?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
FRV: Use generic show_interrupts()
[cris-mirror.git] / net / sunrpc / svcsock.c
blobb7d435c3f19ec537275cb8b413d6c929c4dc8fa6
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/errno.h>
25 #include <linux/fcntl.h>
26 #include <linux/net.h>
27 #include <linux/in.h>
28 #include <linux/inet.h>
29 #include <linux/udp.h>
30 #include <linux/tcp.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/netdevice.h>
34 #include <linux/skbuff.h>
35 #include <linux/file.h>
36 #include <linux/freezer.h>
37 #include <net/sock.h>
38 #include <net/checksum.h>
39 #include <net/ip.h>
40 #include <net/ipv6.h>
41 #include <net/tcp.h>
42 #include <net/tcp_states.h>
43 #include <asm/uaccess.h>
44 #include <asm/ioctls.h>
45
46 #include <linux/sunrpc/types.h>
47 #include <linux/sunrpc/clnt.h>
48 #include <linux/sunrpc/xdr.h>
49 #include <linux/sunrpc/msg_prot.h>
50 #include <linux/sunrpc/svcsock.h>
51 #include <linux/sunrpc/stats.h>
52 #include <linux/sunrpc/xprt.h>
53
54 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
55
56
57 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
58 int *errp, int flags);
59 static void svc_udp_data_ready(struct sock *, int);
60 static int svc_udp_recvfrom(struct svc_rqst *);
61 static int svc_udp_sendto(struct svc_rqst *);
62 static void svc_sock_detach(struct svc_xprt *);
63 static void svc_tcp_sock_detach(struct svc_xprt *);
64 static void svc_sock_free(struct svc_xprt *);
65
66 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
67 struct net *, struct sockaddr *,
68 int, int);
69 #if defined(CONFIG_NFS_V4_1)
70 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73 static void svc_bc_sock_free(struct svc_xprt *xprt);
74 #endif /* CONFIG_NFS_V4_1 */
75
76 #ifdef CONFIG_DEBUG_LOCK_ALLOC
77 static struct lock_class_key svc_key[2];
78 static struct lock_class_key svc_slock_key[2];
79
80 static void svc_reclassify_socket(struct socket *sock)
81 {
82 struct sock *sk = sock->sk;
83 BUG_ON(sock_owned_by_user(sk));
84 switch (sk->sk_family) {
85 case AF_INET:
86 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
87 &svc_slock_key[0],
88 "sk_xprt.xpt_lock-AF_INET-NFSD",
89 &svc_key[0]);
90 break;
91
92 case AF_INET6:
93 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
94 &svc_slock_key[1],
95 "sk_xprt.xpt_lock-AF_INET6-NFSD",
96 &svc_key[1]);
97 break;
98
99 default:
100 BUG();
101 }
102 }
103 #else
104 static void svc_reclassify_socket(struct socket *sock)
105 {
106 }
107 #endif
108
109 /*
110 * Release an skbuff after use
111 */
112 static void svc_release_skb(struct svc_rqst *rqstp)
113 {
114 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
115
116 if (skb) {
117 struct svc_sock *svsk =
118 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
119 rqstp->rq_xprt_ctxt = NULL;
120
121 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
122 skb_free_datagram_locked(svsk->sk_sk, skb);
123 }
124 }
125
126 union svc_pktinfo_u {
127 struct in_pktinfo pkti;
128 struct in6_pktinfo pkti6;
129 };
130 #define SVC_PKTINFO_SPACE \
131 CMSG_SPACE(sizeof(union svc_pktinfo_u))
132
133 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
134 {
135 struct svc_sock *svsk =
136 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
137 switch (svsk->sk_sk->sk_family) {
138 case AF_INET: {
139 struct in_pktinfo *pki = CMSG_DATA(cmh);
140
141 cmh->cmsg_level = SOL_IP;
142 cmh->cmsg_type = IP_PKTINFO;
143 pki->ipi_ifindex = 0;
144 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
145 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
146 }
147 break;
148
149 case AF_INET6: {
150 struct in6_pktinfo *pki = CMSG_DATA(cmh);
151
152 cmh->cmsg_level = SOL_IPV6;
153 cmh->cmsg_type = IPV6_PKTINFO;
154 pki->ipi6_ifindex = 0;
155 ipv6_addr_copy(&pki->ipi6_addr,
156 &rqstp->rq_daddr.addr6);
157 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
158 }
159 break;
160 }
161 }
162
163 /*
164 * send routine intended to be shared by the fore- and back-channel
165 */
166 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
167 struct page *headpage, unsigned long headoffset,
168 struct page *tailpage, unsigned long tailoffset)
169 {
170 int result;
171 int size;
172 struct page **ppage = xdr->pages;
173 size_t base = xdr->page_base;
174 unsigned int pglen = xdr->page_len;
175 unsigned int flags = MSG_MORE;
176 int slen;
177 int len = 0;
178
179 slen = xdr->len;
180
181 /* send head */
182 if (slen == xdr->head[0].iov_len)
183 flags = 0;
184 len = kernel_sendpage(sock, headpage, headoffset,
185 xdr->head[0].iov_len, flags);
186 if (len != xdr->head[0].iov_len)
187 goto out;
188 slen -= xdr->head[0].iov_len;
189 if (slen == 0)
190 goto out;
191
192 /* send page data */
193 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
194 while (pglen > 0) {
195 if (slen == size)
196 flags = 0;
197 result = kernel_sendpage(sock, *ppage, base, size, flags);
198 if (result > 0)
199 len += result;
200 if (result != size)
201 goto out;
202 slen -= size;
203 pglen -= size;
204 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
205 base = 0;
206 ppage++;
207 }
208
209 /* send tail */
210 if (xdr->tail[0].iov_len) {
211 result = kernel_sendpage(sock, tailpage, tailoffset,
212 xdr->tail[0].iov_len, 0);
213 if (result > 0)
214 len += result;
215 }
216
217 out:
218 return len;
219 }
220
221
222 /*
223 * Generic sendto routine
224 */
225 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
226 {
227 struct svc_sock *svsk =
228 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
229 struct socket *sock = svsk->sk_sock;
230 union {
231 struct cmsghdr hdr;
