sha1_file.con commit Merge branch 'jk/remote-pushremote-config-reading' into maint (4097a25)
   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 "string-list.h"
  11#include "delta.h"
  12#include "pack.h"
  13#include "blob.h"
  14#include "commit.h"
  15#include "run-command.h"
  16#include "tag.h"
  17#include "tree.h"
  18#include "tree-walk.h"
  19#include "refs.h"
  20#include "pack-revindex.h"
  21#include "sha1-lookup.h"
  22#include "bulk-checkin.h"
  23#include "streaming.h"
  24#include "dir.h"
  25
  26#ifndef O_NOATIME
  27#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  28#define O_NOATIME 01000000
  29#else
  30#define O_NOATIME 0
  31#endif
  32#endif
  33
  34#define SZ_FMT PRIuMAX
  35static inline uintmax_t sz_fmt(size_t s) { return s; }
  36
  37const unsigned char null_sha1[20];
  38
  39static const char *no_log_pack_access = "no_log_pack_access";
  40static const char *log_pack_access;
  41
  42/*
  43 * This is meant to hold a *small* number of objects that you would
  44 * want read_sha1_file() to be able to return, but yet you do not want
  45 * to write them into the object store (e.g. a browse-only
  46 * application).
  47 */
  48static struct cached_object {
  49        unsigned char sha1[20];
  50        enum object_type type;
  51        void *buf;
  52        unsigned long size;
  53} *cached_objects;
  54static int cached_object_nr, cached_object_alloc;
  55
  56static struct cached_object empty_tree = {
  57        EMPTY_TREE_SHA1_BIN_LITERAL,
  58        OBJ_TREE,
  59        "",
  60        0
  61};
  62
  63static struct packed_git *last_found_pack;
  64
  65static struct cached_object *find_cached_object(const unsigned char *sha1)
  66{
  67        int i;
  68        struct cached_object *co = cached_objects;
  69
  70        for (i = 0; i < cached_object_nr; i++, co++) {
  71                if (!hashcmp(co->sha1, sha1))
  72                        return co;
  73        }
  74        if (!hashcmp(sha1, empty_tree.sha1))
  75                return &empty_tree;
  76        return NULL;
  77}
  78
  79int mkdir_in_gitdir(const char *path)
  80{
  81        if (mkdir(path, 0777)) {
  82                int saved_errno = errno;
  83                struct stat st;
  84                struct strbuf sb = STRBUF_INIT;
  85
  86                if (errno != EEXIST)
  87                        return -1;
  88                /*
  89                 * Are we looking at a path in a symlinked worktree
  90                 * whose original repository does not yet have it?
  91                 * e.g. .git/rr-cache pointing at its original
  92                 * repository in which the user hasn't performed any
  93                 * conflict resolution yet?
  94                 */
  95                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
  96                    strbuf_readlink(&sb, path, st.st_size) ||
  97                    !is_absolute_path(sb.buf) ||
  98                    mkdir(sb.buf, 0777)) {
  99                        strbuf_release(&sb);
 100                        errno = saved_errno;
 101                        return -1;
 102                }
 103                strbuf_release(&sb);
 104        }
 105        return adjust_shared_perm(path);
 106}
 107
 108enum scld_error safe_create_leading_directories(char *path)
 109{
 110        char *next_component = path + offset_1st_component(path);
 111        enum scld_error ret = SCLD_OK;
 112
 113        while (ret == SCLD_OK && next_component) {
 114                struct stat st;
 115                char *slash = next_component, slash_character;
 116
 117                while (*slash && !is_dir_sep(*slash))
 118                        slash++;
 119
 120                if (!*slash)
 121                        break;
 122
 123                next_component = slash + 1;
 124                while (is_dir_sep(*next_component))
 125                        next_component++;
 126                if (!*next_component)
 127                        break;
 128
 129                slash_character = *slash;
 130                *slash = '\0';
 131                if (!stat(path, &st)) {
 132                        /* path exists */
 133                        if (!S_ISDIR(st.st_mode))
 134                                ret = SCLD_EXISTS;
 135                } else if (mkdir(path, 0777)) {
 136                        if (errno == EEXIST &&
 137                            !stat(path, &st) && S_ISDIR(st.st_mode))
 138                                ; /* somebody created it since we checked */
 139                        else if (errno == ENOENT)
 140                                /*
 141                                 * Either mkdir() failed because
 142                                 * somebody just pruned the containing
 143                                 * directory, or stat() failed because
 144                                 * the file that was in our way was
 145                                 * just removed.  Either way, inform
 146                                 * the caller that it might be worth
 147                                 * trying again:
 148                                 */
 149                                ret = SCLD_VANISHED;
 150                        else
 151                                ret = SCLD_FAILED;
 152                } else if (adjust_shared_perm(path)) {
 153                        ret = SCLD_PERMS;
 154                }
 155                *slash = slash_character;
 156        }
 157        return ret;
 158}
 159
 160enum scld_error safe_create_leading_directories_const(const char *path)
 161{
 162        /* path points to cache entries, so xstrdup before messing with it */
 163        char *buf = xstrdup(path);
 164        enum scld_error result = safe_create_leading_directories(buf);
 165        free(buf);
 166        return result;
 167}
 168
 169static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 170{
 171        int i;
 172        for (i = 0; i < 20; i++) {
 173                static char hex[] = "0123456789abcdef";
 174                unsigned int val = sha1[i];
 175                char *pos = pathbuf + i*2 + (i > 0);
 176                *pos++ = hex[val >> 4];
 177                *pos = hex[val & 0xf];
 178        }
 179}
 180
 181/*
 182 * NOTE! This returns a statically allocated buffer, so you have to be
 183 * careful about using it. Do an "xstrdup()" if you need to save the
 184 * filename.
 185 *
 186 * Also note that this returns the location for creating.  Reading
 187 * SHA1 file can happen from any alternate directory listed in the
 188 * DB_ENVIRONMENT environment variable if it is not found in
 189 * the primary object database.
 190 */
 191char *sha1_file_name(const unsigned char *sha1)
 192{
 193        static char buf[PATH_MAX];
 194        const char *objdir;
 195        int len;
 196
 197        objdir = get_object_directory();
 198        len = strlen(objdir);
 199
 200        /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
 201        if (len + 43 > PATH_MAX)
 202                die("insanely long object directory %s", objdir);
 203        memcpy(buf, objdir, len);
 204        buf[len] = '/';
 205        buf[len+3] = '/';
 206        buf[len+42] = '\0';
 207        fill_sha1_path(buf + len + 1, sha1);
 208        return buf;
 209}
 210
 211static char *sha1_get_pack_name(const unsigned char *sha1,
 212                                char **name, char **base, const char *which)
 213{
 214        static const char hex[] = "0123456789abcdef";
 215        char *buf;
 216        int i;
 217
 218        if (!*base) {
 219                const char *sha1_file_directory = get_object_directory();
 220                int len = strlen(sha1_file_directory);
 221                *base = xmalloc(len + 60);
 222                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 223                        sha1_file_directory, which);
 224                *name = *base + len + 11;
 225        }
 226
 227        buf = *name;
 228
 229        for (i = 0; i < 20; i++) {
 230                unsigned int val = *sha1++;
 231                *buf++ = hex[val >> 4];
 232                *buf++ = hex[val & 0xf];
 233        }
 234
 235        return *base;
 236}
 237
 238char *sha1_pack_name(const unsigned char *sha1)
 239{
 240        static char *name, *base;
 241
 242        return sha1_get_pack_name(sha1, &name, &base, "pack");
 243}
 244
 245char *sha1_pack_index_name(const unsigned char *sha1)
 246{
 247        static char *name, *base;
 248
 249        return sha1_get_pack_name(sha1, &name, &base, "idx");
 250}
 251
 252struct alternate_object_database *alt_odb_list;
 253static struct alternate_object_database **alt_odb_tail;
 254
 255static int git_open_noatime(const char *name);
 256
 257/*
 258 * Prepare alternate object database registry.
 259 *
 260 * The variable alt_odb_list points at the list of struct
 261 * alternate_object_database.  The elements on this list come from
 262 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 263 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 264 * whose contents is similar to that environment variable but can be
 265 * LF separated.  Its base points at a statically allocated buffer that
 266 * contains "/the/directory/corresponding/to/.git/objects/...", while
 267 * its name points just after the slash at the end of ".git/objects/"
 268 * in the example above, and has enough space to hold 40-byte hex
 269 * SHA1, an extra slash for the first level indirection, and the
 270 * terminating NUL.
 271 */
 272static int link_alt_odb_entry(const char *entry, const char *relative_base, int depth)
 273{
 274        const char *objdir = get_object_directory();
 275        struct alternate_object_database *ent;
 276        struct alternate_object_database *alt;
 277        int pfxlen, entlen;
 278        struct strbuf pathbuf = STRBUF_INIT;
 279
 280        if (!is_absolute_path(entry) && relative_base) {
 281                strbuf_addstr(&pathbuf, real_path(relative_base));
 282                strbuf_addch(&pathbuf, '/');
 283        }
 284        strbuf_addstr(&pathbuf, entry);
 285
 286        normalize_path_copy(pathbuf.buf, pathbuf.buf);
 287
 288        pfxlen = strlen(pathbuf.buf);
 289
 290        /*
 291         * The trailing slash after the directory name is given by
 292         * this function at the end. Remove duplicates.
 293         */
 294        while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
 295                pfxlen -= 1;
 296
 297        entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
 298        ent = xmalloc(sizeof(*ent) + entlen);
 299        memcpy(ent->base, pathbuf.buf, pfxlen);
 300        strbuf_release(&pathbuf);
 301
 302        ent->name = ent->base + pfxlen + 1;
 303        ent->base[pfxlen + 3] = '/';
 304        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 305
 306        /* Detect cases where alternate disappeared */
 307        if (!is_directory(ent->base)) {
 308                error("object directory %s does not exist; "
 309                      "check .git/objects/info/alternates.",
 310                      ent->base);
 311                free(ent);
 312                return -1;
 313        }
 314
 315        /* Prevent the common mistake of listing the same
 316         * thing twice, or object directory itself.
