packfile.con commit Merge branch 'jt/packmigrate' into next (34a628a)
   1#include "cache.h"
   2#include "mru.h"
   3#include "pack.h"
   4#include "dir.h"
   5#include "mergesort.h"
   6#include "packfile.h"
   7#include "delta.h"
   8#include "list.h"
   9#include "streaming.h"
  10#include "sha1-lookup.h"
  11
  12char *odb_pack_name(struct strbuf *buf,
  13                    const unsigned char *sha1,
  14                    const char *ext)
  15{
  16        strbuf_reset(buf);
  17        strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
  18                    sha1_to_hex(sha1), ext);
  19        return buf->buf;
  20}
  21
  22char *sha1_pack_name(const unsigned char *sha1)
  23{
  24        static struct strbuf buf = STRBUF_INIT;
  25        return odb_pack_name(&buf, sha1, "pack");
  26}
  27
  28char *sha1_pack_index_name(const unsigned char *sha1)
  29{
  30        static struct strbuf buf = STRBUF_INIT;
  31        return odb_pack_name(&buf, sha1, "idx");
  32}
  33
  34static unsigned int pack_used_ctr;
  35static unsigned int pack_mmap_calls;
  36static unsigned int peak_pack_open_windows;
  37static unsigned int pack_open_windows;
  38static unsigned int pack_open_fds;
  39static unsigned int pack_max_fds;
  40static size_t peak_pack_mapped;
  41static size_t pack_mapped;
  42struct packed_git *packed_git;
  43
  44static struct mru packed_git_mru_storage;
  45struct mru *packed_git_mru = &packed_git_mru_storage;
  46
  47#define SZ_FMT PRIuMAX
  48static inline uintmax_t sz_fmt(size_t s) { return s; }
  49
  50void pack_report(void)
  51{
  52        fprintf(stderr,
  53                "pack_report: getpagesize()            = %10" SZ_FMT "\n"
  54                "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
  55                "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
  56                sz_fmt(getpagesize()),
  57                sz_fmt(packed_git_window_size),
  58                sz_fmt(packed_git_limit));
  59        fprintf(stderr,
  60                "pack_report: pack_used_ctr            = %10u\n"
  61                "pack_report: pack_mmap_calls          = %10u\n"
  62                "pack_report: pack_open_windows        = %10u / %10u\n"
  63                "pack_report: pack_mapped              = "
  64                        "%10" SZ_FMT " / %10" SZ_FMT "\n",
  65                pack_used_ctr,
  66                pack_mmap_calls,
  67                pack_open_windows, peak_pack_open_windows,
  68                sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
  69}
  70
  71/*
  72 * Open and mmap the index file at path, perform a couple of
  73 * consistency checks, then record its information to p.  Return 0 on
  74 * success.
  75 */
  76static int check_packed_git_idx(const char *path, struct packed_git *p)
  77{
  78        void *idx_map;
  79        struct pack_idx_header *hdr;
  80        size_t idx_size;
  81        uint32_t version, nr, i, *index;
  82        int fd = git_open(path);
  83        struct stat st;
  84
  85        if (fd < 0)
  86                return -1;
  87        if (fstat(fd, &st)) {
  88                close(fd);
  89                return -1;
  90        }
  91        idx_size = xsize_t(st.st_size);
  92        if (idx_size < 4 * 256 + 20 + 20) {
  93                close(fd);
  94                return error("index file %s is too small", path);
  95        }
  96        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
  97        close(fd);
  98
  99        hdr = idx_map;
 100        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 101                version = ntohl(hdr->idx_version);
 102                if (version < 2 || version > 2) {
 103                        munmap(idx_map, idx_size);
 104                        return error("index file %s is version %"PRIu32
 105                                     " and is not supported by this binary"
 106                                     " (try upgrading GIT to a newer version)",
 107                                     path, version);
 108                }
 109        } else
 110                version = 1;
 111
 112        nr = 0;
 113        index = idx_map;
 114        if (version > 1)
 115                index += 2;  /* skip index header */
 116        for (i = 0; i < 256; i++) {
 117                uint32_t n = ntohl(index[i]);
 118                if (n < nr) {
 119                        munmap(idx_map, idx_size);
 120                        return error("non-monotonic index %s", path);
 121                }
 122                nr = n;
 123        }
 124
 125        if (version == 1) {
 126                /*
 127                 * Total size:
 128                 *  - 256 index entries 4 bytes each
 129                 *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
 130                 *  - 20-byte SHA1 of the packfile
 131                 *  - 20-byte SHA1 file checksum
 132                 */
 133                if (idx_size != 4*256 + nr * 24 + 20 + 20) {
 134                        munmap(idx_map, idx_size);
 135                        return error("wrong index v1 file size in %s", path);
 136                }
 137        } else if (version == 2) {
 138                /*
 139                 * Minimum size:
 140                 *  - 8 bytes of header
 141                 *  - 256 index entries 4 bytes each
 142                 *  - 20-byte sha1 entry * nr
 143                 *  - 4-byte crc entry * nr
 144                 *  - 4-byte offset entry * nr
 145                 *  - 20-byte SHA1 of the packfile
 146                 *  - 20-byte SHA1 file checksum
 147                 * And after the 4-byte offset table might be a
 148                 * variable sized table containing 8-byte entries
 149                 * for offsets larger than 2^31.
 150                 */
 151                unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
 152                unsigned long max_size = min_size;
 153                if (nr)
 154                        max_size += (nr - 1)*8;
 155                if (idx_size < min_size || idx_size > max_size) {
 156                        munmap(idx_map, idx_size);
 157                        return error("wrong index v2 file size in %s", path);
 158                }
 159                if (idx_size != min_size &&
 160                    /*
 161                     * make sure we can deal with large pack offsets.
 162                     * 31-bit signed offset won't be enough, neither
 163                     * 32-bit unsigned one will be.
