Loading...
1/*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
9 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/module.h>
25#include <linux/errno.h>
26#include <linux/fcntl.h>
27#include <linux/net.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/udp.h>
31#include <linux/tcp.h>
32#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/netdevice.h>
35#include <linux/skbuff.h>
36#include <linux/file.h>
37#include <linux/freezer.h>
38#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/ip.h>
41#include <net/ipv6.h>
42#include <net/tcp.h>
43#include <net/tcp_states.h>
44#include <asm/uaccess.h>
45#include <asm/ioctls.h>
46#include <trace/events/skb.h>
47
48#include <linux/sunrpc/types.h>
49#include <linux/sunrpc/clnt.h>
50#include <linux/sunrpc/xdr.h>
51#include <linux/sunrpc/msg_prot.h>
52#include <linux/sunrpc/svcsock.h>
53#include <linux/sunrpc/stats.h>
54#include <linux/sunrpc/xprt.h>
55
56#include "sunrpc.h"
57
58#define RPCDBG_FACILITY RPCDBG_SVCXPRT
59
60
61static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
62 int flags);
63static void svc_udp_data_ready(struct sock *);
64static int svc_udp_recvfrom(struct svc_rqst *);
65static int svc_udp_sendto(struct svc_rqst *);
66static void svc_sock_detach(struct svc_xprt *);
67static void svc_tcp_sock_detach(struct svc_xprt *);
68static void svc_sock_free(struct svc_xprt *);
69
70static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73#if defined(CONFIG_SUNRPC_BACKCHANNEL)
74static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77static void svc_bc_sock_free(struct svc_xprt *xprt);
78#endif /* CONFIG_SUNRPC_BACKCHANNEL */
79
80#ifdef CONFIG_DEBUG_LOCK_ALLOC
81static struct lock_class_key svc_key[2];
82static struct lock_class_key svc_slock_key[2];
83
84static void svc_reclassify_socket(struct socket *sock)
85{
86 struct sock *sk = sock->sk;
87
88 WARN_ON_ONCE(sock_owned_by_user(sk));
89 if (sock_owned_by_user(sk))
90 return;
91
92 switch (sk->sk_family) {
93 case AF_INET:
94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
95 &svc_slock_key[0],
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 &svc_key[0]);
98 break;
99
100 case AF_INET6:
101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
102 &svc_slock_key[1],
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
104 &svc_key[1]);
105 break;
106
107 default:
108 BUG();
109 }
110}
111#else
112static void svc_reclassify_socket(struct socket *sock)
113{
114}
115#endif
116
117/*
118 * Release an skbuff after use
119 */
120static void svc_release_skb(struct svc_rqst *rqstp)
121{
122 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
123
124 if (skb) {
125 struct svc_sock *svsk =
126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
127 rqstp->rq_xprt_ctxt = NULL;
128
129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
130 skb_free_datagram_locked(svsk->sk_sk, skb);
131 }
132}
133
134union svc_pktinfo_u {
135 struct in_pktinfo pkti;
136 struct in6_pktinfo pkti6;
137};
138#define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
140
141static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
142{
143 struct svc_sock *svsk =
144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 switch (svsk->sk_sk->sk_family) {
146 case AF_INET: {
147 struct in_pktinfo *pki = CMSG_DATA(cmh);
148
149 cmh->cmsg_level = SOL_IP;
150 cmh->cmsg_type = IP_PKTINFO;
151 pki->ipi_ifindex = 0;
152 pki->ipi_spec_dst.s_addr =
153 svc_daddr_in(rqstp)->sin_addr.s_addr;
154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
155 }
156 break;
157
158 case AF_INET6: {
159 struct in6_pktinfo *pki = CMSG_DATA(cmh);
160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
161
162 cmh->cmsg_level = SOL_IPV6;
163 cmh->cmsg_type = IPV6_PKTINFO;
164 pki->ipi6_ifindex = daddr->sin6_scope_id;
165 pki->ipi6_addr = daddr->sin6_addr;
166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 }
168 break;
169 }
170}
171
172/*
173 * send routine intended to be shared by the fore- and back-channel
174 */
175int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 struct page *headpage, unsigned long headoffset,
177 struct page *tailpage, unsigned long tailoffset)
178{
179 int result;
180 int size;
181 struct page **ppage = xdr->pages;
182 size_t base = xdr->page_base;
183 unsigned int pglen = xdr->page_len;
184 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
185 int slen;
186 int len = 0;
187
188 slen = xdr->len;
189
190 /* send head */
191 if (slen == xdr->head[0].iov_len)
192 flags = 0;
193 len = kernel_sendpage(sock, headpage, headoffset,
194 xdr->head[0].iov_len, flags);
195 if (len != xdr->head[0].iov_len)
196 goto out;
197 slen -= xdr->head[0].iov_len;
198 if (slen == 0)
199 goto out;
200
201 /* send page data */
202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
203 while (pglen > 0) {
204 if (slen == size)
205 flags = 0;
206 result = kernel_sendpage(sock, *ppage, base, size, flags);
207 if (result > 0)
208 len += result;
209 if (result != size)
210 goto out;
211 slen -= size;
212 pglen -= size;
213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 base = 0;
215 ppage++;
216 }
217
218 /* send tail */
219 if (xdr->tail[0].iov_len) {
220 result = kernel_sendpage(sock, tailpage, tailoffset,
221 xdr->tail[0].iov_len, 0);
222 if (result > 0)
223 len += result;
224 }
225
226out:
227 return len;
228}
229
230
231/*
232 * Generic sendto routine
233 */
234static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
235{
236 struct svc_sock *svsk =
237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 struct socket *sock = svsk->sk_sock;
239 union {
240 struct cmsghdr hdr;
241 long all[SVC_PKTINFO_SPACE / sizeof(long)];
242 } buffer;
243 struct cmsghdr *cmh = &buffer.hdr;
244 int len = 0;
245 unsigned long tailoff;
246 unsigned long headoff;
247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
248
249 if (rqstp->rq_prot == IPPROTO_UDP) {
250 struct msghdr msg = {
251 .msg_name = &rqstp->rq_addr,
252 .msg_namelen = rqstp->rq_addrlen,
253 .msg_control = cmh,
254 .msg_controllen = sizeof(buffer),
255 .msg_flags = MSG_MORE,
256 };
257
258 svc_set_cmsg_data(rqstp, cmh);
259
260 if (sock_sendmsg(sock, &msg) < 0)
261 goto out;
262 }
263
264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
265 headoff = 0;
266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 rqstp->rq_respages[0], tailoff);
268
269out:
270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
273
274 return len;
275}
276
277/*
278 * Report socket names for nfsdfs
279 */
280static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
281{
282 const struct sock *sk = svsk->sk_sk;
283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
284 "udp" : "tcp";
285 int len;
286
287 switch (sk->sk_family) {
288 case PF_INET:
289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
290 proto_name,
291 &inet_sk(sk)->inet_rcv_saddr,
292 inet_sk(sk)->inet_num);
293 break;
294#if IS_ENABLED(CONFIG_IPV6)
295 case PF_INET6:
296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
297 proto_name,
298 &sk->sk_v6_rcv_saddr,
299 inet_sk(sk)->inet_num);
300 break;
301#endif
302 default:
303 len = snprintf(buf, remaining, "*unknown-%d*\n",
304 sk->sk_family);
305 }
306
307 if (len >= remaining) {
308 *buf = '\0';
309 return -ENAMETOOLONG;
310 }
311 return len;
312}
313
314/*
315 * Generic recvfrom routine.
316 */
317static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
318 int buflen)
319{
320 struct svc_sock *svsk =
321 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
322 struct msghdr msg = {
323 .msg_flags = MSG_DONTWAIT,
324 };
325 int len;
326
327 rqstp->rq_xprt_hlen = 0;
328
329 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
330 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
331 msg.msg_flags);
332 /* If we read a full record, then assume there may be more
333 * data to read (stream based sockets only!)
334 */
335 if (len == buflen)
336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
337
338 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
339 svsk, iov[0].iov_base, iov[0].iov_len, len);
340 return len;
341}
342
343static int svc_partial_recvfrom(struct svc_rqst *rqstp,
344 struct kvec *iov, int nr,
345 int buflen, unsigned int base)
346{
347 size_t save_iovlen;
348 void *save_iovbase;
349 unsigned int i;
350 int ret;
351
352 if (base == 0)
353 return svc_recvfrom(rqstp, iov, nr, buflen);
354
355 for (i = 0; i < nr; i++) {
356 if (iov[i].iov_len > base)
357 break;
358 base -= iov[i].iov_len;
359 }
360 save_iovlen = iov[i].iov_len;
361 save_iovbase = iov[i].iov_base;
362 iov[i].iov_len -= base;
363 iov[i].iov_base += base;
364 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
365 iov[i].iov_len = save_iovlen;
366 iov[i].iov_base = save_iovbase;
367 return ret;
368}
369
370/*
371 * Set socket snd and rcv buffer lengths
372 */
373static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
374 unsigned int rcv)
375{
376#if 0
377 mm_segment_t oldfs;
378 oldfs = get_fs(); set_fs(KERNEL_DS);
379 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
380 (char*)&snd, sizeof(snd));
381 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
382 (char*)&rcv, sizeof(rcv));
383#else
384 /* sock_setsockopt limits use to sysctl_?mem_max,
385 * which isn't acceptable. Until that is made conditional
386 * on not having CAP_SYS_RESOURCE or similar, we go direct...
387 * DaveM said I could!
388 */
389 lock_sock(sock->sk);
390 sock->sk->sk_sndbuf = snd * 2;
391 sock->sk->sk_rcvbuf = rcv * 2;
392 sock->sk->sk_write_space(sock->sk);
393 release_sock(sock->sk);
394#endif
395}
396
397static int svc_sock_secure_port(struct svc_rqst *rqstp)
398{
399 return svc_port_is_privileged(svc_addr(rqstp));
400}
401
402static bool sunrpc_waitqueue_active(wait_queue_head_t *wq)
403{
404 if (!wq)
405 return false;
406 /*
407 * There should normally be a memory * barrier here--see
408 * wq_has_sleeper().
409 *
410 * It appears that isn't currently necessary, though, basically
411 * because callers all appear to have sufficient memory barriers
412 * between the time the relevant change is made and the
413 * time they call these callbacks.
414 *
415 * The nfsd code itself doesn't actually explicitly wait on
416 * these waitqueues, but it may wait on them for example in
417 * sendpage() or sendmsg() calls. (And those may be the only
418 * places, since it it uses nonblocking reads.)
419 *
420 * Maybe we should add the memory barriers anyway, but these are
421 * hot paths so we'd need to be convinced there's no sigificant
422 * penalty.
423 */
424 return waitqueue_active(wq);
425}
426
427/*
428 * INET callback when data has been received on the socket.
429 */
430static void svc_udp_data_ready(struct sock *sk)
431{
432 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
433 wait_queue_head_t *wq = sk_sleep(sk);
434
435 if (svsk) {
436 dprintk("svc: socket %p(inet %p), busy=%d\n",
437 svsk, sk,
438 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
439 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
440 svc_xprt_enqueue(&svsk->sk_xprt);
441 }
442 if (sunrpc_waitqueue_active(wq))
443 wake_up_interruptible(wq);
444}
445
446/*
447 * INET callback when space is newly available on the socket.
448 */
449static void svc_write_space(struct sock *sk)
450{
451 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
452 wait_queue_head_t *wq = sk_sleep(sk);
453
454 if (svsk) {
455 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
456 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
457 svc_xprt_enqueue(&svsk->sk_xprt);
458 }
459
460 if (sunrpc_waitqueue_active(wq)) {
461 dprintk("RPC svc_write_space: someone sleeping on %p\n",
462 svsk);
463 wake_up_interruptible(wq);
464 }
465}
466
467static int svc_tcp_has_wspace(struct svc_xprt *xprt)
468{
469 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
470 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
471 int required;
472
473 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
474 return 1;
475 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
476 if (sk_stream_wspace(svsk->sk_sk) >= required ||
477 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
478 atomic_read(&xprt->xpt_reserved) == 0))
479 return 1;
480 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
481 return 0;
482}
483
484static void svc_tcp_write_space(struct sock *sk)
485{
486 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
487 struct socket *sock = sk->sk_socket;
488
489 if (!sk_stream_is_writeable(sk) || !sock)
490 return;
491 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt))
492 clear_bit(SOCK_NOSPACE, &sock->flags);
493 svc_write_space(sk);
494}
495
496static void svc_tcp_adjust_wspace(struct svc_xprt *xprt)
497{
498 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
499
500 if (svc_tcp_has_wspace(xprt))
501 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
502}
503
504/*
505 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
506 */
507static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
508 struct cmsghdr *cmh)
509{
510 struct in_pktinfo *pki = CMSG_DATA(cmh);
511 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
512
513 if (cmh->cmsg_type != IP_PKTINFO)
514 return 0;
515
516 daddr->sin_family = AF_INET;
517 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
518 return 1;
519}
520
521/*
522 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
523 */
524static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
525 struct cmsghdr *cmh)
526{
527 struct in6_pktinfo *pki = CMSG_DATA(cmh);
528 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
529
530 if (cmh->cmsg_type != IPV6_PKTINFO)
531 return 0;
532
533 daddr->sin6_family = AF_INET6;
534 daddr->sin6_addr = pki->ipi6_addr;
535 daddr->sin6_scope_id = pki->ipi6_ifindex;
536 return 1;
537}
538
539/*
540 * Copy the UDP datagram's destination address to the rqstp structure.
541 * The 'destination' address in this case is the address to which the
542 * peer sent the datagram, i.e. our local address. For multihomed
543 * hosts, this can change from msg to msg. Note that only the IP
544 * address changes, the port number should remain the same.
545 */
546static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
547 struct cmsghdr *cmh)
548{
549 switch (cmh->cmsg_level) {
550 case SOL_IP:
551 return svc_udp_get_dest_address4(rqstp, cmh);
552 case SOL_IPV6:
553 return svc_udp_get_dest_address6(rqstp, cmh);
554 }
555
556 return 0;
557}
558
559/*
560 * Receive a datagram from a UDP socket.
561 */
562static int svc_udp_recvfrom(struct svc_rqst *rqstp)
563{
564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
567 struct sk_buff *skb;
568 union {
569 struct cmsghdr hdr;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
571 } buffer;
572 struct cmsghdr *cmh = &buffer.hdr;
573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
575 .msg_control = cmh,
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
578 };
579 size_t len;
580 int err;
581
582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
590 */
591 svc_sock_setbufsize(svsk->sk_sock,
592 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
593 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
594
595 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
596 skb = NULL;
597 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
598 0, 0, MSG_PEEK | MSG_DONTWAIT);
599 if (err >= 0)
600 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
601
602 if (skb == NULL) {
603 if (err != -EAGAIN) {
604 /* possibly an icmp error */
605 dprintk("svc: recvfrom returned error %d\n", -err);
606 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
607 }
608 return 0;
609 }
610 len = svc_addr_len(svc_addr(rqstp));
611 rqstp->rq_addrlen = len;
612 if (skb->tstamp.tv64 == 0) {
613 skb->tstamp = ktime_get_real();
614 /* Don't enable netstamp, sunrpc doesn't
615 need that much accuracy */
616 }
617 svsk->sk_sk->sk_stamp = skb->tstamp;
618 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
619
620 len = skb->len - sizeof(struct udphdr);
621 rqstp->rq_arg.len = len;
622
623 rqstp->rq_prot = IPPROTO_UDP;
624
625 if (!svc_udp_get_dest_address(rqstp, cmh)) {
626 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
627 cmh->cmsg_level, cmh->cmsg_type);
628 goto out_free;
629 }
630 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
631
632 if (skb_is_nonlinear(skb)) {
633 /* we have to copy */
634 local_bh_disable();
635 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
636 local_bh_enable();
637 /* checksum error */
638 goto out_free;
639 }
640 local_bh_enable();
641 skb_free_datagram_locked(svsk->sk_sk, skb);
642 } else {
643 /* we can use it in-place */
644 rqstp->rq_arg.head[0].iov_base = skb->data +
645 sizeof(struct udphdr);
646 rqstp->rq_arg.head[0].iov_len = len;
647 if (skb_checksum_complete(skb))
648 goto out_free;
649 rqstp->rq_xprt_ctxt = skb;
650 }
651
652 rqstp->rq_arg.page_base = 0;
653 if (len <= rqstp->rq_arg.head[0].iov_len) {
654 rqstp->rq_arg.head[0].iov_len = len;
655 rqstp->rq_arg.page_len = 0;
656 rqstp->rq_respages = rqstp->rq_pages+1;
657 } else {
658 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
659 rqstp->rq_respages = rqstp->rq_pages + 1 +
660 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
661 }
662 rqstp->rq_next_page = rqstp->rq_respages+1;
663
664 if (serv->sv_stats)
665 serv->sv_stats->netudpcnt++;
666
667 return len;
668out_free:
669 trace_kfree_skb(skb, svc_udp_recvfrom);
670 skb_free_datagram_locked(svsk->sk_sk, skb);
671 return 0;
672}
673
674static int
675svc_udp_sendto(struct svc_rqst *rqstp)
676{
677 int error;
678
679 error = svc_sendto(rqstp, &rqstp->rq_res);
680 if (error == -ECONNREFUSED)
681 /* ICMP error on earlier request. */
682 error = svc_sendto(rqstp, &rqstp->rq_res);
683
684 return error;
685}
686
687static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
688{
689}
690
691static int svc_udp_has_wspace(struct svc_xprt *xprt)
692{
693 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
694 struct svc_serv *serv = xprt->xpt_server;
695 unsigned long required;
696
697 /*
698 * Set the SOCK_NOSPACE flag before checking the available
699 * sock space.
700 */
701 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
702 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
703 if (required*2 > sock_wspace(svsk->sk_sk))
704 return 0;
705 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
706 return 1;
707}
708
709static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
710{
711 BUG();
712 return NULL;
713}
714
715static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
716 struct net *net,
717 struct sockaddr *sa, int salen,
718 int flags)
719{
720 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
721}
722
723static struct svc_xprt_ops svc_udp_ops = {
724 .xpo_create = svc_udp_create,
725 .xpo_recvfrom = svc_udp_recvfrom,
726 .xpo_sendto = svc_udp_sendto,
727 .xpo_release_rqst = svc_release_skb,
728 .xpo_detach = svc_sock_detach,
729 .xpo_free = svc_sock_free,
730 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
731 .xpo_has_wspace = svc_udp_has_wspace,
732 .xpo_accept = svc_udp_accept,
733 .xpo_secure_port = svc_sock_secure_port,
734};
735
736static struct svc_xprt_class svc_udp_class = {
737 .xcl_name = "udp",
738 .xcl_owner = THIS_MODULE,
739 .xcl_ops = &svc_udp_ops,
740 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
741 .xcl_ident = XPRT_TRANSPORT_UDP,
742};
743
744static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
745{
746 int err, level, optname, one = 1;
747
748 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
749 &svsk->sk_xprt, serv);
750 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
751 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
752 svsk->sk_sk->sk_write_space = svc_write_space;
753
754 /* initialise setting must have enough space to
755 * receive and respond to one request.
756 * svc_udp_recvfrom will re-adjust if necessary
757 */
758 svc_sock_setbufsize(svsk->sk_sock,
759 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
760 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
761
762 /* data might have come in before data_ready set up */
763 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
764 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
765
766 /* make sure we get destination address info */
767 switch (svsk->sk_sk->sk_family) {
768 case AF_INET:
769 level = SOL_IP;
770 optname = IP_PKTINFO;
771 break;
772 case AF_INET6:
773 level = SOL_IPV6;
774 optname = IPV6_RECVPKTINFO;
775 break;
776 default:
777 BUG();
778 }
779 err = kernel_setsockopt(svsk->sk_sock, level, optname,
780 (char *)&one, sizeof(one));
781 dprintk("svc: kernel_setsockopt returned %d\n", err);
782}
783
784/*
785 * A data_ready event on a listening socket means there's a connection
786 * pending. Do not use state_change as a substitute for it.
787 */
788static void svc_tcp_listen_data_ready(struct sock *sk)
789{
790 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
791 wait_queue_head_t *wq;
792
793 dprintk("svc: socket %p TCP (listen) state change %d\n",
794 sk, sk->sk_state);
795
796 /*
797 * This callback may called twice when a new connection
798 * is established as a child socket inherits everything
799 * from a parent LISTEN socket.
800 * 1) data_ready method of the parent socket will be called
801 * when one of child sockets become ESTABLISHED.
802 * 2) data_ready method of the child socket may be called
803 * when it receives data before the socket is accepted.
804 * In case of 2, we should ignore it silently.
805 */
806 if (sk->sk_state == TCP_LISTEN) {
807 if (svsk) {
808 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
809 svc_xprt_enqueue(&svsk->sk_xprt);
810 } else
811 printk("svc: socket %p: no user data\n", sk);
812 }
813
814 wq = sk_sleep(sk);
815 if (sunrpc_waitqueue_active(wq))
816 wake_up_interruptible_all(wq);
817}
818
819/*
820 * A state change on a connected socket means it's dying or dead.
821 */
822static void svc_tcp_state_change(struct sock *sk)
823{
824 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
825 wait_queue_head_t *wq = sk_sleep(sk);
826
827 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
828 sk, sk->sk_state, sk->sk_user_data);
829
830 if (!svsk)
831 printk("svc: socket %p: no user data\n", sk);
832 else {
833 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
834 svc_xprt_enqueue(&svsk->sk_xprt);
835 }
836 if (sunrpc_waitqueue_active(wq))
837 wake_up_interruptible_all(wq);
838}
839
840static void svc_tcp_data_ready(struct sock *sk)
841{
842 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
843 wait_queue_head_t *wq = sk_sleep(sk);
844
845 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
846 sk, sk->sk_user_data);
847 if (svsk) {
848 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
849 svc_xprt_enqueue(&svsk->sk_xprt);
850 }
851 if (sunrpc_waitqueue_active(wq))
852 wake_up_interruptible(wq);
853}
854
855/*
856 * Accept a TCP connection
857 */
858static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
859{
860 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
861 struct sockaddr_storage addr;
862 struct sockaddr *sin = (struct sockaddr *) &addr;
863 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
864 struct socket *sock = svsk->sk_sock;
865 struct socket *newsock;
866 struct svc_sock *newsvsk;
867 int err, slen;
868 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
869
870 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
871 if (!sock)
872 return NULL;
873
874 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
875 err = kernel_accept(sock, &newsock, O_NONBLOCK);
876 if (err < 0) {
877 if (err == -ENOMEM)
878 printk(KERN_WARNING "%s: no more sockets!\n",
879 serv->sv_name);
880 else if (err != -EAGAIN)
881 net_warn_ratelimited("%s: accept failed (err %d)!\n",
882 serv->sv_name, -err);
883 return NULL;
884 }
885 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
886
887 err = kernel_getpeername(newsock, sin, &slen);
888 if (err < 0) {
889 net_warn_ratelimited("%s: peername failed (err %d)!\n",
890 serv->sv_name, -err);
891 goto failed; /* aborted connection or whatever */
892 }
893
894 /* Ideally, we would want to reject connections from unauthorized
895 * hosts here, but when we get encryption, the IP of the host won't
896 * tell us anything. For now just warn about unpriv connections.
