2 * linux/net/sunrpc/xprtsock.c
4 * Client-side transport implementation for sockets.
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
18 #include <linux/types.h>
19 #include <linux/slab.h>
20 #include <linux/capability.h>
21 #include <linux/sched.h>
22 #include <linux/pagemap.h>
23 #include <linux/errno.h>
24 #include <linux/socket.h>
26 #include <linux/net.h>
28 #include <linux/udp.h>
29 #include <linux/tcp.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/file.h>
34 #include <net/checksum.h>
39 * How many times to try sending a request on a socket before waiting
40 * for the socket buffer to clear.
42 #define XS_SENDMSG_RETRY (10U)
45 # undef RPC_DEBUG_DATA
46 # define RPCDBG_FACILITY RPCDBG_TRANS
50 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
52 u8 *buf = (u8 *) packet;
55 dprintk("RPC: %s\n", msg);
56 for (j = 0; j < count && j < 128; j += 4) {
60 dprintk("0x%04x ", j);
62 dprintk("%02x%02x%02x%02x ",
63 buf[j], buf[j+1], buf[j+2], buf[j+3]);
68 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
74 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
76 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
79 .iov_base = xdr->head[0].iov_base + base,
80 .iov_len = len - base,
84 .msg_namelen = addrlen,
85 .msg_flags = XS_SENDMSG_FLAGS,
89 msg.msg_flags |= MSG_MORE;
91 if (likely(iov.iov_len))
92 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
93 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
96 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
99 .iov_base = xdr->tail[0].iov_base + base,
100 .iov_len = len - base,
102 struct msghdr msg = {
103 .msg_flags = XS_SENDMSG_FLAGS,
106 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
110 * xs_sendpages - write pages directly to a socket
111 * @sock: socket to send on
112 * @addr: UDP only -- address of destination
113 * @addrlen: UDP only -- length of destination address
114 * @xdr: buffer containing this request
115 * @base: starting position in the buffer
118 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
120 struct page **ppage = xdr->pages;
121 unsigned int len, pglen = xdr->page_len;
123 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
128 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
130 len = xdr->head[0].iov_len;
131 if (base < len || (addr != NULL && base == 0)) {
132 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
137 if (err != (len - base))
143 if (unlikely(pglen == 0))
145 if (unlikely(base >= pglen)) {
149 if (base || xdr->page_base) {
151 base += xdr->page_base;
152 ppage += base >> PAGE_CACHE_SHIFT;
153 base &= ~PAGE_CACHE_MASK;
156 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
158 int flags = XS_SENDMSG_FLAGS;
160 len = PAGE_CACHE_SIZE;
166 if (pglen != len || xdr->tail[0].iov_len != 0)
169 /* Hmm... We might be dealing with highmem pages */
170 if (PageHighMem(*ppage))
171 sendpage = sock_no_sendpage;
172 err = sendpage(sock, *ppage, base, len, flags);
181 } while ((pglen -= len) != 0);
183 len = xdr->tail[0].iov_len;
185 err = xs_send_tail(sock, xdr, base, len);
196 * xs_nospace - place task on wait queue if transmit was incomplete
197 * @task: task to put to sleep
200 static void xs_nospace(struct rpc_task *task)
202 struct rpc_rqst *req = task->tk_rqstp;
203 struct rpc_xprt *xprt = req->rq_xprt;
205 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
206 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
209 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
210 /* Protect against races with write_space */
211 spin_lock_bh(&xprt->transport_lock);
213 /* Don't race with disconnect */
214 if (!xprt_connected(xprt))
215 task->tk_status = -ENOTCONN;
216 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
217 xprt_wait_for_buffer_space(task);
219 spin_unlock_bh(&xprt->transport_lock);
221 /* Keep holding the socket if it is blocked */
222 rpc_delay(task, HZ>>4);
226 * xs_udp_send_request - write an RPC request to a UDP socket
227 * @task: address of RPC task that manages the state of an RPC request