232 long all[SVC_PKTINFO_SPACE / sizeof(long)];
233 } buffer;
234 struct cmsghdr *cmh = &buffer.hdr;
235 int len = 0;
236 unsigned long tailoff;
237 unsigned long headoff;
238 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
239
240 if (rqstp->rq_prot == IPPROTO_UDP) {
241 struct msghdr msg = {
242 .msg_name = &rqstp->rq_addr,
243 .msg_namelen = rqstp->rq_addrlen,
244 .msg_control = cmh,
245 .msg_controllen = sizeof(buffer),
246 .msg_flags = MSG_MORE,
247 };
248
249 svc_set_cmsg_data(rqstp, cmh);
250
251 if (sock_sendmsg(sock, &msg, 0) < 0)
252 goto out;
253 }
254
255 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
256 headoff = 0;
257 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
258 rqstp->rq_respages[0], tailoff);
259
260 out:
261 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
262 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
263 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
264
265 return len;
266 }
267
268 /*
269 * Report socket names for nfsdfs
270 */
271 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
272 {
273 const struct sock *sk = svsk->sk_sk;
274 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
275 "udp" : "tcp";
276 int len;
277
278 switch (sk->sk_family) {
279 case PF_INET:
280 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
281 proto_name,
282 &inet_sk(sk)->inet_rcv_saddr,
283 inet_sk(sk)->inet_num);
284 break;
285 case PF_INET6:
286 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
287 proto_name,
288 &inet6_sk(sk)->rcv_saddr,
289 inet_sk(sk)->inet_num);
290 break;
291 default:
292 len = snprintf(buf, remaining, "*unknown-%d*\n",
293 sk->sk_family);
294 }
295
296 if (len >= remaining) {
297 *buf = '\0';
298 return -ENAMETOOLONG;
299 }
300 return len;
301 }
302
303 /**
304 * svc_sock_names - construct a list of listener names in a string
305 * @serv: pointer to RPC service
306 * @buf: pointer to a buffer to fill in with socket names
307 * @buflen: size of the buffer to be filled
308 * @toclose: pointer to '\0'-terminated C string containing the name
309 * of a listener to be closed
310 *
311 * Fills in @buf with a '\n'-separated list of names of listener
312 * sockets. If @toclose is not NULL, the socket named by @toclose
313 * is closed, and is not included in the output list.
314 *
315 * Returns positive length of the socket name string, or a negative
316 * errno value on error.
317 */
318 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
319 const char *toclose)
320 {
321 struct svc_sock *svsk, *closesk = NULL;
322 int len = 0;
323
324 if (!serv)
325 return 0;
326
327 spin_lock_bh(&serv->sv_lock);
328 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
329 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
330 if (onelen < 0) {
331 len = onelen;
332 break;
333 }
334 if (toclose && strcmp(toclose, buf + len) == 0) {
335 closesk = svsk;
336 svc_xprt_get(&closesk->sk_xprt);
337 } else
338 len += onelen;
339 }
340 spin_unlock_bh(&serv->sv_lock);
341
342 if (closesk) {
343 /* Should unregister with portmap, but you cannot
344 * unregister just one protocol...
345 */
346 svc_close_xprt(&closesk->sk_xprt);
347 svc_xprt_put(&closesk->sk_xprt);
348 } else if (toclose)
349 return -ENOENT;
350 return len;
351 }
352 EXPORT_SYMBOL_GPL(svc_sock_names);
353
354 /*
355 * Check input queue length
356 */
357 static int svc_recv_available(struct svc_sock *svsk)
358 {
359 struct socket *sock = svsk->sk_sock;
360 int avail, err;
361
362 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
363
364 return (err >= 0)? avail : err;
365 }
366
367 /*
368 * Generic recvfrom routine.
369 */
370 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
371 int buflen)
372 {
373 struct svc_sock *svsk =
374 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
375 struct msghdr msg = {
376 .msg_flags = MSG_DONTWAIT,
377 };
378 int len;
379
380 rqstp->rq_xprt_hlen = 0;
381
382 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
383 msg.msg_flags);
384
385 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
386 svsk, iov[0].iov_base, iov[0].iov_len, len);
387 return len;
388 }
389
390 /*
391 * Set socket snd and rcv buffer lengths
392 */
393 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
394 unsigned int rcv)
395 {
396 #if 0
397 mm_segment_t oldfs;
398 oldfs = get_fs(); set_fs(KERNEL_DS);
399 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
400 (char*)&snd, sizeof(snd));
401 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
402 (char*)&rcv, sizeof(rcv));
403 #else
404 /* sock_setsockopt limits use to sysctl_?mem_max,
405 * which isn't acceptable. Until that is made conditional
406 * on not having CAP_SYS_RESOURCE or similar, we go direct...
407 * DaveM said I could!
408 */
409 lock_sock(sock->sk);
410 sock->sk->sk_sndbuf = snd * 2;
411 sock->sk->sk_rcvbuf = rcv * 2;
412 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
413 sock->sk->sk_write_space(sock->sk);
414 release_sock(sock->sk);
415 #endif
416 }
417 /*
418 * INET callback when data has been received on the socket.
419 */
420 static void svc_udp_data_ready(struct sock *sk, int count)
421 {
422 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
423 wait_queue_head_t *wq = sk_sleep(sk);
424
425 if (svsk) {
426 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
427 svsk, sk, count,
428 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
429 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
430 svc_xprt_enqueue(&svsk->sk_xprt);
431 }
432 if (wq && waitqueue_active(wq))
433 wake_up_interruptible(wq);
434 }
435
436 /*
437 * INET callback when space is newly available on the socket.
438 */
439 static void svc_write_space(struct sock *sk)
440 {
441 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
442 wait_queue_head_t *wq = sk_sleep(sk);
443
444 if (svsk) {
445 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
446 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
447 svc_xprt_enqueue(&svsk->sk_xprt);
448 }
449
450 if (wq && waitqueue_active(wq)) {
451 dprintk("RPC svc_write_space: someone sleeping on %p\n",
452 svsk);
453 wake_up_interruptible(wq);
454 }
455 }
456
457 static void svc_tcp_write_space(struct sock *sk)
458 {
459 struct socket *sock = sk->sk_socket;
460
461 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
462 clear_bit(SOCK_NOSPACE, &sock->flags);
463 svc_write_space(sk);
464 }
465
466 /*
467 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
468 */
469 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
470 struct cmsghdr *cmh)
471 {
472 struct in_pktinfo *pki = CMSG_DATA(cmh);
473 if (cmh->cmsg_type != IP_PKTINFO)
474 return 0;
475 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
476 return 1;
477 }
478
479 /*
480 * See net/ipv6/datagram.c : datagram_recv_ctl
481 */
482 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
483 struct cmsghdr *cmh)
484 {
485 struct in6_pktinfo *pki = CMSG_DATA(cmh);
486 if (cmh->cmsg_type != IPV6_PKTINFO)
487 return 0;
488 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
489 return 1;
490 }
491
492 /*
493 * Copy the UDP datagram's destination address to the rqstp structure.
494 * The 'destination' address in this case is the address to which the
495 * peer sent the datagram, i.e. our local address. For multihomed
496 * hosts, this can change from msg to msg. Note that only the IP
497 * address changes, the port number should remain the same.
498 */
499 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
500 struct cmsghdr *cmh)
501 {
502 switch (cmh->cmsg_level) {
503 case SOL_IP:
504 return svc_udp_get_dest_address4(rqstp, cmh);
505 case SOL_IPV6:
506 return svc_udp_get_dest_address6(rqstp, cmh);
507 }
508
509 return 0;
510 }
511
512 /*
513 * Receive a datagram from a UDP socket.
514 */
515 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
516 {
517 struct svc_sock *svsk =
518 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
519 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
520 struct sk_buff *skb;
521 union {
522 struct cmsghdr hdr;
523 long all[SVC_PKTINFO_SPACE / sizeof(long)];
524 } buffer;
525 struct cmsghdr *cmh = &buffer.hdr;
526 struct msghdr msg = {
527 .msg_name = svc_addr(rqstp),
528 .msg_control = cmh,
529 .msg_controllen = sizeof(buffer),
530 .msg_flags = MSG_DONTWAIT,
531 };
532 size_t len;
533 int err;
534
535 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
536 /* udp sockets need large rcvbuf as all pending
537 * requests are still in that buffer. sndbuf must
538 * also be large enough that there is enough space
539 * for one reply per thread. We count all threads
540 * rather than threads in a particular pool, which
541 * provides an upper bound on the number of threads
542 * which will access the socket.
543 */
544 svc_sock_setbufsize(svsk->sk_sock,
545 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
546 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
547
548 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
549 skb = NULL;
550 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
551 0, 0, MSG_PEEK | MSG_DONTWAIT);
552 if (err >= 0)
553 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
554
555 if (skb == NULL) {
556 if (err != -EAGAIN) {
557 /* possibly an icmp error */
558 dprintk("svc: recvfrom returned error %d\n", -err);
559 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
560 }
561 return -EAGAIN;
562 }
563 len = svc_addr_len(svc_addr(rqstp));
564 if (len == 0)
565 return -EAFNOSUPPORT;
566 rqstp->rq_addrlen = len;
567 if (skb->tstamp.tv64 == 0) {
568 skb->tstamp = ktime_get_real();
569 /* Don't enable netstamp, sunrpc doesn't
570 need that much accuracy */
571 }
572 svsk->sk_sk->sk_stamp = skb->tstamp;
573 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
574
575 len = skb->len - sizeof(struct udphdr);
576 rqstp->rq_arg.len = len;
577
578 rqstp->rq_prot = IPPROTO_UDP;
579
580 if (!svc_udp_get_dest_address(rqstp, cmh)) {
581 if (net_ratelimit())
582 printk(KERN_WARNING
583 "svc: received unknown control message %d/%d; "
584 "dropping RPC reply datagram\n",
585 cmh->cmsg_level, cmh->cmsg_type);
586 skb_free_datagram_locked(svsk->sk_sk, skb);
587 return 0;
588 }
589
590 if (skb_is_nonlinear(skb)) {
591 /* we have to copy */
592 local_bh_disable();
593 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
594 local_bh_enable();
595 /* checksum error */
596 skb_free_datagram_locked(svsk->sk_sk, skb);
597 return 0;
598 }
599 local_bh_enable();
600 skb_free_datagram_locked(svsk->sk_sk, skb);
601 } else {
602 /* we can use it in-place */
603 rqstp->rq_arg.head[0].iov_base = skb->data +
604 sizeof(struct udphdr);
605 rqstp->rq_arg.head[0].iov_len = len;
606 if (skb_checksum_complete(skb)) {
607 skb_free_datagram_locked(svsk->sk_sk, skb);
608 return 0;
609 }
610 rqstp->rq_xprt_ctxt = skb;
611 }
612
613 rqstp->rq_arg.page_base = 0;
614 if (len <= rqstp->rq_arg.head[0].iov_len) {
615 rqstp->rq_arg.head[0].iov_len = len;
616 rqstp->rq_arg.page_len = 0;
617 rqstp->rq_respages = rqstp->rq_pages+1;
618 } else {
619 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
620 rqstp->rq_respages = rqstp->rq_pages + 1 +
621 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
622 }
623
624 if (serv->sv_stats)
625 serv->sv_stats->netudpcnt++;
626
627 return len;
628 }
629
630 static int
631 svc_udp_sendto(struct svc_rqst *rqstp)
632 {
633 int error;
634
635 error = svc_sendto(rqstp, &rqstp->rq_res);
636 if (error == -ECONNREFUSED)
637 /* ICMP error on earlier request. */
638 error = svc_sendto(rqstp, &rqstp->rq_res);
639
640 return error;
641 }
642
643 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
644 {
645 }
646
647 static int svc_udp_has_wspace(struct svc_xprt *xprt)
648 {
649 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
650 struct svc_serv *serv = xprt->xpt_server;
651 unsigned long required;
652
653 /*
654 * Set the SOCK_NOSPACE flag before checking the available
655 * sock space.
656 */
657 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
658 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
659 if (required*2 > sock_wspace(svsk->sk_sk))
660 return 0;
661 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
662 return 1;
663 }
664
665 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
666 {
667 BUG();
668 return NULL;
669 }
670
671 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
672 struct net *net,
673 struct sockaddr *sa, int salen,
674 int flags)
675 {
676 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
677 }
678
679 static struct svc_xprt_ops svc_udp_ops = {
680 .xpo_create = svc_udp_create,
681 .xpo_recvfrom = svc_udp_recvfrom,
682 .xpo_sendto = svc_udp_sendto,
683 .xpo_release_rqst = svc_release_skb,
684 .xpo_detach = svc_sock_detach,
685 .xpo_free = svc_sock_free,
686 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
687 .xpo_has_wspace = svc_udp_has_wspace,
688 .xpo_accept = svc_udp_accept,
689 };
690
691 static struct svc_xprt_class svc_udp_class = {
692 .xcl_name = "udp",
693 .xcl_owner = THIS_MODULE,
694 .xcl_ops = &svc_udp_ops,
695 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
696 };
697
698 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
699 {
700 int err, level, optname, one = 1;
701
702 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
703 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
704 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
705 svsk->sk_sk->sk_write_space = svc_write_space;
706
707 /* initialise setting must have enough space to
708 * receive and respond to one request.
709 * svc_udp_recvfrom will re-adjust if necessary
710 */
711 svc_sock_setbufsize(svsk->sk_sock,
712 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
713 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
714
715 /* data might have come in before data_ready set up */
716 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
717 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
718
719 /* make sure we get destination address info */
720 switch (svsk->sk_sk->sk_family) {
721 case AF_INET:
722 level = SOL_IP;
723 optname = IP_PKTINFO;
724 break;
725 case AF_INET6:
726 level = SOL_IPV6;
727 optname = IPV6_RECVPKTINFO;
728 break;
729 default:
730 BUG();
731 }
732 err = kernel_setsockopt(svsk->sk_sock, level, optname,
733 (char *)&one, sizeof(one));
734 dprintk("svc: kernel_setsockopt returned %d\n", err);
735 }
736
737 /*
738 * A data_ready event on a listening socket means there's a connection
739 * pending. Do not use state_change as a substitute for it.
740 */
741 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
742 {
743 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
744 wait_queue_head_t *wq;
745
746 dprintk("svc: socket %p TCP (listen) state change %d\n",
747 sk, sk->sk_state);
748
749 /*
750 * This callback may called twice when a new connection
751 * is established as a child socket inherits everything
752 * from a parent LISTEN socket.
753 * 1) data_ready method of the parent socket will be called
754 * when one of child sockets become ESTABLISHED.
755 * 2) data_ready method of the child socket may be called
756 * when it receives data before the socket is accepted.
757 * In case of 2, we should ignore it silently.
758 */
759 if (sk->sk_state == TCP_LISTEN) {
760 if (svsk) {
761 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
762 svc_xprt_enqueue(&svsk->sk_xprt);
763 } else
764 printk("svc: socket %p: no user data\n", sk);
765 }
766
767 wq = sk_sleep(sk);
768 if (wq && waitqueue_active(wq))
769 wake_up_interruptible_all(wq);
770 }
771
772 /*
773 * A state change on a connected socket means it's dying or dead.
774 */
775 static void svc_tcp_state_change(struct sock *sk)
776 {
777 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
778 wait_queue_head_t *wq = sk_sleep(sk);
779
780 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
781 sk, sk->sk_state, sk->sk_user_data);
782
783 if (!svsk)
784 printk("svc: socket %p: no user data\n", sk);
785 else {
786 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
787 svc_xprt_enqueue(&svsk->sk_xprt);
788 }
789 if (wq && waitqueue_active(wq))
790 wake_up_interruptible_all(wq);
791 }
792
793 static void svc_tcp_data_ready(struct sock *sk, int count)
794 {
795 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
796 wait_queue_head_t *wq = sk_sleep(sk);
797
798 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
799 sk, sk->sk_user_data);
800 if (svsk) {
801 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
802 svc_xprt_enqueue(&svsk->sk_xprt);
803 }
804 if (wq && waitqueue_active(wq))
805 wake_up_interruptible(wq);
806 }
807
808 /*
809 * Accept a TCP connection
810 */
811 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
812 {
813 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
814 struct sockaddr_storage addr;
815 struct sockaddr *sin = (struct sockaddr *) &addr;
816 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
817 struct socket *sock = svsk->sk_sock;
818 struct socket *newsock;
819 struct svc_sock *newsvsk;
820 int err, slen;
821 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
822
823 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
824 if (!sock)
825 return NULL;
826
827 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
828 err = kernel_accept(sock, &newsock, O_NONBLOCK);
829 if (err < 0) {
830 if (err == -ENOMEM)
831 printk(KERN_WARNING "%s: no more sockets!\n",
832 serv->sv_name);
833 else if (err != -EAGAIN && net_ratelimit())
834 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
835 serv->sv_name, -err);
836 return NULL;
837 }
838 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
839
840 err = kernel_getpeername(newsock, sin, &slen);
841 if (err < 0) {
842 if (net_ratelimit())
843 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
844 serv->sv_name, -err);
845 goto failed; /* aborted connection or whatever */
846 }
847
848 /* Ideally, we would want to reject connections from unauthorized
849 * hosts here, but when we get encryption, the IP of the host won't
850 * tell us anything. For now just warn about unpriv connections.
