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 * Maximum port number to use when requesting a reserved port.
41 #define XS_MAX_RESVPORT (800U)
44 * How many times to try sending a request on a socket before waiting
45 * for the socket buffer to clear.
47 #define XS_SENDMSG_RETRY (10U)
50 # undef RPC_DEBUG_DATA
51 # define RPCDBG_FACILITY RPCDBG_TRANS
55 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
57 u8 *buf = (u8 *) packet;
60 dprintk("RPC: %s\n", msg);
61 for (j = 0; j < count && j < 128; j += 4) {
65 dprintk("0x%04x ", j);
67 dprintk("%02x%02x%02x%02x ",
68 buf[j], buf[j+1], buf[j+2], buf[j+3]);
73 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
79 #define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
81 static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
84 .iov_base = xdr->head[0].iov_base + base,
85 .iov_len = len - base,
89 .msg_namelen = addrlen,
90 .msg_flags = XS_SENDMSG_FLAGS,
94 msg.msg_flags |= MSG_MORE;
96 if (likely(iov.iov_len))
97 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
98 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
101 static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
104 .iov_base = xdr->tail[0].iov_base + base,
105 .iov_len = len - base,
107 struct msghdr msg = {
108 .msg_flags = XS_SENDMSG_FLAGS,
111 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
115 * xs_sendpages - write pages directly to a socket
116 * @sock: socket to send on
117 * @addr: UDP only -- address of destination
118 * @addrlen: UDP only -- length of destination address
119 * @xdr: buffer containing this request
120 * @base: starting position in the buffer
123 static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
125 struct page **ppage = xdr->pages;
126 unsigned int len, pglen = xdr->page_len;
128 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
133 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
135 len = xdr->head[0].iov_len;
136 if (base < len || (addr != NULL && base == 0)) {
137 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
142 if (err != (len - base))
148 if (unlikely(pglen == 0))
150 if (unlikely(base >= pglen)) {
154 if (base || xdr->page_base) {
156 base += xdr->page_base;
157 ppage += base >> PAGE_CACHE_SHIFT;
158 base &= ~PAGE_CACHE_MASK;
161 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
163 int flags = XS_SENDMSG_FLAGS;
165 len = PAGE_CACHE_SIZE;
171 if (pglen != len || xdr->tail[0].iov_len != 0)
174 /* Hmm... We might be dealing with highmem pages */
175 if (PageHighMem(*ppage))
176 sendpage = sock_no_sendpage;
177 err = sendpage(sock, *ppage, base, len, flags);
186 } while ((pglen -= len) != 0);
188 len = xdr->tail[0].iov_len;
190 err = xs_send_tail(sock, xdr, base, len);
201 * xs_nospace - place task on wait queue if transmit was incomplete
202 * @task: task to put to sleep
205 static void xs_nospace(struct rpc_task *task)
207 struct rpc_rqst *req = task->tk_rqstp;
208 struct rpc_xprt *xprt = req->rq_xprt;
210 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
211 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
214 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
215 /* Protect against races with write_space */
216 spin_lock_bh(&xprt->transport_lock);
218 /* Don't race with disconnect */
219 if (!xprt_connected(xprt))
220 task->tk_status = -ENOTCONN;
221 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
222 xprt_wait_for_buffer_space(task);
224 spin_unlock_bh(&xprt->transport_lock);
226 /* Keep holding the socket if it is blocked */
227 rpc_delay(task, HZ>>4);
231 * xs_udp_send_request - write an RPC request to a UDP socket
232 * @task: address of RPC task that manages the state of an RPC request
235 * 0: The request has been sent
236 * EAGAIN: The socket was blocked, please call again later to
237 * complete the request
238 * ENOTCONN: Caller needs to invoke connect logic then call again
239 * other: Some other error occured, the request was not sent
241 static int xs_udp_send_request(struct rpc_task *task)
243 struct rpc_rqst *req = task->tk_rqstp;
244 struct rpc_xprt *xprt = req->rq_xprt;
245 struct xdr_buf *xdr = &req->rq_snd_buf;
248 xs_pktdump("packet data:",
249 req->rq_svec->iov_base,
250 req->rq_svec->iov_len);
252 req->rq_xtime = jiffies;
253 status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
254 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
256 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
257 xdr->len - req->rq_bytes_sent, status);
259 if (likely(status >= (int) req->rq_slen))
262 /* Still some bytes left; set up for a retry later. */
270 /* When the server has died, an ICMP port unreachable message
271 * prompts ECONNREFUSED. */
277 dprintk("RPC: sendmsg returned unrecognized error %d\n",
285 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
287 u32 reclen = buf->len - sizeof(rpc_fraghdr);
288 rpc_fraghdr *base = buf->head[0].iov_base;
289 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
293 * xs_tcp_send_request - write an RPC request to a TCP socket
294 * @task: address of RPC task that manages the state of an RPC request
297 * 0: The request has been sent
298 * EAGAIN: The socket was blocked, please call again later to
299 * complete the request
300 * ENOTCONN: Caller needs to invoke connect logic then call again
301 * other: Some other error occured, the request was not sent
303 * XXX: In the case of soft timeouts, should we eventually give up
304 * if sendmsg is not able to make progress?