230 * 0: The request has been sent
231 * EAGAIN: The socket was blocked, please call again later to
232 * complete the request
233 * ENOTCONN: Caller needs to invoke connect logic then call again
234 * other: Some other error occured, the request was not sent
236 static int xs_udp_send_request(struct rpc_task *task)
238 struct rpc_rqst *req = task->tk_rqstp;
239 struct rpc_xprt *xprt = req->rq_xprt;
240 struct xdr_buf *xdr = &req->rq_snd_buf;
243 xs_pktdump("packet data:",
244 req->rq_svec->iov_base,
245 req->rq_svec->iov_len);
247 req->rq_xtime = jiffies;
248 status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
249 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
251 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
252 xdr->len - req->rq_bytes_sent, status);
254 if (likely(status >= (int) req->rq_slen))
257 /* Still some bytes left; set up for a retry later. */
265 /* When the server has died, an ICMP port unreachable message
266 * prompts ECONNREFUSED. */
272 dprintk("RPC: sendmsg returned unrecognized error %d\n",
280 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
282 u32 reclen = buf->len - sizeof(rpc_fraghdr);
283 rpc_fraghdr *base = buf->head[0].iov_base;
284 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
288 * xs_tcp_send_request - write an RPC request to a TCP socket
289 * @task: address of RPC task that manages the state of an RPC request
292 * 0: The request has been sent
293 * EAGAIN: The socket was blocked, please call again later to
294 * complete the request
295 * ENOTCONN: Caller needs to invoke connect logic then call again
296 * other: Some other error occured, the request was not sent
298 * XXX: In the case of soft timeouts, should we eventually give up
299 * if sendmsg is not able to make progress?
301 static int xs_tcp_send_request(struct rpc_task *task)
303 struct rpc_rqst *req = task->tk_rqstp;
304 struct rpc_xprt *xprt = req->rq_xprt;
305 struct xdr_buf *xdr = &req->rq_snd_buf;
306 int status, retry = 0;
308 xs_encode_tcp_record_marker(&req->rq_snd_buf);
310 xs_pktdump("packet data:",
311 req->rq_svec->iov_base,
312 req->rq_svec->iov_len);
314 /* Continue transmitting the packet/record. We must be careful
315 * to cope with writespace callbacks arriving _after_ we have
316 * called sendmsg(). */
318 req->rq_xtime = jiffies;
319 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
322 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
323 xdr->len - req->rq_bytes_sent, status);
325 if (unlikely(status < 0))
328 /* If we've sent the entire packet, immediately
329 * reset the count of bytes sent. */
330 req->rq_bytes_sent += status;
331 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
332 req->rq_bytes_sent = 0;
337 if (retry++ > XS_SENDMSG_RETRY)
352 dprintk("RPC: sendmsg returned unrecognized error %d\n",
354 xprt_disconnect(xprt);
362 * xs_close - close a socket
366 static void xs_close(struct rpc_xprt *xprt)
368 struct socket *sock = xprt->sock;
369 struct sock *sk = xprt->inet;
374 dprintk("RPC: xs_close xprt %p\n", xprt);
376 write_lock_bh(&sk->sk_callback_lock);
380 sk->sk_user_data = NULL;
381 sk->sk_data_ready = xprt->old_data_ready;
382 sk->sk_state_change = xprt->old_state_change;
383 sk->sk_write_space = xprt->old_write_space;
384 write_unlock_bh(&sk->sk_callback_lock);
392 * xs_destroy - prepare to shutdown a transport
393 * @xprt: doomed transport
396 static void xs_destroy(struct rpc_xprt *xprt)
398 dprintk("RPC: xs_destroy xprt %p\n", xprt);
400 cancel_delayed_work(&xprt->connect_worker);
401 flush_scheduled_work();
403 xprt_disconnect(xprt);
408 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
410 return (struct rpc_xprt *) sk->sk_user_data;
414 * xs_udp_data_ready - "data ready" callback for UDP sockets
415 * @sk: socket with data to read
416 * @len: how much data to read
419 static void xs_udp_data_ready(struct sock *sk, int len)
421 struct rpc_task *task;
422 struct rpc_xprt *xprt;
423 struct rpc_rqst *rovr;
425 int err, repsize, copied;
