sha1_file.con commit Merge branch 'lt/maint-diff-reduce-lstat' into maint (5e04a1e)
   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 * This is used by git-repack in case a newly created pack happens to
 677 * contain the same set of objects as an existing one.  In that case
 678 * the resulting file might be different even if its name would be the
 679 * same.  It is best to close any reference to the old pack before it is
 680 * replaced on disk.  Of course no index pointers nor windows for given pack
 681 * must subsist at this point.  If ever objects from this pack are requested
 682 * again, the new version of the pack will be reinitialized through
 683 * reprepare_packed_git().
 684 */
 685void free_pack_by_name(const char *pack_name)
 686{
 687        struct packed_git *p, **pp = &packed_git;
 688
 689        while (*pp) {
 690                p = *pp;
 691                if (strcmp(pack_name, p->pack_name) == 0) {
 692                        clear_delta_base_cache();
 693                        close_pack_windows(p);
 694                        if (p->pack_fd != -1)
 695                                close(p->pack_fd);
 696                        if (p->index_data)
 697                                munmap((void *)p->index_data, p->index_size);
 698                        free(p->bad_object_sha1);
 699                        *pp = p->next;
 700                        free(p);
 701                        return;
 702                }
 703                pp = &p->next;
 704        }
 705}
 706
 707/*
 708 * Do not call this directly as this leaks p->pack_fd on error return;
 709 * call open_packed_git() instead.
 710 */
 711static int open_packed_git_1(struct packed_git *p)
 712{
 713        struct stat st;
 714        struct pack_header hdr;
 715        unsigned char sha1[20];
 716        unsigned char *idx_sha1;
 717        long fd_flag;
 718
 719        if (!p->index_data && open_pack_index(p))
 720                return error("packfile %s index unavailable", p->pack_name);
 721
 722        p->pack_fd = open(p->pack_name, O_RDONLY);
 723        while (p->pack_fd < 0 && errno == EMFILE && unuse_one_window(p, -1))
 724                p->pack_fd = open(p->pack_name, O_RDONLY);
 725        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 726                return -1;
 727
 728        /* If we created the struct before we had the pack we lack size. */
 729        if (!p->pack_size) {
 730                if (!S_ISREG(st.st_mode))
 731                        return error("packfile %s not a regular file", p->pack_name);
 732                p->pack_size = st.st_size;
 733        } else if (p->pack_size != st.st_size)
 734                return error("packfile %s size changed", p->pack_name);
 735
 736        /* We leave these file descriptors open with sliding mmap;
 737         * there is no point keeping them open across exec(), though.
 738         */
 739        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 740        if (fd_flag < 0)
 741                return error("cannot determine file descriptor flags");
 742        fd_flag |= FD_CLOEXEC;
 743        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 744                return error("cannot set FD_CLOEXEC");
 745
 746        /* Verify we recognize this pack file format. */
 747        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 748                return error("file %s is far too short to be a packfile", p->pack_name);
 749        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 750                return error("file %s is not a GIT packfile", p->pack_name);
 751        if (!pack_version_ok(hdr.hdr_version))
 752                return error("packfile %s is version %"PRIu32" and not"
 753                        " supported (try upgrading GIT to a newer version)",
 754                        p->pack_name, ntohl(hdr.hdr_version));
 755
 756        /* Verify the pack matches its index. */
 757        if (p->num_objects != ntohl(hdr.hdr_entries))
 758                return error("packfile %s claims to have %"PRIu32" objects"
 759                             " while index indicates %"PRIu32" objects",
 760                             p->pack_name, ntohl(hdr.hdr_entries),
 761                             p->num_objects);
 762        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 763                return error("end of packfile %s is unavailable", p->pack_name);
 764        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 765                return error("packfile %s signature is unavailable", p->pack_name);
 766        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 767        if (hashcmp(sha1, idx_sha1))
 768                return error("packfile %s does not match index", p->pack_name);
 769        return 0;
 770}
 771
 772static int open_packed_git(struct packed_git *p)
 773{
 774        if (!open_packed_git_1(p))
 775                return 0;
 776        if (p->pack_fd != -1) {
 777                close(p->pack_fd);
 778                p->pack_fd = -1;
 779        }
 780        return -1;
 781}
 782
 783static int in_window(struct pack_window *win, off_t offset)
 784{
 785        /* We must promise at least 20 bytes (one hash) after the
 786         * offset is available from this window, otherwise the offset
 787         * is not actually in this window and a different window (which
 788         * has that one hash excess) must be used.  This is to support
 789         * the object header and delta base parsing routines below.
 790         */
 791        off_t win_off = win->offset;
 792        return win_off <= offset
 793                && (offset + 20) <= (win_off + win->len);
 794}
 795
 796unsigned char *use_pack(struct packed_git *p,
 797                struct pack_window **w_cursor,
 798                off_t offset,
 799                unsigned int *left)
 800{
 801        struct pack_window *win = *w_cursor;
 802
 803        if (p->pack_fd == -1 && open_packed_git(p))
 804                die("packfile %s cannot be accessed", p->pack_name);
 805
 806        /* Since packfiles end in a hash of their content and it's
 807         * pointless to ask for an offset into the middle of that
 808         * hash, and the in_window function above wouldn't match
 809         * don't allow an offset too close to the end of the file.
 810         */
 811        if (offset > (p->pack_size - 20))
 812                die("offset beyond end of packfile (truncated pack?)");
 813
 814        if (!win || !in_window(win, offset)) {
 815                if (win)
 816                        win->inuse_cnt--;
 817                for (win = p->windows; win; win = win->next) {
 818                        if (in_window(win, offset))
 819                                break;
 820                }
 821                if (!win) {
 822                        size_t window_align = packed_git_window_size / 2;
 823                        off_t len;
 824                        win = xcalloc(1, sizeof(*win));
 825                        win->offset = (offset / window_align) * window_align;
 826                        len = p->pack_size - win->offset;
 827                        if (len > packed_git_window_size)
 828                                len = packed_git_window_size;
 829                        win->len = (size_t)len;
 830                        pack_mapped += win->len;
 831                        while (packed_git_limit < pack_mapped
 832                                && unuse_one_window(p, p->pack_fd))
 833                                ; /* nothing */
 834                        win->base = xmmap(NULL, win->len,
 835                                PROT_READ, MAP_PRIVATE,
 836                                p->pack_fd, win->offset);
 837                        if (win->base == MAP_FAILED)
 838                                die("packfile %s cannot be mapped: %s",
 839                                        p->pack_name,
 840                                        strerror(errno));
 841                        pack_mmap_calls++;
 842                        pack_open_windows++;
 843                        if (pack_mapped > peak_pack_mapped)
 844                                peak_pack_mapped = pack_mapped;
 845                        if (pack_open_windows > peak_pack_open_windows)
 846                                peak_pack_open_windows = pack_open_windows;
 847                        win->next = p->windows;
 848                        p->windows = win;
 849                }
 850        }
 851        if (win != *w_cursor) {
 852                win->last_used = pack_used_ctr++;
 853                win->inuse_cnt++;
 854                *w_cursor = win;
 855        }
 856        offset -= win->offset;
 857        if (left)
 858                *left = win->len - xsize_t(offset);
 859        return win->base + offset;
 860}
 861
 862static struct packed_git *alloc_packed_git(int extra)
 863{
 864        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 865        memset(p, 0, sizeof(*p));
 866        p->pack_fd = -1;
 867        return p;
 868}
 869
 870struct packed_git *add_packed_git(const char *path, int path_len, int local)
 871{
 872        struct stat st;
 873        struct packed_git *p = alloc_packed_git(path_len + 2);
 874
 875        /*
 876         * Make sure a corresponding .pack file exists and that
 877         * the index looks sane.
 878         */
 879        path_len -= strlen(".idx");
 880        if (path_len < 1) {
 881                free(p);
 882                return NULL;
 883        }
 884        memcpy(p->pack_name, path, path_len);
 885
 886        strcpy(p->pack_name + path_len, ".keep");
 887        if (!access(p->pack_name, F_OK))
 888                p->pack_keep = 1;
 889
 890        strcpy(p->pack_name + path_len, ".pack");
 891        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 892                free(p);
 893                return NULL;
 894        }
 895
 896        /* ok, it looks sane as far as we can check without
 897         * actually mapping the pack file.
 898         */
 899        p->pack_size = st.st_size;
 900        p->pack_local = local;
 901        p->mtime = st.st_mtime;
 902        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 903                hashclr(p->sha1);
 904        return p;
 905}
 906
 907struct packed_git *parse_pack_index(unsigned char *sha1)
 908{
 909        const char *idx_path = sha1_pack_index_name(sha1);
 910        const char *path = sha1_pack_name(sha1);
 911        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 912
 913        strcpy(p->pack_name, path);
 914        hashcpy(p->sha1, sha1);
 915        if (check_packed_git_idx(idx_path, p)) {
 916                free(p);
 917                return NULL;
 918        }
 919
 920        return p;
 921}
 922
 923void install_packed_git(struct packed_git *pack)
 924{
 925        pack->next = packed_git;
 926        packed_git = pack;
 927}
 928
 929static void prepare_packed_git_one(char *objdir, int local)
 930{
 931        /* Ensure that this buffer is large enough so that we can
 932           append "/pack/" without clobbering the stack even if
 933           strlen(objdir) were PATH_MAX.  */
 934        char path[PATH_MAX + 1 + 4 + 1 + 1];
 935        int len;
 936        DIR *dir;
 937        struct dirent *de;
 938
 939        sprintf(path, "%s/pack", objdir);
 940        len = strlen(path);
 941        dir = opendir(path);
 942        while (!dir && errno == EMFILE && unuse_one_window(packed_git, -1))
 943                dir = opendir(path);
 944        if (!dir) {
 945                if (errno != ENOENT)
 946                        error("unable to open object pack directory: %s: %s",
 947                              path, strerror(errno));
 948                return;
 949        }
 950        path[len++] = '/';
 951        while ((de = readdir(dir)) != NULL) {
 952                int namelen = strlen(de->d_name);
 953                struct packed_git *p;
 954
 955                if (!has_extension(de->d_name, ".idx"))
 956                        continue;
 957
 958                if (len + namelen + 1 > sizeof(path))
 959                        continue;
 960
 961                /* Don't reopen a pack we already have. */
 962                strcpy(path + len, de->d_name);
 963                for (p = packed_git; p; p = p->next) {
 964                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 965                                break;
 966                }
 967                if (p)
 968                        continue;
 969                /* See if it really is a valid .idx file with corresponding
 970                 * .pack file that we can map.
 971                 */
 972                p = add_packed_git(path, len + namelen, local);
 973                if (!p)
 974                        continue;
 975                install_packed_git(p);
 976        }
 977        closedir(dir);
 978}
 979
 980static int sort_pack(const void *a_, const void *b_)
 981{
 982        struct packed_git *a = *((struct packed_git **)a_);
 983        struct packed_git *b = *((struct packed_git **)b_);
 984        int st;
 985
 986        /*
 987         * Local packs tend to contain objects specific to our
 988         * variant of the project than remote ones.  In addition,
 989         * remote ones could be on a network mounted filesystem.
 990         * Favor local ones for these reasons.
 991         */
 992        st = a->pack_local - b->pack_local;
 993        if (st)
 994                return -st;
 995
 996        /*
 997         * Younger packs tend to contain more recent objects,
 998         * and more recent objects tend to get accessed more
 999         * often.
1000         */
1001        if (a->mtime < b->mtime)
1002                return 1;
1003        else if (a->mtime == b->mtime)
1004                return 0;
1005        return -1;
1006}
1007
1008static void rearrange_packed_git(void)
1009{
1010        struct packed_git **ary, *p;
1011        int i, n;
1012
1013        for (n = 0, p = packed_git; p; p = p->next)
1014                n++;
1015        if (n < 2)
1016                return;
1017
1018        /* prepare an array of packed_git for easier sorting */
1019        ary = xcalloc(n, sizeof(struct packed_git *));
1020        for (n = 0, p = packed_git; p; p = p->next)
1021                ary[n++] = p;
1022
1023        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1024
1025        /* link them back again */
1026        for (i = 0; i < n - 1; i++)
1027                ary[i]->next = ary[i + 1];
1028        ary[n - 1]->next = NULL;
1029        packed_git = ary[0];
1030
1031        free(ary);
1032}
1033
1034static int prepare_packed_git_run_once = 0;
1035void prepare_packed_git(void)
1036{
1037        struct alternate_object_database *alt;
1038
1039        if (prepare_packed_git_run_once)
1040                return;
1041        prepare_packed_git_one(get_object_directory(), 1);
1042        prepare_alt_odb();
1043        for (alt = alt_odb_list; alt; alt = alt->next) {
1044                alt->name[-1] = 0;
1045                prepare_packed_git_one(alt->base, 0);
1046                alt->name[-1] = '/';
1047        }
1048        rearrange_packed_git();
1049        prepare_packed_git_run_once = 1;
1050}
1051
1052void reprepare_packed_git(void)
1053{
1054        discard_revindex();
1055        prepare_packed_git_run_once = 0;
1056        prepare_packed_git();
1057}
1058
1059static void mark_bad_packed_object(struct packed_git *p,
1060                                   const unsigned char *sha1)
1061{
1062        unsigned i;
1063        for (i = 0; i < p->num_bad_objects; i++)
1064                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1065                        return;
1066        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1067        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1068        p->num_bad_objects++;
1069}
1070
1071static int has_packed_and_bad(const unsigned char *sha1)
1072{
1073        struct packed_git *p;
1074        unsigned i;
1075
1076        for (p = packed_git; p; p = p->next)
1077                for (i = 0; i < p->num_bad_objects; i++)
1078                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1079                                return 1;
1080        return 0;
1081}
1082
1083int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1084{
1085        unsigned char real_sha1[20];
1086        hash_sha1_file(map, size, type, real_sha1);
1087        return hashcmp(sha1, real_sha1) ? -1 : 0;
1088}
1089
1090static int git_open_noatime(const char *name)
1091{
1092        static int sha1_file_open_flag = O_NOATIME;
1093        int fd = open(name, O_RDONLY | sha1_file_open_flag);
1094
1095        /* Might the failure be due to O_NOATIME? */
1096        if (fd < 0 && errno != ENOENT && sha1_file_open_flag) {
1097                fd = open(name, O_RDONLY);
1098                if (fd >= 0)
1099                        sha1_file_open_flag = 0;
1100        }
1101        return fd;
1102}
1103
1104static int open_sha1_file(const unsigned char *sha1)
1105{
1106        int fd;
1107        char *name = sha1_file_name(sha1);
1108        struct alternate_object_database *alt;
1109
1110        fd = git_open_noatime(name);
1111        if (fd >= 0)
1112                return fd;
1113
1114        prepare_alt_odb();
1115        errno = ENOENT;
1116        for (alt = alt_odb_list; alt; alt = alt->next) {
1117                name = alt->name;
1118                fill_sha1_path(name, sha1);
1119                fd = git_open_noatime(alt->base);
1120                if (fd >= 0)
1121                        return fd;
1122        }
1123        return -1;
1124}
1125
1126static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1127{
1128        void *map;
1129        int fd;
1130
1131        fd = open_sha1_file(sha1);
1132        map = NULL;
