[PATCH] Fix poll() nfds check.

[Vadim Lobanov <vlobanov@speakeasy.net>]

Hi,

This is a trivial patch to fix the nfds check in the poll system call
implementation. Namely, OPEN_MAX no longer does anything important in
the kernel, and checking that nfds is greater than max_fdset AND greater
than OPEN_MAX therefore just seems wrong.

This brings up a slightly-tangential question: Why do the nfds checks
exist in select()/poll()? They're not strictly necessary, since bad
input will be caught later when we validate all the fds, one by one.
Furthermore, these checks optimize the handling of error cases (which
should be uncommon) while pessimizing correct usage of the syscalls
(which should be more common).

Signed-off-by: Vadim Lobanov <vlobanov@speakeasy.net>

diff -Npru linux-2.6.17-git25/fs/select.c linux-new/fs/select.c
--- linux-2.6.17-git25/fs/select.c	2006-07-05 19:06:56.000000000 -0700
+++ linux-new/fs/select.c	2006-07-05 19:10:51.000000000 -0700
@@ -671,7 +671,7 @@ int do_sys_poll(struct pollfd __user *uf
 	fdt = files_fdtable(current->files);
 	max_fdset = fdt->max_fdset;
 	rcu_read_unlock();
-	if (nfds > max_fdset && nfds > OPEN_MAX)
+	if (nfds > max_fdset)
 		return -EINVAL;

 	poll_initwait(&table);