From: Archie Pusaka <[email protected]>
Whenever we disconnect a L2CAP connection, we would immediately
report a disconnection event (EPOLLHUP) to the upper layer, without
waiting for the response of the other device.
This patch offers an option to wait until we receive a disconnection
response before reporting disconnection event, by using the "how"
parameter in l2cap_sock_shutdown(). Therefore, upper layer can opt
to wait for disconnection response by shutdown(sock, SHUT_WR).
This can be used to enforce proper disconnection order in HID,
where the disconnection of the interrupt channel must be complete
before attempting to disconnect the control channel.
Signed-off-by: Archie Pusaka <[email protected]>
---
net/bluetooth/l2cap_sock.c | 30 +++++++++++++++++++++++-------
1 file changed, 23 insertions(+), 7 deletions(-)
diff --git a/net/bluetooth/l2cap_sock.c b/net/bluetooth/l2cap_sock.c
index 1cea42ee1e922..a995d2c51fa7f 100644
--- a/net/bluetooth/l2cap_sock.c
+++ b/net/bluetooth/l2cap_sock.c
@@ -1271,14 +1271,21 @@ static int l2cap_sock_shutdown(struct socket *sock, int how)
struct l2cap_conn *conn;
int err = 0;
- BT_DBG("sock %p, sk %p", sock, sk);
+ BT_DBG("sock %p, sk %p, how %d", sock, sk, how);
+
+ /* 'how' parameter is mapped to sk_shutdown as follows:
+ * SHUT_RD (0) --> RCV_SHUTDOWN (1)
+ * SHUT_WR (1) --> SEND_SHUTDOWN (2)
+ * SHUT_RDWR (2) --> SHUTDOWN_MASK (3)
+ */
+ how++;
if (!sk)
return 0;
lock_sock(sk);
- if (sk->sk_shutdown)
+ if ((sk->sk_shutdown & how) == how)
goto shutdown_already;
BT_DBG("Handling sock shutdown");
@@ -1301,11 +1308,20 @@ static int l2cap_sock_shutdown(struct socket *sock, int how)
* has already been actioned to close the L2CAP
* link such as by l2cap_disconnection_req().
*/
- if (sk->sk_shutdown)
- goto has_shutdown;
+ if ((sk->sk_shutdown & how) == how)
+ goto shutdown_matched;
}
- sk->sk_shutdown = SHUTDOWN_MASK;
+ /* Try setting the RCV_SHUTDOWN bit, return early if SEND_SHUTDOWN
+ * is already set
+ */
+ if ((how & RCV_SHUTDOWN) && !(sk->sk_shutdown & RCV_SHUTDOWN)) {
+ sk->sk_shutdown |= RCV_SHUTDOWN;
+ if ((sk->sk_shutdown & how) == how)
+ goto shutdown_matched;
+ }
+
+ sk->sk_shutdown |= SEND_SHUTDOWN;
release_sock(sk);
l2cap_chan_lock(chan);
@@ -1335,7 +1351,7 @@ static int l2cap_sock_shutdown(struct socket *sock, int how)
err = bt_sock_wait_state(sk, BT_CLOSED,
sk->sk_lingertime);
-has_shutdown:
+shutdown_matched:
l2cap_chan_put(chan);
sock_put(sk);
@@ -1363,7 +1379,7 @@ static int l2cap_sock_release(struct socket *sock)
bt_sock_unlink(&l2cap_sk_list, sk);
- err = l2cap_sock_shutdown(sock, 2);
+ err = l2cap_sock_shutdown(sock, SHUT_RDWR);
chan = l2cap_pi(sk)->chan;
l2cap_chan_hold(chan);
--
2.26.0.110.g2183baf09c-goog
Hi Archie,
> Whenever we disconnect a L2CAP connection, we would immediately
> report a disconnection event (EPOLLHUP) to the upper layer, without
> waiting for the response of the other device.
>
> This patch offers an option to wait until we receive a disconnection
> response before reporting disconnection event, by using the "how"
> parameter in l2cap_sock_shutdown(). Therefore, upper layer can opt
> to wait for disconnection response by shutdown(sock, SHUT_WR).
>
> This can be used to enforce proper disconnection order in HID,
> where the disconnection of the interrupt channel must be complete
> before attempting to disconnect the control channel.
>
> Signed-off-by: Archie Pusaka <[email protected]>
> ---
>
> net/bluetooth/l2cap_sock.c | 30 +++++++++++++++++++++++-------
> 1 file changed, 23 insertions(+), 7 deletions(-)
the patch looks fine to me. Do we have something in l2cap-tester or l2test that we can verify this with before I apply it.
Regards
Marcel
Hi Marcel,
Let me write a test for that.
However, I cannot seem to run l2cap-tester properly.
On raspberry pi, all of the tests failed to initiate.
On chromeOS, all BREDR tests pass (before and after the change), but
all LE tests timed out on init stage (before and after the change).
I need to find out what went wrong when I execute those tests first.
Thanks,
Archie
On Thu, 23 Apr 2020 at 01:42, Marcel Holtmann <[email protected]> wrote:
>
> Hi Archie,
>
> > Whenever we disconnect a L2CAP connection, we would immediately
> > report a disconnection event (EPOLLHUP) to the upper layer, without
> > waiting for the response of the other device.
> >
> > This patch offers an option to wait until we receive a disconnection
> > response before reporting disconnection event, by using the "how"
> > parameter in l2cap_sock_shutdown(). Therefore, upper layer can opt
> > to wait for disconnection response by shutdown(sock, SHUT_WR).
> >
> > This can be used to enforce proper disconnection order in HID,
> > where the disconnection of the interrupt channel must be complete
> > before attempting to disconnect the control channel.
> >
> > Signed-off-by: Archie Pusaka <[email protected]>
> > ---
> >
> > net/bluetooth/l2cap_sock.c | 30 +++++++++++++++++++++++-------
> > 1 file changed, 23 insertions(+), 7 deletions(-)
>
> the patch looks fine to me. Do we have something in l2cap-tester or l2test that we can verify this with before I apply it.
>
> Regards
>
> Marcel
>
Hi Archie,
> Whenever we disconnect a L2CAP connection, we would immediately
> report a disconnection event (EPOLLHUP) to the upper layer, without
> waiting for the response of the other device.
>
> This patch offers an option to wait until we receive a disconnection
> response before reporting disconnection event, by using the "how"
> parameter in l2cap_sock_shutdown(). Therefore, upper layer can opt
> to wait for disconnection response by shutdown(sock, SHUT_WR).
>
> This can be used to enforce proper disconnection order in HID,
> where the disconnection of the interrupt channel must be complete
> before attempting to disconnect the control channel.
>
> Signed-off-by: Archie Pusaka <[email protected]>
> ---
>
> net/bluetooth/l2cap_sock.c | 30 +++++++++++++++++++++++-------
> 1 file changed, 23 insertions(+), 7 deletions(-)
patch has been applied to bluetooth-next tree.
Regards
Marcel