• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
svc_reclassify_socket(struct socket * sock)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
svc_reclassify_socket(struct socket * sock)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  */
svc_tcp_release_rqst(struct svc_rqst * rqstp)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  */
svc_udp_release_rqst(struct svc_rqst * rqstp)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 
svc_set_cmsg_data(struct svc_rqst * rqstp,struct cmsghdr * cmh)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 
svc_sock_read_payload(struct svc_rqst * rqstp,unsigned int offset,unsigned int length)184 static int svc_sock_read_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  */
svc_one_sock_name(struct svc_sock * svsk,char * buf,int remaining)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
svc_flush_bvec(const struct bio_vec * bvec,size_t size,size_t seek)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
svc_flush_bvec(const struct bio_vec * bvec,size_t size,size_t seek)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  */
svc_tcp_read_msg(struct svc_rqst * rqstp,size_t buflen,size_t seek)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  */
svc_sock_setbufsize(struct svc_sock * svsk,unsigned int nreqs)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 
svc_sock_secure_port(struct svc_rqst * rqstp)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  */
svc_data_ready(struct sock * sk)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  */
svc_write_space(struct sock * sk)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 
svc_tcp_has_wspace(struct svc_xprt * xprt)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 
svc_tcp_kill_temp_xprt(struct svc_xprt * xprt)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  */
svc_udp_get_dest_address4(struct svc_rqst * rqstp,struct cmsghdr * cmh)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  */
svc_udp_get_dest_address6(struct svc_rqst * rqstp,struct cmsghdr * cmh)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  */
svc_udp_get_dest_address(struct svc_rqst * rqstp,struct cmsghdr * cmh)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  */
svc_udp_recvfrom(struct svc_rqst * rqstp)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 	return len;
523 
524 out_recv_err:
525 	if (err != -EAGAIN) {
526 		/* possibly an icmp error */
527 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
528 	}
529 	trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
530 	return 0;
531 out_cmsg_err:
532 	net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
533 			     cmh->cmsg_level, cmh->cmsg_type);
534 	goto out_free;
535 out_bh_enable:
536 	local_bh_enable();
537 out_free:
538 	kfree_skb(skb);
539 	return 0;
540 }
541 
542 /**
543  * svc_udp_sendto - Send out a reply on a UDP socket
544  * @rqstp: completed svc_rqst
545  *
546  * xpt_mutex ensures @rqstp's whole message is written to the socket
547  * without interruption.
548  *
549  * Returns the number of bytes sent, or a negative errno.
550  */
svc_udp_sendto(struct svc_rqst * rqstp)551 static int svc_udp_sendto(struct svc_rqst *rqstp)
552 {
553 	struct svc_xprt *xprt = rqstp->rq_xprt;
554 	struct svc_sock	*svsk = container_of(xprt, struct svc_sock, sk_xprt);
555 	struct xdr_buf *xdr = &rqstp->rq_res;
556 	union {
557 		struct cmsghdr	hdr;
558 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
559 	} buffer;
560 	struct cmsghdr *cmh = &buffer.hdr;
561 	struct msghdr msg = {
562 		.msg_name	= &rqstp->rq_addr,
563 		.msg_namelen	= rqstp->rq_addrlen,
564 		.msg_control	= cmh,
565 		.msg_controllen	= sizeof(buffer),
566 	};
567 	unsigned int sent;
568 	int err;
569 
570 	svc_udp_release_rqst(rqstp);
571 
572 	svc_set_cmsg_data(rqstp, cmh);
573 
574 	mutex_lock(&xprt->xpt_mutex);
575 
576 	if (svc_xprt_is_dead(xprt))
577 		goto out_notconn;
578 
579 	err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
580 	xdr_free_bvec(xdr);
581 	if (err == -ECONNREFUSED) {
582 		/* ICMP error on earlier request. */
583 		err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
584 		xdr_free_bvec(xdr);
585 	}
586 	trace_svcsock_udp_send(xprt, err);
587 
588 	mutex_unlock(&xprt->xpt_mutex);
589 	if (err < 0)
590 		return err;
591 	return sent;
592 
593 out_notconn:
594 	mutex_unlock(&xprt->xpt_mutex);
595 	return -ENOTCONN;
596 }
597 
svc_udp_has_wspace(struct svc_xprt * xprt)598 static int svc_udp_has_wspace(struct svc_xprt *xprt)
599 {
600 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
601 	struct svc_serv	*serv = xprt->xpt_server;
602 	unsigned long required;
603 
604 	/*
605 	 * Set the SOCK_NOSPACE flag before checking the available
606 	 * sock space.
607 	 */
608 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
609 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
610 	if (required*2 > sock_wspace(svsk->sk_sk))
611 		return 0;
612 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
613 	return 1;
614 }
615 
svc_udp_accept(struct svc_xprt * xprt)616 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
617 {
618 	BUG();
619 	return NULL;
620 }
621 
svc_udp_kill_temp_xprt(struct svc_xprt * xprt)622 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
623 {
624 }
625 
svc_udp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)626 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
627 				       struct net *net,
628 				       struct sockaddr *sa, int salen,
629 				       int flags)
630 {
631 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
632 }
633 
634 static const struct svc_xprt_ops svc_udp_ops = {
635 	.xpo_create = svc_udp_create,
636 	.xpo_recvfrom = svc_udp_recvfrom,
637 	.xpo_sendto = svc_udp_sendto,
638 	.xpo_read_payload = svc_sock_read_payload,
639 	.xpo_release_rqst = svc_udp_release_rqst,
640 	.xpo_detach = svc_sock_detach,
641 	.xpo_free = svc_sock_free,
642 	.xpo_has_wspace = svc_udp_has_wspace,
643 	.xpo_accept = svc_udp_accept,
644 	.xpo_secure_port = svc_sock_secure_port,
645 	.xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
646 };
647 
648 static struct svc_xprt_class svc_udp_class = {
649 	.xcl_name = "udp",
650 	.xcl_owner = THIS_MODULE,
651 	.xcl_ops = &svc_udp_ops,
652 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
653 	.xcl_ident = XPRT_TRANSPORT_UDP,
654 };
655 
svc_udp_init(struct svc_sock * svsk,struct svc_serv * serv)656 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
657 {
658 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
659 		      &svsk->sk_xprt, serv);
660 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
661 	svsk->sk_sk->sk_data_ready = svc_data_ready;
662 	svsk->sk_sk->sk_write_space = svc_write_space;
663 
664 	/* initialise setting must have enough space to
665 	 * receive and respond to one request.
666 	 * svc_udp_recvfrom will re-adjust if necessary
667 	 */
668 	svc_sock_setbufsize(svsk, 3);
669 
670 	/* data might have come in before data_ready set up */
671 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
672 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
673 
674 	/* make sure we get destination address info */
675 	switch (svsk->sk_sk->sk_family) {
676 	case AF_INET:
677 		ip_sock_set_pktinfo(svsk->sk_sock->sk);
678 		break;
679 	case AF_INET6:
680 		ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
681 		break;
682 	default:
683 		BUG();
684 	}
685 }
686 
687 /*
688  * A data_ready event on a listening socket means there's a connection
689  * pending. Do not use state_change as a substitute for it.
690  */
svc_tcp_listen_data_ready(struct sock * sk)691 static void svc_tcp_listen_data_ready(struct sock *sk)
692 {
693 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
694 
695 	if (svsk) {
696 		/* Refer to svc_setup_socket() for details. */
697 		rmb();
698 		svsk->sk_odata(sk);
699 	}
700 
701 	/*
702 	 * This callback may called twice when a new connection
703 	 * is established as a child socket inherits everything
704 	 * from a parent LISTEN socket.
705 	 * 1) data_ready method of the parent socket will be called
706 	 *    when one of child sockets become ESTABLISHED.
707 	 * 2) data_ready method of the child socket may be called
708 	 *    when it receives data before the socket is accepted.
709 	 * In case of 2, we should ignore it silently.
710 	 */
711 	if (sk->sk_state == TCP_LISTEN) {
712 		if (svsk) {
713 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
714 			svc_xprt_enqueue(&svsk->sk_xprt);
715 		}
716 	}
717 }
718 
719 /*
720  * A state change on a connected socket means it's dying or dead.
721  */
svc_tcp_state_change(struct sock * sk)722 static void svc_tcp_state_change(struct sock *sk)
723 {
724 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
725 
726 	if (svsk) {
727 		/* Refer to svc_setup_socket() for details. */
728 		rmb();
729 		svsk->sk_ostate(sk);
730 		trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
731 		if (sk->sk_state != TCP_ESTABLISHED) {
732 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
733 			svc_xprt_enqueue(&svsk->sk_xprt);
734 		}
735 	}
736 }
737 
738 /*
739  * Accept a TCP connection
740  */
svc_tcp_accept(struct svc_xprt * xprt)741 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
742 {
743 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
744 	struct sockaddr_storage addr;
745 	struct sockaddr	*sin = (struct sockaddr *) &addr;
746 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
747 	struct socket	*sock = svsk->sk_sock;
748 	struct socket	*newsock;
749 	struct svc_sock	*newsvsk;
750 	int		err, slen;
751 
752 	if (!sock)
753 		return NULL;
754 
755 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
756 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
757 	if (err < 0) {
758 		if (err == -ENOMEM)
759 			printk(KERN_WARNING "%s: no more sockets!\n",
760 			       serv->sv_name);
761 		else if (err != -EAGAIN)
762 			net_warn_ratelimited("%s: accept failed (err %d)!\n",
763 					     serv->sv_name, -err);
764 		trace_svcsock_accept_err(xprt, serv->sv_name, err);
765 		return NULL;
766 	}
767 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
768 
769 	err = kernel_getpeername(newsock, sin);
770 	if (err < 0) {
771 		trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
772 		goto failed;		/* aborted connection or whatever */
773 	}
774 	slen = err;
775 
776 	/* Reset the inherited callbacks before calling svc_setup_socket */
777 	newsock->sk->sk_state_change = svsk->sk_ostate;
778 	newsock->sk->sk_data_ready = svsk->sk_odata;
779 	newsock->sk->sk_write_space = svsk->sk_owspace;
780 
781 	/* make sure that a write doesn't block forever when
782 	 * low on memory
783 	 */
784 	newsock->sk->sk_sndtimeo = HZ*30;
785 
786 	newsvsk = svc_setup_socket(serv, newsock,
787 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
788 	if (IS_ERR(newsvsk))
789 		goto failed;
790 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
791 	err = kernel_getsockname(newsock, sin);
792 	slen = err;
793 	if (unlikely(err < 0))
794 		slen = offsetof(struct sockaddr, sa_data);
795 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
796 
797 	if (sock_is_loopback(newsock->sk))
798 		set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
799 	else
800 		clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
801 	if (serv->sv_stats)
802 		serv->sv_stats->nettcpconn++;
803 
804 	return &newsvsk->sk_xprt;
805 
806 failed:
807 	sock_release(newsock);
808 	return NULL;
809 }
810 
svc_tcp_restore_pages(struct svc_sock * svsk,struct svc_rqst * rqstp)811 static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
812 				    struct svc_rqst *rqstp)
813 {
814 	size_t len = svsk->sk_datalen;
815 	unsigned int i, npages;
816 
817 	if (!len)
818 		return 0;
819 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
820 	for (i = 0; i < npages; i++) {
821 		if (rqstp->rq_pages[i] != NULL)
822 			put_page(rqstp->rq_pages[i]);
823 		BUG_ON(svsk->sk_pages[i] == NULL);
824 		rqstp->rq_pages[i] = svsk->sk_pages[i];
825 		svsk->sk_pages[i] = NULL;
826 	}
827 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
828 	return len;
829 }
830 
svc_tcp_save_pages(struct svc_sock * svsk,struct svc_rqst * rqstp)831 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
832 {
833 	unsigned int i, len, npages;
834 
835 	if (svsk->sk_datalen == 0)
836 		return;
837 	len = svsk->sk_datalen;
838 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
839 	for (i = 0; i < npages; i++) {
840 		svsk->sk_pages[i] = rqstp->rq_pages[i];
841 		rqstp->rq_pages[i] = NULL;
842 	}
843 }
844 
svc_tcp_clear_pages(struct svc_sock * svsk)845 static void svc_tcp_clear_pages(struct svc_sock *svsk)
846 {
847 	unsigned int i, len, npages;
848 
849 	if (svsk->sk_datalen == 0)
850 		goto out;
851 	len = svsk->sk_datalen;
852 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
853 	for (i = 0; i < npages; i++) {
854 		if (svsk->sk_pages[i] == NULL) {
855 			WARN_ON_ONCE(1);
856 			continue;
857 		}
858 		put_page(svsk->sk_pages[i]);
859 		svsk->sk_pages[i] = NULL;
860 	}
861 out:
862 	svsk->sk_tcplen = 0;
863 	svsk->sk_datalen = 0;
864 }
865 
866 /*
867  * Receive fragment record header into sk_marker.
868  */
svc_tcp_read_marker(struct svc_sock * svsk,struct svc_rqst * rqstp)869 static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
870 				   struct svc_rqst *rqstp)
871 {
872 	ssize_t want, len;
873 
874 	/* If we haven't gotten the record length yet,
875 	 * get the next four bytes.
876 	 */
877 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
878 		struct msghdr	msg = { NULL };
879 		struct kvec	iov;
880 
881 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
882 		iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
883 		iov.iov_len  = want;
884 		iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, want);
885 		len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
886 		if (len < 0)
887 			return len;
888 		svsk->sk_tcplen += len;
889 		if (len < want) {
890 			/* call again to read the remaining bytes */
891 			goto err_short;
892 		}
893 		trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
894 		if (svc_sock_reclen(svsk) + svsk->sk_datalen >
895 		    svsk->sk_xprt.xpt_server->sv_max_mesg)
896 			goto err_too_large;
897 	}
898 	return svc_sock_reclen(svsk);
899 
900 err_too_large:
901 	net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
902 			       __func__, svsk->sk_xprt.xpt_server->sv_name,
903 			       svc_sock_reclen(svsk));
904 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
905 err_short:
906 	return -EAGAIN;
907 }
908 
receive_cb_reply(struct svc_sock * svsk,struct svc_rqst * rqstp)909 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
910 {
911 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
912 	struct rpc_rqst *req = NULL;
913 	struct kvec *src, *dst;
914 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
915 	__be32 xid;
916 	__be32 calldir;
917 
918 	xid = *p++;
919 	calldir = *p;
920 
921 	if (!bc_xprt)
922 		return -EAGAIN;
923 	spin_lock(&bc_xprt->queue_lock);
924 	req = xprt_lookup_rqst(bc_xprt, xid);
925 	if (!req)
926 		goto unlock_notfound;
927 
928 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
929 	/*
930 	 * XXX!: cheating for now!  Only copying HEAD.
931 	 * But we know this is good enough for now (in fact, for any
932 	 * callback reply in the forseeable future).
933 	 */
934 	dst = &req->rq_private_buf.head[0];
935 	src = &rqstp->rq_arg.head[0];
936 	if (dst->iov_len < src->iov_len)
937 		goto unlock_eagain; /* whatever; just giving up. */
938 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
939 	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
940 	rqstp->rq_arg.len = 0;
941 	spin_unlock(&bc_xprt->queue_lock);
942 	return 0;
943 unlock_notfound:
944 	printk(KERN_NOTICE
945 		"%s: Got unrecognized reply: "
946 		"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
947 		__func__, ntohl(calldir),
948 		bc_xprt, ntohl(xid));
949 unlock_eagain:
950 	spin_unlock(&bc_xprt->queue_lock);
951 	return -EAGAIN;
952 }
953 
svc_tcp_fragment_received(struct svc_sock * svsk)954 static void svc_tcp_fragment_received(struct svc_sock *svsk)
955 {
956 	/* If we have more data, signal svc_xprt_enqueue() to try again */
957 	svsk->sk_tcplen = 0;
958 	svsk->sk_marker = xdr_zero;
959 }
960 
961 /**
962  * svc_tcp_recvfrom - Receive data from a TCP socket
963  * @rqstp: request structure into which to receive an RPC Call
964  *
965  * Called in a loop when XPT_DATA has been set.
966  *
967  * Read the 4-byte stream record marker, then use the record length
968  * in that marker to set up exactly the resources needed to receive
969  * the next RPC message into @rqstp.
970  *
971  * Returns:
972  *   On success, the number of bytes in a received RPC Call, or
973  *   %0 if a complete RPC Call message was not ready to return
974  *
975  * The zero return case handles partial receives and callback Replies.
976  * The state of a partial receive is preserved in the svc_sock for
977  * the next call to svc_tcp_recvfrom.
978  */
svc_tcp_recvfrom(struct svc_rqst * rqstp)979 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
980 {
981 	struct svc_sock	*svsk =
982 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
983 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
984 	size_t want, base;
985 	ssize_t len;
986 	__be32 *p;
987 	__be32 calldir;
988 
989 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
990 	len = svc_tcp_read_marker(svsk, rqstp);
991 	if (len < 0)
992 		goto error;
993 
994 	base = svc_tcp_restore_pages(svsk, rqstp);
995 	want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
996 	len = svc_tcp_read_msg(rqstp, base + want, base);
997 	if (len >= 0) {
998 		trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
999 		svsk->sk_tcplen += len;
1000 		svsk->sk_datalen += len;
1001 	}
1002 	if (len != want || !svc_sock_final_rec(svsk))
1003 		goto err_incomplete;
1004 	if (svsk->sk_datalen < 8)
1005 		goto err_nuts;
1006 
1007 	rqstp->rq_arg.len = svsk->sk_datalen;
