?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/mm: Add TLB purge to free pmd/pte page interfaces
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blob266a30c8b88b4da05114d0d6a9b33fb519ced3c4
1 /*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
9 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/ip.h>
41 #include <net/ipv6.h>
42 #include <net/tcp.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <trace/events/skb.h>
47
48 #include <linux/sunrpc/types.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/xdr.h>
51 #include <linux/sunrpc/msg_prot.h>
52 #include <linux/sunrpc/svcsock.h>
53 #include <linux/sunrpc/stats.h>
54 #include <linux/sunrpc/xprt.h>
55
56 #include "sunrpc.h"
57
58 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
59
60
61 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
62 int flags);
63 static int svc_udp_recvfrom(struct svc_rqst *);
64 static int svc_udp_sendto(struct svc_rqst *);
65 static void svc_sock_detach(struct svc_xprt *);
66 static void svc_tcp_sock_detach(struct svc_xprt *);
67 static void svc_sock_free(struct svc_xprt *);
68
69 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
70 struct net *, struct sockaddr *,
71 int, int);
72 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
73 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
74 struct net *, struct sockaddr *,
75 int, int);
76 static void svc_bc_sock_free(struct svc_xprt *xprt);
77 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
78
79 #ifdef CONFIG_DEBUG_LOCK_ALLOC
80 static struct lock_class_key svc_key[2];
81 static struct lock_class_key svc_slock_key[2];
82
83 static void svc_reclassify_socket(struct socket *sock)
84 {
85 struct sock *sk = sock->sk;
86
87 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
88 return;
89
90 switch (sk->sk_family) {
91 case AF_INET:
92 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
93 &svc_slock_key[0],
94 "sk_xprt.xpt_lock-AF_INET-NFSD",
95 &svc_key[0]);
96 break;
97
98 case AF_INET6:
99 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
100 &svc_slock_key[1],
101 "sk_xprt.xpt_lock-AF_INET6-NFSD",
102 &svc_key[1]);
103 break;
104
105 default:
106 BUG();
107 }
108 }
109 #else
110 static void svc_reclassify_socket(struct socket *sock)
111 {
112 }
113 #endif
114
115 /*
116 * Release an skbuff after use
117 */
118 static void svc_release_skb(struct svc_rqst *rqstp)
119 {
120 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
121
122 if (skb) {
123 struct svc_sock *svsk =
124 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
125 rqstp->rq_xprt_ctxt = NULL;
126
127 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
128 skb_free_datagram_locked(svsk->sk_sk, skb);
129 }
130 }
131
132 union svc_pktinfo_u {
133 struct in_pktinfo pkti;
134 struct in6_pktinfo pkti6;
135 };
136 #define SVC_PKTINFO_SPACE \
137 CMSG_SPACE(sizeof(union svc_pktinfo_u))
138
139 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
140 {
141 struct svc_sock *svsk =
142 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
143 switch (svsk->sk_sk->sk_family) {
144 case AF_INET: {
145 struct in_pktinfo *pki = CMSG_DATA(cmh);
146
147 cmh->cmsg_level = SOL_IP;
148 cmh->cmsg_type = IP_PKTINFO;
149 pki->ipi_ifindex = 0;
150 pki->ipi_spec_dst.s_addr =
151 svc_daddr_in(rqstp)->sin_addr.s_addr;
152 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
153 }
154 break;
155
156 case AF_INET6: {
157 struct in6_pktinfo *pki = CMSG_DATA(cmh);
158 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
159
160 cmh->cmsg_level = SOL_IPV6;
161 cmh->cmsg_type = IPV6_PKTINFO;
162 pki->ipi6_ifindex = daddr->sin6_scope_id;
163 pki->ipi6_addr = daddr->sin6_addr;
164 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
165 }
166 break;
167 }
168 }
169
170 /*
171 * send routine intended to be shared by the fore- and back-channel
172 */
173 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
174 struct page *headpage, unsigned long headoffset,
175 struct page *tailpage, unsigned long tailoffset)
176 {
177 int result;
178 int size;
179 struct page **ppage = xdr->pages;
180 size_t base = xdr->page_base;
181 unsigned int pglen = xdr->page_len;
182 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
183 int slen;
184 int len = 0;
185
186 slen = xdr->len;
187
188 /* send head */
189 if (slen == xdr->head[0].iov_len)
190 flags = 0;
191 len = kernel_sendpage(sock, headpage, headoffset,
192 xdr->head[0].iov_len, flags);
193 if (len != xdr->head[0].iov_len)
194 goto out;
195 slen -= xdr->head[0].iov_len;
196 if (slen == 0)
197 goto out;
198
199 /* send page data */
200 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
201 while (pglen > 0) {
202 if (slen == size)
203 flags = 0;
204 result = kernel_sendpage(sock, *ppage, base, size, flags);
205 if (result > 0)
206 len += result;
207 if (result != size)
208 goto out;
209 slen -= size;
210 pglen -= size;
211 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
212 base = 0;
213 ppage++;
214 }
215
216 /* send tail */
217 if (xdr->tail[0].iov_len) {
218 result = kernel_sendpage(sock, tailpage, tailoffset,
219 xdr->tail[0].iov_len, 0);
220 if (result > 0)
221 len += result;
222 }
223
224 out:
225 return len;
226 }
227
228
229 /*
230 * Generic sendto routine
231 */
232 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
233 {
234 struct svc_sock *svsk =
235 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
236 struct socket *sock = svsk->sk_sock;
237 union {
238 struct cmsghdr hdr;
239 long all[SVC_PKTINFO_SPACE / sizeof(long)];
240 } buffer;
241 struct cmsghdr *cmh = &buffer.hdr;
242 int len = 0;
243 unsigned long tailoff;
244 unsigned long headoff;
245 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
246
247 if (rqstp->rq_prot == IPPROTO_UDP) {
248 struct msghdr msg = {
249 .msg_name = &rqstp->rq_addr,
250 .msg_namelen = rqstp->rq_addrlen,
251 .msg_control = cmh,
252 .msg_controllen = sizeof(buffer),
253 .msg_flags = MSG_MORE,
254 };
255
256 svc_set_cmsg_data(rqstp, cmh);
257
258 if (sock_sendmsg(sock, &msg) < 0)
259 goto out;
260 }
261
262 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
263 headoff = 0;
264 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
265 rqstp->rq_respages[0], tailoff);
266
267 out:
268 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
269 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
270 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
271
272 return len;
273 }
274
275 /*
276 * Report socket names for nfsdfs
277 */
278 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
279 {
280 const struct sock *sk = svsk->sk_sk;
281 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
282 "udp" : "tcp";
283 int len;
284
285 switch (sk->sk_family) {
286 case PF_INET:
287 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
288 proto_name,
289 &inet_sk(sk)->inet_rcv_saddr,
290 inet_sk(sk)->inet_num);
291 break;
292 #if IS_ENABLED(CONFIG_IPV6)
293 case PF_INET6:
294 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
295 proto_name,
296 &sk->sk_v6_rcv_saddr,
297 inet_sk(sk)->inet_num);
298 break;
299 #endif
300 default:
301 len = snprintf(buf, remaining, "*unknown-%d*\n",
302 sk->sk_family);
303 }
304
305 if (len >= remaining) {
306 *buf = '\0';
307 return -ENAMETOOLONG;
308 }
309 return len;
310 }
311
312 /*
313 * Generic recvfrom routine.
314 */
315 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
316 int buflen)
317 {
318 struct svc_sock *svsk =
319 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
320 struct msghdr msg = {
321 .msg_flags = MSG_DONTWAIT,
322 };
323 int len;
324
325 rqstp->rq_xprt_hlen = 0;
326
327 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
328 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
329 msg.msg_flags);
330 /* If we read a full record, then assume there may be more
331 * data to read (stream based sockets only!)
332 */
333 if (len == buflen)
334 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
335
336 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
337 svsk, iov[0].iov_base, iov[0].iov_len, len);
338 return len;
339 }
340
341 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
342 struct kvec *iov, int nr,
343 int buflen, unsigned int base)
344 {
345 size_t save_iovlen;
346 void *save_iovbase;
347 unsigned int i;
348 int ret;
349
350 if (base == 0)
351 return svc_recvfrom(rqstp, iov, nr, buflen);
352
353 for (i = 0; i < nr; i++) {
354 if (iov[i].iov_len > base)
355 break;
356 base -= iov[i].iov_len;
357 }
358 save_iovlen = iov[i].iov_len;
359 save_iovbase = iov[i].iov_base;
360 iov[i].iov_len -= base;
361 iov[i].iov_base += base;
362 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
363 iov[i].iov_len = save_iovlen;
364 iov[i].iov_base = save_iovbase;
365 return ret;
366 }
367
368 /*
369 * Set socket snd and rcv buffer lengths
370 */
371 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
372 unsigned int rcv)
373 {
374 #if 0
375 mm_segment_t oldfs;
376 oldfs = get_fs(); set_fs(KERNEL_DS);
377 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
378 (char*)&snd, sizeof(snd));
379 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
380 (char*)&rcv, sizeof(rcv));
381 #else
382 /* sock_setsockopt limits use to sysctl_?mem_max,
383 * which isn't acceptable. Until that is made conditional
384 * on not having CAP_SYS_RESOURCE or similar, we go direct...
385 * DaveM said I could!
386 */
387 lock_sock(sock->sk);
388 sock->sk->sk_sndbuf = snd * 2;
389 sock->sk->sk_rcvbuf = rcv * 2;
390 sock->sk->sk_write_space(sock->sk);
391 release_sock(sock->sk);
392 #endif
393 }
394
395 static int svc_sock_secure_port(struct svc_rqst *rqstp)
396 {
397 return svc_port_is_privileged(svc_addr(rqstp));
398 }
399
400 /*
401 * INET callback when data has been received on the socket.
402 */
403 static void svc_data_ready(struct sock *sk)
404 {
405 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
406
407 if (svsk) {
408 dprintk("svc: socket %p(inet %p), busy=%d\n",
409 svsk, sk,
410 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
411
412 /* Refer to svc_setup_socket() for details. */
413 rmb();
414 svsk->sk_odata(sk);
415 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
416 svc_xprt_enqueue(&svsk->sk_xprt);
417 }
418 }
419
420 /*
421 * INET callback when space is newly available on the socket.
422 */
423 static void svc_write_space(struct sock *sk)
424 {
425 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
426
427 if (svsk) {
428 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
429 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
430
431 /* Refer to svc_setup_socket() for details. */
432 rmb();
433 svsk->sk_owspace(sk);
434 svc_xprt_enqueue(&svsk->sk_xprt);
435 }
436 }
437
438 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
439 {
440 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
441
442 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
443 return 1;
444 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
445 }
446
447 static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
448 {
449 struct svc_sock *svsk;
450 struct socket *sock;
451 struct linger no_linger = {
452 .l_onoff = 1,
453 .l_linger = 0,
454 };
455
456 svsk = container_of(xprt, struct svc_sock, sk_xprt);
457 sock = svsk->sk_sock;
458 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
459 (char *)&no_linger, sizeof(no_linger));
460 }
461
462 /*
463 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
464 */
465 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
466 struct cmsghdr *cmh)
467 {
468 struct in_pktinfo *pki = CMSG_DATA(cmh);
469 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
470
471 if (cmh->cmsg_type != IP_PKTINFO)
472 return 0;
473
474 daddr->sin_family = AF_INET;
475 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
476 return 1;
477 }
478
479 /*
480 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
481 */
482 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
483 struct cmsghdr *cmh)
484 {
485 struct in6_pktinfo *pki = CMSG_DATA(cmh);
486 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
487
488 if (cmh->cmsg_type != IPV6_PKTINFO)
489 return 0;
490
491 daddr->sin6_family = AF_INET6;
492 daddr->sin6_addr = pki->ipi6_addr;
493 daddr->sin6_scope_id = pki->ipi6_ifindex;
494 return 1;
495 }
496
497 /*
498 * Copy the UDP datagram's destination address to the rqstp structure.
