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>
42 #include <net/tcp_states.h>
43 #include <asm/uaccess.h>
44 #include <asm/ioctls.h>
46 #include <linux/sunrpc/types.h>
47 #include <linux/sunrpc/clnt.h>
48 #include <linux/sunrpc/xdr.h>
49 #include <linux/sunrpc/svcsock.h>
50 #include <linux/sunrpc/stats.h>
52 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
55 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
56 int *errp, int flags);
57 static void svc_udp_data_ready(struct sock *, int);
58 static int svc_udp_recvfrom(struct svc_rqst *);
59 static int svc_udp_sendto(struct svc_rqst *);
60 static void svc_sock_detach(struct svc_xprt *);
61 static void svc_sock_free(struct svc_xprt *);
63 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
64 struct sockaddr *, int, int);
65 #ifdef CONFIG_DEBUG_LOCK_ALLOC
66 static struct lock_class_key svc_key[2];
67 static struct lock_class_key svc_slock_key[2];
69 static void svc_reclassify_socket(struct socket *sock)
71 struct sock *sk = sock->sk;
72 BUG_ON(sock_owned_by_user(sk));
73 switch (sk->sk_family) {
75 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
77 "sk_xprt.xpt_lock-AF_INET-NFSD",
82 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
84 "sk_xprt.xpt_lock-AF_INET6-NFSD",
93 static void svc_reclassify_socket(struct socket *sock)
99 * Release an skbuff after use
101 static void svc_release_skb(struct svc_rqst *rqstp)
103 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
104 struct svc_deferred_req *dr = rqstp->rq_deferred;
107 struct svc_sock *svsk =
108 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
109 rqstp->rq_xprt_ctxt = NULL;
111 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
112 skb_free_datagram(svsk->sk_sk, skb);
115 rqstp->rq_deferred = NULL;
120 union svc_pktinfo_u {
121 struct in_pktinfo pkti;
122 struct in6_pktinfo pkti6;
124 #define SVC_PKTINFO_SPACE \
125 CMSG_SPACE(sizeof(union svc_pktinfo_u))
127 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
129 struct svc_sock *svsk =
130 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
131 switch (svsk->sk_sk->sk_family) {
133 struct in_pktinfo *pki = CMSG_DATA(cmh);
135 cmh->cmsg_level = SOL_IP;
136 cmh->cmsg_type = IP_PKTINFO;
137 pki->ipi_ifindex = 0;
138 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
139 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
144 struct in6_pktinfo *pki = CMSG_DATA(cmh);
146 cmh->cmsg_level = SOL_IPV6;
147 cmh->cmsg_type = IPV6_PKTINFO;
148 pki->ipi6_ifindex = 0;
149 ipv6_addr_copy(&pki->ipi6_addr,
150 &rqstp->rq_daddr.addr6);
151 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
159 * Generic sendto routine
161 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
163 struct svc_sock *svsk =
164 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
165 struct socket *sock = svsk->sk_sock;
169 long all[SVC_PKTINFO_SPACE / sizeof(long)];
171 struct cmsghdr *cmh = &buffer.hdr;
175 struct page **ppage = xdr->pages;
176 size_t base = xdr->page_base;
177 unsigned int pglen = xdr->page_len;
178 unsigned int flags = MSG_MORE;
179 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
183 if (rqstp->rq_prot == IPPROTO_UDP) {
184 struct msghdr msg = {
185 .msg_name = &rqstp->rq_addr,
186 .msg_namelen = rqstp->rq_addrlen,
188 .msg_controllen = sizeof(buffer),
189 .msg_flags = MSG_MORE,
192 svc_set_cmsg_data(rqstp, cmh);
194 if (sock_sendmsg(sock, &msg, 0) < 0)
199 if (slen == xdr->head[0].iov_len)
201 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
202 xdr->head[0].iov_len, flags);
203 if (len != xdr->head[0].iov_len)
205 slen -= xdr->head[0].iov_len;
210 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
214 result = kernel_sendpage(sock, *ppage, base, size, flags);
221 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
226 if (xdr->tail[0].iov_len) {
227 result = kernel_sendpage(sock, rqstp->rq_respages[0],
228 ((unsigned long)xdr->tail[0].iov_base)
230 xdr->tail[0].iov_len, 0);
236 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
237 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
238 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
244 * Report socket names for nfsdfs
246 static int one_sock_name(char *buf, struct svc_sock *svsk)
250 switch(svsk->sk_sk->sk_family) {
252 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
253 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
255 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
256 inet_sk(svsk->sk_sk)->num);
259 len = sprintf(buf, "*unknown-%d*\n",
260 svsk->sk_sk->sk_family);
266 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
268 struct svc_sock *svsk, *closesk = NULL;
273 spin_lock_bh(&serv->sv_lock);
274 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
275 int onelen = one_sock_name(buf+len, svsk);
276 if (toclose && strcmp(toclose, buf+len) == 0)
281 spin_unlock_bh(&serv->sv_lock);
283 /* Should unregister with portmap, but you cannot
284 * unregister just one protocol...
286 svc_close_xprt(&closesk->sk_xprt);
291 EXPORT_SYMBOL(svc_sock_names);
294 * Check input queue length
296 static int svc_recv_available(struct svc_sock *svsk)
298 struct socket *sock = svsk->sk_sock;
301 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
303 return (err >= 0)? avail : err;
307 * Generic recvfrom routine.
309 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
312 struct svc_sock *svsk =
313 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
314 struct msghdr msg = {
315 .msg_flags = MSG_DONTWAIT,
319 rqstp->rq_xprt_hlen = 0;
321 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
324 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
325 svsk, iov[0].iov_base, iov[0].iov_len, len);
330 * Set socket snd and rcv buffer lengths
332 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
337 oldfs = get_fs(); set_fs(KERNEL_DS);
338 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
339 (char*)&snd, sizeof(snd));
340 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
341 (char*)&rcv, sizeof(rcv));
343 /* sock_setsockopt limits use to sysctl_?mem_max,
344 * which isn't acceptable. Until that is made conditional
345 * on not having CAP_SYS_RESOURCE or similar, we go direct...
346 * DaveM said I could!
349 sock->sk->sk_sndbuf = snd * 2;
350 sock->sk->sk_rcvbuf = rcv * 2;
351 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
352 release_sock(sock->sk);
356 * INET callback when data has been received on the socket.
358 static void svc_udp_data_ready(struct sock *sk, int count)
360 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
363 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
365 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
366 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
367 svc_xprt_enqueue(&svsk->sk_xprt);
369 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
370 wake_up_interruptible(sk->sk_sleep);
374 * INET callback when space is newly available on the socket.
376 static void svc_write_space(struct sock *sk)
378 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
381 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
382 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
383 svc_xprt_enqueue(&svsk->sk_xprt);
386 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
387 dprintk("RPC svc_write_space: someone sleeping on %p\n",
389 wake_up_interruptible(sk->sk_sleep);
394 * Copy the UDP datagram's destination address to the rqstp structure.
395 * The 'destination' address in this case is the address to which the
396 * peer sent the datagram, i.e. our local address. For multihomed
397 * hosts, this can change from msg to msg. Note that only the IP
398 * address changes, the port number should remain the same.
400 static void svc_udp_get_dest_address(struct svc_rqst *rqstp,
403 struct svc_sock *svsk =
404 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
405 switch (svsk->sk_sk->sk_family) {
407 struct in_pktinfo *pki = CMSG_DATA(cmh);
408 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
412 struct in6_pktinfo *pki = CMSG_DATA(cmh);
413 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
420 * Receive a datagram from a UDP socket.
422 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
424 struct svc_sock *svsk =
425 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
426 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
430 long all[SVC_PKTINFO_SPACE / sizeof(long)];
432 struct cmsghdr *cmh = &buffer.hdr;
434 struct msghdr msg = {
435 .msg_name = svc_addr(rqstp),
437 .msg_controllen = sizeof(buffer),
438 .msg_flags = MSG_DONTWAIT,
441 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
442 /* udp sockets need large rcvbuf as all pending
443 * requests are still in that buffer. sndbuf must
444 * also be large enough that there is enough space
445 * for one reply per thread. We count all threads
446 * rather than threads in a particular pool, which
447 * provides an upper bound on the number of threads
448 * which will access the socket.
450 svc_sock_setbufsize(svsk->sk_sock,
451 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
452 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
454 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
456 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
457 0, 0, MSG_PEEK | MSG_DONTWAIT);
459 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
462 if (err != -EAGAIN) {
463 /* possibly an icmp error */
464 dprintk("svc: recvfrom returned error %d\n", -err);
465 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
467 svc_xprt_received(&svsk->sk_xprt);
470 len = svc_addr_len(svc_addr(rqstp));
473 rqstp->rq_addrlen = len;
474 if (skb->tstamp.tv64 == 0) {
475 skb->tstamp = ktime_get_real();
476 /* Don't enable netstamp, sunrpc doesn't
477 need that much accuracy */
479 svsk->sk_sk->sk_stamp = skb->tstamp;
480 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
483 * Maybe more packets - kick another thread ASAP.
485 svc_xprt_received(&svsk->sk_xprt);
487 len = skb->len - sizeof(struct udphdr);
488 rqstp->rq_arg.len = len;
490 rqstp->rq_prot = IPPROTO_UDP;
492 if (cmh->cmsg_level != IPPROTO_IP ||
493 cmh->cmsg_type != IP_PKTINFO) {
495 printk("rpcsvc: received unknown control message:"
497 cmh->cmsg_level, cmh->cmsg_type);
498 skb_free_datagram(svsk->sk_sk, skb);
501 svc_udp_get_dest_address(rqstp, cmh);
503 if (skb_is_nonlinear(skb)) {
504 /* we have to copy */
506 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
509 skb_free_datagram(svsk->sk_sk, skb);
513 skb_free_datagram(svsk->sk_sk, skb);
515 /* we can use it in-place */
516 rqstp->rq_arg.head[0].iov_base = skb->data +
517 sizeof(struct udphdr);
518 rqstp->rq_arg.head[0].iov_len = len;
519 if (skb_checksum_complete(skb)) {
520 skb_free_datagram(svsk->sk_sk, skb);
523 rqstp->rq_xprt_ctxt = skb;
526 rqstp->rq_arg.page_base = 0;
527 if (len <= rqstp->rq_arg.head[0].iov_len) {
528 rqstp->rq_arg.head[0].iov_len = len;
529 rqstp->rq_arg.page_len = 0;
530 rqstp->rq_respages = rqstp->rq_pages+1;
532 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
533 rqstp->rq_respages = rqstp->rq_pages + 1 +
534 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
538 serv->sv_stats->netudpcnt++;
544 svc_udp_sendto(struct svc_rqst *rqstp)
548 error = svc_sendto(rqstp, &rqstp->rq_res);
549 if (error == -ECONNREFUSED)
550 /* ICMP error on earlier request. */
551 error = svc_sendto(rqstp, &rqstp->rq_res);
556 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
560 static int svc_udp_has_wspace(struct svc_xprt *xprt)
562 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
563 struct svc_serv *serv = xprt->xpt_server;
564 unsigned long required;
567 * Set the SOCK_NOSPACE flag before checking the available
570 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
571 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
572 if (required*2 > sock_wspace(svsk->sk_sk))
574 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
578 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
584 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
585 struct sockaddr *sa, int salen,
588 return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
591 static struct svc_xprt_ops svc_udp_ops = {
592 .xpo_create = svc_udp_create,
593 .xpo_recvfrom = svc_udp_recvfrom,
594 .xpo_sendto = svc_udp_sendto,
595 .xpo_release_rqst = svc_release_skb,
596 .xpo_detach = svc_sock_detach,
597 .xpo_free = svc_sock_free,
598 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
599 .xpo_has_wspace = svc_udp_has_wspace,
600 .xpo_accept = svc_udp_accept,
603 static struct svc_xprt_class svc_udp_class = {
605 .xcl_owner = THIS_MODULE,
606 .xcl_ops = &svc_udp_ops,
607 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
610 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
615 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
616 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
617 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
618 svsk->sk_sk->sk_write_space = svc_write_space;
620 /* initialise setting must have enough space to
621 * receive and respond to one request.
