]> git.proxmox.com Git - mirror_frr.git/blob - lib/hash.c
Merge pull request #12798 from donaldsharp/rib_match_multicast
[mirror_frr.git] / lib / hash.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Hash routine.
3 * Copyright (C) 1998 Kunihiro Ishiguro
4 */
5
6 #include <zebra.h>
7 #include <math.h>
8
9 #include "hash.h"
10 #include "memory.h"
11 #include "linklist.h"
12 #include "termtable.h"
13 #include "vty.h"
14 #include "command.h"
15 #include "libfrr.h"
16 #include "frr_pthread.h"
17 #include "libfrr_trace.h"
18
19 DEFINE_MTYPE_STATIC(LIB, HASH, "Hash");
20 DEFINE_MTYPE_STATIC(LIB, HASH_BUCKET, "Hash Bucket");
21 DEFINE_MTYPE_STATIC(LIB, HASH_INDEX, "Hash Index");
22
23 static pthread_mutex_t _hashes_mtx = PTHREAD_MUTEX_INITIALIZER;
24 static struct list *_hashes;
25
26 struct hash *hash_create_size(unsigned int size,
27 unsigned int (*hash_key)(const void *),
28 bool (*hash_cmp)(const void *, const void *),
29 const char *name)
30 {
31 struct hash *hash;
32
33 assert((size & (size - 1)) == 0);
34 hash = XCALLOC(MTYPE_HASH, sizeof(struct hash));
35 hash->index =
36 XCALLOC(MTYPE_HASH_INDEX, sizeof(struct hash_bucket *) * size);
37 hash->size = size;
38 hash->hash_key = hash_key;
39 hash->hash_cmp = hash_cmp;
40 hash->count = 0;
41 hash->name = name ? XSTRDUP(MTYPE_HASH, name) : NULL;
42 hash->stats.empty = hash->size;
43
44 frr_with_mutex (&_hashes_mtx) {
45 if (!_hashes)
46 _hashes = list_new();
47
48 listnode_add(_hashes, hash);
49 }
50
51 return hash;
52 }
53
54 struct hash *hash_create(unsigned int (*hash_key)(const void *),
55 bool (*hash_cmp)(const void *, const void *),
56 const char *name)
57 {
58 return hash_create_size(HASH_INITIAL_SIZE, hash_key, hash_cmp, name);
59 }
60
61 void *hash_alloc_intern(void *arg)
62 {
63 return arg;
64 }
65
66 /*
67 * ssq = ssq + (new^2 - old^2)
68 * = ssq + ((new + old) * (new - old))
69 */
70 #define hash_update_ssq(hz, old, new) \
71 do { \
72 int _adjust = (new + old) * (new - old); \
73 if (_adjust < 0) \
74 atomic_fetch_sub_explicit(&hz->stats.ssq, -_adjust, \
75 memory_order_relaxed); \
76 else \
77 atomic_fetch_add_explicit(&hz->stats.ssq, _adjust, \
78 memory_order_relaxed); \
79 } while (0)
80
81 /* Expand hash if the chain length exceeds the threshold. */
82 static void hash_expand(struct hash *hash)
83 {
84 unsigned int i, new_size;
85 struct hash_bucket *hb, *hbnext, **new_index;
86
87 new_size = hash->size * 2;
88
89 if (hash->max_size && new_size > hash->max_size)
90 return;
91
92 new_index = XCALLOC(MTYPE_HASH_INDEX,
93 sizeof(struct hash_bucket *) * new_size);
94
95 hash->stats.empty = new_size;
96
97 for (i = 0; i < hash->size; i++)
98 for (hb = hash->index[i]; hb; hb = hbnext) {
99 unsigned int h = hb->key & (new_size - 1);
100
101 hbnext = hb->next;
102 hb->next = new_index[h];
103
104 int oldlen = hb->next ? hb->next->len : 0;
105 int newlen = oldlen + 1;
106
107 if (newlen == 1)
108 hash->stats.empty--;
109 else
110 hb->next->len = 0;
111
112 hb->len = newlen;
113
114 hash_update_ssq(hash, oldlen, newlen);
115
116 new_index[h] = hb;
117 }
118
119 /* Switch to new table */
120 XFREE(MTYPE_HASH_INDEX, hash->index);
121 hash->size = new_size;
122 hash->index = new_index;
123 }
124
125 void *hash_get(struct hash *hash, void *data, void *(*alloc_func)(void *))
126 {
127 frrtrace(2, frr_libfrr, hash_get, hash, data);
128
129 unsigned int key;
130 unsigned int index;
131 void *newdata;
132 struct hash_bucket *bucket;
133
134 if (!alloc_func && !hash->count)
135 return NULL;
136
137 key = (*hash->hash_key)(data);
138 index = key & (hash->size - 1);
139
140 for (bucket = hash->index[index]; bucket != NULL;
141 bucket = bucket->next) {
142 if (bucket->key == key && (*hash->hash_cmp)(bucket->data, data))
143 return bucket->data;
144 }
145
146 if (alloc_func) {
147 newdata = (*alloc_func)(data);
148 if (newdata == NULL)
149 return NULL;
150
151 if (HASH_THRESHOLD(hash->count + 1, hash->size)) {
152 hash_expand(hash);
153 index = key & (hash->size - 1);
154 }
155
156 bucket = XCALLOC(MTYPE_HASH_BUCKET, sizeof(struct hash_bucket));
157 bucket->data = newdata;
158 bucket->key = key;
159 bucket->next = hash->index[index];
160 hash->index[index] = bucket;
161 hash->count++;
162
163 frrtrace(3, frr_libfrr, hash_insert, hash, data, key);
164
165 int oldlen = bucket->next ? bucket->next->len : 0;
166 int newlen = oldlen + 1;
167
168 if (newlen == 1)
169 hash->stats.empty--;
170 else
171 bucket->next->len = 0;
