git.oblomov.eu Git - git/blob - packfile.c

   1 #include "cache.h"
   2 #include "list.h"
   3 #include "pack.h"
   4 #include "repository.h"
   5 #include "dir.h"
   6 #include "mergesort.h"
   7 #include "packfile.h"
   8 #include "delta.h"
   9 #include "streaming.h"
  10 #include "sha1-lookup.h"
  11 #include "commit.h"
  12 #include "object.h"
  13 #include "tag.h"
  14 #include "tree-walk.h"
  15 #include "tree.h"
  16 #include "object-store.h"
  17 #include "midx.h"
  18 #include "commit-graph.h"
  19 #include "promisor-remote.h"
  20
  21 char *odb_pack_name(struct strbuf *buf,
  22                     const unsigned char *hash,
  23                     const char *ext)
  24 {
  25         strbuf_reset(buf);
  26         strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
  27                     hash_to_hex(hash), ext);
  28         return buf->buf;
  29 }
  30
  31 char *sha1_pack_name(const unsigned char *sha1)
  32 {
  33         static struct strbuf buf = STRBUF_INIT;
  34         return odb_pack_name(&buf, sha1, "pack");
  35 }
  36
  37 char *sha1_pack_index_name(const unsigned char *sha1)
  38 {
  39         static struct strbuf buf = STRBUF_INIT;
  40         return odb_pack_name(&buf, sha1, "idx");
  41 }
  42
  43 static unsigned int pack_used_ctr;
  44 static unsigned int pack_mmap_calls;
  45 static unsigned int peak_pack_open_windows;
  46 static unsigned int pack_open_windows;
  47 static unsigned int pack_open_fds;
  48 static unsigned int pack_max_fds;
  49 static size_t peak_pack_mapped;
  50 static size_t pack_mapped;
  51
  52 #define SZ_FMT PRIuMAX
  53 static inline uintmax_t sz_fmt(size_t s) { return s; }
  54
  55 void pack_report(void)
  56 {
  57         fprintf(stderr,
  58                 "pack_report: getpagesize()            = %10" SZ_FMT "\n"
  59                 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
  60                 "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
  61                 sz_fmt(getpagesize()),
  62                 sz_fmt(packed_git_window_size),
  63                 sz_fmt(packed_git_limit));
  64         fprintf(stderr,
  65                 "pack_report: pack_used_ctr            = %10u\n"
  66                 "pack_report: pack_mmap_calls          = %10u\n"
  67                 "pack_report: pack_open_windows        = %10u / %10u\n"
  68                 "pack_report: pack_mapped              = "
  69                         "%10" SZ_FMT " / %10" SZ_FMT "\n",
  70                 pack_used_ctr,
  71                 pack_mmap_calls,
  72                 pack_open_windows, peak_pack_open_windows,
  73                 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
  74 }
  75
  76 /*
  77  * Open and mmap the index file at path, perform a couple of
  78  * consistency checks, then record its information to p.  Return 0 on
  79  * success.
  80  */
  81 static int check_packed_git_idx(const char *path, struct packed_git *p)
  82 {
  83         void *idx_map;
  84         size_t idx_size;
  85         int fd = git_open(path), ret;
  86         struct stat st;
  87         const unsigned int hashsz = the_hash_algo->rawsz;
  88
  89         if (fd < 0)
  90                 return -1;
  91         if (fstat(fd, &st)) {
  92                 close(fd);
  93                 return -1;
  94         }
  95         idx_size = xsize_t(st.st_size);
  96         if (idx_size < 4 * 256 + hashsz + hashsz) {
  97                 close(fd);
  98                 return error("index file %s is too small", path);
  99         }
 100         idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
 101         close(fd);
 102
 103         ret = load_idx(path, hashsz, idx_map, idx_size, p);
 104
 105         if (ret)
 106                 munmap(idx_map, idx_size);
 107
 108         return ret;
 109 }
 110
 111 int load_idx(const char *path, const unsigned int hashsz, void *idx_map,
 112              size_t idx_size, struct packed_git *p)
 113 {
 114         struct pack_idx_header *hdr = idx_map;
 115         uint32_t version, nr, i, *index;
 116
 117         if (idx_size < 4 * 256 + hashsz + hashsz)
 118                 return error("index file %s is too small", path);
 119         if (idx_map == NULL)
 120                 return error("empty data");
 121
 122         if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
 123                 version = ntohl(hdr->idx_version);
 124                 if (version < 2 || version > 2)
 125                         return error("index file %s is version %"PRIu32
 126                                      " and is not supported by this binary"
 127                                      " (try upgrading GIT to a newer version)",
 128                                      path, version);
 129         } else
 130                 version = 1;
 131
 132         nr = 0;
 133         index = idx_map;
 134         if (version > 1)
 135                 index += 2;  /* skip index header */
 136         for (i = 0; i < 256; i++) {
 137                 uint32_t n = ntohl(index[i]);
 138                 if (n < nr)
 139                         return error("non-monotonic index %s", path);
 140                 nr = n;
 141         }
 142
 143         if (version == 1) {
 144                 /*
 145                  * Total size:
 146                  *  - 256 index entries 4 bytes each
 147                  *  - 24-byte entries * nr (object ID + 4-byte offset)
 148                  *  - hash of the packfile
 149                  *  - file checksum
 150                  */
 151                 if (idx_size != 4 * 256 + nr * (hashsz + 4) + hashsz + hashsz)
 152                         return error("wrong index v1 file size in %s", path);
 153         } else if (version == 2) {
 154                 /*
 155                  * Minimum size:
 156                  *  - 8 bytes of header
 157                  *  - 256 index entries 4 bytes each
 158                  *  - object ID entry * nr
 159                  *  - 4-byte crc entry * nr
 160                  *  - 4-byte offset entry * nr
 161                  *  - hash of the packfile
 162                  *  - file checksum
 163                  * And after the 4-byte offset table might be a
 164                  * variable sized table containing 8-byte entries
 165                  * for offsets larger than 2^31.
 166                  */
 167                 unsigned long min_size = 8 + 4*256 + nr*(hashsz + 4 + 4) + hashsz + hashsz;
 168                 unsigned long max_size = min_size;
 169                 if (nr)
 170                         max_size += (nr - 1)*8;
 171                 if (idx_size < min_size || idx_size > max_size)
 172                         return error("wrong index v2 file size in %s", path);
 173                 if (idx_size != min_size &&
 174                     /*
 175                      * make sure we can deal with large pack offsets.
 176                      * 31-bit signed offset won't be enough, neither
 177                      * 32-bit unsigned one will be.
 178                      */
 179                     (sizeof(off_t) <= 4))
 180                         return error("pack too large for current definition of off_t in %s", path);
 181                 p->crc_offset = 8 + 4 * 256 + nr * hashsz;
 182         }
 183
 184         p->index_version = version;
 185         p->index_data = idx_map;
 186         p->index_size = idx_size;
 187         p->num_objects = nr;
 188         return 0;
 189 }
 190
 191 int open_pack_index(struct packed_git *p)
 192 {
 193         char *idx_name;
 194         size_t len;
 195         int ret;
 196
 197         if (p->index_data)
 198                 return 0;
 199
 200         if (!strip_suffix(p->pack_name, ".pack", &len))
 201                 BUG("pack_name does not end in .pack");
 202         idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
 203         ret = check_packed_git_idx(idx_name, p);
 204         free(idx_name);
 205         return ret;
 206 }
 207
 208 uint32_t get_pack_fanout(struct packed_git *p, uint32_t value)
 209 {
 210         const uint32_t *level1_ofs = p->index_data;
 211
 212         if (!level1_ofs) {
 213                 if (open_pack_index(p))
 214                         return 0;
 215                 level1_ofs = p->index_data;
 216         }
 217
 218         if (p->index_version > 1) {
 219                 level1_ofs += 2;
 220         }
 221
 222         return ntohl(level1_ofs[value]);
 223 }
 224
 225 static struct packed_git *alloc_packed_git(int extra)
 226 {
 227         struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
 228         memset(p, 0, sizeof(*p));
 229         p->pack_fd = -1;
 230         return p;
 231 }
 232
 233 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
 234 {
 235         const char *path = sha1_pack_name(sha1);
 236         size_t alloc = st_add(strlen(path), 1);
 237         struct packed_git *p = alloc_packed_git(alloc);
 238
 239         memcpy(p->pack_name, path, alloc); /* includes NUL */
 240         hashcpy(p->hash, sha1);
 241         if (check_packed_git_idx(idx_path, p)) {
 242                 free(p);
 243                 return NULL;
 244         }
 245
 246         return p;
 247 }
 248
 249 static void scan_windows(struct packed_git *p,
 250         struct packed_git **lru_p,
 251         struct pack_window **lru_w,
 252         struct pack_window **lru_l)
 253 {
 254         struct pack_window *w, *w_l;
 255
 256         for (w_l = NULL, w = p->windows; w; w = w->next) {
 257                 if (!w->inuse_cnt) {
 258                         if (!*lru_w || w->last_used < (*lru_w)->last_used) {
 259                                 *lru_p = p;
 260                                 *lru_w = w;
 261                                 *lru_l = w_l;
 262                         }
 263                 }
 264                 w_l = w;
 265         }
 266 }
 267
 268 static int unuse_one_window(struct packed_git *current)
 269 {
 270         struct packed_git *p, *lru_p = NULL;
 271         struct pack_window *lru_w = NULL, *lru_l = NULL;
 272
 273         if (current)
 274                 scan_windows(current, &lru_p, &lru_w, &lru_l);
 275         for (p = the_repository->objects->packed_git; p; p = p->next)
 276                 scan_windows(p, &lru_p, &lru_w, &lru_l);
 277         if (lru_p) {
 278                 munmap(lru_w->base, lru_w->len);
 279                 pack_mapped -= lru_w->len;
 280                 if (lru_l)
 281                         lru_l->next = lru_w->next;
 282                 else
 283                         lru_p->windows = lru_w->next;
 284                 free(lru_w);
 285                 pack_open_windows--;
 286                 return 1;
 287         }
 288         return 0;
 289 }
 290
 291 void close_pack_windows(struct packed_git *p)
 292 {
 293         while (p->windows) {
 294                 struct pack_window *w = p->windows;
 295
 296                 if (w->inuse_cnt)
 297                         die("pack '%s' still has open windows to it",
 298                             p->pack_name);
 299                 munmap(w->base, w->len);
 300                 pack_mapped -= w->len;
 301                 pack_open_windows--;
 302                 p->windows = w->next;
 303                 free(w);
 304         }
 305 }
 306
 307 int close_pack_fd(struct packed_git *p)
 308 {
 309         if (p->pack_fd < 0)
 310                 return 0;
 311
 312         close(p->pack_fd);
 313         pack_open_fds--;
 314         p->pack_fd = -1;
 315
 316         return 1;
 317 }
 318
 319 void close_pack_index(struct packed_git *p)
 320 {
 321         if (p->index_data) {
 322                 munmap((void *)p->index_data, p->index_size);
 323                 p->index_data = NULL;
 324         }
 325 }
 326
 327 void close_pack(struct packed_git *p)
 328 {
 329         close_pack_windows(p);
 330         close_pack_fd(p);
 331         close_pack_index(p);
 332 }
 333
 334 void close_object_store(struct raw_object_store *o)
 335 {
 336         struct packed_git *p;
 337
 338         for (p = o->packed_git; p; p = p->next)
 339                 if (p->do_not_close)
 340                         BUG("want to close pack marked 'do-not-close'");
 341                 else
 342                         close_pack(p);
 343
 344         if (o->multi_pack_index) {
 345                 close_midx(o->multi_pack_index);
 346                 o->multi_pack_index = NULL;
 347         }
 348
 349         close_commit_graph(o);
 350 }
 351
 352 void unlink_pack_path(const char *pack_name, int force_delete)
 353 {
 354         static const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
 355         int i;
 356         struct strbuf buf = STRBUF_INIT;
 357         size_t plen;
 358
 359         strbuf_addstr(&buf, pack_name);
 360         strip_suffix_mem(buf.buf, &buf.len, ".pack");
 361         plen = buf.len;
 362
 363         if (!force_delete) {
 364                 strbuf_addstr(&buf, ".keep");
 365                 if (!access(buf.buf, F_OK)) {
 366                         strbuf_release(&buf);
 367                         return;
 368                 }
 369         }
 370
 371         for (i = 0; i < ARRAY_SIZE(exts); i++) {
 372                 strbuf_setlen(&buf, plen);
 373                 strbuf_addstr(&buf, exts[i]);
 374                 unlink(buf.buf);
 375         }
 376
 377         strbuf_release(&buf);
 378 }
 379
 380 /*
 381  * The LRU pack is the one with the oldest MRU window, preferring packs
 382  * with no used windows, or the oldest mtime if it has no windows allocated.
 383  */
 384 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
 385 {
 386         struct pack_window *w, *this_mru_w;
 387         int has_windows_inuse = 0;
 388
 389         /*
 390          * Reject this pack if it has windows and the previously selected
 391          * one does not.  If this pack does not have windows, reject
 392          * it if the pack file is newer than the previously selected one.
 393          */
 394         if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
 395                 return;
 396
 397         for (w = this_mru_w = p->windows; w; w = w->next) {
 398                 /*
 399                  * Reject this pack if any of its windows are in use,
 400                  * but the previously selected pack did not have any
 401                  * inuse windows.  Otherwise, record that this pack
 402                  * has windows in use.
 403                  */
 404                 if (w->inuse_cnt) {
 405                         if (*accept_windows_inuse)
 406                                 has_windows_inuse = 1;
 407                         else
 408                                 return;
 409                 }
 410
 411                 if (w->last_used > this_mru_w->last_used)
 412                         this_mru_w = w;
 413
 414                 /*
 415                  * Reject this pack if it has windows that have been
 416                  * used more recently than the previously selected pack.
 417                  * If the previously selected pack had windows inuse and
 418                  * we have not encountered a window in this pack that is
 419                  * inuse, skip this check since we prefer a pack with no
 420                  * inuse windows to one that has inuse windows.
 421                  */
 422                 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
 423                     this_mru_w->last_used > (*mru_w)->last_used)
 424                         return;
 425         }
 426
 427         /*
 428          * Select this pack.
 429          */
 430         *mru_w = this_mru_w;
 431         *lru_p = p;
 432         *accept_windows_inuse = has_windows_inuse;
 433 }
 434
 435 static int close_one_pack(void)
 436 {
 437         struct packed_git *p, *lru_p = NULL;
 438         struct pack_window *mru_w = NULL;
 439         int accept_windows_inuse = 1;
 440
 441         for (p = the_repository->objects->packed_git; p; p = p->next) {
 442                 if (p->pack_fd == -1)
 443                         continue;
 444                 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
 445         }
 446
 447         if (lru_p)
 448                 return close_pack_fd(lru_p);
 449
 450         return 0;
 451 }
 452
 453 static unsigned int get_max_fd_limit(void)
 454 {
 455 #ifdef RLIMIT_NOFILE
 456         {
 457                 struct rlimit lim;
 458
 459                 if (!getrlimit(RLIMIT_NOFILE, &lim))
 460                         return lim.rlim_cur;
 461         }
 462 #endif
 463
 464 #ifdef _SC_OPEN_MAX
 465         {
 466                 long open_max = sysconf(_SC_OPEN_MAX);
 467                 if (0 < open_max)
 468                         return open_max;
 469                 /*
 470                  * Otherwise, we got -1 for one of the two
 471                  * reasons:
 472                  *
 473                  * (1) sysconf() did not understand _SC_OPEN_MAX
 474                  *     and signaled an error with -1; or
 475                  * (2) sysconf() said there is no limit.
 476                  *
 477                  * We _could_ clear errno before calling sysconf() to
 478                  * tell these two cases apart and return a huge number
 479                  * in the latter case to let the caller cap it to a
 480                  * value that is not so selfish, but letting the
 481                  * fallback OPEN_MAX codepath take care of these cases
 482                  * is a lot simpler.
 483                  */
 484         }
 485 #endif
 486
 487 #ifdef OPEN_MAX
 488         return OPEN_MAX;
 489 #else
 490         return 1; /* see the caller ;-) */
 491 #endif
 492 }
 493
 494 const char *pack_basename(struct packed_git *p)
 495 {
 496         const char *ret = strrchr(p->pack_name, '/');
 497         if (ret)
 498                 ret = ret + 1; /* skip past slash */
 499         else
 500                 ret = p->pack_name; /* we only have a base */
 501         return ret;
 502 }
 503
 504 /*
 505  * Do not call this directly as this leaks p->pack_fd on error return;
 506  * call open_packed_git() instead.
