894027c0cd7e6a3983c55c1564e67dc2957f392e
   1/*
   2 * This handles recursive filename detection with exclude
   3 * files, index knowledge etc..
   4 *
   5 * See Documentation/technical/api-directory-listing.txt
   6 *
   7 * Copyright (C) Linus Torvalds, 2005-2006
   8 *               Junio Hamano, 2005-2006
   9 */
  10#include "cache.h"
  11#include "dir.h"
  12#include "refs.h"
  13#include "wildmatch.h"
  14#include "pathspec.h"
  15
  16struct path_simplify {
  17        int len;
  18        const char *path;
  19};
  20
  21/*
  22 * Tells read_directory_recursive how a file or directory should be treated.
  23 * Values are ordered by significance, e.g. if a directory contains both
  24 * excluded and untracked files, it is listed as untracked because
  25 * path_untracked > path_excluded.
  26 */
  27enum path_treatment {
  28        path_none = 0,
  29        path_recurse,
  30        path_excluded,
  31        path_untracked
  32};
  33
  34static enum path_treatment read_directory_recursive(struct dir_struct *dir,
  35        const char *path, int len,
  36        int check_only, const struct path_simplify *simplify);
  37static int get_dtype(struct dirent *de, const char *path, int len);
  38
  39/* helper string functions with support for the ignore_case flag */
  40int strcmp_icase(const char *a, const char *b)
  41{
  42        return ignore_case ? strcasecmp(a, b) : strcmp(a, b);
  43}
  44
  45int strncmp_icase(const char *a, const char *b, size_t count)
  46{
  47        return ignore_case ? strncasecmp(a, b, count) : strncmp(a, b, count);
  48}
  49
  50int fnmatch_icase(const char *pattern, const char *string, int flags)
  51{
  52        return wildmatch(pattern, string,
  53                         flags | (ignore_case ? WM_CASEFOLD : 0),
  54                         NULL);
  55}
  56
  57int git_fnmatch(const struct pathspec_item *item,
  58                const char *pattern, const char *string,
  59                int prefix)
  60{
  61        if (prefix > 0) {
  62                if (ps_strncmp(item, pattern, string, prefix))
  63                        return WM_NOMATCH;
  64                pattern += prefix;
  65                string += prefix;
  66        }
  67        if (item->flags & PATHSPEC_ONESTAR) {
  68                int pattern_len = strlen(++pattern);
  69                int string_len = strlen(string);
  70                return string_len < pattern_len ||
  71                        ps_strcmp(item, pattern,
  72                                  string + string_len - pattern_len);
  73        }
  74        if (item->magic & PATHSPEC_GLOB)
  75                return wildmatch(pattern, string,
  76                                 WM_PATHNAME |
  77                                 (item->magic & PATHSPEC_ICASE ? WM_CASEFOLD : 0),
  78                                 NULL);
  79        else
  80                /* wildmatch has not learned no FNM_PATHNAME mode yet */
  81                return wildmatch(pattern, string,
  82                                 item->magic & PATHSPEC_ICASE ? WM_CASEFOLD : 0,
  83                                 NULL);
  84}
  85
  86static int fnmatch_icase_mem(const char *pattern, int patternlen,
  87                             const char *string, int stringlen,
  88                             int flags)
  89{
  90        int match_status;
  91        struct strbuf pat_buf = STRBUF_INIT;
  92        struct strbuf str_buf = STRBUF_INIT;
  93        const char *use_pat = pattern;
  94        const char *use_str = string;
  95
  96        if (pattern[patternlen]) {
  97                strbuf_add(&pat_buf, pattern, patternlen);
  98                use_pat = pat_buf.buf;
  99        }
 100        if (string[stringlen]) {
 101                strbuf_add(&str_buf, string, stringlen);
 102                use_str = str_buf.buf;
 103        }
 104
 105        if (ignore_case)
 106                flags |= WM_CASEFOLD;
 107        match_status = wildmatch(use_pat, use_str, flags, NULL);
 108
 109        strbuf_release(&pat_buf);
 110        strbuf_release(&str_buf);
 111
 112        return match_status;
 113}
 114
 115static size_t common_prefix_len(const struct pathspec *pathspec)
 116{
 117        int n;
 118        size_t max = 0;
 119
 120        /*
 121         * ":(icase)path" is treated as a pathspec full of
 122         * wildcard. In other words, only prefix is considered common
 123         * prefix. If the pathspec is abc/foo abc/bar, running in
 124         * subdir xyz, the common prefix is still xyz, not xuz/abc as
 125         * in non-:(icase).
 126         */
 127        GUARD_PATHSPEC(pathspec,
 128                       PATHSPEC_FROMTOP |
 129                       PATHSPEC_MAXDEPTH |
 130                       PATHSPEC_LITERAL |
 131                       PATHSPEC_GLOB |
 132                       PATHSPEC_ICASE |
 133                       PATHSPEC_EXCLUDE);
 134
 135        for (n = 0; n < pathspec->nr; n++) {
 136                size_t i = 0, len = 0, item_len;
 137                if (pathspec->items[n].magic & PATHSPEC_EXCLUDE)
 138                        continue;
 139                if (pathspec->items[n].magic & PATHSPEC_ICASE)
 140                        item_len = pathspec->items[n].prefix;
 141                else
 142                        item_len = pathspec->items[n].nowildcard_len;
 143                while (i < item_len && (n == 0 || i < max)) {
 144                        char c = pathspec->items[n].match[i];
 145                        if (c != pathspec->items[0].match[i])
 146                                break;
 147                        if (c == '/')
 148                                len = i + 1;
 149                        i++;
 150                }
 151                if (n == 0 || len < max) {
 152                        max = len;
 153                        if (!max)
 154                                break;
 155                }
 156        }
 157        return max;
 158}
 159
 160/*
 161 * Returns a copy of the longest leading path common among all
 162 * pathspecs.
 163 */
 164char *common_prefix(const struct pathspec *pathspec)
 165{
 166        unsigned long len = common_prefix_len(pathspec);
 167
 168        return len ? xmemdupz(pathspec->items[0].match, len) : NULL;
 169}
 170
 171int fill_directory(struct dir_struct *dir, const struct pathspec *pathspec)
 172{
 173        size_t len;
 174
 175        /*
 176         * Calculate common prefix for the pathspec, and
 177         * use that to optimize the directory walk
 178         */
 179        len = common_prefix_len(pathspec);
 180
 181        /* Read the directory and prune it */
 182        read_directory(dir, pathspec->nr ? pathspec->_raw[0] : "", len, pathspec);
 183        return len;
 184}
 185
 186int within_depth(const char *name, int namelen,
 187                        int depth, int max_depth)
 188{
 189        const char *cp = name, *cpe = name + namelen;
 190
 191        while (cp < cpe) {
 192                if (*cp++ != '/')
 193                        continue;
 194                depth++;
 195                if (depth > max_depth)
 196                        return 0;
 197        }
 198        return 1;
 199}
 200
 201#define DO_MATCH_EXCLUDE   1
 202#define DO_MATCH_DIRECTORY 2
 203
 204/*
 205 * Does 'match' match the given name?
 206 * A match is found if
 207 *
 208 * (1) the 'match' string is leading directory of 'name', or
 209 * (2) the 'match' string is a wildcard and matches 'name', or
 210 * (3) the 'match' string is exactly the same as 'name'.
 211 *
 212 * and the return value tells which case it was.
 213 *
 214 * It returns 0 when there is no match.
 215 */
 216static int match_pathspec_item(const struct pathspec_item *item, int prefix,
 217                               const char *name, int namelen, unsigned flags)
 218{
 219        /* name/namelen has prefix cut off by caller */
 220        const char *match = item->match + prefix;
 221        int matchlen = item->len - prefix;
 222
 223        /*
 224         * The normal call pattern is:
 225         * 1. prefix = common_prefix_len(ps);
 226         * 2. prune something, or fill_directory
 227         * 3. match_pathspec()
 228         *
 229         * 'prefix' at #1 may be shorter than the command's prefix and
 230         * it's ok for #2 to match extra files. Those extras will be
 231         * trimmed at #3.
 232         *
 233         * Suppose the pathspec is 'foo' and '../bar' running from
 234         * subdir 'xyz'. The common prefix at #1 will be empty, thanks
 235         * to "../". We may have xyz/foo _and_ XYZ/foo after #2. The
 236         * user does not want XYZ/foo, only the "foo" part should be
 237         * case-insensitive. We need to filter out XYZ/foo here. In
 238         * other words, we do not trust the caller on comparing the
 239         * prefix part when :(icase) is involved. We do exact
 240         * comparison ourselves.
 241         *
 242         * Normally the caller (common_prefix_len() in fact) does
 243         * _exact_ matching on name[-prefix+1..-1] and we do not need
 244         * to check that part. Be defensive and check it anyway, in
 245         * case common_prefix_len is changed, or a new caller is
 246         * introduced that does not use common_prefix_len.
 247         *
 248         * If the penalty turns out too high when prefix is really
 249         * long, maybe change it to
 250         * strncmp(match, name, item->prefix - prefix)
 251         */
 252        if (item->prefix && (item->magic & PATHSPEC_ICASE) &&
 253            strncmp(item->match, name - prefix, item->prefix))
 254                return 0;
 255
 256        /* If the match was just the prefix, we matched */
 257        if (!*match)
 258                return MATCHED_RECURSIVELY;
 259
 260        if (matchlen <= namelen && !ps_strncmp(item, match, name, matchlen)) {
 261                if (matchlen == namelen)
 262                        return MATCHED_EXACTLY;
 263
 264                if (match[matchlen-1] == '/' || name[matchlen] == '/')
 265                        return MATCHED_RECURSIVELY;
 266        } else if ((flags & DO_MATCH_DIRECTORY) &&
 267                   match[matchlen - 1] == '/' &&
 268                   namelen == matchlen - 1 &&
 269                   !ps_strncmp(item, match, name, namelen))
 270                return MATCHED_EXACTLY;
 271
 272        if (item->nowildcard_len < item->len &&
 273            !git_fnmatch(item, match, name,
 274                         item->nowildcard_len - prefix))
 275                return MATCHED_FNMATCH;
 276
 277        return 0;
 278}
 279
 280/*
 281 * Given a name and a list of pathspecs, returns the nature of the
 282 * closest (i.e. most specific) match of the name to any of the
 283 * pathspecs.
 284 *
 285 * The caller typically calls this multiple times with the same
 286 * pathspec and seen[] array but with different name/namelen
 287 * (e.g. entries from the index) and is interested in seeing if and
 288 * how each pathspec matches all the names it calls this function
 289 * with.  A mark is left in the seen[] array for each pathspec element
 290 * indicating the closest type of match that element achieved, so if
 291 * seen[n] remains zero after multiple invocations, that means the nth
 292 * pathspec did not match any names, which could indicate that the
 293 * user mistyped the nth pathspec.
 294 */
 295static int do_match_pathspec(const struct pathspec *ps,
 296                             const char *name, int namelen,
 297                             int prefix, char *seen,
 298                             unsigned flags)
 299{
 300        int i, retval = 0, exclude = flags & DO_MATCH_EXCLUDE;
 301
 302        GUARD_PATHSPEC(ps,
 303                       PATHSPEC_FROMTOP |
 304                       PATHSPEC_MAXDEPTH |
 305                       PATHSPEC_LITERAL |
 306                       PATHSPEC_GLOB |
 307                       PATHSPEC_ICASE |
 308                       PATHSPEC_EXCLUDE);
 309
 310        if (!ps->nr) {
 311                if (!ps->recursive ||
 312                    !(ps->magic & PATHSPEC_MAXDEPTH) ||
 313                    ps->max_depth == -1)
 314                        return MATCHED_RECURSIVELY;
 315
 316                if (within_depth(name, namelen, 0, ps->max_depth))
 317                        return MATCHED_EXACTLY;
 318                else
 319                        return 0;
 320        }
 321
 322        name += prefix;
 323        namelen -= prefix;
 324
 325        for (i = ps->nr - 1; i >= 0; i--) {
 326                int how;
 327
 328                if ((!exclude &&   ps->items[i].magic & PATHSPEC_EXCLUDE) ||
 329                    ( exclude && !(ps->items[i].magic & PATHSPEC_EXCLUDE)))
 330                        continue;
 331
 332                if (seen && seen[i] == MATCHED_EXACTLY)
 333                        continue;
 334                /*
 335                 * Make exclude patterns optional and never report
 336                 * "pathspec ':(exclude)foo' matches no files"
 337                 */
 338                if (seen && ps->items[i].magic & PATHSPEC_EXCLUDE)
 339                        seen[i] = MATCHED_FNMATCH;
 340                how = match_pathspec_item(ps->items+i, prefix, name,
 341                                          namelen, flags);
 342                if (ps->recursive &&
 343                    (ps->magic & PATHSPEC_MAXDEPTH) &&
 344                    ps->max_depth != -1 &&
 345                    how && how != MATCHED_FNMATCH) {
 346                        int len = ps->items[i].len;
 347                        if (name[len] == '/')
 348                                len++;
 349                        if (within_depth(name+len, namelen-len, 0, ps->max_depth))
 350                                how = MATCHED_EXACTLY;
 351                        else
 352                                how = 0;
 353                }
 354                if (how) {
 355                        if (retval < how)
 356                                retval = how;
 357                        if (seen && seen[i] < how)
 358                                seen[i] = how;
 359                }
 360        }
 361        return retval;
 362}
 363
 364int match_pathspec(const struct pathspec *ps,
 365                   const char *name, int namelen,
 366                   int prefix, char *seen, int is_dir)
 367{
 368        int positive, negative;
 369        unsigned flags = is_dir ? DO_MATCH_DIRECTORY : 0;
 370        positive = do_match_pathspec(ps, name, namelen,
 371                                     prefix, seen, flags);
 372        if (!(ps->magic & PATHSPEC_EXCLUDE) || !positive)
 373                return positive;
 374        negative = do_match_pathspec(ps, name, namelen,
 375                                     prefix, seen,
 376                                     flags | DO_MATCH_EXCLUDE);
 377        return negative ? 0 : positive;
 378}
 379
 380int report_path_error(const char *ps_matched,
 381                      const struct pathspec *pathspec,
 382                      const char *prefix)
 383{
 384        /*
 385         * Make sure all pathspec matched; otherwise it is an error.
 386         */
 387        struct strbuf sb = STRBUF_INIT;
 388        int num, errors = 0;
 389        for (num = 0; num < pathspec->nr; num++) {
 390                int other, found_dup;
 391
 392                if (ps_matched[num])
 393                        continue;
 394                /*
 395                 * The caller might have fed identical pathspec
 396                 * twice.  Do not barf on such a mistake.
 397                 * FIXME: parse_pathspec should have eliminated
 398                 * duplicate pathspec.
 399                 */
 400                for (found_dup = other = 0;
 401                     !found_dup && other < pathspec->nr;
 402                     other++) {
 403                        if (other == num || !ps_matched[other])
 404                                continue;
 405                        if (!strcmp(pathspec->items[other].original,
 406                                    pathspec->items[num].original))
 407                                /*
 408                                 * Ok, we have a match already.
 409                                 */
 410                                found_dup = 1;
 411                }
 412                if (found_dup)
 413                        continue;
 414
 415                error("pathspec '%s' did not match any file(s) known to git.",
 416                      pathspec->items[num].original);
 417                errors++;
 418        }
 419        strbuf_release(&sb);
 420        return errors;
 421}
 422
 423/*
 424 * Return the length of the "simple" part of a path match limiter.
 425 */
 426int simple_length(const char *match)
 427{
 428        int len = -1;
 429
 430        for (;;) {
 431                unsigned char c = *match++;
 432                len++;
 433                if (c == '\0' || is_glob_special(c))
 434                        return len;
 435        }
 436}
 437
 438int no_wildcard(const char *string)
 439{
 440        return string[simple_length(string)] == '\0';
 441}
 442
 443void parse_exclude_pattern(const char **pattern,
 444                           int *patternlen,
 445                           int *flags,
 446                           int *nowildcardlen)
 447{
 448        const char *p = *pattern;
 449        size_t i, len;
 450
 451        *flags = 0;
 452        if (*p == '!') {
 453                *flags |= EXC_FLAG_NEGATIVE;
 454                p++;
 455        }
 456        len = strlen(p);
 457        if (len && p[len - 1] == '/') {
 458                len--;
 459                *flags |= EXC_FLAG_MUSTBEDIR;
 460        }
 461        for (i = 0; i < len; i++) {
 462                if (p[i] == '/')
 463                        break;
 464        }
 465        if (i == len)
 466                *flags |= EXC_FLAG_NODIR;
 467        *nowildcardlen = simple_length(p);
 468        /*
 469         * we should have excluded the trailing slash from 'p' too,
 470         * but that's one more allocation. Instead just make sure
 471         * nowildcardlen does not exceed real patternlen
 472         */
 473        if (*nowildcardlen > len)
 474                *nowildcardlen = len;
 475        if (*p == '*' && no_wildcard(p + 1))
 476                *flags |= EXC_FLAG_ENDSWITH;
 477        *pattern = p;
 478        *patternlen = len;
 479}
 480
 481void add_exclude(const char *string, const char *base,
 482                 int baselen, struct exclude_list *el, int srcpos)
 483{
 484        struct exclude *x;
 485        int patternlen;
 486        int flags;
 487        int nowildcardlen;
 488
 489        parse_exclude_pattern(&string, &patternlen, &flags, &nowildcardlen);
 490        if (flags & EXC_FLAG_MUSTBEDIR) {
 491                char *s;
 492                x = xmalloc(sizeof(*x) + patternlen + 1);
 493                s = (char *)(x+1);
 494                memcpy(s, string, patternlen);
 495                s[patternlen] = '\0';
 496                x->pattern = s;
 497        } else {
 498                x = xmalloc(sizeof(*x));
 499                x->pattern = string;
 500        }
 501        x->patternlen = patternlen;
 502        x->nowildcardlen = nowildcardlen;
 503        x->base = base;
 504        x->baselen = baselen;
 505        x->flags = flags;
 506        x->srcpos = srcpos;
 507        ALLOC_GROW(el->excludes, el->nr + 1, el->alloc);
 508        el->excludes[el->nr++] = x;
 509        x->el = el;
 510}
 511
 512static void *read_skip_worktree_file_from_index(const char *path, size_t *size)
 513{
 514        int pos, len;
 515        unsigned long sz;
 516        enum object_type type;
 517        void *data;
 518
 519        len = strlen(path);
 520        pos = cache_name_pos(path, len);
 521        if (pos < 0)
 522                return NULL;
 523        if (!ce_skip_worktree(active_cache[pos]))
 524                return NULL;
 525        data = read_sha1_file(active_cache[pos]->sha1, &type, &sz);
 526        if (!data || type != OBJ_BLOB) {
 527                free(data);
 528                return NULL;
 529        }
 530        *size = xsize_t(sz);
 531        return data;
 532}
 533
 534/*
 535 * Frees memory within el which was allocated for exclude patterns and
 536 * the file buffer.  Does not free el itself.
 537 */
 538void clear_exclude_list(struct exclude_list *el)
 539{
 540        int i;
 541
 542        for (i = 0; i < el->nr; i++)
 543                free(el->excludes[i]);
 544        free(el->excludes);
 545        free(el->filebuf);
 546
 547        el->nr = 0;
 548        el->excludes = NULL;
 549        el->filebuf = NULL;
 550}
 551
 552static void trim_trailing_spaces(char *buf)
 553{
 554        char *p, *last_space = NULL;
 555
 556        for (p = buf; *p; p++)
 557                switch (*p) {
 558                case ' ':
 559                        if (!last_space)
 560                                last_space = p;
 561                        break;
 562                case '\\':
 563                        p++;
 564                        if (!*p)
 565                                return;
 566                        /* fallthrough */
 567                default:
 568                        last_space = NULL;
 569                }
 570
 571        if (last_space)
 572                *last_space = '\0';
 573}
 574
 575int add_excludes_from_file_to_list(const char *fname,
 576                                   const char *base,
 577                                   int baselen,
 578                                   struct exclude_list *el,
 579                                   int check_index)
 580{
 581        struct stat st;
 582        int fd, i, lineno = 1;
 583        size_t size = 0;
 584        char *buf, *entry;
 585
 586        fd = open(fname, O_RDONLY);
 587        if (fd < 0 || fstat(fd, &st) < 0) {
 588                if (errno != ENOENT)
 589                        warn_on_inaccessible(fname);
 590                if (0 <= fd)
 591                        close(fd);
 592                if (!check_index ||
 593                    (buf = read_skip_worktree_file_from_index(fname, &size)) == NULL)
 594                        return -1;
 595                if (size == 0) {
 596                        free(buf);
 597                        return 0;
 598                }
 599                if (buf[size-1] != '\n') {
 600                        buf = xrealloc(buf, size+1);
 601                        buf[size++] = '\n';
 602                }
 603        } else {
 604                size = xsize_t(st.st_size);
 605                if (size == 0) {
 606                        close(fd);
 607                        return 0;
 608                }
 609                buf = xmalloc(size+1);
 610                if (read_in_full(fd, buf, size) != size) {
 611                        free(buf);
 612                        close(fd);
 613                        return -1;
 614                }
 615                buf[size++] = '\n';
 616                close(fd);
 617        }
 618
 619        el->filebuf = buf;
 620        entry = buf;
 621        for (i = 0; i < size; i++) {
 622                if (buf[i] == '\n') {
 623                        if (entry != buf + i && entry[0] != '#') {
 624                                buf[i - (i && buf[i-1] == '\r')] = 0;
 625                                trim_trailing_spaces(entry);
 626                                add_exclude(entry, base, baselen, el, lineno);
 627                        }
 628                        lineno++;
 629                        entry = buf + i + 1;
 630                }
 631        }
 632        return 0;
 633}
 634
 635struct exclude_list *add_exclude_list(struct dir_struct *dir,
 636                                      int group_type, const char *src)
 637{
 638        struct exclude_list *el;
 639        struct exclude_list_group *group;
 640
 641        group = &dir->exclude_list_group[group_type];
 642        ALLOC_GROW(group->el, group->nr + 1, group->alloc);
 643        el = &group->el[group->nr++];
 644        memset(el, 0, sizeof(*el));
 645        el->src = src;
 646        return el;
 647}
 648
 649/*
 650 * Used to set up core.excludesfile and .git/info/exclude lists.
 651 */
 652void add_excludes_from_file(struct dir_struct *dir, const char *fname)
 653{
 654        struct exclude_list *el;
 655        el = add_exclude_list(dir, EXC_FILE, fname);
 656        if (add_excludes_from_file_to_list(fname, "", 0, el, 0) < 0)
 657                die("cannot use %s as an exclude file", fname);
 658}
 659
 660int match_basename(const char *basename, int basenamelen,
 661                   const char *pattern, int prefix, int patternlen,
 662                   int flags)
 663{
 664        if (prefix == patternlen) {
 665                if (patternlen == basenamelen &&
 666                    !strncmp_icase(pattern, basename, basenamelen))
 667                        return 1;
 668        } else if (flags & EXC_FLAG_ENDSWITH) {
 669                /* "*literal" matching against "fooliteral" */
 670                if (patternlen - 1 <= basenamelen &&
 671                    !strncmp_icase(pattern + 1,
 672                                   basename + basenamelen - (patternlen - 1),
 673                                   patternlen - 1))
 674                        return 1;
 675        } else {
 676                if (fnmatch_icase_mem(pattern, patternlen,
 677                                      basename, basenamelen,
 678                                      0) == 0)
 679                        return 1;
 680        }
 681        return 0;
 682}
 683
 684int match_pathname(const char *pathname, int pathlen,
 685                   const char *base, int baselen,
 686                   const char *pattern, int prefix, int patternlen,
 687                   int flags)
 688{
 689        const char *name;
 690        int namelen;
 691
 692        /*
 693         * match with FNM_PATHNAME; the pattern has base implicitly
 694         * in front of it.
 695         */
 696        if (*pattern == '/') {
 697                pattern++;
 698                patternlen--;
 699                prefix--;
 700        }
 701
 702        /*
 703         * baselen does not count the trailing slash. base[] may or
 704         * may not end with a trailing slash though.
 705         */
 706        if (pathlen < baselen + 1 ||
 707            (baselen && pathname[baselen] != '/') ||
 708            strncmp_icase(pathname, base, baselen))
 709                return 0;
 710
 711        namelen = baselen ? pathlen - baselen - 1 : pathlen;
 712        name = pathname + pathlen - namelen;
 713
 714        if (prefix) {
 715                /*
 716                 * if the non-wildcard part is longer than the
 717                 * remaining pathname, surely it cannot match.
 718                 */
 719                if (prefix > namelen)
 720                        return 0;
 721
 722                if (strncmp_icase(pattern, name, prefix))
 723                        return 0;
 724                pattern += prefix;
 725                patternlen -= prefix;
 726                name    += prefix;
 727                namelen -= prefix;
 728
 729                /*
 730                 * If the whole pattern did not have a wildcard,
 731                 * then our prefix match is all we need; we
 732                 * do not need to call fnmatch at all.
 733                 */
 734                if (!patternlen && !namelen)
 735                        return 1;
 736                /*
 737                 * This can happen when we ignore some exclude rules
 738                 * on directories in other to see if negative rules
 739                 * may match. E.g.
 740                 *
 741                 * /abc
 742                 * !/abc/def/ghi
 743                 *
 744                 * The pattern of interest is "/abc". On the first
 745                 * try, we should match path "abc" with this pattern
 746                 * in the "if" statement right above, but the caller
 747                 * ignores it.
 748                 *
 749                 * On the second try with paths within "abc",
 750                 * e.g. "abc/xyz", we come here and try to match it
 751                 * with "/abc".
 752                 */
 753                if (!patternlen && namelen && *name == '/')
 754                        return 1;
 755        }
 756
 757        return fnmatch_icase_mem(pattern, patternlen,
 758                                 name, namelen,
 759                                 WM_PATHNAME) == 0;
 760}
 761
 762/*
 763 * Return non-zero if pathname is a directory and an ancestor of the
 764 * literal path in a (negative) pattern. This is used to keep
 765 * descending in "foo" and "foo/bar" when the pattern is
 766 * "!foo/bar/.gitignore". "foo/notbar" will not be descended however.
 767 */
 768static int match_neg_path(const char *pathname, int pathlen, int *dtype,
 769                          const char *base, int baselen,
 770                          const char *pattern, int prefix, int patternlen,
 771                          int flags)
 772{
 773        assert((flags & EXC_FLAG_NEGATIVE) && !(flags & EXC_FLAG_NODIR));
 774
 775        if (*dtype == DT_UNKNOWN)
 776                *dtype = get_dtype(NULL, pathname, pathlen);
 777        if (*dtype != DT_DIR)
 778                return 0;
 779
 780        if (*pattern == '/') {
 781                pattern++;
 782                patternlen--;
 783                prefix--;
 784        }
 785
 786        if (baselen) {
 787                if (((pathlen < baselen && base[pathlen] == '/') ||
 788                     pathlen == baselen) &&
 789                    !strncmp_icase(pathname, base, pathlen))
 790                        return 1;
 791                pathname += baselen + 1;
 792                pathlen  -= baselen + 1;
 793        }
 794
 795
 796        if (prefix &&
 797            ((pathlen < prefix && pattern[pathlen] == '/') &&
 798             !strncmp_icase(pathname, pattern, pathlen)))
 799                return 1;
 800
 801        return 0;
 802}
 803
 804/*
 805 * Scan the given exclude list in reverse to see whether pathname
 806 * should be ignored.  The first match (i.e. the last on the list), if
 807 * any, determines the fate.  Returns the exclude_list element which
 808 * matched, or NULL for undecided.
 809 */
 810static struct exclude *last_exclude_matching_from_list(const char *pathname,
 811                                                       int pathlen,
 812                                                       const char *basename,
 813                                                       int *dtype,
 814                                                       struct exclude_list *el)
 815{
 816        struct exclude *exc = NULL; /* undecided */
 817        int i, matched_negative_path = 0;
 818
 819        if (!el->nr)
 820                return NULL;    /* undefined */
 821
 822        for (i = el->nr - 1; 0 <= i; i--) {
 823                struct exclude *x = el->excludes[i];
 824                const char *exclude = x->pattern;
 825                int prefix = x->nowildcardlen;
 826
 827                if (x->flags & EXC_FLAG_MUSTBEDIR) {
 828                        if (*dtype == DT_UNKNOWN)
 829                                *dtype = get_dtype(NULL, pathname, pathlen);
 830                        if (*dtype != DT_DIR)
 831                                continue;
 832                }
 833
 834                if (x->flags & EXC_FLAG_NODIR) {
 835                        if (match_basename(basename,
 836                                           pathlen - (basename - pathname),
 837                                           exclude, prefix, x->patternlen,
 838                                           x->flags)) {
 839                                exc = x;
 840                                break;
 841                        }
 842                        continue;
 843                }
 844
 845                assert(x->baselen == 0 || x->base[x->baselen - 1] == '/');
 846                if (match_pathname(pathname, pathlen,
 847                                   x->base, x->baselen ? x->baselen - 1 : 0,
 848                                   exclude, prefix, x->patternlen, x->flags)) {
 849                        exc = x;
 850                        break;
 851                }
 852
 853                if ((x->flags & EXC_FLAG_NEGATIVE) && !matched_negative_path &&
 854                    match_neg_path(pathname, pathlen, dtype, x->base,
 855                                   x->baselen ? x->baselen - 1 : 0,
 856                                   exclude, prefix, x->patternlen, x->flags))
 857                        matched_negative_path = 1;
 858        }
 859        if (exc &&
 860            !(exc->flags & EXC_FLAG_NEGATIVE) &&
 861            !(exc->flags & EXC_FLAG_NODIR) &&
 862            matched_negative_path)
 863                exc = NULL;
 864        return exc;
 865}
 866
 867/*
 868 * Scan the list and let the last match determine the fate.
 869 * Return 1 for exclude, 0 for include and -1 for undecided.
 870 */
 871int is_excluded_from_list(const char *pathname,
 872                          int pathlen, const char *basename, int *dtype,
 873                          struct exclude_list *el)
 874{
 875        struct exclude *exclude;
 876        exclude = last_exclude_matching_from_list(pathname, pathlen, basename, dtype, el);
 877        if (exclude)
 878                return exclude->flags & EXC_FLAG_NEGATIVE ? 0 : 1;
 879        return -1; /* undecided */
 880}
 881
 882static struct exclude *last_exclude_matching_from_lists(struct dir_struct *dir,
 883                const char *pathname, int pathlen, const char *basename,
 884                int *dtype_p)
 885{
 886        int i, j;
 887        struct exclude_list_group *group;
 888        struct exclude *exclude;
 889        for (i = EXC_CMDL; i <= EXC_FILE; i++) {
 890                group = &dir->exclude_list_group[i];
 891                for (j = group->nr - 1; j >= 0; j--) {
 892                        exclude = last_exclude_matching_from_list(
 893                                pathname, pathlen, basename, dtype_p,
 894                                &group->el[j]);
 895                        if (exclude)
 896                                return exclude;
 897                }
 898        }
 899        return NULL;
 900}
 901
 902/*
 903 * Loads the per-directory exclude list for the substring of base
 904 * which has a char length of baselen.
 905 */
 906static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
 907{
 908        struct exclude_list_group *group;
 909        struct exclude_list *el;
 910        struct exclude_stack *stk = NULL;
 911        int current;
 912
 913        group = &dir->exclude_list_group[EXC_DIRS];
 914
 915        /*
 916         * Pop the exclude lists from the EXCL_DIRS exclude_list_group
 917         * which originate from directories not in the prefix of the
 918         * path being checked.
 919         */
 920        while ((stk = dir->exclude_stack) != NULL) {
 921                if (stk->baselen <= baselen &&
 922                    !strncmp(dir->basebuf.buf, base, stk->baselen))
 923                        break;
 924                el = &group->el[dir->exclude_stack->exclude_ix];
 925                dir->exclude_stack = stk->prev;
 926                dir->exclude = NULL;
 927                free((char *)el->src); /* see strbuf_detach() below */
 928                clear_exclude_list(el);
 929                free(stk);
 930                group->nr--;
 931        }
 932
 933        /* Skip traversing into sub directories if the parent is excluded */
 934        if (dir->exclude)
 935                return;
 936
 937        /*
 938         * Lazy initialization. All call sites currently just
 939         * memset(dir, 0, sizeof(*dir)) before use. Changing all of
 940         * them seems lots of work for little benefit.
 941         */
 942        if (!dir->basebuf.buf)
 943                strbuf_init(&dir->basebuf, PATH_MAX);
 944
 945        /* Read from the parent directories and push them down. */
 946        current = stk ? stk->baselen : -1;
 947        strbuf_setlen(&dir->basebuf, current < 0 ? 0 : current);
 948        while (current < baselen) {
 949                const char *cp;
 950
 951                stk = xcalloc(1, sizeof(*stk));
 952                if (current < 0) {
 953                        cp = base;
 954                        current = 0;
 955                } else {
 956                        cp = strchr(base + current + 1, '/');
 957                        if (!cp)
 958                                die("oops in prep_exclude");
 959                        cp++;
 960                }
 961                stk->prev = dir->exclude_stack;
 962                stk->baselen = cp - base;
 963                stk->exclude_ix = group->nr;
 964                el = add_exclude_list(dir, EXC_DIRS, NULL);
 965                strbuf_add(&dir->basebuf, base + current, stk->baselen - current);
 966                assert(stk->baselen == dir->basebuf.len);
 967
 968                /* Abort if the directory is excluded */
 969                if (stk->baselen) {
 970                        int dt = DT_DIR;
 971                        dir->basebuf.buf[stk->baselen - 1] = 0;
 972                        dir->exclude = last_exclude_matching_from_lists(dir,
 973                                dir->basebuf.buf, stk->baselen - 1,
 974                                dir->basebuf.buf + current, &dt);
 975                        dir->basebuf.buf[stk->baselen - 1] = '/';
 976                        if (dir->exclude &&
 977                            dir->exclude->flags & EXC_FLAG_NEGATIVE)
 978                                dir->exclude = NULL;
 979                        if (dir->exclude) {
 980                                dir->exclude_stack = stk;
 981                                return;
 982                        }
 983                }
 984
 985                /* Try to read per-directory file */
 986                if (dir->exclude_per_dir) {
 987                        /*
 988                         * dir->basebuf gets reused by the traversal, but we
 989                         * need fname to remain unchanged to ensure the src
 990                         * member of each struct exclude correctly
 991                         * back-references its source file.  Other invocations
 992                         * of add_exclude_list provide stable strings, so we
 993                         * strbuf_detach() and free() here in the caller.
 994                         */
 995                        struct strbuf sb = STRBUF_INIT;
 996                        strbuf_addbuf(&sb, &dir->basebuf);
 997                        strbuf_addstr(&sb, dir->exclude_per_dir);
 998                        el->src = strbuf_detach(&sb, NULL);
 999                        add_excludes_from_file_to_list(el->src, el->src,
1000                                                       stk->baselen, el, 1);
1001                }
1002                dir->exclude_stack = stk;
1003                current = stk->baselen;
1004        }
1005        strbuf_setlen(&dir->basebuf, baselen);
1006}
1007
1008/*
1009 * Loads the exclude lists for the directory containing pathname, then
1010 * scans all exclude lists to determine whether pathname is excluded.
1011 * Returns the exclude_list element which matched, or NULL for
1012 * undecided.
1013 */
1014struct exclude *last_exclude_matching(struct dir_struct *dir,
1015                                             const char *pathname,
1016                                             int *dtype_p)
1017{
1018        int pathlen = strlen(pathname);
1019        const char *basename = strrchr(pathname, '/');
1020        basename = (basename) ? basename+1 : pathname;
1021
1022        prep_exclude(dir, pathname, basename-pathname);
1023
1024        if (dir->exclude)
1025                return dir->exclude;
1026
1027        return last_exclude_matching_from_lists(dir, pathname, pathlen,
1028                        basename, dtype_p);
1029}
1030
1031/*
1032 * Loads the exclude lists for the directory containing pathname, then
1033 * scans all exclude lists to determine whether pathname is excluded.
1034 * Returns 1 if true, otherwise 0.
1035 */
1036int is_excluded(struct dir_struct *dir, const char *pathname, int *dtype_p)
1037{
1038        struct exclude *exclude =
1039                last_exclude_matching(dir, pathname, dtype_p);
1040        if (exclude)
1041                return exclude->flags & EXC_FLAG_NEGATIVE ? 0 : 1;
1042        return 0;
1043}
1044
1045static struct dir_entry *dir_entry_new(const char *pathname, int len)
1046{
1047        struct dir_entry *ent;
1048
1049        ent = xmalloc(sizeof(*ent) + len + 1);
1050        ent->len = len;
1051        memcpy(ent->name, pathname, len);
1052        ent->name[len] = 0;
1053        return ent;
1054}
1055
1056static struct dir_entry *dir_add_name(struct dir_struct *dir, const char *pathname, int len)
1057{
1058        if (cache_file_exists(pathname, len, ignore_case))
1059                return NULL;
1060
1061        ALLOC_GROW(dir->entries, dir->nr+1, dir->alloc);
1062        return dir->entries[dir->nr++] = dir_entry_new(pathname, len);
1063}
1064
1065struct dir_entry *dir_add_ignored(struct dir_struct *dir, const char *pathname, int len)
1066{
1067        if (!cache_name_is_other(pathname, len))
1068                return NULL;
1069
1070        ALLOC_GROW(dir->ignored, dir->ignored_nr+1, dir->ignored_alloc);
1071        return dir->ignored[dir->ignored_nr++] = dir_entry_new(pathname, len);
1072}
1073
1074enum exist_status {
1075        index_nonexistent = 0,
1076        index_directory,
1077        index_gitdir
1078};
1079
1080/*
1081 * Do not use the alphabetically sorted index to look up
1082 * the directory name; instead, use the case insensitive
1083 * directory hash.
1084 */
1085static enum exist_status directory_exists_in_index_icase(const char *dirname, int len)
1086{
1087        const struct cache_entry *ce = cache_dir_exists(dirname, len);
1088        unsigned char endchar;
1089
1090        if (!ce)
1091                return index_nonexistent;
1092        endchar = ce->name[len];
1093
1094        /*
1095         * The cache_entry structure returned will contain this dirname
1096         * and possibly additional path components.
1097         */
1098        if (endchar == '/')
1099                return index_directory;
1100
1101        /*
1102         * If there are no additional path components, then this cache_entry
1103         * represents a submodule.  Submodules, despite being directories,
1104         * are stored in the cache without a closing slash.
1105         */
1106        if (!endchar && S_ISGITLINK(ce->ce_mode))
1107                return index_gitdir;
1108
1109        /* This should never be hit, but it exists just in case. */
1110        return index_nonexistent;
1111}
1112
1113/*
1114 * The index sorts alphabetically by entry name, which
1115 * means that a gitlink sorts as '\0' at the end, while
1116 * a directory (which is defined not as an entry, but as
1117 * the files it contains) will sort with the '/' at the
1118 * end.
