?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
perf bpf: Move perf_event_output() from stdio.h to bpf.h
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blob986f3ed7d1a24800d31713143aebffc45e32fc16
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/udp.h>
43 #include <net/tcp.h>
44 #include <net/tcp_states.h>
45 #include <linux/uaccess.h>
46 #include <asm/ioctls.h>
47 #include <trace/events/skb.h>
48
49 #include <linux/sunrpc/types.h>
50 #include <linux/sunrpc/clnt.h>
51 #include <linux/sunrpc/xdr.h>
52 #include <linux/sunrpc/msg_prot.h>
53 #include <linux/sunrpc/svcsock.h>
54 #include <linux/sunrpc/stats.h>
55 #include <linux/sunrpc/xprt.h>
56
57 #include "sunrpc.h"
58
59 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
60
61
62 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
63 int flags);
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 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
89 return;
90
91 switch (sk->sk_family) {
92 case AF_INET:
93 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
94 &svc_slock_key[0],
95 "sk_xprt.xpt_lock-AF_INET-NFSD",
96 &svc_key[0]);
97 break;
98
99 case AF_INET6:
100 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
101 &svc_slock_key[1],
102 "sk_xprt.xpt_lock-AF_INET6-NFSD",
103 &svc_key[1]);
104 break;
105
106 default:
107 BUG();
108 }
109 }
110 #else
111 static void svc_reclassify_socket(struct socket *sock)
112 {
113 }
114 #endif
115
116 /*
117 * Release an skbuff after use
118 */
119 static void svc_release_skb(struct svc_rqst *rqstp)
120 {
121 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
122
123 if (skb) {
124 struct svc_sock *svsk =
125 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
126 rqstp->rq_xprt_ctxt = NULL;
127
128 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
129 skb_free_datagram_locked(svsk->sk_sk, skb);
130 }
131 }
132
133 static void svc_release_udp_skb(struct svc_rqst *rqstp)
134 {
135 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
136
137 if (skb) {
138 rqstp->rq_xprt_ctxt = NULL;
139
140 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
141 consume_skb(skb);
142 }
143 }
144
145 union svc_pktinfo_u {
146 struct in_pktinfo pkti;
147 struct in6_pktinfo pkti6;
148 };
149 #define SVC_PKTINFO_SPACE \
150 CMSG_SPACE(sizeof(union svc_pktinfo_u))
151
152 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
153 {
154 struct svc_sock *svsk =
155 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
156 switch (svsk->sk_sk->sk_family) {
157 case AF_INET: {
158 struct in_pktinfo *pki = CMSG_DATA(cmh);
159
160 cmh->cmsg_level = SOL_IP;
161 cmh->cmsg_type = IP_PKTINFO;
162 pki->ipi_ifindex = 0;
163 pki->ipi_spec_dst.s_addr =
164 svc_daddr_in(rqstp)->sin_addr.s_addr;
165 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
166 }
167 break;
168
169 case AF_INET6: {
170 struct in6_pktinfo *pki = CMSG_DATA(cmh);
171 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
172
173 cmh->cmsg_level = SOL_IPV6;
174 cmh->cmsg_type = IPV6_PKTINFO;
175 pki->ipi6_ifindex = daddr->sin6_scope_id;
176 pki->ipi6_addr = daddr->sin6_addr;
177 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
178 }
179 break;
180 }
181 }
182
183 /*
184 * send routine intended to be shared by the fore- and back-channel
185 */
186 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
187 struct page *headpage, unsigned long headoffset,
188 struct page *tailpage, unsigned long tailoffset)
189 {
190 int result;
191 int size;
192 struct page **ppage = xdr->pages;
193 size_t base = xdr->page_base;
194 unsigned int pglen = xdr->page_len;
195 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
196 int slen;
197 int len = 0;
198
199 slen = xdr->len;
200
201 /* send head */
202 if (slen == xdr->head[0].iov_len)
203 flags = 0;
204 len = kernel_sendpage(sock, headpage, headoffset,
205 xdr->head[0].iov_len, flags);
206 if (len != xdr->head[0].iov_len)
207 goto out;
208 slen -= xdr->head[0].iov_len;
209 if (slen == 0)
210 goto out;
211
212 /* send page data */
213 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
214 while (pglen > 0) {
215 if (slen == size)
216 flags = 0;
217 result = kernel_sendpage(sock, *ppage, base, size, flags);
218 if (result > 0)
219 len += result;
220 if (result != size)
221 goto out;
222 slen -= size;
223 pglen -= size;
224 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
225 base = 0;
226 ppage++;
227 }
228
229 /* send tail */
230 if (xdr->tail[0].iov_len) {
231 result = kernel_sendpage(sock, tailpage, tailoffset,
232 xdr->tail[0].iov_len, 0);
233 if (result > 0)
234 len += result;
235 }
236
237 out:
238 return len;
239 }
240
241
242 /*
243 * Generic sendto routine
244 */
245 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
246 {
247 struct svc_sock *svsk =
248 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
249 struct socket *sock = svsk->sk_sock;
250 union {
251 struct cmsghdr hdr;
252 long all[SVC_PKTINFO_SPACE / sizeof(long)];
253 } buffer;
254 struct cmsghdr *cmh = &buffer.hdr;
255 int len = 0;
256 unsigned long tailoff;
257 unsigned long headoff;
258 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
259
260 if (rqstp->rq_prot == IPPROTO_UDP) {
261 struct msghdr msg = {
262 .msg_name = &rqstp->rq_addr,
263 .msg_namelen = rqstp->rq_addrlen,
264 .msg_control = cmh,
265 .msg_controllen = sizeof(buffer),
266 .msg_flags = MSG_MORE,
267 };
268
269 svc_set_cmsg_data(rqstp, cmh);
270
271 if (sock_sendmsg(sock, &msg) < 0)
272 goto out;
273 }
274
275 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
276 headoff = 0;
277 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
278 rqstp->rq_respages[0], tailoff);
279
280 out:
281 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
282 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
283 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
284
285 return len;
286 }
287
288 /*
289 * Report socket names for nfsdfs
290 */
291 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
292 {
293 const struct sock *sk = svsk->sk_sk;
294 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
295 "udp" : "tcp";
296 int len;
297
298 switch (sk->sk_family) {
299 case PF_INET:
300 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
301 proto_name,
302 &inet_sk(sk)->inet_rcv_saddr,
303 inet_sk(sk)->inet_num);
304 break;
305 #if IS_ENABLED(CONFIG_IPV6)
306 case PF_INET6:
307 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
308 proto_name,
309 &sk->sk_v6_rcv_saddr,
310 inet_sk(sk)->inet_num);
311 break;
312 #endif
313 default:
314 len = snprintf(buf, remaining, "*unknown-%d*\n",
315 sk->sk_family);
316 }
317
318 if (len >= remaining) {
319 *buf = '\0';
320 return -ENAMETOOLONG;
321 }
322 return len;
323 }
324
325 /*
326 * Generic recvfrom routine.
