xref: /openbmc/linux/net/sunrpc/svcsock.c (revision 79f08d9e)
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 #if IS_ENABLED(CONFIG_IPV6)
295 	case PF_INET6:
296 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
297 				proto_name,
298 				&sk->sk_v6_rcv_saddr,
299 				inet_sk(sk)->inet_num);
300 		break;
301 #endif
302 	default:
303 		len = snprintf(buf, remaining, "*unknown-%d*\n",
304 				sk->sk_family);
305 	}
306 
307 	if (len >= remaining) {
308 		*buf = '\0';
309 		return -ENAMETOOLONG;
310 	}
311 	return len;
312 }
313 
314 /*
315  * Check input queue length
316  */
317 static int svc_recv_available(struct svc_sock *svsk)
318 {
319 	struct socket	*sock = svsk->sk_sock;
320 	int		avail, err;
321 
322 	err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
323 
324 	return (err >= 0)? avail : err;
325 }
326 
327 /*
328  * Generic recvfrom routine.
329  */
330 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
331 			int buflen)
332 {
333 	struct svc_sock *svsk =
334 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
335 	struct msghdr msg = {
336 		.msg_flags	= MSG_DONTWAIT,
337 	};
338 	int len;
339 
340 	rqstp->rq_xprt_hlen = 0;
341 
342 	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
343 				msg.msg_flags);
344 
345 	dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
346 		svsk, iov[0].iov_base, iov[0].iov_len, len);
347 	return len;
348 }
349 
350 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
351 				struct kvec *iov, int nr,
352 				int buflen, unsigned int base)
353 {
354 	size_t save_iovlen;
355 	void *save_iovbase;
356 	unsigned int i;
357 	int ret;
358 
359 	if (base == 0)
360 		return svc_recvfrom(rqstp, iov, nr, buflen);
361 
362 	for (i = 0; i < nr; i++) {
363 		if (iov[i].iov_len > base)
364 			break;
365 		base -= iov[i].iov_len;
366 	}
367 	save_iovlen = iov[i].iov_len;
368 	save_iovbase = iov[i].iov_base;
369 	iov[i].iov_len -= base;
370 	iov[i].iov_base += base;
371 	ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
372 	iov[i].iov_len = save_iovlen;
373 	iov[i].iov_base = save_iovbase;
374 	return ret;
375 }
376 
377 /*
378  * Set socket snd and rcv buffer lengths
379  */
380 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
381 				unsigned int rcv)
382 {
383 #if 0
384 	mm_segment_t	oldfs;
385 	oldfs = get_fs(); set_fs(KERNEL_DS);
386 	sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
387 			(char*)&snd, sizeof(snd));
388 	sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
389 			(char*)&rcv, sizeof(rcv));
390 #else
391 	/* sock_setsockopt limits use to sysctl_?mem_max,
392 	 * which isn't acceptable.  Until that is made conditional
393 	 * on not having CAP_SYS_RESOURCE or similar, we go direct...
394 	 * DaveM said I could!
395 	 */
396 	lock_sock(sock->sk);
397 	sock->sk->sk_sndbuf = snd * 2;
398 	sock->sk->sk_rcvbuf = rcv * 2;
399 	sock->sk->sk_write_space(sock->sk);
400 	release_sock(sock->sk);
401 #endif
402 }
403 /*
404  * INET callback when data has been received on the socket.
405  */
406 static void svc_udp_data_ready(struct sock *sk, int count)
407 {
408 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
409 	wait_queue_head_t *wq = sk_sleep(sk);
410 
411 	if (svsk) {
412 		dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
413 			svsk, sk, count,
414 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
415 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
416 		svc_xprt_enqueue(&svsk->sk_xprt);
417 	}
418 	if (wq && waitqueue_active(wq))
419 		wake_up_interruptible(wq);
420 }
421 
422 /*
423  * INET callback when space is newly available on the socket.
424  */
425 static void svc_write_space(struct sock *sk)
426 {
427 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
428 	wait_queue_head_t *wq = sk_sleep(sk);
429 
430 	if (svsk) {
431 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
432 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
433 		svc_xprt_enqueue(&svsk->sk_xprt);
434 	}
435 
436 	if (wq && waitqueue_active(wq)) {
437 		dprintk("RPC svc_write_space: someone sleeping on %p\n",
438 		       svsk);
439 		wake_up_interruptible(wq);
440 	}
441 }
442 
443 static void svc_tcp_write_space(struct sock *sk)
444 {
445 	struct socket *sock = sk->sk_socket;
446 
447 	if (sk_stream_is_writeable(sk) && sock)
448 		clear_bit(SOCK_NOSPACE, &sock->flags);
449 	svc_write_space(sk);
450 }
451 
452 /*
453  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
454  */
455 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
456 				     struct cmsghdr *cmh)
457 {
458 	struct in_pktinfo *pki = CMSG_DATA(cmh);
459 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
460 
461 	if (cmh->cmsg_type != IP_PKTINFO)
462 		return 0;
463 
464 	daddr->sin_family = AF_INET;
465 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
466 	return 1;
467 }
468 
469 /*
470  * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
471  */
472 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
473 				     struct cmsghdr *cmh)
474 {
475 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
476 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
477 
478 	if (cmh->cmsg_type != IPV6_PKTINFO)
479 		return 0;
480 
481 	daddr->sin6_family = AF_INET6;
482 	daddr->sin6_addr = pki->ipi6_addr;
483 	daddr->sin6_scope_id = pki->ipi6_ifindex;
484 	return 1;
485 }
486 
487 /*
488  * Copy the UDP datagram's destination address to the rqstp structure.
