refs / ref-cache.con commit fsck: handle NULL return of lookup_blob() and lookup_tree() (2720f6d)
   1#include "../cache.h"
   2#include "../refs.h"
   3#include "refs-internal.h"
   4#include "ref-cache.h"
   5#include "../iterator.h"
   6
   7void add_entry_to_dir(struct ref_dir *dir, struct ref_entry *entry)
   8{
   9        ALLOC_GROW(dir->entries, dir->nr + 1, dir->alloc);
  10        dir->entries[dir->nr++] = entry;
  11        /* optimize for the case that entries are added in order */
  12        if (dir->nr == 1 ||
  13            (dir->nr == dir->sorted + 1 &&
  14             strcmp(dir->entries[dir->nr - 2]->name,
  15                    dir->entries[dir->nr - 1]->name) < 0))
  16                dir->sorted = dir->nr;
  17}
  18
  19struct ref_dir *get_ref_dir(struct ref_entry *entry)
  20{
  21        struct ref_dir *dir;
  22        assert(entry->flag & REF_DIR);
  23        dir = &entry->u.subdir;
  24        if (entry->flag & REF_INCOMPLETE) {
  25                if (!dir->cache->fill_ref_dir)
  26                        die("BUG: incomplete ref_store without fill_ref_dir function");
  27
  28                dir->cache->fill_ref_dir(dir->cache->ref_store, dir, entry->name);
  29                entry->flag &= ~REF_INCOMPLETE;
  30        }
  31        return dir;
  32}
  33
  34struct ref_entry *create_ref_entry(const char *refname,
  35                                   const struct object_id *oid, int flag)
  36{
  37        struct ref_entry *ref;
  38
  39        FLEX_ALLOC_STR(ref, name, refname);
  40        oidcpy(&ref->u.value.oid, oid);
  41        oidclr(&ref->u.value.peeled);
  42        ref->flag = flag;
  43        return ref;
  44}
  45
  46struct ref_cache *create_ref_cache(struct ref_store *refs,
  47                                   fill_ref_dir_fn *fill_ref_dir)
  48{
  49        struct ref_cache *ret = xcalloc(1, sizeof(*ret));
  50
  51        ret->ref_store = refs;
  52        ret->fill_ref_dir = fill_ref_dir;
  53        ret->root = create_dir_entry(ret, "", 0, 1);
  54        return ret;
  55}
  56
  57static void clear_ref_dir(struct ref_dir *dir);
  58
  59static void free_ref_entry(struct ref_entry *entry)
  60{
  61        if (entry->flag & REF_DIR) {
  62                /*
  63                 * Do not use get_ref_dir() here, as that might
  64                 * trigger the reading of loose refs.
  65                 */
  66                clear_ref_dir(&entry->u.subdir);
  67        }
  68        free(entry);
  69}
  70
  71void free_ref_cache(struct ref_cache *cache)
  72{
  73        free_ref_entry(cache->root);
  74        free(cache);
  75}
  76
  77/*
  78 * Clear and free all entries in dir, recursively.
  79 */
  80static void clear_ref_dir(struct ref_dir *dir)
  81{
  82        int i;
  83        for (i = 0; i < dir->nr; i++)
  84                free_ref_entry(dir->entries[i]);
  85        FREE_AND_NULL(dir->entries);
  86        dir->sorted = dir->nr = dir->alloc = 0;
  87}
  88
  89struct ref_entry *create_dir_entry(struct ref_cache *cache,
  90                                   const char *dirname, size_t len,
  91                                   int incomplete)
  92{
  93        struct ref_entry *direntry;
  94
  95        FLEX_ALLOC_MEM(direntry, name, dirname, len);
  96        direntry->u.subdir.cache = cache;
  97        direntry->flag = REF_DIR | (incomplete ? REF_INCOMPLETE : 0);
  98        return direntry;
  99}
 100
 101static int ref_entry_cmp(const void *a, const void *b)
 102{
 103        struct ref_entry *one = *(struct ref_entry **)a;
 104        struct ref_entry *two = *(struct ref_entry **)b;
 105        return strcmp(one->name, two->name);
 106}
 107
 108static void sort_ref_dir(struct ref_dir *dir);
 109
 110struct string_slice {
 111        size_t len;
 112        const char *str;
 113};
 114
 115static int ref_entry_cmp_sslice(const void *key_, const void *ent_)
 116{
 117        const struct string_slice *key = key_;
 118        const struct ref_entry *ent = *(const struct ref_entry * const *)ent_;
 119        int cmp = strncmp(key->str, ent->name, key->len);
 120        if (cmp)
 121                return cmp;
 122        return '\0' - (unsigned char)ent->name[key->len];
 123}
 124
 125int search_ref_dir(struct ref_dir *dir, const char *refname, size_t len)
 126{
 127        struct ref_entry **r;
 128        struct string_slice key;
 129
 130        if (refname == NULL || !dir->nr)
 131                return -1;
 132
 133        sort_ref_dir(dir);
 134        key.len = len;
 135        key.str = refname;
 136        r = bsearch(&key, dir->entries, dir->nr, sizeof(*dir->entries),
 137                    ref_entry_cmp_sslice);
 138
 139        if (r == NULL)
 140                return -1;
 141
 142        return r - dir->entries;
 143}
 144
 145/*
 146 * Search for a directory entry directly within dir (without
 147 * recursing).  Sort dir if necessary.  subdirname must be a directory
 148 * name (i.e., end in '/').  If mkdir is set, then create the
 149 * directory if it is missing; otherwise, return NULL if the desired
 150 * directory cannot be found.  dir must already be complete.
 151 */
 152static struct ref_dir *search_for_subdir(struct ref_dir *dir,
 153                                         const char *subdirname, size_t len,
 154                                         int mkdir)
 155{
 156        int entry_index = search_ref_dir(dir, subdirname, len);
 157        struct ref_entry *entry;
 158        if (entry_index == -1) {
 159                if (!mkdir)
 160                        return NULL;
 161                /*
 162                 * Since dir is complete, the absence of a subdir
 163                 * means that the subdir really doesn't exist;
 164                 * therefore, create an empty record for it but mark
 165                 * the record complete.
 166                 */
 167                entry = create_dir_entry(dir->cache, subdirname, len, 0);
 168                add_entry_to_dir(dir, entry);
 169        } else {
 170                entry = dir->entries[entry_index];
 171        }
 172        return get_ref_dir(entry);
 173}
 174
 175/*
 176 * If refname is a reference name, find the ref_dir within the dir
 177 * tree that should hold refname. If refname is a directory name
 178 * (i.e., it ends in '/'), then return that ref_dir itself. dir must
 179 * represent the top-level directory and must already be complete.
 180 * Sort ref_dirs and recurse into subdirectories as necessary. If
 181 * mkdir is set, then create any missing directories; otherwise,
 182 * return NULL if the desired directory cannot be found.
 183 */
 184static struct ref_dir *find_containing_dir(struct ref_dir *dir,
 185                                           const char *refname, int mkdir)
 186{
 187        const char *slash;
 188        for (slash = strchr(refname, '/'); slash; slash = strchr(slash + 1, '/')) {
 189                size_t dirnamelen = slash - refname + 1;
 190                struct ref_dir *subdir;
 191                subdir = search_for_subdir(dir, refname, dirnamelen, mkdir);
 192                if (!subdir) {
 193                        dir = NULL;
 194                        break;
 195                }
 196                dir = subdir;
 197        }
 198
 199        return dir;
 200}
 201
 202struct ref_entry *find_ref_entry(struct ref_dir *dir, const char *refname)
 203{
 204        int entry_index;
 205        struct ref_entry *entry;
 206        dir = find_containing_dir(dir, refname, 0);
 207        if (!dir)
 208                return NULL;
 209        entry_index = search_ref_dir(dir, refname, strlen(refname));
 210        if (entry_index == -1)
 211                return NULL;
 212        entry = dir->entries[entry_index];
 213        return (entry->flag & REF_DIR) ? NULL : entry;
 214}
 215
 216int remove_entry_from_dir(struct ref_dir *dir, const char *refname)
 217{
 218        int refname_len = strlen(refname);
 219        int entry_index;
 220        struct ref_entry *entry;
 221        int is_dir = refname[refname_len - 1] == '/';
 222        if (is_dir) {
 223                /*
 224                 * refname represents a reference directory.  Remove
 225                 * the trailing slash; otherwise we will get the
 226                 * directory *representing* refname rather than the
 227                 * one *containing* it.
