?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.12.70
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blob8c6e9c75c5256552cd2f0a0c5565c119a6227207
1 /*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
9 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/ip.h>
41 #include <net/ipv6.h>
42 #include <net/tcp.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <trace/events/skb.h>
47
48 #include <linux/sunrpc/types.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/xdr.h>
51 #include <linux/sunrpc/msg_prot.h>
52 #include <linux/sunrpc/svcsock.h>
53 #include <linux/sunrpc/stats.h>
54 #include <linux/sunrpc/xprt.h>
55
56 #include "sunrpc.h"
57
58 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
59
60
61 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
62 int flags);
63 static void svc_udp_data_ready(struct sock *, int);
64 static int svc_udp_recvfrom(struct svc_rqst *);
65 static int svc_udp_sendto(struct svc_rqst *);
66 static void svc_sock_detach(struct svc_xprt *);
67 static void svc_tcp_sock_detach(struct svc_xprt *);
68 static void svc_sock_free(struct svc_xprt *);
69
70 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
74 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77 static void svc_bc_sock_free(struct svc_xprt *xprt);
78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
79
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 static struct lock_class_key svc_key[2];
82 static struct lock_class_key svc_slock_key[2];
83
84 static void svc_reclassify_socket(struct socket *sock)
85 {
86 struct sock *sk = sock->sk;
87
88 WARN_ON_ONCE(sock_owned_by_user(sk));
89 if (sock_owned_by_user(sk))
90 return;
91
92 switch (sk->sk_family) {
93 case AF_INET:
94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
95 &svc_slock_key[0],
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 &svc_key[0]);
98 break;
99
100 case AF_INET6:
101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
102 &svc_slock_key[1],
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
104 &svc_key[1]);
105 break;
106
107 default:
108 BUG();
109 }
110 }
111 #else
112 static void svc_reclassify_socket(struct socket *sock)
113 {
114 }
115 #endif
116
117 /*
118 * Release an skbuff after use
119 */
120 static void svc_release_skb(struct svc_rqst *rqstp)
121 {
122 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
123
124 if (skb) {
125 struct svc_sock *svsk =
126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
127 rqstp->rq_xprt_ctxt = NULL;
128
129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
130 skb_free_datagram_locked(svsk->sk_sk, skb);
131 }
132 }
133
134 union svc_pktinfo_u {
135 struct in_pktinfo pkti;
136 struct in6_pktinfo pkti6;
137 };
138 #define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
140
141 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
142 {
143 struct svc_sock *svsk =
144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 switch (svsk->sk_sk->sk_family) {
146 case AF_INET: {
147 struct in_pktinfo *pki = CMSG_DATA(cmh);
148
149 cmh->cmsg_level = SOL_IP;
150 cmh->cmsg_type = IP_PKTINFO;
151 pki->ipi_ifindex = 0;
152 pki->ipi_spec_dst.s_addr =
153 svc_daddr_in(rqstp)->sin_addr.s_addr;
154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
155 }
156 break;
157
158 case AF_INET6: {
159 struct in6_pktinfo *pki = CMSG_DATA(cmh);
160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
161
162 cmh->cmsg_level = SOL_IPV6;
163 cmh->cmsg_type = IPV6_PKTINFO;
164 pki->ipi6_ifindex = daddr->sin6_scope_id;
165 pki->ipi6_addr = daddr->sin6_addr;
166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 }
168 break;
169 }
170 }
171
172 /*
173 * send routine intended to be shared by the fore- and back-channel
174 */
175 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 struct page *headpage, unsigned long headoffset,
177 struct page *tailpage, unsigned long tailoffset)
178 {
179 int result;
180 int size;
181 struct page **ppage = xdr->pages;
182 size_t base = xdr->page_base;
183 unsigned int pglen = xdr->page_len;
184 unsigned int flags = MSG_MORE;
185 int slen;
186 int len = 0;
187
188 slen = xdr->len;
189
190 /* send head */
191 if (slen == xdr->head[0].iov_len)
192 flags = 0;
193 len = kernel_sendpage(sock, headpage, headoffset,
194 xdr->head[0].iov_len, flags);
195 if (len != xdr->head[0].iov_len)
196 goto out;
197 slen -= xdr->head[0].iov_len;
198 if (slen == 0)
199 goto out;
200
201 /* send page data */
202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
203 while (pglen > 0) {
204 if (slen == size)
205 flags = 0;
206 result = kernel_sendpage(sock, *ppage, base, size, flags);
207 if (result > 0)
208 len += result;
209 if (result != size)
210 goto out;
211 slen -= size;
212 pglen -= size;
213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 base = 0;
215 ppage++;
216 }
217
218 /* send tail */
219 if (xdr->tail[0].iov_len) {
220 result = kernel_sendpage(sock, tailpage, tailoffset,
221 xdr->tail[0].iov_len, 0);
222 if (result > 0)
223 len += result;
224 }
225
226 out:
227 return len;
228 }
229
230
231 /*
232 * Generic sendto routine
233 */
234 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
235 {
236 struct svc_sock *svsk =
237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 struct socket *sock = svsk->sk_sock;
239 union {
240 struct cmsghdr hdr;
241 long all[SVC_PKTINFO_SPACE / sizeof(long)];
242 } buffer;
243 struct cmsghdr *cmh = &buffer.hdr;
244 int len = 0;
245 unsigned long tailoff;
246 unsigned long headoff;
247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
248
249 if (rqstp->rq_prot == IPPROTO_UDP) {
250 struct msghdr msg = {
251 .msg_name = &rqstp->rq_addr,
252 .msg_namelen = rqstp->rq_addrlen,
253 .msg_control = cmh,
254 .msg_controllen = sizeof(buffer),
255 .msg_flags = MSG_MORE,
256 };
257
258 svc_set_cmsg_data(rqstp, cmh);
259
260 if (sock_sendmsg(sock, &msg, 0) < 0)
261 goto out;
262 }
263
264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
265 headoff = 0;
266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 rqstp->rq_respages[0], tailoff);
268
269 out:
270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
273
274 return len;
275 }
276
277 /*
278 * Report socket names for nfsdfs
279 */
280 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
281 {
282 const struct sock *sk = svsk->sk_sk;
283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
284 "udp" : "tcp";
285 int len;
286
287 switch (sk->sk_family) {
288 case PF_INET:
289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
290 proto_name,
291 &inet_sk(sk)->inet_rcv_saddr,
292 inet_sk(sk)->inet_num);
293 break;
294 case PF_INET6:
295 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
296 proto_name,
297 &inet6_sk(sk)->rcv_saddr,
298 inet_sk(sk)->inet_num);
299 break;
300 default:
301 len = snprintf(buf, remaining, "*unknown-%d*\n",
302 sk->sk_family);
303 }
304
305 if (len >= remaining) {
306 *buf = '\0';
307 return -ENAMETOOLONG;
308 }
309 return len;
310 }
311
312 /*
313 * Check input queue length
314 */
315 static int svc_recv_available(struct svc_sock *svsk)
316 {
317 struct socket *sock = svsk->sk_sock;
318 int avail, err;
319
320 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
321
322 return (err >= 0)? avail : err;
323 }
324
325 /*
326 * Generic recvfrom routine.
