?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
powerpc/pseries/pci: Remove obsolete SW invalidate
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blobdadfec66dbd8abd301fa3ac307431dabe7999ec6
1 /*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
9 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
28 #include <linux/in.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
38 #include <net/sock.h>
39 #include <net/checksum.h>
40 #include <net/ip.h>
41 #include <net/ipv6.h>
42 #include <net/tcp.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <trace/events/skb.h>
47
48 #include <linux/sunrpc/types.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/xdr.h>
51 #include <linux/sunrpc/msg_prot.h>
52 #include <linux/sunrpc/svcsock.h>
53 #include <linux/sunrpc/stats.h>
54 #include <linux/sunrpc/xprt.h>
55
56 #include "sunrpc.h"
57
58 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
59
60
61 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
62 int flags);
63 static void svc_udp_data_ready(struct sock *);
64 static int svc_udp_recvfrom(struct svc_rqst *);
65 static int svc_udp_sendto(struct svc_rqst *);
66 static void svc_sock_detach(struct svc_xprt *);
67 static void svc_tcp_sock_detach(struct svc_xprt *);
68 static void svc_sock_free(struct svc_xprt *);
69
70 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
74 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77 static void svc_bc_sock_free(struct svc_xprt *xprt);
78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
79
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 static struct lock_class_key svc_key[2];
82 static struct lock_class_key svc_slock_key[2];
83
84 static void svc_reclassify_socket(struct socket *sock)
85 {
86 struct sock *sk = sock->sk;
87
88 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 union svc_pktinfo_u {
134 struct in_pktinfo pkti;
135 struct in6_pktinfo pkti6;
136 };
137 #define SVC_PKTINFO_SPACE \
138 CMSG_SPACE(sizeof(union svc_pktinfo_u))
139
140 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
141 {
142 struct svc_sock *svsk =
143 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
144 switch (svsk->sk_sk->sk_family) {
145 case AF_INET: {
146 struct in_pktinfo *pki = CMSG_DATA(cmh);
147
148 cmh->cmsg_level = SOL_IP;
149 cmh->cmsg_type = IP_PKTINFO;
150 pki->ipi_ifindex = 0;
151 pki->ipi_spec_dst.s_addr =
152 svc_daddr_in(rqstp)->sin_addr.s_addr;
153 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
154 }
155 break;
156
157 case AF_INET6: {
158 struct in6_pktinfo *pki = CMSG_DATA(cmh);
159 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
160
161 cmh->cmsg_level = SOL_IPV6;
162 cmh->cmsg_type = IPV6_PKTINFO;
163 pki->ipi6_ifindex = daddr->sin6_scope_id;
164 pki->ipi6_addr = daddr->sin6_addr;
165 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
166 }
167 break;
168 }
169 }
170
171 /*
172 * send routine intended to be shared by the fore- and back-channel
173 */
174 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
175 struct page *headpage, unsigned long headoffset,
176 struct page *tailpage, unsigned long tailoffset)
177 {
178 int result;
179 int size;
180 struct page **ppage = xdr->pages;
181 size_t base = xdr->page_base;
182 unsigned int pglen = xdr->page_len;
183 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
184 int slen;
185 int len = 0;
186
187 slen = xdr->len;
188
189 /* send head */
190 if (slen == xdr->head[0].iov_len)
191 flags = 0;
192 len = kernel_sendpage(sock, headpage, headoffset,
193 xdr->head[0].iov_len, flags);
194 if (len != xdr->head[0].iov_len)
195 goto out;
196 slen -= xdr->head[0].iov_len;
197 if (slen == 0)
198 goto out;
199
200 /* send page data */
201 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
202 while (pglen > 0) {
203 if (slen == size)
204 flags = 0;
205 result = kernel_sendpage(sock, *ppage, base, size, flags);
206 if (result > 0)
207 len += result;
208 if (result != size)
209 goto out;
210 slen -= size;
211 pglen -= size;
212 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
213 base = 0;
214 ppage++;
215 }
216
217 /* send tail */
218 if (xdr->tail[0].iov_len) {
219 result = kernel_sendpage(sock, tailpage, tailoffset,
220 xdr->tail[0].iov_len, 0);
221 if (result > 0)
222 len += result;
223 }
224
225 out:
226 return len;
227 }
228
229
230 /*
231 * Generic sendto routine
232 */
233 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
234 {
235 struct svc_sock *svsk =
236 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
237 struct socket *sock = svsk->sk_sock;
238 union {
239 struct cmsghdr hdr;
240 long all[SVC_PKTINFO_SPACE / sizeof(long)];
241 } buffer;
242 struct cmsghdr *cmh = &buffer.hdr;
243 int len = 0;
244 unsigned long tailoff;
245 unsigned long headoff;
246 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
247
248 if (rqstp->rq_prot == IPPROTO_UDP) {
249 struct msghdr msg = {
250 .msg_name = &rqstp->rq_addr,
251 .msg_namelen = rqstp->rq_addrlen,
252 .msg_control = cmh,
253 .msg_controllen = sizeof(buffer),
254 .msg_flags = MSG_MORE,
255 };
256
257 svc_set_cmsg_data(rqstp, cmh);
258
259 if (sock_sendmsg(sock, &msg) < 0)
260 goto out;
261 }
262
263 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
264 headoff = 0;
265 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
266 rqstp->rq_respages[0], tailoff);
267
268 out:
269 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
270 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
271 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
272
273 return len;
274 }
275
276 /*
277 * Report socket names for nfsdfs
278 */
279 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
280 {
281 const struct sock *sk = svsk->sk_sk;
282 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
283 "udp" : "tcp";
284 int len;
285
286 switch (sk->sk_family) {
287 case PF_INET:
288 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
289 proto_name,
290 &inet_sk(sk)->inet_rcv_saddr,
291 inet_sk(sk)->inet_num);
292 break;
293 #if IS_ENABLED(CONFIG_IPV6)
294 case PF_INET6:
295 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
296 proto_name,
297 &sk->sk_v6_rcv_saddr,
298 inet_sk(sk)->inet_num);
299 break;
300 #endif
301 default:
302 len = snprintf(buf, remaining, "*unknown-%d*\n",
303 sk->sk_family);
304 }
305
306 if (len >= remaining) {
307 *buf = '\0';
308 return -ENAMETOOLONG;
309 }
310 return len;
311 }
312
313 /*
314 * Generic recvfrom routine.
