xref: /linux/net/sunrpc/svcsock.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 
88 	WARN_ON_ONCE(sock_owned_by_user(sk));
89 	if (sock_owned_by_user(sk))
90 		return;
91 
92 	switch (sk->sk_family) {
93 	case AF_INET:
94 		sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
95 					      &svc_slock_key[0],
96 					      "sk_xprt.xpt_lock-AF_INET-NFSD",
97 					      &svc_key[0]);
98 		break;
99 
100 	case AF_INET6:
101 		sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
102 					      &svc_slock_key[1],
103 					      "sk_xprt.xpt_lock-AF_INET6-NFSD",
104 					      &svc_key[1]);
105 		break;
106 
107 	default:
108 		BUG();
109 	}
110 }
111 #else
112 static void svc_reclassify_socket(struct socket *sock)
113 {
114 }
115 #endif
116 
117 /*
118  * Release an skbuff after use
119  */
120 static void svc_release_skb(struct svc_rqst *rqstp)
121 {
122 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
123 
124 	if (skb) {
125 		struct svc_sock *svsk =
126 			container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
127 		rqstp->rq_xprt_ctxt = NULL;
128 
129 		dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
130 		skb_free_datagram_locked(svsk->sk_sk, skb);
131 	}
132 }
133 
134 union svc_pktinfo_u {
135 	struct in_pktinfo pkti;
136 	struct in6_pktinfo pkti6;
137 };
138 #define SVC_PKTINFO_SPACE \
139 	CMSG_SPACE(sizeof(union svc_pktinfo_u))
140 
141 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
142 {
143 	struct svc_sock *svsk =
144 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 	switch (svsk->sk_sk->sk_family) {
146 	case AF_INET: {
147 			struct in_pktinfo *pki = CMSG_DATA(cmh);
148 
149 			cmh->cmsg_level = SOL_IP;
150 			cmh->cmsg_type = IP_PKTINFO;
151 			pki->ipi_ifindex = 0;
152 			pki->ipi_spec_dst.s_addr =
153 				 svc_daddr_in(rqstp)->sin_addr.s_addr;
154 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
155 		}
156 		break;
157 
158 	case AF_INET6: {
159 			struct in6_pktinfo *pki = CMSG_DATA(cmh);
160 			struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
161 
162 			cmh->cmsg_level = SOL_IPV6;
163 			cmh->cmsg_type = IPV6_PKTINFO;
164 			pki->ipi6_ifindex = daddr->sin6_scope_id;
165 			pki->ipi6_addr = daddr->sin6_addr;
166 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 		}
168 		break;
169 	}
170 }
171 
172 /*
173  * send routine intended to be shared by the fore- and back-channel
174  */
175 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 		    struct page *headpage, unsigned long headoffset,
177 		    struct page *tailpage, unsigned long tailoffset)
178 {
179 	int		result;
180 	int		size;
181 	struct page	**ppage = xdr->pages;
182 	size_t		base = xdr->page_base;
183 	unsigned int	pglen = xdr->page_len;
184 	unsigned int	flags = MSG_MORE;
185 	int		slen;
186 	int		len = 0;
187 
188 	slen = xdr->len;
189 
190 	/* send head */
191 	if (slen == xdr->head[0].iov_len)
192 		flags = 0;
193 	len = kernel_sendpage(sock, headpage, headoffset,
194 				  xdr->head[0].iov_len, flags);
195 	if (len != xdr->head[0].iov_len)
196 		goto out;
197 	slen -= xdr->head[0].iov_len;
198 	if (slen == 0)
199 		goto out;
200 
201 	/* send page data */
202 	size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
203 	while (pglen > 0) {
204 		if (slen == size)
205 			flags = 0;
206 		result = kernel_sendpage(sock, *ppage, base, size, flags);
207 		if (result > 0)
208 			len += result;
209 		if (result != size)
210 			goto out;
211 		slen -= size;
212 		pglen -= size;
213 		size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 		base = 0;
215 		ppage++;
216 	}
217 
218 	/* send tail */
219 	if (xdr->tail[0].iov_len) {
220 		result = kernel_sendpage(sock, tailpage, tailoffset,
221 				   xdr->tail[0].iov_len, 0);
222 		if (result > 0)
223 			len += result;
224 	}
225 
226 out:
227 	return len;
228 }
229 
230 
231 /*
232  * Generic sendto routine
233  */
234 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
235 {
236 	struct svc_sock	*svsk =
237 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 	struct socket	*sock = svsk->sk_sock;
239 	union {
240 		struct cmsghdr	hdr;
241 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
242 	} buffer;
243 	struct cmsghdr *cmh = &buffer.hdr;
244 	int		len = 0;
245 	unsigned long tailoff;
246 	unsigned long headoff;
247 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
248 
249 	if (rqstp->rq_prot == IPPROTO_UDP) {
250 		struct msghdr msg = {
251 			.msg_name	= &rqstp->rq_addr,
252 			.msg_namelen	= rqstp->rq_addrlen,
253 			.msg_control	= cmh,
254 			.msg_controllen	= sizeof(buffer),
255 			.msg_flags	= MSG_MORE,
256 		};
257 
258 		svc_set_cmsg_data(rqstp, cmh);
259 
260 		if (sock_sendmsg(sock, &msg, 0) < 0)
261 			goto out;
262 	}
263 
264 	tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
265 	headoff = 0;
266 	len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 			       rqstp->rq_respages[0], tailoff);
268 
269 out:
270 	dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
271 		svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
272 		xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
273 
274 	return len;
275 }
276 
277 /*
278  * Report socket names for nfsdfs
279  */
280 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
281 {
282 	const struct sock *sk = svsk->sk_sk;
283 	const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
284 							"udp" : "tcp";
285 	int len;
286 
287 	switch (sk->sk_family) {
288 	case PF_INET:
289 		len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
290 				proto_name,
291 				&inet_sk(sk)->inet_rcv_saddr,
292 				inet_sk(sk)->inet_num);
293 		break;
294 	case PF_INET6:
295 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
296 				proto_name,
297 				&inet6_sk(sk)->rcv_saddr,
298 				inet_sk(sk)->inet_num);
299 		break;
300 	default:
301 		len = snprintf(buf, remaining, "*unknown-%d*\n",
302 				sk->sk_family);
303 	}
304 
305 	if (len >= remaining) {
306 		*buf = '\0';
307 		return -ENAMETOOLONG;
308 	}
309 	return len;
310 }
311 
312 /*
313  * Check input queue length
314  */
315 static int svc_recv_available(struct svc_sock *svsk)
316 {
317 	struct socket	*sock = svsk->sk_sock;
318 	int		avail, err;
319 
320 	err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
321 
322 	return (err >= 0)? avail : err;
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 	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
341 				msg.msg_flags);
342 
343 	dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
344 		svsk, iov[0].iov_base, iov[0].iov_len, len);
345 	return len;
346 }
347 
348 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
349 				struct kvec *iov, int nr,
350 				int buflen, unsigned int base)
351 {
352 	size_t save_iovlen;
353 	void *save_iovbase;
354 	unsigned int i;
355 	int ret;
356 
357 	if (base == 0)
358 		return svc_recvfrom(rqstp, iov, nr, buflen);
359 
360 	for (i = 0; i < nr; i++) {
361 		if (iov[i].iov_len > base)
362 			break;
363 		base -= iov[i].iov_len;
364 	}
365 	save_iovlen = iov[i].iov_len;
366 	save_iovbase = iov[i].iov_base;
367 	iov[i].iov_len -= base;
368 	iov[i].iov_base += base;
369 	ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
370 	iov[i].iov_len = save_iovlen;
371 	iov[i].iov_base = save_iovbase;
372 	return ret;
373 }
374 
375 /*
376  * Set socket snd and rcv buffer lengths
377  */
378 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
379 				unsigned int rcv)
380 {
381 #if 0
382 	mm_segment_t	oldfs;
383 	oldfs = get_fs(); set_fs(KERNEL_DS);
384 	sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
385 			(char*)&snd, sizeof(snd));
386 	sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
387 			(char*)&rcv, sizeof(rcv));
388 #else
389 	/* sock_setsockopt limits use to sysctl_?mem_max,
390 	 * which isn't acceptable.  Until that is made conditional
391 	 * on not having CAP_SYS_RESOURCE or similar, we go direct...
392 	 * DaveM said I could!
393 	 */
394 	lock_sock(sock->sk);
395 	sock->sk->sk_sndbuf = snd * 2;
396 	sock->sk->sk_rcvbuf = rcv * 2;
397 	sock->sk->sk_write_space(sock->sk);
398 	release_sock(sock->sk);
399 #endif
400 }
401 /*
402  * INET callback when data has been received on the socket.
403  */
404 static void svc_udp_data_ready(struct sock *sk, int count)
405 {
406 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
407 	wait_queue_head_t *wq = sk_sleep(sk);
408 
409 	if (svsk) {
410 		dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
411 			svsk, sk, count,
412 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
413 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
414 		svc_xprt_enqueue(&svsk->sk_xprt);
415 	}
416 	if (wq && waitqueue_active(wq))
417 		wake_up_interruptible(wq);
418 }
419 
420 /*
421  * INET callback when space is newly available on the socket.
422  */
423 static void svc_write_space(struct sock *sk)
424 {
425 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
426 	wait_queue_head_t *wq = sk_sleep(sk);
427 
428 	if (svsk) {
429 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
430 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
431 		svc_xprt_enqueue(&svsk->sk_xprt);
432 	}
433 
434 	if (wq && waitqueue_active(wq)) {
435 		dprintk("RPC svc_write_space: someone sleeping on %p\n",
436 		       svsk);
437 		wake_up_interruptible(wq);
438 	}
439 }
440 
441 static void svc_tcp_write_space(struct sock *sk)
442 {
443 	struct socket *sock = sk->sk_socket;
444 
445 	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
446 		clear_bit(SOCK_NOSPACE, &sock->flags);
447 	svc_write_space(sk);
448 }
449 
450 /*
451  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
452  */
453 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
454 				     struct cmsghdr *cmh)
455 {
456 	struct in_pktinfo *pki = CMSG_DATA(cmh);
457 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
458 
459 	if (cmh->cmsg_type != IP_PKTINFO)
460 		return 0;
461 
462 	daddr->sin_family = AF_INET;
463 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
464 	return 1;
465 }
466 
467 /*
468  * See net/ipv6/datagram.c : datagram_recv_ctl
469  */
470 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
471 				     struct cmsghdr *cmh)
472 {
473 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
474 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
475 
476 	if (cmh->cmsg_type != IPV6_PKTINFO)
477 		return 0;
478 
479 	daddr->sin6_family = AF_INET6;
480 	daddr->sin6_addr = pki->ipi6_addr;
481 	daddr->sin6_scope_id = pki->ipi6_ifindex;
482 	return 1;
483 }
484 
485 /*
486  * Copy the UDP datagram's destination address to the rqstp structure.
487  * The 'destination' address in this case is the address to which the
488  * peer sent the datagram, i.e. our local address. For multihomed
489  * hosts, this can change from msg to msg. Note that only the IP
490  * address changes, the port number should remain the same.
