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

   1 #include "db.h"
   2 #include "dprintf.h"
   3 #include "io_uring_engine.h"
   4 #include "parse_trigrams.h"
   5 #include "turbopfor.h"
   6 #include "unique_sort.h"
   7
   8 #include <algorithm>
   9 #include <assert.h>
  10 #include <chrono>
  11 #include <fcntl.h>
  12 #include <fnmatch.h>
  13 #include <functional>
  14 #include <getopt.h>
  15 #include <inttypes.h>
  16 #include <iosfwd>
  17 #include <iterator>
  18 #include <limits>
  19 #include <memory>
  20 #include <queue>
  21 #include <regex.h>
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <string>
  27 #include <string_view>
  28 #include <unistd.h>
  29 #include <unordered_map>
  30 #include <unordered_set>
  31 #include <utility>
  32 #include <vector>
  33 #include <zstd.h>
  34
  35 using namespace std;
  36 using namespace std::chrono;
  37
  38 #define DEFAULT_DBPATH "/var/lib/mlocate/plocate.db"
  39
  40 const char *dbpath = DEFAULT_DBPATH;
  41 bool ignore_case = false;
  42 bool only_count = false;
  43 bool print_nul = false;
  44 bool use_debug = false;
  45 bool patterns_are_regex = false;
  46 bool use_extended_regex = false;
  47 int64_t limit_matches = numeric_limits<int64_t>::max();
  48
  49 class Serializer {
  50 public:
  51         bool ready_to_print(int seq) { return next_seq == seq; }
  52         void print_delayed(int seq, const vector<string> msg);
  53         void release_current();
  54
  55 private:
  56         int next_seq = 0;
  57         struct Element {
  58                 int seq;
  59                 vector<string> msg;
  60
  61                 bool operator<(const Element &other) const
  62                 {
  63                         return seq > other.seq;
  64                 }
  65         };
  66         priority_queue<Element> pending;
  67 };
  68
  69 void Serializer::print_delayed(int seq, const vector<string> msg)
  70 {
  71         pending.push(Element{ seq, move(msg) });
  72 }
  73
  74 void Serializer::release_current()
  75 {
  76         ++next_seq;
  77
  78         // See if any delayed prints can now be dealt with.
  79         while (!pending.empty() && pending.top().seq == next_seq) {
  80                 if (limit_matches-- <= 0)
  81                         return;
  82                 for (const string &msg : pending.top().msg) {
  83                         if (print_nul) {
  84                                 printf("%s%c", msg.c_str(), 0);
  85                         } else {
  86                                 printf("%s\n", msg.c_str());
  87                         }
  88                 }
  89                 pending.pop();
  90                 ++next_seq;
  91         }
  92 }
  93
  94 struct Needle {
  95         enum { STRSTR,
  96                REGEX,
  97                GLOB } type;
  98         string str;  // Filled in no matter what.
  99         regex_t re;  // For REGEX.
 100 };
 101
 102 bool matches(const Needle &needle, const char *haystack)
 103 {
 104         if (needle.type == Needle::STRSTR) {
 105                 return strstr(haystack, needle.str.c_str()) != nullptr;
 106         } else if (needle.type == Needle::GLOB) {
 107                 int flags = ignore_case ? FNM_CASEFOLD : 0;
 108                 return fnmatch(needle.str.c_str(), haystack, flags) == 0;
 109         } else {
 110                 assert(needle.type == Needle::REGEX);
 111                 return regexec(&needle.re, haystack, /*nmatch=*/0, /*pmatch=*/nullptr, /*flags=*/0) == 0;
 112         }
 113 }
 114
 115 bool has_access(const char *filename,
 116                 unordered_map<string, bool> *access_rx_cache)
 117 {
 118         const char *end = strchr(filename + 1, '/');
 119         while (end != nullptr) {
 120                 string parent_path(filename, end);
 121                 auto it = access_rx_cache->find(parent_path);
 122                 bool ok;
 123                 if (it == access_rx_cache->end()) {
 124                         ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
 125                         access_rx_cache->emplace(move(parent_path), ok);
 126                 } else {
 127                         ok = it->second;
 128                 }
 129                 if (!ok) {
 130                         return false;
 131                 }
 132                 end = strchr(end + 1, '/');
 133         }
 134
 135         return true;
 136 }
 137
 138 class Corpus {
 139 public:
 140         Corpus(int fd, IOUringEngine *engine);
 141         ~Corpus();
 142         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
 143         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
 144         size_t get_num_filename_blocks() const;
 145         off_t offset_for_block(uint32_t docid) const
 146         {
 147                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
 148         }
 149
 150 public:
 151         const int fd;
 152         IOUringEngine *const engine;
 153
 154         Header hdr;
 155 };
 156
 157 Corpus::Corpus(int fd, IOUringEngine *engine)
 158         : fd(fd), engine(engine)
 159 {
 160         // Enable to test cold-cache behavior (except for access()).
 161         if (false) {
 162                 off_t len = lseek(fd, 0, SEEK_END);
 163                 if (len == -1) {
 164                         perror("lseek");
 165                         exit(1);
 166                 }
 167                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
 168         }
 169
 170         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 171         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 172                 fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
 173                 exit(1);
 174         }
 175         if (hdr.version != 0) {
 176                 fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version);
 177                 exit(1);
 178         }
 179 }
 180
 181 Corpus::~Corpus()
 182 {
 183         close(fd);
 184 }
 185
 186 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 187 {
 188         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 189         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) {
 190                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 191                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 192                         if (trgmptr[i].trgm == trgm) {
 193                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 194                                 return;
 195                         }
 196                 }
 197
 198                 // Not found.
 199                 cb(nullptr, 0);
 200         });
 201 }
 202
 203 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 204 {
 205         // Read the file offset from this docid and the next one.
 206         // This is always allowed, since we have a sentinel block at the end.
 207         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 208                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 209                 off_t offset = ptr[0];
 210                 size_t len = ptr[1] - ptr[0];
 211                 engine->submit_read(fd, len, offset, cb);
 212         });
 213 }
 214
 215 size_t Corpus::get_num_filename_blocks() const
 216 {
 217         return hdr.num_docids;
 218 }
 219
 220 uint64_t scan_file_block(const vector<Needle> &needles, string_view compressed,
 221                          unordered_map<string, bool> *access_rx_cache, int seq,
 222                          Serializer *serializer)
 223 {
 224         uint64_t matched = 0;
 225
 226         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 227         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 228                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 229                 exit(1);
 230         }
 231
 232         string block;
 233         block.resize(uncompressed_len + 1);
 234
 235         size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(),
 236                                      compressed.size());
 237         if (ZSTD_isError(err)) {
 238                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 239                 exit(1);
 240         }
 241         block[block.size() - 1] = '\0';
 242
 243         bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq));
 244         vector<string> delayed;
 245
 246         for (const char *filename = block.data();
 247              filename != block.data() + block.size();
 248              filename += strlen(filename) + 1) {
 249                 bool found = true;
 250                 for (const Needle &needle : needles) {
 251                         if (!matches(needle, filename)) {
 252                                 found = false;
 253                                 break;
 254                         }
 255                 }
 256                 if (found && has_access(filename, access_rx_cache)) {
 257                         if (limit_matches-- <= 0)
 258                                 break;
 259                         ++matched;
 260                         if (only_count)
 261                                 continue;
 262                         if (immediate_print) {
 263                                 if (print_nul) {
 264                                         printf("%s%c", filename, 0);
 265                                 } else {
 266                                         printf("%s\n", filename);
 267                                 }
 268                         } else {
 269                                 delayed.push_back(filename);
 270                         }
 271                 }
 272         }
 273         if (serializer != nullptr && !only_count) {
 274                 if (immediate_print) {
 275                         serializer->release_current();
 276                 } else {
 277                         serializer->print_delayed(seq, move(delayed));
 278                 }
 279         }
 280         return matched;
 281 }
 282
 283 size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 284 {
 285         Serializer docids_in_order;
 286         unordered_map<string, bool> access_rx_cache;
 287         uint64_t matched = 0;
 288         for (size_t i = 0; i < docids.size(); ++i) {
 289                 uint32_t docid = docids[i];
 290                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string_view compressed) {
 291                         matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order);
 292                 });
 293         }
 294         engine->finish();
 295         return matched;
 296 }
 297
 298 // We do this sequentially, as it's faster than scattering
 299 // a lot of I/O through io_uring and hoping the kernel will
 300 // coalesce it plus readahead for us.