851 */
852 if (!svc_port_is_privileged(sin)) {
853 dprintk(KERN_WARNING
854 "%s: connect from unprivileged port: %s\n",
855 serv->sv_name,
856 __svc_print_addr(sin, buf, sizeof(buf)));
857 }
858 dprintk("%s: connect from %s\n", serv->sv_name,
859 __svc_print_addr(sin, buf, sizeof(buf)));
860
861 /* make sure that a write doesn't block forever when
862 * low on memory
863 */
864 newsock->sk->sk_sndtimeo = HZ*30;
865
866 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
867 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
868 goto failed;
869 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
870 err = kernel_getsockname(newsock, sin, &slen);
871 if (unlikely(err < 0)) {
872 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
873 slen = offsetof(struct sockaddr, sa_data);
874 }
875 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
876
877 if (serv->sv_stats)
878 serv->sv_stats->nettcpconn++;
879
880 return &newsvsk->sk_xprt;
881
882 failed:
883 sock_release(newsock);
884 return NULL;
885 }
886
887 /*
888 * Receive data.
889 * If we haven't gotten the record length yet, get the next four bytes.
890 * Otherwise try to gobble up as much as possible up to the complete
891 * record length.
892 */
893 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
894 {
895 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
896 int len;
897
898 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
899 /* sndbuf needs to have room for one request
900 * per thread, otherwise we can stall even when the
901 * network isn't a bottleneck.
902 *
903 * We count all threads rather than threads in a
904 * particular pool, which provides an upper bound
905 * on the number of threads which will access the socket.
906 *
907 * rcvbuf just needs to be able to hold a few requests.
908 * Normally they will be removed from the queue
909 * as soon a a complete request arrives.
910 */
911 svc_sock_setbufsize(svsk->sk_sock,
912 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
913 3 * serv->sv_max_mesg);
914
915 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
916
917 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
918 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
919 struct kvec iov;
920
921 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
922 iov.iov_len = want;
923 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
924 goto error;
925 svsk->sk_tcplen += len;
926
927 if (len < want) {
928 dprintk("svc: short recvfrom while reading record "
929 "length (%d of %d)\n", len, want);
930 goto err_again; /* record header not complete */
931 }
932
933 svsk->sk_reclen = ntohl(svsk->sk_reclen);
934 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
935 /* FIXME: technically, a record can be fragmented,
936 * and non-terminal fragments will not have the top
937 * bit set in the fragment length header.
938 * But apparently no known nfs clients send fragmented
939 * records. */
940 if (net_ratelimit())
941 printk(KERN_NOTICE "RPC: multiple fragments "
942 "per record not supported\n");
943 goto err_delete;
944 }
945
946 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
947 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
948 if (svsk->sk_reclen > serv->sv_max_mesg) {
949 if (net_ratelimit())
950 printk(KERN_NOTICE "RPC: "
951 "fragment too large: 0x%08lx\n",
952 (unsigned long)svsk->sk_reclen);
953 goto err_delete;
954 }
955 }
956
957 /* Check whether enough data is available */
958 len = svc_recv_available(svsk);
959 if (len < 0)
960 goto error;
961
962 if (len < svsk->sk_reclen) {
963 dprintk("svc: incomplete TCP record (%d of %d)\n",
964 len, svsk->sk_reclen);
965 goto err_again; /* record not complete */
966 }
967 len = svsk->sk_reclen;
968 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
969
970 return len;
971 error:
972 if (len == -EAGAIN)
973 dprintk("RPC: TCP recv_record got EAGAIN\n");
974 return len;
975 err_delete:
976 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
977 err_again:
978 return -EAGAIN;
979 }
980
981 static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
982 struct rpc_rqst **reqpp, struct kvec *vec)
983 {
984 struct rpc_rqst *req = NULL;
985 u32 *p;
986 u32 xid;
987 u32 calldir;
988 int len;
989
990 len = svc_recvfrom(rqstp, vec, 1, 8);
991 if (len < 0)
992 goto error;
993
994 p = (u32 *)rqstp->rq_arg.head[0].iov_base;
995 xid = *p++;
996 calldir = *p;
997
998 if (calldir == 0) {
999 /* REQUEST is the most common case */
1000 vec[0] = rqstp->rq_arg.head[0];
1001 } else {
1002 /* REPLY */
1003 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1004
1005 if (bc_xprt)
1006 req = xprt_lookup_rqst(bc_xprt, xid);
1007
1008 if (!req) {
1009 printk(KERN_NOTICE
1010 "%s: Got unrecognized reply: "
1011 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1012 __func__, ntohl(calldir),
1013 bc_xprt, xid);
1014 vec[0] = rqstp->rq_arg.head[0];
1015 goto out;
1016 }
1017
1018 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1019 sizeof(struct xdr_buf));
1020 /* copy the xid and call direction */
1021 memcpy(req->rq_private_buf.head[0].iov_base,
1022 rqstp->rq_arg.head[0].iov_base, 8);
1023 vec[0] = req->rq_private_buf.head[0];
1024 }
1025 out:
1026 vec[0].iov_base += 8;
1027 vec[0].iov_len -= 8;
1028 len = svsk->sk_reclen - 8;
1029 error:
1030 *reqpp = req;
1031 return len;
1032 }
1033
1034 /*
1035 * Receive data from a TCP socket.
1036 */
1037 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1038 {
1039 struct svc_sock *svsk =
1040 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1041 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1042 int len;
1043 struct kvec *vec;
1044 int pnum, vlen;
1045 struct rpc_rqst *req = NULL;
1046
1047 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1048 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1049 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1050 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1051
1052 len = svc_tcp_recv_record(svsk, rqstp);
1053 if (len < 0)
1054 goto error;
1055
1056 vec = rqstp->rq_vec;
1057 vec[0] = rqstp->rq_arg.head[0];
1058 vlen = PAGE_SIZE;
1059
1060 /*
1061 * We have enough data for the whole tcp record. Let's try and read the
1062 * first 8 bytes to get the xid and the call direction. We can use this
1063 * to figure out if this is a call or a reply to a callback. If
1064 * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
1065 * In that case, don't bother with the calldir and just read the data.
1066 * It will be rejected in svc_process.
1067 */
1068 if (len >= 8) {
1069 len = svc_process_calldir(svsk, rqstp, &req, vec);
1070 if (len < 0)
1071 goto err_again;
1072 vlen -= 8;
1073 }
1074
1075 pnum = 1;
1076 while (vlen < len) {
1077 vec[pnum].iov_base = (req) ?
1078 page_address(req->rq_private_buf.pages[pnum - 1]) :
1079 page_address(rqstp->rq_pages[pnum]);
1080 vec[pnum].iov_len = PAGE_SIZE;
1081 pnum++;
1082 vlen += PAGE_SIZE;
1083 }
1084 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1085
1086 /* Now receive data */
1087 len = svc_recvfrom(rqstp, vec, pnum, len);
1088 if (len < 0)
1089 goto err_again;
1090
1091 /*
1092 * Account for the 8 bytes we read earlier
1093 */
1094 len += 8;
1095
1096 if (req) {
1097 xprt_complete_rqst(req->rq_task, len);
1098 len = 0;
1099 goto out;
1100 }
1101 dprintk("svc: TCP complete record (%d bytes)\n", len);
1102 rqstp->rq_arg.len = len;
1103 rqstp->rq_arg.page_base = 0;
1104 if (len <= rqstp->rq_arg.head[0].iov_len) {
1105 rqstp->rq_arg.head[0].iov_len = len;
1106 rqstp->rq_arg.page_len = 0;
1107 } else {
1108 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1109 }
1110
1111 rqstp->rq_xprt_ctxt = NULL;
1112 rqstp->rq_prot = IPPROTO_TCP;
1113
1114 out:
1115 /* Reset TCP read info */
1116 svsk->sk_reclen = 0;
1117 svsk->sk_tcplen = 0;
1118
1119 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1120 if (serv->sv_stats)
1121 serv->sv_stats->nettcpcnt++;
1122
1123 return len;
1124
1125 err_again:
1126 if (len == -EAGAIN) {
1127 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1128 return len;
1129 }
1130 error:
1131 if (len != -EAGAIN) {
1132 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1133 svsk->sk_xprt.xpt_server->sv_name, -len);
1134 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1135 }
1136 return -EAGAIN;
1137 }
1138
1139 /*
1140 * Send out data on TCP socket.
1141 */
1142 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1143 {
1144 struct xdr_buf *xbufp = &rqstp->rq_res;
1145 int sent;
1146 __be32 reclen;
1147
1148 /* Set up the first element of the reply kvec.
1149 * Any other kvecs that may be in use have been taken
1150 * care of by the server implementation itself.
1151 */
1152 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1153 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1154
1155 sent = svc_sendto(rqstp, &rqstp->rq_res);
1156 if (sent != xbufp->len) {
1157 printk(KERN_NOTICE
1158 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1159 "- shutting down socket\n",
1160 rqstp->rq_xprt->xpt_server->sv_name,
1161 (sent<0)?"got error":"sent only",
1162 sent, xbufp->len);
1163 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1164 svc_xprt_enqueue(rqstp->rq_xprt);
1165 sent = -EAGAIN;
1166 }
1167 return sent;
1168 }
1169
1170 /*
1171 * Setup response header. TCP has a 4B record length field.
1172 */
1173 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1174 {
1175 struct kvec *resv = &rqstp->rq_res.head[0];
1176
1177 /* tcp needs a space for the record length... */
1178 svc_putnl(resv, 0);
1179 }
1180
1181 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1182 {
1183 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1184 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1185 int required;
1186
1187 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1188 return 1;
1189 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1190 if (sk_stream_wspace(svsk->sk_sk) >= required)
1191 return 1;
1192 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1193 return 0;
1194 }
1195
1196 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1197 struct net *net,
1198 struct sockaddr *sa, int salen,
1199 int flags)
1200 {
1201 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1202 }
1203
1204 #if defined(CONFIG_NFS_V4_1)
1205 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1206 struct net *, struct sockaddr *,
1207 int, int);
1208 static void svc_bc_sock_free(struct svc_xprt *xprt);
1209
1210 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1211 struct net *net,
1212 struct sockaddr *sa, int salen,
1213 int flags)
1214 {
1215 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1216 }
1217
1218 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1219 {
1220 }
1221
1222 static struct svc_xprt_ops svc_tcp_bc_ops = {
1223 .xpo_create = svc_bc_tcp_create,
