[PATCH v1] netfilter: x_tables: fix ordering of get and update table private

[Rongguang Wei]

From: Rongguang Wei <weirongguang@xxxxxxxxxx>

Meet a kernel panic in ipt_do_table:
PANIC: "Unable to handle kernel paging request at virtual address 00706f746b736564"

and the stack is:
     PC: ffff5e1dbecf0750  [ipt_do_table+1432]
     LR: ffff5e1dbecf04e4  [ipt_do_table+812]
     SP: ffff8021f7643370  PSTATE: 20400009
    X29: ffff8021f7643390  X28: ffff802900c3990c  X27: ffffa0405245a000
    X26: ffff80400ad645a8  X25: ffffa0201c4d8000  X24: ffff5e1dbed00228
    X23: ffff80400ad64738  X22: 0000000000000000  X21: ffff80400ad64000
    X20: ffff802114980ae8  X19: ffff8021f7643570  X18: 00000007ea9ec175
    X17: 0000fffde7b52460  X16: ffff5e1e181e8f20  X15: 0000fffd9a0ae078
    X14: 610d273b56961dbc  X13: 0a08010100007ecb  X12: f5011880fd874f59
    X11: ffff5e1dbed10600  X10: ffffa0405245a000   X9: 569b063f004015d5
     X8: ffff80400ad64738   X7: 0000000000010002   X6: 0000000000000000
     X5: 0000000000000000   X4: 0000000000000000   X3: 0000000000000000
     X2: 0000000000000000   X1: 2e706f746b736564   X0: ffff80400ad65850
[ffff8021f7643390] ipt_do_table at ffff5e1dbecf074c [ip_tables]
[ffff8021f76434d0] iptable_filter_hook at ffff5e1dbfe700a4 [iptable_filter]
[ffff8021f76434f0] nf_hook_slow at ffff5e1e18c31c2c
[ffff8021f7643530] ip_forward at ffff5e1e18c41924
[ffff8021f76435a0] ip_rcv_finish at ffff5e1e18c3fddc
[ffff8021f76435d0] ip_rcv at ffff5e1e18c40214
[ffff8021f7643630] __netif_receive_skb_one_core at ffff5e1e18bbbed4
[ffff8021f7643670] __netif_receive_skb at ffff5e1e18bbbf3c
[ffff8021f7643690] process_backlog at ffff5e1e18bbd52c
[ffff8021f76436f0] __napi_poll at ffff5e1e18bbc464
[ffff8021f7643730] net_rx_action at ffff5e1e18bbc9a8

The panic happend in ipt_do_table function:

	private = READ_ONCE(table->private);
	jumpstack  = (struct ipt_entry **)private->jumpstack[cpu];
	[...]
	jumpstack[stackid++] = e;	// panic here

In vmcore, the cpu is 4, I read the private->jumpstack[cpu] is 007365325f6b6365,
this address between user and kernel address ranges which caused kernel panic.
Also the kmem shows that the private->jumpstack address is free.
It looks like we get a UAF address here.

But in xt_replace_table function:

	private = table->private;
	[...]
	smp_wmb();
	table->private = newtable_info;
	smp_mb();

It seems no chance to get a free private member in ipt_do_table.
May have a ordering error which looks impossible:

	smp_wmb();
	table->private = newtable_info;
	private = table->private;
	smp_mb();

we get table->private after we set new table->private. After that, the
private was free in xt_free_table_info and also used in ipt_do_table.
Here use READ_ONCE to ensure we get private before we set the new one.

Signed-off-by: Rongguang Wei <weirongguang@xxxxxxxxxx>
---
 net/netfilter/x_tables.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/netfilter/x_tables.c b/net/netfilter/x_tables.c
index da5d929c7c85..1ce7a4f268d6 100644
--- a/net/netfilter/x_tables.c
+++ b/net/netfilter/x_tables.c
@@ -1399,7 +1399,7 @@ xt_replace_table(struct xt_table *table,
 
 	/* Do the substitution. */
 	local_bh_disable();
-	private = table->private;
+	private = READ_ONCE(table->private);
 
 	/* Check inside lock: is the old number correct? */
 	if (num_counters != private->number) {
-- 
2.25.1