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>
44 #include <net/tcp_states.h>
45 #include <linux/uaccess.h>
46 #include <asm/ioctls.h>
47 #include <trace/events/skb.h>
49 #include <linux/sunrpc/types.h>
50 #include <linux/sunrpc/clnt.h>
51 #include <linux/sunrpc/xdr.h>
52 #include <linux/sunrpc/msg_prot.h>
53 #include <linux/sunrpc/svcsock.h>
54 #include <linux/sunrpc/stats.h>
55 #include <linux/sunrpc/xprt.h>
59 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
62 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
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 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
91 switch (sk->sk_family) {
93 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
95 "sk_xprt.xpt_lock-AF_INET-NFSD",
100 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
102 "sk_xprt.xpt_lock-AF_INET6-NFSD",
111 static void svc_reclassify_socket(struct socket *sock)
117 * Release an skbuff after use
119 static void svc_release_skb(struct svc_rqst *rqstp)
121 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
124 struct svc_sock *svsk =
125 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
126 rqstp->rq_xprt_ctxt = NULL;
128 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
129 skb_free_datagram_locked(svsk->sk_sk, skb);
133 static void svc_release_udp_skb(struct svc_rqst *rqstp)
135 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
138 rqstp->rq_xprt_ctxt = NULL;
140 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
145 union svc_pktinfo_u {
146 struct in_pktinfo pkti;
147 struct in6_pktinfo pkti6;
149 #define SVC_PKTINFO_SPACE \
150 CMSG_SPACE(sizeof(union svc_pktinfo_u))
152 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
154 struct svc_sock *svsk =
155 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
156 switch (svsk->sk_sk->sk_family) {
158 struct in_pktinfo *pki = CMSG_DATA(cmh);
160 cmh->cmsg_level = SOL_IP;
161 cmh->cmsg_type = IP_PKTINFO;
162 pki->ipi_ifindex = 0;
163 pki->ipi_spec_dst.s_addr =
164 svc_daddr_in(rqstp)->sin_addr.s_addr;
165 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
170 struct in6_pktinfo *pki = CMSG_DATA(cmh);
171 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
173 cmh->cmsg_level = SOL_IPV6;
174 cmh->cmsg_type = IPV6_PKTINFO;
175 pki->ipi6_ifindex = daddr->sin6_scope_id;
176 pki->ipi6_addr = daddr->sin6_addr;
177 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
184 * send routine intended to be shared by the fore- and back-channel
186 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
187 struct page *headpage, unsigned long headoffset,
188 struct page *tailpage, unsigned long tailoffset)
192 struct page **ppage = xdr->pages;
193 size_t base = xdr->page_base;
194 unsigned int pglen = xdr->page_len;
195 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
202 if (slen == xdr->head[0].iov_len)
204 len = kernel_sendpage(sock, headpage, headoffset,
205 xdr->head[0].iov_len, flags);
206 if (len != xdr->head[0].iov_len)
208 slen -= xdr->head[0].iov_len;
213 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
217 result = kernel_sendpage(sock, *ppage, base, size, flags);
224 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
230 if (xdr->tail[0].iov_len) {
231 result = kernel_sendpage(sock, tailpage, tailoffset,
232 xdr->tail[0].iov_len, 0);
243 * Generic sendto routine
245 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
247 struct svc_sock *svsk =
248 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
249 struct socket *sock = svsk->sk_sock;
252 long all[SVC_PKTINFO_SPACE / sizeof(long)];
254 struct cmsghdr *cmh = &buffer.hdr;
256 unsigned long tailoff;
257 unsigned long headoff;
258 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
260 if (rqstp->rq_prot == IPPROTO_UDP) {
261 struct msghdr msg = {
262 .msg_name = &rqstp->rq_addr,
263 .msg_namelen = rqstp->rq_addrlen,
265 .msg_controllen = sizeof(buffer),
266 .msg_flags = MSG_MORE,
269 svc_set_cmsg_data(rqstp, cmh);
271 if (sock_sendmsg(sock, &msg) < 0)
275 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
277 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
278 rqstp->rq_respages[0], tailoff);
281 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
282 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
283 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
289 * Report socket names for nfsdfs
291 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
293 const struct sock *sk = svsk->sk_sk;
294 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
298 switch (sk->sk_family) {
300 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
302 &inet_sk(sk)->inet_rcv_saddr,
303 inet_sk(sk)->inet_num);
305 #if IS_ENABLED(CONFIG_IPV6)
307 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
309 &sk->sk_v6_rcv_saddr,
310 inet_sk(sk)->inet_num);
314 len = snprintf(buf, remaining, "*unknown-%d*\n",
318 if (len >= remaining) {
320 return -ENAMETOOLONG;
326 * Generic recvfrom routine.
328 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = {
334 .msg_flags = MSG_DONTWAIT,
338 rqstp->rq_xprt_hlen = 0;
340 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
341 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, iov, nr, buflen);
342 len = sock_recvmsg(svsk->sk_sock, &msg, msg.msg_flags);
343 /* If we read a full record, then assume there may be more
344 * data to read (stream based sockets only!)
347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
349 dprintk("svc: socket %p recvfrom(%p, %zu) = %d\n",
350 svsk, iov[0].iov_base, iov[0].iov_len, len);
354 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
355 struct kvec *iov, int nr,
356 int buflen, unsigned int base)
364 return svc_recvfrom(rqstp, iov, nr, buflen);
366 for (i = 0; i < nr; i++) {
367 if (iov[i].iov_len > base)
369 base -= iov[i].iov_len;
371 save_iovlen = iov[i].iov_len;
372 save_iovbase = iov[i].iov_base;
373 iov[i].iov_len -= base;
374 iov[i].iov_base += base;
375 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
376 iov[i].iov_len = save_iovlen;
377 iov[i].iov_base = save_iovbase;
382 * Set socket snd and rcv buffer lengths
384 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
386 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
387 struct socket *sock = svsk->sk_sock;
389 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
392 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
393 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
394 sock->sk->sk_write_space(sock->sk);
395 release_sock(sock->sk);
398 static void svc_sock_secure_port(struct svc_rqst *rqstp)
400 if (svc_port_is_privileged(svc_addr(rqstp)))
401 set_bit(RQ_SECURE, &rqstp->rq_flags);
403 clear_bit(RQ_SECURE, &rqstp->rq_flags);
407 * INET callback when data has been received on the socket.
409 static void svc_data_ready(struct sock *sk)
411 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
414 dprintk("svc: socket %p(inet %p), busy=%d\n",
416 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
418 /* Refer to svc_setup_socket() for details. */
421 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
422 svc_xprt_enqueue(&svsk->sk_xprt);
427 * INET callback when space is newly available on the socket.
429 static void svc_write_space(struct sock *sk)
431 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
434 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
435 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
437 /* Refer to svc_setup_socket() for details. */
439 svsk->sk_owspace(sk);
440 svc_xprt_enqueue(&svsk->sk_xprt);
444 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
446 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
448 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
450 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
453 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
455 struct svc_sock *svsk;
457 struct linger no_linger = {
462 svsk = container_of(xprt, struct svc_sock, sk_xprt);
463 sock = svsk->sk_sock;
464 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
465 (char *)&no_linger, sizeof(no_linger));
469 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
471 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
474 struct in_pktinfo *pki = CMSG_DATA(cmh);
475 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
477 if (cmh->cmsg_type != IP_PKTINFO)
480 daddr->sin_family = AF_INET;
481 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
486 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
488 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
491 struct in6_pktinfo *pki = CMSG_DATA(cmh);
492 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
494 if (cmh->cmsg_type != IPV6_PKTINFO)
497 daddr->sin6_family = AF_INET6;
498 daddr->sin6_addr = pki->ipi6_addr;
499 daddr->sin6_scope_id = pki->ipi6_ifindex;
504 * Copy the UDP datagram's destination address to the rqstp structure.
505 * The 'destination' address in this case is the address to which the
506 * peer sent the datagram, i.e. our local address. For multihomed
507 * hosts, this can change from msg to msg. Note that only the IP
508 * address changes, the port number should remain the same.
510 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
513 switch (cmh->cmsg_level) {
515 return svc_udp_get_dest_address4(rqstp, cmh);
517 return svc_udp_get_dest_address6(rqstp, cmh);
524 * Receive a datagram from a UDP socket.
526 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
528 struct svc_sock *svsk =
529 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
530 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
534 long all[SVC_PKTINFO_SPACE / sizeof(long)];
536 struct cmsghdr *cmh = &buffer.hdr;
537 struct msghdr msg = {
538 .msg_name = svc_addr(rqstp),
540 .msg_controllen = sizeof(buffer),
541 .msg_flags = MSG_DONTWAIT,
546 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
547 /* udp sockets need large rcvbuf as all pending
548 * requests are still in that buffer. sndbuf must
549 * also be large enough that there is enough space
550 * for one reply per thread. We count all threads
551 * rather than threads in a particular pool, which
552 * provides an upper bound on the number of threads
553 * which will access the socket.
555 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
557 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
559 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
560 0, 0, MSG_PEEK | MSG_DONTWAIT);
562 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
565 if (err != -EAGAIN) {
566 /* possibly an icmp error */
567 dprintk("svc: recvfrom returned error %d\n", -err);
568 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
572 len = svc_addr_len(svc_addr(rqstp));
573 rqstp->rq_addrlen = len;
574 if (skb->tstamp == 0) {
575 skb->tstamp = ktime_get_real();
576 /* Don't enable netstamp, sunrpc doesn't
577 need that much accuracy */
579 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
580 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
583 rqstp->rq_arg.len = len;
585 rqstp->rq_prot = IPPROTO_UDP;
587 if (!svc_udp_get_dest_address(rqstp, cmh)) {
588 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
589 cmh->cmsg_level, cmh->cmsg_type);
592 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
594 if (skb_is_nonlinear(skb)) {
595 /* we have to copy */
597 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
605 /* we can use it in-place */
606 rqstp->rq_arg.head[0].iov_base = skb->data;
607 rqstp->rq_arg.head[0].iov_len = len;
608 if (skb_checksum_complete(skb))
610 rqstp->rq_xprt_ctxt = skb;
613 rqstp->rq_arg.page_base = 0;
614 if (len <= rqstp->rq_arg.head[0].iov_len) {
615 rqstp->rq_arg.head[0].iov_len = len;
616 rqstp->rq_arg.page_len = 0;
617 rqstp->rq_respages = rqstp->rq_pages+1;
619 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
620 rqstp->rq_respages = rqstp->rq_pages + 1 +
621 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
623 rqstp->rq_next_page = rqstp->rq_respages+1;
626 serv->sv_stats->netudpcnt++;
635 svc_udp_sendto(struct svc_rqst *rqstp)
639 svc_release_udp_skb(rqstp);
641 error = svc_sendto(rqstp, &rqstp->rq_res);
642 if (error == -ECONNREFUSED)
643 /* ICMP error on earlier request. */
644 error = svc_sendto(rqstp, &rqstp->rq_res);
649 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
653 static int svc_udp_has_wspace(struct svc_xprt *xprt)
655 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
656 struct svc_serv *serv = xprt->xpt_server;
657 unsigned long required;
660 * Set the SOCK_NOSPACE flag before checking the available
663 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
664 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
665 if (required*2 > sock_wspace(svsk->sk_sk))
667 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
671 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
677 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
681 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
683 struct sockaddr *sa, int salen,
686 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
689 static const struct svc_xprt_ops svc_udp_ops = {
690 .xpo_create = svc_udp_create,
691 .xpo_recvfrom = svc_udp_recvfrom,
692 .xpo_sendto = svc_udp_sendto,
693 .xpo_release_rqst = svc_release_udp_skb,
694 .xpo_detach = svc_sock_detach,
695 .xpo_free = svc_sock_free,
696 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
697 .xpo_has_wspace = svc_udp_has_wspace,
698 .xpo_accept = svc_udp_accept,
699 .xpo_secure_port = svc_sock_secure_port,
700 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
703 static struct svc_xprt_class svc_udp_class = {
705 .xcl_owner = THIS_MODULE,
706 .xcl_ops = &svc_udp_ops,
707 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
708 .xcl_ident = XPRT_TRANSPORT_UDP,
711 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
713 int err, level, optname, one = 1;
715 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
716 &svsk->sk_xprt, serv);
717 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
718 svsk->sk_sk->sk_data_ready = svc_data_ready;
719 svsk->sk_sk->sk_write_space = svc_write_space;
721 /* initialise setting must have enough space to
722 * receive and respond to one request.
723 * svc_udp_recvfrom will re-adjust if necessary
725 svc_sock_setbufsize(svsk, 3);
727 /* data might have come in before data_ready set up */
728 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
729 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
731 /* make sure we get destination address info */
732 switch (svsk->sk_sk->sk_family) {
735 optname = IP_PKTINFO;
739 optname = IPV6_RECVPKTINFO;
744 err = kernel_setsockopt(svsk->sk_sock, level, optname,
745 (char *)&one, sizeof(one));
746 dprintk("svc: kernel_setsockopt returned %d\n", err);
750 * A data_ready event on a listening socket means there's a connection
751 * pending. Do not use state_change as a substitute for it.
753 static void svc_tcp_listen_data_ready(struct sock *sk)
755 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
757 dprintk("svc: socket %p TCP (listen) state change %d\n",
761 * This callback may called twice when a new connection
762 * is established as a child socket inherits everything
763 * from a parent LISTEN socket.
764 * 1) data_ready method of the parent socket will be called
765 * when one of child sockets become ESTABLISHED.
766 * 2) data_ready method of the child socket may be called
767 * when it receives data before the socket is accepted.
768 * In case of 2, we should ignore it silently and DO NOT
771 if (sk->sk_state != TCP_LISTEN)
775 /* Refer to svc_setup_socket() for details. */
778 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
779 svc_xprt_enqueue(&svsk->sk_xprt);
781 printk("svc: socket %p: no user data\n", sk);
785 * A state change on a connected socket means it's dying or dead.
787 static void svc_tcp_state_change(struct sock *sk)
789 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
791 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
792 sk, sk->sk_state, sk->sk_user_data);
795 printk("svc: socket %p: no user data\n", sk);
797 /* Refer to svc_setup_socket() for details. */
800 if (sk->sk_state != TCP_ESTABLISHED) {
801 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
802 svc_xprt_enqueue(&svsk->sk_xprt);
808 * Accept a TCP connection
810 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
812 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
813 struct sockaddr_storage addr;
814 struct sockaddr *sin = (struct sockaddr *) &addr;
815 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
816 struct socket *sock = svsk->sk_sock;
817 struct socket *newsock;
818 struct svc_sock *newsvsk;
820 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
822 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
826 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
827 err = kernel_accept(sock, &newsock, O_NONBLOCK);
830 printk(KERN_WARNING "%s: no more sockets!\n",
832 else if (err != -EAGAIN)
833 net_warn_ratelimited("%s: accept failed (err %d)!\n",
834 serv->sv_name, -err);
837 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
839 err = kernel_getpeername(newsock, sin);
841 net_warn_ratelimited("%s: peername failed (err %d)!\n",
842 serv->sv_name, -err);
843 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 /* Reset the inherited callbacks before calling svc_setup_socket */
860 newsock->sk->sk_state_change = svsk->sk_ostate;
861 newsock->sk->sk_data_ready = svsk->sk_odata;
862 newsock->sk->sk_write_space = svsk->sk_owspace;
864 /* make sure that a write doesn't block forever when
867 newsock->sk->sk_sndtimeo = HZ*30;
869 newsvsk = svc_setup_socket(serv, newsock,
870 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
873 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
874 err = kernel_getsockname(newsock, sin);
876 if (unlikely(err < 0)) {
877 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
878 slen = offsetof(struct sockaddr, sa_data);
880 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
882 if (sock_is_loopback(newsock->sk))
883 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
