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 <asm/ioctls.h>
48 #include <trace/events/skb.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>
61 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
64 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
66 static int svc_udp_recvfrom(struct svc_rqst *);
67 static int svc_udp_sendto(struct svc_rqst *);
68 static void svc_sock_detach(struct svc_xprt *);
69 static void svc_tcp_sock_detach(struct svc_xprt *);
70 static void svc_sock_free(struct svc_xprt *);
72 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
73 struct net *, struct sockaddr *,
75 #ifdef CONFIG_DEBUG_LOCK_ALLOC
76 static struct lock_class_key svc_key[2];
77 static struct lock_class_key svc_slock_key[2];
79 static void svc_reclassify_socket(struct socket *sock)
81 struct sock *sk = sock->sk;
83 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
86 switch (sk->sk_family) {
88 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
90 "sk_xprt.xpt_lock-AF_INET-NFSD",
95 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
97 "sk_xprt.xpt_lock-AF_INET6-NFSD",
106 static void svc_reclassify_socket(struct socket *sock)
112 * Release an skbuff after use
114 static void svc_release_skb(struct svc_rqst *rqstp)
116 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
119 struct svc_sock *svsk =
120 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
121 rqstp->rq_xprt_ctxt = NULL;
123 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
124 skb_free_datagram_locked(svsk->sk_sk, skb);
128 static void svc_release_udp_skb(struct svc_rqst *rqstp)
130 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
133 rqstp->rq_xprt_ctxt = NULL;
135 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
140 union svc_pktinfo_u {
141 struct in_pktinfo pkti;
142 struct in6_pktinfo pkti6;
144 #define SVC_PKTINFO_SPACE \
145 CMSG_SPACE(sizeof(union svc_pktinfo_u))
147 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
149 struct svc_sock *svsk =
150 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
151 switch (svsk->sk_sk->sk_family) {
153 struct in_pktinfo *pki = CMSG_DATA(cmh);
155 cmh->cmsg_level = SOL_IP;
156 cmh->cmsg_type = IP_PKTINFO;
157 pki->ipi_ifindex = 0;
158 pki->ipi_spec_dst.s_addr =
159 svc_daddr_in(rqstp)->sin_addr.s_addr;
160 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
165 struct in6_pktinfo *pki = CMSG_DATA(cmh);
166 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
168 cmh->cmsg_level = SOL_IPV6;
169 cmh->cmsg_type = IPV6_PKTINFO;
170 pki->ipi6_ifindex = daddr->sin6_scope_id;
171 pki->ipi6_addr = daddr->sin6_addr;
172 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
178 static int svc_sock_read_payload(struct svc_rqst *rqstp, unsigned int offset,
185 * Report socket names for nfsdfs
187 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
189 const struct sock *sk = svsk->sk_sk;
190 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
194 switch (sk->sk_family) {
196 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
198 &inet_sk(sk)->inet_rcv_saddr,
199 inet_sk(sk)->inet_num);
201 #if IS_ENABLED(CONFIG_IPV6)
203 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
205 &sk->sk_v6_rcv_saddr,
206 inet_sk(sk)->inet_num);
210 len = snprintf(buf, remaining, "*unknown-%d*\n",
214 if (len >= remaining) {
216 return -ENAMETOOLONG;
222 * Generic recvfrom routine.
224 static ssize_t svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov,
225 unsigned int nr, size_t buflen, unsigned int base)
227 struct svc_sock *svsk =
228 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
229 struct msghdr msg = { NULL };
232 rqstp->rq_xprt_hlen = 0;
234 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
235 iov_iter_kvec(&msg.msg_iter, READ, iov, nr, buflen);
237 iov_iter_advance(&msg.msg_iter, base);
240 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
241 /* If we read a full record, then assume there may be more
242 * data to read (stream based sockets only!)
245 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
247 dprintk("svc: socket %p recvfrom(%p, %zu) = %zd\n",
248 svsk, iov[0].iov_base, iov[0].iov_len, len);
253 * Set socket snd and rcv buffer lengths
255 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
257 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
258 struct socket *sock = svsk->sk_sock;
260 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
263 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
264 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
265 sock->sk->sk_write_space(sock->sk);
266 release_sock(sock->sk);
269 static void svc_sock_secure_port(struct svc_rqst *rqstp)
271 if (svc_port_is_privileged(svc_addr(rqstp)))
272 set_bit(RQ_SECURE, &rqstp->rq_flags);
274 clear_bit(RQ_SECURE, &rqstp->rq_flags);
278 * INET callback when data has been received on the socket.
