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         }
 927         return r->objects->approximate_object_count;
 928 }
 929
 930 static void *get_next_packed_git(const void *p)
 931 {
 932         return ((const struct packed_git *)p)->next;
 933 }
 934
 935 static void set_next_packed_git(void *p, void *next)
 936 {
 937         ((struct packed_git *)p)->next = next;
 938 }
 939
 940 static int sort_pack(const void *a_, const void *b_)
 941 {
 942         const struct packed_git *a = a_;
 943         const struct packed_git *b = b_;
 944         int st;
 945
 946         /*
 947          * Local packs tend to contain objects specific to our
 948          * variant of the project than remote ones.  In addition,
 949          * remote ones could be on a network mounted filesystem.
 950          * Favor local ones for these reasons.
 951          */
 952         st = a->pack_local - b->pack_local;
 953         if (st)
 954                 return -st;
 955
 956         /*
 957          * Younger packs tend to contain more recent objects,
 958          * and more recent objects tend to get accessed more
 959          * often.
 960          */
 961         if (a->mtime < b->mtime)
 962                 return 1;
 963         else if (a->mtime == b->mtime)
 964                 return 0;
 965         return -1;
 966 }
 967
 968 static void rearrange_packed_git(struct repository *r)
 969 {
 970         r->objects->packed_git = llist_mergesort(
 971                 r->objects->packed_git, get_next_packed_git,
 972                 set_next_packed_git, sort_pack);
 973 }
 974
 975 static void prepare_packed_git_mru(struct repository *r)
 976 {
 977         struct packed_git *p;
 978
 979         INIT_LIST_HEAD(&r->objects->packed_git_mru);
 980
 981         for (p = r->objects->packed_git; p; p = p->next)
 982                 list_add_tail(&p->mru, &r->objects->packed_git_mru);
 983 }
 984
 985 static void prepare_packed_git(struct repository *r)
 986 {
 987         struct object_directory *odb;
 988
 989         if (r->objects->packed_git_initialized)
 990                 return;
 991
 992         prepare_alt_odb(r);
 993         for (odb = r->objects->odb; odb; odb = odb->next) {
 994                 int local = (odb == r->objects->odb);
 995                 prepare_multi_pack_index_one(r, odb->path, local);
 996                 prepare_packed_git_one(r, odb->path, local);
 997         }
 998         rearrange_packed_git(r);
 999
1000         prepare_packed_git_mru(r);
1001         r->objects->packed_git_initialized = 1;
1002 }
1003
1004 void reprepare_packed_git(struct repository *r)
1005 {
1006         struct object_directory *odb;
1007
1008         obj_read_lock();
1009         for (odb = r->objects->odb; odb; odb = odb->next)
1010                 odb_clear_loose_cache(odb);
1011
1012         r->objects->approximate_object_count_valid = 0;
1013         r->objects->packed_git_initialized = 0;
1014         prepare_packed_git(r);
1015         obj_read_unlock();
1016 }
1017
1018 struct packed_git *get_packed_git(struct repository *r)
1019 {
1020         prepare_packed_git(r);
1021         return r->objects->packed_git;
1022 }
1023
1024 struct multi_pack_index *get_multi_pack_index(struct repository *r)
1025 {
1026         prepare_packed_git(r);
1027         return r->objects->multi_pack_index;
1028 }
1029
1030 struct multi_pack_index *get_local_multi_pack_index(struct repository *r)
1031 {
1032         struct multi_pack_index *m = get_multi_pack_index(r);
1033
1034         /* no need to iterate; we always put the local one first (if any) */
1035         if (m && m->local)
1036                 return m;
1037
1038         return NULL;
1039 }
1040
1041 struct packed_git *get_all_packs(struct repository *r)
1042 {
1043         struct multi_pack_index *m;
1044
1045         prepare_packed_git(r);
1046         for (m = r->objects->multi_pack_index; m; m = m->next) {
1047                 uint32_t i;
1048                 for (i = 0; i < m->num_packs; i++)
1049                         prepare_midx_pack(r, m, i);
1050         }
1051
1052         return r->objects->packed_git;
1053 }
1054
1055 struct list_head *get_packed_git_mru(struct repository *r)
1056 {
1057         prepare_packed_git(r);
1058         return &r->objects->packed_git_mru;
1059 }
1060
1061 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1062                 unsigned long len, enum object_type *type, unsigned long *sizep)
1063 {
1064         unsigned shift;
1065         unsigned long size, c;
1066         unsigned long used = 0;
1067
1068         c = buf[used++];
1069         *type = (c >> 4) & 7;
1070         size = c & 15;
1071         shift = 4;
1072         while (c & 0x80) {
1073                 if (len <= used || bitsizeof(long) <= shift) {
1074                         error("bad object header");
1075                         size = used = 0;
1076                         break;
1077                 }
1078                 c = buf[used++];
1079                 size += (c & 0x7f) << shift;
1080                 shift += 7;
1081         }
1082         *sizep = size;
1083         return used;
1084 }
1085
1086 unsigned long get_size_from_delta(struct packed_git *p,
1087                                   struct pack_window **w_curs,
1088                                   off_t curpos)
1089 {
1090         const unsigned char *data;
1091         unsigned char delta_head[20], *in;
1092         git_zstream stream;
1093         int st;
1094
1095         memset(&stream, 0, sizeof(stream));
1096         stream.next_out = delta_head;
1097         stream.avail_out = sizeof(delta_head);
1098
1099         git_inflate_init(&stream);
1100         do {
1101                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1102                 stream.next_in = in;
1103                 /*
1104                  * Note: the window section returned by use_pack() must be
1105                  * available throughout git_inflate()'s unlocked execution. To
1106                  * ensure no other thread will modify the window in the
1107                  * meantime, we rely on the packed_window.inuse_cnt. This
1108                  * counter is incremented before window reading and checked
1109                  * before window disposal.
