sha1_file.con commit mergetool: Add prompt to continue after failing to merge a file (b0169d8)
   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                while (*++pos == '/')
 103                        ;
 104                if (!*pos)
 105                        break;
 106                *--pos = '\0';
 107                if (!stat(path, &st)) {
 108                        /* path exists */
 109                        if (!S_ISDIR(st.st_mode)) {
 110                                *pos = '/';
 111                                return -3;
 112                        }
 113                }
 114                else if (mkdir(path, 0777)) {
 115                        *pos = '/';
 116                        return -1;
 117                }
 118                else if (adjust_shared_perm(path)) {
 119                        *pos = '/';
 120                        return -2;
 121                }
 122                *pos++ = '/';
 123        }
 124        return 0;
 125}
 126
 127int safe_create_leading_directories_const(const char *path)
 128{
 129        /* path points to cache entries, so xstrdup before messing with it */
 130        char *buf = xstrdup(path);
 131        int result = safe_create_leading_directories(buf);
 132        free(buf);
 133        return result;
 134}
 135
 136char *sha1_to_hex(const unsigned char *sha1)
 137{
 138        static int bufno;
 139        static char hexbuffer[4][50];
 140        static const char hex[] = "0123456789abcdef";
 141        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 142        int i;
 143
 144        for (i = 0; i < 20; i++) {
 145                unsigned int val = *sha1++;
 146                *buf++ = hex[val >> 4];
 147                *buf++ = hex[val & 0xf];
 148        }
 149        *buf = '\0';
 150
 151        return buffer;
 152}
 153
 154static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 155{
 156        int i;
 157        for (i = 0; i < 20; i++) {
 158                static char hex[] = "0123456789abcdef";
 159                unsigned int val = sha1[i];
 160                char *pos = pathbuf + i*2 + (i > 0);
 161                *pos++ = hex[val >> 4];
 162                *pos = hex[val & 0xf];
 163        }
 164}
 165
 166/*
 167 * NOTE! This returns a statically allocated buffer, so you have to be
 168 * careful about using it. Do an "xstrdup()" if you need to save the
 169 * filename.
 170 *
 171 * Also note that this returns the location for creating.  Reading
 172 * SHA1 file can happen from any alternate directory listed in the
 173 * DB_ENVIRONMENT environment variable if it is not found in
 174 * the primary object database.
 175 */
 176char *sha1_file_name(const unsigned char *sha1)
 177{
 178        static char *name, *base;
 179
 180        if (!base) {
 181                const char *sha1_file_directory = get_object_directory();
 182                int len = strlen(sha1_file_directory);
 183                base = xmalloc(len + 60);
 184                memcpy(base, sha1_file_directory, len);
 185                memset(base+len, 0, 60);
 186                base[len] = '/';
 187                base[len+3] = '/';
 188                name = base + len + 1;
 189        }
 190        fill_sha1_path(name, sha1);
 191        return base;
 192}
 193
 194static char *sha1_get_pack_name(const unsigned char *sha1,
 195                                char **name, char **base, const char *which)
 196{
 197        static const char hex[] = "0123456789abcdef";
 198        char *buf;
 199        int i;
 200
 201        if (!*base) {
 202                const char *sha1_file_directory = get_object_directory();
 203                int len = strlen(sha1_file_directory);
 204                *base = xmalloc(len + 60);
 205                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 206                        sha1_file_directory, which);
 207                *name = *base + len + 11;
 208        }
 209
 210        buf = *name;
 211
 212        for (i = 0; i < 20; i++) {
 213                unsigned int val = *sha1++;
 214                *buf++ = hex[val >> 4];
 215                *buf++ = hex[val & 0xf];
 216        }
 217
 218        return *base;
 219}
 220
 221char *sha1_pack_name(const unsigned char *sha1)
 222{
 223        static char *name, *base;
 224
 225        return sha1_get_pack_name(sha1, &name, &base, "pack");
 226}
 227
 228char *sha1_pack_index_name(const unsigned char *sha1)
 229{
 230        static char *name, *base;
 231
 232        return sha1_get_pack_name(sha1, &name, &base, "idx");
 233}
 234
 235struct alternate_object_database *alt_odb_list;
 236static struct alternate_object_database **alt_odb_tail;
 237
 238static void read_info_alternates(const char * alternates, int depth);
 239
 240/*
 241 * Prepare alternate object database registry.
 242 *
 243 * The variable alt_odb_list points at the list of struct
 244 * alternate_object_database.  The elements on this list come from
 245 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 246 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 247 * whose contents is similar to that environment variable but can be
 248 * LF separated.  Its base points at a statically allocated buffer that
 249 * contains "/the/directory/corresponding/to/.git/objects/...", while
 250 * its name points just after the slash at the end of ".git/objects/"
 251 * in the example above, and has enough space to hold 40-byte hex
 252 * SHA1, an extra slash for the first level indirection, and the
 253 * terminating NUL.
 254 */
 255static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 256{
 257        const char *objdir = get_object_directory();
 258        struct alternate_object_database *ent;
 259        struct alternate_object_database *alt;
 260        /* 43 = 40-byte + 2 '/' + terminating NUL */
 261        int pfxlen = len;
 262        int entlen = pfxlen + 43;
 263        int base_len = -1;
 264
 265        if (!is_absolute_path(entry) && relative_base) {
 266                /* Relative alt-odb */
 267                if (base_len < 0)
 268                        base_len = strlen(relative_base) + 1;
 269                entlen += base_len;
 270                pfxlen += base_len;
 271        }
 272        ent = xmalloc(sizeof(*ent) + entlen);
 273
 274        if (!is_absolute_path(entry) && relative_base) {
 275                memcpy(ent->base, relative_base, base_len - 1);
 276                ent->base[base_len - 1] = '/';
 277                memcpy(ent->base + base_len, entry, len);
 278        }
 279        else
 280                memcpy(ent->base, entry, pfxlen);
 281
 282        ent->name = ent->base + pfxlen + 1;
 283        ent->base[pfxlen + 3] = '/';
 284        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 285
 286        /* Detect cases where alternate disappeared */
 287        if (!is_directory(ent->base)) {
 288                error("object directory %s does not exist; "
 289                      "check .git/objects/info/alternates.",
 290                      ent->base);
 291                free(ent);
 292                return -1;
 293        }
 294
 295        /* Prevent the common mistake of listing the same
 296         * thing twice, or object directory itself.
 297         */
 298        for (alt = alt_odb_list; alt; alt = alt->next) {
 299                if (!memcmp(ent->base, alt->base, pfxlen)) {
 300                        free(ent);
 301                        return -1;
 302                }
 303        }
 304        if (!memcmp(ent->base, objdir, pfxlen)) {
 305                free(ent);
 306                return -1;
 307        }
 308
 309        /* add the alternate entry */
 310        *alt_odb_tail = ent;
 311        alt_odb_tail = &(ent->next);
 312        ent->next = NULL;
 313
 314        /* recursively add alternates */
 315        read_info_alternates(ent->base, depth + 1);
 316
 317        ent->base[pfxlen] = '/';
 318
 319        return 0;
 320}
 321
 322static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 323                                 const char *relative_base, int depth)
 324{
 325        const char *cp, *last;
 326
 327        if (depth > 5) {
 328                error("%s: ignoring alternate object stores, nesting too deep.",
 329                                relative_base);
 330                return;
 331        }
 332
 333        last = alt;
 334        while (last < ep) {
 335                cp = last;
 336                if (cp < ep && *cp == '#') {
 337                        while (cp < ep && *cp != sep)
 338                                cp++;
 339                        last = cp + 1;
 340                        continue;
 341                }
 342                while (cp < ep && *cp != sep)
 343                        cp++;
 344                if (last != cp) {
 345                        if (!is_absolute_path(last) && depth) {
 346                                error("%s: ignoring relative alternate object store %s",
 347                                                relative_base, last);
 348                        } else {
 349                                link_alt_odb_entry(last, cp - last,
 350                                                relative_base, depth);
 351                        }
 352                }
 353                while (cp < ep && *cp == sep)
 354                        cp++;
 355                last = cp;
 356        }
 357}
 358
 359static void read_info_alternates(const char * relative_base, int depth)
 360{
 361        char *map;
 362        size_t mapsz;
 363        struct stat st;
 364        const char alt_file_name[] = "info/alternates";
 365        /* Given that relative_base is no longer than PATH_MAX,
 366           ensure that "path" has enough space to append "/", the
 367           file name, "info/alternates", and a trailing NUL.  */
 368        char path[PATH_MAX + 1 + sizeof alt_file_name];
 369        int fd;
 370
 371        sprintf(path, "%s/%s", relative_base, alt_file_name);
 372        fd = open(path, O_RDONLY);
 373        if (fd < 0)
 374                return;
 375        if (fstat(fd, &st) || (st.st_size == 0)) {
 376                close(fd);
 377                return;
 378        }
 379        mapsz = xsize_t(st.st_size);
 380        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 381        close(fd);
 382
 383        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 384
 385        munmap(map, mapsz);
 386}
 387
 388void add_to_alternates_file(const char *reference)
 389{
 390        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 391        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
 392        char *alt = mkpath("%s/objects\n", reference);
 393        write_or_die(fd, alt, strlen(alt));
 394        if (commit_lock_file(lock))
 395                die("could not close alternates file");
 396        if (alt_odb_tail)
 397                link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
 398}
 399
 400void foreach_alt_odb(alt_odb_fn fn, void *cb)
 401{
 402        struct alternate_object_database *ent;
 403
 404        prepare_alt_odb();
