2 * linux/net/sunrpc/svcsock.c
4 * These are the RPC server socket internals.
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
39 #include <net/checksum.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <trace/events/skb.h>
48 #include <linux/sunrpc/types.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/xdr.h>
51 #include <linux/sunrpc/msg_prot.h>
52 #include <linux/sunrpc/svcsock.h>
53 #include <linux/sunrpc/stats.h>
54 #include <linux/sunrpc/xprt.h>
58 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
61 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
63 static void svc_udp_data_ready(struct sock *, int);
64 static int svc_udp_recvfrom(struct svc_rqst *);
65 static int svc_udp_sendto(struct svc_rqst *);
66 static void svc_sock_detach(struct svc_xprt *);
67 static void svc_tcp_sock_detach(struct svc_xprt *);
68 static void svc_sock_free(struct svc_xprt *);
70 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
73 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
74 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
77 static void svc_bc_sock_free(struct svc_xprt *xprt);
78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 static struct lock_class_key svc_key[2];
82 static struct lock_class_key svc_slock_key[2];
84 static void svc_reclassify_socket(struct socket *sock)
86 struct sock *sk = sock->sk;
88 WARN_ON_ONCE(sock_owned_by_user(sk));
89 if (sock_owned_by_user(sk))
92 switch (sk->sk_family) {
94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
112 static void svc_reclassify_socket(struct socket *sock)
118 * Release an skbuff after use
120 static void svc_release_skb(struct svc_rqst *rqstp)
122 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
125 struct svc_sock *svsk =
126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
127 rqstp->rq_xprt_ctxt = NULL;
129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
130 skb_free_datagram_locked(svsk->sk_sk, skb);
134 union svc_pktinfo_u {
135 struct in_pktinfo pkti;
136 struct in6_pktinfo pkti6;
138 #define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
141 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
143 struct svc_sock *svsk =
144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 switch (svsk->sk_sk->sk_family) {
147 struct in_pktinfo *pki = CMSG_DATA(cmh);
149 cmh->cmsg_level = SOL_IP;
150 cmh->cmsg_type = IP_PKTINFO;
151 pki->ipi_ifindex = 0;
152 pki->ipi_spec_dst.s_addr =
153 svc_daddr_in(rqstp)->sin_addr.s_addr;
154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
159 struct in6_pktinfo *pki = CMSG_DATA(cmh);
160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
162 cmh->cmsg_level = SOL_IPV6;
163 cmh->cmsg_type = IPV6_PKTINFO;
164 pki->ipi6_ifindex = daddr->sin6_scope_id;
165 pki->ipi6_addr = daddr->sin6_addr;
166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
173 * send routine intended to be shared by the fore- and back-channel
175 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 struct page *headpage, unsigned long headoffset,
177 struct page *tailpage, unsigned long tailoffset)
181 struct page **ppage = xdr->pages;
182 size_t base = xdr->page_base;
183 unsigned int pglen = xdr->page_len;
184 unsigned int flags = MSG_MORE;
191 if (slen == xdr->head[0].iov_len)
193 len = kernel_sendpage(sock, headpage, headoffset,
194 xdr->head[0].iov_len, flags);
195 if (len != xdr->head[0].iov_len)
197 slen -= xdr->head[0].iov_len;
202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
206 result = kernel_sendpage(sock, *ppage, base, size, flags);
213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
219 if (xdr->tail[0].iov_len) {
220 result = kernel_sendpage(sock, tailpage, tailoffset,
221 xdr->tail[0].iov_len, 0);
232 * Generic sendto routine
234 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
236 struct svc_sock *svsk =
237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 struct socket *sock = svsk->sk_sock;
241 long all[SVC_PKTINFO_SPACE / sizeof(long)];
243 struct cmsghdr *cmh = &buffer.hdr;
245 unsigned long tailoff;
246 unsigned long headoff;
247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
249 if (rqstp->rq_prot == IPPROTO_UDP) {
250 struct msghdr msg = {
251 .msg_name = &rqstp->rq_addr,
252 .msg_namelen = rqstp->rq_addrlen,
254 .msg_controllen = sizeof(buffer),
255 .msg_flags = MSG_MORE,
258 svc_set_cmsg_data(rqstp, cmh);
260 if (sock_sendmsg(sock, &msg, 0) < 0)
264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 rqstp->rq_respages[0], tailoff);
270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
278 * Report socket names for nfsdfs
280 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
282 const struct sock *sk = svsk->sk_sk;
283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
287 switch (sk->sk_family) {
289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
291 &inet_sk(sk)->inet_rcv_saddr,
292 inet_sk(sk)->inet_num);
294 #if IS_ENABLED(CONFIG_IPV6)
296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
298 &sk->sk_v6_rcv_saddr,
299 inet_sk(sk)->inet_num);
303 len = snprintf(buf, remaining, "*unknown-%d*\n",
307 if (len >= remaining) {
309 return -ENAMETOOLONG;
315 * Check input queue length
317 static int svc_recv_available(struct svc_sock *svsk)
319 struct socket *sock = svsk->sk_sock;
322 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
324 return (err >= 0)? avail : err;
328 * Generic recvfrom routine.
330 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
333 struct svc_sock *svsk =
334 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
335 struct msghdr msg = {
336 .msg_flags = MSG_DONTWAIT,
340 rqstp->rq_xprt_hlen = 0;
342 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
345 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
346 svsk, iov[0].iov_base, iov[0].iov_len, len);
350 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
351 struct kvec *iov, int nr,
352 int buflen, unsigned int base)
360 return svc_recvfrom(rqstp, iov, nr, buflen);
362 for (i = 0; i < nr; i++) {
363 if (iov[i].iov_len > base)
365 base -= iov[i].iov_len;
367 save_iovlen = iov[i].iov_len;
368 save_iovbase = iov[i].iov_base;
369 iov[i].iov_len -= base;
370 iov[i].iov_base += base;
371 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
372 iov[i].iov_len = save_iovlen;
373 iov[i].iov_base = save_iovbase;
378 * Set socket snd and rcv buffer lengths
380 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
385 oldfs = get_fs(); set_fs(KERNEL_DS);
386 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
387 (char*)&snd, sizeof(snd));
388 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
389 (char*)&rcv, sizeof(rcv));
391 /* sock_setsockopt limits use to sysctl_?mem_max,
392 * which isn't acceptable. Until that is made conditional
393 * on not having CAP_SYS_RESOURCE or similar, we go direct...
394 * DaveM said I could!
397 sock->sk->sk_sndbuf = snd * 2;
398 sock->sk->sk_rcvbuf = rcv * 2;
399 sock->sk->sk_write_space(sock->sk);
400 release_sock(sock->sk);
404 * INET callback when data has been received on the socket.
406 static void svc_udp_data_ready(struct sock *sk, int count)
408 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
409 wait_queue_head_t *wq = sk_sleep(sk);
412 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
414 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
415 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
416 svc_xprt_enqueue(&svsk->sk_xprt);
418 if (wq && waitqueue_active(wq))
419 wake_up_interruptible(wq);
423 * INET callback when space is newly available on the socket.
425 static void svc_write_space(struct sock *sk)
427 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
428 wait_queue_head_t *wq = sk_sleep(sk);
431 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
432 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
433 svc_xprt_enqueue(&svsk->sk_xprt);
436 if (wq && waitqueue_active(wq)) {
437 dprintk("RPC svc_write_space: someone sleeping on %p\n",
439 wake_up_interruptible(wq);
443 static void svc_tcp_write_space(struct sock *sk)
445 struct socket *sock = sk->sk_socket;
447 if (sk_stream_is_writeable(sk) && sock)
448 clear_bit(SOCK_NOSPACE, &sock->flags);
453 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
455 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
458 struct in_pktinfo *pki = CMSG_DATA(cmh);
459 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
461 if (cmh->cmsg_type != IP_PKTINFO)
464 daddr->sin_family = AF_INET;
465 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
470 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
472 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
475 struct in6_pktinfo *pki = CMSG_DATA(cmh);
476 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
478 if (cmh->cmsg_type != IPV6_PKTINFO)
481 daddr->sin6_family = AF_INET6;
482 daddr->sin6_addr = pki->ipi6_addr;
483 daddr->sin6_scope_id = pki->ipi6_ifindex;
488 * Copy the UDP datagram's destination address to the rqstp structure.
489 * The 'destination' address in this case is the address to which the
490 * peer sent the datagram, i.e. our local address. For multihomed
491 * hosts, this can change from msg to msg. Note that only the IP
492 * address changes, the port number should remain the same.
494 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
497 switch (cmh->cmsg_level) {
499 return svc_udp_get_dest_address4(rqstp, cmh);
501 return svc_udp_get_dest_address6(rqstp, cmh);
508 * Receive a datagram from a UDP socket.
510 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
512 struct svc_sock *svsk =
513 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
514 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
518 long all[SVC_PKTINFO_SPACE / sizeof(long)];
520 struct cmsghdr *cmh = &buffer.hdr;
521 struct msghdr msg = {
522 .msg_name = svc_addr(rqstp),
524 .msg_controllen = sizeof(buffer),
525 .msg_flags = MSG_DONTWAIT,
530 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
531 /* udp sockets need large rcvbuf as all pending
532 * requests are still in that buffer. sndbuf must
533 * also be large enough that there is enough space
534 * for one reply per thread. We count all threads
535 * rather than threads in a particular pool, which
536 * provides an upper bound on the number of threads
537 * which will access the socket.
539 svc_sock_setbufsize(svsk->sk_sock,
540 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
541 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
543 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
545 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
546 0, 0, MSG_PEEK | MSG_DONTWAIT);
548 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
551 if (err != -EAGAIN) {
552 /* possibly an icmp error */
553 dprintk("svc: recvfrom returned error %d\n", -err);
554 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
558 len = svc_addr_len(svc_addr(rqstp));
559 rqstp->rq_addrlen = len;
560 if (skb->tstamp.tv64 == 0) {
561 skb->tstamp = ktime_get_real();
562 /* Don't enable netstamp, sunrpc doesn't
563 need that much accuracy */
565 svsk->sk_sk->sk_stamp = skb->tstamp;
566 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
568 len = skb->len - sizeof(struct udphdr);
569 rqstp->rq_arg.len = len;
571 rqstp->rq_prot = IPPROTO_UDP;
573 if (!svc_udp_get_dest_address(rqstp, cmh)) {
574 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
575 cmh->cmsg_level, cmh->cmsg_type);
578 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
580 if (skb_is_nonlinear(skb)) {
581 /* we have to copy */
583 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
589 skb_free_datagram_locked(svsk->sk_sk, skb);
591 /* we can use it in-place */
592 rqstp->rq_arg.head[0].iov_base = skb->data +
593 sizeof(struct udphdr);
594 rqstp->rq_arg.head[0].iov_len = len;
595 if (skb_checksum_complete(skb))
597 rqstp->rq_xprt_ctxt = skb;
600 rqstp->rq_arg.page_base = 0;
601 if (len <= rqstp->rq_arg.head[0].iov_len) {
602 rqstp->rq_arg.head[0].iov_len = len;
603 rqstp->rq_arg.page_len = 0;
604 rqstp->rq_respages = rqstp->rq_pages+1;
606 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
607 rqstp->rq_respages = rqstp->rq_pages + 1 +
608 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
610 rqstp->rq_next_page = rqstp->rq_respages+1;
613 serv->sv_stats->netudpcnt++;
617 trace_kfree_skb(skb, svc_udp_recvfrom);
618 skb_free_datagram_locked(svsk->sk_sk, skb);
623 svc_udp_sendto(struct svc_rqst *rqstp)
627 error = svc_sendto(rqstp, &rqstp->rq_res);
628 if (error == -ECONNREFUSED)
629 /* ICMP error on earlier request. */
630 error = svc_sendto(rqstp, &rqstp->rq_res);
635 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
639 static int svc_udp_has_wspace(struct svc_xprt *xprt)
641 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
642 struct svc_serv *serv = xprt->xpt_server;
643 unsigned long required;
646 * Set the SOCK_NOSPACE flag before checking the available
649 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
650 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
651 if (required*2 > sock_wspace(svsk->sk_sk))
653 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
657 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
663 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
665 struct sockaddr *sa, int salen,
668 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
671 static struct svc_xprt_ops svc_udp_ops = {
672 .xpo_create = svc_udp_create,
673 .xpo_recvfrom = svc_udp_recvfrom,
674 .xpo_sendto = svc_udp_sendto,
675 .xpo_release_rqst = svc_release_skb,
676 .xpo_detach = svc_sock_detach,
677 .xpo_free = svc_sock_free,
678 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
679 .xpo_has_wspace = svc_udp_has_wspace,
680 .xpo_accept = svc_udp_accept,
683 static struct svc_xprt_class svc_udp_class = {
685 .xcl_owner = THIS_MODULE,
686 .xcl_ops = &svc_udp_ops,
687 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
688 .xcl_ident = XPRT_TRANSPORT_UDP,
691 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
693 int err, level, optname, one = 1;
695 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
696 &svsk->sk_xprt, serv);
697 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
698 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
699 svsk->sk_sk->sk_write_space = svc_write_space;
701 /* initialise setting must have enough space to
702 * receive and respond to one request.
703 * svc_udp_recvfrom will re-adjust if necessary
705 svc_sock_setbufsize(svsk->sk_sock,
706 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
707 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
709 /* data might have come in before data_ready set up */
710 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
711 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
713 /* make sure we get destination address info */
714 switch (svsk->sk_sk->sk_family) {
717 optname = IP_PKTINFO;
721 optname = IPV6_RECVPKTINFO;
726 err = kernel_setsockopt(svsk->sk_sock, level, optname,
727 (char *)&one, sizeof(one));
728 dprintk("svc: kernel_setsockopt returned %d\n", err);
732 * A data_ready event on a listening socket means there's a connection
733 * pending. Do not use state_change as a substitute for it.
735 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
737 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
738 wait_queue_head_t *wq;
740 dprintk("svc: socket %p TCP (listen) state change %d\n",
744 * This callback may called twice when a new connection
745 * is established as a child socket inherits everything
746 * from a parent LISTEN socket.
747 * 1) data_ready method of the parent socket will be called
748 * when one of child sockets become ESTABLISHED.
749 * 2) data_ready method of the child socket may be called
750 * when it receives data before the socket is accepted.
751 * In case of 2, we should ignore it silently.
753 if (sk->sk_state == TCP_LISTEN) {
755 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
756 svc_xprt_enqueue(&svsk->sk_xprt);
758 printk("svc: socket %p: no user data\n", sk);
762 if (wq && waitqueue_active(wq))
763 wake_up_interruptible_all(wq);
767 * A state change on a connected socket means it's dying or dead.
769 static void svc_tcp_state_change(struct sock *sk)
771 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
772 wait_queue_head_t *wq = sk_sleep(sk);
774 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
775 sk, sk->sk_state, sk->sk_user_data);
778 printk("svc: socket %p: no user data\n", sk);
780 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
781 svc_xprt_enqueue(&svsk->sk_xprt);
783 if (wq && waitqueue_active(wq))
784 wake_up_interruptible_all(wq);
787 static void svc_tcp_data_ready(struct sock *sk, int count)
789 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
790 wait_queue_head_t *wq = sk_sleep(sk);
792 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
793 sk, sk->sk_user_data);
795 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
796 svc_xprt_enqueue(&svsk->sk_xprt);
798 if (wq && waitqueue_active(wq))
799 wake_up_interruptible(wq);
803 * Accept a TCP connection
805 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
807 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
808 struct sockaddr_storage addr;
809 struct sockaddr *sin = (struct sockaddr *) &addr;
810 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
811 struct socket *sock = svsk->sk_sock;
812 struct socket *newsock;
813 struct svc_sock *newsvsk;
815 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
817 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
821 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
822 err = kernel_accept(sock, &newsock, O_NONBLOCK);
825 printk(KERN_WARNING "%s: no more sockets!\n",
827 else if (err != -EAGAIN)
828 net_warn_ratelimited("%s: accept failed (err %d)!\n",
829 serv->sv_name, -err);
832 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
834 err = kernel_getpeername(newsock, sin, &slen);
836 net_warn_ratelimited("%s: peername failed (err %d)!\n",
837 serv->sv_name, -err);
838 goto failed; /* aborted connection or whatever */
841 /* Ideally, we would want to reject connections from unauthorized
842 * hosts here, but when we get encryption, the IP of the host won't
843 * tell us anything. For now just warn about unpriv connections.
845 if (!svc_port_is_privileged(sin)) {
847 "%s: connect from unprivileged port: %s\n",
849 __svc_print_addr(sin, buf, sizeof(buf)));
851 dprintk("%s: connect from %s\n", serv->sv_name,
852 __svc_print_addr(sin, buf, sizeof(buf)));
854 /* make sure that a write doesn't block forever when
857 newsock->sk->sk_sndtimeo = HZ*30;
859 newsvsk = svc_setup_socket(serv, newsock,
860 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
863 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
864 err = kernel_getsockname(newsock, sin, &slen);
865 if (unlikely(err < 0)) {
866 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
867 slen = offsetof(struct sockaddr, sa_data);
869 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
872 serv->sv_stats->nettcpconn++;
874 return &newsvsk->sk_xprt;
877 sock_release(newsock);
881 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
883 unsigned int i, len, npages;
885 if (svsk->sk_datalen == 0)
887 len = svsk->sk_datalen;
888 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
889 for (i = 0; i < npages; i++) {
890 if (rqstp->rq_pages[i] != NULL)
891 put_page(rqstp->rq_pages[i]);
892 BUG_ON(svsk->sk_pages[i] == NULL);
893 rqstp->rq_pages[i] = svsk->sk_pages[i];
894 svsk->sk_pages[i] = NULL;
896 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
900 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
902 unsigned int i, len, npages;
904 if (svsk->sk_datalen == 0)
906 len = svsk->sk_datalen;
907 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
908 for (i = 0; i < npages; i++) {
909 svsk->sk_pages[i] = rqstp->rq_pages[i];
910 rqstp->rq_pages[i] = NULL;
914 static void svc_tcp_clear_pages(struct svc_sock *svsk)
916 unsigned int i, len, npages;
918 if (svsk->sk_datalen == 0)
920 len = svsk->sk_datalen;
921 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
922 for (i = 0; i < npages; i++) {
923 if (svsk->sk_pages[i] == NULL) {
927 put_page(svsk->sk_pages[i]);
928 svsk->sk_pages[i] = NULL;
932 svsk->sk_datalen = 0;
936 * Receive fragment record header.
937 * If we haven't gotten the record length yet, get the next four bytes.
939 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
941 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
945 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
947 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
950 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
951 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
953 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
955 svsk->sk_tcplen += len;
958 dprintk("svc: short recvfrom while reading record "
959 "length (%d of %d)\n", len, want);
963 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
964 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
966 net_notice_ratelimited("RPC: fragment too large: %d\n",
967 svc_sock_reclen(svsk));
972 return svc_sock_reclen(svsk);
974 dprintk("RPC: TCP recv_record got %d\n", len);
977 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
981 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
983 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
984 struct rpc_rqst *req = NULL;
985 struct kvec *src, *dst;
986 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
994 req = xprt_lookup_rqst(bc_xprt, xid);
998 "%s: Got unrecognized reply: "
999 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1000 __func__, ntohl(calldir),
1005 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1007 * XXX!: cheating for now! Only copying HEAD.
1008 * But we know this is good enough for now (in fact, for any
1009 * callback reply in the forseeable future).
1011 dst = &req->rq_private_buf.head[0];
1012 src = &rqstp->rq_arg.head[0];
1013 if (dst->iov_len < src->iov_len)
1014 return -EAGAIN; /* whatever; just giving up. */
1015 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1016 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1017 rqstp->rq_arg.len = 0;
1021 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1027 vec[i].iov_base = page_address(pages[i]);
1028 vec[i].iov_len = PAGE_SIZE;
1035 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1037 /* If we have more data, signal svc_xprt_enqueue() to try again */
1038 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1039 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1040 dprintk("svc: TCP %s record (%d bytes)\n",
1041 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1042 svc_sock_reclen(svsk));
1043 svsk->sk_tcplen = 0;
1044 svsk->sk_reclen = 0;
1048 * Receive data from a TCP socket.
1050 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1052 struct svc_sock *svsk =
1053 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1054 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1057 unsigned int want, base;
1062 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1063 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1064 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1065 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1067 len = svc_tcp_recv_record(svsk, rqstp);
1071 base = svc_tcp_restore_pages(svsk, rqstp);
1072 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1074 vec = rqstp->rq_vec;
1076 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1077 svsk->sk_datalen + want);
1079 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1080 rqstp->rq_next_page = rqstp->rq_respages + 1;
1082 /* Now receive data */
1083 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1085 svsk->sk_tcplen += len;
1086 svsk->sk_datalen += len;
1088 if (len != want || !svc_sock_final_rec(svsk)) {
1089 svc_tcp_save_pages(svsk, rqstp);
1090 if (len < 0 && len != -EAGAIN)
1093 svc_tcp_fragment_received(svsk);
1095 dprintk("svc: incomplete TCP record (%d of %d)\n",
1096 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1097 svc_sock_reclen(svsk));
1101 if (svsk->sk_datalen < 8) {
1102 svsk->sk_datalen = 0;
1103 goto err_delete; /* client is nuts. */
1106 rqstp->rq_arg.len = svsk->sk_datalen;
1107 rqstp->rq_arg.page_base = 0;
1108 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1109 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1110 rqstp->rq_arg.page_len = 0;
1112 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1114 rqstp->rq_xprt_ctxt = NULL;
1115 rqstp->rq_prot = IPPROTO_TCP;
1117 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1120 len = receive_cb_reply(svsk, rqstp);
1122 /* Reset TCP read info */
1123 svsk->sk_datalen = 0;
1124 svc_tcp_fragment_received(svsk);
1129 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1131 serv->sv_stats->nettcpcnt++;
1133 return rqstp->rq_arg.len;
1138 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1141 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1142 svsk->sk_xprt.xpt_server->sv_name, -len);
1143 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1145 return 0; /* record not complete */
1149 * Send out data on TCP socket.
1151 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1153 struct xdr_buf *xbufp = &rqstp->rq_res;
1157 /* Set up the first element of the reply kvec.
1158 * Any other kvecs that may be in use have been taken
1159 * care of by the server implementation itself.
1161 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1162 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1164 sent = svc_sendto(rqstp, &rqstp->rq_res);
1165 if (sent != xbufp->len) {
1167 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1168 "- shutting down socket\n",
1169 rqstp->rq_xprt->xpt_server->sv_name,
1170 (sent<0)?"got error":"sent only",
1172 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1173 svc_xprt_enqueue(rqstp->rq_xprt);
1180 * Setup response header. TCP has a 4B record length field.
1182 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1184 struct kvec *resv = &rqstp->rq_res.head[0];
1186 /* tcp needs a space for the record length... */
1190 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1192 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1193 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1196 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1198 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1199 if (sk_stream_wspace(svsk->sk_sk) >= required ||
1200 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
1201 atomic_read(&xprt->xpt_reserved) == 0))
1203 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1207 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1209 struct sockaddr *sa, int salen,
1212 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1215 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1216 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1217 struct net *, struct sockaddr *,
1219 static void svc_bc_sock_free(struct svc_xprt *xprt);
1221 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1223 struct sockaddr *sa, int salen,
1226 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1229 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1233 static struct svc_xprt_ops svc_tcp_bc_ops = {
1234 .xpo_create = svc_bc_tcp_create,
1235 .xpo_detach = svc_bc_tcp_sock_detach,
1236 .xpo_free = svc_bc_sock_free,
1237 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1240 static struct svc_xprt_class svc_tcp_bc_class = {