172
173 bucket->len = newlen;
174
175 hash_update_ssq(hash, oldlen, newlen);
176
177 return bucket->data;
178 }
179 return NULL;
180 }
181
182 void *hash_lookup(struct hash *hash, void *data)
183 {
184 return hash_get(hash, data, NULL);
185 }
186
187 unsigned int string_hash_make(const char *str)
188 {
189 unsigned int hash = 0;
190
191 while (*str)
192 hash = (hash * 33) ^ (unsigned int)*str++;
193
194 return hash;
195 }
196
197 void *hash_release(struct hash *hash, void *data)
198 {
199 void *ret = NULL;
200 unsigned int key;
201 unsigned int index;
202 struct hash_bucket *bucket;
203 struct hash_bucket *pp;
204
205 key = (*hash->hash_key)(data);
206 index = key & (hash->size - 1);
207
208 for (bucket = pp = hash->index[index]; bucket; bucket = bucket->next) {
209 if (bucket->key == key
210 && (*hash->hash_cmp)(bucket->data, data)) {
211 int oldlen = hash->index[index]->len;
212 int newlen = oldlen - 1;
213
214 if (bucket == pp)
215 hash->index[index] = bucket->next;
216 else
217 pp->next = bucket->next;
218
219 if (hash->index[index])
220 hash->index[index]->len = newlen;
221 else
222 hash->stats.empty++;
223
224 hash_update_ssq(hash, oldlen, newlen);
225
226 ret = bucket->data;
227 XFREE(MTYPE_HASH_BUCKET, bucket);
228 hash->count--;
229 break;
230 }
231 pp = bucket;
232 }
233
234 frrtrace(3, frr_libfrr, hash_release, hash, data, ret);
235
236 return ret;
237 }
238
239 void hash_iterate(struct hash *hash, void (*func)(struct hash_bucket *, void *),
240 void *arg)
241 {
242 unsigned int i;
243 struct hash_bucket *hb;
244 struct hash_bucket *hbnext;
245
246 for (i = 0; i < hash->size; i++)
247 for (hb = hash->index[i]; hb; hb = hbnext) {
248 /* get pointer to next hash bucket here, in case (*func)
249 * decides to delete hb by calling hash_release
250 */
251 hbnext = hb->next;
252 (*func)(hb, arg);
253 }
254 }
255
256 void hash_walk(struct hash *hash, int (*func)(struct hash_bucket *, void *),
257 void *arg)
258 {
259 unsigned int i;
260 struct hash_bucket *hb;
261 struct hash_bucket *hbnext;
262 int ret = HASHWALK_CONTINUE;
263
264 for (i = 0; i < hash->size; i++) {
265 for (hb = hash->index[i]; hb; hb = hbnext) {
266 /* get pointer to next hash bucket here, in case (*func)
267 * decides to delete hb by calling hash_release
268 */
269 hbnext = hb->next;
270 ret = (*func)(hb, arg);
271 if (ret == HASHWALK_ABORT)
272 return;
273 }
274 }
275 }
276
277 void hash_clean(struct hash *hash, void (*free_func)(void *))
278 {
279 unsigned int i;
280 struct hash_bucket *hb;
281 struct hash_bucket *next;
282
283 for (i = 0; i < hash->size; i++) {
284 for (hb = hash->index[i]; hb; hb = next) {
285 next = hb->next;
286
287 if (free_func)
288 (*free_func)(hb->data);
289
290 XFREE(MTYPE_HASH_BUCKET, hb);
291 hash->count--;
292 }
293 hash->index[i] = NULL;
294 }
295
296 hash->stats.ssq = 0;
297 hash->stats.empty = hash->size;
298 }
299
300 static void hash_to_list_iter(struct hash_bucket *hb, void *arg)
301 {
302 struct list *list = arg;
303
304 listnode_add(list, hb->data);
305 }
306
307 struct list *hash_to_list(struct hash *hash)
308 {
309 struct list *list = list_new();
310
311 hash_iterate(hash, hash_to_list_iter, list);
312 return list;
313 }
314
315 void hash_free(struct hash *hash)
316 {
317 frr_with_mutex (&_hashes_mtx) {
318 if (_hashes) {
319 listnode_delete(_hashes, hash);
320 if (_hashes->count == 0) {
321 list_delete(&_hashes);
322 }
323 }
324 }
325
326 XFREE(MTYPE_HASH, hash->name);
327
328 XFREE(MTYPE_HASH_INDEX, hash->index);
329 XFREE(MTYPE_HASH, hash);
330 }
331
332
333 /* CLI commands ------------------------------------------------------------ */
334
335 DEFUN_NOSH(show_hash_stats,
336 show_hash_stats_cmd,
337 "show debugging hashtable [statistics]",
338 SHOW_STR
339 DEBUG_STR
340 "Statistics about hash tables\n"
341 "Statistics about hash tables\n")
342 {
343 struct hash *h;
344 struct listnode *ln;
345 struct ttable *tt = ttable_new(&ttable_styles[TTSTYLE_BLANK]);
346
347 ttable_add_row(tt, "Hash table|Buckets|Entries|Empty|LF|SD|FLF|SD");
348 tt->style.cell.lpad = 2;
349 tt->style.cell.rpad = 1;
350 tt->style.corner = '+';
351 ttable_restyle(tt);
352 ttable_rowseps(tt, 0, BOTTOM, true, '-');
353
354 /* Summary statistics calculated are:
355 *
356 * - Load factor: This is the number of elements in the table divided
357 * by the number of buckets. Since this hash table implementation
358 * uses chaining, this value can be greater than 1.