1008 	rqstp->rq_arg.page_base = 0;
1009 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1010 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1011 		rqstp->rq_arg.page_len = 0;
1012 	} else
1013 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1014 
1015 	rqstp->rq_xprt_ctxt   = NULL;
1016 	rqstp->rq_prot	      = IPPROTO_TCP;
1017 	if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1018 		set_bit(RQ_LOCAL, &rqstp->rq_flags);
1019 	else
1020 		clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1021 
1022 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1023 	calldir = p[1];
1024 	if (calldir)
1025 		len = receive_cb_reply(svsk, rqstp);
1026 
1027 	/* Reset TCP read info */
1028 	svsk->sk_datalen = 0;
1029 	svc_tcp_fragment_received(svsk);
1030 
1031 	if (len < 0)
1032 		goto error;
1033 
1034 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1035 	if (serv->sv_stats)
1036 		serv->sv_stats->nettcpcnt++;
1037 
1038 	return rqstp->rq_arg.len;
1039 
1040 err_incomplete:
1041 	svc_tcp_save_pages(svsk, rqstp);
1042 	if (len < 0 && len != -EAGAIN)
1043 		goto err_delete;
1044 	if (len == want)
1045 		svc_tcp_fragment_received(svsk);
1046 	else
1047 		trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1048 				svc_sock_reclen(svsk),
1049 				svsk->sk_tcplen - sizeof(rpc_fraghdr));
1050 	goto err_noclose;
1051 error:
1052 	if (len != -EAGAIN)
1053 		goto err_delete;
1054 	trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1055 	return 0;
1056 err_nuts:
1057 	svsk->sk_datalen = 0;
1058 err_delete:
1059 	trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1060 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1061 err_noclose:
1062 	return 0;	/* record not complete */
1063 }
1064 
svc_tcp_send_kvec(struct socket * sock,const struct kvec * vec,int flags)1065 static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1066 			      int flags)
1067 {
1068 	return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1069 			       offset_in_page(vec->iov_base),
1070 			       vec->iov_len, flags);
1071 }
1072 
1073 /*
1074  * kernel_sendpage() is used exclusively to reduce the number of
1075  * copy operations in this path. Therefore the caller must ensure
1076  * that the pages backing @xdr are unchanging.
1077  *
1078  * In addition, the logic assumes that * .bv_len is never larger
1079  * than PAGE_SIZE.
1080  */
svc_tcp_sendmsg(struct socket * sock,struct msghdr * msg,struct xdr_buf * xdr,rpc_fraghdr marker,unsigned int * sentp)1081 static int svc_tcp_sendmsg(struct socket *sock, struct msghdr *msg,
1082 			   struct xdr_buf *xdr, rpc_fraghdr marker,
1083 			   unsigned int *sentp)
1084 {
1085 	const struct kvec *head = xdr->head;
1086 	const struct kvec *tail = xdr->tail;
1087 	struct kvec rm = {
1088 		.iov_base	= &marker,
1089 		.iov_len	= sizeof(marker),
1090 	};
1091 	int flags, ret;
1092 
1093 	*sentp = 0;
1094 	xdr_alloc_bvec(xdr, GFP_KERNEL);
1095 
1096 	msg->msg_flags = MSG_MORE;
1097 	ret = kernel_sendmsg(sock, msg, &rm, 1, rm.iov_len);
1098 	if (ret < 0)
1099 		return ret;
1100 	*sentp += ret;
1101 	if (ret != rm.iov_len)
1102 		return -EAGAIN;
1103 
1104 	flags = head->iov_len < xdr->len ? MSG_MORE | MSG_SENDPAGE_NOTLAST : 0;
1105 	ret = svc_tcp_send_kvec(sock, head, flags);
1106 	if (ret < 0)
1107 		return ret;
1108 	*sentp += ret;
1109 	if (ret != head->iov_len)
1110 		goto out;
1111 
1112 	if (xdr->page_len) {
1113 		unsigned int offset, len, remaining;
1114 		struct bio_vec *bvec;
1115 
1116 		bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1117 		offset = offset_in_page(xdr->page_base);
1118 		remaining = xdr->page_len;
1119 		flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1120 		while (remaining > 0) {
1121 			if (remaining <= PAGE_SIZE && tail->iov_len == 0)
1122 				flags = 0;
1123 
1124 			len = min(remaining, bvec->bv_len - offset);
1125 			ret = kernel_sendpage(sock, bvec->bv_page,
1126 					      bvec->bv_offset + offset,
1127 					      len, flags);
1128 			if (ret < 0)
1129 				return ret;
1130 			*sentp += ret;
1131 			if (ret != len)
1132 				goto out;
1133 			remaining -= len;
1134 			offset = 0;
1135 			bvec++;
1136 		}
1137 	}
1138 
1139 	if (tail->iov_len) {
1140 		ret = svc_tcp_send_kvec(sock, tail, 0);
1141 		if (ret < 0)
1142 			return ret;
1143 		*sentp += ret;
1144 	}
1145 
1146 out:
1147 	return 0;
1148 }
1149 
1150 /**
1151  * svc_tcp_sendto - Send out a reply on a TCP socket
1152  * @rqstp: completed svc_rqst
1153  *
1154  * xpt_mutex ensures @rqstp's whole message is written to the socket
1155  * without interruption.