489  * The 'destination' address in this case is the address to which the
490  * peer sent the datagram, i.e. our local address. For multihomed
491  * hosts, this can change from msg to msg. Note that only the IP
492  * address changes, the port number should remain the same.
493  */
494 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
495 				    struct cmsghdr *cmh)
496 {
497 	switch (cmh->cmsg_level) {
498 	case SOL_IP:
499 		return svc_udp_get_dest_address4(rqstp, cmh);
500 	case SOL_IPV6:
501 		return svc_udp_get_dest_address6(rqstp, cmh);
502 	}
503 
504 	return 0;
505 }
506 
507 /*
508  * Receive a datagram from a UDP socket.
509  */
510 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
511 {
512 	struct svc_sock	*svsk =
513 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
514 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
515 	struct sk_buff	*skb;
516 	union {
517 		struct cmsghdr	hdr;
518 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
519 	} buffer;
520 	struct cmsghdr *cmh = &buffer.hdr;
521 	struct msghdr msg = {
522 		.msg_name = svc_addr(rqstp),
523 		.msg_control = cmh,
524 		.msg_controllen = sizeof(buffer),
525 		.msg_flags = MSG_DONTWAIT,
526 	};
527 	size_t len;
528 	int err;
529 
530 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
531 	    /* udp sockets need large rcvbuf as all pending
532 	     * requests are still in that buffer.  sndbuf must
533 	     * also be large enough that there is enough space
534 	     * for one reply per thread.  We count all threads
535 	     * rather than threads in a particular pool, which
536 	     * provides an upper bound on the number of threads
537 	     * which will access the socket.
538 	     */
539 	    svc_sock_setbufsize(svsk->sk_sock,
540 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
541 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
542 
543 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
544 	skb = NULL;
545 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
546 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
547 	if (err >= 0)
548 		skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
549 
550 	if (skb == NULL) {
551 		if (err != -EAGAIN) {
552 			/* possibly an icmp error */
553 			dprintk("svc: recvfrom returned error %d\n", -err);
554 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
555 		}
556 		return 0;
557 	}
558 	len = svc_addr_len(svc_addr(rqstp));
559 	rqstp->rq_addrlen = len;
560 	if (skb->tstamp.tv64 == 0) {
561 		skb->tstamp = ktime_get_real();
562 		/* Don't enable netstamp, sunrpc doesn't
563 		   need that much accuracy */
564 	}
565 	svsk->sk_sk->sk_stamp = skb->tstamp;
566 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
567 
568 	len  = skb->len - sizeof(struct udphdr);
569 	rqstp->rq_arg.len = len;
570 
571 	rqstp->rq_prot = IPPROTO_UDP;
572 
573 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
574 		net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
575 				     cmh->cmsg_level, cmh->cmsg_type);
576 		goto out_free;
577 	}
578 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
579 
580 	if (skb_is_nonlinear(skb)) {
581 		/* we have to copy */
582 		local_bh_disable();
583 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
584 			local_bh_enable();
585 			/* checksum error */
586 			goto out_free;
587 		}
588 		local_bh_enable();
589 		skb_free_datagram_locked(svsk->sk_sk, skb);
590 	} else {
591 		/* we can use it in-place */
592 		rqstp->rq_arg.head[0].iov_base = skb->data +
593 			sizeof(struct udphdr);
594 		rqstp->rq_arg.head[0].iov_len = len;
595 		if (skb_checksum_complete(skb))
596 			goto out_free;
597 		rqstp->rq_xprt_ctxt = skb;
598 	}
599 
600 	rqstp->rq_arg.page_base = 0;
601 	if (len <= rqstp->rq_arg.head[0].iov_len) {
602 		rqstp->rq_arg.head[0].iov_len = len;
603 		rqstp->rq_arg.page_len = 0;
604 		rqstp->rq_respages = rqstp->rq_pages+1;
605 	} else {
606 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
607 		rqstp->rq_respages = rqstp->rq_pages + 1 +
608 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
609 	}
610 	rqstp->rq_next_page = rqstp->rq_respages+1;
611 
612 	if (serv->sv_stats)
613 		serv->sv_stats->netudpcnt++;
614 
615 	return len;
616 out_free:
617 	trace_kfree_skb(skb, svc_udp_recvfrom);
618 	skb_free_datagram_locked(svsk->sk_sk, skb);
619 	return 0;
620 }
621 
622 static int
623 svc_udp_sendto(struct svc_rqst *rqstp)
624 {
625 	int		error;
626 
627 	error = svc_sendto(rqstp, &rqstp->rq_res);
628 	if (error == -ECONNREFUSED)
629 		/* ICMP error on earlier request. */
630 		error = svc_sendto(rqstp, &rqstp->rq_res);
631 
632 	return error;
633 }
634 
635 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
636 {
637 }
638 
639 static int svc_udp_has_wspace(struct svc_xprt *xprt)
640 {
641 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
642 	struct svc_serv	*serv = xprt->xpt_server;
643 	unsigned long required;
644 
645 	/*
646 	 * Set the SOCK_NOSPACE flag before checking the available
647 	 * sock space.
