2 * linux/net/sunrpc/svcsock.c
4 * These are the RPC server socket internals.
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/errno.h>
25 #include <linux/fcntl.h>
26 #include <linux/net.h>
28 #include <linux/inet.h>
29 #include <linux/udp.h>
30 #include <linux/tcp.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/netdevice.h>
34 #include <linux/skbuff.h>
35 #include <linux/file.h>
36 #include <linux/freezer.h>
38 #include <net/checksum.h>
41 #include <net/tcp_states.h>
42 #include <asm/uaccess.h>
43 #include <asm/ioctls.h>
45 #include <linux/sunrpc/types.h>
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/sunrpc/xdr.h>
48 #include <linux/sunrpc/svcsock.h>
49 #include <linux/sunrpc/stats.h>
51 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
54 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
55 int *errp, int flags);
56 static void svc_udp_data_ready(struct sock *, int);
57 static int svc_udp_recvfrom(struct svc_rqst *);
58 static int svc_udp_sendto(struct svc_rqst *);
59 static void svc_sock_detach(struct svc_xprt *);
60 static void svc_sock_free(struct svc_xprt *);
62 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
63 struct sockaddr *, int, int);
64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
65 static struct lock_class_key svc_key[2];
66 static struct lock_class_key svc_slock_key[2];
68 static void svc_reclassify_socket(struct socket *sock)
70 struct sock *sk = sock->sk;
71 BUG_ON(sock_owned_by_user(sk));
72 switch (sk->sk_family) {
74 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
76 "sk_xprt.xpt_lock-AF_INET-NFSD",
81 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
83 "sk_xprt.xpt_lock-AF_INET6-NFSD",
92 static void svc_reclassify_socket(struct socket *sock)
98 * Release an skbuff after use
100 static void svc_release_skb(struct svc_rqst *rqstp)
102 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
103 struct svc_deferred_req *dr = rqstp->rq_deferred;
106 struct svc_sock *svsk =
107 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
108 rqstp->rq_xprt_ctxt = NULL;
110 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
111 skb_free_datagram(svsk->sk_sk, skb);
114 rqstp->rq_deferred = NULL;
119 union svc_pktinfo_u {
120 struct in_pktinfo pkti;
121 struct in6_pktinfo pkti6;
123 #define SVC_PKTINFO_SPACE \
124 CMSG_SPACE(sizeof(union svc_pktinfo_u))
126 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
128 struct svc_sock *svsk =
129 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
130 switch (svsk->sk_sk->sk_family) {
132 struct in_pktinfo *pki = CMSG_DATA(cmh);
134 cmh->cmsg_level = SOL_IP;
135 cmh->cmsg_type = IP_PKTINFO;
136 pki->ipi_ifindex = 0;
137 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
138 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
143 struct in6_pktinfo *pki = CMSG_DATA(cmh);
145 cmh->cmsg_level = SOL_IPV6;
146 cmh->cmsg_type = IPV6_PKTINFO;
147 pki->ipi6_ifindex = 0;
148 ipv6_addr_copy(&pki->ipi6_addr,
149 &rqstp->rq_daddr.addr6);
150 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
158 * Generic sendto routine
160 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
162 struct svc_sock *svsk =
163 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
164 struct socket *sock = svsk->sk_sock;
168 long all[SVC_PKTINFO_SPACE / sizeof(long)];
170 struct cmsghdr *cmh = &buffer.hdr;
174 struct page **ppage = xdr->pages;
175 size_t base = xdr->page_base;
176 unsigned int pglen = xdr->page_len;
177 unsigned int flags = MSG_MORE;
178 char buf[RPC_MAX_ADDRBUFLEN];
182 if (rqstp->rq_prot == IPPROTO_UDP) {
183 struct msghdr msg = {
184 .msg_name = &rqstp->rq_addr,
185 .msg_namelen = rqstp->rq_addrlen,
187 .msg_controllen = sizeof(buffer),
188 .msg_flags = MSG_MORE,
191 svc_set_cmsg_data(rqstp, cmh);
193 if (sock_sendmsg(sock, &msg, 0) < 0)
198 if (slen == xdr->head[0].iov_len)
200 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
201 xdr->head[0].iov_len, flags);
202 if (len != xdr->head[0].iov_len)
204 slen -= xdr->head[0].iov_len;
209 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
213 result = kernel_sendpage(sock, *ppage, base, size, flags);
220 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
225 if (xdr->tail[0].iov_len) {
226 result = kernel_sendpage(sock, rqstp->rq_respages[0],
227 ((unsigned long)xdr->tail[0].iov_base)
229 xdr->tail[0].iov_len, 0);
235 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
236 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
237 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
243 * Report socket names for nfsdfs
245 static int one_sock_name(char *buf, struct svc_sock *svsk)
249 switch(svsk->sk_sk->sk_family) {
251 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
252 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
254 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
255 inet_sk(svsk->sk_sk)->num);
258 len = sprintf(buf, "*unknown-%d*\n",
259 svsk->sk_sk->sk_family);
265 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
267 struct svc_sock *svsk, *closesk = NULL;
272 spin_lock_bh(&serv->sv_lock);
273 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
274 int onelen = one_sock_name(buf+len, svsk);
275 if (toclose && strcmp(toclose, buf+len) == 0)
280 spin_unlock_bh(&serv->sv_lock);
282 /* Should unregister with portmap, but you cannot
283 * unregister just one protocol...
285 svc_close_xprt(&closesk->sk_xprt);
290 EXPORT_SYMBOL(svc_sock_names);
293 * Check input queue length
295 static int svc_recv_available(struct svc_sock *svsk)
297 struct socket *sock = svsk->sk_sock;
300 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
302 return (err >= 0)? avail : err;
306 * Generic recvfrom routine.
308 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
311 struct svc_sock *svsk =
312 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
313 struct msghdr msg = {
314 .msg_flags = MSG_DONTWAIT,
318 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
321 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
322 svsk, iov[0].iov_base, iov[0].iov_len, len);
327 * Set socket snd and rcv buffer lengths
329 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
334 oldfs = get_fs(); set_fs(KERNEL_DS);
335 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
336 (char*)&snd, sizeof(snd));
337 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
338 (char*)&rcv, sizeof(rcv));
340 /* sock_setsockopt limits use to sysctl_?mem_max,
341 * which isn't acceptable. Until that is made conditional
342 * on not having CAP_SYS_RESOURCE or similar, we go direct...
343 * DaveM said I could!
346 sock->sk->sk_sndbuf = snd * 2;
347 sock->sk->sk_rcvbuf = rcv * 2;
348 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
349 release_sock(sock->sk);
353 * INET callback when data has been received on the socket.
355 static void svc_udp_data_ready(struct sock *sk, int count)
357 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
360 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
362 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
363 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
364 svc_xprt_enqueue(&svsk->sk_xprt);
366 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
367 wake_up_interruptible(sk->sk_sleep);
371 * INET callback when space is newly available on the socket.
373 static void svc_write_space(struct sock *sk)
375 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
378 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
379 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
380 svc_xprt_enqueue(&svsk->sk_xprt);
383 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
384 dprintk("RPC svc_write_space: someone sleeping on %p\n",
386 wake_up_interruptible(sk->sk_sleep);
391 * Copy the UDP datagram's destination address to the rqstp structure.
392 * The 'destination' address in this case is the address to which the
393 * peer sent the datagram, i.e. our local address. For multihomed
394 * hosts, this can change from msg to msg. Note that only the IP
395 * address changes, the port number should remain the same.
397 static void svc_udp_get_dest_address(struct svc_rqst *rqstp,
400 struct svc_sock *svsk =
401 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
402 switch (svsk->sk_sk->sk_family) {
404 struct in_pktinfo *pki = CMSG_DATA(cmh);
405 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
409 struct in6_pktinfo *pki = CMSG_DATA(cmh);
410 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
417 * Receive a datagram from a UDP socket.