1156  *
1157  * Returns the number of bytes sent, or a negative errno.
1158  */
svc_tcp_sendto(struct svc_rqst * rqstp)1159 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1160 {
1161 	struct svc_xprt *xprt = rqstp->rq_xprt;
1162 	struct svc_sock	*svsk = container_of(xprt, struct svc_sock, sk_xprt);
1163 	struct xdr_buf *xdr = &rqstp->rq_res;
1164 	rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1165 					 (u32)xdr->len);
1166 	struct msghdr msg = {
1167 		.msg_flags	= 0,
1168 	};
1169 	unsigned int sent;
1170 	int err;
1171 
1172 	svc_tcp_release_rqst(rqstp);
1173 
1174 	mutex_lock(&xprt->xpt_mutex);
1175 	if (svc_xprt_is_dead(xprt))
1176 		goto out_notconn;
1177 	err = svc_tcp_sendmsg(svsk->sk_sock, &msg, xdr, marker, &sent);
1178 	xdr_free_bvec(xdr);
1179 	trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1180 	if (err < 0 || sent != (xdr->len + sizeof(marker)))
1181 		goto out_close;
1182 	mutex_unlock(&xprt->xpt_mutex);
1183 	return sent;
1184 
1185 out_notconn:
1186 	mutex_unlock(&xprt->xpt_mutex);
1187 	return -ENOTCONN;
1188 out_close:
1189 	pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1190 		  xprt->xpt_server->sv_name,
1191 		  (err < 0) ? "got error" : "sent",
1192 		  (err < 0) ? err : sent, xdr->len);
1193 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
1194 	svc_xprt_enqueue(xprt);
1195 	mutex_unlock(&xprt->xpt_mutex);
1196 	return -EAGAIN;
1197 }
1198 
svc_tcp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)1199 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1200 				       struct net *net,
1201 				       struct sockaddr *sa, int salen,
1202 				       int flags)
1203 {
1204 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1205 }
1206 
1207 static const struct svc_xprt_ops svc_tcp_ops = {
1208 	.xpo_create = svc_tcp_create,
1209 	.xpo_recvfrom = svc_tcp_recvfrom,
1210 	.xpo_sendto = svc_tcp_sendto,
1211 	.xpo_read_payload = svc_sock_read_payload,
1212 	.xpo_release_rqst = svc_tcp_release_rqst,
1213 	.xpo_detach = svc_tcp_sock_detach,
1214 	.xpo_free = svc_sock_free,
1215 	.xpo_has_wspace = svc_tcp_has_wspace,
1216 	.xpo_accept = svc_tcp_accept,
1217 	.xpo_secure_port = svc_sock_secure_port,
1218 	.xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1219 };
1220 
1221 static struct svc_xprt_class svc_tcp_class = {
1222 	.xcl_name = "tcp",
1223 	.xcl_owner = THIS_MODULE,
1224 	.xcl_ops = &svc_tcp_ops,
1225 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1226 	.xcl_ident = XPRT_TRANSPORT_TCP,
1227 };
1228 
svc_init_xprt_sock(void)1229 void svc_init_xprt_sock(void)
1230 {
1231 	svc_reg_xprt_class(&svc_tcp_class);
1232 	svc_reg_xprt_class(&svc_udp_class);
1233 }
1234 
svc_cleanup_xprt_sock(void)1235 void svc_cleanup_xprt_sock(void)
1236 {
1237 	svc_unreg_xprt_class(&svc_tcp_class);
1238 	svc_unreg_xprt_class(&svc_udp_class);
1239 }
1240 
svc_tcp_init(struct svc_sock * svsk,struct svc_serv * serv)1241 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1242 {
1243 	struct sock	*sk = svsk->sk_sk;
1244 
1245 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1246 		      &svsk->sk_xprt, serv);
1247 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1248 	set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1249 	if (sk->sk_state == TCP_LISTEN) {
1250 		strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1251 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1252 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1253 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1254 	} else {
1255 		sk->sk_state_change = svc_tcp_state_change;
1256 		sk->sk_data_ready = svc_data_ready;
1257 		sk->sk_write_space = svc_write_space;
1258 
1259 		svsk->sk_marker = xdr_zero;
1260 		svsk->sk_tcplen = 0;
1261 		svsk->sk_datalen = 0;
1262 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1263 
1264 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1265 
1266 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1267 		switch (sk->sk_state) {
1268 		case TCP_SYN_RECV:
1269 		case TCP_ESTABLISHED:
1270 			break;
1271 		default:
1272 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1273 		}
1274 	}
1275 }
1276 
svc_sock_update_bufs(struct svc_serv * serv)1277 void svc_sock_update_bufs(struct svc_serv *serv)
1278 {
1279 	/*
