?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.168
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blobd0b5a1c47a32445801b2931156c9057c49ae4cdf
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 int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
329 int buflen)
330 {
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = {
334 .msg_flags = MSG_DONTWAIT,
335 };
336 int len;
337
338 rqstp->rq_xprt_hlen = 0;
339
340 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
341 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, iov, nr, buflen);
342 len = sock_recvmsg(svsk->sk_sock, &msg, msg.msg_flags);
343 /* If we read a full record, then assume there may be more
344 * data to read (stream based sockets only!)
345 */
346 if (len == buflen)
347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
348
349 dprintk("svc: socket %p recvfrom(%p, %zu) = %d\n",
350 svsk, iov[0].iov_base, iov[0].iov_len, len);
351 return len;
352 }
353
354 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
355 struct kvec *iov, int nr,
356 int buflen, unsigned int base)
357 {
358 size_t save_iovlen;
359 void *save_iovbase;
360 unsigned int i;
361 int ret;
362
363 if (base == 0)
364 return svc_recvfrom(rqstp, iov, nr, buflen);
365
366 for (i = 0; i < nr; i++) {
367 if (iov[i].iov_len > base)
368 break;
369 base -= iov[i].iov_len;
370 }
371 save_iovlen = iov[i].iov_len;
372 save_iovbase = iov[i].iov_base;
373 iov[i].iov_len -= base;
374 iov[i].iov_base += base;
375 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
376 iov[i].iov_len = save_iovlen;
377 iov[i].iov_base = save_iovbase;
378 return ret;
379 }
380
381 /*
382 * Set socket snd and rcv buffer lengths
383 */
384 static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
385 {
386 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
387 struct socket *sock = svsk->sk_sock;
388
389 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
390
391 lock_sock(sock->sk);
392 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
393 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
394 sock->sk->sk_write_space(sock->sk);
395 release_sock(sock->sk);
396 }
397
398 static void svc_sock_secure_port(struct svc_rqst *rqstp)
399 {
400 if (svc_port_is_privileged(svc_addr(rqstp)))
401 set_bit(RQ_SECURE, &rqstp->rq_flags);
402 else
403 clear_bit(RQ_SECURE, &rqstp->rq_flags);
404 }
405
406 /*
407 * INET callback when data has been received on the socket.
408 */
409 static void svc_data_ready(struct sock *sk)
410 {
411 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
412
413 if (svsk) {
414 dprintk("svc: socket %p(inet %p), busy=%d\n",
415 svsk, sk,
416 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
417
418 /* Refer to svc_setup_socket() for details. */
419 rmb();
420 svsk->sk_odata(sk);
421 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
422 svc_xprt_enqueue(&svsk->sk_xprt);
423 }
424 }
425
426 /*
427 * INET callback when space is newly available on the socket.
428 */
429 static void svc_write_space(struct sock *sk)
430 {
431 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
432
433 if (svsk) {
434 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
435 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
436
437 /* Refer to svc_setup_socket() for details. */
438 rmb();
439 svsk->sk_owspace(sk);
440 svc_xprt_enqueue(&svsk->sk_xprt);
441 }
442 }
443
444 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
445 {
446 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
447
448 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
449 return 1;
450 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
451 }
452
453 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
454 {
455 struct svc_sock *svsk;
456 struct socket *sock;
457 struct linger no_linger = {
458 .l_onoff = 1,
459 .l_linger = 0,
460 };
461
462 svsk = container_of(xprt, struct svc_sock, sk_xprt);
463 sock = svsk->sk_sock;
464 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
465 (char *)&no_linger, sizeof(no_linger));
466 }
467
468 /*
469 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
470 */
471 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
472 struct cmsghdr *cmh)
473 {
474 struct in_pktinfo *pki = CMSG_DATA(cmh);
475 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
476
477 if (cmh->cmsg_type != IP_PKTINFO)
478 return 0;
479
480 daddr->sin_family = AF_INET;
481 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
482 return 1;
483 }
484
485 /*
486 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
487 */
488 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
489 struct cmsghdr *cmh)
490 {
491 struct in6_pktinfo *pki = CMSG_DATA(cmh);
492 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
493
494 if (cmh->cmsg_type != IPV6_PKTINFO)
495 return 0;
496
497 daddr->sin6_family = AF_INET6;
498 daddr->sin6_addr = pki->ipi6_addr;
499 daddr->sin6_scope_id = pki->ipi6_ifindex;
500 return 1;
501 }
502
503 /*
504 * Copy the UDP datagram's destination address to the rqstp structure.
505 * The 'destination' address in this case is the address to which the
506 * peer sent the datagram, i.e. our local address. For multihomed
507 * hosts, this can change from msg to msg. Note that only the IP
508 * address changes, the port number should remain the same.
