static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
{
struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
long idle_time = 0;
if (xprt_connected(xprt))
idle_time = (long)(jiffies - xprt->last_used) / HZ;
seq_printf(seq,
"\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
"%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
0, /* need a local port? */
xprt->stat.bind_count,
xprt->stat.connect_count,
xprt->stat.connect_time,
idle_time,
xprt->stat.sends,
xprt->stat.recvs,
xprt->stat.bad_xids,
xprt->stat.req_u,
xprt->stat.bklog_u,
r_xprt->rx_stats.read_chunk_count,
r_xprt->rx_stats.write_chunk_count,
r_xprt->rx_stats.reply_chunk_count,
r_xprt->rx_stats.total_rdma_request,
r_xprt->rx_stats.total_rdma_reply,
r_xprt->rx_stats.pullup_copy_count,
r_xprt->rx_stats.fixup_copy_count,
r_xprt->rx_stats.hardway_register_count,
r_xprt->rx_stats.failed_marshal_count,
r_xprt->rx_stats.bad_reply_count);
}
/**
* xprt_rdma_send_request - marshal and send an RPC request
* @rqst: RPC message in rq_snd_buf
*
* Caller holds the transport's write lock.
*
* Returns:
* %0 if the RPC message has been sent
* %-ENOTCONN if the caller should reconnect and call again
* %-EAGAIN if the caller should call again
* %-ENOBUFS if the caller should call again after a delay
* %-EIO if a permanent error occurred and the request was not
* sent. Do not try to send this message again.
*/
static int
xprt_rdma_send_request(struct rpc_rqst *rqst)
{
struct rpc_xprt *xprt = rqst->rq_xprt;
struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
int rc = 0;
#if defined(CONFIG_SUNRPC_BACKCHANNEL)
if (unlikely(!rqst->rq_buffer))
return xprt_rdma_bc_send_reply(rqst);
#endif /* CONFIG_SUNRPC_BACKCHANNEL */
if (!xprt_connected(xprt))
goto drop_connection;
if (!xprt_request_get_cong(xprt, rqst))
return -EBADSLT;
rc = rpcrdma_marshal_req(r_xprt, rqst);
if (rc < 0)
goto failed_marshal;
/* Must suppress retransmit to maintain credits */
if (rqst->rq_connect_cookie == xprt->connect_cookie)
goto drop_connection;
rqst->rq_xtime = ktime_get();
__set_bit(RPCRDMA_REQ_F_PENDING, &req->rl_flags);
if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
goto drop_connection;
rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len;
rqst->rq_bytes_sent = 0;
/* An RPC with no reply will throw off credit accounting,
* so drop the connection to reset the credit grant.
*/
if (!rpc_reply_expected(rqst->rq_task))
goto drop_connection;
return 0;
failed_marshal:
if (rc != -ENOTCONN)
return rc;
drop_connection:
xprt_disconnect_done(xprt);
return -ENOTCONN; /* implies disconnect */
}
/**
* xs_tcp_print_stats - display TCP socket-specifc stats
* @xprt: rpc_xprt struct containing statistics
* @seq: output file
*
*/
static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
{
long idle_time = 0;
if (xprt_connected(xprt))
idle_time = (long)(jiffies - xprt->last_used) / HZ;
seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
xprt->port,
xprt->stat.bind_count,
xprt->stat.connect_count,
xprt->stat.connect_time,
idle_time,
xprt->stat.sends,
xprt->stat.recvs,
xprt->stat.bad_xids,
xprt->stat.req_u,
xprt->stat.bklog_u);
}
void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
{
struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
long idle_time = 0;
if (xprt_connected(xprt))
idle_time = (long)(jiffies - xprt->last_used) / HZ;
seq_puts(seq, "\txprt:\trdma ");
seq_printf(seq, "%u %lu %lu %lu %ld %lu %lu %lu %llu %llu ",
0, /* need a local port? */
xprt->stat.bind_count,
xprt->stat.connect_count,
xprt->stat.connect_time / HZ,
idle_time,
xprt->stat.sends,
xprt->stat.recvs,
xprt->stat.bad_xids,
xprt->stat.req_u,
xprt->stat.bklog_u);
seq_printf(seq, "%lu %lu %lu %llu %llu %llu %llu %lu %lu %lu %lu ",
r_xprt->rx_stats.read_chunk_count,
r_xprt->rx_stats.write_chunk_count,
r_xprt->rx_stats.reply_chunk_count,
r_xprt->rx_stats.total_rdma_request,
r_xprt->rx_stats.total_rdma_reply,
r_xprt->rx_stats.pullup_copy_count,
r_xprt->rx_stats.fixup_copy_count,
r_xprt->rx_stats.hardway_register_count,
r_xprt->rx_stats.failed_marshal_count,
r_xprt->rx_stats.bad_reply_count,
r_xprt->rx_stats.nomsg_call_count);
seq_printf(seq, "%lu %lu %lu %lu %lu %lu\n",
r_xprt->rx_stats.mrs_recycled,
r_xprt->rx_stats.mrs_orphaned,
r_xprt->rx_stats.mrs_allocated,
r_xprt->rx_stats.local_inv_needed,
r_xprt->rx_stats.empty_sendctx_q,
r_xprt->rx_stats.reply_waits_for_send);
}
static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
{
struct rpc_xprt *xprt = rd_desc->arg.data;
skb_reader_t desc = {
.skb = skb,
.offset = offset,
.count = len,
.csum = 0
};
dprintk("RPC: xs_tcp_data_recv started\n");
do {
/* Read in a new fragment marker if necessary */
/* Can we ever really expect to get completely empty fragments? */
if (xprt->tcp_flags & XPRT_COPY_RECM) {
xs_tcp_read_fraghdr(xprt, &desc);
continue;
}
/* Read in the xid if necessary */
if (xprt->tcp_flags & XPRT_COPY_XID) {
xs_tcp_read_xid(xprt, &desc);
continue;
}
/* Read in the request data */
if (xprt->tcp_flags & XPRT_COPY_DATA) {
xs_tcp_read_request(xprt, &desc);
continue;
}
/* Skip over any trailing bytes on short reads */
xs_tcp_read_discard(xprt, &desc);
} while (desc.count);
dprintk("RPC: xs_tcp_data_recv done\n");
return len - desc.count;
}
/**
* xs_tcp_data_ready - "data ready" callback for TCP sockets
* @sk: socket with data to read
* @bytes: how much data to read
*
*/
static void xs_tcp_data_ready(struct sock *sk, int bytes)
{
struct rpc_xprt *xprt;
read_descriptor_t rd_desc;
read_lock(&sk->sk_callback_lock);
dprintk("RPC: xs_tcp_data_ready...\n");
if (!(xprt = xprt_from_sock(sk)))
goto out;
if (xprt->shutdown)
goto out;
/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
rd_desc.arg.data = xprt;
rd_desc.count = 65536;
tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
out:
read_unlock(&sk->sk_callback_lock);
}
/**
* xs_tcp_state_change - callback to handle TCP socket state changes
* @sk: socket whose state has changed
*
*/
static void xs_tcp_state_change(struct sock *sk)
{
struct rpc_xprt *xprt;
read_lock(&sk->sk_callback_lock);
if (!(xprt = xprt_from_sock(sk)))
goto out;
dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
dprintk("RPC: state %x conn %d dead %d zapped %d\n",
sk->sk_state, xprt_connected(xprt),
sock_flag(sk, SOCK_DEAD),
sock_flag(sk, SOCK_ZAPPED));
switch (sk->sk_state) {
case TCP_ESTABLISHED:
spin_lock_bh(&xprt->transport_lock);
if (!xprt_test_and_set_connected(xprt)) {
/* Reset TCP record info */
xprt->tcp_offset = 0;
xprt->tcp_reclen = 0;
xprt->tcp_copied = 0;
if (xprt->tcp_flags & XPRT_SRCADDR_PRESENT)
xprt->tcp_flags = XPRT_SRCADDR_PRESENT |
XPRT_COPY_RECM |
XPRT_COPY_XID;
else
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);
break;
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)
queue_work(rpciod_workqueue, &xprt->task_cleanup);
default:
xprt_disconnect(xprt);
}
out:
read_unlock(&sk->sk_callback_lock);
}
/**
* xs_udp_write_space - callback invoked when socket buffer space
* becomes available
* @sk: socket whose state has changed
*
* Called when more output buffer space is available for this socket.