648 	 */
649 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
650 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
651 	if (required*2 > sock_wspace(svsk->sk_sk))
652 		return 0;
653 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
654 	return 1;
655 }
656 
657 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
658 {
659 	BUG();
660 	return NULL;
661 }
662 
663 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
664 				       struct net *net,
665 				       struct sockaddr *sa, int salen,
666 				       int flags)
667 {
668 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
669 }
670 
671 static struct svc_xprt_ops svc_udp_ops = {
672 	.xpo_create = svc_udp_create,
673 	.xpo_recvfrom = svc_udp_recvfrom,
674 	.xpo_sendto = svc_udp_sendto,
675 	.xpo_release_rqst = svc_release_skb,
676 	.xpo_detach = svc_sock_detach,
677 	.xpo_free = svc_sock_free,
678 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
679 	.xpo_has_wspace = svc_udp_has_wspace,
680 	.xpo_accept = svc_udp_accept,
681 };
682 
683 static struct svc_xprt_class svc_udp_class = {
684 	.xcl_name = "udp",
685 	.xcl_owner = THIS_MODULE,
686 	.xcl_ops = &svc_udp_ops,
687 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
688 };
689 
690 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
691 {
692 	int err, level, optname, one = 1;
693 
694 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
695 		      &svsk->sk_xprt, serv);
696 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
697 	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
698 	svsk->sk_sk->sk_write_space = svc_write_space;
699 
700 	/* initialise setting must have enough space to
701 	 * receive and respond to one request.
702 	 * svc_udp_recvfrom will re-adjust if necessary
703 	 */
704 	svc_sock_setbufsize(svsk->sk_sock,
705 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
706 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
707 
708 	/* data might have come in before data_ready set up */
709 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
710 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
711 
712 	/* make sure we get destination address info */
713 	switch (svsk->sk_sk->sk_family) {
714 	case AF_INET:
715 		level = SOL_IP;
716 		optname = IP_PKTINFO;
717 		break;
718 	case AF_INET6:
719 		level = SOL_IPV6;
720 		optname = IPV6_RECVPKTINFO;
721 		break;
722 	default:
723 		BUG();
724 	}
725 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
726 					(char *)&one, sizeof(one));
727 	dprintk("svc: kernel_setsockopt returned %d\n", err);
728 }
729 
730 /*
731  * A data_ready event on a listening socket means there's a connection
732  * pending. Do not use state_change as a substitute for it.
733  */
734 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
735 {
736 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
737 	wait_queue_head_t *wq;
738 
739 	dprintk("svc: socket %p TCP (listen) state change %d\n",
740 		sk, sk->sk_state);
741 
742 	/*
743 	 * This callback may called twice when a new connection
744 	 * is established as a child socket inherits everything
745 	 * from a parent LISTEN socket.
746 	 * 1) data_ready method of the parent socket will be called
747 	 *    when one of child sockets become ESTABLISHED.
748 	 * 2) data_ready method of the child socket may be called
749 	 *    when it receives data before the socket is accepted.
750 	 * In case of 2, we should ignore it silently.
751 	 */
752 	if (sk->sk_state == TCP_LISTEN) {
753 		if (svsk) {
754 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
755 			svc_xprt_enqueue(&svsk->sk_xprt);
756 		} else
757 			printk("svc: socket %p: no user data\n", sk);
758 	}
759 
760 	wq = sk_sleep(sk);
761 	if (wq && waitqueue_active(wq))
762 		wake_up_interruptible_all(wq);
763 }
764 
765 /*
766  * A state change on a connected socket means it's dying or dead.
767  */
768 static void svc_tcp_state_change(struct sock *sk)
769 {
770 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
771 	wait_queue_head_t *wq = sk_sleep(sk);
772 
773 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
774 		sk, sk->sk_state, sk->sk_user_data);
775 
776 	if (!svsk)
777 		printk("svc: socket %p: no user data\n", sk);
778 	else {
779 		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
780 		svc_xprt_enqueue(&svsk->sk_xprt);
781 	}
782 	if (wq && waitqueue_active(wq))
783 		wake_up_interruptible_all(wq);
784 }
785 
786 static void svc_tcp_data_ready(struct sock *sk, int count)
787 {
788 	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
789 	wait_queue_head_t *wq = sk_sleep(sk);
790 
791 	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
792 		sk, sk->sk_user_data);
793 	if (svsk) {
794 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
795 		svc_xprt_enqueue(&svsk->sk_xprt);
796 	}
797 	if (wq && waitqueue_active(wq))
798 		wake_up_interruptible(wq);
799 }
800 
801 /*
802  * Accept a TCP connection
803  */
804 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
805 {
806 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
807 	struct sockaddr_storage addr;
808 	struct sockaddr	*sin = (struct sockaddr *) &addr;
809 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
810 	struct socket	*sock = svsk->sk_sock;
811 	struct socket	*newsock;
812 	struct svc_sock	*newsvsk;
813 	int		err, slen;
814 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
815 
816 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
817 	if (!sock)
818 		return NULL;
819 
820 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
821 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
822 	if (err < 0) {
823 		if (err == -ENOMEM)
824 			printk(KERN_WARNING "%s: no more sockets!\n",
825 			       serv->sv_name);
826 		else if (err != -EAGAIN)
827 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
828 					     serv->sv_name, -err);
829 		return NULL;
830 	}
831 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
832 
833 	err = kernel_getpeername(newsock, sin, &slen);
834 	if (err < 0) {
835 		net_warn_ratelimited("%s: peername failed (err %d)!\n",
836 				     serv->sv_name, -err);
837 		goto failed;		/* aborted connection or whatever */
838 	}
839 
840 	/* Ideally, we would want to reject connections from unauthorized
841 	 * hosts here, but when we get encryption, the IP of the host won't
842 	 * tell us anything.  For now just warn about unpriv connections.
843 	 */
844 	if (!svc_port_is_privileged(sin)) {
845 		dprintk(KERN_WARNING
846 			"%s: connect from unprivileged port: %s\n",
847 			serv->sv_name,
848 			__svc_print_addr(sin, buf, sizeof(buf)));
849 	}
850 	dprintk("%s: connect from %s\n", serv->sv_name,
851 		__svc_print_addr(sin, buf, sizeof(buf)));
852 
853 	/* make sure that a write doesn't block forever when
854 	 * low on memory
855 	 */
856 	newsock->sk->sk_sndtimeo = HZ*30;
857 
858 	newsvsk = svc_setup_socket(serv, newsock,
859 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
860 	if (IS_ERR(newsvsk))
861 		goto failed;
862 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
863 	err = kernel_getsockname(newsock, sin, &slen);
864 	if (unlikely(err < 0)) {
865 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
866 		slen = offsetof(struct sockaddr, sa_data);
867 	}
868 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
869 
870 	if (serv->sv_stats)
871 		serv->sv_stats->nettcpconn++;
872 
873 	return &newsvsk->sk_xprt;
874 
875 failed:
876 	sock_release(newsock);
877 	return NULL;
878 }
879 
880 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
881 {
882 	unsigned int i, len, npages;
883 
884 	if (svsk->sk_datalen == 0)
885 		return 0;
886 	len = svsk->sk_datalen;
887 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
888 	for (i = 0; i < npages; i++) {
889 		if (rqstp->rq_pages[i] != NULL)
890 			put_page(rqstp->rq_pages[i]);
891 		BUG_ON(svsk->sk_pages[i] == NULL);
892 		rqstp->rq_pages[i] = svsk->sk_pages[i];
893 		svsk->sk_pages[i] = NULL;
894 	}
895 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
896 	return len;
897 }
898 
899 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
900 {
901 	unsigned int i, len, npages;
902 
903 	if (svsk->sk_datalen == 0)
904 		return;
905 	len = svsk->sk_datalen;
906 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
907 	for (i = 0; i < npages; i++) {
908 		svsk->sk_pages[i] = rqstp->rq_pages[i];
909 		rqstp->rq_pages[i] = NULL;
910 	}
911 }
912 
913 static void svc_tcp_clear_pages(struct svc_sock *svsk)
914 {
915 	unsigned int i, len, npages;
916 
917 	if (svsk->sk_datalen == 0)
918 		goto out;
919 	len = svsk->sk_datalen;
920 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
921 	for (i = 0; i < npages; i++) {
922 		if (svsk->sk_pages[i] == NULL) {
923 			WARN_ON_ONCE(1);
924 			continue;
925 		}
926 		put_page(svsk->sk_pages[i]);
927 		svsk->sk_pages[i] = NULL;
928 	}
929 out:
930 	svsk->sk_tcplen = 0;
931 	svsk->sk_datalen = 0;
932 }
933 
934 /*
935  * Receive fragment record header.
936  * If we haven't gotten the record length yet, get the next four bytes.
