[PATCH v3 0/3] Introduce getsockname io_uring_cmd

[PATCH v3 0/3] Introduce getsockname io_uring_cmd

[Gabriel Krisman Bertazi]

Since V2:
  - Comment fix (Kuniyuki)
  - Move sockaddr_storage into callee (Kuniyuki)
  - Add r-b tags
Since V1:
  - minor style fixes
  - Resend with (more) maintainers cc'ed
  - rebased to axboe/for-next.
--

This feature has been requested a few times in the liburing repository
and Discord channels, such as in [1,2].  If anything, it also helps
solve a long standing issue in the bind-listen test that results in
occasional test failures.

The patchset is divided in three parts: Patch 1 merges the getpeername
and getsockname implementation in the network layer, making further
patches easier; Patch 2 splits out a helper used by io_uring, like done
for other network commands; Finally, patch 3 plumbs the new command in
io_uring.

The syscall path was tested by booting a Linux distro, which does all
sorts of getsockname/getpeername syscalls.  The io_uring side was tested
with a couple of new liburing subtests available at:

   https://github.com/krisman/liburing.git -b socket

Based on top of Jens' for-next.

[1] https://github.com/axboe/liburing/issues/1356
[2] https://discord.com/channels/1241076672589991966/1241076672589991970/1429975797912830074

---
CC: netdev@vger.kernel.org
CC: io-uring@vger.kernel.org
CC: Jakub Kicinski <kuba@kernel.org>
CC: David S. Miller <davem@davemloft.net>
CC: Eric Dumazet <edumazet@google.com>
CC: Kuniyuki Iwashima <kuniyu@google.com>
CC: Paolo Abeni <pabeni@redhat.com>
CC: Willem de Bruijn <willemb@google.com>
CC: Simon Horman <horms@kernel.org>

Gabriel Krisman Bertazi (3):
  socket: Unify getsockname and getpeername implementation
  socket: Split out a getsockname helper for io_uring
  io_uring: Introduce getsockname io_uring cmd

 include/linux/socket.h        |  6 ++--
 include/uapi/linux/io_uring.h |  1 +
 io_uring/cmd_net.c            | 22 ++++++++++++
 net/compat.c                  |  4 +--
 net/socket.c                  | 67 +++++++++++------------------------
 5 files changed, 49 insertions(+), 51 deletions(-)

-- 
2.51.0

[PATCH v3 1/3] socket: Unify getsockname and getpeername implementation

[Gabriel Krisman Bertazi]

They are already implemented by the same get_name hook in the protocol
level.  Bring the unification one level up to reduce code duplication
in preparation to supporting these as io_uring operations.

Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>

---
v2 to v3:
- Adjust comment (Kuniyuki)
---
 include/linux/socket.h |  4 +--
 net/compat.c           |  4 +--
 net/socket.c           | 55 ++++++++++--------------------------------
 3 files changed, 16 insertions(+), 47 deletions(-)

diff --git a/include/linux/socket.h b/include/linux/socket.h
index 3b262487ec06..937fe331ff1e 100644
--- a/include/linux/socket.h
+++ b/include/linux/socket.h
@@ -454,9 +454,7 @@ extern int __sys_connect(int fd, struct sockaddr __user *uservaddr,
 extern int __sys_listen(int fd, int backlog);
 extern int __sys_listen_socket(struct socket *sock, int backlog);
 extern int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
-			     int __user *usockaddr_len);
-extern int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
-			     int __user *usockaddr_len);
+			     int __user *usockaddr_len, int peer);
 extern int __sys_socketpair(int family, int type, int protocol,
 			    int __user *usockvec);
 extern int __sys_shutdown_sock(struct socket *sock, int how);
diff --git a/net/compat.c b/net/compat.c
index 485db8ee9b28..2c9bd0edac99 100644
--- a/net/compat.c
+++ b/net/compat.c
@@ -460,10 +460,10 @@ COMPAT_SYSCALL_DEFINE2(socketcall, int, call, u32 __user *, args)
 		ret = __sys_accept4(a0, compat_ptr(a1), compat_ptr(a[2]), 0);
 		break;
 	case SYS_GETSOCKNAME:
-		ret = __sys_getsockname(a0, compat_ptr(a1), compat_ptr(a[2]));
+		ret = __sys_getsockname(a0, compat_ptr(a1), compat_ptr(a[2]), 0);
 		break;
 	case SYS_GETPEERNAME:
-		ret = __sys_getpeername(a0, compat_ptr(a1), compat_ptr(a[2]));
+		ret = __sys_getsockname(a0, compat_ptr(a1), compat_ptr(a[2]), 1);
 		break;
 	case SYS_SOCKETPAIR:
 		ret = __sys_socketpair(a0, a1, a[2], compat_ptr(a[3]));
diff --git a/net/socket.c b/net/socket.c
index e8892b218708..208d92ccf0fb 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -2128,12 +2128,11 @@ SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
 }
 
