Skip to content
Snippets Groups Projects
New look: Off

Anbox support

    • Closed Imported Issue created by Administrator @root

    Created by: PureTryOut

    Back in August, @MartijnBraam started packaging Anbox but got stuck somewhere in the process. I just rebased the branch to the latest master and updated some dependencies.

    However one of it's dependencies, dbus-cpp, fails to compile (we package it ourselves). We're guessing it's a musl problem, which might make this quite hard to resolve.

    [  4%] Building CXX object src/core/dbus/CMakeFiles/dbus-cpp.dir/fixture.cpp.o
/home/pmos/build/src/build/src/core/dbus/fixture.cpp:26:29: fatal error: core/posix/exec.h: No such file or directory
 #include <core/posix/exec.h>
                             ^
compilation terminated.
make[2]: *** [src/core/dbus/CMakeFiles/dbus-cpp.dir/build.make:63: src/core/dbus/CMakeFiles/dbus-cpp.dir/fixture.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:1451: src/core/dbus/CMakeFiles/dbus-cpp.dir/all] Error 2
make: *** [Makefile:141: all] Error 2
    To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information

    Child items ...

    • Activity

    • Administrator
      Administrator @root · Imported
      Author Owner

      A quick apt-file search core/posix/exec.h querry on debian suggests that the dependency process-cpp is missing

      By Jonah Brüchert on 2018-03-04T10:38:56

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: PureTryOut

      Good call, thanks for that!

      Some new errors though, this time in process-cpp :unamused:

      [ 11%] Building CXX object src/CMakeFiles/process-cpp.dir/core/posix/child_process.cpp.o
In file included from /usr/include/fortify/stdio.h:20:0,
                 from /usr/include/c++/6.4.0/cstdio:42,
                 from /usr/include/c++/6.4.0/ext/string_conversions.h:43,
                 from /usr/include/c++/6.4.0/bits/basic_string.h:5417,
                 from /usr/include/c++/6.4.0/string:52,
                 from /usr/include/c++/6.4.0/stdexcept:39,
                 from /usr/include/c++/6.4.0/array:39,
                 from /usr/include/c++/6.4.0/tuple:39,
                 from /usr/include/c++/6.4.0/functional:55,
                 from /usr/include/core/connection.h:21,
                 from /usr/include/core/signal.h:21,
                 from /home/pmos/build/src/process-cpp-3.0.1/include/core/posix/signal.h:24,
                 from /home/pmos/build/src/process-cpp-3.0.1/include/core/posix/signalable.h:22,
                 from /home/pmos/build/src/process-cpp-3.0.1/include/core/posix/process_group.h:22,
                 from /home/pmos/build/src/process-cpp-3.0.1/include/core/posix/process.h:22,
                 from /home/pmos/build/src/process-cpp-3.0.1/include/core/posix/child_process.h:22,
                 from /home/pmos/build/src/process-cpp-3.0.1/src/core/posix/child_process.cpp:19:
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:36:5: error: expected identifier before '(' token
     stdin = 1 << 0,
     ^
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:36:5: error: expected '}' before '(' token
In file included from /home/pmos/build/src/process-cpp-3.0.1/include/core/posix/child_process.h:23:0,
                 from /home/pmos/build/src/process-cpp-3.0.1/src/core/posix/child_process.cpp:19:
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:36:18: error: cannot convert 'int' to 'core::posix::StandardStream' in initialization
     stdin = 1 << 0,
                  ^
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:37:19: error: cannot convert 'int' to 'core::posix::StandardStream' in initialization
     stdout = 1 << 1,
                   ^
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:38:19: error: cannot convert 'int' to 'core::posix::StandardStream' in initialization
     stderr = 1 << 2
                   ^
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:39:2: error: extra ';' [-Werror=pedantic]
 };
  ^
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:41:23: error: 'StandardStream' does not name a type
 CORE_POSIX_DLL_PUBLIC StandardStream operator|(StandardStream l, StandardStream r);
                       ^~~~~~~~~~~~~~
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:42:23: error: 'StandardStream' does not name a type
 CORE_POSIX_DLL_PUBLIC StandardStream operator&(StandardStream l, StandardStream r);
                       ^~~~~~~~~~~~~~
/home/pmos/build/src/process-cpp-3.0.1/include/core/posix/standard_stream.h:44:1: error: expected declaration before '}' token
 }
 ^
cc1plus: all warnings being treated as errors
make[2]: *** [src/CMakeFiles/process-cpp.dir/build.make:87: src/CMakeFiles/process-cpp.dir/core/posix/child_process.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:1036: src/CMakeFiles/process-cpp.dir/all] Error 2
make: *** [Makefile:141: all] Error 2

      By * postmarketOS Bot on 2018-03-04T11:09:17

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      could you share the APKBUILD?

      EDIT: never mind, it's in the feature/anbox branch. If someone wants to join in with fixing the build errors:

      git checkout -b feature/anbox origin/feature/anbox
pmbootstrap build process-cpp
# build fails but downloads sources
cd ~/code
tar -xvf ~/.local/var/pmbootstrap/cache_distfiles/process-cpp_3.0.1.orig.tar.gz
cd process-cpp-3.0.1
git init .
git add -A
git commit -m "initial"

      Then repeat until everything is fixed:

      # fix something
git commit -m "fixed xyz"
pmbootstrap build process-cpp --src=.

      By * postmarketOS Bot on 2018-03-04T12:27:12

      Edited by Ghost User
    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      I don't really know how to fix it:

      .../include/core/posix/standard_stream.h:36:5: error: expected identifier before '(' token
     stdin = 1 << 0,
     ^

      There is no ( character in that line, gcc! My guess is that this comes from a macro?

      I've also looked at Debian's patches for this, but nothing that helps with this issue:

      debian/patches/
debian/patches/0001-Don-t-run-tests.patch
debian/patches/0002-Reproducible-documentation.patch

      @PureTryOut: Do you want to try asking about that on stack overflow or C++ IRC/matrix channels?

      By * postmarketOS Bot on 2018-03-04T13:26:15

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: scintill

      It looks like if I comment out these lines in /usr/include/stdio.h, some of the errors go away:

      #define stdin  (stdin)
#define stdout (stdout)
#define stderr (stderr)

      That would be the macro inserting a ( as @ollieparanoid surmised. Apparently the standard does say these symbols should be macros, but my GNU libc stdio.h defines them without parens. The parens muck up the enum definition, I think.

      Maybe the best solution is inserting your own stdio.h into process-cpp's internal include path (don't install it to the system), which does something like this (untested):

      #pragma once

#include_next <stdio.h>
#undef stdin
#undef stdout
#undef stderr
#define stdin  stdin
#define stdout stdout
#define stderr stderr

      When the parens are gone, I don't think there will be adverse effects, but that enum at standard_stream.h:36 should compile.

      By * postmarketOS Bot on 2018-03-05T23:53:41

      Edited by Ghost User
    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      Thanks for the hint @scintill, that helped so much! I ended up adding 3 #undef line just before the namespace that failed to compile in standard_stream.h and that did the trick. With linking against execinfo to fix undefined reference errors for backtrace stuff (which is present in glibc, but not in musl) and fixing one uint -> unsigned int, the process-cpp package compiles now.

      dbus-cpp was an easy fix as well, I only needed to change the subpackages. Now it starts compiling anbox and fails because lxc is missing or something. I've rebased the branch and applied the fixes.

      @PureTryOut, do you want to give it another try?

      By * postmarketOS Bot on 2018-03-06T20:13:18

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: PureTryOut

      Awesome job! Up to the next error, we're close now!

      [100%] Linking CXX executable anbox
/usr/lib/gcc/x86_64-alpine-linux-musl/6.4.0/../../../../x86_64-alpine-linux-musl/bin/ld: cannot open output file anbox: Is a directory
collect2: error: ld returned 1 exit status

      Note that I disabled unit tests entirely, as it requires a framework called gmock which I failed to get to install as of yet. I might look into it again later, but I feel it isn't as important.

      By * postmarketOS Bot on 2018-03-07T11:44:54

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: PureTryOut

      @ollieparanoid suggested trying to build outside of the source directory. That did seem to resolve above linking error, but a new one occurs.

