1#include "builtin.h" 2#include "cache.h" 3#include "repository.h" 4#include "config.h" 5#include "attr.h" 6#include "object.h" 7#include "blob.h" 8#include "commit.h" 9#include "tag.h" 10#include "tree.h" 11#include "delta.h" 12#include "pack.h" 13#include "pack-revindex.h" 14#include "csum-file.h" 15#include "tree-walk.h" 16#include "diff.h" 17#include "revision.h" 18#include "list-objects.h" 19#include "list-objects-filter.h" 20#include "list-objects-filter-options.h" 21#include "pack-objects.h" 22#include "progress.h" 23#include "refs.h" 24#include "streaming.h" 25#include "thread-utils.h" 26#include "pack-bitmap.h" 27#include "delta-islands.h" 28#include "reachable.h" 29#include "sha1-array.h" 30#include "argv-array.h" 31#include "list.h" 32#include "packfile.h" 33#include "object-store.h" 34#include "dir.h" 35#include "midx.h" 36#include "trace2.h" 37 38#define IN_PACK(obj) oe_in_pack(&to_pack, obj) 39#define SIZE(obj) oe_size(&to_pack, obj) 40#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) 41#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) 42#define DELTA(obj) oe_delta(&to_pack, obj) 43#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) 44#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) 45#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) 46#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid) 47#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) 48#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) 49#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) 50 51static const char *pack_usage[] = { 52 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 53 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 54 NULL 55}; 56 57/* 58 * Objects we are going to pack are collected in the `to_pack` structure. 59 * It contains an array (dynamically expanded) of the object data, and a map 60 * that can resolve SHA1s to their position in the array. 61 */ 62static struct packing_data to_pack; 63 64static struct pack_idx_entry **written_list; 65static uint32_t nr_result, nr_written, nr_seen; 66static struct bitmap_index *bitmap_git; 67static uint32_t write_layer; 68 69static int non_empty; 70static int reuse_delta = 1, reuse_object = 1; 71static int keep_unreachable, unpack_unreachable, include_tag; 72static timestamp_t unpack_unreachable_expiration; 73static int pack_loose_unreachable; 74static int local; 75static int have_non_local_packs; 76static int incremental; 77static int ignore_packed_keep_on_disk; 78static int ignore_packed_keep_in_core; 79static int allow_ofs_delta; 80static struct pack_idx_option pack_idx_opts; 81static const char *base_name; 82static int progress = 1; 83static int window = 10; 84static unsigned long pack_size_limit; 85static int depth = 50; 86static int delta_search_threads; 87static int pack_to_stdout; 88static int sparse; 89static int thin; 90static int num_preferred_base; 91static struct progress *progress_state; 92 93static struct packed_git *reuse_packfile; 94static uint32_t reuse_packfile_objects; 95static off_t reuse_packfile_offset; 96 97static int use_bitmap_index_default = 1; 98static int use_bitmap_index = -1; 99static int write_bitmap_index; 100static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE; 101 102static int exclude_promisor_objects; 103 104static int use_delta_islands; 105 106static unsigned long delta_cache_size = 0; 107static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 108static unsigned long cache_max_small_delta_size = 1000; 109 110static unsigned long window_memory_limit = 0; 111 112static struct list_objects_filter_options filter_options; 113 114enum missing_action { 115 MA_ERROR = 0, /* fail if any missing objects are encountered */ 116 MA_ALLOW_ANY, /* silently allow ALL missing objects */ 117 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */ 118}; 119static enum missing_action arg_missing_action; 120static show_object_fn fn_show_object; 121 122/* 123 * stats 124 */ 125static uint32_t written, written_delta; 126static uint32_t reused, reused_delta; 127 128/* 129 * Indexed commits 130 */ 131static struct commit **indexed_commits; 132static unsigned int indexed_commits_nr; 133static unsigned int indexed_commits_alloc; 134 135static void index_commit_for_bitmap(struct commit *commit) 136{ 137 if (indexed_commits_nr >= indexed_commits_alloc) { 138 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; 139 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 140 } 141 142 indexed_commits[indexed_commits_nr++] = commit; 143} 144 145static void *get_delta(struct object_entry *entry) 146{ 147 unsigned long size, base_size, delta_size; 148 void *buf, *base_buf, *delta_buf; 149 enum object_type type; 150 151 buf = read_object_file(&entry->idx.oid, &type, &size); 152 if (!buf) 153 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid)); 154 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type, 155 &base_size); 156 if (!base_buf) 157 die("unable to read %s", 158 oid_to_hex(&DELTA(entry)->idx.oid)); 159 delta_buf = diff_delta(base_buf, base_size, 160 buf, size, &delta_size, 0); 161 /* 162 * We succesfully computed this delta once but dropped it for 163 * memory reasons. Something is very wrong if this time we 164 * recompute and create a different delta. 165 */ 166 if (!delta_buf || delta_size != DELTA_SIZE(entry)) 167 BUG("delta size changed"); 168 free(buf); 169 free(base_buf); 170 return delta_buf; 171} 172 173static unsigned long do_compress(void **pptr, unsigned long size) 174{ 175 git_zstream stream; 176 void *in, *out; 177 unsigned long maxsize; 178 179 git_deflate_init(&stream, pack_compression_level); 180 maxsize = git_deflate_bound(&stream, size); 181 182 in = *pptr; 183 out = xmalloc(maxsize); 184 *pptr = out; 185 186 stream.next_in = in; 187 stream.avail_in = size; 188 stream.next_out = out; 189 stream.avail_out = maxsize; 190 while (git_deflate(&stream, Z_FINISH) == Z_OK) 191 ; /* nothing */ 192 git_deflate_end(&stream); 193 194 free(in); 195 return stream.total_out; 196} 197 198static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f, 199 const struct object_id *oid) 200{ 201 git_zstream stream; 202 unsigned char ibuf[1024 * 16]; 203 unsigned char obuf[1024 * 16]; 204 unsigned long olen = 0; 205 206 git_deflate_init(&stream, pack_compression_level); 207 208 for (;;) { 209 ssize_t readlen; 210 int zret = Z_OK; 211 readlen = read_istream(st, ibuf, sizeof(ibuf)); 212 if (readlen == -1) 213 die(_("unable to read %s"), oid_to_hex(oid)); 214 215 stream.next_in = ibuf; 216 stream.avail_in = readlen; 217 while ((stream.avail_in || readlen == 0) && 218 (zret == Z_OK || zret == Z_BUF_ERROR)) { 219 stream.next_out = obuf; 220 stream.avail_out = sizeof(obuf); 221 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH); 222 hashwrite(f, obuf, stream.next_out - obuf); 223 olen += stream.next_out - obuf; 224 } 225 if (stream.avail_in) 226 die(_("deflate error (%d)"), zret); 227 if (readlen == 0) { 228 if (zret != Z_STREAM_END) 229 die(_("deflate error (%d)"), zret); 230 break; 231 } 232 } 233 git_deflate_end(&stream); 234 return olen; 235} 236 237/* 238 * we are going to reuse the existing object data as is. make 239 * sure it is not corrupt. 240 */ 241static int check_pack_inflate(struct packed_git *p, 242 struct pack_window **w_curs, 243 off_t offset, 244 off_t len, 245 unsigned long expect) 246{ 247 git_zstream stream; 248 unsigned char fakebuf[4096], *in; 249 int st; 250 251 memset(&stream, 0, sizeof(stream)); 252 git_inflate_init(&stream); 253 do { 254 in = use_pack(p, w_curs, offset, &stream.avail_in); 255 stream.next_in = in; 256 stream.next_out = fakebuf; 257 stream.avail_out = sizeof(fakebuf); 258 st = git_inflate(&stream, Z_FINISH); 259 offset += stream.next_in - in; 260 } while (st == Z_OK || st == Z_BUF_ERROR); 261 git_inflate_end(&stream); 262 return (st == Z_STREAM_END && 263 stream.total_out == expect && 264 stream.total_in == len) ? 0 : -1; 265} 266 267static void copy_pack_data(struct hashfile *f, 268 struct packed_git *p, 269 struct pack_window **w_curs, 270 off_t offset, 271 off_t len) 272{ 273 unsigned char *in; 274 unsigned long avail; 275 276 while (len) { 277 in = use_pack(p, w_curs, offset, &avail); 278 if (avail > len) 279 avail = (unsigned long)len; 280 hashwrite(f, in, avail); 281 offset += avail; 282 len -= avail; 283 } 284} 285 286/* Return 0 if we will bust the pack-size limit */ 287static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry, 288 unsigned long limit, int usable_delta) 289{ 290 unsigned long size, datalen; 291 unsigned char header[MAX_PACK_OBJECT_HEADER], 292 dheader[MAX_PACK_OBJECT_HEADER]; 293 unsigned hdrlen; 294 enum object_type type; 295 void *buf; 296 struct git_istream *st = NULL; 297 const unsigned hashsz = the_hash_algo->rawsz; 298 299 if (!usable_delta) { 300 if (oe_type(entry) == OBJ_BLOB && 301 oe_size_greater_than(&to_pack, entry, big_file_threshold) && 302 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 303 buf = NULL; 304 else { 305 buf = read_object_file(&entry->idx.oid, &type, &size); 306 if (!buf) 307 die(_("unable to read %s"), 308 oid_to_hex(&entry->idx.oid)); 309 } 310 /* 311 * make sure no cached delta data remains from a 312 * previous attempt before a pack split occurred. 313 */ 314 FREE_AND_NULL(entry->delta_data); 315 entry->z_delta_size = 0; 316 } else if (entry->delta_data) { 317 size = DELTA_SIZE(entry); 318 buf = entry->delta_data; 319 entry->delta_data = NULL; 320 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 321 OBJ_OFS_DELTA : OBJ_REF_DELTA; 322 } else { 323 buf = get_delta(entry); 324 size = DELTA_SIZE(entry); 325 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 326 OBJ_OFS_DELTA : OBJ_REF_DELTA; 327 } 328 329 if (st) /* large blob case, just assume we don't compress well */ 330 datalen = size; 331 else if (entry->z_delta_size) 332 datalen = entry->z_delta_size; 333 else 334 datalen = do_compress(&buf, size); 335 336 /* 337 * The object header is a byte of 'type' followed by zero or 338 * more bytes of length. 339 */ 340 hdrlen = encode_in_pack_object_header(header, sizeof(header), 341 type, size); 342 343 if (type == OBJ_OFS_DELTA) { 344 /* 345 * Deltas with relative base contain an additional 346 * encoding of the relative offset for the delta 347 * base from this object's position in the pack. 348 */ 349 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 350 unsigned pos = sizeof(dheader) - 1; 351 dheader[pos] = ofs & 127; 352 while (ofs >>= 7) 353 dheader[--pos] = 128 | (--ofs & 127); 354 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 355 if (st) 356 close_istream(st); 357 free(buf); 358 return 0; 359 } 360 hashwrite(f, header, hdrlen); 361 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 362 hdrlen += sizeof(dheader) - pos; 363 } else if (type == OBJ_REF_DELTA) { 364 /* 365 * Deltas with a base reference contain 366 * additional bytes for the base object ID. 367 */ 368 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 369 if (st) 370 close_istream(st); 371 free(buf); 372 return 0; 373 } 374 hashwrite(f, header, hdrlen); 375 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 376 hdrlen += hashsz; 377 } else { 378 if (limit && hdrlen + datalen + hashsz >= limit) { 379 if (st) 380 close_istream(st); 381 free(buf); 382 return 0; 383 } 384 hashwrite(f, header, hdrlen); 385 } 386 if (st) { 387 datalen = write_large_blob_data(st, f, &entry->idx.oid); 388 close_istream(st); 389 } else { 390 hashwrite(f, buf, datalen); 391 free(buf); 392 } 393 394 return hdrlen + datalen; 395} 396 397/* Return 0 if we will bust the pack-size limit */ 398static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry, 399 unsigned long limit, int usable_delta) 400{ 401 struct packed_git *p = IN_PACK(entry); 402 struct pack_window *w_curs = NULL; 403 struct revindex_entry *revidx; 404 off_t offset; 405 enum object_type type = oe_type(entry); 406 off_t datalen; 407 unsigned char header[MAX_PACK_OBJECT_HEADER], 408 dheader[MAX_PACK_OBJECT_HEADER]; 409 unsigned hdrlen; 410 const unsigned hashsz = the_hash_algo->rawsz; 411 unsigned long entry_size = SIZE(entry); 412 413 if (DELTA(entry)) 414 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 415 OBJ_OFS_DELTA : OBJ_REF_DELTA; 416 hdrlen = encode_in_pack_object_header(header, sizeof(header), 417 type, entry_size); 418 419 offset = entry->in_pack_offset; 420 revidx = find_pack_revindex(p, offset); 421 datalen = revidx[1].offset - offset; 422 if (!pack_to_stdout && p->index_version > 1 && 423 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 424 error(_("bad packed object CRC for %s"), 425 oid_to_hex(&entry->idx.oid)); 426 unuse_pack(&w_curs); 427 return write_no_reuse_object(f, entry, limit, usable_delta); 428 } 429 430 offset += entry->in_pack_header_size; 431 datalen -= entry->in_pack_header_size; 432 433 if (!pack_to_stdout && p->index_version == 1 && 434 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 435 error(_("corrupt packed object for %s"), 436 oid_to_hex(&entry->idx.oid)); 437 unuse_pack(&w_curs); 438 return write_no_reuse_object(f, entry, limit, usable_delta); 439 } 440 441 if (type == OBJ_OFS_DELTA) { 442 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 443 unsigned pos = sizeof(dheader) - 1; 444 dheader[pos] = ofs & 127; 445 while (ofs >>= 7) 446 dheader[--pos] = 128 | (--ofs & 127); 447 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 448 unuse_pack(&w_curs); 449 return 0; 450 } 451 hashwrite(f, header, hdrlen); 452 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 453 hdrlen += sizeof(dheader) - pos; 454 reused_delta++; 455 } else if (type == OBJ_REF_DELTA) { 456 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 457 unuse_pack(&w_curs); 458 return 0; 459 } 460 hashwrite(f, header, hdrlen); 461 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 462 hdrlen += hashsz; 463 reused_delta++; 464 } else { 465 if (limit && hdrlen + datalen + hashsz >= limit) { 466 unuse_pack(&w_curs); 467 return 0; 468 } 469 hashwrite(f, header, hdrlen); 470 } 471 copy_pack_data(f, p, &w_curs, offset, datalen); 472 unuse_pack(&w_curs); 473 reused++; 474 return hdrlen + datalen; 475} 476 477/* Return 0 if we will bust the pack-size limit */ 478static off_t write_object(struct hashfile *f, 479 struct object_entry *entry, 480 off_t write_offset) 481{ 482 unsigned long limit; 483 off_t len; 484 int usable_delta, to_reuse; 485 486 if (!pack_to_stdout) 487 crc32_begin(f); 488 489 /* apply size limit if limited packsize and not first object */ 490 if (!pack_size_limit || !nr_written) 491 limit = 0; 492 else if (pack_size_limit <= write_offset) 493 /* 494 * the earlier object did not fit the limit; avoid 495 * mistaking this with unlimited (i.e. limit = 0). 496 */ 497 limit = 1; 498 else 499 limit = pack_size_limit - write_offset; 500 501 if (!DELTA(entry)) 502 usable_delta = 0; /* no delta */ 503 else if (!pack_size_limit) 504 usable_delta = 1; /* unlimited packfile */ 505 else if (DELTA(entry)->idx.offset == (off_t)-1) 506 usable_delta = 0; /* base was written to another pack */ 507 else if (DELTA(entry)->idx.offset) 508 usable_delta = 1; /* base already exists in this pack */ 509 else 510 usable_delta = 0; /* base could end up in another pack */ 511 512 if (!reuse_object) 513 to_reuse = 0; /* explicit */ 514 else if (!IN_PACK(entry)) 515 to_reuse = 0; /* can't reuse what we don't have */ 516 else if (oe_type(entry) == OBJ_REF_DELTA || 517 oe_type(entry) == OBJ_OFS_DELTA) 518 /* check_object() decided it for us ... */ 519 to_reuse = usable_delta; 520 /* ... but pack split may override that */ 521 else if (oe_type(entry) != entry->in_pack_type) 522 to_reuse = 0; /* pack has delta which is unusable */ 523 else if (DELTA(entry)) 524 to_reuse = 0; /* we want to pack afresh */ 525 else 526 to_reuse = 1; /* we have it in-pack undeltified, 527 * and we do not need to deltify it. 528 */ 529 530 if (!to_reuse) 531 len = write_no_reuse_object(f, entry, limit, usable_delta); 532 else 533 len = write_reuse_object(f, entry, limit, usable_delta); 534 if (!len) 535 return 0; 536 537 if (usable_delta) 538 written_delta++; 539 written++; 540 if (!pack_to_stdout) 541 entry->idx.crc32 = crc32_end(f); 542 return len; 543} 544 545enum write_one_status { 546 WRITE_ONE_SKIP = -1, /* already written */ 547 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */ 548 WRITE_ONE_WRITTEN = 1, /* normal */ 549 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */ 550}; 551 552static enum write_one_status write_one(struct hashfile *f, 553 struct object_entry *e, 554 off_t *offset) 555{ 556 off_t size; 557 int recursing; 558 559 /* 560 * we set offset to 1 (which is an impossible value) to mark 561 * the fact that this object is involved in "write its base 562 * first before writing a deltified object" recursion. 563 */ 564 recursing = (e->idx.offset == 1); 565 if (recursing) { 566 warning(_("recursive delta detected for object %s"), 567 oid_to_hex(&e->idx.oid)); 568 return WRITE_ONE_RECURSIVE; 569 } else if (e->idx.offset || e->preferred_base) { 570 /* offset is non zero if object is written already. */ 571 return WRITE_ONE_SKIP; 572 } 573 574 /* if we are deltified, write out base object first. */ 575 if (DELTA(e)) { 576 e->idx.offset = 1; /* now recurse */ 577 switch (write_one(f, DELTA(e), offset)) { 578 case WRITE_ONE_RECURSIVE: 579 /* we cannot depend on this one */ 580 SET_DELTA(e, NULL); 581 break; 582 default: 583 break; 584 case WRITE_ONE_BREAK: 585 e->idx.offset = recursing; 586 return WRITE_ONE_BREAK; 587 } 588 } 589 590 e->idx.offset = *offset; 591 size = write_object(f, e, *offset); 592 if (!size) { 593 e->idx.offset = recursing; 594 return WRITE_ONE_BREAK; 595 } 596 written_list[nr_written++] = &e->idx; 597 598 /* make sure off_t is sufficiently large not to wrap */ 599 if (signed_add_overflows(*offset, size)) 600 die(_("pack too large for current definition of off_t")); 601 *offset += size; 602 return WRITE_ONE_WRITTEN; 603} 604 605static int mark_tagged(const char *path, const struct object_id *oid, int flag, 606 void *cb_data) 607{ 608 struct object_id peeled; 609 struct object_entry *entry = packlist_find(&to_pack, oid, NULL); 610 611 if (entry) 612 entry->tagged = 1; 613 if (!peel_ref(path, &peeled)) { 614 entry = packlist_find(&to_pack, &peeled, NULL); 615 if (entry) 616 entry->tagged = 1; 617 } 618 return 0; 619} 620 621static inline void add_to_write_order(struct object_entry **wo, 622 unsigned int *endp, 623 struct object_entry *e) 624{ 625 if (e->filled || oe_layer(&to_pack, e) != write_layer) 626 return; 627 wo[(*endp)++] = e; 628 e->filled = 1; 629} 630 631static void add_descendants_to_write_order(struct object_entry **wo, 632 unsigned int *endp, 633 struct object_entry *e) 634{ 635 int add_to_order = 1; 636 while (e) { 637 if (add_to_order) { 638 struct object_entry *s; 639 /* add this node... */ 640 add_to_write_order(wo, endp, e); 641 /* all its siblings... */ 642 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) { 643 add_to_write_order(wo, endp, s); 644 } 645 } 646 /* drop down a level to add left subtree nodes if possible */ 647 if (DELTA_CHILD(e)) { 648 add_to_order = 1; 649 e = DELTA_CHILD(e); 650 } else { 651 add_to_order = 0; 652 /* our sibling might have some children, it is next */ 653 if (DELTA_SIBLING(e)) { 654 e = DELTA_SIBLING(e); 655 continue; 656 } 657 /* go back to our parent node */ 658 e = DELTA(e); 659 while (e && !DELTA_SIBLING(e)) { 660 /* we're on the right side of a subtree, keep 661 * going up until we can go right again */ 662 e = DELTA(e); 663 } 664 if (!e) { 665 /* done- we hit our original root node */ 666 return; 667 } 668 /* pass it off to sibling at this level */ 669 e = DELTA_SIBLING(e); 670 } 671 }; 672} 673 674static void add_family_to_write_order(struct object_entry **wo, 675 unsigned int *endp, 676 struct object_entry *e) 677{ 678 struct object_entry *root; 679 680 for (root = e; DELTA(root); root = DELTA(root)) 681 ; /* nothing */ 682 add_descendants_to_write_order(wo, endp, root); 683} 684 685static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end) 686{ 687 unsigned int i, last_untagged; 688 struct object_entry *objects = to_pack.objects; 689 690 for (i = 0; i < to_pack.nr_objects; i++) { 691 if (objects[i].tagged) 692 break; 693 add_to_write_order(wo, wo_end, &objects[i]); 694 } 695 last_untagged = i; 696 697 /* 698 * Then fill all the tagged tips. 699 */ 700 for (; i < to_pack.nr_objects; i++) { 701 if (objects[i].tagged) 702 add_to_write_order(wo, wo_end, &objects[i]); 703 } 704 705 /* 706 * And then all remaining commits and tags. 707 */ 708 for (i = last_untagged; i < to_pack.nr_objects; i++) { 709 if (oe_type(&objects[i]) != OBJ_COMMIT && 710 oe_type(&objects[i]) != OBJ_TAG) 711 continue; 712 add_to_write_order(wo, wo_end, &objects[i]); 713 } 714 715 /* 716 * And then all the trees. 717 */ 718 for (i = last_untagged; i < to_pack.nr_objects; i++) { 719 if (oe_type(&objects[i]) != OBJ_TREE) 720 continue; 721 add_to_write_order(wo, wo_end, &objects[i]); 722 } 723 724 /* 725 * Finally all the rest in really tight order 726 */ 727 for (i = last_untagged; i < to_pack.nr_objects; i++) { 728 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer) 729 add_family_to_write_order(wo, wo_end, &objects[i]); 730 } 731} 732 733static struct object_entry **compute_write_order(void) 734{ 735 uint32_t max_layers = 1; 736 unsigned int i, wo_end; 737 738 struct object_entry **wo; 739 struct object_entry *objects = to_pack.objects; 740 741 for (i = 0; i < to_pack.nr_objects; i++) { 742 objects[i].tagged = 0; 743 objects[i].filled = 0; 744 SET_DELTA_CHILD(&objects[i], NULL); 745 SET_DELTA_SIBLING(&objects[i], NULL); 746 } 747 748 /* 749 * Fully connect delta_child/delta_sibling network. 750 * Make sure delta_sibling is sorted in the original 751 * recency order. 752 */ 753 for (i = to_pack.nr_objects; i > 0;) { 754 struct object_entry *e = &objects[--i]; 755 if (!DELTA(e)) 756 continue; 757 /* Mark me as the first child */ 758 e->delta_sibling_idx = DELTA(e)->delta_child_idx; 759 SET_DELTA_CHILD(DELTA(e), e); 760 } 761 762 /* 763 * Mark objects that are at the tip of tags. 764 */ 765 for_each_tag_ref(mark_tagged, NULL); 766 767 if (use_delta_islands) 768 max_layers = compute_pack_layers(&to_pack); 769 770 ALLOC_ARRAY(wo, to_pack.nr_objects); 771 wo_end = 0; 772 773 for (; write_layer < max_layers; ++write_layer) 774 compute_layer_order(wo, &wo_end); 775 776 if (wo_end != to_pack.nr_objects) 777 die(_("ordered %u objects, expected %"PRIu32), 778 wo_end, to_pack.nr_objects); 779 780 return wo; 781} 782 783static off_t write_reused_pack(struct hashfile *f) 784{ 785 unsigned char buffer[8192]; 786 off_t to_write, total; 787 int fd; 788 789 if (!is_pack_valid(reuse_packfile)) 790 die(_("packfile is invalid: %s"), reuse_packfile->pack_name); 791 792 fd = git_open(reuse_packfile->pack_name); 793 if (fd < 0) 794 die_errno(_("unable to open packfile for reuse: %s"), 795 reuse_packfile->pack_name); 796 797 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1) 798 die_errno(_("unable to seek in reused packfile")); 799 800 if (reuse_packfile_offset < 0) 801 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 802 803 total = to_write = reuse_packfile_offset - sizeof(struct pack_header); 804 805 while (to_write) { 806 int read_pack = xread(fd, buffer, sizeof(buffer)); 807 808 if (read_pack <= 0) 809 die_errno(_("unable to read from reused packfile")); 810 811 if (read_pack > to_write) 812 read_pack = to_write; 813 814 hashwrite(f, buffer, read_pack); 815 to_write -= read_pack; 816 817 /* 818 * We don't know the actual number of objects written, 819 * only how many bytes written, how many bytes total, and 820 * how many objects total. So we can fake it by pretending all 821 * objects we are writing are the same size. This gives us a 822 * smooth progress meter, and at the end it matches the true 823 * answer. 824 */ 825 written = reuse_packfile_objects * 826 (((double)(total - to_write)) / total); 827 display_progress(progress_state, written); 828 } 829 830 close(fd); 831 written = reuse_packfile_objects; 832 display_progress(progress_state, written); 833 return reuse_packfile_offset - sizeof(struct pack_header); 834} 835 836static const char no_split_warning[] = N_( 837"disabling bitmap writing, packs are split due to pack.packSizeLimit" 838); 839 840static void write_pack_file(void) 841{ 842 uint32_t i = 0, j; 843 struct hashfile *f; 844 off_t offset; 845 uint32_t nr_remaining = nr_result; 846 time_t last_mtime = 0; 847 struct object_entry **write_order; 848 849 if (progress > pack_to_stdout) 850 progress_state = start_progress(_("Writing objects"), nr_result); 851 ALLOC_ARRAY(written_list, to_pack.nr_objects); 852 write_order = compute_write_order(); 853 854 do { 855 struct object_id oid; 856 char *pack_tmp_name = NULL; 857 858 if (pack_to_stdout) 859 f = hashfd_throughput(1, "<stdout>", progress_state); 860 else 861 f = create_tmp_packfile(&pack_tmp_name); 862 863 offset = write_pack_header(f, nr_remaining); 864 865 if (reuse_packfile) { 866 off_t packfile_size; 867 assert(pack_to_stdout); 868 869 packfile_size = write_reused_pack(f); 870 offset += packfile_size; 871 } 872 873 nr_written = 0; 874 for (; i < to_pack.nr_objects; i++) { 875 struct object_entry *e = write_order[i]; 876 if (write_one(f, e, &offset) == WRITE_ONE_BREAK) 877 break; 878 display_progress(progress_state, written); 879 } 880 881 /* 882 * Did we write the wrong # entries in the header? 