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 *);
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 static int svc_sock_secure_port(struct svc_rqst *rqstp)
406 return svc_port_is_privileged(svc_addr(rqstp));
410 * INET callback when data has been received on the socket.
412 static void svc_udp_data_ready(struct sock *sk)
414 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
415 wait_queue_head_t *wq = sk_sleep(sk);
418 dprintk("svc: socket %p(inet %p), busy=%d\n",
420 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
421 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
422 svc_xprt_enqueue(&svsk->sk_xprt);
424 if (wq && waitqueue_active(wq))
425 wake_up_interruptible(wq);
429 * INET callback when space is newly available on the socket.
431 static void svc_write_space(struct sock *sk)
433 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
434 wait_queue_head_t *wq = sk_sleep(sk);
437 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
438 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
439 svc_xprt_enqueue(&svsk->sk_xprt);
442 if (wq && waitqueue_active(wq)) {
443 dprintk("RPC svc_write_space: someone sleeping on %p\n",
445 wake_up_interruptible(wq);
449 static void svc_tcp_write_space(struct sock *sk)
451 struct socket *sock = sk->sk_socket;
453 if (sk_stream_is_writeable(sk) && sock)
454 clear_bit(SOCK_NOSPACE, &sock->flags);
459 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
461 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
464 struct in_pktinfo *pki = CMSG_DATA(cmh);
465 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
467 if (cmh->cmsg_type != IP_PKTINFO)
470 daddr->sin_family = AF_INET;
471 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
476 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
478 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
481 struct in6_pktinfo *pki = CMSG_DATA(cmh);
482 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
484 if (cmh->cmsg_type != IPV6_PKTINFO)
487 daddr->sin6_family = AF_INET6;
488 daddr->sin6_addr = pki->ipi6_addr;
489 daddr->sin6_scope_id = pki->ipi6_ifindex;
494 * Copy the UDP datagram's destination address to the rqstp structure.
495 * The 'destination' address in this case is the address to which the
496 * peer sent the datagram, i.e. our local address. For multihomed
497 * hosts, this can change from msg to msg. Note that only the IP
498 * address changes, the port number should remain the same.
500 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
503 switch (cmh->cmsg_level) {
505 return svc_udp_get_dest_address4(rqstp, cmh);
507 return svc_udp_get_dest_address6(rqstp, cmh);
514 * Receive a datagram from a UDP socket.
516 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
518 struct svc_sock *svsk =
519 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
520 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
524 long all[SVC_PKTINFO_SPACE / sizeof(long)];
526 struct cmsghdr *cmh = &buffer.hdr;
527 struct msghdr msg = {
528 .msg_name = svc_addr(rqstp),
530 .msg_controllen = sizeof(buffer),
531 .msg_flags = MSG_DONTWAIT,
536 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
537 /* udp sockets need large rcvbuf as all pending
538 * requests are still in that buffer. sndbuf must
539 * also be large enough that there is enough space
540 * for one reply per thread. We count all threads
541 * rather than threads in a particular pool, which
542 * provides an upper bound on the number of threads
543 * which will access the socket.
