xref: /openbmc/linux/net/sunrpc/svcsock.c (revision f7c35abe)
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/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/ip.h>
41 #include <net/ipv6.h>
42 #include <net/udp.h>
43 #include <net/tcp.h>
44 #include <net/tcp_states.h>
45 #include <linux/uaccess.h>
46 #include <asm/ioctls.h>
47 #include <trace/events/skb.h>
48 
49 #include <linux/sunrpc/types.h>
50 #include <linux/sunrpc/clnt.h>
51 #include <linux/sunrpc/xdr.h>
52 #include <linux/sunrpc/msg_prot.h>
53 #include <linux/sunrpc/svcsock.h>
54 #include <linux/sunrpc/stats.h>
55 #include <linux/sunrpc/xprt.h>
56 
57 #include "sunrpc.h"
58 
59 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
60 
61 
62 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
63 					 int flags);
64 static int		svc_udp_recvfrom(struct svc_rqst *);
65 static int		svc_udp_sendto(struct svc_rqst *);
66 static void		svc_sock_detach(struct svc_xprt *);
67 static void		svc_tcp_sock_detach(struct svc_xprt *);
68 static void		svc_sock_free(struct svc_xprt *);
69 
70 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 					  struct net *, struct sockaddr *,
72 					  int, int);
73 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
74 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 					     struct net *, struct sockaddr *,
76 					     int, int);
77 static void svc_bc_sock_free(struct svc_xprt *xprt);
78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
79 
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 static struct lock_class_key svc_key[2];
82 static struct lock_class_key svc_slock_key[2];
83 
84 static void svc_reclassify_socket(struct socket *sock)
85 {
86 	struct sock *sk = sock->sk;
87 
88 	if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
89 		return;
90 
91 	switch (sk->sk_family) {
92 	case AF_INET:
93 		sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
94 					      &svc_slock_key[0],
95 					      "sk_xprt.xpt_lock-AF_INET-NFSD",
96 					      &svc_key[0]);
97 		break;
98 
99 	case AF_INET6:
100 		sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
101 					      &svc_slock_key[1],
102 					      "sk_xprt.xpt_lock-AF_INET6-NFSD",
103 					      &svc_key[1]);
104 		break;
105 
106 	default:
107 		BUG();
108 	}
109 }
110 #else
111 static void svc_reclassify_socket(struct socket *sock)
112 {
113 }
114 #endif
115 
116 /*
117  * Release an skbuff after use
118  */
119 static void svc_release_skb(struct svc_rqst *rqstp)
120 {
121 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
122 
123 	if (skb) {
124 		struct svc_sock *svsk =
125 			container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
126 		rqstp->rq_xprt_ctxt = NULL;
127 
128 		dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
129 		skb_free_datagram_locked(svsk->sk_sk, skb);
130 	}
131 }
132 
133 static void svc_release_udp_skb(struct svc_rqst *rqstp)
134 {
135 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
136 
137 	if (skb) {
138 		rqstp->rq_xprt_ctxt = NULL;
139 
140 		dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
141 		consume_skb(skb);
142 	}
143 }
144 
145 union svc_pktinfo_u {
146 	struct in_pktinfo pkti;
147 	struct in6_pktinfo pkti6;
148 };
149 #define SVC_PKTINFO_SPACE \
150 	CMSG_SPACE(sizeof(union svc_pktinfo_u))
151 
152 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
153 {
154 	struct svc_sock *svsk =
155 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
156 	switch (svsk->sk_sk->sk_family) {
157 	case AF_INET: {
158 			struct in_pktinfo *pki = CMSG_DATA(cmh);
159 
160 			cmh->cmsg_level = SOL_IP;
161 			cmh->cmsg_type = IP_PKTINFO;
162 			pki->ipi_ifindex = 0;
163 			pki->ipi_spec_dst.s_addr =
164 				 svc_daddr_in(rqstp)->sin_addr.s_addr;
165 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
166 		}
167 		break;
168 
169 	case AF_INET6: {
170 			struct in6_pktinfo *pki = CMSG_DATA(cmh);
171 			struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
172 
173 			cmh->cmsg_level = SOL_IPV6;
174 			cmh->cmsg_type = IPV6_PKTINFO;
175 			pki->ipi6_ifindex = daddr->sin6_scope_id;
176 			pki->ipi6_addr = daddr->sin6_addr;
177 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
178 		}
179 		break;
180 	}
181 }
182 
183 /*
184  * send routine intended to be shared by the fore- and back-channel
185  */
186 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
187 		    struct page *headpage, unsigned long headoffset,
188 		    struct page *tailpage, unsigned long tailoffset)
189 {
190 	int		result;
191 	int		size;
192 	struct page	**ppage = xdr->pages;
193 	size_t		base = xdr->page_base;
194 	unsigned int	pglen = xdr->page_len;
195 	unsigned int	flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
196 	int		slen;
197 	int		len = 0;
198 
199 	slen = xdr->len;
200 
201 	/* send head */
202 	if (slen == xdr->head[0].iov_len)
203 		flags = 0;
204 	len = kernel_sendpage(sock, headpage, headoffset,
205 				  xdr->head[0].iov_len, flags);
206 	if (len != xdr->head[0].iov_len)
207 		goto out;
208 	slen -= xdr->head[0].iov_len;
209 	if (slen == 0)
210 		goto out;
211 
212 	/* send page data */
213 	size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
214 	while (pglen > 0) {
215 		if (slen == size)
216 			flags = 0;
217 		result = kernel_sendpage(sock, *ppage, base, size, flags);
218 		if (result > 0)
219 			len += result;
220 		if (result != size)
221 			goto out;
222 		slen -= size;
223 		pglen -= size;
224 		size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
225 		base = 0;
226 		ppage++;
227 	}
228 
229 	/* send tail */
230 	if (xdr->tail[0].iov_len) {
231 		result = kernel_sendpage(sock, tailpage, tailoffset,
232 				   xdr->tail[0].iov_len, 0);
233 		if (result > 0)
234 			len += result;
235 	}
236 
237 out:
238 	return len;
239 }
240 
241 
242 /*
243  * Generic sendto routine
244  */
245 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
246 {
247 	struct svc_sock	*svsk =
248 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
249 	struct socket	*sock = svsk->sk_sock;
250 	union {
251 		struct cmsghdr	hdr;
252 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
253 	} buffer;
254 	struct cmsghdr *cmh = &buffer.hdr;
255 	int		len = 0;
256 	unsigned long tailoff;