      [100%] Linking CXX executable anbox
/usr/lib/gcc/x86_64-alpine-linux-musl/6.4.0/../../../../x86_64-alpine-linux-musl/bin/ld: ../external/process-cpp-minimal/src/libprocess-cpp.a(backtrace.cpp.o): undefined reference to symbol 'backtrace_symbols'
//usr/lib/libexecinfo.so.1: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
make[2]: *** [src/CMakeFiles/anbox.dir/build.make:119: src/anbox] Error 1
make[1]: *** [CMakeFiles/Makefile2:2134: src/CMakeFiles/anbox.dir/all] Error 2
make: *** [Makefile:130: all] Error 2

      Also, using -j1 (which I did for above) gives me vastly different results than not specifying -j at all. When doing the latter I get the following:

      [ 31%] Linking CXX static library librenderControl_dec.a
In file included from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLDispatch.h:23:0,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLEScontext.h:20,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/GLcommon/GLEScontext.cpp:17:
/home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/include/OpenGLESDispatch/gles_functions.h:18:56: fatal error: OpenGLESDispatch/gles_extensions_functions.h: No such file or directory
 #include "OpenGLESDispatch/gles_extensions_functions.h"
                                                        ^
compilation terminated.
make[2]: *** [external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/build.make:111: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/GLEScontext.cpp.o] Error 1
make[2]: *** Waiting for unfinished jobs....
In file included from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLDispatch.h:23:0,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/GLcommon/GLDispatch.cpp:17:
/home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/include/OpenGLESDispatch/gles_functions.h:18:56: fatal error: OpenGLESDispatch/gles_extensions_functions.h: No such file or directory
 #include "OpenGLESDispatch/gles_extensions_functions.h"
                                                        ^
compilation terminated.
In file included from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLDispatch.h:23:0,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLEScontext.h:20,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLESvalidate.h:20,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/GLcommon/GLESvalidate.cpp:17:
/home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/include/OpenGLESDispatch/gles_functions.h:18:56: fatal error: OpenGLESDispatch/gles_extensions_functions.h: No such file or directory
 #include "OpenGLESDispatch/gles_extensions_functions.h"
                                                        ^
compilation terminated.
make[2]: *** [external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/build.make:63: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/GLDispatch.cpp.o] Error 1
make[2]: *** [external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/build.make:135: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/GLESvalidate.cpp.o] Error 1
In file included from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLDispatch.h:23:0,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLEScontext.h:20,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/TextureUtils.h:21,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/GLcommon/TextureUtils.cpp:16:
/home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/include/OpenGLESDispatch/gles_functions.h:18:56: fatal error: OpenGLESDispatch/gles_extensions_functions.h: No such file or directory
 #include "OpenGLESDispatch/gles_extensions_functions.h"
                                                        ^
compilation terminated.
make[2]: *** [external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/build.make:231: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/TextureUtils.cpp.o] Error 1
[ 31%] Built target renderControl_dec
Generating /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/build/external/android-emugl/host/include/OpenGLESDispatch/gles1_only_functions.h
[ 31%] Built target xdg_test
Generating /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/build/external/android-emugl/host/include/OpenGLESDispatch/gles1_extensions_functions.h
Generating /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/build/external/android-emugl/host/include/OpenGLESDispatch/gles2_only_functions.h
In file included from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLDispatch.h:23:0,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLEScontext.h:20,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/GLcommon/objectNameManager.cpp:18:
/home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/include/OpenGLESDispatch/gles_functions.h:21:51: fatal error: OpenGLESDispatch/gles2_only_functions.h: No such file or directory
 #include "OpenGLESDispatch/gles2_only_functions.h"
                                                   ^
compilation terminated.
make[2]: *** [external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/build.make:303: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/objectNameManager.cpp.o] Error 1
Generating /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/build/external/android-emugl/host/include/OpenGLESDispatch/gles2_extensions_functions.h
In file included from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLDispatch.h:23:0,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/include/GLcommon/GLEScontext.h:20,
                 from /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/libs/Translator/GLcommon/FramebufferData.cpp:19:
/home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/external/android-emugl/host/include/OpenGLESDispatch/gles_functions.h:22:57: fatal error: OpenGLESDispatch/gles2_extensions_functions.h: No such file or directory
 #include "OpenGLESDispatch/gles2_extensions_functions.h"
                                                         ^
compilation terminated.
make[2]: *** [external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/build.make:327: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/FramebufferData.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:1573: external/android-emugl/host/libs/Translator/GLcommon/CMakeFiles/GLcommon.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
Generating /home/pmos/build/src/anbox-03f69c8685e8e05a8c296896431d0645da831092/build/external/android-emugl/host/include/OpenGLESDispatch/gles3_only_functions.h
[ 32%] Linking CXX static library libanbox-protobuf.a
[ 32%] Built target GLHeaders
[ 32%] Built target anbox-protobuf
make: *** [Makefile:130: all] Error 2

      By * postmarketOS Bot on 2018-03-08T09:43:04

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      @PureTryOut: could you upload your latest version uploaded to the feature/anbox branch? I've just tried it, but I hit the cannot open output file anbox: Is a directory error which is already solved in your latest post.