1119 */
1120static enum exist_status directory_exists_in_index(const char *dirname, int len)
1121{
1122        int pos;
1123
1124        if (ignore_case)
1125                return directory_exists_in_index_icase(dirname, len);
1126
1127        pos = cache_name_pos(dirname, len);
1128        if (pos < 0)
1129                pos = -pos-1;
1130        while (pos < active_nr) {
1131                const struct cache_entry *ce = active_cache[pos++];
1132                unsigned char endchar;
1133
1134                if (strncmp(ce->name, dirname, len))
1135                        break;
1136                endchar = ce->name[len];
1137                if (endchar > '/')
1138                        break;
1139                if (endchar == '/')
1140                        return index_directory;
1141                if (!endchar && S_ISGITLINK(ce->ce_mode))
1142                        return index_gitdir;
1143        }
1144        return index_nonexistent;
1145}
1146
1147/*
1148 * When we find a directory when traversing the filesystem, we
1149 * have three distinct cases:
1150 *
1151 *  - ignore it
1152 *  - see it as a directory
1153 *  - recurse into it
1154 *
1155 * and which one we choose depends on a combination of existing
1156 * git index contents and the flags passed into the directory
1157 * traversal routine.
1158 *
1159 * Case 1: If we *already* have entries in the index under that
1160 * directory name, we always recurse into the directory to see
1161 * all the files.
1162 *
1163 * Case 2: If we *already* have that directory name as a gitlink,
1164 * we always continue to see it as a gitlink, regardless of whether
1165 * there is an actual git directory there or not (it might not
1166 * be checked out as a subproject!)
1167 *
1168 * Case 3: if we didn't have it in the index previously, we
1169 * have a few sub-cases:
1170 *
1171 *  (a) if "show_other_directories" is true, we show it as
1172 *      just a directory, unless "hide_empty_directories" is
1173 *      also true, in which case we need to check if it contains any
1174 *      untracked and / or ignored files.
1175 *  (b) if it looks like a git directory, and we don't have
1176 *      'no_gitlinks' set we treat it as a gitlink, and show it
1177 *      as a directory.
1178 *  (c) otherwise, we recurse into it.
1179 */
1180static enum path_treatment treat_directory(struct dir_struct *dir,
1181        const char *dirname, int len, int exclude,
1182        const struct path_simplify *simplify)
1183{
1184        /* The "len-1" is to strip the final '/' */
1185        switch (directory_exists_in_index(dirname, len-1)) {
1186        case index_directory:
1187                return path_recurse;
1188
1189        case index_gitdir:
1190                return path_none;
1191
1192        case index_nonexistent:
1193                if (dir->flags & DIR_SHOW_OTHER_DIRECTORIES)
1194                        break;
1195                if (!(dir->flags & DIR_NO_GITLINKS)) {
1196                        unsigned char sha1[20];
1197                        if (resolve_gitlink_ref(dirname, "HEAD", sha1) == 0)
1198                                return path_untracked;
1199                }
1200                return path_recurse;
1201        }
1202
1203        /* This is the "show_other_directories" case */
1204
1205        if (!(dir->flags & DIR_HIDE_EMPTY_DIRECTORIES))
1206                return exclude ? path_excluded : path_untracked;
1207
1208        return read_directory_recursive(dir, dirname, len, 1, simplify);
1209}
1210
1211/*
1212 * This is an inexact early pruning of any recursive directory
1213 * reading - if the path cannot possibly be in the pathspec,
1214 * return true, and we'll skip it early.
1215 */
1216static int simplify_away(const char *path, int pathlen, const struct path_simplify *simplify)
1217{
1218        if (simplify) {
1219                for (;;) {
1220                        const char *match = simplify->path;
1221                        int len = simplify->len;
1222
1223                        if (!match)
1224                                break;
1225                        if (len > pathlen)
1226                                len = pathlen;
1227                        if (!memcmp(path, match, len))
1228                                return 0;
1229                        simplify++;
1230                }
1231                return 1;
1232        }
1233        return 0;
1234}
1235
1236/*
1237 * This function tells us whether an excluded path matches a
1238 * list of "interesting" pathspecs. That is, whether a path matched
1239 * by any of the pathspecs could possibly be ignored by excluding
1240 * the specified path. This can happen if:
1241 *
1242 *   1. the path is mentioned explicitly in the pathspec
1243 *
1244 *   2. the path is a directory prefix of some element in the
1245 *      pathspec
1246 */
1247static int exclude_matches_pathspec(const char *path, int len,
1248                const struct path_simplify *simplify)
1249{
1250        if (simplify) {
1251                for (; simplify->path; simplify++) {
1252                        if (len == simplify->len
1253                            && !memcmp(path, simplify->path, len))
1254                                return 1;
1255                        if (len < simplify->len
1256                            && simplify->path[len] == '/'
1257                            && !memcmp(path, simplify->path, len))
1258                                return 1;
1259                }
1260        }
1261        return 0;
1262}
1263
1264static int get_index_dtype(const char *path, int len)
1265{
1266        int pos;
1267        const struct cache_entry *ce;
1268
1269        ce = cache_file_exists(path, len, 0);
1270        if (ce) {
1271                if (!ce_uptodate(ce))
1272                        return DT_UNKNOWN;
1273                if (S_ISGITLINK(ce->ce_mode))
1274                        return DT_DIR;
1275                /*
1276                 * Nobody actually cares about the
1277                 * difference between DT_LNK and DT_REG
1278                 */
1279                return DT_REG;
1280        }
1281
1282        /* Try to look it up as a directory */
1283        pos = cache_name_pos(path, len);
1284        if (pos >= 0)
1285                return DT_UNKNOWN;
1286        pos = -pos-1;
1287        while (pos < active_nr) {
1288                ce = active_cache[pos++];
1289                if (strncmp(ce->name, path, len))
1290                        break;
1291                if (ce->name[len] > '/')
1292                        break;
1293                if (ce->name[len] < '/')
1294                        continue;
1295                if (!ce_uptodate(ce))
1296                        break;  /* continue? */
1297                return DT_DIR;
1298        }
1299        return DT_UNKNOWN;
1300}
1301
1302static int get_dtype(struct dirent *de, const char *path, int len)
1303{
1304        int dtype = de ? DTYPE(de) : DT_UNKNOWN;
1305        struct stat st;
1306
1307        if (dtype != DT_UNKNOWN)
1308                return dtype;
1309        dtype = get_index_dtype(path, len);
1310        if (dtype != DT_UNKNOWN)
1311                return dtype;
1312        if (lstat(path, &st))
1313                return dtype;
1314        if (S_ISREG(st.st_mode))
1315                return DT_REG;
1316        if (S_ISDIR(st.st_mode))
1317                return DT_DIR;
1318        if (S_ISLNK(st.st_mode))
1319                return DT_LNK;
1320        return dtype;
1321}
1322
1323static enum path_treatment treat_one_path(struct dir_struct *dir,
1324                                          struct strbuf *path,
1325                                          const struct path_simplify *simplify,
1326                                          int dtype, struct dirent *de)
1327{
1328        int exclude;
1329        int has_path_in_index = !!cache_file_exists(path->buf, path->len, ignore_case);
1330
1331        if (dtype == DT_UNKNOWN)
1332                dtype = get_dtype(de, path->buf, path->len);
1333
1334        /* Always exclude indexed files */
1335        if (dtype != DT_DIR && has_path_in_index)
1336                return path_none;
1337
1338        /*
1339         * When we are looking at a directory P in the working tree,
1340         * there are three cases:
1341         *
1342         * (1) P exists in the index.  Everything inside the directory P in
1343         * the working tree needs to go when P is checked out from the
1344         * index.
1345         *
1346         * (2) P does not exist in the index, but there is P/Q in the index.
