[RFC,v4,01/17] vsock/virtio: read data from non-linear skb

Message ID 20230603204939.1598818-2-AVKrasnov@sberdevices.ru
State New
Headers
Series vsock: MSG_ZEROCOPY flag support |

Commit Message

Arseniy Krasnov June 3, 2023, 8:49 p.m. UTC
  This is preparation patch for non-linear skbuff handling. It replaces
direct calls of 'memcpy_to_msg()' with 'skb_copy_datagram_iter()'. Main
advantage of the second one is that is can handle paged part of the skb
by using 'kmap()' on each page, but if there are no pages in the skb,
it behaves like simple copying to iov iterator. This patch also adds
new field to the control block of skb - this value shows current offset
in the skb to read next portion of data (it doesn't matter linear it or
not). Idea is that 'skb_copy_datagram_iter()' handles both types of
skb internally - it just needs an offset from which to copy data from
the given skb. This offset is incremented on each read from skb. This
approach allows to avoid special handling of non-linear skbs:
1) We can't call 'skb_pull()' on it, because it updates 'data' pointer.
2) We need to update 'data_len' also on each read from this skb.

Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
---
 include/linux/virtio_vsock.h            |  1 +
 net/vmw_vsock/virtio_transport_common.c | 26 +++++++++++++++++--------
 2 files changed, 19 insertions(+), 8 deletions(-)
  

Comments

Bobby Eshleman June 12, 2023, 5:43 p.m. UTC | #1
On Sat, Jun 03, 2023 at 11:49:23PM +0300, Arseniy Krasnov wrote:
> This is preparation patch for non-linear skbuff handling. It replaces
> direct calls of 'memcpy_to_msg()' with 'skb_copy_datagram_iter()'. Main
> advantage of the second one is that is can handle paged part of the skb
> by using 'kmap()' on each page, but if there are no pages in the skb,
> it behaves like simple copying to iov iterator. This patch also adds
> new field to the control block of skb - this value shows current offset
> in the skb to read next portion of data (it doesn't matter linear it or
> not). Idea is that 'skb_copy_datagram_iter()' handles both types of
> skb internally - it just needs an offset from which to copy data from
> the given skb. This offset is incremented on each read from skb. This
> approach allows to avoid special handling of non-linear skbs:
> 1) We can't call 'skb_pull()' on it, because it updates 'data' pointer.
> 2) We need to update 'data_len' also on each read from this skb.
> 
> Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
> ---
>  include/linux/virtio_vsock.h            |  1 +
>  net/vmw_vsock/virtio_transport_common.c | 26 +++++++++++++++++--------
>  2 files changed, 19 insertions(+), 8 deletions(-)
> 
> diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
> index c58453699ee9..17dbb7176e37 100644
> --- a/include/linux/virtio_vsock.h
> +++ b/include/linux/virtio_vsock.h
> @@ -12,6 +12,7 @@
>  struct virtio_vsock_skb_cb {
>  	bool reply;
>  	bool tap_delivered;
> +	u32 frag_off;
>  };
>  
>  #define VIRTIO_VSOCK_SKB_CB(skb) ((struct virtio_vsock_skb_cb *)((skb)->cb))
> diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
> index b769fc258931..5819a9cd4515 100644
> --- a/net/vmw_vsock/virtio_transport_common.c
> +++ b/net/vmw_vsock/virtio_transport_common.c
> @@ -355,7 +355,7 @@ virtio_transport_stream_do_peek(struct vsock_sock *vsk,
>  	spin_lock_bh(&vvs->rx_lock);
>  
>  	skb_queue_walk_safe(&vvs->rx_queue, skb,  tmp) {
> -		off = 0;
> +		off = VIRTIO_VSOCK_SKB_CB(skb)->frag_off;
>  
>  		if (total == len)
>  			break;
> @@ -370,7 +370,10 @@ virtio_transport_stream_do_peek(struct vsock_sock *vsk,
>  			 */
>  			spin_unlock_bh(&vvs->rx_lock);
>  
> -			err = memcpy_to_msg(msg, skb->data + off, bytes);
> +			err = skb_copy_datagram_iter(skb, off,
> +						     &msg->msg_iter,
> +						     bytes);
> +
>  			if (err)
>  				goto out;
>  
> @@ -414,24 +417,28 @@ virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
>  		skb = skb_peek(&vvs->rx_queue);
>  
>  		bytes = len - total;
> -		if (bytes > skb->len)
> -			bytes = skb->len;
> +		if (bytes > skb->len - VIRTIO_VSOCK_SKB_CB(skb)->frag_off)
> +			bytes = skb->len - VIRTIO_VSOCK_SKB_CB(skb)->frag_off;
>  
>  		/* sk_lock is held by caller so no one else can dequeue.
>  		 * Unlock rx_lock since memcpy_to_msg() may sleep.
>  		 */
>  		spin_unlock_bh(&vvs->rx_lock);
>  
> -		err = memcpy_to_msg(msg, skb->data, bytes);
> +		err = skb_copy_datagram_iter(skb,
> +					     VIRTIO_VSOCK_SKB_CB(skb)->frag_off,
> +					     &msg->msg_iter, bytes);
> +
>  		if (err)
>  			goto out;
>  
>  		spin_lock_bh(&vvs->rx_lock);
>  
>  		total += bytes;
> -		skb_pull(skb, bytes);
>  
> -		if (skb->len == 0) {
> +		VIRTIO_VSOCK_SKB_CB(skb)->frag_off += bytes;
> +
> +		if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->frag_off) {
>  			u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
>  
>  			virtio_transport_dec_rx_pkt(vvs, pkt_len);
> @@ -503,7 +510,10 @@ static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
>  				 */
>  				spin_unlock_bh(&vvs->rx_lock);
>  
> -				err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
> +				err = skb_copy_datagram_iter(skb, 0,
> +							     &msg->msg_iter,
> +							     bytes_to_copy);
> +
>  				if (err) {
>  					/* Copy of message failed. Rest of
>  					 * fragments will be freed without copy.
> -- 
> 2.25.1
> 

LGTM.

Reviewed-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
  
Stefano Garzarella June 26, 2023, 3:20 p.m. UTC | #2
On Sat, Jun 03, 2023 at 11:49:23PM +0300, Arseniy Krasnov wrote:
>This is preparation patch for non-linear skbuff handling. It replaces
>direct calls of 'memcpy_to_msg()' with 'skb_copy_datagram_iter()'. Main
>advantage of the second one is that is can handle paged part of the skb
>by using 'kmap()' on each page, but if there are no pages in the skb,
>it behaves like simple copying to iov iterator. This patch also adds
>new field to the control block of skb - this value shows current offset
>in the skb to read next portion of data (it doesn't matter linear it or
>not). Idea is that 'skb_copy_datagram_iter()' handles both types of
>skb internally - it just needs an offset from which to copy data from
>the given skb. This offset is incremented on each read from skb. This
>approach allows to avoid special handling of non-linear skbs:
>1) We can't call 'skb_pull()' on it, because it updates 'data' pointer.
>2) We need to update 'data_len' also on each read from this skb.
>
>Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
>---
> include/linux/virtio_vsock.h            |  1 +
> net/vmw_vsock/virtio_transport_common.c | 26 +++++++++++++++++--------
> 2 files changed, 19 insertions(+), 8 deletions(-)
>
>diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
>index c58453699ee9..17dbb7176e37 100644
>--- a/include/linux/virtio_vsock.h
>+++ b/include/linux/virtio_vsock.h
>@@ -12,6 +12,7 @@
> struct virtio_vsock_skb_cb {
> 	bool reply;
> 	bool tap_delivered;
>+	u32 frag_off;
> };
>
> #define VIRTIO_VSOCK_SKB_CB(skb) ((struct virtio_vsock_skb_cb *)((skb)->cb))
>diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>index b769fc258931..5819a9cd4515 100644
>--- a/net/vmw_vsock/virtio_transport_common.c
>+++ b/net/vmw_vsock/virtio_transport_common.c
>@@ -355,7 +355,7 @@ virtio_transport_stream_do_peek(struct vsock_sock *vsk,
> 	spin_lock_bh(&vvs->rx_lock);
>
> 	skb_queue_walk_safe(&vvs->rx_queue, skb,  tmp) {
>-		off = 0;
>+		off = VIRTIO_VSOCK_SKB_CB(skb)->frag_off;
>
> 		if (total == len)
> 			break;
>@@ -370,7 +370,10 @@ virtio_transport_stream_do_peek(struct vsock_sock *vsk,
> 			 */
> 			spin_unlock_bh(&vvs->rx_lock);
>
>-			err = memcpy_to_msg(msg, skb->data + off, bytes);
>+			err = skb_copy_datagram_iter(skb, off,
>+						     &msg->msg_iter,
>+						     bytes);
>+
> 			if (err)
> 				goto out;
>
>@@ -414,24 +417,28 @@ virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
> 		skb = skb_peek(&vvs->rx_queue);
>
> 		bytes = len - total;
>-		if (bytes > skb->len)
>-			bytes = skb->len;
>+		if (bytes > skb->len - VIRTIO_VSOCK_SKB_CB(skb)->frag_off)
>+			bytes = skb->len - VIRTIO_VSOCK_SKB_CB(skb)->frag_off;

What about storing `VIRTIO_VSOCK_SKB_CB(skb)->frag_off` in a variable?
More for readability than optimization, which I hope the compiler
already does on its own.