306 static int xs_tcp_send_request(struct rpc_task *task)
308 struct rpc_rqst *req = task->tk_rqstp;
309 struct rpc_xprt *xprt = req->rq_xprt;
310 struct xdr_buf *xdr = &req->rq_snd_buf;
311 int status, retry = 0;
313 xs_encode_tcp_record_marker(&req->rq_snd_buf);
315 xs_pktdump("packet data:",
316 req->rq_svec->iov_base,
317 req->rq_svec->iov_len);
319 /* Continue transmitting the packet/record. We must be careful
320 * to cope with writespace callbacks arriving _after_ we have
321 * called sendmsg(). */
323 req->rq_xtime = jiffies;
324 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
327 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
328 xdr->len - req->rq_bytes_sent, status);
330 if (unlikely(status < 0))
333 /* If we've sent the entire packet, immediately
334 * reset the count of bytes sent. */
335 req->rq_bytes_sent += status;
336 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
337 req->rq_bytes_sent = 0;
342 if (retry++ > XS_SENDMSG_RETRY)
357 dprintk("RPC: sendmsg returned unrecognized error %d\n",
359 xprt_disconnect(xprt);
367 * xs_close - close a socket
371 static void xs_close(struct rpc_xprt *xprt)
373 struct socket *sock = xprt->sock;
374 struct sock *sk = xprt->inet;
379 dprintk("RPC: xs_close xprt %p\n", xprt);
381 write_lock_bh(&sk->sk_callback_lock);
385 sk->sk_user_data = NULL;
386 sk->sk_data_ready = xprt->old_data_ready;
387 sk->sk_state_change = xprt->old_state_change;
388 sk->sk_write_space = xprt->old_write_space;
389 write_unlock_bh(&sk->sk_callback_lock);
397 * xs_destroy - prepare to shutdown a transport
398 * @xprt: doomed transport
401 static void xs_destroy(struct rpc_xprt *xprt)
403 dprintk("RPC: xs_destroy xprt %p\n", xprt);
405 cancel_delayed_work(&xprt->connect_worker);
406 flush_scheduled_work();
408 xprt_disconnect(xprt);
413 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
415 return (struct rpc_xprt *) sk->sk_user_data;
419 * xs_udp_data_ready - "data ready" callback for UDP sockets
420 * @sk: socket with data to read
421 * @len: how much data to read
424 static void xs_udp_data_ready(struct sock *sk, int len)
426 struct rpc_task *task;
427 struct rpc_xprt *xprt;
428 struct rpc_rqst *rovr;
430 int err, repsize, copied;
433 read_lock(&sk->sk_callback_lock);
434 dprintk("RPC: xs_udp_data_ready...\n");
435 if (!(xprt = xprt_from_sock(sk)))
438 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
444 repsize = skb->len - sizeof(struct udphdr);
446 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
450 /* Copy the XID from the skb... */
451 xp = skb_header_pointer(skb, sizeof(struct udphdr),
452 sizeof(_xid), &_xid);
456 /* Look up and lock the request corresponding to the given XID */
457 spin_lock(&xprt->transport_lock);
458 rovr = xprt_lookup_rqst(xprt, *xp);
461 task = rovr->rq_task;
463 if ((copied = rovr->rq_private_buf.buflen) > repsize)
466 /* Suck it into the iovec, verify checksum if not done by hw. */
467 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
470 /* Something worked... */
471 dst_confirm(skb->dst);
473 xprt_adjust_cwnd(task, copied);
474 xprt_update_rtt(task);
475 xprt_complete_rqst(task, copied);
478 spin_unlock(&xprt->transport_lock);
480 skb_free_datagram(sk, skb);
482 read_unlock(&sk->sk_callback_lock);
485 static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
487 if (len > desc->count)
489 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
490 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
496 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
501 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
506 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
