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 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
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 socket *sock, unsigned int snd,
389 oldfs = get_fs(); set_fs(KERNEL_DS);
390 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
391 (char*)&snd, sizeof(snd));
392 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
393 (char*)&rcv, sizeof(rcv));
395 /* sock_setsockopt limits use to sysctl_?mem_max,
396 * which isn't acceptable. Until that is made conditional
397 * on not having CAP_SYS_RESOURCE or similar, we go direct...
398 * DaveM said I could!
401 sock->sk->sk_sndbuf = snd * 2;
402 sock->sk->sk_rcvbuf = rcv * 2;
403 sock->sk->sk_write_space(sock->sk);
404 release_sock(sock->sk);
408 static int svc_sock_secure_port(struct svc_rqst *rqstp)
410 return svc_port_is_privileged(svc_addr(rqstp));
414 * INET callback when data has been received on the socket.
416 static void svc_data_ready(struct sock *sk)
418 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
421 dprintk("svc: socket %p(inet %p), busy=%d\n",
423 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
425 /* Refer to svc_setup_socket() for details. */
428 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
429 svc_xprt_enqueue(&svsk->sk_xprt);
434 * INET callback when space is newly available on the socket.
436 static void svc_write_space(struct sock *sk)
438 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
441 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
442 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
444 /* Refer to svc_setup_socket() for details. */
446 svsk->sk_owspace(sk);
447 svc_xprt_enqueue(&svsk->sk_xprt);
451 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
453 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
455 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
457 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
460 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
462 struct svc_sock *svsk;
464 struct linger no_linger = {
469 svsk = container_of(xprt, struct svc_sock, sk_xprt);
470 sock = svsk->sk_sock;
471 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
472 (char *)&no_linger, sizeof(no_linger));
476 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
478 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
481 struct in_pktinfo *pki = CMSG_DATA(cmh);
482 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
484 if (cmh->cmsg_type != IP_PKTINFO)
487 daddr->sin_family = AF_INET;
488 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
493 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
495 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
498 struct in6_pktinfo *pki = CMSG_DATA(cmh);
499 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
501 if (cmh->cmsg_type != IPV6_PKTINFO)
504 daddr->sin6_family = AF_INET6;
505 daddr->sin6_addr = pki->ipi6_addr;
506 daddr->sin6_scope_id = pki->ipi6_ifindex;
511 * Copy the UDP datagram's destination address to the rqstp structure.
512 * The 'destination' address in this case is the address to which the
513 * peer sent the datagram, i.e. our local address. For multihomed
514 * hosts, this can change from msg to msg. Note that only the IP
515 * address changes, the port number should remain the same.
517 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
520 switch (cmh->cmsg_level) {
522 return svc_udp_get_dest_address4(rqstp, cmh);
524 return svc_udp_get_dest_address6(rqstp, cmh);
531 * Receive a datagram from a UDP socket.
533 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
535 struct svc_sock *svsk =
536 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
537 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
541 long all[SVC_PKTINFO_SPACE / sizeof(long)];
543 struct cmsghdr *cmh = &buffer.hdr;
544 struct msghdr msg = {
545 .msg_name = svc_addr(rqstp),
547 .msg_controllen = sizeof(buffer),
548 .msg_flags = MSG_DONTWAIT,
553 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
554 /* udp sockets need large rcvbuf as all pending
555 * requests are still in that buffer. sndbuf must
556 * also be large enough that there is enough space
557 * for one reply per thread. We count all threads
558 * rather than threads in a particular pool, which
559 * provides an upper bound on the number of threads
560 * which will access the socket.