 301 uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus, IOUringEngine *engine)
 302 {
 303         unordered_map<string, bool> access_rx_cache;
 304         uint32_t num_blocks = corpus.get_num_filename_blocks();
 305         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 306         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 307         string compressed;
 308         uint64_t matched = 0;
 309         for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
 310                 uint32_t last_docid = std::min(io_docid + 32, num_blocks);
 311                 size_t io_len = offsets[last_docid] - offsets[io_docid];
 312                 if (compressed.size() < io_len) {
 313                         compressed.resize(io_len);
 314                 }
 315                 complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
 316
 317                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 318                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 319                         size_t len = offsets[docid + 1] - offsets[docid];
 320                         matched += scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr);
 321                         if (limit_matches <= 0)
 322                                 return matched;
 323                 }
 324         }
 325         return matched;
 326 }
 327
 328 // Takes the given posting list, unions it into the parts of the trigram disjunction
 329 // already read; if the list is complete, intersects with “cur_candidates”.
 330 //
 331 // Returns true if the search should be aborted (we are done).
 332 bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
 333 {
 334         if (td->docids.empty()) {
 335                 td->docids = move(decoded);
 336         } else {
 337                 tmp->clear();
 338                 set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
 339                 swap(*tmp, td->docids);
 340         }
 341         if (--td->remaining_trigrams_to_read > 0) {
 342                 // Need to wait for more.
 343                 if (ignore_case) {
 344                         dprintf("  ... %u reads left in OR group %u (%zu docids in list)\n",
 345                                 td->remaining_trigrams_to_read, td->index, td->docids.size());
 346                 }
 347                 return false;
 348         }
 349         if (cur_candidates->empty()) {
 350                 if (ignore_case) {
 351                         dprintf("  ... all reads done for OR group %u (%zu docids)\n",
 352                                 td->index, td->docids.size());
 353                 }
 354                 *cur_candidates = move(td->docids);
 355         } else {
 356                 tmp->clear();
 357                 set_intersection(cur_candidates->begin(), cur_candidates->end(),
 358                                  td->docids.begin(), td->docids.end(),
 359                                  back_inserter(*tmp));
 360                 swap(*cur_candidates, *tmp);
 361                 if (ignore_case) {
 362                         if (cur_candidates->empty()) {
 363                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n",
 364                                         td->index, td->docids.size());
 365                                 return true;
 366                         } else {
 367                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n",
 368                                         td->index, td->docids.size(), cur_candidates->size());
 369                         }
 370                 }
 371         }
 372         return false;
 373 }
 374
 375 void do_search_file(const vector<Needle> &needles, const char *filename)
 376 {
 377         int fd = open(filename, O_RDONLY);
 378         if (fd == -1) {
 379                 perror(filename);
 380                 exit(1);
 381         }
 382
 383         // Drop privileges.
 384         if (setgid(getgid()) != 0) {
 385                 perror("setgid");
 386                 exit(EXIT_FAILURE);
 387         }
 388
 389         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 390         if (access("/", R_OK | X_OK)) {
 391                 // We can't find anything, no need to bother...
 392                 return;
 393         }
 394
 395         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 396         Corpus corpus(fd, &engine);
 397         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 398
 399         vector<TrigramDisjunction> trigram_groups;
 400         if (patterns_are_regex) {
 401                 // We could parse the regex to find trigrams that have to be there
 402                 // (there are actually known algorithms to deal with disjunctions
 403                 // and such, too), but for now, we just go brute force.
 404                 // Using locate with regexes is pretty niche.
 405         } else {
 406                 for (const Needle &needle : needles) {
 407                         parse_trigrams(needle.str, ignore_case, &trigram_groups);
 408                 }
 409         }
 410
 411         unique_sort(
 412                 &trigram_groups,
 413                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
 414                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 415
 416         // Give them names for debugging.
 417         unsigned td_index = 0;
 418         for (TrigramDisjunction &td : trigram_groups) {
 419                 td.index = td_index++;
 420         }
 421
 422         // Collect which trigrams we need to look up in the hash table.
 423         unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
 424         for (TrigramDisjunction &td : trigram_groups) {
 425                 for (uint32_t trgm : td.trigram_alternatives) {
 426                         trigrams_to_lookup[trgm].push_back(&td);
 427                 }
 428         }
 429         if (trigrams_to_lookup.empty()) {
 430                 // Too short for trigram matching. Apply brute force.
 431                 // (We could have searched through all trigrams that matched
 432                 // the pattern and done a union of them, but that's a lot of
 433                 // work for fairly unclear gain.)
 434                 uint64_t matched = scan_all_docids(needles, fd, corpus, &engine);
 435                 if (only_count) {
 436                         printf("%" PRId64 "\n", matched);
 437                 }
 438                 return;
 439         }
 440
 441         // Look them all up on disk.
 442         for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
 443                 corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }](const Trigram *trgmptr, size_t len) {
 444                         if (trgmptr == nullptr) {
 445                                 dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
 446                                 for (TrigramDisjunction *td : *trigram_groups) {
 447                                         --td->remaining_trigrams_to_read;
 448                                         if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
 449                                                 dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
 450                                                 if (only_count) {
 451                                                         printf("0\n");
 452                                                 }
 453                                                 exit(0);
 454                                         }
 455                                 }
 456                                 return;
 457                         }
 458                         for (TrigramDisjunction *td : *trigram_groups) {
 459                                 --td->remaining_trigrams_to_read;
 460                                 td->max_num_docids += trgmptr->num_docids;
 461                                 td->read_trigrams.emplace_back(*trgmptr, len);
 462                         }
 463                 });
 464         }
 465         engine.finish();
 466         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 467
 468         for (TrigramDisjunction &td : trigram_groups) {
 469                 // Reset for reads.
 470                 td.remaining_trigrams_to_read = td.read_trigrams.size();
 471
 472                 if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
 473                         dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
 474                 }
 475         }
 476
 477         // TODO: For case-insensitive (ie. more than one alternative in each),
 478         // prioritize the ones with fewer seeks?
 479         sort(trigram_groups.begin(), trigram_groups.end(),
 480              [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
 481                      return a.max_num_docids < b.max_num_docids;
 482              });
 483
 484         unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
 485         for (TrigramDisjunction &td : trigram_groups) {
 486                 for (uint32_t trgm : td.trigram_alternatives) {
 487                         uses_trigram[trgm].push_back(&td);
 488                 }
 489         }
 490
 491         unordered_set<uint32_t> trigrams_submitted_read;
 492         vector<uint32_t> cur_candidates, tmp, decoded;
 493         bool done = false;
 494         for (TrigramDisjunction &td : trigram_groups) {
 495                 if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
 496                         dprintf("%s has up to %u entries, ignoring the rest (will "
 497                                 "weed out false positives later)\n",
 498                                 print_td(td).c_str(), td.max_num_docids);
 499                         break;
 500                 }
 501
 502                 for (auto &[trgmptr, len] : td.read_trigrams) {
 503                         if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
 504                                 continue;
 505                         }
 506                         trigrams_submitted_read.insert(trgmptr.trgm);
 507                         // Only stay a certain amount ahead, so that we don't spend I/O
 508                         // on reading the latter, large posting lists. We are unlikely
 509                         // to need them anyway, even if they should come in first.