545 svc_sock_setbufsize(svsk->sk_sock,
546 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
547 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
549 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
551 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
552 0, 0, MSG_PEEK | MSG_DONTWAIT);
554 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
557 if (err != -EAGAIN) {
558 /* possibly an icmp error */
559 dprintk("svc: recvfrom returned error %d\n", -err);
560 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
564 len = svc_addr_len(svc_addr(rqstp));
565 rqstp->rq_addrlen = len;
566 if (skb->tstamp.tv64 == 0) {
567 skb->tstamp = ktime_get_real();
568 /* Don't enable netstamp, sunrpc doesn't
569 need that much accuracy */
571 svsk->sk_sk->sk_stamp = skb->tstamp;
572 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
574 len = skb->len - sizeof(struct udphdr);
575 rqstp->rq_arg.len = len;
577 rqstp->rq_prot = IPPROTO_UDP;
579 if (!svc_udp_get_dest_address(rqstp, cmh)) {
580 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
581 cmh->cmsg_level, cmh->cmsg_type);
584 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
586 if (skb_is_nonlinear(skb)) {
587 /* we have to copy */
589 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
595 skb_free_datagram_locked(svsk->sk_sk, skb);
597 /* we can use it in-place */
598 rqstp->rq_arg.head[0].iov_base = skb->data +
599 sizeof(struct udphdr);
600 rqstp->rq_arg.head[0].iov_len = len;
601 if (skb_checksum_complete(skb))
603 rqstp->rq_xprt_ctxt = skb;
606 rqstp->rq_arg.page_base = 0;
607 if (len <= rqstp->rq_arg.head[0].iov_len) {
608 rqstp->rq_arg.head[0].iov_len = len;
609 rqstp->rq_arg.page_len = 0;
610 rqstp->rq_respages = rqstp->rq_pages+1;
612 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
613 rqstp->rq_respages = rqstp->rq_pages + 1 +
614 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
616 rqstp->rq_next_page = rqstp->rq_respages+1;
619 serv->sv_stats->netudpcnt++;
623 trace_kfree_skb(skb, svc_udp_recvfrom);
624 skb_free_datagram_locked(svsk->sk_sk, skb);
629 svc_udp_sendto(struct svc_rqst *rqstp)
633 error = svc_sendto(rqstp, &rqstp->rq_res);
634 if (error == -ECONNREFUSED)
635 /* ICMP error on earlier request. */
636 error = svc_sendto(rqstp, &rqstp->rq_res);
641 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
645 static int svc_udp_has_wspace(struct svc_xprt *xprt)
647 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
648 struct svc_serv *serv = xprt->xpt_server;
649 unsigned long required;
652 * Set the SOCK_NOSPACE flag before checking the available
655 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
656 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
657 if (required*2 > sock_wspace(svsk->sk_sk))
659 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
663 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
669 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
671 struct sockaddr *sa, int salen,
674 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
677 static struct svc_xprt_ops svc_udp_ops = {
678 .xpo_create = svc_udp_create,
679 .xpo_recvfrom = svc_udp_recvfrom,
680 .xpo_sendto = svc_udp_sendto,
681 .xpo_release_rqst = svc_release_skb,
682 .xpo_detach = svc_sock_detach,
683 .xpo_free = svc_sock_free,
684 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
685 .xpo_has_wspace = svc_udp_has_wspace,
686 .xpo_accept = svc_udp_accept,
687 .xpo_secure_port = svc_sock_secure_port,
690 static struct svc_xprt_class svc_udp_class = {
692 .xcl_owner = THIS_MODULE,
693 .xcl_ops = &svc_udp_ops,
694 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
697 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
699 int err, level, optname, one = 1;
701 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
702 &svsk->sk_xprt, serv);
703 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
704 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
705 svsk->sk_sk->sk_write_space = svc_write_space;
707 /* initialise setting must have enough space to
708 * receive and respond to one request.
709 * svc_udp_recvfrom will re-adjust if necessary
711 svc_sock_setbufsize(svsk->sk_sock,
712 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
713 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
715 /* data might have come in before data_ready set up */
716 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
717 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
719 /* make sure we get destination address info */
720 switch (svsk->sk_sk->sk_family) {
723 optname = IP_PKTINFO;
727 optname = IPV6_RECVPKTINFO;
732 err = kernel_setsockopt(svsk->sk_sock, level, optname,
733 (char *)&one, sizeof(one));
734 dprintk("svc: kernel_setsockopt returned %d\n", err);
738 * A data_ready event on a listening socket means there's a connection
739 * pending. Do not use state_change as a substitute for it.
741 static void svc_tcp_listen_data_ready(struct sock *sk)
743 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
744 wait_queue_head_t *wq;
746 dprintk("svc: socket %p TCP (listen) state change %d\n",
750 * This callback may called twice when a new connection
751 * is established as a child socket inherits everything
752 * from a parent LISTEN socket.
