On 10/27/25 3:20 PM, Gabriel Krisman Bertazi wrote:

Jens Axboe [off-list ref] writes:

quoted

On 10/24/25 9:49 AM, Gabriel Krisman Bertazi wrote:

quoted

Introduce a socket-specific io_uring_cmd to support
getsockname/getpeername via io_uring.  I made this an io_uring_cmd
instead of a new operation to avoid polluting the command namespace with
what is exclusively a socket operation.  In addition, since we don't
need to conform to existing interfaces, this merges the
getsockname/getpeername in a single operation, since the implementation
is pretty much the same.

This has been frequently requested, for instance at [1] and more
recently in the project Discord channel. The main use-case is to support
fixed socket file descriptors.

Just two nits below, otherwise looks good!

quoted

diff --git a/io_uring/cmd_net.c b/io_uring/cmd_net.c
index 27a09aa4c9d0..092844358729 100644
--- a/io_uring/cmd_net.c
+++ b/io_uring/cmd_net.c

@@ -132,6 +132,28 @@ static int io_uring_cmd_timestamp(struct socket *sock,
 	return -EAGAIN;
 }
 
+static int io_uring_cmd_getsockname(struct socket *sock,
+				    struct io_uring_cmd *cmd,
+				    unsigned int issue_flags)
+{
+	const struct io_uring_sqe *sqe = cmd->sqe;
+

Random newline.

Done, but this fix will totally ruin the diffstat.  :(

What do you mean, it'll look even better as you're now killing a
redundant line you added :)

quoted

quoted

+	struct sockaddr_storage address;
+	struct sockaddr __user *uaddr;
+	int __user *ulen;
+	unsigned int peer;
+
+	uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	ulen = u64_to_user_ptr(sqe->addr3);
+	peer = READ_ONCE(sqe->optlen);
+
+	if (sqe->ioprio || sqe->__pad1 || sqe->len || sqe->rw_flags)
+		return -EINVAL;

Most/all prep handlers tend to check these first, then proceed with
setting up if not set. Would probably make sense to mirror that here
too.

Ack. will wait a few days for feedback on the network side before the
v2.

Sounds good, thanks.

-- 
Jens Axboe