X-Git-Url: https://git.sesse.net/?p=plocate;a=blobdiff_plain;f=plocate.cpp;h=9606d2f49db1892be4a145e752262bffaac77e0b;hp=390bd21e681dcc7ad6d77458e599bb72c5424e17;hb=6dd0338abec6c820f75d21d441b0a60e3427539f;hpb=e3a362166c8012ebc689b437b5ead928c11fbe3e

diff --git a/plocate.cpp b/plocate.cpp
index 390bd21..9606d2f 100644
--- a/plocate.cpp
+++ b/plocate.cpp
@@ -1,112 +1,456 @@
+#include "access_rx_cache.h"
+#include "complete_pread.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 <algorithm>
+#include <assert.h>
+#include <atomic>
+#include <chrono>
+#include <condition_variable>
+#include <deque>
+#include <fcntl.h>
+#include <functional>
+#include <getopt.h>
+#include <inttypes.h>
+#include <iterator>
+#include <limits>
+#include <locale.h>
+#include <memory>
+#include <mutex>
+#include <regex.h>
+#include <stdint.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
-#include <algorithm>
-#include <unordered_map>
 #include <string>
-#include <vector>
-#include <chrono>
+#include <string_view>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <thread>
+#include <tuple>
 #include <unistd.h>
-#include <fcntl.h>
-#include <sys/mman.h>
-#include <arpa/inet.h>
-#include <endian.h>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+#include <vector>
 #include <zstd.h>
 
-#include "vp4.h"
-
-#define P4NENC_BOUND(n) ((n+127)/128+(n+32)*sizeof(uint32_t))
-
 using namespace std;
 using namespace std::chrono;
 
-#define dprintf(...)
-//#define dprintf(...) fprintf(stderr, __VA_ARGS__);
-	
-static inline uint32_t read_unigram(const string &s, size_t idx)
+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<int64_t>::max();
+int64_t limit_left = numeric_limits<int64_t>::max();
+bool stdout_is_tty = false;
+static bool in_forked_child = false;
+
+steady_clock::time_point start;
+ZSTD_DDict *ddict = nullptr;
+
+class Corpus {
+public:
+	Corpus(int fd, IOUringEngine *engine);
+	~Corpus();
+	void find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> cb);
+	void get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const;
+	size_t get_num_filename_blocks() const;
+	off_t offset_for_block(uint32_t docid) const
+	{
+		return hdr.filename_index_offset_bytes + docid * sizeof(uint64_t);
+	}
+	const Header &get_hdr() const { return hdr; }
+
+public:
+	const int fd;
+	IOUringEngine *const engine;
+
+	Header hdr;
+};
+
+Corpus::Corpus(int fd, IOUringEngine *engine)
+	: fd(fd), engine(engine)
 {
-	if (idx < s.size()) {
-		return (unsigned char)s[idx];
-	} else {
-		return 0;
+	if (flush_cache) {
+		off_t len = lseek(fd, 0, SEEK_END);
+		if (len == -1) {
+			perror("lseek");
+			exit(1);
+		}
+		posix_fadvise(fd, 0, len, POSIX_FADV_DONTNEED);
+	}
+
+	complete_pread(fd, &hdr, sizeof(hdr), /*offset=*/0);
+	if (memcmp(hdr.magic, "\0plocate", 8) != 0) {
+		fprintf(stderr, "plocate.db is corrupt or an old version; please rebuild it.\n");
+		exit(1);
+	}
+	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;
+	}
+	if (hdr.max_version < 2) {
+		// This too. (We ignore the other max_version 2 fields.)
+		hdr.check_visibility = true;
 	}
 }
 
-static inline uint32_t read_trigram(const string &s, size_t start)
+Corpus::~Corpus()
 {
-	return read_unigram(s, start) |
-		(read_unigram(s, start + 1) << 8) |
-		(read_unigram(s, start + 2) << 16);
+	close(fd);
 }
 
