xref: /netbsd-src/external/bsd/openldap/dist/servers/slapd/zn_malloc.c (revision 796c32c94f6e154afc9de0f63da35c91bb739b45)
1 /*	$NetBSD: zn_malloc.c,v 1.1.1.5 2017/02/09 01:46:58 christos Exp $	*/
2 
3 /* zn_malloc.c - zone-based malloc routines */
4 /* $OpenLDAP$*/
5 /* This work is part of OpenLDAP Software <http://www.openldap.org/>.
6  *
7  * Copyright 2003-2016 The OpenLDAP Foundation.
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted only as authorized by the OpenLDAP
12  * Public License.
13  *
14  * A copy of this license is available in the file LICENSE in the
15  * top-level directory of the distribution or, alternatively, at
16  * <http://www.OpenLDAP.org/license.html>.
17  */
18 /* Portions Copyright 2004 IBM Corporation
19  * All rights reserved.
20  * Redistribution and use in source and binary forms, with or without
21  * modification, are permitted only as authorized by the OpenLDAP
22  * Public License.
23  */
24 /* ACKNOWLEDGEMENTS
25  * This work originally developed by Jong-Hyuk Choi for inclusion in
26  * OpenLDAP Software.
27  */
28 
29 #include <sys/cdefs.h>
30 __RCSID("$NetBSD: zn_malloc.c,v 1.1.1.5 2017/02/09 01:46:58 christos Exp $");
31 
32 #include "portable.h"
33 
34 #include <stdio.h>
35 #include <ac/string.h>
36 #include <sys/types.h>
37 #include <fcntl.h>
38 
39 #include "slap.h"
40 
41 #ifdef SLAP_ZONE_ALLOC
42 
43 #include <sys/mman.h>
44 
45 static int slap_zone_cmp(const void *v1, const void *v2);
46 void * slap_replenish_zopool(void *ctx);
47 
48 static void
49 slap_zo_release(void *data)
50 {
51 	struct zone_object *zo = (struct zone_object *)data;
52 	ch_free( zo );
53 }
54 
55 void
56 slap_zn_mem_destroy(
57 	void *ctx
58 )
59 {
60 	struct zone_heap *zh = ctx;
61 	int pad = 2*sizeof(int)-1, pad_shift;
62 	int order_start = -1, i, j;
63 	struct zone_object *zo;
64 
65 	pad_shift = pad - 1;
66 	do {
67 		order_start++;
68 	} while (pad_shift >>= 1);
69 
70 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
71 	for (i = 0; i < zh->zh_zoneorder - order_start + 1; i++) {
72 		zo = LDAP_LIST_FIRST(&zh->zh_free[i]);
73 		while (zo) {
74 			struct zone_object *zo_tmp = zo;
75 			zo = LDAP_LIST_NEXT(zo, zo_link);
76 			LDAP_LIST_REMOVE(zo_tmp, zo_link);
77 			LDAP_LIST_INSERT_HEAD(&zh->zh_zopool, zo_tmp, zo_link);
78 		}
79 	}
80 	ch_free(zh->zh_free);
81 
82 	for (i = 0; i < zh->zh_numzones; i++) {
83 		for (j = 0; j < zh->zh_zoneorder - order_start + 1; j++) {
84 			ch_free(zh->zh_maps[i][j]);
85 		}
86 		ch_free(zh->zh_maps[i]);
87 		munmap(zh->zh_zones[i], zh->zh_zonesize);
88 		ldap_pvt_thread_rdwr_destroy(&zh->zh_znlock[i]);
89 	}
90 	ch_free(zh->zh_maps);
91 	ch_free(zh->zh_zones);
92 	ch_free(zh->zh_seqno);
93 	ch_free(zh->zh_znlock);
94 
95 	avl_free(zh->zh_zonetree, slap_zo_release);
96 
97 	zo = LDAP_LIST_FIRST(&zh->zh_zopool);
98 	while (zo) {
99 		struct zone_object *zo_tmp = zo;
100 		zo = LDAP_LIST_NEXT(zo, zo_link);
101 		if (!zo_tmp->zo_blockhead) {
102 			LDAP_LIST_REMOVE(zo_tmp, zo_link);
103 		}
104 	}
105 	zo = LDAP_LIST_FIRST(&zh->zh_zopool);
106 	while (zo) {
107 		struct zone_object *zo_tmp = zo;
108 		zo = LDAP_LIST_NEXT(zo, zo_link);
109 		ch_free(zo_tmp);
110 	}
111 	ldap_pvt_thread_mutex_unlock(&zh->zh_mutex);
112 	ldap_pvt_thread_rdwr_destroy(&zh->zh_lock);
113 	ldap_pvt_thread_mutex_destroy(&zh->zh_mutex);
114 	ch_free(zh);
115 }
116 
117 void *
118 slap_zn_mem_create(
119 	ber_len_t initsize,
120 	ber_len_t maxsize,
121 	ber_len_t deltasize,
122 	ber_len_t zonesize
123 )
124 {
125 	struct zone_heap *zh = NULL;
126 	ber_len_t zpad;
127 	int pad = 2*sizeof(int)-1, pad_shift;