1110                  *
1111                  * Other worrying sections could be the call to close_pack_fd(),
1112                  * which can close packs even with in-use windows, and to
1113                  * reprepare_packed_git(). Regarding the former, mmap doc says:
1114                  * "closing the file descriptor does not unmap the region". And
1115                  * for the latter, it won't re-open already available packs.
1116                  */
1117                 obj_read_unlock();
1118                 st = git_inflate(&stream, Z_FINISH);
1119                 obj_read_lock();
1120                 curpos += stream.next_in - in;
1121         } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1122                  stream.total_out < sizeof(delta_head));
1123         git_inflate_end(&stream);
1124         if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1125                 error("delta data unpack-initial failed");
1126                 return 0;
1127         }
1128
1129         /* Examine the initial part of the delta to figure out
1130          * the result size.
1131          */
1132         data = delta_head;
1133
1134         /* ignore base size */
1135         get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1136
1137         /* Read the result size */
1138         return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1139 }
1140
1141 int unpack_object_header(struct packed_git *p,
1142                          struct pack_window **w_curs,
1143                          off_t *curpos,
1144                          unsigned long *sizep)
1145 {
1146         unsigned char *base;
1147         unsigned long left;
1148         unsigned long used;
1149         enum object_type type;
1150
1151         /* use_pack() assures us we have [base, base + 20) available
1152          * as a range that we can look at.  (Its actually the hash
1153          * size that is assured.)  With our object header encoding
1154          * the maximum deflated object size is 2^137, which is just
1155          * insane, so we know won't exceed what we have been given.
1156          */
1157         base = use_pack(p, w_curs, *curpos, &left);
1158         used = unpack_object_header_buffer(base, left, &type, sizep);
1159         if (!used) {
1160                 type = OBJ_BAD;
1161         } else
1162                 *curpos += used;
1163
1164         return type;
1165 }
1166
1167 void mark_bad_packed_object(struct packed_git *p, const unsigned char *sha1)
1168 {
1169         unsigned i;
1170         const unsigned hashsz = the_hash_algo->rawsz;
1171         for (i = 0; i < p->num_bad_objects; i++)
1172                 if (hasheq(sha1, p->bad_object_sha1 + hashsz * i))
1173                         return;
1174         p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1175                                       st_mult(GIT_MAX_RAWSZ,
1176                                               st_add(p->num_bad_objects, 1)));
1177         hashcpy(p->bad_object_sha1 + hashsz * p->num_bad_objects, sha1);
1178         p->num_bad_objects++;
1179 }
1180
1181 const struct packed_git *has_packed_and_bad(struct repository *r,
1182                                             const unsigned char *sha1)
1183 {
1184         struct packed_git *p;
1185         unsigned i;
1186
1187         for (p = r->objects->packed_git; p; p = p->next)
1188                 for (i = 0; i < p->num_bad_objects; i++)
1189                         if (hasheq(sha1,
1190                                    p->bad_object_sha1 + the_hash_algo->rawsz * i))
1191                                 return p;
1192         return NULL;
1193 }
1194
1195 off_t get_delta_base(struct packed_git *p,
1196                      struct pack_window **w_curs,
1197                      off_t *curpos,
1198                      enum object_type type,
1199                      off_t delta_obj_offset)
1200 {
1201         unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1202         off_t base_offset;
1203
1204         /* use_pack() assured us we have [base_info, base_info + 20)
1205          * as a range that we can look at without walking off the
1206          * end of the mapped window.  Its actually the hash size
1207          * that is assured.  An OFS_DELTA longer than the hash size
1208          * is stupid, as then a REF_DELTA would be smaller to store.
1209          */
1210         if (type == OBJ_OFS_DELTA) {
1211                 unsigned used = 0;
1212                 unsigned char c = base_info[used++];
1213                 base_offset = c & 127;
1214                 while (c & 128) {
1215                         base_offset += 1;
1216                         if (!base_offset || MSB(base_offset, 7))
1217                                 return 0;  /* overflow */
1218                         c = base_info[used++];
1219                         base_offset = (base_offset << 7) + (c & 127);
1220                 }
1221                 base_offset = delta_obj_offset - base_offset;
1222                 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1223                         return 0;  /* out of bound */
1224                 *curpos += used;
1225         } else if (type == OBJ_REF_DELTA) {
1226                 /* The base entry _must_ be in the same pack */
1227                 base_offset = find_pack_entry_one(base_info, p);
1228                 *curpos += the_hash_algo->rawsz;
1229         } else
1230                 die("I am totally screwed");
1231         return base_offset;
1232 }
1233
1234 /*
1235  * Like get_delta_base above, but we return the sha1 instead of the pack
1236  * offset. This means it is cheaper for REF deltas (we do not have to do
1237  * the final object lookup), but more expensive for OFS deltas (we
1238  * have to load the revidx to convert the offset back into a sha1).