The rest LGTM.

Stefano

>
> 		/* sk_lock is held by caller so no one else can dequeue.
> 		 * Unlock rx_lock since memcpy_to_msg() may sleep.
> 		 */
> 		spin_unlock_bh(&vvs->rx_lock);
>
>-		err = memcpy_to_msg(msg, skb->data, bytes);
>+		err = skb_copy_datagram_iter(skb,
>+					     VIRTIO_VSOCK_SKB_CB(skb)->frag_off,
>+					     &msg->msg_iter, bytes);
>+
> 		if (err)
> 			goto out;
>
> 		spin_lock_bh(&vvs->rx_lock);
>
> 		total += bytes;
>-		skb_pull(skb, bytes);
>
>-		if (skb->len == 0) {
>+		VIRTIO_VSOCK_SKB_CB(skb)->frag_off += bytes;
>+
>+		if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->frag_off) {
> 			u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
>
> 			virtio_transport_dec_rx_pkt(vvs, pkt_len);
>@@ -503,7 +510,10 @@ static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
> 				 */
> 				spin_unlock_bh(&vvs->rx_lock);
>
>-				err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
>+				err = skb_copy_datagram_iter(skb, 0,
>+							     &msg->msg_iter,
>+							     bytes_to_copy);
>+
> 				if (err) {
> 					/* Copy of message failed. Rest of
> 					 * fragments will be freed without copy.
>-- 
>2.25.1
>
  

Patch

diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
index c58453699ee9..17dbb7176e37 100644
--- a/include/linux/virtio_vsock.h
+++ b/include/linux/virtio_vsock.h
@@ -12,6 +12,7 @@ 
 struct virtio_vsock_skb_cb {
 	bool reply;
 	bool tap_delivered;
+	u32 frag_off;
 };
 
 #define VIRTIO_VSOCK_SKB_CB(skb) ((struct virtio_vsock_skb_cb *)((skb)->cb))
diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
index b769fc258931..5819a9cd4515 100644
--- a/net/vmw_vsock/virtio_transport_common.c
+++ b/net/vmw_vsock/virtio_transport_common.c
@@ -355,7 +355,7 @@  virtio_transport_stream_do_peek(struct vsock_sock *vsk,
 	spin_lock_bh(&vvs->rx_lock);
 
 	skb_queue_walk_safe(&vvs->rx_queue, skb,  tmp) {
-		off = 0;
+		off = VIRTIO_VSOCK_SKB_CB(skb)->frag_off;
 
 		if (total == len)
 			break;
@@ -370,7 +370,10 @@  virtio_transport_stream_do_peek(struct vsock_sock *vsk,
 			 */
 			spin_unlock_bh(&vvs->rx_lock);
 
-			err = memcpy_to_msg(msg, skb->data + off, bytes);
+			err = skb_copy_datagram_iter(skb, off,
+						     &msg->msg_iter,
+						     bytes);
+
 			if (err)
 				goto out;
 
@@ -414,24 +417,28 @@  virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
 		skb = skb_peek(&vvs->rx_queue);
 
 		bytes = len - total;
-		if (bytes > skb->len)
-			bytes = skb->len;
+		if (bytes > skb->len - VIRTIO_VSOCK_SKB_CB(skb)->frag_off)
+			bytes = skb->len - VIRTIO_VSOCK_SKB_CB(skb)->frag_off;
 
 		/* sk_lock is held by caller so no one else can dequeue.
 		 * Unlock rx_lock since memcpy_to_msg() may sleep.
 		 */
 		spin_unlock_bh(&vvs->rx_lock);
 
-		err = memcpy_to_msg(msg, skb->data, bytes);
+		err = skb_copy_datagram_iter(skb,
+					     VIRTIO_VSOCK_SKB_CB(skb)->frag_off,
+					     &msg->msg_iter, bytes);
+
 		if (err)
 			goto out;
 
 		spin_lock_bh(&vvs->rx_lock);
 
 		total += bytes;
-		skb_pull(skb, bytes);
 
-		if (skb->len == 0) {
+		VIRTIO_VSOCK_SKB_CB(skb)->frag_off += bytes;
+
+		if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->frag_off) {
 			u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
 
 			virtio_transport_dec_rx_pkt(vvs, pkt_len);
@@ -503,7 +510,10 @@  static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
 				 */
 				spin_unlock_bh(&vvs->rx_lock);
 
-				err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
+				err = skb_copy_datagram_iter(skb, 0,
+							     &msg->msg_iter,
+							     bytes_to_copy);
+
 				if (err) {
 					/* Copy of message failed. Rest of
 					 * fragments will be freed without copy.