#include <elf.h>
#include <errno.h>
#include <fcntl.h>
-#include <linux/random.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdlib.h>
# include <linux/random.h>
#endif
+#include "alloc-util.h"
#include "fd-util.h"
#include "io-util.h"
#include "missing.h"
#include "random-util.h"
+#include "siphash24.h"
#include "time-util.h"
-#if HAS_FEATURE_MEMORY_SANITIZER
-#include <sanitizer/msan_interface.h>
-#endif
-
int rdrand(unsigned long *ret) {
+ /* So, you are a "security researcher", and you wonder why we bother with using raw RDRAND here,
+ * instead of sticking to /dev/urandom or getrandom()?
+ *
+ * Here's why: early boot. On Linux, during early boot the random pool that backs /dev/urandom and
+ * getrandom() is generally not initialized yet. It is very common that initialization of the random
+ * pool takes a longer time (up to many minutes), in particular on embedded devices that have no
+ * explicit hardware random generator, as well as in virtualized environments such as major cloud
+ * installations that do not provide virtio-rng or a similar mechanism.
+ *
+ * In such an environment using getrandom() synchronously means we'd block the entire system boot-up
+ * until the pool is initialized, i.e. *very* long. Using getrandom() asynchronously (GRND_NONBLOCK)
+ * would mean acquiring randomness during early boot would simply fail. Using /dev/urandom would mean
+ * generating many kmsg log messages about our use of it before the random pool is properly
+ * initialized. Neither of these outcomes is desirable.
+ *
+ * Thus, for very specific purposes we use RDRAND instead of either of these three options. RDRAND
+ * provides us quickly and relatively reliably with random values, without having to delay boot,
+ * without triggering warning messages in kmsg.
+ *
+ * Note that we use RDRAND only under very specific circumstances, when the requirements on the
+ * quality of the returned entropy permit it. Specifically, here are some cases where we *do* use
+ * RDRAND:
+ *
+ * • UUID generation: UUIDs are supposed to be universally unique but are not cryptographic
+ * key material. The quality and trust level of RDRAND should hence be OK: UUIDs should be
+ * generated in a way that is reliably unique, but they do not require ultimate trust into
+ * the entropy generator. systemd generates a number of UUIDs during early boot, including
+ * 'invocation IDs' for every unit spawned that identify the specific invocation of the
+ * service globally, and a number of others. Other alternatives for generating these UUIDs
+ * have been considered, but don't really work: for example, hashing uuids from a local
+ * system identifier combined with a counter falls flat because during early boot disk
+ * storage is not yet available (think: initrd) and thus a system-specific ID cannot be
+ * stored or retrieved yet.
+ *
+ * • Hash table seed generation: systemd uses many hash tables internally. Hash tables are
+ * generally assumed to have O(1) access complexity, but can deteriorate to prohibitive
+ * O(n) access complexity if an attacker manages to trigger a large number of hash
+ * collisions. Thus, systemd (as any software employing hash tables should) uses seeded
+ * hash functions for its hash tables, with a seed generated randomly. The hash tables
+ * systemd employs watch the fill level closely and reseed if necessary. This allows use of
+ * a low quality RNG initially, as long as it improves should a hash table be under attack:
+ * the attacker after all needs to to trigger many collisions to exploit it for the purpose
+ * of DoS, but if doing so improves the seed the attack surface is reduced as the attack
+ * takes place.
+ *
+ * Some cases where we do NOT use RDRAND are:
+ *
+ * • Generation of cryptographic key material 🔑
+ *
+ * • Generation of cryptographic salt values 🧂
+ *
+ * This function returns:
+ *
+ * -EOPNOTSUPP → RDRAND is not available on this system 😔
+ * -EAGAIN → The operation failed this time, but is likely to work if you try again a few
+ * times ♻
+ * -EUCLEAN → We got some random value, but it looked strange, so we refused using it.
+ * This failure might or might not be temporary. 😕
+ */
+
#if defined(__i386__) || defined(__x86_64__)
static int have_rdrand = -1;
- unsigned char err;
+ unsigned long v;
+ uint8_t success;
if (have_rdrand < 0) {
uint32_t eax, ebx, ecx, edx;
return -EOPNOTSUPP;
}
- have_rdrand = !!(ecx & (1U << 30));
+/* Compat with old gcc where bit_RDRND didn't exist yet */
+#ifndef bit_RDRND
+#define bit_RDRND (1U << 30)
+#endif
+
+ have_rdrand = !!(ecx & bit_RDRND);
}
if (have_rdrand == 0)
asm volatile("rdrand %0;"
"setc %1"
- : "=r" (*ret),
- "=qm" (err));
-
-#if HAS_FEATURE_MEMORY_SANITIZER
- __msan_unpoison(&err, sizeof(err));
-#endif
-
- if (!err)
+ : "=r" (v),
+ "=qm" (success));
+ msan_unpoison(&success, sizeof(success));
+ if (!success)
return -EAGAIN;
+ /* Apparently on some AMD CPUs RDRAND will sometimes (after a suspend/resume cycle?) report success
+ * via the carry flag but nonetheless return the same fixed value -1 in all cases. This appears to be
+ * a bad bug in the CPU or firmware. Let's deal with that and work-around this by explicitly checking
+ * for this special value (and also 0, just to be sure) and filtering it out. This is a work-around
+ * only however and something AMD really should fix properly. The Linux kernel should probably work
+ * around this issue by turning off RDRAND altogether on those CPUs. See:
+ * https://github.com/systemd/systemd/issues/11810 */
+ if (v == 0 || v == ULONG_MAX)
+ return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
+ "RDRAND returned suspicious value %lx, assuming bad hardware RNG, not using value.", v);
+
+ *ret = v;
return 0;
#else
return -EOPNOTSUPP;
bool got_some = false;
int r;
- /* Gathers some randomness from the kernel (or the CPU if the RANDOM_ALLOW_RDRAND flag is set). This call won't
- * block, unless the RANDOM_BLOCK flag is set. If RANDOM_DONT_DRAIN is set, an error is returned if the random
- * pool is not initialized. Otherwise it will always return some data from the kernel, regardless of whether
- * the random pool is fully initialized or not. */
+ /* Gathers some high-quality randomness from the kernel (or potentially mid-quality randomness from
+ * the CPU if the RANDOM_ALLOW_RDRAND flag is set). This call won't block, unless the RANDOM_BLOCK
+ * flag is set. If RANDOM_MAY_FAIL is set, an error is returned if the random pool is not
+ * initialized. Otherwise it will always return some data from the kernel, regardless of whether the
+ * random pool is fully initialized or not. If RANDOM_EXTEND_WITH_PSEUDO is set, and some but not
+ * enough better quality randomness could be acquired, the rest is filled up with low quality
+ * randomness.
+ *
+ * Of course, when creating cryptographic key material you really shouldn't use RANDOM_ALLOW_DRDRAND
+ * or even RANDOM_EXTEND_WITH_PSEUDO.
+ *
+ * When generating UUIDs it's fine to use RANDOM_ALLOW_RDRAND but not OK to use
+ * RANDOM_EXTEND_WITH_PSEUDO. In fact RANDOM_EXTEND_WITH_PSEUDO is only really fine when invoked via
+ * an "all bets are off" wrapper, such as random_bytes(), see below. */
if (n == 0)
return 0;
if (FLAGS_SET(flags, RANDOM_ALLOW_RDRAND))
- /* Try x86-64' RDRAND intrinsic if we have it. We only use it if high quality randomness is not
- * required, as we don't trust it (who does?). Note that we only do a single iteration of RDRAND here,
- * even though the Intel docs suggest calling this in a tight loop of 10 invocations or so. That's
- * because we don't really care about the quality here. We generally prefer using RDRAND if the caller
- * allows us too, since this way we won't drain the kernel randomness pool if we don't need it, as the
- * pool's entropy is scarce. */
+ /* Try x86-64' RDRAND intrinsic if we have it. We only use it if high quality randomness is
+ * not required, as we don't trust it (who does?). Note that we only do a single iteration of
+ * RDRAND here, even though the Intel docs suggest calling this in a tight loop of 10
+ * invocations or so. That's because we don't really care about the quality here. We
+ * generally prefer using RDRAND if the caller allows us to, since this way we won't upset
+ * the kernel's random subsystem by accessing it before the pool is initialized (after all it
+ * will kmsg log about every attempt to do so)..*/
for (;;) {
unsigned long u;
size_t m;
break;
} else if (errno == EAGAIN) {
- /* The kernel has no entropy whatsoever. Let's remember to use the syscall the next
- * time again though.
+ /* The kernel has no entropy whatsoever. Let's remember to use the syscall
+ * the next time again though.
