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