xref: /linux/net/sunrpc/svcsock.c (revision bcefe12eff5dca6fdfa94ed85e5bee66380d5cd9)
1 /*
2  * linux/net/sunrpc/svcsock.c
3  *
4  * These are the RPC server socket internals.
5  *
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...
9  *
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.
18  *
19  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20  */
21 
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>
27 #include <linux/in.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>
37 #include <net/sock.h>
38 #include <net/checksum.h>
39 #include <net/ip.h>
40 #include <net/ipv6.h>
41 #include <net/tcp.h>
42 #include <net/tcp_states.h>
43 #include <asm/uaccess.h>
44 #include <asm/ioctls.h>
45 
46 #include <linux/sunrpc/types.h>
47 #include <linux/sunrpc/clnt.h>
48 #include <linux/sunrpc/xdr.h>
49 #include <linux/sunrpc/msg_prot.h>
50 #include <linux/sunrpc/svcsock.h>
51 #include <linux/sunrpc/stats.h>
52 #include <linux/sunrpc/xprt.h>
53 
54 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
55 
56 
57 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
58 					 int *errp, int flags);
59 static void		svc_udp_data_ready(struct sock *, int);
60 static int		svc_udp_recvfrom(struct svc_rqst *);
61 static int		svc_udp_sendto(struct svc_rqst *);
62 static void		svc_sock_detach(struct svc_xprt *);
63 static void		svc_tcp_sock_detach(struct svc_xprt *);
64 static void		svc_sock_free(struct svc_xprt *);
65 
66 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
67 					  struct sockaddr *, int, int);
68 #ifdef CONFIG_DEBUG_LOCK_ALLOC
69 static struct lock_class_key svc_key[2];
70 static struct lock_class_key svc_slock_key[2];
71 
72 static void svc_reclassify_socket(struct socket *sock)
73 {
74 	struct sock *sk = sock->sk;
75 	BUG_ON(sock_owned_by_user(sk));
76 	switch (sk->sk_family) {
77 	case AF_INET:
78 		sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
79 					      &svc_slock_key[0],
80 					      "sk_xprt.xpt_lock-AF_INET-NFSD",
81 					      &svc_key[0]);
82 		break;
83 
84 	case AF_INET6:
85 		sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
86 					      &svc_slock_key[1],
87 					      "sk_xprt.xpt_lock-AF_INET6-NFSD",
88 					      &svc_key[1]);
89 		break;
90 
91 	default:
92 		BUG();
93 	}
94 }
95 #else
96 static void svc_reclassify_socket(struct socket *sock)
97 {
98 }
99 #endif
100 
101 /*
102  * Release an skbuff after use
103  */
104 static void svc_release_skb(struct svc_rqst *rqstp)
105 {
106 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
107 
108 	if (skb) {
109 		struct svc_sock *svsk =
110 			container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
111 		rqstp->rq_xprt_ctxt = NULL;
112 
113 		dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
114 		skb_free_datagram_locked(svsk->sk_sk, skb);
115 	}
116 }
117 
118 union svc_pktinfo_u {
119 	struct in_pktinfo pkti;
120 	struct in6_pktinfo pkti6;
121 };
122 #define SVC_PKTINFO_SPACE \
123 	CMSG_SPACE(sizeof(union svc_pktinfo_u))
124 
125 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
126 {
127 	struct svc_sock *svsk =
128 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
129 	switch (svsk->sk_sk->sk_family) {
130 	case AF_INET: {
131 			struct in_pktinfo *pki = CMSG_DATA(cmh);
132 
133 			cmh->cmsg_level = SOL_IP;
134 			cmh->cmsg_type = IP_PKTINFO;
135 			pki->ipi_ifindex = 0;
136 			pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
137 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
138 		}
139 		break;
140 
141 	case AF_INET6: {
142 			struct in6_pktinfo *pki = CMSG_DATA(cmh);
143 
144 			cmh->cmsg_level = SOL_IPV6;
145 			cmh->cmsg_type = IPV6_PKTINFO;
146 			pki->ipi6_ifindex = 0;
147 			ipv6_addr_copy(&pki->ipi6_addr,
148 					&rqstp->rq_daddr.addr6);
149 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
150 		}
151 		break;
152 	}
153 	return;
154 }
155 
156 /*
157  * send routine intended to be shared by the fore- and back-channel
158  */
159 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
160 		    struct page *headpage, unsigned long headoffset,
161 		    struct page *tailpage, unsigned long tailoffset)
162 {
163 	int		result;
164 	int		size;
165 	struct page	**ppage = xdr->pages;
166 	size_t		base = xdr->page_base;
167 	unsigned int	pglen = xdr->page_len;
168 	unsigned int	flags = MSG_MORE;
169 	int		slen;
170 	int		len = 0;
171 
172 	slen = xdr->len;
173 
174 	/* send head */
175 	if (slen == xdr->head[0].iov_len)
176 		flags = 0;
177 	len = kernel_sendpage(sock, headpage, headoffset,
178 				  xdr->head[0].iov_len, flags);
179 	if (len != xdr->head[0].iov_len)
180 		goto out;
181 	slen -= xdr->head[0].iov_len;
182 	if (slen == 0)
183 		goto out;
184 
185 	/* send page data */
186 	size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
187 	while (pglen > 0) {
188 		if (slen == size)
189 			flags = 0;
190 		result = kernel_sendpage(sock, *ppage, base, size, flags);
191 		if (result > 0)
192 			len += result;
193 		if (result != size)
194 			goto out;
195 		slen -= size;
196 		pglen -= size;
197 		size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
198 		base = 0;
199 		ppage++;
200 	}
201 
202 	/* send tail */
203 	if (xdr->tail[0].iov_len) {
204 		result = kernel_sendpage(sock, tailpage, tailoffset,
205 				   xdr->tail[0].iov_len, 0);
206 		if (result > 0)
207 			len += result;
208 	}
209 
210 out:
211 	return len;
212 }
213 
214 
215 /*
216  * Generic sendto routine
217  */
218 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
219 {
220 	struct svc_sock	*svsk =
221 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
222 	struct socket	*sock = svsk->sk_sock;
223 	union {
224 		struct cmsghdr	hdr;
225 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
226 	} buffer;
227 	struct cmsghdr *cmh = &buffer.hdr;
228 	int		len = 0;
229 	unsigned long tailoff;
230 	unsigned long headoff;
231 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
232 
233 	if (rqstp->rq_prot == IPPROTO_UDP) {
234 		struct msghdr msg = {
235 			.msg_name	= &rqstp->rq_addr,
236 			.msg_namelen	= rqstp->rq_addrlen,
237 			.msg_control	= cmh,
238 			.msg_controllen	= sizeof(buffer),
239 			.msg_flags	= MSG_MORE,
240 		};
241 
242 		svc_set_cmsg_data(rqstp, cmh);
243 
244 		if (sock_sendmsg(sock, &msg, 0) < 0)
245 			goto out;
246 	}
247 
248 	tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
249 	headoff = 0;
250 	len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
251 			       rqstp->rq_respages[0], tailoff);
252 
253 out:
254 	dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
255 		svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
256 		xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
257 
258 	return len;
259 }
260 
261 /*
262  * Report socket names for nfsdfs
263  */
264 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
265 {
266 	const struct sock *sk = svsk->sk_sk;
267 	const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
268 							"udp" : "tcp";
269 	int len;
270 
271 	switch (sk->sk_family) {
272 	case PF_INET:
273 		len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
274 				proto_name,
275 				&inet_sk(sk)->rcv_saddr,
276 				inet_sk(sk)->num);
277 		break;
278 	case PF_INET6:
279 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
280 				proto_name,
281 				&inet6_sk(sk)->rcv_saddr,
282 				inet_sk(sk)->num);
283 		break;
284 	default:
285 		len = snprintf(buf, remaining, "*unknown-%d*\n",
286 				sk->sk_family);
287 	}
288 
289 	if (len >= remaining) {
290 		*buf = '\0';
291 		return -ENAMETOOLONG;
292 	}
293 	return len;
294 }
295 
296 /**
297  * svc_sock_names - construct a list of listener names in a string
298  * @serv: pointer to RPC service
299  * @buf: pointer to a buffer to fill in with socket names
300  * @buflen: size of the buffer to be filled
301  * @toclose: pointer to '\0'-terminated C string containing the name
302  *		of a listener to be closed
303  *
304  * Fills in @buf with a '\n'-separated list of names of listener
305  * sockets.  If @toclose is not NULL, the socket named by @toclose
306  * is closed, and is not included in the output list.
307  *
308  * Returns positive length of the socket name string, or a negative
309  * errno value on error.
310  */
311 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
312 		   const char *toclose)
313 {
314 	struct svc_sock *svsk, *closesk = NULL;
315 	int len = 0;
316 
317 	if (!serv)
318 		return 0;
319 
320 	spin_lock_bh(&serv->sv_lock);
321 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
322 		int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
323 		if (onelen < 0) {
324 			len = onelen;
325 			break;
326 		}
327 		if (toclose && strcmp(toclose, buf + len) == 0)
328 			closesk = svsk;
329 		else
330 			len += onelen;
331 	}
332 	spin_unlock_bh(&serv->sv_lock);
333 
334 	if (closesk)
335 		/* Should unregister with portmap, but you cannot
336 		 * unregister just one protocol...
337 		 */
338 		svc_close_xprt(&closesk->sk_xprt);
339 	else if (toclose)
340 		return -ENOENT;
341 	return len;
342 }
343 EXPORT_SYMBOL_GPL(svc_sock_names);
344 
345 /*
346  * Check input queue length
347  */
348 static int svc_recv_available(struct svc_sock *svsk)
349 {
350 	struct socket	*sock = svsk->sk_sock;
351 	int		avail, err;
352 
353 	err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
354 
355 	return (err >= 0)? avail : err;
356 }
357 
358 /*
359  * Generic recvfrom routine.
360  */
361 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
362 			int buflen)
363 {
364 	struct svc_sock *svsk =
365 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
366 	struct msghdr msg = {
367 		.msg_flags	= MSG_DONTWAIT,
368 	};
369 	int len;
370 
371 	rqstp->rq_xprt_hlen = 0;
372 
373 	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
374 				msg.msg_flags);
375 
376 	dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
377 		svsk, iov[0].iov_base, iov[0].iov_len, len);
378 	return len;
379 }
380 
381 /*
382  * Set socket snd and rcv buffer lengths
383  */
384 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
385 				unsigned int rcv)
386 {
387 #if 0
388 	mm_segment_t	oldfs;
389 	oldfs = get_fs(); set_fs(KERNEL_DS);
390 	sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
391 			(char*)&snd, sizeof(snd));
392 	sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
393 			(char*)&rcv, sizeof(rcv));
394 #else
395 	/* sock_setsockopt limits use to sysctl_?mem_max,
396 	 * which isn't acceptable.  Until that is made conditional
397 	 * on not having CAP_SYS_RESOURCE or similar, we go direct...
398 	 * DaveM said I could!
399 	 */
400 	lock_sock(sock->sk);
401 	sock->sk->sk_sndbuf = snd * 2;
402 	sock->sk->sk_rcvbuf = rcv * 2;
403 	sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
404 	sock->sk->sk_write_space(sock->sk);
405 	release_sock(sock->sk);
406 #endif
407 }
408 /*
409  * INET callback when data has been received on the socket.
410  */
411 static void svc_udp_data_ready(struct sock *sk, int count)
412 {
413 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
414 
415 	if (svsk) {
416 		dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
417 			svsk, sk, count,
418 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
419 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
420 		svc_xprt_enqueue(&svsk->sk_xprt);
421 	}
422 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
423 		wake_up_interruptible(sk->sk_sleep);
424 }
425 
426 /*
427  * INET callback when space is newly available on the socket.
428  */
429 static void svc_write_space(struct sock *sk)
430 {
431 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
432 
433 	if (svsk) {
434 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
435 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
436 		svc_xprt_enqueue(&svsk->sk_xprt);
437 	}
438 
439 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
440 		dprintk("RPC svc_write_space: someone sleeping on %p\n",
441 		       svsk);
442 		wake_up_interruptible(sk->sk_sleep);
443 	}
444 }
445 
446 static void svc_tcp_write_space(struct sock *sk)
447 {
448 	struct socket *sock = sk->sk_socket;
449 
450 	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
451 		clear_bit(SOCK_NOSPACE, &sock->flags);
452 	svc_write_space(sk);
453 }
454 
455 /*
456  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
457  */
458 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
459 				     struct cmsghdr *cmh)
460 {
461 	struct in_pktinfo *pki = CMSG_DATA(cmh);
462 	if (cmh->cmsg_type != IP_PKTINFO)
463 		return 0;
464 	rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
465 	return 1;
466 }
467 
468 /*
469  * See net/ipv6/datagram.c : datagram_recv_ctl
470  */
471 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
472 				     struct cmsghdr *cmh)
473 {
474 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
475 	if (cmh->cmsg_type != IPV6_PKTINFO)
476 		return 0;
477 	ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
478 	return 1;
479 }
480 
481 /*
482  * Copy the UDP datagram's destination address to the rqstp structure.
483  * The 'destination' address in this case is the address to which the
484  * peer sent the datagram, i.e. our local address. For multihomed
485  * hosts, this can change from msg to msg. Note that only the IP
486  * address changes, the port number should remain the same.
487  */
488 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
489 				    struct cmsghdr *cmh)
490 {
491 	switch (cmh->cmsg_level) {
492 	case SOL_IP:
493 		return svc_udp_get_dest_address4(rqstp, cmh);
494 	case SOL_IPV6:
495 		return svc_udp_get_dest_address6(rqstp, cmh);
496 	}
497 
498 	return 0;
499 }
500 
501 /*
502  * Receive a datagram from a UDP socket.
503  */
504 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
505 {
506 	struct svc_sock	*svsk =
507 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
508 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
509 	struct sk_buff	*skb;
510 	union {
511 		struct cmsghdr	hdr;
512 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
513 	} buffer;
514 	struct cmsghdr *cmh = &buffer.hdr;
515 	struct msghdr msg = {
516 		.msg_name = svc_addr(rqstp),
517 		.msg_control = cmh,
518 		.msg_controllen = sizeof(buffer),
519 		.msg_flags = MSG_DONTWAIT,
520 	};
521 	size_t len;
522 	int err;
523 
524 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
525 	    /* udp sockets need large rcvbuf as all pending
526 	     * requests are still in that buffer.  sndbuf must
527 	     * also be large enough that there is enough space
528 	     * for one reply per thread.  We count all threads
529 	     * rather than threads in a particular pool, which
530 	     * provides an upper bound on the number of threads
531 	     * which will access the socket.
532 	     */
533 	    svc_sock_setbufsize(svsk->sk_sock,
534 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
535 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
536 
537 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
538 	skb = NULL;
539 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
540 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
541 	if (err >= 0)
542 		skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
543 
544 	if (skb == NULL) {
545 		if (err != -EAGAIN) {
546 			/* possibly an icmp error */
547 			dprintk("svc: recvfrom returned error %d\n", -err);
548 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
549 		}
550 		svc_xprt_received(&svsk->sk_xprt);
551 		return -EAGAIN;
552 	}
553 	len = svc_addr_len(svc_addr(rqstp));
554 	if (len == 0)
555 		return -EAFNOSUPPORT;
556 	rqstp->rq_addrlen = len;
557 	if (skb->tstamp.tv64 == 0) {
558 		skb->tstamp = ktime_get_real();
559 		/* Don't enable netstamp, sunrpc doesn't
560 		   need that much accuracy */
561 	}
562 	svsk->sk_sk->sk_stamp = skb->tstamp;
563 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
564 
565 	/*
566 	 * Maybe more packets - kick another thread ASAP.
567 	 */
568 	svc_xprt_received(&svsk->sk_xprt);
569 
570 	len  = skb->len - sizeof(struct udphdr);
571 	rqstp->rq_arg.len = len;
572 
573 	rqstp->rq_prot = IPPROTO_UDP;
574 
575 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
576 		if (net_ratelimit())
577 			printk(KERN_WARNING
578 				"svc: received unknown control message %d/%d; "
579 				"dropping RPC reply datagram\n",
580 					cmh->cmsg_level, cmh->cmsg_type);
581 		skb_free_datagram_locked(svsk->sk_sk, skb);
582 		return 0;
583 	}
584 
585 	if (skb_is_nonlinear(skb)) {
586 		/* we have to copy */
587 		local_bh_disable();
588 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
589 			local_bh_enable();
590 			/* checksum error */
591 			skb_free_datagram_locked(svsk->sk_sk, skb);
592 			return 0;
593 		}
594 		local_bh_enable();
595 		skb_free_datagram_locked(svsk->sk_sk, skb);
596 	} else {
597 		/* we can use it in-place */
598 		rqstp->rq_arg.head[0].iov_base = skb->data +
599 			sizeof(struct udphdr);
600 		rqstp->rq_arg.head[0].iov_len = len;
601 		if (skb_checksum_complete(skb)) {
602 			skb_free_datagram_locked(svsk->sk_sk, skb);
603 			return 0;
604 		}
605 		rqstp->rq_xprt_ctxt = skb;
606 	}
607 
608 	rqstp->rq_arg.page_base = 0;
609 	if (len <= rqstp->rq_arg.head[0].iov_len) {
610 		rqstp->rq_arg.head[0].iov_len = len;
611 		rqstp->rq_arg.page_len = 0;
612 		rqstp->rq_respages = rqstp->rq_pages+1;
613 	} else {
614 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
615 		rqstp->rq_respages = rqstp->rq_pages + 1 +
616 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
617 	}
618 
619 	if (serv->sv_stats)
620 		serv->sv_stats->netudpcnt++;
621 
622 	return len;
623 }
624 
625 static int
626 svc_udp_sendto(struct svc_rqst *rqstp)
627 {
628 	int		error;
629 
630 	error = svc_sendto(rqstp, &rqstp->rq_res);
631 	if (error == -ECONNREFUSED)
632 		/* ICMP error on earlier request. */
633 		error = svc_sendto(rqstp, &rqstp->rq_res);
634 
635 	return error;
636 }
637 
638 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
639 {
640 }
641 
642 static int svc_udp_has_wspace(struct svc_xprt *xprt)
643 {
644 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
645 	struct svc_serv	*serv = xprt->xpt_server;
646 	unsigned long required;
647 
648 	/*
649 	 * Set the SOCK_NOSPACE flag before checking the available
650 	 * sock space.
651 	 */
652 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
653 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
654 	if (required*2 > sock_wspace(svsk->sk_sk))
655 		return 0;
656 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
657 	return 1;
658 }
659 
660 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
661 {
662 	BUG();
663 	return NULL;
664 }
665 
666 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
667 				       struct sockaddr *sa, int salen,
668 				       int flags)
669 {
670 	return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
671 }
672 
673 static struct svc_xprt_ops svc_udp_ops = {
674 	.xpo_create = svc_udp_create,
675 	.xpo_recvfrom = svc_udp_recvfrom,
676 	.xpo_sendto = svc_udp_sendto,
677 	.xpo_release_rqst = svc_release_skb,
678 	.xpo_detach = svc_sock_detach,
679 	.xpo_free = svc_sock_free,
680 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
681 	.xpo_has_wspace = svc_udp_has_wspace,
682 	.xpo_accept = svc_udp_accept,
683 };
684 
685 static struct svc_xprt_class svc_udp_class = {
686 	.xcl_name = "udp",
687 	.xcl_owner = THIS_MODULE,
688 	.xcl_ops = &svc_udp_ops,
689 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
690 };
691 
692 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
693 {
694 	int err, level, optname, one = 1;
695 
696 	svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
697 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
698 	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
699 	svsk->sk_sk->sk_write_space = svc_write_space;
700 
701 	/* initialise setting must have enough space to
702 	 * receive and respond to one request.
703 	 * svc_udp_recvfrom will re-adjust if necessary
704 	 */
705 	svc_sock_setbufsize(svsk->sk_sock,
706 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
707 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
708 
709 	/* data might have come in before data_ready set up */
710 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
711 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
712 
713 	/* make sure we get destination address info */
714 	switch (svsk->sk_sk->sk_family) {
715 	case AF_INET:
716 		level = SOL_IP;
717 		optname = IP_PKTINFO;
718 		break;
719 	case AF_INET6:
720 		level = SOL_IPV6;
721 		optname = IPV6_RECVPKTINFO;
722 		break;
723 	default:
724 		BUG();
725 	}
726 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
727 					(char *)&one, sizeof(one));
728 	dprintk("svc: kernel_setsockopt returned %d\n", err);
729 }
730 
731 /*
732  * A data_ready event on a listening socket means there's a connection
733  * pending. Do not use state_change as a substitute for it.
734  */
735 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
736 {
737 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
738 
739 	dprintk("svc: socket %p TCP (listen) state change %d\n",
740 		sk, sk->sk_state);
741 
742 	/*
743 	 * This callback may called twice when a new connection
744 	 * is established as a child socket inherits everything
745 	 * from a parent LISTEN socket.
746 	 * 1) data_ready method of the parent socket will be called
747 	 *    when one of child sockets become ESTABLISHED.
748 	 * 2) data_ready method of the child socket may be called
749 	 *    when it receives data before the socket is accepted.
750 	 * In case of 2, we should ignore it silently.
751 	 */
752 	if (sk->sk_state == TCP_LISTEN) {
753 		if (svsk) {
754 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
755 			svc_xprt_enqueue(&svsk->sk_xprt);
756 		} else
757 			printk("svc: socket %p: no user data\n", sk);
758 	}
759 
760 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
761 		wake_up_interruptible_all(sk->sk_sleep);
762 }
763 
764 /*
765  * A state change on a connected socket means it's dying or dead.
766  */
767 static void svc_tcp_state_change(struct sock *sk)
768 {
769 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
770 
771 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
772 		sk, sk->sk_state, sk->sk_user_data);
773 
774 	if (!svsk)
775 		printk("svc: socket %p: no user data\n", sk);
776 	else {
777 		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
778 		svc_xprt_enqueue(&svsk->sk_xprt);
779 	}
780 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
781 		wake_up_interruptible_all(sk->sk_sleep);
782 }
783 
784 static void svc_tcp_data_ready(struct sock *sk, int count)
785 {
786 	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
787 
788 	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
789 		sk, sk->sk_user_data);
790 	if (svsk) {
791 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
792 		svc_xprt_enqueue(&svsk->sk_xprt);
793 	}
794 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
795 		wake_up_interruptible(sk->sk_sleep);
796 }
797 
798 /*
799  * Accept a TCP connection
800  */
801 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
802 {
803 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
804 	struct sockaddr_storage addr;
805 	struct sockaddr	*sin = (struct sockaddr *) &addr;
806 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
807 	struct socket	*sock = svsk->sk_sock;
808 	struct socket	*newsock;
809 	struct svc_sock	*newsvsk;
810 	int		err, slen;
811 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
812 
813 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
814 	if (!sock)
815 		return NULL;
816 
817 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
818 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
819 	if (err < 0) {
820 		if (err == -ENOMEM)
821 			printk(KERN_WARNING "%s: no more sockets!\n",
822 			       serv->sv_name);
823 		else if (err != -EAGAIN && net_ratelimit())
824 			printk(KERN_WARNING "%s: accept failed (err %d)!\n",
825 				   serv->sv_name, -err);
826 		return NULL;
827 	}
828 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
829 
830 	err = kernel_getpeername(newsock, sin, &slen);
831 	if (err < 0) {
832 		if (net_ratelimit())
833 			printk(KERN_WARNING "%s: peername failed (err %d)!\n",
834 				   serv->sv_name, -err);
835 		goto failed;		/* aborted connection or whatever */
836 	}
837 
838 	/* Ideally, we would want to reject connections from unauthorized
839 	 * hosts here, but when we get encryption, the IP of the host won't
840 	 * tell us anything.  For now just warn about unpriv connections.
