> I've finally had time to look at this.  It seems that tail uses mmap, but 
> never calls munmap.  When you execute the second tail, it grabs the same
> mmap'd region, which has not been updated with the new data.

> Adding appropriate munmap()'s seems to fix the problem.

Does this mean that any arbitrary user program that mmap()s a file can
prevent tail from working on it?

-- Richard