 228                 */
 229                char *dirname = xmemdupz(refname, refname_len - 1);
 230                dir = find_containing_dir(dir, dirname, 0);
 231                free(dirname);
 232        } else {
 233                dir = find_containing_dir(dir, refname, 0);
 234        }
 235        if (!dir)
 236                return -1;
 237        entry_index = search_ref_dir(dir, refname, refname_len);
 238        if (entry_index == -1)
 239                return -1;
 240        entry = dir->entries[entry_index];
 241
 242        memmove(&dir->entries[entry_index],
 243                &dir->entries[entry_index + 1],
 244                (dir->nr - entry_index - 1) * sizeof(*dir->entries)
 245                );
 246        dir->nr--;
 247        if (dir->sorted > entry_index)
 248                dir->sorted--;
 249        free_ref_entry(entry);
 250        return dir->nr;
 251}
 252
 253int add_ref_entry(struct ref_dir *dir, struct ref_entry *ref)
 254{
 255        dir = find_containing_dir(dir, ref->name, 1);
 256        if (!dir)
 257                return -1;
 258        add_entry_to_dir(dir, ref);
 259        return 0;
 260}
 261
 262/*
 263 * Emit a warning and return true iff ref1 and ref2 have the same name
 264 * and the same sha1.  Die if they have the same name but different
 265 * sha1s.
 266 */
 267static int is_dup_ref(const struct ref_entry *ref1, const struct ref_entry *ref2)
 268{
 269        if (strcmp(ref1->name, ref2->name))
 270                return 0;
 271
 272        /* Duplicate name; make sure that they don't conflict: */
 273
 274        if ((ref1->flag & REF_DIR) || (ref2->flag & REF_DIR))
 275                /* This is impossible by construction */
 276                die("Reference directory conflict: %s", ref1->name);
 277
 278        if (oidcmp(&ref1->u.value.oid, &ref2->u.value.oid))
 279                die("Duplicated ref, and SHA1s don't match: %s", ref1->name);
 280
 281        warning("Duplicated ref: %s", ref1->name);
 282        return 1;
 283}
 284
 285/*
 286 * Sort the entries in dir non-recursively (if they are not already
 287 * sorted) and remove any duplicate entries.
 288 */
 289static void sort_ref_dir(struct ref_dir *dir)
 290{
 291        int i, j;
 292        struct ref_entry *last = NULL;
 293
 294        /*
 295         * This check also prevents passing a zero-length array to qsort(),
 296         * which is a problem on some platforms.
 297         */
 298        if (dir->sorted == dir->nr)
 299                return;
 300
 301        QSORT(dir->entries, dir->nr, ref_entry_cmp);
 302
 303        /* Remove any duplicates: */
 304        for (i = 0, j = 0; j < dir->nr; j++) {
 305                struct ref_entry *entry = dir->entries[j];
 306                if (last && is_dup_ref(last, entry))
 307                        free_ref_entry(entry);
 308                else
 309                        last = dir->entries[i++] = entry;
 310        }
 311        dir->sorted = dir->nr = i;
 312}
 313
 314enum prefix_state {
 315        /* All refs within the directory would match prefix: */
 316        PREFIX_CONTAINS_DIR,
 317
 318        /* Some, but not all, refs within the directory might match prefix: */
 319        PREFIX_WITHIN_DIR,
 320
 321        /* No refs within the directory could possibly match prefix: */
 322        PREFIX_EXCLUDES_DIR
 323};
 324
 325/*
 326 * Return a `prefix_state` constant describing the relationship
 327 * between the directory with the specified `dirname` and `prefix`.
 328 */
 329static enum prefix_state overlaps_prefix(const char *dirname,
 330                                         const char *prefix)
 331{
 332        while (*prefix && *dirname == *prefix) {
 333                dirname++;
 334                prefix++;
 335        }
 336        if (!*prefix)
 337                return PREFIX_CONTAINS_DIR;
 338        else if (!*dirname)
 339                return PREFIX_WITHIN_DIR;
 340        else
 341                return PREFIX_EXCLUDES_DIR;
 342}
 343
 344/*
 345 * Load all of the refs from `dir` (recursively) that could possibly
 346 * contain references matching `prefix` into our in-memory cache. If
 347 * `prefix` is NULL, prime unconditionally.
 348 */
 349static void prime_ref_dir(struct ref_dir *dir, const char *prefix)
 350{
 351        /*
 352         * The hard work of loading loose refs is done by get_ref_dir(), so we
 353         * just need to recurse through all of the sub-directories. We do not
 354         * even need to care about sorting, as traversal order does not matter
 355         * to us.