 510                         if (engine.get_waiting_reads() >= 5) {
 511                                 engine.finish();
 512                                 if (done)
 513                                         break;
 514                         }
 515                         engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
 516                                 if (done)
 517                                         return;
 518
 519                                 uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 520                                 const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 521                                 size_t num = trgmptr.num_docids;
 522                                 decoded.resize(num);
 523                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
 524
 525                                 assert(uses_trigram.count(trgm) != 0);
 526                                 bool was_empty = cur_candidates.empty();
 527                                 if (ignore_case) {
 528                                         dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 529                                 }
 530
 531                                 for (TrigramDisjunction *td : uses_trigram[trgm]) {
 532                                         done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
 533                                         if (done)
 534                                                 break;
 535                                 }
 536                                 if (!ignore_case) {
 537                                         if (was_empty) {
 538                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 539                                         } else if (cur_candidates.empty()) {
 540                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
 541                                         } else {
 542                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
 543                                         }
 544                                 }
 545                         });
 546                 }
 547         }
 548         engine.finish();
 549         if (done) {
 550                 return;
 551         }
 552         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 553                 1e3 * duration<float>(steady_clock::now() - start).count());
 554
 555         uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
 556         dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
 557                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 558
 559         if (only_count) {
 560                 printf("%" PRId64 "\n", matched);
 561         }
 562 }
 563
 564 string unescape_glob_to_plain_string(const string &needle)
 565 {
 566         string unescaped;
 567         for (size_t i = 0; i < needle.size(); i += read_unigram(needle, i).second) {
 568                 uint32_t ch = read_unigram(needle, i).first;
 569                 assert(ch != WILDCARD_UNIGRAM);
 570                 if (ch == PREMATURE_END_UNIGRAM) {
 571                         fprintf(stderr, "Pattern '%s' ended prematurely\n", needle.c_str());
 572                         exit(1);
 573                 }
 574                 unescaped.push_back(ch);
 575         }
 576         return unescaped;
 577 }
 578
 579 regex_t compile_regex(const string &needle)
 580 {
 581         regex_t re;
 582         int flags = REG_NOSUB;
 583         if (ignore_case) {
 584                 flags |= REG_ICASE;
 585         }
 586         if (use_extended_regex) {
 587                 flags |= REG_EXTENDED;
 588         }
 589         int err = regcomp(&re, needle.c_str(), flags);
 590         if (err != 0) {
 591                 char errbuf[256];
 592                 regerror(err, &re, errbuf, sizeof(errbuf));
 593                 fprintf(stderr, "Error when compiling regex '%s': %s\n", needle.c_str(), errbuf);
 594                 exit(1);
 595         }
 596         return re;
 597 }
 598
 599 void usage()
 600 {
 601         printf(
 602                 "Usage: plocate [OPTION]... PATTERN...\n"
 603                 "\n"
 604                 "  -c, --count            print number of matches instead of the matches\n"
 605                 "  -d, --database DBPATH  search for files in DBPATH\n"
 606                 "                         (default is " DEFAULT_DBPATH ")\n"
 607                 "  -i, --ignore-case      search case-insensitively\n"
 608                 "  -l, --limit LIMIT      stop after LIMIT matches\n"
 609                 "  -0, --null             delimit matches by NUL instead of newline\n"
 610                 "  -r, --regexp           interpret patterns as basic regexps (slow)\n"
 611                 "      --regex            interpret patterns as extended regexps (slow)\n"
 612                 "      --help             print this help\n"
 613                 "      --version          print version information\n");
 614 }
 615
 616 void version()
 617 {
 618         printf("plocate %s\n", PLOCATE_VERSION);
 619         printf("Copyright 2020 Steinar H. Gunderson\n");
 620         printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\n");
 621         printf("This is free software: you are free to change and redistribute it.\n");
 622         printf("There is NO WARRANTY, to the extent permitted by law.\n");
 623         exit(0);
 624 }
 625
 626 int main(int argc, char **argv)
 627 {
 628         constexpr int EXTENDED_REGEX = 1000;
 629         static const struct option long_options[] = {
 630                 { "help", no_argument, 0, 'h' },
 631                 { "count", no_argument, 0, 'c' },
 632                 { "database", required_argument, 0, 'd' },
 633                 { "ignore-case", no_argument, 0, 'i' },
 634                 { "limit", required_argument, 0, 'l' },
 635                 { "null", no_argument, 0, '0' },
 636                 { "version", no_argument, 0, 'V' },
 637                 { "regexp", no_argument, 0, 'r' },
 638                 { "regex", no_argument, 0, EXTENDED_REGEX },
 639                 { "debug", no_argument, 0, 'D' },  // Not documented.
