?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: Introduce gfn_to_hva_memslot_prot
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blobc24a8ff33f8fa1f5a66da0b4f2a57f775c86cb51
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 *);
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 #if IS_ENABLED(CONFIG_IPV6)
295 case PF_INET6:
296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
297 proto_name,
298 &sk->sk_v6_rcv_saddr,
299 inet_sk(sk)->inet_num);
300 break;
301 #endif
302 default:
303 len = snprintf(buf, remaining, "*unknown-%d*\n",
304 sk->sk_family);
305 }
306
307 if (len >= remaining) {
308 *buf = '\0';
309 return -ENAMETOOLONG;
310 }
311 return len;
312 }
313
314 /*
315 * Check input queue length
316 */
317 static int svc_recv_available(struct svc_sock *svsk)
318 {
319 struct socket *sock = svsk->sk_sock;
320 int avail, err;
321
322 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
323
324 return (err >= 0)? avail : err;
325 }
326
327 /*
328 * Generic recvfrom routine.
329 */
330 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
331 int buflen)
332 {
333 struct svc_sock *svsk =
334 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
335 struct msghdr msg = {
336 .msg_flags = MSG_DONTWAIT,
337 };
338 int len;
339
340 rqstp->rq_xprt_hlen = 0;
341
342 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
343 msg.msg_flags);
344
345 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
346 svsk, iov[0].iov_base, iov[0].iov_len, len);
347 return len;
348 }
349
350 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
351 struct kvec *iov, int nr,
352 int buflen, unsigned int base)
353 {
354 size_t save_iovlen;
355 void *save_iovbase;
356 unsigned int i;
357 int ret;
358
359 if (base == 0)
360 return svc_recvfrom(rqstp, iov, nr, buflen);
361
362 for (i = 0; i < nr; i++) {
363 if (iov[i].iov_len > base)
364 break;
365 base -= iov[i].iov_len;
366 }
367 save_iovlen = iov[i].iov_len;
368 save_iovbase = iov[i].iov_base;
369 iov[i].iov_len -= base;
370 iov[i].iov_base += base;
371 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
372 iov[i].iov_len = save_iovlen;
373 iov[i].iov_base = save_iovbase;
374 return ret;
375 }
376
377 /*
378 * Set socket snd and rcv buffer lengths
379 */
380 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
381 unsigned int rcv)
382 {
383 #if 0
384 mm_segment_t oldfs;
385 oldfs = get_fs(); set_fs(KERNEL_DS);
386 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
387 (char*)&snd, sizeof(snd));
388 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
389 (char*)&rcv, sizeof(rcv));
390 #else
391 /* sock_setsockopt limits use to sysctl_?mem_max,
392 * which isn't acceptable. Until that is made conditional
393 * on not having CAP_SYS_RESOURCE or similar, we go direct...
394 * DaveM said I could!
395 */
396 lock_sock(sock->sk);
397 sock->sk->sk_sndbuf = snd * 2;
398 sock->sk->sk_rcvbuf = rcv * 2;
399 sock->sk->sk_write_space(sock->sk);
400 release_sock(sock->sk);
401 #endif
402 }
403
404 static int svc_sock_secure_port(struct svc_rqst *rqstp)
405 {
406 return svc_port_is_privileged(svc_addr(rqstp));
407 }
408
409 /*
410 * INET callback when data has been received on the socket.
411 */
412 static void svc_udp_data_ready(struct sock *sk)
413 {
414 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
415 wait_queue_head_t *wq = sk_sleep(sk);
416
417 if (svsk) {
418 dprintk("svc: socket %p(inet %p), busy=%d\n",
419 svsk, sk,
420 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
421 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
422 svc_xprt_enqueue(&svsk->sk_xprt);
423 }
424 if (wq && waitqueue_active(wq))
425 wake_up_interruptible(wq);
426 }
427
428 /*
429 * INET callback when space is newly available on the socket.
430 */
431 static void svc_write_space(struct sock *sk)
432 {
433 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
434 wait_queue_head_t *wq = sk_sleep(sk);
435
436 if (svsk) {
437 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
438 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
439 svc_xprt_enqueue(&svsk->sk_xprt);
440 }
441
442 if (wq && waitqueue_active(wq)) {
443 dprintk("RPC svc_write_space: someone sleeping on %p\n",
444 svsk);
445 wake_up_interruptible(wq);
446 }
447 }
448
449 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
450 {
451 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
452 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
453 int required;
454
455 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
456 return 1;
457 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
458 if (sk_stream_wspace(svsk->sk_sk) >= required ||
459 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
460 atomic_read(&xprt->xpt_reserved) == 0))
461 return 1;
462 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
463 return 0;
464 }
465
466 static void svc_tcp_write_space(struct sock *sk)
467 {
468 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
469 struct socket *sock = sk->sk_socket;
470
471 if (!sk_stream_is_writeable(sk) || !sock)
472 return;
473 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt))
474 clear_bit(SOCK_NOSPACE, &sock->flags);
475 svc_write_space(sk);
476 }
477
478 static void svc_tcp_adjust_wspace(struct svc_xprt *xprt)
479 {
480 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
481
482 if (svc_tcp_has_wspace(xprt))
483 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
484 }
485
486 /*
487 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
488 */
489 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
490 struct cmsghdr *cmh)
491 {
492 struct in_pktinfo *pki = CMSG_DATA(cmh);
493 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
494
495 if (cmh->cmsg_type != IP_PKTINFO)
496 return 0;
497
498 daddr->sin_family = AF_INET;
499 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
500 return 1;
501 }
502
503 /*
504 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
505 */
506 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
507 struct cmsghdr *cmh)
508 {
509 struct in6_pktinfo *pki = CMSG_DATA(cmh);
510 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
511
512 if (cmh->cmsg_type != IPV6_PKTINFO)
513 return 0;
514
515 daddr->sin6_family = AF_INET6;
516 daddr->sin6_addr = pki->ipi6_addr;
517 daddr->sin6_scope_id = pki->ipi6_ifindex;
518 return 1;
519 }
520
521 /*
522 * Copy the UDP datagram's destination address to the rqstp structure.
523 * The 'destination' address in this case is the address to which the
524 * peer sent the datagram, i.e. our local address. For multihomed
525 * hosts, this can change from msg to msg. Note that only the IP
526 * address changes, the port number should remain the same.
527 */
528 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
529 struct cmsghdr *cmh)
530 {
531 switch (cmh->cmsg_level) {
532 case SOL_IP:
533 return svc_udp_get_dest_address4(rqstp, cmh);
534 case SOL_IPV6:
535 return svc_udp_get_dest_address6(rqstp, cmh);
536 }
537
538 return 0;
539 }
540
541 /*
542 * Receive a datagram from a UDP socket.
543 */
544 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
545 {
546 struct svc_sock *svsk =
547 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
548 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
549 struct sk_buff *skb;
550 union {
551 struct cmsghdr hdr;
552 long all[SVC_PKTINFO_SPACE / sizeof(long)];
553 } buffer;
554 struct cmsghdr *cmh = &buffer.hdr;
555 struct msghdr msg = {
556 .msg_name = svc_addr(rqstp),
557 .msg_control = cmh,
558 .msg_controllen = sizeof(buffer),
559 .msg_flags = MSG_DONTWAIT,
560 };
561 size_t len;
562 int err;
563
564 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
565 /* udp sockets need large rcvbuf as all pending
566 * requests are still in that buffer. sndbuf must
567 * also be large enough that there is enough space
568 * for one reply per thread. We count all threads
569 * rather than threads in a particular pool, which
570 * provides an upper bound on the number of threads
571 * which will access the socket.
572 */
573 svc_sock_setbufsize(svsk->sk_sock,
574 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
575 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
576
577 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
578 skb = NULL;
579 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
580 0, 0, MSG_PEEK | MSG_DONTWAIT);
581 if (err >= 0)
582 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
583
584 if (skb == NULL) {
585 if (err != -EAGAIN) {
586 /* possibly an icmp error */
587 dprintk("svc: recvfrom returned error %d\n", -err);
588 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
589 }
590 return 0;
591 }
592 len = svc_addr_len(svc_addr(rqstp));
593 rqstp->rq_addrlen = len;
594 if (skb->tstamp.tv64 == 0) {
595 skb->tstamp = ktime_get_real();
596 /* Don't enable netstamp, sunrpc doesn't
597 need that much accuracy */
598 }
599 svsk->sk_sk->sk_stamp = skb->tstamp;
600 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
601
602 len = skb->len - sizeof(struct udphdr);
603 rqstp->rq_arg.len = len;
604
605 rqstp->rq_prot = IPPROTO_UDP;
606
607 if (!svc_udp_get_dest_address(rqstp, cmh)) {
608 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
609 cmh->cmsg_level, cmh->cmsg_type);
610 goto out_free;
611 }
612 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
613
614 if (skb_is_nonlinear(skb)) {
615 /* we have to copy */
616 local_bh_disable();
617 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
618 local_bh_enable();
619 /* checksum error */
620 goto out_free;
621 }
622 local_bh_enable();
623 skb_free_datagram_locked(svsk->sk_sk, skb);
624 } else {
625 /* we can use it in-place */
626 rqstp->rq_arg.head[0].iov_base = skb->data +
627 sizeof(struct udphdr);
628 rqstp->rq_arg.head[0].iov_len = len;
629 if (skb_checksum_complete(skb))
630 goto out_free;
631 rqstp->rq_xprt_ctxt = skb;
632 }
633
634 rqstp->rq_arg.page_base = 0;
635 if (len <= rqstp->rq_arg.head[0].iov_len) {
636 rqstp->rq_arg.head[0].iov_len = len;
637 rqstp->rq_arg.page_len = 0;
638 rqstp->rq_respages = rqstp->rq_pages+1;
639 } else {
640 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
641 rqstp->rq_respages = rqstp->rq_pages + 1 +
642 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
643 }
644 rqstp->rq_next_page = rqstp->rq_respages+1;
645
646 if (serv->sv_stats)
647 serv->sv_stats->netudpcnt++;
648
649 return len;
650 out_free:
651 trace_kfree_skb(skb, svc_udp_recvfrom);
652 skb_free_datagram_locked(svsk->sk_sk, skb);
653 return 0;
654 }
655
656 static int
657 svc_udp_sendto(struct svc_rqst *rqstp)
658 {
659 int error;
660
661 error = svc_sendto(rqstp, &rqstp->rq_res);
662 if (error == -ECONNREFUSED)
663 /* ICMP error on earlier request. */
664 error = svc_sendto(rqstp, &rqstp->rq_res);
665
666 return error;
667 }
668
669 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
670 {
671 }
672
673 static int svc_udp_has_wspace(struct svc_xprt *xprt)
674 {
675 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
676 struct svc_serv *serv = xprt->xpt_server;
677 unsigned long required;
678
679 /*
680 * Set the SOCK_NOSPACE flag before checking the available
681 * sock space.
682 */
683 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
684 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
685 if (required*2 > sock_wspace(svsk->sk_sk))
686 return 0;
687 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
688 return 1;
689 }
690
691 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
692 {
693 BUG();
694 return NULL;
695 }
696
697 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
698 struct net *net,
699 struct sockaddr *sa, int salen,
700 int flags)
701 {
702 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
703 }
704
705 static struct svc_xprt_ops svc_udp_ops = {
706 .xpo_create = svc_udp_create,
707 .xpo_recvfrom = svc_udp_recvfrom,
708 .xpo_sendto = svc_udp_sendto,
709 .xpo_release_rqst = svc_release_skb,
710 .xpo_detach = svc_sock_detach,
711 .xpo_free = svc_sock_free,
712 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
713 .xpo_has_wspace = svc_udp_has_wspace,
714 .xpo_accept = svc_udp_accept,
715 .xpo_secure_port = svc_sock_secure_port,
716 };
717
718 static struct svc_xprt_class svc_udp_class = {
719 .xcl_name = "udp",
720 .xcl_owner = THIS_MODULE,
721 .xcl_ops = &svc_udp_ops,
722 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
723 .xcl_ident = XPRT_TRANSPORT_UDP,
724 };
725
726 static 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(sock_net(svsk->sk_sock->sk), &svc_udp_class,
731 &svsk->sk_xprt, serv);
732 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
733 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
734 svsk->sk_sk->sk_write_space = svc_write_space;
735
736 /* initialise setting must have enough space to
737 * receive and respond to one request.
738 * svc_udp_recvfrom will re-adjust if necessary
739 */
740 svc_sock_setbufsize(svsk->sk_sock,
741 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
742 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
743
744 /* data might have come in before data_ready set up */
745 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
746 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
747
748 /* make sure we get destination address info */
749 switch (svsk->sk_sk->sk_family) {
750 case AF_INET:
751 level = SOL_IP;
752 optname = IP_PKTINFO;
753 break;
754 case AF_INET6:
755 level = SOL_IPV6;
756 optname = IPV6_RECVPKTINFO;
757 break;
758 default:
759 BUG();
760 }
761 err = kernel_setsockopt(svsk->sk_sock, level, optname,
762 (char *)&one, sizeof(one));
763 dprintk("svc: kernel_setsockopt returned %d\n", err);
764 }
765
766 /*
767 * A data_ready event on a listening socket means there's a connection
768 * pending. Do not use state_change as a substitute for it.
769 */
770 static void svc_tcp_listen_data_ready(struct sock *sk)
771 {
772 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
773 wait_queue_head_t *wq;
774
775 dprintk("svc: socket %p TCP (listen) state change %d\n",
776 sk, sk->sk_state);
777
778 /*
779 * This callback may called twice when a new connection
780 * is established as a child socket inherits everything
781 * from a parent LISTEN socket.
782 * 1) data_ready method of the parent socket will be called
783 * when one of child sockets become ESTABLISHED.
784 * 2) data_ready method of the child socket may be called
785 * when it receives data before the socket is accepted.
786 * In case of 2, we should ignore it silently.
787 */
788 if (sk->sk_state == TCP_LISTEN) {
789 if (svsk) {
790 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
791 svc_xprt_enqueue(&svsk->sk_xprt);
792 } else
793 printk("svc: socket %p: no user data\n", sk);
794 }
795
796 wq = sk_sleep(sk);
797 if (wq && waitqueue_active(wq))
798 wake_up_interruptible_all(wq);
799 }
800
801 /*
802 * A state change on a connected socket means it's dying or dead.