1239  */
1240 static int get_delta_base_oid(struct packed_git *p,
1241                               struct pack_window **w_curs,
1242                               off_t curpos,
1243                               struct object_id *oid,
1244                               enum object_type type,
1245                               off_t delta_obj_offset)
1246 {
1247         if (type == OBJ_REF_DELTA) {
1248                 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1249                 oidread(oid, base);
1250                 return 0;
1251         } else if (type == OBJ_OFS_DELTA) {
1252                 struct revindex_entry *revidx;
1253                 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1254                                                    type, delta_obj_offset);
1255
1256                 if (!base_offset)
1257                         return -1;
1258
1259                 revidx = find_pack_revindex(p, base_offset);
1260                 if (!revidx)
1261                         return -1;
1262
1263                 return nth_packed_object_id(oid, p, revidx->nr);
1264         } else
1265                 return -1;
1266 }
1267
1268 static int retry_bad_packed_offset(struct repository *r,
1269                                    struct packed_git *p,
1270                                    off_t obj_offset)
1271 {
1272         int type;
1273         struct revindex_entry *revidx;
1274         struct object_id oid;
1275         revidx = find_pack_revindex(p, obj_offset);
1276         if (!revidx)
1277                 return OBJ_BAD;
1278         nth_packed_object_id(&oid, p, revidx->nr);
1279         mark_bad_packed_object(p, oid.hash);
1280         type = oid_object_info(r, &oid, NULL);
1281         if (type <= OBJ_NONE)
1282                 return OBJ_BAD;
1283         return type;
1284 }
1285
1286 #define POI_STACK_PREALLOC 64
1287
1288 static enum object_type packed_to_object_type(struct repository *r,
1289                                               struct packed_git *p,
1290                                               off_t obj_offset,
1291                                               enum object_type type,
1292                                               struct pack_window **w_curs,
1293                                               off_t curpos)
1294 {
1295         off_t small_poi_stack[POI_STACK_PREALLOC];
1296         off_t *poi_stack = small_poi_stack;
1297         int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
1298
1299         while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1300                 off_t base_offset;
1301                 unsigned long size;
1302                 /* Push the object we're going to leave behind */
1303                 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1304                         poi_stack_alloc = alloc_nr(poi_stack_nr);
1305                         ALLOC_ARRAY(poi_stack, poi_stack_alloc);
1306                         COPY_ARRAY(poi_stack, small_poi_stack, poi_stack_nr);
1307                 } else {
1308                         ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1309                 }
1310                 poi_stack[poi_stack_nr++] = obj_offset;
1311                 /* If parsing the base offset fails, just unwind */
1312                 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1313                 if (!base_offset)
1314                         goto unwind;
1315                 curpos = obj_offset = base_offset;
1316                 type = unpack_object_header(p, w_curs, &curpos, &size);
1317                 if (type <= OBJ_NONE) {
1318                         /* If getting the base itself fails, we first
1319                          * retry the base, otherwise unwind */
1320                         type = retry_bad_packed_offset(r, p, base_offset);
1321                         if (type > OBJ_NONE)
1322                                 goto out;
1323                         goto unwind;
1324                 }
1325         }
1326
1327         switch (type) {
1328         case OBJ_BAD:
1329         case OBJ_COMMIT:
1330         case OBJ_TREE:
1331         case OBJ_BLOB:
1332         case OBJ_TAG:
1333                 break;
1334         default:
1335                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1336                       type, (uintmax_t)obj_offset, p->pack_name);
1337                 type = OBJ_BAD;
1338         }
1339
1340 out:
1341         if (poi_stack != small_poi_stack)
1342                 free(poi_stack);
1343         return type;
1344
1345 unwind:
1346         while (poi_stack_nr) {
1347                 obj_offset = poi_stack[--poi_stack_nr];
1348                 type = retry_bad_packed_offset(r, p, obj_offset);
1349                 if (type > OBJ_NONE)
1350                         goto out;
1351         }
1352         type = OBJ_BAD;
1353         goto out;
1354 }
1355
1356 static struct hashmap delta_base_cache;
1357 static size_t delta_base_cached;
1358
1359 static LIST_HEAD(delta_base_cache_lru);
1360
1361 struct delta_base_cache_key {
1362         struct packed_git *p;
1363         off_t base_offset;
1364 };
1365
1366 struct delta_base_cache_entry {
1367         struct hashmap_entry ent;
1368         struct delta_base_cache_key key;
1369         struct list_head lru;
1370         void *data;
1371         unsigned long size;
1372         enum object_type type;
1373 };
1374
1375 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
1376 {
1377         unsigned int hash;
1378
1379         hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
1380         hash += (hash >> 8) + (hash >> 16);
1381         return hash;
1382 }
1383
1384 static struct delta_base_cache_entry *
1385 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
1386 {
1387         struct hashmap_entry entry, *e;
1388         struct delta_base_cache_key key;
1389
1390         if (!delta_base_cache.cmpfn)
1391                 return NULL;
1392
1393         hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
1394         key.p = p;
1395         key.base_offset = base_offset;
1396         e = hashmap_get(&delta_base_cache, &entry, &key);
1397         return e ? container_of(e, struct delta_base_cache_entry, ent) : NULL;
1398 }
1399
1400 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
1401                                    const struct delta_base_cache_key *b)
1402 {
1403         return a->p == b->p && a->base_offset == b->base_offset;
1404 }
1405
1406 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
1407                                      const struct hashmap_entry *va,
1408                                      const struct hashmap_entry *vb,
1409                                      const void *vkey)
1410 {
1411         const struct delta_base_cache_entry *a, *b;
1412         const struct delta_base_cache_key *key = vkey;
1413
1414         a = container_of(va, const struct delta_base_cache_entry, ent);
1415         b = container_of(vb, const struct delta_base_cache_entry, ent);
1416
1417         if (key)
1418                 return !delta_base_cache_key_eq(&a->key, key);
1419         else
1420                 return !delta_base_cache_key_eq(&a->key, &b->key);
1421 }
1422
1423 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1424 {
1425         return !!get_delta_base_cache_entry(p, base_offset);
1426 }
1427
1428 /*
1429  * Remove the entry from the cache, but do _not_ free the associated
1430  * entry data. The caller takes ownership of the "data" buffer, and
1431  * should copy out any fields it wants before detaching.
