sha1_file.con commit Merge branch 'jc/convert' (fb53969)
   1/*
   2 * GIT - The information manager from hell
   3 *
   4 * Copyright (C) Linus Torvalds, 2005
   5 *
   6 * This handles basic git sha1 object files - packing, unpacking,
   7 * creation etc.
   8 */
   9#include "cache.h"
  10#include "delta.h"
  11#include "pack.h"
  12#include "blob.h"
  13#include "commit.h"
  14#include "tag.h"
  15#include "tree.h"
  16#include "tree-walk.h"
  17#include "refs.h"
  18#include "pack-revindex.h"
  19#include "sha1-lookup.h"
  20
  21#ifndef O_NOATIME
  22#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  23#define O_NOATIME 01000000
  24#else
  25#define O_NOATIME 0
  26#endif
  27#endif
  28
  29#define SZ_FMT PRIuMAX
  30static inline uintmax_t sz_fmt(size_t s) { return s; }
  31
  32const unsigned char null_sha1[20];
  33
  34/*
  35 * This is meant to hold a *small* number of objects that you would
  36 * want read_sha1_file() to be able to return, but yet you do not want
  37 * to write them into the object store (e.g. a browse-only
  38 * application).
  39 */
  40static struct cached_object {
  41        unsigned char sha1[20];
  42        enum object_type type;
  43        void *buf;
  44        unsigned long size;
  45} *cached_objects;
  46static int cached_object_nr, cached_object_alloc;
  47
  48static struct cached_object empty_tree = {
  49        EMPTY_TREE_SHA1_BIN_LITERAL,
  50        OBJ_TREE,
  51        "",
  52        0
  53};
  54
  55static struct cached_object *find_cached_object(const unsigned char *sha1)
  56{
  57        int i;
  58        struct cached_object *co = cached_objects;
  59
  60        for (i = 0; i < cached_object_nr; i++, co++) {
  61                if (!hashcmp(co->sha1, sha1))
  62                        return co;
  63        }
  64        if (!hashcmp(sha1, empty_tree.sha1))
  65                return &empty_tree;
  66        return NULL;
  67}
  68
  69int mkdir_in_gitdir(const char *path)
  70{
  71        if (mkdir(path, 0777)) {
  72                int saved_errno = errno;
  73                struct stat st;
  74                struct strbuf sb = STRBUF_INIT;
  75
  76                if (errno != EEXIST)
  77                        return -1;
  78                /*
  79                 * Are we looking at a path in a symlinked worktree
  80                 * whose original repository does not yet have it?
  81                 * e.g. .git/rr-cache pointing at its original
  82                 * repository in which the user hasn't performed any
  83                 * conflict resolution yet?
  84                 */
  85                if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
  86                    strbuf_readlink(&sb, path, st.st_size) ||
  87                    !is_absolute_path(sb.buf) ||
  88                    mkdir(sb.buf, 0777)) {
  89                        strbuf_release(&sb);
  90                        errno = saved_errno;
  91                        return -1;
  92                }
  93                strbuf_release(&sb);
  94        }
  95        return adjust_shared_perm(path);
  96}
  97
  98int safe_create_leading_directories(char *path)
  99{
 100        char *pos = path + offset_1st_component(path);
 101        struct stat st;
 102
 103        while (pos) {
 104                pos = strchr(pos, '/');
 105                if (!pos)
 106                        break;
 107                while (*++pos == '/')
 108                        ;
 109                if (!*pos)
 110                        break;
 111                *--pos = '\0';
 112                if (!stat(path, &st)) {
 113                        /* path exists */
 114                        if (!S_ISDIR(st.st_mode)) {
 115                                *pos = '/';
 116                                return -3;
 117                        }
 118                }
 119                else if (mkdir(path, 0777)) {
 120                        *pos = '/';
 121                        return -1;
 122                }
 123                else if (adjust_shared_perm(path)) {
 124                        *pos = '/';
 125                        return -2;
 126                }
 127                *pos++ = '/';
 128        }
 129        return 0;
 130}
 131
 132int safe_create_leading_directories_const(const char *path)
 133{
 134        /* path points to cache entries, so xstrdup before messing with it */
 135        char *buf = xstrdup(path);
 136        int result = safe_create_leading_directories(buf);
 137        free(buf);
 138        return result;
 139}
 140
 141static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 142{
 143        int i;
 144        for (i = 0; i < 20; i++) {
 145                static char hex[] = "0123456789abcdef";
 146                unsigned int val = sha1[i];
 147                char *pos = pathbuf + i*2 + (i > 0);
 148                *pos++ = hex[val >> 4];
 149                *pos = hex[val & 0xf];
 150        }
 151}
 152
 153/*
 154 * NOTE! This returns a statically allocated buffer, so you have to be
 155 * careful about using it. Do an "xstrdup()" if you need to save the
 156 * filename.
 157 *
 158 * Also note that this returns the location for creating.  Reading
 159 * SHA1 file can happen from any alternate directory listed in the
 160 * DB_ENVIRONMENT environment variable if it is not found in
 161 * the primary object database.
 162 */
 163char *sha1_file_name(const unsigned char *sha1)
 164{
 165        static char buf[PATH_MAX];
 166        const char *objdir;
 167        int len;
 168
 169        objdir = get_object_directory();
 170        len = strlen(objdir);
 171
 172        /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
 173        if (len + 43 > PATH_MAX)
 174                die("insanely long object directory %s", objdir);
 175        memcpy(buf, objdir, len);
 176        buf[len] = '/';
 177        buf[len+3] = '/';
 178        buf[len+42] = '\0';
 179        fill_sha1_path(buf + len + 1, sha1);
 180        return buf;
 181}
 182
 183static char *sha1_get_pack_name(const unsigned char *sha1,
 184                                char **name, char **base, const char *which)
 185{
 186        static const char hex[] = "0123456789abcdef";
 187        char *buf;
 188        int i;
 189
 190        if (!*base) {
 191                const char *sha1_file_directory = get_object_directory();
 192                int len = strlen(sha1_file_directory);
 193                *base = xmalloc(len + 60);
 194                sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
 195                        sha1_file_directory, which);
 196                *name = *base + len + 11;
 197        }
 198
 199        buf = *name;
 200
 201        for (i = 0; i < 20; i++) {
 202                unsigned int val = *sha1++;
 203                *buf++ = hex[val >> 4];
 204                *buf++ = hex[val & 0xf];
 205        }
 206
 207        return *base;
 208}
 209
 210char *sha1_pack_name(const unsigned char *sha1)
 211{
 212        static char *name, *base;
 213
 214        return sha1_get_pack_name(sha1, &name, &base, "pack");
 215}
 216
 217char *sha1_pack_index_name(const unsigned char *sha1)
 218{
 219        static char *name, *base;
 220
 221        return sha1_get_pack_name(sha1, &name, &base, "idx");
 222}
 223
 224struct alternate_object_database *alt_odb_list;
 225static struct alternate_object_database **alt_odb_tail;
 226
 227static void read_info_alternates(const char * alternates, int depth);
 228static int git_open_noatime(const char *name);
 229
 230/*
 231 * Prepare alternate object database registry.
 232 *
 233 * The variable alt_odb_list points at the list of struct
 234 * alternate_object_database.  The elements on this list come from
 235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 237 * whose contents is similar to that environment variable but can be
 238 * LF separated.  Its base points at a statically allocated buffer that
 239 * contains "/the/directory/corresponding/to/.git/objects/...", while
 240 * its name points just after the slash at the end of ".git/objects/"
 241 * in the example above, and has enough space to hold 40-byte hex
 242 * SHA1, an extra slash for the first level indirection, and the
 243 * terminating NUL.
 244 */
 245static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 246{
 247        const char *objdir = get_object_directory();
 248        struct alternate_object_database *ent;
 249        struct alternate_object_database *alt;
 250        /* 43 = 40-byte + 2 '/' + terminating NUL */
 251        int pfxlen = len;
 252        int entlen = pfxlen + 43;
 253        int base_len = -1;
 254
 255        if (!is_absolute_path(entry) && relative_base) {
 256                /* Relative alt-odb */
 257                if (base_len < 0)
 258                        base_len = strlen(relative_base) + 1;
 259                entlen += base_len;
 260                pfxlen += base_len;
 261        }
 262        ent = xmalloc(sizeof(*ent) + entlen);
 263
 264        if (!is_absolute_path(entry) && relative_base) {
 265                memcpy(ent->base, relative_base, base_len - 1);
 266                ent->base[base_len - 1] = '/';
 267                memcpy(ent->base + base_len, entry, len);
 268        }
 269        else
 270                memcpy(ent->base, entry, pfxlen);
 271
 272        ent->name = ent->base + pfxlen + 1;
 273        ent->base[pfxlen + 3] = '/';
 274        ent->base[pfxlen] = ent->base[entlen-1] = 0;
 275
 276        /* Detect cases where alternate disappeared */
 277        if (!is_directory(ent->base)) {
 278                error("object directory %s does not exist; "
 279                      "check .git/objects/info/alternates.",
 280                      ent->base);
 281                free(ent);
 282                return -1;
 283        }
 284
 285        /* Prevent the common mistake of listing the same
 286         * thing twice, or object directory itself.
