ipv4 route dump broken with multiple subnets

ipv4 route dump broken with multiple subnets

[Sabrina Dubroca]

With this setup (no other address assigned):

ip link add type dummy
ip link set dummy0 up
ip a a 10.0.0.1/24 dev dummy0

route dump works:

# ip r
10.0.0.0/24 dev dummy0 proto kernel scope link src 10.0.0.1


If I add another address:

ip a a 10.0.1.1/24 dev dummy0

or a route:

ip r a 10.0.1.0/24 via 10.0.0.2

# ip r
<empty>

It's the same with /proc/net/route instead of ip route.

But with a default route the table appears again.


This patch seems to fix the problem, but I'm not sure it is correct.

---

diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
index 90955455884e..418043dc3a12 100644
--- a/net/ipv4/fib_trie.c
+++ b/net/ipv4/fib_trie.c
@@ -1767,6 +1767,11 @@ int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
 	 */
 	int count = cb->args[2];
 	t_key key = cb->args[3];
+	if (!key) {
+		struct key_vector *n = get_child_rcu(tp, 0);
+		if (n)
+			key = n->key;
+	}
 
 	while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
 		if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
@@ -2276,10 +2281,12 @@ static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
 		pos -= iter->pos;
 		key = iter->key;
 	} else {
+		struct key_vector *n;
 		t = (struct trie *)tb->tb_data;
 		iter->tnode = t->kv;
 		iter->pos = 0;
-		key = 0;
+		n = rcu_dereference(t->kv->tnode[0]);
+		key = n ? n->key : 0;
 	}
 
 	while ((l = leaf_walk_rcu(tp, key)) != NULL) {
@@ -2308,6 +2315,7 @@ static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
 	__acquires(RCU)
 {
 	struct fib_route_iter *iter = seq->private;
+	struct key_vector *n;
 	struct fib_table *tb;
 	struct trie *t;
 
@@ -2323,9 +2331,15 @@ static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
 		return fib_route_get_idx(iter, *pos);
 
 	t = (struct trie *)tb->tb_data;
-	iter->tnode = t->kv;
+	n = rcu_dereference(t->kv->tnode[0]);
+	if (IS_TNODE(n)) {
+		iter->tnode = n;
+		iter->key = n->key;
+	} else {
+		iter->tnode = t->kv;
+		iter->key = 0;
+	}
 	iter->pos = 0;
-	iter->key = 0;
 
 	return SEQ_START_TOKEN;
 }

Re: ipv4 route dump broken with multiple subnets

[Alexander Duyck]

On 03/10/2015 10:07 AM, Sabrina Dubroca wrote:
> With this setup (no other address assigned):
>
> ip link add type dummy
> ip link set dummy0 up
> ip a a 10.0.0.1/24 dev dummy0
>
> route dump works:
>
> # ip r
> 10.0.0.0/24 dev dummy0 proto kernel scope link src 10.0.0.1
>
>
> If I add another address:
>
> ip a a 10.0.1.1/24 dev dummy0
>
> or a route:
>
> ip r a 10.0.1.0/24 via 10.0.0.2
>
> # ip r
> <empty>
>
> It's the same with /proc/net/route instead of ip route.
>
> But with a default route the table appears again.
>
>
> This patch seems to fix the problem, but I'm not sure it is correct.

It looks like I can reproduce it as well if I just add dummy, and then 
drop the default route.  I will look into it and should be able to tell 
you what is going on in the next hour or two.

- Alex

[net-next PATCH] fib_trie: Correctly handle case of key == 0 in leaf_walk_rcu

[Alexander Duyck]

In the case of a trie that had no tnodes with a key of 0 the initial
look-up would fail resulting in an out-of-bounds cindex on the first tnode.
This resulted in an entire trie being skipped.

In order resolve this I have updated the cindex logic in the initial
look-up so that if the key is zero we will always traverse the child zero
path.

Fixes: 8be33e95 ("fib_trie: Fib walk rcu should take a tnode and key instead of a trie and a leaf")
Reported-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
---
 net/ipv4/fib_trie.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/ipv4/fib_trie.c b/net/ipv4/fib_trie.c
index fcfa982..44cab1d 100644
--- a/net/ipv4/fib_trie.c
+++ b/net/ipv4/fib_trie.c
@@ -1530,7 +1530,7 @@ static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
 	do {
 		/* record parent and next child index */
 		pn = n;
-		cindex = get_index(key, pn);
+		cindex = key ? get_index(key, pn) : 0;
 
 		if (cindex >> pn->bits)
 			break;

Re: [net-next PATCH] fib_trie: Correctly handle case of key == 0 in leaf_walk_rcu

[Sabrina Dubroca]

2015-03-10, 11:25:41 -0700, Alexander Duyck wrote:
> In the case of a trie that had no tnodes with a key of 0 the initial
> look-up would fail resulting in an out-of-bounds cindex on the first tnode.
> This resulted in an entire trie being skipped.
> 
> In order resolve this I have updated the cindex logic in the initial
> look-up so that if the key is zero we will always traverse the child zero
> path.
> 
> Fixes: 8be33e95 ("fib_trie: Fib walk rcu should take a tnode and key instead of a trie and a leaf")
> Reported-by: Sabrina Dubroca <sd@queasysnail.net>
> Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
> ---

Tested-by: Sabrina Dubroca <sd@queasysnail.net>

Thanks Alex.

-- 
Sabrina

Re: [net-next PATCH] fib_trie: Correctly handle case of key == 0 in leaf_walk_rcu

[David Miller]

From: Sabrina Dubroca <sd@queasysnail.net>
Date: Tue, 10 Mar 2015 19:48:39 +0100

> 2015-03-10, 11:25:41 -0700, Alexander Duyck wrote:
>> In the case of a trie that had no tnodes with a key of 0 the initial
>> look-up would fail resulting in an out-of-bounds cindex on the first tnode.
>> This resulted in an entire trie being skipped.
>> 
>> In order resolve this I have updated the cindex logic in the initial
>> look-up so that if the key is zero we will always traverse the child zero
>> path.
>> 
>> Fixes: 8be33e95 ("fib_trie: Fib walk rcu should take a tnode and key instead of a trie and a leaf")
>> Reported-by: Sabrina Dubroca <sd@queasysnail.net>
>> Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
>> ---
> 
> Tested-by: Sabrina Dubroca <sd@queasysnail.net>

Applied, thanks everyone.