-bool has_access(const char *filename, unordered_map<string, bool> *access_rx_cache)
+void Corpus::find_trigram(uint32_t trgm, function<void(const Trigram *trgmptr, size_t len)> 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 (!ok) {
-			return false;
+	uint32_t bucket = hash_trigram(trgm, hdr.hashtable_size);
+	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) {
+		const Trigram *trgmptr = reinterpret_cast<const Trigram *>(s.data());
+		for (unsigned i = 0; i < hdr.extra_ht_slots + 1; ++i) {
+			if (trgmptr[i].trgm == trgm) {
+				cb(trgmptr + i, trgmptr[i + 1].offset - trgmptr[i].offset);
+				return;
+			}
 		}
-		end = strchr(end + 1, '/');
-	}
 
-#if 0
-	// Check for rx first in the cache; if that isn't true, check R_OK uncached.
-	// This is roughly the same thing as mlocate does.	
-	auto it = access_rx_cache->find(filename);
-	if (it != access_rx_cache->end() && it->second) {
-		return true;
-	}
+		// Not found.
+		cb(nullptr, 0);
+	});
+}
 
-	return access(filename, R_OK) == 0;
-#endif
-	return true;
+void Corpus::get_compressed_filename_block(uint32_t docid, function<void(string_view)> cb) const
+{
+	// Read the file offset from this docid and the next one.
+	// This is always allowed, since we have a sentinel block at the end.
+	engine->submit_read(fd, sizeof(uint64_t) * 2, offset_for_block(docid), [this, cb{ move(cb) }](string_view s) {
+		const uint64_t *ptr = reinterpret_cast<const uint64_t *>(s.data());
+		off_t offset = ptr[0];
+		size_t len = ptr[1] - ptr[0];
+		engine->submit_read(fd, len, offset, cb);
+	});
 }
 
-struct Trigram {
-	uint32_t trgm;
-	uint32_t num_docids;
-	uint64_t offset;
-};
+size_t Corpus::get_num_filename_blocks() const
+{
+	return hdr.num_docids;
+}
 
-size_t scan_docid(const string &needle, uint32_t docid, const char *data, const uint64_t *filename_offsets, unordered_map<string, bool> *access_rx_cache)
+void scan_file_block(const vector<Needle> &needles, string_view compressed,
+                     AccessRXCache *access_rx_cache, uint64_t seq, ResultReceiver *serializer,
+                     atomic<uint64_t> *matched)
 {
-	const char *compressed = (const char *)(data + filename_offsets[docid]);
-	size_t compressed_size = filename_offsets[docid + 1] - filename_offsets[docid];  // Allowed we have a sentinel block at the end.
-	size_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");
+		exit(1);
+	}
 
 	string block;
-	block.resize(ZSTD_getFrameContentSize(compressed, compressed_size) + 1);
+	block.resize(uncompressed_len + 1);
+
+	static thread_local ZSTD_DCtx *ctx = ZSTD_createDCtx();  // Reused across calls.
+	size_t err;
 
-	ZSTD_decompress(&block[0], block.size(), compressed, compressed_size);
+	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';
 
+	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) {
-		if (strstr(filename, needle.c_str()) == nullptr) {
-			continue;
+	     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;
+			}
 		}
-		if (has_access(filename, access_rx_cache)) {
-			++matched;
-			printf("%s\n", filename);
+
+		bool found = true;
+		for (const Needle &needle : needles) {
+			if (!matches(needle, haystack)) {
+				found = false;
+				break;
+			}
+		}
+		if (found) {
+			if (pending_candidate != nullptr) {
+				test_candidate(pending_candidate, local_seq, local_seq + 1);
+				++local_seq;
+			}
+			pending_candidate = filename;
 		}
 	}
+	if (pending_candidate == nullptr) {
+		serializer->print(seq << 32, 1ULL << 32, "");
+	} else {
+		test_candidate(pending_candidate, local_seq, (seq + 1) << 32);
+	}
+}
+
+size_t scan_docids(const vector<Needle> &needles, const vector<uint32_t> &docids, const Corpus &corpus, IOUringEngine *engine)
+{
+	Serializer docids_in_order;
+	AccessRXCache access_rx_cache(engine, corpus.get_hdr().check_visibility);
+	atomic<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) {
+			scan_file_block(needles, compressed, &access_rx_cache, i, &docids_in_order, &matched);
+		});
+	}
+	engine->finish();
 	return matched;
 }
 
