]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/sunrpc/svcsock.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
svcrpc: minor udp code cleanup
[mirror_ubuntu-artful-kernel.git] / net / sunrpc / svcsock.c
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 *, int);
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 BUG_ON(sock_owned_by_user(sk));
88 switch (sk->sk_family) {
89 case AF_INET:
90 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
91 &svc_slock_key[0],
92 "sk_xprt.xpt_lock-AF_INET-NFSD",
93 &svc_key[0]);
94 break;
95
96 case AF_INET6:
97 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
98 &svc_slock_key[1],
99 "sk_xprt.xpt_lock-AF_INET6-NFSD",
100 &svc_key[1]);
101 break;
102
103 default:
104 BUG();
105 }
106 }
107 #else
108 static void svc_reclassify_socket(struct socket *sock)
109 {
110 }
111 #endif
112
113 /*
114 * Release an skbuff after use
115 */
116 static void svc_release_skb(struct svc_rqst *rqstp)
117 {
118 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
119
120 if (skb) {
121 struct svc_sock *svsk =
122 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
123 rqstp->rq_xprt_ctxt = NULL;
124
125 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
126 skb_free_datagram_locked(svsk->sk_sk, skb);
127 }
128 }
129
130 union svc_pktinfo_u {
131 struct in_pktinfo pkti;
132 struct in6_pktinfo pkti6;
133 };
134 #define SVC_PKTINFO_SPACE \
135 CMSG_SPACE(sizeof(union svc_pktinfo_u))
136
137 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
138 {
139 struct svc_sock *svsk =
140 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
141 switch (svsk->sk_sk->sk_family) {
142 case AF_INET: {
143 struct in_pktinfo *pki = CMSG_DATA(cmh);
144
145 cmh->cmsg_level = SOL_IP;
146 cmh->cmsg_type = IP_PKTINFO;
147 pki->ipi_ifindex = 0;
148 pki->ipi_spec_dst.s_addr =
149 svc_daddr_in(rqstp)->sin_addr.s_addr;
150 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
151 }
152 break;
153
154 case AF_INET6: {
155 struct in6_pktinfo *pki = CMSG_DATA(cmh);
156 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
157
158 cmh->cmsg_level = SOL_IPV6;
159 cmh->cmsg_type = IPV6_PKTINFO;
160 pki->ipi6_ifindex = daddr->sin6_scope_id;
161 pki->ipi6_addr = daddr->sin6_addr;
162 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
163 }
164 break;
165 }
166 }
167
168 /*
169 * send routine intended to be shared by the fore- and back-channel
170 */
171 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
172 struct page *headpage, unsigned long headoffset,
173 struct page *tailpage, unsigned long tailoffset)
174 {
175 int result;
176 int size;
177 struct page **ppage = xdr->pages;
178 size_t base = xdr->page_base;
179 unsigned int pglen = xdr->page_len;
180 unsigned int flags = MSG_MORE;
181 int slen;
182 int len = 0;
183
184 slen = xdr->len;
185
186 /* send head */
187 if (slen == xdr->head[0].iov_len)
188 flags = 0;
189 len = kernel_sendpage(sock, headpage, headoffset,
190 xdr->head[0].iov_len, flags);
191 if (len != xdr->head[0].iov_len)
192 goto out;
193 slen -= xdr->head[0].iov_len;
194 if (slen == 0)
195 goto out;
196
197 /* send page data */
198 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
199 while (pglen > 0) {
200 if (slen == size)
201 flags = 0;
202 result = kernel_sendpage(sock, *ppage, base, size, flags);
203 if (result > 0)
204 len += result;
205 if (result != size)
206 goto out;
207 slen -= size;
208 pglen -= size;
209 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
210 base = 0;
211 ppage++;
212 }
213
214 /* send tail */
215 if (xdr->tail[0].iov_len) {
216 result = kernel_sendpage(sock, tailpage, tailoffset,
217 xdr->tail[0].iov_len, 0);
218 if (result > 0)
219 len += result;
220 }
221
222 out:
223 return len;
224 }
225
226
227 /*
228 * Generic sendto routine
229 */
230 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
231 {
232 struct svc_sock *svsk =
233 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
234 struct socket *sock = svsk->sk_sock;
235 union {
236 struct cmsghdr hdr;
237 long all[SVC_PKTINFO_SPACE / sizeof(long)];
238 } buffer;
239 struct cmsghdr *cmh = &buffer.hdr;
240 int len = 0;
241 unsigned long tailoff;
242 unsigned long headoff;
243 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
244
245 if (rqstp->rq_prot == IPPROTO_UDP) {
246 struct msghdr msg = {
247 .msg_name = &rqstp->rq_addr,
248 .msg_namelen = rqstp->rq_addrlen,
249 .msg_control = cmh,
250 .msg_controllen = sizeof(buffer),
251 .msg_flags = MSG_MORE,
252 };
253
254 svc_set_cmsg_data(rqstp, cmh);
255
256 if (sock_sendmsg(sock, &msg, 0) < 0)
257 goto out;
258 }
259
260 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
261 headoff = 0;
262 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
263 rqstp->rq_respages[0], tailoff);
264
265 out:
266 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
267 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
268 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
269
270 return len;
271 }
272
273 /*
274 * Report socket names for nfsdfs
275 */
276 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
277 {
278 const struct sock *sk = svsk->sk_sk;
279 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
280 "udp" : "tcp";
281 int len;
282
283 switch (sk->sk_family) {
284 case PF_INET:
285 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
286 proto_name,
287 &inet_sk(sk)->inet_rcv_saddr,
288 inet_sk(sk)->inet_num);
289 break;
290 case PF_INET6:
291 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
292 proto_name,
293 &inet6_sk(sk)->rcv_saddr,
294 inet_sk(sk)->inet_num);
295 break;
296 default:
297 len = snprintf(buf, remaining, "*unknown-%d*\n",
298 sk->sk_family);
299 }
300
301 if (len >= remaining) {
302 *buf = '\0';
303 return -ENAMETOOLONG;
304 }
305 return len;
306 }
307
308 /**
309 * svc_sock_names - construct a list of listener names in a string
310 * @serv: pointer to RPC service
311 * @buf: pointer to a buffer to fill in with socket names
312 * @buflen: size of the buffer to be filled
313 * @toclose: pointer to '\0'-terminated C string containing the name
314 * of a listener to be closed
315 *
316 * Fills in @buf with a '\n'-separated list of names of listener
317 * sockets. If @toclose is not NULL, the socket named by @toclose
318 * is closed, and is not included in the output list.
319 *
320 * Returns positive length of the socket name string, or a negative
321 * errno value on error.
322 */
323 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
324 const char *toclose)
325 {
326 struct svc_sock *svsk, *closesk = NULL;
327 int len = 0;
328
329 if (!serv)
330 return 0;
331
332 spin_lock_bh(&serv->sv_lock);
333 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
334 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
335 if (onelen < 0) {
336 len = onelen;
337 break;
338 }
339 if (toclose && strcmp(toclose, buf + len) == 0) {
340 closesk = svsk;
341 svc_xprt_get(&closesk->sk_xprt);
342 } else
343 len += onelen;
344 }
345 spin_unlock_bh(&serv->sv_lock);
346
347 if (closesk) {
348 /* Should unregister with portmap, but you cannot
349 * unregister just one protocol...
350 */
351 svc_close_xprt(&closesk->sk_xprt);
352 svc_xprt_put(&closesk->sk_xprt);
353 } else if (toclose)
354 return -ENOENT;
355 return len;
356 }
357 EXPORT_SYMBOL_GPL(svc_sock_names);
358
359 /*
360 * Check input queue length
361 */
362 static int svc_recv_available(struct svc_sock *svsk)
363 {
364 struct socket *sock = svsk->sk_sock;
365 int avail, err;
366
367 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
368
369 return (err >= 0)? avail : err;
370 }
371
372 /*
373 * Generic recvfrom routine.
374 */
375 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
376 int buflen)
377 {
378 struct svc_sock *svsk =
379 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
380 struct msghdr msg = {
381 .msg_flags = MSG_DONTWAIT,
382 };
383 int len;
384
385 rqstp->rq_xprt_hlen = 0;
386
387 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
388 msg.msg_flags);
389
390 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
391 svsk, iov[0].iov_base, iov[0].iov_len, len);
392 return len;
393 }
394
395 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
396 struct kvec *iov, int nr,
397 int buflen, unsigned int base)
398 {
399 size_t save_iovlen;
400 void *save_iovbase;
401 unsigned int i;
402 int ret;
403
404 if (base == 0)
405 return svc_recvfrom(rqstp, iov, nr, buflen);
406
407 for (i = 0; i < nr; i++) {
408 if (iov[i].iov_len > base)
409 break;
410 base -= iov[i].iov_len;
411 }
412 save_iovlen = iov[i].iov_len;
413 save_iovbase = iov[i].iov_base;
414 iov[i].iov_len -= base;
415 iov[i].iov_base += base;
416 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
417 iov[i].iov_len = save_iovlen;
418 iov[i].iov_base = save_iovbase;
419 return ret;
420 }
421
422 /*
423 * Set socket snd and rcv buffer lengths
424 */
425 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
426 unsigned int rcv)
427 {
428 #if 0
429 mm_segment_t oldfs;
430 oldfs = get_fs(); set_fs(KERNEL_DS);
431 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
432 (char*)&snd, sizeof(snd));
433 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
434 (char*)&rcv, sizeof(rcv));
435 #else
436 /* sock_setsockopt limits use to sysctl_?mem_max,
437 * which isn't acceptable. Until that is made conditional
438 * on not having CAP_SYS_RESOURCE or similar, we go direct...
439 * DaveM said I could!
440 */
441 lock_sock(sock->sk);
442 sock->sk->sk_sndbuf = snd * 2;
443 sock->sk->sk_rcvbuf = rcv * 2;
444 sock->sk->sk_write_space(sock->sk);
445 release_sock(sock->sk);
446 #endif
447 }
448 /*
449 * INET callback when data has been received on the socket.
450 */
451 static void svc_udp_data_ready(struct sock *sk, int count)
452 {
453 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
454 wait_queue_head_t *wq = sk_sleep(sk);
455
456 if (svsk) {
457 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
458 svsk, sk, count,
459 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
460 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
461 svc_xprt_enqueue(&svsk->sk_xprt);
462 }
463 if (wq && waitqueue_active(wq))
464 wake_up_interruptible(wq);
465 }
466
467 /*
468 * INET callback when space is newly available on the socket.
