• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* ----------------------------------------------------------------------------
2 Copyright (c) 2019-2021, Microsoft Research, Daan Leijen
3 This is free software; you can redistribute it and/or modify it under the
4 terms of the MIT license. A copy of the license can be found in the file
5 "LICENSE" at the root of this distribution.
6 -----------------------------------------------------------------------------*/
7 #include "mimalloc.h"
8 #include "mimalloc/internal.h"
9 #include "mimalloc/prim.h"    // _mi_prim_random_buf
10 #include <string.h>       // memset
11 
12 /* ----------------------------------------------------------------------------
13 We use our own PRNG to keep predictable performance of random number generation
14 and to avoid implementations that use a lock. We only use the OS provided
15 random source to initialize the initial seeds. Since we do not need ultimate
16 performance but we do rely on the security (for secret cookies in secure mode)
17 we use a cryptographically secure generator (chacha20).
18 -----------------------------------------------------------------------------*/
19 
20 #define MI_CHACHA_ROUNDS (20)   // perhaps use 12 for better performance?
21 
22 
23 /* ----------------------------------------------------------------------------
24 Chacha20 implementation as the original algorithm with a 64-bit nonce
25 and counter: https://en.wikipedia.org/wiki/Salsa20
26 The input matrix has sixteen 32-bit values:
27 Position  0 to  3: constant key
28 Position  4 to 11: the key
29 Position 12 to 13: the counter.
30 Position 14 to 15: the nonce.
31 
32 The implementation uses regular C code which compiles very well on modern compilers.
33 (gcc x64 has no register spills, and clang 6+ uses SSE instructions)
34 -----------------------------------------------------------------------------*/
35 
rotl(uint32_t x,uint32_t shift)36 static inline uint32_t rotl(uint32_t x, uint32_t shift) {
37   return (x << shift) | (x >> (32 - shift));
38 }
39 
qround(uint32_t x[16],size_t a,size_t b,size_t c,size_t d)40 static inline void qround(uint32_t x[16], size_t a, size_t b, size_t c, size_t d) {
41   x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 16);
42   x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 12);
43   x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 8);
44   x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 7);
45 }
46 
chacha_block(mi_random_ctx_t * ctx)47 static void chacha_block(mi_random_ctx_t* ctx)
48 {
49   // scramble into `x`
50   uint32_t x[16];
51   for (size_t i = 0; i < 16; i++) {
52     x[i] = ctx->input[i];
53   }
54   for (size_t i = 0; i < MI_CHACHA_ROUNDS; i += 2) {
55     qround(x, 0, 4,  8, 12);
56     qround(x, 1, 5,  9, 13);
57     qround(x, 2, 6, 10, 14);
58     qround(x, 3, 7, 11, 15);
59     qround(x, 0, 5, 10, 15);
60     qround(x, 1, 6, 11, 12);
61     qround(x, 2, 7,  8, 13);
62     qround(x, 3, 4,  9, 14);
63   }
64 
65   // add scrambled data to the initial state
66   for (size_t i = 0; i < 16; i++) {
67     ctx->output[i] = x[i] + ctx->input[i];
68   }
69   ctx->output_available = 16;
70 
71   // increment the counter for the next round
72   ctx->input[12] += 1;
73   if (ctx->input[12] == 0) {
74     ctx->input[13] += 1;
75     if (ctx->input[13] == 0) {  // and keep increasing into the nonce
76       ctx->input[14] += 1;
77     }
78   }
79 }
80 
chacha_next32(mi_random_ctx_t * ctx)81 static uint32_t chacha_next32(mi_random_ctx_t* ctx) {
82   if (ctx->output_available <= 0) {
83     chacha_block(ctx);
84     ctx->output_available = 16; // (assign again to suppress static analysis warning)
85   }
86   const uint32_t x = ctx->output[16 - ctx->output_available];
87   ctx->output[16 - ctx->output_available] = 0; // reset once the data is handed out
88   ctx->output_available--;
89   return x;
90 }
91 
read32(const uint8_t * p,size_t idx32)92 static inline uint32_t read32(const uint8_t* p, size_t idx32) {
93   const size_t i = 4*idx32;
