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 "turbopfor.h"
   5 #include "unique_sort.h"
   6
   7 #include <algorithm>
   8 #include <assert.h>
   9 #include <chrono>
  10 #include <fcntl.h>
  11 #include <fnmatch.h>
  12 #include <functional>
  13 #include <getopt.h>
  14 #include <iosfwd>
  15 #include <iterator>
  16 #include <limits>
  17 #include <memory>
  18 #include <queue>
  19 #include <regex.h>
  20 #include <stdint.h>
  21 #include <stdio.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include <string>
  25 #include <string_view>
  26 #include <unistd.h>
  27 #include <unordered_map>
  28 #include <unordered_set>
  29 #include <utility>
  30 #include <vector>
  31 #include <zstd.h>
  32
  33 using namespace std;
  34 using namespace std::chrono;
  35
  36 #define dprintf(...) \
  37         do { \
  38                 if (use_debug) { \
  39                         fprintf(stderr, __VA_ARGS__); \
  40                 } \
  41         } while (false)
  42
  43 #define DEFAULT_DBPATH "/var/lib/mlocate/plocate.db"
  44
  45 const char *dbpath = DEFAULT_DBPATH;
  46 bool ignore_case = false;
  47 bool only_count = false;
  48 bool print_nul = false;
  49 bool use_debug = false;
  50 bool patterns_are_regex = false;
  51 bool use_extended_regex = false;
  52 int64_t limit_matches = numeric_limits<int64_t>::max();
  53
  54 class Serializer {
  55 public:
  56         bool ready_to_print(int seq) { return next_seq == seq; }
  57         void print_delayed(int seq, const vector<string> msg);
  58         void release_current();
  59
  60 private:
  61         int next_seq = 0;
  62         struct Element {
  63                 int seq;
  64                 vector<string> msg;
  65
  66                 bool operator<(const Element &other) const
  67                 {
  68                         return seq > other.seq;
  69                 }
  70         };
  71         priority_queue<Element> pending;
  72 };
  73
  74 void Serializer::print_delayed(int seq, const vector<string> msg)
  75 {
  76         pending.push(Element{ seq, move(msg) });
  77 }
  78
  79 void Serializer::release_current()
  80 {
  81         ++next_seq;
  82
  83         // See if any delayed prints can now be dealt with.
  84         while (!pending.empty() && pending.top().seq == next_seq) {
  85                 if (limit_matches-- <= 0)
  86                         return;
  87                 for (const string &msg : pending.top().msg) {
  88                         if (print_nul) {
  89                                 printf("%s%c", msg.c_str(), 0);
  90                         } else {
  91                                 printf("%s\n", msg.c_str());
  92                         }
  93                 }
  94                 pending.pop();
  95                 ++next_seq;
  96         }
  97 }
  98
  99 struct Needle {
 100         enum { STRSTR,
 101                REGEX,
 102                GLOB } type;
 103         string str;  // Filled in no matter what.
 104         regex_t re;  // For REGEX.
 105 };
 106
 107 bool matches(const Needle &needle, const char *haystack)
 108 {
 109         if (needle.type == Needle::STRSTR) {
 110                 return strstr(haystack, needle.str.c_str()) != nullptr;
 111         } else if (needle.type == Needle::GLOB) {
 112                 int flags = ignore_case ? FNM_CASEFOLD : 0;
 113                 return fnmatch(needle.str.c_str(), haystack, flags) == 0;
 114         } else {
 115                 assert(needle.type == Needle::REGEX);
 116                 return regexec(&needle.re, haystack, /*nmatch=*/0, /*pmatch=*/nullptr, /*flags=*/0) == 0;
 117         }
 118 }
 119
 120 bool has_access(const char *filename,
 121                 unordered_map<string, bool> *access_rx_cache)
 122 {
 123         const char *end = strchr(filename + 1, '/');
 124         while (end != nullptr) {
 125                 string parent_path(filename, end);
 126                 auto it = access_rx_cache->find(parent_path);
 127                 bool ok;
 128                 if (it == access_rx_cache->end()) {
 129                         ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
 130                         access_rx_cache->emplace(move(parent_path), ok);
 131                 } else {
 132                         ok = it->second;
 133                 }
 134                 if (!ok) {
 135                         return false;
 136                 }
 137                 end = strchr(end + 1, '/');
 138         }
 139
 140         return true;
 141 }
 142
 143 class Corpus {
 144 public:
 145         Corpus(int fd, IOUringEngine *engine);
 146         ~Corpus();
 147         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
 148         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
 149         size_t get_num_filename_blocks() const;
 150         off_t offset_for_block(uint32_t docid) const
 151         {
 152                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
 153         }
 154
 155 public:
 156         const int fd;
 157         IOUringEngine *const engine;
 158
 159         Header hdr;
 160 };
 161
 162 Corpus::Corpus(int fd, IOUringEngine *engine)
 163         : fd(fd), engine(engine)
 164 {
 165         // Enable to test cold-cache behavior (except for access()).
 166         if (false) {
 167                 off_t len = lseek(fd, 0, SEEK_END);
 168                 if (len == -1) {
 169                         perror("lseek");
 170                         exit(1);
 171                 }
 172                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
 173         }
 174
 175         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 176         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 177                 fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
 178                 exit(1);
 179         }
 180         if (hdr.version != 0) {
 181                 fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version);
 182                 exit(1);
 183         }
 184 }
 185
 186 Corpus::~Corpus()
 187 {
 188         close(fd);
 189 }
 190
 191 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 192 {
 193         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 194         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) {
 195                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 196                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 197                         if (trgmptr[i].trgm == trgm) {
 198                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 199                                 return;
 200                         }
 201                 }
 202
 203                 // Not found.
 204                 cb(nullptr, 0);
 205         });
 206 }
 207
 208 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 209 {
 210         // Read the file offset from this docid and the next one.
 211         // This is always allowed, since we have a sentinel block at the end.
 212         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 213                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 214                 off_t offset = ptr[0];
 215                 size_t len = ptr[1] - ptr[0];
 216                 engine->submit_read(fd, len, offset, cb);
 217         });
 218 }
 219
 220 size_t Corpus::get_num_filename_blocks() const
 221 {
 222         return hdr.num_docids;
 223 }
 224
 225 uint64_t scan_file_block(const vector<Needle> &needles, string_view compressed,
 226                          unordered_map<string, bool> *access_rx_cache, int seq,
 227                          Serializer *serializer)
 228 {
 229         uint64_t matched = 0;
 230
 231         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 232         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 233                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 234                 exit(1);
 235         }
 236
 237         string block;
 238         block.resize(uncompressed_len + 1);
 239
 240         size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(),
 241                                      compressed.size());
 242         if (ZSTD_isError(err)) {
 243                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 244                 exit(1);
 245         }
 246         block[block.size() - 1] = '\0';
 247
 248         bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq));
 249         vector<string> delayed;
 250
 251         for (const char *filename = block.data();
 252              filename != block.data() + block.size();
 253              filename += strlen(filename) + 1) {
 254                 bool found = true;
 255                 for (const Needle &needle : needles) {
 256                         if (!matches(needle, filename)) {
 257                                 found = false;
 258                                 break;
 259                         }
 260                 }
 261                 if (found && has_access(filename, access_rx_cache)) {
 262                         if (limit_matches-- <= 0)
 263                                 break;
 264                         ++matched;
 265                         if (only_count)
 266                                 continue;
 267                         if (immediate_print) {
 268                                 if (print_nul) {
 269                                         printf("%s%c", filename, 0);
 270                                 } else {
 271                                         printf("%s\n", filename);
 272                                 }
 273                         } else {
 274                                 delayed.push_back(filename);
 275                         }
 276                 }
 277         }
 278         if (serializer != nullptr && !only_count) {
 279                 if (immediate_print) {
 280                         serializer->release_current();
 281                 } else {
 282                         serializer->print_delayed(seq, move(delayed));
 283                 }
 284         }
 285         return matched;
 286 }
 287
 288 size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 289 {
 290         Serializer docids_in_order;
 291         unordered_map<string, bool> access_rx_cache;
 292         uint64_t matched = 0;
 293         for (size_t i = 0; i < docids.size(); ++i) {
 294                 uint32_t docid = docids[i];
 295                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string_view compressed) {
 296                         matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order);
 297                 });
 298         }
 299         engine->finish();
 300         return matched;
 301 }
 302
 303 // We do this sequentially, as it's faster than scattering
 304 // a lot of I/O through io_uring and hoping the kernel will
 305 // coalesce it plus readahead for us.
