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 <map>
  20 #include <memory>
  21 #include <queue>
  22 #include <regex.h>
  23 #include <stdint.h>
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <string.h>
  27 #include <string>
  28 #include <string_view>
  29 #include <unistd.h>
  30 #include <unordered_map>
  31 #include <unordered_set>
  32 #include <utility>
  33 #include <vector>
  34 #include <zstd.h>
  35
  36 using namespace std;
  37 using namespace std::chrono;
  38
  39 #define DEFAULT_DBPATH "/var/lib/mlocate/plocate.db"
  40
  41 const char *dbpath = DEFAULT_DBPATH;
  42 bool ignore_case = false;
  43 bool only_count = false;
  44 bool print_nul = false;
  45 bool use_debug = false;
  46 bool patterns_are_regex = false;
  47 bool use_extended_regex = false;
  48 int64_t limit_matches = numeric_limits<int64_t>::max();
  49 int64_t limit_left = numeric_limits<int64_t>::max();
  50
  51 steady_clock::time_point start;
  52
  53 void apply_limit()
  54 {
  55         if (--limit_left > 0) {
  56                 return;
  57         }
  58         dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
  59                 1e3 * duration<float>(steady_clock::now() - start).count(), limit_matches);
  60         if (only_count) {
  61                 printf("%ld\n", limit_matches);
  62         }
  63         exit(0);
  64 }
  65
  66 class Serializer {
  67 public:
  68         ~Serializer() { assert(limit_left <= 0 || pending.empty()); }
  69         void print(uint64_t seq, uint64_t skip, const string msg);
  70
  71 private:
  72         uint64_t next_seq = 0;
  73         struct Element {
  74                 uint64_t seq, skip;
  75                 string msg;
  76
  77                 bool operator<(const Element &other) const
  78                 {
  79                         return seq > other.seq;
  80                 }
  81         };
  82         priority_queue<Element> pending;
  83 };
  84
  85 void Serializer::print(uint64_t seq, uint64_t skip, const string msg)
  86 {
  87         if (only_count) {
  88                 if (!msg.empty()) {
  89                         apply_limit();
  90                 }
  91                 return;
  92         }
  93
  94         if (next_seq != seq) {
  95                 pending.push(Element{ seq, skip, move(msg) });
  96                 return;
  97         }
  98
  99         if (!msg.empty()) {
 100                 if (print_nul) {
 101                         printf("%s%c", msg.c_str(), 0);
 102                 } else {
 103                         printf("%s\n", msg.c_str());
 104                 }
 105                 apply_limit();
 106         }
 107         next_seq += skip;
 108
 109         // See if any delayed prints can now be dealt with.
 110         while (!pending.empty() && pending.top().seq == next_seq) {
 111                 if (!pending.top().msg.empty()) {
 112                         if (print_nul) {
 113                                 printf("%s%c", pending.top().msg.c_str(), 0);
 114                         } else {
 115                                 printf("%s\n", pending.top().msg.c_str());
 116                         }
 117                         apply_limit();
 118                 }
 119                 next_seq += pending.top().skip;
 120                 pending.pop();
 121         }
 122 }
 123
 124 struct Needle {
 125         enum { STRSTR,
 126                REGEX,
 127                GLOB } type;
 128         string str;  // Filled in no matter what.
 129         regex_t re;  // For REGEX.
 130 };
 131
 132 bool matches(const Needle &needle, const char *haystack)
 133 {
 134         if (needle.type == Needle::STRSTR) {
 135                 return strstr(haystack, needle.str.c_str()) != nullptr;
 136         } else if (needle.type == Needle::GLOB) {
 137                 int flags = ignore_case ? FNM_CASEFOLD : 0;
 138                 return fnmatch(needle.str.c_str(), haystack, flags) == 0;
 139         } else {
 140                 assert(needle.type == Needle::REGEX);
 141                 return regexec(&needle.re, haystack, /*nmatch=*/0, /*pmatch=*/nullptr, /*flags=*/0) == 0;
 142         }
 143 }
 144
 145 class AccessRXCache {
 146 public:
 147         AccessRXCache(IOUringEngine *engine)
 148                 : engine(engine) {}
 149         void check_access(const char *filename, bool allow_async, function<void(bool)> cb);
 150
 151 private:
 152         unordered_map<string, bool> cache;
 153         struct PendingStat {
 154                 string filename;
 155                 function<void(bool)> cb;
 156         };
 157         map<string, vector<PendingStat>> pending_stats;
 158         IOUringEngine *engine;
 159 };
 160
 161 void AccessRXCache::check_access(const char *filename, bool allow_async, function<void(bool)> cb)
 162 {
 163         if (!engine->get_supports_stat()) {
 164                 allow_async = false;
 165         }
 166
 167         for (const char *end = strchr(filename + 1, '/'); end != nullptr; end = strchr(end + 1, '/')) {
 168                 string parent_path(filename, end - filename);  // string_view from C++20.
 169                 auto cache_it = cache.find(parent_path);
 170                 if (cache_it != cache.end()) {
 171                         // Found in the cache.
 172                         if (!cache_it->second) {
 173                                 cb(false);
 174                                 return;
 175                         }
 176                         continue;
 177                 }
 178
 179                 if (!allow_async) {
 180                         bool ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
 181                         cache.emplace(parent_path, ok);
 182                         if (!ok) {
 183                                 cb(false);
 184                                 return;
 185                         }
 186                         continue;
 187                 }
 188
 189                 // We want to call access(), but it could block on I/O. io_uring doesn't support
 190                 // access(), but we can do a dummy asynchonous statx() to populate the kernel's cache,
 191                 // which nearly always makes the next access() instantaneous.
 192
 193                 // See if there's already a pending stat that matches this,
 194                 // or is a subdirectory.
 195                 auto it = pending_stats.lower_bound(parent_path);
 196                 if (it != pending_stats.end() && it->first.size() >= parent_path.size() &&
 197                     it->first.compare(0, parent_path.size(), parent_path) == 0) {
 198                         it->second.emplace_back(PendingStat{ filename, move(cb) });
 199                 } else {
 200                         it = pending_stats.emplace(filename, vector<PendingStat>{}).first;
 201                         engine->submit_stat(filename, [this, it, filename{ strdup(filename) }, cb{ move(cb) }] {
 202                                 // The stat returned, so now do the actual access() calls.
 203                                 // All of them should be in cache, so don't fire off new statx()
 204                                 // calls during that check.
 205                                 check_access(filename, /*allow_async=*/false, move(cb));
 206                                 free(filename);
 207
 208                                 // Call all others that waited for the same stat() to finish.
 209                                 // They may fire off new stat() calls if needed.
 210                                 vector<PendingStat> pending = move(it->second);
 211                                 pending_stats.erase(it);
 212                                 for (PendingStat &ps : pending) {
 213                                         check_access(ps.filename.c_str(), /*allow_async=*/true, move(ps.cb));
 214                                 }
 215                         });
 216                 }
 217                 return;  // The rest will happen in async context.
 218         }
 219
 220         // Passed all checks.
 221         cb(true);
 222 }
 223
 224 class Corpus {
 225 public:
 226         Corpus(int fd, IOUringEngine *engine);
 227         ~Corpus();
 228         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
 229         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
 230         size_t get_num_filename_blocks() const;
 231         off_t offset_for_block(uint32_t docid) const
 232         {
 233                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
 234         }
 235
 236 public:
 237         const int fd;
 238         IOUringEngine *const engine;
 239
 240         Header hdr;
 241 };
 242
 243 Corpus::Corpus(int fd, IOUringEngine *engine)
 244         : fd(fd), engine(engine)
 245 {
 246         // Enable to test cold-cache behavior (except for access()).
 247         if (false) {
 248                 off_t len = lseek(fd, 0, SEEK_END);
 249                 if (len == -1) {
 250                         perror("lseek");
 251                         exit(1);
 252                 }
 253                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
 254         }
 255
 256         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 257         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 258                 fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
 259                 exit(1);
 260         }
 261         if (hdr.version != 0) {
 262                 fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version);
 263                 exit(1);
 264         }
 265 }
 266
 267 Corpus::~Corpus()
 268 {
 269         close(fd);
 270 }
 271
 272 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 273 {
 274         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 275         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) {
 276                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 277                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 278                         if (trgmptr[i].trgm == trgm) {
 279                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 280                                 return;
 281                         }
 282                 }
 283
 284                 // Not found.
 285                 cb(nullptr, 0);
 286         });
 287 }
 288
 289 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 290 {
 291         // Read the file offset from this docid and the next one.
 292         // This is always allowed, since we have a sentinel block at the end.
 293         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 294                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 295                 off_t offset = ptr[0];
 296                 size_t len = ptr[1] - ptr[0];
 297                 engine->submit_read(fd, len, offset, cb);
 298         });
 299 }
 300
 301 size_t Corpus::get_num_filename_blocks() const
 302 {
 303         return hdr.num_docids;
 304 }
 305
 306 void scan_file_block(const vector<Needle> &needles, string_view compressed,
 307                      AccessRXCache *access_rx_cache, uint64_t seq, Serializer *serializer, IOUringEngine *engine,
 308                      size_t *matched)
 309 {
 310         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 311         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 312                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 313                 exit(1);
 314         }
 315
 316         string block;
 317         block.resize(uncompressed_len + 1);
 318
 319         size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(),
 320                                      compressed.size());
 321         if (ZSTD_isError(err)) {
 322                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 323                 exit(1);
 324         }
 325         block[block.size() - 1] = '\0';
 326
 327         auto test_candidate = [&](const char *filename, uint64_t local_seq, uint64_t next_seq) {
 328                 access_rx_cache->check_access(filename, /*allow_async=*/true, [matched, serializer, local_seq, next_seq, filename{ strdup(filename) }](bool ok) {
 329                         if (ok) {
 330                                 ++*matched;
 331                                 serializer->print(local_seq, next_seq - local_seq, filename);
 332                         } else {
 333                                 serializer->print(local_seq, next_seq - local_seq, "");
 334                         }
 335                         free(filename);
 336                 });
 337         };
 338
 339         // We need to know the next sequence number before inserting into Serializer,
 340         // so always buffer one candidate.
 341         const char *pending_candidate = nullptr;
 342
 343         uint64_t local_seq = seq << 32;
 344         for (const char *filename = block.data();
 345              filename != block.data() + block.size();
 346              filename += strlen(filename) + 1) {
 347                 bool found = true;
 348                 for (const Needle &needle : needles) {
 349                         if (!matches(needle, filename)) {
 350                                 found = false;
 351                                 break;
 352                         }
 353                 }
 354                 if (found) {
 355                         if (pending_candidate != nullptr) {
 356                                 test_candidate(pending_candidate, local_seq, local_seq + 1);
 357                                 ++local_seq;
 358                         }
 359                         pending_candidate = filename;
 360                 }
 361         }
 362         if (pending_candidate == nullptr) {
 363                 serializer->print(seq << 32, 1ULL << 32, "");
 364         } else {
 365                 test_candidate(pending_candidate, local_seq, (seq + 1) << 32);
 366         }
 367 }
 368
 369 size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 370 {
 371         Serializer docids_in_order;
 372         AccessRXCache access_rx_cache(engine);
 373         uint64_t matched = 0;
 374         for (size_t i = 0; i < docids.size(); ++i) {
 375                 uint32_t docid = docids[i];
 376                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, engine, &docids_in_order](string_view compressed) {
 377                         scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order, engine, &matched);
 378                 });
 379         }
 380         engine->finish();
 381         return matched;
 382 }
 383
 384 // We do this sequentially, as it's faster than scattering
 385 // a lot of I/O through io_uring and hoping the kernel will
 386 // coalesce it plus readahead for us.
 387 uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus, IOUringEngine *engine)
 388 {
 389         AccessRXCache access_rx_cache(engine);
 390         uint32_t num_blocks = corpus.get_num_filename_blocks();
 391         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 392         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 393         string compressed;
 394         uint64_t matched = 0;
 395         for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
 396                 uint32_t last_docid = std::min(io_docid + 32, num_blocks);
 397                 size_t io_len = offsets[last_docid] - offsets[io_docid];
 398                 if (compressed.size() < io_len) {
 399                         compressed.resize(io_len);
 400                 }
 401                 complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
 402
 403                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 404                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 405                         size_t len = offsets[docid + 1] - offsets[docid];
 406                         scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr, engine, &matched);
 407                 }
 408         }
 409         return matched;
 410 }
 411
 412 // Takes the given posting list, unions it into the parts of the trigram disjunction
 413 // already read; if the list is complete, intersects with “cur_candidates”.
 414 //
 415 // Returns true if the search should be aborted (we are done).
 416 bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
 417 {
 418         if (td->docids.empty()) {
 419                 td->docids = move(decoded);
 420         } else {
 421                 tmp->clear();
 422                 set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
 423                 swap(*tmp, td->docids);
 424         }
 425         if (--td->remaining_trigrams_to_read > 0) {
 426                 // Need to wait for more.
 427                 if (ignore_case) {
 428                         dprintf("  ... %u reads left in OR group %u (%zu docids in list)\n",
 429                                 td->remaining_trigrams_to_read, td->index, td->docids.size());
 430                 }
 431                 return false;
 432         }
 433         if (cur_candidates->empty()) {
 434                 if (ignore_case) {
 435                         dprintf("  ... all reads done for OR group %u (%zu docids)\n",
 436                                 td->index, td->docids.size());
 437                 }
 438                 *cur_candidates = move(td->docids);
 439         } else {
 440                 tmp->clear();
 441                 set_intersection(cur_candidates->begin(), cur_candidates->end(),
 442                                  td->docids.begin(), td->docids.end(),
 443                                  back_inserter(*tmp));
 444                 swap(*cur_candidates, *tmp);
 445                 if (ignore_case) {
 446                         if (cur_candidates->empty()) {
 447                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n",
 448                                         td->index, td->docids.size());
 449                                 return true;
 450                         } else {
 451                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n",
 452                                         td->index, td->docids.size(), cur_candidates->size());
 453                         }
 454                 }
 455         }
 456         return false;
 457 }
 458
 459 void do_search_file(const vector<Needle> &needles, const char *filename)
 460 {
 461         int fd = open(filename, O_RDONLY);
 462         if (fd == -1) {
 463                 perror(filename);
 464                 exit(1);
 465         }
 466
 467         // Drop privileges.
 468         if (setgid(getgid()) != 0) {
 469                 perror("setgid");
 470                 exit(EXIT_FAILURE);
 471         }
 472
 473         start = steady_clock::now();
 474         if (access("/", R_OK | X_OK)) {
 475                 // We can't find anything, no need to bother...
 476                 return;
 477         }
 478
 479         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 480         Corpus corpus(fd, &engine);
 481         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 482
 483         vector<TrigramDisjunction> trigram_groups;
 484         if (patterns_are_regex) {
 485                 // We could parse the regex to find trigrams that have to be there
 486                 // (there are actually known algorithms to deal with disjunctions
 487                 // and such, too), but for now, we just go brute force.
 488                 // Using locate with regexes is pretty niche.
