Feature request: Provide flag for fault injection

[cfcs]

It would be nice to be able to specify flags for introducing varying levels of FS- and storage-related faults. This would allow us to test how backup software fares in the face of failing disks, and to test error recovery code paths that may be hard to reach otherwise.

Some ideas:

• broken reads from blocks
• broken or half-broken directory inodes
• files from previous runs placed in different directories (to simulate files moved to lost+found)
  • these files should have the same content as they would have had with the same PRNG seed passed without the fault-flags, to benchmark identification of previously backed up files from new FS locations

[cfcs]

Sttumbled upon this,

• https://github.com/scylladb/charybdefs
• article: http://www.scylladb.com/2016/02/16/fault-injection-filesystem-software-testing/

Their approach is slightly different, but I think some of their features are worth considering. (kill -9 + testing resumption seems useful)