sha1_file.con commit push: receiver end advertises refs from alternate repositories (d79796b)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This handles basic git sha1 object files - packing, unpacking,
   7 * creation etc.
   8 */
   9#include "cache.h"
  10#include "delta.h"
  11#include "pack.h"
  12#include "blob.h"
  13#include "commit.h"
  14#include "tag.h"
  15#include "tree.h"
  16#include "refs.h"
  17#include "pack-revindex.h"
  18#include "sha1-lookup.h"
  19
  20#ifndef O_NOATIME
  21#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  22#define O_NOATIME 01000000
  23#else
  24#define O_NOATIME 0
  25#endif
  26#endif
  27
  28#ifdef NO_C99_FORMAT
  29#define SZ_FMT "lu"
  30static unsigned long sz_fmt(size_t s) { return (unsigned long)s; }
  31#else
  32#define SZ_FMT "zu"
  33static size_t sz_fmt(size_t s) { return s; }
  34#endif
  35
  36const unsigned char null_sha1[20];
  37
  38const signed char hexval_table[256] = {
  39         -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
  40         -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
  41         -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
  42         -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
  43         -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
  44         -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
  45          0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
  46          8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
  47         -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
  48         -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
  49         -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
  50         -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
  51         -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
  52         -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
  53         -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
  54         -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
  55         -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
  56         -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
  57         -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
  58         -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
  59         -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
  60         -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
  61         -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
  62         -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
  63         -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
  64         -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
  65         -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
  66         -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
  67         -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
  68         -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
  69         -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
  70         -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
  71};
  72
  73int get_sha1_hex(const char *hex, unsigned char *sha1)
  74{
  75        int i;
  76        for (i = 0; i < 20; i++) {
  77                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  78                if (val & ~0xff)
  79                        return -1;
  80                *sha1++ = val;
  81                hex += 2;
  82        }
  83        return 0;
  84}
  85
  86static inline int offset_1st_component(const char *path)
  87{
  88        if (has_dos_drive_prefix(path))
  89                return 2 + (path[2] == '/');
  90        return *path == '/';
  91}
  92
  93int safe_create_leading_directories(char *path)
  94{
  95        char *pos = path + offset_1st_component(path);
  96        struct stat st;
  97
  98        while (pos) {
  99                pos = strchr(pos, '/');
 100                if (!pos)
 101                        break;
 102                *pos = 0;
 103                if (!stat(path, &st)) {
 104                        /* path exists */
 105                        if (!S_ISDIR(st.st_mode)) {
 106                                *pos = '/';
 107                                return -3;
 108                        }
 109                }
 110                else if (mkdir(path, 0777)) {
 111                        *pos = '/';
 112                        return -1;
 113                }
 114                else if (adjust_shared_perm(path)) {
 115                        *pos = '/';
 116                        return -2;
 117                }
 118                *pos++ = '/';
 119        }
 120        return 0;
 121}
 122
 123int safe_create_leading_directories_const(const char *path)
 124{
 125        /* path points to cache entries, so xstrdup before messing with it */
 126        char *buf = xstrdup(path);
 127        int result = safe_create_leading_directories(buf);
 128        free(buf);
 129        return result;
 130}
 131
 132char *sha1_to_hex(const unsigned char *sha1)
 133{
 134        static int bufno;
 135        static char hexbuffer[4][50];
 136        static const char hex[] = "0123456789abcdef";
 137        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 138        int i;
 139
 140        for (i = 0; i < 20; i++) {
 141                unsigned int val = *sha1++;
 142                *buf++ = hex[val >> 4];
 143                *buf++ = hex[val & 0xf];
 144        }
 145        *buf = '\0';
 146
 147        return buffer;
 148}
 149
 150static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 151{
 152        int i;
 153        for (i = 0; i < 20; i++) {
 154                static char hex[] = "0123456789abcdef";
 155                unsigned int val = sha1[i];
 156                char *pos = pathbuf + i*2 + (i > 0);
 157                *pos++ = hex[val >> 4];
 158                *pos = hex[val & 0xf];
 159        }
 160}
 161
 162/*
 163 * NOTE! This returns a statically allocated buffer, so you have to be
 164 * careful about using it. Do an "xstrdup()" if you need to save the
 165 * filename.
 166 *
 167 * Also note that this returns the location for creating.  Reading
 168 * SHA1 file can happen from any alternate directory listed in the
 169 * DB_ENVIRONMENT environment variable if it is not found in
 170 * the primary object database.
 171 */
 172char *sha1_file_name(const unsigned char *sha1)
 173{
 174        static char *name, *base;
 175
 176        if (!base) {
 177                const char *sha1_file_directory = get_object_directory();
 178                int len = strlen(sha1_file_directory);
 179                base = xmalloc(len + 60);
 180                memcpy(base, sha1_file_directory, len);
 181                memset(base+len, 0, 60);
 182                base[len] = '/';
 183                base[len+3] = '/';
 184                name = base + len + 1;
 185        }
 186        fill_sha1_path(name, sha1);
 187        return base;
 188}
 189
 190static char *sha1_get_pack_name(const unsigned char *sha1,
 191                                char **name, char **base, const char *which)
 192{
 193        static const char hex[] = "0123456789abcdef";
 194        char *buf;
 195        int i;
 196
 197        if (!*base) {
 198                const char *sha1_file_directory = get_object_directory();
 199                int len = strlen(sha1_file_directory);
 200                *base = xmalloc(len + 60);
 201                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 202                        sha1_file_directory, which);
 203                *name = *base + len + 11;
 204        }
 205
 206        buf = *name;
 207
 208        for (i = 0; i < 20; i++) {
 209                unsigned int val = *sha1++;
 210                *buf++ = hex[val >> 4];
 211                *buf++ = hex[val & 0xf];
 212        }
 213
 214        return *base;
 215}
 216
 217char *sha1_pack_name(const unsigned char *sha1)
 218{
 219        static char *name, *base;
 220
 221        return sha1_get_pack_name(sha1, &name, &base, "pack");
 222}
 223
 224char *sha1_pack_index_name(const unsigned char *sha1)
 225{
 226        static char *name, *base;
 227
 228        return sha1_get_pack_name(sha1, &name, &base, "idx");
 229}
 230
 231struct alternate_object_database *alt_odb_list;
 232static struct alternate_object_database **alt_odb_tail;
 233
 234static void read_info_alternates(const char * alternates, int depth);
 235
 236/*
 237 * Prepare alternate object database registry.
 238 *
 239 * The variable alt_odb_list points at the list of struct
 240 * alternate_object_database.  The elements on this list come from
 241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 243 * whose contents is similar to that environment variable but can be
 244 * LF separated.  Its base points at a statically allocated buffer that
 245 * contains "/the/directory/corresponding/to/.git/objects/...", while
 246 * its name points just after the slash at the end of ".git/objects/"
 247 * in the example above, and has enough space to hold 40-byte hex
 248 * SHA1, an extra slash for the first level indirection, and the
 249 * terminating NUL.
 250 */
 251static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 252{
 253        const char *objdir = get_object_directory();
 254        struct alternate_object_database *ent;
 255        struct alternate_object_database *alt;
 256        /* 43 = 40-byte + 2 '/' + terminating NUL */
 257        int pfxlen = len;
 258        int entlen = pfxlen + 43;
 259        int base_len = -1;
 260
 261        if (!is_absolute_path(entry) && relative_base) {
 262                /* Relative alt-odb */
 263                if (base_len < 0)
 264                        base_len = strlen(relative_base) + 1;
 265                entlen += base_len;
 266                pfxlen += base_len;
 267        }
 268        ent = xmalloc(sizeof(*ent) + entlen);
 269
 270        if (!is_absolute_path(entry) && relative_base) {
 271                memcpy(ent->base, relative_base, base_len - 1);
 272                ent->base[base_len - 1] = '/';
 273                memcpy(ent->base + base_len, entry, len);
 274        }
 275        else
 276                memcpy(ent->base, entry, pfxlen);
 277
 278        ent->name = ent->base + pfxlen + 1;
 279        ent->base[pfxlen + 3] = '/';
 280        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 281
 282        /* Detect cases where alternate disappeared */
 283        if (!is_directory(ent->base)) {
 284                error("object directory %s does not exist; "
 285                      "check .git/objects/info/alternates.",
 286                      ent->base);
 287                free(ent);
 288                return -1;
 289        }
 290
 291        /* Prevent the common mistake of listing the same
 292         * thing twice, or object directory itself.
 293         */
 294        for (alt = alt_odb_list; alt; alt = alt->next) {
 295                if (!memcmp(ent->base, alt->base, pfxlen)) {
 296                        free(ent);
 297                        return -1;
 298                }
 299        }
 300        if (!memcmp(ent->base, objdir, pfxlen)) {
 301                free(ent);
 302                return -1;
 303        }
 304
 305        /* add the alternate entry */
 306        *alt_odb_tail = ent;
 307        alt_odb_tail = &(ent->next);
 308        ent->next = NULL;
 309
 310        /* recursively add alternates */
 311        read_info_alternates(ent->base, depth + 1);
 312
 313        ent->base[pfxlen] = '/';
 314
 315        return 0;
 316}
 317
 318static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 319                                 const char *relative_base, int depth)
 320{
 321        const char *cp, *last;
 322
 323        if (depth > 5) {
 324                error("%s: ignoring alternate object stores, nesting too deep.",
 325                                relative_base);
 326                return;
 327        }
 328
 329        last = alt;
 330        while (last < ep) {
 331                cp = last;
 332                if (cp < ep && *cp == '#') {
 333                        while (cp < ep && *cp != sep)
 334                                cp++;
 335                        last = cp + 1;
 336                        continue;
 337                }
 338                while (cp < ep && *cp != sep)
 339                        cp++;
 340                if (last != cp) {
 341                        if (!is_absolute_path(last) && depth) {
 342                                error("%s: ignoring relative alternate object store %s",
 343                                                relative_base, last);
 344                        } else {
 345                                link_alt_odb_entry(last, cp - last,
 346                                                relative_base, depth);
 347                        }
 348                }
 349                while (cp < ep && *cp == sep)
 350                        cp++;
 351                last = cp;
 352        }
 353}
 354
 355static void read_info_alternates(const char * relative_base, int depth)
 356{
 357        char *map;
 358        size_t mapsz;
 359        struct stat st;
 360        const char alt_file_name[] = "info/alternates";
 361        /* Given that relative_base is no longer than PATH_MAX,
 362           ensure that "path" has enough space to append "/", the
 363           file name, "info/alternates", and a trailing NUL.  */
 364        char path[PATH_MAX + 1 + sizeof alt_file_name];
 365        int fd;
 366
 367        sprintf(path, "%s/%s", relative_base, alt_file_name);
 368        fd = open(path, O_RDONLY);
 369        if (fd < 0)
 370                return;
 371        if (fstat(fd, &st) || (st.st_size == 0)) {
 372                close(fd);
 373                return;
 374        }
 375        mapsz = xsize_t(st.st_size);
 376        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 377        close(fd);
 378
 379        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 380
 381        munmap(map, mapsz);
 382}
 383
 384void add_to_alternates_file(const char *reference)
 385{
 386        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 387        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), 1);
 388        char *alt = mkpath("%s/objects\n", reference);
 389        write_or_die(fd, alt, strlen(alt));
 390        if (commit_lock_file(lock))
 391                die("could not close alternates file");
 392        if (alt_odb_tail)
 393                link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
 394}
 395
 396void foreach_alt_odb(alt_odb_fn fn, void *cb)
 397{
 398        struct alternate_object_database *ent;
 399
 400        prepare_alt_odb();
 401        for (ent = alt_odb_list; ent; ent = ent->next)
 402                if (fn(ent, cb))
 403                        return;
 404}
 405
 406void prepare_alt_odb(void)
 407{
 408        const char *alt;
 409
 410        if (alt_odb_tail)
 411                return;
 412
 413        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 414        if (!alt) alt = "";
 415
 416        alt_odb_tail = &alt_odb_list;
 417        link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
 418
 419        read_info_alternates(get_object_directory(), 0);
 420}
 421
 422static int has_loose_object(const unsigned char *sha1)
 423{
 424        char *name = sha1_file_name(sha1);
 425        struct alternate_object_database *alt;
 426
 427        if (!access(name, F_OK))
 428                return 1;
 429        prepare_alt_odb();
 430        for (alt = alt_odb_list; alt; alt = alt->next) {
 431                name = alt->name;
 432                fill_sha1_path(name, sha1);
 433                if (!access(alt->base, F_OK))
 434                        return 1;
 435        }
 436        return 0;
 437}
 438
 439static unsigned int pack_used_ctr;
 440static unsigned int pack_mmap_calls;
 441static unsigned int peak_pack_open_windows;
 442static unsigned int pack_open_windows;
 443static size_t peak_pack_mapped;
 444static size_t pack_mapped;
 445struct packed_git *packed_git;
 446
 447void pack_report(void)
 448{
 449        fprintf(stderr,
 450                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 451                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 452                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 453                sz_fmt(getpagesize()),
 454                sz_fmt(packed_git_window_size),
 455                sz_fmt(packed_git_limit));
 456        fprintf(stderr,
 457                "pack_report: pack_used_ctr            = %10u\n"
 458                "pack_report: pack_mmap_calls          = %10u\n"
 459                "pack_report: pack_open_windows        = %10u / %10u\n"
 460                "pack_report: pack_mapped              = "
 461                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 462                pack_used_ctr,
 463                pack_mmap_calls,
 464                pack_open_windows, peak_pack_open_windows,
 465                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 466}
 467
 468static int check_packed_git_idx(const char *path,  struct packed_git *p)
 469{
 470        void *idx_map;
 471        struct pack_idx_header *hdr;
 472        size_t idx_size;
 473        uint32_t version, nr, i, *index;
 474        int fd = open(path, O_RDONLY);
 475        struct stat st;
 476
 477        if (fd < 0)
 478                return -1;
 479        if (fstat(fd, &st)) {
 480                close(fd);
 481                return -1;
 482        }
 483        idx_size = xsize_t(st.st_size);
 484        if (idx_size < 4 * 256 + 20 + 20) {
 485                close(fd);
 486                return error("index file %s is too small", path);
 487        }
 488        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 489        close(fd);
 490
 491        hdr = idx_map;
 492        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 493                version = ntohl(hdr->idx_version);
 494                if (version < 2 || version > 2) {
 495                        munmap(idx_map, idx_size);
 496                        return error("index file %s is version %"PRIu32
 497                                     " and is not supported by this binary"
 498                                     " (try upgrading GIT to a newer version)",
 499                                     path, version);
 500                }
 501        } else
 502                version = 1;
 503
 504        nr = 0;
 505        index = idx_map;
 506        if (version > 1)
 507                index += 2;  /* skip index header */
 508        for (i = 0; i < 256; i++) {
 509                uint32_t n = ntohl(index[i]);
 510                if (n < nr) {
 511                        munmap(idx_map, idx_size);
 512                        return error("non-monotonic index %s", path);
 513                }
 514                nr = n;
 515        }
 516
 517        if (version == 1) {
 518                /*
 519                 * Total size:
 520                 *  - 256 index entries 4 bytes each
 521                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 522                 *  - 20-byte SHA1 of the packfile
 523                 *  - 20-byte SHA1 file checksum
 524                 */
 525                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 526                        munmap(idx_map, idx_size);
 527                        return error("wrong index v1 file size in %s", path);
 528                }
 529        } else if (version == 2) {
 530                /*
 531                 * Minimum size:
 532                 *  - 8 bytes of header
 533                 *  - 256 index entries 4 bytes each
 534                 *  - 20-byte sha1 entry * nr
 535                 *  - 4-byte crc entry * nr
 536                 *  - 4-byte offset entry * nr
 537                 *  - 20-byte SHA1 of the packfile
 538                 *  - 20-byte SHA1 file checksum
 539                 * And after the 4-byte offset table might be a
 540                 * variable sized table containing 8-byte entries
 541                 * for offsets larger than 2^31.
 542                 */
 543                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 544                unsigned long max_size = min_size;
 545                if (nr)
 546                        max_size += (nr - 1)*8;
 547                if (idx_size < min_size || idx_size > max_size) {
 548                        munmap(idx_map, idx_size);
 549                        return error("wrong index v2 file size in %s", path);
 550                }
 551                if (idx_size != min_size &&
 552                    /*
 553                     * make sure we can deal with large pack offsets.
 554                     * 31-bit signed offset won't be enough, neither
 555                     * 32-bit unsigned one will be.
 556                     */
 557                    (sizeof(off_t) <= 4)) {
 558                        munmap(idx_map, idx_size);
 559                        return error("pack too large for current definition of off_t in %s", path);
 560                }
 561        }
 562
 563        p->index_version = version;
 564        p->index_data = idx_map;
 565        p->index_size = idx_size;
 566        p->num_objects = nr;
 567        return 0;
 568}
 569
 570int open_pack_index(struct packed_git *p)
 571{
 572        char *idx_name;
 573        int ret;
 574
 575        if (p->index_data)
 576                return 0;
 577
 578        idx_name = xstrdup(p->pack_name);
 579        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 580        ret = check_packed_git_idx(idx_name, p);
 581        free(idx_name);
 582        return ret;
 583}
 584
 585static void scan_windows(struct packed_git *p,
 586        struct packed_git **lru_p,
 587        struct pack_window **lru_w,
 588        struct pack_window **lru_l)
 589{
 590        struct pack_window *w, *w_l;
 591
 592        for (w_l = NULL, w = p->windows; w; w = w->next) {
 593                if (!w->inuse_cnt) {
 594                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 595                                *lru_p = p;
 596                                *lru_w = w;
 597                                *lru_l = w_l;
 598                        }
 599                }
 600                w_l = w;
 601        }
 602}
 603
 604static int unuse_one_window(struct packed_git *current, int keep_fd)
 605{
 606        struct packed_git *p, *lru_p = NULL;
 607        struct pack_window *lru_w = NULL, *lru_l = NULL;
 608
 609        if (current)
 610                scan_windows(current, &lru_p, &lru_w, &lru_l);
 611        for (p = packed_git; p; p = p->next)
 612                scan_windows(p, &lru_p, &lru_w, &lru_l);
 613        if (lru_p) {
 614                munmap(lru_w->base, lru_w->len);
 615                pack_mapped -= lru_w->len;
 616                if (lru_l)
 617                        lru_l->next = lru_w->next;
 618                else {
 619                        lru_p->windows = lru_w->next;
 620                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 621                                close(lru_p->pack_fd);
 622                                lru_p->pack_fd = -1;
 623                        }
 624                }
 625                free(lru_w);
 626                pack_open_windows--;
 627                return 1;
 628        }
 629        return 0;
 630}
 631
 632void release_pack_memory(size_t need, int fd)
 633{
 634        size_t cur = pack_mapped;
 635        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 636                ; /* nothing */
 637}
 638
 639void close_pack_windows(struct packed_git *p)
 640{
 641        while (p->windows) {
 642                struct pack_window *w = p->windows;
 643
 644                if (w->inuse_cnt)
 645                        die("pack '%s' still has open windows to it",
 646                            p->pack_name);
 647                munmap(w->base, w->len);
 648                pack_mapped -= w->len;
 649                pack_open_windows--;
 650                p->windows = w->next;
 651                free(w);
 652        }
 653}
 654
 655void unuse_pack(struct pack_window **w_cursor)
 656{
 657        struct pack_window *w = *w_cursor;
 658        if (w) {
 659                w->inuse_cnt--;
 660                *w_cursor = NULL;
 661        }
 662}
 663
 664/*
 665 * Do not call this directly as this leaks p->pack_fd on error return;
 666 * call open_packed_git() instead.
 667 */
 668static int open_packed_git_1(struct packed_git *p)
 669{
 670        struct stat st;
 671        struct pack_header hdr;
 672        unsigned char sha1[20];
 673        unsigned char *idx_sha1;
 674        long fd_flag;
 675
 676        if (!p->index_data && open_pack_index(p))
 677                return error("packfile %s index unavailable", p->pack_name);
 678
 679        p->pack_fd = open(p->pack_name, O_RDONLY);
 680        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 681                return -1;
 682
 683        /* If we created the struct before we had the pack we lack size. */
 684        if (!p->pack_size) {
 685                if (!S_ISREG(st.st_mode))
 686                        return error("packfile %s not a regular file", p->pack_name);
 687                p->pack_size = st.st_size;
 688        } else if (p->pack_size != st.st_size)
 689                return error("packfile %s size changed", p->pack_name);
 690
 691        /* We leave these file descriptors open with sliding mmap;
 692         * there is no point keeping them open across exec(), though.
 693         */
 694        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 695        if (fd_flag < 0)
 696                return error("cannot determine file descriptor flags");
 697        fd_flag |= FD_CLOEXEC;
 698        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 699                return error("cannot set FD_CLOEXEC");
 700
 701        /* Verify we recognize this pack file format. */
 702        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 703                return error("file %s is far too short to be a packfile", p->pack_name);
 704        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 705                return error("file %s is not a GIT packfile", p->pack_name);
 706        if (!pack_version_ok(hdr.hdr_version))
 707                return error("packfile %s is version %"PRIu32" and not"
 708                        " supported (try upgrading GIT to a newer version)",
 709                        p->pack_name, ntohl(hdr.hdr_version));
 710
 711        /* Verify the pack matches its index. */
 712        if (p->num_objects != ntohl(hdr.hdr_entries))
 713                return error("packfile %s claims to have %"PRIu32" objects"
 714                             " while index indicates %"PRIu32" objects",
 715                             p->pack_name, ntohl(hdr.hdr_entries),
 716                             p->num_objects);
 717        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 718                return error("end of packfile %s is unavailable", p->pack_name);
 719        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 720                return error("packfile %s signature is unavailable", p->pack_name);
 721        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 722        if (hashcmp(sha1, idx_sha1))
 723                return error("packfile %s does not match index", p->pack_name);
 724        return 0;
 725}
 726
 727static int open_packed_git(struct packed_git *p)
 728{
 729        if (!open_packed_git_1(p))
 730                return 0;
 731        if (p->pack_fd != -1) {
 732                close(p->pack_fd);
 733                p->pack_fd = -1;
 734        }
 735        return -1;
 736}
 737
 738static int in_window(struct pack_window *win, off_t offset)
 739{
 740        /* We must promise at least 20 bytes (one hash) after the
 741         * offset is available from this window, otherwise the offset
 742         * is not actually in this window and a different window (which
 743         * has that one hash excess) must be used.  This is to support
 744         * the object header and delta base parsing routines below.
 745         */
 746        off_t win_off = win->offset;
 747        return win_off <= offset
 748                && (offset + 20) <= (win_off + win->len);
 749}
 750
 751unsigned char* use_pack(struct packed_git *p,
 752                struct pack_window **w_cursor,
 753                off_t offset,
 754                unsigned int *left)
 755{
 756        struct pack_window *win = *w_cursor;
 757
 758        if (p->pack_fd == -1 && open_packed_git(p))
 759                die("packfile %s cannot be accessed", p->pack_name);
 760
 761        /* Since packfiles end in a hash of their content and its
 762         * pointless to ask for an offset into the middle of that
 763         * hash, and the in_window function above wouldn't match
 764         * don't allow an offset too close to the end of the file.
 765         */
 766        if (offset > (p->pack_size - 20))
 767                die("offset beyond end of packfile (truncated pack?)");
 768
 769        if (!win || !in_window(win, offset)) {
 770                if (win)
 771                        win->inuse_cnt--;
 772                for (win = p->windows; win; win = win->next) {
 773                        if (in_window(win, offset))
 774                                break;
 775                }
 776                if (!win) {
 777                        size_t window_align = packed_git_window_size / 2;
 778                        off_t len;
 779                        win = xcalloc(1, sizeof(*win));
 780                        win->offset = (offset / window_align) * window_align;
 781                        len = p->pack_size - win->offset;
 782                        if (len > packed_git_window_size)
 783                                len = packed_git_window_size;
 784                        win->len = (size_t)len;
 785                        pack_mapped += win->len;
 786                        while (packed_git_limit < pack_mapped
 787                                && unuse_one_window(p, p->pack_fd))
 788                                ; /* nothing */
 789                        win->base = xmmap(NULL, win->len,
 790                                PROT_READ, MAP_PRIVATE,
 791                                p->pack_fd, win->offset);
 792                        if (win->base == MAP_FAILED)
 793                                die("packfile %s cannot be mapped: %s",
 794                                        p->pack_name,
 795                                        strerror(errno));
 796                        pack_mmap_calls++;
 797                        pack_open_windows++;
 798                        if (pack_mapped > peak_pack_mapped)
 799                                peak_pack_mapped = pack_mapped;
 800                        if (pack_open_windows > peak_pack_open_windows)
 801                                peak_pack_open_windows = pack_open_windows;
 802                        win->next = p->windows;
 803                        p->windows = win;
 804                }
 805        }
 806        if (win != *w_cursor) {
 807                win->last_used = pack_used_ctr++;
 808                win->inuse_cnt++;
 809                *w_cursor = win;
 810        }
 811        offset -= win->offset;
 812        if (left)
 813                *left = win->len - xsize_t(offset);
 814        return win->base + offset;
 815}
 816
 817static struct packed_git *alloc_packed_git(int extra)
 818{
 819        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 820        memset(p, 0, sizeof(*p));
 821        p->pack_fd = -1;
 822        return p;
 823}
 824
 825struct packed_git *add_packed_git(const char *path, int path_len, int local)
 826{
 827        struct stat st;
 828        struct packed_git *p = alloc_packed_git(path_len + 2);
 829
 830        /*
 831         * Make sure a corresponding .pack file exists and that
 832         * the index looks sane.
 833         */
 834        path_len -= strlen(".idx");
 835        if (path_len < 1) {
 836                free(p);
 837                return NULL;
 838        }
 839        memcpy(p->pack_name, path, path_len);
 840        strcpy(p->pack_name + path_len, ".pack");
 841        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 842                free(p);
 843                return NULL;
 844        }
 845
 846        /* ok, it looks sane as far as we can check without
 847         * actually mapping the pack file.
 848         */
 849        p->pack_size = st.st_size;
 850        p->pack_local = local;
 851        p->mtime = st.st_mtime;
 852        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 853                hashclr(p->sha1);
 854        return p;
 855}
 856
 857struct packed_git *parse_pack_index(unsigned char *sha1)
 858{
 859        const char *idx_path = sha1_pack_index_name(sha1);
 860        const char *path = sha1_pack_name(sha1);
 861        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 862
 863        strcpy(p->pack_name, path);
 864        hashcpy(p->sha1, sha1);
 865        if (check_packed_git_idx(idx_path, p)) {
 866                free(p);
 867                return NULL;
 868        }
 869
 870        return p;
 871}
 872
 873void install_packed_git(struct packed_git *pack)
 874{
 875        pack->next = packed_git;
 876        packed_git = pack;
 877}
 878
 879static void prepare_packed_git_one(char *objdir, int local)
 880{
 881        /* Ensure that this buffer is large enough so that we can
 882           append "/pack/" without clobbering the stack even if
 883           strlen(objdir) were PATH_MAX.  */
 884        char path[PATH_MAX + 1 + 4 + 1 + 1];
 885        int len;
 886        DIR *dir;
 887        struct dirent *de;
 888
 889        sprintf(path, "%s/pack", objdir);
 890        len = strlen(path);
 891        dir = opendir(path);
 892        if (!dir) {
 893                if (errno != ENOENT)
 894                        error("unable to open object pack directory: %s: %s",
 895                              path, strerror(errno));
 896                return;
 897        }
 898        path[len++] = '/';
 899        while ((de = readdir(dir)) != NULL) {
 900                int namelen = strlen(de->d_name);
 901                struct packed_git *p;
 902
 903                if (!has_extension(de->d_name, ".idx"))
 904                        continue;
 905
 906                if (len + namelen + 1 > sizeof(path))
 907                        continue;
 908
 909                /* Don't reopen a pack we already have. */
 910                strcpy(path + len, de->d_name);
 911                for (p = packed_git; p; p = p->next) {
 912                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 913                                break;
 914                }
 915                if (p)
 916                        continue;
 917                /* See if it really is a valid .idx file with corresponding
 918                 * .pack file that we can map.