1280 	 * The number of server threads has changed. Update
1281 	 * rcvbuf and sndbuf accordingly on all sockets
1282 	 */
1283 	struct svc_sock *svsk;
1284 
1285 	spin_lock_bh(&serv->sv_lock);
1286 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1287 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1288 	spin_unlock_bh(&serv->sv_lock);
1289 }
1290 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1291 
1292 /*
1293  * Initialize socket for RPC use and create svc_sock struct
1294  */
svc_setup_socket(struct svc_serv * serv,struct socket * sock,int flags)1295 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1296 						struct socket *sock,
1297 						int flags)
1298 {
1299 	struct svc_sock	*svsk;
1300 	struct sock	*inet;
1301 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1302 	int		err = 0;
1303 
1304 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1305 	if (!svsk)
1306 		return ERR_PTR(-ENOMEM);
1307 
1308 	inet = sock->sk;
1309 
1310 	/* Register socket with portmapper */
1311 	if (pmap_register)
1312 		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1313 				     inet->sk_protocol,
1314 				     ntohs(inet_sk(inet)->inet_sport));
1315 
1316 	if (err < 0) {
1317 		kfree(svsk);
1318 		return ERR_PTR(err);
1319 	}
1320 
1321 	svsk->sk_sock = sock;
1322 	svsk->sk_sk = inet;
1323 	svsk->sk_ostate = inet->sk_state_change;
1324 	svsk->sk_odata = inet->sk_data_ready;
1325 	svsk->sk_owspace = inet->sk_write_space;
1326 	/*
1327 	 * This barrier is necessary in order to prevent race condition
1328 	 * with svc_data_ready(), svc_listen_data_ready() and others
1329 	 * when calling callbacks above.
1330 	 */
1331 	wmb();
1332 	inet->sk_user_data = svsk;
1333 
1334 	/* Initialize the socket */
1335 	if (sock->type == SOCK_DGRAM)
1336 		svc_udp_init(svsk, serv);
1337 	else
1338 		svc_tcp_init(svsk, serv);
1339 
1340 	trace_svcsock_new_socket(sock);
1341 	return svsk;
1342 }
1343 
svc_alien_sock(struct net * net,int fd)1344 bool svc_alien_sock(struct net *net, int fd)
1345 {
1346 	int err;
1347 	struct socket *sock = sockfd_lookup(fd, &err);
1348 	bool ret = false;
1349 
1350 	if (!sock)
1351 		goto out;
1352 	if (sock_net(sock->sk) != net)
1353 		ret = true;
1354 	sockfd_put(sock);
1355 out:
1356 	return ret;
1357 }
1358 EXPORT_SYMBOL_GPL(svc_alien_sock);
1359 
1360 /**
1361  * svc_addsock - add a listener socket to an RPC service
1362  * @serv: pointer to RPC service to which to add a new listener
1363  * @fd: file descriptor of the new listener
1364  * @name_return: pointer to buffer to fill in with name of listener
1365  * @len: size of the buffer
1366  * @cred: credential
1367  *
1368  * Fills in socket name and returns positive length of name if successful.
1369  * Name is terminated with '\n'.  On error, returns a negative errno
1370  * value.
1371  */
svc_addsock(struct svc_serv * serv,const int fd,char * name_return,const size_t len,const struct cred * cred)1372 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1373 		const size_t len, const struct cred *cred)
1374 {
1375 	int err = 0;
1376 	struct socket *so = sockfd_lookup(fd, &err);
1377 	struct svc_sock *svsk = NULL;
1378 	struct sockaddr_storage addr;
1379 	struct sockaddr *sin = (struct sockaddr *)&addr;
1380 	int salen;
1381 
1382 	if (!so)
1383 		return err;
1384 	err = -EAFNOSUPPORT;
1385 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1386 		goto out;
1387 	err =  -EPROTONOSUPPORT;
1388 	if (so->sk->sk_protocol != IPPROTO_TCP &&
1389 	    so->sk->sk_protocol != IPPROTO_UDP)
1390 		goto out;
1391 	err = -EISCONN;
1392 	if (so->state > SS_UNCONNECTED)
1393 		goto out;
1394 	err = -ENOENT;
1395 	if (!try_module_get(THIS_MODULE))
1396 		goto out;
1397 	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1398 	if (IS_ERR(svsk)) {
1399 		module_put(THIS_MODULE);
1400 		err = PTR_ERR(svsk);
1401 		goto out;
1402 	}
1403 	salen = kernel_getsockname(svsk->sk_sock, sin);
1404 	if (salen >= 0)
1405 		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1406 	svsk->sk_xprt.xpt_cred = get_cred(cred);
1407 	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1408 	return svc_one_sock_name(svsk, name_return, len);
1409 out:
1410 	sockfd_put(so);
1411 	return err;
1412 }
1413 EXPORT_SYMBOL_GPL(svc_addsock);
1414 
1415 /*
1416  * Create socket for RPC service.