753 * 1) data_ready method of the parent socket will be called
754 * when one of child sockets become ESTABLISHED.
755 * 2) data_ready method of the child socket may be called
756 * when it receives data before the socket is accepted.
757 * In case of 2, we should ignore it silently.
759 if (sk->sk_state == TCP_LISTEN) {
761 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
762 svc_xprt_enqueue(&svsk->sk_xprt);
764 printk("svc: socket %p: no user data\n", sk);
768 if (wq && waitqueue_active(wq))
769 wake_up_interruptible_all(wq);
773 * A state change on a connected socket means it's dying or dead.
775 static void svc_tcp_state_change(struct sock *sk)
777 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
778 wait_queue_head_t *wq = sk_sleep(sk);
780 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
781 sk, sk->sk_state, sk->sk_user_data);
784 printk("svc: socket %p: no user data\n", sk);
786 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
787 svc_xprt_enqueue(&svsk->sk_xprt);
789 if (wq && waitqueue_active(wq))
790 wake_up_interruptible_all(wq);
793 static void svc_tcp_data_ready(struct sock *sk)
795 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
796 wait_queue_head_t *wq = sk_sleep(sk);
798 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
799 sk, sk->sk_user_data);
801 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
802 svc_xprt_enqueue(&svsk->sk_xprt);
804 if (wq && waitqueue_active(wq))
805 wake_up_interruptible(wq);
809 * Accept a TCP connection
811 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
813 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
814 struct sockaddr_storage addr;
815 struct sockaddr *sin = (struct sockaddr *) &addr;
816 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
817 struct socket *sock = svsk->sk_sock;
818 struct socket *newsock;
819 struct svc_sock *newsvsk;
821 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
823 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
827 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
828 err = kernel_accept(sock, &newsock, O_NONBLOCK);
831 printk(KERN_WARNING "%s: no more sockets!\n",
833 else if (err != -EAGAIN)
834 net_warn_ratelimited("%s: accept failed (err %d)!\n",
835 serv->sv_name, -err);
838 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
840 err = kernel_getpeername(newsock, sin, &slen);
842 net_warn_ratelimited("%s: peername failed (err %d)!\n",
843 serv->sv_name, -err);
844 goto failed; /* aborted connection or whatever */
847 /* Ideally, we would want to reject connections from unauthorized
848 * hosts here, but when we get encryption, the IP of the host won't
849 * tell us anything. For now just warn about unpriv connections.
851 if (!svc_port_is_privileged(sin)) {
852 dprintk("%s: connect from unprivileged port: %s\n",
854 __svc_print_addr(sin, buf, sizeof(buf)));
856 dprintk("%s: connect from %s\n", serv->sv_name,
857 __svc_print_addr(sin, buf, sizeof(buf)));
859 /* make sure that a write doesn't block forever when
862 newsock->sk->sk_sndtimeo = HZ*30;
864 newsvsk = svc_setup_socket(serv, newsock,
865 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
868 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
869 err = kernel_getsockname(newsock, sin, &slen);
870 if (unlikely(err < 0)) {
871 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
872 slen = offsetof(struct sockaddr, sa_data);
874 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
876 if (sock_is_loopback(newsock->sk))
877 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
879 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
881 serv->sv_stats->nettcpconn++;
883 return &newsvsk->sk_xprt;
886 sock_release(newsock);
890 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
892 unsigned int i, len, npages;
894 if (svsk->sk_datalen == 0)
896 len = svsk->sk_datalen;
897 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
898 for (i = 0; i < npages; i++) {
899 if (rqstp->rq_pages[i] != NULL)
900 put_page(rqstp->rq_pages[i]);
901 BUG_ON(svsk->sk_pages[i] == NULL);
902 rqstp->rq_pages[i] = svsk->sk_pages[i];
903 svsk->sk_pages[i] = NULL;
905 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
909 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
911 unsigned int i, len, npages;
913 if (svsk->sk_datalen == 0)
915 len = svsk->sk_datalen;
916 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
917 for (i = 0; i < npages; i++) {
918 svsk->sk_pages[i] = rqstp->rq_pages[i];
919 rqstp->rq_pages[i] = NULL;
923 static void svc_tcp_clear_pages(struct svc_sock *svsk)
925 unsigned int i, len, npages;
927 if (svsk->sk_datalen == 0)
929 len = svsk->sk_datalen;
930 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
931 for (i = 0; i < npages; i++) {
932 if (svsk->sk_pages[i] == NULL) {
936 put_page(svsk->sk_pages[i]);
937 svsk->sk_pages[i] = NULL;
941 svsk->sk_datalen = 0;
945 * Receive fragment record header.