 919                 */
 920                p = add_packed_git(path, len + namelen, local);
 921                if (!p)
 922                        continue;
 923                install_packed_git(p);
 924        }
 925        closedir(dir);
 926}
 927
 928static int sort_pack(const void *a_, const void *b_)
 929{
 930        struct packed_git *a = *((struct packed_git **)a_);
 931        struct packed_git *b = *((struct packed_git **)b_);
 932        int st;
 933
 934        /*
 935         * Local packs tend to contain objects specific to our
 936         * variant of the project than remote ones.  In addition,
 937         * remote ones could be on a network mounted filesystem.
 938         * Favor local ones for these reasons.
 939         */
 940        st = a->pack_local - b->pack_local;
 941        if (st)
 942                return -st;
 943
 944        /*
 945         * Younger packs tend to contain more recent objects,
 946         * and more recent objects tend to get accessed more
 947         * often.
 948         */
 949        if (a->mtime < b->mtime)
 950                return 1;
 951        else if (a->mtime == b->mtime)
 952                return 0;
 953        return -1;
 954}
 955
 956static void rearrange_packed_git(void)
 957{
 958        struct packed_git **ary, *p;
 959        int i, n;
 960
 961        for (n = 0, p = packed_git; p; p = p->next)
 962                n++;
 963        if (n < 2)
 964                return;
 965
 966        /* prepare an array of packed_git for easier sorting */
 967        ary = xcalloc(n, sizeof(struct packed_git *));
 968        for (n = 0, p = packed_git; p; p = p->next)
 969                ary[n++] = p;
 970
 971        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 972
 973        /* link them back again */
 974        for (i = 0; i < n - 1; i++)
 975                ary[i]->next = ary[i + 1];
 976        ary[n - 1]->next = NULL;
 977        packed_git = ary[0];
 978
 979        free(ary);
 980}
 981
 982static int prepare_packed_git_run_once = 0;
 983void prepare_packed_git(void)
 984{
 985        struct alternate_object_database *alt;
 986
 987        if (prepare_packed_git_run_once)
 988                return;
 989        prepare_packed_git_one(get_object_directory(), 1);
 990        prepare_alt_odb();
 991        for (alt = alt_odb_list; alt; alt = alt->next) {
 992                alt->name[-1] = 0;
 993                prepare_packed_git_one(alt->base, 0);
 994                alt->name[-1] = '/';
 995        }
 996        rearrange_packed_git();
 997        prepare_packed_git_run_once = 1;
 998}
 999
1000void reprepare_packed_git(void)
1001{
1002        discard_revindex();
1003        prepare_packed_git_run_once = 0;
1004        prepare_packed_git();
1005}
1006
1007static void mark_bad_packed_object(struct packed_git *p,
1008                                   const unsigned char *sha1)
1009{
1010        unsigned i;
1011        for (i = 0; i < p->num_bad_objects; i++)
1012                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1013                        return;
1014        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1015        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1016        p->num_bad_objects++;
1017}
1018
1019static int has_packed_and_bad(const unsigned char *sha1)
1020{
1021        struct packed_git *p;
1022        unsigned i;
1023
1024        for (p = packed_git; p; p = p->next)
1025                for (i = 0; i < p->num_bad_objects; i++)
1026                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1027                                return 1;
1028        return 0;
1029}
1030
1031int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1032{
1033        unsigned char real_sha1[20];
1034        hash_sha1_file(map, size, type, real_sha1);
1035        return hashcmp(sha1, real_sha1) ? -1 : 0;
1036}
1037
1038static int git_open_noatime(const char *name)
1039{
1040        static int sha1_file_open_flag = O_NOATIME;
1041        int fd = open(name, O_RDONLY | sha1_file_open_flag);
1042
1043        /* Might the failure be due to O_NOATIME? */
1044        if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1045                fd = open(name, O_RDONLY);
1046                if (fd >= 0)
1047                        sha1_file_open_flag = 0;
1048        }
1049        return fd;
1050}
1051
1052static int open_sha1_file(const unsigned char *sha1)
1053{
1054        int fd;
1055        char *name = sha1_file_name(sha1);
1056        struct alternate_object_database *alt;
1057
1058        fd = git_open_noatime(name);
1059        if (fd >= 0)
1060                return fd;
1061
1062        prepare_alt_odb();
1063        errno = ENOENT;
1064        for (alt = alt_odb_list; alt; alt = alt->next) {
1065                name = alt->name;
1066                fill_sha1_path(name, sha1);
1067                fd = git_open_noatime(alt->base);
1068                if (fd >= 0)
1069                        return fd;
1070        }
1071        return -1;
1072}
1073
1074static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1075{
1076        void *map;
1077        int fd;
1078
1079        fd = open_sha1_file(sha1);
1080        map = NULL;
1081        if (fd >= 0) {
1082                struct stat st;
1083
1084                if (!fstat(fd, &st)) {
1085                        *size = xsize_t(st.st_size);
1086                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1087                }
1088                close(fd);
1089        }
1090        return map;
1091}
1092
1093static int legacy_loose_object(unsigned char *map)
1094{
1095        unsigned int word;
1096
1097        /*
1098         * Is it a zlib-compressed buffer? If so, the first byte
1099         * must be 0x78 (15-bit window size, deflated), and the
1100         * first 16-bit word is evenly divisible by 31
1101         */
1102        word = (map[0] << 8) + map[1];
1103        if (map[0] == 0x78 && !(word % 31))
1104                return 1;
1105        else
1106                return 0;
1107}
1108
1109unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
1110{
1111        unsigned shift;
1112        unsigned char c;
1113        unsigned long size;
1114        unsigned long used = 0;
1115
1116        c = buf[used++];
1117        *type = (c >> 4) & 7;
1118        size = c & 15;
1119        shift = 4;
1120        while (c & 0x80) {
1121                if (len <= used)
1122                        return 0;
1123                if (sizeof(long) * 8 <= shift)
1124                        return 0;
1125                c = buf[used++];
1126                size += (c & 0x7f) << shift;
1127                shift += 7;
1128        }
1129        *sizep = size;
1130        return used;
1131}
1132
1133static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1134{
1135        unsigned long size, used;
1136        static const char valid_loose_object_type[8] = {
1137                0, /* OBJ_EXT */
1138                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1139                0, /* "delta" and others are invalid in a loose object */
1140        };
1141        enum object_type type;
1142
1143        /* Get the data stream */
1144        memset(stream, 0, sizeof(*stream));
1145        stream->next_in = map;
1146        stream->avail_in = mapsize;
1147        stream->next_out = buffer;
1148        stream->avail_out = bufsiz;
1149
1150        if (legacy_loose_object(map)) {
1151                inflateInit(stream);
1152                return inflate(stream, 0);
1153        }
1154
1155
1156        /*
1157         * There used to be a second loose object header format which
1158         * was meant to mimic the in-pack format, allowing for direct
1159         * copy of the object data.  This format turned up not to be
1160         * really worth it and we don't write it any longer.  But we
1161         * can still read it.
1162         */
1163        used = unpack_object_header_gently(map, mapsize, &type, &size);
1164        if (!used || !valid_loose_object_type[type])
1165                return -1;
1166        map += used;
1167        mapsize -= used;
1168
1169        /* Set up the stream for the rest.. */
1170        stream->next_in = map;
1171        stream->avail_in = mapsize;
1172        inflateInit(stream);
1173
1174        /* And generate the fake traditional header */
1175        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1176                                         typename(type), size);
1177        return 0;
1178}
1179
1180static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1181{
1182        int bytes = strlen(buffer) + 1;
1183        unsigned char *buf = xmalloc(1+size);
1184        unsigned long n;
1185        int status = Z_OK;
1186
1187        n = stream->total_out - bytes;
1188        if (n > size)
1189                n = size;
1190        memcpy(buf, (char *) buffer + bytes, n);
1191        bytes = n;
1192        if (bytes <= size) {
1193                /*
1194                 * The above condition must be (bytes <= size), not
1195                 * (bytes < size).  In other words, even though we
1196                 * expect no more output and set avail_out to zer0,
1197                 * the input zlib stream may have bytes that express
1198                 * "this concludes the stream", and we *do* want to
1199                 * eat that input.
1200                 *
1201                 * Otherwise we would not be able to test that we
1202                 * consumed all the input to reach the expected size;
1203                 * we also want to check that zlib tells us that all
1204                 * went well with status == Z_STREAM_END at the end.
1205                 */
1206                stream->next_out = buf + bytes;
1207                stream->avail_out = size - bytes;
1208                while (status == Z_OK)
1209                        status = inflate(stream, Z_FINISH);
1210        }
1211        buf[size] = 0;
1212        if (status == Z_STREAM_END && !stream->avail_in) {
1213                inflateEnd(stream);
1214                return buf;
1215        }
1216
1217        if (status < 0)
1218                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1219        else if (stream->avail_in)
1220                error("garbage at end of loose object '%s'",
1221                      sha1_to_hex(sha1));
1222        free(buf);
1223        return NULL;
1224}
1225
1226/*
1227 * We used to just use "sscanf()", but that's actually way
1228 * too permissive for what we want to check. So do an anal
1229 * object header parse by hand.
1230 */
1231static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1232{
1233        char type[10];
1234        int i;
1235        unsigned long size;
1236
1237        /*
1238         * The type can be at most ten bytes (including the
1239         * terminating '\0' that we add), and is followed by
1240         * a space.
