?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
per conn neighbor stall timeout removal
[cor_2_6_31.git] / net / sunrpc / svcsock.c
blob23128ee191ae707973345e3438ccf9e055009bed
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
53 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
54
55
56 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
57 int *errp, int flags);
58 static void svc_udp_data_ready(struct sock *, int);
59 static int svc_udp_recvfrom(struct svc_rqst *);
60 static int svc_udp_sendto(struct svc_rqst *);
61 static void svc_sock_detach(struct svc_xprt *);
62 static void svc_tcp_sock_detach(struct svc_xprt *);
63 static void svc_sock_free(struct svc_xprt *);
64
65 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
66 struct sockaddr *, int, int);
67 #ifdef CONFIG_DEBUG_LOCK_ALLOC
68 static struct lock_class_key svc_key[2];
69 static struct lock_class_key svc_slock_key[2];
70
71 static void svc_reclassify_socket(struct socket *sock)
72 {
73 struct sock *sk = sock->sk;
74 BUG_ON(sock_owned_by_user(sk));
75 switch (sk->sk_family) {
76 case AF_INET:
77 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
78 &svc_slock_key[0],
79 "sk_xprt.xpt_lock-AF_INET-NFSD",
80 &svc_key[0]);
81 break;
82
83 case AF_INET6:
84 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
85 &svc_slock_key[1],
86 "sk_xprt.xpt_lock-AF_INET6-NFSD",
87 &svc_key[1]);
88 break;
89
90 default:
91 BUG();
92 }
93 }
94 #else
95 static void svc_reclassify_socket(struct socket *sock)
96 {
97 }
98 #endif
99
100 /*
101 * Release an skbuff after use
102 */
103 static void svc_release_skb(struct svc_rqst *rqstp)
104 {
105 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
106
107 if (skb) {
108 struct svc_sock *svsk =
109 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
110 rqstp->rq_xprt_ctxt = NULL;
111
112 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
113 skb_free_datagram(svsk->sk_sk, skb);
114 }
115 }
116
117 union svc_pktinfo_u {
118 struct in_pktinfo pkti;
119 struct in6_pktinfo pkti6;
120 };
121 #define SVC_PKTINFO_SPACE \
122 CMSG_SPACE(sizeof(union svc_pktinfo_u))
123
124 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
125 {
126 struct svc_sock *svsk =
127 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
128 switch (svsk->sk_sk->sk_family) {
129 case AF_INET: {
130 struct in_pktinfo *pki = CMSG_DATA(cmh);
131
132 cmh->cmsg_level = SOL_IP;
133 cmh->cmsg_type = IP_PKTINFO;
134 pki->ipi_ifindex = 0;
135 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
136 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
137 }
138 break;
139
140 case AF_INET6: {
141 struct in6_pktinfo *pki = CMSG_DATA(cmh);
142
143 cmh->cmsg_level = SOL_IPV6;
144 cmh->cmsg_type = IPV6_PKTINFO;
145 pki->ipi6_ifindex = 0;
146 ipv6_addr_copy(&pki->ipi6_addr,
147 &rqstp->rq_daddr.addr6);
148 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
149 }
150 break;
151 }
152 return;
153 }
154
155 /*
156 * Generic sendto routine
157 */
158 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
159 {
160 struct svc_sock *svsk =
161 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
162 struct socket *sock = svsk->sk_sock;
163 int slen;
164 union {
165 struct cmsghdr hdr;
166 long all[SVC_PKTINFO_SPACE / sizeof(long)];
167 } buffer;
168 struct cmsghdr *cmh = &buffer.hdr;
169 int len = 0;
170 int result;
171 int size;
172 struct page **ppage = xdr->pages;
173 size_t base = xdr->page_base;
174 unsigned int pglen = xdr->page_len;
175 unsigned int flags = MSG_MORE;
176 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
177
178 slen = xdr->len;
179
180 if (rqstp->rq_prot == IPPROTO_UDP) {
181 struct msghdr msg = {
182 .msg_name = &rqstp->rq_addr,
183 .msg_namelen = rqstp->rq_addrlen,
184 .msg_control = cmh,
185 .msg_controllen = sizeof(buffer),
186 .msg_flags = MSG_MORE,
187 };
188
189 svc_set_cmsg_data(rqstp, cmh);
190
191 if (sock_sendmsg(sock, &msg, 0) < 0)
192 goto out;
193 }
194
195 /* send head */
196 if (slen == xdr->head[0].iov_len)
197 flags = 0;
198 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
199 xdr->head[0].iov_len, flags);
200 if (len != xdr->head[0].iov_len)
201 goto out;
202 slen -= xdr->head[0].iov_len;
203 if (slen == 0)
204 goto out;
205
206 /* send page data */
207 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
208 while (pglen > 0) {
209 if (slen == size)
210 flags = 0;
211 result = kernel_sendpage(sock, *ppage, base, size, flags);
212 if (result > 0)
213 len += result;
214 if (result != size)
215 goto out;
216 slen -= size;
217 pglen -= size;
218 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
219 base = 0;
220 ppage++;
221 }
222 /* send tail */
223 if (xdr->tail[0].iov_len) {
224 result = kernel_sendpage(sock, rqstp->rq_respages[0],
225 ((unsigned long)xdr->tail[0].iov_base)
226 & (PAGE_SIZE-1),
227 xdr->tail[0].iov_len, 0);
