xref: /freebsd/sys/netpfil/ipfw/dn_heap.c (revision 1456f0f9681bbd7fdae7b683553f6c7491508c4e)
1 /*-
2  * Copyright (c) 1998-2002,2010 Luigi Rizzo, Universita` di Pisa
3  * All rights reserved
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 /*
28  * Binary heap and hash tables, used in dummynet
29  *
30  * $FreeBSD$
31  */
32 
33 #include <sys/cdefs.h>
34 #include <sys/param.h>
35 #ifdef _KERNEL
36 __FBSDID("$FreeBSD$");
37 #include <sys/systm.h>
38 #include <sys/malloc.h>
39 #include <sys/kernel.h>
40 #include <netpfil/ipfw/dn_heap.h>
41 #ifndef log
42 #define log(x, arg...)
43 #endif
44 
45 #else /* !_KERNEL */
46 
47 #include <stdio.h>
48 #include <dn_test.h>
49 #include <strings.h>
50 #include <stdlib.h>
51 
52 #include  "dn_heap.h"
53 #define log(x, arg...)	fprintf(stderr, ## arg)
54 #define panic(x...)	fprintf(stderr, ## x), exit(1)
55 #define MALLOC_DEFINE(a, b, c)	volatile int __dummy__ ## a __attribute__((__unused__))
56 static void *my_malloc(int s) {	return malloc(s); }
57 static void my_free(void *p) {	free(p); }
58 #define malloc(s, t, w)	my_malloc(s)
59 #define free(p, t)	my_free(p)
60 #endif /* !_KERNEL */
61 
62 static MALLOC_DEFINE(M_DN_HEAP, "dummynet", "dummynet heap");
63 
64 /*
65  * Heap management functions.
66  *
67  * In the heap, first node is element 0. Children of i are 2i+1 and 2i+2.
68  * Some macros help finding parent/children so we can optimize them.
69  *
70  * heap_init() is called to expand the heap when needed.
71  * Increment size in blocks of 16 entries.
72  * Returns 1 on error, 0 on success
73  */
74 #define HEAP_FATHER(x) ( ( (x) - 1 ) / 2 )
75 #define HEAP_LEFT(x) ( (x)+(x) + 1 )
76 #define	HEAP_SWAP(a, b, buffer) { buffer = a ; a = b ; b = buffer ; }
77 #define HEAP_INCREMENT	15
78 
79 static int
80 heap_resize(struct dn_heap *h, unsigned int new_size)
81 {
82 	struct dn_heap_entry *p;
83 
84 	if ((unsigned int)h->size >= new_size )	/* have enough room */
85 		return 0;
86 #if 1  /* round to the next power of 2 */
87 	new_size |= new_size >> 1;
88 	new_size |= new_size >> 2;
89 	new_size |= new_size >> 4;
90 	new_size |= new_size >> 8;
91 	new_size |= new_size >> 16;
92 #else
93 	new_size = (new_size + HEAP_INCREMENT ) & ~HEAP_INCREMENT;
94 #endif
95 	p = malloc(new_size * sizeof(*p), M_DN_HEAP, M_NOWAIT);
96 	if (p == NULL) {
97 		printf("--- %s, resize %d failed\n", __func__, new_size );
98 		return 1; /* error */
99 	}
100 	if (h->size > 0) {
101 		bcopy(h->p, p, h->size * sizeof(*p) );
102 		free(h->p, M_DN_HEAP);
103 	}
104 	h->p = p;
105 	h->size = new_size;
106 	return 0;
107 }
108 
109 int
110 heap_init(struct dn_heap *h, int size, int ofs)
111 {
112 	if (heap_resize(h, size))
113 		return 1;
114 	h->elements = 0;
115 	h->ofs = ofs;
116 	return 0;
117 }
118 
119 /*
120  * Insert element in heap. Normally, p != NULL, we insert p in
121  * a new position and bubble up. If p == NULL, then the element is
122  * already in place, and key is the position where to start the
123  * bubble-up.
124  * Returns 1 on failure (cannot allocate new heap entry)
125  *
126  * If ofs > 0 the position (index, int) of the element in the heap is
127  * also stored in the element itself at the given offset in bytes.
128  */
129 #define SET_OFFSET(h, i) do {					\
130 	if (h->ofs > 0)						\
131 	    *((int32_t *)((char *)(h->p[i].object) + h->ofs)) = i;	\
132 	} while (0)
133 /*
134  * RESET_OFFSET is used for sanity checks. It sets ofs
135  * to an invalid value.
136  */
137 #define RESET_OFFSET(h, i) do {					\
138 	if (h->ofs > 0)						\
139 	    *((int32_t *)((char *)(h->p[i].object) + h->ofs)) = -16;	\
140 	} while (0)
141 
142 int
143 heap_insert(struct dn_heap *h, uint64_t key1, void *p)
144 {
145 	int son = h->elements;
146 
147 	//log("%s key %llu p %p\n", __FUNCTION__, key1, p);
148 	if (p == NULL) { /* data already there, set starting point */
149 		son = key1;
150 	} else { /* insert new element at the end, possibly resize */
151 		son = h->elements;
152 		if (son == h->size) /* need resize... */
153 			// XXX expand by 16 or so
154 			if (heap_resize(h, h->elements+16) )
155 				return 1; /* failure... */
156 		h->p[son].object = p;
157 		h->p[son].key = key1;
158 		h->elements++;
159 	}
160 	/* make sure that son >= father along the path */
161 	while (son > 0) {
162 		int father = HEAP_FATHER(son);
163 		struct dn_heap_entry tmp;
164 
165 		if (DN_KEY_LT( h->p[father].key, h->p[son].key ) )
166 			break; /* found right position */
167 		/* son smaller than father, swap and repeat */
168 		HEAP_SWAP(h->p[son], h->p[father], tmp);
169 		SET_OFFSET(h, son);
170 		son = father;
171 	}
172 	SET_OFFSET(h, son);
173 	return 0;
174 }
175 
176 /*
177  * remove top element from heap, or obj if obj != NULL
178  */
179 void
180 heap_extract(struct dn_heap *h, void *obj)
181 {
182 	int child, father, max = h->elements - 1;
183 
184 	if (max < 0) {
185 		printf("--- %s: empty heap 0x%p\n", __FUNCTION__, h);
186 		return;
187 	}
188 	if (obj == NULL)
189 		father = 0; /* default: move up smallest child */
190 	else { /* extract specific element, index is at offset */
191 		if (h->ofs <= 0)
192 			panic("%s: extract from middle not set on %p\n",
193 				__FUNCTION__, h);
194 		father = *((int *)((char *)obj + h->ofs));
195 		if (father < 0 || father >= h->elements) {
196 			panic("%s: father %d out of bound 0..%d\n",
197 				__FUNCTION__, father, h->elements);
198 		}
199 	}
200 	/*
201 	 * below, father is the index of the empty element, which
202 	 * we replace at each step with the smallest child until we
203 	 * reach the bottom level.
204 	 */
205 	// XXX why removing RESET_OFFSET increases runtime by 10% ?
206 	RESET_OFFSET(h, father);
207 	while ( (child = HEAP_LEFT(father)) <= max ) {
208 		if (child != max &&
209 		    DN_KEY_LT(h->p[child+1].key, h->p[child].key) )
210 			child++; /* take right child, otherwise left */
211 		h->p[father] = h->p[child];
212 		SET_OFFSET(h, father);
213 		father = child;
214 	}
215 	h->elements--;
216 	if (father != max) {
217 		/*
218 		 * Fill hole with last entry and bubble up,
219 		 * reusing the insert code
220 		 */
221 		h->p[father] = h->p[max];
222 		heap_insert(h, father, NULL);
223 	}
224 }
225 
226 #if 0
227 /*
228  * change object position and update references
229  * XXX this one is never used!
