xref: /linux/net/sunrpc/svcsock.c (revision f69e98a91a01fd7c5755dd710e94a17d6e9f583f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/net/sunrpc/svcsock.c
4  *
5  * These are the RPC server socket internals.
6  *
7  * The server scheduling algorithm does not always distribute the load
8  * evenly when servicing a single client. May need to modify the
9  * svc_xprt_enqueue procedure...
10  *
11  * TCP support is largely untested and may be a little slow. The problem
12  * is that we currently do two separate recvfrom's, one for the 4-byte
13  * record length, and the second for the actual record. This could possibly
14  * be improved by always reading a minimum size of around 100 bytes and
15  * tucking any superfluous bytes away in a temporary store. Still, that
16  * leaves write requests out in the rain. An alternative may be to peek at
17  * the first skb in the queue, and if it matches the next TCP sequence
18  * number, to extract the record marker. Yuck.
19  *
20  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
21  */
22 
23 #include <linux/kernel.h>
24 #include <linux/sched.h>
25 #include <linux/module.h>
26 #include <linux/errno.h>
27 #include <linux/fcntl.h>
28 #include <linux/net.h>
29 #include <linux/in.h>
30 #include <linux/inet.h>
31 #include <linux/udp.h>
32 #include <linux/tcp.h>
33 #include <linux/unistd.h>
34 #include <linux/slab.h>
35 #include <linux/netdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/file.h>
38 #include <linux/freezer.h>
39 #include <net/sock.h>
40 #include <net/checksum.h>
41 #include <net/ip.h>
42 #include <net/ipv6.h>
43 #include <net/udp.h>
44 #include <net/tcp.h>
45 #include <net/tcp_states.h>
46 #include <linux/uaccess.h>
47 #include <linux/highmem.h>
48 #include <asm/ioctls.h>
49 
50 #include <linux/sunrpc/types.h>
51 #include <linux/sunrpc/clnt.h>
52 #include <linux/sunrpc/xdr.h>
53 #include <linux/sunrpc/msg_prot.h>
54 #include <linux/sunrpc/svcsock.h>
55 #include <linux/sunrpc/stats.h>
56 #include <linux/sunrpc/xprt.h>
57 
58 #include <trace/events/sunrpc.h>
59 
60 #include "socklib.h"
61 #include "sunrpc.h"
62 
63 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
64 
65 
66 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
67 					 int flags);
68 static int		svc_udp_recvfrom(struct svc_rqst *);
69 static int		svc_udp_sendto(struct svc_rqst *);
70 static void		svc_sock_detach(struct svc_xprt *);
71 static void		svc_tcp_sock_detach(struct svc_xprt *);
72 static void		svc_sock_free(struct svc_xprt *);
73 
74 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
75 					  struct net *, struct sockaddr *,
76 					  int, int);
77 #ifdef CONFIG_DEBUG_LOCK_ALLOC
78 static struct lock_class_key svc_key[2];
79 static struct lock_class_key svc_slock_key[2];
80 
81 static void svc_reclassify_socket(struct socket *sock)
82 {
83 	struct sock *sk = sock->sk;
84 
85 	if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
86 		return;
87 
88 	switch (sk->sk_family) {
89 	case AF_INET:
90 		sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
91 					      &svc_slock_key[0],
92 					      "sk_xprt.xpt_lock-AF_INET-NFSD",
93 					      &svc_key[0]);
94 		break;
95 
96 	case AF_INET6:
97 		sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
98 					      &svc_slock_key[1],
99 					      "sk_xprt.xpt_lock-AF_INET6-NFSD",
100 					      &svc_key[1]);
101 		break;
102 
103 	default:
104 		BUG();
105 	}
106 }
107 #else
108 static void svc_reclassify_socket(struct socket *sock)
109 {
110 }
111 #endif
112 
113 /**
114  * svc_tcp_release_rqst - Release transport-related resources
115  * @rqstp: request structure with resources to be released
116  *
117  */
118 static void svc_tcp_release_rqst(struct svc_rqst *rqstp)
119 {
120 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
121 
122 	if (skb) {
123 		struct svc_sock *svsk =
124 			container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
125 
126 		rqstp->rq_xprt_ctxt = NULL;
127 		skb_free_datagram_locked(svsk->sk_sk, skb);
128 	}
129 }
130 
131 /**
132  * svc_udp_release_rqst - Release transport-related resources
133  * @rqstp: request structure with resources to be released
134  *
135  */
136 static void svc_udp_release_rqst(struct svc_rqst *rqstp)
137 {
138 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
139 
140 	if (skb) {
141 		rqstp->rq_xprt_ctxt = NULL;
142 		consume_skb(skb);
143 	}
144 }
145 
146 union svc_pktinfo_u {
147 	struct in_pktinfo pkti;
148 	struct in6_pktinfo pkti6;
149 };
150 #define SVC_PKTINFO_SPACE \
151 	CMSG_SPACE(sizeof(union svc_pktinfo_u))
152 
153 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
154 {
155 	struct svc_sock *svsk =
156 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
157 	switch (svsk->sk_sk->sk_family) {
158 	case AF_INET: {
159 			struct in_pktinfo *pki = CMSG_DATA(cmh);
160 
161 			cmh->cmsg_level = SOL_IP;
162 			cmh->cmsg_type = IP_PKTINFO;
163 			pki->ipi_ifindex = 0;
164 			pki->ipi_spec_dst.s_addr =
165 				 svc_daddr_in(rqstp)->sin_addr.s_addr;
166 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 		}
168 		break;
169 
170 	case AF_INET6: {
171 			struct in6_pktinfo *pki = CMSG_DATA(cmh);
172 			struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
173 
174 			cmh->cmsg_level = SOL_IPV6;
175 			cmh->cmsg_type = IPV6_PKTINFO;
176 			pki->ipi6_ifindex = daddr->sin6_scope_id;
177 			pki->ipi6_addr = daddr->sin6_addr;
178 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
179 		}
180 		break;
181 	}
182 }
183 
184 static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
185 				   unsigned int length)
186 {
187 	return 0;
188 }
189 
190 /*
191  * Report socket names for nfsdfs
192  */
193 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
194 {
195 	const struct sock *sk = svsk->sk_sk;
196 	const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
197 							"udp" : "tcp";
198 	int len;
199 
200 	switch (sk->sk_family) {
201 	case PF_INET:
202 		len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
203 				proto_name,
204 				&inet_sk(sk)->inet_rcv_saddr,
205 				inet_sk(sk)->inet_num);
206 		break;
207 #if IS_ENABLED(CONFIG_IPV6)
208 	case PF_INET6:
209 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
210 				proto_name,
211 				&sk->sk_v6_rcv_saddr,
212 				inet_sk(sk)->inet_num);
213 		break;
214 #endif
215 	default:
216 		len = snprintf(buf, remaining, "*unknown-%d*\n",
217 				sk->sk_family);
218 	}
219 
220 	if (len >= remaining) {
221 		*buf = '\0';
222 		return -ENAMETOOLONG;
223 	}
224 	return len;
225 }
226 
227 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
228 static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
229 {
230 	struct bvec_iter bi = {
231 		.bi_size	= size + seek,
232 	};
233 	struct bio_vec bv;
234 
235 	bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
236 	for_each_bvec(bv, bvec, bi, bi)
237 		flush_dcache_page(bv.bv_page);
238 }
239 #else
240 static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
241 				  size_t seek)
242 {
243 }
244 #endif
245 
246 /*
247  * Read from @rqstp's transport socket. The incoming message fills whole
248  * pages in @rqstp's rq_pages array until the last page of the message
249  * has been received into a partial page.