897 */
898 if (!svc_port_is_privileged(sin)) {
899 dprintk("%s: connect from unprivileged port: %s\n",
900 serv->sv_name,
901 __svc_print_addr(sin, buf, sizeof(buf)));
902 }
903 dprintk("%s: connect from %s\n", serv->sv_name,
904 __svc_print_addr(sin, buf, sizeof(buf)));
905
906 /* make sure that a write doesn't block forever when
907 * low on memory
908 */
909 newsock->sk->sk_sndtimeo = HZ*30;
910
911 newsvsk = svc_setup_socket(serv, newsock,
912 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
913 if (IS_ERR(newsvsk))
914 goto failed;
915 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
916 err = kernel_getsockname(newsock, sin, &slen);
917 if (unlikely(err < 0)) {
918 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
919 slen = offsetof(struct sockaddr, sa_data);
920 }
921 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
922
923 if (sock_is_loopback(newsock->sk))
924 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
925 else
926 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
927 if (serv->sv_stats)
928 serv->sv_stats->nettcpconn++;
929
930 return &newsvsk->sk_xprt;
931
932failed:
933 sock_release(newsock);
934 return NULL;
935}
936
937static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
938{
939 unsigned int i, len, npages;
940
941 if (svsk->sk_datalen == 0)
942 return 0;
943 len = svsk->sk_datalen;
944 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
945 for (i = 0; i < npages; i++) {
946 if (rqstp->rq_pages[i] != NULL)
947 put_page(rqstp->rq_pages[i]);
948 BUG_ON(svsk->sk_pages[i] == NULL);
949 rqstp->rq_pages[i] = svsk->sk_pages[i];
950 svsk->sk_pages[i] = NULL;
951 }
952 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
953 return len;
954}
955
956static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
957{
958 unsigned int i, len, npages;
959
960 if (svsk->sk_datalen == 0)
961 return;
962 len = svsk->sk_datalen;
963 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
964 for (i = 0; i < npages; i++) {
965 svsk->sk_pages[i] = rqstp->rq_pages[i];
966 rqstp->rq_pages[i] = NULL;
967 }
968}
969
970static void svc_tcp_clear_pages(struct svc_sock *svsk)
971{
972 unsigned int i, len, npages;
973
974 if (svsk->sk_datalen == 0)
975 goto out;
976 len = svsk->sk_datalen;
977 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
978 for (i = 0; i < npages; i++) {
979 if (svsk->sk_pages[i] == NULL) {
980 WARN_ON_ONCE(1);
981 continue;
982 }
983 put_page(svsk->sk_pages[i]);
984 svsk->sk_pages[i] = NULL;
985 }
986out:
987 svsk->sk_tcplen = 0;
988 svsk->sk_datalen = 0;
989}
990
991/*
992 * Receive fragment record header.
993 * If we haven't gotten the record length yet, get the next four bytes.
994 */
995static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
996{
997 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
998 unsigned int want;
999 int len;
1000
1001 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1002 struct kvec iov;
1003
1004 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1005 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1006 iov.iov_len = want;
1007 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1008 goto error;
1009 svsk->sk_tcplen += len;
1010
1011 if (len < want) {
1012 dprintk("svc: short recvfrom while reading record "
1013 "length (%d of %d)\n", len, want);
1014 return -EAGAIN;
1015 }
1016
1017 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
1018 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1019 serv->sv_max_mesg) {
1020 net_notice_ratelimited("RPC: fragment too large: %d\n",
1021 svc_sock_reclen(svsk));
1022 goto err_delete;
1023 }
1024 }
1025
1026 return svc_sock_reclen(svsk);
1027error:
1028 dprintk("RPC: TCP recv_record got %d\n", len);
1029 return len;
1030err_delete:
1031 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1032 return -EAGAIN;
1033}
1034
1035static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1036{
1037 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1038 struct rpc_rqst *req = NULL;
1039 struct kvec *src, *dst;
1040 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1041 __be32 xid;
1042 __be32 calldir;
1043
1044 xid = *p++;
1045 calldir = *p;
1046
1047 if (!bc_xprt)
1048 return -EAGAIN;
1049 spin_lock_bh(&bc_xprt->transport_lock);
1050 req = xprt_lookup_rqst(bc_xprt, xid);
1051 if (!req)
1052 goto unlock_notfound;
1053
1054 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1055 /*
1056 * XXX!: cheating for now! Only copying HEAD.
1057 * But we know this is good enough for now (in fact, for any
1058 * callback reply in the forseeable future).
1059 */
1060 dst = &req->rq_private_buf.head[0];
1061 src = &rqstp->rq_arg.head[0];
1062 if (dst->iov_len < src->iov_len)
1063 goto unlock_eagain; /* whatever; just giving up. */
1064 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1065 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1066 rqstp->rq_arg.len = 0;
1067 spin_unlock_bh(&bc_xprt->transport_lock);
1068 return 0;
1069unlock_notfound:
1070 printk(KERN_NOTICE
1071 "%s: Got unrecognized reply: "
1072 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1073 __func__, ntohl(calldir),
1074 bc_xprt, ntohl(xid));
1075unlock_eagain:
1076 spin_unlock_bh(&bc_xprt->transport_lock);
1077 return -EAGAIN;
1078}
1079
1080static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1081{
1082 int i = 0;
1083 int t = 0;
1084
1085 while (t < len) {
1086 vec[i].iov_base = page_address(pages[i]);
1087 vec[i].iov_len = PAGE_SIZE;
1088 i++;
1089 t += PAGE_SIZE;
1090 }
1091 return i;
1092}
1093
1094static void svc_tcp_fragment_received(struct svc_sock *svsk)
1095{
1096 /* If we have more data, signal svc_xprt_enqueue() to try again */
1097 dprintk("svc: TCP %s record (%d bytes)\n",
1098 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1099 svc_sock_reclen(svsk));
1100 svsk->sk_tcplen = 0;
1101 svsk->sk_reclen = 0;
1102}
1103
1104/*
1105 * Receive data from a TCP socket.
1106 */
1107static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1108{
1109 struct svc_sock *svsk =
1110 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1111 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1112 int len;
1113 struct kvec *vec;
1114 unsigned int want, base;
1115 __be32 *p;
1116 __be32 calldir;
1117 int pnum;
1118
1119 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1120 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1121 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1122 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1123
1124 len = svc_tcp_recv_record(svsk, rqstp);
1125 if (len < 0)
1126 goto error;
1127
1128 base = svc_tcp_restore_pages(svsk, rqstp);
1129 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1130
1131 vec = rqstp->rq_vec;
1132
1133 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1134 svsk->sk_datalen + want);
1135
1136 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1137 rqstp->rq_next_page = rqstp->rq_respages + 1;
1138
1139 /* Now receive data */
1140 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1141 if (len >= 0) {
1142 svsk->sk_tcplen += len;
1143 svsk->sk_datalen += len;
1144 }
1145 if (len != want || !svc_sock_final_rec(svsk)) {
1146 svc_tcp_save_pages(svsk, rqstp);
1147 if (len < 0 && len != -EAGAIN)
1148 goto err_delete;
1149 if (len == want)
1150 svc_tcp_fragment_received(svsk);
1151 else
1152 dprintk("svc: incomplete TCP record (%d of %d)\n",
1153 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1154 svc_sock_reclen(svsk));
1155 goto err_noclose;
1156 }
1157
1158 if (svsk->sk_datalen < 8) {
1159 svsk->sk_datalen = 0;
1160 goto err_delete; /* client is nuts. */
1161 }
1162
1163 rqstp->rq_arg.len = svsk->sk_datalen;
1164 rqstp->rq_arg.page_base = 0;
1165 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1166 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1167 rqstp->rq_arg.page_len = 0;
1168 } else
1169 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1170
1171 rqstp->rq_xprt_ctxt = NULL;
1172 rqstp->rq_prot = IPPROTO_TCP;
1173 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1174 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1175 else
1176 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1177
1178 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1179 calldir = p[1];
1180 if (calldir)
1181 len = receive_cb_reply(svsk, rqstp);
1182
1183 /* Reset TCP read info */
1184 svsk->sk_datalen = 0;
1185 svc_tcp_fragment_received(svsk);
1186
1187 if (len < 0)
1188 goto error;
1189
1190 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1191 if (serv->sv_stats)
1192 serv->sv_stats->nettcpcnt++;
1193
1194 return rqstp->rq_arg.len;
1195
1196error:
1197 if (len != -EAGAIN)
1198 goto err_delete;
1199 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1200 return 0;
1201err_delete:
1202 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1203 svsk->sk_xprt.xpt_server->sv_name, -len);
1204 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1205err_noclose:
1206 return 0; /* record not complete */
1207}
1208
1209/*
1210 * Send out data on TCP socket.
1211 */
1212static int svc_tcp_sendto(struct svc_rqst *rqstp)
1213{
1214 struct xdr_buf *xbufp = &rqstp->rq_res;
1215 int sent;
1216 __be32 reclen;
1217
1218 /* Set up the first element of the reply kvec.
1219 * Any other kvecs that may be in use have been taken
1220 * care of by the server implementation itself.
1221 */
1222 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1223 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1224
1225 sent = svc_sendto(rqstp, &rqstp->rq_res);
1226 if (sent != xbufp->len) {
1227 printk(KERN_NOTICE
1228 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1229 "- shutting down socket\n",
1230 rqstp->rq_xprt->xpt_server->sv_name,
1231 (sent<0)?"got error":"sent only",
1232 sent, xbufp->len);
1233 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1234 svc_xprt_enqueue(rqstp->rq_xprt);
1235 sent = -EAGAIN;
1236 }
1237 return sent;
1238}
1239
1240/*
1241 * Setup response header. TCP has a 4B record length field.
1242 */
1243static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1244{
1245 struct kvec *resv = &rqstp->rq_res.head[0];
1246
1247 /* tcp needs a space for the record length... */
1248 svc_putnl(resv, 0);
1249}
1250
1251static 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)
1260static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1261 struct net *, struct sockaddr *,
1262 int, int);
1263static void svc_bc_sock_free(struct svc_xprt *xprt);
1264
1265static 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
1273static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1274{
1275}
1276
1277static 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 .xpo_secure_port = svc_sock_secure_port,
1283};
1284
1285static struct svc_xprt_class svc_tcp_bc_class = {
1286 .xcl_name = "tcp-bc",
1287 .xcl_owner = THIS_MODULE,
1288 .xcl_ops = &svc_tcp_bc_ops,
1289 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1290};
1291
1292static void svc_init_bc_xprt_sock(void)
1293{
1294 svc_reg_xprt_class(&svc_tcp_bc_class);
1295}
1296
1297static void svc_cleanup_bc_xprt_sock(void)
1298{
1299 svc_unreg_xprt_class(&svc_tcp_bc_class);
1300}
1301#else /* CONFIG_SUNRPC_BACKCHANNEL */
1302static void svc_init_bc_xprt_sock(void)
1303{
1304}
1305
1306static void svc_cleanup_bc_xprt_sock(void)
1307{
1308}
1309#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1310
1311static struct svc_xprt_ops svc_tcp_ops = {
1312 .xpo_create = svc_tcp_create,
1313 .xpo_recvfrom = svc_tcp_recvfrom,
1314 .xpo_sendto = svc_tcp_sendto,
1315 .xpo_release_rqst = svc_release_skb,
1316 .xpo_detach = svc_tcp_sock_detach,
1317 .xpo_free = svc_sock_free,
1318 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1319 .xpo_has_wspace = svc_tcp_has_wspace,
1320 .xpo_accept = svc_tcp_accept,
1321 .xpo_secure_port = svc_sock_secure_port,
1322 .xpo_adjust_wspace = svc_tcp_adjust_wspace,
1323};
1324
1325static struct svc_xprt_class svc_tcp_class = {
1326 .xcl_name = "tcp",
1327 .xcl_owner = THIS_MODULE,
1328 .xcl_ops = &svc_tcp_ops,
1329 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1330 .xcl_ident = XPRT_TRANSPORT_TCP,
1331};
1332
1333void svc_init_xprt_sock(void)
1334{
1335 svc_reg_xprt_class(&svc_tcp_class);
1336 svc_reg_xprt_class(&svc_udp_class);
1337 svc_init_bc_xprt_sock();
1338}
1339
1340void svc_cleanup_xprt_sock(void)
1341{
1342 svc_unreg_xprt_class(&svc_tcp_class);
1343 svc_unreg_xprt_class(&svc_udp_class);
1344 svc_cleanup_bc_xprt_sock();
1345}
1346
1347static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1348{
1349 struct sock *sk = svsk->sk_sk;
1350
1351 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1352 &svsk->sk_xprt, serv);
1353 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1354 if (sk->sk_state == TCP_LISTEN) {
1355 dprintk("setting up TCP socket for listening\n");
1356 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1357 sk->sk_data_ready = svc_tcp_listen_data_ready;
1358 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1359 } else {
1360 dprintk("setting up TCP socket for reading\n");
1361 sk->sk_state_change = svc_tcp_state_change;
1362 sk->sk_data_ready = svc_tcp_data_ready;
1363 sk->sk_write_space = svc_tcp_write_space;
1364
1365 svsk->sk_reclen = 0;
1366 svsk->sk_tcplen = 0;
1367 svsk->sk_datalen = 0;
1368 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1369
1370 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1371
1372 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1373 if (sk->sk_state != TCP_ESTABLISHED)
1374 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1375 }
1376}
1377
1378void svc_sock_update_bufs(struct svc_serv *serv)
1379{
1380 /*
1381 * The number of server threads has changed. Update
1382 * rcvbuf and sndbuf accordingly on all sockets
1383 */
1384 struct svc_sock *svsk;
1385
1386 spin_lock_bh(&serv->sv_lock);
1387 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1388 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1389 spin_unlock_bh(&serv->sv_lock);
1390}
1391EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1392
1393/*
1394 * Initialize socket for RPC use and create svc_sock struct
1395 */
1396static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1397 struct socket *sock,
1398 int flags)
1399{
1400 struct svc_sock *svsk;
1401 struct sock *inet;
1402 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1403 int err = 0;
1404
1405 dprintk("svc: svc_setup_socket %p\n", sock);
1406 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1407 if (!svsk)
1408 return ERR_PTR(-ENOMEM);
1409
1410 inet = sock->sk;
1411
1412 /* Register socket with portmapper */
1413 if (pmap_register)
1414 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1415 inet->sk_protocol,
1416 ntohs(inet_sk(inet)->inet_sport));
1417
1418 if (err < 0) {
1419 kfree(svsk);
1420 return ERR_PTR(err);
1421 }
1422
1423 inet->sk_user_data = svsk;
1424 svsk->sk_sock = sock;
1425 svsk->sk_sk = inet;
1426 svsk->sk_ostate = inet->sk_state_change;
1427 svsk->sk_odata = inet->sk_data_ready;
1428 svsk->sk_owspace = inet->sk_write_space;
1429
1430 /* Initialize the socket */
1431 if (sock->type == SOCK_DGRAM)
1432 svc_udp_init(svsk, serv);
1433 else {
1434 /* initialise setting must have enough space to
1435 * receive and respond to one request.
1436 */
1437 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1438 4 * serv->sv_max_mesg);
1439 svc_tcp_init(svsk, serv);
1440 }
1441
1442 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1443 svsk, svsk->sk_sk);
1444
1445 return svsk;
1446}
1447
1448bool svc_alien_sock(struct net *net, int fd)
1449{
1450 int err;
1451 struct socket *sock = sockfd_lookup(fd, &err);
1452 bool ret = false;
1453
1454 if (!sock)
1455 goto out;
1456 if (sock_net(sock->sk) != net)
1457 ret = true;
1458 sockfd_put(sock);
1459out:
1460 return ret;
1461}
1462EXPORT_SYMBOL_GPL(svc_alien_sock);
1463
1464/**
1465 * svc_addsock - add a listener socket to an RPC service
1466 * @serv: pointer to RPC service to which to add a new listener
1467 * @fd: file descriptor of the new listener
1468 * @name_return: pointer to buffer to fill in with name of listener
1469 * @len: size of the buffer
1470 *
1471 * Fills in socket name and returns positive length of name if successful.
1472 * Name is terminated with '\n'. On error, returns a negative errno
1473 * value.
1474 */
1475int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1476 const size_t len)
1477{
1478 int err = 0;
1479 struct socket *so = sockfd_lookup(fd, &err);
1480 struct svc_sock *svsk = NULL;
1481 struct sockaddr_storage addr;
1482 struct sockaddr *sin = (struct sockaddr *)&addr;
1483 int salen;
1484
1485 if (!so)
1486 return err;
1487 err = -EAFNOSUPPORT;
1488 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1489 goto out;
1490 err = -EPROTONOSUPPORT;
1491 if (so->sk->sk_protocol != IPPROTO_TCP &&
1492 so->sk->sk_protocol != IPPROTO_UDP)
1493 goto out;
1494 err = -EISCONN;
1495 if (so->state > SS_UNCONNECTED)
1496 goto out;
1497 err = -ENOENT;
1498 if (!try_module_get(THIS_MODULE))
1499 goto out;
1500 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1501 if (IS_ERR(svsk)) {
1502 module_put(THIS_MODULE);
1503 err = PTR_ERR(svsk);
1504 goto out;
1505 }
1506 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1507 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1508 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1509 return svc_one_sock_name(svsk, name_return, len);
1510out:
1511 sockfd_put(so);
1512 return err;
1513}
1514EXPORT_SYMBOL_GPL(svc_addsock);
1515
1516/*
1517 * Create socket for RPC service.
1518 */
1519static 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 = SK_CAN_REUSE; /* 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 svsk = svc_setup_socket(serv, sock, flags);
1591 if (IS_ERR(svsk)) {
1592 error = PTR_ERR(svsk);
1593 goto bummer;
1594 }
1595 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1596 return (struct svc_xprt *)svsk;
1597bummer:
1598 dprintk("svc: svc_create_socket error = %d\n", -error);
1599 sock_release(sock);
1600 return ERR_PTR(error);
1601}
1602
1603/*
1604 * Detach the svc_sock from the socket so that no
1605 * more callbacks occur.
1606 */
1607static void svc_sock_detach(struct svc_xprt *xprt)
1608{
1609 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1610 struct sock *sk = svsk->sk_sk;
1611 wait_queue_head_t *wq;
1612
1613 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1614
1615 /* put back the old socket callbacks */
1616 sk->sk_state_change = svsk->sk_ostate;
1617 sk->sk_data_ready = svsk->sk_odata;
1618 sk->sk_write_space = svsk->sk_owspace;
1619
1620 wq = sk_sleep(sk);
1621 if (sunrpc_waitqueue_active(wq))
1622 wake_up_interruptible(wq);
1623}
1624
1625/*
1626 * Disconnect the socket, and reset the callbacks
1627 */
1628static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1629{
1630 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1631
1632 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1633
1634 svc_sock_detach(xprt);
1635
1636 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1637 svc_tcp_clear_pages(svsk);
1638 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1639 }
1640}
1641
1642/*
1643 * Free the svc_sock's socket resources and the svc_sock itself.
1644 */
1645static void svc_sock_free(struct svc_xprt *xprt)
1646{
1647 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1648 dprintk("svc: svc_sock_free(%p)\n", svsk);
1649
1650 if (svsk->sk_sock->file)
1651 sockfd_put(svsk->sk_sock);
1652 else
1653 sock_release(svsk->sk_sock);
1654 kfree(svsk);
1655}
1656
1657#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1658/*
1659 * Create a back channel svc_xprt which shares the fore channel socket.
1660 */
1661static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1662 int protocol,
1663 struct net *net,
1664 struct sockaddr *sin, int len,
1665 int flags)
1666{
1667 struct svc_sock *svsk;
1668 struct svc_xprt *xprt;
1669
1670 if (protocol != IPPROTO_TCP) {
1671 printk(KERN_WARNING "svc: only TCP sockets"
1672 " supported on shared back channel\n");
1673 return ERR_PTR(-EINVAL);
1674 }
1675
1676 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1677 if (!svsk)
1678 return ERR_PTR(-ENOMEM);
1679
1680 xprt = &svsk->sk_xprt;
1681 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1682
1683 serv->sv_bc_xprt = xprt;
1684
1685 return xprt;
1686}
1687
1688/*
1689 * Free a back channel svc_sock.