327 */
328 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
329 int buflen)
330 {
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = {
334 .msg_flags = MSG_DONTWAIT,
335 };
336 int len;
337
338 rqstp->rq_xprt_hlen = 0;
339
340 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
341 msg.msg_flags);
342
343 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
344 svsk, iov[0].iov_base, iov[0].iov_len, len);
345 return len;
346 }
347
348 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
349 struct kvec *iov, int nr,
350 int buflen, unsigned int base)
351 {
352 size_t save_iovlen;
353 void *save_iovbase;
354 unsigned int i;
355 int ret;
356
357 if (base == 0)
358 return svc_recvfrom(rqstp, iov, nr, buflen);
359
360 for (i = 0; i < nr; i++) {
361 if (iov[i].iov_len > base)
362 break;
363 base -= iov[i].iov_len;
364 }
365 save_iovlen = iov[i].iov_len;
366 save_iovbase = iov[i].iov_base;
367 iov[i].iov_len -= base;
368 iov[i].iov_base += base;
369 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
370 iov[i].iov_len = save_iovlen;
371 iov[i].iov_base = save_iovbase;
372 return ret;
373 }
374
375 /*
376 * Set socket snd and rcv buffer lengths
377 */
378 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
379 unsigned int rcv)
380 {
381 #if 0
382 mm_segment_t oldfs;
383 oldfs = get_fs(); set_fs(KERNEL_DS);
384 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
385 (char*)&snd, sizeof(snd));
386 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
387 (char*)&rcv, sizeof(rcv));
388 #else
389 /* sock_setsockopt limits use to sysctl_?mem_max,
390 * which isn't acceptable. Until that is made conditional
391 * on not having CAP_SYS_RESOURCE or similar, we go direct...
392 * DaveM said I could!
393 */
394 lock_sock(sock->sk);
395 sock->sk->sk_sndbuf = snd * 2;
396 sock->sk->sk_rcvbuf = rcv * 2;
397 sock->sk->sk_write_space(sock->sk);
398 release_sock(sock->sk);
399 #endif
400 }
401 /*
402 * INET callback when data has been received on the socket.
403 */
404 static void svc_udp_data_ready(struct sock *sk, int count)
405 {
406 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
407 wait_queue_head_t *wq = sk_sleep(sk);
408
409 if (svsk) {
410 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
411 svsk, sk, count,
412 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
413 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
414 svc_xprt_enqueue(&svsk->sk_xprt);
415 }
416 if (wq && waitqueue_active(wq))
417 wake_up_interruptible(wq);
418 }
419
420 /*
421 * INET callback when space is newly available on the socket.
422 */
423 static void svc_write_space(struct sock *sk)
424 {
425 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
426 wait_queue_head_t *wq = sk_sleep(sk);
427
428 if (svsk) {
429 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
430 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
431 svc_xprt_enqueue(&svsk->sk_xprt);
432 }
433
434 if (wq && waitqueue_active(wq)) {
435 dprintk("RPC svc_write_space: someone sleeping on %p\n",
436 svsk);
437 wake_up_interruptible(wq);
438 }
439 }
440
441 static void svc_tcp_write_space(struct sock *sk)
442 {
443 struct socket *sock = sk->sk_socket;
444
445 if (sk_stream_is_writeable(sk) && sock)
446 clear_bit(SOCK_NOSPACE, &sock->flags);
447 svc_write_space(sk);
448 }
449
450 /*
451 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
452 */
453 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
454 struct cmsghdr *cmh)
455 {
456 struct in_pktinfo *pki = CMSG_DATA(cmh);
457 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
458
459 if (cmh->cmsg_type != IP_PKTINFO)
460 return 0;
461
462 daddr->sin_family = AF_INET;
463 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
464 return 1;
465 }
466
467 /*
468 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
469 */
470 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
471 struct cmsghdr *cmh)
472 {
473 struct in6_pktinfo *pki = CMSG_DATA(cmh);
474 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
475
476 if (cmh->cmsg_type != IPV6_PKTINFO)
477 return 0;
478
479 daddr->sin6_family = AF_INET6;
480 daddr->sin6_addr = pki->ipi6_addr;
481 daddr->sin6_scope_id = pki->ipi6_ifindex;
482 return 1;
483 }
484
485 /*
486 * Copy the UDP datagram's destination address to the rqstp structure.
487 * The 'destination' address in this case is the address to which the
488 * peer sent the datagram, i.e. our local address. For multihomed
489 * hosts, this can change from msg to msg. Note that only the IP
490 * address changes, the port number should remain the same.
491 */
492 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
493 struct cmsghdr *cmh)
494 {
495 switch (cmh->cmsg_level) {
496 case SOL_IP:
497 return svc_udp_get_dest_address4(rqstp, cmh);
498 case SOL_IPV6:
499 return svc_udp_get_dest_address6(rqstp, cmh);
500 }
501
502 return 0;
503 }
504
505 /*
506 * Receive a datagram from a UDP socket.
507 */
508 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
509 {
510 struct svc_sock *svsk =
511 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
512 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
513 struct sk_buff *skb;
514 union {
515 struct cmsghdr hdr;
516 long all[SVC_PKTINFO_SPACE / sizeof(long)];
517 } buffer;
518 struct cmsghdr *cmh = &buffer.hdr;
519 struct msghdr msg = {
520 .msg_name = svc_addr(rqstp),
521 .msg_control = cmh,
522 .msg_controllen = sizeof(buffer),
523 .msg_flags = MSG_DONTWAIT,
524 };
525 size_t len;
526 int err;
527
528 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
529 /* udp sockets need large rcvbuf as all pending
530 * requests are still in that buffer. sndbuf must
531 * also be large enough that there is enough space
532 * for one reply per thread. We count all threads
533 * rather than threads in a particular pool, which
534 * provides an upper bound on the number of threads
535 * which will access the socket.
536 */
537 svc_sock_setbufsize(svsk->sk_sock,
538 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
539 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
540
541 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
542 skb = NULL;
543 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
544 0, 0, MSG_PEEK | MSG_DONTWAIT);
545 if (err >= 0)
546 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
547
548 if (skb == NULL) {
549 if (err != -EAGAIN) {
550 /* possibly an icmp error */
551 dprintk("svc: recvfrom returned error %d\n", -err);
552 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
553 }
554 return 0;
555 }
556 len = svc_addr_len(svc_addr(rqstp));
557 rqstp->rq_addrlen = len;
558 if (skb->tstamp.tv64 == 0) {
559 skb->tstamp = ktime_get_real();
560 /* Don't enable netstamp, sunrpc doesn't
561 need that much accuracy */
562 }
563 svsk->sk_sk->sk_stamp = skb->tstamp;
564 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
565
566 len = skb->len - sizeof(struct udphdr);
567 rqstp->rq_arg.len = len;
568
569 rqstp->rq_prot = IPPROTO_UDP;
570
571 if (!svc_udp_get_dest_address(rqstp, cmh)) {
572 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
573 cmh->cmsg_level, cmh->cmsg_type);
574 goto out_free;
575 }
576 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
577
578 if (skb_is_nonlinear(skb)) {
579 /* we have to copy */
580 local_bh_disable();
581 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
582 local_bh_enable();
583 /* checksum error */
584 goto out_free;
585 }
586 local_bh_enable();
587 skb_free_datagram_locked(svsk->sk_sk, skb);
588 } else {
589 /* we can use it in-place */
590 rqstp->rq_arg.head[0].iov_base = skb->data +
591 sizeof(struct udphdr);
592 rqstp->rq_arg.head[0].iov_len = len;
593 if (skb_checksum_complete(skb))
594 goto out_free;
595 rqstp->rq_xprt_ctxt = skb;
596 }
597
598 rqstp->rq_arg.page_base = 0;
599 if (len <= rqstp->rq_arg.head[0].iov_len) {
600 rqstp->rq_arg.head[0].iov_len = len;
601 rqstp->rq_arg.page_len = 0;
602 rqstp->rq_respages = rqstp->rq_pages+1;
603 } else {
604 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
605 rqstp->rq_respages = rqstp->rq_pages + 1 +
606 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
607 }
608 rqstp->rq_next_page = rqstp->rq_respages+1;
609
610 if (serv->sv_stats)
611 serv->sv_stats->netudpcnt++;
612
613 return len;
614 out_free:
615 trace_kfree_skb(skb, svc_udp_recvfrom);
616 skb_free_datagram_locked(svsk->sk_sk, skb);
617 return 0;
618 }
619
620 static int
621 svc_udp_sendto(struct svc_rqst *rqstp)
622 {
623 int error;
624
625 error = svc_sendto(rqstp, &rqstp->rq_res);
626 if (error == -ECONNREFUSED)
627 /* ICMP error on earlier request. */
628 error = svc_sendto(rqstp, &rqstp->rq_res);
629
630 return error;
631 }
632
633 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
634 {
635 }
636
637 static int svc_udp_has_wspace(struct svc_xprt *xprt)
638 {
639 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
640 struct svc_serv *serv = xprt->xpt_server;
641 unsigned long required;
642
643 /*
644 * Set the SOCK_NOSPACE flag before checking the available
645 * sock space.
