1 /*
2  * linux/net/sunrpc/svcsock.c
3  *
4  * These are the RPC server socket internals.
5  *
6  * The server scheduling algorithm does not always distribute the load
7  * evenly when servicing a single client. May need to modify the
8  * svc_xprt_enqueue procedure...
9  *
10  * TCP support is largely untested and may be a little slow. The problem
11  * is that we currently do two separate recvfrom's, one for the 4-byte
12  * record length, and the second for the actual record. This could possibly
13  * be improved by always reading a minimum size of around 100 bytes and
14  * tucking any superfluous bytes away in a temporary store. Still, that
15  * leaves write requests out in the rain. An alternative may be to peek at
16  * the first skb in the queue, and if it matches the next TCP sequence
17  * number, to extract the record marker. Yuck.
18  *
19  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20  */
21 
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/ip.h>
41 #include <net/ipv6.h>
42 #include <net/tcp.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 
47 #include <linux/sunrpc/types.h>
48 #include <linux/sunrpc/clnt.h>
49 #include <linux/sunrpc/xdr.h>
50 #include <linux/sunrpc/msg_prot.h>
51 #include <linux/sunrpc/svcsock.h>
52 #include <linux/sunrpc/stats.h>
53 #include <linux/sunrpc/xprt.h>
54 
55 #include "sunrpc.h"
56 
57 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
58 
59 
60 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
61 					 int *errp, int flags);
62 static void		svc_udp_data_ready(struct sock *, int);
63 static int		svc_udp_recvfrom(struct svc_rqst *);
64 static int		svc_udp_sendto(struct svc_rqst *);
65 static void		svc_sock_detach(struct svc_xprt *);
66 static void		svc_tcp_sock_detach(struct svc_xprt *);
67 static void		svc_sock_free(struct svc_xprt *);
68 
69 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
70 					  struct net *, struct sockaddr *,
71 					  int, int);
72 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
73 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
74 					     struct net *, struct sockaddr *,
75 					     int, int);
76 static void svc_bc_sock_free(struct svc_xprt *xprt);
77 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
78 
79 #ifdef CONFIG_DEBUG_LOCK_ALLOC
80 static struct lock_class_key svc_key[2];
81 static struct lock_class_key svc_slock_key[2];
82 
svc_reclassify_socket(struct socket * sock)83 static void svc_reclassify_socket(struct socket *sock)
84 {
85 	struct sock *sk = sock->sk;
86 	BUG_ON(sock_owned_by_user(sk));
87 	switch (sk->sk_family) {
88 	case AF_INET:
89 		sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
90 					      &svc_slock_key[0],
91 					      "sk_xprt.xpt_lock-AF_INET-NFSD",
92 					      &svc_key[0]);
93 		break;
94 
95 	case AF_INET6:
96 		sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
97 					      &svc_slock_key[1],
98 					      "sk_xprt.xpt_lock-AF_INET6-NFSD",
99 					      &svc_key[1]);
100 		break;
101 
102 	default:
103 		BUG();
104 	}
105 }
106 #else
svc_reclassify_socket(struct socket * sock)107 static void svc_reclassify_socket(struct socket *sock)
108 {
109 }
110 #endif
111 
112 /*
113  * Release an skbuff after use
114  */
svc_release_skb(struct svc_rqst * rqstp)115 static void svc_release_skb(struct svc_rqst *rqstp)
116 {
117 	struct sk_buff *skb = rqstp->rq_xprt_ctxt;
118 
119 	if (skb) {
120 		struct svc_sock *svsk =
121 			container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
122 		rqstp->rq_xprt_ctxt = NULL;
123 
124 		dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
125 		skb_free_datagram_locked(svsk->sk_sk, skb);
126 	}
127 }
128 
129 union svc_pktinfo_u {
130 	struct in_pktinfo pkti;
131 	struct in6_pktinfo pkti6;
132 };
133 #define SVC_PKTINFO_SPACE \
134 	CMSG_SPACE(sizeof(union svc_pktinfo_u))
135 
svc_set_cmsg_data(struct svc_rqst * rqstp,struct cmsghdr * cmh)136 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
137 {
138 	struct svc_sock *svsk =
139 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
140 	switch (svsk->sk_sk->sk_family) {
141 	case AF_INET: {
142 			struct in_pktinfo *pki = CMSG_DATA(cmh);
143 
144 			cmh->cmsg_level = SOL_IP;
145 			cmh->cmsg_type = IP_PKTINFO;
146 			pki->ipi_ifindex = 0;
147 			pki->ipi_spec_dst.s_addr =
148 				 svc_daddr_in(rqstp)->sin_addr.s_addr;
149 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
150 		}
151 		break;
152 
153 	case AF_INET6: {
154 			struct in6_pktinfo *pki = CMSG_DATA(cmh);
155 			struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
156 
157 			cmh->cmsg_level = SOL_IPV6;
158 			cmh->cmsg_type = IPV6_PKTINFO;
159 			pki->ipi6_ifindex = daddr->sin6_scope_id;
160 			pki->ipi6_addr = daddr->sin6_addr;
161 			cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
162 		}
163 		break;
164 	}
165 }
166 
167 /*
168  * send routine intended to be shared by the fore- and back-channel
169  */
svc_send_common(struct socket * sock,struct xdr_buf * xdr,struct page * headpage,unsigned long headoffset,struct page * tailpage,unsigned long tailoffset)170 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
171 		    struct page *headpage, unsigned long headoffset,
172 		    struct page *tailpage, unsigned long tailoffset)
173 {
174 	int		result;
175 	int		size;
176 	struct page	**ppage = xdr->pages;
177 	size_t		base = xdr->page_base;
178 	unsigned int	pglen = xdr->page_len;
179 	unsigned int	flags = MSG_MORE;
180 	int		slen;
181 	int		len = 0;
182 
183 	slen = xdr->len;
184 
185 	/* send head */
186 	if (slen == xdr->head[0].iov_len)
187 		flags = 0;
188 	len = kernel_sendpage(sock, headpage, headoffset,
189 				  xdr->head[0].iov_len, flags);
190 	if (len != xdr->head[0].iov_len)
191 		goto out;
192 	slen -= xdr->head[0].iov_len;
193 	if (slen == 0)
194 		goto out;
195 
196 	/* send page data */
197 	size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
198 	while (pglen > 0) {
199 		if (slen == size)
200 			flags = 0;
201 		result = kernel_sendpage(sock, *ppage, base, size, flags);
202 		if (result > 0)
203 			len += result;
204 		if (result != size)
205 			goto out;
206 		slen -= size;
207 		pglen -= size;
208 		size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
209 		base = 0;
210 		ppage++;
211 	}
212 
213 	/* send tail */
214 	if (xdr->tail[0].iov_len) {
215 		result = kernel_sendpage(sock, tailpage, tailoffset,
216 				   xdr->tail[0].iov_len, 0);
217 		if (result > 0)
218 			len += result;
219 	}
220 
221 out:
222 	return len;
223 }
224 
225 
226 /*
227  * Generic sendto routine
228  */
svc_sendto(struct svc_rqst * rqstp,struct xdr_buf * xdr)229 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