 287         */
 288        for (alt = alt_odb_list; alt; alt = alt->next) {
 289                if (!memcmp(ent->base, alt->base, pfxlen)) {
 290                        free(ent);
 291                        return -1;
 292                }
 293        }
 294        if (!memcmp(ent->base, objdir, pfxlen)) {
 295                free(ent);
 296                return -1;
 297        }
 298
 299        /* add the alternate entry */
 300        *alt_odb_tail = ent;
 301        alt_odb_tail = &(ent->next);
 302        ent->next = NULL;
 303
 304        /* recursively add alternates */
 305        read_info_alternates(ent->base, depth + 1);
 306
 307        ent->base[pfxlen] = '/';
 308
 309        return 0;
 310}
 311
 312static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
 313                                 const char *relative_base, int depth)
 314{
 315        const char *cp, *last;
 316
 317        if (depth > 5) {
 318                error("%s: ignoring alternate object stores, nesting too deep.",
 319                                relative_base);
 320                return;
 321        }
 322
 323        last = alt;
 324        while (last < ep) {
 325                cp = last;
 326                if (cp < ep && *cp == '#') {
 327                        while (cp < ep && *cp != sep)
 328                                cp++;
 329                        last = cp + 1;
 330                        continue;
 331                }
 332                while (cp < ep && *cp != sep)
 333                        cp++;
 334                if (last != cp) {
 335                        if (!is_absolute_path(last) && depth) {
 336                                error("%s: ignoring relative alternate object store %s",
 337                                                relative_base, last);
 338                        } else {
 339                                link_alt_odb_entry(last, cp - last,
 340                                                relative_base, depth);
 341                        }
 342                }
 343                while (cp < ep && *cp == sep)
 344                        cp++;
 345                last = cp;
 346        }
 347}
 348
 349static void read_info_alternates(const char * relative_base, int depth)
 350{
 351        char *map;
 352        size_t mapsz;
 353        struct stat st;
 354        const char alt_file_name[] = "info/alternates";
 355        /* Given that relative_base is no longer than PATH_MAX,
 356           ensure that "path" has enough space to append "/", the
 357           file name, "info/alternates", and a trailing NUL.  */
 358        char path[PATH_MAX + 1 + sizeof alt_file_name];
 359        int fd;
 360
 361        sprintf(path, "%s/%s", relative_base, alt_file_name);
 362        fd = git_open_noatime(path);
 363        if (fd < 0)
 364                return;
 365        if (fstat(fd, &st) || (st.st_size == 0)) {
 366                close(fd);
 367                return;
 368        }
 369        mapsz = xsize_t(st.st_size);
 370        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 371        close(fd);
 372
 373        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 374
 375        munmap(map, mapsz);
 376}
 377
 378void add_to_alternates_file(const char *reference)
 379{
 380        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
 381        int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
 382        char *alt = mkpath("%s/objects\n", reference);
 383        write_or_die(fd, alt, strlen(alt));
 384        if (commit_lock_file(lock))
 385                die("could not close alternates file");
 386        if (alt_odb_tail)
 387                link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
 388}
 389
 390void foreach_alt_odb(alt_odb_fn fn, void *cb)
 391{
 392        struct alternate_object_database *ent;
 393
 394        prepare_alt_odb();
 395        for (ent = alt_odb_list; ent; ent = ent->next)
 396                if (fn(ent, cb))
 397                        return;
 398}
 399
 400void prepare_alt_odb(void)
 401{
 402        const char *alt;
 403
 404        if (alt_odb_tail)
 405                return;
 406
 407        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 408        if (!alt) alt = "";
 409
 410        alt_odb_tail = &alt_odb_list;
 411        link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
 412
 413        read_info_alternates(get_object_directory(), 0);
 414}
 415
 416static int has_loose_object_local(const unsigned char *sha1)
 417{
 418        char *name = sha1_file_name(sha1);
 419        return !access(name, F_OK);
 420}
 421
 422int has_loose_object_nonlocal(const unsigned char *sha1)
 423{
 424        struct alternate_object_database *alt;
 425        prepare_alt_odb();
 426        for (alt = alt_odb_list; alt; alt = alt->next) {
 427                fill_sha1_path(alt->name, sha1);
 428                if (!access(alt->base, F_OK))
 429                        return 1;
 430        }
 431        return 0;
 432}
 433
 434static int has_loose_object(const unsigned char *sha1)
 435{
 436        return has_loose_object_local(sha1) ||
 437               has_loose_object_nonlocal(sha1);
 438}
 439
 440static unsigned int pack_used_ctr;
 441static unsigned int pack_mmap_calls;
 442static unsigned int peak_pack_open_windows;
 443static unsigned int pack_open_windows;
 444static unsigned int pack_open_fds;
 445static unsigned int pack_max_fds;
 446static size_t peak_pack_mapped;
 447static size_t pack_mapped;
 448struct packed_git *packed_git;
 449
 450void pack_report(void)
 451{
 452        fprintf(stderr,
 453                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 454                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 455                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 456                sz_fmt(getpagesize()),
 457                sz_fmt(packed_git_window_size),
 458                sz_fmt(packed_git_limit));
 459        fprintf(stderr,
 460                "pack_report: pack_used_ctr            = %10u\n"
 461                "pack_report: pack_mmap_calls          = %10u\n"
 462                "pack_report: pack_open_windows        = %10u / %10u\n"
 463                "pack_report: pack_mapped              = "
 464                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 465                pack_used_ctr,
 466                pack_mmap_calls,
 467                pack_open_windows, peak_pack_open_windows,
 468                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
 469}
 470
 471static int check_packed_git_idx(const char *path,  struct packed_git *p)
 472{
 473        void *idx_map;
 474        struct pack_idx_header *hdr;
 475        size_t idx_size;
 476        uint32_t version, nr, i, *index;
 477        int fd = git_open_noatime(path);
 478        struct stat st;
 479
 480        if (fd < 0)
 481                return -1;
 482        if (fstat(fd, &st)) {
 483                close(fd);
 484                return -1;
 485        }
 486        idx_size = xsize_t(st.st_size);
 487        if (idx_size < 4 * 256 + 20 + 20) {
 488                close(fd);
 489                return error("index file %s is too small", path);
 490        }
 491        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 492        close(fd);
 493
 494        hdr = idx_map;
 495        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 496                version = ntohl(hdr->idx_version);
 497                if (version < 2 || version > 2) {
 498                        munmap(idx_map, idx_size);
 499                        return error("index file %s is version %"PRIu32
 500                                     " and is not supported by this binary"
 501                                     " (try upgrading GIT to a newer version)",
 502                                     path, version);
 503                }
 504        } else
 505                version = 1;
 506
 507        nr = 0;
 508        index = idx_map;
 509        if (version > 1)
 510                index += 2;  /* skip index header */
 511        for (i = 0; i < 256; i++) {
 512                uint32_t n = ntohl(index[i]);
 513                if (n < nr) {
 514                        munmap(idx_map, idx_size);
 515                        return error("non-monotonic index %s", path);
 516                }
 517                nr = n;
 518        }
 519
 520        if (version == 1) {
 521                /*
 522                 * Total size:
 523                 *  - 256 index entries 4 bytes each
 524                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 525                 *  - 20-byte SHA1 of the packfile
 526                 *  - 20-byte SHA1 file checksum
 527                 */
 528                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 529                        munmap(idx_map, idx_size);
 530                        return error("wrong index v1 file size in %s", path);
 531                }
 532        } else if (version == 2) {
 533                /*
 534                 * Minimum size:
 535                 *  - 8 bytes of header
 536                 *  - 256 index entries 4 bytes each
 537                 *  - 20-byte sha1 entry * nr
 538                 *  - 4-byte crc entry * nr
 539                 *  - 4-byte offset entry * nr
 540                 *  - 20-byte SHA1 of the packfile
 541                 *  - 20-byte SHA1 file checksum
 542                 * And after the 4-byte offset table might be a
 543                 * variable sized table containing 8-byte entries
 544                 * for offsets larger than 2^31.
 545                 */
 546                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 547                unsigned long max_size = min_size;
 548                if (nr)
 549                        max_size += (nr - 1)*8;
 550                if (idx_size < min_size || idx_size > max_size) {
 551                        munmap(idx_map, idx_size);
 552                        return error("wrong index v2 file size in %s", path);
 553                }
 554                if (idx_size != min_size &&
 555                    /*
 556                     * make sure we can deal with large pack offsets.
 557                     * 31-bit signed offset won't be enough, neither
 558                     * 32-bit unsigned one will be.
 559                     */
 560                    (sizeof(off_t) <= 4)) {
 561                        munmap(idx_map, idx_size);
 562                        return error("pack too large for current definition of off_t in %s", path);
 563                }
 564        }
 565
 566        p->index_version = version;
 567        p->index_data = idx_map;
 568        p->index_size = idx_size;
 569        p->num_objects = nr;
 570        return 0;
 571}
 572
 573int open_pack_index(struct packed_git *p)
 574{
 575        char *idx_name;
 576        int ret;
 577
 578        if (p->index_data)
 579                return 0;
 580
 581        idx_name = xstrdup(p->pack_name);
 582        strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
 583        ret = check_packed_git_idx(idx_name, p);
 584        free(idx_name);
 585        return ret;
 586}
 587
 588static void scan_windows(struct packed_git *p,
 589        struct packed_git **lru_p,
 590        struct pack_window **lru_w,
 591        struct pack_window **lru_l)
 592{
 593        struct pack_window *w, *w_l;
 594
 595        for (w_l = NULL, w = p->windows; w; w = w->next) {
 596                if (!w->inuse_cnt) {
 597                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 598                                *lru_p = p;
 599                                *lru_w = w;
 600                                *lru_l = w_l;
 601                        }
 602                }
 603                w_l = w;
 604        }
 605}
 606
 607static int unuse_one_window(struct packed_git *current, int keep_fd)
 608{
 609        struct packed_git *p, *lru_p = NULL;
 610        struct pack_window *lru_w = NULL, *lru_l = NULL;
 611
 612        if (current)
 613                scan_windows(current, &lru_p, &lru_w, &lru_l);
 614        for (p = packed_git; p; p = p->next)
 615                scan_windows(p, &lru_p, &lru_w, &lru_l);
 616        if (lru_p) {
 617                munmap(lru_w->base, lru_w->len);
 618                pack_mapped -= lru_w->len;
 619                if (lru_l)
 620                        lru_l->next = lru_w->next;
 621                else {
 622                        lru_p->windows = lru_w->next;
 623                        if (!lru_p->windows && lru_p->pack_fd != -1
 624                                && lru_p->pack_fd != keep_fd) {
 625                                close(lru_p->pack_fd);
 626                                pack_open_fds--;
 627                                lru_p->pack_fd = -1;
 628                        }
 629                }
 630                free(lru_w);
 631                pack_open_windows--;
 632                return 1;
 633        }
 634        return 0;
 635}
 636
 637void release_pack_memory(size_t need, int fd)
 638{
 639        size_t cur = pack_mapped;
 640        while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
 641                ; /* nothing */
 642}
 643
 644void *xmmap(void *start, size_t length,
 645        int prot, int flags, int fd, off_t offset)
 646{
 647        void *ret = mmap(start, length, prot, flags, fd, offset);
 648        if (ret == MAP_FAILED) {
 649                if (!length)
 650                        return NULL;
 651                release_pack_memory(length, fd);
 652                ret = mmap(start, length, prot, flags, fd, offset);
 653                if (ret == MAP_FAILED)
 654                        die_errno("Out of memory? mmap failed");
 655        }
 656        return ret;
 657}
 658
 659void close_pack_windows(struct packed_git *p)
 660{
 661        while (p->windows) {
 662                struct pack_window *w = p->windows;
 663
 664                if (w->inuse_cnt)
 665                        die("pack '%s' still has open windows to it",
 666                            p->pack_name);
 667                munmap(w->base, w->len);
 668                pack_mapped -= w->len;
 669                pack_open_windows--;
 670                p->windows = w->next;
 671                free(w);
 672        }
 673}
 674
 675void unuse_pack(struct pack_window **w_cursor)
 676{
 677        struct pack_window *w = *w_cursor;
 678        if (w) {
 679                w->inuse_cnt--;
 680                *w_cursor = NULL;
 681        }
 682}
 683
 684void close_pack_index(struct packed_git *p)
 685{
 686        if (p->index_data) {
 687                munmap((void *)p->index_data, p->index_size);
 688                p->index_data = NULL;
 689        }
 690}
 691
 692/*
 693 * This is used by git-repack in case a newly created pack happens to
 694 * contain the same set of objects as an existing one.  In that case
 695 * the resulting file might be different even if its name would be the
 696 * same.  It is best to close any reference to the old pack before it is
 697 * replaced on disk.  Of course no index pointers nor windows for given pack
 698 * must subsist at this point.  If ever objects from this pack are requested
 699 * again, the new version of the pack will be reinitialized through
 700 * reprepare_packed_git().
 701 */
 702void free_pack_by_name(const char *pack_name)
 703{
 704        struct packed_git *p, **pp = &packed_git;
 705
 706        while (*pp) {
 707                p = *pp;
 708                if (strcmp(pack_name, p->pack_name) == 0) {
 709                        clear_delta_base_cache();
 710                        close_pack_windows(p);
 711                        if (p->pack_fd != -1) {
 712                                close(p->pack_fd);
 713                                pack_open_fds--;
 714                        }
 715                        close_pack_index(p);
 716                        free(p->bad_object_sha1);
 717                        *pp = p->next;
 718                        free(p);
 719                        return;
 720                }
 721                pp = &p->next;
 722        }
 723}
 724
 725/*
 726 * Do not call this directly as this leaks p->pack_fd on error return;
 727 * call open_packed_git() instead.
 728 */
 729static int open_packed_git_1(struct packed_git *p)
 730{
 731        struct stat st;
 732        struct pack_header hdr;
 733        unsigned char sha1[20];
 734        unsigned char *idx_sha1;
 735        long fd_flag;
 736
 737        if (!p->index_data && open_pack_index(p))
 738                return error("packfile %s index unavailable", p->pack_name);
 739
 740        if (!pack_max_fds) {
 741                struct rlimit lim;
 742                unsigned int max_fds;
 743
 744                if (getrlimit(RLIMIT_NOFILE, &lim))
 745                        die_errno("cannot get RLIMIT_NOFILE");
 746
 747                max_fds = lim.rlim_cur;
 748
 749                /* Save 3 for stdin/stdout/stderr, 22 for work */
 750                if (25 < max_fds)
 751                        pack_max_fds = max_fds - 25;
 752                else
 753                        pack_max_fds = 1;
 754        }
 755
 756        while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
 757                ; /* nothing */
 758
 759        p->pack_fd = git_open_noatime(p->pack_name);
 760        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 761                return -1;
 762        pack_open_fds++;
 763
 764        /* If we created the struct before we had the pack we lack size. */
 765        if (!p->pack_size) {
 766                if (!S_ISREG(st.st_mode))
 767                        return error("packfile %s not a regular file", p->pack_name);
 768                p->pack_size = st.st_size;
 769        } else if (p->pack_size != st.st_size)
 770                return error("packfile %s size changed", p->pack_name);
 771
 772        /* We leave these file descriptors open with sliding mmap;
 773         * there is no point keeping them open across exec(), though.
 774         */
 775        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 776        if (fd_flag < 0)
 777                return error("cannot determine file descriptor flags");
 778        fd_flag |= FD_CLOEXEC;
 779        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 780                return error("cannot set FD_CLOEXEC");
 781
 782        /* Verify we recognize this pack file format. */
 783        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 784                return error("file %s is far too short to be a packfile", p->pack_name);
 785        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 786                return error("file %s is not a GIT packfile", p->pack_name);
 787        if (!pack_version_ok(hdr.hdr_version))
 788                return error("packfile %s is version %"PRIu32" and not"
 789                        " supported (try upgrading GIT to a newer version)",
 790                        p->pack_name, ntohl(hdr.hdr_version));
 791
 792        /* Verify the pack matches its index. */
 793        if (p->num_objects != ntohl(hdr.hdr_entries))
 794                return error("packfile %s claims to have %"PRIu32" objects"
 795                             " while index indicates %"PRIu32" objects",
 796                             p->pack_name, ntohl(hdr.hdr_entries),
 797                             p->num_objects);
 798        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 799                return error("end of packfile %s is unavailable", p->pack_name);
 800        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 801                return error("packfile %s signature is unavailable", p->pack_name);
 802        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 803        if (hashcmp(sha1, idx_sha1))
 804                return error("packfile %s does not match index", p->pack_name);
 805        return 0;
 806}
 807
 808static int open_packed_git(struct packed_git *p)
 809{
 810        if (!open_packed_git_1(p))
 811                return 0;
 812        if (p->pack_fd != -1) {
 813                close(p->pack_fd);
 814                pack_open_fds--;
 815                p->pack_fd = -1;
 816        }
 817        return -1;
 818}
 819
 820static int in_window(struct pack_window *win, off_t offset)
 821{
 822        /* We must promise at least 20 bytes (one hash) after the
 823         * offset is available from this window, otherwise the offset
 824         * is not actually in this window and a different window (which
 825         * has that one hash excess) must be used.  This is to support
 826         * the object header and delta base parsing routines below.
 827         */
 828        off_t win_off = win->offset;
 829        return win_off <= offset
 830                && (offset + 20) <= (win_off + win->len);
 831}
 832
 833unsigned char *use_pack(struct packed_git *p,
 834                struct pack_window **w_cursor,
 835                off_t offset,
 836                unsigned int *left)
 837{
 838        struct pack_window *win = *w_cursor;
 839
 840        /* Since packfiles end in a hash of their content and it's
 841         * pointless to ask for an offset into the middle of that
 842         * hash, and the in_window function above wouldn't match
 843         * don't allow an offset too close to the end of the file.
 844         */
 845        if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
 846                die("packfile %s cannot be accessed", p->pack_name);
 847        if (offset > (p->pack_size - 20))
 848                die("offset beyond end of packfile (truncated pack?)");
 849
 850        if (!win || !in_window(win, offset)) {
 851                if (win)
 852                        win->inuse_cnt--;
 853                for (win = p->windows; win; win = win->next) {
 854                        if (in_window(win, offset))
 855                                break;
 856                }
 857                if (!win) {
 858                        size_t window_align = packed_git_window_size / 2;
 859                        off_t len;
 860
 861                        if (p->pack_fd == -1 && open_packed_git(p))
 862                                die("packfile %s cannot be accessed", p->pack_name);
 863
 864                        win = xcalloc(1, sizeof(*win));
 865                        win->offset = (offset / window_align) * window_align;
 866                        len = p->pack_size - win->offset;
 867                        if (len > packed_git_window_size)
 868                                len = packed_git_window_size;
 869                        win->len = (size_t)len;
 870                        pack_mapped += win->len;
 871                        while (packed_git_limit < pack_mapped
 872                                && unuse_one_window(p, p->pack_fd))
 873                                ; /* nothing */
 874                        win->base = xmmap(NULL, win->len,
 875                                PROT_READ, MAP_PRIVATE,
 876                                p->pack_fd, win->offset);
 877                        if (win->base == MAP_FAILED)
 878                                die("packfile %s cannot be mapped: %s",
 879                                        p->pack_name,
 880                                        strerror(errno));
 881                        if (!win->offset && win->len == p->pack_size
 882                                && !p->do_not_close) {
 883                                close(p->pack_fd);
 884                                pack_open_fds--;
 885                                p->pack_fd = -1;
 886                        }
 887                        pack_mmap_calls++;
 888                        pack_open_windows++;
 889                        if (pack_mapped > peak_pack_mapped)
 890                                peak_pack_mapped = pack_mapped;
 891                        if (pack_open_windows > peak_pack_open_windows)
 892                                peak_pack_open_windows = pack_open_windows;
 893                        win->next = p->windows;
 894                        p->windows = win;
 895                }
 896        }
 897        if (win != *w_cursor) {
 898                win->last_used = pack_used_ctr++;
 899                win->inuse_cnt++;
 900                *w_cursor = win;
 901        }
 902        offset -= win->offset;
 903        if (left)
 904                *left = win->len - xsize_t(offset);
 905        return win->base + offset;
 906}
 907
 908static struct packed_git *alloc_packed_git(int extra)
 909{
 910        struct packed_git *p = xmalloc(sizeof(*p) + extra);
 911        memset(p, 0, sizeof(*p));
 912        p->pack_fd = -1;
 913        return p;
 914}
 915
 916static void try_to_free_pack_memory(size_t size)
 917{
 918        release_pack_memory(size, -1);
 919}
 920
 921struct packed_git *add_packed_git(const char *path, int path_len, int local)
 922{
 923        static int have_set_try_to_free_routine;
 924        struct stat st;
 925        struct packed_git *p = alloc_packed_git(path_len + 2);
 926
 927        if (!have_set_try_to_free_routine) {
 928                have_set_try_to_free_routine = 1;
 929                set_try_to_free_routine(try_to_free_pack_memory);
 930        }
 931
 932        /*
 933         * Make sure a corresponding .pack file exists and that
 934         * the index looks sane.
 935         */
 936        path_len -= strlen(".idx");
 937        if (path_len < 1) {
 938                free(p);
 939                return NULL;
 940        }
 941        memcpy(p->pack_name, path, path_len);
 942
 943        strcpy(p->pack_name + path_len, ".keep");
 944        if (!access(p->pack_name, F_OK))
 945                p->pack_keep = 1;
 946
 947        strcpy(p->pack_name + path_len, ".pack");
 948        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 949                free(p);
 950                return NULL;
 951        }
 952
 953        /* ok, it looks sane as far as we can check without
 954         * actually mapping the pack file.
