1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/net/sunrpc/svcsock.c
5 * These are the RPC server socket internals.
7 * The server scheduling algorithm does not always distribute the load
8 * evenly when servicing a single client. May need to modify the
9 * svc_xprt_enqueue procedure...
11 * TCP support is largely untested and may be a little slow. The problem
12 * is that we currently do two separate recvfrom's, one for the 4-byte
13 * record length, and the second for the actual record. This could possibly
14 * be improved by always reading a minimum size of around 100 bytes and
15 * tucking any superfluous bytes away in a temporary store. Still, that
16 * leaves write requests out in the rain. An alternative may be to peek at
17 * the first skb in the queue, and if it matches the next TCP sequence
18 * number, to extract the record marker. Yuck.
20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
23 #include <linux/kernel.h>
24 #include <linux/sched.h>
25 #include <linux/module.h>
26 #include <linux/errno.h>
27 #include <linux/fcntl.h>
28 #include <linux/net.h>
30 #include <linux/inet.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/unistd.h>
34 #include <linux/slab.h>
35 #include <linux/netdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/file.h>
38 #include <linux/freezer.h>
39 #include <linux/bvec.h>
42 #include <net/checksum.h>
47 #include <net/tcp_states.h>
48 #include <net/tls_prot.h>
49 #include <net/handshake.h>
50 #include <linux/uaccess.h>
51 #include <linux/highmem.h>
52 #include <asm/ioctls.h>
53 #include <linux/key.h>
55 #include <linux/sunrpc/types.h>
56 #include <linux/sunrpc/clnt.h>
57 #include <linux/sunrpc/xdr.h>
58 #include <linux/sunrpc/msg_prot.h>
59 #include <linux/sunrpc/svcsock.h>
60 #include <linux/sunrpc/stats.h>
61 #include <linux/sunrpc/xprt.h>
63 #include <trace/events/sock.h>
64 #include <trace/events/sunrpc.h>
69 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
71 /* To-do: to avoid tying up an nfsd thread while waiting for a
72 * handshake request, the request could instead be deferred.
75 SVC_HANDSHAKE_TO = 5U * HZ
78 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
80 static int svc_udp_recvfrom(struct svc_rqst *);
81 static int svc_udp_sendto(struct svc_rqst *);
82 static void svc_sock_detach(struct svc_xprt *);
83 static void svc_tcp_sock_detach(struct svc_xprt *);
84 static void svc_sock_free(struct svc_xprt *);
86 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
87 struct net *, struct sockaddr *,
89 #ifdef CONFIG_DEBUG_LOCK_ALLOC
90 static struct lock_class_key svc_key[2];
91 static struct lock_class_key svc_slock_key[2];
93 static void svc_reclassify_socket(struct socket *sock)
95 struct sock *sk = sock->sk;
97 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
100 switch (sk->sk_family) {
102 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
104 "sk_xprt.xpt_lock-AF_INET-NFSD",
109 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
111 "sk_xprt.xpt_lock-AF_INET6-NFSD",
120 static void svc_reclassify_socket(struct socket *sock)
126 * svc_tcp_release_ctxt - Release transport-related resources
127 * @xprt: the transport which owned the context
128 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
131 static void svc_tcp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
136 * svc_udp_release_ctxt - Release transport-related resources
137 * @xprt: the transport which owned the context
138 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
141 static void svc_udp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
143 struct sk_buff *skb = ctxt;
149 union svc_pktinfo_u {
150 struct in_pktinfo pkti;
151 struct in6_pktinfo pkti6;
153 #define SVC_PKTINFO_SPACE \
154 CMSG_SPACE(sizeof(union svc_pktinfo_u))
156 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
158 struct svc_sock *svsk =
159 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
160 switch (svsk->sk_sk->sk_family) {
162 struct in_pktinfo *pki = CMSG_DATA(cmh);
164 cmh->cmsg_level = SOL_IP;
165 cmh->cmsg_type = IP_PKTINFO;
166 pki->ipi_ifindex = 0;
167 pki->ipi_spec_dst.s_addr =
168 svc_daddr_in(rqstp)->sin_addr.s_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
174 struct in6_pktinfo *pki = CMSG_DATA(cmh);
175 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
177 cmh->cmsg_level = SOL_IPV6;
178 cmh->cmsg_type = IPV6_PKTINFO;
179 pki->ipi6_ifindex = daddr->sin6_scope_id;
180 pki->ipi6_addr = daddr->sin6_addr;
181 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
187 static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
194 * Report socket names for nfsdfs
196 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
198 const struct sock *sk = svsk->sk_sk;
199 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
203 switch (sk->sk_family) {
205 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
207 &inet_sk(sk)->inet_rcv_saddr,
208 inet_sk(sk)->inet_num);
210 #if IS_ENABLED(CONFIG_IPV6)
212 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
214 &sk->sk_v6_rcv_saddr,
215 inet_sk(sk)->inet_num);
219 len = snprintf(buf, remaining, "*unknown-%d*\n",
223 if (len >= remaining) {
225 return -ENAMETOOLONG;
231 svc_tcp_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
232 struct cmsghdr *cmsg, int ret)
234 u8 content_type = tls_get_record_type(sock->sk, cmsg);
235 u8 level, description;
237 switch (content_type) {
240 case TLS_RECORD_TYPE_DATA:
241 /* TLS sets EOR at the end of each application data
242 * record, even though there might be more frames
243 * waiting to be decrypted.
