Blocking v/s Non-blocking NFS (and iSCSI) file reads/writes.

Blocking v/s Non-blocking NFS (and iSCSI) file reads/writes.

[Ben Greear]

I'm working on writing a program that will be used to stress test
NFS and iSCSI based file systems.  I am currently using a non-threaded,
and non-forking architecture based on non-blocking IO and select() to
do my network traffic generation.  I would like to be able to fit the
file-testing code in the same framework.  However, I'm not sure I can
make this model work with network based file systems....

So, does select() work for NFS reads?  (IE: I open a file-descriptor
on an NFS mounted file system, and start reading.  The network goes
down.  Will select() start not marking that file as read/write-able?)

If I set the file descriptor to be O_NONBLOCK, will it return immediately
if the network is down (regardless of what select told me)?

I have the same questions about an iSCSI based file system...

Does anyone have any suggestions for reading material on this topic,
other than kernel source and patches?

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>          <Ben_Greear@excite.com>
President of Candela Technologies Inc      http://www.candelatech.com
ScryMUD:  http://scry.wanfear.com     http://scry.wanfear.com/~greear

Re: Blocking v/s Non-blocking NFS (and iSCSI) file reads/writes.

[Trond Myklebust]

>>>>> " " == Ben Greear <greearb@candelatech.com> writes:

     > So, does select() work for NFS reads?  (IE: I open a
     > file-descriptor on an NFS mounted file system, and start
     > reading.  The network goes down.  Will select() start not
     > marking that file as read/write-able?)

No.

     > If I set the file descriptor to be O_NONBLOCK, will it return
     > immediately if the network is down (regardless of what select
     > told me)?

No.

The NFS layer knows nothing at all about the network. It relies on the
RPC layer to handle all that for it. There are 2 ways in which it can
do this:
  1) keep the NFS layer in the dark (using the 'hard' mount option)
  2) pass an error back which then propagates back through NFS to the
     user (the 'soft' mount option).

Cheers,
  Trond