kern/76538: nfs-write on gbde partition stalls and continues much later

[Sebastian Dröge]

>Number:         76538
>Category:       kern
>Synopsis:       nfs-write on gbde partition stalls and continues much later
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Fri Jan 21 17:10:17 GMT 2005
>Closed-Date:
>Last-Modified:
>Originator:     Sebastian Dröge
>Release:        5.3
>Organization:
>Environment:
FreeBSD hel.yggdrasil 5.3-RELEASE-p5 FreeBSD 5.3-RELEASE-p5 #1: Tue Jan 18 23:22:54 CET 2005     root at hel.yggdrasil:/usr/obj/usr/src/sys/HEL  i386
>Description:
Hi,
/dev/da0s1d is an partition on a intel raid controller... it's mapped via gbde (with sector size 2048 as in the docs) to /dev/da0s1d.bde
On /dev/da0s1d.bde is an UFS2 fs with softupdates
this is exported with nfs

when I mount this partition on a linux client and write some GB (in many files) to it the transfer stalls at some point (after random time and transfered size).
Now bufdaemon and gbde have both ~40% cpu and on the client it says, that the nfs server isn't responding.
When looking at gkrellm on the server it shows many writes on the harddisks and storcon (an configure/monitor tool for the raid controller) shows in it's cache monitor, that something is writing on the same sectors all the time. I can't write anything or read anything from the partition at that point... it just stalls also
after a while everything turns back to normal operation and everything gets written ok to the disc

Any ideas?
>How-To-Repeat:
1. create a gbde mapping
2. export it with nfs (writeable)
3. write some GB from a client to it
>Fix:
      
>Release-Note:
>Audit-Trail:
>Unformatted: