xref: /linux/net/sunrpc/svcsock.c (revision 08ec212c0f92cbf30e3ecc7349f18151714041d6)
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/tcp.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <trace/events/skb.h>
47 
48 #include <linux/sunrpc/types.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/xdr.h>
51 #include <linux/sunrpc/msg_prot.h>
52 #include <linux/sunrpc/svcsock.h>
53 #include <linux/sunrpc/stats.h>
54 #include <linux/sunrpc/xprt.h>
55 
56 #include "sunrpc.h"
57 
58 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
59 
60 
61 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
62 					 int flags);
63 static void		svc_udp_data_ready(struct sock *, int);
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 	BUG_ON(sock_owned_by_user(sk));
88 	switch (sk->sk_family) {
89 	case AF_INET:
90 		sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
91 					      &svc_slock_key[0],
92 					      "sk_xprt.xpt_lock-AF_INET-NFSD",
93 					      &svc_key[0]);
94 		break;
95 
96 	case AF_INET6:
97 		sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
98 					      &svc_slock_key[1],
99 					      "sk_xprt.xpt_lock-AF_INET6-NFSD",
100 					      &svc_key[1]);
101 		break;
102 
103 	default:
104 		BUG();
105 	}
106 }
107 #else
108 static void svc_reclassify_socket(struct socket *sock)
109 {
110 }
111 #endif
112 
113 /*
114  * Release an skbuff after use
115  */
116 static void svc_release_skb(struct svc_rqst *rqstp)
117 {
118 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
119 
120 	if (skb) {
121 		struct svc_sock *svsk =
122 			container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
123 		rqstp->rq_xprt_ctxt = NULL;
124 
125 		dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
126 		skb_free_datagram_locked(svsk->sk_sk, skb);
127 	}
128 }
129 
130 union svc_pktinfo_u {
131 	struct in_pktinfo pkti;
132 	struct in6_pktinfo pkti6;
133 };
134 #define SVC_PKTINFO_SPACE \
135 	CMSG_SPACE(sizeof(union svc_pktinfo_u))
136 
137 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
138 {
139 	struct svc_sock *svsk =
140 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
141 	switch (svsk->sk_sk->sk_family) {
142 	case AF_INET: {
143 			struct in_pktinfo *pki = CMSG_DATA(cmh);
144 
145 			cmh->cmsg_level = SOL_IP;
146 			cmh->cmsg_type = IP_PKTINFO;
147 			pki->ipi_ifindex = 0;
148 			pki->ipi_spec_dst.s_addr =
149 				 svc_daddr_in(rqstp)->sin_addr.s_addr;
150 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
151 		}
152 		break;
153 
154 	case AF_INET6: {
155 			struct in6_pktinfo *pki = CMSG_DATA(cmh);
156 			struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
157 
158 			cmh->cmsg_level = SOL_IPV6;
159 			cmh->cmsg_type = IPV6_PKTINFO;
160 			pki->ipi6_ifindex = daddr->sin6_scope_id;
161 			pki->ipi6_addr = daddr->sin6_addr;
162 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
163 		}
164 		break;
165 	}
166 }
167 
168 /*
169  * send routine intended to be shared by the fore- and back-channel
170  */
171 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
172 		    struct page *headpage, unsigned long headoffset,
173 		    struct page *tailpage, unsigned long tailoffset)
174 {
175 	int		result;
176 	int		size;
177 	struct page	**ppage = xdr->pages;
178 	size_t		base = xdr->page_base;
179 	unsigned int	pglen = xdr->page_len;
180 	unsigned int	flags = MSG_MORE;
181 	int		slen;
182 	int		len = 0;
183 
184 	slen = xdr->len;
185 
186 	/* send head */
187 	if (slen == xdr->head[0].iov_len)
188 		flags = 0;
189 	len = kernel_sendpage(sock, headpage, headoffset,
190 				  xdr->head[0].iov_len, flags);
191 	if (len != xdr->head[0].iov_len)
192 		goto out;
193 	slen -= xdr->head[0].iov_len;
194 	if (slen == 0)
195 		goto out;
196 
197 	/* send page data */
198 	size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
199 	while (pglen > 0) {
200 		if (slen == size)
201 			flags = 0;
202 		result = kernel_sendpage(sock, *ppage, base, size, flags);
203 		if (result > 0)
204 			len += result;
205 		if (result != size)
206 			goto out;
207 		slen -= size;
208 		pglen -= size;
209 		size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
210 		base = 0;
211 		ppage++;
212 	}
213 
214 	/* send tail */
215 	if (xdr->tail[0].iov_len) {
216 		result = kernel_sendpage(sock, tailpage, tailoffset,
217 				   xdr->tail[0].iov_len, 0);
218 		if (result > 0)
219 			len += result;
220 	}
221 
222 out:
223 	return len;
224 }
225 
226 
227 /*
228  * Generic sendto routine
229  */
230 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
231 {
232 	struct svc_sock	*svsk =
233 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
234 	struct socket	*sock = svsk->sk_sock;
235 	union {
236 		struct cmsghdr	hdr;
237 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
238 	} buffer;
239 	struct cmsghdr *cmh = &buffer.hdr;
240 	int		len = 0;
241 	unsigned long tailoff;
242 	unsigned long headoff;
243 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
244 
245 	if (rqstp->rq_prot == IPPROTO_UDP) {
246 		struct msghdr msg = {
247 			.msg_name	= &rqstp->rq_addr,
248 			.msg_namelen	= rqstp->rq_addrlen,
249 			.msg_control	= cmh,
250 			.msg_controllen	= sizeof(buffer),
251 			.msg_flags	= MSG_MORE,
252 		};
253 
254 		svc_set_cmsg_data(rqstp, cmh);
255 
256 		if (sock_sendmsg(sock, &msg, 0) < 0)
257 			goto out;
258 	}
259 
260 	tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
261 	headoff = 0;
262 	len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
263 			       rqstp->rq_respages[0], tailoff);
264 
265 out:
266 	dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
267 		svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
268 		xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
269 
270 	return len;
271 }
272 
273 /*
274  * Report socket names for nfsdfs
275  */
276 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
277 {
278 	const struct sock *sk = svsk->sk_sk;
279 	const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
280 							"udp" : "tcp";
281 	int len;
282 
283 	switch (sk->sk_family) {
284 	case PF_INET:
285 		len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
286 				proto_name,
287 				&inet_sk(sk)->inet_rcv_saddr,
288 				inet_sk(sk)->inet_num);
289 		break;
290 	case PF_INET6:
291 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
292 				proto_name,
293 				&inet6_sk(sk)->rcv_saddr,
294 				inet_sk(sk)->inet_num);
295 		break;
296 	default:
297 		len = snprintf(buf, remaining, "*unknown-%d*\n",
298 				sk->sk_family);
299 	}
300 
301 	if (len >= remaining) {
302 		*buf = '\0';
303 		return -ENAMETOOLONG;
304 	}
305 	return len;
306 }
307 
308 /*
309  * Check input queue length
310  */
311 static int svc_recv_available(struct svc_sock *svsk)
312 {
313 	struct socket	*sock = svsk->sk_sock;
314 	int		avail, err;
315 
316 	err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
317 
318 	return (err >= 0)? avail : err;
319 }
320 
321 /*
322  * Generic recvfrom routine.
323  */
324 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
325 			int buflen)
326 {
327 	struct svc_sock *svsk =
328 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
329 	struct msghdr msg = {
330 		.msg_flags	= MSG_DONTWAIT,
331 	};
332 	int len;
333 
334 	rqstp->rq_xprt_hlen = 0;
335 
336 	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
337 				msg.msg_flags);
338 
339 	dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
340 		svsk, iov[0].iov_base, iov[0].iov_len, len);
341 	return len;
342 }
343 
344 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
345 				struct kvec *iov, int nr,
346 				int buflen, unsigned int base)
347 {
348 	size_t save_iovlen;
349 	void *save_iovbase;
350 	unsigned int i;
351 	int ret;
352 
353 	if (base == 0)
354 		return svc_recvfrom(rqstp, iov, nr, buflen);
355 
356 	for (i = 0; i < nr; i++) {
357 		if (iov[i].iov_len > base)
358 			break;
359 		base -= iov[i].iov_len;
360 	}
361 	save_iovlen = iov[i].iov_len;
362 	save_iovbase = iov[i].iov_base;
363 	iov[i].iov_len -= base;
364 	iov[i].iov_base += base;
365 	ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
366 	iov[i].iov_len = save_iovlen;
367 	iov[i].iov_base = save_iovbase;
368 	return ret;
369 }
370 
371 /*
372  * Set socket snd and rcv buffer lengths
373  */
374 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
375 				unsigned int rcv)
376 {
377 #if 0
378 	mm_segment_t	oldfs;
379 	oldfs = get_fs(); set_fs(KERNEL_DS);
380 	sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
381 			(char*)&snd, sizeof(snd));
382 	sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
383 			(char*)&rcv, sizeof(rcv));
384 #else
385 	/* sock_setsockopt limits use to sysctl_?mem_max,
386 	 * which isn't acceptable.  Until that is made conditional
387 	 * on not having CAP_SYS_RESOURCE or similar, we go direct...
388 	 * DaveM said I could!
389 	 */
390 	lock_sock(sock->sk);
391 	sock->sk->sk_sndbuf = snd * 2;
392 	sock->sk->sk_rcvbuf = rcv * 2;
393 	sock->sk->sk_write_space(sock->sk);
394 	release_sock(sock->sk);
395 #endif
396 }
397 /*
398  * INET callback when data has been received on the socket.
399  */
400 static void svc_udp_data_ready(struct sock *sk, int count)
401 {
402 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
403 	wait_queue_head_t *wq = sk_sleep(sk);
404 
405 	if (svsk) {
406 		dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
407 			svsk, sk, count,
408 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
409 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
410 		svc_xprt_enqueue(&svsk->sk_xprt);
411 	}
412 	if (wq && waitqueue_active(wq))
413 		wake_up_interruptible(wq);
414 }
415 
416 /*
417  * INET callback when space is newly available on the socket.
418  */
419 static void svc_write_space(struct sock *sk)
420 {
421 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
422 	wait_queue_head_t *wq = sk_sleep(sk);
423 
424 	if (svsk) {
425 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
426 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
427 		svc_xprt_enqueue(&svsk->sk_xprt);
428 	}
429 
430 	if (wq && waitqueue_active(wq)) {
431 		dprintk("RPC svc_write_space: someone sleeping on %p\n",
432 		       svsk);
433 		wake_up_interruptible(wq);
434 	}
435 }
436 
437 static void svc_tcp_write_space(struct sock *sk)
438 {
439 	struct socket *sock = sk->sk_socket;
440 
441 	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
442 		clear_bit(SOCK_NOSPACE, &sock->flags);
443 	svc_write_space(sk);
444 }
445 
446 /*
447  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
448  */
449 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
450 				     struct cmsghdr *cmh)
451 {
452 	struct in_pktinfo *pki = CMSG_DATA(cmh);
453 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
454 
455 	if (cmh->cmsg_type != IP_PKTINFO)
456 		return 0;
457 
458 	daddr->sin_family = AF_INET;
459 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
460 	return 1;
461 }
462 
463 /*
464  * See net/ipv6/datagram.c : datagram_recv_ctl
465  */
466 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
467 				     struct cmsghdr *cmh)
468 {
469 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
470 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
471 
472 	if (cmh->cmsg_type != IPV6_PKTINFO)
473 		return 0;
474 
475 	daddr->sin6_family = AF_INET6;
476 	daddr->sin6_addr = pki->ipi6_addr;
477 	daddr->sin6_scope_id = pki->ipi6_ifindex;
478 	return 1;
479 }
480 
481 /*
482  * Copy the UDP datagram's destination address to the rqstp structure.