841 	 */
842 	if (!svc_port_is_privileged(sin)) {
843 		dprintk(KERN_WARNING
844 			"%s: connect from unprivileged port: %s\n",
845 			serv->sv_name,
846 			__svc_print_addr(sin, buf, sizeof(buf)));
847 	}
848 	dprintk("%s: connect from %s\n", serv->sv_name,
849 		__svc_print_addr(sin, buf, sizeof(buf)));
850 
851 	/* make sure that a write doesn't block forever when
852 	 * low on memory
853 	 */
854 	newsock->sk->sk_sndtimeo = HZ*30;
855 
856 	if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
857 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
858 		goto failed;
859 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
860 	err = kernel_getsockname(newsock, sin, &slen);
861 	if (unlikely(err < 0)) {
862 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
863 		slen = offsetof(struct sockaddr, sa_data);
864 	}
865 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
866 
867 	if (serv->sv_stats)
868 		serv->sv_stats->nettcpconn++;
869 
870 	return &newsvsk->sk_xprt;
871 
872 failed:
873 	sock_release(newsock);
874 	return NULL;
875 }
876 
877 /*
878  * Receive data.
879  * If we haven't gotten the record length yet, get the next four bytes.
880  * Otherwise try to gobble up as much as possible up to the complete
881  * record length.
882  */
883 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
884 {
885 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
886 	int len;
887 
888 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
889 		/* sndbuf needs to have room for one request
890 		 * per thread, otherwise we can stall even when the
891 		 * network isn't a bottleneck.
892 		 *
893 		 * We count all threads rather than threads in a
894 		 * particular pool, which provides an upper bound
895 		 * on the number of threads which will access the socket.
896 		 *
897 		 * rcvbuf just needs to be able to hold a few requests.
898 		 * Normally they will be removed from the queue
899 		 * as soon a a complete request arrives.
900 		 */
901 		svc_sock_setbufsize(svsk->sk_sock,
902 				    (serv->sv_nrthreads+3) * serv->sv_max_mesg,
903 				    3 * serv->sv_max_mesg);
904 
905 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
906 
907 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
908 		int		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
909 		struct kvec	iov;
910 
911 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
912 		iov.iov_len  = want;
913 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
914 			goto error;
915 		svsk->sk_tcplen += len;
916 
917 		if (len < want) {
918 			dprintk("svc: short recvfrom while reading record "
919 				"length (%d of %d)\n", len, want);
920 			svc_xprt_received(&svsk->sk_xprt);
921 			goto err_again; /* record header not complete */
922 		}
923 
924 		svsk->sk_reclen = ntohl(svsk->sk_reclen);
925 		if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
926 			/* FIXME: technically, a record can be fragmented,
927 			 *  and non-terminal fragments will not have the top
928 			 *  bit set in the fragment length header.
929 			 *  But apparently no known nfs clients send fragmented
930 			 *  records. */
931 			if (net_ratelimit())
932 				printk(KERN_NOTICE "RPC: multiple fragments "
933 					"per record not supported\n");
934 			goto err_delete;
935 		}
936 
937 		svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
938 		dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
939 		if (svsk->sk_reclen > serv->sv_max_mesg) {
940 			if (net_ratelimit())
941 				printk(KERN_NOTICE "RPC: "
942 					"fragment too large: 0x%08lx\n",
943 					(unsigned long)svsk->sk_reclen);
944 			goto err_delete;
945 		}
946 	}
947 
948 	/* Check whether enough data is available */
949 	len = svc_recv_available(svsk);
950 	if (len < 0)
951 		goto error;
952 
953 	if (len < svsk->sk_reclen) {
954 		dprintk("svc: incomplete TCP record (%d of %d)\n",
955 			len, svsk->sk_reclen);
956 		svc_xprt_received(&svsk->sk_xprt);
957 		goto err_again;	/* record not complete */
958 	}
959 	len = svsk->sk_reclen;
960 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
961 
962 	return len;
963  error:
964 	if (len == -EAGAIN) {
965 		dprintk("RPC: TCP recv_record got EAGAIN\n");
966 		svc_xprt_received(&svsk->sk_xprt);
967 	}
968 	return len;
969  err_delete:
970 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
971  err_again:
972 	return -EAGAIN;
973 }
974 
975 static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
976 			       struct rpc_rqst **reqpp, struct kvec *vec)
977 {
978 	struct rpc_rqst *req = NULL;
979 	u32 *p;
980 	u32 xid;
981 	u32 calldir;
982 	int len;
983 
984 	len = svc_recvfrom(rqstp, vec, 1, 8);
985 	if (len < 0)
986 		goto error;
987 
988 	p = (u32 *)rqstp->rq_arg.head[0].iov_base;
989 	xid = *p++;
990 	calldir = *p;
991 
992 	if (calldir == 0) {
993 		/* REQUEST is the most common case */
994 		vec[0] = rqstp->rq_arg.head[0];
995 	} else {
996 		/* REPLY */
997 		if (svsk->sk_bc_xprt)
998 			req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid);
999 
1000 		if (!req) {
1001 			printk(KERN_NOTICE
1002 				"%s: Got unrecognized reply: "
1003 				"calldir 0x%x sk_bc_xprt %p xid %08x\n",
1004 				__func__, ntohl(calldir),
1005 				svsk->sk_bc_xprt, xid);
1006 			vec[0] = rqstp->rq_arg.head[0];
1007 			goto out;
1008 		}
1009 
1010 		memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1011 		       sizeof(struct xdr_buf));
1012 		/* copy the xid and call direction */
1013 		memcpy(req->rq_private_buf.head[0].iov_base,
1014 		       rqstp->rq_arg.head[0].iov_base, 8);
1015 		vec[0] = req->rq_private_buf.head[0];
1016 	}
1017  out:
1018 	vec[0].iov_base += 8;
1019 	vec[0].iov_len -= 8;
1020 	len = svsk->sk_reclen - 8;
1021  error:
1022 	*reqpp = req;
1023 	return len;
1024 }
1025 
1026 /*
1027  * Receive data from a TCP socket.
1028  */
1029 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1030 {
1031 	struct svc_sock	*svsk =
1032 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1033 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1034 	int		len;
1035 	struct kvec *vec;
1036 	int pnum, vlen;
1037 	struct rpc_rqst *req = NULL;
1038 
1039 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1040 		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1041 		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1042 		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1043 
1044 	len = svc_tcp_recv_record(svsk, rqstp);
1045 	if (len < 0)
1046 		goto error;
1047 
1048 	vec = rqstp->rq_vec;
1049 	vec[0] = rqstp->rq_arg.head[0];
1050 	vlen = PAGE_SIZE;
1051 
1052 	/*
1053 	 * We have enough data for the whole tcp record. Let's try and read the
1054 	 * first 8 bytes to get the xid and the call direction. We can use this
1055 	 * to figure out if this is a call or a reply to a callback. If
1056 	 * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
1057 	 * In that case, don't bother with the calldir and just read the data.
1058 	 * It will be rejected in svc_process.
1059 	 */
1060 	if (len >= 8) {
1061 		len = svc_process_calldir(svsk, rqstp, &req, vec);
1062 		if (len < 0)
1063 			goto err_again;
1064 		vlen -= 8;
1065 	}
1066 
1067 	pnum = 1;
1068 	while (vlen < len) {
1069 		vec[pnum].iov_base = (req) ?
1070 			page_address(req->rq_private_buf.pages[pnum - 1]) :
1071 			page_address(rqstp->rq_pages[pnum]);
1072 		vec[pnum].iov_len = PAGE_SIZE;
1073 		pnum++;
1074 		vlen += PAGE_SIZE;
1075 	}
1076 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1077 
1078 	/* Now receive data */
1079 	len = svc_recvfrom(rqstp, vec, pnum, len);
1080 	if (len < 0)
1081 		goto err_again;
1082 
1083 	/*
1084 	 * Account for the 8 bytes we read earlier
1085 	 */
1086 	len += 8;
1087 
1088 	if (req) {
1089 		xprt_complete_rqst(req->rq_task, len);
1090 		len = 0;
1091 		goto out;
1092 	}
1093 	dprintk("svc: TCP complete record (%d bytes)\n", len);
1094 	rqstp->rq_arg.len = len;
1095 	rqstp->rq_arg.page_base = 0;
1096 	if (len <= rqstp->rq_arg.head[0].iov_len) {
1097 		rqstp->rq_arg.head[0].iov_len = len;
1098 		rqstp->rq_arg.page_len = 0;
1099 	} else {
1100 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1101 	}
1102 
1103 	rqstp->rq_xprt_ctxt   = NULL;
1104 	rqstp->rq_prot	      = IPPROTO_TCP;
1105 
1106 out:
1107 	/* Reset TCP read info */
1108 	svsk->sk_reclen = 0;
1109 	svsk->sk_tcplen = 0;
1110 
1111 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1112 	svc_xprt_received(&svsk->sk_xprt);
1113 	if (serv->sv_stats)
1114 		serv->sv_stats->nettcpcnt++;
1115 
1116 	return len;
1117 
1118 err_again:
1119 	if (len == -EAGAIN) {
1120 		dprintk("RPC: TCP recvfrom got EAGAIN\n");
1121 		svc_xprt_received(&svsk->sk_xprt);
1122 		return len;
1123 	}
1124 error:
1125 	if (len != -EAGAIN) {
1126 		printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1127 		       svsk->sk_xprt.xpt_server->sv_name, -len);
1128 		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1129 	}
1130 	return -EAGAIN;
1131 }
1132 
1133 /*
1134  * Send out data on TCP socket.
1135  */
1136 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1137 {
1138 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1139 	int sent;
1140 	__be32 reclen;
1141 
1142 	/* Set up the first element of the reply kvec.
1143 	 * Any other kvecs that may be in use have been taken
1144 	 * care of by the server implementation itself.
1145 	 */
1146 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1147 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1148 
1149 	if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
1150 		return -ENOTCONN;
1151 
1152 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1153 	if (sent != xbufp->len) {
1154 		printk(KERN_NOTICE
1155 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1156 		       "- shutting down socket\n",
1157 		       rqstp->rq_xprt->xpt_server->sv_name,
1158 		       (sent<0)?"got error":"sent only",
1159 		       sent, xbufp->len);
1160 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1161 		svc_xprt_enqueue(rqstp->rq_xprt);
1162 		sent = -EAGAIN;
1163 	}
1164 	return sent;
1165 }
1166 
1167 /*
1168  * Setup response header. TCP has a 4B record length field.
1169  */
1170 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1171 {
1172 	struct kvec *resv = &rqstp->rq_res.head[0];
1173 
1174 	/* tcp needs a space for the record length... */
1175 	svc_putnl(resv, 0);
1176 }
1177 
1178 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1179 {
1180 	struct svc_sock *svsk =	container_of(xprt, struct svc_sock, sk_xprt);
1181 	struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1182 	int required;
1183 
1184 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1185 		return 1;
1186 	required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1187 	if (sk_stream_wspace(svsk->sk_sk) >= required)
1188 		return 1;
1189 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1190 	return 0;
1191 }
1192 
1193 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1194 				       struct sockaddr *sa, int salen,
1195 				       int flags)
1196 {
1197 	return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1198 }
1199 
1200 static struct svc_xprt_ops svc_tcp_ops = {
1201 	.xpo_create = svc_tcp_create,
1202 	.xpo_recvfrom = svc_tcp_recvfrom,
1203 	.xpo_sendto = svc_tcp_sendto,
1204 	.xpo_release_rqst = svc_release_skb,
1205 	.xpo_detach = svc_tcp_sock_detach,
1206 	.xpo_free = svc_sock_free,
1207 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1208 	.xpo_has_wspace = svc_tcp_has_wspace,
1209 	.xpo_accept = svc_tcp_accept,
1210 };
1211 
1212 static struct svc_xprt_class svc_tcp_class = {
1213 	.xcl_name = "tcp",
1214 	.xcl_owner = THIS_MODULE,
1215 	.xcl_ops = &svc_tcp_ops,
1216 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1217 };
1218 
1219 void svc_init_xprt_sock(void)
1220 {
1221 	svc_reg_xprt_class(&svc_tcp_class);
1222 	svc_reg_xprt_class(&svc_udp_class);
1223 }
1224 
1225 void svc_cleanup_xprt_sock(void)
1226 {
1227 	svc_unreg_xprt_class(&svc_tcp_class);
1228 	svc_unreg_xprt_class(&svc_udp_class);
1229 }
1230 
1231 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1232 {
1233 	struct sock	*sk = svsk->sk_sk;
1234 
1235 	svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1236 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1237 	if (sk->sk_state == TCP_LISTEN) {
1238 		dprintk("setting up TCP socket for listening\n");
1239 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1240 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1241 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1242 	} else {
1243 		dprintk("setting up TCP socket for reading\n");
1244 		sk->sk_state_change = svc_tcp_state_change;
1245 		sk->sk_data_ready = svc_tcp_data_ready;
1246 		sk->sk_write_space = svc_tcp_write_space;
1247 
1248 		svsk->sk_reclen = 0;
1249 		svsk->sk_tcplen = 0;
1250 
1251 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1252 
1253 		/* initialise setting must have enough space to
1254 		 * receive and respond to one request.
1255 		 * svc_tcp_recvfrom will re-adjust if necessary
1256 		 */
1257 		svc_sock_setbufsize(svsk->sk_sock,
1258 				    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1259 				    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1260 
1261 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1262 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1263 		if (sk->sk_state != TCP_ESTABLISHED)
1264 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1265 	}
1266 }
1267 
1268 void svc_sock_update_bufs(struct svc_serv *serv)
1269 {
1270 	/*
1271 	 * The number of server threads has changed. Update
1272 	 * rcvbuf and sndbuf accordingly on all sockets
1273 	 */
1274 	struct list_head *le;
1275 
1276 	spin_lock_bh(&serv->sv_lock);
1277 	list_for_each(le, &serv->sv_permsocks) {
1278 		struct svc_sock *svsk =
1279 			list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1280 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1281 	}
1282 	list_for_each(le, &serv->sv_tempsocks) {
1283 		struct svc_sock *svsk =
1284 			list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1285 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1286 	}
1287 	spin_unlock_bh(&serv->sv_lock);
1288 }
1289 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1290 
1291 /*
1292  * Initialize socket for RPC use and create svc_sock struct
1293  * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1294  */
1295 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1296 						struct socket *sock,
1297 						int *errp, int flags)
1298 {
1299 	struct svc_sock	*svsk;
1300 	struct sock	*inet;
1301 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1302 
1303 	dprintk("svc: svc_setup_socket %p\n", sock);
1304 	if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1305 		*errp = -ENOMEM;
1306 		return NULL;
1307 	}
1308 
1309 	inet = sock->sk;
1310 
1311 	/* Register socket with portmapper */
1312 	if (*errp >= 0 && pmap_register)
1313 		*errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1314 				     ntohs(inet_sk(inet)->sport));
1315 
1316 	if (*errp < 0) {
1317 		kfree(svsk);
1318 		return NULL;
1319 	}
1320 
1321 	inet->sk_user_data = svsk;
1322 	svsk->sk_sock = sock;
1323 	svsk->sk_sk = inet;
1324 	svsk->sk_ostate = inet->sk_state_change;
1325 	svsk->sk_odata = inet->sk_data_ready;
1326 	svsk->sk_owspace = inet->sk_write_space;
1327 
1328 	/* Initialize the socket */
1329 	if (sock->type == SOCK_DGRAM)
1330 		svc_udp_init(svsk, serv);
1331 	else
1332 		svc_tcp_init(svsk, serv);
1333 
1334 	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1335 				svsk, svsk->sk_sk);
1336 
1337 	return svsk;
1338 }
1339 
1340 /**
1341  * svc_addsock - add a listener socket to an RPC service
1342  * @serv: pointer to RPC service to which to add a new listener
1343  * @fd: file descriptor of the new listener
1344  * @name_return: pointer to buffer to fill in with name of listener
1345  * @len: size of the buffer
1346  *
1347  * Fills in socket name and returns positive length of name if successful.
1348  * Name is terminated with '\n'.  On error, returns a negative errno
1349  * value.
1350  */
1351 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1352 		const size_t len)
1353 {
1354 	int err = 0;
1355 	struct socket *so = sockfd_lookup(fd, &err);
1356 	struct svc_sock *svsk = NULL;
1357 
1358 	if (!so)
1359 		return err;
1360 	if (so->sk->sk_family != AF_INET)
1361 		err =  -EAFNOSUPPORT;
1362 	else if (so->sk->sk_protocol != IPPROTO_TCP &&
1363 	    so->sk->sk_protocol != IPPROTO_UDP)
1364 		err =  -EPROTONOSUPPORT;
1365 	else if (so->state > SS_UNCONNECTED)
1366 		err = -EISCONN;
1367 	else {
1368 		if (!try_module_get(THIS_MODULE))
1369 			err = -ENOENT;
1370 		else
1371 			svsk = svc_setup_socket(serv, so, &err,
1372 						SVC_SOCK_DEFAULTS);
1373 		if (svsk) {
1374 			struct sockaddr_storage addr;
1375 			struct sockaddr *sin = (struct sockaddr *)&addr;
1376 			int salen;
1377 			if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1378 				svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1379 			clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1380 			spin_lock_bh(&serv->sv_lock);
1381 			list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1382 			spin_unlock_bh(&serv->sv_lock);
1383 			svc_xprt_received(&svsk->sk_xprt);
1384 			err = 0;
1385 		} else
1386 			module_put(THIS_MODULE);
1387 	}
1388 	if (err) {
1389 		sockfd_put(so);
1390 		return err;
1391 	}
1392 	return svc_one_sock_name(svsk, name_return, len);
1393 }
1394 EXPORT_SYMBOL_GPL(svc_addsock);
1395 
1396 /*
1397  * Create socket for RPC service.
1398  */
1399 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1400 					  int protocol,
1401 					  struct sockaddr *sin, int len,
1402 					  int flags)
1403 {
1404 	struct svc_sock	*svsk;
1405 	struct socket	*sock;
1406 	int		error;
1407 	int		type;
1408 	struct sockaddr_storage addr;
1409 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1410 	int		newlen;
1411 	int		family;
1412 	int		val;
1413 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1414 
1415 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1416 			serv->sv_program->pg_name, protocol,
1417 			__svc_print_addr(sin, buf, sizeof(buf)));
1418 
1419 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1420 		printk(KERN_WARNING "svc: only UDP and TCP "
1421 				"sockets supported\n");
1422 		return ERR_PTR(-EINVAL);
1423 	}
1424 
1425 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1426 	switch (sin->sa_family) {
1427 	case AF_INET6:
1428 		family = PF_INET6;
1429 		break;
1430 	case AF_INET:
1431 		family = PF_INET;
1432 		break;
1433 	default:
1434 		return ERR_PTR(-EINVAL);
1435 	}
1436 
1437 	error = sock_create_kern(family, type, protocol, &sock);
1438 	if (error < 0)
1439 		return ERR_PTR(error);
1440 
1441 	svc_reclassify_socket(sock);
1442 
1443 	/*
1444 	 * If this is an PF_INET6 listener, we want to avoid
1445 	 * getting requests from IPv4 remotes.  Those should
1446 	 * be shunted to a PF_INET listener via rpcbind.
1447 	 */
1448 	val = 1;
1449 	if (family == PF_INET6)
1450 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1451 					(char *)&val, sizeof(val));
1452 
1453 	if (type == SOCK_STREAM)
1454 		sock->sk->sk_reuse = 1;		/* allow address reuse */
1455 	error = kernel_bind(sock, sin, len);
1456 	if (error < 0)
1457 		goto bummer;
1458 
1459 	newlen = len;
1460 	error = kernel_getsockname(sock, newsin, &newlen);
1461 	if (error < 0)
1462 		goto bummer;
1463 
1464 	if (protocol == IPPROTO_TCP) {
1465 		if ((error = kernel_listen(sock, 64)) < 0)
1466 			goto bummer;
1467 	}
1468 
1469 	if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1470 		svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1471 		return (struct svc_xprt *)svsk;
1472 	}
1473 
1474 bummer:
1475 	dprintk("svc: svc_create_socket error = %d\n", -error);
1476 	sock_release(sock);
1477 	return ERR_PTR(error);
1478 }
1479 
1480 /*
1481  * Detach the svc_sock from the socket so that no
1482  * more callbacks occur.
1483  */
1484 static void svc_sock_detach(struct svc_xprt *xprt)
1485 {
1486 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1487 	struct sock *sk = svsk->sk_sk;
1488 
1489 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1490 
1491 	/* put back the old socket callbacks */
1492 	sk->sk_state_change = svsk->sk_ostate;
1493 	sk->sk_data_ready = svsk->sk_odata;
1494 	sk->sk_write_space = svsk->sk_owspace;
1495 
1496 	if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1497 		wake_up_interruptible(sk->sk_sleep);
1498 }
1499 
1500 /*
1501  * Disconnect the socket, and reset the callbacks
1502  */
1503 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1504 {
1505 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1506 
1507 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1508 
1509 	svc_sock_detach(xprt);
1510 
1511 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1512 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1513 }
1514 
1515 /*
1516  * Free the svc_sock's socket resources and the svc_sock itself.
1517  */
1518 static void svc_sock_free(struct svc_xprt *xprt)
1519 {
1520 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1521 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1522 
1523 	if (svsk->sk_sock->file)
1524 		sockfd_put(svsk->sk_sock);
1525 	else
1526 		sock_release(svsk->sk_sock);
1527 	kfree(svsk);
1528 }
1529 
1530 /*
1531  * Create a svc_xprt.
1532  *
1533  * For internal use only (e.g. nfsv4.1 backchannel).
1534  * Callers should typically use the xpo_create() method.
1535  */
1536 struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot)
1537 {
1538 	struct svc_sock *svsk;
1539 	struct svc_xprt *xprt = NULL;
1540 
1541 	dprintk("svc: %s\n", __func__);
1542 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1543 	if (!svsk)
1544 		goto out;
1545 
1546 	xprt = &svsk->sk_xprt;
1547 	if (prot == IPPROTO_TCP)
1548 		svc_xprt_init(&svc_tcp_class, xprt, serv);
1549 	else if (prot == IPPROTO_UDP)
1550 		svc_xprt_init(&svc_udp_class, xprt, serv);
1551 	else
1552 		BUG();
1553 out:
1554 	dprintk("svc: %s return %p\n", __func__, xprt);
1555 	return xprt;
1556 }
1557 EXPORT_SYMBOL_GPL(svc_sock_create);
1558 
1559 /*
1560  * Destroy a svc_sock.
1561  */
1562 void svc_sock_destroy(struct svc_xprt *xprt)
1563 {
1564 	if (xprt)
1565 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1566 }
1567 EXPORT_SYMBOL_GPL(svc_sock_destroy);
1568