Re: [bkpatch] add sys_sendfile64

[Chris Wedgwood <cw@f00f.org>]

On Sun, Mar 03, 2002 at 04:18:18PM -0500, Benjamin LaHaise wrote:

    The below bitkeeper patch can be pulled from
    	bk://bcrlbits.bkbits.net/linux-2.5

    and adds a sys_sendfile64 call.  Arch maintainers should 
    probably add the entry to their syscall tables if it is 
    appropriate.

Neat.  I made something similar last night but messed it up originally
so never sent it.

    +
    +asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t *offset, size_t count)
    +{
    +	loff_t pos, *ppos = NULL;
    +	ssize_t ret;
    +	if (offset) {
    +		off_t off;
    +		if (unlikely(get_user(off, offset)))
    +			return -EFAULT;
    +		pos = off;
    +		ppos = &pos;

We have a problem if off + count >= 2^32 here.

Ideally i think we need to check for 32-bit (31-bit?) overflow here
and return -EOVERFLOW.  I made a similar patch last night for
sendfile64 which included this check (although I was tired and the
patch was slightly wrong).  Actually, I think wew are missing
EOVERFLOW checks in a number of paths, ideally I'd like to make one
function to check and have all other functions reference that if
people agree that makes sense.

    +	}
    +	ret = common_sendfile(out_fd, in_fd, ppos, count);
    +	if (offset)
    +		put_user((off_t)pos, offset);
    +	return ret;

What is another thread unmapped 'offset' during the system call?  Do
we want to check the result of put_user here and return -EFAULT?
(If so, there are other system calls to consider such as select).




  --cw