280 static void svc_data_ready(struct sock *sk)
282 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
285 dprintk("svc: socket %p(inet %p), busy=%d\n",
287 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
289 /* Refer to svc_setup_socket() for details. */
292 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
293 svc_xprt_enqueue(&svsk->sk_xprt);
298 * INET callback when space is newly available on the socket.
300 static void svc_write_space(struct sock *sk)
302 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
305 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
306 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
308 /* Refer to svc_setup_socket() for details. */
310 svsk->sk_owspace(sk);
311 svc_xprt_enqueue(&svsk->sk_xprt);
315 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
317 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
319 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
321 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
324 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
326 struct svc_sock *svsk;
328 struct linger no_linger = {
333 svsk = container_of(xprt, struct svc_sock, sk_xprt);
334 sock = svsk->sk_sock;
335 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
336 (char *)&no_linger, sizeof(no_linger));
340 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
342 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
345 struct in_pktinfo *pki = CMSG_DATA(cmh);
346 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
348 if (cmh->cmsg_type != IP_PKTINFO)
351 daddr->sin_family = AF_INET;
352 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
357 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
359 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
362 struct in6_pktinfo *pki = CMSG_DATA(cmh);
363 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
365 if (cmh->cmsg_type != IPV6_PKTINFO)
368 daddr->sin6_family = AF_INET6;
369 daddr->sin6_addr = pki->ipi6_addr;
370 daddr->sin6_scope_id = pki->ipi6_ifindex;
375 * Copy the UDP datagram's destination address to the rqstp structure.
376 * The 'destination' address in this case is the address to which the
377 * peer sent the datagram, i.e. our local address. For multihomed
378 * hosts, this can change from msg to msg. Note that only the IP
379 * address changes, the port number should remain the same.
381 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
384 switch (cmh->cmsg_level) {
386 return svc_udp_get_dest_address4(rqstp, cmh);
388 return svc_udp_get_dest_address6(rqstp, cmh);
395 * Receive a datagram from a UDP socket.
397 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
399 struct svc_sock *svsk =
400 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
401 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
405 long all[SVC_PKTINFO_SPACE / sizeof(long)];
407 struct cmsghdr *cmh = &buffer.hdr;
408 struct msghdr msg = {
409 .msg_name = svc_addr(rqstp),
411 .msg_controllen = sizeof(buffer),
412 .msg_flags = MSG_DONTWAIT,
417 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
418 /* udp sockets need large rcvbuf as all pending
419 * requests are still in that buffer. sndbuf must
420 * also be large enough that there is enough space
421 * for one reply per thread. We count all threads
422 * rather than threads in a particular pool, which
423 * provides an upper bound on the number of threads
424 * which will access the socket.
426 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
428 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
430 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
431 0, 0, MSG_PEEK | MSG_DONTWAIT);
433 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
436 if (err != -EAGAIN) {
437 /* possibly an icmp error */
438 dprintk("svc: recvfrom returned error %d\n", -err);
439 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
443 len = svc_addr_len(svc_addr(rqstp));
444 rqstp->rq_addrlen = len;
445 if (skb->tstamp == 0) {
446 skb->tstamp = ktime_get_real();
447 /* Don't enable netstamp, sunrpc doesn't
448 need that much accuracy */
450 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
451 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
454 rqstp->rq_arg.len = len;
456 rqstp->rq_prot = IPPROTO_UDP;
458 if (!svc_udp_get_dest_address(rqstp, cmh)) {
459 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
460 cmh->cmsg_level, cmh->cmsg_type);
463 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
465 if (skb_is_nonlinear(skb)) {
466 /* we have to copy */
468 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
476 /* we can use it in-place */
477 rqstp->rq_arg.head[0].iov_base = skb->data;
478 rqstp->rq_arg.head[0].iov_len = len;
479 if (skb_checksum_complete(skb))
481 rqstp->rq_xprt_ctxt = skb;
484 rqstp->rq_arg.page_base = 0;
485 if (len <= rqstp->rq_arg.head[0].iov_len) {
486 rqstp->rq_arg.head[0].iov_len = len;
487 rqstp->rq_arg.page_len = 0;
488 rqstp->rq_respages = rqstp->rq_pages+1;
490 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
491 rqstp->rq_respages = rqstp->rq_pages + 1 +
492 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
494 rqstp->rq_next_page = rqstp->rq_respages+1;
497 serv->sv_stats->netudpcnt++;
506 * svc_udp_sendto - Send out a reply on a UDP socket
507 * @rqstp: completed svc_rqst
509 * Returns the number of bytes sent, or a negative errno.
511 static int svc_udp_sendto(struct svc_rqst *rqstp)
513 struct svc_xprt *xprt = rqstp->rq_xprt;
514 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
515 struct xdr_buf *xdr = &rqstp->rq_res;
518 long all[SVC_PKTINFO_SPACE / sizeof(long)];
520 struct cmsghdr *cmh = &buffer.hdr;
521 struct msghdr msg = {
522 .msg_name = &rqstp->rq_addr,
523 .msg_namelen = rqstp->rq_addrlen,
525 .msg_controllen = sizeof(buffer),
527 unsigned int uninitialized_var(sent);
530 svc_release_udp_skb(rqstp);
532 svc_set_cmsg_data(rqstp, cmh);
534 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
536 if (err == -ECONNREFUSED) {
537 /* ICMP error on earlier request. */
538 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
546 static int svc_udp_has_wspace(struct svc_xprt *xprt)
548 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
549 struct svc_serv *serv = xprt->xpt_server;
550 unsigned long required;
553 * Set the SOCK_NOSPACE flag before checking the available
556 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
557 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
558 if (required*2 > sock_wspace(svsk->sk_sk))
560 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
564 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
570 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
574 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
576 struct sockaddr *sa, int salen,
579 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
582 static const struct svc_xprt_ops svc_udp_ops = {
583 .xpo_create = svc_udp_create,
584 .xpo_recvfrom = svc_udp_recvfrom,
585 .xpo_sendto = svc_udp_sendto,
586 .xpo_read_payload = svc_sock_read_payload,
587 .xpo_release_rqst = svc_release_udp_skb,
588 .xpo_detach = svc_sock_detach,
589 .xpo_free = svc_sock_free,
590 .xpo_has_wspace = svc_udp_has_wspace,
591 .xpo_accept = svc_udp_accept,
592 .xpo_secure_port = svc_sock_secure_port,
593 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
596 static struct svc_xprt_class svc_udp_class = {
598 .xcl_owner = THIS_MODULE,
599 .xcl_ops = &svc_udp_ops,
600 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
601 .xcl_ident = XPRT_TRANSPORT_UDP,
604 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
606 int err, level, optname, one = 1;
608 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
609 &svsk->sk_xprt, serv);
610 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
611 svsk->sk_sk->sk_data_ready = svc_data_ready;
612 svsk->sk_sk->sk_write_space = svc_write_space;
614 /* initialise setting must have enough space to
615 * receive and respond to one request.
616 * svc_udp_recvfrom will re-adjust if necessary
618 svc_sock_setbufsize(svsk, 3);
620 /* data might have come in before data_ready set up */
621 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
622 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
624 /* make sure we get destination address info */
625 switch (svsk->sk_sk->sk_family) {
628 optname = IP_PKTINFO;
632 optname = IPV6_RECVPKTINFO;
637 err = kernel_setsockopt(svsk->sk_sock, level, optname,
638 (char *)&one, sizeof(one));
639 dprintk("svc: kernel_setsockopt returned %d\n", err);
643 * A data_ready event on a listening socket means there's a connection
644 * pending. Do not use state_change as a substitute for it.
646 static void svc_tcp_listen_data_ready(struct sock *sk)
648 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
650 dprintk("svc: socket %p TCP (listen) state change %d\n",
654 /* Refer to svc_setup_socket() for details. */
660 * This callback may called twice when a new connection
661 * is established as a child socket inherits everything
662 * from a parent LISTEN socket.
663 * 1) data_ready method of the parent socket will be called
664 * when one of child sockets become ESTABLISHED.
665 * 2) data_ready method of the child socket may be called
666 * when it receives data before the socket is accepted.
667 * In case of 2, we should ignore it silently.
669 if (sk->sk_state == TCP_LISTEN) {
671 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
672 svc_xprt_enqueue(&svsk->sk_xprt);
674 printk("svc: socket %p: no user data\n", sk);
679 * A state change on a connected socket means it's dying or dead.
681 static void svc_tcp_state_change(struct sock *sk)
683 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
685 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
686 sk, sk->sk_state, sk->sk_user_data);
689 printk("svc: socket %p: no user data\n", sk);
691 /* Refer to svc_setup_socket() for details. */
694 if (sk->sk_state != TCP_ESTABLISHED) {
695 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
696 svc_xprt_enqueue(&svsk->sk_xprt);
702 * Accept a TCP connection
704 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
706 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
707 struct sockaddr_storage addr;
708 struct sockaddr *sin = (struct sockaddr *) &addr;
709 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
710 struct socket *sock = svsk->sk_sock;
711 struct socket *newsock;
712 struct svc_sock *newsvsk;
714 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
716 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
720 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
721 err = kernel_accept(sock, &newsock, O_NONBLOCK);
724 printk(KERN_WARNING "%s: no more sockets!\n",
726 else if (err != -EAGAIN)
727 net_warn_ratelimited("%s: accept failed (err %d)!\n",
728 serv->sv_name, -err);
731 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
733 err = kernel_getpeername(newsock, sin);
735 net_warn_ratelimited("%s: peername failed (err %d)!\n",
736 serv->sv_name, -err);
737 goto failed; /* aborted connection or whatever */
741 /* Ideally, we would want to reject connections from unauthorized
742 * hosts here, but when we get encryption, the IP of the host won't
743 * tell us anything. For now just warn about unpriv connections.
745 if (!svc_port_is_privileged(sin)) {
746 dprintk("%s: connect from unprivileged port: %s\n",
748 __svc_print_addr(sin, buf, sizeof(buf)));
750 dprintk("%s: connect from %s\n", serv->sv_name,
751 __svc_print_addr(sin, buf, sizeof(buf)));
753 /* Reset the inherited callbacks before calling svc_setup_socket */
754 newsock->sk->sk_state_change = svsk->sk_ostate;
755 newsock->sk->sk_data_ready = svsk->sk_odata;
756 newsock->sk->sk_write_space = svsk->sk_owspace;
758 /* make sure that a write doesn't block forever when
761 newsock->sk->sk_sndtimeo = HZ*30;
763 newsvsk = svc_setup_socket(serv, newsock,
764 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
767 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
768 err = kernel_getsockname(newsock, sin);
770 if (unlikely(err < 0)) {
771 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
772 slen = offsetof(struct sockaddr, sa_data);
774 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
776 if (sock_is_loopback(newsock->sk))
777 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
779 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
781 serv->sv_stats->nettcpconn++;
783 return &newsvsk->sk_xprt;
786 sock_release(newsock);
790 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
792 unsigned int i, len, npages;
794 if (svsk->sk_datalen == 0)
796 len = svsk->sk_datalen;
797 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
798 for (i = 0; i < npages; i++) {
799 if (rqstp->rq_pages[i] != NULL)
800 put_page(rqstp->rq_pages[i]);
801 BUG_ON(svsk->sk_pages[i] == NULL);
802 rqstp->rq_pages[i] = svsk->sk_pages[i];
803 svsk->sk_pages[i] = NULL;
805 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
809 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
811 unsigned int i, len, npages;
813 if (svsk->sk_datalen == 0)
815 len = svsk->sk_datalen;
816 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
817 for (i = 0; i < npages; i++) {
818 svsk->sk_pages[i] = rqstp->rq_pages[i];
819 rqstp->rq_pages[i] = NULL;
823 static void svc_tcp_clear_pages(struct svc_sock *svsk)
825 unsigned int i, len, npages;
827 if (svsk->sk_datalen == 0)
829 len = svsk->sk_datalen;
830 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
831 for (i = 0; i < npages; i++) {
832 if (svsk->sk_pages[i] == NULL) {
836 put_page(svsk->sk_pages[i]);
837 svsk->sk_pages[i] = NULL;
841 svsk->sk_datalen = 0;
845 * Receive fragment record header.
846 * If we haven't gotten the record length yet, get the next four bytes.
