OSDN Git Service

ch_ktls: Issue if connection offload fails
authorRohit Maheshwari <rohitm@chelsio.com>
Tue, 29 Sep 2020 17:44:23 +0000 (23:14 +0530)
committerDavid S. Miller <davem@davemloft.net>
Tue, 29 Sep 2020 20:57:03 +0000 (13:57 -0700)
Since driver first return success to tls_dev_add, if req to HW is
successful, but later if HW returns failure, that connection traffic
fails permanently and connection status remains unknown to stack.

v1->v2:
- removed conn_up from all places.

v2->v3:
- Corrected timeout handling.

Fixes: 34aba2c45024 ("cxgb4/chcr : Register to tls add and del callback")
Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.c
drivers/net/ethernet/chelsio/inline_crypto/ch_ktls/chcr_ktls.h

index 4609f1f..c8db78b 100644 (file)
@@ -125,60 +125,6 @@ out:
        return ret;
 }
 
-static int chcr_ktls_update_connection_state(struct chcr_ktls_info *tx_info,
-                                            int new_state)
-{
-       /* This function can be called from both rx (interrupt context) and tx
-        * queue contexts.
-        */
-       spin_lock_bh(&tx_info->lock);
-       switch (tx_info->connection_state) {
-       case KTLS_CONN_CLOSED:
-               tx_info->connection_state = new_state;
-               break;
-
-       case KTLS_CONN_ACT_OPEN_REQ:
-               /* only go forward if state is greater than current state. */
-               if (new_state <= tx_info->connection_state)
-                       break;
-               /* update to the next state and also initialize TCB */
-               tx_info->connection_state = new_state;
-               fallthrough;
-       case KTLS_CONN_ACT_OPEN_RPL:
-               /* if we are stuck in this state, means tcb init might not
-                * received by HW, try sending it again.
-                */
-               if (!chcr_init_tcb_fields(tx_info))
-                       tx_info->connection_state = KTLS_CONN_SET_TCB_REQ;
-               break;
-
-       case KTLS_CONN_SET_TCB_REQ:
-               /* only go forward if state is greater than current state. */
-               if (new_state <= tx_info->connection_state)
-                       break;
-               /* update to the next state and check if l2t_state is valid  */
-               tx_info->connection_state = new_state;
-               fallthrough;
-       case KTLS_CONN_SET_TCB_RPL:
-               /* Check if l2t state is valid, then move to ready state. */
-               if (cxgb4_check_l2t_valid(tx_info->l2te)) {
-                       tx_info->connection_state = KTLS_CONN_TX_READY;
-                       atomic64_inc(&tx_info->adap->ch_ktls_stats.ktls_tx_ctx);
-               }
-               break;
-
-       case KTLS_CONN_TX_READY:
-               /* nothing to be done here */
-               break;
-
-       default:
-               pr_err("unknown KTLS connection state\n");
-               break;
-       }
-       spin_unlock_bh(&tx_info->lock);
-
-       return tx_info->connection_state;
-}
 /*
  * chcr_ktls_act_open_req: creates TCB entry for ipv4 connection.
  * @sk - tcp socket.
@@ -298,27 +244,17 @@ static int chcr_setup_connection(struct sock *sk,
                return -EINVAL;
 
        tx_info->atid = atid;
-       tx_info->ip_family = sk->sk_family;
 
