?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
[zen-stable.git] / net / sunrpc / svcsock.c
blob71bed1c1c77a168da2299a1a009fa27e83253b1e
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
47 #include <linux/sunrpc/types.h>
48 #include <linux/sunrpc/clnt.h>
49 #include <linux/sunrpc/xdr.h>
50 #include <linux/sunrpc/msg_prot.h>
51 #include <linux/sunrpc/svcsock.h>
52 #include <linux/sunrpc/stats.h>
53 #include <linux/sunrpc/xprt.h>
54
55 #include "sunrpc.h"
56
57 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
58
59
60 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
61 int *errp, int flags);
62 static void svc_udp_data_ready(struct sock *, int);
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 BUG_ON(sock_owned_by_user(sk));
87 switch (sk->sk_family) {
88 case AF_INET:
89 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
90 &svc_slock_key[0],
91 "sk_xprt.xpt_lock-AF_INET-NFSD",
92 &svc_key[0]);
93 break;
94
95 case AF_INET6:
96 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
97 &svc_slock_key[1],
98 "sk_xprt.xpt_lock-AF_INET6-NFSD",
99 &svc_key[1]);
100 break;
101
102 default:
103 BUG();
104 }
105 }
106 #else
107 static void svc_reclassify_socket(struct socket *sock)
108 {
109 }
110 #endif
111
112 /*
113 * Release an skbuff after use
114 */
115 static void svc_release_skb(struct svc_rqst *rqstp)
116 {
117 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
118
119 if (skb) {
120 struct svc_sock *svsk =
121 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
122 rqstp->rq_xprt_ctxt = NULL;
123
124 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
125 skb_free_datagram_locked(svsk->sk_sk, skb);
126 }
127 }
128
129 union svc_pktinfo_u {
130 struct in_pktinfo pkti;
131 struct in6_pktinfo pkti6;
132 };
133 #define SVC_PKTINFO_SPACE \
134 CMSG_SPACE(sizeof(union svc_pktinfo_u))
135
136 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
137 {
138 struct svc_sock *svsk =
139 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
140 switch (svsk->sk_sk->sk_family) {
141 case AF_INET: {
142 struct in_pktinfo *pki = CMSG_DATA(cmh);
143
144 cmh->cmsg_level = SOL_IP;
145 cmh->cmsg_type = IP_PKTINFO;
146 pki->ipi_ifindex = 0;
147 pki->ipi_spec_dst.s_addr =
148 svc_daddr_in(rqstp)->sin_addr.s_addr;
149 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
150 }
151 break;
152
153 case AF_INET6: {
154 struct in6_pktinfo *pki = CMSG_DATA(cmh);
155 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
156
157 cmh->cmsg_level = SOL_IPV6;
158 cmh->cmsg_type = IPV6_PKTINFO;
159 pki->ipi6_ifindex = daddr->sin6_scope_id;
160 ipv6_addr_copy(&pki->ipi6_addr, &daddr->sin6_addr);
161 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
162 }
163 break;
164 }
165 }
166
167 /*
168 * send routine intended to be shared by the fore- and back-channel
169 */
170 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
171 struct page *headpage, unsigned long headoffset,
172 struct page *tailpage, unsigned long tailoffset)
173 {
174 int result;
175 int size;
176 struct page **ppage = xdr->pages;
177 size_t base = xdr->page_base;
178 unsigned int pglen = xdr->page_len;
179 unsigned int flags = MSG_MORE;
180 int slen;
181 int len = 0;
182
183 slen = xdr->len;
184
185 /* send head */
186 if (slen == xdr->head[0].iov_len)
187 flags = 0;
188 len = kernel_sendpage(sock, headpage, headoffset,
189 xdr->head[0].iov_len, flags);
190 if (len != xdr->head[0].iov_len)
191 goto out;
192 slen -= xdr->head[0].iov_len;
193 if (slen == 0)
194 goto out;
195
196 /* send page data */
197 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
198 while (pglen > 0) {
199 if (slen == size)
200 flags = 0;
201 result = kernel_sendpage(sock, *ppage, base, size, flags);
202 if (result > 0)
203 len += result;
204 if (result != size)
205 goto out;
206 slen -= size;
207 pglen -= size;
208 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
209 base = 0;
210 ppage++;
211 }
212
213 /* send tail */
214 if (xdr->tail[0].iov_len) {
215 result = kernel_sendpage(sock, tailpage, tailoffset,
216 xdr->tail[0].iov_len, 0);
217 if (result > 0)
218 len += result;
219 }
220
221 out:
222 return len;
223 }
224
225
226 /*
227 * Generic sendto routine
228 */
229 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
230 {
231 struct svc_sock *svsk =
232 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
233 struct socket *sock = svsk->sk_sock;
234 union {
235 struct cmsghdr hdr;
236 long all[SVC_PKTINFO_SPACE / sizeof(long)];
237 } buffer;
238 struct cmsghdr *cmh = &buffer.hdr;
239 int len = 0;
240 unsigned long tailoff;
241 unsigned long headoff;
242 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
243
244 if (rqstp->rq_prot == IPPROTO_UDP) {
245 struct msghdr msg = {
246 .msg_name = &rqstp->rq_addr,
247 .msg_namelen = rqstp->rq_addrlen,
248 .msg_control = cmh,
249 .msg_controllen = sizeof(buffer),
250 .msg_flags = MSG_MORE,
251 };
252
253 svc_set_cmsg_data(rqstp, cmh);
254
255 if (sock_sendmsg(sock, &msg, 0) < 0)
256 goto out;
257 }
258
259 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
260 headoff = 0;
261 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
262 rqstp->rq_respages[0], tailoff);
263
264 out:
265 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
266 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
267 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
268
269 return len;
270 }
271
272 /*
273 * Report socket names for nfsdfs
274 */
275 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
276 {
277 const struct sock *sk = svsk->sk_sk;
278 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
279 "udp" : "tcp";
280 int len;
281
282 switch (sk->sk_family) {
283 case PF_INET:
284 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
285 proto_name,
286 &inet_sk(sk)->inet_rcv_saddr,
287 inet_sk(sk)->inet_num);
288 break;
289 case PF_INET6:
290 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
291 proto_name,
292 &inet6_sk(sk)->rcv_saddr,
293 inet_sk(sk)->inet_num);
294 break;
295 default:
296 len = snprintf(buf, remaining, "*unknown-%d*\n",
297 sk->sk_family);
298 }
299
300 if (len >= remaining) {
301 *buf = '\0';
302 return -ENAMETOOLONG;
303 }
304 return len;
305 }
306
307 /**
308 * svc_sock_names - construct a list of listener names in a string
309 * @serv: pointer to RPC service
310 * @buf: pointer to a buffer to fill in with socket names
311 * @buflen: size of the buffer to be filled
312 * @toclose: pointer to '\0'-terminated C string containing the name
313 * of a listener to be closed
314 *
315 * Fills in @buf with a '\n'-separated list of names of listener
316 * sockets. If @toclose is not NULL, the socket named by @toclose
317 * is closed, and is not included in the output list.