 356         */
 357        int i;
 358        for (i = 0; i < dir->nr; i++) {
 359                struct ref_entry *entry = dir->entries[i];
 360                if (!(entry->flag & REF_DIR)) {
 361                        /* Not a directory; no need to recurse. */
 362                } else if (!prefix) {
 363                        /* Recurse in any case: */
 364                        prime_ref_dir(get_ref_dir(entry), NULL);
 365                } else {
 366                        switch (overlaps_prefix(entry->name, prefix)) {
 367                        case PREFIX_CONTAINS_DIR:
 368                                /*
 369                                 * Recurse, and from here down we
 370                                 * don't have to check the prefix
 371                                 * anymore:
 372                                 */
 373                                prime_ref_dir(get_ref_dir(entry), NULL);
 374                                break;
 375                        case PREFIX_WITHIN_DIR:
 376                                prime_ref_dir(get_ref_dir(entry), prefix);
 377                                break;
 378                        case PREFIX_EXCLUDES_DIR:
 379                                /* No need to prime this directory. */
 380                                break;
 381                        }
 382                }
 383        }
 384}
 385
 386/*
 387 * A level in the reference hierarchy that is currently being iterated
 388 * through.
 389 */
 390struct cache_ref_iterator_level {
 391        /*
 392         * The ref_dir being iterated over at this level. The ref_dir
 393         * is sorted before being stored here.
 394         */
 395        struct ref_dir *dir;
 396
 397        enum prefix_state prefix_state;
 398
 399        /*
 400         * The index of the current entry within dir (which might
 401         * itself be a directory). If index == -1, then the iteration
 402         * hasn't yet begun. If index == dir->nr, then the iteration
 403         * through this level is over.
 404         */
 405        int index;
 406};
 407
 408/*
 409 * Represent an iteration through a ref_dir in the memory cache. The
 410 * iteration recurses through subdirectories.
 411 */
 412struct cache_ref_iterator {
 413        struct ref_iterator base;
 414
 415        /*
 416         * The number of levels currently on the stack. This is always
 417         * at least 1, because when it becomes zero the iteration is
 418         * ended and this struct is freed.
 419         */
 420        size_t levels_nr;
 421
 422        /* The number of levels that have been allocated on the stack */
 423        size_t levels_alloc;
 424
 425        /*
 426         * Only include references with this prefix in the iteration.
 427         * The prefix is matched textually, without regard for path
 428         * component boundaries.
 429         */
 430        const char *prefix;
 431
 432        /*
 433         * A stack of levels. levels[0] is the uppermost level that is
 434         * being iterated over in this iteration. (This is not
 435         * necessary the top level in the references hierarchy. If we
 436         * are iterating through a subtree, then levels[0] will hold
 437         * the ref_dir for that subtree, and subsequent levels will go
 438         * on from there.)
 439         */
 440        struct cache_ref_iterator_level *levels;
 441};
 442
 443static int cache_ref_iterator_advance(struct ref_iterator *ref_iterator)
 444{
 445        struct cache_ref_iterator *iter =
 446                (struct cache_ref_iterator *)ref_iterator;
 447
 448        while (1) {
 449                struct cache_ref_iterator_level *level =
 450                        &iter->levels[iter->levels_nr - 1];
 451                struct ref_dir *dir = level->dir;
 452                struct ref_entry *entry;
 453                enum prefix_state entry_prefix_state;
 454
 455                if (level->index == -1)
 456                        sort_ref_dir(dir);
 457
 458                if (++level->index == level->dir->nr) {
 459                        /* This level is exhausted; pop up a level */
 460                        if (--iter->levels_nr == 0)
 461                                return ref_iterator_abort(ref_iterator);
 462
 463                        continue;
 464                }
 465
 466                entry = dir->entries[level->index];
 467
 468                if (level->prefix_state == PREFIX_WITHIN_DIR) {
 469                        entry_prefix_state = overlaps_prefix(entry->name, iter->prefix);
 470                        if (entry_prefix_state == PREFIX_EXCLUDES_DIR)
 471                                continue;
 472                } else {
 473                        entry_prefix_state = level->prefix_state;
 474                }
 475
 476                if (entry->flag & REF_DIR) {
 477                        /* push down a level */
 478                        ALLOC_GROW(iter->levels, iter->levels_nr + 1,
 479                                   iter->levels_alloc);
 480
 481                        level = &iter->levels[iter->levels_nr++];
 482                        level->dir = get_ref_dir(entry);
 483                        level->prefix_state = entry_prefix_state;
 484                        level->index = -1;
 485                } else {
 486                        iter->base.refname = entry->name;
 487                        iter->base.oid = &entry->u.value.oid;
 488                        iter->base.flags = entry->flag;
 489                        return ITER_OK;
 490                }
 491        }
 492}
 493
 494enum peel_status peel_entry(struct ref_entry *entry, int repeel)
 495{
 496        enum peel_status status;
 497
 498        if (entry->flag & REF_KNOWS_PEELED) {
 499                if (repeel) {
 500                        entry->flag &= ~REF_KNOWS_PEELED;
 501                        oidclr(&entry->u.value.peeled);
 502                } else {
 503                        return is_null_oid(&entry->u.value.peeled) ?