1432  */
1433 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
1434 {
1435         hashmap_remove(&delta_base_cache, &ent->ent, &ent->key);
1436         list_del(&ent->lru);
1437         delta_base_cached -= ent->size;
1438         free(ent);
1439 }
1440
1441 static void *cache_or_unpack_entry(struct repository *r, struct packed_git *p,
1442                                    off_t base_offset, unsigned long *base_size,
1443                                    enum object_type *type)
1444 {
1445         struct delta_base_cache_entry *ent;
1446
1447         ent = get_delta_base_cache_entry(p, base_offset);
1448         if (!ent)
1449                 return unpack_entry(r, p, base_offset, type, base_size);
1450
1451         if (type)
1452                 *type = ent->type;
1453         if (base_size)
1454                 *base_size = ent->size;
1455         return xmemdupz(ent->data, ent->size);
1456 }
1457
1458 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1459 {
1460         free(ent->data);
1461         detach_delta_base_cache_entry(ent);
1462 }
1463
1464 void clear_delta_base_cache(void)
1465 {
1466         struct list_head *lru, *tmp;
1467         list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1468                 struct delta_base_cache_entry *entry =
1469                         list_entry(lru, struct delta_base_cache_entry, lru);
1470                 release_delta_base_cache(entry);
1471         }
1472 }
1473
1474 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1475         void *base, unsigned long base_size, enum object_type type)
1476 {
1477         struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
1478         struct list_head *lru, *tmp;
1479
1480         /*
1481          * Check required to avoid redundant entries when more than one thread
1482          * is unpacking the same object, in unpack_entry() (since its phases I
1483          * and III might run concurrently across multiple threads).
1484          */
1485         if (in_delta_base_cache(p, base_offset))
1486                 return;
1487
1488         delta_base_cached += base_size;
1489
1490         list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
1491                 struct delta_base_cache_entry *f =
1492                         list_entry(lru, struct delta_base_cache_entry, lru);
1493                 if (delta_base_cached <= delta_base_cache_limit)
1494                         break;
1495                 release_delta_base_cache(f);
1496         }
1497
1498         ent->key.p = p;
1499         ent->key.base_offset = base_offset;
1500         ent->type = type;
1501         ent->data = base;
1502         ent->size = base_size;
1503         list_add_tail(&ent->lru, &delta_base_cache_lru);
1504
1505         if (!delta_base_cache.cmpfn)
1506                 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
1507         hashmap_entry_init(&ent->ent, pack_entry_hash(p, base_offset));
1508         hashmap_add(&delta_base_cache, &ent->ent);
1509 }
1510
1511 int packed_object_info(struct repository *r, struct packed_git *p,
1512                        off_t obj_offset, struct object_info *oi)
1513 {
1514         struct pack_window *w_curs = NULL;
1515         unsigned long size;
1516         off_t curpos = obj_offset;
1517         enum object_type type;
1518
1519         /*
1520          * We always get the representation type, but only convert it to
1521          * a "real" type later if the caller is interested.