318 *
319 * Returns positive length of the socket name string, or a negative
320 * errno value on error.
321 */
322 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
323 const char *toclose)
324 {
325 struct svc_sock *svsk, *closesk = NULL;
326 int len = 0;
327
328 if (!serv)
329 return 0;
330
331 spin_lock_bh(&serv->sv_lock);
332 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
333 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
334 if (onelen < 0) {
335 len = onelen;
336 break;
337 }
338 if (toclose && strcmp(toclose, buf + len) == 0) {
339 closesk = svsk;
340 svc_xprt_get(&closesk->sk_xprt);
341 } else
342 len += onelen;
343 }
344 spin_unlock_bh(&serv->sv_lock);
345
346 if (closesk) {
347 /* Should unregister with portmap, but you cannot
348 * unregister just one protocol...
349 */
350 svc_close_xprt(&closesk->sk_xprt);
351 svc_xprt_put(&closesk->sk_xprt);
352 } else if (toclose)
353 return -ENOENT;
354 return len;
355 }
356 EXPORT_SYMBOL_GPL(svc_sock_names);
357
358 /*
359 * Check input queue length
360 */
361 static int svc_recv_available(struct svc_sock *svsk)
362 {
363 struct socket *sock = svsk->sk_sock;
364 int avail, err;
365
366 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
367
368 return (err >= 0)? avail : err;
369 }
370
371 /*
372 * Generic recvfrom routine.
373 */
374 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
375 int buflen)
376 {
377 struct svc_sock *svsk =
378 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
379 struct msghdr msg = {
380 .msg_flags = MSG_DONTWAIT,
381 };
382 int len;
383
384 rqstp->rq_xprt_hlen = 0;
385
386 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
387 msg.msg_flags);
388
389 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
390 svsk, iov[0].iov_base, iov[0].iov_len, len);
391 return len;
392 }
393
394 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
395 struct kvec *iov, int nr,
396 int buflen, unsigned int base)
397 {
398 size_t save_iovlen;
399 void __user *save_iovbase;
400 unsigned int i;
401 int ret;
402
403 if (base == 0)
404 return svc_recvfrom(rqstp, iov, nr, buflen);
405
406 for (i = 0; i < nr; i++) {
407 if (iov[i].iov_len > base)
408 break;
409 base -= iov[i].iov_len;
410 }
411 save_iovlen = iov[i].iov_len;
412 save_iovbase = iov[i].iov_base;
413 iov[i].iov_len -= base;
414 iov[i].iov_base += base;
415 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
416 iov[i].iov_len = save_iovlen;
417 iov[i].iov_base = save_iovbase;
418 return ret;
419 }
420
421 /*
422 * Set socket snd and rcv buffer lengths
423 */
424 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
425 unsigned int rcv)
426 {
427 #if 0
428 mm_segment_t oldfs;
429 oldfs = get_fs(); set_fs(KERNEL_DS);
430 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
431 (char*)&snd, sizeof(snd));
432 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
433 (char*)&rcv, sizeof(rcv));
434 #else
435 /* sock_setsockopt limits use to sysctl_?mem_max,
436 * which isn't acceptable. Until that is made conditional
437 * on not having CAP_SYS_RESOURCE or similar, we go direct...
438 * DaveM said I could!
439 */
440 lock_sock(sock->sk);
441 sock->sk->sk_sndbuf = snd * 2;
442 sock->sk->sk_rcvbuf = rcv * 2;
443 sock->sk->sk_write_space(sock->sk);
444 release_sock(sock->sk);
445 #endif
446 }
447 /*
448 * INET callback when data has been received on the socket.
449 */
450 static void svc_udp_data_ready(struct sock *sk, int count)
451 {
452 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
453 wait_queue_head_t *wq = sk_sleep(sk);
454
455 if (svsk) {
456 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
457 svsk, sk, count,
458 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
459 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
460 svc_xprt_enqueue(&svsk->sk_xprt);
461 }
462 if (wq && waitqueue_active(wq))
463 wake_up_interruptible(wq);
464 }
465
466 /*
467 * INET callback when space is newly available on the socket.
468 */
469 static void svc_write_space(struct sock *sk)
470 {
471 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
472 wait_queue_head_t *wq = sk_sleep(sk);
473
474 if (svsk) {
475 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
476 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
477 svc_xprt_enqueue(&svsk->sk_xprt);
478 }
479
480 if (wq && waitqueue_active(wq)) {
481 dprintk("RPC svc_write_space: someone sleeping on %p\n",
482 svsk);
483 wake_up_interruptible(wq);
484 }
485 }
486
487 static void svc_tcp_write_space(struct sock *sk)
488 {
489 struct socket *sock = sk->sk_socket;
490
491 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
492 clear_bit(SOCK_NOSPACE, &sock->flags);
493 svc_write_space(sk);
494 }
495
496 /*
497 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
498 */
499 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
500 struct cmsghdr *cmh)
501 {
502 struct in_pktinfo *pki = CMSG_DATA(cmh);
503 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
504
505 if (cmh->cmsg_type != IP_PKTINFO)
506 return 0;
507
508 daddr->sin_family = AF_INET;
509 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
510 return 1;
511 }
512
513 /*
514 * See net/ipv6/datagram.c : datagram_recv_ctl
515 */
516 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
517 struct cmsghdr *cmh)
518 {
519 struct in6_pktinfo *pki = CMSG_DATA(cmh);
520 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
521
522 if (cmh->cmsg_type != IPV6_PKTINFO)
523 return 0;
524
525 daddr->sin6_family = AF_INET6;
526 ipv6_addr_copy(&daddr->sin6_addr, &pki->ipi6_addr);
527 daddr->sin6_scope_id = pki->ipi6_ifindex;
528 return 1;
529 }
530
531 /*
532 * Copy the UDP datagram's destination address to the rqstp structure.
533 * The 'destination' address in this case is the address to which the
534 * peer sent the datagram, i.e. our local address. For multihomed
535 * hosts, this can change from msg to msg. Note that only the IP
536 * address changes, the port number should remain the same.
537 */
538 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
539 struct cmsghdr *cmh)
540 {
541 switch (cmh->cmsg_level) {
542 case SOL_IP:
543 return svc_udp_get_dest_address4(rqstp, cmh);
544 case SOL_IPV6:
545 return svc_udp_get_dest_address6(rqstp, cmh);
546 }
547
548 return 0;
549 }
550
551 /*
552 * Receive a datagram from a UDP socket.
553 */
554 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
555 {
556 struct svc_sock *svsk =
557 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
558 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
559 struct sk_buff *skb;
560 union {
561 struct cmsghdr hdr;
562 long all[SVC_PKTINFO_SPACE / sizeof(long)];
563 } buffer;
564 struct cmsghdr *cmh = &buffer.hdr;
565 struct msghdr msg = {
566 .msg_name = svc_addr(rqstp),
567 .msg_control = cmh,
568 .msg_controllen = sizeof(buffer),
569 .msg_flags = MSG_DONTWAIT,
570 };
571 size_t len;
572 int err;
573
574 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
575 /* udp sockets need large rcvbuf as all pending
576 * requests are still in that buffer. sndbuf must
577 * also be large enough that there is enough space
578 * for one reply per thread. We count all threads
579 * rather than threads in a particular pool, which
580 * provides an upper bound on the number of threads
581 * which will access the socket.
582 */
583 svc_sock_setbufsize(svsk->sk_sock,
584 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
585 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
586
587 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
588 skb = NULL;
589 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
590 0, 0, MSG_PEEK | MSG_DONTWAIT);
591 if (err >= 0)
592 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
593
594 if (skb == NULL) {
595 if (err != -EAGAIN) {
596 /* possibly an icmp error */
597 dprintk("svc: recvfrom returned error %d\n", -err);
598 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
599 }
600 return -EAGAIN;
601 }
602 len = svc_addr_len(svc_addr(rqstp));
603 if (len == 0)
604 return -EAFNOSUPPORT;
605 rqstp->rq_addrlen = len;
606 if (skb->tstamp.tv64 == 0) {
607 skb->tstamp = ktime_get_real();
608 /* Don't enable netstamp, sunrpc doesn't
609 need that much accuracy */
610 }
611 svsk->sk_sk->sk_stamp = skb->tstamp;
612 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
613
614 len = skb->len - sizeof(struct udphdr);
615 rqstp->rq_arg.len = len;
616
617 rqstp->rq_prot = IPPROTO_UDP;
618
619 if (!svc_udp_get_dest_address(rqstp, cmh)) {
620 if (net_ratelimit())
621 printk(KERN_WARNING
622 "svc: received unknown control message %d/%d; "
623 "dropping RPC reply datagram\n",
624 cmh->cmsg_level, cmh->cmsg_type);
625 skb_free_datagram_locked(svsk->sk_sk, skb);
626 return 0;
627 }
628 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
629
630 if (skb_is_nonlinear(skb)) {
631 /* we have to copy */
632 local_bh_disable();
633 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
634 local_bh_enable();
635 /* checksum error */
636 skb_free_datagram_locked(svsk->sk_sk, skb);
637 return 0;
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 sizeof(struct udphdr);
645 rqstp->rq_arg.head[0].iov_len = len;
646 if (skb_checksum_complete(skb)) {
647 skb_free_datagram_locked(svsk->sk_sk, skb);
648 return 0;
649 }
650 rqstp->rq_xprt_ctxt = skb;
651 }
652
653 rqstp->rq_arg.page_base = 0;
654 if (len <= rqstp->rq_arg.head[0].iov_len) {
655 rqstp->rq_arg.head[0].iov_len = len;
656 rqstp->rq_arg.page_len = 0;
657 rqstp->rq_respages = rqstp->rq_pages+1;
658 } else {
659 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
660 rqstp->rq_respages = rqstp->rq_pages + 1 +
661 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
662 }
663
664 if (serv->sv_stats)
665 serv->sv_stats->netudpcnt++;
666
667 return len;
668 }
669
670 static int
671 svc_udp_sendto(struct svc_rqst *rqstp)
672 {
673 int error;
674
675 error = svc_sendto(rqstp, &rqstp->rq_res);
676 if (error == -ECONNREFUSED)
677 /* ICMP error on earlier request. */
678 error = svc_sendto(rqstp, &rqstp->rq_res);
679
680 return error;
681 }
682
683 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
684 {
685 }
686
687 static int svc_udp_has_wspace(struct svc_xprt *xprt)
688 {
689 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
690 struct svc_serv *serv = xprt->xpt_server;
691 unsigned long required;
692
693 /*
694 * Set the SOCK_NOSPACE flag before checking the available
695 * sock space.
