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 enum { 100 WRITE_BITMAP_FALSE = 0, 101 WRITE_BITMAP_QUIET, 102 WRITE_BITMAP_TRUE, 103} write_bitmap_index; 104static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE; 105 106static int exclude_promisor_objects; 107 108static int use_delta_islands; 109 110static unsigned long delta_cache_size = 0; 111static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 112static unsigned long cache_max_small_delta_size = 1000; 113 114static unsigned long window_memory_limit = 0; 115 116static struct list_objects_filter_options filter_options; 117 118enum missing_action { 119 MA_ERROR = 0, /* fail if any missing objects are encountered */ 120 MA_ALLOW_ANY, /* silently allow ALL missing objects */ 121 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */ 122}; 123static enum missing_action arg_missing_action; 124static show_object_fn fn_show_object; 125 126/* 127 * stats 128 */ 129static uint32_t written, written_delta; 130static uint32_t reused, reused_delta; 131 132/* 133 * Indexed commits 134 */ 135static struct commit **indexed_commits; 136static unsigned int indexed_commits_nr; 137static unsigned int indexed_commits_alloc; 138 139static void index_commit_for_bitmap(struct commit *commit) 140{ 141 if (indexed_commits_nr >= indexed_commits_alloc) { 142 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; 143 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 144 } 145 146 indexed_commits[indexed_commits_nr++] = commit; 147} 148 149static void *get_delta(struct object_entry *entry) 150{ 151 unsigned long size, base_size, delta_size; 152 void *buf, *base_buf, *delta_buf; 153 enum object_type type; 154 155 buf = read_object_file(&entry->idx.oid, &type, &size); 156 if (!buf) 157 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid)); 158 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type, 159 &base_size); 160 if (!base_buf) 161 die("unable to read %s", 162 oid_to_hex(&DELTA(entry)->idx.oid)); 163 delta_buf = diff_delta(base_buf, base_size, 164 buf, size, &delta_size, 0); 165 /* 166 * We succesfully computed this delta once but dropped it for 167 * memory reasons. Something is very wrong if this time we 168 * recompute and create a different delta. 169 */ 170 if (!delta_buf || delta_size != DELTA_SIZE(entry)) 171 BUG("delta size changed"); 172 free(buf); 173 free(base_buf); 174 return delta_buf; 175} 176 177static unsigned long do_compress(void **pptr, unsigned long size) 178{ 179 git_zstream stream; 180 void *in, *out; 181 unsigned long maxsize; 182 183 git_deflate_init(&stream, pack_compression_level); 184 maxsize = git_deflate_bound(&stream, size); 185 186 in = *pptr; 187 out = xmalloc(maxsize); 188 *pptr = out; 189 190 stream.next_in = in; 191 stream.avail_in = size; 192 stream.next_out = out; 193 stream.avail_out = maxsize; 194 while (git_deflate(&stream, Z_FINISH) == Z_OK) 195 ; /* nothing */ 196 git_deflate_end(&stream); 197 198 free(in); 199 return stream.total_out; 200} 201 202static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f, 203 const struct object_id *oid) 204{ 205 git_zstream stream; 206 unsigned char ibuf[1024 * 16]; 207 unsigned char obuf[1024 * 16]; 208 unsigned long olen = 0; 209 210 git_deflate_init(&stream, pack_compression_level); 211 212 for (;;) { 213 ssize_t readlen; 214 int zret = Z_OK; 215 readlen = read_istream(st, ibuf, sizeof(ibuf)); 216 if (readlen == -1) 217 die(_("unable to read %s"), oid_to_hex(oid)); 218 219 stream.next_in = ibuf; 220 stream.avail_in = readlen; 221 while ((stream.avail_in || readlen == 0) && 222 (zret == Z_OK || zret == Z_BUF_ERROR)) { 223 stream.next_out = obuf; 224 stream.avail_out = sizeof(obuf); 225 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH); 226 hashwrite(f, obuf, stream.next_out - obuf); 227 olen += stream.next_out - obuf; 228 } 229 if (stream.avail_in) 230 die(_("deflate error (%d)"), zret); 231 if (readlen == 0) { 232 if (zret != Z_STREAM_END) 233 die(_("deflate error (%d)"), zret); 234 break; 235 } 236 } 237 git_deflate_end(&stream); 238 return olen; 239} 240 241/* 242 * we are going to reuse the existing object data as is. make 243 * sure it is not corrupt. 244 */ 245static int check_pack_inflate(struct packed_git *p, 246 struct pack_window **w_curs, 247 off_t offset, 248 off_t len, 249 unsigned long expect) 250{ 251 git_zstream stream; 252 unsigned char fakebuf[4096], *in; 253 int st; 254 255 memset(&stream, 0, sizeof(stream)); 256 git_inflate_init(&stream); 257 do { 258 in = use_pack(p, w_curs, offset, &stream.avail_in); 259 stream.next_in = in; 260 stream.next_out = fakebuf; 261 stream.avail_out = sizeof(fakebuf); 262 st = git_inflate(&stream, Z_FINISH); 263 offset += stream.next_in - in; 264 } while (st == Z_OK || st == Z_BUF_ERROR); 265 git_inflate_end(&stream); 266 return (st == Z_STREAM_END && 267 stream.total_out == expect && 268 stream.total_in == len) ? 0 : -1; 269} 270 271static void copy_pack_data(struct hashfile *f, 272 struct packed_git *p, 273 struct pack_window **w_curs, 274 off_t offset, 275 off_t len) 276{ 277 unsigned char *in; 278 unsigned long avail; 279 280 while (len) { 281 in = use_pack(p, w_curs, offset, &avail); 282 if (avail > len) 283 avail = (unsigned long)len; 284 hashwrite(f, in, avail); 285 offset += avail; 286 len -= avail; 287 } 288} 289 290/* Return 0 if we will bust the pack-size limit */ 291static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry, 292 unsigned long limit, int usable_delta) 293{ 294 unsigned long size, datalen; 295 unsigned char header[MAX_PACK_OBJECT_HEADER], 296 dheader[MAX_PACK_OBJECT_HEADER]; 297 unsigned hdrlen; 298 enum object_type type; 299 void *buf; 300 struct git_istream *st = NULL; 301 const unsigned hashsz = the_hash_algo->rawsz; 302 303 if (!usable_delta) { 304 if (oe_type(entry) == OBJ_BLOB && 305 oe_size_greater_than(&to_pack, entry, big_file_threshold) && 306 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 307 buf = NULL; 308 else { 309 buf = read_object_file(&entry->idx.oid, &type, &size); 310 if (!buf) 311 die(_("unable to read %s"), 312 oid_to_hex(&entry->idx.oid)); 313 } 314 /* 315 * make sure no cached delta data remains from a 316 * previous attempt before a pack split occurred. 317 */ 318 FREE_AND_NULL(entry->delta_data); 319 entry->z_delta_size = 0; 320 } else if (entry->delta_data) { 321 size = DELTA_SIZE(entry); 322 buf = entry->delta_data; 323 entry->delta_data = NULL; 324 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 325 OBJ_OFS_DELTA : OBJ_REF_DELTA; 326 } else { 327 buf = get_delta(entry); 328 size = DELTA_SIZE(entry); 329 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 330 OBJ_OFS_DELTA : OBJ_REF_DELTA; 331 } 332 333 if (st) /* large blob case, just assume we don't compress well */ 334 datalen = size; 335 else if (entry->z_delta_size) 336 datalen = entry->z_delta_size; 337 else 338 datalen = do_compress(&buf, size); 339 340 /* 341 * The object header is a byte of 'type' followed by zero or 342 * more bytes of length. 343 */ 344 hdrlen = encode_in_pack_object_header(header, sizeof(header), 345 type, size); 346 347 if (type == OBJ_OFS_DELTA) { 348 /* 349 * Deltas with relative base contain an additional 350 * encoding of the relative offset for the delta 351 * base from this object's position in the pack. 352 */ 353 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 354 unsigned pos = sizeof(dheader) - 1; 355 dheader[pos] = ofs & 127; 356 while (ofs >>= 7) 357 dheader[--pos] = 128 | (--ofs & 127); 358 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 359 if (st) 360 close_istream(st); 361 free(buf); 362 return 0; 363 } 364 hashwrite(f, header, hdrlen); 365 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 366 hdrlen += sizeof(dheader) - pos; 367 } else if (type == OBJ_REF_DELTA) { 368 /* 369 * Deltas with a base reference contain 370 * additional bytes for the base object ID. 371 */ 372 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 373 if (st) 374 close_istream(st); 375 free(buf); 376 return 0; 377 } 378 hashwrite(f, header, hdrlen); 379 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 380 hdrlen += hashsz; 381 } else { 382 if (limit && hdrlen + datalen + hashsz >= limit) { 383 if (st) 384 close_istream(st); 385 free(buf); 386 return 0; 387 } 388 hashwrite(f, header, hdrlen); 389 } 390 if (st) { 391 datalen = write_large_blob_data(st, f, &entry->idx.oid); 392 close_istream(st); 393 } else { 394 hashwrite(f, buf, datalen); 395 free(buf); 396 } 397 398 return hdrlen + datalen; 399} 400 401/* Return 0 if we will bust the pack-size limit */ 402static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry, 403 unsigned long limit, int usable_delta) 404{ 405 struct packed_git *p = IN_PACK(entry); 406 struct pack_window *w_curs = NULL; 407 struct revindex_entry *revidx; 408 off_t offset; 409 enum object_type type = oe_type(entry); 410 off_t datalen; 411 unsigned char header[MAX_PACK_OBJECT_HEADER], 412 dheader[MAX_PACK_OBJECT_HEADER]; 413 unsigned hdrlen; 414 const unsigned hashsz = the_hash_algo->rawsz; 415 unsigned long entry_size = SIZE(entry); 416 417 if (DELTA(entry)) 418 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 419 OBJ_OFS_DELTA : OBJ_REF_DELTA; 420 hdrlen = encode_in_pack_object_header(header, sizeof(header), 421 type, entry_size); 422 423 offset = entry->in_pack_offset; 424 revidx = find_pack_revindex(p, offset); 425 datalen = revidx[1].offset - offset; 426 if (!pack_to_stdout && p->index_version > 1 && 427 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 428 error(_("bad packed object CRC for %s"), 429 oid_to_hex(&entry->idx.oid)); 430 unuse_pack(&w_curs); 431 return write_no_reuse_object(f, entry, limit, usable_delta); 432 } 433 434 offset += entry->in_pack_header_size; 435 datalen -= entry->in_pack_header_size; 436 437 if (!pack_to_stdout && p->index_version == 1 && 438 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 439 error(_("corrupt packed object for %s"), 440 oid_to_hex(&entry->idx.oid)); 441 unuse_pack(&w_curs); 442 return write_no_reuse_object(f, entry, limit, usable_delta); 443 } 444 445 if (type == OBJ_OFS_DELTA) { 446 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 447 unsigned pos = sizeof(dheader) - 1; 448 dheader[pos] = ofs & 127; 449 while (ofs >>= 7) 450 dheader[--pos] = 128 | (--ofs & 127); 451 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 452 unuse_pack(&w_curs); 453 return 0; 454 } 455 hashwrite(f, header, hdrlen); 456 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 457 hdrlen += sizeof(dheader) - pos; 458 reused_delta++; 459 } else if (type == OBJ_REF_DELTA) { 460 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 461 unuse_pack(&w_curs); 462 return 0; 463 } 464 hashwrite(f, header, hdrlen); 465 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 466 hdrlen += hashsz; 467 reused_delta++; 468 } else { 469 if (limit && hdrlen + datalen + hashsz >= limit) { 470 unuse_pack(&w_curs); 471 return 0; 472 } 473 hashwrite(f, header, hdrlen); 474 } 475 copy_pack_data(f, p, &w_curs, offset, datalen); 476 unuse_pack(&w_curs); 477 reused++; 478 return hdrlen + datalen; 479} 480 481/* Return 0 if we will bust the pack-size limit */ 482static off_t write_object(struct hashfile *f, 483 struct object_entry *entry, 484 off_t write_offset) 485{ 486 unsigned long limit; 487 off_t len; 488 int usable_delta, to_reuse; 489 490 if (!pack_to_stdout) 491 crc32_begin(f); 492 493 /* apply size limit if limited packsize and not first object */ 494 if (!pack_size_limit || !nr_written) 495 limit = 0; 496 else if (pack_size_limit <= write_offset) 497 /* 498 * the earlier object did not fit the limit; avoid 499 * mistaking this with unlimited (i.e. limit = 0). 500 */ 501 limit = 1; 502 else 503 limit = pack_size_limit - write_offset; 504 505 if (!DELTA(entry)) 506 usable_delta = 0; /* no delta */ 507 else if (!pack_size_limit) 508 usable_delta = 1; /* unlimited packfile */ 509 else if (DELTA(entry)->idx.offset == (off_t)-1) 510 usable_delta = 0; /* base was written to another pack */ 511 else if (DELTA(entry)->idx.offset) 512 usable_delta = 1; /* base already exists in this pack */ 513 else 514 usable_delta = 0; /* base could end up in another pack */ 515 516 if (!reuse_object) 517 to_reuse = 0; /* explicit */ 518 else if (!IN_PACK(entry)) 519 to_reuse = 0; /* can't reuse what we don't have */ 520 else if (oe_type(entry) == OBJ_REF_DELTA || 521 oe_type(entry) == OBJ_OFS_DELTA) 522 /* check_object() decided it for us ... */ 523 to_reuse = usable_delta; 524 /* ... but pack split may override that */ 525 else if (oe_type(entry) != entry->in_pack_type) 526 to_reuse = 0; /* pack has delta which is unusable */ 527 else if (DELTA(entry)) 528 to_reuse = 0; /* we want to pack afresh */ 529 else 530 to_reuse = 1; /* we have it in-pack undeltified, 531 * and we do not need to deltify it. 532 */ 533 534 if (!to_reuse) 535 len = write_no_reuse_object(f, entry, limit, usable_delta); 536 else 537 len = write_reuse_object(f, entry, limit, usable_delta); 538 if (!len) 539 return 0; 540 541 if (usable_delta) 542 written_delta++; 543 written++; 544 if (!pack_to_stdout) 545 entry->idx.crc32 = crc32_end(f); 546 return len; 547} 548 549enum write_one_status { 550 WRITE_ONE_SKIP = -1, /* already written */ 551 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */ 552 WRITE_ONE_WRITTEN = 1, /* normal */ 553 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */ 554}; 555 556static enum write_one_status write_one(struct hashfile *f, 557 struct object_entry *e, 558 off_t *offset) 559{ 560 off_t size; 561 int recursing; 562 563 /* 564 * we set offset to 1 (which is an impossible value) to mark 565 * the fact that this object is involved in "write its base 566 * first before writing a deltified object" recursion. 567 */ 568 recursing = (e->idx.offset == 1); 569 if (recursing) { 570 warning(_("recursive delta detected for object %s"), 571 oid_to_hex(&e->idx.oid)); 572 return WRITE_ONE_RECURSIVE; 573 } else if (e->idx.offset || e->preferred_base) { 574 /* offset is non zero if object is written already. */ 575 return WRITE_ONE_SKIP; 576 } 577 578 /* if we are deltified, write out base object first. */ 579 if (DELTA(e)) { 580 e->idx.offset = 1; /* now recurse */ 581 switch (write_one(f, DELTA(e), offset)) { 582 case WRITE_ONE_RECURSIVE: 583 /* we cannot depend on this one */ 584 SET_DELTA(e, NULL); 585 break; 586 default: 587 break; 588 case WRITE_ONE_BREAK: 589 e->idx.offset = recursing; 590 return WRITE_ONE_BREAK; 591 } 592 } 593 594 e->idx.offset = *offset; 595 size = write_object(f, e, *offset); 596 if (!size) { 597 e->idx.offset = recursing; 598 return WRITE_ONE_BREAK; 599 } 600 written_list[nr_written++] = &e->idx; 601 602 /* make sure off_t is sufficiently large not to wrap */ 603 if (signed_add_overflows(*offset, size)) 604 die(_("pack too large for current definition of off_t")); 605 *offset += size; 606 return WRITE_ONE_WRITTEN; 607} 608 609static int mark_tagged(const char *path, const struct object_id *oid, int flag, 610 void *cb_data) 611{ 612 struct object_id peeled; 613 struct object_entry *entry = packlist_find(&to_pack, oid, NULL); 614 615 if (entry) 616 entry->tagged = 1; 617 if (!peel_ref(path, &peeled)) { 618 entry = packlist_find(&to_pack, &peeled, NULL); 619 if (entry) 620 entry->tagged = 1; 621 } 622 return 0; 623} 624 625static inline void add_to_write_order(struct object_entry **wo, 626 unsigned int *endp, 627 struct object_entry *e) 628{ 629 if (e->filled || oe_layer(&to_pack, e) != write_layer) 630 return; 631 wo[(*endp)++] = e; 632 e->filled = 1; 633} 634 635static void add_descendants_to_write_order(struct object_entry **wo, 636 unsigned int *endp, 637 struct object_entry *e) 638{ 639 int add_to_order = 1; 640 while (e) { 641 if (add_to_order) { 642 struct object_entry *s; 643 /* add this node... */ 644 add_to_write_order(wo, endp, e); 645 /* all its siblings... */ 646 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) { 647 add_to_write_order(wo, endp, s); 648 } 649 } 650 /* drop down a level to add left subtree nodes if possible */ 651 if (DELTA_CHILD(e)) { 652 add_to_order = 1; 653 e = DELTA_CHILD(e); 654 } else { 655 add_to_order = 0; 656 /* our sibling might have some children, it is next */ 657 if (DELTA_SIBLING(e)) { 658 e = DELTA_SIBLING(e); 659 continue; 660 } 661 /* go back to our parent node */ 662 e = DELTA(e); 663 while (e && !DELTA_SIBLING(e)) { 664 /* we're on the right side of a subtree, keep 665 * going up until we can go right again */ 666 e = DELTA(e); 667 } 668 if (!e) { 669 /* done- we hit our original root node */ 670 return; 671 } 672 /* pass it off to sibling at this level */ 673 e = DELTA_SIBLING(e); 674 } 675 }; 676} 677 678static void add_family_to_write_order(struct object_entry **wo, 679 unsigned int *endp, 680 struct object_entry *e) 681{ 682 struct object_entry *root; 683 684 for (root = e; DELTA(root); root = DELTA(root)) 685 ; /* nothing */ 686 add_descendants_to_write_order(wo, endp, root); 687} 688 689static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end) 690{ 691 unsigned int i, last_untagged; 692 struct object_entry *objects = to_pack.objects; 693 694 for (i = 0; i < to_pack.nr_objects; i++) { 695 if (objects[i].tagged) 696 break; 697 add_to_write_order(wo, wo_end, &objects[i]); 698 } 699 last_untagged = i; 700 701 /* 702 * Then fill all the tagged tips. 703 */ 704 for (; i < to_pack.nr_objects; i++) { 705 if (objects[i].tagged) 706 add_to_write_order(wo, wo_end, &objects[i]); 707 } 708 709 /* 710 * And then all remaining commits and tags. 711 */ 712 for (i = last_untagged; i < to_pack.nr_objects; i++) { 713 if (oe_type(&objects[i]) != OBJ_COMMIT && 714 oe_type(&objects[i]) != OBJ_TAG) 715 continue; 716 add_to_write_order(wo, wo_end, &objects[i]); 717 } 718 719 /* 720 * And then all the trees. 721 */ 722 for (i = last_untagged; i < to_pack.nr_objects; i++) { 723 if (oe_type(&objects[i]) != OBJ_TREE) 724 continue; 725 add_to_write_order(wo, wo_end, &objects[i]); 726 } 727 728 /* 729 * Finally all the rest in really tight order 730 */ 731 for (i = last_untagged; i < to_pack.nr_objects; i++) { 732 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer) 733 add_family_to_write_order(wo, wo_end, &objects[i]); 734 } 735} 736 737static struct object_entry **compute_write_order(void) 738{ 739 uint32_t max_layers = 1; 740 unsigned int i, wo_end; 741 742 struct object_entry **wo; 743 struct object_entry *objects = to_pack.objects; 744 745 for (i = 0; i < to_pack.nr_objects; i++) { 746 objects[i].tagged = 0; 747 objects[i].filled = 0; 748 SET_DELTA_CHILD(&objects[i], NULL); 749 SET_DELTA_SIBLING(&objects[i], NULL); 750 } 751 752 /* 753 * Fully connect delta_child/delta_sibling network. 754 * Make sure delta_sibling is sorted in the original 755 * recency order. 756 */ 757 for (i = to_pack.nr_objects; i > 0;) { 758 struct object_entry *e = &objects[--i]; 759 if (!DELTA(e)) 760 continue; 761 /* Mark me as the first child */ 762 e->delta_sibling_idx = DELTA(e)->delta_child_idx; 763 SET_DELTA_CHILD(DELTA(e), e); 764 } 765 766 /* 767 * Mark objects that are at the tip of tags. 768 */ 769 for_each_tag_ref(mark_tagged, NULL); 770 771 if (use_delta_islands) 772 max_layers = compute_pack_layers(&to_pack); 773 774 ALLOC_ARRAY(wo, to_pack.nr_objects); 775 wo_end = 0; 776 777 for (; write_layer < max_layers; ++write_layer) 778 compute_layer_order(wo, &wo_end); 779 780 if (wo_end != to_pack.nr_objects) 781 die(_("ordered %u objects, expected %"PRIu32), 782 wo_end, to_pack.nr_objects); 783 784 return wo; 785} 786 787static off_t write_reused_pack(struct hashfile *f) 788{ 789 unsigned char buffer[8192]; 790 off_t to_write, total; 791 int fd; 792 793 if (!is_pack_valid(reuse_packfile)) 794 die(_("packfile is invalid: %s"), reuse_packfile->pack_name); 795 796 fd = git_open(reuse_packfile->pack_name); 797 if (fd < 0) 798 die_errno(_("unable to open packfile for reuse: %s"), 799 reuse_packfile->pack_name); 800 801 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1) 802 die_errno(_("unable to seek in reused packfile")); 803 804 if (reuse_packfile_offset < 0) 805 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 806 807 total = to_write = reuse_packfile_offset - sizeof(struct pack_header); 808 809 while (to_write) { 810 int read_pack = xread(fd, buffer, sizeof(buffer)); 811 812 if (read_pack <= 0) 813 die_errno(_("unable to read from reused packfile")); 814 815 if (read_pack > to_write) 816 read_pack = to_write; 817 818 hashwrite(f, buffer, read_pack); 819 to_write -= read_pack; 820 821 /* 822 * We don't know the actual number of objects written, 823 * only how many bytes written, how many bytes total, and 824 * how many objects total. So we can fake it by pretending all 825 * objects we are writing are the same size. This gives us a 826 * smooth progress meter, and at the end it matches the true 827 * answer. 828 */ 829 written = reuse_packfile_objects * 830 (((double)(total - to_write)) / total); 831 display_progress(progress_state, written); 832 } 833 834 close(fd); 835 written = reuse_packfile_objects; 836 display_progress(progress_state, written); 837 return reuse_packfile_offset - sizeof(struct pack_header); 838} 839 840static const char no_split_warning[] = N_( 841"disabling bitmap writing, packs are split due to pack.packSizeLimit" 842); 843 844static void write_pack_file(void) 845{ 846 uint32_t i = 0, j; 847 struct hashfile *f; 848 off_t offset; 849 uint32_t nr_remaining = nr_result; 850 time_t last_mtime = 0; 851 struct object_entry **write_order; 852 853 if (progress > pack_to_stdout) 854 progress_state = start_progress(_("Writing objects"), nr_result); 855 ALLOC_ARRAY(written_list, to_pack.nr_objects); 856 write_order = compute_write_order(); 857 858 do { 859 struct object_id oid; 860 char *pack_tmp_name = NULL; 861 862 if (pack_to_stdout) 863 f = hashfd_throughput(1, "<stdout>", progress_state); 864 else 865 f = create_tmp_packfile(&pack_tmp_name); 866 867 offset = write_pack_header(f, nr_remaining); 868 869 if (reuse_packfile) { 870 off_t packfile_size; 871 assert(pack_to_stdout); 872 873 packfile_size = write_reused_pack(f); 874 offset += packfile_size; 875 } 876 877 nr_written = 0; 878 for (; i < to_pack.nr_objects; i++) { 879 struct object_entry *e = write_order[i]; 880 if (write_one(f, e, &offset) == WRITE_ONE_BREAK) 881 break; 882 display_progress(progress_state, written); 883 } 884 885 /* 886 * Did we write the wrong # entries in the header? 887 * If so, rewrite it like in fast-import 888 */ 889 if (pack_to_stdout) { 890 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 891 } else if (nr_written == nr_remaining) { 892 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 893 } else { 894 int fd = finalize_hashfile(f, oid.hash, 0); 895 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 896 nr_written, oid.hash, offset); 897 close(fd); 898 if (write_bitmap_index) { 899 if (write_bitmap_index != WRITE_BITMAP_QUIET) 900 warning(_(no_split_warning)); 901 write_bitmap_index = 0; 902 } 903 } 904 905 if (!pack_to_stdout) { 906 struct stat st; 907 struct strbuf tmpname = STRBUF_INIT; 908 909 /* 910 * Packs are runtime accessed in their mtime 911 * order since newer packs are more likely to contain 912 * younger objects. So if we are creating multiple 913 * packs then we should modify the mtime of later ones 914 * to preserve this property. 915 */ 916 if (stat(pack_tmp_name, &st) < 0) { 917 warning_errno(_("failed to stat %s"), pack_tmp_name); 918 } else if (!last_mtime) { 919 last_mtime = st.st_mtime; 920 } else { 921 struct utimbuf utb; 922 utb.actime = st.st_atime; 923 utb.modtime = --last_mtime; 924 if (utime(pack_tmp_name, &utb) < 0) 925 warning_errno(_("failed utime() on %s"), pack_tmp_name); 926 } 927 928 strbuf_addf(&tmpname, "%s-", base_name); 929 930 if (write_bitmap_index) { 931 bitmap_writer_set_checksum(oid.hash); 932 bitmap_writer_build_type_index( 933 &to_pack, written_list, nr_written); 934 } 935 936 finish_tmp_packfile(&tmpname, pack_tmp_name, 937 written_list, nr_written, 938 &pack_idx_opts, oid.hash); 939 940 if (write_bitmap_index) { 941 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid)); 942 943 stop_progress(&progress_state); 944 945 bitmap_writer_show_progress(progress); 946 bitmap_writer_reuse_bitmaps(&to_pack); 947 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 948 bitmap_writer_build(&to_pack); 949 bitmap_writer_finish(written_list, nr_written, 950 tmpname.buf, write_bitmap_options); 951 write_bitmap_index = 0; 952 } 953 954 strbuf_release(&tmpname); 955 free(pack_tmp_name); 956 puts(oid_to_hex(&oid)); 957 } 958 959 /* mark written objects as written to previous pack */ 960 for (j = 0; j < nr_written; j++) { 961 written_list[j]->offset = (off_t)-1; 962 } 963 nr_remaining -= nr_written; 964 } while (nr_remaining && i < to_pack.nr_objects); 965 966 free(written_list); 967 free(write_order); 968 stop_progress(&progress_state); 969 if (written != nr_result) 970 die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 971 written, nr_result); 972 trace2_data_intmax("pack-objects", the_repository, 973 "write_pack_file/wrote", nr_result); 974} 975 976static int no_try_delta(const char *path) 977{ 978 static struct attr_check *check; 979 980 if (!check) 981 check = attr_check_initl("delta", NULL); 982 git_check_attr(the_repository->index, path, check); 983 if (ATTR_FALSE(check->items[0].value)) 984 return 1; 985 return 0; 986} 987 988/* 989 * When adding an object, check whether we have already added it 990 * to our packing list. If so, we can skip. However, if we are 991 * being asked to excludei t, but the previous mention was to include 992 * it, make sure to adjust its flags and tweak our numbers accordingly. 993 * 994 * As an optimization, we pass out the index position where we would have 995 * found the item, since that saves us from having to look it up again a 996 * few lines later when we want to add the new entry. 997 */ 998static int have_duplicate_entry(const struct object_id *oid, 999 int exclude,1000 uint32_t *index_pos)1001{1002 struct object_entry *entry;10031004 entry = packlist_find(&to_pack, oid, index_pos);1005 if (!entry)1006 return 0;10071008 if (exclude) {1009 if (!entry->preferred_base)1010 nr_result--;1011 entry->preferred_base = 1;1012 }10131014 return 1;1015}10161017static int want_found_object(int exclude, struct packed_git *p)1018{1019 if (exclude)1020 return 1;1021 if (incremental)1022 return 0;10231024 /*1025 * When asked to do --local (do not include an object that appears in a1026 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1027 * an object that appears in a pack marked with .keep), finding a pack1028 * that matches the criteria is sufficient for us to decide to omit it.1029 * However, even if this pack does not satisfy the criteria, we need to1030 * make sure no copy of this object appears in _any_ pack that makes us1031 * to omit the object, so we need to check all the packs.1032 *1033 * We can however first check whether these options can possible matter;1034 * if they do not matter we know we want the object in generated pack.1035 * Otherwise, we signal "-1" at the end to tell the caller that we do1036 * not know either way, and it needs to check more packs.1037 */1038 if (!ignore_packed_keep_on_disk &&1039 !ignore_packed_keep_in_core &&1040 (!local || !have_non_local_packs))1041 return 1;10421043 if (local && !p->pack_local)1044 return 0;1045 if (p->pack_local &&1046 ((ignore_packed_keep_on_disk && p->pack_keep) ||1047 (ignore_packed_keep_in_core && p->pack_keep_in_core)))1048 return 0;10491050 /* we don't know yet; keep looking for more packs */1051 return -1;1052}10531054/*1055 * Check whether we want the object in the pack (e.g., we do not want1056 * objects found in non-local stores if the "--local" option was used).1057 *1058 * If the caller already knows an existing pack it wants to take the object1059 * from, that is passed in *found_pack and *found_offset; otherwise this1060 * function finds if there is any pack that has the object and returns the pack1061 * and its offset in these variables.1062 */1063static int want_object_in_pack(const struct object_id *oid,1064 int exclude,1065 struct packed_git **found_pack,1066 off_t *found_offset)1067{1068 int want;1069 struct list_head *pos;1070 struct multi_pack_index *m;10711072 if (!exclude && local && has_loose_object_nonlocal(oid))1073 return 0;10741075 /*1076 * If we already know the pack object lives in, start checks from that1077 * pack - in the usual case when neither --local was given nor .keep files1078 * are present we will determine the answer right now.1079 */1080 if (*found_pack) {1081 want = want_found_object(exclude, *found_pack);1082 if (want != -1)1083 return want;1084 }10851086 for (m = get_multi_pack_index(the_repository); m; m = m->next) {1087 struct pack_entry e;1088 if (fill_midx_entry(the_repository, oid, &e, m)) {1089 struct packed_git *p = e.p;1090 off_t offset;10911092 if (p == *found_pack)1093 offset = *found_offset;1094 else1095 offset = find_pack_entry_one(oid->hash, p);10961097 if (offset) {1098 if (!*found_pack) {1099 if (!is_pack_valid(p))1100 continue;1101 *found_offset = offset;1102 *found_pack = p;1103 }1104 want = want_found_object(exclude, p);1105 if (want != -1)1106 return want;1107 }1108 }1109 }11101111 list_for_each(pos, get_packed_git_mru(the_repository)) {1112 struct packed_git *p = list_entry(pos, struct packed_git, mru);1113 off_t offset;11141115 if (p == *found_pack)1116 offset = *found_offset;1117 else1118 offset = find_pack_entry_one(oid->hash, p);11191120 if (offset) {1121 if (!*found_pack) {1122 if (!is_pack_valid(p))1123 continue;1124 *found_offset = offset;1125 *found_pack = p;1126 }1127 want = want_found_object(exclude, p);1128 if (!exclude && want > 0)1129 list_move(&p->mru,1130 get_packed_git_mru(the_repository));1131 if (want != -1)1132 return want;1133 }1134 }11351136 return 1;1137}11381139static void create_object_entry(const struct object_id *oid,1140 enum object_type type,1141 uint32_t hash,1142 int exclude,1143 int no_try_delta,1144 uint32_t index_pos,1145 struct packed_git *found_pack,1146 off_t found_offset)1147{1148 struct object_entry *entry;11491150 entry = packlist_alloc(&to_pack, oid, index_pos);1151 entry->hash = hash;1152 oe_set_type(entry, type);1153 if (exclude)1154 entry->preferred_base = 1;1155 else1156 nr_result++;1157 if (found_pack) {1158 oe_set_in_pack(&to_pack, entry, found_pack);1159 entry->in_pack_offset = found_offset;1160 }11611162 entry->no_try_delta = no_try_delta;1163}11641165static const char no_closure_warning[] = N_(1166"disabling bitmap writing, as some objects are not being packed"1167);11681169static int add_object_entry(const struct object_id *oid, enum object_type type,1170 const char *name, int exclude)1171{1172 struct packed_git *found_pack = NULL;1173 off_t found_offset = 0;1174 uint32_t index_pos;11751176 display_progress(progress_state, ++nr_seen);11771178 if (have_duplicate_entry(oid, exclude, &index_pos))1179 return 0;11801181 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1182 /* The pack is missing an object, so it will not have closure */1183 if (write_bitmap_index) {1184 if (write_bitmap_index != WRITE_BITMAP_QUIET)1185 warning(_(no_closure_warning));1186 write_bitmap_index = 0;1187 }1188 return 0;1189 }11901191 create_object_entry(oid, type, pack_name_hash(name),1192 exclude, name && no_try_delta(name),1193 index_pos, found_pack, found_offset);1194 return 1;1195}11961197static int add_object_entry_from_bitmap(const struct object_id *oid,1198 enum object_type type,1199 int flags, uint32_t name_hash,1200 struct packed_git *pack, off_t offset)1201{1202 uint32_t index_pos;12031204 display_progress(progress_state, ++nr_seen);12051206 if (have_duplicate_entry(oid, 0, &index_pos))1207 return 0;12081209 if (!want_object_in_pack(oid, 0, &pack, &offset))1210 return 0;12111212 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);1213 return 1;1214}12151216struct pbase_tree_cache {1217 struct object_id oid;1218 int ref;1219 int temporary;1220 void *tree_data;1221 unsigned long tree_size;1222};12231224static struct pbase_tree_cache *(pbase_tree_cache[256]);1225static int pbase_tree_cache_ix(const struct object_id *oid)1226{1227 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);1228}1229static int pbase_tree_cache_ix_incr(int ix)1230{1231 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);1232}12331234static struct pbase_tree {1235 struct pbase_tree *next;1236 /* This is a phony "cache" entry; we are not1237 * going to evict it or find it through _get()1238 * mechanism -- this is for the toplevel node that1239 * would almost always change with any commit.1240 */1241 struct pbase_tree_cache pcache;1242} *pbase_tree;12431244static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1245{1246 struct pbase_tree_cache *ent, *nent;1247 void *data;1248 unsigned long size;1249 enum object_type type;1250 int neigh;1251 int my_ix = pbase_tree_cache_ix(oid);1252 int available_ix = -1;12531254 /* pbase-tree-cache acts as a limited hashtable.1255 * your object will be found at your index or within a few1256 * slots after that slot if it is cached.1257 */1258 for (neigh = 0; neigh < 8; neigh++) {1259 ent = pbase_tree_cache[my_ix];1260 if (ent && oideq(&ent->oid, oid)) {1261 ent->ref++;1262 return ent;1263 }1264 else if (((available_ix < 0) && (!ent || !ent->ref)) ||1265 ((0 <= available_ix) &&1266 (!ent && pbase_tree_cache[available_ix])))1267 available_ix = my_ix;1268 if (!ent)1269 break;1270 my_ix = pbase_tree_cache_ix_incr(my_ix);1271 }12721273 /* Did not find one. Either we got a bogus request or1274 * we need to read and perhaps cache.1275 */1276 data = read_object_file(oid, &type, &size);1277 if (!data)1278 return NULL;1279 if (type != OBJ_TREE) {1280 free(data);1281 return NULL;1282 }12831284 /* We need to either cache or return a throwaway copy */12851286 if (available_ix < 0)1287 ent = NULL;1288 else {1289 ent = pbase_tree_cache[available_ix];1290 my_ix = available_ix;1291 }12921293 if (!ent) {1294 nent = xmalloc(sizeof(*nent));1295 nent->temporary = (available_ix < 0);1296 }1297 else {1298 /* evict and reuse */1299 free(ent->tree_data);1300 nent = ent;1301 }1302 oidcpy(&nent->oid, oid);1303 nent->tree_data = data;1304 nent->tree_size = size;1305 nent->ref = 1;1306 if (!nent->temporary)1307 pbase_tree_cache[my_ix] = nent;1308 return nent;1309}13101311static void pbase_tree_put(struct pbase_tree_cache *cache)1312{1313 if (!cache->temporary) {1314 cache->ref--;1315 return;1316 }1317 free(cache->tree_data);1318 free(cache);1319}13201321static int name_cmp_len(const char *name)1322{1323 int i;1324 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)1325 ;1326 return i;1327}13281329static void add_pbase_object(struct tree_desc *tree,1330 const char *name,1331 int cmplen,1332 const char *fullname)1333{1334 struct name_entry entry;1335 int cmp;13361337 while (tree_entry(tree,&entry)) {1338 if (S_ISGITLINK(entry.mode))1339 continue;1340 cmp = tree_entry_len(&entry) != cmplen ? 1 :1341 memcmp(name, entry.path, cmplen);1342 if (cmp > 0)1343 continue;1344 if (cmp < 0)1345 return;1346 if (name[cmplen] != '/') {1347 add_object_entry(&entry.oid,1348 object_type(entry.mode),1349 fullname, 1);1350 return;1351 }1352 if (S_ISDIR(entry.mode)) {1353 struct tree_desc sub;1354 struct pbase_tree_cache *tree;1355 const char *down = name+cmplen+1;1356 int downlen = name_cmp_len(down);13571358 tree = pbase_tree_get(&entry.oid);1359 if (!tree)1360 return;1361 init_tree_desc(&sub, tree->tree_data, tree->tree_size);13621363 add_pbase_object(&sub, down, downlen, fullname);1364 pbase_tree_put(tree);1365 }1366 }1367}13681369static unsigned *done_pbase_paths;1370static int done_pbase_paths_num;1371static int done_pbase_paths_alloc;1372static int done_pbase_path_pos(unsigned hash)1373{1374 int lo = 0;1375 int hi = done_pbase_paths_num;1376 while (lo < hi) {1377 int mi = lo + (hi - lo) / 2;1378 if (done_pbase_paths[mi] == hash)1379 return mi;1380 if (done_pbase_paths[mi] < hash)1381 hi = mi;1382 else1383 lo = mi + 1;1384 }1385 return -lo-1;1386}13871388static int check_pbase_path(unsigned hash)1389{1390 int pos = done_pbase_path_pos(hash);1391 if (0 <= pos)1392 return 1;1393 pos = -pos - 1;1394 ALLOC_GROW(done_pbase_paths,1395 done_pbase_paths_num + 1,1396 done_pbase_paths_alloc);1397 done_pbase_paths_num++;1398 if (pos < done_pbase_paths_num)1399 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,1400 done_pbase_paths_num - pos - 1);1401 done_pbase_paths[pos] = hash;1402 return 0;1403}14041405static void add_preferred_base_object(const char *name)1406{1407 struct pbase_tree *it;1408 int cmplen;1409 unsigned hash = pack_name_hash(name);14101411 if (!num_preferred_base || check_pbase_path(hash))1412 return;14131414 cmplen = name_cmp_len(name);1415 for (it = pbase_tree; it; it = it->next) {1416 if (cmplen == 0) {1417 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);1418 }1419 else {1420 struct tree_desc tree;1421 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1422 add_pbase_object(&tree, name, cmplen, name);1423 }1424 }1425}14261427static void add_preferred_base(struct object_id *oid)1428{1429 struct pbase_tree *it;1430 void *data;1431 unsigned long size;1432 struct object_id tree_oid;14331434 if (window <= num_preferred_base++)1435 return;14361437 data = read_object_with_reference(the_repository, oid,1438 tree_type, &size, &tree_oid);1439 if (!data)1440 return;14411442 for (it = pbase_tree; it; it = it->next) {1443 if (oideq(&it->pcache.oid, &tree_oid)) {1444 free(data);1445 return;1446 }1447 }14481449 it = xcalloc(1, sizeof(*it));1450 it->next = pbase_tree;1451 pbase_tree = it;14521453 oidcpy(&it->pcache.oid, &tree_oid);1454 it->pcache.tree_data = data;1455 it->pcache.tree_size = size;1456}14571458static void cleanup_preferred_base(void)1459{1460 struct pbase_tree *it;1461 unsigned i;14621463 it = pbase_tree;1464 pbase_tree = NULL;1465 while (it) {1466 struct pbase_tree *tmp = it;1467 it = tmp->next;1468 free(tmp->pcache.tree_data);1469 free(tmp);1470 }14711472 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {1473 if (!pbase_tree_cache[i])1474 continue;1475 free(pbase_tree_cache[i]->tree_data);1476 FREE_AND_NULL(pbase_tree_cache[i]);1477 }14781479 FREE_AND_NULL(done_pbase_paths);1480 done_pbase_paths_num = done_pbase_paths_alloc = 0;1481}14821483/*1484 * Return 1 iff the object specified by "delta" can be sent1485 * literally as a delta against the base in "base_sha1". If1486 * so, then *base_out will point to the entry in our packing1487 * list, or NULL if we must use the external-base list.1488 *1489 * Depth value does not matter - find_deltas() will1490 * never consider reused delta as the base object to1491 * deltify other objects against, in order to avoid1492 * circular deltas.1493 */1494static int can_reuse_delta(const unsigned char *base_sha1,1495 struct object_entry *delta,1496 struct object_entry **base_out)1497{1498 struct object_entry *base;1499 struct object_id base_oid;15001501 if (!base_sha1)1502 return 0;15031504 oidread(&base_oid, base_sha1);15051506 /*1507 * First see if we're already sending the base (or it's explicitly in1508 * our "excluded" list).1509 */1510 base = packlist_find(&to_pack, &base_oid, NULL);1511 if (base) {1512 if (!in_same_island(&delta->idx.oid, &base->idx.oid))1513 return 0;1514 *base_out = base;1515 return 1;1516 }15171518 /*1519 * Otherwise, reachability bitmaps may tell us if the receiver has it,1520 * even if it was buried too deep in history to make it into the1521 * packing list.1522 */1523 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {1524 if (use_delta_islands) {1525 if (!in_same_island(&delta->idx.oid, &base_oid))1526 return 0;1527 }1528 *base_out = NULL;1529 return 1;1530 }15311532 return 0;1533}15341535static void check_object(struct object_entry *entry)1536{1537 unsigned long canonical_size;15381539 if (IN_PACK(entry)) {1540 struct packed_git *p = IN_PACK(entry);1541 struct pack_window *w_curs = NULL;1542 const unsigned char *base_ref = NULL;1543 struct object_entry *base_entry;1544 unsigned long used, used_0;1545 unsigned long avail;1546 off_t ofs;1547 unsigned char *buf, c;1548 enum object_type type;1549 unsigned long in_pack_size;15501551 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);15521553 /*1554 * We want in_pack_type even if we do not reuse delta1555 * since non-delta representations could still be reused.1556 */1557 used = unpack_object_header_buffer(buf, avail,1558 &type,1559 &in_pack_size);1560 if (used == 0)1561 goto give_up;15621563 if (type < 0)1564 BUG("invalid type %d", type);1565 entry->in_pack_type = type;15661567 /*1568 * Determine if this is a delta and if so whether we can1569 * reuse it or not. Otherwise let's find out as cheaply as1570 * possible what the actual type and size for this object is.1571 */1572 switch (entry->in_pack_type) {1573 default:1574 /* Not a delta hence we've already got all we need. */1575 oe_set_type(entry, entry->in_pack_type);1576 SET_SIZE(entry, in_pack_size);1577 entry->in_pack_header_size = used;1578 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)1579 goto give_up;1580 unuse_pack(&w_curs);1581 return;1582 case OBJ_REF_DELTA:1583 if (reuse_delta && !entry->preferred_base)1584 base_ref = use_pack(p, &w_curs,1585 entry->in_pack_offset + used, NULL);1586 entry->in_pack_header_size = used + the_hash_algo->rawsz;1587 break;1588 case OBJ_OFS_DELTA:1589 buf = use_pack(p, &w_curs,1590 entry->in_pack_offset + used, NULL);1591 used_0 = 0;1592 c = buf[used_0++];1593 ofs = c & 127;1594 while (c & 128) {1595 ofs += 1;1596 if (!ofs || MSB(ofs, 7)) {1597 error(_("delta base offset overflow in pack for %s"),1598 oid_to_hex(&entry->idx.oid));1599 goto give_up;1600 }1601 c = buf[used_0++];1602 ofs = (ofs << 7) + (c & 127);1603 }1604 ofs = entry->in_pack_offset - ofs;1605 if (ofs <= 0 || ofs >= entry->in_pack_offset) {1606 error(_("delta base offset out of bound for %s"),1607 oid_to_hex(&entry->idx.oid));1608 goto give_up;1609 }1610 if (reuse_delta && !entry->preferred_base) {1611 struct revindex_entry *revidx;1612 revidx = find_pack_revindex(p, ofs);1613 if (!revidx)1614 goto give_up;1615 base_ref = nth_packed_object_sha1(p, revidx->nr);1616 }1617 entry->in_pack_header_size = used + used_0;1618 break;1619 }16201621 if (can_reuse_delta(base_ref, entry, &base_entry)) {1622 oe_set_type(entry, entry->in_pack_type);1623 SET_SIZE(entry, in_pack_size); /* delta size */1624 SET_DELTA_SIZE(entry, in_pack_size);16251626 if (base_entry) {1627 SET_DELTA(entry, base_entry);1628 entry->delta_sibling_idx = base_entry->delta_child_idx;1629 SET_DELTA_CHILD(base_entry, entry);1630 } else {1631 SET_DELTA_EXT(entry, base_ref);1632 }16331634 unuse_pack(&w_curs);1635 return;1636 }16371638 if (oe_type(entry)) {1639 off_t delta_pos;16401641 /*1642 * This must be a delta and we already know what the1643 * final object type is. Let's extract the actual1644 * object size from the delta header.1645 */1646 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1647 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);1648 if (canonical_size == 0)1649 goto give_up;1650 SET_SIZE(entry, canonical_size);1651 unuse_pack(&w_curs);1652 return;1653 }16541655 /*1656 * No choice but to fall back to the recursive delta walk1657 * with oid_object_info() to find about the object type1658 * at this point...1659 */1660 give_up:1661 unuse_pack(&w_curs);1662 }16631664 oe_set_type(entry,1665 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1666 if (entry->type_valid) {1667 SET_SIZE(entry, canonical_size);1668 } else {1669 /*1670 * Bad object type is checked in prepare_pack(). This is1671 * to permit a missing preferred base object to be ignored1672 * as a preferred base. Doing so can result in a larger1673 * pack file, but the transfer will still take place.1674 */1675 }1676}16771678static int pack_offset_sort(const void *_a, const void *_b)1679{1680 const struct object_entry *a = *(struct object_entry **)_a;1681 const struct object_entry *b = *(struct object_entry **)_b;1682 const struct packed_git *a_in_pack = IN_PACK(a);1683 const struct packed_git *b_in_pack = IN_PACK(b);16841685 /* avoid filesystem trashing with loose objects */1686 if (!a_in_pack && !b_in_pack)1687 return oidcmp(&a->idx.oid, &b->idx.oid);16881689 if (a_in_pack < b_in_pack)1690 return -1;1691 if (a_in_pack > b_in_pack)1692 return 1;1693 return a->in_pack_offset < b->in_pack_offset ? -1 :1694 (a->in_pack_offset > b->in_pack_offset);1695}16961697/*1698 * Drop an on-disk delta we were planning to reuse. Naively, this would1699 * just involve blanking out the "delta" field, but we have to deal1700 * with some extra book-keeping:1701 *1702 * 1. Removing ourselves from the delta_sibling linked list.1703 *1704 * 2. Updating our size/type to the non-delta representation. These were1705 * either not recorded initially (size) or overwritten with the delta type1706 * (type) when check_object() decided to reuse the delta.1707 *1708 * 3. Resetting our delta depth, as we are now a base object.1709 */1710static void drop_reused_delta(struct object_entry *entry)1711{1712 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;1713 struct object_info oi = OBJECT_INFO_INIT;1714 enum object_type type;1715 unsigned long size;17161717 while (*idx) {1718 struct object_entry *oe = &to_pack.objects[*idx - 1];17191720 if (oe == entry)1721 *idx = oe->delta_sibling_idx;1722 else1723 idx = &oe->delta_sibling_idx;1724 }1725 SET_DELTA(entry, NULL);1726 entry->depth = 0;17271728 oi.sizep = &size;1729 oi.typep = &type;1730 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {1731 /*1732 * We failed to get the info from this pack for some reason;1733 * fall back to oid_object_info, which may find another copy.1734 * And if that fails, the error will be recorded in oe_type(entry)1735 * and dealt with in prepare_pack().1736 */1737 oe_set_type(entry,1738 oid_object_info(the_repository, &entry->idx.oid, &size));1739 } else {1740 oe_set_type(entry, type);1741 }1742 SET_SIZE(entry, size);1743}17441745/*1746 * Follow the chain of deltas from this entry onward, throwing away any links1747 * that cause us to hit a cycle (as determined by the DFS state flags in1748 * the entries).1749 *1750 * We also detect too-long reused chains that would violate our --depth1751 * limit.1752 */1753static void break_delta_chains(struct object_entry *entry)1754{1755 /*1756 * The actual depth of each object we will write is stored as an int,1757 * as it cannot exceed our int "depth" limit. But before we break1758 * changes based no that limit, we may potentially go as deep as the1759 * number of objects, which is elsewhere bounded to a uint32_t.1760 */1761 uint32_t total_depth;1762 struct object_entry *cur, *next;17631764 for (cur = entry, total_depth = 0;1765 cur;1766 cur = DELTA(cur), total_depth++) {1767 if (cur->dfs_state == DFS_DONE) {1768 /*1769 * We've already seen this object and know it isn't1770 * part of a cycle. We do need to append its depth1771 * to our count.1772 */1773 total_depth += cur->depth;1774 break;1775 }17761777 /*1778 * We break cycles before looping, so an ACTIVE state (or any1779 * other cruft which made its way into the state variable)1780 * is a bug.1781 */1782 if (cur->dfs_state != DFS_NONE)1783 BUG("confusing delta dfs state in first pass: %d",1784 cur->dfs_state);17851786 /*1787 * Now we know this is the first time we've seen the object. If1788 * it's not a delta, we're done traversing, but we'll mark it1789 * done to save time on future traversals.1790 */1791 if (!DELTA(cur)) {1792 cur->dfs_state = DFS_DONE;1793 break;1794 }17951796 /*1797 * Mark ourselves as active and see if the next step causes1798 * us to cycle to another active object. It's important to do1799 * this _before_ we loop, because it impacts where we make the1800 * cut, and thus how our total_depth counter works.1801 * E.g., We may see a partial loop like:1802 *1803 * A -> B -> C -> D -> B1804 *1805 * Cutting B->C breaks the cycle. But now the depth of A is1806 * only 1, and our total_depth counter is at 3. The size of the1807 * error is always one less than the size of the cycle we1808 * broke. Commits C and D were "lost" from A's chain.1809 *1810 * If we instead cut D->B, then the depth of A is correct at 3.1811 * We keep all commits in the chain that we examined.1812 */1813 cur->dfs_state = DFS_ACTIVE;1814 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {1815 drop_reused_delta(cur);1816 cur->dfs_state = DFS_DONE;1817 break;1818 }1819 }18201821 /*1822 * And now that we've gone all the way to the bottom of the chain, we1823 * need to clear the active flags and set the depth fields as1824 * appropriate. Unlike the loop above, which can quit when it drops a1825 * delta, we need to keep going to look for more depth cuts. So we need1826 * an extra "next" pointer to keep going after we reset cur->delta.1827 */1828 for (cur = entry; cur; cur = next) {1829 next = DELTA(cur);18301831 /*1832 * We should have a chain of zero or more ACTIVE states down to1833 * a final DONE. We can quit after the DONE, because either it1834 * has no bases, or we've already handled them in a previous1835 * call.1836 */1837 if (cur->dfs_state == DFS_DONE)1838 break;1839 else if (cur->dfs_state != DFS_ACTIVE)1840 BUG("confusing delta dfs state in second pass: %d",1841 cur->dfs_state);18421843 /*1844 * If the total_depth is more than depth, then we need to snip1845 * the chain into two or more smaller chains that don't exceed1846 * the maximum depth. Most of the resulting chains will contain1847 * (depth + 1) entries (i.e., depth deltas plus one