327 */
328 static ssize_t svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov,
329 unsigned int nr, size_t buflen, unsigned int base)
330 {
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = { NULL };
334 ssize_t len;
335
336 rqstp->rq_xprt_hlen = 0;
337
338 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
339 iov_iter_kvec(&msg.msg_iter, READ, iov, nr, buflen);
340 if (base != 0) {
341 iov_iter_advance(&msg.msg_iter, base);
342 buflen -= base;
343 }
344 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
345 /* If we read a full record, then assume there may be more
346 * data to read (stream based sockets only!)
347 */
348 if (len == buflen)
349 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
350
351 dprintk("svc: socket %p recvfrom(%p, %zu) = %zd\n",
352 svsk, iov[0].iov_base, iov[0].iov_len, len);
353 return len;
354 }
355
356 /*
357 * Set socket snd and rcv buffer lengths
358 */
359 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
360 unsigned int rcv)
361 {
362 lock_sock(sock->sk);
363 sock->sk->sk_sndbuf = snd * 2;
364 sock->sk->sk_rcvbuf = rcv * 2;
365 sock->sk->sk_write_space(sock->sk);
366 release_sock(sock->sk);
367 }
368
369 static void svc_sock_secure_port(struct svc_rqst *rqstp)
370 {
371 if (svc_port_is_privileged(svc_addr(rqstp)))
372 set_bit(RQ_SECURE, &rqstp->rq_flags);
373 else
374 clear_bit(RQ_SECURE, &rqstp->rq_flags);
375 }
376
377 /*
378 * INET callback when data has been received on the socket.
379 */
380 static void svc_data_ready(struct sock *sk)
381 {
382 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
383
384 if (svsk) {
385 dprintk("svc: socket %p(inet %p), busy=%d\n",
386 svsk, sk,
387 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
388
389 /* Refer to svc_setup_socket() for details. */
390 rmb();
391 svsk->sk_odata(sk);
392 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
393 svc_xprt_enqueue(&svsk->sk_xprt);
394 }
395 }
396
397 /*
398 * INET callback when space is newly available on the socket.
399 */
400 static void svc_write_space(struct sock *sk)
401 {
402 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
403
404 if (svsk) {
405 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
406 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
407
408 /* Refer to svc_setup_socket() for details. */
409 rmb();
410 svsk->sk_owspace(sk);
411 svc_xprt_enqueue(&svsk->sk_xprt);
412 }
413 }
414
415 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
416 {
417 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
418
419 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
420 return 1;
421 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
422 }
423
424 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
425 {
426 struct svc_sock *svsk;
427 struct socket *sock;
428 struct linger no_linger = {
429 .l_onoff = 1,
430 .l_linger = 0,
431 };
432
433 svsk = container_of(xprt, struct svc_sock, sk_xprt);
434 sock = svsk->sk_sock;
435 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
436 (char *)&no_linger, sizeof(no_linger));
437 }
438
439 /*
440 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
441 */
442 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
443 struct cmsghdr *cmh)
444 {
445 struct in_pktinfo *pki = CMSG_DATA(cmh);
446 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
447
448 if (cmh->cmsg_type != IP_PKTINFO)
449 return 0;
450
451 daddr->sin_family = AF_INET;
452 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
453 return 1;
454 }
455
456 /*
457 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
458 */
459 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
460 struct cmsghdr *cmh)
461 {
462 struct in6_pktinfo *pki = CMSG_DATA(cmh);
463 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
464
465 if (cmh->cmsg_type != IPV6_PKTINFO)
466 return 0;
467
468 daddr->sin6_family = AF_INET6;
469 daddr->sin6_addr = pki->ipi6_addr;
470 daddr->sin6_scope_id = pki->ipi6_ifindex;
471 return 1;
472 }
473
474 /*
475 * Copy the UDP datagram's destination address to the rqstp structure.
476 * The 'destination' address in this case is the address to which the
477 * peer sent the datagram, i.e. our local address. For multihomed
478 * hosts, this can change from msg to msg. Note that only the IP
479 * address changes, the port number should remain the same.
480 */
481 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
482 struct cmsghdr *cmh)
483 {
484 switch (cmh->cmsg_level) {
485 case SOL_IP:
486 return svc_udp_get_dest_address4(rqstp, cmh);
487 case SOL_IPV6:
488 return svc_udp_get_dest_address6(rqstp, cmh);
489 }
490
491 return 0;
492 }
493
494 /*
495 * Receive a datagram from a UDP socket.