230 {
231 	struct svc_sock	*svsk =
232 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
233 	struct socket	*sock = svsk->sk_sock;
234 	union {
235 		struct cmsghdr	hdr;
236 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
237 	} buffer;
238 	struct cmsghdr *cmh = &buffer.hdr;
239 	int		len = 0;
240 	unsigned long tailoff;
241 	unsigned long headoff;
242 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
243 
244 	if (rqstp->rq_prot == IPPROTO_UDP) {
245 		struct msghdr msg = {
246 			.msg_name	= &rqstp->rq_addr,
247 			.msg_namelen	= rqstp->rq_addrlen,
248 			.msg_control	= cmh,
249 			.msg_controllen	= sizeof(buffer),
250 			.msg_flags	= MSG_MORE,
251 		};
252 
253 		svc_set_cmsg_data(rqstp, cmh);
254 
255 		if (sock_sendmsg(sock, &msg, 0) < 0)
256 			goto out;
257 	}
258 
259 	tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
260 	headoff = 0;
261 	len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
262 			       rqstp->rq_respages[0], tailoff);
263 
264 out:
265 	dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
266 		svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
267 		xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
268 
269 	return len;
270 }
271 
272 /*
273  * Report socket names for nfsdfs
274  */
svc_one_sock_name(struct svc_sock * svsk,char * buf,int remaining)275 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
276 {
277 	const struct sock *sk = svsk->sk_sk;
278 	const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
279 							"udp" : "tcp";
280 	int len;
281 
282 	switch (sk->sk_family) {
283 	case PF_INET:
284 		len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
285 				proto_name,
286 				&inet_sk(sk)->inet_rcv_saddr,
287 				inet_sk(sk)->inet_num);
288 		break;
289 	case PF_INET6:
290 		len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
291 				proto_name,
292 				&inet6_sk(sk)->rcv_saddr,
293 				inet_sk(sk)->inet_num);
294 		break;
295 	default:
296 		len = snprintf(buf, remaining, "*unknown-%d*\n",
297 				sk->sk_family);
298 	}
299 
300 	if (len >= remaining) {
301 		*buf = '\0';
302 		return -ENAMETOOLONG;
303 	}
304 	return len;
305 }
306 
307 /**
308  * svc_sock_names - construct a list of listener names in a string
309  * @serv: pointer to RPC service
310  * @buf: pointer to a buffer to fill in with socket names
311  * @buflen: size of the buffer to be filled
312  * @toclose: pointer to '\0'-terminated C string containing the name
313  *		of a listener to be closed
314  *
315  * Fills in @buf with a '\n'-separated list of names of listener
316  * sockets.  If @toclose is not NULL, the socket named by @toclose
317  * is closed, and is not included in the output list.
318  *
319  * Returns positive length of the socket name string, or a negative
320  * errno value on error.
321  */
svc_sock_names(struct svc_serv * serv,char * buf,const size_t buflen,const char * toclose)322 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
323 		   const char *toclose)
324 {
325 	struct svc_sock *svsk, *closesk = NULL;
326 	int len = 0;
327 
328 	if (!serv)
329 		return 0;
330 
331 	spin_lock_bh(&serv->sv_lock);
332 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
333 		int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
334 		if (onelen < 0) {
335 			len = onelen;
336 			break;
337 		}
338 		if (toclose && strcmp(toclose, buf + len) == 0) {
339 			closesk = svsk;
340 			svc_xprt_get(&closesk->sk_xprt);
341 		} else
342 			len += onelen;
343 	}
344 	spin_unlock_bh(&serv->sv_lock);
345 
346 	if (closesk) {
347 		/* Should unregister with portmap, but you cannot
348 		 * unregister just one protocol...
349 		 */
350 		svc_close_xprt(&closesk->sk_xprt);
351 		svc_xprt_put(&closesk->sk_xprt);
352 	} else if (toclose)
353 		return -ENOENT;
354 	return len;
355 }
356 EXPORT_SYMBOL_GPL(svc_sock_names);
357 
358 /*
359  * Check input queue length
360  */
svc_recv_available(struct svc_sock * svsk)361 static int svc_recv_available(struct svc_sock *svsk)
362 {
363 	struct socket	*sock = svsk->sk_sock;
364 	int		avail, err;
365 
366 	err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
367 
368 	return (err >= 0)? avail : err;
369 }
370 
371 /*
372  * Generic recvfrom routine.
373  */
svc_recvfrom(struct svc_rqst * rqstp,struct kvec * iov,int nr,int buflen)374 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
375 			int buflen)
376 {
377 	struct svc_sock *svsk =
378 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
379 	struct msghdr msg = {
380 		.msg_flags	= MSG_DONTWAIT,
381 	};
382 	int len;
383 
384 	rqstp->rq_xprt_hlen = 0;
385 
386 	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
387 				msg.msg_flags);
388 
389 	dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
390 		svsk, iov[0].iov_base, iov[0].iov_len, len);
391 	return len;
392 }
393 
svc_partial_recvfrom(struct svc_rqst * rqstp,struct kvec * iov,int nr,int buflen,unsigned int base)394 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
395 				struct kvec *iov, int nr,
396 				int buflen, unsigned int base)
397 {
398 	size_t save_iovlen;
399 	void *save_iovbase;
400 	unsigned int i;
401 	int ret;
402 
403 	if (base == 0)
404 		return svc_recvfrom(rqstp, iov, nr, buflen);
405 
406 	for (i = 0; i < nr; i++) {
407 		if (iov[i].iov_len > base)
408 			break;
409 		base -= iov[i].iov_len;
410 	}
411 	save_iovlen = iov[i].iov_len;
412 	save_iovbase = iov[i].iov_base;
413 	iov[i].iov_len -= base;
414 	iov[i].iov_base += base;
415 	ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
416 	iov[i].iov_len = save_iovlen;
417 	iov[i].iov_base = save_iovbase;
418 	return ret;
419 }
420 
421 /*
422  * Set socket snd and rcv buffer lengths
423  */
svc_sock_setbufsize(struct socket * sock,unsigned int snd,unsigned int rcv)424 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
425 				unsigned int rcv)
426 {
427 #if 0
428 	mm_segment_t	oldfs;
429 	oldfs = get_fs(); set_fs(KERNEL_DS);
430 	sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
431 			(char*)&snd, sizeof(snd));
432 	sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
433 			(char*)&rcv, sizeof(rcv));
434 #else
435 	/* sock_setsockopt limits use to sysctl_?mem_max,
436 	 * which isn't acceptable.  Until that is made conditional
437 	 * on not having CAP_SYS_RESOURCE or similar, we go direct...
438 	 * DaveM said I could!
439 	 */
440 	lock_sock(sock->sk);
441 	sock->sk->sk_sndbuf = snd * 2;
442 	sock->sk->sk_rcvbuf = rcv * 2;
443 	sock->sk->sk_write_space(sock->sk);
444 	release_sock(sock->sk);
445 #endif
446 }
447 /*
448  * INET callback when data has been received on the socket.
