dir.con commit ls-files: add pathspec matching for submodules (75a6315)
   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#include "utf8.h"
  16#include "varint.h"
  17#include "ewah/ewok.h"
  18
  19struct path_simplify {
  20        int len;
  21        const char *path;
  22};
  23
  24/*
  25 * Tells read_directory_recursive how a file or directory should be treated.
  26 * Values are ordered by significance, e.g. if a directory contains both
  27 * excluded and untracked files, it is listed as untracked because
  28 * path_untracked > path_excluded.
  29 */
  30enum path_treatment {
  31        path_none = 0,
  32        path_recurse,
  33        path_excluded,
  34        path_untracked
  35};
  36
  37/*
  38 * Support data structure for our opendir/readdir/closedir wrappers
  39 */
  40struct cached_dir {
  41        DIR *fdir;
  42        struct untracked_cache_dir *untracked;
  43        int nr_files;
  44        int nr_dirs;
  45
  46        struct dirent *de;
  47        const char *file;
  48        struct untracked_cache_dir *ucd;
  49};
  50
  51static enum path_treatment read_directory_recursive(struct dir_struct *dir,
  52        const char *path, int len, struct untracked_cache_dir *untracked,
  53        int check_only, const struct path_simplify *simplify);
  54static int get_dtype(struct dirent *de, const char *path, int len);
  55
  56int fspathcmp(const char *a, const char *b)
  57{
  58        return ignore_case ? strcasecmp(a, b) : strcmp(a, b);
  59}
  60
  61int fspathncmp(const char *a, const char *b, size_t count)
  62{
  63        return ignore_case ? strncasecmp(a, b, count) : strncmp(a, b, count);
  64}
  65
  66int git_fnmatch(const struct pathspec_item *item,
  67                const char *pattern, const char *string,
  68                int prefix)
  69{
  70        if (prefix > 0) {
  71                if (ps_strncmp(item, pattern, string, prefix))
  72                        return WM_NOMATCH;
  73                pattern += prefix;
  74                string += prefix;
  75        }
  76        if (item->flags & PATHSPEC_ONESTAR) {
  77                int pattern_len = strlen(++pattern);
  78                int string_len = strlen(string);
  79                return string_len < pattern_len ||
  80                        ps_strcmp(item, pattern,
  81                                  string + string_len - pattern_len);
  82        }
  83        if (item->magic & PATHSPEC_GLOB)
  84                return wildmatch(pattern, string,
  85                                 WM_PATHNAME |
  86                                 (item->magic & PATHSPEC_ICASE ? WM_CASEFOLD : 0),
  87                                 NULL);
  88        else
  89                /* wildmatch has not learned no FNM_PATHNAME mode yet */
  90                return wildmatch(pattern, string,
  91                                 item->magic & PATHSPEC_ICASE ? WM_CASEFOLD : 0,
  92                                 NULL);
  93}
  94
  95static int fnmatch_icase_mem(const char *pattern, int patternlen,
  96                             const char *string, int stringlen,
  97                             int flags)
  98{
  99        int match_status;
 100        struct strbuf pat_buf = STRBUF_INIT;
 101        struct strbuf str_buf = STRBUF_INIT;
 102        const char *use_pat = pattern;
 103        const char *use_str = string;
 104
 105        if (pattern[patternlen]) {
 106                strbuf_add(&pat_buf, pattern, patternlen);
 107                use_pat = pat_buf.buf;
 108        }
 109        if (string[stringlen]) {
 110                strbuf_add(&str_buf, string, stringlen);
 111                use_str = str_buf.buf;
 112        }
 113
 114        if (ignore_case)
 115                flags |= WM_CASEFOLD;
 116        match_status = wildmatch(use_pat, use_str, flags, NULL);
 117
 118        strbuf_release(&pat_buf);
 119        strbuf_release(&str_buf);
 120
 121        return match_status;
 122}
 123
 124static size_t common_prefix_len(const struct pathspec *pathspec)
 125{
 126        int n;
 127        size_t max = 0;
 128
 129        /*
 130         * ":(icase)path" is treated as a pathspec full of
 131         * wildcard. In other words, only prefix is considered common
 132         * prefix. If the pathspec is abc/foo abc/bar, running in
 133         * subdir xyz, the common prefix is still xyz, not xuz/abc as
 134         * in non-:(icase).
 135         */
 136        GUARD_PATHSPEC(pathspec,
 137                       PATHSPEC_FROMTOP |
 138                       PATHSPEC_MAXDEPTH |
 139                       PATHSPEC_LITERAL |
 140                       PATHSPEC_GLOB |
 141                       PATHSPEC_ICASE |
 142                       PATHSPEC_EXCLUDE);
 143
 144        for (n = 0; n < pathspec->nr; n++) {
 145                size_t i = 0, len = 0, item_len;
 146                if (pathspec->items[n].magic & PATHSPEC_EXCLUDE)
 147                        continue;
 148                if (pathspec->items[n].magic & PATHSPEC_ICASE)
 149                        item_len = pathspec->items[n].prefix;
 150                else
 151                        item_len = pathspec->items[n].nowildcard_len;
 152                while (i < item_len && (n == 0 || i < max)) {
 153                        char c = pathspec->items[n].match[i];
 154                        if (c != pathspec->items[0].match[i])
 155                                break;
 156                        if (c == '/')
 157                                len = i + 1;
 158                        i++;
 159                }
 160                if (n == 0 || len < max) {
 161                        max = len;
 162                        if (!max)
 163                                break;
 164                }
 165        }
 166        return max;
 167}
 168
 169/*
 170 * Returns a copy of the longest leading path common among all
 171 * pathspecs.
 172 */
 173char *common_prefix(const struct pathspec *pathspec)
 174{
 175        unsigned long len = common_prefix_len(pathspec);
 176
 177        return len ? xmemdupz(pathspec->items[0].match, len) : NULL;
 178}
 179
 180int fill_directory(struct dir_struct *dir, const struct pathspec *pathspec)
 181{
 182        size_t len;
 183
 184        /*
 185         * Calculate common prefix for the pathspec, and
 186         * use that to optimize the directory walk
 187         */
 188        len = common_prefix_len(pathspec);
 189
 190        /* Read the directory and prune it */
 191        read_directory(dir, pathspec->nr ? pathspec->_raw[0] : "", len, pathspec);
 192        return len;
 193}
 194
 195int within_depth(const char *name, int namelen,
 196                        int depth, int max_depth)
 197{
 198        const char *cp = name, *cpe = name + namelen;
 199
 200        while (cp < cpe) {
 201                if (*cp++ != '/')
 202                        continue;
 203                depth++;
 204                if (depth > max_depth)
 205                        return 0;
 206        }
 207        return 1;
 208}
 209
 210#define DO_MATCH_EXCLUDE   (1<<0)
 211#define DO_MATCH_DIRECTORY (1<<1)
 212#define DO_MATCH_SUBMODULE (1<<2)
 213
 214/*
 215 * Does 'match' match the given name?
 216 * A match is found if
 217 *
 218 * (1) the 'match' string is leading directory of 'name', or
 219 * (2) the 'match' string is a wildcard and matches 'name', or
 220 * (3) the 'match' string is exactly the same as 'name'.
 221 *
 222 * and the return value tells which case it was.
 223 *
 224 * It returns 0 when there is no match.
 225 */
 226static int match_pathspec_item(const struct pathspec_item *item, int prefix,
 227                               const char *name, int namelen, unsigned flags)
 228{
 229        /* name/namelen has prefix cut off by caller */
 230        const char *match = item->match + prefix;
 231        int matchlen = item->len - prefix;
 232
 233        /*
 234         * The normal call pattern is:
 235         * 1. prefix = common_prefix_len(ps);
 236         * 2. prune something, or fill_directory
 237         * 3. match_pathspec()
 238         *
 239         * 'prefix' at #1 may be shorter than the command's prefix and
 240         * it's ok for #2 to match extra files. Those extras will be
 241         * trimmed at #3.
 242         *
 243         * Suppose the pathspec is 'foo' and '../bar' running from
 244         * subdir 'xyz'. The common prefix at #1 will be empty, thanks
 245         * to "../". We may have xyz/foo _and_ XYZ/foo after #2. The
 246         * user does not want XYZ/foo, only the "foo" part should be
 247         * case-insensitive. We need to filter out XYZ/foo here. In
 248         * other words, we do not trust the caller on comparing the
 249         * prefix part when :(icase) is involved. We do exact
 250         * comparison ourselves.
 251         *
 252         * Normally the caller (common_prefix_len() in fact) does
 253         * _exact_ matching on name[-prefix+1..-1] and we do not need
 254         * to check that part. Be defensive and check it anyway, in
 255         * case common_prefix_len is changed, or a new caller is
 256         * introduced that does not use common_prefix_len.
 257         *
 258         * If the penalty turns out too high when prefix is really
 259         * long, maybe change it to
 260         * strncmp(match, name, item->prefix - prefix)
 261         */
 262        if (item->prefix && (item->magic & PATHSPEC_ICASE) &&
 263            strncmp(item->match, name - prefix, item->prefix))
 264                return 0;
 265
 266        /* If the match was just the prefix, we matched */
 267        if (!*match)
 268                return MATCHED_RECURSIVELY;
 269
 270        if (matchlen <= namelen && !ps_strncmp(item, match, name, matchlen)) {
 271                if (matchlen == namelen)
 272                        return MATCHED_EXACTLY;
 273
 274                if (match[matchlen-1] == '/' || name[matchlen] == '/')
 275                        return MATCHED_RECURSIVELY;
 276        } else if ((flags & DO_MATCH_DIRECTORY) &&
 277                   match[matchlen - 1] == '/' &&
 278                   namelen == matchlen - 1 &&
 279                   !ps_strncmp(item, match, name, namelen))
 280                return MATCHED_EXACTLY;
 281
 282        if (item->nowildcard_len < item->len &&
 283            !git_fnmatch(item, match, name,
 284                         item->nowildcard_len - prefix))
 285                return MATCHED_FNMATCH;
 286
 287        /* Perform checks to see if "name" is a super set of the pathspec */
 288        if (flags & DO_MATCH_SUBMODULE) {
 289                /* name is a literal prefix of the pathspec */
 290                if ((namelen < matchlen) &&
 291                    (match[namelen] == '/') &&
 292                    !ps_strncmp(item, match, name, namelen))
 293                        return MATCHED_RECURSIVELY;
 294
 295                /* name" doesn't match up to the first wild character */
 296                if (item->nowildcard_len < item->len &&
 297                    ps_strncmp(item, match, name,
 298                               item->nowildcard_len - prefix))
 299                        return 0;
 300
 301                /*
 302                 * Here is where we would perform a wildmatch to check if
 303                 * "name" can be matched as a directory (or a prefix) against
 304                 * the pathspec.  Since wildmatch doesn't have this capability
 305                 * at the present we have to punt and say that it is a match,
 306                 * potentially returning a false positive
 307                 * The submodules themselves will be able to perform more
 308                 * accurate matching to determine if the pathspec matches.
 309                 */
 310                return MATCHED_RECURSIVELY;
 311        }
 312
 313        return 0;
 314}
 315
 316/*
 317 * Given a name and a list of pathspecs, returns the nature of the
 318 * closest (i.e. most specific) match of the name to any of the
 319 * pathspecs.
 320 *
 321 * The caller typically calls this multiple times with the same
 322 * pathspec and seen[] array but with different name/namelen
 323 * (e.g. entries from the index) and is interested in seeing if and
 324 * how each pathspec matches all the names it calls this function
 325 * with.  A mark is left in the seen[] array for each pathspec element
 326 * indicating the closest type of match that element achieved, so if
 327 * seen[n] remains zero after multiple invocations, that means the nth
 328 * pathspec did not match any names, which could indicate that the
 329 * user mistyped the nth pathspec.
 330 */
 331static int do_match_pathspec(const struct pathspec *ps,
 332                             const char *name, int namelen,
 333                             int prefix, char *seen,
 334                             unsigned flags)
 335{
 336        int i, retval = 0, exclude = flags & DO_MATCH_EXCLUDE;
 337
 338        GUARD_PATHSPEC(ps,
 339                       PATHSPEC_FROMTOP |
 340                       PATHSPEC_MAXDEPTH |
 341                       PATHSPEC_LITERAL |
 342                       PATHSPEC_GLOB |
 343                       PATHSPEC_ICASE |
 344                       PATHSPEC_EXCLUDE);
 345
 346        if (!ps->nr) {
 347                if (!ps->recursive ||
 348                    !(ps->magic & PATHSPEC_MAXDEPTH) ||
 349                    ps->max_depth == -1)
 350                        return MATCHED_RECURSIVELY;
 351
 352                if (within_depth(name, namelen, 0, ps->max_depth))
 353                        return MATCHED_EXACTLY;
 354                else
 355                        return 0;
 356        }
 357
 358        name += prefix;
 359        namelen -= prefix;
 360
 361        for (i = ps->nr - 1; i >= 0; i--) {
 362                int how;
 363
 364                if ((!exclude &&   ps->items[i].magic & PATHSPEC_EXCLUDE) ||
 365                    ( exclude && !(ps->items[i].magic & PATHSPEC_EXCLUDE)))
 366                        continue;
 367
 368                if (seen && seen[i] == MATCHED_EXACTLY)
 369                        continue;
 370                /*
 371                 * Make exclude patterns optional and never report
 372                 * "pathspec ':(exclude)foo' matches no files"
 373                 */
 374                if (seen && ps->items[i].magic & PATHSPEC_EXCLUDE)
 375                        seen[i] = MATCHED_FNMATCH;
 376                how = match_pathspec_item(ps->items+i, prefix, name,
 377                                          namelen, flags);
 378                if (ps->recursive &&
 379                    (ps->magic & PATHSPEC_MAXDEPTH) &&
 380                    ps->max_depth != -1 &&
 381                    how && how != MATCHED_FNMATCH) {
 382                        int len = ps->items[i].len;
 383                        if (name[len] == '/')
 384                                len++;
 385                        if (within_depth(name+len, namelen-len, 0, ps->max_depth))
 386                                how = MATCHED_EXACTLY;
 387                        else
 388                                how = 0;
 389                }
 390                if (how) {
 391                        if (retval < how)
 392                                retval = how;
 393                        if (seen && seen[i] < how)
 394                                seen[i] = how;
 395                }
 396        }
 397        return retval;
 398}
 399
 400int match_pathspec(const struct pathspec *ps,
 401                   const char *name, int namelen,
 402                   int prefix, char *seen, int is_dir)
 403{
 404        int positive, negative;
 405        unsigned flags = is_dir ? DO_MATCH_DIRECTORY : 0;
 406        positive = do_match_pathspec(ps, name, namelen,
 407                                     prefix, seen, flags);
 408        if (!(ps->magic & PATHSPEC_EXCLUDE) || !positive)
 409                return positive;
 410        negative = do_match_pathspec(ps, name, namelen,
 411                                     prefix, seen,
 412                                     flags | DO_MATCH_EXCLUDE);
 413        return negative ? 0 : positive;
 414}
 415
 416/**
 417 * Check if a submodule is a superset of the pathspec
 418 */
 419int submodule_path_match(const struct pathspec *ps,
 420                         const char *submodule_name,
 421                         char *seen)
 422{
 423        int matched = do_match_pathspec(ps, submodule_name,
 424                                        strlen(submodule_name),
 425                                        0, seen,
 426                                        DO_MATCH_DIRECTORY |
 427                                        DO_MATCH_SUBMODULE);
 428        return matched;
 429}
 430
 431int report_path_error(const char *ps_matched,
 432                      const struct pathspec *pathspec,
 433                      const char *prefix)
 434{
 435        /*
 436         * Make sure all pathspec matched; otherwise it is an error.
 437         */
 438        int num, errors = 0;
 439        for (num = 0; num < pathspec->nr; num++) {
 440                int other, found_dup;
 441
 442                if (ps_matched[num])
 443                        continue;
 444                /*
 445                 * The caller might have fed identical pathspec
 446                 * twice.  Do not barf on such a mistake.
 447                 * FIXME: parse_pathspec should have eliminated
 448                 * duplicate pathspec.
 449                 */
 450                for (found_dup = other = 0;
 451                     !found_dup && other < pathspec->nr;
 452                     other++) {
 453                        if (other == num || !ps_matched[other])
 454                                continue;
 455                        if (!strcmp(pathspec->items[other].original,
 456                                    pathspec->items[num].original))
 457                                /*
 458                                 * Ok, we have a match already.
 459                                 */
 460                                found_dup = 1;
 461                }
 462                if (found_dup)
 463                        continue;
 464
 465                error("pathspec '%s' did not match any file(s) known to git.",
 466                      pathspec->items[num].original);
 467                errors++;
 468        }
 469        return errors;
 470}
 471
 472/*
 473 * Return the length of the "simple" part of a path match limiter.
 474 */
 475int simple_length(const char *match)
 476{
 477        int len = -1;
 478
 479        for (;;) {
 480                unsigned char c = *match++;
 481                len++;
 482                if (c == '\0' || is_glob_special(c))
 483                        return len;
 484        }
 485}
 486
 487int no_wildcard(const char *string)
 488{
 489        return string[simple_length(string)] == '\0';
 490}
 491
 492void parse_exclude_pattern(const char **pattern,
 493                           int *patternlen,
 494                           unsigned *flags,
 495                           int *nowildcardlen)
 496{
 497        const char *p = *pattern;
 498        size_t i, len;
 499
 500        *flags = 0;
 501        if (*p == '!') {
 502                *flags |= EXC_FLAG_NEGATIVE;
 503                p++;
 504        }
 505        len = strlen(p);
 506        if (len && p[len - 1] == '/') {
 507                len--;
 508                *flags |= EXC_FLAG_MUSTBEDIR;
 509        }
 510        for (i = 0; i < len; i++) {
 511                if (p[i] == '/')
 512                        break;
 513        }
 514        if (i == len)
 515                *flags |= EXC_FLAG_NODIR;
 516        *nowildcardlen = simple_length(p);
 517        /*
 518         * we should have excluded the trailing slash from 'p' too,
 519         * but that's one more allocation. Instead just make sure
 520         * nowildcardlen does not exceed real patternlen
 521         */
 522        if (*nowildcardlen > len)
 523                *nowildcardlen = len;
 524        if (*p == '*' && no_wildcard(p + 1))
 525                *flags |= EXC_FLAG_ENDSWITH;
 526        *pattern = p;
 527        *patternlen = len;
 528}
 529
 530void add_exclude(const char *string, const char *base,
 531                 int baselen, struct exclude_list *el, int srcpos)
 532{
 533        struct exclude *x;
 534        int patternlen;
 535        unsigned flags;
 536        int nowildcardlen;
 537
 538        parse_exclude_pattern(&string, &patternlen, &flags, &nowildcardlen);
 539        if (flags & EXC_FLAG_MUSTBEDIR) {
 540                FLEXPTR_ALLOC_MEM(x, pattern, string, patternlen);
 541        } else {
 542                x = xmalloc(sizeof(*x));
 543                x->pattern = string;
 544        }
 545        x->patternlen = patternlen;
 546        x->nowildcardlen = nowildcardlen;
 547        x->base = base;
 548        x->baselen = baselen;
 549        x->flags = flags;
 550        x->srcpos = srcpos;
 551        ALLOC_GROW(el->excludes, el->nr + 1, el->alloc);
 552        el->excludes[el->nr++] = x;
 553        x->el = el;
 554}
 555
 556static void *read_skip_worktree_file_from_index(const char *path, size_t *size,
 557                                                struct sha1_stat *sha1_stat)
 558{
 559        int pos, len;
 560        unsigned long sz;
 561        enum object_type type;
 562        void *data;
 563
 564        len = strlen(path);
 565        pos = cache_name_pos(path, len);
 566        if (pos < 0)
 567                return NULL;
 568        if (!ce_skip_worktree(active_cache[pos]))
 569                return NULL;
 570        data = read_sha1_file(active_cache[pos]->sha1, &type, &sz);
 571        if (!data || type != OBJ_BLOB) {
 572                free(data);
 573                return NULL;
 574        }
 575        *size = xsize_t(sz);
 576        if (sha1_stat) {
 577                memset(&sha1_stat->stat, 0, sizeof(sha1_stat->stat));
 578                hashcpy(sha1_stat->sha1, active_cache[pos]->sha1);
 579        }
 580        return data;
 581}
 582
 583/*
 584 * Frees memory within el which was allocated for exclude patterns and
 585 * the file buffer.  Does not free el itself.
 586 */
 587void clear_exclude_list(struct exclude_list *el)
 588{
 589        int i;
 590
 591        for (i = 0; i < el->nr; i++)
 592                free(el->excludes[i]);
 593        free(el->excludes);
 594        free(el->filebuf);
 595
 596        memset(el, 0, sizeof(*el));
 597}
 598
 599static void trim_trailing_spaces(char *buf)
 600{
 601        char *p, *last_space = NULL;
 602
 603        for (p = buf; *p; p++)
 604                switch (*p) {
 605                case ' ':
 606                        if (!last_space)
 607                                last_space = p;
 608                        break;
 609                case '\\':
 610                        p++;
 611                        if (!*p)
 612                                return;
 613                        /* fallthrough */
 614                default:
 615                        last_space = NULL;
 616                }
 617
 618        if (last_space)
 619                *last_space = '\0';
 620}
 621
 622/*
 623 * Given a subdirectory name and "dir" of the current directory,
 624 * search the subdir in "dir" and return it, or create a new one if it
 625 * does not exist in "dir".
 626 *
 627 * If "name" has the trailing slash, it'll be excluded in the search.
 628 */
 629static struct untracked_cache_dir *lookup_untracked(struct untracked_cache *uc,
 630                                                    struct untracked_cache_dir *dir,
 631                                                    const char *name, int len)
 632{
 633        int first, last;
 634        struct untracked_cache_dir *d;
 635        if (!dir)
 636                return NULL;
 637        if (len && name[len - 1] == '/')
 638                len--;
 639        first = 0;
 640        last = dir->dirs_nr;
 641        while (last > first) {
 642                int cmp, next = (last + first) >> 1;
 643                d = dir->dirs[next];
 644                cmp = strncmp(name, d->name, len);
 645                if (!cmp && strlen(d->name) > len)
 646                        cmp = -1;
 647                if (!cmp)
 648                        return d;
 649                if (cmp < 0) {
 650                        last = next;
 651                        continue;
 652                }
 653                first = next+1;
 654        }
 655
 656        uc->dir_created++;
 657        FLEX_ALLOC_MEM(d, name, name, len);
 658
 659        ALLOC_GROW(dir->dirs, dir->dirs_nr + 1, dir->dirs_alloc);
 660        memmove(dir->dirs + first + 1, dir->dirs + first,
 661                (dir->dirs_nr - first) * sizeof(*dir->dirs));
 662        dir->dirs_nr++;
 663        dir->dirs[first] = d;
 664        return d;
 665}
 666
 667static void do_invalidate_gitignore(struct untracked_cache_dir *dir)
 668{
 669        int i;
 670        dir->valid = 0;
 671        dir->untracked_nr = 0;
 672        for (i = 0; i < dir->dirs_nr; i++)
 673                do_invalidate_gitignore(dir->dirs[i]);
 674}
 675
 676static void invalidate_gitignore(struct untracked_cache *uc,
 677                                 struct untracked_cache_dir *dir)
 678{
 679        uc->gitignore_invalidated++;
 680        do_invalidate_gitignore(dir);
 681}
 682
 683static void invalidate_directory(struct untracked_cache *uc,
 684                                 struct untracked_cache_dir *dir)
 685{
 686        int i;
 687        uc->dir_invalidated++;
 688        dir->valid = 0;
 689        dir->untracked_nr = 0;
 690        for (i = 0; i < dir->dirs_nr; i++)
 691                dir->dirs[i]->recurse = 0;
 692}
 693
 694/*
 695 * Given a file with name "fname", read it (either from disk, or from
 696 * the index if "check_index" is non-zero), parse it and store the
 697 * exclude rules in "el".
 698 *
 699 * If "ss" is not NULL, compute SHA-1 of the exclude file and fill
 700 * stat data from disk (only valid if add_excludes returns zero). If
 701 * ss_valid is non-zero, "ss" must contain good value as input.
 702 */
 703static int add_excludes(const char *fname, const char *base, int baselen,
 704                        struct exclude_list *el, int check_index,
 705                        struct sha1_stat *sha1_stat)
 706{
 707        struct stat st;
 708        int fd, i, lineno = 1;
 709        size_t size = 0;
 710        char *buf, *entry;
 711
 712        fd = open(fname, O_RDONLY);
 713        if (fd < 0 || fstat(fd, &st) < 0) {
 714                if (errno != ENOENT)
 715                        warn_on_inaccessible(fname);
 716                if (0 <= fd)
 717                        close(fd);
 718                if (!check_index ||
 719                    (buf = read_skip_worktree_file_from_index(fname, &size, sha1_stat)) == NULL)
 720                        return -1;
 721                if (size == 0) {
 722                        free(buf);
 723                        return 0;
 724                }
 725                if (buf[size-1] != '\n') {
 726                        buf = xrealloc(buf, st_add(size, 1));
 727                        buf[size++] = '\n';
 728                }
 729        } else {
 730                size = xsize_t(st.st_size);
 731                if (size == 0) {
 732                        if (sha1_stat) {
 733                                fill_stat_data(&sha1_stat->stat, &st);
 734                                hashcpy(sha1_stat->sha1, EMPTY_BLOB_SHA1_BIN);
 735                                sha1_stat->valid = 1;
 736                        }
 737                        close(fd);
 738                        return 0;
 739                }
 740                buf = xmallocz(size);
 741                if (read_in_full(fd, buf, size) != size) {
 742                        free(buf);
 743                        close(fd);
 744                        return -1;
 745                }
 746                buf[size++] = '\n';
 747                close(fd);
 748                if (sha1_stat) {
 749                        int pos;
 750                        if (sha1_stat->valid &&
 751                            !match_stat_data_racy(&the_index, &sha1_stat->stat, &st))
 752                                ; /* no content change, ss->sha1 still good */
 753                        else if (check_index &&
 754                                 (pos = cache_name_pos(fname, strlen(fname))) >= 0 &&
 755                                 !ce_stage(active_cache[pos]) &&
 756                                 ce_uptodate(active_cache[pos]) &&
 757                                 !would_convert_to_git(fname))
 758                                hashcpy(sha1_stat->sha1, active_cache[pos]->sha1);
 759                        else
 760                                hash_sha1_file(buf, size, "blob", sha1_stat->sha1);
 761                        fill_stat_data(&sha1_stat->stat, &st);
 762                        sha1_stat->valid = 1;
 763                }
 764        }
 765
 766        el->filebuf = buf;
 767
 768        if (skip_utf8_bom(&buf, size))
 769                size -= buf - el->filebuf;
 770
 771        entry = buf;
 772
 773        for (i = 0; i < size; i++) {
 774                if (buf[i] == '\n') {
 775                        if (entry != buf + i && entry[0] != '#') {
 776                                buf[i - (i && buf[i-1] == '\r')] = 0;
 777                                trim_trailing_spaces(entry);
 778                                add_exclude(entry, base, baselen, el, lineno);
 779                        }
 780                        lineno++;
 781                        entry = buf + i + 1;
 782                }
 783        }
 784        return 0;
 785}
 786
 787int add_excludes_from_file_to_list(const char *fname, const char *base,
 788                                   int baselen, struct exclude_list *el,
 789                                   int check_index)
 790{
 791        return add_excludes(fname, base, baselen, el, check_index, NULL);
 792}
 793
 794struct exclude_list *add_exclude_list(struct dir_struct *dir,
 795                                      int group_type, const char *src)
 796{
 797        struct exclude_list *el;
 798        struct exclude_list_group *group;
 799
 800        group = &dir->exclude_list_group[group_type];
 801        ALLOC_GROW(group->el, group->nr + 1, group->alloc);
 802        el = &group->el[group->nr++];
 803        memset(el, 0, sizeof(*el));
 804        el->src = src;
 805        return el;
 806}
 807
 808/*
 809 * Used to set up core.excludesfile and .git/info/exclude lists.
 810 */
 811static void add_excludes_from_file_1(struct dir_struct *dir, const char *fname,
 812                                     struct sha1_stat *sha1_stat)
 813{
 814        struct exclude_list *el;
 815        /*
 816         * catch setup_standard_excludes() that's called before
 817         * dir->untracked is assigned. That function behaves
 818         * differently when dir->untracked is non-NULL.
 819         */
 820        if (!dir->untracked)
 821                dir->unmanaged_exclude_files++;
 822        el = add_exclude_list(dir, EXC_FILE, fname);
 823        if (add_excludes(fname, "", 0, el, 0, sha1_stat) < 0)
 824                die("cannot use %s as an exclude file", fname);
 825}
 826
 827void add_excludes_from_file(struct dir_struct *dir, const char *fname)
 828{
 829        dir->unmanaged_exclude_files++; /* see validate_untracked_cache() */
 830        add_excludes_from_file_1(dir, fname, NULL);
 831}
 832
 833int match_basename(const char *basename, int basenamelen,
 834                   const char *pattern, int prefix, int patternlen,
 835                   unsigned flags)
 836{
 837        if (prefix == patternlen) {
 838                if (patternlen == basenamelen &&
 839                    !fspathncmp(pattern, basename, basenamelen))
 840                        return 1;
 841        } else if (flags & EXC_FLAG_ENDSWITH) {
 842                /* "*literal" matching against "fooliteral" */
 843                if (patternlen - 1 <= basenamelen &&
 844                    !fspathncmp(pattern + 1,
 845                                   basename + basenamelen - (patternlen - 1),
 846                                   patternlen - 1))
 847                        return 1;
 848        } else {
 849                if (fnmatch_icase_mem(pattern, patternlen,
 850                                      basename, basenamelen,
 851                                      0) == 0)
 852                        return 1;
 853        }
 854        return 0;
 855}
 856
 857int match_pathname(const char *pathname, int pathlen,
 858                   const char *base, int baselen,
 859                   const char *pattern, int prefix, int patternlen,
 860                   unsigned flags)
 861{
 862        const char *name;
 863        int namelen;
 864
 865        /*
 866         * match with FNM_PATHNAME; the pattern has base implicitly
 867         * in front of it.
 868         */
 869        if (*pattern == '/') {
 870                pattern++;
 871                patternlen--;
 872                prefix--;
 873        }
 874
 875        /*
 876         * baselen does not count the trailing slash. base[] may or
 877         * may not end with a trailing slash though.
 878         */
 879        if (pathlen < baselen + 1 ||
 880            (baselen && pathname[baselen] != '/') ||
 881            fspathncmp(pathname, base, baselen))
 882                return 0;
 883
 884        namelen = baselen ? pathlen - baselen - 1 : pathlen;
 885        name = pathname + pathlen - namelen;
 886
 887        if (prefix) {
 888                /*
 889                 * if the non-wildcard part is longer than the
 890                 * remaining pathname, surely it cannot match.
 891                 */
 892                if (prefix > namelen)
 893                        return 0;
 894
 895                if (fspathncmp(pattern, name, prefix))
 896                        return 0;
 897                pattern += prefix;
 898                patternlen -= prefix;
 899                name    += prefix;
 900                namelen -= prefix;
 901
 902                /*
 903                 * If the whole pattern did not have a wildcard,
 904                 * then our prefix match is all we need; we
 905                 * do not need to call fnmatch at all.
 906                 */
 907                if (!patternlen && !namelen)
 908                        return 1;
 909        }
 910
 911        return fnmatch_icase_mem(pattern, patternlen,
 912                                 name, namelen,
 913                                 WM_PATHNAME) == 0;
 914}
 915
 916/*
 917 * Scan the given exclude list in reverse to see whether pathname
 918 * should be ignored.  The first match (i.e. the last on the list), if
 919 * any, determines the fate.  Returns the exclude_list element which
 920 * matched, or NULL for undecided.
 921 */
 922static struct exclude *last_exclude_matching_from_list(const char *pathname,
 923                                                       int pathlen,
 924                                                       const char *basename,
 925                                                       int *dtype,
 926                                                       struct exclude_list *el)
 927{
 928        struct exclude *exc = NULL; /* undecided */
 929        int i;
 930
 931        if (!el->nr)
 932                return NULL;    /* undefined */
 933
 934        for (i = el->nr - 1; 0 <= i; i--) {
 935                struct exclude *x = el->excludes[i];
 936                const char *exclude = x->pattern;
 937                int prefix = x->nowildcardlen;
 938
 939                if (x->flags & EXC_FLAG_MUSTBEDIR) {
 940                        if (*dtype == DT_UNKNOWN)
 941                                *dtype = get_dtype(NULL, pathname, pathlen);
 942                        if (*dtype != DT_DIR)
 943                                continue;
 944                }
 945
 946                if (x->flags & EXC_FLAG_NODIR) {
 947                        if (match_basename(basename,
 948                                           pathlen - (basename - pathname),
 949                                           exclude, prefix, x->patternlen,
 950                                           x->flags)) {
 951                                exc = x;
 952                                break;
 953                        }
 954                        continue;
 955                }
 956
 957                assert(x->baselen == 0 || x->base[x->baselen - 1] == '/');
 958                if (match_pathname(pathname, pathlen,
 959                                   x->base, x->baselen ? x->baselen - 1 : 0,
 960                                   exclude, prefix, x->patternlen, x->flags)) {
 961                        exc = x;
 962                        break;
 963                }
 964        }
 965        return exc;
 966}
 967
 968/*
 969 * Scan the list and let the last match determine the fate.
 970 * Return 1 for exclude, 0 for include and -1 for undecided.
 971 */
 972int is_excluded_from_list(const char *pathname,
 973                          int pathlen, const char *basename, int *dtype,
 974                          struct exclude_list *el)
 975{
 976        struct exclude *exclude;
 977        exclude = last_exclude_matching_from_list(pathname, pathlen, basename, dtype, el);
 978        if (exclude)
 979                return exclude->flags & EXC_FLAG_NEGATIVE ? 0 : 1;
 980        return -1; /* undecided */
 981}
 982
 983static struct exclude *last_exclude_matching_from_lists(struct dir_struct *dir,
 984                const char *pathname, int pathlen, const char *basename,
 985                int *dtype_p)
 986{
 987        int i, j;
 988        struct exclude_list_group *group;
 989        struct exclude *exclude;
 990        for (i = EXC_CMDL; i <= EXC_FILE; i++) {
 991                group = &dir->exclude_list_group[i];
 992                for (j = group->nr - 1; j >= 0; j--) {
 993                        exclude = last_exclude_matching_from_list(
 994                                pathname, pathlen, basename, dtype_p,
 995                                &group->el[j]);
 996                        if (exclude)
 997                                return exclude;
 998                }
 999        }
1000        return NULL;
1001}
1002
1003/*
1004 * Loads the per-directory exclude list for the substring of base
1005 * which has a char length of baselen.
1006 */
1007static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
1008{
1009        struct exclude_list_group *group;
1010        struct exclude_list *el;
1011        struct exclude_stack *stk = NULL;
1012        struct untracked_cache_dir *untracked;
1013        int current;
1014
1015        group = &dir->exclude_list_group[EXC_DIRS];
1016
1017        /*
1018         * Pop the exclude lists from the EXCL_DIRS exclude_list_group
1019         * which originate from directories not in the prefix of the
1020         * path being checked.
1021         */
1022        while ((stk = dir->exclude_stack) != NULL) {
1023                if (stk->baselen <= baselen &&
1024                    !strncmp(dir->basebuf.buf, base, stk->baselen))
1025                        break;
1026                el = &group->el[dir->exclude_stack->exclude_ix];
1027                dir->exclude_stack = stk->prev;
1028                dir->exclude = NULL;
1029                free((char *)el->src); /* see strbuf_detach() below */
1030                clear_exclude_list(el);
1031                free(stk);
1032                group->nr--;
1033        }
1034
1035        /* Skip traversing into sub directories if the parent is excluded */
1036        if (dir->exclude)
1037                return;
1038
1039        /*
1040         * Lazy initialization. All call sites currently just
1041         * memset(dir, 0, sizeof(*dir)) before use. Changing all of
1042         * them seems lots of work for little benefit.