509 */
510 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
511 struct cmsghdr *cmh)
512 {
513 switch (cmh->cmsg_level) {
514 case SOL_IP:
515 return svc_udp_get_dest_address4(rqstp, cmh);
516 case SOL_IPV6:
517 return svc_udp_get_dest_address6(rqstp, cmh);
518 }
519
520 return 0;
521 }
522
523 /*
524 * Receive a datagram from a UDP socket.
525 */
526 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
527 {
528 struct svc_sock *svsk =
529 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
530 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
531 struct sk_buff *skb;
532 union {
533 struct cmsghdr hdr;
534 long all[SVC_PKTINFO_SPACE / sizeof(long)];
535 } buffer;
536 struct cmsghdr *cmh = &buffer.hdr;
537 struct msghdr msg = {
538 .msg_name = svc_addr(rqstp),
539 .msg_control = cmh,
540 .msg_controllen = sizeof(buffer),
541 .msg_flags = MSG_DONTWAIT,
542 };
543 size_t len;
544 int err;
545
546 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
547 /* udp sockets need large rcvbuf as all pending
548 * requests are still in that buffer. sndbuf must
549 * also be large enough that there is enough space
550 * for one reply per thread. We count all threads
551 * rather than threads in a particular pool, which
552 * provides an upper bound on the number of threads
553 * which will access the socket.
554 */
555 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
556
557 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
558 skb = NULL;
559 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
560 0, 0, MSG_PEEK | MSG_DONTWAIT);
561 if (err >= 0)
562 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
563
564 if (skb == NULL) {
565 if (err != -EAGAIN) {
566 /* possibly an icmp error */
567 dprintk("svc: recvfrom returned error %d\n", -err);
568 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
569 }
570 return 0;
571 }
572 len = svc_addr_len(svc_addr(rqstp));
573 rqstp->rq_addrlen = len;
574 if (skb->tstamp == 0) {
575 skb->tstamp = ktime_get_real();
576 /* Don't enable netstamp, sunrpc doesn't
577 need that much accuracy */
578 }
579 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
580 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
581
582 len = skb->len;
583 rqstp->rq_arg.len = len;
584
585 rqstp->rq_prot = IPPROTO_UDP;
586
587 if (!svc_udp_get_dest_address(rqstp, cmh)) {
588 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
589 cmh->cmsg_level, cmh->cmsg_type);
590 goto out_free;
591 }
592 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
593
594 if (skb_is_nonlinear(skb)) {
595 /* we have to copy */
596 local_bh_disable();
597 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
598 local_bh_enable();
599 /* checksum error */
600 goto out_free;
601 }
602 local_bh_enable();
603 consume_skb(skb);
604 } else {
605 /* we can use it in-place */
606 rqstp->rq_arg.head[0].iov_base = skb->data;
607 rqstp->rq_arg.head[0].iov_len = len;
608 if (skb_checksum_complete(skb))
609 goto out_free;
610 rqstp->rq_xprt_ctxt = skb;
611 }
612
613 rqstp->rq_arg.page_base = 0;
614 if (len <= rqstp->rq_arg.head[0].iov_len) {
615 rqstp->rq_arg.head[0].iov_len = len;
616 rqstp->rq_arg.page_len = 0;
617 rqstp->rq_respages = rqstp->rq_pages+1;
618 } else {
619 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
620 rqstp->rq_respages = rqstp->rq_pages + 1 +
621 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
622 }
623 rqstp->rq_next_page = rqstp->rq_respages+1;
624
625 if (serv->sv_stats)
626 serv->sv_stats->netudpcnt++;
627
628 return len;
629 out_free:
630 kfree_skb(skb);
631 return 0;
632 }
633
634 static int
635 svc_udp_sendto(struct svc_rqst *rqstp)
636 {
637 int error;
638
639 svc_release_udp_skb(rqstp);
640
641 error = svc_sendto(rqstp, &rqstp->rq_res);
642 if (error == -ECONNREFUSED)
643 /* ICMP error on earlier request. */
644 error = svc_sendto(rqstp, &rqstp->rq_res);
645
646 return error;
647 }
648
649 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
650 {
651 }
652
653 static int svc_udp_has_wspace(struct svc_xprt *xprt)
654 {
655 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
656 struct svc_serv *serv = xprt->xpt_server;
657 unsigned long required;
658
659 /*
660 * Set the SOCK_NOSPACE flag before checking the available
661 * sock space.
662 */
663 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
664 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
665 if (required*2 > sock_wspace(svsk->sk_sk))
666 return 0;
667 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
668 return 1;
669 }
670
671 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
672 {
673 BUG();
674 return NULL;
675 }
676
677 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
678 {
679 }
680
681 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
682 struct net *net,
683 struct sockaddr *sa, int salen,
684 int flags)
685 {
686 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
687 }
688
689 static const struct svc_xprt_ops svc_udp_ops = {
690 .xpo_create = svc_udp_create,
691 .xpo_recvfrom = svc_udp_recvfrom,
692 .xpo_sendto = svc_udp_sendto,
693 .xpo_release_rqst = svc_release_udp_skb,
694 .xpo_detach = svc_sock_detach,
695 .xpo_free = svc_sock_free,
696 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
697 .xpo_has_wspace = svc_udp_has_wspace,
698 .xpo_accept = svc_udp_accept,
699 .xpo_secure_port = svc_sock_secure_port,
700 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
701 };
702
703 static struct svc_xprt_class svc_udp_class = {
704 .xcl_name = "udp",
705 .xcl_owner = THIS_MODULE,
706 .xcl_ops = &svc_udp_ops,
707 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
708 .xcl_ident = XPRT_TRANSPORT_UDP,
709 };
710
711 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
712 {
713 int err, level, optname, one = 1;
714
715 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
716 &svsk->sk_xprt, serv);
717 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
718 svsk->sk_sk->sk_data_ready = svc_data_ready;
719 svsk->sk_sk->sk_write_space = svc_write_space;
720
721 /* initialise setting must have enough space to
722 * receive and respond to one request.
723 * svc_udp_recvfrom will re-adjust if necessary
724 */
725 svc_sock_setbufsize(svsk, 3);
726
727 /* data might have come in before data_ready set up */
728 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
729 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
730
731 /* make sure we get destination address info */
732 switch (svsk->sk_sk->sk_family) {
733 case AF_INET:
734 level = SOL_IP;
735 optname = IP_PKTINFO;
736 break;
737 case AF_INET6:
738 level = SOL_IPV6;
739 optname = IPV6_RECVPKTINFO;
740 break;
741 default:
742 BUG();
743 }
744 err = kernel_setsockopt(svsk->sk_sock, level, optname,
745 (char *)&one, sizeof(one));
746 dprintk("svc: kernel_setsockopt returned %d\n", err);
747 }
748
749 /*
750 * A data_ready event on a listening socket means there's a connection
751 * pending. Do not use state_change as a substitute for it.
752 */
753 static void svc_tcp_listen_data_ready(struct sock *sk)
754 {
755 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
756
757 dprintk("svc: socket %p TCP (listen) state change %d\n",
758 sk, sk->sk_state);
759
760 if (svsk) {
761 /* Refer to svc_setup_socket() for details. */
762 rmb();
763 svsk->sk_odata(sk);
764 }
765
766 /*
767 * This callback may called twice when a new connection
768 * is established as a child socket inherits everything
769 * from a parent LISTEN socket.
770 * 1) data_ready method of the parent socket will be called
771 * when one of child sockets become ESTABLISHED.
772 * 2) data_ready method of the child socket may be called
773 * when it receives data before the socket is accepted.
774 * In case of 2, we should ignore it silently.
775 */
776 if (sk->sk_state == TCP_LISTEN) {
777 if (svsk) {
778 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
779 svc_xprt_enqueue(&svsk->sk_xprt);
780 } else
781 printk("svc: socket %p: no user data\n", sk);
782 }
783 }
784
785 /*
786 * A state change on a connected socket means it's dying or dead.
787 */
788 static void svc_tcp_state_change(struct sock *sk)
789 {
790 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
791
792 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
793 sk, sk->sk_state, sk->sk_user_data);
794
795 if (!svsk)
796 printk("svc: socket %p: no user data\n", sk);
797 else {
798 /* Refer to svc_setup_socket() for details. */
799 rmb();
800 svsk->sk_ostate(sk);
801 if (sk->sk_state != TCP_ESTABLISHED) {
802 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
803 svc_xprt_enqueue(&svsk->sk_xprt);
804 }
805 }
806 }
807
808 /*
809 * Accept a TCP connection
810 */
811 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
812 {
813 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
814 struct sockaddr_storage addr;
815 struct sockaddr *sin = (struct sockaddr *) &addr;
816 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
817 struct socket *sock = svsk->sk_sock;
818 struct socket *newsock;
819 struct svc_sock *newsvsk;
820 int err, slen;
821 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
822
823 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
824 if (!sock)
825 return NULL;
826
827 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
828 err = kernel_accept(sock, &newsock, O_NONBLOCK);
829 if (err < 0) {
830 if (err == -ENOMEM)
831 printk(KERN_WARNING "%s: no more sockets!\n",
832 serv->sv_name);
833 else if (err != -EAGAIN)
834 net_warn_ratelimited("%s: accept failed (err %d)!\n",
835 serv->sv_name, -err);
836 return NULL;
837 }
838 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
839
840 err = kernel_getpeername(newsock, sin);
841 if (err < 0) {
842 net_warn_ratelimited("%s: peername failed (err %d)!\n",
843 serv->sv_name, -err);
844 goto failed; /* aborted connection or whatever */
845 }
846 slen = err;
847
848 /* Ideally, we would want to reject connections from unauthorized
849 * hosts here, but when we get encryption, the IP of the host won't
850 * tell us anything. For now just warn about unpriv connections.
851 */
852 if (!svc_port_is_privileged(sin)) {
853 dprintk("%s: connect from unprivileged port: %s\n",
854 serv->sv_name,
855 __svc_print_addr(sin, buf, sizeof(buf)));
856 }
857 dprintk("%s: connect from %s\n", serv->sv_name,
858 __svc_print_addr(sin, buf, sizeof(buf)));
859
860 /* Reset the inherited callbacks before calling svc_setup_socket */
861 newsock->sk->sk_state_change = svsk->sk_ostate;
862 newsock->sk->sk_data_ready = svsk->sk_odata;
863 newsock->sk->sk_write_space = svsk->sk_owspace;
864
865 /* make sure that a write doesn't block forever when
866 * low on memory
867 */
868 newsock->sk->sk_sndtimeo = HZ*30;
869
870 newsvsk = svc_setup_socket(serv, newsock,
871 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
872 if (IS_ERR(newsvsk))
873 goto failed;
874 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
875 err = kernel_getsockname(newsock, sin);
876 slen = err;
877 if (unlikely(err < 0)) {
878 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
879 slen = offsetof(struct sockaddr, sa_data);
880 }
881 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
882
883 if (sock_is_loopback(newsock->sk))
884 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
885 else
886 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
887 if (serv->sv_stats)
888 serv->sv_stats->nettcpconn++;
889
890 return &newsvsk->sk_xprt;
891
892 failed:
893 sock_release(newsock);
894 return NULL;
895 }
896
897 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
898 {
899 unsigned int i, len, npages;
900
901 if (svsk->sk_datalen == 0)
902 return 0;
903 len = svsk->sk_datalen;
904 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
905 for (i = 0; i < npages; i++) {
906 if (rqstp->rq_pages[i] != NULL)
907 put_page(rqstp->rq_pages[i]);
908 BUG_ON(svsk->sk_pages[i] == NULL);
909 rqstp->rq_pages[i] = svsk->sk_pages[i];
910 svsk->sk_pages[i] = NULL;
911 }
912 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
913 return len;
914 }
915
916 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
917 {
918 unsigned int i, len, npages;
919
920 if (svsk->sk_datalen == 0)
921 return;
922 len = svsk->sk_datalen;
923 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
924 for (i = 0; i < npages; i++) {
925 svsk->sk_pages[i] = rqstp->rq_pages[i];
926 rqstp->rq_pages[i] = NULL;
927 }
928 }
929
930 static void svc_tcp_clear_pages(struct svc_sock *svsk)
931 {
932 unsigned int i, len, npages;
933
934 if (svsk->sk_datalen == 0)
935 goto out;
936 len = svsk->sk_datalen;
937 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
938 for (i = 0; i < npages; i++) {
939 if (svsk->sk_pages[i] == NULL) {
940 WARN_ON_ONCE(1);
941 continue;
942 }
943 put_page(svsk->sk_pages[i]);
944 svsk->sk_pages[i] = NULL;
945 }
946 out:
947 svsk->sk_tcplen = 0;
948 svsk->sk_datalen = 0;
949 }
950
951 /*
952 * Receive fragment record header.
953 * If we haven't gotten the record length yet, get the next four bytes.
954 */
955 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
956 {
957 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
958 unsigned int want;
959 int len;
960
961 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
962 struct kvec iov;
963
964 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
965 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
966 iov.iov_len = want;
967 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
968 goto error;
969 svsk->sk_tcplen += len;
970
971 if (len < want) {
972 dprintk("svc: short recvfrom while reading record "
973 "length (%d of %d)\n", len, want);
974 return -EAGAIN;
975 }
976
977 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
978 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
979 serv->sv_max_mesg) {
980 net_notice_ratelimited("RPC: fragment too large: %d\n",
981 svc_sock_reclen(svsk));
982 goto err_delete;
983 }
984 }
985
986 return svc_sock_reclen(svsk);
987 error:
988 dprintk("RPC: TCP recv_record got %d\n", len);
989 return len;
990 err_delete:
991 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
992 return -EAGAIN;
993 }
994
995 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
996 {
997 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
998 struct rpc_rqst *req = NULL;
999 struct kvec *src, *dst;
1000 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1001 __be32 xid;
1002 __be32 calldir;
1003
1004 xid = *p++;
1005 calldir = *p;
1006
1007 if (!bc_xprt)
1008 return -EAGAIN;
1009 spin_lock(&bc_xprt->recv_lock);
1010 req = xprt_lookup_rqst(bc_xprt, xid);
1011 if (!req)
1012 goto unlock_notfound;
1013
1014 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1015 /*
1016 * XXX!: cheating for now! Only copying HEAD.