499 * The 'destination' address in this case is the address to which the
500 * peer sent the datagram, i.e. our local address. For multihomed
501 * hosts, this can change from msg to msg. Note that only the IP
502 * address changes, the port number should remain the same.
503 */
504 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
505 struct cmsghdr *cmh)
506 {
507 switch (cmh->cmsg_level) {
508 case SOL_IP:
509 return svc_udp_get_dest_address4(rqstp, cmh);
510 case SOL_IPV6:
511 return svc_udp_get_dest_address6(rqstp, cmh);
512 }
513
514 return 0;
515 }
516
517 /*
518 * Receive a datagram from a UDP socket.
519 */
520 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
521 {
522 struct svc_sock *svsk =
523 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
524 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
525 struct sk_buff *skb;
526 union {
527 struct cmsghdr hdr;
528 long all[SVC_PKTINFO_SPACE / sizeof(long)];
529 } buffer;
530 struct cmsghdr *cmh = &buffer.hdr;
531 struct msghdr msg = {
532 .msg_name = svc_addr(rqstp),
533 .msg_control = cmh,
534 .msg_controllen = sizeof(buffer),
535 .msg_flags = MSG_DONTWAIT,
536 };
537 size_t len;
538 int err;
539
540 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
541 /* udp sockets need large rcvbuf as all pending
542 * requests are still in that buffer. sndbuf must
543 * also be large enough that there is enough space
544 * for one reply per thread. We count all threads
545 * rather than threads in a particular pool, which
546 * provides an upper bound on the number of threads
547 * which will access the socket.
548 */
549 svc_sock_setbufsize(svsk->sk_sock,
550 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
551 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
552
553 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
554 skb = NULL;
555 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
556 0, 0, MSG_PEEK | MSG_DONTWAIT);
557 if (err >= 0)
558 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
559
560 if (skb == NULL) {
561 if (err != -EAGAIN) {
562 /* possibly an icmp error */
563 dprintk("svc: recvfrom returned error %d\n", -err);
564 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
565 }
566 return 0;
567 }
568 len = svc_addr_len(svc_addr(rqstp));
569 rqstp->rq_addrlen = len;
570 if (skb->tstamp.tv64 == 0) {
571 skb->tstamp = ktime_get_real();
572 /* Don't enable netstamp, sunrpc doesn't
573 need that much accuracy */
574 }
575 svsk->sk_sk->sk_stamp = skb->tstamp;
576 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
577
578 len = skb->len;
579 rqstp->rq_arg.len = len;
580
581 rqstp->rq_prot = IPPROTO_UDP;
582
583 if (!svc_udp_get_dest_address(rqstp, cmh)) {
584 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
585 cmh->cmsg_level, cmh->cmsg_type);
586 goto out_free;
587 }
588 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
589
590 if (skb_is_nonlinear(skb)) {
591 /* we have to copy */
592 local_bh_disable();
593 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
594 local_bh_enable();
595 /* checksum error */
596 goto out_free;
597 }
598 local_bh_enable();
599 skb_free_datagram_locked(svsk->sk_sk, skb);
600 } else {
601 /* we can use it in-place */
602 rqstp->rq_arg.head[0].iov_base = skb->data;
603 rqstp->rq_arg.head[0].iov_len = len;
604 if (skb_checksum_complete(skb))
605 goto out_free;
606 rqstp->rq_xprt_ctxt = skb;
607 }
608
609 rqstp->rq_arg.page_base = 0;
610 if (len <= rqstp->rq_arg.head[0].iov_len) {
611 rqstp->rq_arg.head[0].iov_len = len;
612 rqstp->rq_arg.page_len = 0;
613 rqstp->rq_respages = rqstp->rq_pages+1;
614 } else {
615 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
616 rqstp->rq_respages = rqstp->rq_pages + 1 +
617 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
618 }
619 rqstp->rq_next_page = rqstp->rq_respages+1;
620
621 if (serv->sv_stats)
622 serv->sv_stats->netudpcnt++;
623
624 return len;
625 out_free:
626 trace_kfree_skb(skb, svc_udp_recvfrom);
627 skb_free_datagram_locked(svsk->sk_sk, skb);
628 return 0;
629 }
630
631 static int
632 svc_udp_sendto(struct svc_rqst *rqstp)
633 {
634 int error;
635
636 error = svc_sendto(rqstp, &rqstp->rq_res);
637 if (error == -ECONNREFUSED)
638 /* ICMP error on earlier request. */
639 error = svc_sendto(rqstp, &rqstp->rq_res);
640
641 return error;
642 }
643
644 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
645 {
646 }
647
648 static int svc_udp_has_wspace(struct svc_xprt *xprt)
649 {
650 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
651 struct svc_serv *serv = xprt->xpt_server;
652 unsigned long required;
653
654 /*
655 * Set the SOCK_NOSPACE flag before checking the available
656 * sock space.
