Backport changes to whitelist sockets opened by the zygote.

This is the backport of the following commits :

Commit c5f27a7c:
-----------------------------------------------

Reopen whitelisted zygote file descriptors after a fork.

We don't want these descriptors to be shared post-fork, so we'll
have to close and reopen them when the zygote forks. The set of
open descriptors is checked against a whitelist and it is a fatal
error if a non whitelisted FD is opened. It is also a fatal error
if anything other than a regular file / character device or socket
is opened at the time of forking.

This work is done in two stages :
- An initial list of FDs is constructed and cached prior to the
  first zygote fork.

- On each subsequent fork, we check whether the list of open FDs
  has changed. We are currently tolerant of changes, but in the
  longer term, it should be a fatal error if the set of open file
  descriptors in the zygote changes.

- Post fork, we traverse the list of open descriptors and reopen
  them if necessary.

bug: 30963384

Commit 3764a260:
-----------------------------------------------

Add a whitelist of sockets on fork.

Maintain a whitelist of AF_UNIX sockets that are permitted
to exist at the time of forking. If an open socket does not belong
to the whitelist (or is not AF_UNIX), the process will abort. If an
open socket is whitelisted, it will be redirected to /dev/null after
a sucessful fork. This allows us to unify our handling of the special
zygote sockets (/dev/socket/zygote[_secondary]) with the existing
whitelist of non socket file descriptors.

This change also removes non-fatal ALOGW messages since they have the
side effect of reopening the logging socket.

bug: 30963384

Commit 0b76d6a2:
-----------------------------------------------
fd_utils: Fix broken usage of iterators.

There were two separate issues here :
- RestatInternal was using an iterator after a call to erase(). This
  will not work because it will be invalidated.
- The "standard" for loop idiom for iterating over a map while making
  structural changes to it is broken. Switch to a while loop and treat
  cases where elements are erased differently from cases where they
  aren't.

bug: 31092930
bug: 30963384

Plus additional changes:
-----------------------------------------------
- change std::unordered_map to std::tr1::unordered_map.
- add /dev/alarm and /dev/__properties__ to the whitelist.
- map.erase(iterator) returns void prior to C++11, so need the kludge
  of calling erase(it++).

Change-Id: I694ff66d5f227239b0190ffc2287882b16e336fa