1017 * But we know this is good enough for now (in fact, for any
1018 * callback reply in the forseeable future).
1019 */
1020 dst = &req->rq_private_buf.head[0];
1021 src = &rqstp->rq_arg.head[0];
1022 if (dst->iov_len < src->iov_len)
1023 goto unlock_eagain; /* whatever; just giving up. */
1024 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1025 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1026 rqstp->rq_arg.len = 0;
1027 spin_unlock(&bc_xprt->recv_lock);
1028 return 0;
1029 unlock_notfound:
1030 printk(KERN_NOTICE
1031 "%s: Got unrecognized reply: "
1032 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1033 __func__, ntohl(calldir),
1034 bc_xprt, ntohl(xid));
1035 unlock_eagain:
1036 spin_unlock(&bc_xprt->recv_lock);
1037 return -EAGAIN;
1038 }
1039
1040 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1041 {
1042 int i = 0;
1043 int t = 0;
1044
1045 while (t < len) {
1046 vec[i].iov_base = page_address(pages[i]);
1047 vec[i].iov_len = PAGE_SIZE;
1048 i++;
1049 t += PAGE_SIZE;
1050 }
1051 return i;
1052 }
1053
1054 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1055 {
1056 /* If we have more data, signal svc_xprt_enqueue() to try again */
1057 dprintk("svc: TCP %s record (%d bytes)\n",
1058 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1059 svc_sock_reclen(svsk));
1060 svsk->sk_tcplen = 0;
1061 svsk->sk_reclen = 0;
1062 }
1063
1064 /*
1065 * Receive data from a TCP socket.
1066 */
1067 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1068 {
1069 struct svc_sock *svsk =
1070 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1071 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1072 int len;
1073 struct kvec *vec;
1074 unsigned int want, base;
1075 __be32 *p;
1076 __be32 calldir;
1077 int pnum;
1078
1079 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1080 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1081 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1082 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1083
1084 len = svc_tcp_recv_record(svsk, rqstp);
1085 if (len < 0)
1086 goto error;
1087
1088 base = svc_tcp_restore_pages(svsk, rqstp);
1089 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1090
1091 vec = rqstp->rq_vec;
1092
1093 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1094 svsk->sk_datalen + want);
1095
1096 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1097 rqstp->rq_next_page = rqstp->rq_respages + 1;
1098
1099 /* Now receive data */
1100 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1101 if (len >= 0) {
1102 svsk->sk_tcplen += len;
1103 svsk->sk_datalen += len;
1104 }
1105 if (len != want || !svc_sock_final_rec(svsk)) {
1106 svc_tcp_save_pages(svsk, rqstp);
1107 if (len < 0 && len != -EAGAIN)
1108 goto err_delete;
1109 if (len == want)
1110 svc_tcp_fragment_received(svsk);
1111 else
1112 dprintk("svc: incomplete TCP record (%d of %d)\n",
1113 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1114 svc_sock_reclen(svsk));
1115 goto err_noclose;
1116 }
1117
1118 if (svsk->sk_datalen < 8) {
1119 svsk->sk_datalen = 0;
1120 goto err_delete; /* client is nuts. */
1121 }
1122
1123 rqstp->rq_arg.len = svsk->sk_datalen;
1124 rqstp->rq_arg.page_base = 0;
1125 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1126 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1127 rqstp->rq_arg.page_len = 0;
1128 } else
1129 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1130
1131 rqstp->rq_xprt_ctxt = NULL;
1132 rqstp->rq_prot = IPPROTO_TCP;
1133 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1134 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1135 else
1136 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1137
1138 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1139 calldir = p[1];
1140 if (calldir)
1141 len = receive_cb_reply(svsk, rqstp);
1142
1143 /* Reset TCP read info */
1144 svsk->sk_datalen = 0;
1145 svc_tcp_fragment_received(svsk);
1146
1147 if (len < 0)
1148 goto error;
1149
1150 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1151 if (serv->sv_stats)
1152 serv->sv_stats->nettcpcnt++;
1153
1154 return rqstp->rq_arg.len;
1155
1156 error:
1157 if (len != -EAGAIN)
1158 goto err_delete;
1159 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1160 return 0;
1161 err_delete:
1162 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1163 svsk->sk_xprt.xpt_server->sv_name, -len);
1164 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1165 err_noclose:
1166 return 0; /* record not complete */
1167 }
1168
1169 /*
1170 * Send out data on TCP socket.
1171 */
1172 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1173 {
1174 struct xdr_buf *xbufp = &rqstp->rq_res;
1175 int sent;
1176 __be32 reclen;
1177
1178 svc_release_skb(rqstp);
1179
1180 /* Set up the first element of the reply kvec.
1181 * Any other kvecs that may be in use have been taken
1182 * care of by the server implementation itself.
