tree: 45cd87dd3524578ab3e424f090b4bd6a6530e8e6 [path history] [tgz]
  1. collections/
  2. data/
  3. device/
  4. execution/
  5. fs/
  6. meta/
  7. mm/
  8. signals/
  9. syscalls/
  10. task/
  11. types/
  12. auth.rs
  13. bpf.rs
  14. BUILD.gn
  15. decision-log.md
  16. dynamic_thread_pool.rs
  17. loader.rs
  18. lock.rs
  19. logging.rs
  20. main.rs
  21. mutable_state.rs
  22. README.md
  23. selinux.rs
  24. starnix.gni
  25. testing.rs
  26. trace.rs
  27. vdso-x64.cc
  28. vdso-x64.map
  29. vdso.ld
  30. vmex_resource.rs
src/proc/bin/starnix/README.md

Starnix

Starnix runs unmodified Linux binaries on Fuchsia. We have tested Starnix with glibc and bionic based binaries. Patches that improve Linux compatibility are welcome!

How to run starnix

Configure your build

In order to run starnix, we need to build //src/proc:

fx set core.x64 --with //src/proc,//src/proc:tests
fx build

Note: If you have //vendor/google in your source tree, you might want to add //vendor/google/starnix:tests to the fx set command above to include some additional, non-public tests.

Run Fuchsia

Run Fuchsia as normal, for example using fx serve and ffx emu start --headless.

To monitor starnix, look for log messages with the starnix tag:

ffx log --filter starnix --severity TRACE --select "core/*/starnix*#TRACE"

When running tests, you will need to pass the log selection parameters to fx test instead:

fx test --min-severity-logs TRACE <test name>

The --select arguments contain the moniker for the starnix instance whose minimum severity log level you want to change. This affects the logs emitted by starnix, as opposed to --severity, which affects which logs are filtered for viewing. The changed log level only persists for the duration of the ffx log command.

If you do not care about detailed logging, you can leave out the --severity and just do:

ffx log --filter starnix

Starnix produces a large amount of logs and this can overload archivist's ability to retain them, instead printing messages like:

[starnix] WARNING: Dropped logs count: 5165

If you see this, you can reduce or eliminate the dropped messages by increasing the value of max_cached_original_bytes the the JSON file `//src/diagnostics/archivist/configs/archivist_config.json' in your local checkout. Increasing by a facfor of 10 seems to work well.

Run a Linux binary

To run a Linux binary, ask starnix to start a component that wraps the binary:

ffx starnix start fuchsia-pkg://fuchsia.com/hello-starnix#meta/hello_starnix.cm

If this is the first time you've used the ffx starnix command, you might need to configure ffx to enable the starnix commands. Attempting to run the start command should provide instructions for enabling the starnix commands.

If everything is working, you should see some log messages like the following:

[00064.846853][33707][33709][starnix, starnix] INFO: main
[00064.847640][33707][33709][starnix, starnix] INFO: start_component: fuchsia-pkg://fuchsia.com/hello-starnix#meta/hello_starnix.cm

Run a Linux test binary

Linux test binaries can also be run using the Starnix test runner using the standard fx test command:

fx test exit_test --output

You should see output like:

[==========] Running 3 tests from 1 test suite.
[----------] Global test environment set-up.
[----------] 3 tests from ExitTest
[ RUN      ] ExitTest.Success
[       OK ] ExitTest.Success (4 ms)
[ RUN      ] ExitTest.Failure
[       OK ] ExitTest.Failure (3 ms)
[ RUN      ] ExitTest.CloseFds

If you set the log level to TRACE (e.g., ffx log --severity TRACE --select "core/test*/*/starnix*#TRACE"), you should see the system call handling in the device logs:

[629.603][starnix][D] 1[/data/tests/exit_test] wait4(0x3, 0x1c48095b950, 0x0, 0x0, 0x10, 0x10)
[629.603][starnix][D] 3[/data/tests/exit_test] prctl(0x53564d41, 0x0, 0x700d5ea000, 0x3000, 0x3a506c7a34b, 0xc06913ece9)
[629.603][starnix][D] 3[/data/tests/exit_test] -> 0x0
[629.604][starnix][D] 3[/data/tests/exit_test] exit_group(0x1, 0x3, 0x2b18e3464f8, 0x3000, 0x3a506c7a34b, 0xc06913ece9)
[629.604][starnix][I] exit_group: pid=3 exit_code=1

For GUnit tests (such as the syscall tests in //src/proc/tests/android/gvisor), you can run specific tests with the --test-filter flag. For example,

fx test epoll_test --output --test-filter="EpollTest.AllWritable"

Specifying * as the filter turns on all tests in the binary.

Making changes to Starnix when using ffx

The Starnix instances that ffx connects are static in the component hieararchy. This means that they need to be stopped explicitly in order to be updated when changes have been made to the Starnix runner code.

ffx component stop starnix_kernel

If more than one Starnix instance is running, the above command will list the running Starnix instances and you can stop them individually.

Alternatively, use the following command to stop all the instances at once:

fx shell killall starnix_kernel.cm

Testing

Writing in-process unit tests

Decorate your test function with the #[::fuchsia::test] macro instead of the standard #[test] macro. #[::fuchsia::test] will initialize logging so that failing tests can be debugged more easily.

Running the in-process unit tests

Starnix also has in-process unit tests that can interact with its internals during the test. To run those tests, use the following command:

fx test starnix-tests

Using a locally built syscalls test binary

The syscalls_test test runs a prebuilt binary that has been built with the Android NDK. You can substitute your own prebuilt binary using the starnix_syscalls_test_label GN argument:

fx set core.x64 --args 'starnix_syscalls_test_label="//local/starnix/syscalls"' --with //src/proc,//src/proc:tests

Build your syscalls binary and put the file in //local/starnix/syscalls. (If you are building using the Google-internal build system, be sure to specific the --config=android_x86_64 build flag to build an NDK binary.)

You can then build and run your test as usual:

fx build
fx test syscalls_test

Viewing Inspect for debugging

You can view Inspect data exposed by starnix using ffx inspect.

To view the thread groups currently running, run:

ffx inspect show core/starnix_manager/bionic:root/container/kernel/thread_groups

You can also view the number of syscalls that have been executed (after enabling the “syscall_stats” feature):

ffx inspect show core/starnix_manager/bionic:root:syscall_stats

Tracing

Starnix is integrated with Fuchsia's tracing system. To start a trace with an increased buffer size, run:

ffx trace start --categories "kernel:meta,starnix" --buffer-size 16

Trace files can be visualized and queried with Perfetto. For example, to see the average time spent in starnix during a clock_getres syscall, run the query:

select avg(dur), count(*)
from slice join args using (arg_set_id)
where key='name' and display_value='clock_getres' and name='RunTaskLoop'