496 */
497 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
498 {
499 struct svc_sock *svsk =
500 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
501 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
502 struct sk_buff *skb;
503 union {
504 struct cmsghdr hdr;
505 long all[SVC_PKTINFO_SPACE / sizeof(long)];
506 } buffer;
507 struct cmsghdr *cmh = &buffer.hdr;
508 struct msghdr msg = {
509 .msg_name = svc_addr(rqstp),
510 .msg_control = cmh,
511 .msg_controllen = sizeof(buffer),
512 .msg_flags = MSG_DONTWAIT,
513 };
514 size_t len;
515 int err;
516
517 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
518 /* udp sockets need large rcvbuf as all pending
519 * requests are still in that buffer. sndbuf must
520 * also be large enough that there is enough space
521 * for one reply per thread. We count all threads
522 * rather than threads in a particular pool, which
523 * provides an upper bound on the number of threads
524 * which will access the socket.
525 */
526 svc_sock_setbufsize(svsk->sk_sock,
527 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
528 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
529
530 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
531 skb = NULL;
532 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
533 0, 0, MSG_PEEK | MSG_DONTWAIT);
534 if (err >= 0)
535 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
536
537 if (skb == NULL) {
538 if (err != -EAGAIN) {
539 /* possibly an icmp error */
540 dprintk("svc: recvfrom returned error %d\n", -err);
541 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
542 }
543 return 0;
544 }
545 len = svc_addr_len(svc_addr(rqstp));
546 rqstp->rq_addrlen = len;
547 if (skb->tstamp == 0) {
548 skb->tstamp = ktime_get_real();
549 /* Don't enable netstamp, sunrpc doesn't
550 need that much accuracy */
551 }
552 svsk->sk_sk->sk_stamp = skb->tstamp;
553 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
554
555 len = skb->len;
556 rqstp->rq_arg.len = len;
557
558 rqstp->rq_prot = IPPROTO_UDP;
559
560 if (!svc_udp_get_dest_address(rqstp, cmh)) {
561 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
562 cmh->cmsg_level, cmh->cmsg_type);
563 goto out_free;
564 }
565 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
566
567 if (skb_is_nonlinear(skb)) {
568 /* we have to copy */
569 local_bh_disable();
570 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
571 local_bh_enable();
572 /* checksum error */
573 goto out_free;
574 }
575 local_bh_enable();
576 consume_skb(skb);
577 } else {
578 /* we can use it in-place */
579 rqstp->rq_arg.head[0].iov_base = skb->data;
580 rqstp->rq_arg.head[0].iov_len = len;
581 if (skb_checksum_complete(skb))
582 goto out_free;
583 rqstp->rq_xprt_ctxt = skb;
584 }
585
586 rqstp->rq_arg.page_base = 0;
587 if (len <= rqstp->rq_arg.head[0].iov_len) {
588 rqstp->rq_arg.head[0].iov_len = len;
589 rqstp->rq_arg.page_len = 0;
590 rqstp->rq_respages = rqstp->rq_pages+1;
591 } else {
592 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
593 rqstp->rq_respages = rqstp->rq_pages + 1 +
594 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
595 }
596 rqstp->rq_next_page = rqstp->rq_respages+1;
597
598 if (serv->sv_stats)
599 serv->sv_stats->netudpcnt++;
600
601 return len;
602 out_free:
603 kfree_skb(skb);
604 return 0;
605 }
606
607 static int
608 svc_udp_sendto(struct svc_rqst *rqstp)
609 {
610 int error;
611
612 error = svc_sendto(rqstp, &rqstp->rq_res);
613 if (error == -ECONNREFUSED)
614 /* ICMP error on earlier request. */
615 error = svc_sendto(rqstp, &rqstp->rq_res);
616
617 return error;
618 }
619
620 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
621 {
622 }
623
624 static int svc_udp_has_wspace(struct svc_xprt *xprt)
625 {
626 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
627 struct svc_serv *serv = xprt->xpt_server;
628 unsigned long required;
629
630 /*
631 * Set the SOCK_NOSPACE flag before checking the available
632 * sock space.
633 */
634 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
635 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
636 if (required*2 > sock_wspace(svsk->sk_sk))
637 return 0;
638 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
639 return 1;
640 }
641
642 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
643 {
644 BUG();
645 return NULL;
646 }
647
648 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
649 {
650 }
651
652 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
653 struct net *net,
654 struct sockaddr *sa, int salen,
655 int flags)
656 {
657 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
658 }
659
660 static const struct svc_xprt_ops svc_udp_ops = {
661 .xpo_create = svc_udp_create,
662 .xpo_recvfrom = svc_udp_recvfrom,
663 .xpo_sendto = svc_udp_sendto,
664 .xpo_release_rqst = svc_release_udp_skb,
665 .xpo_detach = svc_sock_detach,
666 .xpo_free = svc_sock_free,
667 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
668 .xpo_has_wspace = svc_udp_has_wspace,
669 .xpo_accept = svc_udp_accept,
670 .xpo_secure_port = svc_sock_secure_port,
671 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
672 };
673
674 static struct svc_xprt_class svc_udp_class = {
675 .xcl_name = "udp",
676 .xcl_owner = THIS_MODULE,
677 .xcl_ops = &svc_udp_ops,
678 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
679 .xcl_ident = XPRT_TRANSPORT_UDP,
680 };
681
682 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
683 {
684 int err, level, optname, one = 1;
685
686 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
687 &svsk->sk_xprt, serv);
688 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
689 svsk->sk_sk->sk_data_ready = svc_data_ready;
690 svsk->sk_sk->sk_write_space = svc_write_space;
691
692 /* initialise setting must have enough space to
693 * receive and respond to one request.