 317         */
 318        for (alt = alt_odb_list; alt; alt = alt->next) {
 319                if (!memcmp(ent->base, alt->base, pfxlen)) {
 320                        free(ent);
 321                        return -1;
 322                }
 323        }
 324        if (!strcmp(ent->base, objdir)) {
 325                free(ent);
 326                return -1;
 327        }
 328
 329        /* add the alternate entry */
 330        *alt_odb_tail = ent;
 331        alt_odb_tail = &(ent->next);
 332        ent->next = NULL;
 333
 334        /* recursively add alternates */
 335        read_info_alternates(ent->base, depth + 1);
 336
 337        ent->base[pfxlen] = '/';
 338
 339        return 0;
 340}
 341
 342static void link_alt_odb_entries(const char *alt, int len, int sep,
 343                                 const char *relative_base, int depth)
 344{
 345        struct string_list entries = STRING_LIST_INIT_NODUP;
 346        char *alt_copy;
 347        int i;
 348
 349        if (depth > 5) {
 350                error("%s: ignoring alternate object stores, nesting too deep.",
 351                                relative_base);
 352                return;
 353        }
 354
 355        alt_copy = xmemdupz(alt, len);
 356        string_list_split_in_place(&entries, alt_copy, sep, -1);
 357        for (i = 0; i < entries.nr; i++) {
 358                const char *entry = entries.items[i].string;
 359                if (entry[0] == '\0' || entry[0] == '#')
 360                        continue;
 361                if (!is_absolute_path(entry) && depth) {
 362                        error("%s: ignoring relative alternate object store %s",
 363                                        relative_base, entry);
 364                } else {
 365                        link_alt_odb_entry(entry, relative_base, depth);
 366                }
 367        }
 368        string_list_clear(&entries, 0);
 369        free(alt_copy);
 370}
 371
 372void read_info_alternates(const char * relative_base, int depth)
 373{
 374        char *map;
 375        size_t mapsz;
 376        struct stat st;
 377        const char alt_file_name[] = "info/alternates";
 378        /* Given that relative_base is no longer than PATH_MAX,
 379           ensure that "path" has enough space to append "/", the
 380           file name, "info/alternates", and a trailing NUL.  */
 381        char path[PATH_MAX + 1 + sizeof alt_file_name];
 382        int fd;
 383
 384        sprintf(path, "%s/%s", relative_base, alt_file_name);
 385        fd = git_open_noatime(path);
 386        if (fd < 0)
 387                return;
 388        if (fstat(fd, &st) || (st.st_size == 0)) {
 389                close(fd);
 390                return;
 391        }
 392        mapsz = xsize_t(st.st_size);
 393        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 394        close(fd);
 395
 396        link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
 397
 398        munmap(map, mapsz);
 399}
 400
 401void add_to_alternates_file(const char *reference)
 402{
 403        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 404        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
 405        char *alt = mkpath("%s\n", reference);
 406        write_or_die(fd, alt, strlen(alt));
 407        if (commit_lock_file(lock))
 408                die("could not close alternates file");
 409        if (alt_odb_tail)
 410                link_alt_odb_entries(alt, strlen(alt), '\n', NULL, 0);
 411}
 412
 413void foreach_alt_odb(alt_odb_fn fn, void *cb)
 414{
 415        struct alternate_object_database *ent;
 416
 417        prepare_alt_odb();
 418        for (ent = alt_odb_list; ent; ent = ent->next)
 419                if (fn(ent, cb))
 420                        return;
 421}
 422
 423void prepare_alt_odb(void)
 424{
 425        const char *alt;
 426
 427        if (alt_odb_tail)
 428                return;
 429
 430        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 431        if (!alt) alt = "";
 432
 433        alt_odb_tail = &alt_odb_list;
 434        link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
 435
 436        read_info_alternates(get_object_directory(), 0);
 437}
 438
 439static int has_loose_object_local(const unsigned char *sha1)
 440{
 441        char *name = sha1_file_name(sha1);
 442        return !access(name, F_OK);
 443}
 444
 445int has_loose_object_nonlocal(const unsigned char *sha1)
 446{
 447        struct alternate_object_database *alt;
 448        prepare_alt_odb();
 449        for (alt = alt_odb_list; alt; alt = alt->next) {
 450                fill_sha1_path(alt->name, sha1);
 451                if (!access(alt->base, F_OK))
 452                        return 1;
 453        }
 454        return 0;
 455}
 456
 457static int has_loose_object(const unsigned char *sha1)
 458{
 459        return has_loose_object_local(sha1) ||
 460               has_loose_object_nonlocal(sha1);
 461}
 462
 463static unsigned int pack_used_ctr;
 464static unsigned int pack_mmap_calls;
 465static unsigned int peak_pack_open_windows;
 466static unsigned int pack_open_windows;
 467static unsigned int pack_open_fds;
 468static unsigned int pack_max_fds;
 469static size_t peak_pack_mapped;
 470static size_t pack_mapped;
 471struct packed_git *packed_git;
 472
 473void pack_report(void)
 474{
 475        fprintf(stderr,
 476                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 477                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 478                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 479                sz_fmt(getpagesize()),
 480                sz_fmt(packed_git_window_size),
 481                sz_fmt(packed_git_limit));
 482        fprintf(stderr,
 483                "pack_report: pack_used_ctr            = %10u\n"
 484                "pack_report: pack_mmap_calls          = %10u\n"
 485                "pack_report: pack_open_windows        = %10u / %10u\n"
 486                "pack_report: pack_mapped              = "
 487                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 488                pack_used_ctr,
 489                pack_mmap_calls,
 490                pack_open_windows, peak_pack_open_windows,
 491                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 492}
 493
 494static int check_packed_git_idx(const char *path,  struct packed_git *p)
 495{
 496        void *idx_map;
 497        struct pack_idx_header *hdr;
 498        size_t idx_size;
 499        uint32_t version, nr, i, *index;
 500        int fd = git_open_noatime(path);
 501        struct stat st;
 502
 503        if (fd < 0)
 504                return -1;
 505        if (fstat(fd, &st)) {
 506                close(fd);
 507                return -1;
 508        }
 509        idx_size = xsize_t(st.st_size);
 510        if (idx_size < 4 * 256 + 20 + 20) {
 511                close(fd);
 512                return error("index file %s is too small", path);
 513        }
 514        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 515        close(fd);
 516
 517        hdr = idx_map;
 518        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 519                version = ntohl(hdr->idx_version);
 520                if (version < 2 || version > 2) {
 521                        munmap(idx_map, idx_size);
 522                        return error("index file %s is version %"PRIu32
 523                                     " and is not supported by this binary"
 524                                     " (try upgrading GIT to a newer version)",
 525                                     path, version);
 526                }
 527        } else
 528                version = 1;
 529
 530        nr = 0;
 531        index = idx_map;
 532        if (version > 1)
 533                index += 2;  /* skip index header */
 534        for (i = 0; i < 256; i++) {
 535                uint32_t n = ntohl(index[i]);
 536                if (n < nr) {
 537                        munmap(idx_map, idx_size);
 538                        return error("non-monotonic index %s", path);
 539                }
 540                nr = n;
 541        }
 542
 543        if (version == 1) {
 544                /*
 545                 * Total size:
 546                 *  - 256 index entries 4 bytes each
 547                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 548                 *  - 20-byte SHA1 of the packfile
 549                 *  - 20-byte SHA1 file checksum
 550                 */
 551                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 552                        munmap(idx_map, idx_size);
 553                        return error("wrong index v1 file size in %s", path);
 554                }
 555        } else if (version == 2) {
 556                /*
 557                 * Minimum size:
 558                 *  - 8 bytes of header
 559                 *  - 256 index entries 4 bytes each
 560                 *  - 20-byte sha1 entry * nr
 561                 *  - 4-byte crc entry * nr
 562                 *  - 4-byte offset entry * nr
 563                 *  - 20-byte SHA1 of the packfile
 564                 *  - 20-byte SHA1 file checksum
 565                 * And after the 4-byte offset table might be a
 566                 * variable sized table containing 8-byte entries
 567                 * for offsets larger than 2^31.
 568                 */
 569                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 570                unsigned long max_size = min_size;
 571                if (nr)
 572                        max_size += (nr - 1)*8;
 573                if (idx_size < min_size || idx_size > max_size) {
 574                        munmap(idx_map, idx_size);
 575                        return error("wrong index v2 file size in %s", path);
 576                }
 577                if (idx_size != min_size &&
 578                    /*
 579                     * make sure we can deal with large pack offsets.
 580                     * 31-bit signed offset won't be enough, neither
 581                     * 32-bit unsigned one will be.
 582                     */
 583                    (sizeof(off_t) <= 4)) {
 584                        munmap(idx_map, idx_size);
 585                        return error("pack too large for current definition of off_t in %s", path);
 586                }
 587        }
 588
 589        p->index_version = version;
 590        p->index_data = idx_map;
 591        p->index_size = idx_size;
 592        p->num_objects = nr;
 593        return 0;
 594}
 595
 596int open_pack_index(struct packed_git *p)
 597{
 598        char *idx_name;
 599        int ret;
 600
 601        if (p->index_data)
 602                return 0;
 603
 604        idx_name = xstrdup(p->pack_name);
 605        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 606        ret = check_packed_git_idx(idx_name, p);
 607        free(idx_name);
 608        return ret;
 609}
 610
 611static void scan_windows(struct packed_git *p,
 612        struct packed_git **lru_p,
 613        struct pack_window **lru_w,
 614        struct pack_window **lru_l)
 615{
 616        struct pack_window *w, *w_l;
 617
 618        for (w_l = NULL, w = p->windows; w; w = w->next) {
 619                if (!w->inuse_cnt) {
 620                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 621                                *lru_p = p;
 622                                *lru_w = w;
 623                                *lru_l = w_l;
 624                        }
 625                }
 626                w_l = w;
 627        }
 628}
 629
 630static int unuse_one_window(struct packed_git *current)
 631{
 632        struct packed_git *p, *lru_p = NULL;
 633        struct pack_window *lru_w = NULL, *lru_l = NULL;
 634
 635        if (current)
 636                scan_windows(current, &lru_p, &lru_w, &lru_l);
 637        for (p = packed_git; p; p = p->next)
 638                scan_windows(p, &lru_p, &lru_w, &lru_l);
 639        if (lru_p) {
 640                munmap(lru_w->base, lru_w->len);
 641                pack_mapped -= lru_w->len;
 642                if (lru_l)
 643                        lru_l->next = lru_w->next;
 644                else
 645                        lru_p->windows = lru_w->next;
 646                free(lru_w);
 647                pack_open_windows--;
 648                return 1;
 649        }
 650        return 0;
 651}
 652
 653void release_pack_memory(size_t need)
 654{
 655        size_t cur = pack_mapped;
 656        while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
 657                ; /* nothing */
 658}
 659
 660void *xmmap(void *start, size_t length,
 661        int prot, int flags, int fd, off_t offset)
 662{
 663        void *ret = mmap(start, length, prot, flags, fd, offset);
 664        if (ret == MAP_FAILED) {
 665                if (!length)
 666                        return NULL;
 667                release_pack_memory(length);
 668                ret = mmap(start, length, prot, flags, fd, offset);
 669                if (ret == MAP_FAILED)
 670                        die_errno("Out of memory? mmap failed");
 671        }
 672        return ret;
 673}
 674
 675void close_pack_windows(struct packed_git *p)
 676{
 677        while (p->windows) {
 678                struct pack_window *w = p->windows;
 679
 680                if (w->inuse_cnt)
 681                        die("pack '%s' still has open windows to it",
 682                            p->pack_name);
 683                munmap(w->base, w->len);
 684                pack_mapped -= w->len;
 685                pack_open_windows--;
 686                p->windows = w->next;
 687                free(w);
 688        }
 689}
 690
 691/*
 692 * The LRU pack is the one with the oldest MRU window, preferring packs
 693 * with no used windows, or the oldest mtime if it has no windows allocated.
 694 */
 695static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
 696{
 697        struct pack_window *w, *this_mru_w;
 698        int has_windows_inuse = 0;
 699
 700        /*
 701         * Reject this pack if it has windows and the previously selected
 702         * one does not.  If this pack does not have windows, reject
 703         * it if the pack file is newer than the previously selected one.
 704         */
 705        if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
 706                return;
 707
 708        for (w = this_mru_w = p->windows; w; w = w->next) {
 709                /*
 710                 * Reject this pack if any of its windows are in use,
 711                 * but the previously selected pack did not have any
 712                 * inuse windows.  Otherwise, record that this pack
 713                 * has windows in use.
 714                 */
 715                if (w->inuse_cnt) {
 716                        if (*accept_windows_inuse)
 717                                has_windows_inuse = 1;
 718                        else
 719                                return;
 720                }
 721
 722                if (w->last_used > this_mru_w->last_used)
 723                        this_mru_w = w;
 724
 725                /*
 726                 * Reject this pack if it has windows that have been
 727                 * used more recently than the previously selected pack.
 728                 * If the previously selected pack had windows inuse and
 729                 * we have not encountered a window in this pack that is
 730                 * inuse, skip this check since we prefer a pack with no
 731                 * inuse windows to one that has inuse windows.
 732                 */
 733                if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
 734                    this_mru_w->last_used > (*mru_w)->last_used)
 735                        return;
 736        }
 737
 738        /*
 739         * Select this pack.
 740         */
 741        *mru_w = this_mru_w;
 742        *lru_p = p;
 743        *accept_windows_inuse = has_windows_inuse;
 744}
 745
 746static int close_one_pack(void)
 747{
 748        struct packed_git *p, *lru_p = NULL;
 749        struct pack_window *mru_w = NULL;
 750        int accept_windows_inuse = 1;
 751
 752        for (p = packed_git; p; p = p->next) {
 753                if (p->pack_fd == -1)
 754                        continue;
 755                find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
 756        }
 757
 758        if (lru_p) {
 759                close(lru_p->pack_fd);
 760                pack_open_fds--;
 761                lru_p->pack_fd = -1;
 762                return 1;
 763        }
 764
 765        return 0;
 766}
 767
 768void unuse_pack(struct pack_window **w_cursor)
 769{
 770        struct pack_window *w = *w_cursor;
 771        if (w) {
 772                w->inuse_cnt--;
 773                *w_cursor = NULL;
 774        }
 775}
 776
 777void close_pack_index(struct packed_git *p)
 778{
 779        if (p->index_data) {
 780                munmap((void *)p->index_data, p->index_size);
 781                p->index_data = NULL;
 782        }
 783}
 784
 785/*
 786 * This is used by git-repack in case a newly created pack happens to
 787 * contain the same set of objects as an existing one.  In that case
 788 * the resulting file might be different even if its name would be the
 789 * same.  It is best to close any reference to the old pack before it is
 790 * replaced on disk.  Of course no index pointers nor windows for given pack
 791 * must subsist at this point.  If ever objects from this pack are requested
 792 * again, the new version of the pack will be reinitialized through
 793 * reprepare_packed_git().
