2 This file is part of GNUnet.
3 (C) 2001, 2002, 2003, 2004, 2006, 2008 Christian Grothoff (and other contributing authors)
5 GNUnet is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published
7 by the Free Software Foundation; either version 2, or (at your
8 option) any later version.
10 GNUnet is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with GNUnet; see the file COPYING. If not, write to the
17 Free Software Foundation, Inc., 59 Temple Place - Suite 330,
18 Boston, MA 02111-1307, USA.
21 * @file util/container_bloomfilter.c
22 * @brief data structure used to reduce disk accesses.
24 * The idea basically: Create a signature for each element in the
25 * database. Add those signatures to a bit array. When doing a lookup,
26 * check if the bit array matches the signature of the requested
27 * element. If yes, address the disk, otherwise return 'not found'.
29 * A property of the bloom filter is that sometimes we will have
30 * a match even if the element is not on the disk (then we do
31 * an unnecessary disk access), but what's most important is that
32 * we never get a single "false negative".
34 * To be able to delete entries from the bloom filter, we maintain
35 * a 4 bit counter in the file on the drive (we still use only one
38 * @author Igor Wronsky
39 * @author Christian Grothoff
43 #include "gnunet_common.h"
44 #include "gnunet_container_lib.h"
45 #include "gnunet_disk_lib.h"
47 struct GNUNET_CONTAINER_BloomFilter
51 * The actual bloomfilter bit array
56 * Filename of the filter
61 * The bit counter file on disk
63 struct GNUNET_DISK_FileHandle *fh;
66 * How many bits we set for each stored element
68 unsigned int addressesPerElement;
71 * Size of bitArray in bytes
79 * Sets a bit active in the bitArray. Increment bit-specific
80 * usage counter on disk only if below 4bit max (==15).
82 * @param bitArray memory area to set the bit in
83 * @param bitIdx which bit to set
86 setBit (char *bitArray, unsigned int bitIdx)
89 unsigned int targetBit;
91 arraySlot = bitIdx / 8;
92 targetBit = (1L << (bitIdx % 8));
93 bitArray[arraySlot] |= targetBit;
97 * Clears a bit from bitArray. Bit is cleared from the array
98 * only if the respective usage counter on the disk hits/is zero.
100 * @param bitArray memory area to set the bit in
101 * @param bitIdx which bit to unset
104 clearBit (char *bitArray, unsigned int bitIdx)
107 unsigned int targetBit;
110 targetBit = (1L << (bitIdx % 8));
111 bitArray[slot] = bitArray[slot] & (~targetBit);
115 * Checks if a bit is active in the bitArray
117 * @param bitArray memory area to set the bit in
118 * @param bitIdx which bit to test
119 * @return GNUNET_YES if the bit is set, GNUNET_NO if not.
122 testBit (char *bitArray, unsigned int bitIdx)
125 unsigned int targetBit;
128 targetBit = (1L << (bitIdx % 8));
129 if (bitArray[slot] & targetBit)
136 * Sets a bit active in the bitArray and increments
137 * bit-specific usage counter on disk (but only if
138 * the counter was below 4 bit max (==15)).
140 * @param bitArray memory area to set the bit in
141 * @param bitIdx which bit to test
142 * @param fh A file to keep the 4 bit address usage counters in
145 incrementBit (char *bitArray, unsigned int bitIdx, const struct GNUNET_DISK_FileHandle *fh)
151 unsigned int targetLoc;
153 setBit (bitArray, bitIdx);
154 if (GNUNET_DISK_handle_invalid (fh))
156 /* Update the counter file on disk */
157 fileSlot = bitIdx / 2;
158 targetLoc = bitIdx % 2;
160 GNUNET_assert (fileSlot ==
161 GNUNET_DISK_file_seek (fh, fileSlot, GNUNET_DISK_SEEK_SET));
162 if (1 != GNUNET_DISK_file_read (fh, &value, 1))
165 high = (value & (~0xF)) >> 4;
177 value = ((high << 4) | low);
178 GNUNET_assert (fileSlot == GNUNET_DISK_file_seek (fh,
180 GNUNET_DISK_SEEK_SET));
181 GNUNET_assert (1 == GNUNET_DISK_file_write (fh, &value, 1));
185 * Clears a bit from bitArray if the respective usage
186 * counter on the disk hits/is zero.
