#include <linux/udp.h>
#include <linux/tcp.h>
#include <linux/sunrpc/clnt.h>
+#include <linux/sunrpc/sched.h>
#include <linux/file.h>
#include <net/sock.h>
#include <net/udp.h>
#include <net/tcp.h>
+/*
+ * xprtsock tunables
+ */
+unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
+unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
+
+unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
+unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
+
/*
* How many times to try sending a request on a socket before waiting
* for the socket buffer to clear.
*/
#define XS_SENDMSG_RETRY (10U)
+/*
+ * Time out for an RPC UDP socket connect. UDP socket connects are
+ * synchronous, but we set a timeout anyway in case of resource
+ * exhaustion on the local host.
+ */
+#define XS_UDP_CONN_TO (5U * HZ)
+
+/*
+ * Wait duration for an RPC TCP connection to be established. Solaris
+ * NFS over TCP uses 60 seconds, for example, which is in line with how
+ * long a server takes to reboot.
+ */
+#define XS_TCP_CONN_TO (60U * HZ)
+
+/*
+ * Wait duration for a reply from the RPC portmapper.
+ */
+#define XS_BIND_TO (60U * HZ)
+
+/*
+ * Delay if a UDP socket connect error occurs. This is most likely some
+ * kind of resource problem on the local host.
+ */
+#define XS_UDP_REEST_TO (2U * HZ)
+
+/*
+ * The reestablish timeout allows clients to delay for a bit before attempting
+ * to reconnect to a server that just dropped our connection.
+ *
+ * We implement an exponential backoff when trying to reestablish a TCP
+ * transport connection with the server. Some servers like to drop a TCP
+ * connection when they are overworked, so we start with a short timeout and
+ * increase over time if the server is down or not responding.
+ */
+#define XS_TCP_INIT_REEST_TO (3U * HZ)
+#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
+
+/*
+ * TCP idle timeout; client drops the transport socket if it is idle
+ * for this long. Note that we also timeout UDP sockets to prevent
+ * holding port numbers when there is no RPC traffic.
+ */
+#define XS_IDLE_DISC_TO (5U * 60 * HZ)
+
#ifdef RPC_DEBUG
# undef RPC_DEBUG_DATA
# define RPCDBG_FACILITY RPCDBG_TRANS
struct sock *sk = xprt->inet;
if (!sk)
- return;
+ goto clear_close_wait;
dprintk("RPC: xs_close xprt %p\n", xprt);
sk->sk_no_check = 0;
sock_release(sock);
+clear_close_wait:
+ smp_mb__before_clear_bit();
+ clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
+ smp_mb__after_clear_bit();
}
/**
xprt->tcp_reclen = 0;
xprt->tcp_copied = 0;
xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
+ xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
xprt_wake_pending_tasks(xprt, 0);
}
spin_unlock_bh(&xprt->transport_lock);
case TCP_SYN_SENT:
case TCP_SYN_RECV:
break;
+ case TCP_CLOSE_WAIT:
+ /* Try to schedule an autoclose RPC calls */
+ set_bit(XPRT_CLOSE_WAIT, &xprt->state);
+ if (test_and_set_bit(XPRT_LOCKED, &xprt->state) == 0)
+ schedule_work(&xprt->task_cleanup);
default:
xprt_disconnect(xprt);
- break;
}
out:
read_unlock(&sk->sk_callback_lock);
read_unlock(&sk->sk_callback_lock);
}
-/**
- * xs_udp_set_buffer_size - set send and receive limits
- * @xprt: generic transport
- *
- * Set socket send and receive limits based on the
- * sndsize and rcvsize fields in the generic transport
- * structure.
- */
-static void xs_udp_set_buffer_size(struct rpc_xprt *xprt)
+static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
{
struct sock *sk = xprt->inet;
}
/**
- * xs_tcp_set_buffer_size - set send and receive limits
+ * xs_udp_set_buffer_size - set send and receive limits
* @xprt: generic transport
+ * @sndsize: requested size of send buffer, in bytes
+ * @rcvsize: requested size of receive buffer, in bytes
*
- * Nothing to do for TCP.
+ * Set socket send and receive buffer size limits.