1241 .xcl_name = "tcp-bc",
1242 .xcl_owner = THIS_MODULE,
1243 .xcl_ops = &svc_tcp_bc_ops,
1244 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1247 static void svc_init_bc_xprt_sock(void)
1249 svc_reg_xprt_class(&svc_tcp_bc_class);
1252 static void svc_cleanup_bc_xprt_sock(void)
1254 svc_unreg_xprt_class(&svc_tcp_bc_class);
1256 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1257 static void svc_init_bc_xprt_sock(void)
1261 static void svc_cleanup_bc_xprt_sock(void)
1264 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1266 static struct svc_xprt_ops svc_tcp_ops = {
1267 .xpo_create = svc_tcp_create,
1268 .xpo_recvfrom = svc_tcp_recvfrom,
1269 .xpo_sendto = svc_tcp_sendto,
1270 .xpo_release_rqst = svc_release_skb,
1271 .xpo_detach = svc_tcp_sock_detach,
1272 .xpo_free = svc_sock_free,
1273 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1274 .xpo_has_wspace = svc_tcp_has_wspace,
1275 .xpo_accept = svc_tcp_accept,
1278 static struct svc_xprt_class svc_tcp_class = {
1280 .xcl_owner = THIS_MODULE,
1281 .xcl_ops = &svc_tcp_ops,
1282 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1283 .xcl_ident = XPRT_TRANSPORT_TCP,
1286 void svc_init_xprt_sock(void)
1288 svc_reg_xprt_class(&svc_tcp_class);
1289 svc_reg_xprt_class(&svc_udp_class);
1290 svc_init_bc_xprt_sock();
1293 void svc_cleanup_xprt_sock(void)
1295 svc_unreg_xprt_class(&svc_tcp_class);
1296 svc_unreg_xprt_class(&svc_udp_class);
1297 svc_cleanup_bc_xprt_sock();
1300 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1302 struct sock *sk = svsk->sk_sk;
1304 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1305 &svsk->sk_xprt, serv);
1306 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1307 if (sk->sk_state == TCP_LISTEN) {
1308 dprintk("setting up TCP socket for listening\n");
1309 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1310 sk->sk_data_ready = svc_tcp_listen_data_ready;
1311 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1313 dprintk("setting up TCP socket for reading\n");
1314 sk->sk_state_change = svc_tcp_state_change;
1315 sk->sk_data_ready = svc_tcp_data_ready;
1316 sk->sk_write_space = svc_tcp_write_space;
1318 svsk->sk_reclen = 0;
1319 svsk->sk_tcplen = 0;
1320 svsk->sk_datalen = 0;
1321 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1323 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1325 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1326 if (sk->sk_state != TCP_ESTABLISHED)
1327 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1331 void svc_sock_update_bufs(struct svc_serv *serv)
1334 * The number of server threads has changed. Update
1335 * rcvbuf and sndbuf accordingly on all sockets
1337 struct svc_sock *svsk;
1339 spin_lock_bh(&serv->sv_lock);
1340 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1341 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1342 spin_unlock_bh(&serv->sv_lock);
1344 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1347 * Initialize socket for RPC use and create svc_sock struct
1348 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1350 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1351 struct socket *sock,
1354 struct svc_sock *svsk;
1356 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1359 dprintk("svc: svc_setup_socket %p\n", sock);
1360 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1362 return ERR_PTR(-ENOMEM);
1366 /* Register socket with portmapper */
1368 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1370 ntohs(inet_sk(inet)->inet_sport));
1374 return ERR_PTR(err);
1377 inet->sk_user_data = svsk;
1378 svsk->sk_sock = sock;
1380 svsk->sk_ostate = inet->sk_state_change;
1381 svsk->sk_odata = inet->sk_data_ready;
1382 svsk->sk_owspace = inet->sk_write_space;
1384 /* Initialize the socket */
1385 if (sock->type == SOCK_DGRAM)
1386 svc_udp_init(svsk, serv);
1388 /* initialise setting must have enough space to
1389 * receive and respond to one request.
1391 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1392 4 * serv->sv_max_mesg);
1393 svc_tcp_init(svsk, serv);
1396 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1402 bool svc_alien_sock(struct net *net, int fd)
1405 struct socket *sock = sockfd_lookup(fd, &err);
1410 if (sock_net(sock->sk) != net)
1416 EXPORT_SYMBOL_GPL(svc_alien_sock);
1419 * svc_addsock - add a listener socket to an RPC service
1420 * @serv: pointer to RPC service to which to add a new listener
1421 * @fd: file descriptor of the new listener
1422 * @name_return: pointer to buffer to fill in with name of listener
1423 * @len: size of the buffer
1425 * Fills in socket name and returns positive length of name if successful.