 306 uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus, IOUringEngine *engine)
 307 {
 308         unordered_map<string, bool> access_rx_cache;
 309         uint32_t num_blocks = corpus.get_num_filename_blocks();
 310         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 311         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 312         string compressed;
 313         uint64_t matched = 0;
 314         for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
 315                 uint32_t last_docid = std::min(io_docid + 32, num_blocks);
 316                 size_t io_len = offsets[last_docid] - offsets[io_docid];
 317                 if (compressed.size() < io_len) {
 318                         compressed.resize(io_len);
 319                 }
 320                 complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
 321
 322                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 323                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 324                         size_t len = offsets[docid + 1] - offsets[docid];
 325                         matched += scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr);
 326                         if (limit_matches <= 0)
 327                                 return matched;
 328                 }
 329         }
 330         return matched;
 331 }
 332
 333 // Takes the given posting list, unions it into the parts of the trigram disjunction
 334 // already read; if the list is complete, intersects with “cur_candidates”.
 335 //
 336 // Returns true if the search should be aborted (we are done).
 337 bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
 338 {
 339         if (td->docids.empty()) {
 340                 td->docids = move(decoded);
 341         } else {
 342                 tmp->clear();
 343                 set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
 344                 swap(*tmp, td->docids);
 345         }
 346         if (--td->remaining_trigrams_to_read > 0) {
 347                 // Need to wait for more.
 348                 if (ignore_case) {
 349                         dprintf("  ... %u reads left in OR group %u (%zu docids in list)\n",
 350                                 td->remaining_trigrams_to_read, td->index, td->docids.size());
 351                 }
 352                 return false;
 353         }
 354         if (cur_candidates->empty()) {
 355                 if (ignore_case) {
 356                         dprintf("  ... all reads done for OR group %u (%zu docids)\n",
 357                                 td->index, td->docids.size());
 358                 }
 359                 *cur_candidates = move(td->docids);
 360         } else {
 361                 tmp->clear();
 362                 set_intersection(cur_candidates->begin(), cur_candidates->end(),
 363                                  td->docids.begin(), td->docids.end(),
 364                                  back_inserter(*tmp));
 365                 swap(*cur_candidates, *tmp);
 366                 if (ignore_case) {
 367                         if (cur_candidates->empty()) {
 368                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n",
 369                                         td->index, td->docids.size());
 370                                 return true;
 371                         } else {
 372                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n",
 373                                         td->index, td->docids.size(), cur_candidates->size());
 374                         }
 375                 }
 376         }
 377         return false;
 378 }
 379
 380 void do_search_file(const vector<Needle> &needles, const char *filename)
 381 {
 382         int fd = open(filename, O_RDONLY);
 383         if (fd == -1) {
 384                 perror(filename);
 385                 exit(1);
 386         }
 387
 388         // Drop privileges.
 389         if (setgid(getgid()) != 0) {
 390                 perror("setgid");
 391                 exit(EXIT_FAILURE);
 392         }
 393
 394         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 395         if (access("/", R_OK | X_OK)) {
 396                 // We can't find anything, no need to bother...
 397                 return;
 398         }
 399
 400         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 401         Corpus corpus(fd, &engine);
 402         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 403
 404         vector<TrigramDisjunction> trigram_groups;
 405         if (patterns_are_regex) {
 406                 // We could parse the regex to find trigrams that have to be there
 407                 // (there are actually known algorithms to deal with disjunctions
 408                 // and such, too), but for now, we just go brute force.
