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 = container_of(xprt, struct svc_sock, sk_xprt);
328 sock_no_linger(svsk->sk_sock->sk);
332 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
334 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
337 struct in_pktinfo *pki = CMSG_DATA(cmh);
338 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
340 if (cmh->cmsg_type != IP_PKTINFO)
343 daddr->sin_family = AF_INET;
344 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
349 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
351 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
354 struct in6_pktinfo *pki = CMSG_DATA(cmh);
355 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
357 if (cmh->cmsg_type != IPV6_PKTINFO)
360 daddr->sin6_family = AF_INET6;
361 daddr->sin6_addr = pki->ipi6_addr;
362 daddr->sin6_scope_id = pki->ipi6_ifindex;
367 * Copy the UDP datagram's destination address to the rqstp structure.
368 * The 'destination' address in this case is the address to which the
369 * peer sent the datagram, i.e. our local address. For multihomed
370 * hosts, this can change from msg to msg. Note that only the IP
371 * address changes, the port number should remain the same.
373 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
376 switch (cmh->cmsg_level) {
378 return svc_udp_get_dest_address4(rqstp, cmh);
380 return svc_udp_get_dest_address6(rqstp, cmh);
387 * Receive a datagram from a UDP socket.
389 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
391 struct svc_sock *svsk =
392 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
393 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
397 long all[SVC_PKTINFO_SPACE / sizeof(long)];
399 struct cmsghdr *cmh = &buffer.hdr;
400 struct msghdr msg = {
401 .msg_name = svc_addr(rqstp),
403 .msg_controllen = sizeof(buffer),
404 .msg_flags = MSG_DONTWAIT,
409 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
410 /* udp sockets need large rcvbuf as all pending
411 * requests are still in that buffer. sndbuf must
412 * also be large enough that there is enough space
413 * for one reply per thread. We count all threads
414 * rather than threads in a particular pool, which
415 * provides an upper bound on the number of threads
416 * which will access the socket.
418 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
420 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
422 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
423 0, 0, MSG_PEEK | MSG_DONTWAIT);
425 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
428 if (err != -EAGAIN) {
429 /* possibly an icmp error */
430 dprintk("svc: recvfrom returned error %d\n", -err);
431 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
435 len = svc_addr_len(svc_addr(rqstp));
436 rqstp->rq_addrlen = len;
437 if (skb->tstamp == 0) {
438 skb->tstamp = ktime_get_real();
439 /* Don't enable netstamp, sunrpc doesn't
440 need that much accuracy */
442 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
443 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
446 rqstp->rq_arg.len = len;
448 rqstp->rq_prot = IPPROTO_UDP;
450 if (!svc_udp_get_dest_address(rqstp, cmh)) {
451 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
452 cmh->cmsg_level, cmh->cmsg_type);
455 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
457 if (skb_is_nonlinear(skb)) {
458 /* we have to copy */
460 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
468 /* we can use it in-place */
469 rqstp->rq_arg.head[0].iov_base = skb->data;
470 rqstp->rq_arg.head[0].iov_len = len;
471 if (skb_checksum_complete(skb))
473 rqstp->rq_xprt_ctxt = skb;
476 rqstp->rq_arg.page_base = 0;
477 if (len <= rqstp->rq_arg.head[0].iov_len) {
478 rqstp->rq_arg.head[0].iov_len = len;
479 rqstp->rq_arg.page_len = 0;
480 rqstp->rq_respages = rqstp->rq_pages+1;
482 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
483 rqstp->rq_respages = rqstp->rq_pages + 1 +
484 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
486 rqstp->rq_next_page = rqstp->rq_respages+1;
489 serv->sv_stats->netudpcnt++;
498 * svc_udp_sendto - Send out a reply on a UDP socket
499 * @rqstp: completed svc_rqst
501 * Returns the number of bytes sent, or a negative errno.
503 static int svc_udp_sendto(struct svc_rqst *rqstp)
505 struct svc_xprt *xprt = rqstp->rq_xprt;
506 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
507 struct xdr_buf *xdr = &rqstp->rq_res;
510 long all[SVC_PKTINFO_SPACE / sizeof(long)];
512 struct cmsghdr *cmh = &buffer.hdr;
513 struct msghdr msg = {
514 .msg_name = &rqstp->rq_addr,
515 .msg_namelen = rqstp->rq_addrlen,
517 .msg_controllen = sizeof(buffer),
519 unsigned int uninitialized_var(sent);
522 svc_release_udp_skb(rqstp);
524 svc_set_cmsg_data(rqstp, cmh);
526 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
528 if (err == -ECONNREFUSED) {
529 /* ICMP error on earlier request. */
530 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
538 static int svc_udp_has_wspace(struct svc_xprt *xprt)
540 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
541 struct svc_serv *serv = xprt->xpt_server;
542 unsigned long required;
545 * Set the SOCK_NOSPACE flag before checking the available
548 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
549 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
550 if (required*2 > sock_wspace(svsk->sk_sk))
552 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
556 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
562 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
566 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
568 struct sockaddr *sa, int salen,
571 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
574 static const struct svc_xprt_ops svc_udp_ops = {
575 .xpo_create = svc_udp_create,
576 .xpo_recvfrom = svc_udp_recvfrom,
577 .xpo_sendto = svc_udp_sendto,
578 .xpo_read_payload = svc_sock_read_payload,
579 .xpo_release_rqst = svc_release_udp_skb,
580 .xpo_detach = svc_sock_detach,
581 .xpo_free = svc_sock_free,
582 .xpo_has_wspace = svc_udp_has_wspace,
583 .xpo_accept = svc_udp_accept,
584 .xpo_secure_port = svc_sock_secure_port,
585 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
588 static struct svc_xprt_class svc_udp_class = {
590 .xcl_owner = THIS_MODULE,
591 .xcl_ops = &svc_udp_ops,
592 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
593 .xcl_ident = XPRT_TRANSPORT_UDP,
596 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
598 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
599 &svsk->sk_xprt, serv);
600 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
601 svsk->sk_sk->sk_data_ready = svc_data_ready;
602 svsk->sk_sk->sk_write_space = svc_write_space;
604 /* initialise setting must have enough space to
605 * receive and respond to one request.
606 * svc_udp_recvfrom will re-adjust if necessary
608 svc_sock_setbufsize(svsk, 3);
610 /* data might have come in before data_ready set up */
611 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
612 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
614 /* make sure we get destination address info */
615 switch (svsk->sk_sk->sk_family) {
617 ip_sock_set_pktinfo(svsk->sk_sock->sk);
620 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
628 * A data_ready event on a listening socket means there's a connection
629 * pending. Do not use state_change as a substitute for it.
631 static void svc_tcp_listen_data_ready(struct sock *sk)
633 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
635 dprintk("svc: socket %p TCP (listen) state change %d\n",
639 /* Refer to svc_setup_socket() for details. */
645 * This callback may called twice when a new connection
646 * is established as a child socket inherits everything
647 * from a parent LISTEN socket.
648 * 1) data_ready method of the parent socket will be called
649 * when one of child sockets become ESTABLISHED.
650 * 2) data_ready method of the child socket may be called
651 * when it receives data before the socket is accepted.
652 * In case of 2, we should ignore it silently.
654 if (sk->sk_state == TCP_LISTEN) {
656 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
657 svc_xprt_enqueue(&svsk->sk_xprt);
659 printk("svc: socket %p: no user data\n", sk);
664 * A state change on a connected socket means it's dying or dead.
666 static void svc_tcp_state_change(struct sock *sk)
668 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
670 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
671 sk, sk->sk_state, sk->sk_user_data);
674 printk("svc: socket %p: no user data\n", sk);
676 /* Refer to svc_setup_socket() for details. */
679 if (sk->sk_state != TCP_ESTABLISHED) {
680 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
681 svc_xprt_enqueue(&svsk->sk_xprt);
687 * Accept a TCP connection
689 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
691 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
692 struct sockaddr_storage addr;
693 struct sockaddr *sin = (struct sockaddr *) &addr;
694 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
695 struct socket *sock = svsk->sk_sock;
696 struct socket *newsock;
697 struct svc_sock *newsvsk;
699 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
701 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
705 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
706 err = kernel_accept(sock, &newsock, O_NONBLOCK);
709 printk(KERN_WARNING "%s: no more sockets!\n",
711 else if (err != -EAGAIN)
712 net_warn_ratelimited("%s: accept failed (err %d)!\n",
713 serv->sv_name, -err);
716 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
718 err = kernel_getpeername(newsock, sin);
720 net_warn_ratelimited("%s: peername failed (err %d)!\n",
721 serv->sv_name, -err);
722 goto failed; /* aborted connection or whatever */
726 /* Ideally, we would want to reject connections from unauthorized
727 * hosts here, but when we get encryption, the IP of the host won't
728 * tell us anything. For now just warn about unpriv connections.
730 if (!svc_port_is_privileged(sin)) {
731 dprintk("%s: connect from unprivileged port: %s\n",
733 __svc_print_addr(sin, buf, sizeof(buf)));
735 dprintk("%s: connect from %s\n", serv->sv_name,
736 __svc_print_addr(sin, buf, sizeof(buf)));
738 /* Reset the inherited callbacks before calling svc_setup_socket */
739 newsock->sk->sk_state_change = svsk->sk_ostate;
740 newsock->sk->sk_data_ready = svsk->sk_odata;
741 newsock->sk->sk_write_space = svsk->sk_owspace;
743 /* make sure that a write doesn't block forever when
746 newsock->sk->sk_sndtimeo = HZ*30;
748 newsvsk = svc_setup_socket(serv, newsock,
749 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
752 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
753 err = kernel_getsockname(newsock, sin);
755 if (unlikely(err < 0)) {
756 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
757 slen = offsetof(struct sockaddr, sa_data);
759 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
761 if (sock_is_loopback(newsock->sk))
762 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
764 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
766 serv->sv_stats->nettcpconn++;
768 return &newsvsk->sk_xprt;
771 sock_release(newsock);
775 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
777 unsigned int i, len, npages;
779 if (svsk->sk_datalen == 0)
781 len = svsk->sk_datalen;
782 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
783 for (i = 0; i < npages; i++) {
784 if (rqstp->rq_pages[i] != NULL)
785 put_page(rqstp->rq_pages[i]);
786 BUG_ON(svsk->sk_pages[i] == NULL);
787 rqstp->rq_pages[i] = svsk->sk_pages[i];
788 svsk->sk_pages[i] = NULL;
790 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
794 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
796 unsigned int i, len, npages;
798 if (svsk->sk_datalen == 0)
800 len = svsk->sk_datalen;
801 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
802 for (i = 0; i < npages; i++) {
803 svsk->sk_pages[i] = rqstp->rq_pages[i];
804 rqstp->rq_pages[i] = NULL;
808 static void svc_tcp_clear_pages(struct svc_sock *svsk)
810 unsigned int i, len, npages;
812 if (svsk->sk_datalen == 0)
814 len = svsk->sk_datalen;
815 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
816 for (i = 0; i < npages; i++) {
817 if (svsk->sk_pages[i] == NULL) {
821 put_page(svsk->sk_pages[i]);
822 svsk->sk_pages[i] = NULL;
826 svsk->sk_datalen = 0;
830 * Receive fragment record header.
831 * If we haven't gotten the record length yet, get the next four bytes.
833 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
835 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
839 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
842 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
843 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
845 len = svc_recvfrom(rqstp, &iov, 1, want, 0);
848 svsk->sk_tcplen += len;
851 dprintk("svc: short recvfrom while reading record "
852 "length (%d of %d)\n", len, want);
856 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
857 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
859 net_notice_ratelimited("RPC: fragment too large: %d\n",
860 svc_sock_reclen(svsk));
865 return svc_sock_reclen(svsk);
867 dprintk("RPC: TCP recv_record got %d\n", len);
870 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
874 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
876 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
877 struct rpc_rqst *req = NULL;
878 struct kvec *src, *dst;
879 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
888 spin_lock(&bc_xprt->queue_lock);
889 req = xprt_lookup_rqst(bc_xprt, xid);
891 goto unlock_notfound;
893 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
895 * XXX!: cheating for now! Only copying HEAD.