1043         */
1044        if (!dir->basebuf.buf)
1045                strbuf_init(&dir->basebuf, PATH_MAX);
1046
1047        /* Read from the parent directories and push them down. */
1048        current = stk ? stk->baselen : -1;
1049        strbuf_setlen(&dir->basebuf, current < 0 ? 0 : current);
1050        if (dir->untracked)
1051                untracked = stk ? stk->ucd : dir->untracked->root;
1052        else
1053                untracked = NULL;
1054
1055        while (current < baselen) {
1056                const char *cp;
1057                struct sha1_stat sha1_stat;
1058
1059                stk = xcalloc(1, sizeof(*stk));
1060                if (current < 0) {
1061                        cp = base;
1062                        current = 0;
1063                } else {
1064                        cp = strchr(base + current + 1, '/');
1065                        if (!cp)
1066                                die("oops in prep_exclude");
1067                        cp++;
1068                        untracked =
1069                                lookup_untracked(dir->untracked, untracked,
1070                                                 base + current,
1071                                                 cp - base - current);
1072                }
1073                stk->prev = dir->exclude_stack;
1074                stk->baselen = cp - base;
1075                stk->exclude_ix = group->nr;
1076                stk->ucd = untracked;
1077                el = add_exclude_list(dir, EXC_DIRS, NULL);
1078                strbuf_add(&dir->basebuf, base + current, stk->baselen - current);
1079                assert(stk->baselen == dir->basebuf.len);
1080
1081                /* Abort if the directory is excluded */
1082                if (stk->baselen) {
1083                        int dt = DT_DIR;
1084                        dir->basebuf.buf[stk->baselen - 1] = 0;
1085                        dir->exclude = last_exclude_matching_from_lists(dir,
1086                                dir->basebuf.buf, stk->baselen - 1,
1087                                dir->basebuf.buf + current, &dt);
1088                        dir->basebuf.buf[stk->baselen - 1] = '/';
1089                        if (dir->exclude &&
1090                            dir->exclude->flags & EXC_FLAG_NEGATIVE)
1091                                dir->exclude = NULL;
1092                        if (dir->exclude) {
1093                                dir->exclude_stack = stk;
1094                                return;
1095                        }
1096                }
1097
1098                /* Try to read per-directory file */
1099                hashclr(sha1_stat.sha1);
1100                sha1_stat.valid = 0;
1101                if (dir->exclude_per_dir &&
1102                    /*
1103                     * If we know that no files have been added in
1104                     * this directory (i.e. valid_cached_dir() has
1105                     * been executed and set untracked->valid) ..
1106                     */
1107                    (!untracked || !untracked->valid ||
1108                     /*
1109                      * .. and .gitignore does not exist before
1110                      * (i.e. null exclude_sha1). Then we can skip
1111                      * loading .gitignore, which would result in
1112                      * ENOENT anyway.
1113                      */
1114                     !is_null_sha1(untracked->exclude_sha1))) {
1115                        /*
1116                         * dir->basebuf gets reused by the traversal, but we
1117                         * need fname to remain unchanged to ensure the src
1118                         * member of each struct exclude correctly
1119                         * back-references its source file.  Other invocations
1120                         * of add_exclude_list provide stable strings, so we
1121                         * strbuf_detach() and free() here in the caller.
1122                         */
1123                        struct strbuf sb = STRBUF_INIT;
1124                        strbuf_addbuf(&sb, &dir->basebuf);
1125                        strbuf_addstr(&sb, dir->exclude_per_dir);
1126                        el->src = strbuf_detach(&sb, NULL);
1127                        add_excludes(el->src, el->src, stk->baselen, el, 1,
1128                                     untracked ? &sha1_stat : NULL);
1129                }
1130                /*
1131                 * NEEDSWORK: when untracked cache is enabled, prep_exclude()
1132                 * will first be called in valid_cached_dir() then maybe many
1133                 * times more in last_exclude_matching(). When the cache is
1134                 * used, last_exclude_matching() will not be called and
1135                 * reading .gitignore content will be a waste.
1136                 *
1137                 * So when it's called by valid_cached_dir() and we can get
1138                 * .gitignore SHA-1 from the index (i.e. .gitignore is not
1139                 * modified on work tree), we could delay reading the
1140                 * .gitignore content until we absolutely need it in
1141                 * last_exclude_matching(). Be careful about ignore rule
1142                 * order, though, if you do that.
1143                 */
1144                if (untracked &&
1145                    hashcmp(sha1_stat.sha1, untracked->exclude_sha1)) {
1146                        invalidate_gitignore(dir->untracked, untracked);
1147                        hashcpy(untracked->exclude_sha1, sha1_stat.sha1);
1148                }
1149                dir->exclude_stack = stk;
1150                current = stk->baselen;
1151        }
1152        strbuf_setlen(&dir->basebuf, baselen);
1153}
1154
1155/*
1156 * Loads the exclude lists for the directory containing pathname, then
1157 * scans all exclude lists to determine whether pathname is excluded.
1158 * Returns the exclude_list element which matched, or NULL for
1159 * undecided.
1160 */
1161struct exclude *last_exclude_matching(struct dir_struct *dir,
1162                                             const char *pathname,
1163                                             int *dtype_p)
1164{
1165        int pathlen = strlen(pathname);
1166        const char *basename = strrchr(pathname, '/');
1167        basename = (basename) ? basename+1 : pathname;
1168
1169        prep_exclude(dir, pathname, basename-pathname);
1170
1171        if (dir->exclude)
1172                return dir->exclude;
1173
1174        return last_exclude_matching_from_lists(dir, pathname, pathlen,
1175                        basename, dtype_p);
1176}
1177
1178/*
1179 * Loads the exclude lists for the directory containing pathname, then
1180 * scans all exclude lists to determine whether pathname is excluded.
1181 * Returns 1 if true, otherwise 0.
1182 */
1183int is_excluded(struct dir_struct *dir, const char *pathname, int *dtype_p)
1184{
1185        struct exclude *exclude =
1186                last_exclude_matching(dir, pathname, dtype_p);
1187        if (exclude)
1188                return exclude->flags & EXC_FLAG_NEGATIVE ? 0 : 1;
1189        return 0;
1190}
1191
1192static struct dir_entry *dir_entry_new(const char *pathname, int len)
1193{
1194        struct dir_entry *ent;
1195
1196        FLEX_ALLOC_MEM(ent, name, pathname, len);
1197        ent->len = len;
1198        return ent;
1199}
1200
1201static struct dir_entry *dir_add_name(struct dir_struct *dir, const char *pathname, int len)
1202{
1203        if (cache_file_exists(pathname, len, ignore_case))
1204                return NULL;
1205
1206        ALLOC_GROW(dir->entries, dir->nr+1, dir->alloc);
1207        return dir->entries[dir->nr++] = dir_entry_new(pathname, len);
1208}
1209
1210struct dir_entry *dir_add_ignored(struct dir_struct *dir, const char *pathname, int len)
1211{
1212        if (!cache_name_is_other(pathname, len))
1213                return NULL;
1214
1215        ALLOC_GROW(dir->ignored, dir->ignored_nr+1, dir->ignored_alloc);
1216        return dir->ignored[dir->ignored_nr++] = dir_entry_new(pathname, len);
1217}
1218
1219enum exist_status {
1220        index_nonexistent = 0,
1221        index_directory,
1222        index_gitdir
1223};
1224
1225/*
1226 * Do not use the alphabetically sorted index to look up
1227 * the directory name; instead, use the case insensitive
1228 * directory hash.
1229 */
1230static enum exist_status directory_exists_in_index_icase(const char *dirname, int len)
1231{
1232        struct cache_entry *ce;
1233
1234        if (cache_dir_exists(dirname, len))
1235                return index_directory;
1236
1237        ce = cache_file_exists(dirname, len, ignore_case);
1238        if (ce && S_ISGITLINK(ce->ce_mode))
1239                return index_gitdir;
1240
1241        return index_nonexistent;
1242}
1243
1244/*
1245 * The index sorts alphabetically by entry name, which
1246 * means that a gitlink sorts as '\0' at the end, while
1247 * a directory (which is defined not as an entry, but as
1248 * the files it contains) will sort with the '/' at the
1249 * end.
1250 */
1251static enum exist_status directory_exists_in_index(const char *dirname, int len)
1252{
1253        int pos;
1254
1255        if (ignore_case)
1256                return directory_exists_in_index_icase(dirname, len);
1257
1258        pos = cache_name_pos(dirname, len);
1259        if (pos < 0)
1260                pos = -pos-1;
1261        while (pos < active_nr) {
1262                const struct cache_entry *ce = active_cache[pos++];
1263                unsigned char endchar;
1264
1265                if (strncmp(ce->name, dirname, len))
1266                        break;
1267                endchar = ce->name[len];
1268                if (endchar > '/')
1269                        break;
1270                if (endchar == '/')
1271                        return index_directory;
1272                if (!endchar && S_ISGITLINK(ce->ce_mode))
1273                        return index_gitdir;
1274        }
1275        return index_nonexistent;
1276}
1277
1278/*
1279 * When we find a directory when traversing the filesystem, we
1280 * have three distinct cases:
1281 *
1282 *  - ignore it
1283 *  - see it as a directory
1284 *  - recurse into it
1285 *
1286 * and which one we choose depends on a combination of existing
1287 * git index contents and the flags passed into the directory
1288 * traversal routine.
1289 *
1290 * Case 1: If we *already* have entries in the index under that
1291 * directory name, we always recurse into the directory to see
1292 * all the files.
1293 *
1294 * Case 2: If we *already* have that directory name as a gitlink,
1295 * we always continue to see it as a gitlink, regardless of whether
1296 * there is an actual git directory there or not (it might not
1297 * be checked out as a subproject!)
1298 *
1299 * Case 3: if we didn't have it in the index previously, we
1300 * have a few sub-cases:
1301 *
1302 *  (a) if "show_other_directories" is true, we show it as
1303 *      just a directory, unless "hide_empty_directories" is
1304 *      also true, in which case we need to check if it contains any
1305 *      untracked and / or ignored files.
1306 *  (b) if it looks like a git directory, and we don't have
1307 *      'no_gitlinks' set we treat it as a gitlink, and show it
1308 *      as a directory.
1309 *  (c) otherwise, we recurse into it.
1310 */
1311static enum path_treatment treat_directory(struct dir_struct *dir,
1312        struct untracked_cache_dir *untracked,
1313        const char *dirname, int len, int baselen, int exclude,
1314        const struct path_simplify *simplify)
1315{
1316        /* The "len-1" is to strip the final '/' */
1317        switch (directory_exists_in_index(dirname, len-1)) {
1318        case index_directory:
1319                return path_recurse;
1320
1321        case index_gitdir:
1322                return path_none;
1323
1324        case index_nonexistent:
1325                if (dir->flags & DIR_SHOW_OTHER_DIRECTORIES)
1326                        break;
1327                if (!(dir->flags & DIR_NO_GITLINKS)) {
1328                        unsigned char sha1[20];
1329                        if (resolve_gitlink_ref(dirname, "HEAD", sha1) == 0)
1330                                return path_untracked;
1331                }
1332                return path_recurse;
1333        }
1334
1335        /* This is the "show_other_directories" case */
1336
1337        if (!(dir->flags & DIR_HIDE_EMPTY_DIRECTORIES))
1338                return exclude ? path_excluded : path_untracked;
1339
1340        untracked = lookup_untracked(dir->untracked, untracked,
1341                                     dirname + baselen, len - baselen);
1342        return read_directory_recursive(dir, dirname, len,
1343                                        untracked, 1, simplify);
1344}
1345
1346/*
1347 * This is an inexact early pruning of any recursive directory
1348 * reading - if the path cannot possibly be in the pathspec,
1349 * return true, and we'll skip it early.
