8e8f250422
We switched from thread-local storage to a mutex-pool in 82639e6f53
because, for highly-threaded processes, the memory used by all the
states could be quite large. I had judged that a mutex-pool should be
fine, but had underestimated the PRNG requirements of some of our jobs.
This change makes rand.c support using either thread-locals or a
mutex-pool. Thread-locals are used if fork-unsafe buffering is enabled.
While not strictly related to fork-safety, we already have the
fork-unsafe control, and it's already set by jobs that care a lot about
PRNG performance, so fits quite nicely here.
Change-Id: Iaf1e0171c70d4c8dbe1e42283ea13df5b613cb2d
Reviewed-on: https://boringssl-review.googlesource.com/c/31564
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
440 lines
15 KiB
C
440 lines
15 KiB
C
/* Copyright (c) 2014, Google Inc.
|
|
*
|
|
* Permission to use, copy, modify, and/or distribute this software for any
|
|
* purpose with or without fee is hereby granted, provided that the above
|
|
* copyright notice and this permission notice appear in all copies.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
|
|
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
|
|
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
|
|
* SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
|
|
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
|
|
* OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
|
|
* CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
|
|
|
|
#include <openssl/rand.h>
|
|
|
|
#include <assert.h>
|
|
#include <limits.h>
|
|
#include <string.h>
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
#include <unistd.h>
|
|
#endif
|
|
|
|
#include <openssl/chacha.h>
|
|
#include <openssl/cpu.h>
|
|
#include <openssl/mem.h>
|
|
|
|
#include "internal.h"
|
|
#include "../../internal.h"
|
|
#include "../delocate.h"
|
|
|
|
|
|
// It's assumed that the operating system always has an unfailing source of
|
|
// entropy which is accessed via |CRYPTO_sysrand|. (If the operating system
|
|
// entropy source fails, it's up to |CRYPTO_sysrand| to abort the process—we
|
|
// don't try to handle it.)
|
|
//
|
|
// In addition, the hardware may provide a low-latency RNG. Intel's rdrand
|
|
// instruction is the canonical example of this. When a hardware RNG is
|
|
// available we don't need to worry about an RNG failure arising from fork()ing
|
|
// the process or moving a VM, so we can keep thread-local RNG state and use it
|
|
// as an additional-data input to CTR-DRBG.
|
|
//
|
|
// (We assume that the OS entropy is safe from fork()ing and VM duplication.
|
|
// This might be a bit of a leap of faith, esp on Windows, but there's nothing
|
|
// that we can do about it.)
|
|
|
|
// kReseedInterval is the number of generate calls made to CTR-DRBG before
|
|
// reseeding.
|
|
static const unsigned kReseedInterval = 4096;
|
|
|
|
// CRNGT_BLOCK_SIZE is the number of bytes in a “block” for the purposes of the
|
|
// continuous random number generator test in FIPS 140-2, section 4.9.2.
|
|
#define CRNGT_BLOCK_SIZE 16
|
|
|
|
#if defined(OPENSSL_X86_64) && !defined(OPENSSL_NO_ASM) && \
|
|
!defined(BORINGSSL_UNSAFE_DETERMINISTIC_MODE)
|
|
|
|
// These functions are defined in asm/rdrand-x86_64.pl
|
|
extern int CRYPTO_rdrand(uint8_t out[8]);
|
|
extern int CRYPTO_rdrand_multiple8_buf(uint8_t *buf, size_t len);
|
|
|
|
static int have_rdrand(void) {
|
|
return (OPENSSL_ia32cap_get()[1] & (1u << 30)) != 0;
|
|
}
|
|
|
|
static int hwrand(uint8_t *buf, const size_t len) {
|
|
if (!have_rdrand()) {
|
|
return 0;
|
|
}
|
|
|
|
const size_t len_multiple8 = len & ~7;
|
|
if (!CRYPTO_rdrand_multiple8_buf(buf, len_multiple8)) {
|
|
return 0;
|
|
}
|
|
const size_t remainder = len - len_multiple8;
|
|
|
|
if (remainder != 0) {
|
|
assert(remainder < 8);
|
|
|
|
uint8_t rand_buf[8];
|
|
if (!CRYPTO_rdrand(rand_buf)) {
|
|
return 0;
|
|
}
|
|
OPENSSL_memcpy(buf + len_multiple8, rand_buf, remainder);
|
|
}
|
|
|
|
#if defined(BORINGSSL_FIPS_BREAK_CRNG)
|
|
// This breaks the "continuous random number generator test" defined in FIPS
|
|
// 140-2, section 4.9.2, and implemented in rand_get_seed().