 507  */
 508 static int open_packed_git_1(struct packed_git *p)
 509 {
 510         struct stat st;
 511         struct pack_header hdr;
 512         unsigned char hash[GIT_MAX_RAWSZ];
 513         unsigned char *idx_hash;
 514         ssize_t read_result;
 515         const unsigned hashsz = the_hash_algo->rawsz;
 516
 517         if (!p->index_data) {
 518                 struct multi_pack_index *m;
 519                 const char *pack_name = pack_basename(p);
 520
 521                 for (m = the_repository->objects->multi_pack_index;
 522                      m; m = m->next) {
 523                         if (midx_contains_pack(m, pack_name))
 524                                 break;
 525                 }
 526
 527                 if (!m && open_pack_index(p))
 528                         return error("packfile %s index unavailable", p->pack_name);
 529         }
 530
 531         if (!pack_max_fds) {
 532                 unsigned int max_fds = get_max_fd_limit();
 533
 534                 /* Save 3 for stdin/stdout/stderr, 22 for work */
 535                 if (25 < max_fds)
 536                         pack_max_fds = max_fds - 25;
 537                 else
 538                         pack_max_fds = 1;
 539         }
 540
 541         while (pack_max_fds <= pack_open_fds && close_one_pack())
 542                 ; /* nothing */
 543
 544         p->pack_fd = git_open(p->pack_name);
 545         if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
 546                 return -1;
 547         pack_open_fds++;
 548
 549         /* If we created the struct before we had the pack we lack size. */
 550         if (!p->pack_size) {
 551                 if (!S_ISREG(st.st_mode))
 552                         return error("packfile %s not a regular file", p->pack_name);
 553                 p->pack_size = st.st_size;
 554         } else if (p->pack_size != st.st_size)
 555                 return error("packfile %s size changed", p->pack_name);
 556
 557         /* Verify we recognize this pack file format. */
 558         read_result = read_in_full(p->pack_fd, &hdr, sizeof(hdr));
 559         if (read_result < 0)
 560                 return error_errno("error reading from %s", p->pack_name);
 561         if (read_result != sizeof(hdr))
 562                 return error("file %s is far too short to be a packfile", p->pack_name);
 563         if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
 564                 return error("file %s is not a GIT packfile", p->pack_name);
 565         if (!pack_version_ok(hdr.hdr_version))
 566                 return error("packfile %s is version %"PRIu32" and not"
 567                         " supported (try upgrading GIT to a newer version)",
 568                         p->pack_name, ntohl(hdr.hdr_version));
 569
 570         /* Skip index checking if in multi-pack-index */
 571         if (!p->index_data)
 572                 return 0;
 573
 574         /* Verify the pack matches its index. */
 575         if (p->num_objects != ntohl(hdr.hdr_entries))
 576                 return error("packfile %s claims to have %"PRIu32" objects"
 577                              " while index indicates %"PRIu32" objects",
 578                              p->pack_name, ntohl(hdr.hdr_entries),
 579                              p->num_objects);
 580         read_result = pread_in_full(p->pack_fd, hash, hashsz,
 581                                         p->pack_size - hashsz);
 582         if (read_result < 0)
 583                 return error_errno("error reading from %s", p->pack_name);
 584         if (read_result != hashsz)
 585                 return error("packfile %s signature is unavailable", p->pack_name);
 586         idx_hash = ((unsigned char *)p->index_data) + p->index_size - hashsz * 2;
 587         if (!hasheq(hash, idx_hash))
 588                 return error("packfile %s does not match index", p->pack_name);
 589         return 0;
 590 }
 591
 592 static int open_packed_git(struct packed_git *p)
 593 {
 594         if (!open_packed_git_1(p))
 595                 return 0;
 596         close_pack_fd(p);
 597         return -1;
 598 }
 599
 600 static int in_window(struct pack_window *win, off_t offset)
 601 {
 602         /* We must promise at least one full hash after the
 603          * offset is available from this window, otherwise the offset
 604          * is not actually in this window and a different window (which
 605          * has that one hash excess) must be used.  This is to support
 606          * the object header and delta base parsing routines below.
 607          */
 608         off_t win_off = win->offset;
 609         return win_off <= offset
 610                 && (offset + the_hash_algo->rawsz) <= (win_off + win->len);
 611 }
 612
 613 unsigned char *use_pack(struct packed_git *p,
 614                 struct pack_window **w_cursor,
 615                 off_t offset,
 616                 unsigned long *left)
 617 {
 618         struct pack_window *win = *w_cursor;
 619
 620         /* Since packfiles end in a hash of their content and it's
 621          * pointless to ask for an offset into the middle of that
 622          * hash, and the in_window function above wouldn't match
 623          * don't allow an offset too close to the end of the file.
 624          */
 625         if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
 626                 die("packfile %s cannot be accessed", p->pack_name);
 627         if (offset > (p->pack_size - the_hash_algo->rawsz))
 628                 die("offset beyond end of packfile (truncated pack?)");
 629         if (offset < 0)
 630                 die(_("offset before end of packfile (broken .idx?)"));
 631
 632         if (!win || !in_window(win, offset)) {
 633                 if (win)
 634                         win->inuse_cnt--;
 635                 for (win = p->windows; win; win = win->next) {
 636                         if (in_window(win, offset))
 637                                 break;
 638                 }
 639                 if (!win) {
 640                         size_t window_align = packed_git_window_size / 2;
 641                         off_t len;
 642
 643                         if (p->pack_fd == -1 && open_packed_git(p))
 644                                 die("packfile %s cannot be accessed", p->pack_name);
 645
 646                         win = xcalloc(1, sizeof(*win));
 647                         win->offset = (offset / window_align) * window_align;
 648                         len = p->pack_size - win->offset;
 649                         if (len > packed_git_window_size)
 650                                 len = packed_git_window_size;
 651                         win->len = (size_t)len;
 652                         pack_mapped += win->len;
 653                         while (packed_git_limit < pack_mapped
 654                                 && unuse_one_window(p))
 655                                 ; /* nothing */
 656                         win->base = xmmap_gently(NULL, win->len,
 657                                 PROT_READ, MAP_PRIVATE,
 658                                 p->pack_fd, win->offset);
 659                         if (win->base == MAP_FAILED)
 660                                 die_errno("packfile %s cannot be mapped",
 661                                           p->pack_name);
 662                         if (!win->offset && win->len == p->pack_size
 663                                 && !p->do_not_close)
 664                                 close_pack_fd(p);
 665                         pack_mmap_calls++;
 666                         pack_open_windows++;
 667                         if (pack_mapped > peak_pack_mapped)
 668                                 peak_pack_mapped = pack_mapped;
 669                         if (pack_open_windows > peak_pack_open_windows)
 670                                 peak_pack_open_windows = pack_open_windows;
 671                         win->next = p->windows;
 672                         p->windows = win;
 673                 }
 674         }
 675         if (win != *w_cursor) {
 676                 win->last_used = pack_used_ctr++;
 677                 win->inuse_cnt++;
 678                 *w_cursor = win;
 679         }
 680         offset -= win->offset;
 681         if (left)
 682                 *left = win->len - xsize_t(offset);
 683         return win->base + offset;
 684 }
 685
 686 void unuse_pack(struct pack_window **w_cursor)
 687 {
 688         struct pack_window *w = *w_cursor;
 689         if (w) {
 690                 w->inuse_cnt--;
 691                 *w_cursor = NULL;
 692         }
 693 }
 694
 695 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
 696 {
 697         struct stat st;
 698         size_t alloc;
 699         struct packed_git *p;
 700
 701         /*
 702          * Make sure a corresponding .pack file exists and that
 703          * the index looks sane.
 704          */
 705         if (!strip_suffix_mem(path, &path_len, ".idx"))
 706                 return NULL;
 707
 708         /*
 709          * ".promisor" is long enough to hold any suffix we're adding (and
 710          * the use xsnprintf double-checks that)
 711          */
 712         alloc = st_add3(path_len, strlen(".promisor"), 1);
 713         p = alloc_packed_git(alloc);
 714         memcpy(p->pack_name, path, path_len);
 715
 716         xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
 717         if (!access(p->pack_name, F_OK))
 718                 p->pack_keep = 1;
 719
 720         xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
 721         if (!access(p->pack_name, F_OK))
 722                 p->pack_promisor = 1;
 723
 724         xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
 725         if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
 726                 free(p);
 727                 return NULL;
 728         }
 729
 730         /* ok, it looks sane as far as we can check without
 731          * actually mapping the pack file.
