| 1 | #define USE_THE_REPOSITORY_VARIABLE |
| 2 | #define DISABLE_SIGN_COMPARE_WARNINGS |
| 3 | |
| 4 | #include "git-compat-util.h" |
| 5 | #include "gettext.h" |
| 6 | #include "hex.h" |
| 7 | #include "lockfile.h" |
| 8 | #include "tree.h" |
| 9 | #include "tree-walk.h" |
| 10 | #include "cache-tree.h" |
| 11 | #include "object-file.h" |
| 12 | #include "odb.h" |
| 13 | #include "odb/transaction.h" |
| 14 | #include "read-cache-ll.h" |
| 15 | #include "replace-object.h" |
| 16 | #include "repository.h" |
| 17 | #include "promisor-remote.h" |
| 18 | #include "trace.h" |
| 19 | #include "trace2.h" |
| 20 | |
| 21 | #ifndef DEBUG_CACHE_TREE |
| 22 | #define DEBUG_CACHE_TREE 0 |
| 23 | #endif |
| 24 | |
| 25 | struct cache_tree *cache_tree(void) |
| 26 | { |
| 27 | struct cache_tree *it = xcalloc(1, sizeof(struct cache_tree)); |
| 28 | it->entry_count = -1; |
| 29 | return it; |
| 30 | } |
| 31 | |
| 32 | void cache_tree_free(struct cache_tree **it_p) |
| 33 | { |
| 34 | int i; |
| 35 | struct cache_tree *it = *it_p; |
| 36 | |
| 37 | if (!it) |
| 38 | return; |
| 39 | for (i = 0; i < it->subtree_nr; i++) |
| 40 | if (it->down[i]) { |
| 41 | cache_tree_free(&it->down[i]->cache_tree); |
| 42 | free(it->down[i]); |
| 43 | } |
| 44 | free(it->down); |
| 45 | free(it); |
| 46 | *it_p = NULL; |
| 47 | } |
| 48 | |
| 49 | static int subtree_name_cmp(const char *one, int onelen, |
| 50 | const char *two, int twolen) |
| 51 | { |
| 52 | if (onelen < twolen) |
| 53 | return -1; |
| 54 | if (twolen < onelen) |
| 55 | return 1; |
| 56 | return memcmp(one, two, onelen); |
| 57 | } |
| 58 | |
| 59 | int cache_tree_subtree_pos(struct cache_tree *it, const char *path, int pathlen) |
| 60 | { |
| 61 | struct cache_tree_sub **down = it->down; |
| 62 | int lo, hi; |
| 63 | lo = 0; |
| 64 | hi = it->subtree_nr; |
| 65 | while (lo < hi) { |
| 66 | int mi = lo + (hi - lo) / 2; |
| 67 | struct cache_tree_sub *mdl = down[mi]; |
| 68 | int cmp = subtree_name_cmp(path, pathlen, |
| 69 | mdl->name, mdl->namelen); |
| 70 | if (!cmp) |
| 71 | return mi; |
| 72 | if (cmp < 0) |
| 73 | hi = mi; |
| 74 | else |
| 75 | lo = mi + 1; |
| 76 | } |
| 77 | return -lo-1; |
| 78 | } |
| 79 | |
| 80 | static struct cache_tree_sub *find_subtree(struct cache_tree *it, |
| 81 | const char *path, |
| 82 | int pathlen, |
| 83 | int create) |
| 84 | { |
| 85 | struct cache_tree_sub *down; |
| 86 | int pos = cache_tree_subtree_pos(it, path, pathlen); |
| 87 | if (0 <= pos) |
| 88 | return it->down[pos]; |
| 89 | if (!create) |
| 90 | return NULL; |
| 91 | |
| 92 | pos = -pos-1; |
| 93 | ALLOC_GROW(it->down, it->subtree_nr + 1, it->subtree_alloc); |
| 94 | it->subtree_nr++; |
| 95 | |
| 96 | FLEX_ALLOC_MEM(down, name, path, pathlen); |
| 97 | down->cache_tree = NULL; |
| 98 | down->namelen = pathlen; |
| 99 | |
| 100 | if (pos < it->subtree_nr) |
| 101 | MOVE_ARRAY(it->down + pos + 1, it->down + pos, |
| 102 | it->subtree_nr - pos - 1); |
| 103 | it->down[pos] = down; |
| 104 | return down; |
| 105 | } |
| 106 | |
| 107 | struct cache_tree_sub *cache_tree_sub(struct cache_tree *it, const char *path) |
| 108 | { |
| 109 | int pathlen = strlen(path); |
| 110 | return find_subtree(it, path, pathlen, 1); |
| 111 | } |
| 112 | |
| 113 | static int do_invalidate_path(struct cache_tree *it, const char *path) |
| 114 | { |
| 115 | /* a/b/c |
| 116 | * ==> invalidate self |
| 117 | * ==> find "a", have it invalidate "b/c" |
| 118 | * a |
| 119 | * ==> invalidate self |
| 120 | * ==> if "a" exists as a subtree, remove it. |
| 121 | */ |
| 122 | const char *slash; |
| 123 | int namelen; |
| 124 | struct cache_tree_sub *down; |
| 125 | |
| 126 | #if DEBUG_CACHE_TREE |
| 127 | fprintf(stderr, "cache-tree invalidate <%s>\n", path); |
| 128 | #endif |
| 129 | |
| 130 | if (!it) |
| 131 | return 0; |
| 132 | slash = strchrnul(path, '/'); |
| 133 | namelen = slash - path; |
| 134 | it->entry_count = -1; |
| 135 | if (!*slash) { |
| 136 | int pos; |
| 137 | pos = cache_tree_subtree_pos(it, path, namelen); |
| 138 | if (0 <= pos) { |
| 139 | cache_tree_free(&it->down[pos]->cache_tree); |
| 140 | free(it->down[pos]); |
| 141 | /* 0 1 2 3 4 5 |
| 142 | * ^ ^subtree_nr = 6 |
| 143 | * pos |
| 144 | * move 4 and 5 up one place (2 entries) |
| 145 | * 2 = 6 - 3 - 1 = subtree_nr - pos - 1 |
| 146 | */ |
| 147 | MOVE_ARRAY(it->down + pos, it->down + pos + 1, |
| 148 | it->subtree_nr - pos - 1); |
| 149 | it->subtree_nr--; |
| 150 | } |
| 151 | return 1; |
| 152 | } |
| 153 | down = find_subtree(it, path, namelen, 0); |
| 154 | if (down) |
| 155 | do_invalidate_path(down->cache_tree, slash + 1); |