896 * But we know this is good enough for now (in fact, for any
897 * callback reply in the forseeable future).
899 dst = &req->rq_private_buf.head[0];
900 src = &rqstp->rq_arg.head[0];
901 if (dst->iov_len < src->iov_len)
902 goto unlock_eagain; /* whatever; just giving up. */
903 memcpy(dst->iov_base, src->iov_base, src->iov_len);
904 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
905 rqstp->rq_arg.len = 0;
906 spin_unlock(&bc_xprt->queue_lock);
910 "%s: Got unrecognized reply: "
911 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
912 __func__, ntohl(calldir),
913 bc_xprt, ntohl(xid));
915 spin_unlock(&bc_xprt->queue_lock);
919 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
925 vec[i].iov_base = page_address(pages[i]);
926 vec[i].iov_len = PAGE_SIZE;
933 static void svc_tcp_fragment_received(struct svc_sock *svsk)
935 /* If we have more data, signal svc_xprt_enqueue() to try again */
936 dprintk("svc: TCP %s record (%d bytes)\n",
937 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
938 svc_sock_reclen(svsk));
944 * Receive data from a TCP socket.
946 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
948 struct svc_sock *svsk =
949 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
950 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
953 unsigned int want, base;
958 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
959 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
960 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
961 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
963 len = svc_tcp_recv_record(svsk, rqstp);
967 base = svc_tcp_restore_pages(svsk, rqstp);
968 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
972 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want);
974 rqstp->rq_respages = &rqstp->rq_pages[pnum];
975 rqstp->rq_next_page = rqstp->rq_respages + 1;
977 /* Now receive data */
978 len = svc_recvfrom(rqstp, vec, pnum, base + want, base);
980 svsk->sk_tcplen += len;
981 svsk->sk_datalen += len;
983 if (len != want || !svc_sock_final_rec(svsk)) {
984 svc_tcp_save_pages(svsk, rqstp);
985 if (len < 0 && len != -EAGAIN)
988 svc_tcp_fragment_received(svsk);
990 dprintk("svc: incomplete TCP record (%d of %d)\n",
991 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
992 svc_sock_reclen(svsk));
996 if (svsk->sk_datalen < 8) {
997 svsk->sk_datalen = 0;
998 goto err_delete; /* client is nuts. */
1001 rqstp->rq_arg.len = svsk->sk_datalen;
1002 rqstp->rq_arg.page_base = 0;
1003 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1004 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1005 rqstp->rq_arg.page_len = 0;
1007 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1009 rqstp->rq_xprt_ctxt = NULL;
1010 rqstp->rq_prot = IPPROTO_TCP;
1011 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1012 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1014 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1016 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1019 len = receive_cb_reply(svsk, rqstp);
1021 /* Reset TCP read info */
1022 svsk->sk_datalen = 0;
1023 svc_tcp_fragment_received(svsk);
1028 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1030 serv->sv_stats->nettcpcnt++;
1032 return rqstp->rq_arg.len;
1037 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1040 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1041 svsk->sk_xprt.xpt_server->sv_name, -len);
1042 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1044 return 0; /* record not complete */
1048 * svc_tcp_sendto - Send out a reply on a TCP socket
1049 * @rqstp: completed svc_rqst
1051 * Returns the number of bytes sent, or a negative errno.
