mirror of
https://github.com/microsoft/mimalloc.git
synced 2024-12-26 21:04:27 +08:00
updated random cookie generation using OS primitives and chacha20
This commit is contained in:
parent
36d168a2d9
commit
ba87a39d9f
@ -18,6 +18,7 @@ include("cmake/mimalloc-config-version.cmake")
|
||||
|
||||
set(mi_sources
|
||||
src/stats.c
|
||||
src/random.c
|
||||
src/os.c
|
||||
src/arena.c
|
||||
src/memory.c
|
||||
@ -115,7 +116,7 @@ endif()
|
||||
|
||||
# extra needed libraries
|
||||
if(WIN32)
|
||||
list(APPEND mi_libraries psapi shell32 user32)
|
||||
list(APPEND mi_libraries psapi shell32 user32 bcrypt)
|
||||
else()
|
||||
list(APPEND mi_libraries pthread)
|
||||
find_library(LIBRT rt)
|
||||
|
@ -129,7 +129,7 @@
|
||||
<CompileAs>Default</CompileAs>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<AdditionalDependencies>$(ProjectDir)\..\..\bin\mimalloc-redirect.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<AdditionalDependencies>$(ProjectDir)\..\..\bin\mimalloc-redirect.lib;bcrypt.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<IgnoreSpecificDefaultLibraries>
|
||||
</IgnoreSpecificDefaultLibraries>
|
||||
<ModuleDefinitionFile>
|
||||
@ -195,7 +195,7 @@
|
||||
<Link>
|
||||
<EnableCOMDATFolding>true</EnableCOMDATFolding>
|
||||
<OptimizeReferences>true</OptimizeReferences>
|
||||
<AdditionalDependencies>$(ProjectDir)\..\..\bin\mimalloc-redirect.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<AdditionalDependencies>$(ProjectDir)\..\..\bin\mimalloc-redirect.lib;bcrypt.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<ModuleDefinitionFile>
|
||||
</ModuleDefinitionFile>
|
||||
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
|
||||
@ -244,6 +244,7 @@
|
||||
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\page.c" />
|
||||
<ClCompile Include="..\..\src\random.c" />
|
||||
<ClCompile Include="..\..\src\segment.c" />
|
||||
<ClCompile Include="..\..\src\stats.c" />
|
||||
</ItemGroup>
|
||||
|
@ -73,5 +73,8 @@
|
||||
<ClCompile Include="..\..\src\arena.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\random.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
</ItemGroup>
|
||||
</Project>
|
@ -229,6 +229,7 @@
|
||||
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\page.c" />
|
||||
<ClCompile Include="..\..\src\random.c" />
|
||||
<ClCompile Include="..\..\src\segment.c" />
|
||||
<ClCompile Include="..\..\src\os.c" />
|
||||
<ClCompile Include="..\..\src\stats.c" />
|
||||
|
@ -56,6 +56,9 @@
|
||||
<ClCompile Include="..\..\src\arena.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\random.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClInclude Include="$(ProjectDir)..\..\include\mimalloc.h">
|
||||
|
@ -247,6 +247,7 @@
|
||||
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\page.c" />
|
||||
<ClCompile Include="..\..\src\random.c" />
|
||||
<ClCompile Include="..\..\src\segment.c" />
|
||||
<ClCompile Include="..\..\src\stats.c" />
|
||||
</ItemGroup>
|
||||
|
@ -46,6 +46,9 @@
|
||||
<ClCompile Include="..\..\src\bitmap.inc.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\random.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClInclude Include="$(ProjectDir)..\..\include\mimalloc.h">
|
||||
|
@ -232,6 +232,7 @@
|
||||
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\page.c" />
|
||||
<ClCompile Include="..\..\src\random.c" />
|
||||
<ClCompile Include="..\..\src\segment.c" />
|
||||
<ClCompile Include="..\..\src\os.c" />
|
||||
<ClCompile Include="..\..\src\stats.c" />
|
||||
|
@ -49,6 +49,9 @@
|
||||
<ClCompile Include="..\..\src\bitmap.inc.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
<ClCompile Include="..\..\src\random.c">
|
||||
<Filter>Source Files</Filter>
|
||||
</ClCompile>
|
||||
</ItemGroup>
|
||||
<ItemGroup>
|
||||
<ClInclude Include="$(ProjectDir)..\..\include\mimalloc.h">
|
||||
|
@ -42,12 +42,17 @@ void _mi_trace_message(const char* fmt, ...);
|
||||
void _mi_options_init(void);
|
||||
void _mi_fatal_error(const char* fmt, ...) mi_attr_noreturn;
|
||||
|
||||
// "init.c"
|
||||
// random.c
|
||||
void _mi_random_init(mi_random_ctx_t* ctx);
|
||||
void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* new_ctx);
|
||||
uintptr_t _mi_random_next(mi_random_ctx_t* ctx);
|
||||
uintptr_t _mi_heap_random_next(mi_heap_t* heap);
|
||||
static inline uintptr_t _mi_random_shuffle(uintptr_t x);
|
||||
|
||||
// init.c
|
||||
extern mi_stats_t _mi_stats_main;
|
||||
extern const mi_page_t _mi_page_empty;
|
||||
bool _mi_is_main_thread(void);
|
||||
uintptr_t _mi_random_shuffle(uintptr_t x);
|
||||
uintptr_t _mi_random_init(uintptr_t seed /* can be zero */);
|
||||
bool _mi_preloading(); // true while the C runtime is not ready
|
||||
|
||||
// os.c
|
||||
@ -100,7 +105,6 @@ uint8_t _mi_bsr(uintptr_t x); // bit-scan-right, used on BSD i
|
||||
// "heap.c"
|
||||
void _mi_heap_destroy_pages(mi_heap_t* heap);
|
||||
void _mi_heap_collect_abandon(mi_heap_t* heap);
|
||||
uintptr_t _mi_heap_random(mi_heap_t* heap);
|
||||
void _mi_heap_set_default_direct(mi_heap_t* heap);
|
||||
|
||||
// "stats.c"
|
||||
@ -454,6 +458,29 @@ static inline void mi_block_set_next(const mi_page_t* page, mi_block_t* block, c
|
||||
#endif
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------
|
||||
// Fast "random" shuffle
|
||||
// -------------------------------------------------------------------
|
||||
|
||||
static inline uintptr_t _mi_random_shuffle(uintptr_t x) {
|
||||
mi_assert_internal(x!=0);
|
||||
#if (MI_INTPTR_SIZE==8)
|
||||
// by Sebastiano Vigna, see: <http://xoshiro.di.unimi.it/splitmix64.c>
|
||||
x ^= x >> 30;
|
||||
x *= 0xbf58476d1ce4e5b9UL;
|
||||
x ^= x >> 27;
|
||||
x *= 0x94d049bb133111ebUL;
|
||||
x ^= x >> 31;
|
||||
#elif (MI_INTPTR_SIZE==4)
|
||||
// by Chris Wellons, see: <https://nullprogram.com/blog/2018/07/31/>
|
||||
x ^= x >> 16;
|
||||
x *= 0x7feb352dUL;
|
||||
x ^= x >> 15;
|
||||
x *= 0x846ca68bUL;
|
||||
x ^= x >> 16;
|
||||
#endif
|
||||
return x;
|
||||
}
|
||||
|
||||
// -------------------------------------------------------------------
|
||||
// Optimize numa node access for the common case (= one node)
|
||||
|
@ -76,6 +76,7 @@ terms of the MIT license. A copy of the license can be found in the file
|
||||
#endif
|
||||
|
||||
#define MI_INTPTR_SIZE (1<<MI_INTPTR_SHIFT)
|
||||
#define MI_INTPTR_BITS (MI_INTPTR_SIZE*8)
|
||||
|
||||
#define KiB ((size_t)1024)
|
||||
#define MiB (KiB*KiB)
|
||||
@ -273,6 +274,14 @@ typedef struct mi_page_queue_s {
|
||||
|
||||
#define MI_BIN_FULL (MI_BIN_HUGE+1)
|
||||
|
||||
// Random context
|
||||
typedef struct mi_random_cxt_s {
|
||||
uint32_t input[16];
|
||||
uint32_t output[16];
|
||||
int output_available;
|
||||
} mi_random_ctx_t;
|
||||
|
||||
|
||||
// A heap owns a set of pages.
|
||||
struct mi_heap_s {
|
||||
mi_tld_t* tld;
|
||||
@ -281,7 +290,7 @@ struct mi_heap_s {
|
||||
volatile _Atomic(mi_block_t*) thread_delayed_free;
|
||||
uintptr_t thread_id; // thread this heap belongs too
|
||||
uintptr_t cookie;
|
||||
uintptr_t random; // random number used for secure allocation
|
||||
mi_random_ctx_t random; // random number used for secure allocation
|
||||
size_t page_count; // total number of pages in the `pages` queues.
|
||||
bool no_reclaim; // `true` if this heap should not reclaim abandoned pages
|
||||
};
|
||||
|
14
src/heap.c
14
src/heap.c
@ -184,12 +184,6 @@ mi_heap_t* mi_heap_get_backing(void) {
|
||||
return bheap;
|
||||
}
|
||||
|
||||
uintptr_t _mi_heap_random(mi_heap_t* heap) {
|
||||
uintptr_t r = heap->random;
|
||||
heap->random = _mi_random_shuffle(r);
|
||||
return r;
|
||||
}
|
||||
|
||||
mi_heap_t* mi_heap_new(void) {
|
||||
mi_heap_t* bheap = mi_heap_get_backing();
|
||||
mi_heap_t* heap = mi_heap_malloc_tp(bheap, mi_heap_t);
|
||||
@ -197,12 +191,16 @@ mi_heap_t* mi_heap_new(void) {
|
||||
memcpy(heap, &_mi_heap_empty, sizeof(mi_heap_t));
|
||||
heap->tld = bheap->tld;
|
||||
heap->thread_id = _mi_thread_id();
|
||||
heap->cookie = ((uintptr_t)heap ^ _mi_heap_random(bheap)) | 1;
|
||||
heap->random = _mi_heap_random(bheap);
|
||||
_mi_random_split(&bheap->random, &heap->random);
|
||||
heap->cookie = _mi_heap_random_next(heap) | 1;
|
||||
heap->no_reclaim = true; // don't reclaim abandoned pages or otherwise destroy is unsafe
|
||||
return heap;
|
||||
}
|
||||
|
||||
uintptr_t _mi_heap_random_next(mi_heap_t* heap) {
|
||||
return _mi_random_next(&heap->random);
|
||||
}
|
||||
|
||||
// zero out the page queues
|
||||
static void mi_heap_reset_pages(mi_heap_t* heap) {
|
||||
mi_assert_internal(mi_heap_is_initialized(heap));
|
||||
|
77
src/init.c
77
src/init.c
@ -85,7 +85,7 @@ const mi_heap_t _mi_heap_empty = {
|
||||
ATOMIC_VAR_INIT(NULL),
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
{ {0}, {0}, 0 },
|
||||
0,
|
||||
false
|
||||
};
|
||||
@ -116,7 +116,7 @@ mi_heap_t _mi_heap_main = {
|
||||
#else
|
||||
0xCDCDCDCDUL,
|
||||
#endif
|
||||
0, // random
|
||||
{ {0}, {0}, 0 }, // random
|
||||
0, // page count
|
||||
false // can reclaim
|
||||
};
|
||||
@ -125,66 +125,6 @@ bool _mi_process_is_initialized = false; // set to `true` in `mi_process_init`.
|
||||
|
||||
mi_stats_t _mi_stats_main = { MI_STATS_NULL };
|
||||
|
||||
/* -----------------------------------------------------------
|
||||
Initialization of random numbers
|
||||
----------------------------------------------------------- */
|
||||
|
||||
#if defined(_WIN32)
|
||||
#include <windows.h>
|
||||
#elif defined(__APPLE__)
|
||||
#include <mach/mach_time.h>
|
||||
#else
|
||||
#include <time.h>
|
||||
#endif
|
||||
|
||||
uintptr_t _mi_random_shuffle(uintptr_t x) {
|
||||
#if (MI_INTPTR_SIZE==8)
|
||||
// by Sebastiano Vigna, see: <http://xoshiro.di.unimi.it/splitmix64.c>
|
||||
x ^= x >> 30;
|
||||
x *= 0xbf58476d1ce4e5b9UL;
|
||||
x ^= x >> 27;
|
||||
x *= 0x94d049bb133111ebUL;
|
||||
x ^= x >> 31;
|
||||
#elif (MI_INTPTR_SIZE==4)
|
||||
// by Chris Wellons, see: <https://nullprogram.com/blog/2018/07/31/>
|
||||
x ^= x >> 16;
|
||||
x *= 0x7feb352dUL;
|
||||
x ^= x >> 15;
|
||||
x *= 0x846ca68bUL;
|
||||
x ^= x >> 16;
|
||||
#endif
|
||||
return x;
|
||||
}
|
||||
|
||||
uintptr_t _mi_random_init(uintptr_t seed /* can be zero */) {
|
||||
#ifdef __wasi__ // no ASLR when using WebAssembly, and time granularity may be coarse
|
||||
uintptr_t x;
|
||||
arc4random_buf(&x, sizeof x);
|
||||
#else
|
||||
// Hopefully, ASLR makes our function address random
|
||||
uintptr_t x = (uintptr_t)((void*)&_mi_random_init);
|
||||
x ^= seed;
|
||||
// xor with high res time
|
||||
#if defined(_WIN32)
|
||||
LARGE_INTEGER pcount;
|
||||
QueryPerformanceCounter(&pcount);
|
||||
x ^= (uintptr_t)(pcount.QuadPart);
|
||||
#elif defined(__APPLE__)
|
||||
x ^= (uintptr_t)mach_absolute_time();
|
||||
#else
|
||||
struct timespec time;
|
||||
clock_gettime(CLOCK_MONOTONIC, &time);
|
||||
x ^= (uintptr_t)time.tv_sec;
|
||||
x ^= (uintptr_t)time.tv_nsec;
|
||||
#endif
|
||||
// and do a few randomization steps
|
||||
uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1;
|
||||
for (uintptr_t i = 0; i < max; i++) {
|
||||
x = _mi_random_shuffle(x);
|
||||
}
|
||||
#endif
|
||||
return x;
|
||||
}
|
||||
|
||||
/* -----------------------------------------------------------
|
||||
Initialization and freeing of the thread local heaps
|
||||
@ -214,8 +154,8 @@ static bool _mi_heap_init(void) {
|
||||
mi_heap_t* heap = &td->heap;
|
||||
memcpy(heap, &_mi_heap_empty, sizeof(*heap));
|
||||
heap->thread_id = _mi_thread_id();
|
||||
heap->random = _mi_random_init(heap->thread_id);
|
||||
heap->cookie = ((uintptr_t)heap ^ _mi_heap_random(heap)) | 1;
|
||||
_mi_random_init(&heap->random);
|
||||
heap->cookie = _mi_heap_random_next(heap) | 1;
|
||||
heap->tld = tld;
|
||||
memset(tld, 0, sizeof(*tld));
|
||||
tld->heap_backing = heap;
|
||||
@ -451,16 +391,15 @@ void mi_process_init(void) mi_attr_noexcept {
|
||||
// access _mi_heap_default before setting _mi_process_is_initialized to ensure
|
||||
// that the TLS slot is allocated without getting into recursion on macOS
|
||||
// when using dynamic linking with interpose.
|
||||
mi_heap_t* h = mi_get_default_heap();
|
||||
mi_get_default_heap();
|
||||
_mi_process_is_initialized = true;
|
||||
|
||||
_mi_heap_main.thread_id = _mi_thread_id();
|
||||
_mi_verbose_message("process init: 0x%zx\n", _mi_heap_main.thread_id);
|
||||
uintptr_t random = _mi_random_init(_mi_heap_main.thread_id) ^ (uintptr_t)h;
|
||||
#ifndef __APPLE__
|
||||
_mi_heap_main.cookie = (uintptr_t)&_mi_heap_main ^ random;
|
||||
_mi_random_init(&_mi_heap_main.random);
|
||||
#ifndef __APPLE__ // TODO: fix this? cannot update cookie if allocation already happened..
|
||||
_mi_heap_main.cookie = _mi_heap_random_next(&_mi_heap_main);
|
||||
#endif
|
||||
_mi_heap_main.random = _mi_random_shuffle(random);
|
||||
mi_process_setup_auto_thread_done();
|
||||
_mi_os_init();
|
||||
#if (MI_DEBUG)
|
||||
|
@ -79,7 +79,7 @@ typedef union mi_region_info_u {
|
||||
struct {
|
||||
bool valid;
|
||||
bool is_large;
|
||||
int numa_node;
|
||||
short numa_node;
|
||||
};
|
||||
} mi_region_info_t;
|
||||
|
||||
|
8
src/os.c
8
src/os.c
@ -409,8 +409,8 @@ static void* mi_os_get_aligned_hint(size_t try_alignment, size_t size) {
|
||||
if (hint == 0 || hint > ((intptr_t)30<<40)) { // try to wrap around after 30TiB (area after 32TiB is used for huge OS pages)
|
||||
intptr_t init = ((intptr_t)4 << 40); // start at 4TiB area
|
||||
#if (MI_SECURE>0 || MI_DEBUG==0) // security: randomize start of aligned allocations unless in debug mode
|
||||
uintptr_t r = _mi_random_init((uintptr_t)&mi_os_get_aligned_hint ^ hint);
|
||||
init = init + (MI_SEGMENT_SIZE * ((r>>17) & 0xFFFF)); // (randomly 0-64k)*4MiB == 0 to 256GiB
|
||||
uintptr_t r = _mi_heap_random_next(mi_get_default_heap());
|
||||
init = init + (MI_SEGMENT_SIZE * ((r>>17) & 0xFFFFF)); // (randomly 20 bits)*4MiB == 0 to 4TiB
|
||||
#endif
|
||||
mi_atomic_cas_strong(mi_atomic_cast(uintptr_t, &aligned_base), init, hint + size);
|
||||
hint = mi_atomic_add(&aligned_base, size); // this may still give 0 or > 30TiB but that is ok, it is a hint after all
|
||||
@ -909,8 +909,8 @@ static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) {
|
||||
// Initialize the start address after the 32TiB area
|
||||
start = ((uintptr_t)32 << 40); // 32TiB virtual start address
|
||||
#if (MI_SECURE>0 || MI_DEBUG==0) // security: randomize start of huge pages unless in debug mode
|
||||
uintptr_t r = _mi_random_init((uintptr_t)&mi_os_claim_huge_pages);
|
||||
start = start + ((uintptr_t)MI_HUGE_OS_PAGE_SIZE * ((r>>17) & 0x3FF)); // (randomly 0-1024)*1GiB == 0 to 1TiB
|
||||
uintptr_t r = _mi_heap_random_next(mi_get_default_heap());
|
||||
start = start + ((uintptr_t)MI_HUGE_OS_PAGE_SIZE * ((r>>17) & 0x0FFF)); // (randomly 12bits)*1GiB == between 0 to 4TiB
|
||||
#endif
|
||||
}
|
||||
end = start + size;
|
||||
|
14
src/page.c
14
src/page.c
@ -475,11 +475,12 @@ static void mi_page_free_list_extend_secure(mi_heap_t* const heap, mi_page_t* co
|
||||
|
||||
// and initialize the free list by randomly threading through them
|
||||
// set up first element
|
||||
size_t current = _mi_heap_random(heap) % slice_count;
|
||||
const uintptr_t r = _mi_heap_random_next(heap);
|
||||
size_t current = r % slice_count;
|
||||
counts[current]--;
|
||||
mi_block_t* const free_start = blocks[current];
|
||||
// and iterate through the rest
|
||||
uintptr_t rnd = heap->random;
|
||||
// and iterate through the rest; use `random_shuffle` for performance
|
||||
uintptr_t rnd = _mi_random_shuffle(r);
|
||||
for (size_t i = 1; i < extend; i++) {
|
||||
// call random_shuffle only every INTPTR_SIZE rounds
|
||||
const size_t round = i%MI_INTPTR_SIZE;
|
||||
@ -499,8 +500,7 @@ static void mi_page_free_list_extend_secure(mi_heap_t* const heap, mi_page_t* co
|
||||
}
|
||||
// prepend to the free list (usually NULL)
|
||||
mi_block_set_next(page, blocks[current], page->free); // end of the list
|
||||
page->free = free_start;
|
||||
heap->random = _mi_random_shuffle(rnd);
|
||||
page->free = free_start;
|
||||
}
|
||||
|
||||
static mi_decl_noinline void mi_page_free_list_extend( mi_page_t* const page, const size_t extend, mi_stats_t* const stats)
|
||||
@ -608,7 +608,7 @@ static void mi_page_init(mi_heap_t* heap, mi_page_t* page, size_t block_size, mi
|
||||
mi_assert_internal(page_size / block_size < (1L<<16));
|
||||
page->reserved = (uint16_t)(page_size / block_size);
|
||||
#ifdef MI_ENCODE_FREELIST
|
||||
page->cookie = _mi_heap_random(heap) | 1;
|
||||
page->cookie = _mi_heap_random_next(heap) | 1;
|
||||
#endif
|
||||
page->is_zero = page->is_zero_init;
|
||||
|
||||
@ -710,7 +710,7 @@ static inline mi_page_t* mi_find_free_page(mi_heap_t* heap, size_t size) {
|
||||
mi_page_queue_t* pq = mi_page_queue(heap,size);
|
||||
mi_page_t* page = pq->first;
|
||||
if (page != NULL) {
|
||||
if ((MI_SECURE >= 3) && page->capacity < page->reserved && ((_mi_heap_random(heap) & 1) == 1)) {
|
||||
if ((MI_SECURE >= 3) && page->capacity < page->reserved && ((_mi_heap_random_next(heap) & 1) == 1)) {
|
||||
// in secure mode, we extend half the time to increase randomness
|
||||
mi_page_extend_free(heap, page, heap->tld);
|
||||
mi_assert_internal(mi_page_immediate_available(page));
|
||||
|
290
src/random.c
Normal file
290
src/random.c
Normal file
@ -0,0 +1,290 @@
|
||||
/* ----------------------------------------------------------------------------
|
||||
Copyright (c) 2019, Microsoft Research, Daan Leijen
|
||||
This is free software; you can redistribute it and/or modify it under the
|
||||
terms of the MIT license. A copy of the license can be found in the file
|
||||
"LICENSE" at the root of this distribution.
|
||||
-----------------------------------------------------------------------------*/
|
||||
#include "mimalloc.h"
|
||||
#include "mimalloc-internal.h"
|
||||
|
||||
#include <string.h> // memset
|
||||
|
||||
/* ----------------------------------------------------------------------------
|
||||
We use our own PRNG to keep predictable performance of random number generation
|
||||
and to avoid implementations that use a lock. We only use the OS provided
|
||||
random source to initialize the initial seeds. Since we do not need ultimate
|
||||
performance but we do rely on the security (for secret cookies in secure mode)
|
||||
we use a cryptographically secure generator (chacha20).
|
||||
-----------------------------------------------------------------------------*/
|
||||
|
||||
#define MI_CHACHA_ROUNDS (20) // perhaps use 12 for better performance?
|
||||
|
||||
|
||||
/* ----------------------------------------------------------------------------
|
||||
Chacha20 implementation as the original algorithm with a 64-bit nonce
|
||||
and counter: https://en.wikipedia.org/wiki/Salsa20
|
||||
The input matrix has sixteen 32-bit values:
|
||||
Position 0 to 3: constant key
|
||||
Position 4 to 11: the key
|
||||
Position 12 to 13: the counter.
|
||||
Position 14 to 15: the nonce.
|
||||
|
||||
The implementation uses regular C code which compiles very well on modern compilers.
|
||||
(gcc x64 has no register spills, and clang 6+ uses SSE instructions)
|
||||
-----------------------------------------------------------------------------*/
|
||||
|
||||
static inline uint32_t rotl(uint32_t x, uint32_t shift) {
|
||||
return (x << shift) | (x >> (32 - shift));
|
||||
}
|
||||
|
||||
static inline void qround(uint32_t x[16], size_t a, size_t b, size_t c, size_t d) {
|
||||
x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 16);
|
||||
x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 12);
|
||||
x[a] += x[b]; x[d] = rotl(x[d] ^ x[a], 8);
|
||||
x[c] += x[d]; x[b] = rotl(x[b] ^ x[c], 7);
|
||||
}
|
||||
|
||||
static void chacha_block(mi_random_ctx_t* r)
|
||||
{
|
||||
// scramble into `x`
|
||||
uint32_t x[16];
|
||||
for (size_t i = 0; i < 16; i++) {
|
||||
x[i] = r->input[i];
|
||||
}
|
||||
for (size_t i = 0; i < MI_CHACHA_ROUNDS; i += 2) {
|
||||
qround(x, 0, 4, 8, 12);
|
||||
qround(x, 1, 5, 9, 13);
|
||||
qround(x, 2, 6, 10, 14);
|
||||
qround(x, 3, 7, 11, 15);
|
||||
qround(x, 0, 5, 10, 15);
|
||||
qround(x, 1, 6, 11, 12);
|
||||
qround(x, 2, 7, 8, 13);
|
||||
qround(x, 3, 4, 9, 14);
|
||||
}
|
||||
|
||||
// add scrambled data to the initial state
|
||||
for (size_t i = 0; i < 16; i++) {
|
||||
r->output[i] = x[i] + r->input[i];
|
||||
}
|
||||
r->output_available = 16;
|
||||
|
||||
// increment the counter for the next round
|
||||
r->input[12] += 1;
|
||||
if (r->input[12] == 0) {
|
||||
r->input[13] += 1;
|
||||
if (r->input[13] == 0) { // and keep increasing into the nonce
|
||||
r->input[14] += 1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
static uint32_t chacha_next32(mi_random_ctx_t* r) {
|
||||
if (r->output_available <= 0) {
|
||||
chacha_block(r);
|
||||
r->output_available = 16; // (assign again to suppress static analysis warning)
|
||||
}
|
||||
r->output_available--;
|
||||
const uint32_t x = r->output[r->output_available];
|
||||
r->output[r->output_available] = 0; // reset once the data is handed out
|
||||
return x;
|
||||
}
|
||||
|
||||
static inline uint32_t read32(const uint8_t* p, size_t idx32) {
|
||||
const size_t i = 4*idx32;
|
||||
return ((uint32_t)p[i+0] | (uint32_t)p[i+1] << 8 | (uint32_t)p[i+2] << 16 | (uint32_t)p[i+3] << 24);
|
||||
}
|
||||
|
||||
static void chacha_init(mi_random_ctx_t* r, const uint8_t key[32], uint64_t nonce)
|
||||
{
|
||||
// since we only use chacha for randomness (and not encryption) we
|
||||
// do not _need_ to read 32-bit values as little endian but we do anyways
|
||||
// just for being compatible :-)
|
||||
memset(r, 0, sizeof(*r));
|
||||
for (size_t i = 0; i < 4; i++) {
|
||||
const uint8_t* sigma = (uint8_t*)"expand 32-byte k";
|
||||
r->input[i] = read32(sigma,i);
|
||||
}
|
||||
for (size_t i = 0; i < 8; i++) {
|
||||
r->input[i + 4] = read32(key,i);
|
||||
}
|
||||
r->input[12] = 0;
|
||||
r->input[13] = 0;
|
||||
r->input[14] = (uint32_t)nonce;
|
||||
r->input[15] = (uint32_t)(nonce >> 32);
|
||||
}
|
||||
|
||||
static void chacha_split(mi_random_ctx_t* r, uint64_t nonce, mi_random_ctx_t* init) {
|
||||
memset(init, 0, sizeof(*init));
|
||||
memcpy(init->input, r->input, sizeof(init->input));
|
||||
init->input[12] = 0;
|
||||
init->input[13] = 0;
|
||||
init->input[14] = (uint32_t)nonce;
|
||||
init->input[15] = (uint32_t)(nonce >> 32);
|
||||
mi_assert_internal(r->input[14] != init->input[14] || r->input[15] != init->input[15]); // do not reuse nonces!