 955         */
 956        p->pack_size = st.st_size;
 957        p->pack_local = local;
 958        p->mtime = st.st_mtime;
 959        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 960                hashclr(p->sha1);
 961        return p;
 962}
 963
 964struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
 965{
 966        const char *path = sha1_pack_name(sha1);
 967        struct packed_git *p = alloc_packed_git(strlen(path) + 1);
 968
 969        strcpy(p->pack_name, path);
 970        hashcpy(p->sha1, sha1);
 971        if (check_packed_git_idx(idx_path, p)) {
 972                free(p);
 973                return NULL;
 974        }
 975
 976        return p;
 977}
 978
 979void install_packed_git(struct packed_git *pack)
 980{
 981        if (pack->pack_fd != -1)
 982                pack_open_fds++;
 983
 984        pack->next = packed_git;
 985        packed_git = pack;
 986}
 987
 988static void prepare_packed_git_one(char *objdir, int local)
 989{
 990        /* Ensure that this buffer is large enough so that we can
 991           append "/pack/" without clobbering the stack even if
 992           strlen(objdir) were PATH_MAX.  */
 993        char path[PATH_MAX + 1 + 4 + 1 + 1];
 994        int len;
 995        DIR *dir;
 996        struct dirent *de;
 997
 998        sprintf(path, "%s/pack", objdir);
 999        len = strlen(path);
1000        dir = opendir(path);
1001        if (!dir) {
1002                if (errno != ENOENT)
1003                        error("unable to open object pack directory: %s: %s",
1004                              path, strerror(errno));
1005                return;
1006        }
1007        path[len++] = '/';
1008        while ((de = readdir(dir)) != NULL) {
1009                int namelen = strlen(de->d_name);
1010                struct packed_git *p;
1011
1012                if (!has_extension(de->d_name, ".idx"))
1013                        continue;
1014
1015                if (len + namelen + 1 > sizeof(path))
1016                        continue;
1017
1018                /* Don't reopen a pack we already have. */
1019                strcpy(path + len, de->d_name);
1020                for (p = packed_git; p; p = p->next) {
1021                        if (!memcmp(path, p->pack_name, len + namelen - 4))
1022                                break;
1023                }
1024                if (p)
1025                        continue;
1026                /* See if it really is a valid .idx file with corresponding
1027                 * .pack file that we can map.
1028                 */
1029                p = add_packed_git(path, len + namelen, local);
1030                if (!p)
1031                        continue;
1032                install_packed_git(p);
1033        }
1034        closedir(dir);
1035}
1036
1037static int sort_pack(const void *a_, const void *b_)
1038{
1039        struct packed_git *a = *((struct packed_git **)a_);
1040        struct packed_git *b = *((struct packed_git **)b_);
1041        int st;
1042
1043        /*
1044         * Local packs tend to contain objects specific to our
1045         * variant of the project than remote ones.  In addition,
1046         * remote ones could be on a network mounted filesystem.
1047         * Favor local ones for these reasons.
1048         */
1049        st = a->pack_local - b->pack_local;
1050        if (st)
1051                return -st;
1052
1053        /*
1054         * Younger packs tend to contain more recent objects,
1055         * and more recent objects tend to get accessed more
1056         * often.
1057         */
1058        if (a->mtime < b->mtime)
1059                return 1;
1060        else if (a->mtime == b->mtime)
1061                return 0;
1062        return -1;
1063}
1064
1065static void rearrange_packed_git(void)
1066{
1067        struct packed_git **ary, *p;
1068        int i, n;
1069
1070        for (n = 0, p = packed_git; p; p = p->next)
1071                n++;
1072        if (n < 2)
1073                return;
1074
1075        /* prepare an array of packed_git for easier sorting */
1076        ary = xcalloc(n, sizeof(struct packed_git *));
1077        for (n = 0, p = packed_git; p; p = p->next)
1078                ary[n++] = p;
1079
1080        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1081
1082        /* link them back again */
1083        for (i = 0; i < n - 1; i++)
1084                ary[i]->next = ary[i + 1];
1085        ary[n - 1]->next = NULL;
1086        packed_git = ary[0];
1087
1088        free(ary);
1089}
1090
1091static int prepare_packed_git_run_once = 0;
1092void prepare_packed_git(void)
1093{
1094        struct alternate_object_database *alt;
1095
1096        if (prepare_packed_git_run_once)
1097                return;
1098        prepare_packed_git_one(get_object_directory(), 1);
1099        prepare_alt_odb();
1100        for (alt = alt_odb_list; alt; alt = alt->next) {
1101                alt->name[-1] = 0;
1102                prepare_packed_git_one(alt->base, 0);
1103                alt->name[-1] = '/';
1104        }
1105        rearrange_packed_git();
1106        prepare_packed_git_run_once = 1;
1107}
1108
1109void reprepare_packed_git(void)
1110{
1111        discard_revindex();
1112        prepare_packed_git_run_once = 0;
1113        prepare_packed_git();
1114}
1115
1116static void mark_bad_packed_object(struct packed_git *p,
1117                                   const unsigned char *sha1)
1118{
1119        unsigned i;
1120        for (i = 0; i < p->num_bad_objects; i++)
1121                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1122                        return;
1123        p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1124        hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1125        p->num_bad_objects++;
1126}
1127
1128static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1129{
1130        struct packed_git *p;
1131        unsigned i;
1132
1133        for (p = packed_git; p; p = p->next)
1134                for (i = 0; i < p->num_bad_objects; i++)
1135                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1136                                return p;
1137        return NULL;
1138}
1139
1140int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1141{
1142        unsigned char real_sha1[20];
1143        hash_sha1_file(map, size, type, real_sha1);
1144        return hashcmp(sha1, real_sha1) ? -1 : 0;
1145}
1146
1147static int git_open_noatime(const char *name)
1148{
1149        static int sha1_file_open_flag = O_NOATIME;
1150
1151        for (;;) {
1152                int fd = open(name, O_RDONLY | sha1_file_open_flag);
1153                if (fd >= 0)
1154                        return fd;
1155
1156                /* Might the failure be due to O_NOATIME? */
1157                if (errno != ENOENT && sha1_file_open_flag) {
1158                        sha1_file_open_flag = 0;
1159                        continue;
1160                }
1161
1162                return -1;
1163        }
1164}
1165
1166static int open_sha1_file(const unsigned char *sha1)
1167{
1168        int fd;
1169        char *name = sha1_file_name(sha1);
1170        struct alternate_object_database *alt;
1171
1172        fd = git_open_noatime(name);
1173        if (fd >= 0)
1174                return fd;
1175
1176        prepare_alt_odb();
1177        errno = ENOENT;
1178        for (alt = alt_odb_list; alt; alt = alt->next) {
1179                name = alt->name;
1180                fill_sha1_path(name, sha1);
1181                fd = git_open_noatime(alt->base);
1182                if (fd >= 0)
1183                        return fd;
1184        }
1185        return -1;
1186}
1187
1188static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1189{
1190        void *map;
1191        int fd;
1192
1193        fd = open_sha1_file(sha1);
1194        map = NULL;
1195        if (fd >= 0) {
1196                struct stat st;
1197
1198                if (!fstat(fd, &st)) {
1199                        *size = xsize_t(st.st_size);
1200                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1201                }
1202                close(fd);
1203        }
1204        return map;
1205}
1206
1207static int legacy_loose_object(unsigned char *map)
1208{
1209        unsigned int word;
1210
1211        /*
1212         * Is it a zlib-compressed buffer? If so, the first byte
1213         * must be 0x78 (15-bit window size, deflated), and the
1214         * first 16-bit word is evenly divisible by 31
1215         */
1216        word = (map[0] << 8) + map[1];
1217        if (map[0] == 0x78 && !(word % 31))
1218                return 1;
1219        else
1220                return 0;
1221}
1222
1223unsigned long unpack_object_header_buffer(const unsigned char *buf,
1224                unsigned long len, enum object_type *type, unsigned long *sizep)
1225{
1226        unsigned shift;
1227        unsigned long size, c;
1228        unsigned long used = 0;
1229
1230        c = buf[used++];
1231        *type = (c >> 4) & 7;
1232        size = c & 15;
1233        shift = 4;
1234        while (c & 0x80) {
1235                if (len <= used || bitsizeof(long) <= shift) {
1236                        error("bad object header");
1237                        return 0;
1238                }
1239                c = buf[used++];
1240                size += (c & 0x7f) << shift;
1241                shift += 7;
1242        }
1243        *sizep = size;
1244        return used;
1245}
1246
1247static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1248{
1249        unsigned long size, used;
1250        static const char valid_loose_object_type[8] = {
1251                0, /* OBJ_EXT */
1252                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1253                0, /* "delta" and others are invalid in a loose object */
1254        };
1255        enum object_type type;
1256
1257        /* Get the data stream */
1258        memset(stream, 0, sizeof(*stream));
1259        stream->next_in = map;
1260        stream->avail_in = mapsize;
1261        stream->next_out = buffer;
1262        stream->avail_out = bufsiz;
1263
1264        if (legacy_loose_object(map)) {
1265                git_inflate_init(stream);
1266                return git_inflate(stream, 0);
1267        }
1268
1269
1270        /*
1271         * There used to be a second loose object header format which
1272         * was meant to mimic the in-pack format, allowing for direct
1273         * copy of the object data.  This format turned up not to be
1274         * really worth it and we don't write it any longer.  But we
1275         * can still read it.
1276         */
1277        used = unpack_object_header_buffer(map, mapsize, &type, &size);
1278        if (!used || !valid_loose_object_type[type])
1279                return -1;
1280        map += used;
1281        mapsize -= used;
1282
1283        /* Set up the stream for the rest.. */
1284        stream->next_in = map;
1285        stream->avail_in = mapsize;
1286        git_inflate_init(stream);
1287
1288        /* And generate the fake traditional header */
1289        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1290                                         typename(type), size);
1291        return 0;
1292}
1293
1294static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1295{
1296        int bytes = strlen(buffer) + 1;
1297        unsigned char *buf = xmallocz(size);
1298        unsigned long n;
1299        int status = Z_OK;
1300
1301        n = stream->total_out - bytes;
1302        if (n > size)
1303                n = size;
1304        memcpy(buf, (char *) buffer + bytes, n);
1305        bytes = n;
1306        if (bytes <= size) {
1307                /*
1308                 * The above condition must be (bytes <= size), not
1309                 * (bytes < size).  In other words, even though we
1310                 * expect no more output and set avail_out to zero,
1311                 * the input zlib stream may have bytes that express
1312                 * "this concludes the stream", and we *do* want to
1313                 * eat that input.
1314                 *
1315                 * Otherwise we would not be able to test that we
1316                 * consumed all the input to reach the expected size;
1317                 * we also want to check that zlib tells us that all
1318                 * went well with status == Z_STREAM_END at the end.
1319                 */
1320                stream->next_out = buf + bytes;
1321                stream->avail_out = size - bytes;
1322                while (status == Z_OK)
1323                        status = git_inflate(stream, Z_FINISH);
1324        }
1325        if (status == Z_STREAM_END && !stream->avail_in) {
1326                git_inflate_end(stream);
1327                return buf;
1328        }
1329
1330        if (status < 0)
1331                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1332        else if (stream->avail_in)
1333                error("garbage at end of loose object '%s'",
1334                      sha1_to_hex(sha1));
1335        free(buf);
1336        return NULL;
1337}
1338
1339/*
1340 * We used to just use "sscanf()", but that's actually way
1341 * too permissive for what we want to check. So do an anal
1342 * object header parse by hand.
