two concurrency questions

I searched the documentation and can't find the specific answers to
these questions.

1: Are user defined functions (eg pgplsql) atomic? I.e. if my function
updates a bunch of rows and one update fails for one reason or another
does it automatically roll-back to the state before the function was called?

2: If an update causes a trigger and the trigger updates a row and the
trigger fails, what happens? Is there a way to atomize the whole process?

Thanks,

Jeff Patterson
jpat@mpip.org
The Melanoma Patients' Information Page
http://www.mpip.org