_gently(map, mapsize, &type, &size);
1051        if (!used || !valid_loose_object_type[type])
1052                return -1;
1053        map += used;
1054        mapsize -= used;
1055
1056        /* Set up the stream for the rest.. */
1057        stream->next_in = map;
1058        stream->avail_in = mapsize;
1059        inflateInit(stream);
1060
1061        /* And generate the fake traditional header */
1062        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1063                                         typename(type), size);
1064        return 0;
1065}
1066
1067static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1068{
1069        int bytes = strlen(buffer) + 1;
1070        unsigned char *buf = xmalloc(1+size);
1071        unsigned long n;
1072        int status = Z_OK;
1073
1074        n = stream->total_out - bytes;
1075        if (n > size)
1076                n = size;
1077        memcpy(buf, (char *) buffer + bytes, n);
1078        bytes = n;
1079        if (bytes <= size) {
1080                /*
1081                 * The above condition must be (bytes <= size), not
1082                 * (bytes < size).  In other words, even though we
1083                 * expect no more output and set avail_out to zer0,
1084                 * the input zlib stream may have bytes that express
1085                 * "this concludes the stream", and we *do* want to
1086                 * eat that input.
1087                 *
1088                 * Otherwise we would not be able to test that we
1089                 * consumed all the input to reach the expected size;
1090                 * we also want to check that zlib tells us that all
1091                 * went well with status == Z_STREAM_END at the end.
1092                 */
1093                stream->next_out = buf + bytes;
1094                stream->avail_out = size - bytes;
1095                while (status == Z_OK)
1096                        status = inflate(stream, Z_FINISH);
1097        }
1098        buf[size] = 0;
1099        if (status == Z_STREAM_END && !stream->avail_in) {
1100                inflateEnd(stream);
1101                return buf;
1102        }
1103
1104        if (status < 0)
1105                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1106        else if (stream->avail_in)
1107                error("garbage at end of loose object '%s'",
1108                      sha1_to_hex(sha1));
1109        free(buf);
1110        return NULL;
1111}
1112
1113/*
1114 * We used to just use "sscanf()", but that's actually way
1115 * too permissive for what we want to check. So do an anal
1116 * object header parse by hand.
1117 */
1118static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1119{
1120        char type[10];
1121        int i;
1122        unsigned long size;
1123
1124        /*
1125         * The type can be at most ten bytes (including the 
1126         * terminating '\0' that we add), and is followed by
1127         * a space.
1128         */
1129        i = 0;
1130        for (;;) {
1131                char c = *hdr++;
1132                if (c == ' ')
1133                        break;
1134                type[i++] = c;
1135                if (i >= sizeof(type))
1136                        return -1;
1137        }
1138        type[i] = 0;
1139
1140        /*
1141         * The length must follow immediately, and be in canonical
1142         * decimal format (ie "010" is not valid).
1143         */
1144        size = *hdr++ - '0';
1145        if (size > 9)
1146                return -1;
1147        if (size) {
1148                for (;;) {
1149                        unsigned long c = *hdr - '0';
1150                        if (c > 9)
1151                                break;
1152                        hdr++;
1153                        size = size * 10 + c;
1154                }
1155        }
1156        *sizep = size;
1157
1158        /*
1159         * The length must be followed by a zero byte
1160         */
1161        return *hdr ? -1 : type_from_string(type);
1162}
1163
1164static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1165{
1166        int ret;
1167        z_stream stream;
1168        char hdr[8192];
1169
1170        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1171        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1172                return NULL;
1173
1174        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1175}
1176
1177unsigned long get_size_from_delta(struct packed_git *p,
1178                                  struct pack_window **w_curs,
1179                                  off_t curpos)
1180{
1181        const unsigned char *data;
1182        unsigned char delta_head[20], *in;
1183        z_stream stream;
1184        int st;
1185
1186        memset(&stream, 0, sizeof(stream));
1187        stream.next_out = delta_head;
1188        stream.avail_out = sizeof(delta_head);
1189
1190        inflateInit(&stream);
1191        do {
1192                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1193                stream.next_in = in;
1194                st = inflate(&stream, Z_FINISH);
1195                curpos += stream.next_in - in;
1196        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1197                 stream.total_out < sizeof(delta_head));
1198        inflateEnd(&stream);
1199        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1200                die("delta data unpack-initial failed");
1201
1202        /* Examine the initial part of the delta to figure out
1203         * the result size.
1204         */
1205        data = delta_head;
1206
1207        /* ignore base size */
1208        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1209
1210        /* Read the result size */
1211        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1212}
1213
1214static off_t get_delta_base(struct packed_git *p,
1215                                    struct pack_window **w_curs,
1216                                    off_t *curpos,
1217                                    enum object_type type,
1218                                    off_t delta_obj_offset)
1219{
1220        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1221        off_t base_offset;
1222
1223        /* use_pack() assured us we have [base_info, base_info + 20)
1224         * as a range that we can look at without walking off the
1225         * end of the mapped window.  Its actually the hash size
1226         * that is assured.  An OFS_DELTA longer than the hash size
1227         * is stupid, as then a REF_DELTA would be smaller to store.
