df displays total space incorrectly... sometimes

[Mikhail T.]

Hello!

I have a pair of systems (foo and bar) mounting each other's ZFS shares.
One on of them I recently started getting errors from some applications
complaining about disk being full. An examination with `df -m' on the
NFS client (foo) revealed:

    bar:/bar    353010 922208 -569198   261%    /bar

The same `df -m' run on the server shows:

    bar        4547314 922208 3625105    20%    /bar

Note, that the Used-column (second) matches. Note also, that the Total
column on the NFS-client is remainder of the real value seen on the
server. Hex makes this easier to see:

    % printf '%X %X\n' 4547314 353010
    4562F2 562F2

The difference is exactly 0x400000. If it were df itself messing up the
number, it would've done the same thing on the server as well. What's
broken on the client side?

I don't think, I saw this, when I first switched to ZFS (and giant
storage-sizes with it) from multiple single drives. But the recent
kernel (foo is running 10.3 from April 18) displays this behavior...

Things work fine in the other direction, where bar is the client and foo
is the server:

    foo:/foo  4778760  919 4777840     0%    /foo

The kernel on bar is from today (10.3 as of May 8th) -- was this problem
fixed in the three weeks between these dates?

Thanks!

    -mi