848 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
850 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
854 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
857 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
858 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
860 len = svc_recvfrom(rqstp, &iov, 1, want, 0);
863 svsk->sk_tcplen += len;
866 dprintk("svc: short recvfrom while reading record "
867 "length (%d of %d)\n", len, want);
871 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
872 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
874 net_notice_ratelimited("RPC: fragment too large: %d\n",
875 svc_sock_reclen(svsk));
880 return svc_sock_reclen(svsk);
882 dprintk("RPC: TCP recv_record got %d\n", len);
885 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
889 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
891 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
892 struct rpc_rqst *req = NULL;
893 struct kvec *src, *dst;
894 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
903 spin_lock(&bc_xprt->queue_lock);
904 req = xprt_lookup_rqst(bc_xprt, xid);
906 goto unlock_notfound;
908 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
910 * XXX!: cheating for now! Only copying HEAD.
911 * But we know this is good enough for now (in fact, for any
912 * callback reply in the forseeable future).
914 dst = &req->rq_private_buf.head[0];
915 src = &rqstp->rq_arg.head[0];
916 if (dst->iov_len < src->iov_len)
917 goto unlock_eagain; /* whatever; just giving up. */
918 memcpy(dst->iov_base, src->iov_base, src->iov_len);
919 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
920 rqstp->rq_arg.len = 0;
921 spin_unlock(&bc_xprt->queue_lock);
925 "%s: Got unrecognized reply: "
926 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
927 __func__, ntohl(calldir),
928 bc_xprt, ntohl(xid));
930 spin_unlock(&bc_xprt->queue_lock);
934 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
940 vec[i].iov_base = page_address(pages[i]);
941 vec[i].iov_len = PAGE_SIZE;
948 static void svc_tcp_fragment_received(struct svc_sock *svsk)
950 /* If we have more data, signal svc_xprt_enqueue() to try again */
951 dprintk("svc: TCP %s record (%d bytes)\n",
952 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
953 svc_sock_reclen(svsk));
959 * Receive data from a TCP socket.
961 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
963 struct svc_sock *svsk =
964 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
965 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
968 unsigned int want, base;
973 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
974 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
975 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
976 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
978 len = svc_tcp_recv_record(svsk, rqstp);
982 base = svc_tcp_restore_pages(svsk, rqstp);
983 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
987 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want);
989 rqstp->rq_respages = &rqstp->rq_pages[pnum];
990 rqstp->rq_next_page = rqstp->rq_respages + 1;
992 /* Now receive data */
993 len = svc_recvfrom(rqstp, vec, pnum, base + want, base);
995 svsk->sk_tcplen += len;
996 svsk->sk_datalen += len;
998 if (len != want || !svc_sock_final_rec(svsk)) {
999 svc_tcp_save_pages(svsk, rqstp);
1000 if (len < 0 && len != -EAGAIN)
1003 svc_tcp_fragment_received(svsk);
1005 dprintk("svc: incomplete TCP record (%d of %d)\n",
1006 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1007 svc_sock_reclen(svsk));
1011 if (svsk->sk_datalen < 8) {
1012 svsk->sk_datalen = 0;
1013 goto err_delete; /* client is nuts. */
1016 rqstp->rq_arg.len = svsk->sk_datalen;
1017 rqstp->rq_arg.page_base = 0;
1018 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1019 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1020 rqstp->rq_arg.page_len = 0;
1022 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1024 rqstp->rq_xprt_ctxt = NULL;
1025 rqstp->rq_prot = IPPROTO_TCP;
1026 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1027 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1029 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1031 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1034 len = receive_cb_reply(svsk, rqstp);
1036 /* Reset TCP read info */
1037 svsk->sk_datalen = 0;
1038 svc_tcp_fragment_received(svsk);
1043 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1045 serv->sv_stats->nettcpcnt++;
1047 return rqstp->rq_arg.len;
1052 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1055 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1056 svsk->sk_xprt.xpt_server->sv_name, -len);
1057 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1059 return 0; /* record not complete */
1063 * svc_tcp_sendto - Send out a reply on a TCP socket
1064 * @rqstp: completed svc_rqst
1066 * Returns the number of bytes sent, or a negative errno.
1068 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1070 struct svc_xprt *xprt = rqstp->rq_xprt;
1071 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1072 struct xdr_buf *xdr = &rqstp->rq_res;
1073 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1075 struct msghdr msg = {
1078 unsigned int uninitialized_var(sent);
1081 svc_release_skb(rqstp);
1083 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, marker, &sent);
1085 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1090 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1091 xprt->xpt_server->sv_name,
1092 (err < 0) ? "got error" : "sent",
1093 (err < 0) ? err : sent, xdr->len);
1094 set_bit(XPT_CLOSE, &xprt->xpt_flags);
1095 svc_xprt_enqueue(xprt);
1099 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1101 struct sockaddr *sa, int salen,
1104 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1107 static const struct svc_xprt_ops svc_tcp_ops = {
1108 .xpo_create = svc_tcp_create,
1109 .xpo_recvfrom = svc_tcp_recvfrom,
1110 .xpo_sendto = svc_tcp_sendto,
1111 .xpo_read_payload = svc_sock_read_payload,
1112 .xpo_release_rqst = svc_release_skb,
1113 .xpo_detach = svc_tcp_sock_detach,
1114 .xpo_free = svc_sock_free,
1115 .xpo_has_wspace = svc_tcp_has_wspace,
1116 .xpo_accept = svc_tcp_accept,
1117 .xpo_secure_port = svc_sock_secure_port,
1118 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1121 static struct svc_xprt_class svc_tcp_class = {
1123 .xcl_owner = THIS_MODULE,
1124 .xcl_ops = &svc_tcp_ops,
1125 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1126 .xcl_ident = XPRT_TRANSPORT_TCP,
1129 void svc_init_xprt_sock(void)
1131 svc_reg_xprt_class(&svc_tcp_class);
1132 svc_reg_xprt_class(&svc_udp_class);
1135 void svc_cleanup_xprt_sock(void)
1137 svc_unreg_xprt_class(&svc_tcp_class);
1138 svc_unreg_xprt_class(&svc_udp_class);
1141 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1143 struct sock *sk = svsk->sk_sk;
1145 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1146 &svsk->sk_xprt, serv);
1147 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1148 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1149 if (sk->sk_state == TCP_LISTEN) {
1150 dprintk("setting up TCP socket for listening\n");
1151 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1152 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1153 sk->sk_data_ready = svc_tcp_listen_data_ready;
1154 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1156 dprintk("setting up TCP socket for reading\n");
1157 sk->sk_state_change = svc_tcp_state_change;
1158 sk->sk_data_ready = svc_data_ready;
1159 sk->sk_write_space = svc_write_space;
1161 svsk->sk_reclen = 0;
1162 svsk->sk_tcplen = 0;
1163 svsk->sk_datalen = 0;
1164 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1166 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1168 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1169 switch (sk->sk_state) {
1171 case TCP_ESTABLISHED:
1174 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1179 void svc_sock_update_bufs(struct svc_serv *serv)
1182 * The number of server threads has changed. Update
1183 * rcvbuf and sndbuf accordingly on all sockets
1185 struct svc_sock *svsk;
1187 spin_lock_bh(&serv->sv_lock);
1188 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1189 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1190 spin_unlock_bh(&serv->sv_lock);
1192 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1195 * Initialize socket for RPC use and create svc_sock struct
1197 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1198 struct socket *sock,
1201 struct svc_sock *svsk;
1203 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1206 dprintk("svc: svc_setup_socket %p\n", sock);
1207 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1209 return ERR_PTR(-ENOMEM);
1213 /* Register socket with portmapper */
1215 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1217 ntohs(inet_sk(inet)->inet_sport));
1221 return ERR_PTR(err);
1224 svsk->sk_sock = sock;
1226 svsk->sk_ostate = inet->sk_state_change;
1227 svsk->sk_odata = inet->sk_data_ready;
1228 svsk->sk_owspace = inet->sk_write_space;
1230 * This barrier is necessary in order to prevent race condition
1231 * with svc_data_ready(), svc_listen_data_ready() and others
1232 * when calling callbacks above.
1235 inet->sk_user_data = svsk;
1237 /* Initialize the socket */
1238 if (sock->type == SOCK_DGRAM)
1239 svc_udp_init(svsk, serv);
1241 svc_tcp_init(svsk, serv);
1243 dprintk("svc: svc_setup_socket created %p (inet %p), "
1244 "listen %d close %d\n",
1246 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1247 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1252 bool svc_alien_sock(struct net *net, int fd)
1255 struct socket *sock = sockfd_lookup(fd, &err);
1260 if (sock_net(sock->sk) != net)
1266 EXPORT_SYMBOL_GPL(svc_alien_sock);
1269 * svc_addsock - add a listener socket to an RPC service
1270 * @serv: pointer to RPC service to which to add a new listener
1271 * @fd: file descriptor of the new listener
1272 * @name_return: pointer to buffer to fill in with name of listener
1273 * @len: size of the buffer
1276 * Fills in socket name and returns positive length of name if successful.
