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

   1 #include <fcntl.h>
   2 #include <getopt.h>
   3 #include <stdio.h>
   4 #include <string.h>
   5 #include <unistd.h>
   6 #include <zstd.h>
   7 #include <getopt.h>
   8 #include <stdlib.h>
   9 #include <algorithm>
  10 #include <chrono>
  11 #include <functional>
  12 #include <memory>
  13 #include <string>
  14 #include <unordered_map>
  15 #include <vector>
  16 #include <stdint.h>
  17 #include <iosfwd>
  18 #include <iterator>
  19 #include <limits>
  20 #include <queue>
  21 #include <string_view>
  22 #include <utility>
  23
  24 #include "db.h"
  25 #include "io_uring_engine.h"
  26
  27 using namespace std;
  28 using namespace std::chrono;
  29
  30 #define dprintf(...)
  31 //#define dprintf(...) fprintf(stderr, __VA_ARGS__);
  32
  33 #include "turbopfor.h"
  34
  35 const char *dbpath = "/var/lib/mlocate/plocate.db";
  36 bool only_count = false;
  37 bool print_nul = false;
  38 int64_t limit_matches = numeric_limits<int64_t>::max();
  39
  40 class Serializer {
  41 public:
  42         bool ready_to_print(int seq) { return next_seq == seq; }
  43         void print_delayed(int seq, const vector<string> msg);
  44         void release_current();
  45
  46 private:
  47         int next_seq = 0;
  48         struct Element {
  49                 int seq;
  50                 vector<string> msg;
  51
  52                 bool operator<(const Element &other) const
  53                 {
  54                         return seq > other.seq;
  55                 }
  56         };
  57         priority_queue<Element> pending;
  58 };
  59
  60 void Serializer::print_delayed(int seq, const vector<string> msg)
  61 {
  62         pending.push(Element{ seq, move(msg) });
  63 }
  64
  65 void Serializer::release_current()
  66 {
  67         ++next_seq;
  68
  69         // See if any delayed prints can now be dealt with.
  70         while (!pending.empty() && pending.top().seq == next_seq) {
  71                 if (limit_matches-- <= 0)
  72                         return;
  73                 for (const string &msg : pending.top().msg) {
  74                         if (print_nul) {
  75                                 printf("%s%c", msg.c_str(), 0);
  76                         } else {
  77                                 printf("%s\n", msg.c_str());
  78                         }
  79                 }
  80                 pending.pop();
  81                 ++next_seq;
  82         }
  83 }
  84
  85 static inline uint32_t read_unigram(const string &s, size_t idx)
  86 {
  87         if (idx < s.size()) {
  88                 return (unsigned char)s[idx];
  89         } else {
  90                 return 0;
  91         }
  92 }
  93
  94 static inline uint32_t read_trigram(const string &s, size_t start)
  95 {
  96         return read_unigram(s, start) | (read_unigram(s, start + 1) << 8) |
  97                 (read_unigram(s, start + 2) << 16);
  98 }
  99
 100 bool has_access(const char *filename,
 101                 unordered_map<string, bool> *access_rx_cache)
 102 {
 103         const char *end = strchr(filename + 1, '/');
 104         while (end != nullptr) {
 105                 string parent_path(filename, end);
 106                 auto it = access_rx_cache->find(parent_path);
 107                 bool ok;
 108                 if (it == access_rx_cache->end()) {
 109                         ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
 110                         access_rx_cache->emplace(move(parent_path), ok);
 111                 } else {
 112                         ok = it->second;
 113                 }
 114                 if (!ok) {
 115                         return false;
 116                 }
 117                 end = strchr(end + 1, '/');
 118         }
 119
 120         return true;
 121 }
 122
 123 class Corpus {
 124 public:
 125         Corpus(int fd, IOUringEngine *engine);
 126         ~Corpus();
 127         void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
 128         void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
 129         size_t get_num_filename_blocks() const;
 130         off_t offset_for_block(uint32_t docid) const
 131         {
 132                 return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
 133         }
 134
 135 public:
 136         const int fd;
 137         IOUringEngine *const engine;
 138
 139         Header hdr;
 140 };
 141
 142 Corpus::Corpus(int fd, IOUringEngine *engine)
 143         : fd(fd), engine(engine)
 144 {
 145         // Enable to test cold-cache behavior (except for access()).
