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