-void do_search_file(const string &needle, const char *filename)
+struct WorkerThread {
+	thread t;
+
+	// We use a result queue instead of synchronizing Serializer,
+	// since a lock on it becomes a huge choke point if there are
+	// lots of threads.
+	mutex result_mu;
+	struct Result {
+		uint64_t seq;
+		uint64_t skip;
+		string msg;
+	};
+	vector<Result> results;
+};
+
+class WorkerThreadReceiver : public ResultReceiver {
+public:
+	WorkerThreadReceiver(WorkerThread *wt)
+		: wt(wt) {}
+
+	void print(uint64_t seq, uint64_t skip, const string msg) override
+	{
+		lock_guard<mutex> lock(wt->result_mu);
+		if (msg.empty() && !wt->results.empty() && wt->results.back().seq + wt->results.back().skip == seq) {
+			wt->results.back().skip += skip;
+		} else {
+			wt->results.emplace_back(WorkerThread::Result{ seq, skip, move(msg) });
+		}
+	}
+
+private:
+	WorkerThread *wt;
+};
+
+void deliver_results(WorkerThread *wt, Serializer *serializer)
+{
+	vector<WorkerThread::Result> results;
+	{
+		lock_guard<mutex> lock(wt->result_mu);
+		results = move(wt->results);
+	}
+	for (const WorkerThread::Result &result : results) {
+		serializer->print(result.seq, result.skip, move(result.msg));
+	}
+}
+
+// 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. Since we assume that
+// we will primarily be CPU-bound, we'll be firing up one
+// worker thread for each spare core (the last one will
+// only be doing I/O). access() is still synchronous.
+uint64_t scan_all_docids(const vector<Needle> &needles, int fd, const Corpus &corpus)
 {
-	int fd = open(filename, O_RDONLY);
+	{
+		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, corpus.get_hdr().check_visibility);
+	Serializer serializer;
+	uint32_t num_blocks = corpus.get_num_filename_blocks();
+	unique_ptr<uint64_t[]> offsets(new uint64_t[num_blocks + 1]);
+	complete_pread(fd, offsets.get(), (num_blocks + 1) * sizeof(uint64_t), corpus.offset_for_block(0));
+	atomic<uint64_t> matched{ 0 };
+
+	mutex mu;
+	condition_variable queue_added, queue_removed;
+	deque<tuple<int, int, string>> work_queue;  // Under mu.
+	bool done = false;  // Under mu.
+
+	unsigned num_threads = max<int>(sysconf(_SC_NPROCESSORS_ONLN) - 1, 1);
+	dprintf("Using %u worker threads for linear scan.\n", num_threads);
+	unique_ptr<WorkerThread[]> threads(new WorkerThread[num_threads]);
+	for (unsigned i = 0; i < num_threads; ++i) {
+		threads[i].t = thread([&threads, &mu, &queue_added, &queue_removed, &work_queue, &done, &offsets, &needles, &access_rx_cache, &matched, i] {
+			// regcomp() takes a lock on the regex, so each thread will need its own.
+			const vector<Needle> *use_needles = &needles;
+			vector<Needle> recompiled_needles;
+			if (i != 0 && patterns_are_regex) {
+				recompiled_needles = needles;
+				for (Needle &needle : recompiled_needles) {
+					needle.re = compile_regex(needle.str);
+				}
+				use_needles = &recompiled_needles;
+			}
+
+			WorkerThreadReceiver receiver(&threads[i]);
+			for (;;) {
+				uint32_t io_docid, last_docid;
+				string compressed;
+
+				{
+					unique_lock<mutex> lock(mu);
+					queue_added.wait(lock, [&work_queue, &done] { return !work_queue.empty() || done; });
+					if (done && work_queue.empty()) {
+						return;
+					}
+					tie(io_docid, last_docid, compressed) = move(work_queue.front());
+					work_queue.pop_front();
+					queue_removed.notify_all();
+				}
+
+				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];
+					scan_file_block(*use_needles, { &compressed[relative_offset], len }, &access_rx_cache, docid, &receiver, &matched);
+				}
+			}
+		});
+	}
+
+	string compressed;
+	for (uint32_t io_docid = 0; io_docid < num_blocks; io_docid += 32) {
+		uint32_t last_docid = std::min(io_docid + 32, num_blocks);
+		size_t io_len = offsets[last_docid] - offsets[io_docid];
+		if (compressed.size() < io_len) {
+			compressed.resize(io_len);
+		}
+		complete_pread(fd, &compressed[0], io_len, offsets[io_docid]);
+
+		{
+			unique_lock<mutex> lock(mu);
+			queue_removed.wait(lock, [&work_queue] { return work_queue.size() < 256; });  // Allow ~2MB of data queued up.
+			work_queue.emplace_back(io_docid, last_docid, move(compressed));
+			queue_added.notify_one();  // Avoid the thundering herd.
+		}
+
+		// Pick up some results, so that we are sure that we won't just overload.
+		// (Seemingly, going through all of these causes slowness with many threads,
+		// but taking only one is OK.)
+		unsigned i = io_docid / 32;
+		deliver_results(&threads[i % num_threads], &serializer);
+	}
+	{
+		lock_guard<mutex> lock(mu);
+		done = true;
+		queue_added.notify_all();
+	}
+	for (unsigned i = 0; i < num_threads; ++i) {
+		threads[i].t.join();
+		deliver_results(&threads[i], &serializer);
+	}
+	return matched;
+}
+
+// Takes the given posting list, unions it into the parts of the trigram disjunction
+// already read; if the list is complete, intersects with “cur_candidates”.
+//
+// Returns true if the search should be aborted (we are done).
+bool new_posting_list_read(TrigramDisjunction *td, vector<uint32_t> decoded, vector<uint32_t> *cur_candidates, vector<uint32_t> *tmp)
+{
+	if (td->docids.empty()) {
+		td->docids = move(decoded);
+	} else {
+		tmp->clear();
+		set_union(decoded.begin(), decoded.end(), td->docids.begin(), td->docids.end(), back_inserter(*tmp));
+		swap(*tmp, td->docids);
+	}
+	if (--td->remaining_trigrams_to_read > 0) {
+		// 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());
+		}
+		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());
+		}
+		*cur_candidates = move(td->docids);
+	} else {
+		tmp->clear();
+		set_intersection(cur_candidates->begin(), cur_candidates->end(),
+		                 td->docids.begin(), td->docids.end(),
+		                 back_inserter(*tmp));
+		swap(*cur_candidates, *tmp);
+		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());
+				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());
+			}
+		}
+	}
+	return false;
+}
+
+uint64_t do_search_file(const vector<Needle> &needles, const std::string &filename)
+{
+	int fd = open(filename.c_str(), O_RDONLY);
 	if (fd == -1) {
-		perror(filename);
+		perror(filename.c_str());
 		exit(1);
 	}
 
