On Tue, 25 Nov 2003, David P. Reese, Jr. wrote:

> I'm trying to build distributions on an LFS volume on 1.6ZF and one of my
> nbinstall proc's is getting hung in state "uvn_fp".  When it gets hung
> I see the following on the console:
>
> lfs_writevnodes: starting over
> ino 248385 bytes == 0 offset 0

I'm seeing that too. It's easily reproducible with "cp -pR xsrc ..."
It only gets about 100 Mb done before hanging. It wasn't doing that
earlier this month, with 1.6ZE.

> I was able to get a backtrace of the process in DDB
>
> trace: pid 28454 at 0xe138c6f8
> ltsleep(c0803680,204,c0314c15,0,e09bb8a0) at netbsd:ltsleep+0x3fb
> uvn_findpage(e09bb8a0,0,0,e138c940,0) at netbsd:uvn_findpage+0x82
> uvn_findpages(e09bb8a0,0,0,e138c8dc,e138c940) at netbsd:uvn_findpages+0xaf
> genfs_getpages(e138ca00,6,0,c0366cc0,0) at netbsd:genfs_getpages+0x41c
> ubc_fault(e138cb30,e036e000,e138cab0,1,0) at netbsd:ubc_fault+0x1a0
> uvm_fault(c0366d00,e036e000,0,2,0) at netbsd:uvm_fault+0x2f8
> trap() at netbsd:trap+0x38f
> --- trap (number 2202) ---
>
> Is it possible to unhang the process?  It's not responding to SIGKILL.

I don't think a process can be killed while it's blocking on I/O. You
just have to unmount as many file systems as you can, and reboot.

Frederick