315 */
316 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
317 int buflen)
318 {
319 struct svc_sock *svsk =
320 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
321 struct msghdr msg = {
322 .msg_flags = MSG_DONTWAIT,
323 };
324 int len;
325
326 rqstp->rq_xprt_hlen = 0;
327
328 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
329 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
330 msg.msg_flags);
331 /* If we read a full record, then assume there may be more
332 * data to read (stream based sockets only!)
333 */
334 if (len == buflen)
335 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
336
337 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
338 svsk, iov[0].iov_base, iov[0].iov_len, len);
339 return len;
340 }
341
342 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
343 struct kvec *iov, int nr,
344 int buflen, unsigned int base)
345 {
346 size_t save_iovlen;
347 void *save_iovbase;
348 unsigned int i;
349 int ret;
350
351 if (base == 0)
352 return svc_recvfrom(rqstp, iov, nr, buflen);
353
354 for (i = 0; i < nr; i++) {
355 if (iov[i].iov_len > base)
356 break;
357 base -= iov[i].iov_len;
358 }
359 save_iovlen = iov[i].iov_len;
360 save_iovbase = iov[i].iov_base;
361 iov[i].iov_len -= base;
362 iov[i].iov_base += base;
363 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
364 iov[i].iov_len = save_iovlen;
365 iov[i].iov_base = save_iovbase;
366 return ret;
367 }
368
369 /*
370 * Set socket snd and rcv buffer lengths
371 */
372 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
373 unsigned int rcv)
374 {
375 #if 0
376 mm_segment_t oldfs;
377 oldfs = get_fs(); set_fs(KERNEL_DS);
378 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
379 (char*)&snd, sizeof(snd));
380 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
381 (char*)&rcv, sizeof(rcv));
382 #else
383 /* sock_setsockopt limits use to sysctl_?mem_max,
384 * which isn't acceptable. Until that is made conditional
385 * on not having CAP_SYS_RESOURCE or similar, we go direct...
386 * DaveM said I could!
387 */
388 lock_sock(sock->sk);
389 sock->sk->sk_sndbuf = snd * 2;
390 sock->sk->sk_rcvbuf = rcv * 2;
391 sock->sk->sk_write_space(sock->sk);
392 release_sock(sock->sk);
393 #endif
394 }
395
396 static int svc_sock_secure_port(struct svc_rqst *rqstp)
397 {
398 return svc_port_is_privileged(svc_addr(rqstp));
399 }
400
401 static bool sunrpc_waitqueue_active(wait_queue_head_t *wq)
402 {
403 if (!wq)
404 return false;
405 /*
406 * There should normally be a memory * barrier here--see
407 * wq_has_sleeper().
408 *
409 * It appears that isn't currently necessary, though, basically
410 * because callers all appear to have sufficient memory barriers
411 * between the time the relevant change is made and the
412 * time they call these callbacks.
413 *
414 * The nfsd code itself doesn't actually explicitly wait on
415 * these waitqueues, but it may wait on them for example in
416 * sendpage() or sendmsg() calls. (And those may be the only
417 * places, since it it uses nonblocking reads.)
418 *
419 * Maybe we should add the memory barriers anyway, but these are
420 * hot paths so we'd need to be convinced there's no sigificant
421 * penalty.
422 */
423 return waitqueue_active(wq);
424 }
425
426 /*
427 * INET callback when data has been received on the socket.
428 */
429 static void svc_udp_data_ready(struct sock *sk)
430 {
431 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
432 wait_queue_head_t *wq = sk_sleep(sk);
433
434 if (svsk) {
435 dprintk("svc: socket %p(inet %p), busy=%d\n",
436 svsk, sk,
437 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
438 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
439 svc_xprt_enqueue(&svsk->sk_xprt);
440 }
441 if (sunrpc_waitqueue_active(wq))
442 wake_up_interruptible(wq);
443 }
444
445 /*
446 * INET callback when space is newly available on the socket.
447 */
448 static void svc_write_space(struct sock *sk)
449 {
450 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
451 wait_queue_head_t *wq = sk_sleep(sk);
452
453 if (svsk) {
454 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
455 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
456 svc_xprt_enqueue(&svsk->sk_xprt);
457 }
458
459 if (sunrpc_waitqueue_active(wq)) {
460 dprintk("RPC svc_write_space: someone sleeping on %p\n",
461 svsk);
462 wake_up_interruptible(wq);
463 }
464 }
465
466 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
467 {
468 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
469 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
470 int required;
471
472 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
473 return 1;
474 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
475 if (sk_stream_wspace(svsk->sk_sk) >= required ||
476 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
477 atomic_read(&xprt->xpt_reserved) == 0))
478 return 1;
479 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
480 return 0;
481 }
482
483 static void svc_tcp_write_space(struct sock *sk)
484 {
485 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
486 struct socket *sock = sk->sk_socket;
487
488 if (!sk_stream_is_writeable(sk) || !sock)
489 return;
490 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt))
491 clear_bit(SOCK_NOSPACE, &sock->flags);
492 svc_write_space(sk);
493 }
494
495 static void svc_tcp_adjust_wspace(struct svc_xprt *xprt)
496 {
497 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
498
499 if (svc_tcp_has_wspace(xprt))
500 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
501 }
502
503 /*
504 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
505 */
506 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
507 struct cmsghdr *cmh)
508 {
509 struct in_pktinfo *pki = CMSG_DATA(cmh);
510 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
511
512 if (cmh->cmsg_type != IP_PKTINFO)
513 return 0;
514
515 daddr->sin_family = AF_INET;
516 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
517 return 1;
518 }
519
520 /*
521 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
522 */
523 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
524 struct cmsghdr *cmh)
525 {
526 struct in6_pktinfo *pki = CMSG_DATA(cmh);
527 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
528
529 if (cmh->cmsg_type != IPV6_PKTINFO)
530 return 0;
531
532 daddr->sin6_family = AF_INET6;
533 daddr->sin6_addr = pki->ipi6_addr;
534 daddr->sin6_scope_id = pki->ipi6_ifindex;
535 return 1;
536 }
537
538 /*
539 * Copy the UDP datagram's destination address to the rqstp structure.