128 	int size_shift;
129 	int order = -1, order_start = -1, order_end = -1;
130 	int i, j;
131 	struct zone_object *zo;
132 
133 	Debug(LDAP_DEBUG_NONE,
134 		"--> slap_zn_mem_create: initsize=%d, maxsize=%d\n",
135 		initsize, maxsize, 0);
136 	Debug(LDAP_DEBUG_NONE,
137 		"++> slap_zn_mem_create: deltasize=%d, zonesize=%d\n",
138 		deltasize, zonesize, 0);
139 
140 	zh = (struct zone_heap *)ch_calloc(1, sizeof(struct zone_heap));
141 
142 	zh->zh_fd = open("/dev/zero", O_RDWR);
143 
144 	if ( zonesize ) {
145 		zh->zh_zonesize = zonesize;
146 	} else {
147 		zh->zh_zonesize = SLAP_ZONE_SIZE;
148 	}
149 
150 	zpad = zh->zh_zonesize - 1;
151 	zh->zh_numzones = ((initsize + zpad) & ~zpad) / zh->zh_zonesize;
152 
153 	if ( maxsize && maxsize >= initsize ) {
154 		zh->zh_maxzones = ((maxsize + zpad) & ~zpad) / zh->zh_zonesize;
155 	} else {
156 		zh->zh_maxzones = ((initsize + zpad) & ~zpad) / zh->zh_zonesize;
157 	}
158 
159 	if ( deltasize ) {
160 		zh->zh_deltazones = ((deltasize + zpad) & ~zpad) / zh->zh_zonesize;
161 	} else {
162 		zh->zh_deltazones = ((SLAP_ZONE_DELTA+zpad) & ~zpad) / zh->zh_zonesize;
163 	}
164 
165 	size_shift = zh->zh_zonesize - 1;
166 	do {
167 		order_end++;
168 	} while (size_shift >>= 1);
169 
170 	pad_shift = pad - 1;
171 	do {
172 		order_start++;
173 	} while (pad_shift >>= 1);
174 
175 	order = order_end - order_start + 1;
176 
177 	zh->zh_zones = (void **)ch_malloc(zh->zh_maxzones * sizeof(void*));
178 	zh->zh_znlock = (ldap_pvt_thread_rdwr_t *)ch_malloc(
179 						zh->zh_maxzones * sizeof(ldap_pvt_thread_rdwr_t *));
180 	zh->zh_maps = (unsigned char ***)ch_malloc(
181 					zh->zh_maxzones * sizeof(unsigned char**));
182 
183 	zh->zh_zoneorder = order_end;
184 	zh->zh_free = (struct zh_freelist *)
185 					ch_malloc(order * sizeof(struct zh_freelist));
186 	zh->zh_seqno = (unsigned long *)ch_calloc(zh->zh_maxzones,
187 											sizeof(unsigned long));
188 	for (i = 0; i < order; i++) {
189 		LDAP_LIST_INIT(&zh->zh_free[i]);
190 	}
191 	LDAP_LIST_INIT(&zh->zh_zopool);
192 
193 	for (i = 0; i < zh->zh_numzones; i++) {
194 		zh->zh_zones[i] = mmap(0, zh->zh_zonesize, PROT_READ | PROT_WRITE,
195 							MAP_PRIVATE, zh->zh_fd, 0);
196 		zh->zh_maps[i] = (unsigned char **)
197 					ch_malloc(order * sizeof(unsigned char *));
198 		for (j = 0; j < order; j++) {
199 			int shiftamt = order_start + 1 + j;
200 			int nummaps = zh->zh_zonesize >> shiftamt;
201 			assert(nummaps);
202 			nummaps >>= 3;
203 			if (!nummaps) nummaps = 1;
204 			zh->zh_maps[i][j] = (unsigned char *)ch_malloc(nummaps);
205 			memset(zh->zh_maps[i][j], 0, nummaps);
206 		}
207 
208 		if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
209 			slap_replenish_zopool(zh);
210 		}
211 		zo = LDAP_LIST_FIRST(&zh->zh_zopool);
212 		LDAP_LIST_REMOVE(zo, zo_link);
213 		zo->zo_ptr = zh->zh_zones[i];
214 		zo->zo_idx = i;
215 		LDAP_LIST_INSERT_HEAD(&zh->zh_free[order-1], zo, zo_link);
216 
217 		if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
218 			slap_replenish_zopool(zh);
219 		}
220 		zo = LDAP_LIST_FIRST(&zh->zh_zopool);
221 		LDAP_LIST_REMOVE(zo, zo_link);
222 		zo->zo_ptr = zh->zh_zones[i];
223 		zo->zo_siz = zh->zh_zonesize;
224 		zo->zo_idx = i;
225 		avl_insert(&zh->zh_zonetree, zo, slap_zone_cmp, avl_dup_error);
226 		ldap_pvt_thread_rdwr_init(&zh->zh_znlock[i]);
227 	}
228 
229 	LDAP_STAILQ_INIT(&zh->zh_latency_history_queue);
230 	ldap_pvt_thread_mutex_init(&zh->zh_mutex);
231 	ldap_pvt_thread_rdwr_init(&zh->zh_lock);
232 
233 	return zh;
234 }
235 
236 void *
237 slap_zn_malloc(
238     ber_len_t	size,