250  */
251 static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
252 				size_t seek)
253 {
254 	struct svc_sock *svsk =
255 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
256 	struct bio_vec *bvec = rqstp->rq_bvec;
257 	struct msghdr msg = { NULL };
258 	unsigned int i;
259 	ssize_t len;
260 	size_t t;
261 
262 	rqstp->rq_xprt_hlen = 0;
263 
264 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
265 
266 	for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) {
267 		bvec[i].bv_page = rqstp->rq_pages[i];
268 		bvec[i].bv_len = PAGE_SIZE;
269 		bvec[i].bv_offset = 0;
270 	}
271 	rqstp->rq_respages = &rqstp->rq_pages[i];
272 	rqstp->rq_next_page = rqstp->rq_respages + 1;
273 
274 	iov_iter_bvec(&msg.msg_iter, READ, bvec, i, buflen);
275 	if (seek) {
276 		iov_iter_advance(&msg.msg_iter, seek);
277 		buflen -= seek;
278 	}
279 	len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
280 	if (len > 0)
281 		svc_flush_bvec(bvec, len, seek);
282 
283 	/* If we read a full record, then assume there may be more
284 	 * data to read (stream based sockets only!)
285 	 */
286 	if (len == buflen)
287 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
288 
289 	return len;
290 }
291 
292 /*
293  * Set socket snd and rcv buffer lengths
294  */
295 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
296 {
297 	unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
298 	struct socket *sock = svsk->sk_sock;
299 
300 	nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
301 
302 	lock_sock(sock->sk);
303 	sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
304 	sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
305 	sock->sk->sk_write_space(sock->sk);
306 	release_sock(sock->sk);
307 }
308 
309 static void svc_sock_secure_port(struct svc_rqst *rqstp)
310 {
311 	if (svc_port_is_privileged(svc_addr(rqstp)))
312 		set_bit(RQ_SECURE, &rqstp->rq_flags);
313 	else
314 		clear_bit(RQ_SECURE, &rqstp->rq_flags);
315 }
316 
317 /*
318  * INET callback when data has been received on the socket.
319  */
320 static void svc_data_ready(struct sock *sk)
321 {
322 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
323 
324 	if (svsk) {
325 		/* Refer to svc_setup_socket() for details. */
326 		rmb();
327 		svsk->sk_odata(sk);
328 		trace_svcsock_data_ready(&svsk->sk_xprt, 0);
329 		if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
330 			svc_xprt_enqueue(&svsk->sk_xprt);
331 	}
332 }
333 
334 /*
335  * INET callback when space is newly available on the socket.
336  */
337 static void svc_write_space(struct sock *sk)
338 {
339 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
340 
341 	if (svsk) {
342 		/* Refer to svc_setup_socket() for details. */
343 		rmb();
344 		trace_svcsock_write_space(&svsk->sk_xprt, 0);
345 		svsk->sk_owspace(sk);
346 		svc_xprt_enqueue(&svsk->sk_xprt);
347 	}
348 }
349 
350 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
351 {
352 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
353 
354 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
355 		return 1;
356 	return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
357 }
358 
359 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
360 {
361 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
362 
363 	sock_no_linger(svsk->sk_sock->sk);
364 }
365 
366 /*
367  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
368  */
369 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
370 				     struct cmsghdr *cmh)
371 {
372 	struct in_pktinfo *pki = CMSG_DATA(cmh);
373 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
374 
375 	if (cmh->cmsg_type != IP_PKTINFO)
376 		return 0;
377 
378 	daddr->sin_family = AF_INET;
379 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
380 	return 1;
381 }
382 
383 /*
384  * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
385  */
386 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
387 				     struct cmsghdr *cmh)
388 {
389 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
390 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
391 
392 	if (cmh->cmsg_type != IPV6_PKTINFO)
393 		return 0;
394 
395 	daddr->sin6_family = AF_INET6;
396 	daddr->sin6_addr = pki->ipi6_addr;
397 	daddr->sin6_scope_id = pki->ipi6_ifindex;
398 	return 1;
399 }
400 
401 /*
402  * Copy the UDP datagram's destination address to the rqstp structure.
403  * The 'destination' address in this case is the address to which the
404  * peer sent the datagram, i.e. our local address. For multihomed
405  * hosts, this can change from msg to msg. Note that only the IP
406  * address changes, the port number should remain the same.
407  */
408 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
409 				    struct cmsghdr *cmh)
410 {
411 	switch (cmh->cmsg_level) {
412 	case SOL_IP:
413 		return svc_udp_get_dest_address4(rqstp, cmh);
414 	case SOL_IPV6:
415 		return svc_udp_get_dest_address6(rqstp, cmh);
416 	}
417 
418 	return 0;
419 }
420 
421 /**
422  * svc_udp_recvfrom - Receive a datagram from a UDP socket.
423  * @rqstp: request structure into which to receive an RPC Call
424  *
425  * Called in a loop when XPT_DATA has been set.
426  *
427  * Returns:
428  *   On success, the number of bytes in a received RPC Call, or
429  *   %0 if a complete RPC Call message was not ready to return
430  */
431 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
432 {
433 	struct svc_sock	*svsk =
434 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
435 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
436 	struct sk_buff	*skb;
437 	union {
438 		struct cmsghdr	hdr;
439 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
440 	} buffer;
441 	struct cmsghdr *cmh = &buffer.hdr;
442 	struct msghdr msg = {
443 		.msg_name = svc_addr(rqstp),
444 		.msg_control = cmh,
445 		.msg_controllen = sizeof(buffer),
446 		.msg_flags = MSG_DONTWAIT,
447 	};
448 	size_t len;
449 	int err;
450 
451 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
452 	    /* udp sockets need large rcvbuf as all pending
453 	     * requests are still in that buffer.  sndbuf must
454 	     * also be large enough that there is enough space
455 	     * for one reply per thread.  We count all threads
456 	     * rather than threads in a particular pool, which
457 	     * provides an upper bound on the number of threads
458 	     * which will access the socket.
459 	     */
460 	    svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
461 
462 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
463 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
464 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
465 	if (err < 0)
466 		goto out_recv_err;
467 	skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
468 	if (!skb)
469 		goto out_recv_err;
470 
471 	len = svc_addr_len(svc_addr(rqstp));
472 	rqstp->rq_addrlen = len;
473 	if (skb->tstamp == 0) {
474 		skb->tstamp = ktime_get_real();
475 		/* Don't enable netstamp, sunrpc doesn't
476 		   need that much accuracy */
477 	}
478 	sock_write_timestamp(svsk->sk_sk, skb->tstamp);
479 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
480 
481 	len = skb->len;
482 	rqstp->rq_arg.len = len;
483 	trace_svcsock_udp_recv(&svsk->sk_xprt, len);
484 
485 	rqstp->rq_prot = IPPROTO_UDP;
486 
487 	if (!svc_udp_get_dest_address(rqstp, cmh))
488 		goto out_cmsg_err;
489 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
490 
491 	if (skb_is_nonlinear(skb)) {
492 		/* we have to copy */
493 		local_bh_disable();
494 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
495 			goto out_bh_enable;
496 		local_bh_enable();
497 		consume_skb(skb);
498 	} else {
499 		/* we can use it in-place */
500 		rqstp->rq_arg.head[0].iov_base = skb->data;
501 		rqstp->rq_arg.head[0].iov_len = len;
502 		if (skb_checksum_complete(skb))
503 			goto out_free;
504 		rqstp->rq_xprt_ctxt = skb;
505 	}
506 
507 	rqstp->rq_arg.page_base = 0;
508 	if (len <= rqstp->rq_arg.head[0].iov_len) {
509 		rqstp->rq_arg.head[0].iov_len = len;
510 		rqstp->rq_arg.page_len = 0;
511 		rqstp->rq_respages = rqstp->rq_pages+1;
512 	} else {
513 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
514 		rqstp->rq_respages = rqstp->rq_pages + 1 +
515 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
516 	}
517 	rqstp->rq_next_page = rqstp->rq_respages+1;
518 
519 	if (serv->sv_stats)
520 		serv->sv_stats->netudpcnt++;
521 
522 	svc_xprt_received(rqstp->rq_xprt);
523 	return len;
524 
525 out_recv_err:
526 	if (err != -EAGAIN) {
527 		/* possibly an icmp error */
528 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
529 	}
530 	trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
531 	goto out_clear_busy;
532 out_cmsg_err:
533 	net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
534 			     cmh->cmsg_level, cmh->cmsg_type);
535 	goto out_free;
536 out_bh_enable:
537 	local_bh_enable();
538 out_free:
539 	kfree_skb(skb);
540 out_clear_busy:
541 	svc_xprt_received(rqstp->rq_xprt);
542 	return 0;
543 }
544 
545 /**
546  * svc_udp_sendto - Send out a reply on a UDP socket
547  * @rqstp: completed svc_rqst
548  *
549  * xpt_mutex ensures @rqstp's whole message is written to the socket
550  * without interruption.
551  *
552  * Returns the number of bytes sent, or a negative errno.
553  */
554 static int svc_udp_sendto(struct svc_rqst *rqstp)
555 {
556 	struct svc_xprt *xprt = rqstp->rq_xprt;
557 	struct svc_sock	*svsk = container_of(xprt, struct svc_sock, sk_xprt);
558 	struct xdr_buf *xdr = &rqstp->rq_res;
559 	union {
560 		struct cmsghdr	hdr;
561 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
562 	} buffer;
563 	struct cmsghdr *cmh = &buffer.hdr;
564 	struct msghdr msg = {
565 		.msg_name	= &rqstp->rq_addr,
566 		.msg_namelen	= rqstp->rq_addrlen,
567 		.msg_control	= cmh,
568 		.msg_controllen	= sizeof(buffer),
569 	};
570 	unsigned int sent;
571 	int err;
572 
573 	svc_udp_release_rqst(rqstp);
574 
575 	svc_set_cmsg_data(rqstp, cmh);
576 
577 	mutex_lock(&xprt->xpt_mutex);
578 
579 	if (svc_xprt_is_dead(xprt))
580 		goto out_notconn;
581 
582 	err = xdr_alloc_bvec(xdr, GFP_KERNEL);
583 	if (err < 0)
584 		goto out_unlock;
585 
586 	err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
587 	if (err == -ECONNREFUSED) {
588 		/* ICMP error on earlier request. */
589 		err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
590 	}
591 	xdr_free_bvec(xdr);
592 	trace_svcsock_udp_send(xprt, err);
593 out_unlock:
594 	mutex_unlock(&xprt->xpt_mutex);
595 	if (err < 0)
596 		return err;
597 	return sent;
598 
599 out_notconn:
600 	mutex_unlock(&xprt->xpt_mutex);
601 	return -ENOTCONN;
602 }
603 
604 static int svc_udp_has_wspace(struct svc_xprt *xprt)
605 {
606 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
607 	struct svc_serv	*serv = xprt->xpt_server;
608 	unsigned long required;
609 
610 	/*
611 	 * Set the SOCK_NOSPACE flag before checking the available
612 	 * sock space.
613 	 */
614 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
615 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
616 	if (required*2 > sock_wspace(svsk->sk_sk))
617 		return 0;
618 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
619 	return 1;
620 }
621 
622 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
623 {
624 	BUG();
625 	return NULL;
626 }
627 
628 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
629 {
630 }
631 
632 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
633 				       struct net *net,
634 				       struct sockaddr *sa, int salen,
635 				       int flags)
636 {
637 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
638 }
639 
640 static const struct svc_xprt_ops svc_udp_ops = {
641 	.xpo_create = svc_udp_create,
642 	.xpo_recvfrom = svc_udp_recvfrom,
643 	.xpo_sendto = svc_udp_sendto,
644 	.xpo_result_payload = svc_sock_result_payload,
645 	.xpo_release_rqst = svc_udp_release_rqst,
646 	.xpo_detach = svc_sock_detach,
647 	.xpo_free = svc_sock_free,
648 	.xpo_has_wspace = svc_udp_has_wspace,
649 	.xpo_accept = svc_udp_accept,
650 	.xpo_secure_port = svc_sock_secure_port,
651 	.xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
652 };
653 
654 static struct svc_xprt_class svc_udp_class = {
655 	.xcl_name = "udp",
656 	.xcl_owner = THIS_MODULE,
657 	.xcl_ops = &svc_udp_ops,
658 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
659 	.xcl_ident = XPRT_TRANSPORT_UDP,
660 };
661 
662 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
663 {
664 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
665 		      &svsk->sk_xprt, serv);
666 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
667 	svsk->sk_sk->sk_data_ready = svc_data_ready;
668 	svsk->sk_sk->sk_write_space = svc_write_space;
669 
670 	/* initialise setting must have enough space to
671 	 * receive and respond to one request.
672 	 * svc_udp_recvfrom will re-adjust if necessary
673 	 */
674 	svc_sock_setbufsize(svsk, 3);
675 
676 	/* data might have come in before data_ready set up */
677 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
678 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
679 
680 	/* make sure we get destination address info */
681 	switch (svsk->sk_sk->sk_family) {
682 	case AF_INET:
683 		ip_sock_set_pktinfo(svsk->sk_sock->sk);
684 		break;
685 	case AF_INET6:
686 		ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
687 		break;
688 	default:
689 		BUG();
690 	}
691 }
692 
693 /*
694  * A data_ready event on a listening socket means there's a connection
695  * pending. Do not use state_change as a substitute for it.
696  */
697 static void svc_tcp_listen_data_ready(struct sock *sk)
698 {
699 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
700 
701 	if (svsk) {
702 		/* Refer to svc_setup_socket() for details. */
703 		rmb();
704 		svsk->sk_odata(sk);
705 	}
706 
707 	/*
708 	 * This callback may called twice when a new connection
709 	 * is established as a child socket inherits everything
710 	 * from a parent LISTEN socket.
711 	 * 1) data_ready method of the parent socket will be called
712 	 *    when one of child sockets become ESTABLISHED.
713 	 * 2) data_ready method of the child socket may be called
714 	 *    when it receives data before the socket is accepted.
715 	 * In case of 2, we should ignore it silently.
716 	 */
717 	if (sk->sk_state == TCP_LISTEN) {
718 		if (svsk) {
719 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
720 			svc_xprt_enqueue(&svsk->sk_xprt);
721 		}
722 	}
723 }
724 
725 /*
726  * A state change on a connected socket means it's dying or dead.
727  */
728 static void svc_tcp_state_change(struct sock *sk)
729 {
730 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
731 
732 	if (svsk) {
733 		/* Refer to svc_setup_socket() for details. */
734 		rmb();
735 		svsk->sk_ostate(sk);
736 		trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
737 		if (sk->sk_state != TCP_ESTABLISHED)
738 			svc_xprt_deferred_close(&svsk->sk_xprt);
739 	}
740 }
741 
742 /*
743  * Accept a TCP connection
744  */
745 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
746 {
747 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
748 	struct sockaddr_storage addr;
749 	struct sockaddr	*sin = (struct sockaddr *) &addr;
750 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
751 	struct socket	*sock = svsk->sk_sock;
752 	struct socket	*newsock;
753 	struct svc_sock	*newsvsk;
754 	int		err, slen;
755 
756 	if (!sock)
757 		return NULL;
758 
759 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
760 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
761 	if (err < 0) {
762 		if (err == -ENOMEM)
763 			printk(KERN_WARNING "%s: no more sockets!\n",
764 			       serv->sv_name);
765 		else if (err != -EAGAIN)
766 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
767 					     serv->sv_name, -err);
768 		trace_svcsock_accept_err(xprt, serv->sv_name, err);
769 		return NULL;
770 	}
771 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
772 
773 	err = kernel_getpeername(newsock, sin);
774 	if (err < 0) {
775 		trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
776 		goto failed;		/* aborted connection or whatever */
777 	}
778 	slen = err;
779 
780 	/* Reset the inherited callbacks before calling svc_setup_socket */
781 	newsock->sk->sk_state_change = svsk->sk_ostate;
782 	newsock->sk->sk_data_ready = svsk->sk_odata;
783 	newsock->sk->sk_write_space = svsk->sk_owspace;
784 
785 	/* make sure that a write doesn't block forever when
786 	 * low on memory
787 	 */
788 	newsock->sk->sk_sndtimeo = HZ*30;
789 
790 	newsvsk = svc_setup_socket(serv, newsock,
791 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
792 	if (IS_ERR(newsvsk))
793 		goto failed;
794 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
795 	err = kernel_getsockname(newsock, sin);
796 	slen = err;
797 	if (unlikely(err < 0))
798 		slen = offsetof(struct sockaddr, sa_data);
799 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
800 
801 	if (sock_is_loopback(newsock->sk))
802 		set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
803 	else
804 		clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
805 	if (serv->sv_stats)
806 		serv->sv_stats->nettcpconn++;
807 
808 	return &newsvsk->sk_xprt;
809 
810 failed:
811 	sock_release(newsock);
812 	return NULL;
813 }
814 
815 static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
816 				    struct svc_rqst *rqstp)
817 {
818 	size_t len = svsk->sk_datalen;
819 	unsigned int i, npages;
820 
821 	if (!len)
822 		return 0;
823 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
824 	for (i = 0; i < npages; i++) {
825 		if (rqstp->rq_pages[i] != NULL)
826 			put_page(rqstp->rq_pages[i]);
827 		BUG_ON(svsk->sk_pages[i] == NULL);
828 		rqstp->rq_pages[i] = svsk->sk_pages[i];
829 		svsk->sk_pages[i] = NULL;
830 	}
831 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
832 	return len;
833 }
834 
835 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
836 {
837 	unsigned int i, len, npages;
838 
839 	if (svsk->sk_datalen == 0)
840 		return;
841 	len = svsk->sk_datalen;
842 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
843 	for (i = 0; i < npages; i++) {
844 		svsk->sk_pages[i] = rqstp->rq_pages[i];
845 		rqstp->rq_pages[i] = NULL;
846 	}
847 }
848 
849 static void svc_tcp_clear_pages(struct svc_sock *svsk)
850 {
851 	unsigned int i, len, npages;
852 
853 	if (svsk->sk_datalen == 0)
854 		goto out;
855 	len = svsk->sk_datalen;
856 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
857 	for (i = 0; i < npages; i++) {
858 		if (svsk->sk_pages[i] == NULL) {
859 			WARN_ON_ONCE(1);
860 			continue;
861 		}
862 		put_page(svsk->sk_pages[i]);
863 		svsk->sk_pages[i] = NULL;
864 	}
865 out:
866 	svsk->sk_tcplen = 0;
867 	svsk->sk_datalen = 0;
868 }
869 
870 /*
871  * Receive fragment record header into sk_marker.
872  */
873 static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
874 				   struct svc_rqst *rqstp)
875 {
876 	ssize_t want, len;
877 
878 	/* If we haven't gotten the record length yet,
879 	 * get the next four bytes.
880 	 */
881 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
882 		struct msghdr	msg = { NULL };
883 		struct kvec	iov;
884 
885 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
886 		iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
887 		iov.iov_len  = want;
888 		iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, want);
889 		len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
890 		if (len < 0)
891 			return len;
892 		svsk->sk_tcplen += len;
893 		if (len < want) {
894 			/* call again to read the remaining bytes */
895 			goto err_short;
896 		}
897 		trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
898 		if (svc_sock_reclen(svsk) + svsk->sk_datalen >
899 		    svsk->sk_xprt.xpt_server->sv_max_mesg)
900 			goto err_too_large;
901 	}
902 	return svc_sock_reclen(svsk);
903 
904 err_too_large:
905 	net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
906 			       __func__, svsk->sk_xprt.xpt_server->sv_name,
907 			       svc_sock_reclen(svsk));
908 	svc_xprt_deferred_close(&svsk->sk_xprt);
909 err_short:
910 	return -EAGAIN;
911 }
912 
913 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
914 {
915 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
916 	struct rpc_rqst *req = NULL;
917 	struct kvec *src, *dst;
918 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
919 	__be32 xid;
920 	__be32 calldir;
921 
922 	xid = *p++;
923 	calldir = *p;
924 
925 	if (!bc_xprt)
926 		return -EAGAIN;
927 	spin_lock(&bc_xprt->queue_lock);
928 	req = xprt_lookup_rqst(bc_xprt, xid);
929 	if (!req)
930 		goto unlock_notfound;
931 
932 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
933 	/*
934 	 * XXX!: cheating for now!  Only copying HEAD.
935 	 * But we know this is good enough for now (in fact, for any
936 	 * callback reply in the forseeable future).
937 	 */
938 	dst = &req->rq_private_buf.head[0];
939 	src = &rqstp->rq_arg.head[0];
940 	if (dst->iov_len < src->iov_len)
941 		goto unlock_eagain; /* whatever; just giving up. */
942 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
943 	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
944 	rqstp->rq_arg.len = 0;
945 	spin_unlock(&bc_xprt->queue_lock);
946 	return 0;
947 unlock_notfound:
948 	printk(KERN_NOTICE
949 		"%s: Got unrecognized reply: "
950 		"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
951 		__func__, ntohl(calldir),
952 		bc_xprt, ntohl(xid));
953 unlock_eagain:
954 	spin_unlock(&bc_xprt->queue_lock);
955 	return -EAGAIN;
956 }
957 
958 static void svc_tcp_fragment_received(struct svc_sock *svsk)
959 {
960 	/* If we have more data, signal svc_xprt_enqueue() to try again */
961 	svsk->sk_tcplen = 0;
962 	svsk->sk_marker = xdr_zero;
963 }
964 
965 /**
966  * svc_tcp_recvfrom - Receive data from a TCP socket
967  * @rqstp: request structure into which to receive an RPC Call
968  *
969  * Called in a loop when XPT_DATA has been set.
970  *
971  * Read the 4-byte stream record marker, then use the record length
972  * in that marker to set up exactly the resources needed to receive
973  * the next RPC message into @rqstp.
974  *
975  * Returns:
976  *   On success, the number of bytes in a received RPC Call, or
977  *   %0 if a complete RPC Call message was not ready to return
978  *
979  * The zero return case handles partial receives and callback Replies.
980  * The state of a partial receive is preserved in the svc_sock for
981  * the next call to svc_tcp_recvfrom.
982  */
983 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
984 {
985 	struct svc_sock	*svsk =
986 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
987 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
988 	size_t want, base;
989 	ssize_t len;
990 	__be32 *p;
991 	__be32 calldir;
992 
993 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
994 	len = svc_tcp_read_marker(svsk, rqstp);
995 	if (len < 0)
996 		goto error;
997 
998 	base = svc_tcp_restore_pages(svsk, rqstp);
999 	want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1000 	len = svc_tcp_read_msg(rqstp, base + want, base);
1001 	if (len >= 0) {
1002 		trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1003 		svsk->sk_tcplen += len;
1004 		svsk->sk_datalen += len;
1005 	}
1006 	if (len != want || !svc_sock_final_rec(svsk))
1007 		goto err_incomplete;
1008 	if (svsk->sk_datalen < 8)
1009 		goto err_nuts;
1010 
1011 	rqstp->rq_arg.len = svsk->sk_datalen;
1012 	rqstp->rq_arg.page_base = 0;
1013 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1014 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1015 		rqstp->rq_arg.page_len = 0;
1016 	} else
1017 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1018 
1019 	rqstp->rq_xprt_ctxt   = NULL;
1020 	rqstp->rq_prot	      = IPPROTO_TCP;
1021 	if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1022 		set_bit(RQ_LOCAL, &rqstp->rq_flags);
1023 	else
1024 		clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1025 
1026 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1027 	calldir = p[1];
1028 	if (calldir)
1029 		len = receive_cb_reply(svsk, rqstp);
1030 
1031 	/* Reset TCP read info */
1032 	svsk->sk_datalen = 0;
1033 	svc_tcp_fragment_received(svsk);
1034 
1035 	if (len < 0)
1036 		goto error;
1037 
1038 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1039 	if (serv->sv_stats)
1040 		serv->sv_stats->nettcpcnt++;
1041 
1042 	svc_xprt_received(rqstp->rq_xprt);
1043 	return rqstp->rq_arg.len;
1044 
1045 err_incomplete:
1046 	svc_tcp_save_pages(svsk, rqstp);
1047 	if (len < 0 && len != -EAGAIN)
1048 		goto err_delete;
1049 	if (len == want)
1050 		svc_tcp_fragment_received(svsk);
1051 	else
1052 		trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1053 				svc_sock_reclen(svsk),
1054 				svsk->sk_tcplen - sizeof(rpc_fraghdr));
1055 	goto err_noclose;
1056 error:
1057 	if (len != -EAGAIN)
1058 		goto err_delete;
1059 	trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1060 	goto err_noclose;
1061 err_nuts:
1062 	svsk->sk_datalen = 0;
1063 err_delete:
1064 	trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1065 	svc_xprt_deferred_close(&svsk->sk_xprt);
1066 err_noclose:
1067 	svc_xprt_received(rqstp->rq_xprt);
1068 	return 0;	/* record not complete */
1069 }
1070 
1071 static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1072 			      int flags)
1073 {
1074 	return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1075 			       offset_in_page(vec->iov_base),
1076 			       vec->iov_len, flags);
1077 }
1078 
1079 /*
1080  * kernel_sendpage() is used exclusively to reduce the number of
1081  * copy operations in this path. Therefore the caller must ensure
1082  * that the pages backing @xdr are unchanging.
1083  *
1084  * In addition, the logic assumes that * .bv_len is never larger
1085  * than PAGE_SIZE.
1086  */
1087 static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
1088 			   rpc_fraghdr marker, unsigned int *sentp)
1089 {
1090 	const struct kvec *head = xdr->head;
1091 	const struct kvec *tail = xdr->tail;
1092 	struct kvec rm = {
1093 		.iov_base	= &marker,
1094 		.iov_len	= sizeof(marker),
1095 	};
1096 	struct msghdr msg = {
1097 		.msg_flags	= 0,
1098 	};
1099 	int ret;
1100 
1101 	*sentp = 0;
1102 	ret = xdr_alloc_bvec(xdr, GFP_KERNEL);
1103 	if (ret < 0)
1104 		return ret;
1105 
1106 	ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len);
1107 	if (ret < 0)
1108 		return ret;
1109 	*sentp += ret;
1110 	if (ret != rm.iov_len)
1111 		return -EAGAIN;
1112 
1113 	ret = svc_tcp_send_kvec(sock, head, 0);
1114 	if (ret < 0)
1115 		return ret;
1116 	*sentp += ret;
1117 	if (ret != head->iov_len)
1118 		goto out;
1119 
1120 	if (xdr->page_len) {
1121 		unsigned int offset, len, remaining;
1122 		struct bio_vec *bvec;
1123 
1124 		bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1125 		offset = offset_in_page(xdr->page_base);
1126 		remaining = xdr->page_len;
1127 		while (remaining > 0) {
1128 			len = min(remaining, bvec->bv_len - offset);
1129 			ret = kernel_sendpage(sock, bvec->bv_page,
1130 					      bvec->bv_offset + offset,
1131 					      len, 0);
1132 			if (ret < 0)
1133 				return ret;
1134 			*sentp += ret;
1135 			if (ret != len)
1136 				goto out;
1137 			remaining -= len;
1138 			offset = 0;
1139 			bvec++;
1140 		}
1141 	}
1142 
1143 	if (tail->iov_len) {
1144 		ret = svc_tcp_send_kvec(sock, tail, 0);
1145 		if (ret < 0)
1146 			return ret;
1147 		*sentp += ret;
1148 	}
1149 
1150 out:
1151 	return 0;
1152 }
1153 
1154 /**
1155  * svc_tcp_sendto - Send out a reply on a TCP socket
1156  * @rqstp: completed svc_rqst
1157  *
1158  * xpt_mutex ensures @rqstp's whole message is written to the socket
1159  * without interruption.
1160  *
1161  * Returns the number of bytes sent, or a negative errno.
