/src/openssl31/crypto/lhash/lhash.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright 1995-2023 The OpenSSL Project Authors. All Rights Reserved. |
3 | | * |
4 | | * Licensed under the Apache License 2.0 (the "License"). You may not use |
5 | | * this file except in compliance with the License. You can obtain a copy |
6 | | * in the file LICENSE in the source distribution or at |
7 | | * https://www.openssl.org/source/license.html |
8 | | */ |
9 | | |
10 | | #include <stdio.h> |
11 | | #include <string.h> |
12 | | #include <stdlib.h> |
13 | | #include <openssl/crypto.h> |
14 | | #include <openssl/lhash.h> |
15 | | #include <openssl/err.h> |
16 | | #include "crypto/ctype.h" |
17 | | #include "crypto/lhash.h" |
18 | | #include "lhash_local.h" |
19 | | |
20 | | /* |
21 | | * A hashing implementation that appears to be based on the linear hashing |
22 | | * algorithm: |
23 | | * https://en.wikipedia.org/wiki/Linear_hashing |
24 | | * |
25 | | * Litwin, Witold (1980), "Linear hashing: A new tool for file and table |
26 | | * addressing", Proc. 6th Conference on Very Large Databases: 212-223 |
27 | | * https://hackthology.com/pdfs/Litwin-1980-Linear_Hashing.pdf |
28 | | * |
29 | | * From the Wikipedia article "Linear hashing is used in the BDB Berkeley |
30 | | * database system, which in turn is used by many software systems such as |
31 | | * OpenLDAP, using a C implementation derived from the CACM article and first |
32 | | * published on the Usenet in 1988 by Esmond Pitt." |
33 | | * |
34 | | * The CACM paper is available here: |
35 | | * https://pdfs.semanticscholar.org/ff4d/1c5deca6269cc316bfd952172284dbf610ee.pdf |
36 | | */ |
37 | | |
38 | | #undef MIN_NODES |
39 | 2.66M | #define MIN_NODES 16 |
40 | 305k | #define UP_LOAD (2*LH_LOAD_MULT) /* load times 256 (default 2) */ |
41 | 305k | #define DOWN_LOAD (LH_LOAD_MULT) /* load times 256 (default 1) */ |
42 | | |
43 | | static int expand(OPENSSL_LHASH *lh); |
44 | | static void contract(OPENSSL_LHASH *lh); |
45 | | static OPENSSL_LH_NODE **getrn(OPENSSL_LHASH *lh, const void *data, unsigned long *rhash); |
46 | | |
47 | | OPENSSL_LHASH *OPENSSL_LH_new(OPENSSL_LH_HASHFUNC h, OPENSSL_LH_COMPFUNC c) |
48 | 305k | { |
49 | 305k | OPENSSL_LHASH *ret; |
50 | | |
51 | 305k | if ((ret = OPENSSL_zalloc(sizeof(*ret))) == NULL) { |
52 | | /* |
53 | | * Do not set the error code, because the ERR code uses LHASH |
54 | | * and we want to avoid possible endless error loop. |
55 | | * ERR_raise(ERR_LIB_CRYPTO, ERR_R_MALLOC_FAILURE); |
56 | | */ |
57 | 0 | return NULL; |
58 | 0 | } |
59 | 305k | if ((ret->b = OPENSSL_zalloc(sizeof(*ret->b) * MIN_NODES)) == NULL) |
60 | 0 | goto err; |
61 | 305k | ret->comp = ((c == NULL) ? (OPENSSL_LH_COMPFUNC)strcmp : c); |
62 | 305k | ret->hash = ((h == NULL) ? (OPENSSL_LH_HASHFUNC)OPENSSL_LH_strhash : h); |