1350 */
1351static int simplify_away(const char *path, int pathlen, const struct path_simplify *simplify)
1352{
1353        if (simplify) {
1354                for (;;) {
1355                        const char *match = simplify->path;
1356                        int len = simplify->len;
1357
1358                        if (!match)
1359                                break;
1360                        if (len > pathlen)
1361                                len = pathlen;
1362                        if (!memcmp(path, match, len))
1363                                return 0;
1364                        simplify++;
1365                }
1366                return 1;
1367        }
1368        return 0;
1369}
1370
1371/*
1372 * This function tells us whether an excluded path matches a
1373 * list of "interesting" pathspecs. That is, whether a path matched
1374 * by any of the pathspecs could possibly be ignored by excluding
1375 * the specified path. This can happen if:
1376 *
1377 *   1. the path is mentioned explicitly in the pathspec
1378 *
1379 *   2. the path is a directory prefix of some element in the
1380 *      pathspec
1381 */
1382static int exclude_matches_pathspec(const char *path, int len,
1383                const struct path_simplify *simplify)
1384{
1385        if (simplify) {
1386                for (; simplify->path; simplify++) {
1387                        if (len == simplify->len
1388                            && !memcmp(path, simplify->path, len))
1389                                return 1;
1390                        if (len < simplify->len
1391                            && simplify->path[len] == '/'
1392                            && !memcmp(path, simplify->path, len))
1393                                return 1;
1394                }
1395        }
1396        return 0;
1397}
1398
1399static int get_index_dtype(const char *path, int len)
1400{
1401        int pos;
1402        const struct cache_entry *ce;
1403
1404        ce = cache_file_exists(path, len, 0);
1405        if (ce) {
1406                if (!ce_uptodate(ce))
1407                        return DT_UNKNOWN;
1408                if (S_ISGITLINK(ce->ce_mode))
1409                        return DT_DIR;
1410                /*
1411                 * Nobody actually cares about the
1412                 * difference between DT_LNK and DT_REG
1413                 */
1414                return DT_REG;
1415        }
1416
1417        /* Try to look it up as a directory */
1418        pos = cache_name_pos(path, len);
1419        if (pos >= 0)
1420                return DT_UNKNOWN;
1421        pos = -pos-1;
1422        while (pos < active_nr) {
1423                ce = active_cache[pos++];
1424                if (strncmp(ce->name, path, len))
1425                        break;
1426                if (ce->name[len] > '/')
1427                        break;
1428                if (ce->name[len] < '/')
1429                        continue;
1430                if (!ce_uptodate(ce))
1431                        break;  /* continue? */
1432                return DT_DIR;
1433        }
1434        return DT_UNKNOWN;
1435}
1436
1437static int get_dtype(struct dirent *de, const char *path, int len)
1438{
1439        int dtype = de ? DTYPE(de) : DT_UNKNOWN;
1440        struct stat st;
1441
1442        if (dtype != DT_UNKNOWN)
1443                return dtype;
1444        dtype = get_index_dtype(path, len);
1445        if (dtype != DT_UNKNOWN)
1446                return dtype;
1447        if (lstat(path, &st))
1448                return dtype;
1449        if (S_ISREG(st.st_mode))
1450                return DT_REG;
1451        if (S_ISDIR(st.st_mode))
1452                return DT_DIR;
1453        if (S_ISLNK(st.st_mode))
1454                return DT_LNK;
1455        return dtype;
1456}
1457
1458static enum path_treatment treat_one_path(struct dir_struct *dir,
1459                                          struct untracked_cache_dir *untracked,
1460                                          struct strbuf *path,
1461                                          int baselen,
1462                                          const struct path_simplify *simplify,
1463                                          int dtype, struct dirent *de)
1464{
1465        int exclude;
1466        int has_path_in_index = !!cache_file_exists(path->buf, path->len, ignore_case);
1467
1468        if (dtype == DT_UNKNOWN)
1469                dtype = get_dtype(de, path->buf, path->len);
1470
1471        /* Always exclude indexed files */
1472        if (dtype != DT_DIR && has_path_in_index)
1473                return path_none;
1474
1475        /*
1476         * When we are looking at a directory P in the working tree,
1477         * there are three cases:
1478         *
1479         * (1) P exists in the index.  Everything inside the directory P in
1480         * the working tree needs to go when P is checked out from the
1481         * index.
1482         *
1483         * (2) P does not exist in the index, but there is P/Q in the index.
1484         * We know P will stay a directory when we check out the contents
1485         * of the index, but we do not know yet if there is a directory
1486         * P/Q in the working tree to be killed, so we need to recurse.
1487         *
1488         * (3) P does not exist in the index, and there is no P/Q in the index
1489         * to require P to be a directory, either.  Only in this case, we
1490         * know that everything inside P will not be killed without
1491         * recursing.
1492         */
1493        if ((dir->flags & DIR_COLLECT_KILLED_ONLY) &&
1494            (dtype == DT_DIR) &&
1495            !has_path_in_index &&
1496            (directory_exists_in_index(path->buf, path->len) == index_nonexistent))
1497                return path_none;
1498
1499        exclude = is_excluded(dir, path->buf, &dtype);
1500
1501        /*
1502         * Excluded? If we don't explicitly want to show
1503         * ignored files, ignore it
1504         */
1505        if (exclude && !(dir->flags & (DIR_SHOW_IGNORED|DIR_SHOW_IGNORED_TOO)))
1506                return path_excluded;
1507
1508        switch (dtype) {
1509        default:
1510                return path_none;
1511        case DT_DIR:
1512                strbuf_addch(path, '/');
1513                return treat_directory(dir, untracked, path->buf, path->len,
1514                                       baselen, exclude, simplify);
1515        case DT_REG:
1516        case DT_LNK:
1517                return exclude ? path_excluded : path_untracked;
1518        }
1519}
1520
1521static enum path_treatment treat_path_fast(struct dir_struct *dir,
1522                                           struct untracked_cache_dir *untracked,
1523                                           struct cached_dir *cdir,
1524                                           struct strbuf *path,
1525                                           int baselen,
1526                                           const struct path_simplify *simplify)
1527{
1528        strbuf_setlen(path, baselen);
1529        if (!cdir->ucd) {
1530                strbuf_addstr(path, cdir->file);
1531                return path_untracked;
1532        }
1533        strbuf_addstr(path, cdir->ucd->name);
1534        /* treat_one_path() does this before it calls treat_directory() */
1535        strbuf_complete(path, '/');
1536        if (cdir->ucd->check_only)
1537                /*
1538                 * check_only is set as a result of treat_directory() getting
1539                 * to its bottom. Verify again the same set of directories
1540                 * with check_only set.
1541                 */
1542                return read_directory_recursive(dir, path->buf, path->len,
1543                                                cdir->ucd, 1, simplify);
1544        /*
1545         * We get path_recurse in the first run when
1546         * directory_exists_in_index() returns index_nonexistent. We
1547         * are sure that new changes in the index does not impact the
1548         * outcome. Return now.
1549         */
1550        return path_recurse;
1551}
1552
1553static enum path_treatment treat_path(struct dir_struct *dir,
1554                                      struct untracked_cache_dir *untracked,
1555                                      struct cached_dir *cdir,
1556                                      struct strbuf *path,
1557                                      int baselen,
1558                                      const struct path_simplify *simplify)
1559{
1560        int dtype;
1561        struct dirent *de = cdir->de;
1562
1563        if (!de)
1564                return treat_path_fast(dir, untracked, cdir, path,
1565                                       baselen, simplify);
1566        if (is_dot_or_dotdot(de->d_name) || !strcmp(de->d_name, ".git"))
1567                return path_none;
1568        strbuf_setlen(path, baselen);
1569        strbuf_addstr(path, de->d_name);
1570        if (simplify_away(path->buf, path->len, simplify))
1571                return path_none;
1572
1573        dtype = DTYPE(de);
1574        return treat_one_path(dir, untracked, path, baselen, simplify, dtype, de);
1575}
1576
1577static void add_untracked(struct untracked_cache_dir *dir, const char *name)
1578{
1579        if (!dir)
1580                return;
1581        ALLOC_GROW(dir->untracked, dir->untracked_nr + 1,
1582                   dir->untracked_alloc);
1583        dir->untracked[dir->untracked_nr++] = xstrdup(name);
1584}
1585
1586static int valid_cached_dir(struct dir_struct *dir,
1587                            struct untracked_cache_dir *untracked,
1588                            struct strbuf *path,
1589                            int check_only)
1590{
1591        struct stat st;
1592
1593        if (!untracked)
1594                return 0;
1595
1596        if (stat(path->len ? path->buf : ".", &st)) {
1597                invalidate_directory(dir->untracked, untracked);
1598                memset(&untracked->stat_data, 0, sizeof(untracked->stat_data));
1599                return 0;
1600        }
1601        if (!untracked->valid ||
1602            match_stat_data_racy(&the_index, &untracked->stat_data, &st)) {
1603                if (untracked->valid)
1604                        invalidate_directory(dir->untracked, untracked);
1605                fill_stat_data(&untracked->stat_data, &st);
1606                return 0;
1607        }
1608
1609        if (untracked->check_only != !!check_only) {
1610                invalidate_directory(dir->untracked, untracked);
1611                return 0;
1612        }
1613
1614        /*
1615         * prep_exclude will be called eventually on this directory,
1616         * but it's called much later in last_exclude_matching(). We
1617         * need it now to determine the validity of the cache for this
1618         * path. The next calls will be nearly no-op, the way
1619         * prep_exclude() is designed.
1620         */
1621        if (path->len && path->buf[path->len - 1] != '/') {
1622                strbuf_addch(path, '/');
1623                prep_exclude(dir, path->buf, path->len);
1624                strbuf_setlen(path, path->len - 1);
1625        } else
1626                prep_exclude(dir, path->buf, path->len);
1627
1628        /* hopefully prep_exclude() haven't invalidated this entry... */
1629        return untracked->valid;
1630}
1631
1632static int open_cached_dir(struct cached_dir *cdir,
1633                           struct dir_struct *dir,
1634                           struct untracked_cache_dir *untracked,
1635                           struct strbuf *path,
1636                           int check_only)
1637{
1638        memset(cdir, 0, sizeof(*cdir));
1639        cdir->untracked = untracked;
1640        if (valid_cached_dir(dir, untracked, path, check_only))
1641                return 0;
1642        cdir->fdir = opendir(path->len ? path->buf : ".");
1643        if (dir->untracked)
1644                dir->untracked->dir_opened++;
1645        if (!cdir->fdir)
1646                return -1;
1647        return 0;
1648}
1649
1650static int read_cached_dir(struct cached_dir *cdir)
1651{
1652        if (cdir->fdir) {
1653                cdir->de = readdir(cdir->fdir);
1654                if (!cdir->de)
1655                        return -1;
1656                return 0;
1657        }
1658        while (cdir->nr_dirs < cdir->untracked->dirs_nr) {
1659                struct untracked_cache_dir *d = cdir->untracked->dirs[cdir->nr_dirs];
1660                if (!d->recurse) {
1661                        cdir->nr_dirs++;
1662                        continue;
1663                }
1664                cdir->ucd = d;
1665                cdir->nr_dirs++;
1666                return 0;
1667        }
1668        cdir->ucd = NULL;
1669        if (cdir->nr_files < cdir->untracked->untracked_nr) {
1670                struct untracked_cache_dir *d = cdir->untracked;
1671                cdir->file = d->untracked[cdir->nr_files++];
1672                return 0;
1673        }
1674        return -1;
1675}
1676
1677static void close_cached_dir(struct cached_dir *cdir)
1678{
1679        if (cdir->fdir)
1680                closedir(cdir->fdir);
1681        /*
1682         * We have gone through this directory and found no untracked
1683         * entries. Mark it valid.
1684         */
1685        if (cdir->untracked) {
1686                cdir->untracked->valid = 1;
1687                cdir->untracked->recurse = 1;
1688        }
1689}
1690
1691/*
1692 * Read a directory tree. We currently ignore anything but
1693 * directories, regular files and symlinks. That's because git
1694 * doesn't handle them at all yet. Maybe that will change some
1695 * day.
1696 *
1697 * Also, we ignore the name ".git" (even if it is not a directory).
1698 * That likely will not change.
1699 *
1700 * Returns the most significant path_treatment value encountered in the scan.