257 	unsigned long headoff;
258 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
259 
260 	if (rqstp->rq_prot == IPPROTO_UDP) {
261 		struct msghdr msg = {
262 			.msg_name	= &rqstp->rq_addr,
263 			.msg_namelen	= rqstp->rq_addrlen,
264 			.msg_control	= cmh,
265 			.msg_controllen	= sizeof(buffer),
266 			.msg_flags	= MSG_MORE,
267 		};
268 
269 		svc_set_cmsg_data(rqstp, cmh);
270 
271 		if (sock_sendmsg(sock, &msg) < 0)
272 			goto out;
273 	}
274 
275 	tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
276 	headoff = 0;
277 	len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
278 			       rqstp->rq_respages[0], tailoff);
279 
280 out:
281 	dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
282 		svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
283 		xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
284 
285 	return len;
286 }
287 
288 /*
289  * Report socket names for nfsdfs
290  */
291 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
292 {
293 	const struct sock *sk = svsk->sk_sk;
294 	const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
295 							"udp" : "tcp";
296 	int len;
297 
298 	switch (sk->sk_family) {
299 	case PF_INET:
300 		len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
301 				proto_name,
302 				&inet_sk(sk)->inet_rcv_saddr,
303 				inet_sk(sk)->inet_num);
304 		break;
305 #if IS_ENABLED(CONFIG_IPV6)
306 	case PF_INET6:
307 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
308 				proto_name,
309 				&sk->sk_v6_rcv_saddr,
310 				inet_sk(sk)->inet_num);
311 		break;
312 #endif
313 	default:
314 		len = snprintf(buf, remaining, "*unknown-%d*\n",
315 				sk->sk_family);
316 	}
317 
318 	if (len >= remaining) {
319 		*buf = '\0';
320 		return -ENAMETOOLONG;
321 	}
322 	return len;
323 }
324 
325 /*
326  * Generic recvfrom routine.
327  */
328 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
329 			int buflen)
330 {
331 	struct svc_sock *svsk =
332 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 	struct msghdr msg = {
334 		.msg_flags	= MSG_DONTWAIT,
335 	};
336 	int len;
337 
338 	rqstp->rq_xprt_hlen = 0;
339 
340 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
341 	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
342 				msg.msg_flags);
343 	/* If we read a full record, then assume there may be more
344 	 * data to read (stream based sockets only!)
345 	 */
346 	if (len == buflen)
347 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
348 
349 	dprintk("svc: socket %p recvfrom(%p, %zu) = %d\n",
350 		svsk, iov[0].iov_base, iov[0].iov_len, len);
351 	return len;
352 }
353 
354 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
355 				struct kvec *iov, int nr,
356 				int buflen, unsigned int base)
357 {
358 	size_t save_iovlen;
359 	void *save_iovbase;
360 	unsigned int i;
361 	int ret;
362 
363 	if (base == 0)
364 		return svc_recvfrom(rqstp, iov, nr, buflen);
365 
366 	for (i = 0; i < nr; i++) {
367 		if (iov[i].iov_len > base)
368 			break;
369 		base -= iov[i].iov_len;
370 	}
371 	save_iovlen = iov[i].iov_len;
372 	save_iovbase = iov[i].iov_base;
373 	iov[i].iov_len -= base;
374 	iov[i].iov_base += base;
375 	ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
376 	iov[i].iov_len = save_iovlen;
377 	iov[i].iov_base = save_iovbase;
378 	return ret;
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_write_space(sock->sk);
404 	release_sock(sock->sk);
405 #endif
406 }
407 
408 static int svc_sock_secure_port(struct svc_rqst *rqstp)
409 {
410 	return svc_port_is_privileged(svc_addr(rqstp));
411 }
412 
413 /*
414  * INET callback when data has been received on the socket.
415  */
416 static void svc_data_ready(struct sock *sk)
417 {
418 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
419 
420 	if (svsk) {
421 		dprintk("svc: socket %p(inet %p), busy=%d\n",
422 			svsk, sk,
423 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
424 		svsk->sk_odata(sk);
425 		if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
426 			svc_xprt_enqueue(&svsk->sk_xprt);
427 	}
428 }
429 
430 /*
431  * INET callback when space is newly available on the socket.
432  */
433 static void svc_write_space(struct sock *sk)
434 {
435 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
436 
437 	if (svsk) {
438 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
439 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
440 		svsk->sk_owspace(sk);
441 		svc_xprt_enqueue(&svsk->sk_xprt);
442 	}
443 }
444 
445 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
446 {
447 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
448 
449 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
450 		return 1;
451 	return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
452 }
453 
454 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
455 {
456 	struct svc_sock *svsk;
457 	struct socket *sock;
458 	struct linger no_linger = {
459 		.l_onoff = 1,
460 		.l_linger = 0,
461 	};
462 
463 	svsk = container_of(xprt, struct svc_sock, sk_xprt);
464 	sock = svsk->sk_sock;
465 	kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
466 			  (char *)&no_linger, sizeof(no_linger));
467 }
468 
469 /*
470  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
471  */
472 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
473 				     struct cmsghdr *cmh)
474 {
475 	struct in_pktinfo *pki = CMSG_DATA(cmh);
476 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
477 
478 	if (cmh->cmsg_type != IP_PKTINFO)
479 		return 0;
480 
481 	daddr->sin_family = AF_INET;
482 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
483 	return 1;
484 }
485 
486 /*
487  * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
488  */
489 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
490 				     struct cmsghdr *cmh)
491 {
492 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
493 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
494 
495 	if (cmh->cmsg_type != IPV6_PKTINFO)
496 		return 0;
497 
498 	daddr->sin6_family = AF_INET6;
499 	daddr->sin6_addr = pki->ipi6_addr;
500 	daddr->sin6_scope_id = pki->ipi6_ifindex;
501 	return 1;
502 }
503 
504 /*
505  * Copy the UDP datagram's destination address to the rqstp structure.
506  * The 'destination' address in this case is the address to which the
507  * peer sent the datagram, i.e. our local address. For multihomed
508  * hosts, this can change from msg to msg. Note that only the IP
509  * address changes, the port number should remain the same.
510  */
511 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
512 				    struct cmsghdr *cmh)
513 {
514 	switch (cmh->cmsg_level) {
515 	case SOL_IP:
516 		return svc_udp_get_dest_address4(rqstp, cmh);
517 	case SOL_IPV6:
518 		return svc_udp_get_dest_address6(rqstp, cmh);
519 	}
520 
521 	return 0;
522 }
523 
524 /*
525  * Receive a datagram from a UDP socket.
526  */
527 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
528 {
529 	struct svc_sock	*svsk =
530 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
531 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
532 	struct sk_buff	*skb;
533 	union {
534 		struct cmsghdr	hdr;
535 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
536 	} buffer;
537 	struct cmsghdr *cmh = &buffer.hdr;
538 	struct msghdr msg = {
539 		.msg_name = svc_addr(rqstp),
540 		.msg_control = cmh,
541 		.msg_controllen = sizeof(buffer),
542 		.msg_flags = MSG_DONTWAIT,
543 	};
544 	size_t len;
545 	int err;
546 
547 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
548 	    /* udp sockets need large rcvbuf as all pending
549 	     * requests are still in that buffer.  sndbuf must
550 	     * also be large enough that there is enough space
551 	     * for one reply per thread.  We count all threads
552 	     * rather than threads in a particular pool, which
553 	     * provides an upper bound on the number of threads
554 	     * which will access the socket.
555 	     */
556 	    svc_sock_setbufsize(svsk->sk_sock,
557 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
558 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
559 
560 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
561 	skb = NULL;
562 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
563 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
564 	if (err >= 0)
565 		skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
566 
567 	if (skb == NULL) {
568 		if (err != -EAGAIN) {
569 			/* possibly an icmp error */
570 			dprintk("svc: recvfrom returned error %d\n", -err);
571 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
572 		}
573 		return 0;
574 	}
575 	len = svc_addr_len(svc_addr(rqstp));
576 	rqstp->rq_addrlen = len;
577 	if (skb->tstamp == 0) {
578 		skb->tstamp = ktime_get_real();
579 		/* Don't enable netstamp, sunrpc doesn't
580 		   need that much accuracy */
581 	}
582 	svsk->sk_sk->sk_stamp = skb->tstamp;
583 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
584 
585 	len  = skb->len;
586 	rqstp->rq_arg.len = len;
587 
588 	rqstp->rq_prot = IPPROTO_UDP;
589 
590 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
591 		net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
592 				     cmh->cmsg_level, cmh->cmsg_type);
593 		goto out_free;
594 	}
595 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
596 
597 	if (skb_is_nonlinear(skb)) {
598 		/* we have to copy */
599 		local_bh_disable();
600 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
601 			local_bh_enable();
602 			/* checksum error */
603 			goto out_free;
604 		}
605 		local_bh_enable();
606 		consume_skb(skb);
607 	} else {
608 		/* we can use it in-place */
609 		rqstp->rq_arg.head[0].iov_base = skb->data;
610 		rqstp->rq_arg.head[0].iov_len = len;
611 		if (skb_checksum_complete(skb))
612 			goto out_free;
613 		rqstp->rq_xprt_ctxt = skb;
614 	}
615 
616 	rqstp->rq_arg.page_base = 0;
617 	if (len <= rqstp->rq_arg.head[0].iov_len) {
618 		rqstp->rq_arg.head[0].iov_len = len;
619 		rqstp->rq_arg.page_len = 0;
620 		rqstp->rq_respages = rqstp->rq_pages+1;
621 	} else {
622 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
623 		rqstp->rq_respages = rqstp->rq_pages + 1 +
624 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
625 	}
626 	rqstp->rq_next_page = rqstp->rq_respages+1;
627 
628 	if (serv->sv_stats)
629 		serv->sv_stats->netudpcnt++;
630 
631 	return len;
632 out_free:
633 	kfree_skb(skb);
634 	return 0;
635 }
636 
637 static int
638 svc_udp_sendto(struct svc_rqst *rqstp)
639 {
640 	int		error;
641 
642 	error = svc_sendto(rqstp, &rqstp->rq_res);
643 	if (error == -ECONNREFUSED)
644 		/* ICMP error on earlier request. */
645 		error = svc_sendto(rqstp, &rqstp->rq_res);
646 
647 	return error;
648 }
649 
650 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
651 {
652 }
653 
654 static int svc_udp_has_wspace(struct svc_xprt *xprt)
655 {
656 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
657 	struct svc_serv	*serv = xprt->xpt_server;
658 	unsigned long required;
659 
660 	/*
661 	 * Set the SOCK_NOSPACE flag before checking the available
662 	 * sock space.
663 	 */
664 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
665 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
666 	if (required*2 > sock_wspace(svsk->sk_sk))
667 		return 0;
668 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
669 	return 1;
670 }
671 
672 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
673 {
674 	BUG();
675 	return NULL;
676 }
677 
678 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
679 {
680 }
681 
682 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
683 				       struct net *net,
684 				       struct sockaddr *sa, int salen,
685 				       int flags)
686 {
687 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
688 }
689 
690 static struct svc_xprt_ops svc_udp_ops = {
691 	.xpo_create = svc_udp_create,
692 	.xpo_recvfrom = svc_udp_recvfrom,
693 	.xpo_sendto = svc_udp_sendto,
694 	.xpo_release_rqst = svc_release_udp_skb,
695 	.xpo_detach = svc_sock_detach,
696 	.xpo_free = svc_sock_free,
697 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
698 	.xpo_has_wspace = svc_udp_has_wspace,
699 	.xpo_accept = svc_udp_accept,
700 	.xpo_secure_port = svc_sock_secure_port,
701 	.xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
702 };
703 
704 static struct svc_xprt_class svc_udp_class = {
705 	.xcl_name = "udp",
706 	.xcl_owner = THIS_MODULE,
707 	.xcl_ops = &svc_udp_ops,
708 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
709 	.xcl_ident = XPRT_TRANSPORT_UDP,
710 };
711 
712 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
713 {
714 	int err, level, optname, one = 1;
715 
716 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
717 		      &svsk->sk_xprt, serv);
718 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
719 	svsk->sk_sk->sk_data_ready = svc_data_ready;
720 	svsk->sk_sk->sk_write_space = svc_write_space;
721 
722 	/* initialise setting must have enough space to
723 	 * receive and respond to one request.
724 	 * svc_udp_recvfrom will re-adjust if necessary
725 	 */
726 	svc_sock_setbufsize(svsk->sk_sock,
727 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
728 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
729 
730 	/* data might have come in before data_ready set up */
731 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
732 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
733 
734 	/* make sure we get destination address info */
735 	switch (svsk->sk_sk->sk_family) {
736 	case AF_INET:
737 		level = SOL_IP;
738 		optname = IP_PKTINFO;
739 		break;
740 	case AF_INET6:
741 		level = SOL_IPV6;
742 		optname = IPV6_RECVPKTINFO;
743 		break;
744 	default:
745 		BUG();
746 	}
747 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
748 					(char *)&one, sizeof(one));
749 	dprintk("svc: kernel_setsockopt returned %d\n", err);
750 }
751 
752 /*
753  * A data_ready event on a listening socket means there's a connection
754  * pending. Do not use state_change as a substitute for it.
755  */
756 static void svc_tcp_listen_data_ready(struct sock *sk)
757 {
758 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
759 
760 	dprintk("svc: socket %p TCP (listen) state change %d\n",
761 		sk, sk->sk_state);
762 
763 	if (svsk)
764 		svsk->sk_odata(sk);
765 	/*
766 	 * This callback may called twice when a new connection
767 	 * is established as a child socket inherits everything
768 	 * from a parent LISTEN socket.
769 	 * 1) data_ready method of the parent socket will be called
770 	 *    when one of child sockets become ESTABLISHED.
771 	 * 2) data_ready method of the child socket may be called
772 	 *    when it receives data before the socket is accepted.
773 	 * In case of 2, we should ignore it silently.
774 	 */
775 	if (sk->sk_state == TCP_LISTEN) {
776 		if (svsk) {
777 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
778 			svc_xprt_enqueue(&svsk->sk_xprt);
779 		} else
780 			printk("svc: socket %p: no user data\n", sk);
781 	}
782 }
783 
784 /*
785  * A state change on a connected socket means it's dying or dead.
786  */
787 static void svc_tcp_state_change(struct sock *sk)
788 {
789 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
790 
791 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
792 		sk, sk->sk_state, sk->sk_user_data);
793 
794 	if (!svsk)
795 		printk("svc: socket %p: no user data\n", sk);
796 	else {
797 		svsk->sk_ostate(sk);
798 		if (sk->sk_state != TCP_ESTABLISHED) {
799 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
800 			svc_xprt_enqueue(&svsk->sk_xprt);
801 		}
802 	}
803 }
804 
805 /*
806  * Accept a TCP connection
807  */
808 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
809 {
810 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
811 	struct sockaddr_storage addr;
812 	struct sockaddr	*sin = (struct sockaddr *) &addr;
813 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
814 	struct socket	*sock = svsk->sk_sock;
815 	struct socket	*newsock;
816 	struct svc_sock	*newsvsk;
817 	int		err, slen;
818 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
819 
820 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
821 	if (!sock)
822 		return NULL;
823 
824 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
825 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
826 	if (err < 0) {
827 		if (err == -ENOMEM)
828 			printk(KERN_WARNING "%s: no more sockets!\n",
829 			       serv->sv_name);
830 		else if (err != -EAGAIN)
831 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
832 					     serv->sv_name, -err);
833 		return NULL;
834 	}
835 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
836 
837 	err = kernel_getpeername(newsock, sin, &slen);
838 	if (err < 0) {
839 		net_warn_ratelimited("%s: peername failed (err %d)!\n",
840 				     serv->sv_name, -err);
841 		goto failed;		/* aborted connection or whatever */
842 	}
843 
844 	/* Ideally, we would want to reject connections from unauthorized
845 	 * hosts here, but when we get encryption, the IP of the host won't
846 	 * tell us anything.  For now just warn about unpriv connections.
847 	 */
848 	if (!svc_port_is_privileged(sin)) {
849 		dprintk("%s: connect from unprivileged port: %s\n",
850 			serv->sv_name,
851 			__svc_print_addr(sin, buf, sizeof(buf)));
852 	}
853 	dprintk("%s: connect from %s\n", serv->sv_name,
854 		__svc_print_addr(sin, buf, sizeof(buf)));
855 
856 	/* Reset the inherited callbacks before calling svc_setup_socket */
857 	newsock->sk->sk_state_change = svsk->sk_ostate;
858 	newsock->sk->sk_data_ready = svsk->sk_odata;
859 	newsock->sk->sk_write_space = svsk->sk_owspace;
860 
861 	/* make sure that a write doesn't block forever when
862 	 * low on memory
863 	 */
864 	newsock->sk->sk_sndtimeo = HZ*30;
865 
866 	newsvsk = svc_setup_socket(serv, newsock,
867 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
868 	if (IS_ERR(newsvsk))
869 		goto failed;
870 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
871 	err = kernel_getsockname(newsock, sin, &slen);
872 	if (unlikely(err < 0)) {
873 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
874 		slen = offsetof(struct sockaddr, sa_data);
875 	}
876 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
877 
878 	if (sock_is_loopback(newsock->sk))
879 		set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
880 	else
881 		clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
882 	if (serv->sv_stats)
883 		serv->sv_stats->nettcpconn++;
884 
885 	return &newsvsk->sk_xprt;
886 
887 failed:
888 	sock_release(newsock);
889 	return NULL;
890 }
891 
892 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
893 {
894 	unsigned int i, len, npages;
895 
896 	if (svsk->sk_datalen == 0)
897 		return 0;
898 	len = svsk->sk_datalen;
899 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
900 	for (i = 0; i < npages; i++) {
901 		if (rqstp->rq_pages[i] != NULL)
902 			put_page(rqstp->rq_pages[i]);
903 		BUG_ON(svsk->sk_pages[i] == NULL);
904 		rqstp->rq_pages[i] = svsk->sk_pages[i];
905 		svsk->sk_pages[i] = NULL;
906 	}
907 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
908 	return len;
909 }
910 
911 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
912 {
913 	unsigned int i, len, npages;
914 
915 	if (svsk->sk_datalen == 0)
916 		return;
917 	len = svsk->sk_datalen;
918 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
919 	for (i = 0; i < npages; i++) {
920 		svsk->sk_pages[i] = rqstp->rq_pages[i];
921 		rqstp->rq_pages[i] = NULL;
922 	}
923 }
924 
925 static void svc_tcp_clear_pages(struct svc_sock *svsk)
926 {
927 	unsigned int i, len, npages;
928 
929 	if (svsk->sk_datalen == 0)
930 		goto out;
931 	len = svsk->sk_datalen;
932 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
933 	for (i = 0; i < npages; i++) {
934 		if (svsk->sk_pages[i] == NULL) {
935 			WARN_ON_ONCE(1);
936 			continue;
937 		}
938 		put_page(svsk->sk_pages[i]);
939 		svsk->sk_pages[i] = NULL;
940 	}
941 out:
942 	svsk->sk_tcplen = 0;
943 	svsk->sk_datalen = 0;
944 }
945 
946 /*
947  * Receive fragment record header.
948  * If we haven't gotten the record length yet, get the next four bytes.
949  */
950 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
951 {
952 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
953 	unsigned int want;
954 	int len;
955 
956 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
957 		struct kvec	iov;
958 
959 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
960 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
961 		iov.iov_len  = want;
962 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
963 			goto error;
964 		svsk->sk_tcplen += len;
965 
966 		if (len < want) {
967 			dprintk("svc: short recvfrom while reading record "
968 				"length (%d of %d)\n", len, want);
969 			return -EAGAIN;
970 		}
971 
972 		dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
973 		if (svc_sock_reclen(svsk) + svsk->sk_datalen >
974 							serv->sv_max_mesg) {
975 			net_notice_ratelimited("RPC: fragment too large: %d\n",
976 					svc_sock_reclen(svsk));
977 			goto err_delete;
978 		}
979 	}
980 
981 	return svc_sock_reclen(svsk);
982 error:
983 	dprintk("RPC: TCP recv_record got %d\n", len);
984 	return len;
985 err_delete:
986 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
987 	return -EAGAIN;
988 }
989 
990 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
991 {
992 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
993 	struct rpc_rqst *req = NULL;
994 	struct kvec *src, *dst;
995 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
996 	__be32 xid;
997 	__be32 calldir;
998 
999 	xid = *p++;
1000 	calldir = *p;
1001 
1002 	if (!bc_xprt)
1003 		return -EAGAIN;
1004 	spin_lock_bh(&bc_xprt->transport_lock);
1005 	req = xprt_lookup_rqst(bc_xprt, xid);
1006 	if (!req)
1007 		goto unlock_notfound;
1008 
1009 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1010 	/*
1011 	 * XXX!: cheating for now!  Only copying HEAD.
1012 	 * But we know this is good enough for now (in fact, for any
1013 	 * callback reply in the forseeable future).
1014 	 */
1015 	dst = &req->rq_private_buf.head[0];
1016 	src = &rqstp->rq_arg.head[0];
1017 	if (dst->iov_len < src->iov_len)
1018 		goto unlock_eagain; /* whatever; just giving up. */
1019 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1020 	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1021 	rqstp->rq_arg.len = 0;
1022 	spin_unlock_bh(&bc_xprt->transport_lock);
1023 	return 0;
1024 unlock_notfound:
1025 	printk(KERN_NOTICE
1026 		"%s: Got unrecognized reply: "
1027 		"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1028 		__func__, ntohl(calldir),
1029 		bc_xprt, ntohl(xid));
1030 unlock_eagain:
1031 	spin_unlock_bh(&bc_xprt->transport_lock);
1032 	return -EAGAIN;
1033 }
1034 
1035 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1036 {
1037 	int i = 0;
1038 	int t = 0;
1039 
1040 	while (t < len) {
1041 		vec[i].iov_base = page_address(pages[i]);
1042 		vec[i].iov_len = PAGE_SIZE;
1043 		i++;
1044 		t += PAGE_SIZE;
1045 	}
1046 	return i;
1047 }
1048 
1049 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1050 {
1051 	/* If we have more data, signal svc_xprt_enqueue() to try again */
1052 	dprintk("svc: TCP %s record (%d bytes)\n",
1053 		svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1054 		svc_sock_reclen(svsk));
1055 	svsk->sk_tcplen = 0;
1056 	svsk->sk_reclen = 0;
1057 }
1058 
1059 /*
1060  * Receive data from a TCP socket.
1061  */
1062 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1063 {
1064 	struct svc_sock	*svsk =
1065 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1066 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1067 	int		len;
1068 	struct kvec *vec;
1069 	unsigned int want, base;
1070 	__be32 *p;
1071 	__be32 calldir;
1072 	int pnum;
1073 
1074 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1075 		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1076 		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1077 		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1078 
1079 	len = svc_tcp_recv_record(svsk, rqstp);
1080 	if (len < 0)
1081 		goto error;
1082 
1083 	base = svc_tcp_restore_pages(svsk, rqstp);
1084 	want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1085 
1086 	vec = rqstp->rq_vec;
1087 
1088 	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1089 						svsk->sk_datalen + want);
1090 
1091 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1092 	rqstp->rq_next_page = rqstp->rq_respages + 1;
1093 
1094 	/* Now receive data */
1095 	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1096 	if (len >= 0) {
1097 		svsk->sk_tcplen += len;
1098 		svsk->sk_datalen += len;
1099 	}
1100 	if (len != want || !svc_sock_final_rec(svsk)) {
1101 		svc_tcp_save_pages(svsk, rqstp);
1102 		if (len < 0 && len != -EAGAIN)
1103 			goto err_delete;
1104 		if (len == want)
1105 			svc_tcp_fragment_received(svsk);
1106 		else
1107 			dprintk("svc: incomplete TCP record (%d of %d)\n",
1108 				(int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1109 				svc_sock_reclen(svsk));
1110 		goto err_noclose;
1111 	}
1112 
1113 	if (svsk->sk_datalen < 8) {
1114 		svsk->sk_datalen = 0;
1115 		goto err_delete; /* client is nuts. */
1116 	}
1117 
1118 	rqstp->rq_arg.len = svsk->sk_datalen;
1119 	rqstp->rq_arg.page_base = 0;
1120 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1121 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1122 		rqstp->rq_arg.page_len = 0;
1123 	} else
1124 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1125 
1126 	rqstp->rq_xprt_ctxt   = NULL;
1127 	rqstp->rq_prot	      = IPPROTO_TCP;
1128 	if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1129 		set_bit(RQ_LOCAL, &rqstp->rq_flags);
1130 	else
1131 		clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1132 
1133 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1134 	calldir = p[1];
1135 	if (calldir)
1136 		len = receive_cb_reply(svsk, rqstp);
1137 
1138 	/* Reset TCP read info */
1139 	svsk->sk_datalen = 0;
1140 	svc_tcp_fragment_received(svsk);
1141 
1142 	if (len < 0)
1143 		goto error;
1144 
1145 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1146 	if (serv->sv_stats)
1147 		serv->sv_stats->nettcpcnt++;
1148 
1149 	return rqstp->rq_arg.len;
1150 
1151 error:
1152 	if (len != -EAGAIN)
1153 		goto err_delete;
1154 	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1155 	return 0;
1156 err_delete:
1157 	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1158 	       svsk->sk_xprt.xpt_server->sv_name, -len);
1159 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1160 err_noclose:
1161 	return 0;	/* record not complete */
1162 }
1163 
1164 /*
1165  * Send out data on TCP socket.
1166  */
1167 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1168 {
1169 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1170 	int sent;
1171 	__be32 reclen;
1172 
1173 	/* Set up the first element of the reply kvec.
1174 	 * Any other kvecs that may be in use have been taken
1175 	 * care of by the server implementation itself.
1176 	 */
1177 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1178 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1179 
1180 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1181 	if (sent != xbufp->len) {
1182 		printk(KERN_NOTICE
1183 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1184 		       "- shutting down socket\n",
1185 		       rqstp->rq_xprt->xpt_server->sv_name,
1186 		       (sent<0)?"got error":"sent only",
1187 		       sent, xbufp->len);
1188 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1189 		svc_xprt_enqueue(rqstp->rq_xprt);
1190 		sent = -EAGAIN;
1191 	}
1192 	return sent;
1193 }
1194 
1195 /*
1196  * Setup response header. TCP has a 4B record length field.
1197  */
1198 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1199 {
1200 	struct kvec *resv = &rqstp->rq_res.head[0];
1201 
1202 	/* tcp needs a space for the record length... */
1203 	svc_putnl(resv, 0);
1204 }
1205 
1206 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1207 				       struct net *net,
1208 				       struct sockaddr *sa, int salen,
1209 				       int flags)
1210 {
1211 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1212 }
1213 
1214 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1215 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1216 					     struct net *, struct sockaddr *,
1217 					     int, int);
1218 static void svc_bc_sock_free(struct svc_xprt *xprt);
1219 
1220 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1221 				       struct net *net,
1222 				       struct sockaddr *sa, int salen,
1223 				       int flags)
1224 {
1225 	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1226 }
1227 
1228 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1229 {
1230 }
1231 
1232 static struct svc_xprt_ops svc_tcp_bc_ops = {
1233 	.xpo_create = svc_bc_tcp_create,
1234 	.xpo_detach = svc_bc_tcp_sock_detach,
1235 	.xpo_free = svc_bc_sock_free,
1236 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1237 	.xpo_secure_port = svc_sock_secure_port,
1238 };
1239 
1240 static struct svc_xprt_class svc_tcp_bc_class = {
1241 	.xcl_name = "tcp-bc",
1242 	.xcl_owner = THIS_MODULE,
1243 	.xcl_ops = &svc_tcp_bc_ops,