63 | 305k | ret->num_nodes = MIN_NODES / 2; |
64 | 305k | ret->num_alloc_nodes = MIN_NODES; |
65 | 305k | ret->pmax = MIN_NODES / 2; |
66 | 305k | ret->up_load = UP_LOAD; |
67 | 305k | ret->down_load = DOWN_LOAD; |
68 | 305k | return ret; |
69 | | |
70 | 0 | err: |
71 | 0 | OPENSSL_free(ret->b); |
72 | 0 | OPENSSL_free(ret); |
73 | 0 | return NULL; |
74 | 305k | } |
75 | | |
76 | | void OPENSSL_LH_free(OPENSSL_LHASH *lh) |
77 | 305k | { |
78 | 305k | if (lh == NULL) |
79 | 133 | return; |
80 | | |
81 | 305k | OPENSSL_LH_flush(lh); |
82 | 305k | OPENSSL_free(lh->b); |
83 | 305k | OPENSSL_free(lh); |
84 | 305k | } |
85 | | |
86 | | void OPENSSL_LH_flush(OPENSSL_LHASH *lh) |
87 | 307k | { |
88 | 307k | unsigned int i; |
89 | 307k | OPENSSL_LH_NODE *n, *nn; |
90 | | |
91 | 307k | if (lh == NULL) |
92 | 0 | return; |
93 | | |
94 | 3.39M | for (i = 0; i < lh->num_nodes; i++) { |
95 | 3.08M | n = lh->b[i]; |
96 | 3.94M | while (n != NULL) { |
97 | 864k | nn = n->next; |
98 | 864k | OPENSSL_free(n); |
99 | 864k | n = nn; |
100 | 864k | } |
101 | 3.08M | lh->b[i] = NULL; |
102 | 3.08M | } |
103 | | |
104 | 307k | lh->num_items = 0; |
105 | 307k | } |
106 | | |
107 | | void *OPENSSL_LH_insert(OPENSSL_LHASH *lh, void *data) |
108 | 3.17M | { |
109 | 3.17M | unsigned long hash; |
110 | 3.17M | OPENSSL_LH_NODE *nn, **rn; |
111 | 3.17M | void *ret; |
112 | | |
113 | 3.17M | lh->error = 0; |
114 | 3.17M | if ((lh->up_load <= (lh->num_items * LH_LOAD_MULT / lh->num_nodes)) && !expand(lh)) |
115 | 0 | return NULL; /* 'lh->error++' already done in 'expand' */ |
116 | | |
117 | 3.17M | rn = getrn(lh, data, &hash); |
118 | | |
119 | 3.17M | if (*rn == NULL) { |
120 | 2.61M | if ((nn = OPENSSL_malloc(sizeof(*nn))) == NULL) { |
121 | 0 | lh->error++; |
122 | 0 | return NULL; |
123 | 0 | } |
124 | 2.61M | nn->data = data; |
125 | 2.61M | nn->next = NULL; |
126 | 2.61M | nn->hash = hash; |
127 | 2.61M | *rn = nn; |
128 | 2.61M | ret = NULL; |
129 | 2.61M | lh->num_items++; |
130 | 2.61M | } else { /* replace same key */ |
131 | 563k | ret = (*rn)->data; |
132 | 563k | (*rn)->data = data; |
133 | 563k | } |
134 | 3.17M | return ret; |
135 | 3.17M | } |
136 | | |
137 | | void *OPENSSL_LH_delete(OPENSSL_LHASH *lh, const void *data) |
138 | 1.74M | { |
139 | 1.74M | unsigned long hash; |
140 | 1.74M | OPENSSL_LH_NODE *nn, **rn; |
141 | 1.74M | void *ret; |
142 | | |
143 | 1.74M | lh->error = 0; |
144 | 1.74M | rn = getrn(lh, data, &hash); |
145 | | |
146 | 1.74M | if (*rn == NULL) { |
147 | 144 | return NULL; |
148 | 1.74M | } else { |
149 | 1.74M | nn = *rn; |
150 | 1.74M | *rn = nn->next; |
151 | 1.74M | ret = nn->data; |
152 | 1.74M | OPENSSL_free(nn); |
153 | 1.74M | } |
154 | | |
155 | 1.74M | lh->num_items--; |
156 | 1.74M | if ((lh->num_nodes > MIN_NODES) && |