562 svc_sock_setbufsize(svsk->sk_sock,
563 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
564 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
566 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
568 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
569 0, 0, MSG_PEEK | MSG_DONTWAIT);
571 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
574 if (err != -EAGAIN) {
575 /* possibly an icmp error */
576 dprintk("svc: recvfrom returned error %d\n", -err);
577 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
581 len = svc_addr_len(svc_addr(rqstp));
582 rqstp->rq_addrlen = len;
583 if (skb->tstamp == 0) {
584 skb->tstamp = ktime_get_real();
585 /* Don't enable netstamp, sunrpc doesn't
586 need that much accuracy */
588 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
589 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
592 rqstp->rq_arg.len = len;
594 rqstp->rq_prot = IPPROTO_UDP;
596 if (!svc_udp_get_dest_address(rqstp, cmh)) {
597 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
598 cmh->cmsg_level, cmh->cmsg_type);
601 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
603 if (skb_is_nonlinear(skb)) {
604 /* we have to copy */
606 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
614 /* we can use it in-place */
615 rqstp->rq_arg.head[0].iov_base = skb->data;
616 rqstp->rq_arg.head[0].iov_len = len;
617 if (skb_checksum_complete(skb))
619 rqstp->rq_xprt_ctxt = skb;
622 rqstp->rq_arg.page_base = 0;
623 if (len <= rqstp->rq_arg.head[0].iov_len) {
624 rqstp->rq_arg.head[0].iov_len = len;
625 rqstp->rq_arg.page_len = 0;
626 rqstp->rq_respages = rqstp->rq_pages+1;
628 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
629 rqstp->rq_respages = rqstp->rq_pages + 1 +
630 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
632 rqstp->rq_next_page = rqstp->rq_respages+1;
635 serv->sv_stats->netudpcnt++;
644 svc_udp_sendto(struct svc_rqst *rqstp)
648 error = svc_sendto(rqstp, &rqstp->rq_res);
649 if (error == -ECONNREFUSED)
650 /* ICMP error on earlier request. */
651 error = svc_sendto(rqstp, &rqstp->rq_res);
656 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
660 static int svc_udp_has_wspace(struct svc_xprt *xprt)
662 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
663 struct svc_serv *serv = xprt->xpt_server;
664 unsigned long required;
667 * Set the SOCK_NOSPACE flag before checking the available
670 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
671 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
672 if (required*2 > sock_wspace(svsk->sk_sk))
674 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
678 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
684 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
688 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
690 struct sockaddr *sa, int salen,
693 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
696 static const struct svc_xprt_ops svc_udp_ops = {
697 .xpo_create = svc_udp_create,
698 .xpo_recvfrom = svc_udp_recvfrom,
699 .xpo_sendto = svc_udp_sendto,
700 .xpo_release_rqst = svc_release_udp_skb,
701 .xpo_detach = svc_sock_detach,
702 .xpo_free = svc_sock_free,
703 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
704 .xpo_has_wspace = svc_udp_has_wspace,
705 .xpo_accept = svc_udp_accept,
706 .xpo_secure_port = svc_sock_secure_port,
707 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
710 static struct svc_xprt_class svc_udp_class = {
712 .xcl_owner = THIS_MODULE,
713 .xcl_ops = &svc_udp_ops,
714 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
715 .xcl_ident = XPRT_TRANSPORT_UDP,
718 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
720 int err, level, optname, one = 1;
722 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
723 &svsk->sk_xprt, serv);
724 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
725 svsk->sk_sk->sk_data_ready = svc_data_ready;
726 svsk->sk_sk->sk_write_space = svc_write_space;
728 /* initialise setting must have enough space to
729 * receive and respond to one request.
730 * svc_udp_recvfrom will re-adjust if necessary
732 svc_sock_setbufsize(svsk->sk_sock,
733 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
734 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
736 /* data might have come in before data_ready set up */
737 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
738 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
740 /* make sure we get destination address info */
741 switch (svsk->sk_sk->sk_family) {
744 optname = IP_PKTINFO;
748 optname = IPV6_RECVPKTINFO;
753 err = kernel_setsockopt(svsk->sk_sock, level, optname,
754 (char *)&one, sizeof(one));
755 dprintk("svc: kernel_setsockopt returned %d\n", err);
759 * A data_ready event on a listening socket means there's a connection
760 * pending. Do not use state_change as a substitute for it.
762 static void svc_tcp_listen_data_ready(struct sock *sk)
764 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
766 dprintk("svc: socket %p TCP (listen) state change %d\n",
770 * This callback may called twice when a new connection
771 * is established as a child socket inherits everything
772 * from a parent LISTEN socket.
773 * 1) data_ready method of the parent socket will be called
774 * when one of child sockets become ESTABLISHED.
775 * 2) data_ready method of the child socket may be called
776 * when it receives data before the socket is accepted.
777 * In case of 2, we should ignore it silently and DO NOT
780 if (sk->sk_state != TCP_LISTEN)
784 /* Refer to svc_setup_socket() for details. */
787 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
788 svc_xprt_enqueue(&svsk->sk_xprt);
790 printk("svc: socket %p: no user data\n", sk);
794 * A state change on a connected socket means it's dying or dead.
796 static void svc_tcp_state_change(struct sock *sk)
798 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
800 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
801 sk, sk->sk_state, sk->sk_user_data);
804 printk("svc: socket %p: no user data\n", sk);
806 /* Refer to svc_setup_socket() for details. */
809 if (sk->sk_state != TCP_ESTABLISHED) {
810 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
811 svc_xprt_enqueue(&svsk->sk_xprt);
817 * Accept a TCP connection
819 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
821 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
822 struct sockaddr_storage addr;
823 struct sockaddr *sin = (struct sockaddr *) &addr;
824 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
825 struct socket *sock = svsk->sk_sock;
826 struct socket *newsock;
827 struct svc_sock *newsvsk;
829 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
831 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
835 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
836 err = kernel_accept(sock, &newsock, O_NONBLOCK);
839 printk(KERN_WARNING "%s: no more sockets!\n",
841 else if (err != -EAGAIN)
842 net_warn_ratelimited("%s: accept failed (err %d)!\n",
843 serv->sv_name, -err);
846 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
848 err = kernel_getpeername(newsock, sin, &slen);
850 net_warn_ratelimited("%s: peername failed (err %d)!\n",
851 serv->sv_name, -err);
852 goto failed; /* aborted connection or whatever */
855 /* Ideally, we would want to reject connections from unauthorized
856 * hosts here, but when we get encryption, the IP of the host won't
857 * tell us anything. For now just warn about unpriv connections.
859 if (!svc_port_is_privileged(sin)) {
860 dprintk("%s: connect from unprivileged port: %s\n",
862 __svc_print_addr(sin, buf, sizeof(buf)));
864 dprintk("%s: connect from %s\n", serv->sv_name,
865 __svc_print_addr(sin, buf, sizeof(buf)));
867 /* Reset the inherited callbacks before calling svc_setup_socket */
868 newsock->sk->sk_state_change = svsk->sk_ostate;
869 newsock->sk->sk_data_ready = svsk->sk_odata;
870 newsock->sk->sk_write_space = svsk->sk_owspace;
872 /* make sure that a write doesn't block forever when
875 newsock->sk->sk_sndtimeo = HZ*30;
877 newsvsk = svc_setup_socket(serv, newsock,
878 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
881 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
882 err = kernel_getsockname(newsock, sin, &slen);
883 if (unlikely(err < 0)) {
884 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
885 slen = offsetof(struct sockaddr, sa_data);
887 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
889 if (sock_is_loopback(newsock->sk))
890 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
892 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
894 serv->sv_stats->nettcpconn++;
896 return &newsvsk->sk_xprt;
899 sock_release(newsock);
903 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
905 unsigned int i, len, npages;
907 if (svsk->sk_datalen == 0)
909 len = svsk->sk_datalen;
910 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
911 for (i = 0; i < npages; i++) {
912 if (rqstp->rq_pages[i] != NULL)
913 put_page(rqstp->rq_pages[i]);
914 BUG_ON(svsk->sk_pages[i] == NULL);
915 rqstp->rq_pages[i] = svsk->sk_pages[i];
916 svsk->sk_pages[i] = NULL;
918 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
922 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
924 unsigned int i, len, npages;
926 if (svsk->sk_datalen == 0)
928 len = svsk->sk_datalen;
929 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
930 for (i = 0; i < npages; i++) {
931 svsk->sk_pages[i] = rqstp->rq_pages[i];
932 rqstp->rq_pages[i] = NULL;
936 static void svc_tcp_clear_pages(struct svc_sock *svsk)
938 unsigned int i, len, npages;
940 if (svsk->sk_datalen == 0)
942 len = svsk->sk_datalen;
943 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
944 for (i = 0; i < npages; i++) {
945 if (svsk->sk_pages[i] == NULL) {
949 put_page(svsk->sk_pages[i]);
950 svsk->sk_pages[i] = NULL;
954 svsk->sk_datalen = 0;
958 * Receive fragment record header.
959 * If we haven't gotten the record length yet, get the next four bytes.
961 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
963 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
967 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
970 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
971 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
973 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
975 svsk->sk_tcplen += len;
978 dprintk("svc: short recvfrom while reading record "
979 "length (%d of %d)\n", len, want);
983 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
984 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
986 net_notice_ratelimited("RPC: fragment too large: %d\n",
987 svc_sock_reclen(svsk));
992 return svc_sock_reclen(svsk);
994 dprintk("RPC: TCP recv_record got %d\n", len);
997 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1001 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1003 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1004 struct rpc_rqst *req = NULL;
1005 struct kvec *src, *dst;
1006 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1015 spin_lock(&bc_xprt->recv_lock);
1016 req = xprt_lookup_rqst(bc_xprt, xid);
1018 goto unlock_notfound;
1020 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1022 * XXX!: cheating for now! Only copying HEAD.
1023 * But we know this is good enough for now (in fact, for any
1024 * callback reply in the forseeable future).
1026 dst = &req->rq_private_buf.head[0];
1027 src = &rqstp->rq_arg.head[0];
1028 if (dst->iov_len < src->iov_len)
1029 goto unlock_eagain; /* whatever; just giving up. */
1030 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1031 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1032 rqstp->rq_arg.len = 0;
1033 spin_unlock(&bc_xprt->recv_lock);
1037 "%s: Got unrecognized reply: "
1038 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1039 __func__, ntohl(calldir),
1040 bc_xprt, ntohl(xid));
1042 spin_unlock(&bc_xprt->recv_lock);
1046 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1052 vec[i].iov_base = page_address(pages[i]);
1053 vec[i].iov_len = PAGE_SIZE;
1060 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1062 /* If we have more data, signal svc_xprt_enqueue() to try again */
1063 dprintk("svc: TCP %s record (%d bytes)\n",
1064 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1065 svc_sock_reclen(svsk));
1066 svsk->sk_tcplen = 0;
1067 svsk->sk_reclen = 0;
1071 * Receive data from a TCP socket.
1073 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1075 struct svc_sock *svsk =
1076 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1077 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1080 unsigned int want, base;
1085 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1086 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1087 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1088 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1090 len = svc_tcp_recv_record(svsk, rqstp);
1094 base = svc_tcp_restore_pages(svsk, rqstp);
1095 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1097 vec = rqstp->rq_vec;
1099 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1100 svsk->sk_datalen + want);
1102 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1103 rqstp->rq_next_page = rqstp->rq_respages + 1;
1105 /* Now receive data */
1106 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1108 svsk->sk_tcplen += len;
1109 svsk->sk_datalen += len;
1111 if (len != want || !svc_sock_final_rec(svsk)) {
1112 svc_tcp_save_pages(svsk, rqstp);
1113 if (len < 0 && len != -EAGAIN)
1116 svc_tcp_fragment_received(svsk);
1118 dprintk("svc: incomplete TCP record (%d of %d)\n",
1119 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1120 svc_sock_reclen(svsk));
1124 if (svsk->sk_datalen < 8) {
1125 svsk->sk_datalen = 0;
1126 goto err_delete; /* client is nuts. */
1129 rqstp->rq_arg.len = svsk->sk_datalen;
1130 rqstp->rq_arg.page_base = 0;
1131 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1132 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1133 rqstp->rq_arg.page_len = 0;
1135 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1137 rqstp->rq_xprt_ctxt = NULL;
1138 rqstp->rq_prot = IPPROTO_TCP;
1139 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1140 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1142 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1144 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1147 len = receive_cb_reply(svsk, rqstp);
1149 /* Reset TCP read info */
1150 svsk->sk_datalen = 0;
1151 svc_tcp_fragment_received(svsk);
1156 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1158 serv->sv_stats->nettcpcnt++;
1160 return rqstp->rq_arg.len;
1165 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1168 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1169 svsk->sk_xprt.xpt_server->sv_name, -len);
1170 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1172 return 0; /* record not complete */
1176 * Send out data on TCP socket.
1178 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1180 struct xdr_buf *xbufp = &rqstp->rq_res;
1184 /* Set up the first element of the reply kvec.
1185 * Any other kvecs that may be in use have been taken
1186 * care of by the server implementation itself.
