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

   1 #include "access_rx_cache.h"
   2 #include "complete_pread.h"
   3 #include "db.h"
   4 #include "dprintf.h"
   5 #include "io_uring_engine.h"
   6 #include "needle.h"
   7 #include "parse_trigrams.h"
   8 #include "serializer.h"
   9 #include "turbopfor.h"
  10 #include "unique_sort.h"
  11
  12 #include <algorithm>
  13 #include <assert.h>
  14 #include <atomic>
  15 #include <chrono>
  16 #include <condition_variable>
  17 #include <deque>
  18 #include <fcntl.h>
  19 #include <functional>
  20 #include <getopt.h>
  21 #include <inttypes.h>
  22 #include <iterator>
  23 #include <limits>
  24 #include <locale.h>
  25 #include <memory>
  26 #include <mutex>
  27 #include <regex.h>
  28 #include <stdint.h>
  29 #include <stdio.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <string>
  33 #include <string_view>
  34 #include <sys/stat.h>
  35 #include <sys/types.h>
  36 #include <sys/wait.h>
  37 #include <thread>
  38 #include <tuple>
  39 #include <unistd.h>
  40 #include <unordered_map>
  41 #include <unordered_set>
  42 #include <utility>
  43 #include <vector>
  44 #include <zstd.h>
  45
  46 using namespace std;
  47 using namespace std::chrono;
  48
  49 bool ignore_case = false;
  50 bool only_count = false;
  51 bool print_nul = false;
  52 bool use_debug = false;
  53 bool flush_cache = false;
  54 bool patterns_are_regex = false;
  55 bool use_extended_regex = false;
  56 bool match_basename = false;
  57 bool check_existence = false;
  58 bool ignore_visibility = false;
  59 int64_t limit_matches = numeric_limits<int64_t>::max();
  60 int64_t limit_left = numeric_limits<int64_t>::max();
  61 bool stdout_is_tty = false;
  62 bool literal_printing = false;
  63 static bool in_forked_child = false;
  64
  65 steady_clock::time_point start;
  66 ZSTD_DDict *ddict = nullptr;
  67
  68 class Corpus {
  69 public:
  70         Corpus(int fd, const char *filename_for_errors, IOUringEngine *engine);
  71         ~Corpus();
  72         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
  73         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
  74         size_t get_num_filename_blocks() const;
  75         off_t offset_for_block(uint32_t docid) const
  76         {
  77                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
  78         }
  79         const Header &get_hdr() const { return hdr; }
  80
  81 public:
  82         const int fd;
  83         IOUringEngine *const engine;
  84
  85         Header hdr;
  86 };
  87
  88 Corpus::Corpus(int fd, const char *filename_for_errors, IOUringEngine *engine)
  89         : fd(fd), engine(engine)
  90 {
  91         if (flush_cache) {
  92                 off_t len = lseek(fd, 0, SEEK_END);
  93                 if (len == -1) {
  94                         perror("lseek");
  95                         exit(1);
  96                 }
  97                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
  98         }
  99
 100         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 101         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 102                 fprintf(stderr, "%s: database is corrupt or not a plocate database; please rebuild it.\n", filename_for_errors);
 103                 exit(1);
 104         }
 105         if (hdr.version != 0 && hdr.version != 1) {
 106                 fprintf(stderr, "%s: has version %u, expected 0 or 1; please rebuild it.\n", filename_for_errors, hdr.version);
 107                 exit(1);
 108         }
 109         if (hdr.version == 0) {
 110                 // These will be junk data.
 111                 hdr.zstd_dictionary_offset_bytes = 0;
 112                 hdr.zstd_dictionary_length_bytes = 0;
 113         }
 114         if (hdr.max_version < 2) {
 115                 // This too. (We ignore the other max_version 2 fields.)
 116                 hdr.check_visibility = true;
 117         }
 118         if (ignore_visibility) {
 119                 hdr.check_visibility = false;
 120         }
 121 }
 122
 123 Corpus::~Corpus()
 124 {
 125         close(fd);
 126 }
 127
 128 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 129 {
 130         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 131         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) {
 132                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 133                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 134                         if (trgmptr[i].trgm == trgm) {
 135                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 136                                 return;
 137                         }
 138                 }
 139
 140                 // Not found.
 141                 cb(nullptr, 0);
 142         });
 143 }
 144
 145 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 146 {
 147         // Read the file offset from this docid and the next one.
 148         // This is always allowed, since we have a sentinel block at the end.
 149         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 150                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 151                 off_t offset = ptr[0];
 152                 size_t len = ptr[1] - ptr[0];
 153                 engine->submit_read(fd, len, offset, cb);
 154         });
 155 }
 156
 157 size_t Corpus::get_num_filename_blocks() const
 158 {
 159         return hdr.num_docids;
 160 }
 161
 162 template<class T>
 163 void stat_if_needed(const char *filename, bool access_ok, IOUringEngine *engine, T cb)
 164 {
 165         if (!access_ok || !check_existence) {
 166                 // Doesn't have access or doesn't care about existence, so no need to stat.
 167                 cb(access_ok);
 168         } else if (engine == nullptr || !engine->get_supports_stat()) {
 169                 // Do a synchronous stat.
 170                 struct stat buf;
 171                 bool ok = lstat(filename, &buf) == 0;
 172                 cb(ok);
 173         } else {
 174                 engine->submit_stat(filename, cb);
 175         }
 176 }
 177
 178 void scan_file_block(const vector<Needle> &needles, string_view compressed,
 179                      IOUringEngine *engine, AccessRXCache *access_rx_cache, uint64_t seq, ResultReceiver *serializer,
 180                      atomic<uint64_t> *matched)
 181 {
 182         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 183         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 184                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 185                 exit(1);
 186         }
 187
 188         string block;
 189         block.resize(uncompressed_len + 1);
 190
 191         static thread_local ZSTD_DCtx *ctx = ZSTD_createDCtx();  // Reused across calls.
 192         size_t err;
 193
 194         if (ddict != nullptr) {
 195                 err = ZSTD_decompress_usingDDict(ctx, &block[0], block.size(), compressed.data(),
 196                                                  compressed.size(), ddict);
 197         } else {
 198                 err = ZSTD_decompressDCtx(ctx, &block[0], block.size(), compressed.data(),
 199                                           compressed.size());
 200         }
 201         if (ZSTD_isError(err)) {
 202                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 203                 exit(1);
 204         }
 205         block[block.size() - 1] = '\0';
 206
 207         auto test_candidate = [&](const char *filename, uint64_t local_seq, uint64_t next_seq) {
 208                 access_rx_cache->check_access(filename, /*allow_async=*/true, [matched, engine, serializer, local_seq, next_seq, filename{ strdup(filename) }](bool ok) {
 209                         stat_if_needed(filename, ok, engine, [matched, serializer, local_seq, next_seq, filename](bool ok) {
 210                                 if (ok) {
 211                                         ++*matched;
 212                                         serializer->print(local_seq, next_seq - local_seq, filename);
 213                                 } else {
 214                                         serializer->print(local_seq, next_seq - local_seq, "");
 215                                 }
 216                                 free(filename);
 217                         });
 218                 });
 219         };
 220
 221         // We need to know the next sequence number before inserting into Serializer,
 222         // so always buffer one candidate.
 223         const char *pending_candidate = nullptr;
 224
 225         uint64_t local_seq = seq << 32;
 226         for (const char *filename = block.data();
 227              filename != block.data() + block.size();
 228              filename += strlen(filename) + 1) {
 229                 const char *haystack = filename;
 230                 if (match_basename) {
 231                         haystack = strrchr(filename, '/');
 232                         if (haystack == nullptr) {
 233                                 haystack = filename;
 234                         } else {
 235                                 ++haystack;
 236                         }
 237                 }
 238
 239                 bool found = true;
 240                 for (const Needle &needle : needles) {
 241                         if (!matches(needle, haystack)) {
 242                                 found = false;
 243                                 break;
 244                         }
 245                 }
 246                 if (found) {
 247                         if (pending_candidate != nullptr) {
 248                                 test_candidate(pending_candidate, local_seq, local_seq + 1);
 249                                 ++local_seq;
 250                         }
 251                         pending_candidate = filename;
 252                 }
 253         }
 254         if (pending_candidate == nullptr) {
 255                 serializer->print(seq << 32, 1ULL << 32, "");
 256         } else {
 257                 test_candidate(pending_candidate, local_seq, (seq + 1) << 32);
 258         }
 259 }
 260
 261 size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 262 {
 263         Serializer docids_in_order;
 264         AccessRXCache access_rx_cache(engine, corpus.get_hdr().check_visibility);
 265         atomic<uint64_t> matched{ 0 };
 266         for (size_t i = 0; i < docids.size(); ++i) {
 267                 uint32_t docid = docids[i];
 268                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, engine, &docids_in_order](string_view compressed) {
 269                         scan_file_block(needles, compressed, engine, &access_rx_cache, i, &docids_in_order, &matched);
 270                 });
 271         }
 272         engine->finish();
 273         return matched;
 274 }
 275
 276 struct WorkerThread {
 277         thread t;
 278
 279         // We use a result queue instead of synchronizing Serializer,
 280         // since a lock on it becomes a huge choke point if there are
 281         // lots of threads.
 282         mutex result_mu;
 283         struct Result {
 284                 uint64_t seq;
 285                 uint64_t skip;
 286                 string msg;
 287         };
 288         vector<Result> results;
 289 };
 290
 291 class WorkerThreadReceiver : public ResultReceiver {
 292 public:
 293         WorkerThreadReceiver(WorkerThread *wt)
 294                 : wt(wt) {}
 295
 296         void print(uint64_t seq, uint64_t skip, const string msg) override
 297         {
 298                 lock_guard<mutex> lock(wt->result_mu);
 299                 if (msg.empty() && !wt->results.empty() && wt->results.back().seq + wt->results.back().skip == seq) {
 300                         wt->results.back().skip += skip;
 301                 } else {
 302                         wt->results.emplace_back(WorkerThread::Result{ seq, skip, move(msg) });
 303                 }
 304         }
 305
 306 private:
 307         WorkerThread *wt;
 308 };
 309
 310 void deliver_results(WorkerThread *wt, Serializer *serializer)
 311 {
 312         vector<WorkerThread::Result> results;
 313         {
 314                 lock_guard<mutex> lock(wt->result_mu);
 315                 results = move(wt->results);
 316         }
 317         for (const WorkerThread::Result &result : results) {
 318                 serializer->print(result.seq, result.skip, move(result.msg));
 319         }
 320 }
 321
 322 // We do this sequentially, as it's faster than scattering
 323 // a lot of I/O through io_uring and hoping the kernel will
 324 // coalesce it plus readahead for us. Since we assume that
 325 // we will primarily be CPU-bound, we'll be firing up one
 326 // worker thread for each spare core (the last one will
 327 // only be doing I/O). access() is still synchronous.
 328 uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus)
 329 {
 330         {
 331                 const Header &hdr = corpus.get_hdr();
 332                 if (hdr.zstd_dictionary_length_bytes > 0) {
 333                         string dictionary;
 334                         dictionary.resize(hdr.zstd_dictionary_length_bytes);
 335                         complete_pread(fd, &dictionary[0], hdr.zstd_dictionary_length_bytes, hdr.zstd_dictionary_offset_bytes);
 336                         ddict = ZSTD_createDDict(dictionary.data(), dictionary.size());
 337                 }
 338         }
 339
 340         AccessRXCache access_rx_cache(nullptr, corpus.get_hdr().check_visibility);
 341         Serializer serializer;
 342         uint32_t num_blocks = corpus.get_num_filename_blocks();
 343         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 344         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 345         atomic<uint64_t> matched{ 0 };
 346
 347         mutex mu;
 348         condition_variable queue_added, queue_removed;
 349         deque<tuple<int, int, string>> work_queue;  // Under mu.
 350         bool done = false;  // Under mu.
 351
 352         unsigned num_threads = max<int>(sysconf(_SC_NPROCESSORS_ONLN) - 1, 1);
 353         dprintf("Using %u worker threads for linear scan.\n", num_threads);
 354         unique_ptr<WorkerThread[]> threads(new WorkerThread[num_threads]);
 355         for (unsigned i = 0; i < num_threads; ++i) {
 356                 threads[i].t = thread([&threads, &mu, &queue_added, &queue_removed, &work_queue, &done, &offsets, &needles, &access_rx_cache, &matched, i] {
 357                         // regcomp() takes a lock on the regex, so each thread will need its own.
 358                         const vector<Needle> *use_needles = &needles;
 359                         vector<Needle> recompiled_needles;
 360                         if (i != 0 && patterns_are_regex) {
 361                                 recompiled_needles = needles;
 362                                 for (Needle &needle : recompiled_needles) {
 363                                         needle.re = compile_regex(needle.str);
 364                                 }
 365                                 use_needles = &recompiled_needles;
 366                         }
 367
 368                         WorkerThreadReceiver receiver(&threads[i]);
 369                         for (;;) {
 370                                 uint32_t io_docid, last_docid;
 371                                 string compressed;
 372
 373                                 {
 374                                         unique_lock lock(mu);
 375                                         queue_added.wait(lock, [&work_queue, &done] { return !work_queue.empty() || done; });
 376                                         if (done && work_queue.empty()) {
 377                                                 return;
 378                                         }
 379                                         tie(io_docid, last_docid, compressed) = move(work_queue.front());
 380                                         work_queue.pop_front();
 381                                         queue_removed.notify_all();
 382                                 }
 383
 384                                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 385                                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 386                                         size_t len = offsets[docid + 1] - offsets[docid];
 387                                         // IOUringEngine isn't thread-safe, so we do any needed stat()s synchronously (nullptr engine).