 794 */
 795void free_pack_by_name(const char *pack_name)
 796{
 797        struct packed_git *p, **pp = &packed_git;
 798
 799        while (*pp) {
 800                p = *pp;
 801                if (strcmp(pack_name, p->pack_name) == 0) {
 802                        clear_delta_base_cache();
 803                        close_pack_windows(p);
 804                        if (p->pack_fd != -1) {
 805                                close(p->pack_fd);
 806                                pack_open_fds--;
 807                        }
 808                        close_pack_index(p);
 809                        free(p->bad_object_sha1);
 810                        *pp = p->next;
 811                        if (last_found_pack == p)
 812                                last_found_pack = NULL;
 813                        free(p);
 814                        return;
 815                }
 816                pp = &p->next;
 817        }
 818}
 819
 820static unsigned int get_max_fd_limit(void)
 821{
 822#ifdef RLIMIT_NOFILE
 823        {
 824                struct rlimit lim;
 825
 826                if (!getrlimit(RLIMIT_NOFILE, &lim))
 827                        return lim.rlim_cur;
 828        }
 829#endif
 830
 831#ifdef _SC_OPEN_MAX
 832        {
 833                long open_max = sysconf(_SC_OPEN_MAX);
 834                if (0 < open_max)
 835                        return open_max;
 836                /*
 837                 * Otherwise, we got -1 for one of the two
 838                 * reasons:
 839                 *
 840                 * (1) sysconf() did not understand _SC_OPEN_MAX
 841                 *     and signaled an error with -1; or
 842                 * (2) sysconf() said there is no limit.
 843                 *
 844                 * We _could_ clear errno before calling sysconf() to
 845                 * tell these two cases apart and return a huge number
 846                 * in the latter case to let the caller cap it to a
 847                 * value that is not so selfish, but letting the
 848                 * fallback OPEN_MAX codepath take care of these cases
 849                 * is a lot simpler.
 850                 */
 851        }
 852#endif
 853
 854#ifdef OPEN_MAX
 855        return OPEN_MAX;
 856#else
 857        return 1; /* see the caller ;-) */
 858#endif
 859}
 860
 861/*
 862 * Do not call this directly as this leaks p->pack_fd on error return;
 863 * call open_packed_git() instead.
 864 */
 865static int open_packed_git_1(struct packed_git *p)
 866{
 867        struct stat st;
 868        struct pack_header hdr;
 869        unsigned char sha1[20];
 870        unsigned char *idx_sha1;
 871        long fd_flag;
 872
 873        if (!p->index_data && open_pack_index(p))
 874                return error("packfile %s index unavailable", p->pack_name);
 875
 876        if (!pack_max_fds) {
 877                unsigned int max_fds = get_max_fd_limit();
 878
 879                /* Save 3 for stdin/stdout/stderr, 22 for work */
 880                if (25 < max_fds)
 881                        pack_max_fds = max_fds - 25;
 882                else
 883                        pack_max_fds = 1;
 884        }
 885
 886        while (pack_max_fds <= pack_open_fds && close_one_pack())
 887                ; /* nothing */
 888
 889        p->pack_fd = git_open_noatime(p->pack_name);
 890        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 891                return -1;
 892        pack_open_fds++;
 893
 894        /* If we created the struct before we had the pack we lack size. */
 895        if (!p->pack_size) {
 896                if (!S_ISREG(st.st_mode))
 897                        return error("packfile %s not a regular file", p->pack_name);
 898                p->pack_size = st.st_size;
 899        } else if (p->pack_size != st.st_size)
 900                return error("packfile %s size changed", p->pack_name);
 901
 902        /* We leave these file descriptors open with sliding mmap;
 903         * there is no point keeping them open across exec(), though.
 904         */
 905        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 906        if (fd_flag < 0)
 907                return error("cannot determine file descriptor flags");
 908        fd_flag |= FD_CLOEXEC;
 909        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 910                return error("cannot set FD_CLOEXEC");
 911
 912        /* Verify we recognize this pack file format. */
 913        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 914                return error("file %s is far too short to be a packfile", p->pack_name);
 915        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 916                return error("file %s is not a GIT packfile", p->pack_name);
 917        if (!pack_version_ok(hdr.hdr_version))
 918                return error("packfile %s is version %"PRIu32" and not"
 919                        " supported (try upgrading GIT to a newer version)",
 920                        p->pack_name, ntohl(hdr.hdr_version));
 921
 922        /* Verify the pack matches its index. */
 923        if (p->num_objects != ntohl(hdr.hdr_entries))
 924                return error("packfile %s claims to have %"PRIu32" objects"
 925                             " while index indicates %"PRIu32" objects",
 926                             p->pack_name, ntohl(hdr.hdr_entries),
 927                             p->num_objects);
 928        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 929                return error("end of packfile %s is unavailable", p->pack_name);
 930        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 931                return error("packfile %s signature is unavailable", p->pack_name);
 932        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 933        if (hashcmp(sha1, idx_sha1))
 934                return error("packfile %s does not match index", p->pack_name);
 935        return 0;
 936}
 937
 938static int open_packed_git(struct packed_git *p)
 939{
 940        if (!open_packed_git_1(p))
 941                return 0;
 942        if (p->pack_fd != -1) {
 943                close(p->pack_fd);
 944                pack_open_fds--;
 945                p->pack_fd = -1;
 946        }
 947        return -1;
 948}
 949
 950static int in_window(struct pack_window *win, off_t offset)
 951{
 952        /* We must promise at least 20 bytes (one hash) after the
 953         * offset is available from this window, otherwise the offset
 954         * is not actually in this window and a different window (which
 955         * has that one hash excess) must be used.  This is to support
 956         * the object header and delta base parsing routines below.
 957         */
 958        off_t win_off = win->offset;
 959        return win_off <= offset
 960                && (offset + 20) <= (win_off + win->len);
 961}
 962
 963unsigned char *use_pack(struct packed_git *p,
 964                struct pack_window **w_cursor,
 965                off_t offset,
 966                unsigned long *left)
 967{
 968        struct pack_window *win = *w_cursor;
 969
 970        /* Since packfiles end in a hash of their content and it's
 971         * pointless to ask for an offset into the middle of that
 972         * hash, and the in_window function above wouldn't match
 973         * don't allow an offset too close to the end of the file.
 974         */
 975        if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
 976                die("packfile %s cannot be accessed", p->pack_name);
 977        if (offset > (p->pack_size - 20))
 978                die("offset beyond end of packfile (truncated pack?)");
 979
 980        if (!win || !in_window(win, offset)) {
 981                if (win)
 982                        win->inuse_cnt--;
 983                for (win = p->windows; win; win = win->next) {
 984                        if (in_window(win, offset))
 985                                break;
 986                }
 987                if (!win) {
 988                        size_t window_align = packed_git_window_size / 2;
 989                        off_t len;
 990
 991                        if (p->pack_fd == -1 && open_packed_git(p))
 992                                die("packfile %s cannot be accessed", p->pack_name);
 993
 994                        win = xcalloc(1, sizeof(*win));
 995                        win->offset = (offset / window_align) * window_align;
 996                        len = p->pack_size - win->offset;
 997                        if (len > packed_git_window_size)
 998                                len = packed_git_window_size;
 999                        win->len = (size_t)len;
1000                        pack_mapped += win->len;
1001                        while (packed_git_limit < pack_mapped
1002                                && unuse_one_window(p))
1003                                ; /* nothing */
1004                        win->base = xmmap(NULL, win->len,
1005                                PROT_READ, MAP_PRIVATE,
1006                                p->pack_fd, win->offset);
1007                        if (win->base == MAP_FAILED)
1008                                die("packfile %s cannot be mapped: %s",
1009                                        p->pack_name,
1010                                        strerror(errno));
1011                        if (!win->offset && win->len == p->pack_size
1012                                && !p->do_not_close) {
1013                                close(p->pack_fd);
1014                                pack_open_fds--;
1015                                p->pack_fd = -1;
1016                        }
1017                        pack_mmap_calls++;
1018                        pack_open_windows++;
1019                        if (pack_mapped > peak_pack_mapped)
1020                                peak_pack_mapped = pack_mapped;
1021                        if (pack_open_windows > peak_pack_open_windows)
1022                                peak_pack_open_windows = pack_open_windows;
1023                        win->next = p->windows;
1024                        p->windows = win;
1025                }
1026        }
1027        if (win != *w_cursor) {
1028                win->last_used = pack_used_ctr++;
1029                win->inuse_cnt++;
1030                *w_cursor = win;
1031        }
1032        offset -= win->offset;
1033        if (left)
1034                *left = win->len - xsize_t(offset);
1035        return win->base + offset;
1036}
1037
1038static struct packed_git *alloc_packed_git(int extra)
1039{
1040        struct packed_git *p = xmalloc(sizeof(*p) + extra);
1041        memset(p, 0, sizeof(*p));
1042        p->pack_fd = -1;
1043        return p;
1044}
1045
1046static void try_to_free_pack_memory(size_t size)
1047{
1048        release_pack_memory(size);
1049}
1050
1051struct packed_git *add_packed_git(const char *path, int path_len, int local)
1052{
1053        static int have_set_try_to_free_routine;
1054        struct stat st;
1055        struct packed_git *p = alloc_packed_git(path_len + 2);
1056
1057        if (!have_set_try_to_free_routine) {
1058                have_set_try_to_free_routine = 1;
1059                set_try_to_free_routine(try_to_free_pack_memory);
1060        }
1061
1062        /*
1063         * Make sure a corresponding .pack file exists and that
1064         * the index looks sane.
1065         */
1066        path_len -= strlen(".idx");
1067        if (path_len < 1) {
1068                free(p);
1069                return NULL;
1070        }
1071        memcpy(p->pack_name, path, path_len);
1072
1073        strcpy(p->pack_name + path_len, ".keep");
1074        if (!access(p->pack_name, F_OK))
1075                p->pack_keep = 1;
1076
1077        strcpy(p->pack_name + path_len, ".pack");
1078        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1079                free(p);
1080                return NULL;
1081        }
1082
1083        /* ok, it looks sane as far as we can check without
1084         * actually mapping the pack file.
1085         */
1086        p->pack_size = st.st_size;
1087        p->pack_local = local;
1088        p->mtime = st.st_mtime;
1089        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1090                hashclr(p->sha1);
1091        return p;
1092}
1093
1094struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1095{
1096        const char *path = sha1_pack_name(sha1);
1097        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
1098
1099        strcpy(p->pack_name, path);
1100        hashcpy(p->sha1, sha1);
1101        if (check_packed_git_idx(idx_path, p)) {
1102                free(p);
1103                return NULL;
1104        }
1105
1106        return p;
1107}
1108
1109void install_packed_git(struct packed_git *pack)
1110{
1111        if (pack->pack_fd != -1)
1112                pack_open_fds++;
1113
1114        pack->next = packed_git;
1115        packed_git = pack;
1116}
1117
1118void (*report_garbage)(const char *desc, const char *path);
1119
1120static void report_helper(const struct string_list *list,
1121                          int seen_bits, int first, int last)
1122{
1123        const char *msg;
1124        switch (seen_bits) {
1125        case 0:
1126                msg = "no corresponding .idx nor .pack";
1127                break;
1128        case 1:
1129                msg = "no corresponding .idx";
1130                break;
1131        case 2:
1132                msg = "no corresponding .pack";
1133                break;
1134        default:
1135                return;
1136        }
1137        for (; first < last; first++)
1138                report_garbage(msg, list->items[first].string);
1139}
1140
1141static void report_pack_garbage(struct string_list *list)
1142{
1143        int i, baselen = -1, first = 0, seen_bits = 0;
1144
1145        if (!report_garbage)
1146                return;
1147
1148        sort_string_list(list);
1149
1150        for (i = 0; i < list->nr; i++) {
1151                const char *path = list->items[i].string;
1152                if (baselen != -1 &&
1153                    strncmp(path, list->items[first].string, baselen)) {
1154                        report_helper(list, seen_bits, first, i);
1155                        baselen = -1;
1156                        seen_bits = 0;
1157                }
1158                if (baselen == -1) {
1159                        const char *dot = strrchr(path, '.');
1160                        if (!dot) {
1161                                report_garbage("garbage found", path);
1162                                continue;
1163                        }
1164                        baselen = dot - path + 1;
1165                        first = i;
1166                }
1167                if (!strcmp(path + baselen, "pack"))
1168                        seen_bits |= 1;
1169                else if (!strcmp(path + baselen, "idx"))
1170                        seen_bits |= 2;
1171        }
1172        report_helper(list, seen_bits, first, list->nr);
1173}
1174
1175static void prepare_packed_git_one(char *objdir, int local)
1176{
1177        /* Ensure that this buffer is large enough so that we can
1178           append "/pack/" without clobbering the stack even if
1179           strlen(objdir) were PATH_MAX.  */
1180        char path[PATH_MAX + 1 + 4 + 1 + 1];
1181        int len;
1182        DIR *dir;
1183        struct dirent *de;
1184        struct string_list garbage = STRING_LIST_INIT_DUP;
1185
1186        sprintf(path, "%s/pack", objdir);
1187        len = strlen(path);
1188        dir = opendir(path);
1189        if (!dir) {
1190                if (errno != ENOENT)
1191                        error("unable to open object pack directory: %s: %s",
1192                              path, strerror(errno));
1193                return;
1194        }
1195        path[len++] = '/';
1196        while ((de = readdir(dir)) != NULL) {
1197                int namelen = strlen(de->d_name);
1198                struct packed_git *p;
1199
1200                if (len + namelen + 1 > sizeof(path)) {
1201                        if (report_garbage) {
1202                                struct strbuf sb = STRBUF_INIT;
1203                                strbuf_addf(&sb, "%.*s/%s", len - 1, path, de->d_name);
1204                                report_garbage("path too long", sb.buf);
1205                                strbuf_release(&sb);
1206                        }
1207                        continue;
1208                }
1209
1210                if (is_dot_or_dotdot(de->d_name))
1211                        continue;
1212
1213                strcpy(path + len, de->d_name);
1214
1215                if (has_extension(de->d_name, ".idx")) {
1216                        /* Don't reopen a pack we already have. */
1217                        for (p = packed_git; p; p = p->next) {
1218                                if (!memcmp(path, p->pack_name, len + namelen - 4))
1219                                        break;
1220                        }
1221                        if (p == NULL &&
1222                            /*
1223                             * See if it really is a valid .idx file with
1224                             * corresponding .pack file that we can map.