      By * postmarketOS Bot on 2018-03-13T21:54:43

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: PureTryOut

      @ollieparanoid done. Also rebased to latest master for good measure.

      By * postmarketOS Bot on 2018-03-13T22:02:01

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      Thanks! You were pretty close already, with some more fixing it compiles now :smiley: Back to you, @PureTryOut!

      By * postmarketOS Bot on 2018-03-13T23:41:34

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      @PureTryOut: what was the status on this? In case we could get it working in time, it would be nice for the 1 year blog post as well.

      By * postmarketOS Bot on 2018-04-09T20:34:13

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: PureTryOut

      That would be awesome indeed. I believe there are just a few things left to be done:

      • write an OpenRC service file, so far there are only systemd ones out there
      • write a .desktop file
      • create a basic Android image to use in Anbox

      I don't have any experience with OpenRC init files really. I could learn how to do it, but it might be better if someone else tried. We can take the .desktop file from Arch here. For the Android image we could use the pre-built one by the Anbox project, but I'm not sure if we want that (we probably do as building our own image has some insane pc requirements).

      EDIT: It seems I had a .desktop file locally, just uncommitted. I pushed it to the branch.

      Right now it fails to build again, it seems some Boost libraries are updated. Some C++ experts out here?

      In file included from /home/pmos/build/src/anbox-f68725cff813700e448fb3b2f3d57d88e2c7daae/src/anbox/cmds/session_manager.cpp:56:0:
/usr/include/core/dbus/asio/executor.h:29:7: error: using typedef-name 'boost::asio::io_service' after 'class'
 class io_service;
       ^~~~~~~~~~
In file included from /usr/include/boost/asio.hpp:69:0,
                 from /home/pmos/build/src/anbox-f68725cff813700e448fb3b2f3d57d88e2c7daae/src/anbox/runtime.h:21,
                 from /home/pmos/build/src/anbox-f68725cff813700e448fb3b2f3d57d88e2c7daae/src/anbox/audio/server.h:21,
                 from /home/pmos/build/src/anbox-f68725cff813700e448fb3b2f3d57d88e2c7daae/src/anbox/cmds/session_manager.cpp:26:
/usr/include/boost/asio/io_service.hpp:27:20: note: 'boost::asio::io_service' has a previous declaration here
 typedef io_context io_service;
                    ^~~~~~~~~~
make[2]: *** [src/CMakeFiles/anbox-core.dir/build.make:1363: src/CMakeFiles/anbox-core.dir/anbox/cmds/session_manager.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:2088: src/CMakeFiles/anbox-core.dir/all] Error 2
make: *** [Makefile:130: all] Error 2

      By * postmarketOS Bot on 2018-04-09T22:07:42

      Edited by Ghost User
    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      As discussed in #postmarketOS, I think it would be good to make it work with the pre-built Android images first (possibly look into building our own ones later if it makes sense). That leaves the OpenRC service and .desktop files. Could you take a shot at them?

      Build errors are fixed now, with patches from @tmn505 posted in https://github.com/anbox/anbox/issues/637 (thanks a lot for those!).

      @PureTryOut: updated the branch. make sure to force rebuild dbus-cpp first:

      $ pmbootstrap build --strict --force dbus-cpp
$ pmbootstrap build --strict anbox

      By * postmarketOS Bot on 2018-04-12T20:00:22

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      Slowly making progress, here's the first screenshot!

      anbox_wip

      That's the part running as user and it's saying in the terminal that it can't connect to the session daemon and then gives up after a few seconds, but still it's satisfying to have at least some visual :grin:

      Status:

      • Container manager starts, mounts the android image and runs in background
      • Session manager doesn't run yet: it requests a new container from container manager, and the container can't be set up because of an error coming from lxc (it would fail earlier because a fifo can not be set up, but I have a workaround for that now).

      lxc_cgfs - cgroups/cgfs.c:cgfs_init:2363 - cgroupfs failed to detect cgroup metadata

      I've added a temporary README.anbox file to the branch with instructions on how to start everything, how to debug, where the log file is and what is left to do: https://github.com/postmarketOS/pmbootstrap/blob/feature/anbox/README.anbox

      By * postmarketOS Bot on 2018-04-25T18:23:34

      Edited by Ghost User
    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: PureTryOut

      I've added a service file, some post-install and post-deinstall hooks and made the loop kernel module autoload, so step 1 in README.anbox now happens automatically on boot.

