X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=plocate.cpp;h=dfb12d42573d4c32dac3d8b0b9aaf4311535f2ac;hb=436bb5d60934c534b2ddab599236493fbff4da23;hp=f721b91264f9f2f6d642d0b628d2905a2c7aee21;hpb=efd7545c8ee2177aa13cf3ec8423d0e725c6a16d;p=plocate diff --git a/plocate.cpp b/plocate.cpp index f721b91..dfb12d4 100644 --- a/plocate.cpp +++ b/plocate.cpp @@ -1,17 +1,22 @@ #include "db.h" +#include "dprintf.h" #include "io_uring_engine.h" #include "parse_trigrams.h" +#include "turbopfor.h" #include "unique_sort.h" #include #include #include #include +#include #include #include +#include #include #include #include +#include #include #include #include @@ -31,28 +36,44 @@ using namespace std; using namespace std::chrono; -#define dprintf(...) -//#define dprintf(...) fprintf(stderr, __VA_ARGS__); - -#include "turbopfor.h" +#define DEFAULT_DBPATH "/var/lib/mlocate/plocate.db" -const char *dbpath = "/var/lib/mlocate/plocate.db"; +const char *dbpath = DEFAULT_DBPATH; bool ignore_case = false; bool only_count = false; bool print_nul = false; +bool use_debug = false; +bool patterns_are_regex = false; +bool use_extended_regex = false; int64_t limit_matches = numeric_limits::max(); +int64_t limit_left = numeric_limits::max(); + +steady_clock::time_point start; +ZSTD_DDict *ddict = nullptr; + +void apply_limit() +{ + if (--limit_left > 0) { + return; + } + dprintf("Done in %.1f ms, found %" PRId64 " matches.\n", + 1e3 * duration(steady_clock::now() - start).count(), limit_matches); + if (only_count) { + printf("%" PRId64 "\n", limit_matches); + } + exit(0); +} class Serializer { public: - bool ready_to_print(int seq) { return next_seq == seq; } - void print_delayed(int seq, const vector msg); - void release_current(); + ~Serializer() { assert(limit_left <= 0 || pending.empty()); } + void print(uint64_t seq, uint64_t skip, const string msg); private: - int next_seq = 0; + uint64_t next_seq = 0; struct Element { - int seq; - vector msg; + uint64_t seq, skip; + string msg; bool operator<(const Element &other) const { @@ -62,34 +83,49 @@ private: priority_queue pending; }; -void Serializer::print_delayed(int seq, const vector msg) +void Serializer::print(uint64_t seq, uint64_t skip, const string msg) { - pending.push(Element{ seq, move(msg) }); -} + if (only_count) { + if (!msg.empty()) { + apply_limit(); + } + return; + } -void Serializer::release_current() -{ - ++next_seq; + if (next_seq != seq) { + pending.push(Element{ seq, skip, move(msg) }); + return; + } + + if (!msg.empty()) { + if (print_nul) { + printf("%s%c", msg.c_str(), 0); + } else { + printf("%s\n", msg.c_str()); + } + apply_limit(); + } + next_seq += skip; // See if any delayed prints can now be dealt with. while (!pending.empty() && pending.top().seq == next_seq) { - if (limit_matches-- <= 0) - return; - for (const string &msg : pending.top().msg) { + if (!pending.top().msg.empty()) { if (print_nul) { - printf("%s%c", msg.c_str(), 0); + printf("%s%c", pending.top().msg.c_str(), 0); } else { - printf("%s\n", msg.c_str()); + printf("%s\n", pending.top().msg.c_str()); } + apply_limit(); } + next_seq += pending.top().skip; pending.pop(); - ++next_seq; } } struct Needle { enum { STRSTR, - REGEX } type; + REGEX, + GLOB } type; string str; // Filled in no matter what. regex_t re; // For REGEX. }; @@ -98,33 +134,92 @@ bool matches(const Needle &needle, const char *haystack) { if (needle.type == Needle::STRSTR) { return strstr(haystack, needle.str.c_str()) != nullptr; + } else if (needle.type == Needle::GLOB) { + int flags = ignore_case ? FNM_CASEFOLD : 0; + return fnmatch(needle.str.c_str(), haystack, flags) == 0; } else { assert(needle.type == Needle::REGEX); return regexec(&needle.re, haystack, /*nmatch=*/0, /*pmatch=*/nullptr, /*flags=*/0) == 0; } } -bool has_access(const char *filename, - unordered_map *access_rx_cache) +class AccessRXCache { +public: + AccessRXCache(IOUringEngine *engine) + : engine(engine) {} + void check_access(const char *filename, bool allow_async, function cb); + +private: + unordered_map cache; + struct PendingStat { + string filename; + function cb; + }; + map> pending_stats; + IOUringEngine *engine; +}; + +void AccessRXCache::check_access(const char *filename, bool allow_async, function cb) { - const char *end = strchr(filename + 1, '/'); - while (end != nullptr) { - string parent_path(filename, end); - auto it = access_rx_cache->find(parent_path); - bool ok; - if (it == access_rx_cache->end()) { - ok = access(parent_path.c_str(), R_OK | X_OK) == 0; - access_rx_cache->emplace(move(parent_path), ok); - } else { - ok = it->second; + if (engine == nullptr || !engine->get_supports_stat()) { + allow_async = false; + } + + for (const char *end = strchr(filename + 1, '/'); end != nullptr; end = strchr(end + 1, '/')) { + string parent_path(filename, end - filename); // string_view from C++20. + auto cache_it = cache.find(parent_path); + if (cache_it != cache.end()) { + // Found in the cache. + if (!cache_it->second) { + cb(false); + return; + } + continue; + } + + if (!allow_async) { + bool ok = access(parent_path.c_str(), R_OK | X_OK) == 0; + cache.emplace(parent_path, ok); + if (!ok) { + cb(false); + return; + } + continue; } - if (!ok) { - return false; + + // We want to call access(), but it could block on I/O. io_uring doesn't support + // access(), but we can do a dummy asynchonous statx() to populate the kernel's cache, + // which nearly always makes the next access() instantaneous. + + // See if there's already a pending stat that matches this, + // or is a subdirectory. + auto it = pending_stats.lower_bound(parent_path); + if (it != pending_stats.end() && it->first.size() >= parent_path.size() && + it->first.compare(0, parent_path.size(), parent_path) == 0) { + it->second.emplace_back(PendingStat{ filename, move(cb) }); + } else { + it = pending_stats.emplace(filename, vector{}).first; + engine->submit_stat(filename, [this, it, filename{ strdup(filename) }, cb{ move(cb) }] { + // The stat returned, so now do the actual access() calls. + // All of them should be in cache, so don't fire off new statx() + // calls during that check. + check_access(filename, /*allow_async=*/false, move(cb)); + free(filename); + + // Call all others that waited for the same stat() to finish. + // They may fire off new stat() calls if needed. + vector pending = move(it->second); + pending_stats.erase(it); + for (PendingStat &ps : pending) { + check_access(ps.filename.c_str(), /*allow_async=*/true, move(ps.cb)); + } + }); } - end = strchr(end + 1, '/'); + return; // The rest will happen in async context. } - return true; + // Passed all checks. + cb(true); } class Corpus { @@ -138,6 +233,7 @@ public: { return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t); } + const Header &get_hdr() const { return hdr; } public: const int fd; @@ -164,10 +260,15 @@ Corpus::Corpus(int fd, IOUringEngine *engine) fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n"); exit(1); } - if (hdr.version != 0) { - fprintf(stderr, "plocate.db has version %u, expected 0; please rebuild it.\n", hdr.version); + if (hdr.version != 0 && hdr.version != 1) { + fprintf(stderr, "plocate.db has version %u, expected 0 or 1; please rebuild it.\n", hdr.version); exit(1); } + if (hdr.version == 0) { + // These will be junk data. + hdr.zstd_dictionary_offset_bytes = 0; + hdr.zstd_dictionary_length_bytes = 0; + } } Corpus::~Corpus() @@ -209,12 +310,10 @@ size_t Corpus::get_num_filename_blocks() const return hdr.num_docids; } -uint64_t scan_file_block(const vector &needles, string_view compressed, - unordered_map *access_rx_cache, int seq, - Serializer *serializer) +void scan_file_block(const vector &needles, string_view compressed, + AccessRXCache *access_rx_cache, uint64_t seq, Serializer *serializer, + uint64_t *matched) { - uint64_t matched = 0; - unsigned long long uncompressed_len = ZSTD_getFrameContentSize(compressed.data(), compressed.size()); if (uncompressed_len == ZSTD_CONTENTSIZE_UNKNOWN || uncompressed_len == ZSTD_CONTENTSIZE_ERROR) { fprintf(stderr, "ZSTD_getFrameContentSize() failed\n"); @@ -224,17 +323,39 @@ uint64_t scan_file_block(const vector &needles, string_view compressed, string block; block.resize(uncompressed_len + 1); - size_t err = ZSTD_decompress(&block[0], block.size(), compressed.data(), - compressed.size()); + static ZSTD_DCtx *ctx = ZSTD_createDCtx(); // Reused across calls. + size_t err; + + if (ddict != nullptr) { + err = ZSTD_decompress_usingDDict(ctx, &block[0], block.size(), compressed.data(), + compressed.size(), ddict); + } else { + err = ZSTD_decompressDCtx(ctx, &block[0], block.size(), compressed.data(), + compressed.size()); + } if (ZSTD_isError(err)) { fprintf(stderr, "ZSTD_decompress(): %s\n", ZSTD_getErrorName(err)); exit(1); } block[block.size() - 1] = '\0'; - bool immediate_print = (serializer == nullptr || serializer->ready_to_print(seq)); - vector delayed; + auto test_candidate = [&](const char *filename, uint64_t local_seq, uint64_t next_seq) { + access_rx_cache->check_access(filename, /*allow_async=*/true, [matched, serializer, local_seq, next_seq, filename{ strdup(filename) }](bool ok) { + if (ok) { + ++*matched; + serializer->print(local_seq, next_seq - local_seq, filename); + } else { + serializer->print(local_seq, next_seq - local_seq, ""); + } + free(filename); + }); + }; + + // We need to know the next sequence number before inserting into Serializer, + // so always buffer one candidate. + const char *pending_candidate = nullptr; + uint64_t local_seq = seq << 32; for (const char *filename = block.data(); filename != block.data() + block.size(); filename += strlen(filename) + 1) { @@ -245,42 +366,30 @@ uint64_t scan_file_block(const vector &needles, string_view compressed, break; } } - if (found && has_access(filename, access_rx_cache)) { - if (limit_matches-- <= 0) - break; - ++matched; - if (only_count) - continue; - if (immediate_print) { - if (print_nul) { - printf("%s%c", filename, 0); - } else { - printf("%s\n", filename); - } - } else { - delayed.push_back(filename); + if (found) { + if (pending_candidate != nullptr) { + test_candidate(pending_candidate, local_seq, local_seq + 1); + ++local_seq; } + pending_candidate = filename; } } - if (serializer != nullptr && !only_count) { - if (immediate_print) { - serializer->release_current(); - } else { - serializer->print_delayed(seq, move(delayed)); - } + if (pending_candidate == nullptr) { + serializer->print(seq << 32, 1ULL << 32, ""); + } else { + test_candidate(pending_candidate, local_seq, (seq + 1) << 32); } - return matched; } size_t scan_docids(const vector &needles, const vector &docids, const Corpus &corpus, IOUringEngine *engine) { Serializer docids_in_order; - unordered_map access_rx_cache; + AccessRXCache access_rx_cache(engine); uint64_t matched = 0; for (size_t i = 0; i < docids.size(); ++i) { uint32_t docid = docids[i]; corpus.get_compressed_filename_block(docid, [i, &matched, &needles, &access_rx_cache, &docids_in_order](string_view compressed) { - matched += scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order); + scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order, &matched); }); } engine->finish(); @@ -290,9 +399,20 @@ size_t scan_docids(const vector &needles, const vector &docids // We do this sequentially, as it's faster than scattering // a lot of I/O through io_uring and hoping the kernel will // coalesce it plus readahead for us. -uint64_t scan_all_docids(const vector &needles, int fd, const Corpus &corpus, IOUringEngine *engine) +uint64_t scan_all_docids(const vector &needles, int fd, const Corpus &corpus) { - unordered_map access_rx_cache; + { + const Header &hdr = corpus.get_hdr(); + if (hdr.zstd_dictionary_length_bytes > 0) { + string dictionary; + dictionary.resize(hdr.zstd_dictionary_length_bytes); + complete_pread(fd, &dictionary[0], hdr.zstd_dictionary_length_bytes, hdr.zstd_dictionary_offset_bytes); + ddict = ZSTD_createDDict(dictionary.data(), dictionary.size()); + } + } + + AccessRXCache access_rx_cache(nullptr); + Serializer serializer; // Mostly dummy; handles only the limit. uint32_t num_blocks = corpus.get_num_filename_blocks(); unique_ptr offsets(new uint64_t[num_blocks + 1]); complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0)); @@ -309,9 +429,7 @@ uint64_t scan_all_docids(const vector &needles, int fd, const Corpus &co for (uint32_t docid = io_docid; docid < last_docid; ++docid) { size_t relative_offset = offsets[docid] - offsets[io_docid]; size_t len = offsets[docid + 1] - offsets[docid]; - matched += scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, 0, nullptr); - if (limit_matches <= 0) - return matched; + scan_file_block(needles, { &compressed[relative_offset], len }, &access_rx_cache, docid, &serializer, &matched); } } return matched; @@ -334,14 +452,14 @@ bool new_posting_list_read(TrigramDisjunction *td, vector decoded, vec // Need to wait for more. if (ignore_case) { dprintf(" ... %u reads left in OR group %u (%zu docids in list)\n", - td->remaining_trigrams_to_read, td->index, td->docids.size()); + td->remaining_trigrams_to_read, td->index, td->docids.size()); } return false; } if (cur_candidates->empty()) { if (ignore_case) { dprintf(" ... all reads done for OR group %u (%zu docids)\n", - td->index, td->docids.size()); + td->index, td->docids.size()); } *cur_candidates = move(td->docids); } else { @@ -353,11 +471,11 @@ bool new_posting_list_read(TrigramDisjunction *td, vector decoded, vec if (ignore_case) { if (cur_candidates->empty()) { dprintf(" ... all reads done for OR group %u (%zu docids), intersected (none left, search is done)\n", - td->index, td->docids.size()); + td->index, td->docids.size()); return true; } else { dprintf(" ... all reads done for OR group %u (%zu docids), intersected (%zu left)\n", - td->index, td->docids.size(), cur_candidates->size()); + td->index, td->docids.size(), cur_candidates->size()); } } } @@ -378,7 +496,7 @@ void do_search_file(const vector &needles, const char *filename) exit(EXIT_FAILURE); } - steady_clock::time_point start __attribute__((unused)) = steady_clock::now(); + start = steady_clock::now(); if (access("/", R_OK | X_OK)) { // We can't find anything, no need to bother... return; @@ -389,11 +507,17 @@ void do_search_file(const vector &needles, const char *filename) dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration(steady_clock::now() - start).count()); vector trigram_groups; - for (const Needle &needle : needles) { - if (needle.str.size() < 3) - continue; - parse_trigrams(needle.str, ignore_case, &trigram_groups); + if (patterns_are_regex) { + // We could parse the regex to find trigrams that have to be there + // (there are actually known algorithms to deal with disjunctions + // and such, too), but for now, we just go brute force. + // Using locate with regexes is pretty niche. + } else { + for (const Needle &needle : needles) { + parse_trigrams(needle.str, ignore_case, &trigram_groups); + } } + unique_sort( &trigram_groups, [](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; }, @@ -417,13 +541,28 @@ void do_search_file(const vector &needles, const char *filename) // (We could have searched through all trigrams that matched // the pattern and done a union of them, but that's a lot of // work for fairly unclear