 405        for (ent = alt_odb_list; ent; ent = ent->next)
 406                if (fn(ent, cb))
 407                        return;
 408}
 409
 410void prepare_alt_odb(void)
 411{
 412        const char *alt;
 413
 414        if (alt_odb_tail)
 415                return;
 416
 417        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 418        if (!alt) alt = "";
 419
 420        alt_odb_tail = &alt_odb_list;
 421        link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
 422
 423        read_info_alternates(get_object_directory(), 0);
 424}
 425
 426static int has_loose_object_local(const unsigned char *sha1)
 427{
 428        char *name = sha1_file_name(sha1);
 429        return !access(name, F_OK);
 430}
 431
 432int has_loose_object_nonlocal(const unsigned char *sha1)
 433{
 434        struct alternate_object_database *alt;
 435        prepare_alt_odb();
 436        for (alt = alt_odb_list; alt; alt = alt->next) {
 437                fill_sha1_path(alt->name, sha1);
 438                if (!access(alt->base, F_OK))
 439                        return 1;
 440        }
 441        return 0;
 442}
 443
 444static int has_loose_object(const unsigned char *sha1)
 445{
 446        return has_loose_object_local(sha1) ||
 447               has_loose_object_nonlocal(sha1);
 448}
 449
 450static unsigned int pack_used_ctr;
 451static unsigned int pack_mmap_calls;
 452static unsigned int peak_pack_open_windows;
 453static unsigned int pack_open_windows;
 454static size_t peak_pack_mapped;
 455static size_t pack_mapped;
 456struct packed_git *packed_git;
 457
 458void pack_report(void)
 459{
 460        fprintf(stderr,
 461                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 462                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 463                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 464                sz_fmt(getpagesize()),
 465                sz_fmt(packed_git_window_size),
 466                sz_fmt(packed_git_limit));
 467        fprintf(stderr,
 468                "pack_report: pack_used_ctr            = %10u\n"
 469                "pack_report: pack_mmap_calls          = %10u\n"
 470                "pack_report: pack_open_windows        = %10u / %10u\n"
 471                "pack_report: pack_mapped              = "
 472                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 473                pack_used_ctr,
 474                pack_mmap_calls,
 475                pack_open_windows, peak_pack_open_windows,
 476                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 477}
 478
 479static int check_packed_git_idx(const char *path,  struct packed_git *p)
 480{
 481        void *idx_map;
 482        struct pack_idx_header *hdr;
 483        size_t idx_size;
 484        uint32_t version, nr, i, *index;
 485        int fd = open(path, O_RDONLY);
 486        struct stat st;
 487
 488        if (fd < 0)
 489                return -1;
 490        if (fstat(fd, &st)) {
 491                close(fd);
 492                return -1;
 493        }
 494        idx_size = xsize_t(st.st_size);
 495        if (idx_size < 4 * 256 + 20 + 20) {
 496                close(fd);
 497                return error("index file %s is too small", path);
 498        }
 499        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 500        close(fd);
 501
 502        hdr = idx_map;
 503        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 504                version = ntohl(hdr->idx_version);
 505                if (version < 2 || version > 2) {
 506                        munmap(idx_map, idx_size);
 507                        return error("index file %s is version %"PRIu32
 508                                     " and is not supported by this binary"
 509                                     " (try upgrading GIT to a newer version)",
 510                                     path, version);
 511                }
 512        } else
 513                version = 1;
 514
 515        nr = 0;
 516        index = idx_map;
 517        if (version > 1)
 518                index += 2;  /* skip index header */
 519        for (i = 0; i < 256; i++) {
 520                uint32_t n = ntohl(index[i]);
 521                if (n < nr) {
 522                        munmap(idx_map, idx_size);
 523                        return error("non-monotonic index %s", path);
 524                }
 525                nr = n;
 526        }
 527
 528        if (version == 1) {
 529                /*
 530                 * Total size:
 531                 *  - 256 index entries 4 bytes each
 532                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 533                 *  - 20-byte SHA1 of the packfile
 534                 *  - 20-byte SHA1 file checksum
 535                 */
 536                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 537                        munmap(idx_map, idx_size);
 538                        return error("wrong index v1 file size in %s", path);
 539                }
 540        } else if (version == 2) {
 541                /*
 542                 * Minimum size:
 543                 *  - 8 bytes of header
 544                 *  - 256 index entries 4 bytes each
 545                 *  - 20-byte sha1 entry * nr
 546                 *  - 4-byte crc entry * nr
 547                 *  - 4-byte offset entry * nr
 548                 *  - 20-byte SHA1 of the packfile
 549                 *  - 20-byte SHA1 file checksum
 550                 * And after the 4-byte offset table might be a
 551                 * variable sized table containing 8-byte entries
 552                 * for offsets larger than 2^31.
 553                 */
 554                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 555                unsigned long max_size = min_size;
 556                if (nr)
 557                        max_size += (nr - 1)*8;
 558                if (idx_size < min_size || idx_size > max_size) {
 559                        munmap(idx_map, idx_size);
 560                        return error("wrong index v2 file size in %s", path);
 561                }
 562                if (idx_size != min_size &&
 563                    /*
 564                     * make sure we can deal with large pack offsets.
 565                     * 31-bit signed offset won't be enough, neither
 566                     * 32-bit unsigned one will be.
 567                     */
 568                    (sizeof(off_t) <= 4)) {
 569                        munmap(idx_map, idx_size);
 570                        return error("pack too large for current definition of off_t in %s", path);
 571                }
 572        }
 573
 574        p->index_version = version;
 575        p->index_data = idx_map;
 576        p->index_size = idx_size;
 577        p->num_objects = nr;
 578        return 0;
 579}
 580
 581int open_pack_index(struct packed_git *p)
 582{
 583        char *idx_name;
 584        int ret;
 585
 586        if (p->index_data)
 587                return 0;
 588
 589        idx_name = xstrdup(p->pack_name);
 590        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 591        ret = check_packed_git_idx(idx_name, p);
 592        free(idx_name);
 593        return ret;
 594}
 595
 596static void scan_windows(struct packed_git *p,
 597        struct packed_git **lru_p,
 598        struct pack_window **lru_w,
 599        struct pack_window **lru_l)
 600{
 601        struct pack_window *w, *w_l;
 602
 603        for (w_l = NULL, w = p->windows; w; w = w->next) {
 604                if (!w->inuse_cnt) {
 605                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 606                                *lru_p = p;
 607                                *lru_w = w;
 608                                *lru_l = w_l;
 609                        }
 610                }
 611                w_l = w;
 612        }
 613}
 614
 615static int unuse_one_window(struct packed_git *current, int keep_fd)
 616{
 617        struct packed_git *p, *lru_p = NULL;
 618        struct pack_window *lru_w = NULL, *lru_l = NULL;
 619
 620        if (current)
 621                scan_windows(current, &lru_p, &lru_w, &lru_l);
 622        for (p = packed_git; p; p = p->next)
 623                scan_windows(p, &lru_p, &lru_w, &lru_l);
 624        if (lru_p) {
 625                munmap(lru_w->base, lru_w->len);
 626                pack_mapped -= lru_w->len;
 627                if (lru_l)
 628                        lru_l->next = lru_w->next;
 629                else {
 630                        lru_p->windows = lru_w->next;
 631                        if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
 632                                close(lru_p->pack_fd);
 633                                lru_p->pack_fd = -1;
 634                        }
 635                }
 636                free(lru_w);
 637                pack_open_windows--;
 638                return 1;
 639        }
 640        return 0;
 641}
 642
 643void release_pack_memory(size_t need, int fd)
 644{
 645        size_t cur = pack_mapped;
 646        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 647                ; /* nothing */
 648}
 649
 650void close_pack_windows(struct packed_git *p)
 651{
 652        while (p->windows) {
 653                struct pack_window *w = p->windows;
 654
 655                if (w->inuse_cnt)
 656                        die("pack '%s' still has open windows to it",
 657                            p->pack_name);
 658                munmap(w->base, w->len);
 659                pack_mapped -= w->len;
 660                pack_open_windows--;
 661                p->windows = w->next;
 662                free(w);
 663        }
 664}
 665
 666void unuse_pack(struct pack_window **w_cursor)
 667{
 668        struct pack_window *w = *w_cursor;
 669        if (w) {
 670                w->inuse_cnt--;
 671                *w_cursor = NULL;
 672        }
 673}
 674
 675/*
 676 * Do not call this directly as this leaks p->pack_fd on error return;
 677 * call open_packed_git() instead.
 678 */
 679static int open_packed_git_1(struct packed_git *p)
 680{
 681        struct stat st;
 682        struct pack_header hdr;
 683        unsigned char sha1[20];
 684        unsigned char *idx_sha1;
 685        long fd_flag;
 686
 687        if (!p->index_data && open_pack_index(p))
 688                return error("packfile %s index unavailable", p->pack_name);
 689
 690        p->pack_fd = open(p->pack_name, O_RDONLY);
 691        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 692                return -1;
 693
 694        /* If we created the struct before we had the pack we lack size. */
 695        if (!p->pack_size) {
 696                if (!S_ISREG(st.st_mode))
 697                        return error("packfile %s not a regular file", p->pack_name);
 698                p->pack_size = st.st_size;
 699        } else if (p->pack_size != st.st_size)
 700                return error("packfile %s size changed", p->pack_name);
 701
 702        /* We leave these file descriptors open with sliding mmap;
 703         * there is no point keeping them open across exec(), though.