 732          */
 733         p->pack_size = st.st_size;
 734         p->pack_local = local;
 735         p->mtime = st.st_mtime;
 736         if (path_len < the_hash_algo->hexsz ||
 737             get_sha1_hex(path + path_len - the_hash_algo->hexsz, p->hash))
 738                 hashclr(p->hash);
 739         return p;
 740 }
 741
 742 void install_packed_git(struct repository *r, struct packed_git *pack)
 743 {
 744         if (pack->pack_fd != -1)
 745                 pack_open_fds++;
 746
 747         pack->next = r->objects->packed_git;
 748         r->objects->packed_git = pack;
 749
 750         hashmap_entry_init(&pack->packmap_ent, strhash(pack->pack_name));
 751         hashmap_add(&r->objects->pack_map, &pack->packmap_ent);
 752 }
 753
 754 void (*report_garbage)(unsigned seen_bits, const char *path);
 755
 756 static void report_helper(const struct string_list *list,
 757                           int seen_bits, int first, int last)
 758 {
 759         if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
 760                 return;
 761
 762         for (; first < last; first++)
 763                 report_garbage(seen_bits, list->items[first].string);
 764 }
 765
 766 static void report_pack_garbage(struct string_list *list)
 767 {
 768         int i, baselen = -1, first = 0, seen_bits = 0;
 769
 770         if (!report_garbage)
 771                 return;
 772
 773         string_list_sort(list);
 774
 775         for (i = 0; i < list->nr; i++) {
 776                 const char *path = list->items[i].string;
 777                 if (baselen != -1 &&
 778                     strncmp(path, list->items[first].string, baselen)) {
 779                         report_helper(list, seen_bits, first, i);
 780                         baselen = -1;
 781                         seen_bits = 0;
 782                 }
 783                 if (baselen == -1) {
 784                         const char *dot = strrchr(path, '.');
 785                         if (!dot) {
 786                                 report_garbage(PACKDIR_FILE_GARBAGE, path);
 787                                 continue;
 788                         }
 789                         baselen = dot - path + 1;
 790                         first = i;
 791                 }
 792                 if (!strcmp(path + baselen, "pack"))
 793                         seen_bits |= 1;
 794                 else if (!strcmp(path + baselen, "idx"))
 795                         seen_bits |= 2;
 796         }
 797         report_helper(list, seen_bits, first, list->nr);
 798 }
 799
 800 void for_each_file_in_pack_dir(const char *objdir,
 801                                each_file_in_pack_dir_fn fn,
 802                                void *data)
 803 {
 804         struct strbuf path = STRBUF_INIT;
 805         size_t dirnamelen;
 806         DIR *dir;
 807         struct dirent *de;
 808
 809         strbuf_addstr(&path, objdir);
 810         strbuf_addstr(&path, "/pack");
 811         dir = opendir(path.buf);
 812         if (!dir) {
 813                 if (errno != ENOENT)
 814                         error_errno("unable to open object pack directory: %s",
 815                                     path.buf);
 816                 strbuf_release(&path);
 817                 return;
 818         }
 819         strbuf_addch(&path, '/');
 820         dirnamelen = path.len;
 821         while ((de = readdir(dir)) != NULL) {
 822                 if (is_dot_or_dotdot(de->d_name))
 823                         continue;
 824
 825                 strbuf_setlen(&path, dirnamelen);
 826                 strbuf_addstr(&path, de->d_name);
 827
 828                 fn(path.buf, path.len, de->d_name, data);
 829         }
 830
 831         closedir(dir);
 832         strbuf_release(&path);
 833 }
 834
 835 struct prepare_pack_data {
 836         struct repository *r;
 837         struct string_list *garbage;
 838         int local;
 839         struct multi_pack_index *m;
 840 };
 841
 842 static void prepare_pack(const char *full_name, size_t full_name_len,
 843                          const char *file_name, void *_data)
 844 {
 845         struct prepare_pack_data *data = (struct prepare_pack_data *)_data;
 846         struct packed_git *p;
 847         size_t base_len = full_name_len;
 848
 849         if (strip_suffix_mem(full_name, &base_len, ".idx") &&
 850             !(data->m && midx_contains_pack(data->m, file_name))) {
 851                 struct hashmap_entry hent;
 852                 char *pack_name = xstrfmt("%.*s.pack", (int)base_len, full_name);
 853                 unsigned int hash = strhash(pack_name);
 854                 hashmap_entry_init(&hent, hash);
 855
 856                 /* Don't reopen a pack we already have. */
 857                 if (!hashmap_get(&data->r->objects->pack_map, &hent, pack_name)) {
 858                         p = add_packed_git(full_name, full_name_len, data->local);
 859                         if (p)
 860                                 install_packed_git(data->r, p);
 861                 }
 862                 free(pack_name);
 863         }
 864
 865         if (!report_garbage)
 866                 return;
 867
 868         if (!strcmp(file_name, "multi-pack-index"))
 869                 return;
 870         if (ends_with(file_name, ".idx") ||
 871             ends_with(file_name, ".pack") ||
 872             ends_with(file_name, ".bitmap") ||
 873             ends_with(file_name, ".keep") ||
 874             ends_with(file_name, ".promisor"))
 875                 string_list_append(data->garbage, full_name);
 876         else
 877                 report_garbage(PACKDIR_FILE_GARBAGE, full_name);
 878 }
 879
 880 static void prepare_packed_git_one(struct repository *r, char *objdir, int local)
 881 {
 882         struct prepare_pack_data data;
 883         struct string_list garbage = STRING_LIST_INIT_DUP;
 884
 885         data.m = r->objects->multi_pack_index;
 886
 887         /* look for the multi-pack-index for this object directory */
 888         while (data.m && strcmp(data.m->object_dir, objdir))
 889                 data.m = data.m->next;
 890
 891         data.r = r;
 892         data.garbage = &garbage;
 893         data.local = local;
 894
 895         for_each_file_in_pack_dir(objdir, prepare_pack, &data);
 896
 897         report_pack_garbage(data.garbage);
 898         string_list_clear(data.garbage, 0);
 899 }
 900
 901 static void prepare_packed_git(struct repository *r);
 902 /*
 903  * Give a fast, rough count of the number of objects in the repository. This
 904  * ignores loose objects completely. If you have a lot of them, then either
 905  * you should repack because your performance will be awful, or they are
 906  * all unreachable objects about to be pruned, in which case they're not really
 907  * interesting as a measure of repo size in the first place.
 908  */
 909 unsigned long repo_approximate_object_count(struct repository *r)
 910 {
 911         if (!r->objects->approximate_object_count_valid) {
 912                 unsigned long count;
 913                 struct multi_pack_index *m;
 914                 struct packed_git *p;
 915
 916                 prepare_packed_git(r);
 917                 count = 0;
 918                 for (m = get_multi_pack_index(r); m; m = m->next)
 919                         count += m->num_objects;
 920                 for (p = r->objects->packed_git; p; p = p->next) {
 921                         if (open_pack_index(p))
 922                                 continue;
 923                         count += p->num_objects;
 924                 }
 925                 r->objects->approximate_object_count = count;
 926                 r->objects->approximate_object_count_valid = 1;
 927         }
 928         return r->objects->approximate_object_count;
 929 }
 930
 931 static void *get_next_packed_git(const void *p)
 932 {
 933         return ((const struct packed_git *)p)->next;
 934 }
 935
 936 static void set_next_packed_git(void *p, void *next)
 937 {
 938         ((struct packed_git *)p)->next = next;
 939 }
 940
 941 static int sort_pack(const void *a_, const void *b_)
 942 {
 943         const struct packed_git *a = a_;
 944         const struct packed_git *b = b_;
 945         int st;
 946
 947         /*
 948          * Local packs tend to contain objects specific to our
 949          * variant of the project than remote ones.  In addition,
 950          * remote ones could be on a network mounted filesystem.
 951          * Favor local ones for these reasons.
 952          */
 953         st = a->pack_local - b->pack_local;
 954         if (st)
 955                 return -st;
 956
 957         /*
 958          * Younger packs tend to contain more recent objects,
 959          * and more recent objects tend to get accessed more
 960          * often.
 961          */
 962         if (a->mtime < b->mtime)
 963                 return 1;
 964         else if (a->mtime == b->mtime)
 965                 return 0;
 966         return -1;
 967 }
 968
 969 static void rearrange_packed_git(struct repository *r)
 970 {
 971         r->objects->packed_git = llist_mergesort(
 972                 r->objects->packed_git, get_next_packed_git,
 973                 set_next_packed_git, sort_pack);
 974 }
 975
 976 static void prepare_packed_git_mru(struct repository *r)
 977 {
 978         struct packed_git *p;
 979
 980         INIT_LIST_HEAD(&r->objects->packed_git_mru);
 981
 982         for (p = r->objects->packed_git; p; p = p->next)
 983                 list_add_tail(&p->mru, &r->objects->packed_git_mru);
 984 }
 985
 986 static void prepare_packed_git(struct repository *r)
 987 {
 988         struct object_directory *odb;
 989
 990         if (r->objects->packed_git_initialized)
 991                 return;
 992
 993         prepare_alt_odb(r);
 994         for (odb = r->objects->odb; odb; odb = odb->next) {
 995                 int local = (odb == r->objects->odb);
 996                 prepare_multi_pack_index_one(r, odb->path, local);
 997                 prepare_packed_git_one(r, odb->path, local);
 998         }
 999         rearrange_packed_git(r);
1000
1001         prepare_packed_git_mru(r);
1002         r->objects->packed_git_initialized = 1;
1003 }
1004
1005 void reprepare_packed_git(struct repository *r)
1006 {
1007         struct object_directory *odb;
1008
1009         obj_read_lock();
1010         for (odb = r->objects->odb; odb; odb = odb->next)
1011                 odb_clear_loose_cache(odb);
1012
1013         r->objects->approximate_object_count_valid = 0;
1014         r->objects->packed_git_initialized = 0;
1015         prepare_packed_git(r);
1016         obj_read_unlock();
1017 }
1018
1019 struct packed_git *get_packed_git(struct repository *r)
1020 {
1021         prepare_packed_git(r);
1022         return r->objects->packed_git;
1023 }
1024
1025 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1026 {
1027         prepare_packed_git(r);
1028         return r->objects->multi_pack_index;
1029 }
1030
1031 struct multi_pack_index *get_local_multi_pack_index(struct repository *r)
1032 {
1033         struct multi_pack_index *m = get_multi_pack_index(r);
1034
1035         /* no need to iterate; we always put the local one first (if any) */
1036         if (m && m->local)
1037                 return m;
1038
1039         return NULL;
1040 }
1041
1042 struct packed_git *get_all_packs(struct repository *r)
1043 {
1044         struct multi_pack_index *m;
1045
1046         prepare_packed_git(r);
1047         for (m = r->objects->multi_pack_index; m; m = m->next) {
1048                 uint32_t i;
1049                 for (i = 0; i < m->num_packs; i++)
1050                         prepare_midx_pack(r, m, i);
1051         }
1052
1053         return r->objects->packed_git;
1054 }
1055
1056 struct list_head *get_packed_git_mru(struct repository *r)
1057 {
1058         prepare_packed_git(r);
1059         return &r->objects->packed_git_mru;
1060 }
1061
1062 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1063                 unsigned long len, enum object_type *type, unsigned long *sizep)
1064 {
1065         unsigned shift;
1066         unsigned long size, c;
1067         unsigned long used = 0;
1068
1069         c = buf[used++];
1070         *type = (c >> 4) & 7;
1071         size = c & 15;
1072         shift = 4;
1073         while (c & 0x80) {
1074                 if (len <= used || bitsizeof(long) <= shift) {
1075                         error("bad object header");
1076                         size = used = 0;
1077                         break;
1078                 }
1079                 c = buf[used++];
1080                 size += (c & 0x7f) << shift;
1081                 shift += 7;
1082         }
1083         *sizep = size;
1084         return used;
1085 }
1086
1087 unsigned long get_size_from_delta(struct packed_git *p,
1088                                   struct pack_window **w_curs,
1089                                   off_t curpos)
1090 {
1091         const unsigned char *data;
1092         unsigned char delta_head[20], *in;
1093         git_zstream stream;
1094         int st;
1095
1096         memset(&stream, 0, sizeof(stream));
1097         stream.next_out = delta_head;
1098         stream.avail_out = sizeof(delta_head);
1099
1100         git_inflate_init(&stream);
1101         do {
1102                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1103                 stream.next_in = in;
1104                 /*
1105                  * Note: the window section returned by use_pack() must be
1106                  * available throughout git_inflate()'s unlocked execution. To
1107                  * ensure no other thread will modify the window in the
1108                  * meantime, we rely on the packed_window.inuse_cnt. This
1109                  * counter is incremented before window reading and checked
1110                  * before window disposal.
