xref: /openbmc/linux/net/sunrpc/svcsock.c (revision 133f9794)
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 	iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, iov, nr, buflen);
342 	len = sock_recvmsg(svsk->sk_sock, &msg, 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 	lock_sock(sock->sk);
388 	sock->sk->sk_sndbuf = snd * 2;
389 	sock->sk->sk_rcvbuf = rcv * 2;
390 	sock->sk->sk_write_space(sock->sk);
391 	release_sock(sock->sk);
392 }
393 
394 static int svc_sock_secure_port(struct svc_rqst *rqstp)
395 {
396 	return svc_port_is_privileged(svc_addr(rqstp));
397 }
398 
399 /*
400  * INET callback when data has been received on the socket.
401  */
402 static void svc_data_ready(struct sock *sk)
403 {
404 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
405 
406 	if (svsk) {
407 		dprintk("svc: socket %p(inet %p), busy=%d\n",
408 			svsk, sk,
409 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
410 
411 		/* Refer to svc_setup_socket() for details. */
412 		rmb();
413 		svsk->sk_odata(sk);
414 		if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
415 			svc_xprt_enqueue(&svsk->sk_xprt);
416 	}
417 }
418 
419 /*
420  * INET callback when space is newly available on the socket.
421  */
422 static void svc_write_space(struct sock *sk)
423 {
424 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
425 
426 	if (svsk) {
427 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
428 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
429 
430 		/* Refer to svc_setup_socket() for details. */
431 		rmb();
432 		svsk->sk_owspace(sk);
433 		svc_xprt_enqueue(&svsk->sk_xprt);
434 	}
435 }
436 
437 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
438 {
439 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
440 
441 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
442 		return 1;
443 	return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
444 }
445 
446 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
447 {
448 	struct svc_sock *svsk;
449 	struct socket *sock;
450 	struct linger no_linger = {
451 		.l_onoff = 1,
452 		.l_linger = 0,
453 	};
454 
455 	svsk = container_of(xprt, struct svc_sock, sk_xprt);
456 	sock = svsk->sk_sock;
457 	kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
458 			  (char *)&no_linger, sizeof(no_linger));
459 }
460 
461 /*
462  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
463  */
464 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
465 				     struct cmsghdr *cmh)
466 {
467 	struct in_pktinfo *pki = CMSG_DATA(cmh);
468 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
469 
470 	if (cmh->cmsg_type != IP_PKTINFO)
471 		return 0;
472 
473 	daddr->sin_family = AF_INET;
474 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
475 	return 1;
476 }
477 
478 /*
479  * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
480  */
481 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
482 				     struct cmsghdr *cmh)
483 {
484 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
485 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
486 
487 	if (cmh->cmsg_type != IPV6_PKTINFO)
488 		return 0;
489 
490 	daddr->sin6_family = AF_INET6;
491 	daddr->sin6_addr = pki->ipi6_addr;
492 	daddr->sin6_scope_id = pki->ipi6_ifindex;
493 	return 1;
494 }
495 
496 /*
497  * Copy the UDP datagram's destination address to the rqstp structure.
498  * The 'destination' address in this case is the address to which the
499  * peer sent the datagram, i.e. our local address. For multihomed
500  * hosts, this can change from msg to msg. Note that only the IP
501  * address changes, the port number should remain the same.
502  */
503 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
504 				    struct cmsghdr *cmh)
505 {
506 	switch (cmh->cmsg_level) {
507 	case SOL_IP:
508 		return svc_udp_get_dest_address4(rqstp, cmh);
509 	case SOL_IPV6:
510 		return svc_udp_get_dest_address6(rqstp, cmh);
511 	}
512 
513 	return 0;
514 }
515 
516 /*
517  * Receive a datagram from a UDP socket.
518  */
519 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
520 {
521 	struct svc_sock	*svsk =
522 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
523 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
524 	struct sk_buff	*skb;
525 	union {
526 		struct cmsghdr	hdr;
527 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
528 	} buffer;
529 	struct cmsghdr *cmh = &buffer.hdr;
530 	struct msghdr msg = {
531 		.msg_name = svc_addr(rqstp),
532 		.msg_control = cmh,
533 		.msg_controllen = sizeof(buffer),
534 		.msg_flags = MSG_DONTWAIT,
535 	};
536 	size_t len;
537 	int err;
538 
539 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
540 	    /* udp sockets need large rcvbuf as all pending
541 	     * requests are still in that buffer.  sndbuf must
542 	     * also be large enough that there is enough space
543 	     * for one reply per thread.  We count all threads
544 	     * rather than threads in a particular pool, which
545 	     * provides an upper bound on the number of threads
546 	     * which will access the socket.
547 	     */
548 	    svc_sock_setbufsize(svsk->sk_sock,
549 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
550 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
551 
552 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
553 	skb = NULL;
554 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
555 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
556 	if (err >= 0)
557 		skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
558 
559 	if (skb == NULL) {
560 		if (err != -EAGAIN) {
561 			/* possibly an icmp error */
562 			dprintk("svc: recvfrom returned error %d\n", -err);
563 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
564 		}
565 		return 0;
566 	}
567 	len = svc_addr_len(svc_addr(rqstp));
568 	rqstp->rq_addrlen = len;
569 	if (skb->tstamp == 0) {
570 		skb->tstamp = ktime_get_real();
571 		/* Don't enable netstamp, sunrpc doesn't
572 		   need that much accuracy */
573 	}
574 	svsk->sk_sk->sk_stamp = skb->tstamp;
575 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
576 
577 	len  = skb->len;
578 	rqstp->rq_arg.len = len;
579 
580 	rqstp->rq_prot = IPPROTO_UDP;
581 
582 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
583 		net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
584 				     cmh->cmsg_level, cmh->cmsg_type);
585 		goto out_free;
586 	}
587 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
588 
589 	if (skb_is_nonlinear(skb)) {
590 		/* we have to copy */
591 		local_bh_disable();
592 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
593 			local_bh_enable();
594 			/* checksum error */
595 			goto out_free;
596 		}
597 		local_bh_enable();
598 		consume_skb(skb);
599 	} else {
600 		/* we can use it in-place */
601 		rqstp->rq_arg.head[0].iov_base = skb->data;
602 		rqstp->rq_arg.head[0].iov_len = len;
603 		if (skb_checksum_complete(skb))
604 			goto out_free;
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 	rqstp->rq_next_page = rqstp->rq_respages+1;
619 
620 	if (serv->sv_stats)
621 		serv->sv_stats->netudpcnt++;
622 
623 	return len;
624 out_free:
625 	kfree_skb(skb);
626 	return 0;
627 }
628 
629 static int
630 svc_udp_sendto(struct svc_rqst *rqstp)
631 {
632 	int		error;
633 
634 	error = svc_sendto(rqstp, &rqstp->rq_res);
635 	if (error == -ECONNREFUSED)
636 		/* ICMP error on earlier request. */
637 		error = svc_sendto(rqstp, &rqstp->rq_res);
638 
639 	return error;
640 }
641 
642 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
643 {
644 }
645 
646 static int svc_udp_has_wspace(struct svc_xprt *xprt)
647 {
648 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
649 	struct svc_serv	*serv = xprt->xpt_server;
650 	unsigned long required;
651 
652 	/*
653 	 * Set the SOCK_NOSPACE flag before checking the available
654 	 * sock space.
655 	 */
656 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
657 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
658 	if (required*2 > sock_wspace(svsk->sk_sk))
659 		return 0;
660 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
661 	return 1;
662 }
663 
664 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
665 {
666 	BUG();
667 	return NULL;
668 }
669 
670 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
671 {
672 }
673 
674 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
675 				       struct net *net,
676 				       struct sockaddr *sa, int salen,
677 				       int flags)
678 {
679 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
680 }
681 
682 static const struct svc_xprt_ops svc_udp_ops = {
683 	.xpo_create = svc_udp_create,
684 	.xpo_recvfrom = svc_udp_recvfrom,
685 	.xpo_sendto = svc_udp_sendto,
686 	.xpo_release_rqst = svc_release_udp_skb,
687 	.xpo_detach = svc_sock_detach,
688 	.xpo_free = svc_sock_free,
689 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
690 	.xpo_has_wspace = svc_udp_has_wspace,
691 	.xpo_accept = svc_udp_accept,
692 	.xpo_secure_port = svc_sock_secure_port,
693 	.xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
694 };
695 
696 static struct svc_xprt_class svc_udp_class = {
697 	.xcl_name = "udp",
698 	.xcl_owner = THIS_MODULE,
699 	.xcl_ops = &svc_udp_ops,
700 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
701 	.xcl_ident = XPRT_TRANSPORT_UDP,
702 };
703 
704 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
705 {
706 	int err, level, optname, one = 1;
707 
708 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
709 		      &svsk->sk_xprt, serv);
710 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
711 	svsk->sk_sk->sk_data_ready = svc_data_ready;
712 	svsk->sk_sk->sk_write_space = svc_write_space;
713 
714 	/* initialise setting must have enough space to
715 	 * receive and respond to one request.
716 	 * svc_udp_recvfrom will re-adjust if necessary
717 	 */
718 	svc_sock_setbufsize(svsk->sk_sock,
719 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
720 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
721 
722 	/* data might have come in before data_ready set up */
723 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
724 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
725 
726 	/* make sure we get destination address info */
727 	switch (svsk->sk_sk->sk_family) {
728 	case AF_INET:
729 		level = SOL_IP;
730 		optname = IP_PKTINFO;
731 		break;
732 	case AF_INET6:
733 		level = SOL_IPV6;
734 		optname = IPV6_RECVPKTINFO;
735 		break;
736 	default:
737 		BUG();
738 	}
739 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
740 					(char *)&one, sizeof(one));
741 	dprintk("svc: kernel_setsockopt returned %d\n", err);
742 }
743 
744 /*
745  * A data_ready event on a listening socket means there's a connection
746  * pending. Do not use state_change as a substitute for it.
747  */
748 static void svc_tcp_listen_data_ready(struct sock *sk)
749 {
750 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
751 
752 	dprintk("svc: socket %p TCP (listen) state change %d\n",
753 		sk, sk->sk_state);
754 
755 	if (svsk) {
756 		/* Refer to svc_setup_socket() for details. */
757 		rmb();
758 		svsk->sk_odata(sk);
759 	}
760 
761 	/*
762 	 * This callback may called twice when a new connection
763 	 * is established as a child socket inherits everything
764 	 * from a parent LISTEN socket.
765 	 * 1) data_ready method of the parent socket will be called
766 	 *    when one of child sockets become ESTABLISHED.
767 	 * 2) data_ready method of the child socket may be called
768 	 *    when it receives data before the socket is accepted.
769 	 * In case of 2, we should ignore it silently.
770 	 */
771 	if (sk->sk_state == TCP_LISTEN) {
772 		if (svsk) {
773 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
774 			svc_xprt_enqueue(&svsk->sk_xprt);
775 		} else
776 			printk("svc: socket %p: no user data\n", sk);
777 	}
778 }
779 
780 /*
781  * A state change on a connected socket means it's dying or dead.
782  */
783 static void svc_tcp_state_change(struct sock *sk)
784 {
785 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
786 
787 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
788 		sk, sk->sk_state, sk->sk_user_data);
789 
790 	if (!svsk)
791 		printk("svc: socket %p: no user data\n", sk);
792 	else {
793 		/* Refer to svc_setup_socket() for details. */
794 		rmb();
795 		svsk->sk_ostate(sk);
796 		if (sk->sk_state != TCP_ESTABLISHED) {
797 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
798 			svc_xprt_enqueue(&svsk->sk_xprt);
799 		}
800 	}
801 }
802 
803 /*
804  * Accept a TCP connection
805  */
806 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
807 {
808 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
809 	struct sockaddr_storage addr;
810 	struct sockaddr	*sin = (struct sockaddr *) &addr;
811 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
812 	struct socket	*sock = svsk->sk_sock;
813 	struct socket	*newsock;
814 	struct svc_sock	*newsvsk;
815 	int		err, slen;
816 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
817 
818 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
819 	if (!sock)
820 		return NULL;
821 
822 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
823 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
824 	if (err < 0) {
825 		if (err == -ENOMEM)
826 			printk(KERN_WARNING "%s: no more sockets!\n",
827 			       serv->sv_name);
828 		else if (err != -EAGAIN)
829 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
830 					     serv->sv_name, -err);
831 		return NULL;
832 	}
833 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
834 
835 	err = kernel_getpeername(newsock, sin, &slen);
836 	if (err < 0) {
837 		net_warn_ratelimited("%s: peername failed (err %d)!\n",
838 				     serv->sv_name, -err);
839 		goto failed;		/* aborted connection or whatever */
840 	}
841 
842 	/* Ideally, we would want to reject connections from unauthorized
843 	 * hosts here, but when we get encryption, the IP of the host won't
844 	 * tell us anything.  For now just warn about unpriv connections.
845 	 */
846 	if (!svc_port_is_privileged(sin)) {
847 		dprintk("%s: connect from unprivileged port: %s\n",
848 			serv->sv_name,
849 			__svc_print_addr(sin, buf, sizeof(buf)));
850 	}
851 	dprintk("%s: connect from %s\n", serv->sv_name,
852 		__svc_print_addr(sin, buf, sizeof(buf)));
853 
854 	/* Reset the inherited callbacks before calling svc_setup_socket */
855 	newsock->sk->sk_state_change = svsk->sk_ostate;
856 	newsock->sk->sk_data_ready = svsk->sk_odata;
857 	newsock->sk->sk_write_space = svsk->sk_owspace;
858 
859 	/* make sure that a write doesn't block forever when
860 	 * low on memory
861 	 */
862 	newsock->sk->sk_sndtimeo = HZ*30;
863 
864 	newsvsk = svc_setup_socket(serv, newsock,
865 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
866 	if (IS_ERR(newsvsk))
867 		goto failed;
868 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
869 	err = kernel_getsockname(newsock, sin, &slen);
870 	if (unlikely(err < 0)) {
871 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
872 		slen = offsetof(struct sockaddr, sa_data);
873 	}
874 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
875 
876 	if (sock_is_loopback(newsock->sk))
877 		set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
878 	else
879 		clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
880 	if (serv->sv_stats)
881 		serv->sv_stats->nettcpconn++;
882 
883 	return &newsvsk->sk_xprt;
884 
885 failed:
886 	sock_release(newsock);
887 	return NULL;
888 }
889 
890 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
891 {
892 	unsigned int i, len, npages;
893 
894 	if (svsk->sk_datalen == 0)
895 		return 0;
896 	len = svsk->sk_datalen;
897 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
898 	for (i = 0; i < npages; i++) {
899 		if (rqstp->rq_pages[i] != NULL)
900 			put_page(rqstp->rq_pages[i]);
901 		BUG_ON(svsk->sk_pages[i] == NULL);
902 		rqstp->rq_pages[i] = svsk->sk_pages[i];
903 		svsk->sk_pages[i] = NULL;
904 	}
905 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
906 	return len;
907 }
908 
909 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
910 {
911 	unsigned int i, len, npages;
912 
913 	if (svsk->sk_datalen == 0)
914 		return;
915 	len = svsk->sk_datalen;
916 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
917 	for (i = 0; i < npages; i++) {
918 		svsk->sk_pages[i] = rqstp->rq_pages[i];
919 		rqstp->rq_pages[i] = NULL;
920 	}
921 }
922 
923 static void svc_tcp_clear_pages(struct svc_sock *svsk)
924 {
925 	unsigned int i, len, npages;
926 
927 	if (svsk->sk_datalen == 0)
928 		goto out;
929 	len = svsk->sk_datalen;
930 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
931 	for (i = 0; i < npages; i++) {
932 		if (svsk->sk_pages[i] == NULL) {
933 			WARN_ON_ONCE(1);
934 			continue;
935 		}
936 		put_page(svsk->sk_pages[i]);
937 		svsk->sk_pages[i] = NULL;
938 	}
939 out:
940 	svsk->sk_tcplen = 0;
941 	svsk->sk_datalen = 0;
942 }
943 
944 /*
945  * Receive fragment record header.
946  * If we haven't gotten the record length yet, get the next four bytes.
947  */
948 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
949 {
950 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
951 	unsigned int want;
952 	int len;
953 
954 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
955 		struct kvec	iov;
956 
957 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
958 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
959 		iov.iov_len  = want;
960 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
961 			goto error;
962 		svsk->sk_tcplen += len;
963 
964 		if (len < want) {
965 			dprintk("svc: short recvfrom while reading record "
966 				"length (%d of %d)\n", len, want);
967 			return -EAGAIN;
968 		}
969 
970 		dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
971 		if (svc_sock_reclen(svsk) + svsk->sk_datalen >
972 							serv->sv_max_mesg) {
973 			net_notice_ratelimited("RPC: fragment too large: %d\n",
974 					svc_sock_reclen(svsk));
975 			goto err_delete;
976 		}
977 	}
978 
979 	return svc_sock_reclen(svsk);
980 error:
981 	dprintk("RPC: TCP recv_record got %d\n", len);
982 	return len;
983 err_delete:
984 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
985 	return -EAGAIN;
986 }
987 
988 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
989 {
990 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
991 	struct rpc_rqst *req = NULL;
992 	struct kvec *src, *dst;
993 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
994 	__be32 xid;
995 	__be32 calldir;
996 
997 	xid = *p++;
998 	calldir = *p;
999 
1000 	if (!bc_xprt)
1001 		return -EAGAIN;
1002 	spin_lock(&bc_xprt->recv_lock);
1003 	req = xprt_lookup_rqst(bc_xprt, xid);
1004 	if (!req)
1005 		goto unlock_notfound;
1006 
1007 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1008 	/*
1009 	 * XXX!: cheating for now!  Only copying HEAD.
1010 	 * But we know this is good enough for now (in fact, for any
1011 	 * callback reply in the forseeable future).
1012 	 */
1013 	dst = &req->rq_private_buf.head[0];
1014 	src = &rqstp->rq_arg.head[0];
1015 	if (dst->iov_len < src->iov_len)
1016 		goto unlock_eagain; /* whatever; just giving up. */
1017 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1018 	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1019 	rqstp->rq_arg.len = 0;
1020 	spin_unlock(&bc_xprt->recv_lock);
1021 	return 0;
1022 unlock_notfound:
1023 	printk(KERN_NOTICE
1024 		"%s: Got unrecognized reply: "
1025 		"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1026 		__func__, ntohl(calldir),
1027 		bc_xprt, ntohl(xid));
1028 unlock_eagain:
1029 	spin_unlock(&bc_xprt->recv_lock);
1030 	return -EAGAIN;
1031 }
1032 
1033 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1034 {
1035 	int i = 0;
1036 	int t = 0;
1037 
1038 	while (t < len) {
1039 		vec[i].iov_base = page_address(pages[i]);
1040 		vec[i].iov_len = PAGE_SIZE;
1041 		i++;
1042 		t += PAGE_SIZE;
1043 	}
1044 	return i;
1045 }
1046 
1047 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1048 {
1049 	/* If we have more data, signal svc_xprt_enqueue() to try again */
1050 	dprintk("svc: TCP %s record (%d bytes)\n",
1051 		svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1052 		svc_sock_reclen(svsk));
1053 	svsk->sk_tcplen = 0;
1054 	svsk->sk_reclen = 0;
1055 }
1056 
1057 /*
1058  * Receive data from a TCP socket.
1059  */
1060 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1061 {
1062 	struct svc_sock	*svsk =
1063 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1064 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1065 	int		len;
1066 	struct kvec *vec;
1067 	unsigned int want, base;
1068 	__be32 *p;
1069 	__be32 calldir;
1070 	int pnum;
1071 
1072 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1073 		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1074 		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1075 		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1076 
1077 	len = svc_tcp_recv_record(svsk, rqstp);
1078 	if (len < 0)
1079 		goto error;
1080 
1081 	base = svc_tcp_restore_pages(svsk, rqstp);
1082 	want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1083 
1084 	vec = rqstp->rq_vec;
1085 
1086 	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1087 						svsk->sk_datalen + want);
1088 
1089 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1090 	rqstp->rq_next_page = rqstp->rq_respages + 1;
1091 
1092 	/* Now receive data */
1093 	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1094 	if (len >= 0) {
1095 		svsk->sk_tcplen += len;
1096 		svsk->sk_datalen += len;
1097 	}
1098 	if (len != want || !svc_sock_final_rec(svsk)) {
1099 		svc_tcp_save_pages(svsk, rqstp);
1100 		if (len < 0 && len != -EAGAIN)
1101 			goto err_delete;
1102 		if (len == want)
1103 			svc_tcp_fragment_received(svsk);
1104 		else
1105 			dprintk("svc: incomplete TCP record (%d of %d)\n",
1106 				(int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1107 				svc_sock_reclen(svsk));
1108 		goto err_noclose;
1109 	}
1110 
1111 	if (svsk->sk_datalen < 8) {
1112 		svsk->sk_datalen = 0;
1113 		goto err_delete; /* client is nuts. */
1114 	}
1115 
1116 	rqstp->rq_arg.len = svsk->sk_datalen;
1117 	rqstp->rq_arg.page_base = 0;
1118 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1119 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1120 		rqstp->rq_arg.page_len = 0;
1121 	} else
1122 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1123 
1124 	rqstp->rq_xprt_ctxt   = NULL;
1125 	rqstp->rq_prot	      = IPPROTO_TCP;
1126 	if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1127 		set_bit(RQ_LOCAL, &rqstp->rq_flags);
1128 	else
1129 		clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1130 
1131 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1132 	calldir = p[1];
1133 	if (calldir)
1134 		len = receive_cb_reply(svsk, rqstp);
1135 
1136 	/* Reset TCP read info */
1137 	svsk->sk_datalen = 0;
1138 	svc_tcp_fragment_received(svsk);
1139 
1140 	if (len < 0)
1141 		goto error;
1142 
1143 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1144 	if (serv->sv_stats)
1145 		serv->sv_stats->nettcpcnt++;
1146 
1147 	return rqstp->rq_arg.len;
1148 
1149 error:
1150 	if (len != -EAGAIN)
1151 		goto err_delete;
1152 	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1153 	return 0;
1154 err_delete:
1155 	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1156 	       svsk->sk_xprt.xpt_server->sv_name, -len);
1157 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1158 err_noclose:
1159 	return 0;	/* record not complete */
1160 }
1161 
1162 /*
1163  * Send out data on TCP socket.
1164  */
1165 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1166 {
1167 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1168 	int sent;
1169 	__be32 reclen;
1170 
1171 	/* Set up the first element of the reply kvec.
1172 	 * Any other kvecs that may be in use have been taken
1173 	 * care of by the server implementation itself.
1174 	 */
1175 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1176 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1177 
1178 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1179 	if (sent != xbufp->len) {
1180 		printk(KERN_NOTICE
1181 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1182 		       "- shutting down socket\n",
1183 		       rqstp->rq_xprt->xpt_server->sv_name,
1184 		       (sent<0)?"got error":"sent only",
1185 		       sent, xbufp->len);
1186 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1187 		svc_xprt_enqueue(rqstp->rq_xprt);
1188 		sent = -EAGAIN;
1189 	}
1190 	return sent;
1191 }
1192 
1193 /*
1194  * Setup response header. TCP has a 4B record length field.