657 */
658 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
659 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
660 if (required*2 > sock_wspace(svsk->sk_sk))
661 return 0;
662 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
663 return 1;
664 }
665
666 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
667 {
668 BUG();
669 return NULL;
670 }
671
672 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
673 {
674 }
675
676 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
677 struct net *net,
678 struct sockaddr *sa, int salen,
679 int flags)
680 {
681 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
682 }
683
684 static struct svc_xprt_ops svc_udp_ops = {
685 .xpo_create = svc_udp_create,
686 .xpo_recvfrom = svc_udp_recvfrom,
687 .xpo_sendto = svc_udp_sendto,
688 .xpo_release_rqst = svc_release_skb,
689 .xpo_detach = svc_sock_detach,
690 .xpo_free = svc_sock_free,
691 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
692 .xpo_has_wspace = svc_udp_has_wspace,
693 .xpo_accept = svc_udp_accept,
694 .xpo_secure_port = svc_sock_secure_port,
695 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
696 };
697
698 static struct svc_xprt_class svc_udp_class = {
699 .xcl_name = "udp",
700 .xcl_owner = THIS_MODULE,
701 .xcl_ops = &svc_udp_ops,
702 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
703 .xcl_ident = XPRT_TRANSPORT_UDP,
704 };
705
706 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
707 {
708 int err, level, optname, one = 1;
709
710 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
711 &svsk->sk_xprt, serv);
712 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
713 svsk->sk_sk->sk_data_ready = svc_data_ready;
714 svsk->sk_sk->sk_write_space = svc_write_space;
715
716 /* initialise setting must have enough space to
717 * receive and respond to one request.
718 * svc_udp_recvfrom will re-adjust if necessary
719 */
720 svc_sock_setbufsize(svsk->sk_sock,
721 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
722 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
723
724 /* data might have come in before data_ready set up */
725 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
726 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
727
728 /* make sure we get destination address info */
729 switch (svsk->sk_sk->sk_family) {
730 case AF_INET:
731 level = SOL_IP;
732 optname = IP_PKTINFO;
733 break;
734 case AF_INET6:
735 level = SOL_IPV6;
736 optname = IPV6_RECVPKTINFO;
737 break;
738 default:
739 BUG();
740 }
741 err = kernel_setsockopt(svsk->sk_sock, level, optname,
742 (char *)&one, sizeof(one));
743 dprintk("svc: kernel_setsockopt returned %d\n", err);
744 }
745
746 /*
747 * A data_ready event on a listening socket means there's a connection
748 * pending. Do not use state_change as a substitute for it.
749 */
750 static void svc_tcp_listen_data_ready(struct sock *sk)
751 {
752 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
753
754 dprintk("svc: socket %p TCP (listen) state change %d\n",
755 sk, sk->sk_state);
756
757 if (svsk) {
758 /* Refer to svc_setup_socket() for details. */
759 rmb();
760 svsk->sk_odata(sk);
761 }
762
763 /*
764 * This callback may called twice when a new connection
765 * is established as a child socket inherits everything
766 * from a parent LISTEN socket.
767 * 1) data_ready method of the parent socket will be called
768 * when one of child sockets become ESTABLISHED.
769 * 2) data_ready method of the child socket may be called
770 * when it receives data before the socket is accepted.
771 * In case of 2, we should ignore it silently.
772 */
773 if (sk->sk_state == TCP_LISTEN) {
774 if (svsk) {
775 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
776 svc_xprt_enqueue(&svsk->sk_xprt);
777 } else
778 printk("svc: socket %p: no user data\n", sk);
779 }
780 }
781
782 /*
783 * A state change on a connected socket means it's dying or dead.
784 */
785 static void svc_tcp_state_change(struct sock *sk)
786 {
787 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
788
789 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
790 sk, sk->sk_state, sk->sk_user_data);
791
792 if (!svsk)
793 printk("svc: socket %p: no user data\n", sk);
794 else {
795 /* Refer to svc_setup_socket() for details. */
796 rmb();
797 svsk->sk_ostate(sk);
798 if (sk->sk_state != TCP_ESTABLISHED) {
799 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
800 svc_xprt_enqueue(&svsk->sk_xprt);
801 }
802 }
803 }
804
805 /*
806 * Accept a TCP connection
807 */
808 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
809 {
810 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
811 struct sockaddr_storage addr;
812 struct sockaddr *sin = (struct sockaddr *) &addr;
813 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
814 struct socket *sock = svsk->sk_sock;
815 struct socket *newsock;
816 struct svc_sock *newsvsk;
817 int err, slen;
818 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
819
820 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
821 if (!sock)
822 return NULL;
823
824 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
825 err = kernel_accept(sock, &newsock, O_NONBLOCK);
826 if (err < 0) {
827 if (err == -ENOMEM)
828 printk(KERN_WARNING "%s: no more sockets!\n",
829 serv->sv_name);
830 else if (err != -EAGAIN)
831 net_warn_ratelimited("%s: accept failed (err %d)!\n",
832 serv->sv_name, -err);
833 return NULL;
834 }
835 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
836
837 err = kernel_getpeername(newsock, sin, &slen);
838 if (err < 0) {
839 net_warn_ratelimited("%s: peername failed (err %d)!\n",
840 serv->sv_name, -err);
841 goto failed; /* aborted connection or whatever */
842 }
843
844 /* Ideally, we would want to reject connections from unauthorized
845 * hosts here, but when we get encryption, the IP of the host won't
846 * tell us anything. For now just warn about unpriv connections.
847 */
848 if (!svc_port_is_privileged(sin)) {
849 dprintk("%s: connect from unprivileged port: %s\n",
850 serv->sv_name,
851 __svc_print_addr(sin, buf, sizeof(buf)));
852 }
853 dprintk("%s: connect from %s\n", serv->sv_name,
854 __svc_print_addr(sin, buf, sizeof(buf)));
855
856 /* Reset the inherited callbacks before calling svc_setup_socket */
857 newsock->sk->sk_state_change = svsk->sk_ostate;
858 newsock->sk->sk_data_ready = svsk->sk_odata;
859 newsock->sk->sk_write_space = svsk->sk_owspace;
860
861 /* make sure that a write doesn't block forever when
862 * low on memory
863 */
864 newsock->sk->sk_sndtimeo = HZ*30;
865
866 newsvsk = svc_setup_socket(serv, newsock,
867 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
868 if (IS_ERR(newsvsk))
869 goto failed;
870 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
871 err = kernel_getsockname(newsock, sin, &slen);
872 if (unlikely(err < 0)) {
873 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
874 slen = offsetof(struct sockaddr, sa_data);
875 }
876 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
877
878 if (sock_is_loopback(newsock->sk))
879 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
880 else
881 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
882 if (serv->sv_stats)
883 serv->sv_stats->nettcpconn++;
884
885 return &newsvsk->sk_xprt;
886
887 failed:
888 sock_release(newsock);
889 return NULL;
890 }
891
892 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
893 {
894 unsigned int i, len, npages;
895
896 if (svsk->sk_datalen == 0)
897 return 0;
898 len = svsk->sk_datalen;
899 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
900 for (i = 0; i < npages; i++) {
901 if (rqstp->rq_pages[i] != NULL)
902 put_page(rqstp->rq_pages[i]);
903 BUG_ON(svsk->sk_pages[i] == NULL);
904 rqstp->rq_pages[i] = svsk->sk_pages[i];
905 svsk->sk_pages[i] = NULL;
906 }
907 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
908 return len;
909 }
910
911 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
912 {
913 unsigned int i, len, npages;
914
915 if (svsk->sk_datalen == 0)
916 return;
917 len = svsk->sk_datalen;
918 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
919 for (i = 0; i < npages; i++) {
920 svsk->sk_pages[i] = rqstp->rq_pages[i];
921 rqstp->rq_pages[i] = NULL;
922 }
923 }
924
925 static void svc_tcp_clear_pages(struct svc_sock *svsk)
926 {
927 unsigned int i, len, npages;
928
929 if (svsk->sk_datalen == 0)
930 goto out;
931 len = svsk->sk_datalen;
932 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
933 for (i = 0; i < npages; i++) {
934 if (svsk->sk_pages[i] == NULL) {
935 WARN_ON_ONCE(1);
936 continue;
937 }
938 put_page(svsk->sk_pages[i]);
939 svsk->sk_pages[i] = NULL;
940 }
941 out:
942 svsk->sk_tcplen = 0;
943 svsk->sk_datalen = 0;
944 }
945
946 /*
947 * Receive fragment record header.
948 * If we haven't gotten the record length yet, get the next four bytes.
949 */
950 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
951 {
952 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
953 unsigned int want;
954 int len;
955
956 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
957 struct kvec iov;
958
959 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
960 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
961 iov.iov_len = want;
962 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
963 goto error;
964 svsk->sk_tcplen += len;
965
966 if (len < want) {
967 dprintk("svc: short recvfrom while reading record "
968 "length (%d of %d)\n", len, want);
969 return -EAGAIN;
970 }
971
972 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
973 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
974 serv->sv_max_mesg) {
975 net_notice_ratelimited("RPC: fragment too large: %d\n",
976 svc_sock_reclen(svsk));
977 goto err_delete;
978 }
979 }
980
981 return svc_sock_reclen(svsk);
982 error:
983 dprintk("RPC: TCP recv_record got %d\n", len);
984 return len;
985 err_delete:
986 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
987 return -EAGAIN;
988 }
989
990 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
991 {
992 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
993 struct rpc_rqst *req = NULL;
994 struct kvec *src, *dst;
995 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
996 __be32 xid;
997 __be32 calldir;
998
999 xid = *p++;
1000 calldir = *p;
1001
1002 if (!bc_xprt)
1003 return -EAGAIN;
1004 spin_lock_bh(&bc_xprt->transport_lock);
1005 req = xprt_lookup_rqst(bc_xprt, xid);
1006 if (!req)
1007 goto unlock_notfound;
1008
1009 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1010 /*
1011 * XXX!: cheating for now! Only copying HEAD.
1012 * But we know this is good enough for now (in fact, for any
1013 * callback reply in the forseeable future).
1014 */
1015 dst = &req->rq_private_buf.head[0];
1016 src = &rqstp->rq_arg.head[0];
1017 if (dst->iov_len < src->iov_len)
1018 goto unlock_eagain; /* whatever; just giving up. */
1019 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1020 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1021 rqstp->rq_arg.len = 0;
1022 spin_unlock_bh(&bc_xprt->transport_lock);
1023 return 0;
1024 unlock_notfound:
1025 printk(KERN_NOTICE
1026 "%s: Got unrecognized reply: "
1027 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1028 __func__, ntohl(calldir),
1029 bc_xprt, ntohl(xid));
1030 unlock_eagain:
1031 spin_unlock_bh(&bc_xprt->transport_lock);
1032 return -EAGAIN;
1033 }
1034
1035 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1036 {
1037 int i = 0;
1038 int t = 0;
1039
1040 while (t < len) {
1041 vec[i].iov_base = page_address(pages[i]);
1042 vec[i].iov_len = PAGE_SIZE;
1043 i++;
1044 t += PAGE_SIZE;
1045 }
1046 return i;
1047 }
1048
1049 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1050 {
1051 /* If we have more data, signal svc_xprt_enqueue() to try again */
1052 dprintk("svc: TCP %s record (%d bytes)\n",
1053 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1054 svc_sock_reclen(svsk));
1055 svsk->sk_tcplen = 0;
1056 svsk->sk_reclen = 0;
1057 }
1058
1059 /*
1060 * Receive data from a TCP socket.
1061 */
1062 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1063 {
1064 struct svc_sock *svsk =
1065 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1066 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1067 int len;
1068 struct kvec *vec;
1069 unsigned int want, base;
1070 __be32 *p;
1071 __be32 calldir;
1072 int pnum;
1073
1074 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1075 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1076 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1077 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1078
1079 len = svc_tcp_recv_record(svsk, rqstp);
1080 if (len < 0)
1081 goto error;
1082
1083 base = svc_tcp_restore_pages(svsk, rqstp);
1084 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1085
1086 vec = rqstp->rq_vec;
1087
1088 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1089 svsk->sk_datalen + want);
1090
1091 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1092 rqstp->rq_next_page = rqstp->rq_respages + 1;
1093
1094 /* Now receive data */
1095 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1096 if (len >= 0) {
1097 svsk->sk_tcplen += len;
1098 svsk->sk_datalen += len;
1099 }
1100 if (len != want || !svc_sock_final_rec(svsk)) {
1101 svc_tcp_save_pages(svsk, rqstp);
1102 if (len < 0 && len != -EAGAIN)
1103 goto err_delete;
1104 if (len == want)
1105 svc_tcp_fragment_received(svsk);
1106 else
1107 dprintk("svc: incomplete TCP record (%d of %d)\n",
1108 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1109 svc_sock_reclen(svsk));
1110 goto err_noclose;
1111 }
1112
1113 if (svsk->sk_datalen < 8) {
1114 svsk->sk_datalen = 0;
1115 goto err_delete; /* client is nuts. */
1116 }
1117
1118 rqstp->rq_arg.len = svsk->sk_datalen;
1119 rqstp->rq_arg.page_base = 0;
1120 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1121 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1122 rqstp->rq_arg.page_len = 0;
1123 } else
1124 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1125
1126 rqstp->rq_xprt_ctxt = NULL;
1127 rqstp->rq_prot = IPPROTO_TCP;
1128 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1129 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1130 else
1131 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1132
1133 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1134 calldir = p[1];
1135 if (calldir)
1136 len = receive_cb_reply(svsk, rqstp);
1137
1138 /* Reset TCP read info */
1139 svsk->sk_datalen = 0;
1140 svc_tcp_fragment_received(svsk);
1141
1142 if (len < 0)
1143 goto error;
1144
1145 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1146 if (serv->sv_stats)
1147 serv->sv_stats->nettcpcnt++;
1148
1149 return rqstp->rq_arg.len;
1150
1151 error:
1152 if (len != -EAGAIN)
1153 goto err_delete;
1154 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1155 return 0;
1156 err_delete:
1157 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1158 svsk->sk_xprt.xpt_server->sv_name, -len);
1159 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1160 err_noclose:
1161 return 0; /* record not complete */
1162 }
1163
1164 /*
1165 * Send out data on TCP socket.
1166 */
1167 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1168 {
1169 struct xdr_buf *xbufp = &rqstp->rq_res;
1170 int sent;
1171 __be32 reclen;
1172
1173 /* Set up the first element of the reply kvec.
1174 * Any other kvecs that may be in use have been taken
1175 * care of by the server implementation itself.
1176 */
1177 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1178 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1179
1180 sent = svc_sendto(rqstp, &rqstp->rq_res);
1181 if (sent != xbufp->len) {
1182 printk(KERN_NOTICE
1183 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1184 "- shutting down socket\n",
1185 rqstp->rq_xprt->xpt_server->sv_name,
1186 (sent<0)?"got error":"sent only",
1187 sent, xbufp->len);
1188 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1189 svc_xprt_enqueue(rqstp->rq_xprt);
1190 sent = -EAGAIN;
1191 }
1192 return sent;
1193 }
1194
1195 /*
1196 * Setup response header. TCP has a 4B record length field.
1197 */
1198 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1199 {
1200 struct kvec *resv = &rqstp->rq_res.head[0];
1201
1202 /* tcp needs a space for the record length... */
1203 svc_putnl(resv, 0);
1204 }
1205
1206 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1207 struct net *net,
1208 struct sockaddr *sa, int salen,
1209 int flags)
1210 {
1211 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1212 }
1213
1214 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1215 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1216 struct net *, struct sockaddr *,
1217 int, int);
1218 static void svc_bc_sock_free(struct svc_xprt *xprt);
1219
1220 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1221 struct net *net,
1222 struct sockaddr *sa, int salen,
1223 int flags)
1224 {
1225 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1226 }
1227
1228 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1229 {
1230 }
1231
1232 static struct svc_xprt_ops svc_tcp_bc_ops = {
1233 .xpo_create = svc_bc_tcp_create,
1234 .xpo_detach = svc_bc_tcp_sock_detach,
1235 .xpo_free = svc_bc_sock_free,
1236 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1237 .xpo_secure_port = svc_sock_secure_port,
1238 };
1239
1240 static struct svc_xprt_class svc_tcp_bc_class = {
1241 .xcl_name = "tcp-bc",
1242 .xcl_owner = THIS_MODULE,
1243 .xcl_ops = &svc_tcp_bc_ops,
1244 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1245 };
1246
1247 static void svc_init_bc_xprt_sock(void)