|
||||
chacha_block(init);
|
||||
}
|
||||
|
||||
|
||||
/* ----------------------------------------------------------------------------
|
||||
Random interface
|
||||
-----------------------------------------------------------------------------*/
|
||||
|
||||
#if MI_DEBUG>1
|
||||
static bool mi_random_is_initialized(mi_random_ctx_t* ctx) {
|
||||
return (ctx != NULL && ctx->input[0] != 0);
|
||||
}
|
||||
#endif
|
||||
|
||||
void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* new_ctx) {
|
||||
mi_assert_internal(mi_random_is_initialized(ctx));
|
||||
mi_assert_internal(ctx != new_ctx);
|
||||
chacha_split(ctx, (uintptr_t)new_ctx /*nonce*/, new_ctx);
|
||||
}
|
||||
|
||||
uintptr_t _mi_random_next(mi_random_ctx_t* ctx) {
|
||||
mi_assert_internal(mi_random_is_initialized(ctx));
|
||||
#if MI_INTPTR_SIZE <= 4
|
||||
return chacha_next32(ctx);
|
||||
#elif MI_INTPTR_SIZE == 8
|
||||
return (((uintptr_t)chacha_next32(ctx) << 32) | chacha_next32(ctx));
|
||||
#else
|
||||
# error "define mi_random_next for this platform"
|
||||
#endif
|
||||
}
|
||||
|
||||
|
||||
/* ----------------------------------------------------------------------------
|
||||
To initialize a fresh random context we rely on the OS:
|
||||
- windows: BCryptGenRandom
|
||||
- bsd,wasi: arc4random_buf
|
||||
- linux: getrandom
|
||||
If we cannot get good randomness, we fall back to weak randomness based on a timer and ASLR.
|
||||
-----------------------------------------------------------------------------*/
|
||||
|
||||
#if defined(_WIN32)
|
||||
#pragma comment (lib,"bcrypt.lib")
|
||||
#include <bcrypt.h>
|
||||
static bool os_random_buf(void* buf, size_t buf_len) {
|
||||
return (BCryptGenRandom(NULL, (PUCHAR)buf, (ULONG)buf_len, BCRYPT_USE_SYSTEM_PREFERRED_RNG) >= 0);
|
||||
}
|
||||
/*
|
||||
#define SystemFunction036 NTAPI SystemFunction036
|
||||
#include <NTSecAPI.h>
|
||||
#undef SystemFunction036
|
||||
static bool os_random_buf(void* buf, size_t buf_len) {
|
||||
RtlGenRandom(buf, (ULONG)buf_len);
|
||||
return true;
|
||||
}
|
||||
*/
|
||||
#elif defined(ANDROID) || defined(XP_DARWIN) || defined(__DragonFly__) || \
|
||||
defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || \
|
||||
defined(__wasi__)
|
||||
#include <stdlib.h>
|
||||
static bool os_random_buf(void* buf, size_t buf_len) {
|
||||
arc4random_buf(buf, buf_len);
|
||||
return true;
|
||||
}
|
||||
#elif defined(__linux__)
|
||||
#include <sys/random.h>
|
||||
static bool os_random_buf(void* buf, size_t buf_len) {
|
||||
return (getrandom(buf, buf_len, GRND_NONBLOCK) == (ssize_t)buf_len);
|
||||
}
|
||||
#else
|
||||
static bool os_random_buf(void* buf, size_t buf_len) {
|
||||
return false;
|
||||
}
|
||||
#endif
|
||||
|
||||
#if defined(_WIN32)
|
||||
#include <windows.h>
|
||||
#elif defined(__APPLE__)
|
||||
#include <mach/mach_time.h>
|
||||
#else
|
||||
#include <time.h>
|
||||
#endif
|
||||
|
||||
static uintptr_t os_random_weak(uintptr_t extra_seed) {
|
||||
uintptr_t x = (uintptr_t)&os_random_weak ^ extra_seed; // ASLR makes the address random
|
||||
#if defined(_WIN32)
|
||||
LARGE_INTEGER pcount;
|
||||
QueryPerformanceCounter(&pcount);
|
||||
x ^= (uintptr_t)(pcount.QuadPart);
|
||||
#elif defined(__APPLE__)
|
||||
x ^= (uintptr_t)mach_absolute_time();
|
||||
#else
|
||||
struct timespec time;
|
||||
clock_gettime(CLOCK_MONOTONIC, &time);
|
||||
x ^= (uintptr_t)time.tv_sec;
|
||||
x ^= (uintptr_t)time.tv_nsec;
|
||||
#endif
|
||||
// and do a few randomization steps
|
||||
uintptr_t max = ((x ^ (x >> 17)) & 0x0F) + 1;
|
||||
for (uintptr_t i = 0; i < max; i++) {
|
||||
x = _mi_random_shuffle(x);
|
||||
}
|
||||
mi_assert_internal(x != 0);
|
||||
return x;
|
||||
}
|
||||
|
||||
void _mi_random_init(mi_random_ctx_t* ctx) {
|
||||
uint8_t key[32];
|
||||
if (!os_random_buf(key, sizeof(key))) {
|
||||
// if we fail to get random data from the OS, we fall back to a
|
||||
// weak random source based on the current time
|
||||
uintptr_t x = os_random_weak(0);
|
||||
for (size_t i = 0; i < 8; i++) { // key is eight 32-bit words.
