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>
47 #include <linux/sunrpc/types.h>
48 #include <linux/sunrpc/clnt.h>
49 #include <linux/sunrpc/xdr.h>
50 #include <linux/sunrpc/msg_prot.h>
51 #include <linux/sunrpc/svcsock.h>
52 #include <linux/sunrpc/stats.h>
53 #include <linux/sunrpc/xprt.h>
57 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
60 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
61 int *errp, int flags);
62 static void svc_udp_data_ready(struct sock *, int);
63 static int svc_udp_recvfrom(struct svc_rqst *);
64 static int svc_udp_sendto(struct svc_rqst *);
65 static void svc_sock_detach(struct svc_xprt *);
66 static void svc_tcp_sock_detach(struct svc_xprt *);
67 static void svc_sock_free(struct svc_xprt *);
69 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
70 struct net *, struct sockaddr *,
72 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
73 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
74 struct net *, struct sockaddr *,
76 static void svc_bc_sock_free(struct svc_xprt *xprt);
77 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
79 #ifdef CONFIG_DEBUG_LOCK_ALLOC
80 static struct lock_class_key svc_key[2];
81 static struct lock_class_key svc_slock_key[2];
83 static void svc_reclassify_socket(struct socket *sock)
85 struct sock *sk = sock->sk;
86 BUG_ON(sock_owned_by_user(sk));
87 switch (sk->sk_family) {
89 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
91 "sk_xprt.xpt_lock-AF_INET-NFSD",
96 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
98 "sk_xprt.xpt_lock-AF_INET6-NFSD",
107 static void svc_reclassify_socket(struct socket *sock)
113 * Release an skbuff after use
115 static void svc_release_skb(struct svc_rqst *rqstp)
117 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
120 struct svc_sock *svsk =
121 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
122 rqstp->rq_xprt_ctxt = NULL;
124 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
125 skb_free_datagram_locked(svsk->sk_sk, skb);
129 union svc_pktinfo_u {
130 struct in_pktinfo pkti;
131 struct in6_pktinfo pkti6;
133 #define SVC_PKTINFO_SPACE \
134 CMSG_SPACE(sizeof(union svc_pktinfo_u))
136 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
138 struct svc_sock *svsk =
139 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
140 switch (svsk->sk_sk->sk_family) {
142 struct in_pktinfo *pki = CMSG_DATA(cmh);
144 cmh->cmsg_level = SOL_IP;
145 cmh->cmsg_type = IP_PKTINFO;
146 pki->ipi_ifindex = 0;
147 pki->ipi_spec_dst.s_addr =
148 svc_daddr_in(rqstp)->sin_addr.s_addr;
149 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
154 struct in6_pktinfo *pki = CMSG_DATA(cmh);
155 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
157 cmh->cmsg_level = SOL_IPV6;
158 cmh->cmsg_type = IPV6_PKTINFO;
159 pki->ipi6_ifindex = daddr->sin6_scope_id;
160 pki->ipi6_addr = daddr->sin6_addr;
161 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
168 * send routine intended to be shared by the fore- and back-channel
170 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
171 struct page *headpage, unsigned long headoffset,
172 struct page *tailpage, unsigned long tailoffset)
176 struct page **ppage = xdr->pages;
177 size_t base = xdr->page_base;
178 unsigned int pglen = xdr->page_len;
179 unsigned int flags = MSG_MORE;
186 if (slen == xdr->head[0].iov_len)
188 len = kernel_sendpage(sock, headpage, headoffset,
189 xdr->head[0].iov_len, flags);
190 if (len != xdr->head[0].iov_len)
192 slen -= xdr->head[0].iov_len;
197 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
201 result = kernel_sendpage(sock, *ppage, base, size, flags);
208 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 if (xdr->tail[0].iov_len) {
215 result = kernel_sendpage(sock, tailpage, tailoffset,
216 xdr->tail[0].iov_len, 0);
227 * Generic sendto routine
229 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
231 struct svc_sock *svsk =
232 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
233 struct socket *sock = svsk->sk_sock;
236 long all[SVC_PKTINFO_SPACE / sizeof(long)];
238 struct cmsghdr *cmh = &buffer.hdr;
240 unsigned long tailoff;
241 unsigned long headoff;
242 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
244 if (rqstp->rq_prot == IPPROTO_UDP) {
245 struct msghdr msg = {
246 .msg_name = &rqstp->rq_addr,
247 .msg_namelen = rqstp->rq_addrlen,
249 .msg_controllen = sizeof(buffer),
250 .msg_flags = MSG_MORE,
253 svc_set_cmsg_data(rqstp, cmh);
255 if (sock_sendmsg(sock, &msg, 0) < 0)