646 */
647 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
648 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
649 if (required*2 > sock_wspace(svsk->sk_sk))
650 return 0;
651 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
652 return 1;
653 }
654
655 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
656 {
657 BUG();
658 return NULL;
659 }
660
661 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
662 struct net *net,
663 struct sockaddr *sa, int salen,
664 int flags)
665 {
666 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
667 }
668
669 static struct svc_xprt_ops svc_udp_ops = {
670 .xpo_create = svc_udp_create,
671 .xpo_recvfrom = svc_udp_recvfrom,
672 .xpo_sendto = svc_udp_sendto,
673 .xpo_release_rqst = svc_release_skb,
674 .xpo_detach = svc_sock_detach,
675 .xpo_free = svc_sock_free,
676 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
677 .xpo_has_wspace = svc_udp_has_wspace,
678 .xpo_accept = svc_udp_accept,
679 };
680
681 static struct svc_xprt_class svc_udp_class = {
682 .xcl_name = "udp",
683 .xcl_owner = THIS_MODULE,
684 .xcl_ops = &svc_udp_ops,
685 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
686 .xcl_ident = XPRT_TRANSPORT_UDP,
687 };
688
689 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
690 {
691 int err, level, optname, one = 1;
692
693 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
694 &svsk->sk_xprt, serv);
695 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
696 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
697 svsk->sk_sk->sk_write_space = svc_write_space;
698
699 /* initialise setting must have enough space to
700 * receive and respond to one request.
701 * svc_udp_recvfrom will re-adjust if necessary
702 */
703 svc_sock_setbufsize(svsk->sk_sock,
704 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
705 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
706
707 /* data might have come in before data_ready set up */
708 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
709 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
710
711 /* make sure we get destination address info */
712 switch (svsk->sk_sk->sk_family) {
713 case AF_INET:
714 level = SOL_IP;
715 optname = IP_PKTINFO;
716 break;
717 case AF_INET6:
718 level = SOL_IPV6;
719 optname = IPV6_RECVPKTINFO;
720 break;
721 default:
722 BUG();
723 }
724 err = kernel_setsockopt(svsk->sk_sock, level, optname,
725 (char *)&one, sizeof(one));
726 dprintk("svc: kernel_setsockopt returned %d\n", err);
727 }
728
729 /*
730 * A data_ready event on a listening socket means there's a connection
731 * pending. Do not use state_change as a substitute for it.
732 */
733 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
734 {
735 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
736 wait_queue_head_t *wq;
737
738 dprintk("svc: socket %p TCP (listen) state change %d\n",
739 sk, sk->sk_state);
740
741 /*
742 * This callback may called twice when a new connection
743 * is established as a child socket inherits everything
744 * from a parent LISTEN socket.
745 * 1) data_ready method of the parent socket will be called
746 * when one of child sockets become ESTABLISHED.
747 * 2) data_ready method of the child socket may be called
748 * when it receives data before the socket is accepted.
749 * In case of 2, we should ignore it silently.
750 */
751 if (sk->sk_state == TCP_LISTEN) {
752 if (svsk) {
753 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
754 svc_xprt_enqueue(&svsk->sk_xprt);
755 } else
756 printk("svc: socket %p: no user data\n", sk);
757 }
758
759 wq = sk_sleep(sk);
760 if (wq && waitqueue_active(wq))
761 wake_up_interruptible_all(wq);
762 }
763
764 /*
765 * A state change on a connected socket means it's dying or dead.
766 */
767 static void svc_tcp_state_change(struct sock *sk)
768 {
769 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
770 wait_queue_head_t *wq = sk_sleep(sk);
771
772 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
773 sk, sk->sk_state, sk->sk_user_data);
774
775 if (!svsk)
776 printk("svc: socket %p: no user data\n", sk);
777 else {
778 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
779 svc_xprt_enqueue(&svsk->sk_xprt);
780 }
781 if (wq && waitqueue_active(wq))
782 wake_up_interruptible_all(wq);
783 }
784
785 static void svc_tcp_data_ready(struct sock *sk, int count)
786 {
787 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
788 wait_queue_head_t *wq = sk_sleep(sk);
789
790 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
791 sk, sk->sk_user_data);
792 if (svsk) {
793 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
794 svc_xprt_enqueue(&svsk->sk_xprt);
795 }
796 if (wq && waitqueue_active(wq))
797 wake_up_interruptible(wq);
798 }
799
800 /*
801 * Accept a TCP connection
802 */
803 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
804 {
805 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
806 struct sockaddr_storage addr;
807 struct sockaddr *sin = (struct sockaddr *) &addr;
808 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
809 struct socket *sock = svsk->sk_sock;
810 struct socket *newsock;
811 struct svc_sock *newsvsk;
812 int err, slen;
813 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
814
815 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
816 if (!sock)
817 return NULL;
818
819 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
820 err = kernel_accept(sock, &newsock, O_NONBLOCK);
821 if (err < 0) {
822 if (err == -ENOMEM)
823 printk(KERN_WARNING "%s: no more sockets!\n",
824 serv->sv_name);
825 else if (err != -EAGAIN)
826 net_warn_ratelimited("%s: accept failed (err %d)!\n",
827 serv->sv_name, -err);
828 return NULL;
829 }
830 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
831
832 err = kernel_getpeername(newsock, sin, &slen);
833 if (err < 0) {
834 net_warn_ratelimited("%s: peername failed (err %d)!\n",
835 serv->sv_name, -err);
836 goto failed; /* aborted connection or whatever */
837 }
838
839 /* Ideally, we would want to reject connections from unauthorized
840 * hosts here, but when we get encryption, the IP of the host won't
841 * tell us anything. For now just warn about unpriv connections.
