hv_sock: Fix incorrect readable byte reporting in hvs_stream_has_data()

 * In f0c5827, the hvs_stream_has_data() function was modified to accurately return the number of readable bytes.
 * However, this caused an issue in the WSL2 environment where standard input redirection from Windows to Linux
   (e.g., echo hi | wsl -u root sh) silently failed, returning error code 1 without any warning.
 * This issue occurs when the payload of the current packet has been fully transmitted, but the descriptor has
   not yet been cleaned up in the recvmsg() path.
 * If the next packet is already waiting in the VMBus ring buffer, need_refill evaluated to false, incorrectly
   returning -EIO.
 * As a result, POLLERR was propagated to user space, leading to a premature termination of the stream and a
   broken pipe connection.
 * In 95e754d, this was modified to return 1 instead of -EIO to resolve the issue.
 * However, while returning an artificial positive byte count fixes the stdin pipe regression, it incorrectly reports
   the stream state to other hvsock users.
 * Since WSL2 interoperability uses blocking frame control messages on the hvsock channel, this false readability
   can manifest as an unexpected EOF or a protocol error.
 * Consequently, although rare, executable interoperability between WSL2 and the Windows host can be broken.
 * To address this, if the current descriptor is already consumed, we use the hv_pkt_iter_next() function to advance
   the iterator to the next packet, returning 0 if there are no more available packets.
 * Otherwise, it updates the cached receive state of the next packet and returns the length of the unread payload.
 * This approach prevents false errors caused by a stale recv_desc state while maintaining accurate reporting of unread
   bytes for the next packet.
 * This commit fixes f0c5827 and 95e754d.

Bug: Nevuly/WSL2-Rolling-Kernel-Issue#42, Nevuly/WSL2-Rolling-Kernel-Issue#43
Fixes: f0c5827, 95e754d.
Test: Build and type `echo echo hi |  wsl -u root sh` in Windows CMD
Signed-off-by: Yang Jeong Hun <onyxclover9931@gmail.com>

Author: Nevuly

net/vmw_vsock/hyperv_transport.c

@@ -703,18 +703,15 @@ static s64 hvs_stream_has_data(struct vsock_sock *vsk)
 	switch (hvs_channel_readable_payload(hvs->chan)) {
 	case 1:
 		need_refill = !hvs->recv_desc;
-		/*
-		 * recv_desc is not null, but recv_data_len is 0.
-		 * The current packet is drained but the ring buffer has more data.
-		 * return 1 to trigger EPOLLIN so recvmsg() can clean up the
-		 * current descriptor and fetch the next packet.
-		 */
-		if (!need_refill)
-			return 1;
-
-		hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
-		if (!hvs->recv_desc)
-			return -ENOBUFS;
+		if (!need_refill) {
+			hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc);
+			if (!hvs->recv_desc)
+				return 0;
+		} else {
+			hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
+			if (!hvs->recv_desc)
+				return -ENOBUFS;
+		}
 
 		ret = hvs_update_recv_data(hvs);
 		if (ret)