359 * This number provides information on how 'full' the table is, but
360 * does not provide information on how evenly distributed the
361 * elements are.
362 * Notably, a load factor >= 1 does not imply that every bucket has
363 * an element; with a pathological hash function, all elements could
364 * be in a single bucket.
365 *
366 * - Full load factor: this is the number of elements in the table
367 * divided by the number of buckets that have some elements in them.
368 *
369 * - Std. Dev.: This is the standard deviation calculated from the
370 * relevant load factor. If the load factor is the mean of number of
371 * elements per bucket, the standard deviation measures how much any
372 * particular bucket is likely to deviate from the mean.
373 * As a rule of thumb this number should be less than 2, and ideally
374 * <= 1 for optimal performance. A number larger than 3 generally
375 * indicates a poor hash function.
376 */
377
378 double lf; // load factor
379 double flf; // full load factor
380 double var; // overall variance
381 double fvar; // full variance
382 double stdv; // overall stddev
383 double fstdv; // full stddev
384
385 long double x2; // h->count ^ 2
386 long double ldc; // (long double) h->count
387 long double full; // h->size - h->stats.empty
388 long double ssq; // ssq casted to long double
389
390 pthread_mutex_lock(&_hashes_mtx);
391 if (!_hashes) {
392 pthread_mutex_unlock(&_hashes_mtx);
393 ttable_del(tt);
394 vty_out(vty, "No hash tables in use.\n");
395 return CMD_SUCCESS;
396 }
397
398 for (ALL_LIST_ELEMENTS_RO(_hashes, ln, h)) {
399 if (!h->name)
400 continue;
401
402 ssq = (long double)h->stats.ssq;
403 x2 = h->count * h->count;
404 ldc = (long double)h->count;
405 full = h->size - h->stats.empty;
406 lf = h->count / (double)h->size;
407 flf = full ? h->count / (double)(full) : 0;
408 var = ldc ? (1.0 / ldc) * (ssq - x2 / ldc) : 0;
409 fvar = full ? (1.0 / full) * (ssq - x2 / full) : 0;
410 var = (var < .0001) ? 0 : var;
411 fvar = (fvar < .0001) ? 0 : fvar;
412 stdv = sqrt(var);
413 fstdv = sqrt(fvar);
414
415 ttable_add_row(tt, "%s|%d|%ld|%.0f%%|%.2lf|%.2lf|%.2lf|%.2lf",
416 h->name, h->size, h->count,
417 (h->stats.empty / (double)h->size) * 100, lf,
418 stdv, flf, fstdv);
419 }
420 pthread_mutex_unlock(&_hashes_mtx);
421
422 /* display header */
423 char header[] = "Showing hash table statistics for ";
424 char underln[sizeof(header) + strlen(frr_protonameinst)];
425 memset(underln, '-', sizeof(underln));
426 underln[sizeof(underln) - 1] = '\0';
427 vty_out(vty, "%s%s\n", header, frr_protonameinst);
428 vty_out(vty, "%s\n", underln);
429
430 vty_out(vty, "# allocated: %d\n", _hashes->count);
431 vty_out(vty, "# named: %d\n\n", tt->nrows - 1);
432
433 if (tt->nrows > 1) {
434 ttable_colseps(tt, 0, RIGHT, true, '|');
435 char *table = ttable_dump(tt, "\n");
436 vty_out(vty, "%s\n", table);
437 XFREE(MTYPE_TMP, table);
438 } else
439 vty_out(vty, "No named hash tables to display.\n");
440
441 ttable_del(tt);
442
443 return CMD_SUCCESS;
444 }
445
446 void hash_cmd_init(void)
447 {
448 install_element(ENABLE_NODE, &show_hash_stats_cmd);
449 }