Peter Staubach wrote:
> Venkatesh KS wrote:
>> Folks,
>>
>> I am trying to understand how nfs client handles a server reboot while 
>> issuing multiple
>> unstable writes followed by a commit.
>>
>> Say a user requests a large write requests.
>> nfs client breaks it into multiple 32k requests, each 32k write 
>> request being unstable
>> and finally a commit call before the nfs client returns back to the user.
>>
>> If the server reboots in between these write calls, will the nfs 
>> client know abt it.
>> The write before this may not have made it to disk. If the nfs client 
>> doenst know
>> abt it then it goes ahead with commit call and returns.
> 
> I would suggest checking out RFC1813 for information on this topic.

There is also some explanation of this in the Linux NFS FAQ:

   http://nfs.sourceforge.net/

and we've found Brent Callaghan's "NFS Illustrated" to have some useful 
explanation of reboot recovery with unstable writes.