1111                  *
1112                  * Other worrying sections could be the call to close_pack_fd(),
1113                  * which can close packs even with in-use windows, and to
1114                  * reprepare_packed_git(). Regarding the former, mmap doc says:
1115                  * "closing the file descriptor does not unmap the region". And
1116                  * for the latter, it won't re-open already available packs.
1117                  */
1118                 obj_read_unlock();
1119                 st = git_inflate(&stream, Z_FINISH);
1120                 obj_read_lock();
1121                 curpos += stream.next_in - in;
1122         } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1123                  stream.total_out < sizeof(delta_head));
1124         git_inflate_end(&stream);
1125         if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1126                 error("delta data unpack-initial failed");
1127                 return 0;
1128         }
1129
1130         /* Examine the initial part of the delta to figure out
1131          * the result size.
1132          */
1133         data = delta_head;
1134
1135         /* ignore base size */
1136         get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1137
1138         /* Read the result size */
1139         return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1140 }
1141
1142 int unpack_object_header(struct packed_git *p,
1143                          struct pack_window **w_curs,
1144                          off_t *curpos,
1145                          unsigned long *sizep)
1146 {
1147         unsigned char *base;
1148         unsigned long left;
1149         unsigned long used;
1150         enum object_type type;
1151
1152         /* use_pack() assures us we have [base, base + 20) available
1153          * as a range that we can look at.  (Its actually the hash
1154          * size that is assured.)  With our object header encoding
1155          * the maximum deflated object size is 2^137, which is just
1156          * insane, so we know won't exceed what we have been given.
1157          */
1158         base = use_pack(p, w_curs, *curpos, &left);
1159         used = unpack_object_header_buffer(base, left, &type, sizep);
1160         if (!used) {
1161                 type = OBJ_BAD;
1162         } else
1163                 *curpos += used;
1164
1165         return type;
1166 }
1167
1168 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1169 {
1170         unsigned i;
1171         const unsigned hashsz = the_hash_algo->rawsz;
1172         for (i = 0; i < p->num_bad_objects; i++)
1173                 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1174                         return;
1175         p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1176                                       st_mult(GIT_MAX_RAWSZ,
1177                                               st_add(p->num_bad_objects, 1)));
1178         hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1179         p->num_bad_objects++;
1180 }
1181
1182 const struct packed_git *has_packed_and_bad(struct repository *r,
1183                                             const unsigned char *sha1)
1184 {
1185         struct packed_git *p;
1186         unsigned i;
1187
1188         for (p = r->objects->packed_git; p; p = p->next)
1189                 for (i = 0; i < p->num_bad_objects; i++)
1190                         if (hasheq(sha1,
1191                                    p->bad_object_sha1 + the_hash_algo->rawsz * i))
1192                                 return p;
1193         return NULL;
1194 }
1195
1196 off_t get_delta_base(struct packed_git *p,
1197                      struct pack_window **w_curs,
1198                      off_t *curpos,
1199                      enum object_type type,
1200                      off_t delta_obj_offset)
1201 {
1202         unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1203         off_t base_offset;
1204
1205         /* use_pack() assured us we have [base_info, base_info + 20)
1206          * as a range that we can look at without walking off the
1207          * end of the mapped window.  Its actually the hash size
1208          * that is assured.  An OFS_DELTA longer than the hash size
1209          * is stupid, as then a REF_DELTA would be smaller to store.
1210          */
1211         if (type == OBJ_OFS_DELTA) {
1212                 unsigned used = 0;
1213                 unsigned char c = base_info[used++];
1214                 base_offset = c & 127;
1215                 while (c & 128) {
1216                         base_offset += 1;
1217                         if (!base_offset || MSB(base_offset, 7))
1218                                 return 0;  /* overflow */
1219                         c = base_info[used++];
1220                         base_offset = (base_offset << 7) + (c & 127);
1221                 }
1222                 base_offset = delta_obj_offset - base_offset;
1223                 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1224                         return 0;  /* out of bound */
1225                 *curpos += used;
1226         } else if (type == OBJ_REF_DELTA) {
1227                 /* The base entry _must_ be in the same pack */
1228                 base_offset = find_pack_entry_one(base_info, p);
1229                 *curpos += the_hash_algo->rawsz;
1230         } else
1231                 die("I am totally screwed");
1232         return base_offset;
1233 }
1234
1235 /*
1236  * Like get_delta_base above, but we return the sha1 instead of the pack
1237  * offset. This means it is cheaper for REF deltas (we do not have to do
1238  * the final object lookup), but more expensive for OFS deltas (we
1239  * have to load the revidx to convert the offset back into a sha1).