 164                     */
 165                    (sizeof(off_t) <= 4)) {
 166                        munmap(idx_map, idx_size);
 167                        return error("pack too large for current definition of off_t in %s", path);
 168                }
 169        }
 170
 171        p->index_version = version;
 172        p->index_data = idx_map;
 173        p->index_size = idx_size;
 174        p->num_objects = nr;
 175        return 0;
 176}
 177
 178int open_pack_index(struct packed_git *p)
 179{
 180        char *idx_name;
 181        size_t len;
 182        int ret;
 183
 184        if (p->index_data)
 185                return 0;
 186
 187        if (!strip_suffix(p->pack_name, ".pack", &len))
 188                die("BUG: pack_name does not end in .pack");
 189        idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
 190        ret = check_packed_git_idx(idx_name, p);
 191        free(idx_name);
 192        return ret;
 193}
 194
 195static struct packed_git *alloc_packed_git(int extra)
 196{
 197        struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
 198        memset(p, 0, sizeof(*p));
 199        p->pack_fd = -1;
 200        return p;
 201}
 202
 203struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
 204{
 205        const char *path = sha1_pack_name(sha1);
 206        size_t alloc = st_add(strlen(path), 1);
 207        struct packed_git *p = alloc_packed_git(alloc);
 208
 209        memcpy(p->pack_name, path, alloc); /* includes NUL */
 210        hashcpy(p->sha1, sha1);
 211        if (check_packed_git_idx(idx_path, p)) {
 212                free(p);
 213                return NULL;
 214        }
 215
 216        return p;
 217}
 218
 219static void scan_windows(struct packed_git *p,
 220        struct packed_git **lru_p,
 221        struct pack_window **lru_w,
 222        struct pack_window **lru_l)
 223{
 224        struct pack_window *w, *w_l;
 225
 226        for (w_l = NULL, w = p->windows; w; w = w->next) {
 227                if (!w->inuse_cnt) {
 228                        if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 229                                *lru_p = p;
 230                                *lru_w = w;
 231                                *lru_l = w_l;
 232                        }
 233                }
 234                w_l = w;
 235        }
 236}
 237
 238static int unuse_one_window(struct packed_git *current)
 239{
 240        struct packed_git *p, *lru_p = NULL;
 241        struct pack_window *lru_w = NULL, *lru_l = NULL;
 242
 243        if (current)
 244                scan_windows(current, &lru_p, &lru_w, &lru_l);
 245        for (p = packed_git; p; p = p->next)
 246                scan_windows(p, &lru_p, &lru_w, &lru_l);
 247        if (lru_p) {
 248                munmap(lru_w->base, lru_w->len);
 249                pack_mapped -= lru_w->len;
 250                if (lru_l)
 251                        lru_l->next = lru_w->next;
 252                else
 253                        lru_p->windows = lru_w->next;
 254                free(lru_w);
 255                pack_open_windows--;
 256                return 1;
 257        }
 258        return 0;
 259}
 260
 261void release_pack_memory(size_t need)
 262{
 263        size_t cur = pack_mapped;
 264        while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
 265                ; /* nothing */
 266}
 267
 268void close_pack_windows(struct packed_git *p)
 269{
 270        while (p->windows) {
 271                struct pack_window *w = p->windows;
 272
 273                if (w->inuse_cnt)
 274                        die("pack '%s' still has open windows to it",
 275                            p->pack_name);
 276                munmap(w->base, w->len);
 277                pack_mapped -= w->len;
 278                pack_open_windows--;
 279                p->windows = w->next;
 280                free(w);
 281        }
 282}
 283
 284static int close_pack_fd(struct packed_git *p)
 285{
 286        if (p->pack_fd < 0)
 287                return 0;
 288
 289        close(p->pack_fd);
 290        pack_open_fds--;
 291        p->pack_fd = -1;
 292
 293        return 1;
 294}
 295
 296void close_pack_index(struct packed_git *p)
 297{
 298        if (p->index_data) {
 299                munmap((void *)p->index_data, p->index_size);
 300                p->index_data = NULL;
 301        }
 302}
 303
 304static void close_pack(struct packed_git *p)
 305{
 306        close_pack_windows(p);
 307        close_pack_fd(p);
 308        close_pack_index(p);
 309}
 310
 311void close_all_packs(void)
 312{
 313        struct packed_git *p;
 314
 315        for (p = packed_git; p; p = p->next)
 316                if (p->do_not_close)
 317                        die("BUG: want to close pack marked 'do-not-close'");
 318                else
 319                        close_pack(p);
 320}
 321
 322/*
 323 * The LRU pack is the one with the oldest MRU window, preferring packs
 324 * with no used windows, or the oldest mtime if it has no windows allocated.
 325 */
 326static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
 327{
 328        struct pack_window *w, *this_mru_w;
 329        int has_windows_inuse = 0;
 330
 331        /*
 332         * Reject this pack if it has windows and the previously selected
 333         * one does not.  If this pack does not have windows, reject
 334         * it if the pack file is newer than the previously selected one.
 335         */
 336        if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
 337                return;
 338
 339        for (w = this_mru_w = p->windows; w; w = w->next) {
 340                /*
 341                 * Reject this pack if any of its windows are in use,
 342                 * but the previously selected pack did not have any
 343                 * inuse windows.  Otherwise, record that this pack
 344                 * has windows in use.
 345                 */
 346                if (w->inuse_cnt) {
 347                        if (*accept_windows_inuse)
 348                                has_windows_inuse = 1;
 349                        else
 350                                return;
 351                }
 352
 353                if (w->last_used > this_mru_w->last_used)
 354                        this_mru_w = w;
 355
 356                /*
 357                 * Reject this pack if it has windows that have been
 358                 * used more recently than the previously selected pack.
 359                 * If the previously selected pack had windows inuse and
 360                 * we have not encountered a window in this pack that is
 361                 * inuse, skip this check since we prefer a pack with no
 362                 * inuse windows to one that has inuse windows.
 363                 */
 364                if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
 365                    this_mru_w->last_used > (*mru_w)->last_used)
 366                        return;
 367        }
 368
 369        /*
 370         * Select this pack.
 371         */
 372        *mru_w = this_mru_w;
 373        *lru_p = p;
 374        *accept_windows_inuse = has_windows_inuse;
 375}
 376
 377static int close_one_pack(void)
 378{
 379        struct packed_git *p, *lru_p = NULL;
 380        struct pack_window *mru_w = NULL;
 381        int accept_windows_inuse = 1;
 382
 383        for (p = packed_git; p; p = p->next) {
 384                if (p->pack_fd == -1)
 385                        continue;
 386                find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
 387        }
 388
 389        if (lru_p)
 390                return close_pack_fd(lru_p);
 391
 392        return 0;
 393}
 394
 395static unsigned int get_max_fd_limit(void)
 396{
 397#ifdef RLIMIT_NOFILE
 398        {
 399                struct rlimit lim;
 400
 401                if (!getrlimit(RLIMIT_NOFILE, &lim))
 402                        return lim.rlim_cur;
 403        }
 404#endif
 405
 406#ifdef _SC_OPEN_MAX
 407        {
 408                long open_max = sysconf(_SC_OPEN_MAX);
 409                if (0 < open_max)
 410                        return open_max;
 411                /*
 412                 * Otherwise, we got -1 for one of the two
 413                 * reasons:
 414                 *
 415                 * (1) sysconf() did not understand _SC_OPEN_MAX
 416                 *     and signaled an error with -1; or
 417                 * (2) sysconf() said there is no limit.
 418                 *
 419                 * We _could_ clear errno before calling sysconf() to
 420                 * tell these two cases apart and return a huge number
 421                 * in the latter case to let the caller cap it to a
 422                 * value that is not so selfish, but letting the
 423                 * fallback OPEN_MAX codepath take care of these cases
 424                 * is a lot simpler.
 425                 */
 426        }
 427#endif
 428
 429#ifdef OPEN_MAX
 430        return OPEN_MAX;
 431#else
 432        return 1; /* see the caller ;-) */
 433#endif
 434}
 435
 436/*
 437 * Do not call this directly as this leaks p->pack_fd on error return;
 438 * call open_packed_git() instead.
 439 */
 440static int open_packed_git_1(struct packed_git *p)
 441{
 442        struct stat st;
 443        struct pack_header hdr;
 444        unsigned char sha1[20];
 445        unsigned char *idx_sha1;
 446        long fd_flag;
 447
 448        if (!p->index_data && open_pack_index(p))
 449                return error("packfile %s index unavailable", p->pack_name);
 450
 451        if (!pack_max_fds) {
 452                unsigned int max_fds = get_max_fd_limit();
 453
 454                /* Save 3 for stdin/stdout/stderr, 22 for work */
 455                if (25 < max_fds)
 456                        pack_max_fds = max_fds - 25;
 457                else
 458                        pack_max_fds = 1;
 459        }
 460
 461        while (pack_max_fds <= pack_open_fds && close_one_pack())
 462                ; /* nothing */
 463
 464        p->pack_fd = git_open(p->pack_name);
 465        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 466                return -1;
 467        pack_open_fds++;
 468
 469        /* If we created the struct before we had the pack we lack size. */
 470        if (!p->pack_size) {
 471                if (!S_ISREG(st.st_mode))
 472                        return error("packfile %s not a regular file", p->pack_name);
 473                p->pack_size = st.st_size;
 474        } else if (p->pack_size != st.st_size)
 475                return error("packfile %s size changed", p->pack_name);
 476
 477        /* We leave these file descriptors open with sliding mmap;
 478         * there is no point keeping them open across exec(), though.
 479         */
 480        fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
 481        if (fd_flag < 0)
 482                return error("cannot determine file descriptor flags");
 483        fd_flag |= FD_CLOEXEC;
 484        if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
 485                return error("cannot set FD_CLOEXEC");
 486
 487        /* Verify we recognize this pack file format. */
 488        if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
 489                return error("file %s is far too short to be a packfile", p->pack_name);
 490        if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 491                return error("file %s is not a GIT packfile", p->pack_name);
 492        if (!pack_version_ok(hdr.hdr_version))
 493                return error("packfile %s is version %"PRIu32" and not"
 494                        " supported (try upgrading GIT to a newer version)",
 495                        p->pack_name, ntohl(hdr.hdr_version));
 496
 497        /* Verify the pack matches its index. */
 498        if (p->num_objects != ntohl(hdr.hdr_entries))
 499                return error("packfile %s claims to have %"PRIu32" objects"
 500                             " while index indicates %"PRIu32" objects",
 501                             p->pack_name, ntohl(hdr.hdr_entries),
 502                             p->num_objects);
 503        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
 504                return error("end of packfile %s is unavailable", p->pack_name);
 505        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
 506                return error("packfile %s signature is unavailable", p->pack_name);
 507        idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
 508        if (hashcmp(sha1, idx_sha1))
 509                return error("packfile %s does not match index", p->pack_name);
 510        return 0;
 511}
 512
 513static int open_packed_git(struct packed_git *p)
 514{
 515        if (!open_packed_git_1(p))
 516                return 0;
 517        close_pack_fd(p);
 518        return -1;
 519}
 520
 521static int in_window(struct pack_window *win, off_t offset)
 522{
 523        /* We must promise at least 20 bytes (one hash) after the
 524         * offset is available from this window, otherwise the offset
 525         * is not actually in this window and a different window (which
 526         * has that one hash excess) must be used.  This is to support
 527         * the object header and delta base parsing routines below.
 528         */
 529        off_t win_off = win->offset;
 530        return win_off <= offset
 531                && (offset + 20) <= (win_off + win->len);
 532}
 533
 534unsigned char *use_pack(struct packed_git *p,
 535                struct pack_window **w_cursor,
 536                off_t offset,
 537                unsigned long *left)
 538{
 539        struct pack_window *win = *w_cursor;
 540
 541        /* Since packfiles end in a hash of their content and it's
 542         * pointless to ask for an offset into the middle of that
 543         * hash, and the in_window function above wouldn't match
 544         * don't allow an offset too close to the end of the file.
 545         */
 546        if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
 547                die("packfile %s cannot be accessed", p->pack_name);
 548        if (offset > (p->pack_size - 20))
 549                die("offset beyond end of packfile (truncated pack?)");
 550        if (offset < 0)
 551                die(_("offset before end of packfile (broken .idx?)"));
 552
 553        if (!win || !in_window(win, offset)) {
 554                if (win)
 555                        win->inuse_cnt--;
 556                for (win = p->windows; win; win = win->next) {
 557                        if (in_window(win, offset))
 558                                break;
 559                }
 560                if (!win) {
 561                        size_t window_align = packed_git_window_size / 2;
 562                        off_t len;
 563
 564                        if (p->pack_fd == -1 && open_packed_git(p))
 565                                die("packfile %s cannot be accessed", p->pack_name);
 566
 567                        win = xcalloc(1, sizeof(*win));
 568                        win->offset = (offset / window_align) * window_align;
 569                        len = p->pack_size - win->offset;
 570                        if (len > packed_git_window_size)
 571                                len = packed_git_window_size;
 572                        win->len = (size_t)len;
 573                        pack_mapped += win->len;
 574                        while (packed_git_limit < pack_mapped
 575                                && unuse_one_window(p))
 576                                ; /* nothing */
 577                        win->base = xmmap(NULL, win->len,
 578                                PROT_READ, MAP_PRIVATE,
 579                                p->pack_fd, win->offset);
 580                        if (win->base == MAP_FAILED)
 581                                die_errno("packfile %s cannot be mapped",
 582                                          p->pack_name);
 583                        if (!win->offset && win->len == p->pack_size
 584                                && !p->do_not_close)
 585                                close_pack_fd(p);
 586                        pack_mmap_calls++;
 587                        pack_open_windows++;
 588                        if (pack_mapped > peak_pack_mapped)
 589                                peak_pack_mapped = pack_mapped;
 590                        if (pack_open_windows > peak_pack_open_windows)
 591                                peak_pack_open_windows = pack_open_windows;
 592                        win->next = p->windows;
 593                        p->windows = win;
 594                }
 595        }
 596        if (win != *w_cursor) {
 597                win->last_used = pack_used_ctr++;
 598                win->inuse_cnt++;
 599                *w_cursor = win;
 600        }
 601        offset -= win->offset;
 602        if (left)
 603                *left = win->len - xsize_t(offset);
 604        return win->base + offset;
 605}
 606
 607void unuse_pack(struct pack_window **w_cursor)
 608{
 609        struct pack_window *w = *w_cursor;
 610        if (w) {
 611                w->inuse_cnt--;
 612                *w_cursor = NULL;
 613        }
 614}
 615
 616static void try_to_free_pack_memory(size_t size)
 617{
 618        release_pack_memory(size);
 619}
 620
 621struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
 622{
 623        static int have_set_try_to_free_routine;
 624        struct stat st;
 625        size_t alloc;
 626        struct packed_git *p;
 627
 628        if (!have_set_try_to_free_routine) {
 629                have_set_try_to_free_routine = 1;
 630                set_try_to_free_routine(try_to_free_pack_memory);
 631        }
 632
 633        /*
 634         * Make sure a corresponding .pack file exists and that
 635         * the index looks sane.
 636         */
 637        if (!strip_suffix_mem(path, &path_len, ".idx"))
 638                return NULL;
 639
 640        /*
 641         * ".pack" is long enough to hold any suffix we're adding (and
 642         * the use xsnprintf double-checks that)
 643         */
 644        alloc = st_add3(path_len, strlen(".pack"), 1);
 645        p = alloc_packed_git(alloc);
 646        memcpy(p->pack_name, path, path_len);
 647
 648        xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
 649        if (!access(p->pack_name, F_OK))
 650                p->pack_keep = 1;
 651
 652        xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
 653        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 654                free(p);
 655                return NULL;
 656        }
 657
 658        /* ok, it looks sane as far as we can check without
 659         * actually mapping the pack file.
 660         */
 661        p->pack_size = st.st_size;
 662        p->pack_local = local;
 663        p->mtime = st.st_mtime;
 664        if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
 665                hashclr(p->sha1);
 666        return p;
 667}
 668
 669void install_packed_git(struct packed_git *pack)
 670{
 671        if (pack->pack_fd != -1)
 672                pack_open_fds++;
 673
 674        pack->next = packed_git;
 675        packed_git = pack;
 676}
 677
 678void (*report_garbage)(unsigned seen_bits, const char *path);
 679
 680static void report_helper(const struct string_list *list,
 681                          int seen_bits, int first, int last)
 682{
 683        if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
 684                return;
 685
 686        for (; first < last; first++)
 687                report_garbage(seen_bits, list->items[first].string);
 688}
 689
 690static void report_pack_garbage(struct string_list *list)
 691{
 692        int i, baselen = -1, first = 0, seen_bits = 0;
 693
 694        if (!report_garbage)
 695                return;
 696
 697        string_list_sort(list);
 698
 699        for (i = 0; i < list->nr; i++) {
 700                const char *path = list->items[i].string;
 701                if (baselen != -1 &&
 702                    strncmp(path, list->items[first].string, baselen)) {
 703                        report_helper(list, seen_bits, first, i);
 704                        baselen = -1;
 705                        seen_bits = 0;
 706                }
 707                if (baselen == -1) {
 708                        const char *dot = strrchr(path, '.');
 709                        if (!dot) {
 710                                report_garbage(PACKDIR_FILE_GARBAGE, path);
 711                                continue;
 712                        }
 713                        baselen = dot - path + 1;
 714                        first = i;
 715                }
 716                if (!strcmp(path + baselen, "pack"))
 717                        seen_bits |= 1;
 718                else if (!strcmp(path + baselen, "idx"))
 719                        seen_bits |= 2;
 720        }
 721        report_helper(list, seen_bits, first, list->nr);
 722}
 723
 724static void prepare_packed_git_one(char *objdir, int local)
 725{
 726        struct strbuf path = STRBUF_INIT;
 727        size_t dirnamelen;
 728        DIR *dir;
 729        struct dirent *de;
 730        struct string_list garbage = STRING_LIST_INIT_DUP;
 731
 732        strbuf_addstr(&path, objdir);
 733        strbuf_addstr(&path, "/pack");
 734        dir = opendir(path.buf);
 735        if (!dir) {
 736                if (errno != ENOENT)
 737                        error_errno("unable to open object pack directory: %s",
 738                                    path.buf);
 739                strbuf_release(&path);
 740                return;
 741        }
 742        strbuf_addch(&path, '/');
 743        dirnamelen = path.len;
 744        while ((de = readdir(dir)) != NULL) {
 745                struct packed_git *p;
 746                size_t base_len;
 747
 748                if (is_dot_or_dotdot(de->d_name))
 749                        continue;
 750
 751                strbuf_setlen(&path, dirnamelen);
 752                strbuf_addstr(&path, de->d_name);
 753
 754                base_len = path.len;
 755                if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
 756                        /* Don't reopen a pack we already have. */
 757                        for (p = packed_git; p; p = p->next) {
 758                                size_t len;
 759                                if (strip_suffix(p->pack_name, ".pack", &len) &&
 760                                    len == base_len &&
 761                                    !memcmp(p->pack_name, path.buf, len))
 762                                        break;
 763                        }
 764                        if (p == NULL &&
 765                            /*
 766                             * See if it really is a valid .idx file with
 767                             * corresponding .pack file that we can map.
 768                             */
 769                            (p = add_packed_git(path.buf, path.len, local)) != NULL)
 770                                install_packed_git(p);
 771                }
 772
 773                if (!report_garbage)
 774                        continue;
 775
 776                if (ends_with(de->d_name, ".idx") ||
 777                    ends_with(de->d_name, ".pack") ||
 778                    ends_with(de->d_name, ".bitmap") ||
 779                    ends_with(de->d_name, ".keep"))
 780                        string_list_append(&garbage, path.buf);
 781                else
 782                        report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
 783        }
 784        closedir(dir);
 785        report_pack_garbage(&garbage);
 786        string_list_clear(&garbage, 0);
 787        strbuf_release(&path);
 788}
 789
 790static int approximate_object_count_valid;
 791
 792/*
 793 * Give a fast, rough count of the number of objects in the repository. This
 794 * ignores loose objects completely. If you have a lot of them, then either
 795 * you should repack because your performance will be awful, or they are
 796 * all unreachable objects about to be pruned, in which case they're not really
 797 * interesting as a measure of repo size in the first place.
 798 */
 799unsigned long approximate_object_count(void)
 800{
 801        static unsigned long count;
 802        if (!approximate_object_count_valid) {
 803                struct packed_git *p;
 804
 805                prepare_packed_git();
 806                count = 0;
 807                for (p = packed_git; p; p = p->next) {
 808                        if (open_pack_index(p))
 809                                continue;
 810                        count += p->num_objects;
 811                }
 812        }
 813        return count;
 814}
 815
 816static void *get_next_packed_git(const void *p)
 817{
 818        return ((const struct packed_git *)p)->next;
 819}
 820
 821static void set_next_packed_git(void *p, void *next)
 822{
 823        ((struct packed_git *)p)->next = next;
 824}
 825
 826static int sort_pack(const void *a_, const void *b_)
 827{
 828        const struct packed_git *a = a_;
 829        const struct packed_git *b = b_;
 830        int st;
 831
 832        /*
 833         * Local packs tend to contain objects specific to our
 834         * variant of the project than remote ones.  In addition,
 835         * remote ones could be on a network mounted filesystem.
 836         * Favor local ones for these reasons.
 837         */
 838        st = a->pack_local - b->pack_local;
 839        if (st)
 840                return -st;
 841
 842        /*
 843         * Younger packs tend to contain more recent objects,
 844         * and more recent objects tend to get accessed more
 845         * often.
 846         */
 847        if (a->mtime < b->mtime)
 848                return 1;
 849        else if (a->mtime == b->mtime)
 850                return 0;
 851        return -1;
 852}
 853
 854static void rearrange_packed_git(void)
 855{
 856        packed_git = llist_mergesort(packed_git, get_next_packed_git,
 857                                     set_next_packed_git, sort_pack);
 858}
 859
 860static void prepare_packed_git_mru(void)
 861{
 862        struct packed_git *p;
 863
 864        mru_clear(packed_git_mru);
 865        for (p = packed_git; p; p = p->next)
 866                mru_append(packed_git_mru, p);
 867}
 868
 869static int prepare_packed_git_run_once = 0;
 870void prepare_packed_git(void)
 871{
 872        struct alternate_object_database *alt;
 873
 874        if (prepare_packed_git_run_once)
 875                return;
 876        prepare_packed_git_one(get_object_directory(), 1);
 877        prepare_alt_odb();
 878        for (alt = alt_odb_list; alt; alt = alt->next)
 879                prepare_packed_git_one(alt->path, 0);
 880        rearrange_packed_git();
 881        prepare_packed_git_mru();
 882        prepare_packed_git_run_once = 1;
 883}
 884
 885void reprepare_packed_git(void)
 886{
 887        approximate_object_count_valid = 0;
 888        prepare_packed_git_run_once = 0;
 889        prepare_packed_git();
 890}
 891
 892unsigned long unpack_object_header_buffer(const unsigned char *buf,
 893                unsigned long len, enum object_type *type, unsigned long *sizep)
 894{
 895        unsigned shift;
 896        unsigned long size, c;
 897        unsigned long used = 0;
 898
 899        c = buf[used++];
 900        *type = (c >> 4) & 7;
 901        size = c & 15;
 902        shift = 4;
 903        while (c & 0x80) {
 904                if (len <= used || bitsizeof(long) <= shift) {
 905                        error("bad object header");
 906                        size = used = 0;
 907                        break;
 908                }
 909                c = buf[used++];
 910                size += (c & 0x7f) << shift;
 911                shift += 7;
 912        }
 913        *sizep = size;
 914        return used;
 915}
 916
 917unsigned long get_size_from_delta(struct packed_git *p,
 918                                  struct pack_window **w_curs,
 919                                  off_t curpos)
 920{
 921        const unsigned char *data;
 922        unsigned char delta_head[20], *in;
 923        git_zstream stream;
 924        int st;
 925
 926        memset(&stream, 0, sizeof(stream));
 927        stream.next_out = delta_head;
 928        stream.avail_out = sizeof(delta_head);
 929
 930        git_inflate_init(&stream);
 931        do {
 932                in = use_pack(p, w_curs, curpos, &stream.avail_in);
 933                stream.next_in = in;
 934                st = git_inflate(&stream, Z_FINISH);
 935                curpos += stream.next_in - in;
 936        } while ((st == Z_OK || st == Z_BUF_ERROR) &&
 937                 stream.total_out < sizeof(delta_head));
 938        git_inflate_end(&stream);
 939        if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
 940                error("delta data unpack-initial failed");
 941                return 0;
 942        }
 943
 944        /* Examine the initial part of the delta to figure out
 945         * the result size.