428 read_lock(&sk->sk_callback_lock);
429 dprintk("RPC: xs_udp_data_ready...\n");
430 if (!(xprt = xprt_from_sock(sk)))
433 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
439 repsize = skb->len - sizeof(struct udphdr);
441 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
445 /* Copy the XID from the skb... */
446 xp = skb_header_pointer(skb, sizeof(struct udphdr),
447 sizeof(_xid), &_xid);
451 /* Look up and lock the request corresponding to the given XID */
452 spin_lock(&xprt->transport_lock);
453 rovr = xprt_lookup_rqst(xprt, *xp);
456 task = rovr->rq_task;
458 if ((copied = rovr->rq_private_buf.buflen) > repsize)
461 /* Suck it into the iovec, verify checksum if not done by hw. */
462 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
465 /* Something worked... */
466 dst_confirm(skb->dst);
468 xprt_adjust_cwnd(task, copied);
469 xprt_update_rtt(task);
470 xprt_complete_rqst(task, copied);
473 spin_unlock(&xprt->transport_lock);
475 skb_free_datagram(sk, skb);
477 read_unlock(&sk->sk_callback_lock);
480 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
482 if (len > desc->count)
484 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
485 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
491 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
496 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
501 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
502 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
503 used = xs_tcp_copy_data(desc, p, len);
504 xprt->tcp_offset += used;
508 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
509 if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
510 xprt->tcp_flags |= XPRT_LAST_FRAG;
512 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
513 xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
515 xprt->tcp_flags &= ~XPRT_COPY_RECM;
516 xprt->tcp_offset = 0;
518 /* Sanity check of the record length */
519 if (unlikely(xprt->tcp_reclen < 4)) {
520 dprintk("RPC: invalid TCP record fragment length\n");
521 xprt_disconnect(xprt);
524 dprintk("RPC: reading TCP record fragment of length %d\n",
528 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
530 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
531 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
532 if (xprt->tcp_offset == xprt->tcp_reclen) {
533 xprt->tcp_flags |= XPRT_COPY_RECM;
534 xprt->tcp_offset = 0;
535 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
536 xprt->tcp_flags &= ~XPRT_COPY_DATA;
537 xprt->tcp_flags |= XPRT_COPY_XID;
538 xprt->tcp_copied = 0;
543 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
548 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
549 dprintk("RPC: reading XID (%Zu bytes)\n", len);
550 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
551 used = xs_tcp_copy_data(desc, p, len);
552 xprt->tcp_offset += used;
555 xprt->tcp_flags &= ~XPRT_COPY_XID;
556 xprt->tcp_flags |= XPRT_COPY_DATA;
557 xprt->tcp_copied = 4;
558 dprintk("RPC: reading reply for XID %08x\n",
559 ntohl(xprt->tcp_xid));
560 xs_tcp_check_recm(xprt);
563 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
565 struct rpc_rqst *req;
566 struct xdr_buf *rcvbuf;
570 /* Find and lock the request corresponding to this xid */
571 spin_lock(&xprt->transport_lock);
572 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
574 xprt->tcp_flags &= ~XPRT_COPY_DATA;
575 dprintk("RPC: XID %08x request not found!\n",
576 ntohl(xprt->tcp_xid));
577 spin_unlock(&xprt->transport_lock);
581 rcvbuf = &req->rq_private_buf;
583 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
584 skb_reader_t my_desc;
586 len = xprt->tcp_reclen - xprt->tcp_offset;
587 memcpy(&my_desc, desc, sizeof(my_desc));
589 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
590 &my_desc, xs_tcp_copy_data);
594 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
595 desc, xs_tcp_copy_data);
598 xprt->tcp_copied += r;
599 xprt->tcp_offset += r;
602 /* Error when copying to the receive buffer,
603 * usually because we weren't able to allocate
604 * additional buffer pages. All we can do now
605 * is turn off XPRT_COPY_DATA, so the request
606 * will not receive any additional updates,
608 * Any remaining data from this record will
611 xprt->tcp_flags &= ~XPRT_COPY_DATA;
612 dprintk("RPC: XID %08x truncated request\n",
613 ntohl(xprt->tcp_xid));
614 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
615 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
619 dprintk("RPC: XID %08x read %Zd bytes\n",
620 ntohl(xprt->tcp_xid), r);
621 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
622 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
624 if (xprt->tcp_copied == req->rq_private_buf.buflen)
625 xprt->tcp_flags &= ~XPRT_COPY_DATA;
626 else if (xprt->tcp_offset == xprt->tcp_reclen) {
627 if (xprt->tcp_flags & XPRT_LAST_FRAG)
628 xprt->tcp_flags &= ~XPRT_COPY_DATA;
632 if (!(xprt->tcp_flags & XPRT_COPY_DATA))
633 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
634 spin_unlock(&xprt->transport_lock);
635 xs_tcp_check_recm(xprt);
638 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
642 len = xprt->tcp_reclen - xprt->tcp_offset;
643 if (len > desc->count)
647 xprt->tcp_offset += len;
648 dprintk("RPC: discarded %Zu bytes\n", len);
649 xs_tcp_check_recm(xprt);
652 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
654 struct rpc_xprt *xprt = rd_desc->arg.data;
655 skb_reader_t desc = {
662 dprintk("RPC: xs_tcp_data_recv started\n");
664 /* Read in a new fragment marker if necessary */
665 /* Can we ever really expect to get completely empty fragments? */
666 if (xprt->tcp_flags & XPRT_COPY_RECM) {
667 xs_tcp_read_fraghdr(xprt, &desc);
670 /* Read in the xid if necessary */
671 if (xprt->tcp_flags & XPRT_COPY_XID) {
672 xs_tcp_read_xid(xprt, &desc);
675 /* Read in the request data */
676 if (xprt->tcp_flags & XPRT_COPY_DATA) {
677 xs_tcp_read_request(xprt, &desc);
680 /* Skip over any trailing bytes on short reads */
681 xs_tcp_read_discard(xprt, &desc);
682 } while (desc.count);
683 dprintk("RPC: xs_tcp_data_recv done\n");
684 return len - desc.count;
688 * xs_tcp_data_ready - "data ready" callback for TCP sockets
689 * @sk: socket with data to read
690 * @bytes: how much data to read
693 static void xs_tcp_data_ready(struct sock *sk, int bytes)
695 struct rpc_xprt *xprt;
696 read_descriptor_t rd_desc;
698 read_lock(&sk->sk_callback_lock);
699 dprintk("RPC: xs_tcp_data_ready...\n");
700 if (!(xprt = xprt_from_sock(sk)))
705 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
706 rd_desc.arg.data = xprt;
707 rd_desc.count = 65536;
708 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
710 read_unlock(&sk->sk_callback_lock);
714 * xs_tcp_state_change - callback to handle TCP socket state changes
715 * @sk: socket whose state has changed
718 static void xs_tcp_state_change(struct sock *sk)
720 struct rpc_xprt *xprt;
722 read_lock(&sk->sk_callback_lock);
723 if (!(xprt = xprt_from_sock(sk)))
725 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
726 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
727 sk->sk_state, xprt_connected(xprt),
728 sock_flag(sk, SOCK_DEAD),
729 sock_flag(sk, SOCK_ZAPPED));
731 switch (sk->sk_state) {
732 case TCP_ESTABLISHED:
733 spin_lock_bh(&xprt->transport_lock);
734 if (!xprt_test_and_set_connected(xprt)) {
735 /* Reset TCP record info */
736 xprt->tcp_offset = 0;
737 xprt->tcp_reclen = 0;
738 xprt->tcp_copied = 0;
739 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
740 xprt_wake_pending_tasks(xprt, 0);
742 spin_unlock_bh(&xprt->transport_lock);
748 xprt_disconnect(xprt);
752 read_unlock(&sk->sk_callback_lock);
756 * xs_udp_write_space - callback invoked when socket buffer space
758 * @sk: socket whose state has changed
760 * Called when more output buffer space is available for this socket.
761 * We try not to wake our writers until they can make "significant"
762 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
763 * with a bunch of small requests.
765 static void xs_udp_write_space(struct sock *sk)
767 read_lock(&sk->sk_callback_lock);
769 /* from net/core/sock.c:sock_def_write_space */
770 if (sock_writeable(sk)) {
772 struct rpc_xprt *xprt;
774 if (unlikely(!(sock = sk->sk_socket)))
776 if (unlikely(!(xprt = xprt_from_sock(sk))))
778 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
781 xprt_write_space(xprt);
785 read_unlock(&sk->sk_callback_lock);
789 * xs_tcp_write_space - callback invoked when socket buffer space
791 * @sk: socket whose state has changed
793 * Called when more output buffer space is available for this socket.
794 * We try not to wake our writers until they can make "significant"
795 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
796 * with a bunch of small requests.
798 static void xs_tcp_write_space(struct sock *sk)
800 read_lock(&sk->sk_callback_lock);
802 /* from net/core/stream.c:sk_stream_write_space */
803 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
805 struct rpc_xprt *xprt;
807 if (unlikely(!(sock = sk->sk_socket)))
809 if (unlikely(!(xprt = xprt_from_sock(sk))))
811 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
814 xprt_write_space(xprt);
818 read_unlock(&sk->sk_callback_lock);
822 * xs_udp_set_buffer_size - set send and receive limits
823 * @xprt: generic transport
825 * Set socket send and receive limits based on the
826 * sndsize and rcvsize fields in the generic transport
829 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt)
831 struct sock *sk = xprt->inet;
834 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
835 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
838 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
839 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
840 sk->sk_write_space(sk);
845 * xs_tcp_set_buffer_size - set send and receive limits
846 * @xprt: generic transport
848 * Nothing to do for TCP.
850 static void xs_tcp_set_buffer_size(struct rpc_xprt *xprt)
856 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
857 * @task: task that timed out
859 * Adjust the congestion window after a retransmit timeout has occurred.
861 static void xs_udp_timer(struct rpc_task *task)
863 xprt_adjust_cwnd(task, -ETIMEDOUT);
866 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
868 struct sockaddr_in myaddr = {
869 .sin_family = AF_INET,
872 unsigned short port = xprt->port;
875 myaddr.sin_port = htons(port);
876 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
880 dprintk("RPC: xs_bindresvport bound to port %u\n",
884 if (port <= xprt_min_resvport)
885 port = xprt_max_resvport;
888 } while (err == -EADDRINUSE && port != xprt->port);
890 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
895 * xs_udp_connect_worker - set up a UDP socket
896 * @args: RPC transport to connect
898 * Invoked by a work queue tasklet.
900 static void xs_udp_connect_worker(void *args)
902 struct rpc_xprt *xprt = (struct rpc_xprt *) args;
903 struct socket *sock = xprt->sock;
904 int err, status = -EIO;
906 if (xprt->shutdown || xprt->addr.sin_port == 0)
909 dprintk("RPC: xs_udp_connect_worker for xprt %p\n", xprt);
911 /* Start by resetting any existing state */
914 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
915 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
919 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
925 struct sock *sk = sock->sk;
927 write_lock_bh(&sk->sk_callback_lock);
929 sk->sk_user_data = xprt;
930 xprt->old_data_ready = sk->sk_data_ready;
931 xprt->old_state_change = sk->sk_state_change;
932 xprt->old_write_space = sk->sk_write_space;
933 sk->sk_data_ready = xs_udp_data_ready;
934 sk->sk_write_space = xs_udp_write_space;
935 sk->sk_no_check = UDP_CSUM_NORCV;
937 xprt_set_connected(xprt);
939 /* Reset to new socket */
943 write_unlock_bh(&sk->sk_callback_lock);
945 xs_udp_set_buffer_size(xprt);
948 xprt_wake_pending_tasks(xprt, status);
949 xprt_clear_connecting(xprt);
953 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
954 * @args: RPC transport to connect
956 * Invoked by a work queue tasklet.