228
229 if (result > 0)
230 len += result;
231 }
232 out:
233 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
234 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
235 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
236
237 return len;
238 }
239
240 /*
241 * Report socket names for nfsdfs
242 */
243 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
244 {
245 const struct sock *sk = svsk->sk_sk;
246 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
247 "udp" : "tcp";
248 int len;
249
250 switch (sk->sk_family) {
251 case PF_INET:
252 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
253 proto_name,
254 &inet_sk(sk)->rcv_saddr,
255 inet_sk(sk)->num);
256 break;
257 case PF_INET6:
258 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
259 proto_name,
260 &inet6_sk(sk)->rcv_saddr,
261 inet_sk(sk)->num);
262 break;
263 default:
264 len = snprintf(buf, remaining, "*unknown-%d*\n",
265 sk->sk_family);
266 }
267
268 if (len >= remaining) {
269 *buf = '\0';
270 return -ENAMETOOLONG;
271 }
272 return len;
273 }
274
275 /**
276 * svc_sock_names - construct a list of listener names in a string
277 * @serv: pointer to RPC service
278 * @buf: pointer to a buffer to fill in with socket names
279 * @buflen: size of the buffer to be filled
280 * @toclose: pointer to '\0'-terminated C string containing the name
281 * of a listener to be closed
282 *
283 * Fills in @buf with a '\n'-separated list of names of listener
284 * sockets. If @toclose is not NULL, the socket named by @toclose
285 * is closed, and is not included in the output list.
286 *
287 * Returns positive length of the socket name string, or a negative
288 * errno value on error.
289 */
290 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
291 const char *toclose)
292 {
293 struct svc_sock *svsk, *closesk = NULL;
294 int len = 0;
295
296 if (!serv)
297 return 0;
298
299 spin_lock_bh(&serv->sv_lock);
300 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
301 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
302 if (onelen < 0) {
303 len = onelen;
304 break;
305 }
306 if (toclose && strcmp(toclose, buf + len) == 0)
307 closesk = svsk;
308 else
309 len += onelen;
310 }
311 spin_unlock_bh(&serv->sv_lock);
312
313 if (closesk)
314 /* Should unregister with portmap, but you cannot
315 * unregister just one protocol...
316 */
317 svc_close_xprt(&closesk->sk_xprt);
318 else if (toclose)
319 return -ENOENT;
320 return len;
321 }
322 EXPORT_SYMBOL_GPL(svc_sock_names);
323
324 /*
325 * Check input queue length
326 */
327 static int svc_recv_available(struct svc_sock *svsk)
328 {
329 struct socket *sock = svsk->sk_sock;
330 int avail, err;
331
332 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
333
334 return (err >= 0)? avail : err;
335 }
336
337 /*
338 * Generic recvfrom routine.
339 */
340 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
341 int buflen)
342 {
343 struct svc_sock *svsk =
344 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
345 struct msghdr msg = {
346 .msg_flags = MSG_DONTWAIT,
347 };
348 int len;
349
350 rqstp->rq_xprt_hlen = 0;
351
352 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
353 msg.msg_flags);
354
355 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
356 svsk, iov[0].iov_base, iov[0].iov_len, len);
357 return len;
358 }
359
360 /*
361 * Set socket snd and rcv buffer lengths
362 */
363 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
364 unsigned int rcv)
365 {
366 #if 0
367 mm_segment_t oldfs;
368 oldfs = get_fs(); set_fs(KERNEL_DS);
369 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
370 (char*)&snd, sizeof(snd));
371 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
372 (char*)&rcv, sizeof(rcv));
373 #else
374 /* sock_setsockopt limits use to sysctl_?mem_max,
375 * which isn't acceptable. Until that is made conditional
376 * on not having CAP_SYS_RESOURCE or similar, we go direct...
377 * DaveM said I could!
378 */
379 lock_sock(sock->sk);
380 sock->sk->sk_sndbuf = snd * 2;
381 sock->sk->sk_rcvbuf = rcv * 2;
382 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
383 sock->sk->sk_write_space(sock->sk);
384 release_sock(sock->sk);
385 #endif
386 }
387 /*
388 * INET callback when data has been received on the socket.
389 */
390 static void svc_udp_data_ready(struct sock *sk, int count)
391 {
392 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
393
394 if (svsk) {
395 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
396 svsk, sk, count,
397 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
398 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
399 svc_xprt_enqueue(&svsk->sk_xprt);
400 }
401 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
402 wake_up_interruptible(sk->sk_sleep);
403 }
404
405 /*
406 * INET callback when space is newly available on the socket.
