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>
40 #include <net/checksum.h>
45 #include <net/tcp_states.h>
46 #include <linux/uaccess.h>
47 #include <linux/highmem.h>
48 #include <asm/ioctls.h>
50 #include <linux/sunrpc/types.h>
51 #include <linux/sunrpc/clnt.h>
52 #include <linux/sunrpc/xdr.h>
53 #include <linux/sunrpc/msg_prot.h>
54 #include <linux/sunrpc/svcsock.h>
55 #include <linux/sunrpc/stats.h>
56 #include <linux/sunrpc/xprt.h>
58 #include <trace/events/sunrpc.h>
63 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
66 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
68 static int svc_udp_recvfrom(struct svc_rqst *);
69 static int svc_udp_sendto(struct svc_rqst *);
70 static void svc_sock_detach(struct svc_xprt *);
71 static void svc_tcp_sock_detach(struct svc_xprt *);
72 static void svc_sock_free(struct svc_xprt *);
74 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
77 #ifdef CONFIG_DEBUG_LOCK_ALLOC
78 static struct lock_class_key svc_key[2];
79 static struct lock_class_key svc_slock_key[2];
81 static void svc_reclassify_socket(struct socket *sock)
83 struct sock *sk = sock->sk;
85 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
88 switch (sk->sk_family) {
90 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
92 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
99 "sk_xprt.xpt_lock-AF_INET6-NFSD",
108 static void svc_reclassify_socket(struct socket *sock)
114 * svc_tcp_release_rqst - Release transport-related resources
115 * @rqstp: request structure with resources to be released
118 static void svc_tcp_release_rqst(struct svc_rqst *rqstp)
120 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
123 struct svc_sock *svsk =
124 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
126 rqstp->rq_xprt_ctxt = NULL;
127 skb_free_datagram_locked(svsk->sk_sk, skb);
132 * svc_udp_release_rqst - Release transport-related resources
133 * @rqstp: request structure with resources to be released
136 static void svc_udp_release_rqst(struct svc_rqst *rqstp)
138 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
141 rqstp->rq_xprt_ctxt = NULL;
146 union svc_pktinfo_u {
147 struct in_pktinfo pkti;
148 struct in6_pktinfo pkti6;
150 #define SVC_PKTINFO_SPACE \
151 CMSG_SPACE(sizeof(union svc_pktinfo_u))
153 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
155 struct svc_sock *svsk =
156 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
157 switch (svsk->sk_sk->sk_family) {
159 struct in_pktinfo *pki = CMSG_DATA(cmh);
161 cmh->cmsg_level = SOL_IP;
162 cmh->cmsg_type = IP_PKTINFO;
163 pki->ipi_ifindex = 0;
164 pki->ipi_spec_dst.s_addr =
165 svc_daddr_in(rqstp)->sin_addr.s_addr;
166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
171 struct in6_pktinfo *pki = CMSG_DATA(cmh);
172 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
174 cmh->cmsg_level = SOL_IPV6;
175 cmh->cmsg_type = IPV6_PKTINFO;
176 pki->ipi6_ifindex = daddr->sin6_scope_id;
177 pki->ipi6_addr = daddr->sin6_addr;
178 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
184 static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
191 * Report socket names for nfsdfs
193 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
195 const struct sock *sk = svsk->sk_sk;
196 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
200 switch (sk->sk_family) {
202 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
204 &inet_sk(sk)->inet_rcv_saddr,
205 inet_sk(sk)->inet_num);
207 #if IS_ENABLED(CONFIG_IPV6)
209 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
211 &sk->sk_v6_rcv_saddr,
212 inet_sk(sk)->inet_num);
216 len = snprintf(buf, remaining, "*unknown-%d*\n",
220 if (len >= remaining) {
222 return -ENAMETOOLONG;
227 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
228 static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
230 struct bvec_iter bi = {
231 .bi_size = size + seek,
235 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
236 for_each_bvec(bv, bvec, bi, bi)
237 flush_dcache_page(bv.bv_page);
240 static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
247 * Read from @rqstp's transport socket. The incoming message fills whole
248 * pages in @rqstp's rq_pages array until the last page of the message
249 * has been received into a partial page.