1690 */
1691static void svc_bc_sock_free(struct svc_xprt *xprt)
1692{
1693 if (xprt)
1694 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1695}
1696#endif /* CONFIG_SUNRPC_BACKCHANNEL */
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/errno.h>
25#include <linux/fcntl.h>
26#include <linux/net.h>
27#include <linux/in.h>
28#include <linux/inet.h>
29#include <linux/udp.h>
30#include <linux/tcp.h>
31#include <linux/unistd.h>
32#include <linux/slab.h>
33#include <linux/netdevice.h>
34#include <linux/skbuff.h>
35#include <linux/file.h>
36#include <linux/freezer.h>
37#include <net/sock.h>
38#include <net/checksum.h>
39#include <net/ip.h>
40#include <net/ipv6.h>
41#include <net/tcp.h>
42#include <net/tcp_states.h>
43#include <asm/uaccess.h>
44#include <asm/ioctls.h>
45
46#include <linux/sunrpc/types.h>
47#include <linux/sunrpc/clnt.h>
48#include <linux/sunrpc/xdr.h>
49#include <linux/sunrpc/msg_prot.h>
50#include <linux/sunrpc/svcsock.h>
51#include <linux/sunrpc/stats.h>
52#include <linux/sunrpc/xprt.h>
53
54#include "sunrpc.h"
55
56#define RPCDBG_FACILITY RPCDBG_SVCXPRT
57
58
59static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
60 int *errp, int flags);
61static void svc_udp_data_ready(struct sock *, int);
62static int svc_udp_recvfrom(struct svc_rqst *);
63static int svc_udp_sendto(struct svc_rqst *);
64static void svc_sock_detach(struct svc_xprt *);
65static void svc_tcp_sock_detach(struct svc_xprt *);
66static void svc_sock_free(struct svc_xprt *);
67
68static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
69 struct net *, struct sockaddr *,
70 int, int);
71#if defined(CONFIG_SUNRPC_BACKCHANNEL)
72static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
73 struct net *, struct sockaddr *,
74 int, int);
75static void svc_bc_sock_free(struct svc_xprt *xprt);
76#endif /* CONFIG_SUNRPC_BACKCHANNEL */
77
78#ifdef CONFIG_DEBUG_LOCK_ALLOC
79static struct lock_class_key svc_key[2];
80static struct lock_class_key svc_slock_key[2];
81
82static void svc_reclassify_socket(struct socket *sock)
83{
84 struct sock *sk = sock->sk;
85 BUG_ON(sock_owned_by_user(sk));
86 switch (sk->sk_family) {
87 case AF_INET:
88 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
89 &svc_slock_key[0],
90 "sk_xprt.xpt_lock-AF_INET-NFSD",
91 &svc_key[0]);
92 break;
93
94 case AF_INET6:
95 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
96 &svc_slock_key[1],
97 "sk_xprt.xpt_lock-AF_INET6-NFSD",
98 &svc_key[1]);
99 break;
100
101 default:
102 BUG();
103 }
104}
105#else
106static void svc_reclassify_socket(struct socket *sock)
107{
108}
109#endif
110
111/*
112 * Release an skbuff after use
113 */
114static void svc_release_skb(struct svc_rqst *rqstp)
115{
116 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
117
118 if (skb) {
119 struct svc_sock *svsk =
120 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
121 rqstp->rq_xprt_ctxt = NULL;
122
123 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
124 skb_free_datagram_locked(svsk->sk_sk, skb);
125 }
126}
127
128union svc_pktinfo_u {
129 struct in_pktinfo pkti;
130 struct in6_pktinfo pkti6;
131};
132#define SVC_PKTINFO_SPACE \
133 CMSG_SPACE(sizeof(union svc_pktinfo_u))
134
135static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
136{
137 struct svc_sock *svsk =
138 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
139 switch (svsk->sk_sk->sk_family) {
140 case AF_INET: {
141 struct in_pktinfo *pki = CMSG_DATA(cmh);
142
143 cmh->cmsg_level = SOL_IP;
144 cmh->cmsg_type = IP_PKTINFO;
145 pki->ipi_ifindex = 0;
146 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
147 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
148 }
149 break;
150
151 case AF_INET6: {
152 struct in6_pktinfo *pki = CMSG_DATA(cmh);
153
154 cmh->cmsg_level = SOL_IPV6;
155 cmh->cmsg_type = IPV6_PKTINFO;
156 pki->ipi6_ifindex = 0;
157 ipv6_addr_copy(&pki->ipi6_addr,
158 &rqstp->rq_daddr.addr6);
159 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
160 }
161 break;
162 }
163}
164
165/*
166 * send routine intended to be shared by the fore- and back-channel
167 */
168int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
169 struct page *headpage, unsigned long headoffset,
170 struct page *tailpage, unsigned long tailoffset)
171{
172 int result;
173 int size;
174 struct page **ppage = xdr->pages;
175 size_t base = xdr->page_base;
176 unsigned int pglen = xdr->page_len;
177 unsigned int flags = MSG_MORE;
178 int slen;
179 int len = 0;
180
181 slen = xdr->len;
182
183 /* send head */
184 if (slen == xdr->head[0].iov_len)
185 flags = 0;
186 len = kernel_sendpage(sock, headpage, headoffset,
187 xdr->head[0].iov_len, flags);
188 if (len != xdr->head[0].iov_len)
189 goto out;
190 slen -= xdr->head[0].iov_len;
191 if (slen == 0)
192 goto out;
193
194 /* send page data */
195 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
196 while (pglen > 0) {
197 if (slen == size)
198 flags = 0;
199 result = kernel_sendpage(sock, *ppage, base, size, flags);
200 if (result > 0)
201 len += result;
202 if (result != size)
203 goto out;
204 slen -= size;
205 pglen -= size;
206 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
207 base = 0;
208 ppage++;
209 }
210
211 /* send tail */
212 if (xdr->tail[0].iov_len) {
213 result = kernel_sendpage(sock, tailpage, tailoffset,
214 xdr->tail[0].iov_len, 0);
215 if (result > 0)
216 len += result;
217 }
218
219out:
220 return len;
221}
222
223
224/*
225 * Generic sendto routine
226 */
227static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
228{
229 struct svc_sock *svsk =
230 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
231 struct socket *sock = svsk->sk_sock;
232 union {
233 struct cmsghdr hdr;
234 long all[SVC_PKTINFO_SPACE / sizeof(long)];
235 } buffer;
236 struct cmsghdr *cmh = &buffer.hdr;
237 int len = 0;
238 unsigned long tailoff;
239 unsigned long headoff;
240 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
241
242 if (rqstp->rq_prot == IPPROTO_UDP) {
243 struct msghdr msg = {
244 .msg_name = &rqstp->rq_addr,
245 .msg_namelen = rqstp->rq_addrlen,
246 .msg_control = cmh,
247 .msg_controllen = sizeof(buffer),
248 .msg_flags = MSG_MORE,
249 };
250
251 svc_set_cmsg_data(rqstp, cmh);
252
253 if (sock_sendmsg(sock, &msg, 0) < 0)
254 goto out;
255 }
256
257 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
258 headoff = 0;
259 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
260 rqstp->rq_respages[0], tailoff);
261
262out:
263 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
264 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
265 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
266
267 return len;
268}
269
270/*
271 * Report socket names for nfsdfs
272 */
273static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
274{
275 const struct sock *sk = svsk->sk_sk;
276 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
277 "udp" : "tcp";
278 int len;
279
280 switch (sk->sk_family) {
281 case PF_INET:
282 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
283 proto_name,
284 &inet_sk(sk)->inet_rcv_saddr,
285 inet_sk(sk)->inet_num);
286 break;
287 case PF_INET6:
288 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
289 proto_name,
290 &inet6_sk(sk)->rcv_saddr,
291 inet_sk(sk)->inet_num);
292 break;
293 default:
294 len = snprintf(buf, remaining, "*unknown-%d*\n",
295 sk->sk_family);
296 }
297
298 if (len >= remaining) {
299 *buf = '\0';
300 return -ENAMETOOLONG;
301 }
302 return len;
303}
304
305/**
306 * svc_sock_names - construct a list of listener names in a string
307 * @serv: pointer to RPC service
308 * @buf: pointer to a buffer to fill in with socket names
309 * @buflen: size of the buffer to be filled
310 * @toclose: pointer to '\0'-terminated C string containing the name
311 * of a listener to be closed
312 *
313 * Fills in @buf with a '\n'-separated list of names of listener
314 * sockets. If @toclose is not NULL, the socket named by @toclose
315 * is closed, and is not included in the output list.
316 *
317 * Returns positive length of the socket name string, or a negative
318 * errno value on error.
319 */
320int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
321 const char *toclose)
322{
323 struct svc_sock *svsk, *closesk = NULL;
324 int len = 0;
325
326 if (!serv)
327 return 0;
328
329 spin_lock_bh(&serv->sv_lock);
330 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
331 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
332 if (onelen < 0) {
333 len = onelen;
334 break;
335 }
336 if (toclose && strcmp(toclose, buf + len) == 0) {
337 closesk = svsk;
338 svc_xprt_get(&closesk->sk_xprt);
339 } else
340 len += onelen;
341 }
342 spin_unlock_bh(&serv->sv_lock);
343
344 if (closesk) {
345 /* Should unregister with portmap, but you cannot
346 * unregister just one protocol...
347 */
348 svc_close_xprt(&closesk->sk_xprt);
349 svc_xprt_put(&closesk->sk_xprt);
350 } else if (toclose)
351 return -ENOENT;
352 return len;
353}
354EXPORT_SYMBOL_GPL(svc_sock_names);
355
356/*
357 * Check input queue length
358 */
359static int svc_recv_available(struct svc_sock *svsk)
360{
361 struct socket *sock = svsk->sk_sock;
362 int avail, err;
363
364 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
365
366 return (err >= 0)? avail : err;
367}
368
369/*
370 * Generic recvfrom routine.
371 */
372static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
373 int buflen)
374{
375 struct svc_sock *svsk =
376 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
377 struct msghdr msg = {
378 .msg_flags = MSG_DONTWAIT,
379 };
380 int len;
381
382 rqstp->rq_xprt_hlen = 0;
383
384 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
385 msg.msg_flags);
386
387 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
388 svsk, iov[0].iov_base, iov[0].iov_len, len);
389 return len;
390}
391
392static int svc_partial_recvfrom(struct svc_rqst *rqstp,
393 struct kvec *iov, int nr,
394 int buflen, unsigned int base)
395{
396 size_t save_iovlen;
397 void __user *save_iovbase;
398 unsigned int i;
399 int ret;
400
401 if (base == 0)
402 return svc_recvfrom(rqstp, iov, nr, buflen);
403
404 for (i = 0; i < nr; i++) {
405 if (iov[i].iov_len > base)
406 break;
407 base -= iov[i].iov_len;
408 }
409 save_iovlen = iov[i].iov_len;
410 save_iovbase = iov[i].iov_base;
411 iov[i].iov_len -= base;
412 iov[i].iov_base += base;
413 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
414 iov[i].iov_len = save_iovlen;
415 iov[i].iov_base = save_iovbase;
416 return ret;
417}
418
419/*
420 * Set socket snd and rcv buffer lengths
421 */
422static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
423 unsigned int rcv)
424{
425#if 0
426 mm_segment_t oldfs;
427 oldfs = get_fs(); set_fs(KERNEL_DS);
428 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
429 (char*)&snd, sizeof(snd));
430 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
431 (char*)&rcv, sizeof(rcv));
432#else
433 /* sock_setsockopt limits use to sysctl_?mem_max,
434 * which isn't acceptable. Until that is made conditional
435 * on not having CAP_SYS_RESOURCE or similar, we go direct...
436 * DaveM said I could!
437 */
438 lock_sock(sock->sk);
439 sock->sk->sk_sndbuf = snd * 2;
440 sock->sk->sk_rcvbuf = rcv * 2;
441 sock->sk->sk_write_space(sock->sk);
442 release_sock(sock->sk);
443#endif
444}
445/*
446 * INET callback when data has been received on the socket.
447 */
448static void svc_udp_data_ready(struct sock *sk, int count)
449{
450 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
451 wait_queue_head_t *wq = sk_sleep(sk);
452
453 if (svsk) {
454 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
455 svsk, sk, count,
456 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
457 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
458 svc_xprt_enqueue(&svsk->sk_xprt);
459 }
460 if (wq && waitqueue_active(wq))
461 wake_up_interruptible(wq);
462}
463
464/*
465 * INET callback when space is newly available on the socket.