*
- * If RANDOM_DONT_DRAIN is set, return an error so that random_bytes() can produce some
- * pseudo-random bytes instead. Otherwise, fall back to /dev/urandom, which we know is empty,
- * but the kernel will produce some bytes for us on a best-effort basis. */
+ * If RANDOM_MAY_FAIL is set, return an error so that random_bytes() can
+ * produce some pseudo-random bytes instead. Otherwise, fall back to
+ * /dev/urandom, which we know is empty, but the kernel will produce some
+ * bytes for us on a best-effort basis. */
have_syscall = true;
if (got_some && FLAGS_SET(flags, RANDOM_EXTEND_WITH_PSEUDO)) {
return 0;
}
- if (FLAGS_SET(flags, RANDOM_DONT_DRAIN))
+ if (FLAGS_SET(flags, RANDOM_MAY_FAIL))
return -ENODATA;
/* Use /dev/urandom instead */
return;
#if HAVE_SYS_AUXV_H
- /* The kernel provides us with 16 bytes of entropy in auxv, so let's
- * try to make use of that to seed the pseudo-random generator. It's
- * better than nothing... */
+ /* The kernel provides us with 16 bytes of entropy in auxv, so let's try to make use of that to seed
+ * the pseudo-random generator. It's better than nothing... But let's first hash it to make it harder
+ * to recover the original value by watching any pseudo-random bits we generate. After all the
+ * AT_RANDOM data might be used by other stuff too (in particular: ASLR), and we probably shouldn't
+ * leak the seed for that. */
- auxv = (const void*) getauxval(AT_RANDOM);
+ auxv = ULONG_TO_PTR(getauxval(AT_RANDOM));
if (auxv) {
- assert_cc(sizeof(x) <= 16);
- memcpy(&x, auxv, sizeof(x));
+ static const uint8_t auxval_hash_key[16] = {
+ 0x92, 0x6e, 0xfe, 0x1b, 0xcf, 0x00, 0x52, 0x9c, 0xcc, 0x42, 0xcf, 0xdc, 0x94, 0x1f, 0x81, 0x0f
+ };
+
+ x = (unsigned) siphash24(auxv, 16, auxval_hash_key);
} else
#endif
x = 0;
void pseudo_random_bytes(void *p, size_t n) {
uint8_t *q;
+ /* This returns pseudo-random data using libc's rand() function. You probably never want to call this
+ * directly, because why would you use this if you can get better stuff cheaply? Use random_bytes()
+ * instead, see below: it will fall back to this function if there's nothing better to get, but only
+ * then. */
+
initialize_srand();
for (q = p; q < (uint8_t*) p + n; q += RAND_STEP) {
void random_bytes(void *p, size_t n) {
- if (genuine_random_bytes(p, n, RANDOM_EXTEND_WITH_PSEUDO|RANDOM_DONT_DRAIN|RANDOM_ALLOW_RDRAND) >= 0)
+ /* This returns high quality randomness if we can get it cheaply. If we can't because for some reason
+ * it is not available we'll try some crappy fallbacks.
+ *
+ * What this function will do:
+ *
+ * • This function will preferably use the CPU's RDRAND operation, if it is available, in
+ * order to return "mid-quality" random values cheaply.
+ *
+ * • Use getrandom() with GRND_NONBLOCK, to return high-quality random values if they are
+ * cheaply available.
+ *
+ * • This function will return pseudo-random data, generated via libc rand() if nothing
+ * better is available.
+ *
+ * • This function will work fine in early boot
+ *
+ * • This function will always succeed
+ *
+ * What this function won't do:
+ *
+ * • This function will never fail: it will give you randomness no matter what. It might not
+ * be high quality, but it will return some, possibly generated via libc's rand() call.
+ *
+ * • This function will never block: if the only way to get good randomness is a blocking,
+ * synchronous getrandom() we'll instead provide you with pseudo-random data.
+ *
+ * This function is hence great for things like seeding hash tables, generating random numeric UNIX
+ * user IDs (that are checked for collisions before use) and such.
+ *
+ * This function is hence not useful for generating UUIDs or cryptographic key material.
+ */
+
+ if (genuine_random_bytes(p, n, RANDOM_EXTEND_WITH_PSEUDO|RANDOM_MAY_FAIL|RANDOM_ALLOW_RDRAND) >= 0)
return;
/* If for some reason some user made /dev/urandom unavailable to us, or the kernel has no entropy, use a PRNG instead. */