86764 - Closing a TCP socket may fail with EINVAL.

Status: RESOLVED DUPLICATE of bug 86941

(NSPR :: NSPR, defect, P1)

Description

[Wan-Teh Chang]

On Tru64 Unix, closing a TCP socket sometimes fails with the
EINVAL error but the socket is actually closed.  Compaq
confirmed this is a bug.  The patch will not be available
for another year because we just missed the last patch release.

Here are the tracking numbers associated with the
tcp socket close() defect:

Request ID: BCGM711RJ
Field ID:      CFS77854
PTR:            117-1-14950
QAR:           81633 and 81431

The following is from the Tru64 Unix Networking Group.

>    1. A clear description of the nature of the problem and the 
>       solution. 

   Problem description: 

   A problem exists such that the close() system call, when 
   called to close a TCP socket, may return an error status 
   with errno set to EINVAL.  Contrary to the return status, 
   the system call does successfully close the socket. 

   Solution: 

   The solution is a change to TCP processing of the close() 
   request, such that a successful close() completes with a 
   success status. 

   Workaround: 

   Because the socket was successfully closed, an application 
   may safely ignore the EINVAL error. 

   Tru64 UNIX versions affected: 

   V5.1 and earlier, although the problem is most likely to 
   occur on V5.0, V5.0A, and V5.1. 

   Kernel modules affected: 

   tcp_usrreq.o 
   inet.mod 

>    2. Under what conditions and with what iPlanet products the 
>       problem will occur. 

   The problem is most likely to occur when a socket has been 
   shutdown() for writing, followed by the remote end of the 
   connection being closed, followed by the socket being closed() 
   within 1 second of the remote end closing. 

>   3. Why the fix to the NSPR library is the best solution. 

   It frees those applications that use it from being dependent 
   on a patch to the base operating system.

Comment 1

[Wan-Teh Chang]

Attachment: Proposed patch.

Comment 2

[Wan-Teh Chang]

I opened a new bug (bug #86941) with Compaq-approved bug
description.

*** This bug has been marked as a duplicate of 86941 ***