469 */
470 static void svc_write_space(struct sock *sk)
471 {
472 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
473 wait_queue_head_t *wq = sk_sleep(sk);
474
475 if (svsk) {
476 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
477 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
478 svc_xprt_enqueue(&svsk->sk_xprt);
479 }
480
481 if (wq && waitqueue_active(wq)) {
482 dprintk("RPC svc_write_space: someone sleeping on %p\n",
483 svsk);
484 wake_up_interruptible(wq);
485 }
486 }
487
488 static void svc_tcp_write_space(struct sock *sk)
489 {
490 struct socket *sock = sk->sk_socket;
491
492 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
493 clear_bit(SOCK_NOSPACE, &sock->flags);
494 svc_write_space(sk);
495 }
496
497 /*
498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
499 */
500 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
501 struct cmsghdr *cmh)
502 {
503 struct in_pktinfo *pki = CMSG_DATA(cmh);
504 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
505
506 if (cmh->cmsg_type != IP_PKTINFO)
507 return 0;
508
509 daddr->sin_family = AF_INET;
510 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
511 return 1;
512 }
513
514 /*
515 * See net/ipv6/datagram.c : datagram_recv_ctl
516 */
517 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
518 struct cmsghdr *cmh)
519 {
520 struct in6_pktinfo *pki = CMSG_DATA(cmh);
521 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
522
523 if (cmh->cmsg_type != IPV6_PKTINFO)
524 return 0;
525
526 daddr->sin6_family = AF_INET6;
527 daddr->sin6_addr = pki->ipi6_addr;
528 daddr->sin6_scope_id = pki->ipi6_ifindex;
529 return 1;
530 }
531
532 /*
533 * Copy the UDP datagram's destination address to the rqstp structure.
534 * The 'destination' address in this case is the address to which the
535 * peer sent the datagram, i.e. our local address. For multihomed
536 * hosts, this can change from msg to msg. Note that only the IP
537 * address changes, the port number should remain the same.
538 */
539 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
540 struct cmsghdr *cmh)
541 {
542 switch (cmh->cmsg_level) {
543 case SOL_IP:
544 return svc_udp_get_dest_address4(rqstp, cmh);
545 case SOL_IPV6:
546 return svc_udp_get_dest_address6(rqstp, cmh);
547 }
548
549 return 0;
550 }
551
552 /*
553 * Receive a datagram from a UDP socket.
554 */
555 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
556 {
557 struct svc_sock *svsk =
558 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
559 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
560 struct sk_buff *skb;
561 union {
562 struct cmsghdr hdr;
563 long all[SVC_PKTINFO_SPACE / sizeof(long)];
564 } buffer;
565 struct cmsghdr *cmh = &buffer.hdr;
566 struct msghdr msg = {
567 .msg_name = svc_addr(rqstp),
568 .msg_control = cmh,
569 .msg_controllen = sizeof(buffer),
570 .msg_flags = MSG_DONTWAIT,
571 };
572 size_t len;
573 int err;
574
575 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
576 /* udp sockets need large rcvbuf as all pending
577 * requests are still in that buffer. sndbuf must
578 * also be large enough that there is enough space
579 * for one reply per thread. We count all threads
580 * rather than threads in a particular pool, which
581 * provides an upper bound on the number of threads
582 * which will access the socket.
583 */
584 svc_sock_setbufsize(svsk->sk_sock,
585 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
586 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
587
588 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
589 skb = NULL;
590 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
591 0, 0, MSG_PEEK | MSG_DONTWAIT);
592 if (err >= 0)
593 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
594
595 if (skb == NULL) {
596 if (err != -EAGAIN) {
597 /* possibly an icmp error */
598 dprintk("svc: recvfrom returned error %d\n", -err);
599 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
600 }
601 return -EAGAIN;
602 }
603 len = svc_addr_len(svc_addr(rqstp));
604 if (len == 0)
605 return -EAFNOSUPPORT;
606 rqstp->rq_addrlen = len;
607 if (skb->tstamp.tv64 == 0) {
608 skb->tstamp = ktime_get_real();
609 /* Don't enable netstamp, sunrpc doesn't
610 need that much accuracy */
611 }
612 svsk->sk_sk->sk_stamp = skb->tstamp;
613 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
614
615 len = skb->len - sizeof(struct udphdr);
616 rqstp->rq_arg.len = len;
617
618 rqstp->rq_prot = IPPROTO_UDP;
619
620 if (!svc_udp_get_dest_address(rqstp, cmh)) {
621 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
622 cmh->cmsg_level, cmh->cmsg_type);
623 goto out_free;
624 }
625 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
626
627 if (skb_is_nonlinear(skb)) {
628 /* we have to copy */
629 local_bh_disable();
630 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
631 local_bh_enable();
632 /* checksum error */
633 goto out_free;
634 }
635 local_bh_enable();
636 skb_free_datagram_locked(svsk->sk_sk, skb);
637 } else {
638 /* we can use it in-place */
639 rqstp->rq_arg.head[0].iov_base = skb->data +
640 sizeof(struct udphdr);
641 rqstp->rq_arg.head[0].iov_len = len;
642 if (skb_checksum_complete(skb))
643 goto out_free;
644 rqstp->rq_xprt_ctxt = skb;
645 }
646
647 rqstp->rq_arg.page_base = 0;
648 if (len <= rqstp->rq_arg.head[0].iov_len) {
649 rqstp->rq_arg.head[0].iov_len = len;
650 rqstp->rq_arg.page_len = 0;
651 rqstp->rq_respages = rqstp->rq_pages+1;
652 } else {
653 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
654 rqstp->rq_respages = rqstp->rq_pages + 1 +
655 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
656 }
657
658 if (serv->sv_stats)
659 serv->sv_stats->netudpcnt++;
660
661 return len;
662 out_free:
663 trace_kfree_skb(skb, svc_udp_recvfrom);
664 skb_free_datagram_locked(svsk->sk_sk, skb);
665 return 0;
666 }
667
668 static int
669 svc_udp_sendto(struct svc_rqst *rqstp)
670 {
671 int error;
672
673 error = svc_sendto(rqstp, &rqstp->rq_res);
674 if (error == -ECONNREFUSED)
675 /* ICMP error on earlier request. */
676 error = svc_sendto(rqstp, &rqstp->rq_res);
677
678 return error;
679 }
680
681 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
682 {
683 }
684
685 static int svc_udp_has_wspace(struct svc_xprt *xprt)
686 {
687 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
688 struct svc_serv *serv = xprt->xpt_server;
689 unsigned long required;
690
691 /*
692 * Set the SOCK_NOSPACE flag before checking the available
693 * sock space.
694 */
695 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
696 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
697 if (required*2 > sock_wspace(svsk->sk_sk))
698 return 0;
699 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
700 return 1;
701 }
702
703 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
704 {
705 BUG();
706 return NULL;
707 }
708
709 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
710 struct net *net,
711 struct sockaddr *sa, int salen,
712 int flags)
713 {
714 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
715 }
716
717 static struct svc_xprt_ops svc_udp_ops = {
718 .xpo_create = svc_udp_create,
719 .xpo_recvfrom = svc_udp_recvfrom,
720 .xpo_sendto = svc_udp_sendto,
721 .xpo_release_rqst = svc_release_skb,
722 .xpo_detach = svc_sock_detach,
723 .xpo_free = svc_sock_free,
724 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
725 .xpo_has_wspace = svc_udp_has_wspace,
726 .xpo_accept = svc_udp_accept,
727 };
728
729 static struct svc_xprt_class svc_udp_class = {
730 .xcl_name = "udp",
731 .xcl_owner = THIS_MODULE,
732 .xcl_ops = &svc_udp_ops,
733 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
734 };
735
736 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
737 {
738 int err, level, optname, one = 1;
739
740 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
741 &svsk->sk_xprt, serv);
742 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
743 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
744 svsk->sk_sk->sk_write_space = svc_write_space;
745
746 /* initialise setting must have enough space to
747 * receive and respond to one request.
748 * svc_udp_recvfrom will re-adjust if necessary
749 */
750 svc_sock_setbufsize(svsk->sk_sock,
751 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
752 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
753
754 /* data might have come in before data_ready set up */
755 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
756 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
757
758 /* make sure we get destination address info */
759 switch (svsk->sk_sk->sk_family) {
760 case AF_INET:
761 level = SOL_IP;
762 optname = IP_PKTINFO;
763 break;
764 case AF_INET6:
765 level = SOL_IPV6;
766 optname = IPV6_RECVPKTINFO;
767 break;
768 default:
769 BUG();
770 }
771 err = kernel_setsockopt(svsk->sk_sock, level, optname,
772 (char *)&one, sizeof(one));
773 dprintk("svc: kernel_setsockopt returned %d\n", err);
774 }
775
776 /*
777 * A data_ready event on a listening socket means there's a connection
778 * pending. Do not use state_change as a substitute for it.
779 */
780 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
781 {
782 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
783 wait_queue_head_t *wq;
784
785 dprintk("svc: socket %p TCP (listen) state change %d\n",
786 sk, sk->sk_state);
787
788 /*
789 * This callback may called twice when a new connection
790 * is established as a child socket inherits everything
791 * from a parent LISTEN socket.
792 * 1) data_ready method of the parent socket will be called
793 * when one of child sockets become ESTABLISHED.
794 * 2) data_ready method of the child socket may be called
795 * when it receives data before the socket is accepted.
796 * In case of 2, we should ignore it silently.
797 */
798 if (sk->sk_state == TCP_LISTEN) {
799 if (svsk) {
800 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
801 svc_xprt_enqueue(&svsk->sk_xprt);
802 } else
803 printk("svc: socket %p: no user data\n", sk);
804 }
805
806 wq = sk_sleep(sk);
807 if (wq && waitqueue_active(wq))
808 wake_up_interruptible_all(wq);
809 }
810
811 /*
812 * A state change on a connected socket means it's dying or dead.
813 */
814 static void svc_tcp_state_change(struct sock *sk)
815 {
816 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
817 wait_queue_head_t *wq = sk_sleep(sk);
818
819 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
820 sk, sk->sk_state, sk->sk_user_data);
821
822 if (!svsk)
823 printk("svc: socket %p: no user data\n", sk);
824 else {
825 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
826 svc_xprt_enqueue(&svsk->sk_xprt);
827 }
828 if (wq && waitqueue_active(wq))
829 wake_up_interruptible_all(wq);
830 }
831
832 static void svc_tcp_data_ready(struct sock *sk, int count)
833 {
834 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
835 wait_queue_head_t *wq = sk_sleep(sk);
836
837 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
838 sk, sk->sk_user_data);
839 if (svsk) {
840 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
841 svc_xprt_enqueue(&svsk->sk_xprt);
842 }
843 if (wq && waitqueue_active(wq))
844 wake_up_interruptible(wq);
845 }
846
847 /*
848 * Accept a TCP connection
849 */
850 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
851 {
852 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
853 struct sockaddr_storage addr;
854 struct sockaddr *sin = (struct sockaddr *) &addr;
855 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
856 struct socket *sock = svsk->sk_sock;
857 struct socket *newsock;
858 struct svc_sock *newsvsk;
859 int err, slen;
860 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
861
862 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
863 if (!sock)
864 return NULL;
865
866 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
867 err = kernel_accept(sock, &newsock, O_NONBLOCK);
868 if (err < 0) {
869 if (err == -ENOMEM)
870 printk(KERN_WARNING "%s: no more sockets!\n",
871 serv->sv_name);
872 else if (err != -EAGAIN)
873 net_warn_ratelimited("%s: accept failed (err %d)!\n",
874 serv->sv_name, -err);
875 return NULL;
876 }
877 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
878
879 err = kernel_getpeername(newsock, sin, &slen);
880 if (err < 0) {
881 net_warn_ratelimited("%s: peername failed (err %d)!\n",
882 serv->sv_name, -err);
883 goto failed; /* aborted connection or whatever */
884 }
885
886 /* Ideally, we would want to reject connections from unauthorized
887 * hosts here, but when we get encryption, the IP of the host won't
888 * tell us anything. For now just warn about unpriv connections.
889 */
890 if (!svc_port_is_privileged(sin)) {
891 dprintk(KERN_WARNING
892 "%s: connect from unprivileged port: %s\n",
893 serv->sv_name,
894 __svc_print_addr(sin, buf, sizeof(buf)));
895 }
896 dprintk("%s: connect from %s\n", serv->sv_name,
897 __svc_print_addr(sin, buf, sizeof(buf)));
898
899 /* make sure that a write doesn't block forever when
900 * low on memory
901 */
902 newsock->sk->sk_sndtimeo = HZ*30;
903
904 newsvsk = svc_setup_socket(serv, newsock,
905 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
906 if (IS_ERR(newsvsk))
907 goto failed;
908 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
909 err = kernel_getsockname(newsock, sin, &slen);
910 if (unlikely(err < 0)) {
911 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
912 slen = offsetof(struct sockaddr, sa_data);
913 }
914 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
915
916 if (serv->sv_stats)
917 serv->sv_stats->nettcpconn++;
918
919 return &newsvsk->sk_xprt;
920
921 failed:
922 sock_release(newsock);
923 return NULL;
924 }
925
926 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
927 {
928 unsigned int i, len, npages;
929
930 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
931 return 0;
932 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
933 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
934 for (i = 0; i < npages; i++) {
935 if (rqstp->rq_pages[i] != NULL)
936 put_page(rqstp->rq_pages[i]);
937 BUG_ON(svsk->sk_pages[i] == NULL);
938 rqstp->rq_pages[i] = svsk->sk_pages[i];
939 svsk->sk_pages[i] = NULL;
940 }
941 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
942 return len;
943 }
944
945 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
946 {
947 unsigned int i, len, npages;
948
949 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
950 return;
951 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
952 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
953 for (i = 0; i < npages; i++) {
954 svsk->sk_pages[i] = rqstp->rq_pages[i];
955 rqstp->rq_pages[i] = NULL;
956 }
957 }
958
959 static void svc_tcp_clear_pages(struct svc_sock *svsk)
960 {
961 unsigned int i, len, npages;
962
963 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
964 goto out;
965 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
966 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
967 for (i = 0; i < npages; i++) {
968 BUG_ON(svsk->sk_pages[i] == NULL);
969 put_page(svsk->sk_pages[i]);
970 svsk->sk_pages[i] = NULL;
971 }
972 out:
973 svsk->sk_tcplen = 0;
974 }
975
976 /*
977 * Receive data.
978 * If we haven't gotten the record length yet, get the next four bytes.
979 * Otherwise try to gobble up as much as possible up to the complete
980 * record length.
981 */
982 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
983 {
984 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
985 unsigned int want;
986 int len;
987
988 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
989
990 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
991 struct kvec iov;
992
993 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
994 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
995 iov.iov_len = want;
996 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
997 goto error;
998 svsk->sk_tcplen += len;
999
1000 if (len < want) {
1001 dprintk("svc: short recvfrom while reading record "
1002 "length (%d of %d)\n", len, want);
1003 return -EAGAIN;
1004 }
1005
1006 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1007 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1008 /* FIXME: technically, a record can be fragmented,
1009 * and non-terminal fragments will not have the top
1010 * bit set in the fragment length header.
1011 * But apparently no known nfs clients send fragmented
1012 * records. */
1013 net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
1014 goto err_delete;
1015 }
1016
1017 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1018 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1019 if (svsk->sk_reclen > serv->sv_max_mesg) {
1020 net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
1021 (unsigned long)svsk->sk_reclen);
1022 goto err_delete;
1023 }
1024 }
1025
1026 if (svsk->sk_reclen < 8)
1027 goto err_delete; /* client is nuts. */
1028
1029 len = svsk->sk_reclen;
1030
1031 return len;
1032 error:
1033 dprintk("RPC: TCP recv_record got %d\n", len);
1034 return len;
1035 err_delete:
1036 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1037 return -EAGAIN;
1038 }
1039
1040 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1041 {
1042 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1043 struct rpc_rqst *req = NULL;
1044 struct kvec *src, *dst;
1045 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1046 __be32 xid;
1047 __be32 calldir;
1048
1049 xid = *p++;
1050 calldir = *p;
1051
1052 if (bc_xprt)
1053 req = xprt_lookup_rqst(bc_xprt, xid);
1054
1055 if (!req) {
1056 printk(KERN_NOTICE
1057 "%s: Got unrecognized reply: "
1058 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1059 __func__, ntohl(calldir),
1060 bc_xprt, xid);
1061 return -EAGAIN;
1062 }
1063
1064 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1065 /*
1066 * XXX!: cheating for now! Only copying HEAD.
1067 * But we know this is good enough for now (in fact, for any
1068 * callback reply in the forseeable future).
1069 */
1070 dst = &req->rq_private_buf.head[0];
1071 src = &rqstp->rq_arg.head[0];
1072 if (dst->iov_len < src->iov_len)
1073 return -EAGAIN; /* whatever; just giving up. */
1074 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1075 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1076 rqstp->rq_arg.len = 0;
1077 return 0;
1078 }
1079
1080 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1081 {
1082 int i = 0;
1083 int t = 0;
1084
1085 while (t < len) {
1086 vec[i].iov_base = page_address(pages[i]);
1087 vec[i].iov_len = PAGE_SIZE;
1088 i++;
1089 t += PAGE_SIZE;
1090 }
1091 return i;
1092 }
1093
1094
1095 /*
1096 * Receive data from a TCP socket.
1097 */
1098 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1099 {
1100 struct svc_sock *svsk =
1101 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1102 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1103 int len;
1104 struct kvec *vec;
1105 unsigned int want, base;
1106 __be32 *p;
1107 __be32 calldir;
1108 int pnum;
1109
1110 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1111 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1112 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1113 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1114
1115 len = svc_tcp_recv_record(svsk, rqstp);
1116 if (len < 0)
1117 goto error;
1118
1119 base = svc_tcp_restore_pages(svsk, rqstp);
1120 want = svsk->sk_reclen - base;
1121
1122 vec = rqstp->rq_vec;
1123
1124 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1125 svsk->sk_reclen);
1126
1127 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1128
1129 /* Now receive data */
1130 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1131 if (len >= 0)
1132 svsk->sk_tcplen += len;
1133 if (len != want) {
1134 svc_tcp_save_pages(svsk, rqstp);
1135 if (len < 0 && len != -EAGAIN)
1136 goto err_other;
1137 dprintk("svc: incomplete TCP record (%d of %d)\n",
1138 svsk->sk_tcplen, svsk->sk_reclen);
1139 goto err_noclose;
1140 }
1141
1142 rqstp->rq_arg.len = svsk->sk_reclen;
1143 rqstp->rq_arg.page_base = 0;
1144 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1145 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1146 rqstp->rq_arg.page_len = 0;
1147 } else
1148 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1149
1150 rqstp->rq_xprt_ctxt = NULL;
1151 rqstp->rq_prot = IPPROTO_TCP;
1152
1153 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1154 calldir = p[1];
1155 if (calldir)
1156 len = receive_cb_reply(svsk, rqstp);
1157
1158 /* Reset TCP read info */
1159 svsk->sk_reclen = 0;
1160 svsk->sk_tcplen = 0;
1161 /* If we have more data, signal svc_xprt_enqueue() to try again */
1162 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1163 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1164
1165 if (len < 0)
1166 goto error;
1167
1168 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1169 if (serv->sv_stats)
1170 serv->sv_stats->nettcpcnt++;
1171
1172 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1173 return rqstp->rq_arg.len;
1174
1175 error:
1176 if (len != -EAGAIN)
1177 goto err_other;
1178 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1179 return -EAGAIN;
1180 err_other:
1181 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1182 svsk->sk_xprt.xpt_server->sv_name, -len);
1183 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1184 err_noclose:
1185 return -EAGAIN; /* record not complete */
1186 }
1187
1188 /*
1189 * Send out data on TCP socket.
1190 */
1191 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1192 {
1193 struct xdr_buf *xbufp = &rqstp->rq_res;
1194 int sent;
1195 __be32 reclen;
1196
1197 /* Set up the first element of the reply kvec.
1198 * Any other kvecs that may be in use have been taken
1199 * care of by the server implementation itself.
1200 */
1201 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1202 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1203
1204 sent = svc_sendto(rqstp, &rqstp->rq_res);
1205 if (sent != xbufp->len) {
1206 printk(KERN_NOTICE
1207 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1208 "- shutting down socket\n",
1209 rqstp->rq_xprt->xpt_server->sv_name,
1210 (sent<0)?"got error":"sent only",
1211 sent, xbufp->len);
1212 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1213 svc_xprt_enqueue(rqstp->rq_xprt);
1214 sent = -EAGAIN;
1215 }
1216 return sent;
1217 }
1218
1219 /*
1220 * Setup response header. TCP has a 4B record length field.
1221 */
1222 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1223 {
1224 struct kvec *resv = &rqstp->rq_res.head[0];
1225
1226 /* tcp needs a space for the record length... */
1227 svc_putnl(resv, 0);
1228 }
1229
1230 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1231 {
1232 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1233 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1234 int required;
1235
1236 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1237 return 1;
1238 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1239 if (sk_stream_wspace(svsk->sk_sk) >= required)
1240 return 1;
1241 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1242 return 0;
1243 }
1244
1245 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1246 struct net *net,
1247 struct sockaddr *sa, int salen,
1248 int flags)
1249 {
1250 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1251 }
1252
1253 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1254 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1255 struct net *, struct sockaddr *,
1256 int, int);
1257 static void svc_bc_sock_free(struct svc_xprt *xprt);
1258
1259 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1260 struct net *net,
1261 struct sockaddr *sa, int salen,
1262 int flags)
1263 {
1264 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1265 }
1266
1267 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1268 {
1269 }
1270
1271 static struct svc_xprt_ops svc_tcp_bc_ops = {
1272 .xpo_create = svc_bc_tcp_create,
1273 .xpo_detach = svc_bc_tcp_sock_detach,
1274 .xpo_free = svc_bc_sock_free,
1275 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1276 };
1277
1278 static struct svc_xprt_class svc_tcp_bc_class = {
1279 .xcl_name = "tcp-bc",
1280 .xcl_owner = THIS_MODULE,
1281 .xcl_ops = &svc_tcp_bc_ops,
1282 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1283 };
1284
1285 static void svc_init_bc_xprt_sock(void)
1286 {
1287 svc_reg_xprt_class(&svc_tcp_bc_class);
1288 }
1289
1290 static void svc_cleanup_bc_xprt_sock(void)
1291 {
1292 svc_unreg_xprt_class(&svc_tcp_bc_class);
1293 }
1294 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1295 static void svc_init_bc_xprt_sock(void)
1296 {
1297 }
1298
1299 static void svc_cleanup_bc_xprt_sock(void)
1300 {
1301 }
1302 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1303
1304 static struct svc_xprt_ops svc_tcp_ops = {
1305 .xpo_create = svc_tcp_create,
1306 .xpo_recvfrom = svc_tcp_recvfrom,
1307 .xpo_sendto = svc_tcp_sendto,
1308 .xpo_release_rqst = svc_release_skb,
1309 .xpo_detach = svc_tcp_sock_detach,
1310 .xpo_free = svc_sock_free,
1311 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1312 .xpo_has_wspace = svc_tcp_has_wspace,
1313 .xpo_accept = svc_tcp_accept,
1314 };
1315
1316 static struct svc_xprt_class svc_tcp_class = {
1317 .xcl_name = "tcp",
1318 .xcl_owner = THIS_MODULE,
1319 .xcl_ops = &svc_tcp_ops,
1320 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1321 };
1322
1323 void svc_init_xprt_sock(void)
1324 {
1325 svc_reg_xprt_class(&svc_tcp_class);
1326 svc_reg_xprt_class(&svc_udp_class);
1327 svc_init_bc_xprt_sock();
1328 }
1329
1330 void svc_cleanup_xprt_sock(void)
1331 {
1332 svc_unreg_xprt_class(&svc_tcp_class);
1333 svc_unreg_xprt_class(&svc_udp_class);
1334 svc_cleanup_bc_xprt_sock();
1335 }
1336
1337 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1338 {
1339 struct sock *sk = svsk->sk_sk;
1340
1341 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1342 &svsk->sk_xprt, serv);
1343 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1344 if (sk->sk_state == TCP_LISTEN) {
1345 dprintk("setting up TCP socket for listening\n");
1346 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1347 sk->sk_data_ready = svc_tcp_listen_data_ready;
1348 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1349 } else {
1350 dprintk("setting up TCP socket for reading\n");
1351 sk->sk_state_change = svc_tcp_state_change;
1352 sk->sk_data_ready = svc_tcp_data_ready;
1353 sk->sk_write_space = svc_tcp_write_space;
1354
1355 svsk->sk_reclen = 0;
1356 svsk->sk_tcplen = 0;
1357 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1358
1359 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1360
1361 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1362 if (sk->sk_state != TCP_ESTABLISHED)
1363 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1364 }
1365 }
1366
1367 void svc_sock_update_bufs(struct svc_serv *serv)
1368 {
1369 /*
1370 * The number of server threads has changed. Update
1371 * rcvbuf and sndbuf accordingly on all sockets
1372 */
1373 struct svc_sock *svsk;
1374
1375 spin_lock_bh(&serv->sv_lock);
1376 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1377 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1378 spin_unlock_bh(&serv->sv_lock);
1379 }
1380 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1381
1382 /*
1383 * Initialize socket for RPC use and create svc_sock struct
1384 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1385 */
1386 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1387 struct socket *sock,
1388 int flags)
1389 {
1390 struct svc_sock *svsk;
1391 struct sock *inet;
1392 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1393 int err = 0;
1394
1395 dprintk("svc: svc_setup_socket %p\n", sock);
1396 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1397 if (!svsk)
1398 return ERR_PTR(-ENOMEM);
1399
1400 inet = sock->sk;
1401
1402 /* Register socket with portmapper */
1403 if (pmap_register)
1404 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1405 inet->sk_protocol,
1406 ntohs(inet_sk(inet)->inet_sport));
1407
1408 if (err < 0) {
1409 kfree(svsk);
1410 return ERR_PTR(err);
1411 }
1412
1413 inet->sk_user_data = svsk;
1414 svsk->sk_sock = sock;
1415 svsk->sk_sk = inet;
1416 svsk->sk_ostate = inet->sk_state_change;
1417 svsk->sk_odata = inet->sk_data_ready;
1418 svsk->sk_owspace = inet->sk_write_space;
1419
1420 /* Initialize the socket */
1421 if (sock->type == SOCK_DGRAM)
1422 svc_udp_init(svsk, serv);
1423 else {
1424 /* initialise setting must have enough space to
1425 * receive and respond to one request.
1426 */
1427 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1428 4 * serv->sv_max_mesg);
1429 svc_tcp_init(svsk, serv);
1430 }
1431
1432 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1433 svsk, svsk->sk_sk);
1434
1435 return svsk;
1436 }
1437
1438 /**
1439 * svc_addsock - add a listener socket to an RPC service
1440 * @serv: pointer to RPC service to which to add a new listener
1441 * @fd: file descriptor of the new listener
1442 * @name_return: pointer to buffer to fill in with name of listener
1443 * @len: size of the buffer
1444 *
1445 * Fills in socket name and returns positive length of name if successful.
1446 * Name is terminated with '\n'. On error, returns a negative errno
1447 * value.
1448 */
1449 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1450 const size_t len)
1451 {
1452 int err = 0;
1453 struct socket *so = sockfd_lookup(fd, &err);
1454 struct svc_sock *svsk = NULL;
1455 struct sockaddr_storage addr;
1456 struct sockaddr *sin = (struct sockaddr *)&addr;
1457 int salen;
1458
1459 if (!so)
1460 return err;
1461 err = -EAFNOSUPPORT;
1462 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1463 goto out;
1464 err = -EPROTONOSUPPORT;
1465 if (so->sk->sk_protocol != IPPROTO_TCP &&
1466 so->sk->sk_protocol != IPPROTO_UDP)
1467 goto out;
1468 err = -EISCONN;
1469 if (so->state > SS_UNCONNECTED)
1470 goto out;
1471 err = -ENOENT;
1472 if (!try_module_get(THIS_MODULE))
1473 goto out;
1474 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1475 if (IS_ERR(svsk)) {
1476 module_put(THIS_MODULE);
1477 err = PTR_ERR(svsk);
1478 goto out;
1479 }
1480 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1481 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1482 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1483 return svc_one_sock_name(svsk, name_return, len);
1484 out:
1485 sockfd_put(so);
1486 return err;
1487 }
1488 EXPORT_SYMBOL_GPL(svc_addsock);
1489
1490 /*
1491 * Create socket for RPC service.