* We try not to wake our writers until they can make "significant"
* progress, otherwise we'll waste resources thrashing kernel_sendmsg
* with a bunch of small requests.
*/
static void xs_udp_write_space(struct sock *sk)
{
read_lock(&sk->sk_callback_lock);
/* from net/core/sock.c:sock_def_write_space */
if (sock_writeable(sk)) {
struct socket *sock;
struct rpc_xprt *xprt;
if (unlikely(!(sock = sk->sk_socket)))
goto out;
if (unlikely(!(xprt = xprt_from_sock(sk))))
goto out;
if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
goto out;
xprt_write_space(xprt);
}
out:
read_unlock(&sk->sk_callback_lock);
}
static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
{
struct kvec iov = {
.iov_base = xdr->head[0].iov_base + base,
.iov_len = len - base,
};
struct msghdr msg = {
.msg_name = addr,
.msg_namelen = addrlen,
.msg_flags = XS_SENDMSG_FLAGS,
};
if (xdr->len > len)
msg.msg_flags |= MSG_MORE;
if (likely(iov.iov_len))
return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
return kernel_sendmsg(sock, &msg, NULL, 0, 0);
}
static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
{
struct kvec iov = {
.iov_base = xdr->tail[0].iov_base + base,
.iov_len = len - base,
};
struct msghdr msg = {
.msg_flags = XS_SENDMSG_FLAGS,
};
return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
}
/**
* xs_sendpages - write pages directly to a socket
* @sock: socket to send on
* @addr: UDP only -- address of destination
* @addrlen: UDP only -- length of destination address
* @xdr: buffer containing this request
* @base: starting position in the buffer
*
*/
static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
{
struct page **ppage = xdr->pages;
unsigned int len, pglen = xdr->page_len;
int err, ret = 0;
ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
if (unlikely(!sock))
return -ENOTCONN;
clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
len = xdr->head[0].iov_len;
if (base < len || (addr != NULL && base == 0)) {
err = xs_send_head(sock, addr, addrlen, xdr, base, len);
if (ret == 0)
ret = err;
else if (err > 0)
ret += err;
if (err != (len - base))
goto out;
base = 0;
} else
base -= len;
if (unlikely(pglen == 0))
goto copy_tail;
if (unlikely(base >= pglen)) {
base -= pglen;
goto copy_tail;
}
if (base || xdr->page_base) {
pglen -= base;
base += xdr->page_base;
ppage += base >> PAGE_CACHE_SHIFT;
base &= ~PAGE_CACHE_MASK;
}
sendpage = kernel_sendpage;
do {
int flags = XS_SENDMSG_FLAGS;
len = PAGE_CACHE_SIZE;
if (base)
len -= base;
if (pglen < len)
len = pglen;
if (pglen != len || xdr->tail[0].iov_len != 0)
flags |= MSG_MORE;
/* Hmm... We might be dealing with highmem pages */
if (PageHighMem(*ppage))
sendpage = sock_no_sendpage;
err = sendpage(sock, *ppage, base, len, flags);
if (ret == 0)
ret = err;
else if (err > 0)
ret += err;
if (err != len)
goto out;
base = 0;
ppage++;
} while ((pglen -= len) != 0);
copy_tail:
len = xdr->tail[0].iov_len;
if (base < len) {
err = xs_send_tail(sock, xdr, base, len);
if (ret == 0)
ret = err;
else if (err > 0)
ret += err;
}
out:
return ret;
}
/**
* xs_nospace - place task on wait queue if transmit was incomplete
* @task: task to put to sleep
*
*/
static void xs_nospace(struct rpc_task *task)
{
struct rpc_rqst *req = task->tk_rqstp;
struct rpc_xprt *xprt = req->rq_xprt;
dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
task->tk_pid, req->rq_slen - req->rq_bytes_sent,
req->rq_slen);
if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
/* Protect against races with write_space */
spin_lock_bh(&xprt->transport_lock);
/* Don't race with disconnect */
if (!xprt_connected(xprt))
task->tk_status = -ENOTCONN;
else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