 146         if (false) {
 147                 off_t len = lseek(fd, 0, SEEK_END);
 148                 if (len == -1) {
 149                         perror("lseek");
 150                         exit(1);
 151                 }
 152                 posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
 153         }
 154
 155         complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
 156         if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
 157                 fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
 158                 exit(1);
 159         }
 160         if (hdr.version != 0) {
 161                 fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version);
 162                 exit(1);
 163         }
 164 }
 165
 166 Corpus::~Corpus()
 167 {
 168         close(fd);
 169 }
 170
 171 void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb)
 172 {
 173         uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
 174         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) {
 175                 const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
 176                 for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
 177                         if (trgmptr[i].trgm == trgm) {
 178                                 cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
 179                                 return;
 180                         }
 181                 }
 182
 183                 // Not found.
 184                 cb(nullptr, 0);
 185         });
 186 }
 187
 188 void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
 189 {
 190         // Read the file offset from this docid and the next one.
 191         // This is always allowed, since we have a sentinel block at the end.
 192         engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
 193                 const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
 194                 off_t offset = ptr[0];
 195                 size_t len = ptr[1] - ptr[0];
 196                 engine->submit_read(fd, len, offset, cb);
 197         });
 198 }
 199
 200 size_t Corpus::get_num_filename_blocks() const
 201 {
 202         return hdr.num_docids;
 203 }
 204
 205 uint64_t scan_file_block(const vector<string> &needles, string_view compressed,
 206                          unordered_map<string, bool> *access_rx_cache, int seq,
 207                          Serializer *serializer)
 208 {
 209         uint64_t matched = 0;
 210
 211         unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size());
 212         if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) {
 213                 fprintf(stderr, "ZSTD_getFrameContentSize() failed\n");
 214                 exit(1);
 215         }
 216
 217         string block;
 218         block.resize(uncompressed_len + 1);
 219
 220         size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(),
 221                                      compressed.size());
 222         if (ZSTD_isError(err)) {
 223                 fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err));
 224                 exit(1);
 225         }
 226         block[block.size() - 1] = '\0';
 227
 228         bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq));
 229         vector<string> delayed;
 230
 231         for (const char *filename = block.data();
 232              filename != block.data() + block.size();
 233              filename += strlen(filename) + 1) {
 234                 bool found = true;
 235                 for (const string &needle : needles) {
 236                         if (strstr(filename, needle.c_str()) == nullptr) {
 237                                 found = false;
 238                                 break;
 239                         }
 240                 }
 241                 if (found && has_access(filename, access_rx_cache)) {
 242                         if (limit_matches-- <= 0)
 243                                 break;
 244                         ++matched;
 245                         if (only_count)
 246                                 continue;
 247                         if (immediate_print) {
 248                                 if (print_nul) {
 249                                         printf("%s%c", filename, 0);
 250                                 } else {
 251                                         printf("%s\n", filename);
 252                                 }
 253                         } else {
 254                                 delayed.push_back(filename);
 255                         }
 256                 }
 257         }
 258         if (serializer != nullptr && !only_count) {
 259                 if (immediate_print) {
 260                         serializer->release_current();
 261                 } else {
 262                         serializer->print_delayed(seq, move(delayed));
 263                 }
 264         }
 265         return matched;
 266 }
 267
 268 size_t scan_docids(const vector<string> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
 269 {
 270         Serializer docids_in_order;
 271         unordered_map<string, bool> access_rx_cache;
 272         uint64_t matched = 0;
 273         for (size_t i = 0; i < docids.size(); ++i) {
 274                 uint32_t docid = docids[i];
 275                 corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string_view compressed) {
 276                         matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order);
 277                 });
 278         }
 279         engine->finish();
 280         return matched;
 281 }
 282
 283 // We do this sequentially, as it's faster than scattering
 284 // a lot of I/O through io_uring and hoping the kernel will
 285 // coalesce it plus readahead for us.
