git.sesse.net Git - plocate/blob - plocate.cpp

   1 #include "db.h"
   2 #include "io_uring_engine.h"
   3 #include "parse_trigrams.h"
   4 #include "unique_sort.h"
   5
   6 #include <algorithm>
   7 #include <assert.h>
   8 #include <chrono>
   9 #include <fcntl.h>
  10 #include <fnmatch.h>
  11 #include <functional>
  12 #include <getopt.h>
  13 #include <iosfwd>
  14 #include <iterator>
  15 #include <limits>
  16 #include <memory>
  17 #include <queue>
  18 #include <regex.h>
  19 #include <stdint.h>
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <string.h>
  23 #include <string>
  24 #include <string_view>
  25 #include <unistd.h>
  26 #include <unordered_map>
  27 #include <unordered_set>
  28 #include <utility>
  29 #include <vector>
  30 #include <zstd.h>
  31
  32 using namespace std;
  33 using namespace std::chrono;
  34
  35 #define dprintf(...) \
  36         do { \
  37                 if (use_debug) { \
  38                         fprintf(stderr, __VA_ARGS__); \
  39                 } \
  40         } while (false)
  41
  42 #include "turbopfor.h"
  43
  44 const char *dbpath = "/var/lib/mlocate/plocate.db";
  45 bool ignore_case = false;
  46 bool only_count = false;
  47 bool print_nul = false;
  48 bool use_debug = false;
  49 int64_t limit_matches = numeric_limits<int64_t>::max();
  50
  51 class Serializer {
  52 public:
  53         bool ready_to_print(int seq) { return next_seq == seq; }
  54         void print_delayed(int seq, const vector<string> msg);
  55         void release_current();
  56
  57 private:
  58         int next_seq = 0;
  59         struct Element {
  60                 int seq;
  61                 vector<string> msg;
  62
  63                 bool operator<(const Element &other) const
  64                 {
  65                         return seq > other.seq;
  66                 }
  67         };
  68         priority_queue<Element> pending;
  69 };
  70
  71 void Serializer::print_delayed(int seq, const vector<string> msg)
  72 {
  73         pending.push(Element{ seq, move(msg) });
  74 }
  75
  76 void Serializer::release_current()
  77 {
  78         ++next_seq;
  79
  80         // See if any delayed prints can now be dealt with.
  81         while (!pending.empty() && pending.top().seq == next_seq) {
  82                 if (limit_matches-- <= 0)
  83                         return;
  84                 for (const string &msg : pending.top().msg) {
  85                         if (print_nul) {
  86                                 printf("%s%c", msg.c_str(), 0);
  87                         } else {
  88                                 printf("%s\n", msg.c_str());
  89                         }
  90                 }
  91                 pending.pop();
  92                 ++next_seq;
  93         }
  94 }
  95
  96 struct Needle {
  97         enum { STRSTR,
  98                REGEX,  // Not currently used.
  99                GLOB } type;
 100         string str;  // Filled in no matter what.
 101         regex_t re;  // For REGEX.
 102 };
 103
 104 bool matches(const Needle &needle, const char *haystack)
 105 {
 106         if (needle.type == Needle::STRSTR) {
 107                 return strstr(haystack, needle.str.c_str()) != nullptr;
 108         } else if (needle.type == Needle::GLOB) {
 109                 int flags = ignore_case ? FNM_CASEFOLD : 0;
 110                 return fnmatch(needle.str.c_str(), haystack, flags) == 0;
 111         } else {
 112                 assert(needle.type == Needle::REGEX);
 113                 return regexec(&needle.re, haystack, /*nmatch=*/0, /*pmatch=*/nullptr, /*flags=*/0) == 0;
 114         }
 115 }
 116
 117 bool has_access(const char *filename,
 118                 unordered_map<string, bool> *access_rx_cache)
 119 {
 120         const char *end = strchr(filename + 1, '/');
 121         while (end != nullptr) {
 122                 string parent_path(filename, end);
 123                 auto it = access_rx_cache->find(parent_path);
 124                 bool ok;
 125                 if (it == access_rx_cache->end()) {
 126                         ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
 127                         access_rx_cache->emplace(move(parent_path), ok);
 128                 } else {
 129                         ok = it->second;
 130                 }
 131                 if (!ok) {
 132                         return false;
 133                 }
 134                 end = strchr(end + 1, '/');
 135         }
 136
 137         return true;
 138 }
 139
 140 class Corpus {
 141 public:
 142         Corpus(int fd, IOUringEngine *engine);
 143         ~Corpus();
 144         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
 145         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
 146         size_t get_num_filename_blocks() const;
 147         off_t offset_for_block(uint32_t docid) const
 148         {
 149                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
 150         }
 151
 152 public:
 153         const int fd;
 154         IOUringEngine *const engine;
 155
 156         Header hdr;
 157 };
 158
 159 Corpus::Corpus(int fd, IOUringEngine *engine)
 160         : fd(fd), engine(engine)
 161 {
 162         // Enable to test cold-cache behavior (except for access()).
 163         if (false) {
 164                 off_t len = lseek(fd, 0, SEEK_END);
 165                 if (len == -1) {
 166                         perror("lseek");
 167                         exit(1);
 168                 }
 169                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
 170         }
 171
 172         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 173         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 174                 fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
 175                 exit(1);
 176         }
 177         if (hdr.version != 0) {
 178                 fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version);
 179                 exit(1);
 180         }
 181 }
 182
 183 Corpus::~Corpus()
 184 {
 185         close(fd);
 186 }
 187
 188 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 189 {
 190         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 191         engine->submit_read(fd, sizeof(Trigram) * (hdr.extra_ht_slots + 2), hdr.hash_table_offset_bytes + sizeof(Trigram) * bucket, [this, trgm, cb{ move(cb) }](string_view s) {
 192                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 193                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 194                         if (trgmptr[i].trgm == trgm) {
 195                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 196                                 return;
 197                         }
 198                 }
 199
 200                 // Not found.
 201                 cb(nullptr, 0);
 202         });
 203 }
 204
 205 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 206 {
 207         // Read the file offset from this docid and the next one.
 208         // This is always allowed, since we have a sentinel block at the end.
 209         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 210                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 211                 off_t offset = ptr[0];
 212                 size_t len = ptr[1] - ptr[0];
 213                 engine->submit_read(fd, len, offset, cb);
 214         });
 215 }
 216
 217 size_t Corpus::get_num_filename_blocks() const
 218 {
 219         return hdr.num_docids;
 220 }
 221
 222 uint64_t scan_file_block(const vector<Needle> &needles, string_view compressed,
 223                          unordered_map<string, bool> *access_rx_cache, int seq,
 224                          Serializer *serializer)
 225 {
 226         uint64_t matched = 0;
 227
 228         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 229         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 230                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 231                 exit(1);
 232         }
 233
 234         string block;
 235         block.resize(uncompressed_len + 1);
 236
 237         size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(),
 238                                      compressed.size());
 239         if (ZSTD_isError(err)) {
 240                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 241                 exit(1);
 242         }
 243         block[block.size() - 1] = '\0';
 244
 245         bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq));
 246         vector<string> delayed;
 247
 248         for (const char *filename = block.data();
 249              filename != block.data() + block.size();
 250              filename += strlen(filename) + 1) {
 251                 bool found = true;
 252                 for (const Needle &needle : needles) {
 253                         if (!matches(needle, filename)) {
 254                                 found = false;
 255                                 break;
 256                         }
 257                 }
 258                 if (found && has_access(filename, access_rx_cache)) {
 259                         if (limit_matches-- <= 0)
 260                                 break;
 261                         ++matched;
 262                         if (only_count)
 263                                 continue;
 264                         if (immediate_print) {
 265                                 if (print_nul) {
 266                                         printf("%s%c", filename, 0);
 267                                 } else {
 268                                         printf("%s\n", filename);
 269                                 }
 270                         } else {
 271                                 delayed.push_back(filename);
 272                         }
 273                 }
 274         }
 275         if (serializer != nullptr && !only_count) {
 276                 if (immediate_print) {
 277                         serializer->release_current();
 278                 } else {
 279                         serializer->print_delayed(seq, move(delayed));
 280                 }
 281         }
 282         return matched;
 283 }
 284
 285 size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 286 {
 287         Serializer docids_in_order;
 288         unordered_map<string, bool> access_rx_cache;
 289         uint64_t matched = 0;
 290         for (size_t i = 0; i < docids.size(); ++i) {
 291                 uint32_t docid = docids[i];
 292                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string_view compressed) {
 293                         matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order);
 294                 });
 295         }
 296         engine->finish();
 297         return matched;
 298 }
 299
 300 // We do this sequentially, as it's faster than scattering
 301 // a lot of I/O through io_uring and hoping the kernel will
 302 // coalesce it plus readahead for us.
 303 uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus, IOUringEngine *engine)
 304 {
 305         unordered_map<string, bool> access_rx_cache;
 306         uint32_t num_blocks = corpus.get_num_filename_blocks();
 307         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 308         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 309         string compressed;
 310         uint64_t matched = 0;
 311         for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
 312                 uint32_t last_docid = std::min(io_docid + 32, num_blocks);
 313                 size_t io_len = offsets[last_docid] - offsets[io_docid];
 314                 if (compressed.size() < io_len) {
 315                         compressed.resize(io_len);
 316                 }
 317                 complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
 318
 319                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 320                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 321                         size_t len = offsets[docid + 1] - offsets[docid];
 322                         matched += scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr);
 323                         if (limit_matches <= 0)
 324                                 return matched;
 325                 }
 326         }
 327         return matched;
 328 }
 329
 330 // Takes the given posting list, unions it into the parts of the trigram disjunction
 331 // already read; if the list is complete, intersects with “cur_candidates”.
 332 //
 333 // Returns true if the search should be aborted (we are done).
 334 bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
 335 {
 336         if (td->docids.empty()) {
 337                 td->docids = move(decoded);
 338         } else {
 339                 tmp->clear();
 340                 set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
 341                 swap(*tmp, td->docids);
 342         }
 343         if (--td->remaining_trigrams_to_read > 0) {
 344                 // Need to wait for more.
 345                 if (ignore_case) {
 346                         dprintf("  ... %u reads left in OR group %u (%zu docids in list)\n",
 347                                 td->remaining_trigrams_to_read, td->index, td->docids.size());
 348                 }
 349                 return false;
 350         }
 351         if (cur_candidates->empty()) {
 352                 if (ignore_case) {
 353                         dprintf("  ... all reads done for OR group %u (%zu docids)\n",
 354                                 td->index, td->docids.size());
 355                 }
 356                 *cur_candidates = move(td->docids);
 357         } else {
 358                 tmp->clear();
 359                 set_intersection(cur_candidates->begin(), cur_candidates->end(),
 360                                  td->docids.begin(), td->docids.end(),
 361                                  back_inserter(*tmp));
 362                 swap(*cur_candidates, *tmp);
 363                 if (ignore_case) {
 364                         if (cur_candidates->empty()) {
 365                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n",
 366                                         td->index, td->docids.size());
 367                                 return true;
 368                         } else {
 369                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n",
 370                                         td->index, td->docids.size(), cur_candidates->size());
 371                         }
 372                 }
 373         }
 374         return false;
 375 }
 376
 377 void do_search_file(const vector<Needle> &needles, const char *filename)
 378 {
 379         int fd = open(filename, O_RDONLY);
 380         if (fd == -1) {
 381                 perror(filename);
 382                 exit(1);
 383         }
 384
 385         // Drop privileges.
 386         if (setgid(getgid()) != 0) {
 387                 perror("setgid");
 388                 exit(EXIT_FAILURE);
 389         }
 390
 391         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 392         if (access("/", R_OK | X_OK)) {
 393                 // We can't find anything, no need to bother...
 394                 return;
 395         }
 396
 397         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 398         Corpus corpus(fd, &engine);
 399         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 400
 401         vector<TrigramDisjunction> trigram_groups;
 402         for (const Needle &needle : needles) {
 403                 if (needle.str.size() < 3)
 404                         continue;
 405                 parse_trigrams(needle.str, ignore_case, &trigram_groups);
 406         }
 407         unique_sort(
 408                 &trigram_groups,
 409                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
 410                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 411
 412         // Give them names for debugging.
 413         unsigned td_index = 0;
 414         for (TrigramDisjunction &td : trigram_groups) {
 415                 td.index = td_index++;
 416         }
 417
 418         // Collect which trigrams we need to look up in the hash table.
 419         unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
 420         for (TrigramDisjunction &td : trigram_groups) {
 421                 for (uint32_t trgm : td.trigram_alternatives) {
 422                         trigrams_to_lookup[trgm].push_back(&td);
 423                 }
 424         }
 425         if (trigrams_to_lookup.empty()) {
 426                 // Too short for trigram matching. Apply brute force.
 427                 // (We could have searched through all trigrams that matched
 428                 // the pattern and done a union of them, but that's a lot of
 429                 // work for fairly unclear gain.)
 430                 uint64_t matched = scan_all_docids(needles, fd, corpus, &engine);
 431                 if (only_count) {
 432                         printf("%zu\n", matched);
 433                 }
 434                 return;
 435         }
 436
 437         // Look them all up on disk.
 438         for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
 439                 corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }](const Trigram *trgmptr, size_t len) {
 440                         if (trgmptr == nullptr) {
 441                                 dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
 442                                 for (TrigramDisjunction *td : *trigram_groups) {
 443                                         --td->remaining_trigrams_to_read;
 444                                         if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
 445                                                 dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
 446                                                 if (only_count) {
 447                                                         printf("0\n");
 448                                                 }
 449                                                 exit(0);
 450                                         }
 451                                 }
 452                                 return;
 453                         }
 454                         for (TrigramDisjunction *td : *trigram_groups) {
 455                                 --td->remaining_trigrams_to_read;
 456                                 td->max_num_docids += trgmptr->num_docids;
 457                                 td->read_trigrams.emplace_back(*trgmptr, len);
 458                         }
 459                 });
 460         }
 461         engine.finish();
 462         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 463
 464         for (TrigramDisjunction &td : trigram_groups) {
 465                 // Reset for reads.
 466                 td.remaining_trigrams_to_read = td.read_trigrams.size();
 467
 468                 if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
 469                         dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
 470                 }
 471         }
 472
 473         // TODO: For case-insensitive (ie. more than one alternative in each),
 474         // prioritize the ones with fewer seeks?
 475         sort(trigram_groups.begin(), trigram_groups.end(),
 476              [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
 477                      return a.max_num_docids < b.max_num_docids;
 478              });
 479
 480         unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
 481         for (TrigramDisjunction &td : trigram_groups) {
 482                 for (uint32_t trgm : td.trigram_alternatives) {
 483                         uses_trigram[trgm].push_back(&td);
 484                 }
 485         }
 486
 487         unordered_set<uint32_t> trigrams_submitted_read;
 488         vector<uint32_t> cur_candidates, tmp, decoded;
 489         bool done = false;
 490         for (TrigramDisjunction &td : trigram_groups) {
 491                 if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
 492                         dprintf("%s has up to %u entries, ignoring the rest (will "
 493                                 "weed out false positives later)\n",
 494                                 print_td(td).c_str(), td.max_num_docids);
 495                         break;
 496                 }
 497
 498                 for (auto &[trgmptr, len] : td.read_trigrams) {
 499                         if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
 500                                 continue;
 501                         }
 502                         trigrams_submitted_read.insert(trgmptr.trgm);
 503                         // Only stay a certain amount ahead, so that we don't spend I/O
 504                         // on reading the latter, large posting lists. We are unlikely
 505                         // to need them anyway, even if they should come in first.
 506                         if (engine.get_waiting_reads() >= 5) {
 507                                 engine.finish();
 508                                 if (done)
 509                                         break;
 510                         }
 511                         engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
 512                                 if (done)
 513                                         return;
 514
 515                                 uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 516                                 const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 517                                 size_t num = trgmptr.num_docids;
 518                                 decoded.resize(num);
 519                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
 520
 521                                 assert(uses_trigram.count(trgm) != 0);
 522                                 bool was_empty = cur_candidates.empty();
 523                                 if (ignore_case) {
 524                                         dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 525                                 }
 526
 527                                 for (TrigramDisjunction *td : uses_trigram[trgm]) {
 528                                         done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
 529                                         if (done)
 530                                                 break;
 531                                 }
 532                                 if (!ignore_case) {
 533                                         if (was_empty) {
 534                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 535                                         } else if (cur_candidates.empty()) {
 536                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
 537                                         } else {
 538                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
 539                                         }
 540                                 }
 541                         });
 542                 }
 543         }
 544         engine.finish();
 545         if (done) {
 546                 return;
 547         }
 548         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 549                 1e3 * duration<float>(steady_clock::now() - start).count());
 550
 551         uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
 552         dprintf("Done in %.1f ms, found %zu matches.\n",
 553                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 554
 555         if (only_count) {
 556                 printf("%zu\n", matched);
 557         }
 558 }
 559
 560 string unescape_glob_to_plain_string(const string &needle)
 561 {
 562         string unescaped;
 563         for (size_t i = 0; i < needle.size(); i += read_unigram(needle, i).second) {
 564                 uint32_t ch = read_unigram(needle, i).first;
 565                 assert(ch != WILDCARD_UNIGRAM);
 566                 if (ch == PREMATURE_END_UNIGRAM) {
 567                         fprintf(stderr, "Pattern '%s' ended prematurely\n", needle.c_str());
 568                         exit(1);
 569                 }
 570                 unescaped.push_back(ch);
 571         }
 572         return unescaped;
 573 }
 574
 575 void usage()
 576 {
 577         // The help text comes from mlocate.
 578         printf("Usage: plocate [OPTION]... PATTERN...\n");
 579         printf("\n");
 580         printf("  -c, --count            only print number of found entries\n");
 581         printf("  -d, --database DBPATH  use DBPATH instead of default database (which is\n");
 582         printf("                         %s)\n", dbpath);
 583         printf("  -h, --help             print this help\n");
 584         printf("  -i, --ignore-case      ignore case distinctions when matching patterns\n");
 585         printf("  -l, --limit, -n LIMIT  limit output (or counting) to LIMIT entries\n");
 586         printf("  -0, --null             separate entries with NUL on output\n");
 587         printf("  -V, --version          print version information\n");
 588 }
 589
 590 void version()
 591 {
 592         printf("plocate %s\n", PLOCATE_VERSION);
 593         printf("Copyright 2020 Steinar H. Gunderson\n");
 594         printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\n");
 595         printf("This is free software: you are free to change and redistribute it.\n");
 596         printf("There is NO WARRANTY, to the extent permitted by law.\n");
 597         exit(0);
 598 }
 599
 600 int main(int argc, char **argv)
 601 {
 602         static const struct option long_options[] = {
 603                 { "help", no_argument, 0, 'h' },
 604                 { "count", no_argument, 0, 'c' },
 605                 { "database", required_argument, 0, 'd' },
 606                 { "ignore-case", no_argument, 0, 'i' },
 607                 { "limit", required_argument, 0, 'l' },
 608                 { "null", no_argument, 0, '0' },
 609                 { "version", no_argument, 0, 'V' },
 610                 { "debug", no_argument, 0, 'D' },  // Not documented.
 611                 { 0, 0, 0, 0 }
 612         };
 613
 614         setlocale(LC_ALL, "");
 615         for (;;) {
 616                 int option_index = 0;
 617                 int c = getopt_long(argc, argv, "cd:hil:n:0VD", long_options, &option_index);
 618                 if (c == -1) {
 619                         break;
 620                 }
 621                 switch (c) {
 622                 case 'c':
 623                         only_count = true;
 624                         break;
 625                 case 'd':
 626                         dbpath = strdup(optarg);
 627                         break;
 628                 case 'h':
 629                         usage();
 630                         exit(0);
 631                 case 'i':
 632                         ignore_case = true;
 633                         break;
 634                 case 'l':
 635                 case 'n':
 636                         limit_matches = atoll(optarg);
 637                         break;
 638                 case '0':
 639                         print_nul = true;
 640                         break;
 641                 case 'D':
 642                         use_debug = true;
 643                         break;
 644                 case 'V':
 645                         version();
 646                         break;
 647                 default:
 648                         exit(1);
 649                 }
 650         }
 651
 652         if (use_debug) {
 653                 // Debug information would leak information about which files exist,
 654                 // so drop setgid before we open the file; one would either need to run
 655                 // as root, or use a locally-built file.
 656                 if (setgid(getgid()) != 0) {
 657                         perror("setgid");
 658                         exit(EXIT_FAILURE);
 659                 }
 660         }
 661
 662         vector<Needle> needles;
 663         for (int i = optind; i < argc; ++i) {
 664                 Needle needle;
 665                 needle.str = argv[i];
 666
 667                 // See if there are any wildcard characters, which indicates we should treat it
 668                 // as an (anchored) glob.
 669                 bool any_wildcard = false;
 670                 for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) {
 671                         if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) {
 672                                 any_wildcard = true;
 673                                 break;
 674                         }
 675                 }
 676
 677                 if (any_wildcard) {
 678                         needle.type = Needle::GLOB;
 679                 } else if (ignore_case) {
 680                         // strcasestr() doesn't handle locales correctly (even though LSB
 681                         // claims it should), but somehow, fnmatch() does, and it's about
 682                         // the same speed as using a regex.
 683                         needle.type = Needle::GLOB;
 684                         needle.str = "*" + needle.str + "*";
 685                 } else {
 686                         needle.type = Needle::STRSTR;
 687                         needle.str = unescape_glob_to_plain_string(needle.str);
 688                 }
 689                 needles.push_back(move(needle));
 690         }
 691         if (needles.empty()) {
 692                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 693                 exit(0);
 694         }
 695         do_search_file(needles, dbpath);
 696 }