239 	void *ctx
240 )
241 {
242 	struct zone_heap *zh = ctx;
243 	ber_len_t size_shift;
244 	int pad = 2*sizeof(int)-1, pad_shift;
245 	int order = -1, order_start = -1;
246 	struct zone_object *zo, *zo_new, *zo_left, *zo_right;
247 	ber_len_t *ptr, *new;
248 	int idx;
249 	unsigned long diff;
250 	int i, j, k;
251 
252 	Debug(LDAP_DEBUG_NONE,
253 		"--> slap_zn_malloc: size=%d\n", size, 0, 0);
254 
255 	if (!zh) return ber_memalloc_x(size, NULL);
256 
257 	/* round up to doubleword boundary */
258 	size += 2*sizeof(ber_len_t) + pad;
259 	size &= ~pad;
260 
261 	size_shift = size - 1;
262 	do {
263 		order++;
264 	} while (size_shift >>= 1);
265 
266 	pad_shift = pad - 1;
267 	do {
268 		order_start++;
269 	} while (pad_shift >>= 1);
270 
271 retry:
272 
273 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
274 	for (i = order; i <= zh->zh_zoneorder &&
275 			LDAP_LIST_EMPTY(&zh->zh_free[i-order_start]); i++);
276 
277 	if (i == order) {
278 		zo_new = LDAP_LIST_FIRST(&zh->zh_free[i-order_start]);
279 		LDAP_LIST_REMOVE(zo_new, zo_link);
280 		ptr = zo_new->zo_ptr;
281 		idx = zo_new->zo_idx;
282 		diff = (unsigned long)((char*)ptr -
283 				(char*)zh->zh_zones[idx]) >> (order + 1);
284 		zh->zh_maps[idx][order-order_start][diff>>3] |= (1 << (diff & 0x7));
285 		*ptr++ = zh->zh_seqno[idx];
286 		*ptr++ = size - 2*sizeof(ber_len_t);
287 		zo_new->zo_ptr = NULL;
288 		zo_new->zo_idx = -1;
289 		LDAP_LIST_INSERT_HEAD(&zh->zh_zopool, zo_new, zo_link);
290 		ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
291 		Debug(LDAP_DEBUG_NONE, "slap_zn_malloc: returning 0x%x, 0x%x\n",
292 				ptr, (int)ptr>>(zh->zh_zoneorder+1), 0);
293 		return((void*)ptr);
294 	} else if (i <= zh->zh_zoneorder) {
295 		for (j = i; j > order; j--) {
296 			zo_left = LDAP_LIST_FIRST(&zh->zh_free[j-order_start]);
297 			LDAP_LIST_REMOVE(zo_left, zo_link);
298 			if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
299 				slap_replenish_zopool(zh);
300 			}
301 			zo_right = LDAP_LIST_FIRST(&zh->zh_zopool);
302 			LDAP_LIST_REMOVE(zo_right, zo_link);
303 			zo_right->zo_ptr = zo_left->zo_ptr + (1 << j);
304 			zo_right->zo_idx = zo_left->zo_idx;
305 			Debug(LDAP_DEBUG_NONE,
306 				"slap_zn_malloc: split (left=0x%x, right=0x%x)\n",
307 				zo_left->zo_ptr, zo_right->zo_ptr, 0);
308 			if (j == order + 1) {
309 				ptr = zo_left->zo_ptr;
310 				diff = (unsigned long)((char*)ptr -
311 						(char*)zh->zh_zones[zo_left->zo_idx]) >> (order+1);
312 				zh->zh_maps[zo_left->zo_idx][order-order_start][diff>>3] |=
313 						(1 << (diff & 0x7));
314 				*ptr++ = zh->zh_seqno[zo_left->zo_idx];
315 				*ptr++ = size - 2*sizeof(ber_len_t);
316 				LDAP_LIST_INSERT_HEAD(
317 						&zh->zh_free[j-1-order_start], zo_right, zo_link);
318 				LDAP_LIST_INSERT_HEAD(&zh->zh_zopool, zo_left, zo_link);
319 				ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
320 				Debug(LDAP_DEBUG_NONE,
321 					"slap_zn_malloc: returning 0x%x, 0x%x\n",
322 					ptr, (int)ptr>>(zh->zh_zoneorder+1), 0);
323 				return((void*)ptr);
324 			} else {
325 				LDAP_LIST_INSERT_HEAD(
326 						&zh->zh_free[j-1-order_start], zo_right, zo_link);
327 				LDAP_LIST_INSERT_HEAD(
328 						&zh->zh_free[j-1-order_start], zo_left, zo_link);
329 			}
330 		}
331 		assert(0);
332 	} else {
333 
334 		if ( zh->zh_maxzones < zh->zh_numzones + zh->zh_deltazones ) {
335 			ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
336 			Debug( LDAP_DEBUG_TRACE,
337 				"zn_malloc %lu: ch_malloc\n",
338 				(long)size, 0, 0);
339 			Debug(LDAP_DEBUG_NONE,
340 				"slap_zn_malloc: returning 0x%x, 0x%x\n",