| 156 | return 1; |
| 157 | } |
| 158 | |
| 159 | void cache_tree_invalidate_path(struct index_state *istate, const char *path) |
| 160 | { |
| 161 | if (do_invalidate_path(istate->cache_tree, path)) |
| 162 | istate->cache_changed |= CACHE_TREE_CHANGED; |
| 163 | } |
| 164 | |
| 165 | /* |
| 166 | * Check whether this_ce and the next entry in the index form a D/F |
| 167 | * conflict ("path" vs "path/file"). Returns the conflicting "path/..." |
| 168 | * name when one is found, or NULL otherwise. |
| 169 | * |
| 170 | * The cache is sorted, so "path/file" sorts after "path" and the |
| 171 | * conflict is usually visible as adjacent entries. But other entries |
| 172 | * can sort between them -- e.g. "path-internal" sits between "path" |
| 173 | * and "path/file" because '-' (0x2D) precedes '/' (0x2F) -- so when |
| 174 | * the immediately following entry shares our prefix but starts with a |
| 175 | * character that sorts before '/', binary search for "path/" instead. |
| 176 | */ |
| 177 | static const char *find_df_conflict(struct index_state *istate, |
| 178 | const struct cache_entry *this_ce, |
| 179 | const struct cache_entry *next_ce) |
| 180 | { |
| 181 | const char *this_name = this_ce->name; |
| 182 | const char *next_name = next_ce->name; |
| 183 | int this_len = ce_namelen(this_ce); |
| 184 | const struct cache_entry *other; |
| 185 | struct strbuf probe = STRBUF_INIT; |
| 186 | int pos; |
| 187 | |
| 188 | if (this_len >= ce_namelen(next_ce) || |
| 189 | next_name[this_len] > '/' || |
| 190 | strncmp(this_name, next_name, this_len)) |
| 191 | return NULL; |
| 192 | |
| 193 | if (next_name[this_len] == '/') |
| 194 | return next_name; |
| 195 | |
| 196 | strbuf_add(&probe, this_name, this_len); |
| 197 | strbuf_addch(&probe, '/'); |
| 198 | pos = index_name_pos_sparse(istate, probe.buf, probe.len); |
| 199 | strbuf_release(&probe); |
| 200 | |
| 201 | if (pos < 0) |
| 202 | pos = -pos - 1; |
| 203 | if (pos >= (int)istate->cache_nr) |
| 204 | return NULL; |
| 205 | other = istate->cache[pos]; |
| 206 | if (ce_namelen(other) > this_len && |
| 207 | other->name[this_len] == '/' && |
| 208 | !strncmp(this_name, other->name, this_len)) |
| 209 | return other->name; |
| 210 | return NULL; |
| 211 | } |
| 212 | |
| 213 | static int verify_cache(struct index_state *istate, int flags) |
| 214 | { |
| 215 | unsigned i, funny; |
| 216 | int silent = flags & WRITE_TREE_SILENT; |
| 217 | |
| 218 | /* Verify that the tree is merged */ |
| 219 | funny = 0; |
| 220 | for (i = 0; i < istate->cache_nr; i++) { |
| 221 | const struct cache_entry *ce = istate->cache[i]; |
| 222 | if (ce_stage(ce)) { |
| 223 | if (silent) |
| 224 | return -1; |
| 225 | if (10 < ++funny) { |
| 226 | fprintf(stderr, "...\n"); |
| 227 | break; |
| 228 | } |
| 229 | fprintf(stderr, "%s: unmerged (%s)\n", |
| 230 | ce->name, oid_to_hex(&ce->oid)); |
| 231 | } |
| 232 | } |
| 233 | if (funny) |
| 234 | return -1; |
| 235 | |
| 236 | /* Also verify that the cache does not have path and path/file |
| 237 | * at the same time. At this point we know the cache has only |
| 238 | * stage 0 entries. |
| 239 | */ |
| 240 | funny = 0; |
| 241 | for (i = 0; i + 1 < istate->cache_nr; i++) { |
| 242 | const struct cache_entry *this_ce = istate->cache[i]; |
| 243 | const struct cache_entry *next_ce = istate->cache[i + 1]; |
| 244 | const char *conflict_name; |
| 245 | |
| 246 | conflict_name = find_df_conflict(istate, this_ce, next_ce); |
| 247 | if (conflict_name) { |
| 248 | if (10 < ++funny) { |
| 249 | fprintf(stderr, "...\n"); |
| 250 | break; |
| 251 | } |
| 252 | fprintf(stderr, "You have both %s and %s\n", |
| 253 | this_ce->name, conflict_name); |
| 254 | } |
| 255 | } |
| 256 | if (funny) |
| 257 | return -1; |
| 258 | return 0; |
| 259 | } |
| 260 | |
| 261 | static void discard_unused_subtrees(struct cache_tree *it) |
| 262 | { |
| 263 | struct cache_tree_sub **down = it->down; |
| 264 | int nr = it->subtree_nr; |
| 265 | int dst, src; |
| 266 | for (dst = src = 0; src < nr; src++) { |
| 267 | struct cache_tree_sub *s = down[src]; |
| 268 | if (s->used) |
| 269 | down[dst++] = s; |
| 270 | else { |
| 271 | cache_tree_free(&s->cache_tree); |
| 272 | free(s); |
| 273 | it->subtree_nr--; |
| 274 | } |
| 275 | } |
| 276 | } |
| 277 | |
| 278 | int cache_tree_fully_valid(struct cache_tree *it) |
| 279 | { |
| 280 | int i; |
| 281 | if (!it) |
| 282 | return 0; |
| 283 | if (it->entry_count < 0 || |
| 284 | !odb_has_object(the_repository->objects, &it->oid, |
| 285 | ODB_HAS_OBJECT_RECHECK_PACKED | ODB_HAS_OBJECT_FETCH_PROMISOR)) |
| 286 | return 0; |
| 287 | for (i = 0; i < it->subtree_nr; i++) { |
| 288 | if (!cache_tree_fully_valid(it->down[i]->cache_tree)) |
| 289 | return 0; |
| 290 | } |
| 291 | return 1; |
| 292 | } |
| 293 | |
| 294 | static int must_check_existence(const struct cache_entry *ce) |
| 295 | { |
| 296 | return !(repo_has_promisor_remote(the_repository) && ce_skip_worktree(ce)); |
| 297 | } |
| 298 | |
| 299 | static int update_one(struct cache_tree *it, |
| 300 | struct cache_entry **cache, |
| 301 | int entries, |
| 302 | const char *base, |
| 303 | int baselen, |
| 304 | int *skip_count, |
| 305 | int flags) |
| 306 | { |
| 307 | struct strbuf buffer; |
| 308 | int missing_ok = flags & WRITE_TREE_MISSING_OK; |
| 309 | int dryrun = flags & WRITE_TREE_DRY_RUN; |
| 310 | int repair = flags & WRITE_TREE_REPAIR; |
| 311 | int to_invalidate = 0; |
| 312 | int i; |
| 313 | |
| 314 | assert(!(dryrun && repair)); |
| 315 | |
| 316 | *skip_count = 0; |
| 317 | |
| 318 | /* |
| 319 | * If the first entry of this region is a sparse directory |
| 320 | * entry corresponding exactly to 'base', then this cache_tree |
| 321 | * struct is a "leaf" in the data structure, pointing to the |
| 322 | * tree OID specified in the entry. |
| 323 | */ |
| 324 | if (entries > 0) { |
| 325 | const struct cache_entry *ce = cache[0]; |
| 326 | |
| 327 | if (S_ISSPARSEDIR(ce->ce_mode) && |
| 328 | ce->ce_namelen == baselen && |
| 329 | !strncmp(ce->name, base, baselen)) { |
| 330 | it->entry_count = 1; |
| 331 | oidcpy(&it->oid, &ce->oid); |
| 332 | return 1; |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | if (0 <= it->entry_count && |
| 337 | odb_has_object(the_repository->objects, &it->oid, |
| 338 | ODB_HAS_OBJECT_RECHECK_PACKED | ODB_HAS_OBJECT_FETCH_PROMISOR)) |
| 339 | return it->entry_count; |
| 340 | |
| 341 | /* |
| 342 | * We first scan for subtrees and update them; we start by |
| 343 | * marking existing subtrees -- the ones that are unmarked |
| 344 | * should not be in the result. |
| 345 | */ |
| 346 | for (i = 0; i < it->subtree_nr; i++) |
| 347 | it->down[i]->used = 0; |
| 348 | |
| 349 | /* |
| 350 | * Find the subtrees and update them. |
| 351 | */ |
| 352 | i = 0; |
| 353 | while (i < entries) { |
| 354 | const struct cache_entry *ce = cache[i]; |
| 355 | struct cache_tree_sub *sub; |
| 356 | const char *path, *slash; |
| 357 | int pathlen, sublen, subcnt, subskip; |
| 358 | |
| 359 | path = ce->name; |
| 360 | pathlen = ce_namelen(ce); |
| 361 | if (pathlen <= baselen || memcmp(base, path, baselen)) |
| 362 | break; /* at the end of this level */ |
| 363 | |
| 364 | slash = strchr(path + baselen, '/'); |
| 365 | if (!slash) { |
| 366 | i++; |
| 367 | continue; |
| 368 | } |
| 369 | /* |
| 370 | * a/bbb/c (base = a/, slash = /c) |
| 371 | * ==> |
| 372 | * path+baselen = bbb/c, sublen = 3 |
| 373 | */ |
| 374 | sublen = slash - (path + baselen); |
| 375 | sub = find_subtree(it, path + baselen, sublen, 1); |
| 376 | if (!sub->cache_tree) |
| 377 | sub->cache_tree = cache_tree(); |
| 378 | subcnt = update_one(sub->cache_tree, |
| 379 | cache + i, entries - i, |
| 380 | path, |
| 381 | baselen + sublen + 1, |
| 382 | &subskip, |
| 383 | flags); |
| 384 | if (subcnt < 0) |
| 385 | return subcnt; |
| 386 | if (!subcnt) |
| 387 | die("index cache-tree records empty sub-tree"); |
| 388 | i += subcnt; |
| 389 | sub->count = subcnt; /* to be used in the next loop */ |
| 390 | *skip_count += subskip; |
| 391 | sub->used = 1; |
| 392 | } |
| 393 | |
| 394 | discard_unused_subtrees(it); |
| 395 | |
| 396 | /* |
| 397 | * Then write out the tree object for this level. |
| 398 | */ |
| 399 | strbuf_init(&buffer, 8192); |
| 400 | |
| 401 | i = 0; |
| 402 | while (i < entries) { |
| 403 | const struct cache_entry *ce = cache[i]; |
| 404 | struct cache_tree_sub *sub = NULL; |
| 405 | const char *path, *slash; |
| 406 | int pathlen, entlen; |
| 407 | const struct object_id *oid; |
| 408 | unsigned mode; |
| 409 | int expected_missing = 0; |
| 410 | int contains_ita = 0; |
| 411 | int ce_missing_ok; |
| 412 | |
| 413 | path = ce->name; |
| 414 | pathlen = ce_namelen(ce); |
| 415 | if (pathlen <= baselen || memcmp(base, path, baselen)) |
| 416 | break; /* at the end of this level */ |
| 417 | |
| 418 | slash = strchr(path + baselen, '/'); |
| 419 | if (slash) { |
| 420 | entlen = slash - (path + baselen); |
| 421 | sub = find_subtree(it, path + baselen, entlen, 0); |
| 422 | if (!sub) |
| 423 | die("cache-tree.c: '%.*s' in '%s' not found", |
| 424 | entlen, path + baselen, path); |
| 425 | i += sub->count; |
| 426 | oid = &sub->cache_tree->oid; |
| 427 | mode = S_IFDIR; |
| 428 | contains_ita = sub->cache_tree->entry_count < 0; |
| 429 | if (contains_ita) { |
| 430 | to_invalidate = 1; |
| 431 | expected_missing = 1; |
| 432 | } |
| 433 | } |
| 434 | else { |
| 435 | oid = &ce->oid; |
| 436 | mode = ce->ce_mode; |
| 437 | entlen = pathlen - baselen; |
| 438 | i++; |
| 439 | } |
| 440 | |
| 441 | ce_missing_ok = mode == S_IFGITLINK || missing_ok || |
| 442 | !must_check_existence(ce); |
| 443 | if (is_null_oid(oid) || |
| 444 | (!ce_missing_ok && |
| 445 | !odb_has_object(the_repository->objects, oid, |
| 446 | ODB_HAS_OBJECT_RECHECK_PACKED | ODB_HAS_OBJECT_FETCH_PROMISOR))) { |
| 447 | strbuf_release(&buffer); |
| 448 | if (expected_missing) |
| 449 | return -1; |
| 450 | return error("invalid object %06o %s for '%.*s'", |
| 451 | mode, oid_to_hex(oid), entlen+baselen, path); |
| 452 | } |
| 453 | |
| 454 | /* |
| 455 | * CE_REMOVE entries are removed before the index is |
| 456 | * written to disk. Skip them to remain consistent |
| 457 | * with the future on-disk index. |
| 458 | */ |
| 459 | if (ce->ce_flags & CE_REMOVE) { |
| 460 | *skip_count = *skip_count + 1; |
| 461 | continue; |
| 462 | } |
| 463 | |
| 464 | /* |
| 465 | * CE_INTENT_TO_ADD entries exist in on-disk index but |
| 466 | * they are not part of generated trees. Invalidate up |
| 467 | * to root to force cache-tree users to read elsewhere. |
| 468 | */ |
| 469 | if (!sub && ce_intent_to_add(ce)) { |
| 470 | to_invalidate = 1; |
| 471 | continue; |
| 472 | } |
| 473 | |
| 474 | /* |
| 475 | * "sub" can be an empty tree if all subentries are i-t-a. |
| 476 | */ |
| 477 | if (contains_ita && is_empty_tree_oid(oid, the_repository->hash_algo)) |
| 478 | continue; |
| 479 | |
| 480 | strbuf_grow(&buffer, entlen + 100); |
| 481 | strbuf_addf(&buffer, "%o %.*s%c", mode, entlen, path + baselen, '\0'); |
| 482 | strbuf_add(&buffer, oid->hash, the_hash_algo->rawsz); |
| 483 | |
| 484 | #if DEBUG_CACHE_TREE |
| 485 | fprintf(stderr, "cache-tree update-one %o %.*s\n", |
| 486 | mode, entlen, path + baselen); |
| 487 | #endif |
| 488 | } |
| 489 | |
| 490 | if (repair) { |
| 491 | struct object_id oid; |
| 492 | hash_object_file(the_hash_algo, buffer.buf, buffer.len, |
| 493 | OBJ_TREE, &oid); |
| 494 | if (odb_has_object(the_repository->objects, &oid, ODB_HAS_OBJECT_RECHECK_PACKED)) |
| 495 | oidcpy(&it->oid, &oid); |
| 496 | else |
| 497 | to_invalidate = 1; |
| 498 | } else if (dryrun) { |
| 499 | hash_object_file(the_hash_algo, buffer.buf, buffer.len, |
| 500 | OBJ_TREE, &it->oid); |
| 501 | } else if (odb_write_object_ext(the_repository->objects, buffer.buf, buffer.len, OBJ_TREE, |
| 502 | &it->oid, NULL, flags & WRITE_TREE_SILENT ? ODB_WRITE_OBJECT_SILENT : 0)) { |
| 503 | strbuf_release(&buffer); |
| 504 | return -1; |
| 505 | } |
| 506 | |
| 507 | strbuf_release(&buffer); |
| 508 | it->entry_count = to_invalidate ? -1 : i - *skip_count; |
| 509 | #if DEBUG_CACHE_TREE |
| 510 | fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n", |
| 511 | it->entry_count, it->subtree_nr, |
| 512 | oid_to_hex(&it->oid)); |
| 513 | #endif |
| 514 | return i; |
| 515 | } |
| 516 | |
| 517 | int cache_tree_update(struct index_state *istate, int flags) |
| 518 | { |
| 519 | int inflight = !!the_repository->objects->transaction; |
| 520 | struct odb_transaction *transaction; |
| 521 | int skip, i; |
| 522 | |
| 523 | i = verify_cache(istate, flags); |
| 524 | |
| 525 | if (i) |
| 526 | return i; |
| 527 | |
| 528 | if (!istate->cache_tree) |
| 529 | istate->cache_tree = cache_tree(); |
| 530 | |
| 531 | if (!(flags & WRITE_TREE_MISSING_OK) && repo_has_promisor_remote(the_repository)) |
| 532 | prefetch_cache_entries(istate, must_check_existence); |
| 533 | |
| 534 | trace_performance_enter(); |
| 535 | trace2_region_enter("cache_tree", "update", istate->repo); |
| 536 | if (!inflight) |
| 537 | odb_transaction_begin_or_die(the_repository->objects, &transaction, 0); |
| 538 | i = update_one(istate->cache_tree, istate->cache, istate->cache_nr, |
| 539 | "", 0, &skip, flags); |
| 540 | if (!inflight) |
| 541 | odb_transaction_commit(transaction); |
| 542 | trace2_region_leave("cache_tree", "update", istate->repo); |
| 543 | trace_performance_leave("cache_tree_update"); |
| 544 | if (i < 0) |
| 545 | return i; |
| 546 | istate->cache_changed |= CACHE_TREE_CHANGED; |
| 547 | return 0; |
| 548 | } |
| 549 | |
| 550 | static void write_one(struct strbuf *buffer, struct cache_tree *it, |
| 551 | const char *path, int pathlen) |
| 552 | { |
| 553 | int i; |
| 554 | |
| 555 | /* One "cache-tree" entry consists of the following: |
| 556 | * path (NUL terminated) |
| 557 | * entry_count, subtree_nr ("%d %d\n") |
| 558 | * tree-sha1 (missing if invalid) |
| 559 | * subtree_nr "cache-tree" entries for subtrees. |
| 560 | */ |
| 561 | strbuf_grow(buffer, pathlen + 100); |
| 562 | strbuf_add(buffer, path, pathlen); |
| 563 | strbuf_addf(buffer, "%c%d %d\n", 0, it->entry_count, it->subtree_nr); |
| 564 | |
| 565 | #if DEBUG_CACHE_TREE |
| 566 | if (0 <= it->entry_count) |
| 567 | fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n", |
| 568 | pathlen, path, it->entry_count, it->subtree_nr, |
| 569 | oid_to_hex(&it->oid)); |
| 570 | else |
| 571 | fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n", |
| 572 | pathlen, path, it->subtree_nr); |
| 573 | #endif |
| 574 | |
| 575 | if (0 <= it->entry_count) { |
| 576 | strbuf_add(buffer, it->oid.hash, the_hash_algo->rawsz); |
| 577 | } |
| 578 | for (i = 0; i < it->subtree_nr; i++) { |
| 579 | struct cache_tree_sub *down = it->down[i]; |
| 580 | if (i) { |
| 581 | struct cache_tree_sub *prev = it->down[i-1]; |
| 582 | if (subtree_name_cmp(down->name, down->namelen, |
| 583 | prev->name, prev->namelen) <= 0) |
| 584 | die("fatal - unsorted cache subtree"); |
| 585 | } |
| 586 | write_one(buffer, down->cache_tree, down->name, down->namelen); |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | void cache_tree_write(struct strbuf *sb, struct cache_tree *root) |
| 591 | { |
| 592 | trace2_region_enter("cache_tree", "write", the_repository); |
| 593 | write_one(sb, root, "", 0); |
| 594 | trace2_region_leave("cache_tree", "write", the_repository); |
| 595 | } |
| 596 | |
| 597 | static int parse_int(const char **ptr, unsigned long *len_p, int *out) |
| 598 | { |
| 599 | const char *s = *ptr; |
| 600 | unsigned long len = *len_p; |
| 601 | int ret = 0; |
| 602 | int sign = 1; |
| 603 | |
| 604 | while (len && *s == '-') { |
| 605 | sign *= -1; |
| 606 | s++; |
| 607 | len--; |
| 608 | } |
| 609 | |
| 610 | while (len) { |
| 611 | if (!isdigit(*s)) |
| 612 | break; |
| 613 | ret *= 10; |
| 614 | ret += *s - '0'; |
| 615 | s++; |
| 616 | len--; |
| 617 | } |
| 618 | |
| 619 | if (s == *ptr) |
| 620 | return -1; |
| 621 | |
| 622 | *ptr = s; |
| 623 | *len_p = len; |
| 624 | *out = sign * ret; |
| 625 | return 0; |
| 626 | } |
| 627 | |
| 628 | static struct cache_tree *read_one(const char **buffer, unsigned long *size_p) |
| 629 | { |
| 630 | const char *buf = *buffer; |
| 631 | unsigned long size = *size_p; |
| 632 | struct cache_tree *it; |
| 633 | int i, subtree_nr; |
| 634 | const unsigned rawsz = the_hash_algo->rawsz; |
| 635 | |
| 636 | it = NULL; |
| 637 | /* skip name, but make sure name exists */ |
| 638 | while (size && *buf) { |
| 639 | size--; |
| 640 | buf++; |
| 641 | } |
| 642 | if (!size) |
| 643 | goto free_return; |
| 644 | buf++; size--; |
| 645 | it = cache_tree(); |
| 646 | |
| 647 | if (parse_int(&buf, &size, &it->entry_count) < 0) |
| 648 | goto free_return; |
| 649 | if (!size || *buf != ' ') |
| 650 | goto free_return; |
| 651 | buf++; size--; |
| 652 | if (parse_int(&buf, &size, &subtree_nr) < 0) |
| 653 | goto free_return; |
| 654 | if (!size || *buf != '\n') |
| 655 | goto free_return; |
| 656 | buf++; size--; |
| 657 | if (0 <= it->entry_count) { |
| 658 | if (size < rawsz) |
| 659 | goto free_return; |
| 660 | oidread(&it->oid, (const unsigned char *)buf, |
| 661 | the_repository->hash_algo); |
| 662 | buf += rawsz; |
| 663 | size -= rawsz; |
| 664 | } |
| 665 | |
| 666 | #if DEBUG_CACHE_TREE |
| 667 | if (0 <= it->entry_count) |
| 668 | fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n", |
| 669 | *buffer, it->entry_count, subtree_nr, |
| 670 | oid_to_hex(&it->oid)); |
| 671 | else |
| 672 | fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n", |
| 673 | *buffer, subtree_nr); |
| 674 | #endif |
| 675 | |
| 676 | /* |
| 677 | * Just a heuristic -- we do not add directories that often but |
| 678 | * we do not want to have to extend it immediately when we do, |
| 679 | * hence +2. |
| 680 | */ |
| 681 | it->subtree_alloc = subtree_nr + 2; |
| 682 | CALLOC_ARRAY(it->down, it->subtree_alloc); |
| 683 | for (i = 0; i < subtree_nr; i++) { |
| 684 | /* read each subtree */ |
| 685 | struct cache_tree *sub; |
| 686 | struct cache_tree_sub *subtree; |
| 687 | const char *name = buf; |
| 688 | |
| 689 | sub = read_one(&buf, &size); |
| 690 | if (!sub) |
| 691 | goto free_return; |
| 692 | subtree = cache_tree_sub(it, name); |
| 693 | subtree->cache_tree = sub; |
| 694 | } |
| 695 | if (subtree_nr != it->subtree_nr) |
| 696 | die("cache-tree: internal error"); |
| 697 | *buffer = buf; |
| 698 | *size_p = size; |