188 * @param bitArray memory area to set the bit in
189 * @param bitIdx which bit to test
190 * @param fh A file to keep the 4bit address usage counters in
193 decrementBit (char *bitArray, unsigned int bitIdx, const struct GNUNET_DISK_FileHandle *fh)
199 unsigned int targetLoc;
201 if (GNUNET_DISK_handle_invalid (fh))
202 return; /* cannot decrement! */
203 /* Each char slot in the counter file holds two 4 bit counters */
204 fileSlot = bitIdx / 2;
205 targetLoc = bitIdx % 2;
206 GNUNET_DISK_file_seek (fh, fileSlot, GNUNET_DISK_SEEK_SET);
207 if (1 != GNUNET_DISK_file_read (fh, &value, 1))
210 high = (value & 0xF0) >> 4;
212 /* decrement, but once we have reached the max, never go back! */
215 if ((low > 0) && (low < 0xF))
219 clearBit (bitArray, bitIdx);
224 if ((high > 0) && (high < 0xF))
228 clearBit (bitArray, bitIdx);
231 value = ((high << 4) | low);
232 GNUNET_DISK_file_seek (fh, fileSlot, GNUNET_DISK_SEEK_SET);
233 GNUNET_assert (1 == GNUNET_DISK_file_write (fh, &value, 1));
236 #define BUFFSIZE 65536
239 * Creates a file filled with zeroes
241 * @param fh the file handle
242 * @param size the size of the file
243 * @return GNUNET_OK if created ok, GNUNET_SYSERR otherwise
246 makeEmptyFile (const struct GNUNET_DISK_FileHandle *fh,
250 size_t bytesleft = size;
253 if (GNUNET_DISK_handle_invalid (fh))
254 return GNUNET_SYSERR;
255 buffer = GNUNET_malloc (BUFFSIZE);
256 memset (buffer, 0, BUFFSIZE);
257 GNUNET_DISK_file_seek (fh, 0, GNUNET_DISK_SEEK_SET);
259 while (bytesleft > 0)
261 if (bytesleft > BUFFSIZE)
263 res = GNUNET_DISK_file_write (fh, buffer, BUFFSIZE);
264 bytesleft -= BUFFSIZE;
268 res = GNUNET_DISK_file_write (fh, buffer, bytesleft);
271 GNUNET_assert (res != GNUNET_SYSERR);
273 GNUNET_free (buffer);
277 /* ************** GNUNET_CONTAINER_BloomFilter iterator ********* */
280 * Iterator (callback) method to be called by the
281 * bloomfilter iterator on each bit that is to be
282 * set or tested for the key.
285 * @param bf the filter to manipulate
286 * @param bit the current bit
288 typedef void (*BitIterator) (void *cls,
289 struct GNUNET_CONTAINER_BloomFilter * bf,
293 * Call an iterator for each bit that the bloomfilter
294 * must test or set for this element.
296 * @param bf the filter
297 * @param callback the method to call
298 * @param arg extra argument to callback
299 * @param key the key for which we iterate over the BF bits
302 iterateBits (struct GNUNET_CONTAINER_BloomFilter *bf,
303 BitIterator callback, void *arg, const GNUNET_HashCode * key)
305 GNUNET_HashCode tmp[2];
308 unsigned int slot = 0;
310 bitCount = bf->addressesPerElement;
311 memcpy (&tmp[0], key, sizeof (GNUNET_HashCode));
315 while (slot < (sizeof (GNUNET_HashCode) / sizeof (uint32_t)))
319 (((uint32_t *) &tmp[round & 1])[slot]) &
320 ((bf->bitArraySize * 8) - 1));
328 GNUNET_CRYPTO_hash (&tmp[round & 1], sizeof (GNUNET_HashCode),
329 &tmp[(round + 1) & 1]);
337 * Callback: increment bit
339 * @param cls not used
340 * @param bf the filter to manipulate
341 * @param bit the bit to increment
344 incrementBitCallback (void *cls,
345 struct GNUNET_CONTAINER_BloomFilter *bf,
348 incrementBit (bf->bitArray, bit, bf->fh);
352 * Callback: decrement bit
354 * @param cls not used
355 * @param bf the filter to manipulate
356 * @param bit the bit to decrement
359 decrementBitCallback (void *cls,
360 struct GNUNET_CONTAINER_BloomFilter *bf,
363 decrementBit (bf->bitArray, bit, bf->fh);
367 * Callback: test if all bits are set
369 * @param cls pointer set to GNUNET_NO if bit is not set
370 * @param bf the filter
371 * @param bit the bit to test
374 testBitCallback (void *cls,
375 struct GNUNET_CONTAINER_BloomFilter *bf, unsigned int bit)
378 if (GNUNET_NO == testBit (bf->bitArray, bit))
382 /* *********************** INTERFACE **************** */
385 * Load a bloom-filter from a file.
387 * @param filename the name of the file (or the prefix)
388 * @param size the size of the bloom-filter (number of
389 * bytes of storage space to use)
390 * @param k the number of GNUNET_CRYPTO_hash-functions to apply per
391 * element (number of bits set per element in the set)
392 * @return the bloomfilter
394 struct GNUNET_CONTAINER_BloomFilter *
395 GNUNET_CONTAINER_bloomfilter_load (const char *filename,
399 struct GNUNET_CONTAINER_BloomFilter *bf;
405 if ((k == 0) || (size == 0))
412 size = ui; /* make sure it's a power of 2 */
414 bf = GNUNET_malloc (sizeof (struct GNUNET_CONTAINER_BloomFilter));
415 /* Try to open a bloomfilter file */
416 if (filename != NULL)
418 bf->fh = GNUNET_DISK_file_open (filename, GNUNET_DISK_OPEN_READWRITE
419 | GNUNET_DISK_OPEN_CREATE,
420 GNUNET_DISK_PERM_USER_READ | GNUNET_DISK_PERM_USER_WRITE);
426 bf->filename = GNUNET_strdup (filename);
434 bf->bitArray = GNUNET_malloc_large (size);
435 bf->bitArraySize = size;
436 bf->addressesPerElement = k;
437 memset (bf->bitArray, 0, bf->bitArraySize);
439 if (bf->filename != NULL)
441 /* Read from the file what bits we can */
442 rbuff = GNUNET_malloc (BUFFSIZE);
444 while (pos < size * 8)
448 res = GNUNET_DISK_file_read (bf->fh, rbuff, BUFFSIZE);
451 GNUNET_log_strerror_file (GNUNET_ERROR_TYPE_WARNING,
456 break; /* is ok! we just did not use that many bits yet */
457 for (i = 0; i < res; i++)
459 if ((rbuff[i] & 0x0F) != 0)
460 setBit (bf->bitArray, pos + i * 2);
461 if ((rbuff[i] & 0xF0) != 0)
462 setBit (bf->bitArray, pos + i * 2 + 1);
466 pos += BUFFSIZE * 2; /* 2 bits per byte in the buffer */
475 * Create a bloom filter from raw bits.
477 * @param data the raw bits in memory (maybe NULL,
478 * in which case all bits should be considered
480 * @param size the size of the bloom-filter (number of
481 * bytes of storage space to use); also size of data
482 * -- unless data is NULL
483 * @param k the number of GNUNET_CRYPTO_hash-functions to apply per
484 * element (number of bits set per element in the set)
485 * @return the bloomfilter
487 struct GNUNET_CONTAINER_BloomFilter *
488 GNUNET_CONTAINER_bloomfilter_init (const char *data,
492 struct GNUNET_CONTAINER_BloomFilter *bf;
495 if ((k == 0) || (size == 0))
505 bf = GNUNET_malloc (sizeof (struct GNUNET_CONTAINER_BloomFilter));
508 bf->bitArray = GNUNET_malloc_large (size);
509 bf->bitArraySize = size;
510 bf->addressesPerElement = k;
512 memcpy (bf->bitArray, data, size);
514 memset (bf->bitArray, 0, bf->bitArraySize);
520 * Copy the raw data of this bloomfilter into
521 * the given data array.
523 * @param data where to write the data
524 * @param size the size of the given data array
525 * @return GNUNET_SYSERR if the data array is not big enough
528 GNUNET_CONTAINER_bloomfilter_get_raw_data (struct GNUNET_CONTAINER_BloomFilter
533 return GNUNET_SYSERR;
535 if (bf->bitArraySize != size)
536 return GNUNET_SYSERR;
537 memcpy (data, bf->bitArray, size);
542 * Free the space associated with a filter
543 * in memory, flush to drive if needed (do not
544 * free the space on the drive)
546 * @param bf the filter
549 GNUNET_CONTAINER_bloomfilter_free (struct GNUNET_CONTAINER_BloomFilter *bf)
553 if (bf->filename != NULL)
555 GNUNET_DISK_file_close (bf->fh);
556 GNUNET_free (bf->filename);
558 GNUNET_free (bf->bitArray);
563 * Reset a bloom filter to empty. Clears the file on disk.
565 * @param bf the filter
568 GNUNET_CONTAINER_bloomfilter_clear (struct GNUNET_CONTAINER_BloomFilter *bf)
573 memset (bf->bitArray, 0, bf->bitArraySize);
574 if (bf->filename != NULL)
575 makeEmptyFile (bf->fh, bf->bitArraySize * 4);
580 * Test if an element is in the filter.
582 * @param e the element
583 * @param bf the filter
584 * @return GNUNET_YES if the element is in the filter, GNUNET_NO if not
587 GNUNET_CONTAINER_bloomfilter_test (struct GNUNET_CONTAINER_BloomFilter *bf,
588 const GNUNET_HashCode * e)
595 iterateBits (bf, &testBitCallback, &res, e);
600 * Add an element to the filter
602 * @param bf the filter
603 * @param e the element
606 GNUNET_CONTAINER_bloomfilter_add (struct GNUNET_CONTAINER_BloomFilter *bf,
607 const GNUNET_HashCode * e)
612 iterateBits (bf, &incrementBitCallback, NULL, e);
617 * Or the entries of the given raw data array with the
618 * data of the given bloom filter. Assumes that
619 * the size of the data array and the current filter
622 * @param bf the filter
623 * @param data the data to or-in
624 * @param size number of bytes in data
627 GNUNET_CONTAINER_bloomfilter_or (struct GNUNET_CONTAINER_BloomFilter *bf,
635 if (bf->bitArraySize != size)
636 return GNUNET_SYSERR;
637 /* FIXME: we could do this 4-8x faster by
638 going over int/long arrays */
639 for (i = 0; i < size; i++)
640 bf->bitArray[i] |= data[i];
645 * Remove an element from the filter.
647 * @param bf the filter
648 * @param e the element to remove
651 GNUNET_CONTAINER_bloomfilter_remove (struct GNUNET_CONTAINER_BloomFilter *bf,
652 const GNUNET_HashCode * e)
656 if (bf->filename == NULL)
658 iterateBits (bf, &decrementBitCallback, NULL, e);
662 * Resize a bloom filter. Note that this operation
663 * is pretty costly. Essentially, the bloom filter
664 * needs to be completely re-build.
666 * @param bf the filter
667 * @param iterator an iterator over all elements stored in the BF
668 * @param iterator_arg argument to the iterator function
669 * @param size the new size for the filter
670 * @param k the new number of GNUNET_CRYPTO_hash-function to apply per element
673 GNUNET_CONTAINER_bloomfilter_resize (struct GNUNET_CONTAINER_BloomFilter *bf,
674 GNUNET_HashCodeIterator iterator,
682 GNUNET_free (bf->bitArray);
686 size = i; /* make sure it's a power of 2 */
688 bf->bitArraySize = size;
689 bf->bitArray = GNUNET_malloc (size);
690 memset (bf->bitArray, 0, bf->bitArraySize);
691 if (bf->filename != NULL)
692 makeEmptyFile (bf->fh, bf->bitArraySize * 4);
693 while (GNUNET_YES == iterator (iterator_arg, &hc))
694 GNUNET_CONTAINER_bloomfilter_add (bf, &hc);
697 /* end of container_bloomfilter.c */