|
|
OPENSSL_memset(buf, 0, len);
|
|
#endif
|
|
|
|
return 1;
|
|
}
|
|
|
|
#else
|
|
|
|
static int hwrand(uint8_t *buf, size_t len) {
|
|
return 0;
|
|
}
|
|
|
|
#endif
|
|
|
|
// rand_state contains an RNG state. State object are managed in one of two
|
|
// ways, depending on whether |RAND_enable_fork_unsafe_buffering| has been
|
|
// called: if it has been called then thread-local storage is used to keep a
|
|
// per-thread state. Otherwise a mutex-protected pool of state objects is used.
|
|
struct rand_state {
|
|
CTR_DRBG_STATE drbg;
|
|
// next forms a NULL-terminated linked-list of all free |rand_state| objects
|
|
// in a pool. This is unused if using thread-local states.
|
|
struct rand_state *next;
|
|
// calls is the number of generate calls made on |drbg| since it was last
|
|
// (re)seeded. This is bound by
|
|
// |kReseedInterval - 1 + SIZE_MAX / CTR_DRBG_MAX_GENERATE_LENGTH|.
|
|
size_t calls;
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
// prev_all and next_all form another NULL-terminated linked-list, this time
|
|
// of all |rand_state| objects that have been allocated including those that
|
|
// might currently be in use.
|
|
struct rand_state *prev_all, *next_all;
|
|
// last_block contains the previous block from |CRYPTO_sysrand|.
|
|
uint8_t last_block[CRNGT_BLOCK_SIZE];
|
|
// last_block_valid is non-zero iff |last_block| contains data from
|
|
// |CRYPTO_sysrand|.
|
|
int last_block_valid;
|
|
#endif
|
|
};
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
|
|
static void rand_get_seed(struct rand_state *state,
|
|
uint8_t seed[CTR_DRBG_ENTROPY_LEN]) {
|
|
if (!state->last_block_valid) {
|
|
if (!hwrand(state->last_block, sizeof(state->last_block))) {
|
|
CRYPTO_sysrand(state->last_block, sizeof(state->last_block));
|
|
}
|
|
state->last_block_valid = 1;
|
|
}
|
|
|
|
// We overread from /dev/urandom or RDRAND by a factor of 10 and XOR to
|
|
// whiten.
|
|
#define FIPS_OVERREAD 10
|
|
uint8_t entropy[CTR_DRBG_ENTROPY_LEN * FIPS_OVERREAD];
|
|
|
|
if (!hwrand(entropy, sizeof(entropy))) {
|
|
CRYPTO_sysrand(entropy, sizeof(entropy));
|
|
}
|
|
|
|
// See FIPS 140-2, section 4.9.2. This is the “continuous random number
|
|
// generator test” which causes the program to randomly abort. Hopefully the
|
|
// rate of failure is small enough not to be a problem in practice.
|
|
if (CRYPTO_memcmp(state->last_block, entropy, CRNGT_BLOCK_SIZE) == 0) {
|
|
fprintf(stderr, "CRNGT failed.\n");
|
|
BORINGSSL_FIPS_abort();
|
|
}
|
|
|
|
for (size_t i = CRNGT_BLOCK_SIZE; i < sizeof(entropy);
|
|
i += CRNGT_BLOCK_SIZE) {
|
|
if (CRYPTO_memcmp(entropy + i - CRNGT_BLOCK_SIZE, entropy + i,
|
|
CRNGT_BLOCK_SIZE) == 0) {
|
|
fprintf(stderr, "CRNGT failed.\n");
|
|
BORINGSSL_FIPS_abort();
|
|
}
|
|
}
|
|
OPENSSL_memcpy(state->last_block,
|
|
entropy + sizeof(entropy) - CRNGT_BLOCK_SIZE,
|
|
CRNGT_BLOCK_SIZE);
|
|
|
|
OPENSSL_memcpy(seed, entropy, CTR_DRBG_ENTROPY_LEN);
|
|
|
|
for (size_t i = 1; i < FIPS_OVERREAD; i++) {
|
|
for (size_t j = 0; j < CTR_DRBG_ENTROPY_LEN; j++) {
|
|
seed[j] ^= entropy[CTR_DRBG_ENTROPY_LEN * i + j];
|
|
}
|
|
}
|
|
}
|
|
|
|
#else
|
|
|
|
static void rand_get_seed(struct rand_state *state,
|
|
uint8_t seed[CTR_DRBG_ENTROPY_LEN]) {
|
|
// If not in FIPS mode, we don't overread from the system entropy source and
|
|
// we don't depend only on the hardware RDRAND.
|
|
CRYPTO_sysrand(seed, CTR_DRBG_ENTROPY_LEN);
|
|
}
|
|
|
|
#endif
|
|
|
|
// rand_state_free_list is a list of currently free, |rand_state| structures.
|
|
// (It is only used if a mutex-pool is being used to manage |rand_state|
|
|
// objects.) When a thread needs a |rand_state| it picks the head element of
|
|
// this list and allocs a new one if the list is empty. Once it's finished, it
|
|
// pushes the state back onto the front of the list.
|
|
//
|
|
// Since we don't free |rand_state| objects, the number of objects in memory
|
|
// will eventually equal the maximum concurrency of |RAND_bytes| in the
|
|
// mutex-pool model.
|
|
DEFINE_BSS_GET(struct rand_state *, rand_state_free_list);
|
|
|
|
// rand_state_lock protects |rand_state_free_list| (and |rand_state_all_list|,
|
|
// in FIPS mode).
|
|
DEFINE_STATIC_MUTEX(rand_state_lock);
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
// rand_state_all_list is the head of a linked-list of all |rand_state| objects
|
|
// in the process. This is needed because FIPS requires that they be zeroed on
|
|
// process exit.
|
|
DEFINE_BSS_GET(struct rand_state *, rand_state_all_list);
|
|
|
|
// rand_drbg_lock is taken in write mode by |rand_state_clear_all|, and
|
|
// in read mode by any operation on the |drbg| member of |rand_state|.
|
|
// This ensures that, in the event that a thread races destructor functions, we
|
|
// never return bogus random data. At worst, the thread will deadlock.
|
|
DEFINE_STATIC_MUTEX(rand_drbg_lock);
|
|
|
|
static void rand_state_clear_all(void) __attribute__((destructor));
|
|
static void rand_state_clear_all(void) {
|
|
CRYPTO_STATIC_MUTEX_lock_write(rand_drbg_lock_bss_get());
|
|
CRYPTO_STATIC_MUTEX_lock_write(rand_state_lock_bss_get());
|
|
for (struct rand_state *cur = *rand_state_all_list_bss_get();
|
|
cur != NULL; cur = cur->next_all) {
|
|
CTR_DRBG_clear(&cur->drbg);
|
|
}
|
|
// Both locks are deliberately left locked so that any threads that are still
|
|
// running will hang if they try to call |RAND_bytes|.
|
|
}
|
|
#endif
|
|
|
|
// rand_state_free frees a |rand_state|. This is called when a thread exits if
|
|
// we're using thread-local states.
|
|
static void rand_state_free(void *state_in) {
|
|
struct rand_state *state = state_in;
|
|
if (state_in == NULL) {
|
|
return;
|
|
}
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
CRYPTO_STATIC_MUTEX_lock_write(rand_state_lock_bss_get());
|
|
if (state->prev_all != NULL) {
|
|
state->prev_all->next_all = state->next_all;
|
|
} else {
|
|
*rand_state_all_list_bss_get() = state->next_all;
|
|
}
|
|
|
|
if (state->next_all != NULL) {
|
|
state->next_all->prev_all = state->prev_all;
|
|
}
|
|
CRYPTO_STATIC_MUTEX_unlock_write(rand_state_lock_bss_get());
|
|
|
|
CTR_DRBG_clear(&state->drbg);
|
|
#endif
|
|
|
|
OPENSSL_free(state);
|
|
}
|
|
|
|
// rand_state_init seeds a |rand_state|.
|
|
static void rand_state_init(struct rand_state *state) {
|
|
OPENSSL_memset(state, 0, sizeof(struct rand_state));
|
|
uint8_t seed[CTR_DRBG_ENTROPY_LEN];
|
|
rand_get_seed(state, seed);
|
|
if (!CTR_DRBG_init(&state->drbg, seed, NULL, 0)) {
|
|
abort();
|
|
}
|
|
}
|
|
|
|
// rand_state_get returns a usable |rand_state|, or NULL if memory is exhausted.
|
|
//
|
|
// If a pool is being used, it pops a |rand_state| from the head of
|
|
// |rand_state_free_list| and returns it. If the list is empty, it
|
|
// creates a fresh |rand_state| and returns that instead.
|
|
//
|
|
// Alternatively, if thread-local states are being used, it returns the current
|
|
// thread's state object and creates it if needed.
|
|
static struct rand_state *rand_state_get(const int fork_unsafe_buffering) {
|
|
struct rand_state *state = NULL;
|
|
if (fork_unsafe_buffering) {
|
|
// Thread-local storage is used in this case. This is unrelated to fork-
|
|
// safety and we are overloading this global control to also identify
|
|
// processes that really care about PRNG speed.
|
|
state = CRYPTO_get_thread_local(OPENSSL_THREAD_LOCAL_RAND);
|
|
} else {
|
|
// Otherwise a mutex-protected pool of states is used.
|
|
CRYPTO_STATIC_MUTEX_lock_write(rand_state_lock_bss_get());
|
|
state = *rand_state_free_list_bss_get();
|
|
if (state != NULL) {
|
|
*rand_state_free_list_bss_get() = state->next;
|
|
}
|
|
CRYPTO_STATIC_MUTEX_unlock_write(rand_state_lock_bss_get());
|
|
}
|
|
|
|
if (state != NULL) {
|
|
return state;
|
|
}
|
|
|
|
state = OPENSSL_malloc(sizeof(struct rand_state));
|
|
if (state == NULL) {
|
|
return NULL;
|
|
}
|
|
|
|
rand_state_init(state);
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
CRYPTO_STATIC_MUTEX_lock_write(rand_state_lock_bss_get());
|
|
state->next_all = *rand_state_all_list_bss_get();
|
|
if (state->next_all) {
|
|
state->next_all->prev_all = state;
|
|
}
|
|
*rand_state_all_list_bss_get() = state;
|
|
CRYPTO_STATIC_MUTEX_unlock_write(rand_state_lock_bss_get());
|
|
#endif
|
|
|
|
if (fork_unsafe_buffering &&
|
|
!CRYPTO_set_thread_local(OPENSSL_THREAD_LOCAL_RAND, state,
|
|
rand_state_free)) {
|
|
rand_state_free(state);
|
|
return NULL;
|
|
}
|
|
|
|
return state;
|
|
}
|
|
|
|
// rand_state_put pushes |state| onto |rand_state_free_list| if the pool is
|
|
// being used. May only be called if the pool is being used.
|
|
static void rand_state_put(struct rand_state *state) {
|
|
CRYPTO_STATIC_MUTEX_lock_write(rand_state_lock_bss_get());
|
|
state->next = *rand_state_free_list_bss_get();
|
|
*rand_state_free_list_bss_get() = state;
|
|
CRYPTO_STATIC_MUTEX_unlock_write(rand_state_lock_bss_get());
|
|
}
|
|
|
|
void RAND_bytes_with_additional_data(uint8_t *out, size_t out_len,
|
|
const uint8_t user_additional_data[32]) {
|
|
if (out_len == 0) {
|
|
return;
|
|
}
|
|
|
|
const int fork_unsafe_buffering = rand_fork_unsafe_buffering_enabled();
|
|
|
|
// Additional data is mixed into every CTR-DRBG call to protect, as best we
|
|
// can, against forks & VM clones. We do not over-read this information and
|
|
// don't reseed with it so, from the point of view of FIPS, this doesn't
|
|
// provide “prediction resistance”. But, in practice, it does.
