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>
60 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
63 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
65 static int svc_udp_recvfrom(struct svc_rqst *);
66 static int svc_udp_sendto(struct svc_rqst *);
67 static void svc_sock_detach(struct svc_xprt *);
68 static void svc_tcp_sock_detach(struct svc_xprt *);
69 static void svc_sock_free(struct svc_xprt *);
71 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
72 struct net *, struct sockaddr *,
74 #ifdef CONFIG_DEBUG_LOCK_ALLOC
75 static struct lock_class_key svc_key[2];
76 static struct lock_class_key svc_slock_key[2];
78 static void svc_reclassify_socket(struct socket *sock)
80 struct sock *sk = sock->sk;
82 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
85 switch (sk->sk_family) {
87 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
89 "sk_xprt.xpt_lock-AF_INET-NFSD",
94 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
96 "sk_xprt.xpt_lock-AF_INET6-NFSD",
105 static void svc_reclassify_socket(struct socket *sock)
111 * Release an skbuff after use
113 static void svc_release_skb(struct svc_rqst *rqstp)
115 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
118 struct svc_sock *svsk =
119 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
120 rqstp->rq_xprt_ctxt = NULL;
122 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
123 skb_free_datagram_locked(svsk->sk_sk, skb);
127 static void svc_release_udp_skb(struct svc_rqst *rqstp)
129 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
132 rqstp->rq_xprt_ctxt = NULL;
134 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
139 union svc_pktinfo_u {
140 struct in_pktinfo pkti;
141 struct in6_pktinfo pkti6;
143 #define SVC_PKTINFO_SPACE \
144 CMSG_SPACE(sizeof(union svc_pktinfo_u))
146 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
148 struct svc_sock *svsk =
149 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
150 switch (svsk->sk_sk->sk_family) {
152 struct in_pktinfo *pki = CMSG_DATA(cmh);
154 cmh->cmsg_level = SOL_IP;
155 cmh->cmsg_type = IP_PKTINFO;
156 pki->ipi_ifindex = 0;
157 pki->ipi_spec_dst.s_addr =
158 svc_daddr_in(rqstp)->sin_addr.s_addr;
159 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
164 struct in6_pktinfo *pki = CMSG_DATA(cmh);
165 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
167 cmh->cmsg_level = SOL_IPV6;
168 cmh->cmsg_type = IPV6_PKTINFO;
169 pki->ipi6_ifindex = daddr->sin6_scope_id;
170 pki->ipi6_addr = daddr->sin6_addr;
171 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
178 * send routine intended to be shared by the fore- and back-channel
180 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
181 struct page *headpage, unsigned long headoffset,
182 struct page *tailpage, unsigned long tailoffset)
186 struct page **ppage = xdr->pages;
187 size_t base = xdr->page_base;
188 unsigned int pglen = xdr->page_len;
189 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
196 if (slen == xdr->head[0].iov_len)
198 len = kernel_sendpage(sock, headpage, headoffset,
199 xdr->head[0].iov_len, flags);
200 if (len != xdr->head[0].iov_len)
202 slen -= xdr->head[0].iov_len;
207 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
211 result = kernel_sendpage(sock, *ppage, base, size, flags);
218 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
224 if (xdr->tail[0].iov_len) {
225 result = kernel_sendpage(sock, tailpage, tailoffset,
226 xdr->tail[0].iov_len, 0);
237 * Generic sendto routine
239 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
241 struct svc_sock *svsk =
242 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
243 struct socket *sock = svsk->sk_sock;
246 long all[SVC_PKTINFO_SPACE / sizeof(long)];
248 struct cmsghdr *cmh = &buffer.hdr;
250 unsigned long tailoff;
251 unsigned long headoff;
252 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
254 if (rqstp->rq_prot == IPPROTO_UDP) {
255 struct msghdr msg = {
256 .msg_name = &rqstp->rq_addr,
257 .msg_namelen = rqstp->rq_addrlen,
259 .msg_controllen = sizeof(buffer),
260 .msg_flags = MSG_MORE,
263 svc_set_cmsg_data(rqstp, cmh);
265 if (sock_sendmsg(sock, &msg) < 0)
269 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
271 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
272 rqstp->rq_respages[0], tailoff);
275 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
276 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
277 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
282 static int svc_sock_read_payload(struct svc_rqst *rqstp, unsigned int offset,
289 * Report socket names for nfsdfs
291 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
293 const struct sock *sk = svsk->sk_sk;
294 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
298 switch (sk->sk_family) {
300 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
302 &inet_sk(sk)->inet_rcv_saddr,
303 inet_sk(sk)->inet_num);
305 #if IS_ENABLED(CONFIG_IPV6)
307 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
309 &sk->sk_v6_rcv_saddr,
310 inet_sk(sk)->inet_num);
314 len = snprintf(buf, remaining, "*unknown-%d*\n",
318 if (len >= remaining) {
320 return -ENAMETOOLONG;
326 * Generic recvfrom routine.
328 static ssize_t svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov,
329 unsigned int nr, size_t buflen, unsigned int base)
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = { NULL };
336 rqstp->rq_xprt_hlen = 0;
338 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
339 iov_iter_kvec(&msg.msg_iter, READ, iov, nr, buflen);
341 iov_iter_advance(&msg.msg_iter, base);
344 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
345 /* If we read a full record, then assume there may be more
346 * data to read (stream based sockets only!)
349 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
351 dprintk("svc: socket %p recvfrom(%p, %zu) = %zd\n",
352 svsk, iov[0].iov_base, iov[0].iov_len, len);
357 * Set socket snd and rcv buffer lengths
359 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
361 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
362 struct socket *sock = svsk->sk_sock;
364 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
367 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
368 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
369 sock->sk->sk_write_space(sock->sk);
370 release_sock(sock->sk);
373 static void svc_sock_secure_port(struct svc_rqst *rqstp)
375 if (svc_port_is_privileged(svc_addr(rqstp)))
376 set_bit(RQ_SECURE, &rqstp->rq_flags);
378 clear_bit(RQ_SECURE, &rqstp->rq_flags);
382 * INET callback when data has been received on the socket.
384 static void svc_data_ready(struct sock *sk)
386 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
389 dprintk("svc: socket %p(inet %p), busy=%d\n",
391 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
393 /* Refer to svc_setup_socket() for details. */
396 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
397 svc_xprt_enqueue(&svsk->sk_xprt);
402 * INET callback when space is newly available on the socket.
404 static void svc_write_space(struct sock *sk)
406 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
409 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
410 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
412 /* Refer to svc_setup_socket() for details. */
414 svsk->sk_owspace(sk);
415 svc_xprt_enqueue(&svsk->sk_xprt);
419 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
421 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
423 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
425 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
428 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
430 struct svc_sock *svsk;
432 struct linger no_linger = {
437 svsk = container_of(xprt, struct svc_sock, sk_xprt);
438 sock = svsk->sk_sock;
439 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
440 (char *)&no_linger, sizeof(no_linger));
444 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
446 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
449 struct in_pktinfo *pki = CMSG_DATA(cmh);
450 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
452 if (cmh->cmsg_type != IP_PKTINFO)
455 daddr->sin_family = AF_INET;
456 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
461 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
463 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
466 struct in6_pktinfo *pki = CMSG_DATA(cmh);
467 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
469 if (cmh->cmsg_type != IPV6_PKTINFO)
472 daddr->sin6_family = AF_INET6;
473 daddr->sin6_addr = pki->ipi6_addr;
474 daddr->sin6_scope_id = pki->ipi6_ifindex;
479 * Copy the UDP datagram's destination address to the rqstp structure.
