Hacker News new | past | comments | ask | show | jobs | submit login

TL;DR: He overflowed inotify queue of /etc/init which is the Upstart configuration directory being monitored. Upstart doesn't deal with the overflow, exits, and causes kernel panic.

The bug is not fixed because in order to trigger it you need root to spam file operations in /etc/init, which implies bigger problems elsewhere. If you have root and want to see panics, just echo c >/proc/sysrq-trigger.

Rachel is a she I think you will find. A better TLDR is that init can die due to valid response from the kernel. She found this in a production use case. Perhaps involving fast changing services.

> Upstart doesn't deal with the overflow,

Actually it's libnih that doesn't deal with the overflow. The difference is that any other application using that library will also abort on that condition. Let the new bug hunt begin!

Actually, it watches all of /etc/. Rachel wrote this:

"If you poke around in the source, you can find that it actually registers watches for the entire directory of its config file for various reasons, so it winds up following /etc (due to /etc/init.conf) and /etc/init (its "job dir")."

> Upstart doesn't deal with the overflow

Tecnically Upstart doesn't even see the overflow, libnih blows up without notifying its caller.

Why is this getting downvoted? This is an important aspect to the bug.

All software has bugs, and Upstart is no exception, so it should not be a great surprise or tragedy that a bug is found. If you could do this as a non-root user, that would be worthy of the "I just can't believe Upstart could have a bug!" type post. But this is a pretty standard this-should-never-happen-in-a-normal-system bug.

Sometimes when a bug has a very low likelihood of happening in a normal system, it gets ignored; in this case the bug was noticed, but the bug needed work, which was never done. I know rachel says it came up on a production system, but that does not mean it was operating under normal circumstances. Still, this bug should be fixed.

Not that I need to mention it, but I will because i'm a dick: This problem would not happen with sysvinit. One more reason for me to keep using Slackware...

Yeah, while the digging in "step 5" is interesting, I think the reach of this bug is fairly negligible: "root user can break the system".

EDIT: perhaps not so negligible after all. See justincormack's comment.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact