net: disable netpoll on fresh napis

[ Upstream commit 96e97bc ]

napi_disable() makes sure to set the NAPI_STATE_NPSVC bit to prevent
netpoll from accessing rings before init is complete. However, the
same is not done for fresh napi instances in netif_napi_add(),
even though we expect NAPI instances to be added as disabled.

This causes crashes during driver reconfiguration (enabling XDP,
changing the channel count) - if there is any printk() after
netif_napi_add() but before napi_enable().

To ensure memory ordering is correct we need to use RCU accessors.

Reported-by: Rob Sherwood <rsher@fb.com>
Fixes: 2d8bff1 ("netpoll: Close race condition between poll_one_napi and napi_disable")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

net/core/dev.c

@@ -4849,13 +4849,14 @@ void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
 		pr_err_once("netif_napi_add() called with weight %d on device %s\n",
 			    weight, dev->name);
 	napi->weight = weight;
-	list_add(&napi->dev_list, &dev->napi_list);
 	napi->dev = dev;
 #ifdef CONFIG_NETPOLL
 	spin_lock_init(&napi->poll_lock);
 	napi->poll_owner = -1;
 #endif
 	set_bit(NAPI_STATE_SCHED, &napi->state);
+	set_bit(NAPI_STATE_NPSVC, &napi->state);
+	list_add_rcu(&napi->dev_list, &dev->napi_list);
 }
 EXPORT_SYMBOL(netif_napi_add);
 

net/core/netpoll.c

@@ -178,7 +178,7 @@ static void poll_napi(struct net_device *dev)
 {
 	struct napi_struct *napi;
 
-	list_for_each_entry(napi, &dev->napi_list, dev_list) {
+	list_for_each_entry_rcu(napi, &dev->napi_list, dev_list) {
 		if (napi->poll_owner != smp_processor_id() &&
 		    spin_trylock(&napi->poll_lock)) {
 			poll_one_napi(napi);