@@ -116,105 +460,508 @@ void do_search_file(const string &needle, const char *filename)
 		exit(EXIT_FAILURE);
 	}
 
-	//steady_clock::time_point start = steady_clock::now();
+	start = steady_clock::now();
 	if (access("/", R_OK | X_OK)) {
 		// We can't find anything, no need to bother...
-		return;
+		return 0;
 	}
 
-	off_t len = lseek(fd, 0, SEEK_END);
-	if (len == -1) {
-		perror("lseek");
-		exit(1);
-	}
-	const char *data = (char *)mmap(nullptr, len, PROT_READ, MAP_SHARED, fd, /*offset=*/0);
-	if (data == MAP_FAILED) {
-		perror("mmap");
-		exit(1);
+	IOUringEngine engine(/*slop_bytes=*/16);  // 16 slop bytes as described in turbopfor.h.
+	Corpus corpus(fd, &engine);
+	dprintf("Corpus init done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
+
+	vector<TrigramDisjunction> 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);
+		}
 	}
-	uint64_t num_trigrams = *(const uint64_t *)data;
-	uint64_t filename_index_offset = *(const uint64_t *)(data + sizeof(uint64_t));
 
-	const Trigram *trgm_begin = (Trigram *)(data + sizeof(uint64_t) * 2);
-	const Trigram *trgm_end = trgm_begin + num_trigrams;
+	unique_sort(
+		&trigram_groups,
+		[](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives < b.trigram_alternatives; },
+		[](const TrigramDisjunction &a, const TrigramDisjunction &b) { return a.trigram_alternatives == b.trigram_alternatives; });
 
-	vector<const Trigram *> trigrams;
-	for (size_t i = 0; i < needle.size() - 2; ++i) {
-		uint32_t trgm = read_trigram(needle, i);
-		const Trigram *trgmptr = lower_bound(trgm_begin, trgm_end, trgm, [](const Trigram &trgm, uint32_t t) {
-			return trgm.trgm < t;
-		});
-		if (trgmptr == trgm_end || trgmptr->trgm != trgm) {
-			dprintf("trigram %06x isn't found, we abort the search\n", trgm);
-			munmap((void *)data, len);
-			close(fd);
-			return;
+	// Give them names for debugging.
+	unsigned td_index = 0;
+	for (TrigramDisjunction &td : trigram_groups) {
+		td.index = td_index++;
+	}
+
+	// Collect which trigrams we need to look up in the hash table.
+	unordered_map<uint32_t, vector<TrigramDisjunction *>> trigrams_to_lookup;
+	for (TrigramDisjunction &td : trigram_groups) {
+		for (uint32_t trgm : td.trigram_alternatives) {
+			trigrams_to_lookup[trgm].push_back(&td);
 		}
-		trigrams.push_back(trgmptr);
 	}
-	sort(trigrams.begin(), trigrams.end());
+	if (trigrams_to_lookup.empty()) {
+		// Too short for trigram matching. Apply brute force.
+		// (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);
+		dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
+		        1e3 * duration<float>(steady_clock::now() - start).count(), matched);
+		return matched;
+	}
+
+	// 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.
 	{
-		auto last = unique(trigrams.begin(), trigrams.end());
-		trigrams.erase(last, trigrams.end());
+		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<float>(steady_clock::now() - start).count());
+			});
+		}
 	}
-	sort(trigrams.begin(), trigrams.end(), [&](const Trigram *a, const Trigram *b) {
-		return a->num_docids < b->num_docids;
-	});
 