1183 */
1184 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1185 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1186
1187 sent = svc_sendto(rqstp, &rqstp->rq_res);
1188 if (sent != xbufp->len) {
1189 printk(KERN_NOTICE
1190 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1191 "- shutting down socket\n",
1192 rqstp->rq_xprt->xpt_server->sv_name,
1193 (sent<0)?"got error":"sent only",
1194 sent, xbufp->len);
1195 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1196 svc_xprt_enqueue(rqstp->rq_xprt);
1197 sent = -EAGAIN;
1198 }
1199 return sent;
1200 }
1201
1202 /*
1203 * Setup response header. TCP has a 4B record length field.
1204 */
1205 void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1206 {
1207 struct kvec *resv = &rqstp->rq_res.head[0];
1208
1209 /* tcp needs a space for the record length... */
1210 svc_putnl(resv, 0);
1211 }
1212
1213 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1214 struct net *net,
1215 struct sockaddr *sa, int salen,
1216 int flags)
1217 {
1218 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1219 }
1220
1221 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1222 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1223 struct net *, struct sockaddr *,
1224 int, int);
1225 static void svc_bc_sock_free(struct svc_xprt *xprt);
1226
1227 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1228 struct net *net,
1229 struct sockaddr *sa, int salen,
1230 int flags)
1231 {
1232 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1233 }
1234
1235 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1236 {
1237 }
1238
1239 static const struct svc_xprt_ops svc_tcp_bc_ops = {
1240 .xpo_create = svc_bc_tcp_create,
1241 .xpo_detach = svc_bc_tcp_sock_detach,
1242 .xpo_free = svc_bc_sock_free,
1243 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1244 .xpo_secure_port = svc_sock_secure_port,
1245 };
1246
1247 static struct svc_xprt_class svc_tcp_bc_class = {
1248 .xcl_name = "tcp-bc",
1249 .xcl_owner = THIS_MODULE,
1250 .xcl_ops = &svc_tcp_bc_ops,
1251 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1252 };
1253
1254 static void svc_init_bc_xprt_sock(void)
1255 {
1256 svc_reg_xprt_class(&svc_tcp_bc_class);
1257 }
1258
1259 static void svc_cleanup_bc_xprt_sock(void)
1260 {
1261 svc_unreg_xprt_class(&svc_tcp_bc_class);
1262 }
1263 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1264 static void svc_init_bc_xprt_sock(void)
1265 {
1266 }
1267
1268 static void svc_cleanup_bc_xprt_sock(void)
1269 {
1270 }
1271 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1272
1273 static const struct svc_xprt_ops svc_tcp_ops = {
1274 .xpo_create = svc_tcp_create,
1275 .xpo_recvfrom = svc_tcp_recvfrom,
1276 .xpo_sendto = svc_tcp_sendto,
1277 .xpo_release_rqst = svc_release_skb,
1278 .xpo_detach = svc_tcp_sock_detach,
1279 .xpo_free = svc_sock_free,
1280 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1281 .xpo_has_wspace = svc_tcp_has_wspace,
1282 .xpo_accept = svc_tcp_accept,
1283 .xpo_secure_port = svc_sock_secure_port,
1284 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1285 };
1286
1287 static struct svc_xprt_class svc_tcp_class = {
1288 .xcl_name = "tcp",
1289 .xcl_owner = THIS_MODULE,
1290 .xcl_ops = &svc_tcp_ops,
1291 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1292 .xcl_ident = XPRT_TRANSPORT_TCP,
1293 };
1294
1295 void svc_init_xprt_sock(void)
1296 {
1297 svc_reg_xprt_class(&svc_tcp_class);
1298 svc_reg_xprt_class(&svc_udp_class);
1299 svc_init_bc_xprt_sock();
1300 }
1301
1302 void svc_cleanup_xprt_sock(void)
1303 {
1304 svc_unreg_xprt_class(&svc_tcp_class);
1305 svc_unreg_xprt_class(&svc_udp_class);
1306 svc_cleanup_bc_xprt_sock();
1307 }
1308
1309 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1310 {
1311 struct sock *sk = svsk->sk_sk;
1312
1313 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1314 &svsk->sk_xprt, serv);
1315 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1316 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1317 if (sk->sk_state == TCP_LISTEN) {
1318 dprintk("setting up TCP socket for listening\n");
1319 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1320 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1321 sk->sk_data_ready = svc_tcp_listen_data_ready;
1322 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1323 } else {
1324 dprintk("setting up TCP socket for reading\n");
1325 sk->sk_state_change = svc_tcp_state_change;
1326 sk->sk_data_ready = svc_data_ready;
1327 sk->sk_write_space = svc_write_space;
1328
1329 svsk->sk_reclen = 0;