1347         * We know P will stay a directory when we check out the contents
1348         * of the index, but we do not know yet if there is a directory
1349         * P/Q in the working tree to be killed, so we need to recurse.
1350         *
1351         * (3) P does not exist in the index, and there is no P/Q in the index
1352         * to require P to be a directory, either.  Only in this case, we
1353         * know that everything inside P will not be killed without
1354         * recursing.
1355         */
1356        if ((dir->flags & DIR_COLLECT_KILLED_ONLY) &&
1357            (dtype == DT_DIR) &&
1358            !has_path_in_index &&
1359            (directory_exists_in_index(path->buf, path->len) == index_nonexistent))
1360                return path_none;
1361
1362        exclude = is_excluded(dir, path->buf, &dtype);
1363
1364        /*
1365         * Excluded? If we don't explicitly want to show
1366         * ignored files, ignore it
1367         */
1368        if (exclude && !(dir->flags & (DIR_SHOW_IGNORED|DIR_SHOW_IGNORED_TOO)))
1369                return path_excluded;
1370
1371        switch (dtype) {
1372        default:
1373                return path_none;
1374        case DT_DIR:
1375                strbuf_addch(path, '/');
1376                return treat_directory(dir, path->buf, path->len, exclude,
1377                        simplify);
1378        case DT_REG:
1379        case DT_LNK:
1380                return exclude ? path_excluded : path_untracked;
1381        }
1382}
1383
1384static enum path_treatment treat_path(struct dir_struct *dir,
1385                                      struct dirent *de,
1386                                      struct strbuf *path,
1387                                      int baselen,
1388                                      const struct path_simplify *simplify)
1389{
1390        int dtype;
1391
1392        if (is_dot_or_dotdot(de->d_name) || !strcmp(de->d_name, ".git"))
1393                return path_none;
1394        strbuf_setlen(path, baselen);
1395        strbuf_addstr(path, de->d_name);
1396        if (simplify_away(path->buf, path->len, simplify))
1397                return path_none;
1398
1399        dtype = DTYPE(de);
1400        return treat_one_path(dir, path, simplify, dtype, de);
1401}
1402
1403/*
1404 * Read a directory tree. We currently ignore anything but
1405 * directories, regular files and symlinks. That's because git
1406 * doesn't handle them at all yet. Maybe that will change some
1407 * day.
1408 *
1409 * Also, we ignore the name ".git" (even if it is not a directory).
1410 * That likely will not change.
1411 *
1412 * Returns the most significant path_treatment value encountered in the scan.
1413 */
1414static enum path_treatment read_directory_recursive(struct dir_struct *dir,
1415                                    const char *base, int baselen,
1416                                    int check_only,
1417                                    const struct path_simplify *simplify)
1418{
1419        DIR *fdir;
1420        enum path_treatment state, subdir_state, dir_state = path_none;
1421        struct dirent *de;
1422        struct strbuf path = STRBUF_INIT;
1423
1424        strbuf_add(&path, base, baselen);
1425
1426        fdir = opendir(path.len ? path.buf : ".");
1427        if (!fdir)
1428                goto out;
1429
1430        while ((de = readdir(fdir)) != NULL) {
1431                /* check how the file or directory should be treated */
1432                state = treat_path(dir, de, &path, baselen, simplify);
1433                if (state > dir_state)
1434                        dir_state = state;
1435
1436                /* recurse into subdir if instructed by treat_path */
1437                if (state == path_recurse) {
1438                        subdir_state = read_directory_recursive(dir, path.buf,
1439                                path.len, check_only, simplify);
1440                        if (subdir_state > dir_state)
1441                                dir_state = subdir_state;
1442                }
1443
1444                if (check_only) {
1445                        /* abort early if maximum state has been reached */
1446                        if (dir_state == path_untracked)
1447                                break;
1448                        /* skip the dir_add_* part */
1449                        continue;
1450                }
1451
1452                /* add the path to the appropriate result list */
1453                switch (state) {
1454                case path_excluded:
1455                        if (dir->flags & DIR_SHOW_IGNORED)
1456                                dir_add_name(dir, path.buf, path.len);
1457                        else if ((dir->flags & DIR_SHOW_IGNORED_TOO) ||
1458                                ((dir->flags & DIR_COLLECT_IGNORED) &&
1459                                exclude_matches_pathspec(path.buf, path.len,
1460                                        simplify)))
1461                                dir_add_ignored(dir, path.buf, path.len);
1462                        break;
1463
1464                case path_untracked:
1465                        if (!(dir->flags & DIR_SHOW_IGNORED))
1466                                dir_add_name(dir, path.buf, path.len);
1467                        break;
1468
1469                default:
1470                        break;
1471                }
1472        }
1473        closedir(fdir);
1474 out:
1475        strbuf_release(&path);
1476
1477        return dir_state;
1478}
1479
1480static int cmp_name(const void *p1, const void *p2)
1481{
1482        const struct dir_entry *e1 = *(const struct dir_entry **)p1;
1483        const struct dir_entry *e2 = *(const struct dir_entry **)p2;
1484
1485        return name_compare(e1->name, e1->len, e2->name, e2->len);
1486}
1487
1488static struct path_simplify *create_simplify(const char **pathspec)
1489{
1490        int nr, alloc = 0;
1491        struct path_simplify *simplify = NULL;
1492
1493        if (!pathspec)
1494                return NULL;
1495
1496        for (nr = 0 ; ; nr++) {
1497                const char *match;
1498                ALLOC_GROW(simplify, nr + 1, alloc);
1499                match = *pathspec++;
1500                if (!match)
1501                        break;
1502                simplify[nr].path = match;
1503                simplify[nr].len = simple_length(match);
1504        }
1505        simplify[nr].path = NULL;
1506        simplify[nr].len = 0;
1507        return simplify;
1508}
1509
1510static void free_simplify(struct path_simplify *simplify)
1511{
1512        free(simplify);
1513}
1514
1515static int treat_leading_path(struct dir_struct *dir,
1516                              const char *path, int len,
1517                              const struct path_simplify *simplify)
1518{
1519        struct strbuf sb = STRBUF_INIT;
1520        int baselen, rc = 0;
1521        const char *cp;
1522        int old_flags = dir->flags;
1523
1524        while (len && path[len - 1] == '/')
1525                len--;
1526        if (!len)
1527                return 1;
1528        baselen = 0;
1529        dir->flags &= ~DIR_SHOW_OTHER_DIRECTORIES;
1530        while (1) {
1531                cp = path + baselen + !!baselen;
1532                cp = memchr(cp, '/', path + len - cp);
1533                if (!cp)
1534                        baselen = len;
1535                else
1536                        baselen = cp - path;
1537                strbuf_setlen(&sb, 0);
1538                strbuf_add(&sb, path, baselen);
1539                if (!is_directory(sb.buf))
1540                        break;
1541                if (simplify_away(sb.buf, sb.len, simplify))
1542                        break;
1543                if (treat_one_path(dir, &sb, simplify,
1544                                   DT_DIR, NULL) == path_none)
1545                        break; /* do not recurse into it */
1546                if (len <= baselen) {
1547                        rc = 1;
1548                        break; /* finished checking */
1549                }
1550        }
1551        strbuf_release(&sb);
1552        dir->flags = old_flags;
1553        return rc;
1554}
1555
1556int read_directory(struct dir_struct *dir, const char *path, int len, const struct pathspec *pathspec)
1557{
1558        struct path_simplify *simplify;
1559
1560        /*
1561         * Check out create_simplify()
1562         */
1563        if (pathspec)
1564                GUARD_PATHSPEC(pathspec,
1565                               PATHSPEC_FROMTOP |
1566                               PATHSPEC_MAXDEPTH |
1567                               PATHSPEC_LITERAL |
1568                               PATHSPEC_GLOB |
1569                               PATHSPEC_ICASE |
1570                               PATHSPEC_EXCLUDE);
1571
1572        if (has_symlink_leading_path(path, len))
1573                return dir->nr;
1574
1575        /*
1576         * exclude patterns are treated like positive ones in
1577         * create_simplify. Usually exclude patterns should be a
1578         * subset of positive ones, which has no impacts on
1579         * create_simplify().