341 				ptr, (int)ptr>>(zh->zh_zoneorder+1), 0);
342 			return (void*)ch_malloc(size);
343 		}
344 
345 		for (i = zh->zh_numzones; i < zh->zh_numzones+zh->zh_deltazones; i++) {
346 			zh->zh_zones[i] = mmap(0, zh->zh_zonesize, PROT_READ | PROT_WRITE,
347 								MAP_PRIVATE, zh->zh_fd, 0);
348 			zh->zh_maps[i] = (unsigned char **)
349 						ch_malloc((zh->zh_zoneorder - order_start + 1) *
350 						sizeof(unsigned char *));
351 			for (j = 0; j < zh->zh_zoneorder-order_start+1; j++) {
352 				int shiftamt = order_start + 1 + j;
353 				int nummaps = zh->zh_zonesize >> shiftamt;
354 				assert(nummaps);
355 				nummaps >>= 3;
356 				if (!nummaps) nummaps = 1;
357 				zh->zh_maps[i][j] = (unsigned char *)ch_malloc(nummaps);
358 				memset(zh->zh_maps[i][j], 0, nummaps);
359 			}
360 
361 			if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
362 				slap_replenish_zopool(zh);
363 			}
364 			zo = LDAP_LIST_FIRST(&zh->zh_zopool);
365 			LDAP_LIST_REMOVE(zo, zo_link);
366 			zo->zo_ptr = zh->zh_zones[i];
367 			zo->zo_idx = i;
368 			LDAP_LIST_INSERT_HEAD(&zh->
369 						zh_free[zh->zh_zoneorder-order_start],zo,zo_link);
370 
371 			if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
372 				slap_replenish_zopool(zh);
373 			}
374 			zo = LDAP_LIST_FIRST(&zh->zh_zopool);
375 			LDAP_LIST_REMOVE(zo, zo_link);
376 			zo->zo_ptr = zh->zh_zones[i];
377 			zo->zo_siz = zh->zh_zonesize;
378 			zo->zo_idx = i;
379 			avl_insert(&zh->zh_zonetree, zo, slap_zone_cmp, avl_dup_error);
380 			ldap_pvt_thread_rdwr_init(&zh->zh_znlock[i]);
381 		}
382 		zh->zh_numzones += zh->zh_deltazones;
383 		ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
384 		goto retry;
385 	}
386 }
387 
388 void *
389 slap_zn_calloc( ber_len_t n, ber_len_t size, void *ctx )
390 {
391 	void *new;
392 
393 	new = slap_zn_malloc( n*size, ctx );
394 	if ( new ) {
395 		memset( new, 0, n*size );
396 	}
397 	return new;
398 }
399 
400 void *
401 slap_zn_realloc(void *ptr, ber_len_t size, void *ctx)
402 {
403 	struct zone_heap *zh = ctx;
404 	int pad = 2*sizeof(int)-1, pad_shift;
405 	int order_start = -1, order = -1;
406 	struct zone_object zoi, *zoo;
407 	ber_len_t *p = (ber_len_t *)ptr, *new;
408 	unsigned long diff;
409 	int i;
410 	void *newptr = NULL;
411 	struct zone_heap *zone = NULL;
412 
413 	Debug(LDAP_DEBUG_NONE,
414 		"--> slap_zn_realloc: ptr=0x%x, size=%d\n", ptr, size, 0);
415 
416 	if (ptr == NULL)
417 		return slap_zn_malloc(size, zh);
418 
419 	zoi.zo_ptr = p;
420 	zoi.zo_idx = -1;
421 
422 	if (zh) {
423 		ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
424 		zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
425 		ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
426 	}
427 
428 	/* Not our memory? */
429 	if (!zoo) {
430 		/* duplicate of realloc behavior, oh well */
431 		new = ber_memrealloc_x(ptr, size, NULL);
432 		if (new) {
433 			return new;
434 		}
435 		Debug(LDAP_DEBUG_ANY, "ch_realloc of %lu bytes failed\n",
436 				(long) size, 0, 0);
437 		assert(0);
438 		exit( EXIT_FAILURE );
439 	}
440 
441 	assert(zoo->zo_idx != -1);
442 
443 	zone = zh->zh_zones[zoo->zo_idx];
444 
445 	if (size == 0) {
446 		slap_zn_free(ptr, zh);
447 		return NULL;
448 	}
449 
450 	newptr = slap_zn_malloc(size, zh);
451 	if (size < p[-1]) {
452 		AC_MEMCPY(newptr, ptr, size);
453 	} else {
454 		AC_MEMCPY(newptr, ptr, p[-1]);
455 	}
456 	slap_zn_free(ptr, zh);
457 	return newptr;
458 }
459 
460 void
461 slap_zn_free(void *ptr, void *ctx)
462 {
463 	struct zone_heap *zh = ctx;
464 	int size, size_shift, order_size;
465 	int pad = 2*sizeof(int)-1, pad_shift;
466 	ber_len_t *p = (ber_len_t *)ptr, *tmpp;
467 	int order_start = -1, order = -1;
468 	struct zone_object zoi, *zoo, *zo;
469 	unsigned long diff;
470 	int i, k, inserted = 0, idx;
471 	struct zone_heap *zone = NULL;
472 
473 	zoi.zo_ptr = p;
474 	zoi.zo_idx = -1;
475 
476 	Debug(LDAP_DEBUG_NONE, "--> slap_zn_free: ptr=0x%x\n", ptr, 0, 0);
477 
478 	if (zh) {
479 		ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
480 		zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
481 		ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
482 	}
483 
484 	if (!zoo) {
485 		ber_memfree_x(ptr, NULL);
486 	} else {
487 		idx = zoo->zo_idx;
488 		assert(idx != -1);
489 		zone = zh->zh_zones[idx];
490 
491 		size = *(--p);
492 		size_shift = size + 2*sizeof(ber_len_t) - 1;
493 		do {
494 			order++;
495 		} while (size_shift >>= 1);
496 
497 		pad_shift = pad - 1;
498 		do {
499 			order_start++;
500 		} while (pad_shift >>= 1);
501 
502 		ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
503 		for (i = order, tmpp = p; i <= zh->zh_zoneorder; i++) {
504 			order_size = 1 << (i+1);
505 			diff = (unsigned long)((char*)tmpp - (char*)zone) >> (i+1);
506 			zh->zh_maps[idx][i-order_start][diff>>3] &= (~(1 << (diff & 0x7)));
507 			if (diff == ((diff>>1)<<1)) {
508 				if (!(zh->zh_maps[idx][i-order_start][(diff+1)>>3] &
509 						(1<<((diff+1)&0x7)))) {
510 					zo = LDAP_LIST_FIRST(&zh->zh_free[i-order_start]);
511 					while (zo) {
512 						if ((char*)zo->zo_ptr == (char*)tmpp) {
513 							LDAP_LIST_REMOVE( zo, zo_link );
514 						} else if ((char*)zo->zo_ptr ==
515 								(char*)tmpp + order_size) {
516 							LDAP_LIST_REMOVE(zo, zo_link);
517 							break;
518 						}
519 						zo = LDAP_LIST_NEXT(zo, zo_link);
520 					}
521 					if (zo) {
522 						if (i < zh->zh_zoneorder) {
523 							inserted = 1;
524 							zo->zo_ptr = tmpp;
525 							Debug(LDAP_DEBUG_NONE,
526 								"slap_zn_free: merging 0x%x\n",
527 								zo->zo_ptr, 0, 0);
528 							LDAP_LIST_INSERT_HEAD(&zh->zh_free[i-order_start+1],
529 									zo, zo_link);
530 						}
531 						continue;
532 					} else {
533 						if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
534 							slap_replenish_zopool(zh);
535 						}
536 						zo = LDAP_LIST_FIRST(&zh->zh_zopool);
537 						LDAP_LIST_REMOVE(zo, zo_link);
538 						zo->zo_ptr = tmpp;
539 						zo->zo_idx = idx;
540 						Debug(LDAP_DEBUG_NONE,
541 							"slap_zn_free: merging 0x%x\n",
542 							zo->zo_ptr, 0, 0);
543 						LDAP_LIST_INSERT_HEAD(&zh->zh_free[i-order_start],
544 								zo, zo_link);
545 						break;
546 
547 						Debug(LDAP_DEBUG_ANY, "slap_zn_free: "
548 							"free object not found while bit is clear.\n",
549 							0, 0, 0);
550 						assert(zo != NULL);
551 
552 					}
553 				} else {
554 					if (!inserted) {
555 						if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
556 							slap_replenish_zopool(zh);
557 						}
558 						zo = LDAP_LIST_FIRST(&zh->zh_zopool);
559 						LDAP_LIST_REMOVE(zo, zo_link);
560 						zo->zo_ptr = tmpp;
561 						zo->zo_idx = idx;
562 						Debug(LDAP_DEBUG_NONE,
563 							"slap_zn_free: merging 0x%x\n",
564 							zo->zo_ptr, 0, 0);
565 						LDAP_LIST_INSERT_HEAD(&zh->zh_free[i-order_start],
566 								zo, zo_link);
567 					}
568 					break;
569 				}
570 			} else {
571 				if (!(zh->zh_maps[idx][i-order_start][(diff-1)>>3] &
572 						(1<<((diff-1)&0x7)))) {
573 					zo = LDAP_LIST_FIRST(&zh->zh_free[i-order_start]);
574 					while (zo) {
575 						if ((char*)zo->zo_ptr == (char*)tmpp) {
576 							LDAP_LIST_REMOVE(zo, zo_link);
