2 *--------------------------------------------------------------------------------------
3 * The "hash seed" feature was added in Perl 5.8.1 to perturb the results
4 * to avoid "algorithmic complexity attacks".
6 * If USE_HASH_SEED is defined, hash randomisation is done by default
7 * If USE_HASH_SEED_EXPLICIT is defined, hash randomisation is done
8 * only if the environment variable PERL_HASH_SEED is set.
9 * (see also perl.c:perl_parse() and S_init_tls_and_interp() and util.c:get_hash_seed())
12 #ifndef PERL_SEEN_HV_FUNC_H /* compile once */
13 #define PERL_SEEN_HV_FUNC_H
16 || defined(PERL_HASH_FUNC_SDBM) \
17 || defined(PERL_HASH_FUNC_DJB2) \
18 || defined(PERL_HASH_FUNC_SUPERFAST) \
19 || defined(PERL_HASH_FUNC_MURMUR3) \
20 || defined(PERL_HASH_FUNC_ONE_AT_A_TIME) \
21 || defined(PERL_HASH_FUNC_ONE_AT_A_TIME_HARD) \
22 || defined(PERL_HASH_FUNC_ONE_AT_A_TIME_OLD) \
25 #define PERL_HASH_FUNC_SIPHASH
27 #define PERL_HASH_FUNC_ONE_AT_A_TIME_HARD
31 #if defined(PERL_HASH_FUNC_SIPHASH)
32 # define PERL_HASH_FUNC "SIPHASH_2_4"
33 # define PERL_HASH_SEED_BYTES 16
34 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_siphash_2_4(PERL_HASH_SEED,(U8*)(str),(len))
35 #elif defined(PERL_HASH_FUNC_SUPERFAST)
36 # define PERL_HASH_FUNC "SUPERFAST"
37 # define PERL_HASH_SEED_BYTES 4
38 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_superfast(PERL_HASH_SEED,(U8*)(str),(len))
39 #elif defined(PERL_HASH_FUNC_MURMUR3)
40 # define PERL_HASH_FUNC "MURMUR3"
41 # define PERL_HASH_SEED_BYTES 4
42 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_murmur3(PERL_HASH_SEED,(U8*)(str),(len))
43 #elif defined(PERL_HASH_FUNC_DJB2)
44 # define PERL_HASH_FUNC "DJB2"
45 # define PERL_HASH_SEED_BYTES 4
46 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_djb2(PERL_HASH_SEED,(U8*)(str),(len))
47 #elif defined(PERL_HASH_FUNC_SDBM)
48 # define PERL_HASH_FUNC "SDBM"
49 # define PERL_HASH_SEED_BYTES 4
50 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_sdbm(PERL_HASH_SEED,(U8*)(str),(len))
51 #elif defined(PERL_HASH_FUNC_ONE_AT_A_TIME_HARD)
52 # define PERL_HASH_FUNC "ONE_AT_A_TIME_HARD"
53 # define PERL_HASH_SEED_BYTES 8
54 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_one_at_a_time_hard(PERL_HASH_SEED,(U8*)(str),(len))
55 #elif defined(PERL_HASH_FUNC_ONE_AT_A_TIME)
56 # define PERL_HASH_FUNC "ONE_AT_A_TIME"
57 # define PERL_HASH_SEED_BYTES 4
58 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_one_at_a_time(PERL_HASH_SEED,(U8*)(str),(len))
59 #elif defined(PERL_HASH_FUNC_ONE_AT_A_TIME_OLD)
60 # define PERL_HASH_FUNC "ONE_AT_A_TIME_OLD"
61 # define PERL_HASH_SEED_BYTES 4
62 # define PERL_HASH(hash,str,len) (hash)= S_perl_hash_old_one_at_a_time(PERL_HASH_SEED,(U8*)(str),(len))
66 #error "No hash function defined!"
68 #ifndef PERL_HASH_SEED_BYTES
69 #error "PERL_HASH_SEED_BYTES not defined"
71 #ifndef PERL_HASH_FUNC
72 #error "PERL_HASH_FUNC not defined"
75 #ifndef PERL_HASH_SEED
76 # if defined(USE_HASH_SEED) || defined(USE_HASH_SEED_EXPLICIT)
77 # define PERL_HASH_SEED PL_hash_seed
78 # elif PERL_HASH_SEED_BYTES == 4
79 # define PERL_HASH_SEED "PeRl"
80 # elif PERL_HASH_SEED_BYTES == 16
81 # define PERL_HASH_SEED "PeRlHaShhAcKpErl"
83 # error "No PERL_HASH_SEED definition for " PERL_HASH_FUNC
87 /*-----------------------------------------------------------------------------
88 * Endianess, misalignment capabilities and util macros
90 * The following 3 macros are defined in this section. The other macros defined
91 * are only needed to help derive these 3.
93 * U8TO32_LE(x) Read a little endian unsigned 32-bit int
94 * UNALIGNED_SAFE Defined if READ_UINT32 works on non-word boundaries
95 * ROTL32(x,r) Rotate x left by r bits
98 #if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \
99 || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__)
100 #define U8TO16_LE(d) (*((const U16 *) (d)))
103 #if !defined (U8TO16_LE)
104 #define U8TO16_LE(d) ((((const U8 *)(d))[1] << 8)\
105 +((const U8 *)(d))[0])
109 /* Now find best way we can to READ_UINT32 */
110 #if (BYTEORDER == 0x1234 || BYTEORDER == 0x12345678) && U32SIZE == 4
111 /* CPU endian matches murmurhash algorithm, so read 32-bit word directly */
112 #define U8TO32_LE(ptr) (*((U32*)(ptr)))
113 #elif BYTEORDER == 0x4321 || BYTEORDER == 0x87654321
114 /* TODO: Add additional cases below where a compiler provided bswap32 is available */
115 #if defined(__GNUC__) && (__GNUC__>4 || (__GNUC__==4 && __GNUC_MINOR__>=3))
116 #define U8TO32_LE(ptr) (__builtin_bswap32(*((U32*)(ptr))))
118 /* Without a known fast bswap32 we're just as well off doing this */
119 #define U8TO32_LE(ptr) (ptr[0]|ptr[1]<<8|ptr[2]<<16|ptr[3]<<24)
120 #define UNALIGNED_SAFE
123 /* Unknown endianess so last resort is to read individual bytes */
124 #define U8TO32_LE(ptr) (ptr[0]|ptr[1]<<8|ptr[2]<<16|ptr[3]<<24)
125 /* Since we're not doing word-reads we can skip the messing about with realignment */
126 #define UNALIGNED_SAFE
129 /* Find best way to ROTL32 */
130 #if defined(_MSC_VER)
131 #include <stdlib.h> /* Microsoft put _rotl declaration in here */
132 #define ROTL32(x,r) _rotl(x,r)
134 /* gcc recognises this code and generates a rotate instruction for CPUs with one */
135 #define ROTL32(x,r) (((U32)x << r) | ((U32)x >> (32 - r)))
139 /* This is SipHash by Jean-Philippe Aumasson and Daniel J. Bernstein.
140 * The authors claim it is relatively secure compared to the alternatives
141 * and that performance wise it is a suitable hash for languages like Perl.
144 * https://www.131002.net/siphash/
146 * This implementation seems to perform slightly slower than one-at-a-time for
147 * short keys, but degrades slower for longer keys. Murmur Hash outperforms it
148 * regardless of keys size.
156 /* This probably isn't going to work, but failing with a compiler error due to
157 lack of uint64_t is no worse than failing right now with an #error. */
158 #define U64TYPE uint64_t
162 #define ROTL64(x,b) (U64TYPE)( ((x) << (b)) | ( (x) >> (64 - (b))) )
164 #define U8TO64_LE(p) \
165 (((U64TYPE)((p)[0]) ) | \
166 ((U64TYPE)((p)[1]) << 8) | \
167 ((U64TYPE)((p)[2]) << 16) | \
168 ((U64TYPE)((p)[3]) << 24) | \
169 ((U64TYPE)((p)[4]) << 32) | \
170 ((U64TYPE)((p)[5]) << 40) | \
171 ((U64TYPE)((p)[6]) << 48) | \
172 ((U64TYPE)((p)[7]) << 56))
176 v0 += v1; v1=ROTL64(v1,13); v1 ^= v0; v0=ROTL64(v0,32); \
177 v2 += v3; v3=ROTL64(v3,16); v3 ^= v2; \
178 v0 += v3; v3=ROTL64(v3,21); v3 ^= v0; \
179 v2 += v1; v1=ROTL64(v1,17); v1 ^= v2; v2=ROTL64(v2,32); \
184 PERL_STATIC_INLINE U32
185 S_perl_hash_siphash_2_4(const unsigned char * const seed, const unsigned char *in, const STRLEN inlen) {
186 /* "somepseudorandomlygeneratedbytes" */
187 U64TYPE v0 = 0x736f6d6570736575ULL;
188 U64TYPE v1 = 0x646f72616e646f6dULL;
189 U64TYPE v2 = 0x6c7967656e657261ULL;
190 U64TYPE v3 = 0x7465646279746573ULL;
193 U64TYPE k0 = ((U64TYPE*)seed)[0];
194 U64TYPE k1 = ((U64TYPE*)seed)[1];
196 const int left = inlen & 7;
197 const U8 *end = in + inlen - left;
199 b = ( ( U64TYPE )(inlen) ) << 56;
205 for ( ; in != end; in += 8 )
216 case 7: b |= ( ( U64TYPE )in[ 6] ) << 48;
217 case 6: b |= ( ( U64TYPE )in[ 5] ) << 40;
218 case 5: b |= ( ( U64TYPE )in[ 4] ) << 32;
219 case 4: b |= ( ( U64TYPE )in[ 3] ) << 24;
220 case 3: b |= ( ( U64TYPE )in[ 2] ) << 16;
221 case 2: b |= ( ( U64TYPE )in[ 1] ) << 8;
222 case 1: b |= ( ( U64TYPE )in[ 0] ); break;
236 b = v0 ^ v1 ^ v2 ^ v3;
237 return (U32)(b & U32_MAX);
239 #endif /* defined(HAS_QUAD) */
241 /* FYI: This is the "Super-Fast" algorithm mentioned by Bob Jenkins in
242 * (http://burtleburtle.net/bob/hash/doobs.html)
243 * It is by Paul Hsieh (c) 2004 and is analysed here
244 * http://www.azillionmonkeys.com/qed/hash.html
245 * license terms are here:
246 * http://www.azillionmonkeys.com/qed/weblicense.html
250 PERL_STATIC_INLINE U32
251 S_perl_hash_superfast(const unsigned char * const seed, const unsigned char *str, STRLEN len) {
252 U32 hash = *((U32*)seed) + len;
257 for (;len > 0; len--) {
258 hash += U8TO16_LE (str);
259 tmp = (U8TO16_LE (str+2) << 11) ^ hash;
260 hash = (hash << 16) ^ tmp;
261 str += 2 * sizeof (U16);
265 /* Handle end cases */
267 case 3: hash += U8TO16_LE (str);
269 hash ^= str[sizeof (U16)] << 18;
272 case 2: hash += U8TO16_LE (str);
276 case 1: hash += *str;
280 /* Force "avalanching" of final 127 bits */
286 return (hash + (hash >> 6));
290 /*-----------------------------------------------------------------------------
291 * MurmurHash3 was written by Austin Appleby, and is placed in the public
294 * This implementation was originally written by Shane Day, and is also public domain,
295 * and was modified to function as a macro similar to other perl hash functions by
298 * This is a portable ANSI C implementation of MurmurHash3_x86_32 (Murmur3A)
299 * with support for progressive processing.
301 * If you want to understand the MurmurHash algorithm you would be much better
302 * off reading the original source. Just point your browser at:
303 * http://code.google.com/p/smhasher/source/browse/trunk/MurmurHash3.cpp
307 * We can only process entire 32 bit chunks of input, except for the very end
308 * that may be shorter.
310 * To handle endianess I simply use a macro that reads a U32 and define
311 * that macro to be a direct read on little endian machines, a read and swap
312 * on big endian machines, or a byte-by-byte read if the endianess is unknown.
316 /*-----------------------------------------------------------------------------
317 * Core murmurhash algorithm macros */
319 #define MURMUR_C1 (0xcc9e2d51)
320 #define MURMUR_C2 (0x1b873593)
321 #define MURMUR_C3 (0xe6546b64)
322 #define MURMUR_C4 (0x85ebca6b)
323 #define MURMUR_C5 (0xc2b2ae35)
325 /* This is the main processing body of the algorithm. It operates
326 * on each full 32-bits of input. */
327 #define MURMUR_DOBLOCK(h1, k1) STMT_START { \
329 k1 = ROTL32(k1,15); \
333 h1 = ROTL32(h1,13); \
334 h1 = h1 * 5 + MURMUR_C3; \
338 /* Append unaligned bytes to carry, forcing hash churn if we have 4 bytes */
339 /* cnt=bytes to process, h1=name of h1 var, c=carry, n=bytes in c, ptr/len=payload */
340 #define MURMUR_DOBYTES(cnt, h1, c, n, ptr, len) STMT_START { \
341 int MURMUR_DOBYTES_i = cnt; \
342 while(MURMUR_DOBYTES_i--) { \
343 c = c>>8 | *ptr++<<24; \
346 MURMUR_DOBLOCK(h1, c); \
353 /* now we create the hash function */
354 PERL_STATIC_INLINE U32
355 S_perl_hash_murmur3(const unsigned char * const seed, const unsigned char *ptr, STRLEN len) {
356 U32 h1 = *((U32*)seed);
360 const unsigned char *end;
361 int bytes_in_carry = 0; /* bytes in carry */
362 I32 total_length= len;
364 #if defined(UNALIGNED_SAFE)