507 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
508 used = xs_tcp_copy_data(desc, p, len);
509 xprt->tcp_offset += used;
513 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
514 if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
515 xprt->tcp_flags |= XPRT_LAST_FRAG;
517 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
518 xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
520 xprt->tcp_flags &= ~XPRT_COPY_RECM;
521 xprt->tcp_offset = 0;
523 /* Sanity check of the record length */
524 if (unlikely(xprt->tcp_reclen < 4)) {
525 dprintk("RPC: invalid TCP record fragment length\n");
526 xprt_disconnect(xprt);
529 dprintk("RPC: reading TCP record fragment of length %d\n",
533 static void xs_tcp_check_recm(struct rpc_xprt *xprt)
535 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
536 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
537 if (xprt->tcp_offset == xprt->tcp_reclen) {
538 xprt->tcp_flags |= XPRT_COPY_RECM;
539 xprt->tcp_offset = 0;
540 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
541 xprt->tcp_flags &= ~XPRT_COPY_DATA;
542 xprt->tcp_flags |= XPRT_COPY_XID;
543 xprt->tcp_copied = 0;
548 static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
553 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
554 dprintk("RPC: reading XID (%Zu bytes)\n", len);
555 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
556 used = xs_tcp_copy_data(desc, p, len);
557 xprt->tcp_offset += used;
560 xprt->tcp_flags &= ~XPRT_COPY_XID;
561 xprt->tcp_flags |= XPRT_COPY_DATA;
562 xprt->tcp_copied = 4;
563 dprintk("RPC: reading reply for XID %08x\n",
564 ntohl(xprt->tcp_xid));
565 xs_tcp_check_recm(xprt);
568 static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
570 struct rpc_rqst *req;
571 struct xdr_buf *rcvbuf;
575 /* Find and lock the request corresponding to this xid */
576 spin_lock(&xprt->transport_lock);
577 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
579 xprt->tcp_flags &= ~XPRT_COPY_DATA;
580 dprintk("RPC: XID %08x request not found!\n",
581 ntohl(xprt->tcp_xid));
582 spin_unlock(&xprt->transport_lock);
586 rcvbuf = &req->rq_private_buf;
588 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
589 skb_reader_t my_desc;
591 len = xprt->tcp_reclen - xprt->tcp_offset;
592 memcpy(&my_desc, desc, sizeof(my_desc));
594 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
595 &my_desc, xs_tcp_copy_data);
599 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
600 desc, xs_tcp_copy_data);
603 xprt->tcp_copied += r;
604 xprt->tcp_offset += r;
607 /* Error when copying to the receive buffer,
608 * usually because we weren't able to allocate
609 * additional buffer pages. All we can do now
610 * is turn off XPRT_COPY_DATA, so the request
611 * will not receive any additional updates,
613 * Any remaining data from this record will
616 xprt->tcp_flags &= ~XPRT_COPY_DATA;
617 dprintk("RPC: XID %08x truncated request\n",
618 ntohl(xprt->tcp_xid));
619 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
620 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
624 dprintk("RPC: XID %08x read %Zd bytes\n",
625 ntohl(xprt->tcp_xid), r);
626 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
627 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
629 if (xprt->tcp_copied == req->rq_private_buf.buflen)
630 xprt->tcp_flags &= ~XPRT_COPY_DATA;
631 else if (xprt->tcp_offset == xprt->tcp_reclen) {
632 if (xprt->tcp_flags & XPRT_LAST_FRAG)
633 xprt->tcp_flags &= ~XPRT_COPY_DATA;
637 if (!(xprt->tcp_flags & XPRT_COPY_DATA))
638 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