251 static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
254 struct svc_sock *svsk =
255 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
256 struct bio_vec *bvec = rqstp->rq_bvec;
257 struct msghdr msg = { NULL };
262 rqstp->rq_xprt_hlen = 0;
264 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
266 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) {
267 bvec[i].bv_page = rqstp->rq_pages[i];
268 bvec[i].bv_len = PAGE_SIZE;
269 bvec[i].bv_offset = 0;
271 rqstp->rq_respages = &rqstp->rq_pages[i];
272 rqstp->rq_next_page = rqstp->rq_respages + 1;
274 iov_iter_bvec(&msg.msg_iter, READ, bvec, i, buflen);
276 iov_iter_advance(&msg.msg_iter, seek);
279 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
281 svc_flush_bvec(bvec, len, seek);
283 /* If we read a full record, then assume there may be more
284 * data to read (stream based sockets only!)
287 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
293 * Set socket snd and rcv buffer lengths
295 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
297 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
298 struct socket *sock = svsk->sk_sock;
300 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
303 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
304 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
305 sock->sk->sk_write_space(sock->sk);
306 release_sock(sock->sk);
309 static void svc_sock_secure_port(struct svc_rqst *rqstp)
311 if (svc_port_is_privileged(svc_addr(rqstp)))
312 set_bit(RQ_SECURE, &rqstp->rq_flags);
314 clear_bit(RQ_SECURE, &rqstp->rq_flags);
318 * INET callback when data has been received on the socket.
320 static void svc_data_ready(struct sock *sk)
322 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
325 /* Refer to svc_setup_socket() for details. */
328 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
329 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
330 svc_xprt_enqueue(&svsk->sk_xprt);
335 * INET callback when space is newly available on the socket.
337 static void svc_write_space(struct sock *sk)
339 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
342 /* Refer to svc_setup_socket() for details. */
344 trace_svcsock_write_space(&svsk->sk_xprt, 0);
345 svsk->sk_owspace(sk);
346 svc_xprt_enqueue(&svsk->sk_xprt);
350 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
352 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
354 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
356 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
359 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
361 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
363 sock_no_linger(svsk->sk_sock->sk);
367 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
369 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
372 struct in_pktinfo *pki = CMSG_DATA(cmh);
373 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
375 if (cmh->cmsg_type != IP_PKTINFO)
378 daddr->sin_family = AF_INET;
379 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
384 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
386 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
389 struct in6_pktinfo *pki = CMSG_DATA(cmh);
390 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
392 if (cmh->cmsg_type != IPV6_PKTINFO)
395 daddr->sin6_family = AF_INET6;
396 daddr->sin6_addr = pki->ipi6_addr;
397 daddr->sin6_scope_id = pki->ipi6_ifindex;
402 * Copy the UDP datagram's destination address to the rqstp structure.
403 * The 'destination' address in this case is the address to which the
404 * peer sent the datagram, i.e. our local address. For multihomed
405 * hosts, this can change from msg to msg. Note that only the IP
406 * address changes, the port number should remain the same.
408 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
411 switch (cmh->cmsg_level) {
413 return svc_udp_get_dest_address4(rqstp, cmh);
415 return svc_udp_get_dest_address6(rqstp, cmh);
422 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
423 * @rqstp: request structure into which to receive an RPC Call
425 * Called in a loop when XPT_DATA has been set.
428 * On success, the number of bytes in a received RPC Call, or
429 * %0 if a complete RPC Call message was not ready to return
431 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
433 struct svc_sock *svsk =
434 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
435 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
439 long all[SVC_PKTINFO_SPACE / sizeof(long)];
441 struct cmsghdr *cmh = &buffer.hdr;
442 struct msghdr msg = {
443 .msg_name = svc_addr(rqstp),
445 .msg_controllen = sizeof(buffer),
446 .msg_flags = MSG_DONTWAIT,
451 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
452 /* udp sockets need large rcvbuf as all pending
453 * requests are still in that buffer. sndbuf must
454 * also be large enough that there is enough space
455 * for one reply per thread. We count all threads
456 * rather than threads in a particular pool, which
457 * provides an upper bound on the number of threads
458 * which will access the socket.
460 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
462 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
463 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
464 0, 0, MSG_PEEK | MSG_DONTWAIT);
467 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
471 len = svc_addr_len(svc_addr(rqstp));
472 rqstp->rq_addrlen = len;
473 if (skb->tstamp == 0) {
474 skb->tstamp = ktime_get_real();
475 /* Don't enable netstamp, sunrpc doesn't
476 need that much accuracy */
478 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
479 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
482 rqstp->rq_arg.len = len;
483 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
485 rqstp->rq_prot = IPPROTO_UDP;
487 if (!svc_udp_get_dest_address(rqstp, cmh))
489 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
491 if (skb_is_nonlinear(skb)) {
492 /* we have to copy */
494 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
499 /* we can use it in-place */
500 rqstp->rq_arg.head[0].iov_base = skb->data;
501 rqstp->rq_arg.head[0].iov_len = len;
502 if (skb_checksum_complete(skb))
504 rqstp->rq_xprt_ctxt = skb;
507 rqstp->rq_arg.page_base = 0;
508 if (len <= rqstp->rq_arg.head[0].iov_len) {
509 rqstp->rq_arg.head[0].iov_len = len;
510 rqstp->rq_arg.page_len = 0;
511 rqstp->rq_respages = rqstp->rq_pages+1;
513 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
514 rqstp->rq_respages = rqstp->rq_pages + 1 +
515 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
517 rqstp->rq_next_page = rqstp->rq_respages+1;
520 serv->sv_stats->netudpcnt++;
522 svc_xprt_received(rqstp->rq_xprt);
526 if (err != -EAGAIN) {
527 /* possibly an icmp error */
528 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
530 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
533 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
534 cmh->cmsg_level, cmh->cmsg_type);
541 svc_xprt_received(rqstp->rq_xprt);
546 * svc_udp_sendto - Send out a reply on a UDP socket
547 * @rqstp: completed svc_rqst
549 * xpt_mutex ensures @rqstp's whole message is written to the socket
550 * without interruption.
552 * Returns the number of bytes sent, or a negative errno.
554 static int svc_udp_sendto(struct svc_rqst *rqstp)
556 struct svc_xprt *xprt = rqstp->rq_xprt;
557 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
558 struct xdr_buf *xdr = &rqstp->rq_res;
561 long all[SVC_PKTINFO_SPACE / sizeof(long)];
563 struct cmsghdr *cmh = &buffer.hdr;
564 struct msghdr msg = {
565 .msg_name = &rqstp->rq_addr,
566 .msg_namelen = rqstp->rq_addrlen,
568 .msg_controllen = sizeof(buffer),
573 svc_udp_release_rqst(rqstp);
575 svc_set_cmsg_data(rqstp, cmh);
577 mutex_lock(&xprt->xpt_mutex);
579 if (svc_xprt_is_dead(xprt))
582 err = xdr_alloc_bvec(xdr, GFP_KERNEL);
586 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
587 if (err == -ECONNREFUSED) {
588 /* ICMP error on earlier request. */
589 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
592 trace_svcsock_udp_send(xprt, err);
594 mutex_unlock(&xprt->xpt_mutex);
600 mutex_unlock(&xprt->xpt_mutex);
604 static int svc_udp_has_wspace(struct svc_xprt *xprt)
606 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
607 struct svc_serv *serv = xprt->xpt_server;
608 unsigned long required;
611 * Set the SOCK_NOSPACE flag before checking the available
614 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
615 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
616 if (required*2 > sock_wspace(svsk->sk_sk))
618 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
622 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
628 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
632 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
634 struct sockaddr *sa, int salen,
637 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
640 static const struct svc_xprt_ops svc_udp_ops = {
641 .xpo_create = svc_udp_create,
642 .xpo_recvfrom = svc_udp_recvfrom,
643 .xpo_sendto = svc_udp_sendto,
644 .xpo_result_payload = svc_sock_result_payload,
645 .xpo_release_rqst = svc_udp_release_rqst,
646 .xpo_detach = svc_sock_detach,
647 .xpo_free = svc_sock_free,
648 .xpo_has_wspace = svc_udp_has_wspace,
649 .xpo_accept = svc_udp_accept,
650 .xpo_secure_port = svc_sock_secure_port,
651 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
654 static struct svc_xprt_class svc_udp_class = {
656 .xcl_owner = THIS_MODULE,
657 .xcl_ops = &svc_udp_ops,
658 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
659 .xcl_ident = XPRT_TRANSPORT_UDP,
662 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
664 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
665 &svsk->sk_xprt, serv);
666 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
667 svsk->sk_sk->sk_data_ready = svc_data_ready;
668 svsk->sk_sk->sk_write_space = svc_write_space;
670 /* initialise setting must have enough space to
671 * receive and respond to one request.
672 * svc_udp_recvfrom will re-adjust if necessary
674 svc_sock_setbufsize(svsk, 3);
676 /* data might have come in before data_ready set up */
677 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
678 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
680 /* make sure we get destination address info */
681 switch (svsk->sk_sk->sk_family) {
683 ip_sock_set_pktinfo(svsk->sk_sock->sk);
686 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
694 * A data_ready event on a listening socket means there's a connection
695 * pending. Do not use state_change as a substitute for it.
697 static void svc_tcp_listen_data_ready(struct sock *sk)
699 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
702 /* Refer to svc_setup_socket() for details. */
708 * This callback may called twice when a new connection
709 * is established as a child socket inherits everything
710 * from a parent LISTEN socket.
711 * 1) data_ready method of the parent socket will be called
712 * when one of child sockets become ESTABLISHED.
713 * 2) data_ready method of the child socket may be called
714 * when it receives data before the socket is accepted.
715 * In case of 2, we should ignore it silently.
717 if (sk->sk_state == TCP_LISTEN) {
719 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
720 svc_xprt_enqueue(&svsk->sk_xprt);
726 * A state change on a connected socket means it's dying or dead.
728 static void svc_tcp_state_change(struct sock *sk)
730 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
733 /* Refer to svc_setup_socket() for details. */
736 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
737 if (sk->sk_state != TCP_ESTABLISHED)
738 svc_xprt_deferred_close(&svsk->sk_xprt);
743 * Accept a TCP connection
745 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
747 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
748 struct sockaddr_storage addr;
749 struct sockaddr *sin = (struct sockaddr *) &addr;
750 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
751 struct socket *sock = svsk->sk_sock;
752 struct socket *newsock;
753 struct svc_sock *newsvsk;
759 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
760 err = kernel_accept(sock, &newsock, O_NONBLOCK);
763 printk(KERN_WARNING "%s: no more sockets!\n",
765 else if (err != -EAGAIN)
766 net_warn_ratelimited("%s: accept failed (err %d)!\n",
767 serv->sv_name, -err);
768 trace_svcsock_accept_err(xprt, serv->sv_name, err);
771 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
773 err = kernel_getpeername(newsock, sin);
775 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
776 goto failed; /* aborted connection or whatever */
780 /* Reset the inherited callbacks before calling svc_setup_socket */
781 newsock->sk->sk_state_change = svsk->sk_ostate;
782 newsock->sk->sk_data_ready = svsk->sk_odata;
783 newsock->sk->sk_write_space = svsk->sk_owspace;
785 /* make sure that a write doesn't block forever when
788 newsock->sk->sk_sndtimeo = HZ*30;
790 newsvsk = svc_setup_socket(serv, newsock,
791 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
794 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
795 err = kernel_getsockname(newsock, sin);
797 if (unlikely(err < 0))
798 slen = offsetof(struct sockaddr, sa_data);
799 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
801 if (sock_is_loopback(newsock->sk))
802 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
804 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
806 serv->sv_stats->nettcpconn++;
808 return &newsvsk->sk_xprt;
811 sock_release(newsock);
815 static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
816 struct svc_rqst *rqstp)
818 size_t len = svsk->sk_datalen;
819 unsigned int i, npages;
823 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
824 for (i = 0; i < npages; i++) {
825 if (rqstp->rq_pages[i] != NULL)