230  */
231 static void
232 heap_move(struct dn_heap *h, uint64_t new_key, void *object)
233 {
234 	int temp, i, max = h->elements-1;
235 	struct dn_heap_entry *p, buf;
236 
237 	if (h->ofs <= 0)
238 		panic("cannot move items on this heap");
239 	p = h->p;	/* shortcut */
240 
241 	i = *((int *)((char *)object + h->ofs));
242 	if (DN_KEY_LT(new_key, p[i].key) ) { /* must move up */
243 		p[i].key = new_key;
244 		for (; i>0 &&
245 		    DN_KEY_LT(new_key, p[(temp = HEAP_FATHER(i))].key);
246 		    i = temp ) { /* bubble up */
247 			HEAP_SWAP(p[i], p[temp], buf);
248 			SET_OFFSET(h, i);
249 		}
250 	} else {		/* must move down */
251 		p[i].key = new_key;
252 		while ( (temp = HEAP_LEFT(i)) <= max ) {
253 			/* found left child */
254 			if (temp != max &&
255 			    DN_KEY_LT(p[temp+1].key, p[temp].key))
256 				temp++; /* select child with min key */
257 			if (DN_KEY_LT(>p[temp].key, new_key)) {
258 				/* go down */
259 				HEAP_SWAP(p[i], p[temp], buf);
260 				SET_OFFSET(h, i);
261 			} else
262 				break;
263 			i = temp;
264 		}
265 	}
266 	SET_OFFSET(h, i);
267 }
268 #endif /* heap_move, unused */
269 
270 /*
271  * heapify() will reorganize data inside an array to maintain the
272  * heap property. It is needed when we delete a bunch of entries.
273  */
274 static void
275 heapify(struct dn_heap *h)
276 {
277 	int i;
278 
279 	for (i = 0; i < h->elements; i++ )
280 		heap_insert(h, i , NULL);
281 }
282 
283 int
284 heap_scan(struct dn_heap *h, int (*fn)(void *, uintptr_t),
285 	uintptr_t arg)
286 {
287 	int i, ret, found;
288 
289 	for (i = found = 0 ; i < h->elements ;) {
290 		ret = fn(h->p[i].object, arg);
291 		if (ret & HEAP_SCAN_DEL) {
292 			h->elements-- ;
293 			h->p[i] = h->p[h->elements] ;
294 			found++ ;
295 		} else
296 			i++ ;
297 		if (ret & HEAP_SCAN_END)
298 			break;
299 	}
300 	if (found)
301 		heapify(h);
302 	return found;
303 }
304 
305 /*
306  * cleanup the heap and free data structure
307  */
308 void
309 heap_free(struct dn_heap *h)
310 {
311 	if (h->size >0 )
312 		free(h->p, M_DN_HEAP);
313 	bzero(h, sizeof(*h) );
314 }
315 
316 /*
317  * hash table support.
318  */
319 
320 struct dn_ht {
321         int buckets;            /* how many buckets, really buckets - 1*/
322         int entries;            /* how many entries */
323         int ofs;	        /* offset of link field */
324         uint32_t (*hash)(uintptr_t, int, void *arg);
325         int (*match)(void *_el, uintptr_t key, int, void *);
326         void *(*newh)(uintptr_t, int, void *);
327         void **ht;              /* bucket heads */
328 };
329 /*
330  * Initialize, allocating bucket pointers inline.
331  * Recycle previous record if possible.
332  * If the 'newh' function is not supplied, we assume that the
333  * key passed to ht_find is the same object to be stored in.
334  */
335 struct dn_ht *
336 dn_ht_init(struct dn_ht *ht, int buckets, int ofs,
337         uint32_t (*h)(uintptr_t, int, void *),
338         int (*match)(void *, uintptr_t, int, void *),
339 	void *(*newh)(uintptr_t, int, void *))
340 {
341 	int l;
342 
343 	/*
344 	 * Notes about rounding bucket size to a power of two.
345 	 * Given the original bucket size, we compute the nearest lower and
346 	 * higher power of two, minus 1  (respectively b_min and b_max) because
347 	 * this value will be used to do an AND with the index returned
348 	 * by hash function.
349 	 * To choice between these two values, the original bucket size is
350 	 * compared with b_min. If the original size is greater than 4/3 b_min,
351 	 * we round the bucket size to b_max, else to b_min.
352 	 * This ratio try to round to the nearest power of two, advantaging
353 	 * the greater size if the different between two power is relatively
354 	 * big.
355 	 * Rounding the bucket size to a power of two avoid the use of
356 	 * module when calculating the correct bucket.
357 	 * The ht->buckets variable store the bucket size - 1 to simply
358 	 * do an AND between the index returned by hash function and ht->bucket
359 	 * instead of a module.
360 	 */
361 	int b_min; /* min buckets */
362 	int b_max; /* max buckets */
363 	int b_ori; /* original buckets */
364 
365 	if (h == NULL || match == NULL) {
366 		printf("--- missing hash or match function");
367 		return NULL;
368 	}
369 	if (buckets < 1 || buckets > 65536)
370 		return NULL;
371 
372 	b_ori = buckets;
373 	/* calculate next power of 2, - 1*/
374 	buckets |= buckets >> 1;
375 	buckets |= buckets >> 2;
376 	buckets |= buckets >> 4;
377 	buckets |= buckets >> 8;
378 	buckets |= buckets >> 16;
379 
380 	b_max = buckets; /* Next power */
381 	b_min = buckets >> 1; /* Previous power */
382 
383 	/* Calculate the 'nearest' bucket size */
384 	if (b_min * 4000 / 3000 < b_ori)
385 		buckets = b_max;
386 	else
387 		buckets = b_min;
388 
389 	if (ht) {	/* see if we can reuse */
390 		if (buckets <= ht->buckets) {
391 			ht->buckets = buckets;
392 		} else {
393 			/* free pointers if not allocated inline */
394 			if (ht->ht != (void *)(ht + 1))
395 				free(ht->ht, M_DN_HEAP);
396 			free(ht, M_DN_HEAP);
397 			ht = NULL;
398 		}
399 	}
400 	if (ht == NULL) {
401 		/* Allocate buckets + 1 entries because buckets is use to
402 		 * do the AND with the index returned by hash function
403 		 */
404 		l = sizeof(*ht) + (buckets + 1) * sizeof(void **);
405 		ht = malloc(l, M_DN_HEAP, M_NOWAIT | M_ZERO);
406 	}
407 	if (ht) {
408 		ht->ht = (void **)(ht + 1);
409 		ht->buckets = buckets;
410 		ht->ofs = ofs;
411 		ht->hash = h;
412 		ht->match = match;
413 		ht->newh = newh;
414 	}
415 	return ht;
416 }
417 
418 /* dummy callback for dn_ht_free to unlink all */
419 static int
420 do_del(void *obj, void *arg)
421 {
422 	(void)obj;
423 	(void)arg;
424 	return DNHT_SCAN_DEL;
425 }
426 
427 void
428 dn_ht_free(struct dn_ht *ht, int flags)
429 {
430 	if (ht == NULL)
431 		return;
432 	if (flags & DNHT_REMOVE) {
433 		(void)dn_ht_scan(ht, do_del, NULL);
434 	} else {
435 		if (ht->ht && ht->ht != (void *)(ht + 1))
436 			free(ht->ht, M_DN_HEAP);
437 		free(ht, M_DN_HEAP);
438 	}
439 }
440 
441 int
442 dn_ht_entries(struct dn_ht *ht)
443 {
444 	return ht ? ht->entries : 0;
445 }
446 
447 /* lookup and optionally create or delete element */
448 void *
449 dn_ht_find(struct dn_ht *ht, uintptr_t key, int flags, void *arg)
450 {
451 	int i;
452 	void **pp, *p;
453 
454 	if (ht == NULL)	/* easy on an empty hash */
455 		return NULL;
456 	i = (ht->buckets == 1) ? 0 :
457 		(ht->hash(key, flags, arg) & ht->buckets);
458 
459 	for (pp = &ht->ht[i]; (p = *pp); pp = (void **)((char *)p + ht->ofs)) {
460 		if (flags & DNHT_MATCH_PTR) {
461 			if (key == (uintptr_t)p)
462 				break;
463 		} else if (ht->match(p, key, flags, arg)) /* found match */
464 			break;
465 	}
466 	if (p) {
467 		if (flags & DNHT_REMOVE) {
468 			/* link in the next element */
469 			*pp = *(void **)((char *)p + ht->ofs);
470 			*(void **)((char *)p + ht->ofs) = NULL;
471 			ht->entries--;
472 		}
473 	} else if (flags & DNHT_INSERT) {
474 		// printf("%s before calling new, bucket %d ofs %d\n",
475 		//	__FUNCTION__, i, ht->ofs);
476 		p = ht->newh ? ht->newh(key, flags, arg) : (void *)key;
477 		// printf("%s newh returns %p\n", __FUNCTION__, p);
478 		if (p) {
479 			ht->entries++;
480 			*(void **)((char *)p + ht->ofs) = ht->ht[i];
481 			ht->ht[i] = p;
482 		}
483 	}
484 	return p;
485 }
486 
487 /*
488  * do a scan with the option to delete the object. Extract next before
489  * running the callback because the element may be destroyed there.
490  */
491 int
492 dn_ht_scan(struct dn_ht *ht, int (*fn)(void *, void *), void *arg)
493 {
494 	int i, ret, found = 0;
495 	void **curp, *cur, *next;
496 
497 	if (ht == NULL || fn == NULL)
498 		return 0;
499 	for (i = 0; i <= ht->buckets; i++) {
500 		curp = &ht->ht[i];
501 		while ( (cur = *curp) != NULL) {
502 			next = *(void **)((char *)cur + ht->ofs);
503 			ret = fn(cur, arg);
504 			if (ret & DNHT_SCAN_DEL) {
505 				found++;
506 				ht->entries--;
507 				*curp = next;
508 			} else {
509 				curp = (void **)((char *)cur + ht->ofs);
510 			}
511 			if (ret & DNHT_SCAN_END)
512 				return found;
513 		}
514 	}
515 	return found;
516 }
517 
518 /*
519  * Similar to dn_ht_scan(), except that the scan is performed only
520  * in the bucket 'bucket'. The function returns a correct bucket number if
521  * the original is invalid.
522  * If the callback returns DNHT_SCAN_END, the function move the ht->ht[i]
523  * pointer to the last entry processed. Moreover, the bucket number passed
524  * by caller is decremented, because usually the caller increment it.
525  */
526 int
527 dn_ht_scan_bucket(struct dn_ht *ht, int *bucket, int (*fn)(void *, void *),
528 		 void *arg)
529 {
530 	int i, ret, found = 0;
531 	void **curp, *cur, *next;
532 
533 	if (ht == NULL || fn == NULL)
534 		return 0;
535 	if (*bucket > ht->buckets)
536 		*bucket = 0;
537 	i = *bucket;
538 
539 	curp = &ht->ht[i];
540 	while ( (cur = *curp) != NULL) {
541 		next = *(void **)((char *)cur + ht->ofs);
542 		ret = fn(cur, arg);
543 		if (ret & DNHT_SCAN_DEL) {
544 			found++;
545 			ht->entries--;
546 			*curp = next;
547 		} else {
548 			curp = (void **)((char *)cur + ht->ofs);
549 		}
550 		if (ret & DNHT_SCAN_END)
551 			return found;
552 	}
553 	return found;
554 }
555