 704         */
 705        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 706        if (fd_flag < 0)
 707                return error("cannot determine file descriptor flags");
 708        fd_flag |= FD_CLOEXEC;
 709        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 710                return error("cannot set FD_CLOEXEC");
 711
 712        /* Verify we recognize this pack file format. */
 713        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 714                return error("file %s is far too short to be a packfile", p->pack_name);
 715        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 716                return error("file %s is not a GIT packfile", p->pack_name);
 717        if (!pack_version_ok(hdr.hdr_version))
 718                return error("packfile %s is version %"PRIu32" and not"
 719                        " supported (try upgrading GIT to a newer version)",
 720                        p->pack_name, ntohl(hdr.hdr_version));
 721
 722        /* Verify the pack matches its index. */
 723        if (p->num_objects != ntohl(hdr.hdr_entries))
 724                return error("packfile %s claims to have %"PRIu32" objects"
 725                             " while index indicates %"PRIu32" objects",
 726                             p->pack_name, ntohl(hdr.hdr_entries),
 727                             p->num_objects);
 728        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 729                return error("end of packfile %s is unavailable", p->pack_name);
 730        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 731                return error("packfile %s signature is unavailable", p->pack_name);
 732        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 733        if (hashcmp(sha1, idx_sha1))
 734                return error("packfile %s does not match index", p->pack_name);
 735        return 0;
 736}
 737
 738static int open_packed_git(struct packed_git *p)
 739{
 740        if (!open_packed_git_1(p))
 741                return 0;
 742        if (p->pack_fd != -1) {
 743                close(p->pack_fd);
 744                p->pack_fd = -1;
 745        }
 746        return -1;
 747}
 748
 749static int in_window(struct pack_window *win, off_t offset)
 750{
 751        /* We must promise at least 20 bytes (one hash) after the
 752         * offset is available from this window, otherwise the offset
 753         * is not actually in this window and a different window (which
 754         * has that one hash excess) must be used.  This is to support
 755         * the object header and delta base parsing routines below.
 756         */
 757        off_t win_off = win->offset;
 758        return win_off <= offset
 759                && (offset + 20) <= (win_off + win->len);
 760}
 761
 762unsigned char* use_pack(struct packed_git *p,
 763                struct pack_window **w_cursor,
 764                off_t offset,
 765                unsigned int *left)
 766{
 767        struct pack_window *win = *w_cursor;
 768
 769        if (p->pack_fd == -1 && open_packed_git(p))
 770                die("packfile %s cannot be accessed", p->pack_name);
 771
 772        /* Since packfiles end in a hash of their content and its
 773         * pointless to ask for an offset into the middle of that
 774         * hash, and the in_window function above wouldn't match
 775         * don't allow an offset too close to the end of the file.
 776         */
 777        if (offset > (p->pack_size - 20))
 778                die("offset beyond end of packfile (truncated pack?)");
 779
 780        if (!win || !in_window(win, offset)) {
 781                if (win)
 782                        win->inuse_cnt--;
 783                for (win = p->windows; win; win = win->next) {
 784                        if (in_window(win, offset))
 785                                break;
 786                }
 787                if (!win) {
 788                        size_t window_align = packed_git_window_size / 2;
 789                        off_t len;
 790                        win = xcalloc(1, sizeof(*win));
 791                        win->offset = (offset / window_align) * window_align;
 792                        len = p->pack_size - win->offset;
 793                        if (len > packed_git_window_size)
 794                                len = packed_git_window_size;
 795                        win->len = (size_t)len;
 796                        pack_mapped += win->len;
 797                        while (packed_git_limit < pack_mapped
 798                                && unuse_one_window(p, p->pack_fd))
 799                                ; /* nothing */
 800                        win->base = xmmap(NULL, win->len,
 801                                PROT_READ, MAP_PRIVATE,
 802                                p->pack_fd, win->offset);
 803                        if (win->base == MAP_FAILED)
 804                                die("packfile %s cannot be mapped: %s",
 805                                        p->pack_name,
 806                                        strerror(errno));
 807                        pack_mmap_calls++;
 808                        pack_open_windows++;
 809                        if (pack_mapped > peak_pack_mapped)
 810                                peak_pack_mapped = pack_mapped;
 811                        if (pack_open_windows > peak_pack_open_windows)
 812                                peak_pack_open_windows = pack_open_windows;
 813                        win->next = p->windows;
 814                        p->windows = win;
 815                }
 816        }
 817        if (win != *w_cursor) {
 818                win->last_used = pack_used_ctr++;
 819                win->inuse_cnt++;
 820                *w_cursor = win;
 821        }
 822        offset -= win->offset;
 823        if (left)
 824                *left = win->len - xsize_t(offset);
 825        return win->base + offset;
 826}
 827
 828static struct packed_git *alloc_packed_git(int extra)
 829{
 830        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 831        memset(p, 0, sizeof(*p));
 832        p->pack_fd = -1;
 833        return p;
 834}
 835
 836struct packed_git *add_packed_git(const char *path, int path_len, int local)
 837{
 838        struct stat st;
 839        struct packed_git *p = alloc_packed_git(path_len + 2);
 840
 841        /*
 842         * Make sure a corresponding .pack file exists and that
 843         * the index looks sane.
 844         */
 845        path_len -= strlen(".idx");
 846        if (path_len < 1) {
 847                free(p);
 848                return NULL;
 849        }
 850        memcpy(p->pack_name, path, path_len);
 851
 852        strcpy(p->pack_name + path_len, ".keep");
 853        if (!access(p->pack_name, F_OK))
 854                p->pack_keep = 1;
 855
 856        strcpy(p->pack_name + path_len, ".pack");
 857        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 858                free(p);
 859                return NULL;
 860        }
 861
 862        /* ok, it looks sane as far as we can check without
 863         * actually mapping the pack file.
 864         */
 865        p->pack_size = st.st_size;
 866        p->pack_local = local;
 867        p->mtime = st.st_mtime;
 868        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 869                hashclr(p->sha1);
 870        return p;
 871}
 872
 873struct packed_git *parse_pack_index(unsigned char *sha1)
 874{
 875        const char *idx_path = sha1_pack_index_name(sha1);
 876        const char *path = sha1_pack_name(sha1);
 877        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 878
 879        strcpy(p->pack_name, path);
 880        hashcpy(p->sha1, sha1);
 881        if (check_packed_git_idx(idx_path, p)) {
 882                free(p);
 883                return NULL;
 884        }
 885
 886        return p;
 887}
 888
 889void install_packed_git(struct packed_git *pack)
 890{
 891        pack->next = packed_git;
 892        packed_git = pack;
 893}
 894
 895static void prepare_packed_git_one(char *objdir, int local)
 896{
 897        /* Ensure that this buffer is large enough so that we can
 898           append "/pack/" without clobbering the stack even if
 899           strlen(objdir) were PATH_MAX.  */
 900        char path[PATH_MAX + 1 + 4 + 1 + 1];
 901        int len;
 902        DIR *dir;
 903        struct dirent *de;
 904
 905        sprintf(path, "%s/pack", objdir);
 906        len = strlen(path);
 907        dir = opendir(path);
 908        if (!dir) {
 909                if (errno != ENOENT)
 910                        error("unable to open object pack directory: %s: %s",
 911                              path, strerror(errno));
 912                return;
 913        }
 914        path[len++] = '/';
 915        while ((de = readdir(dir)) != NULL) {
 916                int namelen = strlen(de->d_name);
 917                struct packed_git *p;
 918
 919                if (!has_extension(de->d_name, ".idx"))
 920                        continue;
 921
 922                if (len + namelen + 1 > sizeof(path))
 923                        continue;
 924
 925                /* Don't reopen a pack we already have. */
 926                strcpy(path + len, de->d_name);
 927                for (p = packed_git; p; p = p->next) {
 928                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 929                                break;
 930                }
 931                if (p)
 932                        continue;
 933                /* See if it really is a valid .idx file with corresponding
 934                 * .pack file that we can map.
 935                 */
 936                p = add_packed_git(path, len + namelen, local);
 937                if (!p)
 938                        continue;
 939                install_packed_git(p);
 940        }
 941        closedir(dir);
 942}
 943
 944static int sort_pack(const void *a_, const void *b_)
 945{
 946        struct packed_git *a = *((struct packed_git **)a_);
 947        struct packed_git *b = *((struct packed_git **)b_);
 948        int st;
 949
 950        /*
 951         * Local packs tend to contain objects specific to our
 952         * variant of the project than remote ones.  In addition,
 953         * remote ones could be on a network mounted filesystem.
 954         * Favor local ones for these reasons.
 955         */
 956        st = a->pack_local - b->pack_local;
 957        if (st)
 958                return -st;
 959
 960        /*
 961         * Younger packs tend to contain more recent objects,
 962         * and more recent objects tend to get accessed more
 963         * often.