-	vector<uint32_t> in1, in2, out;
-	for (const Trigram *trgmptr : trigrams) {
-		//uint32_t trgm = trgmptr->trgm;
-		size_t num = trgmptr->num_docids;
-		unsigned char *pldata = (unsigned char *)(data + trgmptr->offset);
-		if (in1.empty()) {
-			in1.resize(num + 128);
-			p4nd1dec128v32((unsigned char *)pldata, num, &in1[0]);
-			in1.resize(num);
-			dprintf("trigram '%c%c%c' decoded to %zu entries\n", trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, num);
-		} else {
-			if (num > in1.size() * 100) {
-				dprintf("trigram '%c%c%c' has %zu entries, ignoring the rest (will weed out false positives later)\n",
-					trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, num);
-				break;
+	// Look them all up on disk.
+	bool should_early_exit = false;
+	for (auto &[trgm, trigram_groups] : trigrams_to_lookup) {
+		corpus.find_trigram(trgm, [trgm{ trgm }, trigram_groups{ &trigram_groups }, &should_early_exit](const Trigram *trgmptr, size_t len) {
+			if (trgmptr == nullptr) {
+				dprintf("trigram %s isn't found\n", print_trigram(trgm).c_str());
+				for (TrigramDisjunction *td : *trigram_groups) {
+					--td->remaining_trigrams_to_read;
+
+					// If we now know this trigram group doesn't match anything at all,
+					// we can do early exit; however, if we're in a forked child,
+					// that would confuse the parent process (since we don't write
+					// our count to the pipe), so we wait until we're back in to the
+					// regular (non-async) context. This is a fairly rare case anyway,
+					// and the gains from dropping the remaining trigram reads are limited.
+					if (td->remaining_trigrams_to_read == 0 && td->read_trigrams.empty()) {
+						if (in_forked_child) {
+							should_early_exit = true;
+						} else {
+							dprintf("zero matches in %s, so we are done\n", print_td(*td).c_str());
+							if (only_count) {
+								printf("0\n");
+							}
+							exit(0);
+						}
+					}
+				}
+				return;
+			}
+			for (TrigramDisjunction *td : *trigram_groups) {
+				--td->remaining_trigrams_to_read;
+				td->max_num_docids += trgmptr->num_docids;
+				td->read_trigrams.emplace_back(*trgmptr, len);
 			}
+		});
+	}
+	engine.finish();
+	dprintf("Hashtable lookups done after %.1f ms.\n", 1e3 * duration<float>(steady_clock::now() - start).count());
+
+	if (should_early_exit) {
+		return 0;
+	}
+
+	for (TrigramDisjunction &td : trigram_groups) {
+		// Reset for reads.
+		td.remaining_trigrams_to_read = td.read_trigrams.size();
+
+		if (ignore_case) {  // If case-sensitive, they'll all be pretty obvious single-entry groups.
+			dprintf("OR group %u (max_num_docids=%u): %s\n", td.index, td.max_num_docids, print_td(td).c_str());
+		}
+	}
+
+	// TODO: For case-insensitive (ie. more than one alternative in each),
+	// prioritize the ones with fewer seeks?
+	sort(trigram_groups.begin(), trigram_groups.end(),
+	     [&](const TrigramDisjunction &a, const TrigramDisjunction &b) {
+		     return a.max_num_docids < b.max_num_docids;
+	     });
+
+	unordered_map<uint32_t, vector<TrigramDisjunction *>> uses_trigram;
+	for (TrigramDisjunction &td : trigram_groups) {