-       if (sk->sk_family == AF_INET) {
-               tx_info->ip_family = AF_INET;
+       if (tx_info->ip_family == AF_INET) {
                ret = chcr_ktls_act_open_req(sk, tx_info, atid);
 #if IS_ENABLED(CONFIG_IPV6)
        } else {
-               if (!sk->sk_ipv6only &&
-                   ipv6_addr_type(&sk->sk_v6_daddr) == IPV6_ADDR_MAPPED) {
-                       tx_info->ip_family = AF_INET;
-                       ret = chcr_ktls_act_open_req(sk, tx_info, atid);
-               } else {
-                       tx_info->ip_family = AF_INET6;
-                       ret = cxgb4_clip_get(tx_info->netdev,
-                                            (const u32 *)
-                                            &sk->sk_v6_rcv_saddr.s6_addr,
-                                            1);
-                       if (ret)
-                               goto out;
-                       ret = chcr_ktls_act_open_req6(sk, tx_info, atid);
-               }
+               ret = cxgb4_clip_get(tx_info->netdev, (const u32 *)
+                                    &sk->sk_v6_rcv_saddr,
+                                    1);
+               if (ret)
+                       return ret;
+               ret = chcr_ktls_act_open_req6(sk, tx_info, atid);
 #endif
        }
 
@@ -326,16 +262,21 @@ static int chcr_setup_connection(struct sock *sk,
         * success, if any other return type clear atid and return that failure.
         */
        if (ret) {
-               if (ret == NET_XMIT_CN)
+               if (ret == NET_XMIT_CN) {
                        ret = 0;
-               else
+               } else {
+#if IS_ENABLED(CONFIG_IPV6)
+                       /* clear clip entry */
+                       if (tx_info->ip_family == AF_INET6)
+                               cxgb4_clip_release(tx_info->netdev,
+                                                  (const u32 *)
+                                                  &sk->sk_v6_rcv_saddr,
+                                                  1);
+#endif
                        cxgb4_free_atid(t, atid);
-               goto out;
+               }
        }
 
-       /* update the connection state */
-       chcr_ktls_update_connection_state(tx_info, KTLS_CONN_ACT_OPEN_REQ);
-out:
        return ret;
 }
 
@@ -396,15 +337,9 @@ static void chcr_ktls_dev_del(struct net_device *netdev,
        struct chcr_ktls_ofld_ctx_tx *tx_ctx =
                                chcr_get_ktls_tx_context(tls_ctx);
        struct chcr_ktls_info *tx_info = tx_ctx->chcr_info;
-       struct sock *sk;
 
        if (!tx_info)
                return;
-       sk = tx_info->sk;
-
-       spin_lock(&tx_info->lock);
-       tx_info->connection_state = KTLS_CONN_CLOSED;
-       spin_unlock(&tx_info->lock);
 
        /* clear l2t entry */
        if (tx_info->l2te)
@@ -413,8 +348,8 @@ static void chcr_ktls_dev_del(struct net_device *netdev,
 #if IS_ENABLED(CONFIG_IPV6)
        /* clear clip entry */
        if (tx_info->ip_family == AF_INET6)
-               cxgb4_clip_release(netdev,
-                                  (const u32 *)&sk->sk_v6_daddr.in6_u.u6_addr8,
+               cxgb4_clip_release(netdev, (const u32 *)
+                                  &tx_info->sk->sk_v6_rcv_saddr,
                                   1);
 #endif
 
@@ -461,30 +396,22 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
 
        pi = netdev_priv(netdev);
        adap = pi->adapter;
+
+       atomic64_inc(&adap->ch_ktls_stats.ktls_tx_connection_open);
        if (direction == TLS_OFFLOAD_CTX_DIR_RX) {
                pr_err("not expecting for RX direction\n");
-               ret = -EINVAL;
                goto out;
        }
-       if (tx_ctx->chcr_info) {
-               ret = -EINVAL;
+
+       if (tx_ctx->chcr_info)
                goto out;
-       }
 
        tx_info = kvzalloc(sizeof(*tx_info), GFP_KERNEL);
-       if (!tx_info) {
-               ret = -ENOMEM;
+       if (!tx_info)
                goto out;
-       }
-
-       spin_lock_init(&tx_info->lock);
-
-       /* clear connection state */
-       spin_lock(&tx_info->lock);
-       tx_info->connection_state = KTLS_CONN_CLOSED;
-       spin_unlock(&tx_info->lock);
 
