linux: enable PIE (ASLR) for chrome executable

linux: enable PIE (ASLR) for chrome executable

I had been concerned that this wouldn't work with our -fPIC
bits we build internally, but the difference is actually a
performance once, not a correctness one.

(Retry of r85286, with ARM and ia32 disabled.)

BUG=57908
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=85437

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=86333

[Chris Evans, 2011-05-12 19:21:40]

On 2011/05/12 19:14:33, Evan Martin wrote:

LGTM

One allegation is that there might be a performance concern, possibly different
on 32-bit vs 64-bit too. Why not do a land / unland pair to cycle the main bots
and see if anything goes red?

[Evan Martin, 2011-05-12 19:23:35]

I belive we already build everything with -fPIC to make the PDF library work. 
But I will definitely watch the bots.

[raymes, 2011-05-12 19:24:46]

You should also be compiling with -fPIE.

On Thu, May 12, 2011 at 12:14 PM,  <evan@chromium.org> wrote:
> Reviewers: Rahul Chaturvedi, raymes, Chris Evans, agl,
>
> Description:
> linux: enable PIE (ASLR) for chrome executable
>
> I had been concerned that this wouldn't work with our -fPIC
> bits we build internally, but the difference is actually a
> performance once, not a correctness one.
>
> BUG=57908
>
>
> Please review this at http://codereview.chromium.org/6973032/
>
> SVN Base: svn://svn.chromium.org/chrome/trunk/src
>
> Affected files:
>  M chrome/chrome_exe.gypi
>
>
> Index: chrome/chrome_exe.gypi
> diff --git a/chrome/chrome_exe.gypi b/chrome/chrome_exe.gypi
> index
>
a2f0a4991d348ee8a9380a8c896ddc0438f2b96e..2f6d8a57c5e8b65ddff1b8ab304f3c6bdc5a2218
> 100644
> --- a/chrome/chrome_exe.gypi
> +++ b/chrome/chrome_exe.gypi
> @@ -199,6 +199,8 @@
>             # pie executables. Currently -pie breaks remote debugging.
>             ['disable_pie==1', {
>               'ldflags' : ['-nopie'],
> +            }, {
> +              'ldflags' : ['-pie'],
>             }],
>             ['use_system_xdg_utils==0', {
>               'copies': [
>
>
>

[rkc, 2011-05-12 19:27:10]

On 2011/05/12 19:23:35, Evan Martin wrote:
> I belive we already build everything with -fPIC to make the PDF library work. 
> But I will definitely watch the bots.

So I am not sure I understand the motivation for this code. The flag itself is a
hack that will hopefully be removed soon. The idea is to not enable or disable
pie with a flag, but to specifically disable it _if the flag is used.

[Evan Martin, 2011-05-12 19:29:14]

On 2011/05/12 19:24:46, raymes wrote:
> You should also be compiling with -fPIE.

We compile with -fPIC for complicated reasons (some subset of objects need to be
built into a shared library and our build system doesn't support building them
separately).

I experimented with this locally (building some -fPIC objects, some .so, linking
them together into a -pie executable) and verified I do get ASLR.  Do you know
if it will break anything?

[Evan Martin, 2011-05-12 19:30:07]

On 2011/05/12 19:27:10, Rahul Chaturvedi wrote:
> So I am not sure I understand the motivation for this code. The flag itself is
a
> hack that will hopefully be removed soon. The idea is to not enable or disable
> pie with a flag, but to specifically disable it _if the flag is used.

Since the flag defaults to 0, the "else" branch is currently always taken.  If
we eventually remove the flag to disable it, this code can be adjusted to
unconditionally apply the flag.

[raymes, 2011-05-12 19:35:53]

I think it will still work but it might limit some gcc optimizations.
Cary knows about the subtleties, lets check with him before we go
ahead.

> We compile with -fPIC for complicated reasons (some subset of objects need
> to be
> built into a shared library and our build system doesn't support building
> them
> separately).
>
> I experimented with this locally (building some -fPIC objects, some .so,
> linking
> them together into a -pie executable) and verified I do get ASLR.  Do you
> know
> if it will break anything?C

On Thu, May 12, 2011 at 12:29 PM,  <evan@chromium.org> wrote:
> On 2011/05/12 19:24:46, raymes wrote:
>>
>> You should also be compiling with -fPIE.
>
> We compile with -fPIC for complicated reasons (some subset of objects need
> to be
> built into a shared library and our build system doesn't support building
> them
> separately).
>
> I experimented with this locally (building some -fPIC objects, some .so,
> linking
> them together into a -pie executable) and verified I do get ASLR.  Do you
> know
> if it will break anything?
>
> http://codereview.chromium.org/6973032/
>

[Evan Martin, 2011-05-12 19:38:09]

On 2011/05/12 19:35:53, raymes wrote:
> I think it will still work but it might limit some gcc optimizations.

Yes, this is what I concluded from reading the (short, uninformative) resources
I found online.  I figure we're already paying for -fPIC currently so adding
-pie here doesn't make things any *worse* -- as long as correctness isn't
harmed.

> Cary knows about the subtleties, lets check with him before we go
> ahead.
> 
> > We compile with -fPIC for complicated reasons (some subset of objects need
> > to be
> > built into a shared library and our build system doesn't support building
> > them
> > separately).
> >
> > I experimented with this locally (building some -fPIC objects, some .so,
> > linking
> > them together into a -pie executable) and verified I do get ASLR.  Do
you
> > know
> > if it will break anything?C
> 
> On Thu, May 12, 2011 at 12:29 PM,  <mailto:evan@chromium.org> wrote:
> > On 2011/05/12 19:24:46, raymes wrote:
> >>
> >> You should also be compiling with -fPIE.
> >
> > We compile with -fPIC for complicated reasons (some subset of objects need
> > to be
> > built into a shared library and our build system doesn't support building
> > them
> > separately).
> >
> > I experimented with this locally (building some -fPIC objects, some .so,
> > linking
> > them together into a -pie executable) and verified I do get ASLR. &nbsp;Do
you
> > know
> > if it will break anything?
> >
> > http://codereview.chromium.org/6973032/
> >

[raymes, 2011-05-12 21:25:52]

I just verified with Cary that there shouldn't be any issue building with fPIC
over fPIE.

LGTM with nit comment.

Thanks.

http://codereview.chromium.org/6973032/diff/1/chrome/chrome_exe.gypi
File chrome/chrome_exe.gypi (right):

http://codereview.chromium.org/6973032/diff/1/chrome/chrome_exe.gypi#newcode199
chrome/chrome_exe.gypi:199: # pie executables. Currently -pie breaks remote
debugging.
Lets remove this TODO then?

[Evan Martin, 2011-05-18 17:12:29]

PTAL, gonna enable only for x64 right now until we figure out the other
problems.

[raymes, 2011-05-21 23:51:01]

LGTM

(Sorry for the delay, didn't realise you wanted another review.)

On Wed, May 18, 2011 at 10:12 AM,  <evan@chromium.org> wrote:
> PTAL, gonna enable only for x64 right now until we figure out the other
> problems.
>
> http://codereview.chromium.org/6973032/
>