Re: [PATCH 4 of 5] IPv6: fix lifetime calculation on temporary address creation

[Benoit Boissinot <benoit.boissinot@ens-lyon.org>]

On Wed, Apr 02, 2008 at 08:41:14AM +0900, YOSHIFUJI Hideaki / 吉藤英明 wrote:
> In article <20080401215657.GP475@pirzuine> (at Tue, 1 Apr 2008 23:56:57 +0200), Benoit Boissinot <benoit.boissinot@ens-lyon.org> says:
> 
> > On Fri, Mar 28, 2008 at 12:04:01PM +0900, YOSHIFUJI Hideaki / 吉藤英明 wrote:
> > > In article <ec551b4a5bb25cda00ff.1206305172@pirzuine> (at Sun, 23 Mar 2008 21:46:12 +0100), Benoit Boissinot <benoit.boissinot@ens-lyon.org> says:
> > > 
> > > > IPv6: fix lifetime calculation on temporary address creation
> > [snip]
> > > > +	now = jiffies;
> > > > +	elapsed = (now - ifp->tstamp) / HZ;
> > > > +	if (elapsed >= ifp->valid_lft)
> > > > +		tmp_valid_lft = 0;
> > > > +	else
> > > > +		tmp_valid_lft = min_t(__u32,
> > > > +				      ifp->valid_lft - elapsed,
> > > > +				      idev->cnf.temp_valid_lft);
> > > > +	if (elapsed >= ifp->prefered_lft)
> > > > +		tmp_prefered_lft = 0;
> > > > +	else
> > > > +		tmp_prefered_lft = min_t(__u32,
> > > > +					 ifp->prefered_lft - elapsed,
> > > > +					 idev->cnf.temp_prefered_lft - desync_factor / HZ);
> > > 
> > > Basically I agree, but it is possible to expire the temporary
> > > address AFTER public address, which is not good.  Please fix this.
> > 
> > do you mean because of the rounding of 'elapsed' ? otherwise I don't see
> > what the problem is, sorry.
> 
> Right.  Maybe we could substruct "now" by adj = (now - ifp->tstamp) % HZ;
>    now = jiffies
>    elapsed = (now - ifp->tstamp) / HZ;
>    now -= (now - ifp->tstamp) % HZ;

Why not just simply round it above:
elapsed = (now - ifp->tstamp + HZ - 1) / HZ; /* round it up */
and the rest stays the same.

Or do we care about having a lifetime a little bit (<1s) shorter ?

updated patch below:

IPv6: fix lifetime calculation on temporary address creation

The lifetime calculation was buggy since it copied the tstamp
from the associated public address.
If (now - ifp->prefered_lft)/HZ (ie the elapsed time since the
timestamp was set in the public address) was greater than
temp_prefered_lft, you would always get deprecated addresses.

This patch corrects the lifetime calculation by setting the tstamp
to "now" and calculating the remaining time from the public address.

Signed-off-by: Benoit Boissinot <benoit.boissinot@ens-lyon.org>

diff -r 00affc24c178 net/ipv6/addrconf.c
--- a/net/ipv6/addrconf.c	Sat Mar 22 00:39:16 2008 +0100
+++ b/net/ipv6/addrconf.c	Wed Apr 02 02:14:13 2008 +0200
@@ -775,8 +775,7 @@
 {
 	struct inet6_dev *idev = ifp->idev;
 	struct in6_addr addr, *tmpaddr;
-	unsigned long tmp_prefered_lft, tmp_valid_lft, tmp_cstamp, tmp_tstamp;
-	unsigned long regen_advance;
+	unsigned long tmp_prefered_lft, tmp_valid_lft, elapsed, regen_advance;
 	int tmp_plen;
 	int ret = 0;
 	int max_addresses;
@@ -825,16 +824,21 @@
 		goto out;
 	}
 	memcpy(&addr.s6_addr[8], idev->rndid, 8);
-	tmp_valid_lft = min_t(__u32,
-			      ifp->valid_lft,
-			      idev->cnf.temp_valid_lft);
-	tmp_prefered_lft = min_t(__u32,
-				 ifp->prefered_lft,
-				 idev->cnf.temp_prefered_lft - desync_factor / HZ);
+	elapsed = (jiffies - ifp->tstamp + HZ - 1) / HZ; /* round above */
+	if (elapsed >= ifp->valid_lft)
+		tmp_valid_lft = 0;
+	else
+		tmp_valid_lft = min_t(__u32,
+				      ifp->valid_lft - elapsed,
+				      idev->cnf.temp_valid_lft);
+	if (elapsed >= ifp->prefered_lft)
+		tmp_prefered_lft = 0;
+	else
+		tmp_prefered_lft = min_t(__u32,
+					 ifp->prefered_lft - elapsed,
+					 idev->cnf.temp_prefered_lft - desync_factor / HZ);
 	tmp_plen = ifp->prefix_len;
 	max_addresses = idev->cnf.max_addresses;
-	tmp_cstamp = ifp->cstamp;
-	tmp_tstamp = ifp->tstamp;
 	spin_unlock_bh(&ifp->lock);
 
 	regen_advance = idev->cnf.regen_max_retry *
@@ -878,8 +882,6 @@
 	ift->ifpub = ifp;
 	ift->valid_lft = tmp_valid_lft;
 	ift->prefered_lft = tmp_prefered_lft;
-	ift->cstamp = tmp_cstamp;
-	ift->tstamp = tmp_tstamp;
 	spin_unlock_bh(&ift->lock);
 
 	addrconf_dad_start(ift, 0);

-- 
:wq