803 */
804 static void svc_tcp_state_change(struct sock *sk)
805 {
806 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
807 wait_queue_head_t *wq = sk_sleep(sk);
808
809 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
810 sk, sk->sk_state, sk->sk_user_data);
811
812 if (!svsk)
813 printk("svc: socket %p: no user data\n", sk);
814 else {
815 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
816 svc_xprt_enqueue(&svsk->sk_xprt);
817 }
818 if (wq && waitqueue_active(wq))
819 wake_up_interruptible_all(wq);
820 }
821
822 static void svc_tcp_data_ready(struct sock *sk)
823 {
824 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
825 wait_queue_head_t *wq = sk_sleep(sk);
826
827 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
828 sk, sk->sk_user_data);
829 if (svsk) {
830 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
831 svc_xprt_enqueue(&svsk->sk_xprt);
832 }
833 if (wq && waitqueue_active(wq))
834 wake_up_interruptible(wq);
835 }
836
837 /*
838 * Accept a TCP connection
839 */
840 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
841 {
842 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
843 struct sockaddr_storage addr;
844 struct sockaddr *sin = (struct sockaddr *) &addr;
845 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
846 struct socket *sock = svsk->sk_sock;
847 struct socket *newsock;
848 struct svc_sock *newsvsk;
849 int err, slen;
850 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
851
852 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
853 if (!sock)
854 return NULL;
855
856 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
857 err = kernel_accept(sock, &newsock, O_NONBLOCK);
858 if (err < 0) {
859 if (err == -ENOMEM)
860 printk(KERN_WARNING "%s: no more sockets!\n",
861 serv->sv_name);
862 else if (err != -EAGAIN)
863 net_warn_ratelimited("%s: accept failed (err %d)!\n",
864 serv->sv_name, -err);
865 return NULL;
866 }
867 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
868
869 err = kernel_getpeername(newsock, sin, &slen);
870 if (err < 0) {
871 net_warn_ratelimited("%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("%s: connect from unprivileged port: %s\n",
882 serv->sv_name,
883 __svc_print_addr(sin, buf, sizeof(buf)));
884 }
885 dprintk("%s: connect from %s\n", serv->sv_name,
886 __svc_print_addr(sin, buf, sizeof(buf)));
887
888 /* make sure that a write doesn't block forever when
889 * low on memory
890 */
891 newsock->sk->sk_sndtimeo = HZ*30;
892
893 newsvsk = svc_setup_socket(serv, newsock,
894 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
895 if (IS_ERR(newsvsk))
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 (sock_is_loopback(newsock->sk))
906 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
907 else
908 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
909 if (serv->sv_stats)
910 serv->sv_stats->nettcpconn++;
911
912 return &newsvsk->sk_xprt;
913
914 failed:
915 sock_release(newsock);
916 return NULL;
917 }
918
919 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
920 {
921 unsigned int i, len, npages;
922
923 if (svsk->sk_datalen == 0)
924 return 0;
925 len = svsk->sk_datalen;
926 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
927 for (i = 0; i < npages; i++) {
928 if (rqstp->rq_pages[i] != NULL)
929 put_page(rqstp->rq_pages[i]);
930 BUG_ON(svsk->sk_pages[i] == NULL);
931 rqstp->rq_pages[i] = svsk->sk_pages[i];
932 svsk->sk_pages[i] = NULL;
933 }
934 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
935 return len;
936 }
937
938 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
939 {
940 unsigned int i, len, npages;
941
942 if (svsk->sk_datalen == 0)
943 return;
944 len = svsk->sk_datalen;
945 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
946 for (i = 0; i < npages; i++) {
947 svsk->sk_pages[i] = rqstp->rq_pages[i];
948 rqstp->rq_pages[i] = NULL;
949 }
950 }
951
952 static void svc_tcp_clear_pages(struct svc_sock *svsk)
953 {
954 unsigned int i, len, npages;
955
956 if (svsk->sk_datalen == 0)
957 goto out;
958 len = svsk->sk_datalen;
959 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
960 for (i = 0; i < npages; i++) {
961 if (svsk->sk_pages[i] == NULL) {
962 WARN_ON_ONCE(1);
963 continue;
964 }
965 put_page(svsk->sk_pages[i]);
966 svsk->sk_pages[i] = NULL;
967 }
968 out:
969 svsk->sk_tcplen = 0;
970 svsk->sk_datalen = 0;
971 }
972
973 /*
974 * Receive fragment record header.
975 * If we haven't gotten the record length yet, get the next four bytes.
