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.
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:
allow_crash_reporter(domain_name)
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.
_metrics.enableAnalyticsReporting
or by manually creating the file /data/misc/metrics/enabled
(for testing only).BRILLO_CRASH_SERVER
make variable should be set in the product.mk
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
.BRILLO_PRODUCT_ID
make variable should be set in the product.mk
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
.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.