1701 */
1702static enum path_treatment read_directory_recursive(struct dir_struct *dir,
1703                                    const char *base, int baselen,
1704                                    struct untracked_cache_dir *untracked, int check_only,
1705                                    const struct path_simplify *simplify)
1706{
1707        struct cached_dir cdir;
1708        enum path_treatment state, subdir_state, dir_state = path_none;
1709        struct strbuf path = STRBUF_INIT;
1710
1711        strbuf_add(&path, base, baselen);
1712
1713        if (open_cached_dir(&cdir, dir, untracked, &path, check_only))
1714                goto out;
1715
1716        if (untracked)
1717                untracked->check_only = !!check_only;
1718
1719        while (!read_cached_dir(&cdir)) {
1720                /* check how the file or directory should be treated */
1721                state = treat_path(dir, untracked, &cdir, &path, baselen, simplify);
1722
1723                if (state > dir_state)
1724                        dir_state = state;
1725
1726                /* recurse into subdir if instructed by treat_path */
1727                if (state == path_recurse) {
1728                        struct untracked_cache_dir *ud;
1729                        ud = lookup_untracked(dir->untracked, untracked,
1730                                              path.buf + baselen,
1731                                              path.len - baselen);
1732                        subdir_state =
1733                                read_directory_recursive(dir, path.buf, path.len,
1734                                                         ud, check_only, simplify);
1735                        if (subdir_state > dir_state)
1736                                dir_state = subdir_state;
1737                }
1738
1739                if (check_only) {
1740                        /* abort early if maximum state has been reached */
1741                        if (dir_state == path_untracked) {
1742                                if (cdir.fdir)
1743                                        add_untracked(untracked, path.buf + baselen);
1744                                break;
1745                        }
1746                        /* skip the dir_add_* part */
1747                        continue;
1748                }
1749
1750                /* add the path to the appropriate result list */
1751                switch (state) {
1752                case path_excluded:
1753                        if (dir->flags & DIR_SHOW_IGNORED)
1754                                dir_add_name(dir, path.buf, path.len);
1755                        else if ((dir->flags & DIR_SHOW_IGNORED_TOO) ||
1756                                ((dir->flags & DIR_COLLECT_IGNORED) &&
1757                                exclude_matches_pathspec(path.buf, path.len,
1758                                        simplify)))
1759                                dir_add_ignored(dir, path.buf, path.len);
1760                        break;
1761
1762                case path_untracked:
1763                        if (dir->flags & DIR_SHOW_IGNORED)
1764                                break;
1765                        dir_add_name(dir, path.buf, path.len);
1766                        if (cdir.fdir)
1767                                add_untracked(untracked, path.buf + baselen);
1768                        break;
1769
1770                default:
1771                        break;
1772                }
1773        }
1774        close_cached_dir(&cdir);
1775 out:
1776        strbuf_release(&path);
1777
1778        return dir_state;
1779}
1780
1781static int cmp_name(const void *p1, const void *p2)
1782{
1783        const struct dir_entry *e1 = *(const struct dir_entry **)p1;
1784        const struct dir_entry *e2 = *(const struct dir_entry **)p2;
1785
1786        return name_compare(e1->name, e1->len, e2->name, e2->len);
1787}
1788
1789static struct path_simplify *create_simplify(const char **pathspec)
1790{
1791        int nr, alloc = 0;
1792        struct path_simplify *simplify = NULL;
1793
1794        if (!pathspec)
1795                return NULL;
1796
1797        for (nr = 0 ; ; nr++) {
1798                const char *match;
1799                ALLOC_GROW(simplify, nr + 1, alloc);
1800                match = *pathspec++;
1801                if (!match)
1802                        break;
1803                simplify[nr].path = match;
1804                simplify[nr].len = simple_length(match);
1805        }
1806        simplify[nr].path = NULL;
1807        simplify[nr].len = 0;
1808        return simplify;
1809}
1810
1811static void free_simplify(struct path_simplify *simplify)
1812{
1813        free(simplify);
1814}
1815
1816static int treat_leading_path(struct dir_struct *dir,
1817                              const char *path, int len,
1818                              const struct path_simplify *simplify)
1819{
1820        struct strbuf sb = STRBUF_INIT;
1821        int baselen, rc = 0;
1822        const char *cp;
1823        int old_flags = dir->flags;
1824
1825        while (len && path[len - 1] == '/')
1826                len--;
1827        if (!len)
1828                return 1;
1829        baselen = 0;
1830        dir->flags &= ~DIR_SHOW_OTHER_DIRECTORIES;
1831        while (1) {
1832                cp = path + baselen + !!baselen;
1833                cp = memchr(cp, '/', path + len - cp);
1834                if (!cp)
1835                        baselen = len;
1836                else
1837                        baselen = cp - path;
1838                strbuf_setlen(&sb, 0);
1839                strbuf_add(&sb, path, baselen);
1840                if (!is_directory(sb.buf))
1841                        break;
1842                if (simplify_away(sb.buf, sb.len, simplify))
1843                        break;
1844                if (treat_one_path(dir, NULL, &sb, baselen, simplify,
1845                                   DT_DIR, NULL) == path_none)
1846                        break; /* do not recurse into it */
1847                if (len <= baselen) {
1848                        rc = 1;
1849                        break; /* finished checking */
1850                }
1851        }
1852        strbuf_release(&sb);
1853        dir->flags = old_flags;
1854        return rc;
1855}
1856
1857static const char *get_ident_string(void)
1858{
1859        static struct strbuf sb = STRBUF_INIT;
1860        struct utsname uts;
1861
1862        if (sb.len)
1863                return sb.buf;
1864        if (uname(&uts) < 0)
1865                die_errno(_("failed to get kernel name and information"));
1866        strbuf_addf(&sb, "Location %s, system %s", get_git_work_tree(),
1867                    uts.sysname);
1868        return sb.buf;
1869}
1870
1871static int ident_in_untracked(const struct untracked_cache *uc)
1872{
1873        /*
1874         * Previous git versions may have saved many NUL separated
1875         * strings in the "ident" field, but it is insane to manage
1876         * many locations, so just take care of the first one.
1877         */
1878
1879        return !strcmp(uc->ident.buf, get_ident_string());
1880}
1881
1882static void set_untracked_ident(struct untracked_cache *uc)
1883{
1884        strbuf_reset(&uc->ident);
1885        strbuf_addstr(&uc->ident, get_ident_string());
1886
1887        /*
1888         * This strbuf used to contain a list of NUL separated
1889         * strings, so save NUL too for backward compatibility.
1890         */
1891        strbuf_addch(&uc->ident, 0);
1892}
1893
1894static void new_untracked_cache(struct index_state *istate)
1895{
1896        struct untracked_cache *uc = xcalloc(1, sizeof(*uc));
1897        strbuf_init(&uc->ident, 100);
1898        uc->exclude_per_dir = ".gitignore";
1899        /* should be the same flags used by git-status */
1900        uc->dir_flags = DIR_SHOW_OTHER_DIRECTORIES | DIR_HIDE_EMPTY_DIRECTORIES;
1901        set_untracked_ident(uc);
1902        istate->untracked = uc;
1903        istate->cache_changed |= UNTRACKED_CHANGED;
1904}
1905
1906void add_untracked_cache(struct index_state *istate)
1907{
1908        if (!istate->untracked) {
1909                new_untracked_cache(istate);
1910        } else {
1911                if (!ident_in_untracked(istate->untracked)) {
1912                        free_untracked_cache(istate->untracked);
1913                        new_untracked_cache(istate);
1914                }
1915        }
1916}
1917
1918void remove_untracked_cache(struct index_state *istate)
1919{
1920        if (istate->untracked) {
1921                free_untracked_cache(istate->untracked);
1922                istate->untracked = NULL;
1923                istate->cache_changed |= UNTRACKED_CHANGED;
1924        }
1925}
1926
1927static struct untracked_cache_dir *validate_untracked_cache(struct dir_struct *dir,
1928                                                      int base_len,
1929                                                      const struct pathspec *pathspec)
1930{
1931        struct untracked_cache_dir *root;
1932
1933        if (!dir->untracked || getenv("GIT_DISABLE_UNTRACKED_CACHE"))
1934                return NULL;
1935
1936        /*
1937         * We only support $GIT_DIR/info/exclude and core.excludesfile
1938         * as the global ignore rule files. Any other additions
1939         * (e.g. from command line) invalidate the cache. This
1940         * condition also catches running setup_standard_excludes()
1941         * before setting dir->untracked!
1942         */
1943        if (dir->unmanaged_exclude_files)
1944                return NULL;
1945
1946        /*
1947         * Optimize for the main use case only: whole-tree git
1948         * status. More work involved in treat_leading_path() if we
1949         * use cache on just a subset of the worktree. pathspec
1950         * support could make the matter even worse.
1951         */
1952        if (base_len || (pathspec && pathspec->nr))
1953                return NULL;
1954
1955        /* Different set of flags may produce different results */
1956        if (dir->flags != dir->untracked->dir_flags ||
1957            /*
1958             * See treat_directory(), case index_nonexistent. Without
1959             * this flag, we may need to also cache .git file content
1960             * for the resolve_gitlink_ref() call, which we don't.
1961             */
1962            !(dir->flags & DIR_SHOW_OTHER_DIRECTORIES) ||
1963            /* We don't support collecting ignore files */
1964            (dir->flags & (DIR_SHOW_IGNORED | DIR_SHOW_IGNORED_TOO |
1965                           DIR_COLLECT_IGNORED)))
1966                return NULL;
1967
1968        /*
1969         * If we use .gitignore in the cache and now you change it to
1970         * .gitexclude, everything will go wrong.
1971         */
1972        if (dir->exclude_per_dir != dir->untracked->exclude_per_dir &&
1973            strcmp(dir->exclude_per_dir, dir->untracked->exclude_per_dir))
1974                return NULL;
1975
1976        /*
1977         * EXC_CMDL is not considered in the cache. If people set it,
1978         * skip the cache.
1979         */
1980        if (dir->exclude_list_group[EXC_CMDL].nr)
1981                return NULL;
1982
1983        if (!ident_in_untracked(dir->untracked)) {
1984                warning(_("Untracked cache is disabled on this system or location."));
1985                return NULL;
1986        }
1987
1988        if (!dir->untracked->root) {
1989                const int len = sizeof(*dir->untracked->root);
1990                dir->untracked->root = xmalloc(len);
1991                memset(dir->untracked->root, 0, len);
1992        }
1993
1994        /* Validate $GIT_DIR/info/exclude and core.excludesfile */
1995        root = dir->untracked->root;
1996        if (hashcmp(dir->ss_info_exclude.sha1,
1997                    dir->untracked->ss_info_exclude.sha1)) {
1998                invalidate_gitignore(dir->untracked, root);
1999                dir->untracked->ss_info_exclude = dir->ss_info_exclude;
2000        }
2001        if (hashcmp(dir->ss_excludes_file.sha1,
2002                    dir->untracked->ss_excludes_file.sha1)) {
2003                invalidate_gitignore(dir->untracked, root);
2004                dir->untracked->ss_excludes_file = dir->ss_excludes_file;
2005        }
2006
2007        /* Make sure this directory is not dropped out at saving phase */
2008        root->recurse = 1;
2009        return root;
2010}
2011
2012int read_directory(struct dir_struct *dir, const char *path, int len, const struct pathspec *pathspec)
2013{
2014        struct path_simplify *simplify;
2015        struct untracked_cache_dir *untracked;
2016
2017        /*
2018         * Check out create_simplify()
2019         */
2020        if (pathspec)
2021                GUARD_PATHSPEC(pathspec,
2022                               PATHSPEC_FROMTOP |
2023                               PATHSPEC_MAXDEPTH |
2024                               PATHSPEC_LITERAL |
2025                               PATHSPEC_GLOB |
2026                               PATHSPEC_ICASE |
2027                               PATHSPEC_EXCLUDE);
2028
2029        if (has_symlink_leading_path(path, len))
2030                return dir->nr;
2031
2032        /*
2033         * exclude patterns are treated like positive ones in
2034         * create_simplify. Usually exclude patterns should be a
2035         * subset of positive ones, which has no impacts on
2036         * create_simplify().
2037         */
2038        simplify = create_simplify(pathspec ? pathspec->_raw : NULL);
2039        untracked = validate_untracked_cache(dir, len, pathspec);
2040        if (!untracked)
2041                /*
2042                 * make sure untracked cache code path is disabled,
2043                 * e.g. prep_exclude()
2044                 */
2045                dir->untracked = NULL;
2046        if (!len || treat_leading_path(dir, path, len, simplify))
2047                read_directory_recursive(dir, path, len, untracked, 0, simplify);
2048        free_simplify(simplify);
2049        qsort(dir->entries, dir->nr, sizeof(struct dir_entry *), cmp_name);
2050        qsort(dir->ignored, dir->ignored_nr, sizeof(struct dir_entry *), cmp_name);
2051        if (dir->untracked) {
2052                static struct trace_key trace_untracked_stats = TRACE_KEY_INIT(UNTRACKED_STATS);
2053                trace_printf_key(&trace_untracked_stats,
2054                                 "node creation: %u\n"
2055                                 "gitignore invalidation: %u\n"
2056                                 "directory invalidation: %u\n"
2057                                 "opendir: %u\n",
2058                                 dir->untracked->dir_created,
2059                                 dir->untracked->gitignore_invalidated,
2060                                 dir->untracked->dir_invalidated,
2061                                 dir->untracked->dir_opened);
2062                if (dir->untracked == the_index.untracked &&
2063                    (dir->untracked->dir_opened ||
2064                     dir->untracked->gitignore_invalidated ||
2065                     dir->untracked->dir_invalidated))
2066                        the_index.cache_changed |= UNTRACKED_CHANGED;
2067                if (dir->untracked != the_index.untracked) {
2068                        free(dir->untracked);
2069                        dir->untracked = NULL;
2070                }
2071        }
2072        return dir->nr;
2073}
2074
2075int file_exists(const char *f)
2076{
2077        struct stat sb;
2078        return lstat(f, &sb) == 0;
2079}
2080
2081static int cmp_icase(char a, char b)
2082{
2083        if (a == b)
2084                return 0;
2085        if (ignore_case)
2086                return toupper(a) - toupper(b);
2087        return a - b;
2088}
2089
2090/*
2091 * Given two normalized paths (a trailing slash is ok), if subdir is
2092 * outside dir, return -1.  Otherwise return the offset in subdir that
2093 * can be used as relative path to dir.
