/*
* Connection established.
* We get here for both outgoing and incoming connection.
*/
void rds_iw_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
{
const struct rds_iw_connect_private *dp = NULL;
struct rds_iw_connection *ic = conn->c_transport_data;
struct rds_iw_device *rds_iwdev;
int err;
if (event->param.conn.private_data_len) {
dp = event->param.conn.private_data;
rds_iw_set_protocol(conn,
RDS_PROTOCOL(dp->dp_protocol_major,
dp->dp_protocol_minor));
rds_iw_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
}
/* update ib_device with this local ipaddr & conn */
rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
err = rds_iw_update_cm_id(rds_iwdev, ic->i_cm_id);
if (err)
printk(KERN_ERR "rds_iw_update_ipaddr failed (%d)\n", err);
rds_iw_add_conn(rds_iwdev, conn);
/* If the peer gave us the last packet it saw, process this as if
* we had received a regular ACK. */
if (dp && dp->dp_ack_seq)
rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
printk(KERN_NOTICE "RDS/IW: connected to %pI4<->%pI4 version %u.%u%s\n",
&conn->c_laddr, &conn->c_faddr,
RDS_PROTOCOL_MAJOR(conn->c_version),
RDS_PROTOCOL_MINOR(conn->c_version),
ic->i_flowctl ? ", flow control" : "");
rds_connect_complete(conn);
}
/*
* Connection established.
* We get here for both outgoing and incoming connection.
*/
void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
{
const struct rds_ib_connect_private *dp = NULL;
struct rds_ib_connection *ic = conn->c_transport_data;
struct ib_qp_attr qp_attr;
int err;
if (event->param.conn.private_data_len >= sizeof(*dp)) {
dp = event->param.conn.private_data;
/* make sure it isn't empty data */
if (dp->dp_protocol_major) {
rds_ib_set_protocol(conn,
RDS_PROTOCOL(dp->dp_protocol_major,
dp->dp_protocol_minor));
rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
}
}
if (conn->c_version < RDS_PROTOCOL(3, 1)) {
printk(KERN_NOTICE "RDS/IB: Connection to %pI4 version %u.%u failed,"
" no longer supported\n",
&conn->c_faddr,
RDS_PROTOCOL_MAJOR(conn->c_version),
RDS_PROTOCOL_MINOR(conn->c_version));
rds_conn_destroy(conn);
return;
} else {
printk(KERN_NOTICE "RDS/IB: connected to %pI4 version %u.%u%s\n",
&conn->c_faddr,
RDS_PROTOCOL_MAJOR(conn->c_version),
RDS_PROTOCOL_MINOR(conn->c_version),
ic->i_flowctl ? ", flow control" : "");
}
/*
* Init rings and fill recv. this needs to wait until protocol negotiation
* is complete, since ring layout is different from 3.0 to 3.1.
*/
rds_ib_send_init_ring(ic);
rds_ib_recv_init_ring(ic);
/* Post receive buffers - as a side effect, this will update
* the posted credit count. */
rds_ib_recv_refill(conn, 1, GFP_KERNEL);
/* Tune RNR behavior */
rds_ib_tune_rnr(ic, &qp_attr);
qp_attr.qp_state = IB_QPS_RTS;
err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
if (err)
printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err);
/* update ib_device with this local ipaddr */
err = rds_ib_update_ipaddr(ic->rds_ibdev, conn->c_laddr);
if (err)
printk(KERN_ERR "rds_ib_update_ipaddr failed (%d)\n",
err);
/* If the peer gave us the last packet it saw, process this as if
* we had received a regular ACK. */
if (dp) {
/* dp structure start is not guaranteed to be 8 bytes aligned.
* Since dp_ack_seq is 64-bit extended load operations can be
* used so go through get_unaligned to avoid unaligned errors.