157 | 1.74M | (lh->down_load >= (lh->num_items * LH_LOAD_MULT / lh->num_nodes))) |
158 | 378k | contract(lh); |
159 | | |
160 | 1.74M | return ret; |
161 | 1.74M | } |
162 | | |
163 | | void *OPENSSL_LH_retrieve(OPENSSL_LHASH *lh, const void *data) |
164 | 311M | { |
165 | 311M | unsigned long hash; |
166 | 311M | OPENSSL_LH_NODE **rn; |
167 | | |
168 | 311M | if (lh->error != 0) |
169 | 0 | lh->error = 0; |
170 | | |
171 | 311M | rn = getrn(lh, data, &hash); |
172 | | |
173 | 311M | return *rn == NULL ? NULL : (*rn)->data; |
174 | 311M | } |
175 | | |
176 | | static void doall_util_fn(OPENSSL_LHASH *lh, int use_arg, |
177 | | OPENSSL_LH_DOALL_FUNC func, |
178 | | OPENSSL_LH_DOALL_FUNCARG func_arg, void *arg) |
179 | 10.7M | { |
180 | 10.7M | int i; |
181 | 10.7M | OPENSSL_LH_NODE *a, *n; |
182 | | |
183 | 10.7M | if (lh == NULL) |
184 | 0 | return; |
185 | | |
186 | | /* |
187 | | * reverse the order so we search from 'top to bottom' We were having |
188 | | * memory leaks otherwise |
189 | | */ |
190 | 1.07G | for (i = lh->num_nodes - 1; i >= 0; i--) { |
191 | 1.06G | a = lh->b[i]; |
192 | 3.16G | while (a != NULL) { |
193 | 2.10G | n = a->next; |
194 | 2.10G | if (use_arg) |
195 | 2.10G | func_arg(a->data, arg); |
196 | 226k | else |
197 | 226k | func(a->data); |
198 | 2.10G | a = n; |
199 | 2.10G | } |
200 | 1.06G | } |
201 | 10.7M | } |
202 | | |
203 | | void OPENSSL_LH_doall(OPENSSL_LHASH *lh, OPENSSL_LH_DOALL_FUNC func) |
204 | 14.0k | { |
205 | 14.0k | doall_util_fn(lh, 0, func, (OPENSSL_LH_DOALL_FUNCARG)0, NULL); |
206 | 14.0k | } |
207 | | |
208 | | void OPENSSL_LH_doall_arg(OPENSSL_LHASH *lh, OPENSSL_LH_DOALL_FUNCARG func, void *arg) |
209 | 8.13M | { |
210 | 8.13M | doall_util_fn(lh, 1, (OPENSSL_LH_DOALL_FUNC)0, func, arg); |
211 | 8.13M | } |
212 | | |
213 | | static int expand(OPENSSL_LHASH *lh) |
214 | 1.00M | { |
215 | 1.00M | OPENSSL_LH_NODE **n, **n1, **n2, *np; |
216 | 1.00M | unsigned int p, pmax, nni, j; |
217 | 1.00M | unsigned long hash; |
218 | | |
219 | 1.00M | nni = lh->num_alloc_nodes; |
220 | 1.00M | p = lh->p; |
221 | 1.00M | pmax = lh->pmax; |
222 | 1.00M | if (p + 1 >= pmax) { |
223 | 17.3k | j = nni * 2; |
224 | 17.3k | n = OPENSSL_realloc(lh->b, sizeof(OPENSSL_LH_NODE *) * j); |
225 | 17.3k | if (n == NULL) { |
226 | 0 | lh->error++; |
227 | 0 | return 0; |
228 | 0 | } |
229 | 17.3k | lh->b = n; |
230 | 17.3k | memset(n + nni, 0, sizeof(*n) * (j - nni)); |
231 | 17.3k | lh->pmax = nni; |
232 | 17.3k | lh->num_alloc_nodes = j; |
233 | 17.3k | lh->p = 0; |
234 | 986k | } else { |
235 | 986k | lh->p++; |
236 | 986k | } |
237 | | |
238 | 1.00M | lh->num_nodes++; |
239 | 1.00M | n1 = &(lh->b[p]); |
240 | 1.00M | n2 = &(lh->b[p + pmax]); |
241 | 1.00M | *n2 = NULL; |
242 | | |
243 | 3.98M | for (np = *n1; np != NULL;) { |
244 | 2.98M | hash = np->hash; |