245 msg->msg_flags &= ~MSG_EOR;
247 case TLS_RECORD_TYPE_ALERT:
248 tls_alert_recv(sock->sk, msg, &level, &description);
249 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
253 /* discard this record type */
260 svc_tcp_sock_recv_cmsg(struct svc_sock *svsk, struct msghdr *msg)
264 u8 buf[CMSG_SPACE(sizeof(u8))];
266 struct socket *sock = svsk->sk_sock;
269 msg->msg_control = &u;
270 msg->msg_controllen = sizeof(u);
271 ret = sock_recvmsg(sock, msg, MSG_DONTWAIT);
272 if (unlikely(msg->msg_controllen != sizeof(u)))
273 ret = svc_tcp_sock_process_cmsg(sock, msg, &u.cmsg, ret);
277 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
278 static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
280 struct bvec_iter bi = {
281 .bi_size = size + seek,
285 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
286 for_each_bvec(bv, bvec, bi, bi)
287 flush_dcache_page(bv.bv_page);
290 static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
297 * Read from @rqstp's transport socket. The incoming message fills whole
298 * pages in @rqstp's rq_pages array until the last page of the message
299 * has been received into a partial page.
301 static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
304 struct svc_sock *svsk =
305 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
306 struct bio_vec *bvec = rqstp->rq_bvec;
307 struct msghdr msg = { NULL };
312 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
314 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE)
315 bvec_set_page(&bvec[i], rqstp->rq_pages[i], PAGE_SIZE, 0);
316 rqstp->rq_respages = &rqstp->rq_pages[i];
317 rqstp->rq_next_page = rqstp->rq_respages + 1;
319 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen);
321 iov_iter_advance(&msg.msg_iter, seek);
324 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
326 svc_flush_bvec(bvec, len, seek);
328 /* If we read a full record, then assume there may be more
329 * data to read (stream based sockets only!)
332 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
338 * Set socket snd and rcv buffer lengths
340 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
342 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
343 struct socket *sock = svsk->sk_sock;
345 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
348 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
349 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
350 sock->sk->sk_write_space(sock->sk);
351 release_sock(sock->sk);
354 static void svc_sock_secure_port(struct svc_rqst *rqstp)
356 if (svc_port_is_privileged(svc_addr(rqstp)))
357 set_bit(RQ_SECURE, &rqstp->rq_flags);
359 clear_bit(RQ_SECURE, &rqstp->rq_flags);
363 * INET callback when data has been received on the socket.
365 static void svc_data_ready(struct sock *sk)
367 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
369 trace_sk_data_ready(sk);
372 /* Refer to svc_setup_socket() for details. */
375 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
376 if (test_bit(XPT_HANDSHAKE, &svsk->sk_xprt.xpt_flags))
378 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
379 svc_xprt_enqueue(&svsk->sk_xprt);
384 * INET callback when space is newly available on the socket.