1228         */
1229        if (type == OBJ_OFS_DELTA) {
1230                unsigned used = 0;
1231                unsigned char c = base_info[used++];
1232                base_offset = c & 127;
1233                while (c & 128) {
1234                        base_offset += 1;
1235                        if (!base_offset || MSB(base_offset, 7))
1236                                die("offset value overflow for delta base object");
1237                        c = base_info[used++];
1238                        base_offset = (base_offset << 7) + (c & 127);
1239                }
1240                base_offset = delta_obj_offset - base_offset;
1241                if (base_offset >= delta_obj_offset)
1242                        die("delta base offset out of bound");
1243                *curpos += used;
1244        } else if (type == OBJ_REF_DELTA) {
1245                /* The base entry _must_ be in the same pack */
1246                base_offset = find_pack_entry_one(base_info, p);
1247                if (!base_offset)
1248                        die("failed to find delta-pack base object %s",
1249                                sha1_to_hex(base_info));
1250                *curpos += 20;
1251        } else
1252                die("I am totally screwed");
1253        return base_offset;
1254}
1255
1256/* forward declaration for a mutually recursive function */
1257static int packed_object_info(struct packed_git *p, off_t offset,
1258                              unsigned long *sizep);
1259
1260static int packed_delta_info(struct packed_git *p,
1261                             struct pack_window **w_curs,
1262                             off_t curpos,
1263                             enum object_type type,
1264                             off_t obj_offset,
1265                             unsigned long *sizep)
1266{
1267        off_t base_offset;
1268
1269        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1270        type = packed_object_info(p, base_offset, NULL);
1271
1272        /* We choose to only get the type of the base object and
1273         * ignore potentially corrupt pack file that expects the delta
1274         * based on a base with a wrong size.  This saves tons of
1275         * inflate() calls.
1276         */
1277        if (sizep)
1278                *sizep = get_size_from_delta(p, w_curs, curpos);
1279
1280        return type;
1281}
1282
1283static int unpack_object_header(struct packed_git *p,
1284                                struct pack_window **w_curs,
1285                                off_t *curpos,
1286                                unsigned long *sizep)
1287{
1288        unsigned char *base;
1289        unsigned int left;
1290        unsigned long used;
1291        enum object_type type;
1292
1293        /* use_pack() assures us we have [base, base + 20) available
1294         * as a range that we can look at at.  (Its actually the hash
1295         * size that is assured.)  With our object header encoding
1296         * the maximum deflated object size is 2^137, which is just
1297         * insane, so we know won't exceed what we have been given.
1298         */
1299        base = use_pack(p, w_curs, *curpos, &left);
1300        used = unpack_object_header_gently(base, left, &type, sizep);
1301        if (!used)
1302                die("object offset outside of pack file");
1303        *curpos += used;
1304
1305        return type;
1306}
1307
1308const char *packed_object_info_detail(struct packed_git *p,
1309                                      off_t obj_offset,
1310                                      unsigned long *size,
1311                                      unsigned long *store_size,
1312                                      unsigned int *delta_chain_length,
1313                                      unsigned char *base_sha1)
1314{
1315        struct pack_window *w_curs = NULL;
1316        off_t curpos;
1317        unsigned long dummy;
1318        unsigned char *next_sha1;
1319        enum object_type type;
1320
1321        *delta_chain_length = 0;
1322        curpos = obj_offset;
1323        type = unpack_object_header(p, &w_curs, &curpos, size);
1324
1325        for (;;) {
1326                switch (type) {
1327                default:
1328                        die("pack %s contains unknown object type %d",
1329                            p->pack_name, type);
1330                case OBJ_COMMIT:
1331                case OBJ_TREE:
1332                case OBJ_BLOB:
1333                case OBJ_TAG:
1334                        *store_size = 0; /* notyet */
1335                        unuse_pack(&w_curs);
1336                        return typename(type);
1337                case OBJ_OFS_DELTA:
1338                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1339                        if (*delta_chain_length == 0) {
1340                                /* TODO: find base_sha1 as pointed by curpos */
1341                                hashclr(base_sha1);
1342                        }
1343                        break;
1344                case OBJ_REF_DELTA:
1345                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1346                        if (*delta_chain_length == 0)
1347                                hashcpy(base_sha1, next_sha1);
1348                        obj_offset = find_pack_entry_one(next_sha1, p);
1349                        break;
1350                }
1351                (*delta_chain_length)++;
1352                curpos = obj_offset;
1353                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1354        }
1355}
1356
1357static int packed_object_info(struct packed_git *p, off_t obj_offset,
1358                              unsigned long *sizep)
1359{
1360        struct pack_window *w_curs = NULL;
1361        unsigned long size;
1362        off_t curpos = obj_offset;
1363        enum object_type type;
1364
1365        type = unpack_object_header(p, &w_curs, &curpos, &size);
1366
1367        switch (type) {
1368        case OBJ_OFS_DELTA:
1369        case OBJ_REF_DELTA:
1370                type = packed_delta_info(p, &w_curs, curpos,
1371                                         type, obj_offset, sizep);
1372                break;
1373        case OBJ_COMMIT:
1374        case OBJ_TREE:
1375        case OBJ_BLOB:
1376        case OBJ_TAG:
1377                if (sizep)
1378                        *sizep = size;
1379                break;
1380        default:
1381                die("pack %s contains unknown object type %d",
1382                    p->pack_name, type);
1383        }
1384        unuse_pack(&w_curs);
1385        return type;
1386}
1387
1388static void *unpack_compressed_entry(struct packed_git *p,
1389                                    struct pack_window **w_curs,
1390                                    off_t curpos,
1391                                    unsigned long size)
1392{
1393        int st;
1394        z_stream stream;
1395        unsigned char *buffer, *in;
1396
1397        buffer = xmalloc(size + 1);
1398        buffer[size] = 0;
1399        memset(&stream, 0, sizeof(stream));
1400        stream.next_out = buffer;
1401        stream.avail_out = size;
1402
1403        inflateInit(&stream);
1404        do {
1405                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1406                stream.next_in = in;
1407                st = inflate(&stream, Z_FINISH);
1408                curpos += stream.next_in - in;
1409        } while (st == Z_OK || st == Z_BUF_ERROR);
1410        inflateEnd(&stream);
1411        if ((st != Z_STREAM_END) || stream.total_out != size) {
1412                free(buffer);
1413                return NULL;
1414        }
1415
1416        return buffer;
1417}
1418
1419#define MAX_DELTA_CACHE (256)
1420
1421static size_t delta_base_cached;
1422
1423static struct delta_base_cache_lru_list {
1424        struct delta_base_cache_lru_list *prev;
1425        struct delta_base_cache_lru_list *next;
1426} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1427
1428static struct delta_base_cache_entry {
1429        struct delta_base_cache_lru_list lru;
1430        void *data;
1431        struct packed_git *p;
1432        off_t base_offset;
1433        unsigned long size;
1434        enum object_type type;
1435} delta_base_cache[MAX_DELTA_CACHE];
1436
1437static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1438{
1439        unsigned long hash;
1440
1441        hash = (unsigned long)p + (unsigned long)base_offset;
1442        hash += (hash >> 8) + (hash >> 16);
1443        return hash % MAX_DELTA_CACHE;
1444}
1445
1446static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1447        unsigned long *base_size, enum object_type *type, int keep_cache)
1448{
1449        void *ret;
1450        unsigned long hash = pack_entry_hash(p, base_offset);
1451        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1452
1453        ret = ent->data;
1454        if (ret && ent->p == p && ent->base_offset == base_offset)
1455                goto found_cache_entry;
1456        return unpack_entry(p, base_offset, type, base_size);
1457
1458found_cache_entry:
1459        if (!keep_cache) {
1460                ent->data = NULL;
1461                ent->lru.next->prev = ent->lru.prev;
1462                ent->lru.prev->next = ent->lru.next;
1463                delta_base_cached -= ent->size;
1464        }
1465        else {
1466                ret = xmalloc(ent->size + 1);
1467                memcpy(ret, ent->data, ent->size);
1468                ((char *)ret)[ent->size] = 0;
1469        }
1470        *type = ent->type;
1471        *base_size = ent->size;
1472        return ret;
1473}
1474
1475static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1476{
1477        if (ent->data) {
1478                free(ent->data);
1479                ent->data = NULL;
1480                ent->lru.next->prev = ent->lru.prev;
1481                ent->lru.prev->next = ent->lru.next;
1482                delta_base_cached -= ent->size;
1483        }
1484}
1485
1486static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1487        void *base, unsigned long base_size, enum object_type type)
1488{
1489        unsigned long hash = pack_entry_hash(p, base_offset);
1490        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1491        struct delta_base_cache_lru_list *lru;
1492
1493        release_delta_base_cache(ent);
1494        delta_base_cached += base_size;
1495
1496        for (lru = delta_base_cache_lru.next;
1497             delta_base_cached > delta_base_cache_limit
1498             && lru != &delta_base_cache_lru;
1499             lru = lru->next) {
1500                struct delta_base_cache_entry *f = (void *)lru;
1501                if (f->type == OBJ_BLOB)
1502                        release_delta_base_cache(f);
1503        }
1504        for (lru = delta_base_cache_lru.next;
1505             delta_base_cached > delta_base_cache_limit
1506             && lru != &delta_base_cache_lru;
1507             lru = lru->next) {
1508                struct delta_base_cache_entry *f = (void *)lru;
1509                release_delta_base_cache(f);
1510        }
1511
1512        ent->p = p;
1513        ent->base_offset = base_offset;
1514        ent->type = type;
1515        ent->data = base;
1516        ent->size = base_size;
1517        ent->lru.next = &delta_base_cache_lru;
1518        ent->lru.prev = delta_base_cache_lru.prev;
1519        delta_base_cache_lru.prev->next = &ent->lru;
1520        delta_base_cache_lru.prev = &ent->lru;
1521}
1522
1523static void *unpack_delta_entry(struct packed_git *p,
1524                                struct pack_window **w_curs,
1525                                off_t curpos,
1526                                unsigned long delta_size,