842 */
843 if (!svc_port_is_privileged(sin)) {
844 dprintk(KERN_WARNING
845 "%s: connect from unprivileged port: %s\n",
846 serv->sv_name,
847 __svc_print_addr(sin, buf, sizeof(buf)));
848 }
849 dprintk("%s: connect from %s\n", serv->sv_name,
850 __svc_print_addr(sin, buf, sizeof(buf)));
851
852 /* make sure that a write doesn't block forever when
853 * low on memory
854 */
855 newsock->sk->sk_sndtimeo = HZ*30;
856
857 newsvsk = svc_setup_socket(serv, newsock,
858 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
859 if (IS_ERR(newsvsk))
860 goto failed;
861 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
862 err = kernel_getsockname(newsock, sin, &slen);
863 if (unlikely(err < 0)) {
864 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
865 slen = offsetof(struct sockaddr, sa_data);
866 }
867 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
868
869 if (serv->sv_stats)
870 serv->sv_stats->nettcpconn++;
871
872 return &newsvsk->sk_xprt;
873
874 failed:
875 sock_release(newsock);
876 return NULL;
877 }
878
879 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
880 {
881 unsigned int i, len, npages;
882
883 if (svsk->sk_datalen == 0)
884 return 0;
885 len = svsk->sk_datalen;
886 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
887 for (i = 0; i < npages; i++) {
888 if (rqstp->rq_pages[i] != NULL)
889 put_page(rqstp->rq_pages[i]);
890 BUG_ON(svsk->sk_pages[i] == NULL);
891 rqstp->rq_pages[i] = svsk->sk_pages[i];
892 svsk->sk_pages[i] = NULL;
893 }
894 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
895 return len;
896 }
897
898 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
899 {
900 unsigned int i, len, npages;
901
902 if (svsk->sk_datalen == 0)
903 return;
904 len = svsk->sk_datalen;
905 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
906 for (i = 0; i < npages; i++) {
907 svsk->sk_pages[i] = rqstp->rq_pages[i];
908 rqstp->rq_pages[i] = NULL;
909 }
910 }
911
912 static void svc_tcp_clear_pages(struct svc_sock *svsk)
913 {
914 unsigned int i, len, npages;
915
916 if (svsk->sk_datalen == 0)
917 goto out;
918 len = svsk->sk_datalen;
919 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
920 for (i = 0; i < npages; i++) {
921 if (svsk->sk_pages[i] == NULL) {
922 WARN_ON_ONCE(1);
923 continue;
924 }
925 put_page(svsk->sk_pages[i]);
926 svsk->sk_pages[i] = NULL;
927 }
928 out:
929 svsk->sk_tcplen = 0;
930 svsk->sk_datalen = 0;
931 }
932
933 /*
934 * Receive fragment record header.
935 * If we haven't gotten the record length yet, get the next four bytes.
936 */
937 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
938 {
939 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
940 unsigned int want;
941 int len;
942
943 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
944
945 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
946 struct kvec iov;
947
948 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
949 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
950 iov.iov_len = want;
951 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
952 goto error;
953 svsk->sk_tcplen += len;
954
955 if (len < want) {
956 dprintk("svc: short recvfrom while reading record "
957 "length (%d of %d)\n", len, want);
958 return -EAGAIN;
959 }
960
961 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
962 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
963 serv->sv_max_mesg) {
964 net_notice_ratelimited("RPC: fragment too large: %d\n",
965 svc_sock_reclen(svsk));
966 goto err_delete;
967 }
968 }
969
970 return svc_sock_reclen(svsk);
971 error:
972 dprintk("RPC: TCP recv_record got %d\n", len);
973 return len;
974 err_delete:
975 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
976 return -EAGAIN;
977 }
978
979 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
980 {
981 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
982 struct rpc_rqst *req = NULL;
983 struct kvec *src, *dst;
984 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
985 __be32 xid;
986 __be32 calldir;
987
988 xid = *p++;
989 calldir = *p;
990
991 if (bc_xprt)
992 req = xprt_lookup_rqst(bc_xprt, xid);
993
994 if (!req) {
995 printk(KERN_NOTICE
996 "%s: Got unrecognized reply: "
997 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
998 __func__, ntohl(calldir),
999 bc_xprt, xid);
1000 return -EAGAIN;
1001 }
1002
1003 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1004 /*
1005 * XXX!: cheating for now! Only copying HEAD.
1006 * But we know this is good enough for now (in fact, for any
1007 * callback reply in the forseeable future).
1008 */
1009 dst = &req->rq_private_buf.head[0];
1010 src = &rqstp->rq_arg.head[0];
1011 if (dst->iov_len < src->iov_len)
1012 return -EAGAIN; /* whatever; just giving up. */
1013 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1014 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1015 rqstp->rq_arg.len = 0;
1016 return 0;
1017 }
1018
1019 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1020 {
1021 int i = 0;
1022 int t = 0;
1023
1024 while (t < len) {
1025 vec[i].iov_base = page_address(pages[i]);
1026 vec[i].iov_len = PAGE_SIZE;
1027 i++;
1028 t += PAGE_SIZE;
1029 }
1030 return i;
1031 }
1032
1033 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1034 {
1035 /* If we have more data, signal svc_xprt_enqueue() to try again */
1036 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1037 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1038 dprintk("svc: TCP %s record (%d bytes)\n",
1039 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1040 svc_sock_reclen(svsk));
1041 svsk->sk_tcplen = 0;
1042 svsk->sk_reclen = 0;
1043 }
1044
1045 /*
1046 * Receive data from a TCP socket.
1047 */
1048 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1049 {
1050 struct svc_sock *svsk =
1051 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1052 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1053 int len;
1054 struct kvec *vec;
1055 unsigned int want, base;
1056 __be32 *p;
1057 __be32 calldir;
1058 int pnum;
1059
1060 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1061 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1062 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1063 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1064
1065 len = svc_tcp_recv_record(svsk, rqstp);
1066 if (len < 0)
1067 goto error;
1068
1069 base = svc_tcp_restore_pages(svsk, rqstp);
1070 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1071
1072 vec = rqstp->rq_vec;
1073
1074 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1075 svsk->sk_datalen + want);
1076
1077 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1078 rqstp->rq_next_page = rqstp->rq_respages + 1;
1079
1080 /* Now receive data */
1081 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1082 if (len >= 0) {
1083 svsk->sk_tcplen += len;
1084 svsk->sk_datalen += len;
1085 }
1086 if (len != want || !svc_sock_final_rec(svsk)) {
1087 svc_tcp_save_pages(svsk, rqstp);
1088 if (len < 0 && len != -EAGAIN)
1089 goto err_delete;
1090 if (len == want)
1091 svc_tcp_fragment_received(svsk);
1092 else
1093 dprintk("svc: incomplete TCP record (%d of %d)\n",
1094 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1095 svc_sock_reclen(svsk));
1096 goto err_noclose;
1097 }
1098
1099 if (svsk->sk_datalen < 8) {
1100 svsk->sk_datalen = 0;
1101 goto err_delete; /* client is nuts. */
1102 }
1103
1104 rqstp->rq_arg.len = svsk->sk_datalen;
1105 rqstp->rq_arg.page_base = 0;
1106 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1107 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1108 rqstp->rq_arg.page_len = 0;
1109 } else
1110 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1111
1112 rqstp->rq_xprt_ctxt = NULL;
1113 rqstp->rq_prot = IPPROTO_TCP;
1114
1115 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1116 calldir = p[1];
1117 if (calldir)
1118 len = receive_cb_reply(svsk, rqstp);
1119
1120 /* Reset TCP read info */
1121 svsk->sk_datalen = 0;
1122 svc_tcp_fragment_received(svsk);
1123
1124 if (len < 0)
1125 goto error;
1126
1127 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1128 if (serv->sv_stats)
1129 serv->sv_stats->nettcpcnt++;
1130
1131 return rqstp->rq_arg.len;
1132
1133 error:
1134 if (len != -EAGAIN)
1135 goto err_delete;
1136 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1137 return 0;
1138 err_delete:
1139 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1140 svsk->sk_xprt.xpt_server->sv_name, -len);
1141 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1142 err_noclose:
1143 return 0; /* record not complete */
1144 }
1145
1146 /*
1147 * Send out data on TCP socket.
1148 */
1149 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1150 {
1151 struct xdr_buf *xbufp = &rqstp->rq_res;
1152 int sent;
1153 __be32 reclen;
1154
1155 /* Set up the first element of the reply kvec.
1156 * Any other kvecs that may be in use have been taken
1157 * care of by the server implementation itself.
1158 */
1159 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1160 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1161
1162 sent = svc_sendto(rqstp, &rqstp->rq_res);
1163 if (sent != xbufp->len) {
1164 printk(KERN_NOTICE
1165 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1166 "- shutting down socket\n",
1167 rqstp->rq_xprt->xpt_server->sv_name,
1168 (sent<0)?"got error":"sent only",
1169 sent, xbufp->len);
1170 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1171 svc_xprt_enqueue(rqstp->rq_xprt);
1172 sent = -EAGAIN;
1173 }
1174 return sent;
1175 }
1176
1177 /*
1178 * Setup response header. TCP has a 4B record length field.
1179 */
1180 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1181 {
1182 struct kvec *resv = &rqstp->rq_res.head[0];
1183
1184 /* tcp needs a space for the record length... */
1185 svc_putnl(resv, 0);
1186 }
1187
1188 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1189 {
1190 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1191 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1192 int required;
1193
1194 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1195 return 1;
1196 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1197 if (sk_stream_wspace(svsk->sk_sk) >= required ||
1198 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
1199 atomic_read(&xprt->xpt_reserved) == 0))
1200 return 1;
1201 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1202 return 0;
1203 }
1204
1205 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1206 struct net *net,
1207 struct sockaddr *sa, int salen,
1208 int flags)
1209 {
1210 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1211 }
1212
1213 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1214 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1215 struct net *, struct sockaddr *,
1216 int, int);
1217 static void svc_bc_sock_free(struct svc_xprt *xprt);
1218
1219 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1220 struct net *net,
1221 struct sockaddr *sa, int salen,
1222 int flags)
1223 {
1224 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1225 }
1226
1227 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1228 {
1229 }
1230
1231 static struct svc_xprt_ops svc_tcp_bc_ops = {
1232 .xpo_create = svc_bc_tcp_create,
1233 .xpo_detach = svc_bc_tcp_sock_detach,
1234 .xpo_free = svc_bc_sock_free,
1235 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1236 };
1237
1238 static struct svc_xprt_class svc_tcp_bc_class = {
1239 .xcl_name = "tcp-bc",
1240 .xcl_owner = THIS_MODULE,
1241 .xcl_ops = &svc_tcp_bc_ops,
1242 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1243 };
1244
1245 static void svc_init_bc_xprt_sock(void)
1246 {
1247 svc_reg_xprt_class(&svc_tcp_bc_class);
1248 }
1249
1250 static void svc_cleanup_bc_xprt_sock(void)
1251 {
1252 svc_unreg_xprt_class(&svc_tcp_bc_class);
1253 }
1254 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1255 static void svc_init_bc_xprt_sock(void)
1256 {
1257 }
1258