937  */
938 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
939 {
940 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
941 	unsigned int want;
942 	int len;
943 
944 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
945 
946 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
947 		struct kvec	iov;
948 
949 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
950 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
951 		iov.iov_len  = want;
952 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
953 			goto error;
954 		svsk->sk_tcplen += len;
955 
956 		if (len < want) {
957 			dprintk("svc: short recvfrom while reading record "
958 				"length (%d of %d)\n", len, want);
959 			return -EAGAIN;
960 		}
961 
962 		dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
963 		if (svc_sock_reclen(svsk) + svsk->sk_datalen >
964 							serv->sv_max_mesg) {
965 			net_notice_ratelimited("RPC: fragment too large: %d\n",
966 					svc_sock_reclen(svsk));
967 			goto err_delete;
968 		}
969 	}
970 
971 	return svc_sock_reclen(svsk);
972 error:
973 	dprintk("RPC: TCP recv_record got %d\n", len);
974 	return len;
975 err_delete:
976 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
977 	return -EAGAIN;
978 }
979 
980 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
981 {
982 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
983 	struct rpc_rqst *req = NULL;
984 	struct kvec *src, *dst;
985 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
986 	__be32 xid;
987 	__be32 calldir;
988 
989 	xid = *p++;
990 	calldir = *p;
991 
992 	if (bc_xprt)
993 		req = xprt_lookup_rqst(bc_xprt, xid);
994 
995 	if (!req) {
996 		printk(KERN_NOTICE
997 			"%s: Got unrecognized reply: "
998 			"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
999 			__func__, ntohl(calldir),
1000 			bc_xprt, xid);
1001 		return -EAGAIN;
1002 	}
1003 
1004 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1005 	/*
1006 	 * XXX!: cheating for now!  Only copying HEAD.
1007 	 * But we know this is good enough for now (in fact, for any
1008 	 * callback reply in the forseeable future).
1009 	 */
1010 	dst = &req->rq_private_buf.head[0];
1011 	src = &rqstp->rq_arg.head[0];
1012 	if (dst->iov_len < src->iov_len)
1013 		return -EAGAIN; /* whatever; just giving up. */
1014 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1015 	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1016 	rqstp->rq_arg.len = 0;
1017 	return 0;
1018 }
1019 
1020 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1021 {
1022 	int i = 0;
1023 	int t = 0;
1024 
1025 	while (t < len) {
1026 		vec[i].iov_base = page_address(pages[i]);
1027 		vec[i].iov_len = PAGE_SIZE;
1028 		i++;
1029 		t += PAGE_SIZE;
1030 	}
1031 	return i;
1032 }
1033 
1034 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1035 {
1036 	/* If we have more data, signal svc_xprt_enqueue() to try again */
1037 	if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1038 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1039 	dprintk("svc: TCP %s record (%d bytes)\n",
1040 		svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1041 		svc_sock_reclen(svsk));
1042 	svsk->sk_tcplen = 0;
1043 	svsk->sk_reclen = 0;
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 = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1072 
1073 	vec = rqstp->rq_vec;
1074 
1075 	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1076 						svsk->sk_datalen + want);
1077 
1078 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1079 	rqstp->rq_next_page = rqstp->rq_respages + 1;
1080 
1081 	/* Now receive data */
1082 	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1083 	if (len >= 0) {
1084 		svsk->sk_tcplen += len;
1085 		svsk->sk_datalen += len;
1086 	}
1087 	if (len != want || !svc_sock_final_rec(svsk)) {
1088 		svc_tcp_save_pages(svsk, rqstp);
1089 		if (len < 0 && len != -EAGAIN)
1090 			goto err_delete;
1091 		if (len == want)
1092 			svc_tcp_fragment_received(svsk);
1093 		else
1094 			dprintk("svc: incomplete TCP record (%d of %d)\n",
1095 				(int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1096 				svc_sock_reclen(svsk));
1097 		goto err_noclose;
1098 	}
1099 
1100 	if (svsk->sk_datalen < 8) {
1101 		svsk->sk_datalen = 0;
1102 		goto err_delete; /* client is nuts. */
1103 	}
1104 
1105 	rqstp->rq_arg.len = svsk->sk_datalen;
1106 	rqstp->rq_arg.page_base = 0;
1107 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1108 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1109 		rqstp->rq_arg.page_len = 0;
1110 	} else
1111 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1112 
1113 	rqstp->rq_xprt_ctxt   = NULL;
1114 	rqstp->rq_prot	      = IPPROTO_TCP;
1115 
1116 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1117 	calldir = p[1];
1118 	if (calldir)
1119 		len = receive_cb_reply(svsk, rqstp);
1120 
1121 	/* Reset TCP read info */
1122 	svsk->sk_datalen = 0;
1123 	svc_tcp_fragment_received(svsk);
1124 
1125 	if (len < 0)
1126 		goto error;
1127 
1128 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1129 	if (serv->sv_stats)
1130 		serv->sv_stats->nettcpcnt++;
1131 
1132 	return rqstp->rq_arg.len;
1133 
1134 error:
1135 	if (len != -EAGAIN)
1136 		goto err_delete;
1137 	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1138 	return 0;
1139 err_delete:
1140 	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1141 	       svsk->sk_xprt.xpt_server->sv_name, -len);
1142 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1143 err_noclose:
1144 	return 0;	/* record not complete */
1145 }
1146 
1147 /*
1148  * Send out data on TCP socket.