1417  */
svc_create_socket(struct svc_serv * serv,int protocol,struct net * net,struct sockaddr * sin,int len,int flags)1418 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1419 					  int protocol,
1420 					  struct net *net,
1421 					  struct sockaddr *sin, int len,
1422 					  int flags)
1423 {
1424 	struct svc_sock	*svsk;
1425 	struct socket	*sock;
1426 	int		error;
1427 	int		type;
1428 	struct sockaddr_storage addr;
1429 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1430 	int		newlen;
1431 	int		family;
1432 
1433 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1434 		printk(KERN_WARNING "svc: only UDP and TCP "
1435 				"sockets supported\n");
1436 		return ERR_PTR(-EINVAL);
1437 	}
1438 
1439 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1440 	switch (sin->sa_family) {
1441 	case AF_INET6:
1442 		family = PF_INET6;
1443 		break;
1444 	case AF_INET:
1445 		family = PF_INET;
1446 		break;
1447 	default:
1448 		return ERR_PTR(-EINVAL);
1449 	}
1450 
1451 	error = __sock_create(net, family, type, protocol, &sock, 1);
1452 	if (error < 0)
1453 		return ERR_PTR(error);
1454 
1455 	svc_reclassify_socket(sock);
1456 
1457 	/*
1458 	 * If this is an PF_INET6 listener, we want to avoid
1459 	 * getting requests from IPv4 remotes.  Those should
1460 	 * be shunted to a PF_INET listener via rpcbind.
1461 	 */
1462 	if (family == PF_INET6)
1463 		ip6_sock_set_v6only(sock->sk);
1464 	if (type == SOCK_STREAM)
1465 		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1466 	error = kernel_bind(sock, sin, len);
1467 	if (error < 0)
1468 		goto bummer;
1469 
1470 	error = kernel_getsockname(sock, newsin);
1471 	if (error < 0)
1472 		goto bummer;
1473 	newlen = error;
1474 
1475 	if (protocol == IPPROTO_TCP) {
1476 		if ((error = kernel_listen(sock, 64)) < 0)
1477 			goto bummer;
1478 	}
1479 
1480 	svsk = svc_setup_socket(serv, sock, flags);
1481 	if (IS_ERR(svsk)) {
1482 		error = PTR_ERR(svsk);
1483 		goto bummer;
1484 	}
1485 	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1486 	return (struct svc_xprt *)svsk;
1487 bummer:
1488 	sock_release(sock);
1489 	return ERR_PTR(error);
1490 }
1491 
1492 /*
1493  * Detach the svc_sock from the socket so that no
1494  * more callbacks occur.
1495  */
svc_sock_detach(struct svc_xprt * xprt)1496 static void svc_sock_detach(struct svc_xprt *xprt)
1497 {
1498 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1499 	struct sock *sk = svsk->sk_sk;
1500 
1501 	/* put back the old socket callbacks */
1502 	lock_sock(sk);
1503 	sk->sk_state_change = svsk->sk_ostate;
1504 	sk->sk_data_ready = svsk->sk_odata;
1505 	sk->sk_write_space = svsk->sk_owspace;
1506 	sk->sk_user_data = NULL;
1507 	release_sock(sk);
1508 }
1509 
1510 /*
1511  * Disconnect the socket, and reset the callbacks
1512  */
svc_tcp_sock_detach(struct svc_xprt * xprt)1513 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1514 {
1515 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1516 
1517 	svc_sock_detach(xprt);
1518 
1519 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1520 		svc_tcp_clear_pages(svsk);
1521 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1522 	}
1523 }
1524 
1525 /*
1526  * Free the svc_sock's socket resources and the svc_sock itself.
1527  */
svc_sock_free(struct svc_xprt * xprt)1528 static void svc_sock_free(struct svc_xprt *xprt)
1529 {
1530 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1531 
1532 	if (svsk->sk_sock->file)
1533 		sockfd_put(svsk->sk_sock);
1534 	else
1535 		sock_release(svsk->sk_sock);
1536 	kfree(svsk);
1537 }
1538