2 Bacula® - The Network Backup Solution
4 Copyright (C) 2003-2008 Free Software Foundation Europe e.V.
6 The main author of Bacula is Kern Sibbald, with contributions from
7 many others, a complete list can be found in the file AUTHORS.
8 This program is Free Software; you can redistribute it and/or
9 modify it under the terms of version two of the GNU General Public
10 License as published by the Free Software Foundation and included
13 This program is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
23 Bacula® is a registered trademark of John Walker.
24 The licensor of Bacula is the Free Software Foundation Europe
25 (FSFE), Fiduciary Program, Sumatrastrasse 25, 8006 Zürich,
26 Switzerland, email:ftf@fsfeurope.org.
29 * Bacula hash table routines
31 * htable is a hash table of items (pointers). This code is
32 * adapted and enhanced from code I wrote in 1982 for a
33 * relocatable linker. At that time, the hash table size
34 * was fixed and a primary number, which essentially provides
35 * the randomness. In this program, the hash table can grow when
36 * it gets too full, so the table size here is a binary number. The
37 * hashing is provided using an idea from Tcl where the initial
38 * hash code is "randomized" using a simple calculation from
39 * a random number generator that multiplies by a big number
40 * (I multiply by a prime number, while Tcl did not)
41 * then shifts the result down and does the binary division
42 * by masking. Increasing the size of the hash table is simple.
43 * Just create a new larger table, walk the old table and
44 * re-hash insert each entry into the new table.
47 * Kern Sibbald, July MMIII
57 /* ===================================================================
63 * This subroutine gets a big buffer.
65 void htable::malloc_buf(int size)
69 hmem = (struct h_mem *)malloc(size);
72 hmem->next = this->mem;
74 hmem->mem = mem->first;
75 hmem->rem = (char *)hmem + size - hmem->mem;
76 Dmsg2(200, "malloc buf size=%d rem=%d\n", size, hmem->rem);
79 /* This routine frees the whole tree */
80 void htable::hash_free()
82 struct h_mem *hmem, *rel;
84 for (hmem=mem; hmem; ) {
93 char *htable::hash_malloc(int size)
97 int asize = BALIGN(size);
99 if (mem->rem < asize) {
101 if (total_size >= 1000000) {
115 return (char *)malloc(size);
123 * Create hash of key, stored in hash then
124 * create and return the pseudo random bucket index
126 void htable::hash_index(char *key)
129 for (char *p=key; *p; p++) {
130 hash += (hash << 3) + (uint32_t)*p;
132 /* Multiply by large prime number, take top bits, mask for remainder */
133 index = ((hash * 1103515249) >> rshift) & mask;
134 Dmsg2(100, "Leave hash_index hash=0x%x index=%d\n", hash, index);
138 * tsize is the estimated number of entries in the hash table
140 htable::htable(void *item, void *link, int tsize)
142 init(item, link, tsize);
145 void htable::init(void *item, void *link, int tsize)
153 for (pwr=0; tsize; pwr++) {
156 loffset = (char *)link - (char *)item;
157 mask = ~((~0)<<pwr); /* 3 bits => table size = 8 */
158 rshift = 30 - pwr; /* start using bits 28, 29, 30 */
159 num_items = 0; /* number of entries in table */
160 buckets = 1<<pwr; /* hash table size -- power of two */
161 max_items = buckets * 4; /* allow average 4 entries per chain */
162 table = (hlink **)malloc(buckets * sizeof(hlink *));
163 memset(table, 0, buckets * sizeof(hlink *));
170 malloc_buf(1000000); /* ***FIXME*** need variable or some estimate */
174 uint32_t htable::size()
180 * Take each hash link and walk down the chain of items
181 * that hash there counting them (i.e. the hits),
182 * then report that number.
183 * Obiously, the more hits in a chain, the more time
184 * it takes to reference them. Empty chains are not so
185 * hot either -- as it means unused or wasted space.
194 printf("\n\nNumItems=%d\nTotal buckets=%d\n", num_items, buckets);
195 printf("Hits/bucket: buckets\n");
196 for (i=0; i < MAX_COUNT; i++) {
199 for (i=0; i<(int)buckets; i++) {
203 p = (hlink *)(p->next);
213 for (i=0; i < MAX_COUNT; i++) {
214 printf("%2d: %d\n",i, hits[i]);
216 printf("buckets=%d num_items=%d max_items=%d\n", buckets, num_items, max_items);
217 printf("max hits in a bucket = %d\n", max);
219 printf("total bytes malloced = %d\n", total_size);
220 printf("total blocks malloced = %d\n", blocks);
224 void htable::grow_table()
226 Dmsg1(100, "Grow called old size = %d\n", buckets);
227 /* Setup a bigger table */
228 htable *big = (htable *)malloc(sizeof(htable));
229 big->loffset = loffset;
230 big->mask = mask<<1 | 1;
231 big->rshift = rshift - 1;
233 big->buckets = buckets * 2;
234 big->max_items = big->buckets * 4;
235 big->table = (hlink **)malloc(big->buckets * sizeof(hlink *));
236 memset(big->table, 0, big->buckets * sizeof(hlink *));
239 /* Insert all the items in the new hash table */
240 Dmsg1(100, "Before copy num_items=%d\n", num_items);
242 * We walk through the old smaller tree getting items,
243 * but since we are overwriting the colision links, we must
244 * explicitly save the item->next pointer and walk each
245 * colision chain ourselves. We do use next() for getting
246 * to the next bucket.
