1. Create a new sandbox type which allows access to Unix sockets in the Mac...

1. Create a new sandbox type which allows access to Unix sockets in the Mac
renderer sandbox to support running Native Client.
2. Put the Native Client sel_ldr (which contains the user's untrusted code)
into a new Mac sandbox type.
3. Open /dev/random in SandboxWarmup().
4. Remove the "--nosandbox" flag when running Mac tests.

See http://codereview.chromium.org/1234003/show and
http://codereview.chromium.org/1525005/show which were both reverted because
of problems on Mac 10.6. This change is identical except for the ";NACL"
lines in renderer.sb files and the corresponding lines in nacl-loader.sb.
Unix socket support for the sandbox changed considerably from 10.5 to 10.6.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=327
TEST=nacl_ui_tests still pass while running in the sandbox.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=43473

[John Grabowski, 2010-04-01 22:33:08]

Could you be a little more specific about which tests you ran before sending
this up for review, and on what platforms?

[Mark Schneckloth, 2010-04-01 22:42:47]

Hi,

I borrowed someone's (personal) 10.6 laptop and ran a bunch of experiments
to determine the 10.6 flags that open the smallest hole.  (I could not find
a Google-owned 10.6 Mac that I could use.)  I then ran our Native Client
behavior tests on my 10.5 laptop and on the borrowed 10.6 laptop.  The
Native Client behavior tests are a superset of the nacl_ui_tests in Chrome.

Mark


On Thu, Apr 1, 2010 at 3:33 PM, <jrg@chromium.org> wrote:

> Could you be a little more specific about which tests you ran before
> sending
> this up for review, and on what platforms?
>
>
>
> http://codereview.chromium.org/1559012
>

[John Grabowski, 2010-04-01 23:42:13]

LGTM

I disagree with Jeremy about splitting the sandbox profile out since
cut-and-paste sucks.  I also think we'll forget to update both when we update
one.  But oh well.

[Mark Schneckloth, 2010-04-02 00:59:07]

I entered a bug for Jeremy to refactor the configs using common stuff in
included files to avoid replication.

Mark
- Sent from my phone.

On Apr 1, 2010 4:42 PM, <jrg@chromium.org> wrote:

LGTM

I disagree with Jeremy about splitting the sandbox profile out since
cut-and-paste sucks.  I also think we'll forget to update both when we
update
one.  But oh well.




http://codereview.chromium.org/1559012