[PATCH][2/4] poll()/select() timeout behavior

[PATCH][2/4] poll()/select() timeout behavior

[Bill Rugolsky Jr.]

This patch cause select() to return EINVAL when passed an un-normalized
timeval usec, and fixes up the range check in preparation for the patch
that follows.

Please apply.

	Bill Rugolsky

--- linux/fs/select.c	2004-02-20 14:22:53.641191351 -0500
+++ linux/fs/select.c	2004-02-20 14:23:01.891295743 -0500
@@ -291,8 +291,6 @@
  * Update: ERESTARTSYS breaks at least the xview clock binary, so
  * I'm trying ERESTARTNOHAND which restart only when you want to.
  */
-#define MAX_SELECT_SECONDS \
-	((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
 
 asmlinkage long
 sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp)
@@ -312,10 +310,10 @@
 			goto out_nofds;
 
 		ret = -EINVAL;
-		if (sec < 0 || usec < 0)
+		if (sec < 0 || usec < 0 || usec >= 1000000)
 			goto out_nofds;
 
-		if ((unsigned long) sec < MAX_SELECT_SECONDS) {
+		if ((unsigned long) sec < (MAX_SCHEDULE_TIMEOUT-1) / HZ - 1) {
 			timeout = ROUND_UP(usec, 1000000/HZ);
 			timeout += sec * (unsigned long) HZ;
 		}

Re: [PATCH][2/4] poll()/select() timeout behavior

[Andrew Morton]

"Bill Rugolsky Jr." <brugolsky@telemetry-investments.com> wrote:
>
> This patch cause select() to return EINVAL when passed an un-normalized
> timeval usec

Why?  The current behaviour is to do the right thing when passed a tv_usec
which is greater than 1,000,000.  Why not retain that?