419 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
421 struct svc_sock *svsk =
422 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
423 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
427 long all[SVC_PKTINFO_SPACE / sizeof(long)];
429 struct cmsghdr *cmh = &buffer.hdr;
431 struct msghdr msg = {
432 .msg_name = svc_addr(rqstp),
434 .msg_controllen = sizeof(buffer),
435 .msg_flags = MSG_DONTWAIT,
438 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
439 /* udp sockets need large rcvbuf as all pending
440 * requests are still in that buffer. sndbuf must
441 * also be large enough that there is enough space
442 * for one reply per thread. We count all threads
443 * rather than threads in a particular pool, which
444 * provides an upper bound on the number of threads
445 * which will access the socket.
447 svc_sock_setbufsize(svsk->sk_sock,
448 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
449 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
451 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
453 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
454 0, 0, MSG_PEEK | MSG_DONTWAIT);
456 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
459 if (err != -EAGAIN) {
460 /* possibly an icmp error */
461 dprintk("svc: recvfrom returned error %d\n", -err);
462 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
464 svc_xprt_received(&svsk->sk_xprt);
467 len = svc_addr_len(svc_addr(rqstp));
470 rqstp->rq_addrlen = len;
471 if (skb->tstamp.tv64 == 0) {
472 skb->tstamp = ktime_get_real();
473 /* Don't enable netstamp, sunrpc doesn't
474 need that much accuracy */
476 svsk->sk_sk->sk_stamp = skb->tstamp;
477 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
480 * Maybe more packets - kick another thread ASAP.
482 svc_xprt_received(&svsk->sk_xprt);
484 len = skb->len - sizeof(struct udphdr);
485 rqstp->rq_arg.len = len;
487 rqstp->rq_prot = IPPROTO_UDP;
489 if (cmh->cmsg_level != IPPROTO_IP ||
490 cmh->cmsg_type != IP_PKTINFO) {
492 printk("rpcsvc: received unknown control message:"
494 cmh->cmsg_level, cmh->cmsg_type);
495 skb_free_datagram(svsk->sk_sk, skb);
498 svc_udp_get_dest_address(rqstp, cmh);
500 if (skb_is_nonlinear(skb)) {
501 /* we have to copy */
503 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
506 skb_free_datagram(svsk->sk_sk, skb);
510 skb_free_datagram(svsk->sk_sk, skb);
512 /* we can use it in-place */
513 rqstp->rq_arg.head[0].iov_base = skb->data +
514 sizeof(struct udphdr);
515 rqstp->rq_arg.head[0].iov_len = len;
516 if (skb_checksum_complete(skb)) {
517 skb_free_datagram(svsk->sk_sk, skb);
520 rqstp->rq_xprt_ctxt = skb;
523 rqstp->rq_arg.page_base = 0;
524 if (len <= rqstp->rq_arg.head[0].iov_len) {
525 rqstp->rq_arg.head[0].iov_len = len;
526 rqstp->rq_arg.page_len = 0;
527 rqstp->rq_respages = rqstp->rq_pages+1;
529 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
530 rqstp->rq_respages = rqstp->rq_pages + 1 +
531 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
535 serv->sv_stats->netudpcnt++;
541 svc_udp_sendto(struct svc_rqst *rqstp)
545 error = svc_sendto(rqstp, &rqstp->rq_res);
546 if (error == -ECONNREFUSED)
547 /* ICMP error on earlier request. */
548 error = svc_sendto(rqstp, &rqstp->rq_res);
553 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
557 static int svc_udp_has_wspace(struct svc_xprt *xprt)
559 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
560 struct svc_serv *serv = xprt->xpt_server;
561 unsigned long required;
564 * Set the SOCK_NOSPACE flag before checking the available
567 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
568 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
569 if (required*2 > sock_wspace(svsk->sk_sk))
571 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
575 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
581 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
582 struct sockaddr *sa, int salen,
585 return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
588 static struct svc_xprt_ops svc_udp_ops = {
589 .xpo_create = svc_udp_create,
590 .xpo_recvfrom = svc_udp_recvfrom,
591 .xpo_sendto = svc_udp_sendto,
592 .xpo_release_rqst = svc_release_skb,
593 .xpo_detach = svc_sock_detach,
594 .xpo_free = svc_sock_free,
595 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
596 .xpo_has_wspace = svc_udp_has_wspace,
597 .xpo_accept = svc_udp_accept,
600 static struct svc_xprt_class svc_udp_class = {
602 .xcl_owner = THIS_MODULE,
603 .xcl_ops = &svc_udp_ops,
604 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
607 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
612 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
613 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
614 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
615 svsk->sk_sk->sk_write_space = svc_write_space;
617 /* initialise setting must have enough space to
618 * receive and respond to one request.