259 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
261 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
262 rqstp->rq_respages[0], tailoff);
265 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
266 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
267 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
273 * Report socket names for nfsdfs
275 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
277 const struct sock *sk = svsk->sk_sk;
278 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
282 switch (sk->sk_family) {
284 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
286 &inet_sk(sk)->inet_rcv_saddr,
287 inet_sk(sk)->inet_num);
290 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
292 &inet6_sk(sk)->rcv_saddr,
293 inet_sk(sk)->inet_num);
296 len = snprintf(buf, remaining, "*unknown-%d*\n",
300 if (len >= remaining) {
302 return -ENAMETOOLONG;
308 * svc_sock_names - construct a list of listener names in a string
309 * @serv: pointer to RPC service
310 * @buf: pointer to a buffer to fill in with socket names
311 * @buflen: size of the buffer to be filled
312 * @toclose: pointer to '\0'-terminated C string containing the name
313 * of a listener to be closed
315 * Fills in @buf with a '\n'-separated list of names of listener
316 * sockets. If @toclose is not NULL, the socket named by @toclose
317 * is closed, and is not included in the output list.
319 * Returns positive length of the socket name string, or a negative
320 * errno value on error.
322 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
325 struct svc_sock *svsk, *closesk = NULL;
331 spin_lock_bh(&serv->sv_lock);
332 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
333 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
338 if (toclose && strcmp(toclose, buf + len) == 0) {
340 svc_xprt_get(&closesk->sk_xprt);
344 spin_unlock_bh(&serv->sv_lock);
347 /* Should unregister with portmap, but you cannot
348 * unregister just one protocol...
350 svc_close_xprt(&closesk->sk_xprt);
351 svc_xprt_put(&closesk->sk_xprt);
356 EXPORT_SYMBOL_GPL(svc_sock_names);
359 * Check input queue length
361 static int svc_recv_available(struct svc_sock *svsk)
363 struct socket *sock = svsk->sk_sock;
366 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
368 return (err >= 0)? avail : err;
372 * Generic recvfrom routine.
374 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
377 struct svc_sock *svsk =
378 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
379 struct msghdr msg = {
380 .msg_flags = MSG_DONTWAIT,
384 rqstp->rq_xprt_hlen = 0;
386 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
389 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
390 svsk, iov[0].iov_base, iov[0].iov_len, len);
394 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
395 struct kvec *iov, int nr,
396 int buflen, unsigned int base)
404 return svc_recvfrom(rqstp, iov, nr, buflen);
406 for (i = 0; i < nr; i++) {
407 if (iov[i].iov_len > base)
409 base -= iov[i].iov_len;
411 save_iovlen = iov[i].iov_len;
412 save_iovbase = iov[i].iov_base;
413 iov[i].iov_len -= base;
414 iov[i].iov_base += base;
415 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
416 iov[i].iov_len = save_iovlen;
417 iov[i].iov_base = save_iovbase;
422 * Set socket snd and rcv buffer lengths
424 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
429 oldfs = get_fs(); set_fs(KERNEL_DS);
430 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
431 (char*)&snd, sizeof(snd));
432 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
433 (char*)&rcv, sizeof(rcv));
435 /* sock_setsockopt limits use to sysctl_?mem_max,
436 * which isn't acceptable. Until that is made conditional
437 * on not having CAP_SYS_RESOURCE or similar, we go direct...
438 * DaveM said I could!
441 sock->sk->sk_sndbuf = snd * 2;
442 sock->sk->sk_rcvbuf = rcv * 2;
443 sock->sk->sk_write_space(sock->sk);
444 release_sock(sock->sk);
448 * INET callback when data has been received on the socket.
450 static void svc_udp_data_ready(struct sock *sk, int count)
452 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
453 wait_queue_head_t *wq = sk_sleep(sk);
456 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
458 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
459 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
460 svc_xprt_enqueue(&svsk->sk_xprt);
462 if (wq && waitqueue_active(wq))
463 wake_up_interruptible(wq);
467 * INET callback when space is newly available on the socket.
469 static void svc_write_space(struct sock *sk)
471 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
472 wait_queue_head_t *wq = sk_sleep(sk);
475 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
476 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
477 svc_xprt_enqueue(&svsk->sk_xprt);
480 if (wq && waitqueue_active(wq)) {
481 dprintk("RPC svc_write_space: someone sleeping on %p\n",
483 wake_up_interruptible(wq);
487 static void svc_tcp_write_space(struct sock *sk)
489 struct socket *sock = sk->sk_socket;
491 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
492 clear_bit(SOCK_NOSPACE, &sock->flags);
497 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
499 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
502 struct in_pktinfo *pki = CMSG_DATA(cmh);
503 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
505 if (cmh->cmsg_type != IP_PKTINFO)
508 daddr->sin_family = AF_INET;
509 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
514 * See net/ipv6/datagram.c : datagram_recv_ctl
516 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
519 struct in6_pktinfo *pki = CMSG_DATA(cmh);
520 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
522 if (cmh->cmsg_type != IPV6_PKTINFO)
525 daddr->sin6_family = AF_INET6;
526 daddr->sin6_addr = pki->ipi6_addr;
527 daddr->sin6_scope_id = pki->ipi6_ifindex;
532 * Copy the UDP datagram's destination address to the rqstp structure.
533 * The 'destination' address in this case is the address to which the
534 * peer sent the datagram, i.e. our local address. For multihomed
535 * hosts, this can change from msg to msg. Note that only the IP
536 * address changes, the port number should remain the same.
538 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
541 switch (cmh->cmsg_level) {
543 return svc_udp_get_dest_address4(rqstp, cmh);
545 return svc_udp_get_dest_address6(rqstp, cmh);
552 * Receive a datagram from a UDP socket.
554 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
556 struct svc_sock *svsk =
557 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
558 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
562 long all[SVC_PKTINFO_SPACE / sizeof(long)];
564 struct cmsghdr *cmh = &buffer.hdr;
565 struct msghdr msg = {
566 .msg_name = svc_addr(rqstp),
568 .msg_controllen = sizeof(buffer),
569 .msg_flags = MSG_DONTWAIT,
574 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
575 /* udp sockets need large rcvbuf as all pending
576 * requests are still in that buffer. sndbuf must
577 * also be large enough that there is enough space
578 * for one reply per thread. We count all threads
579 * rather than threads in a particular pool, which
580 * provides an upper bound on the number of threads
581 * which will access the socket.
583 svc_sock_setbufsize(svsk->sk_sock,
584 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
585 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
587 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
589 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
590 0, 0, MSG_PEEK | MSG_DONTWAIT);
592 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
595 if (err != -EAGAIN) {
596 /* possibly an icmp error */
597 dprintk("svc: recvfrom returned error %d\n", -err);
598 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
602 len = svc_addr_len(svc_addr(rqstp));
604 return -EAFNOSUPPORT;
605 rqstp->rq_addrlen = len;
606 if (skb->tstamp.tv64 == 0) {
607 skb->tstamp = ktime_get_real();
608 /* Don't enable netstamp, sunrpc doesn't
609 need that much accuracy */
611 svsk->sk_sk->sk_stamp = skb->tstamp;
612 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
614 len = skb->len - sizeof(struct udphdr);
615 rqstp->rq_arg.len = len;
617 rqstp->rq_prot = IPPROTO_UDP;
619 if (!svc_udp_get_dest_address(rqstp, cmh)) {
620 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
621 cmh->cmsg_level, cmh->cmsg_type);
622 skb_free_datagram_locked(svsk->sk_sk, skb);
625 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
627 if (skb_is_nonlinear(skb)) {
628 /* we have to copy */
630 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
633 skb_free_datagram_locked(svsk->sk_sk, skb);
637 skb_free_datagram_locked(svsk->sk_sk, skb);
639 /* we can use it in-place */
640 rqstp->rq_arg.head[0].iov_base = skb->data +
641 sizeof(struct udphdr);
642 rqstp->rq_arg.head[0].iov_len = len;
643 if (skb_checksum_complete(skb)) {
644 skb_free_datagram_locked(svsk->sk_sk, skb);
647 rqstp->rq_xprt_ctxt = skb;
650 rqstp->rq_arg.page_base = 0;
651 if (len <= rqstp->rq_arg.head[0].iov_len) {
652 rqstp->rq_arg.head[0].iov_len = len;
653 rqstp->rq_arg.page_len = 0;
654 rqstp->rq_respages = rqstp->rq_pages+1;
656 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
657 rqstp->rq_respages = rqstp->rq_pages + 1 +
658 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
662 serv->sv_stats->netudpcnt++;
668 svc_udp_sendto(struct svc_rqst *rqstp)
672 error = svc_sendto(rqstp, &rqstp->rq_res);
673 if (error == -ECONNREFUSED)
674 /* ICMP error on earlier request. */
675 error = svc_sendto(rqstp, &rqstp->rq_res);
680 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
684 static int svc_udp_has_wspace(struct svc_xprt *xprt)
686 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
687 struct svc_serv *serv = xprt->xpt_server;
688 unsigned long required;
691 * Set the SOCK_NOSPACE flag before checking the available
694 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
695 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
696 if (required*2 > sock_wspace(svsk->sk_sk))
698 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
702 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
708 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
710 struct sockaddr *sa, int salen,
713 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
716 static struct svc_xprt_ops svc_udp_ops = {
717 .xpo_create = svc_udp_create,
718 .xpo_recvfrom = svc_udp_recvfrom,
719 .xpo_sendto = svc_udp_sendto,
720 .xpo_release_rqst = svc_release_skb,
721 .xpo_detach = svc_sock_detach,
722 .xpo_free = svc_sock_free,
723 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
724 .xpo_has_wspace = svc_udp_has_wspace,
725 .xpo_accept = svc_udp_accept,
728 static struct svc_xprt_class svc_udp_class = {
730 .xcl_owner = THIS_MODULE,
731 .xcl_ops = &svc_udp_ops,
732 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
735 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
737 int err, level, optname, one = 1;
739 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
740 &svsk->sk_xprt, serv);
741 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
742 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
743 svsk->sk_sk->sk_write_space = svc_write_space;
745 /* initialise setting must have enough space to
746 * receive and respond to one request.
747 * svc_udp_recvfrom will re-adjust if necessary
749 svc_sock_setbufsize(svsk->sk_sock,
750 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
751 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
753 /* data might have come in before data_ready set up */
754 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
755 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
757 /* make sure we get destination address info */
758 switch (svsk->sk_sk->sk_family) {
761 optname = IP_PKTINFO;
765 optname = IPV6_RECVPKTINFO;
770 err = kernel_setsockopt(svsk->sk_sock, level, optname,
771 (char *)&one, sizeof(one));
772 dprintk("svc: kernel_setsockopt returned %d\n", err);
776 * A data_ready event on a listening socket means there's a connection
777 * pending. Do not use state_change as a substitute for it.
779 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
781 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
782 wait_queue_head_t *wq;
784 dprintk("svc: socket %p TCP (listen) state change %d\n",
788 * This callback may called twice when a new connection
789 * is established as a child socket inherits everything
790 * from a parent LISTEN socket.
791 * 1) data_ready method of the parent socket will be called
792 * when one of child sockets become ESTABLISHED.
793 * 2) data_ready method of the child socket may be called
794 * when it receives data before the socket is accepted.
795 * In case of 2, we should ignore it silently.
797 if (sk->sk_state == TCP_LISTEN) {
799 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
800 svc_xprt_enqueue(&svsk->sk_xprt);
802 printk("svc: socket %p: no user data\n", sk);
806 if (wq && waitqueue_active(wq))
807 wake_up_interruptible_all(wq);
811 * A state change on a connected socket means it's dying or dead.