2094 */
2095int dir_inside_of(const char *subdir, const char *dir)
2096{
2097        int offset = 0;
2098
2099        assert(dir && subdir && *dir && *subdir);
2100
2101        while (*dir && *subdir && !cmp_icase(*dir, *subdir)) {
2102                dir++;
2103                subdir++;
2104                offset++;
2105        }
2106
2107        /* hel[p]/me vs hel[l]/yeah */
2108        if (*dir && *subdir)
2109                return -1;
2110
2111        if (!*subdir)
2112                return !*dir ? offset : -1; /* same dir */
2113
2114        /* foo/[b]ar vs foo/[] */
2115        if (is_dir_sep(dir[-1]))
2116                return is_dir_sep(subdir[-1]) ? offset : -1;
2117
2118        /* foo[/]bar vs foo[] */
2119        return is_dir_sep(*subdir) ? offset + 1 : -1;
2120}
2121
2122int is_inside_dir(const char *dir)
2123{
2124        char *cwd;
2125        int rc;
2126
2127        if (!dir)
2128                return 0;
2129
2130        cwd = xgetcwd();
2131        rc = (dir_inside_of(cwd, dir) >= 0);
2132        free(cwd);
2133        return rc;
2134}
2135
2136int is_empty_dir(const char *path)
2137{
2138        DIR *dir = opendir(path);
2139        struct dirent *e;
2140        int ret = 1;
2141
2142        if (!dir)
2143                return 0;
2144
2145        while ((e = readdir(dir)) != NULL)
2146                if (!is_dot_or_dotdot(e->d_name)) {
2147                        ret = 0;
2148                        break;
2149                }
2150
2151        closedir(dir);
2152        return ret;
2153}
2154
2155static int remove_dir_recurse(struct strbuf *path, int flag, int *kept_up)
2156{
2157        DIR *dir;
2158        struct dirent *e;
2159        int ret = 0, original_len = path->len, len, kept_down = 0;
2160        int only_empty = (flag & REMOVE_DIR_EMPTY_ONLY);
2161        int keep_toplevel = (flag & REMOVE_DIR_KEEP_TOPLEVEL);
2162        unsigned char submodule_head[20];
2163
2164        if ((flag & REMOVE_DIR_KEEP_NESTED_GIT) &&
2165            !resolve_gitlink_ref(path->buf, "HEAD", submodule_head)) {
2166                /* Do not descend and nuke a nested git work tree. */
2167                if (kept_up)
2168                        *kept_up = 1;
2169                return 0;
2170        }
2171
2172        flag &= ~REMOVE_DIR_KEEP_TOPLEVEL;
2173        dir = opendir(path->buf);
2174        if (!dir) {
2175                if (errno == ENOENT)
2176                        return keep_toplevel ? -1 : 0;
2177                else if (errno == EACCES && !keep_toplevel)
2178                        /*
2179                         * An empty dir could be removable even if it
2180                         * is unreadable:
2181                         */
2182                        return rmdir(path->buf);
2183                else
2184                        return -1;
2185        }
2186        strbuf_complete(path, '/');
2187
2188        len = path->len;
2189        while ((e = readdir(dir)) != NULL) {
2190                struct stat st;
2191                if (is_dot_or_dotdot(e->d_name))
2192                        continue;
2193
2194                strbuf_setlen(path, len);
2195                strbuf_addstr(path, e->d_name);
2196                if (lstat(path->buf, &st)) {
2197                        if (errno == ENOENT)
2198                                /*
2199                                 * file disappeared, which is what we
2200                                 * wanted anyway
2201                                 */
2202                                continue;
2203                        /* fall thru */
2204                } else if (S_ISDIR(st.st_mode)) {
2205                        if (!remove_dir_recurse(path, flag, &kept_down))
2206                                continue; /* happy */
2207                } else if (!only_empty &&
2208                           (!unlink(path->buf) || errno == ENOENT)) {
2209                        continue; /* happy, too */
2210                }
2211
2212                /* path too long, stat fails, or non-directory still exists */
2213                ret = -1;
2214                break;
2215        }
2216        closedir(dir);
2217
2218        strbuf_setlen(path, original_len);
2219        if (!ret && !keep_toplevel && !kept_down)
2220                ret = (!rmdir(path->buf) || errno == ENOENT) ? 0 : -1;
2221        else if (kept_up)
2222                /*
2223                 * report the uplevel that it is not an error that we
2224                 * did not rmdir() our directory.
2225                 */
2226                *kept_up = !ret;
2227        return ret;
2228}
2229
2230int remove_dir_recursively(struct strbuf *path, int flag)
2231{
2232        return remove_dir_recurse(path, flag, NULL);
2233}
2234
2235static GIT_PATH_FUNC(git_path_info_exclude, "info/exclude")
2236
2237void setup_standard_excludes(struct dir_struct *dir)
2238{
2239        const char *path;
2240
2241        dir->exclude_per_dir = ".gitignore";
2242
2243        /* core.excludefile defaulting to $XDG_HOME/git/ignore */
2244        if (!excludes_file)
2245                excludes_file = xdg_config_home("ignore");
2246        if (excludes_file && !access_or_warn(excludes_file, R_OK, 0))
2247                add_excludes_from_file_1(dir, excludes_file,
2248                                         dir->untracked ? &dir->ss_excludes_file : NULL);
2249
2250        /* per repository user preference */
2251        path = git_path_info_exclude();
2252        if (!access_or_warn(path, R_OK, 0))
2253                add_excludes_from_file_1(dir, path,
2254                                         dir->untracked ? &dir->ss_info_exclude : NULL);
2255}
2256
2257int remove_path(const char *name)
2258{
2259        char *slash;
2260
2261        if (unlink(name) && errno != ENOENT && errno != ENOTDIR)
2262                return -1;
2263
2264        slash = strrchr(name, '/');
2265        if (slash) {
2266                char *dirs = xstrdup(name);
2267                slash = dirs + (slash - name);
2268                do {
2269                        *slash = '\0';
2270                } while (rmdir(dirs) == 0 && (slash = strrchr(dirs, '/')));
2271                free(dirs);
2272        }
2273        return 0;
2274}
2275
2276/*
2277 * Frees memory within dir which was allocated for exclude lists and
2278 * the exclude_stack.  Does not free dir itself.
2279 */
2280void clear_directory(struct dir_struct *dir)
2281{
2282        int i, j;
2283        struct exclude_list_group *group;
2284        struct exclude_list *el;
2285        struct exclude_stack *stk;
2286
2287        for (i = EXC_CMDL; i <= EXC_FILE; i++) {
2288                group = &dir->exclude_list_group[i];
2289                for (j = 0; j < group->nr; j++) {
2290                        el = &group->el[j];
2291                        if (i == EXC_DIRS)
2292                                free((char *)el->src);
2293                        clear_exclude_list(el);
2294                }
2295                free(group->el);
2296        }
2297
2298        stk = dir->exclude_stack;
2299        while (stk) {
2300                struct exclude_stack *prev = stk->prev;
2301                free(stk);
2302                stk = prev;
2303        }
2304        strbuf_release(&dir->basebuf);
2305}
2306
2307struct ondisk_untracked_cache {
2308        struct stat_data info_exclude_stat;
2309        struct stat_data excludes_file_stat;
2310        uint32_t dir_flags;
2311        unsigned char info_exclude_sha1[20];
2312        unsigned char excludes_file_sha1[20];
2313        char exclude_per_dir[FLEX_ARRAY];
2314};
2315
2316#define ouc_size(len) (offsetof(struct ondisk_untracked_cache, exclude_per_dir) + len + 1)
2317
2318struct write_data {
2319        int index;         /* number of written untracked_cache_dir */
2320        struct ewah_bitmap *check_only; /* from untracked_cache_dir */
2321        struct ewah_bitmap *valid;      /* from untracked_cache_dir */
2322        struct ewah_bitmap *sha1_valid; /* set if exclude_sha1 is not null */
2323        struct strbuf out;
2324        struct strbuf sb_stat;
2325        struct strbuf sb_sha1;
2326};
2327
2328static void stat_data_to_disk(struct stat_data *to, const struct stat_data *from)
2329{
2330        to->sd_ctime.sec  = htonl(from->sd_ctime.sec);
2331        to->sd_ctime.nsec = htonl(from->sd_ctime.nsec);
2332        to->sd_mtime.sec  = htonl(from->sd_mtime.sec);
2333        to->sd_mtime.nsec = htonl(from->sd_mtime.nsec);
2334        to->sd_dev        = htonl(from->sd_dev);
2335        to->sd_ino        = htonl(from->sd_ino);
2336        to->sd_uid        = htonl(from->sd_uid);
2337        to->sd_gid        = htonl(from->sd_gid);
2338        to->sd_size       = htonl(from->sd_size);
2339}
2340
2341static void write_one_dir(struct untracked_cache_dir *untracked,
2342                          struct write_data *wd)
2343{
2344        struct stat_data stat_data;
2345        struct strbuf *out = &wd->out;
2346        unsigned char intbuf[16];
2347        unsigned int intlen, value;
2348        int i = wd->index++;
2349
2350        /*
2351         * untracked_nr should be reset whenever valid is clear, but
2352         * for safety..