407 */
408 static void svc_write_space(struct sock *sk)
409 {
410 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
411
412 if (svsk) {
413 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
414 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
415 svc_xprt_enqueue(&svsk->sk_xprt);
416 }
417
418 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
419 dprintk("RPC svc_write_space: someone sleeping on %p\n",
420 svsk);
421 wake_up_interruptible(sk->sk_sleep);
422 }
423 }
424
425 static void svc_tcp_write_space(struct sock *sk)
426 {
427 struct socket *sock = sk->sk_socket;
428
429 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
430 clear_bit(SOCK_NOSPACE, &sock->flags);
431 svc_write_space(sk);
432 }
433
434 /*
435 * Copy the UDP datagram's destination address to the rqstp structure.
436 * The 'destination' address in this case is the address to which the
437 * peer sent the datagram, i.e. our local address. For multihomed
438 * hosts, this can change from msg to msg. Note that only the IP
439 * address changes, the port number should remain the same.
440 */
441 static void svc_udp_get_dest_address(struct svc_rqst *rqstp,
442 struct cmsghdr *cmh)
443 {
444 struct svc_sock *svsk =
445 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
446 switch (svsk->sk_sk->sk_family) {
447 case AF_INET: {
448 struct in_pktinfo *pki = CMSG_DATA(cmh);
449 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
450 break;
451 }
452 case AF_INET6: {
453 struct in6_pktinfo *pki = CMSG_DATA(cmh);
454 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
455 break;
456 }
457 }
458 }
459
460 /*
461 * Receive a datagram from a UDP socket.
462 */
463 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
464 {
465 struct svc_sock *svsk =
466 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
467 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
468 struct sk_buff *skb;
469 union {
470 struct cmsghdr hdr;
471 long all[SVC_PKTINFO_SPACE / sizeof(long)];
472 } buffer;
473 struct cmsghdr *cmh = &buffer.hdr;
474 struct msghdr msg = {
475 .msg_name = svc_addr(rqstp),
476 .msg_control = cmh,
477 .msg_controllen = sizeof(buffer),
478 .msg_flags = MSG_DONTWAIT,
479 };
480 size_t len;
481 int err;
482
483 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
484 /* udp sockets need large rcvbuf as all pending
485 * requests are still in that buffer. sndbuf must
486 * also be large enough that there is enough space
487 * for one reply per thread. We count all threads
488 * rather than threads in a particular pool, which
489 * provides an upper bound on the number of threads
490 * which will access the socket.
491 */
492 svc_sock_setbufsize(svsk->sk_sock,
493 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
494 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
495
496 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
497 skb = NULL;
498 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
499 0, 0, MSG_PEEK | MSG_DONTWAIT);
500 if (err >= 0)
501 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
502
503 if (skb == NULL) {
504 if (err != -EAGAIN) {
505 /* possibly an icmp error */
506 dprintk("svc: recvfrom returned error %d\n", -err);
507 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
508 }
509 svc_xprt_received(&svsk->sk_xprt);
510 return -EAGAIN;
511 }
512 len = svc_addr_len(svc_addr(rqstp));
513 if (len == 0)
514 return -EAFNOSUPPORT;
515 rqstp->rq_addrlen = len;
516 if (skb->tstamp.tv64 == 0) {
517 skb->tstamp = ktime_get_real();
518 /* Don't enable netstamp, sunrpc doesn't
519 need that much accuracy */
520 }
521 svsk->sk_sk->sk_stamp = skb->tstamp;
522 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
523
524 /*
525 * Maybe more packets - kick another thread ASAP.
526 */
527 svc_xprt_received(&svsk->sk_xprt);
528
529 len = skb->len - sizeof(struct udphdr);
530 rqstp->rq_arg.len = len;
531
532 rqstp->rq_prot = IPPROTO_UDP;
533
534 if (cmh->cmsg_level != IPPROTO_IP ||
535 cmh->cmsg_type != IP_PKTINFO) {
536 if (net_ratelimit())
537 printk("rpcsvc: received unknown control message:"
538 "%d/%d\n",
539 cmh->cmsg_level, cmh->cmsg_type);
540 skb_free_datagram(svsk->sk_sk, skb);
541 return 0;
542 }
543 svc_udp_get_dest_address(rqstp, cmh);
544
545 if (skb_is_nonlinear(skb)) {
546 /* we have to copy */
547 local_bh_disable();
548 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
549 local_bh_enable();
550 /* checksum error */
551 skb_free_datagram(svsk->sk_sk, skb);
552 return 0;
553 }
554 local_bh_enable();
555 skb_free_datagram(svsk->sk_sk, skb);
556 } else {
557 /* we can use it in-place */
558 rqstp->rq_arg.head[0].iov_base = skb->data +
559 sizeof(struct udphdr);
560 rqstp->rq_arg.head[0].iov_len = len;
561 if (skb_checksum_complete(skb)) {
562 skb_free_datagram(svsk->sk_sk, skb);
563 return 0;
564 }
565 rqstp->rq_xprt_ctxt = skb;
566 }
567
568 rqstp->rq_arg.page_base = 0;
569 if (len <= rqstp->rq_arg.head[0].iov_len) {
570 rqstp->rq_arg.head[0].iov_len = len;
571 rqstp->rq_arg.page_len = 0;
572 rqstp->rq_respages = rqstp->rq_pages+1;
573 } else {
574 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
575 rqstp->rq_respages = rqstp->rq_pages + 1 +
576 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
577 }
578
579 if (serv->sv_stats)
580 serv->sv_stats->netudpcnt++;
581
582 return len;
583 }
584
585 static int
586 svc_udp_sendto(struct svc_rqst *rqstp)
587 {
588 int error;
589
590 error = svc_sendto(rqstp, &rqstp->rq_res);
591 if (error == -ECONNREFUSED)
592 /* ICMP error on earlier request. */
593 error = svc_sendto(rqstp, &rqstp->rq_res);
594
595 return error;
596 }
597
598 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
599 {
600 }
601
602 static int svc_udp_has_wspace(struct svc_xprt *xprt)
603 {
604 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
605 struct svc_serv *serv = xprt->xpt_server;
606 unsigned long required;
607
608 /*
609 * Set the SOCK_NOSPACE flag before checking the available
610 * sock space.
611 */
612 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
613 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
614 if (required*2 > sock_wspace(svsk->sk_sk))
615 return 0;
616 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
617 return 1;
618 }
619
620 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
621 {
622 BUG();
623 return NULL;
624 }
625
626 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
627 struct sockaddr *sa, int salen,
628 int flags)
629 {
630 return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
631 }
632
633 static struct svc_xprt_ops svc_udp_ops = {
634 .xpo_create = svc_udp_create,
635 .xpo_recvfrom = svc_udp_recvfrom,
636 .xpo_sendto = svc_udp_sendto,
637 .xpo_release_rqst = svc_release_skb,
638 .xpo_detach = svc_sock_detach,
639 .xpo_free = svc_sock_free,
640 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
641 .xpo_has_wspace = svc_udp_has_wspace,
642 .xpo_accept = svc_udp_accept,
643 };
644
645 static struct svc_xprt_class svc_udp_class = {
646 .xcl_name = "udp",
647 .xcl_owner = THIS_MODULE,
648 .xcl_ops = &svc_udp_ops,
649 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
650 };
651
652 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
653 {
654 int one = 1;
655 mm_segment_t oldfs;
656
657 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
658 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
659 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
660 svsk->sk_sk->sk_write_space = svc_write_space;
661
662 /* initialise setting must have enough space to
663 * receive and respond to one request.
664 * svc_udp_recvfrom will re-adjust if necessary
665 */
666 svc_sock_setbufsize(svsk->sk_sock,
667 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
668 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
669
670 /* data might have come in before data_ready set up */
671 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
672 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
673
674 oldfs = get_fs();
675 set_fs(KERNEL_DS);
676 /* make sure we get destination address info */
677 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
678 (char __user *)&one, sizeof(one));
679 set_fs(oldfs);
680 }
681
682 /*
683 * A data_ready event on a listening socket means there's a connection
684 * pending. Do not use state_change as a substitute for it.
685 */
686 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
687 {
688 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
689
690 dprintk("svc: socket %p TCP (listen) state change %d\n",
691 sk, sk->sk_state);
692
693 /*
694 * This callback may called twice when a new connection
695 * is established as a child socket inherits everything
696 * from a parent LISTEN socket.
697 * 1) data_ready method of the parent socket will be called
698 * when one of child sockets become ESTABLISHED.
699 * 2) data_ready method of the child socket may be called
700 * when it receives data before the socket is accepted.
701 * In case of 2, we should ignore it silently.
702 */
703 if (sk->sk_state == TCP_LISTEN) {
704 if (svsk) {
705 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
706 svc_xprt_enqueue(&svsk->sk_xprt);
707 } else
708 printk("svc: socket %p: no user data\n", sk);
709 }
710
711 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
712 wake_up_interruptible_all(sk->sk_sleep);
713 }
714
715 /*
716 * A state change on a connected socket means it's dying or dead.