976 */
977 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
978 {
979 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
980 unsigned int want;
981 int len;
982
983 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
984
985 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
986 struct kvec iov;
987
988 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
989 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
990 iov.iov_len = want;
991 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
992 goto error;
993 svsk->sk_tcplen += len;
994
995 if (len < want) {
996 dprintk("svc: short recvfrom while reading record "
997 "length (%d of %d)\n", len, want);
998 return -EAGAIN;
999 }
1000
1001 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
1002 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1003 serv->sv_max_mesg) {
1004 net_notice_ratelimited("RPC: fragment too large: %d\n",
1005 svc_sock_reclen(svsk));
1006 goto err_delete;
1007 }
1008 }
1009
1010 return svc_sock_reclen(svsk);
1011 error:
1012 dprintk("RPC: TCP recv_record got %d\n", len);
1013 return len;
1014 err_delete:
1015 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1016 return -EAGAIN;
1017 }
1018
1019 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1020 {
1021 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1022 struct rpc_rqst *req = NULL;
1023 struct kvec *src, *dst;
1024 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1025 __be32 xid;
1026 __be32 calldir;
1027
1028 xid = *p++;
1029 calldir = *p;
1030
1031 if (bc_xprt)
1032 req = xprt_lookup_rqst(bc_xprt, xid);
1033
1034 if (!req) {
1035 printk(KERN_NOTICE
1036 "%s: Got unrecognized reply: "
1037 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1038 __func__, ntohl(calldir),
1039 bc_xprt, xid);
1040 return -EAGAIN;
1041 }
1042
1043 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1044 /*
1045 * XXX!: cheating for now! Only copying HEAD.
1046 * But we know this is good enough for now (in fact, for any
1047 * callback reply in the forseeable future).
1048 */
1049 dst = &req->rq_private_buf.head[0];
1050 src = &rqstp->rq_arg.head[0];
1051 if (dst->iov_len < src->iov_len)
1052 return -EAGAIN; /* whatever; just giving up. */
1053 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1054 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1055 rqstp->rq_arg.len = 0;
1056 return 0;
1057 }
1058
1059 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1060 {
1061 int i = 0;
1062 int t = 0;
1063
1064 while (t < len) {
1065 vec[i].iov_base = page_address(pages[i]);
1066 vec[i].iov_len = PAGE_SIZE;
1067 i++;
1068 t += PAGE_SIZE;
1069 }
1070 return i;
1071 }
1072
1073 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1074 {
1075 /* If we have more data, signal svc_xprt_enqueue() to try again */
1076 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1077 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1078 dprintk("svc: TCP %s record (%d bytes)\n",
1079 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1080 svc_sock_reclen(svsk));
1081 svsk->sk_tcplen = 0;
1082 svsk->sk_reclen = 0;
1083 }
1084
1085 /*
1086 * Receive data from a TCP socket.
1087 */
1088 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1089 {
1090 struct svc_sock *svsk =
1091 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1092 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1093 int len;
1094 struct kvec *vec;
1095 unsigned int want, base;
1096 __be32 *p;
1097 __be32 calldir;
1098 int pnum;
1099
1100 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1101 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1102 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1103 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1104
1105 len = svc_tcp_recv_record(svsk, rqstp);
1106 if (len < 0)
1107 goto error;
1108
1109 base = svc_tcp_restore_pages(svsk, rqstp);
1110 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1111
1112 vec = rqstp->rq_vec;
1113
1114 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1115 svsk->sk_datalen + want);
1116
1117 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1118 rqstp->rq_next_page = rqstp->rq_respages + 1;
1119
1120 /* Now receive data */
1121 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1122 if (len >= 0) {
1123 svsk->sk_tcplen += len;
1124 svsk->sk_datalen += len;
1125 }
1126 if (len != want || !svc_sock_final_rec(svsk)) {
1127 svc_tcp_save_pages(svsk, rqstp);
1128 if (len < 0 && len != -EAGAIN)
1129 goto err_delete;
1130 if (len == want)
1131 svc_tcp_fragment_received(svsk);
1132 else
1133 dprintk("svc: incomplete TCP record (%d of %d)\n",
1134 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1135 svc_sock_reclen(svsk));
1136 goto err_noclose;
1137 }
1138
1139 if (svsk->sk_datalen < 8) {
1140 svsk->sk_datalen = 0;
1141 goto err_delete; /* client is nuts. */
1142 }
1143
1144 rqstp->rq_arg.len = svsk->sk_datalen;
1145 rqstp->rq_arg.page_base = 0;
1146 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1147 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1148 rqstp->rq_arg.page_len = 0;
1149 } else
1150 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1151
1152 rqstp->rq_xprt_ctxt = NULL;
1153 rqstp->rq_prot = IPPROTO_TCP;
1154 rqstp->rq_local = !!test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags);
1155
1156 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1157 calldir = p[1];
1158 if (calldir)
1159 len = receive_cb_reply(svsk, rqstp);
1160
1161 /* Reset TCP read info */
1162 svsk->sk_datalen = 0;
1163 svc_tcp_fragment_received(svsk);
1164
1165 if (len < 0)
1166 goto error;
1167
1168 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1169 if (serv->sv_stats)
1170 serv->sv_stats->nettcpcnt++;
1171
1172 return rqstp->rq_arg.len;
1173
1174 error:
1175 if (len != -EAGAIN)
1176 goto err_delete;
1177 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1178 return 0;
1179 err_delete:
1180 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1181 svsk->sk_xprt.xpt_server->sv_name, -len);
1182 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1183 err_noclose:
1184 return 0; /* record not complete */
1185 }
1186
1187 /*
1188 * Send out data on TCP socket.
1189 */
1190 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1191 {
1192 struct xdr_buf *xbufp = &rqstp->rq_res;
1193 int sent;
1194 __be32 reclen;
1195
1196 /* Set up the first element of the reply kvec.
1197 * Any other kvecs that may be in use have been taken
1198 * care of by the server implementation itself.
1199 */
1200 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1201 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1202
1203 sent = svc_sendto(rqstp, &rqstp->rq_res);
1204 if (sent != xbufp->len) {
1205 printk(KERN_NOTICE
1206 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1207 "- shutting down socket\n",
1208 rqstp->rq_xprt->xpt_server->sv_name,
1209 (sent<0)?"got error":"sent only",
1210 sent, xbufp->len);
1211 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1212 svc_xprt_enqueue(rqstp->rq_xprt);
1213 sent = -EAGAIN;
1214 }
1215 return sent;
1216 }
1217
1218 /*
1219 * Setup response header. TCP has a 4B record length field.
1220 */
1221 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1222 {
1223 struct kvec *resv = &rqstp->rq_res.head[0];
1224
1225 /* tcp needs a space for the record length... */
1226 svc_putnl(resv, 0);
1227 }
1228
1229 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1230 struct net *net,
1231 struct sockaddr *sa, int salen,
1232 int flags)
1233 {
1234 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1235 }
1236
1237 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1238 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1239 struct net *, struct sockaddr *,
1240 int, int);
1241 static void svc_bc_sock_free(struct svc_xprt *xprt);
1242
1243 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1244 struct net *net,
1245 struct sockaddr *sa, int salen,
1246 int flags)
1247 {
1248 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1249 }
1250
1251 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1252 {
1253 }
1254
1255 static struct svc_xprt_ops svc_tcp_bc_ops = {
1256 .xpo_create = svc_bc_tcp_create,
1257 .xpo_detach = svc_bc_tcp_sock_detach,
1258 .xpo_free = svc_bc_sock_free,
1259 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1260 .xpo_secure_port = svc_sock_secure_port,
1261 };
1262
1263 static struct svc_xprt_class svc_tcp_bc_class = {
1264 .xcl_name = "tcp-bc",
1265 .xcl_owner = THIS_MODULE,
1266 .xcl_ops = &svc_tcp_bc_ops,
1267 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1268 };
1269
1270 static void svc_init_bc_xprt_sock(void)
1271 {
1272 svc_reg_xprt_class(&svc_tcp_bc_class);
1273 }
1274
1275 static void svc_cleanup_bc_xprt_sock(void)
1276 {
1277 svc_unreg_xprt_class(&svc_tcp_bc_class);
1278 }
1279 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1280 static void svc_init_bc_xprt_sock(void)
1281 {
1282 }
1283
1284 static void svc_cleanup_bc_xprt_sock(void)
1285 {
1286 }
1287 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1288
1289 static struct svc_xprt_ops svc_tcp_ops = {
1290 .xpo_create = svc_tcp_create,
1291 .xpo_recvfrom = svc_tcp_recvfrom,
1292 .xpo_sendto = svc_tcp_sendto,
1293 .xpo_release_rqst = svc_release_skb,
1294 .xpo_detach = svc_tcp_sock_detach,
1295 .xpo_free = svc_sock_free,
1296 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1297 .xpo_has_wspace = svc_tcp_has_wspace,
1298 .xpo_accept = svc_tcp_accept,
1299 .xpo_secure_port = svc_sock_secure_port,
1300 .xpo_adjust_wspace = svc_tcp_adjust_wspace,
1301 };
1302
1303 static struct svc_xprt_class svc_tcp_class = {
1304 .xcl_name = "tcp",
1305 .xcl_owner = THIS_MODULE,
1306 .xcl_ops = &svc_tcp_ops,
1307 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1308 .xcl_ident = XPRT_TRANSPORT_TCP,
1309 };
1310
1311 void svc_init_xprt_sock(void)
1312 {
1313 svc_reg_xprt_class(&svc_tcp_class);
1314 svc_reg_xprt_class(&svc_udp_class);
1315 svc_init_bc_xprt_sock();
1316 }
1317
1318 void svc_cleanup_xprt_sock(void)
1319 {
1320 svc_unreg_xprt_class(&svc_tcp_class);
1321 svc_unreg_xprt_class(&svc_udp_class);
1322 svc_cleanup_bc_xprt_sock();
1323 }
1324
1325 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1326 {
1327 struct sock *sk = svsk->sk_sk;
1328
1329 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1330 &svsk->sk_xprt, serv);
1331 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1332 if (sk->sk_state == TCP_LISTEN) {
1333 dprintk("setting up TCP socket for listening\n");
1334 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1335 sk->sk_data_ready = svc_tcp_listen_data_ready;
1336 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1337 } else {
1338 dprintk("setting up TCP socket for reading\n");
1339 sk->sk_state_change = svc_tcp_state_change;
1340 sk->sk_data_ready = svc_tcp_data_ready;
1341 sk->sk_write_space = svc_tcp_write_space;
1342
1343 svsk->sk_reclen = 0;
1344 svsk->sk_tcplen = 0;
1345 svsk->sk_datalen = 0;
1346 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1347
1348 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1349
1350 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1351 if (sk->sk_state != TCP_ESTABLISHED)
1352 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1353 }
1354 }
1355
1356 void svc_sock_update_bufs(struct svc_serv *serv)
1357 {
1358 /*
1359 * The number of server threads has changed. Update
1360 * rcvbuf and sndbuf accordingly on all sockets
1361 */
1362 struct svc_sock *svsk;
1363
1364 spin_lock_bh(&serv->sv_lock);
1365 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1366 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1367 spin_unlock_bh(&serv->sv_lock);
1368 }
1369 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1370
1371 /*
1372 * Initialize socket for RPC use and create svc_sock struct
1373 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1374 */
1375 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1376 struct socket *sock,
1377 int flags)
1378 {
1379 struct svc_sock *svsk;
1380 struct sock *inet;
1381 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1382 int err = 0;
1383
1384 dprintk("svc: svc_setup_socket %p\n", sock);
1385 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1386 if (!svsk)
1387 return ERR_PTR(-ENOMEM);
1388
1389 inet = sock->sk;
1390
1391 /* Register socket with portmapper */
1392 if (pmap_register)
1393 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1394 inet->sk_protocol,
1395 ntohs(inet_sk(inet)->inet_sport));
1396
1397 if (err < 0) {
1398 kfree(svsk);
1399 return ERR_PTR(err);
1400 }
1401
1402 inet->sk_user_data = svsk;
1403 svsk->sk_sock = sock;
1404 svsk->sk_sk = inet;
1405 svsk->sk_ostate = inet->sk_state_change;
1406 svsk->sk_odata = inet->sk_data_ready;
1407 svsk->sk_owspace = inet->sk_write_space;
1408
1409 /* Initialize the socket */
1410 if (sock->type == SOCK_DGRAM)
1411 svc_udp_init(svsk, serv);
1412 else {
1413 /* initialise setting must have enough space to
1414 * receive and respond to one request.
1415 */
1416 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1417 4 * serv->sv_max_mesg);
1418 svc_tcp_init(svsk, serv);
1419 }
1420
1421 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1422 svsk, svsk->sk_sk);
1423
1424 return svsk;
1425 }
1426
1427 bool svc_alien_sock(struct net *net, int fd)
1428 {
1429 int err;
1430 struct socket *sock = sockfd_lookup(fd, &err);
1431 bool ret = false;
1432
1433 if (!sock)
1434 goto out;
1435 if (sock_net(sock->sk) != net)
1436 ret = true;
1437 sockfd_put(sock);
1438 out:
1439 return ret;
1440 }
1441 EXPORT_SYMBOL_GPL(svc_alien_sock);
1442
1443 /**
1444 * svc_addsock - add a listener socket to an RPC service
1445 * @serv: pointer to RPC service to which to add a new listener
1446 * @fd: file descriptor of the new listener
1447 * @name_return: pointer to buffer to fill in with name of listener
1448 * @len: size of the buffer
1449 *
1450 * Fills in socket name and returns positive length of name if successful.
1451 * Name is terminated with '\n'. On error, returns a negative errno
1452 * value.
1453 */
1454 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1455 const size_t len)
1456 {
1457 int err = 0;
1458 struct socket *so = sockfd_lookup(fd, &err);
1459 struct svc_sock *svsk = NULL;
1460 struct sockaddr_storage addr;
1461 struct sockaddr *sin = (struct sockaddr *)&addr;
1462 int salen;
1463
1464 if (!so)
1465 return err;
1466 err = -EAFNOSUPPORT;
1467 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1468 goto out;
1469 err = -EPROTONOSUPPORT;
1470 if (so->sk->sk_protocol != IPPROTO_TCP &&
1471 so->sk->sk_protocol != IPPROTO_UDP)
1472 goto out;
1473 err = -EISCONN;
1474 if (so->state > SS_UNCONNECTED)
1475 goto out;
1476 err = -ENOENT;
1477 if (!try_module_get(THIS_MODULE))
1478 goto out;
1479 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1480 if (IS_ERR(svsk)) {
1481 module_put(THIS_MODULE);
1482 err = PTR_ERR(svsk);
1483 goto out;
1484 }
1485 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1486 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1487 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1488 return svc_one_sock_name(svsk, name_return, len);
1489 out:
1490 sockfd_put(so);
1491 return err;
1492 }
1493 EXPORT_SYMBOL_GPL(svc_addsock);
1494
1495 /*
1496 * Create socket for RPC service.
