Hence the i_size_write() in there is deadlocky.  Go back to the old way.

 
Forward-port from 2.4:
 
The following patch pulls an NFS server IP address off root_server_path
(handed out via the DHCP root-path option), if it is present.  For example,
you can do this sort of thing in dhcpd.conf:
 
  root-path = 192.168.1.33:/tftpboot/yip.zImage
 
This lets you mount your root filesystem off a different machine than you
booted from, without needing to use kernel command-line parameters.
 
The patch appears to be backwards compatible.
 
RFC2132 says this about the root-path option:
 
   This option specifies the path-name that contains the client's root
   disk.  The path is formatted as a character string consisting of
   characters from the NVT ASCII character set.
 
This is sufficiently vague to allow the path-name to include an IP-address.
Also, I found some documentation for FreeBSD saying it does this too, so it
must be right, because those FreeBSD guys are really smart...  :-)
 
The only downside of the patch is that the summary that ipconfig prints can
be a little odd when the kernel command line overrides whatever ipconfig gets
from (say) DHCP.  The address from the kernel command line seems to get
stripped off early, so ipconfig reports it, but it doesn't report the kernel
command line NFS path, since that's handled a bit later...  This small
cosmetic problem looks difficult to "fix" without rewriting quite a bit of
stuff...

       read due to EOF.

     This is wanted in order to allow the NFS client to send more requests before
     is has to block and wait for replies.
     This is mainly useful if you have a WAN and want to ensure that the bandwidth
     is being used efficiently.

    where this causes a padding error: xdr_encode_fhandle() and unx_marshal()

    are placed last on the ordered list nlm_block (problem reported by Olaf
    Kirch).

    by using the client address in addition to the value of the NLM cookie field.

    were requesting a blocking lock when we get a reply from the server asking us to
    block.

    When a server receives that callback it should deallocate the corresponding blocked
    lock using the nlmsvc_grant_reply function.

     of compound_decode_hdr_maxsz.
NFSv4: fix a printk() typo (spotted by Linda Dunaphant).
NFSv4: Ensure that nfs4_open_reclaim() copies the value of the new stateid back into
     the shared nfsv4 state structure.
NFSv4: Don't leak NFS4ERR_WRONGSEC errors back into nfs_lookup().
RPC,NFS,Lockd: Mark the debugging code as "unlikely" so that gcc moves it out of the
     mainline code paths.

     full error codes. Should allow the "mount" program to print more useful error
     diagnostics.

     port. Fixes a privilege bug when CONFIG_SECURITY is set.
RPC: When trying to reconnect to a TCP port, try to bind() to the last used port number
     in order to ensure that the servers NFS replay cache recognizes this as being the
     same mount as before.

     ensure that the timeout shift is clamped to a maximum value of 8.

RPC: Fix by Olaf Kirch to the rpc scheduler to ensure sync tasks respect the
     "intr" mount flag.

     of an unnecessary global spinlock.

     problem.

      ones labelled as belonging to our file. This fixes a bug in which msync(MS_SYNC)
      will fail to flush the pages to disk.

     mounting. This should allow us to use AUTH_SYS credentials when mounting,
     (even when the user requests RPCSEC_GSS authentication) due to the hack
     described in RFC2623.

     Remove the broken NFS_INO_FAKE_ROOT hack.

      if NFSv4 is selected. Remove need for user to select SUNRPC_GSS, and the
      crypto options.
      Make NFSv3 a recommended option.

      is requested by the application (or memory pressure).

     longer relies on ctime for correctness in avoiding
     update races.

VFS: allow filesystems to disable inode_update_time() on
     a per-inode basis.

into ppc970.osdl.org:/home/torvalds/v2.5/linux

into redhat.com:/spare/repo/net-drivers-2.5

* the scan-channels message seemed to always give invalid output.
  Look at the constant, and discover we are sending another
  message entirely.  Fix the constant (CARM_MSG_IOCTL).

* s/MISC_SYNC_TIME/MISC_SET_TIME/

* list some additional messages

* bump version number

* Not setting cb->skb = NULL after releasing skb to OS or during
  initialization of cbs.  Reported by Deepak Saxena
  [dsaxena@plexity.net].

into redhat.com:/spare/repo/net-drivers-2.5