1225                             */
1226                            (p = add_packed_git(path, len + namelen, local)) != NULL)
1227                                install_packed_git(p);
1228                }
1229
1230                if (!report_garbage)
1231                        continue;
1232
1233                if (has_extension(de->d_name, ".idx") ||
1234                    has_extension(de->d_name, ".pack") ||
1235                    has_extension(de->d_name, ".keep"))
1236                        string_list_append(&garbage, path);
1237                else
1238                        report_garbage("garbage found", path);
1239        }
1240        closedir(dir);
1241        report_pack_garbage(&garbage);
1242        string_list_clear(&garbage, 0);
1243}
1244
1245static int sort_pack(const void *a_, const void *b_)
1246{
1247        struct packed_git *a = *((struct packed_git **)a_);
1248        struct packed_git *b = *((struct packed_git **)b_);
1249        int st;
1250
1251        /*
1252         * Local packs tend to contain objects specific to our
1253         * variant of the project than remote ones.  In addition,
1254         * remote ones could be on a network mounted filesystem.
1255         * Favor local ones for these reasons.
1256         */
1257        st = a->pack_local - b->pack_local;
1258        if (st)
1259                return -st;
1260
1261        /*
1262         * Younger packs tend to contain more recent objects,
1263         * and more recent objects tend to get accessed more
1264         * often.
1265         */
1266        if (a->mtime < b->mtime)
1267                return 1;
1268        else if (a->mtime == b->mtime)
1269                return 0;
1270        return -1;
1271}
1272
1273static void rearrange_packed_git(void)
1274{
1275        struct packed_git **ary, *p;
1276        int i, n;
1277
1278        for (n = 0, p = packed_git; p; p = p->next)
1279                n++;
1280        if (n < 2)
1281                return;
1282
1283        /* prepare an array of packed_git for easier sorting */
1284        ary = xcalloc(n, sizeof(struct packed_git *));
1285        for (n = 0, p = packed_git; p; p = p->next)
1286                ary[n++] = p;
1287
1288        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1289
1290        /* link them back again */
1291        for (i = 0; i < n - 1; i++)
1292                ary[i]->next = ary[i + 1];
1293        ary[n - 1]->next = NULL;
1294        packed_git = ary[0];
1295
1296        free(ary);
1297}
1298
1299static int prepare_packed_git_run_once = 0;
1300void prepare_packed_git(void)
1301{
1302        struct alternate_object_database *alt;
1303
1304        if (prepare_packed_git_run_once)
1305                return;
1306        prepare_packed_git_one(get_object_directory(), 1);
1307        prepare_alt_odb();
1308        for (alt = alt_odb_list; alt; alt = alt->next) {
1309                alt->name[-1] = 0;
1310                prepare_packed_git_one(alt->base, 0);
1311                alt->name[-1] = '/';
1312        }
1313        rearrange_packed_git();
1314        prepare_packed_git_run_once = 1;
1315}
1316
1317void reprepare_packed_git(void)
1318{
1319        discard_revindex();
1320        prepare_packed_git_run_once = 0;
1321        prepare_packed_git();
1322}
1323
1324static void mark_bad_packed_object(struct packed_git *p,
1325                                   const unsigned char *sha1)
1326{
1327        unsigned i;
1328        for (i = 0; i < p->num_bad_objects; i++)
1329                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1330                        return;
1331        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1332        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1333        p->num_bad_objects++;
1334}
1335
1336static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1337{
1338        struct packed_git *p;
1339        unsigned i;
1340
1341        for (p = packed_git; p; p = p->next)
1342                for (i = 0; i < p->num_bad_objects; i++)
1343                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1344                                return p;
1345        return NULL;
1346}
1347
1348/*
1349 * With an in-core object data in "map", rehash it to make sure the
1350 * object name actually matches "sha1" to detect object corruption.
1351 * With "map" == NULL, try reading the object named with "sha1" using
1352 * the streaming interface and rehash it to do the same.
1353 */
1354int check_sha1_signature(const unsigned char *sha1, void *map,
1355                         unsigned long size, const char *type)
1356{
1357        unsigned char real_sha1[20];
1358        enum object_type obj_type;
1359        struct git_istream *st;
1360        git_SHA_CTX c;
1361        char hdr[32];
1362        int hdrlen;
1363
1364        if (map) {
1365                hash_sha1_file(map, size, type, real_sha1);
1366                return hashcmp(sha1, real_sha1) ? -1 : 0;
1367        }
1368
1369        st = open_istream(sha1, &obj_type, &size, NULL);
1370        if (!st)
1371                return -1;
1372
1373        /* Generate the header */
1374        hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1375
1376        /* Sha1.. */
1377        git_SHA1_Init(&c);
1378        git_SHA1_Update(&c, hdr, hdrlen);
1379        for (;;) {
1380                char buf[1024 * 16];
1381                ssize_t readlen = read_istream(st, buf, sizeof(buf));
1382
1383                if (readlen < 0) {
1384                        close_istream(st);
1385                        return -1;
1386                }
1387                if (!readlen)
1388                        break;
1389                git_SHA1_Update(&c, buf, readlen);
1390        }
1391        git_SHA1_Final(real_sha1, &c);
1392        close_istream(st);
1393        return hashcmp(sha1, real_sha1) ? -1 : 0;
1394}
1395
1396static int git_open_noatime(const char *name)
1397{
1398        static int sha1_file_open_flag = O_NOATIME;
1399
1400        for (;;) {
1401                int fd = open(name, O_RDONLY | sha1_file_open_flag);
1402                if (fd >= 0)
1403                        return fd;
1404
1405                /* Might the failure be due to O_NOATIME? */
1406                if (errno != ENOENT && sha1_file_open_flag) {
1407                        sha1_file_open_flag = 0;
1408                        continue;
1409                }
1410
1411                return -1;
1412        }
1413}
1414
1415static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1416{
1417        char *name = sha1_file_name(sha1);
1418        struct alternate_object_database *alt;
1419
1420        if (!lstat(name, st))
1421                return 0;
1422
1423        prepare_alt_odb();
1424        errno = ENOENT;
1425        for (alt = alt_odb_list; alt; alt = alt->next) {
1426                name = alt->name;
1427                fill_sha1_path(name, sha1);
1428                if (!lstat(alt->base, st))
1429                        return 0;
1430        }
1431
1432        return -1;
1433}
1434
1435static int open_sha1_file(const unsigned char *sha1)
1436{
1437        int fd;
1438        char *name = sha1_file_name(sha1);
1439        struct alternate_object_database *alt;
1440
1441        fd = git_open_noatime(name);
1442        if (fd >= 0)
1443                return fd;
1444
1445        prepare_alt_odb();
1446        errno = ENOENT;
1447        for (alt = alt_odb_list; alt; alt = alt->next) {
1448                name = alt->name;
1449                fill_sha1_path(name, sha1);
1450                fd = git_open_noatime(alt->base);
1451                if (fd >= 0)
1452                        return fd;
1453        }
1454        return -1;
1455}
1456
1457void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1458{
1459        void *map;
1460        int fd;
1461
1462        fd = open_sha1_file(sha1);
1463        map = NULL;
1464        if (fd >= 0) {
1465                struct stat st;
1466
1467                if (!fstat(fd, &st)) {
1468                        *size = xsize_t(st.st_size);
1469                        if (!*size) {
1470                                /* mmap() is forbidden on empty files */
1471                                error("object file %s is empty", sha1_file_name(sha1));
1472                                return NULL;
1473                        }
1474                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1475                }
1476                close(fd);
1477        }
1478        return map;
1479}
1480
1481unsigned long unpack_object_header_buffer(const unsigned char *buf,
1482                unsigned long len, enum object_type *type, unsigned long *sizep)
1483{
1484        unsigned shift;
1485        unsigned long size, c;
1486        unsigned long used = 0;
1487
1488        c = buf[used++];
1489        *type = (c >> 4) & 7;
1490        size = c & 15;
1491        shift = 4;
1492        while (c & 0x80) {
1493                if (len <= used || bitsizeof(long) <= shift) {
1494                        error("bad object header");
1495                        size = used = 0;
1496                        break;
1497                }
1498                c = buf[used++];
1499                size += (c & 0x7f) << shift;
1500                shift += 7;
1501        }
1502        *sizep = size;
1503        return used;
1504}
1505
1506int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1507{
1508        /* Get the data stream */
1509        memset(stream, 0, sizeof(*stream));
1510        stream->next_in = map;
1511        stream->avail_in = mapsize;
1512        stream->next_out = buffer;
1513        stream->avail_out = bufsiz;
1514
1515        git_inflate_init(stream);
1516        return git_inflate(stream, 0);
1517}
1518
1519static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1520{
1521        int bytes = strlen(buffer) + 1;
1522        unsigned char *buf = xmallocz(size);
1523        unsigned long n;
1524        int status = Z_OK;
1525
1526        n = stream->total_out - bytes;
1527        if (n > size)
1528                n = size;
1529        memcpy(buf, (char *) buffer + bytes, n);
1530        bytes = n;
1531        if (bytes <= size) {
1532                /*
1533                 * The above condition must be (bytes <= size), not
1534                 * (bytes < size).  In other words, even though we
1535                 * expect no more output and set avail_out to zero,
1536                 * the input zlib stream may have bytes that express
1537                 * "this concludes the stream", and we *do* want to
1538                 * eat that input.
1539                 *
1540                 * Otherwise we would not be able to test that we
1541                 * consumed all the input to reach the expected size;
1542                 * we also want to check that zlib tells us that all
1543                 * went well with status == Z_STREAM_END at the end.
1544                 */
1545                stream->next_out = buf + bytes;
1546                stream->avail_out = size - bytes;
1547                while (status == Z_OK)
1548                        status = git_inflate(stream, Z_FINISH);
1549        }
1550        if (status == Z_STREAM_END && !stream->avail_in) {
1551                git_inflate_end(stream);
1552                return buf;
1553        }
1554
1555        if (status < 0)
1556                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1557        else if (stream->avail_in)
1558                error("garbage at end of loose object '%s'",
1559                      sha1_to_hex(sha1));
1560        free(buf);
1561        return NULL;
1562}
1563
1564/*
1565 * We used to just use "sscanf()", but that's actually way
1566 * too permissive for what we want to check. So do an anal
1567 * object header parse by hand.
1568 */
1569int parse_sha1_header(const char *hdr, unsigned long *sizep)
1570{
1571        char type[10];
1572        int i;
1573        unsigned long size;
1574
1575        /*
1576         * The type can be at most ten bytes (including the
1577         * terminating '\0' that we add), and is followed by
1578         * a space.
1579         */
1580        i = 0;
1581        for (;;) {
1582                char c = *hdr++;
1583                if (c == ' ')
1584                        break;
1585                type[i++] = c;
1586                if (i >= sizeof(type))
1587                        return -1;
1588        }
1589        type[i] = 0;
1590
1591        /*
1592         * The length must follow immediately, and be in canonical
1593         * decimal format (ie "010" is not valid).
1594         */
1595        size = *hdr++ - '0';
1596        if (size > 9)
1597                return -1;
1598        if (size) {
1599                for (;;) {
1600                        unsigned long c = *hdr - '0';
1601                        if (c > 9)
1602                                break;
1603                        hdr++;
1604                        size = size * 10 + c;
1605                }
1606        }
1607        *sizep = size;
1608
1609        /*
1610         * The length must be followed by a zero byte
1611         */
1612        return *hdr ? -1 : type_from_string(type);
1613}
1614
1615static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1616{
1617        int ret;
1618        git_zstream stream;
1619        char hdr[8192];
1620
1621        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1622        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1623                return NULL;
1624
1625        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1626}
1627
1628unsigned long get_size_from_delta(struct packed_git *p,
1629                                  struct pack_window **w_curs,
1630                                  off_t curpos)
1631{
1632        const unsigned char *data;
1633        unsigned char delta_head[20], *in;
1634        git_zstream stream;
1635        int st;
1636
1637        memset(&stream, 0, sizeof(stream));
1638        stream.next_out = delta_head;
1639        stream.avail_out = sizeof(delta_head);
1640
1641        git_inflate_init(&stream);
1642        do {
1643                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1644                stream.next_in = in;
1645                st = git_inflate(&stream, Z_FINISH);
1646                curpos += stream.next_in - in;
1647        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1648                 stream.total_out < sizeof(delta_head));
1649        git_inflate_end(&stream);
1650        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1651                error("delta data unpack-initial failed");
1652                return 0;
1653        }
1654
1655        /* Examine the initial part of the delta to figure out
1656         * the result size.
1657         */
1658        data = delta_head;
1659
1660        /* ignore base size */
1661        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1662
1663        /* Read the result size */
1664        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1665}
1666
1667static off_t get_delta_base(struct packed_git *p,
1668                                    struct pack_window **w_curs,
1669                                    off_t *curpos,
1670                                    enum object_type type,
1671                                    off_t delta_obj_offset)
1672{
1673        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1674        off_t base_offset;
1675
1676        /* use_pack() assured us we have [base_info, base_info + 20)
1677         * as a range that we can look at without walking off the
1678         * end of the mapped window.  Its actually the hash size
1679         * that is assured.  An OFS_DELTA longer than the hash size
1680         * is stupid, as then a REF_DELTA would be smaller to store.
1681         */
1682        if (type == OBJ_OFS_DELTA) {
1683                unsigned used = 0;
1684                unsigned char c = base_info[used++];
1685                base_offset = c & 127;
1686                while (c & 128) {
1687                        base_offset += 1;
1688                        if (!base_offset || MSB(base_offset, 7))
1689                                return 0;  /* overflow */
1690                        c = base_info[used++];
1691                        base_offset = (base_offset << 7) + (c & 127);
1692                }
1693                base_offset = delta_obj_offset - base_offset;
1694                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1695                        return 0;  /* out of bound */
1696                *curpos += used;
1697        } else if (type == OBJ_REF_DELTA) {
1698                /* The base entry _must_ be in the same pack */
1699                base_offset = find_pack_entry_one(base_info, p);
1700                *curpos += 20;
1701        } else
1702                die("I am totally screwed");
1703        return base_offset;
1704}
1705
1706/*
1707 * Like get_delta_base above, but we return the sha1 instead of the pack
1708 * offset. This means it is cheaper for REF deltas (we do not have to do
1709 * the final object lookup), but more expensive for OFS deltas (we
1710 * have to load the revidx to convert the offset back into a sha1).
1711 */
1712static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1713                                                struct pack_window **w_curs,
1714                                                off_t curpos,
1715                                                enum object_type type,
1716                                                off_t delta_obj_offset)
1717{
1718        if (type == OBJ_REF_DELTA) {
1719                unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1720                return base;
1721        } else if (type == OBJ_OFS_DELTA) {
1722                struct revindex_entry *revidx;
1723                off_t base_offset = get_delta_base(p, w_curs, &curpos,
1724                                                   type, delta_obj_offset);
1725
1726                if (!base_offset)
1727                        return NULL;
1728
1729                revidx = find_pack_revindex(p, base_offset);
1730                if (!revidx)
1731                        return NULL;
1732
1733                return nth_packed_object_sha1(p, revidx->nr);
1734        } else
1735                return NULL;
1736}
1737
1738int unpack_object_header(struct packed_git *p,
1739                         struct pack_window **w_curs,
1740                         off_t *curpos,
1741                         unsigned long *sizep)
1742{
1743        unsigned char *base;
1744        unsigned long left;
1745        unsigned long used;
1746        enum object_type type;
1747
1748        /* use_pack() assures us we have [base, base + 20) available
1749         * as a range that we can look at.  (Its actually the hash
1750         * size that is assured.)  With our object header encoding
1751         * the maximum deflated object size is 2^137, which is just
1752         * insane, so we know won't exceed what we have been given.
1753         */
1754        base = use_pack(p, w_curs, *curpos, &left);
1755        used = unpack_object_header_buffer(base, left, &type, sizep);
1756        if (!used) {
1757                type = OBJ_BAD;
1758        } else
1759                *curpos += used;
1760
1761        return type;
1762}
1763
1764static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1765{
1766        int type;
1767        struct revindex_entry *revidx;
1768        const unsigned char *sha1;
1769        revidx = find_pack_revindex(p, obj_offset);
1770        if (!revidx)
1771                return OBJ_BAD;
1772        sha1 = nth_packed_object_sha1(p, revidx->nr);
1773        mark_bad_packed_object(p, sha1);
1774        type = sha1_object_info(sha1, NULL);
1775        if (type <= OBJ_NONE)
1776                return OBJ_BAD;
1777        return type;
1778}
1779
1780#define POI_STACK_PREALLOC 64
1781
1782static enum object_type packed_to_object_type(struct packed_git *p,
1783                                              off_t obj_offset,
1784                                              enum object_type type,
1785                                              struct pack_window **w_curs,
1786                                              off_t curpos)
1787{
1788        off_t small_poi_stack[POI_STACK_PREALLOC];
1789        off_t *poi_stack = small_poi_stack;
1790        int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1791
1792        while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1793                off_t base_offset;
1794                unsigned long size;
1795                /* Push the object we're going to leave behind */
1796                if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1797                        poi_stack_alloc = alloc_nr(poi_stack_nr);
1798                        poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1799                        memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1800                } else {
1801                        ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1802                }
1803                poi_stack[poi_stack_nr++] = obj_offset;
1804                /* If parsing the base offset fails, just unwind */
1805                base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1806                if (!base_offset)
1807                        goto unwind;
1808                curpos = obj_offset = base_offset;
1809                type = unpack_object_header(p, w_curs, &curpos, &size);
1810                if (type <= OBJ_NONE) {
1811                        /* If getting the base itself fails, we first
1812                         * retry the base, otherwise unwind */
1813                        type = retry_bad_packed_offset(p, base_offset);
1814                        if (type > OBJ_NONE)
1815                                goto out;
1816                        goto unwind;
1817                }
1818        }
1819
1820        switch (type) {
1821        case OBJ_BAD:
1822        case OBJ_COMMIT:
1823        case OBJ_TREE:
1824        case OBJ_BLOB:
1825        case OBJ_TAG:
1826                break;
1827        default:
1828                error("unknown object type %i at offset %"PRIuMAX" in %s",
1829                      type, (uintmax_t)obj_offset, p->pack_name);
1830                type = OBJ_BAD;
1831        }
1832
1833out:
1834        if (poi_stack != small_poi_stack)
1835                free(poi_stack);
1836        return type;
1837
1838unwind:
1839        while (poi_stack_nr) {
1840                obj_offset = poi_stack[--poi_stack_nr];
1841                type = retry_bad_packed_offset(p, obj_offset);
1842                if (type > OBJ_NONE)
1843                        goto out;
1844        }
1845        type = OBJ_BAD;
1846        goto out;
1847}
1848
1849static int packed_object_info(struct packed_git *p, off_t obj_offset,
1850                              struct object_info *oi)
1851{
1852        struct pack_window *w_curs = NULL;
1853        unsigned long size;
1854        off_t curpos = obj_offset;
1855        enum object_type type;
1856
1857        /*
1858         * We always get the representation type, but only convert it to
1859         * a "real" type later if the caller is interested.
1860         */
1861        type = unpack_object_header(p, &w_curs, &curpos, &size);
1862
1863        if (oi->sizep) {
1864                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1865                        off_t tmp_pos = curpos;
1866                        off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1867                                                           type, obj_offset);
1868                        if (!base_offset) {
1869                                type = OBJ_BAD;
1870                                goto out;
1871                        }
1872                        *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1873                        if (*oi->sizep == 0) {
1874                                type = OBJ_BAD;
1875                                goto out;
1876                        }
1877                } else {
1878                        *oi->sizep = size;
1879                }
1880        }
1881
1882        if (oi->disk_sizep) {
1883                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1884                *oi->disk_sizep = revidx[1].offset - obj_offset;
1885        }
1886
1887        if (oi->typep) {
1888                *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
1889                if (*oi->typep < 0) {
1890                        type = OBJ_BAD;
1891                        goto out;
1892                }
1893        }
1894
1895        if (oi->delta_base_sha1) {
1896                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1897                        const unsigned char *base;
1898
1899                        base = get_delta_base_sha1(p, &w_curs, curpos,
1900                                                   type, obj_offset);
1901                        if (!base) {
1902                                type = OBJ_BAD;
1903                                goto out;
1904                        }
1905
1906                        hashcpy(oi->delta_base_sha1, base);
1907                } else
1908                        hashclr(oi->delta_base_sha1);
1909        }
1910
1911out:
1912        unuse_pack(&w_curs);
1913        return type;
1914}
1915
1916static void *unpack_compressed_entry(struct packed_git *p,
1917                                    struct pack_window **w_curs,
1918                                    off_t curpos,
1919                                    unsigned long size)
1920{
1921        int st;
1922        git_zstream stream;
1923        unsigned char *buffer, *in;
1924
1925        buffer = xmallocz(size);
1926        memset(&stream, 0, sizeof(stream));
1927        stream.next_out = buffer;
1928        stream.avail_out = size + 1;
1929
1930        git_inflate_init(&stream);
1931        do {
1932                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1933                stream.next_in = in;
1934                st = git_inflate(&stream, Z_FINISH);
1935                if (!stream.avail_out)
1936                        break; /* the payload is larger than it should be */
1937                curpos += stream.next_in - in;
1938        } while (st == Z_OK || st == Z_BUF_ERROR);
1939        git_inflate_end(&stream);
1940        if ((st != Z_STREAM_END) || stream.total_out != size) {
1941                free(buffer);
1942                return NULL;
1943        }
1944
1945        return buffer;
1946}
1947
1948#define MAX_DELTA_CACHE (256)
1949
1950static size_t delta_base_cached;
1951
1952static struct delta_base_cache_lru_list {
1953        struct delta_base_cache_lru_list *prev;
1954        struct delta_base_cache_lru_list *next;
1955} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1956
1957static struct delta_base_cache_entry {
1958        struct delta_base_cache_lru_list lru;
1959        void *data;
1960        struct packed_git *p;
1961        off_t base_offset;
1962        unsigned long size;
1963        enum object_type type;
1964} delta_base_cache[MAX_DELTA_CACHE];
1965
1966static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1967{
1968        unsigned long hash;
1969
1970        hash = (unsigned long)p + (unsigned long)base_offset;
1971        hash += (hash >> 8) + (hash >> 16);
1972        return hash % MAX_DELTA_CACHE;
1973}
1974
1975static struct delta_base_cache_entry *
1976get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1977{
1978        unsigned long hash = pack_entry_hash(p, base_offset);
1979        return delta_base_cache + hash;
1980}
1981
1982static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
1983                                     struct packed_git *p, off_t base_offset)
1984{
1985        return (ent->data && ent->p == p && ent->base_offset == base_offset);
1986}
1987
1988static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1989{
1990        struct delta_base_cache_entry *ent;
1991        ent = get_delta_base_cache_entry(p, base_offset);
1992        return eq_delta_base_cache_entry(ent, p, base_offset);
1993}
1994
1995static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1996{
1997        ent->data = NULL;
1998        ent->lru.next->prev = ent->lru.prev;
1999        ent->lru.prev->next = ent->lru.next;
2000        delta_base_cached -= ent->size;
2001}
2002
2003static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2004        unsigned long *base_size, enum object_type *type, int keep_cache)
2005{
2006        struct delta_base_cache_entry *ent;
2007        void *ret;
2008
2009        ent = get_delta_base_cache_entry(p, base_offset);
2010
2011        if (!eq_delta_base_cache_entry(ent, p, base_offset))
2012                return unpack_entry(p, base_offset, type, base_size);
2013
2014        ret = ent->data;
2015
2016        if (!keep_cache)
2017                clear_delta_base_cache_entry(ent);
2018        else
2019                ret = xmemdupz(ent->data, ent->size);
2020        *type = ent->type;
2021        *base_size = ent->size;
2022        return ret;
2023}
2024
2025static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2026{
2027        if (ent->data) {
2028                free(ent->data);
2029                ent->data = NULL;
2030                ent->lru.next->prev = ent->lru.prev;
2031                ent->lru.prev->next = ent->lru.next;
2032                delta_base_cached -= ent->size;
2033        }
2034}
2035
2036void clear_delta_base_cache(void)
2037{
2038        unsigned long p;
2039        for (p = 0; p < MAX_DELTA_CACHE; p++)
2040                release_delta_base_cache(&delta_base_cache[p]);
2041}
2042
2043static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2044        void *base, unsigned long base_size, enum object_type type)
2045{
2046        unsigned long hash = pack_entry_hash(p, base_offset);
2047        struct delta_base_cache_entry *ent = delta_base_cache + hash;
2048        struct delta_base_cache_lru_list *lru;
2049
2050        release_delta_base_cache(ent);
2051        delta_base_cached += base_size;
2052
2053        for (lru = delta_base_cache_lru.next;
2054             delta_base_cached > delta_base_cache_limit
2055             && lru != &delta_base_cache_lru;
2056             lru = lru->next) {
2057                struct delta_base_cache_entry *f = (void *)lru;
2058                if (f->type == OBJ_BLOB)
2059                        release_delta_base_cache(f);
2060        }
2061        for (lru = delta_base_cache_lru.next;
2062             delta_base_cached > delta_base_cache_limit
2063             && lru != &delta_base_cache_lru;
2064             lru = lru->next) {
2065                struct delta_base_cache_entry *f = (void *)lru;
2066                release_delta_base_cache(f);
2067        }
2068
2069        ent->p = p;
2070        ent->base_offset = base_offset;
2071        ent->type = type;
2072        ent->data = base;
2073        ent->size = base_size;
2074        ent->lru.next = &delta_base_cache_lru;
2075        ent->lru.prev = delta_base_cache_lru.prev;
2076        delta_base_cache_lru.prev->next = &ent->lru;
2077        delta_base_cache_lru.prev = &ent->lru;
2078}
2079
2080static void *read_object(const unsigned char *sha1, enum object_type *type,
2081                         unsigned long *size);
2082
2083static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2084{
2085        static FILE *log_file;
2086
2087        if (!log_pack_access)
2088                log_pack_access = getenv("GIT_TRACE_PACK_ACCESS");
2089        if (!log_pack_access)
2090                log_pack_access = no_log_pack_access;
2091        if (log_pack_access == no_log_pack_access)
2092                return;
2093
2094        if (!log_file) {
2095                log_file = fopen(log_pack_access, "w");
2096                if (!log_file) {
2097                        error("cannot open pack access log '%s' for writing: %s",
2098                              log_pack_access, strerror(errno));
2099                        log_pack_access = no_log_pack_access;
2100                        return;
2101                }
2102        }
2103        fprintf(log_file, "%s %"PRIuMAX"\n",
2104                p->pack_name, (uintmax_t)obj_offset);
2105        fflush(log_file);
2106}
2107
2108int do_check_packed_object_crc;
2109
2110#define UNPACK_ENTRY_STACK_PREALLOC 64
2111struct unpack_entry_stack_ent {
2112        off_t obj_offset;
2113        off_t curpos;
2114        unsigned long size;
2115};
2116
2117void *unpack_entry(struct packed_git *p, off_t obj_offset,
2118                   enum object_type *final_type, unsigned long *final_size)
2119{
2120        struct pack_window *w_curs = NULL;
2121        off_t curpos = obj_offset;
2122        void *data = NULL;
2123        unsigned long size;
2124        enum object_type type;
2125        struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2126        struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2127        int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2128        int base_from_cache = 0;
2129
2130        if (log_pack_access != no_log_pack_access)
2131                write_pack_access_log(p, obj_offset);
2132
2133        /* PHASE 1: drill down to the innermost base object */
2134        for (;;) {
2135                off_t base_offset;
2136                int i;
2137                struct delta_base_cache_entry *ent;
2138
2139                ent = get_delta_base_cache_entry(p, curpos);
2140                if (eq_delta_base_cache_entry(ent, p, curpos)) {
2141                        type = ent->type;
2142                        data = ent->data;
2143                        size = ent->size;
2144                        clear_delta_base_cache_entry(ent);
2145                        base_from_cache = 1;
2146                        break;
2147                }
2148
2149                if (do_check_packed_object_crc && p->index_version > 1) {
2150                        struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2151                        unsigned long len = revidx[1].offset - obj_offset;
2152                        if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2153                                const unsigned char *sha1 =
2154                                        nth_packed_object_sha1(p, revidx->nr);
2155                                error("bad packed object CRC for %s",
2156                                      sha1_to_hex(sha1));
2157                                mark_bad_packed_object(p, sha1);
2158                                unuse_pack(&w_curs);
2159                                return NULL;
2160                        }
2161                }
2162
2163                type = unpack_object_header(p, &w_curs, &curpos, &size);
2164                if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2165                        break;
2166
2167                base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2168                if (!base_offset) {
2169                        error("failed to validate delta base reference "
2170                              "at offset %"PRIuMAX" from %s",
2171                              (uintmax_t)curpos, p->pack_name);
2172                        /* bail to phase 2, in hopes of recovery */
2173                        data = NULL;
2174                        break;
2175                }
2176
2177                /* push object, proceed to base */
2178                if (delta_stack_nr >= delta_stack_alloc
2179                    && delta_stack == small_delta_stack) {
2180                        delta_stack_alloc = alloc_nr(delta_stack_nr);
2181                        delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2182                        memcpy(delta_stack, small_delta_stack,
2183                               sizeof(*delta_stack)*delta_stack_nr);
2184                } else {
2185                        ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2186                }
2187                i = delta_stack_nr++;
2188                delta_stack[i].obj_offset = obj_offset;
2189                delta_stack[i].curpos = curpos;
2190                delta_stack[i].size = size;
2191
2192                curpos = obj_offset = base_offset;
2193        }
2194
2195        /* PHASE 2: handle the base */
2196        switch (type) {
2197        case OBJ_OFS_DELTA:
2198        case OBJ_REF_DELTA:
2199                if (data)
2200                        die("BUG in unpack_entry: left loop at a valid delta");
2201                break;
2202        case OBJ_COMMIT:
2203        case OBJ_TREE:
2204        case OBJ_BLOB:
2205        case OBJ_TAG:
2206                if (!base_from_cache)
2207                        data = unpack_compressed_entry(p, &w_curs, curpos, size);
2208                break;
2209        default:
2210                data = NULL;
2211                error("unknown object type %i at offset %"PRIuMAX" in %s",
2212                      type, (uintmax_t)obj_offset, p->pack_name);
2213        }
2214
2215        /* PHASE 3: apply deltas in order */
2216
2217        /* invariants:
2218         *   'data' holds the base data, or NULL if there was corruption
2219         */
2220        while (delta_stack_nr) {
2221                void *delta_data;
2222                void *base = data;
2223                unsigned long delta_size, base_size = size;
2224                int i;
2225
2226                data = NULL;
2227
2228                if (base)
2229                        add_delta_base_cache(p, obj_offset, base, base_size, type);
2230
2231                if (!base) {
2232                        /*
2233                         * We're probably in deep shit, but let's try to fetch
2234                         * the required base anyway from another pack or loose.
2235                         * This is costly but should happen only in the presence
2236                         * of a corrupted pack, and is better than failing outright.
2237                         */
2238                        struct revindex_entry *revidx;
2239                        const unsigned char *base_sha1;
2240                        revidx = find_pack_revindex(p, obj_offset);
2241                        if (revidx) {
2242                                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2243                                error("failed to read delta base object %s"
2244                                      " at offset %"PRIuMAX" from %s",
2245                                      sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2246                                      p->pack_name);
2247                                mark_bad_packed_object(p, base_sha1);
2248                                base = read_object(base_sha1, &type, &base_size);
2249                        }
2250                }
2251
2252                i = --delta_stack_nr;
2253                obj_offset = delta_stack[i].obj_offset;
2254                curpos = delta_stack[i].curpos;
2255                delta_size = delta_stack[i].size;
2256
2257                if (!base)
2258                        continue;
2259
2260                delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2261
2262                if (!delta_data) {
2263                        error("failed to unpack compressed delta "
2264                              "at offset %"PRIuMAX" from %s",
2265                              (uintmax_t)curpos, p->pack_name);
2266                        data = NULL;
2267                        continue;
2268                }
2269
2270                data = patch_delta(base, base_size,
2271                                   delta_data, delta_size,
2272                                   &size);
2273
2274                /*
2275                 * We could not apply the delta; warn the user, but keep going.
2276                 * Our failure will be noticed either in the next iteration of
2277                 * the loop, or if this is the final delta, in the caller when
2278                 * we return NULL. Those code paths will take care of making
2279                 * a more explicit warning and retrying with another copy of
2280                 * the object.
2281                 */
2282                if (!data)
2283                        error("failed to apply delta");
2284
2285                free(delta_data);
2286        }
2287
2288        *final_type = type;
2289        *final_size = size;
2290
2291        unuse_pack(&w_curs);
2292        return data;
2293}
2294
2295const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2296                                            uint32_t n)
2297{
2298        const unsigned char *index = p->index_data;
2299        if (!index) {
2300                if (open_pack_index(p))
2301                        return NULL;
2302                index = p->index_data;
2303        }
2304        if (n >= p->num_objects)
2305                return NULL;
2306        index += 4 * 256;
2307        if (p->index_version == 1) {
2308                return index + 24 * n + 4;
2309        } else {
2310                index += 8;
2311                return index + 20 * n;
2312        }
2313}
2314
2315off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2316{
2317        const unsigned char *index = p->index_data;
2318        index += 4 * 256;
2319        if (p->index_version == 1) {
2320                return ntohl(*((uint32_t *)(index + 24 * n)));
2321        } else {
2322                uint32_t off;
2323                index += 8 + p->num_objects * (20 + 4);
2324                off = ntohl(*((uint32_t *)(index + 4 * n)));
2325                if (!(off & 0x80000000))
2326                        return off;
2327                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2328                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2329                                   ntohl(*((uint32_t *)(index + 4)));
2330        }
2331}
2332
2333off_t find_pack_entry_one(const unsigned char *sha1,
2334                                  struct packed_git *p)
2335{
2336        const uint32_t *level1_ofs = p->index_data;
2337        const unsigned char *index = p->index_data;
2338        unsigned hi, lo, stride;
2339        static int use_lookup = -1;
2340        static int debug_lookup = -1;
2341
2342        if (debug_lookup < 0)
2343                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2344
2345        if (!index) {
2346                if (open_pack_index(p))
2347                        return 0;
2348                level1_ofs = p->index_data;
2349                index = p->index_data;
2350        }
2351        if (p->index_version > 1) {
2352                level1_ofs += 2;
2353                index += 8;
2354        }
2355        index += 4 * 256;
2356        hi = ntohl(level1_ofs[*sha1]);
2357        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2358        if (p->index_version > 1) {
2359                stride = 20;
2360        } else {
2361                stride = 24;
2362                index += 4;
2363        }
2364
2365        if (debug_lookup)
2366                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2367                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2368
2369        if (use_lookup < 0)
2370                use_lookup = !!getenv("GIT_USE_LOOKUP");
2371        if (use_lookup) {
2372                int pos = sha1_entry_pos(index, stride, 0,
2373                                         lo, hi, p->num_objects, sha1);
2374                if (pos < 0)
2375                        return 0;
2376                return nth_packed_object_offset(p, pos);
2377        }
2378
2379        do {
2380                unsigned mi = (lo + hi) / 2;
2381                int cmp = hashcmp(index + mi * stride, sha1);
2382
2383                if (debug_lookup)
2384                        printf("lo %u hi %u rg %u mi %u\n",
2385                               lo, hi, hi - lo, mi);
2386                if (!cmp)
2387                        return nth_packed_object_offset(p, mi);
2388                if (cmp > 0)
2389                        hi = mi;
2390                else
2391                        lo = mi+1;
2392        } while (lo < hi);
2393        return 0;
2394}
2395
2396int is_pack_valid(struct packed_git *p)
2397{
2398        /* An already open pack is known to be valid. */
2399        if (p->pack_fd != -1)
2400                return 1;
2401
2402        /* If the pack has one window completely covering the
2403         * file size, the pack is known to be valid even if
2404         * the descriptor is not currently open.