1241         */
1242        i = 0;
1243        for (;;) {
1244                char c = *hdr++;
1245                if (c == ' ')
1246                        break;
1247                type[i++] = c;
1248                if (i >= sizeof(type))
1249                        return -1;
1250        }
1251        type[i] = 0;
1252
1253        /*
1254         * The length must follow immediately, and be in canonical
1255         * decimal format (ie "010" is not valid).
1256         */
1257        size = *hdr++ - '0';
1258        if (size > 9)
1259                return -1;
1260        if (size) {
1261                for (;;) {
1262                        unsigned long c = *hdr - '0';
1263                        if (c > 9)
1264                                break;
1265                        hdr++;
1266                        size = size * 10 + c;
1267                }
1268        }
1269        *sizep = size;
1270
1271        /*
1272         * The length must be followed by a zero byte
1273         */
1274        return *hdr ? -1 : type_from_string(type);
1275}
1276
1277static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1278{
1279        int ret;
1280        z_stream stream;
1281        char hdr[8192];
1282
1283        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1284        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1285                return NULL;
1286
1287        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1288}
1289
1290unsigned long get_size_from_delta(struct packed_git *p,
1291                                  struct pack_window **w_curs,
1292                                  off_t curpos)
1293{
1294        const unsigned char *data;
1295        unsigned char delta_head[20], *in;
1296        z_stream stream;
1297        int st;
1298
1299        memset(&stream, 0, sizeof(stream));
1300        stream.next_out = delta_head;
1301        stream.avail_out = sizeof(delta_head);
1302
1303        inflateInit(&stream);
1304        do {
1305                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1306                stream.next_in = in;
1307                st = inflate(&stream, Z_FINISH);
1308                curpos += stream.next_in - in;
1309        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1310                 stream.total_out < sizeof(delta_head));
1311        inflateEnd(&stream);
1312        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
1313                die("delta data unpack-initial failed");
1314
1315        /* Examine the initial part of the delta to figure out
1316         * the result size.
1317         */
1318        data = delta_head;
1319
1320        /* ignore base size */
1321        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1322
1323        /* Read the result size */
1324        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1325}
1326
1327static off_t get_delta_base(struct packed_git *p,
1328                                    struct pack_window **w_curs,
1329                                    off_t *curpos,
1330                                    enum object_type type,
1331                                    off_t delta_obj_offset)
1332{
1333        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1334        off_t base_offset;
1335
1336        /* use_pack() assured us we have [base_info, base_info + 20)
1337         * as a range that we can look at without walking off the
1338         * end of the mapped window.  Its actually the hash size
1339         * that is assured.  An OFS_DELTA longer than the hash size
1340         * is stupid, as then a REF_DELTA would be smaller to store.
1341         */
1342        if (type == OBJ_OFS_DELTA) {
1343                unsigned used = 0;
1344                unsigned char c = base_info[used++];
1345                base_offset = c & 127;
1346                while (c & 128) {
1347                        base_offset += 1;
1348                        if (!base_offset || MSB(base_offset, 7))
1349                                return 0;  /* overflow */
1350                        c = base_info[used++];
1351                        base_offset = (base_offset << 7) + (c & 127);
1352                }
1353                base_offset = delta_obj_offset - base_offset;
1354                if (base_offset >= delta_obj_offset)
1355                        return 0;  /* out of bound */
1356                *curpos += used;
1357        } else if (type == OBJ_REF_DELTA) {
1358                /* The base entry _must_ be in the same pack */
1359                base_offset = find_pack_entry_one(base_info, p);
1360                *curpos += 20;
1361        } else
1362                die("I am totally screwed");
1363        return base_offset;
1364}
1365
1366/* forward declaration for a mutually recursive function */
1367static int packed_object_info(struct packed_git *p, off_t offset,
1368                              unsigned long *sizep);
1369
1370static int packed_delta_info(struct packed_git *p,
1371                             struct pack_window **w_curs,
1372                             off_t curpos,
1373                             enum object_type type,
1374                             off_t obj_offset,
1375                             unsigned long *sizep)
1376{
1377        off_t base_offset;
1378
1379        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1380        type = packed_object_info(p, base_offset, NULL);
1381
1382        /* We choose to only get the type of the base object and
1383         * ignore potentially corrupt pack file that expects the delta
1384         * based on a base with a wrong size.  This saves tons of
1385         * inflate() calls.
1386         */
1387        if (sizep)
1388                *sizep = get_size_from_delta(p, w_curs, curpos);
1389
1390        return type;
1391}
1392
1393static int unpack_object_header(struct packed_git *p,
1394                                struct pack_window **w_curs,
1395                                off_t *curpos,
1396                                unsigned long *sizep)
1397{
1398        unsigned char *base;
1399        unsigned int left;
1400        unsigned long used;
1401        enum object_type type;
1402
1403        /* use_pack() assures us we have [base, base + 20) available
1404         * as a range that we can look at at.  (Its actually the hash
1405         * size that is assured.)  With our object header encoding
1406         * the maximum deflated object size is 2^137, which is just
1407         * insane, so we know won't exceed what we have been given.
1408         */
1409        base = use_pack(p, w_curs, *curpos, &left);
1410        used = unpack_object_header_gently(base, left, &type, sizep);
1411        if (!used)
1412                die("object offset outside of pack file");
1413        *curpos += used;
1414
1415        return type;
1416}
1417
1418const char *packed_object_info_detail(struct packed_git *p,
1419                                      off_t obj_offset,
1420                                      unsigned long *size,
1421                                      unsigned long *store_size,
1422                                      unsigned int *delta_chain_length,
1423                                      unsigned char *base_sha1)
1424{
1425        struct pack_window *w_curs = NULL;
1426        off_t curpos;
1427        unsigned long dummy;
1428        unsigned char *next_sha1;
1429        enum object_type type;
1430        struct revindex_entry *revidx;
1431
1432        *delta_chain_length = 0;
1433        curpos = obj_offset;
1434        type = unpack_object_header(p, &w_curs, &curpos, size);
1435
1436        revidx = find_pack_revindex(p, obj_offset);
1437        *store_size = revidx[1].offset - obj_offset;
1438
1439        for (;;) {
1440                switch (type) {
1441                default:
1442                        die("pack %s contains unknown object type %d",
1443                            p->pack_name, type);
1444                case OBJ_COMMIT:
1445                case OBJ_TREE:
1446                case OBJ_BLOB:
1447                case OBJ_TAG:
1448                        unuse_pack(&w_curs);
1449                        return typename(type);
1450                case OBJ_OFS_DELTA:
1451                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1452                        if (!obj_offset)
1453                                die("pack %s contains bad delta base reference of type %s",
1454                                    p->pack_name, typename(type));
1455                        if (*delta_chain_length == 0) {
1456                                revidx = find_pack_revindex(p, obj_offset);
1457                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1458                        }
1459                        break;
1460                case OBJ_REF_DELTA:
1461                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1462                        if (*delta_chain_length == 0)
1463                                hashcpy(base_sha1, next_sha1);
1464                        obj_offset = find_pack_entry_one(next_sha1, p);
1465                        break;
1466                }
1467                (*delta_chain_length)++;
1468                curpos = obj_offset;
1469                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1470        }
1471}
1472
1473static int packed_object_info(struct packed_git *p, off_t obj_offset,
1474                              unsigned long *sizep)
1475{
1476        struct pack_window *w_curs = NULL;
1477        unsigned long size;
1478        off_t curpos = obj_offset;
1479        enum object_type type;
1480
1481        type = unpack_object_header(p, &w_curs, &curpos, &size);
1482
1483        switch (type) {
1484        case OBJ_OFS_DELTA:
1485        case OBJ_REF_DELTA:
1486                type = packed_delta_info(p, &w_curs, curpos,
1487                                         type, obj_offset, sizep);
1488                break;
1489        case OBJ_COMMIT:
1490        case OBJ_TREE:
1491        case OBJ_BLOB:
1492        case OBJ_TAG:
1493                if (sizep)
1494                        *sizep = size;
1495                break;
1496        default:
1497                die("pack %s contains unknown object type %d",
1498                    p->pack_name, type);
1499        }
1500        unuse_pack(&w_curs);
1501        return type;
1502}
1503
1504static void *unpack_compressed_entry(struct packed_git *p,
1505                                    struct pack_window **w_curs,
1506                                    off_t curpos,
1507                                    unsigned long size)
1508{
1509        int st;
1510        z_stream stream;
1511        unsigned char *buffer, *in;
1512
1513        buffer = xmalloc(size + 1);
1514        buffer[size] = 0;
1515        memset(&stream, 0, sizeof(stream));
1516        stream.next_out = buffer;
1517        stream.avail_out = size;
1518
1519        inflateInit(&stream);
1520        do {
1521                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1522                stream.next_in = in;
1523                st = inflate(&stream, Z_FINISH);
1524                curpos += stream.next_in - in;
1525        } while (st == Z_OK || st == Z_BUF_ERROR);
1526        inflateEnd(&stream);
1527        if ((st != Z_STREAM_END) || stream.total_out != size) {
1528                free(buffer);
1529                return NULL;
1530        }
1531
1532        return buffer;
1533}
1534
1535#define MAX_DELTA_CACHE (256)
1536
1537static size_t delta_base_cached;
1538
1539static struct delta_base_cache_lru_list {
1540        struct delta_base_cache_lru_list *prev;
1541        struct delta_base_cache_lru_list *next;
1542} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1543
1544static struct delta_base_cache_entry {
1545        struct delta_base_cache_lru_list lru;
1546        void *data;
1547        struct packed_git *p;
1548        off_t base_offset;
1549        unsigned long size;
1550        enum object_type type;
1551} delta_base_cache[MAX_DELTA_CACHE];
1552
1553static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1554{
1555        unsigned long hash;
1556
1557        hash = (unsigned long)p + (unsigned long)base_offset;
1558        hash += (hash >> 8) + (hash >> 16);
1559        return hash % MAX_DELTA_CACHE;
1560}
1561
1562static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1563        unsigned long *base_size, enum object_type *type, int keep_cache)
1564{
1565        void *ret;
1566        unsigned long hash = pack_entry_hash(p, base_offset);
1567        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1568
1569        ret = ent->data;
1570        if (ret && ent->p == p && ent->base_offset == base_offset)
1571                goto found_cache_entry;
1572        return unpack_entry(p, base_offset, type, base_size);
1573
1574found_cache_entry:
1575        if (!keep_cache) {
1576                ent->data = NULL;
1577                ent->lru.next->prev = ent->lru.prev;
1578                ent->lru.prev->next = ent->lru.next;
1579                delta_base_cached -= ent->size;
1580        } else {
1581                ret = xmemdupz(ent->data, ent->size);
1582        }
1583        *type = ent->type;
1584        *base_size = ent->size;
1585        return ret;
1586}
1587
1588static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1589{
1590        if (ent->data) {
1591                free(ent->data);
1592                ent->data = NULL;
1593                ent->lru.next->prev = ent->lru.prev;
1594                ent->lru.prev->next = ent->lru.next;
1595                delta_base_cached -= ent->size;
1596        }
1597}
1598
1599static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1600        void *base, unsigned long base_size, enum object_type type)
1601{
1602        unsigned long hash = pack_entry_hash(p, base_offset);
1603        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1604        struct delta_base_cache_lru_list *lru;
1605
1606        release_delta_base_cache(ent);
1607        delta_base_cached += base_size;
1608
1609        for (lru = delta_base_cache_lru.next;
1610             delta_base_cached > delta_base_cache_limit
1611             && lru != &delta_base_cache_lru;
1612             lru = lru->next) {
1613                struct delta_base_cache_entry *f = (void *)lru;
1614                if (f->type == OBJ_BLOB)
1615                        release_delta_base_cache(f);
1616        }
1617        for (lru = delta_base_cache_lru.next;
1618             delta_base_cached > delta_base_cache_limit
1619             && lru != &delta_base_cache_lru;
1620             lru = lru->next) {
1621                struct delta_base_cache_entry *f = (void *)lru;
1622                release_delta_base_cache(f);
1623        }
1624
1625        ent->p = p;
1626        ent->base_offset = base_offset;
1627        ent->type = type;
1628        ent->data = base;
1629        ent->size = base_size;
1630        ent->lru.next = &delta_base_cache_lru;
1631        ent->lru.prev = delta_base_cache_lru.prev;
1632        delta_base_cache_lru.prev->next = &ent->lru;
1633        delta_base_cache_lru.prev = &ent->lru;
1634}
1635
1636static void *unpack_delta_entry(struct packed_git *p,
1637                                struct pack_window **w_curs,
1638                                off_t curpos,
1639                                unsigned long delta_size,
1640                                off_t obj_offset,
1641                                enum object_type *type,
1642                                unsigned long *sizep)
1643{
1644        void *delta_data, *result, *base;
1645        unsigned long base_size;
1646        off_t base_offset;
1647
1648        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1649        if (!base_offset) {
1650                error("failed to validate delta base reference "
1651                      "at offset %"PRIuMAX" from %s",
1652                      (uintmax_t)curpos, p->pack_name);
1653                return NULL;
1654        }
1655        unuse_pack(w_curs);
1656        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1657        if (!base) {
1658                /*
1659                 * We're probably in deep shit, but let's try to fetch
1660                 * the required base anyway from another pack or loose.
1661                 * This is costly but should happen only in the presence
1662                 * of a corrupted pack, and is better than failing outright.
1663                 */
1664                struct revindex_entry *revidx = find_pack_revindex(p, base_offset);
1665                const unsigned char *base_sha1 =
1666                                        nth_packed_object_sha1(p, revidx->nr);
1667                error("failed to read delta base object %s"
1668                      " at offset %"PRIuMAX" from %s",
1669                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1670                      p->pack_name);
1671                mark_bad_packed_object(p, base_sha1);
1672                base = read_object(base_sha1, type, &base_size);
1673                if (!base)
1674                        return NULL;
1675        }
1676
1677        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1678        if (!delta_data) {
1679                error("failed to unpack compressed delta "
1680                      "at offset %"PRIuMAX" from %s",
1681                      (uintmax_t)curpos, p->pack_name);
1682                free(base);
1683                return NULL;
1684        }
1685        result = patch_delta(base, base_size,
1686                             delta_data, delta_size,
1687                             sizep);
1688        if (!result)
1689                die("failed to apply delta");
1690        free(delta_data);
1691        add_delta_base_cache(p, base_offset, base, base_size, *type);
1692        return result;
1693}
1694
1695void *unpack_entry(struct packed_git *p, off_t obj_offset,
1696                   enum object_type *type, unsigned long *sizep)
1697{
1698        struct pack_window *w_curs = NULL;
1699        off_t curpos = obj_offset;
1700        void *data;
1701
1702        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1703        switch (*type) {
1704        case OBJ_OFS_DELTA:
1705        case OBJ_REF_DELTA:
1706                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1707                                          obj_offset, type, sizep);
1708                break;
1709        case OBJ_COMMIT:
1710        case OBJ_TREE:
1711        case OBJ_BLOB:
1712        case OBJ_TAG:
1713                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1714                break;
1715        default:
1716                data = NULL;
1717                error("unknown object type %i at offset %"PRIuMAX" in %s",
1718                      *type, (uintmax_t)obj_offset, p->pack_name);
1719        }
1720        unuse_pack(&w_curs);
1721        return data;
1722}
1723
1724const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1725                                            uint32_t n)
1726{
1727        const unsigned char *index = p->index_data;
1728        if (!index) {
1729                if (open_pack_index(p))
1730                        return NULL;
1731                index = p->index_data;
1732        }
1733        if (n >= p->num_objects)
1734                return NULL;
1735        index += 4 * 256;
1736        if (p->index_version == 1) {
1737                return index + 24 * n + 4;
1738        } else {
1739                index += 8;
1740                return index + 20 * n;
1741        }
1742}
1743
1744off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1745{
1746        const unsigned char *index = p->index_data;
1747        index += 4 * 256;
1748        if (p->index_version == 1) {
1749                return ntohl(*((uint32_t *)(index + 24 * n)));
1750        } else {
1751                uint32_t off;
1752                index += 8 + p->num_objects * (20 + 4);
1753                off = ntohl(*((uint32_t *)(index + 4 * n)));
1754                if (!(off & 0x80000000))
1755                        return off;
1756                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1757                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1758                                   ntohl(*((uint32_t *)(index + 4)));
1759        }
1760}
1761
1762off_t find_pack_entry_one(const unsigned char *sha1,
1763                                  struct packed_git *p)
1764{
1765        const uint32_t *level1_ofs = p->index_data;
1766        const unsigned char *index = p->index_data;
1767        unsigned hi, lo, stride;
1768        static int use_lookup = -1;
1769        static int debug_lookup = -1;
1770
1771        if (debug_lookup < 0)
1772                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1773
1774        if (!index) {
1775                if (open_pack_index(p))
1776                        return 0;
1777                level1_ofs = p->index_data;
1778                index = p->index_data;
1779        }
1780        if (p->index_version > 1) {
1781                level1_ofs += 2;
1782                index += 8;
1783        }
1784        index += 4 * 256;
1785        hi = ntohl(level1_ofs[*sha1]);
1786        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1787        if (p->index_version > 1) {
1788                stride = 20;
1789        } else {
1790                stride = 24;
1791                index += 4;
1792        }
1793
1794        if (debug_lookup)
1795                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1796                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1797
1798        if (use_lookup < 0)
1799                use_lookup = !!getenv("GIT_USE_LOOKUP");
1800        if (use_lookup) {
1801                int pos = sha1_entry_pos(index, stride, 0,
1802                                         lo, hi, p->num_objects, sha1);
1803                if (pos < 0)
1804                        return 0;
1805                return nth_packed_object_offset(p, pos);
1806        }
1807
1808        do {
1809                unsigned mi = (lo + hi) / 2;
1810                int cmp = hashcmp(index + mi * stride, sha1);
1811
1812                if (debug_lookup)
1813                        printf("lo %u hi %u rg %u mi %u\n",
1814                               lo, hi, hi - lo, mi);
1815                if (!cmp)
1816                        return nth_packed_object_offset(p, mi);
1817                if (cmp > 0)
1818                        hi = mi;
1819                else
1820                        lo = mi+1;
1821        } while (lo < hi);
1822        return 0;
1823}
1824
1825int matches_pack_name(struct packed_git *p, const char *name)
1826{
1827        const char *last_c, *c;
1828
1829        if (!strcmp(p->pack_name, name))
1830                return 1;
1831
1832        for (c = p->pack_name, last_c = c; *c;)
1833                if (*c == '/')
1834                        last_c = ++c;
1835                else
1836                        ++c;
1837        if (!strcmp(last_c, name))
1838                return 1;
1839
1840        return 0;
1841}
1842
1843static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1844{
1845        static struct packed_git *last_found = (void *)1;
1846        struct packed_git *p;
1847        off_t offset;
1848
1849        prepare_packed_git();
1850        if (!packed_git)
1851                return 0;
1852        p = (last_found == (void *)1) ? packed_git : last_found;
1853
1854        do {
1855                if (ignore_packed) {
1856                        const char **ig;
1857                        for (ig = ignore_packed; *ig; ig++)
1858                                if (matches_pack_name(p, *ig))
1859                                        break;
1860                        if (*ig)
1861                                goto next;
1862                }
1863
1864                if (p->num_bad_objects) {
1865                        unsigned i;
1866                        for (i = 0; i < p->num_bad_objects; i++)
1867                                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1868                                        goto next;
1869                }
1870
1871                offset = find_pack_entry_one(sha1, p);
1872                if (offset) {
1873                        /*
1874                         * We are about to tell the caller where they can
1875                         * locate the requested object.  We better make
1876                         * sure the packfile is still here and can be
1877                         * accessed before supplying that answer, as
1878                         * it may have been deleted since the index
1879                         * was loaded!