1053 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1055 struct svc_xprt *xprt = rqstp->rq_xprt;
1056 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1057 struct xdr_buf *xdr = &rqstp->rq_res;
1058 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1060 struct msghdr msg = {
1063 unsigned int uninitialized_var(sent);
1066 svc_release_skb(rqstp);
1068 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, marker, &sent);
1070 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1075 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1076 xprt->xpt_server->sv_name,
1077 (err < 0) ? "got error" : "sent",
1078 (err < 0) ? err : sent, xdr->len);
1079 set_bit(XPT_CLOSE, &xprt->xpt_flags);
1080 svc_xprt_enqueue(xprt);
1084 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1086 struct sockaddr *sa, int salen,
1089 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1092 static const struct svc_xprt_ops svc_tcp_ops = {
1093 .xpo_create = svc_tcp_create,
1094 .xpo_recvfrom = svc_tcp_recvfrom,
1095 .xpo_sendto = svc_tcp_sendto,
1096 .xpo_read_payload = svc_sock_read_payload,
1097 .xpo_release_rqst = svc_release_skb,
1098 .xpo_detach = svc_tcp_sock_detach,
1099 .xpo_free = svc_sock_free,
1100 .xpo_has_wspace = svc_tcp_has_wspace,
1101 .xpo_accept = svc_tcp_accept,
1102 .xpo_secure_port = svc_sock_secure_port,
1103 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1106 static struct svc_xprt_class svc_tcp_class = {
1108 .xcl_owner = THIS_MODULE,
1109 .xcl_ops = &svc_tcp_ops,
1110 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1111 .xcl_ident = XPRT_TRANSPORT_TCP,
1114 void svc_init_xprt_sock(void)
1116 svc_reg_xprt_class(&svc_tcp_class);
1117 svc_reg_xprt_class(&svc_udp_class);
1120 void svc_cleanup_xprt_sock(void)
1122 svc_unreg_xprt_class(&svc_tcp_class);
1123 svc_unreg_xprt_class(&svc_udp_class);
1126 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1128 struct sock *sk = svsk->sk_sk;
1130 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1131 &svsk->sk_xprt, serv);
1132 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1133 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1134 if (sk->sk_state == TCP_LISTEN) {
1135 dprintk("setting up TCP socket for listening\n");
1136 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1137 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1138 sk->sk_data_ready = svc_tcp_listen_data_ready;
1139 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1141 dprintk("setting up TCP socket for reading\n");
1142 sk->sk_state_change = svc_tcp_state_change;
1143 sk->sk_data_ready = svc_data_ready;
1144 sk->sk_write_space = svc_write_space;
1146 svsk->sk_reclen = 0;
1147 svsk->sk_tcplen = 0;
1148 svsk->sk_datalen = 0;
1149 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1151 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1153 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1154 switch (sk->sk_state) {
1156 case TCP_ESTABLISHED:
1159 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1164 void svc_sock_update_bufs(struct svc_serv *serv)
1167 * The number of server threads has changed. Update
1168 * rcvbuf and sndbuf accordingly on all sockets
1170 struct svc_sock *svsk;
1172 spin_lock_bh(&serv->sv_lock);
1173 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1174 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1175 spin_unlock_bh(&serv->sv_lock);
1177 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1180 * Initialize socket for RPC use and create svc_sock struct
1182 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1183 struct socket *sock,
1186 struct svc_sock *svsk;
1188 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1191 dprintk("svc: svc_setup_socket %p\n", sock);
1192 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1194 return ERR_PTR(-ENOMEM);
1198 /* Register socket with portmapper */
1200 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1202 ntohs(inet_sk(inet)->inet_sport));
1206 return ERR_PTR(err);
1209 svsk->sk_sock = sock;
1211 svsk->sk_ostate = inet->sk_state_change;
1212 svsk->sk_odata = inet->sk_data_ready;
1213 svsk->sk_owspace = inet->sk_write_space;
1215 * This barrier is necessary in order to prevent race condition
1216 * with svc_data_ready(), svc_listen_data_ready() and others
1217 * when calling callbacks above.
1220 inet->sk_user_data = svsk;
1222 /* Initialize the socket */
1223 if (sock->type == SOCK_DGRAM)
1224 svc_udp_init(svsk, serv);
1226 svc_tcp_init(svsk, serv);
1228 dprintk("svc: svc_setup_socket created %p (inet %p), "
1229 "listen %d close %d\n",
1231 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1232 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1237 bool svc_alien_sock(struct net *net, int fd)
1240 struct socket *sock = sockfd_lookup(fd, &err);
1245 if (sock_net(sock->sk) != net)
1251 EXPORT_SYMBOL_GPL(svc_alien_sock);
1254 * svc_addsock - add a listener socket to an RPC service
1255 * @serv: pointer to RPC service to which to add a new listener
1256 * @fd: file descriptor of the new listener
1257 * @name_return: pointer to buffer to fill in with name of listener
1258 * @len: size of the buffer
1261 * Fills in socket name and returns positive length of name if successful.
1262 * Name is terminated with '\n'. On error, returns a negative errno
1265 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1266 const size_t len, const struct cred *cred)
1269 struct socket *so = sockfd_lookup(fd, &err);
1270 struct svc_sock *svsk = NULL;
1271 struct sockaddr_storage addr;
1272 struct sockaddr *sin = (struct sockaddr *)&addr;
1277 err = -EAFNOSUPPORT;
1278 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1280 err = -EPROTONOSUPPORT;
1281 if (so->sk->sk_protocol != IPPROTO_TCP &&
1282 so->sk->sk_protocol != IPPROTO_UDP)
1285 if (so->state > SS_UNCONNECTED)
1288 if (!try_module_get(THIS_MODULE))
1290 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1292 module_put(THIS_MODULE);
1293 err = PTR_ERR(svsk);
1296 salen = kernel_getsockname(svsk->sk_sock, sin);
1298 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1299 svsk->sk_xprt.xpt_cred = get_cred(cred);
1300 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1301 return svc_one_sock_name(svsk, name_return, len);
1306 EXPORT_SYMBOL_GPL(svc_addsock);
1309 * Create socket for RPC service.
1311 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1314 struct sockaddr *sin, int len,
1317 struct svc_sock *svsk;
1318 struct socket *sock;
1321 struct sockaddr_storage addr;
1322 struct sockaddr *newsin = (struct sockaddr *)&addr;
1325 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1327 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1328 serv->sv_program->pg_name, protocol,
1329 __svc_print_addr(sin, buf, sizeof(buf)));
1331 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1332 printk(KERN_WARNING "svc: only UDP and TCP "
1333 "sockets supported\n");
1334 return ERR_PTR(-EINVAL);
1337 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1338 switch (sin->sa_family) {
1346 return ERR_PTR(-EINVAL);
1349 error = __sock_create(net, family, type, protocol, &sock, 1);
1351 return ERR_PTR(error);
1353 svc_reclassify_socket(sock);
1356 * If this is an PF_INET6 listener, we want to avoid
1357 * getting requests from IPv4 remotes. Those should
1358 * be shunted to a PF_INET listener via rpcbind.
1360 if (family == PF_INET6)
1361 ip6_sock_set_v6only(sock->sk);
1362 if (type == SOCK_STREAM)
1363 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1364 error = kernel_bind(sock, sin, len);
1368 error = kernel_getsockname(sock, newsin);
1373 if (protocol == IPPROTO_TCP) {
1374 if ((error = kernel_listen(sock, 64)) < 0)
1378 svsk = svc_setup_socket(serv, sock, flags);
1380 error = PTR_ERR(svsk);
1383 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1384 return (struct svc_xprt *)svsk;
1386 dprintk("svc: svc_create_socket error = %d\n", -error);
1388 return ERR_PTR(error);
1392 * Detach the svc_sock from the socket so that no
1393 * more callbacks occur.
1395 static void svc_sock_detach(struct svc_xprt *xprt)
1397 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1398 struct sock *sk = svsk->sk_sk;
1400 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1402 /* put back the old socket callbacks */
1404 sk->sk_state_change = svsk->sk_ostate;
1405 sk->sk_data_ready = svsk->sk_odata;
1406 sk->sk_write_space = svsk->sk_owspace;
1407 sk->sk_user_data = NULL;
1412 * Disconnect the socket, and reset the callbacks
1414 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1416 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1418 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1420 svc_sock_detach(xprt);
1422 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1423 svc_tcp_clear_pages(svsk);
1424 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1429 * Free the svc_sock's socket resources and the svc_sock itself.
1431 static void svc_sock_free(struct svc_xprt *xprt)
1433 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1434 dprintk("svc: svc_sock_free(%p)\n", svsk);
1436 if (svsk->sk_sock->file)
1437 sockfd_put(svsk->sk_sock);
1439 sock_release(svsk->sk_sock);