4 #include <unordered_map>
11 #include <arpa/inet.h>
17 #define P4NENC_BOUND(n) ((n+127)/128+(n+32)*sizeof(uint32_t))
20 using namespace std::chrono;
23 //#define dprintf(...) fprintf(stderr, __VA_ARGS__);
25 static inline uint32_t read_unigram(const string &s, size_t idx)
28 return (unsigned char)s[idx];
34 static inline uint32_t read_trigram(const string &s, size_t start)
36 return read_unigram(s, start) |
37 (read_unigram(s, start + 1) << 8) |
38 (read_unigram(s, start + 2) << 16);
41 bool has_access(const char *filename, unordered_map<string, bool> *access_rx_cache)
43 const char *end = strchr(filename + 1, '/');
44 while (end != nullptr) {
45 string parent_path(filename, end);
46 auto it = access_rx_cache->find(parent_path);
48 if (it == access_rx_cache->end()) {
49 ok = access(parent_path.c_str(), R_OK | X_OK) == 0;
50 access_rx_cache->emplace(move(parent_path), ok);
57 end = strchr(end + 1, '/');
61 // Check for rx first in the cache; if that isn't true, check R_OK uncached.
62 // This is roughly the same thing as mlocate does.
63 auto it = access_rx_cache->find(filename);
64 if (it != access_rx_cache->end() && it->second) {
68 return access(filename, R_OK) == 0;
79 void do_search_file(const string &needle, const char *filename)
81 int fd = open(filename, O_RDONLY);
88 if (setgid(getgid()) != 0) {
93 //steady_clock::time_point start = steady_clock::now();
94 if (access("/", R_OK | X_OK)) {
95 // We can't find anything, no need to bother...
99 off_t len = lseek(fd, 0, SEEK_END);
104 const char *data = (char *)mmap(nullptr, len, PROT_READ, MAP_SHARED, fd, /*offset=*/0);
105 if (data == MAP_FAILED) {
109 uint64_t num_trigrams = *(const uint64_t *)data;
110 uint64_t filename_index_offset = *(const uint64_t *)(data + sizeof(uint64_t));
112 const Trigram *trgm_begin = (Trigram *)(data + sizeof(uint64_t) * 2);
113 const Trigram *trgm_end = trgm_begin + num_trigrams;
115 vector<const Trigram *> trigrams;
116 for (size_t i = 0; i < needle.size() - 2; ++i) {
117 uint32_t trgm = read_trigram(needle, i);
118 const Trigram *trgmptr = lower_bound(trgm_begin, trgm_end, trgm, [](const Trigram &trgm, uint32_t t) {
119 return trgm.trgm < t;
121 if (trgmptr == trgm_end || trgmptr->trgm != trgm) {
122 dprintf("trigram %06x isn't found, we abort the search\n", trgm);
123 munmap((void *)data, len);
127 trigrams.push_back(trgmptr);
129 sort(trigrams.begin(), trigrams.end());
131 auto last = unique(trigrams.begin(), trigrams.end());
132 trigrams.erase(last, trigrams.end());
134 sort(trigrams.begin(), trigrams.end(), [&](const Trigram *a, const Trigram *b) {
135 return a->num_docids < b->num_docids;
138 vector<uint32_t> in1, in2, out;
139 for (const Trigram *trgmptr : trigrams) {
140 //uint32_t trgm = trgmptr->trgm;
141 size_t num = trgmptr->num_docids;
142 unsigned char *pldata = (unsigned char *)(data + trgmptr->offset);
144 in1.resize(num + 128);
145 p4nd1dec128v32((unsigned char *)pldata, num, &in1[0]);
147 dprintf("trigram '%c%c%c' decoded to %zu entries\n", trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, num);
149 if (num > in1.size() * 100) {
150 dprintf("trigram '%c%c%c' has %zu entries, ignoring the rest (will weed out false positives later)\n",
151 trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, num);
155 if (in2.size() < num + 128) {
156 in2.resize(num + 128);
158 p4nd1dec128v32((unsigned char *)pldata, num, &in2[0]);
161 set_intersection(in1.begin(), in1.end(), in2.begin(), in2.begin() + num, back_inserter(out));
163 dprintf("trigram '%c%c%c' decoded to %zu entries, %zu left\n", trgm & 0xff, (trgm >> 8) & 0xff, (trgm >> 16) & 0xff, num, in1.size());
165 dprintf("no matches (intersection list is empty)\n");
170 steady_clock::time_point end = steady_clock::now();
172 dprintf("Intersection took %.1f ms. Doing final verification and printing:\n",
173 1e3 * duration<float>(end - start).count());
175 unordered_map<string, bool> access_rx_cache;
177 const uint64_t *filename_offsets = (const uint64_t *)(data + filename_index_offset);
179 for (uint32_t docid : in1) {
180 const char *compressed = (const char *)(data + filename_offsets[docid]);
181 size_t compressed_size = filename_offsets[docid + 1] - filename_offsets[docid]; // Allowed we have a sentinel block at the end.
184 block.resize(ZSTD_getFrameContentSize(compressed, compressed_size) + 1);
186 ZSTD_decompress(&block[0], block.size(), compressed, compressed_size);
187 block[block.size() - 1] = '\0';
189 for (const char *filename = block.data();
190 filename != block.data() + block.size();
191 filename += strlen(filename) + 1) {
192 if (strstr(filename, needle.c_str()) == nullptr) {
195 if (has_access(filename, &access_rx_cache)) {
197 printf("%s\n", filename);
201 end = steady_clock::now();
202 dprintf("Done in %.1f ms, found %d matches.\n",
203 1e3 * duration<float>(end - start).count(), matched);
205 munmap((void *)data, len);
209 int main(int argc, char **argv)
211 //do_search_file(argv[1], "all.trgm");
212 do_search_file(argv[1], "/var/lib/mlocate/plocate.db");