 504                                PEEL_NON_TAG : PEEL_PEELED;
 505                }
 506        }
 507        if (entry->flag & REF_ISBROKEN)
 508                return PEEL_BROKEN;
 509        if (entry->flag & REF_ISSYMREF)
 510                return PEEL_IS_SYMREF;
 511
 512        status = peel_object(entry->u.value.oid.hash, entry->u.value.peeled.hash);
 513        if (status == PEEL_PEELED || status == PEEL_NON_TAG)
 514                entry->flag |= REF_KNOWS_PEELED;
 515        return status;
 516}
 517
 518static int cache_ref_iterator_peel(struct ref_iterator *ref_iterator,
 519                                   struct object_id *peeled)
 520{
 521        struct cache_ref_iterator *iter =
 522                (struct cache_ref_iterator *)ref_iterator;
 523        struct cache_ref_iterator_level *level;
 524        struct ref_entry *entry;
 525
 526        level = &iter->levels[iter->levels_nr - 1];
 527
 528        if (level->index == -1)
 529                die("BUG: peel called before advance for cache iterator");
 530
 531        entry = level->dir->entries[level->index];
 532
 533        if (peel_entry(entry, 0))
 534                return -1;
 535        oidcpy(peeled, &entry->u.value.peeled);
 536        return 0;
 537}
 538
 539static int cache_ref_iterator_abort(struct ref_iterator *ref_iterator)
 540{
 541        struct cache_ref_iterator *iter =
 542                (struct cache_ref_iterator *)ref_iterator;
 543
 544        free((char *)iter->prefix);
 545        free(iter->levels);
 546        base_ref_iterator_free(ref_iterator);
 547        return ITER_DONE;
 548}
 549
 550static struct ref_iterator_vtable cache_ref_iterator_vtable = {
 551        cache_ref_iterator_advance,
 552        cache_ref_iterator_peel,
 553        cache_ref_iterator_abort
 554};
 555
 556struct ref_iterator *cache_ref_iterator_begin(struct ref_cache *cache,
 557                                              const char *prefix,
 558                                              int prime_dir)
 559{
 560        struct ref_dir *dir;
 561        struct cache_ref_iterator *iter;
 562        struct ref_iterator *ref_iterator;
 563        struct cache_ref_iterator_level *level;
 564
 565        dir = get_ref_dir(cache->root);
 566        if (prefix && *prefix)
 567                dir = find_containing_dir(dir, prefix, 0);
 568        if (!dir)
 569                /* There's nothing to iterate over. */
 570                return empty_ref_iterator_begin();
 571
 572        if (prime_dir)
 573                prime_ref_dir(dir, prefix);
 574
 575        iter = xcalloc(1, sizeof(*iter));
 576        ref_iterator = &iter->base;
 577        base_ref_iterator_init(ref_iterator, &cache_ref_iterator_vtable);
 578        ALLOC_GROW(iter->levels, 10, iter->levels_alloc);
 579
 580        iter->levels_nr = 1;
 581        level = &iter->levels[0];
 582        level->index = -1;
 583        level->dir = dir;
 584
 585        if (prefix && *prefix) {
 586                iter->prefix = xstrdup(prefix);
 587                level->prefix_state = PREFIX_WITHIN_DIR;
 588        } else {
 589                level->prefix_state = PREFIX_CONTAINS_DIR;
 590        }
 591
 592        return ref_iterator;
 593}