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
80 * Get size of the bloom filter.
82 * @param bf the filter
83 * @return number of bytes used for the data of the bloom filter
86 GNUNET_CONTAINER_bloomfilter_get_size (const struct GNUNET_CONTAINER_BloomFilter
91 return bf->bitArraySize;
96 * Copy an existing memory. Any association with a file
97 * on-disk will be lost in the process.
98 * @param bf the filter to copy
99 * @return copy of the bf
101 struct GNUNET_CONTAINER_BloomFilter *
102 GNUNET_CONTAINER_bloomfilter_copy (const struct GNUNET_CONTAINER_BloomFilter
105 return GNUNET_CONTAINER_bloomfilter_init (bf->bitArray,
107 bf->addressesPerElement);
112 * Sets a bit active in the bitArray. Increment bit-specific
113 * usage counter on disk only if below 4bit max (==15).
115 * @param bitArray memory area to set the bit in
116 * @param bitIdx which bit to set
119 setBit (char *bitArray, unsigned int bitIdx)
122 unsigned int targetBit;
124 arraySlot = bitIdx / 8;
125 targetBit = (1L << (bitIdx % 8));
126 bitArray[arraySlot] |= targetBit;
130 * Clears a bit from bitArray. Bit is cleared from the array
131 * only if the respective usage counter on the disk hits/is zero.
133 * @param bitArray memory area to set the bit in
134 * @param bitIdx which bit to unset
137 clearBit (char *bitArray, unsigned int bitIdx)
140 unsigned int targetBit;
143 targetBit = (1L << (bitIdx % 8));
144 bitArray[slot] = bitArray[slot] & (~targetBit);
148 * Checks if a bit is active in the bitArray
150 * @param bitArray memory area to set the bit in
151 * @param bitIdx which bit to test
152 * @return GNUNET_YES if the bit is set, GNUNET_NO if not.
155 testBit (char *bitArray, unsigned int bitIdx)
158 unsigned int targetBit;
161 targetBit = (1L << (bitIdx % 8));
162 if (bitArray[slot] & targetBit)
169 * Sets a bit active in the bitArray and increments
170 * bit-specific usage counter on disk (but only if
171 * the counter was below 4 bit max (==15)).
173 * @param bitArray memory area to set the bit in
174 * @param bitIdx which bit to test
175 * @param fh A file to keep the 4 bit address usage counters in
178 incrementBit (char *bitArray, unsigned int bitIdx,
179 const struct GNUNET_DISK_FileHandle *fh)
185 unsigned int targetLoc;
187 setBit (bitArray, bitIdx);
188 if (GNUNET_DISK_handle_invalid (fh))
190 /* Update the counter file on disk */
191 fileSlot = bitIdx / 2;
192 targetLoc = bitIdx % 2;
194 GNUNET_assert (fileSlot ==
195 GNUNET_DISK_file_seek (fh, fileSlot, GNUNET_DISK_SEEK_SET));
196 if (1 != GNUNET_DISK_file_read (fh, &value, 1))
199 high = (value & (~0xF)) >> 4;
211 value = ((high << 4) | low);
212 GNUNET_assert (fileSlot == GNUNET_DISK_file_seek (fh,
214 GNUNET_DISK_SEEK_SET));
215 GNUNET_assert (1 == GNUNET_DISK_file_write (fh, &value, 1));
219 * Clears a bit from bitArray if the respective usage
220 * counter on the disk hits/is zero.