        tx_info->sk = sk;
+       spin_lock_init(&tx_info->lock);
        /* initialize tid and atid to -1, 0 is a also a valid id. */
        tx_info->tid = -1;
        tx_info->atid = -1;
@@ -495,10 +422,12 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
        tx_info->tx_chan = pi->tx_chan;
        tx_info->smt_idx = pi->smt_idx;
        tx_info->port_id = pi->port_id;
+       tx_info->prev_ack = 0;
+       tx_info->prev_win = 0;
 
        tx_info->rx_qid = chcr_get_first_rx_qid(adap);
        if (unlikely(tx_info->rx_qid < 0))
-               goto out2;
+               goto free_tx_info;
 
        tx_info->prev_seq = start_offload_tcp_sn;
        tx_info->tcp_start_seq_number = start_offload_tcp_sn;
@@ -506,18 +435,22 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
        /* save crypto keys */
        ret = chcr_ktls_save_keys(tx_info, crypto_info, direction);
        if (ret < 0)
-               goto out2;
+               goto free_tx_info;
 
        /* get peer ip */
        if (sk->sk_family == AF_INET) {
                memcpy(daaddr, &sk->sk_daddr, 4);
+               tx_info->ip_family = AF_INET;
 #if IS_ENABLED(CONFIG_IPV6)
        } else {
                if (!sk->sk_ipv6only &&
-                   ipv6_addr_type(&sk->sk_v6_daddr) == IPV6_ADDR_MAPPED)
+                   ipv6_addr_type(&sk->sk_v6_daddr) == IPV6_ADDR_MAPPED) {
                        memcpy(daaddr, &sk->sk_daddr, 4);
-               else
+                       tx_info->ip_family = AF_INET;
+               } else {
                        memcpy(daaddr, sk->sk_v6_daddr.in6_u.u6_addr8, 16);
+                       tx_info->ip_family = AF_INET6;
+               }
 #endif
        }
 
@@ -525,13 +458,13 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
        dst = sk_dst_get(sk);
        if (!dst) {
                pr_err("DST entry not found\n");
-               goto out2;
+               goto free_tx_info;
        }
        n = dst_neigh_lookup(dst, daaddr);
        if (!n || !n->dev) {
                pr_err("neighbour not found\n");
                dst_release(dst);
-               goto out2;
+               goto free_tx_info;
        }
        tx_info->l2te  = cxgb4_l2t_get(adap->l2t, n, n->dev, 0);
 
@@ -540,31 +473,86 @@ static int chcr_ktls_dev_add(struct net_device *netdev, struct sock *sk,
 
        if (!tx_info->l2te) {
                pr_err("l2t entry not found\n");
-               goto out2;
+               goto free_tx_info;
        }
 
-       tx_ctx->chcr_info = tx_info;
+       /* Driver shouldn't be removed until any single connection exists */
+       if (!try_module_get(THIS_MODULE))
+               goto free_l2t;
 
+       init_completion(&tx_info->completion);
        /* create a filter and call cxgb4_l2t_send to send the packet out, which
         * will take care of updating l2t entry in hw if not already done.
         */
-       ret = chcr_setup_connection(sk, tx_info);
-       if (ret)
-               goto out2;
+       tx_info->open_state = CH_KTLS_OPEN_PENDING;
 
-       /* Driver shouldn't be removed until any single connection exists */
-       if (!try_module_get(THIS_MODULE)) {
-               ret = -EINVAL;
-               goto out2;
+       if (chcr_setup_connection(sk, tx_info))
+               goto put_module;
+
+       /* Wait for reply */
+       wait_for_completion_timeout(&tx_info->completion, 30 * HZ);
+       spin_lock_bh(&tx_info->lock);
+       if (tx_info->open_state) {
+               /* need to wait for hw response, can't free tx_info yet. */
+               if (tx_info->open_state == CH_KTLS_OPEN_PENDING)
+                       tx_info->pending_close = true;
+               /* free the lock after the cleanup */
+               goto put_module;
        }
+       spin_unlock_bh(&tx_info->lock);
+
+       /* initialize tcb */
+       reinit_completion(&tx_info->completion);
+       /* mark it pending for hw response */
+       tx_info->open_state = CH_KTLS_OPEN_PENDING;
+
+       if (chcr_init_tcb_fields(tx_info))
+               goto free_tid;
+
+       /* Wait for reply */
+       wait_for_completion_timeout(&tx_info->completion, 30 * HZ);
+       spin_lock_bh(&tx_info->lock);
+       if (tx_info->open_state) {
+               /* need to wait for hw response, can't free tx_info yet. */
+               tx_info->pending_close = true;
+               /* free the lock after cleanup */
+               goto free_tid;
+       }
+       spin_unlock_bh(&tx_info->lock);
+
+       if (!cxgb4_check_l2t_valid(tx_info->l2te))
+               goto free_tid;
+
+       atomic64_inc(&adap->ch_ktls_stats.ktls_tx_ctx);
+       tx_ctx->chcr_info = tx_info;
 
-       atomic64_inc(&adap->ch_ktls_stats.ktls_tx_connection_open);
        return 0;
-out2:
-       kvfree(tx_info);
+
+free_tid:
+       chcr_ktls_mark_tcb_close(tx_info);
+#if IS_ENABLED(CONFIG_IPV6)
+       /* clear clip entry */
+       if (tx_info->ip_family == AF_INET6)
+               cxgb4_clip_release(netdev, (const u32 *)
+                                  &sk->sk_v6_rcv_saddr,
+                                  1);
+#endif
+       cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
+                        tx_info->tid, tx_info->ip_family);
+
+put_module:
+       /* release module refcount */
+       module_put(THIS_MODULE);
+free_l2t:
+       cxgb4_l2t_release(tx_info->l2te);
+free_tx_info:
+       if (tx_info->pending_close)
+               spin_unlock_bh(&tx_info->lock);
+       else
+               kvfree(tx_info);
 out:
        atomic64_inc(&adap->ch_ktls_stats.ktls_tx_connection_fail);
-       return ret;
+       return -1;
 }
 
 /*
@@ -627,20 +615,39 @@ static int chcr_ktls_cpl_act_open_rpl(struct adapter *adap,
        tx_info = lookup_atid(t, atid);
 
        if (!tx_info || tx_info->atid != atid) {
-               pr_err("tx_info or atid is not correct\n");
+               pr_err("%s: incorrect tx_info or atid\n", __func__);
                return -1;
        }
 
+       cxgb4_free_atid(t, atid);
+       tx_info->atid = -1;
+
+       spin_lock(&tx_info->lock);
+       /* HW response is very close, finish pending cleanup */
+       if (tx_info->pending_close) {
+               spin_unlock(&tx_info->lock);
+               if (!status) {
+                       /* it's a late success, tcb status is establised,
+                        * mark it close.
+                        */
+                       chcr_ktls_mark_tcb_close(tx_info);
+                       cxgb4_remove_tid(&tx_info->adap->tids, tx_info->tx_chan,
+                                        tid, tx_info->ip_family);
+               }
+               kvfree(tx_info);
+               return 0;
+       }
+
        if (!status) {
                tx_info->tid = tid;
                cxgb4_insert_tid(t, tx_info, tx_info->tid, tx_info->ip_family);
-
-               cxgb4_free_atid(t, atid);
-               tx_info->atid = -1;
-               /* update the connection state */
-               chcr_ktls_update_connection_state(tx_info,
-                                                 KTLS_CONN_ACT_OPEN_RPL);
+               tx_info->open_state = CH_KTLS_OPEN_SUCCESS;
+       } else {
+               tx_info->open_state = CH_KTLS_OPEN_FAILURE;
        }
+       spin_unlock(&tx_info->lock);
+
+       complete(&tx_info->completion);
        return 0;
 }
 