639 spin_unlock(&xprt->transport_lock);
640 xs_tcp_check_recm(xprt);
643 static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
647 len = xprt->tcp_reclen - xprt->tcp_offset;
648 if (len > desc->count)
652 xprt->tcp_offset += len;
653 dprintk("RPC: discarded %Zu bytes\n", len);
654 xs_tcp_check_recm(xprt);
657 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
659 struct rpc_xprt *xprt = rd_desc->arg.data;
660 skb_reader_t desc = {
667 dprintk("RPC: xs_tcp_data_recv started\n");
669 /* Read in a new fragment marker if necessary */
670 /* Can we ever really expect to get completely empty fragments? */
671 if (xprt->tcp_flags & XPRT_COPY_RECM) {
672 xs_tcp_read_fraghdr(xprt, &desc);
675 /* Read in the xid if necessary */
676 if (xprt->tcp_flags & XPRT_COPY_XID) {
677 xs_tcp_read_xid(xprt, &desc);
680 /* Read in the request data */
681 if (xprt->tcp_flags & XPRT_COPY_DATA) {
682 xs_tcp_read_request(xprt, &desc);
685 /* Skip over any trailing bytes on short reads */
686 xs_tcp_read_discard(xprt, &desc);
687 } while (desc.count);
688 dprintk("RPC: xs_tcp_data_recv done\n");
689 return len - desc.count;
693 * xs_tcp_data_ready - "data ready" callback for TCP sockets
694 * @sk: socket with data to read
695 * @bytes: how much data to read
698 static void xs_tcp_data_ready(struct sock *sk, int bytes)
700 struct rpc_xprt *xprt;
701 read_descriptor_t rd_desc;
703 read_lock(&sk->sk_callback_lock);
704 dprintk("RPC: xs_tcp_data_ready...\n");
705 if (!(xprt = xprt_from_sock(sk)))
710 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
711 rd_desc.arg.data = xprt;
712 rd_desc.count = 65536;
713 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
715 read_unlock(&sk->sk_callback_lock);
719 * xs_tcp_state_change - callback to handle TCP socket state changes
720 * @sk: socket whose state has changed
723 static void xs_tcp_state_change(struct sock *sk)
725 struct rpc_xprt *xprt;
727 read_lock(&sk->sk_callback_lock);
728 if (!(xprt = xprt_from_sock(sk)))
730 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
731 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
732 sk->sk_state, xprt_connected(xprt),
733 sock_flag(sk, SOCK_DEAD),
734 sock_flag(sk, SOCK_ZAPPED));
736 switch (sk->sk_state) {
737 case TCP_ESTABLISHED:
738 spin_lock_bh(&xprt->transport_lock);
739 if (!xprt_test_and_set_connected(xprt)) {
740 /* Reset TCP record info */
741 xprt->tcp_offset = 0;
742 xprt->tcp_reclen = 0;
743 xprt->tcp_copied = 0;
744 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
745 xprt_wake_pending_tasks(xprt, 0);
747 spin_unlock_bh(&xprt->transport_lock);
753 xprt_disconnect(xprt);
757 read_unlock(&sk->sk_callback_lock);
761 * xs_udp_write_space - callback invoked when socket buffer space
763 * @sk: socket whose state has changed
765 * Called when more output buffer space is available for this socket.
766 * We try not to wake our writers until they can make "significant"
767 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
768 * with a bunch of small requests.
770 static void xs_udp_write_space(struct sock *sk)
772 read_lock(&sk->sk_callback_lock);
774 /* from net/core/sock.c:sock_def_write_space */
775 if (sock_writeable(sk)) {
777 struct rpc_xprt *xprt;
779 if (unlikely(!(sock = sk->sk_socket)))
781 if (unlikely(!(xprt = xprt_from_sock(sk))))
783 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
786 xprt_write_space(xprt);
790 read_unlock(&sk->sk_callback_lock);
794 * xs_tcp_write_space - callback invoked when socket buffer space
796 * @sk: socket whose state has changed
798 * Called when more output buffer space is available for this socket.
