On Mon, 27 Sep 2004 21:56:10 +1000
Herbert Xu [off-list ref] wrote:

David S. Miller [off-list ref] wrote:

quoted

4) The controversial/RFC patch, dorking with neigh_forced_gc()

+			if (n->nud_state -= NUD_INCOMPLETE &&
+			    reap_incomplete == 0 &&
+			    time_after(jiffies,
+				       n->used + n->parms->retrans_time)) {
+				num_incomplete++;
+				goto next_ent;

That should either be time_before, or you need to swap the arguments.

Good catch, and it means that the code basically behaved
as if the NUD_INCOMPLETE tests weren't even there.

So, as mentioned in another email, this is what I'm using
in the end:

# This is a BitKeeper generated diff -Nru style patch.
#
# ChangeSet
#   2004/09/27 11:46:12-07:00 davem@nuts.davemloft.net 
#   [NET]: Remove INCOMPLETE checks from neigh_forced_gc().
#   
#   Signed-off-by: David S. Miller [off-list ref]
# 
# net/core/neighbour.c
#   2004/09/27 11:45:40-07:00 davem@nuts.davemloft.net +3 -11
#   [NET]: Remove INCOMPLETE checks from neigh_forced_gc().
# 
diff -Nru a/net/core/neighbour.c b/net/core/neighbour.c

--- a/net/core/neighbour.c	2004-09-27 11:55:57 -07:00
+++ b/net/core/neighbour.c	2004-09-27 11:55:57 -07:00

@@ -123,20 +123,12 @@
 		np = &tbl->hash_buckets[i];
 		while ((n = *np) != NULL) {
 			/* Neighbour record may be discarded if:
-			   - nobody refers to it.
-			   - it is not permanent
-			   - (NEW and probably wrong)
-			     INCOMPLETE entries are kept at least for
-			     n->parms->retrans_time, otherwise we could
-			     flood network with resolution requests.
-			     It is not clear, what is better table overflow
-			     or flooding.
+			 * - nobody refers to it.
+			 * - it is not permanent
 			 */
 			write_lock(&n->lock);
 			if (atomic_read(&n->refcnt) == 1 &&
-			    !(n->nud_state & NUD_PERMANENT) &&
-			    (n->nud_state != NUD_INCOMPLETE ||
-			     time_after(jiffies, n->used + n->parms->retrans_time))) {
+			    !(n->nud_state & NUD_PERMANENT)) {
 				*np	= n->next;
 				n->dead = 1;
 				shrunk	= 1;