696 */
697 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
698 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
699 if (required*2 > sock_wspace(svsk->sk_sk))
700 return 0;
701 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
702 return 1;
703 }
704
705 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
706 {
707 BUG();
708 return NULL;
709 }
710
711 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
712 struct net *net,
713 struct sockaddr *sa, int salen,
714 int flags)
715 {
716 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
717 }
718
719 static struct svc_xprt_ops svc_udp_ops = {
720 .xpo_create = svc_udp_create,
721 .xpo_recvfrom = svc_udp_recvfrom,
722 .xpo_sendto = svc_udp_sendto,
723 .xpo_release_rqst = svc_release_skb,
724 .xpo_detach = svc_sock_detach,
725 .xpo_free = svc_sock_free,
726 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
727 .xpo_has_wspace = svc_udp_has_wspace,
728 .xpo_accept = svc_udp_accept,
729 };
730
731 static struct svc_xprt_class svc_udp_class = {
732 .xcl_name = "udp",
733 .xcl_owner = THIS_MODULE,
734 .xcl_ops = &svc_udp_ops,
735 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
736 };
737
738 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
739 {
740 int err, level, optname, one = 1;
741
742 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
743 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
744 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
745 svsk->sk_sk->sk_write_space = svc_write_space;
746
747 /* initialise setting must have enough space to
748 * receive and respond to one request.
749 * svc_udp_recvfrom will re-adjust if necessary
750 */
751 svc_sock_setbufsize(svsk->sk_sock,
752 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
753 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
754
755 /* data might have come in before data_ready set up */
756 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
757 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
758
759 /* make sure we get destination address info */
760 switch (svsk->sk_sk->sk_family) {
761 case AF_INET:
762 level = SOL_IP;
763 optname = IP_PKTINFO;
764 break;
765 case AF_INET6:
766 level = SOL_IPV6;
767 optname = IPV6_RECVPKTINFO;
768 break;
769 default:
770 BUG();
771 }
772 err = kernel_setsockopt(svsk->sk_sock, level, optname,
773 (char *)&one, sizeof(one));
774 dprintk("svc: kernel_setsockopt returned %d\n", err);
775 }
776
777 /*
778 * A data_ready event on a listening socket means there's a connection
779 * pending. Do not use state_change as a substitute for it.
780 */
781 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
782 {
783 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
784 wait_queue_head_t *wq;
785
786 dprintk("svc: socket %p TCP (listen) state change %d\n",
787 sk, sk->sk_state);
788
789 /*
790 * This callback may called twice when a new connection
791 * is established as a child socket inherits everything
792 * from a parent LISTEN socket.
793 * 1) data_ready method of the parent socket will be called
794 * when one of child sockets become ESTABLISHED.
795 * 2) data_ready method of the child socket may be called
796 * when it receives data before the socket is accepted.
797 * In case of 2, we should ignore it silently.
798 */
799 if (sk->sk_state == TCP_LISTEN) {
800 if (svsk) {
801 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
802 svc_xprt_enqueue(&svsk->sk_xprt);
803 } else
804 printk("svc: socket %p: no user data\n", sk);
805 }
806
807 wq = sk_sleep(sk);
808 if (wq && waitqueue_active(wq))
809 wake_up_interruptible_all(wq);
810 }
811
812 /*
813 * A state change on a connected socket means it's dying or dead.
814 */
815 static void svc_tcp_state_change(struct sock *sk)
816 {
817 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
818 wait_queue_head_t *wq = sk_sleep(sk);
819
820 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
821 sk, sk->sk_state, sk->sk_user_data);
822
823 if (!svsk)
824 printk("svc: socket %p: no user data\n", sk);
825 else {
826 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
827 svc_xprt_enqueue(&svsk->sk_xprt);
828 }
829 if (wq && waitqueue_active(wq))
830 wake_up_interruptible_all(wq);
831 }
832
833 static void svc_tcp_data_ready(struct sock *sk, int count)
834 {
835 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
836 wait_queue_head_t *wq = sk_sleep(sk);
837
838 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
839 sk, sk->sk_user_data);
840 if (svsk) {
841 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
842 svc_xprt_enqueue(&svsk->sk_xprt);
843 }
844 if (wq && waitqueue_active(wq))
845 wake_up_interruptible(wq);
846 }
847
848 /*
849 * Accept a TCP connection
850 */
851 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
852 {
853 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
854 struct sockaddr_storage addr;
855 struct sockaddr *sin = (struct sockaddr *) &addr;
856 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
857 struct socket *sock = svsk->sk_sock;
858 struct socket *newsock;
859 struct svc_sock *newsvsk;
860 int err, slen;
861 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
862
863 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
864 if (!sock)
865 return NULL;
866
867 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
868 err = kernel_accept(sock, &newsock, O_NONBLOCK);
869 if (err < 0) {
870 if (err == -ENOMEM)
871 printk(KERN_WARNING "%s: no more sockets!\n",
872 serv->sv_name);
873 else if (err != -EAGAIN && net_ratelimit())
874 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
875 serv->sv_name, -err);
876 return NULL;
877 }
878 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
879
880 err = kernel_getpeername(newsock, sin, &slen);
881 if (err < 0) {
882 if (net_ratelimit())
883 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
884 serv->sv_name, -err);
885 goto failed; /* aborted connection or whatever */
886 }
887
888 /* Ideally, we would want to reject connections from unauthorized
889 * hosts here, but when we get encryption, the IP of the host won't
890 * tell us anything. For now just warn about unpriv connections.
891 */
892 if (!svc_port_is_privileged(sin)) {
893 dprintk(KERN_WARNING
894 "%s: connect from unprivileged port: %s\n",
895 serv->sv_name,
896 __svc_print_addr(sin, buf, sizeof(buf)));
897 }
898 dprintk("%s: connect from %s\n", serv->sv_name,
899 __svc_print_addr(sin, buf, sizeof(buf)));
900
901 /* make sure that a write doesn't block forever when
902 * low on memory
903 */
904 newsock->sk->sk_sndtimeo = HZ*30;
905
906 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
907 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
908 goto failed;
909 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
910 err = kernel_getsockname(newsock, sin, &slen);
911 if (unlikely(err < 0)) {
912 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
913 slen = offsetof(struct sockaddr, sa_data);
914 }
915 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
916
917 if (serv->sv_stats)
918 serv->sv_stats->nettcpconn++;
919
920 return &newsvsk->sk_xprt;
921
922 failed:
923 sock_release(newsock);
924 return NULL;
925 }
926
927 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
928 {
929 unsigned int i, len, npages;
930
931 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
932 return 0;
933 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
934 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
935 for (i = 0; i < npages; i++) {
936 if (rqstp->rq_pages[i] != NULL)
937 put_page(rqstp->rq_pages[i]);
938 BUG_ON(svsk->sk_pages[i] == NULL);
939 rqstp->rq_pages[i] = svsk->sk_pages[i];
940 svsk->sk_pages[i] = NULL;
941 }
942 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
943 return len;
944 }
945
946 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
947 {
948 unsigned int i, len, npages;
949
950 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
951 return;
952 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
953 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
954 for (i = 0; i < npages; i++) {
955 svsk->sk_pages[i] = rqstp->rq_pages[i];
956 rqstp->rq_pages[i] = NULL;
957 }
958 }
959
960 static void svc_tcp_clear_pages(struct svc_sock *svsk)
961 {
962 unsigned int i, len, npages;
963
964 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
965 goto out;
966 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
967 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
968 for (i = 0; i < npages; i++) {
969 BUG_ON(svsk->sk_pages[i] == NULL);
970 put_page(svsk->sk_pages[i]);
971 svsk->sk_pages[i] = NULL;
972 }
973 out:
974 svsk->sk_tcplen = 0;
975 }
976
977 /*
978 * Receive data.
979 * If we haven't gotten the record length yet, get the next four bytes.
980 * Otherwise try to gobble up as much as possible up to the complete
981 * record length.
982 */
983 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
984 {
985 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
986 unsigned int want;
987 int len;
988
989 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
990
991 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
992 struct kvec iov;
993
994 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
995 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
996 iov.iov_len = want;
997 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
998 goto error;
999 svsk->sk_tcplen += len;
1000
1001 if (len < want) {
1002 dprintk("svc: short recvfrom while reading record "
1003 "length (%d of %d)\n", len, want);
1004 return -EAGAIN;
1005 }
1006
1007 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1008 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1009 /* FIXME: technically, a record can be fragmented,
1010 * and non-terminal fragments will not have the top
1011 * bit set in the fragment length header.
1012 * But apparently no known nfs clients send fragmented
1013 * records. */
1014 if (net_ratelimit())
1015 printk(KERN_NOTICE "RPC: multiple fragments "
1016 "per record not supported\n");
1017 goto err_delete;
1018 }
1019
1020 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1021 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1022 if (svsk->sk_reclen > serv->sv_max_mesg) {
1023 if (net_ratelimit())
1024 printk(KERN_NOTICE "RPC: "
1025 "fragment too large: 0x%08lx\n",
1026 (unsigned long)svsk->sk_reclen);
1027 goto err_delete;
1028 }
1029 }
1030
1031 if (svsk->sk_reclen < 8)
1032 goto err_delete; /* client is nuts. */
1033
1034 len = svsk->sk_reclen;
1035
1036 return len;
1037 error:
1038 dprintk("RPC: TCP recv_record got %d\n", len);
1039 return len;
1040 err_delete:
1041 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1042 return -EAGAIN;
1043 }
1044
1045 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1046 {
1047 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1048 struct rpc_rqst *req = NULL;
1049 struct kvec *src, *dst;
1050 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1051 __be32 xid;
1052 __be32 calldir;
1053
1054 xid = *p++;
1055 calldir = *p;
1056
1057 if (bc_xprt)
1058 req = xprt_lookup_rqst(bc_xprt, xid);
1059
1060 if (!req) {
1061 printk(KERN_NOTICE
1062 "%s: Got unrecognized reply: "
1063 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1064 __func__, ntohl(calldir),
1065 bc_xprt, xid);
1066 return -EAGAIN;
1067 }
1068
1069 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1070 /*
1071 * XXX!: cheating for now! Only copying HEAD.
1072 * But we know this is good enough for now (in fact, for any
1073 * callback reply in the forseeable future).
1074 */
1075 dst = &req->rq_private_buf.head[0];
1076 src = &rqstp->rq_arg.head[0];
1077 if (dst->iov_len < src->iov_len)
1078 return -EAGAIN; /* whatever; just giving up. */
1079 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1080 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1081 rqstp->rq_arg.len = 0;
1082 return 0;
1083 }
1084
1085 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1086 {
1087 int i = 0;
1088 int t = 0;
1089
1090 while (t < len) {
1091 vec[i].iov_base = page_address(pages[i]);
1092 vec[i].iov_len = PAGE_SIZE;
1093 i++;
1094 t += PAGE_SIZE;
1095 }
1096 return i;
1097 }
1098
1099
1100 /*
1101 * Receive data from a TCP socket.
1102 */
1103 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1104 {
1105 struct svc_sock *svsk =
1106 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1107 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1108 int len;
1109 struct kvec *vec;
1110 unsigned int want, base;
1111 __be32 *p;
1112 __be32 calldir;
1113 int pnum;
1114
1115 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1116 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1117 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1118 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1119
1120 len = svc_tcp_recv_record(svsk, rqstp);
1121 if (len < 0)
1122 goto error;
1123
1124 base = svc_tcp_restore_pages(svsk, rqstp);
1125 want = svsk->sk_reclen - base;
1126
1127 vec = rqstp->rq_vec;
1128
1129 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1130 svsk->sk_reclen);
1131
1132 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1133
1134 /* Now receive data */
1135 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1136 if (len >= 0)
1137 svsk->sk_tcplen += len;
1138 if (len != want) {
1139 if (len < 0 && len != -EAGAIN)
1140 goto err_other;
1141 svc_tcp_save_pages(svsk, rqstp);
1142 dprintk("svc: incomplete TCP record (%d of %d)\n",
1143 svsk->sk_tcplen, svsk->sk_reclen);
1144 goto err_noclose;
1145 }
1146
1147 rqstp->rq_arg.len = svsk->sk_reclen;
1148 rqstp->rq_arg.page_base = 0;
1149 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1150 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1151 rqstp->rq_arg.page_len = 0;
1152 } else
1153 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1154
1155 rqstp->rq_xprt_ctxt = NULL;
1156 rqstp->rq_prot = IPPROTO_TCP;
1157
1158 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1159 calldir = p[1];
1160 if (calldir)
1161 len = receive_cb_reply(svsk, rqstp);
1162
1163 /* Reset TCP read info */
1164 svsk->sk_reclen = 0;
1165 svsk->sk_tcplen = 0;
1166 /* If we have more data, signal svc_xprt_enqueue() to try again */
1167 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1168 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1169
1170 if (len < 0)
1171 goto error;
1172
1173 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1174 if (serv->sv_stats)
1175 serv->sv_stats->nettcpcnt++;
1176
1177 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1178 return rqstp->rq_arg.len;
1179
1180 error:
1181 if (len != -EAGAIN)
1182 goto err_other;
1183 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1184 return -EAGAIN;
1185 err_other:
1186 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1187 svsk->sk_xprt.xpt_server->sv_name, -len);
1188 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1189 err_noclose:
1190 return -EAGAIN; /* record not complete */
1191 }
1192
1193 /*
1194 * Send out data on TCP socket.
1195 */
1196 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1197 {
1198 struct xdr_buf *xbufp = &rqstp->rq_res;
1199 int sent;
1200 __be32 reclen;
1201
1202 /* Set up the first element of the reply kvec.
1203 * Any other kvecs that may be in use have been taken
1204 * care of by the server implementation itself.
1205 */
1206 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1207 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1208
1209 sent = svc_sendto(rqstp, &rqstp->rq_res);
1210 if (sent != xbufp->len) {
1211 printk(KERN_NOTICE
1212 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1213 "- shutting down socket\n",
1214 rqstp->rq_xprt->xpt_server->sv_name,
1215 (sent<0)?"got error":"sent only",
1216 sent, xbufp->len);
1217 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1218 svc_xprt_enqueue(rqstp->rq_xprt);
1219 sent = -EAGAIN;
1220 }
1221 return sent;
1222 }
1223
1224 /*
1225 * Setup response header. TCP has a 4B record length field.
1226 */
1227 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1228 {
1229 struct kvec *resv = &rqstp->rq_res.head[0];
1230
1231 /* tcp needs a space for the record length... */
1232 svc_putnl(resv, 0);
1233 }
1234
1235 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1236 {
1237 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1238 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1239 int required;
1240
1241 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1242 return 1;
1243 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1244 if (sk_stream_wspace(svsk->sk_sk) >= required)
1245 return 1;
1246 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1247 return 0;
1248 }
1249
1250 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1251 struct net *net,
1252 struct sockaddr *sa, int salen,
1253 int flags)
1254 {
1255 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1256 }
1257
1258 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1259 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1260 struct net *, struct sockaddr *,
1261 int, int);
1262 static void svc_bc_sock_free(struct svc_xprt *xprt);
1263
1264 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1265 struct net *net,
1266 struct sockaddr *sa, int salen,
1267 int flags)
1268 {
1269 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1270 }
1271
1272 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1273 {
1274 }
1275
1276 static struct svc_xprt_ops svc_tcp_bc_ops = {
1277 .xpo_create = svc_bc_tcp_create,
1278 .xpo_detach = svc_bc_tcp_sock_detach,
1279 .xpo_free = svc_bc_sock_free,
1280 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
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 };
1320
1321 static struct svc_xprt_class svc_tcp_class = {
1322 .xcl_name = "tcp",
1323 .xcl_owner = THIS_MODULE,
1324 .xcl_ops = &svc_tcp_ops,
1325 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1326 };
1327
1328 void svc_init_xprt_sock(void)
1329 {
1330 svc_reg_xprt_class(&svc_tcp_class);
1331 svc_reg_xprt_class(&svc_udp_class);
1332 svc_init_bc_xprt_sock();
1333 }
1334
1335 void svc_cleanup_xprt_sock(void)
1336 {
1337 svc_unreg_xprt_class(&svc_tcp_class);
1338 svc_unreg_xprt_class(&svc_udp_class);
1339 svc_cleanup_bc_xprt_sock();
1340 }
1341
1342 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1343 {
1344 struct sock *sk = svsk->sk_sk;
1345
1346 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1347 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1348 if (sk->sk_state == TCP_LISTEN) {
1349 dprintk("setting up TCP socket for listening\n");
1350 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1351 sk->sk_data_ready = svc_tcp_listen_data_ready;
1352 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1353 } else {
1354 dprintk("setting up TCP socket for reading\n");
1355 sk->sk_state_change = svc_tcp_state_change;
1356 sk->sk_data_ready = svc_tcp_data_ready;
1357 sk->sk_write_space = svc_tcp_write_space;
1358
1359 svsk->sk_reclen = 0;
1360 svsk->sk_tcplen = 0;
1361 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1362
1363 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1364
1365 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1366 if (sk->sk_state != TCP_ESTABLISHED)
1367 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1368 }
1369 }
1370
1371 void svc_sock_update_bufs(struct svc_serv *serv)
1372 {
1373 /*
1374 * The number of server threads has changed. Update
1375 * rcvbuf and sndbuf accordingly on all sockets
1376 */
1377 struct svc_sock *svsk;
1378
1379 spin_lock_bh(&serv->sv_lock);
1380 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1381 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1382 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1383 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1384 spin_unlock_bh(&serv->sv_lock);
1385 }
1386 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1387
1388 /*
1389 * Initialize socket for RPC use and create svc_sock struct
1390 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1391 */
1392 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1393 struct socket *sock,
1394 int *errp, int flags)
1395 {
1396 struct svc_sock *svsk;
1397 struct sock *inet;
1398 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1399
1400 dprintk("svc: svc_setup_socket %p\n", sock);
1401 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1402 *errp = -ENOMEM;
1403 return NULL;
1404 }
1405
1406 inet = sock->sk;
1407
1408 /* Register socket with portmapper */
1409 if (*errp >= 0 && pmap_register)
1410 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1411 ntohs(inet_sk(inet)->inet_sport));
1412
1413 if (*errp < 0) {
1414 kfree(svsk);
1415 return NULL;
1416 }
1417
1418 inet->sk_user_data = svsk;
1419 svsk->sk_sock = sock;
1420 svsk->sk_sk = inet;
1421 svsk->sk_ostate = inet->sk_state_change;
1422 svsk->sk_odata = inet->sk_data_ready;
1423 svsk->sk_owspace = inet->sk_write_space;
1424
1425 /* Initialize the socket */
1426 if (sock->type == SOCK_DGRAM)
1427 svc_udp_init(svsk, serv);
1428 else {
1429 /* initialise setting must have enough space to
1430 * receive and respond to one request.
1431 */
1432 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1433 4 * serv->sv_max_mesg);
1434 svc_tcp_init(svsk, serv);
1435 }
1436
1437 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1438 svsk, svsk->sk_sk);
1439
1440 return svsk;
1441 }
1442
1443 /**
1444 * svc_addsock - add a listener socket to an RPC service
1445 * @serv: pointer to RPC service to which to add a new listener
1446 * @fd: file descriptor of the new listener
1447 * @name_return: pointer to buffer to fill in with name of listener
1448 * @len: size of the buffer
1449 *
1450 * Fills in socket name and returns positive length of name if successful.
1451 * Name is terminated with '\n'. On error, returns a negative errno
1452 * value.
1453 */
1454 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1455 const size_t len)
1456 {
1457 int err = 0;
1458 struct socket *so = sockfd_lookup(fd, &err);
1459 struct svc_sock *svsk = NULL;
1460
1461 if (!so)
1462 return err;
1463 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1464 err = -EAFNOSUPPORT;
1465 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1466 so->sk->sk_protocol != IPPROTO_UDP)
1467 err = -EPROTONOSUPPORT;
1468 else if (so->state > SS_UNCONNECTED)
1469 err = -EISCONN;
1470 else {
1471 if (!try_module_get(THIS_MODULE))
1472 err = -ENOENT;
1473 else
1474 svsk = svc_setup_socket(serv, so, &err,
1475 SVC_SOCK_DEFAULTS);
1476 if (svsk) {
1477 struct sockaddr_storage addr;
1478 struct sockaddr *sin = (struct sockaddr *)&addr;
1479 int salen;
1480 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1481 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1482 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1483 spin_lock_bh(&serv->sv_lock);
1484 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1485 spin_unlock_bh(&serv->sv_lock);
1486 svc_xprt_received(&svsk->sk_xprt);
1487 err = 0;
1488 } else
1489 module_put(THIS_MODULE);
1490 }
1491 if (err) {
1492 sockfd_put(so);
1493 return err;
1494 }
1495 return svc_one_sock_name(svsk, name_return, len);
1496 }
1497 EXPORT_SYMBOL_GPL(svc_addsock);
1498
1499 /*
1500 * Create socket for RPC service.
