When using select/poll/epoll() with a non-blocking ISOTP socket to wait for when non-blocking write is possible, false EPOLLOUT event is sometimes returned. This can happen at least after sending a message which must be split to multiple CAN frames. The reason is that isotp_sendmsg() returns -EAGAIN when tx.state is not equal to ISOTP_IDLE and this behavior is not reflected in datagram_poll(), which is used in isotp_ops. This is fixed by introducing ISOTP-specific poll function, which suppresses the EPOLLOUT events in that case. This can be tested with a program available at https://github.com/wentasah/isotp-poll-test/blob/fd095b2242c49dc5d3e36faf5ac9f4f47fd002c7/isotp-poll-test.c which can trigger the problem on a vcan interface. When running the program as: ./isotp-poll-test -s 123 -d 321 -o it starts sending ISOTP messages that include increasing ASCII numbers. poll() is used to wait before the next transmission. With current mainline Linux, once the message length is greater than 7 bytes, write() returns -EAGAIN and the program terminates. This should not happen, because the previous poll() reported that the write() would not block. After applying this patch, the above command doesn't fail - if one runs some ISOTP reader such as: isotprecv -l -s 321 -d 123 vcan0 This test program can also show another problem. When running: ./isotp-poll-test -s 321 -d 123 -i -a and then in another terminal: ./isotp-poll-test -s 123 -d 321 -o The first program receives the messages and uses the counter values to check for lost messages. After a random number of iterations a lost message is always detected. I believe that ISOTP should be reliable protocol, at least on vcan, shouldn't it? Anyway, this patch doesn't try to address this problem. Signed-off-by: Michal Sojka <michal.sojka@xxxxxxx> Reported-by: Jakub Jira <jirajak2@xxxxxxxxxxx> --- Changelog: v2: Added waiting for isotp-specific wait queue: poll_wait(file, &so->wait, wait). --- net/can/isotp.c | 17 ++++++++++++++++- 1 file changed, 16 insertions(+), 1 deletion(-) diff --git a/net/can/isotp.c b/net/can/isotp.c index 9bc344851704..ec163e36ac53 100644 --- a/net/can/isotp.c +++ b/net/can/isotp.c @@ -1608,6 +1608,21 @@ static int isotp_init(struct sock *sk) return 0; } +static __poll_t isotp_poll(struct file *file, struct socket *sock, poll_table *wait) +{ + struct sock *sk = sock->sk; + struct isotp_sock *so = isotp_sk(sk); + + __poll_t mask = datagram_poll(file, sock, wait); + poll_wait(file, &so->wait, wait); + + /* Check for false positives due to TX state */ + if ((mask & EPOLLWRNORM) && (so->tx.state != ISOTP_IDLE)) + mask &= ~(EPOLLOUT | EPOLLWRNORM); + + return mask; +} + static int isotp_sock_no_ioctlcmd(struct socket *sock, unsigned int cmd, unsigned long arg) { @@ -1623,7 +1638,7 @@ static const struct proto_ops isotp_ops = { .socketpair = sock_no_socketpair, .accept = sock_no_accept, .getname = isotp_getname, - .poll = datagram_poll, + .poll = isotp_poll, .ioctl = isotp_sock_no_ioctlcmd, .gettstamp = sock_gettstamp, .listen = sock_no_listen, -- 2.39.2