885 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
887 serv->sv_stats->nettcpconn++;
889 return &newsvsk->sk_xprt;
892 sock_release(newsock);
896 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
898 unsigned int i, len, npages;
900 if (svsk->sk_datalen == 0)
902 len = svsk->sk_datalen;
903 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
904 for (i = 0; i < npages; i++) {
905 if (rqstp->rq_pages[i] != NULL)
906 put_page(rqstp->rq_pages[i]);
907 BUG_ON(svsk->sk_pages[i] == NULL);
908 rqstp->rq_pages[i] = svsk->sk_pages[i];
909 svsk->sk_pages[i] = NULL;
911 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
915 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
917 unsigned int i, len, npages;
919 if (svsk->sk_datalen == 0)
921 len = svsk->sk_datalen;
922 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
923 for (i = 0; i < npages; i++) {
924 svsk->sk_pages[i] = rqstp->rq_pages[i];
925 rqstp->rq_pages[i] = NULL;
929 static void svc_tcp_clear_pages(struct svc_sock *svsk)
931 unsigned int i, len, npages;
933 if (svsk->sk_datalen == 0)
935 len = svsk->sk_datalen;
936 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
937 for (i = 0; i < npages; i++) {
938 if (svsk->sk_pages[i] == NULL) {
942 put_page(svsk->sk_pages[i]);
943 svsk->sk_pages[i] = NULL;
947 svsk->sk_datalen = 0;
951 * Receive fragment record header.
952 * If we haven't gotten the record length yet, get the next four bytes.
954 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
956 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
960 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
963 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
964 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
966 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
968 svsk->sk_tcplen += len;
971 dprintk("svc: short recvfrom while reading record "
972 "length (%d of %d)\n", len, want);
976 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
977 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
979 net_notice_ratelimited("RPC: fragment too large: %d\n",
980 svc_sock_reclen(svsk));
985 return svc_sock_reclen(svsk);
987 dprintk("RPC: TCP recv_record got %d\n", len);
990 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
994 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
996 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
997 struct rpc_rqst *req = NULL;
998 struct kvec *src, *dst;
999 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1008 spin_lock(&bc_xprt->recv_lock);
1009 req = xprt_lookup_rqst(bc_xprt, xid);
1011 goto unlock_notfound;
1013 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1015 * XXX!: cheating for now! Only copying HEAD.
1016 * But we know this is good enough for now (in fact, for any
1017 * callback reply in the forseeable future).
1019 dst = &req->rq_private_buf.head[0];
1020 src = &rqstp->rq_arg.head[0];
1021 if (dst->iov_len < src->iov_len)
1022 goto unlock_eagain; /* whatever; just giving up. */
1023 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1024 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1025 rqstp->rq_arg.len = 0;
1026 spin_unlock(&bc_xprt->recv_lock);
1030 "%s: Got unrecognized reply: "
1031 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1032 __func__, ntohl(calldir),
1033 bc_xprt, ntohl(xid));
1035 spin_unlock(&bc_xprt->recv_lock);
1039 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1045 vec[i].iov_base = page_address(pages[i]);
1046 vec[i].iov_len = PAGE_SIZE;
1053 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1055 /* If we have more data, signal svc_xprt_enqueue() to try again */
1056 dprintk("svc: TCP %s record (%d bytes)\n",
1057 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1058 svc_sock_reclen(svsk));
1059 svsk->sk_tcplen = 0;
1060 svsk->sk_reclen = 0;
1064 * Receive data from a TCP socket.
1066 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1068 struct svc_sock *svsk =
1069 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1070 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1073 unsigned int want, base;
1078 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1079 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1080 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1081 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1083 len = svc_tcp_recv_record(svsk, rqstp);
1087 base = svc_tcp_restore_pages(svsk, rqstp);
1088 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1090 vec = rqstp->rq_vec;
1092 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1093 svsk->sk_datalen + want);
1095 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1096 rqstp->rq_next_page = rqstp->rq_respages + 1;
1098 /* Now receive data */
1099 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1101 svsk->sk_tcplen += len;
1102 svsk->sk_datalen += len;
1104 if (len != want || !svc_sock_final_rec(svsk)) {
1105 svc_tcp_save_pages(svsk, rqstp);
1106 if (len < 0 && len != -EAGAIN)
1109 svc_tcp_fragment_received(svsk);
1111 dprintk("svc: incomplete TCP record (%d of %d)\n",
1112 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1113 svc_sock_reclen(svsk));
1117 if (svsk->sk_datalen < 8) {
1118 svsk->sk_datalen = 0;
1119 goto err_delete; /* client is nuts. */
1122 rqstp->rq_arg.len = svsk->sk_datalen;
1123 rqstp->rq_arg.page_base = 0;
1124 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1125 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1126 rqstp->rq_arg.page_len = 0;
1128 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1130 rqstp->rq_xprt_ctxt = NULL;
1131 rqstp->rq_prot = IPPROTO_TCP;
1132 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1133 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1135 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1137 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1140 len = receive_cb_reply(svsk, rqstp);
1142 /* Reset TCP read info */
1143 svsk->sk_datalen = 0;
1144 svc_tcp_fragment_received(svsk);
1149 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1151 serv->sv_stats->nettcpcnt++;
1153 return rqstp->rq_arg.len;
1158 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1161 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1162 svsk->sk_xprt.xpt_server->sv_name, -len);
1163 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1165 return 0; /* record not complete */
1169 * Send out data on TCP socket.
1171 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1173 struct xdr_buf *xbufp = &rqstp->rq_res;
1177 svc_release_skb(rqstp);
1179 /* Set up the first element of the reply kvec.
1180 * Any other kvecs that may be in use have been taken
1181 * care of by the server implementation itself.
1183 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1184 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1186 sent = svc_sendto(rqstp, &rqstp->rq_res);
1187 if (sent != xbufp->len) {
1189 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1190 "- shutting down socket\n",
1191 rqstp->rq_xprt->xpt_server->sv_name,
1192 (sent<0)?"got error":"sent only",
1194 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1195 svc_xprt_enqueue(rqstp->rq_xprt);
1202 * Setup response header. TCP has a 4B record length field.
1204 void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1206 struct kvec *resv = &rqstp->rq_res.head[0];
1208 /* tcp needs a space for the record length... */
1212 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1214 struct sockaddr *sa, int salen,
1217 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1220 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1221 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1222 struct net *, struct sockaddr *,
1224 static void svc_bc_sock_free(struct svc_xprt *xprt);
1226 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1228 struct sockaddr *sa, int salen,
1231 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1234 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1238 static const struct svc_xprt_ops svc_tcp_bc_ops = {
1239 .xpo_create = svc_bc_tcp_create,
1240 .xpo_detach = svc_bc_tcp_sock_detach,
1241 .xpo_free = svc_bc_sock_free,
1242 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1243 .xpo_secure_port = svc_sock_secure_port,
1246 static struct svc_xprt_class svc_tcp_bc_class = {
1247 .xcl_name = "tcp-bc",
1248 .xcl_owner = THIS_MODULE,
1249 .xcl_ops = &svc_tcp_bc_ops,
1250 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1253 static void svc_init_bc_xprt_sock(void)
1255 svc_reg_xprt_class(&svc_tcp_bc_class);
1258 static void svc_cleanup_bc_xprt_sock(void)
1260 svc_unreg_xprt_class(&svc_tcp_bc_class);
1262 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1263 static void svc_init_bc_xprt_sock(void)
1267 static void svc_cleanup_bc_xprt_sock(void)
1270 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1272 static const struct svc_xprt_ops svc_tcp_ops = {
1273 .xpo_create = svc_tcp_create,
1274 .xpo_recvfrom = svc_tcp_recvfrom,
1275 .xpo_sendto = svc_tcp_sendto,
1276 .xpo_release_rqst = svc_release_skb,
1277 .xpo_detach = svc_tcp_sock_detach,
1278 .xpo_free = svc_sock_free,
1279 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1280 .xpo_has_wspace = svc_tcp_has_wspace,
1281 .xpo_accept = svc_tcp_accept,
1282 .xpo_secure_port = svc_sock_secure_port,
1283 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1286 static struct svc_xprt_class svc_tcp_class = {
1288 .xcl_owner = THIS_MODULE,
1289 .xcl_ops = &svc_tcp_ops,
1290 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1291 .xcl_ident = XPRT_TRANSPORT_TCP,
1294 void svc_init_xprt_sock(void)
1296 svc_reg_xprt_class(&svc_tcp_class);
1297 svc_reg_xprt_class(&svc_udp_class);
1298 svc_init_bc_xprt_sock();
1301 void svc_cleanup_xprt_sock(void)
1303 svc_unreg_xprt_class(&svc_tcp_class);
1304 svc_unreg_xprt_class(&svc_udp_class);
1305 svc_cleanup_bc_xprt_sock();
1308 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1310 struct sock *sk = svsk->sk_sk;
1312 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1313 &svsk->sk_xprt, serv);
1314 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1315 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1316 if (sk->sk_state == TCP_LISTEN) {
1317 dprintk("setting up TCP socket for listening\n");
1318 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1319 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1320 sk->sk_data_ready = svc_tcp_listen_data_ready;
1321 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1323 dprintk("setting up TCP socket for reading\n");
1324 sk->sk_state_change = svc_tcp_state_change;
1325 sk->sk_data_ready = svc_data_ready;
1326 sk->sk_write_space = svc_write_space;
1328 svsk->sk_reclen = 0;
1329 svsk->sk_tcplen = 0;
1330 svsk->sk_datalen = 0;
1331 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1333 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1335 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1336 switch (sk->sk_state) {
1338 case TCP_ESTABLISHED:
1341 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1346 void svc_sock_update_bufs(struct svc_serv *serv)
1349 * The number of server threads has changed. Update
1350 * rcvbuf and sndbuf accordingly on all sockets
1352 struct svc_sock *svsk;
1354 spin_lock_bh(&serv->sv_lock);
1355 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1356 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1357 spin_unlock_bh(&serv->sv_lock);
1359 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1362 * Initialize socket for RPC use and create svc_sock struct
1364 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1365 struct socket *sock,
1368 struct svc_sock *svsk;
1370 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1373 dprintk("svc: svc_setup_socket %p\n", sock);
1374 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1376 return ERR_PTR(-ENOMEM);
1380 /* Register socket with portmapper */
1382 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1384 ntohs(inet_sk(inet)->inet_sport));
1388 return ERR_PTR(err);
1391 svsk->sk_sock = sock;
1393 svsk->sk_ostate = inet->sk_state_change;
1394 svsk->sk_odata = inet->sk_data_ready;
1395 svsk->sk_owspace = inet->sk_write_space;
1397 * This barrier is necessary in order to prevent race condition
1398 * with svc_data_ready(), svc_listen_data_ready() and others
1399 * when calling callbacks above.
1402 inet->sk_user_data = svsk;
1404 /* Initialize the socket */
1405 if (sock->type == SOCK_DGRAM)
1406 svc_udp_init(svsk, serv);
1408 svc_tcp_init(svsk, serv);
1410 dprintk("svc: svc_setup_socket created %p (inet %p), "
1411 "listen %d close %d\n",
1413 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1414 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1419 bool svc_alien_sock(struct net *net, int fd)
1422 struct socket *sock = sockfd_lookup(fd, &err);
1427 if (sock_net(sock->sk) != net)
1433 EXPORT_SYMBOL_GPL(svc_alien_sock);
1436 * svc_addsock - add a listener socket to an RPC service
1437 * @serv: pointer to RPC service to which to add a new listener
1438 * @fd: file descriptor of the new listener
1439 * @name_return: pointer to buffer to fill in with name of listener
1440 * @len: size of the buffer
1442 * Fills in socket name and returns positive length of name if successful.