826 put_page(rqstp->rq_pages[i]);
827 BUG_ON(svsk->sk_pages[i] == NULL);
828 rqstp->rq_pages[i] = svsk->sk_pages[i];
829 svsk->sk_pages[i] = NULL;
831 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
835 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
837 unsigned int i, len, npages;
839 if (svsk->sk_datalen == 0)
841 len = svsk->sk_datalen;
842 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
843 for (i = 0; i < npages; i++) {
844 svsk->sk_pages[i] = rqstp->rq_pages[i];
845 rqstp->rq_pages[i] = NULL;
849 static void svc_tcp_clear_pages(struct svc_sock *svsk)
851 unsigned int i, len, npages;
853 if (svsk->sk_datalen == 0)
855 len = svsk->sk_datalen;
856 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
857 for (i = 0; i < npages; i++) {
858 if (svsk->sk_pages[i] == NULL) {
862 put_page(svsk->sk_pages[i]);
863 svsk->sk_pages[i] = NULL;
867 svsk->sk_datalen = 0;
871 * Receive fragment record header into sk_marker.
873 static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
874 struct svc_rqst *rqstp)
878 /* If we haven't gotten the record length yet,
879 * get the next four bytes.
881 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
882 struct msghdr msg = { NULL };
885 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
886 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
888 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, want);
889 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
892 svsk->sk_tcplen += len;
894 /* call again to read the remaining bytes */
897 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
898 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
899 svsk->sk_xprt.xpt_server->sv_max_mesg)
902 return svc_sock_reclen(svsk);
905 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
906 __func__, svsk->sk_xprt.xpt_server->sv_name,
907 svc_sock_reclen(svsk));
908 svc_xprt_deferred_close(&svsk->sk_xprt);
913 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
915 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
916 struct rpc_rqst *req = NULL;
917 struct kvec *src, *dst;
918 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
927 spin_lock(&bc_xprt->queue_lock);
928 req = xprt_lookup_rqst(bc_xprt, xid);
930 goto unlock_notfound;
932 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
934 * XXX!: cheating for now! Only copying HEAD.
935 * But we know this is good enough for now (in fact, for any
936 * callback reply in the forseeable future).
938 dst = &req->rq_private_buf.head[0];
939 src = &rqstp->rq_arg.head[0];
940 if (dst->iov_len < src->iov_len)
941 goto unlock_eagain; /* whatever; just giving up. */
942 memcpy(dst->iov_base, src->iov_base, src->iov_len);
943 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
944 rqstp->rq_arg.len = 0;
945 spin_unlock(&bc_xprt->queue_lock);
949 "%s: Got unrecognized reply: "
950 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
951 __func__, ntohl(calldir),
952 bc_xprt, ntohl(xid));
954 spin_unlock(&bc_xprt->queue_lock);
958 static void svc_tcp_fragment_received(struct svc_sock *svsk)
960 /* If we have more data, signal svc_xprt_enqueue() to try again */
962 svsk->sk_marker = xdr_zero;
966 * svc_tcp_recvfrom - Receive data from a TCP socket
967 * @rqstp: request structure into which to receive an RPC Call
969 * Called in a loop when XPT_DATA has been set.
971 * Read the 4-byte stream record marker, then use the record length
972 * in that marker to set up exactly the resources needed to receive
973 * the next RPC message into @rqstp.
976 * On success, the number of bytes in a received RPC Call, or
977 * %0 if a complete RPC Call message was not ready to return
979 * The zero return case handles partial receives and callback Replies.
980 * The state of a partial receive is preserved in the svc_sock for
981 * the next call to svc_tcp_recvfrom.
983 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
985 struct svc_sock *svsk =
986 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
987 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
993 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
994 len = svc_tcp_read_marker(svsk, rqstp);
998 base = svc_tcp_restore_pages(svsk, rqstp);
999 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1000 len = svc_tcp_read_msg(rqstp, base + want, base);
1002 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1003 svsk->sk_tcplen += len;
1004 svsk->sk_datalen += len;
1006 if (len != want || !svc_sock_final_rec(svsk))
1007 goto err_incomplete;
1008 if (svsk->sk_datalen < 8)
1011 rqstp->rq_arg.len = svsk->sk_datalen;
1012 rqstp->rq_arg.page_base = 0;
1013 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1014 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1015 rqstp->rq_arg.page_len = 0;
1017 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1019 rqstp->rq_xprt_ctxt = NULL;
1020 rqstp->rq_prot = IPPROTO_TCP;
1021 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1022 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1024 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1026 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1029 len = receive_cb_reply(svsk, rqstp);
1031 /* Reset TCP read info */
1032 svsk->sk_datalen = 0;
1033 svc_tcp_fragment_received(svsk);
1038 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1040 serv->sv_stats->nettcpcnt++;
1042 svc_xprt_received(rqstp->rq_xprt);
1043 return rqstp->rq_arg.len;
1046 svc_tcp_save_pages(svsk, rqstp);
1047 if (len < 0 && len != -EAGAIN)
1050 svc_tcp_fragment_received(svsk);
1052 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1053 svc_sock_reclen(svsk),
1054 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1059 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1062 svsk->sk_datalen = 0;
1064 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1065 svc_xprt_deferred_close(&svsk->sk_xprt);
1067 svc_xprt_received(rqstp->rq_xprt);
1068 return 0; /* record not complete */
1071 static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1074 return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1075 offset_in_page(vec->iov_base),
1076 vec->iov_len, flags);
1080 * kernel_sendpage() is used exclusively to reduce the number of
1081 * copy operations in this path. Therefore the caller must ensure
1082 * that the pages backing @xdr are unchanging.
1084 * In addition, the logic assumes that * .bv_len is never larger
1087 static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
1088 rpc_fraghdr marker, unsigned int *sentp)
1090 const struct kvec *head = xdr->head;
1091 const struct kvec *tail = xdr->tail;
1093 .iov_base = &marker,
1094 .iov_len = sizeof(marker),
1096 struct msghdr msg = {
1102 ret = xdr_alloc_bvec(xdr, GFP_KERNEL);
1106 ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len);
1110 if (ret != rm.iov_len)
1113 ret = svc_tcp_send_kvec(sock, head, 0);
1117 if (ret != head->iov_len)
1120 if (xdr->page_len) {
1121 unsigned int offset, len, remaining;
1122 struct bio_vec *bvec;
1124 bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1125 offset = offset_in_page(xdr->page_base);
1126 remaining = xdr->page_len;
1127 while (remaining > 0) {
1128 len = min(remaining, bvec->bv_len - offset);
1129 ret = kernel_sendpage(sock, bvec->bv_page,
1130 bvec->bv_offset + offset,
1143 if (tail->iov_len) {
1144 ret = svc_tcp_send_kvec(sock, tail, 0);
1155 * svc_tcp_sendto - Send out a reply on a TCP socket
1156 * @rqstp: completed svc_rqst
1158 * xpt_mutex ensures @rqstp's whole message is written to the socket
1159 * without interruption.
1161 * Returns the number of bytes sent, or a negative errno.
1163 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1165 struct svc_xprt *xprt = rqstp->rq_xprt;
1166 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1167 struct xdr_buf *xdr = &rqstp->rq_res;
1168 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1173 svc_tcp_release_rqst(rqstp);
1175 atomic_inc(&svsk->sk_sendqlen);
1176 mutex_lock(&xprt->xpt_mutex);
1177 if (svc_xprt_is_dead(xprt))
1179 tcp_sock_set_cork(svsk->sk_sk, true);
1180 err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent);
1182 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1183 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1185 if (atomic_dec_and_test(&svsk->sk_sendqlen))
1186 tcp_sock_set_cork(svsk->sk_sk, false);
1187 mutex_unlock(&xprt->xpt_mutex);
1191 atomic_dec(&svsk->sk_sendqlen);
1192 mutex_unlock(&xprt->xpt_mutex);
1195 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1196 xprt->xpt_server->sv_name,
1197 (err < 0) ? "got error" : "sent",
1198 (err < 0) ? err : sent, xdr->len);
1199 svc_xprt_deferred_close(xprt);
1200 atomic_dec(&svsk->sk_sendqlen);
1201 mutex_unlock(&xprt->xpt_mutex);
1205 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1207 struct sockaddr *sa, int salen,
1210 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1213 static const struct svc_xprt_ops svc_tcp_ops = {
1214 .xpo_create = svc_tcp_create,
1215 .xpo_recvfrom = svc_tcp_recvfrom,
1216 .xpo_sendto = svc_tcp_sendto,
1217 .xpo_result_payload = svc_sock_result_payload,
1218 .xpo_release_rqst = svc_tcp_release_rqst,
1219 .xpo_detach = svc_tcp_sock_detach,
1220 .xpo_free = svc_sock_free,
1221 .xpo_has_wspace = svc_tcp_has_wspace,
1222 .xpo_accept = svc_tcp_accept,
1223 .xpo_secure_port = svc_sock_secure_port,
1224 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1227 static struct svc_xprt_class svc_tcp_class = {
1229 .xcl_owner = THIS_MODULE,
1230 .xcl_ops = &svc_tcp_ops,
1231 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1232 .xcl_ident = XPRT_TRANSPORT_TCP,
1235 void svc_init_xprt_sock(void)
1237 svc_reg_xprt_class(&svc_tcp_class);
1238 svc_reg_xprt_class(&svc_udp_class);
1241 void svc_cleanup_xprt_sock(void)
1243 svc_unreg_xprt_class(&svc_tcp_class);
1244 svc_unreg_xprt_class(&svc_udp_class);
1247 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1249 struct sock *sk = svsk->sk_sk;
1251 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1252 &svsk->sk_xprt, serv);
1253 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1254 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1255 if (sk->sk_state == TCP_LISTEN) {
1256 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1257 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1258 sk->sk_data_ready = svc_tcp_listen_data_ready;
1259 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1261 sk->sk_state_change = svc_tcp_state_change;
1262 sk->sk_data_ready = svc_data_ready;
1263 sk->sk_write_space = svc_write_space;
1265 svsk->sk_marker = xdr_zero;
1266 svsk->sk_tcplen = 0;
1267 svsk->sk_datalen = 0;
1268 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1270 tcp_sock_set_nodelay(sk);
1272 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1273 switch (sk->sk_state) {
1275 case TCP_ESTABLISHED:
1278 svc_xprt_deferred_close(&svsk->sk_xprt);
1283 void svc_sock_update_bufs(struct svc_serv *serv)
1286 * The number of server threads has changed. Update
1287 * rcvbuf and sndbuf accordingly on all sockets
1289 struct svc_sock *svsk;
1291 spin_lock_bh(&serv->sv_lock);
1292 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1293 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1294 spin_unlock_bh(&serv->sv_lock);
1296 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1299 * Initialize socket for RPC use and create svc_sock struct
1301 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1302 struct socket *sock,
1305 struct svc_sock *svsk;
1307 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1310 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1312 return ERR_PTR(-ENOMEM);
1316 /* Register socket with portmapper */
1318 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1320 ntohs(inet_sk(inet)->inet_sport));
1324 return ERR_PTR(err);
1327 svsk->sk_sock = sock;
1329 svsk->sk_ostate = inet->sk_state_change;
1330 svsk->sk_odata = inet->sk_data_ready;
1331 svsk->sk_owspace = inet->sk_write_space;
1333 * This barrier is necessary in order to prevent race condition
1334 * with svc_data_ready(), svc_listen_data_ready() and others
1335 * when calling callbacks above.
