Re: [PATCH net-net] tun: Assign missing bpf_net_context.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 9/12/24 3:17 PM, Breno Leitao wrote:
On Thu, Sep 12, 2024 at 02:28:47PM +0200, Sebastian Andrzej Siewior wrote:
On 2024-09-12 05:06:36 [-0700], Breno Leitao wrote:
Hello Sebastian, Jakub,
Hi,

I've seen some crashes in 6.11-rc7 that seems related to 401cb7dae8130
("net: Reference bpf_redirect_info via task_struct on PREEMPT_RT.").

Basically bpf_net_context is NULL, and it is being dereferenced by
bpf_net_ctx->ri.kern_flags (offset 0x38) in the following code.

	static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
	{
		struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
		if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {

That said, it means that bpf_net_ctx_get() is returning NULL.

This stack is coming from the bpf function bpf_redirect()
	BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
	{
	      struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();


Since I don't think there is XDP involved, I wondering if we need some
preotection before calling bpf_redirect()

This origins in netkit_xmit(). If my memory serves me, then Daniel told
me that netkit is not doing any redirect and therefore does not need
"this". This must have been during one of the first "designs"/ versions.

Right, I've seen several crashes related to this, and in all of them it
is through netkit_xmit() -> netkit_run() ->  bpf_prog_run()

If you are saying, that this is possible then something must be done.
Either assign a context or reject the bpf program.

If we want to assign a context, do you meant something like the
following?

Author: Breno Leitao <leitao@xxxxxxxxxx>
Date:   Thu Sep 12 06:11:28 2024 -0700

     netkit: Assign missing bpf_net_context.
During the introduction of struct bpf_net_context handling for
     XDP-redirect, the netkit driver has been missed.
Set the bpf_net_context before invoking netkit_xmit() program within the
     netkit driver.
Fixes: 401cb7dae8130 ("net: Reference bpf_redirect_info via task_struct on PREEMPT_RT.")
     Signed-off-by: Breno Leitao <leitao@xxxxxxxxxx>

Oh well, quite annoying that we need this context now everywhere also outside of XDP :(
Sebastian, do you see any way where this could be noop for !PREEMPT_RT?

Anyway, fix looks good to me, thanks!

Acked-by: Daniel Borkmann <daniel@xxxxxxxxxxxxx>

diff --git a/drivers/net/netkit.c b/drivers/net/netkit.c
index 79232f5cc088..f8af57b7a1e8 100644
--- a/drivers/net/netkit.c
+++ b/drivers/net/netkit.c
@@ -65,6 +65,7 @@ static struct netkit *netkit_priv(const struct net_device *dev)
static netdev_tx_t netkit_xmit(struct sk_buff *skb, struct net_device *dev)
  {
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
  	struct netkit *nk = netkit_priv(dev);
  	enum netkit_action ret = READ_ONCE(nk->policy);
  	netdev_tx_t ret_dev = NET_XMIT_SUCCESS;
@@ -72,6 +73,7 @@ static netdev_tx_t netkit_xmit(struct sk_buff *skb, struct net_device *dev)
  	struct net_device *peer;
  	int len = skb->len;
+ bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
  	rcu_read_lock();
  	peer = rcu_dereference(nk->peer);
  	if (unlikely(!peer || !(peer->flags & IFF_UP) ||
@@ -110,6 +112,7 @@ static netdev_tx_t netkit_xmit(struct sk_buff *skb, struct net_device *dev)
  		break;
  	}
  	rcu_read_unlock();
+	bpf_net_ctx_clear(bpf_net_ctx);
  	return ret_dev;
  }






[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux