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>
16 #include <linux/types.h>
17 #include <linux/slab.h>
18 #include <linux/capability.h>
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/errno.h>
22 #include <linux/socket.h>
24 #include <linux/net.h>
26 #include <linux/udp.h>
27 #include <linux/tcp.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/file.h>
32 #include <net/checksum.h>
37 * Maximum port number to use when requesting a reserved port.
39 #define XS_MAX_RESVPORT (800U)
42 # undef RPC_DEBUG_DATA
43 # define RPCDBG_FACILITY RPCDBG_TRANS
47 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
49 u8 *buf = (u8 *) packet;
52 dprintk("RPC: %s\n", msg);
53 for (j = 0; j < count && j < 128; j += 4) {
57 dprintk("0x%04x ", j);
59 dprintk("%02x%02x%02x%02x ",
60 buf[j], buf[j+1], buf[j+2], buf[j+3]);
65 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
71 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
73 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
76 .iov_base = xdr->head[0].iov_base + base,
77 .iov_len = len - base,
81 .msg_namelen = addrlen,
82 .msg_flags = XS_SENDMSG_FLAGS,
86 msg.msg_flags |= MSG_MORE;
88 if (likely(iov.iov_len))
89 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
90 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
93 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
96 .iov_base = xdr->tail[0].iov_base + base,
97 .iov_len = len - base,
100 .msg_flags = XS_SENDMSG_FLAGS,
103 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
107 * xs_sendpages - write pages directly to a socket
108 * @sock: socket to send on
109 * @addr: UDP only -- address of destination
110 * @addrlen: UDP only -- length of destination address
111 * @xdr: buffer containing this request
112 * @base: starting position in the buffer
115 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
117 struct page **ppage = xdr->pages;
118 unsigned int len, pglen = xdr->page_len;
120 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
122 len = xdr->head[0].iov_len;
123 if (base < len || (addr != NULL && base == 0)) {
124 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
129 if (err != (len - base))
135 if (unlikely(pglen == 0))
137 if (unlikely(base >= pglen)) {
141 if (base || xdr->page_base) {
143 base += xdr->page_base;
144 ppage += base >> PAGE_CACHE_SHIFT;
145 base &= ~PAGE_CACHE_MASK;
148 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
150 int flags = XS_SENDMSG_FLAGS;
152 len = PAGE_CACHE_SIZE;
158 if (pglen != len || xdr->tail[0].iov_len != 0)
161 /* Hmm... We might be dealing with highmem pages */
162 if (PageHighMem(*ppage))
163 sendpage = sock_no_sendpage;
164 err = sendpage(sock, *ppage, base, len, flags);
173 } while ((pglen -= len) != 0);
175 len = xdr->tail[0].iov_len;
177 err = xs_send_tail(sock, xdr, base, len);
188 * xs_sendmsg - write an RPC request to a socket
189 * @xprt: generic transport
190 * @req: the RPC request to write
193 static int xs_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
195 struct socket *sock = xprt->sock;
196 struct xdr_buf *xdr = &req->rq_snd_buf;
197 struct sockaddr *addr = NULL;
205 xs_pktdump("packet data:",
206 req->rq_svec->iov_base,
207 req->rq_svec->iov_len);
209 /* For UDP, we need to provide an address */
211 addr = (struct sockaddr *) &xprt->addr;
212 addrlen = sizeof(xprt->addr);
214 /* Don't repeat bytes */
215 skip = req->rq_bytes_sent;
217 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
218 result = xs_sendpages(sock, addr, addrlen, xdr, skip);
220 dprintk("RPC: xs_sendmsg(%d) = %d\n", xdr->len - skip, result);
227 /* When the server has died, an ICMP port unreachable message
228 * prompts ECONNREFUSED. */
234 /* connection broken */
245 * xs_send_request - write an RPC request to a socket
246 * @task: address of RPC task that manages the state of an RPC request
249 * 0: The request has been sent
250 * EAGAIN: The socket was blocked, please call again later to
251 * complete the request
252 * other: Some other error occured, the request was not sent
254 * XXX: In the case of soft timeouts, should we eventually give up
255 * if the socket is not able to make progress?
257 static int xs_send_request(struct rpc_task *task)
259 struct rpc_rqst *req = task->tk_rqstp;
260 struct rpc_xprt *xprt = req->rq_xprt;
261 int status, retry = 0;
263 /* set up everything as needed. */
264 /* Write the record marker */
266 u32 *marker = req->rq_svec[0].iov_base;
268 *marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
271 /* Continue transmitting the packet/record. We must be careful
272 * to cope with writespace callbacks arriving _after_ we have
276 req->rq_xtime = jiffies;
277 status = xs_sendmsg(xprt, req);
283 req->rq_bytes_sent += status;
285 /* If we've sent the entire packet, immediately
286 * reset the count of bytes sent. */
287 if (req->rq_bytes_sent >= req->rq_slen) {
288 req->rq_bytes_sent = 0;
292 if (status >= req->rq_slen)
298 dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
299 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
307 if (status == -EAGAIN) {
308 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
309 /* Protect against races with xs_write_space */
310 spin_lock_bh(&xprt->transport_lock);
311 /* Don't race with disconnect */
312 if (!xprt_connected(xprt))
313 task->tk_status = -ENOTCONN;
314 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) {
315 task->tk_timeout = req->rq_timeout;
316 rpc_sleep_on(&xprt->pending, task, NULL, NULL);
318 spin_unlock_bh(&xprt->transport_lock);
321 /* Keep holding the socket if it is blocked */
322 rpc_delay(task, HZ>>4);
328 * xs_close - close a socket
332 static void xs_close(struct rpc_xprt *xprt)
334 struct socket *sock = xprt->sock;
335 struct sock *sk = xprt->inet;
340 dprintk("RPC: xs_close xprt %p\n", xprt);
342 write_lock_bh(&sk->sk_callback_lock);
346 sk->sk_user_data = NULL;
347 sk->sk_data_ready = xprt->old_data_ready;
348 sk->sk_state_change = xprt->old_state_change;
349 sk->sk_write_space = xprt->old_write_space;
350 write_unlock_bh(&sk->sk_callback_lock);
358 * xs_destroy - prepare to shutdown a transport
359 * @xprt: doomed transport
362 static void xs_destroy(struct rpc_xprt *xprt)
364 dprintk("RPC: xs_destroy xprt %p\n", xprt);
366 cancel_delayed_work(&xprt->sock_connect);
367 flush_scheduled_work();
369 xprt_disconnect(xprt);
374 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
376 return (struct rpc_xprt *) sk->sk_user_data;
380 * xs_udp_data_ready - "data ready" callback for UDP sockets
381 * @sk: socket with data to read
382 * @len: how much data to read
385 static void xs_udp_data_ready(struct sock *sk, int len)
387 struct rpc_task *task;
388 struct rpc_xprt *xprt;
389 struct rpc_rqst *rovr;
391 int err, repsize, copied;
394 read_lock(&sk->sk_callback_lock);
395 dprintk("RPC: xs_udp_data_ready...\n");
396 if (!(xprt = xprt_from_sock(sk)))
399 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
405 repsize = skb->len - sizeof(struct udphdr);
407 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
411 /* Copy the XID from the skb... */
412 xp = skb_header_pointer(skb, sizeof(struct udphdr),
413 sizeof(_xid), &_xid);
417 /* Look up and lock the request corresponding to the given XID */
418 spin_lock(&xprt->transport_lock);
419 rovr = xprt_lookup_rqst(xprt, *xp);
422 task = rovr->rq_task;
424 dprintk("RPC: %4d received reply\n", task->tk_pid);
426 if ((copied = rovr->rq_private_buf.buflen) > repsize)
429 /* Suck it into the iovec, verify checksum if not done by hw. */
430 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
433 /* Something worked... */
434 dst_confirm(skb->dst);
436 xprt_complete_rqst(xprt, rovr, copied);
439 spin_unlock(&xprt->transport_lock);
441 skb_free_datagram(sk, skb);
443 read_unlock(&sk->sk_callback_lock);
446 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
448 if (len > desc->count)
450 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
451 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
457 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
462 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
467 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
468 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
469 used = xs_tcp_copy_data(desc, p, len);
470 xprt->tcp_offset += used;
473 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
474 if (xprt->tcp_reclen & 0x80000000)
475 xprt->tcp_flags |= XPRT_LAST_FRAG;
477 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
478 xprt->tcp_reclen &= 0x7fffffff;
479 xprt->tcp_flags &= ~XPRT_COPY_RECM;
480 xprt->tcp_offset = 0;
481 /* Sanity check of the record length */
482 if (xprt->tcp_reclen < 4) {
483 dprintk("RPC: invalid TCP record fragment length\n");
484 xprt_disconnect(xprt);
