OpenAFS getcwd() sometimes returns ENOENT

[andersk]

Every few hours, a server gets stuck in a state where certain users receive ENOENT from getcwd(), resulting in errors like

shell-init: error retrieving current directory: getcwd: cannot access parent directories: No such file or directory
job-working-directory: error retrieving current directory: getcwd: cannot access parent directories: No such file or directory
job-working-directory: error retrieving current directory: getcwd: cannot access parent directories: No such file or directory

Sometimes the problem goes away on its own, but usually the server needs to be rebooted.

I attempted to debug this with upstream on ​2013-06-11 and ​2013-06-16. Using crash, I found that we’re ending up with two dentries referring to the same directory inode, which is ​bad. We’re all confused.