Re: [PATCH v1 net 11/16] net: Fix a data-race around sysctl_mem.

[Steven Rostedt]

On Tue, 5 Jul 2022 22:21:25 -0700
Kuniyuki Iwashima <kuniyu@xxxxxxxxxx> wrote:

> --- a/include/trace/events/sock.h
> +++ b/include/trace/events/sock.h
> @@ -122,9 +122,9 @@ TRACE_EVENT(sock_exceed_buf_limit,
>  
>  	TP_printk("proto:%s sysctl_mem=%ld,%ld,%ld allocated=%ld sysctl_rmem=%d rmem_alloc=%d sysctl_wmem=%d wmem_alloc=%d wmem_queued=%d kind=%s",
>  		__entry->name,
> -		__entry->sysctl_mem[0],
> -		__entry->sysctl_mem[1],
> -		__entry->sysctl_mem[2],
> +		READ_ONCE(__entry->sysctl_mem[0]),
> +		READ_ONCE(__entry->sysctl_mem[1]),
> +		READ_ONCE(__entry->sysctl_mem[2]),

This is not reading anything to do with sysctl. It's reading the content of
what was recorded in the ring buffer.

That is, the READ_ONCE() here is not necessary, and if anything will break
user space parsing, as this is exported to user space to tell it how to
read the binary format in the ring buffer.

-- Steve


>  		__entry->allocated,
>  		__entry->sysctl_rmem,
>  		__entry->rmem_alloc,