487 dprintk("RPC: reading TCP record fragment of length %d\n",
491 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
493 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
494 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
495 if (xprt->tcp_offset == xprt->tcp_reclen) {
496 xprt->tcp_flags |= XPRT_COPY_RECM;
497 xprt->tcp_offset = 0;
498 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
499 xprt->tcp_flags &= ~XPRT_COPY_DATA;
500 xprt->tcp_flags |= XPRT_COPY_XID;
501 xprt->tcp_copied = 0;
506 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
511 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
512 dprintk("RPC: reading XID (%Zu bytes)\n", len);
513 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
514 used = xs_tcp_copy_data(desc, p, len);
515 xprt->tcp_offset += used;
518 xprt->tcp_flags &= ~XPRT_COPY_XID;
519 xprt->tcp_flags |= XPRT_COPY_DATA;
520 xprt->tcp_copied = 4;
521 dprintk("RPC: reading reply for XID %08x\n",
522 ntohl(xprt->tcp_xid));
523 xs_tcp_check_recm(xprt);
526 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
528 struct rpc_rqst *req;
529 struct xdr_buf *rcvbuf;
533 /* Find and lock the request corresponding to this xid */
534 spin_lock(&xprt->transport_lock);
535 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
537 xprt->tcp_flags &= ~XPRT_COPY_DATA;
538 dprintk("RPC: XID %08x request not found!\n",
539 ntohl(xprt->tcp_xid));
540 spin_unlock(&xprt->transport_lock);
544 rcvbuf = &req->rq_private_buf;
546 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
547 skb_reader_t my_desc;
549 len = xprt->tcp_reclen - xprt->tcp_offset;
550 memcpy(&my_desc, desc, sizeof(my_desc));
552 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
553 &my_desc, xs_tcp_copy_data);
557 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
558 desc, xs_tcp_copy_data);
561 xprt->tcp_copied += r;
562 xprt->tcp_offset += r;
565 /* Error when copying to the receive buffer,
566 * usually because we weren't able to allocate
567 * additional buffer pages. All we can do now
568 * is turn off XPRT_COPY_DATA, so the request
569 * will not receive any additional updates,
571 * Any remaining data from this record will
574 xprt->tcp_flags &= ~XPRT_COPY_DATA;
575 dprintk("RPC: XID %08x truncated request\n",
576 ntohl(xprt->tcp_xid));
577 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
578 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
582 dprintk("RPC: XID %08x read %Zd bytes\n",
583 ntohl(xprt->tcp_xid), r);
584 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
585 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
587 if (xprt->tcp_copied == req->rq_private_buf.buflen)
588 xprt->tcp_flags &= ~XPRT_COPY_DATA;
589 else if (xprt->tcp_offset == xprt->tcp_reclen) {
590 if (xprt->tcp_flags & XPRT_LAST_FRAG)
591 xprt->tcp_flags &= ~XPRT_COPY_DATA;
595 if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
596 dprintk("RPC: %4d received reply complete\n",
597 req->rq_task->tk_pid);
598 xprt_complete_rqst(xprt, req, xprt->tcp_copied);
600 spin_unlock(&xprt->transport_lock);
601 xs_tcp_check_recm(xprt);
604 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
608 len = xprt->tcp_reclen - xprt->tcp_offset;
609 if (len > desc->count)
613 xprt->tcp_offset += len;
614 dprintk("RPC: discarded %Zu bytes\n", len);
615 xs_tcp_check_recm(xprt);
618 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
620 struct rpc_xprt *xprt = rd_desc->arg.data;
621 skb_reader_t desc = {
628 dprintk("RPC: xs_tcp_data_recv started\n");
630 /* Read in a new fragment marker if necessary */
631 /* Can we ever really expect to get completely empty fragments? */
632 if (xprt->tcp_flags & XPRT_COPY_RECM) {
633 xs_tcp_read_fraghdr(xprt, &desc);
636 /* Read in the xid if necessary */
637 if (xprt->tcp_flags & XPRT_COPY_XID) {
638 xs_tcp_read_xid(xprt, &desc);
641 /* Read in the request data */
642 if (xprt->tcp_flags & XPRT_COPY_DATA) {
643 xs_tcp_read_request(xprt, &desc);
646 /* Skip over any trailing bytes on short reads */
647 xs_tcp_read_discard(xprt, &desc);
648 } while (desc.count);
649 dprintk("RPC: xs_tcp_data_recv done\n");
650 return len - desc.count;
654 * xs_tcp_data_ready - "data ready" callback for TCP sockets
655 * @sk: socket with data to read
656 * @bytes: how much data to read
659 static void xs_tcp_data_ready(struct sock *sk, int bytes)
661 struct rpc_xprt *xprt;
662 read_descriptor_t rd_desc;
664 read_lock(&sk->sk_callback_lock);
665 dprintk("RPC: xs_tcp_data_ready...\n");
666 if (!(xprt = xprt_from_sock(sk)))
671 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