483  * The 'destination' address in this case is the address to which the
484  * peer sent the datagram, i.e. our local address. For multihomed
485  * hosts, this can change from msg to msg. Note that only the IP
486  * address changes, the port number should remain the same.
487  */
488 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
489 				    struct cmsghdr *cmh)
490 {
491 	switch (cmh->cmsg_level) {
492 	case SOL_IP:
493 		return svc_udp_get_dest_address4(rqstp, cmh);
494 	case SOL_IPV6:
495 		return svc_udp_get_dest_address6(rqstp, cmh);
496 	}
497 
498 	return 0;
499 }
500 
501 /*
502  * Receive a datagram from a UDP socket.
503  */
504 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
505 {
506 	struct svc_sock	*svsk =
507 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
508 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
509 	struct sk_buff	*skb;
510 	union {
511 		struct cmsghdr	hdr;
512 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
513 	} buffer;
514 	struct cmsghdr *cmh = &buffer.hdr;
515 	struct msghdr msg = {
516 		.msg_name = svc_addr(rqstp),
517 		.msg_control = cmh,
518 		.msg_controllen = sizeof(buffer),
519 		.msg_flags = MSG_DONTWAIT,
520 	};
521 	size_t len;
522 	int err;
523 
524 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
525 	    /* udp sockets need large rcvbuf as all pending
526 	     * requests are still in that buffer.  sndbuf must
527 	     * also be large enough that there is enough space
528 	     * for one reply per thread.  We count all threads
529 	     * rather than threads in a particular pool, which
530 	     * provides an upper bound on the number of threads
531 	     * which will access the socket.
532 	     */
533 	    svc_sock_setbufsize(svsk->sk_sock,
534 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
535 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
536 
537 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
538 	skb = NULL;
539 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
540 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
541 	if (err >= 0)
542 		skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
543 
544 	if (skb == NULL) {
545 		if (err != -EAGAIN) {
546 			/* possibly an icmp error */
547 			dprintk("svc: recvfrom returned error %d\n", -err);
548 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
549 		}
550 		return 0;
551 	}
552 	len = svc_addr_len(svc_addr(rqstp));
553 	rqstp->rq_addrlen = len;
554 	if (skb->tstamp.tv64 == 0) {
555 		skb->tstamp = ktime_get_real();
556 		/* Don't enable netstamp, sunrpc doesn't
557 		   need that much accuracy */
558 	}
559 	svsk->sk_sk->sk_stamp = skb->tstamp;
560 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
561 
562 	len  = skb->len - sizeof(struct udphdr);
563 	rqstp->rq_arg.len = len;
564 
565 	rqstp->rq_prot = IPPROTO_UDP;
566 
567 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
568 		net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
569 				     cmh->cmsg_level, cmh->cmsg_type);
570 		goto out_free;
571 	}
572 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
573 
574 	if (skb_is_nonlinear(skb)) {
575 		/* we have to copy */
576 		local_bh_disable();
577 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
578 			local_bh_enable();
579 			/* checksum error */
580 			goto out_free;
581 		}
582 		local_bh_enable();
583 		skb_free_datagram_locked(svsk->sk_sk, skb);
584 	} else {
585 		/* we can use it in-place */
586 		rqstp->rq_arg.head[0].iov_base = skb->data +
587 			sizeof(struct udphdr);
588 		rqstp->rq_arg.head[0].iov_len = len;
589 		if (skb_checksum_complete(skb))
590 			goto out_free;
591 		rqstp->rq_xprt_ctxt = skb;
592 	}
593 
594 	rqstp->rq_arg.page_base = 0;
595 	if (len <= rqstp->rq_arg.head[0].iov_len) {
596 		rqstp->rq_arg.head[0].iov_len = len;
597 		rqstp->rq_arg.page_len = 0;
598 		rqstp->rq_respages = rqstp->rq_pages+1;
599 	} else {
600 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
601 		rqstp->rq_respages = rqstp->rq_pages + 1 +
602 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
603 	}
604 
605 	if (serv->sv_stats)
606 		serv->sv_stats->netudpcnt++;
607 
608 	return len;
609 out_free:
610 	trace_kfree_skb(skb, svc_udp_recvfrom);
611 	skb_free_datagram_locked(svsk->sk_sk, skb);
612 	return 0;
613 }
614 
615 static int
616 svc_udp_sendto(struct svc_rqst *rqstp)
617 {
618 	int		error;
619 
620 	error = svc_sendto(rqstp, &rqstp->rq_res);
621 	if (error == -ECONNREFUSED)
622 		/* ICMP error on earlier request. */
623 		error = svc_sendto(rqstp, &rqstp->rq_res);
624 
625 	return error;
626 }
627 
628 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
629 {
630 }
631 
632 static int svc_udp_has_wspace(struct svc_xprt *xprt)
633 {
634 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
635 	struct svc_serv	*serv = xprt->xpt_server;
636 	unsigned long required;
637 
638 	/*
639 	 * Set the SOCK_NOSPACE flag before checking the available
640 	 * sock space.
