sha1_file.con commit Teach tree_entry_interesting() that the tree entries are sorted. (7d2f667)
   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
  17#ifndef O_NOATIME
  18#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
  19#define O_NOATIME 01000000
  20#else
  21#define O_NOATIME 0
  22#endif
  23#endif
  24
  25#ifdef NO_C99_FORMAT
  26#define SZ_FMT "lu"
  27#else
  28#define SZ_FMT "zu"
  29#endif
  30
  31const unsigned char null_sha1[20];
  32
  33static unsigned int sha1_file_open_flag = O_NOATIME;
  34
  35signed char hexval_table[256] = {
  36         -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
  37         -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
  38         -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
  39         -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
  40         -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
  41         -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
  42          0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
  43          8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
  44         -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
  45         -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
  46         -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
  47         -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
  48         -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
  49         -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
  50         -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
  51         -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
  52         -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
  53         -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
  54         -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
  55         -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
  56         -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
  57         -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
  58         -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
  59         -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
  60         -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
  61         -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
  62         -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
  63         -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
  64         -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
  65         -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
  66         -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
  67         -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
  68};
  69
  70int get_sha1_hex(const char *hex, unsigned char *sha1)
  71{
  72        int i;
  73        for (i = 0; i < 20; i++) {
  74                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
  75                if (val & ~0xff)
  76                        return -1;
  77                *sha1++ = val;
  78                hex += 2;
  79        }
  80        return 0;
  81}
  82
  83int safe_create_leading_directories(char *path)
  84{
  85        char *pos = path;
  86        struct stat st;
  87
  88        if (*pos == '/')
  89                pos++;
  90
  91        while (pos) {
  92                pos = strchr(pos, '/');
  93                if (!pos)
  94                        break;
  95                *pos = 0;
  96                if (!stat(path, &st)) {
  97                        /* path exists */
  98                        if (!S_ISDIR(st.st_mode)) {
  99                                *pos = '/';
 100                                return -3;
 101                        }
 102                }
 103                else if (mkdir(path, 0777)) {
 104                        *pos = '/';
 105                        return -1;
 106                }
 107                else if (adjust_shared_perm(path)) {
 108                        *pos = '/';
 109                        return -2;
 110                }
 111                *pos++ = '/';
 112        }
 113        return 0;
 114}
 115
 116char * sha1_to_hex(const unsigned char *sha1)
 117{
 118        static int bufno;
 119        static char hexbuffer[4][50];
 120        static const char hex[] = "0123456789abcdef";
 121        char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
 122        int i;
 123
 124        for (i = 0; i < 20; i++) {
 125                unsigned int val = *sha1++;
 126                *buf++ = hex[val >> 4];
 127                *buf++ = hex[val & 0xf];
 128        }
 129        *buf = '\0';
 130
 131        return buffer;
 132}
 133
 134static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
 135{
 136        int i;
 137        for (i = 0; i < 20; i++) {
 138                static char hex[] = "0123456789abcdef";
 139                unsigned int val = sha1[i];
 140                char *pos = pathbuf + i*2 + (i > 0);
 141                *pos++ = hex[val >> 4];
 142                *pos = hex[val & 0xf];
 143        }
 144}
 145
 146/*
 147 * NOTE! This returns a statically allocated buffer, so you have to be
 148 * careful about using it. Do a "xstrdup()" if you need to save the
 149 * filename.
 150 *
 151 * Also note that this returns the location for creating.  Reading
 152 * SHA1 file can happen from any alternate directory listed in the
 153 * DB_ENVIRONMENT environment variable if it is not found in
 154 * the primary object database.
 155 */
 156char *sha1_file_name(const unsigned char *sha1)
 157{
 158        static char *name, *base;
 159
 160        if (!base) {
 161                const char *sha1_file_directory = get_object_directory();
 162                int len = strlen(sha1_file_directory);
 163                base = xmalloc(len + 60);
 164                memcpy(base, sha1_file_directory, len);
 165                memset(base+len, 0, 60);
 166                base[len] = '/';
 167                base[len+3] = '/';
 168                name = base + len + 1;
 169        }
 170        fill_sha1_path(name, sha1);
 171        return base;
 172}
 173
 174char *sha1_pack_name(const unsigned char *sha1)
 175{
 176        static const char hex[] = "0123456789abcdef";
 177        static char *name, *base, *buf;
 178        int i;
 179
 180        if (!base) {
 181                const char *sha1_file_directory = get_object_directory();
 182                int len = strlen(sha1_file_directory);
 183                base = xmalloc(len + 60);
 184                sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
 185                name = base + len + 11;
 186        }
 187
 188        buf = name;
 189
 190        for (i = 0; i < 20; i++) {
 191                unsigned int val = *sha1++;
 192                *buf++ = hex[val >> 4];
 193                *buf++ = hex[val & 0xf];
 194        }
 195        
 196        return base;
 197}
 198
 199char *sha1_pack_index_name(const unsigned char *sha1)
 200{
 201        static const char hex[] = "0123456789abcdef";
 202        static char *name, *base, *buf;
 203        int i;
 204
 205        if (!base) {
 206                const char *sha1_file_directory = get_object_directory();
 207                int len = strlen(sha1_file_directory);
 208                base = xmalloc(len + 60);
 209                sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
 210                name = base + len + 11;
 211        }
 212
 213        buf = name;
 214
 215        for (i = 0; i < 20; i++) {
 216                unsigned int val = *sha1++;
 217                *buf++ = hex[val >> 4];
 218                *buf++ = hex[val & 0xf];
 219        }
 220        
 221        return base;
 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);
 228
 229/*
 230 * Prepare alternate object database registry.
 231 *
 232 * The variable alt_odb_list points at the list of struct
 233 * alternate_object_database.  The elements on this list come from
 234 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 235 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 236 * whose contents is similar to that environment variable but can be
 237 * LF separated.  Its base points at a statically allocated buffer that
 238 * contains "/the/directory/corresponding/to/.git/objects/...", while
 239 * its name points just after the slash at the end of ".git/objects/"
 240 * in the example above, and has enough space to hold 40-byte hex
 241 * SHA1, an extra slash for the first level indirection, and the
 242 * terminating NUL.
 243 */
 244static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
 245{
 246        struct stat st;
 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 (*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 (*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 (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
 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 ((*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        char path[PATH_MAX];
 355        int fd;
 356
 357        sprintf(path, "%s/info/alternates", relative_base);
 358        fd = open(path, O_RDONLY);
 359        if (fd < 0)
 360                return;
 361        if (fstat(fd, &st) || (st.st_size == 0)) {
 362                close(fd);
 363                return;
 364        }
 365        mapsz = xsize_t(st.st_size);
 366        map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
 367        close(fd);
 368
 369        link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
 370
 371        munmap(map, mapsz);
 372}
 373
 374void prepare_alt_odb(void)
 375{
 376        const char *alt;
 377
 378        alt = getenv(ALTERNATE_DB_ENVIRONMENT);
 379        if (!alt) alt = "";
 380
 381        if (alt_odb_tail)
 382                return;
 383        alt_odb_tail = &alt_odb_list;
 384        link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
 385
 386        read_info_alternates(get_object_directory(), 0);
 387}
 388
 389static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
 390{
 391        char *name = sha1_file_name(sha1);
 392        struct alternate_object_database *alt;
 393
 394        if (!stat(name, st))
 395                return name;
 396        prepare_alt_odb();
 397        for (alt = alt_odb_list; alt; alt = alt->next) {
 398                name = alt->name;
 399                fill_sha1_path(name, sha1);
 400                if (!stat(alt->base, st))
 401                        return alt->base;
 402        }
 403        return NULL;
 404}
 405
 406static unsigned int pack_used_ctr;
 407static unsigned int pack_mmap_calls;
 408static unsigned int peak_pack_open_windows;
 409static unsigned int pack_open_windows;
 410static size_t peak_pack_mapped;
 411static size_t pack_mapped;
 412struct packed_git *packed_git;
 413
 414void pack_report()
 415{
 416        fprintf(stderr,
 417                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
 418                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
 419                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
 420                (size_t) getpagesize(),
 421                packed_git_window_size,
 422                packed_git_limit);
 423        fprintf(stderr,
 424                "pack_report: pack_used_ctr            = %10u\n"
 425                "pack_report: pack_mmap_calls          = %10u\n"
 426                "pack_report: pack_open_windows        = %10u / %10u\n"
 427                "pack_report: pack_mapped              = "
 428                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
 429                pack_used_ctr,
 430                pack_mmap_calls,
 431                pack_open_windows, peak_pack_open_windows,
 432                pack_mapped, peak_pack_mapped);
 433}
 434
 435static int check_packed_git_idx(const char *path,  struct packed_git *p)
 436{
 437        void *idx_map;
 438        struct pack_idx_header *hdr;
 439        size_t idx_size;
 440        uint32_t nr, i, *index;
 441        int fd = open(path, O_RDONLY);
 442        struct stat st;
 443
 444        if (fd < 0)
 445                return -1;
 446        if (fstat(fd, &st)) {
 447                close(fd);
 448                return -1;
 449        }
 450        idx_size = xsize_t(st.st_size);
 451        if (idx_size < 4 * 256 + 20 + 20) {
 452                close(fd);
 453                return error("index file %s is too small", path);
 454        }
 455        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 456        close(fd);
 457
 458        /* a future index format would start with this, as older git
 459         * binaries would fail the non-monotonic index check below.
