On 6/6/23 15:46, Guillaume Nault wrote:
On Tue, Jun 06, 2023 at 08:24:54AM +0200, Mirsad Goran Todorovac wrote:
On 5/31/23 20:11, Guillaume Nault wrote:
I believe this condition should be relaxed to allow the case where
->sk_bound_dev_if is oif's master device (and maybe there are other
VRF cases to also consider).
I've tried something like this, but something makes the kernel stuck
here:
TEST: ping out, blocked by route - ns-B loopback IPv6 [ OK ]
TEST: ping out, device bind, blocked by route - ns-B loopback IPv6 [ OK ]
TEST: ping in, blocked by route - ns-A loopback IPv6 [ OK ]
TEST: ping out, unreachable route - ns-B loopback IPv6 [ OK ]
TEST: ping out, device bind, unreachable route - ns-B loopback IPv6 [ OK ]
#################################################################
With VRF
[hanged process and kernel won't shutdown]
The code is:
---
net/ipv6/ping.c | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/net/ipv6/ping.c b/net/ipv6/ping.c
index c4835dbdfcff..81293e902293 100644
--- a/net/ipv6/ping.c
+++ b/net/ipv6/ping.c
@@ -73,6 +73,9 @@ static int ping_v6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
struct rt6_info *rt;
struct pingfakehdr pfh;
struct ipcm6_cookie ipc6;
+ struct net *net = sock_net(sk);
+ struct net_device *dev = NULL;
+ struct net_device *mdev = NULL;
err = ping_common_sendmsg(AF_INET6, msg, len, &user_icmph,
sizeof(user_icmph));
@@ -111,10 +114,17 @@ static int ping_v6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
else if (!oif)
oif = np->ucast_oif;
+ if (oif) {
+ dev = dev_get_by_index(net, oif);
+ mdev = netdev_master_upper_dev_get(dev);
+ }
+
addr_type = ipv6_addr_type(daddr);
if ((__ipv6_addr_needs_scope_id(addr_type) && !oif) ||
(addr_type & IPV6_ADDR_MAPPED) ||
- (oif && sk->sk_bound_dev_if && oif != sk->sk_bound_dev_if))
+ (oif && sk->sk_bound_dev_if && oif != sk->sk_bound_dev_if &&
+ !(mdev && sk->sk_bound_dev_if &&
+ mdev != dev_get_by_index(net, sk->sk_bound_dev_if))))
return -EINVAL;
ipcm6_init_sk(&ipc6, np);
I am obviously doing something very stupid.
The problem is that dev_get_by_index() holds a reference on 'dev' which
your code never releases. Also netdev_master_upper_dev_get() needs rtnl
protection. These should have generated some kernel oops.
You can try this instead:
-------- >8 --------
diff --git a/net/ipv6/ping.c b/net/ipv6/ping.c
index c4835dbdfcff..f804c11e2146 100644
--- a/net/ipv6/ping.c
+++ b/net/ipv6/ping.c
@@ -114,7 +114,8 @@ static int ping_v6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
addr_type = ipv6_addr_type(daddr);
if ((__ipv6_addr_needs_scope_id(addr_type) && !oif) ||
(addr_type & IPV6_ADDR_MAPPED) ||
- (oif && sk->sk_bound_dev_if && oif != sk->sk_bound_dev_if))
+ (oif && sk->sk_bound_dev_if && oif != sk->sk_bound_dev_if &&
+ l3mdev_master_ifindex_by_index(sock_net(sk), oif) != sk->sk_bound_dev_if))
return -EINVAL;
ipcm6_init_sk(&ipc6, np);
The problem appears to be fixed:
# ./fcnal-test.sh
[...]
TEST: ping out, vrf device+address bind - ns-B loopback IPv6 [ OK ]
TEST: ping out, vrf device+address bind - ns-B IPv6 LLA [ OK ]
TEST: ping in - ns-A IPv6 [ OK ]
[...]
Tests passed: 888
Tests failed: 0
#
The test passed in both environments that manifested the bug.
Tested-by: Mirsad Todorovac <mirsad.todorovac@xxxxxxxxxxxx>
However, test on my AMD Ubuntu 22.04 box with 6.4-rc5 commit a4d7d7011219
has shown additional four failed tests:
root@host # grep -n FAIL ../fcnal-test-4.log
90:TEST: ping local, VRF bind - VRF IP [FAIL]
92:TEST: ping local, device bind - ns-A IP [FAIL]
116:TEST: ping local, VRF bind - VRF IP [FAIL]
118:TEST: ping local, device bind - ns-A IP [FAIL]
root@host #
But you would probably want me to file a separate bug report?
Best regards,
Mirsad