+		for (uint32_t trgm : td.trigram_alternatives) {
+			uses_trigram[trgm].push_back(&td);
+		}
+	}
+
+	unordered_set<uint32_t> trigrams_submitted_read;
+	vector<uint32_t> cur_candidates, tmp, decoded;
+	bool done = false;
+	for (TrigramDisjunction &td : trigram_groups) {
+		if (!cur_candidates.empty() && td.max_num_docids > cur_candidates.size() * 100) {
+			dprintf("%s has up to %u entries, ignoring the rest (will "
+			        "weed out false positives later)\n",
+			        print_td(td).c_str(), td.max_num_docids);
+			break;
+		}
 
-			if (in2.size() < num + 128) {
-				in2.resize(num + 128);
+		for (auto &[trgmptr, len] : td.read_trigrams) {
+			if (trigrams_submitted_read.count(trgmptr.trgm) != 0) {
+				continue;
 			}
-			p4nd1dec128v32((unsigned char *)pldata, num, &in2[0]);
-
-			out.clear();
-			set_intersection(in1.begin(), in1.end(), in2.begin(), in2.begin() + num, back_inserter(out));
-			swap(in1, out);
-			dprintf("trigram '%c%c%c' decoded to %zu entries, %zu left\n", trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, num, in1.size());
-			if (in1.empty()) {
-				dprintf("no matches (intersection list is empty)\n");
-				break;
+			trigrams_submitted_read.insert(trgmptr.trgm);
+			// Only stay a certain amount ahead, so that we don't spend I/O
+			// on reading the latter, large posting lists. We are unlikely
+			// to need them anyway, even if they should come in first.
+			if (engine.get_waiting_reads() >= 5) {
+				engine.finish();
+				if (done)
+					break;
 			}
+			engine.submit_read(fd, len, trgmptr.offset, [trgmptr{ trgmptr }, len{ len }, &done, &cur_candidates, &tmp, &decoded, &uses_trigram](string_view s) {
+				if (done)
+					return;
+
+				uint32_t trgm = trgmptr.trgm;
+				const unsigned char *pldata = reinterpret_cast<const unsigned char *>(s.data());
+				size_t num = trgmptr.num_docids;
+				decoded.resize(num);
+				decode_pfor_delta1_128(pldata, num, /*interleaved=*/true, &decoded[0]);
+
+				assert(uses_trigram.count(trgm) != 0);
+				bool was_empty = cur_candidates.empty();
+				if (ignore_case) {
+					dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
+				}
+
+				for (TrigramDisjunction *td : uses_trigram[trgm]) {
+					done |= new_posting_list_read(td, decoded, &cur_candidates, &tmp);
+					if (done)
+						break;
+				}
+				if (!ignore_case) {
+					if (was_empty) {
+						dprintf("trigram %s (%zu bytes) decoded to %zu entries\n", print_trigram(trgm).c_str(), len, num);
+					} else if (cur_candidates.empty()) {
+						dprintf("trigram %s (%zu bytes) decoded to %zu entries (none left, search is done)\n", print_trigram(trgm).c_str(), len, num);
+					} else {
+						dprintf("trigram %s (%zu bytes) decoded to %zu entries (%zu left)\n", print_trigram(trgm).c_str(), len, num, cur_candidates.size());
+					}
+				}
+			});
 		}
 	}
-	steady_clock::time_point end = steady_clock::now();
+	engine.finish();
+	if (done) {
+		return 0;
+	}
+	dprintf("Intersection done after %.1f ms. Doing final verification and printing:\n",
+	        1e3 * duration<float>(steady_clock::now() - start).count());
 