1343 */
1344static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1345{
1346        char type[10];
1347        int i;
1348        unsigned long size;
1349
1350        /*
1351         * The type can be at most ten bytes (including the
1352         * terminating '\0' that we add), and is followed by
1353         * a space.
1354         */
1355        i = 0;
1356        for (;;) {
1357                char c = *hdr++;
1358                if (c == ' ')
1359                        break;
1360                type[i++] = c;
1361                if (i >= sizeof(type))
1362                        return -1;
1363        }
1364        type[i] = 0;
1365
1366        /*
1367         * The length must follow immediately, and be in canonical
1368         * decimal format (ie "010" is not valid).
1369         */
1370        size = *hdr++ - '0';
1371        if (size > 9)
1372                return -1;
1373        if (size) {
1374                for (;;) {
1375                        unsigned long c = *hdr - '0';
1376                        if (c > 9)
1377                                break;
1378                        hdr++;
1379                        size = size * 10 + c;
1380                }
1381        }
1382        *sizep = size;
1383
1384        /*
1385         * The length must be followed by a zero byte
1386         */
1387        return *hdr ? -1 : type_from_string(type);
1388}
1389
1390static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1391{
1392        int ret;
1393        z_stream stream;
1394        char hdr[8192];
1395
1396        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1397        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1398                return NULL;
1399
1400        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1401}
1402
1403unsigned long get_size_from_delta(struct packed_git *p,
1404                                  struct pack_window **w_curs,
1405                                  off_t curpos)
1406{
1407        const unsigned char *data;
1408        unsigned char delta_head[20], *in;
1409        z_stream stream;
1410        int st;
1411
1412        memset(&stream, 0, sizeof(stream));
1413        stream.next_out = delta_head;
1414        stream.avail_out = sizeof(delta_head);
1415
1416        git_inflate_init(&stream);
1417        do {
1418                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1419                stream.next_in = in;
1420                st = git_inflate(&stream, Z_FINISH);
1421                curpos += stream.next_in - in;
1422        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1423                 stream.total_out < sizeof(delta_head));
1424        git_inflate_end(&stream);
1425        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1426                error("delta data unpack-initial failed");
1427                return 0;
1428        }
1429
1430        /* Examine the initial part of the delta to figure out
1431         * the result size.
1432         */
1433        data = delta_head;
1434
1435        /* ignore base size */
1436        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1437
1438        /* Read the result size */
1439        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1440}
1441
1442static off_t get_delta_base(struct packed_git *p,
1443                                    struct pack_window **w_curs,
1444                                    off_t *curpos,
1445                                    enum object_type type,
1446                                    off_t delta_obj_offset)
1447{
1448        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1449        off_t base_offset;
1450
1451        /* use_pack() assured us we have [base_info, base_info + 20)
1452         * as a range that we can look at without walking off the
1453         * end of the mapped window.  Its actually the hash size
1454         * that is assured.  An OFS_DELTA longer than the hash size
1455         * is stupid, as then a REF_DELTA would be smaller to store.
1456         */
1457        if (type == OBJ_OFS_DELTA) {
1458                unsigned used = 0;
1459                unsigned char c = base_info[used++];
1460                base_offset = c & 127;
1461                while (c & 128) {
1462                        base_offset += 1;
1463                        if (!base_offset || MSB(base_offset, 7))
1464                                return 0;  /* overflow */
1465                        c = base_info[used++];
1466                        base_offset = (base_offset << 7) + (c & 127);
1467                }
1468                base_offset = delta_obj_offset - base_offset;
1469                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1470                        return 0;  /* out of bound */
1471                *curpos += used;
1472        } else if (type == OBJ_REF_DELTA) {
1473                /* The base entry _must_ be in the same pack */
1474                base_offset = find_pack_entry_one(base_info, p);
1475                *curpos += 20;
1476        } else
1477                die("I am totally screwed");
1478        return base_offset;
1479}
1480
1481/* forward declaration for a mutually recursive function */
1482static int packed_object_info(struct packed_git *p, off_t offset,
1483                              unsigned long *sizep);
1484
1485static int packed_delta_info(struct packed_git *p,
1486                             struct pack_window **w_curs,
1487                             off_t curpos,
1488                             enum object_type type,
1489                             off_t obj_offset,
1490                             unsigned long *sizep)
1491{
1492        off_t base_offset;
1493
1494        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1495        if (!base_offset)
1496                return OBJ_BAD;
1497        type = packed_object_info(p, base_offset, NULL);
1498        if (type <= OBJ_NONE) {
1499                struct revindex_entry *revidx;
1500                const unsigned char *base_sha1;
1501                revidx = find_pack_revindex(p, base_offset);
1502                if (!revidx)
1503                        return OBJ_BAD;
1504                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1505                mark_bad_packed_object(p, base_sha1);
1506                type = sha1_object_info(base_sha1, NULL);
1507                if (type <= OBJ_NONE)
1508                        return OBJ_BAD;
1509        }
1510
1511        /* We choose to only get the type of the base object and
1512         * ignore potentially corrupt pack file that expects the delta
1513         * based on a base with a wrong size.  This saves tons of
1514         * inflate() calls.
1515         */
1516        if (sizep) {
1517                *sizep = get_size_from_delta(p, w_curs, curpos);
1518                if (*sizep == 0)
1519                        type = OBJ_BAD;
1520        }
1521
1522        return type;
1523}
1524
1525static int unpack_object_header(struct packed_git *p,
1526                                struct pack_window **w_curs,
1527                                off_t *curpos,
1528                                unsigned long *sizep)
1529{
1530        unsigned char *base;
1531        unsigned int left;
1532        unsigned long used;
1533        enum object_type type;
1534
1535        /* use_pack() assures us we have [base, base + 20) available
1536         * as a range that we can look at.  (Its actually the hash
1537         * size that is assured.)  With our object header encoding
1538         * the maximum deflated object size is 2^137, which is just
1539         * insane, so we know won't exceed what we have been given.
1540         */
1541        base = use_pack(p, w_curs, *curpos, &left);
1542        used = unpack_object_header_buffer(base, left, &type, sizep);
1543        if (!used) {
1544                type = OBJ_BAD;
1545        } else
1546                *curpos += used;
1547
1548        return type;
1549}
1550
1551const char *packed_object_info_detail(struct packed_git *p,
1552                                      off_t obj_offset,
1553                                      unsigned long *size,
1554                                      unsigned long *store_size,
1555                                      unsigned int *delta_chain_length,
1556                                      unsigned char *base_sha1)
1557{
1558        struct pack_window *w_curs = NULL;
1559        off_t curpos;
1560        unsigned long dummy;
1561        unsigned char *next_sha1;
1562        enum object_type type;
1563        struct revindex_entry *revidx;
1564
1565        *delta_chain_length = 0;
1566        curpos = obj_offset;
1567        type = unpack_object_header(p, &w_curs, &curpos, size);
1568
1569        revidx = find_pack_revindex(p, obj_offset);
1570        *store_size = revidx[1].offset - obj_offset;
1571
1572        for (;;) {
1573                switch (type) {
1574                default:
1575                        die("pack %s contains unknown object type %d",
1576                            p->pack_name, type);
1577                case OBJ_COMMIT:
1578                case OBJ_TREE:
1579                case OBJ_BLOB:
1580                case OBJ_TAG:
1581                        unuse_pack(&w_curs);
1582                        return typename(type);
1583                case OBJ_OFS_DELTA:
1584                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1585                        if (!obj_offset)
1586                                die("pack %s contains bad delta base reference of type %s",
1587                                    p->pack_name, typename(type));
1588                        if (*delta_chain_length == 0) {
1589                                revidx = find_pack_revindex(p, obj_offset);
1590                                hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1591                        }
1592                        break;
1593                case OBJ_REF_DELTA:
1594                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1595                        if (*delta_chain_length == 0)
1596                                hashcpy(base_sha1, next_sha1);
1597                        obj_offset = find_pack_entry_one(next_sha1, p);
1598                        break;
1599                }
1600                (*delta_chain_length)++;
1601                curpos = obj_offset;
1602                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1603        }
1604}
1605
1606static int packed_object_info(struct packed_git *p, off_t obj_offset,
1607                              unsigned long *sizep)
1608{
1609        struct pack_window *w_curs = NULL;
1610        unsigned long size;
1611        off_t curpos = obj_offset;
1612        enum object_type type;
1613
1614        type = unpack_object_header(p, &w_curs, &curpos, &size);
1615
1616        switch (type) {
1617        case OBJ_OFS_DELTA:
1618        case OBJ_REF_DELTA:
1619                type = packed_delta_info(p, &w_curs, curpos,
1620                                         type, obj_offset, sizep);
1621                break;
1622        case OBJ_COMMIT:
1623        case OBJ_TREE:
1624        case OBJ_BLOB:
1625        case OBJ_TAG:
1626                if (sizep)
1627                        *sizep = size;
1628                break;
1629        default:
1630                error("unknown object type %i at offset %"PRIuMAX" in %s",
1631                      type, (uintmax_t)obj_offset, p->pack_name);
1632                type = OBJ_BAD;
1633        }
1634        unuse_pack(&w_curs);
1635        return type;
1636}
1637
1638static void *unpack_compressed_entry(struct packed_git *p,
1639                                    struct pack_window **w_curs,
1640                                    off_t curpos,
1641                                    unsigned long size)
1642{
1643        int st;
1644        z_stream stream;
1645        unsigned char *buffer, *in;
1646
1647        buffer = xmallocz(size);
1648        memset(&stream, 0, sizeof(stream));
1649        stream.next_out = buffer;
1650        stream.avail_out = size + 1;
1651
1652        git_inflate_init(&stream);
1653        do {
1654                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1655                stream.next_in = in;
1656                st = git_inflate(&stream, Z_FINISH);
1657                if (!stream.avail_out)
1658                        break; /* the payload is larger than it should be */
1659                curpos += stream.next_in - in;
1660        } while (st == Z_OK || st == Z_BUF_ERROR);
1661        git_inflate_end(&stream);
1662        if ((st != Z_STREAM_END) || stream.total_out != size) {
1663                free(buffer);
1664                return NULL;
1665        }
1666
1667        return buffer;
1668}
1669
1670#define MAX_DELTA_CACHE (256)
1671
1672static size_t delta_base_cached;
1673
1674static struct delta_base_cache_lru_list {
1675        struct delta_base_cache_lru_list *prev;
1676        struct delta_base_cache_lru_list *next;
1677} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1678
1679static struct delta_base_cache_entry {
1680        struct delta_base_cache_lru_list lru;
1681        void *data;
1682        struct packed_git *p;
1683        off_t base_offset;
1684        unsigned long size;
1685        enum object_type type;
1686} delta_base_cache[MAX_DELTA_CACHE];
1687
1688static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1689{
1690        unsigned long hash;
1691
1692        hash = (unsigned long)p + (unsigned long)base_offset;