| 699 | return it; |
| 700 | |
| 701 | free_return: |
| 702 | cache_tree_free(&it); |
| 703 | return NULL; |
| 704 | } |
| 705 | |
| 706 | struct cache_tree *cache_tree_read(const char *buffer, unsigned long size) |
| 707 | { |
| 708 | struct cache_tree *result; |
| 709 | |
| 710 | if (buffer[0]) |
| 711 | return NULL; /* not the whole tree */ |
| 712 | |
| 713 | trace2_region_enter("cache_tree", "read", the_repository); |
| 714 | result = read_one(&buffer, &size); |
| 715 | trace2_region_leave("cache_tree", "read", the_repository); |
| 716 | |
| 717 | return result; |
| 718 | } |
| 719 | |
| 720 | static struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path) |
| 721 | { |
| 722 | if (!it) |
| 723 | return NULL; |
| 724 | while (*path) { |
| 725 | const char *slash; |
| 726 | struct cache_tree_sub *sub; |
| 727 | |
| 728 | slash = strchrnul(path, '/'); |
| 729 | /* |
| 730 | * Between path and slash is the name of the subtree |
| 731 | * to look for. |
| 732 | */ |
| 733 | sub = find_subtree(it, path, slash - path, 0); |
| 734 | if (!sub) |
| 735 | return NULL; |
| 736 | it = sub->cache_tree; |
| 737 | |
| 738 | path = slash; |
| 739 | while (*path == '/') |
| 740 | path++; |
| 741 | } |
| 742 | return it; |
| 743 | } |
| 744 | |
| 745 | static int write_index_as_tree_internal(struct object_id *oid, |
| 746 | struct index_state *index_state, |
| 747 | int cache_tree_valid, |
| 748 | int flags, |
| 749 | const char *prefix) |
| 750 | { |
| 751 | if (flags & WRITE_TREE_IGNORE_CACHE_TREE) { |
| 752 | cache_tree_free(&index_state->cache_tree); |
| 753 | cache_tree_valid = 0; |
| 754 | } |
| 755 | |
| 756 | if (!cache_tree_valid && cache_tree_update(index_state, flags) < 0) |
| 757 | return WRITE_TREE_UNMERGED_INDEX; |
| 758 | |
| 759 | if (prefix) { |
| 760 | struct cache_tree *subtree; |
| 761 | subtree = cache_tree_find(index_state->cache_tree, prefix); |
| 762 | if (!subtree) |
| 763 | return WRITE_TREE_PREFIX_ERROR; |
| 764 | oidcpy(oid, &subtree->oid); |
| 765 | } |
| 766 | else |
| 767 | oidcpy(oid, &index_state->cache_tree->oid); |
| 768 | |
| 769 | return 0; |
| 770 | } |
| 771 | |
| 772 | struct tree *write_in_core_index_as_tree(struct repository *repo, |
| 773 | struct index_state *index_state) { |
| 774 | struct object_id o; |
| 775 | int was_valid, ret; |
| 776 | |
| 777 | was_valid = index_state->cache_tree && |
| 778 | cache_tree_fully_valid(index_state->cache_tree); |
| 779 | |
| 780 | ret = write_index_as_tree_internal(&o, index_state, was_valid, 0, NULL); |
| 781 | if (ret == WRITE_TREE_UNMERGED_INDEX) { |
| 782 | int i; |
| 783 | bug("there are unmerged index entries:"); |
| 784 | for (i = 0; i < index_state->cache_nr; i++) { |
| 785 | const struct cache_entry *ce = index_state->cache[i]; |
| 786 | if (ce_stage(ce)) |
| 787 | bug("%d %.*s", ce_stage(ce), |
| 788 | (int)ce_namelen(ce), ce->name); |
| 789 | } |
| 790 | BUG("unmerged index entries when writing in-core index"); |
| 791 | } |
| 792 | |
| 793 | return lookup_tree(repo, &index_state->cache_tree->oid); |
| 794 | } |
| 795 | |
| 796 | |
| 797 | int write_index_as_tree(struct object_id *oid, struct index_state *index_state, const char *index_path, int flags, const char *prefix) |
| 798 | { |
| 799 | int entries, was_valid; |
| 800 | struct lock_file lock_file = LOCK_INIT; |
| 801 | int ret; |
| 802 | |
| 803 | hold_lock_file_for_update(&lock_file, index_path, LOCK_DIE_ON_ERROR); |
| 804 | |
| 805 | entries = read_index_from(index_state, index_path, |
| 806 | repo_get_git_dir(the_repository)); |
| 807 | if (entries < 0) { |
| 808 | ret = WRITE_TREE_UNREADABLE_INDEX; |
| 809 | goto out; |
| 810 | } |
| 811 | |
| 812 | was_valid = !(flags & WRITE_TREE_IGNORE_CACHE_TREE) && |
| 813 | index_state->cache_tree && |
| 814 | cache_tree_fully_valid(index_state->cache_tree); |
| 815 | |
| 816 | ret = write_index_as_tree_internal(oid, index_state, was_valid, flags, |
| 817 | prefix); |
| 818 | if (!ret && !was_valid) { |
| 819 | write_locked_index(index_state, &lock_file, COMMIT_LOCK); |
| 820 | /* Not being able to write is fine -- we are only interested |
| 821 | * in updating the cache-tree part, and if the next caller |
| 822 | * ends up using the old index with unupdated cache-tree part |
| 823 | * it misses the work we did here, but that is just a |
| 824 | * performance penalty and not a big deal. |
| 825 | */ |
| 826 | } |
| 827 | |
| 828 | out: |
| 829 | rollback_lock_file(&lock_file); |
| 830 | return ret; |
| 831 | } |
| 832 | |
| 833 | static void prime_cache_tree_sparse_dir(struct cache_tree *it, |
| 834 | struct tree *tree) |
| 835 | { |
| 836 | |
| 837 | oidcpy(&it->oid, &tree->object.oid); |
| 838 | it->entry_count = 1; |
| 839 | } |
| 840 | |
| 841 | static void prime_cache_tree_rec(struct repository *r, |
| 842 | struct cache_tree *it, |
| 843 | struct tree *tree, |
| 844 | struct strbuf *tree_path) |
| 845 | { |
| 846 | struct tree_desc desc; |
| 847 | struct name_entry entry; |
| 848 | int cnt; |
| 849 | size_t base_path_len = tree_path->len; |
| 850 | |
| 851 | oidcpy(&it->oid, &tree->object.oid); |
| 852 | |
| 853 | init_tree_desc(&desc, &tree->object.oid, tree->buffer, tree->size); |
| 854 | cnt = 0; |
| 855 | while (tree_entry(&desc, &entry)) { |
| 856 | if (!S_ISDIR(entry.mode)) |
| 857 | cnt++; |
| 858 | else { |
| 859 | struct cache_tree_sub *sub; |
| 860 | struct tree *subtree = lookup_tree(r, &entry.oid); |
| 861 | |
| 862 | if (repo_parse_tree(the_repository, subtree) < 0) |
| 863 | exit(128); |
| 864 | sub = cache_tree_sub(it, entry.path); |
| 865 | sub->cache_tree = cache_tree(); |
| 866 | |
| 867 | /* |
| 868 | * Recursively-constructed subtree path is only needed when working |
| 869 | * in a sparse index (where it's used to determine whether the |
| 870 | * subtree is a sparse directory in the index). |
| 871 | */ |
| 872 | if (r->index->sparse_index) { |
| 873 | strbuf_setlen(tree_path, base_path_len); |
| 874 | strbuf_add(tree_path, entry.path, entry.pathlen); |
| 875 | strbuf_addch(tree_path, '/'); |
| 876 | } |
| 877 | |
| 878 | /* |
| 879 | * If a sparse index is in use, the directory being processed may be |
| 880 | * sparse. To confirm that, we can check whether an entry with that |
| 881 | * exact name exists in the index. If it does, the created subtree |
| 882 | * should be sparse. Otherwise, cache tree expansion should continue |
| 883 | * as normal. |
| 884 | */ |
| 885 | if (r->index->sparse_index && |
| 886 | index_entry_exists(r->index, tree_path->buf, tree_path->len)) |
| 887 | prime_cache_tree_sparse_dir(sub->cache_tree, subtree); |
| 888 | else |
| 889 | prime_cache_tree_rec(r, sub->cache_tree, subtree, tree_path); |
| 890 | cnt += sub->cache_tree->entry_count; |
| 891 | } |
| 892 | } |
| 893 | |
| 894 | it->entry_count = cnt; |
| 895 | } |
| 896 | |
| 897 | void prime_cache_tree(struct repository *r, |
| 898 | struct index_state *istate, |
| 899 | struct tree *tree) |
| 900 | { |
| 901 | struct strbuf tree_path = STRBUF_INIT; |
| 902 | |
| 903 | trace2_region_enter("cache-tree", "prime_cache_tree", r); |
| 904 | cache_tree_free(&istate->cache_tree); |
| 905 | istate->cache_tree = cache_tree(); |
| 906 | |
| 907 | prime_cache_tree_rec(r, istate->cache_tree, tree, &tree_path); |
| 908 | strbuf_release(&tree_path); |
| 909 | istate->cache_changed |= CACHE_TREE_CHANGED; |
| 910 | trace2_region_leave("cache-tree", "prime_cache_tree", r); |
| 911 | } |
| 912 | |
| 913 | /* |
| 914 | * find the cache_tree that corresponds to the current level without |
| 915 | * exploding the full path into textual form. The root of the |
| 916 | * cache tree is given as "root", and our current level is "info". |
| 917 | * (1) When at root level, info->prev is NULL, so it is "root" itself. |
| 918 | * (2) Otherwise, find the cache_tree that corresponds to one level |
| 919 | * above us, and find ourselves in there. |
| 920 | */ |
| 921 | static struct cache_tree *find_cache_tree_from_traversal(struct cache_tree *root, |
| 922 | struct traverse_info *info) |
| 923 | { |
| 924 | struct cache_tree *our_parent; |
| 925 | |
| 926 | if (!info->prev) |
| 927 | return root; |
| 928 | our_parent = find_cache_tree_from_traversal(root, info->prev); |
| 929 | return cache_tree_find(our_parent, info->name); |
| 930 | } |
| 931 | |
| 932 | int cache_tree_matches_traversal(struct cache_tree *root, |
| 933 | struct name_entry *ent, |
| 934 | struct traverse_info *info) |
| 935 | { |
| 936 | struct cache_tree *it; |
| 937 | |
| 938 | it = find_cache_tree_from_traversal(root, info); |
| 939 | it = cache_tree_find(it, ent->path); |
| 940 | if (it && it->entry_count > 0 && oideq(&ent->oid, &it->oid)) |
| 941 | return it->entry_count; |
| 942 | return 0; |
| 943 | } |
| 944 | |
| 945 | static int verify_one_sparse(struct index_state *istate, |
| 946 | struct strbuf *path, |
| 947 | int pos) |
| 948 | { |
| 949 | struct cache_entry *ce = istate->cache[pos]; |
| 950 | if (!S_ISSPARSEDIR(ce->ce_mode)) |
| 951 | return error(_("directory '%s' is present in index, but not sparse"), |
| 952 | path->buf); |
| 953 | return 0; |
| 954 | } |
| 955 | |
| 956 | /* |
| 957 | * Returns: |
| 958 | * 0 - Verification completed. |
| 959 | * 1 - Restart verification - a call to ensure_full_index() freed the cache |