 946         */
 947        data = delta_head;
 948
 949        /* ignore base size */
 950        get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
 951
 952        /* Read the result size */
 953        return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
 954}
 955
 956int unpack_object_header(struct packed_git *p,
 957                         struct pack_window **w_curs,
 958                         off_t *curpos,
 959                         unsigned long *sizep)
 960{
 961        unsigned char *base;
 962        unsigned long left;
 963        unsigned long used;
 964        enum object_type type;
 965
 966        /* use_pack() assures us we have [base, base + 20) available
 967         * as a range that we can look at.  (Its actually the hash
 968         * size that is assured.)  With our object header encoding
 969         * the maximum deflated object size is 2^137, which is just
 970         * insane, so we know won't exceed what we have been given.
 971         */
 972        base = use_pack(p, w_curs, *curpos, &left);
 973        used = unpack_object_header_buffer(base, left, &type, sizep);
 974        if (!used) {
 975                type = OBJ_BAD;
 976        } else
 977                *curpos += used;
 978
 979        return type;
 980}
 981
 982void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
 983{
 984        unsigned i;
 985        for (i = 0; i < p->num_bad_objects; i++)
 986                if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
 987                        return;
 988        p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
 989                                      st_mult(GIT_MAX_RAWSZ,
 990                                              st_add(p->num_bad_objects, 1)));
 991        hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
 992        p->num_bad_objects++;
 993}
 994
 995const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
 996{
 997        struct packed_git *p;
 998        unsigned i;
 999
1000        for (p = packed_git; p; p = p->next)
1001                for (i = 0; i < p->num_bad_objects; i++)
1002                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1003                                return p;
1004        return NULL;
1005}
1006
1007static off_t get_delta_base(struct packed_git *p,
1008                                    struct pack_window **w_curs,
1009                                    off_t *curpos,
1010                                    enum object_type type,
1011                                    off_t delta_obj_offset)
1012{
1013        unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1014        off_t base_offset;
1015
1016        /* use_pack() assured us we have [base_info, base_info + 20)
1017         * as a range that we can look at without walking off the
1018         * end of the mapped window.  Its actually the hash size
1019         * that is assured.  An OFS_DELTA longer than the hash size
1020         * is stupid, as then a REF_DELTA would be smaller to store.
1021         */
1022        if (type == OBJ_OFS_DELTA) {
1023                unsigned used = 0;
1024                unsigned char c = base_info[used++];
1025                base_offset = c & 127;
1026                while (c & 128) {
1027                        base_offset += 1;
1028                        if (!base_offset || MSB(base_offset, 7))
1029                                return 0;  /* overflow */
1030                        c = base_info[used++];
1031                        base_offset = (base_offset << 7) + (c & 127);
1032                }
1033                base_offset = delta_obj_offset - base_offset;
1034                if (base_offset <= 0 || base_offset >= delta_obj_offset)
1035                        return 0;  /* out of bound */
1036                *curpos += used;
1037        } else if (type == OBJ_REF_DELTA) {
1038                /* The base entry _must_ be in the same pack */
1039                base_offset = find_pack_entry_one(base_info, p);
1040                *curpos += 20;
1041        } else
1042                die("I am totally screwed");
1043        return base_offset;
1044}
1045
1046/*
1047 * Like get_delta_base above, but we return the sha1 instead of the pack
1048 * offset. This means it is cheaper for REF deltas (we do not have to do
1049 * the final object lookup), but more expensive for OFS deltas (we
1050 * have to load the revidx to convert the offset back into a sha1).
1051 */
1052static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1053                                                struct pack_window **w_curs,
1054                                                off_t curpos,
1055                                                enum object_type type,
1056                                                off_t delta_obj_offset)
1057{
1058        if (type == OBJ_REF_DELTA) {
1059                unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1060                return base;
1061        } else if (type == OBJ_OFS_DELTA) {
1062                struct revindex_entry *revidx;
1063                off_t base_offset = get_delta_base(p, w_curs, &curpos,
1064                                                   type, delta_obj_offset);
1065
1066                if (!base_offset)
1067                        return NULL;
1068
1069                revidx = find_pack_revindex(p, base_offset);
1070                if (!revidx)
1071                        return NULL;
1072
1073                return nth_packed_object_sha1(p, revidx->nr);
1074        } else
1075                return NULL;
1076}
1077
1078static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1079{
1080        int type;
1081        struct revindex_entry *revidx;
1082        const unsigned char *sha1;
1083        revidx = find_pack_revindex(p, obj_offset);
1084        if (!revidx)
1085                return OBJ_BAD;
1086        sha1 = nth_packed_object_sha1(p, revidx->nr);
1087        mark_bad_packed_object(p, sha1);
1088        type = sha1_object_info(sha1, NULL);
1089        if (type <= OBJ_NONE)
1090                return OBJ_BAD;
1091        return type;
1092}
1093
1094#define POI_STACK_PREALLOC 64
1095
1096static enum object_type packed_to_object_type(struct packed_git *p,
1097                                              off_t obj_offset,
1098                                              enum object_type type,
1099                                              struct pack_window **w_curs,
1100                                              off_t curpos)
1101{
1102        off_t small_poi_stack[POI_STACK_PREALLOC];
1103        off_t *poi_stack = small_poi_stack;
1104        int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1105
1106        while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1107                off_t base_offset;
1108                unsigned long size;
1109                /* Push the object we're going to leave behind */
1110                if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1111                        poi_stack_alloc = alloc_nr(poi_stack_nr);
1112                        ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1113                        memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1114                } else {
1115                        ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1116                }
1117                poi_stack[poi_stack_nr++] = obj_offset;
1118                /* If parsing the base offset fails, just unwind */
1119                base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1120                if (!base_offset)
1121                        goto unwind;
1122                curpos = obj_offset = base_offset;
1123                type = unpack_object_header(p, w_curs, &curpos, &size);
1124                if (type <= OBJ_NONE) {
1125                        /* If getting the base itself fails, we first
1126                         * retry the base, otherwise unwind */
1127                        type = retry_bad_packed_offset(p, base_offset);
1128                        if (type > OBJ_NONE)
1129                                goto out;
1130                        goto unwind;
1131                }
1132        }
1133
1134        switch (type) {
1135        case OBJ_BAD:
1136        case OBJ_COMMIT:
1137        case OBJ_TREE:
1138        case OBJ_BLOB:
1139        case OBJ_TAG:
1140                break;
1141        default:
1142                error("unknown object type %i at offset %"PRIuMAX" in %s",
1143                      type, (uintmax_t)obj_offset, p->pack_name);
1144                type = OBJ_BAD;
1145        }
1146
1147out:
1148        if (poi_stack != small_poi_stack)
1149                free(poi_stack);
1150        return type;
1151
1152unwind:
1153        while (poi_stack_nr) {
1154                obj_offset = poi_stack[--poi_stack_nr];
1155                type = retry_bad_packed_offset(p, obj_offset);
1156                if (type > OBJ_NONE)
1157                        goto out;
1158        }
1159        type = OBJ_BAD;
1160        goto out;
1161}
1162