1693        hash += (hash >> 8) + (hash >> 16);
1694        return hash % MAX_DELTA_CACHE;
1695}
1696
1697static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1698        unsigned long *base_size, enum object_type *type, int keep_cache)
1699{
1700        void *ret;
1701        unsigned long hash = pack_entry_hash(p, base_offset);
1702        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1703
1704        ret = ent->data;
1705        if (!ret || ent->p != p || ent->base_offset != base_offset)
1706                return unpack_entry(p, base_offset, type, base_size);
1707
1708        if (!keep_cache) {
1709                ent->data = NULL;
1710                ent->lru.next->prev = ent->lru.prev;
1711                ent->lru.prev->next = ent->lru.next;
1712                delta_base_cached -= ent->size;
1713        } else {
1714                ret = xmemdupz(ent->data, ent->size);
1715        }
1716        *type = ent->type;
1717        *base_size = ent->size;
1718        return ret;
1719}
1720
1721static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1722{
1723        if (ent->data) {
1724                free(ent->data);
1725                ent->data = NULL;
1726                ent->lru.next->prev = ent->lru.prev;
1727                ent->lru.prev->next = ent->lru.next;
1728                delta_base_cached -= ent->size;
1729        }
1730}
1731
1732void clear_delta_base_cache(void)
1733{
1734        unsigned long p;
1735        for (p = 0; p < MAX_DELTA_CACHE; p++)
1736                release_delta_base_cache(&delta_base_cache[p]);
1737}
1738
1739static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1740        void *base, unsigned long base_size, enum object_type type)
1741{
1742        unsigned long hash = pack_entry_hash(p, base_offset);
1743        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1744        struct delta_base_cache_lru_list *lru;
1745
1746        release_delta_base_cache(ent);
1747        delta_base_cached += base_size;
1748
1749        for (lru = delta_base_cache_lru.next;
1750             delta_base_cached > delta_base_cache_limit
1751             && lru != &delta_base_cache_lru;
1752             lru = lru->next) {
1753                struct delta_base_cache_entry *f = (void *)lru;
1754                if (f->type == OBJ_BLOB)
1755                        release_delta_base_cache(f);
1756        }
1757        for (lru = delta_base_cache_lru.next;
1758             delta_base_cached > delta_base_cache_limit
1759             && lru != &delta_base_cache_lru;
1760             lru = lru->next) {
1761                struct delta_base_cache_entry *f = (void *)lru;
1762                release_delta_base_cache(f);
1763        }
1764
1765        ent->p = p;
1766        ent->base_offset = base_offset;
1767        ent->type = type;
1768        ent->data = base;
1769        ent->size = base_size;
1770        ent->lru.next = &delta_base_cache_lru;
1771        ent->lru.prev = delta_base_cache_lru.prev;
1772        delta_base_cache_lru.prev->next = &ent->lru;
1773        delta_base_cache_lru.prev = &ent->lru;
1774}
1775
1776static void *read_object(const unsigned char *sha1, enum object_type *type,
1777                         unsigned long *size);
1778
1779static void *unpack_delta_entry(struct packed_git *p,
1780                                struct pack_window **w_curs,
1781                                off_t curpos,
1782                                unsigned long delta_size,
1783                                off_t obj_offset,
1784                                enum object_type *type,
1785                                unsigned long *sizep)
1786{
1787        void *delta_data, *result, *base;
1788        unsigned long base_size;
1789        off_t base_offset;
1790
1791        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1792        if (!base_offset) {
1793                error("failed to validate delta base reference "
1794                      "at offset %"PRIuMAX" from %s",
1795                      (uintmax_t)curpos, p->pack_name);
1796                return NULL;
1797        }
1798        unuse_pack(w_curs);
1799        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1800        if (!base) {
1801                /*
1802                 * We're probably in deep shit, but let's try to fetch
1803                 * the required base anyway from another pack or loose.
1804                 * This is costly but should happen only in the presence
1805                 * of a corrupted pack, and is better than failing outright.
1806                 */
1807                struct revindex_entry *revidx;
1808                const unsigned char *base_sha1;
1809                revidx = find_pack_revindex(p, base_offset);
1810                if (!revidx)
1811                        return NULL;
1812                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1813                error("failed to read delta base object %s"
1814                      " at offset %"PRIuMAX" from %s",
1815                      sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1816                      p->pack_name);
1817                mark_bad_packed_object(p, base_sha1);
1818                base = read_object(base_sha1, type, &base_size);
1819                if (!base)
1820                        return NULL;
1821        }
1822
1823        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1824        if (!delta_data) {
1825                error("failed to unpack compressed delta "
1826                      "at offset %"PRIuMAX" from %s",
1827                      (uintmax_t)curpos, p->pack_name);
1828                free(base);
1829                return NULL;
1830        }
1831        result = patch_delta(base, base_size,
1832                             delta_data, delta_size,
1833                             sizep);
1834        if (!result)
1835                die("failed to apply delta");
1836        free(delta_data);
1837        add_delta_base_cache(p, base_offset, base, base_size, *type);
1838        return result;
1839}
1840
1841int do_check_packed_object_crc;
1842
1843void *unpack_entry(struct packed_git *p, off_t obj_offset,
1844                   enum object_type *type, unsigned long *sizep)
1845{
1846        struct pack_window *w_curs = NULL;
1847        off_t curpos = obj_offset;
1848        void *data;
1849
1850        if (do_check_packed_object_crc && p->index_version > 1) {
1851                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1852                unsigned long len = revidx[1].offset - obj_offset;
1853                if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1854                        const unsigned char *sha1 =
1855                                nth_packed_object_sha1(p, revidx->nr);
1856                        error("bad packed object CRC for %s",
1857                              sha1_to_hex(sha1));
1858                        mark_bad_packed_object(p, sha1);
1859                        unuse_pack(&w_curs);
1860                        return NULL;
1861                }
1862        }
1863
1864        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1865        switch (*type) {
1866        case OBJ_OFS_DELTA:
1867        case OBJ_REF_DELTA:
1868                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1869                                          obj_offset, type, sizep);
1870                break;
1871        case OBJ_COMMIT:
1872        case OBJ_TREE:
1873        case OBJ_BLOB:
1874        case OBJ_TAG:
1875                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1876                break;
1877        default:
1878                data = NULL;
1879                error("unknown object type %i at offset %"PRIuMAX" in %s",
1880                      *type, (uintmax_t)obj_offset, p->pack_name);
1881        }
1882        unuse_pack(&w_curs);
1883        return data;
1884}
1885
1886const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1887                                            uint32_t n)
1888{
1889        const unsigned char *index = p->index_data;
1890        if (!index) {
1891                if (open_pack_index(p))
1892                        return NULL;
1893                index = p->index_data;
1894        }
1895        if (n >= p->num_objects)
1896                return NULL;
1897        index += 4 * 256;
1898        if (p->index_version == 1) {
1899                return index + 24 * n + 4;
1900        } else {
1901                index += 8;
1902                return index + 20 * n;
1903        }
1904}
1905
1906off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1907{
1908        const unsigned char *index = p->index_data;
1909        index += 4 * 256;
1910        if (p->index_version == 1) {
1911                return ntohl(*((uint32_t *)(index + 24 * n)));
1912        } else {
1913                uint32_t off;
1914                index += 8 + p->num_objects * (20 + 4);
1915                off = ntohl(*((uint32_t *)(index + 4 * n)));
1916                if (!(off & 0x80000000))
1917                        return off;
1918                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1919                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1920                                   ntohl(*((uint32_t *)(index + 4)));
1921        }
1922}
1923
1924off_t find_pack_entry_one(const unsigned char *sha1,
1925                                  struct packed_git *p)
1926{
1927        const uint32_t *level1_ofs = p->index_data;
1928        const unsigned char *index = p->index_data;
1929        unsigned hi, lo, stride;
1930        static int use_lookup = -1;
1931        static int debug_lookup = -1;
1932
1933        if (debug_lookup < 0)
1934                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1935
1936        if (!index) {
1937                if (open_pack_index(p))
1938                        return 0;
1939                level1_ofs = p->index_data;
1940                index = p->index_data;
1941        }
1942        if (p->index_version > 1) {
1943                level1_ofs += 2;
1944                index += 8;
1945        }
1946        index += 4 * 256;
1947        hi = ntohl(level1_ofs[*sha1]);
1948        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1949        if (p->index_version > 1) {
1950                stride = 20;
1951        } else {
1952                stride = 24;
1953                index += 4;
1954        }
1955
1956        if (debug_lookup)
1957                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1958                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1959
1960        if (use_lookup < 0)
1961                use_lookup = !!getenv("GIT_USE_LOOKUP");
1962        if (use_lookup) {
1963                int pos = sha1_entry_pos(index, stride, 0,
1964                                         lo, hi, p->num_objects, sha1);
1965                if (pos < 0)
1966                        return 0;
1967                return nth_packed_object_offset(p, pos);
1968        }
1969
1970        do {
1971                unsigned mi = (lo + hi) / 2;
1972                int cmp = hashcmp(index + mi * stride, sha1);
1973
1974                if (debug_lookup)
1975                        printf("lo %u hi %u rg %u mi %u\n",
1976                               lo, hi, hi - lo, mi);
1977                if (!cmp)
1978                        return nth_packed_object_offset(p, mi);
1979                if (cmp > 0)
1980                        hi = mi;
1981                else
1982                        lo = mi+1;
1983        } while (lo < hi);
1984        return 0;
1985}
1986
1987static int is_pack_valid(struct packed_git *p)
1988{
1989        /* An already open pack is known to be valid. */
1990        if (p->pack_fd != -1)
1991                return 1;
1992
1993        /* If the pack has one window completely covering the
1994         * file size, the pack is known to be valid even if
1995         * the descriptor is not currently open.
1996         */
1997        if (p->windows) {
1998                struct pack_window *w = p->windows;
1999
2000                if (!w->offset && w->len == p->pack_size)
2001                        return 1;
2002        }
2003
2004        /* Force the pack to open to prove its valid. */
2005        return !open_packed_git(p);
2006}
2007
2008static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2009{
2010        static struct packed_git *last_found = (void *)1;
2011        struct packed_git *p;
2012        off_t offset;
2013
2014        prepare_packed_git();
2015        if (!packed_git)
2016                return 0;
2017        p = (last_found == (void *)1) ? packed_git : last_found;
2018
2019        do {
2020                if (p->num_bad_objects) {
2021                        unsigned i;
2022                        for (i = 0; i < p->num_bad_objects; i++)
2023                                if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2024                                        goto next;
2025                }
2026
2027                offset = find_pack_entry_one(sha1, p);
2028                if (offset) {
2029                        /*
2030                         * We are about to tell the caller where they can
2031                         * locate the requested object.  We better make
2032                         * sure the packfile is still here and can be
2033                         * accessed before supplying that answer, as
2034                         * it may have been deleted since the index
2035                         * was loaded!