base), and1848 * the last chain (i.e., the one containing entry) will contain1849 * whatever entries are left over, namely1850 * (total_depth % (depth + 1)) of them.1851 *1852 * Since we are iterating towards decreasing depth, we need to1853 * decrement total_depth as we go, and we need to write to the1854 * entry what its final depth will be after all of the1855 * snipping. Since we're snipping into chains of length (depth1856 * + 1) entries, the final depth of an entry will be its1857 * original depth modulo (depth + 1). Any time we encounter an1858 * entry whose final depth is supposed to be zero, we snip it1859 * from its delta base, thereby making it so.1860 */1861 cur->depth = (total_depth--) % (depth + 1);1862 if (!cur->depth)1863 drop_reused_delta(cur);18641865 cur->dfs_state = DFS_DONE;1866 }1867}18681869static void get_object_details(void)1870{1871 uint32_t i;1872 struct object_entry **sorted_by_offset;18731874 if (progress)1875 progress_state = start_progress(_("Counting objects"),1876 to_pack.nr_objects);18771878 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));1879 for (i = 0; i < to_pack.nr_objects; i++)1880 sorted_by_offset[i] = to_pack.objects + i;1881 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18821883 for (i = 0; i < to_pack.nr_objects; i++) {1884 struct object_entry *entry = sorted_by_offset[i];1885 check_object(entry);1886 if (entry->type_valid &&1887 oe_size_greater_than(&to_pack, entry, big_file_threshold))1888 entry->no_try_delta = 1;1889 display_progress(progress_state, i + 1);1890 }1891 stop_progress(&progress_state);18921893 /*1894 * This must happen in a second pass, since we rely on the delta1895 * information for the whole list being completed.1896 */1897 for (i = 0; i < to_pack.nr_objects; i++)1898 break_delta_chains(&to_pack.objects[i]);18991900 free(sorted_by_offset);1901}19021903/*1904 * We search for deltas in a list sorted by type, by filename hash, and then1905 * by size, so that we see progressively smaller and smaller files.1906 * That's because we prefer deltas to be from the bigger file1907 * to the smaller -- deletes are potentially cheaper, but perhaps1908 * more importantly, the bigger file is likely the more recent1909 * one. The deepest deltas are therefore the oldest objects which are1910 * less susceptible to be accessed often.1911 */1912static int type_size_sort(const void *_a, const void *_b)1913{1914 const struct object_entry *a = *(struct object_entry **)_a;1915 const struct object_entry *b = *(struct object_entry **)_b;1916 const enum object_type a_type = oe_type(a);1917 const enum object_type b_type = oe_type(b);1918 const unsigned long a_size = SIZE(a);1919 const unsigned long b_size = SIZE(b);19201921 if (a_type > b_type)1922 return -1;1923 if (a_type < b_type)1924 return 1;1925 if (a->hash > b->hash)1926 return -1;1927 if (a->hash < b->hash)1928 return 1;1929 if (a->preferred_base > b->preferred_base)1930 return -1;1931 if (a->preferred_base < b->preferred_base)1932 return 1;1933 if (use_delta_islands) {1934 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);1935 if (island_cmp)1936 return island_cmp;1937 }1938 if (a_size > b_size)1939 return -1;1940 if (a_size < b_size)1941 return 1;1942 return a < b ? -1 : (a > b); /* newest first */1943}19441945struct unpacked {1946 struct object_entry *entry;1947 void *data;1948 struct delta_index *index;1949 unsigned depth;1950};19511952static int delta_cacheable(unsigned long src_size, unsigned long trg_size,1953 unsigned long delta_size)1954{1955 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1956 return 0;19571958 if (delta_size < cache_max_small_delta_size)1959 return 1;19601961 /* cache delta, if objects are large enough compared to delta size */1962 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))1963 return 1;19641965 return 0;1966}19671968/* Protect delta_cache_size */1969static pthread_mutex_t cache_mutex;1970#define cache_lock() pthread_mutex_lock(&cache_mutex)1971#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19721973/*1974 * Protect object list partitioning (e.g. struct thread_param) and1975 * progress_state1976 */1977static pthread_mutex_t progress_mutex;1978#define progress_lock() pthread_mutex_lock(&progress_mutex)1979#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19801981/*1982 * Access to struct object_entry is unprotected since each thread owns1983 * a portion of the main object list. Just don't access object entries1984 * ahead in the list because they can be stolen and would need1985 * progress_mutex for protection.1986 */19871988/*1989 * Return the size of the object without doing any delta1990 * reconstruction (so non-deltas are true object sizes, but deltas1991 * return the size of the delta data).1992 */1993unsigned long oe_get_size_slow(struct packing_data *pack,1994 const struct object_entry *e)1995{1996 struct packed_git *p;1997 struct pack_window *w_curs;1998 unsigned char *buf;1999 enum object_type type;2000 unsigned long used, avail, size;20012002 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {2003 packing_data_lock(&to_pack);2004 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)2005 die(_("unable to get size of %s"),2006 oid_to_hex(&e->idx.oid));2007 packing_data_unlock(&to_pack);2008 return size;2009 }20102011 p = oe_in_pack(pack, e);2012 if (!p)2013 BUG("when e->type is a delta, it must belong to a pack");20142015 packing_data_lock(&to_pack);2016 w_curs = NULL;2017 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);2018 used = unpack_object_header_buffer(buf, avail, &type, &size);2019 if (used == 0)2020 die(_("unable to parse object header of %s"),2021 oid_to_hex(&e->idx.oid));20222023 unuse_pack(&w_curs);2024 packing_data_unlock(&to_pack);2025 return size;2026}20272028static int try_delta(struct unpacked *trg, struct unpacked *src,2029 unsigned max_depth, unsigned long *mem_usage)2030{2031 struct object_entry *trg_entry = trg->entry;2032 struct object_entry *src_entry = src->entry;2033 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2034 unsigned ref_depth;2035 enum object_type type;2036 void *delta_buf;20372038 /* Don't bother doing diffs between different types */2039 if (oe_type(trg_entry) != oe_type(src_entry))2040 return -1;20412042 /*2043 * We do not bother to try a delta that we discarded on an2044 * earlier try, but only when reusing delta data. Note that2045 * src_entry that is marked as the preferred_base should always2046 * be considered, as even if we produce a suboptimal delta against2047 * it, we will still save the transfer cost, as we already know2048 * the other side has it and we won't send src_entry at all.2049 */2050 if (reuse_delta && IN_PACK(trg_entry) &&2051 IN_PACK(trg_entry) == IN_PACK(src_entry) &&2052 !src_entry->preferred_base &&2053 trg_entry->in_pack_type != OBJ_REF_DELTA &&2054 trg_entry->in_pack_type != OBJ_OFS_DELTA)2055 return 0;20562057 /* Let's not bust the allowed depth. */2058 if (src->depth >= max_depth)2059 return 0;20602061 /* Now some size filtering heuristics. */2062 trg_size = SIZE(trg_entry);2063 if (!DELTA(trg_entry)) {2064 max_size = trg_size/2 - the_hash_algo->rawsz;2065 ref_depth = 1;2066 } else {2067 max_size = DELTA_SIZE(trg_entry);2068 ref_depth = trg->depth;2069 }2070 max_size = (uint64_t)max_size * (max_depth - src->depth) /2071 (max_depth - ref_depth + 1);2072 if (max_size == 0)2073 return 0;2074 src_size = SIZE(src_entry);2075 sizediff = src_size < trg_size ? trg_size - src_size : 0;2076 if (sizediff >= max_size)2077 return 0;2078 if (trg_size < src_size / 32)2079 return 0;20802081 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2082 return 0;20832084 /* Load data if not already done */2085 if (!trg->data) {2086 packing_data_lock(&to_pack);2087 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);2088 packing_data_unlock(&to_pack);2089 if (!trg->data)2090 die(_("object %s cannot be read"),2091 oid_to_hex(&trg_entry->idx.oid));2092 if (sz != trg_size)2093 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2094 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2095 (uintmax_t)trg_size);2096 *mem_usage += sz;2097 }2098 if (!src->data) {2099 packing_data_lock(&to_pack);2100 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);2101 packing_data_unlock(&to_pack);2102 if (!src->data) {2103 if (src_entry->preferred_base) {2104 static int warned = 0;2105 if (!warned++)2106 warning(_("object %s cannot be read"),2107 oid_to_hex(&src_entry->idx.oid));2108 /*2109 * Those objects are not included in the2110 * resulting pack. Be resilient and ignore2111 * them if they can't be read, in case the2112 * pack could be created nevertheless.2113 */2114 return 0;2115 }2116 die(_("object %s cannot be read"),2117 oid_to_hex(&src_entry->idx.oid));2118 }2119 if (sz != src_size)2120 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2121 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2122 (uintmax_t)src_size);2123 *mem_usage += sz;2124 }2125 if (!src->index) {2126 src->index = create_delta_index(src->data, src_size);2127 if (!src->index) {2128 static int warned = 0;2129 if (!warned++)2130 warning(_("suboptimal pack - out of memory"));2131 return 0;2132 }2133 *mem_usage += sizeof_delta_index(src->index);2134 }21352136 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2137 if (!delta_buf)2138 return 0;21392140 if (DELTA(trg_entry)) {2141 /* Prefer only shallower same-sized deltas. */2142 if (delta_size == DELTA_SIZE(trg_entry) &&2143 src->depth + 1 >= trg->depth) {2144 free(delta_buf);2145 return 0;2146 }2147 }21482149 /*2150 * Handle memory allocation outside of the cache2151 * accounting lock. Compiler will optimize the strangeness2152 * away when NO_PTHREADS is defined.2153 */2154 free(trg_entry->delta_data);2155 cache_lock();2156 if (trg_entry->delta_data) {2157 delta_cache_size -= DELTA_SIZE(trg_entry);2158 trg_entry->delta_data = NULL;2159 }2160 if (delta_cacheable(src_size, trg_size, delta_size)) {2161 delta_cache_size += delta_size;2162 cache_unlock();2163 trg_entry->delta_data = xrealloc(delta_buf, delta_size);2164 } else {2165 cache_unlock();2166 free(delta_buf);2167 }21682169 SET_DELTA(trg_entry, src_entry);2170 SET_DELTA_SIZE(trg_entry, delta_size);2171 trg->depth = src->depth + 1;21722173 return 1;2174}21752176static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)2177{2178 struct object_entry *child = DELTA_CHILD(me);2179 unsigned int m = n;2180 while (child) {2181 const unsigned int c = check_delta_limit(child, n + 1);2182 if (m < c)2183 m = c;2184 child = DELTA_SIBLING(child);2185 }2186 return m;2187}21882189static unsigned long free_unpacked(struct unpacked *n)2190{2191 unsigned long freed_mem = sizeof_delta_index(n->index);2192 free_delta_index(n->index);2193 n->index = NULL;2194 if (n->data) {2195 freed_mem += SIZE(n->entry);2196 FREE_AND_NULL(n->data);2197 }2198 n->entry = NULL;2199 n->depth = 0;2200 return freed_mem;2201}22022203static void find_deltas(struct object_entry **list, unsigned *list_size,2204 int window, int depth, unsigned *processed)2205{2206 uint32_t i, idx = 0, count = 0;2207 struct unpacked *array;2208 unsigned long mem_usage = 0;22092210 array = xcalloc(window, sizeof(struct unpacked));22112212 for (;;) {2213 struct object_entry *entry;2214 struct unpacked *n = array + idx;2215 int j, max_depth, best_base = -1;22162217 progress_lock();2218 if (!