1163static struct hashmap delta_base_cache;
1164static size_t delta_base_cached;
1165
1166static LIST_HEAD(delta_base_cache_lru);
1167
1168struct delta_base_cache_key {
1169        struct packed_git *p;
1170        off_t base_offset;
1171};
1172
1173struct delta_base_cache_entry {
1174        struct hashmap hash;
1175        struct delta_base_cache_key key;
1176        struct list_head lru;
1177        void *data;
1178        unsigned long size;
1179        enum object_type type;
1180};
1181
1182static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1183{
1184        unsigned int hash;
1185
1186        hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1187        hash += (hash >> 8) + (hash >> 16);
1188        return hash;
1189}
1190
1191static struct delta_base_cache_entry *
1192get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1193{
1194        struct hashmap_entry entry;
1195        struct delta_base_cache_key key;
1196
1197        if (!delta_base_cache.cmpfn)
1198                return NULL;
1199
1200        hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1201        key.p = p;
1202        key.base_offset = base_offset;
1203        return hashmap_get(&delta_base_cache, &entry, &key);
1204}
1205
1206static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1207                                   const struct delta_base_cache_key *b)
1208{
1209        return a->p == b->p && a->base_offset == b->base_offset;
1210}
1211
1212static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1213                                     const void *va, const void *vb,
1214                                     const void *vkey)
1215{
1216        const struct delta_base_cache_entry *a = va, *b = vb;
1217        const struct delta_base_cache_key *key = vkey;
1218        if (key)
1219                return !delta_base_cache_key_eq(&a->key, key);
1220        else
1221                return !delta_base_cache_key_eq(&a->key, &b->key);
1222}
1223
1224static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1225{
1226        return !!get_delta_base_cache_entry(p, base_offset);
1227}
1228
1229/*
1230 * Remove the entry from the cache, but do _not_ free the associated
1231 * entry data. The caller takes ownership of the "data" buffer, and
1232 * should copy out any fields it wants before detaching.
1233 */
1234static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1235{
1236        hashmap_remove(&delta_base_cache, ent, &ent->key);
1237        list_del(&ent->lru);
1238        delta_base_cached -= ent->size;
1239        free(ent);
1240}
1241
1242static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1243        unsigned long *base_size, enum object_type *type)
1244{
1245        struct delta_base_cache_entry *ent;
1246
1247        ent = get_delta_base_cache_entry(p, base_offset);
1248        if (!ent)
1249                return unpack_entry(p, base_offset, type, base_size);
1250
1251        if (type)
1252                *type = ent->type;
1253        if (base_size)
1254                *base_size = ent->size;
1255        return xmemdupz(ent->data, ent->size);
1256}
1257
1258static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1259{
1260        free(ent->data);
1261        detach_delta_base_cache_entry(ent);
1262}
1263
1264void clear_delta_base_cache(void)
1265{
1266        struct list_head *lru, *tmp;
1267        list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1268                struct delta_base_cache_entry *entry =
1269                        list_entry(lru, struct delta_base_cache_entry, lru);
1270                release_delta_base_cache(entry);
1271        }
1272}
1273
1274static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1275        void *base, unsigned long base_size, enum object_type type)
1276{
1277        struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1278        struct list_head *lru, *tmp;
1279
1280        delta_base_cached += base_size;
1281
1282        list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1283                struct delta_base_cache_entry *f =
1284                        list_entry(lru, struct delta_base_cache_entry, lru);
1285                if (delta_base_cached <= delta_base_cache_limit)
1286                        break;
1287                release_delta_base_cache(f);
1288        }
1289
1290        ent->key.p = p;
1291        ent->key.base_offset = base_offset;
1292        ent->type = type;
1293        ent->data = base;
1294        ent->size = base_size;
1295        list_add_tail(&ent->lru, &delta_base_cache_lru);
1296
1297        if (!delta_base_cache.cmpfn)
1298                hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1299        hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
1300        hashmap_add(&delta_base_cache, ent);
1301}
1302
1303int packed_object_info(struct packed_git *p, off_t obj_offset,
1304                       struct object_info *oi)
1305{
1306        struct pack_window *w_curs = NULL;
1307        unsigned long size;
1308        off_t curpos = obj_offset;
1309        enum object_type type;
1310
1311        /*
1312         * We always get the representation type, but only convert it to
1313         * a "real" type later if the caller is interested.
1314         */
1315        if (oi->contentp) {
1316                *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
1317                                                      &type);
1318                if (!*oi->contentp)
1319                        type = OBJ_BAD;
1320        } else {
1321                type = unpack_object_header(p, &w_curs, &curpos, &size);
1322        }
1323
1324        if (!oi->contentp && oi->sizep) {
1325                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1326                        off_t tmp_pos = curpos;
1327                        off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1328                                                           type, obj_offset);
1329                        if (!base_offset) {
1330                                type = OBJ_BAD;
1331                                goto out;
1332                        }
1333                        *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1334                        if (*oi->sizep == 0) {
1335                                type = OBJ_BAD;
1336                                goto out;
1337                        }
1338                } else {
1339                        *oi->sizep = size;
1340                }
1341        }
1342
1343        if (oi->disk_sizep) {
1344                struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1345                *oi->disk_sizep = revidx[1].offset - obj_offset;
1346        }
1347
1348        if (oi->typep || oi->typename) {
1349                enum object_type ptot;
1350                ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
1351                                             curpos);
1352                if (oi->typep)
1353                        *oi->typep = ptot;
1354                if (oi->typename) {
1355                        const char *tn = typename(ptot);
1356                        if (tn)
1357                                strbuf_addstr(oi->typename, tn);
1358                }
1359                if (ptot < 0) {
1360                        type = OBJ_BAD;
1361                        goto out;
1362                }
1363        }
1364
1365        if (oi->delta_base_sha1) {
1366                if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1367                        const unsigned char *base;
1368
1369                        base = get_delta_base_sha1(p, &w_curs, curpos,
1370                                                   type, obj_offset);
1371                        if (!base) {
1372                                type = OBJ_BAD;
1373                                goto out;
1374                        }
1375
1376                        hashcpy(oi->delta_base_sha1, base);
1377                } else
1378                        hashclr(oi->delta_base_sha1);
1379        }
1380
1381        oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1382                                                          OI_PACKED;
1383
1384out:
1385        unuse_pack(&w_curs);
1386        return type;
1387}
1388
1389static void *unpack_compressed_entry(struct packed_git *p,
1390                                    struct pack_window **w_curs,
1391                                    off_t curpos,
1392                                    unsigned long size)
1393{
1394        int st;
1395        git_zstream stream;
1396        unsigned char *buffer, *in;
1397
1398        buffer = xmallocz_gently(size);
1399        if (!buffer)
1400                return NULL;
1401        memset(&stream, 0, sizeof(stream));
1402        stream.next_out = buffer;
1403        stream.avail_out = size + 1;
1404
1405        git_inflate_init(&stream);
1406        do {
1407                in = use_pack(p, w_curs, curpos, &stream.avail_in);
1408                stream.next_in = in;
1409                st = git_inflate(&stream, Z_FINISH);
1410                if (!stream.avail_out)
