Merge branch 'vsock-virtio-vhost-msg_zerocopy-preparations'

Arseniy Krasnov says:

====================
vsock/virtio/vhost: MSG_ZEROCOPY preparations

this patchset is first of three parts of another big patchset for
MSG_ZEROCOPY flag support:
https://lore.kernel.org/netdev/20230701063947.3422088-1-AVKrasnov@sberdevices.ru/

During review of this series, Stefano Garzarella <sgarzare@redhat.com>
suggested to split it for three parts to simplify review and merging:

1) virtio and vhost updates (for fragged skbs) <--- this patchset
2) AF_VSOCK updates (allows to enable MSG_ZEROCOPY mode and read
   tx completions) and update for Documentation/.
3) Updates for tests and utils.

This series enables handling of fragged skbs in virtio and vhost parts.
Newly logic won't be triggered, because SO_ZEROCOPY options is still
impossible to enable at this moment (next bunch of patches from big
set above will enable it).
====================

Link: https://lore.kernel.org/r/20230916130918.4105122-1-avkrasnov@salutedevices.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

drivers/vhost/vsock.c

@@ -114,6 +114,7 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
 		struct sk_buff *skb;
 		unsigned out, in;
 		size_t nbytes;
+		u32 offset;
 		int head;
 
 		skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
@@ -156,7 +157,8 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
 		}
 
 		iov_iter_init(&iov_iter, ITER_DEST, &vq->iov[out], in, iov_len);
-		payload_len = skb->len;
+		offset = VIRTIO_VSOCK_SKB_CB(skb)->offset;
+		payload_len = skb->len - offset;
 		hdr = virtio_vsock_hdr(skb);
 
 		/* If the packet is greater than the space available in the
@@ -197,8 +199,10 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
 			break;
 		}
 
-		nbytes = copy_to_iter(skb->data, payload_len, &iov_iter);
-		if (nbytes != payload_len) {
+		if (skb_copy_datagram_iter(skb,
+					   offset,
+					   &iov_iter,
+					   payload_len)) {
 			kfree_skb(skb);
 			vq_err(vq, "Faulted on copying pkt buf\n");
 			break;
@@ -212,13 +216,13 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
 		vhost_add_used(vq, head, sizeof(*hdr) + payload_len);
 		added = true;
 
-		skb_pull(skb, payload_len);
+		VIRTIO_VSOCK_SKB_CB(skb)->offset += payload_len;
 		total_len += payload_len;
 
 		/* If we didn't send all the payload we can requeue the packet
 		 * to send it with the next available buffer.
 		 */
-		if (skb->len > 0) {
+		if (VIRTIO_VSOCK_SKB_CB(skb)->offset < skb->len) {
 			hdr->flags |= cpu_to_le32(flags_to_restore);
 
 			/* We are queueing the same skb to handle

include/linux/virtio_vsock.h

@@ -12,6 +12,7 @@
 struct virtio_vsock_skb_cb {
 	bool reply;
 	bool tap_delivered;
+	u32 offset;
 };
 
 #define VIRTIO_VSOCK_SKB_CB(skb) ((struct virtio_vsock_skb_cb *)((skb)->cb))
@@ -159,6 +160,15 @@ struct virtio_transport {
 
 	/* Takes ownership of the packet */
 	int (*send_pkt)(struct sk_buff *skb);
+
+	/* Used in MSG_ZEROCOPY mode. Checks, that provided data
+	 * (number of buffers) could be transmitted with zerocopy
+	 * mode. If this callback is not implemented for the current
+	 * transport - this means that this transport doesn't need
+	 * extra checks and can perform zerocopy transmission by
+	 * default.
+	 */
+	bool (*can_msgzerocopy)(int bufs_num);
 };
 
 ssize_t

include/trace/events/vsock_virtio_transport_common.h

@@ -43,15 +43,17 @@ TRACE_EVENT(virtio_transport_alloc_pkt,
 		 __u32 len,
 		 __u16 type,
 		 __u16 op,
-		 __u32 flags
+		 __u32 flags,
+		 bool zcopy
 	),
 	TP_ARGS(
 		src_cid, src_port,
 		dst_cid, dst_port,
 		len,
 		type,
 		op,
-		flags
+		flags,
+		zcopy
 	),
 	TP_STRUCT__entry(
 		__field(__u32, src_cid)
@@ -62,6 +64,7 @@ TRACE_EVENT(virtio_transport_alloc_pkt,
 		__field(__u16, type)
 		__field(__u16, op)
 		__field(__u32, flags)
+		__field(bool, zcopy)
 	),
 	TP_fast_assign(
 		__entry->src_cid = src_cid;
@@ -72,14 +75,15 @@ TRACE_EVENT(virtio_transport_alloc_pkt,
 		__entry->type = type;
 		__entry->op = op;
 		__entry->flags = flags;
+		__entry->zcopy = zcopy;
 	),
-	TP_printk("%u:%u -> %u:%u len=%u type=%s op=%s flags=%#x",
+	TP_printk("%u:%u -> %u:%u len=%u type=%s op=%s flags=%#x zcopy=%s",
 		  __entry->src_cid, __entry->src_port,
 		  __entry->dst_cid, __entry->dst_port,
 		  __entry->len,
 		  show_type(__entry->type),
 		  show_op(__entry->op),
-		  __entry->flags)
+		  __entry->flags, __entry->zcopy ? "true" : "false")
 );
 
 TRACE_EVENT(virtio_transport_recv_pkt,

net/vmw_vsock/virtio_transport.c

@@ -63,6 +63,17 @@ struct virtio_vsock {
 
 	u32 guest_cid;
 	bool seqpacket_allow;
+
+	/* These fields are used only in tx path in function
+	 * 'virtio_transport_send_pkt_work()', so to save
+	 * stack space in it, place both of them here. Each
+	 * pointer from 'out_sgs' points to the corresponding
+	 * element in 'out_bufs' - this is initialized in
+	 * 'virtio_vsock_probe()'. Both fields are protected
+	 * by 'tx_lock'. +1 is needed for packet header.
+	 */
+	struct scatterlist *out_sgs[MAX_SKB_FRAGS + 1];
+	struct scatterlist out_bufs[MAX_SKB_FRAGS + 1];
 };
 
 static u32 virtio_transport_get_local_cid(void)
@@ -100,8 +111,8 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 	vq = vsock->vqs[VSOCK_VQ_TX];
 
 	for (;;) {
-		struct scatterlist hdr, buf, *sgs[2];
 		int ret, in_sg = 0, out_sg = 0;
+		struct scatterlist **sgs;
 		struct sk_buff *skb;
 		bool reply;
 
@@ -111,12 +122,43 @@ virtio_transport_send_pkt_work(struct work_struct *work)
 
 		virtio_transport_deliver_tap_pkt(skb);
 		reply = virtio_vsock_skb_reply(skb);
-
-		sg_init_one(&hdr, virtio_vsock_hdr(skb), sizeof(*virtio_vsock_hdr(skb)));
-		sgs[out_sg++] = &hdr;
-		if (skb->len > 0) {
-			sg_init_one(&buf, skb->data, skb->len);
-			sgs[out_sg++] = &buf;
+		sgs = vsock->out_sgs;
+		sg_init_one(sgs[out_sg], virtio_vsock_hdr(skb),
+			    sizeof(*virtio_vsock_hdr(skb)));
+		out_sg++;
+
+		if (!skb_is_nonlinear(skb)) {
+			if (skb->len > 0) {
+				sg_init_one(sgs[out_sg], skb->data, skb->len);
+				out_sg++;
+			}
+		} else {
+			struct skb_shared_info *si;
+			int i;
+
+			/* If skb is nonlinear, then its buffer must contain
+			 * only header and nothing more. Data is stored in
+			 * the fragged part.
+			 */
+			WARN_ON_ONCE(skb_headroom(skb) != sizeof(*virtio_vsock_hdr(skb)));
+
+			si = skb_shinfo(skb);
+
+			for (i = 0; i < si->nr_frags; i++) {
+				skb_frag_t *skb_frag = &si->frags[i];
+				void *va;
+
+				/* We will use 'page_to_virt()' for the userspace page
+				 * here, because virtio or dma-mapping layers will call
+				 * 'virt_to_phys()' later to fill the buffer descriptor.
+				 * We don't touch memory at "virtual" address of this page.
+				 */
+				va = page_to_virt(skb_frag->bv_page);
+				sg_init_one(sgs[out_sg],
+					    va + skb_frag->bv_offset,
+					    skb_frag->bv_len);
+				out_sg++;
+			}
 		}
 
 		ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, skb, GFP_KERNEL);
@@ -413,6 +455,37 @@ static void virtio_vsock_rx_done(struct virtqueue *vq)
 	queue_work(virtio_vsock_workqueue, &vsock->rx_work);
 }
 
+static bool virtio_transport_can_msgzerocopy(int bufs_num)
+{
+	struct virtio_vsock *vsock;
+	bool res = false;
+
+	rcu_read_lock();
+
+	vsock = rcu_dereference(the_virtio_vsock);
+	if (vsock) {
+		struct virtqueue *vq = vsock->vqs[VSOCK_VQ_TX];
+
+		/* Check that tx queue is large enough to keep whole
+		 * data to send. This is needed, because when there is
+		 * not enough free space in the queue, current skb to
+		 * send will be reinserted to the head of tx list of
+		 * the socket to retry transmission later, so if skb
+		 * is bigger than whole queue, it will be reinserted
+		 * again and again, thus blocking other skbs to be sent.
+		 * Each page of the user provided buffer will be added
+		 * as a single buffer to the tx virtqueue, so compare
+		 * number of pages against maximum capacity of the queue.
+		 */
+		if (bufs_num <= vq->num_max)
+			res = true;
+	}
+
+	rcu_read_unlock();
+
+	return res;
+}
+
 static bool virtio_transport_seqpacket_allow(u32 remote_cid);
 
 static struct virtio_transport virtio_transport = {
@@ -462,6 +535,7 @@ static struct virtio_transport virtio_transport = {
 	},
 
 	.send_pkt = virtio_transport_send_pkt,
+	.can_msgzerocopy = virtio_transport_can_msgzerocopy,
 };
 
 static bool virtio_transport_seqpacket_allow(u32 remote_cid)
@@ -621,6 +695,7 @@ static int virtio_vsock_probe(struct virtio_device *vdev)
 {
 	struct virtio_vsock *vsock = NULL;
 	int ret;
+	int i;
 
 	ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
 	if (ret)
@@ -663,6 +738,9 @@ static int virtio_vsock_probe(struct virtio_device *vdev)
 	if (ret < 0)
 		goto out;
 
+	for (i = 0; i < ARRAY_SIZE(vsock->out_sgs); i++)
+		vsock->out_sgs[i] = &vsock->out_bufs[i];
+
 	rcu_assign_pointer(the_virtio_vsock, vsock);
 
 	mutex_unlock(&the_virtio_vsock_mutex);