1492 */
1493 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1494 int protocol,
1495 struct net *net,
1496 struct sockaddr *sin, int len,
1497 int flags)
1498 {
1499 struct svc_sock *svsk;
1500 struct socket *sock;
1501 int error;
1502 int type;
1503 struct sockaddr_storage addr;
1504 struct sockaddr *newsin = (struct sockaddr *)&addr;
1505 int newlen;
1506 int family;
1507 int val;
1508 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1509
1510 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1511 serv->sv_program->pg_name, protocol,
1512 __svc_print_addr(sin, buf, sizeof(buf)));
1513
1514 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1515 printk(KERN_WARNING "svc: only UDP and TCP "
1516 "sockets supported\n");
1517 return ERR_PTR(-EINVAL);
1518 }
1519
1520 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1521 switch (sin->sa_family) {
1522 case AF_INET6:
1523 family = PF_INET6;
1524 break;
1525 case AF_INET:
1526 family = PF_INET;
1527 break;
1528 default:
1529 return ERR_PTR(-EINVAL);
1530 }
1531
1532 error = __sock_create(net, family, type, protocol, &sock, 1);
1533 if (error < 0)
1534 return ERR_PTR(error);
1535
1536 svc_reclassify_socket(sock);
1537
1538 /*
1539 * If this is an PF_INET6 listener, we want to avoid
1540 * getting requests from IPv4 remotes. Those should
1541 * be shunted to a PF_INET listener via rpcbind.
1542 */
1543 val = 1;
1544 if (family == PF_INET6)
1545 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1546 (char *)&val, sizeof(val));
1547
1548 if (type == SOCK_STREAM)
1549 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1550 error = kernel_bind(sock, sin, len);
1551 if (error < 0)
1552 goto bummer;
1553
1554 newlen = len;
1555 error = kernel_getsockname(sock, newsin, &newlen);
1556 if (error < 0)
1557 goto bummer;
1558
1559 if (protocol == IPPROTO_TCP) {
1560 if ((error = kernel_listen(sock, 64)) < 0)
1561 goto bummer;
1562 }
1563
1564 svsk = svc_setup_socket(serv, sock, flags);
1565 if (IS_ERR(svsk)) {
1566 error = PTR_ERR(svsk);
1567 goto bummer;
1568 }
1569 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1570 return (struct svc_xprt *)svsk;
1571 bummer:
1572 dprintk("svc: svc_create_socket error = %d\n", -error);
1573 sock_release(sock);
1574 return ERR_PTR(error);
1575 }
1576
1577 /*
1578 * Detach the svc_sock from the socket so that no
1579 * more callbacks occur.
1580 */
1581 static void svc_sock_detach(struct svc_xprt *xprt)
1582 {
1583 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1584 struct sock *sk = svsk->sk_sk;
1585 wait_queue_head_t *wq;
1586
1587 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1588
1589 /* put back the old socket callbacks */
1590 sk->sk_state_change = svsk->sk_ostate;
1591 sk->sk_data_ready = svsk->sk_odata;
1592 sk->sk_write_space = svsk->sk_owspace;
1593
1594 wq = sk_sleep(sk);
1595 if (wq && waitqueue_active(wq))
1596 wake_up_interruptible(wq);
1597 }
1598
1599 /*
1600 * Disconnect the socket, and reset the callbacks
1601 */
1602 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1603 {
1604 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1605
1606 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1607
1608 svc_sock_detach(xprt);
1609
1610 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1611 svc_tcp_clear_pages(svsk);
1612 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1613 }
1614 }
1615
1616 /*
1617 * Free the svc_sock's socket resources and the svc_sock itself.
1618 */
1619 static void svc_sock_free(struct svc_xprt *xprt)
1620 {
1621 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1622 dprintk("svc: svc_sock_free(%p)\n", svsk);
1623
1624 if (svsk->sk_sock->file)
1625 sockfd_put(svsk->sk_sock);
1626 else
1627 sock_release(svsk->sk_sock);
1628 kfree(svsk);
1629 }
1630
1631 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1632 /*
1633 * Create a back channel svc_xprt which shares the fore channel socket.
1634 */
1635 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1636 int protocol,
1637 struct net *net,
1638 struct sockaddr *sin, int len,
1639 int flags)
1640 {
1641 struct svc_sock *svsk;
1642 struct svc_xprt *xprt;
1643
1644 if (protocol != IPPROTO_TCP) {
1645 printk(KERN_WARNING "svc: only TCP sockets"
1646 " supported on shared back channel\n");
1647 return ERR_PTR(-EINVAL);
1648 }
1649
1650 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1651 if (!svsk)
1652 return ERR_PTR(-ENOMEM);
1653
1654 xprt = &svsk->sk_xprt;
1655 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1656
1657 serv->sv_bc_xprt = xprt;
1658
1659 return xprt;
1660 }
1661
1662 /*
1663 * Free a back channel svc_sock.
1664 */
1665 static void svc_bc_sock_free(struct svc_xprt *xprt)
1666 {
1667 if (xprt)
1668 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1669 }
1670 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Ubuntu Artful kernel mirror