248 for (void *item=first(); item; ) {
249 void *ni = ((hlink *)((char *)item+loffset))->next; /* save link overwritten by insert */
250 Dmsg1(100, "Grow insert: %s\n", ((hlink *)((char *)item+loffset))->key);
251 big->insert(((hlink *)((char *)item+loffset))->key, item);
253 item = (void *)((char *)ni-loffset);
259 Dmsg1(100, "After copy new num_items=%d\n", big->num_items);
260 if (num_items != big->num_items) {
261 Dmsg0(000, "****** Big problems num_items mismatch ******\n");
264 memcpy(this, big, sizeof(htable)); /* move everything across */
266 Dmsg0(100, "Exit grow.\n");
269 bool htable::insert(char *key, void *item)
273 return false; /* already exists */
275 ASSERT(index < buckets);
276 Dmsg2(100, "Insert: hash=%p index=%d\n", hash, index);
277 hp = (hlink *)(((char *)item)+loffset);
278 Dmsg4(100, "Insert hp=%p index=%d item=%p offset=%u\n", hp,
279 index, item, loffset);
280 hp->next = table[index];
284 Dmsg3(100, "Insert hp->next=%p hp->hash=0x%x hp->key=%s\n",
285 hp->next, hp->hash, hp->key);
287 if (++num_items >= max_items) {
288 Dmsg2(100, "num_items=%d max_items=%d\n", num_items, max_items);
291 Dmsg3(100, "Leave insert index=%d num_items=%d key=%s\n", index, num_items, key);
295 void *htable::lookup(char *key)
298 for (hlink *hp=table[index]; hp; hp=(hlink *)hp->next) {
299 // Dmsg2(100, "hp=%p key=%s\n", hp, hp->key);
300 if (hash == hp->hash && strcmp(key, hp->key) == 0) {
301 Dmsg1(100, "lookup return %p\n", ((char *)hp)-loffset);
302 return ((char *)hp)-loffset;
310 Dmsg1(100, "Enter next: walkptr=%p\n", walkptr);
312 walkptr = (hlink *)(walkptr->next);
314 while (!walkptr && walk_index < buckets) {
315 walkptr = table[walk_index++];
317 Dmsg3(100, "new walkptr=%p next=%p inx=%d\n", walkptr,
318 walkptr->next, walk_index-1);
322 Dmsg2(100, "next: rtn %p walk_index=%d\n",
323 ((char *)walkptr)-loffset, walk_index);
324 return ((char *)walkptr)-loffset;
326 Dmsg0(100, "next: return NULL\n");
330 void *htable::first()
332 Dmsg0(100, "Enter first\n");
333 walkptr = table[0]; /* get first bucket */
334 walk_index = 1; /* Point to next index */
335 while (!walkptr && walk_index < buckets) {
336 walkptr = table[walk_index++]; /* go to next bucket */
338 Dmsg3(100, "first new walkptr=%p next=%p inx=%d\n", walkptr,
339 walkptr->next, walk_index-1);
343 Dmsg1(100, "Leave first walkptr=%p\n", walkptr);
344 return ((char *)walkptr)-loffset;
346 Dmsg0(100, "Leave first walkptr=NULL\n");
350 /* Destroy the table and its contents */
351 void htable::destroy()
368 Dmsg0(100, "Done destroy.\n");
380 #define NITEMS 5000000
386 MYJCR *save_jcr = NULL, *item;
390 jcrtbl = (htable *)malloc(sizeof(htable));
391 jcrtbl->init(jcr, &jcr->link, NITEMS);
393 Dmsg1(000, "Inserting %d items\n", NITEMS);
394 for (int i=0; i<NITEMS; i++) {
396 len = sprintf(mkey, "This is htable item %d", i) + 1;
398 jcr = (MYJCR *)jcrtbl->hash_malloc(sizeof(MYJCR));
399 jcr->key = (char *)jcrtbl->hash_malloc(len);
400 memcpy(jcr->key, mkey, len);
401 Dmsg2(100, "link=%p jcr=%p\n", jcr->link, jcr);
403 jcrtbl->insert(jcr->key, jcr);
408 if (!(item = (MYJCR *)jcrtbl->lookup(save_jcr->key))) {
409 printf("Bad news: %s not found.\n", save_jcr->key);
411 printf("Item 10's key is: %s\n", item->key);
415 printf("Walk the hash table:\n");
416 foreach_htable (jcr, jcrtbl) {
417 // printf("htable item = %s\n", jcr->key);
423 printf("Got %d items -- %s\n", count, count==NITEMS?"OK":"***ERROR***");
424 printf("Calling destroy\n");
428 printf("Freed jcrtbl\n");