577 						} else if ((char*)tmpp == zo->zo_ptr + order_size) {
578 							LDAP_LIST_REMOVE(zo, zo_link);
579 							tmpp = zo->zo_ptr;
580 							break;
581 						}
582 						zo = LDAP_LIST_NEXT(zo, zo_link);
583 					}
584 					if (zo) {
585 						if (i < zh->zh_zoneorder) {
586 							inserted = 1;
587 							Debug(LDAP_DEBUG_NONE,
588 								"slap_zn_free: merging 0x%x\n",
589 								zo->zo_ptr, 0, 0);
590 							LDAP_LIST_INSERT_HEAD(&zh->zh_free[i-order_start+1],
591 									zo, zo_link);
592 							continue;
593 						}
594 					} else {
595 						if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
596 							slap_replenish_zopool(zh);
597 						}
598 						zo = LDAP_LIST_FIRST(&zh->zh_zopool);
599 						LDAP_LIST_REMOVE(zo, zo_link);
600 						zo->zo_ptr = tmpp;
601 						zo->zo_idx = idx;
602 						Debug(LDAP_DEBUG_NONE,
603 							"slap_zn_free: merging 0x%x\n",
604 							zo->zo_ptr, 0, 0);
605 						LDAP_LIST_INSERT_HEAD(&zh->zh_free[i-order_start],
606 								zo, zo_link);
607 						break;
608 
609 						Debug(LDAP_DEBUG_ANY, "slap_zn_free: "
610 							"free object not found while bit is clear.\n",
611 							0, 0, 0 );
612 						assert(zo != NULL);
613 
614 					}
615 				} else {
616 					if ( !inserted ) {
617 						if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
618 							slap_replenish_zopool(zh);
619 						}
620 						zo = LDAP_LIST_FIRST(&zh->zh_zopool);
621 						LDAP_LIST_REMOVE(zo, zo_link);
622 						zo->zo_ptr = tmpp;
623 						zo->zo_idx = idx;
624 						Debug(LDAP_DEBUG_NONE,
625 							"slap_zn_free: merging 0x%x\n",
626 							zo->zo_ptr, 0, 0);
627 						LDAP_LIST_INSERT_HEAD(&zh->zh_free[i-order_start],
628 								zo, zo_link);
629 					}
630 					break;
631 				}
632 			}
633 		}
634 		ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
635 	}
636 }
637 
638 static int
639 slap_zone_cmp(const void *v1, const void *v2)
640 {
641 	const struct zone_object *zo1 = v1;
642 	const struct zone_object *zo2 = v2;
643 	char *ptr1;
644 	char *ptr2;
645 	ber_len_t zpad;
646 
647 	zpad = zo2->zo_siz - 1;
648 	ptr1 = (char*)(((unsigned long)zo1->zo_ptr + zpad) & ~zpad);
649 	ptr2 = (char*)zo2->zo_ptr + ((char*)ptr1 - (char*)zo1->zo_ptr);
650 	ptr2 = (char*)(((unsigned long)ptr2 + zpad) & ~zpad);
651 	return (int)((char*)ptr1 - (char*)ptr2);
652 }
653 
654 void *
655 slap_replenish_zopool(
656 	void *ctx
657 )
658 {
659 	struct zone_heap* zh = ctx;
660 	struct zone_object *zo_block;
661 	int i;
662 
663 	zo_block = (struct zone_object *)ch_malloc(
664 					SLAP_ZONE_ZOBLOCK * sizeof(struct zone_object));
665 
666 	if ( zo_block == NULL ) {
667 		return NULL;
668 	}
669 
670 	zo_block[0].zo_blockhead = 1;
671 	LDAP_LIST_INSERT_HEAD(&zh->zh_zopool, &zo_block[0], zo_link);
672 	for (i = 1; i < SLAP_ZONE_ZOBLOCK; i++) {
673 		zo_block[i].zo_blockhead = 0;
674 		LDAP_LIST_INSERT_HEAD(&zh->zh_zopool, &zo_block[i], zo_link );
675 	}
676 
677 	return zo_block;
678 }
679 
680 int
681 slap_zn_invalidate(
682 	void *ctx,
683 	void *ptr
684 )
685 {
686 	struct zone_heap* zh = ctx;
687 	struct zone_object zoi, *zoo;
688 	struct zone_heap *zone = NULL;
689 	int seqno = *((ber_len_t*)ptr - 2);
690 	int idx = -1, rc = 0;
691 	int pad = 2*sizeof(int)-1, pad_shift;
692 	int order_start = -1, i;
693 	struct zone_object *zo;
694 
695 	pad_shift = pad - 1;
696 	do {
697 		order_start++;
698 	} while (pad_shift >>= 1);
699 
700 	zoi.zo_ptr = ptr;
701 	zoi.zo_idx = -1;
702 
703 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
704 	zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
705 