1527                                off_t obj_offset,
1528                                enum object_type *type,
1529                                unsigned long *sizep)
1530{
1531        void *delta_data, *result, *base;
1532        unsigned long base_size;
1533        off_t base_offset;
1534
1535        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1536        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1537        if (!base)
1538                die("failed to read delta base object"
1539                    " at %"PRIuMAX" from %s",
1540                    (uintmax_t)base_offset, p->pack_name);
1541
1542        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1543        result = patch_delta(base, base_size,
1544                             delta_data, delta_size,
1545                             sizep);
1546        if (!result)
1547                die("failed to apply delta");
1548        free(delta_data);
1549        add_delta_base_cache(p, base_offset, base, base_size, *type);
1550        return result;
1551}
1552
1553void *unpack_entry(struct packed_git *p, off_t obj_offset,
1554                   enum object_type *type, unsigned long *sizep)
1555{
1556        struct pack_window *w_curs = NULL;
1557        off_t curpos = obj_offset;
1558        void *data;
1559
1560        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1561        switch (*type) {
1562        case OBJ_OFS_DELTA:
1563        case OBJ_REF_DELTA:
1564                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1565                                          obj_offset, type, sizep);
1566                break;
1567        case OBJ_COMMIT:
1568        case OBJ_TREE:
1569        case OBJ_BLOB:
1570        case OBJ_TAG:
1571                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1572                break;
1573        default:
1574                die("unknown object type %i in %s", *type, p->pack_name);
1575        }
1576        unuse_pack(&w_curs);
1577        return data;
1578}
1579
1580const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1581                                            uint32_t n)
1582{
1583        const unsigned char *index = p->index_data;
1584        if (n >= p->num_objects)
1585                return NULL;
1586        index += 4 * 256;
1587        if (p->index_version == 1) {
1588                return index + 24 * n + 4;
1589        } else {
1590                index += 8;
1591                return index + 20 * n;
1592        }
1593}
1594
1595static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1596{
1597        const unsigned char *index = p->index_data;
1598        index += 4 * 256;
1599        if (p->index_version == 1) {
1600                return ntohl(*((uint32_t *)(index + 24 * n)));
1601        } else {
1602                uint32_t off;
1603                index += 8 + p->num_objects * (20 + 4);
1604                off = ntohl(*((uint32_t *)(index + 4 * n)));
1605                if (!(off & 0x80000000))
1606                        return off;
1607                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1608                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1609                                   ntohl(*((uint32_t *)(index + 4)));
1610        }
1611}
1612
1613off_t find_pack_entry_one(const unsigned char *sha1,
1614                                  struct packed_git *p)
1615{
1616        const uint32_t *level1_ofs = p->index_data;
1617        const unsigned char *index = p->index_data;
1618        unsigned hi, lo;
1619
1620        if (p->index_version > 1) {
1621                level1_ofs += 2;
1622                index += 8;
1623        }
1624        index += 4 * 256;
1625        hi = ntohl(level1_ofs[*sha1]);
1626        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1627
1628        do {
1629                unsigned mi = (lo + hi) / 2;
1630                unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1631                int cmp = hashcmp(index + x, sha1);
1632                if (!cmp)
1633                        return nth_packed_object_offset(p, mi);
1634                if (cmp > 0)
1635                        hi = mi;
1636                else
1637                        lo = mi+1;
1638        } while (lo < hi);
1639        return 0;
1640}
1641
1642static int matches_pack_name(struct packed_git *p, const char *ig)
1643{
1644        const char *last_c, *c;
1645
1646        if (!strcmp(p->pack_name, ig))
1647                return 0;
1648
1649        for (c = p->pack_name, last_c = c; *c;)
1650                if (*c == '/')
1651                        last_c = ++c;
1652                else
1653                        ++c;
1654        if (!strcmp(last_c, ig))
1655                return 0;
1656
1657        return 1;
1658}
1659
1660static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1661{
1662        struct packed_git *p;
1663        off_t offset;
1664
1665        prepare_packed_git();
1666
1667        for (p = packed_git; p; p = p->next) {
1668                if (ignore_packed) {
1669                        const char **ig;
1670                        for (ig = ignore_packed; *ig; ig++)
1671                                if (!matches_pack_name(p, *ig))
1672                                        break;
1673                        if (*ig)
1674                                continue;
1675                }
1676                offset = find_pack_entry_one(sha1, p);
1677                if (offset) {
1678                        /*
1679                         * We are about to tell the caller where they can
1680                         * locate the requested object.  We better make
1681                         * sure the packfile is still here and can be
1682                         * accessed before supplying that answer, as
1683                         * it may have been deleted since the index
1684                         * was loaded!