1133        if (fd >= 0) {
1134                struct stat st;
1135
1136                if (!fstat(fd, &st)) {
1137                        *size = xsize_t(st.st_size);
1138                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1139                }
1140                close(fd);
1141        }
1142        return map;
1143}
1144
1145static int legacy_loose_object(unsigned char *map)
1146{
1147        unsigned int word;
1148
1149        /*
1150         * Is it a zlib-compressed buffer? If so, the first byte
1151         * must be 0x78 (15-bit window size, deflated), and the
1152         * first 16-bit word is evenly divisible by 31
1153         */
1154        word = (map[0] << 8) + map[1];
1155        if (map[0] == 0x78 && !(word % 31))
1156                return 1;
1157        else
1158                return 0;
1159}
1160
1161unsigned long unpack_object_header_buffer(const unsigned char *buf,
1162                unsigned long len, enum object_type *type, unsigned long *sizep)
1163{
1164        unsigned shift;
1165        unsigned char c;
1166        unsigned long size;
1167        unsigned long used = 0;
1168
1169        c = buf[used++];
1170        *type = (c >> 4) & 7;
1171        size = c & 15;
1172        shift = 4;
1173        while (c & 0x80) {
1174                if (len <= used || sizeof(long) * 8 <= shift) {
1175                        error("bad object header");
1176                        return 0;
1177                }
1178                c = buf[used++];
1179                size += (c & 0x7f) << shift;
1180                shift += 7;
1181        }
1182        *sizep = size;
1183        return used;
1184}
1185
1186static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1187{
1188        unsigned long size, used;
1189        static const char valid_loose_object_type[8] = {
1190                0, /* OBJ_EXT */
1191                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1192                0, /* "delta" and others are invalid in a loose object */
1193        };
1194        enum object_type type;
1195
1196        /* Get the data stream */
1197        memset(stream, 0, sizeof(*stream));
1198        stream->next_in = map;
1199        stream->avail_in = mapsize;
1200        stream->next_out = buffer;
1201        stream->avail_out = bufsiz;
1202
1203        if (legacy_loose_object(map)) {
1204                git_inflate_init(stream);
1205                return git_inflate(stream, 0);
1206        }
1207
1208
1209        /*
1210         * There used to be a second loose object header format which
1211         * was meant to mimic the in-pack format, allowing for direct
1212         * copy of the object data.  This format turned up not to be
1213         * really worth it and we don't write it any longer.  But we
1214         * can still read it.
1215         */
1216        used = unpack_object_header_buffer(map, mapsize, &type, &size);
1217        if (!used || !valid_loose_object_type[type])
1218                return -1;
1219        map += used;
1220        mapsize -= used;
1221
1222        /* Set up the stream for the rest.. */
1223        stream->next_in = map;
1224        stream->avail_in = mapsize;
1225        git_inflate_init(stream);
1226
1227        /* And generate the fake traditional header */
1228        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1229                                         typename(type), size);
1230        return 0;
1231}
1232
1233static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1234{
1235        int bytes = strlen(buffer) + 1;
1236        unsigned char *buf = xmalloc(1+size);
1237        unsigned long n;
1238        int status = Z_OK;
1239
1240        n = stream->total_out - bytes;
1241        if (n > size)
1242                n = size;
1243        memcpy(buf, (char *) buffer + bytes, n);
1244        bytes = n;
1245        if (bytes <= size) {
1246                /*
1247                 * The above condition must be (bytes <= size), not
1248                 * (bytes < size).  In other words, even though we
1249                 * expect no more output and set avail_out to zer0,
1250                 * the input zlib stream may have bytes that express
1251                 * "this concludes the stream", and we *do* want to
1252                 * eat that input.
1253                 *
1254                 * Otherwise we would not be able to test that we
1255                 * consumed all the input to reach the expected size;
1256                 * we also want to check that zlib tells us that all
1257                 * went well with status == Z_STREAM_END at the end.
1258                 */
1259                stream->next_out = buf + bytes;
1260                stream->avail_out = size - bytes;
1261                while (status == Z_OK)
1262                        status = git_inflate(stream, Z_FINISH);
1263        }
1264        buf[size] = 0;
1265        if (status == Z_STREAM_END && !stream->avail_in) {
1266                git_inflate_end(stream);
1267                return buf;
1268        }
1269
1270        if (status < 0)
1271                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1272        else if (stream->avail_in)
1273                error("garbage at end of loose object '%s'",
1274                      sha1_to_hex(sha1));
1275        free(buf);
1276        return NULL;
1277}
1278
1279/*
1280 * We used to just use "sscanf()", but that's actually way
1281 * too permissive for what we want to check. So do an anal
1282 * object header parse by hand.
1283 */
1284static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1285{
1286        char type[10];
1287        int i;
1288        unsigned long size;
1289
1290        /*
1291         * The type can be at most ten bytes (including the
1292         * terminating '\0' that we add), and is followed by
1293         * a space.
1294         */
1295        i = 0;
1296        for (;;) {
1297                char c = *hdr++;
1298                if (c == ' ')
1299                        break;
1300                type[i++] = c;
1301                if (i >= sizeof(type))
1302                        return -1;
1303        }
1304        type[i] = 0;
1305
1306        /*
1307         * The length must follow immediately, and be in canonical
1308         * decimal format (ie "010" is not valid).
1309         */
1310        size = *hdr++ - '0';
1311        if (size > 9)
1312                return -1;
1313        if (size) {
1314                for (;;) {
1315                        unsigned long c = *hdr - '0';
1316                        if (c > 9)
1317                                break;
1318                        hdr++;
1319                        size = size * 10 + c;
1320                }
1321        }
1322        *sizep = size;
1323
1324        /*
1325         * The length must be followed by a zero byte
1326         */
1327        return *hdr ? -1 : type_from_string(type);
1328}
1329
1330static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1331{
1332        int ret;
1333        z_stream stream;
1334        char hdr[8192];
1335
1336        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1337        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1338                return NULL;
1339
1340        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1341}
1342
1343unsigned long get_size_from_delta(struct packed_git *p,
1344                                  struct pack_window **w_curs,
1345                                  off_t curpos)
1346{
1347        const unsigned char *data;
1348        unsigned char delta_head[20], *in;
1349        z_stream stream;
1350        int st;
1351
1352        memset(&stream, 0, sizeof(stream));
1353        stream.next_out = delta_head;
1354        stream.avail_out = sizeof(delta_head);
1355
1356        git_inflate_init(&stream);
1357        do {
1358                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1359                stream.next_in = in;
1360                st = git_inflate(&stream, Z_FINISH);
1361                curpos += stream.next_in - in;
1362        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1363                 stream.total_out < sizeof(delta_head));
1364        git_inflate_end(&stream);
1365        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1366                error("delta data unpack-initial failed");
1367                return 0;
1368        }
1369
1370        /* Examine the initial part of the delta to figure out
1371         * the result size.