 286 uint64_t scan_all_docids(const vector<string> &needles, int fd, const Corpus &corpus, IOUringEngine *engine)
 287 {
 288         unordered_map<string, bool> access_rx_cache;
 289         uint32_t num_blocks = corpus.get_num_filename_blocks();
 290         unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
 291         complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
 292         string compressed;
 293         uint64_t matched = 0;
 294         for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
 295                 uint32_t last_docid = std::min(io_docid + 32, num_blocks);
 296                 size_t io_len = offsets[last_docid] - offsets[io_docid];
 297                 if (compressed.size() < io_len) {
 298                         compressed.resize(io_len);
 299                 }
 300                 complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
 301
 302                 for (uint32_t docid = io_docid; docid < last_docid; ++docid) {
 303                         size_t relative_offset = offsets[docid] - offsets[io_docid];
 304                         size_t len = offsets[docid + 1] - offsets[docid];
 305                         matched += scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr);
 306                         if (limit_matches <= 0)
 307                                 return matched;
 308                 }
 309         }
 310         return matched;
 311 }
 312
 313 void do_search_file(const vector<string> &needles, const char *filename)
 314 {
 315         int fd = open(filename, O_RDONLY);
 316         if (fd == -1) {
 317                 perror(filename);
 318                 exit(1);
 319         }
 320
 321         // Drop privileges.
 322         if (setgid(getgid()) != 0) {
 323                 perror("setgid");
 324                 exit(EXIT_FAILURE);
 325         }
 326
 327         steady_clock::time_point start __attribute__((unused)) = steady_clock::now();
 328         if (access("/", R_OK | X_OK)) {
 329                 // We can't find anything, no need to bother...
 330                 return;
 331         }
 332
 333         IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
 334         Corpus corpus(fd, &engine);
 335         dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 336
 337         vector<pair<Trigram, size_t>> trigrams;
 338         uint64_t shortest_so_far = numeric_limits<uint32_t>::max();
 339         for (const string &needle : needles) {
 340                 if (needle.size() < 3)
 341                         continue;
 342                 for (size_t i = 0; i < needle.size() - 2; ++i) {
 343                         uint32_t trgm = read_trigram(needle, i);
 344                         corpus.find_trigram(trgm, [trgm, &trigrams, &shortest_so_far](const Trigram *trgmptr, size_t len) {
 345                                 if (trgmptr == nullptr) {
 346                                         dprintf("trigram '%c%c%c' isn't found, we abort the search\n",
 347                                                 trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff);
 348                                         if (only_count) {
 349                                                 printf("0\n");
 350                                         }
 351                                         exit(0);
 352                                 }
 353                                 if (trgmptr->num_docids > shortest_so_far * 100) {
 354                                         dprintf("not loading trigram '%c%c%c' with %u docids, it would be ignored later anyway\n",
 355                                                 trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff,
 356                                                 trgmptr->num_docids);
 357                                 } else {
 358                                         trigrams.emplace_back(*trgmptr, len);
 359                                         shortest_so_far = std::min<uint64_t>(shortest_so_far, trgmptr->num_docids);
 360                                 }
 361                         });
 362                 }
 363         }
 364         engine.finish();
 365         dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
 366
 367         if (trigrams.empty()) {
 368                 // Too short for trigram matching. Apply brute force.
 369                 // (We could have searched through all trigrams that matched
 370                 // the pattern and done a union of them, but that's a lot of
 371                 // work for fairly unclear gain.)
 372                 uint64_t matched = scan_all_docids(needles, fd, corpus, &engine);
 373                 printf("%zu\n", matched);
 374                 return;
 375         }
 376         sort(trigrams.begin(), trigrams.end());
 377         {
 378                 auto last = unique(trigrams.begin(), trigrams.end());
 379                 trigrams.erase(last, trigrams.end());
 380         }
 381         sort(trigrams.begin(), trigrams.end(),
 382              [&](const pair<Trigram, size_t> &a, const pair<Trigram, size_t> &b) {
 383                      return a.first.num_docids < b.first.num_docids;
 384              });
 385
 386         vector<uint32_t> in1, in2, out;
 387         bool done = false;
 388         for (auto [trgmptr, len] : trigrams) {
 389                 if (!in1.empty() && trgmptr.num_docids > in1.size() * 100) {
 390                         uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 391                         dprintf("trigram '%c%c%c' (%zu bytes) has %u entries, ignoring the rest (will "
 392                                 "weed out false positives later)\n",
 393                                 trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff,
 394                                 len, trgmptr.num_docids);
 395                         break;
 396                 }
 397
 398                 // Only stay a certain amount ahead, so that we don't spend I/O
 399                 // on reading the latter, large posting lists. We are unlikely
 400                 // to need them anyway, even if they should come in first.