883 * If so, rewrite it like in fast-import 884 */ 885 if (pack_to_stdout) { 886 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 887 } else if (nr_written == nr_remaining) { 888 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 889 } else { 890 int fd = finalize_hashfile(f, oid.hash, 0); 891 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 892 nr_written, oid.hash, offset); 893 close(fd); 894 if (write_bitmap_index) { 895 warning(_(no_split_warning)); 896 write_bitmap_index = 0; 897 } 898 } 899 900 if (!pack_to_stdout) { 901 struct stat st; 902 struct strbuf tmpname = STRBUF_INIT; 903 904 /* 905 * Packs are runtime accessed in their mtime 906 * order since newer packs are more likely to contain 907 * younger objects. So if we are creating multiple 908 * packs then we should modify the mtime of later ones 909 * to preserve this property. 910 */ 911 if (stat(pack_tmp_name, &st) < 0) { 912 warning_errno(_("failed to stat %s"), pack_tmp_name); 913 } else if (!last_mtime) { 914 last_mtime = st.st_mtime; 915 } else { 916 struct utimbuf utb; 917 utb.actime = st.st_atime; 918 utb.modtime = --last_mtime; 919 if (utime(pack_tmp_name, &utb) < 0) 920 warning_errno(_("failed utime() on %s"), pack_tmp_name); 921 } 922 923 strbuf_addf(&tmpname, "%s-", base_name); 924 925 if (write_bitmap_index) { 926 bitmap_writer_set_checksum(oid.hash); 927 bitmap_writer_build_type_index( 928 &to_pack, written_list, nr_written); 929 } 930 931 finish_tmp_packfile(&tmpname, pack_tmp_name, 932 written_list, nr_written, 933 &pack_idx_opts, oid.hash); 934 935 if (write_bitmap_index) { 936 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid)); 937 938 stop_progress(&progress_state); 939 940 bitmap_writer_show_progress(progress); 941 bitmap_writer_reuse_bitmaps(&to_pack); 942 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 943 bitmap_writer_build(&to_pack); 944 bitmap_writer_finish(written_list, nr_written, 945 tmpname.buf, write_bitmap_options); 946 write_bitmap_index = 0; 947 } 948 949 strbuf_release(&tmpname); 950 free(pack_tmp_name); 951 puts(oid_to_hex(&oid)); 952 } 953 954 /* mark written objects as written to previous pack */ 955 for (j = 0; j < nr_written; j++) { 956 written_list[j]->offset = (off_t)-1; 957 } 958 nr_remaining -= nr_written; 959 } while (nr_remaining && i < to_pack.nr_objects); 960 961 free(written_list); 962 free(write_order); 963 stop_progress(&progress_state); 964 if (written != nr_result) 965 die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 966 written, nr_result); 967 trace2_data_intmax("pack-objects", the_repository, 968 "write_pack_file/wrote", nr_result); 969} 970 971static int no_try_delta(const char *path) 972{ 973 static struct attr_check *check; 974 975 if (!check) 976 check = attr_check_initl("delta", NULL); 977 git_check_attr(the_repository->index, path, check); 978 if (ATTR_FALSE(check->items[0].value)) 979 return 1; 980 return 0; 981} 982 983/* 984 * When adding an object, check whether we have already added it 985 * to our packing list. If so, we can skip. However, if we are 986 * being asked to excludei t, but the previous mention was to include 987 * it, make sure to adjust its flags and tweak our numbers accordingly. 988 * 989 * As an optimization, we pass out the index position where we would have 990 * found the item, since that saves us from having to look it up again a 991 * few lines later when we want to add the new entry. 992 */ 993static int have_duplicate_entry(const struct object_id *oid, 994 int exclude, 995 uint32_t *index_pos) 996{ 997 struct object_entry *entry; 998 999 entry = packlist_find(&to_pack, oid, index_pos);1000 if (!entry)1001 return 0;10021003 if (exclude) {1004 if (!entry->preferred_base)1005 nr_result--;1006 entry->preferred_base = 1;1007 }10081009 return 1;1010}10111012static int want_found_object(int exclude, struct packed_git *p)1013{1014 if (exclude)1015 return 1;1016 if (incremental)1017 return 0;10181019 /*1020 * When asked to do --local (do not include an object that appears in a1021 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1022 * an object that appears in a pack marked with .keep), finding a pack1023 * that matches the criteria is sufficient for us to decide to omit it.1024 * However, even if this pack does not satisfy the criteria, we need to1025 * make sure no copy of this object appears in _any_ pack that makes us1026 * to omit the object, so we need to check all the packs.1027 *1028 * We can however first check whether these options can possible matter;1029 * if they do not matter we know we want the object in generated pack.1030 * Otherwise, we signal "-1" at the end to tell the caller that we do1031 * not know either way, and it needs to check more packs.1032 */1033 if (!ignore_packed_keep_on_disk &&1034 !ignore_packed_keep_in_core &&1035 (!local || !have_non_local_packs))1036 return 1;10371038 if (local && !p->pack_local)1039 return 0;1040 if (p->pack_local &&1041 ((ignore_packed_keep_on_disk && p->pack_keep) ||1042 (ignore_packed_keep_in_core && p->pack_keep_in_core)))1043 return 0;10441045 /* we don't know yet; keep looking for more packs */1046 return -1;1047}10481049/*1050 * Check whether we want the object in the pack (e.g., we do not want1051 * objects found in non-local stores if the "--local" option was used).1052 *1053 * If the caller already knows an existing pack it wants to take the object1054 * from, that is passed in *found_pack and *found_offset; otherwise this1055 * function finds if there is any pack that has the object and returns the pack1056 * and its offset in these variables.1057 */1058static int want_object_in_pack(const struct object_id *oid,1059 int exclude,1060 struct packed_git **found_pack,1061 off_t *found_offset)1062{1063 int want;1064 struct list_head *pos;1065 struct multi_pack_index *m;10661067 if (!exclude && local && has_loose_object_nonlocal(oid))1068 return 0;10691070 /*1071 * If we already know the pack object lives in, start checks from that1072 * pack - in the usual case when neither --local was given nor .keep files1073 * are present we will determine the answer right now.1074 */1075 if (*found_pack) {1076 want = want_found_object(exclude, *found_pack);1077 if (want != -1)1078 return want;1079 }10801081 for (m = get_multi_pack_index(the_repository); m; m = m->next) {1082 struct pack_entry e;1083 if (fill_midx_entry(the_repository, oid, &e, m)) {1084 struct packed_git *p = e.p;1085 off_t offset;10861087 if (p == *found_pack)1088 offset = *found_offset;1089 else1090 offset = find_pack_entry_one(oid->hash, p);10911092 if (offset) {1093 if (!*found_pack) {1094 if (!is_pack_valid(p))1095 continue;1096 *found_offset = offset;1097 *found_pack = p;1098 }1099 want = want_found_object(exclude, p);1100 if (want != -1)1101 return want;1102 }1103 }1104 }11051106 list_for_each(pos, get_packed_git_mru(the_repository)) {1107 struct packed_git *p = list_entry(pos, struct packed_git, mru);1108 off_t offset;11091110 if (p == *found_pack)1111 offset = *found_offset;1112 else1113 offset = find_pack_entry_one(oid->hash, p);11141115 if (offset) {1116 if (!*found_pack) {1117 if (!is_pack_valid(p))1118 continue;1119 *found_offset = offset;1120 *found_pack = p;1121 }1122 want = want_found_object(exclude, p);1123 if (!exclude && want > 0)1124 list_move(&p->mru,1125 get_packed_git_mru(the_repository));1126 if (want != -1)1127 return want;1128 }1129 }11301131 return 1;1132}11331134static void create_object_entry(const struct object_id *oid,1135 enum object_type type,1136 uint32_t hash,1137 int exclude,1138 int no_try_delta,1139 uint32_t index_pos,1140 struct packed_git *found_pack,1141 off_t found_offset)1142{1143 struct object_entry *entry;11441145 entry = packlist_alloc(&to_pack, oid->hash, index_pos);1146 entry->hash = hash;1147 oe_set_type(entry, type);1148 if (exclude)1149 entry->preferred_base = 1;1150 else1151 nr_result++;1152 if (found_pack) {1153 oe_set_in_pack(&to_pack, entry, found_pack);1154 entry->in_pack_offset = found_offset;1155 }11561157 entry->no_try_delta = no_try_delta;1158}11591160static const char no_closure_warning[] = N_(1161"disabling bitmap writing, as some objects are not being packed"1162);11631164static int add_object_entry(const struct object_id *oid, enum object_type type,1165 const char *name, int exclude)1166{1167 struct packed_git *found_pack = NULL;1168 off_t found_offset = 0;1169 uint32_t index_pos;11701171 display_progress(progress_state, ++nr_seen);11721173 if (have_duplicate_entry(oid, exclude, &index_pos))1174 return 0;11751176 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1177 /* The pack is missing an object, so it will not have closure */1178 if (write_bitmap_index) {1179 warning(_(no_closure_warning));1180 write_bitmap_index = 0;1181 }1182 return 0;1183 }11841185 create_object_entry(oid, type, pack_name_hash(name),1186 exclude, name && no_try_delta(name),1187 index_pos, found_pack, found_offset);1188 return 1;1189}11901191static int add_object_entry_from_bitmap(const struct object_id *oid,1192 enum object_type type,1193 int flags, uint32_t name_hash,1194 struct packed_git *pack, off_t offset)1195{1196 uint32_t index_pos;11971198 display_progress(progress_state, ++nr_seen);11991200 if (have_duplicate_entry(oid, 0, &index_pos))1201 return 0;12021203 if (!want_object_in_pack(oid, 0, &pack, &offset))1204 return 0;12051206 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);1207 return 1;1208}12091210struct pbase_tree_cache {1211 struct object_id oid;1212 int ref;1213 int temporary;1214 void *tree_data;1215 unsigned long tree_size;1216};12171218static struct pbase_tree_cache *(pbase_tree_cache[256]);1219static int pbase_tree_cache_ix(const struct object_id *oid)1220{1221 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);1222}1223static int pbase_tree_cache_ix_incr(int ix)1224{1225 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);1226}12271228static struct pbase_tree {1229 struct pbase_tree *next;1230 /* This is a phony "cache" entry; we are not1231 * going to evict it or find it through _get()1232 * mechanism -- this is for the toplevel node that1233 * would almost always change with any commit.1234 */1235 struct pbase_tree_cache pcache;1236} *pbase_tree;12371238static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1239{1240 struct pbase_tree_cache *ent, *nent;1241 void *data;1242 unsigned long size;1243 enum object_type type;1244 int neigh;1245 int my_ix = pbase_tree_cache_ix(oid);1246 int available_ix = -1;12471248 /* pbase-tree-cache acts as a limited hashtable.1249 * your object will be found at your index or within a few1250 * slots after that slot if it is cached.1251 */1252 for (neigh = 0; neigh < 8; neigh++) {1253 ent = pbase_tree_cache[my_ix];1254 if (ent && oideq(&ent->oid, oid)) {1255 ent->ref++;1256 return ent;1257 }1258 else if (((available_ix < 0) && (!ent || !ent->ref)) ||1259 ((0 <= available_ix) &&1260 (!ent && pbase_tree_cache[available_ix])))1261 available_ix = my_ix;1262 if (!ent)1263 break;1264 my_ix = pbase_tree_cache_ix_incr(my_ix);1265 }12661267 /* Did not find one. Either we got a bogus request or1268 * we need to read and perhaps cache.1269 */1270 data = read_object_file(oid, &type, &size);1271 if (!data)1272 return NULL;1273 if (type != OBJ_TREE) {1274 free(data);1275 return NULL;1276 }12771278 /* We need to either cache or return a throwaway copy */12791280 if (available_ix < 0)1281 ent = NULL;1282 else {1283 ent = pbase_tree_cache[available_ix];1284 my_ix = available_ix;1285 }12861287 if (!ent) {1288 nent = xmalloc(sizeof(*nent));1289 nent->temporary = (available_ix < 0);1290 }1291 else {1292 /* evict and reuse */1293 free(ent->tree_data);1294 nent = ent;1295 }1296 oidcpy(&nent->oid, oid);1297 nent->tree_data = data;1298 nent->tree_size = size;1299 nent->ref = 1;1300 if (!nent->temporary)1301 pbase_tree_cache[my_ix] = nent;1302 return nent;1303}13041305static void pbase_tree_put(struct pbase_tree_cache *cache)1306{1307 if (!cache->temporary) {1308 cache->ref--;1309 return;1310 }1311 free(cache->tree_data);1312 free(cache);1313}13141315static int name_cmp_len(const char *name)1316{1317 int i;1318 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)1319 ;1320 return i;1321}13221323static void add_pbase_object(struct tree_desc *tree,1324 const char *name,1325 int cmplen,1326 const char *fullname)1327{1328 struct name_entry entry;1329 int cmp;13301331 while (tree_entry(tree,&entry)) {1332 if (S_ISGITLINK(entry.mode))1333 continue;1334 cmp = tree_entry_len(&entry) != cmplen ? 1 :1335 memcmp(name, entry.path, cmplen);1336 if (cmp > 0)1337 continue;1338 if (cmp < 0)1339 return;1340 if (name[cmplen] != '/') {1341 add_object_entry(&entry.oid,1342 object_type(entry.mode),1343 fullname, 1);1344 return;1345 }1346 if (S_ISDIR(entry.mode)) {1347 struct tree_desc sub;1348 struct pbase_tree_cache *tree;1349 const char *down = name+cmplen+1;1350 int downlen = name_cmp_len(down);13511352 tree = pbase_tree_get(&entry.oid);1353 if (!tree)1354 return;1355 init_tree_desc(&sub, tree->tree_data, tree->tree_size);13561357 add_pbase_object(&sub, down, downlen, fullname);1358 pbase_tree_put(tree);1359 }1360 }1361}13621363static unsigned *done_pbase_paths;1364static int done_pbase_paths_num;1365static int done_pbase_paths_alloc;1366static int done_pbase_path_pos(unsigned hash)1367{1368 int lo = 0;1369 int hi = done_pbase_paths_num;1370 while (lo < hi) {1371 int mi = lo + (hi - lo) / 2;1372 if (done_pbase_paths[mi] == hash)1373 return mi;1374 if (done_pbase_paths[mi] < hash)1375 hi = mi;1376 else1377 lo = mi + 1;1378 }1379 return -lo-1;1380}13811382static int check_pbase_path(unsigned hash)1383{1384 int pos = done_pbase_path_pos(hash);1385 if (0 <= pos)1386 return 1;1387 pos = -pos - 1;1388 ALLOC_GROW(done_pbase_paths,1389 done_pbase_paths_num + 1,1390 done_pbase_paths_alloc);1391 done_pbase_paths_num++;1392 if (pos < done_pbase_paths_num)1393 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,1394 done_pbase_paths_num - pos - 1);1395 done_pbase_paths[pos] = hash;1396 return 0;1397}13981399static void add_preferred_base_object(const char *name)1400{1401 struct pbase_tree *it;1402 int cmplen;1403 unsigned hash = pack_name_hash(name);14041405 if (!num_preferred_base || check_pbase_path(hash))1406 return;14071408 cmplen = name_cmp_len(name);1409 for (it = pbase_tree; it; it = it->next) {1410 if (cmplen == 0) {1411 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);1412 }1413 else {1414 struct tree_desc tree;1415 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1416 add_pbase_object(&tree, name, cmplen, name);1417 }1418 }1419}14201421static void add_preferred_base(struct object_id *oid)1422{1423 struct pbase_tree *it;1424 void *data;1425 unsigned long size;1426 struct object_id tree_oid;14271428 if (window <= num_preferred_base++)1429 return;14301431 data = read_object_with_reference(the_repository, oid,1432 tree_type, &size, &tree_oid);1433 if (!data)1434 return;14351436 for (it = pbase_tree; it; it = it->next) {1437 if (oideq(&it->pcache.oid, &tree_oid)) {1438 free(data);1439 return;1440 }1441 }14421443 it = xcalloc(1, sizeof(*it));1444 it->next = pbase_tree;1445 pbase_tree = it;14461447 oidcpy(&it->pcache.oid, &tree_oid);1448 it->pcache.tree_data = data;1449 it->pcache.tree_size = size;1450}14511452static void cleanup_preferred_base(void)1453{1454 struct pbase_tree *it;1455 unsigned i;14561457 it = pbase_tree;1458 pbase_tree = NULL;1459 while (it) {1460 struct pbase_tree *tmp = it;1461 it = tmp->next;1462 free(tmp->pcache.tree_data);1463 free(tmp);1464 }14651466 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {1467 if (!pbase_tree_cache[i])1468 continue;1469 free(pbase_tree_cache[i]->tree_data);1470 FREE_AND_NULL(pbase_tree_cache[i]);1471 }14721473 FREE_AND_NULL(done_pbase_paths);1474 done_pbase_paths_num = done_pbase_paths_alloc = 0;1475}14761477/*1478 * Return 1 iff the object specified by "delta" can be sent1479 * literally as a delta against the base in "base_sha1". If1480 * so, then *base_out will point to the entry in our packing1481 * list, or NULL if we must use the external-base list.1482 *1483 * Depth value does not matter - find_deltas() will1484 * never consider reused delta as the base object to1485 * deltify other objects against, in order to avoid1486 * circular deltas.1487 */1488static int can_reuse_delta(const unsigned char *base_sha1,1489 struct object_entry *delta,1490 struct object_entry **base_out)1491{1492 struct object_entry *base;1493 struct object_id base_oid;14941495 if (!base_sha1)1496 return 0;14971498 oidread(&base_oid, base_sha1);14991500 /*1501 * First see if we're already sending the base (or it's explicitly in1502 * our "excluded" list).1503 */1504 base = packlist_find(&to_pack, &base_oid, NULL);1505 if (base) {1506 if (!in_same_island(&delta->idx.oid, &base->idx.oid))1507 return 0;1508 *base_out = base;1509 return 1;1510 }15111512 /*1513 * Otherwise, reachability bitmaps may tell us if the receiver has it,1514 * even if it was buried too deep in history to make it into the1515 * packing list.1516 */1517 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {1518 if (use_delta_islands) {1519 if (!in_same_island(&delta->idx.oid, &base_oid))1520 return 0;1521 }1522 *base_out = NULL;1523 return 1;1524 }15251526 return 0;1527}15281529static void check_object(struct object_entry *entry)1530{1531 unsigned long canonical_size;15321533 if (IN_PACK(entry)) {1534 struct packed_git *p = IN_PACK(entry);1535 struct pack_window *w_curs = NULL;1536 const unsigned char *base_ref = NULL;1537 struct object_entry *base_entry;1538 unsigned long used, used_0;1539 unsigned long avail;1540 off_t ofs;1541 unsigned char *buf, c;1542 enum object_type type;1543 unsigned long in_pack_size;15441545 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);15461547 /*1548 * We want in_pack_type even if we do not reuse delta1549 * since non-delta representations could still be reused.1550 */1551 used = unpack_object_header_buffer(buf, avail,1552 &type,1553 &in_pack_size);1554 if (used == 0)1555 goto give_up;15561557 if (type < 0)1558 BUG("invalid type %d", type);1559 entry->in_pack_type = type;15601561 /*1562 * Determine if this is a delta and if so whether we can1563 * reuse it or not. Otherwise let's find out as cheaply as1564 * possible what the actual type and size for this object is.1565 */1566 switch (entry->in_pack_type) {1567 default:1568 /* Not a delta hence we've already got all we need. */1569 oe_set_type(entry, entry->in_pack_type);1570 SET_SIZE(entry, in_pack_size);1571 entry->in_pack_header_size = used;1572 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)1573 goto give_up;1574 unuse_pack(&w_curs);1575 return;1576 case OBJ_REF_DELTA:1577 if (reuse_delta && !entry->preferred_base)1578 base_ref = use_pack(p, &w_curs,1579 entry->in_pack_offset + used, NULL);1580 entry->in_pack_header_size = used + the_hash_algo->rawsz;1581 break;1582 case OBJ_OFS_DELTA:1583 buf = use_pack(p, &w_curs,1584 entry->in_pack_offset + used, NULL);1585 used_0 = 0;1586 c = buf[used_0++];1587 ofs = c & 127;1588 while (c & 128) {1589 ofs += 1;1590 if (!ofs || MSB(ofs, 7)) {1591 error(_("delta base offset overflow in pack for %s"),1592 oid_to_hex(&entry->idx.oid));1593 goto give_up;1594 }1595 c = buf[used_0++];1596 ofs = (ofs << 7) + (c & 127);1597 }1598 ofs = entry->in_pack_offset - ofs;1599 if (ofs <= 0 || ofs >= entry->in_pack_offset) {1600 error(_("delta base offset out of bound for %s"),1601 oid_to_hex(&entry->idx.oid));1602 goto give_up;1603 }1604 if (reuse_delta && !entry->preferred_base) {1605 struct revindex_entry *revidx;1606 revidx = find_pack_revindex(p, ofs);1607 if (!revidx)1608 goto give_up;1609 base_ref = nth_packed_object_sha1(p, revidx->nr);1610 }1611 entry->in_pack_header_size = used + used_0;1612 break;1613 }16141615 if (can_reuse_delta(base_ref, entry, &base_entry)) {1616 oe_set_type(entry, entry->in_pack_type);1617 SET_SIZE(entry, in_pack_size); /* delta size */1618 SET_DELTA_SIZE(entry, in_pack_size);16191620 if (base_entry) {1621 SET_DELTA(entry, base_entry);1622 entry->delta_sibling_idx = base_entry->delta_child_idx;1623 SET_DELTA_CHILD(base_entry, entry);1624 } else {1625 SET_DELTA_EXT(entry, base_ref);1626 }16271628 unuse_pack(&w_curs);1629 return;1630 }16311632 if (oe_type(entry)) {1633 off_t delta_pos;16341635 /*1636 * This must be a delta and we already know what the1637 * final object type is. Let's extract the actual1638 * object size from the delta header.1639 */1640 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1641 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);1642 if (canonical_size == 0)1643 goto give_up;1644 SET_SIZE(entry, canonical_size);1645 unuse_pack(&w_curs);1646 return;1647 }16481649 /*1650 * No choice but to fall back to the recursive delta walk1651 * with oid_object_info() to find about the object type1652 * at this point...1653 */1654 give_up:1655 unuse_pack(&w_curs);1656 }16571658 oe_set_type(entry,1659 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1660 if (entry->type_valid) {1661 SET_SIZE(entry, canonical_size);1662 } else {1663 /*1664 * Bad object type is checked in prepare_pack(). This is1665 * to permit a missing preferred base object to be ignored1666 * as a preferred base. Doing so can result in a larger1667 * pack file, but the transfer will still take place.1668 */1669 }1670}16711672static int pack_offset_sort(const void *_a, const void *_b)1673{1674 const struct object_entry *a = *(struct object_entry **)_a;1675 const struct object_entry *b = *(struct object_entry **)_b;1676 const struct packed_git *a_in_pack = IN_PACK(a);1677 const struct packed_git *b_in_pack = IN_PACK(b);16781679 /* avoid filesystem trashing with loose objects */1680 if (!a_in_pack && !b_in_pack)1681 return oidcmp(&a->idx.oid, &b->idx.oid);16821683 if (a_in_pack < b_in_pack)1684 return -1;1685 if (a_in_pack > b_in_pack)1686 return 1;1687 return a->in_pack_offset < b->in_pack_offset ? -1 :1688 (a->in_pack_offset > b->in_pack_offset);1689}16901691/*1692 * Drop an on-disk delta we were planning to reuse. Naively, this would1693 * just involve blanking out the "delta" field, but we have to deal1694 * with some extra book-keeping:1695 *1696 * 1. Removing ourselves from the delta_sibling linked list.1697 *1698 * 2. Updating our size/type to the non-delta representation. These were1699 * either not recorded initially (size) or overwritten with the delta type1700 * (type) when check_object() decided to reuse the delta.1701 *1702 * 3. Resetting our delta depth, as we are now a base object.1703 */1704static void drop_reused_delta(struct object_entry *entry)1705{1706 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;1707 struct object_info oi = OBJECT_INFO_INIT;1708 enum object_type type;1709 unsigned long size;17101711 while (*idx) {1712 struct object_entry *oe = &to_pack.objects[*idx - 1];17131714 if (oe == entry)1715 *idx = oe->delta_sibling_idx;1716 else1717 idx = &oe->delta_sibling_idx;1718 }1719 SET_DELTA(entry, NULL);1720 entry->depth = 0;17211722 oi.sizep = &size;1723 oi.typep = &type;1724 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {1725 /*1726 * We failed to get the info from this pack for some reason;1727 * fall back to oid_object_info, which may find another copy.1728 * And if that fails, the error will be recorded in oe_type(entry)1729 * and dealt with in prepare_pack().1730 */1731 oe_set_type(entry,1732 oid_object_info(the_repository, &entry->idx.oid, &size));1733 } else {1734 oe_set_type(entry, type);1735 }1736 SET_SIZE(entry, size);1737}17381739/*1740 * Follow the chain of deltas from this entry onward, throwing away any links1741 * that cause us to hit a cycle (as determined by the DFS state flags in1742 * the entries).1743 *1744 * We also detect too-long reused chains that would violate our --depth1745 * limit.1746 */1747static void break_delta_chains(struct object_entry *entry)1748{1749 /*1750 * The actual depth of each object we will write is stored as an int,1751 * as it cannot exceed our int "depth" limit. But before we break1752 * changes based no that limit, we may potentially go as deep as the1753 * number of objects, which is elsewhere bounded to a uint32_t.1754 */1755 uint32_t total_depth;1756 struct object_entry *cur, *next;17571758 for (cur = entry, total_depth = 0;1759 cur;1760 cur = DELTA(cur), total_depth++) {1761 if (cur->dfs_state == DFS_DONE) {1762 /*1763 * We've already seen this object and know it isn't1764 * part of a cycle. We do need to append its depth1765 * to our count.1766 */1767 total_depth += cur->depth;1768 break;1769 }17701771 /*1772 * We break cycles before looping, so an ACTIVE state (or any1773 * other cruft which made its way into the state variable)1774 * is a bug.1775 */1776 if (cur->dfs_state != DFS_NONE)1777 BUG("confusing delta dfs state in first pass: %d",1778 cur->dfs_state);17791780 /*1781 * Now we know this is the first time we've seen the object. If1782 * it's not a delta, we're done traversing, but we'll mark it1783 * done to save time on future traversals.1784 */1785 if (!DELTA(cur)) {1786 cur->dfs_state = DFS_DONE;1787 break;1788 }17891790 /*1791 * Mark ourselves as active and see if the next step causes1792 * us to cycle to another active object. It's important to do1793 * this _before_ we loop, because it impacts where we make the1794 * cut, and thus how our total_depth counter works.1795 * E.g., We may see a partial loop like:1796 *1797 * A -> B -> C -> D -> B1798 *1799 * Cutting B->C breaks the cycle. But now the depth of A is1800 * only 1, and our total_depth counter is at 3. The size of the1801 * error is always one less than the size of the cycle we1802 * broke. Commits C and D were "lost" from A's chain.1803 *1804 * If we instead cut D->B, then the depth of A is correct at 3.1805 * We keep all commits in the chain that we examined.1806 */1807 cur->dfs_state = DFS_ACTIVE;1808 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {1809 drop_reused_delta(cur);1810 cur->dfs_state = DFS_DONE;1811 break;1812 }1813 }18141815 /*1816 * And now that we've gone all the way to the bottom of the chain, we1817 * need to clear the active flags and set the depth fields as1818 * appropriate. Unlike the loop above, which can quit when it drops a1819 * delta, we need to keep going to look for more depth cuts. So we need1820 * an extra "next" pointer to keep going after we reset cur->delta.1821 */1822 for (cur = entry; cur; cur = next) {1823 next = DELTA(cur);18241825 /*1826 * We should have a chain of zero or more ACTIVE states down to1827 * a final DONE. We can quit after the DONE, because either it1828 * has no bases, or we've already handled them in a previous1829 * call.1830 */1831 if (cur->dfs_state == DFS_DONE)1832 break;1833 else if (cur->dfs_state != DFS_ACTIVE)1834 BUG("confusing delta dfs state in second pass: %d",1835 cur->dfs_state);18361837 /*1838 * If the total_depth is more than depth, then we need to snip1839 * the chain into two or more smaller chains that don't exceed1840 * the maximum depth. Most of the resulting chains will contain1841 * (depth + 1) entries (i.e., depth deltas plus one base), and1842 * the last chain (i.e., the one containing entry) will contain1843 * whatever entries are left over, namely1844 * (total_depth % (depth + 1)) of them.1845 *1846 * Since we are iterating towards decreasing depth, we need to1847 * decrement total_depth as we go, and we need to write to the1848 * entry what its final depth will be after all of the1849 * snipping. Since we're snipping into chains of length (depth1850 * + 1) entries, the final depth of an entry will be its1851 * original depth modulo (depth + 1). Any time we encounter an1852 * entry whose final depth is supposed to be zero, we snip it1853 * from its delta base, thereby making it so.1854 */1855 cur->depth = (total_depth--) % (depth + 1);1856 if (!cur->depth)1857 drop_reused_delta(cur);18581859 cur->dfs_state = DFS_DONE;1860 }1861}18621863static void get_object_details(void)1864{1865 uint32_t i;1866 struct object_entry **sorted_by_offset;18671868 if (progress)1869 progress_state = start_progress(_("Counting objects"),1870 to_pack.nr_objects);18711872 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));1873 for (i = 0; i < to_pack.nr_objects; i++)1874 sorted_by_offset[i] = to_pack.objects + i;1875 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18761877 for (i = 0; i < to_pack.nr_objects; i++) {1878 struct object_entry *entry = sorted_by_offset[i];1879 check_object(entry);1880 if (entry->type_valid &&1881 oe_size_greater_than(&to_pack, entry, big_file_threshold))1882 entry->no_try_delta = 1;1883 display_progress(progress_state, i + 1);1884 }1885 stop_progress(&progress_state);18861887 /*1888 * This must happen in a second pass, since we rely on the delta1889 * information for the whole list being completed.1890 */1891 for (i = 0; i < to_pack.nr_objects; i++)1892 break_delta_chains(&to_pack.objects[i]);18931894 free(sorted_by_offset);1895}18961897/*1898 * We search for deltas in a list sorted by type, by filename hash, and then1899 * by size, so that we see progressively smaller and smaller files.1900 * That's because we prefer deltas to be from the bigger file1901 * to the smaller -- deletes are potentially cheaper, but perhaps1902 * more importantly, the bigger file is likely the more recent1903 * one. The deepest deltas are therefore the oldest objects which are1904 * less susceptible to be accessed often.1905 */1906static int type_size_sort(const void *_a, const void *_b)1907{1908 const struct object_entry *a = *(struct object_entry **)_a;1909 const struct object_entry *b = *(struct object_entry **)_b;1910 const enum object_type a_type = oe_type(a);1911 const enum object_type b_type = oe_type(b);1912 const unsigned long a_size = SIZE(a);1913 const unsigned long b_size = SIZE(b);19141915 if (a_type > b_type)1916 return -1;1917 if (a_type < b_type)1918 return 1;1919 if (a->hash > b->hash)1920 return -1;1921 if (a->hash < b->hash)1922 return 1;1923 if (a->preferred_base > b->preferred_base)1924 return -1;1925 if (a->preferred_base < b->preferred_base)1926 return 1;1927 if (use_delta_islands) {1928 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);1929 if (island_cmp)1930 return island_cmp;1931 }1932 if (a_size > b_size)1933 return -1;1934 if (a_size < b_size)1935 return 1;1936 return a < b ? -1 : (a > b); /* newest first */1937}19381939struct unpacked {1940 struct object_entry *entry;1941 void *data;1942 struct delta_index *index;1943 unsigned depth;1944};19451946static int delta_cacheable(unsigned long src_size, unsigned long trg_size,1947 unsigned long delta_size)1948{1949 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1950 return 0;19511952 if (delta_size < cache_max_small_delta_size)1953 return 1;19541955 /* cache delta, if objects are large enough compared to delta size */1956 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))1957 return 1;19581959 return 0;1960}19611962/* Protect delta_cache_size */1963static pthread_mutex_t cache_mutex;1964#define cache_lock() pthread_mutex_lock(&cache_mutex)1965#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19661967/*1968 * Protect object list partitioning (e.g. struct thread_param) and1969 * progress_state1970 */1971static pthread_mutex_t progress_mutex;1972#define progress_lock() pthread_mutex_lock(&progress_mutex)1973#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19741975/*1976 * Access to struct object_entry is unprotected since each thread owns1977 * a portion of the main object list. Just don't access object entries1978 * ahead in the list because they can be stolen and would need1979 * progress_mutex for protection.1980 */19811982/*1983 * Return the size of the object without doing any delta1984 * reconstruction (so non-deltas are true object sizes, but deltas1985 * return the size of the delta data).1986 */1987unsigned long oe_get_size_slow(struct packing_data *pack,1988 const struct object_entry *e)1989{1990 struct packed_git *p;1991 struct pack_window *w_curs;1992 unsigned char *buf;1993 enum object_type type;1994 unsigned long used, avail, size;19951996 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1997 packing_data_lock(&to_pack);1998 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)1999 die(_("unable to get size of %s"),2000 oid_to_hex(&e->idx.oid));2001 packing_data_unlock(&to_pack);2002 return size;2003 }20042005 p = oe_in_pack(pack, e);2006 if (!p)2007 BUG("when e->type is a delta, it must belong to a pack");20082009 packing_data_lock(&to_pack);2010 w_curs = NULL;2011 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);2012 used = unpack_object_header_buffer(buf, avail, &type, &size);2013 if (used == 0)2014 die(_("unable to parse object header of %s"),2015 oid_to_hex(&e->idx.oid));20162017 unuse_pack(&w_curs);2018 packing_data_unlock(&to_pack);2019 return size;2020}20212022static int try_delta(struct unpacked *trg, struct unpacked *src,2023 unsigned max_depth, unsigned long *mem_usage)2024{2025 struct object_entry *trg_entry = trg->entry;2026 struct object_entry *src_entry = src->entry;2027 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2028 unsigned ref_depth;2029 enum object_type type;2030 void *delta_buf;20312032 /* Don't bother doing diffs between different types */2033 if (oe_type(trg_entry) != oe_type(src_entry))2034 return -1;20352036 /*2037 * We do not bother to try a delta that we discarded on an2038 * earlier try, but only when reusing delta data. Note that2039 * src_entry that is marked as the preferred_base should always2040 * be considered, as even if we produce a suboptimal delta against2041 * it, we will still save the transfer cost, as we already know2042 * the other side has it and we won't send src_entry at all.2043 */2044 if (reuse_delta && IN_PACK(trg_entry) &&2045 IN_PACK(trg_entry) == IN_PACK(src_entry) &&2046 !src_entry->preferred_base &&2047 trg_entry->in_pack_type != OBJ_REF_DELTA &&2048 trg_entry->in_pack_type != OBJ_OFS_DELTA)2049 return 0;20502051 /* Let's not bust the allowed depth. */2052 if (src->depth >= max_depth)2053 return 0;20542055 /* Now some size filtering heuristics. */2056 trg_size = SIZE(trg_entry);2057 if (!DELTA(trg_entry)) {2058 max_size = trg_size/2 - the_hash_algo->rawsz;2059 ref_depth = 1;2060 } else {2061 max_size = DELTA_SIZE(trg_entry);2062 ref_depth = trg->depth;2063 }2064 max_size = (uint64_t)max_size * (max_depth - src->depth) /2065 (max_depth - ref_depth + 1);2066 if (max_size == 0)2067 return 0;2068 src_size = SIZE(src_entry);2069 sizediff = src_size < trg_size ? trg_size - src_size : 0;2070 if (sizediff >= max_size)2071 return 0;2072 if (trg_size < src_size / 32)2073 return 0;20742075 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2076 return 0;20772078 /* Load data if not already done */2079 if (!trg->data) {2080 packing_data_lock(&to_pack);2081 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);2082 packing_data_unlock(&to_pack);2083 if (!trg->data)2084 die(_("object %s cannot be read"),2085 oid_to_hex(&trg_entry->idx.oid));2086 if (sz != trg_size)2087 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2088 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2089 (uintmax_t)trg_size);2090 *mem_usage += sz;2091 }2092 if (!src->data) {2093 packing_data_lock(&to_pack);2094 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);2095 packing_data_unlock(&to_pack);2096 if (!src->data) {2097 if (src_entry->preferred_base) {2098 static int warned = 0;2099 if (!warned++)2100 warning(_("object %s cannot be read"),2101 oid_to_hex(&src_entry->idx.oid));2102 /*2103 * Those objects are not included in the2104 * resulting pack. Be resilient and ignore2105 * them if they can't be read, in case the2106 * pack could be created nevertheless.2107 */2108 return 0;2109 }2110 die(_("object %s cannot be read"),2111 oid_to_hex(&src_entry->idx.oid));2112 }2113 if (sz != src_size)2114 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2115 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2116 (uintmax_t)src_size);2117 *mem_usage += sz;2118 }2119 if (!src->index) {2120 src->index = create_delta_index(src->data, src_size);2121 if (!src->index) {2122 static int warned = 0;2123 if (!warned++)2124 warning(_("suboptimal pack - out of memory"));2125 return 0;2126 }2127 *mem_usage += sizeof_delta_index(src->index);2128 }21292130 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2131 if (!delta_buf)2132 return 0;21332134 if (DELTA(trg_entry)) {2135 /* Prefer only shallower same-sized deltas. */2136 if (delta_size == DELTA_SIZE(trg_entry) &&2137 src->depth + 1 >= trg->depth) {2138 free(delta_buf);2139 return 0;2140 }2141 }21422143 /*2144 * Handle memory allocation outside of the cache2145 * accounting lock. Compiler will optimize the strangeness2146 * away when NO_PTHREADS is defined.2147 */2148 free(trg_entry->delta_data);2149 cache_lock();2150 if (trg_entry->delta_data) {2151 delta_cache_size -= DELTA_SIZE(trg_entry);2152 trg_entry->delta_data = NULL;2153 }2154 if (delta_cacheable(src_size, trg_size, delta_size)) {2155 delta_cache_size += delta_size;2156 cache_unlock();2157 trg_entry->delta_data = xrealloc(delta_buf, delta_size);2158 } else {2159 cache_unlock();2160 free(delta_buf);2161 }21622163 SET_DELTA(trg_entry, src_entry);2164 SET_DELTA_SIZE(trg_entry, delta_size);2165 trg->depth = src->depth + 1;21662167 return 1;2168}21692170static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)2171{2172 struct object_entry *child = DELTA_CHILD(me);2173 unsigned int m = n;2174 while (child) {2175 const unsigned int c = check_delta_limit(child, n + 1);2176 if (m < c)2177 m = c;2178 child = DELTA_SIBLING(child);2179 }2180 return m;2181}21822183static unsigned long free_unpacked(struct unpacked *n)2184{2185 unsigned long freed_mem = sizeof_delta_index(n->index);2186 free_delta_index(n->index);2187 n->index = NULL;2188 if (n->data) {2189 freed_mem += SIZE(n->entry);2190 FREE_AND_NULL(n->data);2191 }2192 n->entry = NULL;2193 n->depth = 0;2194 return freed_mem;2195}21962197static void find_deltas(struct object_entry **list, unsigned *list_size,2198 int window, int depth, unsigned *processed)2199{2200 uint32_t i, idx = 0, count = 0;2201 struct unpacked *array;2202 unsigned long mem_usage = 0;22032204 array = xcalloc(window, sizeof(struct unpacked));22052206 for (;;) {2207 struct object_entry *entry;2208 struct unpacked *n = array + idx;2209 int j, max_depth, best_base = -1;22102211 progress_lock();2212 if (!*list_size) {2213 progress_unlock();2214 break;2215 }2216 entry = *list++;2217 (*list_size)--;2218 if (!entry->preferred_base) {2219 (*processed)++;2220 display_progress(progress_state, *processed);2221 }2222 progress_unlock();22232224 mem_usage -= free_unpacked(n);2225 n->entry = entry;22262227 while (window_memory_limit &&2228 mem_usage > window_memory_limit &&2229 count > 1) {2230 const uint32_t tail = (idx + window - count) % window;2231 mem_usage -= free_unpacked(array + tail);2232 count--;2233 }22342235 /* We do not compute delta to *create* objects we are not2236 * going to pack.2237 */2238 if (entry->preferred_base)2239 goto next;22402241 /*2242 * If the current object is at pack edge, take the depth the2243 * objects that depend on the current object into account2244 * otherwise they would become too deep.2245 */2246 max_depth = depth;2247 if (DELTA_CHILD(entry)) {2248 max_depth -= check_delta_limit(entry, 0);2249 if (max_depth <= 0)2250 goto next;2251 }22522253 j = window;2254 while (--j > 0) {2255 int ret;2256 uint32_t other_idx = idx + j;2257 struct unpacked *m;2258 if (other_idx >= window)2259 other_idx -= window;2260 m = array + other_idx;2261 if (!m->entry)2262 break;2263 ret = try_delta(n, m, max_depth, &mem_usage);2264 if (ret < 0)2265 break;2266 else if (ret > 0)2267 best_base = other_idx;2268 }22692270 /*2271 * If we decided to cache the delta data, then it is best2272 * to compress it right away. First because we have to do2273 * it anyway, and doing it here while we're threaded will2274 * save a lot of time in the non threaded write phase,2275 * as well as allow for caching more deltas within2276 * the same cache size limit.2277 * ...2278 * But only if not writing to stdout, since in that case2279 * the network is most likely throttling writes anyway,2280 * and therefore it is best to go to the write phase ASAP2281 * instead, as we can afford spending more time compressing2282 * between writes at that moment.2283 */2284 if (entry->delta_data && !pack_to_stdout) {2285 unsigned long size;22862287 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));2288 if (size < (1U << OE_Z_DELTA_BITS)) {2289 entry->z_delta_size = size;2290 cache_lock();2291 delta_cache_size -= DELTA_SIZE(entry);2292 delta_cache_size += entry->z_delta_size;2293 cache_unlock();2294 } else {2295 FREE_AND_NULL(entry->delta_data);2296 entry->z_delta_size = 0;2297 }2298 }22992300 /* if we made n a delta, and if n is already at max2301 * depth, leaving it in the window is pointless. we2302 * should evict it first.2303 */2304 if (DELTA(entry) && max_depth <= n->depth)2305 continue;23062307 /*2308 * Move the best delta base up in the window, after the2309 * currently deltified object, to keep it longer. It will2310 * be the first base object to be attempted next.2311 */2312 if (DELTA(entry)) {2313 struct unpacked swap = array[best_base];2314 int dist = (window + idx - best_base) % window;2315 int dst = best_base;2316 while (dist--) {2317 int src = (dst + 1) % window;2318 array[dst] = array[src];2319 dst = src;2320 }2321 array[dst] = swap;2322 }23232324 next:2325 idx++;2326 if (count + 1 < window)2327 count++;2328 if (idx >= window)2329 idx = 0;2330 }23312332 for (i = 0; i < window; ++i) {2333 free_delta_index(array[i].index);2334 free(array[i].data);2335 }2336 free(array);2337}23382339static void try_to_free_from_threads(size_t size)2340{2341 packing_data_lock(&to_pack);2342 release_pack_memory(size);2343 packing_data_unlock(&to_pack);2344}23452346static try_to_free_t old_try_to_free_routine;23472348/*2349 * The main object list is split into smaller lists, each is handed to2350 * one worker.2351 *2352 * The main thread waits on the condition that (at least) one of the workers2353 * has stopped working (which is indicated in the .working member of2354 * struct thread_params).2355 *2356 * When a work thread has completed its work, it sets .working to 0 and2357 * signals the main thread and waits on the condition that .data_ready2358 * becomes 1.2359 *2360 * The main thread steals half of the work from the worker that has2361 * most work left to hand it to the idle worker.2362 */23632364struct thread_params {2365 pthread_t thread;2366 struct object_entry **list;2367 unsigned list_size;2368 unsigned remaining;2369 int window;2370 int depth;2371 int working;2372 int data_ready;2373 pthread_mutex_t mutex;2374 pthread_cond_t cond;2375 unsigned *processed;2376};23772378static pthread_cond_t progress_cond;23792380/*2381 * Mutex and conditional variable can't be statically-initialized on Windows.2382 */2383static void init_threaded_search(void)2384{2385 pthread_mutex_init(&cache_mutex, NULL);2386 pthread_mutex_init(&progress_mutex, NULL);2387 pthread_cond_init(&progress_cond, NULL);2388 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);2389}23902391static void cleanup_threaded_search(void)2392{2393 set_try_to_free_routine(old_try_to_free_routine);2394 pthread_cond_destroy(&progress_cond);2395 pthread_mutex_destroy(&cache_mutex);2396 pthread_mutex_destroy(&progress_mutex);2397}23982399static void *threaded_find_deltas(void *arg)2400{2401 struct thread_params *me = arg;24022403 progress_lock();2404 while (me->remaining) {2405 progress_unlock();24062407 find_deltas(me->list, &me->remaining,2408 me->window, me->depth, me->processed);24092410 progress_lock();2411 me->working = 0;2412 pthread_cond_signal(&progress_cond);2413 progress_unlock();24142415 /*2416 * We must not set ->data_ready before we wait on the2417 * condition because the main thread may have set it to 12418 * before we get here. In order to be sure that new2419 * work is available if we see 1 in ->data_ready, it2420 * was initialized to 0 before this thread was spawned2421 * and we reset it to 0 right away.2422 */2423 pthread_mutex_lock(&me->mutex);2424 while (!me->data_ready)2425 pthread_cond_wait(&me->cond, &me->mutex);2426 me->data_ready = 0;2427 pthread_mutex_unlock(&me->mutex);24282429 progress_lock();2430 }2431 progress_unlock();2432 /* leave ->working 1 so that this doesn't get more work assigned */2433 return NULL;2434}24352436static void ll_find_deltas(struct object_entry **list, unsigned list_size,2437 int window, int depth, unsigned *processed)2438{2439 struct thread_params *p;2440 int i, ret, active_threads = 0;24412442 init_threaded_search();24432444 if (delta_search_threads <= 1) {2445 find_deltas(list, &list_size, window, depth, processed);2446 cleanup_threaded_search();2447 return;2448 }2449 if (progress > pack_to_stdout)2450 fprintf_ln(stderr, _("Delta compression using up to %d threads"),2451 delta_search_threads);2452 p = xcalloc(delta_search_threads, sizeof(*p));24532454 /* Partition the work amongst work threads. */2455 for (i = 0; i < delta_search_threads; i++) {2456 unsigned sub_size = list_size / (delta_search_threads - i);24572458 /* don't use too small segments or no deltas will be found */2459 if (sub_size < 2*window && i+1 < delta_search_threads)2460 sub_size = 0;24612462 p[i].window = window;2463 p[i].depth = depth;2464 p[i].processed = processed;2465 p[i].working = 1;2466 p[i].data_ready = 0;24672468 /* try to split chunks on "path" boundaries */2469 while (sub_size && sub_size < list_size &&2470 list[sub_size]->hash &&2471 list[sub_size]->hash == list[sub_size-1]->hash)2472 sub_size++;24732474 p[i].list = list;2475 p[i].list_size = sub_size;2476 p[i].remaining = sub_size;24772478 list += sub_size;2479 list_size -= sub_size;2480 }24812482 /* Start work threads. */2483 for (i = 0; i < delta_search_threads; i++) {2484 if (!p[i].list_size)2485 continue;2486 pthread_mutex_init(&p[i].mutex, NULL);2487 pthread_cond_init(&p[i].cond, NULL);2488 ret = pthread_create(&p[i].thread, NULL,2489 threaded_find_deltas, &p[i]);2490 if (ret)2491 die(_("unable to create thread: %s"), strerror(ret));2492 active_threads++;2493 }24942495 /*2496 * Now let's wait for work completion. Each time a thread is done2497 * with its work, we steal half of the remaining work from the2498 * thread with the largest number of unprocessed objects and give2499 * it to that newly idle thread. This ensure good load balancing2500 * until the remaining object list segments are simply too short2501 * to be worth splitting anymore.2502 */2503 while (active_threads) {2504 struct thread_params *target = NULL;2505 struct thread_params *victim = NULL;2506 unsigned sub_size = 0;25072508 progress_lock();2509 for (;;) {2510 for (i = 0; !target && i < delta_search_threads; i++)2511 if (!p[i].working)2512 target = &p[i];2513 if (target)2514 break;2515 pthread_cond_wait(&progress_cond, &progress_mutex);2516 }25172518 for (i = 0; i < delta_search_threads; i++)2519 if (p[i].remaining > 2*window &&2520 (!victim || victim->remaining < p[i].remaining))2521 victim = &p[i];2522 if (victim) {2523 sub_size = victim->remaining / 2;2524 list = victim->list + victim->list_size - sub_size;2525 while (sub_size && list[0]->hash &&2526 list[0]->hash == list[-1]->hash) {2527 list++;2528 sub_size--;2529 }2530 if (!sub_size) {2531 /*2532 * It is possible for some "paths" to have2533 * so many objects that no hash boundary2534 * might be found. Let's just steal the2535 * exact half in that case.2536 */2537 sub_size = victim->remaining / 2;2538 list -= sub_size;2539 }2540 target->list = list;2541 victim->list_size -= sub_size;2542 victim->remaining -= sub_size;2543 }2544 target->list_size = sub_size;2545 target->remaining = sub_size;2546 target->working = 1;2547 progress_unlock();25482549 pthread_mutex_lock(&target->mutex);2550 target->data_ready = 1;2551 pthread_cond_signal(&target->cond);2552 pthread_mutex_unlock(&target->mutex);25532554 if (!sub_size) {2555 pthread_join(target->thread, NULL);2556 pthread_cond_destroy(&target->cond);2557 pthread_mutex_destroy(&target->mutex);2558 active_threads--;2559 }2560 }2561 cleanup_threaded_search();2562 free(p);2563}25642565static void add_tag_chain(const struct object_id *oid)2566{2567 struct tag *tag;25682569 /*2570 * We catch duplicates already in add_object_entry(), but we'd2571 * prefer to do this extra check to avoid having to parse the2572 * tag at all if we already know that it's being packed (e.g., if2573 * it was included via bitmaps, we would not have parsed it2574 * previously).2575 */2576 if (packlist_find(&to_pack, oid, NULL))2577 return;25782579 tag = lookup_tag(the_repository, oid);2580 while (1) {2581 if (!tag || parse_tag(tag) || !tag->tagged)2582 die(_("unable to pack objects reachable from tag %s"),2583 oid_to_hex(oid));25842585 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);25862587 if (tag->tagged->type != OBJ_TAG)2588 return;25892590 tag = (struct tag *)tag->tagged;2591 }2592}25932594static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)2595{2596 struct object_id peeled;25972598 if (starts_with(path, "refs/tags/") && /* is a tag? */2599 !peel_ref(path, &peeled) && /* peelable? */2600 packlist_find(&to_pack, &peeled, NULL)) /* object packed? */2601 add_tag_chain(oid);2602 return 0;2603}26042605static void prepare_pack(int window, int depth)2606{2607 struct object_entry **delta_list;2608 uint32_t i, nr_deltas;2609 unsigned n;26102611 if (use_delta_islands)2612 resolve_tree_islands(the_repository, progress, &to_pack);26132614 get_object_details();26152616 /*2617 * If we're locally repacking then we need to be doubly careful2618 * from now on in order to make sure no stealth corruption gets2619 * propagated to the new pack. Clients receiving streamed packs2620 * should validate everything they get anyway so no need to incur2621 * the additional cost here in that case.2622 */2623 if (!pack_to_stdout)2624 do_check_packed_object_crc = 1;26252626 if (!to_pack.nr_objects || !window || !depth)2627 return;26282629 ALLOC_ARRAY(delta_list, to_pack.nr_objects);2630 nr_deltas = n = 0;26312632 for (i = 0; i < to_pack.nr_objects; i++) {2633 struct object_entry *entry = to_pack.objects + i;26342635 if (DELTA(entry))2636 /* This happens if we decided to reuse existing2637 * delta from a pack. "reuse_delta &&" is implied.2638 */2639 continue;26402641 if (!entry->type_valid ||2642 oe_size_less_than(&to_pack, entry, 50))2643 continue;26442645 if (entry->no_try_delta)2646 continue;26472648 if (!entry->preferred_base) {2649 nr_deltas++;2650 if (oe_type(entry) < 0)2651 die(_("unable to get type of object %s"),2652 oid_to_hex(&entry->idx.oid));2653 } else {2654 if (oe_type(entry) < 0) {2655 /*2656 * This object is not found, but we2657 * don't have to include it anyway.2658 */2659 continue;2660 }2661 }26622663 delta_list[n++] = entry;2664 }26652666 if (nr_deltas && n > 1) {2667 unsigned nr_done = 0;2668 if (progress)2669 progress_state = start_progress(_("Compressing objects"),2670 nr_deltas);2671 QSORT(delta_list, n, type_size_sort);2672 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2673 stop_progress(&progress_state);2674 if (nr_done != nr_deltas)2675 die(_("inconsistency with delta count"));2676 }2677 free(delta_list);2678}26792680static int git_pack_config(const char *k, const char *v, void *cb)2681{2682 if (!strcmp(k, "pack.window")) {2683 window = git_config_int(k, v);2684 return 0;2685 }2686 if (!strcmp(k, "pack.windowmemory")) {2687 window_memory_limit = git_config_ulong(k, v);2688 return 0;2689 }2690 if (!strcmp(k, "pack.depth")) {2691 depth = git_config_int(k, v);2692 return 0;2693 }2694 if (!strcmp(k, "pack.deltacachesize")) {2695 max_delta_cache_size = git_config_int(k, v);2696 return 0;2697 }2698 if (!strcmp(k, "pack.deltacachelimit")) {2699 cache_max_small_delta_size = git_config_int(k, v);2700 return 0;2701 }2702 if (!strcmp(k, "pack.writebitmaphashcache")) {2703 if (git_config_bool(k, v))2704 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2705 else2706 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2707 }2708 if (!strcmp(k, "pack.usebitmaps")) {2709 use_bitmap_index_default = git_config_bool(k, v);2710 return 0;2711 }2712 if (!strcmp(k, "pack.usesparse")) {2713 sparse = git_config_bool(k, v);2714 return 0;2715 }2716 if (!strcmp(k, "pack.threads")) {2717 delta_search_threads = git_config_int(k, v);2718 if (delta_search_threads < 0)2719 die(_("invalid number of threads specified (%d)"),2720 delta_search_threads);2721 if (!HAVE_THREADS && delta_search_threads != 1) {2722 warning(_("no threads support, ignoring %s"), k);2723 delta_search_threads = 0;2724 }2725 return 0;2726 }2727 if (!strcmp(k, "pack.indexversion")) {2728 pack_idx_opts.version = git_config_int(k, v);2729 if (pack_idx_opts.version > 2)2730 die(_("bad pack.indexversion=%"PRIu32),2731 pack_idx_opts.version);2732 return 0;2733 }2734 return git_default_config(k, v, cb);2735}27362737static void read_object_list_from_stdin(void)2738{2739 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];2740 struct object_id oid;2741 const char *p;27422743 for (;;) {2744 if (!fgets(line, sizeof(line), stdin)) {2745 if (feof(stdin))2746 break;2747 if (!ferror(stdin))2748 die("BUG: fgets returned NULL, not EOF, not error!");2749 if (errno != EINTR)2750 die_errno("fgets");2751 clearerr(stdin);2752 continue;2753 }2754 if (line[0] == '-') {2755 if (get_oid_hex(line+1, &oid))2756 die(_("expected edge object ID, got garbage:\n %s"),2757 line);2758 add_preferred_base(&oid);2759 continue;2760 }2761 if (parse_oid_hex(line, &oid, &p))2762 die(_("expected object ID, got garbage:\n %s"), line);27632764 add_preferred_base_object(p + 1);2765 add_object_entry(&oid, OBJ_NONE, p + 1, 0);2766 }2767}27682769/* Remember to update object flag allocation in object.h */2770#define OBJECT_ADDED (1u<<20)27712772static void show_commit(struct commit *commit, void *data)2773{2774 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);2775 commit->object.flags |= OBJECT_ADDED;27762777 if (write_bitmap_index)2778 index_commit_for_bitmap(commit);27792780 if (use_delta_islands)2781 propagate_island_marks(commit);2782}27832784static void show_object(struct object *obj, const char *name, void *data)2785{2786 add_preferred_base_object(name);2787 add_object_entry(&obj->oid, obj->type, name, 0);2788 obj->flags |= OBJECT_ADDED;27892790 if (use_delta_islands) {2791 const char *p;2792 unsigned depth;2793 struct object_entry *ent;27942795 /* the empty string is a root tree, which is depth 0 */2796 depth = *name ? 1 : 0;2797 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))2798 depth++;27992800 ent = packlist_find(&to_pack, &obj->oid, NULL);2801 if (ent && depth > oe_tree_depth(&to_pack, ent))2802 oe_set_tree_depth(&to_pack, ent, depth);2803 }2804}28052806static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)2807{2808 assert(arg_missing_action == MA_ALLOW_ANY);28092810 /*2811 * Quietly ignore ALL missing objects. This avoids problems with2812 * staging them now and getting an odd error later.2813 */2814 if (!has_object_file(&obj->oid))2815 return;28162817 show_object(obj, name, data);2818}28192820static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)2821{2822 assert(arg_missing_action == MA_ALLOW_PROMISOR);28232824 /*2825 * Quietly ignore EXPECTED missing objects. This avoids problems with2826 * staging them now and getting an odd error later.2827 */2828 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))2829 return;28302831 show_object(obj, name, data);2832}28332834static int option_parse_missing_action(const struct option *opt,2835 const char *arg, int unset)2836{2837 assert(arg);2838 assert(!unset);28392840 if (!strcmp(arg, "error")) {2841 arg_missing_action = MA_ERROR;2842 fn_show_object = show_object;2843 return 0;2844 }28452846 if (!strcmp(arg, "allow-any")) {2847 arg_missing_action = MA_ALLOW_ANY;2848 fetch_if_missing = 0;2849 fn_show_object = show_object__ma_allow_any;2850 return 0;2851 }28522853 if (!strcmp(arg, "allow-promisor")) {2854 arg_missing_action = MA_ALLOW_PROMISOR;2855 fetch_if_missing = 0;2856 fn_show_object = show_object__ma_allow_promisor;2857 return 0;2858 }28592860 die(_("invalid value for --missing"));2861 return 0;2862}28632864static void show_edge(struct commit *commit)2865{2866 add_preferred_base(&commit->object.oid);2867}28682869struct in_pack_object {2870 off_t offset;2871 struct object *object;2872};28732874struct in_pack {2875 unsigned int alloc;2876 unsigned int nr;2877 struct in_pack_object *array;2878};28792880static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)2881{2882 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);2883 in_pack->array[in_pack->nr].object = object;2884 in_pack->nr++;2885}28862887/*2888 * Compare the objects in the offset order, in order to emulate the2889 * "git rev-list --objects" output that produced the pack originally.2890 */2891static int ofscmp(const void *a_, const void *b_)2892{2893 struct in_pack_object *a = (struct in_pack_object *)a_;2894 struct in_pack_object *b = (struct in_pack_object *)b_;28952896 if (a->offset < b->offset)2897 return -1;2898 else if (a->offset > b->offset)2899 return 1;2900 else2901 return oidcmp(&a->object->oid, &b->object->oid);2902}29032904static void add_objects_in_unpacked_packs(void)2905{2906 struct packed_git *p;2907 struct in_pack in_pack;2908 uint32_t i;29092910 memset(&in_pack, 0, sizeof(in_pack));29112912 for (p = get_all_packs(the_repository); p; p = p->next) {2913 struct object_id oid;2914 struct object *o;29152916 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)2917 continue;2918 if (open_pack_index(p))2919 die(_("cannot open pack index"));29202921 ALLOC_GROW(in_pack.array,2922 in_pack.nr + p->num_objects,2923 in_pack.alloc);29242925 for (i = 0; i < p->num_objects; i++) {2926 nth_packed_object_oid(&oid, p, i);2927 o = lookup_unknown_object(&oid);2928 if (!(o->flags & OBJECT_ADDED))2929 mark_in_pack_object(o, p, &in_pack);2930 o->flags |= OBJECT_ADDED;2931 }2932 }29332934 if (in_pack.nr) {2935 QSORT(in_pack.array, in_pack.nr, ofscmp);2936 for (i = 0; i < in_pack.nr; i++) {2937 struct object *o = in_pack.array[i].object;2938 add_object_entry(&o->oid, o->type, "", 0);2939 }2940 }2941 free(in_pack.array);2942}29432944static int add_loose_object(const struct object_id *oid, const char *path,2945 void *data)2946{2947 enum object_type type = oid_object_info(the_repository, oid, NULL);29482949 if (type < 0) {2950 warning(_("loose object at %s could not be examined"), path);2951 return 0;2952 }29532954 add_object_entry(oid, type, "", 0);2955 return 0;2956}29572958/*2959 * We actually don't even have to worry about reachability here.2960 * add_object_entry will weed out duplicates, so we just add every2961 * loose object we find.2962 */2963static void add_unreachable_loose_objects(void)2964{2965 for_each_loose_file_in_objdir(get_object_directory(),2966 add_loose_object,2967 NULL, NULL, NULL);2968}29692970static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2971{2972 static struct packed_git *last_found = (void *)1;2973 struct packed_git *p;29742975 p = (last_found != (void *)1) ? last_found :2976 get_all_packs(the_repository);29772978 while (p) {2979 if ((!p->pack_local || p->pack_keep ||2980 p->pack_keep_in_core) &&2981 find_pack_entry_one(oid->hash, p)) {2982 last_found = p;2983 return 1;2984 }2985 if (p == last_found)2986 p = get_all_packs(the_repository);2987 else2988 p = p->next;2989 if (p == last_found)2990 p = p->next;2991 }2992 return 0;2993}29942995/*2996 * Store a list of sha1s that are should not be discarded2997 * because they are either written too recently, or are2998 * reachable from another object that was.2999 *3000 * This is filled by get_object_list.3001 */3002static struct oid_array recent_objects;30033004static int loosened_object_can_be_discarded(const struct object_id *oid,3005 timestamp_t mtime)3006{3007 if (!unpack_unreachable_expiration)3008 return 0;3009 if (mtime > unpack_unreachable_expiration)3010 return 0;3011 if (oid_array_lookup(&recent_objects, oid) >= 0)3012 return 0;3013 return 1;3014}30153016static void loosen_unused_packed_objects(void)3017{3018 struct packed_git *p;3019 uint32_t i;3020 struct object_id oid;30213022 for (p = get_all_packs(the_repository); p; p = p->next) {3023 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)3024 continue;30253026 if (open_pack_index(p))3027 die(_("cannot open pack index"));30283029 for (i = 0; i < p->num_objects; i++) {3030 nth_packed_object_oid(&oid, p, i);3031 if (!packlist_find(&to_pack, &oid, NULL) &&3032 !has_sha1_pack_kept_or_nonlocal(&oid) &&3033 !loosened_object_can_be_discarded(&oid, p->mtime))3034 if (force_object_loose(&oid, p->mtime))3035 die(_("unable to force loose object"));3036 }3037 }3038}30393040/*3041 * This tracks any options which pack-reuse code expects to be on, or which a3042 * reader of the pack might not understand, and which would therefore prevent3043 * blind reuse of what we have on disk.3044 */3045static int pack_options_allow_reuse(void)3046{3047 return pack_to_stdout &&3048 allow_ofs_delta &&3049 !ignore_packed_keep_on_disk &&3050 !ignore_packed_keep_in_core &&3051 (!local || !have_non_local_packs) &&3052 !incremental;3053}30543055static int get_object_list_from_bitmap(struct rev_info *revs)3056{3057 if (!(bitmap_git = prepare_bitmap_walk(revs)))3058 return -1;30593060 if (pack_options_allow_reuse() &&3061 !reuse_partial_packfile_from_bitmap(3062 bitmap_git,3063 &reuse_packfile,3064 &reuse_packfile_objects,3065 &reuse_packfile_offset)) {3066 assert(reuse_packfile_objects);3067 nr_result += reuse_packfile_objects;3068 display_progress(progress_state, nr_result);3069 }30703071 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3072 return 0;3073}30743075static void record_recent_object(struct object *obj,3076 const char *name,3077 void *data)3078{3079 oid_array_append(&recent_objects, &obj->oid);3080}30813082static void record_recent_commit(struct commit *commit, void *data)3083{3084 oid_array_append(&recent_objects, &commit->object.oid);3085}30863087static void get_object_list(int ac, const char **av)3088{3089 struct rev_info revs;3090 struct setup_revision_opt s_r_opt = {3091 .allow_exclude_promisor_objects = 1,3092 };3093 char line[1000];3094 int flags = 0;3095 int save_warning;30963097 repo_init_revisions(the_repository, &revs, NULL);3098 save_commit_buffer = 0;3099 setup_revisions(ac, av, &revs, &s_r_opt);31003101 /* make sure shallows are read */3102 is_repository_shallow(the_repository);31033104 save_warning = warn_on_object_refname_ambiguity;3105 warn_on_object_refname_ambiguity = 0;31063107 while (fgets(line, sizeof(line), stdin) != NULL) {3108 int len = strlen(line);3109 if (len && line[len - 1] == '\n')3110 line[--len] = 0;3111 if (!len)3112 break;3113 if (*line == '-') {3114 if (!strcmp(line, "--not")) {3115 flags ^= UNINTERESTING;3116 write_bitmap_index = 0;3117 continue;3118 }3119 if (starts_with(line, "--shallow ")) {3120 struct object_id oid;3121 if (get_oid_hex(line + 10, &oid))3122 die("not an SHA-1 '%s'", line + 10);3123 register_shallow(the_repository, &oid);3124 use_bitmap_index = 0;3125 continue;3126 }3127 die(_("not a rev '%s'"), line);3128 }3129 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3130 die(_("bad revision '%s'"), line);3131 }31323133 warn_on_object_refname_ambiguity = save_warning;31343135 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))3136 return;31373138 if (use_delta_islands)3139 load_delta_islands(the_repository, progress);31403141 if (prepare_revision_walk(&revs))3142 die(_("revision walk setup failed"));3143 mark_edges_uninteresting(&revs, show_edge, sparse);31443145 if (!fn_show_object)3146 fn_show_object = show_object;3147 traverse_commit_list_filtered(&filter_options, &revs,3148 show_commit, fn_show_object, NULL,3149 NULL);31503151 if (unpack_unreachable_expiration) {3152 revs.ignore_missing_links = 1;3153 if (add_unseen_recent_objects_to_traversal(&revs,3154 unpack_unreachable_expiration))3155 die(_("unable to add recent objects"));3156 if (prepare_revision_walk(&revs))3157 die(_("revision walk setup failed"));3158 traverse_commit_list(&revs, record_recent_commit,3159 record_recent_object, NULL);3160 }31613162 if (keep_unreachable)3163 add_objects_in_unpacked_packs();3164 if (pack_loose_unreachable)3165 add_unreachable_loose_objects();3166 if (unpack_unreachable)3167 loosen_unused_packed_objects();31683169 oid_array_clear(&recent_objects);3170}31713172static void add_extra_kept_packs(const struct string_list *names)3173{3174 struct packed_git *p;31753176 if (!names->nr)3177 return;31783179 for (p = get_all_packs(the_repository); p; p = p->next) {3180 const char *name = basename(p->pack_name);3181 int i;31823183 if (!p->pack_local)3184 continue;31853186 for (i = 0; i < names->nr; i++)3187 if (!fspathcmp(name, names->items[i].string))3188 break;31893190 if (i < names->nr) {3191 p->pack_keep_in_core = 1;3192 ignore_packed_keep_in_core = 1;3193 continue;3194 }3195 }3196}31973198static int option_parse_index_version(const struct option *opt,3199 const char *arg, int unset)3200{3201 char *c;3202 const char *val = arg;32033204 BUG_ON_OPT_NEG(unset);32053206 pack_idx_opts.version = strtoul(val, &c, 10);3207 if (pack_idx_opts.version > 2)3208 die(_("unsupported index version %s"), val);3209 if (*c == ',' && c[1])3210 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);3211 if (*c || pack_idx_opts.off32_limit & 0x80000000)3212 die(_("bad index version '%s'"), val);3213 return 0;3214}32153216static int option_parse_unpack_unreachable(const struct option *opt,3217 const char *arg, int unset)3218{3219 if (unset) {3220 unpack_unreachable = 0;3221 unpack_unreachable_expiration = 0;3222 }3223 else {3224 unpack_unreachable = 1;3225 if (arg)3226 unpack_unreachable_expiration = approxidate(arg);3227 }3228 return 0;3229}32303231int cmd_pack_objects(int argc, const char **argv, const char *prefix)3232{3233 int use_internal_rev_list = 0;3234 int shallow = 0;3235 int all_progress_implied = 0;3236 struct argv_array rp = ARGV_ARRAY_INIT;3237 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;3238 int rev_list_index = 0;3239 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3240 struct option pack_objects_options[] = {3241 OPT_SET_INT('q', "quiet", &progress,3242 N_("do not show progress meter"), 0),3243 OPT_SET_INT(0, "progress", &progress,3244 N_("show progress meter"), 1),3245 OPT_SET_INT(0, "all-progress", &progress,3246 N_("show progress meter during object writing phase"), 2),3247 OPT_BOOL(0, "all-progress-implied",3248 &all_progress_implied,3249 N_("similar to --all-progress when progress meter is shown")),3250 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),3251 N_("write the pack index file in the specified idx format version"),3252 PARSE_OPT_NONEG, option_parse_index_version },3253 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,3254 N_("maximum size of each output pack file")),3255 OPT_BOOL(0, "local", &local,3256 N_("ignore borrowed objects from alternate object store")),3257 OPT_BOOL(0, "incremental", &incremental,3258 N_("ignore packed objects")),3259 OPT_INTEGER(0, "window", &window,3260 N_("limit pack window by objects")),3261 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,3262 N_("limit pack window by memory in addition to object limit")),3263 OPT_INTEGER(0, "depth", &depth,3264 N_("maximum length of delta chain allowed in the resulting pack")),3265 OPT_BOOL(0, "reuse-delta", &reuse_delta,3266 N_("reuse existing deltas")),3267 OPT_BOOL(0, "reuse-object", &reuse_object,3268 N_("reuse existing objects")),3269 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,3270 N_("use OFS_DELTA objects")),3271 OPT_INTEGER(0, "threads", &delta_search_threads,3272 N_("use threads when searching for best delta matches")),3273 OPT_BOOL(0, "non-empty", &non_empty,3274 N_("do not create an empty pack output")),3275 OPT_BOOL(0, "revs", &use_internal_rev_list,3276 N_("read revision arguments from standard input")),3277 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,3278 N_("limit the objects to those that are not yet packed"),3279 1, PARSE_OPT_NONEG),3280 OPT_SET_INT_F(0, "all", &rev_list_all,3281 N_("include objects reachable from any reference"),3282 1, PARSE_OPT_NONEG),3283 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,3284 N_("include objects referred by reflog entries"),3285 1, PARSE_OPT_NONEG),3286 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,3287 N_("include objects referred to by the index"),3288 1, PARSE_OPT_NONEG),3289 OPT_BOOL(0, "stdout", &pack_to_stdout,3290 N_("output pack to stdout")),3291 OPT_BOOL(0, "include-tag", &include_tag,3292 N_("include tag objects that refer to objects to be packed")),3293 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,3294 N_("keep unreachable objects")),3295 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,3296 N_("pack loose unreachable objects")),3297 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),3298 N_("unpack unreachable objects newer than <time>"),3299 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3300 OPT_BOOL(0, "sparse", &sparse,3301 N_("use the sparse reachability algorithm")),3302 OPT_BOOL(0, "thin", &thin,3303 N_("create thin packs")),3304 OPT_BOOL(0, "shallow", &shallow,3305 N_("create packs suitable for shallow fetches")),3306 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,3307 N_("ignore packs that have companion .keep file")),3308 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),3309 N_("ignore this pack")),3310 OPT_INTEGER(0, "compression", &pack_compression_level,3311 N_("pack compression level")),3312 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,3313 N_("do not hide commits by grafts"), 0),3314 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,3315 N_("use a bitmap index if available to speed up counting objects")),3316 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,3317 N_("write a bitmap index together with the pack index")),3318 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3319 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),3320 N_("handling for missing objects"), PARSE_OPT_NONEG,3321 option_parse_missing_action },3322 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,3323 N_("do not pack objects in promisor packfiles")),3324 OPT_BOOL(0, "delta-islands", &use_delta_islands,3325 N_("respect islands during delta compression")),3326 OPT_END(),3327 };33283329 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))3330 BUG("too many dfs states, increase OE_DFS_STATE_BITS");33313332 read_replace_refs = 0;33333334 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", 0);3335 reset_pack_idx_option(&pack_idx_opts);3336 git_config(git_pack_config, NULL);33373338 progress = isatty(2);3339 argc = parse_options(argc, argv, prefix, pack_objects_options,3340 pack_usage, 0);33413342 if (argc) {3343 base_name = argv[0];3344 argc--;3345 }3346 if (pack_to_stdout != !base_name || argc)3347 usage_with_options(pack_usage, pack_objects_options);33483349 if (depth >= (1 << OE_DEPTH_BITS)) {3350 warning(_("delta chain depth %d is too deep, forcing %d"),3351 depth, (1 << OE_DEPTH_BITS) - 1);3352 depth = (1 << OE_DEPTH_BITS) - 1;3353 }3354 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {3355 warning(_("pack.deltaCacheLimit is too high, forcing %d"),3356 (1U << OE_Z_DELTA_BITS) - 1);3357 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;3358 }33593360 argv_array_push(&rp, "pack-objects");3361 if (thin) {3362 use_internal_rev_list = 1;3363 argv_array_push(&rp, shallow3364 ? "--objects-edge-aggressive"3365 : "--objects-edge");3366 } else3367 argv_array_push(&rp, "--objects");33683369 if (rev_list_all) {3370 use_internal_rev_list = 1;3371 argv_array_push(&rp, "--all");3372 }3373 if (rev_list_reflog) {3374 use_internal_rev_list = 1;3375 argv_array_push(&rp, "--reflog");3376 }3377 if (rev_list_index) {3378 use_internal_rev_list = 1;3379 argv_array_push(&rp, "--indexed-objects");3380 }3381 if (rev_list_unpacked) {3382 use_internal_rev_list = 1;3383 argv_array_push(&rp, "--unpacked");3384 }33853386 if (exclude_promisor_objects) {3387 use_internal_rev_list = 1;3388 fetch_if_missing = 0;3389 argv_array_push(&rp, "--exclude-promisor-objects");3390 }3391 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)3392 use_internal_rev_list = 1;33933394 if (!reuse_object)3395 reuse_delta = 0;3396 if (pack_compression_level == -1)3397 pack_compression_level = Z_DEFAULT_COMPRESSION;3398 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)3399 die(_("bad pack compression level %d"), pack_compression_level);34003401 if (!delta_search_threads) /* --threads=0 means autodetect */3402 delta_search_threads = online_cpus();34033404 if (!HAVE_THREADS && delta_search_threads != 1)3405 warning(_("no threads support, ignoring --threads"));3406 if (!pack_to_stdout && !pack_size_limit)3407 pack_size_limit = pack_size_limit_cfg;3408 if (pack_to_stdout && pack_size_limit)3409 die(_("--max-pack-size cannot be used to build a pack for transfer"));3410 if (pack_size_limit && pack_size_limit < 1024*1024) {3411 warning(_("minimum pack size limit is 1 MiB"));3412 pack_size_limit = 1024*1024;3413 }34143415 if (!pack_to_stdout && thin)3416 die(_("--thin cannot be used to build an indexable pack"));34173418 if (keep_unreachable && unpack_unreachable)3419 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3420 if (!rev_list_all || !rev_list_reflog || !rev_list_index)3421 unpack_unreachable_expiration = 0;34223423 if (filter_options.choice) {3424 if (!pack_to_stdout)3425 die(_("cannot use --filter without --stdout"));3426 use_bitmap_index = 0;3427 }34283429 /*3430 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3431 *3432 * - to produce good pack (with bitmap index not-yet-packed objects are3433 * packed in suboptimal order).3434 *3435 * - to use more robust pack-generation codepath (avoiding possible3436 * bugs in bitmap code and possible bitmap index corruption).3437 */3438 if (!pack_to_stdout)3439 use_bitmap_index_default = 0;34403441 if (use_bitmap_index < 0)3442 use_bitmap_index = use_bitmap_index_default;34433444 /* "hard" reasons not to use bitmaps; these just won't work at all */3445 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))3446 use_bitmap_index = 0;34473448 if (pack_to_stdout || !rev_list_all)3449 write_bitmap_index = 0;34503451 if (use_delta_islands)3452 argv_array_push(&rp, "--topo-order");34533454 if (progress && all_progress_implied)3455 progress = 2;34563457 add_extra_kept_packs(&keep_pack_list);3458 if (ignore_packed_keep_on_disk) {3459 struct packed_git *p;3460 for (p = get_all_packs(the_repository); p; p = p->next)3461 if (p->pack_local && p->pack_keep)3462 break;3463 if (!p) /* no keep-able packs found */3464 ignore_packed_keep_on_disk = 0;3465 }3466 if (local) {3467 /*3468 * unlike ignore_packed_keep_on_disk above, we do not3469 * want to unset "local" based on looking at packs, as3470 * it also covers non-local objects3471 */3472 struct packed_git *p;3473 for (p = get_all_packs(the_repository); p; p = p->next) {3474 if (!p->pack_local) {3475 have_non_local_packs = 1;3476 break;3477 }3478 }3479 }34803481 trace2_region_enter("pack-objects", "enumerate-objects",3482 the_repository);3483 prepare_packing_data(the_repository, &to_pack);34843485 if (progress)3486 progress_state = start_progress(_("Enumerating objects"), 0);3487 if (!use_internal_rev_list)3488 read_object_list_from_stdin();3489 else {3490 get_object_list(rp.argc, rp.argv);3491 argv_array_clear(&rp);3492 }3493 cleanup_preferred_base();3494 if (include_tag && nr_result)3495 for_each_ref(add_ref_tag, NULL);3496 stop_progress(&progress_state);3497 trace2_region_leave("pack-objects", "enumerate-objects",3498 the_repository);34993500 if (non_empty && !nr_result)3501 return 0;3502 if (nr_result) {3503 trace2_region_enter("pack-objects", "prepare-pack",3504 the_repository);3505 prepare_pack(window, depth);3506 trace2_region_leave("pack-objects", "prepare-pack",3507 the_repository);3508 }35093510 trace2_region_enter("pack-objects", "write-pack-file", the_repository);3511 write_pack_file();3512 trace2_region_leave("pack-objects", "write-pack-file", the_repository);35133514 if (progress)3515 fprintf_ln(stderr,3516 _("Total %"PRIu32" (delta %"PRIu32"),"3517 " reused %"PRIu32" (delta %"PRIu32")"),3518 written, written_delta, reused, reused_delta);3519 return 0;3520}