 460         * give a nicer warning to the user if we can.
 461         */
 462        hdr = idx_map;
 463        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 464                munmap(idx_map, idx_size);
 465                return error("index file %s is a newer version"
 466                        " and is not supported by this binary"
 467                        " (try upgrading GIT to a newer version)",
 468                        path);
 469        }
 470
 471        nr = 0;
 472        index = idx_map;
 473        for (i = 0; i < 256; i++) {
 474                uint32_t n = ntohl(index[i]);
 475                if (n < nr) {
 476                        munmap(idx_map, idx_size);
 477                        return error("non-monotonic index %s", path);
 478                }
 479                nr = n;
 480        }
 481
 482        /*
 483         * Total size:
 484         *  - 256 index entries 4 bytes each
 485         *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 486         *  - 20-byte SHA1 of the packfile
 487         *  - 20-byte SHA1 file checksum
 488         */
 489        if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 490                munmap(idx_map, idx_size);
 491                return error("wrong index file size in %s", path);
 492        }
 493
 494        p->index_version = 1;
 495        p->index_data = idx_map;
 496        p->index_size = idx_size;
 497        return 0;
 498}
 499
 500static void scan_windows(struct packed_git *p,
 501        struct packed_git **lru_p,
 502        struct pack_window **lru_w,
 503        struct pack_window **lru_l)
 504{
 505        struct pack_window *w, *w_l;
 506
 507        for (w_l = NULL, w = p->windows; w; w = w->next) {
 508                if (!w->inuse_cnt) {
 509                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 510                                *lru_p = p;
 511                                *lru_w = w;
 512                                *lru_l = w_l;
 513                        }
 514                }
 515                w_l = w;
 516        }
 517}
 518
 519static int unuse_one_window(struct packed_git *current)
 520{
 521        struct packed_git *p, *lru_p = NULL;
 522        struct pack_window *lru_w = NULL, *lru_l = NULL;
 523
 524        if (current)
 525                scan_windows(current, &lru_p, &lru_w, &lru_l);
 526        for (p = packed_git; p; p = p->next)
 527                scan_windows(p, &lru_p, &lru_w, &lru_l);
 528        if (lru_p) {
 529                munmap(lru_w->base, lru_w->len);
 530                pack_mapped -= lru_w->len;
 531                if (lru_l)
 532                        lru_l->next = lru_w->next;
 533                else {
 534                        lru_p->windows = lru_w->next;
 535                        if (!lru_p->windows && lru_p != current) {
 536                                close(lru_p->pack_fd);
 537                                lru_p->pack_fd = -1;
 538                        }
 539                }
 540                free(lru_w);
 541                pack_open_windows--;
 542                return 1;
 543        }
 544        return 0;
 545}
 546
 547void release_pack_memory(size_t need)
 548{
 549        size_t cur = pack_mapped;
 550        while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
 551                ; /* nothing */
 552}
 553
 554void unuse_pack(struct pack_window **w_cursor)
 555{
 556        struct pack_window *w = *w_cursor;
 557        if (w) {
 558                w->inuse_cnt--;
 559                *w_cursor = NULL;
 560        }
 561}
 562
 563/*
 564 * Do not call this directly as this leaks p->pack_fd on error return;
 565 * call open_packed_git() instead.
 566 */
 567static int open_packed_git_1(struct packed_git *p)
 568{
 569        struct stat st;
 570        struct pack_header hdr;
 571        unsigned char sha1[20];
 572        unsigned char *idx_sha1;
 573        long fd_flag;
 574
 575        p->pack_fd = open(p->pack_name, O_RDONLY);
 576        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 577                return -1;
 578
 579        /* If we created the struct before we had the pack we lack size. */
 580        if (!p->pack_size) {
 581                if (!S_ISREG(st.st_mode))
 582                        return error("packfile %s not a regular file", p->pack_name);
 583                p->pack_size = st.st_size;
 584        } else if (p->pack_size != st.st_size)
 585                return error("packfile %s size changed", p->pack_name);
 586
 587        /* We leave these file descriptors open with sliding mmap;
 588         * there is no point keeping them open across exec(), though.
 589         */
 590        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 591        if (fd_flag < 0)
 592                return error("cannot determine file descriptor flags");
 593        fd_flag |= FD_CLOEXEC;
 594        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 595                return error("cannot set FD_CLOEXEC");
 596
 597        /* Verify we recognize this pack file format. */
 598        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 599                return error("file %s is far too short to be a packfile", p->pack_name);
 600        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 601                return error("file %s is not a GIT packfile", p->pack_name);
 602        if (!pack_version_ok(hdr.hdr_version))
 603                return error("packfile %s is version %u and not supported"
 604                        " (try upgrading GIT to a newer version)",
 605                        p->pack_name, ntohl(hdr.hdr_version));
 606
 607        /* Verify the pack matches its index. */
 608        if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
 609                return error("packfile %s claims to have %u objects"
 610                        " while index size indicates %u objects",
 611                        p->pack_name, ntohl(hdr.hdr_entries),
 612                        num_packed_objects(p));
 613        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 614                return error("end of packfile %s is unavailable", p->pack_name);
 615        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 616                return error("packfile %s signature is unavailable", p->pack_name);
 617        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 618        if (hashcmp(sha1, idx_sha1))
 619                return error("packfile %s does not match index", p->pack_name);
 620        return 0;
 621}
 622
 623static int open_packed_git(struct packed_git *p)
 624{
 625        if (!open_packed_git_1(p))
 626                return 0;
 627        if (p->pack_fd != -1) {
 628                close(p->pack_fd);
 629                p->pack_fd = -1;
 630        }
 631        return -1;
 632}
 633
 634static int in_window(struct pack_window *win, off_t offset)
 635{
 636        /* We must promise at least 20 bytes (one hash) after the
 637         * offset is available from this window, otherwise the offset
 638         * is not actually in this window and a different window (which
 639         * has that one hash excess) must be used.  This is to support
 640         * the object header and delta base parsing routines below.
 641         */
 642        off_t win_off = win->offset;
 643        return win_off <= offset
 644                && (offset + 20) <= (win_off + win->len);
 645}
 646
 647unsigned char* use_pack(struct packed_git *p,
 648                struct pack_window **w_cursor,
 649                off_t offset,
 650                unsigned int *left)
 651{
 652        struct pack_window *win = *w_cursor;
 653
 654        if (p->pack_fd == -1 && open_packed_git(p))
 655                die("packfile %s cannot be accessed", p->pack_name);
 656
 657        /* Since packfiles end in a hash of their content and its
 658         * pointless to ask for an offset into the middle of that
 659         * hash, and the in_window function above wouldn't match
 660         * don't allow an offset too close to the end of the file.
 661         */
 662        if (offset > (p->pack_size - 20))
 663                die("offset beyond end of packfile (truncated pack?)");
 664
 665        if (!win || !in_window(win, offset)) {
 666                if (win)
 667                        win->inuse_cnt--;
 668                for (win = p->windows; win; win = win->next) {
 669                        if (in_window(win, offset))
 670                                break;
 671                }
 672                if (!win) {
 673                        size_t window_align = packed_git_window_size / 2;
 674                        off_t len;
 675                        win = xcalloc(1, sizeof(*win));
 676                        win->offset = (offset / window_align) * window_align;
 677                        len = p->pack_size - win->offset;
 678                        if (len > packed_git_window_size)
 679                                len = packed_git_window_size;
 680                        win->len = (size_t)len;
 681                        pack_mapped += win->len;
 682                        while (packed_git_limit < pack_mapped
 683                                && unuse_one_window(p))
 684                                ; /* nothing */
 685                        win->base = xmmap(NULL, win->len,
 686                                PROT_READ, MAP_PRIVATE,
 687                                p->pack_fd, win->offset);
 688                        if (win->base == MAP_FAILED)
 689                                die("packfile %s cannot be mapped: %s",
 690                                        p->pack_name,
 691                                        strerror(errno));
 692                        pack_mmap_calls++;
 693                        pack_open_windows++;
 694                        if (pack_mapped > peak_pack_mapped)
 695                                peak_pack_mapped = pack_mapped;
 696                        if (pack_open_windows > peak_pack_open_windows)
 697                                peak_pack_open_windows = pack_open_windows;
 698                        win->next = p->windows;
 699                        p->windows = win;
 700                }
 701        }
 702        if (win != *w_cursor) {
 703                win->last_used = pack_used_ctr++;
 704                win->inuse_cnt++;
 705                *w_cursor = win;
 706        }
 707        offset -= win->offset;
 708        if (left)
 709                *left = win->len - xsize_t(offset);
 710        return win->base + offset;
 711}
 712
 713struct packed_git *add_packed_git(const char *path, int path_len, int local)
 714{
 715        struct stat st;
 716        struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
 717
 718        /*
 719         * Make sure a corresponding .pack file exists and that
 720         * the index looks sane.
 721         */
 722        path_len -= strlen(".idx");
 723        if (path_len < 1)
 724                return NULL;
 725        memcpy(p->pack_name, path, path_len);
 726        strcpy(p->pack_name + path_len, ".pack");
 727        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
 728            check_packed_git_idx(path, p)) {
 729                free(p);
 730                return NULL;
 731        }
 732
 733        /* ok, it looks sane as far as we can check without
 734         * actually mapping the pack file.
 735         */
 736        p->pack_size = st.st_size;
 737        p->next = NULL;
 738        p->windows = NULL;
 739        p->pack_fd = -1;
 740        p->pack_local = local;
 741        p->mtime = st.st_mtime;
 742        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 743                hashclr(p->sha1);
 744        return p;
 745}
 746
 747struct packed_git *parse_pack_index(unsigned char *sha1)
 748{
 749        char *path = sha1_pack_index_name(sha1);
 750        return parse_pack_index_file(sha1, path);
 751}
 752
 753struct packed_git *parse_pack_index_file(const unsigned char *sha1,
 754                                         const char *idx_path)
 755{
 756        const char *path = sha1_pack_name(sha1);
 757        struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
 758
 759        if (check_packed_git_idx(idx_path, p)) {
 760                free(p);
 761                return NULL;
 762        }
 763
 764        strcpy(p->pack_name, path);
 765        p->pack_size = 0;
 766        p->next = NULL;
 767        p->windows = NULL;
 768        p->pack_fd = -1;
 769        hashcpy(p->sha1, sha1);
 770        return p;
 771}
 772
 773void install_packed_git(struct packed_git *pack)
 774{
 775        pack->next = packed_git;
 776        packed_git = pack;
 777}
 778
 779static void prepare_packed_git_one(char *objdir, int local)
 780{
 781        char path[PATH_MAX];
 782        int len;
 783        DIR *dir;
 784        struct dirent *de;
 785
 786        sprintf(path, "%s/pack", objdir);
 787        len = strlen(path);
 788        dir = opendir(path);
 789        if (!dir) {
 790                if (errno != ENOENT)
 791                        error("unable to open object pack directory: %s: %s",
 792                              path, strerror(errno));
 793                return;
 794        }
 795        path[len++] = '/';
 796        while ((de = readdir(dir)) != NULL) {
 797                int namelen = strlen(de->d_name);
 798                struct packed_git *p;
 799
 800                if (!has_extension(de->d_name, ".idx"))
 801                        continue;
 802
 803                /* Don't reopen a pack we already have. */
 804                strcpy(path + len, de->d_name);
 805                for (p = packed_git; p; p = p->next) {
 806                        if (!memcmp(path, p->pack_name, len + namelen - 4))
 807                                break;
 808                }
 809                if (p)
 810                        continue;
 811                /* See if it really is a valid .idx file with corresponding
 812                 * .pack file that we can map.
 813                 */
 814                p = add_packed_git(path, len + namelen, local);
 815                if (!p)
 816                        continue;
 817                install_packed_git(p);
 818        }
 819        closedir(dir);
 820}
 821
 822static int sort_pack(const void *a_, const void *b_)
 823{
 824        struct packed_git *a = *((struct packed_git **)a_);
 825        struct packed_git *b = *((struct packed_git **)b_);
 826        int st;
 827
 828        /*
 829         * Local packs tend to contain objects specific to our
 830         * variant of the project than remote ones.  In addition,
 831         * remote ones could be on a network mounted filesystem.
 832         * Favor local ones for these reasons.
 833         */
 834        st = a->pack_local - b->pack_local;
 835        if (st)
 836                return -st;
 837
 838        /*
 839         * Younger packs tend to contain more recent objects,
 840         * and more recent objects tend to get accessed more
 841         * often.
 842         */
 843        if (a->mtime < b->mtime)
 844                return 1;
 845        else if (a->mtime == b->mtime)
 846                return 0;
 847        return -1;
 848}
 849
 850static void rearrange_packed_git(void)
 851{
 852        struct packed_git **ary, *p;
 853        int i, n;
 854
 855        for (n = 0, p = packed_git; p; p = p->next)
 856                n++;
 857        if (n < 2)
 858                return;
 859
 860        /* prepare an array of packed_git for easier sorting */
 861        ary = xcalloc(n, sizeof(struct packed_git *));
 862        for (n = 0, p = packed_git; p; p = p->next)
 863                ary[n++] = p;
 864
 865        qsort(ary, n, sizeof(struct packed_git *), sort_pack);
 866
 867        /* link them back again */
 868        for (i = 0; i < n - 1; i++)
 869                ary[i]->next = ary[i + 1];
 870        ary[n - 1]->next = NULL;
 871        packed_git = ary[0];
 872
 873        free(ary);
 874}
 875
 876static int prepare_packed_git_run_once = 0;
 877void prepare_packed_git(void)
 878{
 879        struct alternate_object_database *alt;
 880
 881        if (prepare_packed_git_run_once)
 882                return;
 883        prepare_packed_git_one(get_object_directory(), 1);
 884        prepare_alt_odb();
 885        for (alt = alt_odb_list; alt; alt = alt->next) {
 886                alt->name[-1] = 0;
 887                prepare_packed_git_one(alt->base, 0);
 888                alt->name[-1] = '/';
 889        }
 890        rearrange_packed_git();
 891        prepare_packed_git_run_once = 1;
 892}
 893
 894void reprepare_packed_git(void)
 895{
 896        prepare_packed_git_run_once = 0;
 897        prepare_packed_git();
 898}
 899
 900int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
 901{
 902        unsigned char real_sha1[20];
 903        hash_sha1_file(map, size, type, real_sha1);
 904        return hashcmp(sha1, real_sha1) ? -1 : 0;
 905}
 906
 907void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
 908{
 909        struct stat st;
 910        void *map;
 911        int fd;
 912        char *filename = find_sha1_file(sha1, &st);
 913
 914        if (!filename) {
 915                return NULL;
 916        }
 917
 918        fd = open(filename, O_RDONLY | sha1_file_open_flag);
 919        if (fd < 0) {
 920                /* See if it works without O_NOATIME */
 921                switch (sha1_file_open_flag) {
 922                default:
 923                        fd = open(filename, O_RDONLY);
 924                        if (fd >= 0)
 925                                break;
 926                /* Fallthrough */
 927                case 0:
 928                        return NULL;
 929                }
 930
 931                /* If it failed once, it will probably fail again.
 932                 * Stop using O_NOATIME
 933                 */
 934                sha1_file_open_flag = 0;
 935        }
 936        *size = xsize_t(st.st_size);
 937        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 938        close(fd);
 939        return map;
 940}
 941
 942int legacy_loose_object(unsigned char *map)
 943{
 944        unsigned int word;
 945
 946        /*
 947         * Is it a zlib-compressed buffer? If so, the first byte
 948         * must be 0x78 (15-bit window size, deflated), and the
 949         * first 16-bit word is evenly divisible by 31
 950         */
 951        word = (map[0] << 8) + map[1];
 952        if (map[0] == 0x78 && !(word % 31))
 953                return 1;
 954        else
 955                return 0;
 956}
 957
 958unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
 959{
 960        unsigned shift;
 961        unsigned char c;
 962        unsigned long size;
 963        unsigned long used = 0;
 964
 965        c = buf[used++];
 966        *type = (c >> 4) & 7;
 967        size = c & 15;
 968        shift = 4;
 969        while (c & 0x80) {
 970                if (len <= used)
 971                        return 0;
 972                if (sizeof(long) * 8 <= shift)
 973                        return 0;
 974                c = buf[used++];
 975                size += (c & 0x7f) << shift;
 976                shift += 7;
 977        }
 978        *sizep = size;
 979        return used;
 980}
 981
 982static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
 983{
 984        unsigned long size, used;
 985        static const char valid_loose_object_type[8] = {
 986                0, /* OBJ_EXT */
 987                1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
 988                0, /* "delta" and others are invalid in a loose object */
 989        };
 990        enum object_type type;
 991
 992        /* Get the data stream */
 993        memset(stream, 0, sizeof(*stream));
 994        stream->next_in = map;
 995        stream->avail_in = mapsize;
 996        stream->next_out = buffer;
 997        stream->avail_out = bufsiz;
 998
 999        if (legacy_loose_object(map)) {
1000                inflateInit(stream);
1001                return inflate(stream, 0);
1002        }
1003
1004        used = unpack_object_header_gently(map, mapsize, &type, &size);
1005        if (!used || !valid_loose_object_type[type])
1006                return -1;
1007        map += used;
1008        mapsize -= used;
1009
1010        /* Set up the stream for the rest.. */
1011        stream->next_in = map;
1012        stream->avail_in = mapsize;
1013        inflateInit(stream);
1014
1015        /* And generate the fake traditional header */
1016        stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1017                                         typename(type), size);
1018        return 0;
1019}
1020
1021static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1022{
1023        int bytes = strlen(buffer) + 1;
1024        unsigned char *buf = xmalloc(1+size);
1025        unsigned long n;
1026        int status = Z_OK;
1027
1028        n = stream->total_out - bytes;
1029        if (n > size)
1030                n = size;
1031        memcpy(buf, (char *) buffer + bytes, n);
1032        bytes = n;
1033        if (bytes <= size) {
1034                /*
1035                 * The above condition must be (bytes <= size), not
1036                 * (bytes < size).  In other words, even though we
1037                 * expect no more output and set avail_out to zer0,
1038                 * the input zlib stream may have bytes that express
1039                 * "this concludes the stream", and we *do* want to
1040                 * eat that input.
1041                 *
1042                 * Otherwise we would not be able to test that we
1043                 * consumed all the input to reach the expected size;
1044                 * we also want to check that zlib tells us that all
1045                 * went well with status == Z_STREAM_END at the end.
1046                 */
1047                stream->next_out = buf + bytes;
1048                stream->avail_out = size - bytes;
1049                while (status == Z_OK)
1050                        status = inflate(stream, Z_FINISH);
1051        }
1052        buf[size] = 0;
1053        if (status == Z_STREAM_END && !stream->avail_in) {
1054                inflateEnd(stream);
1055                return buf;
1056        }
1057
1058        if (status < 0)
1059                error("corrupt loose object '%s'", sha1_to_hex(sha1));
1060        else if (stream->avail_in)
1061                error("garbage at end of loose object '%s'",
1062                      sha1_to_hex(sha1));
1063        free(buf);
1064        return NULL;
1065}
1066
1067/*
1068 * We used to just use "sscanf()", but that's actually way
1069 * too permissive for what we want to check. So do an anal
1070 * object header parse by hand.
1071 */
1072static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1073{
1074        char type[10];
1075        int i;
1076        unsigned long size;
1077
1078        /*
1079         * The type can be at most ten bytes (including the 
1080         * terminating '\0' that we add), and is followed by
1081         * a space.
1082         */
1083        i = 0;
1084        for (;;) {
1085                char c = *hdr++;
1086                if (c == ' ')
1087                        break;
1088                type[i++] = c;
1089                if (i >= sizeof(type))
1090                        return -1;
1091        }
1092        type[i] = 0;
1093
1094        /*
1095         * The length must follow immediately, and be in canonical
1096         * decimal format (ie "010" is not valid).
1097         */
1098        size = *hdr++ - '0';
1099        if (size > 9)
1100                return -1;
1101        if (size) {
1102                for (;;) {
1103                        unsigned long c = *hdr - '0';
1104                        if (c > 9)
1105                                break;
1106                        hdr++;
1107                        size = size * 10 + c;
1108                }
1109        }
1110        *sizep = size;
1111
1112        /*
1113         * The length must be followed by a zero byte
1114         */
1115        return *hdr ? -1 : type_from_string(type);
1116}
1117
1118static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1119{
1120        int ret;
1121        z_stream stream;
1122        char hdr[8192];
1123
1124        ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1125        if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1126                return NULL;
1127
1128        return unpack_sha1_rest(&stream, hdr, *size, sha1);
1129}
1130
1131static off_t get_delta_base(struct packed_git *p,
1132                                    struct pack_window **w_curs,
1133                                    off_t *curpos,
1134                                    enum object_type type,
1135                                    off_t delta_obj_offset)
1136{
1137        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1138        off_t base_offset;
1139
1140        /* use_pack() assured us we have [base_info, base_info + 20)
1141         * as a range that we can look at without walking off the
1142         * end of the mapped window.  Its actually the hash size
1143         * that is assured.  An OFS_DELTA longer than the hash size
1144         * is stupid, as then a REF_DELTA would be smaller to store.
1145         */
1146        if (type == OBJ_OFS_DELTA) {
1147                unsigned used = 0;
1148                unsigned char c = base_info[used++];
1149                base_offset = c & 127;
1150                while (c & 128) {
1151                        base_offset += 1;
1152                        if (!base_offset || base_offset & ~(~0UL >> 7))
1153                                die("offset value overflow for delta base object");
1154                        c = base_info[used++];
1155                        base_offset = (base_offset << 7) + (c & 127);
1156                }
1157                base_offset = delta_obj_offset - base_offset;
1158                if (base_offset >= delta_obj_offset)
1159                        die("delta base offset out of bound");
1160                *curpos += used;
1161        } else if (type == OBJ_REF_DELTA) {
1162                /* The base entry _must_ be in the same pack */
1163                base_offset = find_pack_entry_one(base_info, p);
1164                if (!base_offset)
1165                        die("failed to find delta-pack base object %s",
1166                                sha1_to_hex(base_info));
1167                *curpos += 20;
1168        } else
1169                die("I am totally screwed");
1170        return base_offset;
1171}
1172
1173/* forward declaration for a mutually recursive function */
1174static int packed_object_info(struct packed_git *p, off_t offset,
1175                              unsigned long *sizep);
1176
1177static int packed_delta_info(struct packed_git *p,
1178                             struct pack_window **w_curs,
1179                             off_t curpos,
1180                             enum object_type type,
1181                             off_t obj_offset,
1182                             unsigned long *sizep)
1183{
1184        off_t base_offset;
1185
1186        base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1187        type = packed_object_info(p, base_offset, NULL);
1188
1189        /* We choose to only get the type of the base object and
1190         * ignore potentially corrupt pack file that expects the delta
1191         * based on a base with a wrong size.  This saves tons of
1192         * inflate() calls.
1193         */
1194        if (sizep) {
1195                const unsigned char *data;
1196                unsigned char delta_head[20], *in;
1197                z_stream stream;
1198                int st;
1199
1200                memset(&stream, 0, sizeof(stream));
1201                stream.next_out = delta_head;
1202                stream.avail_out = sizeof(delta_head);
1203
1204                inflateInit(&stream);
1205                do {
1206                        in = use_pack(p, w_curs, curpos, &stream.avail_in);
1207                        stream.next_in = in;
1208                        st = inflate(&stream, Z_FINISH);
1209                        curpos += stream.next_in - in;
1210                } while ((st == Z_OK || st == Z_BUF_ERROR)
1211                        && stream.total_out < sizeof(delta_head));
1212                inflateEnd(&stream);
1213                if ((st != Z_STREAM_END) &&
1214                    stream.total_out != sizeof(delta_head))
1215                        die("delta data unpack-initial failed");
1216
1217                /* Examine the initial part of the delta to figure out
1218                 * the result size.
1219                 */
1220                data = delta_head;
1221
1222                /* ignore base size */
1223                get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1224
1225                /* Read the result size */
1226                *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1227        }
1228
1229        return type;
1230}
1231
1232static int unpack_object_header(struct packed_git *p,
1233                                struct pack_window **w_curs,
1234                                off_t *curpos,
1235                                unsigned long *sizep)
1236{
1237        unsigned char *base;
1238        unsigned int left;
1239        unsigned long used;
1240        enum object_type type;
1241
1242        /* use_pack() assures us we have [base, base + 20) available
1243         * as a range that we can look at at.  (Its actually the hash
1244         * size that is assured.)  With our object header encoding
1245         * the maximum deflated object size is 2^137, which is just
1246         * insane, so we know won't exceed what we have been given.