706 	if (zoo) {
707 		idx = zoo->zo_idx;
708 		assert(idx != -1);
709 		madvise(zh->zh_zones[idx], zh->zh_zonesize, MADV_DONTNEED);
710 		for (i = 0; i < zh->zh_zoneorder - order_start + 1; i++) {
711 			int shiftamt = order_start + 1 + i;
712 			int nummaps = zh->zh_zonesize >> shiftamt;
713 			assert(nummaps);
714 			nummaps >>= 3;
715 			if (!nummaps) nummaps = 1;
716 			memset(zh->zh_maps[idx][i], 0, nummaps);
717 			zo = LDAP_LIST_FIRST(&zh->zh_free[i]);
718 			while (zo) {
719 				struct zone_object *zo_tmp = zo;
720 				zo = LDAP_LIST_NEXT(zo, zo_link);
721 				if (zo_tmp && zo_tmp->zo_idx == idx) {
722 					LDAP_LIST_REMOVE(zo_tmp, zo_link);
723 					LDAP_LIST_INSERT_HEAD(&zh->zh_zopool, zo_tmp, zo_link);
724 				}
725 			}
726 		}
727 		if (LDAP_LIST_EMPTY(&zh->zh_zopool)) {
728 			slap_replenish_zopool(zh);
729 		}
730 		zo = LDAP_LIST_FIRST(&zh->zh_zopool);
731 		LDAP_LIST_REMOVE(zo, zo_link);
732 		zo->zo_ptr = zh->zh_zones[idx];
733 		zo->zo_idx = idx;
734 		LDAP_LIST_INSERT_HEAD(&zh->zh_free[zh->zh_zoneorder-order_start],
735 								zo, zo_link);
736 		zh->zh_seqno[idx]++;
737 	} else {
738 		Debug(LDAP_DEBUG_NONE, "zone not found for (ctx=0x%x, ptr=0x%x) !\n",
739 				ctx, ptr, 0);
740 	}
741 
742 	ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
743 	Debug(LDAP_DEBUG_NONE, "zone %d invalidate\n", idx, 0, 0);
744 	return rc;
745 }
746 
747 int
748 slap_zn_validate(
749 	void *ctx,
750 	void *ptr,
751 	int seqno
752 )
753 {
754 	struct zone_heap* zh = ctx;
755 	struct zone_object zoi, *zoo;
756 	struct zone_heap *zone = NULL;
757 	int idx, rc = 0;
758 
759 	zoi.zo_ptr = ptr;
760 	zoi.zo_idx = -1;
761 
762 	zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
763 
764 	if (zoo) {
765 		idx = zoo->zo_idx;
766 		assert(idx != -1);
767 		assert(seqno <= zh->zh_seqno[idx]);
768 		rc = (seqno == zh->zh_seqno[idx]);
769 	}
770 
771 	return rc;
772 }
773 
774 int slap_zh_rlock(
775 	void *ctx
776 )
777 {
778 	struct zone_heap* zh = ctx;
779 	ldap_pvt_thread_rdwr_rlock(&zh->zh_lock);
780 }
781 
782 int slap_zh_runlock(
783 	void *ctx
784 )
785 {
786 	struct zone_heap* zh = ctx;
787 	ldap_pvt_thread_rdwr_runlock(&zh->zh_lock);
788 }
789 
790 int slap_zh_wlock(
791 	void *ctx
792 )
793 {
794 	struct zone_heap* zh = ctx;
795 	ldap_pvt_thread_rdwr_wlock(&zh->zh_lock);
796 }
797 
798 int slap_zh_wunlock(
799 	void *ctx
800 )
801 {
802 	struct zone_heap* zh = ctx;
803 	ldap_pvt_thread_rdwr_wunlock(&zh->zh_lock);
804 }
805 
806 int slap_zn_rlock(
807 	void *ctx,
808 	void *ptr
809 )
810 {
811 	struct zone_heap* zh = ctx;
812 	struct zone_object zoi, *zoo;
813 	struct zone_heap *zone = NULL;
814 	int idx;
815 
816 	zoi.zo_ptr = ptr;
817 	zoi.zo_idx = -1;
818 
819 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
820 	zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
821 	ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
822 
823 	if (zoo) {
824 		idx = zoo->zo_idx;
825 		assert(idx != -1);
826 		ldap_pvt_thread_rdwr_rlock(&zh->zh_znlock[idx]);
827 	}
828 }
829 
830 int slap_zn_runlock(
831 	void *ctx,
832 	void *ptr
833 )
834 {
835 	struct zone_heap* zh = ctx;
836 	struct zone_object zoi, *zoo;
837 	struct zone_heap *zone = NULL;
838 	int idx;
839 
840 	zoi.zo_ptr = ptr;
841 	zoi.zo_idx = -1;
842 
843 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
844 	zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
845 	ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
846 
847 	if (zoo) {
848 		idx = zoo->zo_idx;
849 		assert(idx != -1);
850 		ldap_pvt_thread_rdwr_runlock(&zh->zh_znlock[idx]);
851 	}
852 }
853 
