usercrash¶

Description¶

The usercrash tool gets the crash info for the crashing process in userspace. The collected information is saved as usercrash_xx under /var/log/usercrashes/.

Design¶

usercrash is designed using a Client/Server model. The server is autostarted at boot. The client is configured in core_pattern, which will be triggered when a crash occurs in userspace. The client then sends the crash event to the server. The server checks the files under /var/log/usercrashes/ and creates a new file usercrash_xx (xx means the index of the crash file). Then it sends the file descriptor (fd) to the client. The client is responsible for collecting crash information and saving it in the crashlog file. After the saving work is done, the client notifies server and the server will clean up.

The work flow diagram:

+--------------------------------------------------+
|                                                  |
|        Server                    Client          |
|           +                         +            |
|           |  Send crash event       |            |
|           | <-----------------------+            |
|           |                         |            |
|    Create usercrash_xx              |            |
|           |                         |            |
|           | Send usercrash_xx fd    |            |
|           +-----------------------> |            |
|           |                         |            |
|           |                   Fill usercrash_xx  |
|           |                         |            |
|           |  Notify completion      |            |
|           | <-----------------------+            |
|           |                         |            |
|       Clean up                      |            |
|           |                         |            |
|           v                         v            |
|                                                  |
+--------------------------------------------------+

Usage¶

The server is launched automatically at boot, and the client is configured in core_pattern or coredump-wrapper. In prepare.service, it will check the content of /proc/sys/kernel/core_pattern. If there is coredump-wrapper, which means that core_pattern has been set in systemd, no need to do it again. Otherwise, the content should be changed by:

$ echo "|/usr/bin/usercrash_c %p %e %s" > /proc/sys/kernel/core_pattern

That means client will be triggered once userspace crash occurs. Then the event will be sent to server from client.

The debugger is an independent tool to dump the debug information of the specific process, including backtrace, stack, opened files, registers value, memory content around registers, and etc.

$ debugger <pid>

Note

You need to be root to use the debugger.

Souce Code¶

client.c : This file is the implementation for client of usercrash, which is responsible for delivering the usercrash event to the server, and collecting crash information and saving it to the crashfile.
crash_dump.c : This file is the implementation for dumping the crash information, including backtrace stack, opened files, registers value, memory content around registers, and etc.
debugger.c : This file is to implement a tool, which runs in command line to dump the process information list above.
protocol.c : This file is the socket protocol implement file.
server.c : This file is the implement file for server of usercrash, which is responsible for creating the crashfile and handle the events from client.