386 static void svc_write_space(struct sock *sk)
388 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
391 /* Refer to svc_setup_socket() for details. */
393 trace_svcsock_write_space(&svsk->sk_xprt, 0);
394 svsk->sk_owspace(sk);
395 svc_xprt_enqueue(&svsk->sk_xprt);
399 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
401 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
403 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
405 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
408 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
410 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
412 sock_no_linger(svsk->sk_sock->sk);
416 * svc_tcp_handshake_done - Handshake completion handler
417 * @data: address of xprt to wake
418 * @status: status of handshake
419 * @peerid: serial number of key containing the remote peer's identity
421 * If a security policy is specified as an export option, we don't
422 * have a specific export here to check. So we set a "TLS session
423 * is present" flag on the xprt and let an upper layer enforce local
426 static void svc_tcp_handshake_done(void *data, int status, key_serial_t peerid)
428 struct svc_xprt *xprt = data;
429 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
432 if (peerid != TLS_NO_PEERID)
433 set_bit(XPT_PEER_AUTH, &xprt->xpt_flags);
434 set_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
436 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
437 complete_all(&svsk->sk_handshake_done);
441 * svc_tcp_handshake - Perform a transport-layer security handshake
442 * @xprt: connected transport endpoint
445 static void svc_tcp_handshake(struct svc_xprt *xprt)
447 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
448 struct sock *sk = svsk->sk_sock->sk;
449 struct tls_handshake_args args = {
450 .ta_sock = svsk->sk_sock,
451 .ta_done = svc_tcp_handshake_done,
456 trace_svc_tls_upcall(xprt);
458 clear_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
459 init_completion(&svsk->sk_handshake_done);
461 ret = tls_server_hello_x509(&args, GFP_KERNEL);
463 trace_svc_tls_not_started(xprt);
467 ret = wait_for_completion_interruptible_timeout(&svsk->sk_handshake_done,
470 if (tls_handshake_cancel(sk)) {
471 trace_svc_tls_timed_out(xprt);
476 if (!test_bit(XPT_TLS_SESSION, &xprt->xpt_flags)) {
477 trace_svc_tls_unavailable(xprt);
481 /* Mark the transport ready in case the remote sent RPC
482 * traffic before the kernel received the handshake
483 * completion downcall.
485 set_bit(XPT_DATA, &xprt->xpt_flags);
486 svc_xprt_enqueue(xprt);
490 set_bit(XPT_CLOSE, &xprt->xpt_flags);
492 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
493 set_bit(XPT_DATA, &xprt->xpt_flags);
494 svc_xprt_enqueue(xprt);
498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
500 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
503 struct in_pktinfo *pki = CMSG_DATA(cmh);
504 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
506 if (cmh->cmsg_type != IP_PKTINFO)
509 daddr->sin_family = AF_INET;
510 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
515 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
517 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
520 struct in6_pktinfo *pki = CMSG_DATA(cmh);
521 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
523 if (cmh->cmsg_type != IPV6_PKTINFO)
526 daddr->sin6_family = AF_INET6;
527 daddr->sin6_addr = pki->ipi6_addr;
528 daddr->sin6_scope_id = pki->ipi6_ifindex;
533 * Copy the UDP datagram's destination address to the rqstp structure.
534 * The 'destination' address in this case is the address to which the
535 * peer sent the datagram, i.e. our local address. For multihomed
536 * hosts, this can change from msg to msg. Note that only the IP
537 * address changes, the port number should remain the same.
539 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
542 switch (cmh->cmsg_level) {
544 return svc_udp_get_dest_address4(rqstp, cmh);
546 return svc_udp_get_dest_address6(rqstp, cmh);
553 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
554 * @rqstp: request structure into which to receive an RPC Call
556 * Called in a loop when XPT_DATA has been set.
559 * On success, the number of bytes in a received RPC Call, or
560 * %0 if a complete RPC Call message was not ready to return
562 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
572 struct cmsghdr *cmh = &buffer.hdr;
573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
591 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
593 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
594 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
595 0, 0, MSG_PEEK | MSG_DONTWAIT);
598 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err);
602 len = svc_addr_len(svc_addr(rqstp));
603 rqstp->rq_addrlen = len;
604 if (skb->tstamp == 0) {
605 skb->tstamp = ktime_get_real();
606 /* Don't enable netstamp, sunrpc doesn't
607 need that much accuracy */
609 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
610 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
613 rqstp->rq_arg.len = len;
614 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
616 rqstp->rq_prot = IPPROTO_UDP;
618 if (!svc_udp_get_dest_address(rqstp, cmh))
620 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
622 if (skb_is_nonlinear(skb)) {
623 /* we have to copy */
625 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
630 /* we can use it in-place */
631 rqstp->rq_arg.head[0].iov_base = skb->data;
632 rqstp->rq_arg.head[0].iov_len = len;
633 if (skb_checksum_complete(skb))
635 rqstp->rq_xprt_ctxt = skb;
638 rqstp->rq_arg.page_base = 0;
639 if (len <= rqstp->rq_arg.head[0].iov_len) {
640 rqstp->rq_arg.head[0].iov_len = len;
641 rqstp->rq_arg.page_len = 0;
642 rqstp->rq_respages = rqstp->rq_pages+1;
644 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
645 rqstp->rq_respages = rqstp->rq_pages + 1 +
646 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
648 rqstp->rq_next_page = rqstp->rq_respages+1;
651 serv->sv_stats->netudpcnt++;
653 svc_sock_secure_port(rqstp);
654 svc_xprt_received(rqstp->rq_xprt);
658 if (err != -EAGAIN) {
659 /* possibly an icmp error */
660 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
662 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
665 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
666 cmh->cmsg_level, cmh->cmsg_type);
673 svc_xprt_received(rqstp->rq_xprt);
678 * svc_udp_sendto - Send out a reply on a UDP socket
679 * @rqstp: completed svc_rqst
681 * xpt_mutex ensures @rqstp's whole message is written to the socket
682 * without interruption.
684 * Returns the number of bytes sent, or a negative errno.
686 static int svc_udp_sendto(struct svc_rqst *rqstp)
688 struct svc_xprt *xprt = rqstp->rq_xprt;
689 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
690 struct xdr_buf *xdr = &rqstp->rq_res;
693 long all[SVC_PKTINFO_SPACE / sizeof(long)];
695 struct cmsghdr *cmh = &buffer.hdr;
696 struct msghdr msg = {
697 .msg_name = &rqstp->rq_addr,
698 .msg_namelen = rqstp->rq_addrlen,
700 .msg_flags = MSG_SPLICE_PAGES,
701 .msg_controllen = sizeof(buffer),
706 svc_udp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
707 rqstp->rq_xprt_ctxt = NULL;
709 svc_set_cmsg_data(rqstp, cmh);
711 mutex_lock(&xprt->xpt_mutex);
713 if (svc_xprt_is_dead(xprt))
716 count = xdr_buf_to_bvec(rqstp->rq_bvec,
717 ARRAY_SIZE(rqstp->rq_bvec), xdr);
719 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
721 err = sock_sendmsg(svsk->sk_sock, &msg);
722 if (err == -ECONNREFUSED) {
723 /* ICMP error on earlier request. */
724 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
726 err = sock_sendmsg(svsk->sk_sock, &msg);
729 trace_svcsock_udp_send(xprt, err);
731 mutex_unlock(&xprt->xpt_mutex);
735 mutex_unlock(&xprt->xpt_mutex);
739 static int svc_udp_has_wspace(struct svc_xprt *xprt)
741 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
742 struct svc_serv *serv = xprt->xpt_server;
743 unsigned long required;
746 * Set the SOCK_NOSPACE flag before checking the available
749 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
750 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
751 if (required*2 > sock_wspace(svsk->sk_sk))
753 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
757 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
763 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
767 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
769 struct sockaddr *sa, int salen,
772 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
775 static const struct svc_xprt_ops svc_udp_ops = {
776 .xpo_create = svc_udp_create,
777 .xpo_recvfrom = svc_udp_recvfrom,
778 .xpo_sendto = svc_udp_sendto,
779 .xpo_result_payload = svc_sock_result_payload,
780 .xpo_release_ctxt = svc_udp_release_ctxt,
781 .xpo_detach = svc_sock_detach,
782 .xpo_free = svc_sock_free,
783 .xpo_has_wspace = svc_udp_has_wspace,
784 .xpo_accept = svc_udp_accept,
785 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
788 static struct svc_xprt_class svc_udp_class = {
790 .xcl_owner = THIS_MODULE,
791 .xcl_ops = &svc_udp_ops,
792 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
793 .xcl_ident = XPRT_TRANSPORT_UDP,
796 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
798 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
799 &svsk->sk_xprt, serv);
800 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
801 svsk->sk_sk->sk_data_ready = svc_data_ready;
802 svsk->sk_sk->sk_write_space = svc_write_space;
804 /* initialise setting must have enough space to
805 * receive and respond to one request.
806 * svc_udp_recvfrom will re-adjust if necessary
808 svc_sock_setbufsize(svsk, 3);
810 /* data might have come in before data_ready set up */
811 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
812 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
814 /* make sure we get destination address info */
815 switch (svsk->sk_sk->sk_family) {
817 ip_sock_set_pktinfo(svsk->sk_sock->sk);
820 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
828 * A data_ready event on a listening socket means there's a connection
829 * pending. Do not use state_change as a substitute for it.
831 static void svc_tcp_listen_data_ready(struct sock *sk)
833 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
835 trace_sk_data_ready(sk);
838 * This callback may called twice when a new connection
839 * is established as a child socket inherits everything
840 * from a parent LISTEN socket.
841 * 1) data_ready method of the parent socket will be called
842 * when one of child sockets become ESTABLISHED.
843 * 2) data_ready method of the child socket may be called
844 * when it receives data before the socket is accepted.
845 * In case of 2, we should ignore it silently and DO NOT
848 if (sk->sk_state != TCP_LISTEN)
852 /* Refer to svc_setup_socket() for details. */
855 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
856 svc_xprt_enqueue(&svsk->sk_xprt);
861 * A state change on a connected socket means it's dying or dead.
863 static void svc_tcp_state_change(struct sock *sk)
865 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
868 /* Refer to svc_setup_socket() for details. */
871 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
872 if (sk->sk_state != TCP_ESTABLISHED)
873 svc_xprt_deferred_close(&svsk->sk_xprt);
878 * Accept a TCP connection
880 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
882 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
883 struct sockaddr_storage addr;
884 struct sockaddr *sin = (struct sockaddr *) &addr;
885 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
886 struct socket *sock = svsk->sk_sock;
887 struct socket *newsock;
888 struct svc_sock *newsvsk;
894 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
895 err = kernel_accept(sock, &newsock, O_NONBLOCK);
898 trace_svcsock_accept_err(xprt, serv->sv_name, err);
901 if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL)))
904 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
906 err = kernel_getpeername(newsock, sin);
908 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
909 goto failed; /* aborted connection or whatever */
913 /* Reset the inherited callbacks before calling svc_setup_socket */
914 newsock->sk->sk_state_change = svsk->sk_ostate;
915 newsock->sk->sk_data_ready = svsk->sk_odata;
916 newsock->sk->sk_write_space = svsk->sk_owspace;
918 /* make sure that a write doesn't block forever when
921 newsock->sk->sk_sndtimeo = HZ*30;
923 newsvsk = svc_setup_socket(serv, newsock,
924 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
927 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
928 err = kernel_getsockname(newsock, sin);
930 if (unlikely(err < 0))
931 slen = offsetof(struct sockaddr, sa_data);
932 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
934 if (sock_is_loopback(newsock->sk))
935 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
937 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
939 serv->sv_stats->nettcpconn++;
941 return &newsvsk->sk_xprt;
948 static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
949 struct svc_rqst *rqstp)
951 size_t len = svsk->sk_datalen;
952 unsigned int i, npages;
956 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
957 for (i = 0; i < npages; i++) {
958 if (rqstp->rq_pages[i] != NULL)
959 put_page(rqstp->rq_pages[i]);
960 BUG_ON(svsk->sk_pages[i] == NULL);
961 rqstp->rq_pages[i] = svsk->sk_pages[i];
962 svsk->sk_pages[i] = NULL;
964 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
968 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
970 unsigned int i, len, npages;
972 if (svsk->sk_datalen == 0)
974 len = svsk->sk_datalen;
975 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 for (i = 0; i < npages; i++) {
977 svsk->sk_pages[i] = rqstp->rq_pages[i];
978 rqstp->rq_pages[i] = NULL;
982 static void svc_tcp_clear_pages(struct svc_sock *svsk)
984 unsigned int i, len, npages;
986 if (svsk->sk_datalen == 0)
988 len = svsk->sk_datalen;
989 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
990 for (i = 0; i < npages; i++) {
991 if (svsk->sk_pages[i] == NULL) {
995 put_page(svsk->sk_pages[i]);
996 svsk->sk_pages[i] = NULL;
1000 svsk->sk_datalen = 0;
1004 * Receive fragment record header into sk_marker.
1006 static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
1007 struct svc_rqst *rqstp)
1011 /* If we haven't gotten the record length yet,
1012 * get the next four bytes.
1014 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1015 struct msghdr msg = { NULL };
1018 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1019 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
1021 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
1022 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
1025 svsk->sk_tcplen += len;
1027 /* call again to read the remaining bytes */
1030 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
1031 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1032 svsk->sk_xprt.xpt_server->sv_max_mesg)
1035 return svc_sock_reclen(svsk);
1038 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
1039 __func__, svsk->sk_xprt.xpt_server->sv_name,
1040 svc_sock_reclen(svsk));
1041 svc_xprt_deferred_close(&svsk->sk_xprt);
1046 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1048 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1049 struct rpc_rqst *req = NULL;
1050 struct kvec *src, *dst;
1051 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1060 spin_lock(&bc_xprt->queue_lock);
1061 req = xprt_lookup_rqst(bc_xprt, xid);
1063 goto unlock_notfound;
1065 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1067 * XXX!: cheating for now! Only copying HEAD.
1068 * But we know this is good enough for now (in fact, for any
1069 * callback reply in the forseeable future).
1071 dst = &req->rq_private_buf.head[0];
1072 src = &rqstp->rq_arg.head[0];
1073 if (dst->iov_len < src->iov_len)
1074 goto unlock_eagain; /* whatever; just giving up. */
1075 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1076 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1077 rqstp->rq_arg.len = 0;
1078 spin_unlock(&bc_xprt->queue_lock);
1082 "%s: Got unrecognized reply: "
1083 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1084 __func__, ntohl(calldir),
1085 bc_xprt, ntohl(xid));
1087 spin_unlock(&bc_xprt->queue_lock);
1091 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1093 /* If we have more data, signal svc_xprt_enqueue() to try again */
1094 svsk->sk_tcplen = 0;
1095 svsk->sk_marker = xdr_zero;
1098 tcp_set_rcvlowat(svsk->sk_sk, 1);
1102 * svc_tcp_recvfrom - Receive data from a TCP socket
1103 * @rqstp: request structure into which to receive an RPC Call
1105 * Called in a loop when XPT_DATA has been set.
1107 * Read the 4-byte stream record marker, then use the record length
1108 * in that marker to set up exactly the resources needed to receive
1109 * the next RPC message into @rqstp.
1112 * On success, the number of bytes in a received RPC Call, or
1113 * %0 if a complete RPC Call message was not ready to return
1115 * The zero return case handles partial receives and callback Replies.
1116 * The state of a partial receive is preserved in the svc_sock for
1117 * the next call to svc_tcp_recvfrom.
1119 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1121 struct svc_sock *svsk =
1122 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1123 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1129 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1130 len = svc_tcp_read_marker(svsk, rqstp);
1134 base = svc_tcp_restore_pages(svsk, rqstp);
1135 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1136 len = svc_tcp_read_msg(rqstp, base + want, base);
1138 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1139 svsk->sk_tcplen += len;
1140 svsk->sk_datalen += len;
1142 if (len != want || !svc_sock_final_rec(svsk))
1143 goto err_incomplete;
1144 if (svsk->sk_datalen < 8)
1147 rqstp->rq_arg.len = svsk->sk_datalen;
1148 rqstp->rq_arg.page_base = 0;
1149 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1150 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1151 rqstp->rq_arg.page_len = 0;
1153 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1155 rqstp->rq_xprt_ctxt = NULL;
1156 rqstp->rq_prot = IPPROTO_TCP;
1157 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1158 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1160 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1162 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1165 len = receive_cb_reply(svsk, rqstp);
1167 /* Reset TCP read info */
1168 svsk->sk_datalen = 0;
1169 svc_tcp_fragment_received(svsk);
1174 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1176 serv->sv_stats->nettcpcnt++;
1178 svc_sock_secure_port(rqstp);
1179 svc_xprt_received(rqstp->rq_xprt);
1180 return rqstp->rq_arg.len;
1183 svc_tcp_save_pages(svsk, rqstp);
1184 if (len < 0 && len != -EAGAIN)
1187 svc_tcp_fragment_received(svsk);
1189 /* Avoid more ->sk_data_ready() calls until the rest
1190 * of the message has arrived. This reduces service
1191 * thread wake-ups on large incoming messages. */
1192 tcp_set_rcvlowat(svsk->sk_sk,
1193 svc_sock_reclen(svsk) - svsk->sk_tcplen);
1195 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1196 svc_sock_reclen(svsk),
1197 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1203 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1206 svsk->sk_datalen = 0;
1208 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1209 svc_xprt_deferred_close(&svsk->sk_xprt);
1211 svc_xprt_received(rqstp->rq_xprt);
1212 return 0; /* record not complete */
1216 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1217 * copy operations in this path. Therefore the caller must ensure
1218 * that the pages backing @xdr are unchanging.
1220 * Note that the send is non-blocking. The caller has incremented
1221 * the reference count on each page backing the RPC message, and
1222 * the network layer will "put" these pages when transmission is
1225 * This is safe for our RPC services because the memory backing
1226 * the head and tail components is never kmalloc'd. These always
1227 * come from pages in the svc_rqst::rq_pages array.
1229 static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1230 rpc_fraghdr marker, unsigned int *sentp)
1232 struct msghdr msg = {
1233 .msg_flags = MSG_SPLICE_PAGES,
1241 /* The stream record marker is copied into a temporary page
1242 * fragment buffer so that it can be included in rq_bvec.
1244 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1248 memcpy(buf, &marker, sizeof(marker));
1249 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1251 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1252 ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1254 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1255 1 + count, sizeof(marker) + rqstp->rq_res.len);
1256 ret = sock_sendmsg(svsk->sk_sock, &msg);
1264 * svc_tcp_sendto - Send out a reply on a TCP socket
1265 * @rqstp: completed svc_rqst
1267 * xpt_mutex ensures @rqstp's whole message is written to the socket
1268 * without interruption.
1270 * Returns the number of bytes sent, or a negative errno.
1272 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1274 struct svc_xprt *xprt = rqstp->rq_xprt;
1275 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1276 struct xdr_buf *xdr = &rqstp->rq_res;
1277 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1282 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1283 rqstp->rq_xprt_ctxt = NULL;
1285 mutex_lock(&xprt->xpt_mutex);
1286 if (svc_xprt_is_dead(xprt))
1288 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1289 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1290 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1292 mutex_unlock(&xprt->xpt_mutex);
1296 mutex_unlock(&xprt->xpt_mutex);
1299 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1300 xprt->xpt_server->sv_name,
1301 (err < 0) ? "got error" : "sent",
1302 (err < 0) ? err : sent, xdr->len);
1303 svc_xprt_deferred_close(xprt);
1304 mutex_unlock(&xprt->xpt_mutex);
1308 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1310 struct sockaddr *sa, int salen,
1313 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1316 static const struct svc_xprt_ops svc_tcp_ops = {
1317 .xpo_create = svc_tcp_create,
1318 .xpo_recvfrom = svc_tcp_recvfrom,
1319 .xpo_sendto = svc_tcp_sendto,
1320 .xpo_result_payload = svc_sock_result_payload,
1321 .xpo_release_ctxt = svc_tcp_release_ctxt,
1322 .xpo_detach = svc_tcp_sock_detach,
1323 .xpo_free = svc_sock_free,
1324 .xpo_has_wspace = svc_tcp_has_wspace,
1325 .xpo_accept = svc_tcp_accept,
1326 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1327 .xpo_handshake = svc_tcp_handshake,
1330 static struct svc_xprt_class svc_tcp_class = {
1332 .xcl_owner = THIS_MODULE,
1333 .xcl_ops = &svc_tcp_ops,
1334 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1335 .xcl_ident = XPRT_TRANSPORT_TCP,
1338 void svc_init_xprt_sock(void)
1340 svc_reg_xprt_class(&svc_tcp_class);
1341 svc_reg_xprt_class(&svc_udp_class);
1344 void svc_cleanup_xprt_sock(void)
1346 svc_unreg_xprt_class(&svc_tcp_class);
1347 svc_unreg_xprt_class(&svc_udp_class);
1350 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1352 struct sock *sk = svsk->sk_sk;
1354 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1355 &svsk->sk_xprt, serv);
1356 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1357 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1358 if (sk->sk_state == TCP_LISTEN) {
1359 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1360 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1361 sk->sk_data_ready = svc_tcp_listen_data_ready;
1362 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1364 sk->sk_state_change = svc_tcp_state_change;
1365 sk->sk_data_ready = svc_data_ready;
1366 sk->sk_write_space = svc_write_space;
1368 svsk->sk_marker = xdr_zero;
1369 svsk->sk_tcplen = 0;
1370 svsk->sk_datalen = 0;
1371 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1373 tcp_sock_set_nodelay(sk);
1375 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1376 switch (sk->sk_state) {
1378 case TCP_ESTABLISHED:
1381 svc_xprt_deferred_close(&svsk->sk_xprt);
1386 void svc_sock_update_bufs(struct svc_serv *serv)
1389 * The number of server threads has changed. Update
1390 * rcvbuf and sndbuf accordingly on all sockets
1392 struct svc_sock *svsk;
1394 spin_lock_bh(&serv->sv_lock);
1395 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1396 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1397 spin_unlock_bh(&serv->sv_lock);
1399 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1402 * Initialize socket for RPC use and create svc_sock struct
1404 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1405 struct socket *sock,
1408 struct svc_sock *svsk;
1410 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1412 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1414 return ERR_PTR(-ENOMEM);
1418 if (pmap_register) {
1421 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1423 ntohs(inet_sk(inet)->inet_sport));
1426 return ERR_PTR(err);
1430 svsk->sk_sock = sock;
1432 svsk->sk_ostate = inet->sk_state_change;
1433 svsk->sk_odata = inet->sk_data_ready;
1434 svsk->sk_owspace = inet->sk_write_space;
1436 * This barrier is necessary in order to prevent race condition
1437 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1438 * when calling callbacks above.