958 static void xs_tcp_connect_worker(void *args)
960 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
961 struct socket *sock = xprt->sock;
962 int err, status = -EIO;
964 if (xprt->shutdown || xprt->addr.sin_port == 0)
967 dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt);
969 /* Start by resetting any existing socket state */
972 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
973 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
977 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
983 struct sock *sk = sock->sk;
985 write_lock_bh(&sk->sk_callback_lock);
987 sk->sk_user_data = xprt;
988 xprt->old_data_ready = sk->sk_data_ready;
989 xprt->old_state_change = sk->sk_state_change;
990 xprt->old_write_space = sk->sk_write_space;
991 sk->sk_data_ready = xs_tcp_data_ready;
992 sk->sk_state_change = xs_tcp_state_change;
993 sk->sk_write_space = xs_tcp_write_space;
994 tcp_sk(sk)->nonagle = 1;
996 xprt_clear_connected(xprt);
998 /* Reset to new socket */
1002 write_unlock_bh(&sk->sk_callback_lock);
1005 /* Tell the socket layer to start connecting... */
1006 status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1007 sizeof(xprt->addr), O_NONBLOCK);
1008 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1009 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1018 xprt_wake_pending_tasks(xprt, status);
1020 xprt_clear_connecting(xprt);
1024 * xs_connect - connect a socket to a remote endpoint
1025 * @task: address of RPC task that manages state of connect request
1027 * TCP: If the remote end dropped the connection, delay reconnecting.
1029 static void xs_connect(struct rpc_task *task)
1031 struct rpc_xprt *xprt = task->tk_xprt;
1033 if (xprt_test_and_set_connecting(xprt))
1036 if (xprt->sock != NULL) {
1037 dprintk("RPC: xs_connect delayed xprt %p\n", xprt);
1038 schedule_delayed_work(&xprt->connect_worker,
1039 RPC_REESTABLISH_TIMEOUT);
1041 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1042 schedule_work(&xprt->connect_worker);
1044 /* flush_scheduled_work can sleep... */
1045 if (!RPC_IS_ASYNC(task))
1046 flush_scheduled_work();
1050 static struct rpc_xprt_ops xs_udp_ops = {
1051 .set_buffer_size = xs_udp_set_buffer_size,
1052 .reserve_xprt = xprt_reserve_xprt_cong,
1053 .release_xprt = xprt_release_xprt_cong,
1054 .connect = xs_connect,
1055 .send_request = xs_udp_send_request,
1056 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1057 .timer = xs_udp_timer,
1058 .release_request = xprt_release_rqst_cong,
1060 .destroy = xs_destroy,
1063 static struct rpc_xprt_ops xs_tcp_ops = {
1064 .set_buffer_size = xs_tcp_set_buffer_size,
1065 .reserve_xprt = xprt_reserve_xprt,
1066 .release_xprt = xprt_release_xprt,
1067 .connect = xs_connect,
1068 .send_request = xs_tcp_send_request,
1069 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1071 .destroy = xs_destroy,
1075 * xs_setup_udp - Set up transport to use a UDP socket
1076 * @xprt: transport to set up
1077 * @to: timeout parameters
1080 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1082 size_t slot_table_size;
1084 dprintk("RPC: setting up udp-ipv4 transport...\n");
1086 xprt->max_reqs = xprt_udp_slot_table_entries;
1087 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1088 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1089 if (xprt->slot == NULL)
1091 memset(xprt->slot, 0, slot_table_size);
1093 xprt->prot = IPPROTO_UDP;
1094 xprt->port = xprt_max_resvport;
1096 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1097 /* XXX: header size can vary due to auth type, IPv6, etc. */
1098 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1100 INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1102 xprt->ops = &xs_udp_ops;
1105 xprt->timeout = *to;
1107 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1113 * xs_setup_tcp - Set up transport to use a TCP socket
1114 * @xprt: transport to set up
1115 * @to: timeout parameters
1118 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1120 size_t slot_table_size;
1122 dprintk("RPC: setting up tcp-ipv4 transport...\n");
1124 xprt->max_reqs = xprt_tcp_slot_table_entries;
1125 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1126 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1127 if (xprt->slot == NULL)
1129 memset(xprt->slot, 0, slot_table_size);
1131 xprt->prot = IPPROTO_TCP;
1132 xprt->port = xprt_max_resvport;
1133 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1134 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1135 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1137 INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1139 xprt->ops = &xs_tcp_ops;
1142 xprt->timeout = *to;
1144 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);