1372         */
1373        data = delta_head;
1374
1375        /* ignore base size */
1376        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1377
1378        /* Read the result size */
1379        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1380}
1381
1382static off_t get_delta_base(struct packed_git *p,
1383                                    struct pack_window **w_curs,
1384                                    off_t *curpos,
1385                                    enum object_type type,
1386                                    off_t delta_obj_offset)
1387{
1388        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1389        off_t base_offset;
1390
1391        /* use_pack() assured us we have [base_info, base_info + 20)
1392         * as a range that we can look at without walking off the
1393         * end of the mapped window.  Its actually the hash size
1394         * that is assured.  An OFS_DELTA longer than the hash size
1395         * is stupid, as then a REF_DELTA would be smaller to store.
1396         */
1397        if (type == OBJ_OFS_DELTA) {
1398                unsigned used = 0;
1399                unsigned char c = base_info[used++];
1400                base_offset = c & 127;
1401                while (c & 128) {
1402                        base_offset += 1;
1403                        if (!base_offset || MSB(base_offset, 7))
1404                                return 0;  /* overflow */
1405                        c = base_info[used++];
1406                        base_offset = (base_offset << 7) + (c & 127);
1407                }
1408                base_offset = delta_obj_offset - base_offset;
1409                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1410                        return 0;  /* out of bound */
1411                *curpos += used;
1412        } else if (type == OBJ_REF_DELTA) {
1413                /* The base entry _must_ be in the same pack */
1414                base_offset = find_pack_entry_one(base_info, p);
1415                *curpos += 20;
1416        } else
1417                die("I am totally screwed");
1418        return base_offset;
1419}
1420
1421/* forward declaration for a mutually recursive function */
1422static int packed_object_info(struct packed_git *p, off_t offset,
1423                              unsigned long *sizep);
1424
1425static int packed_delta_info(struct packed_git *p,
1426                             struct pack_window **w_curs,
1427                             off_t curpos,
1428                             enum object_type type,
1429                             off_t obj_offset,
1430                             unsigned long *sizep)
1431{
1432        off_t base_offset;
1433
1434        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1435        if (!base_offset)
1436                return OBJ_BAD;
1437        type = packed_object_info(p, base_offset, NULL);
1438        if (type <= OBJ_NONE) {
1439                struct revindex_entry *revidx;
1440                const unsigned char *base_sha1;
1441                revidx = find_pack_revindex(p, base_offset);
1442                if (!revidx)
1443                        return OBJ_BAD;
1444                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1445                mark_bad_packed_object(p, base_sha1);
1446                type = sha1_object_info(base_sha1, NULL);
1447                if (type <= OBJ_NONE)
1448                        return OBJ_BAD;
1449        }
1450
1451        /* We choose to only get the type of the base object and
1452         * ignore potentially corrupt pack file that expects the delta
1453         * based on a base with a wrong size.  This saves tons of
1454         * inflate() calls.
1455         */
1456        if (sizep) {
1457                *sizep = get_size_from_delta(p, w_curs, curpos);
1458                if (*sizep == 0)
1459                        type = OBJ_BAD;
1460        }
1461
1462        return type;
1463}
1464
1465static int unpack_object_header(struct packed_git *p,
1466                                struct pack_window **w_curs,
1467                                off_t *curpos,
1468                                unsigned long *sizep)
1469{
1470        unsigned char *base;
1471        unsigned int left;
1472        unsigned long used;
1473        enum object_type type;
1474
1475        /* use_pack() assures us we have [base, base + 20) available
1476         * as a range that we can look at at.  (Its actually the hash
1477         * size that is assured.)  With our object header encoding
1478         * the maximum deflated object size is 2^137, which is just
1479         * insane, so we know won't exceed what we have been given.
1480         */
1481        base = use_pack(p, w_curs, *curpos, &left);
1482        used = unpack_object_header_buffer(base, left, &type, sizep);
1483        if (!used) {
1484                type = OBJ_BAD;
1485        } else
1486                *curpos += used;
1487
1488        return type;
1489}
1490
1491const char *packed_object_info_detail(struct packed_git *p,
1492                                      off_t obj_offset,
1493                                      unsigned long *size,
1494                                      unsigned long *store_size,
1495                                      unsigned int *delta_chain_length,
1496                                      unsigned char *base_sha1)
1497{
1498        struct pack_window *w_curs = NULL;
1499        off_t curpos;
1500        unsigned long dummy;
1501        unsigned char *next_sha1;
1502        enum object_type type;
1503        struct revindex_entry *revidx;
1504
1505        *delta_chain_length = 0;
1506        curpos = obj_offset;
1507        type = unpack_object_header(p, &w_curs, &curpos, size);
1508
1509        revidx = find_pack_revindex(p, obj_offset);
1510        *store_size = revidx[1].offset - obj_offset;
1511
1512        for (;;) {
1513                switch (type) {
1514                default:
1515                        die("pack %s contains unknown object type %d",
1516                            p->pack_name, type);
1517                case OBJ_COMMIT:
1518                case OBJ_TREE:
1519                case OBJ_BLOB:
1520                case OBJ_TAG:
1521                        unuse_pack(&w_curs);
1522                        return typename(type);
1523                case OBJ_OFS_DELTA:
1524                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1525                        if (!obj_offset)
1526                                die("pack %s contains bad delta base reference of type %s",
1527                                    p->pack_name, typename(type));
1528                        if (*delta_chain_length == 0) {
1529                                revidx = find_pack_revindex(p, obj_offset);
1530                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1531                        }
1532                        break;
1533                case OBJ_REF_DELTA:
1534                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1535                        if (*delta_chain_length == 0)
1536                                hashcpy(base_sha1, next_sha1);
1537                        obj_offset = find_pack_entry_one(next_sha1, p);
1538                        break;
1539                }
1540                (*delta_chain_length)++;
1541                curpos = obj_offset;
1542                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1543        }
1544}
1545
1546static int packed_object_info(struct packed_git *p, off_t obj_offset,
1547                              unsigned long *sizep)
1548{
1549        struct pack_window *w_curs = NULL;
1550        unsigned long size;
1551        off_t curpos = obj_offset;
1552        enum object_type type;
1553
1554        type = unpack_object_header(p, &w_curs, &curpos, &size);
1555
1556        switch (type) {
1557        case OBJ_OFS_DELTA:
1558        case OBJ_REF_DELTA:
1559                type = packed_delta_info(p, &w_curs, curpos,
1560                                         type, obj_offset, sizep);
1561                break;
1562        case OBJ_COMMIT:
1563        case OBJ_TREE:
1564        case OBJ_BLOB:
1565        case OBJ_TAG:
1566                if (sizep)
1567                        *sizep = size;
1568                break;
1569        default:
1570                error("unknown object type %i at offset %"PRIuMAX" in %s",
1571                      type, (uintmax_t)obj_offset, p->pack_name);
1572                type = OBJ_BAD;
1573        }
1574        unuse_pack(&w_curs);
1575        return type;
1576}
1577
1578static void *unpack_compressed_entry(struct packed_git *p,
1579                                    struct pack_window **w_curs,
1580                                    off_t curpos,
1581                                    unsigned long size)
1582{
1583        int st;
1584        z_stream stream;
1585        unsigned char *buffer, *in;
1586
1587        buffer = xmalloc(size + 1);
1588        buffer[size] = 0;
1589        memset(&stream, 0, sizeof(stream));
1590        stream.next_out = buffer;
1591        stream.avail_out = size;
1592
1593        git_inflate_init(&stream);