619 * svc_udp_recvfrom will re-adjust if necessary
621 svc_sock_setbufsize(svsk->sk_sock,
622 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
623 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
625 /* data might have come in before data_ready set up */
626 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
627 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
631 /* make sure we get destination address info */
632 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
633 (char __user *)&one, sizeof(one));
638 * A data_ready event on a listening socket means there's a connection
639 * pending. Do not use state_change as a substitute for it.
641 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
643 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
645 dprintk("svc: socket %p TCP (listen) state change %d\n",
649 * This callback may called twice when a new connection
650 * is established as a child socket inherits everything
651 * from a parent LISTEN socket.
652 * 1) data_ready method of the parent socket will be called
653 * when one of child sockets become ESTABLISHED.
654 * 2) data_ready method of the child socket may be called
655 * when it receives data before the socket is accepted.
656 * In case of 2, we should ignore it silently.
658 if (sk->sk_state == TCP_LISTEN) {
660 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
661 svc_xprt_enqueue(&svsk->sk_xprt);
663 printk("svc: socket %p: no user data\n", sk);
666 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
667 wake_up_interruptible_all(sk->sk_sleep);
671 * A state change on a connected socket means it's dying or dead.
673 static void svc_tcp_state_change(struct sock *sk)
675 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
677 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
678 sk, sk->sk_state, sk->sk_user_data);
681 printk("svc: socket %p: no user data\n", sk);
683 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
684 svc_xprt_enqueue(&svsk->sk_xprt);
686 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
687 wake_up_interruptible_all(sk->sk_sleep);
690 static void svc_tcp_data_ready(struct sock *sk, int count)
692 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
694 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
695 sk, sk->sk_user_data);
697 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
698 svc_xprt_enqueue(&svsk->sk_xprt);
700 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
701 wake_up_interruptible(sk->sk_sleep);
705 * Accept a TCP connection
707 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
709 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
710 struct sockaddr_storage addr;
711 struct sockaddr *sin = (struct sockaddr *) &addr;
712 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
713 struct socket *sock = svsk->sk_sock;
714 struct socket *newsock;
715 struct svc_sock *newsvsk;
717 char buf[RPC_MAX_ADDRBUFLEN];
719 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
723 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
724 err = kernel_accept(sock, &newsock, O_NONBLOCK);
727 printk(KERN_WARNING "%s: no more sockets!\n",
729 else if (err != -EAGAIN && net_ratelimit())
730 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
731 serv->sv_name, -err);
734 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
736 err = kernel_getpeername(newsock, sin, &slen);
739 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
740 serv->sv_name, -err);
741 goto failed; /* aborted connection or whatever */
744 /* Ideally, we would want to reject connections from unauthorized
745 * hosts here, but when we get encryption, the IP of the host won't
746 * tell us anything. For now just warn about unpriv connections.
748 if (!svc_port_is_privileged(sin)) {
750 "%s: connect from unprivileged port: %s\n",
752 __svc_print_addr(sin, buf, sizeof(buf)));
754 dprintk("%s: connect from %s\n", serv->sv_name,
755 __svc_print_addr(sin, buf, sizeof(buf)));
757 /* make sure that a write doesn't block forever when
760 newsock->sk->sk_sndtimeo = HZ*30;
762 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
763 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
765 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
766 err = kernel_getsockname(newsock, sin, &slen);
767 if (unlikely(err < 0)) {
768 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
769 slen = offsetof(struct sockaddr, sa_data);
771 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
774 serv->sv_stats->nettcpconn++;
776 return &newsvsk->sk_xprt;
779 sock_release(newsock);
784 * Receive data from a TCP socket.
