No redzone created before first global

[GoogleCodeExporter]

AddressSanitizer does not seem to create a redzone before the first global 
variable. Consider the following example:

int global1 = 123;
int global2 = 456;

int main()
{
    int v = *(&global1 - 1);
    return 0;
}

ASAN does not catch the invalid read because no readzone exists before global1. 
Reading from &global2 - 1 does trigger an ASAN error report. Is this expected 
behaviour? I build with Clang 3.5 on a Ubuntu 14.04 x64 installation: clang 
-fsanitize=address -g -O0 test.c

Thanks!

Original issue reported on code.google.com by chrhaus...@googlemail.com on 23 Mar 2015 at 6:56

[GoogleCodeExporter]

yes, this is known.
We do not plan any fix (no easy one is seen)

Original comment by konstant...@gmail.com on 23 Mar 2015 at 7:32

[GoogleCodeExporter]

Thanks for clarifying.

Original comment by chrhaus...@googlemail.com on 23 Mar 2015 at 7:37

[GoogleCodeExporter]

We can insert dummy globals in asan_preinit.o and build it with 
-fsanitize=address. This would only work for executables though.

Original comment by tetra20...@gmail.com on 24 Mar 2015 at 7:06

[GoogleCodeExporter]

I don't think there is any guarantee that globals from asan_preinit.o will 
appear at the start of .bss/.data.

Original comment by euge...@google.com on 30 Mar 2015 at 10:51

[GoogleCodeExporter]

I thought linker arranges file data in the order they appear on cmdline.

Original comment by tetra20...@gmail.com on 30 Mar 2015 at 11:00

[GoogleCodeExporter]

With -fdata-sections and linker GC it may not be as simple as that.
But I guess in most cases (if not always, with the _current_ linker 
implementation) it will be at the start of the output section. In any case, a 
dummy global won't hurt.

Original comment by euge...@google.com on 30 Mar 2015 at 11:04