      However, I don't get to the point you have gotten @ollieparanoid (using the instructions in README.anbox). When I try to load anbox session-manager I just get the following:

      [ 2018-04-30 14:33:27] [client.cpp:49@start] Failed to start container: Failed to start container: Failed to start container
[ 2018-04-30 14:33:27] [session_manager.cpp:162@operator()] Lost connection to container manager, terminating.

      Nothing happens on the screen either. Also, the result isn't any different if I don't create the fifo file.

      By * postmarketOS Bot on 2018-04-30T15:18:23

      Edited by Ghost User
    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      Thanks for the streamlining! Yeah, the session manager does not work yet, I get the same output and from anbox session-manager. And /var/lib/anbox/logs/container.log has the real error message:

      Current fatal error in step 4: lxc_cgfs - cgroups/cgfs.c:cgfs_init:2363 - cgroupfs failed to detect cgroup metadata

      The loading screen appears, when you just ignore that and try to launch an Android app anyway (and of course it fails because session manager is not running):

      $ export DISPLAY=:0
$ anbox launch --package=org.anbox.appmgr --component=org.anbox.appmgr.AppViewActivity

      My understanding is, that the container manager uses the LXC library to set up the containers. And it passes a bunch of configuration variables to LXC (which is quite readable when looking at the anbox source actually), and one of them is lxc.cgroup.use. This variable is empty, and then in the LXC code lxc_cgroup_load_meta() fails because it is empty. I had something along the lines of that noted down, but I didn't have time yet to look into it further.

      By * postmarketOS Bot on 2018-05-01T00:37:22

    • Administrator
      Administrator @root · Imported
      Author Owner

      Created by: ollieparanoid

      Also right now we are running anbox in privileged mode, which means the container manager and session manager need root rights as I understand, but we could also make it run without root rights which would be better of course. The official snap packaging uses the non-privileged mode as well from what I can tell. In the anbox code, it passes different configuration options to LXC, so this might even avoid the error above.

      Finally, another idea for debugging I had was installing Ubuntu in a VM and running anbox there as snap (which is the supported installation method by upstream), then figuring out what the value for lxc.cgroup.use is in Ubuntu, so we can emulate it. The config options may also be set by a global config file or something, so I'm not entirely sure how it would work in Ubuntu.

      By * postmarketOS Bot on 2018-05-01T00:41:29

    • Administrator mentioned in commit pmbootstrap@7afaf8da · Imported

      mentioned in commit pmbootstrap@7afaf8da

      By Oliver Smith on 2018-08-31T20:20:50

    • Administrator moved from pmbootstrap#1286 (closed) · Imported

      moved from pmbootstrap#1286 (closed)

      By Bart Ribbers on 2018-09-08T22:31:55

    • Administrator
      Administrator @root · Imported
      Author Owner

      Anbox now uses libsystemd's sd-bus instead of dbus-cpp. We would either need to package sd-bus (possibly difficult due to musl) or convince them to use a different dbus binding/implementation.

      By Grant Miller on 2019-01-23T18:48:13

    • Administrator
      Administrator @root · Imported
      Author Owner

      Damn, that is disappointing... Besides the musl issue, is sd-bus even usable without systemd? Hopefully we can convince him to use a different dbus binding.

      By Bart Ribbers on 2019-01-23T22:51:53

    • Administrator
      Administrator @root · Imported
      Author Owner

      From the Anbox thread:

      libsystemd should be even possible on systems which don't use systemd as init system

      By Grant Miller on 2019-01-24T01:51:35

    • Administrator mentioned in merge request !174 (merged) · Imported

      mentioned in merge request !174 (merged)

      By Grant Miller on 2019-01-25T00:41:34

    • Administrator
      Administrator @root · Imported
      Author Owner

      I just rebased the feature/anbox branch on the latest master, it hadn't been touched for a year lol. Thanks for your work so far @GrantM11235, hopefully we can get Anbox running any time soon.

      By Bart Ribbers on 2019-05-10T15:13:48

    • Administrator
      Administrator @root · Imported
      Author Owner

      The last rebase left a bit of mess in dbus-cpp, process-cpp, and properties-cpp so I rebased it again to clean that up.

      Alpine now has elogind in their aports, which makes things a bit easier for us.

      I rebased wip/anbox-libelogind and merged it in to feature/anbox.

      By Grant Miller on 2019-05-11T05:01:16

    • Administrator
      Administrator @root · Imported
      Author Owner

      Thanks!

      I've noticed you added a (removed by upstream) udev rules file to basically modprobe(?) the required kernel modules. Might it not be preferable to built those modules in the kernel? If anything, we should at least make anbox-modules a separate package.

      By Bart Ribbers on 2019-05-11T05:43:24

      Edited by Administrator
    • Administrator mentioned in merge request !432 (closed) · Imported

      mentioned in merge request !432 (closed)

      By Grant Miller on 2019-06-11T15:22:53

      • Administrator
        Administrator @root · Imported
        Author Owner

        On my branch I made anbox compile again !823.

        I manage to get past the LXC "cannot start" error using:

        # this fails, as there is no container to start; but it seems to fix something related to cgroups?
sudo rc-service lxc start

# the command from `rc-service anbox-container-manager start`
sudo anbox container-manager --priviledged --data-path=/var/lib/anbox/

        (with service anbox-container-manager disabled)

        But then when I run anbox session-manager, it seems to work correctly but a lot of errors appear in /var/lib/anbox/logs/console.log (EDIT: I posted a better log latter. For reference, here is the long, old one and an extract)

        Now, anbox session-manager keeps trying to start the session, but fails. anbox launch --package=org.anbox.appmgr --component=org.anbox.appmgr.AppViewActivity waits for the session manager a bit then give up.

        It seems to go fairly far, though, as I could enable the network bridge using https://github.com/anbox/anbox/issues/270: ip addr shows a new interface vethU4WD5X@if11: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue master anbox0 state UP group default qlen 1000

        By Antoine Fontaine on 2019-12-30T22:37:06

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner

        I've definitely got post the LXC problem: the last line in /var/lib/anbox/logs/container.log is lxccontainer - lxccontainer.c:wait_on_daemonized_start:877 - Container is in "RUNNING" state. That's great!

        Now the following error seems to be the blocking one:

        12-30 22:43:20.379   105   105 E System  : ******************************************
12-30 22:43:20.380   105   105 E System  : ************ Failure starting system services
12-30 22:43:20.380   105   105 E System  : java.lang.RuntimeException: Failed to create service com.android.server.am.ActivityManagerService$Lifecycle:
 service constructor threw an exception
12-30 22:43:20.380   105   105 E System  :      at com.android.server.SystemServiceManager.startService(SystemServiceManager.java:103)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.SystemServer.startBootstrapServices(SystemServer.java:419)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.SystemServer.run(SystemServer.java:332)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.SystemServer.main(SystemServer.java:219)
12-30 22:43:20.380   105   105 E System  :      at java.lang.reflect.Method.invoke(Native Method)
12-30 22:43:20.380   105   105 E System  :      at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:886)
12-30 22:43:20.380   105   105 E System  :      at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:776)
12-30 22:43:20.380   105   105 E System  : Caused by: java.lang.reflect.InvocationTargetException
12-30 22:43:20.380   105   105 E System  :      at java.lang.reflect.Constructor.newInstance0(Native Method)
12-30 22:43:20.380   105   105 E System  :      at java.lang.reflect.Constructor.newInstance(Constructor.java:430)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.SystemServiceManager.startService(SystemServiceManager.java:92)
12-30 22:43:20.380   105   105 E System  :      ... 6 more
12-30 22:43:20.380   105   105 E System  : Caused by: java.lang.IllegalArgumentException: Invalid argument: 105
12-30 22:43:20.380   105   105 E System  :      at android.os.Process.setThreadGroup(Native Method)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.UiThread.<init>(UiThread.java:35)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.UiThread.ensureThreadLocked(UiThread.java:40)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.UiThread.get(UiThread.java:49)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.am.ActivityManagerService$UiHandler.<init>(ActivityManagerService.java:1607)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.am.ActivityManagerService.<init>(ActivityManagerService.java:2629)
12-30 22:43:20.380   105   105 E System  :      at com.android.server.am.ActivityManagerService$Lifecycle.<init>(ActivityManagerService.java:2603)
12-30 22:43:20.380   105   105 E System  :      ... 9 more
12-30 22:43:20.380   105   105 D AndroidRuntime: Shutting down VM

        It seems to be a problem with Zygote, which is part of the Android image. I wonder if another image would work better. It seems odd that the image would not know how to talk to itself, especially as it works elsewhere.