@@ -658,12 +665,22 @@ static int chcr_ktls_cpl_set_tcb_rpl(struct adapter *adap, unsigned char *input)
 
        t = &adap->tids;
        tx_info = lookup_tid(t, tid);
+
        if (!tx_info || tx_info->tid != tid) {
-               pr_err("tx_info or atid is not correct\n");
+               pr_err("%s: incorrect tx_info or tid\n", __func__);
                return -1;
        }
-       /* update the connection state */
-       chcr_ktls_update_connection_state(tx_info, KTLS_CONN_SET_TCB_RPL);
+
+       spin_lock(&tx_info->lock);
+       if (tx_info->pending_close) {
+               spin_unlock(&tx_info->lock);
+               kvfree(tx_info);
+               return 0;
+       }
+       tx_info->open_state = false;
+       spin_unlock(&tx_info->lock);
+
+       complete(&tx_info->completion);
        return 0;
 }
 
@@ -1845,7 +1862,6 @@ static int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
        u32 tls_end_offset, tcp_seq;
        struct tls_context *tls_ctx;
        struct sk_buff *local_skb;
-       int new_connection_state;
        struct sge_eth_txq *q;
        struct adapter *adap;
        unsigned long flags;
@@ -1868,15 +1884,6 @@ static int chcr_ktls_xmit(struct sk_buff *skb, struct net_device *dev)
        if (unlikely(!tx_info))
                goto out;
 
-       /* check the connection state, we don't need to pass new connection
-        * state, state machine will check and update the new state if it is
-        * stuck due to responses not received from HW.
-        * Start the tx handling only if state is KTLS_CONN_TX_READY.
-        */
-       new_connection_state = chcr_ktls_update_connection_state(tx_info, 0);
-       if (new_connection_state != KTLS_CONN_TX_READY)
-               goto out;
-
        /* don't touch the original skb, make a new skb to extract each records
         * and send them separately.
         */
index 4ae6ae3..c1651b1 100644 (file)
 #define CHCR_KTLS_WR_SIZE      (CHCR_PLAIN_TX_DATA_LEN +\
                                 sizeof(struct cpl_tx_sec_pdu))
 
-enum chcr_ktls_conn_state {
-       KTLS_CONN_CLOSED,
-       KTLS_CONN_ACT_OPEN_REQ,
-       KTLS_CONN_ACT_OPEN_RPL,
-       KTLS_CONN_SET_TCB_REQ,
-       KTLS_CONN_SET_TCB_RPL,
-       KTLS_CONN_TX_READY,
+enum ch_ktls_open_state {
+       CH_KTLS_OPEN_SUCCESS = 0,
+       CH_KTLS_OPEN_PENDING = 1,
+       CH_KTLS_OPEN_FAILURE = 2,
 };
 
 struct chcr_ktls_info {
        struct sock *sk;
-       spinlock_t lock; /* state machine lock */
+       spinlock_t lock; /* lock for pending_close */
        struct ktls_key_ctx key_ctx;
        struct adapter *adap;
        struct l2t_entry *l2te;
        struct net_device *netdev;
+       struct completion completion;
        u64 iv;
        u64 record_no;
        int tid;
@@ -58,13 +56,14 @@ struct chcr_ktls_info {
        u32 tcp_start_seq_number;
        u32 scmd0_short_seqno_numivs;
        u32 scmd0_short_ivgen_hdrlen;
-       enum chcr_ktls_conn_state connection_state;
        u16 prev_win;
        u8 tx_chan;
        u8 smt_idx;
        u8 port_id;
        u8 ip_family;
        u8 first_qset;
+       enum ch_ktls_open_state open_state;
+       bool pending_close;
 };
 
 struct chcr_ktls_ofld_ctx_tx {