365 /* Handle carry: commented out as its only used in incremental mode - it never fires for us
368 MURMUR_DOBYTES(i, h1, carry, bytes_in_carry, ptr, len);
372 /* This CPU handles unaligned word access */
373 /* Process 32-bit chunks */
375 for( ; ptr < end ; ptr+=4) {
377 MURMUR_DOBLOCK(h1, k1);
380 /* This CPU does not handle unaligned word access */
382 /* Consume enough so that the next data byte is word aligned */
383 int i = -(long)ptr & 3;
385 MURMUR_DOBYTES(i, h1, carry, bytes_in_carry, ptr, len);
388 /* We're now aligned. Process in aligned blocks. Specialise for each possible carry count */
390 switch(bytes_in_carry) { /* how many bytes in carry */
391 case 0: /* c=[----] w=[3210] b=[3210]=w c'=[----] */
392 for( ; ptr < end ; ptr+=4) {
394 MURMUR_DOBLOCK(h1, k1);
397 case 1: /* c=[0---] w=[4321] b=[3210]=c>>24|w<<8 c'=[4---] */
398 for( ; ptr < end ; ptr+=4) {
400 carry = U8TO32_LE(ptr);
402 MURMUR_DOBLOCK(h1, k1);
405 case 2: /* c=[10--] w=[5432] b=[3210]=c>>16|w<<16 c'=[54--] */
406 for( ; ptr < end ; ptr+=4) {
408 carry = U8TO32_LE(ptr);
410 MURMUR_DOBLOCK(h1, k1);
413 case 3: /* c=[210-] w=[6543] b=[3210]=c>>8|w<<24 c'=[654-] */
414 for( ; ptr < end ; ptr+=4) {
416 carry = U8TO32_LE(ptr);
418 MURMUR_DOBLOCK(h1, k1);
422 /* Advance over whole 32-bit chunks, possibly leaving 1..3 bytes */
425 /* Append any remaining bytes into carry */
426 MURMUR_DOBYTES(len, h1, carry, bytes_in_carry, ptr, len);
428 if (bytes_in_carry) {
429 k1 = carry >> ( 4 - bytes_in_carry ) * 8;
447 PERL_STATIC_INLINE U32
448 S_perl_hash_djb2(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
449 const unsigned char * const end = (const unsigned char *)str + len;
450 U32 hash = *((U32*)seed + len);
452 hash = ((hash << 5) + hash) + *str++;
457 PERL_STATIC_INLINE U32
458 S_perl_hash_sdbm(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
459 const unsigned char * const end = (const unsigned char *)str + len;
460 U32 hash = *((U32*)seed + len);
462 hash = (hash << 6) + (hash << 16) - hash + *str++;
468 /* This is the "One-at-a-Time" algorithm by Bob Jenkins
469 * from requirements by Colin Plumb.
470 * (http://burtleburtle.net/bob/hash/doobs.html)
471 * With seed/len tweak.
473 PERL_STATIC_INLINE U32
474 S_perl_hash_one_at_a_time(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
475 const unsigned char * const end = (const unsigned char *)str + len;
476 U32 hash = *((U32*)seed) + len;
479 hash += (hash << 10);
483 hash ^= (hash >> 11);
484 return (hash + (hash << 15));
487 /* Derived from "One-at-a-Time" algorithm by Bob Jenkins */
488 PERL_STATIC_INLINE U32
489 S_perl_hash_one_at_a_time_hard(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
490 const unsigned char * const end = (const unsigned char *)str + len;
491 U32 hash = *((U32*)seed) + len;
494 hash += (hash << 10);
499 hash += (hash << 10);
503 hash += (hash << 10);
507 hash += (hash << 10);
511 hash += (hash << 10);
515 hash += (hash << 10);
519 hash ^= (hash >> 11);
520 return (hash + (hash << 15));
523 PERL_STATIC_INLINE U32
524 S_perl_hash_old_one_at_a_time(const unsigned char * const seed, const unsigned char *str, const STRLEN len) {
525 const unsigned char * const end = (const unsigned char *)str + len;
526 U32 hash = *((U32*)seed);
529 hash += (hash << 10);
533 hash ^= (hash >> 11);
534 return (hash + (hash << 15));
537 /* legacy - only mod_perl should be doing this. */
538 #ifdef PERL_HASH_INTERNAL_ACCESS
539 #define PERL_HASH_INTERNAL(hash,str,len) PERL_HASH(hash,str,len)
542 #endif /*compile once*/
546 * c-indentation-style: bsd
548 * indent-tabs-mode: nil
551 * ex: set ts=8 sts=4 sw=4 et: