1
$Id: 085_sessreg_implement_hostname_hashing.diff 689 2005-10-19 22:11:30Z dnusinow $
3
The sessreg program assumes that hostnames in utmp entries are unique in
4
the last four characters. When two entries are "test:0" and "fist:0", for
5
example, this assumption fails.
7
Rectify this problem by hashing the entire hostname field from the utmp
8
entry, so that we can tell unlike hostnames apart with much greater
9
reliability. There is still a possiblility of one hash collision in 2^32.
11
This patch by Maximiliano Curia and Damián Viano, using a public-domain
12
hash algorithm by Bob Jenkins.
14
Not submitted to XFree86.
16
Index: sessreg/sessreg.c
17
===================================================================
18
--- sessreg/sessreg.c.orig 2005-10-18 19:45:25.000000000 -0400
19
+++ sessreg/sessreg.c 2005-10-18 21:36:02.000000000 -0400
21
const char *host, Time_t date, int addp);
25
+/* used for hashing ut_id */
26
+typedef unsigned long int ub4; /* unsigned 4-byte quantities */
27
+typedef unsigned char ub1; /* unsigned 1-byte quantities */
29
+#define hashsize(n) ((ub4)1<<(n))
30
+#define hashmask(n) (hashsize(n)-1)
32
+ub4 hash(register ub1 *k, register ub4 length, register ub4 initval);
36
int wflag, uflag, lflag;
37
char *wtmp_file, *utmp_file, *line;
40
bzero (u->ut_name, sizeof (u->ut_name));
45
- * this is a bit crufty, but
46
- * follows the apparent conventions in
47
- * the ttys file. ut_id is only 4 bytes
48
- * long, and the last 4 bytes of the line
49
- * name are written into it, left justified.
50
+ * The ut_id is 4 bytes long. We make a hash of the line
51
+ * received, preceding it by ":" to prevent clashing with
55
- if (i >= sizeof (u->ut_id))
56
- i -= sizeof (u->ut_id);
59
- (void) strncpy (u->ut_id, line + i, sizeof (u->ut_id));
62
+ h = hash(line, strlen(line),0x9e3779b9);
63
+ h = (h & hashmask((sizeof(u->ut_id)-sizeof(char))*8));
64
+ (void) strncpy (u->ut_id + 1,(char *) &h, sizeof (u->ut_id)-sizeof(char));
68
+ * Clearing ut_id may result in race conditions leading to corrupted
69
+ * utmp entries and and potential security holes.
71
+ /* TODO: CHECK this */
72
bzero (u->ut_id, sizeof (u->ut_id));
74
u->ut_pid = getppid ();
82
+--------------------------------------------------------------------
83
+mix -- mix 3 32-bit values reversibly.
84
+For every delta with one or two bits set, and the deltas of all three
85
+ high bits or all three low bits, whether the original value of a,b,c
86
+ is almost all zero or is uniformly distributed,
87
+* If mix() is run forward or backward, at least 32 bits in a,b,c
88
+ have at least 1/4 probability of changing.
89
+* If mix() is run forward, every bit of c will change between 1/3 and
90
+ 2/3 of the time. (Well, 22/100 and 78/100 for some 2-bit deltas.)
91
+mix() was built out of 36 single-cycle latency instructions in a
92
+ structure that could supported 2x parallelism, like so:
94
+ a -= c; x = (c>>13);
98
+ c -= b; x = (b>>13);
100
+ Unfortunately, superscalar Pentiums and Sparcs can't take advantage
101
+ of that parallelism. They've also turned some of those single-cycle
102
+ latency instructions into multi-cycle latency instructions. Still,
103
+ this is the fastest good hash I could find. There were about 2^^68
104
+ to choose from. I only looked at a billion or so.