1594        do {
1595                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1596                stream.next_in = in;
1597                st = git_inflate(&stream, Z_FINISH);
1598                curpos += stream.next_in - in;
1599        } while (st == Z_OK || st == Z_BUF_ERROR);
1600        git_inflate_end(&stream);
1601        if ((st != Z_STREAM_END) || stream.total_out != size) {
1602                free(buffer);
1603                return NULL;
1604        }
1605
1606        return buffer;
1607}
1608
1609#define MAX_DELTA_CACHE (256)
1610
1611static size_t delta_base_cached;
1612
1613static struct delta_base_cache_lru_list {
1614        struct delta_base_cache_lru_list *prev;
1615        struct delta_base_cache_lru_list *next;
1616} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1617
1618static struct delta_base_cache_entry {
1619        struct delta_base_cache_lru_list lru;
1620        void *data;
1621        struct packed_git *p;
1622        off_t base_offset;
1623        unsigned long size;
1624        enum object_type type;
1625} delta_base_cache[MAX_DELTA_CACHE];
1626
1627static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1628{
1629        unsigned long hash;
1630
1631        hash = (unsigned long)p + (unsigned long)base_offset;
1632        hash += (hash >> 8) + (hash >> 16);
1633        return hash % MAX_DELTA_CACHE;
1634}
1635
1636static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1637        unsigned long *base_size, enum object_type *type, int keep_cache)
1638{
1639        void *ret;
1640        unsigned long hash = pack_entry_hash(p, base_offset);
1641        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1642
1643        ret = ent->data;
1644        if (!ret || ent->p != p || ent->base_offset != base_offset)
1645                return unpack_entry(p, base_offset, type, base_size);
1646
1647        if (!keep_cache) {
1648                ent->data = NULL;
1649                ent->lru.next->prev = ent->lru.prev;
1650                ent->lru.prev->next = ent->lru.next;
1651                delta_base_cached -= ent->size;
1652        } else {
1653                ret = xmemdupz(ent->data, ent->size);
1654        }
1655        *type = ent->type;
1656        *base_size = ent->size;
1657        return ret;
1658}
1659
1660static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1661{
1662        if (ent->data) {
1663                free(ent->data);
1664                ent->data = NULL;
1665                ent->lru.next->prev = ent->lru.prev;
1666                ent->lru.prev->next = ent->lru.next;
1667                delta_base_cached -= ent->size;
1668        }
1669}
1670
1671void clear_delta_base_cache(void)
1672{
1673        unsigned long p;
1674        for (p = 0; p < MAX_DELTA_CACHE; p++)
1675                release_delta_base_cache(&delta_base_cache[p]);
1676}
1677
1678static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1679        void *base, unsigned long base_size, enum object_type type)
1680{
1681        unsigned long hash = pack_entry_hash(p, base_offset);
1682        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1683        struct delta_base_cache_lru_list *lru;
1684
1685        release_delta_base_cache(ent);
1686        delta_base_cached += base_size;
1687
1688        for (lru = delta_base_cache_lru.next;
1689             delta_base_cached > delta_base_cache_limit
1690             && lru != &delta_base_cache_lru;
1691             lru = lru->next) {
1692                struct delta_base_cache_entry *f = (void *)lru;
1693                if (f->type == OBJ_BLOB)
1694                        release_delta_base_cache(f);
1695        }
1696        for (lru = delta_base_cache_lru.next;
1697             delta_base_cached > delta_base_cache_limit
1698             && lru != &delta_base_cache_lru;
1699             lru = lru->next) {
1700                struct delta_base_cache_entry *f = (void *)lru;
1701                release_delta_base_cache(f);
1702        }
1703
1704        ent->p = p;
1705        ent->base_offset = base_offset;
1706        ent->type = type;
1707        ent->data = base;
1708        ent->size = base_size;
1709        ent->lru.next = &delta_base_cache_lru;
1710        ent->lru.prev = delta_base_cache_lru.prev;
1711        delta_base_cache_lru.prev->next = &ent->lru;
1712        delta_base_cache_lru.prev = &ent->lru;
1713}
1714
1715static void *read_object(const unsigned char *sha1, enum object_type *type,
1716                         unsigned long *size);
1717
1718static void *unpack_delta_entry(struct packed_git *p,
1719                                struct pack_window **w_curs,
1720                                off_t curpos,
1721                                unsigned long delta_size,
1722                                off_t obj_offset,
1723                                enum object_type *type,
1724                                unsigned long *sizep)
1725{
1726        void *delta_data, *result, *base;
1727        unsigned long base_size;
1728        off_t base_offset;
1729
1730        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1731        if (!base_offset) {
1732                error("failed to validate delta base reference "
1733                      "at offset %"PRIuMAX" from %s",
1734                      (uintmax_t)curpos, p->pack_name);
1735                return NULL;
1736        }
1737        unuse_pack(w_curs);
1738        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1739        if (!base) {
1740                /*
1741                 * We're probably in deep shit, but let's try to fetch
1742                 * the required base anyway from another pack or loose.
1743                 * This is costly but should happen only in the presence
1744                 * of a corrupted pack, and is better than failing outright.
1745                 */
1746                struct revindex_entry *revidx;
1747                const unsigned char *base_sha1;
1748                revidx = find_pack_revindex(p, base_offset);
1749                if (!revidx)
1750                        return NULL;
1751                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1752                error("failed to read delta base object %s"
1753                      " at offset %"PRIuMAX" from %s",
1754                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1755                      p->pack_name);
1756                mark_bad_packed_object(p, base_sha1);
1757                base = read_object(base_sha1, type, &base_size);
1758                if (!base)
1759                        return NULL;
1760        }
1761
1762        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1763        if (!delta_data) {
1764                error("failed to unpack compressed delta "
1765                      "at offset %"PRIuMAX" from %s",
1766                      (uintmax_t)curpos, p->pack_name);
1767                free(base);
1768                return NULL;
1769        }
1770        result = patch_delta(base, base_size,
1771                             delta_data, delta_size,
1772                             sizep);
1773        if (!result)
1774                die("failed to apply delta");
1775        free(delta_data);
1776        add_delta_base_cache(p, base_offset, base, base_size, *type);
1777        return result;
1778}
1779
1780int do_check_packed_object_crc;
1781
1782void *unpack_entry(struct packed_git *p, off_t obj_offset,
1783                   enum object_type *type, unsigned long *sizep)
1784{
1785        struct pack_window *w_curs = NULL;
1786        off_t curpos = obj_offset;
1787        void *data;
1788
1789        if (do_check_packed_object_crc && p->index_version > 1) {
1790                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1791                unsigned long len = revidx[1].offset - obj_offset;
1792                if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1793                        const unsigned char *sha1 =
1794                                nth_packed_object_sha1(p, revidx->nr);
1795                        error("bad packed object CRC for %s",
1796                              sha1_to_hex(sha1));
1797                        mark_bad_packed_object(p, sha1);
1798                        unuse_pack(&w_curs);
1799                        return NULL;
1800                }
1801        }
1802
1803        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1804        switch (*type) {
1805        case OBJ_OFS_DELTA:
1806        case OBJ_REF_DELTA:
1807                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1808                                          obj_offset, type, sizep);
1809                break;
1810        case OBJ_COMMIT:
1811        case OBJ_TREE:
1812        case OBJ_BLOB:
1813        case OBJ_TAG:
1814                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1815                break;
1816        default:
1817                data = NULL;
1818                error("unknown object type %i at offset %"PRIuMAX" in %s",
1819                      *type, (uintmax_t)obj_offset, p->pack_name);
1820        }
1821        unuse_pack(&w_curs);
1822        return data;
1823}
1824
1825const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1826                                            uint32_t n)
1827{
1828        const unsigned char *index = p->index_data;
1829        if (!index) {
1830                if (open_pack_index(p))
1831                        return NULL;
1832                index = p->index_data;
1833        }
1834        if (n >= p->num_objects)
1835                return NULL;
1836        index += 4 * 256;
1837        if (p->index_version == 1) {
1838                return index + 24 * n + 4;
1839        } else {
1840                index += 8;
1841                return index + 20 * n;
1842        }
1843}
1844
1845off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1846{
1847        const unsigned char *index = p->index_data;
1848        index += 4 * 256;
1849        if (p->index_version == 1) {
1850                return ntohl(*((uint32_t *)(index + 24 * n)));
1851        } else {
1852                uint32_t off;
1853                index += 8 + p->num_objects * (20 + 4);
1854                off = ntohl(*((uint32_t *)(index + 4 * n)));
1855                if (!(off & 0x80000000))
1856                        return off;
1857                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1858                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1859                                   ntohl(*((uint32_t *)(index + 4)));
1860        }
1861}
1862
1863off_t find_pack_entry_one(const unsigned char *sha1,
1864                                  struct packed_git *p)
1865{
1866        const uint32_t *level1_ofs = p->index_data;
1867        const unsigned char *index = p->index_data;
1868        unsigned hi, lo, stride;
1869        static int use_lookup = -1;
1870        static int debug_lookup = -1;
1871
1872        if (debug_lookup < 0)
1873                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1874
1875        if (!index) {
1876                if (open_pack_index(p))
1877                        return 0;
1878                level1_ofs = p->index_data;
1879                index = p->index_data;
1880        }
1881        if (p->index_version > 1) {
1882                level1_ofs += 2;
1883                index += 8;
1884        }
1885        index += 4 * 256;
1886        hi = ntohl(level1_ofs[*sha1]);
1887        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1888        if (p->index_version > 1) {
1889                stride = 20;
1890        } else {
1891                stride = 24;
1892                index += 4;
1893        }
1894
1895        if (debug_lookup)
1896                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1897                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1898
1899        if (use_lookup < 0)
1900                use_lookup = !!getenv("GIT_USE_LOOKUP");
1901        if (use_lookup) {
1902                int pos = sha1_entry_pos(index, stride, 0,
1903                                         lo, hi, p->num_objects, sha1);
1904                if (pos < 0)
1905                        return 0;
1906                return nth_packed_object_offset(p, pos);
1907        }
1908
1909        do {
1910                unsigned mi = (lo + hi) / 2;
1911                int cmp = hashcmp(index + mi * stride, sha1);
1912
1913                if (debug_lookup)
1914                        printf("lo %u hi %u rg %u mi %u\n",
1915                               lo, hi, hi - lo, mi);
1916                if (!cmp)
1917                        return nth_packed_object_offset(p, mi);
1918                if (cmp > 0)
1919                        hi = mi;
1920                else
1921                        lo = mi+1;
1922        } while (lo < hi);
1923        return 0;
1924}
1925
1926static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1927{
1928        static struct packed_git *last_found = (void *)1;
1929        struct packed_git *p;
1930        off_t offset;
1931
1932        prepare_packed_git();
1933        if (!packed_git)
1934                return 0;
1935        p = (last_found == (void *)1) ? packed_git : last_found;
1936
1937        do {
1938                if (p->num_bad_objects) {
1939                        unsigned i;
1940                        for (i = 0; i < p->num_bad_objects; i++)
1941                                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1942                                        goto next;
1943                }
1944
1945                offset = find_pack_entry_one(sha1, p);
1946                if (offset) {
1947                        /*
1948                         * We are about to tell the caller where they can
1949                         * locate the requested object.  We better make
1950                         * sure the packfile is still here and can be
1951                         * accessed before supplying that answer, as
1952                         * it may have been deleted since the index
1953                         * was loaded!
1954                         */
1955                        if (p->pack_fd == -1 && open_packed_git(p)) {
1956                                error("packfile %s cannot be accessed", p->pack_name);
1957                                goto next;
1958                        }
1959                        e->offset = offset;
1960                        e->p = p;
1961                        hashcpy(e->sha1, sha1);
1962                        last_found = p;
1963                        return 1;
1964                }
1965
1966                next:
1967                if (p == last_found)
1968                        p = packed_git;
1969                else
1970                        p = p->next;
1971                if (p == last_found)
1972                        p = p->next;
1973        } while (p);
1974        return 0;
1975}
1976
1977struct packed_git *find_sha1_pack(const unsigned char *sha1,
1978                                  struct packed_git *packs)
1979{
1980        struct packed_git *p;
1981
1982        for (p = packs; p; p = p->next) {
1983                if (find_pack_entry_one(sha1, p))
1984                        return p;
1985        }
1986        return NULL;
1987
1988}
1989
1990static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1991{
1992        int status;
1993        unsigned long mapsize, size;
1994        void *map;
1995        z_stream stream;
1996        char hdr[32];
1997
1998        map = map_sha1_file(sha1, &mapsize);
1999        if (!map)
2000                return error("unable to find %s", sha1_to_hex(sha1));
2001        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2002                status = error("unable to unpack %s header",
2003                               sha1_to_hex(sha1));
2004        else if ((status = parse_sha1_header(hdr, &size)) < 0)
2005                status = error("unable to parse %s header", sha1_to_hex(sha1));
2006        else if (sizep)
2007                *sizep = size;
2008        git_inflate_end(&stream);
2009        munmap(map, mapsize);
2010        return status;
2011}
2012
2013int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2014{
2015        struct pack_entry e;
2016        int status;
2017
2018        if (!find_pack_entry(sha1, &e)) {
2019                /* Most likely it's a loose object. */
2020                status = sha1_loose_object_info(sha1, sizep);
2021                if (status >= 0)
2022                        return status;
2023
2024                /* Not a loose object; someone else may have just packed it. */
2025                reprepare_packed_git();
2026                if (!find_pack_entry(sha1, &e))
2027                        return status;
2028        }
2029
2030        status = packed_object_info(e.p, e.offset, sizep);
2031        if (status < 0) {
2032                mark_bad_packed_object(e.p, sha1);
2033                status = sha1_object_info(sha1, sizep);
2034        }
2035
2036        return status;
2037}
2038
2039static void *read_packed_sha1(const unsigned char *sha1,
2040                              enum object_type *type, unsigned long *size)
2041{
2042        struct pack_entry e;
2043        void *data;
2044
2045        if (!find_pack_entry(sha1, &e))
2046                return NULL;
2047        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2048        if (!data) {
2049                /*
2050                 * We're probably in deep shit, but let's try to fetch
2051                 * the required object anyway from another pack or loose.
2052                 * This should happen only in the presence of a corrupted
2053                 * pack, and is better than failing outright.
2054                 */
2055                error("failed to read object %s at offset %"PRIuMAX" from %s",
2056                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2057                mark_bad_packed_object(e.p, sha1);
2058                data = read_object(sha1, type, size);
2059        }
2060        return data;
2061}
2062
2063/*
2064 * This is meant to hold a *small* number of objects that you would
2065 * want read_sha1_file() to be able to return, but yet you do not want
2066 * to write them into the object store (e.g. a browse-only
2067 * application).