694 * svc_udp_recvfrom will re-adjust if necessary
695 */
696 svc_sock_setbufsize(svsk->sk_sock,
697 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
698 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
699
700 /* data might have come in before data_ready set up */
701 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
702 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
703
704 /* make sure we get destination address info */
705 switch (svsk->sk_sk->sk_family) {
706 case AF_INET:
707 level = SOL_IP;
708 optname = IP_PKTINFO;
709 break;
710 case AF_INET6:
711 level = SOL_IPV6;
712 optname = IPV6_RECVPKTINFO;
713 break;
714 default:
715 BUG();
716 }
717 err = kernel_setsockopt(svsk->sk_sock, level, optname,
718 (char *)&one, sizeof(one));
719 dprintk("svc: kernel_setsockopt returned %d\n", err);
720 }
721
722 /*
723 * A data_ready event on a listening socket means there's a connection
724 * pending. Do not use state_change as a substitute for it.
725 */
726 static void svc_tcp_listen_data_ready(struct sock *sk)
727 {
728 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
729
730 dprintk("svc: socket %p TCP (listen) state change %d\n",
731 sk, sk->sk_state);
732
733 if (svsk) {
734 /* Refer to svc_setup_socket() for details. */
735 rmb();
736 svsk->sk_odata(sk);
737 }
738
739 /*
740 * This callback may called twice when a new connection
741 * is established as a child socket inherits everything
742 * from a parent LISTEN socket.
743 * 1) data_ready method of the parent socket will be called
744 * when one of child sockets become ESTABLISHED.
745 * 2) data_ready method of the child socket may be called
746 * when it receives data before the socket is accepted.
747 * In case of 2, we should ignore it silently.
748 */
749 if (sk->sk_state == TCP_LISTEN) {
750 if (svsk) {
751 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
752 svc_xprt_enqueue(&svsk->sk_xprt);
753 } else
754 printk("svc: socket %p: no user data\n", sk);
755 }
756 }
757
758 /*
759 * A state change on a connected socket means it's dying or dead.
760 */
761 static void svc_tcp_state_change(struct sock *sk)
762 {
763 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
764
765 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
766 sk, sk->sk_state, sk->sk_user_data);
767
768 if (!svsk)
769 printk("svc: socket %p: no user data\n", sk);
770 else {
771 /* Refer to svc_setup_socket() for details. */
772 rmb();
773 svsk->sk_ostate(sk);
774 if (sk->sk_state != TCP_ESTABLISHED) {
775 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
776 svc_xprt_enqueue(&svsk->sk_xprt);
777 }
778 }
779 }
780
781 /*
782 * Accept a TCP connection
783 */
784 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
785 {
786 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
787 struct sockaddr_storage addr;
788 struct sockaddr *sin = (struct sockaddr *) &addr;
789 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
790 struct socket *sock = svsk->sk_sock;
791 struct socket *newsock;
792 struct svc_sock *newsvsk;
793 int err, slen;
794 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
795
796 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
797 if (!sock)
798 return NULL;
799
800 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
801 err = kernel_accept(sock, &newsock, O_NONBLOCK);
802 if (err < 0) {
803 if (err == -ENOMEM)
804 printk(KERN_WARNING "%s: no more sockets!\n",
805 serv->sv_name);
806 else if (err != -EAGAIN)
807 net_warn_ratelimited("%s: accept failed (err %d)!\n",
808 serv->sv_name, -err);
809 return NULL;
810 }
811 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
812
813 err = kernel_getpeername(newsock, sin);
814 if (err < 0) {
815 net_warn_ratelimited("%s: peername failed (err %d)!\n",
816 serv->sv_name, -err);
817 goto failed; /* aborted connection or whatever */
818 }
819 slen = err;
820
821 /* Ideally, we would want to reject connections from unauthorized
822 * hosts here, but when we get encryption, the IP of the host won't
823 * tell us anything. For now just warn about unpriv connections.
824 */
825 if (!svc_port_is_privileged(sin)) {
826 dprintk("%s: connect from unprivileged port: %s\n",
827 serv->sv_name,
828 __svc_print_addr(sin, buf, sizeof(buf)));
829 }
830 dprintk("%s: connect from %s\n", serv->sv_name,
831 __svc_print_addr(sin, buf, sizeof(buf)));
832
833 /* Reset the inherited callbacks before calling svc_setup_socket */
834 newsock->sk->sk_state_change = svsk->sk_ostate;
835 newsock->sk->sk_data_ready = svsk->sk_odata;
836 newsock->sk->sk_write_space = svsk->sk_owspace;
837
838 /* make sure that a write doesn't block forever when
839 * low on memory
840 */
841 newsock->sk->sk_sndtimeo = HZ*30;
842
843 newsvsk = svc_setup_socket(serv, newsock,
844 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
845 if (IS_ERR(newsvsk))
846 goto failed;
847 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
848 err = kernel_getsockname(newsock, sin);
849 slen = err;
850 if (unlikely(err < 0)) {
851 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
852 slen = offsetof(struct sockaddr, sa_data);
853 }
854 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
855
856 if (sock_is_loopback(newsock->sk))
857 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
858 else
859 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
860 if (serv->sv_stats)
861 serv->sv_stats->nettcpconn++;
862
863 return &newsvsk->sk_xprt;
864
865 failed:
866 sock_release(newsock);
867 return NULL;
868 }
869
870 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
871 {
872 unsigned int i, len, npages;
873
874 if (svsk->sk_datalen == 0)
875 return 0;
876 len = svsk->sk_datalen;
877 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
878 for (i = 0; i < npages; i++) {
879 if (rqstp->rq_pages[i] != NULL)
880 put_page(rqstp->rq_pages[i]);
881 BUG_ON(svsk->sk_pages[i] == NULL);
882 rqstp->rq_pages[i] = svsk->sk_pages[i];
883 svsk->sk_pages[i] = NULL;
884 }
885 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
886 return len;
887 }
888
889 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
890 {
891 unsigned int i, len, npages;
892
893 if (svsk->sk_datalen == 0)
894 return;
895 len = svsk->sk_datalen;
896 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
897 for (i = 0; i < npages; i++) {
898 svsk->sk_pages[i] = rqstp->rq_pages[i];
899 rqstp->rq_pages[i] = NULL;
900 }
901 }
902
903 static void svc_tcp_clear_pages(struct svc_sock *svsk)
904 {
905 unsigned int i, len, npages;
906
907 if (svsk->sk_datalen == 0)
908 goto out;
909 len = svsk->sk_datalen;
910 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
911 for (i = 0; i < npages; i++) {
912 if (svsk->sk_pages[i] == NULL) {
913 WARN_ON_ONCE(1);
914 continue;
915 }
916 put_page(svsk->sk_pages[i]);
917 svsk->sk_pages[i] = NULL;
918 }
919 out:
920 svsk->sk_tcplen = 0;
921 svsk->sk_datalen = 0;
922 }
923
924 /*
925 * Receive fragment record header.
926 * If we haven't gotten the record length yet, get the next four bytes.
927 */
928 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
929 {
930 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
931 unsigned int want;
932 int len;
933
934 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
935 struct kvec iov;
936
937 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
938 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
939 iov.iov_len = want;
940 len = svc_recvfrom(rqstp, &iov, 1, want, 0);
941 if (len < 0)
942 goto error;
943 svsk->sk_tcplen += len;
944
945 if (len < want) {
946 dprintk("svc: short recvfrom while reading record "
947 "length (%d of %d)\n", len, want);
948 return -EAGAIN;
949 }
950
951 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
952 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
953 serv->sv_max_mesg) {
954 net_notice_ratelimited("RPC: fragment too large: %d\n",
955 svc_sock_reclen(svsk));
956 goto err_delete;
957 }
958 }
959
960 return svc_sock_reclen(svsk);
961 error:
962 dprintk("RPC: TCP recv_record got %d\n", len);
963 return len;
964 err_delete:
965 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
966 return -EAGAIN;
967 }
968
969 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
970 {
971 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
972 struct rpc_rqst *req = NULL;
973 struct kvec *src, *dst;
974 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
975 __be32 xid;
976 __be32 calldir;
977
978 xid = *p++;
979 calldir = *p;
980
981 if (!bc_xprt)
982 return -EAGAIN;
983 spin_lock(&bc_xprt->queue_lock);
984 req = xprt_lookup_rqst(bc_xprt, xid);
985 if (!req)
986 goto unlock_notfound;
987
988 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
989 /*
990 * XXX!: cheating for now! Only copying HEAD.
991 * But we know this is good enough for now (in fact, for any
992 * callback reply in the forseeable future).
993 */
994 dst = &req->rq_private_buf.head[0];
995 src = &rqstp->rq_arg.head[0];
996 if (dst->iov_len < src->iov_len)
997 goto unlock_eagain; /* whatever; just giving up. */
998 memcpy(dst->iov_base, src->iov_base, src->iov_len);
999 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1000 rqstp->rq_arg.len = 0;
1001 spin_unlock(&bc_xprt->queue_lock);
1002 return 0;
1003 unlock_notfound:
1004 printk(KERN_NOTICE
1005 "%s: Got unrecognized reply: "
1006 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1007 __func__, ntohl(calldir),
1008 bc_xprt, ntohl(xid));
1009 unlock_eagain:
1010 spin_unlock(&bc_xprt->queue_lock);
1011 return -EAGAIN;
1012 }
1013
1014 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1015 {
1016 int i = 0;
1017 int t = 0;
1018
1019 while (t < len) {
1020 vec[i].iov_base = page_address(pages[i]);
1021 vec[i].iov_len = PAGE_SIZE;
1022 i++;
1023 t += PAGE_SIZE;
1024 }
1025 return i;
1026 }
1027
1028 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1029 {
1030 /* If we have more data, signal svc_xprt_enqueue() to try again */
1031 dprintk("svc: TCP %s record (%d bytes)\n",
1032 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1033 svc_sock_reclen(svsk));
1034 svsk->sk_tcplen = 0;
1035 svsk->sk_reclen = 0;
1036 }
1037
1038 /*
1039 * Receive data from a TCP socket.