491  */
492 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
493 				    struct cmsghdr *cmh)
494 {
495 	switch (cmh->cmsg_level) {
496 	case SOL_IP:
497 		return svc_udp_get_dest_address4(rqstp, cmh);
498 	case SOL_IPV6:
499 		return svc_udp_get_dest_address6(rqstp, cmh);
500 	}
501 
502 	return 0;
503 }
504 
505 /*
506  * Receive a datagram from a UDP socket.
507  */
508 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
509 {
510 	struct svc_sock	*svsk =
511 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
512 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
513 	struct sk_buff	*skb;
514 	union {
515 		struct cmsghdr	hdr;
516 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
517 	} buffer;
518 	struct cmsghdr *cmh = &buffer.hdr;
519 	struct msghdr msg = {
520 		.msg_name = svc_addr(rqstp),
521 		.msg_control = cmh,
522 		.msg_controllen = sizeof(buffer),
523 		.msg_flags = MSG_DONTWAIT,
524 	};
525 	size_t len;
526 	int err;
527 
528 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
529 	    /* udp sockets need large rcvbuf as all pending
530 	     * requests are still in that buffer.  sndbuf must
531 	     * also be large enough that there is enough space
532 	     * for one reply per thread.  We count all threads
533 	     * rather than threads in a particular pool, which
534 	     * provides an upper bound on the number of threads
535 	     * which will access the socket.
536 	     */
537 	    svc_sock_setbufsize(svsk->sk_sock,
538 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
539 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
540 
541 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
542 	skb = NULL;
543 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
544 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
545 	if (err >= 0)
546 		skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
547 
548 	if (skb == NULL) {
549 		if (err != -EAGAIN) {
550 			/* possibly an icmp error */
551 			dprintk("svc: recvfrom returned error %d\n", -err);
552 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
553 		}
554 		return 0;
555 	}
556 	len = svc_addr_len(svc_addr(rqstp));
557 	rqstp->rq_addrlen = len;
558 	if (skb->tstamp.tv64 == 0) {
559 		skb->tstamp = ktime_get_real();
560 		/* Don't enable netstamp, sunrpc doesn't
561 		   need that much accuracy */
562 	}
563 	svsk->sk_sk->sk_stamp = skb->tstamp;
564 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
565 
566 	len  = skb->len - sizeof(struct udphdr);
567 	rqstp->rq_arg.len = len;
568 
569 	rqstp->rq_prot = IPPROTO_UDP;
570 
571 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
572 		net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
573 				     cmh->cmsg_level, cmh->cmsg_type);
574 		goto out_free;
575 	}
576 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
577 
578 	if (skb_is_nonlinear(skb)) {
579 		/* we have to copy */
580 		local_bh_disable();
581 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
582 			local_bh_enable();
583 			/* checksum error */
584 			goto out_free;
585 		}
586 		local_bh_enable();
587 		skb_free_datagram_locked(svsk->sk_sk, skb);
588 	} else {
589 		/* we can use it in-place */
590 		rqstp->rq_arg.head[0].iov_base = skb->data +
591 			sizeof(struct udphdr);
592 		rqstp->rq_arg.head[0].iov_len = len;
593 		if (skb_checksum_complete(skb))
594 			goto out_free;
595 		rqstp->rq_xprt_ctxt = skb;
596 	}
597 
598 	rqstp->rq_arg.page_base = 0;
599 	if (len <= rqstp->rq_arg.head[0].iov_len) {
600 		rqstp->rq_arg.head[0].iov_len = len;
601 		rqstp->rq_arg.page_len = 0;
602 		rqstp->rq_respages = rqstp->rq_pages+1;
603 	} else {
604 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
605 		rqstp->rq_respages = rqstp->rq_pages + 1 +
606 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
607 	}
608 
609 	if (serv->sv_stats)
610 		serv->sv_stats->netudpcnt++;
611 
612 	return len;
613 out_free:
614 	trace_kfree_skb(skb, svc_udp_recvfrom);
615 	skb_free_datagram_locked(svsk->sk_sk, skb);
616 	return 0;
617 }
618 
619 static int
620 svc_udp_sendto(struct svc_rqst *rqstp)
621 {
622 	int		error;
623 
624 	error = svc_sendto(rqstp, &rqstp->rq_res);
625 	if (error == -ECONNREFUSED)
626 		/* ICMP error on earlier request. */
627 		error = svc_sendto(rqstp, &rqstp->rq_res);
628 
629 	return error;
630 }
631 
632 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
633 {
634 }
635 
636 static int svc_udp_has_wspace(struct svc_xprt *xprt)
637 {
638 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
639 	struct svc_serv	*serv = xprt->xpt_server;
640 	unsigned long required;
641 
642 	/*
643 	 * Set the SOCK_NOSPACE flag before checking the available
644 	 * sock space.
645 	 */
646 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
647 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
648 	if (required*2 > sock_wspace(svsk->sk_sk))
649 		return 0;
650 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
651 	return 1;
652 }
653 
654 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
655 {
656 	BUG();
657 	return NULL;
658 }
659 
660 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
661 				       struct net *net,
662 				       struct sockaddr *sa, int salen,
663 				       int flags)
664 {
665 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
666 }
667 
668 static struct svc_xprt_ops svc_udp_ops = {
669 	.xpo_create = svc_udp_create,
670 	.xpo_recvfrom = svc_udp_recvfrom,
671 	.xpo_sendto = svc_udp_sendto,
672 	.xpo_release_rqst = svc_release_skb,
673 	.xpo_detach = svc_sock_detach,
674 	.xpo_free = svc_sock_free,
675 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
676 	.xpo_has_wspace = svc_udp_has_wspace,
677 	.xpo_accept = svc_udp_accept,
678 };
679 
680 static struct svc_xprt_class svc_udp_class = {
681 	.xcl_name = "udp",
682 	.xcl_owner = THIS_MODULE,
683 	.xcl_ops = &svc_udp_ops,
684 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
685 };
686 
687 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
688 {
689 	int err, level, optname, one = 1;
690 
691 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
692 		      &svsk->sk_xprt, serv);
693 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
694 	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
695 	svsk->sk_sk->sk_write_space = svc_write_space;
696 
697 	/* initialise setting must have enough space to
698 	 * receive and respond to one request.
699 	 * svc_udp_recvfrom will re-adjust if necessary
700 	 */
701 	svc_sock_setbufsize(svsk->sk_sock,
702 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
703 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
704 
705 	/* data might have come in before data_ready set up */
706 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
707 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
708 
709 	/* make sure we get destination address info */
710 	switch (svsk->sk_sk->sk_family) {
711 	case AF_INET:
712 		level = SOL_IP;
713 		optname = IP_PKTINFO;
714 		break;
715 	case AF_INET6:
716 		level = SOL_IPV6;
717 		optname = IPV6_RECVPKTINFO;
718 		break;
719 	default:
720 		BUG();
721 	}
722 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
723 					(char *)&one, sizeof(one));
724 	dprintk("svc: kernel_setsockopt returned %d\n", err);
725 }
726 
727 /*
728  * A data_ready event on a listening socket means there's a connection
729  * pending. Do not use state_change as a substitute for it.
730  */
731 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
732 {
733 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
734 	wait_queue_head_t *wq;
735 
736 	dprintk("svc: socket %p TCP (listen) state change %d\n",
737 		sk, sk->sk_state);
738 
739 	/*
740 	 * This callback may called twice when a new connection
741 	 * is established as a child socket inherits everything
742 	 * from a parent LISTEN socket.
743 	 * 1) data_ready method of the parent socket will be called
744 	 *    when one of child sockets become ESTABLISHED.
745 	 * 2) data_ready method of the child socket may be called
746 	 *    when it receives data before the socket is accepted.
747 	 * In case of 2, we should ignore it silently.
748 	 */
749 	if (sk->sk_state == TCP_LISTEN) {
750 		if (svsk) {
751 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
752 			svc_xprt_enqueue(&svsk->sk_xprt);
753 		} else
754 			printk("svc: socket %p: no user data\n", sk);
755 	}
756 
757 	wq = sk_sleep(sk);
758 	if (wq && waitqueue_active(wq))
759 		wake_up_interruptible_all(wq);
760 }
761 
762 /*
763  * A state change on a connected socket means it's dying or dead.
764  */
765 static void svc_tcp_state_change(struct sock *sk)
766 {
767 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
768 	wait_queue_head_t *wq = sk_sleep(sk);
769 
770 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
771 		sk, sk->sk_state, sk->sk_user_data);
772 
773 	if (!svsk)
774 		printk("svc: socket %p: no user data\n", sk);
775 	else {
776 		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
777 		svc_xprt_enqueue(&svsk->sk_xprt);
778 	}
779 	if (wq && waitqueue_active(wq))
780 		wake_up_interruptible_all(wq);
781 }
782 
783 static void svc_tcp_data_ready(struct sock *sk, int count)
784 {
785 	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
786 	wait_queue_head_t *wq = sk_sleep(sk);
787 
788 	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
789 		sk, sk->sk_user_data);
790 	if (svsk) {
791 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
792 		svc_xprt_enqueue(&svsk->sk_xprt);
793 	}
794 	if (wq && waitqueue_active(wq))
795 		wake_up_interruptible(wq);
796 }
797 
798 /*
799  * Accept a TCP connection
800  */
801 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
802 {
803 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
804 	struct sockaddr_storage addr;
805 	struct sockaddr	*sin = (struct sockaddr *) &addr;
806 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
807 	struct socket	*sock = svsk->sk_sock;
808 	struct socket	*newsock;
809 	struct svc_sock	*newsvsk;
810 	int		err, slen;
811 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
812 
813 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
814 	if (!sock)
815 		return NULL;
816 
817 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
818 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
819 	if (err < 0) {
820 		if (err == -ENOMEM)
821 			printk(KERN_WARNING "%s: no more sockets!\n",
822 			       serv->sv_name);
823 		else if (err != -EAGAIN)
824 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
825 					     serv->sv_name, -err);
826 		return NULL;
827 	}
828 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
829 
830 	err = kernel_getpeername(newsock, sin, &slen);
831 	if (err < 0) {
832 		net_warn_ratelimited("%s: peername failed (err %d)!\n",
833 				     serv->sv_name, -err);
834 		goto failed;		/* aborted connection or whatever */
835 	}
836 
837 	/* Ideally, we would want to reject connections from unauthorized
838 	 * hosts here, but when we get encryption, the IP of the host won't
839 	 * tell us anything.  For now just warn about unpriv connections.