466 */
467static void svc_write_space(struct sock *sk)
468{
469 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
470 wait_queue_head_t *wq = sk_sleep(sk);
471
472 if (svsk) {
473 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
474 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
475 svc_xprt_enqueue(&svsk->sk_xprt);
476 }
477
478 if (wq && waitqueue_active(wq)) {
479 dprintk("RPC svc_write_space: someone sleeping on %p\n",
480 svsk);
481 wake_up_interruptible(wq);
482 }
483}
484
485static void svc_tcp_write_space(struct sock *sk)
486{
487 struct socket *sock = sk->sk_socket;
488
489 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
490 clear_bit(SOCK_NOSPACE, &sock->flags);
491 svc_write_space(sk);
492}
493
494/*
495 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
496 */
497static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
498 struct cmsghdr *cmh)
499{
500 struct in_pktinfo *pki = CMSG_DATA(cmh);
501 if (cmh->cmsg_type != IP_PKTINFO)
502 return 0;
503 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
504 return 1;
505}
506
507/*
508 * See net/ipv6/datagram.c : datagram_recv_ctl
509 */
510static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
511 struct cmsghdr *cmh)
512{
513 struct in6_pktinfo *pki = CMSG_DATA(cmh);
514 if (cmh->cmsg_type != IPV6_PKTINFO)
515 return 0;
516 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
517 return 1;
518}
519
520/*
521 * Copy the UDP datagram's destination address to the rqstp structure.
522 * The 'destination' address in this case is the address to which the
523 * peer sent the datagram, i.e. our local address. For multihomed
524 * hosts, this can change from msg to msg. Note that only the IP
525 * address changes, the port number should remain the same.
526 */
527static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
528 struct cmsghdr *cmh)
529{
530 switch (cmh->cmsg_level) {
531 case SOL_IP:
532 return svc_udp_get_dest_address4(rqstp, cmh);
533 case SOL_IPV6:
534 return svc_udp_get_dest_address6(rqstp, cmh);
535 }
536
537 return 0;
538}
539
540/*
541 * Receive a datagram from a UDP socket.
542 */
543static int svc_udp_recvfrom(struct svc_rqst *rqstp)
544{
545 struct svc_sock *svsk =
546 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
547 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
548 struct sk_buff *skb;
549 union {
550 struct cmsghdr hdr;
551 long all[SVC_PKTINFO_SPACE / sizeof(long)];
552 } buffer;
553 struct cmsghdr *cmh = &buffer.hdr;
554 struct msghdr msg = {
555 .msg_name = svc_addr(rqstp),
556 .msg_control = cmh,
557 .msg_controllen = sizeof(buffer),
558 .msg_flags = MSG_DONTWAIT,
559 };
560 size_t len;
561 int err;
562
563 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
564 /* udp sockets need large rcvbuf as all pending
565 * requests are still in that buffer. sndbuf must
566 * also be large enough that there is enough space
567 * for one reply per thread. We count all threads
568 * rather than threads in a particular pool, which
569 * provides an upper bound on the number of threads
570 * which will access the socket.
571 */
572 svc_sock_setbufsize(svsk->sk_sock,
573 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
574 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
575
576 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
577 skb = NULL;
578 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
579 0, 0, MSG_PEEK | MSG_DONTWAIT);
580 if (err >= 0)
581 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
582
583 if (skb == NULL) {
584 if (err != -EAGAIN) {
585 /* possibly an icmp error */
586 dprintk("svc: recvfrom returned error %d\n", -err);
587 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
588 }
589 return -EAGAIN;
590 }
591 len = svc_addr_len(svc_addr(rqstp));
592 if (len == 0)
593 return -EAFNOSUPPORT;
594 rqstp->rq_addrlen = len;
595 if (skb->tstamp.tv64 == 0) {
596 skb->tstamp = ktime_get_real();
597 /* Don't enable netstamp, sunrpc doesn't
598 need that much accuracy */
599 }
600 svsk->sk_sk->sk_stamp = skb->tstamp;
601 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
602
603 len = skb->len - sizeof(struct udphdr);
604 rqstp->rq_arg.len = len;
605
606 rqstp->rq_prot = IPPROTO_UDP;
607
608 if (!svc_udp_get_dest_address(rqstp, cmh)) {
609 if (net_ratelimit())
610 printk(KERN_WARNING
611 "svc: received unknown control message %d/%d; "
612 "dropping RPC reply datagram\n",
613 cmh->cmsg_level, cmh->cmsg_type);
614 skb_free_datagram_locked(svsk->sk_sk, skb);
615 return 0;
616 }
617
618 if (skb_is_nonlinear(skb)) {
619 /* we have to copy */
620 local_bh_disable();
621 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
622 local_bh_enable();
623 /* checksum error */
624 skb_free_datagram_locked(svsk->sk_sk, skb);
625 return 0;
626 }
627 local_bh_enable();
628 skb_free_datagram_locked(svsk->sk_sk, skb);
629 } else {
630 /* we can use it in-place */
631 rqstp->rq_arg.head[0].iov_base = skb->data +
632 sizeof(struct udphdr);
633 rqstp->rq_arg.head[0].iov_len = len;
634 if (skb_checksum_complete(skb)) {
635 skb_free_datagram_locked(svsk->sk_sk, skb);
636 return 0;
637 }
638 rqstp->rq_xprt_ctxt = skb;
639 }
640
641 rqstp->rq_arg.page_base = 0;
642 if (len <= rqstp->rq_arg.head[0].iov_len) {
643 rqstp->rq_arg.head[0].iov_len = len;
644 rqstp->rq_arg.page_len = 0;
645 rqstp->rq_respages = rqstp->rq_pages+1;
646 } else {
647 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
648 rqstp->rq_respages = rqstp->rq_pages + 1 +
649 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
650 }
651
652 if (serv->sv_stats)
653 serv->sv_stats->netudpcnt++;
654
655 return len;
656}
657
658static int
659svc_udp_sendto(struct svc_rqst *rqstp)
660{
661 int error;
662
663 error = svc_sendto(rqstp, &rqstp->rq_res);
664 if (error == -ECONNREFUSED)
665 /* ICMP error on earlier request. */
666 error = svc_sendto(rqstp, &rqstp->rq_res);
667
668 return error;
669}
670
671static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
672{
673}
674
675static int svc_udp_has_wspace(struct svc_xprt *xprt)
676{
677 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
678 struct svc_serv *serv = xprt->xpt_server;
679 unsigned long required;
680
681 /*
682 * Set the SOCK_NOSPACE flag before checking the available
683 * sock space.
684 */
685 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
686 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
687 if (required*2 > sock_wspace(svsk->sk_sk))
688 return 0;
689 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
690 return 1;
691}
692
693static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
694{
695 BUG();
696 return NULL;
697}
698
699static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
700 struct net *net,
701 struct sockaddr *sa, int salen,
702 int flags)
703{
704 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
705}
706
707static struct svc_xprt_ops svc_udp_ops = {
708 .xpo_create = svc_udp_create,
709 .xpo_recvfrom = svc_udp_recvfrom,
710 .xpo_sendto = svc_udp_sendto,
711 .xpo_release_rqst = svc_release_skb,
712 .xpo_detach = svc_sock_detach,
713 .xpo_free = svc_sock_free,
714 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
715 .xpo_has_wspace = svc_udp_has_wspace,
716 .xpo_accept = svc_udp_accept,
717};
718
719static struct svc_xprt_class svc_udp_class = {
720 .xcl_name = "udp",
721 .xcl_owner = THIS_MODULE,
722 .xcl_ops = &svc_udp_ops,
723 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
724};
725
726static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
727{
728 int err, level, optname, one = 1;
729
730 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
731 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
732 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
733 svsk->sk_sk->sk_write_space = svc_write_space;
734
735 /* initialise setting must have enough space to
736 * receive and respond to one request.
737 * svc_udp_recvfrom will re-adjust if necessary
738 */
739 svc_sock_setbufsize(svsk->sk_sock,
740 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
741 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
742
743 /* data might have come in before data_ready set up */
744 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
745 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
746
747 /* make sure we get destination address info */
748 switch (svsk->sk_sk->sk_family) {
749 case AF_INET:
750 level = SOL_IP;
751 optname = IP_PKTINFO;
752 break;
753 case AF_INET6:
754 level = SOL_IPV6;
755 optname = IPV6_RECVPKTINFO;
756 break;
757 default:
758 BUG();
759 }
760 err = kernel_setsockopt(svsk->sk_sock, level, optname,
761 (char *)&one, sizeof(one));
762 dprintk("svc: kernel_setsockopt returned %d\n", err);
763}
764
765/*
766 * A data_ready event on a listening socket means there's a connection
767 * pending. Do not use state_change as a substitute for it.
768 */
769static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
770{
771 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
772 wait_queue_head_t *wq;
773
774 dprintk("svc: socket %p TCP (listen) state change %d\n",
775 sk, sk->sk_state);
776
777 /*
778 * This callback may called twice when a new connection
779 * is established as a child socket inherits everything
780 * from a parent LISTEN socket.
781 * 1) data_ready method of the parent socket will be called
782 * when one of child sockets become ESTABLISHED.
783 * 2) data_ready method of the child socket may be called
784 * when it receives data before the socket is accepted.
785 * In case of 2, we should ignore it silently.
786 */
787 if (sk->sk_state == TCP_LISTEN) {
788 if (svsk) {
789 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
790 svc_xprt_enqueue(&svsk->sk_xprt);
791 } else
792 printk("svc: socket %p: no user data\n", sk);
793 }
794
795 wq = sk_sleep(sk);
796 if (wq && waitqueue_active(wq))
797 wake_up_interruptible_all(wq);
798}
799
800/*
801 * A state change on a connected socket means it's dying or dead.
802 */
803static void svc_tcp_state_change(struct sock *sk)
804{
805 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
806 wait_queue_head_t *wq = sk_sleep(sk);
807
808 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
809 sk, sk->sk_state, sk->sk_user_data);
810
811 if (!svsk)
812 printk("svc: socket %p: no user data\n", sk);
813 else {
814 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
815 svc_xprt_enqueue(&svsk->sk_xprt);
816 }
817 if (wq && waitqueue_active(wq))
818 wake_up_interruptible_all(wq);
819}
820
821static void svc_tcp_data_ready(struct sock *sk, int count)
822{
823 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
824 wait_queue_head_t *wq = sk_sleep(sk);
825
826 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
827 sk, sk->sk_user_data);
828 if (svsk) {
829 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
830 svc_xprt_enqueue(&svsk->sk_xprt);
831 }
832 if (wq && waitqueue_active(wq))
833 wake_up_interruptible(wq);
834}
835
836/*
837 * Accept a TCP connection
838 */
839static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
840{
841 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
842 struct sockaddr_storage addr;
843 struct sockaddr *sin = (struct sockaddr *) &addr;
844 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
845 struct socket *sock = svsk->sk_sock;
846 struct socket *newsock;
847 struct svc_sock *newsvsk;
848 int err, slen;
849 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
850
851 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
852 if (!sock)
853 return NULL;
854
855 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
856 err = kernel_accept(sock, &newsock, O_NONBLOCK);
857 if (err < 0) {
858 if (err == -ENOMEM)
859 printk(KERN_WARNING "%s: no more sockets!\n",
860 serv->sv_name);
861 else if (err != -EAGAIN && net_ratelimit())
862 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
863 serv->sv_name, -err);
864 return NULL;
865 }
866 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
867
868 err = kernel_getpeername(newsock, sin, &slen);
869 if (err < 0) {
870 if (net_ratelimit())
871 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
872 serv->sv_name, -err);
873 goto failed; /* aborted connection or whatever */
874 }
875
876 /* Ideally, we would want to reject connections from unauthorized
877 * hosts here, but when we get encryption, the IP of the host won't
878 * tell us anything. For now just warn about unpriv connections.
