[ ssic-linux-Bugs-1718329 ] Problem with migration of process using tcp socket

[SourceForge.net]

Bugs item #1718329, was opened at 2007-05-13 17:25
Message generated for change (Comment added) made by rogertsang
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=405834&aid=1718329&group_id=32541

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Process Management
Group: v1.9.1
Status: Open
>Resolution: Fixed
Priority: 5
Private: No
Submitted By: ch0hlik (ch0hlik)
Assigned to: Roger Tsang (rogertsang)
Summary: Problem with migration of process using tcp socket

Initial Comment:
For example I use this little tcp server:
http://www.cs.rpi.edu/courses/sysprog/sockets/server.c
$ gcc -o server server.c

On the first node 10.0.0.1:
$ ./server 8888

$ migrate 2 server_pid

On the second node 10.0.0.2:
$ telnet 10.0.0.1 8888
Trying 10.0.0.1...
Connected to openone (10.0.0.1).
Escape character is '^]'.
Connection closed by foreign host.

After this server show:
ERROR reading from socket: Invalid argument

In 1.2.2 I'm the same steps and it works fine.
In 1.9.3 problem looks just like in 1.9.2 or worse. (1.9.3 from here: http://radian.org/~roger/OPENSSI-FC/kernel/2.6.11/)

----------------------------------------------------------------------

>Comment By: Roger Tsang (rogertsang)
Date: 2009-10-26 23:53

Message:
checked-in fix

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2009-07-04 17:11

Message:
Finally got around to this...  kernel-2.6 socket_file_ops no longer provide
read() file operation that the existing OpenSSI code depends on.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-12-06 02:37

Message:
Logged In: YES
user_id=1246761
Originator: NO

I don't have documentation.  Try code inspection... ghost sockets get
created during migration and they inherit ssi socket ops used mostly to
ship syscalls to origin node.

----------------------------------------------------------------------

Comment By: ch0hlik (ch0hlik)
Date: 2007-12-05 10:36

Message:
Logged In: YES
user_id=1771064
Originator: YES

Roger,
Is there is any possibility to recive documentation for code used to
modify sockets inside kernel?
Thank you and regards,
RJ

----------------------------------------------------------------------

Comment By: ch0hlik (ch0hlik)
Date: 2007-12-05 10:33

Message:
Logged In: YES
user_id=1771064
Originator: YES

Roger,
Is there is any possibility to recive documentation for code used to
modify sockets inside kernel?
Thank you and regards,
RJ

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-10-29 01:36

Message:
Logged In: YES
user_id=1246761
Originator: NO

VPROC recreates the socket for SOCKET_I() during process migration.  No
-ERFB_TRYAGAIN errors were encountered, so the problem is somewhere in
rmtsock.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-08-14 02:24

Message:
Logged In: YES
user_id=1246761
Originator: NO

This is looking more like a revisit of the rmtfb_getsvr_handle() SSI-1.9.1
bug for which rmtfb_tryagain_handle() was written in SSI-1.9.2 to fix
remote tty but maybe should be extended for rmtsock if it is a case of
client/server ino mismatch.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-08-13 03:05

Message:
Logged In: YES
user_id=1246761
Originator: NO

Looking at the code again... In SSI-1.9 the base kernel-2.6 inode
structure no longer directly references its socket structure unlike in
SSI-1.2.  Unless the contents of this socket structure is exported to the
destination node along with the inode SSI-1.9 rmtsocksvr_* operations that
use SOCKET_I() are still working with bugus information from the local
node's sock_inode_cache.  This explains why TCP sockets did not get
recreated properly on the destination migration node.

----------------------------------------------------------------------

Comment By: ch0hlik (ch0hlik)
Date: 2007-06-06 15:18

Message:
Logged In: YES
user_id=1771064
Originator: YES

Is there is any possibilities that socket migration will be supported in
1.9.2 od 1.9.3?

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-05-14 03:10

Message:
Logged In: YES
user_id=1246761
Originator: NO

According to reop_import_pathinit() the current OpenSSI implementation
does not support socket migration.  However for some reason existing
outbound connections continue even after process migration.  The migration
code looks pretty much the same in SSI-1.2 and SSI-1.9.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2007-05-13 23:59

Message:
Logged In: YES
user_id=1246761
Originator: NO

In 2.6.11-ssi TCP socket did not get recreated properly at the destination
node.

At origin node:
server  236031 root    3u  IPv4 121308792                TCP *:8888
(LISTEN)

At destination node:
server  236031 root    3u  sock    0,2          121308792 can't identify
protocol

When migrating the process back to the origin node, the listen socket
reappears.

----------------------------------------------------------------------

You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=405834&aid=1718329&group_id=32541

------------------------------------------------------------------------------
Come build with us! The BlackBerry(R) Developer Conference in SF, CA
is the only developer event you need to attend this year. Jumpstart your
developing skills, take BlackBerry mobile applications to market and stay
ahead of the curve. Join us from November 9 - 12, 2009. Register now!
http://p.sf.net/sfu/devconference
_______________________________________________
ssic-linux-devel mailing list
ssic-linux-devel@...
https://lists.sourceforge.net/lists/listinfo/ssic-linux-devel