Merge branch 'rewrite-crash_reporter' into merge-crash_reporter

Initial import of crash_reporter from platform/system/core

BUG: 29548040
tree: 865dca4a11fa42b8e57b1e4392a0d3bcf5bc5932
  1. dbus_bindings/
  2. init/
  3. .project_alias
  4. 99-crash-reporter.rules
  7. crash_collector.h
  9. crash_collector_test.h
  11. crash_reporter.rc
  12. crash_reporter_logs.conf
  14. crash_sender
  16. kernel_collector.h
  18. kernel_collector_test.h
  21. kernel_warning_collector.h
  23. OWNERS
  24. periodic_scheduler
  29. udev_collector.h
  32. unclean_shutdown_collector.h
  35. user_collector.h
  37. warn_collector.l
  38. warn_collector_test.c


crash_reporter is a deamon running on the device that saves the call stack of crashing programs. It makes use of the Breakpad library.

During a build, Breakpad symbol files are generated for all binaries. They are packaged into a zip file when running m dist, so that a developer can upload them to the crash server.

On a device, if the user has opted in to metrics and crash reporting, a Breakpad minidump is generated when an executable crashes, which is then uploaded to the crash server.

On the crash server, it compares the minidump's signature to the symbol files that the developer has uploaded, and extracts and symbolizes the stack trace from the minidump.

SELinux policies

In order to correctly generate a minidump, crash_reporter needs to be given the proper SELinux permissions for accessing the domain of the crashing executable. By default, crash_reporter has only been given access to a select number of system domains, such as metricsd, weave, and update_engine. If a developer wants their executable's crashes to be caught by crash_reporter, they will have to set their SELinux policies in their .te file to allow crash_reporter access to their domain. This can be done through a simple macro:


Replace domain_name with whatever domain is assigned to the executable in the file_contexts file.


crash_reporter has a few different configuration options that have to be set.

  • Crashes are only handled and uploaded if analytics reporting is enabled, either via the weave call to set _metrics.enableAnalyticsReporting or by manually creating the file /data/misc/metrics/enabled (for testing only).
  • The BRILLO_CRASH_SERVER make variable should be set in the file to the URL of the crash server. For Brillo builds, it is set automatically through the product configuration. Setting this variable will populate the /etc/os-release.d/crash_server file on the device, which is read by crash_sender.
  • The BRILLO_PRODUCT_ID make variable should be set in the file to the product's ID. For Brillo builds, it is set automatically through the product configuration. Setting this variable will populate the /etc/os-release.d/product_id, which is read by crash_sender.

Uploading crash reports in eng builds

By default, crash reports are only uploaded to the server for production user and userdebug images. In eng builds, with crash reporting enabled the device will generate minidumps for any crashing executables but will not send them to the crash server. If a developer does want to force an upload, they can do so by issuing the command SECONDS_SEND_SPREAD=5 FORCE_OFFICIAL=1 crash_sender from an ADB shell. This will send the report to the server, with the image_type field set to force-official so that these reports can be differentiated from normal reports.