2036                         */
2037                        if (!is_pack_valid(p)) {
2038                                error("packfile %s cannot be accessed", p->pack_name);
2039                                goto next;
2040                        }
2041                        e->offset = offset;
2042                        e->p = p;
2043                        hashcpy(e->sha1, sha1);
2044                        last_found = p;
2045                        return 1;
2046                }
2047
2048                next:
2049                if (p == last_found)
2050                        p = packed_git;
2051                else
2052                        p = p->next;
2053                if (p == last_found)
2054                        p = p->next;
2055        } while (p);
2056        return 0;
2057}
2058
2059struct packed_git *find_sha1_pack(const unsigned char *sha1,
2060                                  struct packed_git *packs)
2061{
2062        struct packed_git *p;
2063
2064        for (p = packs; p; p = p->next) {
2065                if (find_pack_entry_one(sha1, p))
2066                        return p;
2067        }
2068        return NULL;
2069
2070}
2071
2072static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2073{
2074        int status;
2075        unsigned long mapsize, size;
2076        void *map;
2077        z_stream stream;
2078        char hdr[32];
2079
2080        map = map_sha1_file(sha1, &mapsize);
2081        if (!map)
2082                return error("unable to find %s", sha1_to_hex(sha1));
2083        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2084                status = error("unable to unpack %s header",
2085                               sha1_to_hex(sha1));
2086        else if ((status = parse_sha1_header(hdr, &size)) < 0)
2087                status = error("unable to parse %s header", sha1_to_hex(sha1));
2088        else if (sizep)
2089                *sizep = size;
2090        git_inflate_end(&stream);
2091        munmap(map, mapsize);
2092        return status;
2093}
2094
2095int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2096{
2097        struct cached_object *co;
2098        struct pack_entry e;
2099        int status;
2100
2101        co = find_cached_object(sha1);
2102        if (co) {
2103                if (sizep)
2104                        *sizep = co->size;
2105                return co->type;
2106        }
2107
2108        if (!find_pack_entry(sha1, &e)) {
2109                /* Most likely it's a loose object. */
2110                status = sha1_loose_object_info(sha1, sizep);
2111                if (status >= 0)
2112                        return status;
2113
2114                /* Not a loose object; someone else may have just packed it. */
2115                reprepare_packed_git();
2116                if (!find_pack_entry(sha1, &e))
2117                        return status;
2118        }
2119
2120        status = packed_object_info(e.p, e.offset, sizep);
2121        if (status < 0) {
2122                mark_bad_packed_object(e.p, sha1);
2123                status = sha1_object_info(sha1, sizep);
2124        }
2125
2126        return status;
2127}
2128
2129static void *read_packed_sha1(const unsigned char *sha1,
2130                              enum object_type *type, unsigned long *size)
2131{
2132        struct pack_entry e;
2133        void *data;
2134
2135        if (!find_pack_entry(sha1, &e))
2136                return NULL;
2137        data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2138        if (!data) {
2139                /*
2140                 * We're probably in deep shit, but let's try to fetch
2141                 * the required object anyway from another pack or loose.
2142                 * This should happen only in the presence of a corrupted
2143                 * pack, and is better than failing outright.
2144                 */
2145                error("failed to read object %s at offset %"PRIuMAX" from %s",
2146                      sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2147                mark_bad_packed_object(e.p, sha1);
2148                data = read_object(sha1, type, size);
2149        }
2150        return data;
2151}
2152
2153int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2154                      unsigned char *sha1)
2155{
2156        struct cached_object *co;
2157
2158        hash_sha1_file(buf, len, typename(type), sha1);
2159        if (has_sha1_file(sha1) || find_cached_object(sha1))
2160                return 0;
2161        if (cached_object_alloc <= cached_object_nr) {
2162                cached_object_alloc = alloc_nr(cached_object_alloc);
2163                cached_objects = xrealloc(cached_objects,
2164                                          sizeof(*cached_objects) *
2165                                          cached_object_alloc);
2166        }
2167        co = &cached_objects[cached_object_nr++];
2168        co->size = len;
2169        co->type = type;
2170        co->buf = xmalloc(len);
2171        memcpy(co->buf, buf, len);
2172        hashcpy(co->sha1, sha1);
2173        return 0;
2174}
2175
2176static void *read_object(const unsigned char *sha1, enum object_type *type,
2177                         unsigned long *size)
2178{
2179        unsigned long mapsize;
2180        void *map, *buf;
2181        struct cached_object *co;
2182
2183        co = find_cached_object(sha1);
2184        if (co) {
2185                *type = co->type;
2186                *size = co->size;
2187                return xmemdupz(co->buf, co->size);
2188        }
2189
2190        buf = read_packed_sha1(sha1, type, size);
2191        if (buf)
2192                return buf;
2193        map = map_sha1_file(sha1, &mapsize);
2194        if (map) {
2195                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2196                munmap(map, mapsize);
2197                return buf;
2198        }
2199        reprepare_packed_git();
2200        return read_packed_sha1(sha1, type, size);
2201}
2202
2203/*
2204 * This function dies on corrupt objects; the callers who want to
2205 * deal with them should arrange to call read_object() and give error
2206 * messages themselves.
2207 */
2208void *read_sha1_file_repl(const unsigned char *sha1,
2209                          enum object_type *type,
2210                          unsigned long *size,
2211                          const unsigned char **replacement)
2212{
2213        const unsigned char *repl = lookup_replace_object(sha1);
2214        void *data;
2215        char *path;
2216        const struct packed_git *p;
2217
2218        errno = 0;
2219        data = read_object(repl, type, size);
2220        if (data) {
2221                if (replacement)
2222                        *replacement = repl;
2223                return data;
2224        }
2225
2226        if (errno && errno != ENOENT)
2227                die_errno("failed to read object %s", sha1_to_hex(sha1));
2228
2229        /* die if we replaced an object with one that does not exist */
2230        if (repl != sha1)
2231                die("replacement %s not found for %s",
2232                    sha1_to_hex(repl), sha1_to_hex(sha1));
2233
2234        if (has_loose_object(repl)) {
2235                path = sha1_file_name(sha1);
2236                die("loose object %s (stored in %s) is corrupt",
2237                    sha1_to_hex(repl), path);
2238        }
2239
2240        if ((p = has_packed_and_bad(repl)) != NULL)
2241                die("packed object %s (stored in %s) is corrupt",
2242                    sha1_to_hex(repl), p->pack_name);
2243
2244        return NULL;
2245}
2246
2247void *read_object_with_reference(const unsigned char *sha1,
2248                                 const char *required_type_name,
2249                                 unsigned long *size,
2250                                 unsigned char *actual_sha1_return)
2251{
2252        enum object_type type, required_type;
2253        void *buffer;
2254        unsigned long isize;
2255        unsigned char actual_sha1[20];
2256
2257        required_type = type_from_string(required_type_name);
2258        hashcpy(actual_sha1, sha1);
2259        while (1) {
2260                int ref_length = -1;
2261                const char *ref_type = NULL;
2262
2263                buffer = read_sha1_file(actual_sha1, &type, &isize);
2264                if (!buffer)
2265                        return NULL;
2266                if (type == required_type) {
2267                        *size = isize;
2268                        if (actual_sha1_return)
2269                                hashcpy(actual_sha1_return, actual_sha1);
2270                        return buffer;
2271                }
2272                /* Handle references */
2273                else if (type == OBJ_COMMIT)
2274                        ref_type = "tree ";
2275                else if (type == OBJ_TAG)
2276                        ref_type = "object ";
2277                else {
2278                        free(buffer);
2279                        return NULL;
2280                }
2281                ref_length = strlen(ref_type);
2282
2283                if (ref_length + 40 > isize ||
2284                    memcmp(buffer, ref_type, ref_length) ||
2285                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2286                        free(buffer);
2287                        return NULL;
2288                }
2289                free(buffer);
2290                /* Now we have the ID of the referred-to object in
2291                 * actual_sha1.  Check again. */
2292        }
2293}
2294
2295static void write_sha1_file_prepare(const void *buf, unsigned long len,
2296                                    const char *type, unsigned char *sha1,
2297                                    char *hdr, int *hdrlen)
2298{
2299        git_SHA_CTX c;
2300
2301        /* Generate the header */
2302        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2303
2304        /* Sha1.. */
2305        git_SHA1_Init(&c);
2306        git_SHA1_Update(&c, hdr, *hdrlen);
2307        git_SHA1_Update(&c, buf, len);
2308        git_SHA1_Final(sha1, &c);
2309}
2310
2311/*
2312 * Move the just written object into its final resting place.
2313 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2314 * "moving" is only a part of what it does, when no patch between
2315 * master to pu changes the call sites of this function.
2316 */
2317int move_temp_to_file(const char *tmpfile, const char *filename)
2318{
2319        int ret = 0;
2320
2321        if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2322                goto try_rename;
2323        else if (link(tmpfile, filename))
2324                ret = errno;
2325
2326        /*
2327         * Coda hack - coda doesn't like cross-directory links,
2328         * so we fall back to a rename, which will mean that it
2329         * won't be able to check collisions, but that's not a
2330         * big deal.
2331         *
2332         * The same holds for FAT formatted media.
2333         *
2334         * When this succeeds, we just return.  We have nothing
2335         * left to unlink.