*/
-static void xs_tcp_set_buffer_size(struct rpc_xprt *xprt)
+static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
{
- return;
+ xprt->sndsize = 0;
+ if (sndsize)
+ xprt->sndsize = sndsize + 1024;
+ xprt->rcvsize = 0;
+ if (rcvsize)
+ xprt->rcvsize = rcvsize + 1024;
+
+ xs_udp_do_set_buffer_size(xprt);
}
/**
xprt_adjust_cwnd(task, -ETIMEDOUT);
}
+/**
+ * xs_set_port - reset the port number in the remote endpoint address
+ * @xprt: generic transport
+ * @port: new port number
+ *
+ */
+static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
+{
+ dprintk("RPC: setting port for xprt %p to %u\n", xprt, port);
+ xprt->addr.sin_port = htons(port);
+}
+
static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
{
struct sockaddr_in myaddr = {
sk->sk_data_ready = xs_udp_data_ready;
sk->sk_write_space = xs_udp_write_space;
sk->sk_no_check = UDP_CSUM_NORCV;
+ sk->sk_allocation = GFP_ATOMIC;
xprt_set_connected(xprt);
write_unlock_bh(&sk->sk_callback_lock);
}
- xs_udp_set_buffer_size(xprt);
+ xs_udp_do_set_buffer_size(xprt);
status = 0;
out:
xprt_wake_pending_tasks(xprt, status);
sk->sk_data_ready = xs_tcp_data_ready;
sk->sk_state_change = xs_tcp_state_change;
sk->sk_write_space = xs_tcp_write_space;
+ sk->sk_allocation = GFP_ATOMIC;
/* socket options */
sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
* @task: address of RPC task that manages state of connect request
*
* TCP: If the remote end dropped the connection, delay reconnecting.
+ *
+ * UDP socket connects are synchronous, but we use a work queue anyway
+ * to guarantee that even unprivileged user processes can set up a
+ * socket on a privileged port.
+ *
+ * If a UDP socket connect fails, the delay behavior here prevents
+ * retry floods (hard mounts).
*/
static void xs_connect(struct rpc_task *task)
{
return;
if (xprt->sock != NULL) {
- dprintk("RPC: xs_connect delayed xprt %p\n", xprt);
+ dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n",
+ xprt, xprt->reestablish_timeout / HZ);
schedule_delayed_work(&xprt->connect_worker,
- RPC_REESTABLISH_TIMEOUT);
+ xprt->reestablish_timeout);
+ xprt->reestablish_timeout <<= 1;
+ if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
+ xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
} else {
dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
schedule_work(&xprt->connect_worker);
.set_buffer_size = xs_udp_set_buffer_size,
.reserve_xprt = xprt_reserve_xprt_cong,
.release_xprt = xprt_release_xprt_cong,
+ .set_port = xs_set_port,
.connect = xs_connect,
+ .buf_alloc = rpc_malloc,
+ .buf_free = rpc_free,
.send_request = xs_udp_send_request,
.set_retrans_timeout = xprt_set_retrans_timeout_rtt,
.timer = xs_udp_timer,
};
static struct rpc_xprt_ops xs_tcp_ops = {
- .set_buffer_size = xs_tcp_set_buffer_size,
.reserve_xprt = xprt_reserve_xprt,
.release_xprt = xprt_release_xprt,
+ .set_port = xs_set_port,
.connect = xs_connect,
+ .buf_alloc = rpc_malloc,
+ .buf_free = rpc_free,
.send_request = xs_tcp_send_request,
.set_retrans_timeout = xprt_set_retrans_timeout_def,
.close = xs_close,
xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
+ xprt->bind_timeout = XS_BIND_TO;
+ xprt->connect_timeout = XS_UDP_CONN_TO;
+ xprt->reestablish_timeout = XS_UDP_REEST_TO;
+ xprt->idle_timeout = XS_IDLE_DISC_TO;
xprt->ops = &xs_udp_ops;
xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
+ xprt->bind_timeout = XS_BIND_TO;
+ xprt->connect_timeout = XS_TCP_CONN_TO;
+ xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
+ xprt->idle_timeout = XS_IDLE_DISC_TO;
xprt->ops = &xs_tcp_ops;