1338 inet->sk_user_data = svsk;
1340 /* Initialize the socket */
1341 if (sock->type == SOCK_DGRAM)
1342 svc_udp_init(svsk, serv);
1344 svc_tcp_init(svsk, serv);
1346 trace_svcsock_new_socket(sock);
1350 bool svc_alien_sock(struct net *net, int fd)
1353 struct socket *sock = sockfd_lookup(fd, &err);
1358 if (sock_net(sock->sk) != net)
1364 EXPORT_SYMBOL_GPL(svc_alien_sock);
1367 * svc_addsock - add a listener socket to an RPC service
1368 * @serv: pointer to RPC service to which to add a new listener
1369 * @fd: file descriptor of the new listener
1370 * @name_return: pointer to buffer to fill in with name of listener
1371 * @len: size of the buffer
1374 * Fills in socket name and returns positive length of name if successful.
1375 * Name is terminated with '\n'. On error, returns a negative errno
1378 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1379 const size_t len, const struct cred *cred)
1382 struct socket *so = sockfd_lookup(fd, &err);
1383 struct svc_sock *svsk = NULL;
1384 struct sockaddr_storage addr;
1385 struct sockaddr *sin = (struct sockaddr *)&addr;
1390 err = -EAFNOSUPPORT;
1391 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1393 err = -EPROTONOSUPPORT;
1394 if (so->sk->sk_protocol != IPPROTO_TCP &&
1395 so->sk->sk_protocol != IPPROTO_UDP)
1398 if (so->state > SS_UNCONNECTED)
1401 if (!try_module_get(THIS_MODULE))
1403 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1405 module_put(THIS_MODULE);
1406 err = PTR_ERR(svsk);
1409 salen = kernel_getsockname(svsk->sk_sock, sin);
1411 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1412 svsk->sk_xprt.xpt_cred = get_cred(cred);
1413 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1414 return svc_one_sock_name(svsk, name_return, len);
1419 EXPORT_SYMBOL_GPL(svc_addsock);
1422 * Create socket for RPC service.
1424 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1427 struct sockaddr *sin, int len,
1430 struct svc_sock *svsk;
1431 struct socket *sock;
1434 struct sockaddr_storage addr;
1435 struct sockaddr *newsin = (struct sockaddr *)&addr;
1439 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1440 printk(KERN_WARNING "svc: only UDP and TCP "
1441 "sockets supported\n");
1442 return ERR_PTR(-EINVAL);
1445 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1446 switch (sin->sa_family) {
1454 return ERR_PTR(-EINVAL);
1457 error = __sock_create(net, family, type, protocol, &sock, 1);
1459 return ERR_PTR(error);
1461 svc_reclassify_socket(sock);
1464 * If this is an PF_INET6 listener, we want to avoid
1465 * getting requests from IPv4 remotes. Those should
1466 * be shunted to a PF_INET listener via rpcbind.
1468 if (family == PF_INET6)
1469 ip6_sock_set_v6only(sock->sk);
1470 if (type == SOCK_STREAM)
1471 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1472 error = kernel_bind(sock, sin, len);
1476 error = kernel_getsockname(sock, newsin);
1481 if (protocol == IPPROTO_TCP) {
1482 if ((error = kernel_listen(sock, 64)) < 0)
1486 svsk = svc_setup_socket(serv, sock, flags);
1488 error = PTR_ERR(svsk);
1491 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1492 return (struct svc_xprt *)svsk;
1495 return ERR_PTR(error);
1499 * Detach the svc_sock from the socket so that no
1500 * more callbacks occur.
1502 static void svc_sock_detach(struct svc_xprt *xprt)
1504 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1505 struct sock *sk = svsk->sk_sk;
1507 /* put back the old socket callbacks */
1509 sk->sk_state_change = svsk->sk_ostate;
1510 sk->sk_data_ready = svsk->sk_odata;
1511 sk->sk_write_space = svsk->sk_owspace;
1512 sk->sk_user_data = NULL;
1517 * Disconnect the socket, and reset the callbacks
1519 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1521 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1523 svc_sock_detach(xprt);
1525 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1526 svc_tcp_clear_pages(svsk);
1527 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1532 * Free the svc_sock's socket resources and the svc_sock itself.
1534 static void svc_sock_free(struct svc_xprt *xprt)
1536 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1538 if (svsk->sk_sock->file)
1539 sockfd_put(svsk->sk_sock);
1541 sock_release(svsk->sk_sock);