1580         */
1581        simplify = create_simplify(pathspec ? pathspec->_raw : NULL);
1582        if (!len || treat_leading_path(dir, path, len, simplify))
1583                read_directory_recursive(dir, path, len, 0, simplify);
1584        free_simplify(simplify);
1585        qsort(dir->entries, dir->nr, sizeof(struct dir_entry *), cmp_name);
1586        qsort(dir->ignored, dir->ignored_nr, sizeof(struct dir_entry *), cmp_name);
1587        return dir->nr;
1588}
1589
1590int file_exists(const char *f)
1591{
1592        struct stat sb;
1593        return lstat(f, &sb) == 0;
1594}
1595
1596/*
1597 * Given two normalized paths (a trailing slash is ok), if subdir is
1598 * outside dir, return -1.  Otherwise return the offset in subdir that
1599 * can be used as relative path to dir.
1600 */
1601int dir_inside_of(const char *subdir, const char *dir)
1602{
1603        int offset = 0;
1604
1605        assert(dir && subdir && *dir && *subdir);
1606
1607        while (*dir && *subdir && *dir == *subdir) {
1608                dir++;
1609                subdir++;
1610                offset++;
1611        }
1612
1613        /* hel[p]/me vs hel[l]/yeah */
1614        if (*dir && *subdir)
1615                return -1;
1616
1617        if (!*subdir)
1618                return !*dir ? offset : -1; /* same dir */
1619
1620        /* foo/[b]ar vs foo/[] */
1621        if (is_dir_sep(dir[-1]))
1622                return is_dir_sep(subdir[-1]) ? offset : -1;
1623
1624        /* foo[/]bar vs foo[] */
1625        return is_dir_sep(*subdir) ? offset + 1 : -1;
1626}
1627
1628int is_inside_dir(const char *dir)
1629{
1630        char *cwd;
1631        int rc;
1632
1633        if (!dir)
1634                return 0;
1635
1636        cwd = xgetcwd();
1637        rc = (dir_inside_of(cwd, dir) >= 0);
1638        free(cwd);
1639        return rc;
1640}
1641
1642int is_empty_dir(const char *path)
1643{
1644        DIR *dir = opendir(path);
1645        struct dirent *e;
1646        int ret = 1;
1647
1648        if (!dir)
1649                return 0;
1650
1651        while ((e = readdir(dir)) != NULL)
1652                if (!is_dot_or_dotdot(e->d_name)) {
1653                        ret = 0;
1654                        break;
1655                }
1656
1657        closedir(dir);
1658        return ret;
1659}
1660
1661static int remove_dir_recurse(struct strbuf *path, int flag, int *kept_up)
1662{
1663        DIR *dir;
1664        struct dirent *e;
1665        int ret = 0, original_len = path->len, len, kept_down = 0;
1666        int only_empty = (flag & REMOVE_DIR_EMPTY_ONLY);
1667        int keep_toplevel = (flag & REMOVE_DIR_KEEP_TOPLEVEL);
1668        unsigned char submodule_head[20];
1669
1670        if ((flag & REMOVE_DIR_KEEP_NESTED_GIT) &&
1671            !resolve_gitlink_ref(path->buf, "HEAD", submodule_head)) {
1672                /* Do not descend and nuke a nested git work tree. */
1673                if (kept_up)
1674                        *kept_up = 1;
1675                return 0;
1676        }
1677
1678        flag &= ~REMOVE_DIR_KEEP_TOPLEVEL;
1679        dir = opendir(path->buf);
1680        if (!dir) {
1681                if (errno == ENOENT)
1682                        return keep_toplevel ? -1 : 0;
1683                else if (errno == EACCES && !keep_toplevel)
1684                        /*
1685                         * An empty dir could be removable even if it
1686                         * is unreadable:
1687                         */
1688                        return rmdir(path->buf);
1689                else
1690                        return -1;
1691        }
1692        if (path->buf[original_len - 1] != '/')
1693                strbuf_addch(path, '/');
1694
1695        len = path->len;
1696        while ((e = readdir(dir)) != NULL) {
1697                struct stat st;
1698                if (is_dot_or_dotdot(e->d_name))
1699                        continue;
1700
1701                strbuf_setlen(path, len);
1702                strbuf_addstr(path, e->d_name);
1703                if (lstat(path->buf, &st)) {
1704                        if (errno == ENOENT)
1705                                /*
1706                                 * file disappeared, which is what we
1707                                 * wanted anyway
1708                                 */
1709                                continue;
1710                        /* fall thru */
1711                } else if (S_ISDIR(st.st_mode)) {
1712                        if (!remove_dir_recurse(path, flag, &kept_down))
1713                                continue; /* happy */
1714                } else if (!only_empty &&
1715                           (!unlink(path->buf) || errno == ENOENT)) {
1716                        continue; /* happy, too */
1717                }
1718
1719                /* path too long, stat fails, or non-directory still exists */
1720                ret = -1;
1721                break;
1722        }
1723        closedir(dir);
1724
1725        strbuf_setlen(path, original_len);
1726        if (!ret && !keep_toplevel && !kept_down)
1727                ret = (!rmdir(path->buf) || errno == ENOENT) ? 0 : -1;
1728        else if (kept_up)
1729                /*
1730                 * report the uplevel that it is not an error that we
1731                 * did not rmdir() our directory.
1732                 */
1733                *kept_up = !ret;
1734        return ret;
1735}
1736
1737int remove_dir_recursively(struct strbuf *path, int flag)
1738{
1739        return remove_dir_recurse(path, flag, NULL);
1740}
1741
1742void setup_standard_excludes(struct dir_struct *dir)
1743{
1744        const char *path;
1745        char *xdg_path;
1746
1747        dir->exclude_per_dir = ".gitignore";
1748        path = git_path("info/exclude");
1749        if (!excludes_file) {
1750                home_config_paths(NULL, &xdg_path, "ignore");
1751                excludes_file = xdg_path;
1752        }
1753        if (!access_or_warn(path, R_OK, 0))
1754                add_excludes_from_file(dir, path);
1755        if (excludes_file && !access_or_warn(excludes_file, R_OK, 0))
1756                add_excludes_from_file(dir, excludes_file);
1757}
1758
1759int remove_path(const char *name)
1760{
1761        char *slash;
1762
1763        if (unlink(name) && errno != ENOENT && errno != ENOTDIR)
1764                return -1;
1765
1766        slash = strrchr(name, '/');
1767        if (slash) {
1768                char *dirs = xstrdup(name);
1769                slash = dirs + (slash - name);
1770                do {
1771                        *slash = '\0';
1772                } while (rmdir(dirs) == 0 && (slash = strrchr(dirs, '/')));
1773                free(dirs);
1774        }
1775        return 0;
1776}
1777
1778/*
1779 * Frees memory within dir which was allocated for exclude lists and
1780 * the exclude_stack.  Does not free dir itself.
1781 */
1782void clear_directory(struct dir_struct *dir)
1783{
1784        int i, j;
1785        struct exclude_list_group *group;
1786        struct exclude_list *el;
1787        struct exclude_stack *stk;
1788
1789        for (i = EXC_CMDL; i <= EXC_FILE; i++) {
1790                group = &dir->exclude_list_group[i];
1791                for (j = 0; j < group->nr; j++) {
1792                        el = &group->el[j];
1793                        if (i == EXC_DIRS)
1794                                free((char *)el->src);
1795                        clear_exclude_list(el);
1796                }
1797                free(group->el);
1798        }
1799
1800        stk = dir->exclude_stack;
1801        while (stk) {
1802                struct exclude_stack *prev = stk->prev;
1803                free(stk);
1804                stk = prev;
1805        }
1806        strbuf_release(&dir->basebuf);
1807}