222 * @param bitArray memory area to set the bit in
223 * @param bitIdx which bit to test
224 * @param fh A file to keep the 4bit address usage counters in
227 decrementBit (char *bitArray, unsigned int bitIdx,
228 const struct GNUNET_DISK_FileHandle *fh)
234 unsigned int targetLoc;
236 if (GNUNET_DISK_handle_invalid (fh))
237 return; /* cannot decrement! */
238 /* Each char slot in the counter file holds two 4 bit counters */
239 fileSlot = bitIdx / 2;
240 targetLoc = bitIdx % 2;
241 GNUNET_DISK_file_seek (fh, fileSlot, GNUNET_DISK_SEEK_SET);
242 if (1 != GNUNET_DISK_file_read (fh, &value, 1))
245 high = (value & 0xF0) >> 4;
247 /* decrement, but once we have reached the max, never go back! */
250 if ((low > 0) && (low < 0xF))
254 clearBit (bitArray, bitIdx);
259 if ((high > 0) && (high < 0xF))
263 clearBit (bitArray, bitIdx);
266 value = ((high << 4) | low);
267 GNUNET_DISK_file_seek (fh, fileSlot, GNUNET_DISK_SEEK_SET);
268 GNUNET_assert (1 == GNUNET_DISK_file_write (fh, &value, 1));
271 #define BUFFSIZE 65536
274 * Creates a file filled with zeroes
276 * @param fh the file handle
277 * @param size the size of the file
278 * @return GNUNET_OK if created ok, GNUNET_SYSERR otherwise
281 makeEmptyFile (const struct GNUNET_DISK_FileHandle *fh, size_t size)
284 size_t bytesleft = size;
287 if (GNUNET_DISK_handle_invalid (fh))
288 return GNUNET_SYSERR;
289 buffer = GNUNET_malloc (BUFFSIZE);
290 memset (buffer, 0, BUFFSIZE);
291 GNUNET_DISK_file_seek (fh, 0, GNUNET_DISK_SEEK_SET);
293 while (bytesleft > 0)
295 if (bytesleft > BUFFSIZE)
297 res = GNUNET_DISK_file_write (fh, buffer, BUFFSIZE);
298 bytesleft -= BUFFSIZE;
302 res = GNUNET_DISK_file_write (fh, buffer, bytesleft);
305 GNUNET_assert (res != GNUNET_SYSERR);
307 GNUNET_free (buffer);
311 /* ************** GNUNET_CONTAINER_BloomFilter iterator ********* */
314 * Iterator (callback) method to be called by the
315 * bloomfilter iterator on each bit that is to be
316 * set or tested for the key.
319 * @param bf the filter to manipulate
320 * @param bit the current bit
322 typedef void (*BitIterator) (void *cls,
323 const struct GNUNET_CONTAINER_BloomFilter * bf,
327 * Call an iterator for each bit that the bloomfilter
328 * must test or set for this element.
330 * @param bf the filter
331 * @param callback the method to call
332 * @param arg extra argument to callback
333 * @param key the key for which we iterate over the BF bits
336 iterateBits (const struct GNUNET_CONTAINER_BloomFilter *bf,
337 BitIterator callback, void *arg, const GNUNET_HashCode * key)
339 GNUNET_HashCode tmp[2];
342 unsigned int slot = 0;
344 bitCount = bf->addressesPerElement;
345 memcpy (&tmp[0], key, sizeof (GNUNET_HashCode));
349 while (slot < (sizeof (GNUNET_HashCode) / sizeof (uint32_t)))
353 (((uint32_t *) & tmp[round & 1])[slot]) &
354 ((bf->bitArraySize * 8) - 1));
362 GNUNET_CRYPTO_hash (&tmp[round & 1], sizeof (GNUNET_HashCode),
363 &tmp[(round + 1) & 1]);
371 * Callback: increment bit
373 * @param cls pointer to writeable form of bf
374 * @param bf the filter to manipulate
375 * @param bit the bit to increment
378 incrementBitCallback (void *cls,
379 const struct GNUNET_CONTAINER_BloomFilter *bf,
382 struct GNUNET_CONTAINER_BloomFilter *b = cls;
384 incrementBit (b->bitArray, bit, bf->fh);
388 * Callback: decrement bit
390 * @param cls pointer to writeable form of bf
391 * @param bf the filter to manipulate
392 * @param bit the bit to decrement
395 decrementBitCallback (void *cls,
396 const struct GNUNET_CONTAINER_BloomFilter *bf,
399 struct GNUNET_CONTAINER_BloomFilter *b = cls;
401 decrementBit (b->bitArray, bit, bf->fh);
405 * Callback: test if all bits are set
407 * @param cls pointer set to GNUNET_NO if bit is not set
408 * @param bf the filter
409 * @param bit the bit to test
412 testBitCallback (void *cls,
413 const struct GNUNET_CONTAINER_BloomFilter *bf,
418 if (GNUNET_NO == testBit (bf->bitArray, bit))
422 /* *********************** INTERFACE **************** */
425 * Load a bloom-filter from a file.
427 * @param filename the name of the file (or the prefix)
428 * @param size the size of the bloom-filter (number of
429 * bytes of storage space to use)
430 * @param k the number of GNUNET_CRYPTO_hash-functions to apply per
431 * element (number of bits set per element in the set)
432 * @return the bloomfilter
434 struct GNUNET_CONTAINER_BloomFilter *
435 GNUNET_CONTAINER_bloomfilter_load (const char *filename,
436 size_t size, unsigned int k)
438 struct GNUNET_CONTAINER_BloomFilter *bf;
444 GNUNET_assert (NULL != filename);
445 if ((k == 0) || (size == 0))
452 size = ui; /* make sure it's a power of 2 */
454 bf = GNUNET_malloc (sizeof (struct GNUNET_CONTAINER_BloomFilter));
455 /* Try to open a bloomfilter file */
456 bf->fh = GNUNET_DISK_file_open (filename, GNUNET_DISK_OPEN_READWRITE
457 | GNUNET_DISK_OPEN_CREATE,
458 GNUNET_DISK_PERM_USER_READ |
459 GNUNET_DISK_PERM_USER_WRITE);
465 bf->filename = GNUNET_strdup (filename);
467 bf->bitArray = GNUNET_malloc_large (size);
468 if (bf->bitArray == NULL)
471 GNUNET_DISK_file_close (bf->fh);
472 GNUNET_free (bf->filename);
476 bf->bitArraySize = size;
477 bf->addressesPerElement = k;
478 memset (bf->bitArray, 0, bf->bitArraySize);
480 /* Read from the file what bits we can */
481 rbuff = GNUNET_malloc (BUFFSIZE);
483 while (pos < size * 8)
487 res = GNUNET_DISK_file_read (bf->fh, rbuff, BUFFSIZE);
490 GNUNET_log_strerror_file (GNUNET_ERROR_TYPE_WARNING,
491 "read", bf->filename);
494 break; /* is ok! we just did not use that many bits yet */
495 for (i = 0; i < res; i++)
497 if ((rbuff[i] & 0x0F) != 0)
498 setBit (bf->bitArray, pos + i * 2);
499 if ((rbuff[i] & 0xF0) != 0)
500 setBit (bf->bitArray, pos + i * 2 + 1);
504 pos += BUFFSIZE * 2; /* 2 bits per byte in the buffer */
512 * Create a bloom filter from raw bits.
514 * @param data the raw bits in memory (maybe NULL,
515 * in which case all bits should be considered
517 * @param size the size of the bloom-filter (number of
518 * bytes of storage space to use); also size of data
519 * -- unless data is NULL
520 * @param k the number of GNUNET_CRYPTO_hash-functions to apply per
521 * element (number of bits set per element in the set)
522 * @return the bloomfilter
524 struct GNUNET_CONTAINER_BloomFilter *
525 GNUNET_CONTAINER_bloomfilter_init (const char *data,
526 size_t size, unsigned int k)
528 struct GNUNET_CONTAINER_BloomFilter *bf;
531 if ((k == 0) || (size == 0))
541 bf = GNUNET_malloc (sizeof (struct GNUNET_CONTAINER_BloomFilter));
544 bf->bitArray = GNUNET_malloc_large (size);
545 if (bf->bitArray == NULL)
550 bf->bitArraySize = size;
551 bf->addressesPerElement = k;
553 memcpy (bf->bitArray, data, size);
555 memset (bf->bitArray, 0, bf->bitArraySize);
561 * Copy the raw data of this bloomfilter into
562 * the given data array.
564 * @param bf bloomfilter to take the raw data from
565 * @param data where to write the data
566 * @param size the size of the given data array
567 * @return GNUNET_SYSERR if the data array is not big enough
570 GNUNET_CONTAINER_bloomfilter_get_raw_data (const struct
571 GNUNET_CONTAINER_BloomFilter *bf,
572 char *data, size_t size)
575 return GNUNET_SYSERR;
576 if (bf->bitArraySize != size)
577 return GNUNET_SYSERR;
578 memcpy (data, bf->bitArray, size);
583 * Free the space associated with a filter
584 * in memory, flush to drive if needed (do not
585 * free the space on the drive)
587 * @param bf the filter
590 GNUNET_CONTAINER_bloomfilter_free (struct GNUNET_CONTAINER_BloomFilter *bf)
595 GNUNET_DISK_file_close (bf->fh);
596 GNUNET_free_non_null (bf->filename);
597 GNUNET_free (bf->bitArray);
602 * Reset a bloom filter to empty. Clears the file on disk.