xprt_wait_for_buffer_space(task);
spin_unlock_bh(&xprt->transport_lock);
} else
/* Keep holding the socket if it is blocked */
rpc_delay(task, HZ>>4);
}
/**
* xs_udp_send_request - write an RPC request to a UDP socket
* @task: address of RPC task that manages the state of an RPC request
*
* Return values:
* 0: The request has been sent
* EAGAIN: The socket was blocked, please call again later to
* complete the request
* ENOTCONN: Caller needs to invoke connect logic then call again
* other: Some other error occured, the request was not sent
*/
static int xs_udp_send_request(struct rpc_task *task)
{
struct rpc_rqst *req = task->tk_rqstp;
struct rpc_xprt *xprt = req->rq_xprt;
struct xdr_buf *xdr = &req->rq_snd_buf;
int status;
xs_pktdump("packet data:",
req->rq_svec->iov_base,
req->rq_svec->iov_len);
req->rq_xtime = jiffies;
status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
sizeof(xprt->addr), xdr, req->rq_bytes_sent);
dprintk("RPC: xs_udp_send_request(%u) = %d\n",
xdr->len - req->rq_bytes_sent, status);
if (likely(status >= (int) req->rq_slen))
return 0;
/* Still some bytes left; set up for a retry later. */
if (status > 0)
status = -EAGAIN;
switch (status) {
case -ENETUNREACH:
case -EPIPE:
case -ECONNREFUSED:
/* When the server has died, an ICMP port unreachable message
* prompts ECONNREFUSED. */
break;
case -EAGAIN:
xs_nospace(task);
break;
default:
dprintk("RPC: sendmsg returned unrecognized error %d\n",
-status);
break;
}
return status;
}
static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
{
u32 reclen = buf->len - sizeof(rpc_fraghdr);
rpc_fraghdr *base = buf->head[0].iov_base;
*base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
}
/**
* xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
* @args: RPC transport to connect
*
* Invoked by a work queue tasklet.
*/
static void xs_tcp_connect_worker(void *args)
{
struct rpc_xprt *xprt = (struct rpc_xprt *)args;
struct socket *sock = xprt->sock;
int err, status = -EIO;
if (xprt->shutdown || xprt->addr.sin_port == 0)
goto out;
dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt);
if (!xprt->sock) {
/* start from scratch */
if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
goto out;
}
xs_reclassify_socket(sock);
if (xs_bind(xprt, sock)) {
sock_release(sock);
goto out;
}
} else
/* "close" the socket, preserving the local port */
xs_tcp_reuse_connection(xprt);
if (!xprt->inet) {
struct sock *sk = sock->sk;
write_lock_bh(&sk->sk_callback_lock);
sk->sk_user_data = xprt;
xprt->old_data_ready = sk->sk_data_ready;
xprt->old_state_change = sk->sk_state_change;
xprt->old_write_space = sk->sk_write_space;
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;
sock_reset_flag(sk, SOCK_LINGER);
tcp_sk(sk)->linger2 = 0;
tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
xprt_clear_connected(xprt);
/* Reset to new socket */
xprt->sock = sock;
xprt->inet = sk;
write_unlock_bh(&sk->sk_callback_lock);
}
/* Tell the socket layer to start connecting... */
xprt->stat.connect_count++;
xprt->stat.connect_start = jiffies;
status = kernel_connect(sock, (struct sockaddr *) &xprt->addr,
sizeof(xprt->addr), O_NONBLOCK);
dprintk("RPC: %p connect status %d connected %d sock state %d\n",
xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
if (status < 0) {
switch (status) {
case -EINPROGRESS:
case -EALREADY:
goto out_clear;
case -ECONNREFUSED:
case -ECONNRESET:
/* retry with existing socket, after a delay */
break;
case -ENETUNREACH:
status = -ENOTCONN;
break;
default:
/* get rid of existing socket, and retry */
xs_close(xprt);
break;
}
}
out:
xprt_wake_pending_tasks(xprt, status);
out_clear:
xprt_clear_connecting(xprt);
}
