net/smc: fixes for -net
v2 cover net/smc: fixes for -net
Fixes for -net, addressing two races in SMC receive path and add a missing cleanup when the link group creating fails with ISM devices and a VLAN id.
v1->v2:
- added fixes tags
Karsten Graul (2):
net/smc: receive returns without data net/smc: receive pending data after RCV_SHUTDOWN
Ursula Braun (1):
net/smc: fix SMCD link group creation with VLAN id
net/smc/smc_core.c | 5 ++++- net/smc/smc_rx.c | 29 +++++++++++++++++++++-------- 2 files changed, 25 insertions(+), 9 deletions(-)
Signature · 1 line
2.17.1
PATCH v2 1/3 net/smc: fix SMCD link group creation with VLAN id
From: Ursula Braun <[email protected]>
If creation of an SMCD link group with VLAN id fails, the initial smc_ism_get_vlan() step has to be reverted as well.
--- net/smc/smc_core.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
net/smc/smc_core.c +4 -1
@@ -213,7 +213,7 @@ static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini) lgr =kzalloc (sizeof (*lgr),GFP_KERNEL );if (!lgr) { rc =SMC_CLC_DECL_MEM ; - goto out; + goto ism_put_vlan; } lgr->is_smcd = ini->is_smcd; lgr->sync_err =0 ;
@@ -289,6 +289,9 @@ static int smc_lgr_create(struct smc_sock *smc, struct smc_init_info *ini)smc_llc_link_clear (lnk); free_lgr:kfree (lgr); +ism_put_vlan: +if (ini->is_smcd && ini->vlan_id) +smc_ism_put_vlan (ini->ism_dev, ini->vlan_id); out:if (rc <0 ) {if (rc == -ENOMEM )
Signature · 1 line
2.17.1
PATCH v2 2/3 net/smc: receive returns without data
In reply to Karsten Graulsmc_cdc_rxed_any_close_or_senddone() is used as an end condition for the receive loop. This conflicts with smc_cdc_msg_recv_action() which could run in parallel and set the bits checked by smc_cdc_rxed_any_close_or_senddone() before the receive is processed. In that case we could return from receive with no data, although data is available. The same applies to smc_rx_wait().
Fix this by checking for RCV_SHUTDOWN only, which is set in smc_cdc_msg_recv_action() after the receive was actually processed.
--- net/smc/smc_rx.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
net/smc/smc_rx.c +1 -3
@@ -211,8 +211,7 @@ int smc_rx_wait(struct smc_sock *smc, long *timeo, rc =sk_wait_event (sk, timeo, sk->sk_err || sk->sk_shutdown &RCV_SHUTDOWN || -fcrit (conn) || -smc_cdc_rxed_any_close_or_senddone (conn), +fcrit (conn), &wait);remove_wait_queue (sk_sleep (sk), &wait);sk_clear_bit (SOCKWQ_ASYNC_WAITDATA , sk);
@@ -310,7 +309,6 @@ int smc_rx_recvmsg(struct smc_sock *smc, struct msghdr *msg,smc_rx_update_cons (smc,0 );if (sk->sk_shutdown &RCV_SHUTDOWN || -smc_cdc_rxed_any_close_or_senddone (conn) || conn->local_tx_ctrl.conn_state_flags.peer_conn_abort)break ;
Signature · 1 line
2.17.1
PATCH v2 3/3 net/smc: receive pending data after RCV_SHUTDOWN
In reply to Karsten Graulsmc_rx_recvmsg() first checks if data is available, and then if RCV_SHUTDOWN is set. There is a race when smc_cdc_msg_recv_action() runs in between these 2 checks, receives data and sets RCV_SHUTDOWN. In that case smc_rx_recvmsg() would return from receive without to process the available data.
Fix that with a final check for data available if RCV_SHUTDOWN is set. Move the check for data into a function and call it twice. And use the existing helper smc_rx_data_available().
--- net/smc/smc_rx.c | 25 ++++++++++++++++++++----- 1 file changed, 20 insertions(+), 5 deletions(-)
net/smc/smc_rx.c +20 -5
@@ -261,6 +261,18 @@ static int smc_rx_recv_urg(struct smc_sock *smc, struct msghdr *msg, int len,return -EAGAIN ; } +static bool smc_rx_recvmsg_data_available (struct smc_sock *smc) +{ +struct smc_connection *conn = &smc->conn; + +if (smc_rx_data_available (conn)) +return true; +else if (conn->urg_state ==SMC_URG_VALID ) +/* we received a single urgent Byte - skip */ +smc_rx_update_cons (smc,0 ); +return false; +} +/* smc_rx_recvmsg - receive data from RMBE * @msg: copy data to receive buffer * @pipe: copy data to pipe if set - indicates splice() call
@@ -302,15 +314,18 @@ int smc_rx_recvmsg(struct smc_sock *smc, struct msghdr *msg,if (read_done >= target || (pipe && read_done))break ; -if (atomic_read (&conn->bytes_to_rcv)) +if (smc_rx_recvmsg_data_available (smc)) goto copy; -else if (conn->urg_state ==SMC_URG_VALID ) -/* we received a single urgent Byte - skip */ -smc_rx_update_cons (smc,0 );if (sk->sk_shutdown &RCV_SHUTDOWN || - conn->local_tx_ctrl.conn_state_flags.peer_conn_abort) + conn->local_tx_ctrl.conn_state_flags.peer_conn_abort) { +/* smc_cdc_msg_recv_action() could have run after +* above smc_rx_recvmsg_data_available() +*/ +if (smc_rx_recvmsg_data_available (smc)) + goto copy;break ; + }if (read_done) { if (sk->sk_err ||
Signature · 1 line
2.17.1
Jakub Kicinski
AppliedQuoted 6 lines · Karsten Graul
Fixes for -net, addressing two races in SMC receive path and add a missing cleanup when the link group creating fails with ISM devices and a VLAN id.
v1->v2:
- added fixes tags
Applied, queued for stable