[RFC PATCH RESEND bpf-next 1/4] bpf: Rollback to text_poke when arch not supported ftrace direct call

[Xu Kuohai]

On 12/20/2022 10:13 AM, Pu Lehui wrote:
> From: Pu Lehui <pulehui at huawei.com>
> 
> The current bpf trampoline attach to kernel functions via ftrace direct
> call API, while text_poke is applied for bpf2bpf attach and tail call
> optimization. For architectures that do not support ftrace direct call,
> text_poke is still able to attach bpf trampoline to kernel functions.
> Let's relax it by rollback to text_poke when architecture not supported.
> 
> Signed-off-by: Pu Lehui <pulehui at huawei.com>
> ---
>   kernel/bpf/trampoline.c | 8 ++------
>   1 file changed, 2 insertions(+), 6 deletions(-)
> 
> diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
> index d6395215b849..386197a7952c 100644
> --- a/kernel/bpf/trampoline.c
> +++ b/kernel/bpf/trampoline.c
> @@ -228,15 +228,11 @@ static int modify_fentry(struct bpf_trampoline *tr, void *old_addr, void *new_ad
>   static int register_fentry(struct bpf_trampoline *tr, void *new_addr)
>   {
>   	void *ip = tr->func.addr;
> -	unsigned long faddr;
>   	int ret;
>   
> -	faddr = ftrace_location((unsigned long)ip);
> -	if (faddr) {
> -		if (!tr->fops)
> -			return -ENOTSUPP;
> +	if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS) &&
> +	    !!ftrace_location((unsigned long)ip))
>   		tr->func.ftrace_managed = true;
> -	}
>

After this patch, a kernel function with true trace_location will be patched
by bpf when CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS is disabled, which means
that a kernel function may be patched by both bpf and ftrace in a mutually
unaware way. This will cause ftrace and bpf_arch_text_poke to fail in a
somewhat random way if the function to be patched was already patched
by the other.

>   	if (bpf_trampoline_module_get(tr))
>   		return -ENOENT;