1247         */
1248        base = use_pack(p, w_curs, *curpos, &left);
1249        used = unpack_object_header_gently(base, left, &type, sizep);
1250        if (!used)
1251                die("object offset outside of pack file");
1252        *curpos += used;
1253
1254        return type;
1255}
1256
1257const char *packed_object_info_detail(struct packed_git *p,
1258                                      off_t obj_offset,
1259                                      unsigned long *size,
1260                                      unsigned long *store_size,
1261                                      unsigned int *delta_chain_length,
1262                                      unsigned char *base_sha1)
1263{
1264        struct pack_window *w_curs = NULL;
1265        off_t curpos;
1266        unsigned long dummy;
1267        unsigned char *next_sha1;
1268        enum object_type type;
1269
1270        *delta_chain_length = 0;
1271        curpos = obj_offset;
1272        type = unpack_object_header(p, &w_curs, &curpos, size);
1273
1274        for (;;) {
1275                switch (type) {
1276                default:
1277                        die("pack %s contains unknown object type %d",
1278                            p->pack_name, type);
1279                case OBJ_COMMIT:
1280                case OBJ_TREE:
1281                case OBJ_BLOB:
1282                case OBJ_TAG:
1283                        *store_size = 0; /* notyet */
1284                        unuse_pack(&w_curs);
1285                        return typename(type);
1286                case OBJ_OFS_DELTA:
1287                        obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1288                        if (*delta_chain_length == 0) {
1289                                /* TODO: find base_sha1 as pointed by curpos */
1290                                hashclr(base_sha1);
1291                        }
1292                        break;
1293                case OBJ_REF_DELTA:
1294                        next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1295                        if (*delta_chain_length == 0)
1296                                hashcpy(base_sha1, next_sha1);
1297                        obj_offset = find_pack_entry_one(next_sha1, p);
1298                        break;
1299                }
1300                (*delta_chain_length)++;
1301                curpos = obj_offset;
1302                type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1303        }
1304}
1305
1306static int packed_object_info(struct packed_git *p, off_t obj_offset,
1307                              unsigned long *sizep)
1308{
1309        struct pack_window *w_curs = NULL;
1310        unsigned long size;
1311        off_t curpos = obj_offset;
1312        enum object_type type;
1313
1314        type = unpack_object_header(p, &w_curs, &curpos, &size);
1315
1316        switch (type) {
1317        case OBJ_OFS_DELTA:
1318        case OBJ_REF_DELTA:
1319                type = packed_delta_info(p, &w_curs, curpos,
1320                                         type, obj_offset, sizep);
1321                break;
1322        case OBJ_COMMIT:
1323        case OBJ_TREE:
1324        case OBJ_BLOB:
1325        case OBJ_TAG:
1326                if (sizep)
1327                        *sizep = size;
1328                break;
1329        default:
1330                die("pack %s contains unknown object type %d",
1331                    p->pack_name, type);
1332        }
1333        unuse_pack(&w_curs);
1334        return type;
1335}
1336
1337static void *unpack_compressed_entry(struct packed_git *p,
1338                                    struct pack_window **w_curs,
1339                                    off_t curpos,
1340                                    unsigned long size)
1341{
1342        int st;
1343        z_stream stream;
1344        unsigned char *buffer, *in;
1345
1346        buffer = xmalloc(size + 1);
1347        buffer[size] = 0;
1348        memset(&stream, 0, sizeof(stream));
1349        stream.next_out = buffer;
1350        stream.avail_out = size;
1351
1352        inflateInit(&stream);
1353        do {
1354                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1355                stream.next_in = in;
1356                st = inflate(&stream, Z_FINISH);
1357                curpos += stream.next_in - in;
1358        } while (st == Z_OK || st == Z_BUF_ERROR);
1359        inflateEnd(&stream);
1360        if ((st != Z_STREAM_END) || stream.total_out != size) {
1361                free(buffer);
1362                return NULL;
1363        }
1364
1365        return buffer;
1366}
1367
1368#define MAX_DELTA_CACHE (256)
1369
1370static size_t delta_base_cached;
1371
1372static struct delta_base_cache_lru_list {
1373        struct delta_base_cache_lru_list *prev;
1374        struct delta_base_cache_lru_list *next;
1375} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1376
1377static struct delta_base_cache_entry {
1378        struct delta_base_cache_lru_list lru;
1379        void *data;
1380        struct packed_git *p;
1381        off_t base_offset;
1382        unsigned long size;
1383        enum object_type type;
1384} delta_base_cache[MAX_DELTA_CACHE];
1385
1386static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1387{
1388        unsigned long hash;
1389
1390        hash = (unsigned long)p + (unsigned long)base_offset;
1391        hash += (hash >> 8) + (hash >> 16);
1392        return hash % MAX_DELTA_CACHE;
1393}
1394
1395static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1396        unsigned long *base_size, enum object_type *type, int keep_cache)
1397{
1398        void *ret;
1399        unsigned long hash = pack_entry_hash(p, base_offset);
1400        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1401
1402        ret = ent->data;
1403        if (ret && ent->p == p && ent->base_offset == base_offset)
1404                goto found_cache_entry;
1405        return unpack_entry(p, base_offset, type, base_size);
1406
1407found_cache_entry:
1408        if (!keep_cache) {
1409                ent->data = NULL;
1410                ent->lru.next->prev = ent->lru.prev;
1411                ent->lru.prev->next = ent->lru.next;
1412                delta_base_cached -= ent->size;
1413        }
1414        else {
1415                ret = xmalloc(ent->size + 1);
1416                memcpy(ret, ent->data, ent->size);
1417                ((char *)ret)[ent->size] = 0;
1418        }
1419        *type = ent->type;
1420        *base_size = ent->size;
1421        return ret;
1422}
1423
1424static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1425{
1426        if (ent->data) {
1427                free(ent->data);
1428                ent->data = NULL;
1429                ent->lru.next->prev = ent->lru.prev;
1430                ent->lru.prev->next = ent->lru.next;
1431                delta_base_cached -= ent->size;
1432        }
1433}
1434
1435static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1436        void *base, unsigned long base_size, enum object_type type)
1437{
1438        unsigned long hash = pack_entry_hash(p, base_offset);
1439        struct delta_base_cache_entry *ent = delta_base_cache + hash;
1440        struct delta_base_cache_lru_list *lru;
1441
1442        release_delta_base_cache(ent);
1443        delta_base_cached += base_size;
1444
1445        for (lru = delta_base_cache_lru.next;
1446             delta_base_cached > delta_base_cache_limit
1447             && lru != &delta_base_cache_lru;
1448             lru = lru->next) {
1449                struct delta_base_cache_entry *f = (void *)lru;
1450                if (f->type == OBJ_BLOB)
1451                        release_delta_base_cache(f);
1452        }
1453        for (lru = delta_base_cache_lru.next;
1454             delta_base_cached > delta_base_cache_limit
1455             && lru != &delta_base_cache_lru;
1456             lru = lru->next) {
1457                struct delta_base_cache_entry *f = (void *)lru;
1458                release_delta_base_cache(f);
1459        }
1460
1461        ent->p = p;
1462        ent->base_offset = base_offset;
1463        ent->type = type;
1464        ent->data = base;
1465        ent->size = base_size;
1466        ent->lru.next = &delta_base_cache_lru;
1467        ent->lru.prev = delta_base_cache_lru.prev;
1468        delta_base_cache_lru.prev->next = &ent->lru;
1469        delta_base_cache_lru.prev = &ent->lru;
1470}
1471
1472static void *unpack_delta_entry(struct packed_git *p,
1473                                struct pack_window **w_curs,
1474                                off_t curpos,
1475                                unsigned long delta_size,
1476                                off_t obj_offset,
1477                                enum object_type *type,
1478                                unsigned long *sizep)
1479{
1480        void *delta_data, *result, *base;
1481        unsigned long base_size;
1482        off_t base_offset;
1483
1484        base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1485        base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1486        if (!base)
1487                die("failed to read delta base object"
1488                    " at %"PRIuMAX" from %s",
1489                    (uintmax_t)base_offset, p->pack_name);
1490
1491        delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1492        result = patch_delta(base, base_size,
1493                             delta_data, delta_size,
1494                             sizep);
1495        if (!result)
1496                die("failed to apply delta");
1497        free(delta_data);
1498        add_delta_base_cache(p, base_offset, base, base_size, *type);
1499        return result;
1500}
1501
1502void *unpack_entry(struct packed_git *p, off_t obj_offset,
1503                   enum object_type *type, unsigned long *sizep)
1504{
1505        struct pack_window *w_curs = NULL;
1506        off_t curpos = obj_offset;
1507        void *data;
1508
1509        *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1510        switch (*type) {
1511        case OBJ_OFS_DELTA:
1512        case OBJ_REF_DELTA:
1513                data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1514                                          obj_offset, type, sizep);
1515                break;
1516        case OBJ_COMMIT:
1517        case OBJ_TREE:
1518        case OBJ_BLOB:
1519        case OBJ_TAG:
1520                data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1521                break;
1522        default:
1523                die("unknown object type %i in %s", *type, p->pack_name);
1524        }
1525        unuse_pack(&w_curs);
1526        return data;
1527}
1528
1529uint32_t num_packed_objects(const struct packed_git *p)
1530{
1531        /* See check_packed_git_idx() */
1532        return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
1533}
1534
1535int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
1536                           unsigned char* sha1)
1537{
1538        const unsigned char *index = p->index_data;
1539        index += 4 * 256;
1540        if (num_packed_objects(p) <= n)
1541                return -1;
1542        hashcpy(sha1, index + 24 * n + 4);
1543        return 0;
1544}
1545
1546off_t find_pack_entry_one(const unsigned char *sha1,
1547                                  struct packed_git *p)
1548{
1549        const uint32_t *level1_ofs = p->index_data;
1550        int hi = ntohl(level1_ofs[*sha1]);
1551        int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1552        const unsigned char *index = p->index_data;
1553
1554        index += 4 * 256;
1555
1556        do {
1557                int mi = (lo + hi) / 2;
1558                int cmp = hashcmp(index + 24 * mi + 4, sha1);
1559                if (!cmp)
1560                        return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
1561                if (cmp > 0)
1562                        hi = mi;
1563                else
1564                        lo = mi+1;
1565        } while (lo < hi);
1566        return 0;
1567}
1568
1569static int matches_pack_name(struct packed_git *p, const char *ig)
1570{
1571        const char *last_c, *c;
1572
1573        if (!strcmp(p->pack_name, ig))
1574                return 0;
1575
1576        for (c = p->pack_name, last_c = c; *c;)
1577                if (*c == '/')
1578                        last_c = ++c;
1579                else
1580                        ++c;
1581        if (!strcmp(last_c, ig))
1582                return 0;
1583
1584        return 1;
1585}
1586
1587static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1588{
1589        struct packed_git *p;
1590        off_t offset;
1591
1592        prepare_packed_git();
1593
1594        for (p = packed_git; p; p = p->next) {
1595                if (ignore_packed) {
1596                        const char **ig;
1597                        for (ig = ignore_packed; *ig; ig++)
1598                                if (!matches_pack_name(p, *ig))
1599                                        break;
1600                        if (*ig)
1601                                continue;
1602                }
1603                offset = find_pack_entry_one(sha1, p);
1604                if (offset) {
1605                        /*
1606                         * We are about to tell the caller where they can
1607                         * locate the requested object.  We better make
1608                         * sure the packfile is still here and can be
1609                         * accessed before supplying that answer, as
1610                         * it may have been deleted since the index
1611                         * was loaded!
1612                         */
1613                        if (p->pack_fd == -1 && open_packed_git(p)) {
1614                                error("packfile %s cannot be accessed", p->pack_name);
1615                                continue;
1616                        }
1617                        e->offset = offset;
1618                        e->p = p;
1619                        hashcpy(e->sha1, sha1);
1620                        return 1;
1621                }
1622        }
1623        return 0;
1624}
1625
1626struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1627                                  struct packed_git *packs)
1628{
1629        struct packed_git *p;
1630
1631        for (p = packs; p; p = p->next) {
1632                if (find_pack_entry_one(sha1, p))
1633                        return p;
1634        }
1635        return NULL;
1636
1637}
1638
1639static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
1640{
1641        int status;
1642        unsigned long mapsize, size;
1643        void *map;
1644        z_stream stream;
1645        char hdr[32];
1646
1647        map = map_sha1_file(sha1, &mapsize);
1648        if (!map)
1649                return error("unable to find %s", sha1_to_hex(sha1));
1650        if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1651                status = error("unable to unpack %s header",
1652                               sha1_to_hex(sha1));
1653        else if ((status = parse_sha1_header(hdr, &size)) < 0)
1654                status = error("unable to parse %s header", sha1_to_hex(sha1));
1655        else if (sizep)
1656                *sizep = size;
1657        inflateEnd(&stream);
1658        munmap(map, mapsize);
1659        return status;
1660}
1661
1662int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1663{
1664        struct pack_entry e;
1665
1666        if (!find_pack_entry(sha1, &e, NULL)) {
1667                reprepare_packed_git();
1668                if (!find_pack_entry(sha1, &e, NULL))
1669                        return sha1_loose_object_info(sha1, sizep);
1670        }
1671        return packed_object_info(e.p, e.offset, sizep);
1672}
1673
1674static void *read_packed_sha1(const unsigned char *sha1,
1675                              enum object_type *type, unsigned long *size)
1676{
1677        struct pack_entry e;
1678
1679        if (!find_pack_entry(sha1, &e, NULL))
1680                return NULL;
1681        else
1682                return cache_or_unpack_entry(e.p, e.offset, size, type, 1);
1683}
1684
1685/*
1686 * This is meant to hold a *small* number of objects that you would
1687 * want read_sha1_file() to be able to return, but yet you do not want
1688 * to write them into the object store (e.g. a browse-only
1689 * application).
1690 */
1691static struct cached_object {
1692        unsigned char sha1[20];
1693        enum object_type type;
1694        void *buf;
1695        unsigned long size;
1696} *cached_objects;
1697static int cached_object_nr, cached_object_alloc;
1698
1699static struct cached_object *find_cached_object(const unsigned char *sha1)
1700{
1701        int i;
1702        struct cached_object *co = cached_objects;
1703
1704        for (i = 0; i < cached_object_nr; i++, co++) {
1705                if (!hashcmp(co->sha1, sha1))
1706                        return co;
1707        }
1708        return NULL;
1709}
1710
1711int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1712                      unsigned char *sha1)
1713{
1714        struct cached_object *co;
1715
1716        hash_sha1_file(buf, len, typename(type), sha1);
1717        if (has_sha1_file(sha1) || find_cached_object(sha1))
1718                return 0;
1719        if (cached_object_alloc <= cached_object_nr) {
1720                cached_object_alloc = alloc_nr(cached_object_alloc);
1721                cached_objects = xrealloc(cached_objects,
1722                                          sizeof(*cached_objects) *
1723                                          cached_object_alloc);
1724        }
1725        co = &cached_objects[cached_object_nr++];
1726        co->size = len;
1727        co->type = type;
1728        co->buf = xmalloc(len);
1729        memcpy(co->buf, buf, len);
1730        hashcpy(co->sha1, sha1);
1731        return 0;
1732}
1733
1734void *read_sha1_file(const unsigned char *sha1, enum object_type *type,
1735                     unsigned long *size)
1736{
1737        unsigned long mapsize;
1738        void *map, *buf;
1739        struct cached_object *co;
1740
1741        co = find_cached_object(sha1);
1742        if (co) {
1743                buf = xmalloc(co->size + 1);
1744                memcpy(buf, co->buf, co->size);
1745                ((char*)buf)[co->size] = 0;
1746                *type = co->type;
1747                *size = co->size;
1748                return buf;
1749        }
1750
1751        buf = read_packed_sha1(sha1, type, size);
1752        if (buf)
1753                return buf;
1754        map = map_sha1_file(sha1, &mapsize);
1755        if (map) {
1756                buf = unpack_sha1_file(map, mapsize, type, size, sha1);
1757                munmap(map, mapsize);
1758                return buf;
1759        }
1760        reprepare_packed_git();
1761        return read_packed_sha1(sha1, type, size);
1762}
1763
1764void *read_object_with_reference(const unsigned char *sha1,
1765                                 const char *required_type_name,
1766                                 unsigned long *size,
1767                                 unsigned char *actual_sha1_return)
1768{
1769        enum object_type type, required_type;
1770        void *buffer;
1771        unsigned long isize;
1772        unsigned char actual_sha1[20];
1773
1774        required_type = type_from_string(required_type_name);
1775        hashcpy(actual_sha1, sha1);
1776        while (1) {
1777                int ref_length = -1;
1778                const char *ref_type = NULL;
1779
1780                buffer = read_sha1_file(actual_sha1, &type, &isize);
1781                if (!buffer)
1782                        return NULL;
1783                if (type == required_type) {
1784                        *size = isize;
1785                        if (actual_sha1_return)
1786                                hashcpy(actual_sha1_return, actual_sha1);
1787                        return buffer;
1788                }
1789                /* Handle references */
1790                else if (type == OBJ_COMMIT)
1791                        ref_type = "tree ";
1792                else if (type == OBJ_TAG)
1793                        ref_type = "object ";
1794                else {
1795                        free(buffer);
1796                        return NULL;
1797                }
1798                ref_length = strlen(ref_type);
1799
1800                if (memcmp(buffer, ref_type, ref_length) ||
1801                    get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1802                        free(buffer);
1803                        return NULL;
1804                }
1805                free(buffer);
1806                /* Now we have the ID of the referred-to object in
1807                 * actual_sha1.  Check again. */
1808        }
1809}
1810
1811static void write_sha1_file_prepare(const void *buf, unsigned long len,
1812                                    const char *type, unsigned char *sha1,
1813                                    char *hdr, int *hdrlen)
1814{
1815        SHA_CTX c;
1816
1817        /* Generate the header */
1818        *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
1819
1820        /* Sha1.. */
1821        SHA1_Init(&c);
1822        SHA1_Update(&c, hdr, *hdrlen);
1823        SHA1_Update(&c, buf, len);
1824        SHA1_Final(sha1, &c);
1825}
1826
1827/*
1828 * Link the tempfile to the final place, possibly creating the
1829 * last directory level as you do so.
1830 *
1831 * Returns the errno on failure, 0 on success.
1832 */
1833static int link_temp_to_file(const char *tmpfile, const char *filename)
1834{
1835        int ret;
1836        char *dir;
1837
1838        if (!link(tmpfile, filename))
1839                return 0;
1840
1841        /*
1842         * Try to mkdir the last path component if that failed.
1843         *
1844         * Re-try the "link()" regardless of whether the mkdir
1845         * succeeds, since a race might mean that somebody
1846         * else succeeded.
1847         */
1848        ret = errno;
1849        dir = strrchr(filename, '/');
1850        if (dir) {
1851                *dir = 0;
1852                if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1853                        *dir = '/';
1854                        return -2;
1855                }
1856                *dir = '/';
1857                if (!link(tmpfile, filename))
1858                        return 0;
1859                ret = errno;
1860        }
1861        return ret;
1862}
1863
1864/*
1865 * Move the just written object into its final resting place
1866 */
1867int move_temp_to_file(const char *tmpfile, const char *filename)
1868{
1869        int ret = link_temp_to_file(tmpfile, filename);
1870
1871        /*
1872         * Coda hack - coda doesn't like cross-directory links,
1873         * so we fall back to a rename, which will mean that it
1874         * won't be able to check collisions, but that's not a
1875         * big deal.
1876         *
1877         * The same holds for FAT formatted media.
1878         *
1879         * When this succeeds, we just return 0. We have nothing
1880         * left to unlink.
1881         */
1882        if (ret && ret != EEXIST) {
1883                if (!rename(tmpfile, filename))
1884                        return 0;
1885                ret = errno;
1886        }
1887        unlink(tmpfile);
1888        if (ret) {
1889                if (ret != EEXIST) {
1890                        return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1891                }
1892                /* FIXME!!! Collision check here ? */
1893        }
1894
1895        return 0;
1896}
1897
1898static int write_buffer(int fd, const void *buf, size_t len)
1899{
1900        if (write_in_full(fd, buf, len) < 0)
1901                return error("file write error (%s)", strerror(errno));
1902        return 0;
1903}
1904
1905static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1906{
1907        int hdr_len;
1908        unsigned char c;
1909
1910        c = (type << 4) | (len & 15);
1911        len >>= 4;
1912        hdr_len = 1;
1913        while (len) {
1914                *hdr++ = c | 0x80;
1915                hdr_len++;
1916                c = (len & 0x7f);
1917                len >>= 7;
1918        }
1919        *hdr = c;
1920        return hdr_len;
1921}
1922
1923static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1924{
1925        int obj_type, hdrlen;
1926
1927        if (use_legacy_headers) {
1928                while (deflate(stream, 0) == Z_OK)
1929                        /* nothing */;
1930                return;
1931        }
1932        obj_type = type_from_string(type);
1933        hdrlen = write_binary_header(stream->next_out, obj_type, len);
1934        stream->total_out = hdrlen;
1935        stream->next_out += hdrlen;
1936        stream->avail_out -= hdrlen;
1937}
1938
1939int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1940                   unsigned char *sha1)
1941{
1942        char hdr[32];
1943        int hdrlen;
1944        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1945        return 0;
1946}
1947
1948int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1949{
1950        int size, ret;
1951        unsigned char *compressed;
1952        z_stream stream;
1953        unsigned char sha1[20];
1954        char *filename;
1955        static char tmpfile[PATH_MAX];
1956        char hdr[32];
1957        int fd, hdrlen;
1958
1959        /* Normally if we have it in the pack then we do not bother writing
1960         * it out into .git/objects/??/?{38} file.
1961         */
1962        write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1963        filename = sha1_file_name(sha1);
1964        if (returnsha1)
1965                hashcpy(returnsha1, sha1);
1966        if (has_sha1_file(sha1))
1967                return 0;
1968        fd = open(filename, O_RDONLY);
1969        if (fd >= 0) {
1970                /*
1971                 * FIXME!!! We might do collision checking here, but we'd
1972                 * need to uncompress the old file and check it. Later.
1973                 */
1974                close(fd);
1975                return 0;
1976        }
1977
1978        if (errno != ENOENT) {
1979                return error("sha1 file %s: %s\n", filename, strerror(errno));
1980        }
1981
1982        snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1983
1984        fd = mkstemp(tmpfile);
1985        if (fd < 0) {
1986                if (errno == EPERM)
1987                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1988                else
1989                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1990        }
1991
1992        /* Set it up */
1993        memset(&stream, 0, sizeof(stream));
1994        deflateInit(&stream, zlib_compression_level);
1995        size = 8 + deflateBound(&stream, len+hdrlen);
1996        compressed = xmalloc(size);
1997
1998        /* Compress it */
1999        stream.next_out = compressed;
2000        stream.avail_out = size;
2001
2002        /* First header.. */
2003        stream.next_in = (unsigned char *)hdr;
2004        stream.avail_in = hdrlen;
2005        setup_object_header(&stream, type, len);
2006
2007        /* Then the data itself.. */
2008        stream.next_in = buf;
2009        stream.avail_in = len;
2010        ret = deflate(&stream, Z_FINISH);
2011        if (ret != Z_STREAM_END)
2012                die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2013
2014        ret = deflateEnd(&stream);
2015        if (ret != Z_OK)
2016                die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2017
2018        size = stream.total_out;
2019
2020        if (write_buffer(fd, compressed, size) < 0)
2021                die("unable to write sha1 file");
2022        fchmod(fd, 0444);
2023        close(fd);
2024        free(compressed);
2025
2026        return move_temp_to_file(tmpfile, filename);
2027}
2028
2029/*
2030 * We need to unpack and recompress the object for writing
2031 * it out to a different file.
