GitLab

Our current memwipe implementation (see #7352 (moved)) was chosen for being very-likely-to-be-safe, not for its speed. We could get much faster, probably: All we really need is a memwipe, plus some way to tell the compiler that the memwipe shouldn't be eliminated.

One GCC-only option is

void memwipe(void *p, uint8_t b, size_t sz) __attribtue__((noinline));
void memwipe(void *p, uint8_t b, size_t sz)
{
   memset(p, b, sz);
}

Another, crazier GCC-only option is

inline void 
memwipe(void *p, uint8_t b, size_t sz)
{
   memset(p, b, sz);
   asm("" ::: "memory");
}

And on windows, one could probably say

void
memwipe(void *p, uint8_t b, size_t sz)
{
   SecureZeroMemory(p, sz);
}

if b == 0.

Probably, we shouldn't bother to do any of these unless it turns out that OPENSSL_cleanse() is showing up in our profiles.