| // Copyright 2020 The Fuchsia Authors. All rights reserved. | |
| // Use of this source code is governed by a BSD-style license that can be | |
| // found in the LICENSE file. | |
| #include <fuchsia/appmgr/llcpp/fidl.h> | |
| #include <lib/async-loop/cpp/loop.h> | |
| #include <lib/async-loop/default.h> | |
| #include <lib/fdio/directory.h> | |
| #include <lib/fidl-async/cpp/bind.h> | |
| #include <lib/fit/function.h> | |
| #include <fbl/string.h> | |
| /* | |
| * The startup component exists to start the `appmgr` component that exists | |
| * inside the `core` realm. This arrangement solves the problem that in some | |
| * builds `appmgr` is not included. It would be more straightforward to have | |
| * `appmgr` in a direct `eager` lineage from the root component, but that would | |
| * cause the system to crash in product configurations that don't include | |
| * `appmgr` because failure to start an `eager` child is fatal to a parent. | |
| * startup is not a child of the `core` component because that component is | |
| * itself stored in pkgfs, which may not be included in the build. | |
| * | |
| * startup works by using a capability routed to it from `appmgr`. startup | |
| * connects to this capability, tries to send a request, and exits. Success or | |
| * failure of the request is irrelevant, startup is just making the component | |
| * manager resolve and start `appmgr`, if it is present. | |
| */ | |
| int main() { | |
| async::Loop loop((async::Loop(&kAsyncLoopConfigAttachToCurrentThread))); | |
| zx::channel remote; | |
| zx::channel local; | |
| zx_status_t status = zx::channel::create(0, &local, &remote); | |
| if (status != ZX_OK) { | |
| exit(-1); | |
| } | |
| auto path = fbl::String("/svc/fuchsia.appmgr.Startup"); | |
| status = fdio_service_connect(path.data(), remote.release()); | |
| if (status != ZX_OK) { | |
| // This failed, presumably appmgr is not available. | |
| return 0; | |
| } | |
| llcpp::fuchsia::appmgr::Startup::SyncClient client(std::move(local)); | |
| client.LaunchAppmgr(); | |
| // We ignore the result here. A failure probably indicates we're running on a | |
| // config that doesn't include appmgr. | |
| // | |
| // In the future we may want to wait indefinitely without exiting if | |
| // components are required to have an active client to keep them running. As | |
| // of this writing components are not actively halted if they have no | |
| // clients, but this may change. | |
| } |