1149  */
1150 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1151 {
1152 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1153 	int sent;
1154 	__be32 reclen;
1155 
1156 	/* Set up the first element of the reply kvec.
1157 	 * Any other kvecs that may be in use have been taken
1158 	 * care of by the server implementation itself.
1159 	 */
1160 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1161 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1162 
1163 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1164 	if (sent != xbufp->len) {
1165 		printk(KERN_NOTICE
1166 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1167 		       "- shutting down socket\n",
1168 		       rqstp->rq_xprt->xpt_server->sv_name,
1169 		       (sent<0)?"got error":"sent only",
1170 		       sent, xbufp->len);
1171 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1172 		svc_xprt_enqueue(rqstp->rq_xprt);
1173 		sent = -EAGAIN;
1174 	}
1175 	return sent;
1176 }
1177 
1178 /*
1179  * Setup response header. TCP has a 4B record length field.
1180  */
1181 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1182 {
1183 	struct kvec *resv = &rqstp->rq_res.head[0];
1184 
1185 	/* tcp needs a space for the record length... */
1186 	svc_putnl(resv, 0);
1187 }
1188 
1189 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1190 {
1191 	struct svc_sock *svsk =	container_of(xprt, struct svc_sock, sk_xprt);
1192 	struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1193 	int required;
1194 
1195 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1196 		return 1;
1197 	required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1198 	if (sk_stream_wspace(svsk->sk_sk) >= required ||
1199 	    (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
1200 	     atomic_read(&xprt->xpt_reserved) == 0))
1201 		return 1;
1202 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1203 	return 0;
1204 }
1205 
1206 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1207 				       struct net *net,
1208 				       struct sockaddr *sa, int salen,
1209 				       int flags)
1210 {
1211 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1212 }
1213 
1214 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1215 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1216 					     struct net *, struct sockaddr *,
1217 					     int, int);
1218 static void svc_bc_sock_free(struct svc_xprt *xprt);
1219 
1220 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1221 				       struct net *net,
1222 				       struct sockaddr *sa, int salen,
1223 				       int flags)
1224 {
1225 	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1226 }
1227 
1228 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1229 {
1230 }
1231 
1232 static struct svc_xprt_ops svc_tcp_bc_ops = {
1233 	.xpo_create = svc_bc_tcp_create,
1234 	.xpo_detach = svc_bc_tcp_sock_detach,
1235 	.xpo_free = svc_bc_sock_free,
1236 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1237 };
1238 
1239 static struct svc_xprt_class svc_tcp_bc_class = {
1240 	.xcl_name = "tcp-bc",
1241 	.xcl_owner = THIS_MODULE,
1242 	.xcl_ops = &svc_tcp_bc_ops,
1243 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1244 };
1245 
1246 static void svc_init_bc_xprt_sock(void)
1247 {
1248 	svc_reg_xprt_class(&svc_tcp_bc_class);
1249 }
1250 
1251 static void svc_cleanup_bc_xprt_sock(void)
1252 {
1253 	svc_unreg_xprt_class(&svc_tcp_bc_class);
1254 }
1255 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1256 static void svc_init_bc_xprt_sock(void)
1257 {
1258 }
1259 
1260 static void svc_cleanup_bc_xprt_sock(void)
1261 {
1262 }
1263 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1264 
1265 static struct svc_xprt_ops svc_tcp_ops = {
1266 	.xpo_create = svc_tcp_create,
1267 	.xpo_recvfrom = svc_tcp_recvfrom,
1268 	.xpo_sendto = svc_tcp_sendto,
1269 	.xpo_release_rqst = svc_release_skb,
1270 	.xpo_detach = svc_tcp_sock_detach,
1271 	.xpo_free = svc_sock_free,
1272 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1273 	.xpo_has_wspace = svc_tcp_has_wspace,
1274 	.xpo_accept = svc_tcp_accept,
1275 };
1276 
1277 static struct svc_xprt_class svc_tcp_class = {
1278 	.xcl_name = "tcp",
1279 	.xcl_owner = THIS_MODULE,
1280 	.xcl_ops = &svc_tcp_ops,
1281 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1282 };
1283 
1284 void svc_init_xprt_sock(void)
1285 {
1286 	svc_reg_xprt_class(&svc_tcp_class);
1287 	svc_reg_xprt_class(&svc_udp_class);
1288 	svc_init_bc_xprt_sock();
1289 }
1290 
1291 void svc_cleanup_xprt_sock(void)
1292 {
1293 	svc_unreg_xprt_class(&svc_tcp_class);
1294 	svc_unreg_xprt_class(&svc_udp_class);
1295 	svc_cleanup_bc_xprt_sock();
1296 }
1297 
1298 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1299 {
1300 	struct sock	*sk = svsk->sk_sk;
1301 
1302 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1303 		      &svsk->sk_xprt, serv);