1259 static void svc_cleanup_bc_xprt_sock(void)
1260 {
1261 }
1262 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1263
1264 static struct svc_xprt_ops svc_tcp_ops = {
1265 .xpo_create = svc_tcp_create,
1266 .xpo_recvfrom = svc_tcp_recvfrom,
1267 .xpo_sendto = svc_tcp_sendto,
1268 .xpo_release_rqst = svc_release_skb,
1269 .xpo_detach = svc_tcp_sock_detach,
1270 .xpo_free = svc_sock_free,
1271 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1272 .xpo_has_wspace = svc_tcp_has_wspace,
1273 .xpo_accept = svc_tcp_accept,
1274 };
1275
1276 static struct svc_xprt_class svc_tcp_class = {
1277 .xcl_name = "tcp",
1278 .xcl_owner = THIS_MODULE,
1279 .xcl_ops = &svc_tcp_ops,
1280 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1281 .xcl_ident = XPRT_TRANSPORT_TCP,
1282 };
1283
1284 void svc_init_xprt_sock(void)
1285 {
1286 svc_reg_xprt_class(&svc_tcp_class);
1287 svc_reg_xprt_class(&svc_udp_class);
1288 svc_init_bc_xprt_sock();
1289 }
1290
1291 void svc_cleanup_xprt_sock(void)
1292 {
1293 svc_unreg_xprt_class(&svc_tcp_class);
1294 svc_unreg_xprt_class(&svc_udp_class);
1295 svc_cleanup_bc_xprt_sock();
1296 }
1297
1298 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1299 {
1300 struct sock *sk = svsk->sk_sk;
1301
1302 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1303 &svsk->sk_xprt, serv);
1304 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1305 if (sk->sk_state == TCP_LISTEN) {
1306 dprintk("setting up TCP socket for listening\n");
1307 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1308 sk->sk_data_ready = svc_tcp_listen_data_ready;
1309 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1310 } else {
1311 dprintk("setting up TCP socket for reading\n");
1312 sk->sk_state_change = svc_tcp_state_change;
1313 sk->sk_data_ready = svc_tcp_data_ready;
1314 sk->sk_write_space = svc_tcp_write_space;
1315
1316 svsk->sk_reclen = 0;
1317 svsk->sk_tcplen = 0;
1318 svsk->sk_datalen = 0;
1319 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1320
1321 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1322
1323 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1324 if (sk->sk_state != TCP_ESTABLISHED)
1325 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1326 }
1327 }
1328
1329 void svc_sock_update_bufs(struct svc_serv *serv)
1330 {
1331 /*
1332 * The number of server threads has changed. Update
1333 * rcvbuf and sndbuf accordingly on all sockets
1334 */
1335 struct svc_sock *svsk;
1336
1337 spin_lock_bh(&serv->sv_lock);
1338 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1339 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1340 spin_unlock_bh(&serv->sv_lock);
1341 }
1342 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1343
1344 /*
1345 * Initialize socket for RPC use and create svc_sock struct
1346 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1347 */
1348 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1349 struct socket *sock,
1350 int flags)
1351 {
1352 struct svc_sock *svsk;
1353 struct sock *inet;
1354 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1355 int err = 0;
1356
1357 dprintk("svc: svc_setup_socket %p\n", sock);
1358 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1359 if (!svsk)
1360 return ERR_PTR(-ENOMEM);
1361
1362 inet = sock->sk;
1363
1364 /* Register socket with portmapper */
1365 if (pmap_register)
1366 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1367 inet->sk_protocol,
1368 ntohs(inet_sk(inet)->inet_sport));
1369
1370 if (err < 0) {
1371 kfree(svsk);
1372 return ERR_PTR(err);
1373 }
1374
1375 inet->sk_user_data = svsk;
1376 svsk->sk_sock = sock;
1377 svsk->sk_sk = inet;
1378 svsk->sk_ostate = inet->sk_state_change;
1379 svsk->sk_odata = inet->sk_data_ready;
1380 svsk->sk_owspace = inet->sk_write_space;
1381
1382 /* Initialize the socket */
1383 if (sock->type == SOCK_DGRAM)
1384 svc_udp_init(svsk, serv);
1385 else {
1386 /* initialise setting must have enough space to
1387 * receive and respond to one request.
1388 */
1389 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1390 4 * serv->sv_max_mesg);
1391 svc_tcp_init(svsk, serv);
1392 }
1393
1394 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1395 svsk, svsk->sk_sk);
1396
1397 return svsk;
1398 }
1399
1400 bool svc_alien_sock(struct net *net, int fd)
1401 {
1402 int err;
1403 struct socket *sock = sockfd_lookup(fd, &err);
1404 bool ret = false;
1405
1406 if (!sock)
1407 goto out;
1408 if (sock_net(sock->sk) != net)
1409 ret = true;
1410 sockfd_put(sock);
1411 out:
1412 return ret;
1413 }
1414 EXPORT_SYMBOL_GPL(svc_alien_sock);
1415
1416 /**
1417 * svc_addsock - add a listener socket to an RPC service
1418 * @serv: pointer to RPC service to which to add a new listener
1419 * @fd: file descriptor of the new listener
1420 * @name_return: pointer to buffer to fill in with name of listener
1421 * @len: size of the buffer
1422 *
1423 * Fills in socket name and returns positive length of name if successful.
1424 * Name is terminated with '\n'. On error, returns a negative errno
1425 * value.
1426 */
1427 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1428 const size_t len)
1429 {
1430 int err = 0;
1431 struct socket *so = sockfd_lookup(fd, &err);
1432 struct svc_sock *svsk = NULL;
1433 struct sockaddr_storage addr;
1434 struct sockaddr *sin = (struct sockaddr *)&addr;
1435 int salen;
1436
1437 if (!so)
1438 return err;
1439 err = -EAFNOSUPPORT;
1440 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1441 goto out;
1442 err = -EPROTONOSUPPORT;
1443 if (so->sk->sk_protocol != IPPROTO_TCP &&
1444 so->sk->sk_protocol != IPPROTO_UDP)
1445 goto out;
1446 err = -EISCONN;
1447 if (so->state > SS_UNCONNECTED)
1448 goto out;
1449 err = -ENOENT;
1450 if (!try_module_get(THIS_MODULE))
1451 goto out;
1452 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1453 if (IS_ERR(svsk)) {
1454 module_put(THIS_MODULE);
1455 err = PTR_ERR(svsk);
1456 goto out;
1457 }
1458 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1459 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1460 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1461 return svc_one_sock_name(svsk, name_return, len);
1462 out:
1463 sockfd_put(so);
1464 return err;
1465 }
1466 EXPORT_SYMBOL_GPL(svc_addsock);
1467
1468 /*
1469 * Create socket for RPC service.
