GitLab

=== API

The current ModernLinker loading primitives look like:
    LoadLibraryCreateRelros(path, loadAddress, libInfo);
    LoadLibraryUseRelros(path, loadAddress, fd);

This change replaces them with:
    LoadLibrary(path, loadAddress, libInfo, boolean spawnRelroRegion);
    UseRelros(libInfo);

The UseRelros() can be called *while* the library code is running,
atomically replacing the RELRO region with its version backed by shared
memory, if all stars align.

=== Motivation

The main motivation is to allow the App Zygote to provide the RELRO
region on systems with App Zygote enabled, and use the usual mechanism
when App Zygote is not available.

=== Implementation highlights

With this change the ModernLinker still waits for mLibInfo to arrive
from the browser process, as usual, but RELRO sharing already happens
after the library is loaded. I believe, there is no code path to replace
concurrently with execution.

Semantically there are a few notable changes:

1. When spawned, the RELRO FD is sealed as PROT_READ, making it
   impossible to make it writable ever again even by privileged
   processes. This way we should not worry where the code runs, in the
   Browser process or in the App Zygote.

2. Library load ranges are explicitly including the range of
   PT_GNU_RELRO segment. It should not affect anything now, but can
   potentially eliminate hard to debug crashes in distant future.

The main changes are in modern_linker_jni.cc where 'class NativeLibInfo'
is introduced to hold basically the same information as the Java-side
LibInfo, with the ability to perform the primitives described above, and
import/export internally between Java-LibInfo and NativeLibInfo.

Retrospectively the move into the class does *not* provide much useful
encapsulation at the moment, but seems like a good direction if we later
decide to use it for LegacyLinker, mock out parts for testing. The most
part I dislike most is that the class behaves like two independent parts
(LoadInfo and RelroInfo), some operations affect one, but not the other.
I did not attempt to make the split, as things are still a bit coupled
sometimes, maybe later.

OK, this was long.

Bug: 1154224
Change-Id: I88af048a298310b56b3205a2b2dd765e2c69759a
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2560308Reviewed-by: Benoit L <lizeb@chromium.org>
Commit-Queue: Egor Pasko <pasko@chromium.org>
Cr-Commit-Position: refs/heads/master@{#833245}