641 	 */
642 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
643 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
644 	if (required*2 > sock_wspace(svsk->sk_sk))
645 		return 0;
646 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
647 	return 1;
648 }
649 
650 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
651 {
652 	BUG();
653 	return NULL;
654 }
655 
656 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
657 				       struct net *net,
658 				       struct sockaddr *sa, int salen,
659 				       int flags)
660 {
661 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
662 }
663 
664 static struct svc_xprt_ops svc_udp_ops = {
665 	.xpo_create = svc_udp_create,
666 	.xpo_recvfrom = svc_udp_recvfrom,
667 	.xpo_sendto = svc_udp_sendto,
668 	.xpo_release_rqst = svc_release_skb,
669 	.xpo_detach = svc_sock_detach,
670 	.xpo_free = svc_sock_free,
671 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
672 	.xpo_has_wspace = svc_udp_has_wspace,
673 	.xpo_accept = svc_udp_accept,
674 };
675 
676 static struct svc_xprt_class svc_udp_class = {
677 	.xcl_name = "udp",
678 	.xcl_owner = THIS_MODULE,
679 	.xcl_ops = &svc_udp_ops,
680 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
681 };
682 
683 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
684 {
685 	int err, level, optname, one = 1;
686 
687 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
688 		      &svsk->sk_xprt, serv);
689 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
690 	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
691 	svsk->sk_sk->sk_write_space = svc_write_space;
692 
693 	/* initialise setting must have enough space to
694 	 * receive and respond to one request.
695 	 * svc_udp_recvfrom will re-adjust if necessary
696 	 */
697 	svc_sock_setbufsize(svsk->sk_sock,
698 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
699 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
700 
701 	/* data might have come in before data_ready set up */
702 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
703 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
704 
705 	/* make sure we get destination address info */
706 	switch (svsk->sk_sk->sk_family) {
707 	case AF_INET:
708 		level = SOL_IP;
709 		optname = IP_PKTINFO;
710 		break;
711 	case AF_INET6:
712 		level = SOL_IPV6;
713 		optname = IPV6_RECVPKTINFO;
714 		break;
715 	default:
716 		BUG();
717 	}
718 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
719 					(char *)&one, sizeof(one));
720 	dprintk("svc: kernel_setsockopt returned %d\n", err);
721 }
722 
723 /*
724  * A data_ready event on a listening socket means there's a connection
725  * pending. Do not use state_change as a substitute for it.
726  */
727 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
728 {
729 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
730 	wait_queue_head_t *wq;
731 
732 	dprintk("svc: socket %p TCP (listen) state change %d\n",
733 		sk, sk->sk_state);
734 
735 	/*
736 	 * This callback may called twice when a new connection
737 	 * is established as a child socket inherits everything
738 	 * from a parent LISTEN socket.
739 	 * 1) data_ready method of the parent socket will be called
740 	 *    when one of child sockets become ESTABLISHED.
741 	 * 2) data_ready method of the child socket may be called
742 	 *    when it receives data before the socket is accepted.
743 	 * In case of 2, we should ignore it silently.
744 	 */
745 	if (sk->sk_state == TCP_LISTEN) {
746 		if (svsk) {
747 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
748 			svc_xprt_enqueue(&svsk->sk_xprt);
749 		} else
750 			printk("svc: socket %p: no user data\n", sk);
751 	}
752 
753 	wq = sk_sleep(sk);
754 	if (wq && waitqueue_active(wq))
755 		wake_up_interruptible_all(wq);
756 }
757 
758 /*
759  * A state change on a connected socket means it's dying or dead.
760  */
761 static void svc_tcp_state_change(struct sock *sk)
762 {
763 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
764 	wait_queue_head_t *wq = sk_sleep(sk);
765 
766 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
767 		sk, sk->sk_state, sk->sk_user_data);
768 
769 	if (!svsk)
770 		printk("svc: socket %p: no user data\n", sk);
771 	else {
772 		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
773 		svc_xprt_enqueue(&svsk->sk_xprt);
774 	}
775 	if (wq && waitqueue_active(wq))
776 		wake_up_interruptible_all(wq);
777 }
778 
779 static void svc_tcp_data_ready(struct sock *sk, int count)
780 {
781 	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
782 	wait_queue_head_t *wq = sk_sleep(sk);
783 
784 	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
785 		sk, sk->sk_user_data);
786 	if (svsk) {
787 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
788 		svc_xprt_enqueue(&svsk->sk_xprt);
789 	}
790 	if (wq && waitqueue_active(wq))
791 		wake_up_interruptible(wq);
792 }
793 
794 /*
795  * Accept a TCP connection
796  */
797 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
798 {
799 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
800 	struct sockaddr_storage addr;
801 	struct sockaddr	*sin = (struct sockaddr *) &addr;
802 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
803 	struct socket	*sock = svsk->sk_sock;
804 	struct socket	*newsock;
805 	struct svc_sock	*newsvsk;
806 	int		err, slen;
807 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
808 
809 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
810 	if (!sock)
811 		return NULL;
812 
813 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
814 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
815 	if (err < 0) {
816 		if (err == -ENOMEM)
817 			printk(KERN_WARNING "%s: no more sockets!\n",
818 			       serv->sv_name);
819 		else if (err != -EAGAIN)
820 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
821 					     serv->sv_name, -err);
822 		return NULL;
823 	}
824 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
825 
826 	err = kernel_getpeername(newsock, sin, &slen);
827 	if (err < 0) {
828 		net_warn_ratelimited("%s: peername failed (err %d)!\n",
829 				     serv->sv_name, -err);
830 		goto failed;		/* aborted connection or whatever */
831 	}
832 
833 	/* Ideally, we would want to reject connections from unauthorized
834 	 * hosts here, but when we get encryption, the IP of the host won't
835 	 * tell us anything.  For now just warn about unpriv connections.
836 	 */
837 	if (!svc_port_is_privileged(sin)) {
838 		dprintk(KERN_WARNING
839 			"%s: connect from unprivileged port: %s\n",
840 			serv->sv_name,
841 			__svc_print_addr(sin, buf, sizeof(buf)));
842 	}
843 	dprintk("%s: connect from %s\n", serv->sv_name,
844 		__svc_print_addr(sin, buf, sizeof(buf)));
845 
846 	/* make sure that a write doesn't block forever when
847 	 * low on memory
848 	 */
849 	newsock->sk->sk_sndtimeo = HZ*30;
850 
851 	newsvsk = svc_setup_socket(serv, newsock,
852 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
853 	if (IS_ERR(newsvsk))
854 		goto failed;
855 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
856 	err = kernel_getsockname(newsock, sin, &slen);
857 	if (unlikely(err < 0)) {
858 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
859 		slen = offsetof(struct sockaddr, sa_data);
860 	}
861 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
862 
863 	if (serv->sv_stats)
864 		serv->sv_stats->nettcpconn++;
865 
866 	return &newsvsk->sk_xprt;
867 
868 failed:
869 	sock_release(newsock);
870 	return NULL;
871 }
872 
873 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
874 {
875 	unsigned int i, len, npages;
876 
877 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
878 		return 0;
879 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
880 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
881 	for (i = 0; i < npages; i++) {
882 		if (rqstp->rq_pages[i] != NULL)
883 			put_page(rqstp->rq_pages[i]);
884 		BUG_ON(svsk->sk_pages[i] == NULL);
885 		rqstp->rq_pages[i] = svsk->sk_pages[i];
886 		svsk->sk_pages[i] = NULL;
887 	}
888 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
889 	return len;
890 }
891 
892 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
893 {
894 	unsigned int i, len, npages;
895 
896 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
897 		return;
898 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
899 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
900 	for (i = 0; i < npages; i++) {
901 		svsk->sk_pages[i] = rqstp->rq_pages[i];
902 		rqstp->rq_pages[i] = NULL;
903 	}
904 }
905 
906 static void svc_tcp_clear_pages(struct svc_sock *svsk)
907 {
908 	unsigned int i, len, npages;
909 
910 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
911 		goto out;
912 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
913 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
914 	for (i = 0; i < npages; i++) {
915 		BUG_ON(svsk->sk_pages[i] == NULL);
916 		put_page(svsk->sk_pages[i]);
917 		svsk->sk_pages[i] = NULL;
918 	}
919 out:
920 	svsk->sk_tcplen = 0;
921 }
922 
923 /*
924  * Receive data.
925  * If we haven't gotten the record length yet, get the next four bytes.
926  * Otherwise try to gobble up as much as possible up to the complete
927  * record length.
928  */
929 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
930 {
931 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
932 	unsigned int want;
933 	int len;
934 
935 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
936 
937 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
938 		struct kvec	iov;
939 
940 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
941 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
942 		iov.iov_len  = want;
943 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
944 			goto error;
945 		svsk->sk_tcplen += len;
946 
947 		if (len < want) {
948 			dprintk("svc: short recvfrom while reading record "
949 				"length (%d of %d)\n", len, want);
950 			return -EAGAIN;
951 		}
952 
953 		svsk->sk_reclen = ntohl(svsk->sk_reclen);
954 		if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
955 			/* FIXME: technically, a record can be fragmented,
956 			 *  and non-terminal fragments will not have the top
957 			 *  bit set in the fragment length header.
958 			 *  But apparently no known nfs clients send fragmented
959 			 *  records. */
960 			net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
961 			goto err_delete;
962 		}
963 
964 		svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
965 		dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
966 		if (svsk->sk_reclen > serv->sv_max_mesg) {
967 			net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
968 					       (unsigned long)svsk->sk_reclen);
969 			goto err_delete;
970 		}
971 	}
972 
973 	if (svsk->sk_reclen < 8)
974 		goto err_delete; /* client is nuts. */
975 
976 	len = svsk->sk_reclen;
977 
978 	return len;
979 error:
980 	dprintk("RPC: TCP recv_record got %d\n", len);
981 	return len;
982 err_delete:
983 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
984 	return -EAGAIN;
985 }
986 
987 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
988 {
989 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
990 	struct rpc_rqst *req = NULL;
991 	struct kvec *src, *dst;
992 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
993 	__be32 xid;
994 	__be32 calldir;
995 
996 	xid = *p++;
997 	calldir = *p;
998 
999 	if (bc_xprt)
1000 		req = xprt_lookup_rqst(bc_xprt, xid);
1001 
1002 	if (!req) {
1003 		printk(KERN_NOTICE
1004 			"%s: Got unrecognized reply: "
1005 			"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1006 			__func__, ntohl(calldir),
1007 			bc_xprt, xid);
1008 		return -EAGAIN;
1009 	}
1010 
1011 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1012 	/*
1013 	 * XXX!: cheating for now!  Only copying HEAD.
1014 	 * But we know this is good enough for now (in fact, for any
1015 	 * callback reply in the forseeable future).
1016 	 */
1017 	dst = &req->rq_private_buf.head[0];
1018 	src = &rqstp->rq_arg.head[0];
1019 	if (dst->iov_len < src->iov_len)
1020 		return -EAGAIN; /* whatever; just giving up. */
1021 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1022 	xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1023 	rqstp->rq_arg.len = 0;
1024 	return 0;
1025 }
1026 
1027 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1028 {
1029 	int i = 0;
1030 	int t = 0;
1031 
1032 	while (t < len) {
1033 		vec[i].iov_base = page_address(pages[i]);
1034 		vec[i].iov_len = PAGE_SIZE;
1035 		i++;
1036 		t += PAGE_SIZE;
1037 	}
1038 	return i;
1039 }
1040 
1041 
1042 /*
1043  * Receive data from a TCP socket.