946 * If we haven't gotten the record length yet, get the next four bytes.
948 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
950 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
954 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
956 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
959 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
960 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
962 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
964 svsk->sk_tcplen += len;
967 dprintk("svc: short recvfrom while reading record "
968 "length (%d of %d)\n", len, want);
972 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
973 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
975 net_notice_ratelimited("RPC: fragment too large: %d\n",
976 svc_sock_reclen(svsk));
981 return svc_sock_reclen(svsk);
983 dprintk("RPC: TCP recv_record got %d\n", len);
986 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
990 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
992 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
993 struct rpc_rqst *req = NULL;
994 struct kvec *src, *dst;
995 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1003 req = xprt_lookup_rqst(bc_xprt, xid);
1007 "%s: Got unrecognized reply: "
1008 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1009 __func__, ntohl(calldir),
1014 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1016 * XXX!: cheating for now! Only copying HEAD.
1017 * But we know this is good enough for now (in fact, for any
1018 * callback reply in the forseeable future).
1020 dst = &req->rq_private_buf.head[0];
1021 src = &rqstp->rq_arg.head[0];
1022 if (dst->iov_len < src->iov_len)
1023 return -EAGAIN; /* whatever; just giving up. */
1024 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1025 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1026 rqstp->rq_arg.len = 0;
1030 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1036 vec[i].iov_base = page_address(pages[i]);
1037 vec[i].iov_len = PAGE_SIZE;
1044 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1046 /* If we have more data, signal svc_xprt_enqueue() to try again */
1047 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1048 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1049 dprintk("svc: TCP %s record (%d bytes)\n",
1050 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1051 svc_sock_reclen(svsk));
1052 svsk->sk_tcplen = 0;
1053 svsk->sk_reclen = 0;
1057 * Receive data from a TCP socket.
1059 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1061 struct svc_sock *svsk =
1062 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1063 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1066 unsigned int want, base;
1071 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1072 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1073 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1074 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1076 len = svc_tcp_recv_record(svsk, rqstp);
1080 base = svc_tcp_restore_pages(svsk, rqstp);
1081 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1083 vec = rqstp->rq_vec;
1085 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1086 svsk->sk_datalen + want);
1088 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1089 rqstp->rq_next_page = rqstp->rq_respages + 1;
1091 /* Now receive data */
1092 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1094 svsk->sk_tcplen += len;
1095 svsk->sk_datalen += len;
1097 if (len != want || !svc_sock_final_rec(svsk)) {
1098 svc_tcp_save_pages(svsk, rqstp);
1099 if (len < 0 && len != -EAGAIN)
1102 svc_tcp_fragment_received(svsk);
1104 dprintk("svc: incomplete TCP record (%d of %d)\n",
1105 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1106 svc_sock_reclen(svsk));
1110 if (svsk->sk_datalen < 8) {
1111 svsk->sk_datalen = 0;
1112 goto err_delete; /* client is nuts. */
1115 rqstp->rq_arg.len = svsk->sk_datalen;
1116 rqstp->rq_arg.page_base = 0;
1117 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1118 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1119 rqstp->rq_arg.page_len = 0;
1121 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1123 rqstp->rq_xprt_ctxt = NULL;
1124 rqstp->rq_prot = IPPROTO_TCP;
1125 rqstp->rq_local = !!test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags);
1127 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1130 len = receive_cb_reply(svsk, rqstp);
1132 /* Reset TCP read info */
1133 svsk->sk_datalen = 0;
1134 svc_tcp_fragment_received(svsk);
1139 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1141 serv->sv_stats->nettcpcnt++;
1143 return rqstp->rq_arg.len;
1148 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1151 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1152 svsk->sk_xprt.xpt_server->sv_name, -len);
1153 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1155 return 0; /* record not complete */
1159 * Send out data on TCP socket.
1161 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1163 struct xdr_buf *xbufp = &rqstp->rq_res;
1167 /* Set up the first element of the reply kvec.
1168 * Any other kvecs that may be in use have been taken
1169 * care of by the server implementation itself.
1171 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1172 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1174 sent = svc_sendto(rqstp, &rqstp->rq_res);
1175 if (sent != xbufp->len) {
1177 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1178 "- shutting down socket\n",
1179 rqstp->rq_xprt->xpt_server->sv_name,
1180 (sent<0)?"got error":"sent only",
1182 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1183 svc_xprt_enqueue(rqstp->rq_xprt);
1190 * Setup response header. TCP has a 4B record length field.
1192 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1194 struct kvec *resv = &rqstp->rq_res.head[0];
1196 /* tcp needs a space for the record length... */
1200 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1202 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1203 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1206 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1208 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1209 if (sk_stream_wspace(svsk->sk_sk) >= required ||
1210 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
1211 atomic_read(&xprt->xpt_reserved) == 0))
1213 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1217 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1219 struct sockaddr *sa, int salen,
1222 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1225 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1226 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1227 struct net *, struct sockaddr *,
1229 static void svc_bc_sock_free(struct svc_xprt *xprt);
1231 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1233 struct sockaddr *sa, int salen,
1236 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1239 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1243 static struct svc_xprt_ops svc_tcp_bc_ops = {
1244 .xpo_create = svc_bc_tcp_create,
1245 .xpo_detach = svc_bc_tcp_sock_detach,
1246 .xpo_free = svc_bc_sock_free,
1247 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1248 .xpo_secure_port = svc_sock_secure_port,
1251 static struct svc_xprt_class svc_tcp_bc_class = {
1252 .xcl_name = "tcp-bc",
1253 .xcl_owner = THIS_MODULE,
1254 .xcl_ops = &svc_tcp_bc_ops,
1255 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1258 static void svc_init_bc_xprt_sock(void)
1260 svc_reg_xprt_class(&svc_tcp_bc_class);
1263 static void svc_cleanup_bc_xprt_sock(void)
1265 svc_unreg_xprt_class(&svc_tcp_bc_class);
1267 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1268 static void svc_init_bc_xprt_sock(void)
1272 static void svc_cleanup_bc_xprt_sock(void)
1275 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1277 static struct svc_xprt_ops svc_tcp_ops = {
1278 .xpo_create = svc_tcp_create,
1279 .xpo_recvfrom = svc_tcp_recvfrom,
1280 .xpo_sendto = svc_tcp_sendto,
1281 .xpo_release_rqst = svc_release_skb,
1282 .xpo_detach = svc_tcp_sock_detach,
1283 .xpo_free = svc_sock_free,
1284 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1285 .xpo_has_wspace = svc_tcp_has_wspace,
1286 .xpo_accept = svc_tcp_accept,
1287 .xpo_secure_port = svc_sock_secure_port,
1290 static struct svc_xprt_class svc_tcp_class = {
1292 .xcl_owner = THIS_MODULE,
1293 .xcl_ops = &svc_tcp_ops,
1294 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1297 void svc_init_xprt_sock(void)
1299 svc_reg_xprt_class(&svc_tcp_class);
1300 svc_reg_xprt_class(&svc_udp_class);
1301 svc_init_bc_xprt_sock();
1304 void svc_cleanup_xprt_sock(void)
1306 svc_unreg_xprt_class(&svc_tcp_class);
1307 svc_unreg_xprt_class(&svc_udp_class);
1308 svc_cleanup_bc_xprt_sock();
1311 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1313 struct sock *sk = svsk->sk_sk;
1315 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1316 &svsk->sk_xprt, serv);