2353         */
2354        if (!untracked->valid) {
2355                untracked->untracked_nr = 0;
2356                untracked->check_only = 0;
2357        }
2358
2359        if (untracked->check_only)
2360                ewah_set(wd->check_only, i);
2361        if (untracked->valid) {
2362                ewah_set(wd->valid, i);
2363                stat_data_to_disk(&stat_data, &untracked->stat_data);
2364                strbuf_add(&wd->sb_stat, &stat_data, sizeof(stat_data));
2365        }
2366        if (!is_null_sha1(untracked->exclude_sha1)) {
2367                ewah_set(wd->sha1_valid, i);
2368                strbuf_add(&wd->sb_sha1, untracked->exclude_sha1, 20);
2369        }
2370
2371        intlen = encode_varint(untracked->untracked_nr, intbuf);
2372        strbuf_add(out, intbuf, intlen);
2373
2374        /* skip non-recurse directories */
2375        for (i = 0, value = 0; i < untracked->dirs_nr; i++)
2376                if (untracked->dirs[i]->recurse)
2377                        value++;
2378        intlen = encode_varint(value, intbuf);
2379        strbuf_add(out, intbuf, intlen);
2380
2381        strbuf_add(out, untracked->name, strlen(untracked->name) + 1);
2382
2383        for (i = 0; i < untracked->untracked_nr; i++)
2384                strbuf_add(out, untracked->untracked[i],
2385                           strlen(untracked->untracked[i]) + 1);
2386
2387        for (i = 0; i < untracked->dirs_nr; i++)
2388                if (untracked->dirs[i]->recurse)
2389                        write_one_dir(untracked->dirs[i], wd);
2390}
2391
2392void write_untracked_extension(struct strbuf *out, struct untracked_cache *untracked)
2393{
2394        struct ondisk_untracked_cache *ouc;
2395        struct write_data wd;
2396        unsigned char varbuf[16];
2397        int varint_len;
2398        size_t len = strlen(untracked->exclude_per_dir);
2399
2400        FLEX_ALLOC_MEM(ouc, exclude_per_dir, untracked->exclude_per_dir, len);
2401        stat_data_to_disk(&ouc->info_exclude_stat, &untracked->ss_info_exclude.stat);
2402        stat_data_to_disk(&ouc->excludes_file_stat, &untracked->ss_excludes_file.stat);
2403        hashcpy(ouc->info_exclude_sha1, untracked->ss_info_exclude.sha1);
2404        hashcpy(ouc->excludes_file_sha1, untracked->ss_excludes_file.sha1);
2405        ouc->dir_flags = htonl(untracked->dir_flags);
2406
2407        varint_len = encode_varint(untracked->ident.len, varbuf);
2408        strbuf_add(out, varbuf, varint_len);
2409        strbuf_addbuf(out, &untracked->ident);
2410
2411        strbuf_add(out, ouc, ouc_size(len));
2412        free(ouc);
2413        ouc = NULL;
2414
2415        if (!untracked->root) {
2416                varint_len = encode_varint(0, varbuf);
2417                strbuf_add(out, varbuf, varint_len);
2418                return;
2419        }
2420
2421        wd.index      = 0;
2422        wd.check_only = ewah_new();
2423        wd.valid      = ewah_new();
2424        wd.sha1_valid = ewah_new();
2425        strbuf_init(&wd.out, 1024);
2426        strbuf_init(&wd.sb_stat, 1024);
2427        strbuf_init(&wd.sb_sha1, 1024);
2428        write_one_dir(untracked->root, &wd);
2429
2430        varint_len = encode_varint(wd.index, varbuf);
2431        strbuf_add(out, varbuf, varint_len);
2432        strbuf_addbuf(out, &wd.out);
2433        ewah_serialize_strbuf(wd.valid, out);
2434        ewah_serialize_strbuf(wd.check_only, out);
2435        ewah_serialize_strbuf(wd.sha1_valid, out);
2436        strbuf_addbuf(out, &wd.sb_stat);
2437        strbuf_addbuf(out, &wd.sb_sha1);
2438        strbuf_addch(out, '\0'); /* safe guard for string lists */
2439
2440        ewah_free(wd.valid);
2441        ewah_free(wd.check_only);
2442        ewah_free(wd.sha1_valid);
2443        strbuf_release(&wd.out);
2444        strbuf_release(&wd.sb_stat);
2445        strbuf_release(&wd.sb_sha1);
2446}
2447
2448static void free_untracked(struct untracked_cache_dir *ucd)
2449{
2450        int i;
2451        if (!ucd)
2452                return;
2453        for (i = 0; i < ucd->dirs_nr; i++)
2454                free_untracked(ucd->dirs[i]);
2455        for (i = 0; i < ucd->untracked_nr; i++)
2456                free(ucd->untracked[i]);
2457        free(ucd->untracked);
2458        free(ucd->dirs);
2459        free(ucd);
2460}
2461
2462void free_untracked_cache(struct untracked_cache *uc)
2463{
2464        if (uc)
2465                free_untracked(uc->root);
2466        free(uc);
2467}
2468
2469struct read_data {
2470        int index;
2471        struct untracked_cache_dir **ucd;
2472        struct ewah_bitmap *check_only;
2473        struct ewah_bitmap *valid;
2474        struct ewah_bitmap *sha1_valid;
2475        const unsigned char *data;
2476        const unsigned char *end;
2477};
2478
2479static void stat_data_from_disk(struct stat_data *to, const struct stat_data *from)
2480{
2481        to->sd_ctime.sec  = get_be32(&from->sd_ctime.sec);
2482        to->sd_ctime.nsec = get_be32(&from->sd_ctime.nsec);
2483        to->sd_mtime.sec  = get_be32(&from->sd_mtime.sec);
2484        to->sd_mtime.nsec = get_be32(&from->sd_mtime.nsec);
2485        to->sd_dev        = get_be32(&from->sd_dev);
2486        to->sd_ino        = get_be32(&from->sd_ino);
2487        to->sd_uid        = get_be32(&from->sd_uid);
2488        to->sd_gid        = get_be32(&from->sd_gid);
2489        to->sd_size       = get_be32(&from->sd_size);
2490}
2491
2492static int read_one_dir(struct untracked_cache_dir **untracked_,
2493                        struct read_data *rd)
2494{
2495        struct untracked_cache_dir ud, *untracked;
2496        const unsigned char *next, *data = rd->data, *end = rd->end;
2497        unsigned int value;
2498        int i, len;
2499
2500        memset(&ud, 0, sizeof(ud));
2501
2502        next = data;
2503        value = decode_varint(&next);
2504        if (next > end)
2505                return -1;
2506        ud.recurse         = 1;
2507        ud.untracked_alloc = value;
2508        ud.untracked_nr    = value;
2509        if (ud.untracked_nr)
2510                ALLOC_ARRAY(ud.untracked, ud.untracked_nr);
2511        data = next;
2512
2513        next = data;
2514        ud.dirs_alloc = ud.dirs_nr = decode_varint(&next);
2515        if (next > end)
2516                return -1;
2517        ALLOC_ARRAY(ud.dirs, ud.dirs_nr);
2518        data = next;
2519
2520        len = strlen((const char *)data);
2521        next = data + len + 1;
2522        if (next > rd->end)
2523                return -1;
2524        *untracked_ = untracked = xmalloc(st_add(sizeof(*untracked), len));
2525        memcpy(untracked, &ud, sizeof(ud));
2526        memcpy(untracked->name, data, len + 1);
2527        data = next;
2528
2529        for (i = 0; i < untracked->untracked_nr; i++) {
2530                len = strlen((const char *)data);
2531                next = data + len + 1;
2532                if (next > rd->end)
2533                        return -1;
2534                untracked->untracked[i] = xstrdup((const char*)data);
2535                data = next;
2536        }
2537
2538        rd->ucd[rd->index++] = untracked;
2539        rd->data = data;
2540
2541        for (i = 0; i < untracked->dirs_nr; i++) {
2542                len = read_one_dir(untracked->dirs + i, rd);
2543                if (len < 0)
2544                        return -1;
2545        }
2546        return 0;
2547}
2548
2549static void set_check_only(size_t pos, void *cb)
2550{
2551        struct read_data *rd = cb;
2552        struct untracked_cache_dir *ud = rd->ucd[pos];
2553        ud->check_only = 1;
2554}
2555
2556static void read_stat(size_t pos, void *cb)
2557{
2558        struct read_data *rd = cb;
2559        struct untracked_cache_dir *ud = rd->ucd[pos];
2560        if (rd->data + sizeof(struct stat_data) > rd->end) {
2561                rd->data = rd->end + 1;
2562                return;
2563        }
2564        stat_data_from_disk(&ud->stat_data, (struct stat_data *)rd->data);
2565        rd->data += sizeof(struct stat_data);
2566        ud->valid = 1;
2567}
2568
2569static void read_sha1(size_t pos, void *cb)
2570{
2571        struct read_data *rd = cb;
2572        struct untracked_cache_dir *ud = rd->ucd[pos];
2573        if (rd->data + 20 > rd->end) {
2574                rd->data = rd->end + 1;
2575                return;
2576        }
2577        hashcpy(ud->exclude_sha1, rd->data);
2578        rd->data += 20;
2579}
2580
2581static void load_sha1_stat(struct sha1_stat *sha1_stat,
2582                           const struct stat_data *stat,
2583                           const unsigned char *sha1)
2584{
2585        stat_data_from_disk(&sha1_stat->stat, stat);
2586        hashcpy(sha1_stat->sha1, sha1);
2587        sha1_stat->valid = 1;
2588}
2589
2590struct untracked_cache *read_untracked_extension(const void *data, unsigned long sz)
2591{
2592        const struct ondisk_untracked_cache *ouc;
2593        struct untracked_cache *uc;
2594        struct read_data rd;
2595        const unsigned char *next = data, *end = (const unsigned char *)data + sz;
2596        const char *ident;
2597        int ident_len, len;
2598
2599        if (sz <= 1 || end[-1] != '\0')
2600                return NULL;
2601        end--;
2602
2603        ident_len = decode_varint(&next);
2604        if (next + ident_len > end)
2605                return NULL;
2606        ident = (const char *)next;
2607        next += ident_len;
2608
2609        ouc = (const struct ondisk_untracked_cache *)next;
2610        if (next + ouc_size(0) > end)
2611                return NULL;
2612
2613        uc = xcalloc(1, sizeof(*uc));
2614        strbuf_init(&uc->ident, ident_len);
2615        strbuf_add(&uc->ident, ident, ident_len);
2616        load_sha1_stat(&uc->ss_info_exclude, &ouc->info_exclude_stat,
2617                       ouc->info_exclude_sha1);
2618        load_sha1_stat(&uc->ss_excludes_file, &ouc->excludes_file_stat,
2619                       ouc->excludes_file_sha1);
2620        uc->dir_flags = get_be32(&ouc->dir_flags);
2621        uc->exclude_per_dir = xstrdup(ouc->exclude_per_dir);
2622        /* NUL after exclude_per_dir is covered by sizeof(*ouc) */
2623        next += ouc_size(strlen(ouc->exclude_per_dir));
2624        if (next >= end)
2625                goto done2;
2626
2627        len = decode_varint(&next);
2628        if (next > end || len == 0)
2629                goto done2;
2630
2631        rd.valid      = ewah_new();
2632        rd.check_only = ewah_new();
2633        rd.sha1_valid = ewah_new();
2634        rd.data       = next;
2635        rd.end        = end;
2636        rd.index      = 0;
2637        ALLOC_ARRAY(rd.ucd, len);
2638
2639        if (read_one_dir(&uc->root, &rd) || rd.index != len)
2640                goto done;
2641
2642        next = rd.data;
2643        len = ewah_read_mmap(rd.valid, next, end - next);
2644        if (len < 0)
2645                goto done;
2646
2647        next += len;
2648        len = ewah_read_mmap(rd.check_only, next, end - next);
2649        if (len < 0)
2650                goto done;
2651
2652        next += len;
2653        len = ewah_read_mmap(rd.sha1_valid, next, end - next);
2654        if (len < 0)
2655                goto done;
2656
2657        ewah_each_bit(rd.check_only, set_check_only, &rd);
2658        rd.data = next + len;
2659        ewah_each_bit(rd.valid, read_stat, &rd);
2660        ewah_each_bit(rd.sha1_valid, read_sha1, &rd);
2661        next = rd.data;
2662
2663done:
2664        free(rd.ucd);
2665        ewah_free(rd.valid);
2666        ewah_free(rd.check_only);
2667        ewah_free(rd.sha1_valid);
2668done2:
2669        if (next != end) {
2670                free_untracked_cache(uc);
2671                uc = NULL;
2672        }
2673        return uc;
2674}
2675
2676static void invalidate_one_directory(struct untracked_cache *uc,
2677                                     struct untracked_cache_dir *ucd)
2678{
2679        uc->dir_invalidated++;
2680        ucd->valid = 0;
2681        ucd->untracked_nr = 0;
2682}
2683
2684/*
2685 * Normally when an entry is added or removed from a directory,
2686 * invalidating that directory is enough. No need to touch its
2687 * ancestors. When a directory is shown as "foo/bar/" in git-status
2688 * however, deleting or adding an entry may have cascading effect.
2689 *
2690 * Say the "foo/bar/file" has become untracked, we need to tell the
2691 * untracked_cache_dir of "foo" that "bar/" is not an untracked
2692 * directory any more (because "bar" is managed by foo as an untracked
2693 * "file").
2694 *
2695 * Similarly, if "foo/bar/file" moves from untracked to tracked and it
2696 * was the last untracked entry in the entire "foo", we should show
2697 * "foo/" instead. Which means we have to invalidate past "bar" up to
2698 * "foo".
2699 *
2700 * This function traverses all directories from root to leaf. If there
2701 * is a chance of one of the above cases happening, we invalidate back
2702 * to root. Otherwise we just invalidate the leaf. There may be a more
2703 * sophisticated way than checking for SHOW_OTHER_DIRECTORIES to
2704 * detect these cases and avoid unnecessary invalidation, for example,
2705 * checking for the untracked entry named "bar/" in "foo", but for now
2706 * stick to something safe and simple.
2707 */
2708static int invalidate_one_component(struct untracked_cache *uc,
2709                                    struct untracked_cache_dir *dir,
2710                                    const char *path, int len)
2711{
2712        const char *rest = strchr(path, '/');
2713
2714        if (rest) {
2715                int component_len = rest - path;
2716                struct untracked_cache_dir *d =
2717                        lookup_untracked(uc, dir, path, component_len);
2718                int ret =
2719                        invalidate_one_component(uc, d, rest + 1,
2720                                                 len - (component_len + 1));
2721                if (ret)
2722                        invalidate_one_directory(uc, dir);
2723                return ret;
2724        }
2725
2726        invalidate_one_directory(uc, dir);
2727        return uc->dir_flags & DIR_SHOW_OTHER_DIRECTORIES;
2728}
2729
2730void untracked_cache_invalidate_path(struct index_state *istate,
2731                                     const char *path)
2732{
2733        if (!istate->untracked || !istate->untracked->root)
2734                return;
2735        invalidate_one_component(istate->untracked, istate->untracked->root,
2736                                 path, strlen(path));
2737}
2738
2739void untracked_cache_remove_from_index(struct index_state *istate,
2740                                       const char *path)
2741{
2742        untracked_cache_invalidate_path(istate, path);
2743}
2744
2745void untracked_cache_add_to_index(struct index_state *istate,
2746                                  const char *path)
2747{
2748        untracked_cache_invalidate_path(istate, path);
2749}