1248 {
1249 svc_reg_xprt_class(&svc_tcp_bc_class);
1250 }
1251
1252 static void svc_cleanup_bc_xprt_sock(void)
1253 {
1254 svc_unreg_xprt_class(&svc_tcp_bc_class);
1255 }
1256 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1257 static void svc_init_bc_xprt_sock(void)
1258 {
1259 }
1260
1261 static void svc_cleanup_bc_xprt_sock(void)
1262 {
1263 }
1264 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1265
1266 static struct svc_xprt_ops svc_tcp_ops = {
1267 .xpo_create = svc_tcp_create,
1268 .xpo_recvfrom = svc_tcp_recvfrom,
1269 .xpo_sendto = svc_tcp_sendto,
1270 .xpo_release_rqst = svc_release_skb,
1271 .xpo_detach = svc_tcp_sock_detach,
1272 .xpo_free = svc_sock_free,
1273 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1274 .xpo_has_wspace = svc_tcp_has_wspace,
1275 .xpo_accept = svc_tcp_accept,
1276 .xpo_secure_port = svc_sock_secure_port,
1277 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1278 };
1279
1280 static struct svc_xprt_class svc_tcp_class = {
1281 .xcl_name = "tcp",
1282 .xcl_owner = THIS_MODULE,
1283 .xcl_ops = &svc_tcp_ops,
1284 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1285 .xcl_ident = XPRT_TRANSPORT_TCP,
1286 };
1287
1288 void svc_init_xprt_sock(void)
1289 {
1290 svc_reg_xprt_class(&svc_tcp_class);
1291 svc_reg_xprt_class(&svc_udp_class);
1292 svc_init_bc_xprt_sock();
1293 }
1294
1295 void svc_cleanup_xprt_sock(void)
1296 {
1297 svc_unreg_xprt_class(&svc_tcp_class);
1298 svc_unreg_xprt_class(&svc_udp_class);
1299 svc_cleanup_bc_xprt_sock();
1300 }
1301
1302 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1303 {
1304 struct sock *sk = svsk->sk_sk;
1305
1306 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1307 &svsk->sk_xprt, serv);
1308 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1309 if (sk->sk_state == TCP_LISTEN) {
1310 dprintk("setting up TCP socket for listening\n");
1311 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1312 sk->sk_data_ready = svc_tcp_listen_data_ready;
1313 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1314 } else {
1315 dprintk("setting up TCP socket for reading\n");
1316 sk->sk_state_change = svc_tcp_state_change;
1317 sk->sk_data_ready = svc_data_ready;
1318 sk->sk_write_space = svc_write_space;
1319
1320 svsk->sk_reclen = 0;
1321 svsk->sk_tcplen = 0;
1322 svsk->sk_datalen = 0;
1323 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1324
1325 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1326
1327 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1328 switch (sk->sk_state) {
1329 case TCP_SYN_RECV:
1330 case TCP_ESTABLISHED:
1331 break;
1332 default:
1333 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1334 }
1335 }
1336 }
1337
1338 void svc_sock_update_bufs(struct svc_serv *serv)
1339 {
1340 /*
1341 * The number of server threads has changed. Update
1342 * rcvbuf and sndbuf accordingly on all sockets
1343 */
1344 struct svc_sock *svsk;
1345
1346 spin_lock_bh(&serv->sv_lock);
1347 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1348 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1349 spin_unlock_bh(&serv->sv_lock);
1350 }
1351 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1352
1353 /*
1354 * Initialize socket for RPC use and create svc_sock struct
1355 */
1356 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1357 struct socket *sock,
1358 int flags)
1359 {
1360 struct svc_sock *svsk;
1361 struct sock *inet;
1362 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1363 int err = 0;
1364
1365 dprintk("svc: svc_setup_socket %p\n", sock);
1366 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1367 if (!svsk)
1368 return ERR_PTR(-ENOMEM);
1369
1370 inet = sock->sk;
1371
1372 /* Register socket with portmapper */
1373 if (pmap_register)
1374 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1375 inet->sk_protocol,
1376 ntohs(inet_sk(inet)->inet_sport));
1377
1378 if (err < 0) {
1379 kfree(svsk);
1380 return ERR_PTR(err);
1381 }
1382
1383 svsk->sk_sock = sock;
1384 svsk->sk_sk = inet;
1385 svsk->sk_ostate = inet->sk_state_change;
1386 svsk->sk_odata = inet->sk_data_ready;
1387 svsk->sk_owspace = inet->sk_write_space;
1388 /*
1389 * This barrier is necessary in order to prevent race condition
1390 * with svc_data_ready(), svc_listen_data_ready() and others
1391 * when calling callbacks above.
1392 */
1393 wmb();
1394 inet->sk_user_data = svsk;
1395
1396 /* Initialize the socket */
1397 if (sock->type == SOCK_DGRAM)
1398 svc_udp_init(svsk, serv);
1399 else
1400 svc_tcp_init(svsk, serv);
1401
1402 dprintk("svc: svc_setup_socket created %p (inet %p), "
1403 "listen %d close %d\n",
1404 svsk, svsk->sk_sk,
1405 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1406 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1407
1408 return svsk;
1409 }
1410
1411 bool svc_alien_sock(struct net *net, int fd)
1412 {
1413 int err;
1414 struct socket *sock = sockfd_lookup(fd, &err);
1415 bool ret = false;
1416
1417 if (!sock)
1418 goto out;
1419 if (sock_net(sock->sk) != net)
1420 ret = true;
1421 sockfd_put(sock);
1422 out:
1423 return ret;
1424 }
1425 EXPORT_SYMBOL_GPL(svc_alien_sock);
1426
1427 /**
1428 * svc_addsock - add a listener socket to an RPC service
1429 * @serv: pointer to RPC service to which to add a new listener
1430 * @fd: file descriptor of the new listener
1431 * @name_return: pointer to buffer to fill in with name of listener
1432 * @len: size of the buffer
1433 *
1434 * Fills in socket name and returns positive length of name if successful.
1435 * Name is terminated with '\n'. On error, returns a negative errno
1436 * value.