840 	 */
841 	if (!svc_port_is_privileged(sin)) {
842 		dprintk(KERN_WARNING
843 			"%s: connect from unprivileged port: %s\n",
844 			serv->sv_name,
845 			__svc_print_addr(sin, buf, sizeof(buf)));
846 	}
847 	dprintk("%s: connect from %s\n", serv->sv_name,
848 		__svc_print_addr(sin, buf, sizeof(buf)));
849 
850 	/* make sure that a write doesn't block forever when
851 	 * low on memory
852 	 */
853 	newsock->sk->sk_sndtimeo = HZ*30;
854 
855 	newsvsk = svc_setup_socket(serv, newsock,
856 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
857 	if (IS_ERR(newsvsk))
858 		goto failed;
859 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
860 	err = kernel_getsockname(newsock, sin, &slen);
861 	if (unlikely(err < 0)) {
862 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
863 		slen = offsetof(struct sockaddr, sa_data);
864 	}
865 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
866 
867 	if (serv->sv_stats)
868 		serv->sv_stats->nettcpconn++;
869 
870 	return &newsvsk->sk_xprt;
871 
872 failed:
873 	sock_release(newsock);
874 	return NULL;
875 }
876 
877 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
878 {
879 	unsigned int i, len, npages;
880 
881 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
882 		return 0;
883 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
884 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
885 	for (i = 0; i < npages; i++) {
886 		if (rqstp->rq_pages[i] != NULL)
887 			put_page(rqstp->rq_pages[i]);
888 		BUG_ON(svsk->sk_pages[i] == NULL);
889 		rqstp->rq_pages[i] = svsk->sk_pages[i];
890 		svsk->sk_pages[i] = NULL;
891 	}
892 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
893 	return len;
894 }
895 
896 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
897 {
898 	unsigned int i, len, npages;
899 
900 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
901 		return;
902 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
903 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
904 	for (i = 0; i < npages; i++) {
905 		svsk->sk_pages[i] = rqstp->rq_pages[i];
906 		rqstp->rq_pages[i] = NULL;
907 	}
908 }
909 
910 static void svc_tcp_clear_pages(struct svc_sock *svsk)
911 {
912 	unsigned int i, len, npages;
913 
914 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
915 		goto out;
916 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
917 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
918 	for (i = 0; i < npages; i++) {
919 		BUG_ON(svsk->sk_pages[i] == NULL);
920 		put_page(svsk->sk_pages[i]);
921 		svsk->sk_pages[i] = NULL;
922 	}
923 out:
924 	svsk->sk_tcplen = 0;
925 }
926 
927 /*
928  * Receive data.
929  * If we haven't gotten the record length yet, get the next four bytes.
930  * Otherwise try to gobble up as much as possible up to the complete
931  * record length.
932  */
933 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
934 {
935 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
936 	unsigned int want;
937 	int len;
938 
939 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
940 
941 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
942 		struct kvec	iov;
943 
944 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
945 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
946 		iov.iov_len  = want;
947 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
948 			goto error;
949 		svsk->sk_tcplen += len;
950 
951 		if (len < want) {
952 			dprintk("svc: short recvfrom while reading record "
953 				"length (%d of %d)\n", len, want);
954 			return -EAGAIN;
955 		}
956 
957 		svsk->sk_reclen = ntohl(svsk->sk_reclen);
958 		if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
959 			/* FIXME: technically, a record can be fragmented,
960 			 *  and non-terminal fragments will not have the top
961 			 *  bit set in the fragment length header.
962 			 *  But apparently no known nfs clients send fragmented
963 			 *  records. */
964 			net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
965 			goto err_delete;
966 		}
967 
968 		svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
969 		dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
970 		if (svsk->sk_reclen > serv->sv_max_mesg) {
971 			net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
972 					       (unsigned long)svsk->sk_reclen);
973 			goto err_delete;
974 		}
975 	}
976 
977 	if (svsk->sk_reclen < 8)
978 		goto err_delete; /* client is nuts. */
979 
980 	len = svsk->sk_reclen;
981 
982 	return len;
983 error:
984 	dprintk("RPC: TCP recv_record got %d\n", len);
985 	return len;
986 err_delete:
987 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
988 	return -EAGAIN;
989 }
990 
991 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
992 {
993 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
994 	struct rpc_rqst *req = NULL;
995 	struct kvec *src, *dst;
996 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
997 	__be32 xid;
998 	__be32 calldir;
999 
1000 	xid = *p++;
1001 	calldir = *p;
1002 
1003 	if (bc_xprt)
1004 		req = xprt_lookup_rqst(bc_xprt, xid);
1005 
1006 	if (!req) {
1007 		printk(KERN_NOTICE
1008 			"%s: Got unrecognized reply: "
1009 			"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1010 			__func__, ntohl(calldir),
1011 			bc_xprt, xid);
1012 		return -EAGAIN;
1013 	}
1014 
1015 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1016 	/*
1017 	 * XXX!: cheating for now!  Only copying HEAD.
1018 	 * But we know this is good enough for now (in fact, for any
1019 	 * callback reply in the forseeable future).
1020 	 */
1021 	dst = &req->rq_private_buf.head[0];
1022 	src = &rqstp->rq_arg.head[0];
1023 	if (dst->iov_len < src->iov_len)
1024 		return -EAGAIN; /* whatever; just giving up. */
1025 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1026 	xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1027 	rqstp->rq_arg.len = 0;
1028 	return 0;
1029 }
1030 
1031 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1032 {
1033 	int i = 0;
1034 	int t = 0;
1035 
1036 	while (t < len) {
1037 		vec[i].iov_base = page_address(pages[i]);
1038 		vec[i].iov_len = PAGE_SIZE;
1039 		i++;
1040 		t += PAGE_SIZE;
1041 	}
1042 	return i;
1043 }
1044 
1045 
1046 /*
1047  * Receive data from a TCP socket.
1048  */
1049 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1050 {
1051 	struct svc_sock	*svsk =
1052 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1053 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1054 	int		len;
1055 	struct kvec *vec;
1056 	unsigned int want, base;
1057 	__be32 *p;
1058 	__be32 calldir;
1059 	int pnum;
1060 
1061 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1062 		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1063 		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1064 		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1065 
1066 	len = svc_tcp_recv_record(svsk, rqstp);
1067 	if (len < 0)
1068 		goto error;
1069 
1070 	base = svc_tcp_restore_pages(svsk, rqstp);
1071 	want = svsk->sk_reclen - base;
1072 
1073 	vec = rqstp->rq_vec;
1074 
1075 	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1076 						svsk->sk_reclen);
1077 
1078 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1079 
1080 	/* Now receive data */
1081 	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1082 	if (len >= 0)
1083 		svsk->sk_tcplen += len;
1084 	if (len != want) {
1085 		svc_tcp_save_pages(svsk, rqstp);
1086 		if (len < 0 && len != -EAGAIN)
1087 			goto err_other;
1088 		dprintk("svc: incomplete TCP record (%d of %d)\n",
1089 			svsk->sk_tcplen, svsk->sk_reclen);
1090 		goto err_noclose;
1091 	}
1092 
1093 	rqstp->rq_arg.len = svsk->sk_reclen;
1094 	rqstp->rq_arg.page_base = 0;
1095 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1096 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1097 		rqstp->rq_arg.page_len = 0;
1098 	} else
1099 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1100 
1101 	rqstp->rq_xprt_ctxt   = NULL;
1102 	rqstp->rq_prot	      = IPPROTO_TCP;
1103 
1104 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1105 	calldir = p[1];
1106 	if (calldir)
1107 		len = receive_cb_reply(svsk, rqstp);
1108 
1109 	/* Reset TCP read info */
1110 	svsk->sk_reclen = 0;
1111 	svsk->sk_tcplen = 0;
1112 	/* If we have more data, signal svc_xprt_enqueue() to try again */
1113 	if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1114 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1115 
1116 	if (len < 0)
1117 		goto error;
1118 
1119 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1120 	if (serv->sv_stats)
1121 		serv->sv_stats->nettcpcnt++;
1122 
1123 	dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1124 	return rqstp->rq_arg.len;
1125 
1126 error:
1127 	if (len != -EAGAIN)
1128 		goto err_other;
1129 	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1130 	return 0;
1131 err_other:
1132 	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1133 	       svsk->sk_xprt.xpt_server->sv_name, -len);
1134 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1135 err_noclose:
1136 	return 0;	/* record not complete */
1137 }
1138 
1139 /*
1140  * Send out data on TCP socket.
1141  */
1142 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1143 {
1144 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1145 	int sent;
1146 	__be32 reclen;
1147 
1148 	/* Set up the first element of the reply kvec.
1149 	 * Any other kvecs that may be in use have been taken
1150 	 * care of by the server implementation itself.
1151 	 */
1152 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1153 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1154 
1155 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1156 	if (sent != xbufp->len) {
1157 		printk(KERN_NOTICE
1158 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1159 		       "- shutting down socket\n",
1160 		       rqstp->rq_xprt->xpt_server->sv_name,
1161 		       (sent<0)?"got error":"sent only",
1162 		       sent, xbufp->len);
1163 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1164 		svc_xprt_enqueue(rqstp->rq_xprt);
1165 		sent = -EAGAIN;
1166 	}
1167 	return sent;
1168 }
1169 
1170 /*
1171  * Setup response header. TCP has a 4B record length field.
1172  */
1173 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1174 {
1175 	struct kvec *resv = &rqstp->rq_res.head[0];
1176 
1177 	/* tcp needs a space for the record length... */
1178 	svc_putnl(resv, 0);
1179 }
1180 
1181 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1182 {
1183 	struct svc_sock *svsk =	container_of(xprt, struct svc_sock, sk_xprt);
1184 	struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1185 	int required;
1186 
1187 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1188 		return 1;
1189 	required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1190 	if (sk_stream_wspace(svsk->sk_sk) >= required)
1191 		return 1;
1192 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1193 	return 0;
1194 }
1195 
1196 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1197 				       struct net *net,
1198 				       struct sockaddr *sa, int salen,
1199 				       int flags)
1200 {
1201 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1202 }
1203 
1204 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1205 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1206 					     struct net *, struct sockaddr *,
1207 					     int, int);
1208 static void svc_bc_sock_free(struct svc_xprt *xprt);
1209 
1210 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1211 				       struct net *net,
1212 				       struct sockaddr *sa, int salen,
1213 				       int flags)
1214 {
1215 	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1216 }
1217 
1218 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1219 {
1220 }
1221 
1222 static struct svc_xprt_ops svc_tcp_bc_ops = {
1223 	.xpo_create = svc_bc_tcp_create,
1224 	.xpo_detach = svc_bc_tcp_sock_detach,
1225 	.xpo_free = svc_bc_sock_free,
1226 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1227 };
1228 
1229 static struct svc_xprt_class svc_tcp_bc_class = {
1230 	.xcl_name = "tcp-bc",
1231 	.xcl_owner = THIS_MODULE,
1232 	.xcl_ops = &svc_tcp_bc_ops,
1233 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1234 };
1235 
1236 static void svc_init_bc_xprt_sock(void)
1237 {
1238 	svc_reg_xprt_class(&svc_tcp_bc_class);
1239 }
1240 
1241 static void svc_cleanup_bc_xprt_sock(void)
1242 {
1243 	svc_unreg_xprt_class(&svc_tcp_bc_class);
1244 }
1245 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1246 static void svc_init_bc_xprt_sock(void)
1247 {
1248 }
1249 
1250 static void svc_cleanup_bc_xprt_sock(void)
1251 {
1252 }
1253 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1254 
1255 static struct svc_xprt_ops svc_tcp_ops = {
1256 	.xpo_create = svc_tcp_create,
1257 	.xpo_recvfrom = svc_tcp_recvfrom,
1258 	.xpo_sendto = svc_tcp_sendto,
1259 	.xpo_release_rqst = svc_release_skb,
1260 	.xpo_detach = svc_tcp_sock_detach,
1261 	.xpo_free = svc_sock_free,
1262 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1263 	.xpo_has_wspace = svc_tcp_has_wspace,
1264 	.xpo_accept = svc_tcp_accept,
1265 };
1266 
1267 static struct svc_xprt_class svc_tcp_class = {
1268 	.xcl_name = "tcp",
1269 	.xcl_owner = THIS_MODULE,
1270 	.xcl_ops = &svc_tcp_ops,
1271 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1272 };
1273 
1274 void svc_init_xprt_sock(void)
1275 {
1276 	svc_reg_xprt_class(&svc_tcp_class);
1277 	svc_reg_xprt_class(&svc_udp_class);
1278 	svc_init_bc_xprt_sock();
1279 }
1280 
1281 void svc_cleanup_xprt_sock(void)
1282 {
1283 	svc_unreg_xprt_class(&svc_tcp_class);
1284 	svc_unreg_xprt_class(&svc_udp_class);
1285 	svc_cleanup_bc_xprt_sock();
1286 }
1287 
1288 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1289 {
1290 	struct sock	*sk = svsk->sk_sk;
1291 
1292 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1293 		      &svsk->sk_xprt, serv);
1294 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1295 	if (sk->sk_state == TCP_LISTEN) {
1296 		dprintk("setting up TCP socket for listening\n");
1297 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1298 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1299 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1300 	} else {
1301 		dprintk("setting up TCP socket for reading\n");
1302 		sk->sk_state_change = svc_tcp_state_change;
1303 		sk->sk_data_ready = svc_tcp_data_ready;
1304 		sk->sk_write_space = svc_tcp_write_space;
1305 
1306 		svsk->sk_reclen = 0;
1307 		svsk->sk_tcplen = 0;
1308 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1309 
1310 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1311 
1312 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1313 		if (sk->sk_state != TCP_ESTABLISHED)
1314 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1315 	}
1316 }
1317 
1318 void svc_sock_update_bufs(struct svc_serv *serv)
1319 {
1320 	/*
1321 	 * The number of server threads has changed. Update
1322 	 * rcvbuf and sndbuf accordingly on all sockets
1323 	 */
1324 	struct svc_sock *svsk;
1325 
1326 	spin_lock_bh(&serv->sv_lock);
1327 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1328 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1329 	spin_unlock_bh(&serv->sv_lock);
1330 }
1331 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1332 
1333 /*
1334  * Initialize socket for RPC use and create svc_sock struct
1335  * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1336  */
1337 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1338 						struct socket *sock,
1339 						int flags)
1340 {
1341 	struct svc_sock	*svsk;
1342 	struct sock	*inet;
1343 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1344 	int		err = 0;
1345 
1346 	dprintk("svc: svc_setup_socket %p\n", sock);
1347 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1348 	if (!svsk)
1349 		return ERR_PTR(-ENOMEM);
1350 
1351 	inet = sock->sk;
1352 
1353 	/* Register socket with portmapper */
1354 	if (pmap_register)
1355 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1356 				     inet->sk_protocol,
1357 				     ntohs(inet_sk(inet)->inet_sport));
1358 
1359 	if (err < 0) {
1360 		kfree(svsk);
1361 		return ERR_PTR(err);
1362 	}
1363 
1364 	inet->sk_user_data = svsk;
1365 	svsk->sk_sock = sock;
1366 	svsk->sk_sk = inet;
1367 	svsk->sk_ostate = inet->sk_state_change;
1368 	svsk->sk_odata = inet->sk_data_ready;
1369 	svsk->sk_owspace = inet->sk_write_space;
1370 
1371 	/* Initialize the socket */
1372 	if (sock->type == SOCK_DGRAM)
1373 		svc_udp_init(svsk, serv);
1374 	else {
1375 		/* initialise setting must have enough space to
1376 		 * receive and respond to one request.
1377 		 */
1378 		svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1379 					4 * serv->sv_max_mesg);
1380 		svc_tcp_init(svsk, serv);
1381 	}
1382 
1383 	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1384 				svsk, svsk->sk_sk);
1385 
1386 	return svsk;
1387 }
1388 
1389 /**
1390  * svc_addsock - add a listener socket to an RPC service
1391  * @serv: pointer to RPC service to which to add a new listener
1392  * @fd: file descriptor of the new listener
1393  * @name_return: pointer to buffer to fill in with name of listener
1394  * @len: size of the buffer
1395  *
1396  * Fills in socket name and returns positive length of name if successful.
1397  * Name is terminated with '\n'.  On error, returns a negative errno
1398  * value.
1399  */
1400 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1401 		const size_t len)
1402 {
1403 	int err = 0;
1404 	struct socket *so = sockfd_lookup(fd, &err);
1405 	struct svc_sock *svsk = NULL;
1406 	struct sockaddr_storage addr;
1407 	struct sockaddr *sin = (struct sockaddr *)&addr;
1408 	int salen;
1409 
1410 	if (!so)
1411 		return err;
1412 	err = -EAFNOSUPPORT;
1413 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1414 		goto out;
1415 	err =  -EPROTONOSUPPORT;
1416 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1417 	    so->sk->sk_protocol != IPPROTO_UDP)
1418 		goto out;
1419 	err = -EISCONN;
1420 	if (so->state > SS_UNCONNECTED)
1421 		goto out;
1422 	err = -ENOENT;
1423 	if (!try_module_get(THIS_MODULE))
1424 		goto out;
1425 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1426 	if (IS_ERR(svsk)) {
1427 		module_put(THIS_MODULE);
1428 		err = PTR_ERR(svsk);
1429 		goto out;
1430 	}
1431 	if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1432 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1433 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1434 	return svc_one_sock_name(svsk, name_return, len);
1435 out:
1436 	sockfd_put(so);
1437 	return err;
1438 }
1439 EXPORT_SYMBOL_GPL(svc_addsock);
1440 
1441 /*
1442  * Create socket for RPC service.
1443  */
1444 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1445 					  int protocol,
1446 					  struct net *net,
1447 					  struct sockaddr *sin, int len,
1448 					  int flags)
1449 {
1450 	struct svc_sock	*svsk;
1451 	struct socket	*sock;
1452 	int		error;
1453 	int		type;
1454 	struct sockaddr_storage addr;
1455 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1456 	int		newlen;
1457 	int		family;
1458 	int		val;
1459 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1460 
1461 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1462 			serv->sv_program->pg_name, protocol,
1463 			__svc_print_addr(sin, buf, sizeof(buf)));
1464 
1465 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1466 		printk(KERN_WARNING "svc: only UDP and TCP "
1467 				"sockets supported\n");
1468 		return ERR_PTR(-EINVAL);
1469 	}
1470 
1471 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1472 	switch (sin->sa_family) {
1473 	case AF_INET6:
1474 		family = PF_INET6;
1475 		break;
1476 	case AF_INET:
1477 		family = PF_INET;
1478 		break;
1479 	default:
1480 		return ERR_PTR(-EINVAL);
1481 	}
1482 
1483 	error = __sock_create(net, family, type, protocol, &sock, 1);
1484 	if (error < 0)
1485 		return ERR_PTR(error);
1486 
1487 	svc_reclassify_socket(sock);
1488 
1489 	/*
1490 	 * If this is an PF_INET6 listener, we want to avoid
1491 	 * getting requests from IPv4 remotes.  Those should
1492 	 * be shunted to a PF_INET listener via rpcbind.
1493 	 */
1494 	val = 1;
1495 	if (family == PF_INET6)
1496 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1497 					(char *)&val, sizeof(val));
1498 
1499 	if (type == SOCK_STREAM)
1500 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1501 	error = kernel_bind(sock, sin, len);
1502 	if (error < 0)
1503 		goto bummer;
1504 
1505 	newlen = len;
1506 	error = kernel_getsockname(sock, newsin, &newlen);
1507 	if (error < 0)
1508 		goto bummer;
1509 
1510 	if (protocol == IPPROTO_TCP) {
1511 		if ((error = kernel_listen(sock, 64)) < 0)
1512 			goto bummer;
1513 	}
1514 
1515 	svsk = svc_setup_socket(serv, sock, flags);
1516 	if (IS_ERR(svsk)) {
1517 		error = PTR_ERR(svsk);
1518 		goto bummer;
1519 	}
1520 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1521 	return (struct svc_xprt *)svsk;
1522 bummer:
1523 	dprintk("svc: svc_create_socket error = %d\n", -error);
1524 	sock_release(sock);
1525 	return ERR_PTR(error);
1526 }
1527 
1528 /*
1529  * Detach the svc_sock from the socket so that no
1530  * more callbacks occur.
1531  */
1532 static void svc_sock_detach(struct svc_xprt *xprt)
1533 {
1534 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1535 	struct sock *sk = svsk->sk_sk;
1536 	wait_queue_head_t *wq;
1537 
1538 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1539 
1540 	/* put back the old socket callbacks */
1541 	sk->sk_state_change = svsk->sk_ostate;
1542 	sk->sk_data_ready = svsk->sk_odata;
1543 	sk->sk_write_space = svsk->sk_owspace;
1544 
1545 	wq = sk_sleep(sk);
1546 	if (wq && waitqueue_active(wq))
1547 		wake_up_interruptible(wq);
1548 }
1549 
1550 /*
1551  * Disconnect the socket, and reset the callbacks
1552  */
1553 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1554 {
1555 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1556 
1557 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1558 
1559 	svc_sock_detach(xprt);
1560 
1561 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1562 		svc_tcp_clear_pages(svsk);
1563 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1564 	}
1565 }
1566 
1567 /*
1568  * Free the svc_sock's socket resources and the svc_sock itself.
1569  */
1570 static void svc_sock_free(struct svc_xprt *xprt)
1571 {
1572 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1573 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1574 
1575 	if (svsk->sk_sock->file)
1576 		sockfd_put(svsk->sk_sock);
1577 	else
1578 		sock_release(svsk->sk_sock);
1579 	kfree(svsk);
1580 }
1581 
1582 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1583 /*
1584  * Create a back channel svc_xprt which shares the fore channel socket.
1585  */
1586 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1587 					     int protocol,
1588 					     struct net *net,
1589 					     struct sockaddr *sin, int len,
1590 					     int flags)
1591 {
1592 	struct svc_sock *svsk;
1593 	struct svc_xprt *xprt;
1594 
1595 	if (protocol != IPPROTO_TCP) {
1596 		printk(KERN_WARNING "svc: only TCP sockets"
1597 			" supported on shared back channel\n");
1598 		return ERR_PTR(-EINVAL);
1599 	}
1600 
1601 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1602 	if (!svsk)
1603 		return ERR_PTR(-ENOMEM);
1604 
1605 	xprt = &svsk->sk_xprt;
1606 	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1607 
1608 	serv->sv_bc_xprt = xprt;
1609 
1610 	return xprt;
1611 }
1612 
1613 /*
1614  * Free a back channel svc_sock.
1615  */
1616 static void svc_bc_sock_free(struct svc_xprt *xprt)
1617 {
1618 	if (xprt)
1619 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1620 }
1621 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1622