net: ipv4: Set skb->dev for output route resolution
authorIdo Schimmel <idosch@mellanox.com>
Thu, 20 Dec 2018 17:03:27 +0000 (17:03 +0000)
committerDavid S. Miller <davem@davemloft.net>
Fri, 21 Dec 2018 00:42:39 +0000 (16:42 -0800)
commit21f9477537059344e4705719137f6695de93b662
tree448b5a47c5c8c3027dd972a402bd8887522693de
parent639c1b2625af66501b24c8082231a83d4f5ad56b
net: ipv4: Set skb->dev for output route resolution

When user requests to resolve an output route, the kernel synthesizes
an skb where the relevant parameters (e.g., source address) are set. The
skb is then passed to ip_route_output_key_hash_rcu() which might call
into the flow dissector in case a multipath route was hit and a nexthop
needs to be selected based on the multipath hash.

Since both 'skb->dev' and 'skb->sk' are not set, a warning is triggered
in the flow dissector [1]. The warning is there to prevent codepaths
from silently falling back to the standard flow dissector instead of the
BPF one.

Therefore, instead of removing the warning, set 'skb->dev' to the
loopback device, as its not used for anything but resolving the correct
namespace.

[1]
WARNING: CPU: 1 PID: 24819 at net/core/flow_dissector.c:764 __skb_flow_dissect+0x314/0x16b0
...
RSP: 0018:ffffa0df41fdf650 EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff8bcded232000 RCX: 0000000000000000
RDX: ffffa0df41fdf7e0 RSI: ffffffff98e415a0 RDI: ffff8bcded232000
RBP: ffffa0df41fdf760 R08: 0000000000000000 R09: 0000000000000000
R10: ffffa0df41fdf7e8 R11: ffff8bcdf27a3000 R12: ffffffff98e415a0
R13: ffffa0df41fdf7e0 R14: ffffffff98dd2980 R15: ffffa0df41fdf7e0
FS:  00007f46f6897680(0000) GS:ffff8bcdf7a80000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000055933e95f9a0 CR3: 000000021e636000 CR4: 00000000001006e0
Call Trace:
 fib_multipath_hash+0x28c/0x2d0
 ? fib_multipath_hash+0x28c/0x2d0
 fib_select_path+0x241/0x32f
 ? __fib_lookup+0x6a/0xb0
 ip_route_output_key_hash_rcu+0x650/0xa30
 ? __alloc_skb+0x9b/0x1d0
 inet_rtm_getroute+0x3f7/0xb80
 ? __alloc_pages_nodemask+0x11c/0x2c0
 rtnetlink_rcv_msg+0x1d9/0x2f0
 ? rtnl_calcit.isra.24+0x120/0x120
 netlink_rcv_skb+0x54/0x130
 rtnetlink_rcv+0x15/0x20
 netlink_unicast+0x20a/0x2c0
 netlink_sendmsg+0x2d1/0x3d0
 sock_sendmsg+0x39/0x50
 ___sys_sendmsg+0x2a0/0x2f0
 ? filemap_map_pages+0x16b/0x360
 ? __handle_mm_fault+0x108e/0x13d0
 __sys_sendmsg+0x63/0xa0
 ? __sys_sendmsg+0x63/0xa0
 __x64_sys_sendmsg+0x1f/0x30
 do_syscall_64+0x5a/0x120
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Fixes: d0e13a1488ad ("flow_dissector: lookup netns by skb->sk if skb->dev is NULL")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/route.c