 489         } else {
 490                 for (const Needle &needle : needles) {
 491                         parse_trigrams(needle.str, ignore_case, &trigram_groups);
 492                 }
 493         }
 494
 495         unique_sort(
 496                 &trigram_groups,
 497                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
 498                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 499
 500         // Give them names for debugging.
 501         unsigned td_index = 0;
 502         for (TrigramDisjunction &td : trigram_groups) {
 503                 td.index = td_index++;
 504         }
 505
 506         // Collect which trigrams we need to look up in the hash table.
 507         unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
 508         for (TrigramDisjunction &td : trigram_groups) {
 509                 for (uint32_t trgm : td.trigram_alternatives) {
 510                         trigrams_to_lookup[trgm].push_back(&td);
 511                 }
 512         }
 513         if (trigrams_to_lookup.empty()) {
 514                 // Too short for trigram matching. Apply brute force.
 515                 // (We could have searched through all trigrams that matched
 516                 // the pattern and done a union of them, but that's a lot of
 517                 // work for fairly unclear gain.)
 518                 uint64_t matched = scan_all_docids(needles, fd, corpus, &engine);
 519                 if (only_count) {
 520                         printf("%" PRId64 "\n", matched);
 521                 }
 522                 return;
 523         }
 524
 525         // Look them all up on disk.
 526         for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
 527                 corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }](const Trigram *trgmptr, size_t len) {
 528                         if (trgmptr == nullptr) {
 529                                 dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
 530                                 for (TrigramDisjunction *td : *trigram_groups) {
 531                                         --td->remaining_trigrams_to_read;
 532                                         if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
 533                                                 dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
 534                                                 if (only_count) {
 535                                                         printf("0\n");
 536                                                 }
 537                                                 exit(0);
 538                                         }
 539                                 }
 540                                 return;
 541                         }
 542                         for (TrigramDisjunction *td : *trigram_groups) {
 543                                 --td->remaining_trigrams_to_read;
 544                                 td->max_num_docids += trgmptr->num_docids;
 545                                 td->read_trigrams.emplace_back(*trgmptr, len);
 546                         }
 547                 });
 548         }
 549         engine.finish();
 550         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 551
 552         for (TrigramDisjunction &td : trigram_groups) {
 553                 // Reset for reads.
 554                 td.remaining_trigrams_to_read = td.read_trigrams.size();
 555
 556                 if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
 557                         dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
 558                 }
 559         }
 560
 561         // TODO: For case-insensitive (ie. more than one alternative in each),
 562         // prioritize the ones with fewer seeks?
 563         sort(trigram_groups.begin(), trigram_groups.end(),
 564              [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
 565                      return a.max_num_docids < b.max_num_docids;
 566              });
 567
 568         unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
 569         for (TrigramDisjunction &td : trigram_groups) {
 570                 for (uint32_t trgm : td.trigram_alternatives) {
 571                         uses_trigram[trgm].push_back(&td);
 572                 }
 573         }
 574
 575         unordered_set<uint32_t> trigrams_submitted_read;
 576         vector<uint32_t> cur_candidates, tmp, decoded;
 577         bool done = false;
 578         for (TrigramDisjunction &td : trigram_groups) {
 579                 if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
 580                         dprintf("%s has up to %u entries, ignoring the rest (will "
 581                                 "weed out false positives later)\n",
 582                                 print_td(td).c_str(), td.max_num_docids);
 583                         break;
 584                 }
 585
 586                 for (auto &[trgmptr, len] : td.read_trigrams) {
 587                         if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
 588                                 continue;
 589                         }
 590                         trigrams_submitted_read.insert(trgmptr.trgm);
 591                         // Only stay a certain amount ahead, so that we don't spend I/O
 592                         // on reading the latter, large posting lists. We are unlikely
 593                         // to need them anyway, even if they should come in first.
 594                         if (engine.get_waiting_reads() >= 5) {
 595                                 engine.finish();
 596                                 if (done)
 597                                         break;
 598                         }
 599                         engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
 600                                 if (done)
 601                                         return;
 602
 603                                 uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 604                                 const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 605                                 size_t num = trgmptr.num_docids;
 606                                 decoded.resize(num);
 607                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
 608
 609                                 assert(uses_trigram.count(trgm) != 0);
 610                                 bool was_empty = cur_candidates.empty();
 611                                 if (ignore_case) {
 612                                         dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 613                                 }
 614
 615                                 for (TrigramDisjunction *td : uses_trigram[trgm]) {
 616                                         done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
 617                                         if (done)
 618                                                 break;
 619                                 }
 620                                 if (!ignore_case) {
 621                                         if (was_empty) {
 622                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 623                                         } else if (cur_candidates.empty()) {
 624                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
 625                                         } else {
 626                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
 627                                         }
 628                                 }
 629                         });
 630                 }
 631         }
 632         engine.finish();
 633         if (done) {
 634                 return;
 635         }
 636         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 637                 1e3 * duration<float>(steady_clock::now() - start).count());
 638
 639         uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
 640         dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
 641                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 642
 643         if (only_count) {
 644                 printf("%" PRId64 "\n", matched);
 645         }
 646 }
 647
 648 string unescape_glob_to_plain_string(const string &needle)
 649 {
 650         string unescaped;
 651         for (size_t i = 0; i < needle.size(); i += read_unigram(needle, i).second) {
 652                 uint32_t ch = read_unigram(needle, i).first;
 653                 assert(ch != WILDCARD_UNIGRAM);
 654                 if (ch == PREMATURE_END_UNIGRAM) {
 655                         fprintf(stderr, "Pattern '%s' ended prematurely\n", needle.c_str());
 656                         exit(1);
 657                 }
 658                 unescaped.push_back(ch);
 659         }
 660         return unescaped;
 661 }
 662
 663 regex_t compile_regex(const string &needle)
 664 {
 665         regex_t re;
 666         int flags = REG_NOSUB;
 667         if (ignore_case) {
 668                 flags |= REG_ICASE;
 669         }
 670         if (use_extended_regex) {
 671                 flags |= REG_EXTENDED;
 672         }
 673         int err = regcomp(&re, needle.c_str(), flags);
 674         if (err != 0) {
 675                 char errbuf[256];
 676                 regerror(err, &re, errbuf, sizeof(errbuf));
 677                 fprintf(stderr, "Error when compiling regex '%s': %s\n", needle.c_str(), errbuf);
 678                 exit(1);
 679         }
 680         return re;
 681 }
 682
 683 void usage()
 684 {
 685         printf(
 686                 "Usage: plocate [OPTION]... PATTERN...\n"
 687                 "\n"
 688                 "  -c, --count            print number of matches instead of the matches\n"
 689                 "  -d, --database DBPATH  search for files in DBPATH\n"
 690                 "                         (default is " DEFAULT_DBPATH ")\n"
 691                 "  -i, --ignore-case      search case-insensitively\n"
 692                 "  -l, --limit LIMIT      stop after LIMIT matches\n"
 693                 "  -0, --null             delimit matches by NUL instead of newline\n"
 694                 "  -r, --regexp           interpret patterns as basic regexps (slow)\n"
 695                 "      --regex            interpret patterns as extended regexps (slow)\n"
 696                 "      --help             print this help\n"
 697                 "      --version          print version information\n");
 698 }
 699
 700 void version()
 701 {
 702         printf("plocate %s\n", PLOCATE_VERSION);
 703         printf("Copyright 2020 Steinar H. Gunderson\n");
 704         printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\n");
 705         printf("This is free software: you are free to change and redistribute it.\n");
 706         printf("There is NO WARRANTY, to the extent permitted by law.\n");
 707         exit(0);
 708 }
 709
 710 int main(int argc, char **argv)
 711 {
 712         constexpr int EXTENDED_REGEX = 1000;
 713         static const struct option long_options[] = {
 714                 { "help", no_argument, 0, 'h' },
 715                 { "count", no_argument, 0, 'c' },
 716                 { "database", required_argument, 0, 'd' },
 717                 { "ignore-case", no_argument, 0, 'i' },
 718                 { "limit", required_argument, 0, 'l' },
 719                 { "null", no_argument, 0, '0' },
 720                 { "version", no_argument, 0, 'V' },
 721                 { "regexp", no_argument, 0, 'r' },
 722                 { "regex", no_argument, 0, EXTENDED_REGEX },
 723                 { "debug", no_argument, 0, 'D' },  // Not documented.