1685                         */
1686                        if (p->pack_fd == -1 && open_packed_git(p)) {
1687                                error("packfile %s cannot be accessed", p->pack_name);
1688                                continue;
1689                        }
1690                        e->offset = offset;
1691                        e->p = p;
1692                        hashcpy(e->sha1, sha1);
1693                        return 1;
1694                }
1695        }
1696        return 0;
1697}
1698
1699struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1700                                  struct packed_git *packs)
1701{
1702        struct packed_git *p;
1703
1704        for (p = packs; p; p = p->next) {
1705                if (find_pack_entry_one(sha1, p))
1706                        return p;
1707        }
1708        return NULL;
1709
1710}
1711
1712static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1713{
1714        int status;
1715        unsigned long mapsize, size;
1716        void *map;
1717        z_stream stream;
1718        char hdr[32];
1719
1720        map = map_sha1_file(sha1, &mapsize);
1721        if (!map)
1722                return error("unable to find %s", sha1_to_hex(sha1));
1723        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1724                status = error("unable to unpack %s header",
1725                               sha1_to_hex(sha1));
1726        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1727                status = error("unable to parse %s header", sha1_to_hex(sha1));
1728        else if (sizep)
1729                *sizep = size;
1730        inflateEnd(&stream);
1731        munmap(map, mapsize);
1732        return status;
1733}
1734
1735int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1736{
1737        struct pack_entry e;
1738
1739        if (!find_pack_entry(sha1, &e, NULL)) {
1740                reprepare_packed_git();
1741                if (!find_pack_entry(sha1, &e, NULL))
1742                        return sha1_loose_object_info(sha1, sizep);
1743        }
1744        return packed_object_info(e.p, e.offset, sizep);
1745}
1746
1747static void *read_packed_sha1(const unsigned char *sha1,
1748                              enum object_type *type, unsigned long *size)
1749{
1750        struct pack_entry e;
1751
1752        if (!find_pack_entry(sha1, &e, NULL))
1753                return NULL;
1754        else
1755                return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1756}
1757
1758/*
1759 * This is meant to hold a *small* number of objects that you would
1760 * want read_sha1_file() to be able to return, but yet you do not want
1761 * to write them into the object store (e.g. a browse-only
1762 * application).
1763 */
1764static struct cached_object {
1765        unsigned char sha1[20];
1766        enum object_type type;
1767        void *buf;
1768        unsigned long size;
1769} *cached_objects;
1770static int cached_object_nr, cached_object_alloc;
1771
1772static struct cached_object *find_cached_object(const unsigned char *sha1)
1773{
1774        int i;
1775        struct cached_object *co = cached_objects;
1776
1777        for (i = 0; i < cached_object_nr; i++, co++) {
1778                if (!hashcmp(co->sha1, sha1))
1779                        return co;
1780        }
1781        return NULL;
1782}
1783
1784int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1785                      unsigned char *sha1)
1786{
1787        struct cached_object *co;
1788
1789        hash_sha1_file(buf, len, typename(type), sha1);
1790        if (has_sha1_file(sha1) || find_cached_object(sha1))
1791                return 0;
1792        if (cached_object_alloc <= cached_object_nr) {
1793                cached_object_alloc = alloc_nr(cached_object_alloc);
1794                cached_objects = xrealloc(cached_objects,
1795                                          sizeof(*cached_objects) *
1796                                          cached_object_alloc);
1797        }
1798        co = &cached_objects[cached_object_nr++];
1799        co->size = len;
1800        co->type = type;
1801        co->buf = xmalloc(len);
1802        memcpy(co->buf, buf, len);
1803        hashcpy(co->sha1, sha1);
1804        return 0;
1805}
1806
1807void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1808                     unsigned long *size)
1809{
1810        unsigned long mapsize;
1811        void *map, *buf;
1812        struct cached_object *co;
1813
1814        co = find_cached_object(sha1);
1815        if (co) {
1816                buf = xmalloc(co->size + 1);
1817                memcpy(buf, co->buf, co->size);
1818                ((char*)buf)[co->size] = 0;
1819                *type = co->type;
1820                *size = co->size;
1821                return buf;
1822        }
1823
1824        buf = read_packed_sha1(sha1, type, size);
1825        if (buf)
1826                return buf;
1827        map = map_sha1_file(sha1, &mapsize);
1828        if (map) {
1829                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1830                munmap(map, mapsize);
1831                return buf;
1832        }
1833        reprepare_packed_git();
1834        return read_packed_sha1(sha1, type, size);
1835}
1836
1837void *read_object_with_reference(const unsigned char *sha1,
1838                                 const char *required_type_name,
1839                                 unsigned long *size,
1840                                 unsigned char *actual_sha1_return)
1841{
1842        enum object_type type, required_type;
1843        void *buffer;
1844        unsigned long isize;
1845        unsigned char actual_sha1[20];
1846
1847        required_type = type_from_string(required_type_name);
1848        hashcpy(actual_sha1, sha1);
1849        while (1) {
1850                int ref_length = -1;
1851                const char *ref_type = NULL;
1852
1853                buffer = read_sha1_file(actual_sha1, &type, &isize);
1854                if (!buffer)
1855                        return NULL;
1856                if (type == required_type) {
1857                        *size = isize;
1858                        if (actual_sha1_return)
1859                                hashcpy(actual_sha1_return, actual_sha1);
1860                        return buffer;
1861                }
1862                /* Handle references */
1863                else if (type == OBJ_COMMIT)
1864                        ref_type = "tree ";
1865                else if (type == OBJ_TAG)
1866                        ref_type = "object ";
1867                else {
1868                        free(buffer);
1869                        return NULL;
1870                }
1871                ref_length = strlen(ref_type);
1872
1873                if (memcmp(buffer, ref_type, ref_length) ||
1874                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1875                        free(buffer);
1876                        return NULL;
1877                }
1878                free(buffer);
1879                /* Now we have the ID of the referred-to object in
1880                 * actual_sha1.  Check again. */
1881        }
1882}
1883
1884static void write_sha1_file_prepare(const void *buf, unsigned long len,
1885                                    const char *type, unsigned char *sha1,
1886                                    char *hdr, int *hdrlen)
1887{
1888        SHA_CTX c;
1889
1890        /* Generate the header */
1891        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1892
1893        /* Sha1.. */
1894        SHA1_Init(&c);
1895        SHA1_Update(&c, hdr, *hdrlen);
1896        SHA1_Update(&c, buf, len);
1897        SHA1_Final(sha1, &c);
1898}
1899
1900/*
1901 * Link the tempfile to the final place, possibly creating the
1902 * last directory level as you do so.