1188 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1189 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1191 sent = svc_sendto(rqstp, &rqstp->rq_res);
1192 if (sent != xbufp->len) {
1194 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1195 "- shutting down socket\n",
1196 rqstp->rq_xprt->xpt_server->sv_name,
1197 (sent<0)?"got error":"sent only",
1199 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1200 svc_xprt_enqueue(rqstp->rq_xprt);
1207 * Setup response header. TCP has a 4B record length field.
1209 void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1211 struct kvec *resv = &rqstp->rq_res.head[0];
1213 /* tcp needs a space for the record length... */
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 const 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 const 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,
1288 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1291 static struct svc_xprt_class svc_tcp_class = {
1293 .xcl_owner = THIS_MODULE,
1294 .xcl_ops = &svc_tcp_ops,
1295 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1296 .xcl_ident = XPRT_TRANSPORT_TCP,
1299 void svc_init_xprt_sock(void)
1301 svc_reg_xprt_class(&svc_tcp_class);
1302 svc_reg_xprt_class(&svc_udp_class);
1303 svc_init_bc_xprt_sock();
1306 void svc_cleanup_xprt_sock(void)
1308 svc_unreg_xprt_class(&svc_tcp_class);
1309 svc_unreg_xprt_class(&svc_udp_class);
1310 svc_cleanup_bc_xprt_sock();
1313 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1315 struct sock *sk = svsk->sk_sk;
1317 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1318 &svsk->sk_xprt, serv);
1319 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1320 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1321 if (sk->sk_state == TCP_LISTEN) {
1322 dprintk("setting up TCP socket for listening\n");
1323 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1324 sk->sk_data_ready = svc_tcp_listen_data_ready;
1325 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1327 dprintk("setting up TCP socket for reading\n");
1328 sk->sk_state_change = svc_tcp_state_change;
1329 sk->sk_data_ready = svc_data_ready;
1330 sk->sk_write_space = svc_write_space;
1332 svsk->sk_reclen = 0;
1333 svsk->sk_tcplen = 0;
1334 svsk->sk_datalen = 0;
1335 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1337 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1339 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1340 switch (sk->sk_state) {
1342 case TCP_ESTABLISHED:
1345 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1350 void svc_sock_update_bufs(struct svc_serv *serv)
1353 * The number of server threads has changed. Update
1354 * rcvbuf and sndbuf accordingly on all sockets
1356 struct svc_sock *svsk;
1358 spin_lock_bh(&serv->sv_lock);
1359 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1360 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1361 spin_unlock_bh(&serv->sv_lock);
1363 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1366 * Initialize socket for RPC use and create svc_sock struct
1368 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1369 struct socket *sock,
1372 struct svc_sock *svsk;
1374 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1377 dprintk("svc: svc_setup_socket %p\n", sock);
1378 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1380 return ERR_PTR(-ENOMEM);
1384 /* Register socket with portmapper */
1386 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1388 ntohs(inet_sk(inet)->inet_sport));
1392 return ERR_PTR(err);
1395 svsk->sk_sock = sock;
1397 svsk->sk_ostate = inet->sk_state_change;
1398 svsk->sk_odata = inet->sk_data_ready;
1399 svsk->sk_owspace = inet->sk_write_space;
1401 * This barrier is necessary in order to prevent race condition
1402 * with svc_data_ready(), svc_listen_data_ready() and others
1403 * when calling callbacks above.
1406 inet->sk_user_data = svsk;
1408 /* Initialize the socket */
1409 if (sock->type == SOCK_DGRAM)
1410 svc_udp_init(svsk, serv);
1412 svc_tcp_init(svsk, serv);
1414 dprintk("svc: svc_setup_socket created %p (inet %p), "
1415 "listen %d close %d\n",
1417 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1418 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1423 bool svc_alien_sock(struct net *net, int fd)
1426 struct socket *sock = sockfd_lookup(fd, &err);
1431 if (sock_net(sock->sk) != net)
1437 EXPORT_SYMBOL_GPL(svc_alien_sock);
1440 * svc_addsock - add a listener socket to an RPC service
1441 * @serv: pointer to RPC service to which to add a new listener
1442 * @fd: file descriptor of the new listener
1443 * @name_return: pointer to buffer to fill in with name of listener
1444 * @len: size of the buffer
1446 * Fills in socket name and returns positive length of name if successful.