1497 */
1498 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1499 int protocol,
1500 struct net *net,
1501 struct sockaddr *sin, int len,
1502 int flags)
1503 {
1504 struct svc_sock *svsk;
1505 struct socket *sock;
1506 int error;
1507 int type;
1508 struct sockaddr_storage addr;
1509 struct sockaddr *newsin = (struct sockaddr *)&addr;
1510 int newlen;
1511 int family;
1512 int val;
1513 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1514
1515 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1516 serv->sv_program->pg_name, protocol,
1517 __svc_print_addr(sin, buf, sizeof(buf)));
1518
1519 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1520 printk(KERN_WARNING "svc: only UDP and TCP "
1521 "sockets supported\n");
1522 return ERR_PTR(-EINVAL);
1523 }
1524
1525 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1526 switch (sin->sa_family) {
1527 case AF_INET6:
1528 family = PF_INET6;
1529 break;
1530 case AF_INET:
1531 family = PF_INET;
1532 break;
1533 default:
1534 return ERR_PTR(-EINVAL);
1535 }
1536
1537 error = __sock_create(net, family, type, protocol, &sock, 1);
1538 if (error < 0)
1539 return ERR_PTR(error);
1540
1541 svc_reclassify_socket(sock);
1542
1543 /*
1544 * If this is an PF_INET6 listener, we want to avoid
1545 * getting requests from IPv4 remotes. Those should
1546 * be shunted to a PF_INET listener via rpcbind.
1547 */
1548 val = 1;
1549 if (family == PF_INET6)
1550 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1551 (char *)&val, sizeof(val));
1552
1553 if (type == SOCK_STREAM)
1554 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1555 error = kernel_bind(sock, sin, len);
1556 if (error < 0)
1557 goto bummer;
1558
1559 newlen = len;
1560 error = kernel_getsockname(sock, newsin, &newlen);
1561 if (error < 0)
1562 goto bummer;
1563
1564 if (protocol == IPPROTO_TCP) {
1565 if ((error = kernel_listen(sock, 64)) < 0)
1566 goto bummer;
1567 }
1568
1569 svsk = svc_setup_socket(serv, sock, flags);
1570 if (IS_ERR(svsk)) {
1571 error = PTR_ERR(svsk);
1572 goto bummer;
1573 }
1574 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1575 return (struct svc_xprt *)svsk;
1576 bummer:
1577 dprintk("svc: svc_create_socket error = %d\n", -error);
1578 sock_release(sock);
1579 return ERR_PTR(error);
1580 }
1581
1582 /*
1583 * Detach the svc_sock from the socket so that no
1584 * more callbacks occur.
1585 */
1586 static void svc_sock_detach(struct svc_xprt *xprt)
1587 {
1588 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1589 struct sock *sk = svsk->sk_sk;
1590 wait_queue_head_t *wq;
1591
1592 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1593
1594 /* put back the old socket callbacks */
1595 sk->sk_state_change = svsk->sk_ostate;
1596 sk->sk_data_ready = svsk->sk_odata;
1597 sk->sk_write_space = svsk->sk_owspace;
1598
1599 wq = sk_sleep(sk);
1600 if (wq && waitqueue_active(wq))
1601 wake_up_interruptible(wq);
1602 }
1603
1604 /*
1605 * Disconnect the socket, and reset the callbacks
1606 */
1607 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1608 {
1609 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1610
1611 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1612
1613 svc_sock_detach(xprt);
1614
1615 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1616 svc_tcp_clear_pages(svsk);
1617 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1618 }
1619 }
1620
1621 /*
1622 * Free the svc_sock's socket resources and the svc_sock itself.
1623 */
1624 static void svc_sock_free(struct svc_xprt *xprt)
1625 {
1626 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1627 dprintk("svc: svc_sock_free(%p)\n", svsk);
1628
1629 if (svsk->sk_sock->file)
1630 sockfd_put(svsk->sk_sock);
1631 else
1632 sock_release(svsk->sk_sock);
1633 kfree(svsk);
1634 }
1635
1636 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1637 /*
1638 * Create a back channel svc_xprt which shares the fore channel socket.
1639 */
1640 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1641 int protocol,
1642 struct net *net,
1643 struct sockaddr *sin, int len,
1644 int flags)
1645 {
1646 struct svc_sock *svsk;
1647 struct svc_xprt *xprt;
1648
1649 if (protocol != IPPROTO_TCP) {
1650 printk(KERN_WARNING "svc: only TCP sockets"
1651 " supported on shared back channel\n");
1652 return ERR_PTR(-EINVAL);
1653 }
1654
1655 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1656 if (!svsk)
1657 return ERR_PTR(-ENOMEM);
1658
1659 xprt = &svsk->sk_xprt;
1660 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1661
1662 serv->sv_bc_xprt = xprt;
1663
1664 return xprt;
1665 }
1666
1667 /*
1668 * Free a back channel svc_sock.
1669 */
1670 static void svc_bc_sock_free(struct svc_xprt *xprt)
1671 {
1672 if (xprt)
1673 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1674 }
1675 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support