 401                 if (engine.get_waiting_reads() >= 5) {
 402                         engine.finish();
 403                         if (done)
 404                                 break;
 405                 }
 406                 engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &in1, &in2, &out](string_view s) {
 407                         if (done)
 408                                 return;
 409                         uint32_t trgm __attribute__((unused)) = trgmptr.trgm;
 410                         size_t num = trgmptr.num_docids;
 411                         const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
 412                         if (in1.empty()) {
 413                                 in1.resize(num + 128);
 414                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &in1[0]);
 415                                 in1.resize(num);
 416                                 dprintf("trigram '%c%c%c' (%zu bytes) decoded to %zu entries\n", trgm & 0xff,
 417                                         (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, len, num);
 418                         } else {
 419                                 if (in2.size() < num + 128) {
 420                                         in2.resize(num + 128);
 421                                 }
 422                                 decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &in2[0]);
 423
 424                                 out.clear();
 425                                 set_intersection(in1.begin(), in1.end(), in2.begin(), in2.begin() + num,
 426                                                  back_inserter(out));
 427                                 swap(in1, out);
 428                                 dprintf("trigram '%c%c%c' (%zu bytes) decoded to %zu entries, %zu left\n",
 429                                         trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff,
 430                                         len, num, in1.size());
 431                                 if (in1.empty()) {
 432                                         dprintf("no matches (intersection list is empty)\n");
 433                                         done = true;
 434                                 }
 435                         }
 436                 });
 437         }
 438         engine.finish();
 439         if (done) {
 440                 return;
 441         }
 442         dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
 443                 1e3 * duration<float>(steady_clock::now() - start).count());
 444
 445         uint64_t matched = scan_docids(needles, in1, corpus, &engine);
 446         dprintf("Done in %.1f ms, found %zu matches.\n",
 447                 1e3 * duration<float>(steady_clock::now() - start).count(), matched);
 448
 449         if (only_count) {
 450                 printf("%zu\n", matched);
 451         }
 452 }
 453
 454 void usage()
 455 {
 456         // The help text comes from mlocate.
 457         printf("Usage: plocate [OPTION]... PATTERN...\n");
 458         printf("\n");
 459         printf("  -c, --count            only print number of found entries\n");
 460         printf("  -d, --database DBPATH  use DBPATH instead of default database (which is\n");
 461         printf("                         %s)\n", dbpath);
 462         printf("  -h, --help             print this help\n");
 463         printf("  -l, --limit, -n LIMIT  limit output (or counting) to LIMIT entries\n");
 464         printf("  -0, --null             separate entries with NUL on output\n");
 465 }
 466
 467 int main(int argc, char **argv)
 468 {
 469         static const struct option long_options[] = {
 470                 { "help", no_argument, 0, 'h' },
 471                 { "count", no_argument, 0, 'c' },
 472                 { "database", required_argument, 0, 'd' },
 473                 { "limit", required_argument, 0, 'l' },
 474                 { nullptr, required_argument, 0, 'n' },
 475                 { "null", no_argument, 0, '0' },
 476                 { 0, 0, 0, 0 }
 477         };
 478
 479         for (;;) {
 480                 int option_index = 0;
 481                 int c = getopt_long(argc, argv, "cd:hl:n:0", long_options, &option_index);
 482                 if (c == -1) {
 483                         break;
 484                 }
 485                 switch (c) {
 486                 case 'c':
 487                         only_count = true;
 488                         break;
 489                 case 'd':
 490                         dbpath = strdup(optarg);
 491                         break;
 492                 case 'h':
 493                         usage();
 494                         exit(0);
 495                 case 'l':
 496                 case 'n':
 497                         limit_matches = atoll(optarg);
 498                         break;
 499                 case '0':
 500                         print_nul = true;
 501                         break;
 502                 default:
 503                         exit(1);
 504                 }
 505         }
 506
 507         vector<string> needles;
 508         for (int i = optind; i < argc; ++i) {
 509                 needles.push_back(argv[i]);
 510         }
 511         if (needles.empty()) {
 512                 fprintf(stderr, "plocate: no pattern to search for specified\n");
 513                 exit(0);
 514         }
 515         do_search_file(needles, dbpath);
 516 }