2032 */
2033static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
2034{
2035        size_t size;
2036        z_stream stream;
2037        unsigned char *unpacked;
2038        unsigned long len;
2039        enum object_type type;
2040        char hdr[32];
2041        int hdrlen;
2042        void *buf;
2043
2044        /* need to unpack and recompress it by itself */
2045        unpacked = read_packed_sha1(sha1, &type, &len);
2046        if (!unpacked)
2047                error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2048
2049        hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2050
2051        /* Set it up */
2052        memset(&stream, 0, sizeof(stream));
2053        deflateInit(&stream, zlib_compression_level);
2054        size = deflateBound(&stream, len + hdrlen);
2055        buf = xmalloc(size);
2056
2057        /* Compress it */
2058        stream.next_out = buf;
2059        stream.avail_out = size;
2060
2061        /* First header.. */
2062        stream.next_in = (void *)hdr;
2063        stream.avail_in = hdrlen;
2064        while (deflate(&stream, 0) == Z_OK)
2065                /* nothing */;
2066
2067        /* Then the data itself.. */
2068        stream.next_in = unpacked;
2069        stream.avail_in = len;
2070        while (deflate(&stream, Z_FINISH) == Z_OK)
2071                /* nothing */;
2072        deflateEnd(&stream);
2073        free(unpacked);
2074
2075        *objsize = stream.total_out;
2076        return buf;
2077}
2078
2079int write_sha1_to_fd(int fd, const unsigned char *sha1)
2080{
2081        int retval;
2082        unsigned long objsize;
2083        void *buf = map_sha1_file(sha1, &objsize);
2084
2085        if (buf) {
2086                retval = write_buffer(fd, buf, objsize);
2087                munmap(buf, objsize);
2088                return retval;
2089        }
2090
2091        buf = repack_object(sha1, &objsize);
2092        retval = write_buffer(fd, buf, objsize);
2093        free(buf);
2094        return retval;
2095}
2096
2097int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
2098                       size_t bufsize, size_t *bufposn)
2099{
2100        char tmpfile[PATH_MAX];
2101        int local;
2102        z_stream stream;
2103        unsigned char real_sha1[20];
2104        unsigned char discard[4096];
2105        int ret;
2106        SHA_CTX c;
2107
2108        snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
2109
2110        local = mkstemp(tmpfile);
2111        if (local < 0) {
2112                if (errno == EPERM)
2113                        return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2114                else
2115                        return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2116        }
2117
2118        memset(&stream, 0, sizeof(stream));
2119
2120        inflateInit(&stream);
2121
2122        SHA1_Init(&c);
2123
2124        do {
2125                ssize_t size;
2126                if (*bufposn) {
2127                        stream.avail_in = *bufposn;
2128                        stream.next_in = (unsigned char *) buffer;
2129                        do {
2130                                stream.next_out = discard;
2131                                stream.avail_out = sizeof(discard);
2132                                ret = inflate(&stream, Z_SYNC_FLUSH);
2133                                SHA1_Update(&c, discard, sizeof(discard) -
2134                                            stream.avail_out);
2135                        } while (stream.avail_in && ret == Z_OK);
2136                        if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
2137                                die("unable to write sha1 file");
2138                        memmove(buffer, buffer + *bufposn - stream.avail_in,
2139                                stream.avail_in);
2140                        *bufposn = stream.avail_in;
2141                        if (ret != Z_OK)
2142                                break;
2143                }
2144                size = xread(fd, buffer + *bufposn, bufsize - *bufposn);
2145                if (size <= 0) {
2146                        close(local);
2147                        unlink(tmpfile);
2148                        if (!size)
2149                                return error("Connection closed?");
2150                        perror("Reading from connection");
2151                        return -1;
2152                }
2153                *bufposn += size;
2154        } while (1);
2155        inflateEnd(&stream);
2156
2157        close(local);
2158        SHA1_Final(real_sha1, &c);
2159        if (ret != Z_STREAM_END) {
2160                unlink(tmpfile);
2161                return error("File %s corrupted", sha1_to_hex(sha1));
2162        }
2163        if (hashcmp(sha1, real_sha1)) {
2164                unlink(tmpfile);
2165                return error("File %s has bad hash", sha1_to_hex(sha1));
2166        }
2167
2168        return move_temp_to_file(tmpfile, sha1_file_name(sha1));
2169}
2170
2171int has_pack_index(const unsigned char *sha1)
2172{
2173        struct stat st;
2174        if (stat(sha1_pack_index_name(sha1), &st))
2175                return 0;
2176        return 1;
2177}
2178
2179int has_pack_file(const unsigned char *sha1)
2180{
2181        struct stat st;
2182        if (stat(sha1_pack_name(sha1), &st))
2183                return 0;
2184        return 1;
2185}
2186
2187int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
2188{
2189        struct pack_entry e;
2190        return find_pack_entry(sha1, &e, ignore_packed);
2191}
2192
2193int has_sha1_file(const unsigned char *sha1)
2194{
2195        struct stat st;
2196        struct pack_entry e;
2197
2198        if (find_pack_entry(sha1, &e, NULL))
2199                return 1;
2200        return find_sha1_file(sha1, &st) ? 1 : 0;
2201}
2202
2203/*
2204 * reads from fd as long as possible into a supplied buffer of size bytes.
2205 * If necessary the buffer's size is increased using realloc()
2206 *
2207 * returns 0 if anything went fine and -1 otherwise
2208 *
2209 * NOTE: both buf and size may change, but even when -1 is returned
2210 * you still have to free() it yourself.
2211 */
2212int read_pipe(int fd, char** return_buf, unsigned long* return_size)
2213{
2214        char* buf = *return_buf;
2215        unsigned long size = *return_size;
2216        int iret;
2217        unsigned long off = 0;
2218
2219        do {
2220                iret = xread(fd, buf + off, size - off);
2221                if (iret > 0) {
2222                        off += iret;
2223                        if (off == size) {
2224                                size *= 2;
2225                                buf = xrealloc(buf, size);
2226                        }
2227                }
2228        } while (iret > 0);
2229
2230        *return_buf = buf;
2231        *return_size = off;
2232
2233        if (iret < 0)
2234                return -1;
2235        return 0;
2236}
2237
2238int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
2239{
2240        unsigned long size = 4096;
2241        char *buf = xmalloc(size);
2242        int ret;
2243
2244        if (read_pipe(fd, &buf, &size)) {
2245                free(buf);
2246                return -1;
2247        }
2248
2249        if (!type)
2250                type = blob_type;
2251        if (write_object)
2252                ret = write_sha1_file(buf, size, type, sha1);
2253        else
2254                ret = hash_sha1_file(buf, size, type, sha1);
2255        free(buf);
2256        return ret;
2257}
2258
2259int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2260             enum object_type type, const char *path)
2261{
2262        size_t size = xsize_t(st->st_size);
2263        void *buf = NULL;
2264        int ret, re_allocated = 0;
2265
2266        if (size)
2267                buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2268        close(fd);
2269
2270        if (!type)
2271                type = OBJ_BLOB;
2272
2273        /*
2274         * Convert blobs to git internal format
2275         */
2276        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
2277                unsigned long nsize = size;
2278                char *nbuf = buf;
2279                if (convert_to_git(path, &nbuf, &nsize)) {
2280                        if (size)
2281                                munmap(buf, size);
2282                        size = nsize;
2283                        buf = nbuf;
2284                        re_allocated = 1;
2285                }
2286        }
2287
2288        if (write_object)
2289                ret = write_sha1_file(buf, size, typename(type), sha1);
2290        else
2291                ret = hash_sha1_file(buf, size, typename(type), sha1);
2292        if (re_allocated) {
2293                free(buf);
2294                return ret;
2295        }
2296        if (size)
2297                munmap(buf, size);
2298        return ret;
2299}
2300
2301int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2302{
2303        int fd;
2304        char *target;
2305        size_t len;
2306
2307        switch (st->st_mode & S_IFMT) {
2308        case S_IFREG:
2309                fd = open(path, O_RDONLY);
2310                if (fd < 0)
2311                        return error("open(\"%s\"): %s", path,
2312                                     strerror(errno));
2313                if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path) < 0)
2314                        return error("%s: failed to insert into database",
2315                                     path);
2316                break;
2317        case S_IFLNK:
2318                len = xsize_t(st->st_size);
2319                target = xmalloc(len + 1);
2320                if (readlink(path, target, len + 1) != st->st_size) {
2321                        char *errstr = strerror(errno);
2322                        free(target);
2323                        return error("readlink(\"%s\"): %s", path,
2324                                     errstr);
2325                }
2326                if (!write_object)
2327                        hash_sha1_file(target, len, blob_type, sha1);
2328                else if (write_sha1_file(target, len, blob_type, sha1))
2329                        return error("%s: failed to insert into database",
2330                                     path);
2331                free(target);
2332                break;
2333        default:
2334                return error("%s: unsupported file type", path);
2335        }
2336        return 0;
2337}
2338
2339int read_pack_header(int fd, struct pack_header *header)
2340{
2341        char *c = (char*)header;
2342        ssize_t remaining = sizeof(struct pack_header);
2343        do {
2344                ssize_t r = xread(fd, c, remaining);
2345                if (r <= 0)
2346                        /* "eof before pack header was fully read" */
2347                        return PH_ERROR_EOF;
2348                remaining -= r;
2349                c += r;
2350        } while (remaining > 0);
2351        if (header->hdr_signature != htonl(PACK_SIGNATURE))
2352                /* "protocol error (pack signature mismatch detected)" */
2353                return PH_ERROR_PACK_SIGNATURE;
2354        if (!pack_version_ok(header->hdr_version))
2355                /* "protocol error (pack version unsupported)" */
2356                return PH_ERROR_PROTOCOL;
2357        return 0;
2358}