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_ctxt - Release transport-related resources
115 * @xprt: the transport which owned the context
116 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
119 static void svc_tcp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
124 * svc_udp_release_ctxt - Release transport-related resources
125 * @xprt: the transport which owned the context
126 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
129 static void svc_udp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
131 struct sk_buff *skb = ctxt;
137 union svc_pktinfo_u {
138 struct in_pktinfo pkti;
139 struct in6_pktinfo pkti6;
141 #define SVC_PKTINFO_SPACE \
142 CMSG_SPACE(sizeof(union svc_pktinfo_u))
144 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
146 struct svc_sock *svsk =
147 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
148 switch (svsk->sk_sk->sk_family) {
150 struct in_pktinfo *pki = CMSG_DATA(cmh);
152 cmh->cmsg_level = SOL_IP;
153 cmh->cmsg_type = IP_PKTINFO;
154 pki->ipi_ifindex = 0;
155 pki->ipi_spec_dst.s_addr =
156 svc_daddr_in(rqstp)->sin_addr.s_addr;
157 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
162 struct in6_pktinfo *pki = CMSG_DATA(cmh);
163 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
165 cmh->cmsg_level = SOL_IPV6;
166 cmh->cmsg_type = IPV6_PKTINFO;
167 pki->ipi6_ifindex = daddr->sin6_scope_id;
168 pki->ipi6_addr = daddr->sin6_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
175 static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
182 * Report socket names for nfsdfs
184 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
186 const struct sock *sk = svsk->sk_sk;
187 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
191 switch (sk->sk_family) {
193 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
195 &inet_sk(sk)->inet_rcv_saddr,
196 inet_sk(sk)->inet_num);
198 #if IS_ENABLED(CONFIG_IPV6)
200 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
202 &sk->sk_v6_rcv_saddr,
203 inet_sk(sk)->inet_num);
207 len = snprintf(buf, remaining, "*unknown-%d*\n",
211 if (len >= remaining) {
213 return -ENAMETOOLONG;
218 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
219 static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
221 struct bvec_iter bi = {
222 .bi_size = size + seek,
226 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
227 for_each_bvec(bv, bvec, bi, bi)
228 flush_dcache_page(bv.bv_page);
231 static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
238 * Read from @rqstp's transport socket. The incoming message fills whole
239 * pages in @rqstp's rq_pages array until the last page of the message
240 * has been received into a partial page.
242 static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
245 struct svc_sock *svsk =
246 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
247 struct bio_vec *bvec = rqstp->rq_bvec;
248 struct msghdr msg = { NULL };
253 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
255 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) {
256 bvec[i].bv_page = rqstp->rq_pages[i];
257 bvec[i].bv_len = PAGE_SIZE;
258 bvec[i].bv_offset = 0;
260 rqstp->rq_respages = &rqstp->rq_pages[i];
261 rqstp->rq_next_page = rqstp->rq_respages + 1;
263 iov_iter_bvec(&msg.msg_iter, READ, bvec, i, buflen);
265 iov_iter_advance(&msg.msg_iter, seek);
268 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
270 svc_flush_bvec(bvec, len, seek);
272 /* If we read a full record, then assume there may be more
273 * data to read (stream based sockets only!)
276 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
282 * Set socket snd and rcv buffer lengths
284 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
286 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
287 struct socket *sock = svsk->sk_sock;
289 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
292 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
293 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
294 sock->sk->sk_write_space(sock->sk);
295 release_sock(sock->sk);
298 static void svc_sock_secure_port(struct svc_rqst *rqstp)
300 if (svc_port_is_privileged(svc_addr(rqstp)))
301 set_bit(RQ_SECURE, &rqstp->rq_flags);
303 clear_bit(RQ_SECURE, &rqstp->rq_flags);
307 * INET callback when data has been received on the socket.
309 static void svc_data_ready(struct sock *sk)
311 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
314 /* Refer to svc_setup_socket() for details. */
317 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
318 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
319 svc_xprt_enqueue(&svsk->sk_xprt);
324 * INET callback when space is newly available on the socket.
326 static void svc_write_space(struct sock *sk)
328 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
331 /* Refer to svc_setup_socket() for details. */
333 trace_svcsock_write_space(&svsk->sk_xprt, 0);
334 svsk->sk_owspace(sk);
335 svc_xprt_enqueue(&svsk->sk_xprt);
339 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
341 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
343 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
345 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
348 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
350 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
352 sock_no_linger(svsk->sk_sock->sk);
356 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
358 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
361 struct in_pktinfo *pki = CMSG_DATA(cmh);
362 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
364 if (cmh->cmsg_type != IP_PKTINFO)
367 daddr->sin_family = AF_INET;
368 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
373 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
375 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
378 struct in6_pktinfo *pki = CMSG_DATA(cmh);
379 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
381 if (cmh->cmsg_type != IPV6_PKTINFO)
384 daddr->sin6_family = AF_INET6;
385 daddr->sin6_addr = pki->ipi6_addr;
386 daddr->sin6_scope_id = pki->ipi6_ifindex;
391 * Copy the UDP datagram's destination address to the rqstp structure.
392 * The 'destination' address in this case is the address to which the
393 * peer sent the datagram, i.e. our local address. For multihomed
394 * hosts, this can change from msg to msg. Note that only the IP
395 * address changes, the port number should remain the same.
397 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
400 switch (cmh->cmsg_level) {
402 return svc_udp_get_dest_address4(rqstp, cmh);
404 return svc_udp_get_dest_address6(rqstp, cmh);
411 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
412 * @rqstp: request structure into which to receive an RPC Call
414 * Called in a loop when XPT_DATA has been set.
417 * On success, the number of bytes in a received RPC Call, or
418 * %0 if a complete RPC Call message was not ready to return
420 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
422 struct svc_sock *svsk =
423 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
424 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
428 long all[SVC_PKTINFO_SPACE / sizeof(long)];
430 struct cmsghdr *cmh = &buffer.hdr;
431 struct msghdr msg = {
432 .msg_name = svc_addr(rqstp),
434 .msg_controllen = sizeof(buffer),
435 .msg_flags = MSG_DONTWAIT,
440 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
441 /* udp sockets need large rcvbuf as all pending
442 * requests are still in that buffer. sndbuf must
443 * also be large enough that there is enough space
444 * for one reply per thread. We count all threads
445 * rather than threads in a particular pool, which
446 * provides an upper bound on the number of threads
447 * which will access the socket.
449 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
451 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
452 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
453 0, 0, MSG_PEEK | MSG_DONTWAIT);
456 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
460 len = svc_addr_len(svc_addr(rqstp));
461 rqstp->rq_addrlen = len;
462 if (skb->tstamp == 0) {
463 skb->tstamp = ktime_get_real();
464 /* Don't enable netstamp, sunrpc doesn't
465 need that much accuracy */
467 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
468 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
471 rqstp->rq_arg.len = len;
472 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
474 rqstp->rq_prot = IPPROTO_UDP;
476 if (!svc_udp_get_dest_address(rqstp, cmh))
478 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
480 if (skb_is_nonlinear(skb)) {
481 /* we have to copy */
483 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
488 /* we can use it in-place */
489 rqstp->rq_arg.head[0].iov_base = skb->data;
490 rqstp->rq_arg.head[0].iov_len = len;
491 if (skb_checksum_complete(skb))
493 rqstp->rq_xprt_ctxt = skb;
496 rqstp->rq_arg.page_base = 0;
497 if (len <= rqstp->rq_arg.head[0].iov_len) {
498 rqstp->rq_arg.head[0].iov_len = len;
499 rqstp->rq_arg.page_len = 0;
500 rqstp->rq_respages = rqstp->rq_pages+1;
502 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
503 rqstp->rq_respages = rqstp->rq_pages + 1 +
504 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
506 rqstp->rq_next_page = rqstp->rq_respages+1;
509 serv->sv_stats->netudpcnt++;
511 svc_xprt_received(rqstp->rq_xprt);
515 if (err != -EAGAIN) {
516 /* possibly an icmp error */
517 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
519 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
522 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
523 cmh->cmsg_level, cmh->cmsg_type);
530 svc_xprt_received(rqstp->rq_xprt);
535 * svc_udp_sendto - Send out a reply on a UDP socket
536 * @rqstp: completed svc_rqst
538 * xpt_mutex ensures @rqstp's whole message is written to the socket
539 * without interruption.
541 * Returns the number of bytes sent, or a negative errno.
543 static int svc_udp_sendto(struct svc_rqst *rqstp)
545 struct svc_xprt *xprt = rqstp->rq_xprt;
546 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
547 struct xdr_buf *xdr = &rqstp->rq_res;
550 long all[SVC_PKTINFO_SPACE / sizeof(long)];
552 struct cmsghdr *cmh = &buffer.hdr;
553 struct msghdr msg = {
554 .msg_name = &rqstp->rq_addr,
555 .msg_namelen = rqstp->rq_addrlen,
557 .msg_controllen = sizeof(buffer),
562 svc_udp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
563 rqstp->rq_xprt_ctxt = NULL;
565 svc_set_cmsg_data(rqstp, cmh);
567 mutex_lock(&xprt->xpt_mutex);
569 if (svc_xprt_is_dead(xprt))
572 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
574 if (err == -ECONNREFUSED) {
575 /* ICMP error on earlier request. */
576 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
579 trace_svcsock_udp_send(xprt, err);
581 mutex_unlock(&xprt->xpt_mutex);
587 mutex_unlock(&xprt->xpt_mutex);
591 static int svc_udp_has_wspace(struct svc_xprt *xprt)
593 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
594 struct svc_serv *serv = xprt->xpt_server;
595 unsigned long required;
598 * Set the SOCK_NOSPACE flag before checking the available
601 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
602 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
603 if (required*2 > sock_wspace(svsk->sk_sk))
605 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
609 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
615 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
619 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
621 struct sockaddr *sa, int salen,
624 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
627 static const struct svc_xprt_ops svc_udp_ops = {
628 .xpo_create = svc_udp_create,
629 .xpo_recvfrom = svc_udp_recvfrom,
630 .xpo_sendto = svc_udp_sendto,
631 .xpo_result_payload = svc_sock_result_payload,
632 .xpo_release_ctxt = svc_udp_release_ctxt,
633 .xpo_detach = svc_sock_detach,
634 .xpo_free = svc_sock_free,
635 .xpo_has_wspace = svc_udp_has_wspace,
636 .xpo_accept = svc_udp_accept,
637 .xpo_secure_port = svc_sock_secure_port,
638 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
641 static struct svc_xprt_class svc_udp_class = {
643 .xcl_owner = THIS_MODULE,
644 .xcl_ops = &svc_udp_ops,
645 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
646 .xcl_ident = XPRT_TRANSPORT_UDP,
649 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
651 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
652 &svsk->sk_xprt, serv);
653 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
654 svsk->sk_sk->sk_data_ready = svc_data_ready;
655 svsk->sk_sk->sk_write_space = svc_write_space;
657 /* initialise setting must have enough space to
658 * receive and respond to one request.
659 * svc_udp_recvfrom will re-adjust if necessary
661 svc_sock_setbufsize(svsk, 3);
663 /* data might have come in before data_ready set up */
664 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
665 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
667 /* make sure we get destination address info */
668 switch (svsk->sk_sk->sk_family) {
670 ip_sock_set_pktinfo(svsk->sk_sock->sk);
673 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
681 * A data_ready event on a listening socket means there's a connection
682 * pending. Do not use state_change as a substitute for it.
684 static void svc_tcp_listen_data_ready(struct sock *sk)
686 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
689 * This callback may called twice when a new connection
690 * is established as a child socket inherits everything
691 * from a parent LISTEN socket.
692 * 1) data_ready method of the parent socket will be called
693 * when one of child sockets become ESTABLISHED.
694 * 2) data_ready method of the child socket may be called
695 * when it receives data before the socket is accepted.
696 * In case of 2, we should ignore it silently and DO NOT
699 if (sk->sk_state != TCP_LISTEN)
703 /* Refer to svc_setup_socket() for details. */
706 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
707 svc_xprt_enqueue(&svsk->sk_xprt);
712 * A state change on a connected socket means it's dying or dead.
714 static void svc_tcp_state_change(struct sock *sk)
716 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
719 /* Refer to svc_setup_socket() for details. */
722 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
723 if (sk->sk_state != TCP_ESTABLISHED)
724 svc_xprt_deferred_close(&svsk->sk_xprt);
729 * Accept a TCP connection
731 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
733 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
734 struct sockaddr_storage addr;
735 struct sockaddr *sin = (struct sockaddr *) &addr;
736 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
737 struct socket *sock = svsk->sk_sock;
738 struct socket *newsock;
739 struct svc_sock *newsvsk;
745 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
746 err = kernel_accept(sock, &newsock, O_NONBLOCK);
749 printk(KERN_WARNING "%s: no more sockets!\n",
751 else if (err != -EAGAIN)
752 net_warn_ratelimited("%s: accept failed (err %d)!\n",
753 serv->sv_name, -err);
754 trace_svcsock_accept_err(xprt, serv->sv_name, err);
757 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
759 err = kernel_getpeername(newsock, sin);
761 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
762 goto failed; /* aborted connection or whatever */
766 /* Reset the inherited callbacks before calling svc_setup_socket */
767 newsock->sk->sk_state_change = svsk->sk_ostate;
768 newsock->sk->sk_data_ready = svsk->sk_odata;
769 newsock->sk->sk_write_space = svsk->sk_owspace;
771 /* make sure that a write doesn't block forever when
774 newsock->sk->sk_sndtimeo = HZ*30;
776 newsvsk = svc_setup_socket(serv, newsock,
777 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
780 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
781 err = kernel_getsockname(newsock, sin);
783 if (unlikely(err < 0))
784 slen = offsetof(struct sockaddr, sa_data);
785 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
787 if (sock_is_loopback(newsock->sk))
788 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
790 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
792 serv->sv_stats->nettcpconn++;
794 return &newsvsk->sk_xprt;
797 sock_release(newsock);
801 static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
802 struct svc_rqst *rqstp)
804 size_t len = svsk->sk_datalen;
805 unsigned int i, npages;
809 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
810 for (i = 0; i < npages; i++) {
811 if (rqstp->rq_pages[i] != NULL)
812 put_page(rqstp->rq_pages[i]);
813 BUG_ON(svsk->sk_pages[i] == NULL);
814 rqstp->rq_pages[i] = svsk->sk_pages[i];
815 svsk->sk_pages[i] = NULL;
817 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
821 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
823 unsigned int i, len, npages;
825 if (svsk->sk_datalen == 0)
827 len = svsk->sk_datalen;
828 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
829 for (i = 0; i < npages; i++) {
830 svsk->sk_pages[i] = rqstp->rq_pages[i];
831 rqstp->rq_pages[i] = NULL;
835 static void svc_tcp_clear_pages(struct svc_sock *svsk)
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 if (svsk->sk_pages[i] == NULL) {
848 put_page(svsk->sk_pages[i]);
849 svsk->sk_pages[i] = NULL;
853 svsk->sk_datalen = 0;
857 * Receive fragment record header into sk_marker.
859 static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
860 struct svc_rqst *rqstp)
864 /* If we haven't gotten the record length yet,
865 * get the next four bytes.
867 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
868 struct msghdr msg = { NULL };
871 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
872 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
874 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, want);
875 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
878 svsk->sk_tcplen += len;
880 /* call again to read the remaining bytes */
883 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
884 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
885 svsk->sk_xprt.xpt_server->sv_max_mesg)
888 return svc_sock_reclen(svsk);
891 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
892 __func__, svsk->sk_xprt.xpt_server->sv_name,
893 svc_sock_reclen(svsk));
894 svc_xprt_deferred_close(&svsk->sk_xprt);
899 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
901 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
902 struct rpc_rqst *req = NULL;
903 struct kvec *src, *dst;
904 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
913 spin_lock(&bc_xprt->queue_lock);
914 req = xprt_lookup_rqst(bc_xprt, xid);
916 goto unlock_notfound;
918 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
920 * XXX!: cheating for now! Only copying HEAD.
921 * But we know this is good enough for now (in fact, for any
922 * callback reply in the forseeable future).
924 dst = &req->rq_private_buf.head[0];
925 src = &rqstp->rq_arg.head[0];
926 if (dst->iov_len < src->iov_len)
927 goto unlock_eagain; /* whatever; just giving up. */
928 memcpy(dst->iov_base, src->iov_base, src->iov_len);
929 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
930 rqstp->rq_arg.len = 0;
931 spin_unlock(&bc_xprt->queue_lock);
935 "%s: Got unrecognized reply: "
936 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
937 __func__, ntohl(calldir),
938 bc_xprt, ntohl(xid));
940 spin_unlock(&bc_xprt->queue_lock);
944 static void svc_tcp_fragment_received(struct svc_sock *svsk)
946 /* If we have more data, signal svc_xprt_enqueue() to try again */
948 svsk->sk_marker = xdr_zero;
952 * svc_tcp_recvfrom - Receive data from a TCP socket
953 * @rqstp: request structure into which to receive an RPC Call
955 * Called in a loop when XPT_DATA has been set.
957 * Read the 4-byte stream record marker, then use the record length
958 * in that marker to set up exactly the resources needed to receive
959 * the next RPC message into @rqstp.
962 * On success, the number of bytes in a received RPC Call, or
963 * %0 if a complete RPC Call message was not ready to return
965 * The zero return case handles partial receives and callback Replies.
966 * The state of a partial receive is preserved in the svc_sock for
967 * the next call to svc_tcp_recvfrom.
969 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
971 struct svc_sock *svsk =
972 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
973 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
979 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
980 len = svc_tcp_read_marker(svsk, rqstp);
984 base = svc_tcp_restore_pages(svsk, rqstp);
985 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
986 len = svc_tcp_read_msg(rqstp, base + want, base);
988 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
989 svsk->sk_tcplen += len;
990 svsk->sk_datalen += len;
992 if (len != want || !svc_sock_final_rec(svsk))
994 if (svsk->sk_datalen < 8)
997 rqstp->rq_arg.len = svsk->sk_datalen;
998 rqstp->rq_arg.page_base = 0;
999 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1000 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1001 rqstp->rq_arg.page_len = 0;
1003 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1005 rqstp->rq_xprt_ctxt = NULL;
1006 rqstp->rq_prot = IPPROTO_TCP;
1007 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1008 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1010 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1012 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1015 len = receive_cb_reply(svsk, rqstp);
1017 /* Reset TCP read info */
1018 svsk->sk_datalen = 0;
1019 svc_tcp_fragment_received(svsk);
1024 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1026 serv->sv_stats->nettcpcnt++;
1028 svc_xprt_received(rqstp->rq_xprt);
1029 return rqstp->rq_arg.len;
1032 svc_tcp_save_pages(svsk, rqstp);
1033 if (len < 0 && len != -EAGAIN)
1036 svc_tcp_fragment_received(svsk);
1038 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1039 svc_sock_reclen(svsk),
1040 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1045 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1048 svsk->sk_datalen = 0;
1050 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1051 svc_xprt_deferred_close(&svsk->sk_xprt);
1053 svc_xprt_received(rqstp->rq_xprt);
1054 return 0; /* record not complete */
1057 static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1060 return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1061 offset_in_page(vec->iov_base),
1062 vec->iov_len, flags);
1066 * kernel_sendpage() is used exclusively to reduce the number of
1067 * copy operations in this path. Therefore the caller must ensure
1068 * that the pages backing @xdr are unchanging.
1070 * In addition, the logic assumes that * .bv_len is never larger
1073 static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
1074 rpc_fraghdr marker, unsigned int *sentp)
1076 const struct kvec *head = xdr->head;
1077 const struct kvec *tail = xdr->tail;
1079 .iov_base = &marker,
1080 .iov_len = sizeof(marker),
1082 struct msghdr msg = {
1088 ret = xdr_alloc_bvec(xdr, GFP_KERNEL);
1092 ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len);
1096 if (ret != rm.iov_len)
1099 ret = svc_tcp_send_kvec(sock, head, 0);
1103 if (ret != head->iov_len)
1106 if (xdr->page_len) {
1107 unsigned int offset, len, remaining;
1108 struct bio_vec *bvec;
1110 bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1111 offset = offset_in_page(xdr->page_base);
1112 remaining = xdr->page_len;
1113 while (remaining > 0) {
1114 len = min(remaining, bvec->bv_len - offset);
1115 ret = kernel_sendpage(sock, bvec->bv_page,
1116 bvec->bv_offset + offset,
1129 if (tail->iov_len) {
1130 ret = svc_tcp_send_kvec(sock, tail, 0);
1141 * svc_tcp_sendto - Send out a reply on a TCP socket
1142 * @rqstp: completed svc_rqst
1144 * xpt_mutex ensures @rqstp's whole message is written to the socket
1145 * without interruption.
1147 * Returns the number of bytes sent, or a negative errno.
1149 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1151 struct svc_xprt *xprt = rqstp->rq_xprt;
1152 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1153 struct xdr_buf *xdr = &rqstp->rq_res;
1154 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1159 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1160 rqstp->rq_xprt_ctxt = NULL;
1162 atomic_inc(&svsk->sk_sendqlen);
1163 mutex_lock(&xprt->xpt_mutex);
1164 if (svc_xprt_is_dead(xprt))
1166 tcp_sock_set_cork(svsk->sk_sk, true);
1167 err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent);
1169 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1170 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1172 if (atomic_dec_and_test(&svsk->sk_sendqlen))
1173 tcp_sock_set_cork(svsk->sk_sk, false);
1174 mutex_unlock(&xprt->xpt_mutex);
1178 atomic_dec(&svsk->sk_sendqlen);
1179 mutex_unlock(&xprt->xpt_mutex);
1182 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1183 xprt->xpt_server->sv_name,
1184 (err < 0) ? "got error" : "sent",
1185 (err < 0) ? err : sent, xdr->len);
1186 svc_xprt_deferred_close(xprt);
1187 atomic_dec(&svsk->sk_sendqlen);
1188 mutex_unlock(&xprt->xpt_mutex);
1192 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1194 struct sockaddr *sa, int salen,
1197 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1200 static const struct svc_xprt_ops svc_tcp_ops = {
1201 .xpo_create = svc_tcp_create,
1202 .xpo_recvfrom = svc_tcp_recvfrom,
1203 .xpo_sendto = svc_tcp_sendto,
1204 .xpo_result_payload = svc_sock_result_payload,
1205 .xpo_release_ctxt = svc_tcp_release_ctxt,
1206 .xpo_detach = svc_tcp_sock_detach,
1207 .xpo_free = svc_sock_free,
1208 .xpo_has_wspace = svc_tcp_has_wspace,
1209 .xpo_accept = svc_tcp_accept,
1210 .xpo_secure_port = svc_sock_secure_port,
1211 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1214 static struct svc_xprt_class svc_tcp_class = {
1216 .xcl_owner = THIS_MODULE,
1217 .xcl_ops = &svc_tcp_ops,
1218 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1219 .xcl_ident = XPRT_TRANSPORT_TCP,
1222 void svc_init_xprt_sock(void)
1224 svc_reg_xprt_class(&svc_tcp_class);
1225 svc_reg_xprt_class(&svc_udp_class);
1228 void svc_cleanup_xprt_sock(void)
1230 svc_unreg_xprt_class(&svc_tcp_class);
1231 svc_unreg_xprt_class(&svc_udp_class);
1234 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1236 struct sock *sk = svsk->sk_sk;
1238 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1239 &svsk->sk_xprt, serv);
1240 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1241 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1242 if (sk->sk_state == TCP_LISTEN) {
1243 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1244 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1245 sk->sk_data_ready = svc_tcp_listen_data_ready;
1246 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1248 sk->sk_state_change = svc_tcp_state_change;
1249 sk->sk_data_ready = svc_data_ready;
1250 sk->sk_write_space = svc_write_space;
1252 svsk->sk_marker = xdr_zero;
1253 svsk->sk_tcplen = 0;
1254 svsk->sk_datalen = 0;
1255 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1257 tcp_sock_set_nodelay(sk);
1259 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1260 switch (sk->sk_state) {
1262 case TCP_ESTABLISHED:
1265 svc_xprt_deferred_close(&svsk->sk_xprt);
1270 void svc_sock_update_bufs(struct svc_serv *serv)
1273 * The number of server threads has changed. Update
1274 * rcvbuf and sndbuf accordingly on all sockets
1276 struct svc_sock *svsk;
1278 spin_lock_bh(&serv->sv_lock);
1279 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1280 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1281 spin_unlock_bh(&serv->sv_lock);
1283 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1286 * Initialize socket for RPC use and create svc_sock struct
1288 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1289 struct socket *sock,
1292 struct svc_sock *svsk;
1294 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1297 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1299 return ERR_PTR(-ENOMEM);
1303 /* Register socket with portmapper */
1305 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1307 ntohs(inet_sk(inet)->inet_sport));
1311 return ERR_PTR(err);
1314 svsk->sk_sock = sock;
1316 svsk->sk_ostate = inet->sk_state_change;
1317 svsk->sk_odata = inet->sk_data_ready;
1318 svsk->sk_owspace = inet->sk_write_space;
1320 * This barrier is necessary in order to prevent race condition
1321 * with svc_data_ready(), svc_listen_data_ready() and others
1322 * when calling callbacks above.
1325 inet->sk_user_data = svsk;
1327 /* Initialize the socket */
1328 if (sock->type == SOCK_DGRAM)
1329 svc_udp_init(svsk, serv);
1331 svc_tcp_init(svsk, serv);
1333 trace_svcsock_new_socket(sock);
1337 bool svc_alien_sock(struct net *net, int fd)
1340 struct socket *sock = sockfd_lookup(fd, &err);
1345 if (sock_net(sock->sk) != net)
1351 EXPORT_SYMBOL_GPL(svc_alien_sock);
1354 * svc_addsock - add a listener socket to an RPC service
1355 * @serv: pointer to RPC service to which to add a new listener
1356 * @fd: file descriptor of the new listener
1357 * @name_return: pointer to buffer to fill in with name of listener
1358 * @len: size of the buffer
1361 * Fills in socket name and returns positive length of name if successful.
1362 * Name is terminated with '\n'. On error, returns a negative errno
1365 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1366 const size_t len, const struct cred *cred)
1369 struct socket *so = sockfd_lookup(fd, &err);
1370 struct svc_sock *svsk = NULL;
1371 struct sockaddr_storage addr;
1372 struct sockaddr *sin = (struct sockaddr *)&addr;
1377 err = -EAFNOSUPPORT;
1378 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1380 err = -EPROTONOSUPPORT;
1381 if (so->sk->sk_protocol != IPPROTO_TCP &&
1382 so->sk->sk_protocol != IPPROTO_UDP)
1385 if (so->state > SS_UNCONNECTED)
1388 if (!try_module_get(THIS_MODULE))
1390 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1392 module_put(THIS_MODULE);
1393 err = PTR_ERR(svsk);
1396 salen = kernel_getsockname(svsk->sk_sock, sin);
1398 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1399 svsk->sk_xprt.xpt_cred = get_cred(cred);
1400 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1401 return svc_one_sock_name(svsk, name_return, len);
1406 EXPORT_SYMBOL_GPL(svc_addsock);
1409 * Create socket for RPC service.
1411 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1414 struct sockaddr *sin, int len,
1417 struct svc_sock *svsk;
1418 struct socket *sock;
1421 struct sockaddr_storage addr;
1422 struct sockaddr *newsin = (struct sockaddr *)&addr;
1426 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1427 printk(KERN_WARNING "svc: only UDP and TCP "
1428 "sockets supported\n");
1429 return ERR_PTR(-EINVAL);
1432 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1433 switch (sin->sa_family) {
1441 return ERR_PTR(-EINVAL);
1444 error = __sock_create(net, family, type, protocol, &sock, 1);
1446 return ERR_PTR(error);
1448 svc_reclassify_socket(sock);
1451 * If this is an PF_INET6 listener, we want to avoid
1452 * getting requests from IPv4 remotes. Those should
1453 * be shunted to a PF_INET listener via rpcbind.
1455 if (family == PF_INET6)
1456 ip6_sock_set_v6only(sock->sk);
1457 if (type == SOCK_STREAM)
1458 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1459 error = kernel_bind(sock, sin, len);
1463 error = kernel_getsockname(sock, newsin);
1468 if (protocol == IPPROTO_TCP) {
1469 if ((error = kernel_listen(sock, 64)) < 0)
1473 svsk = svc_setup_socket(serv, sock, flags);
1475 error = PTR_ERR(svsk);
1478 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1479 return (struct svc_xprt *)svsk;
1482 return ERR_PTR(error);
1486 * Detach the svc_sock from the socket so that no
1487 * more callbacks occur.
1489 static void svc_sock_detach(struct svc_xprt *xprt)
1491 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1492 struct sock *sk = svsk->sk_sk;
1494 /* put back the old socket callbacks */
1496 sk->sk_state_change = svsk->sk_ostate;
1497 sk->sk_data_ready = svsk->sk_odata;
1498 sk->sk_write_space = svsk->sk_owspace;
1499 sk->sk_user_data = NULL;
1504 * Disconnect the socket, and reset the callbacks
1506 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1508 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1510 svc_sock_detach(xprt);
1512 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1513 svc_tcp_clear_pages(svsk);
1514 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1519 * Free the svc_sock's socket resources and the svc_sock itself.
1521 static void svc_sock_free(struct svc_xprt *xprt)
1523 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1525 if (svsk->sk_sock->file)
1526 sockfd_put(svsk->sk_sock);
1528 sock_release(svsk->sk_sock);