449  */
svc_udp_data_ready(struct sock * sk,int count)450 static void svc_udp_data_ready(struct sock *sk, int count)
451 {
452 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
453 	wait_queue_head_t *wq = sk_sleep(sk);
454 
455 	if (svsk) {
456 		dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
457 			svsk, sk, count,
458 			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
459 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
460 		svc_xprt_enqueue(&svsk->sk_xprt);
461 	}
462 	if (wq && waitqueue_active(wq))
463 		wake_up_interruptible(wq);
464 }
465 
466 /*
467  * INET callback when space is newly available on the socket.
468  */
svc_write_space(struct sock * sk)469 static void svc_write_space(struct sock *sk)
470 {
471 	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
472 	wait_queue_head_t *wq = sk_sleep(sk);
473 
474 	if (svsk) {
475 		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
476 			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
477 		svc_xprt_enqueue(&svsk->sk_xprt);
478 	}
479 
480 	if (wq && waitqueue_active(wq)) {
481 		dprintk("RPC svc_write_space: someone sleeping on %p\n",
482 		       svsk);
483 		wake_up_interruptible(wq);
484 	}
485 }
486 
svc_tcp_write_space(struct sock * sk)487 static void svc_tcp_write_space(struct sock *sk)
488 {
489 	struct socket *sock = sk->sk_socket;
490 
491 	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
492 		clear_bit(SOCK_NOSPACE, &sock->flags);
493 	svc_write_space(sk);
494 }
495 
496 /*
497  * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
498  */
svc_udp_get_dest_address4(struct svc_rqst * rqstp,struct cmsghdr * cmh)499 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
500 				     struct cmsghdr *cmh)
501 {
502 	struct in_pktinfo *pki = CMSG_DATA(cmh);
503 	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
504 
505 	if (cmh->cmsg_type != IP_PKTINFO)
506 		return 0;
507 
508 	daddr->sin_family = AF_INET;
509 	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
510 	return 1;
511 }
512 
513 /*
514  * See net/ipv6/datagram.c : datagram_recv_ctl
515  */
svc_udp_get_dest_address6(struct svc_rqst * rqstp,struct cmsghdr * cmh)516 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
517 				     struct cmsghdr *cmh)
518 {
519 	struct in6_pktinfo *pki = CMSG_DATA(cmh);
520 	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
521 
522 	if (cmh->cmsg_type != IPV6_PKTINFO)
523 		return 0;
524 
525 	daddr->sin6_family = AF_INET6;
526 	daddr->sin6_addr = pki->ipi6_addr;
527 	daddr->sin6_scope_id = pki->ipi6_ifindex;
528 	return 1;
529 }
530 
531 /*
532  * Copy the UDP datagram's destination address to the rqstp structure.
533  * The 'destination' address in this case is the address to which the
534  * peer sent the datagram, i.e. our local address. For multihomed
535  * hosts, this can change from msg to msg. Note that only the IP
536  * address changes, the port number should remain the same.
537  */
svc_udp_get_dest_address(struct svc_rqst * rqstp,struct cmsghdr * cmh)538 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
539 				    struct cmsghdr *cmh)
540 {
541 	switch (cmh->cmsg_level) {
542 	case SOL_IP:
543 		return svc_udp_get_dest_address4(rqstp, cmh);
544 	case SOL_IPV6:
545 		return svc_udp_get_dest_address6(rqstp, cmh);
546 	}
547 
548 	return 0;
549 }
550 
551 /*
552  * Receive a datagram from a UDP socket.
553  */
svc_udp_recvfrom(struct svc_rqst * rqstp)554 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
555 {
556 	struct svc_sock	*svsk =
557 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
558 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
559 	struct sk_buff	*skb;
560 	union {
561 		struct cmsghdr	hdr;
562 		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
563 	} buffer;
564 	struct cmsghdr *cmh = &buffer.hdr;
565 	struct msghdr msg = {
566 		.msg_name = svc_addr(rqstp),
567 		.msg_control = cmh,
568 		.msg_controllen = sizeof(buffer),
569 		.msg_flags = MSG_DONTWAIT,
570 	};
571 	size_t len;
572 	int err;
573 
574 	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
575 	    /* udp sockets need large rcvbuf as all pending
576 	     * requests are still in that buffer.  sndbuf must
577 	     * also be large enough that there is enough space
578 	     * for one reply per thread.  We count all threads
579 	     * rather than threads in a particular pool, which
580 	     * provides an upper bound on the number of threads
581 	     * which will access the socket.
582 	     */
583 	    svc_sock_setbufsize(svsk->sk_sock,
584 				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
585 				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
586 
587 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
588 	skb = NULL;
589 	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
590 			     0, 0, MSG_PEEK | MSG_DONTWAIT);
591 	if (err >= 0)
592 		skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
593 
594 	if (skb == NULL) {
595 		if (err != -EAGAIN) {
596 			/* possibly an icmp error */
597 			dprintk("svc: recvfrom returned error %d\n", -err);
598 			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
599 		}
600 		return -EAGAIN;
601 	}
602 	len = svc_addr_len(svc_addr(rqstp));
603 	if (len == 0)
604 		return -EAFNOSUPPORT;
605 	rqstp->rq_addrlen = len;
606 	if (skb->tstamp.tv64 == 0) {
607 		skb->tstamp = ktime_get_real();
608 		/* Don't enable netstamp, sunrpc doesn't
609 		   need that much accuracy */
610 	}
611 	svsk->sk_sk->sk_stamp = skb->tstamp;
612 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
613 
614 	len  = skb->len - sizeof(struct udphdr);
615 	rqstp->rq_arg.len = len;
616 
617 	rqstp->rq_prot = IPPROTO_UDP;
618 
619 	if (!svc_udp_get_dest_address(rqstp, cmh)) {
620 		if (net_ratelimit())
621 			printk(KERN_WARNING
622 				"svc: received unknown control message %d/%d; "
623 				"dropping RPC reply datagram\n",
624 					cmh->cmsg_level, cmh->cmsg_type);
625 		skb_free_datagram_locked(svsk->sk_sk, skb);
626 		return 0;
627 	}
628 	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
629 
630 	if (skb_is_nonlinear(skb)) {
631 		/* we have to copy */
632 		local_bh_disable();
633 		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
634 			local_bh_enable();
635 			/* checksum error */
636 			skb_free_datagram_locked(svsk->sk_sk, skb);
637 			return 0;
638 		}
639 		local_bh_enable();
640 		skb_free_datagram_locked(svsk->sk_sk, skb);
641 	} else {
642 		/* we can use it in-place */
643 		rqstp->rq_arg.head[0].iov_base = skb->data +
644 			sizeof(struct udphdr);
645 		rqstp->rq_arg.head[0].iov_len = len;
646 		if (skb_checksum_complete(skb)) {
647 			skb_free_datagram_locked(svsk->sk_sk, skb);
648 			return 0;
649 		}
650 		rqstp->rq_xprt_ctxt = skb;
651 	}
652 
653 	rqstp->rq_arg.page_base = 0;
654 	if (len <= rqstp->rq_arg.head[0].iov_len) {
655 		rqstp->rq_arg.head[0].iov_len = len;
656 		rqstp->rq_arg.page_len = 0;
657 		rqstp->rq_respages = rqstp->rq_pages+1;
658 	} else {
659 		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
660 		rqstp->rq_respages = rqstp->rq_pages + 1 +
661 			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
662 	}
663 
664 	if (serv->sv_stats)
665 		serv->sv_stats->netudpcnt++;
666 
667 	return len;
668 }
669 
670 static int
svc_udp_sendto(struct svc_rqst * rqstp)671 svc_udp_sendto(struct svc_rqst *rqstp)
672 {
673 	int		error;
674 
675 	error = svc_sendto(rqstp, &rqstp->rq_res);
676 	if (error == -ECONNREFUSED)
677 		/* ICMP error on earlier request. */
678 		error = svc_sendto(rqstp, &rqstp->rq_res);
679 
680 	return error;
681 }
682 
svc_udp_prep_reply_hdr(struct svc_rqst * rqstp)683 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
684 {
685 }
686 
svc_udp_has_wspace(struct svc_xprt * xprt)687 static int svc_udp_has_wspace(struct svc_xprt *xprt)
688 {
689 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
690 	struct svc_serv	*serv = xprt->xpt_server;
691 	unsigned long required;
692 
693 	/*
694 	 * Set the SOCK_NOSPACE flag before checking the available
695 	 * sock space.
