X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=plocate.cpp;h=c2bb17ca1452e5b63468c7e45ae5a305db1dc059;hb=97c57726bc81c979386c623ad7586c2677fdd865;hp=926bbf1977831562217829cc07e7aaddf56ca4f8;hpb=36b3de205b731a061e84a54c5318039b3a3d31cb;p=plocate diff --git a/plocate.cpp b/plocate.cpp index 926bbf1..c2bb17c 100644 --- a/plocate.cpp +++ b/plocate.cpp @@ -1,27 +1,28 @@ +#include "access_rx_cache.h" #include "db.h" #include "dprintf.h" #include "io_uring_engine.h" +#include "needle.h" #include "parse_trigrams.h" +#include "serializer.h" #include "turbopfor.h" #include "unique_sort.h" #include -#include #include +#include #include #include +#include #include -#include #include #include #include -#include #include #include -#include +#include #include #include -#include #include #include #include @@ -30,6 +31,7 @@ #include #include #include +#include #include #include #include @@ -47,195 +49,16 @@ bool ignore_case = false; bool only_count = false; bool print_nul = false; bool use_debug = false; +bool flush_cache = false; bool patterns_are_regex = false; bool use_extended_regex = false; +bool match_basename = false; int64_t limit_matches = numeric_limits::max(); int64_t limit_left = numeric_limits::max(); steady_clock::time_point start; ZSTD_DDict *ddict = nullptr; -regex_t compile_regex(const string &needle); - -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 ResultReceiver { -public: - virtual ~ResultReceiver() = default; - virtual void print(uint64_t seq, uint64_t skip, const string msg) = 0; -}; - -class Serializer : public ResultReceiver { -public: - ~Serializer() { assert(limit_left <= 0 || pending.empty()); } - void print(uint64_t seq, uint64_t skip, const string msg) override; - -private: - uint64_t next_seq = 0; - struct Element { - uint64_t seq, skip; - string msg; - - bool operator<(const Element &other) const - { - return seq > other.seq; - } - }; - priority_queue pending; -}; - -void Serializer::print(uint64_t seq, uint64_t skip, const string msg) -{ - if (only_count) { - if (!msg.empty()) { - apply_limit(); - } - return; - } - - 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 (!pending.top().msg.empty()) { - if (print_nul) { - printf("%s%c", pending.top().msg.c_str(), 0); - } else { - printf("%s\n", pending.top().msg.c_str()); - } - apply_limit(); - } - next_seq += pending.top().skip; - pending.pop(); - } -} - -struct Needle { - enum { STRSTR, - REGEX, - GLOB } type; - string str; // Filled in no matter what. - regex_t re; // For REGEX. -}; - -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; - } -} - -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; - mutex mu; -}; - -void AccessRXCache::check_access(const char *filename, bool allow_async, function cb) -{ - lock_guard lock(mu); - 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; - } - - // 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)); - } - }); - } - return; // The rest will happen in async context. - } - - // Passed all checks. - cb(true); -} - class Corpus { public: Corpus(int fd, IOUringEngine *engine); @@ -259,8 +82,7 @@ public: Corpus::Corpus(int fd, IOUringEngine *engine) : fd(fd), engine(engine) { - // Enable to test cold-cache behavior (except for access()). - if (false) { + if (flush_cache) { off_t len = lseek(fd, 0, SEEK_END); if (len == -1) { perror("lseek"); @@ -373,9 +195,19 @@ void scan_file_block(const vector &needles, string_view compressed, for (const char *filename = block.data(); filename != block.data() + block.size(); filename += strlen(filename) + 1) { + const char *haystack = filename; + if (match_basename) { + haystack = strrchr(filename, '/'); + if (haystack == nullptr) { + haystack = filename; + } else { + ++haystack; + } + } + bool found = true; for (const Needle &needle : needles) { - if (!matches(needle, filename)) { + if (!matches(needle, haystack)) { found = false; break; } @@ -399,7 +231,7 @@ size_t scan_docids(const vector &needles, const vector &docids { Serializer docids_in_order; AccessRXCache access_rx_cache(engine); - atomic matched{0}; + atomic 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) { @@ -427,7 +259,8 @@ struct WorkerThread { class WorkerThreadReceiver : public ResultReceiver { public: - WorkerThreadReceiver(WorkerThread *wt) : wt(wt) {} + WorkerThreadReceiver(WorkerThread *wt) + : wt(wt) {} void print(uint64_t seq, uint64_t skip, const string msg) override { @@ -478,7 +311,7 @@ uint64_t scan_all_docids(const vector &needles, int fd, const Corpus &co 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)); - atomic matched{0}; + atomic matched{ 0 }; mutex mu; condition_variable queue_added, queue_removed; @@ -667,6 +500,8 @@ void do_search_file(const vector &needles, const char *filename) // 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); + dprintf("Done in %.1f ms, found %" PRId64 " matches.\n", + 1e3 * duration(steady_clock::now() - start).count(), matched); if (only_count) { printf("%" PRId64 "\n", matched); } @@ -811,46 +646,12 @@ void do_search_file(const vector &needles, const char *filename) } } -string unescape_glob_to_plain_string(const string &needle) -{ - 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 flags = REG_NOSUB; - if (ignore_case) { - 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 '%s': %s\n", needle.c_str(), errbuf); - exit(1); - } - return re; -} - void usage() { printf( "Usage: plocate [OPTION]... PATTERN...\n" "\n" + " -b, --basename search only the file name portion of path names\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" @@ -859,6 +660,7 @@ void usage() " -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" + " -w, --wholename search the entire path name (default; see -b)\n" " --help print this help\n" " --version print version information\n"); } @@ -876,9 +678,11 @@ void version() int main(int argc, char **argv) { constexpr int EXTENDED_REGEX = 1000; + constexpr int FLUSH_CACHE = 1001; static const struct option long_options[] = { { "help", no_argument, 0, 'h' }, { "count", no_argument, 0, 'c' }, + { "basename", no_argument, 0, 'b' }, { "database", required_argument, 0, 'd' }, { "ignore-case", no_argument, 0, 'i' }, { "limit", required_argument, 0, 'l' }, @@ -886,18 +690,24 @@ int main(int argc, char **argv) { "version", no_argument, 0, 'V' }, { "regexp", no_argument, 0, 'r' }, { "regex", no_argument, 0, EXTENDED_REGEX }, + { "wholename", no_argument, 0, 'w' }, { "debug", no_argument, 0, 'D' }, // Not documented. + // Enable to test cold-cache behavior (except for access()). Not documented. + { "flush-cache", no_argument, 0, FLUSH_CACHE }, { 0, 0, 0, 0 } }; setlocale(LC_ALL, ""); for (;;) { int option_index = 0; - int c = getopt_long(argc, argv, "cd:hil:n:0VD", long_options, &option_index); + int c = getopt_long(argc, argv, "bcd:hil:n:0rwVD", long_options, &option_index); if (c == -1) { break; } switch (c) { + case 'b': + match_basename = true; + break; case 'c': only_count = true; break; @@ -928,9 +738,15 @@ int main(int argc, char **argv) patterns_are_regex = true; use_extended_regex = true; break; + case 'w': + match_basename = false; // No-op unless -b is given first. + break; case 'D': use_debug = true; break; + case FLUSH_CACHE: + flush_cache = true; + break; case 'V': version(); break; @@ -939,10 +755,12 @@ int main(int argc, char **argv) } } - if (use_debug) { + if (use_debug || flush_cache) { // 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. + // as root, or use a locally-built file. Doing the same thing for + // flush_cache is mostly paranoia, in an attempt to prevent random users + // from making plocate slow for everyone else. if (setgid(getgid()) != 0) { perror("setgid"); exit(EXIT_FAILURE);