 640                 { 0, 0, 0, 0 }
 641         };
 642
 643         setlocale(LC_ALL, "");
 644         for (;;) {
 645                 int option_index = 0;
 646                 int c = getopt_long(argc, argv, "cd:hil:n:0VD", long_options, &option_index);
 647                 if (c == -1) {
 648                         break;
 649                 }
 650                 switch (c) {
 651                 case 'c':
 652                         only_count = true;
 653                         break;
 654                 case 'd':
 655                         dbpath = strdup(optarg);
 656                         break;
 657                 case 'h':
 658                         usage();
 659                         exit(0);
 660                 case 'i':
 661                         ignore_case = true;
 662                         break;
 663                 case 'l':
 664                 case 'n':
 665                         limit_matches = atoll(optarg);
 666                         if (limit_matches <= 0) {
 667                                 fprintf(stderr, "Error: limit must be a strictly positive number.\n");
 668                                 exit(1);
 669                         }
 670                         break;
 671                 case '0':
 672                         print_nul = true;
 673                         break;
 674                 case 'r':
 675                         patterns_are_regex = true;
 676                         break;
 677                 case EXTENDED_REGEX:
 678                         patterns_are_regex = true;
 679                         use_extended_regex = true;
 680                         break;
 681                 case 'D':
 682                         use_debug = true;
 683                         break;
 684                 case 'V':
 685                         version();
 686                         break;
 687                 default:
 688                         exit(1);
 689                 }
 690         }
 691
 692         if (use_debug) {
 693                 // Debug information would leak information about which files exist,
 694                 // so drop setgid before we open the file; one would either need to run
 695                 // as root, or use a locally-built file.
 696                 if (setgid(getgid()) != 0) {
 697                         perror("setgid");
 698                         exit(EXIT_FAILURE);
 699                 }
 700         }
 701
 702         vector<Needle> needles;
 703         for (int i = optind; i < argc; ++i) {
 704                 Needle needle;
 705                 needle.str = argv[i];
 706
 707                 // See if there are any wildcard characters, which indicates we should treat it
 708                 // as an (anchored) glob.
 709                 bool any_wildcard = false;
 710                 for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) {
 711                         if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) {
 712                                 any_wildcard = true;
 713                                 break;
 714                         }
 715                 }
 716
 717                 if (patterns_are_regex) {
 718                         needle.type = Needle::REGEX;
 719                         needle.re = compile_regex(needle.str);
 720                 } else if (any_wildcard) {
 721                         needle.type = Needle::GLOB;
 722                 } else if (ignore_case) {
 723                         // strcasestr() doesn't handle locales correctly (even though LSB
 724                         // claims it should), but somehow, fnmatch() does, and it's about
 725                         // the same speed as using a regex.
 726                         needle.type = Needle::GLOB;
 727                         needle.str = "*" + needle.str + "*";
 728                 } else {
 729                         needle.type = Needle::STRSTR;
 730                         needle.str = unescape_glob_to_plain_string(needle.str);
 731                 }
 732                 needles.push_back(move(needle));
 733         }
 734         if (needles.empty()) {
 735                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 736                 exit(0);
 737         }
 738         do_search_file(needles, dbpath);
 739 }