696 	 */
697 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
698 	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
699 	if (required*2 > sock_wspace(svsk->sk_sk))
700 		return 0;
701 	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
702 	return 1;
703 }
704 
svc_udp_accept(struct svc_xprt * xprt)705 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
706 {
707 	BUG();
708 	return NULL;
709 }
710 
svc_udp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)711 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
712 				       struct net *net,
713 				       struct sockaddr *sa, int salen,
714 				       int flags)
715 {
716 	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
717 }
718 
719 static struct svc_xprt_ops svc_udp_ops = {
720 	.xpo_create = svc_udp_create,
721 	.xpo_recvfrom = svc_udp_recvfrom,
722 	.xpo_sendto = svc_udp_sendto,
723 	.xpo_release_rqst = svc_release_skb,
724 	.xpo_detach = svc_sock_detach,
725 	.xpo_free = svc_sock_free,
726 	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
727 	.xpo_has_wspace = svc_udp_has_wspace,
728 	.xpo_accept = svc_udp_accept,
729 };
730 
731 static struct svc_xprt_class svc_udp_class = {
732 	.xcl_name = "udp",
733 	.xcl_owner = THIS_MODULE,
734 	.xcl_ops = &svc_udp_ops,
735 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
736 };
737 
svc_udp_init(struct svc_sock * svsk,struct svc_serv * serv)738 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
739 {
740 	int err, level, optname, one = 1;
741 
742 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
743 		      &svsk->sk_xprt, serv);
744 	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
745 	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
746 	svsk->sk_sk->sk_write_space = svc_write_space;
747 
748 	/* initialise setting must have enough space to
749 	 * receive and respond to one request.
750 	 * svc_udp_recvfrom will re-adjust if necessary
751 	 */
752 	svc_sock_setbufsize(svsk->sk_sock,
753 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
754 			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
755 
756 	/* data might have come in before data_ready set up */
757 	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
758 	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
759 
760 	/* make sure we get destination address info */
761 	switch (svsk->sk_sk->sk_family) {
762 	case AF_INET:
763 		level = SOL_IP;
764 		optname = IP_PKTINFO;
765 		break;
766 	case AF_INET6:
767 		level = SOL_IPV6;
768 		optname = IPV6_RECVPKTINFO;
769 		break;
770 	default:
771 		BUG();
772 	}
773 	err = kernel_setsockopt(svsk->sk_sock, level, optname,
774 					(char *)&one, sizeof(one));
775 	dprintk("svc: kernel_setsockopt returned %d\n", err);
776 }
777 
778 /*
779  * A data_ready event on a listening socket means there's a connection
780  * pending. Do not use state_change as a substitute for it.
781  */
svc_tcp_listen_data_ready(struct sock * sk,int count_unused)782 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
783 {
784 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
785 	wait_queue_head_t *wq;
786 
787 	dprintk("svc: socket %p TCP (listen) state change %d\n",
788 		sk, sk->sk_state);
789 
790 	/*
791 	 * This callback may called twice when a new connection
792 	 * is established as a child socket inherits everything
793 	 * from a parent LISTEN socket.
794 	 * 1) data_ready method of the parent socket will be called
795 	 *    when one of child sockets become ESTABLISHED.
796 	 * 2) data_ready method of the child socket may be called
797 	 *    when it receives data before the socket is accepted.
798 	 * In case of 2, we should ignore it silently.
799 	 */
800 	if (sk->sk_state == TCP_LISTEN) {
801 		if (svsk) {
802 			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
803 			svc_xprt_enqueue(&svsk->sk_xprt);
804 		} else
805 			printk("svc: socket %p: no user data\n", sk);
806 	}
807 
808 	wq = sk_sleep(sk);
809 	if (wq && waitqueue_active(wq))
810 		wake_up_interruptible_all(wq);
811 }
812 
813 /*
814  * A state change on a connected socket means it's dying or dead.
815  */
svc_tcp_state_change(struct sock * sk)816 static void svc_tcp_state_change(struct sock *sk)
817 {
818 	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
819 	wait_queue_head_t *wq = sk_sleep(sk);
820 
821 	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
822 		sk, sk->sk_state, sk->sk_user_data);
823 
824 	if (!svsk)
825 		printk("svc: socket %p: no user data\n", sk);
826 	else {
827 		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
828 		svc_xprt_enqueue(&svsk->sk_xprt);
829 	}
830 	if (wq && waitqueue_active(wq))
831 		wake_up_interruptible_all(wq);
832 }
833 
svc_tcp_data_ready(struct sock * sk,int count)834 static void svc_tcp_data_ready(struct sock *sk, int count)
835 {
836 	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
837 	wait_queue_head_t *wq = sk_sleep(sk);
838 
839 	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
840 		sk, sk->sk_user_data);
841 	if (svsk) {
842 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
843 		svc_xprt_enqueue(&svsk->sk_xprt);
844 	}
845 	if (wq && waitqueue_active(wq))
846 		wake_up_interruptible(wq);
847 }
848 
849 /*
850  * Accept a TCP connection
851  */
svc_tcp_accept(struct svc_xprt * xprt)852 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
853 {
854 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
855 	struct sockaddr_storage addr;
856 	struct sockaddr	*sin = (struct sockaddr *) &addr;
857 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
858 	struct socket	*sock = svsk->sk_sock;
859 	struct socket	*newsock;
860 	struct svc_sock	*newsvsk;
861 	int		err, slen;
862 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
863 
864 	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
865 	if (!sock)
866 		return NULL;
867 
868 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
869 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
870 	if (err < 0) {
871 		if (err == -ENOMEM)
872 			printk(KERN_WARNING "%s: no more sockets!\n",
873 			       serv->sv_name);
874 		else if (err != -EAGAIN && net_ratelimit())
875 			printk(KERN_WARNING "%s: accept failed (err %d)!\n",
876 				   serv->sv_name, -err);
877 		return NULL;
878 	}
879 	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
880 
881 	err = kernel_getpeername(newsock, sin, &slen);
882 	if (err < 0) {
883 		if (net_ratelimit())
884 			printk(KERN_WARNING "%s: peername failed (err %d)!\n",
885 				   serv->sv_name, -err);
886 		goto failed;		/* aborted connection or whatever */
887 	}
888 
889 	/* Ideally, we would want to reject connections from unauthorized
890 	 * hosts here, but when we get encryption, the IP of the host won't
891 	 * tell us anything.  For now just warn about unpriv connections.