*list_size) {2219 progress_unlock();2220 break;2221 }2222 entry = *list++;2223 (*list_size)--;2224 if (!entry->preferred_base) {2225 (*processed)++;2226 display_progress(progress_state, *processed);2227 }2228 progress_unlock();22292230 mem_usage -= free_unpacked(n);2231 n->entry = entry;22322233 while (window_memory_limit &&2234 mem_usage > window_memory_limit &&2235 count > 1) {2236 const uint32_t tail = (idx + window - count) % window;2237 mem_usage -= free_unpacked(array + tail);2238 count--;2239 }22402241 /* We do not compute delta to *create* objects we are not2242 * going to pack.2243 */2244 if (entry->preferred_base)2245 goto next;22462247 /*2248 * If the current object is at pack edge, take the depth the2249 * objects that depend on the current object into account2250 * otherwise they would become too deep.2251 */2252 max_depth = depth;2253 if (DELTA_CHILD(entry)) {2254 max_depth -= check_delta_limit(entry, 0);2255 if (max_depth <= 0)2256 goto next;2257 }22582259 j = window;2260 while (--j > 0) {2261 int ret;2262 uint32_t other_idx = idx + j;2263 struct unpacked *m;2264 if (other_idx >= window)2265 other_idx -= window;2266 m = array + other_idx;2267 if (!m->entry)2268 break;2269 ret = try_delta(n, m, max_depth, &mem_usage);2270 if (ret < 0)2271 break;2272 else if (ret > 0)2273 best_base = other_idx;2274 }22752276 /*2277 * If we decided to cache the delta data, then it is best2278 * to compress it right away. First because we have to do2279 * it anyway, and doing it here while we're threaded will2280 * save a lot of time in the non threaded write phase,2281 * as well as allow for caching more deltas within2282 * the same cache size limit.2283 * ...2284 * But only if not writing to stdout, since in that case2285 * the network is most likely throttling writes anyway,2286 * and therefore it is best to go to the write phase ASAP2287 * instead, as we can afford spending more time compressing2288 * between writes at that moment.2289 */2290 if (entry->delta_data && !pack_to_stdout) {2291 unsigned long size;22922293 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));2294 if (size < (1U << OE_Z_DELTA_BITS)) {2295 entry->z_delta_size = size;2296 cache_lock();2297 delta_cache_size -= DELTA_SIZE(entry);2298 delta_cache_size += entry->z_delta_size;2299 cache_unlock();2300 } else {2301 FREE_AND_NULL(entry->delta_data);2302 entry->z_delta_size = 0;2303 }2304 }23052306 /* if we made n a delta, and if n is already at max2307 * depth, leaving it in the window is pointless. we2308 * should evict it first.2309 */2310 if (DELTA(entry) && max_depth <= n->depth)2311 continue;23122313 /*2314 * Move the best delta base up in the window, after the2315 * currently deltified object, to keep it longer. It will2316 * be the first base object to be attempted next.2317 */2318 if (DELTA(entry)) {2319 struct unpacked swap = array[best_base];2320 int dist = (window + idx - best_base) % window;2321 int dst = best_base;2322 while (dist--) {2323 int src = (dst + 1) % window;2324 array[dst] = array[src];2325 dst = src;2326 }2327 array[dst] = swap;2328 }23292330 next:2331 idx++;2332 if (count + 1 < window)2333 count++;2334 if (idx >= window)2335 idx = 0;2336 }23372338 for (i = 0; i < window; ++i) {2339 free_delta_index(array[i].index);2340 free(array[i].data);2341 }2342 free(array);2343}23442345static void try_to_free_from_threads(size_t size)2346{2347 packing_data_lock(&to_pack);2348 release_pack_memory(size);2349 packing_data_unlock(&to_pack);2350}23512352static try_to_free_t old_try_to_free_routine;23532354/*2355 * The main object list is split into smaller lists, each is handed to2356 * one worker.2357 *2358 * The main thread waits on the condition that (at least) one of the workers2359 * has stopped working (which is indicated in the .working member of2360 * struct thread_params).2361 *2362 * When a work thread has completed its work, it sets .working to 0 and2363 * signals the main thread and waits on the condition that .data_ready2364 * becomes 1.2365 *2366 * The main thread steals half of the work from the worker that has2367 * most work left to hand it to the idle worker.2368 */23692370struct thread_params {2371 pthread_t thread;2372 struct object_entry **list;2373 unsigned list_size;2374 unsigned remaining;2375 int window;2376 int depth;2377 int working;2378 int data_ready;2379 pthread_mutex_t mutex;2380 pthread_cond_t cond;2381 unsigned *processed;2382};23832384static pthread_cond_t progress_cond;23852386/*2387 * Mutex and conditional variable can't be statically-initialized on Windows.2388 */2389static void init_threaded_search(void)2390{2391 pthread_mutex_init(&cache_mutex, NULL);2392 pthread_mutex_init(&progress_mutex, NULL);2393 pthread_cond_init(&progress_cond, NULL);2394 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);2395}23962397static void cleanup_threaded_search(void)2398{2399 set_try_to_free_routine(old_try_to_free_routine);2400 pthread_cond_destroy(&progress_cond);2401 pthread_mutex_destroy(&cache_mutex);2402 pthread_mutex_destroy(&progress_mutex);2403}24042405static void *threaded_find_deltas(void *arg)2406{2407 struct thread_params *me = arg;24082409 progress_lock();2410 while (me->remaining) {2411 progress_unlock();24122413 find_deltas(me->list, &me->remaining,2414 me->window, me->depth, me->processed);24152416 progress_lock();2417 me->working = 0;2418 pthread_cond_signal(&progress_cond);2419 progress_unlock();24202421 /*2422 * We must not set ->data_ready before we wait on the2423 * condition because the main thread may have set it to 12424 * before we get here. In order to be sure that new2425 * work is available if we see 1 in ->data_ready, it2426 * was initialized to 0 before this thread was spawned2427 * and we reset it to 0 right away.2428 */2429 pthread_mutex_lock(&me->mutex);2430 while (!me->data_ready)2431 pthread_cond_wait(&me->cond, &me->mutex);2432 me->data_ready = 0;2433 pthread_mutex_unlock(&me->mutex);24342435 progress_lock();2436 }2437 progress_unlock();2438 /* leave ->working 1 so that this doesn't get more work assigned */2439 return NULL;2440}24412442static void ll_find_deltas(struct object_entry **list, unsigned list_size,2443 int window, int depth, unsigned *processed)2444{2445 struct thread_params *p;2446 int i, ret, active_threads = 0;24472448 init_threaded_search();24492450 if (delta_search_threads <= 1) {2451 find_deltas(list, &list_size, window, depth, processed);2452 cleanup_threaded_search();2453 return;2454 }2455 if (progress > pack_to_stdout)2456 fprintf_ln(stderr, _("Delta compression using up to %d threads"),2457 delta_search_threads);2458 p = xcalloc(delta_search_threads, sizeof(*p));24592460 /* Partition the work amongst work threads. */2461 for (i = 0; i < delta_search_threads; i++) {2462 unsigned sub_size = list_size / (delta_search_threads - i);24632464 /* don't use too small segments or no deltas will be found */2465 if (sub_size < 2*window && i+1 < delta_search_threads)2466 sub_size = 0;24672468 p[i].window = window;2469 p[i].depth = depth;2470 p[i].processed = processed;2471 p[i].working = 1;2472 p[i].data_ready = 0;24732474 /* try to split chunks on "path" boundaries */2475 while (sub_size && sub_size < list_size &&2476 list[sub_size]->hash &&2477 list[sub_size]->hash == list[sub_size-1]->hash)2478 sub_size++;24792480 p[i].list = list;2481 p[i].list_size = sub_size;2482 p[i].remaining = sub_size;24832484 list += sub_size;2485 list_size -= sub_size;2486 }24872488 /* Start work threads. */2489 for (i = 0; i < delta_search_threads; i++) {2490 if (!p[i].list_size)2491 continue;2492 pthread_mutex_init(&p[i].mutex, NULL);2493 pthread_cond_init(&p[i].cond, NULL);2494 ret = pthread_create(&p[i].thread, NULL,2495 threaded_find_deltas, &p[i]);2496 if (ret)2497 die(_("unable to create thread: %s"), strerror(ret));2498 active_threads++;2499 }25002501 /*2502 * Now let's wait for work completion. Each time a thread is done2503 * with its work, we steal half of the remaining work from the2504 * thread with the largest number of unprocessed objects and give2505 * it to that newly idle thread. This ensure good load balancing2506 * until the remaining object list segments are simply too short2507 * to be worth splitting anymore.2508 */2509 while (active_threads) {2510 struct thread_params *target = NULL;2511 struct thread_params *victim = NULL;2512 unsigned sub_size = 0;25132514 progress_lock();2515 for (;;) {2516 for (i = 0; !target && i < delta_search_threads; i++)2517 if (!p[i].working)2518 target = &p[i];2519 if (target)2520 break;2521 pthread_cond_wait(&progress_cond, &progress_mutex);2522 }25232524 for (i = 0; i < delta_search_threads; i++)2525 if (p[i].remaining > 2*window &&2526 (!victim || victim->remaining < p[i].remaining))2527 victim = &p[i];2528 if (victim) {2529 sub_size = victim->remaining / 2;2530 list = victim->list + victim->list_size - sub_size;2531 while (sub_size && list[0]->hash &&2532 list[0]->hash == list[-1]->hash) {2533 list++;2534 sub_size--;2535 }2536 if (!sub_size) {2537 /*2538 * It is possible for some "paths" to have2539 * so many objects that no hash boundary2540 * might be found. Let's just steal the2541 * exact half in that case.2542 */2543 sub_size = victim->remaining / 2;2544 list -= sub_size;2545 }2546 target->list = list;2547 victim->list_size -= sub_size;2548 victim->remaining -= sub_size;2549 }2550 target->list_size = sub_size;2551 target->remaining = sub_size;2552 target->working = 1;2553 progress_unlock();25542555 pthread_mutex_lock(&target->mutex);2556 target->data_ready = 1;2557 pthread_cond_signal(&target->cond);2558 pthread_mutex_unlock(&target->mutex);25592560 if (!sub_size) {2561 pthread_join(target->thread, NULL);2562 pthread_cond_destroy(&target->cond);2563 pthread_mutex_destroy(&target->mutex);2564 active_threads--;2565 }2566 }2567 cleanup_threaded_search();2568 free(p);2569}25702571static void add_tag_chain(const struct object_id *oid)2572{2573 struct tag *tag;25742575 /*2576 * We catch duplicates already in add_object_entry(), but we'd2577 * prefer to do this extra check to avoid having to parse the2578 * tag at all if we already know that it's being packed (e.g., if2579 * it was included via bitmaps, we would not have parsed it2580 * previously).2581 */2582 if (packlist_find(&to_pack, oid, NULL))2583 return;25842585 tag = lookup_tag(the_repository, oid);2586 while (1) {2587 if (!tag || parse_tag(tag) || !tag->tagged)2588 die(_("unable to pack objects reachable from tag %s"),2589 oid_to_hex(oid));25902591 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);25922593 if (tag->tagged->type != OBJ_TAG)2594 return;25952596 tag = (struct tag *)tag->tagged;2597 }2598}25992600static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)2601{2602 struct object_id peeled;26032604 if (starts_with(path, "refs/tags/") && /* is a tag? */2605 !peel_ref(path, &peeled) && /* peelable? */2606 packlist_find(&to_pack, &peeled, NULL)) /* object packed? */2607 add_tag_chain(oid);2608 return 0;2609}26102611static void prepare_pack(int window, int depth)2612{2613 struct object_entry **delta_list;2614 uint32_t i, nr_deltas;2615 unsigned n;26162617 if (use_delta_islands)2618 resolve_tree_islands(the_repository, progress, &to_pack);26192620 get_object_details();26212622 /*2623 * If we're locally repacking then we need to be doubly careful2624 * from now on in order to make sure no stealth corruption gets2625 * propagated to the new pack. Clients receiving streamed packs2626 * should validate everything they get anyway so no need to incur2627 * the additional cost here in that case.2628 */2629 if (!pack_to_stdout)2630 do_check_packed_object_crc = 1;26312632 if (!to_pack.nr_objects || !window || !depth)2633 return;26342635 ALLOC_ARRAY(delta_list, to_pack.nr_objects);2636 nr_deltas = n = 0;26372638 for (i = 0; i < to_pack.nr_objects; i++) {2639 struct object_entry *entry = to_pack.objects + i;26402641 if (DELTA(entry))2642 /* This happens if we decided to reuse existing2643 * delta from a pack. "reuse_delta &&" is implied.2644 */2645 continue;26462647 if (!entry->type_valid ||2648 oe_size_less_than(&to_pack, entry, 50))2649 continue;26502651 if (entry->no_try_delta)2652 continue;26532654 if (!entry->preferred_base) {2655 nr_deltas++;2656 if (oe_type(entry) < 0)2657 die(_("unable to get type of object %s"),2658 oid_to_hex(&entry->idx.oid));2659 } else {2660 if (oe_type(entry) < 0) {2661 /*2662 * This object is not found, but we2663 * don't have to include it anyway.2664 */2665 continue;2666 }2667 }26682669 delta_list[n++] = entry;2670 }26712672 if (nr_deltas && n > 1) {2673 unsigned nr_done = 0;2674 if (progress)2675 progress_state = start_progress(_("Compressing objects"),2676 nr_deltas);2677 QSORT(delta_list, n, type_size_sort);2678 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2679 stop_progress(&progress_state);2680 if (nr_done != nr_deltas)2681 die(_("inconsistency with delta count"));2682 }2683 free(delta_list);2684}26852686static int git_pack_config(const char *k, const char *v, void *cb)2687{2688 if (!strcmp(k, "pack.window")) {2689 window = git_config_int(k, v);2690 return 0;2691 }2692 if (!strcmp(k, "pack.windowmemory")) {2693 window_memory_limit = git_config_ulong(k, v);2694 return 0;2695 }2696 if (!strcmp(k, "pack.depth")) {2697 depth = git_config_int(k, v);2698 return 0;2699 }2700 if (!strcmp(k, "pack.deltacachesize")) {2701 max_delta_cache_size = git_config_int(k, v);2702 return 0;2703 }2704 if (!strcmp(k, "pack.deltacachelimit")) {2705 