gain.) - uint64_t matched = scan_all_docids(needles, fd, corpus, &engine); + uint64_t matched = scan_all_docids(needles, fd, corpus); if (only_count) { - printf("%zu\n", matched); + printf("%" PRId64 "\n", matched); } return; } + // Sneak in fetching the dictionary, if present. It's not necessarily clear + // exactly where it would be cheapest to get it, but it needs to be present + // before we can decode any of the posting lists. Most likely, it's + // in the same filesystem block as the header anyway, so it should be + // present in the cache. + { + const Header &hdr = corpus.get_hdr(); + if (hdr.zstd_dictionary_length_bytes > 0) { + engine.submit_read(fd, hdr.zstd_dictionary_length_bytes, hdr.zstd_dictionary_offset_bytes, [](string_view s) { + ddict = ZSTD_createDDict(s.data(), s.size()); + dprintf("Dictionary initialized after %.1f ms.\n", 1e3 * duration(steady_clock::now() - start).count()); + }); + } + } + // Look them all up on disk. for (auto &[trgm, trigram_groups] : trigrams_to_lookup) { corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }](const Trigram *trgmptr, size_t len) { @@ -539,49 +678,44 @@ void do_search_file(const vector &needles, const char *filename) 1e3 * duration(steady_clock::now() - start).count()); uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine); - dprintf("Done in %.1f ms, found %zu matches.\n", + dprintf("Done in %.1f ms, found %" PRId64 " matches.\n", 1e3 * duration(steady_clock::now() - start).count(), matched); if (only_count) { - printf("%zu\n", matched); + printf("%" PRId64 "\n", matched); } } -regex_t needle_to_regex(const string &needle) +string unescape_glob_to_plain_string(const string &needle) { - string escaped_needle; - for (char ch : needle) { - switch (ch) { - // Directly from what regex(7) considers an “atom”. - case '^': - case '.': - case '[': - case '$': - case '(': - case ')': - case '|': - case '*': - case '+': - case '?': - case '{': - case '\\': - escaped_needle.push_back('\\'); - // Fall through. - default: - escaped_needle.push_back(ch); + string unescaped; + for (size_t i = 0; i < needle.size(); i += read_unigram(needle, i).second) { + uint32_t ch = read_unigram(needle, i).first; + assert(ch != WILDCARD_UNIGRAM); + if (ch == PREMATURE_END_UNIGRAM) { + fprintf(stderr, "Pattern '%s' ended prematurely\n", needle.c_str()); + exit(1); } + unescaped.push_back(ch); } + return unescaped; +} + +regex_t compile_regex(const string &needle) +{ regex_t re; - int err; + int flags = REG_NOSUB; if (ignore_case) { - err = regcomp(&re, escaped_needle.c_str(), REG_NOSUB | REG_ICASE); - } else { - err = regcomp(&re, escaped_needle.c_str(), REG_NOSUB); + flags |= REG_ICASE; + } + if (use_extended_regex) { + flags |= REG_EXTENDED; } + int err = regcomp(&re, needle.c_str(), flags); if (err != 0) { char errbuf[256]; regerror(err, &re, errbuf, sizeof(errbuf)); - fprintf(stderr, "Error when compiling regex for '%s': %s\n", needle.c_str(), errbuf); + fprintf(stderr, "Error when compiling regex '%s': %s\n", needle.c_str(), errbuf); exit(1); } return re; @@ -589,35 +723,52 @@ regex_t needle_to_regex(const string &needle) void usage() { - // The help text comes from mlocate. - printf("Usage: plocate [OPTION]... PATTERN...\n"); - printf("\n"); - printf(" -c, --count only print number of found entries\n"); - printf(" -d, --database DBPATH use DBPATH instead of default database (which is\n"); - printf(" %s)\n", dbpath); - printf(" -h, --help print this help\n"); - printf(" -i, --ignore-case ignore case distinctions when matching patterns\n"); - printf(" -l, --limit, -n LIMIT limit output (or counting) to LIMIT entries\n"); - printf(" -0, --null separate entries with NUL on output\n"); + printf( + "Usage: plocate [OPTION]... PATTERN...