1411                        break; /* the payload is larger than it should be */
1412                curpos += stream.next_in - in;
1413        } while (st == Z_OK || st == Z_BUF_ERROR);
1414        git_inflate_end(&stream);
1415        if ((st != Z_STREAM_END) || stream.total_out != size) {
1416                free(buffer);
1417                return NULL;
1418        }
1419
1420        return buffer;
1421}
1422
1423static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1424{
1425        static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1426        trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1427                         p->pack_name, (uintmax_t)obj_offset);
1428}
1429
1430int do_check_packed_object_crc;
1431
1432#define UNPACK_ENTRY_STACK_PREALLOC 64
1433struct unpack_entry_stack_ent {
1434        off_t obj_offset;
1435        off_t curpos;
1436        unsigned long size;
1437};
1438
1439static void *read_object(const unsigned char *sha1, enum object_type *type,
1440                         unsigned long *size)
1441{
1442        struct object_info oi = OBJECT_INFO_INIT;
1443        void *content;
1444        oi.typep = type;
1445        oi.sizep = size;
1446        oi.contentp = &content;
1447
1448        if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1449                return NULL;
1450        return content;
1451}
1452
1453void *unpack_entry(struct packed_git *p, off_t obj_offset,
1454                   enum object_type *final_type, unsigned long *final_size)
1455{
1456        struct pack_window *w_curs = NULL;
1457        off_t curpos = obj_offset;
1458        void *data = NULL;
1459        unsigned long size;
1460        enum object_type type;
1461        struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1462        struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1463        int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1464        int base_from_cache = 0;
1465
1466        write_pack_access_log(p, obj_offset);
1467
1468        /* PHASE 1: drill down to the innermost base object */
1469        for (;;) {
1470                off_t base_offset;
1471                int i;
1472                struct delta_base_cache_entry *ent;
1473
1474                ent = get_delta_base_cache_entry(p, curpos);
1475                if (ent) {
1476                        type = ent->type;
1477                        data = ent->data;
1478                        size = ent->size;
1479                        detach_delta_base_cache_entry(ent);
1480                        base_from_cache = 1;
1481                        break;
1482                }
1483
1484                if (do_check_packed_object_crc && p->index_version > 1) {
1485                        struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1486                        off_t len = revidx[1].offset - obj_offset;
1487                        if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1488                                const unsigned char *sha1 =
1489                                        nth_packed_object_sha1(p, revidx->nr);
1490                                error("bad packed object CRC for %s",
1491                                      sha1_to_hex(sha1));
1492                                mark_bad_packed_object(p, sha1);
1493                                data = NULL;
1494                                goto out;
1495                        }
1496                }
1497
1498                type = unpack_object_header(p, &w_curs, &curpos, &size);
1499                if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1500                        break;
1501
1502                base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1503                if (!base_offset) {
1504                        error("failed to validate delta base reference "
1505                              "at offset %"PRIuMAX" from %s",
1506                              (uintmax_t)curpos, p->pack_name);
1507                        /* bail to phase 2, in hopes of recovery */
1508                        data = NULL;
1509                        break;
1510                }
1511
1512                /* push object, proceed to base */
1513                if (delta_stack_nr >= delta_stack_alloc
1514                    && delta_stack == small_delta_stack) {
1515                        delta_stack_alloc = alloc_nr(delta_stack_nr);
1516                        ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1517                        memcpy(delta_stack, small_delta_stack,
1518                               sizeof(*delta_stack)*delta_stack_nr);
1519                } else {
1520                        ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1521                }
1522                i = delta_stack_nr++;
1523                delta_stack[i].obj_offset = obj_offset;
1524                delta_stack[i].curpos = curpos;
1525                delta_stack[i].size = size;
1526
1527                curpos = obj_offset = base_offset;
1528        }
1529
1530        /* PHASE 2: handle the base */
1531        switch (type) {
1532        case OBJ_OFS_DELTA:
1533        case OBJ_REF_DELTA:
1534                if (data)
1535                        die("BUG: unpack_entry: left loop at a valid delta");
1536                break;
1537        case OBJ_COMMIT:
1538        case OBJ_TREE:
1539        case OBJ_BLOB:
1540        case OBJ_TAG:
1541                if (!base_from_cache)
1542                        data = unpack_compressed_entry(p, &w_curs, curpos, size);
1543                break;
1544        default:
1545                data = NULL;
1546                error("unknown object type %i at offset %"PRIuMAX" in %s",
1547                      type, (uintmax_t)obj_offset, p->pack_name);
1548        }
1549
1550        /* PHASE 3: apply deltas in order */
1551
1552        /* invariants:
1553         *   'data' holds the base data, or NULL if there was corruption
1554         */
1555        while (delta_stack_nr) {
1556                void *delta_data;
1557                void *base = data;
1558                void *external_base = NULL;
1559                unsigned long delta_size, base_size = size;
1560                int i;
1561
1562                data = NULL;
1563
1564                if (base)
1565                        add_delta_base_cache(p, obj_offset, base, base_size, type);
1566
1567                if (!base) {
1568                        /*
1569                         * We're probably in deep shit, but let's try to fetch
1570                         * the required base anyway from another pack or loose.
1571                         * This is costly but should happen only in the presence
1572                         * of a corrupted pack, and is better than failing outright.
1573                         */
1574                        struct revindex_entry *revidx;
1575                        const unsigned char *base_sha1;
1576                        revidx = find_pack_revindex(p, obj_offset);
1577                        if (revidx) {
1578                                base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1579                                error("failed to read delta base object %s"
1580                                      " at offset %"PRIuMAX" from %s",
1581                                      sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
1582                                      p->pack_name);
1583                                mark_bad_packed_object(p, base_sha1);
1584                                base = read_object(base_sha1, &type, &base_size);
1585                                external_base = base;
1586                        }
1587                }
1588
1589                i = --delta_stack_nr;
1590                obj_offset = delta_stack[i].obj_offset;
1591                curpos = delta_stack[i].curpos;
1592                delta_size = delta_stack[i].size;
1593
1594                if (!base)
1595                        continue;
1596
1597                delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1598
1599                if (!delta_data) {
1600                        error("failed to unpack compressed delta "
1601                              "at offset %"PRIuMAX" from %s",
1602                              (uintmax_t)curpos, p->pack_name);
1603                        data = NULL;
1604                        free(external_base);
1605                        continue;
1606                }
1607
1608                data = patch_delta(base, base_size,
1609                                   delta_data, delta_size,
1610                                   &size);
1611
1612                /*
1613                 * We could not apply the delta; warn the user, but keep going.
1614                 * Our failure will be noticed either in the next iteration of
1615                 * the loop, or if this is the final delta, in the caller when
1616                 * we return NULL. Those code paths will take care of making
1617                 * a more explicit warning and retrying with another copy of
1618                 * the object.
1619                 */
1620                if (!data)
1621                        error("failed to apply delta");
1622
1623                free(delta_data);
1624                free(external_base);
1625        }
1626
1627        if (final_type)
1628                *final_type = type;
1629        if (final_size)
1630                *final_size = size;
1631
1632out:
1633        unuse_pack(&w_curs);
1634
1635        if (delta_stack != small_delta_stack)
1636                free(delta_stack);
1637
1638        return data;
1639}
1640
1641const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1642                                            uint32_t n)
1643{
1644        const unsigned char *index = p->index_data;
1645        if (!index) {
1646                if (open_pack_index(p))
1647                        return NULL;
1648                index = p->index_data;
1649        }
1650        if (n >= p->num_objects)
1651                return NULL;
1652        index += 4 * 256;
1653        if (p->index_version == 1) {
1654                return index + 24 * n + 4;
1655        } else {
1656                index += 8;
1657                return index + 20 * n;
1658        }
1659}
1660
1661const struct object_id *nth_packed_object_oid(struct object_id *oid,
1662                                              struct packed_git *p,
1663                                              uint32_t n)
1664{
1665        const unsigned char *hash = nth_packed_object_sha1(p, n);
1666        if (!hash)
1667                return NULL;
1668        hashcpy(oid->hash, hash);
1669        return oid;
1670}
1671
1672void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1673{
1674        const unsigned char *ptr = vptr;
1675        const unsigned char *start = p->index_data;
1676        const unsigned char *end = start + p->index_size;
1677        if (ptr < start)
1678                die(_("offset before start of pack index for %s (corrupt index?)"),
1679                    p->pack_name);
1680        /* No need to check for underflow; .idx files must be at least 8 bytes */
1681        if (ptr >= end - 8)
1682                die(_("offset beyond end of pack index for %s (truncated index?)"),
1683                    p->pack_name);
1684}
1685
1686off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1687{
1688        const unsigned char *index = p->index_data;
1689        index += 4 * 256;
1690        if (p->index_version == 1) {
1691                return ntohl(*((uint32_t *)(index + 24 * n)));
1692        } else {
1693                uint32_t off;
1694                index += 8 + p->num_objects * (20 + 4);
1695                off = ntohl(*((uint32_t *)(index + 4 * n)));
1696                if (!(off & 0x80000000))
1697                        return off;
1698                index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1699                check_pack_index_ptr(p, index);
1700                return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1701                                   ntohl(*((uint32_t *)(index + 4)));
1702        }
1703}
1704
1705off_t find_pack_entry_one(const unsigned char *sha1,
1706                                  struct packed_git *p)
1707{
1708        const uint32_t *level1_ofs = p->index_data;
1709        const unsigned char *index = p->index_data;
1710        unsigned hi, lo, stride;
1711        static int debug_lookup = -1;
1712
1713        if (debug_lookup < 0)
1714                debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1715
1716        if (!index) {
1717                if (open_pack_index(p))
1718                        return 0;
1719                level1_ofs = p->index_data;
1720                index = p->index_data;
1721        }
1722        if (p->index_version > 1) {
1723                level1_ofs += 2;
1724                index += 8;
1725        }
1726        index += 4 * 256;
1727        hi = ntohl(level1_ofs[*sha1]);
1728        lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1729        if (p->index_version > 1) {
1730                stride = 20;
1731        } else {
1732                stride = 24;
1733                index += 4;
1734        }
1735
1736        if (debug_lookup)
1737                printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1738                       sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1739
1740        while (lo < hi) {
1741                unsigned mi = (lo + hi) / 2;
1742                int cmp = hashcmp(index + mi * stride, sha1);
1743
1744                if (debug_lookup)
1745                        printf("lo %u hi %u rg %u mi %u\n",
1746                               lo, hi, hi - lo, mi);
1747                if (!cmp)
1748                        return nth_packed_object_offset(p, mi);
1749                if (cmp > 0)
1750                        hi = mi;
1751                else
1752                        lo = mi+1;
1753        }
1754        return 0;
1755}
1756
1757int is_pack_valid(struct packed_git *p)
1758{
1759        /* An already open pack is known to be valid. */
1760        if (p->pack_fd != -1)
1761                return 1;
1762
1763        /* If the pack has one window completely covering the
1764         * file size, the pack is known to be valid even if
1765         * the descriptor is not currently open.
1766         */
1767        if (p->windows) {
1768                struct pack_window *w = p->windows;
1769
1770                if (!w->offset && w->len == p->pack_size)
1771                        return 1;
1772        }
1773
1774        /* Force the pack to open to prove its valid. */
1775        return !open_packed_git(p);
1776}
1777
1778struct packed_git *find_sha1_pack(const unsigned char *sha1,
1779                                  struct packed_git *packs)
1780{
1781        struct packed_git *p;
1782
1783        for (p = packs; p; p = p->next) {
1784                if (find_pack_entry_one(sha1, p))
1785                        return p;
1786        }
1787        return NULL;
1788
1789}
1790
1791static int fill_pack_entry(const unsigned char *sha1,
1792                           struct pack_entry *e,
1793                           struct packed_git *p)
1794{
1795        off_t offset;
1796
1797        if (p->num_bad_objects) {
1798                unsigned i;
1799                for (i = 0; i < p->num_bad_objects; i++)
1800                        if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1801                                return 0;
1802        }
1803
1804        offset = find_pack_entry_one(sha1, p);
1805        if (!offset)
1806                return 0;
1807
1808        /*
1809         * We are about to tell the caller where they can locate the
1810         * requested object.  We better make sure the packfile is
1811         * still here and can be accessed before supplying that
1812         * answer, as it may have been deleted since the index was
1813         * loaded!
1814         */
1815        if (!is_pack_valid(p))
1816                return 0;
1817        e->offset = offset;
1818        e->p = p;
1819        hashcpy(e->sha1, sha1);
1820        return 1;
1821}
1822
1823/*
1824 * Iff a pack file contains the object named by sha1, return true and
1825 * store its location to e.
1826 */
1827int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1828{
1829        struct mru_entry *p;
1830
1831        prepare_packed_git();
1832        if (!packed_git)
1833                return 0;
1834
1835        for (p = packed_git_mru->head; p; p = p->next) {
1836                if (fill_pack_entry(sha1, e, p->item)) {
1837                        mru_mark(packed_git_mru, p);
1838                        return 1;
1839                }
1840        }
1841        return 0;
1842}
1843
1844int has_sha1_pack(const unsigned char *sha1)
1845{
1846        struct pack_entry e;
1847        return find_pack_entry(sha1, &e);
1848}
1849
1850int has_pack_index(const unsigned char *sha1)
1851{
1852        struct stat st;
1853        if (stat(sha1_pack_index_name(sha1), &st))
1854                return 0;
1855        return 1;
1856}
1857
1858static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
1859{
1860        uint32_t i;
1861        int r = 0;
1862
1863        for (i = 0; i < p->num_objects; i++) {
1864                struct object_id oid;
1865
1866                if (!nth_packed_object_oid(&oid, p, i))
1867                        return error("unable to get sha1 of object %u in %s",
1868                                     i, p->pack_name);
1869
1870                r = cb(&oid, p, i, data);
1871                if (r)
1872                        break;
1873        }
1874        return r;
1875}
1876
1877int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
1878{
1879        struct packed_git *p;
1880        int r = 0;
1881        int pack_errors = 0;
1882
1883        prepare_packed_git();
1884        for (p = packed_git; p; p = p->next) {
1885                if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
1886                        continue;
1887                if (open_pack_index(p)) {
1888                        pack_errors = 1;
1889                        continue;
1890                }
1891                r = for_each_object_in_pack(p, cb, data);
1892                if (r)
1893                        break;
1894        }
1895        return r ? r : pack_errors;
1896}