892 	 */
893 	if (!svc_port_is_privileged(sin)) {
894 		dprintk(KERN_WARNING
895 			"%s: connect from unprivileged port: %s\n",
896 			serv->sv_name,
897 			__svc_print_addr(sin, buf, sizeof(buf)));
898 	}
899 	dprintk("%s: connect from %s\n", serv->sv_name,
900 		__svc_print_addr(sin, buf, sizeof(buf)));
901 
902 	/* make sure that a write doesn't block forever when
903 	 * low on memory
904 	 */
905 	newsock->sk->sk_sndtimeo = HZ*30;
906 
907 	if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
908 				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
909 		goto failed;
910 	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
911 	err = kernel_getsockname(newsock, sin, &slen);
912 	if (unlikely(err < 0)) {
913 		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
914 		slen = offsetof(struct sockaddr, sa_data);
915 	}
916 	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
917 
918 	if (serv->sv_stats)
919 		serv->sv_stats->nettcpconn++;
920 
921 	return &newsvsk->sk_xprt;
922 
923 failed:
924 	sock_release(newsock);
925 	return NULL;
926 }
927 
svc_tcp_restore_pages(struct svc_sock * svsk,struct svc_rqst * rqstp)928 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
929 {
930 	unsigned int i, len, npages;
931 
932 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
933 		return 0;
934 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
935 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
936 	for (i = 0; i < npages; i++) {
937 		if (rqstp->rq_pages[i] != NULL)
938 			put_page(rqstp->rq_pages[i]);
939 		BUG_ON(svsk->sk_pages[i] == NULL);
940 		rqstp->rq_pages[i] = svsk->sk_pages[i];
941 		svsk->sk_pages[i] = NULL;
942 	}
943 	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
944 	return len;
945 }
946 
svc_tcp_save_pages(struct svc_sock * svsk,struct svc_rqst * rqstp)947 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
948 {
949 	unsigned int i, len, npages;
950 
951 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
952 		return;
953 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
954 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
955 	for (i = 0; i < npages; i++) {
956 		svsk->sk_pages[i] = rqstp->rq_pages[i];
957 		rqstp->rq_pages[i] = NULL;
958 	}
959 }
960 
svc_tcp_clear_pages(struct svc_sock * svsk)961 static void svc_tcp_clear_pages(struct svc_sock *svsk)
962 {
963 	unsigned int i, len, npages;
964 
965 	if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
966 		goto out;
967 	len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
968 	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
969 	for (i = 0; i < npages; i++) {
970 		BUG_ON(svsk->sk_pages[i] == NULL);
971 		put_page(svsk->sk_pages[i]);
972 		svsk->sk_pages[i] = NULL;
973 	}
974 out:
975 	svsk->sk_tcplen = 0;
976 }
977 
978 /*
979  * Receive data.
980  * If we haven't gotten the record length yet, get the next four bytes.
981  * Otherwise try to gobble up as much as possible up to the complete
982  * record length.
983  */
svc_tcp_recv_record(struct svc_sock * svsk,struct svc_rqst * rqstp)984 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
985 {
986 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
987 	unsigned int want;
988 	int len;
989 
990 	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
991 
992 	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
993 		struct kvec	iov;
994 
995 		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
996 		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
997 		iov.iov_len  = want;
998 		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
999 			goto error;
1000 		svsk->sk_tcplen += len;
1001 
1002 		if (len < want) {
1003 			dprintk("svc: short recvfrom while reading record "
1004 				"length (%d of %d)\n", len, want);
1005 			return -EAGAIN;
1006 		}
1007 
1008 		svsk->sk_reclen = ntohl(svsk->sk_reclen);
1009 		if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1010 			/* FIXME: technically, a record can be fragmented,
1011 			 *  and non-terminal fragments will not have the top
1012 			 *  bit set in the fragment length header.
1013 			 *  But apparently no known nfs clients send fragmented
1014 			 *  records. */
1015 			if (net_ratelimit())
1016 				printk(KERN_NOTICE "RPC: multiple fragments "
1017 					"per record not supported\n");
1018 			goto err_delete;
1019 		}
1020 
1021 		svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1022 		dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1023 		if (svsk->sk_reclen > serv->sv_max_mesg) {
1024 			if (net_ratelimit())
1025 				printk(KERN_NOTICE "RPC: "
1026 					"fragment too large: 0x%08lx\n",
1027 					(unsigned long)svsk->sk_reclen);
1028 			goto err_delete;
1029 		}
1030 	}
1031 
1032 	if (svsk->sk_reclen < 8)
1033 		goto err_delete; /* client is nuts. */
1034 
1035 	len = svsk->sk_reclen;
1036 
1037 	return len;
1038 error:
1039 	dprintk("RPC: TCP recv_record got %d\n", len);
1040 	return len;
1041 err_delete:
1042 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1043 	return -EAGAIN;
1044 }
1045 
receive_cb_reply(struct svc_sock * svsk,struct svc_rqst * rqstp)1046 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1047 {
1048 	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1049 	struct rpc_rqst *req = NULL;
1050 	struct kvec *src, *dst;
1051 	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1052 	__be32 xid;
1053 	__be32 calldir;
1054 
1055 	xid = *p++;
1056 	calldir = *p;
1057 
1058 	if (bc_xprt)
1059 		req = xprt_lookup_rqst(bc_xprt, xid);
1060 
1061 	if (!req) {
1062 		printk(KERN_NOTICE
1063 			"%s: Got unrecognized reply: "
1064 			"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1065 			__func__, ntohl(calldir),
1066 			bc_xprt, xid);
1067 		return -EAGAIN;
1068 	}
1069 
1070 	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1071 	/*
1072 	 * XXX!: cheating for now!  Only copying HEAD.
1073 	 * But we know this is good enough for now (in fact, for any
1074 	 * callback reply in the forseeable future).
1075 	 */
1076 	dst = &req->rq_private_buf.head[0];
1077 	src = &rqstp->rq_arg.head[0];
1078 	if (dst->iov_len < src->iov_len)
1079 		return -EAGAIN; /* whatever; just giving up. */
1080 	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1081 	xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1082 	rqstp->rq_arg.len = 0;
1083 	return 0;
1084 }
1085 
copy_pages_to_kvecs(struct kvec * vec,struct page ** pages,int len)1086 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1087 {
1088 	int i = 0;
1089 	int t = 0;
1090 
1091 	while (t < len) {
1092 		vec[i].iov_base = page_address(pages[i]);
1093 		vec[i].iov_len = PAGE_SIZE;
1094 		i++;
1095 		t += PAGE_SIZE;
1096 	}
1097 	return i;
1098 }
1099 
1100 
1101 /*
1102  * Receive data from a TCP socket.
1103  */
svc_tcp_recvfrom(struct svc_rqst * rqstp)1104 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1105 {
1106 	struct svc_sock	*svsk =
1107 		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1108 	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1109 	int		len;
1110 	struct kvec *vec;
1111 	unsigned int want, base;
1112 	__be32 *p;
1113 	__be32 calldir;
1114 	int pnum;
1115 
1116 	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1117 		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1118 		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1119 		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1120 
1121 	len = svc_tcp_recv_record(svsk, rqstp);
1122 	if (len < 0)
1123 		goto error;
1124 
1125 	base = svc_tcp_restore_pages(svsk, rqstp);
1126 	want = svsk->sk_reclen - base;
1127 
1128 	vec = rqstp->rq_vec;
1129 
1130 	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1131 						svsk->sk_reclen);
1132 
1133 	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1134 
1135 	/* Now receive data */
1136 	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1137 	if (len >= 0)
1138 		svsk->sk_tcplen += len;
1139 	if (len != want) {
1140 		svc_tcp_save_pages(svsk, rqstp);
1141 		if (len < 0 && len != -EAGAIN)
1142 			goto err_other;
1143 		dprintk("svc: incomplete TCP record (%d of %d)\n",
1144 			svsk->sk_tcplen, svsk->sk_reclen);
1145 		goto err_noclose;
1146 	}
1147 
1148 	rqstp->rq_arg.len = svsk->sk_reclen;
1149 	rqstp->rq_arg.page_base = 0;
1150 	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1151 		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1152 		rqstp->rq_arg.page_len = 0;
1153 	} else
1154 		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1155 
1156 	rqstp->rq_xprt_ctxt   = NULL;
1157 	rqstp->rq_prot	      = IPPROTO_TCP;
1158 
1159 	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1160 	calldir = p[1];
1161 	if (calldir)
1162 		len = receive_cb_reply(svsk, rqstp);
1163 
1164 	/* Reset TCP read info */
1165 	svsk->sk_reclen = 0;
1166 	svsk->sk_tcplen = 0;
1167 	/* If we have more data, signal svc_xprt_enqueue() to try again */
1168 	if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1169 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1170 
1171 	if (len < 0)
1172 		goto error;
1173 
1174 	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1175 	if (serv->sv_stats)
1176 		serv->sv_stats->nettcpcnt++;
1177 
1178 	dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1179 	return rqstp->rq_arg.len;
1180 
1181 error:
1182 	if (len != -EAGAIN)
1183 		goto err_other;
1184 	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1185 	return -EAGAIN;
1186 err_other:
1187 	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1188 	       svsk->sk_xprt.xpt_server->sv_name, -len);
1189 	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1190 err_noclose:
1191 	return -EAGAIN;	/* record not complete */
1192 }
1193 
1194 /*
1195  * Send out data on TCP socket.
1196  */
svc_tcp_sendto(struct svc_rqst * rqstp)1197 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1198 {
1199 	struct xdr_buf	*xbufp = &rqstp->rq_res;
1200 	int sent;
1201 	__be32 reclen;
1202 
1203 	/* Set up the first element of the reply kvec.
1204 	 * Any other kvecs that may be in use have been taken
1205 	 * care of by the server implementation itself.
1206 	 */
1207 	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1208 	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1209 
1210 	sent = svc_sendto(rqstp, &rqstp->rq_res);
1211 	if (sent != xbufp->len) {
1212 		printk(KERN_NOTICE
1213 		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1214 		       "- shutting down socket\n",
1215 		       rqstp->rq_xprt->xpt_server->sv_name,
1216 		       (sent<0)?"got error":"sent only",
1217 		       sent, xbufp->len);
1218 		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1219 		svc_xprt_enqueue(rqstp->rq_xprt);
1220 		sent = -EAGAIN;
1221 	}
1222 	return sent;
1223 }
1224 
1225 /*
1226  * Setup response header. TCP has a 4B record length field.
1227  */
svc_tcp_prep_reply_hdr(struct svc_rqst * rqstp)1228 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1229 {
1230 	struct kvec *resv = &rqstp->rq_res.head[0];
1231 
1232 	/* tcp needs a space for the record length... */
1233 	svc_putnl(resv, 0);
1234 }
1235 
svc_tcp_has_wspace(struct svc_xprt * xprt)1236 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1237 {
1238 	struct svc_sock *svsk =	container_of(xprt, struct svc_sock, sk_xprt);
1239 	struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1240 	int required;
1241 
1242 	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1243 		return 1;
1244 	required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1245 	if (sk_stream_wspace(svsk->sk_sk) >= required)
1246 		return 1;
1247 	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1248 	return 0;
1249 }
1250 
svc_tcp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)1251 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1252 				       struct net *net,
1253 				       struct sockaddr *sa, int salen,
1254 				       int flags)
1255 {
1256 	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1257 }
1258 
1259 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1260 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1261 					     struct net *, struct sockaddr *,
1262 					     int, int);
1263 static void svc_bc_sock_free(struct svc_xprt *xprt);
1264 
svc_bc_tcp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)1265 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1266 				       struct net *net,
1267 				       struct sockaddr *sa, int salen,
1268 				       int flags)
1269 {
1270 	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1271 }
1272 
svc_bc_tcp_sock_detach(struct svc_xprt * xprt)1273 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1274 {
1275 }
1276 
1277 static struct svc_xprt_ops svc_tcp_bc_ops = {
1278 	.xpo_create = svc_bc_tcp_create,
1279 	.xpo_detach = svc_bc_tcp_sock_detach,
1280 	.xpo_free = svc_bc_sock_free,
1281 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1282 };
1283 
1284 static struct svc_xprt_class svc_tcp_bc_class = {
1285 	.xcl_name = "tcp-bc",
1286 	.xcl_owner = THIS_MODULE,
1287 	.xcl_ops = &svc_tcp_bc_ops,
1288 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1289 };
1290 
svc_init_bc_xprt_sock(void)1291 static void svc_init_bc_xprt_sock(void)
1292 {
1293 	svc_reg_xprt_class(&svc_tcp_bc_class);
1294 }
1295 
svc_cleanup_bc_xprt_sock(void)1296 static void svc_cleanup_bc_xprt_sock(void)
1297 {
1298 	svc_unreg_xprt_class(&svc_tcp_bc_class);
1299 }
1300 #else /* CONFIG_SUNRPC_BACKCHANNEL */
svc_init_bc_xprt_sock(void)1301 static void svc_init_bc_xprt_sock(void)
1302 {
1303 }
1304 
svc_cleanup_bc_xprt_sock(void)1305 static void svc_cleanup_bc_xprt_sock(void)
1306 {
1307 }
1308 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1309 
1310 static struct svc_xprt_ops svc_tcp_ops = {
1311 	.xpo_create = svc_tcp_create,
1312 	.xpo_recvfrom = svc_tcp_recvfrom,
1313 	.xpo_sendto = svc_tcp_sendto,
1314 	.xpo_release_rqst = svc_release_skb,
1315 	.xpo_detach = svc_tcp_sock_detach,
1316 	.xpo_free = svc_sock_free,
1317 	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1318 	.xpo_has_wspace = svc_tcp_has_wspace,
1319 	.xpo_accept = svc_tcp_accept,
1320 };
1321 
1322 static struct svc_xprt_class svc_tcp_class = {
1323 	.xcl_name = "tcp",
1324 	.xcl_owner = THIS_MODULE,
1325 	.xcl_ops = &svc_tcp_ops,
1326 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1327 };
1328 
svc_init_xprt_sock(void)1329 void svc_init_xprt_sock(void)
1330 {
1331 	svc_reg_xprt_class(&svc_tcp_class);
1332 	svc_reg_xprt_class(&svc_udp_class);
1333 	svc_init_bc_xprt_sock();
1334 }
1335 
svc_cleanup_xprt_sock(void)1336 void svc_cleanup_xprt_sock(void)
1337 {
1338 	svc_unreg_xprt_class(&svc_tcp_class);
1339 	svc_unreg_xprt_class(&svc_udp_class);
1340 	svc_cleanup_bc_xprt_sock();
1341 }
1342 
svc_tcp_init(struct svc_sock * svsk,struct svc_serv * serv)1343 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1344 {
1345 	struct sock	*sk = svsk->sk_sk;
1346 
1347 	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1348 		      &svsk->sk_xprt, serv);
1349 	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1350 	if (sk->sk_state == TCP_LISTEN) {
1351 		dprintk("setting up TCP socket for listening\n");
1352 		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1353 		sk->sk_data_ready = svc_tcp_listen_data_ready;
1354 		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1355 	} else {
1356 		dprintk("setting up TCP socket for reading\n");
1357 		sk->sk_state_change = svc_tcp_state_change;
1358 		sk->sk_data_ready = svc_tcp_data_ready;
1359 		sk->sk_write_space = svc_tcp_write_space;
1360 
1361 		svsk->sk_reclen = 0;
1362 		svsk->sk_tcplen = 0;
1363 		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1364 
1365 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1366 
1367 		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1368 		if (sk->sk_state != TCP_ESTABLISHED)
1369 			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1370 	}
1371 }
1372 
svc_sock_update_bufs(struct svc_serv * serv)1373 void svc_sock_update_bufs(struct svc_serv *serv)
1374 {
1375 	/*
1376 	 * The number of server threads has changed. Update
1377 	 * rcvbuf and sndbuf accordingly on all sockets
1378 	 */
1379 	struct svc_sock *svsk;
1380 
1381 	spin_lock_bh(&serv->sv_lock);
1382 	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1383 		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1384 	spin_unlock_bh(&serv->sv_lock);
1385 }
1386 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1387 
1388 /*
1389  * Initialize socket for RPC use and create svc_sock struct
1390  * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1391  */
svc_setup_socket(struct svc_serv * serv,struct socket * sock,int * errp,int flags)1392 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1393 						struct socket *sock,
1394 						int *errp, int flags)
1395 {
1396 	struct svc_sock	*svsk;
1397 	struct sock	*inet;
1398 	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1399 
1400 	dprintk("svc: svc_setup_socket %p\n", sock);
1401 	if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1402 		*errp = -ENOMEM;
1403 		return NULL;
1404 	}
1405 
1406 	inet = sock->sk;
1407 
1408 	/* Register socket with portmapper */
1409 	if (*errp >= 0 && pmap_register)
1410 		*errp = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1411 				     inet->sk_protocol,
1412 				     ntohs(inet_sk(inet)->inet_sport));
1413 
1414 	if (*errp < 0) {
1415 		kfree(svsk);
1416 		return NULL;
1417 	}
1418 
1419 	inet->sk_user_data = svsk;
1420 	svsk->sk_sock = sock;
1421 	svsk->sk_sk = inet;
1422 	svsk->sk_ostate = inet->sk_state_change;
1423 	svsk->sk_odata = inet->sk_data_ready;
1424 	svsk->sk_owspace = inet->sk_write_space;
1425 
1426 	/* Initialize the socket */
1427 	if (sock->type == SOCK_DGRAM)
1428 		svc_udp_init(svsk, serv);
1429 	else {
1430 		/* initialise setting must have enough space to
1431 		 * receive and respond to one request.
1432 		 */
1433 		svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1434 					4 * serv->sv_max_mesg);
1435 		svc_tcp_init(svsk, serv);
1436 	}
1437 
1438 	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1439 				svsk, svsk->sk_sk);
1440 
1441 	return svsk;
1442 }
1443 
svc_alien_sock(struct net * net,int fd)1444 bool svc_alien_sock(struct net *net, int fd)
1445 {
1446 	int err;
1447 	struct socket *sock = sockfd_lookup(fd, &err);
1448 	bool ret = false;
1449 
1450 	if (!sock)
1451 		goto out;
1452 	if (sock_net(sock->sk) != net)
1453 		ret = true;
1454 	sockfd_put(sock);
1455 out:
1456 	return ret;
1457 }
1458 EXPORT_SYMBOL_GPL(svc_alien_sock);
1459 
1460 /**
1461  * svc_addsock - add a listener socket to an RPC service
1462  * @serv: pointer to RPC service to which to add a new listener
1463  * @fd: file descriptor of the new listener
1464  * @name_return: pointer to buffer to fill in with name of listener
1465  * @len: size of the buffer
1466  *
1467  * Fills in socket name and returns positive length of name if successful.
1468  * Name is terminated with '\n'.  On error, returns a negative errno
1469  * value.