2405         */
2406        if (p->windows) {
2407                struct pack_window *w = p->windows;
2408
2409                if (!w->offset && w->len == p->pack_size)
2410                        return 1;
2411        }
2412
2413        /* Force the pack to open to prove its valid. */
2414        return !open_packed_git(p);
2415}
2416
2417static int fill_pack_entry(const unsigned char *sha1,
2418                           struct pack_entry *e,
2419                           struct packed_git *p)
2420{
2421        off_t offset;
2422
2423        if (p->num_bad_objects) {
2424                unsigned i;
2425                for (i = 0; i < p->num_bad_objects; i++)
2426                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2427                                return 0;
2428        }
2429
2430        offset = find_pack_entry_one(sha1, p);
2431        if (!offset)
2432                return 0;
2433
2434        /*
2435         * We are about to tell the caller where they can locate the
2436         * requested object.  We better make sure the packfile is
2437         * still here and can be accessed before supplying that
2438         * answer, as it may have been deleted since the index was
2439         * loaded!
2440         */
2441        if (!is_pack_valid(p)) {
2442                warning("packfile %s cannot be accessed", p->pack_name);
2443                return 0;
2444        }
2445        e->offset = offset;
2446        e->p = p;
2447        hashcpy(e->sha1, sha1);
2448        return 1;
2449}
2450
2451static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2452{
2453        struct packed_git *p;
2454
2455        prepare_packed_git();
2456        if (!packed_git)
2457                return 0;
2458
2459        if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2460                return 1;
2461
2462        for (p = packed_git; p; p = p->next) {
2463                if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2464                        continue;
2465
2466                last_found_pack = p;
2467                return 1;
2468        }
2469        return 0;
2470}
2471
2472struct packed_git *find_sha1_pack(const unsigned char *sha1,
2473                                  struct packed_git *packs)
2474{
2475        struct packed_git *p;
2476
2477        for (p = packs; p; p = p->next) {
2478                if (find_pack_entry_one(sha1, p))
2479                        return p;
2480        }
2481        return NULL;
2482
2483}
2484
2485static int sha1_loose_object_info(const unsigned char *sha1,
2486                                  struct object_info *oi)
2487{
2488        int status;
2489        unsigned long mapsize, size;
2490        void *map;
2491        git_zstream stream;
2492        char hdr[32];
2493
2494        if (oi->delta_base_sha1)
2495                hashclr(oi->delta_base_sha1);
2496
2497        /*
2498         * If we don't care about type or size, then we don't
2499         * need to look inside the object at all. Note that we
2500         * do not optimize out the stat call, even if the
2501         * caller doesn't care about the disk-size, since our
2502         * return value implicitly indicates whether the
2503         * object even exists.
2504         */
2505        if (!oi->typep && !oi->sizep) {
2506                struct stat st;
2507                if (stat_sha1_file(sha1, &st) < 0)
2508                        return -1;
2509                if (oi->disk_sizep)
2510                        *oi->disk_sizep = st.st_size;
2511                return 0;
2512        }
2513
2514        map = map_sha1_file(sha1, &mapsize);
2515        if (!map)
2516                return -1;
2517        if (oi->disk_sizep)
2518                *oi->disk_sizep = mapsize;
2519        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2520                status = error("unable to unpack %s header",
2521                               sha1_to_hex(sha1));
2522        else if ((status = parse_sha1_header(hdr, &size)) < 0)
2523                status = error("unable to parse %s header", sha1_to_hex(sha1));
2524        else if (oi->sizep)
2525                *oi->sizep = size;
2526        git_inflate_end(&stream);
2527        munmap(map, mapsize);
2528        if (oi->typep)
2529                *oi->typep = status;
2530        return 0;
2531}
2532
2533int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2534{
2535        struct cached_object *co;
2536        struct pack_entry e;
2537        int rtype;
2538        const unsigned char *real = lookup_replace_object_extended(sha1, flags);
2539
2540        co = find_cached_object(real);
2541        if (co) {
2542                if (oi->typep)
2543                        *(oi->typep) = co->type;
2544                if (oi->sizep)
2545                        *(oi->sizep) = co->size;
2546                if (oi->disk_sizep)
2547                        *(oi->disk_sizep) = 0;
2548                if (oi->delta_base_sha1)
2549                        hashclr(oi->delta_base_sha1);
2550                oi->whence = OI_CACHED;
2551                return 0;
2552        }
2553
2554        if (!find_pack_entry(real, &e)) {
2555                /* Most likely it's a loose object. */
2556                if (!sha1_loose_object_info(real, oi)) {
2557                        oi->whence = OI_LOOSE;
2558                        return 0;
2559                }
2560
2561                /* Not a loose object; someone else may have just packed it. */
2562                reprepare_packed_git();
2563                if (!find_pack_entry(real, &e))
2564                        return -1;
2565        }
2566
2567        rtype = packed_object_info(e.p, e.offset, oi);
2568        if (rtype < 0) {
2569                mark_bad_packed_object(e.p, real);
2570                return sha1_object_info_extended(real, oi, 0);
2571        } else if (in_delta_base_cache(e.p, e.offset)) {
2572                oi->whence = OI_DBCACHED;
2573        } else {
2574                oi->whence = OI_PACKED;
2575                oi->u.packed.offset = e.offset;
2576                oi->u.packed.pack = e.p;
2577                oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2578                                         rtype == OBJ_OFS_DELTA);
2579        }
2580
2581        return 0;
2582}
2583
2584/* returns enum object_type or negative */
2585int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2586{
2587        enum object_type type;
2588        struct object_info oi = {NULL};
2589
2590        oi.typep = &type;
2591        oi.sizep = sizep;
2592        if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
2593                return -1;
2594        return type;
2595}
2596
2597static void *read_packed_sha1(const unsigned char *sha1,
2598                              enum object_type *type, unsigned long *size)
2599{
2600        struct pack_entry e;
2601        void *data;
2602
2603        if (!find_pack_entry(sha1, &e))
2604                return NULL;
2605        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2606        if (!data) {
2607                /*
2608                 * We're probably in deep shit, but let's try to fetch
2609                 * the required object anyway from another pack or loose.
2610                 * This should happen only in the presence of a corrupted
2611                 * pack, and is better than failing outright.
2612                 */
2613                error("failed to read object %s at offset %"PRIuMAX" from %s",
2614                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2615                mark_bad_packed_object(e.p, sha1);
2616                data = read_object(sha1, type, size);
2617        }
2618        return data;
2619}
2620
2621int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2622                      unsigned char *sha1)
2623{
2624        struct cached_object *co;
2625
2626        hash_sha1_file(buf, len, typename(type), sha1);
2627        if (has_sha1_file(sha1) || find_cached_object(sha1))
2628                return 0;
2629        if (cached_object_alloc <= cached_object_nr) {
2630                cached_object_alloc = alloc_nr(cached_object_alloc);
2631                cached_objects = xrealloc(cached_objects,
2632                                          sizeof(*cached_objects) *
2633                                          cached_object_alloc);
2634        }
2635        co = &cached_objects[cached_object_nr++];
2636        co->size = len;
2637        co->type = type;
2638        co->buf = xmalloc(len);
2639        memcpy(co->buf, buf, len);
2640        hashcpy(co->sha1, sha1);
2641        return 0;
2642}
2643
2644static void *read_object(const unsigned char *sha1, enum object_type *type,
2645                         unsigned long *size)
2646{
2647        unsigned long mapsize;
2648        void *map, *buf;
2649        struct cached_object *co;
2650
2651        co = find_cached_object(sha1);
2652        if (co) {
2653                *type = co->type;
2654                *size = co->size;
2655                return xmemdupz(co->buf, co->size);
2656        }
2657
2658        buf = read_packed_sha1(sha1, type, size);
2659        if (buf)
2660                return buf;
2661        map = map_sha1_file(sha1, &mapsize);
2662        if (map) {
2663                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2664                munmap(map, mapsize);
2665                return buf;
2666        }
2667        reprepare_packed_git();
2668        return read_packed_sha1(sha1, type, size);
2669}
2670
2671/*
2672 * This function dies on corrupt objects; the callers who want to
2673 * deal with them should arrange to call read_object() and give error
2674 * messages themselves.
2675 */
2676void *read_sha1_file_extended(const unsigned char *sha1,
2677                              enum object_type *type,
2678                              unsigned long *size,
2679                              unsigned flag)
2680{
2681        void *data;
2682        char *path;
2683        const struct packed_git *p;
2684        const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
2685
2686        errno = 0;
2687        data = read_object(repl, type, size);
2688        if (data)
2689                return data;
2690
2691        if (errno && errno != ENOENT)
2692                die_errno("failed to read object %s", sha1_to_hex(sha1));
2693
2694        /* die if we replaced an object with one that does not exist */
2695        if (repl != sha1)
2696                die("replacement %s not found for %s",
2697                    sha1_to_hex(repl), sha1_to_hex(sha1));
2698
2699        if (has_loose_object(repl)) {
2700                path = sha1_file_name(sha1);
2701                die("loose object %s (stored in %s) is corrupt",
2702                    sha1_to_hex(repl), path);
2703        }
2704
2705        if ((p = has_packed_and_bad(repl)) != NULL)
2706                die("packed object %s (stored in %s) is corrupt",
2707                    sha1_to_hex(repl), p->pack_name);
2708
2709        return NULL;
2710}
2711
2712void *read_object_with_reference(const unsigned char *sha1,
2713                                 const char *required_type_name,
2714                                 unsigned long *size,
2715                                 unsigned char *actual_sha1_return)
2716{
2717        enum object_type type, required_type;
2718        void *buffer;
2719        unsigned long isize;
2720        unsigned char actual_sha1[20];
2721
2722        required_type = type_from_string(required_type_name);
2723        hashcpy(actual_sha1, sha1);
2724        while (1) {
2725                int ref_length = -1;
2726                const char *ref_type = NULL;
2727
2728                buffer = read_sha1_file(actual_sha1, &type, &isize);
2729                if (!buffer)
2730                        return NULL;
2731                if (type == required_type) {
2732                        *size = isize;
2733                        if (actual_sha1_return)
2734                                hashcpy(actual_sha1_return, actual_sha1);
2735                        return buffer;
2736                }
2737                /* Handle references */
2738                else if (type == OBJ_COMMIT)
2739                        ref_type = "tree ";
2740                else if (type == OBJ_TAG)
2741                        ref_type = "object ";
2742                else {
2743                        free(buffer);
2744                        return NULL;
2745                }
2746                ref_length = strlen(ref_type);
2747
2748                if (ref_length + 40 > isize ||
2749                    memcmp(buffer, ref_type, ref_length) ||
2750                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2751                        free(buffer);
2752                        return NULL;
2753                }
2754                free(buffer);
2755                /* Now we have the ID of the referred-to object in
2756                 * actual_sha1.  Check again. */
2757        }
2758}
2759
2760static void write_sha1_file_prepare(const void *buf, unsigned long len,
2761                                    const char *type, unsigned char *sha1,
2762                                    char *hdr, int *hdrlen)
2763{
2764        git_SHA_CTX c;
2765
2766        /* Generate the header */
2767        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2768
2769        /* Sha1.. */
2770        git_SHA1_Init(&c);
2771        git_SHA1_Update(&c, hdr, *hdrlen);
2772        git_SHA1_Update(&c, buf, len);
2773        git_SHA1_Final(sha1, &c);
2774}
2775
2776/*
2777 * Move the just written object into its final resting place.
2778 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2779 * "moving" is only a part of what it does, when no patch between
2780 * master to pu changes the call sites of this function.
2781 */
2782int move_temp_to_file(const char *tmpfile, const char *filename)
2783{
2784        int ret = 0;
2785
2786        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2787                goto try_rename;
2788        else if (link(tmpfile, filename))
2789                ret = errno;
2790
2791        /*
2792         * Coda hack - coda doesn't like cross-directory links,
2793         * so we fall back to a rename, which will mean that it
2794         * won't be able to check collisions, but that's not a
2795         * big deal.
2796         *
2797         * The same holds for FAT formatted media.
2798         *
2799         * When this succeeds, we just return.  We have nothing
2800         * left to unlink.
