Re: [PATCH] x86/ptrace: Remove questionable TS_COMPAT usage in ptrace

[Pedro Alves]

On 06/18/2016 06:02 PM, Andy Lutomirski wrote:

> Yuck.  I should have dug in to the history.  Why not just
> unconditionally sign-extend eax when set by a 32-bit tracer?

No idea.

> 
> Do you know how to acquire a copy of erestartsys-trap.c?  The old
> links appear to be broken.

That's part of the ptrace testsuite project, still in cvs, though the url changed:

 $ https://sourceware.org/systemtap/wiki/utrace/tests

 $ cvs -d :pserver:anoncvs:anoncvs@xxxxxxxxxxxxxx:/cvs/systemtap co ptrace-tests

Can't seem to find a cvsweb interface for that.

I think it'd be great to move these to the selftests infrastructure
directly in the kernel tree.  However, nobody's has ever managed to
find energy for that.

> 
> Also, while I have your attention: when gdb restores old state like
> this, does it do it with individual calls to PTRACE_POKEUSER or does
> it use SETREGSET or similar to do it all at once?  I'm asking because
> I have some other code (fsgsbase) that's on hold until I can figure
> out how to keep it from breaking gdb if and when gdb writes to fs and
> fs_base.
> 

It depends on which register you're accessing, and on kernel version.
But on a recent kernel, it should be using PTRACE_SETREGS / PTRACE_SETREGSET,
thus storing a whole register set in one go.  (And it's likely we could
get rid of PTRACE_POKE fallback paths by now.)
To write to the debug registers (dr0-dr7), PTRACE_POKEUSER is always used.

There's code that coordinates with glibc's libthread_db.so that ends
up _reading_ fs_base/gs_base, and gdb uses PTRACE_PEEKUSER for that,
though there's a pending series that changes it, exposing fs_base/gs_base
as just another register in gdb's register cache:

 https://www.sourceware.org/ml/gdb-patches/2015-11/msg00076.html
 https://www.sourceware.org/ml/gdb-patches/2015-11/msg00077.html

Guess that makes fs_base/gs_base user-writable, if the kernel allows it.

Thanks,
Pedro Alves