672 rd_desc.arg.data = xprt;
673 rd_desc.count = 65536;
674 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
676 read_unlock(&sk->sk_callback_lock);
680 * xs_tcp_state_change - callback to handle TCP socket state changes
681 * @sk: socket whose state has changed
684 static void xs_tcp_state_change(struct sock *sk)
686 struct rpc_xprt *xprt;
688 read_lock(&sk->sk_callback_lock);
689 if (!(xprt = xprt_from_sock(sk)))
691 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
692 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
693 sk->sk_state, xprt_connected(xprt),
694 sock_flag(sk, SOCK_DEAD),
695 sock_flag(sk, SOCK_ZAPPED));
697 switch (sk->sk_state) {
698 case TCP_ESTABLISHED:
699 spin_lock_bh(&xprt->transport_lock);
700 if (!xprt_test_and_set_connected(xprt)) {
701 /* Reset TCP record info */
702 xprt->tcp_offset = 0;
703 xprt->tcp_reclen = 0;
704 xprt->tcp_copied = 0;
705 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
706 xprt_wake_pending_tasks(xprt, 0);
708 spin_unlock_bh(&xprt->transport_lock);
714 xprt_disconnect(xprt);
718 read_unlock(&sk->sk_callback_lock);
722 * xs_write_space - callback invoked when socket buffer space becomes
724 * @sk: socket whose state has changed
726 * Called when more output buffer space is available for this socket.
727 * We try not to wake our writers until they can make "significant"
728 * progress, otherwise we'll waste resources thrashing sock_sendmsg
729 * with a bunch of small requests.
731 static void xs_write_space(struct sock *sk)
733 struct rpc_xprt *xprt;
736 read_lock(&sk->sk_callback_lock);
737 if (!(xprt = xprt_from_sock(sk)) || !(sock = sk->sk_socket))
742 /* Wait until we have enough socket memory */
744 /* from net/core/stream.c:sk_stream_write_space */
745 if (sk_stream_wspace(sk) < sk_stream_min_wspace(sk))
748 /* from net/core/sock.c:sock_def_write_space */
749 if (!sock_writeable(sk))
753 if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))
756 spin_lock_bh(&xprt->transport_lock);
758 rpc_wake_up_task(xprt->snd_task);
759 spin_unlock_bh(&xprt->transport_lock);
761 read_unlock(&sk->sk_callback_lock);
765 * xs_set_buffer_size - set send and receive limits
766 * @xprt: generic transport
768 * Set socket send and receive limits based on the
769 * sndsize and rcvsize fields in the generic transport
770 * structure. This applies only to UDP sockets.
772 static void xs_set_buffer_size(struct rpc_xprt *xprt)
774 struct sock *sk = xprt->inet;
779 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
780 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
783 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
784 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
785 sk->sk_write_space(sk);
789 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
791 struct sockaddr_in myaddr = {
792 .sin_family = AF_INET,
796 /* Were we already bound to a given port? Try to reuse it */
799 myaddr.sin_port = htons(port);
800 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
804 dprintk("RPC: xs_bindresvport bound to port %u\n",
809 port = XS_MAX_RESVPORT;
810 } while (err == -EADDRINUSE && port != xprt->port);
812 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
816 static struct socket *xs_create(struct rpc_xprt *xprt, int proto, int resvport)
821 dprintk("RPC: xs_create(%s %d)\n",
822 (proto == IPPROTO_UDP)? "udp" : "tcp", proto);
824 type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
826 if ((err = sock_create_kern(PF_INET, type, proto, &sock)) < 0) {
827 dprintk("RPC: can't create socket (%d).\n", -err);
831 /* If the caller has the capability, bind to a reserved port */
832 if (resvport && xs_bindresvport(xprt, sock) < 0)
842 static void xs_bind(struct rpc_xprt *xprt, struct socket *sock)
844 struct sock *sk = sock->sk;
849 write_lock_bh(&sk->sk_callback_lock);
850 sk->sk_user_data = xprt;
851 xprt->old_data_ready = sk->sk_data_ready;
852 xprt->old_state_change = sk->sk_state_change;
853 xprt->old_write_space = sk->sk_write_space;
854 if (xprt->prot == IPPROTO_UDP) {
855 sk->sk_data_ready = xs_udp_data_ready;
856 sk->sk_no_check = UDP_CSUM_NORCV;
857 xprt_set_connected(xprt);
859 tcp_sk(sk)->nonagle = 1; /* disable Nagle's algorithm */
860 sk->sk_data_ready = xs_tcp_data_ready;
861 sk->sk_state_change = xs_tcp_state_change;
862 xprt_clear_connected(xprt);
864 sk->sk_write_space = xs_write_space;
866 /* Reset to new socket */
869 write_unlock_bh(&sk->sk_callback_lock);
875 * xs_connect_worker - try to connect a socket to a remote endpoint
876 * @args: RPC transport to connect
878 * Invoked by a work queue tasklet.
880 static void xs_connect_worker(void *args)
882 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
883 struct socket *sock = xprt->sock;
886 if (xprt->shutdown || xprt->addr.sin_port == 0)
889 dprintk("RPC: xs_connect_worker xprt %p\n", xprt);
892 * Start by resetting any existing state
895 sock = xs_create(xprt, xprt->prot, xprt->resvport);
897 /* couldn't create socket or bind to reserved port;
898 * this is likely a permanent error, so cause an abort */
902 xs_set_buffer_size(xprt);
909 * Tell the socket layer to start connecting...
911 status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
912 sizeof(xprt->addr), O_NONBLOCK);
913 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
914 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
923 xprt_wake_pending_tasks(xprt, status);
925 xprt_clear_connecting(xprt);
929 * xs_connect - connect a socket to a remote endpoint
930 * @task: address of RPC task that manages state of connect request
932 * TCP: If the remote end dropped the connection, delay reconnecting.
934 static void xs_connect(struct rpc_task *task)
936 struct rpc_xprt *xprt = task->tk_xprt;
938 if (!xprt_test_and_set_connecting(xprt)) {
939 if (xprt->sock != NULL) {
940 dprintk("RPC: xs_connect delayed xprt %p\n", xprt);
941 schedule_delayed_work(&xprt->sock_connect,
942 RPC_REESTABLISH_TIMEOUT);
944 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
945 schedule_work(&xprt->sock_connect);
946 /* flush_scheduled_work can sleep... */
947 if (!RPC_IS_ASYNC(task))
948 flush_scheduled_work();
953 static struct rpc_xprt_ops xs_ops = {
954 .set_buffer_size = xs_set_buffer_size,
955 .connect = xs_connect,
956 .send_request = xs_send_request,
958 .destroy = xs_destroy,
961 extern unsigned int xprt_udp_slot_table_entries;
962 extern unsigned int xprt_tcp_slot_table_entries;
965 * xs_setup_udp - Set up transport to use a UDP socket
966 * @xprt: transport to set up
967 * @to: timeout parameters
970 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
972 size_t slot_table_size;
974 dprintk("RPC: setting up udp-ipv4 transport...\n");
976 xprt->max_reqs = xprt_udp_slot_table_entries;
977 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
978 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
979 if (xprt->slot == NULL)
981 memset(xprt->slot, 0, slot_table_size);
983 xprt->prot = IPPROTO_UDP;
984 xprt->port = XS_MAX_RESVPORT;
987 xprt->cwnd = RPC_INITCWND;
988 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
989 /* XXX: header size can vary due to auth type, IPv6, etc. */
990 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
992 INIT_WORK(&xprt->sock_connect, xs_connect_worker, xprt);
999 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1005 * xs_setup_tcp - Set up transport to use a TCP socket
1006 * @xprt: transport to set up
1007 * @to: timeout parameters
1010 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1012 size_t slot_table_size;
1014 dprintk("RPC: setting up tcp-ipv4 transport...\n");
1016 xprt->max_reqs = xprt_tcp_slot_table_entries;
1017 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1018 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1019 if (xprt->slot == NULL)
1021 memset(xprt->slot, 0, slot_table_size);
1023 xprt->prot = IPPROTO_TCP;
1024 xprt->port = XS_MAX_RESVPORT;
1027 xprt->cwnd = RPC_MAXCWND(xprt);
1028 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1029 xprt->max_payload = (1U << 31) - 1;
1031 INIT_WORK(&xprt->sock_connect, xs_connect_worker, xprt);
1033 xprt->ops = &xs_ops;
1036 xprt->timeout = *to;
1038 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);