Re: [PATCH 0/1] MR3379: ntdll: Match Windows used block filling.

[Paul Gofman (＠gofman)]

On Tue Jul 25 12:12:10 2023 +0000, Rémi Bernon wrote:
>     if (flags & HEAP_FREE_CHECKING_ENABLED)
>     {
>         UINT *p32, tmp = 0;
>         size = 4 + 3;
>         p = pHeapAlloc( GetProcessHeap(), 0, size );
>         ok( !!p, "HeapAlloc failed\n" );
>         p32 = (UINT *)p;
>         ok( p32[0] == 0xbaadf00d, "got %#x\n", p32[0] );
>         memcpy( &tmp, p + size - 3, 3 );
>         ok( tmp != 0xadf00d, "got %#x\n", tmp );
>         memset( p, 0xcc, size );
>         size += 2 * 4;
>         p = pHeapReAlloc( GetProcessHeap(), 0, p, size );
>         ok( !!p, "HeapReAlloc failed\n" );
>         p32 = (UINT *)p;
>         ok( p32[0] == 0xcccccccc, "got %#x\n", p32[0] );
>         ok( p32[1] << 8 == 0xcccccc00, "got %#x\n", p32[1] );
>         if (flags & HEAP_TAIL_CHECKING_ENABLED)
>             ok( p32[1] >> 24 == 0xab, "got %#x\n", p32[1] );
>         ok( p32[2] == 0xbaadf00d, "got %#x\n", p32[2] );
>         memcpy( &tmp, p + size - 3, 3 );
>         ok( tmp != 0xadf00d, "got %#x\n", tmp );
>         ret = pHeapFree( GetProcessHeap(), 0, p );
>         ok( ret, "failed.\n" );
>     }
>
> I think this would be simpler, the loops seem unnecessary.
> I removed the non-inplace check with 0xba, as I don't think it is
> reliable and hopefully doesn't matter?

I don't think that the test with 0xba matters too much, but it also seems to me that it is reliable. That is the high byte of BLOCK_FILL_USED (0xbaadf00d), and it gets there when the new block is allocated out of place and gets filled before 39 bytes are copied from the old location, so not a leftover from previous allocation at those addresses (testing those would be indeed flaky).