|
||||
_mi_warning_message("unable to use secure randomness\n");
|
||||
x = _mi_random_shuffle(x);
|
||||
((uint32_t*)key)[i] = (uint32_t)x;
|
||||
}
|
||||
}
|
||||
chacha_init(ctx, key, (uintptr_t)ctx /*nonce*/ );
|
||||
}
|
||||
|
||||
/* --------------------------------------------------------
|
||||
test vectors from <https://tools.ietf.org/html/rfc8439>
|
||||
----------------------------------------------------------- */
|
||||
/*
|
||||
static bool array_equals(uint32_t* x, uint32_t* y, size_t n) {
|
||||
for (size_t i = 0; i < n; i++) {
|
||||
if (x[i] != y[i]) return false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
static void chacha_test(void)
|
||||
{
|
||||
uint32_t x[4] = { 0x11111111, 0x01020304, 0x9b8d6f43, 0x01234567 };
|
||||
uint32_t x_out[4] = { 0xea2a92f4, 0xcb1cf8ce, 0x4581472e, 0x5881c4bb };
|
||||
qround(x, 0, 1, 2, 3);
|
||||
mi_assert_internal(array_equals(x, x_out, 4));
|
||||
|
||||
uint32_t y[16] = {
|
||||
0x879531e0, 0xc5ecf37d, 0x516461b1, 0xc9a62f8a,
|
||||
0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0x2a5f714c,
|
||||
0x53372767, 0xb00a5631, 0x974c541a, 0x359e9963,
|
||||
0x5c971061, 0x3d631689, 0x2098d9d6, 0x91dbd320 };
|
||||
uint32_t y_out[16] = {
|
||||
0x879531e0, 0xc5ecf37d, 0xbdb886dc, 0xc9a62f8a,
|
||||
0x44c20ef3, 0x3390af7f, 0xd9fc690b, 0xcfacafd2,
|
||||
0xe46bea80, 0xb00a5631, 0x974c541a, 0x359e9963,
|
||||
0x5c971061, 0xccc07c79, 0x2098d9d6, 0x91dbd320 };
|
||||
qround(y, 2, 7, 8, 13);
|
||||
mi_assert_internal(array_equals(y, y_out, 16));
|
||||
|
||||
mi_random_ctx_t r = {
|
||||
{ 0x61707865, 0x3320646e, 0x79622d32, 0x6b206574,
|
||||
0x03020100, 0x07060504, 0x0b0a0908, 0x0f0e0d0c,
|
||||
0x13121110, 0x17161514, 0x1b1a1918, 0x1f1e1d1c,
|
||||
0x00000001, 0x09000000, 0x4a000000, 0x00000000 },
|
||||
{0},
|
||||
0
|
||||
};
|
||||
uint32_t r_out[16] = {
|
||||
0xe4e7f110, 0x15593bd1, 0x1fdd0f50, 0xc47120a3,
|
||||
0xc7f4d1c7, 0x0368c033, 0x9aaa2204, 0x4e6cd4c3,
|
||||
0x466482d2, 0x09aa9f07, 0x05d7c214, 0xa2028bd9,
|
||||
0xd19c12b5, 0xb94e16de, 0xe883d0cb, 0x4e3c50a2 };
|
||||
chacha_block(&r);
|
||||
mi_assert_internal(array_equals(r.output, r_out, 16));
|
||||
}
|
||||
*/
|
@ -14,6 +14,7 @@ terms of the MIT license. A copy of the license can be found in the file
|
||||
// it will override all the standard library allocation
|
||||
// functions (on Unix's).
|
||||
#include "stats.c"
|
||||
#include "random.c"
|
||||
#include "os.c"
|
||||
#include "arena.c"
|
||||
#include "memory.c"
|
||||
|
Loading…
x
Reference in New Issue
Block a user