717 */
718 static void svc_tcp_state_change(struct sock *sk)
719 {
720 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
721
722 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
723 sk, sk->sk_state, sk->sk_user_data);
724
725 if (!svsk)
726 printk("svc: socket %p: no user data\n", sk);
727 else {
728 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
729 svc_xprt_enqueue(&svsk->sk_xprt);
730 }
731 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
732 wake_up_interruptible_all(sk->sk_sleep);
733 }
734
735 static void svc_tcp_data_ready(struct sock *sk, int count)
736 {
737 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
738
739 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
740 sk, sk->sk_user_data);
741 if (svsk) {
742 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
743 svc_xprt_enqueue(&svsk->sk_xprt);
744 }
745 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
746 wake_up_interruptible(sk->sk_sleep);
747 }
748
749 /*
750 * Accept a TCP connection
751 */
752 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
753 {
754 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
755 struct sockaddr_storage addr;
756 struct sockaddr *sin = (struct sockaddr *) &addr;
757 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
758 struct socket *sock = svsk->sk_sock;
759 struct socket *newsock;
760 struct svc_sock *newsvsk;
761 int err, slen;
762 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
763
764 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
765 if (!sock)
766 return NULL;
767
768 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
769 err = kernel_accept(sock, &newsock, O_NONBLOCK);
770 if (err < 0) {
771 if (err == -ENOMEM)
772 printk(KERN_WARNING "%s: no more sockets!\n",
773 serv->sv_name);
774 else if (err != -EAGAIN && net_ratelimit())
775 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
776 serv->sv_name, -err);
777 return NULL;
778 }
779 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
780
781 err = kernel_getpeername(newsock, sin, &slen);
782 if (err < 0) {
783 if (net_ratelimit())
784 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
785 serv->sv_name, -err);
786 goto failed; /* aborted connection or whatever */
787 }
788
789 /* Ideally, we would want to reject connections from unauthorized
790 * hosts here, but when we get encryption, the IP of the host won't
791 * tell us anything. For now just warn about unpriv connections.
792 */
793 if (!svc_port_is_privileged(sin)) {
794 dprintk(KERN_WARNING
795 "%s: connect from unprivileged port: %s\n",
796 serv->sv_name,
797 __svc_print_addr(sin, buf, sizeof(buf)));
798 }
799 dprintk("%s: connect from %s\n", serv->sv_name,
800 __svc_print_addr(sin, buf, sizeof(buf)));
801
802 /* make sure that a write doesn't block forever when
803 * low on memory
804 */
805 newsock->sk->sk_sndtimeo = HZ*30;
806
807 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
808 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
809 goto failed;
810 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
811 err = kernel_getsockname(newsock, sin, &slen);
812 if (unlikely(err < 0)) {
813 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
814 slen = offsetof(struct sockaddr, sa_data);
815 }
816 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
817
818 if (serv->sv_stats)
819 serv->sv_stats->nettcpconn++;
820
821 return &newsvsk->sk_xprt;
822
823 failed:
824 sock_release(newsock);
825 return NULL;
826 }
827
828 /*
829 * Receive data from a TCP socket.
830 */
831 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
832 {
833 struct svc_sock *svsk =
834 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
835 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
836 int len;
837 struct kvec *vec;
838 int pnum, vlen;
839
840 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
841 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
842 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
843 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
844
845 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
846 /* sndbuf needs to have room for one request
847 * per thread, otherwise we can stall even when the
848 * network isn't a bottleneck.
849 *
850 * We count all threads rather than threads in a
851 * particular pool, which provides an upper bound
852 * on the number of threads which will access the socket.
853 *
854 * rcvbuf just needs to be able to hold a few requests.
855 * Normally they will be removed from the queue
856 * as soon a a complete request arrives.
857 */
858 svc_sock_setbufsize(svsk->sk_sock,
859 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
860 3 * serv->sv_max_mesg);
861
862 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
863
864 /* Receive data. If we haven't got the record length yet, get
865 * the next four bytes. Otherwise try to gobble up as much as
866 * possible up to the complete record length.
867 */
868 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
869 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
870 struct kvec iov;
871
872 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
873 iov.iov_len = want;
874 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
875 goto error;
876 svsk->sk_tcplen += len;
877
878 if (len < want) {
879 dprintk("svc: short recvfrom while reading record "
880 "length (%d of %d)\n", len, want);
881 svc_xprt_received(&svsk->sk_xprt);
882 return -EAGAIN; /* record header not complete */
883 }
884
885 svsk->sk_reclen = ntohl(svsk->sk_reclen);
886 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
887 /* FIXME: technically, a record can be fragmented,
888 * and non-terminal fragments will not have the top
889 * bit set in the fragment length header.
890 * But apparently no known nfs clients send fragmented
891 * records. */
892 if (net_ratelimit())
893 printk(KERN_NOTICE "RPC: multiple fragments "
894 "per record not supported\n");
895 goto err_delete;
896 }
897 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
898 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
899 if (svsk->sk_reclen > serv->sv_max_mesg) {
900 if (net_ratelimit())
901 printk(KERN_NOTICE "RPC: "
902 "fragment too large: 0x%08lx\n",
903 (unsigned long)svsk->sk_reclen);
904 goto err_delete;
905 }
906 }
907
908 /* Check whether enough data is available */
909 len = svc_recv_available(svsk);
910 if (len < 0)
911 goto error;
912
913 if (len < svsk->sk_reclen) {
914 dprintk("svc: incomplete TCP record (%d of %d)\n",
915 len, svsk->sk_reclen);
916 svc_xprt_received(&svsk->sk_xprt);
917 return -EAGAIN; /* record not complete */
918 }
919 len = svsk->sk_reclen;
920 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
921
922 vec = rqstp->rq_vec;
923 vec[0] = rqstp->rq_arg.head[0];
924 vlen = PAGE_SIZE;
925 pnum = 1;
926 while (vlen < len) {
927 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
928 vec[pnum].iov_len = PAGE_SIZE;
929 pnum++;
930 vlen += PAGE_SIZE;
931 }
932 rqstp->rq_respages = &rqstp->rq_pages[pnum];
933
934 /* Now receive data */
935 len = svc_recvfrom(rqstp, vec, pnum, len);
936 if (len < 0)
937 goto error;
938
939 dprintk("svc: TCP complete record (%d bytes)\n", len);
940 rqstp->rq_arg.len = len;
941 rqstp->rq_arg.page_base = 0;
942 if (len <= rqstp->rq_arg.head[0].iov_len) {
943 rqstp->rq_arg.head[0].iov_len = len;
944 rqstp->rq_arg.page_len = 0;
945 } else {
946 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
947 }
948
949 rqstp->rq_xprt_ctxt = NULL;
950 rqstp->rq_prot = IPPROTO_TCP;
951
952 /* Reset TCP read info */
953 svsk->sk_reclen = 0;
954 svsk->sk_tcplen = 0;
955
956 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
957 svc_xprt_received(&svsk->sk_xprt);
958 if (serv->sv_stats)
959 serv->sv_stats->nettcpcnt++;
960
961 return len;
962
963 err_delete:
964 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
965 return -EAGAIN;
966
967 error:
968 if (len == -EAGAIN) {
969 dprintk("RPC: TCP recvfrom got EAGAIN\n");
970 svc_xprt_received(&svsk->sk_xprt);
971 } else {
972 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
973 svsk->sk_xprt.xpt_server->sv_name, -len);
974 goto err_delete;
975 }
976
977 return len;
978 }
979
980 /*
981 * Send out data on TCP socket.
982 */
983 static int svc_tcp_sendto(struct svc_rqst *rqstp)
984 {
985 struct xdr_buf *xbufp = &rqstp->rq_res;
986 int sent;
987 __be32 reclen;
988
989 /* Set up the first element of the reply kvec.
990 * Any other kvecs that may be in use have been taken
991 * care of by the server implementation itself.
992 */
993 reclen = htonl(0x80000000|((xbufp->len ) - 4));
994 memcpy(xbufp->head[0].iov_base, &reclen, 4);
995
996 if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
997 return -ENOTCONN;
998
999 sent = svc_sendto(rqstp, &rqstp->rq_res);
1000 if (sent != xbufp->len) {
1001 printk(KERN_NOTICE
1002 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1003 "- shutting down socket\n",
1004 rqstp->rq_xprt->xpt_server->sv_name,
1005 (sent<0)?"got error":"sent only",
1006 sent, xbufp->len);
1007 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1008 svc_xprt_enqueue(rqstp->rq_xprt);
1009 sent = -EAGAIN;
1010 }
1011 return sent;
1012 }
1013
1014 /*
1015 * Setup response header. TCP has a 4B record length field.
1016 */
1017 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1018 {
1019 struct kvec *resv = &rqstp->rq_res.head[0];
1020
1021 /* tcp needs a space for the record length... */
1022 svc_putnl(resv, 0);
1023 }
1024
1025 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1026 {
1027 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1028 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1029 int required;
1030
1031 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1032 return 1;
1033 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1034 if (sk_stream_wspace(svsk->sk_sk) >= required)
1035 return 1;
1036 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1037 return 0;
1038 }
1039
1040 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1041 struct sockaddr *sa, int salen,
1042 int flags)
1043 {
1044 return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1045 }
1046
1047 static struct svc_xprt_ops svc_tcp_ops = {
1048 .xpo_create = svc_tcp_create,
1049 .xpo_recvfrom = svc_tcp_recvfrom,
1050 .xpo_sendto = svc_tcp_sendto,
1051 .xpo_release_rqst = svc_release_skb,
1052 .xpo_detach = svc_tcp_sock_detach,
1053 .xpo_free = svc_sock_free,
1054 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1055 .xpo_has_wspace = svc_tcp_has_wspace,
1056 .xpo_accept = svc_tcp_accept,
1057 };
1058
1059 static struct svc_xprt_class svc_tcp_class = {
1060 .xcl_name = "tcp",
1061 .xcl_owner = THIS_MODULE,
1062 .xcl_ops = &svc_tcp_ops,
1063 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1064 };
1065
1066 void svc_init_xprt_sock(void)
1067 {
1068 svc_reg_xprt_class(&svc_tcp_class);
1069 svc_reg_xprt_class(&svc_udp_class);
1070 }
1071
1072 void svc_cleanup_xprt_sock(void)
1073 {
1074 svc_unreg_xprt_class(&svc_tcp_class);
1075 svc_unreg_xprt_class(&svc_udp_class);
1076 }
1077
1078 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1079 {
1080 struct sock *sk = svsk->sk_sk;
1081
1082 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1083 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1084 if (sk->sk_state == TCP_LISTEN) {
1085 dprintk("setting up TCP socket for listening\n");
1086 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1087 sk->sk_data_ready = svc_tcp_listen_data_ready;
1088 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1089 } else {
1090 dprintk("setting up TCP socket for reading\n");
1091 sk->sk_state_change = svc_tcp_state_change;
1092 sk->sk_data_ready = svc_tcp_data_ready;
1093 sk->sk_write_space = svc_tcp_write_space;
1094
1095 svsk->sk_reclen = 0;
1096 svsk->sk_tcplen = 0;
1097
1098 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1099
1100 /* initialise setting must have enough space to
1101 * receive and respond to one request.