1330 svsk->sk_tcplen = 0;
1331 svsk->sk_datalen = 0;
1332 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1333
1334 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1335
1336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1337 switch (sk->sk_state) {
1338 case TCP_SYN_RECV:
1339 case TCP_ESTABLISHED:
1340 break;
1341 default:
1342 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1343 }
1344 }
1345 }
1346
1347 void svc_sock_update_bufs(struct svc_serv *serv)
1348 {
1349 /*
1350 * The number of server threads has changed. Update
1351 * rcvbuf and sndbuf accordingly on all sockets
1352 */
1353 struct svc_sock *svsk;
1354
1355 spin_lock_bh(&serv->sv_lock);
1356 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1357 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1358 spin_unlock_bh(&serv->sv_lock);
1359 }
1360 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1361
1362 /*
1363 * Initialize socket for RPC use and create svc_sock struct
1364 */
1365 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1366 struct socket *sock,
1367 int flags)
1368 {
1369 struct svc_sock *svsk;
1370 struct sock *inet;
1371 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1372 int err = 0;
1373
1374 dprintk("svc: svc_setup_socket %p\n", sock);
1375 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1376 if (!svsk)
1377 return ERR_PTR(-ENOMEM);
1378
1379 inet = sock->sk;
1380
1381 /* Register socket with portmapper */
1382 if (pmap_register)
1383 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1384 inet->sk_protocol,
1385 ntohs(inet_sk(inet)->inet_sport));
1386
1387 if (err < 0) {
1388 kfree(svsk);
1389 return ERR_PTR(err);
1390 }
1391
1392 svsk->sk_sock = sock;
1393 svsk->sk_sk = inet;
1394 svsk->sk_ostate = inet->sk_state_change;
1395 svsk->sk_odata = inet->sk_data_ready;
1396 svsk->sk_owspace = inet->sk_write_space;
1397 /*
1398 * This barrier is necessary in order to prevent race condition
1399 * with svc_data_ready(), svc_listen_data_ready() and others
1400 * when calling callbacks above.
1401 */
1402 wmb();
1403 inet->sk_user_data = svsk;
1404
1405 /* Initialize the socket */
1406 if (sock->type == SOCK_DGRAM)
1407 svc_udp_init(svsk, serv);
1408 else
1409 svc_tcp_init(svsk, serv);
1410
1411 dprintk("svc: svc_setup_socket created %p (inet %p), "
1412 "listen %d close %d\n",
1413 svsk, svsk->sk_sk,
1414 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1415 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1416
1417 return svsk;
1418 }
1419
1420 bool svc_alien_sock(struct net *net, int fd)
1421 {
1422 int err;
1423 struct socket *sock = sockfd_lookup(fd, &err);
1424 bool ret = false;
1425
1426 if (!sock)
1427 goto out;
1428 if (sock_net(sock->sk) != net)
1429 ret = true;
1430 sockfd_put(sock);
1431 out:
1432 return ret;
1433 }
1434 EXPORT_SYMBOL_GPL(svc_alien_sock);
1435
1436 /**
1437 * svc_addsock - add a listener socket to an RPC service
1438 * @serv: pointer to RPC service to which to add a new listener
1439 * @fd: file descriptor of the new listener
1440 * @name_return: pointer to buffer to fill in with name of listener
1441 * @len: size of the buffer
1442 *
1443 * Fills in socket name and returns positive length of name if successful.
1444 * Name is terminated with '\n'. On error, returns a negative errno
1445 * value.
1446 */
1447 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1448 const size_t len)
1449 {
1450 int err = 0;
1451 struct socket *so = sockfd_lookup(fd, &err);
1452 struct svc_sock *svsk = NULL;
1453 struct sockaddr_storage addr;
1454 struct sockaddr *sin = (struct sockaddr *)&addr;
1455 int salen;
1456
1457 if (!so)
1458 return err;
1459 err = -EAFNOSUPPORT;
1460 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1461 goto out;
1462 err = -EPROTONOSUPPORT;
1463 if (so->sk->sk_protocol != IPPROTO_TCP &&
1464 so->sk->sk_protocol != IPPROTO_UDP)
1465 goto out;
1466 err = -EISCONN;
1467 if (so->state > SS_UNCONNECTED)
1468 goto out;
1469 err = -ENOENT;
1470 if (!try_module_get(THIS_MODULE))
1471 goto out;
1472 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1473 if (IS_ERR(svsk)) {
1474 module_put(THIS_MODULE);
1475 err = PTR_ERR(svsk);
1476 goto out;
1477 }
1478 salen = kernel_getsockname(svsk->sk_sock, sin);
1479 if (salen >= 0)
1480 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1481 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1482 return svc_one_sock_name(svsk, name_return, len);
1483 out:
1484 sockfd_put(so);
1485 return err;
1486 }
1487 EXPORT_SYMBOL_GPL(svc_addsock);
1488
1489 /*
1490 * Create socket for RPC service.