1443 * Name is terminated with '\n'. On error, returns a negative errno
1446 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1450 struct socket *so = sockfd_lookup(fd, &err);
1451 struct svc_sock *svsk = NULL;
1452 struct sockaddr_storage addr;
1453 struct sockaddr *sin = (struct sockaddr *)&addr;
1458 err = -EAFNOSUPPORT;
1459 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1461 err = -EPROTONOSUPPORT;
1462 if (so->sk->sk_protocol != IPPROTO_TCP &&
1463 so->sk->sk_protocol != IPPROTO_UDP)
1466 if (so->state > SS_UNCONNECTED)
1469 if (!try_module_get(THIS_MODULE))
1471 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1473 module_put(THIS_MODULE);
1474 err = PTR_ERR(svsk);
1477 salen = kernel_getsockname(svsk->sk_sock, sin);
1479 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1480 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1481 return svc_one_sock_name(svsk, name_return, len);
1486 EXPORT_SYMBOL_GPL(svc_addsock);
1489 * Create socket for RPC service.
1491 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1494 struct sockaddr *sin, int len,
1497 struct svc_sock *svsk;
1498 struct socket *sock;
1501 struct sockaddr_storage addr;
1502 struct sockaddr *newsin = (struct sockaddr *)&addr;
1506 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1508 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1509 serv->sv_program->pg_name, protocol,
1510 __svc_print_addr(sin, buf, sizeof(buf)));
1512 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1513 printk(KERN_WARNING "svc: only UDP and TCP "
1514 "sockets supported\n");
1515 return ERR_PTR(-EINVAL);
1518 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1519 switch (sin->sa_family) {
1527 return ERR_PTR(-EINVAL);
1530 error = __sock_create(net, family, type, protocol, &sock, 1);
1532 return ERR_PTR(error);
1534 svc_reclassify_socket(sock);
1537 * If this is an PF_INET6 listener, we want to avoid
1538 * getting requests from IPv4 remotes. Those should
1539 * be shunted to a PF_INET listener via rpcbind.
1542 if (family == PF_INET6)
1543 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1544 (char *)&val, sizeof(val));
1546 if (type == SOCK_STREAM)
1547 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1548 error = kernel_bind(sock, sin, len);
1552 error = kernel_getsockname(sock, newsin);
1557 if (protocol == IPPROTO_TCP) {
1558 if ((error = kernel_listen(sock, 64)) < 0)
1562 svsk = svc_setup_socket(serv, sock, flags);
1564 error = PTR_ERR(svsk);
1567 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1568 return (struct svc_xprt *)svsk;
1570 dprintk("svc: svc_create_socket error = %d\n", -error);
1572 return ERR_PTR(error);
1576 * Detach the svc_sock from the socket so that no
1577 * more callbacks occur.
1579 static void svc_sock_detach(struct svc_xprt *xprt)
1581 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1582 struct sock *sk = svsk->sk_sk;
1584 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1586 /* put back the old socket callbacks */
1588 sk->sk_state_change = svsk->sk_ostate;
1589 sk->sk_data_ready = svsk->sk_odata;
1590 sk->sk_write_space = svsk->sk_owspace;
1591 sk->sk_user_data = NULL;
1596 * Disconnect the socket, and reset the callbacks
1598 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1600 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1602 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1604 svc_sock_detach(xprt);
1606 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1607 svc_tcp_clear_pages(svsk);
1608 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1613 * Free the svc_sock's socket resources and the svc_sock itself.
1615 static void svc_sock_free(struct svc_xprt *xprt)
1617 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1618 dprintk("svc: svc_sock_free(%p)\n", svsk);
1620 if (svsk->sk_sock->file)
1621 sockfd_put(svsk->sk_sock);
1623 sock_release(svsk->sk_sock);
1627 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1629 * Create a back channel svc_xprt which shares the fore channel socket.
1631 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1634 struct sockaddr *sin, int len,
1637 struct svc_sock *svsk;
1638 struct svc_xprt *xprt;
1640 if (protocol != IPPROTO_TCP) {
1641 printk(KERN_WARNING "svc: only TCP sockets"
1642 " supported on shared back channel\n");
1643 return ERR_PTR(-EINVAL);
1646 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1648 return ERR_PTR(-ENOMEM);
1650 xprt = &svsk->sk_xprt;
1651 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1652 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1654 serv->sv_bc_xprt = xprt;
1660 * Free a back channel svc_sock.
1662 static void svc_bc_sock_free(struct svc_xprt *xprt)
1665 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1667 #endif /* CONFIG_SUNRPC_BACKCHANNEL */