1903 *
1904 * Returns the errno on failure, 0 on success.
1905 */
1906static int link_temp_to_file(const char *tmpfile, const char *filename)
1907{
1908        int ret;
1909        char *dir;
1910
1911        if (!link(tmpfile, filename))
1912                return 0;
1913
1914        /*
1915         * Try to mkdir the last path component if that failed.
1916         *
1917         * Re-try the "link()" regardless of whether the mkdir
1918         * succeeds, since a race might mean that somebody
1919         * else succeeded.
1920         */
1921        ret = errno;
1922        dir = strrchr(filename, '/');
1923        if (dir) {
1924                *dir = 0;
1925                if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1926                        *dir = '/';
1927                        return -2;
1928                }
1929                *dir = '/';
1930                if (!link(tmpfile, filename))
1931                        return 0;
1932                ret = errno;
1933        }
1934        return ret;
1935}
1936
1937/*
1938 * Move the just written object into its final resting place
1939 */
1940int move_temp_to_file(const char *tmpfile, const char *filename)
1941{
1942        int ret = link_temp_to_file(tmpfile, filename);
1943
1944        /*
1945         * Coda hack - coda doesn't like cross-directory links,
1946         * so we fall back to a rename, which will mean that it
1947         * won't be able to check collisions, but that's not a
1948         * big deal.
1949         *
1950         * The same holds for FAT formatted media.
1951         *
1952         * When this succeeds, we just return 0. We have nothing
1953         * left to unlink.
1954         */
1955        if (ret && ret != EEXIST) {
1956                if (!rename(tmpfile, filename))
1957                        return 0;
1958                ret = errno;
1959        }
1960        unlink(tmpfile);
1961        if (ret) {
1962                if (ret != EEXIST) {
1963                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1964                }
1965                /* FIXME!!! Collision check here ? */
1966        }
1967
1968        return 0;
1969}
1970
1971static int write_buffer(int fd, const void *buf, size_t len)
1972{
1973        if (write_in_full(fd, buf, len) < 0)
1974                return error("file write error (%s)", strerror(errno));
1975        return 0;
1976}
1977
1978static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1979{
1980        int hdr_len;
1981        unsigned char c;
1982
1983        c = (type << 4) | (len & 15);
1984        len >>= 4;
1985        hdr_len = 1;
1986        while (len) {
1987                *hdr++ = c | 0x80;
1988                hdr_len++;
1989                c = (len & 0x7f);
1990                len >>= 7;
1991        }
1992        *hdr = c;
1993        return hdr_len;
1994}
1995
1996static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1997{
1998        int obj_type, hdrlen;
1999
2000        if (use_legacy_headers) {
2001                while (deflate(stream, 0) == Z_OK)
2002                        /* nothing */;
2003                return;
2004        }
2005        obj_type = type_from_string(type);
2006        hdrlen = write_binary_header(stream->next_out, obj_type, len);
2007        stream->total_out = hdrlen;
2008        stream->next_out += hdrlen;
2009        stream->avail_out -= hdrlen;
2010}
2011
2012int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2013                   unsigned char *sha1)
2014{
2015        char hdr[32];
2016        int hdrlen;
2017        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2018        return 0;
2019}
2020
2021int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2022{
2023        int size, ret;
2024        unsigned char *compressed;
2025        z_stream stream;
2026        unsigned char sha1[20];
2027        char *filename;
2028        static char tmpfile[PATH_MAX];
2029        char hdr[32];
2030        int fd, hdrlen;
2031
2032        /* Normally if we have it in the pack then we do not bother writing
2033         * it out into .git/objects/??/?{38} file.
2034         */
2035        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2036        filename = sha1_file_name(sha1);
2037        if (returnsha1)
2038                hashcpy(returnsha1, sha1);
2039        if (has_sha1_file(sha1))
2040                return 0;
2041        fd = open(filename, O_RDONLY);
2042        if (fd >= 0) {
2043                /*
2044                 * FIXME!!! We might do collision checking here, but we'd
2045                 * need to uncompress the old file and check it. Later.
2046                 */
2047                close(fd);
2048                return 0;
2049        }
2050
2051        if (errno != ENOENT) {
2052                return error("sha1 file %s: %s\n", filename, strerror(errno));
2053        }
2054
2055        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2056
2057        fd = mkstemp(tmpfile);
2058        if (fd < 0) {
2059                if (errno == EPERM)
2060                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2061                else
2062                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2063        }
2064
2065        /* Set it up */
2066        memset(&stream, 0, sizeof(stream));
2067        deflateInit(&stream, zlib_compression_level);
2068        size = 8 + deflateBound(&stream, len+hdrlen);
2069        compressed = xmalloc(size);
2070
2071        /* Compress it */
2072        stream.next_out = compressed;
2073        stream.avail_out = size;
2074
2075        /* First header.. */
2076        stream.next_in = (unsigned char *)hdr;
2077        stream.avail_in = hdrlen;
2078        setup_object_header(&stream, type, len);
2079
2080        /* Then the data itself.. */
2081        stream.next_in = buf;
2082        stream.avail_in = len;
2083        ret = deflate(&stream, Z_FINISH);
2084        if (ret != Z_STREAM_END)
2085                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2086
2087        ret = deflateEnd(&stream);
2088        if (ret != Z_OK)
2089                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2090
2091        size = stream.total_out;
2092
2093        if (write_buffer(fd, compressed, size) < 0)
2094                die("unable to write sha1 file");
2095        fchmod(fd, 0444);
2096        if (close(fd))
2097                die("unable to write sha1 file");
2098        free(compressed);
2099
2100        return move_temp_to_file(tmpfile, filename);
2101}
2102
2103/*
2104 * We need to unpack and recompress the object for writing
2105 * it out to a different file.