1044  */
1045 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1046 {
1047 	struct svc_sock	*svsk =
1048 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1049 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1050 	int		len;
1051 	struct kvec *vec;
1052 	unsigned int want, base;
1053 	__be32 *p;
1054 	__be32 calldir;
1055 	int pnum;
1056 
1057 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1058 		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1059 		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1060 		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1061 
1062 	len = svc_tcp_recv_record(svsk, rqstp);
1063 	if (len < 0)
1064 		goto error;
1065 
1066 	base = svc_tcp_restore_pages(svsk, rqstp);
1067 	want = svsk->sk_reclen - base;
1068 
1069 	vec = rqstp->rq_vec;
1070 
1071 	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1072 						svsk->sk_reclen);
1073 
1074 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1075 
1076 	/* Now receive data */
1077 	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1078 	if (len >= 0)
1079 		svsk->sk_tcplen += len;
1080 	if (len != want) {
1081 		svc_tcp_save_pages(svsk, rqstp);
1082 		if (len < 0 && len != -EAGAIN)
1083 			goto err_other;
1084 		dprintk("svc: incomplete TCP record (%d of %d)\n",
1085 			svsk->sk_tcplen, svsk->sk_reclen);
1086 		goto err_noclose;
1087 	}
1088 
1089 	rqstp->rq_arg.len = svsk->sk_reclen;
1090 	rqstp->rq_arg.page_base = 0;
1091 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1092 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1093 		rqstp->rq_arg.page_len = 0;
1094 	} else
1095 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1096 
1097 	rqstp->rq_xprt_ctxt   = NULL;
1098 	rqstp->rq_prot	      = IPPROTO_TCP;
1099 
1100 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1101 	calldir = p[1];
1102 	if (calldir)
1103 		len = receive_cb_reply(svsk, rqstp);
1104 
1105 	/* Reset TCP read info */
1106 	svsk->sk_reclen = 0;
1107 	svsk->sk_tcplen = 0;
1108 	/* If we have more data, signal svc_xprt_enqueue() to try again */
1109 	if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1110 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1111 
1112 	if (len < 0)
1113 		goto error;
1114 
1115 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1116 	if (serv->sv_stats)
1117 		serv->sv_stats->nettcpcnt++;
1118 
1119 	dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1120 	return rqstp->rq_arg.len;
1121 
1122 error:
1123 	if (len != -EAGAIN)
1124 		goto err_other;
1125 	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1126 	return 0;
1127 err_other:
1128 	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1129 	       svsk->sk_xprt.xpt_server->sv_name, -len);
1130 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1131 err_noclose:
1132 	return 0;	/* record not complete */
1133 }
1134 
1135 /*
1136  * Send out data on TCP socket.
1137  */
1138 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1139 {
1140 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1141 	int sent;
1142 	__be32 reclen;
1143 
1144 	/* Set up the first element of the reply kvec.
1145 	 * Any other kvecs that may be in use have been taken
1146 	 * care of by the server implementation itself.
1147 	 */
1148 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1149 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1150 
1151 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1152 	if (sent != xbufp->len) {
1153 		printk(KERN_NOTICE
1154 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1155 		       "- shutting down socket\n",
1156 		       rqstp->rq_xprt->xpt_server->sv_name,
1157 		       (sent<0)?"got error":"sent only",
1158 		       sent, xbufp->len);
1159 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1160 		svc_xprt_enqueue(rqstp->rq_xprt);
1161 		sent = -EAGAIN;
1162 	}
1163 	return sent;
1164 }
1165 
1166 /*
1167  * Setup response header. TCP has a 4B record length field.
1168  */
1169 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1170 {
1171 	struct kvec *resv = &rqstp->rq_res.head[0];
1172 
1173 	/* tcp needs a space for the record length... */
1174 	svc_putnl(resv, 0);
1175 }
1176 
1177 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1178 {
1179 	struct svc_sock *svsk =	container_of(xprt, struct svc_sock, sk_xprt);
1180 	struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1181 	int required;
1182 
1183 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1184 		return 1;
1185 	required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1186 	if (sk_stream_wspace(svsk->sk_sk) >= required)
1187 		return 1;
1188 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1189 	return 0;
1190 }
1191 
1192 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1193 				       struct net *net,
1194 				       struct sockaddr *sa, int salen,
1195 				       int flags)
1196 {
1197 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1198 }
1199 
1200 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1201 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1202 					     struct net *, struct sockaddr *,
1203 					     int, int);
1204 static void svc_bc_sock_free(struct svc_xprt *xprt);
1205 
1206 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1207 				       struct net *net,
1208 				       struct sockaddr *sa, int salen,
1209 				       int flags)
1210 {
1211 	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1212 }
1213 
1214 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1215 {
1216 }
1217 
1218 static struct svc_xprt_ops svc_tcp_bc_ops = {
1219 	.xpo_create = svc_bc_tcp_create,
1220 	.xpo_detach = svc_bc_tcp_sock_detach,
1221 	.xpo_free = svc_bc_sock_free,
1222 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1223 };
1224 
1225 static struct svc_xprt_class svc_tcp_bc_class = {
1226 	.xcl_name = "tcp-bc",
1227 	.xcl_owner = THIS_MODULE,
1228 	.xcl_ops = &svc_tcp_bc_ops,
1229 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1230 };
1231 
1232 static void svc_init_bc_xprt_sock(void)
1233 {
1234 	svc_reg_xprt_class(&svc_tcp_bc_class);
1235 }
1236 
1237 static void svc_cleanup_bc_xprt_sock(void)
1238 {
1239 	svc_unreg_xprt_class(&svc_tcp_bc_class);
1240 }