245 | 2.98M | if ((hash % nni) != p) { /* move it */ |
246 | 1.30M | *n1 = (*n1)->next; |
247 | 1.30M | np->next = *n2; |
248 | 1.30M | *n2 = np; |
249 | 1.30M | } else |
250 | 1.67M | n1 = &((*n1)->next); |
251 | 2.98M | np = *n1; |
252 | 2.98M | } |
253 | | |
254 | 1.00M | return 1; |
255 | 1.00M | } |
256 | | |
257 | | static void contract(OPENSSL_LHASH *lh) |
258 | 378k | { |
259 | 378k | OPENSSL_LH_NODE **n, *n1, *np; |
260 | | |
261 | 378k | np = lh->b[lh->p + lh->pmax - 1]; |
262 | 378k | lh->b[lh->p + lh->pmax - 1] = NULL; /* 24/07-92 - eay - weird but :-( */ |
263 | 378k | if (lh->p == 0) { |
264 | 6.31k | n = OPENSSL_realloc(lh->b, |
265 | 6.31k | (unsigned int)(sizeof(OPENSSL_LH_NODE *) * lh->pmax)); |
266 | 6.31k | if (n == NULL) { |
267 | | /* fputs("realloc error in lhash",stderr); */ |
268 | 0 | lh->error++; |
269 | 6.31k | } else { |
270 | 6.31k | lh->b = n; |
271 | 6.31k | } |
272 | 6.31k | lh->num_alloc_nodes /= 2; |
273 | 6.31k | lh->pmax /= 2; |
274 | 6.31k | lh->p = lh->pmax - 1; |
275 | 6.31k | } else |
276 | 372k | lh->p--; |
277 | | |
278 | 378k | lh->num_nodes--; |
279 | | |
280 | 378k | n1 = lh->b[(int)lh->p]; |
281 | 378k | if (n1 == NULL) |
282 | 105k | lh->b[(int)lh->p] = np; |
283 | 273k | else { |
284 | 458k | while (n1->next != NULL) |
285 | 185k | n1 = n1->next; |
286 | 273k | n1->next = np; |
287 | 273k | } |
288 | 378k | } |
289 | | |
290 | | static OPENSSL_LH_NODE **getrn(OPENSSL_LHASH *lh, |
291 | | const void *data, unsigned long *rhash) |
292 | 298M | { |
293 | 298M | OPENSSL_LH_NODE **ret, *n1; |
294 | 298M | unsigned long hash, nn; |
295 | 298M | OPENSSL_LH_COMPFUNC cf; |
296 | | |
297 | 298M | hash = (*(lh->hash)) (data); |
298 | 298M | *rhash = hash; |
299 | | |
300 | 298M | nn = hash % lh->pmax; |
301 | 298M | if (nn < lh->p) |
302 | 192M | nn = hash % lh->num_alloc_nodes; |
303 | | |
304 | 298M | cf = lh->comp; |
305 | 298M | ret = &(lh->b[(int)nn]); |
306 | 1.03G | for (n1 = *ret; n1 != NULL; n1 = n1->next) { |
307 | 964M | if (n1->hash != hash) { |
308 | 739M | ret = &(n1->next); |
309 | 739M | continue; |
310 | 739M | } |
311 | 224M | if (cf(n1->data, data) == 0) |
312 | 224M | break; |
313 | 127k | ret = &(n1->next); |
314 | 127k | } |
315 | 298M | return ret; |
316 | 298M | } |
317 | | |
318 | | /* |
319 | | * The following hash seems to work very well on normal text strings no |
320 | | * collisions on /usr/dict/words and it distributes on %2^n quite well, not |
321 | | * as good as MD5, but still good. |
322 | | */ |
323 | | unsigned long OPENSSL_LH_strhash(const char *c) |
324 | 28.3M | { |
325 | 28.3M | unsigned long ret = 0; |
326 | 28.3M | long n; |
327 | 28.3M | unsigned long v; |
328 | 28.3M | int r; |
329 | | |
330 | 28.3M | if ((c == NULL) || (*c == '\0')) |
331 | 11.9M | return ret; |
332 | | |
333 | 16.3M | n = 0x100; |
334 | 614M | while (*c) { |
335 | 598M | v = n | (*c); |
336 | 598M | n += 0x100; |
337 | 598M | r = (int)((v >> 2) ^ v) & 0x0f; |
338 | | /* cast to uint64_t to avoid 32 bit shift of 32 bit value */ |
339 | 598M | ret = (ret << r) | (unsigned long)((uint64_t)ret >> (32 - r)); |
340 | 598M | ret &= 0xFFFFFFFFL; |
341 | 598M | ret ^= v * v; |
342 | 598M | c++; |
343 | 598M | } |
344 | 16.3M | return (ret >> 16) ^ ret; |
345 | 28.3M | } |
346 | | |
347 | | /* |
348 | | * Case insensitive string hashing. |
349 | | * |
350 | | * The lower/upper case bit is masked out (forcing all letters to be capitals). |
351 | | * The major side effect on non-alpha characters is mapping the symbols and |
352 | | * digits into the control character range (which should be harmless). |
353 | | * The duplication (with respect to the hash value) of printable characters |
354 | | * are that '`', '{', '|', '}' and '~' map to '@', '[', '\', ']' and '^' |
355 | | * respectively (which seems tolerable). |
356 | | * |
357 | | * For EBCDIC, the alpha mapping is to lower case, most symbols go to control |
358 | | * characters. The only duplication is '0' mapping to '^', which is better |
359 | | * than for ASCII. |
360 | | */ |
361 | | unsigned long ossl_lh_strcasehash(const char *c) |
362 | 273M | { |
363 | 273M | unsigned long ret = 0; |
364 | 273M | long n; |
365 | 273M | unsigned long v; |
366 | 273M | int r; |
367 | | #if defined(CHARSET_EBCDIC) && !defined(CHARSET_EBCDIC_TEST) |
368 | | const long int case_adjust = ~0x40; |
369 | | #else |
370 | 273M | const long int case_adjust = ~0x20; |
371 | 273M | #endif |
372 | | |
373 | 273M | if (c == NULL || *c == '\0') |
374 | 495 | return ret; |
375 | | |
376 | 1.67G | for (n = 0x100; *c != '\0'; n += 0x100) { |
377 | 1.40G | v = n | (case_adjust & *c); |
378 | 1.40G | r = (int)((v >> 2) ^ v) & 0x0f; |
379 | | /* cast to uint64_t to avoid 32 bit shift of 32 bit value */ |
380 | 1.40G | ret = (ret << r) | (unsigned long)((uint64_t)ret >> (32 - r)); |
381 | 1.40G | ret &= 0xFFFFFFFFL; |
382 | 1.40G | ret ^= v * v; |
383 | 1.40G | c++; |
384 | 1.40G | } |
385 | 273M | return (ret >> 16) ^ ret; |
386 | 273M | } |
387 | | |
388 | | unsigned long OPENSSL_LH_num_items(const OPENSSL_LHASH *lh) |
389 | 2.92M | { |
390 | 2.92M | return lh ? lh->num_items : 0; |
391 | 2.92M | } |
392 | | |
393 | | unsigned long OPENSSL_LH_get_down_load(const OPENSSL_LHASH *lh) |
394 | 81.4k | { |
395 | 81.4k | return lh->down_load; |
396 | 81.4k | } |
397 | | |
398 | | void OPENSSL_LH_set_down_load(OPENSSL_LHASH *lh, unsigned long down_load) |
399 | 222k | { |
400 | 222k | lh->down_load = down_load; |
401 | 222k | } |
402 | | |
403 | | int OPENSSL_LH_error(OPENSSL_LHASH *lh) |
404 | 1.34M | { |
405 | 1.34M | return lh->error; |
406 | 1.34M | } |