854 int slap_zn_wlock(
855 	void *ctx,
856 	void *ptr
857 )
858 {
859 	struct zone_heap* zh = ctx;
860 	struct zone_object zoi, *zoo;
861 	struct zone_heap *zone = NULL;
862 	int idx;
863 
864 	zoi.zo_ptr = ptr;
865 	zoi.zo_idx = -1;
866 
867 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
868 	zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
869 	ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
870 
871 	if (zoo) {
872 		idx = zoo->zo_idx;
873 		assert(idx != -1);
874 		ldap_pvt_thread_rdwr_wlock(&zh->zh_znlock[idx]);
875 	}
876 }
877 
878 int slap_zn_wunlock(
879 	void *ctx,
880 	void *ptr
881 )
882 {
883 	struct zone_heap* zh = ctx;
884 	struct zone_object zoi, *zoo;
885 	struct zone_heap *zone = NULL;
886 	int idx;
887 
888 	zoi.zo_ptr = ptr;
889 	zoi.zo_idx = -1;
890 
891 	ldap_pvt_thread_mutex_lock( &zh->zh_mutex );
892 	zoo = avl_find(zh->zh_zonetree, &zoi, slap_zone_cmp);
893 	ldap_pvt_thread_mutex_unlock( &zh->zh_mutex );
894 
895 	if (zoo) {
896 		idx = zoo->zo_idx;
897 		assert(idx != -1);
898 		ldap_pvt_thread_rdwr_wunlock(&zh->zh_znlock[idx]);
899 	}
900 }
901 
902 #define T_SEC_IN_USEC 1000000
903 
904 static int
905 slap_timediff(struct timeval *tv_begin, struct timeval *tv_end)
906 {
907 	uint64_t t_begin, t_end, t_diff;
908 
909 	t_begin = T_SEC_IN_USEC * tv_begin->tv_sec + tv_begin->tv_usec;
910 	t_end  = T_SEC_IN_USEC * tv_end->tv_sec  + tv_end->tv_usec;
911 	t_diff  = t_end - t_begin;
912 
913 	if ( t_diff < 0 )
914 		t_diff = 0;
915 
916 	return (int)t_diff;
917 }
918 
919 void
920 slap_set_timing(struct timeval *tv_set)
921 {
922 	gettimeofday(tv_set, (struct timezone *)NULL);
923 }
924 
925 int
926 slap_measure_timing(struct timeval *tv_set, struct timeval *tv_measure)
927 {
928 	gettimeofday(tv_measure, (struct timezone *)NULL);
929 	return(slap_timediff(tv_set, tv_measure));
930 }
931 
932 #define EMA_WEIGHT 0.999000
933 #define SLAP_ZN_LATENCY_HISTORY_QLEN 500
934 int
935 slap_zn_latency_history(void* ctx, int ea_latency)
936 {
937 /* TODO: monitor /proc/stat (swap) as well */
938 	struct zone_heap* zh = ctx;
939 	double t_diff = 0.0;
940 
941 	zh->zh_ema_latency = (double)ea_latency * (1.0 - EMA_WEIGHT)
942 					+ zh->zh_ema_latency * EMA_WEIGHT;
943 	if (!zh->zh_swapping && zh->zh_ema_samples++ % 100 == 99) {
944 		struct zone_latency_history *zlh_entry;
945 		zlh_entry = ch_calloc(1, sizeof(struct zone_latency_history));
946 		zlh_entry->zlh_latency = zh->zh_ema_latency;
947 		LDAP_STAILQ_INSERT_TAIL(
948 				&zh->zh_latency_history_queue, zlh_entry, zlh_next);
949 		zh->zh_latency_history_qlen++;
950 		while (zh->zh_latency_history_qlen > SLAP_ZN_LATENCY_HISTORY_QLEN) {
951 			struct zone_latency_history *zlh;
952 			zlh = LDAP_STAILQ_FIRST(&zh->zh_latency_history_queue);
953 			LDAP_STAILQ_REMOVE_HEAD(
954 					&zh->zh_latency_history_queue, zlh_next);
955 			zh->zh_latency_history_qlen--;
956 			ch_free(zlh);
957 		}
958 		if (zh->zh_latency_history_qlen == SLAP_ZN_LATENCY_HISTORY_QLEN) {
959 			struct zone_latency_history *zlh_first, *zlh_last;
960 			zlh_first = LDAP_STAILQ_FIRST(&zh->zh_latency_history_queue);
961 			zlh_last = LDAP_STAILQ_LAST(&zh->zh_latency_history_queue,
962 						zone_latency_history, zlh_next);
963 			t_diff = zlh_last->zlh_latency - zlh_first->zlh_latency;
964 		}
965 		if (t_diff >= 2000) {
966 			zh->zh_latency_jump++;
967 		} else {
968 			zh->zh_latency_jump = 0;
969 		}
970 		if (zh->zh_latency_jump > 3) {
971 			zh->zh_latency_jump = 0;
972 			zh->zh_swapping = 1;
973 		}
974 	}
975 	return zh->zh_swapping;
976 }
977 #endif /* SLAP_ZONE_ALLOC */
978