2801         */
2802        if (ret && ret != EEXIST) {
2803        try_rename:
2804                if (!rename(tmpfile, filename))
2805                        goto out;
2806                ret = errno;
2807        }
2808        unlink_or_warn(tmpfile);
2809        if (ret) {
2810                if (ret != EEXIST) {
2811                        return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2812                }
2813                /* FIXME!!! Collision check here ? */
2814        }
2815
2816out:
2817        if (adjust_shared_perm(filename))
2818                return error("unable to set permission to '%s'", filename);
2819        return 0;
2820}
2821
2822static int write_buffer(int fd, const void *buf, size_t len)
2823{
2824        if (write_in_full(fd, buf, len) < 0)
2825                return error("file write error (%s)", strerror(errno));
2826        return 0;
2827}
2828
2829int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2830                   unsigned char *sha1)
2831{
2832        char hdr[32];
2833        int hdrlen;
2834        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2835        return 0;
2836}
2837
2838/* Finalize a file on disk, and close it. */
2839static void close_sha1_file(int fd)
2840{
2841        if (fsync_object_files)
2842                fsync_or_die(fd, "sha1 file");
2843        if (close(fd) != 0)
2844                die_errno("error when closing sha1 file");
2845}
2846
2847/* Size of directory component, including the ending '/' */
2848static inline int directory_size(const char *filename)
2849{
2850        const char *s = strrchr(filename, '/');
2851        if (!s)
2852                return 0;
2853        return s - filename + 1;
2854}
2855
2856/*
2857 * This creates a temporary file in the same directory as the final
2858 * 'filename'
2859 *
2860 * We want to avoid cross-directory filename renames, because those
2861 * can have problems on various filesystems (FAT, NFS, Coda).
2862 */
2863static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2864{
2865        int fd, dirlen = directory_size(filename);
2866
2867        if (dirlen + 20 > bufsiz) {
2868                errno = ENAMETOOLONG;
2869                return -1;
2870        }
2871        memcpy(buffer, filename, dirlen);
2872        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2873        fd = git_mkstemp_mode(buffer, 0444);
2874        if (fd < 0 && dirlen && errno == ENOENT) {
2875                /* Make sure the directory exists */
2876                memcpy(buffer, filename, dirlen);
2877                buffer[dirlen-1] = 0;
2878                if (mkdir(buffer, 0777) && errno != EEXIST)
2879                        return -1;
2880                if (adjust_shared_perm(buffer))
2881                        return -1;
2882
2883                /* Try again */
2884                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2885                fd = git_mkstemp_mode(buffer, 0444);
2886        }
2887        return fd;
2888}
2889
2890static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2891                              const void *buf, unsigned long len, time_t mtime)
2892{
2893        int fd, ret;
2894        unsigned char compressed[4096];
2895        git_zstream stream;
2896        git_SHA_CTX c;
2897        unsigned char parano_sha1[20];
2898        char *filename;
2899        static char tmp_file[PATH_MAX];
2900
2901        filename = sha1_file_name(sha1);
2902        fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2903        if (fd < 0) {
2904                if (errno == EACCES)
2905                        return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2906                else
2907                        return error("unable to create temporary file: %s", strerror(errno));
2908        }
2909
2910        /* Set it up */
2911        memset(&stream, 0, sizeof(stream));
2912        git_deflate_init(&stream, zlib_compression_level);
2913        stream.next_out = compressed;
2914        stream.avail_out = sizeof(compressed);
2915        git_SHA1_Init(&c);
2916
2917        /* First header.. */
2918        stream.next_in = (unsigned char *)hdr;
2919        stream.avail_in = hdrlen;
2920        while (git_deflate(&stream, 0) == Z_OK)
2921                ; /* nothing */
2922        git_SHA1_Update(&c, hdr, hdrlen);
2923
2924        /* Then the data itself.. */
2925        stream.next_in = (void *)buf;
2926        stream.avail_in = len;
2927        do {
2928                unsigned char *in0 = stream.next_in;
2929                ret = git_deflate(&stream, Z_FINISH);
2930                git_SHA1_Update(&c, in0, stream.next_in - in0);
2931                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2932                        die("unable to write sha1 file");
2933                stream.next_out = compressed;
2934                stream.avail_out = sizeof(compressed);
2935        } while (ret == Z_OK);
2936
2937        if (ret != Z_STREAM_END)
2938                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2939        ret = git_deflate_end_gently(&stream);
2940        if (ret != Z_OK)
2941                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2942        git_SHA1_Final(parano_sha1, &c);
2943        if (hashcmp(sha1, parano_sha1) != 0)
2944                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2945
2946        close_sha1_file(fd);
2947
2948        if (mtime) {
2949                struct utimbuf utb;
2950                utb.actime = mtime;
2951                utb.modtime = mtime;
2952                if (utime(tmp_file, &utb) < 0)
2953                        warning("failed utime() on %s: %s",
2954                                tmp_file, strerror(errno));
2955        }
2956
2957        return move_temp_to_file(tmp_file, filename);
2958}
2959
2960int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2961{
2962        unsigned char sha1[20];
2963        char hdr[32];
2964        int hdrlen;
2965
2966        /* Normally if we have it in the pack then we do not bother writing
2967         * it out into .git/objects/??/?{38} file.
2968         */
2969        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2970        if (returnsha1)
2971                hashcpy(returnsha1, sha1);
2972        if (has_sha1_file(sha1))
2973                return 0;
2974        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2975}
2976
2977int force_object_loose(const unsigned char *sha1, time_t mtime)
2978{
2979        void *buf;
2980        unsigned long len;
2981        enum object_type type;
2982        char hdr[32];
2983        int hdrlen;
2984        int ret;
2985
2986        if (has_loose_object(sha1))
2987                return 0;
2988        buf = read_packed_sha1(sha1, &type, &len);
2989        if (!buf)
2990                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2991        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2992        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2993        free(buf);
2994
2995        return ret;
2996}
2997
2998int has_pack_index(const unsigned char *sha1)
2999{
3000        struct stat st;
3001        if (stat(sha1_pack_index_name(sha1), &st))
3002                return 0;
3003        return 1;
3004}
3005
3006int has_sha1_pack(const unsigned char *sha1)
3007{
3008        struct pack_entry e;
3009        return find_pack_entry(sha1, &e);
3010}
3011
3012int has_sha1_file(const unsigned char *sha1)
3013{
3014        struct pack_entry e;
3015
3016        if (find_pack_entry(sha1, &e))
3017                return 1;
3018        if (has_loose_object(sha1))
3019                return 1;
3020        reprepare_packed_git();
3021        return find_pack_entry(sha1, &e);
3022}
3023
3024static void check_tree(const void *buf, size_t size)
3025{
3026        struct tree_desc desc;
3027        struct name_entry entry;
3028
3029        init_tree_desc(&desc, buf, size);
3030        while (tree_entry(&desc, &entry))
3031                /* do nothing
3032                 * tree_entry() will die() on malformed entries */
3033                ;
3034}
3035
3036static void check_commit(const void *buf, size_t size)
3037{
3038        struct commit c;
3039        memset(&c, 0, sizeof(c));
3040        if (parse_commit_buffer(&c, buf, size))
3041                die("corrupt commit");
3042}
3043
3044static void check_tag(const void *buf, size_t size)
3045{
3046        struct tag t;
3047        memset(&t, 0, sizeof(t));
3048        if (parse_tag_buffer(&t, buf, size))
3049                die("corrupt tag");
3050}
3051
3052static int index_mem(unsigned char *sha1, void *buf, size_t size,
3053                     enum object_type type,
3054                     const char *path, unsigned flags)
3055{
3056        int ret, re_allocated = 0;
3057        int write_object = flags & HASH_WRITE_OBJECT;
3058
3059        if (!type)
3060                type = OBJ_BLOB;
3061
3062        /*
3063         * Convert blobs to git internal format
3064         */
3065        if ((type == OBJ_BLOB) && path) {
3066                struct strbuf nbuf = STRBUF_INIT;
3067                if (convert_to_git(path, buf, size, &nbuf,
3068                                   write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3069                        buf = strbuf_detach(&nbuf, &size);
3070                        re_allocated = 1;
3071                }
3072        }
3073        if (flags & HASH_FORMAT_CHECK) {
3074                if (type == OBJ_TREE)
3075                        check_tree(buf, size);
3076                if (type == OBJ_COMMIT)
3077                        check_commit(buf, size);
3078                if (type == OBJ_TAG)
3079                        check_tag(buf, size);
3080        }
3081
3082        if (write_object)
3083                ret = write_sha1_file(buf, size, typename(type), sha1);
3084        else
3085                ret = hash_sha1_file(buf, size, typename(type), sha1);
3086        if (re_allocated)
3087                free(buf);
3088        return ret;
3089}
3090
3091static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3092                      const char *path, unsigned flags)
3093{
3094        struct strbuf sbuf = STRBUF_INIT;
3095        int ret;
3096
3097        if (strbuf_read(&sbuf, fd, 4096) >= 0)
3098                ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3099        else
3100                ret = -1;
3101        strbuf_release(&sbuf);
3102        return ret;
3103}
3104
3105#define SMALL_FILE_SIZE (32*1024)
3106
3107static int index_core(unsigned char *sha1, int fd, size_t size,
3108                      enum object_type type, const char *path,
3109                      unsigned flags)
3110{
3111        int ret;
3112
3113        if (!size) {
3114                ret = index_mem(sha1, NULL, size, type, path, flags);
3115        } else if (size <= SMALL_FILE_SIZE) {
3116                char *buf = xmalloc(size);
3117                if (size == read_in_full(fd, buf, size))
3118                        ret = index_mem(sha1, buf, size, type, path, flags);
3119                else
3120                        ret = error("short read %s", strerror(errno));
3121                free(buf);
3122        } else {
3123                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3124                ret = index_mem(sha1, buf, size, type, path, flags);
3125                munmap(buf, size);
3126        }
3127        return ret;
3128}
3129
3130/*
3131 * This creates one packfile per large blob unless bulk-checkin
3132 * machinery is "plugged".
3133 *
3134 * This also bypasses the usual "convert-to-git" dance, and that is on
3135 * purpose. We could write a streaming version of the converting
3136 * functions and insert that before feeding the data to fast-import
3137 * (or equivalent in-core API described above). However, that is
3138 * somewhat complicated, as we do not know the size of the filter
3139 * result, which we need to know beforehand when writing a git object.
3140 * Since the primary motivation for trying to stream from the working
3141 * tree file and to avoid mmaping it in core is to deal with large
3142 * binary blobs, they generally do not want to get any conversion, and
3143 * callers should avoid this code path when filters are requested.
3144 */
3145static int index_stream(unsigned char *sha1, int fd, size_t size,
3146                        enum object_type type, const char *path,
3147                        unsigned flags)
3148{
3149        return index_bulk_checkin(sha1, fd, size, type, path, flags);
3150}
3151
3152int index_fd(unsigned char *sha1, int fd, struct stat *st,
3153             enum object_type type, const char *path, unsigned flags)
3154{
3155        int ret;
3156        size_t size = xsize_t(st->st_size);
3157
3158        if (!S_ISREG(st->st_mode))
3159                ret = index_pipe(sha1, fd, type, path, flags);
3160        else if (size <= big_file_threshold || type != OBJ_BLOB ||
3161                 (path && would_convert_to_git(path, NULL, 0, 0)))
3162                ret = index_core(sha1, fd, size, type, path, flags);
3163        else
3164                ret = index_stream(sha1, fd, size, type, path, flags);
3165        close(fd);
3166        return ret;
3167}
3168
3169int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3170{
3171        int fd;
3172        struct strbuf sb = STRBUF_INIT;
3173
3174        switch (st->st_mode & S_IFMT) {
3175        case S_IFREG:
3176                fd = open(path, O_RDONLY);
3177                if (fd < 0)
3178                        return error("open(\"%s\"): %s", path,
3179                                     strerror(errno));
3180                if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3181                        return error("%s: failed to insert into database",
3182                                     path);
3183                break;
3184        case S_IFLNK:
3185                if (strbuf_readlink(&sb, path, st->st_size)) {
3186                        char *errstr = strerror(errno);
3187                        return error("readlink(\"%s\"): %s", path,
3188                                     errstr);
3189                }
3190                if (!(flags & HASH_WRITE_OBJECT))
3191                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3192                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3193                        return error("%s: failed to insert into database",
3194                                     path);
3195                strbuf_release(&sb);
3196                break;
3197        case S_IFDIR:
3198                return resolve_gitlink_ref(path, "HEAD", sha1);
3199        default:
3200                return error("%s: unsupported file type", path);
3201        }
3202        return 0;
3203}
3204
3205int read_pack_header(int fd, struct pack_header *header)
3206{
3207        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3208                /* "eof before pack header was fully read" */
3209                return PH_ERROR_EOF;
3210
3211        if (header->hdr_signature != htonl(PACK_SIGNATURE))
3212                /* "protocol error (pack signature mismatch detected)" */
3213                return PH_ERROR_PACK_SIGNATURE;
3214        if (!pack_version_ok(header->hdr_version))
3215                /* "protocol error (pack version unsupported)" */
3216                return PH_ERROR_PROTOCOL;
3217        return 0;
3218}
3219
3220void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3221{
3222        enum object_type type = sha1_object_info(sha1, NULL);
3223        if (type < 0)
3224                die("%s is not a valid object", sha1_to_hex(sha1));
3225        if (type != expect)
3226                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3227                    typename(expect));
3228}