94   return ((uint32_t)p[i+0] | (uint32_t)p[i+1] << 8 | (uint32_t)p[i+2] << 16 | (uint32_t)p[i+3] << 24);
95 }
96 
chacha_init(mi_random_ctx_t * ctx,const uint8_t key[32],uint64_t nonce)97 static void chacha_init(mi_random_ctx_t* ctx, const uint8_t key[32], uint64_t nonce)
98 {
99   // since we only use chacha for randomness (and not encryption) we
100   // do not _need_ to read 32-bit values as little endian but we do anyways
101   // just for being compatible :-)
102   memset(ctx, 0, sizeof(*ctx));
103   for (size_t i = 0; i < 4; i++) {
104     const uint8_t* sigma = (uint8_t*)"expand 32-byte k";
105     ctx->input[i] = read32(sigma,i);
106   }
107   for (size_t i = 0; i < 8; i++) {
108     ctx->input[i + 4] = read32(key,i);
109   }
110   ctx->input[12] = 0;
111   ctx->input[13] = 0;
112   ctx->input[14] = (uint32_t)nonce;
113   ctx->input[15] = (uint32_t)(nonce >> 32);
114 }
115 
chacha_split(mi_random_ctx_t * ctx,uint64_t nonce,mi_random_ctx_t * ctx_new)116 static void chacha_split(mi_random_ctx_t* ctx, uint64_t nonce, mi_random_ctx_t* ctx_new) {
117   memset(ctx_new, 0, sizeof(*ctx_new));
118   _mi_memcpy(ctx_new->input, ctx->input, sizeof(ctx_new->input));
119   ctx_new->input[12] = 0;
120   ctx_new->input[13] = 0;
121   ctx_new->input[14] = (uint32_t)nonce;
122   ctx_new->input[15] = (uint32_t)(nonce >> 32);
123   mi_assert_internal(ctx->input[14] != ctx_new->input[14] || ctx->input[15] != ctx_new->input[15]); // do not reuse nonces!
124   chacha_block(ctx_new);
125 }
126 
127 
128 /* ----------------------------------------------------------------------------
129 Random interface
130 -----------------------------------------------------------------------------*/
131 
132 #if MI_DEBUG>1
mi_random_is_initialized(mi_random_ctx_t * ctx)133 static bool mi_random_is_initialized(mi_random_ctx_t* ctx) {
134   return (ctx != NULL && ctx->input[0] != 0);
135 }
136 #endif
137 
_mi_random_split(mi_random_ctx_t * ctx,mi_random_ctx_t * ctx_new)138 void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* ctx_new) {
139   mi_assert_internal(mi_random_is_initialized(ctx));
140   mi_assert_internal(ctx != ctx_new);
141   chacha_split(ctx, (uintptr_t)ctx_new /*nonce*/, ctx_new);
142 }
143 
_mi_random_next(mi_random_ctx_t * ctx)144 uintptr_t _mi_random_next(mi_random_ctx_t* ctx) {
145   mi_assert_internal(mi_random_is_initialized(ctx));
146   #if MI_INTPTR_SIZE <= 4
147     return chacha_next32(ctx);
148   #elif MI_INTPTR_SIZE == 8
149     return (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
150   #else
151   # error "define mi_random_next for this platform"
152   #endif
153 }
154 
155 
156 /* ----------------------------------------------------------------------------
157 To initialize a fresh random context.
158 If we cannot get good randomness, we fall back to weak randomness based on a timer and ASLR.
159 -----------------------------------------------------------------------------*/
160 
_mi_os_random_weak(uintptr_t extra_seed)161 uintptr_t _mi_os_random_weak(uintptr_t extra_seed) {
162   uintptr_t x = (uintptr_t)&_mi_os_random_weak ^ extra_seed; // ASLR makes the address random
163   x ^= _mi_prim_clock_now();
164   // and do a few randomization steps
165   uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1;
166   for (uintptr_t i = 0; i < max; i++) {
167     x = _mi_random_shuffle(x);
168   }
169   mi_assert_internal(x != 0);
170   return x;
171 }
172 
mi_random_init_ex(mi_random_ctx_t * ctx,bool use_weak)173 static void mi_random_init_ex(mi_random_ctx_t* ctx, bool use_weak) {
174   uint8_t key[32] = {0};
175   if (use_weak || !_mi_prim_random_buf(key, sizeof(key))) {
176     // if we fail to get random data from the OS, we fall back to a
177     // weak random source based on the current time
178     #if !defined(__wasi__)
179     if (!use_weak) { _mi_warning_message("unable to use secure randomness\n"); }
180     #endif
181     uintptr_t x = _mi_os_random_weak(0);
182     for (size_t i = 0; i < 8; i++) {  // key is eight 32-bit words.