 964         */
 965        if (a->mtime < b->mtime)
 966                return 1;
 967        else if (a->mtime == b->mtime)
 968                return 0;
 969        return -1;
 970}
 971
 972static void rearrange_packed_git(void)
 973{
 974        struct packed_git **ary, *p;
 975        int i, n;
 976
 977        for (n = 0, p = packed_git; p; p = p->next)
 978                n++;
 979        if (n < 2)
 980                return;
 981
 982        /* prepare an array of packed_git for easier sorting */
 983        ary = xcalloc(n, sizeof(struct packed_git *));
 984        for (n = 0, p = packed_git; p; p = p->next)
 985                ary[n++] = p;
 986
 987        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 988
 989        /* link them back again */
 990        for (i = 0; i < n - 1; i++)
 991                ary[i]->next = ary[i + 1];
 992        ary[n - 1]->next = NULL;
 993        packed_git = ary[0];
 994
 995        free(ary);
 996}
 997
 998static int prepare_packed_git_run_once = 0;
 999void prepare_packed_git(void)
1000{
1001        struct alternate_object_database *alt;
1002
1003        if (prepare_packed_git_run_once)
1004                return;
1005        prepare_packed_git_one(get_object_directory(), 1);
1006        prepare_alt_odb();
1007        for (alt = alt_odb_list; alt; alt = alt->next) {
1008                alt->name[-1] = 0;
1009                prepare_packed_git_one(alt->base, 0);
1010                alt->name[-1] = '/';
1011        }
1012        rearrange_packed_git();
1013        prepare_packed_git_run_once = 1;
1014}
1015
1016void reprepare_packed_git(void)
1017{
1018        discard_revindex();
1019        prepare_packed_git_run_once = 0;
1020        prepare_packed_git();
1021}
1022
1023static void mark_bad_packed_object(struct packed_git *p,
1024                                   const unsigned char *sha1)
1025{
1026        unsigned i;
1027        for (i = 0; i < p->num_bad_objects; i++)
1028                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1029                        return;
1030        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1031        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1032        p->num_bad_objects++;
1033}
1034
1035static int has_packed_and_bad(const unsigned char *sha1)
1036{
1037        struct packed_git *p;
1038        unsigned i;
1039
1040        for (p = packed_git; p; p = p->next)
1041                for (i = 0; i < p->num_bad_objects; i++)
1042                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1043                                return 1;
1044        return 0;
1045}
1046
1047int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1048{
1049        unsigned char real_sha1[20];
1050        hash_sha1_file(map, size, type, real_sha1);
1051        return hashcmp(sha1, real_sha1) ? -1 : 0;
1052}
1053
1054static int git_open_noatime(const char *name)
1055{
1056        static int sha1_file_open_flag = O_NOATIME;
1057        int fd = open(name, O_RDONLY | sha1_file_open_flag);
1058
1059        /* Might the failure be due to O_NOATIME? */
1060        if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1061                fd = open(name, O_RDONLY);
1062                if (fd >= 0)
1063                        sha1_file_open_flag = 0;
1064        }
1065        return fd;
1066}
1067
1068static int open_sha1_file(const unsigned char *sha1)
1069{
1070        int fd;
1071        char *name = sha1_file_name(sha1);
1072        struct alternate_object_database *alt;
1073
1074        fd = git_open_noatime(name);
1075        if (fd >= 0)
1076                return fd;
1077
1078        prepare_alt_odb();
1079        errno = ENOENT;
1080        for (alt = alt_odb_list; alt; alt = alt->next) {
1081                name = alt->name;
1082                fill_sha1_path(name, sha1);
1083                fd = git_open_noatime(alt->base);
1084                if (fd >= 0)
1085                        return fd;
1086        }
1087        return -1;
1088}
1089
1090static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1091{
1092        void *map;
1093        int fd;
1094
1095        fd = open_sha1_file(sha1);
1096        map = NULL;
1097        if (fd >= 0) {
1098                struct stat st;
1099
1100                if (!fstat(fd, &st)) {
1101                        *size = xsize_t(st.st_size);
1102                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1103                }
1104                close(fd);
1105        }
1106        return map;
1107}
1108
1109static int legacy_loose_object(unsigned char *map)
1110{
1111        unsigned int word;
1112
1113        /*
1114         * Is it a zlib-compressed buffer? If so, the first byte
1115         * must be 0x78 (15-bit window size, deflated), and the
1116         * first 16-bit word is evenly divisible by 31
1117         */
1118        word = (map[0] << 8) + map[1];
1119        if (map[0] == 0x78 && !(word % 31))
1120                return 1;
1121        else
1122                return 0;
1123}
1124
1125unsigned long unpack_object_header_buffer(const unsigned char *buf,
1126                unsigned long len, enum object_type *type, unsigned long *sizep)
1127{
1128        unsigned shift;
1129        unsigned char c;
1130        unsigned long size;
1131        unsigned long used = 0;
1132
1133        c = buf[used++];
1134        *type = (c >> 4) & 7;
1135        size = c & 15;
1136        shift = 4;
1137        while (c & 0x80) {
1138                if (len <= used || sizeof(long) * 8 <= shift) {
1139                        error("bad object header");
1140                        return 0;
1141                }
1142                c = buf[used++];
1143                size += (c & 0x7f) << shift;
1144                shift += 7;
1145        }
1146        *sizep = size;
1147        return used;
1148}
1149
1150static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1151{
1152        unsigned long size, used;
1153        static const char valid_loose_object_type[8] = {
1154                0, /* OBJ_EXT */
1155                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1156                0, /* "delta" and others are invalid in a loose object */
1157        };
1158        enum object_type type;
1159
1160        /* Get the data stream */
1161        memset(stream, 0, sizeof(*stream));
1162        stream->next_in = map;
1163        stream->avail_in = mapsize;
1164        stream->next_out = buffer;
1165        stream->avail_out = bufsiz;
1166
1167        if (legacy_loose_object(map)) {
1168                inflateInit(stream);
1169                return inflate(stream, 0);
1170        }
1171
1172
1173        /*
1174         * There used to be a second loose object header format which
1175         * was meant to mimic the in-pack format, allowing for direct
1176         * copy of the object data.  This format turned up not to be
1177         * really worth it and we don't write it any longer.  But we
1178         * can still read it.
1179         */
1180        used = unpack_object_header_buffer(map, mapsize, &type, &size);
1181        if (!used || !valid_loose_object_type[type])
1182                return -1;
1183        map += used;
1184        mapsize -= used;
1185
1186        /* Set up the stream for the rest.. */
1187        stream->next_in = map;
1188        stream->avail_in = mapsize;
1189        inflateInit(stream);
1190
1191        /* And generate the fake traditional header */
1192        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1193                                         typename(type), size);
1194        return 0;
1195}
1196
1197static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1198{
1199        int bytes = strlen(buffer) + 1;
1200        unsigned char *buf = xmalloc(1+size);
1201        unsigned long n;
1202        int status = Z_OK;
1203
1204        n = stream->total_out - bytes;
1205        if (n > size)
1206                n = size;
1207        memcpy(buf, (char *) buffer + bytes, n);
1208        bytes = n;
1209        if (bytes <= size) {
1210                /*
1211                 * The above condition must be (bytes <= size), not
1212                 * (bytes < size).  In other words, even though we
1213                 * expect no more output and set avail_out to zer0,
1214                 * the input zlib stream may have bytes that express
1215                 * "this concludes the stream", and we *do* want to
1216                 * eat that input.
1217                 *
1218                 * Otherwise we would not be able to test that we
1219                 * consumed all the input to reach the expected size;
1220                 * we also want to check that zlib tells us that all
1221                 * went well with status == Z_STREAM_END at the end.
1222                 */
1223                stream->next_out = buf + bytes;
1224                stream->avail_out = size - bytes;
1225                while (status == Z_OK)
1226                        status = inflate(stream, Z_FINISH);
1227        }
1228        buf[size] = 0;
1229        if (status == Z_STREAM_END && !stream->avail_in) {
1230                inflateEnd(stream);
1231                return buf;
1232        }
1233
1234        if (status < 0)
1235                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1236        else if (stream->avail_in)
1237                error("garbage at end of loose object '%s'",
1238                      sha1_to_hex(sha1));
1239        free(buf);
1240        return NULL;
1241}
1242
1243/*
1244 * We used to just use "sscanf()", but that's actually way
1245 * too permissive for what we want to check. So do an anal
1246 * object header parse by hand.
1247 */
1248static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1249{
1250        char type[10];
1251        int i;
1252        unsigned long size;
1253
1254        /*
1255         * The type can be at most ten bytes (including the
1256         * terminating '\0' that we add), and is followed by
1257         * a space.
1258         */
1259        i = 0;
1260        for (;;) {
1261                char c = *hdr++;
1262                if (c == ' ')
1263                        break;
1264                type[i++] = c;
1265                if (i >= sizeof(type))
1266                        return -1;
1267        }
1268        type[i] = 0;
1269
1270        /*
1271         * The length must follow immediately, and be in canonical
1272         * decimal format (ie "010" is not valid).
1273         */
1274        size = *hdr++ - '0';
1275        if (size > 9)
1276                return -1;
1277        if (size) {
1278                for (;;) {
1279                        unsigned long c = *hdr - '0';
1280                        if (c > 9)
1281                                break;
1282                        hdr++;
1283                        size = size * 10 + c;
1284                }
1285        }
1286        *sizep = size;
1287
1288        /*
1289         * The length must be followed by a zero byte
1290         */
1291        return *hdr ? -1 : type_from_string(type);
1292}
1293
1294static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1295{
1296        int ret;
1297        z_stream stream;
1298        char hdr[8192];
1299
1300        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1301        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1302                return NULL;
1303
1304        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1305}
1306
1307unsigned long get_size_from_delta(struct packed_git *p,
1308                                  struct pack_window **w_curs,
1309                                  off_t curpos)
1310{
1311        const unsigned char *data;
1312        unsigned char delta_head[20], *in;
1313        z_stream stream;
1314        int st;
1315
1316        memset(&stream, 0, sizeof(stream));
1317        stream.next_out = delta_head;
1318        stream.avail_out = sizeof(delta_head);
1319
1320        inflateInit(&stream);
1321        do {
1322                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1323                stream.next_in = in;
1324                st = inflate(&stream, Z_FINISH);
1325                curpos += stream.next_in - in;
1326        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1327                 stream.total_out < sizeof(delta_head));
1328        inflateEnd(&stream);
1329        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1330                error("delta data unpack-initial failed");
1331                return 0;
1332        }
1333
1334        /* Examine the initial part of the delta to figure out
1335         * the result size.
1336         */
1337        data = delta_head;
1338
1339        /* ignore base size */
1340        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1341
1342        /* Read the result size */
1343        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1344}
1345
1346static off_t get_delta_base(struct packed_git *p,
1347                                    struct pack_window **w_curs,
1348                                    off_t *curpos,
1349                                    enum object_type type,
1350                                    off_t delta_obj_offset)
1351{
1352        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1353        off_t base_offset;
1354
1355        /* use_pack() assured us we have [base_info, base_info + 20)
1356         * as a range that we can look at without walking off the
1357         * end of the mapped window.  Its actually the hash size
1358         * that is assured.  An OFS_DELTA longer than the hash size
1359         * is stupid, as then a REF_DELTA would be smaller to store.
1360         */
1361        if (type == OBJ_OFS_DELTA) {
1362                unsigned used = 0;
1363                unsigned char c = base_info[used++];
1364                base_offset = c & 127;
1365                while (c & 128) {
1366                        base_offset += 1;
1367                        if (!base_offset || MSB(base_offset, 7))
1368                                return 0;  /* overflow */
1369                        c = base_info[used++];
1370                        base_offset = (base_offset << 7) + (c & 127);
1371                }
1372                base_offset = delta_obj_offset - base_offset;
1373                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1374                        return 0;  /* out of bound */
1375                *curpos += used;
1376        } else if (type == OBJ_REF_DELTA) {
1377                /* The base entry _must_ be in the same pack */
1378                base_offset = find_pack_entry_one(base_info, p);
1379                *curpos += 20;
1380        } else
1381                die("I am totally screwed");
1382        return base_offset;
1383}
1384
1385/* forward declaration for a mutually recursive function */
1386static int packed_object_info(struct packed_git *p, off_t offset,
1387                              unsigned long *sizep);
1388
1389static int packed_delta_info(struct packed_git *p,
1390                             struct pack_window **w_curs,
1391                             off_t curpos,
1392                             enum object_type type,
1393                             off_t obj_offset,
1394                             unsigned long *sizep)
1395{
1396        off_t base_offset;
1397
1398        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1399        if (!base_offset)
1400                return OBJ_BAD;
1401        type = packed_object_info(p, base_offset, NULL);
1402        if (type <= OBJ_NONE) {
1403                struct revindex_entry *revidx;
1404                const unsigned char *base_sha1;
1405                revidx = find_pack_revindex(p, base_offset);
1406                if (!revidx)
1407                        return OBJ_BAD;
1408                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1409                mark_bad_packed_object(p, base_sha1);
1410                type = sha1_object_info(base_sha1, NULL);
1411                if (type <= OBJ_NONE)
1412                        return OBJ_BAD;
1413        }
1414
1415        /* We choose to only get the type of the base object and
1416         * ignore potentially corrupt pack file that expects the delta
1417         * based on a base with a wrong size.  This saves tons of
1418         * inflate() calls.
1419         */
1420        if (sizep) {
1421                *sizep = get_size_from_delta(p, w_curs, curpos);
1422                if (*sizep == 0)
1423                        type = OBJ_BAD;
1424        }
1425
1426        return type;
1427}
1428
1429static int unpack_object_header(struct packed_git *p,
1430                                struct pack_window **w_curs,
1431                                off_t *curpos,
1432                                unsigned long *sizep)
1433{
1434        unsigned char *base;
1435        unsigned int left;
1436        unsigned long used;
1437        enum object_type type;
1438
1439        /* use_pack() assures us we have [base, base + 20) available
1440         * as a range that we can look at at.  (Its actually the hash
1441         * size that is assured.)  With our object header encoding
1442         * the maximum deflated object size is 2^137, which is just
1443         * insane, so we know won't exceed what we have been given.
1444         */
1445        base = use_pack(p, w_curs, *curpos, &left);
1446        used = unpack_object_header_buffer(base, left, &type, sizep);
1447        if (!used) {
1448                type = OBJ_BAD;
1449        } else
1450                *curpos += used;
1451
1452        return type;
1453}
1454
1455const char *packed_object_info_detail(struct packed_git *p,
1456                                      off_t obj_offset,
1457                                      unsigned long *size,
1458                                      unsigned long *store_size,
1459                                      unsigned int *delta_chain_length,
1460                                      unsigned char *base_sha1)
1461{
1462        struct pack_window *w_curs = NULL;
1463        off_t curpos;
1464        unsigned long dummy;
1465        unsigned char *next_sha1;
1466        enum object_type type;
1467        struct revindex_entry *revidx;
1468
1469        *delta_chain_length = 0;
1470        curpos = obj_offset;
1471        type = unpack_object_header(p, &w_curs, &curpos, size);
1472
1473        revidx = find_pack_revindex(p, obj_offset);
1474        *store_size = revidx[1].offset - obj_offset;
1475
1476        for (;;) {
1477                switch (type) {
1478                default:
1479                        die("pack %s contains unknown object type %d",
1480                            p->pack_name, type);
1481                case OBJ_COMMIT:
1482                case OBJ_TREE:
1483                case OBJ_BLOB:
1484                case OBJ_TAG:
1485                        unuse_pack(&w_curs);
1486                        return typename(type);
1487                case OBJ_OFS_DELTA:
1488                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1489                        if (!obj_offset)
1490                                die("pack %s contains bad delta base reference of type %s",
1491                                    p->pack_name, typename(type));
1492                        if (*delta_chain_length == 0) {
1493                                revidx = find_pack_revindex(p, obj_offset);
1494                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1495                        }
1496                        break;
1497                case OBJ_REF_DELTA:
1498                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1499                        if (*delta_chain_length == 0)
1500                                hashcpy(base_sha1, next_sha1);
1501                        obj_offset = find_pack_entry_one(next_sha1, p);
1502                        break;
1503                }
1504                (*delta_chain_length)++;
1505                curpos = obj_offset;
1506                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1507        }
1508}
1509
1510static int packed_object_info(struct packed_git *p, off_t obj_offset,
1511                              unsigned long *sizep)
1512{
1513        struct pack_window *w_curs = NULL;
1514        unsigned long size;
1515        off_t curpos = obj_offset;
1516        enum object_type type;
1517
1518        type = unpack_object_header(p, &w_curs, &curpos, &size);
1519
1520        switch (type) {
1521        case OBJ_OFS_DELTA:
1522        case OBJ_REF_DELTA:
1523                type = packed_delta_info(p, &w_curs, curpos,
1524                                         type, obj_offset, sizep);
1525                break;
1526        case OBJ_COMMIT:
1527        case OBJ_TREE:
1528        case OBJ_BLOB:
1529        case OBJ_TAG:
1530                if (sizep)
1531                        *sizep = size;
1532                break;
1533        default:
1534                error("unknown object type %i at offset %"PRIuMAX" in %s",
1535                      type, (uintmax_t)obj_offset, p->pack_name);
1536                type = OBJ_BAD;
1537        }
1538        unuse_pack(&w_curs);
1539        return type;
1540}
1541
1542static void *unpack_compressed_entry(struct packed_git *p,
1543                                    struct pack_window **w_curs,
1544                                    off_t curpos,
1545                                    unsigned long size)
1546{
1547        int st;
1548        z_stream stream;
1549        unsigned char *buffer, *in;
1550
1551        buffer = xmalloc(size + 1);
1552        buffer[size] = 0;
1553        memset(&stream, 0, sizeof(stream));
1554        stream.next_out = buffer;
1555        stream.avail_out = size;
1556
1557        inflateInit(&stream);
1558        do {
1559                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1560                stream.next_in = in;
1561                st = inflate(&stream, Z_FINISH);
1562                curpos += stream.next_in - in;
1563        } while (st == Z_OK || st == Z_BUF_ERROR);
1564        inflateEnd(&stream);
1565        if ((st != Z_STREAM_END) || stream.total_out != size) {
1566                free(buffer);
1567                return NULL;
1568        }
1569
1570        return buffer;
1571}
1572
1573#define MAX_DELTA_CACHE (256)
1574
1575static size_t delta_base_cached;
1576
1577static struct delta_base_cache_lru_list {
1578        struct delta_base_cache_lru_list *prev;
1579        struct delta_base_cache_lru_list *next;
1580} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1581
1582static struct delta_base_cache_entry {
1583        struct delta_base_cache_lru_list lru;
1584        void *data;
1585        struct packed_git *p;
1586        off_t base_offset;
1587        unsigned long size;
1588        enum object_type type;
1589} delta_base_cache[MAX_DELTA_CACHE];
1590
1591static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1592{
1593        unsigned long hash;
1594
1595        hash = (unsigned long)p + (unsigned long)base_offset;
1596        hash += (hash >> 8) + (hash >> 16);
1597        return hash % MAX_DELTA_CACHE;
1598}
1599
1600static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1601        unsigned long *base_size, enum object_type *type, int keep_cache)
1602{
1603        void *ret;
1604        unsigned long hash = pack_entry_hash(p, base_offset);
1605        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1606
1607        ret = ent->data;
1608        if (!ret || ent->p != p || ent->base_offset != base_offset)
1609                return unpack_entry(p, base_offset, type, base_size);
1610
1611        if (!keep_cache) {
1612                ent->data = NULL;
1613                ent->lru.next->prev = ent->lru.prev;
1614                ent->lru.prev->next = ent->lru.next;
1615                delta_base_cached -= ent->size;
1616        } else {
1617                ret = xmemdupz(ent->data, ent->size);
1618        }
1619        *type = ent->type;
1620        *base_size = ent->size;
1621        return ret;
1622}
1623
1624static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1625{
1626        if (ent->data) {
1627                free(ent->data);
1628                ent->data = NULL;
1629                ent->lru.next->prev = ent->lru.prev;
1630                ent->lru.prev->next = ent->lru.next;
1631                delta_base_cached -= ent->size;
1632        }
1633}
1634