-	dprintf("Intersection took %.1f ms. Doing final verification and printing:\n",
-		1e3 * duration<float>(end - start).count());
+	uint64_t matched = scan_docids(needles, cur_candidates, corpus, &engine);
+	dprintf("Done in %.1f ms, found %" PRId64 " matches.\n",
+	        1e3 * duration<float>(steady_clock::now() - start).count(), matched);
+	return matched;
+}
 
-	unordered_map<string, bool> access_rx_cache;
+// Run do_search_file() in a child process.
+//
+// The reason for this is that we're not robust against malicious input, so we need
+// to drop privileges after opening the file. (Otherwise, we could fall prey to an attack
+// where a user does locate -d badfile.db:/var/lib/plocate/plocate.db, badfile.db contains
+// a buffer overflow that takes over the process, and then uses the elevated privileges
+// to print out otherwise inaccessible paths.) We solve this by forking and treating the
+// child process as untrusted after it has dropped its privileges (which it does before
+// reading any data from the file); it returns a single 64-bit number over a pipe,
+// and that's it. The parent keeps its privileges, and can then fork out new children
+// without fear of being taken over. (The child keeps stdout for outputting results.)
+//
+// The count is returned over the pipe, because it's needed both for --limit and --count.
+uint64_t do_search_file_in_child(const vector<Needle> &needles, const std::string &filename)
+{
+	int pipefd[2];
+	if (pipe(pipefd) == -1) {
+		perror("pipe");
+		exit(EXIT_FAILURE);
+	}
 
-	const uint64_t *filename_offsets = (const uint64_t *)(data + filename_index_offset);
-	int matched = 0;
-	for (uint32_t docid : in1) {
-		matched += scan_docid(needle, docid, data, filename_offsets, &access_rx_cache);
+	pid_t child_pid = fork();
+	switch (child_pid) {
+	case 0: {
+		// Child.
+		close(pipefd[0]);
+		in_forked_child = true;
+		uint64_t matched = do_search_file(needles, filename);
+		int ret;
+		do {
+			ret = write(pipefd[1], &matched, sizeof(matched));
+		} while (ret == -1 && errno == EINTR);
+		if (ret != sizeof(matched)) {
+			perror("write");
+			_exit(EXIT_FAILURE);
+		}
+		_exit(EXIT_SUCCESS);
+	}
+	case -1:
+		// Error.
+		perror("fork");
+		exit(EXIT_FAILURE);
+	default:
+		// Parent.
+		close(pipefd[1]);
+		break;
 	}
-	end = steady_clock::now();
-	dprintf("Done in %.1f ms, found %d matches.\n",
-		1e3 * duration<float>(end - start).count(), matched);
 
