[PATCH] net/ceph/messenger: print TCP connect error

[PATCH] net/ceph/messenger: print TCP connect error

[Max Kellermann]

A non-blocking connect() usually returns `-EINPROGRESS` and fails
later, asynchronously; the actual error is then only visible as
`sk->sk_err` in the sk_state_change() callback and was silently
dropped.  All the user got was:

 libceph: mon4 (1)192.168.1.42:6789 socket closed (con state V1_BANNER)

Save the error in the new field `ceph_connection.error_code` and let
con_fault() print it next to `error_msg`:

 libceph: mon4 (1)192.168.1.42:6789 socket closed (con state V1_BANNER) (-ECONNREFUSED)

Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
---
 include/linux/ceph/messenger.h |  1 +
 net/ceph/messenger.c           | 23 ++++++++++++++++++-----
 2 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/include/linux/ceph/messenger.h b/include/linux/ceph/messenger.h
index 6aa4c6478c9f..f3aa977d7886 100644
--- a/include/linux/ceph/messenger.h
+++ b/include/linux/ceph/messenger.h
@@ -482,6 +482,7 @@ struct ceph_connection {
 
 	unsigned long flags;  /* CEPH_CON_F_* */
 	const char *error_msg;  /* error message, if any */
+	int error_code;  /* errno (negative), if any */
 
 	struct ceph_entity_name peer_name; /* peer name */
 	struct ceph_entity_addr peer_addr; /* peer address */
diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c
index 212e7797f9e4..f989e043686d 100644
--- a/net/ceph/messenger.c
+++ b/net/ceph/messenger.c
@@ -398,6 +398,7 @@ static void ceph_sock_state_change(struct sock *sk)
 	case TCP_CLOSE_WAIT:
 		dout("%s TCP_CLOSE_WAIT\n", __func__);
 		con_sock_state_closing(con);
+		WRITE_ONCE(con->error_code, -READ_ONCE(sk->sk_err));
 		ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED);
 		queue_con(con);
 		break;
@@ -448,8 +449,10 @@ int ceph_tcp_connect(struct ceph_connection *con)
 	ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
 			       SOCK_STREAM, IPPROTO_TCP, &sock);
 	memalloc_noio_restore(noio_flag);
-	if (ret)
+	if (ret) {
+		con->error_code = ret;
 		return ret;
+	}
 	sock->sk->sk_allocation = GFP_NOFS;
 	sock->sk->sk_use_task_frag = false;
 
@@ -467,9 +470,9 @@ int ceph_tcp_connect(struct ceph_connection *con)
 		     ceph_pr_addr(&con->peer_addr),
 		     sock->sk->sk_state);
 	} else if (ret < 0) {
-		pr_err("connect %s error %d\n",
-		       ceph_pr_addr(&con->peer_addr), ret);
 		sock_release(sock);
+		/* reported by con_fault() */
+		con->error_code = ret;
 		return ret;
 	}
 
@@ -501,6 +504,7 @@ int ceph_con_close_socket(struct ceph_connection *con)
 	 * shut the socket down.
 	 */
 	ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED);
+	con->error_code = 0;
 
 	con_sock_state_closed(con);
 	return rc;
@@ -1613,12 +1617,21 @@ static void ceph_con_workfn(struct work_struct *work)
  */
 static void con_fault(struct ceph_connection *con)
 {
+	int error_code;
+
 	dout("fault %p state %d to peer %s\n",
 	     con, con->state, ceph_pr_addr(&con->peer_addr));
 
-	pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
-		ceph_pr_addr(&con->peer_addr), con->error_msg);
+	error_code = READ_ONCE(con->error_code);
+	if (error_code)
+		pr_warn("%s%lld %s %s (%pe)\n", ENTITY_NAME(con->peer_name),
+			ceph_pr_addr(&con->peer_addr), con->error_msg,
+			ERR_PTR(error_code));
+	else
+		pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
+			ceph_pr_addr(&con->peer_addr), con->error_msg);
 	con->error_msg = NULL;
+	con->error_code = 0;
 
 	WARN_ON(con->state == CEPH_CON_S_STANDBY ||
 		con->state == CEPH_CON_S_CLOSED);
-- 
2.47.3