2336         */
2337        if (ret && ret != EEXIST) {
2338        try_rename:
2339                if (!rename(tmpfile, filename))
2340                        goto out;
2341                ret = errno;
2342        }
2343        unlink_or_warn(tmpfile);
2344        if (ret) {
2345                if (ret != EEXIST) {
2346                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2347                }
2348                /* FIXME!!! Collision check here ? */
2349        }
2350
2351out:
2352        if (adjust_shared_perm(filename))
2353                return error("unable to set permission to '%s'", filename);
2354        return 0;
2355}
2356
2357static int write_buffer(int fd, const void *buf, size_t len)
2358{
2359        if (write_in_full(fd, buf, len) < 0)
2360                return error("file write error (%s)", strerror(errno));
2361        return 0;
2362}
2363
2364int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2365                   unsigned char *sha1)
2366{
2367        char hdr[32];
2368        int hdrlen;
2369        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2370        return 0;
2371}
2372
2373/* Finalize a file on disk, and close it. */
2374static void close_sha1_file(int fd)
2375{
2376        if (fsync_object_files)
2377                fsync_or_die(fd, "sha1 file");
2378        if (close(fd) != 0)
2379                die_errno("error when closing sha1 file");
2380}
2381
2382/* Size of directory component, including the ending '/' */
2383static inline int directory_size(const char *filename)
2384{
2385        const char *s = strrchr(filename, '/');
2386        if (!s)
2387                return 0;
2388        return s - filename + 1;
2389}
2390
2391/*
2392 * This creates a temporary file in the same directory as the final
2393 * 'filename'
2394 *
2395 * We want to avoid cross-directory filename renames, because those
2396 * can have problems on various filesystems (FAT, NFS, Coda).
2397 */
2398static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2399{
2400        int fd, dirlen = directory_size(filename);
2401
2402        if (dirlen + 20 > bufsiz) {
2403                errno = ENAMETOOLONG;
2404                return -1;
2405        }
2406        memcpy(buffer, filename, dirlen);
2407        strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2408        fd = git_mkstemp_mode(buffer, 0444);
2409        if (fd < 0 && dirlen && errno == ENOENT) {
2410                /* Make sure the directory exists */
2411                memcpy(buffer, filename, dirlen);
2412                buffer[dirlen-1] = 0;
2413                if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2414                        return -1;
2415
2416                /* Try again */
2417                strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2418                fd = git_mkstemp_mode(buffer, 0444);
2419        }
2420        return fd;
2421}
2422
2423static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2424                              const void *buf, unsigned long len, time_t mtime)
2425{
2426        int fd, ret;
2427        unsigned char compressed[4096];
2428        z_stream stream;
2429        git_SHA_CTX c;
2430        unsigned char parano_sha1[20];
2431        char *filename;
2432        static char tmpfile[PATH_MAX];
2433
2434        filename = sha1_file_name(sha1);
2435        fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2436        if (fd < 0) {
2437                if (errno == EACCES)
2438                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2439                else
2440                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2441        }
2442
2443        /* Set it up */
2444        memset(&stream, 0, sizeof(stream));
2445        deflateInit(&stream, zlib_compression_level);
2446        stream.next_out = compressed;
2447        stream.avail_out = sizeof(compressed);
2448        git_SHA1_Init(&c);
2449
2450        /* First header.. */
2451        stream.next_in = (unsigned char *)hdr;
2452        stream.avail_in = hdrlen;
2453        while (deflate(&stream, 0) == Z_OK)
2454                /* nothing */;
2455        git_SHA1_Update(&c, hdr, hdrlen);
2456
2457        /* Then the data itself.. */
2458        stream.next_in = (void *)buf;
2459        stream.avail_in = len;
2460        do {
2461                unsigned char *in0 = stream.next_in;
2462                ret = deflate(&stream, Z_FINISH);
2463                git_SHA1_Update(&c, in0, stream.next_in - in0);
2464                if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2465                        die("unable to write sha1 file");
2466                stream.next_out = compressed;
2467                stream.avail_out = sizeof(compressed);
2468        } while (ret == Z_OK);
2469
2470        if (ret != Z_STREAM_END)
2471                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2472        ret = deflateEnd(&stream);
2473        if (ret != Z_OK)
2474                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2475        git_SHA1_Final(parano_sha1, &c);
2476        if (hashcmp(sha1, parano_sha1) != 0)
2477                die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2478
2479        close_sha1_file(fd);
2480
2481        if (mtime) {
2482                struct utimbuf utb;
2483                utb.actime = mtime;
2484                utb.modtime = mtime;
2485                if (utime(tmpfile, &utb) < 0)
2486                        warning("failed utime() on %s: %s",
2487                                tmpfile, strerror(errno));
2488        }
2489
2490        return move_temp_to_file(tmpfile, filename);
2491}
2492
2493int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2494{
2495        unsigned char sha1[20];
2496        char hdr[32];
2497        int hdrlen;
2498
2499        /* Normally if we have it in the pack then we do not bother writing
2500         * it out into .git/objects/??/?{38} file.
2501         */
2502        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2503        if (returnsha1)
2504                hashcpy(returnsha1, sha1);
2505        if (has_sha1_file(sha1))
2506                return 0;
2507        return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2508}
2509
2510int force_object_loose(const unsigned char *sha1, time_t mtime)
2511{
2512        void *buf;
2513        unsigned long len;
2514        enum object_type type;
2515        char hdr[32];
2516        int hdrlen;
2517        int ret;
2518
2519        if (has_loose_object(sha1))
2520                return 0;
2521        buf = read_packed_sha1(sha1, &type, &len);
2522        if (!buf)
2523                return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2524        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2525        ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2526        free(buf);
2527
2528        return ret;
2529}
2530
2531int has_pack_index(const unsigned char *sha1)
2532{
2533        struct stat st;
2534        if (stat(sha1_pack_index_name(sha1), &st))
2535                return 0;
2536        return 1;
2537}
2538
2539int has_sha1_pack(const unsigned char *sha1)
2540{
2541        struct pack_entry e;
2542        return find_pack_entry(sha1, &e);
2543}
2544
2545int has_sha1_file(const unsigned char *sha1)
2546{
2547        struct pack_entry e;
2548
2549        if (find_pack_entry(sha1, &e))
2550                return 1;
2551        return has_loose_object(sha1);
2552}
2553
2554static void check_tree(const void *buf, size_t size)
2555{
2556        struct tree_desc desc;
2557        struct name_entry entry;
2558
2559        init_tree_desc(&desc, buf, size);
2560        while (tree_entry(&desc, &entry))
2561                /* do nothing
2562                 * tree_entry() will die() on malformed entries */
2563                ;
2564}
2565
2566static void check_commit(const void *buf, size_t size)
2567{
2568        struct commit c;
2569        memset(&c, 0, sizeof(c));
2570        if (parse_commit_buffer(&c, buf, size))
2571                die("corrupt commit");
2572}
2573
2574static void check_tag(const void *buf, size_t size)
2575{
2576        struct tag t;
2577        memset(&t, 0, sizeof(t));
2578        if (parse_tag_buffer(&t, buf, size))
2579                die("corrupt tag");
2580}
2581
2582static int index_mem(unsigned char *sha1, void *buf, size_t size,
2583                     int write_object, enum object_type type,
2584                     const char *path, int format_check)
2585{
2586        int ret, re_allocated = 0;
2587
2588        if (!type)
2589                type = OBJ_BLOB;
2590
2591        /*
2592         * Convert blobs to git internal format
2593         */
2594        if ((type == OBJ_BLOB) && path) {
2595                struct strbuf nbuf = STRBUF_INIT;
2596                if (convert_to_git(path, buf, size, &nbuf,
2597                                   write_object ? safe_crlf : 0)) {
2598                        buf = strbuf_detach(&nbuf, &size);
2599                        re_allocated = 1;
2600                }
2601        }
2602        if (format_check) {
2603                if (type == OBJ_TREE)
2604                        check_tree(buf, size);
2605                if (type == OBJ_COMMIT)
2606                        check_commit(buf, size);
2607                if (type == OBJ_TAG)
2608                        check_tag(buf, size);
2609        }
2610
2611        if (write_object)
2612                ret = write_sha1_file(buf, size, typename(type), sha1);
2613        else
2614                ret = hash_sha1_file(buf, size, typename(type), sha1);
2615        if (re_allocated)
2616                free(buf);
2617        return ret;
2618}
2619
2620#define SMALL_FILE_SIZE (32*1024)
2621
2622int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2623             enum object_type type, const char *path, int format_check)
2624{
2625        int ret;
2626        size_t size = xsize_t(st->st_size);
2627
2628        if (!S_ISREG(st->st_mode)) {
2629                struct strbuf sbuf = STRBUF_INIT;
2630                if (strbuf_read(&sbuf, fd, 4096) >= 0)
2631                        ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2632                                        type, path, format_check);
2633                else
2634                        ret = -1;
2635                strbuf_release(&sbuf);
2636        } else if (!size) {
2637                ret = index_mem(sha1, NULL, size, write_object, type, path,
2638                                format_check);
2639        } else if (size <= SMALL_FILE_SIZE) {
2640                char *buf = xmalloc(size);
2641                if (size == read_in_full(fd, buf, size))
2642                        ret = index_mem(sha1, buf, size, write_object, type,
2643                                        path, format_check);
2644                else
2645                        ret = error("short read %s", strerror(errno));
2646                free(buf);
2647        } else {
2648                void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2649                ret = index_mem(sha1, buf, size, write_object, type, path,
2650                                format_check);
2651                munmap(buf, size);
2652        }
2653        close(fd);
2654        return ret;
2655}
2656
2657int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2658{
2659        int fd;
2660        struct strbuf sb = STRBUF_INIT;
2661
2662        switch (st->st_mode & S_IFMT) {
2663        case S_IFREG:
2664                fd = open(path, O_RDONLY);
2665                if (fd < 0)
2666                        return error("open(\"%s\"): %s", path,
2667                                     strerror(errno));
2668                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path, 0) < 0)
2669                        return error("%s: failed to insert into database",
2670                                     path);
2671                break;
2672        case S_IFLNK:
2673                if (strbuf_readlink(&sb, path, st->st_size)) {
2674                        char *errstr = strerror(errno);
2675                        return error("readlink(\"%s\"): %s", path,
2676                                     errstr);
2677                }
2678                if (!write_object)
2679                        hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2680                else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2681                        return error("%s: failed to insert into database",
2682                                     path);
2683                strbuf_release(&sb);
2684                break;
2685        case S_IFDIR:
2686                return resolve_gitlink_ref(path, "HEAD", sha1);
2687        default:
2688                return error("%s: unsupported file type", path);
2689        }
2690        return 0;
2691}
2692
2693int read_pack_header(int fd, struct pack_header *header)
2694{
2695        if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2696                /* "eof before pack header was fully read" */
2697                return PH_ERROR_EOF;
2698
2699        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2700                /* "protocol error (pack signature mismatch detected)" */
2701                return PH_ERROR_PACK_SIGNATURE;
2702        if (!pack_version_ok(header->hdr_version))
2703                /* "protocol error (pack version unsupported)" */
2704                return PH_ERROR_PROTOCOL;
2705        return 0;
2706}
2707
2708void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2709{
2710        enum object_type type = sha1_object_info(sha1, NULL);
2711        if (type < 0)
2712                die("%s is not a valid object", sha1_to_hex(sha1));
2713        if (type != expect)
2714                die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2715                    typename(expect));
2716}