 388                                         scan_file_block(*use_needles, { &compressed[relative_offset], len }, /*engine=*/nullptr, &access_rx_cache, docid, &receiver, &matched);
 389                                 }
 390                         }
 391                 });
 392         }
 393
 394         string compressed;
 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                 {
 404                         unique_lock lock(mu);
 405                         queue_removed.wait(lock, [&work_queue] { return work_queue.size() < 256; });  // Allow ~2MB of data queued up.
 406                         work_queue.emplace_back(io_docid, last_docid, move(compressed));
 407                         queue_added.notify_one();  // Avoid the thundering herd.
 408                 }
 409
 410                 // Pick up some results, so that we are sure that we won't just overload.
 411                 // (Seemingly, going through all of these causes slowness with many threads,
 412                 // but taking only one is OK.)
 413                 unsigned i = io_docid / 32;
 414                 deliver_results(&threads[i % num_threads], &serializer);
 415         }
 416         {
 417                 lock_guard<mutex> lock(mu);
 418                 done = true;
 419                 queue_added.notify_all();
 420         }
 421         for (unsigned i = 0; i < num_threads; ++i) {
 422                 threads[i].t.join();
 423                 deliver_results(&threads[i], &serializer);
 424         }
 425         return matched;
 426 }
 427
 428 // Takes the given posting list, unions it into the parts of the trigram disjunction
 429 // already read; if the list is complete, intersects with “cur_candidates”.
 430 //
 431 // Returns true if the search should be aborted (we are done).
 432 bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
 433 {
 434         if (td->docids.empty()) {
 435                 td->docids = move(decoded);
 436         } else {
 437                 tmp->clear();
 438                 set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
 439                 swap(*tmp, td->docids);
 440         }
 441         if (--td->remaining_trigrams_to_read > 0) {
 442                 // Need to wait for more.
 443                 if (ignore_case) {
 444                         dprintf("  ... %u reads left in OR group %u (%zu docids in list)\n",
 445                                 td->remaining_trigrams_to_read, td->index, td->docids.size());
 446                 }
 447                 return false;
 448         }
 449         if (cur_candidates->empty()) {
 450                 if (ignore_case) {
 451                         dprintf("  ... all reads done for OR group %u (%zu docids)\n",
 452                                 td->index, td->docids.size());
 453                 }
 454                 *cur_candidates = move(td->docids);
 455         } else {
 456                 tmp->clear();
 457                 set_intersection(cur_candidates->begin(), cur_candidates->end(),
 458                                  td->docids.begin(), td->docids.end(),
 459                                  back_inserter(*tmp));
 460                 swap(*cur_candidates, *tmp);
 461                 if (ignore_case) {
 462                         if (cur_candidates->empty()) {
 463                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n",
 464                                         td->index, td->docids.size());
 465                                 return true;
 466                         } else {
 467                                 dprintf("  ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n",
 468                                         td->index, td->docids.size(), cur_candidates->size());
 469                         }
 470                 }
 471         }
 472         return false;
 473 }
 474
 475 uint64_t do_search_file(const vector<Needle> &needles, const std::string &filename)
 476 {
 477         int fd = open(filename.c_str(), O_RDONLY);
 478         if (fd == -1) {
 479                 perror(filename.c_str());
 480                 exit(1);
 481         }
 482
 483         // Drop privileges.
 484         if (setgid(getgid()) != 0) {
 485                 perror("setgid");
 486                 exit(EXIT_FAILURE);
 487         }
 488
 489         start = steady_clock::now();
 490         if (access("/", R_OK | X_OK)) {
 491                 // We can't find anything, no need to bother...
 492                 return 0;
 493         }
 494
 495         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 496         Corpus corpus(fd, filename.c_str(), &engine);
 497         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 498
 499         vector<TrigramDisjunction> trigram_groups;
 500         if (patterns_are_regex) {
 501                 // We could parse the regex to find trigrams that have to be there
 502                 // (there are actually known algorithms to deal with disjunctions
 503                 // and such, too), but for now, we just go brute force.
 504                 // Using locate with regexes is pretty niche.
 505         } else {
 506                 for (const Needle &needle : needles) {
 507                         parse_trigrams(needle.str, ignore_case, &trigram_groups);
 508                 }
 509         }
 510
 511         unique_sort(
 512                 &trigram_groups,
 513                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
 514                 [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 515
 516         // Give them names for debugging.
 517         unsigned td_index = 0;
 518         for (TrigramDisjunction &td : trigram_groups) {
 519                 td.index = td_index++;
 520         }
 521
 522         // Collect which trigrams we need to look up in the hash table.