*/
__be64 dp_ack_seq = get_unaligned(&dp->dp_ack_seq);
if (dp_ack_seq)
rds_send_drop_acked(conn, be64_to_cpu(dp_ack_seq),
NULL);
}
rds_connect_complete(conn);
}
/* the core send_sem serializes this with other xmit and shutdown */
static int rds_tcp_sendmsg(struct socket *sock, void *data, unsigned int len)
{
struct kvec vec = {
.iov_base = data,
.iov_len = len,
};
struct msghdr msg = {
.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL,
};
return kernel_sendmsg(sock, &msg, &vec, 1, vec.iov_len);
}
/* the core send_sem serializes this with other xmit and shutdown */
int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm,
unsigned int hdr_off, unsigned int sg, unsigned int off)
{
struct rds_tcp_connection *tc = conn->c_transport_data;
int done = 0;
int ret = 0;
if (hdr_off == 0) {
/*
* m_ack_seq is set to the sequence number of the last byte of
* header and data. see rds_tcp_is_acked().
*/
tc->t_last_sent_nxt = rds_tcp_snd_nxt(tc);
rm->m_ack_seq = tc->t_last_sent_nxt +
sizeof(struct rds_header) +
be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1;
smp_mb__before_clear_bit();
set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags);
tc->t_last_expected_una = rm->m_ack_seq + 1;
rdsdebug("rm %p tcp nxt %u ack_seq %llu\n",
rm, rds_tcp_snd_nxt(tc),
(unsigned long long)rm->m_ack_seq);
}
if (hdr_off < sizeof(struct rds_header)) {
/* see rds_tcp_write_space() */
set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags);
ret = rds_tcp_sendmsg(tc->t_sock,
(void *)&rm->m_inc.i_hdr + hdr_off,
sizeof(rm->m_inc.i_hdr) - hdr_off);
if (ret < 0)
goto out;
done += ret;
if (hdr_off + done != sizeof(struct rds_header))
goto out;
}
while (sg < rm->data.op_nents) {
ret = tc->t_sock->ops->sendpage(tc->t_sock,
sg_page(&rm->data.op_sg[sg]),
rm->data.op_sg[sg].offset + off,
rm->data.op_sg[sg].length - off,
MSG_DONTWAIT|MSG_NOSIGNAL);
rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->data.op_sg[sg]),
rm->data.op_sg[sg].offset + off, rm->data.op_sg[sg].length - off,
ret);
if (ret <= 0)
break;
off += ret;
done += ret;
if (off == rm->data.op_sg[sg].length) {
off = 0;
sg++;
}
}
out:
if (ret <= 0) {
/* write_space will hit after EAGAIN, all else fatal */
if (ret == -EAGAIN) {
rds_tcp_stats_inc(s_tcp_sndbuf_full);
ret = 0;
} else {
printk(KERN_WARNING "RDS/tcp: send to %pI4 "
"returned %d, disconnecting and reconnecting\n",
&conn->c_faddr, ret);
rds_conn_drop(conn);
}
}
if (done == 0)
done = ret;
return done;
}
/*
* rm->m_ack_seq is set to the tcp sequence number that corresponds to the
* last byte of the message, including the header. This means that the
* entire message has been received if rm->m_ack_seq is "before" the next
* unacked byte of the TCP sequence space. We have to do very careful
* wrapping 32bit comparisons here.
*/
static int rds_tcp_is_acked(struct rds_message *rm, uint64_t ack)
{
if (!test_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags))
return 0;
return (__s32)((u32)rm->m_ack_seq - (u32)ack) < 0;
}
void rds_tcp_write_space(struct sock *sk)
{
void (*write_space)(struct sock *sk);
struct rds_connection *conn;
struct rds_tcp_connection *tc;
read_lock_bh(&sk->sk_callback_lock);
conn = sk->sk_user_data;
if (!conn) {
write_space = sk->sk_write_space;
goto out;
}
tc = conn->c_transport_data;
rdsdebug("write_space for tc %p\n", tc);
write_space = tc->t_orig_write_space;
rds_tcp_stats_inc(s_tcp_write_space_calls);
rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc));
tc->t_last_seen_una = rds_tcp_snd_una(tc);
rds_send_drop_acked(conn, rds_tcp_snd_una(tc), rds_tcp_is_acked);
if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf)
queue_delayed_work(rds_wq, &conn->c_send_w, 0);
out:
read_unlock_bh(&sk->sk_callback_lock);
/*
* write_space is only called when data leaves tcp's send queue if
* SOCK_NOSPACE is set. We set SOCK_NOSPACE every time we put
* data in tcp's send queue because we use write_space to parse the
* sequence numbers and notice that rds messages have been fully
* received.