813 static void svc_tcp_state_change(struct sock *sk)
815 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
816 wait_queue_head_t *wq = sk_sleep(sk);
818 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
819 sk, sk->sk_state, sk->sk_user_data);
822 printk("svc: socket %p: no user data\n", sk);
824 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
825 svc_xprt_enqueue(&svsk->sk_xprt);
827 if (wq && waitqueue_active(wq))
828 wake_up_interruptible_all(wq);
831 static void svc_tcp_data_ready(struct sock *sk, int count)
833 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
834 wait_queue_head_t *wq = sk_sleep(sk);
836 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
837 sk, sk->sk_user_data);
839 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
840 svc_xprt_enqueue(&svsk->sk_xprt);
842 if (wq && waitqueue_active(wq))
843 wake_up_interruptible(wq);
847 * Accept a TCP connection
849 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
851 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
852 struct sockaddr_storage addr;
853 struct sockaddr *sin = (struct sockaddr *) &addr;
854 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
855 struct socket *sock = svsk->sk_sock;
856 struct socket *newsock;
857 struct svc_sock *newsvsk;
859 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
861 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
865 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
866 err = kernel_accept(sock, &newsock, O_NONBLOCK);
869 printk(KERN_WARNING "%s: no more sockets!\n",
871 else if (err != -EAGAIN)
872 net_warn_ratelimited("%s: accept failed (err %d)!\n",
873 serv->sv_name, -err);
876 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
878 err = kernel_getpeername(newsock, sin, &slen);
880 net_warn_ratelimited("%s: peername failed (err %d)!\n",
881 serv->sv_name, -err);
882 goto failed; /* aborted connection or whatever */
885 /* Ideally, we would want to reject connections from unauthorized
886 * hosts here, but when we get encryption, the IP of the host won't
887 * tell us anything. For now just warn about unpriv connections.
889 if (!svc_port_is_privileged(sin)) {
891 "%s: connect from unprivileged port: %s\n",
893 __svc_print_addr(sin, buf, sizeof(buf)));
895 dprintk("%s: connect from %s\n", serv->sv_name,
896 __svc_print_addr(sin, buf, sizeof(buf)));
898 /* make sure that a write doesn't block forever when
901 newsock->sk->sk_sndtimeo = HZ*30;
903 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
904 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
906 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
907 err = kernel_getsockname(newsock, sin, &slen);
908 if (unlikely(err < 0)) {
909 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
910 slen = offsetof(struct sockaddr, sa_data);
912 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
915 serv->sv_stats->nettcpconn++;
917 return &newsvsk->sk_xprt;
920 sock_release(newsock);
924 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
926 unsigned int i, len, npages;
928 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
930 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
931 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
932 for (i = 0; i < npages; i++) {
933 if (rqstp->rq_pages[i] != NULL)
934 put_page(rqstp->rq_pages[i]);
935 BUG_ON(svsk->sk_pages[i] == NULL);
936 rqstp->rq_pages[i] = svsk->sk_pages[i];
937 svsk->sk_pages[i] = NULL;
939 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
943 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
945 unsigned int i, len, npages;
947 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
949 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
950 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 for (i = 0; i < npages; i++) {
952 svsk->sk_pages[i] = rqstp->rq_pages[i];
953 rqstp->rq_pages[i] = NULL;
957 static void svc_tcp_clear_pages(struct svc_sock *svsk)
959 unsigned int i, len, npages;
961 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
963 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
964 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
965 for (i = 0; i < npages; i++) {
966 BUG_ON(svsk->sk_pages[i] == NULL);
967 put_page(svsk->sk_pages[i]);
968 svsk->sk_pages[i] = NULL;
976 * If we haven't gotten the record length yet, get the next four bytes.
977 * Otherwise try to gobble up as much as possible up to the complete
980 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
982 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
986 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
988 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
991 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
992 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
994 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
996 svsk->sk_tcplen += len;
999 dprintk("svc: short recvfrom while reading record "
1000 "length (%d of %d)\n", len, want);
1004 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1005 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1006 /* FIXME: technically, a record can be fragmented,
1007 * and non-terminal fragments will not have the top
1008 * bit set in the fragment length header.
1009 * But apparently no known nfs clients send fragmented
1011 net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
1015 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1016 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1017 if (svsk->sk_reclen > serv->sv_max_mesg) {
1018 net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
1019 (unsigned long)svsk->sk_reclen);
1024 if (svsk->sk_reclen < 8)
1025 goto err_delete; /* client is nuts. */
1027 len = svsk->sk_reclen;
1031 dprintk("RPC: TCP recv_record got %d\n", len);
1034 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1038 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1040 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1041 struct rpc_rqst *req = NULL;
1042 struct kvec *src, *dst;
1043 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1051 req = xprt_lookup_rqst(bc_xprt, xid);
1055 "%s: Got unrecognized reply: "
1056 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1057 __func__, ntohl(calldir),
1062 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1064 * XXX!: cheating for now! Only copying HEAD.