1522          */
1523         if (oi->contentp) {
1524                 *oi->contentp = cache_or_unpack_entry(r, p, obj_offset, oi->sizep,
1525                                                       &type);
1526                 if (!*oi->contentp)
1527                         type = OBJ_BAD;
1528         } else {
1529                 type = unpack_object_header(p, &w_curs, &curpos, &size);
1530         }
1531
1532         if (!oi->contentp && oi->sizep) {
1533                 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1534                         off_t tmp_pos = curpos;
1535                         off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
1536                                                            type, obj_offset);
1537                         if (!base_offset) {
1538                                 type = OBJ_BAD;
1539                                 goto out;
1540                         }
1541                         *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
1542                         if (*oi->sizep == 0) {
1543                                 type = OBJ_BAD;
1544                                 goto out;
1545                         }
1546                 } else {
1547                         *oi->sizep = size;
1548                 }
1549         }
1550
1551         if (oi->disk_sizep) {
1552                 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1553                 *oi->disk_sizep = revidx[1].offset - obj_offset;
1554         }
1555
1556         if (oi->typep || oi->type_name) {
1557                 enum object_type ptot;
1558                 ptot = packed_to_object_type(r, p, obj_offset,
1559                                              type, &w_curs, curpos);
1560                 if (oi->typep)
1561                         *oi->typep = ptot;
1562                 if (oi->type_name) {
1563                         const char *tn = type_name(ptot);
1564                         if (tn)
1565                                 strbuf_addstr(oi->type_name, tn);
1566                 }
1567                 if (ptot < 0) {
1568                         type = OBJ_BAD;
1569                         goto out;
1570                 }
1571         }
1572
1573         if (oi->delta_base_oid) {
1574                 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1575                         if (get_delta_base_oid(p, &w_curs, curpos,
1576                                                oi->delta_base_oid,
1577                                                type, obj_offset) < 0) {
1578                                 type = OBJ_BAD;
1579                                 goto out;
1580                         }
1581                 } else
1582                         oidclr(oi->delta_base_oid);
1583         }
1584
1585         oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
1586                                                           OI_PACKED;
1587
1588 out:
1589         unuse_pack(&w_curs);
1590         return type;
1591 }
1592
1593 static void *unpack_compressed_entry(struct packed_git *p,
1594                                     struct pack_window **w_curs,
1595                                     off_t curpos,
1596                                     unsigned long size)
1597 {
1598         int st;
1599         git_zstream stream;
1600         unsigned char *buffer, *in;
1601
1602         buffer = xmallocz_gently(size);
1603         if (!buffer)
1604                 return NULL;
1605         memset(&stream, 0, sizeof(stream));
1606         stream.next_out = buffer;
1607         stream.avail_out = size + 1;
1608
1609         git_inflate_init(&stream);
1610         do {
1611                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1612                 stream.next_in = in;
1613                 /*
1614                  * Note: we must ensure the window section returned by
1615                  * use_pack() will be available throughout git_inflate()'s
1616                  * unlocked execution. Please refer to the comment at
1617                  * get_size_from_delta() to see how this is done.
1618                  */
1619                 obj_read_unlock();
1620                 st = git_inflate(&stream, Z_FINISH);
1621                 obj_read_lock();
1622                 if (!stream.avail_out)
1623                         break; /* the payload is larger than it should be */
1624                 curpos += stream.next_in - in;
1625         } while (st == Z_OK || st == Z_BUF_ERROR);
1626         git_inflate_end(&stream);
1627         if ((st != Z_STREAM_END) || stream.total_out != size) {
1628                 free(buffer);
1629                 return NULL;
1630         }
1631
1632         /* versions of zlib can clobber unconsumed portion of outbuf */
1633         buffer[size] = '\0';
1634
1635         return buffer;
1636 }
1637
1638 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1639 {
1640         static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
1641         trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
1642                          p->pack_name, (uintmax_t)obj_offset);
1643 }
1644
1645 int do_check_packed_object_crc;
1646
1647 #define UNPACK_ENTRY_STACK_PREALLOC 64
1648 struct unpack_entry_stack_ent {
1649         off_t obj_offset;
1650         off_t curpos;
1651         unsigned long size;
1652 };
1653
1654 static void *read_object(struct repository *r,
1655                          const struct object_id *oid,
1656                          enum object_type *type,
1657                          unsigned long *size)
1658 {
1659         struct object_info oi = OBJECT_INFO_INIT;
1660         void *content;
1661         oi.typep = type;
1662         oi.sizep = size;
1663         oi.contentp = &content;
1664
1665         if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1666                 return NULL;
1667         return content;
1668 }
1669
1670 void *unpack_entry(struct repository *r, struct packed_git *p, off_t obj_offset,
1671                    enum object_type *final_type, unsigned long *final_size)
1672 {
1673         struct pack_window *w_curs = NULL;
1674         off_t curpos = obj_offset;
1675         void *data = NULL;
1676         unsigned long size;
1677         enum object_type type;
1678         struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
1679         struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
1680         int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
1681         int base_from_cache = 0;
1682
1683         write_pack_access_log(p, obj_offset);
1684
1685         /* PHASE 1: drill down to the innermost base object */
1686         for (;;) {
1687                 off_t base_offset;
1688                 int i;
1689                 struct delta_base_cache_entry *ent;
1690
1691                 ent = get_delta_base_cache_entry(p, curpos);
1692                 if (ent) {
1693                         type = ent->type;
1694                         data = ent->data;
1695                         size = ent->size;
1696                         detach_delta_base_cache_entry(ent);
1697                         base_from_cache = 1;
1698                         break;
1699                 }
1700
1701                 if (do_check_packed_object_crc && p->index_version > 1) {
1702                         struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1703                         off_t len = revidx[1].offset - obj_offset;
1704                         if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1705                                 struct object_id oid;
1706                                 nth_packed_object_id(&oid, p, revidx->nr);
1707                                 error("bad packed object CRC for %s",
1708                                       oid_to_hex(&oid));
1709                                 mark_bad_packed_object(p, oid.hash);
1710                                 data = NULL;
1711                                 goto out;
1712                         }
1713                 }
1714
1715                 type = unpack_object_header(p, &w_curs, &curpos, &size);
1716                 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
1717                         break;
1718
1719                 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1720                 if (!base_offset) {
1721                         error("failed to validate delta base reference "
1722                               "at offset %"PRIuMAX" from %s",
1723                               (uintmax_t)curpos, p->pack_name);
1724                         /* bail to phase 2, in hopes of recovery */
1725                         data = NULL;
1726                         break;
1727                 }
1728
1729                 /* push object, proceed to base */
1730                 if (delta_stack_nr >= delta_stack_alloc
1731                     && delta_stack == small_delta_stack) {
1732                         delta_stack_alloc = alloc_nr(delta_stack_nr);
1733                         ALLOC_ARRAY(delta_stack, delta_stack_alloc);
1734                         COPY_ARRAY(delta_stack, small_delta_stack,
1735                                    delta_stack_nr);
1736                 } else {
1737                         ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
1738                 }
1739                 i = delta_stack_nr++;
1740                 delta_stack[i].obj_offset = obj_offset;
1741                 delta_stack[i].curpos = curpos;
1742                 delta_stack[i].size = size;
1743
1744                 curpos = obj_offset = base_offset;
1745         }
1746
1747         /* PHASE 2: handle the base */
1748         switch (type) {
1749         case OBJ_OFS_DELTA:
1750         case OBJ_REF_DELTA:
1751                 if (data)
1752                         BUG("unpack_entry: left loop at a valid delta");
1753                 break;
1754         case OBJ_COMMIT:
1755         case OBJ_TREE:
1756         case OBJ_BLOB:
1757         case OBJ_TAG:
1758                 if (!base_from_cache)
1759                         data = unpack_compressed_entry(p, &w_curs, curpos, size);
1760                 break;
1761         default:
1762                 data = NULL;
1763                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1764                       type, (uintmax_t)obj_offset, p->pack_name);
1765         }
1766
1767         /* PHASE 3: apply deltas in order */
1768
1769         /* invariants:
1770          *   'data' holds the base data, or NULL if there was corruption
1771          */
1772         while (delta_stack_nr) {
1773                 void *delta_data;
1774                 void *base = data;
1775                 void *external_base = NULL;
1776                 unsigned long delta_size, base_size = size;
1777                 int i;
1778
1779                 data = NULL;
1780
1781                 if (base)
1782                         add_delta_base_cache(p, obj_offset, base, base_size, type);
1783
1784                 if (!base) {
1785                         /*
1786                          * We're probably in deep shit, but let's try to fetch
1787                          * the required base anyway from another pack or loose.
1788                          * This is costly but should happen only in the presence
1789                          * of a corrupted pack, and is better than failing outright.
1790                          */
1791                         struct revindex_entry *revidx;
1792                         struct object_id base_oid;
1793                         revidx = find_pack_revindex(p, obj_offset);
1794                         if (revidx) {
1795                                 nth_packed_object_id(&base_oid, p, revidx->nr);
1796                                 error("failed to read delta base object %s"
1797                                       " at offset %"PRIuMAX" from %s",
1798                                       oid_to_hex(&base_oid), (uintmax_t)obj_offset,
1799                                       p->pack_name);
1800                                 mark_bad_packed_object(p, base_oid.hash);
1801                                 base = read_object(r, &base_oid, &type, &base_size);
1802                                 external_base = base;
1803                         }
1804                 }
1805
1806                 i = --delta_stack_nr;
1807                 obj_offset = delta_stack[i].obj_offset;
1808                 curpos = delta_stack[i].curpos;
1809                 delta_size = delta_stack[i].size;
1810
1811                 if (!base)
1812                         continue;
1813
1814                 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
1815
1816                 if (!delta_data) {
1817                         error("failed to unpack compressed delta "
1818                               "at offset %"PRIuMAX" from %s",
1819                               (uintmax_t)curpos, p->pack_name);
1820                         data = NULL;
1821                         free(external_base);
1822                         continue;
1823                 }
1824
1825                 data = patch_delta(base, base_size,
1826                                    delta_data, delta_size,
1827                                    &size);
1828
1829                 /*
1830                  * We could not apply the delta; warn the user, but keep going.
1831                  * Our failure will be noticed either in the next iteration of
1832                  * the loop, or if this is the final delta, in the caller when
1833                  * we return NULL. Those code paths will take care of making
1834                  * a more explicit warning and retrying with another copy of
1835                  * the object.