1195  */
1196 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1197 {
1198 	struct kvec *resv = &rqstp->rq_res.head[0];
1199 
1200 	/* tcp needs a space for the record length... */
1201 	svc_putnl(resv, 0);
1202 }
1203 
1204 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1205 				       struct net *net,
1206 				       struct sockaddr *sa, int salen,
1207 				       int flags)
1208 {
1209 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1210 }
1211 
1212 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1213 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1214 					     struct net *, struct sockaddr *,
1215 					     int, int);
1216 static void svc_bc_sock_free(struct svc_xprt *xprt);
1217 
1218 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1219 				       struct net *net,
1220 				       struct sockaddr *sa, int salen,
1221 				       int flags)
1222 {
1223 	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1224 }
1225 
1226 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1227 {
1228 }
1229 
1230 static const struct svc_xprt_ops svc_tcp_bc_ops = {
1231 	.xpo_create = svc_bc_tcp_create,
1232 	.xpo_detach = svc_bc_tcp_sock_detach,
1233 	.xpo_free = svc_bc_sock_free,
1234 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1235 	.xpo_secure_port = svc_sock_secure_port,
1236 };
1237 
1238 static struct svc_xprt_class svc_tcp_bc_class = {
1239 	.xcl_name = "tcp-bc",
1240 	.xcl_owner = THIS_MODULE,
1241 	.xcl_ops = &svc_tcp_bc_ops,
1242 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1243 };
1244 
1245 static void svc_init_bc_xprt_sock(void)
1246 {
1247 	svc_reg_xprt_class(&svc_tcp_bc_class);
1248 }
1249 
1250 static void svc_cleanup_bc_xprt_sock(void)
1251 {
1252 	svc_unreg_xprt_class(&svc_tcp_bc_class);
1253 }
1254 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1255 static void svc_init_bc_xprt_sock(void)
1256 {
1257 }
1258 
1259 static void svc_cleanup_bc_xprt_sock(void)
1260 {
1261 }
1262 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1263 
1264 static const struct svc_xprt_ops svc_tcp_ops = {
1265 	.xpo_create = svc_tcp_create,
1266 	.xpo_recvfrom = svc_tcp_recvfrom,
1267 	.xpo_sendto = svc_tcp_sendto,
1268 	.xpo_release_rqst = svc_release_skb,
1269 	.xpo_detach = svc_tcp_sock_detach,
1270 	.xpo_free = svc_sock_free,
1271 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1272 	.xpo_has_wspace = svc_tcp_has_wspace,
1273 	.xpo_accept = svc_tcp_accept,
1274 	.xpo_secure_port = svc_sock_secure_port,
1275 	.xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1276 };
1277 
1278 static struct svc_xprt_class svc_tcp_class = {
1279 	.xcl_name = "tcp",
1280 	.xcl_owner = THIS_MODULE,
1281 	.xcl_ops = &svc_tcp_ops,
1282 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1283 	.xcl_ident = XPRT_TRANSPORT_TCP,
1284 };
1285 
1286 void svc_init_xprt_sock(void)
1287 {
1288 	svc_reg_xprt_class(&svc_tcp_class);
1289 	svc_reg_xprt_class(&svc_udp_class);
1290 	svc_init_bc_xprt_sock();
1291 }
1292 
1293 void svc_cleanup_xprt_sock(void)
1294 {
1295 	svc_unreg_xprt_class(&svc_tcp_class);
1296 	svc_unreg_xprt_class(&svc_udp_class);
1297 	svc_cleanup_bc_xprt_sock();
1298 }
1299 
1300 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1301 {
1302 	struct sock	*sk = svsk->sk_sk;
1303 
1304 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1305 		      &svsk->sk_xprt, serv);
1306 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1307 	set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1308 	if (sk->sk_state == TCP_LISTEN) {
1309 		dprintk("setting up TCP socket for listening\n");
1310 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1311 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1312 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1313 	} else {
1314 		dprintk("setting up TCP socket for reading\n");
1315 		sk->sk_state_change = svc_tcp_state_change;
1316 		sk->sk_data_ready = svc_data_ready;
1317 		sk->sk_write_space = svc_write_space;
1318 
1319 		svsk->sk_reclen = 0;
1320 		svsk->sk_tcplen = 0;
1321 		svsk->sk_datalen = 0;
1322 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1323 
1324 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1325 
1326 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1327 		switch (sk->sk_state) {
1328 		case TCP_SYN_RECV:
1329 		case TCP_ESTABLISHED:
1330 			break;
1331 		default:
1332 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1333 		}
1334 	}
1335 }
1336 
1337 void svc_sock_update_bufs(struct svc_serv *serv)
1338 {
1339 	/*
1340 	 * The number of server threads has changed. Update
1341 	 * rcvbuf and sndbuf accordingly on all sockets
1342 	 */
1343 	struct svc_sock *svsk;
1344 
1345 	spin_lock_bh(&serv->sv_lock);
1346 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1347 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1348 	spin_unlock_bh(&serv->sv_lock);
1349 }
1350 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1351 
1352 /*
1353  * Initialize socket for RPC use and create svc_sock struct
1354  */
1355 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1356 						struct socket *sock,
1357 						int flags)
1358 {
1359 	struct svc_sock	*svsk;
1360 	struct sock	*inet;
1361 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1362 	int		err = 0;
1363 
1364 	dprintk("svc: svc_setup_socket %p\n", sock);
1365 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1366 	if (!svsk)
1367 		return ERR_PTR(-ENOMEM);
1368 
1369 	inet = sock->sk;
1370 
1371 	/* Register socket with portmapper */
1372 	if (pmap_register)
1373 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1374 				     inet->sk_protocol,
1375 				     ntohs(inet_sk(inet)->inet_sport));
1376 
1377 	if (err < 0) {
1378 		kfree(svsk);
1379 		return ERR_PTR(err);
1380 	}
1381 
1382 	svsk->sk_sock = sock;
1383 	svsk->sk_sk = inet;
1384 	svsk->sk_ostate = inet->sk_state_change;
1385 	svsk->sk_odata = inet->sk_data_ready;
1386 	svsk->sk_owspace = inet->sk_write_space;
1387 	/*
1388 	 * This barrier is necessary in order to prevent race condition
1389 	 * with svc_data_ready(), svc_listen_data_ready() and others
1390 	 * when calling callbacks above.
1391 	 */
1392 	wmb();
1393 	inet->sk_user_data = svsk;
1394 
1395 	/* Initialize the socket */
1396 	if (sock->type == SOCK_DGRAM)
1397 		svc_udp_init(svsk, serv);
1398 	else
1399 		svc_tcp_init(svsk, serv);
1400 
1401 	dprintk("svc: svc_setup_socket created %p (inet %p), "
1402 			"listen %d close %d\n",
1403 			svsk, svsk->sk_sk,
1404 			test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1405 			test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1406 
1407 	return svsk;
1408 }
1409 
1410 bool svc_alien_sock(struct net *net, int fd)
1411 {
1412 	int err;
1413 	struct socket *sock = sockfd_lookup(fd, &err);
1414 	bool ret = false;
1415 
1416 	if (!sock)
1417 		goto out;
1418 	if (sock_net(sock->sk) != net)
1419 		ret = true;
1420 	sockfd_put(sock);
1421 out:
1422 	return ret;
1423 }
1424 EXPORT_SYMBOL_GPL(svc_alien_sock);
1425 
1426 /**
1427  * svc_addsock - add a listener socket to an RPC service
1428  * @serv: pointer to RPC service to which to add a new listener
1429  * @fd: file descriptor of the new listener
1430  * @name_return: pointer to buffer to fill in with name of listener
1431  * @len: size of the buffer
1432  *
1433  * Fills in socket name and returns positive length of name if successful.
1434  * Name is terminated with '\n'.  On error, returns a negative errno
1435  * value.
1436  */
1437 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1438 		const size_t len)
1439 {
1440 	int err = 0;
1441 	struct socket *so = sockfd_lookup(fd, &err);
1442 	struct svc_sock *svsk = NULL;
1443 	struct sockaddr_storage addr;
1444 	struct sockaddr *sin = (struct sockaddr *)&addr;
1445 	int salen;
1446 
1447 	if (!so)
1448 		return err;
1449 	err = -EAFNOSUPPORT;
1450 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1451 		goto out;
1452 	err =  -EPROTONOSUPPORT;
1453 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1454 	    so->sk->sk_protocol != IPPROTO_UDP)
1455 		goto out;
1456 	err = -EISCONN;
1457 	if (so->state > SS_UNCONNECTED)
1458 		goto out;
1459 	err = -ENOENT;
1460 	if (!try_module_get(THIS_MODULE))
1461 		goto out;
1462 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1463 	if (IS_ERR(svsk)) {
1464 		module_put(THIS_MODULE);
1465 		err = PTR_ERR(svsk);
1466 		goto out;
1467 	}
1468 	if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1469 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1470 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1471 	return svc_one_sock_name(svsk, name_return, len);
1472 out:
1473 	sockfd_put(so);
1474 	return err;
1475 }
1476 EXPORT_SYMBOL_GPL(svc_addsock);
1477 
1478 /*
1479  * Create socket for RPC service.
1480  */
1481 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1482 					  int protocol,
1483 					  struct net *net,
1484 					  struct sockaddr *sin, int len,
1485 					  int flags)
1486 {
1487 	struct svc_sock	*svsk;
1488 	struct socket	*sock;
1489 	int		error;
1490 	int		type;
1491 	struct sockaddr_storage addr;
1492 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1493 	int		newlen;
1494 	int		family;
1495 	int		val;
1496 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1497 
1498 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1499 			serv->sv_program->pg_name, protocol,
1500 			__svc_print_addr(sin, buf, sizeof(buf)));
1501 
1502 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1503 		printk(KERN_WARNING "svc: only UDP and TCP "
1504 				"sockets supported\n");
1505 		return ERR_PTR(-EINVAL);
1506 	}
1507 
1508 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1509 	switch (sin->sa_family) {
1510 	case AF_INET6:
1511 		family = PF_INET6;
1512 		break;
1513 	case AF_INET:
1514 		family = PF_INET;
1515 		break;
1516 	default:
1517 		return ERR_PTR(-EINVAL);
1518 	}
1519 
1520 	error = __sock_create(net, family, type, protocol, &sock, 1);
1521 	if (error < 0)
1522 		return ERR_PTR(error);
1523 
1524 	svc_reclassify_socket(sock);
1525 
1526 	/*
1527 	 * If this is an PF_INET6 listener, we want to avoid
1528 	 * getting requests from IPv4 remotes.  Those should
1529 	 * be shunted to a PF_INET listener via rpcbind.
1530 	 */
1531 	val = 1;
1532 	if (family == PF_INET6)
1533 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1534 					(char *)&val, sizeof(val));
1535 
1536 	if (type == SOCK_STREAM)
1537 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1538 	error = kernel_bind(sock, sin, len);
1539 	if (error < 0)
1540 		goto bummer;
1541 
1542 	newlen = len;
1543 	error = kernel_getsockname(sock, newsin, &newlen);
1544 	if (error < 0)
1545 		goto bummer;
1546 
1547 	if (protocol == IPPROTO_TCP) {
1548 		if ((error = kernel_listen(sock, 64)) < 0)
1549 			goto bummer;
1550 	}
1551 
1552 	svsk = svc_setup_socket(serv, sock, flags);
1553 	if (IS_ERR(svsk)) {
1554 		error = PTR_ERR(svsk);
1555 		goto bummer;
1556 	}
1557 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1558 	return (struct svc_xprt *)svsk;
1559 bummer:
1560 	dprintk("svc: svc_create_socket error = %d\n", -error);
1561 	sock_release(sock);
1562 	return ERR_PTR(error);
1563 }
1564 
1565 /*
1566  * Detach the svc_sock from the socket so that no
1567  * more callbacks occur.
1568  */
1569 static void svc_sock_detach(struct svc_xprt *xprt)
1570 {
1571 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1572 	struct sock *sk = svsk->sk_sk;
1573 
1574 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1575 
1576 	/* put back the old socket callbacks */
1577 	lock_sock(sk);
1578 	sk->sk_state_change = svsk->sk_ostate;
1579 	sk->sk_data_ready = svsk->sk_odata;
1580 	sk->sk_write_space = svsk->sk_owspace;
1581 	sk->sk_user_data = NULL;
1582 	release_sock(sk);
1583 }
1584 
1585 /*
1586  * Disconnect the socket, and reset the callbacks
1587  */
1588 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1589 {
1590 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1591 
1592 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1593 
1594 	svc_sock_detach(xprt);
1595 
1596 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1597 		svc_tcp_clear_pages(svsk);
1598 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1599 	}
1600 }
1601 
1602 /*
1603  * Free the svc_sock's socket resources and the svc_sock itself.
1604  */
1605 static void svc_sock_free(struct svc_xprt *xprt)
1606 {
1607 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1608 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1609 
1610 	if (svsk->sk_sock->file)
1611 		sockfd_put(svsk->sk_sock);
1612 	else
1613 		sock_release(svsk->sk_sock);
1614 	kfree(svsk);
1615 }
1616 
1617 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1618 /*
1619  * Create a back channel svc_xprt which shares the fore channel socket.
1620  */
1621 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1622 					     int protocol,
1623 					     struct net *net,
1624 					     struct sockaddr *sin, int len,
1625 					     int flags)
1626 {
1627 	struct svc_sock *svsk;
1628 	struct svc_xprt *xprt;
1629 
1630 	if (protocol != IPPROTO_TCP) {
1631 		printk(KERN_WARNING "svc: only TCP sockets"
1632 			" supported on shared back channel\n");
1633 		return ERR_PTR(-EINVAL);
1634 	}
1635 
1636 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1637 	if (!svsk)
1638 		return ERR_PTR(-ENOMEM);
1639 
1640 	xprt = &svsk->sk_xprt;
1641 	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1642 	set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1643 
1644 	serv->sv_bc_xprt = xprt;
1645 
1646 	return xprt;
1647 }
1648 
1649 /*
1650  * Free a back channel svc_sock.
1651  */
1652 static void svc_bc_sock_free(struct svc_xprt *xprt)
1653 {
1654 	if (xprt)
1655 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1656 }
1657 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1658