1441 inet->sk_user_data = svsk;
1443 /* Initialize the socket */
1444 if (sock->type == SOCK_DGRAM)
1445 svc_udp_init(svsk, serv);
1447 svc_tcp_init(svsk, serv);
1449 trace_svcsock_new(svsk, sock);
1454 * svc_addsock - add a listener socket to an RPC service
1455 * @serv: pointer to RPC service to which to add a new listener
1456 * @net: caller's network namespace
1457 * @fd: file descriptor of the new listener
1458 * @name_return: pointer to buffer to fill in with name of listener
1459 * @len: size of the buffer
1462 * Fills in socket name and returns positive length of name if successful.
1463 * Name is terminated with '\n'. On error, returns a negative errno
1466 int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1467 char *name_return, const size_t len, const struct cred *cred)
1470 struct socket *so = sockfd_lookup(fd, &err);
1471 struct svc_sock *svsk = NULL;
1472 struct sockaddr_storage addr;
1473 struct sockaddr *sin = (struct sockaddr *)&addr;
1479 if (sock_net(so->sk) != net)
1481 err = -EAFNOSUPPORT;
1482 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1484 err = -EPROTONOSUPPORT;
1485 if (so->sk->sk_protocol != IPPROTO_TCP &&
1486 so->sk->sk_protocol != IPPROTO_UDP)
1489 if (so->state > SS_UNCONNECTED)
1492 if (!try_module_get(THIS_MODULE))
1494 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1496 module_put(THIS_MODULE);
1497 err = PTR_ERR(svsk);
1500 salen = kernel_getsockname(svsk->sk_sock, sin);
1502 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1503 svsk->sk_xprt.xpt_cred = get_cred(cred);
1504 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1505 return svc_one_sock_name(svsk, name_return, len);
1510 EXPORT_SYMBOL_GPL(svc_addsock);
1513 * Create socket for RPC service.
1515 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1518 struct sockaddr *sin, int len,
1521 struct svc_sock *svsk;
1522 struct socket *sock;
1525 struct sockaddr_storage addr;
1526 struct sockaddr *newsin = (struct sockaddr *)&addr;
1530 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1531 printk(KERN_WARNING "svc: only UDP and TCP "
1532 "sockets supported\n");
1533 return ERR_PTR(-EINVAL);
1536 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1537 switch (sin->sa_family) {
1545 return ERR_PTR(-EINVAL);
1548 error = __sock_create(net, family, type, protocol, &sock, 1);
1550 return ERR_PTR(error);
1552 svc_reclassify_socket(sock);
1555 * If this is an PF_INET6 listener, we want to avoid
1556 * getting requests from IPv4 remotes. Those should
1557 * be shunted to a PF_INET listener via rpcbind.
1559 if (family == PF_INET6)
1560 ip6_sock_set_v6only(sock->sk);
1561 if (type == SOCK_STREAM)
1562 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1563 error = kernel_bind(sock, sin, len);
1567 error = kernel_getsockname(sock, newsin);
1572 if (protocol == IPPROTO_TCP) {
1573 if ((error = kernel_listen(sock, 64)) < 0)
1577 svsk = svc_setup_socket(serv, sock, flags);
1579 error = PTR_ERR(svsk);
1582 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1583 return (struct svc_xprt *)svsk;
1586 return ERR_PTR(error);
1590 * Detach the svc_sock from the socket so that no
1591 * more callbacks occur.
1593 static void svc_sock_detach(struct svc_xprt *xprt)
1595 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1596 struct sock *sk = svsk->sk_sk;
1598 /* put back the old socket callbacks */
1600 sk->sk_state_change = svsk->sk_ostate;
1601 sk->sk_data_ready = svsk->sk_odata;
1602 sk->sk_write_space = svsk->sk_owspace;
1603 sk->sk_user_data = NULL;
1608 * Disconnect the socket, and reset the callbacks
1610 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1612 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1614 tls_handshake_close(svsk->sk_sock);
1616 svc_sock_detach(xprt);
1618 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1619 svc_tcp_clear_pages(svsk);
1620 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1625 * Free the svc_sock's socket resources and the svc_sock itself.
1627 static void svc_sock_free(struct svc_xprt *xprt)
1629 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1630 struct page_frag_cache *pfc = &svsk->sk_frag_cache;
1631 struct socket *sock = svsk->sk_sock;
1633 trace_svcsock_free(svsk, sock);
1635 tls_handshake_cancel(sock->sk);
1641 __page_frag_cache_drain(virt_to_head_page(pfc->va),