1240  */
1241 static int get_delta_base_oid(struct packed_git *p,
1242                               struct pack_window **w_curs,
1243                               off_t curpos,
1244                               struct object_id *oid,
1245                               enum object_type type,
1246                               off_t delta_obj_offset)
1247 {
1248         if (type == OBJ_REF_DELTA) {
1249                 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1250                 oidread(oid, base);
1251                 return 0;
1252         } else if (type == OBJ_OFS_DELTA) {
1253                 struct revindex_entry *revidx;
1254                 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1255                                                    type, delta_obj_offset);
1256
1257                 if (!base_offset)
1258                         return -1;
1259
1260                 revidx = find_pack_revindex(p, base_offset);
1261                 if (!revidx)
1262                         return -1;
1263
1264                 return nth_packed_object_id(oid, p, revidx->nr);
1265         } else
1266                 return -1;
1267 }
1268
1269 static int retry_bad_packed_offset(struct repository *r,
1270                                    struct packed_git *p,
1271                                    off_t obj_offset)
1272 {
1273         int type;
1274         struct revindex_entry *revidx;
1275         struct object_id oid;
1276         revidx = find_pack_revindex(p, obj_offset);
1277         if (!revidx)
1278                 return OBJ_BAD;
1279         nth_packed_object_id(&oid, p, revidx->nr);
1280         mark_bad_packed_object(p, oid.hash);
1281         type = oid_object_info(r, &oid, NULL);
1282         if (type <= OBJ_NONE)
1283                 return OBJ_BAD;
1284         return type;
1285 }
1286
1287 #define POI_STACK_PREALLOC 64
1288
1289 static enum object_type packed_to_object_type(struct repository *r,
1290                                               struct packed_git *p,
1291                                               off_t obj_offset,
1292                                               enum object_type type,
1293                                               struct pack_window **w_curs,
1294                                               off_t curpos)
1295 {
1296         off_t small_poi_stack[POI_STACK_PREALLOC];
1297         off_t *poi_stack = small_poi_stack;
1298         int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1299
1300         while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1301                 off_t base_offset;
1302                 unsigned long size;
1303                 /* Push the object we're going to leave behind */
1304                 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1305                         poi_stack_alloc = alloc_nr(poi_stack_nr);
1306                         ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1307                         COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1308                 } else {
1309                         ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1310                 }
1311                 poi_stack[poi_stack_nr++] = obj_offset;
1312                 /* If parsing the base offset fails, just unwind */
1313                 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1314                 if (!base_offset)
1315                         goto unwind;
1316                 curpos = obj_offset = base_offset;
1317                 type = unpack_object_header(p, w_curs, &curpos, &size);
1318                 if (type <= OBJ_NONE) {
1319                         /* If getting the base itself fails, we first
1320                          * retry the base, otherwise unwind */
1321                         type = retry_bad_packed_offset(r, p, base_offset);
1322                         if (type > OBJ_NONE)
1323                                 goto out;
1324                         goto unwind;
1325                 }
1326         }
1327
1328         switch (type) {
1329         case OBJ_BAD:
1330         case OBJ_COMMIT:
1331         case OBJ_TREE:
1332         case OBJ_BLOB:
1333         case OBJ_TAG:
1334                 break;
1335         default:
1336                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1337                       type, (uintmax_t)obj_offset, p->pack_name);
1338                 type = OBJ_BAD;
1339         }
1340
1341 out:
1342         if (poi_stack != small_poi_stack)
1343                 free(poi_stack);
1344         return type;
1345
1346 unwind:
1347         while (poi_stack_nr) {
1348                 obj_offset = poi_stack[--poi_stack_nr];
1349                 type = retry_bad_packed_offset(r, p, obj_offset);
1350                 if (type > OBJ_NONE)
1351                         goto out;
1352         }
1353         type = OBJ_BAD;
1354         goto out;
1355 }
1356
1357 static struct hashmap delta_base_cache;
1358 static size_t delta_base_cached;
1359
1360 static LIST_HEAD(delta_base_cache_lru);
1361
1362 struct delta_base_cache_key {
1363         struct packed_git *p;
1364         off_t base_offset;
1365 };
1366
1367 struct delta_base_cache_entry {
1368         struct hashmap_entry ent;
1369         struct delta_base_cache_key key;
1370         struct list_head lru;
1371         void *data;
1372         unsigned long size;
1373         enum object_type type;
1374 };
1375
1376 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1377 {
1378         unsigned int hash;
1379
1380         hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1381         hash += (hash >> 8) + (hash >> 16);
1382         return hash;
1383 }
1384
1385 static struct delta_base_cache_entry *
1386 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1387 {
1388         struct hashmap_entry entry, *e;
1389         struct delta_base_cache_key key;
1390
1391         if (!delta_base_cache.cmpfn)
1392                 return NULL;
1393
1394         hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1395         key.p = p;
1396         key.base_offset = base_offset;
1397         e = hashmap_get(&delta_base_cache, &entry, &key);
1398         return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1399 }
1400
1401 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1402                                    const struct delta_base_cache_key *b)
1403 {
1404         return a->p == b->p && a->base_offset == b->base_offset;
1405 }
1406
1407 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1408                                      const struct hashmap_entry *va,
1409                                      const struct hashmap_entry *vb,
1410                                      const void *vkey)
1411 {
1412         const struct delta_base_cache_entry *a, *b;
1413         const struct delta_base_cache_key *key = vkey;
1414
1415         a = container_of(va, const struct delta_base_cache_entry, ent);
1416         b = container_of(vb, const struct delta_base_cache_entry, ent);
1417
1418         if (key)
1419                 return !delta_base_cache_key_eq(&a->key, key);
1420         else
1421                 return !delta_base_cache_key_eq(&a->key, &b->key);
1422 }
1423
1424 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1425 {
1426         return !!get_delta_base_cache_entry(p, base_offset);
1427 }
1428
1429 /*
1430  * Remove the entry from the cache, but do _not_ free the associated
1431  * entry data. The caller takes ownership of the "data" buffer, and
1432  * should copy out any fields it wants before detaching.
1433  */
1434 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1435 {
1436         hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1437         list_del(&ent->lru);
1438         delta_base_cached -= ent->size;
1439         free(ent);
1440 }
1441
1442 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1443                                    off_t base_offset, unsigned long *base_size,
1444                                    enum object_type *type)
1445 {
1446         struct delta_base_cache_entry *ent;
1447
1448         ent = get_delta_base_cache_entry(p, base_offset);
1449         if (!ent)
1450                 return unpack_entry(r, p, base_offset, type, base_size);
1451
1452         if (type)
1453                 *type = ent->type;
1454         if (base_size)
1455                 *base_size = ent->size;
1456         return xmemdupz(ent->data, ent->size);
1457 }
1458
1459 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1460 {
1461         free(ent->data);
1462         detach_delta_base_cache_entry(ent);
1463 }
1464
1465 void clear_delta_base_cache(void)
1466 {
1467         struct list_head *lru, *tmp;
1468         list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1469                 struct delta_base_cache_entry *entry =
1470                         list_entry(lru, struct delta_base_cache_entry, lru);
1471                 release_delta_base_cache(entry);
1472         }
1473 }
1474
1475 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1476         void *base, unsigned long base_size, enum object_type type)
1477 {
1478         struct delta_base_cache_entry *ent;
1479         struct list_head *lru, *tmp;
1480
1481         /*
1482          * Check required to avoid redundant entries when more than one thread
1483          * is unpacking the same object, in unpack_entry() (since its phases I
1484          * and III might run concurrently across multiple threads).
1485          */
1486         if (in_delta_base_cache(p, base_offset)) {
1487                 free(base);
1488                 return;
1489         }
1490
1491         delta_base_cached += base_size;
1492
1493         list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1494                 struct delta_base_cache_entry *f =
1495                         list_entry(lru, struct delta_base_cache_entry, lru);
1496                 if (delta_base_cached <= delta_base_cache_limit)
1497                         break;
1498                 release_delta_base_cache(f);
1499         }
1500
1501         ent = xmalloc(sizeof(*ent));
1502         ent->key.p = p;
1503         ent->key.base_offset = base_offset;
1504         ent->type = type;
1505         ent->data = base;
1506         ent->size = base_size;
1507         list_add_tail(&ent->lru, &delta_base_cache_lru);
1508
1509         if (!delta_base_cache.cmpfn)
1510                 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1511         hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1512         hashmap_add(&delta_base_cache, &ent->ent);
1513 }
1514
1515 int packed_object_info(struct repository *r, struct packed_git *p,
1516                        off_t obj_offset, struct object_info *oi)
1517 {
1518         struct pack_window *w_curs = NULL;
1519         unsigned long size;
1520         off_t curpos = obj_offset;
1521         enum object_type type;
1522
1523         /*
1524          * We always get the representation type, but only convert it to
1525          * a "real" type later if the caller is interested.
1526          */
1527         if (oi->contentp) {
1528                 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1529                                                       &type);
1530                 if (!*oi->contentp)
1531                         type = OBJ_BAD;
1532         } else {
1533                 type = unpack_object_header(p, &w_curs, &curpos, &size);
1534         }
1535
1536         if (!oi->contentp && oi->sizep) {
1537                 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1538                         off_t tmp_pos = curpos;
1539                         off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1540                                                            type, obj_offset);
1541                         if (!base_offset) {
1542                                 type = OBJ_BAD;
1543                                 goto out;
1544                         }
1545                         *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1546                         if (*oi->sizep == 0) {
1547                                 type = OBJ_BAD;
1548                                 goto out;
1549                         }
1550                 } else {
1551                         *oi->sizep = size;
1552                 }
1553         }
1554
1555         if (oi->disk_sizep) {
1556                 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1557                 *oi->disk_sizep = revidx[1].offset - obj_offset;
1558         }
1559
1560         if (oi->typep || oi->type_name) {
1561                 enum object_type ptot;
1562                 ptot = packed_to_object_type(r, p, obj_offset,
1563                                              type, &w_curs, curpos);
1564                 if (oi->typep)
1565                         *oi->typep = ptot;
1566                 if (oi->type_name) {
1567                         const char *tn = type_name(ptot);
1568                         if (tn)
1569                                 strbuf_addstr(oi->type_name, tn);
1570                 }
1571                 if (ptot < 0) {
1572                         type = OBJ_BAD;
1573                         goto out;
1574                 }
1575         }
1576
1577         if (oi->delta_base_oid) {
1578                 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1579                         if (get_delta_base_oid(p, &w_curs, curpos,
1580                                                oi->delta_base_oid,
1581                                                type, obj_offset) < 0) {
1582                                 type = OBJ_BAD;
1583                                 goto out;
1584                         }
1585                 } else
1586                         oidclr(oi->delta_base_oid);
1587         }
1588
1589         oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1590                                                           OI_PACKED;
1591
1592 out:
1593         unuse_pack(&w_curs);
1594         return type;
1595 }
1596
1597 static void *unpack_compressed_entry(struct packed_git *p,
1598                                     struct pack_window **w_curs,
1599                                     off_t curpos,
1600                                     unsigned long size)
1601 {
1602         int st;
1603         git_zstream stream;
1604         unsigned char *buffer, *in;
1605
1606         buffer = xmallocz_gently(size);
1607         if (!buffer)
1608                 return NULL;
1609         memset(&stream, 0, sizeof(stream));
1610         stream.next_out = buffer;
1611         stream.avail_out = size + 1;
1612
1613         git_inflate_init(&stream);
1614         do {
1615                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1616                 stream.next_in = in;
1617                 /*
1618                  * Note: we must ensure the window section returned by
1619                  * use_pack() will be available throughout git_inflate()'s
1620                  * unlocked execution. Please refer to the comment at
1621                  * get_size_from_delta() to see how this is done.
1622                  */
1623                 obj_read_unlock();
1624                 st = git_inflate(&stream, Z_FINISH);
1625                 obj_read_lock();
1626                 if (!stream.avail_out)
1627                         break; /* the payload is larger than it should be */
1628                 curpos += stream.next_in - in;
1629         } while (st == Z_OK || st == Z_BUF_ERROR);
1630         git_inflate_end(&stream);
1631         if ((st != Z_STREAM_END) || stream.total_out != size) {
1632                 free(buffer);
1633                 return NULL;
1634         }
1635
1636         /* versions of zlib can clobber unconsumed portion of outbuf */
1637         buffer[size] = '\0';
1638
1639         return buffer;
1640 }
1641
1642 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1643 {
1644         static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1645         trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1646                          p->pack_name, (uintmax_t)obj_offset);
1647 }
1648
1649 int do_check_packed_object_crc;
1650
1651 #define UNPACK_ENTRY_STACK_PREALLOC 64
1652 struct unpack_entry_stack_ent {
1653         off_t obj_offset;
1654         off_t curpos;
1655         unsigned long size;
1656 };
1657
1658 static void *read_object(struct repository *r,
1659                          const struct object_id *oid,
1660                          enum object_type *type,
1661                          unsigned long *size)
1662 {
1663         struct object_info oi = OBJECT_INFO_INIT;
1664         void *content;
1665         oi.typep = type;
1666         oi.sizep = size;
1667         oi.contentp = &content;
1668
1669         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1670                 return NULL;
1671         return content;
1672 }
1673
1674 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1675                    enum object_type *final_type, unsigned long *final_size)
1676 {
1677         struct pack_window *w_curs = NULL;
1678         off_t curpos = obj_offset;
1679         void *data = NULL;
1680         unsigned long size;
1681         enum object_type type;
1682         struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1683         struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1684         int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1685         int base_from_cache = 0;
1686
1687         write_pack_access_log(p, obj_offset);
1688
1689         /* PHASE 1: drill down to the innermost base object */
1690         for (;;) {
1691                 off_t base_offset;
1692                 int i;
1693                 struct delta_base_cache_entry *ent;
1694
1695                 ent = get_delta_base_cache_entry(p, curpos);
1696                 if (ent) {
1697                         type = ent->type;
1698                         data = ent->data;
1699                         size = ent->size;
1700                         detach_delta_base_cache_entry(ent);
1701                         base_from_cache = 1;
1702                         break;
1703                 }
1704
1705                 if (do_check_packed_object_crc && p->index_version > 1) {
1706                         struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1707                         off_t len = revidx[1].offset - obj_offset;
1708                         if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1709                                 struct object_id oid;
1710                                 nth_packed_object_id(&oid, p, revidx->nr);
1711                                 error("bad packed object CRC for %s",
1712                                       oid_to_hex(&oid));
1713                                 mark_bad_packed_object(p, oid.hash);
1714                                 data = NULL;
1715                                 goto out;
1716                         }
1717                 }
1718
1719                 type = unpack_object_header(p, &w_curs, &curpos, &size);
1720                 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1721                         break;
1722
1723                 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1724                 if (!base_offset) {
1725                         error("failed to validate delta base reference "
1726                               "at offset %"PRIuMAX" from %s",
1727                               (uintmax_t)curpos, p->pack_name);
1728                         /* bail to phase 2, in hopes of recovery */
1729                         data = NULL;
1730                         break;
1731                 }
1732
1733                 /* push object, proceed to base */
1734                 if (delta_stack_nr >= delta_stack_alloc
1735                     && delta_stack == small_delta_stack) {
1736                         delta_stack_alloc = alloc_nr(delta_stack_nr);
1737                         ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1738                         COPY_ARRAY(delta_stack, small_delta_stack,
1739                                    delta_stack_nr);
1740                 } else {
1741                         ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1742                 }
1743                 i = delta_stack_nr++;
1744                 delta_stack[i].obj_offset = obj_offset;
1745                 delta_stack[i].curpos = curpos;
1746                 delta_stack[i].size = size;
1747
1748                 curpos = obj_offset = base_offset;
1749         }
1750
1751         /* PHASE 2: handle the base */
1752         switch (type) {
1753         case OBJ_OFS_DELTA:
1754         case OBJ_REF_DELTA:
1755                 if (data)
1756                         BUG("unpack_entry: left loop at a valid delta");
1757                 break;
1758         case OBJ_COMMIT:
1759         case OBJ_TREE:
1760         case OBJ_BLOB:
1761         case OBJ_TAG:
1762                 if (!base_from_cache)
1763                         data = unpack_compressed_entry(p, &w_curs, curpos, size);
1764                 break;
1765         default:
1766                 data = NULL;
1767                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1768                       type, (uintmax_t)obj_offset, p->pack_name);
1769         }
1770
1771         /* PHASE 3: apply deltas in order */
1772
1773         /* invariants:
1774          *   'data' holds the base data, or NULL if there was corruption
1775          */
1776         while (delta_stack_nr) {
1777                 void *delta_data;
1778                 void *base = data;
1779                 void *external_base = NULL;
1780                 unsigned long delta_size, base_size = size;
1781                 int i;
1782                 off_t base_obj_offset = obj_offset;
1783
1784                 data = NULL;
1785
1786                 if (!base) {
1787                         /*
1788                          * We're probably in deep shit, but let's try to fetch
1789                          * the required base anyway from another pack or loose.
1790                          * This is costly but should happen only in the presence
1791                          * of a corrupted pack, and is better than failing outright.
1792                          */
1793                         struct revindex_entry *revidx;
1794                         struct object_id base_oid;
1795                         revidx = find_pack_revindex(p, obj_offset);
1796                         if (revidx) {
1797                                 nth_packed_object_id(&base_oid, p, revidx->nr);
1798                                 error("failed to read delta base object %s"
1799                                       " at offset %"PRIuMAX" from %s",
1800                                       oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1801                                       p->pack_name);
1802                                 mark_bad_packed_object(p, base_oid.hash);
1803                                 base = read_object(r, &base_oid, &type, &base_size);
1804                                 external_base = base;
1805                         }
1806                 }
1807
1808                 i = --delta_stack_nr;
1809                 obj_offset = delta_stack[i].obj_offset;
1810                 curpos = delta_stack[i].curpos;
1811                 delta_size = delta_stack[i].size;
1812
1813                 if (!base)
1814                         continue;
1815
1816                 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1817
1818                 if (!delta_data) {
1819                         error("failed to unpack compressed delta "
1820                               "at offset %"PRIuMAX" from %s",
1821                               (uintmax_t)curpos, p->pack_name);
1822                         data = NULL;
1823                 } else {
1824                         data = patch_delta(base, base_size, delta_data,
1825                                            delta_size, &size);
1826
1827                         /*
1828                          * We could not apply the delta; warn the user, but
1829                          * keep going. Our failure will be noticed either in
1830                          * the next iteration of the loop, or if this is the
1831                          * final delta, in the caller when we return NULL.
1832                          * Those code paths will take care of making a more
1833                          * explicit warning and retrying with another copy of
1834                          * the object.
1835                          */
1836                         if (!data)
1837                                 error("failed to apply delta");
1838                 }
1839
1840                 /*
1841                  * We delay adding `base` to the cache until the end of the loop
1842                  * because unpack_compressed_entry() momentarily releases the
1843                  * obj_read_mutex, giving another thread the chance to access
1844                  * the cache. Therefore, if `base` was already there, this other
1845                  * thread could free() it (e.g. to make space for another entry)
1846                  * before we are done using it.