 523         unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
 524         for (TrigramDisjunction &td : trigram_groups) {
 525                 for (uint32_t trgm : td.trigram_alternatives) {
 526                         trigrams_to_lookup[trgm].push_back(&td);
 527                 }
 528         }
 529         if (trigrams_to_lookup.empty()) {
 530                 // Too short for trigram matching. Apply brute force.
 531                 // (We could have searched through all trigrams that matched
 532                 // the pattern and done a union of them, but that's a lot of
 533                 // work for fairly unclear gain.)
 534                 uint64_t matched = scan_all_docids(needles, fd, corpus);
 535                 dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
 536                         1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 537                 return matched;
 538         }
 539
 540         // Sneak in fetching the dictionary, if present. It's not necessarily clear
 541         // exactly where it would be cheapest to get it, but it needs to be present
 542         // before we can decode any of the posting lists. Most likely, it's
 543         // in the same filesystem block as the header anyway, so it should be
 544         // present in the cache.
 545         {
 546                 const Header &hdr = corpus.get_hdr();
 547                 if (hdr.zstd_dictionary_length_bytes > 0) {
 548                         engine.submit_read(fd, hdr.zstd_dictionary_length_bytes, hdr.zstd_dictionary_offset_bytes, [](string_view s) {
 549                                 ddict = ZSTD_createDDict(s.data(), s.size());
 550                                 dprintf("Dictionary initialized after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 551                         });
 552                 }
 553         }
 554
 555         // Look them all up on disk.
 556         bool should_early_exit = false;
 557         for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
 558                 corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }, &should_early_exit](const Trigram *trgmptr, size_t len) {
 559                         if (trgmptr == nullptr) {
 560                                 dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
 561                                 for (TrigramDisjunction *td : *trigram_groups) {
 562                                         --td->remaining_trigrams_to_read;
 563
 564                                         // If we now know this trigram group doesn't match anything at all,
 565                                         // we can do early exit; however, if we're in a forked child,
 566                                         // that would confuse the parent process (since we don't write
 567                                         // our count to the pipe), so we wait until we're back in to the
 568                                         // regular (non-async) context. This is a fairly rare case anyway,
 569                                         // and the gains from dropping the remaining trigram reads are limited.
 570                                         if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
 571                                                 if (in_forked_child) {
 572                                                         should_early_exit = true;
 573                                                 } else {
 574                                                         dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
 575                                                         if (only_count) {
 576                                                                 printf("0\n");
 577                                                         }
 578                                                         exit(1);
 579                                                 }
 580                                         }
 581                                 }
 582                                 return;
 583                         }
 584                         for (TrigramDisjunction *td : *trigram_groups) {
 585                                 --td->remaining_trigrams_to_read;
 586                                 td->max_num_docids += trgmptr->num_docids;
 587                                 td->read_trigrams.emplace_back(*trgmptr, len);
 588                         }
 589                 });
 590         }
 591         engine.finish();
 592         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 593
 594         if (should_early_exit) {
 595                 return 0;
 596         }
 597
 598         for (TrigramDisjunction &td : trigram_groups) {
 599                 // Reset for reads.
 600                 td.remaining_trigrams_to_read = td.read_trigrams.size();
 601
 602                 if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
 603                         dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
 604                 }
 605         }
 606
 607         // TODO: For case-insensitive (ie. more than one alternative in each),
 608         // prioritize the ones with fewer seeks?
 609         sort(trigram_groups.begin(), trigram_groups.end(),
 610              [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
 611                      return a.max_num_docids < b.max_num_docids;
 612              });
 613
 614         unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
 615         for (TrigramDisjunction &td : trigram_groups) {
 616                 for (uint32_t trgm : td.trigram_alternatives) {
 617                         uses_trigram[trgm].push_back(&td);
 618                 }
 619         }
 620
 621         unordered_set<uint32_t> trigrams_submitted_read;
 622         vector<uint32_t> cur_candidates, tmp, decoded;
 623         bool done = false;
 624         for (TrigramDisjunction &td : trigram_groups) {
 625                 if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
 626                         dprintf("%s has up to %u entries, ignoring the rest (will "
 627                                 "weed out false positives later)\n",
 628                                 print_td(td).c_str(), td.max_num_docids);
 629                         break;
 630                 }
 631
 632                 for (auto &[trgmptr, len] : td.read_trigrams) {
 633                         if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
 634                                 continue;
 635                         }
 636                         trigrams_submitted_read.insert(trgmptr.trgm);
 637                         // Only stay a certain amount ahead, so that we don't spend I/O
 638                         // on reading the latter, large posting lists. We are unlikely
 639                         // to need them anyway, even if they should come in first.