1470  */
svc_addsock(struct svc_serv * serv,const int fd,char * name_return,const size_t len)1471 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1472 		const size_t len)
1473 {
1474 	int err = 0;
1475 	struct socket *so = sockfd_lookup(fd, &err);
1476 	struct svc_sock *svsk = NULL;
1477 
1478 	if (!so)
1479 		return err;
1480 	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1481 		err =  -EAFNOSUPPORT;
1482 	else if (so->sk->sk_protocol != IPPROTO_TCP &&
1483 	    so->sk->sk_protocol != IPPROTO_UDP)
1484 		err =  -EPROTONOSUPPORT;
1485 	else if (so->state > SS_UNCONNECTED)
1486 		err = -EISCONN;
1487 	else {
1488 		if (!try_module_get(THIS_MODULE))
1489 			err = -ENOENT;
1490 		else
1491 			svsk = svc_setup_socket(serv, so, &err,
1492 						SVC_SOCK_DEFAULTS);
1493 		if (svsk) {
1494 			struct sockaddr_storage addr;
1495 			struct sockaddr *sin = (struct sockaddr *)&addr;
1496 			int salen;
1497 			if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1498 				svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1499 			clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1500 			spin_lock_bh(&serv->sv_lock);
1501 			list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1502 			spin_unlock_bh(&serv->sv_lock);
1503 			svc_xprt_received(&svsk->sk_xprt);
1504 			err = 0;
1505 		} else
1506 			module_put(THIS_MODULE);
1507 	}
1508 	if (err) {
1509 		sockfd_put(so);
1510 		return err;
1511 	}
1512 	return svc_one_sock_name(svsk, name_return, len);
1513 }
1514 EXPORT_SYMBOL_GPL(svc_addsock);
1515 
1516 /*
1517  * Create socket for RPC service.
1518  */
svc_create_socket(struct svc_serv * serv,int protocol,struct net * net,struct sockaddr * sin,int len,int flags)1519 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1520 					  int protocol,
1521 					  struct net *net,
1522 					  struct sockaddr *sin, int len,
1523 					  int flags)
1524 {
1525 	struct svc_sock	*svsk;
1526 	struct socket	*sock;
1527 	int		error;
1528 	int		type;
1529 	struct sockaddr_storage addr;
1530 	struct sockaddr *newsin = (struct sockaddr *)&addr;
1531 	int		newlen;
1532 	int		family;
1533 	int		val;
1534 	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1535 
1536 	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1537 			serv->sv_program->pg_name, protocol,
1538 			__svc_print_addr(sin, buf, sizeof(buf)));
1539 
1540 	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1541 		printk(KERN_WARNING "svc: only UDP and TCP "
1542 				"sockets supported\n");
1543 		return ERR_PTR(-EINVAL);
1544 	}
1545 
1546 	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1547 	switch (sin->sa_family) {
1548 	case AF_INET6:
1549 		family = PF_INET6;
1550 		break;
1551 	case AF_INET:
1552 		family = PF_INET;
1553 		break;
1554 	default:
1555 		return ERR_PTR(-EINVAL);
1556 	}
1557 
1558 	error = __sock_create(net, family, type, protocol, &sock, 1);
1559 	if (error < 0)
1560 		return ERR_PTR(error);
1561 
1562 	svc_reclassify_socket(sock);
1563 
1564 	/*
1565 	 * If this is an PF_INET6 listener, we want to avoid
1566 	 * getting requests from IPv4 remotes.  Those should
1567 	 * be shunted to a PF_INET listener via rpcbind.
1568 	 */
1569 	val = 1;
1570 	if (family == PF_INET6)
1571 		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1572 					(char *)&val, sizeof(val));
1573 
1574 	if (type == SOCK_STREAM)
1575 		sock->sk->sk_reuse = 1;		/* allow address reuse */
1576 	error = kernel_bind(sock, sin, len);
1577 	if (error < 0)
1578 		goto bummer;
1579 
1580 	newlen = len;
1581 	error = kernel_getsockname(sock, newsin, &newlen);
1582 	if (error < 0)
1583 		goto bummer;
1584 
1585 	if (protocol == IPPROTO_TCP) {
1586 		if ((error = kernel_listen(sock, 64)) < 0)
1587 			goto bummer;
1588 	}
1589 
1590 	if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1591 		svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1592 		return (struct svc_xprt *)svsk;
1593 	}
1594 
1595 bummer:
1596 	dprintk("svc: svc_create_socket error = %d\n", -error);
1597 	sock_release(sock);
1598 	return ERR_PTR(error);
1599 }
1600 
1601 /*
1602  * Detach the svc_sock from the socket so that no
1603  * more callbacks occur.
1604  */
svc_sock_detach(struct svc_xprt * xprt)1605 static void svc_sock_detach(struct svc_xprt *xprt)
1606 {
1607 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1608 	struct sock *sk = svsk->sk_sk;
1609 	wait_queue_head_t *wq;
1610 
1611 	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1612 
1613 	/* put back the old socket callbacks */
1614 	sk->sk_state_change = svsk->sk_ostate;
1615 	sk->sk_data_ready = svsk->sk_odata;
1616 	sk->sk_write_space = svsk->sk_owspace;
1617 
1618 	wq = sk_sleep(sk);
1619 	if (wq && waitqueue_active(wq))
1620 		wake_up_interruptible(wq);
1621 }
1622 
1623 /*
1624  * Disconnect the socket, and reset the callbacks
1625  */
svc_tcp_sock_detach(struct svc_xprt * xprt)1626 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1627 {
1628 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1629 
1630 	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1631 
1632 	svc_sock_detach(xprt);
1633 
1634 	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1635 		svc_tcp_clear_pages(svsk);
1636 		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1637 	}
1638 }
1639 
1640 /*
1641  * Free the svc_sock's socket resources and the svc_sock itself.
1642  */
svc_sock_free(struct svc_xprt * xprt)1643 static void svc_sock_free(struct svc_xprt *xprt)
1644 {
1645 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1646 	dprintk("svc: svc_sock_free(%p)\n", svsk);
1647 
1648 	if (svsk->sk_sock->file)
1649 		sockfd_put(svsk->sk_sock);
1650 	else
1651 		sock_release(svsk->sk_sock);
1652 	kfree(svsk);
1653 }
1654 
1655 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1656 /*
1657  * Create a back channel svc_xprt which shares the fore channel socket.
1658  */
svc_bc_create_socket(struct svc_serv * serv,int protocol,struct net * net,struct sockaddr * sin,int len,int flags)1659 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1660 					     int protocol,
1661 					     struct net *net,
1662 					     struct sockaddr *sin, int len,
1663 					     int flags)
1664 {
1665 	struct svc_sock *svsk;
1666 	struct svc_xprt *xprt;
1667 
1668 	if (protocol != IPPROTO_TCP) {
1669 		printk(KERN_WARNING "svc: only TCP sockets"
1670 			" supported on shared back channel\n");
1671 		return ERR_PTR(-EINVAL);
1672 	}
1673 
1674 	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1675 	if (!svsk)
1676 		return ERR_PTR(-ENOMEM);
1677 
1678 	xprt = &svsk->sk_xprt;
1679 	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1680 
1681 	serv->sv_bc_xprt = xprt;
1682 
1683 	return xprt;
1684 }
1685 
1686 /*
1687  * Free a back channel svc_sock.
1688  */
svc_bc_sock_free(struct svc_xprt * xprt)1689 static void svc_bc_sock_free(struct svc_xprt *xprt)
1690 {
1691 	if (xprt)
1692 		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1693 }
1694 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1695