[LU-14290] Convert fault-injection framework to match the model using in linux

Related — Mon, 4 Jan 2021 17:24:49 +0000

Lustre has a fault-injection framework which contains functionality which is similar to the fault-injection in Linux, though in a very different form.

It also contains functionality - particularly for encouraging race conditions - which is not present in Linux.

The two frameworks have substantially different user-space interfaces. Lustre allows a single fault to be enabled at a time, by writing a magic number to a magic file. Linux allows each possible fault to be enabled independently by using a named directory in debugfs which contains files for controlling various aspects of the fault trigger.

The goal of this improvement is to provide enhancements to the Linux fault-injection framework which allow races to be encouraged while using the interface style of the rest of that framework. Then convert all lustre faults to use the linux framework.

We would need to preserve the current lustre interface, at least for a while, so old test suites could run on new code.

Whamcloud Community JIRA

[LU-14290] Convert fault-injection framework to match the model using in linux