105
+--------------------------------------------------------------------
107
+#define mix(a,b,c) \
109
+ a -= b; a -= c; a ^= (c>>13); \
110
+ b -= c; b -= a; b ^= (a<<8); \
111
+ c -= a; c -= b; c ^= (b>>13); \
112
+ a -= b; a -= c; a ^= (c>>12); \
113
+ b -= c; b -= a; b ^= (a<<16); \
114
+ c -= a; c -= b; c ^= (b>>5); \
115
+ a -= b; a -= c; a ^= (c>>3); \
116
+ b -= c; b -= a; b ^= (a<<10); \
117
+ c -= a; c -= b; c ^= (b>>15); \
121
+--------------------------------------------------------------------
122
+hash() -- hash a variable-length key into a 32-bit value
123
+ k : the key (the unaligned variable-length array of bytes)
124
+ len : the length of the key, counting by bytes
125
+ initval : can be any 4-byte value
126
+Returns a 32-bit value. Every bit of the key affects every bit of
127
+the return value. Every 1-bit and 2-bit delta achieves avalanche.
128
+About 6*len+35 instructions.
130
+The best hash table sizes are powers of 2. There is no need to do
131
+mod a prime (mod is sooo slow!). If you need less than 32 bits,
132
+use a bitmask. For example, if you need only 10 bits, do
133
+ h = (h & hashmask(10));
134
+In which case, the hash table should have hashsize(10) elements.
136
+If you are hashing n strings (ub1 **)k, do it like this:
137
+ for (i=0, h=0; i<n; ++i) h = hash( k[i], len[i], h);
139
+By Bob Jenkins, 1996. bob_jenkins@burtleburtle.net. You may use this
140
+code any way you wish, private, educational, or commercial. It's free.
142
+[On 27 May 2004, Bob Jenkins further clarified the above statement.
144
+ From: Bob Jenkins <bob_jenkins@burtleburtle.net>
145
+ Date: Thu, 27 May 2004 22:33:06 -0700
146
+ To: Margarita Manterola <marga@marga.com.ar>
147
+ Subject: Re: Hash function
149
+ The algorithm is public domain. I ask that I be referenced as the
150
+ source of the algorithm, but I can't enforce that, since being public
151
+ domain means I've reserved no rights at all.
153
+-- Branden Robinson, 2004-06-06]
155
+See http://burtleburtle.net/bob/hash/evahash.html
156
+Use for hash table lookup, or anything where one collision in 2^^32 is
157
+acceptable. Do NOT use for cryptographic purposes.
158
+--------------------------------------------------------------------
162
+hash(register ub1 *k, register ub4 length, register ub4 initval)
164
+ register ub4 a,b,c,len;
166
+ /* Set up the internal state */
168
+ a = b = 0x9e3779b9; /* the golden ratio; an arbitrary value */
169
+ c = initval; /* the previous hash value */
171
+ /*---------------------------------------- handle most of the key */
174
+ a += (k[0] +((ub4)k[1]<<8) +((ub4)k[2]<<16) +((ub4)k[3]<<24));
175
+ b += (k[4] +((ub4)k[5]<<8) +((ub4)k[6]<<16) +((ub4)k[7]<<24));
176
+ c += (k[8] +((ub4)k[9]<<8) +((ub4)k[10]<<16)+((ub4)k[11]<<24));
178
+ k += 12; len -= 12;
181
+ /*------------------------------------- handle the last 11 bytes */
183
+ switch(len) /* all the case statements fall through */
185
+ case 11: c+=((ub4)k[10]<<24);
186
+ case 10: c+=((ub4)k[9]<<16);
187
+ case 9 : c+=((ub4)k[8]<<8);
188
+ /* the first byte of c is reserved for the length */
189
+ case 8 : b+=((ub4)k[7]<<24);
190
+ case 7 : b+=((ub4)k[6]<<16);
191
+ case 6 : b+=((ub4)k[5]<<8);
193
+ case 4 : a+=((ub4)k[3]<<24);
194
+ case 3 : a+=((ub4)k[2]<<16);
195
+ case 2 : a+=((ub4)k[1]<<8);
197
+ /* case 0: nothing left to add */
200
+ /*-------------------------------------------- report the result */