879 */
880 if (!svc_port_is_privileged(sin)) {
881 dprintk(KERN_WARNING
882 "%s: connect from unprivileged port: %s\n",
883 serv->sv_name,
884 __svc_print_addr(sin, buf, sizeof(buf)));
885 }
886 dprintk("%s: connect from %s\n", serv->sv_name,
887 __svc_print_addr(sin, buf, sizeof(buf)));
888
889 /* make sure that a write doesn't block forever when
890 * low on memory
891 */
892 newsock->sk->sk_sndtimeo = HZ*30;
893
894 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
895 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
896 goto failed;
897 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
898 err = kernel_getsockname(newsock, sin, &slen);
899 if (unlikely(err < 0)) {
900 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
901 slen = offsetof(struct sockaddr, sa_data);
902 }
903 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
904
905 if (serv->sv_stats)
906 serv->sv_stats->nettcpconn++;
907
908 return &newsvsk->sk_xprt;
909
910failed:
911 sock_release(newsock);
912 return NULL;
913}
914
915static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
916{
917 unsigned int i, len, npages;
918
919 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
920 return 0;
921 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
922 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
923 for (i = 0; i < npages; i++) {
924 if (rqstp->rq_pages[i] != NULL)
925 put_page(rqstp->rq_pages[i]);
926 BUG_ON(svsk->sk_pages[i] == NULL);
927 rqstp->rq_pages[i] = svsk->sk_pages[i];
928 svsk->sk_pages[i] = NULL;
929 }
930 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
931 return len;
932}
933
934static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
935{
936 unsigned int i, len, npages;
937
938 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
939 return;
940 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
941 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
942 for (i = 0; i < npages; i++) {
943 svsk->sk_pages[i] = rqstp->rq_pages[i];
944 rqstp->rq_pages[i] = NULL;
945 }
946}
947
948static void svc_tcp_clear_pages(struct svc_sock *svsk)
949{
950 unsigned int i, len, npages;
951
952 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
953 goto out;
954 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
955 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
956 for (i = 0; i < npages; i++) {
957 BUG_ON(svsk->sk_pages[i] == NULL);
958 put_page(svsk->sk_pages[i]);
959 svsk->sk_pages[i] = NULL;
960 }
961out:
962 svsk->sk_tcplen = 0;
963}
964
965/*
966 * Receive data.
967 * If we haven't gotten the record length yet, get the next four bytes.
968 * Otherwise try to gobble up as much as possible up to the complete
969 * record length.
970 */
971static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
972{
973 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
974 unsigned int want;
975 int len;
976
977 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
978
979 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
980 struct kvec iov;
981
982 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
983 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
984 iov.iov_len = want;
985 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
986 goto error;
987 svsk->sk_tcplen += len;
988
989 if (len < want) {
990 dprintk("svc: short recvfrom while reading record "
991 "length (%d of %d)\n", len, want);
992 return -EAGAIN;
993 }
994
995 svsk->sk_reclen = ntohl(svsk->sk_reclen);
996 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
997 /* FIXME: technically, a record can be fragmented,
998 * and non-terminal fragments will not have the top
999 * bit set in the fragment length header.
1000 * But apparently no known nfs clients send fragmented
1001 * records. */
1002 if (net_ratelimit())
1003 printk(KERN_NOTICE "RPC: multiple fragments "
1004 "per record not supported\n");
1005 goto err_delete;
1006 }
1007
1008 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1009 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1010 if (svsk->sk_reclen > serv->sv_max_mesg) {
1011 if (net_ratelimit())
1012 printk(KERN_NOTICE "RPC: "
1013 "fragment too large: 0x%08lx\n",
1014 (unsigned long)svsk->sk_reclen);
1015 goto err_delete;
1016 }
1017 }
1018
1019 if (svsk->sk_reclen < 8)
1020 goto err_delete; /* client is nuts. */
1021
1022 len = svsk->sk_reclen;
1023
1024 return len;
1025error:
1026 dprintk("RPC: TCP recv_record got %d\n", len);
1027 return len;
1028err_delete:
1029 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1030 return -EAGAIN;
1031}
1032
1033static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1034{
1035 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1036 struct rpc_rqst *req = NULL;
1037 struct kvec *src, *dst;
1038 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1039 __be32 xid;
1040 __be32 calldir;
1041
1042 xid = *p++;
1043 calldir = *p;
1044
1045 if (bc_xprt)
1046 req = xprt_lookup_rqst(bc_xprt, xid);
1047
1048 if (!req) {
1049 printk(KERN_NOTICE
1050 "%s: Got unrecognized reply: "
1051 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1052 __func__, ntohl(calldir),
1053 bc_xprt, xid);
1054 return -EAGAIN;
1055 }
1056
1057 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1058 /*
1059 * XXX!: cheating for now! Only copying HEAD.
1060 * But we know this is good enough for now (in fact, for any
1061 * callback reply in the forseeable future).
1062 */
1063 dst = &req->rq_private_buf.head[0];
1064 src = &rqstp->rq_arg.head[0];
1065 if (dst->iov_len < src->iov_len)
1066 return -EAGAIN; /* whatever; just giving up. */
1067 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1068 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1069 rqstp->rq_arg.len = 0;
1070 return 0;
1071}
1072
1073static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1074{
1075 int i = 0;
1076 int t = 0;
1077
1078 while (t < len) {
1079 vec[i].iov_base = page_address(pages[i]);
1080 vec[i].iov_len = PAGE_SIZE;
1081 i++;
1082 t += PAGE_SIZE;
1083 }
1084 return i;
1085}
1086
1087
1088/*
1089 * Receive data from a TCP socket.
1090 */
1091static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1092{
1093 struct svc_sock *svsk =
1094 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1095 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1096 int len;
1097 struct kvec *vec;
1098 unsigned int want, base;
1099 __be32 *p;
1100 __be32 calldir;
1101 int pnum;
1102
1103 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1104 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1105 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1106 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1107
1108 len = svc_tcp_recv_record(svsk, rqstp);
1109 if (len < 0)
1110 goto error;
1111
1112 base = svc_tcp_restore_pages(svsk, rqstp);
1113 want = svsk->sk_reclen - base;
1114
1115 vec = rqstp->rq_vec;
1116
1117 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1118 svsk->sk_reclen);
1119
1120 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1121
1122 /* Now receive data */
1123 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1124 if (len >= 0)
1125 svsk->sk_tcplen += len;
1126 if (len != want) {
1127 if (len < 0 && len != -EAGAIN)
1128 goto err_other;
1129 svc_tcp_save_pages(svsk, rqstp);
1130 dprintk("svc: incomplete TCP record (%d of %d)\n",
1131 svsk->sk_tcplen, svsk->sk_reclen);
1132 goto err_noclose;
1133 }
1134
1135 rqstp->rq_arg.len = svsk->sk_reclen;
1136 rqstp->rq_arg.page_base = 0;
1137 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1138 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1139 rqstp->rq_arg.page_len = 0;
1140 } else
1141 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1142
1143 rqstp->rq_xprt_ctxt = NULL;
1144 rqstp->rq_prot = IPPROTO_TCP;
1145
1146 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1147 calldir = p[1];
1148 if (calldir)
1149 len = receive_cb_reply(svsk, rqstp);
1150
1151 /* Reset TCP read info */
1152 svsk->sk_reclen = 0;
1153 svsk->sk_tcplen = 0;
1154 /* If we have more data, signal svc_xprt_enqueue() to try again */
1155 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1156 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1157
1158 if (len < 0)
1159 goto error;
1160
1161 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1162 if (serv->sv_stats)
1163 serv->sv_stats->nettcpcnt++;
1164
1165 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1166 return rqstp->rq_arg.len;
1167
1168error:
1169 if (len != -EAGAIN)
1170 goto err_other;
1171 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1172 return -EAGAIN;
1173err_other:
1174 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1175 svsk->sk_xprt.xpt_server->sv_name, -len);
1176 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1177err_noclose:
1178 return -EAGAIN; /* record not complete */
1179}
1180
1181/*
1182 * Send out data on TCP socket.
1183 */
1184static int svc_tcp_sendto(struct svc_rqst *rqstp)
1185{
1186 struct xdr_buf *xbufp = &rqstp->rq_res;
1187 int sent;
1188 __be32 reclen;
1189
1190 /* Set up the first element of the reply kvec.
1191 * Any other kvecs that may be in use have been taken
1192 * care of by the server implementation itself.
1193 */
1194 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1195 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1196
1197 sent = svc_sendto(rqstp, &rqstp->rq_res);
1198 if (sent != xbufp->len) {
1199 printk(KERN_NOTICE
1200 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1201 "- shutting down socket\n",
1202 rqstp->rq_xprt->xpt_server->sv_name,
1203 (sent<0)?"got error":"sent only",
1204 sent, xbufp->len);
1205 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1206 svc_xprt_enqueue(rqstp->rq_xprt);
1207 sent = -EAGAIN;
1208 }
1209 return sent;
1210}
1211
1212/*
1213 * Setup response header. TCP has a 4B record length field.
1214 */
1215static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1216{
1217 struct kvec *resv = &rqstp->rq_res.head[0];
1218
1219 /* tcp needs a space for the record length... */
1220 svc_putnl(resv, 0);
1221}
1222
1223static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1224{
1225 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1226 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1227 int required;
1228
1229 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1230 return 1;
1231 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1232 if (sk_stream_wspace(svsk->sk_sk) >= required)
1233 return 1;
1234 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1235 return 0;
1236}
1237
1238static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1239 struct net *net,
1240 struct sockaddr *sa, int salen,
1241 int flags)
1242{
1243 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1244}
1245
1246#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1247static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1248 struct net *, struct sockaddr *,
1249 int, int);
1250static void svc_bc_sock_free(struct svc_xprt *xprt);
1251
1252static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1253 struct net *net,
1254 struct sockaddr *sa, int salen,
1255 int flags)
1256{
1257 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1258}
1259
1260static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1261{
1262}
1263
1264static struct svc_xprt_ops svc_tcp_bc_ops = {
1265 .xpo_create = svc_bc_tcp_create,
1266 .xpo_detach = svc_bc_tcp_sock_detach,
1267 .xpo_free = svc_bc_sock_free,
1268 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1269};
1270
1271static struct svc_xprt_class svc_tcp_bc_class = {
1272 .xcl_name = "tcp-bc",
1273 .xcl_owner = THIS_MODULE,
1274 .xcl_ops = &svc_tcp_bc_ops,
1275 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1276};
1277
1278static void svc_init_bc_xprt_sock(void)
1279{
1280 svc_reg_xprt_class(&svc_tcp_bc_class);
1281}
1282
1283static void svc_cleanup_bc_xprt_sock(void)
1284{
1285 svc_unreg_xprt_class(&svc_tcp_bc_class);
1286}
1287#else /* CONFIG_SUNRPC_BACKCHANNEL */
1288static void svc_init_bc_xprt_sock(void)
1289{
1290}
1291
1292static void svc_cleanup_bc_xprt_sock(void)
1293{
1294}
1295#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1296
1297static struct svc_xprt_ops svc_tcp_ops = {
1298 .xpo_create = svc_tcp_create,
1299 .xpo_recvfrom = svc_tcp_recvfrom,
1300 .xpo_sendto = svc_tcp_sendto,
1301 .xpo_release_rqst = svc_release_skb,
1302 .xpo_detach = svc_tcp_sock_detach,
1303 .xpo_free = svc_sock_free,
1304 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1305 .xpo_has_wspace = svc_tcp_has_wspace,
1306 .xpo_accept = svc_tcp_accept,
1307};
1308
1309static struct svc_xprt_class svc_tcp_class = {
1310 .xcl_name = "tcp",
1311 .xcl_owner = THIS_MODULE,
1312 .xcl_ops = &svc_tcp_ops,
1313 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1314};
1315
1316void svc_init_xprt_sock(void)
1317{
1318 svc_reg_xprt_class(&svc_tcp_class);
1319 svc_reg_xprt_class(&svc_udp_class);
1320 svc_init_bc_xprt_sock();
1321}
1322
1323void svc_cleanup_xprt_sock(void)
1324{
1325 svc_unreg_xprt_class(&svc_tcp_class);
1326 svc_unreg_xprt_class(&svc_udp_class);
1327 svc_cleanup_bc_xprt_sock();
1328}
1329
1330static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1331{
1332 struct sock *sk = svsk->sk_sk;
1333
1334 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1335 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1336 if (sk->sk_state == TCP_LISTEN) {
1337 dprintk("setting up TCP socket for listening\n");
1338 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1339 sk->sk_data_ready = svc_tcp_listen_data_ready;
1340 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1341 } else {
1342 dprintk("setting up TCP socket for reading\n");
1343 sk->sk_state_change = svc_tcp_state_change;
1344 sk->sk_data_ready = svc_tcp_data_ready;
1345 sk->sk_write_space = svc_tcp_write_space;
1346
1347 svsk->sk_reclen = 0;
1348 svsk->sk_tcplen = 0;
1349 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1350
1351 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1352
1353 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1354 if (sk->sk_state != TCP_ESTABLISHED)
1355 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1356 }
1357}
1358
1359void svc_sock_update_bufs(struct svc_serv *serv)
1360{
1361 /*
1362 * The number of server threads has changed. Update
1363 * rcvbuf and sndbuf accordingly on all sockets
1364 */
1365 struct svc_sock *svsk;
1366
1367 spin_lock_bh(&serv->sv_lock);
1368 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1369 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1370 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1371 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1372 spin_unlock_bh(&serv->sv_lock);
1373}
1374EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1375
1376/*
1377 * Initialize socket for RPC use and create svc_sock struct
1378 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1379 */
1380static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1381 struct socket *sock,
1382 int *errp, int flags)
1383{
1384 struct svc_sock *svsk;
1385 struct sock *inet;
1386 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1387
1388 dprintk("svc: svc_setup_socket %p\n", sock);
1389 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1390 *errp = -ENOMEM;
1391 return NULL;
1392 }
1393
1394 inet = sock->sk;
1395
1396 /* Register socket with portmapper */
1397 if (*errp >= 0 && pmap_register)
1398 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1399 ntohs(inet_sk(inet)->inet_sport));
1400
1401 if (*errp < 0) {
1402 kfree(svsk);
1403 return NULL;
1404 }
1405
1406 inet->sk_user_data = svsk;
1407 svsk->sk_sock = sock;
1408 svsk->sk_sk = inet;
1409 svsk->sk_ostate = inet->sk_state_change;
1410 svsk->sk_odata = inet->sk_data_ready;
1411 svsk->sk_owspace = inet->sk_write_space;
1412
1413 /* Initialize the socket */
1414 if (sock->type == SOCK_DGRAM)
1415 svc_udp_init(svsk, serv);
1416 else {
1417 /* initialise setting must have enough space to
1418 * receive and respond to one request.
1419 */
1420 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1421 4 * serv->sv_max_mesg);
1422 svc_tcp_init(svsk, serv);
1423 }
1424
1425 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1426 svsk, svsk->sk_sk);
1427
1428 return svsk;
1429}
1430
1431/**
1432 * svc_addsock - add a listener socket to an RPC service
1433 * @serv: pointer to RPC service to which to add a new listener
1434 * @fd: file descriptor of the new listener
1435 * @name_return: pointer to buffer to fill in with name of listener
1436 * @len: size of the buffer
1437 *
1438 * Fills in socket name and returns positive length of name if successful.
1439 * Name is terminated with '\n'. On error, returns a negative errno
1440 * value.
1441 */
1442int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1443 const size_t len)
1444{
1445 int err = 0;
1446 struct socket *so = sockfd_lookup(fd, &err);
1447 struct svc_sock *svsk = NULL;
1448
1449 if (!so)
1450 return err;
1451 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1452 err = -EAFNOSUPPORT;
1453 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1454 so->sk->sk_protocol != IPPROTO_UDP)
1455 err = -EPROTONOSUPPORT;
1456 else if (so->state > SS_UNCONNECTED)
1457 err = -EISCONN;
1458 else {
1459 if (!try_module_get(THIS_MODULE))
1460 err = -ENOENT;
1461 else
1462 svsk = svc_setup_socket(serv, so, &err,
1463 SVC_SOCK_DEFAULTS);
1464 if (svsk) {
1465 struct sockaddr_storage addr;
1466 struct sockaddr *sin = (struct sockaddr *)&addr;
1467 int salen;
1468 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1469 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1470 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1471 spin_lock_bh(&serv->sv_lock);
1472 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1473 spin_unlock_bh(&serv->sv_lock);
1474 svc_xprt_received(&svsk->sk_xprt);
1475 err = 0;
1476 } else
1477 module_put(THIS_MODULE);
1478 }
1479 if (err) {
1480 sockfd_put(so);
1481 return err;
1482 }
1483 return svc_one_sock_name(svsk, name_return, len);
1484}
1485EXPORT_SYMBOL_GPL(svc_addsock);
1486
1487/*
1488 * Create socket for RPC service.
1489 */
1490static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1491 int protocol,
1492 struct net *net,
1493 struct sockaddr *sin, int len,
1494 int flags)
1495{
1496 struct svc_sock *svsk;
1497 struct socket *sock;
1498 int error;
1499 int type;
1500 struct sockaddr_storage addr;
1501 struct sockaddr *newsin = (struct sockaddr *)&addr;
1502 int newlen;
1503 int family;
1504 int val;
1505 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1506
1507 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1508 serv->sv_program->pg_name, protocol,
1509 __svc_print_addr(sin, buf, sizeof(buf)));
1510
1511 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1512 printk(KERN_WARNING "svc: only UDP and TCP "
1513 "sockets supported\n");
1514 return ERR_PTR(-EINVAL);
1515 }
1516
1517 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1518 switch (sin->sa_family) {
1519 case AF_INET6:
1520 family = PF_INET6;
1521 break;
1522 case AF_INET:
1523 family = PF_INET;
1524 break;
1525 default:
1526 return ERR_PTR(-EINVAL);
1527 }
1528
1529 error = __sock_create(net, family, type, protocol, &sock, 1);
1530 if (error < 0)
1531 return ERR_PTR(error);
1532
1533 svc_reclassify_socket(sock);
1534
1535 /*
1536 * If this is an PF_INET6 listener, we want to avoid
1537 * getting requests from IPv4 remotes. Those should
1538 * be shunted to a PF_INET listener via rpcbind.
1539 */
1540 val = 1;
1541 if (family == PF_INET6)
1542 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1543 (char *)&val, sizeof(val));
1544
1545 if (type == SOCK_STREAM)
1546 sock->sk->sk_reuse = 1; /* allow address reuse */
1547 error = kernel_bind(sock, sin, len);
1548 if (error < 0)
1549 goto bummer;
1550
1551 newlen = len;
1552 error = kernel_getsockname(sock, newsin, &newlen);
1553 if (error < 0)
1554 goto bummer;
1555
1556 if (protocol == IPPROTO_TCP) {
1557 if ((error = kernel_listen(sock, 64)) < 0)
1558 goto bummer;
1559 }
1560
1561 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1562 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1563 return (struct svc_xprt *)svsk;
1564 }
1565
1566bummer:
1567 dprintk("svc: svc_create_socket error = %d\n", -error);
1568 sock_release(sock);
1569 return ERR_PTR(error);
1570}
1571
1572/*
1573 * Detach the svc_sock from the socket so that no
1574 * more callbacks occur.
1575 */
1576static void svc_sock_detach(struct svc_xprt *xprt)
1577{
1578 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1579 struct sock *sk = svsk->sk_sk;
1580 wait_queue_head_t *wq;
1581
1582 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1583
1584 /* put back the old socket callbacks */
1585 sk->sk_state_change = svsk->sk_ostate;
1586 sk->sk_data_ready = svsk->sk_odata;
1587 sk->sk_write_space = svsk->sk_owspace;
1588
1589 wq = sk_sleep(sk);
1590 if (wq && waitqueue_active(wq))
1591 wake_up_interruptible(wq);
1592}
1593
1594/*
1595 * Disconnect the socket, and reset the callbacks
1596 */
1597static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1598{
1599 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1600
1601 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1602
1603 svc_sock_detach(xprt);
1604
1605 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1606 svc_tcp_clear_pages(svsk);
1607 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1608 }
1609}
1610
1611/*
1612 * Free the svc_sock's socket resources and the svc_sock itself.
1613 */
1614static void svc_sock_free(struct svc_xprt *xprt)
1615{
1616 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1617 dprintk("svc: svc_sock_free(%p)\n", svsk);
1618
1619 if (svsk->sk_sock->file)
1620 sockfd_put(svsk->sk_sock);
1621 else
1622 sock_release(svsk->sk_sock);
1623 kfree(svsk);
1624}
1625
1626#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1627/*
1628 * Create a back channel svc_xprt which shares the fore channel socket.
1629 */
1630static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1631 int protocol,
1632 struct net *net,
1633 struct sockaddr *sin, int len,
1634 int flags)
1635{
1636 struct svc_sock *svsk;
1637 struct svc_xprt *xprt;
1638
1639 if (protocol != IPPROTO_TCP) {
1640 printk(KERN_WARNING "svc: only TCP sockets"
1641 " supported on shared back channel\n");
1642 return ERR_PTR(-EINVAL);
1643 }
1644
1645 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1646 if (!svsk)
1647 return ERR_PTR(-ENOMEM);
1648
1649 xprt = &svsk->sk_xprt;
1650 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1651
1652 serv->sv_bc_xprt = xprt;
1653
1654 return xprt;
1655}
1656
1657/*
1658 * Free a back channel svc_sock.
1659 */
1660static void svc_bc_sock_free(struct svc_xprt *xprt)
1661{
1662 if (xprt)
1663 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1664}
1665#endif /* CONFIG_SUNRPC_BACKCHANNEL */