]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
smc: fix mutex unlocks during link group creation
authorUrsula Braun <ubraun@linux.vnet.ibm.com>
Wed, 25 Oct 2017 09:01:44 +0000 (11:01 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 26 Oct 2017 09:00:29 +0000 (18:00 +0900)
Link group creation is synchronized with the smc_create_lgr_pending
lock. In smc_listen_work() this mutex is sometimes unlocked, even
though it has not been locked before. This issue will surface in
presence of the SMC rendezvous code.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/smc/af_smc.c

index 745f145d4c4d43316f3cb30083066456497fec3a..70f7640f509095ee2a6b959a021f1e938bab97b5 100644 (file)
@@ -808,7 +808,7 @@ static void smc_listen_work(struct work_struct *work)
                rc = local_contact;
                if (rc == -ENOMEM)
                        reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
-               goto decline_rdma;
+               goto decline_rdma_unlock;
        }
        link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
 
@@ -816,7 +816,7 @@ static void smc_listen_work(struct work_struct *work)
        rc = smc_buf_create(new_smc);
        if (rc) {
                reason_code = SMC_CLC_DECL_MEM;
-               goto decline_rdma;
+               goto decline_rdma_unlock;
        }
 
        smc_close_init(new_smc);
@@ -831,7 +831,7 @@ static void smc_listen_work(struct work_struct *work)
                                             buf_desc->mr_rx[SMC_SINGLE_LINK]);
                        if (rc) {
                                reason_code = SMC_CLC_DECL_INTERR;
-                               goto decline_rdma;
+                               goto decline_rdma_unlock;
                        }
                }
        }
@@ -839,15 +839,15 @@ static void smc_listen_work(struct work_struct *work)
 
        rc = smc_clc_send_accept(new_smc, local_contact);
        if (rc)
-               goto out_err;
+               goto out_err_unlock;
 
        /* receive SMC Confirm CLC message */
        reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
                                       SMC_CLC_CONFIRM);
        if (reason_code < 0)
-               goto out_err;
+               goto out_err_unlock;
        if (reason_code > 0)
-               goto decline_rdma;
+               goto decline_rdma_unlock;
        smc_conn_save_peer_info(new_smc, &cclc);
        if (local_contact == SMC_FIRST_CONTACT)
                smc_link_save_peer_info(link, &cclc);
@@ -855,34 +855,34 @@ static void smc_listen_work(struct work_struct *work)
        rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
        if (rc) {
                reason_code = SMC_CLC_DECL_INTERR;
-               goto decline_rdma;
+               goto decline_rdma_unlock;
        }
 
        if (local_contact == SMC_FIRST_CONTACT) {
                rc = smc_ib_ready_link(link);
                if (rc) {
                        reason_code = SMC_CLC_DECL_INTERR;
-                       goto decline_rdma;
+                       goto decline_rdma_unlock;
                }
                /* QP confirmation over RoCE fabric */
                reason_code = smc_serv_conf_first_link(new_smc);
                if (reason_code < 0) {
                        /* peer is not aware of a problem */
                        rc = reason_code;
-                       goto out_err;
+                       goto out_err_unlock;
                }
                if (reason_code > 0)
-                       goto decline_rdma;
+                       goto decline_rdma_unlock;
        }
 
        smc_tx_init(new_smc);
+       mutex_unlock(&smc_create_lgr_pending);
 
 out_connected:
        sk_refcnt_debug_inc(newsmcsk);
        if (newsmcsk->sk_state == SMC_INIT)
                newsmcsk->sk_state = SMC_ACTIVE;
 enqueue:
-       mutex_unlock(&smc_create_lgr_pending);
        lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
        if (lsmc->sk.sk_state == SMC_LISTEN) {
                smc_accept_enqueue(&lsmc->sk, newsmcsk);
@@ -896,6 +896,8 @@ static void smc_listen_work(struct work_struct *work)
        sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
        return;
 
+decline_rdma_unlock:
+       mutex_unlock(&smc_create_lgr_pending);
 decline_rdma:
        /* RDMA setup failed, switch back to TCP */
        smc_conn_free(&new_smc->conn);
@@ -907,6 +909,8 @@ static void smc_listen_work(struct work_struct *work)
        }
        goto out_connected;
 
+out_err_unlock:
+       mutex_unlock(&smc_create_lgr_pending);
 out_err:
        newsmcsk->sk_state = SMC_CLOSED;
        smc_conn_free(&new_smc->conn);