 724                 { 0, 0, 0, 0 }
 725         };
 726
 727         setlocale(LC_ALL, "");
 728         for (;;) {
 729                 int option_index = 0;
 730                 int c = getopt_long(argc, argv, "cd:hil:n:0VD", long_options, &option_index);
 731                 if (c == -1) {
 732                         break;
 733                 }
 734                 switch (c) {
 735                 case 'c':
 736                         only_count = true;
 737                         break;
 738                 case 'd':
 739                         dbpath = strdup(optarg);
 740                         break;
 741                 case 'h':
 742                         usage();
 743                         exit(0);
 744                 case 'i':
 745                         ignore_case = true;
 746                         break;
 747                 case 'l':
 748                 case 'n':
 749                         limit_matches = limit_left = atoll(optarg);
 750                         if (limit_matches <= 0) {
 751                                 fprintf(stderr, "Error: limit must be a strictly positive number.\n");
 752                                 exit(1);
 753                         }
 754                         break;
 755                 case '0':
 756                         print_nul = true;
 757                         break;
 758                 case 'r':
 759                         patterns_are_regex = true;
 760                         break;
 761                 case EXTENDED_REGEX:
 762                         patterns_are_regex = true;
 763                         use_extended_regex = true;
 764                         break;
 765                 case 'D':
 766                         use_debug = true;
 767                         break;
 768                 case 'V':
 769                         version();
 770                         break;
 771                 default:
 772                         exit(1);
 773                 }
 774         }
 775
 776         if (use_debug) {
 777                 // Debug information would leak information about which files exist,
 778                 // so drop setgid before we open the file; one would either need to run
 779                 // as root, or use a locally-built file.
 780                 if (setgid(getgid()) != 0) {
 781                         perror("setgid");
 782                         exit(EXIT_FAILURE);
 783                 }
 784         }
 785
 786         vector<Needle> needles;
 787         for (int i = optind; i < argc; ++i) {
 788                 Needle needle;
 789                 needle.str = argv[i];
 790
 791                 // See if there are any wildcard characters, which indicates we should treat it
 792                 // as an (anchored) glob.
 793                 bool any_wildcard = false;
 794                 for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) {
 795                         if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) {
 796                                 any_wildcard = true;
 797                                 break;
 798                         }
 799                 }
 800
 801                 if (patterns_are_regex) {
 802                         needle.type = Needle::REGEX;
 803                         needle.re = compile_regex(needle.str);
 804                 } else if (any_wildcard) {
 805                         needle.type = Needle::GLOB;
 806                 } else if (ignore_case) {
 807                         // strcasestr() doesn't handle locales correctly (even though LSB
 808                         // claims it should), but somehow, fnmatch() does, and it's about
 809                         // the same speed as using a regex.
 810                         needle.type = Needle::GLOB;
 811                         needle.str = "*" + needle.str + "*";
 812                 } else {
 813                         needle.type = Needle::STRSTR;
 814                         needle.str = unescape_glob_to_plain_string(needle.str);
 815                 }
 816                 needles.push_back(move(needle));
 817         }
 818         if (needles.empty()) {
 819                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 820                 exit(0);
 821         }
 822         do_search_file(needles, dbpath);
 823 }