1102 * svc_tcp_recvfrom will re-adjust if necessary
1103 */
1104 svc_sock_setbufsize(svsk->sk_sock,
1105 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1106 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1107
1108 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1109 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1110 if (sk->sk_state != TCP_ESTABLISHED)
1111 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1112 }
1113 }
1114
1115 void svc_sock_update_bufs(struct svc_serv *serv)
1116 {
1117 /*
1118 * The number of server threads has changed. Update
1119 * rcvbuf and sndbuf accordingly on all sockets
1120 */
1121 struct list_head *le;
1122
1123 spin_lock_bh(&serv->sv_lock);
1124 list_for_each(le, &serv->sv_permsocks) {
1125 struct svc_sock *svsk =
1126 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1127 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1128 }
1129 list_for_each(le, &serv->sv_tempsocks) {
1130 struct svc_sock *svsk =
1131 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1132 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1133 }
1134 spin_unlock_bh(&serv->sv_lock);
1135 }
1136 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1137
1138 /*
1139 * Initialize socket for RPC use and create svc_sock struct
1140 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1141 */
1142 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1143 struct socket *sock,
1144 int *errp, int flags)
1145 {
1146 struct svc_sock *svsk;
1147 struct sock *inet;
1148 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1149
1150 dprintk("svc: svc_setup_socket %p\n", sock);
1151 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1152 *errp = -ENOMEM;
1153 return NULL;
1154 }
1155
1156 inet = sock->sk;
1157
1158 /* Register socket with portmapper */
1159 if (*errp >= 0 && pmap_register)
1160 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1161 ntohs(inet_sk(inet)->sport));
1162
1163 if (*errp < 0) {
1164 kfree(svsk);
1165 return NULL;
1166 }
1167
1168 inet->sk_user_data = svsk;
1169 svsk->sk_sock = sock;
1170 svsk->sk_sk = inet;
1171 svsk->sk_ostate = inet->sk_state_change;
1172 svsk->sk_odata = inet->sk_data_ready;
1173 svsk->sk_owspace = inet->sk_write_space;
1174
1175 /* Initialize the socket */
1176 if (sock->type == SOCK_DGRAM)
1177 svc_udp_init(svsk, serv);
1178 else
1179 svc_tcp_init(svsk, serv);
1180
1181 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1182 svsk, svsk->sk_sk);
1183
1184 return svsk;
1185 }
1186
1187 /**
1188 * svc_addsock - add a listener socket to an RPC service
1189 * @serv: pointer to RPC service to which to add a new listener
1190 * @fd: file descriptor of the new listener
1191 * @name_return: pointer to buffer to fill in with name of listener
1192 * @len: size of the buffer
1193 *
1194 * Fills in socket name and returns positive length of name if successful.
1195 * Name is terminated with '\n'. On error, returns a negative errno
1196 * value.
1197 */
1198 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1199 const size_t len)
1200 {
1201 int err = 0;
1202 struct socket *so = sockfd_lookup(fd, &err);
1203 struct svc_sock *svsk = NULL;
1204
1205 if (!so)
1206 return err;
1207 if (so->sk->sk_family != AF_INET)
1208 err = -EAFNOSUPPORT;
1209 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1210 so->sk->sk_protocol != IPPROTO_UDP)
1211 err = -EPROTONOSUPPORT;
1212 else if (so->state > SS_UNCONNECTED)
1213 err = -EISCONN;
1214 else {
1215 if (!try_module_get(THIS_MODULE))
1216 err = -ENOENT;
1217 else
1218 svsk = svc_setup_socket(serv, so, &err,
1219 SVC_SOCK_DEFAULTS);
1220 if (svsk) {
1221 struct sockaddr_storage addr;
1222 struct sockaddr *sin = (struct sockaddr *)&addr;
1223 int salen;
1224 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1225 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1226 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1227 spin_lock_bh(&serv->sv_lock);
1228 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1229 spin_unlock_bh(&serv->sv_lock);
1230 svc_xprt_received(&svsk->sk_xprt);
1231 err = 0;
1232 } else
1233 module_put(THIS_MODULE);
1234 }
1235 if (err) {
1236 sockfd_put(so);
1237 return err;
1238 }
1239 return svc_one_sock_name(svsk, name_return, len);
1240 }
1241 EXPORT_SYMBOL_GPL(svc_addsock);
1242
1243 /*
1244 * Create socket for RPC service.