786 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
788 struct svc_sock *svsk =
789 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
790 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
795 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
796 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
797 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
798 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
800 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
801 /* sndbuf needs to have room for one request
802 * per thread, otherwise we can stall even when the
803 * network isn't a bottleneck.
805 * We count all threads rather than threads in a
806 * particular pool, which provides an upper bound
807 * on the number of threads which will access the socket.
809 * rcvbuf just needs to be able to hold a few requests.
810 * Normally they will be removed from the queue
811 * as soon a a complete request arrives.
813 svc_sock_setbufsize(svsk->sk_sock,
814 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
815 3 * serv->sv_max_mesg);
817 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
819 /* Receive data. If we haven't got the record length yet, get
820 * the next four bytes. Otherwise try to gobble up as much as
821 * possible up to the complete record length.
823 if (svsk->sk_tcplen < 4) {
824 unsigned long want = 4 - svsk->sk_tcplen;
827 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
829 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
831 svsk->sk_tcplen += len;
834 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
836 svc_xprt_received(&svsk->sk_xprt);
837 return -EAGAIN; /* record header not complete */
840 svsk->sk_reclen = ntohl(svsk->sk_reclen);
841 if (!(svsk->sk_reclen & 0x80000000)) {
842 /* FIXME: technically, a record can be fragmented,
843 * and non-terminal fragments will not have the top
844 * bit set in the fragment length header.
845 * But apparently no known nfs clients send fragmented
848 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
850 (unsigned long) svsk->sk_reclen);
853 svsk->sk_reclen &= 0x7fffffff;
854 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
855 if (svsk->sk_reclen > serv->sv_max_mesg) {
857 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
859 (unsigned long) svsk->sk_reclen);
864 /* Check whether enough data is available */
865 len = svc_recv_available(svsk);
869 if (len < svsk->sk_reclen) {
870 dprintk("svc: incomplete TCP record (%d of %d)\n",
871 len, svsk->sk_reclen);
872 svc_xprt_received(&svsk->sk_xprt);
873 return -EAGAIN; /* record not complete */
875 len = svsk->sk_reclen;
876 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
879 vec[0] = rqstp->rq_arg.head[0];
883 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
884 vec[pnum].iov_len = PAGE_SIZE;
888 rqstp->rq_respages = &rqstp->rq_pages[pnum];
890 /* Now receive data */
891 len = svc_recvfrom(rqstp, vec, pnum, len);
895 dprintk("svc: TCP complete record (%d bytes)\n", len);
896 rqstp->rq_arg.len = len;
897 rqstp->rq_arg.page_base = 0;
898 if (len <= rqstp->rq_arg.head[0].iov_len) {
899 rqstp->rq_arg.head[0].iov_len = len;
900 rqstp->rq_arg.page_len = 0;
902 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
905 rqstp->rq_xprt_ctxt = NULL;
906 rqstp->rq_prot = IPPROTO_TCP;
908 /* Reset TCP read info */
912 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
913 svc_xprt_received(&svsk->sk_xprt);
915 serv->sv_stats->nettcpcnt++;
920 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
924 if (len == -EAGAIN) {
925 dprintk("RPC: TCP recvfrom got EAGAIN\n");
926 svc_xprt_received(&svsk->sk_xprt);
928 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
929 svsk->sk_xprt.xpt_server->sv_name, -len);
937 * Send out data on TCP socket.
939 static int svc_tcp_sendto(struct svc_rqst *rqstp)
941 struct xdr_buf *xbufp = &rqstp->rq_res;
945 /* Set up the first element of the reply kvec.
946 * Any other kvecs that may be in use have been taken
947 * care of by the server implementation itself.
949 reclen = htonl(0x80000000|((xbufp->len ) - 4));
950 memcpy(xbufp->head[0].iov_base, &reclen, 4);
952 if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
955 sent = svc_sendto(rqstp, &rqstp->rq_res);
956 if (sent != xbufp->len) {
958 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
959 "- shutting down socket\n",
960 rqstp->rq_xprt->xpt_server->sv_name,
961 (sent<0)?"got error":"sent only",
963 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
964 svc_xprt_enqueue(rqstp->rq_xprt);
971 * Setup response header. TCP has a 4B record length field.
973 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
975 struct kvec *resv = &rqstp->rq_res.head[0];
977 /* tcp needs a space for the record length... */
981 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
983 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
984 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
989 * Set the SOCK_NOSPACE flag before checking the available
992 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
993 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
994 wspace = sk_stream_wspace(svsk->sk_sk);
996 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
998 if (required * 2 > wspace)
1001 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1005 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1006 struct sockaddr *sa, int salen,
1009 return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1012 static struct svc_xprt_ops svc_tcp_ops = {
1013 .xpo_create = svc_tcp_create,
1014 .xpo_recvfrom = svc_tcp_recvfrom,
1015 .xpo_sendto = svc_tcp_sendto,
1016 .xpo_release_rqst = svc_release_skb,
1017 .xpo_detach = svc_sock_detach,
1018 .xpo_free = svc_sock_free,
1019 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1020 .xpo_has_wspace = svc_tcp_has_wspace,
1021 .xpo_accept = svc_tcp_accept,
1024 static struct svc_xprt_class svc_tcp_class = {
1026 .xcl_owner = THIS_MODULE,
1027 .xcl_ops = &svc_tcp_ops,
1028 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1031 void svc_init_xprt_sock(void)
1033 svc_reg_xprt_class(&svc_tcp_class);
1034 svc_reg_xprt_class(&svc_udp_class);
1037 void svc_cleanup_xprt_sock(void)
1039 svc_unreg_xprt_class(&svc_tcp_class);
1040 svc_unreg_xprt_class(&svc_udp_class);
1043 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1045 struct sock *sk = svsk->sk_sk;
1046 struct tcp_sock *tp = tcp_sk(sk);
1048 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1049 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1050 if (sk->sk_state == TCP_LISTEN) {
1051 dprintk("setting up TCP socket for listening\n");
1052 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1053 sk->sk_data_ready = svc_tcp_listen_data_ready;
1054 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1056 dprintk("setting up TCP socket for reading\n");
1057 sk->sk_state_change = svc_tcp_state_change;
1058 sk->sk_data_ready = svc_tcp_data_ready;
1059 sk->sk_write_space = svc_write_space;
1061 svsk->sk_reclen = 0;
1062 svsk->sk_tcplen = 0;
1064 tp->nonagle = 1; /* disable Nagle's algorithm */
1066 /* initialise setting must have enough space to
1067 * receive and respond to one request.
1068 * svc_tcp_recvfrom will re-adjust if necessary
1070 svc_sock_setbufsize(svsk->sk_sock,
1071 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1072 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1074 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1075 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1076 if (sk->sk_state != TCP_ESTABLISHED)
1077 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1081 void svc_sock_update_bufs(struct svc_serv *serv)
1084 * The number of server threads has changed. Update
1085 * rcvbuf and sndbuf accordingly on all sockets
1087 struct list_head *le;
1089 spin_lock_bh(&serv->sv_lock);
1090 list_for_each(le, &serv->sv_permsocks) {
1091 struct svc_sock *svsk =
1092 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1093 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1095 list_for_each(le, &serv->sv_tempsocks) {
1096 struct svc_sock *svsk =
1097 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1098 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1100 spin_unlock_bh(&serv->sv_lock);
1104 * Initialize socket for RPC use and create svc_sock struct
1105 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1107 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1108 struct socket *sock,
1109 int *errp, int flags)
1111 struct svc_sock *svsk;
1113 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1115 dprintk("svc: svc_setup_socket %p\n", sock);
1116 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1123 /* Register socket with portmapper */
1124 if (*errp >= 0 && pmap_register)
1125 *errp = svc_register(serv, inet->sk_protocol,
1126 ntohs(inet_sk(inet)->sport));
1133 inet->sk_user_data = svsk;
1134 svsk->sk_sock = sock;
1136 svsk->sk_ostate = inet->sk_state_change;
1137 svsk->sk_odata = inet->sk_data_ready;
1138 svsk->sk_owspace = inet->sk_write_space;
1140 /* Initialize the socket */
1141 if (sock->type == SOCK_DGRAM)
1142 svc_udp_init(svsk, serv);
1144 svc_tcp_init(svsk, serv);
1146 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1152 int svc_addsock(struct svc_serv *serv,
1158 struct socket *so = sockfd_lookup(fd, &err);
1159 struct svc_sock *svsk = NULL;
1163 if (so->sk->sk_family != AF_INET)
1164 err = -EAFNOSUPPORT;
1165 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1166 so->sk->sk_protocol != IPPROTO_UDP)
1167 err = -EPROTONOSUPPORT;
1168 else if (so->state > SS_UNCONNECTED)
1171 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
1173 struct sockaddr_storage addr;
1174 struct sockaddr *sin = (struct sockaddr *)&addr;
1176 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1177 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1178 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1179 spin_lock_bh(&serv->sv_lock);
1180 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1181 spin_unlock_bh(&serv->sv_lock);
1182 svc_xprt_received(&svsk->sk_xprt);
1190 if (proto) *proto = so->sk->sk_protocol;
1191 return one_sock_name(name_return, svsk);
1193 EXPORT_SYMBOL_GPL(svc_addsock);
1196 * Create socket for RPC service.
1198 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1200 struct sockaddr *sin, int len,
1203 struct svc_sock *svsk;
1204 struct socket *sock;
1207 char buf[RPC_MAX_ADDRBUFLEN];
1208 struct sockaddr_storage addr;
1209 struct sockaddr *newsin = (struct sockaddr *)&addr;
1212 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1213 serv->sv_program->pg_name, protocol,
1214 __svc_print_addr(sin, buf, sizeof(buf)));
1216 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1217 printk(KERN_WARNING "svc: only UDP and TCP "
1218 "sockets supported\n");
1219 return ERR_PTR(-EINVAL);
1221 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1223 error = sock_create_kern(sin->sa_family, type, protocol, &sock);
1225 return ERR_PTR(error);
1227 svc_reclassify_socket(sock);
1229 if (type == SOCK_STREAM)
1230 sock->sk->sk_reuse = 1; /* allow address reuse */
1231 error = kernel_bind(sock, sin, len);
1236 error = kernel_getsockname(sock, newsin, &newlen);
1240 if (protocol == IPPROTO_TCP) {
1241 if ((error = kernel_listen(sock, 64)) < 0)
1245 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1246 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1247 return (struct svc_xprt *)svsk;
1251 dprintk("svc: svc_create_socket error = %d\n", -error);
1253 return ERR_PTR(error);
1257 * Detach the svc_sock from the socket so that no
1258 * more callbacks occur.
1260 static void svc_sock_detach(struct svc_xprt *xprt)
1262 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1263 struct sock *sk = svsk->sk_sk;
1265 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1267 /* put back the old socket callbacks */
1268 sk->sk_state_change = svsk->sk_ostate;
1269 sk->sk_data_ready = svsk->sk_odata;
1270 sk->sk_write_space = svsk->sk_owspace;
1274 * Free the svc_sock's socket resources and the svc_sock itself.
1276 static void svc_sock_free(struct svc_xprt *xprt)
1278 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1279 dprintk("svc: svc_sock_free(%p)\n", svsk);
1281 if (svsk->sk_sock->file)
1282 sockfd_put(svsk->sk_sock);
1284 sock_release(svsk->sk_sock);