Re: [PATCH net-next v3 2/3] sock: add MSG_ZEROCOPY notification mechanism based on msg_control

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

zijianzhang@ wrote:
> From: Zijian Zhang <zijianzhang@bytedance.com>
> 
> The MSG_ZEROCOPY flag enables copy avoidance for socket send calls.
> However, zerocopy is not a free lunch. Apart from the management of user
> pages, the combination of poll + recvmsg to receive notifications incurs
> unignorable overhead in the applications. The overhead of such sometimes
> might be more than the CPU savings from zerocopy. We try to solve this
> problem with a new notification mechanism based on msgcontrol.
> This new mechanism aims to reduce the overhead associated with receiving
> notifications by embedding them directly into user arguments passed with
> each sendmsg control message. By doing so, we can significantly reduce
> the complexity and overhead for managing notifications. In an ideal
> pattern, the user will keep calling sendmsg with SCM_ZC_NOTIFICATION
> msg_control, and the notification will be delivered as soon as possible.
> 
> Signed-off-by: Zijian Zhang <zijianzhang@bytedance.com>
> Signed-off-by: Xiaochun Lu <xiaochun.lu@bytedance.com>

> +#include <linux/types.h>
> +
>  /*
>   * Desired design of maximum size and alignment (see RFC2553)
>   */
> @@ -35,4 +37,12 @@ struct __kernel_sockaddr_storage {
>  #define SOCK_TXREHASH_DISABLED	0
>  #define SOCK_TXREHASH_ENABLED	1
>  
> +#define SOCK_ZC_INFO_MAX 128
> +
> +struct zc_info_elem {
> +	__u32 lo;
> +	__u32 hi;
> +	__u8 zerocopy;
> +};
> +
>  #endif /* _UAPI_LINUX_SOCKET_H */
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 8d6e638b5426..15da609be026 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -2842,6 +2842,74 @@ int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
>  	case SCM_RIGHTS:
>  	case SCM_CREDENTIALS:
>  		break;
> +	case SCM_ZC_NOTIFICATION: {
> +		int ret, i = 0;
> +		int cmsg_data_len, zc_info_elem_num;
> +		void __user	*usr_addr;
> +		struct zc_info_elem zc_info_kern[SOCK_ZC_INFO_MAX];
> +		unsigned long flags;
> +		struct sk_buff_head *q, local_q;
> +		struct sk_buff *skb, *tmp;
> +		struct sock_exterr_skb *serr;

minor: reverse xmas tree

> +
> +		if (!sock_flag(sk, SOCK_ZEROCOPY) || sk->sk_family == PF_RDS)
> +			return -EINVAL;

Is this mechanism supported for PF_RDS?
The next patch fails on PF_RDS + '-n'

> +
> +		cmsg_data_len = cmsg->cmsg_len - sizeof(struct cmsghdr);
> +		if (cmsg_data_len % sizeof(struct zc_info_elem))
> +			return -EINVAL;
> +
> +		zc_info_elem_num = cmsg_data_len / sizeof(struct zc_info_elem);
> +		if (!zc_info_elem_num || zc_info_elem_num > SOCK_ZC_INFO_MAX)
> +			return -EINVAL;
> +
> +		if (in_compat_syscall())
> +			usr_addr = compat_ptr(*(compat_uptr_t *)CMSG_DATA(cmsg));
> +		else
> +			usr_addr = (void __user *)*(void **)CMSG_DATA(cmsg);

The main design issue with this series is this indirection, rather
than passing the array of notifications as cmsg.

This trick circumvents having to deal with compat issues and having to
figure out copy_to_user in ____sys_sendmsg (as msg_control is an
in-kernel copy).

This is quite hacky, from an API design PoV.

As is passing a pointer, but expecting msg_controllen to hold the
length not of the pointer, but of the pointed to user buffer.

I had also hoped for more significant savings. Especially with the
higher syscall overhead due to meltdown and spectre mitigations vs
when MSG_ZEROCOPY was introduced and I last tried this optimization.

> +		if (!access_ok(usr_addr, cmsg_data_len))
> +			return -EFAULT;
> +
> +		q = &sk->sk_error_queue;
> +		skb_queue_head_init(&local_q);
> +		spin_lock_irqsave(&q->lock, flags);
> +		skb = skb_peek(q);
> +		while (skb && i < zc_info_elem_num) {
> +			struct sk_buff *skb_next = skb_peek_next(skb, q);
> +
> +			serr = SKB_EXT_ERR(skb);
> +			if (serr->ee.ee_errno == 0 &&
> +			    serr->ee.ee_origin == SO_EE_ORIGIN_ZEROCOPY) {
> +				zc_info_kern[i].hi = serr->ee.ee_data;
> +				zc_info_kern[i].lo = serr->ee.ee_info;
> +				zc_info_kern[i].zerocopy = !(serr->ee.ee_code
> +								& SO_EE_CODE_ZEROCOPY_COPIED);
> +				__skb_unlink(skb, q);
> +				__skb_queue_tail(&local_q, skb);
> +				i++;
> +			}
> +			skb = skb_next;
> +		}
> +		spin_unlock_irqrestore(&q->lock, flags);
> +
> +		ret = copy_to_user(usr_addr,
> +				   zc_info_kern,
> +					i * sizeof(struct zc_info_elem));
> +
> +		if (unlikely(ret)) {
> +			spin_lock_irqsave(&q->lock, flags);
> +			skb_queue_reverse_walk_safe(&local_q, skb, tmp) {
> +				__skb_unlink(skb, &local_q);
> +				__skb_queue_head(q, skb);
> +			}

Can just list_splice_init?

> +			spin_unlock_irqrestore(&q->lock, flags);
> +			return -EFAULT;
> +		}
> +
> +		while ((skb = __skb_dequeue(&local_q)))
> +			consume_skb(skb);
> +		break;
> +	}
>  	default:
>  		return -EINVAL;
>  	}
> -- 
> 2.20.1
>