540 * The 'destination' address in this case is the address to which the
541 * peer sent the datagram, i.e. our local address. For multihomed
542 * hosts, this can change from msg to msg. Note that only the IP
543 * address changes, the port number should remain the same.
544 */
545 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
546 struct cmsghdr *cmh)
547 {
548 switch (cmh->cmsg_level) {
549 case SOL_IP:
550 return svc_udp_get_dest_address4(rqstp, cmh);
551 case SOL_IPV6:
552 return svc_udp_get_dest_address6(rqstp, cmh);
553 }
554
555 return 0;
556 }
557
558 /*
559 * Receive a datagram from a UDP socket.
560 */
561 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
562 {
563 struct svc_sock *svsk =
564 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
565 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
566 struct sk_buff *skb;
567 union {
568 struct cmsghdr hdr;
569 long all[SVC_PKTINFO_SPACE / sizeof(long)];
570 } buffer;
571 struct cmsghdr *cmh = &buffer.hdr;
572 struct msghdr msg = {
573 .msg_name = svc_addr(rqstp),
574 .msg_control = cmh,
575 .msg_controllen = sizeof(buffer),
576 .msg_flags = MSG_DONTWAIT,
577 };
578 size_t len;
579 int err;
580
581 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
582 /* udp sockets need large rcvbuf as all pending
583 * requests are still in that buffer. sndbuf must
584 * also be large enough that there is enough space
585 * for one reply per thread. We count all threads
586 * rather than threads in a particular pool, which
587 * provides an upper bound on the number of threads
588 * which will access the socket.
589 */
590 svc_sock_setbufsize(svsk->sk_sock,
591 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
592 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
593
594 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
595 skb = NULL;
596 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
597 0, 0, MSG_PEEK | MSG_DONTWAIT);
598 if (err >= 0)
599 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
600
601 if (skb == NULL) {
602 if (err != -EAGAIN) {
603 /* possibly an icmp error */
604 dprintk("svc: recvfrom returned error %d\n", -err);
605 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
606 }
607 return 0;
608 }
609 len = svc_addr_len(svc_addr(rqstp));
610 rqstp->rq_addrlen = len;
611 if (skb->tstamp.tv64 == 0) {
612 skb->tstamp = ktime_get_real();
613 /* Don't enable netstamp, sunrpc doesn't
614 need that much accuracy */
615 }
616 svsk->sk_sk->sk_stamp = skb->tstamp;
617 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
618
619 len = skb->len;
620 rqstp->rq_arg.len = len;
621
622 rqstp->rq_prot = IPPROTO_UDP;
623
624 if (!svc_udp_get_dest_address(rqstp, cmh)) {
625 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
626 cmh->cmsg_level, cmh->cmsg_type);
627 goto out_free;
628 }
629 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
630
631 if (skb_is_nonlinear(skb)) {
632 /* we have to copy */
633 local_bh_disable();
634 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
635 local_bh_enable();
636 /* checksum error */
637 goto out_free;
638 }
639 local_bh_enable();
640 skb_free_datagram_locked(svsk->sk_sk, skb);
641 } else {
642 /* we can use it in-place */
643 rqstp->rq_arg.head[0].iov_base = skb->data;
644 rqstp->rq_arg.head[0].iov_len = len;
645 if (skb_checksum_complete(skb))
646 goto out_free;
647 rqstp->rq_xprt_ctxt = skb;
648 }
649
650 rqstp->rq_arg.page_base = 0;
651 if (len <= rqstp->rq_arg.head[0].iov_len) {
652 rqstp->rq_arg.head[0].iov_len = len;
653 rqstp->rq_arg.page_len = 0;
654 rqstp->rq_respages = rqstp->rq_pages+1;
655 } else {
656 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
657 rqstp->rq_respages = rqstp->rq_pages + 1 +
658 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
659 }
660 rqstp->rq_next_page = rqstp->rq_respages+1;
661
662 if (serv->sv_stats)
663 serv->sv_stats->netudpcnt++;
664
665 return len;
666 out_free:
667 trace_kfree_skb(skb, svc_udp_recvfrom);
668 skb_free_datagram_locked(svsk->sk_sk, skb);
669 return 0;
670 }
671
672 static int
673 svc_udp_sendto(struct svc_rqst *rqstp)
674 {
675 int error;
676
677 error = svc_sendto(rqstp, &rqstp->rq_res);
678 if (error == -ECONNREFUSED)
679 /* ICMP error on earlier request. */
680 error = svc_sendto(rqstp, &rqstp->rq_res);
681
682 return error;
683 }
684
685 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
686 {
687 }
688
689 static int svc_udp_has_wspace(struct svc_xprt *xprt)
690 {
691 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
692 struct svc_serv *serv = xprt->xpt_server;
693 unsigned long required;
694
695 /*
696 * Set the SOCK_NOSPACE flag before checking the available
697 * sock space.
698 */
699 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
700 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
701 if (required*2 > sock_wspace(svsk->sk_sk))
702 return 0;
703 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
704 return 1;
705 }
706
707 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
708 {
709 BUG();
710 return NULL;
711 }
712
713 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
714 struct net *net,
715 struct sockaddr *sa, int salen,
716 int flags)
717 {
718 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
719 }
720
721 static struct svc_xprt_ops svc_udp_ops = {
722 .xpo_create = svc_udp_create,
723 .xpo_recvfrom = svc_udp_recvfrom,
724 .xpo_sendto = svc_udp_sendto,
725 .xpo_release_rqst = svc_release_skb,
726 .xpo_detach = svc_sock_detach,
727 .xpo_free = svc_sock_free,
728 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
729 .xpo_has_wspace = svc_udp_has_wspace,
730 .xpo_accept = svc_udp_accept,
731 .xpo_secure_port = svc_sock_secure_port,
732 };
733
734 static struct svc_xprt_class svc_udp_class = {
735 .xcl_name = "udp",
736 .xcl_owner = THIS_MODULE,
737 .xcl_ops = &svc_udp_ops,
738 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
739 .xcl_ident = XPRT_TRANSPORT_UDP,
740 };
741
742 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
743 {
744 int err, level, optname, one = 1;
745
746 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
747 &svsk->sk_xprt, serv);
748 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
749 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
750 svsk->sk_sk->sk_write_space = svc_write_space;
751
752 /* initialise setting must have enough space to
753 * receive and respond to one request.
754 * svc_udp_recvfrom will re-adjust if necessary
755 */
756 svc_sock_setbufsize(svsk->sk_sock,
757 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
758 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
759
760 /* data might have come in before data_ready set up */
761 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
762 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
763
764 /* make sure we get destination address info */
765 switch (svsk->sk_sk->sk_family) {
766 case AF_INET:
767 level = SOL_IP;
768 optname = IP_PKTINFO;
769 break;
770 case AF_INET6:
771 level = SOL_IPV6;
772 optname = IPV6_RECVPKTINFO;
773 break;
774 default:
775 BUG();
776 }
777 err = kernel_setsockopt(svsk->sk_sock, level, optname,
778 (char *)&one, sizeof(one));
779 dprintk("svc: kernel_setsockopt returned %d\n", err);
780 }
781
782 /*
783 * A data_ready event on a listening socket means there's a connection
784 * pending. Do not use state_change as a substitute for it.
785 */
786 static void svc_tcp_listen_data_ready(struct sock *sk)
787 {
788 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
789 wait_queue_head_t *wq;
790
791 dprintk("svc: socket %p TCP (listen) state change %d\n",
792 sk, sk->sk_state);
793
794 /*
795 * This callback may called twice when a new connection
796 * is established as a child socket inherits everything
797 * from a parent LISTEN socket.
798 * 1) data_ready method of the parent socket will be called
799 * when one of child sockets become ESTABLISHED.
800 * 2) data_ready method of the child socket may be called
801 * when it receives data before the socket is accepted.
802 * In case of 2, we should ignore it silently.
803 */
804 if (sk->sk_state == TCP_LISTEN) {
805 if (svsk) {
806 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
807 svc_xprt_enqueue(&svsk->sk_xprt);
808 } else
809 printk("svc: socket %p: no user data\n", sk);
810 }
811
812 wq = sk_sleep(sk);
813 if (sunrpc_waitqueue_active(wq))
814 wake_up_interruptible_all(wq);
815 }
816
817 /*
818 * A state change on a connected socket means it's dying or dead.
819 */
820 static void svc_tcp_state_change(struct sock *sk)
821 {
822 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
823 wait_queue_head_t *wq = sk_sleep(sk);
824
825 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
826 sk, sk->sk_state, sk->sk_user_data);
827
828 if (!svsk)
829 printk("svc: socket %p: no user data\n", sk);
830 else {
831 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
832 svc_xprt_enqueue(&svsk->sk_xprt);
833 }
834 if (sunrpc_waitqueue_active(wq))
835 wake_up_interruptible_all(wq);
836 }
837
838 static void svc_tcp_data_ready(struct sock *sk)
839 {
840 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
841 wait_queue_head_t *wq = sk_sleep(sk);
842
843 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
844 sk, sk->sk_user_data);
845 if (svsk) {
846 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
847 svc_xprt_enqueue(&svsk->sk_xprt);
848 }
849 if (sunrpc_waitqueue_active(wq))
850 wake_up_interruptible(wq);
851 }
852
853 /*
854 * Accept a TCP connection
855 */
856 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
857 {
858 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
859 struct sockaddr_storage addr;
860 struct sockaddr *sin = (struct sockaddr *) &addr;
861 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
862 struct socket *sock = svsk->sk_sock;
863 struct socket *newsock;
864 struct svc_sock *newsvsk;
865 int err, slen;
866 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
867
868 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
869 if (!sock)
870 return NULL;
871
872 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
873 err = kernel_accept(sock, &newsock, O_NONBLOCK);
874 if (err < 0) {
875 if (err == -ENOMEM)
876 printk(KERN_WARNING "%s: no more sockets!\n",
877 serv->sv_name);
878 else if (err != -EAGAIN)
879 net_warn_ratelimited("%s: accept failed (err %d)!\n",
880 serv->sv_name, -err);
881 return NULL;
882 }
883 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
884
885 err = kernel_getpeername(newsock, sin, &slen);
886 if (err < 0) {
887 net_warn_ratelimited("%s: peername failed (err %d)!\n",
888 serv->sv_name, -err);
889 goto failed; /* aborted connection or whatever */
890 }
891
892 /* Ideally, we would want to reject connections from unauthorized
893 * hosts here, but when we get encryption, the IP of the host won't
894 * tell us anything. For now just warn about unpriv connections.
