Jeongjun Park wrote:
In the previous commit, bpf_net_context handling was added to
tun_sendmsg() and do_xdp_generic(), but if you write code like this,
bpf_net_context overlaps in the call trace below, causing various
memory corruptions.
I'm no expert on this code, but commit 401cb7dae813 that introduced
bpf_net_ctx_set explicitly states that nested calls are allowed.
And the function does imply that:
static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
{
struct task_struct *tsk = current;
if (tsk->bpf_net_context != NULL)
return NULL;
bpf_net_ctx->ri.kern_flags = 0;
tsk->bpf_net_context = bpf_net_ctx;
return bpf_net_ctx;
}
<Call trace>
...
tun_sendmsg() // bpf_net_ctx_set()
tun_xdp_one()
do_xdp_generic() // bpf_net_ctx_set() <-- nested
...
This patch removes the bpf_net_context handling that exists in
do_xdp_generic() and modifies it to handle it in the parent function.
Is tun_xdp_one missing? That also calls do_xdp_generic.