1470 */
1471 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1472 int protocol,
1473 struct net *net,
1474 struct sockaddr *sin, int len,
1475 int flags)
1476 {
1477 struct svc_sock *svsk;
1478 struct socket *sock;
1479 int error;
1480 int type;
1481 struct sockaddr_storage addr;
1482 struct sockaddr *newsin = (struct sockaddr *)&addr;
1483 int newlen;
1484 int family;
1485 int val;
1486 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1487
1488 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1489 serv->sv_program->pg_name, protocol,
1490 __svc_print_addr(sin, buf, sizeof(buf)));
1491
1492 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1493 printk(KERN_WARNING "svc: only UDP and TCP "
1494 "sockets supported\n");
1495 return ERR_PTR(-EINVAL);
1496 }
1497
1498 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1499 switch (sin->sa_family) {
1500 case AF_INET6:
1501 family = PF_INET6;
1502 break;
1503 case AF_INET:
1504 family = PF_INET;
1505 break;
1506 default:
1507 return ERR_PTR(-EINVAL);
1508 }
1509
1510 error = __sock_create(net, family, type, protocol, &sock, 1);
1511 if (error < 0)
1512 return ERR_PTR(error);
1513
1514 svc_reclassify_socket(sock);
1515
1516 /*
1517 * If this is an PF_INET6 listener, we want to avoid
1518 * getting requests from IPv4 remotes. Those should
1519 * be shunted to a PF_INET listener via rpcbind.
1520 */
1521 val = 1;
1522 if (family == PF_INET6)
1523 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1524 (char *)&val, sizeof(val));
1525
1526 if (type == SOCK_STREAM)
1527 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1528 error = kernel_bind(sock, sin, len);
1529 if (error < 0)
1530 goto bummer;
1531
1532 newlen = len;
1533 error = kernel_getsockname(sock, newsin, &newlen);
1534 if (error < 0)
1535 goto bummer;
1536
1537 if (protocol == IPPROTO_TCP) {
1538 if ((error = kernel_listen(sock, 64)) < 0)
1539 goto bummer;
1540 }
1541
1542 svsk = svc_setup_socket(serv, sock, flags);
1543 if (IS_ERR(svsk)) {
1544 error = PTR_ERR(svsk);
1545 goto bummer;
1546 }
1547 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1548 return (struct svc_xprt *)svsk;
1549 bummer:
1550 dprintk("svc: svc_create_socket error = %d\n", -error);
1551 sock_release(sock);
1552 return ERR_PTR(error);
1553 }
1554
1555 /*
1556 * Detach the svc_sock from the socket so that no
1557 * more callbacks occur.
1558 */
1559 static void svc_sock_detach(struct svc_xprt *xprt)
1560 {
1561 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1562 struct sock *sk = svsk->sk_sk;
1563 wait_queue_head_t *wq;
1564
1565 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1566
1567 /* put back the old socket callbacks */
1568 sk->sk_state_change = svsk->sk_ostate;
1569 sk->sk_data_ready = svsk->sk_odata;
1570 sk->sk_write_space = svsk->sk_owspace;
1571
1572 wq = sk_sleep(sk);
1573 if (wq && waitqueue_active(wq))
1574 wake_up_interruptible(wq);
1575 }
1576
1577 /*
1578 * Disconnect the socket, and reset the callbacks
1579 */
1580 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1581 {
1582 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1583
1584 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1585
1586 svc_sock_detach(xprt);
1587
1588 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1589 svc_tcp_clear_pages(svsk);
1590 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1591 }
1592 }
1593
1594 /*
1595 * Free the svc_sock's socket resources and the svc_sock itself.
1596 */
1597 static void svc_sock_free(struct svc_xprt *xprt)
1598 {
1599 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1600 dprintk("svc: svc_sock_free(%p)\n", svsk);
1601
1602 if (svsk->sk_sock->file)
1603 sockfd_put(svsk->sk_sock);
1604 else
1605 sock_release(svsk->sk_sock);
1606 kfree(svsk);
1607 }
1608
1609 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1610 /*
1611 * Create a back channel svc_xprt which shares the fore channel socket.
1612 */
1613 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1614 int protocol,
1615 struct net *net,
1616 struct sockaddr *sin, int len,
1617 int flags)
1618 {
1619 struct svc_sock *svsk;
1620 struct svc_xprt *xprt;
1621
1622 if (protocol != IPPROTO_TCP) {
1623 printk(KERN_WARNING "svc: only TCP sockets"
1624 " supported on shared back channel\n");
1625 return ERR_PTR(-EINVAL);
1626 }
1627
1628 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1629 if (!svsk)
1630 return ERR_PTR(-ENOMEM);
1631
1632 xprt = &svsk->sk_xprt;
1633 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1634
1635 serv->sv_bc_xprt = xprt;
1636
1637 return xprt;
1638 }
1639
1640 /*
1641 * Free a back channel svc_sock.
1642 */
1643 static void svc_bc_sock_free(struct svc_xprt *xprt)
1644 {
1645 if (xprt)
1646 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1647 }
1648 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support