1437 */
1438 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1439 const size_t len)
1440 {
1441 int err = 0;
1442 struct socket *so = sockfd_lookup(fd, &err);
1443 struct svc_sock *svsk = NULL;
1444 struct sockaddr_storage addr;
1445 struct sockaddr *sin = (struct sockaddr *)&addr;
1446 int salen;
1447
1448 if (!so)
1449 return err;
1450 err = -EAFNOSUPPORT;
1451 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1452 goto out;
1453 err = -EPROTONOSUPPORT;
1454 if (so->sk->sk_protocol != IPPROTO_TCP &&
1455 so->sk->sk_protocol != IPPROTO_UDP)
1456 goto out;
1457 err = -EISCONN;
1458 if (so->state > SS_UNCONNECTED)
1459 goto out;
1460 err = -ENOENT;
1461 if (!try_module_get(THIS_MODULE))
1462 goto out;
1463 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1464 if (IS_ERR(svsk)) {
1465 module_put(THIS_MODULE);
1466 err = PTR_ERR(svsk);
1467 goto out;
1468 }
1469 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1470 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1471 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1472 return svc_one_sock_name(svsk, name_return, len);
1473 out:
1474 sockfd_put(so);
1475 return err;
1476 }
1477 EXPORT_SYMBOL_GPL(svc_addsock);
1478
1479 /*
1480 * Create socket for RPC service.
1481 */
1482 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1483 int protocol,
1484 struct net *net,
1485 struct sockaddr *sin, int len,
1486 int flags)
1487 {
1488 struct svc_sock *svsk;
1489 struct socket *sock;
1490 int error;
1491 int type;
1492 struct sockaddr_storage addr;
1493 struct sockaddr *newsin = (struct sockaddr *)&addr;
1494 int newlen;
1495 int family;
1496 int val;
1497 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1498
1499 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1500 serv->sv_program->pg_name, protocol,
1501 __svc_print_addr(sin, buf, sizeof(buf)));
1502
1503 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1504 printk(KERN_WARNING "svc: only UDP and TCP "
1505 "sockets supported\n");
1506 return ERR_PTR(-EINVAL);
1507 }
1508
1509 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1510 switch (sin->sa_family) {
1511 case AF_INET6:
1512 family = PF_INET6;
1513 break;
1514 case AF_INET:
1515 family = PF_INET;
1516 break;
1517 default:
1518 return ERR_PTR(-EINVAL);
1519 }
1520
1521 error = __sock_create(net, family, type, protocol, &sock, 1);
1522 if (error < 0)
1523 return ERR_PTR(error);
1524
1525 svc_reclassify_socket(sock);
1526
1527 /*
1528 * If this is an PF_INET6 listener, we want to avoid
1529 * getting requests from IPv4 remotes. Those should
1530 * be shunted to a PF_INET listener via rpcbind.
1531 */
1532 val = 1;
1533 if (family == PF_INET6)
1534 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1535 (char *)&val, sizeof(val));
1536
1537 if (type == SOCK_STREAM)
1538 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1539 error = kernel_bind(sock, sin, len);
1540 if (error < 0)
1541 goto bummer;
1542
1543 newlen = len;
1544 error = kernel_getsockname(sock, newsin, &newlen);
1545 if (error < 0)
1546 goto bummer;
1547
1548 if (protocol == IPPROTO_TCP) {
1549 if ((error = kernel_listen(sock, 64)) < 0)
1550 goto bummer;
1551 }
1552
1553 svsk = svc_setup_socket(serv, sock, flags);
1554 if (IS_ERR(svsk)) {
1555 error = PTR_ERR(svsk);
1556 goto bummer;
1557 }
1558 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1559 return (struct svc_xprt *)svsk;
1560 bummer:
1561 dprintk("svc: svc_create_socket error = %d\n", -error);
1562 sock_release(sock);
1563 return ERR_PTR(error);
1564 }
1565
1566 /*
1567 * Detach the svc_sock from the socket so that no
1568 * more callbacks occur.
1569 */
1570 static void svc_sock_detach(struct svc_xprt *xprt)
1571 {
1572 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1573 struct sock *sk = svsk->sk_sk;
1574
1575 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1576
1577 /* put back the old socket callbacks */
1578 lock_sock(sk);
1579 sk->sk_state_change = svsk->sk_ostate;
1580 sk->sk_data_ready = svsk->sk_odata;
1581 sk->sk_write_space = svsk->sk_owspace;
1582 sk->sk_user_data = NULL;
1583 release_sock(sk);
1584 }
1585
1586 /*
1587 * Disconnect the socket, and reset the callbacks
1588 */
1589 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1590 {
1591 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1592
1593 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1594
1595 svc_sock_detach(xprt);
1596
1597 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1598 svc_tcp_clear_pages(svsk);
1599 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1600 }
1601 }
1602
1603 /*
1604 * Free the svc_sock's socket resources and the svc_sock itself.
1605 */
1606 static void svc_sock_free(struct svc_xprt *xprt)
1607 {
1608 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1609 dprintk("svc: svc_sock_free(%p)\n", svsk);
1610
1611 if (svsk->sk_sock->file)
1612 sockfd_put(svsk->sk_sock);
1613 else
1614 sock_release(svsk->sk_sock);
1615 kfree(svsk);
1616 }
1617
1618 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1619 /*
1620 * Create a back channel svc_xprt which shares the fore channel socket.
1621 */
1622 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1623 int protocol,
1624 struct net *net,
1625 struct sockaddr *sin, int len,
1626 int flags)
1627 {
1628 struct svc_sock *svsk;
1629 struct svc_xprt *xprt;
1630
1631 if (protocol != IPPROTO_TCP) {
1632 printk(KERN_WARNING "svc: only TCP sockets"
1633 " supported on shared back channel\n");
1634 return ERR_PTR(-EINVAL);
1635 }
1636
1637 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1638 if (!svsk)
1639 return ERR_PTR(-ENOMEM);
1640
1641 xprt = &svsk->sk_xprt;
1642 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1643
1644 serv->sv_bc_xprt = xprt;
1645
1646 return xprt;
1647 }
1648
1649 /*
1650 * Free a back channel svc_sock.
1651 */
1652 static void svc_bc_sock_free(struct svc_xprt *xprt)
1653 {
1654 if (xprt)
1655 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1656 }
1657 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support