| 960 | * tree that is being verified and verification needs to be restarted from |
| 961 | * the new toplevel cache tree. |
| 962 | * -1 - Verification failed. |
| 963 | */ |
| 964 | static int verify_one(struct repository *r, |
| 965 | struct index_state *istate, |
| 966 | struct cache_tree *it, |
| 967 | struct strbuf *path) |
| 968 | { |
| 969 | int i, pos, len = path->len; |
| 970 | struct strbuf tree_buf = STRBUF_INIT; |
| 971 | struct object_id new_oid; |
| 972 | int ret; |
| 973 | |
| 974 | for (i = 0; i < it->subtree_nr; i++) { |
| 975 | strbuf_addf(path, "%s/", it->down[i]->name); |
| 976 | ret = verify_one(r, istate, it->down[i]->cache_tree, path); |
| 977 | if (ret) |
| 978 | goto out; |
| 979 | |
| 980 | strbuf_setlen(path, len); |
| 981 | } |
| 982 | |
| 983 | if (it->entry_count < 0 || |
| 984 | /* no verification on tests (t7003) that replace trees */ |
| 985 | lookup_replace_object(r, &it->oid) != &it->oid) { |
| 986 | ret = 0; |
| 987 | goto out; |
| 988 | } |
| 989 | |
| 990 | if (path->len) { |
| 991 | /* |
| 992 | * If the index is sparse and the cache tree is not |
| 993 | * index_name_pos() may trigger ensure_full_index() which will |
| 994 | * free the tree that is being verified. |
| 995 | */ |
| 996 | int is_sparse = istate->sparse_index; |
| 997 | pos = index_name_pos(istate, path->buf, path->len); |
| 998 | if (is_sparse && !istate->sparse_index) { |
| 999 | ret = 1; |
| 1000 | goto out; |
| 1001 | } |
| 1002 | |
| 1003 | if (pos >= 0) { |
| 1004 | ret = verify_one_sparse(istate, path, pos); |
| 1005 | goto out; |
| 1006 | } |
| 1007 | |
| 1008 | pos = -pos - 1; |
| 1009 | } else { |
| 1010 | pos = 0; |
| 1011 | } |
| 1012 | |
| 1013 | if (it->entry_count + pos > istate->cache_nr) { |
| 1014 | ret = error(_("corrupted cache-tree has entries not present in index")); |
| 1015 | goto out; |
| 1016 | } |
| 1017 | |
| 1018 | i = 0; |
| 1019 | while (i < it->entry_count) { |
| 1020 | struct cache_entry *ce = istate->cache[pos + i]; |
| 1021 | const char *slash; |
| 1022 | struct cache_tree_sub *sub = NULL; |
| 1023 | const struct object_id *oid; |
| 1024 | const char *name; |
| 1025 | unsigned mode; |
| 1026 | int entlen; |
| 1027 | |
| 1028 | if (ce->ce_flags & (CE_STAGEMASK | CE_INTENT_TO_ADD | CE_REMOVE)) { |
| 1029 | ret = error(_("%s with flags 0x%x should not be in cache-tree"), |
| 1030 | ce->name, ce->ce_flags); |
| 1031 | goto out; |
| 1032 | } |
| 1033 | |
| 1034 | name = ce->name + path->len; |
| 1035 | slash = strchr(name, '/'); |
| 1036 | if (slash) { |
| 1037 | entlen = slash - name; |
| 1038 | |
| 1039 | sub = find_subtree(it, ce->name + path->len, entlen, 0); |
| 1040 | if (!sub || sub->cache_tree->entry_count < 0) { |
| 1041 | ret = error(_("bad subtree '%.*s'"), entlen, name); |
| 1042 | goto out; |
| 1043 | } |
| 1044 | |
| 1045 | oid = &sub->cache_tree->oid; |
| 1046 | mode = S_IFDIR; |
| 1047 | i += sub->cache_tree->entry_count; |
| 1048 | } else { |
| 1049 | oid = &ce->oid; |
| 1050 | mode = ce->ce_mode; |
| 1051 | entlen = ce_namelen(ce) - path->len; |
| 1052 | i++; |
| 1053 | } |
| 1054 | strbuf_addf(&tree_buf, "%o %.*s%c", mode, entlen, name, '\0'); |
| 1055 | strbuf_add(&tree_buf, oid->hash, r->hash_algo->rawsz); |
| 1056 | } |
| 1057 | |
| 1058 | hash_object_file(r->hash_algo, tree_buf.buf, tree_buf.len, OBJ_TREE, |
| 1059 | &new_oid); |
| 1060 | |
| 1061 | if (!oideq(&new_oid, &it->oid)) { |
| 1062 | ret = error(_("cache-tree for path %.*s does not match. " |
| 1063 | "Expected %s got %s"), len, path->buf, |
| 1064 | oid_to_hex(&new_oid), oid_to_hex(&it->oid)); |
| 1065 | goto out; |
| 1066 | } |
| 1067 | |
| 1068 | ret = 0; |
| 1069 | out: |
| 1070 | strbuf_setlen(path, len); |
| 1071 | strbuf_release(&tree_buf); |
| 1072 | return ret; |
| 1073 | } |
| 1074 | |
| 1075 | int cache_tree_verify(struct repository *r, struct index_state *istate) |
| 1076 | { |
| 1077 | struct strbuf path = STRBUF_INIT; |
| 1078 | int ret; |
| 1079 | |
| 1080 | if (!istate->cache_tree) { |
| 1081 | ret = 0; |
| 1082 | goto out; |
| 1083 | } |
| 1084 | |
| 1085 | ret = verify_one(r, istate, istate->cache_tree, &path); |
| 1086 | if (ret < 0) |
| 1087 | goto out; |
| 1088 | if (ret > 0) { |
| 1089 | strbuf_reset(&path); |
| 1090 | |
| 1091 | ret = verify_one(r, istate, istate->cache_tree, &path); |
| 1092 | if (ret < 0) |
| 1093 | goto out; |
| 1094 | if (ret > 0) |
| 1095 | BUG("ensure_full_index() called twice while verifying cache tree"); |
| 1096 | } |
| 1097 | |
| 1098 | ret = 0; |
| 1099 | |
| 1100 | out: |
| 1101 | strbuf_release(&path); |
| 1102 | return ret; |
| 1103 | } |