2106 */
2107static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2108{
2109        size_t size;
2110        z_stream stream;
2111        unsigned char *unpacked;
2112        unsigned long len;
2113        enum object_type type;
2114        char hdr[32];
2115        int hdrlen;
2116        void *buf;
2117
2118        /* need to unpack and recompress it by itself */
2119        unpacked = read_packed_sha1(sha1, &type, &len);
2120        if (!unpacked)
2121                error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2122
2123        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2124
2125        /* Set it up */
2126        memset(&stream, 0, sizeof(stream));
2127        deflateInit(&stream, zlib_compression_level);
2128        size = deflateBound(&stream, len + hdrlen);
2129        buf = xmalloc(size);
2130
2131        /* Compress it */
2132        stream.next_out = buf;
2133        stream.avail_out = size;
2134
2135        /* First header.. */
2136        stream.next_in = (void *)hdr;
2137        stream.avail_in = hdrlen;
2138        while (deflate(&stream, 0) == Z_OK)
2139                /* nothing */;
2140
2141        /* Then the data itself.. */
2142        stream.next_in = unpacked;
2143        stream.avail_in = len;
2144        while (deflate(&stream, Z_FINISH) == Z_OK)
2145                /* nothing */;
2146        deflateEnd(&stream);
2147        free(unpacked);
2148
2149        *objsize = stream.total_out;
2150        return buf;
2151}
2152
2153int write_sha1_to_fd(int fd, const unsigned char *sha1)
2154{
2155        int retval;
2156        unsigned long objsize;
2157        void *buf = map_sha1_file(sha1, &objsize);
2158
2159        if (buf) {
2160                retval = write_buffer(fd, buf, objsize);
2161                munmap(buf, objsize);
2162                return retval;
2163        }
2164
2165        buf = repack_object(sha1, &objsize);
2166        retval = write_buffer(fd, buf, objsize);
2167        free(buf);
2168        return retval;
2169}
2170
2171int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2172                       size_t bufsize, size_t *bufposn)
2173{
2174        char tmpfile[PATH_MAX];
2175        int local;
2176        z_stream stream;
2177        unsigned char real_sha1[20];
2178        unsigned char discard[4096];
2179        int ret;
2180        SHA_CTX c;
2181
2182        snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
2183
2184        local = mkstemp(tmpfile);
2185        if (local < 0) {
2186                if (errno == EPERM)
2187                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2188                else
2189                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2190        }
2191
2192        memset(&stream, 0, sizeof(stream));
2193
2194        inflateInit(&stream);
2195
2196        SHA1_Init(&c);
2197
2198        do {
2199                ssize_t size;
2200                if (*bufposn) {
2201                        stream.avail_in = *bufposn;
2202                        stream.next_in = (unsigned char *) buffer;
2203                        do {
2204                                stream.next_out = discard;
2205                                stream.avail_out = sizeof(discard);
2206                                ret = inflate(&stream, Z_SYNC_FLUSH);
2207                                SHA1_Update(&c, discard, sizeof(discard) -
2208                                            stream.avail_out);
2209                        } while (stream.avail_in && ret == Z_OK);
2210                        if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2211                                die("unable to write sha1 file");
2212                        memmove(buffer, buffer + *bufposn - stream.avail_in,
2213                                stream.avail_in);
2214                        *bufposn = stream.avail_in;
2215                        if (ret != Z_OK)
2216                                break;
2217                }
2218                size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2219                if (size <= 0) {
2220                        close(local);
2221                        unlink(tmpfile);
2222                        if (!size)
2223                                return error("Connection closed?");
2224                        perror("Reading from connection");
2225                        return -1;
2226                }
2227                *bufposn += size;
2228        } while (1);
2229        inflateEnd(&stream);
2230
2231        fchmod(local, 0444);
2232        if (close(local) != 0)
2233                die("unable to write sha1 file");
2234        SHA1_Final(real_sha1, &c);
2235        if (ret != Z_STREAM_END) {
2236                unlink(tmpfile);
2237                return error("File %s corrupted", sha1_to_hex(sha1));
2238        }
2239        if (hashcmp(sha1, real_sha1)) {
2240                unlink(tmpfile);
2241                return error("File %s has bad hash", sha1_to_hex(sha1));
2242        }
2243
2244        return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2245}
2246
2247int has_pack_index(const unsigned char *sha1)
2248{
2249        struct stat st;
2250        if (stat(sha1_pack_index_name(sha1), &st))
2251                return 0;
2252        return 1;
2253}
2254
2255int has_pack_file(const unsigned char *sha1)
2256{
2257        struct stat st;
2258        if (stat(sha1_pack_name(sha1), &st))
2259                return 0;
2260        return 1;
2261}
2262
2263int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2264{
2265        struct pack_entry e;
2266        return find_pack_entry(sha1, &e, ignore_packed);
2267}
2268
2269int has_sha1_file(const unsigned char *sha1)
2270{
2271        struct stat st;
2272        struct pack_entry e;
2273
2274        if (find_pack_entry(sha1, &e, NULL))
2275                return 1;
2276        return find_sha1_file(sha1, &st) ? 1 : 0;
2277}
2278
2279/*
2280 * reads from fd as long as possible into a supplied buffer of size bytes.