\n" + "\n" + " -c, --count print number of matches instead of the matches\n" + " -d, --database DBPATH search for files in DBPATH\n" + " (default is " DEFAULT_DBPATH ")\n" + " -i, --ignore-case search case-insensitively\n" + " -l, --limit LIMIT stop after LIMIT matches\n" + " -0, --null delimit matches by NUL instead of newline\n" + " -r, --regexp interpret patterns as basic regexps (slow)\n" + " --regex interpret patterns as extended regexps (slow)\n" + " --help print this help\n" + " --version print version information\n"); +} + +void version() +{ + printf("plocate %s\n", PLOCATE_VERSION); + printf("Copyright 2020 Steinar H. Gunderson\n"); + printf("License GPLv2+: GNU GPL version 2 or later .\n"); + printf("This is free software: you are free to change and redistribute it.\n"); + printf("There is NO WARRANTY, to the extent permitted by law.\n"); + exit(0); } int main(int argc, char **argv) { + constexpr int EXTENDED_REGEX = 1000; static const struct option long_options[] = { { "help", no_argument, 0, 'h' }, { "count", no_argument, 0, 'c' }, { "database", required_argument, 0, 'd' }, { "ignore-case", no_argument, 0, 'i' }, { "limit", required_argument, 0, 'l' }, - { nullptr, required_argument, 0, 'n' }, { "null", no_argument, 0, '0' }, + { "version", no_argument, 0, 'V' }, + { "regexp", no_argument, 0, 'r' }, + { "regex", no_argument, 0, EXTENDED_REGEX }, + { "debug", no_argument, 0, 'D' }, // Not documented. { 0, 0, 0, 0 } }; setlocale(LC_ALL, ""); for (;;) { int option_index = 0; - int c = getopt_long(argc, argv, "cd:hil:n:0", long_options, &option_index); + int c = getopt_long(argc, argv, "cd:hil:n:0VD", long_options, &option_index); if (c == -1) { break; } @@ -636,30 +787,72 @@ int main(int argc, char **argv) break; case 'l': case 'n': - limit_matches = atoll(optarg); + limit_matches = limit_left = atoll(optarg); + if (limit_matches <= 0) { + fprintf(stderr, "Error: limit must be a strictly positive number.\n"); + exit(1); + } break; case '0': print_nul = true; break; + case 'r': + patterns_are_regex = true; + break; + case EXTENDED_REGEX: + patterns_are_regex = true; + use_extended_regex = true; + break; + case 'D': + use_debug = true; + break; + case 'V': + version(); + break; default: exit(1); } } + if (use_debug) { + // Debug information would leak information about which files exist, + // so drop setgid before we open the file; one would either need to run + // as root, or use a locally-built file. + if (setgid(getgid()) != 0) { + perror("setgid"); + exit(EXIT_FAILURE); + } + } + vector needles; for (int i = optind; i < argc; ++i) { Needle needle; needle.str = argv[i]; - if (ignore_case) { - // strcasestr() doesn't handle locales correctly (even though LSB - // claims it should), but somehow, the glibc regex engine does. - // It's much slower than strstr() for non-case-sensitive searches, though - // (even though it really ought to be faster, since it can precompile), - // so only use it for that. + + // See if there are any wildcard characters, which indicates we should treat it + // as an (anchored) glob. + bool any_wildcard = false; + for (size_t i = 0; i < needle.str.size(); i += read_unigram(needle.str, i).second) { + if (read_unigram(needle.str, i).first == WILDCARD_UNIGRAM) { + any_wildcard = true; + break; + } + } + + if (patterns_are_regex) { needle.type = Needle::REGEX; - needle.re = needle_to_regex(argv[i]); + needle.re = compile_regex(needle.str); + } else if (any_wildcard) { + needle.type = Needle::GLOB; + } else if (ignore_case) { + // strcasestr() doesn't handle locales correctly (even though LSB + // claims it should), but somehow, fnmatch() does, and it's about + // the same speed as using a regex. + needle.type = Needle::GLOB; + needle.str = "*" + needle.str + "*"; } else { needle.type = Needle::STRSTR; + needle.str = unescape_glob_to_plain_string(needle.str); } needles.push_back(move(needle)); }