cache_max_small_delta_size = git_config_int(k, v);2706 return 0;2707 }2708 if (!strcmp(k, "pack.writebitmaphashcache")) {2709 if (git_config_bool(k, v))2710 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2711 else2712 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2713 }2714 if (!strcmp(k, "pack.usebitmaps")) {2715 use_bitmap_index_default = git_config_bool(k, v);2716 return 0;2717 }2718 if (!strcmp(k, "pack.usesparse")) {2719 sparse = git_config_bool(k, v);2720 return 0;2721 }2722 if (!strcmp(k, "pack.threads")) {2723 delta_search_threads = git_config_int(k, v);2724 if (delta_search_threads < 0)2725 die(_("invalid number of threads specified (%d)"),2726 delta_search_threads);2727 if (!HAVE_THREADS && delta_search_threads != 1) {2728 warning(_("no threads support, ignoring %s"), k);2729 delta_search_threads = 0;2730 }2731 return 0;2732 }2733 if (!strcmp(k, "pack.indexversion")) {2734 pack_idx_opts.version = git_config_int(k, v);2735 if (pack_idx_opts.version > 2)2736 die(_("bad pack.indexversion=%"PRIu32),2737 pack_idx_opts.version);2738 return 0;2739 }2740 return git_default_config(k, v, cb);2741}27422743static void read_object_list_from_stdin(void)2744{2745 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];2746 struct object_id oid;2747 const char *p;27482749 for (;;) {2750 if (!fgets(line, sizeof(line), stdin)) {2751 if (feof(stdin))2752 break;2753 if (!ferror(stdin))2754 die("BUG: fgets returned NULL, not EOF, not error!");2755 if (errno != EINTR)2756 die_errno("fgets");2757 clearerr(stdin);2758 continue;2759 }2760 if (line[0] == '-') {2761 if (get_oid_hex(line+1, &oid))2762 die(_("expected edge object ID, got garbage:\n %s"),2763 line);2764 add_preferred_base(&oid);2765 continue;2766 }2767 if (parse_oid_hex(line, &oid, &p))2768 die(_("expected object ID, got garbage:\n %s"), line);27692770 add_preferred_base_object(p + 1);2771 add_object_entry(&oid, OBJ_NONE, p + 1, 0);2772 }2773}27742775/* Remember to update object flag allocation in object.h */2776#define OBJECT_ADDED (1u<<20)27772778static void show_commit(struct commit *commit, void *data)2779{2780 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);2781 commit->object.flags |= OBJECT_ADDED;27822783 if (write_bitmap_index)2784 index_commit_for_bitmap(commit);27852786 if (use_delta_islands)2787 propagate_island_marks(commit);2788}27892790static void show_object(struct object *obj, const char *name, void *data)2791{2792 add_preferred_base_object(name);2793 add_object_entry(&obj->oid, obj->type, name, 0);2794 obj->flags |= OBJECT_ADDED;27952796 if (use_delta_islands) {2797 const char *p;2798 unsigned depth;2799 struct object_entry *ent;28002801 /* the empty string is a root tree, which is depth 0 */2802 depth = *name ? 1 : 0;2803 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))2804 depth++;28052806 ent = packlist_find(&to_pack, &obj->oid, NULL);2807 if (ent && depth > oe_tree_depth(&to_pack, ent))2808 oe_set_tree_depth(&to_pack, ent, depth);2809 }2810}28112812static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)2813{2814 assert(arg_missing_action == MA_ALLOW_ANY);28152816 /*2817 * Quietly ignore ALL missing objects. This avoids problems with2818 * staging them now and getting an odd error later.2819 */2820 if (!has_object_file(&obj->oid))2821 return;28222823 show_object(obj, name, data);2824}28252826static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)2827{2828 assert(arg_missing_action == MA_ALLOW_PROMISOR);28292830 /*2831 * Quietly ignore EXPECTED missing objects. This avoids problems with2832 * staging them now and getting an odd error later.2833 */2834 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))2835 return;28362837 show_object(obj, name, data);2838}28392840static int option_parse_missing_action(const struct option *opt,2841 const char *arg, int unset)2842{2843 assert(arg);2844 assert(!unset);28452846 if (!strcmp(arg, "error")) {2847 arg_missing_action = MA_ERROR;2848 fn_show_object = show_object;2849 return 0;2850 }28512852 if (!strcmp(arg, "allow-any")) {2853 arg_missing_action = MA_ALLOW_ANY;2854 fetch_if_missing = 0;2855 fn_show_object = show_object__ma_allow_any;2856 return 0;2857 }28582859 if (!strcmp(arg, "allow-promisor")) {2860 arg_missing_action = MA_ALLOW_PROMISOR;2861 fetch_if_missing = 0;2862 fn_show_object = show_object__ma_allow_promisor;2863 return 0;2864 }28652866 die(_("invalid value for --missing"));2867 return 0;2868}28692870static void show_edge(struct commit *commit)2871{2872 add_preferred_base(&commit->object.oid);2873}28742875struct in_pack_object {2876 off_t offset;2877 struct object *object;2878};28792880struct in_pack {2881 unsigned int alloc;2882 unsigned int nr;2883 struct in_pack_object *array;2884};28852886static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)2887{2888 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);2889 in_pack->array[in_pack->nr].object = object;2890 in_pack->nr++;2891}28922893/*2894 * Compare the objects in the offset order, in order to emulate the2895 * "git rev-list --objects" output that produced the pack originally.2896 */2897static int ofscmp(const void *a_, const void *b_)2898{2899 struct in_pack_object *a = (struct in_pack_object *)a_;2900 struct in_pack_object *b = (struct in_pack_object *)b_;29012902 if (a->offset < b->offset)2903 return -1;2904 else if (a->offset > b->offset)2905 return 1;2906 else2907 return oidcmp(&a->object->oid, &b->object->oid);2908}29092910static void add_objects_in_unpacked_packs(void)2911{2912 struct packed_git *p;2913 struct in_pack in_pack;2914 uint32_t i;29152916 memset(&in_pack, 0, sizeof(in_pack));29172918 for (p = get_all_packs(the_repository); p; p = p->next) {2919 struct object_id oid;2920 struct object *o;29212922 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)2923 continue;2924 if (open_pack_index(p))2925 die(_("cannot open pack index"));29262927 ALLOC_GROW(in_pack.array,2928 in_pack.nr + p->num_objects,2929 in_pack.alloc);29302931 for (i = 0; i < p->num_objects; i++) {2932 nth_packed_object_oid(&oid, p, i);2933 o = lookup_unknown_object(&oid);2934 if (!(o->flags & OBJECT_ADDED))2935 mark_in_pack_object(o, p, &in_pack);2936 o->flags |= OBJECT_ADDED;2937 }2938 }29392940 if (in_pack.nr) {2941 QSORT(in_pack.array, in_pack.nr, ofscmp);2942 for (i = 0; i < in_pack.nr; i++) {2943 struct object *o = in_pack.array[i].object;2944 add_object_entry(&o->oid, o->type, "", 0);2945 }2946 }2947 free(in_pack.array);2948}29492950static int add_loose_object(const struct object_id *oid, const char *path,2951 void *data)2952{2953 enum object_type type = oid_object_info(the_repository, oid, NULL);29542955 if (type < 0) {2956 warning(_("loose object at %s could not be examined"), path);2957 return 0;2958 }29592960 add_object_entry(oid, type, "", 0);2961 return 0;2962}29632964/*2965 * We actually don't even have to worry about reachability here.2966 * add_object_entry will weed out duplicates, so we just add every2967 * loose object we find.2968 */2969static void add_unreachable_loose_objects(void)2970{2971 for_each_loose_file_in_objdir(get_object_directory(),2972 add_loose_object,2973 NULL, NULL, NULL);2974}29752976static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2977{2978 static struct packed_git *last_found = (void *)1;2979 struct packed_git *p;29802981 p = (last_found != (void *)1) ? last_found :2982 get_all_packs(the_repository);29832984 while (p) {2985 if ((!p->pack_local || p->pack_keep ||2986 p->pack_keep_in_core) &&2987 find_pack_entry_one(oid->hash, p)) {2988 last_found = p;2989 return 1;2990 }2991 if (p == last_found)2992 p = get_all_packs(the_repository);2993 else2994 p = p->next;2995 if (p == last_found)2996 p = p->next;2997 }2998 return 0;2999}30003001/*3002 * Store a list of sha1s that are should not be discarded3003 * because they are either written too recently, or are3004 * reachable from another object that was.3005 *3006 * This is filled by get_object_list.3007 */3008static struct oid_array recent_objects;30093010static int loosened_object_can_be_discarded(const struct object_id *oid,3011 timestamp_t mtime)3012{3013 if (!unpack_unreachable_expiration)3014 return 0;3015 if (mtime > unpack_unreachable_expiration)3016 return 0;3017 if (oid_array_lookup(&recent_objects, oid) >= 0)3018 return 0;3019 return 1;3020}30213022static void loosen_unused_packed_objects(void)3023{3024 struct packed_git *p;3025 uint32_t i;3026 struct object_id oid;30273028 for (p = get_all_packs(the_repository); p; p = p->next) {3029 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)3030 continue;30313032 if (open_pack_index(p))3033 die(_("cannot open pack index"));30343035 for (i = 0; i < p->num_objects; i++) {3036 nth_packed_object_oid(&oid, p, i);3037 if (!packlist_find(&to_pack, &oid, NULL) &&3038 !has_sha1_pack_kept_or_nonlocal(&oid) &&3039 !loosened_object_can_be_discarded(&oid, p->mtime))3040 if (force_object_loose(&oid, p->mtime))3041 die(_("unable to force loose object"));3042 }3043 }3044}30453046/*3047 * This tracks any options which pack-reuse code expects to be on, or which a3048 * reader of the pack might not understand, and which would therefore prevent3049 * blind reuse of what we have on disk.3050 */3051static int pack_options_allow_reuse(void)3052{3053 return pack_to_stdout &&3054 allow_ofs_delta &&3055 !ignore_packed_keep_on_disk &&3056 !ignore_packed_keep_in_core &&3057 (!local || !have_non_local_packs) &&3058 !incremental;3059}30603061static int get_object_list_from_bitmap(struct rev_info *revs)3062{3063 if (!(bitmap_git = prepare_bitmap_walk(revs)))3064 return -1;30653066 if (pack_options_allow_reuse() &&3067 !reuse_partial_packfile_from_bitmap(3068 bitmap_git,3069 &reuse_packfile,3070 &reuse_packfile_objects,3071 &reuse_packfile_offset)) {3072 assert(reuse_packfile_objects);3073 nr_result += reuse_packfile_objects;3074 display_progress(progress_state, nr_result);3075 }30763077 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3078 return 0;3079}30803081static void record_recent_object(struct object *obj,3082 const char *name,3083 void *data)3084{3085 oid_array_append(&recent_objects, &obj->oid);3086}30873088static void record_recent_commit(struct commit *commit, void *data)3089{3090 oid_array_append(&recent_objects, &commit->object.oid);3091}30923093static void get_object_list(int ac, const char **av)3094{3095 struct rev_info revs;3096 struct setup_revision_opt s_r_opt = {3097 .allow_exclude_promisor_objects = 1,3098 };3099 char line[1000];3100 int flags = 0;3101 int save_warning;31023103 repo_init_revisions(the_repository, &revs, NULL);3104 save_commit_buffer = 0;3105 setup_revisions(ac, av, &revs, &s_r_opt);31063107 /* make sure shallows are read */3108 is_repository_shallow(the_repository);31093110 save_warning = warn_on_object_refname_ambiguity;3111 warn_on_object_refname_ambiguity = 0;31123113 while (fgets(line, sizeof(line), stdin) != NULL) {3114 int len = strlen(line);3115 if (len && line[len - 1] == '\n')3116 line[--len] = 0;3117 if (!len)3118 break;3119 if (*line == '-') {3120 if (!strcmp(line, "--not")) {3121 flags ^= UNINTERESTING;3122 write_bitmap_index = 0;3123 continue;3124 }3125 if (starts_with(line, "--shallow ")) {3126 struct object_id oid;3127 if (get_oid_hex(line + 10, &oid))3128 die("not an SHA-1 '%s'", line + 10);3129 register_shallow(the_repository, &oid);3130 use_bitmap_index = 0;3131 continue;3132 }3133 die(_("not a rev '%s'"), line);3134 }3135 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3136 die(_("bad revision '%s'"), line);3137 }31383139 warn_on_object_refname_ambiguity = save_warning;31403141 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))3142 return;31433144 if (use_delta_islands)3145 load_delta_islands(the_repository, progress);31463147 if (prepare_revision_walk(&revs))3148 die(_("revision walk setup failed"));3149 mark_edges_uninteresting(&revs, show_edge, sparse);31503151 if (!fn_show_object)3152 fn_show_object = show_object;3153 traverse_commit_list_filtered(&filter_options, &revs,3154 show_commit, fn_show_object, NULL,3155 NULL);31563157 if (unpack_unreachable_expiration) {3158 revs.ignore_missing_links = 1;3159 if (add_unseen_recent_objects_to_traversal(&revs,3160 unpack_unreachable_expiration))3161 die(_("unable to add recent objects"));3162 if (prepare_revision_walk(&revs))3163 die(_("revision walk setup failed"));3164 traverse_commit_list(&revs, record_recent_commit,3165 record_recent_object, NULL);3166 }31673168 if (keep_unreachable)3169 add_objects_in_unpacked_packs();3170 if (pack_loose_unreachable)3171 add_unreachable_loose_objects();3172 if (unpack_unreachable)3173 loosen_unused_packed_objects();31743175 oid_array_clear(&recent_objects);3176}31773178static void add_extra_kept_packs(const struct string_list *names)3179{3180 struct packed_git *p;31813182 if (!names->nr)3183 return;31843185 for (p = get_all_packs(the_repository); p; p = p->next) {3186 const char *name = basename(p->pack_name);3187 int i;31883189 if (!p->pack_local)3190 continue;31913192 for (i = 0; i < names->nr; i++)3193 if (!fspathcmp(name, names->items[i].string))3194 break;31953196 if (i < names->nr) {3197 p->pack_keep_in_core = 1;3198 ignore_packed_keep_in_core = 1;3199 continue;3200 }3201 }3202}32033204static int option_parse_index_version(const struct option *opt,3205 const char *arg, int unset)3206{3207 char *c;3208 const char *val = arg;32093210 BUG_ON_OPT_NEG(unset);32113212 pack_idx_opts.version = strtoul(val, &c, 10);3213 if (pack_idx_opts.version > 2)3214 die(_("unsupported index version %s"), val);3215 if (*c == ',' && c[1])3216 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);3217 if (*c || pack_idx_opts.off32_limit & 0x80000000)3218 die(_("bad index version '%s'"), val);3219 return 0;3220}32213222static int option_parse_unpack_unreachable(const struct option *opt,3223 const char *arg, int unset)3224{3225 if (unset) {3226 unpack_unreachable = 0;3227 unpack_unreachable_expiration = 0;3228 }3229 else {3230 unpack_unreachable = 1;3231 if (arg)3232 unpack_unreachable_expiration = approxidate(arg);3233 }3234 return 0;3235}32363237int cmd_pack_objects(int argc, const char **argv, const char *prefix)3238{3239 int use_internal_rev_list = 0;3240 int shallow = 0;3241 int all_progress_implied = 0;3242 struct argv_array rp = ARGV_ARRAY_INIT;3243 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;3244 int rev_list_index = 0;3245 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3246 struct option pack_objects_options[] = {3247 OPT_SET_INT('q', "quiet", &progress,3248 N_("do not show progress meter"), 0),3249 OPT_SET_INT(0, "progress", &progress,3250 N_("show progress meter"), 1),3251 OPT_SET_INT(0, "all-progress", &progress,3252 N_("show progress meter during object writing phase"), 2),3253 OPT_BOOL(0, "all-progress-implied",3254 &all_progress_implied,3255 N_("similar to --all-progress when progress meter is shown")),3256 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),3257 N_("write the pack index file in the specified idx format version"),3258 PARSE_OPT_NONEG, option_parse_index_version },3259 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,3260 N_("maximum size of each output pack file")),3261 OPT_BOOL(0, "local", &local,3262 N_("ignore borrowed objects from alternate object store")),3263 OPT_BOOL(0, "incremental", &incremental,3264 N_("ignore packed objects")),3265 OPT_INTEGER(0, "window", &window,3266 N_("limit pack window by objects")),3267 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,3268 N_("limit pack window by memory in addition to object limit")),3269 OPT_INTEGER(0, "depth", &depth,3270 N_("maximum length of delta chain allowed in the resulting pack")),3271 OPT_BOOL(0, "reuse-delta", &reuse_delta,3272 N_("reuse existing deltas")),3273 OPT_BOOL(0, "reuse-object", &reuse_object,3274 N_("reuse existing objects")),3275 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,3276 N_("use OFS_DELTA objects")),3277 OPT_INTEGER(0, "threads", &delta_search_threads,3278 N_("use threads when searching for best delta matches")),3279 OPT_BOOL(0, "non-empty", &non_empty,3280 N_("do not create an empty pack output")),3281 OPT_BOOL(0, "revs", &use_internal_rev_list,3282 N_("read revision arguments from standard input")),3283 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,3284 N_("limit the objects to those that are not yet packed"),3285 1, PARSE_OPT_NONEG),3286 OPT_SET_INT_F(0, "all", &rev_list_all,3287 N_("include objects reachable from any reference"),3288 1, PARSE_OPT_NONEG),3289 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,3290 N_("include objects referred by reflog entries"),3291 1, PARSE_OPT_NONEG),3292 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,3293 N_("include objects referred to by the index"),3294 1, PARSE_OPT_NONEG),3295 OPT_BOOL(0, "stdout", &pack_to_stdout,3296 N_("output pack to stdout")),3297 OPT_BOOL(0, "include-tag", &include_tag,3298 N_("include tag objects that refer to objects to be packed")),3299 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,3300 N_("keep unreachable objects")),3301 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,3302 N_("pack loose unreachable objects")),3303 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),3304 N_("unpack unreachable objects newer than <time>"),3305 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3306 OPT_BOOL(0, "sparse", &sparse,3307 N_("use the sparse reachability algorithm")),3308 OPT_BOOL(0, "thin", &thin,3309 N_("create thin packs")),3310 OPT_BOOL(0, "shallow", &shallow,3311 N_("create packs suitable for shallow fetches")),3312 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,3313 N_("ignore packs that have companion .keep file")),3314 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),3315 N_("ignore this pack")),3316 OPT_INTEGER(0, "compression", &pack_compression_level,3317 N_("pack compression level")),3318 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,3319 N_("do not hide commits by grafts"), 0),3320 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,3321 N_("use a bitmap index if available to speed up counting objects")),3322 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index,3323 N_("write a bitmap index together with the pack index"),3324 WRITE_BITMAP_TRUE),3325 OPT_SET_INT_F(0, "write-bitmap-index-quiet",3326 &write_bitmap_index,3327 N_("write a bitmap index if possible"),3328 WRITE_BITMAP_QUIET, PARSE_OPT_HIDDEN),3329 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3330 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),3331 N_("handling for missing objects"), PARSE_OPT_NONEG,3332 option_parse_missing_action },3333 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,3334 N_("do not pack objects in promisor packfiles")),3335 OPT_BOOL(0, "delta-islands", &use_delta_islands,3336 N_("respect islands during delta compression")),3337 OPT_END(),3338 };33393340 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))3341 BUG("too many dfs states, increase OE_DFS_STATE_BITS");33423343 read_replace_refs = 0;33443345 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", 0);3346 reset_pack_idx_option(&pack_idx_opts);3347 git_config(git_pack_config, NULL);33483349 progress = isatty(2);3350 argc = parse_options(argc, argv, prefix, pack_objects_options,3351 pack_usage, 0);33523353 if (argc) {3354 base_name = argv[0];3355 argc--;3356 }3357 if (pack_to_stdout != !base_name || argc)3358 usage_with_options(pack_usage, pack_objects_options);33593360 if (depth >= (1 << OE_DEPTH_BITS)) {3361 warning(_("delta chain depth %d is too deep, forcing %d"),3362 depth, (1 << OE_DEPTH_BITS) - 1);3363 depth = (1 << OE_DEPTH_BITS) - 1;3364 }3365 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {3366 warning(_("pack.deltaCacheLimit is too high, forcing %d"),3367 (1U << OE_Z_DELTA_BITS) - 1);3368 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;3369 }33703371 argv_array_push(&rp, "pack-objects");3372 if (thin) {3373 use_internal_rev_list = 1;3374 argv_array_push(&rp, shallow3375 ? "--objects-edge-aggressive"3376 : "--objects-edge");3377 } else3378 argv_array_push(&rp, "--objects");33793380 if (rev_list_all) {3381 use_internal_rev_list = 1;3382 argv_array_push(&rp, "--all");3383 }3384 if (rev_list_reflog) {3385 use_internal_rev_list = 1;3386 argv_array_push(&rp, "--reflog");3387 }3388 if (rev_list_index) {3389 use_internal_rev_list = 1;3390 argv_array_push(&rp, "--indexed-objects");3391 }3392 if (rev_list_unpacked) {3393 use_internal_rev_list = 1;3394 argv_array_push(&rp, "--unpacked");3395 }33963397 if (exclude_promisor_objects) {3398 use_internal_rev_list = 1;3399 fetch_if_missing = 0;3400 argv_array_push(&rp, "--exclude-promisor-objects");3401 }3402 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)3403 use_internal_rev_list = 1;34043405 if (!reuse_object)3406 reuse_delta = 0;3407 if (pack_compression_level == -1)3408 pack_compression_level = Z_DEFAULT_COMPRESSION;3409 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)3410 die(_("bad pack compression level %d"), pack_compression_level);34113412 if (!delta_search_threads) /* --threads=0 means autodetect */3413 delta_search_threads = online_cpus();34143415 if (!HAVE_THREADS && delta_search_threads != 1)3416 warning(_("no threads support, ignoring --threads"));3417 if (!pack_to_stdout && !pack_size_limit)3418 pack_size_limit = pack_size_limit_cfg;3419 if (pack_to_stdout && pack_size_limit)3420 die(_("--max-pack-size cannot be used to build a pack for transfer"));3421 if (pack_size_limit && pack_size_limit < 1024*1024) {3422 warning(_("minimum pack size limit is 1 MiB"));3423 pack_size_limit = 1024*1024;3424 }34253426 if (!pack_to_stdout && thin)3427 die(_("--thin cannot be used to build an indexable pack"));34283429 if (keep_unreachable && unpack_unreachable)3430 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3431 if (!rev_list_all || !rev_list_reflog || !rev_list_index)3432 unpack_unreachable_expiration = 0;34333434 if (filter_options.choice) {3435 if (!pack_to_stdout)3436 die(_("cannot use --filter without --stdout"));3437 use_bitmap_index = 0;3438 }34393440 /*3441 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3442 *3443 * - to produce good pack (with bitmap index not-yet-packed objects are3444 * packed in suboptimal order).3445 *3446 * - to use more robust pack-generation codepath (avoiding possible3447 * bugs in bitmap code and possible bitmap index corruption).3448 */3449 if (!pack_to_stdout)3450 use_bitmap_index_default = 0;34513452 if (use_bitmap_index < 0)3453 use_bitmap_index = use_bitmap_index_default;34543455 /* "hard" reasons not to use bitmaps; these just won't work at all */3456 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))3457 use_bitmap_index = 0;34583459 if (pack_to_stdout || !rev_list_all)3460 write_bitmap_index = 0;34613462 if (use_delta_islands)3463 argv_array_push(&rp, "--topo-order");34643465 if (progress && all_progress_implied)3466 progress = 2;34673468 add_extra_kept_packs(&keep_pack_list);3469 if (ignore_packed_keep_on_disk) {3470 struct packed_git *p;3471 for (p = get_all_packs(the_repository); p; p = p->next)3472 if (p->pack_local && p->pack_keep)3473 break;3474 if (!p) /* no keep-able packs found */3475 ignore_packed_keep_on_disk = 0;3476 }3477 if (local) {3478 /*3479 * unlike ignore_packed_keep_on_disk above, we do not3480 * want to unset "local" based on looking at packs, as3481 * it also covers non-local objects3482 */3483 struct packed_git *p;3484 for (p = get_all_packs(the_repository); p; p = p->next) {3485 if (!p->pack_local) {3486 have_non_local_packs = 1;3487 break;3488 }3489 }3490 }34913492 trace2_region_enter("pack-objects", "enumerate-objects",3493 the_repository);3494 prepare_packing_data(the_repository, &to_pack);34953496 if (progress)3497 progress_state = start_progress(_("Enumerating objects"), 0);3498 if (!use_internal_rev_list)3499 read_object_list_from_stdin();3500 else {3501 get_object_list(rp.argc, rp.argv);3502 argv_array_clear(&rp);3503 }3504 cleanup_preferred_base();3505 if (include_tag && nr_result)3506 for_each_ref(add_ref_tag, NULL);3507 stop_progress(&progress_state);3508 trace2_region_leave("pack-objects", "enumerate-objects",3509 the_repository);35103511 if (non_empty && !nr_result)3512 return 0;3513 if (nr_result) {3514 trace2_region_enter("pack-objects", "prepare-pack",3515 the_repository);3516 prepare_pack(window, depth);3517 trace2_region_leave("pack-objects", "prepare-pack",3518 the_repository);3519 }35203521 trace2_region_enter("pack-objects", "write-pack-file", the_repository);3522 write_pack_file();3523 trace2_region_leave("pack-objects", "write-pack-file", the_repository);35243525 if (progress)3526 fprintf_ln(stderr,3527 _("Total %"PRIu32" (delta %"PRIu32"),"3528 " reused %"PRIu32" (delta %"PRIu32")"),3529 written, written_delta, reused, reused_delta);3530 return 0;3531}