1880                         */
1881                        if (p->pack_fd == -1 && open_packed_git(p)) {
1882                                error("packfile %s cannot be accessed", p->pack_name);
1883                                goto next;
1884                        }
1885                        e->offset = offset;
1886                        e->p = p;
1887                        hashcpy(e->sha1, sha1);
1888                        last_found = p;
1889                        return 1;
1890                }
1891
1892                next:
1893                if (p == last_found)
1894                        p = packed_git;
1895                else
1896                        p = p->next;
1897                if (p == last_found)
1898                        p = p->next;
1899        } while (p);
1900        return 0;
1901}
1902
1903struct packed_git *find_sha1_pack(const unsigned char *sha1,
1904                                  struct packed_git *packs)
1905{
1906        struct packed_git *p;
1907
1908        for (p = packs; p; p = p->next) {
1909                if (find_pack_entry_one(sha1, p))
1910                        return p;
1911        }
1912        return NULL;
1913
1914}
1915
1916static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1917{
1918        int status;
1919        unsigned long mapsize, size;
1920        void *map;
1921        z_stream stream;
1922        char hdr[32];
1923
1924        map = map_sha1_file(sha1, &mapsize);
1925        if (!map)
1926                return error("unable to find %s", sha1_to_hex(sha1));
1927        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1928                status = error("unable to unpack %s header",
1929                               sha1_to_hex(sha1));
1930        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1931                status = error("unable to parse %s header", sha1_to_hex(sha1));
1932        else if (sizep)
1933                *sizep = size;
1934        inflateEnd(&stream);
1935        munmap(map, mapsize);
1936        return status;
1937}
1938
1939int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1940{
1941        struct pack_entry e;
1942        int status;
1943
1944        if (!find_pack_entry(sha1, &e, NULL)) {
1945                /* Most likely it's a loose object. */
1946                status = sha1_loose_object_info(sha1, sizep);
1947                if (status >= 0)
1948                        return status;
1949
1950                /* Not a loose object; someone else may have just packed it. */
1951                reprepare_packed_git();
1952                if (!find_pack_entry(sha1, &e, NULL))
1953                        return status;
1954        }
1955        return packed_object_info(e.p, e.offset, sizep);
1956}
1957
1958static void *read_packed_sha1(const unsigned char *sha1,
1959                              enum object_type *type, unsigned long *size)
1960{
1961        struct pack_entry e;
1962        void *data;
1963
1964        if (!find_pack_entry(sha1, &e, NULL))
1965                return NULL;
1966        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1967        if (!data) {
1968                /*
1969                 * We're probably in deep shit, but let's try to fetch
1970                 * the required object anyway from another pack or loose.
1971                 * This should happen only in the presence of a corrupted
1972                 * pack, and is better than failing outright.
1973                 */
1974                error("failed to read object %s at offset %"PRIuMAX" from %s",
1975                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
1976                mark_bad_packed_object(e.p, sha1);
1977                data = read_object(sha1, type, size);
1978        }
1979        return data;
1980}
1981
1982/*
1983 * This is meant to hold a *small* number of objects that you would
1984 * want read_sha1_file() to be able to return, but yet you do not want
1985 * to write them into the object store (e.g. a browse-only
1986 * application).
1987 */
1988static struct cached_object {
1989        unsigned char sha1[20];
1990        enum object_type type;
1991        void *buf;
1992        unsigned long size;
1993} *cached_objects;
1994static int cached_object_nr, cached_object_alloc;
1995
1996static struct cached_object empty_tree = {
1997        /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
1998        "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
1999        "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2000        OBJ_TREE,
2001        "",
2002        0
2003};
2004
2005static struct cached_object *find_cached_object(const unsigned char *sha1)
2006{
2007        int i;
2008        struct cached_object *co = cached_objects;
2009
2010        for (i = 0; i < cached_object_nr; i++, co++) {
2011                if (!hashcmp(co->sha1, sha1))
2012                        return co;
2013        }
2014        if (!hashcmp(sha1, empty_tree.sha1))
2015                return &empty_tree;
2016        return NULL;
2017}
2018
2019int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2020                      unsigned char *sha1)
2021{
2022        struct cached_object *co;
2023
2024        hash_sha1_file(buf, len, typename(type), sha1);
2025        if (has_sha1_file(sha1) || find_cached_object(sha1))
2026                return 0;
2027        if (cached_object_alloc <= cached_object_nr) {
2028                cached_object_alloc = alloc_nr(cached_object_alloc);
2029                cached_objects = xrealloc(cached_objects,
2030                                          sizeof(*cached_objects) *
2031                                          cached_object_alloc);
2032        }
2033        co = &cached_objects[cached_object_nr++];
2034        co->size = len;
2035        co->type = type;
2036        co->buf = xmalloc(len);
2037        memcpy(co->buf, buf, len);
2038        hashcpy(co->sha1, sha1);
2039        return 0;
2040}
2041
2042void *read_object(const unsigned char *sha1, enum object_type *type,
2043                  unsigned long *size)
2044{
2045        unsigned long mapsize;
2046        void *map, *buf;
2047        struct cached_object *co;
2048
2049        co = find_cached_object(sha1);
2050        if (co) {
2051                *type = co->type;
2052                *size = co->size;
2053                return xmemdupz(co->buf, co->size);
2054        }
2055
2056        buf = read_packed_sha1(sha1, type, size);
2057        if (buf)
2058                return buf;
2059        map = map_sha1_file(sha1, &mapsize);
2060        if (map) {
2061                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2062                munmap(map, mapsize);
2063                return buf;
2064        }
2065        reprepare_packed_git();
2066        return read_packed_sha1(sha1, type, size);
2067}
2068
2069void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2070                     unsigned long *size)
2071{
2072        void *data = read_object(sha1, type, size);
2073        /* legacy behavior is to die on corrupted objects */
2074        if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2075                die("object %s is corrupted", sha1_to_hex(sha1));
2076        return data;
2077}
2078
2079void *read_object_with_reference(const unsigned char *sha1,
2080                                 const char *required_type_name,
2081                                 unsigned long *size,
2082                                 unsigned char *actual_sha1_return)
2083{
2084        enum object_type type, required_type;
2085        void *buffer;
2086        unsigned long isize;
2087        unsigned char actual_sha1[20];
2088
2089        required_type = type_from_string(required_type_name);
2090        hashcpy(actual_sha1, sha1);
2091        while (1) {
2092                int ref_length = -1;
2093                const char *ref_type = NULL;
2094
2095                buffer = read_sha1_file(actual_sha1, &type, &isize);
2096                if (!buffer)
2097                        return NULL;
2098                if (type == required_type) {
2099                        *size = isize;
2100                        if (actual_sha1_return)
2101                                hashcpy(actual_sha1_return, actual_sha1);
2102                        return buffer;
2103                }
2104                /* Handle references */
2105                else if (type == OBJ_COMMIT)
2106                        ref_type = "tree ";
2107                else if (type == OBJ_TAG)
2108                        ref_type = "object ";
2109                else {
2110                        free(buffer);
2111                        return NULL;
2112                }
2113                ref_length = strlen(ref_type);
2114
2115                if (ref_length + 40 > isize ||
2116                    memcmp(buffer, ref_type, ref_length) ||
2117                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2118                        free(buffer);
2119                        return NULL;
2120                }
2121                free(buffer);
2122                /* Now we have the ID of the referred-to object in
2123                 * actual_sha1.  Check again. */
2124        }
2125}
2126
2127static void write_sha1_file_prepare(const void *buf, unsigned long len,
2128                                    const char *type, unsigned char *sha1,
2129                                    char *hdr, int *hdrlen)
2130{
2131        SHA_CTX c;
2132
2133        /* Generate the header */
2134        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2135
2136        /* Sha1.. */
2137        SHA1_Init(&c);
2138        SHA1_Update(&c, hdr, *hdrlen);
2139        SHA1_Update(&c, buf, len);
2140        SHA1_Final(sha1, &c);
2141}
2142
2143/*
2144 * Move the just written object into its final resting place
2145 */
2146int move_temp_to_file(const char *tmpfile, const char *filename)
2147{
2148        int ret = link(tmpfile, filename);
2149
2150        /*
2151         * Coda hack - coda doesn't like cross-directory links,
2152         * so we fall back to a rename, which will mean that it
2153         * won't be able to check collisions, but that's not a
2154         * big deal.