*
* tcp's write_space clears SOCK_NOSPACE if the send queue has more
* than a certain amount of space. So we need to set it again *after*
* we call tcp's write_space or else we might only get called on the
* first of a series of incoming tcp acks.
*/
write_space(sk);
if (sk->sk_socket)
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
}
static int rds_tcp_sendmsg(struct socket *sock, void *data, unsigned int len)
{
struct kvec vec = {
.iov_base = data,
.iov_len = len,
};
struct msghdr msg = {
.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL,
};
return kernel_sendmsg(sock, &msg, &vec, 1, vec.iov_len);
}
int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm,
unsigned int hdr_off, unsigned int sg, unsigned int off)
{
struct rds_tcp_connection *tc = conn->c_transport_data;
int done = 0;
int ret = 0;
if (hdr_off == 0) {
tc->t_last_sent_nxt = rds_tcp_snd_nxt(tc);
rm->m_ack_seq = tc->t_last_sent_nxt +
sizeof(struct rds_header) +
be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1;
smp_mb__before_clear_bit();
set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags);
tc->t_last_expected_una = rm->m_ack_seq + 1;
rdsdebug("rm %p tcp nxt %u ack_seq %llu\n",
rm, rds_tcp_snd_nxt(tc),
(unsigned long long)rm->m_ack_seq);
}
if (hdr_off < sizeof(struct rds_header)) {
set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags);
ret = rds_tcp_sendmsg(tc->t_sock,
(void *)&rm->m_inc.i_hdr + hdr_off,
sizeof(rm->m_inc.i_hdr) - hdr_off);
if (ret < 0)
goto out;
done += ret;
if (hdr_off + done != sizeof(struct rds_header))
goto out;
}
while (sg < rm->data.op_nents) {
ret = tc->t_sock->ops->sendpage(tc->t_sock,
sg_page(&rm->data.op_sg[sg]),
rm->data.op_sg[sg].offset + off,
rm->data.op_sg[sg].length - off,
MSG_DONTWAIT|MSG_NOSIGNAL);
rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->data.op_sg[sg]),
rm->data.op_sg[sg].offset + off, rm->data.op_sg[sg].length - off,
ret);
if (ret <= 0)
break;
off += ret;
done += ret;
if (off == rm->data.op_sg[sg].length) {
off = 0;
sg++;
}
}
out:
if (ret <= 0) {
if (ret == -EAGAIN) {
rds_tcp_stats_inc(s_tcp_sndbuf_full);
ret = 0;
} else {
printk(KERN_WARNING "RDS/tcp: send to %pI4 "
"returned %d, disconnecting and reconnecting\n",
&conn->c_faddr, ret);
rds_conn_drop(conn);
}
}
if (done == 0)
done = ret;
return done;
}
static int rds_tcp_is_acked(struct rds_message *rm, uint64_t ack)
{
if (!test_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags))
return 0;
return (__s32)((u32)rm->m_ack_seq - (u32)ack) < 0;
}
void rds_tcp_write_space(struct sock *sk)
{
void (*write_space)(struct sock *sk);
struct rds_connection *conn;
struct rds_tcp_connection *tc;
read_lock_bh(&sk->sk_callback_lock);
conn = sk->sk_user_data;
if (!conn) {
write_space = sk->sk_write_space;
goto out;
}
tc = conn->c_transport_data;
rdsdebug("write_space for tc %p\n", tc);
write_space = tc->t_orig_write_space;
rds_tcp_stats_inc(s_tcp_write_space_calls);
rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc));
tc->t_last_seen_una = rds_tcp_snd_una(tc);
rds_send_drop_acked(conn, rds_tcp_snd_una(tc), rds_tcp_is_acked);
if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf)
queue_delayed_work(rds_wq, &conn->c_send_w, 0);
out:
read_unlock_bh(&sk->sk_callback_lock);
write_space(sk);
if (sk->sk_socket)
set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
}