1501 */
1502 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1503 int protocol,
1504 struct net *net,
1505 struct sockaddr *sin, int len,
1506 int flags)
1507 {
1508 struct svc_sock *svsk;
1509 struct socket *sock;
1510 int error;
1511 int type;
1512 struct sockaddr_storage addr;
1513 struct sockaddr *newsin = (struct sockaddr *)&addr;
1514 int newlen;
1515 int family;
1516 int val;
1517 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1518
1519 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1520 serv->sv_program->pg_name, protocol,
1521 __svc_print_addr(sin, buf, sizeof(buf)));
1522
1523 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1524 printk(KERN_WARNING "svc: only UDP and TCP "
1525 "sockets supported\n");
1526 return ERR_PTR(-EINVAL);
1527 }
1528
1529 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1530 switch (sin->sa_family) {
1531 case AF_INET6:
1532 family = PF_INET6;
1533 break;
1534 case AF_INET:
1535 family = PF_INET;
1536 break;
1537 default:
1538 return ERR_PTR(-EINVAL);
1539 }
1540
1541 error = __sock_create(net, family, type, protocol, &sock, 1);
1542 if (error < 0)
1543 return ERR_PTR(error);
1544
1545 svc_reclassify_socket(sock);
1546
1547 /*
1548 * If this is an PF_INET6 listener, we want to avoid
1549 * getting requests from IPv4 remotes. Those should
1550 * be shunted to a PF_INET listener via rpcbind.
1551 */
1552 val = 1;
1553 if (family == PF_INET6)
1554 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1555 (char *)&val, sizeof(val));
1556
1557 if (type == SOCK_STREAM)
1558 sock->sk->sk_reuse = 1; /* allow address reuse */
1559 error = kernel_bind(sock, sin, len);
1560 if (error < 0)
1561 goto bummer;
1562
1563 newlen = len;
1564 error = kernel_getsockname(sock, newsin, &newlen);
1565 if (error < 0)
1566 goto bummer;
1567
1568 if (protocol == IPPROTO_TCP) {
1569 if ((error = kernel_listen(sock, 64)) < 0)
1570 goto bummer;
1571 }
1572
1573 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1574 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1575 return (struct svc_xprt *)svsk;
1576 }
1577
1578 bummer:
1579 dprintk("svc: svc_create_socket error = %d\n", -error);
1580 sock_release(sock);
1581 return ERR_PTR(error);
1582 }
1583
1584 /*
1585 * Detach the svc_sock from the socket so that no
1586 * more callbacks occur.
1587 */
1588 static void svc_sock_detach(struct svc_xprt *xprt)
1589 {
1590 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1591 struct sock *sk = svsk->sk_sk;
1592 wait_queue_head_t *wq;
1593
1594 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1595
1596 /* put back the old socket callbacks */
1597 sk->sk_state_change = svsk->sk_ostate;
1598 sk->sk_data_ready = svsk->sk_odata;
1599 sk->sk_write_space = svsk->sk_owspace;
1600
1601 wq = sk_sleep(sk);
1602 if (wq && waitqueue_active(wq))
1603 wake_up_interruptible(wq);
1604 }
1605
1606 /*
1607 * Disconnect the socket, and reset the callbacks
1608 */
1609 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1610 {
1611 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1612
1613 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1614
1615 svc_sock_detach(xprt);
1616
1617 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1618 svc_tcp_clear_pages(svsk);
1619 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1620 }
1621 }
1622
1623 /*
1624 * Free the svc_sock's socket resources and the svc_sock itself.
1625 */
1626 static void svc_sock_free(struct svc_xprt *xprt)
1627 {
1628 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1629 dprintk("svc: svc_sock_free(%p)\n", svsk);
1630
1631 if (svsk->sk_sock->file)
1632 sockfd_put(svsk->sk_sock);
1633 else
1634 sock_release(svsk->sk_sock);
1635 kfree(svsk);
1636 }
1637
1638 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1639 /*
1640 * Create a back channel svc_xprt which shares the fore channel socket.
1641 */
1642 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1643 int protocol,
1644 struct net *net,
1645 struct sockaddr *sin, int len,
1646 int flags)
1647 {
1648 struct svc_sock *svsk;
1649 struct svc_xprt *xprt;
1650
1651 if (protocol != IPPROTO_TCP) {
1652 printk(KERN_WARNING "svc: only TCP sockets"
1653 " supported on shared back channel\n");
1654 return ERR_PTR(-EINVAL);
1655 }
1656
1657 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1658 if (!svsk)
1659 return ERR_PTR(-ENOMEM);
1660
1661 xprt = &svsk->sk_xprt;
1662 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1663
1664 serv->sv_bc_xprt = xprt;
1665
1666 return xprt;
1667 }
1668
1669 /*
1670 * Free a back channel svc_sock.
1671 */
1672 static void svc_bc_sock_free(struct svc_xprt *xprt)
1673 {
1674 if (xprt)
1675 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1676 }
1677 #endif /* CONFIG_SUNRPC_BACKCHANNEL */