Bug 186310

Description of problem:
I think that bug 154678 has introduced a regression into the maximum number of
nfs mounts that we can have at the moment

Version-Release number of selected component (if applicable):
RHEL3 U6 forward

How reproducible:
always

Steps to Reproduce:
1.boot a U6 or newer kernel
2.begin mounting 1000 NFS mounts
3.
  
Actual results:
~512 NFS mounts are mounted before EADDRINUSE is returned

1000 NFS mounts are mounted

Additional info:
I think this is an inadvertent error in the xprt_bindresvport function changed
in the kernel as of U6.  By increasing the max port range scan from 800 to 1023
the kernel fully overlapped the port range which the C library uses to scan port
ranges in the userspace version of bindresvport.  Since the mount program uses a
reserved port to initiate mounts to NFS servers, which it obtains via
bindresvport (which only scans ports 512 through 1023), the new kernel behavior
allows all the ports that mount can legitimately use to become allocated for
handling NFS mounts.  The functional offshoot of this is that, while the kernel
could establish a new NFS socket for additional mounts, the mount program in
user space itself cannot find a secure port through  bindresvport to initiate a
mount to a new NFS server.  The net of this is that whereas previous releases
supported 800 NFS mounts, post U6 kernels can only support ~512, depending on
port utilization from other applications.

I've attached a patch to the kernel to correct this.  There are several
alternate methods to fix this I think (have mount not use a reserved port,
change glibc so that bindresvport scans all reserved ports, etc).  But this
solution struck me as having the least impact so late in the RHEL3 release schedule.