1836                  */
1837                 if (!data)
1838                         error("failed to apply delta");
1839
1840                 free(delta_data);
1841                 free(external_base);
1842         }
1843
1844         if (final_type)
1845                 *final_type = type;
1846         if (final_size)
1847                 *final_size = size;
1848
1849 out:
1850         unuse_pack(&w_curs);
1851
1852         if (delta_stack != small_delta_stack)
1853                 free(delta_stack);
1854
1855         return data;
1856 }
1857
1858 int bsearch_pack(const struct object_id *oid, const struct packed_git *p, uint32_t *result)
1859 {
1860         const unsigned char *index_fanout = p->index_data;
1861         const unsigned char *index_lookup;
1862         const unsigned int hashsz = the_hash_algo->rawsz;
1863         int index_lookup_width;
1864
1865         if (!index_fanout)
1866                 BUG("bsearch_pack called without a valid pack-index");
1867
1868         index_lookup = index_fanout + 4 * 256;
1869         if (p->index_version == 1) {
1870                 index_lookup_width = hashsz + 4;
1871                 index_lookup += 4;
1872         } else {
1873                 index_lookup_width = hashsz;
1874                 index_fanout += 8;
1875                 index_lookup += 8;
1876         }
1877
1878         return bsearch_hash(oid->hash, (const uint32_t*)index_fanout,
1879                             index_lookup, index_lookup_width, result);
1880 }
1881
1882 int nth_packed_object_id(struct object_id *oid,
1883                          struct packed_git *p,
1884                          uint32_t n)
1885 {
1886         const unsigned char *index = p->index_data;
1887         const unsigned int hashsz = the_hash_algo->rawsz;
1888         if (!index) {
1889                 if (open_pack_index(p))
1890                         return -1;
1891                 index = p->index_data;
1892         }
1893         if (n >= p->num_objects)
1894                 return -1;
1895         index += 4 * 256;
1896         if (p->index_version == 1) {
1897                 oidread(oid, index + (hashsz + 4) * n + 4);
1898         } else {
1899                 index += 8;
1900                 oidread(oid, index + hashsz * n);
1901         }
1902         return 0;
1903 }
1904
1905 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
1906 {
1907         const unsigned char *ptr = vptr;
1908         const unsigned char *start = p->index_data;
1909         const unsigned char *end = start + p->index_size;
1910         if (ptr < start)
1911                 die(_("offset before start of pack index for %s (corrupt index?)"),
1912                     p->pack_name);
1913         /* No need to check for underflow; .idx files must be at least 8 bytes */
1914         if (ptr >= end - 8)
1915                 die(_("offset beyond end of pack index for %s (truncated index?)"),
1916                     p->pack_name);
1917 }
1918
1919 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1920 {
1921         const unsigned char *index = p->index_data;
1922         const unsigned int hashsz = the_hash_algo->rawsz;
1923         index += 4 * 256;
1924         if (p->index_version == 1) {
1925                 return ntohl(*((uint32_t *)(index + (hashsz + 4) * n)));
1926         } else {
1927                 uint32_t off;
1928                 index += 8 + p->num_objects * (hashsz + 4);
1929                 off = ntohl(*((uint32_t *)(index + 4 * n)));
1930                 if (!(off & 0x80000000))
1931                         return off;
1932                 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1933                 check_pack_index_ptr(p, index);
1934                 return get_be64(index);
1935         }
1936 }
1937
1938 off_t find_pack_entry_one(const unsigned char *sha1,
1939                                   struct packed_git *p)
1940 {
1941         const unsigned char *index = p->index_data;
1942         struct object_id oid;
1943         uint32_t result;
1944
1945         if (!index) {
1946                 if (open_pack_index(p))
1947                         return 0;
1948         }
1949
1950         hashcpy(oid.hash, sha1);
1951         if (bsearch_pack(&oid, p, &result))
1952                 return nth_packed_object_offset(p, result);
1953         return 0;
1954 }
1955
1956 int is_pack_valid(struct packed_git *p)
1957 {
1958         /* An already open pack is known to be valid. */
1959         if (p->pack_fd != -1)
1960                 return 1;
1961
1962         /* If the pack has one window completely covering the
1963          * file size, the pack is known to be valid even if
1964          * the descriptor is not currently open.
1965          */
1966         if (p->windows) {
1967                 struct pack_window *w = p->windows;
1968
1969                 if (!w->offset && w->len == p->pack_size)
1970                         return 1;
1971         }
1972
1973         /* Force the pack to open to prove its valid. */
1974         return !open_packed_git(p);
1975 }
1976
1977 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1978                                   struct packed_git *packs)
1979 {
1980         struct packed_git *p;
1981
1982         for (p = packs; p; p = p->next) {
1983                 if (find_pack_entry_one(sha1, p))
1984                         return p;
1985         }
1986         return NULL;
1987
1988 }
1989
1990 static int fill_pack_entry(const struct object_id *oid,
1991                            struct pack_entry *e,
1992                            struct packed_git *p)
1993 {
1994         off_t offset;
1995
1996         if (p->num_bad_objects) {
1997                 unsigned i;
1998                 for (i = 0; i < p->num_bad_objects; i++)
1999                         if (hasheq(oid->hash,
2000                                    p->bad_object_sha1 + the_hash_algo->rawsz * i))
2001                                 return 0;
2002         }
2003
2004         offset = find_pack_entry_one(oid->hash, p);
2005         if (!offset)
2006                 return 0;
2007
2008         /*
2009          * We are about to tell the caller where they can locate the
2010          * requested object.  We better make sure the packfile is
2011          * still here and can be accessed before supplying that
2012          * answer, as it may have been deleted since the index was
2013          * loaded!
2014          */
2015         if (!is_pack_valid(p))
2016                 return 0;
2017         e->offset = offset;
2018         e->p = p;
2019         return 1;
2020 }
2021
2022 int find_pack_entry(struct repository *r, const struct object_id *oid, struct pack_entry *e)
2023 {
2024         struct list_head *pos;
2025         struct multi_pack_index *m;
2026
2027         prepare_packed_git(r);
2028         if (!r->objects->packed_git && !r->objects->multi_pack_index)
2029                 return 0;
2030
2031         for (m = r->objects->multi_pack_index; m; m = m->next) {
2032                 if (fill_midx_entry(r, oid, e, m))
2033                         return 1;
2034         }
2035
2036         list_for_each(pos, &r->objects->packed_git_mru) {
2037                 struct packed_git *p = list_entry(pos, struct packed_git, mru);
2038                 if (!p->multi_pack_index && fill_pack_entry(oid, e, p)) {
2039                         list_move(&p->mru, &r->objects->packed_git_mru);
2040                         return 1;
2041                 }
2042         }
2043         return 0;
2044 }
2045
2046 int has_object_pack(const struct object_id *oid)
2047 {
2048         struct pack_entry e;
2049         return find_pack_entry(the_repository, oid, &e);
2050 }
2051
2052 int has_pack_index(const unsigned char *sha1)
2053 {
2054         struct stat st;
2055         if (stat(sha1_pack_index_name(sha1), &st))
2056                 return 0;
2057         return 1;
2058 }
2059
2060 int for_each_object_in_pack(struct packed_git *p,
2061                             each_packed_object_fn cb, void *data,
2062                             enum for_each_object_flags flags)
2063 {
2064         uint32_t i;
2065         int r = 0;
2066
2067         if (flags & FOR_EACH_OBJECT_PACK_ORDER) {
2068                 if (load_pack_revindex(p))
2069                         return -1;
2070         }
2071
2072         for (i = 0; i < p->num_objects; i++) {
2073                 uint32_t pos;
2074                 struct object_id oid;
2075
2076                 if (flags & FOR_EACH_OBJECT_PACK_ORDER)
2077                         pos = p->revindex[i].nr;
2078                 else
2079                         pos = i;
2080
2081                 if (nth_packed_object_id(&oid, p, pos) < 0)
2082                         return error("unable to get sha1 of object %u in %s",
2083                                      pos, p->pack_name);
2084
2085                 r = cb(&oid, p, pos, data);
2086                 if (r)
2087                         break;
2088         }
2089         return r;
2090 }
2091
2092 int for_each_packed_object(each_packed_object_fn cb, void *data,
2093                            enum for_each_object_flags flags)
2094 {
2095         struct packed_git *p;
2096         int r = 0;
2097         int pack_errors = 0;
2098
2099         prepare_packed_git(the_repository);
2100         for (p = get_all_packs(the_repository); p; p = p->next) {
2101                 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
2102                         continue;
2103                 if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
2104                     !p->pack_promisor)
2105                         continue;
2106                 if (open_pack_index(p)) {
2107                         pack_errors = 1;
2108                         continue;
2109                 }
2110                 r = for_each_object_in_pack(p, cb, data, flags);
2111                 if (r)
2112                         break;
2113         }
2114         return r ? r : pack_errors;
2115 }
2116
2117 static int add_promisor_object(const struct object_id *oid,
2118                                struct packed_git *pack,
2119                                uint32_t pos,
2120                                void *set_)
2121 {
2122         struct oidset *set = set_;
2123         struct object *obj = parse_object(the_repository, oid);
2124         if (!obj)
2125                 return 1;
2126
2127         oidset_insert(set, oid);
2128
2129         /*
2130          * If this is a tree, commit, or tag, the objects it refers
2131          * to are also promisor objects. (Blobs refer to no objects->)
2132          */
2133         if (obj->type == OBJ_TREE) {
2134                 struct tree *tree = (struct tree *)obj;
2135                 struct tree_desc desc;
2136                 struct name_entry entry;
2137                 if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
2138                         /*
2139                          * Error messages are given when packs are
2140                          * verified, so do not print any here.
2141                          */
2142                         return 0;
2143                 while (tree_entry_gently(&desc, &entry))
2144                         oidset_insert(set, &entry.oid);
2145         } else if (obj->type == OBJ_COMMIT) {
2146                 struct commit *commit = (struct commit *) obj;
2147                 struct commit_list *parents = commit->parents;
2148
2149                 oidset_insert(set, get_commit_tree_oid(commit));
2150                 for (; parents; parents = parents->next)
2151                         oidset_insert(set, &parents->item->object.oid);
2152         } else if (obj->type == OBJ_TAG) {
2153                 struct tag *tag = (struct tag *) obj;
2154                 oidset_insert(set, get_tagged_oid(tag));
2155         }
2156         return 0;
2157 }
2158
2159 int is_promisor_object(const struct object_id *oid)
2160 {
2161         static struct oidset promisor_objects;
2162         static int promisor_objects_prepared;
2163
2164         if (!promisor_objects_prepared) {
2165                 if (has_promisor_remote()) {
2166                         for_each_packed_object(add_promisor_object,
2167                                                &promisor_objects,
2168                                                FOR_EACH_OBJECT_PROMISOR_ONLY);
2169                 }
2170                 promisor_objects_prepared = 1;
2171         }
2172         return oidset_contains(&promisor_objects, oid);
2173 }