1447 * Name is terminated with '\n'. On error, returns a negative errno
1450 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1454 struct socket *so = sockfd_lookup(fd, &err);
1455 struct svc_sock *svsk = NULL;
1456 struct sockaddr_storage addr;
1457 struct sockaddr *sin = (struct sockaddr *)&addr;
1462 err = -EAFNOSUPPORT;
1463 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1465 err = -EPROTONOSUPPORT;
1466 if (so->sk->sk_protocol != IPPROTO_TCP &&
1467 so->sk->sk_protocol != IPPROTO_UDP)
1470 if (so->state > SS_UNCONNECTED)
1473 if (!try_module_get(THIS_MODULE))
1475 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1477 module_put(THIS_MODULE);
1478 err = PTR_ERR(svsk);
1481 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1482 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1483 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1484 return svc_one_sock_name(svsk, name_return, len);
1489 EXPORT_SYMBOL_GPL(svc_addsock);
1492 * Create socket for RPC service.
1494 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1497 struct sockaddr *sin, int len,
1500 struct svc_sock *svsk;
1501 struct socket *sock;
1504 struct sockaddr_storage addr;
1505 struct sockaddr *newsin = (struct sockaddr *)&addr;
1509 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1511 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1512 serv->sv_program->pg_name, protocol,
1513 __svc_print_addr(sin, buf, sizeof(buf)));
1515 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1516 printk(KERN_WARNING "svc: only UDP and TCP "
1517 "sockets supported\n");
1518 return ERR_PTR(-EINVAL);
1521 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1522 switch (sin->sa_family) {
1530 return ERR_PTR(-EINVAL);
1533 error = __sock_create(net, family, type, protocol, &sock, 1);
1535 return ERR_PTR(error);
1537 svc_reclassify_socket(sock);
1540 * If this is an PF_INET6 listener, we want to avoid
1541 * getting requests from IPv4 remotes. Those should
1542 * be shunted to a PF_INET listener via rpcbind.
1545 if (family == PF_INET6)
1546 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1547 (char *)&val, sizeof(val));
1549 if (type == SOCK_STREAM)
1550 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1551 error = kernel_bind(sock, sin, len);
1556 error = kernel_getsockname(sock, newsin, &newlen);
1560 if (protocol == IPPROTO_TCP) {
1561 if ((error = kernel_listen(sock, 64)) < 0)
1565 svsk = svc_setup_socket(serv, sock, flags);
1567 error = PTR_ERR(svsk);
1570 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1571 return (struct svc_xprt *)svsk;
1573 dprintk("svc: svc_create_socket error = %d\n", -error);
1575 return ERR_PTR(error);
1579 * Detach the svc_sock from the socket so that no
1580 * more callbacks occur.
1582 static void svc_sock_detach(struct svc_xprt *xprt)
1584 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1585 struct sock *sk = svsk->sk_sk;
1587 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1589 /* put back the old socket callbacks */
1591 sk->sk_state_change = svsk->sk_ostate;
1592 sk->sk_data_ready = svsk->sk_odata;
1593 sk->sk_write_space = svsk->sk_owspace;
1594 sk->sk_user_data = NULL;
1599 * Disconnect the socket, and reset the callbacks
1601 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1603 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1605 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1607 svc_sock_detach(xprt);
1609 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1610 svc_tcp_clear_pages(svsk);
1611 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1616 * Free the svc_sock's socket resources and the svc_sock itself.
1618 static void svc_sock_free(struct svc_xprt *xprt)
1620 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1621 dprintk("svc: svc_sock_free(%p)\n", svsk);
1623 if (svsk->sk_sock->file)
1624 sockfd_put(svsk->sk_sock);
1626 sock_release(svsk->sk_sock);
1630 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1632 * Create a back channel svc_xprt which shares the fore channel socket.
1634 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1637 struct sockaddr *sin, int len,
1640 struct svc_sock *svsk;
1641 struct svc_xprt *xprt;
1643 if (protocol != IPPROTO_TCP) {
1644 printk(KERN_WARNING "svc: only TCP sockets"
1645 " supported on shared back channel\n");
1646 return ERR_PTR(-EINVAL);
1649 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1651 return ERR_PTR(-ENOMEM);
1653 xprt = &svsk->sk_xprt;
1654 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1655 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
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 */