1040 */
1041 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1042 {
1043 struct svc_sock *svsk =
1044 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1045 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1046 int len;
1047 struct kvec *vec;
1048 unsigned int want, base;
1049 __be32 *p;
1050 __be32 calldir;
1051 int pnum;
1052
1053 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1054 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1055 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1056 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1057
1058 len = svc_tcp_recv_record(svsk, rqstp);
1059 if (len < 0)
1060 goto error;
1061
1062 base = svc_tcp_restore_pages(svsk, rqstp);
1063 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1064
1065 vec = rqstp->rq_vec;
1066
1067 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want);
1068
1069 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1070 rqstp->rq_next_page = rqstp->rq_respages + 1;
1071
1072 /* Now receive data */
1073 len = svc_recvfrom(rqstp, vec, pnum, base + want, base);
1074 if (len >= 0) {
1075 svsk->sk_tcplen += len;
1076 svsk->sk_datalen += len;
1077 }
1078 if (len != want || !svc_sock_final_rec(svsk)) {
1079 svc_tcp_save_pages(svsk, rqstp);
1080 if (len < 0 && len != -EAGAIN)
1081 goto err_delete;
1082 if (len == want)
1083 svc_tcp_fragment_received(svsk);
1084 else
1085 dprintk("svc: incomplete TCP record (%d of %d)\n",
1086 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1087 svc_sock_reclen(svsk));
1088 goto err_noclose;
1089 }
1090
1091 if (svsk->sk_datalen < 8) {
1092 svsk->sk_datalen = 0;
1093 goto err_delete; /* client is nuts. */
1094 }
1095
1096 rqstp->rq_arg.len = svsk->sk_datalen;
1097 rqstp->rq_arg.page_base = 0;
1098 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1099 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1100 rqstp->rq_arg.page_len = 0;
1101 } else
1102 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1103
1104 rqstp->rq_xprt_ctxt = NULL;
1105 rqstp->rq_prot = IPPROTO_TCP;
1106 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1107 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1108 else
1109 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1110
1111 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1112 calldir = p[1];
1113 if (calldir)
1114 len = receive_cb_reply(svsk, rqstp);
1115
1116 /* Reset TCP read info */
1117 svsk->sk_datalen = 0;
1118 svc_tcp_fragment_received(svsk);
1119
1120 if (len < 0)
1121 goto error;
1122
1123 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1124 if (serv->sv_stats)
1125 serv->sv_stats->nettcpcnt++;
1126
1127 return rqstp->rq_arg.len;
1128
1129 error:
1130 if (len != -EAGAIN)
1131 goto err_delete;
1132 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1133 return 0;
1134 err_delete:
1135 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1136 svsk->sk_xprt.xpt_server->sv_name, -len);
1137 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1138 err_noclose:
1139 return 0; /* record not complete */
1140 }
1141
1142 /*
1143 * Send out data on TCP socket.
1144 */
1145 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1146 {
1147 struct xdr_buf *xbufp = &rqstp->rq_res;
1148 int sent;
1149 __be32 reclen;
1150
1151 /* Set up the first element of the reply kvec.
1152 * Any other kvecs that may be in use have been taken
1153 * care of by the server implementation itself.
1154 */
1155 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1156 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1157
1158 sent = svc_sendto(rqstp, &rqstp->rq_res);
1159 if (sent != xbufp->len) {
1160 printk(KERN_NOTICE
1161 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1162 "- shutting down socket\n",
1163 rqstp->rq_xprt->xpt_server->sv_name,
1164 (sent<0)?"got error":"sent only",
1165 sent, xbufp->len);
1166 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1167 svc_xprt_enqueue(rqstp->rq_xprt);
1168 sent = -EAGAIN;
1169 }
1170 return sent;
1171 }
1172
1173 /*
1174 * Setup response header. TCP has a 4B record length field.
1175 */
1176 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1177 {
1178 struct kvec *resv = &rqstp->rq_res.head[0];
1179
1180 /* tcp needs a space for the record length... */
1181 svc_putnl(resv, 0);
1182 }
1183
1184 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1185 struct net *net,
1186 struct sockaddr *sa, int salen,
1187 int flags)
1188 {
1189 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1190 }
1191
1192 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1193 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1194 struct net *, struct sockaddr *,
1195 int, int);
1196 static void svc_bc_sock_free(struct svc_xprt *xprt);
1197
1198 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1199 struct net *net,
1200 struct sockaddr *sa, int salen,
1201 int flags)
1202 {
1203 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1204 }
1205
1206 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1207 {
1208 }
1209
1210 static const struct svc_xprt_ops svc_tcp_bc_ops = {
1211 .xpo_create = svc_bc_tcp_create,
1212 .xpo_detach = svc_bc_tcp_sock_detach,
1213 .xpo_free = svc_bc_sock_free,
1214 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1215 .xpo_secure_port = svc_sock_secure_port,
1216 };
1217
1218 static struct svc_xprt_class svc_tcp_bc_class = {
1219 .xcl_name = "tcp-bc",
1220 .xcl_owner = THIS_MODULE,
1221 .xcl_ops = &svc_tcp_bc_ops,
1222 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1223 };
1224
1225 static void svc_init_bc_xprt_sock(void)
1226 {
1227 svc_reg_xprt_class(&svc_tcp_bc_class);
1228 }
1229
1230 static void svc_cleanup_bc_xprt_sock(void)
1231 {
1232 svc_unreg_xprt_class(&svc_tcp_bc_class);
1233 }
1234 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1235 static void svc_init_bc_xprt_sock(void)
1236 {
1237 }
1238
1239 static void svc_cleanup_bc_xprt_sock(void)
1240 {
1241 }
1242 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1243
1244 static const struct svc_xprt_ops svc_tcp_ops = {
1245 .xpo_create = svc_tcp_create,
1246 .xpo_recvfrom = svc_tcp_recvfrom,
1247 .xpo_sendto = svc_tcp_sendto,
1248 .xpo_release_rqst = svc_release_skb,
1249 .xpo_detach = svc_tcp_sock_detach,
1250 .xpo_free = svc_sock_free,
1251 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1252 .xpo_has_wspace = svc_tcp_has_wspace,
1253 .xpo_accept = svc_tcp_accept,
1254 .xpo_secure_port = svc_sock_secure_port,
1255 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1256 };
1257
1258 static struct svc_xprt_class svc_tcp_class = {
1259 .xcl_name = "tcp",
1260 .xcl_owner = THIS_MODULE,
1261 .xcl_ops = &svc_tcp_ops,
1262 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1263 .xcl_ident = XPRT_TRANSPORT_TCP,
1264 };
1265
1266 void svc_init_xprt_sock(void)
1267 {
1268 svc_reg_xprt_class(&svc_tcp_class);
1269 svc_reg_xprt_class(&svc_udp_class);
1270 svc_init_bc_xprt_sock();
1271 }
1272
1273 void svc_cleanup_xprt_sock(void)
1274 {
1275 svc_unreg_xprt_class(&svc_tcp_class);
1276 svc_unreg_xprt_class(&svc_udp_class);
1277 svc_cleanup_bc_xprt_sock();
1278 }
1279
1280 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1281 {
1282 struct sock *sk = svsk->sk_sk;
1283
1284 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1285 &svsk->sk_xprt, serv);
1286 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1287 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1288 if (sk->sk_state == TCP_LISTEN) {
1289 dprintk("setting up TCP socket for listening\n");
1290 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1291 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1292 sk->sk_data_ready = svc_tcp_listen_data_ready;
1293 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1294 } else {
1295 dprintk("setting up TCP socket for reading\n");
1296 sk->sk_state_change = svc_tcp_state_change;
1297 sk->sk_data_ready = svc_data_ready;
1298 sk->sk_write_space = svc_write_space;
1299
1300 svsk->sk_reclen = 0;
1301 svsk->sk_tcplen = 0;
1302 svsk->sk_datalen = 0;
1303 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1304
1305 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1306
1307 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1308 switch (sk->sk_state) {
1309 case TCP_SYN_RECV:
1310 case TCP_ESTABLISHED:
1311 break;
1312 default:
1313 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1314 }
1315 }
1316 }
1317
1318 void svc_sock_update_bufs(struct svc_serv *serv)
1319 {
1320 /*
1321 * The number of server threads has changed. Update
1322 * rcvbuf and sndbuf accordingly on all sockets
1323 */
1324 struct svc_sock *svsk;
1325
1326 spin_lock_bh(&serv->sv_lock);
1327 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1328 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1329 spin_unlock_bh(&serv->sv_lock);
1330 }
1331 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1332
1333 /*
1334 * Initialize socket for RPC use and create svc_sock struct
1335 */
1336 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1337 struct socket *sock,
1338 int flags)
1339 {
1340 struct svc_sock *svsk;
1341 struct sock *inet;
1342 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1343 int err = 0;
1344
1345 dprintk("svc: svc_setup_socket %p\n", sock);
1346 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1347 if (!svsk)
1348 return ERR_PTR(-ENOMEM);
1349
1350 inet = sock->sk;
1351
1352 /* Register socket with portmapper */
1353 if (pmap_register)
1354 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1355 inet->sk_protocol,
1356 ntohs(inet_sk(inet)->inet_sport));
1357
1358 if (err < 0) {
1359 kfree(svsk);
1360 return ERR_PTR(err);
1361 }
1362
1363 svsk->sk_sock = sock;
1364 svsk->sk_sk = inet;
1365 svsk->sk_ostate = inet->sk_state_change;
1366 svsk->sk_odata = inet->sk_data_ready;
1367 svsk->sk_owspace = inet->sk_write_space;
1368 /*
1369 * This barrier is necessary in order to prevent race condition
1370 * with svc_data_ready(), svc_listen_data_ready() and others
1371 * when calling callbacks above.
1372 */
1373 wmb();
1374 inet->sk_user_data = svsk;
1375
1376 /* Initialize the socket */
1377 if (sock->type == SOCK_DGRAM)
1378 svc_udp_init(svsk, serv);
1379 else
1380 svc_tcp_init(svsk, serv);
1381
1382 dprintk("svc: svc_setup_socket created %p (inet %p), "
1383 "listen %d close %d\n",
1384 svsk, svsk->sk_sk,
1385 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1386 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1387
1388 return svsk;
1389 }
1390
1391 bool svc_alien_sock(struct net *net, int fd)
1392 {
1393 int err;
1394 struct socket *sock = sockfd_lookup(fd, &err);
1395 bool ret = false;
1396
1397 if (!sock)
1398 goto out;
1399 if (sock_net(sock->sk) != net)
1400 ret = true;
1401 sockfd_put(sock);
1402 out:
1403 return ret;
1404 }
1405 EXPORT_SYMBOL_GPL(svc_alien_sock);
1406
1407 /**
1408 * svc_addsock - add a listener socket to an RPC service
1409 * @serv: pointer to RPC service to which to add a new listener
1410 * @fd: file descriptor of the new listener
1411 * @name_return: pointer to buffer to fill in with name of listener
1412 * @len: size of the buffer
1413 *
1414 * Fills in socket name and returns positive length of name if successful.
1415 * Name is terminated with '\n'. On error, returns a negative errno
1416 * value.
1417 */
1418 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1419 const size_t len)
1420 {
1421 int err = 0;
1422 struct socket *so = sockfd_lookup(fd, &err);
1423 struct svc_sock *svsk = NULL;
1424 struct sockaddr_storage addr;
1425 struct sockaddr *sin = (struct sockaddr *)&addr;
1426 int salen;
1427
1428 if (!so)
1429 return err;
1430 err = -EAFNOSUPPORT;
1431 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1432 goto out;
1433 err = -EPROTONOSUPPORT;
1434 if (so->sk->sk_protocol != IPPROTO_TCP &&
1435 so->sk->sk_protocol != IPPROTO_UDP)
1436 goto out;
1437 err = -EISCONN;
1438 if (so->state > SS_UNCONNECTED)
1439 goto out;
1440 err = -ENOENT;
1441 if (!try_module_get(THIS_MODULE))
1442 goto out;
1443 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1444 if (IS_ERR(svsk)) {
1445 module_put(THIS_MODULE);
1446 err = PTR_ERR(svsk);
1447 goto out;
1448 }
1449 salen = kernel_getsockname(svsk->sk_sock, sin);
1450 if (salen >= 0)
1451 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1452 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1453 return svc_one_sock_name(svsk, name_return, len);
1454 out:
1455 sockfd_put(so);
1456 return err;
1457 }
1458 EXPORT_SYMBOL_GPL(svc_addsock);
1459
1460 /*
1461 * Create socket for RPC service.