 640                         if (engine.get_waiting_reads() >= 5) {
 641                                 engine.finish();
 642                                 if (done)
 643                                         break;
 644                         }
 645                         engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
 646                                 if (done)
 647                                         return;
 648
 649                                 uint32_t trgm = trgmptr.trgm;
 650                                 const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 651                                 size_t num = trgmptr.num_docids;
 652                                 decoded.resize(num);
 653                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
 654
 655                                 assert(uses_trigram.count(trgm) != 0);
 656                                 bool was_empty = cur_candidates.empty();
 657                                 if (ignore_case) {
 658                                         dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 659                                 }
 660
 661                                 for (TrigramDisjunction *td : uses_trigram[trgm]) {
 662                                         done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
 663                                         if (done)
 664                                                 break;
 665                                 }
 666                                 if (!ignore_case) {
 667                                         if (was_empty) {
 668                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
 669                                         } else if (cur_candidates.empty()) {
 670                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
 671                                         } else {
 672                                                 dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
 673                                         }
 674                                 }
 675                         });
 676                 }
 677         }
 678         engine.finish();
 679         if (done) {
 680                 return 0;
 681         }
 682         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 683                 1e3 * duration<float>(steady_clock::now() - start).count());
 684
 685         uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
 686         dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
 687                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 688         return matched;
 689 }
 690
 691 // Run do_search_file() in a child process.
 692 //
 693 // The reason for this is that we're not robust against malicious input, so we need
 694 // to drop privileges after opening the file. (Otherwise, we could fall prey to an attack
 695 // where a user does locate -d badfile.db:/var/lib/plocate/plocate.db, badfile.db contains
 696 // a buffer overflow that takes over the process, and then uses the elevated privileges
 697 // to print out otherwise inaccessible paths.) We solve this by forking and treating the
 698 // child process as untrusted after it has dropped its privileges (which it does before
 699 // reading any data from the file); it returns a single 64-bit number over a pipe,
 700 // and that's it. The parent keeps its privileges, and can then fork out new children
 701 // without fear of being taken over. (The child keeps stdout for outputting results.)
 702 //
 703 // The count is returned over the pipe, because it's needed both for --limit and --count.
 704 uint64_t do_search_file_in_child(const vector<Needle> &needles, const std::string &filename)
 705 {
 706         int pipefd[2];
 707         if (pipe(pipefd) == -1) {
 708                 perror("pipe");
 709                 exit(EXIT_FAILURE);
 710         }
 711
 712         pid_t child_pid = fork();
 713         switch (child_pid) {
 714         case 0: {
 715                 // Child.
 716                 close(pipefd[0]);
 717                 in_forked_child = true;
 718                 uint64_t matched = do_search_file(needles, filename);
 719                 int ret;
 720                 do {
 721                         ret = write(pipefd[1], &matched, sizeof(matched));
 722                 } while (ret == -1 && errno == EINTR);
 723                 if (ret != sizeof(matched)) {
 724                         perror("write");
 725                         _exit(EXIT_FAILURE);
 726                 }
 727                 fflush(stdout);
 728                 _exit(EXIT_SUCCESS);
 729         }
 730         case -1:
 731                 // Error.
 732                 perror("fork");
 733                 exit(EXIT_FAILURE);
 734         default:
 735                 // Parent.
 736                 close(pipefd[1]);
 737                 break;
 738         }
 739
 740         // Wait for the child to finish.
 741         int wstatus;
 742         pid_t err;
 743         do {
 744                 err = waitpid(child_pid, &wstatus, 0);
 745         } while (err == -1 && errno == EINTR);
 746         if (err == -1) {
 747                 perror("waitpid");
 748                 exit(EXIT_FAILURE);
 749         }
 750         if (WIFEXITED(wstatus)) {
 751                 if (WEXITSTATUS(wstatus) != 0) {
 752                         // The child has probably already printed out its error, so just propagate the exit status.
 753                         exit(WEXITSTATUS(wstatus));
 754                 }
 755                 // Success!
 756         } else if (!WIFEXITED(wstatus)) {
 757                 fprintf(stderr, "FATAL: Child died unexpectedly while processing %s\n", filename.c_str());
 758                 exit(1);
 759         }
 760
 761         // Now get the number of matches from the child.
 762         uint64_t matched;
 763         int ret;
 764         do {
 765                 ret = read(pipefd[0], &matched, sizeof(matched));
 766         } while (ret == -1 && errno == EINTR);
 767         if (ret == -1) {
 768                 perror("read");
 769                 exit(EXIT_FAILURE);
 770         } else if (ret != sizeof(matched)) {
 771                 fprintf(stderr, "FATAL: Short read through pipe (got %d bytes)\n", ret);
 772                 exit(EXIT_FAILURE);
 773         }
 774         close(pipefd[0]);
 775         return matched;
 776 }
 777
 778 // Parses a colon-separated list of strings and appends them onto the given vector.