604 * @param bf the filter
607 GNUNET_CONTAINER_bloomfilter_clear (struct GNUNET_CONTAINER_BloomFilter *bf)
612 memset (bf->bitArray, 0, bf->bitArraySize);
613 if (bf->filename != NULL)
614 makeEmptyFile (bf->fh, bf->bitArraySize * 4);
619 * Test if an element is in the filter.
621 * @param e the element
622 * @param bf the filter
623 * @return GNUNET_YES if the element is in the filter, GNUNET_NO if not
626 GNUNET_CONTAINER_bloomfilter_test (const struct GNUNET_CONTAINER_BloomFilter
627 *bf, const GNUNET_HashCode * e)
634 iterateBits (bf, &testBitCallback, &res, e);
639 * Add an element to the filter
641 * @param bf the filter
642 * @param e the element
645 GNUNET_CONTAINER_bloomfilter_add (struct GNUNET_CONTAINER_BloomFilter *bf,
646 const GNUNET_HashCode * e)
651 iterateBits (bf, &incrementBitCallback, bf, e);
656 * Or the entries of the given raw data array with the
657 * data of the given bloom filter. Assumes that
658 * the size of the data array and the current filter
661 * @param bf the filter
662 * @param data the data to or-in
663 * @param size number of bytes in data
666 GNUNET_CONTAINER_bloomfilter_or (struct GNUNET_CONTAINER_BloomFilter *bf,
667 const char *data, size_t size)
671 unsigned long long *fc;
672 const unsigned long long *dc;
676 if (bf->bitArraySize != size)
677 return GNUNET_SYSERR;
678 fc = (unsigned long long *) bf->bitArray;
679 dc = (const unsigned long long *) data;
680 n = size / sizeof (unsigned long long);
682 for (i = 0; i < n; i++)
684 for (i = n * sizeof (unsigned long long); i < size; i++)
685 bf->bitArray[i] |= data[i];
690 * Or the entries of the given raw data array with the
691 * data of the given bloom filter. Assumes that
692 * the size of the data array and the current filter
695 * @param bf the filter
696 * @param to_or the bloomfilter to or-in
697 * @param size number of bytes in data
700 GNUNET_CONTAINER_bloomfilter_or2 (struct GNUNET_CONTAINER_BloomFilter *bf,
701 const struct GNUNET_CONTAINER_BloomFilter
706 unsigned long long *fc;
707 const unsigned long long *dc;
711 if (bf->bitArraySize != size)
712 return GNUNET_SYSERR;
713 fc = (unsigned long long *) bf->bitArray;
714 dc = (const unsigned long long *) to_or->bitArray;
715 n = size / sizeof (unsigned long long);
717 for (i = 0; i < n; i++)
719 for (i = n * sizeof (unsigned long long); i < size; i++)
720 bf->bitArray[i] |= to_or->bitArray[i];
725 * Remove an element from the filter.
727 * @param bf the filter
728 * @param e the element to remove
731 GNUNET_CONTAINER_bloomfilter_remove (struct GNUNET_CONTAINER_BloomFilter *bf,
732 const GNUNET_HashCode * e)
736 if (bf->filename == NULL)
738 iterateBits (bf, &decrementBitCallback, bf, e);
742 * Resize a bloom filter. Note that this operation
743 * is pretty costly. Essentially, the bloom filter
744 * needs to be completely re-build.
746 * @param bf the filter
747 * @param iterator an iterator over all elements stored in the BF
748 * @param iterator_cls argument to the iterator function
749 * @param size the new size for the filter
750 * @param k the new number of GNUNET_CRYPTO_hash-function to apply per element
753 GNUNET_CONTAINER_bloomfilter_resize (struct GNUNET_CONTAINER_BloomFilter *bf,
754 GNUNET_HashCodeIterator iterator,
756 size_t size, unsigned int k)
761 GNUNET_free (bf->bitArray);
765 size = i; /* make sure it's a power of 2 */
767 bf->bitArraySize = size;
768 bf->bitArray = GNUNET_malloc (size);
769 memset (bf->bitArray, 0, bf->bitArraySize);
770 if (bf->filename != NULL)
771 makeEmptyFile (bf->fh, bf->bitArraySize * 4);
772 while (GNUNET_YES == iterator (iterator_cls, &hc))
773 GNUNET_CONTAINER_bloomfilter_add (bf, &hc);
776 /* end of container_bloomfilter.c */