2281 * If necessary the buffer's size is increased using realloc()
2282 *
2283 * returns 0 if anything went fine and -1 otherwise
2284 *
2285 * NOTE: both buf and size may change, but even when -1 is returned
2286 * you still have to free() it yourself.
2287 */
2288int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2289{
2290        char* buf = *return_buf;
2291        unsigned long size = *return_size;
2292        ssize_t iret;
2293        unsigned long off = 0;
2294
2295        do {
2296                iret = xread(fd, buf + off, size - off);
2297                if (iret > 0) {
2298                        off += iret;
2299                        if (off == size) {
2300                                size *= 2;
2301                                buf = xrealloc(buf, size);
2302                        }
2303                }
2304        } while (iret > 0);
2305
2306        *return_buf = buf;
2307        *return_size = off;
2308
2309        if (iret < 0)
2310                return -1;
2311        return 0;
2312}
2313
2314int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2315{
2316        unsigned long size = 4096;
2317        char *buf = xmalloc(size);
2318        int ret;
2319
2320        if (read_pipe(fd, &buf, &size)) {
2321                free(buf);
2322                return -1;
2323        }
2324
2325        if (!type)
2326                type = blob_type;
2327        if (write_object)
2328                ret = write_sha1_file(buf, size, type, sha1);
2329        else
2330                ret = hash_sha1_file(buf, size, type, sha1);
2331        free(buf);
2332        return ret;
2333}
2334
2335int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2336             enum object_type type, const char *path)
2337{
2338        size_t size = xsize_t(st->st_size);
2339        void *buf = NULL;
2340        int ret, re_allocated = 0;
2341
2342        if (size)
2343                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2344        close(fd);
2345
2346        if (!type)
2347                type = OBJ_BLOB;
2348
2349        /*
2350         * Convert blobs to git internal format
2351         */
2352        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2353                unsigned long nsize = size;
2354                char *nbuf = convert_to_git(path, buf, &nsize);
2355                if (nbuf) {
2356                        munmap(buf, size);
2357                        size = nsize;
2358                        buf = nbuf;
2359                        re_allocated = 1;
2360                }
2361        }
2362
2363        if (write_object)
2364                ret = write_sha1_file(buf, size, typename(type), sha1);
2365        else
2366                ret = hash_sha1_file(buf, size, typename(type), sha1);
2367        if (re_allocated) {
2368                free(buf);
2369                return ret;
2370        }
2371        if (size)
2372                munmap(buf, size);
2373        return ret;
2374}
2375
2376int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2377{
2378        int fd;
2379        char *target;
2380        size_t len;
2381
2382        switch (st->st_mode & S_IFMT) {
2383        case S_IFREG:
2384                fd = open(path, O_RDONLY);
2385                if (fd < 0)
2386                        return error("open(\"%s\"): %s", path,
2387                                     strerror(errno));
2388                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2389                        return error("%s: failed to insert into database",
2390                                     path);
2391                break;
2392        case S_IFLNK:
2393                len = xsize_t(st->st_size);
2394                target = xmalloc(len + 1);
2395                if (readlink(path, target, len + 1) != st->st_size) {
2396                        char *errstr = strerror(errno);
2397                        free(target);
2398                        return error("readlink(\"%s\"): %s", path,
2399                                     errstr);
2400                }
2401                if (!write_object)
2402                        hash_sha1_file(target, len, blob_type, sha1);
2403                else if (write_sha1_file(target, len, blob_type, sha1))
2404                        return error("%s: failed to insert into database",
2405                                     path);
2406                free(target);
2407                break;
2408        case S_IFDIR:
2409                return resolve_gitlink_ref(path, "HEAD", sha1);
2410        default:
2411                return error("%s: unsupported file type", path);
2412        }
2413        return 0;
2414}
2415
2416int read_pack_header(int fd, struct pack_header *header)
2417{
2418        char *c = (char*)header;
2419        ssize_t remaining = sizeof(struct pack_header);
2420        do {
2421                ssize_t r = xread(fd, c, remaining);
2422                if (r <= 0)
2423                        /* "eof before pack header was fully read" */
2424                        return PH_ERROR_EOF;
2425                remaining -= r;
2426                c += r;
2427        } while (remaining > 0);
2428        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2429                /* "protocol error (pack signature mismatch detected)" */
2430                return PH_ERROR_PACK_SIGNATURE;
2431        if (!pack_version_ok(header->hdr_version))
2432                /* "protocol error (pack version unsupported)" */
2433                return PH_ERROR_PROTOCOL;
2434        return 0;
2435}