|
|
uint8_t additional_data[32];
|
|
if (!hwrand(additional_data, sizeof(additional_data))) {
|
|
// Without a hardware RNG to save us from address-space duplication, the OS
|
|
// entropy is used. This can be expensive (one read per |RAND_bytes| call)
|
|
// and so can be disabled by applications that we have ensured don't fork
|
|
// and aren't at risk of VM cloning.
|
|
if (!fork_unsafe_buffering) {
|
|
CRYPTO_sysrand(additional_data, sizeof(additional_data));
|
|
} else {
|
|
OPENSSL_memset(additional_data, 0, sizeof(additional_data));
|
|
}
|
|
}
|
|
|
|
for (size_t i = 0; i < sizeof(additional_data); i++) {
|
|
additional_data[i] ^= user_additional_data[i];
|
|
}
|
|
|
|
struct rand_state stack_state;
|
|
struct rand_state *state = rand_state_get(fork_unsafe_buffering);
|
|
|
|
if (state == NULL) {
|
|
// If the system is out of memory, use an ephemeral state on the
|
|
// stack.
|
|
state = &stack_state;
|
|
rand_state_init(state);
|
|
}
|
|
|
|
if (state->calls >= kReseedInterval) {
|
|
uint8_t seed[CTR_DRBG_ENTROPY_LEN];
|
|
rand_get_seed(state, seed);
|
|
#if defined(BORINGSSL_FIPS)
|
|
// Take a read lock around accesses to |state->drbg|. This is needed to
|
|
// avoid returning bad entropy if we race with
|
|
// |rand_state_clear_all|.
|
|
//
|
|
// This lock must be taken after any calls to |CRYPTO_sysrand| to avoid a
|
|
// bug on ppc64le. glibc may implement pthread locks by wrapping user code
|
|
// in a hardware transaction, but, on some older versions of glibc and the
|
|
// kernel, syscalls made with |syscall| did not abort the transaction.
|
|
CRYPTO_STATIC_MUTEX_lock_read(rand_drbg_lock_bss_get());
|
|
#endif
|
|
if (!CTR_DRBG_reseed(&state->drbg, seed, NULL, 0)) {
|
|
abort();
|
|
}
|
|
state->calls = 0;
|
|
} else {
|
|
#if defined(BORINGSSL_FIPS)
|
|
CRYPTO_STATIC_MUTEX_lock_read(rand_drbg_lock_bss_get());
|
|
#endif
|
|
}
|
|
|
|
int first_call = 1;
|
|
while (out_len > 0) {
|
|
size_t todo = out_len;
|
|
if (todo > CTR_DRBG_MAX_GENERATE_LENGTH) {
|
|
todo = CTR_DRBG_MAX_GENERATE_LENGTH;
|
|
}
|
|
|
|
if (!CTR_DRBG_generate(&state->drbg, out, todo, additional_data,
|
|
first_call ? sizeof(additional_data) : 0)) {
|
|
abort();
|
|
}
|
|
|
|
out += todo;
|
|
out_len -= todo;
|
|
// Though we only check before entering the loop, this cannot add enough to
|
|
// overflow a |size_t|.
|
|
state->calls++;
|
|
first_call = 0;
|
|
}
|
|
|
|
if (state == &stack_state) {
|
|
CTR_DRBG_clear(&state->drbg);
|
|
}
|
|
|
|
#if defined(BORINGSSL_FIPS)
|
|
CRYPTO_STATIC_MUTEX_unlock_read(rand_drbg_lock_bss_get());
|
|
#endif
|
|
|
|
if (!fork_unsafe_buffering && state != &stack_state) {
|
|
rand_state_put(state);
|
|
}
|
|
}
|
|
|
|
int RAND_bytes(uint8_t *out, size_t out_len) {
|
|
static const uint8_t kZeroAdditionalData[32] = {0};
|
|
RAND_bytes_with_additional_data(out, out_len, kZeroAdditionalData);
|
|
return 1;
|
|
}
|
|
|
|
int RAND_pseudo_bytes(uint8_t *buf, size_t len) {
|
|
return RAND_bytes(buf, len);
|
|
}
|