183       x = _mi_random_shuffle(x);
184       ((uint32_t*)key)[i] = (uint32_t)x;
185     }
186     ctx->weak = true;
187   }
188   else {
189     ctx->weak = false;
190   }
191   chacha_init(ctx, key, (uintptr_t)ctx /*nonce*/ );
192 }
193 
_mi_random_init(mi_random_ctx_t * ctx)194 void _mi_random_init(mi_random_ctx_t* ctx) {
195   mi_random_init_ex(ctx, false);
196 }
197 
_mi_random_init_weak(mi_random_ctx_t * ctx)198 void _mi_random_init_weak(mi_random_ctx_t * ctx) {
199   mi_random_init_ex(ctx, true);
200 }
201 
_mi_random_reinit_if_weak(mi_random_ctx_t * ctx)202 void _mi_random_reinit_if_weak(mi_random_ctx_t * ctx) {
203   if (ctx->weak) {
204     _mi_random_init(ctx);
205   }
206 }
207 
208 /* --------------------------------------------------------
209 test vectors from <https://tools.ietf.org/html/rfc8439>
210 ----------------------------------------------------------- */
211 /*
212 static bool array_equals(uint32_t* x, uint32_t* y, size_t n) {
213   for (size_t i = 0; i < n; i++) {
214     if (x[i] != y[i]) return false;
215   }
216   return true;
217 }
218 static void chacha_test(void)
219 {
220   uint32_t x[4] = { 0x11111111, 0x01020304, 0x9b8d6f43, 0x01234567 };
221   uint32_t x_out[4] = { 0xea2a92f4, 0xcb1cf8ce, 0x4581472e, 0x5881c4bb };
222   qround(x, 0, 1, 2, 3);
223   mi_assert_internal(array_equals(x, x_out, 4));
224 
225   uint32_t y[16] = {
226        0x879531e0,  0xc5ecf37d,  0x516461b1,  0xc9a62f8a,
227        0x44c20ef3,  0x3390af7f,  0xd9fc690b,  0x2a5f714c,
228        0x53372767,  0xb00a5631,  0x974c541a,  0x359e9963,
229        0x5c971061,  0x3d631689,  0x2098d9d6,  0x91dbd320 };
230   uint32_t y_out[16] = {
231        0x879531e0,  0xc5ecf37d,  0xbdb886dc,  0xc9a62f8a,
232        0x44c20ef3,  0x3390af7f,  0xd9fc690b,  0xcfacafd2,
233        0xe46bea80,  0xb00a5631,  0x974c541a,  0x359e9963,
234        0x5c971061,  0xccc07c79,  0x2098d9d6,  0x91dbd320 };
235   qround(y, 2, 7, 8, 13);
236   mi_assert_internal(array_equals(y, y_out, 16));
237 
238   mi_random_ctx_t r = {
239     { 0x61707865, 0x3320646e, 0x79622d32, 0x6b206574,
240       0x03020100, 0x07060504, 0x0b0a0908, 0x0f0e0d0c,
241       0x13121110, 0x17161514, 0x1b1a1918, 0x1f1e1d1c,
242       0x00000001, 0x09000000, 0x4a000000, 0x00000000 },
243     {0},
244     0
245   };
246   uint32_t r_out[16] = {
247        0xe4e7f110, 0x15593bd1, 0x1fdd0f50, 0xc47120a3,
248        0xc7f4d1c7, 0x0368c033, 0x9aaa2204, 0x4e6cd4c3,
249        0x466482d2, 0x09aa9f07, 0x05d7c214, 0xa2028bd9,
250        0xd19c12b5, 0xb94e16de, 0xe883d0cb, 0x4e3c50a2 };
251   chacha_block(&r);
252   mi_assert_internal(array_equals(r.output, r_out, 16));
253 }
254 */
255