In MountEmulatedStorage() don't create a mount namespace unless actually mounting.
When the zygote starts, it creates its own mount namespace in
nativeUnmountStorageOnInit(). When the zygote forks a new process, unless the
new process actually has permission to access emulated storage (and thus it
needs to be mounted), there is no reason to create another new mount namespace
in the child.
This supports the WebView zygote, which does not have CAP_SYS_ADMIN to perform
mount operations. But since it only forks isolated_app processes, which do not
have access to storage, it does not need to handle mounting.
Test: m checkbuild
Test: angler boots
Bug: 21643067
Change-Id: Ieb75cc3009ed26b7366213409d5fad836f597084
1 file changed