1 | #include "filecache.hfa" |
---|
2 | |
---|
3 | #include <assert.h> |
---|
4 | #include <string.h> |
---|
5 | |
---|
6 | #include <stdlib.hfa> |
---|
7 | |
---|
8 | #include <errno.h> |
---|
9 | #include <unistd.h> |
---|
10 | extern "C" { |
---|
11 | #include <fcntl.h> |
---|
12 | #include <ftw.h> |
---|
13 | } |
---|
14 | |
---|
15 | #include "options.hfa" |
---|
16 | |
---|
17 | static inline uint32_t murmur_32_scramble(uint32_t k) { |
---|
18 | k *= 0xcc9e2d51; |
---|
19 | k = (k << 15) | (k >> 17); |
---|
20 | k *= 0x1b873593; |
---|
21 | return k; |
---|
22 | } |
---|
23 | |
---|
24 | uint32_t murmur3_32(const uint8_t* key, size_t len, uint32_t seed) |
---|
25 | { |
---|
26 | uint32_t h = seed; |
---|
27 | uint32_t k; |
---|
28 | /* Read in groups of 4. */ |
---|
29 | for (size_t i = len >> 2; i; i--) { |
---|
30 | // Here is a source of differing results across endiannesses. |
---|
31 | // A swap here has no effects on hash properties though. |
---|
32 | memcpy(&k, key, sizeof(uint32_t)); |
---|
33 | key += sizeof(uint32_t); |
---|
34 | h ^= murmur_32_scramble(k); |
---|
35 | h = (h << 13) | (h >> 19); |
---|
36 | h = h * 5 + 0xe6546b64; |
---|
37 | } |
---|
38 | /* Read the rest. */ |
---|
39 | k = 0; |
---|
40 | for (size_t i = len & 3; i; i--) { |
---|
41 | k <<= 8; |
---|
42 | k |= key[i - 1]; |
---|
43 | } |
---|
44 | // A swap is *not* necessary here because the preceding loop already |
---|
45 | // places the low bytes in the low places according to whatever endianness |
---|
46 | // we use. Swaps only apply when the memory is copied in a chunk. |
---|
47 | h ^= murmur_32_scramble(k); |
---|
48 | /* Finalize. */ |
---|
49 | h ^= len; |
---|
50 | h ^= h >> 16; |
---|
51 | h *= 0x85ebca6b; |
---|
52 | h ^= h >> 13; |
---|
53 | h *= 0xc2b2ae35; |
---|
54 | h ^= h >> 16; |
---|
55 | return h; |
---|
56 | } |
---|
57 | |
---|
58 | |
---|
59 | struct { |
---|
60 | cache_line * entries; |
---|
61 | size_t size; |
---|
62 | } file_cache; |
---|
63 | |
---|
64 | void ?{}( cache_line & this ) with( this ) { |
---|
65 | file = 0p; |
---|
66 | size = 0; |
---|
67 | fd = 0; |
---|
68 | } |
---|
69 | |
---|
70 | [int fd, size_t size] get_file( * const char file, size_t len ) { |
---|
71 | uint32_t idx = murmur3_32( (const uint8_t *)file, len, options.hash_seed ) % file_cache.size; |
---|
72 | |
---|
73 | for(int i = 0;; i++) { |
---|
74 | assert( i < file_cache.size ); |
---|
75 | cache_line & entry = file_cache.entries[idx]; |
---|
76 | if( !entry.file ) return [-1, 0]; |
---|
77 | #if !defined(REJECT_CONFLICTS) |
---|
78 | if( strncmp(entry.file, file, len) != 0 ) { |
---|
79 | idx = (idx + 1) % file_cache.size; |
---|
80 | continue; |
---|
81 | } |
---|
82 | #endif |
---|
83 | return [entry.fd, entry.size]; |
---|
84 | } |
---|
85 | } |
---|
86 | |
---|
87 | int put_file( cache_line & entry ) { |
---|
88 | uint32_t idx = murmur3_32( (const uint8_t *)entry.file, strlen(entry.file), options.hash_seed ) % file_cache.size; |
---|
89 | |
---|
90 | int i = 0; |
---|
91 | for(;file_cache.entries[idx].file; i++ ) { |
---|
92 | assert( i < file_cache.size ); |
---|
93 | idx = (idx + 1) % file_cache.size; |
---|
94 | } |
---|
95 | |
---|
96 | file_cache.entries[idx] = entry; |
---|
97 | return i > 0 ? 1 : 0; |
---|
98 | } |
---|
99 | |
---|
100 | // int ftw(const char *dirpath, int (*fn) (const char *fpath, const struct stat *sb, int typeflag), int nopenfd) |
---|
101 | void fill_cache( const char * path ) { |
---|
102 | int ret; |
---|
103 | size_t fcount = 0; |
---|
104 | size_t fsize = 16; |
---|
105 | cache_line * raw = 0p; |
---|
106 | raw = alloc(raw, fsize, true); |
---|
107 | // Step 1 get a dense array of all files |
---|
108 | int walk(const char *fpath, const struct stat *sb, int typeflag) { |
---|
109 | if(typeflag != FTW_F) return 0; |
---|
110 | |
---|
111 | int idx = fcount; |
---|
112 | fcount++; |
---|
113 | if(fcount > fsize) { |
---|
114 | fsize *= 2; |
---|
115 | raw = alloc(raw, fsize, true); |
---|
116 | } |
---|
117 | |
---|
118 | raw[idx].file = strdup(fpath+2); |
---|
119 | raw[idx].size = sb->st_size; |
---|
120 | raw[idx].fd = open( fpath, options.open_flags ); |
---|
121 | if(raw[idx].fd < 0) { |
---|
122 | abort( "open file error: (%d) %s\n", (int)errno, strerror(errno) ); |
---|
123 | } |
---|
124 | return 0; |
---|
125 | } |
---|
126 | |
---|
127 | ret = ftw(path, walk, 10); |
---|
128 | if(ret < 0) { |
---|
129 | abort( "ftw error: (%d) %s\n", (int)errno, strerror(errno) ); |
---|
130 | } |
---|
131 | |
---|
132 | if(fcount == 0) { |
---|
133 | abort("No file found in path %s\n", path); |
---|
134 | } |
---|
135 | |
---|
136 | // Step 2 create the cache |
---|
137 | file_cache.size = options.file_cache_size > 0 ? options.file_cache_size : fsize; |
---|
138 | if( file_cache.size < fcount ) { |
---|
139 | abort("File Cache too small\n"); |
---|
140 | } |
---|
141 | |
---|
142 | file_cache.entries = anew(fsize); |
---|
143 | |
---|
144 | // Step 3 fill the cache |
---|
145 | int conflicts = 0; |
---|
146 | for(i; fcount) { |
---|
147 | printf("Added file %s\n", raw[i].file); |
---|
148 | conflicts += put_file( raw[i] ); |
---|
149 | } |
---|
150 | printf("Filled cache from path \"%s\" with %zu files\n", path, fcount); |
---|
151 | if( conflicts > 0 ) { |
---|
152 | printf("Found %d conflicts (seed: %u)\n", conflicts, options.hash_seed); |
---|
153 | #if defined(REJECT_CONFLICTS) |
---|
154 | abort("Conflicts found in the cache"); |
---|
155 | #endif |
---|
156 | } |
---|
157 | |
---|
158 | // Step 4 clean up |
---|
159 | free( raw ); |
---|
160 | } |
---|
161 | |
---|
162 | void close_cache() { |
---|
163 | for(idx; file_cache.size) { |
---|
164 | cache_line & entry = file_cache.entries[idx]; |
---|
165 | if( !entry.file ) continue; |
---|
166 | |
---|
167 | int ret = close( entry.fd ); |
---|
168 | if(ret < 0) { |
---|
169 | abort( "close file error: (%d) %s\n", (int)errno, strerror(errno) ); |
---|
170 | } |
---|
171 | } |
---|
172 | |
---|
173 | delete( file_cache.entries ); |
---|
174 | } |
---|