480 * The 'destination' address in this case is the address to which the
481 * peer sent the datagram, i.e. our local address. For multihomed
482 * hosts, this can change from msg to msg. Note that only the IP
483 * address changes, the port number should remain the same.
485 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
488 switch (cmh->cmsg_level) {
490 return svc_udp_get_dest_address4(rqstp, cmh);
492 return svc_udp_get_dest_address6(rqstp, cmh);
499 * Receive a datagram from a UDP socket.
501 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
503 struct svc_sock *svsk =
504 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
505 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
509 long all[SVC_PKTINFO_SPACE / sizeof(long)];
511 struct cmsghdr *cmh = &buffer.hdr;
512 struct msghdr msg = {
513 .msg_name = svc_addr(rqstp),
515 .msg_controllen = sizeof(buffer),
516 .msg_flags = MSG_DONTWAIT,
521 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
522 /* udp sockets need large rcvbuf as all pending
523 * requests are still in that buffer. sndbuf must
524 * also be large enough that there is enough space
525 * for one reply per thread. We count all threads
526 * rather than threads in a particular pool, which
527 * provides an upper bound on the number of threads
528 * which will access the socket.
530 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
532 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
534 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
535 0, 0, MSG_PEEK | MSG_DONTWAIT);
537 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
540 if (err != -EAGAIN) {
541 /* possibly an icmp error */
542 dprintk("svc: recvfrom returned error %d\n", -err);
543 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
547 len = svc_addr_len(svc_addr(rqstp));
548 rqstp->rq_addrlen = len;
549 if (skb->tstamp == 0) {
550 skb->tstamp = ktime_get_real();
551 /* Don't enable netstamp, sunrpc doesn't
552 need that much accuracy */
554 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
555 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
558 rqstp->rq_arg.len = len;
560 rqstp->rq_prot = IPPROTO_UDP;
562 if (!svc_udp_get_dest_address(rqstp, cmh)) {
563 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
564 cmh->cmsg_level, cmh->cmsg_type);
567 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
569 if (skb_is_nonlinear(skb)) {
570 /* we have to copy */
572 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
580 /* we can use it in-place */
581 rqstp->rq_arg.head[0].iov_base = skb->data;
582 rqstp->rq_arg.head[0].iov_len = len;
583 if (skb_checksum_complete(skb))
585 rqstp->rq_xprt_ctxt = skb;
588 rqstp->rq_arg.page_base = 0;
589 if (len <= rqstp->rq_arg.head[0].iov_len) {
590 rqstp->rq_arg.head[0].iov_len = len;
591 rqstp->rq_arg.page_len = 0;
592 rqstp->rq_respages = rqstp->rq_pages+1;
594 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
595 rqstp->rq_respages = rqstp->rq_pages + 1 +
596 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
598 rqstp->rq_next_page = rqstp->rq_respages+1;
601 serv->sv_stats->netudpcnt++;
610 svc_udp_sendto(struct svc_rqst *rqstp)
614 svc_release_udp_skb(rqstp);
616 error = svc_sendto(rqstp, &rqstp->rq_res);
617 if (error == -ECONNREFUSED)
618 /* ICMP error on earlier request. */
619 error = svc_sendto(rqstp, &rqstp->rq_res);
624 static int svc_udp_has_wspace(struct svc_xprt *xprt)
626 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
627 struct svc_serv *serv = xprt->xpt_server;
628 unsigned long required;
631 * Set the SOCK_NOSPACE flag before checking the available
634 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
635 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
636 if (required*2 > sock_wspace(svsk->sk_sk))
638 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
642 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
648 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
652 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
654 struct sockaddr *sa, int salen,
657 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
660 static const struct svc_xprt_ops svc_udp_ops = {
661 .xpo_create = svc_udp_create,
662 .xpo_recvfrom = svc_udp_recvfrom,
663 .xpo_sendto = svc_udp_sendto,
664 .xpo_read_payload = svc_sock_read_payload,
665 .xpo_release_rqst = svc_release_udp_skb,
666 .xpo_detach = svc_sock_detach,
667 .xpo_free = svc_sock_free,
668 .xpo_has_wspace = svc_udp_has_wspace,
669 .xpo_accept = svc_udp_accept,
670 .xpo_secure_port = svc_sock_secure_port,
671 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
674 static struct svc_xprt_class svc_udp_class = {
676 .xcl_owner = THIS_MODULE,
677 .xcl_ops = &svc_udp_ops,
678 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
679 .xcl_ident = XPRT_TRANSPORT_UDP,
682 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
684 int err, level, optname, one = 1;
686 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
687 &svsk->sk_xprt, serv);
688 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
689 svsk->sk_sk->sk_data_ready = svc_data_ready;
690 svsk->sk_sk->sk_write_space = svc_write_space;
692 /* initialise setting must have enough space to
693 * receive and respond to one request.
694 * svc_udp_recvfrom will re-adjust if necessary
696 svc_sock_setbufsize(svsk, 3);
698 /* data might have come in before data_ready set up */
699 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
700 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
702 /* make sure we get destination address info */
703 switch (svsk->sk_sk->sk_family) {
706 optname = IP_PKTINFO;
710 optname = IPV6_RECVPKTINFO;
715 err = kernel_setsockopt(svsk->sk_sock, level, optname,
716 (char *)&one, sizeof(one));
717 dprintk("svc: kernel_setsockopt returned %d\n", err);
721 * A data_ready event on a listening socket means there's a connection
722 * pending. Do not use state_change as a substitute for it.
724 static void svc_tcp_listen_data_ready(struct sock *sk)
726 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
728 dprintk("svc: socket %p TCP (listen) state change %d\n",
732 /* Refer to svc_setup_socket() for details. */
738 * This callback may called twice when a new connection
739 * is established as a child socket inherits everything
740 * from a parent LISTEN socket.
741 * 1) data_ready method of the parent socket will be called
742 * when one of child sockets become ESTABLISHED.
743 * 2) data_ready method of the child socket may be called
744 * when it receives data before the socket is accepted.
745 * In case of 2, we should ignore it silently.
747 if (sk->sk_state == TCP_LISTEN) {
749 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
750 svc_xprt_enqueue(&svsk->sk_xprt);
752 printk("svc: socket %p: no user data\n", sk);
757 * A state change on a connected socket means it's dying or dead.
759 static void svc_tcp_state_change(struct sock *sk)
761 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
763 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
764 sk, sk->sk_state, sk->sk_user_data);
767 printk("svc: socket %p: no user data\n", sk);
769 /* Refer to svc_setup_socket() for details. */
772 if (sk->sk_state != TCP_ESTABLISHED) {
773 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
774 svc_xprt_enqueue(&svsk->sk_xprt);
780 * Accept a TCP connection
782 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
784 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
785 struct sockaddr_storage addr;
786 struct sockaddr *sin = (struct sockaddr *) &addr;
787 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
788 struct socket *sock = svsk->sk_sock;
789 struct socket *newsock;
790 struct svc_sock *newsvsk;
792 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
794 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
798 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
799 err = kernel_accept(sock, &newsock, O_NONBLOCK);
802 printk(KERN_WARNING "%s: no more sockets!\n",
804 else if (err != -EAGAIN)
805 net_warn_ratelimited("%s: accept failed (err %d)!\n",
806 serv->sv_name, -err);
809 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
811 err = kernel_getpeername(newsock, sin);
813 net_warn_ratelimited("%s: peername failed (err %d)!\n",
814 serv->sv_name, -err);
815 goto failed; /* aborted connection or whatever */
819 /* Ideally, we would want to reject connections from unauthorized
820 * hosts here, but when we get encryption, the IP of the host won't
821 * tell us anything. For now just warn about unpriv connections.
823 if (!svc_port_is_privileged(sin)) {
824 dprintk("%s: connect from unprivileged port: %s\n",
826 __svc_print_addr(sin, buf, sizeof(buf)));
828 dprintk("%s: connect from %s\n", serv->sv_name,
829 __svc_print_addr(sin, buf, sizeof(buf)));
831 /* Reset the inherited callbacks before calling svc_setup_socket */
832 newsock->sk->sk_state_change = svsk->sk_ostate;
833 newsock->sk->sk_data_ready = svsk->sk_odata;
834 newsock->sk->sk_write_space = svsk->sk_owspace;
836 /* make sure that a write doesn't block forever when
839 newsock->sk->sk_sndtimeo = HZ*30;
841 newsvsk = svc_setup_socket(serv, newsock,
842 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
845 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
846 err = kernel_getsockname(newsock, sin);
848 if (unlikely(err < 0)) {
849 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
850 slen = offsetof(struct sockaddr, sa_data);
852 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
854 if (sock_is_loopback(newsock->sk))
855 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
857 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
859 serv->sv_stats->nettcpconn++;
861 return &newsvsk->sk_xprt;
864 sock_release(newsock);
868 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
870 unsigned int i, len, npages;
872 if (svsk->sk_datalen == 0)
874 len = svsk->sk_datalen;
875 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
876 for (i = 0; i < npages; i++) {
877 if (rqstp->rq_pages[i] != NULL)
878 put_page(rqstp->rq_pages[i]);
879 BUG_ON(svsk->sk_pages[i] == NULL);
880 rqstp->rq_pages[i] = svsk->sk_pages[i];
881 svsk->sk_pages[i] = NULL;
883 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
887 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
889 unsigned int i, len, npages;
891 if (svsk->sk_datalen == 0)
893 len = svsk->sk_datalen;
894 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
895 for (i = 0; i < npages; i++) {
896 svsk->sk_pages[i] = rqstp->rq_pages[i];
897 rqstp->rq_pages[i] = NULL;
901 static void svc_tcp_clear_pages(struct svc_sock *svsk)
903 unsigned int i, len, npages;
905 if (svsk->sk_datalen == 0)
907 len = svsk->sk_datalen;
908 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
909 for (i = 0; i < npages; i++) {
910 if (svsk->sk_pages[i] == NULL) {
914 put_page(svsk->sk_pages[i]);
915 svsk->sk_pages[i] = NULL;
919 svsk->sk_datalen = 0;
923 * Receive fragment record header.
924 * If we haven't gotten the record length yet, get the next four bytes.
926 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
928 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
932 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
935 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
936 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
938 len = svc_recvfrom(rqstp, &iov, 1, want, 0);
941 svsk->sk_tcplen += len;
944 dprintk("svc: short recvfrom while reading record "
945 "length (%d of %d)\n", len, want);
949 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
950 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
952 net_notice_ratelimited("RPC: fragment too large: %d\n",
953 svc_sock_reclen(svsk));
958 return svc_sock_reclen(svsk);
960 dprintk("RPC: TCP recv_record got %d\n", len);
963 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
967 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
969 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
970 struct rpc_rqst *req = NULL;
971 struct kvec *src, *dst;
972 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
981 spin_lock(&bc_xprt->queue_lock);
982 req = xprt_lookup_rqst(bc_xprt, xid);
984 goto unlock_notfound;
986 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
988 * XXX!: cheating for now! Only copying HEAD.
989 * But we know this is good enough for now (in fact, for any
990 * callback reply in the forseeable future).
992 dst = &req->rq_private_buf.head[0];
993 src = &rqstp->rq_arg.head[0];
994 if (dst->iov_len < src->iov_len)
995 goto unlock_eagain; /* whatever; just giving up. */
996 memcpy(dst->iov_base, src->iov_base, src->iov_len);
997 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
998 rqstp->rq_arg.len = 0;
999 spin_unlock(&bc_xprt->queue_lock);
1003 "%s: Got unrecognized reply: "
1004 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1005 __func__, ntohl(calldir),
1006 bc_xprt, ntohl(xid));
1008 spin_unlock(&bc_xprt->queue_lock);
1012 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1018 vec[i].iov_base = page_address(pages[i]);
1019 vec[i].iov_len = PAGE_SIZE;
1026 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1028 /* If we have more data, signal svc_xprt_enqueue() to try again */
1029 dprintk("svc: TCP %s record (%d bytes)\n",
1030 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1031 svc_sock_reclen(svsk));
1032 svsk->sk_tcplen = 0;
1033 svsk->sk_reclen = 0;
1037 * Receive data from a TCP socket.
1039 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1041 struct svc_sock *svsk =
1042 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1043 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1046 unsigned int want, base;
1051 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1052 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1053 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1054 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1056 len = svc_tcp_recv_record(svsk, rqstp);
1060 base = svc_tcp_restore_pages(svsk, rqstp);
1061 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1063 vec = rqstp->rq_vec;
1065 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want);
1067 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1068 rqstp->rq_next_page = rqstp->rq_respages + 1;
1070 /* Now receive data */
1071 len = svc_recvfrom(rqstp, vec, pnum, base + want, base);
1073 svsk->sk_tcplen += len;
1074 svsk->sk_datalen += len;
1076 if (len != want || !svc_sock_final_rec(svsk)) {
1077 svc_tcp_save_pages(svsk, rqstp);
1078 if (len < 0 && len != -EAGAIN)
1081 svc_tcp_fragment_received(svsk);
1083 dprintk("svc: incomplete TCP record (%d of %d)\n",
1084 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1085 svc_sock_reclen(svsk));
1089 if (svsk->sk_datalen < 8) {
1090 svsk->sk_datalen = 0;
1091 goto err_delete; /* client is nuts. */
1094 rqstp->rq_arg.len = svsk->sk_datalen;
1095 rqstp->rq_arg.page_base = 0;
1096 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1097 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1098 rqstp->rq_arg.page_len = 0;
1100 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1102 rqstp->rq_xprt_ctxt = NULL;
1103 rqstp->rq_prot = IPPROTO_TCP;
1104 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1105 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1107 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1109 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1112 len = receive_cb_reply(svsk, rqstp);
1114 /* Reset TCP read info */
1115 svsk->sk_datalen = 0;
1116 svc_tcp_fragment_received(svsk);
1121 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1123 serv->sv_stats->nettcpcnt++;
1125 return rqstp->rq_arg.len;
1130 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1133 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1134 svsk->sk_xprt.xpt_server->sv_name, -len);
1135 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1137 return 0; /* record not complete */
1141 * Send out data on TCP socket.
1143 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1145 struct xdr_buf *xbufp = &rqstp->rq_res;
1149 svc_release_skb(rqstp);
1151 /* Set up the first element of the reply kvec.
1152 * Any other kvecs that may be in use have been taken
1153 * care of by the server implementation itself.
1155 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1156 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1158 sent = svc_sendto(rqstp, &rqstp->rq_res);
1159 if (sent != xbufp->len) {
1161 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1162 "- shutting down socket\n",
1163 rqstp->rq_xprt->xpt_server->sv_name,
1164 (sent<0)?"got error":"sent only",
1166 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1167 svc_xprt_enqueue(rqstp->rq_xprt);
1173 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1175 struct sockaddr *sa, int salen,
1178 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1181 static const struct svc_xprt_ops svc_tcp_ops = {
1182 .xpo_create = svc_tcp_create,
1183 .xpo_recvfrom = svc_tcp_recvfrom,
1184 .xpo_sendto = svc_tcp_sendto,
1185 .xpo_read_payload = svc_sock_read_payload,
1186 .xpo_release_rqst = svc_release_skb,
1187 .xpo_detach = svc_tcp_sock_detach,
1188 .xpo_free = svc_sock_free,
1189 .xpo_has_wspace = svc_tcp_has_wspace,
1190 .xpo_accept = svc_tcp_accept,
1191 .xpo_secure_port = svc_sock_secure_port,
1192 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1195 static struct svc_xprt_class svc_tcp_class = {
1197 .xcl_owner = THIS_MODULE,
1198 .xcl_ops = &svc_tcp_ops,
1199 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1200 .xcl_ident = XPRT_TRANSPORT_TCP,
1203 void svc_init_xprt_sock(void)
1205 svc_reg_xprt_class(&svc_tcp_class);
1206 svc_reg_xprt_class(&svc_udp_class);
1209 void svc_cleanup_xprt_sock(void)
1211 svc_unreg_xprt_class(&svc_tcp_class);
1212 svc_unreg_xprt_class(&svc_udp_class);
1215 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1217 struct sock *sk = svsk->sk_sk;
1219 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1220 &svsk->sk_xprt, serv);
1221 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1222 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1223 if (sk->sk_state == TCP_LISTEN) {
1224 dprintk("setting up TCP socket for listening\n");
1225 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1226 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1227 sk->sk_data_ready = svc_tcp_listen_data_ready;
1228 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1230 dprintk("setting up TCP socket for reading\n");
1231 sk->sk_state_change = svc_tcp_state_change;
1232 sk->sk_data_ready = svc_data_ready;
1233 sk->sk_write_space = svc_write_space;
1235 svsk->sk_reclen = 0;
1236 svsk->sk_tcplen = 0;
1237 svsk->sk_datalen = 0;
1238 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1240 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1242 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1243 switch (sk->sk_state) {
1245 case TCP_ESTABLISHED:
1248 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1253 void svc_sock_update_bufs(struct svc_serv *serv)
1256 * The number of server threads has changed. Update
1257 * rcvbuf and sndbuf accordingly on all sockets
1259 struct svc_sock *svsk;
1261 spin_lock_bh(&serv->sv_lock);
1262 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1263 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1264 spin_unlock_bh(&serv->sv_lock);
1266 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1269 * Initialize socket for RPC use and create svc_sock struct
1271 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1272 struct socket *sock,
1275 struct svc_sock *svsk;
1277 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1280 dprintk("svc: svc_setup_socket %p\n", sock);
1281 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1283 return ERR_PTR(-ENOMEM);
1287 /* Register socket with portmapper */
1289 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1291 ntohs(inet_sk(inet)->inet_sport));
1295 return ERR_PTR(err);
1298 svsk->sk_sock = sock;
1300 svsk->sk_ostate = inet->sk_state_change;
1301 svsk->sk_odata = inet->sk_data_ready;
1302 svsk->sk_owspace = inet->sk_write_space;
1304 * This barrier is necessary in order to prevent race condition
1305 * with svc_data_ready(), svc_listen_data_ready() and others
1306 * when calling callbacks above.
1309 inet->sk_user_data = svsk;
1311 /* Initialize the socket */
1312 if (sock->type == SOCK_DGRAM)
1313 svc_udp_init(svsk, serv);
1315 svc_tcp_init(svsk, serv);
1317 dprintk("svc: svc_setup_socket created %p (inet %p), "
1318 "listen %d close %d\n",
1320 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1321 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1326 bool svc_alien_sock(struct net *net, int fd)
1329 struct socket *sock = sockfd_lookup(fd, &err);
1334 if (sock_net(sock->sk) != net)
1340 EXPORT_SYMBOL_GPL(svc_alien_sock);
1343 * svc_addsock - add a listener socket to an RPC service
1344 * @serv: pointer to RPC service to which to add a new listener
1345 * @fd: file descriptor of the new listener
1346 * @name_return: pointer to buffer to fill in with name of listener
1347 * @len: size of the buffer
1350 * Fills in socket name and returns positive length of name if successful.