1245 */
1246 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1247 int protocol,
1248 struct sockaddr *sin, int len,
1249 int flags)
1250 {
1251 struct svc_sock *svsk;
1252 struct socket *sock;
1253 int error;
1254 int type;
1255 struct sockaddr_storage addr;
1256 struct sockaddr *newsin = (struct sockaddr *)&addr;
1257 int newlen;
1258 int family;
1259 int val;
1260 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1261
1262 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1263 serv->sv_program->pg_name, protocol,
1264 __svc_print_addr(sin, buf, sizeof(buf)));
1265
1266 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1267 printk(KERN_WARNING "svc: only UDP and TCP "
1268 "sockets supported\n");
1269 return ERR_PTR(-EINVAL);
1270 }
1271
1272 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1273 switch (sin->sa_family) {
1274 case AF_INET6:
1275 family = PF_INET6;
1276 break;
1277 case AF_INET:
1278 family = PF_INET;
1279 break;
1280 default:
1281 return ERR_PTR(-EINVAL);
1282 }
1283
1284 error = sock_create_kern(family, type, protocol, &sock);
1285 if (error < 0)
1286 return ERR_PTR(error);
1287
1288 svc_reclassify_socket(sock);
1289
1290 /*
1291 * If this is an PF_INET6 listener, we want to avoid
1292 * getting requests from IPv4 remotes. Those should
1293 * be shunted to a PF_INET listener via rpcbind.
1294 */
1295 val = 1;
1296 if (family == PF_INET6)
1297 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1298 (char *)&val, sizeof(val));
1299
1300 if (type == SOCK_STREAM)
1301 sock->sk->sk_reuse = 1; /* allow address reuse */
1302 error = kernel_bind(sock, sin, len);
1303 if (error < 0)
1304 goto bummer;
1305
1306 newlen = len;
1307 error = kernel_getsockname(sock, newsin, &newlen);
1308 if (error < 0)
1309 goto bummer;
1310
1311 if (protocol == IPPROTO_TCP) {
1312 if ((error = kernel_listen(sock, 64)) < 0)
1313 goto bummer;
1314 }
1315
1316 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1317 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1318 return (struct svc_xprt *)svsk;
1319 }
1320
1321 bummer:
1322 dprintk("svc: svc_create_socket error = %d\n", -error);
1323 sock_release(sock);
1324 return ERR_PTR(error);
1325 }
1326
1327 /*
1328 * Detach the svc_sock from the socket so that no
1329 * more callbacks occur.
1330 */
1331 static void svc_sock_detach(struct svc_xprt *xprt)
1332 {
1333 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1334 struct sock *sk = svsk->sk_sk;
1335
1336 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1337
1338 /* put back the old socket callbacks */
1339 sk->sk_state_change = svsk->sk_ostate;
1340 sk->sk_data_ready = svsk->sk_odata;
1341 sk->sk_write_space = svsk->sk_owspace;
1342
1343 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1344 wake_up_interruptible(sk->sk_sleep);
1345 }
1346
1347 /*
1348 * Disconnect the socket, and reset the callbacks
1349 */
1350 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1351 {
1352 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1353
1354 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1355
1356 svc_sock_detach(xprt);
1357
1358 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1359 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1360 }
1361
1362 /*
1363 * Free the svc_sock's socket resources and the svc_sock itself.
1364 */
1365 static void svc_sock_free(struct svc_xprt *xprt)
1366 {
1367 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1368 dprintk("svc: svc_sock_free(%p)\n", svsk);
1369
1370 if (svsk->sk_sock->file)
1371 sockfd_put(svsk->sk_sock);
1372 else
1373 sock_release(svsk->sk_sock);
1374 kfree(svsk);
1375 }
1376
1377 /*
1378 * Create a svc_xprt.
1379 *
1380 * For internal use only (e.g. nfsv4.1 backchannel).
1381 * Callers should typically use the xpo_create() method.
1382 */
1383 struct svc_xprt *svc_sock_create(struct svc_serv *serv, int prot)
1384 {
1385 struct svc_sock *svsk;
1386 struct svc_xprt *xprt = NULL;
1387
1388 dprintk("svc: %s\n", __func__);
1389 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1390 if (!svsk)
1391 goto out;
1392
1393 xprt = &svsk->sk_xprt;
1394 if (prot == IPPROTO_TCP)
1395 svc_xprt_init(&svc_tcp_class, xprt, serv);
1396 else if (prot == IPPROTO_UDP)
1397 svc_xprt_init(&svc_udp_class, xprt, serv);
1398 else
1399 BUG();
1400 out:
1401 dprintk("svc: %s return %p\n", __func__, xprt);
1402 return xprt;
1403 }
1404 EXPORT_SYMBOL_GPL(svc_sock_create);
1405
1406 /*
1407 * Destroy a svc_sock.
1408 */
1409 void svc_sock_destroy(struct svc_xprt *xprt)
1410 {
1411 if (xprt)
1412 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1413 }
1414 EXPORT_SYMBOL_GPL(svc_sock_destroy);
connection oriented routing - ARCHIVE NO UPDATES HERE