        The odd part is that the invalid argument is different each time, "105", "152", "196", "240", "284", "329". It seems like it is increasing by between 40 and 50 each time. It looks a bit like an amount of process started in the LXC container between two attempts.

        By Antoine Fontaine on 2019-12-30T22:55:54

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner

        Here's a much cleaner log: https://paste.gnugen.ch/raw/9G4q.

        I obtained this one by running anbox launch --package=org.anbox.appmgr --component=org.anbox.appmgr.AppViewActivity without running the session manager before.

        By Antoine Fontaine on 2019-12-30T22:44:44

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner

        At that time, I could get adb shell to run, but anbox could not start the application manager. It was apparently because the Android Application Manager could not start Zygote, the "app embryo" from which all apps are forked. I couldn't adb install either.

        By Antoine Fontaine on 2020-02-15T03:58:51

        Edited by Administrator
      • Please register or sign in to reply
      • Administrator
        Administrator @root · Imported
        Author Owner

        After @afontain's MR is merged into the anbox branch, it seems like a good idea to me to create a new MR to merge the anbox branch into master.

        Because then we don't need to deal with two branches anymore, and contributions should become easier (e.g. rebasing the kernel with that one change is annoying).

        What do you think about that, everybody?

        By Oliver Smith on 2019-12-30T22:19:16

        Edited by Ghost User
      • Administrator
        Administrator @root · Imported
        Author Owner

        I'd suggest only merging the kernel change, or the kernel change and the two dependencies; but not anbox. It brings the same benefits, and allows merging Anbox itself only when it'll be functional.

        I also believe the two dependencies (and maybe Anbox too!) should be merged to Alpine itself, if possible. I could do the upstreaming work for that, if you want.

        By Antoine Fontaine on 2019-12-30T22:28:49

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner

        I'd suggest only merging the kernel change, or the kernel change and the two dependencies; but not anbox. It brings the same benefits, and allows merging Anbox itself only when it'll be functional.

        Fine with me, would you like to make MRs to prepare the kernel change + two dependencies to get merged into master?

        I also believe the two dependencies (and maybe Anbox too!) should be merged to Alpine itself, if possible. I could do the upstreaming work for that, if you want.

        If you are willing to maintain it in Alpine, then it would be great. But I'd still wait with that, until it is at least somewhat usable in postmarketOS, otherwise the dependencies aren't really useful in Alpine either.

        By Oliver Smith on 2020-01-03T01:00:34

      • Please register or sign in to reply
    • Administrator mentioned in issue postmarketos.org#124 (closed) · Imported

      mentioned in issue postmarketos.org#124 (closed)

      By Oliver Smith on 2020-02-07T18:10:48

    • Administrator
      Administrator @root · Imported
      Author Owner

      Two days ago, I finally got Anbox to run on my computer. That's fantastic!

      It seems the biggest issue was the openRC service file

      #!/sbin/openrc-run

supervisor=supervise-daemon

name="Anbox container manager"
command=/usr/bin/anbox
command_args="container-manager --daemon --privileged --data-path=/var/lib/anbox"
-command_background=true
-pidfile=/run/anbox-container-manager.pid
-
-depend() {
-	need net
-	after firewall
-}

supervisor=supervise-daemon

name="Anbox container manager"
command=/usr/bin/anbox
command_args="container-manager --daemon --privileged --data-path=/var/lib/anbox"
+
+depend() {
+	need localmount sysfs cgroups
+	after firewall
+}

      I try to reduce the amount of patches required. For example, I noticed the boost patch isn't required anymore and I sent a PR for some musl fixes.

      image

      Follow up work include:

      1. merging kernel changes (I couldn't use a postmarketOS kernel, so I think someone else should do it);
      2. building an android image(?);
      3. getting Anbox merged into Alpine;
      4. creating an anbox0 interface automatically (if possible independently from Network Manager);
      5. checking that it works with binderfs (CONFIG_ANDROID_BINDERFS in kernel config);
      6. making Anbox compile on architectures other than x86_64;
      7. rc-update add anbox container manager automatically on install (with a postmarketos-anbox package?).

      By Antoine Fontaine on 2020-02-15T01:16:23

      Edited by Administrator
    • Administrator
      Administrator @root · Imported
      Author Owner

      We also need testers! While it works on my computer, there is a high probability that I've forgotten something and that it will fail elsewhere.

      To test, you must at the moment:

      • have a x86_64 device
      • have a kernel with the required options. You may need to rebuild it;
      • Build the packages with abuild or maybe something like this with pmbootstrap:
      $ git clone https://gitlab.alpinelinux.org/afontain/aports/ -b anbox
$ cp -r aports/testing/cmake-extras aports/testing/properties-cpp aports/testing/anbox* "$(pmbootstrap config aports)/main"
$ pmbootstrap build anbox

      then after building and installing the package,

      • start the anbox container manager: rc-service anbox-container-manager start
      • finally, anbox-launch to start Anbox.

      I have included a lot of debugging instructions on the wiki page. Please make use of it!

      By Antoine Fontaine on 2020-02-15T03:48:20

      Edited by Administrator
    • Administrator
      Administrator @root · Imported
      Author Owner

      progress report, 4 days later:

      1. I haven't touched pmOS' kernel configs;
      2. Although building an android image looks nearly impossible, repackaging the image works very well, I think it work well enough for now;
      3. there has been some progress to merge Anbox into Alpine, while the image will live in postmarketOS repositories since it is not built from source. I am currently waiting for review on both;
      4. we do create a network interface automatically using network manager as of now (anbox-networkmanager subpackage). It would be nice to do it with openRC's networking scripts, as network manager isn't everyone's favourite and the current way has an issue: network manager needs to be started, so installation can't be done at pmbootstrap install stage;
      5. I checked if it works with binderfs: the answer is that it actually doesn't work;
      6. I haven't got any upstream answer about making Anbox compile on architectures other than x86_64;
      7. we autoenable the container manager with the postmarketos-anbox metapackage.

      also, the out of the box support is much better, as I got F-Droid bundled into the android image, as a privileged system app. That way, there is no need for adb install FDroid.apk or to enable installation from unknown sources!

      The big downside is that, as of now, it seems to only work on my computer.

      By Antoine Fontaine on 2020-02-21T23:49:52

      Edited by Administrator
    • Administrator
      Administrator @root · Imported
      Author Owner

      6 days after the previous status report, progress has been made:

      • I updated two pmOS kernel configs: linux-postmarketos-stable and linux-postmarketos-allwinner (the latter being used for the pinephone);
      • I managed to make Anbox work out of the box on x86_64, at least for linux-postmarkeos-stable with my computer and device-tablet-x64uefi;
      • I found a way to compile Anbox on every architecture supported by postmarketOS, which is great! I sent a PR and opened an issue upstream, which is awaiting an answer;
      • @Minecrell could build newer versions of the Android images, which means the images will be more up to date and that they will all come from the same release: this will help avoid bugs creeping up because the images are different;
      • I could get Anbox merged into Alpine;

      There is still work though:

      • Anbox doesn't work under QEMU yet (tried on both x86_64 and aarch64)
      • Nobody has succeeded running Anbox on his device so far, including x86_64 (the last attempts were before I managed to make it work out of the box);
      • I have yet to see the Wayland backend functionning.
      • Although I tried to make it work on my pinephone, there has been no success made so far to run Anbox on any ARM device. I'll try again with Minecrell's image later.

      Later will also come considerations such as "how can we make the virtual keyboard pop out on plasma mobile" (phosh should work fine) or "how to force the wayland backend so that it can fit in phosh and not look blurry" or "how to issue a back button" (that's done using the escape key on desktop).

      Progress is being made!

      By Antoine Fontaine on 2020-02-22T11:50:10

      Edited by Administrator
    • Administrator
      Administrator @root · Imported
      Author Owner

      Anbox is merged into both Alpine and pmaports so I think it is time to close this issue! :tada:

      Further improvements/fixes should go into separate issues (and some have already been made).

      By Minecrell on 2020-03-07T19:37:04

    • Administrator closed · Imported

      closed

      By Minecrell on 2020-03-07T19:37:04

    • Administrator mentioned in merge request !2212 (merged) · Imported

      mentioned in merge request !2212 (merged)

      By Pablo Correa Gomez on 2021-07-18T19:01:06

    • Administrator mentioned in issue #2198 (closed) · Imported

      mentioned in issue #2198 (closed)

      By Daniel Fancsali on 2023-12-13T20:25:13

    • Administrator mentioned in issue #2477 (closed) · Imported

      mentioned in issue #2477 (closed)

      By mrintrepide on 2023-12-26T23:22:16

    • Ghost User mentioned in merge request !6106

      mentioned in merge request !6106

    Please register or sign in to reply