1462 */
1463 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1464 int protocol,
1465 struct net *net,
1466 struct sockaddr *sin, int len,
1467 int flags)
1468 {
1469 struct svc_sock *svsk;
1470 struct socket *sock;
1471 int error;
1472 int type;
1473 struct sockaddr_storage addr;
1474 struct sockaddr *newsin = (struct sockaddr *)&addr;
1475 int newlen;
1476 int family;
1477 int val;
1478 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1479
1480 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1481 serv->sv_program->pg_name, protocol,
1482 __svc_print_addr(sin, buf, sizeof(buf)));
1483
1484 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1485 printk(KERN_WARNING "svc: only UDP and TCP "
1486 "sockets supported\n");
1487 return ERR_PTR(-EINVAL);
1488 }
1489
1490 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1491 switch (sin->sa_family) {
1492 case AF_INET6:
1493 family = PF_INET6;
1494 break;
1495 case AF_INET:
1496 family = PF_INET;
1497 break;
1498 default:
1499 return ERR_PTR(-EINVAL);
1500 }
1501
1502 error = __sock_create(net, family, type, protocol, &sock, 1);
1503 if (error < 0)
1504 return ERR_PTR(error);
1505
1506 svc_reclassify_socket(sock);
1507
1508 /*
1509 * If this is an PF_INET6 listener, we want to avoid
1510 * getting requests from IPv4 remotes. Those should
1511 * be shunted to a PF_INET listener via rpcbind.
1512 */
1513 val = 1;
1514 if (family == PF_INET6)
1515 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1516 (char *)&val, sizeof(val));
1517
1518 if (type == SOCK_STREAM)
1519 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1520 error = kernel_bind(sock, sin, len);
1521 if (error < 0)
1522 goto bummer;
1523
1524 error = kernel_getsockname(sock, newsin);
1525 if (error < 0)
1526 goto bummer;
1527 newlen = error;
1528
1529 if (protocol == IPPROTO_TCP) {
1530 if ((error = kernel_listen(sock, 64)) < 0)
1531 goto bummer;
1532 }
1533
1534 svsk = svc_setup_socket(serv, sock, flags);
1535 if (IS_ERR(svsk)) {
1536 error = PTR_ERR(svsk);
1537 goto bummer;
1538 }
1539 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1540 return (struct svc_xprt *)svsk;
1541 bummer:
1542 dprintk("svc: svc_create_socket error = %d\n", -error);
1543 sock_release(sock);
1544 return ERR_PTR(error);
1545 }
1546
1547 /*
1548 * Detach the svc_sock from the socket so that no
1549 * more callbacks occur.
1550 */
1551 static void svc_sock_detach(struct svc_xprt *xprt)
1552 {
1553 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1554 struct sock *sk = svsk->sk_sk;
1555
1556 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1557
1558 /* put back the old socket callbacks */
1559 lock_sock(sk);
1560 sk->sk_state_change = svsk->sk_ostate;
1561 sk->sk_data_ready = svsk->sk_odata;
1562 sk->sk_write_space = svsk->sk_owspace;
1563 sk->sk_user_data = NULL;
1564 release_sock(sk);
1565 }
1566
1567 /*
1568 * Disconnect the socket, and reset the callbacks
1569 */
1570 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1571 {
1572 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1573
1574 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1575
1576 svc_sock_detach(xprt);
1577
1578 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1579 svc_tcp_clear_pages(svsk);
1580 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1581 }
1582 }
1583
1584 /*
1585 * Free the svc_sock's socket resources and the svc_sock itself.
1586 */
1587 static void svc_sock_free(struct svc_xprt *xprt)
1588 {
1589 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1590 dprintk("svc: svc_sock_free(%p)\n", svsk);
1591
1592 if (svsk->sk_sock->file)
1593 sockfd_put(svsk->sk_sock);
1594 else
1595 sock_release(svsk->sk_sock);
1596 kfree(svsk);
1597 }
1598
1599 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1600 /*
1601 * Create a back channel svc_xprt which shares the fore channel socket.
1602 */
1603 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1604 int protocol,
1605 struct net *net,
1606 struct sockaddr *sin, int len,
1607 int flags)
1608 {
1609 struct svc_sock *svsk;
1610 struct svc_xprt *xprt;
1611
1612 if (protocol != IPPROTO_TCP) {
1613 printk(KERN_WARNING "svc: only TCP sockets"
1614 " supported on shared back channel\n");
1615 return ERR_PTR(-EINVAL);
1616 }
1617
1618 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1619 if (!svsk)
1620 return ERR_PTR(-ENOMEM);
1621
1622 xprt = &svsk->sk_xprt;
1623 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1624 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1625
1626 serv->sv_bc_xprt = xprt;
1627
1628 return xprt;
1629 }
1630
1631 /*
1632 * Free a back channel svc_sock.
1633 */
1634 static void svc_bc_sock_free(struct svc_xprt *xprt)
1635 {
1636 if (xprt)
1637 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1638 }
1639 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support