622 * svc_udp_recvfrom will re-adjust if necessary
624 svc_sock_setbufsize(svsk->sk_sock,
625 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
626 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
628 /* data might have come in before data_ready set up */
629 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
630 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
634 /* make sure we get destination address info */
635 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
636 (char __user *)&one, sizeof(one));
641 * A data_ready event on a listening socket means there's a connection
642 * pending. Do not use state_change as a substitute for it.
644 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
646 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
648 dprintk("svc: socket %p TCP (listen) state change %d\n",
652 * This callback may called twice when a new connection
653 * is established as a child socket inherits everything
654 * from a parent LISTEN socket.
655 * 1) data_ready method of the parent socket will be called
656 * when one of child sockets become ESTABLISHED.
657 * 2) data_ready method of the child socket may be called
658 * when it receives data before the socket is accepted.
659 * In case of 2, we should ignore it silently.
661 if (sk->sk_state == TCP_LISTEN) {
663 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
664 svc_xprt_enqueue(&svsk->sk_xprt);
666 printk("svc: socket %p: no user data\n", sk);
669 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
670 wake_up_interruptible_all(sk->sk_sleep);
674 * A state change on a connected socket means it's dying or dead.
676 static void svc_tcp_state_change(struct sock *sk)
678 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
680 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
681 sk, sk->sk_state, sk->sk_user_data);
684 printk("svc: socket %p: no user data\n", sk);
686 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
687 svc_xprt_enqueue(&svsk->sk_xprt);
689 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
690 wake_up_interruptible_all(sk->sk_sleep);
693 static void svc_tcp_data_ready(struct sock *sk, int count)
695 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
697 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
698 sk, sk->sk_user_data);
700 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
701 svc_xprt_enqueue(&svsk->sk_xprt);
703 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
704 wake_up_interruptible(sk->sk_sleep);
708 * Accept a TCP connection
710 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
712 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
713 struct sockaddr_storage addr;
714 struct sockaddr *sin = (struct sockaddr *) &addr;
715 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
716 struct socket *sock = svsk->sk_sock;
717 struct socket *newsock;
718 struct svc_sock *newsvsk;
720 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
722 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
726 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
727 err = kernel_accept(sock, &newsock, O_NONBLOCK);
730 printk(KERN_WARNING "%s: no more sockets!\n",
732 else if (err != -EAGAIN && net_ratelimit())
733 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
734 serv->sv_name, -err);
737 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
739 err = kernel_getpeername(newsock, sin, &slen);
742 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
743 serv->sv_name, -err);
744 goto failed; /* aborted connection or whatever */
747 /* Ideally, we would want to reject connections from unauthorized
748 * hosts here, but when we get encryption, the IP of the host won't
749 * tell us anything. For now just warn about unpriv connections.
751 if (!svc_port_is_privileged(sin)) {
753 "%s: connect from unprivileged port: %s\n",
755 __svc_print_addr(sin, buf, sizeof(buf)));
757 dprintk("%s: connect from %s\n", serv->sv_name,
758 __svc_print_addr(sin, buf, sizeof(buf)));
760 /* make sure that a write doesn't block forever when
763 newsock->sk->sk_sndtimeo = HZ*30;
765 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
766 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
768 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
769 err = kernel_getsockname(newsock, sin, &slen);
770 if (unlikely(err < 0)) {
771 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
772 slen = offsetof(struct sockaddr, sa_data);
774 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
777 serv->sv_stats->nettcpconn++;
779 return &newsvsk->sk_xprt;
782 sock_release(newsock);
787 * Receive data from a TCP socket.
789 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
791 struct svc_sock *svsk =
792 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
793 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
798 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
799 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
800 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
801 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
803 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
804 /* sndbuf needs to have room for one request
805 * per thread, otherwise we can stall even when the
806 * network isn't a bottleneck.
808 * We count all threads rather than threads in a
809 * particular pool, which provides an upper bound
810 * on the number of threads which will access the socket.
812 * rcvbuf just needs to be able to hold a few requests.
813 * Normally they will be removed from the queue
814 * as soon a a complete request arrives.
816 svc_sock_setbufsize(svsk->sk_sock,
817 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
818 3 * serv->sv_max_mesg);
820 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
822 /* Receive data. If we haven't got the record length yet, get
823 * the next four bytes. Otherwise try to gobble up as much as
824 * possible up to the complete record length.
826 if (svsk->sk_tcplen < 4) {
827 unsigned long want = 4 - svsk->sk_tcplen;
830 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
832 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
834 svsk->sk_tcplen += len;
837 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
839 svc_xprt_received(&svsk->sk_xprt);
840 return -EAGAIN; /* record header not complete */
843 svsk->sk_reclen = ntohl(svsk->sk_reclen);
844 if (!(svsk->sk_reclen & 0x80000000)) {
845 /* FIXME: technically, a record can be fragmented,
846 * and non-terminal fragments will not have the top
847 * bit set in the fragment length header.
848 * But apparently no known nfs clients send fragmented
851 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
853 (unsigned long) svsk->sk_reclen);
856 svsk->sk_reclen &= 0x7fffffff;
857 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
858 if (svsk->sk_reclen > serv->sv_max_mesg) {
860 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
862 (unsigned long) svsk->sk_reclen);
867 /* Check whether enough data is available */
868 len = svc_recv_available(svsk);
872 if (len < svsk->sk_reclen) {
873 dprintk("svc: incomplete TCP record (%d of %d)\n",
874 len, svsk->sk_reclen);
875 svc_xprt_received(&svsk->sk_xprt);
876 return -EAGAIN; /* record not complete */
878 len = svsk->sk_reclen;
879 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
882 vec[0] = rqstp->rq_arg.head[0];
886 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
887 vec[pnum].iov_len = PAGE_SIZE;
891 rqstp->rq_respages = &rqstp->rq_pages[pnum];
893 /* Now receive data */
894 len = svc_recvfrom(rqstp, vec, pnum, len);
898 dprintk("svc: TCP complete record (%d bytes)\n", len);
899 rqstp->rq_arg.len = len;
900 rqstp->rq_arg.page_base = 0;
901 if (len <= rqstp->rq_arg.head[0].iov_len) {
902 rqstp->rq_arg.head[0].iov_len = len;
903 rqstp->rq_arg.page_len = 0;
905 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
908 rqstp->rq_xprt_ctxt = NULL;
909 rqstp->rq_prot = IPPROTO_TCP;
911 /* Reset TCP read info */
915 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
916 svc_xprt_received(&svsk->sk_xprt);
918 serv->sv_stats->nettcpcnt++;
923 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
927 if (len == -EAGAIN) {
928 dprintk("RPC: TCP recvfrom got EAGAIN\n");
929 svc_xprt_received(&svsk->sk_xprt);
931 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
932 svsk->sk_xprt.xpt_server->sv_name, -len);
940 * Send out data on TCP socket.
942 static int svc_tcp_sendto(struct svc_rqst *rqstp)
944 struct xdr_buf *xbufp = &rqstp->rq_res;
948 /* Set up the first element of the reply kvec.
949 * Any other kvecs that may be in use have been taken
950 * care of by the server implementation itself.
952 reclen = htonl(0x80000000|((xbufp->len ) - 4));
953 memcpy(xbufp->head[0].iov_base, &reclen, 4);
955 if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
958 sent = svc_sendto(rqstp, &rqstp->rq_res);
959 if (sent != xbufp->len) {
961 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
962 "- shutting down socket\n",
963 rqstp->rq_xprt->xpt_server->sv_name,
964 (sent<0)?"got error":"sent only",
966 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
967 svc_xprt_enqueue(rqstp->rq_xprt);
974 * Setup response header. TCP has a 4B record length field.
976 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
978 struct kvec *resv = &rqstp->rq_res.head[0];
980 /* tcp needs a space for the record length... */
984 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
986 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
987 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
992 * Set the SOCK_NOSPACE flag before checking the available
995 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
996 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
997 wspace = sk_stream_wspace(svsk->sk_sk);
999 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
1001 if (required * 2 > wspace)
1004 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1008 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1009 struct sockaddr *sa, int salen,
1012 return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1015 static struct svc_xprt_ops svc_tcp_ops = {
1016 .xpo_create = svc_tcp_create,
1017 .xpo_recvfrom = svc_tcp_recvfrom,
1018 .xpo_sendto = svc_tcp_sendto,
1019 .xpo_release_rqst = svc_release_skb,
1020 .xpo_detach = svc_sock_detach,
1021 .xpo_free = svc_sock_free,
1022 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1023 .xpo_has_wspace = svc_tcp_has_wspace,
1024 .xpo_accept = svc_tcp_accept,
1027 static struct svc_xprt_class svc_tcp_class = {
1029 .xcl_owner = THIS_MODULE,
1030 .xcl_ops = &svc_tcp_ops,
1031 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1034 void svc_init_xprt_sock(void)
1036 svc_reg_xprt_class(&svc_tcp_class);
1037 svc_reg_xprt_class(&svc_udp_class);
1040 void svc_cleanup_xprt_sock(void)
1042 svc_unreg_xprt_class(&svc_tcp_class);
1043 svc_unreg_xprt_class(&svc_udp_class);
1046 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1048 struct sock *sk = svsk->sk_sk;
1050 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1051 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1052 if (sk->sk_state == TCP_LISTEN) {
1053 dprintk("setting up TCP socket for listening\n");
1054 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1055 sk->sk_data_ready = svc_tcp_listen_data_ready;
1056 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1058 dprintk("setting up TCP socket for reading\n");
1059 sk->sk_state_change = svc_tcp_state_change;
1060 sk->sk_data_ready = svc_tcp_data_ready;
1061 sk->sk_write_space = svc_write_space;
1063 svsk->sk_reclen = 0;
1064 svsk->sk_tcplen = 0;
1066 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1068 /* initialise setting must have enough space to
1069 * receive and respond to one request.
1070 * svc_tcp_recvfrom will re-adjust if necessary
1072 svc_sock_setbufsize(svsk->sk_sock,
1073 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1074 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1076 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1077 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1078 if (sk->sk_state != TCP_ESTABLISHED)
1079 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1083 void svc_sock_update_bufs(struct svc_serv *serv)
1086 * The number of server threads has changed. Update
1087 * rcvbuf and sndbuf accordingly on all sockets
1089 struct list_head *le;
1091 spin_lock_bh(&serv->sv_lock);
1092 list_for_each(le, &serv->sv_permsocks) {
1093 struct svc_sock *svsk =
1094 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1095 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1097 list_for_each(le, &serv->sv_tempsocks) {
1098 struct svc_sock *svsk =
1099 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1100 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1102 spin_unlock_bh(&serv->sv_lock);
1104 EXPORT_SYMBOL(svc_sock_update_bufs);
1107 * Initialize socket for RPC use and create svc_sock struct
1108 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1110 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1111 struct socket *sock,
1112 int *errp, int flags)
1114 struct svc_sock *svsk;
1116 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1118 dprintk("svc: svc_setup_socket %p\n", sock);
1119 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1126 /* Register socket with portmapper */
1127 if (*errp >= 0 && pmap_register)
1128 *errp = svc_register(serv, inet->sk_protocol,
1129 ntohs(inet_sk(inet)->sport));
1136 inet->sk_user_data = svsk;
1137 svsk->sk_sock = sock;
1139 svsk->sk_ostate = inet->sk_state_change;
1140 svsk->sk_odata = inet->sk_data_ready;
1141 svsk->sk_owspace = inet->sk_write_space;
1143 /* Initialize the socket */
1144 if (sock->type == SOCK_DGRAM)
1145 svc_udp_init(svsk, serv);
1147 svc_tcp_init(svsk, serv);
1149 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1155 int svc_addsock(struct svc_serv *serv,
1161 struct socket *so = sockfd_lookup(fd, &err);
1162 struct svc_sock *svsk = NULL;
1166 if (so->sk->sk_family != AF_INET)
1167 err = -EAFNOSUPPORT;
1168 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1169 so->sk->sk_protocol != IPPROTO_UDP)
1170 err = -EPROTONOSUPPORT;
1171 else if (so->state > SS_UNCONNECTED)
1174 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
1176 struct sockaddr_storage addr;
1177 struct sockaddr *sin = (struct sockaddr *)&addr;
1179 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1180 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1181 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1182 spin_lock_bh(&serv->sv_lock);
1183 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1184 spin_unlock_bh(&serv->sv_lock);
1185 svc_xprt_received(&svsk->sk_xprt);
1193 if (proto) *proto = so->sk->sk_protocol;
1194 return one_sock_name(name_return, svsk);
1196 EXPORT_SYMBOL_GPL(svc_addsock);
1199 * Create socket for RPC service.
1201 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1203 struct sockaddr *sin, int len,
1206 struct svc_sock *svsk;
1207 struct socket *sock;
1210 struct sockaddr_storage addr;
1211 struct sockaddr *newsin = (struct sockaddr *)&addr;
1213 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1215 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1216 serv->sv_program->pg_name, protocol,
1217 __svc_print_addr(sin, buf, sizeof(buf)));
1219 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1220 printk(KERN_WARNING "svc: only UDP and TCP "
1221 "sockets supported\n");
1222 return ERR_PTR(-EINVAL);
1224 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1226 error = sock_create_kern(sin->sa_family, type, protocol, &sock);
1228 return ERR_PTR(error);
1230 svc_reclassify_socket(sock);
1232 if (type == SOCK_STREAM)
1233 sock->sk->sk_reuse = 1; /* allow address reuse */
1234 error = kernel_bind(sock, sin, len);
1239 error = kernel_getsockname(sock, newsin, &newlen);
1243 if (protocol == IPPROTO_TCP) {
1244 if ((error = kernel_listen(sock, 64)) < 0)
1248 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1249 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1250 return (struct svc_xprt *)svsk;
1254 dprintk("svc: svc_create_socket error = %d\n", -error);
1256 return ERR_PTR(error);
1260 * Detach the svc_sock from the socket so that no
1261 * more callbacks occur.
1263 static void svc_sock_detach(struct svc_xprt *xprt)
1265 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1266 struct sock *sk = svsk->sk_sk;
1268 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1270 /* put back the old socket callbacks */
1271 sk->sk_state_change = svsk->sk_ostate;
1272 sk->sk_data_ready = svsk->sk_odata;
1273 sk->sk_write_space = svsk->sk_owspace;
1277 * Free the svc_sock's socket resources and the svc_sock itself.
1279 static void svc_sock_free(struct svc_xprt *xprt)
1281 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1282 dprintk("svc: svc_sock_free(%p)\n", svsk);
1284 if (svsk->sk_sock->file)
1285 sockfd_put(svsk->sk_sock);
1287 sock_release(svsk->sk_sock);