1635static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1636        void *base, unsigned long base_size, enum object_type type)
1637{
1638        unsigned long hash = pack_entry_hash(p, base_offset);
1639        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1640        struct delta_base_cache_lru_list *lru;
1641
1642        release_delta_base_cache(ent);
1643        delta_base_cached += base_size;
1644
1645        for (lru = delta_base_cache_lru.next;
1646             delta_base_cached > delta_base_cache_limit
1647             && lru != &delta_base_cache_lru;
1648             lru = lru->next) {
1649                struct delta_base_cache_entry *f = (void *)lru;
1650                if (f->type == OBJ_BLOB)
1651                        release_delta_base_cache(f);
1652        }
1653        for (lru = delta_base_cache_lru.next;
1654             delta_base_cached > delta_base_cache_limit
1655             && lru != &delta_base_cache_lru;
1656             lru = lru->next) {
1657                struct delta_base_cache_entry *f = (void *)lru;
1658                release_delta_base_cache(f);
1659        }
1660
1661        ent->p = p;
1662        ent->base_offset = base_offset;
1663        ent->type = type;
1664        ent->data = base;
1665        ent->size = base_size;
1666        ent->lru.next = &delta_base_cache_lru;
1667        ent->lru.prev = delta_base_cache_lru.prev;
1668        delta_base_cache_lru.prev->next = &ent->lru;
1669        delta_base_cache_lru.prev = &ent->lru;
1670}
1671
1672static void *unpack_delta_entry(struct packed_git *p,
1673                                struct pack_window **w_curs,
1674                                off_t curpos,
1675                                unsigned long delta_size,
1676                                off_t obj_offset,
1677                                enum object_type *type,
1678                                unsigned long *sizep)
1679{
1680        void *delta_data, *result, *base;
1681        unsigned long base_size;
1682        off_t base_offset;
1683
1684        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1685        if (!base_offset) {
1686                error("failed to validate delta base reference "
1687                      "at offset %"PRIuMAX" from %s",
1688                      (uintmax_t)curpos, p->pack_name);
1689                return NULL;
1690        }
1691        unuse_pack(w_curs);
1692        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1693        if (!base) {
1694                /*
1695                 * We're probably in deep shit, but let's try to fetch
1696                 * the required base anyway from another pack or loose.
1697                 * This is costly but should happen only in the presence
1698                 * of a corrupted pack, and is better than failing outright.
1699                 */
1700                struct revindex_entry *revidx;
1701                const unsigned char *base_sha1;
1702                revidx = find_pack_revindex(p, base_offset);
1703                if (!revidx)
1704                        return NULL;
1705                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1706                error("failed to read delta base object %s"
1707                      " at offset %"PRIuMAX" from %s",
1708                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1709                      p->pack_name);
1710                mark_bad_packed_object(p, base_sha1);
1711                base = read_object(base_sha1, type, &base_size);
1712                if (!base)
1713                        return NULL;
1714        }
1715
1716        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1717        if (!delta_data) {
1718                error("failed to unpack compressed delta "
1719                      "at offset %"PRIuMAX" from %s",
1720                      (uintmax_t)curpos, p->pack_name);
1721                free(base);
1722                return NULL;
1723        }
1724        result = patch_delta(base, base_size,
1725                             delta_data, delta_size,
1726                             sizep);
1727        if (!result)
1728                die("failed to apply delta");
1729        free(delta_data);
1730        add_delta_base_cache(p, base_offset, base, base_size, *type);
1731        return result;
1732}
1733
1734int do_check_packed_object_crc;
1735
1736void *unpack_entry(struct packed_git *p, off_t obj_offset,
1737                   enum object_type *type, unsigned long *sizep)
1738{
1739        struct pack_window *w_curs = NULL;
1740        off_t curpos = obj_offset;
1741        void *data;
1742
1743        if (do_check_packed_object_crc && p->index_version > 1) {
1744                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1745                unsigned long len = revidx[1].offset - obj_offset;
1746                if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1747                        const unsigned char *sha1 =
1748                                nth_packed_object_sha1(p, revidx->nr);
1749                        error("bad packed object CRC for %s",
1750                              sha1_to_hex(sha1));
1751                        mark_bad_packed_object(p, sha1);
1752                        return NULL;
1753                }
1754        }
1755
1756        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1757        switch (*type) {
1758        case OBJ_OFS_DELTA:
1759        case OBJ_REF_DELTA:
1760                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1761                                          obj_offset, type, sizep);
1762                break;
1763        case OBJ_COMMIT:
1764        case OBJ_TREE:
1765        case OBJ_BLOB:
1766        case OBJ_TAG:
1767                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1768                break;
1769        default:
1770                data = NULL;
1771                error("unknown object type %i at offset %"PRIuMAX" in %s",
1772                      *type, (uintmax_t)obj_offset, p->pack_name);
1773        }
1774        unuse_pack(&w_curs);
1775        return data;
1776}
1777
1778const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1779                                            uint32_t n)
1780{
1781        const unsigned char *index = p->index_data;
1782        if (!index) {
1783                if (open_pack_index(p))
1784                        return NULL;
1785                index = p->index_data;
1786        }
1787        if (n >= p->num_objects)
1788                return NULL;
1789        index += 4 * 256;
1790        if (p->index_version == 1) {
1791                return index + 24 * n + 4;
1792        } else {
1793                index += 8;
1794                return index + 20 * n;
1795        }
1796}
1797
1798off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1799{
1800        const unsigned char *index = p->index_data;
1801        index += 4 * 256;
1802        if (p->index_version == 1) {
1803                return ntohl(*((uint32_t *)(index + 24 * n)));
1804        } else {
1805                uint32_t off;
1806                index += 8 + p->num_objects * (20 + 4);
1807                off = ntohl(*((uint32_t *)(index + 4 * n)));
1808                if (!(off & 0x80000000))
1809                        return off;
1810                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1811                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1812                                   ntohl(*((uint32_t *)(index + 4)));
1813        }
1814}
1815
1816off_t find_pack_entry_one(const unsigned char *sha1,
1817                                  struct packed_git *p)
1818{
1819        const uint32_t *level1_ofs = p->index_data;
1820        const unsigned char *index = p->index_data;
1821        unsigned hi, lo, stride;
1822        static int use_lookup = -1;
1823        static int debug_lookup = -1;
1824
1825        if (debug_lookup < 0)
1826                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1827
1828        if (!index) {
1829                if (open_pack_index(p))
1830                        return 0;
1831                level1_ofs = p->index_data;
1832                index = p->index_data;
1833        }
1834        if (p->index_version > 1) {
1835                level1_ofs += 2;
1836                index += 8;
1837        }
1838        index += 4 * 256;
1839        hi = ntohl(level1_ofs[*sha1]);
1840        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1841        if (p->index_version > 1) {
1842                stride = 20;
1843        } else {
1844                stride = 24;
1845                index += 4;
1846        }
1847
1848        if (debug_lookup)
1849                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1850                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1851
1852        if (use_lookup < 0)
1853                use_lookup = !!getenv("GIT_USE_LOOKUP");
1854        if (use_lookup) {
1855                int pos = sha1_entry_pos(index, stride, 0,
1856                                         lo, hi, p->num_objects, sha1);
1857                if (pos < 0)
1858                        return 0;
1859                return nth_packed_object_offset(p, pos);
1860        }
1861
1862        do {
1863                unsigned mi = (lo + hi) / 2;
1864                int cmp = hashcmp(index + mi * stride, sha1);
1865
1866                if (debug_lookup)
1867                        printf("lo %u hi %u rg %u mi %u\n",
1868                               lo, hi, hi - lo, mi);
1869                if (!cmp)
1870                        return nth_packed_object_offset(p, mi);
1871                if (cmp > 0)
1872                        hi = mi;
1873                else
1874                        lo = mi+1;
1875        } while (lo < hi);
1876        return 0;
1877}
1878
1879int matches_pack_name(struct packed_git *p, const char *name)
1880{
1881        const char *last_c, *c;
1882
1883        if (!strcmp(p->pack_name, name))
1884                return 1;
1885
1886        for (c = p->pack_name, last_c = c; *c;)
1887                if (*c == '/')
1888                        last_c = ++c;
1889                else
1890                        ++c;
1891        if (!strcmp(last_c, name))
1892                return 1;
1893
1894        return 0;
1895}
1896
1897static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1898{
1899        static struct packed_git *last_found = (void *)1;
1900        struct packed_git *p;
1901        off_t offset;
1902
1903        prepare_packed_git();
1904        if (!packed_git)
1905                return 0;
1906        p = (last_found == (void *)1) ? packed_git : last_found;
1907
1908        do {
1909                if (ignore_packed) {
1910                        const char **ig;
1911                        for (ig = ignore_packed; *ig; ig++)
1912                                if (matches_pack_name(p, *ig))
1913                                        break;
1914                        if (*ig)
1915                                goto next;
1916                }
1917
1918                if (p->num_bad_objects) {
1919                        unsigned i;
1920                        for (i = 0; i < p->num_bad_objects; i++)
1921                                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1922                                        goto next;
1923                }
1924
1925                offset = find_pack_entry_one(sha1, p);
1926                if (offset) {
1927                        /*
1928                         * We are about to tell the caller where they can
1929                         * locate the requested object.  We better make
1930                         * sure the packfile is still here and can be
1931                         * accessed before supplying that answer, as
1932                         * it may have been deleted since the index
1933                         * was loaded!