1162  */
1163 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1164 {
1165 	struct svc_xprt *xprt = rqstp->rq_xprt;
1166 	struct svc_sock	*svsk = container_of(xprt, struct svc_sock, sk_xprt);
1167 	struct xdr_buf *xdr = &rqstp->rq_res;
1168 	rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1169 					 (u32)xdr->len);
1170 	unsigned int sent;
1171 	int err;
1172 
1173 	svc_tcp_release_rqst(rqstp);
1174 
1175 	atomic_inc(&svsk->sk_sendqlen);
1176 	mutex_lock(&xprt->xpt_mutex);
1177 	if (svc_xprt_is_dead(xprt))
1178 		goto out_notconn;
1179 	tcp_sock_set_cork(svsk->sk_sk, true);
1180 	err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent);
1181 	xdr_free_bvec(xdr);
1182 	trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1183 	if (err < 0 || sent != (xdr->len + sizeof(marker)))
1184 		goto out_close;
1185 	if (atomic_dec_and_test(&svsk->sk_sendqlen))
1186 		tcp_sock_set_cork(svsk->sk_sk, false);
1187 	mutex_unlock(&xprt->xpt_mutex);
1188 	return sent;
1189 
1190 out_notconn:
1191 	atomic_dec(&svsk->sk_sendqlen);
1192 	mutex_unlock(&xprt->xpt_mutex);
1193 	return -ENOTCONN;
1194 out_close:
1195 	pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1196 		  xprt->xpt_server->sv_name,
1197 		  (err < 0) ? "got error" : "sent",
1198 		  (err < 0) ? err : sent, xdr->len);
1199 	svc_xprt_deferred_close(xprt);
1200 	atomic_dec(&svsk->sk_sendqlen);
1201 	mutex_unlock(&xprt->xpt_mutex);
1202 	return -EAGAIN;
1203 }
1204 
1205 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1206 				       struct net *net,
1207 				       struct sockaddr *sa, int salen,
1208 				       int flags)
1209 {
1210 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1211 }
1212 
1213 static const struct svc_xprt_ops svc_tcp_ops = {
1214 	.xpo_create = svc_tcp_create,
1215 	.xpo_recvfrom = svc_tcp_recvfrom,
1216 	.xpo_sendto = svc_tcp_sendto,
1217 	.xpo_result_payload = svc_sock_result_payload,
1218 	.xpo_release_rqst = svc_tcp_release_rqst,
1219 	.xpo_detach = svc_tcp_sock_detach,
1220 	.xpo_free = svc_sock_free,
1221 	.xpo_has_wspace = svc_tcp_has_wspace,
1222 	.xpo_accept = svc_tcp_accept,
1223 	.xpo_secure_port = svc_sock_secure_port,
1224 	.xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1225 };
1226 
1227 static struct svc_xprt_class svc_tcp_class = {
1228 	.xcl_name = "tcp",
1229 	.xcl_owner = THIS_MODULE,
1230 	.xcl_ops = &svc_tcp_ops,
1231 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1232 	.xcl_ident = XPRT_TRANSPORT_TCP,
1233 };
1234 
1235 void svc_init_xprt_sock(void)
1236 {
1237 	svc_reg_xprt_class(&svc_tcp_class);
1238 	svc_reg_xprt_class(&svc_udp_class);
1239 }
1240 
1241 void svc_cleanup_xprt_sock(void)
1242 {
1243 	svc_unreg_xprt_class(&svc_tcp_class);
1244 	svc_unreg_xprt_class(&svc_udp_class);
1245 }
1246 
1247 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1248 {
1249 	struct sock	*sk = svsk->sk_sk;
1250 
1251 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1252 		      &svsk->sk_xprt, serv);
1253 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1254 	set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1255 	if (sk->sk_state == TCP_LISTEN) {
1256 		strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1257 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1258 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1259 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1260 	} else {
1261 		sk->sk_state_change = svc_tcp_state_change;
1262 		sk->sk_data_ready = svc_data_ready;
1263 		sk->sk_write_space = svc_write_space;
1264 
1265 		svsk->sk_marker = xdr_zero;
1266 		svsk->sk_tcplen = 0;
1267 		svsk->sk_datalen = 0;
1268 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1269 
1270 		tcp_sock_set_nodelay(sk);
1271 
1272 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1273 		switch (sk->sk_state) {
1274 		case TCP_SYN_RECV:
1275 		case TCP_ESTABLISHED:
1276 			break;
1277 		default:
1278 			svc_xprt_deferred_close(&svsk->sk_xprt);
1279 		}
1280 	}
1281 }
1282 
1283 void svc_sock_update_bufs(struct svc_serv *serv)
1284 {
1285 	/*
1286 	 * The number of server threads has changed. Update
1287 	 * rcvbuf and sndbuf accordingly on all sockets
1288 	 */
1289 	struct svc_sock *svsk;
1290 
1291 	spin_lock_bh(&serv->sv_lock);
1292 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1293 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1294 	spin_unlock_bh(&serv->sv_lock);
1295 }
1296 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1297 
1298 /*
1299  * Initialize socket for RPC use and create svc_sock struct
1300  */
1301 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1302 						struct socket *sock,
1303 						int flags)
1304 {
1305 	struct svc_sock	*svsk;
1306 	struct sock	*inet;
1307 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1308 	int		err = 0;
1309 
1310 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1311 	if (!svsk)
1312 		return ERR_PTR(-ENOMEM);
1313 
1314 	inet = sock->sk;
1315 
1316 	/* Register socket with portmapper */
1317 	if (pmap_register)
1318 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1319 				     inet->sk_protocol,
1320 				     ntohs(inet_sk(inet)->inet_sport));
1321 
1322 	if (err < 0) {
1323 		kfree(svsk);
1324 		return ERR_PTR(err);
1325 	}
1326 
1327 	svsk->sk_sock = sock;
1328 	svsk->sk_sk = inet;
1329 	svsk->sk_ostate = inet->sk_state_change;
1330 	svsk->sk_odata = inet->sk_data_ready;
1331 	svsk->sk_owspace = inet->sk_write_space;
1332 	/*
1333 	 * This barrier is necessary in order to prevent race condition
1334 	 * with svc_data_ready(), svc_listen_data_ready() and others
1335 	 * when calling callbacks above.
1336 	 */
1337 	wmb();
1338 	inet->sk_user_data = svsk;
1339 
1340 	/* Initialize the socket */
1341 	if (sock->type == SOCK_DGRAM)
1342 		svc_udp_init(svsk, serv);
1343 	else
1344 		svc_tcp_init(svsk, serv);
1345 
1346 	trace_svcsock_new_socket(sock);
1347 	return svsk;
1348 }
1349 
1350 bool svc_alien_sock(struct net *net, int fd)
1351 {
1352 	int err;
1353 	struct socket *sock = sockfd_lookup(fd, &err);
1354 	bool ret = false;
1355 
1356 	if (!sock)
1357 		goto out;
1358 	if (sock_net(sock->sk) != net)
1359 		ret = true;
1360 	sockfd_put(sock);
1361 out:
1362 	return ret;
1363 }
1364 EXPORT_SYMBOL_GPL(svc_alien_sock);
1365 
1366 /**
1367  * svc_addsock - add a listener socket to an RPC service
1368  * @serv: pointer to RPC service to which to add a new listener
1369  * @fd: file descriptor of the new listener
1370  * @name_return: pointer to buffer to fill in with name of listener
1371  * @len: size of the buffer
1372  * @cred: credential
1373  *
1374  * Fills in socket name and returns positive length of name if successful.
1375  * Name is terminated with '\n'.  On error, returns a negative errno
1376  * value.
1377  */
1378 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1379 		const size_t len, const struct cred *cred)
1380 {
1381 	int err = 0;
1382 	struct socket *so = sockfd_lookup(fd, &err);
1383 	struct svc_sock *svsk = NULL;
1384 	struct sockaddr_storage addr;
1385 	struct sockaddr *sin = (struct sockaddr *)&addr;
1386 	int salen;
1387 
1388 	if (!so)
1389 		return err;
1390 	err = -EAFNOSUPPORT;
1391 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1392 		goto out;
1393 	err =  -EPROTONOSUPPORT;
1394 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1395 	    so->sk->sk_protocol != IPPROTO_UDP)
1396 		goto out;
1397 	err = -EISCONN;
1398 	if (so->state > SS_UNCONNECTED)
1399 		goto out;
1400 	err = -ENOENT;
1401 	if (!try_module_get(THIS_MODULE))
1402 		goto out;
1403 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1404 	if (IS_ERR(svsk)) {
1405 		module_put(THIS_MODULE);
1406 		err = PTR_ERR(svsk);
1407 		goto out;
1408 	}
1409 	salen = kernel_getsockname(svsk->sk_sock, sin);
1410 	if (salen >= 0)
1411 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1412 	svsk->sk_xprt.xpt_cred = get_cred(cred);
1413 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1414 	return svc_one_sock_name(svsk, name_return, len);
1415 out:
1416 	sockfd_put(so);
1417 	return err;
1418 }
1419 EXPORT_SYMBOL_GPL(svc_addsock);
1420 
1421 /*
1422  * Create socket for RPC service.
1423  */
1424 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1425 					  int protocol,
1426 					  struct net *net,
1427 					  struct sockaddr *sin, int len,
1428 					  int flags)
1429 {
1430 	struct svc_sock	*svsk;
1431 	struct socket	*sock;
1432 	int		error;
1433 	int		type;
1434 	struct sockaddr_storage addr;
1435 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1436 	int		newlen;
1437 	int		family;
1438 
1439 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1440 		printk(KERN_WARNING "svc: only UDP and TCP "
1441 				"sockets supported\n");
1442 		return ERR_PTR(-EINVAL);
1443 	}
1444 
1445 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1446 	switch (sin->sa_family) {
1447 	case AF_INET6:
1448 		family = PF_INET6;
1449 		break;
1450 	case AF_INET:
1451 		family = PF_INET;
1452 		break;
1453 	default:
1454 		return ERR_PTR(-EINVAL);
1455 	}
1456 
1457 	error = __sock_create(net, family, type, protocol, &sock, 1);
1458 	if (error < 0)
1459 		return ERR_PTR(error);
1460 
1461 	svc_reclassify_socket(sock);
1462 
1463 	/*
1464 	 * If this is an PF_INET6 listener, we want to avoid
1465 	 * getting requests from IPv4 remotes.  Those should
1466 	 * be shunted to a PF_INET listener via rpcbind.
1467 	 */
1468 	if (family == PF_INET6)
1469 		ip6_sock_set_v6only(sock->sk);
1470 	if (type == SOCK_STREAM)
1471 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1472 	error = kernel_bind(sock, sin, len);
1473 	if (error < 0)
1474 		goto bummer;
1475 
1476 	error = kernel_getsockname(sock, newsin);
1477 	if (error < 0)
1478 		goto bummer;
1479 	newlen = error;
1480 
1481 	if (protocol == IPPROTO_TCP) {
1482 		if ((error = kernel_listen(sock, 64)) < 0)
1483 			goto bummer;
1484 	}
1485 
1486 	svsk = svc_setup_socket(serv, sock, flags);
1487 	if (IS_ERR(svsk)) {
1488 		error = PTR_ERR(svsk);
1489 		goto bummer;
1490 	}
1491 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1492 	return (struct svc_xprt *)svsk;
1493 bummer:
1494 	sock_release(sock);
1495 	return ERR_PTR(error);
1496 }
1497 
1498 /*
1499  * Detach the svc_sock from the socket so that no
1500  * more callbacks occur.
1501  */
1502 static void svc_sock_detach(struct svc_xprt *xprt)
1503 {
1504 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1505 	struct sock *sk = svsk->sk_sk;
1506 
1507 	/* put back the old socket callbacks */
1508 	lock_sock(sk);
1509 	sk->sk_state_change = svsk->sk_ostate;
1510 	sk->sk_data_ready = svsk->sk_odata;
1511 	sk->sk_write_space = svsk->sk_owspace;
1512 	sk->sk_user_data = NULL;
1513 	release_sock(sk);
1514 }
1515 
1516 /*
1517  * Disconnect the socket, and reset the callbacks
1518  */
1519 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1520 {
1521 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1522 
1523 	svc_sock_detach(xprt);
1524 
1525 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1526 		svc_tcp_clear_pages(svsk);
1527 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1528 	}
1529 }
1530 
1531 /*
1532  * Free the svc_sock's socket resources and the svc_sock itself.
1533  */
1534 static void svc_sock_free(struct svc_xprt *xprt)
1535 {
1536 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1537 
1538 	if (svsk->sk_sock->file)
1539 		sockfd_put(svsk->sk_sock);
1540 	else
1541 		sock_release(svsk->sk_sock);
1542 	kfree(svsk);
1543 }
1544