1847                  */
1848                 if (!external_base)
1849                         add_delta_base_cache(p, base_obj_offset, base, base_size, type);
1850
1851                 free(delta_data);
1852                 free(external_base);
1853         }
1854
1855         if (final_type)
1856                 *final_type = type;
1857         if (final_size)
1858                 *final_size = size;
1859
1860 out:
1861         unuse_pack(&w_curs);
1862
1863         if (delta_stack != small_delta_stack)
1864                 free(delta_stack);
1865
1866         return data;
1867 }
1868
1869 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1870 {
1871         const unsigned char *index_fanout = p->index_data;
1872         const unsigned char *index_lookup;
1873         const unsigned int hashsz = the_hash_algo->rawsz;
1874         int index_lookup_width;
1875
1876         if (!index_fanout)
1877                 BUG("bsearch_pack called without a valid pack-index");
1878
1879         index_lookup = index_fanout + 4 * 256;
1880         if (p->index_version == 1) {
1881                 index_lookup_width = hashsz + 4;
1882                 index_lookup += 4;
1883         } else {
1884                 index_lookup_width = hashsz;
1885                 index_fanout += 8;
1886                 index_lookup += 8;
1887         }
1888
1889         return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1890                             index_lookup, index_lookup_width, result);
1891 }
1892
1893 int nth_packed_object_id(struct object_id *oid,
1894                          struct packed_git *p,
1895                          uint32_t n)
1896 {
1897         const unsigned char *index = p->index_data;
1898         const unsigned int hashsz = the_hash_algo->rawsz;
1899         if (!index) {
1900                 if (open_pack_index(p))
1901                         return -1;
1902                 index = p->index_data;
1903         }
1904         if (n >= p->num_objects)
1905                 return -1;
1906         index += 4 * 256;
1907         if (p->index_version == 1) {
1908                 oidread(oid, index + (hashsz + 4) * n + 4);
1909         } else {
1910                 index += 8;
1911                 oidread(oid, index + hashsz * n);
1912         }
1913         return 0;
1914 }
1915
1916 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1917 {
1918         const unsigned char *ptr = vptr;
1919         const unsigned char *start = p->index_data;
1920         const unsigned char *end = start + p->index_size;
1921         if (ptr < start)
1922                 die(_("offset before start of pack index for %s (corrupt index?)"),
1923                     p->pack_name);
1924         /* No need to check for underflow; .idx files must be at least 8 bytes */
1925         if (ptr >= end - 8)
1926                 die(_("offset beyond end of pack index for %s (truncated index?)"),
1927                     p->pack_name);
1928 }
1929
1930 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1931 {
1932         const unsigned char *index = p->index_data;
1933         const unsigned int hashsz = the_hash_algo->rawsz;
1934         index += 4 * 256;
1935         if (p->index_version == 1) {
1936                 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1937         } else {
1938                 uint32_t off;
1939                 index += 8 + p->num_objects * (hashsz + 4);
1940                 off = ntohl(*((uint32_t *)(index + 4 * n)));
1941                 if (!(off & 0x80000000))
1942                         return off;
1943                 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1944                 check_pack_index_ptr(p, index);
1945                 return get_be64(index);
1946         }
1947 }
1948
1949 off_t find_pack_entry_one(const unsigned char *sha1,
1950                                   struct packed_git *p)
1951 {
1952         const unsigned char *index = p->index_data;
1953         struct object_id oid;
1954         uint32_t result;
1955
1956         if (!index) {
1957                 if (open_pack_index(p))
1958                         return 0;
1959         }
1960
1961         hashcpy(oid.hash, sha1);
1962         if (bsearch_pack(&oid, p, &result))
1963                 return nth_packed_object_offset(p, result);
1964         return 0;
1965 }
1966
1967 int is_pack_valid(struct packed_git *p)
1968 {
1969         /* An already open pack is known to be valid. */
1970         if (p->pack_fd != -1)
1971                 return 1;
1972
1973         /* If the pack has one window completely covering the
1974          * file size, the pack is known to be valid even if
1975          * the descriptor is not currently open.
1976          */
1977         if (p->windows) {
1978                 struct pack_window *w = p->windows;
1979
1980                 if (!w->offset && w->len == p->pack_size)
1981                         return 1;
1982         }
1983
1984         /* Force the pack to open to prove its valid. */
1985         return !open_packed_git(p);
1986 }
1987
1988 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1989                                   struct packed_git *packs)
1990 {
1991         struct packed_git *p;
1992
1993         for (p = packs; p; p = p->next) {
1994                 if (find_pack_entry_one(sha1, p))
1995                         return p;
1996         }
1997         return NULL;
1998
1999 }
2000
2001 static int fill_pack_entry(const struct object_id *oid,
2002                            struct pack_entry *e,
2003                            struct packed_git *p)
2004 {
2005         off_t offset;
2006
2007         if (p->num_bad_objects) {
2008                 unsigned i;
2009                 for (i = 0; i < p->num_bad_objects; i++)
2010                         if (hasheq(oid->hash,
2011                                    p->bad_object_sha1 + the_hash_algo->rawsz * i))
2012                                 return 0;
2013         }
2014
2015         offset = find_pack_entry_one(oid->hash, p);
2016         if (!offset)
2017                 return 0;
2018
2019         /*
2020          * We are about to tell the caller where they can locate the
2021          * requested object.  We better make sure the packfile is
2022          * still here and can be accessed before supplying that
2023          * answer, as it may have been deleted since the index was
2024          * loaded!
2025          */
2026         if (!is_pack_valid(p))
2027                 return 0;
2028         e->offset = offset;
2029         e->p = p;
2030         return 1;
2031 }
2032
2033 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2034 {
2035         struct list_head *pos;
2036         struct multi_pack_index *m;
2037
2038         prepare_packed_git(r);
2039         if (!r->objects->packed_git && !r->objects->multi_pack_index)
2040                 return 0;
2041
2042         for (m = r->objects->multi_pack_index; m; m = m->next) {
2043                 if (fill_midx_entry(r, oid, e, m))
2044                         return 1;
2045         }
2046
2047         list_for_each(pos, &r->objects->packed_git_mru) {
2048                 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2049                 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2050                         list_move(&p->mru, &r->objects->packed_git_mru);
2051                         return 1;
2052                 }
2053         }
2054         return 0;
2055 }
2056
2057 int has_object_pack(const struct object_id *oid)
2058 {
2059         struct pack_entry e;
2060         return find_pack_entry(the_repository, oid, &e);
2061 }
2062
2063 int has_pack_index(const unsigned char *sha1)
2064 {
2065         struct stat st;
2066         if (stat(sha1_pack_index_name(sha1), &st))
2067                 return 0;
2068         return 1;
2069 }
2070
2071 int for_each_object_in_pack(struct packed_git *p,
2072                             each_packed_object_fn cb, void *data,
2073                             enum for_each_object_flags flags)
2074 {
2075         uint32_t i;
2076         int r = 0;
2077
2078         if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2079                 if (load_pack_revindex(p))
2080                         return -1;
2081         }
2082
2083         for (i = 0; i < p->num_objects; i++) {
2084                 uint32_t pos;
2085                 struct object_id oid;
2086
2087                 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2088                         pos = p->revindex[i].nr;
2089                 else
2090                         pos = i;
2091
2092                 if (nth_packed_object_id(&oid, p, pos) < 0)
2093                         return error("unable to get sha1 of object %u in %s",
2094                                      pos, p->pack_name);
2095
2096                 r = cb(&oid, p, pos, data);
2097                 if (r)
2098                         break;
2099         }
2100         return r;
2101 }
2102
2103 int for_each_packed_object(each_packed_object_fn cb, void *data,
2104                            enum for_each_object_flags flags)
2105 {
2106         struct packed_git *p;
2107         int r = 0;
2108         int pack_errors = 0;
2109
2110         prepare_packed_git(the_repository);
2111         for (p = get_all_packs(the_repository); p; p = p->next) {
2112                 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2113                         continue;
2114                 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2115                     !p->pack_promisor)
2116                         continue;
2117                 if (open_pack_index(p)) {
2118                         pack_errors = 1;
2119                         continue;
2120                 }
2121                 r = for_each_object_in_pack(p, cb, data, flags);
2122                 if (r)
2123                         break;
2124         }
2125         return r ? r : pack_errors;
2126 }
2127
2128 static int add_promisor_object(const struct object_id *oid,
2129                                struct packed_git *pack,
2130                                uint32_t pos,
2131                                void *set_)
2132 {
2133         struct oidset *set = set_;
2134         struct object *obj = parse_object(the_repository, oid);
2135         if (!obj)
2136                 return 1;
2137
2138         oidset_insert(set, oid);
2139
2140         /*
2141          * If this is a tree, commit, or tag, the objects it refers
2142          * to are also promisor objects. (Blobs refer to no objects->)
2143          */
2144         if (obj->type == OBJ_TREE) {
2145                 struct tree *tree = (struct tree *)obj;
2146                 struct tree_desc desc;
2147                 struct name_entry entry;
2148                 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2149                         /*
2150                          * Error messages are given when packs are
2151                          * verified, so do not print any here.
2152                          */
2153                         return 0;
2154                 while (tree_entry_gently(&desc, &entry))
2155                         oidset_insert(set, &entry.oid);
2156         } else if (obj->type == OBJ_COMMIT) {
2157                 struct commit *commit = (struct commit *) obj;
2158                 struct commit_list *parents = commit->parents;
2159
2160                 oidset_insert(set, get_commit_tree_oid(commit));
2161                 for (; parents; parents = parents->next)
2162                         oidset_insert(set, &parents->item->object.oid);
2163         } else if (obj->type == OBJ_TAG) {
2164                 struct tag *tag = (struct tag *) obj;
2165                 oidset_insert(set, get_tagged_oid(tag));
2166         }
2167         return 0;
2168 }
2169
2170 int is_promisor_object(const struct object_id *oid)
2171 {
2172         static struct oidset promisor_objects;
2173         static int promisor_objects_prepared;
2174
2175         if (!promisor_objects_prepared) {
2176                 if (has_promisor_remote()) {
2177                         for_each_packed_object(add_promisor_object,
2178                                                &promisor_objects,
2179                                                FOR_EACH_OBJECT_PROMISOR_ONLY);
2180                 }
2181                 promisor_objects_prepared = 1;
2182         }
2183         return oidset_contains(&promisor_objects, oid);
2184 }