1065 * But we know this is good enough for now (in fact, for any
1066 * callback reply in the forseeable future).
1068 dst = &req->rq_private_buf.head[0];
1069 src = &rqstp->rq_arg.head[0];
1070 if (dst->iov_len < src->iov_len)
1071 return -EAGAIN; /* whatever; just giving up. */
1072 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1073 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1074 rqstp->rq_arg.len = 0;
1078 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1084 vec[i].iov_base = page_address(pages[i]);
1085 vec[i].iov_len = PAGE_SIZE;
1094 * Receive data from a TCP socket.
1096 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1098 struct svc_sock *svsk =
1099 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1100 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1103 unsigned int want, base;
1108 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1109 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1110 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1111 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1113 len = svc_tcp_recv_record(svsk, rqstp);
1117 base = svc_tcp_restore_pages(svsk, rqstp);
1118 want = svsk->sk_reclen - base;
1120 vec = rqstp->rq_vec;
1122 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1125 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1127 /* Now receive data */
1128 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1130 svsk->sk_tcplen += len;
1132 if (len < 0 && len != -EAGAIN)
1134 svc_tcp_save_pages(svsk, rqstp);
1135 dprintk("svc: incomplete TCP record (%d of %d)\n",
1136 svsk->sk_tcplen, svsk->sk_reclen);
1140 rqstp->rq_arg.len = svsk->sk_reclen;
1141 rqstp->rq_arg.page_base = 0;
1142 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1143 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1144 rqstp->rq_arg.page_len = 0;
1146 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1148 rqstp->rq_xprt_ctxt = NULL;
1149 rqstp->rq_prot = IPPROTO_TCP;
1151 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1154 len = receive_cb_reply(svsk, rqstp);
1156 /* Reset TCP read info */
1157 svsk->sk_reclen = 0;
1158 svsk->sk_tcplen = 0;
1159 /* If we have more data, signal svc_xprt_enqueue() to try again */
1160 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1161 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1166 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1168 serv->sv_stats->nettcpcnt++;
1170 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1171 return rqstp->rq_arg.len;
1176 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1179 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1180 svsk->sk_xprt.xpt_server->sv_name, -len);
1181 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1183 return -EAGAIN; /* record not complete */
1187 * Send out data on TCP socket.
1189 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1191 struct xdr_buf *xbufp = &rqstp->rq_res;
1195 /* Set up the first element of the reply kvec.
1196 * Any other kvecs that may be in use have been taken
1197 * care of by the server implementation itself.
1199 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1200 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1202 sent = svc_sendto(rqstp, &rqstp->rq_res);
1203 if (sent != xbufp->len) {
1205 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1206 "- shutting down socket\n",
1207 rqstp->rq_xprt->xpt_server->sv_name,
1208 (sent<0)?"got error":"sent only",
1210 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1211 svc_xprt_enqueue(rqstp->rq_xprt);
1218 * Setup response header. TCP has a 4B record length field.
1220 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1222 struct kvec *resv = &rqstp->rq_res.head[0];
1224 /* tcp needs a space for the record length... */
1228 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1230 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1231 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1234 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1236 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1237 if (sk_stream_wspace(svsk->sk_sk) >= required)
1239 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1243 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1245 struct sockaddr *sa, int salen,
1248 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1251 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1252 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1253 struct net *, struct sockaddr *,
1255 static void svc_bc_sock_free(struct svc_xprt *xprt);
1257 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1259 struct sockaddr *sa, int salen,
1262 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1265 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1269 static struct svc_xprt_ops svc_tcp_bc_ops = {