-	munmap((void *)data, len);
-	close(fd);
+	// Wait for the child to finish.
+	int wstatus;
+	pid_t err;
+	do {
+		err = waitpid(child_pid, &wstatus, 0);
+	} while (err == -1 && errno == EINTR);
+	if (err == -1) {
+		perror("waitpid");
+		exit(EXIT_FAILURE);
+	}
+	if (WIFEXITED(wstatus)) {
+		if (WEXITSTATUS(wstatus) != 0) {
+			// The child has probably already printed out its error, so just propagate the exit status.
+			exit(WEXITSTATUS(wstatus));
+		}
+		// Success!
+	} else if (!WIFEXITED(wstatus)) {
+		fprintf(stderr, "FATAL: Child died unexpectedly while processing %s\n", filename.c_str());
+		exit(1);
+	}
+
+	// Now get the number of matches from the child.
+	uint64_t matched;
+	int ret;
+	do {
+		ret = read(pipefd[0], &matched, sizeof(matched));
+	} while (ret == -1 && errno == EINTR);
+	if (ret == -1) {
+		perror("read");
+		exit(EXIT_FAILURE);
+	} else if (ret != sizeof(matched)) {
+		fprintf(stderr, "FATAL: Short read through pipe (got %d bytes)\n", ret);
+		exit(EXIT_FAILURE);
+	}
+	close(pipefd[0]);
+	return matched;
+}
+
+// Parses a colon-separated list of strings and appends them onto the given vector.
+// Backslash escapes whatever comes after it.
+void parse_dbpaths(const char *ptr, vector<string> *output)
+{
+	string str;
+	while (*ptr != '\0') {
+		if (*ptr == '\\') {
+			if (ptr[1] == '\0') {
+				fprintf(stderr, "ERROR: Escape character at the end of string\n");
+				exit(EXIT_FAILURE);
+			}
+			// Escape.
+			str.push_back(ptr[1]);
+			ptr += 2;
+			continue;
+		}
+		if (*ptr == ':') {
+			// Separator.
+			output->push_back(move(str));
+			++ptr;
+			continue;
+		}
+		str.push_back(*ptr++);
+	}
+	output->push_back(move(str));
+}
+
+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 " DBFILE ")\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"
+		"  -w, --wholename        search the entire path name (default; see -b)\n"
+		"      --help             print this help\n"
+		"      --version          print version information\n");
+}
+
+void version()
+{
+	printf("%s %s\n", PACKAGE_NAME, PACKAGE_VERSION);
+	printf("Copyright 2020 Steinar H. Gunderson\n");
+	printf("License GPLv2+: GNU GPL version 2 or later <https://gnu.org/licenses/gpl.html>.\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)
 {
-	//do_search_file(argv[1], "all.trgm");
-	do_search_file(argv[1], "/var/lib/mlocate/plocate.db");
+	vector<string> dbpaths;
+
+	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' },
+		{ "null", no_argument, 0, '0' },
+		{ "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, "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;
+		case 'd':
+			parse_dbpaths(optarg, &dbpaths);
+			break;
+		case 'h':
+			usage();
+			exit(0);
+		case 'i':
+			ignore_case = true;
+			break;
+		case 'l':
+		case 'n':
+			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 '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;
+		default:
+			exit(1);
+		}
+	}
+
+	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. 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);
+		}
+	}
+
+	if (!print_nul) {
+		stdout_is_tty = isatty(1);
+	}
+
+	vector<Needle> needles;
+	for (int i = optind; i < argc; ++i) {
+		Needle needle;
+		needle.str = argv[i];
+
+		// 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 = 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));
+	}
+	if (needles.empty()) {
+		fprintf(stderr, "plocate: no pattern to search for specified\n");
+		exit(0);
+	}
+
+	if (dbpaths.empty()) {
+		// No -d given, so use our default. Note that this happens
+		// even if LOCATE_PATH exists, to match mlocate behavior.
+		dbpaths.push_back(DBFILE);
+	}
+
+	const char *locate_path = getenv("LOCATE_PATH");
+	if (locate_path != nullptr) {
+		parse_dbpaths(locate_path, &dbpaths);
+	}
+
+	uint64_t matched = 0;
+	for (size_t i = 0; i < dbpaths.size(); ++i) {
+		uint64_t this_matched;
+		if (i != dbpaths.size() - 1) {
+			this_matched = do_search_file_in_child(needles, dbpaths[i]);
+		} else {
+			this_matched = do_search_file(needles, dbpaths[i]);
+		}
+		matched += this_matched;
+		limit_left -= this_matched;
+	}
+	if (only_count) {
+		printf("%" PRId64 "\n", matched);
+	}
 }