1934                         */
1935                        if (p->pack_fd == -1 && open_packed_git(p)) {
1936                                error("packfile %s cannot be accessed", p->pack_name);
1937                                goto next;
1938                        }
1939                        e->offset = offset;
1940                        e->p = p;
1941                        hashcpy(e->sha1, sha1);
1942                        last_found = p;
1943                        return 1;
1944                }
1945
1946                next:
1947                if (p == last_found)
1948                        p = packed_git;
1949                else
1950                        p = p->next;
1951                if (p == last_found)
1952                        p = p->next;
1953        } while (p);
1954        return 0;
1955}
1956
1957struct packed_git *find_sha1_pack(const unsigned char *sha1,
1958                                  struct packed_git *packs)
1959{
1960        struct packed_git *p;
1961
1962        for (p = packs; p; p = p->next) {
1963                if (find_pack_entry_one(sha1, p))
1964                        return p;
1965        }
1966        return NULL;
1967
1968}
1969
1970static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1971{
1972        int status;
1973        unsigned long mapsize, size;
1974        void *map;
1975        z_stream stream;
1976        char hdr[32];
1977
1978        map = map_sha1_file(sha1, &mapsize);
1979        if (!map)
1980                return error("unable to find %s", sha1_to_hex(sha1));
1981        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1982                status = error("unable to unpack %s header",
1983                               sha1_to_hex(sha1));
1984        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1985                status = error("unable to parse %s header", sha1_to_hex(sha1));
1986        else if (sizep)
1987                *sizep = size;
1988        inflateEnd(&stream);
1989        munmap(map, mapsize);
1990        return status;
1991}
1992
1993int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1994{
1995        struct pack_entry e;
1996        int status;
1997
1998        if (!find_pack_entry(sha1, &e, NULL)) {
1999                /* Most likely it's a loose object. */
2000                status = sha1_loose_object_info(sha1, sizep);
2001                if (status >= 0)
2002                        return status;
2003
2004                /* Not a loose object; someone else may have just packed it. */
2005                reprepare_packed_git();
2006                if (!find_pack_entry(sha1, &e, NULL))
2007                        return status;
2008        }
2009
2010        status = packed_object_info(e.p, e.offset, sizep);
2011        if (status < 0) {
2012                mark_bad_packed_object(e.p, sha1);
2013                status = sha1_object_info(sha1, sizep);
2014        }
2015
2016        return status;
2017}
2018
2019static void *read_packed_sha1(const unsigned char *sha1,
2020                              enum object_type *type, unsigned long *size)
2021{
2022        struct pack_entry e;
2023        void *data;
2024
2025        if (!find_pack_entry(sha1, &e, NULL))
2026                return NULL;
2027        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2028        if (!data) {
2029                /*
2030                 * We're probably in deep shit, but let's try to fetch
2031                 * the required object anyway from another pack or loose.
2032                 * This should happen only in the presence of a corrupted
2033                 * pack, and is better than failing outright.
2034                 */
2035                error("failed to read object %s at offset %"PRIuMAX" from %s",
2036                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2037                mark_bad_packed_object(e.p, sha1);
2038                data = read_object(sha1, type, size);
2039        }
2040        return data;
2041}
2042
2043/*
2044 * This is meant to hold a *small* number of objects that you would
2045 * want read_sha1_file() to be able to return, but yet you do not want
2046 * to write them into the object store (e.g. a browse-only
2047 * application).
2048 */
2049static struct cached_object {
2050        unsigned char sha1[20];
2051        enum object_type type;
2052        void *buf;
2053        unsigned long size;
2054} *cached_objects;
2055static int cached_object_nr, cached_object_alloc;
2056
2057static struct cached_object empty_tree = {
2058        /* empty tree sha1: 4b825dc642cb6eb9a060e54bf8d69288fbee4904 */
2059        "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60"
2060        "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04",
2061        OBJ_TREE,
2062        "",
2063        0
2064};
2065
2066static struct cached_object *find_cached_object(const unsigned char *sha1)
2067{
2068        int i;
2069        struct cached_object *co = cached_objects;
2070
2071        for (i = 0; i < cached_object_nr; i++, co++) {
2072                if (!hashcmp(co->sha1, sha1))
2073                        return co;
2074        }
2075        if (!hashcmp(sha1, empty_tree.sha1))
2076                return &empty_tree;
2077        return NULL;
2078}
2079
2080int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2081                      unsigned char *sha1)
2082{
2083        struct cached_object *co;
2084
2085        hash_sha1_file(buf, len, typename(type), sha1);
2086        if (has_sha1_file(sha1) || find_cached_object(sha1))
2087                return 0;
2088        if (cached_object_alloc <= cached_object_nr) {
2089                cached_object_alloc = alloc_nr(cached_object_alloc);
2090                cached_objects = xrealloc(cached_objects,
2091                                          sizeof(*cached_objects) *
2092                                          cached_object_alloc);
2093        }
2094        co = &cached_objects[cached_object_nr++];
2095        co->size = len;
2096        co->type = type;
2097        co->buf = xmalloc(len);
2098        memcpy(co->buf, buf, len);
2099        hashcpy(co->sha1, sha1);
2100        return 0;
2101}
2102
2103void *read_object(const unsigned char *sha1, enum object_type *type,
2104                  unsigned long *size)
2105{
2106        unsigned long mapsize;
2107        void *map, *buf;
2108        struct cached_object *co;
2109
2110        co = find_cached_object(sha1);
2111        if (co) {
2112                *type = co->type;
2113                *size = co->size;
2114                return xmemdupz(co->buf, co->size);
2115        }
2116
2117        buf = read_packed_sha1(sha1, type, size);
2118        if (buf)
2119                return buf;
2120        map = map_sha1_file(sha1, &mapsize);
2121        if (map) {
2122                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2123                munmap(map, mapsize);
2124                return buf;
2125        }
2126        reprepare_packed_git();
2127        return read_packed_sha1(sha1, type, size);
2128}
2129
2130void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2131                     unsigned long *size)
2132{
2133        void *data = read_object(sha1, type, size);
2134        /* legacy behavior is to die on corrupted objects */
2135        if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2136                die("object %s is corrupted", sha1_to_hex(sha1));
2137        return data;
2138}
2139
2140void *read_object_with_reference(const unsigned char *sha1,
2141                                 const char *required_type_name,
2142                                 unsigned long *size,
2143                                 unsigned char *actual_sha1_return)
2144{
2145        enum object_type type, required_type;
2146        void *buffer;
2147        unsigned long isize;
2148        unsigned char actual_sha1[20];
2149
2150        required_type = type_from_string(required_type_name);
2151        hashcpy(actual_sha1, sha1);
2152        while (1) {
2153                int ref_length = -1;
2154                const char *ref_type = NULL;
2155
2156                buffer = read_sha1_file(actual_sha1, &type, &isize);
2157                if (!buffer)
2158                        return NULL;
2159                if (type == required_type) {
2160                        *size = isize;
2161                        if (actual_sha1_return)
2162                                hashcpy(actual_sha1_return, actual_sha1);
2163                        return buffer;
2164                }
2165                /* Handle references */
2166                else if (type == OBJ_COMMIT)
2167                        ref_type = "tree ";
2168                else if (type == OBJ_TAG)
2169                        ref_type = "object ";
2170                else {
2171                        free(buffer);
2172                        return NULL;
2173                }
2174                ref_length = strlen(ref_type);
2175
2176                if (ref_length + 40 > isize ||
2177                    memcmp(buffer, ref_type, ref_length) ||
2178                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2179                        free(buffer);
2180                        return NULL;
2181                }
2182                free(buffer);
2183                /* Now we have the ID of the referred-to object in
2184                 * actual_sha1.  Check again. */
2185        }
2186}
2187
2188static void write_sha1_file_prepare(const void *buf, unsigned long len,
2189                                    const char *type, unsigned char *sha1,
2190                                    char *hdr, int *hdrlen)
2191{
2192        git_SHA_CTX c;
2193
2194        /* Generate the header */
2195        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2196
2197        /* Sha1.. */
2198        git_SHA1_Init(&c);
2199        git_SHA1_Update(&c, hdr, *hdrlen);
2200        git_SHA1_Update(&c, buf, len);
2201        git_SHA1_Final(sha1, &c);
2202}
2203
2204/*
2205 * Move the just written object into its final resting place
2206 */
2207int move_temp_to_file(const char *tmpfile, const char *filename)
2208{
2209        int ret = 0;
2210        if (link(tmpfile, filename))
2211                ret = errno;
2212
2213        /*
2214         * Coda hack - coda doesn't like cross-directory links,
2215         * so we fall back to a rename, which will mean that it
2216         * won't be able to check collisions, but that's not a
2217         * big deal.