799 * We try not to wake our writers until they can make "significant"
800 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
801 * with a bunch of small requests.
803 static void xs_tcp_write_space(struct sock *sk)
805 read_lock(&sk->sk_callback_lock);
807 /* from net/core/stream.c:sk_stream_write_space */
808 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
810 struct rpc_xprt *xprt;
812 if (unlikely(!(sock = sk->sk_socket)))
814 if (unlikely(!(xprt = xprt_from_sock(sk))))
816 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
819 xprt_write_space(xprt);
823 read_unlock(&sk->sk_callback_lock);
827 * xs_udp_set_buffer_size - set send and receive limits
828 * @xprt: generic transport
830 * Set socket send and receive limits based on the
831 * sndsize and rcvsize fields in the generic transport
834 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt)
836 struct sock *sk = xprt->inet;
839 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
840 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
843 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
844 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
845 sk->sk_write_space(sk);
850 * xs_tcp_set_buffer_size - set send and receive limits
851 * @xprt: generic transport
853 * Nothing to do for TCP.
855 static void xs_tcp_set_buffer_size(struct rpc_xprt *xprt)
861 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
862 * @task: task that timed out
864 * Adjust the congestion window after a retransmit timeout has occurred.
866 static void xs_udp_timer(struct rpc_task *task)
868 xprt_adjust_cwnd(task, -ETIMEDOUT);
871 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
873 struct sockaddr_in myaddr = {
874 .sin_family = AF_INET,
878 /* Were we already bound to a given port? Try to reuse it */
881 myaddr.sin_port = htons(port);
882 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
886 dprintk("RPC: xs_bindresvport bound to port %u\n",
891 port = XS_MAX_RESVPORT;
892 } while (err == -EADDRINUSE && port != xprt->port);
894 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
899 * xs_udp_connect_worker - set up a UDP socket
900 * @args: RPC transport to connect
902 * Invoked by a work queue tasklet.
904 static void xs_udp_connect_worker(void *args)
906 struct rpc_xprt *xprt = (struct rpc_xprt *) args;
907 struct socket *sock = xprt->sock;
908 int err, status = -EIO;
910 if (xprt->shutdown || xprt->addr.sin_port == 0)
913 dprintk("RPC: xs_udp_connect_worker for xprt %p\n", xprt);
915 /* Start by resetting any existing state */
918 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
919 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
923 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
929 struct sock *sk = sock->sk;
931 write_lock_bh(&sk->sk_callback_lock);
933 sk->sk_user_data = xprt;
934 xprt->old_data_ready = sk->sk_data_ready;
935 xprt->old_state_change = sk->sk_state_change;
936 xprt->old_write_space = sk->sk_write_space;
937 sk->sk_data_ready = xs_udp_data_ready;
938 sk->sk_write_space = xs_udp_write_space;
939 sk->sk_no_check = UDP_CSUM_NORCV;
941 xprt_set_connected(xprt);
943 /* Reset to new socket */
947 write_unlock_bh(&sk->sk_callback_lock);
949 xs_udp_set_buffer_size(xprt);
952 xprt_wake_pending_tasks(xprt, status);
953 xprt_clear_connecting(xprt);
957 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
958 * @args: RPC transport to connect
960 * Invoked by a work queue tasklet.
962 static void xs_tcp_connect_worker(void *args)
964 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
965 struct socket *sock = xprt->sock;
966 int err, status = -EIO;
968 if (xprt->shutdown || xprt->addr.sin_port == 0)
971 dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt);
973 /* Start by resetting any existing socket state */
976 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
977 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
981 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
987 struct sock *sk = sock->sk;
989 write_lock_bh(&sk->sk_callback_lock);
991 sk->sk_user_data = xprt;
992 xprt->old_data_ready = sk->sk_data_ready;
993 xprt->old_state_change = sk->sk_state_change;
994 xprt->old_write_space = sk->sk_write_space;
995 sk->sk_data_ready = xs_tcp_data_ready;
996 sk->sk_state_change = xs_tcp_state_change;
997 sk->sk_write_space = xs_tcp_write_space;
998 tcp_sk(sk)->nonagle = 1;
1000 xprt_clear_connected(xprt);
1002 /* Reset to new socket */
1006 write_unlock_bh(&sk->sk_callback_lock);
1009 /* Tell the socket layer to start connecting... */
1010 status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1011 sizeof(xprt->addr), O_NONBLOCK);
1012 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1013 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1022 xprt_wake_pending_tasks(xprt, status);
1024 xprt_clear_connecting(xprt);
1028 * xs_connect - connect a socket to a remote endpoint
1029 * @task: address of RPC task that manages state of connect request
1031 * TCP: If the remote end dropped the connection, delay reconnecting.
1033 static void xs_connect(struct rpc_task *task)
1035 struct rpc_xprt *xprt = task->tk_xprt;
1037 if (xprt_test_and_set_connecting(xprt))
1040 if (xprt->sock != NULL) {
1041 dprintk("RPC: xs_connect delayed xprt %p\n", xprt);
1042 schedule_delayed_work(&xprt->connect_worker,
1043 RPC_REESTABLISH_TIMEOUT);
1045 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1046 schedule_work(&xprt->connect_worker);
1048 /* flush_scheduled_work can sleep... */
1049 if (!RPC_IS_ASYNC(task))
1050 flush_scheduled_work();
1054 static struct rpc_xprt_ops xs_udp_ops = {
1055 .set_buffer_size = xs_udp_set_buffer_size,
1056 .reserve_xprt = xprt_reserve_xprt_cong,
1057 .release_xprt = xprt_release_xprt_cong,
1058 .connect = xs_connect,
1059 .send_request = xs_udp_send_request,
1060 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1061 .timer = xs_udp_timer,
1062 .release_request = xprt_release_rqst_cong,
1064 .destroy = xs_destroy,
1067 static struct rpc_xprt_ops xs_tcp_ops = {
1068 .set_buffer_size = xs_tcp_set_buffer_size,
1069 .reserve_xprt = xprt_reserve_xprt,
1070 .release_xprt = xprt_release_xprt,
1071 .connect = xs_connect,
1072 .send_request = xs_tcp_send_request,
1073 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1075 .destroy = xs_destroy,
1078 extern unsigned int xprt_udp_slot_table_entries;
1079 extern unsigned int xprt_tcp_slot_table_entries;
1082 * xs_setup_udp - Set up transport to use a UDP socket
1083 * @xprt: transport to set up
1084 * @to: timeout parameters
1087 int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1089 size_t slot_table_size;
1091 dprintk("RPC: setting up udp-ipv4 transport...\n");
1093 xprt->max_reqs = xprt_udp_slot_table_entries;
1094 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1095 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1096 if (xprt->slot == NULL)
1098 memset(xprt->slot, 0, slot_table_size);
1100 xprt->prot = IPPROTO_UDP;
1101 xprt->port = XS_MAX_RESVPORT;
1103 xprt->cwnd = RPC_INITCWND;
1104 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1105 /* XXX: header size can vary due to auth type, IPv6, etc. */
1106 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1108 INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1110 xprt->ops = &xs_udp_ops;
1113 xprt->timeout = *to;
1115 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1121 * xs_setup_tcp - Set up transport to use a TCP socket
1122 * @xprt: transport to set up
1123 * @to: timeout parameters
1126 int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1128 size_t slot_table_size;
1130 dprintk("RPC: setting up tcp-ipv4 transport...\n");
1132 xprt->max_reqs = xprt_tcp_slot_table_entries;
1133 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1134 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1135 if (xprt->slot == NULL)
1137 memset(xprt->slot, 0, slot_table_size);
1139 xprt->prot = IPPROTO_TCP;
1140 xprt->port = XS_MAX_RESVPORT;
1141 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1142 xprt->cwnd = RPC_MAXCWND(xprt);
1143 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1144 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1146 INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1148 xprt->ops = &xs_tcp_ops;
1151 xprt->timeout = *to;
1153 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);