 /*
- *	Get the local address ('name') of a socket object. Move the obtained
- *	name to user space.
+ *	Get the remote or local address ('name') of a socket object. Move the
+ *	obtained name to user space.
  */
-
 int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
-		      int __user *usockaddr_len)
+		      int __user *usockaddr_len, int peer)
 {
 	struct socket *sock;
 	struct sockaddr_storage address;
@@ -2146,11 +2145,14 @@ int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
 	if (unlikely(!sock))
 		return -ENOTSOCK;
 
-	err = security_socket_getsockname(sock);
+	if (peer)
+		err = security_socket_getpeername(sock);
+	else
+		err = security_socket_getsockname(sock);
 	if (err)
 		return err;
 
-	err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 0);
+	err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, peer);
 	if (err < 0)
 		return err;
 
@@ -2161,44 +2163,13 @@ int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
 		int __user *, usockaddr_len)
 {
-	return __sys_getsockname(fd, usockaddr, usockaddr_len);
-}
-
-/*
- *	Get the remote address ('name') of a socket object. Move the obtained
- *	name to user space.
- */
-
-int __sys_getpeername(int fd, struct sockaddr __user *usockaddr,
-		      int __user *usockaddr_len)
-{
-	struct socket *sock;
-	struct sockaddr_storage address;
-	CLASS(fd, f)(fd);
-	int err;
-
-	if (fd_empty(f))
-		return -EBADF;
-	sock = sock_from_file(fd_file(f));
-	if (unlikely(!sock))
-		return -ENOTSOCK;
-
-	err = security_socket_getpeername(sock);
-	if (err)
-		return err;
-
-	err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 1);
-	if (err < 0)
-		return err;
-
-	/* "err" is actually length in this case */
-	return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
+	return __sys_getsockname(fd, usockaddr, usockaddr_len, 0);
 }
 
 SYSCALL_DEFINE3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
 		int __user *, usockaddr_len)
 {
-	return __sys_getpeername(fd, usockaddr, usockaddr_len);
+	return __sys_getsockname(fd, usockaddr, usockaddr_len, 1);
 }
 
 /*
@@ -3162,12 +3133,12 @@ SYSCALL_DEFINE2(socketcall, int, call, unsigned long __user *, args)
 	case SYS_GETSOCKNAME:
 		err =
 		    __sys_getsockname(a0, (struct sockaddr __user *)a1,
-				      (int __user *)a[2]);
+				      (int __user *)a[2], 0);
 		break;
 	case SYS_GETPEERNAME:
 		err =
-		    __sys_getpeername(a0, (struct sockaddr __user *)a1,
-				      (int __user *)a[2]);
+		    __sys_getsockname(a0, (struct sockaddr __user *)a1,
+				      (int __user *)a[2], 1);
 		break;
 	case SYS_SOCKETPAIR:
 		err = __sys_socketpair(a0, a1, a[2], (int __user *)a[3]);
-- 
2.51.0

[PATCH v3 2/3] socket: Split out a getsockname helper for io_uring

[Gabriel Krisman Bertazi]

Similar to getsockopt, split out a helper to check security and issue
the operation from the main handler that can be used by io_uring.

Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>

---
v2 -> v3:
- Move sockaddr_storage parameter into do_getsockname(Kuniyuki)
---
 include/linux/socket.h |  2 ++
 net/socket.c           | 36 ++++++++++++++++++++----------------
 2 files changed, 22 insertions(+), 16 deletions(-)

diff --git a/include/linux/socket.h b/include/linux/socket.h
index 937fe331ff1e..8d580074ddea 100644
--- a/include/linux/socket.h
+++ b/include/linux/socket.h
@@ -453,6 +453,8 @@ extern int __sys_connect(int fd, struct sockaddr __user *uservaddr,
 			 int addrlen);
 extern int __sys_listen(int fd, int backlog);
 extern int __sys_listen_socket(struct socket *sock, int backlog);
+extern int do_getsockname(struct socket *sock, int peer,
+			  struct sockaddr __user *usockaddr, int __user *usockaddr_len);
 extern int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
 			     int __user *usockaddr_len, int peer);
 extern int __sys_socketpair(int family, int type, int protocol,
diff --git a/net/socket.c b/net/socket.c
index 208d92ccf0fb..89bac0a17e5a 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -2127,39 +2127,43 @@ SYSCALL_DEFINE3(connect, int, fd, struct sockaddr __user *, uservaddr,
 	return __sys_connect(fd, uservaddr, addrlen);
 }
 
-/*
- *	Get the remote or local address ('name') of a socket object. Move the
- *	obtained name to user space.
- */
-int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
-		      int __user *usockaddr_len, int peer)
+int do_getsockname(struct socket *sock, int peer,
+		   struct sockaddr __user *usockaddr, int __user *usockaddr_len)
 {
-	struct socket *sock;
 	struct sockaddr_storage address;
-	CLASS(fd, f)(fd);
 	int err;
 
-	if (fd_empty(f))
-		return -EBADF;
-	sock = sock_from_file(fd_file(f));
-	if (unlikely(!sock))
-		return -ENOTSOCK;
-
 	if (peer)
 		err = security_socket_getpeername(sock);
 	else
 		err = security_socket_getsockname(sock);
 	if (err)
 		return err;
-
 	err = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, peer);
 	if (err < 0)
 		return err;
-
 	/* "err" is actually length in this case */
 	return move_addr_to_user(&address, err, usockaddr, usockaddr_len);
 }
 
+/*
+ *	Get the remote or local address ('name') of a socket object. Move the
+ *	obtained name to user space.
+ */
+int __sys_getsockname(int fd, struct sockaddr __user *usockaddr,
+		      int __user *usockaddr_len, int peer)
+{
+	struct socket *sock;
+	CLASS(fd, f)(fd);
+
+	if (fd_empty(f))
+		return -EBADF;
+	sock = sock_from_file(fd_file(f));
+	if (unlikely(!sock))
+		return -ENOTSOCK;
+	return do_getsockname(sock, peer, usockaddr, usockaddr_len);
+}
+
 SYSCALL_DEFINE3(getsockname, int, fd, struct sockaddr __user *, usockaddr,
 		int __user *, usockaddr_len)
 {
-- 
2.51.0

[PATCH v3 3/3] io_uring: Introduce getsockname io_uring cmd

[Gabriel Krisman Bertazi]

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.

[1] https://github.com/axboe/liburing/issues/1356

Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de>

---
v2->v3:
Don't pass sockaddr_storage pointer parameter
---
 include/uapi/linux/io_uring.h |  1 +
 io_uring/cmd_net.c            | 22 ++++++++++++++++++++++
 2 files changed, 23 insertions(+)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 3d921cbb84f8..6a97c5376019 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -1010,6 +1010,7 @@ enum io_uring_socket_op {
 	SOCKET_URING_OP_GETSOCKOPT,
 	SOCKET_URING_OP_SETSOCKOPT,
 	SOCKET_URING_OP_TX_TIMESTAMP,
+	SOCKET_URING_OP_GETSOCKNAME,
 };
 
 /*
diff --git a/io_uring/cmd_net.c b/io_uring/cmd_net.c
index 27a09aa4c9d0..a2d76157df4f 100644
--- a/io_uring/cmd_net.c
+++ b/io_uring/cmd_net.c
@@ -132,6 +132,26 @@ 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;
+	struct sockaddr __user *uaddr;
+	unsigned int peer;
+	int __user *ulen;
+
+	if (sqe->ioprio || sqe->__pad1 || sqe->len || sqe->rw_flags)
+		return -EINVAL;
+
+	uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	ulen = u64_to_user_ptr(sqe->addr3);
+	peer = READ_ONCE(sqe->optlen);
+	if (peer > 1)
+		return -EINVAL;
+	return do_getsockname(sock, 0, uaddr, ulen);
+}
+
 int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags)
 {
 	struct socket *sock = cmd->file->private_data;
@@ -159,6 +179,8 @@ int io_uring_cmd_sock(struct io_uring_cmd *cmd, unsigned int issue_flags)
 		return io_uring_cmd_setsockopt(sock, cmd, issue_flags);
 	case SOCKET_URING_OP_TX_TIMESTAMP:
 		return io_uring_cmd_timestamp(sock, cmd, issue_flags);
+	case SOCKET_URING_OP_GETSOCKNAME:
+		return io_uring_cmd_getsockname(sock, cmd, issue_flags);
 	default:
 		return -EOPNOTSUPP;
 	}
-- 
2.51.0

Re: [PATCH v3 3/3] io_uring: Introduce getsockname io_uring cmd

[Stefan Metzmacher]

Hi Gabriel,

> +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;
> +	struct sockaddr __user *uaddr;
> +	unsigned int peer;
> +	int __user *ulen;
> +
> +	if (sqe->ioprio || sqe->__pad1 || sqe->len || sqe->rw_flags)
> +		return -EINVAL;
> +
> +	uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
> +	ulen = u64_to_user_ptr(sqe->addr3);
> +	peer = READ_ONCE(sqe->optlen);
> +	if (peer > 1)
> +		return -EINVAL;
> +	return do_getsockname(sock, 0, uaddr, ulen);

I guess this should actually pass down 'peer' instead of '0'?

metze

Re: [PATCH v3 3/3] io_uring: Introduce getsockname io_uring cmd

[Gabriel Krisman Bertazi]

Stefan Metzmacher <metze@samba.org> writes:

> Hi Gabriel,
>
>> +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;
>> +	struct sockaddr __user *uaddr;
>> +	unsigned int peer;
>> +	int __user *ulen;
>> +
>> +	if (sqe->ioprio || sqe->__pad1 || sqe->len || sqe->rw_flags)
>> +		return -EINVAL;
>> +
>> +	uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
>> +	ulen = u64_to_user_ptr(sqe->addr3);
>> +	peer = READ_ONCE(sqe->optlen);
>> +	if (peer > 1)
>> +		return -EINVAL;
>> +	return do_getsockname(sock, 0, uaddr, ulen);
>
> I guess this should actually pass down 'peer' instead of '0'?

Thanks for the catch.  I guess two wrongs *do* make a right somethings.
The getpeername test was peeking at the wrong socket, effectively
testing the local port against itself, and it thus succeeded. Updated
the test and will send a v4.

-- 
Gabriel Krisman Bertazi