1241 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1242 static void svc_init_bc_xprt_sock(void)
1243 {
1244 }
1245 
1246 static void svc_cleanup_bc_xprt_sock(void)
1247 {
1248 }
1249 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1250 
1251 static struct svc_xprt_ops svc_tcp_ops = {
1252 	.xpo_create = svc_tcp_create,
1253 	.xpo_recvfrom = svc_tcp_recvfrom,
1254 	.xpo_sendto = svc_tcp_sendto,
1255 	.xpo_release_rqst = svc_release_skb,
1256 	.xpo_detach = svc_tcp_sock_detach,
1257 	.xpo_free = svc_sock_free,
1258 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1259 	.xpo_has_wspace = svc_tcp_has_wspace,
1260 	.xpo_accept = svc_tcp_accept,
1261 };
1262 
1263 static struct svc_xprt_class svc_tcp_class = {
1264 	.xcl_name = "tcp",
1265 	.xcl_owner = THIS_MODULE,
1266 	.xcl_ops = &svc_tcp_ops,
1267 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1268 };
1269 
1270 void svc_init_xprt_sock(void)
1271 {
1272 	svc_reg_xprt_class(&svc_tcp_class);
1273 	svc_reg_xprt_class(&svc_udp_class);
1274 	svc_init_bc_xprt_sock();
1275 }
1276 
1277 void svc_cleanup_xprt_sock(void)
1278 {
1279 	svc_unreg_xprt_class(&svc_tcp_class);
1280 	svc_unreg_xprt_class(&svc_udp_class);
1281 	svc_cleanup_bc_xprt_sock();
1282 }
1283 
1284 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1285 {
1286 	struct sock	*sk = svsk->sk_sk;
1287 
1288 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1289 		      &svsk->sk_xprt, serv);
1290 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1291 	if (sk->sk_state == TCP_LISTEN) {
1292 		dprintk("setting up TCP socket for listening\n");
1293 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1294 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1295 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1296 	} else {
1297 		dprintk("setting up TCP socket for reading\n");
1298 		sk->sk_state_change = svc_tcp_state_change;
1299 		sk->sk_data_ready = svc_tcp_data_ready;
1300 		sk->sk_write_space = svc_tcp_write_space;
1301 
1302 		svsk->sk_reclen = 0;
1303 		svsk->sk_tcplen = 0;
1304 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1305 
1306 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1307 
1308 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1309 		if (sk->sk_state != TCP_ESTABLISHED)
1310 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1311 	}
1312 }
1313 
1314 void svc_sock_update_bufs(struct svc_serv *serv)
1315 {
1316 	/*
1317 	 * The number of server threads has changed. Update
1318 	 * rcvbuf and sndbuf accordingly on all sockets
1319 	 */
1320 	struct svc_sock *svsk;
1321 
1322 	spin_lock_bh(&serv->sv_lock);
1323 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1324 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1325 	spin_unlock_bh(&serv->sv_lock);
1326 }
1327 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1328 
1329 /*
1330  * Initialize socket for RPC use and create svc_sock struct
1331  * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1332  */
1333 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1334 						struct socket *sock,
1335 						int flags)
1336 {
1337 	struct svc_sock	*svsk;
1338 	struct sock	*inet;
1339 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1340 	int		err = 0;
1341 
1342 	dprintk("svc: svc_setup_socket %p\n", sock);
1343 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1344 	if (!svsk)
1345 		return ERR_PTR(-ENOMEM);
1346 
1347 	inet = sock->sk;
1348 
1349 	/* Register socket with portmapper */
1350 	if (pmap_register)
1351 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1352 				     inet->sk_protocol,
1353 				     ntohs(inet_sk(inet)->inet_sport));
1354 
1355 	if (err < 0) {
1356 		kfree(svsk);
1357 		return ERR_PTR(err);
1358 	}
1359 
1360 	inet->sk_user_data = svsk;
1361 	svsk->sk_sock = sock;
1362 	svsk->sk_sk = inet;
1363 	svsk->sk_ostate = inet->sk_state_change;
1364 	svsk->sk_odata = inet->sk_data_ready;
1365 	svsk->sk_owspace = inet->sk_write_space;
1366 
1367 	/* Initialize the socket */
1368 	if (sock->type == SOCK_DGRAM)
1369 		svc_udp_init(svsk, serv);
1370 	else {
1371 		/* initialise setting must have enough space to
1372 		 * receive and respond to one request.
1373 		 */
1374 		svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1375 					4 * serv->sv_max_mesg);
1376 		svc_tcp_init(svsk, serv);
1377 	}
1378 
1379 	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1380 				svsk, svsk->sk_sk);
1381 
1382 	return svsk;
1383 }
1384 
1385 /**
1386  * svc_addsock - add a listener socket to an RPC service
1387  * @serv: pointer to RPC service to which to add a new listener
1388  * @fd: file descriptor of the new listener
1389  * @name_return: pointer to buffer to fill in with name of listener
1390  * @len: size of the buffer
1391  *
1392  * Fills in socket name and returns positive length of name if successful.
1393  * Name is terminated with '\n'.  On error, returns a negative errno
1394  * value.
1395  */
1396 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1397 		const size_t len)
1398 {
1399 	int err = 0;
1400 	struct socket *so = sockfd_lookup(fd, &err);
1401 	struct svc_sock *svsk = NULL;
1402 	struct sockaddr_storage addr;
1403 	struct sockaddr *sin = (struct sockaddr *)&addr;
1404 	int salen;
1405 
1406 	if (!so)
1407 		return err;
1408 	err = -EAFNOSUPPORT;
1409 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1410 		goto out;
1411 	err =  -EPROTONOSUPPORT;
1412 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1413 	    so->sk->sk_protocol != IPPROTO_UDP)
1414 		goto out;
1415 	err = -EISCONN;
1416 	if (so->state > SS_UNCONNECTED)
1417 		goto out;
1418 	err = -ENOENT;
1419 	if (!try_module_get(THIS_MODULE))
1420 		goto out;
1421 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1422 	if (IS_ERR(svsk)) {
1423 		module_put(THIS_MODULE);
1424 		err = PTR_ERR(svsk);
1425 		goto out;
1426 	}
1427 	if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1428 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1429 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1430 	return svc_one_sock_name(svsk, name_return, len);
1431 out:
1432 	sockfd_put(so);
1433 	return err;
1434 }
1435 EXPORT_SYMBOL_GPL(svc_addsock);
1436 
1437 /*
1438  * Create socket for RPC service.
1439  */
1440 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1441 					  int protocol,
1442 					  struct net *net,
1443 					  struct sockaddr *sin, int len,
1444 					  int flags)
1445 {
1446 	struct svc_sock	*svsk;
1447 	struct socket	*sock;
1448 	int		error;
1449 	int		type;
1450 	struct sockaddr_storage addr;
1451 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1452 	int		newlen;
1453 	int		family;
1454 	int		val;
1455 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1456 
1457 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1458 			serv->sv_program->pg_name, protocol,
1459 			__svc_print_addr(sin, buf, sizeof(buf)));
1460 
1461 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1462 		printk(KERN_WARNING "svc: only UDP and TCP "
1463 				"sockets supported\n");
1464 		return ERR_PTR(-EINVAL);
1465 	}
1466 
1467 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1468 	switch (sin->sa_family) {
1469 	case AF_INET6:
1470 		family = PF_INET6;
1471 		break;
1472 	case AF_INET:
1473 		family = PF_INET;
1474 		break;
1475 	default:
1476 		return ERR_PTR(-EINVAL);
1477 	}
1478 
1479 	error = __sock_create(net, family, type, protocol, &sock, 1);
1480 	if (error < 0)
1481 		return ERR_PTR(error);
1482 
1483 	svc_reclassify_socket(sock);
1484 
1485 	/*
1486 	 * If this is an PF_INET6 listener, we want to avoid
1487 	 * getting requests from IPv4 remotes.  Those should
1488 	 * be shunted to a PF_INET listener via rpcbind.
1489 	 */
1490 	val = 1;
1491 	if (family == PF_INET6)
1492 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1493 					(char *)&val, sizeof(val));
1494 
1495 	if (type == SOCK_STREAM)
1496 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1497 	error = kernel_bind(sock, sin, len);
1498 	if (error < 0)
1499 		goto bummer;
1500 
1501 	newlen = len;
1502 	error = kernel_getsockname(sock, newsin, &newlen);
1503 	if (error < 0)
1504 		goto bummer;
1505 
1506 	if (protocol == IPPROTO_TCP) {
1507 		if ((error = kernel_listen(sock, 64)) < 0)
1508 			goto bummer;
1509 	}
1510 
1511 	svsk = svc_setup_socket(serv, sock, flags);
1512 	if (IS_ERR(svsk)) {
1513 		error = PTR_ERR(svsk);
1514 		goto bummer;
1515 	}
1516 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1517 	return (struct svc_xprt *)svsk;
1518 bummer:
1519 	dprintk("svc: svc_create_socket error = %d\n", -error);
1520 	sock_release(sock);
1521 	return ERR_PTR(error);
1522 }
1523 
1524 /*
1525  * Detach the svc_sock from the socket so that no
1526  * more callbacks occur.
1527  */
1528 static void svc_sock_detach(struct svc_xprt *xprt)
1529 {
1530 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1531 	struct sock *sk = svsk->sk_sk;
1532 	wait_queue_head_t *wq;
1533 
1534 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1535 
1536 	/* put back the old socket callbacks */
1537 	sk->sk_state_change = svsk->sk_ostate;
1538 	sk->sk_data_ready = svsk->sk_odata;
1539 	sk->sk_write_space = svsk->sk_owspace;
1540 
1541 	wq = sk_sleep(sk);
1542 	if (wq && waitqueue_active(wq))
1543 		wake_up_interruptible(wq);
1544 }
1545 
1546 /*
1547  * Disconnect the socket, and reset the callbacks
1548  */
1549 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1550 {
1551 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1552 
1553 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1554 
1555 	svc_sock_detach(xprt);
1556 
1557 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1558 		svc_tcp_clear_pages(svsk);
1559 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1560 	}
1561 }
1562 
1563 /*
1564  * Free the svc_sock's socket resources and the svc_sock itself.
1565  */
1566 static void svc_sock_free(struct svc_xprt *xprt)
1567 {
1568 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1569 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1570 
1571 	if (svsk->sk_sock->file)
1572 		sockfd_put(svsk->sk_sock);
1573 	else
1574 		sock_release(svsk->sk_sock);
1575 	kfree(svsk);
1576 }
1577 
1578 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1579 /*
1580  * Create a back channel svc_xprt which shares the fore channel socket.
1581  */
1582 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1583 					     int protocol,
1584 					     struct net *net,
1585 					     struct sockaddr *sin, int len,
1586 					     int flags)
1587 {
1588 	struct svc_sock *svsk;
1589 	struct svc_xprt *xprt;
1590 
1591 	if (protocol != IPPROTO_TCP) {
1592 		printk(KERN_WARNING "svc: only TCP sockets"
1593 			" supported on shared back channel\n");
1594 		return ERR_PTR(-EINVAL);
1595 	}
1596 
1597 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1598 	if (!svsk)
1599 		return ERR_PTR(-ENOMEM);
1600 
1601 	xprt = &svsk->sk_xprt;
1602 	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1603 
1604 	serv->sv_bc_xprt = xprt;
1605 
1606 	return xprt;
1607 }
1608 
1609 /*
1610  * Free a back channel svc_sock.
1611  */
1612 static void svc_bc_sock_free(struct svc_xprt *xprt)
1613 {
1614 	if (xprt)
1615 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1616 }
1617 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1618