Re: [2.4.19] read(2) and page aligned buffers

[Adam Kropelin <akropel1@rochester.rr.com>]

On Thu, Nov 07, 2002 at 01:54:21AM -0500, Clayton Weaver wrote:
> Here is wrap_read() (assume that the appropriate #includes are there):

Your code is broken, and for exactly the same reason I told you
yesterday.

>       default: /* partial read */
> 	switch(errno) {
> 	case EINTR:    /* interrupted by signal */
> 	case EAGAIN:   /* O_NONBLOCK ? */
> 	  retval += tmpret;
> 	  break;

I repeat: Checking errno when read() has returned something other than
-1 is ILLEGAL. Period. This check is triggering early, thus giving your
supposed early EOF.

This is not even remotely a kernel issue.

--Adam