1224 .xpo_detach = svc_bc_tcp_sock_detach,
1225 .xpo_free = svc_bc_sock_free,
1226 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1227 };
1228
1229 static struct svc_xprt_class svc_tcp_bc_class = {
1230 .xcl_name = "tcp-bc",
1231 .xcl_owner = THIS_MODULE,
1232 .xcl_ops = &svc_tcp_bc_ops,
1233 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1234 };
1235
1236 static void svc_init_bc_xprt_sock(void)
1237 {
1238 svc_reg_xprt_class(&svc_tcp_bc_class);
1239 }
1240
1241 static void svc_cleanup_bc_xprt_sock(void)
1242 {
1243 svc_unreg_xprt_class(&svc_tcp_bc_class);
1244 }
1245 #else /* CONFIG_NFS_V4_1 */
1246 static void svc_init_bc_xprt_sock(void)
1247 {
1248 }
1249
1250 static void svc_cleanup_bc_xprt_sock(void)
1251 {
1252 }
1253 #endif /* CONFIG_NFS_V4_1 */
1254
1255 static struct svc_xprt_ops svc_tcp_ops = {
1256 .xpo_create = svc_tcp_create,
1257 .xpo_recvfrom = svc_tcp_recvfrom,
1258 .xpo_sendto = svc_tcp_sendto,
1259 .xpo_release_rqst = svc_release_skb,
1260 .xpo_detach = svc_tcp_sock_detach,
1261 .xpo_free = svc_sock_free,
1262 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1263 .xpo_has_wspace = svc_tcp_has_wspace,
1264 .xpo_accept = svc_tcp_accept,
1265 };
1266
1267 static struct svc_xprt_class svc_tcp_class = {
1268 .xcl_name = "tcp",
1269 .xcl_owner = THIS_MODULE,
1270 .xcl_ops = &svc_tcp_ops,
1271 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1272 };
1273
1274 void svc_init_xprt_sock(void)
1275 {
1276 svc_reg_xprt_class(&svc_tcp_class);
1277 svc_reg_xprt_class(&svc_udp_class);
1278 svc_init_bc_xprt_sock();
1279 }
1280
1281 void svc_cleanup_xprt_sock(void)
1282 {
1283 svc_unreg_xprt_class(&svc_tcp_class);
1284 svc_unreg_xprt_class(&svc_udp_class);
1285 svc_cleanup_bc_xprt_sock();
1286 }
1287
1288 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1289 {
1290 struct sock *sk = svsk->sk_sk;
1291
1292 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1293 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1294 if (sk->sk_state == TCP_LISTEN) {
1295 dprintk("setting up TCP socket for listening\n");
1296 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1297 sk->sk_data_ready = svc_tcp_listen_data_ready;
1298 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1299 } else {
1300 dprintk("setting up TCP socket for reading\n");
1301 sk->sk_state_change = svc_tcp_state_change;
1302 sk->sk_data_ready = svc_tcp_data_ready;
1303 sk->sk_write_space = svc_tcp_write_space;
1304
1305 svsk->sk_reclen = 0;
1306 svsk->sk_tcplen = 0;
1307
1308 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1309
1310 /* initialise setting must have enough space to
1311 * receive and respond to one request.
1312 * svc_tcp_recvfrom will re-adjust if necessary
1313 */
1314 svc_sock_setbufsize(svsk->sk_sock,
1315 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1316 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1317
1318 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1319 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1320 if (sk->sk_state != TCP_ESTABLISHED)
1321 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1322 }
1323 }
1324
1325 void svc_sock_update_bufs(struct svc_serv *serv)
1326 {
1327 /*
1328 * The number of server threads has changed. Update
1329 * rcvbuf and sndbuf accordingly on all sockets
1330 */
1331 struct svc_sock *svsk;
1332
1333 spin_lock_bh(&serv->sv_lock);
1334 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1335 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1336 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1337 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1338 spin_unlock_bh(&serv->sv_lock);
1339 }
1340 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1341
1342 /*
1343 * Initialize socket for RPC use and create svc_sock struct
1344 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1345 */
1346 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1347 struct socket *sock,
1348 int *errp, int flags)
1349 {
1350 struct svc_sock *svsk;
1351 struct sock *inet;
1352 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1353
1354 dprintk("svc: svc_setup_socket %p\n", sock);
1355 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1356 *errp = -ENOMEM;
1357 return NULL;
1358 }
1359
1360 inet = sock->sk;
1361
1362 /* Register socket with portmapper */
1363 if (*errp >= 0 && pmap_register)
1364 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1365 ntohs(inet_sk(inet)->inet_sport));
1366
1367 if (*errp < 0) {
1368 kfree(svsk);
1369 return NULL;
1370 }
1371
1372 inet->sk_user_data = svsk;
1373 svsk->sk_sock = sock;
1374 svsk->sk_sk = inet;
1375 svsk->sk_ostate = inet->sk_state_change;
1376 svsk->sk_odata = inet->sk_data_ready;
1377 svsk->sk_owspace = inet->sk_write_space;
1378
1379 /* Initialize the socket */
1380 if (sock->type == SOCK_DGRAM)
1381 svc_udp_init(svsk, serv);
1382 else
1383 svc_tcp_init(svsk, serv);
1384
1385 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1386 svsk, svsk->sk_sk);
1387
1388 return svsk;
1389 }
1390
1391 /**
1392 * svc_addsock - add a listener socket to an RPC service
1393 * @serv: pointer to RPC service to which to add a new listener
1394 * @fd: file descriptor of the new listener
1395 * @name_return: pointer to buffer to fill in with name of listener
1396 * @len: size of the buffer
1397 *
1398 * Fills in socket name and returns positive length of name if successful.
1399 * Name is terminated with '\n'. On error, returns a negative errno
1400 * value.
1401 */
1402 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1403 const size_t len)
1404 {
1405 int err = 0;
1406 struct socket *so = sockfd_lookup(fd, &err);
1407 struct svc_sock *svsk = NULL;
1408
1409 if (!so)
1410 return err;
1411 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1412 err = -EAFNOSUPPORT;
1413 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1414 so->sk->sk_protocol != IPPROTO_UDP)
1415 err = -EPROTONOSUPPORT;
1416 else if (so->state > SS_UNCONNECTED)
1417 err = -EISCONN;
1418 else {
1419 if (!try_module_get(THIS_MODULE))
1420 err = -ENOENT;
1421 else
1422 svsk = svc_setup_socket(serv, so, &err,
1423 SVC_SOCK_DEFAULTS);
1424 if (svsk) {
1425 struct sockaddr_storage addr;
1426 struct sockaddr *sin = (struct sockaddr *)&addr;
1427 int salen;
1428 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1429 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1430 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1431 spin_lock_bh(&serv->sv_lock);
1432 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1433 spin_unlock_bh(&serv->sv_lock);
1434 svc_xprt_received(&svsk->sk_xprt);
1435 err = 0;
1436 } else
1437 module_put(THIS_MODULE);
1438 }
1439 if (err) {
1440 sockfd_put(so);
1441 return err;
1442 }
1443 return svc_one_sock_name(svsk, name_return, len);
1444 }
1445 EXPORT_SYMBOL_GPL(svc_addsock);
1446
1447 /*
1448 * Create socket for RPC service.
1449 */
1450 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1451 int protocol,
1452 struct net *net,
1453 struct sockaddr *sin, int len,
1454 int flags)
1455 {
1456 struct svc_sock *svsk;
1457 struct socket *sock;
1458 int error;
1459 int type;
1460 struct sockaddr_storage addr;
1461 struct sockaddr *newsin = (struct sockaddr *)&addr;
1462 int newlen;
1463 int family;
1464 int val;
1465 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1466
1467 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1468 serv->sv_program->pg_name, protocol,
1469 __svc_print_addr(sin, buf, sizeof(buf)));
1470
1471 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1472 printk(KERN_WARNING "svc: only UDP and TCP "
1473 "sockets supported\n");
1474 return ERR_PTR(-EINVAL);
1475 }
1476
1477 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1478 switch (sin->sa_family) {
1479 case AF_INET6:
1480 family = PF_INET6;
1481 break;
1482 case AF_INET:
1483 family = PF_INET;
1484 break;
1485 default:
1486 return ERR_PTR(-EINVAL);
1487 }
1488
1489 error = __sock_create(net, family, type, protocol, &sock, 1);
1490 if (error < 0)
1491 return ERR_PTR(error);
1492
1493 svc_reclassify_socket(sock);
1494
1495 /*
1496 * If this is an PF_INET6 listener, we want to avoid
1497 * getting requests from IPv4 remotes. Those should
1498 * be shunted to a PF_INET listener via rpcbind.
1499 */
1500 val = 1;
1501 if (family == PF_INET6)
1502 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1503 (char *)&val, sizeof(val));
1504
1505 if (type == SOCK_STREAM)
1506 sock->sk->sk_reuse = 1; /* allow address reuse */
1507 error = kernel_bind(sock, sin, len);
1508 if (error < 0)
1509 goto bummer;
1510
1511 newlen = len;
1512 error = kernel_getsockname(sock, newsin, &newlen);
1513 if (error < 0)
1514 goto bummer;
1515
1516 if (protocol == IPPROTO_TCP) {
1517 if ((error = kernel_listen(sock, 64)) < 0)
1518 goto bummer;
1519 }
1520
1521 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1522 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1523 return (struct svc_xprt *)svsk;
1524 }
1525
1526 bummer:
1527 dprintk("svc: svc_create_socket error = %d\n", -error);
1528 sock_release(sock);
1529 return ERR_PTR(error);
1530 }
1531
1532 /*
1533 * Detach the svc_sock from the socket so that no
1534 * more callbacks occur.
1535 */
1536 static void svc_sock_detach(struct svc_xprt *xprt)
1537 {
1538 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1539 struct sock *sk = svsk->sk_sk;
1540 wait_queue_head_t *wq;
1541
1542 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1543
1544 /* put back the old socket callbacks */
1545 sk->sk_state_change = svsk->sk_ostate;
1546 sk->sk_data_ready = svsk->sk_odata;
1547 sk->sk_write_space = svsk->sk_owspace;
1548
1549 wq = sk_sleep(sk);
1550 if (wq && waitqueue_active(wq))
1551 wake_up_interruptible(wq);
1552 }
1553
1554 /*
1555 * Disconnect the socket, and reset the callbacks
1556 */
1557 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1558 {
1559 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1560
1561 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1562
1563 svc_sock_detach(xprt);
1564
1565 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1566 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1567 }
1568
1569 /*
1570 * Free the svc_sock's socket resources and the svc_sock itself.
1571 */
1572 static void svc_sock_free(struct svc_xprt *xprt)
1573 {
1574 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1575 dprintk("svc: svc_sock_free(%p)\n", svsk);
1576
1577 if (svsk->sk_sock->file)
1578 sockfd_put(svsk->sk_sock);
1579 else
1580 sock_release(svsk->sk_sock);
1581 kfree(svsk);
1582 }
1583
1584 #if defined(CONFIG_NFS_V4_1)
1585 /*
1586 * Create a back channel svc_xprt which shares the fore channel socket.
1587 */
1588 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1589 int protocol,
1590 struct net *net,
1591 struct sockaddr *sin, int len,
1592 int flags)
1593 {
1594 struct svc_sock *svsk;
1595 struct svc_xprt *xprt;
1596
1597 if (protocol != IPPROTO_TCP) {
1598 printk(KERN_WARNING "svc: only TCP sockets"
1599 " supported on shared back channel\n");
1600 return ERR_PTR(-EINVAL);
1601 }
1602
1603 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1604 if (!svsk)
1605 return ERR_PTR(-ENOMEM);
1606
1607 xprt = &svsk->sk_xprt;
1608 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1609
1610 serv->sv_bc_xprt = xprt;
1611
1612 return xprt;
1613 }
1614
1615 /*
1616 * Free a back channel svc_sock.
1617 */
1618 static void svc_bc_sock_free(struct svc_xprt *xprt)
1619 {
1620 if (xprt)
1621 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1622 }
1623 #endif /* CONFIG_NFS_V4_1 */
CRIS linux kernel tree