1244 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1245 };
1246 
1247 static void svc_init_bc_xprt_sock(void)
1248 {
1249 	svc_reg_xprt_class(&svc_tcp_bc_class);
1250 }
1251 
1252 static void svc_cleanup_bc_xprt_sock(void)
1253 {
1254 	svc_unreg_xprt_class(&svc_tcp_bc_class);
1255 }
1256 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1257 static void svc_init_bc_xprt_sock(void)
1258 {
1259 }
1260 
1261 static void svc_cleanup_bc_xprt_sock(void)
1262 {
1263 }
1264 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1265 
1266 static struct svc_xprt_ops svc_tcp_ops = {
1267 	.xpo_create = svc_tcp_create,
1268 	.xpo_recvfrom = svc_tcp_recvfrom,
1269 	.xpo_sendto = svc_tcp_sendto,
1270 	.xpo_release_rqst = svc_release_skb,
1271 	.xpo_detach = svc_tcp_sock_detach,
1272 	.xpo_free = svc_sock_free,
1273 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1274 	.xpo_has_wspace = svc_tcp_has_wspace,
1275 	.xpo_accept = svc_tcp_accept,
1276 	.xpo_secure_port = svc_sock_secure_port,
1277 	.xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1278 };
1279 
1280 static struct svc_xprt_class svc_tcp_class = {
1281 	.xcl_name = "tcp",
1282 	.xcl_owner = THIS_MODULE,
1283 	.xcl_ops = &svc_tcp_ops,
1284 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1285 	.xcl_ident = XPRT_TRANSPORT_TCP,
1286 };
1287 
1288 void svc_init_xprt_sock(void)
1289 {
1290 	svc_reg_xprt_class(&svc_tcp_class);
1291 	svc_reg_xprt_class(&svc_udp_class);
1292 	svc_init_bc_xprt_sock();
1293 }
1294 
1295 void svc_cleanup_xprt_sock(void)
1296 {
1297 	svc_unreg_xprt_class(&svc_tcp_class);
1298 	svc_unreg_xprt_class(&svc_udp_class);
1299 	svc_cleanup_bc_xprt_sock();
1300 }
1301 
1302 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1303 {
1304 	struct sock	*sk = svsk->sk_sk;
1305 
1306 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1307 		      &svsk->sk_xprt, serv);
1308 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1309 	set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1310 	if (sk->sk_state == TCP_LISTEN) {
1311 		dprintk("setting up TCP socket for listening\n");
1312 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1313 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1314 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1315 	} else {
1316 		dprintk("setting up TCP socket for reading\n");
1317 		sk->sk_state_change = svc_tcp_state_change;
1318 		sk->sk_data_ready = svc_data_ready;
1319 		sk->sk_write_space = svc_write_space;
1320 
1321 		svsk->sk_reclen = 0;
1322 		svsk->sk_tcplen = 0;
1323 		svsk->sk_datalen = 0;
1324 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1325 
1326 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1327 
1328 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1329 		switch (sk->sk_state) {
1330 		case TCP_SYN_RECV:
1331 		case TCP_ESTABLISHED:
1332 			break;
1333 		default:
1334 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1335 		}
1336 	}
1337 }
1338 
1339 void svc_sock_update_bufs(struct svc_serv *serv)
1340 {
1341 	/*
1342 	 * The number of server threads has changed. Update
1343 	 * rcvbuf and sndbuf accordingly on all sockets
1344 	 */
1345 	struct svc_sock *svsk;
1346 
1347 	spin_lock_bh(&serv->sv_lock);
1348 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1349 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1350 	spin_unlock_bh(&serv->sv_lock);
1351 }
1352 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1353 
1354 /*
1355  * Initialize socket for RPC use and create svc_sock struct
1356  */
1357 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1358 						struct socket *sock,
1359 						int flags)
1360 {
1361 	struct svc_sock	*svsk;
1362 	struct sock	*inet;
1363 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1364 	int		err = 0;
1365 
1366 	dprintk("svc: svc_setup_socket %p\n", sock);
1367 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1368 	if (!svsk)
1369 		return ERR_PTR(-ENOMEM);
1370 
1371 	inet = sock->sk;
1372 
1373 	/* Register socket with portmapper */
1374 	if (pmap_register)
1375 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1376 				     inet->sk_protocol,
1377 				     ntohs(inet_sk(inet)->inet_sport));
1378 
1379 	if (err < 0) {
1380 		kfree(svsk);
1381 		return ERR_PTR(err);
1382 	}
1383 
1384 	inet->sk_user_data = svsk;
1385 	svsk->sk_sock = sock;
1386 	svsk->sk_sk = inet;
1387 	svsk->sk_ostate = inet->sk_state_change;
1388 	svsk->sk_odata = inet->sk_data_ready;
1389 	svsk->sk_owspace = inet->sk_write_space;
1390 
1391 	/* Initialize the socket */
1392 	if (sock->type == SOCK_DGRAM)
1393 		svc_udp_init(svsk, serv);
1394 	else
1395 		svc_tcp_init(svsk, serv);
1396 
1397 	dprintk("svc: svc_setup_socket created %p (inet %p), "
1398 			"listen %d close %d\n",
1399 			svsk, svsk->sk_sk,
1400 			test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1401 			test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1402 
1403 	return svsk;
1404 }
1405 
1406 bool svc_alien_sock(struct net *net, int fd)
1407 {
1408 	int err;
1409 	struct socket *sock = sockfd_lookup(fd, &err);
1410 	bool ret = false;
1411 
1412 	if (!sock)
1413 		goto out;
1414 	if (sock_net(sock->sk) != net)
1415 		ret = true;
1416 	sockfd_put(sock);
1417 out:
1418 	return ret;
1419 }
1420 EXPORT_SYMBOL_GPL(svc_alien_sock);
1421 
1422 /**
1423  * svc_addsock - add a listener socket to an RPC service
1424  * @serv: pointer to RPC service to which to add a new listener
1425  * @fd: file descriptor of the new listener
1426  * @name_return: pointer to buffer to fill in with name of listener
1427  * @len: size of the buffer
1428  *
1429  * Fills in socket name and returns positive length of name if successful.
1430  * Name is terminated with '\n'.  On error, returns a negative errno
1431  * value.
1432  */
1433 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1434 		const size_t len)
1435 {
1436 	int err = 0;
1437 	struct socket *so = sockfd_lookup(fd, &err);
1438 	struct svc_sock *svsk = NULL;
1439 	struct sockaddr_storage addr;
1440 	struct sockaddr *sin = (struct sockaddr *)&addr;
1441 	int salen;
1442 
1443 	if (!so)
1444 		return err;
1445 	err = -EAFNOSUPPORT;
1446 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1447 		goto out;
1448 	err =  -EPROTONOSUPPORT;
1449 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1450 	    so->sk->sk_protocol != IPPROTO_UDP)
1451 		goto out;
1452 	err = -EISCONN;
1453 	if (so->state > SS_UNCONNECTED)
1454 		goto out;
1455 	err = -ENOENT;
1456 	if (!try_module_get(THIS_MODULE))
1457 		goto out;
1458 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1459 	if (IS_ERR(svsk)) {
1460 		module_put(THIS_MODULE);
1461 		err = PTR_ERR(svsk);
1462 		goto out;
1463 	}
1464 	if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1465 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1466 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1467 	return svc_one_sock_name(svsk, name_return, len);
1468 out:
1469 	sockfd_put(so);
1470 	return err;
1471 }
1472 EXPORT_SYMBOL_GPL(svc_addsock);
1473 
1474 /*
1475  * Create socket for RPC service.
1476  */
1477 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1478 					  int protocol,
1479 					  struct net *net,
1480 					  struct sockaddr *sin, int len,
1481 					  int flags)
1482 {
1483 	struct svc_sock	*svsk;
1484 	struct socket	*sock;
1485 	int		error;
1486 	int		type;
1487 	struct sockaddr_storage addr;
1488 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1489 	int		newlen;
1490 	int		family;
1491 	int		val;
1492 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1493 
1494 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1495 			serv->sv_program->pg_name, protocol,
1496 			__svc_print_addr(sin, buf, sizeof(buf)));
1497 
1498 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1499 		printk(KERN_WARNING "svc: only UDP and TCP "
1500 				"sockets supported\n");
1501 		return ERR_PTR(-EINVAL);
1502 	}
1503 
1504 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1505 	switch (sin->sa_family) {
1506 	case AF_INET6:
1507 		family = PF_INET6;
1508 		break;
1509 	case AF_INET:
1510 		family = PF_INET;
1511 		break;
1512 	default:
1513 		return ERR_PTR(-EINVAL);
1514 	}
1515 
1516 	error = __sock_create(net, family, type, protocol, &sock, 1);
1517 	if (error < 0)
1518 		return ERR_PTR(error);
1519 
1520 	svc_reclassify_socket(sock);
1521 
1522 	/*
1523 	 * If this is an PF_INET6 listener, we want to avoid
1524 	 * getting requests from IPv4 remotes.  Those should
1525 	 * be shunted to a PF_INET listener via rpcbind.
1526 	 */
1527 	val = 1;
1528 	if (family == PF_INET6)
1529 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1530 					(char *)&val, sizeof(val));
1531 
1532 	if (type == SOCK_STREAM)
1533 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1534 	error = kernel_bind(sock, sin, len);
1535 	if (error < 0)
1536 		goto bummer;
1537 
1538 	newlen = len;
1539 	error = kernel_getsockname(sock, newsin, &newlen);
1540 	if (error < 0)
1541 		goto bummer;
1542 
1543 	if (protocol == IPPROTO_TCP) {
1544 		if ((error = kernel_listen(sock, 64)) < 0)
1545 			goto bummer;
1546 	}
1547 
1548 	svsk = svc_setup_socket(serv, sock, flags);
1549 	if (IS_ERR(svsk)) {
1550 		error = PTR_ERR(svsk);
1551 		goto bummer;
1552 	}
1553 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1554 	return (struct svc_xprt *)svsk;
1555 bummer:
1556 	dprintk("svc: svc_create_socket error = %d\n", -error);
1557 	sock_release(sock);
1558 	return ERR_PTR(error);
1559 }
1560 
1561 /*
1562  * Detach the svc_sock from the socket so that no
1563  * more callbacks occur.
1564  */
1565 static void svc_sock_detach(struct svc_xprt *xprt)
1566 {
1567 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1568 	struct sock *sk = svsk->sk_sk;
1569 
1570 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1571 
1572 	/* put back the old socket callbacks */
1573 	lock_sock(sk);
1574 	sk->sk_state_change = svsk->sk_ostate;
1575 	sk->sk_data_ready = svsk->sk_odata;
1576 	sk->sk_write_space = svsk->sk_owspace;
1577 	sk->sk_user_data = NULL;
1578 	release_sock(sk);
1579 }
1580 
1581 /*
1582  * Disconnect the socket, and reset the callbacks
1583  */
1584 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1585 {
1586 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1587 
1588 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1589 
1590 	svc_sock_detach(xprt);
1591 
1592 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1593 		svc_tcp_clear_pages(svsk);
1594 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1595 	}
1596 }
1597 
1598 /*
1599  * Free the svc_sock's socket resources and the svc_sock itself.
1600  */
1601 static void svc_sock_free(struct svc_xprt *xprt)
1602 {
1603 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1604 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1605 
1606 	if (svsk->sk_sock->file)
1607 		sockfd_put(svsk->sk_sock);
1608 	else
1609 		sock_release(svsk->sk_sock);
1610 	kfree(svsk);
1611 }
1612 
1613 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1614 /*
1615  * Create a back channel svc_xprt which shares the fore channel socket.
1616  */
1617 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1618 					     int protocol,
1619 					     struct net *net,
1620 					     struct sockaddr *sin, int len,
1621 					     int flags)
1622 {
1623 	struct svc_sock *svsk;
1624 	struct svc_xprt *xprt;
1625 
1626 	if (protocol != IPPROTO_TCP) {
1627 		printk(KERN_WARNING "svc: only TCP sockets"
1628 			" supported on shared back channel\n");
1629 		return ERR_PTR(-EINVAL);
1630 	}
1631 
1632 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1633 	if (!svsk)
1634 		return ERR_PTR(-ENOMEM);
1635 
1636 	xprt = &svsk->sk_xprt;
1637 	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1638 
1639 	serv->sv_bc_xprt = xprt;
1640 
1641 	return xprt;
1642 }
1643 
1644 /*
1645  * Free a back channel svc_sock.
1646  */
1647 static void svc_bc_sock_free(struct svc_xprt *xprt)
1648 {
1649 	if (xprt)
1650 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1651 }
1652 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1653