2218         *
2219         * The same holds for FAT formatted media.
2220         *
2221         * When this succeeds, we just return 0. We have nothing
2222         * left to unlink.
2223         */
2224        if (ret && ret != EEXIST) {
2225                if (!rename(tmpfile, filename))
2226                        return 0;
2227                ret = errno;
2228        }
2229        unlink(tmpfile);
2230        if (ret) {
2231                if (ret != EEXIST) {
2232                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2233                }
2234                /* FIXME!!! Collision check here ? */
2235        }
2236
2237        return 0;
2238}
2239
2240static int write_buffer(int fd, const void *buf, size_t len)
2241{
2242        if (write_in_full(fd, buf, len) < 0)
2243                return error("file write error (%s)", strerror(errno));
2244        return 0;
2245}
2246
2247int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2248                   unsigned char *sha1)
2249{
2250        char hdr[32];
2251        int hdrlen;
2252        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2253        return 0;
2254}
2255
2256/* Finalize a file on disk, and close it. */
2257static void close_sha1_file(int fd)
2258{
2259        if (fsync_object_files)
2260                fsync_or_die(fd, "sha1 file");
2261        fchmod(fd, 0444);
2262        if (close(fd) != 0)
2263                die("unable to write sha1 file");
2264}
2265
2266/* Size of directory component, including the ending '/' */
2267static inline int directory_size(const char *filename)
2268{
2269        const char *s = strrchr(filename, '/');
2270        if (!s)
2271                return 0;
2272        return s - filename + 1;
2273}
2274
2275/*
2276 * This creates a temporary file in the same directory as the final
2277 * 'filename'
2278 *
2279 * We want to avoid cross-directory filename renames, because those
2280 * can have problems on various filesystems (FAT, NFS, Coda).
2281 */
2282static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2283{
2284        int fd, dirlen = directory_size(filename);
2285
2286        if (dirlen + 20 > bufsiz) {
2287                errno = ENAMETOOLONG;
2288                return -1;
2289        }
2290        memcpy(buffer, filename, dirlen);
2291        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2292        fd = mkstemp(buffer);
2293        if (fd < 0 && dirlen) {
2294                /* Make sure the directory exists */
2295                memcpy(buffer, filename, dirlen);
2296                buffer[dirlen-1] = 0;
2297                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2298                        return -1;
2299
2300                /* Try again */
2301                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2302                fd = mkstemp(buffer);
2303        }
2304        return fd;
2305}
2306
2307static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2308                              void *buf, unsigned long len, time_t mtime)
2309{
2310        int fd, size, ret;
2311        unsigned char *compressed;
2312        z_stream stream;
2313        char *filename;
2314        static char tmpfile[PATH_MAX];
2315
2316        filename = sha1_file_name(sha1);
2317        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2318        if (fd < 0) {
2319                if (errno == EPERM)
2320                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2321                else
2322                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2323        }
2324
2325        /* Set it up */
2326        memset(&stream, 0, sizeof(stream));
2327        deflateInit(&stream, zlib_compression_level);
2328        size = 8 + deflateBound(&stream, len+hdrlen);
2329        compressed = xmalloc(size);
2330
2331        /* Compress it */
2332        stream.next_out = compressed;
2333        stream.avail_out = size;
2334
2335        /* First header.. */
2336        stream.next_in = (unsigned char *)hdr;
2337        stream.avail_in = hdrlen;
2338        while (deflate(&stream, 0) == Z_OK)
2339                /* nothing */;
2340
2341        /* Then the data itself.. */
2342        stream.next_in = buf;
2343        stream.avail_in = len;
2344        ret = deflate(&stream, Z_FINISH);
2345        if (ret != Z_STREAM_END)
2346                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2347
2348        ret = deflateEnd(&stream);
2349        if (ret != Z_OK)
2350                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2351
2352        size = stream.total_out;
2353
2354        if (write_buffer(fd, compressed, size) < 0)
2355                die("unable to write sha1 file");
2356        close_sha1_file(fd);
2357        free(compressed);
2358
2359        if (mtime) {
2360                struct utimbuf utb;
2361                utb.actime = mtime;
2362                utb.modtime = mtime;
2363                if (utime(tmpfile, &utb) < 0)
2364                        warning("failed utime() on %s: %s",
2365                                tmpfile, strerror(errno));
2366        }
2367
2368        return move_temp_to_file(tmpfile, filename);
2369}
2370
2371int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2372{
2373        unsigned char sha1[20];
2374        char hdr[32];
2375        int hdrlen;
2376
2377        /* Normally if we have it in the pack then we do not bother writing
2378         * it out into .git/objects/??/?{38} file.
2379         */
2380        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2381        if (returnsha1)
2382                hashcpy(returnsha1, sha1);
2383        if (has_sha1_file(sha1))
2384                return 0;
2385        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2386}
2387
2388int force_object_loose(const unsigned char *sha1, time_t mtime)
2389{
2390        void *buf;
2391        unsigned long len;
2392        enum object_type type;
2393        char hdr[32];
2394        int hdrlen;
2395        int ret;
2396
2397        if (has_loose_object(sha1))
2398                return 0;
2399        buf = read_packed_sha1(sha1, &type, &len);
2400        if (!buf)
2401                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2402        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2403        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2404        free(buf);
2405
2406        return ret;
2407}
2408
2409int has_pack_index(const unsigned char *sha1)
2410{
2411        struct stat st;
2412        if (stat(sha1_pack_index_name(sha1), &st))
2413                return 0;
2414        return 1;
2415}
2416
2417int has_pack_file(const unsigned char *sha1)
2418{
2419        struct stat st;
2420        if (stat(sha1_pack_name(sha1), &st))
2421                return 0;
2422        return 1;
2423}
2424
2425int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2426{
2427        struct pack_entry e;
2428        return find_pack_entry(sha1, &e, ignore_packed);
2429}
2430
2431int has_sha1_file(const unsigned char *sha1)
2432{
2433        struct pack_entry e;
2434
2435        if (find_pack_entry(sha1, &e, NULL))
2436                return 1;
2437        return has_loose_object(sha1);
2438}
2439
2440static int index_mem(unsigned char *sha1, void *buf, size_t size,
2441                     int write_object, enum object_type type, const char *path)
2442{
2443        int ret, re_allocated = 0;
2444
2445        if (!type)
2446                type = OBJ_BLOB;
2447
2448        /*
2449         * Convert blobs to git internal format
2450         */
2451        if ((type == OBJ_BLOB) && path) {
2452                struct strbuf nbuf = STRBUF_INIT;
2453                if (convert_to_git(path, buf, size, &nbuf,
2454                                   write_object ? safe_crlf : 0)) {
2455                        buf = strbuf_detach(&nbuf, &size);
2456                        re_allocated = 1;
2457                }
2458        }
2459
2460        if (write_object)
2461                ret = write_sha1_file(buf, size, typename(type), sha1);
2462        else
2463                ret = hash_sha1_file(buf, size, typename(type), sha1);
2464        if (re_allocated)
2465                free(buf);
2466        return ret;
2467}
2468
2469int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2470             enum object_type type, const char *path)
2471{
2472        int ret;
2473        size_t size = xsize_t(st->st_size);
2474
2475        if (!S_ISREG(st->st_mode)) {
2476                struct strbuf sbuf = STRBUF_INIT;
2477                if (strbuf_read(&sbuf, fd, 4096) >= 0)
2478                        ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2479                                        type, path);
2480                else
2481                        ret = -1;
2482                strbuf_release(&sbuf);
2483        } else if (size) {
2484                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2485                ret = index_mem(sha1, buf, size, write_object, type, path);
2486                munmap(buf, size);
2487        } else
2488                ret = index_mem(sha1, NULL, size, write_object, type, path);
2489        close(fd);
2490        return ret;
2491}
2492
2493int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2494{
2495        int fd;
2496        char *target;
2497        size_t len;
2498
2499        switch (st->st_mode & S_IFMT) {
2500        case S_IFREG:
2501                fd = open(path, O_RDONLY);
2502                if (fd < 0)
2503                        return error("open(\"%s\"): %s", path,
2504                                     strerror(errno));
2505                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2506                        return error("%s: failed to insert into database",
2507                                     path);
2508                break;
2509        case S_IFLNK:
2510                len = xsize_t(st->st_size);
2511                target = xmalloc(len + 1);
2512                if (readlink(path, target, len + 1) != st->st_size) {
2513                        char *errstr = strerror(errno);
2514                        free(target);
2515                        return error("readlink(\"%s\"): %s", path,
2516                                     errstr);
2517                }
2518                if (!write_object)
2519                        hash_sha1_file(target, len, blob_type, sha1);
2520                else if (write_sha1_file(target, len, blob_type, sha1))
2521                        return error("%s: failed to insert into database",
2522                                     path);
2523                free(target);
2524                break;
2525        case S_IFDIR:
2526                return resolve_gitlink_ref(path, "HEAD", sha1);
2527        default:
2528                return error("%s: unsupported file type", path);
2529        }
2530        return 0;
2531}
2532
2533int read_pack_header(int fd, struct pack_header *header)
2534{
2535        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2536                /* "eof before pack header was fully read" */
2537                return PH_ERROR_EOF;
2538
2539        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2540                /* "protocol error (pack signature mismatch detected)" */
2541                return PH_ERROR_PACK_SIGNATURE;
2542        if (!pack_version_ok(header->hdr_version))
2543                /* "protocol error (pack version unsupported)" */
2544                return PH_ERROR_PROTOCOL;
2545        return 0;
2546}