2155         *
2156         * The same holds for FAT formatted media.
2157         *
2158         * When this succeeds, we just return 0. We have nothing
2159         * left to unlink.
2160         */
2161        if (ret && ret != EEXIST) {
2162                if (!rename(tmpfile, filename))
2163                        return 0;
2164                ret = errno;
2165        }
2166        unlink(tmpfile);
2167        if (ret) {
2168                if (ret != EEXIST) {
2169                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2170                }
2171                /* FIXME!!! Collision check here ? */
2172        }
2173
2174        return 0;
2175}
2176
2177static int write_buffer(int fd, const void *buf, size_t len)
2178{
2179        if (write_in_full(fd, buf, len) < 0)
2180                return error("file write error (%s)", strerror(errno));
2181        return 0;
2182}
2183
2184int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2185                   unsigned char *sha1)
2186{
2187        char hdr[32];
2188        int hdrlen;
2189        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2190        return 0;
2191}
2192
2193/* Finalize a file on disk, and close it. */
2194static void close_sha1_file(int fd)
2195{
2196        if (fsync_object_files)
2197                fsync_or_die(fd, "sha1 file");
2198        fchmod(fd, 0444);
2199        if (close(fd) != 0)
2200                die("unable to write sha1 file");
2201}
2202
2203/* Size of directory component, including the ending '/' */
2204static inline int directory_size(const char *filename)
2205{
2206        const char *s = strrchr(filename, '/');
2207        if (!s)
2208                return 0;
2209        return s - filename + 1;
2210}
2211
2212/*
2213 * This creates a temporary file in the same directory as the final
2214 * 'filename'
2215 *
2216 * We want to avoid cross-directory filename renames, because those
2217 * can have problems on various filesystems (FAT, NFS, Coda).
2218 */
2219static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2220{
2221        int fd, dirlen = directory_size(filename);
2222
2223        if (dirlen + 20 > bufsiz) {
2224                errno = ENAMETOOLONG;
2225                return -1;
2226        }
2227        memcpy(buffer, filename, dirlen);
2228        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2229        fd = mkstemp(buffer);
2230        if (fd < 0 && dirlen) {
2231                /* Make sure the directory exists */
2232                memcpy(buffer, filename, dirlen);
2233                buffer[dirlen-1] = 0;
2234                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2235                        return -1;
2236
2237                /* Try again */
2238                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2239                fd = mkstemp(buffer);
2240        }
2241        return fd;
2242}
2243
2244static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2245                              void *buf, unsigned long len, time_t mtime)
2246{
2247        int fd, size, ret;
2248        unsigned char *compressed;
2249        z_stream stream;
2250        char *filename;
2251        static char tmpfile[PATH_MAX];
2252
2253        filename = sha1_file_name(sha1);
2254        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2255        if (fd < 0) {
2256                if (errno == EPERM)
2257                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2258                else
2259                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2260        }
2261
2262        /* Set it up */
2263        memset(&stream, 0, sizeof(stream));
2264        deflateInit(&stream, zlib_compression_level);
2265        size = 8 + deflateBound(&stream, len+hdrlen);
2266        compressed = xmalloc(size);
2267
2268        /* Compress it */
2269        stream.next_out = compressed;
2270        stream.avail_out = size;
2271
2272        /* First header.. */
2273        stream.next_in = (unsigned char *)hdr;
2274        stream.avail_in = hdrlen;
2275        while (deflate(&stream, 0) == Z_OK)
2276                /* nothing */;
2277
2278        /* Then the data itself.. */
2279        stream.next_in = buf;
2280        stream.avail_in = len;
2281        ret = deflate(&stream, Z_FINISH);
2282        if (ret != Z_STREAM_END)
2283                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2284
2285        ret = deflateEnd(&stream);
2286        if (ret != Z_OK)
2287                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2288
2289        size = stream.total_out;
2290
2291        if (write_buffer(fd, compressed, size) < 0)
2292                die("unable to write sha1 file");
2293        close_sha1_file(fd);
2294        free(compressed);
2295
2296        if (mtime) {
2297                struct utimbuf utb;
2298                utb.actime = mtime;
2299                utb.modtime = mtime;
2300                if (utime(tmpfile, &utb) < 0)
2301                        warning("failed utime() on %s: %s",
2302                                tmpfile, strerror(errno));
2303        }
2304
2305        return move_temp_to_file(tmpfile, filename);
2306}
2307
2308int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2309{
2310        unsigned char sha1[20];
2311        char hdr[32];
2312        int hdrlen;
2313
2314        /* Normally if we have it in the pack then we do not bother writing
2315         * it out into .git/objects/??/?{38} file.
2316         */
2317        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2318        if (returnsha1)
2319                hashcpy(returnsha1, sha1);
2320        if (has_sha1_file(sha1))
2321                return 0;
2322        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2323}
2324
2325int force_object_loose(const unsigned char *sha1, time_t mtime)
2326{
2327        void *buf;
2328        unsigned long len;
2329        enum object_type type;
2330        char hdr[32];
2331        int hdrlen;
2332
2333        if (has_loose_object(sha1))
2334                return 0;
2335        buf = read_packed_sha1(sha1, &type, &len);
2336        if (!buf)
2337                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2338        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2339        return write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2340}
2341
2342int has_pack_index(const unsigned char *sha1)
2343{
2344        struct stat st;
2345        if (stat(sha1_pack_index_name(sha1), &st))
2346                return 0;
2347        return 1;
2348}
2349
2350int has_pack_file(const unsigned char *sha1)
2351{
2352        struct stat st;
2353        if (stat(sha1_pack_name(sha1), &st))
2354                return 0;
2355        return 1;
2356}
2357
2358int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2359{
2360        struct pack_entry e;
2361        return find_pack_entry(sha1, &e, ignore_packed);
2362}
2363
2364int has_sha1_file(const unsigned char *sha1)
2365{
2366        struct pack_entry e;
2367
2368        if (find_pack_entry(sha1, &e, NULL))
2369                return 1;
2370        return has_loose_object(sha1);
2371}
2372
2373static int index_mem(unsigned char *sha1, void *buf, size_t size,
2374                     int write_object, enum object_type type, const char *path)
2375{
2376        int ret, re_allocated = 0;
2377
2378        if (!type)
2379                type = OBJ_BLOB;
2380
2381        /*
2382         * Convert blobs to git internal format
2383         */
2384        if ((type == OBJ_BLOB) && path) {
2385                struct strbuf nbuf;
2386                strbuf_init(&nbuf, 0);
2387                if (convert_to_git(path, buf, size, &nbuf,
2388                                   write_object ? safe_crlf : 0)) {
2389                        buf = strbuf_detach(&nbuf, &size);
2390                        re_allocated = 1;
2391                }
2392        }
2393
2394        if (write_object)
2395                ret = write_sha1_file(buf, size, typename(type), sha1);
2396        else
2397                ret = hash_sha1_file(buf, size, typename(type), sha1);
2398        if (re_allocated)
2399                free(buf);
2400        return ret;
2401}
2402
2403int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2404             enum object_type type, const char *path)
2405{
2406        int ret;
2407        size_t size = xsize_t(st->st_size);
2408
2409        if (!S_ISREG(st->st_mode)) {
2410                struct strbuf sbuf;
2411                strbuf_init(&sbuf, 0);
2412                if (strbuf_read(&sbuf, fd, 4096) >= 0)
2413                        ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2414                                        type, path);
2415                else
2416                        ret = -1;
2417                strbuf_release(&sbuf);
2418        } else if (size) {
2419                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2420                ret = index_mem(sha1, buf, size, write_object, type, path);
2421                munmap(buf, size);
2422        } else
2423                ret = index_mem(sha1, NULL, size, write_object, type, path);
2424        close(fd);
2425        return ret;
2426}
2427
2428int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2429{
2430        int fd;
2431        char *target;
2432        size_t len;
2433
2434        switch (st->st_mode & S_IFMT) {
2435        case S_IFREG:
2436                fd = open(path, O_RDONLY);
2437                if (fd < 0)
2438                        return error("open(\"%s\"): %s", path,
2439                                     strerror(errno));
2440                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2441                        return error("%s: failed to insert into database",
2442                                     path);
2443                break;
2444        case S_IFLNK:
2445                len = xsize_t(st->st_size);
2446                target = xmalloc(len + 1);
2447                if (readlink(path, target, len + 1) != st->st_size) {
2448                        char *errstr = strerror(errno);
2449                        free(target);
2450                        return error("readlink(\"%s\"): %s", path,
2451                                     errstr);
2452                }
2453                if (!write_object)
2454                        hash_sha1_file(target, len, blob_type, sha1);
2455                else if (write_sha1_file(target, len, blob_type, sha1))
2456                        return error("%s: failed to insert into database",
2457                                     path);
2458                free(target);
2459                break;
2460        case S_IFDIR:
2461                return resolve_gitlink_ref(path, "HEAD", sha1);
2462        default:
2463                return error("%s: unsupported file type", path);
2464        }
2465        return 0;
2466}
2467
2468int read_pack_header(int fd, struct pack_header *header)
2469{
2470        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2471                /* "eof before pack header was fully read" */
2472                return PH_ERROR_EOF;
2473
2474        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2475                /* "protocol error (pack signature mismatch detected)" */
2476                return PH_ERROR_PACK_SIGNATURE;
2477        if (!pack_version_ok(header->hdr_version))
2478                /* "protocol error (pack version unsupported)" */
2479                return PH_ERROR_PROTOCOL;
2480        return 0;
2481}