How to handle interruped sys call "connect" ?

[Chris Torek]

In article <7380.272cbb05 at jetson.uh.edu> elee6i5 at jetson.uh.edu writes:
>Now the question is how to handle interruped system call "connect"

I will answer this first because this one is easy.  Just call connect
again.  Any decent kernel (possibly not including some distributed by
certain major manufacturers) will handle this correctly.

(Another possibility is to use nonblocking connects, but this gets hairy.)

None of this is necessary, though:

>I am trying to set up a system wherein a client tries to send many files
>for a remote server's consumption. The way I go about is
>	- open a new socket
>	- connect
>	- fork a child to send data
>   The parent contiues this in a loop. The children after sending the
>   data close their sockets.
>
>But the parent should also close the sockets on which data transfer has been
>completed in order to clear up the system table entry. 

All you need do is close the socket in the parent immediately after
forking.  Closing a socket, like closing a pipe, has no effect unless
you have the last instance of the descriptor.  In this case, the child
has one also.

>When connect is retried (called again) if it returns with EINTR,
>it fails again  with EISCONN (socket already connected or previous
>connect attempt not completed ;  whatever it means)

You have an indecent or defective kernel (probably based on the 4.2BSD
implementation rather than one of the more reasonable BSD releases).
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 405 2750)
Domain:	chris at cs.umd.edu	Path:	uunet!mimsy!chris