two bugs and a patch

[Nickolai Zeldovich]

> Here is another patch for you to try:
> [..]  If the call is free:ed retry the whole hash-bucket chain.

Well, my arlad now has a record uptime of 1.5 days, so it seems to
have helped.

I seem to have ran into a new bug though, with the latest checkout
of arla from CVS:

  freebsd% cd ~ ; ls | wc -l
        44
  freebsd% /usr/arla/bin/aklog
  freebsd% ls | wc -l
  ls: .: Bad file descriptor
         0
  freebsd% fs flushv . ; ls | wc -l
        44

Experimentally, backing out fcache.c, messages.c and messages.h to
versions 1.309, 1.230 and 1.16 made the problem disappear.

Thanks again,

-- kolya