1426 * Name is terminated with '\n'. On error, returns a negative errno
1429 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1433 struct socket *so = sockfd_lookup(fd, &err);
1434 struct svc_sock *svsk = NULL;
1435 struct sockaddr_storage addr;
1436 struct sockaddr *sin = (struct sockaddr *)&addr;
1441 err = -EAFNOSUPPORT;
1442 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1444 err = -EPROTONOSUPPORT;
1445 if (so->sk->sk_protocol != IPPROTO_TCP &&
1446 so->sk->sk_protocol != IPPROTO_UDP)
1449 if (so->state > SS_UNCONNECTED)
1452 if (!try_module_get(THIS_MODULE))
1454 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1456 module_put(THIS_MODULE);
1457 err = PTR_ERR(svsk);
1460 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1461 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1462 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1463 return svc_one_sock_name(svsk, name_return, len);
1468 EXPORT_SYMBOL_GPL(svc_addsock);
1471 * Create socket for RPC service.
1473 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1476 struct sockaddr *sin, int len,
1479 struct svc_sock *svsk;
1480 struct socket *sock;
1483 struct sockaddr_storage addr;
1484 struct sockaddr *newsin = (struct sockaddr *)&addr;
1488 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1490 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1491 serv->sv_program->pg_name, protocol,
1492 __svc_print_addr(sin, buf, sizeof(buf)));
1494 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1495 printk(KERN_WARNING "svc: only UDP and TCP "
1496 "sockets supported\n");
1497 return ERR_PTR(-EINVAL);
1500 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1501 switch (sin->sa_family) {
1509 return ERR_PTR(-EINVAL);
1512 error = __sock_create(net, family, type, protocol, &sock, 1);
1514 return ERR_PTR(error);
1516 svc_reclassify_socket(sock);
1519 * If this is an PF_INET6 listener, we want to avoid
1520 * getting requests from IPv4 remotes. Those should
1521 * be shunted to a PF_INET listener via rpcbind.
1524 if (family == PF_INET6)
1525 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1526 (char *)&val, sizeof(val));
1528 if (type == SOCK_STREAM)
1529 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1530 error = kernel_bind(sock, sin, len);
1535 error = kernel_getsockname(sock, newsin, &newlen);
1539 if (protocol == IPPROTO_TCP) {
1540 if ((error = kernel_listen(sock, 64)) < 0)
1544 svsk = svc_setup_socket(serv, sock, flags);
1546 error = PTR_ERR(svsk);
1549 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1550 return (struct svc_xprt *)svsk;
1552 dprintk("svc: svc_create_socket error = %d\n", -error);
1554 return ERR_PTR(error);
1558 * Detach the svc_sock from the socket so that no
1559 * more callbacks occur.
1561 static void svc_sock_detach(struct svc_xprt *xprt)
1563 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1564 struct sock *sk = svsk->sk_sk;
1565 wait_queue_head_t *wq;
1567 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1569 /* put back the old socket callbacks */
1570 sk->sk_state_change = svsk->sk_ostate;
1571 sk->sk_data_ready = svsk->sk_odata;
1572 sk->sk_write_space = svsk->sk_owspace;
1575 if (wq && waitqueue_active(wq))
1576 wake_up_interruptible(wq);
1580 * Disconnect the socket, and reset the callbacks
1582 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1584 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1586 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1588 svc_sock_detach(xprt);
1590 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1591 svc_tcp_clear_pages(svsk);
1592 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1597 * Free the svc_sock's socket resources and the svc_sock itself.
1599 static void svc_sock_free(struct svc_xprt *xprt)
1601 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1602 dprintk("svc: svc_sock_free(%p)\n", svsk);
1604 if (svsk->sk_sock->file)
1605 sockfd_put(svsk->sk_sock);
1607 sock_release(svsk->sk_sock);
1611 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1613 * Create a back channel svc_xprt which shares the fore channel socket.
1615 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1618 struct sockaddr *sin, int len,
1621 struct svc_sock *svsk;
1622 struct svc_xprt *xprt;
1624 if (protocol != IPPROTO_TCP) {
1625 printk(KERN_WARNING "svc: only TCP sockets"
1626 " supported on shared back channel\n");
1627 return ERR_PTR(-EINVAL);
1630 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1632 return ERR_PTR(-ENOMEM);
1634 xprt = &svsk->sk_xprt;
1635 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1637 serv->sv_bc_xprt = xprt;
1643 * Free a back channel svc_sock.
1645 static void svc_bc_sock_free(struct svc_xprt *xprt)
1648 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1650 #endif /* CONFIG_SUNRPC_BACKCHANNEL */