1351 * Name is terminated with '\n'. On error, returns a negative errno
1354 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1355 const size_t len, const struct cred *cred)
1358 struct socket *so = sockfd_lookup(fd, &err);
1359 struct svc_sock *svsk = NULL;
1360 struct sockaddr_storage addr;
1361 struct sockaddr *sin = (struct sockaddr *)&addr;
1366 err = -EAFNOSUPPORT;
1367 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1369 err = -EPROTONOSUPPORT;
1370 if (so->sk->sk_protocol != IPPROTO_TCP &&
1371 so->sk->sk_protocol != IPPROTO_UDP)
1374 if (so->state > SS_UNCONNECTED)
1377 if (!try_module_get(THIS_MODULE))
1379 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1381 module_put(THIS_MODULE);
1382 err = PTR_ERR(svsk);
1385 salen = kernel_getsockname(svsk->sk_sock, sin);
1387 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1388 svsk->sk_xprt.xpt_cred = get_cred(cred);
1389 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1390 return svc_one_sock_name(svsk, name_return, len);
1395 EXPORT_SYMBOL_GPL(svc_addsock);
1398 * Create socket for RPC service.
1400 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1403 struct sockaddr *sin, int len,
1406 struct svc_sock *svsk;
1407 struct socket *sock;
1410 struct sockaddr_storage addr;
1411 struct sockaddr *newsin = (struct sockaddr *)&addr;
1415 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1417 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1418 serv->sv_program->pg_name, protocol,
1419 __svc_print_addr(sin, buf, sizeof(buf)));
1421 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1422 printk(KERN_WARNING "svc: only UDP and TCP "
1423 "sockets supported\n");
1424 return ERR_PTR(-EINVAL);
1427 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1428 switch (sin->sa_family) {
1436 return ERR_PTR(-EINVAL);
1439 error = __sock_create(net, family, type, protocol, &sock, 1);
1441 return ERR_PTR(error);
1443 svc_reclassify_socket(sock);
1446 * If this is an PF_INET6 listener, we want to avoid
1447 * getting requests from IPv4 remotes. Those should
1448 * be shunted to a PF_INET listener via rpcbind.
1451 if (family == PF_INET6)
1452 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1453 (char *)&val, sizeof(val));
1455 if (type == SOCK_STREAM)
1456 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1457 error = kernel_bind(sock, sin, len);
1461 error = kernel_getsockname(sock, newsin);
1466 if (protocol == IPPROTO_TCP) {
1467 if ((error = kernel_listen(sock, 64)) < 0)
1471 svsk = svc_setup_socket(serv, sock, flags);
1473 error = PTR_ERR(svsk);
1476 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1477 return (struct svc_xprt *)svsk;
1479 dprintk("svc: svc_create_socket error = %d\n", -error);
1481 return ERR_PTR(error);
1485 * Detach the svc_sock from the socket so that no
1486 * more callbacks occur.
1488 static void svc_sock_detach(struct svc_xprt *xprt)
1490 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1491 struct sock *sk = svsk->sk_sk;
1493 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1495 /* put back the old socket callbacks */
1497 sk->sk_state_change = svsk->sk_ostate;
1498 sk->sk_data_ready = svsk->sk_odata;
1499 sk->sk_write_space = svsk->sk_owspace;
1500 sk->sk_user_data = NULL;
1505 * Disconnect the socket, and reset the callbacks
1507 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1509 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1511 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1513 svc_sock_detach(xprt);
1515 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1516 svc_tcp_clear_pages(svsk);
1517 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1522 * Free the svc_sock's socket resources and the svc_sock itself.
1524 static void svc_sock_free(struct svc_xprt *xprt)
1526 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1527 dprintk("svc: svc_sock_free(%p)\n", svsk);
1529 if (svsk->sk_sock->file)
1530 sockfd_put(svsk->sk_sock);
1532 sock_release(svsk->sk_sock);