1277 * Name is terminated with '\n'. On error, returns a negative errno
1280 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1281 const size_t len, const struct cred *cred)
1284 struct socket *so = sockfd_lookup(fd, &err);
1285 struct svc_sock *svsk = NULL;
1286 struct sockaddr_storage addr;
1287 struct sockaddr *sin = (struct sockaddr *)&addr;
1292 err = -EAFNOSUPPORT;
1293 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1295 err = -EPROTONOSUPPORT;
1296 if (so->sk->sk_protocol != IPPROTO_TCP &&
1297 so->sk->sk_protocol != IPPROTO_UDP)
1300 if (so->state > SS_UNCONNECTED)
1303 if (!try_module_get(THIS_MODULE))
1305 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1307 module_put(THIS_MODULE);
1308 err = PTR_ERR(svsk);
1311 salen = kernel_getsockname(svsk->sk_sock, sin);
1313 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1314 svsk->sk_xprt.xpt_cred = get_cred(cred);
1315 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1316 return svc_one_sock_name(svsk, name_return, len);
1321 EXPORT_SYMBOL_GPL(svc_addsock);
1324 * Create socket for RPC service.
1326 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1329 struct sockaddr *sin, int len,
1332 struct svc_sock *svsk;
1333 struct socket *sock;
1336 struct sockaddr_storage addr;
1337 struct sockaddr *newsin = (struct sockaddr *)&addr;
1341 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1343 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1344 serv->sv_program->pg_name, protocol,
1345 __svc_print_addr(sin, buf, sizeof(buf)));
1347 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1348 printk(KERN_WARNING "svc: only UDP and TCP "
1349 "sockets supported\n");
1350 return ERR_PTR(-EINVAL);
1353 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1354 switch (sin->sa_family) {
1362 return ERR_PTR(-EINVAL);
1365 error = __sock_create(net, family, type, protocol, &sock, 1);
1367 return ERR_PTR(error);
1369 svc_reclassify_socket(sock);
1372 * If this is an PF_INET6 listener, we want to avoid
1373 * getting requests from IPv4 remotes. Those should
1374 * be shunted to a PF_INET listener via rpcbind.
1377 if (family == PF_INET6)
1378 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1379 (char *)&val, sizeof(val));
1381 if (type == SOCK_STREAM)
1382 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1383 error = kernel_bind(sock, sin, len);
1387 error = kernel_getsockname(sock, newsin);
1392 if (protocol == IPPROTO_TCP) {
1393 if ((error = kernel_listen(sock, 64)) < 0)
1397 svsk = svc_setup_socket(serv, sock, flags);
1399 error = PTR_ERR(svsk);
1402 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1403 return (struct svc_xprt *)svsk;
1405 dprintk("svc: svc_create_socket error = %d\n", -error);
1407 return ERR_PTR(error);
1411 * Detach the svc_sock from the socket so that no
1412 * more callbacks occur.
1414 static void svc_sock_detach(struct svc_xprt *xprt)
1416 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1417 struct sock *sk = svsk->sk_sk;
1419 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1421 /* put back the old socket callbacks */
1423 sk->sk_state_change = svsk->sk_ostate;
1424 sk->sk_data_ready = svsk->sk_odata;
1425 sk->sk_write_space = svsk->sk_owspace;
1426 sk->sk_user_data = NULL;
1431 * Disconnect the socket, and reset the callbacks
1433 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1435 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1437 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1439 svc_sock_detach(xprt);
1441 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1442 svc_tcp_clear_pages(svsk);
1443 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1448 * Free the svc_sock's socket resources and the svc_sock itself.
1450 static void svc_sock_free(struct svc_xprt *xprt)
1452 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1453 dprintk("svc: svc_sock_free(%p)\n", svsk);
1455 if (svsk->sk_sock->file)
1456 sockfd_put(svsk->sk_sock);
1458 sock_release(svsk->sk_sock);