895 */
896 if (!svc_port_is_privileged(sin)) {
897 dprintk("%s: connect from unprivileged port: %s\n",
898 serv->sv_name,
899 __svc_print_addr(sin, buf, sizeof(buf)));
900 }
901 dprintk("%s: connect from %s\n", serv->sv_name,
902 __svc_print_addr(sin, buf, sizeof(buf)));
903
904 /* make sure that a write doesn't block forever when
905 * low on memory
906 */
907 newsock->sk->sk_sndtimeo = HZ*30;
908
909 newsvsk = svc_setup_socket(serv, newsock,
910 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
911 if (IS_ERR(newsvsk))
912 goto failed;
913 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
914 err = kernel_getsockname(newsock, sin, &slen);
915 if (unlikely(err < 0)) {
916 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
917 slen = offsetof(struct sockaddr, sa_data);
918 }
919 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
920
921 if (sock_is_loopback(newsock->sk))
922 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
923 else
924 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
925 if (serv->sv_stats)
926 serv->sv_stats->nettcpconn++;
927
928 return &newsvsk->sk_xprt;
929
930 failed:
931 sock_release(newsock);
932 return NULL;
933 }
934
935 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
936 {
937 unsigned int i, len, npages;
938
939 if (svsk->sk_datalen == 0)
940 return 0;
941 len = svsk->sk_datalen;
942 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
943 for (i = 0; i < npages; i++) {
944 if (rqstp->rq_pages[i] != NULL)
945 put_page(rqstp->rq_pages[i]);
946 BUG_ON(svsk->sk_pages[i] == NULL);
947 rqstp->rq_pages[i] = svsk->sk_pages[i];
948 svsk->sk_pages[i] = NULL;
949 }
950 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
951 return len;
952 }
953
954 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
955 {
956 unsigned int i, len, npages;
957
958 if (svsk->sk_datalen == 0)
959 return;
960 len = svsk->sk_datalen;
961 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
962 for (i = 0; i < npages; i++) {
963 svsk->sk_pages[i] = rqstp->rq_pages[i];
964 rqstp->rq_pages[i] = NULL;
965 }
966 }
967
968 static void svc_tcp_clear_pages(struct svc_sock *svsk)
969 {
970 unsigned int i, len, npages;
971
972 if (svsk->sk_datalen == 0)
973 goto out;
974 len = svsk->sk_datalen;
975 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 for (i = 0; i < npages; i++) {
977 if (svsk->sk_pages[i] == NULL) {
978 WARN_ON_ONCE(1);
979 continue;
980 }
981 put_page(svsk->sk_pages[i]);
982 svsk->sk_pages[i] = NULL;
983 }
984 out:
985 svsk->sk_tcplen = 0;
986 svsk->sk_datalen = 0;
987 }
988
989 /*
990 * Receive fragment record header.
991 * If we haven't gotten the record length yet, get the next four bytes.
992 */
993 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
994 {
995 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
996 unsigned int want;
997 int len;
998
999 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1000 struct kvec iov;
1001
1002 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1003 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1004 iov.iov_len = want;
1005 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1006 goto error;
1007 svsk->sk_tcplen += len;
1008
1009 if (len < want) {
1010 dprintk("svc: short recvfrom while reading record "
1011 "length (%d of %d)\n", len, want);
1012 return -EAGAIN;
1013 }
1014
1015 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
1016 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1017 serv->sv_max_mesg) {
1018 net_notice_ratelimited("RPC: fragment too large: %d\n",
1019 svc_sock_reclen(svsk));
1020 goto err_delete;
1021 }
1022 }
1023
1024 return svc_sock_reclen(svsk);
1025 error:
1026 dprintk("RPC: TCP recv_record got %d\n", len);
1027 return len;
1028 err_delete:
1029 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1030 return -EAGAIN;
1031 }
1032
1033 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1034 {
1035 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1036 struct rpc_rqst *req = NULL;
1037 struct kvec *src, *dst;
1038 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1039 __be32 xid;
1040 __be32 calldir;
1041
1042 xid = *p++;
1043 calldir = *p;
1044
1045 if (!bc_xprt)
1046 return -EAGAIN;
1047 spin_lock_bh(&bc_xprt->transport_lock);
1048 req = xprt_lookup_rqst(bc_xprt, xid);
1049 if (!req)
1050 goto unlock_notfound;
1051
1052 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1053 /*
1054 * XXX!: cheating for now! Only copying HEAD.
1055 * But we know this is good enough for now (in fact, for any
1056 * callback reply in the forseeable future).
1057 */
1058 dst = &req->rq_private_buf.head[0];
1059 src = &rqstp->rq_arg.head[0];
1060 if (dst->iov_len < src->iov_len)
1061 goto unlock_eagain; /* whatever; just giving up. */
1062 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1063 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1064 rqstp->rq_arg.len = 0;
1065 spin_unlock_bh(&bc_xprt->transport_lock);
1066 return 0;
1067 unlock_notfound:
1068 printk(KERN_NOTICE
1069 "%s: Got unrecognized reply: "
1070 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1071 __func__, ntohl(calldir),
1072 bc_xprt, ntohl(xid));
1073 unlock_eagain:
1074 spin_unlock_bh(&bc_xprt->transport_lock);
1075 return -EAGAIN;
1076 }
1077
1078 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1079 {
1080 int i = 0;
1081 int t = 0;
1082
1083 while (t < len) {
1084 vec[i].iov_base = page_address(pages[i]);
1085 vec[i].iov_len = PAGE_SIZE;
1086 i++;
1087 t += PAGE_SIZE;
1088 }
1089 return i;
1090 }
1091
1092 static void svc_tcp_fragment_received(struct svc_sock *svsk)
1093 {
1094 /* If we have more data, signal svc_xprt_enqueue() to try again */
1095 dprintk("svc: TCP %s record (%d bytes)\n",
1096 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1097 svc_sock_reclen(svsk));
1098 svsk->sk_tcplen = 0;
1099 svsk->sk_reclen = 0;
1100 }
1101
1102 /*
1103 * Receive data from a TCP socket.
1104 */
1105 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1106 {
1107 struct svc_sock *svsk =
1108 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1109 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1110 int len;
1111 struct kvec *vec;
1112 unsigned int want, base;
1113 __be32 *p;
1114 __be32 calldir;
1115 int pnum;
1116
1117 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1118 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1119 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1120 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1121
1122 len = svc_tcp_recv_record(svsk, rqstp);
1123 if (len < 0)
1124 goto error;
1125
1126 base = svc_tcp_restore_pages(svsk, rqstp);
1127 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1128
1129 vec = rqstp->rq_vec;
1130
1131 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1132 svsk->sk_datalen + want);
1133
1134 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1135 rqstp->rq_next_page = rqstp->rq_respages + 1;
1136
1137 /* Now receive data */
1138 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1139 if (len >= 0) {
1140 svsk->sk_tcplen += len;
1141 svsk->sk_datalen += len;
1142 }
1143 if (len != want || !svc_sock_final_rec(svsk)) {
1144 svc_tcp_save_pages(svsk, rqstp);
1145 if (len < 0 && len != -EAGAIN)
1146 goto err_delete;
1147 if (len == want)
1148 svc_tcp_fragment_received(svsk);
1149 else
1150 dprintk("svc: incomplete TCP record (%d of %d)\n",
1151 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1152 svc_sock_reclen(svsk));
1153 goto err_noclose;
1154 }
1155
1156 if (svsk->sk_datalen < 8) {
1157 svsk->sk_datalen = 0;
1158 goto err_delete; /* client is nuts. */
1159 }
1160
1161 rqstp->rq_arg.len = svsk->sk_datalen;
1162 rqstp->rq_arg.page_base = 0;
1163 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1164 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1165 rqstp->rq_arg.page_len = 0;
1166 } else
1167 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1168
1169 rqstp->rq_xprt_ctxt = NULL;
1170 rqstp->rq_prot = IPPROTO_TCP;
1171 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1172 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1173 else
1174 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1175
1176 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1177 calldir = p[1];
1178 if (calldir)
1179 len = receive_cb_reply(svsk, rqstp);
1180
1181 /* Reset TCP read info */
1182 svsk->sk_datalen = 0;
1183 svc_tcp_fragment_received(svsk);
1184
1185 if (len < 0)
1186 goto error;
1187
1188 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1189 if (serv->sv_stats)
1190 serv->sv_stats->nettcpcnt++;
1191
1192 return rqstp->rq_arg.len;
1193
1194 error:
1195 if (len != -EAGAIN)
1196 goto err_delete;
1197 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1198 return 0;
1199 err_delete:
1200 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1201 svsk->sk_xprt.xpt_server->sv_name, -len);
1202 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1203 err_noclose:
1204 return 0; /* record not complete */
1205 }
1206
1207 /*
1208 * Send out data on TCP socket.
1209 */
1210 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1211 {
1212 struct xdr_buf *xbufp = &rqstp->rq_res;
1213 int sent;
1214 __be32 reclen;
1215
1216 /* Set up the first element of the reply kvec.
1217 * Any other kvecs that may be in use have been taken
1218 * care of by the server implementation itself.
1219 */
1220 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1221 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1222
1223 sent = svc_sendto(rqstp, &rqstp->rq_res);
1224 if (sent != xbufp->len) {
1225 printk(KERN_NOTICE
1226 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1227 "- shutting down socket\n",
1228 rqstp->rq_xprt->xpt_server->sv_name,
1229 (sent<0)?"got error":"sent only",
1230 sent, xbufp->len);
1231 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1232 svc_xprt_enqueue(rqstp->rq_xprt);
1233 sent = -EAGAIN;
1234 }
1235 return sent;
1236 }
1237
1238 /*
1239 * Setup response header. TCP has a 4B record length field.
1240 */
1241 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1242 {
1243 struct kvec *resv = &rqstp->rq_res.head[0];
1244
1245 /* tcp needs a space for the record length... */
1246 svc_putnl(resv, 0);
1247 }
1248
1249 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1250 struct net *net,
1251 struct sockaddr *sa, int salen,
1252 int flags)
1253 {
1254 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1255 }
1256
1257 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1258 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1259 struct net *, struct sockaddr *,
1260 int, int);
1261 static void svc_bc_sock_free(struct svc_xprt *xprt);
1262
1263 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1264 struct net *net,
1265 struct sockaddr *sa, int salen,
1266 int flags)
1267 {
1268 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1269 }
1270
1271 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1272 {
1273 }
1274
1275 static struct svc_xprt_ops svc_tcp_bc_ops = {
1276 .xpo_create = svc_bc_tcp_create,
1277 .xpo_detach = svc_bc_tcp_sock_detach,
1278 .xpo_free = svc_bc_sock_free,
1279 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1280 .xpo_secure_port = svc_sock_secure_port,
1281 };
1282
1283 static struct svc_xprt_class svc_tcp_bc_class = {
1284 .xcl_name = "tcp-bc",
1285 .xcl_owner = THIS_MODULE,
1286 .xcl_ops = &svc_tcp_bc_ops,
1287 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1288 };
1289
1290 static void svc_init_bc_xprt_sock(void)
1291 {
1292 svc_reg_xprt_class(&svc_tcp_bc_class);
1293 }
1294
1295 static void svc_cleanup_bc_xprt_sock(void)
1296 {
1297 svc_unreg_xprt_class(&svc_tcp_bc_class);
1298 }
1299 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1300 static void svc_init_bc_xprt_sock(void)
1301 {
1302 }
1303
1304 static void svc_cleanup_bc_xprt_sock(void)
1305 {
1306 }
1307 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1308
1309 static struct svc_xprt_ops svc_tcp_ops = {
1310 .xpo_create = svc_tcp_create,
1311 .xpo_recvfrom = svc_tcp_recvfrom,
1312 .xpo_sendto = svc_tcp_sendto,
1313 .xpo_release_rqst = svc_release_skb,
1314 .xpo_detach = svc_tcp_sock_detach,
1315 .xpo_free = svc_sock_free,
1316 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1317 .xpo_has_wspace = svc_tcp_has_wspace,
1318 .xpo_accept = svc_tcp_accept,
1319 .xpo_secure_port = svc_sock_secure_port,
1320 .xpo_adjust_wspace = svc_tcp_adjust_wspace,
1321 };
1322
1323 static struct svc_xprt_class svc_tcp_class = {
1324 .xcl_name = "tcp",
1325 .xcl_owner = THIS_MODULE,
1326 .xcl_ops = &svc_tcp_ops,
1327 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1328 .xcl_ident = XPRT_TRANSPORT_TCP,
1329 };
1330
1331 void svc_init_xprt_sock(void)
1332 {
1333 svc_reg_xprt_class(&svc_tcp_class);
1334 svc_reg_xprt_class(&svc_udp_class);
1335 svc_init_bc_xprt_sock();
1336 }
1337
1338 void svc_cleanup_xprt_sock(void)
1339 {
1340 svc_unreg_xprt_class(&svc_tcp_class);
1341 svc_unreg_xprt_class(&svc_udp_class);
1342 svc_cleanup_bc_xprt_sock();
1343 }
1344
1345 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1346 {
1347 struct sock *sk = svsk->sk_sk;
1348
1349 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1350 &svsk->sk_xprt, serv);
1351 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1352 if (sk->sk_state == TCP_LISTEN) {
1353 dprintk("setting up TCP socket for listening\n");
1354 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1355 sk->sk_data_ready = svc_tcp_listen_data_ready;
1356 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1357 } else {
1358 dprintk("setting up TCP socket for reading\n");
1359 sk->sk_state_change = svc_tcp_state_change;
1360 sk->sk_data_ready = svc_tcp_data_ready;
1361 sk->sk_write_space = svc_tcp_write_space;
1362
1363 svsk->sk_reclen = 0;
1364 svsk->sk_tcplen = 0;
1365 svsk->sk_datalen = 0;
1366 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1367
1368 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1369
1370 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1371 if (sk->sk_state != TCP_ESTABLISHED)
1372 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1373 }
1374 }
1375
1376 void svc_sock_update_bufs(struct svc_serv *serv)
1377 {
1378 /*
1379 * The number of server threads has changed. Update
1380 * rcvbuf and sndbuf accordingly on all sockets
1381 */
1382 struct svc_sock *svsk;
1383
1384 spin_lock_bh(&serv->sv_lock);
1385 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1386 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1387 spin_unlock_bh(&serv->sv_lock);
1388 }
1389 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1390
1391 /*
1392 * Initialize socket for RPC use and create svc_sock struct
1393 */
1394 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1395 struct socket *sock,
1396 int flags)
1397 {
1398 struct svc_sock *svsk;
1399 struct sock *inet;
1400 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1401 int err = 0;
1402
1403 dprintk("svc: svc_setup_socket %p\n", sock);
1404 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1405 if (!svsk)
1406 return ERR_PTR(-ENOMEM);
1407
1408 inet = sock->sk;
1409
1410 /* Register socket with portmapper */
1411 if (pmap_register)
1412 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1413 inet->sk_protocol,
1414 ntohs(inet_sk(inet)->inet_sport));
1415
1416 if (err < 0) {
1417 kfree(svsk);
1418 return ERR_PTR(err);
1419 }
1420
1421 inet->sk_user_data = svsk;
1422 svsk->sk_sock = sock;
1423 svsk->sk_sk = inet;
1424 svsk->sk_ostate = inet->sk_state_change;
1425 svsk->sk_odata = inet->sk_data_ready;
1426 svsk->sk_owspace = inet->sk_write_space;
1427
1428 /* Initialize the socket */
1429 if (sock->type == SOCK_DGRAM)
1430 svc_udp_init(svsk, serv);
1431 else {
1432 /* initialise setting must have enough space to
1433 * receive and respond to one request.
1434 */
1435 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1436 4 * serv->sv_max_mesg);
1437 svc_tcp_init(svsk, serv);
1438 }
1439
1440 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1441 svsk, svsk->sk_sk);
1442
1443 return svsk;
1444 }
1445
1446 bool svc_alien_sock(struct net *net, int fd)
1447 {
1448 int err;
1449 struct socket *sock = sockfd_lookup(fd, &err);
1450 bool ret = false;
1451
1452 if (!sock)
1453 goto out;
1454 if (sock_net(sock->sk) != net)
1455 ret = true;
1456 sockfd_put(sock);
1457 out:
1458 return ret;
1459 }
1460 EXPORT_SYMBOL_GPL(svc_alien_sock);
1461
1462 /**
1463 * svc_addsock - add a listener socket to an RPC service
1464 * @serv: pointer to RPC service to which to add a new listener
1465 * @fd: file descriptor of the new listener
1466 * @name_return: pointer to buffer to fill in with name of listener
1467 * @len: size of the buffer
1468 *
1469 * Fills in socket name and returns positive length of name if successful.
1470 * Name is terminated with '\n'. On error, returns a negative errno
1471 * value.
