[LSF/MM TOPIC] drgn, a scriptable debugger

[LSF/MM TOPIC] drgn, a scriptable debugger

[Omar Sandoval]

Hello,

I've already sent in an attend request for the I/O track, but I'd also
like to propose my project, drgn [1], as a topic for the BPF/tracing
track.

Basically, drgn is a more intuitive replacement for crash(8) which
exposes kernel types and variables to Python. For example, one can
trivially access common kernel structures:

>>> init = find_task(prog, 1)
>>> fd = fget(init, 3)
>>> fd.f_inode.i_sb.s_type.name
(const char *)0xffffffff862bf29c = "devtmpfs"

Since it's just Python, it's also easy to do more complex analysis:

>>> sum(1 for _ in list_for_each(fd.f_inode.i_sb.s_inodes.address_of_()))
491

I've been using this to debug issues in production at Facebook. Since
the project is relatively new, some of it is still in flux. I'm
currently rewriting the core components in C for performance reasons.
Next after that is improved documentation as well as stack trace
support, which is the biggest missing feature. Regardless, it's already
proven quite useful, so I'd love a chance to talk about it and gather
feedback.

Thanks!

1: https://github.com/osandov/drgn