1270 .xpo_create = svc_bc_tcp_create,
1271 .xpo_detach = svc_bc_tcp_sock_detach,
1272 .xpo_free = svc_bc_sock_free,
1273 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1276 static struct svc_xprt_class svc_tcp_bc_class = {
1277 .xcl_name = "tcp-bc",
1278 .xcl_owner = THIS_MODULE,
1279 .xcl_ops = &svc_tcp_bc_ops,
1280 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1283 static void svc_init_bc_xprt_sock(void)
1285 svc_reg_xprt_class(&svc_tcp_bc_class);
1288 static void svc_cleanup_bc_xprt_sock(void)
1290 svc_unreg_xprt_class(&svc_tcp_bc_class);
1292 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1293 static void svc_init_bc_xprt_sock(void)
1297 static void svc_cleanup_bc_xprt_sock(void)
1300 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1302 static struct svc_xprt_ops svc_tcp_ops = {
1303 .xpo_create = svc_tcp_create,
1304 .xpo_recvfrom = svc_tcp_recvfrom,
1305 .xpo_sendto = svc_tcp_sendto,
1306 .xpo_release_rqst = svc_release_skb,
1307 .xpo_detach = svc_tcp_sock_detach,
1308 .xpo_free = svc_sock_free,
1309 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1310 .xpo_has_wspace = svc_tcp_has_wspace,
1311 .xpo_accept = svc_tcp_accept,
1314 static struct svc_xprt_class svc_tcp_class = {
1316 .xcl_owner = THIS_MODULE,
1317 .xcl_ops = &svc_tcp_ops,
1318 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1321 void svc_init_xprt_sock(void)
1323 svc_reg_xprt_class(&svc_tcp_class);
1324 svc_reg_xprt_class(&svc_udp_class);
1325 svc_init_bc_xprt_sock();
1328 void svc_cleanup_xprt_sock(void)
1330 svc_unreg_xprt_class(&svc_tcp_class);
1331 svc_unreg_xprt_class(&svc_udp_class);
1332 svc_cleanup_bc_xprt_sock();
1335 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1337 struct sock *sk = svsk->sk_sk;
1339 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1340 &svsk->sk_xprt, serv);
1341 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1342 if (sk->sk_state == TCP_LISTEN) {
1343 dprintk("setting up TCP socket for listening\n");
1344 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1345 sk->sk_data_ready = svc_tcp_listen_data_ready;
1346 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1348 dprintk("setting up TCP socket for reading\n");
1349 sk->sk_state_change = svc_tcp_state_change;
1350 sk->sk_data_ready = svc_tcp_data_ready;
1351 sk->sk_write_space = svc_tcp_write_space;
1353 svsk->sk_reclen = 0;
1354 svsk->sk_tcplen = 0;
1355 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1357 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1359 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1360 if (sk->sk_state != TCP_ESTABLISHED)
1361 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1365 void svc_sock_update_bufs(struct svc_serv *serv)
1368 * The number of server threads has changed. Update
1369 * rcvbuf and sndbuf accordingly on all sockets
1371 struct svc_sock *svsk;
1373 spin_lock_bh(&serv->sv_lock);
1374 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1375 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1376 spin_unlock_bh(&serv->sv_lock);
1378 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1381 * Initialize socket for RPC use and create svc_sock struct
1382 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1384 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1385 struct socket *sock,
1386 int *errp, int flags)
1388 struct svc_sock *svsk;
1390 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1392 dprintk("svc: svc_setup_socket %p\n", sock);
1393 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1400 /* Register socket with portmapper */
1401 if (*errp >= 0 && pmap_register)
1402 *errp = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1404 ntohs(inet_sk(inet)->inet_sport));
1411 inet->sk_user_data = svsk;
1412 svsk->sk_sock = sock;
1414 svsk->sk_ostate = inet->sk_state_change;
1415 svsk->sk_odata = inet->sk_data_ready;
1416 svsk->sk_owspace = inet->sk_write_space;
1418 /* Initialize the socket */
1419 if (sock->type == SOCK_DGRAM)
1420 svc_udp_init(svsk, serv);
1422 /* initialise setting must have enough space to
1423 * receive and respond to one request.
1425 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1426 4 * serv->sv_max_mesg);
1427 svc_tcp_init(svsk, serv);
1430 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1437 * svc_addsock - add a listener socket to an RPC service
1438 * @serv: pointer to RPC service to which to add a new listener
1439 * @fd: file descriptor of the new listener
1440 * @name_return: pointer to buffer to fill in with name of listener
1441 * @len: size of the buffer
1443 * Fills in socket name and returns positive length of name if successful.
1444 * Name is terminated with '\n'. On error, returns a negative errno
1447 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1451 struct socket *so = sockfd_lookup(fd, &err);
1452 struct svc_sock *svsk = NULL;
1456 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1457 err = -EAFNOSUPPORT;
1458 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1459 so->sk->sk_protocol != IPPROTO_UDP)
1460 err = -EPROTONOSUPPORT;
1461 else if (so->state > SS_UNCONNECTED)
1464 if (!try_module_get(THIS_MODULE))
1467 svsk = svc_setup_socket(serv, so, &err,
1470 struct sockaddr_storage addr;
1471 struct sockaddr *sin = (struct sockaddr *)&addr;
1473 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1474 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1475 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1476 spin_lock_bh(&serv->sv_lock);
1477 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1478 spin_unlock_bh(&serv->sv_lock);
1479 svc_xprt_received(&svsk->sk_xprt);
1482 module_put(THIS_MODULE);
1488 return svc_one_sock_name(svsk, name_return, len);
1490 EXPORT_SYMBOL_GPL(svc_addsock);
1493 * Create socket for RPC service.
1495 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1498 struct sockaddr *sin, int len,
1501 struct svc_sock *svsk;
1502 struct socket *sock;
1505 struct sockaddr_storage addr;
1506 struct sockaddr *newsin = (struct sockaddr *)&addr;
1510 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1512 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1513 serv->sv_program->pg_name, protocol,
1514 __svc_print_addr(sin, buf, sizeof(buf)));
1516 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1517 printk(KERN_WARNING "svc: only UDP and TCP "
1518 "sockets supported\n");
1519 return ERR_PTR(-EINVAL);
1522 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1523 switch (sin->sa_family) {
1531 return ERR_PTR(-EINVAL);
1534 error = __sock_create(net, family, type, protocol, &sock, 1);
1536 return ERR_PTR(error);
1538 svc_reclassify_socket(sock);
1541 * If this is an PF_INET6 listener, we want to avoid
1542 * getting requests from IPv4 remotes. Those should
1543 * be shunted to a PF_INET listener via rpcbind.
1546 if (family == PF_INET6)
1547 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1548 (char *)&val, sizeof(val));
1550 if (type == SOCK_STREAM)
1551 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1552 error = kernel_bind(sock, sin, len);
1557 error = kernel_getsockname(sock, newsin, &newlen);
1561 if (protocol == IPPROTO_TCP) {
1562 if ((error = kernel_listen(sock, 64)) < 0)
1566 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1567 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1568 return (struct svc_xprt *)svsk;
1572 dprintk("svc: svc_create_socket error = %d\n", -error);
1574 return ERR_PTR(error);
1578 * Detach the svc_sock from the socket so that no
1579 * more callbacks occur.
1581 static void svc_sock_detach(struct svc_xprt *xprt)
1583 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1584 struct sock *sk = svsk->sk_sk;
1585 wait_queue_head_t *wq;
1587 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1589 /* put back the old socket callbacks */
1590 sk->sk_state_change = svsk->sk_ostate;
1591 sk->sk_data_ready = svsk->sk_odata;
1592 sk->sk_write_space = svsk->sk_owspace;
1595 if (wq && waitqueue_active(wq))
1596 wake_up_interruptible(wq);
1600 * Disconnect the socket, and reset the callbacks
1602 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1604 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1606 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1608 svc_sock_detach(xprt);
1610 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1611 svc_tcp_clear_pages(svsk);
1612 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1617 * Free the svc_sock's socket resources and the svc_sock itself.
1619 static void svc_sock_free(struct svc_xprt *xprt)
1621 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1622 dprintk("svc: svc_sock_free(%p)\n", svsk);
1624 if (svsk->sk_sock->file)
1625 sockfd_put(svsk->sk_sock);
1627 sock_release(svsk->sk_sock);
1631 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1633 * Create a back channel svc_xprt which shares the fore channel socket.
1635 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1638 struct sockaddr *sin, int len,
1641 struct svc_sock *svsk;
1642 struct svc_xprt *xprt;
1644 if (protocol != IPPROTO_TCP) {
1645 printk(KERN_WARNING "svc: only TCP sockets"
1646 " supported on shared back channel\n");
1647 return ERR_PTR(-EINVAL);
1650 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1652 return ERR_PTR(-ENOMEM);
1654 xprt = &svsk->sk_xprt;
1655 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1657 serv->sv_bc_xprt = xprt;
1663 * Free a back channel svc_sock.
1665 static void svc_bc_sock_free(struct svc_xprt *xprt)
1668 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1670 #endif /* CONFIG_SUNRPC_BACKCHANNEL */