1472 */
1473 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1474 const size_t len)
1475 {
1476 int err = 0;
1477 struct socket *so = sockfd_lookup(fd, &err);
1478 struct svc_sock *svsk = NULL;
1479 struct sockaddr_storage addr;
1480 struct sockaddr *sin = (struct sockaddr *)&addr;
1481 int salen;
1482
1483 if (!so)
1484 return err;
1485 err = -EAFNOSUPPORT;
1486 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1487 goto out;
1488 err = -EPROTONOSUPPORT;
1489 if (so->sk->sk_protocol != IPPROTO_TCP &&
1490 so->sk->sk_protocol != IPPROTO_UDP)
1491 goto out;
1492 err = -EISCONN;
1493 if (so->state > SS_UNCONNECTED)
1494 goto out;
1495 err = -ENOENT;
1496 if (!try_module_get(THIS_MODULE))
1497 goto out;
1498 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1499 if (IS_ERR(svsk)) {
1500 module_put(THIS_MODULE);
1501 err = PTR_ERR(svsk);
1502 goto out;
1503 }
1504 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1505 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1506 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1507 return svc_one_sock_name(svsk, name_return, len);
1508 out:
1509 sockfd_put(so);
1510 return err;
1511 }
1512 EXPORT_SYMBOL_GPL(svc_addsock);
1513
1514 /*
1515 * Create socket for RPC service.
1516 */
1517 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1518 int protocol,
1519 struct net *net,
1520 struct sockaddr *sin, int len,
1521 int flags)
1522 {
1523 struct svc_sock *svsk;
1524 struct socket *sock;
1525 int error;
1526 int type;
1527 struct sockaddr_storage addr;
1528 struct sockaddr *newsin = (struct sockaddr *)&addr;
1529 int newlen;
1530 int family;
1531 int val;
1532 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1533
1534 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1535 serv->sv_program->pg_name, protocol,
1536 __svc_print_addr(sin, buf, sizeof(buf)));
1537
1538 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1539 printk(KERN_WARNING "svc: only UDP and TCP "
1540 "sockets supported\n");
1541 return ERR_PTR(-EINVAL);
1542 }
1543
1544 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1545 switch (sin->sa_family) {
1546 case AF_INET6:
1547 family = PF_INET6;
1548 break;
1549 case AF_INET:
1550 family = PF_INET;
1551 break;
1552 default:
1553 return ERR_PTR(-EINVAL);
1554 }
1555
1556 error = __sock_create(net, family, type, protocol, &sock, 1);
1557 if (error < 0)
1558 return ERR_PTR(error);
1559
1560 svc_reclassify_socket(sock);
1561
1562 /*
1563 * If this is an PF_INET6 listener, we want to avoid
1564 * getting requests from IPv4 remotes. Those should
1565 * be shunted to a PF_INET listener via rpcbind.
1566 */
1567 val = 1;
1568 if (family == PF_INET6)
1569 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1570 (char *)&val, sizeof(val));
1571
1572 if (type == SOCK_STREAM)
1573 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1574 error = kernel_bind(sock, sin, len);
1575 if (error < 0)
1576 goto bummer;
1577
1578 newlen = len;
1579 error = kernel_getsockname(sock, newsin, &newlen);
1580 if (error < 0)
1581 goto bummer;
1582
1583 if (protocol == IPPROTO_TCP) {
1584 if ((error = kernel_listen(sock, 64)) < 0)
1585 goto bummer;
1586 }
1587
1588 svsk = svc_setup_socket(serv, sock, flags);
1589 if (IS_ERR(svsk)) {
1590 error = PTR_ERR(svsk);
1591 goto bummer;
1592 }
1593 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1594 return (struct svc_xprt *)svsk;
1595 bummer:
1596 dprintk("svc: svc_create_socket error = %d\n", -error);
1597 sock_release(sock);
1598 return ERR_PTR(error);
1599 }
1600
1601 /*
1602 * Detach the svc_sock from the socket so that no
1603 * more callbacks occur.
1604 */
1605 static void svc_sock_detach(struct svc_xprt *xprt)
1606 {
1607 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1608 struct sock *sk = svsk->sk_sk;
1609 wait_queue_head_t *wq;
1610
1611 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1612
1613 /* put back the old socket callbacks */
1614 sk->sk_state_change = svsk->sk_ostate;
1615 sk->sk_data_ready = svsk->sk_odata;
1616 sk->sk_write_space = svsk->sk_owspace;
1617
1618 wq = sk_sleep(sk);
1619 if (sunrpc_waitqueue_active(wq))
1620 wake_up_interruptible(wq);
1621 }
1622
1623 /*
1624 * Disconnect the socket, and reset the callbacks
1625 */
1626 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1627 {
1628 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1629
1630 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1631
1632 svc_sock_detach(xprt);
1633
1634 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1635 svc_tcp_clear_pages(svsk);
1636 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1637 }
1638 }
1639
1640 /*
1641 * Free the svc_sock's socket resources and the svc_sock itself.
1642 */
1643 static void svc_sock_free(struct svc_xprt *xprt)
1644 {
1645 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1646 dprintk("svc: svc_sock_free(%p)\n", svsk);
1647
1648 if (svsk->sk_sock->file)
1649 sockfd_put(svsk->sk_sock);
1650 else
1651 sock_release(svsk->sk_sock);
1652 kfree(svsk);
1653 }
1654
1655 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1656 /*
1657 * Create a back channel svc_xprt which shares the fore channel socket.
1658 */
1659 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1660 int protocol,
1661 struct net *net,
1662 struct sockaddr *sin, int len,
1663 int flags)
1664 {
1665 struct svc_sock *svsk;
1666 struct svc_xprt *xprt;
1667
1668 if (protocol != IPPROTO_TCP) {
1669 printk(KERN_WARNING "svc: only TCP sockets"
1670 " supported on shared back channel\n");
1671 return ERR_PTR(-EINVAL);
1672 }
1673
1674 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1675 if (!svsk)
1676 return ERR_PTR(-ENOMEM);
1677
1678 xprt = &svsk->sk_xprt;
1679 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1680
1681 serv->sv_bc_xprt = xprt;
1682
1683 return xprt;
1684 }
1685
1686 /*
1687 * Free a back channel svc_sock.
1688 */
1689 static void svc_bc_sock_free(struct svc_xprt *xprt)
1690 {
1691 if (xprt)
1692 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1693 }
1694 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux with added FPC-III support