2068 */
2069static struct cached_object {
2070        unsigned char sha1[20];
2071        enum object_type type;
2072        void *buf;
2073        unsigned long size;
2074} *cached_objects;
2075static int cached_object_nr, cached_object_alloc;
2076
2077static struct cached_object empty_tree = {
2078        EMPTY_TREE_SHA1_BIN,
2079        OBJ_TREE,
2080        "",
2081        0
2082};
2083
2084static struct cached_object *find_cached_object(const unsigned char *sha1)
2085{
2086        int i;
2087        struct cached_object *co = cached_objects;
2088
2089        for (i = 0; i < cached_object_nr; i++, co++) {
2090                if (!hashcmp(co->sha1, sha1))
2091                        return co;
2092        }
2093        if (!hashcmp(sha1, empty_tree.sha1))
2094                return &empty_tree;
2095        return NULL;
2096}
2097
2098int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2099                      unsigned char *sha1)
2100{
2101        struct cached_object *co;
2102
2103        hash_sha1_file(buf, len, typename(type), sha1);
2104        if (has_sha1_file(sha1) || find_cached_object(sha1))
2105                return 0;
2106        if (cached_object_alloc <= cached_object_nr) {
2107                cached_object_alloc = alloc_nr(cached_object_alloc);
2108                cached_objects = xrealloc(cached_objects,
2109                                          sizeof(*cached_objects) *
2110                                          cached_object_alloc);
2111        }
2112        co = &cached_objects[cached_object_nr++];
2113        co->size = len;
2114        co->type = type;
2115        co->buf = xmalloc(len);
2116        memcpy(co->buf, buf, len);
2117        hashcpy(co->sha1, sha1);
2118        return 0;
2119}
2120
2121static void *read_object(const unsigned char *sha1, enum object_type *type,
2122                         unsigned long *size)
2123{
2124        unsigned long mapsize;
2125        void *map, *buf;
2126        struct cached_object *co;
2127
2128        co = find_cached_object(sha1);
2129        if (co) {
2130                *type = co->type;
2131                *size = co->size;
2132                return xmemdupz(co->buf, co->size);
2133        }
2134
2135        buf = read_packed_sha1(sha1, type, size);
2136        if (buf)
2137                return buf;
2138        map = map_sha1_file(sha1, &mapsize);
2139        if (map) {
2140                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2141                munmap(map, mapsize);
2142                return buf;
2143        }
2144        reprepare_packed_git();
2145        return read_packed_sha1(sha1, type, size);
2146}
2147
2148void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
2149                     unsigned long *size)
2150{
2151        void *data = read_object(sha1, type, size);
2152        /* legacy behavior is to die on corrupted objects */
2153        if (!data && (has_loose_object(sha1) || has_packed_and_bad(sha1)))
2154                die("object %s is corrupted", sha1_to_hex(sha1));
2155        return data;
2156}
2157
2158void *read_object_with_reference(const unsigned char *sha1,
2159                                 const char *required_type_name,
2160                                 unsigned long *size,
2161                                 unsigned char *actual_sha1_return)
2162{
2163        enum object_type type, required_type;
2164        void *buffer;
2165        unsigned long isize;
2166        unsigned char actual_sha1[20];
2167
2168        required_type = type_from_string(required_type_name);
2169        hashcpy(actual_sha1, sha1);
2170        while (1) {
2171                int ref_length = -1;
2172                const char *ref_type = NULL;
2173
2174                buffer = read_sha1_file(actual_sha1, &type, &isize);
2175                if (!buffer)
2176                        return NULL;
2177                if (type == required_type) {
2178                        *size = isize;
2179                        if (actual_sha1_return)
2180                                hashcpy(actual_sha1_return, actual_sha1);
2181                        return buffer;
2182                }
2183                /* Handle references */
2184                else if (type == OBJ_COMMIT)
2185                        ref_type = "tree ";
2186                else if (type == OBJ_TAG)
2187                        ref_type = "object ";
2188                else {
2189                        free(buffer);
2190                        return NULL;
2191                }
2192                ref_length = strlen(ref_type);
2193
2194                if (ref_length + 40 > isize ||
2195                    memcmp(buffer, ref_type, ref_length) ||
2196                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2197                        free(buffer);
2198                        return NULL;
2199                }
2200                free(buffer);
2201                /* Now we have the ID of the referred-to object in
2202                 * actual_sha1.  Check again. */
2203        }
2204}
2205
2206static void write_sha1_file_prepare(const void *buf, unsigned long len,
2207                                    const char *type, unsigned char *sha1,
2208                                    char *hdr, int *hdrlen)
2209{
2210        git_SHA_CTX c;
2211
2212        /* Generate the header */
2213        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2214
2215        /* Sha1.. */
2216        git_SHA1_Init(&c);
2217        git_SHA1_Update(&c, hdr, *hdrlen);
2218        git_SHA1_Update(&c, buf, len);
2219        git_SHA1_Final(sha1, &c);
2220}
2221
2222/*
2223 * Move the just written object into its final resting place.
2224 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2225 * "moving" is only a part of what it does, when no patch between
2226 * master to pu changes the call sites of this function.
2227 */
2228int move_temp_to_file(const char *tmpfile, const char *filename)
2229{
2230        int ret = 0;
2231
2232        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2233                goto try_rename;
2234        else if (link(tmpfile, filename))
2235                ret = errno;
2236
2237        /*
2238         * Coda hack - coda doesn't like cross-directory links,
2239         * so we fall back to a rename, which will mean that it
2240         * won't be able to check collisions, but that's not a
2241         * big deal.
2242         *
2243         * The same holds for FAT formatted media.
2244         *
2245         * When this succeeds, we just return.  We have nothing
2246         * left to unlink.
2247         */
2248        if (ret && ret != EEXIST) {
2249        try_rename:
2250                if (!rename(tmpfile, filename))
2251                        goto out;
2252                ret = errno;
2253        }
2254        unlink_or_warn(tmpfile);
2255        if (ret) {
2256                if (ret != EEXIST) {
2257                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2258                }
2259                /* FIXME!!! Collision check here ? */
2260        }
2261
2262out:
2263        if (set_shared_perm(filename, (S_IFREG|0444)))
2264                return error("unable to set permission to '%s'", filename);
2265        return 0;
2266}
2267
2268static int write_buffer(int fd, const void *buf, size_t len)
2269{
2270        if (write_in_full(fd, buf, len) < 0)
2271                return error("file write error (%s)", strerror(errno));
2272        return 0;
2273}
2274
2275int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2276                   unsigned char *sha1)
2277{
2278        char hdr[32];
2279        int hdrlen;
2280        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2281        return 0;
2282}
2283
2284/* Finalize a file on disk, and close it. */
2285static void close_sha1_file(int fd)
2286{
2287        if (fsync_object_files)
2288                fsync_or_die(fd, "sha1 file");
2289        if (close(fd) != 0)
2290                die("error when closing sha1 file (%s)", strerror(errno));
2291}
2292
2293/* Size of directory component, including the ending '/' */
2294static inline int directory_size(const char *filename)
2295{
2296        const char *s = strrchr(filename, '/');
2297        if (!s)
2298                return 0;
2299        return s - filename + 1;
2300}
2301
2302/*
2303 * This creates a temporary file in the same directory as the final
2304 * 'filename'
2305 *
2306 * We want to avoid cross-directory filename renames, because those
2307 * can have problems on various filesystems (FAT, NFS, Coda).
2308 */
2309static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2310{
2311        int fd, dirlen = directory_size(filename);
2312
2313        if (dirlen + 20 > bufsiz) {
2314                errno = ENAMETOOLONG;
2315                return -1;
2316        }
2317        memcpy(buffer, filename, dirlen);
2318        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2319        fd = mkstemp(buffer);
2320        if (fd < 0 && dirlen && errno == ENOENT) {
2321                /* Make sure the directory exists */
2322                memcpy(buffer, filename, dirlen);
2323                buffer[dirlen-1] = 0;
2324                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2325                        return -1;
2326
2327                /* Try again */
2328                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2329                fd = mkstemp(buffer);
2330        }
2331        return fd;
2332}
2333
2334static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2335                              void *buf, unsigned long len, time_t mtime)
2336{
2337        int fd, ret;
2338        size_t size;
2339        unsigned char *compressed;
2340        z_stream stream;
2341        char *filename;
2342        static char tmpfile[PATH_MAX];
2343
2344        filename = sha1_file_name(sha1);
2345        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2346        while (fd < 0 && errno == EMFILE && unuse_one_window(packed_git, -1))
2347                fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2348        if (fd < 0) {
2349                if (errno == EACCES)
2350                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2351                else
2352                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2353        }
2354
2355        /* Set it up */
2356        memset(&stream, 0, sizeof(stream));
2357        deflateInit(&stream, zlib_compression_level);
2358        size = 8 + deflateBound(&stream, len+hdrlen);
2359        compressed = xmalloc(size);
2360
2361        /* Compress it */
2362        stream.next_out = compressed;
2363        stream.avail_out = size;
2364
2365        /* First header.. */
2366        stream.next_in = (unsigned char *)hdr;
2367        stream.avail_in = hdrlen;
2368        while (deflate(&stream, 0) == Z_OK)
2369                /* nothing */;
2370
2371        /* Then the data itself.. */
2372        stream.next_in = buf;
2373        stream.avail_in = len;
2374        ret = deflate(&stream, Z_FINISH);
2375        if (ret != Z_STREAM_END)
2376                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2377
2378        ret = deflateEnd(&stream);
2379        if (ret != Z_OK)
2380                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2381
2382        size = stream.total_out;
2383
2384        if (write_buffer(fd, compressed, size) < 0)
2385                die("unable to write sha1 file");
2386        close_sha1_file(fd);
2387        free(compressed);
2388
2389        if (mtime) {
2390                struct utimbuf utb;
2391                utb.actime = mtime;
2392                utb.modtime = mtime;
2393                if (utime(tmpfile, &utb) < 0)
2394                        warning("failed utime() on %s: %s",
2395                                tmpfile, strerror(errno));
2396        }
2397
2398        return move_temp_to_file(tmpfile, filename);
2399}
2400
2401int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2402{
2403        unsigned char sha1[20];
2404        char hdr[32];
2405        int hdrlen;
2406
2407        /* Normally if we have it in the pack then we do not bother writing
2408         * it out into .git/objects/??/?{38} file.
2409         */
2410        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2411        if (returnsha1)
2412                hashcpy(returnsha1, sha1);
2413        if (has_sha1_file(sha1))
2414                return 0;
2415        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2416}
2417
2418int force_object_loose(const unsigned char *sha1, time_t mtime)
2419{
2420        void *buf;
2421        unsigned long len;
2422        enum object_type type;
2423        char hdr[32];
2424        int hdrlen;
2425        int ret;
2426
2427        if (has_loose_object(sha1))
2428                return 0;
2429        buf = read_packed_sha1(sha1, &type, &len);
2430        if (!buf)
2431                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2432        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2433        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2434        free(buf);
2435
2436        return ret;
2437}
2438
2439int has_pack_index(const unsigned char *sha1)
2440{
2441        struct stat st;
2442        if (stat(sha1_pack_index_name(sha1), &st))
2443                return 0;
2444        return 1;
2445}
2446
2447int has_pack_file(const unsigned char *sha1)
2448{
2449        struct stat st;
2450        if (stat(sha1_pack_name(sha1), &st))
2451                return 0;
2452        return 1;
2453}
2454
2455int has_sha1_pack(const unsigned char *sha1)
2456{
2457        struct pack_entry e;
2458        return find_pack_entry(sha1, &e);
2459}
2460
2461int has_sha1_file(const unsigned char *sha1)
2462{
2463        struct pack_entry e;
2464
2465        if (find_pack_entry(sha1, &e))
2466                return 1;
2467        return has_loose_object(sha1);
2468}
2469
2470static int index_mem(unsigned char *sha1, void *buf, size_t size,
2471                     int write_object, enum object_type type, const char *path)
2472{
2473        int ret, re_allocated = 0;
2474
2475        if (!type)
2476                type = OBJ_BLOB;
2477
2478        /*
2479         * Convert blobs to git internal format
2480         */
2481        if ((type == OBJ_BLOB) && path) {
2482                struct strbuf nbuf = STRBUF_INIT;
2483                if (convert_to_git(path, buf, size, &nbuf,
2484                                   write_object ? safe_crlf : 0)) {
2485                        buf = strbuf_detach(&nbuf, &size);
2486                        re_allocated = 1;
2487                }
2488        }
2489
2490        if (write_object)
2491                ret = write_sha1_file(buf, size, typename(type), sha1);
2492        else
2493                ret = hash_sha1_file(buf, size, typename(type), sha1);
2494        if (re_allocated)
2495                free(buf);
2496        return ret;
2497}
2498
2499int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2500             enum object_type type, const char *path)
2501{
2502        int ret;
2503        size_t size = xsize_t(st->st_size);
2504
2505        if (!S_ISREG(st->st_mode)) {
2506                struct strbuf sbuf = STRBUF_INIT;
2507                if (strbuf_read(&sbuf, fd, 4096) >= 0)
2508                        ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2509                                        type, path);
2510                else
2511                        ret = -1;
2512                strbuf_release(&sbuf);
2513        } else if (size) {
2514                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2515                ret = index_mem(sha1, buf, size, write_object, type, path);
2516                munmap(buf, size);
2517        } else
2518                ret = index_mem(sha1, NULL, size, write_object, type, path);
2519        close(fd);
2520        return ret;
2521}
2522
2523int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2524{
2525        int fd;
2526        struct strbuf sb = STRBUF_INIT;
2527
2528        switch (st->st_mode & S_IFMT) {
2529        case S_IFREG:
2530                fd = open(path, O_RDONLY);
2531                if (fd < 0)
2532                        return error("open(\"%s\"): %s", path,
2533                                     strerror(errno));
2534                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2535                        return error("%s: failed to insert into database",
2536                                     path);
2537                break;
2538        case S_IFLNK:
2539                if (strbuf_readlink(&sb, path, st->st_size)) {
2540                        char *errstr = strerror(errno);
2541                        return error("readlink(\"%s\"): %s", path,
2542                                     errstr);
2543                }
2544                if (!write_object)
2545                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2546                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2547                        return error("%s: failed to insert into database",
2548                                     path);
2549                strbuf_release(&sb);
2550                break;
2551        case S_IFDIR:
2552                return resolve_gitlink_ref(path, "HEAD", sha1);
2553        default:
2554                return error("%s: unsupported file type", path);
2555        }
2556        return 0;
2557}
2558
2559int read_pack_header(int fd, struct pack_header *header)
2560{
2561        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2562                /* "eof before pack header was fully read" */
2563                return PH_ERROR_EOF;
2564
2565        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2566                /* "protocol error (pack signature mismatch detected)" */
2567                return PH_ERROR_PACK_SIGNATURE;
2568        if (!pack_version_ok(header->hdr_version))
2569                /* "protocol error (pack version unsupported)" */
2570                return PH_ERROR_PROTOCOL;
2571        return 0;
2572}