 409                 // Using locate with regexes is pretty niche.
 410         } else {
 411                 for (const Needle &needle : needles) {
 412                         parse_trigrams(needle.str, ignore_case, &trigram_groups);
 413                 }
 414         }
 415
 416         unique_sort(
 417                 &trigram_groups,
 418                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
 419                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 420
 421         // Give them names for debugging.
 422         unsigned td_index = 0;
 423         for (TrigramDisjunction &td : trigram_groups) {
 424                 td.index = td_index++;
 425         }
 426
 427         // Collect which trigrams we need to look up in the hash table.
 428         unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
 429         for (TrigramDisjunction &td : trigram_groups) {
 430                 for (uint32_t trgm : td.trigram_alternatives) {
 431                         trigrams_to_lookup[trgm].push_back(&td);
 432                 }
 433         }
 434         if (trigrams_to_lookup.empty()) {
 435                 // Too short for trigram matching. Apply brute force.
 436                 // (We could have searched through all trigrams that matched
 437                 // the pattern and done a union of them, but that's a lot of
 438                 // work for fairly unclear gain.)
 439                 uint64_t matched = scan_all_docids(needles, fd, corpus, &engine);
 440                 if (only_count) {
 441                         printf("%zu\n", matched);
 442                 }
 443                 return;
 444         }
 445
 446         // Look them all up on disk.
 447         for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
 448                 corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }](const Trigram *trgmptr, size_t len) {
 449                         if (trgmptr == nullptr) {
 450                                 dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
 451                                 for (TrigramDisjunction *td : *trigram_groups) {
 452                                         --td->remaining_trigrams_to_read;
 453                                         if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
 454                                                 dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
 455                                                 if (only_count) {
 456                                                         printf("0\n");
 457                                                 }
 458                                                 exit(0);
 459                                         }
 460                                 }
 461                                 return;
 462                         }
 463                         for (TrigramDisjunction *td : *trigram_groups) {
 464                                 --td->remaining_trigrams_to_read;
 465                                 td->max_num_docids += trgmptr->num_docids;
 466                                 td->read_trigrams.emplace_back(*trgmptr, len);
 467                         }
 468                 });
 469         }
 470         engine.finish();
 471         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 472
 473         for (TrigramDisjunction &td : trigram_groups) {
 474                 // Reset for reads.
 475                 td.remaining_trigrams_to_read = td.read_trigrams.size();
 476
 477                 if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
 478                         dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
 479                 }
 480         }
 481
 482         // TODO: For case-insensitive (ie. more than one alternative in each),
 483         // prioritize the ones with fewer seeks?
 484         sort(trigram_groups.begin(), trigram_groups.end(),
 485              [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
 486                      return a.max_num_docids < b.max_num_docids;
 487              });
 488
 489         unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
 490         for (TrigramDisjunction &td : trigram_groups) {
 491                 for (uint32_t trgm : td.trigram_alternatives) {
 492                         uses_trigram[trgm].push_back(&td);
 493                 }
 494         }
 495
 496         unordered_set<uint32_t> trigrams_submitted_read;
 497         vector<uint32_t> cur_candidates, tmp, decoded;
 498         bool done = false;
 499         for (TrigramDisjunction &td : trigram_groups) {
 500                 if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
 501                         dprintf("%s has up to %u entries, ignoring the rest (will "
 502                                 "weed out false positives later)\n",
 503                                 print_td(td).c_str(), td.max_num_docids);
 504                         break;
 505                 }
 506
 507                 for (auto &[trgmptr, len] : td.read_trigrams) {
 508                         if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
 509                                 continue;
 510                         }
 511                         trigrams_submitted_read.insert(trgmptr.trgm);
 512                         // Only stay a certain amount ahead, so that we don't spend I/O
 513                         // on reading the latter, large posting lists. We are unlikely
 514                         // to need them anyway, even if they should come in first.