1304 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1305 	if (sk->sk_state == TCP_LISTEN) {
1306 		dprintk("setting up TCP socket for listening\n");
1307 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1308 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1309 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1310 	} else {
1311 		dprintk("setting up TCP socket for reading\n");
1312 		sk->sk_state_change = svc_tcp_state_change;
1313 		sk->sk_data_ready = svc_tcp_data_ready;
1314 		sk->sk_write_space = svc_tcp_write_space;
1315 
1316 		svsk->sk_reclen = 0;
1317 		svsk->sk_tcplen = 0;
1318 		svsk->sk_datalen = 0;
1319 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1320 
1321 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1322 
1323 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1324 		if (sk->sk_state != TCP_ESTABLISHED)
1325 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1326 	}
1327 }
1328 
1329 void svc_sock_update_bufs(struct svc_serv *serv)
1330 {
1331 	/*
1332 	 * The number of server threads has changed. Update
1333 	 * rcvbuf and sndbuf accordingly on all sockets
1334 	 */
1335 	struct svc_sock *svsk;
1336 
1337 	spin_lock_bh(&serv->sv_lock);
1338 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1339 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1340 	spin_unlock_bh(&serv->sv_lock);
1341 }
1342 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1343 
1344 /*
1345  * Initialize socket for RPC use and create svc_sock struct
1346  * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1347  */
1348 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1349 						struct socket *sock,
1350 						int flags)
1351 {
1352 	struct svc_sock	*svsk;
1353 	struct sock	*inet;
1354 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1355 	int		err = 0;
1356 
1357 	dprintk("svc: svc_setup_socket %p\n", sock);
1358 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1359 	if (!svsk)
1360 		return ERR_PTR(-ENOMEM);
1361 
1362 	inet = sock->sk;
1363 
1364 	/* Register socket with portmapper */
1365 	if (pmap_register)
1366 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1367 				     inet->sk_protocol,
1368 				     ntohs(inet_sk(inet)->inet_sport));
1369 
1370 	if (err < 0) {
1371 		kfree(svsk);
1372 		return ERR_PTR(err);
1373 	}
1374 
1375 	inet->sk_user_data = svsk;
1376 	svsk->sk_sock = sock;
1377 	svsk->sk_sk = inet;
1378 	svsk->sk_ostate = inet->sk_state_change;
1379 	svsk->sk_odata = inet->sk_data_ready;
1380 	svsk->sk_owspace = inet->sk_write_space;
1381 
1382 	/* Initialize the socket */
1383 	if (sock->type == SOCK_DGRAM)
1384 		svc_udp_init(svsk, serv);
1385 	else {
1386 		/* initialise setting must have enough space to
1387 		 * receive and respond to one request.
1388 		 */
1389 		svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1390 					4 * serv->sv_max_mesg);
1391 		svc_tcp_init(svsk, serv);
1392 	}
1393 
1394 	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1395 				svsk, svsk->sk_sk);
1396 
1397 	return svsk;
1398 }
1399 
1400 /**
1401  * svc_addsock - add a listener socket to an RPC service
1402  * @serv: pointer to RPC service to which to add a new listener
1403  * @fd: file descriptor of the new listener
1404  * @name_return: pointer to buffer to fill in with name of listener
1405  * @len: size of the buffer
1406  *
1407  * Fills in socket name and returns positive length of name if successful.
1408  * Name is terminated with '\n'.  On error, returns a negative errno
1409  * value.
1410  */
1411 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1412 		const size_t len)
1413 {
1414 	int err = 0;
1415 	struct socket *so = sockfd_lookup(fd, &err);
1416 	struct svc_sock *svsk = NULL;
1417 	struct sockaddr_storage addr;
1418 	struct sockaddr *sin = (struct sockaddr *)&addr;
1419 	int salen;
1420 
1421 	if (!so)
1422 		return err;
1423 	err = -EAFNOSUPPORT;
1424 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1425 		goto out;
1426 	err =  -EPROTONOSUPPORT;
1427 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1428 	    so->sk->sk_protocol != IPPROTO_UDP)
1429 		goto out;
1430 	err = -EISCONN;
1431 	if (so->state > SS_UNCONNECTED)
1432 		goto out;
1433 	err = -ENOENT;
1434 	if (!try_module_get(THIS_MODULE))
1435 		goto out;
1436 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1437 	if (IS_ERR(svsk)) {
1438 		module_put(THIS_MODULE);
1439 		err = PTR_ERR(svsk);
1440 		goto out;
1441 	}
1442 	if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1443 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1444 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1445 	return svc_one_sock_name(svsk, name_return, len);
1446 out:
1447 	sockfd_put(so);
1448 	return err;
1449 }
1450 EXPORT_SYMBOL_GPL(svc_addsock);
1451 
1452 /*
1453  * Create socket for RPC service.
1454  */
1455 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1456 					  int protocol,
1457 					  struct net *net,
1458 					  struct sockaddr *sin, int len,
1459 					  int flags)
1460 {
1461 	struct svc_sock	*svsk;
1462 	struct socket	*sock;
1463 	int		error;
1464 	int		type;
1465 	struct sockaddr_storage addr;
1466 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1467 	int		newlen;
1468 	int		family;
1469 	int		val;
1470 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1471 
1472 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1473 			serv->sv_program->pg_name, protocol,
1474 			__svc_print_addr(sin, buf, sizeof(buf)));
1475 
1476 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1477 		printk(KERN_WARNING "svc: only UDP and TCP "
1478 				"sockets supported\n");
1479 		return ERR_PTR(-EINVAL);
1480 	}
1481 
1482 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1483 	switch (sin->sa_family) {
1484 	case AF_INET6:
1485 		family = PF_INET6;
1486 		break;
1487 	case AF_INET:
1488 		family = PF_INET;
1489 		break;
1490 	default:
1491 		return ERR_PTR(-EINVAL);
1492 	}
1493 
1494 	error = __sock_create(net, family, type, protocol, &sock, 1);
1495 	if (error < 0)
1496 		return ERR_PTR(error);
1497 
1498 	svc_reclassify_socket(sock);
1499 
1500 	/*
1501 	 * If this is an PF_INET6 listener, we want to avoid
1502 	 * getting requests from IPv4 remotes.  Those should
1503 	 * be shunted to a PF_INET listener via rpcbind.
1504 	 */
1505 	val = 1;
1506 	if (family == PF_INET6)
1507 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1508 					(char *)&val, sizeof(val));
1509 
1510 	if (type == SOCK_STREAM)
1511 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1512 	error = kernel_bind(sock, sin, len);
1513 	if (error < 0)
1514 		goto bummer;
1515 
1516 	newlen = len;
1517 	error = kernel_getsockname(sock, newsin, &newlen);
1518 	if (error < 0)
1519 		goto bummer;
1520 
1521 	if (protocol == IPPROTO_TCP) {
1522 		if ((error = kernel_listen(sock, 64)) < 0)
1523 			goto bummer;
1524 	}
1525 
1526 	svsk = svc_setup_socket(serv, sock, flags);
1527 	if (IS_ERR(svsk)) {
1528 		error = PTR_ERR(svsk);
1529 		goto bummer;
1530 	}
1531 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1532 	return (struct svc_xprt *)svsk;
1533 bummer:
1534 	dprintk("svc: svc_create_socket error = %d\n", -error);
1535 	sock_release(sock);
1536 	return ERR_PTR(error);
1537 }
1538 
1539 /*
1540  * Detach the svc_sock from the socket so that no
1541  * more callbacks occur.
1542  */
1543 static void svc_sock_detach(struct svc_xprt *xprt)
1544 {
1545 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1546 	struct sock *sk = svsk->sk_sk;
1547 	wait_queue_head_t *wq;
1548 
1549 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1550 
1551 	/* put back the old socket callbacks */
1552 	sk->sk_state_change = svsk->sk_ostate;
1553 	sk->sk_data_ready = svsk->sk_odata;
1554 	sk->sk_write_space = svsk->sk_owspace;
1555 
1556 	wq = sk_sleep(sk);
1557 	if (wq && waitqueue_active(wq))
1558 		wake_up_interruptible(wq);
1559 }
1560 
1561 /*
1562  * Disconnect the socket, and reset the callbacks
1563  */
1564 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1565 {
1566 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1567 
1568 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1569 
1570 	svc_sock_detach(xprt);
1571 
1572 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1573 		svc_tcp_clear_pages(svsk);
1574 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1575 	}
1576 }
1577 
1578 /*
1579  * Free the svc_sock's socket resources and the svc_sock itself.
1580  */
1581 static void svc_sock_free(struct svc_xprt *xprt)
1582 {
1583 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1584 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1585 
1586 	if (svsk->sk_sock->file)
1587 		sockfd_put(svsk->sk_sock);
1588 	else
1589 		sock_release(svsk->sk_sock);
1590 	kfree(svsk);
1591 }
1592 
1593 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1594 /*
1595  * Create a back channel svc_xprt which shares the fore channel socket.
1596  */
1597 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1598 					     int protocol,
1599 					     struct net *net,
1600 					     struct sockaddr *sin, int len,
1601 					     int flags)
1602 {
1603 	struct svc_sock *svsk;
1604 	struct svc_xprt *xprt;
1605 
1606 	if (protocol != IPPROTO_TCP) {
1607 		printk(KERN_WARNING "svc: only TCP sockets"
1608 			" supported on shared back channel\n");
1609 		return ERR_PTR(-EINVAL);
1610 	}
1611 
1612 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1613 	if (!svsk)
1614 		return ERR_PTR(-ENOMEM);
1615 
1616 	xprt = &svsk->sk_xprt;
1617 	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1618 
1619 	serv->sv_bc_xprt = xprt;
1620 
1621 	return xprt;
1622 }
1623 
1624 /*
1625  * Free a back channel svc_sock.
1626  */
1627 static void svc_bc_sock_free(struct svc_xprt *xprt)
1628 {
1629 	if (xprt)
1630 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1631 }
1632 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1633