1317 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1318 if (sk->sk_state == TCP_LISTEN) {
1319 dprintk("setting up TCP socket for listening\n");
1320 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1321 sk->sk_data_ready = svc_tcp_listen_data_ready;
1322 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1324 dprintk("setting up TCP socket for reading\n");
1325 sk->sk_state_change = svc_tcp_state_change;
1326 sk->sk_data_ready = svc_tcp_data_ready;
1327 sk->sk_write_space = svc_tcp_write_space;
1329 svsk->sk_reclen = 0;
1330 svsk->sk_tcplen = 0;
1331 svsk->sk_datalen = 0;
1332 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1334 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1337 if (sk->sk_state != TCP_ESTABLISHED)
1338 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1342 void svc_sock_update_bufs(struct svc_serv *serv)
1345 * The number of server threads has changed. Update
1346 * rcvbuf and sndbuf accordingly on all sockets
1348 struct svc_sock *svsk;
1350 spin_lock_bh(&serv->sv_lock);
1351 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1352 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1353 spin_unlock_bh(&serv->sv_lock);
1355 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1358 * Initialize socket for RPC use and create svc_sock struct
1359 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1361 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1362 struct socket *sock,
1365 struct svc_sock *svsk;
1367 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1370 dprintk("svc: svc_setup_socket %p\n", sock);
1371 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1373 return ERR_PTR(-ENOMEM);
1377 /* Register socket with portmapper */
1379 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1381 ntohs(inet_sk(inet)->inet_sport));
1385 return ERR_PTR(err);
1388 inet->sk_user_data = svsk;
1389 svsk->sk_sock = sock;
1391 svsk->sk_ostate = inet->sk_state_change;
1392 svsk->sk_odata = inet->sk_data_ready;
1393 svsk->sk_owspace = inet->sk_write_space;
1395 /* Initialize the socket */
1396 if (sock->type == SOCK_DGRAM)
1397 svc_udp_init(svsk, serv);
1399 /* initialise setting must have enough space to
1400 * receive and respond to one request.
1402 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1403 4 * serv->sv_max_mesg);
1404 svc_tcp_init(svsk, serv);
1407 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1413 bool svc_alien_sock(struct net *net, int fd)
1416 struct socket *sock = sockfd_lookup(fd, &err);
1421 if (sock_net(sock->sk) != net)
1427 EXPORT_SYMBOL_GPL(svc_alien_sock);
1430 * svc_addsock - add a listener socket to an RPC service
1431 * @serv: pointer to RPC service to which to add a new listener
1432 * @fd: file descriptor of the new listener
1433 * @name_return: pointer to buffer to fill in with name of listener
1434 * @len: size of the buffer
1436 * Fills in socket name and returns positive length of name if successful.
1437 * Name is terminated with '\n'. On error, returns a negative errno
1440 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1444 struct socket *so = sockfd_lookup(fd, &err);
1445 struct svc_sock *svsk = NULL;
1446 struct sockaddr_storage addr;
1447 struct sockaddr *sin = (struct sockaddr *)&addr;
1452 err = -EAFNOSUPPORT;
1453 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1455 err = -EPROTONOSUPPORT;
1456 if (so->sk->sk_protocol != IPPROTO_TCP &&
1457 so->sk->sk_protocol != IPPROTO_UDP)
1460 if (so->state > SS_UNCONNECTED)
1463 if (!try_module_get(THIS_MODULE))
1465 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1467 module_put(THIS_MODULE);
1468 err = PTR_ERR(svsk);
1471 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1472 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1473 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1474 return svc_one_sock_name(svsk, name_return, len);
1479 EXPORT_SYMBOL_GPL(svc_addsock);
1482 * Create socket for RPC service.
1484 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1487 struct sockaddr *sin, int len,
1490 struct svc_sock *svsk;
1491 struct socket *sock;
1494 struct sockaddr_storage addr;
1495 struct sockaddr *newsin = (struct sockaddr *)&addr;
1499 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1501 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1502 serv->sv_program->pg_name, protocol,
1503 __svc_print_addr(sin, buf, sizeof(buf)));
1505 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1506 printk(KERN_WARNING "svc: only UDP and TCP "
1507 "sockets supported\n");
1508 return ERR_PTR(-EINVAL);
1511 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1512 switch (sin->sa_family) {
1520 return ERR_PTR(-EINVAL);
1523 error = __sock_create(net, family, type, protocol, &sock, 1);
1525 return ERR_PTR(error);
1527 svc_reclassify_socket(sock);
1530 * If this is an PF_INET6 listener, we want to avoid
1531 * getting requests from IPv4 remotes. Those should
1532 * be shunted to a PF_INET listener via rpcbind.
1535 if (family == PF_INET6)
1536 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1537 (char *)&val, sizeof(val));
1539 if (type == SOCK_STREAM)
1540 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1541 error = kernel_bind(sock, sin, len);
1546 error = kernel_getsockname(sock, newsin, &newlen);
1550 if (protocol == IPPROTO_TCP) {
1551 if ((error = kernel_listen(sock, 64)) < 0)
1555 svsk = svc_setup_socket(serv, sock, flags);
1557 error = PTR_ERR(svsk);
1560 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1561 return (struct svc_xprt *)svsk;
1563 dprintk("svc: svc_create_socket error = %d\n", -error);
1565 return ERR_PTR(error);
1569 * Detach the svc_sock from the socket so that no
1570 * more callbacks occur.
1572 static void svc_sock_detach(struct svc_xprt *xprt)
1574 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1575 struct sock *sk = svsk->sk_sk;
1576 wait_queue_head_t *wq;
1578 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1580 /* put back the old socket callbacks */
1581 sk->sk_state_change = svsk->sk_ostate;
1582 sk->sk_data_ready = svsk->sk_odata;
1583 sk->sk_write_space = svsk->sk_owspace;
1586 if (wq && waitqueue_active(wq))
1587 wake_up_interruptible(wq);
1591 * Disconnect the socket, and reset the callbacks
1593 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1595 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1597 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1599 svc_sock_detach(xprt);
1601 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1602 svc_tcp_clear_pages(svsk);
1603 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1608 * Free the svc_sock's socket resources and the svc_sock itself.
1610 static void svc_sock_free(struct svc_xprt *xprt)
1612 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1613 dprintk("svc: svc_sock_free(%p)\n", svsk);
1615 if (svsk->sk_sock->file)
1616 sockfd_put(svsk->sk_sock);
1618 sock_release(svsk->sk_sock);
1622 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1624 * Create a back channel svc_xprt which shares the fore channel socket.
1626 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1629 struct sockaddr *sin, int len,
1632 struct svc_sock *svsk;
1633 struct svc_xprt *xprt;
1635 if (protocol != IPPROTO_TCP) {
1636 printk(KERN_WARNING "svc: only TCP sockets"
1637 " supported on shared back channel\n");
1638 return ERR_PTR(-EINVAL);
1641 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1643 return ERR_PTR(-ENOMEM);
1645 xprt = &svsk->sk_xprt;
1646 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1648 serv->sv_bc_xprt = xprt;
1654 * Free a back channel svc_sock.
1656 static void svc_bc_sock_free(struct svc_xprt *xprt)
1659 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1661 #endif /* CONFIG_SUNRPC_BACKCHANNEL */