 515                         if (engine.get_waiting_reads() >= 5) {
 516                                 engine.finish();
 517                                 if (done)
 518                                         break;
 519                         }
 520                         engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
 521                                 if (done)
 522                                         return;
 523
 524                                 uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 525                                 const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 526                                 size_t num = trgmptr.num_docids;
 527                                 decoded.resize(num);
 528                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
 529
 530                                 assert(uses_trigram.count(trgm) != 0);
 531                                 bool was_empty = cur_candidates.empty();
 532                                 if (ignore_case) {
 533                                         dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 534                                 }
 535
 536                                 for (TrigramDisjunction *td : uses_trigram[trgm]) {
 537                                         done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
 538                                         if (done)
 539                                                 break;
 540                                 }
 541                                 if (!ignore_case) {
 542                                         if (was_empty) {
 543                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 544                                         } else if (cur_candidates.empty()) {
 545                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
 546                                         } else {
 547                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
 548                                         }
 549                                 }
 550                         });
 551                 }
 552         }
 553         engine.finish();
 554         if (done) {
 555                 return;
 556         }
 557         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 558                 1e3 * duration<float>(steady_clock::now() - start).count());
 559
 560         uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
 561         dprintf("Done in %.1f ms, found %zu matches.\n",
 562                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 563
 564         if (only_count) {
 565                 printf("%zu\n", matched);
 566         }
 567 }
 568
 569 string unescape_glob_to_plain_string(const string &needle)
 570 {
 571         string unescaped;
 572         for (size_t i = 0; i < needle.size(); i += read_unigram(needle, i).second) {
 573                 uint32_t ch = read_unigram(needle, i).first;
 574                 assert(ch != WILDCARD_UNIGRAM);
 575                 if (ch == PREMATURE_END_UNIGRAM) {
 576                         fprintf(stderr, "Pattern '%s' ended prematurely\n", needle.c_str());
 577                         exit(1);
 578                 }
 579                 unescaped.push_back(ch);
 580         }
 581         return unescaped;
 582 }
 583
 584 regex_t compile_regex(const string &needle)
 585 {
 586         regex_t re;
 587         int flags = REG_NOSUB;
 588         if (ignore_case) {
 589                 flags |= REG_ICASE;
 590         }
 591         if (use_extended_regex) {
 592                 flags |= REG_EXTENDED;
 593         }
 594         int err = regcomp(&re, needle.c_str(), flags);
 595         if (err != 0) {
 596                 char errbuf[256];
 597                 regerror(err, &re, errbuf, sizeof(errbuf));
 598                 fprintf(stderr, "Error when compiling regex '%s': %s\n", needle.c_str(), errbuf);
 599                 exit(1);
 600         }
 601         return re;
 602 }
 603
 604 void usage()
 605 {
 606         printf(
 607                 "Usage: plocate [OPTION]... PATTERN...\n"
 608                 "\n"
 609                 "  -c, --count            print number of matches instead of the matches\n"
 610                 "  -d, --database DBPATH  search for files in DBPATH\n"
 611                 "                         (default is " DEFAULT_DBPATH ")\n"
 612                 "  -i, --ignore-case      search case-insensitively\n"
 613                 "  -l, --limit LIMIT      stop after LIMIT matches\n"
 614                 "  -0, --null             delimit matches by NUL instead of newline\n"
 615                 "  -r, --regexp           interpret patterns as basic regexps (slow)\n"
 616                 "      --regex            interpret patterns as extended regexps (slow)\n"
 617                 "      --help             print this help\n"
 618                 "      --version          print version information\n");
 619 }
 620
 621 void version()
 622 {
 623         printf("plocate %s\n", PLOCATE_VERSION);
 624         printf("Copyright 2020 Steinar H. Gunderson\n");
 625         printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\n");
 626         printf("This is free software: you are free to change and redistribute it.\n");
 627         printf("There is NO WARRANTY, to the extent permitted by law.\n");
 628         exit(0);
 629 }
 630
 631 int main(int argc, char **argv)
 632 {
 633         constexpr int EXTENDED_REGEX = 1000;
 634         static const struct option long_options[] = {
 635                 { "help", no_argument, 0, 'h' },
 636                 { "count", no_argument, 0, 'c' },
 637                 { "database", required_argument, 0, 'd' },
 638                 { "ignore-case", no_argument, 0, 'i' },
 639                 { "limit", required_argument, 0, 'l' },
 640                 { "null", no_argument, 0, '0' },
 641                 { "version", no_argument, 0, 'V' },
 642                 { "regexp", no_argument, 0, 'r' },
 643                 { "regex", no_argument, 0, EXTENDED_REGEX },
 644                 { "debug", no_argument, 0, 'D' },  // Not documented.
 645                 { 0, 0, 0, 0 }
 646         };
 647
 648         setlocale(LC_ALL, "");
 649         for (;;) {
 650                 int option_index = 0;
 651                 int c = getopt_long(argc, argv, "cd:hil:n:0VD", long_options, &option_index);
 652                 if (c == -1) {
 653                         break;
 654                 }
 655                 switch (c) {
 656                 case 'c':
 657                         only_count = true;
 658                         break;
 659                 case 'd':
 660                         dbpath = strdup(optarg);
 661                         break;
 662                 case 'h':
 663                         usage();
 664                         exit(0);
 665                 case 'i':
 666                         ignore_case = true;
 667                         break;
 668                 case 'l':
 669                 case 'n':
 670                         limit_matches = atoll(optarg);
 671                         if (limit_matches <= 0) {
 672                                 fprintf(stderr, "Error: limit must be a strictly positive number.\n");
 673                                 exit(1);
 674                         }
 675                         break;
 676                 case '0':
 677                         print_nul = true;
 678                         break;
 679                 case 'r':
 680                         patterns_are_regex = true;
 681                         break;
 682                 case EXTENDED_REGEX:
 683                         patterns_are_regex = true;
 684                         use_extended_regex = true;
 685                         break;
 686                 case 'D':
 687                         use_debug = true;
 688                         break;
 689                 case 'V':
 690                         version();
 691                         break;
 692                 default:
 693                         exit(1);
 694                 }
 695         }
 696
 697         if (use_debug) {
 698                 // Debug information would leak information about which files exist,
 699                 // so drop setgid before we open the file; one would either need to run
 700                 // as root, or use a locally-built file.
 701                 if (setgid(getgid()) != 0) {
 702                         perror("setgid");
 703                         exit(EXIT_FAILURE);
 704                 }
 705         }
 706
 707         vector<Needle> needles;
 708         for (int i = optind; i < argc; ++i) {
 709                 Needle needle;
 710                 needle.str = argv[i];
 711
 712                 // See if there are any wildcard characters, which indicates we should treat it
 713                 // as an (anchored) glob.
 714                 bool any_wildcard = false;
 715                 for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) {
 716                         if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) {
 717                                 any_wildcard = true;
 718                                 break;
 719                         }
 720                 }
 721
 722                 if (patterns_are_regex) {
 723                         needle.type = Needle::REGEX;
 724                         needle.re = compile_regex(needle.str);
 725                 } else if (any_wildcard) {
 726                         needle.type = Needle::GLOB;
 727                 } else if (ignore_case) {
 728                         // strcasestr() doesn't handle locales correctly (even though LSB
 729                         // claims it should), but somehow, fnmatch() does, and it's about
 730                         // the same speed as using a regex.
 731                         needle.type = Needle::GLOB;
 732                         needle.str = "*" + needle.str + "*";
 733                 } else {
 734                         needle.type = Needle::STRSTR;
 735                         needle.str = unescape_glob_to_plain_string(needle.str);
 736                 }
 737                 needles.push_back(move(needle));
 738         }
 739         if (needles.empty()) {
 740                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 741                 exit(0);
 742         }
 743         do_search_file(needles, dbpath);
 744 }