 779 // Backslash escapes whatever comes after it.
 780 void parse_dbpaths(const char *ptr, vector<string> *output)
 781 {
 782         string str;
 783         while (*ptr != '\0') {
 784                 if (*ptr == '\\') {
 785                         if (ptr[1] == '\0') {
 786                                 fprintf(stderr, "ERROR: Escape character at the end of string\n");
 787                                 exit(EXIT_FAILURE);
 788                         }
 789                         // Escape.
 790                         str.push_back(ptr[1]);
 791                         ptr += 2;
 792                         continue;
 793                 }
 794                 if (*ptr == ':') {
 795                         // Separator.
 796                         output->push_back(move(str));
 797                         ++ptr;
 798                         continue;
 799                 }
 800                 str.push_back(*ptr++);
 801         }
 802         output->push_back(move(str));
 803 }
 804
 805 void usage()
 806 {
 807         printf(
 808                 "Usage: plocate [OPTION]... PATTERN...\n"
 809                 "\n"
 810                 "  -b, --basename         search only the file name portion of path names\n"
 811                 "  -c, --count            print number of matches instead of the matches\n"
 812                 "  -d, --database DBPATH  search for files in DBPATH\n"
 813                 "                         (default is " DBFILE ")\n"
 814                 "  -i, --ignore-case      search case-insensitively\n"
 815                 "  -l, --limit LIMIT      stop after LIMIT matches\n"
 816                 "  -0, --null             delimit matches by NUL instead of newline\n"
 817                 "  -N, --literal          do not quote filenames, even if printing to a tty\n"
 818                 "  -r, --regexp           interpret patterns as basic regexps (slow)\n"
 819                 "      --regex            interpret patterns as extended regexps (slow)\n"
 820                 "  -w, --wholename        search the entire path name (default; see -b)\n"
 821                 "      --help             print this help\n"
 822                 "      --version          print version information\n");
 823 }
 824
 825 void version()
 826 {
 827         printf("%s %s\n", PACKAGE_NAME, PACKAGE_VERSION);
 828         printf("Copyright 2020 Steinar H. Gunderson\n");
 829         printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\n");
 830         printf("This is free software: you are free to change and redistribute it.\n");
 831         printf("There is NO WARRANTY, to the extent permitted by law.\n");
 832         exit(0);
 833 }
 834
 835 int main(int argc, char **argv)
 836 {
 837         vector<string> dbpaths;
 838
 839         constexpr int EXTENDED_REGEX = 1000;
 840         constexpr int FLUSH_CACHE = 1001;
 841         constexpr int IGNORE_VISIBILITY = 1002;
 842         static const struct option long_options[] = {
 843                 { "help", no_argument, 0, 'h' },
 844                 { "count", no_argument, 0, 'c' },
 845                 { "all", no_argument, 0, 'A' },
 846                 { "basename", no_argument, 0, 'b' },
 847                 { "database", required_argument, 0, 'd' },
 848                 { "existing", no_argument, 0, 'e' },
 849                 { "ignore-case", no_argument, 0, 'i' },
 850                 { "limit", required_argument, 0, 'l' },
 851                 { "literal", no_argument, 0, 'N' },
 852                 { "null", no_argument, 0, '0' },
 853                 { "version", no_argument, 0, 'V' },
 854                 { "regexp", no_argument, 0, 'r' },
 855                 { "regex", no_argument, 0, EXTENDED_REGEX },
 856                 { "wholename", no_argument, 0, 'w' },
 857                 { "debug", no_argument, 0, 'D' },  // Not documented.
 858                 // Enable to test cold-cache behavior (except for access()). Not documented.
 859                 { "flush-cache", no_argument, 0, FLUSH_CACHE },
 860                 // Mostly useful to dump out the entire database, even if the given directories
 861                 // are gone. Disables sgid due to security. Not documented.
 862                 { "ignore-visibility", no_argument, 0, IGNORE_VISIBILITY },
 863                 { 0, 0, 0, 0 }
 864         };
 865
 866         setlocale(LC_ALL, "");
 867         for (;;) {
 868                 int option_index = 0;
 869                 int c = getopt_long(argc, argv, "Abcd:ehil:n:N0rwVD", long_options, &option_index);
 870                 if (c == -1) {
 871                         break;
 872                 }
 873                 switch (c) {
 874                 case 'A':
 875                         // Ignored.
 876                         break;
 877                 case 'b':
 878                         match_basename = true;
 879                         break;
 880                 case 'c':
 881                         only_count = true;
 882                         break;
 883                 case 'd':
 884                         parse_dbpaths(optarg, &dbpaths);
 885                         break;
 886                 case 'e':
 887                         check_existence = true;
 888                         break;
 889                 case 'h':
 890                         usage();
 891                         exit(0);
 892                 case 'i':
 893                         ignore_case = true;
 894                         break;
 895                 case 'l':
 896                 case 'n':
 897                         limit_matches = limit_left = atoll(optarg);
 898                         if (limit_matches <= 0) {
 899                                 fprintf(stderr, "Error: limit must be a strictly positive number.\n");
 900                                 exit(1);
 901                         }
 902                         break;
 903                 case 'N':
 904                         literal_printing = true;
 905                         break;
 906                 case '0':
 907                         print_nul = true;
 908                         break;
 909                 case 'r':
 910                         patterns_are_regex = true;
 911                         break;
 912                 case EXTENDED_REGEX:
 913                         patterns_are_regex = true;
 914                         use_extended_regex = true;
 915                         break;
 916                 case 'w':
 917                         match_basename = false;  // No-op unless -b is given first.
 918                         break;
 919                 case 'D':
 920                         use_debug = true;
 921                         break;
 922                 case FLUSH_CACHE:
 923                         flush_cache = true;
 924                         break;
 925                 case 'V':
 926                         version();
 927                         break;
 928                 case IGNORE_VISIBILITY:
 929                         ignore_visibility = true;
 930                         break;
 931                 default:
 932                         exit(1);
 933                 }
 934         }
 935
 936         if (use_debug || flush_cache || ignore_visibility) {
 937                 // Debug information would leak information about which files exist,
 938                 // so drop setgid before we open the file; one would either need to run
 939                 // as root, or use a locally-built file. Doing the same thing for
 940                 // flush_cache is mostly paranoia, in an attempt to prevent random users
 941                 // from making plocate slow for everyone else.
 942                 // --ignore-visibility is obvious; if we allowed to keep sgid with
 943                 // that flag on, it would subvert the entire security model.
 944                 if (setgid(getgid()) != 0) {
 945                         perror("setgid");
 946                         exit(EXIT_FAILURE);
 947                 }
 948         }
 949
 950         if (!print_nul) {
 951                 stdout_is_tty = isatty(1);
 952         }
 953
 954         vector<Needle> needles;
 955         for (int i = optind; i < argc; ++i) {
 956                 Needle needle;
 957                 needle.str = argv[i];
 958
 959                 // See if there are any wildcard characters, which indicates we should treat it
 960                 // as an (anchored) glob.
 961                 bool any_wildcard = false;
 962                 for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) {
 963                         if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) {
 964                                 any_wildcard = true;
 965                                 break;
 966                         }
 967                 }
 968
 969                 if (patterns_are_regex) {
 970                         needle.type = Needle::REGEX;
 971                         needle.re = compile_regex(needle.str);
 972                 } else if (any_wildcard) {
 973                         needle.type = Needle::GLOB;
 974                 } else if (ignore_case) {
 975                         // strcasestr() doesn't handle locales correctly (even though LSB
 976                         // claims it should), but somehow, fnmatch() does, and it's about
 977                         // the same speed as using a regex.
 978                         needle.type = Needle::GLOB;
 979                         needle.str = "*" + needle.str + "*";
 980                 } else {
 981                         needle.type = Needle::STRSTR;
 982                         needle.str = unescape_glob_to_plain_string(needle.str);
 983                 }
 984                 needles.push_back(move(needle));
 985         }
 986         if (needles.empty()) {
 987                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 988                 exit(1);
 989         }
 990
 991         if (dbpaths.empty()) {
 992                 // No -d given, so use our default. Note that this happens
 993                 // even if LOCATE_PATH exists, to match mlocate behavior.
 994                 dbpaths.push_back(DBFILE);
 995         }
 996
 997         const char *locate_path = getenv("LOCATE_PATH");
 998         if (locate_path != nullptr) {
 999                 parse_dbpaths(locate_path, &dbpaths);
1000         }
1001
1002         uint64_t matched = 0;
1003         for (size_t i = 0; i < dbpaths.size(); ++i) {
1004                 uint64_t this_matched;
1005                 if (i != dbpaths.size() - 1) {
1006                         this_matched = do_search_file_in_child(needles, dbpaths[i]);
1007                 } else {
1008                         this_matched = do_search_file(needles, dbpaths[i]);
1009                 }
1010                 matched += this_matched;
1011                 limit_left -= this_matched;
1012         }
1013         if (only_count) {
1014                 printf("%" PRId64 "\n", matched);
1015         }
1016
1017         return matched == 0;
1018 }