1491 */
1492 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1493 int protocol,
1494 struct net *net,
1495 struct sockaddr *sin, int len,
1496 int flags)
1497 {
1498 struct svc_sock *svsk;
1499 struct socket *sock;
1500 int error;
1501 int type;
1502 struct sockaddr_storage addr;
1503 struct sockaddr *newsin = (struct sockaddr *)&addr;
1504 int newlen;
1505 int family;
1506 int val;
1507 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1508
1509 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1510 serv->sv_program->pg_name, protocol,
1511 __svc_print_addr(sin, buf, sizeof(buf)));
1512
1513 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1514 printk(KERN_WARNING "svc: only UDP and TCP "
1515 "sockets supported\n");
1516 return ERR_PTR(-EINVAL);
1517 }
1518
1519 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1520 switch (sin->sa_family) {
1521 case AF_INET6:
1522 family = PF_INET6;
1523 break;
1524 case AF_INET:
1525 family = PF_INET;
1526 break;
1527 default:
1528 return ERR_PTR(-EINVAL);
1529 }
1530
1531 error = __sock_create(net, family, type, protocol, &sock, 1);
1532 if (error < 0)
1533 return ERR_PTR(error);
1534
1535 svc_reclassify_socket(sock);
1536
1537 /*
1538 * If this is an PF_INET6 listener, we want to avoid
1539 * getting requests from IPv4 remotes. Those should
1540 * be shunted to a PF_INET listener via rpcbind.
1541 */
1542 val = 1;
1543 if (family == PF_INET6)
1544 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1545 (char *)&val, sizeof(val));
1546
1547 if (type == SOCK_STREAM)
1548 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1549 error = kernel_bind(sock, sin, len);
1550 if (error < 0)
1551 goto bummer;
1552
1553 error = kernel_getsockname(sock, newsin);
1554 if (error < 0)
1555 goto bummer;
1556 newlen = error;
1557
1558 if (protocol == IPPROTO_TCP) {
1559 if ((error = kernel_listen(sock, 64)) < 0)
1560 goto bummer;
1561 }
1562
1563 svsk = svc_setup_socket(serv, sock, flags);
1564 if (IS_ERR(svsk)) {
1565 error = PTR_ERR(svsk);
1566 goto bummer;
1567 }
1568 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1569 return (struct svc_xprt *)svsk;
1570 bummer:
1571 dprintk("svc: svc_create_socket error = %d\n", -error);
1572 sock_release(sock);
1573 return ERR_PTR(error);
1574 }
1575
1576 /*
1577 * Detach the svc_sock from the socket so that no
1578 * more callbacks occur.
1579 */
1580 static void svc_sock_detach(struct svc_xprt *xprt)
1581 {
1582 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1583 struct sock *sk = svsk->sk_sk;
1584
1585 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1586
1587 /* put back the old socket callbacks */
1588 lock_sock(sk);
1589 sk->sk_state_change = svsk->sk_ostate;
1590 sk->sk_data_ready = svsk->sk_odata;
1591 sk->sk_write_space = svsk->sk_owspace;
1592 sk->sk_user_data = NULL;
1593 release_sock(sk);
1594 }
1595
1596 /*
1597 * Disconnect the socket, and reset the callbacks
1598 */
1599 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1600 {
1601 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1602
1603 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1604
1605 svc_sock_detach(xprt);
1606
1607 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1608 svc_tcp_clear_pages(svsk);
1609 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1610 }
1611 }
1612
1613 /*
1614 * Free the svc_sock's socket resources and the svc_sock itself.
1615 */
1616 static void svc_sock_free(struct svc_xprt *xprt)
1617 {
1618 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1619 dprintk("svc: svc_sock_free(%p)\n", svsk);
1620
1621 if (svsk->sk_sock->file)
1622 sockfd_put(svsk->sk_sock);
1623 else
1624 sock_release(svsk->sk_sock);
1625 kfree(svsk);
1626 }
1627
1628 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1629 /*
1630 * Create a back channel svc_xprt which shares the fore channel socket.
1631 */
1632 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1633 int protocol,
1634 struct net *net,
1635 struct sockaddr *sin, int len,
1636 int flags)
1637 {
1638 struct svc_sock *svsk;
1639 struct svc_xprt *xprt;
1640
1641 if (protocol != IPPROTO_TCP) {
1642 printk(KERN_WARNING "svc: only TCP sockets"
1643 " supported on shared back channel\n");
1644 return ERR_PTR(-EINVAL);
1645 }
1646
1647 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1648 if (!svsk)
1649 return ERR_PTR(-ENOMEM);
1650
1651 xprt = &svsk->sk_xprt;
1652 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1653 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1654
1655 serv->sv_bc_xprt = xprt;
1656
1657 return xprt;
1658 }
1659
1660 /*
1661 * Free a back channel svc_sock.
1662 */
1663 static void svc_bc_sock_free(struct svc_xprt *xprt)
1664 {
1665 if (xprt)
1666 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1667 }
1668 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support