?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
NFS rename client back channel transport field
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sunrpc / svcsock.c
blobd265aa700bb3dc9fd036ceef3e3650f8e9b59533
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 else
337 len += onelen;
338 }
339 spin_unlock_bh(&serv->sv_lock);
340
341 if (closesk)
342 /* Should unregister with portmap, but you cannot
343 * unregister just one protocol...
344 */
345 svc_close_xprt(&closesk->sk_xprt);
346 else if (toclose)
347 return -ENOENT;
348 return len;
349 }
350 EXPORT_SYMBOL_GPL(svc_sock_names);
351
352 /*
353 * Check input queue length
354 */
355 static int svc_recv_available(struct svc_sock *svsk)
356 {
357 struct socket *sock = svsk->sk_sock;
358 int avail, err;
359
360 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
361
362 return (err >= 0)? avail : err;
363 }
364
365 /*
366 * Generic recvfrom routine.
367 */
368 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
369 int buflen)
370 {
371 struct svc_sock *svsk =
372 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
373 struct msghdr msg = {
374 .msg_flags = MSG_DONTWAIT,
375 };
376 int len;
377
378 rqstp->rq_xprt_hlen = 0;
379
380 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
381 msg.msg_flags);
382
383 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
384 svsk, iov[0].iov_base, iov[0].iov_len, len);
385 return len;
386 }
387
388 /*
389 * Set socket snd and rcv buffer lengths
390 */
391 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
392 unsigned int rcv)
393 {
394 #if 0
395 mm_segment_t oldfs;
396 oldfs = get_fs(); set_fs(KERNEL_DS);
397 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
398 (char*)&snd, sizeof(snd));
399 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
400 (char*)&rcv, sizeof(rcv));
401 #else
402 /* sock_setsockopt limits use to sysctl_?mem_max,
403 * which isn't acceptable. Until that is made conditional
404 * on not having CAP_SYS_RESOURCE or similar, we go direct...
405 * DaveM said I could!
406 */
407 lock_sock(sock->sk);
408 sock->sk->sk_sndbuf = snd * 2;
409 sock->sk->sk_rcvbuf = rcv * 2;
410 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
411 sock->sk->sk_write_space(sock->sk);
412 release_sock(sock->sk);
413 #endif
414 }
415 /*
416 * INET callback when data has been received on the socket.
417 */
418 static void svc_udp_data_ready(struct sock *sk, int count)
419 {
420 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
421
422 if (svsk) {
423 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
424 svsk, sk, count,
425 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
426 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
427 svc_xprt_enqueue(&svsk->sk_xprt);
428 }
429 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
430 wake_up_interruptible(sk_sleep(sk));
431 }
432
433 /*
434 * INET callback when space is newly available on the socket.
435 */
436 static void svc_write_space(struct sock *sk)
437 {
438 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
439
440 if (svsk) {
441 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
442 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
443 svc_xprt_enqueue(&svsk->sk_xprt);
444 }
445
446 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) {
447 dprintk("RPC svc_write_space: someone sleeping on %p\n",
448 svsk);
449 wake_up_interruptible(sk_sleep(sk));
450 }
451 }
452
453 static void svc_tcp_write_space(struct sock *sk)
454 {
455 struct socket *sock = sk->sk_socket;
456
457 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
458 clear_bit(SOCK_NOSPACE, &sock->flags);
459 svc_write_space(sk);
460 }
461
462 /*
463 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
464 */
465 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
466 struct cmsghdr *cmh)
467 {
468 struct in_pktinfo *pki = CMSG_DATA(cmh);
469 if (cmh->cmsg_type != IP_PKTINFO)
470 return 0;
471 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
472 return 1;
473 }
474
475 /*
476 * See net/ipv6/datagram.c : datagram_recv_ctl
477 */
478 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
479 struct cmsghdr *cmh)
480 {
481 struct in6_pktinfo *pki = CMSG_DATA(cmh);
482 if (cmh->cmsg_type != IPV6_PKTINFO)
483 return 0;
484 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
485 return 1;
486 }
487
488 /*
489 * Copy the UDP datagram's destination address to the rqstp structure.
490 * The 'destination' address in this case is the address to which the
491 * peer sent the datagram, i.e. our local address. For multihomed
492 * hosts, this can change from msg to msg. Note that only the IP
493 * address changes, the port number should remain the same.
494 */
495 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
496 struct cmsghdr *cmh)
497 {
498 switch (cmh->cmsg_level) {
499 case SOL_IP:
500 return svc_udp_get_dest_address4(rqstp, cmh);
501 case SOL_IPV6:
502 return svc_udp_get_dest_address6(rqstp, cmh);
503 }
504
505 return 0;
506 }
507
508 /*
509 * Receive a datagram from a UDP socket.
510 */
511 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
512 {
513 struct svc_sock *svsk =
514 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
515 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
516 struct sk_buff *skb;
517 union {
518 struct cmsghdr hdr;
519 long all[SVC_PKTINFO_SPACE / sizeof(long)];
520 } buffer;
521 struct cmsghdr *cmh = &buffer.hdr;
522 struct msghdr msg = {
523 .msg_name = svc_addr(rqstp),
524 .msg_control = cmh,
525 .msg_controllen = sizeof(buffer),
526 .msg_flags = MSG_DONTWAIT,
527 };
528 size_t len;
529 int err;
530
531 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
532 /* udp sockets need large rcvbuf as all pending
533 * requests are still in that buffer. sndbuf must
534 * also be large enough that there is enough space
535 * for one reply per thread. We count all threads
536 * rather than threads in a particular pool, which
537 * provides an upper bound on the number of threads
538 * which will access the socket.
539 */
540 svc_sock_setbufsize(svsk->sk_sock,
541 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
542 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
543
544 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
545 skb = NULL;
546 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
547 0, 0, MSG_PEEK | MSG_DONTWAIT);
548 if (err >= 0)
549 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
550
551 if (skb == NULL) {
552 if (err != -EAGAIN) {
553 /* possibly an icmp error */
554 dprintk("svc: recvfrom returned error %d\n", -err);
555 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
556 }
557 return -EAGAIN;
558 }
559 len = svc_addr_len(svc_addr(rqstp));
560 if (len == 0)
561 return -EAFNOSUPPORT;
562 rqstp->rq_addrlen = len;
563 if (skb->tstamp.tv64 == 0) {
564 skb->tstamp = ktime_get_real();
565 /* Don't enable netstamp, sunrpc doesn't
566 need that much accuracy */
567 }
568 svsk->sk_sk->sk_stamp = skb->tstamp;
569 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
570
571 len = skb->len - sizeof(struct udphdr);
572 rqstp->rq_arg.len = len;
573
574 rqstp->rq_prot = IPPROTO_UDP;
575
576 if (!svc_udp_get_dest_address(rqstp, cmh)) {
577 if (net_ratelimit())
578 printk(KERN_WARNING
579 "svc: received unknown control message %d/%d; "
580 "dropping RPC reply datagram\n",
581 cmh->cmsg_level, cmh->cmsg_type);
582 skb_free_datagram_locked(svsk->sk_sk, skb);
583 return 0;
584 }
585
586 if (skb_is_nonlinear(skb)) {
587 /* we have to copy */
588 local_bh_disable();
589 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
590 local_bh_enable();
591 /* checksum error */
592 skb_free_datagram_locked(svsk->sk_sk, skb);
593 return 0;
594 }
595 local_bh_enable();
596 skb_free_datagram_locked(svsk->sk_sk, skb);
597 } else {
598 /* we can use it in-place */
599 rqstp->rq_arg.head[0].iov_base = skb->data +
600 sizeof(struct udphdr);
601 rqstp->rq_arg.head[0].iov_len = len;
602 if (skb_checksum_complete(skb)) {
603 skb_free_datagram_locked(svsk->sk_sk, skb);
604 return 0;
605 }
606 rqstp->rq_xprt_ctxt = skb;
607 }
608
609 rqstp->rq_arg.page_base = 0;
610 if (len <= rqstp->rq_arg.head[0].iov_len) {
611 rqstp->rq_arg.head[0].iov_len = len;
612 rqstp->rq_arg.page_len = 0;
613 rqstp->rq_respages = rqstp->rq_pages+1;
614 } else {
615 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
616 rqstp->rq_respages = rqstp->rq_pages + 1 +
617 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
618 }
619
620 if (serv->sv_stats)
621 serv->sv_stats->netudpcnt++;
622
623 return len;
624 }
625
626 static int
627 svc_udp_sendto(struct svc_rqst *rqstp)
628 {
629 int error;
630
631 error = svc_sendto(rqstp, &rqstp->rq_res);
632 if (error == -ECONNREFUSED)
633 /* ICMP error on earlier request. */
634 error = svc_sendto(rqstp, &rqstp->rq_res);
635
636 return error;
637 }
638
639 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
640 {
641 }
642
643 static int svc_udp_has_wspace(struct svc_xprt *xprt)
644 {
645 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
646 struct svc_serv *serv = xprt->xpt_server;
647 unsigned long required;
648
649 /*
650 * Set the SOCK_NOSPACE flag before checking the available
651 * sock space.
652 */
653 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
654 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
655 if (required*2 > sock_wspace(svsk->sk_sk))
656 return 0;
657 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
658 return 1;
659 }
660
661 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
662 {
663 BUG();
664 return NULL;
665 }
666
667 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
668 struct net *net,
669 struct sockaddr *sa, int salen,
670 int flags)
671 {
672 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
673 }
674
675 static struct svc_xprt_ops svc_udp_ops = {
676 .xpo_create = svc_udp_create,
677 .xpo_recvfrom = svc_udp_recvfrom,
678 .xpo_sendto = svc_udp_sendto,
679 .xpo_release_rqst = svc_release_skb,
680 .xpo_detach = svc_sock_detach,
681 .xpo_free = svc_sock_free,
682 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
683 .xpo_has_wspace = svc_udp_has_wspace,
684 .xpo_accept = svc_udp_accept,
685 };
686
687 static struct svc_xprt_class svc_udp_class = {
688 .xcl_name = "udp",
689 .xcl_owner = THIS_MODULE,
690 .xcl_ops = &svc_udp_ops,
691 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
692 };
693
694 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
695 {
696 int err, level, optname, one = 1;
697
698 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
699 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
700 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
701 svsk->sk_sk->sk_write_space = svc_write_space;
702
703 /* initialise setting must have enough space to
704 * receive and respond to one request.
705 * svc_udp_recvfrom will re-adjust if necessary
706 */
707 svc_sock_setbufsize(svsk->sk_sock,
708 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
709 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
710
711 /* data might have come in before data_ready set up */
712 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
713 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
714
715 /* make sure we get destination address info */
716 switch (svsk->sk_sk->sk_family) {
717 case AF_INET:
718 level = SOL_IP;
719 optname = IP_PKTINFO;
720 break;
721 case AF_INET6:
722 level = SOL_IPV6;
723 optname = IPV6_RECVPKTINFO;
724 break;
725 default:
726 BUG();
727 }
728 err = kernel_setsockopt(svsk->sk_sock, level, optname,
729 (char *)&one, sizeof(one));
730 dprintk("svc: kernel_setsockopt returned %d\n", err);
731 }
732
733 /*
734 * A data_ready event on a listening socket means there's a connection
735 * pending. Do not use state_change as a substitute for it.
736 */
737 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
738 {
739 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
740
741 dprintk("svc: socket %p TCP (listen) state change %d\n",
742 sk, sk->sk_state);
743
744 /*
745 * This callback may called twice when a new connection
746 * is established as a child socket inherits everything
747 * from a parent LISTEN socket.
748 * 1) data_ready method of the parent socket will be called
749 * when one of child sockets become ESTABLISHED.
750 * 2) data_ready method of the child socket may be called
751 * when it receives data before the socket is accepted.
752 * In case of 2, we should ignore it silently.
753 */
754 if (sk->sk_state == TCP_LISTEN) {
755 if (svsk) {
756 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
757 svc_xprt_enqueue(&svsk->sk_xprt);
758 } else
759 printk("svc: socket %p: no user data\n", sk);
760 }
761
762 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
763 wake_up_interruptible_all(sk_sleep(sk));
764 }
765
766 /*
767 * A state change on a connected socket means it's dying or dead.
768 */
769 static void svc_tcp_state_change(struct sock *sk)
770 {
771 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
772
773 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
774 sk, sk->sk_state, sk->sk_user_data);
775
776 if (!svsk)
777 printk("svc: socket %p: no user data\n", sk);
778 else {
779 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
780 svc_xprt_enqueue(&svsk->sk_xprt);
781 }
782 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
783 wake_up_interruptible_all(sk_sleep(sk));
784 }
785
786 static void svc_tcp_data_ready(struct sock *sk, int count)
787 {
788 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
789
790 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
791 sk, sk->sk_user_data);
792 if (svsk) {
793 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
794 svc_xprt_enqueue(&svsk->sk_xprt);
795 }
796 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
797 wake_up_interruptible(sk_sleep(sk));
798 }
799
800 /*
801 * Accept a TCP connection
802 */
803 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
804 {
805 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
806 struct sockaddr_storage addr;
807 struct sockaddr *sin = (struct sockaddr *) &addr;
808 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
809 struct socket *sock = svsk->sk_sock;
810 struct socket *newsock;
811 struct svc_sock *newsvsk;
812 int err, slen;
813 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
814
815 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
816 if (!sock)
817 return NULL;
818
819 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
820 err = kernel_accept(sock, &newsock, O_NONBLOCK);
821 if (err < 0) {
822 if (err == -ENOMEM)
823 printk(KERN_WARNING "%s: no more sockets!\n",
824 serv->sv_name);
825 else if (err != -EAGAIN && net_ratelimit())
826 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
827 serv->sv_name, -err);
828 return NULL;
829 }
830 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
831
832 err = kernel_getpeername(newsock, sin, &slen);
833 if (err < 0) {
834 if (net_ratelimit())
835 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
836 serv->sv_name, -err);
837 goto failed; /* aborted connection or whatever */
838 }
839
840 /* Ideally, we would want to reject connections from unauthorized
841 * hosts here, but when we get encryption, the IP of the host won't
842 * tell us anything. For now just warn about unpriv connections.
843 */
844 if (!svc_port_is_privileged(sin)) {
845 dprintk(KERN_WARNING
846 "%s: connect from unprivileged port: %s\n",
847 serv->sv_name,
848 __svc_print_addr(sin, buf, sizeof(buf)));
849 }
850 dprintk("%s: connect from %s\n", serv->sv_name,
851 __svc_print_addr(sin, buf, sizeof(buf)));
852
853 /* make sure that a write doesn't block forever when
854 * low on memory
855 */
856 newsock->sk->sk_sndtimeo = HZ*30;
857
858 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
859 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
860 goto failed;
861 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
862 err = kernel_getsockname(newsock, sin, &slen);
863 if (unlikely(err < 0)) {
864 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
865 slen = offsetof(struct sockaddr, sa_data);
866 }
867 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
868
869 if (serv->sv_stats)
870 serv->sv_stats->nettcpconn++;
871
872 return &newsvsk->sk_xprt;
873
874 failed:
875 sock_release(newsock);
876 return NULL;
877 }
878
879 /*
880 * Receive data.
881 * If we haven't gotten the record length yet, get the next four bytes.
882 * Otherwise try to gobble up as much as possible up to the complete
883 * record length.
884 */
885 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
886 {
887 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
888 int len;
889
890 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
891 /* sndbuf needs to have room for one request
892 * per thread, otherwise we can stall even when the
893 * network isn't a bottleneck.
894 *
895 * We count all threads rather than threads in a
896 * particular pool, which provides an upper bound
897 * on the number of threads which will access the socket.
898 *
899 * rcvbuf just needs to be able to hold a few requests.
900 * Normally they will be removed from the queue
901 * as soon a a complete request arrives.
902 */
903 svc_sock_setbufsize(svsk->sk_sock,
904 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
905 3 * serv->sv_max_mesg);
906
907 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
908
909 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
910 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
911 struct kvec iov;
912
913 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
914 iov.iov_len = want;
915 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
916 goto error;
917 svsk->sk_tcplen += len;
918
919 if (len < want) {
920 dprintk("svc: short recvfrom while reading record "
921 "length (%d of %d)\n", len, want);
922 goto err_again; /* record header not complete */
923 }
924
925 svsk->sk_reclen = ntohl(svsk->sk_reclen);
926 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
927 /* FIXME: technically, a record can be fragmented,
928 * and non-terminal fragments will not have the top
929 * bit set in the fragment length header.
930 * But apparently no known nfs clients send fragmented
931 * records. */
932 if (net_ratelimit())
933 printk(KERN_NOTICE "RPC: multiple fragments "
934 "per record not supported\n");
935 goto err_delete;
936 }
937
938 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
939 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
940 if (svsk->sk_reclen > serv->sv_max_mesg) {
941 if (net_ratelimit())
942 printk(KERN_NOTICE "RPC: "
943 "fragment too large: 0x%08lx\n",
944 (unsigned long)svsk->sk_reclen);
945 goto err_delete;
946 }
947 }
948
949 /* Check whether enough data is available */
950 len = svc_recv_available(svsk);
951 if (len < 0)
952 goto error;
953
954 if (len < svsk->sk_reclen) {
955 dprintk("svc: incomplete TCP record (%d of %d)\n",
956 len, svsk->sk_reclen);
957 goto err_again; /* record not complete */
958 }
959 len = svsk->sk_reclen;
960 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
961
962 return len;
963 error:
964 if (len == -EAGAIN)
965 dprintk("RPC: TCP recv_record got EAGAIN\n");
966 return len;
967 err_delete:
968 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
969 err_again:
970 return -EAGAIN;
971 }
972
973 static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
974 struct rpc_rqst **reqpp, struct kvec *vec)
975 {
976 struct rpc_rqst *req = NULL;
977 u32 *p;
978 u32 xid;
979 u32 calldir;
980 int len;
981
982 len = svc_recvfrom(rqstp, vec, 1, 8);
983 if (len < 0)
984 goto error;
985
986 p = (u32 *)rqstp->rq_arg.head[0].iov_base;
987 xid = *p++;
988 calldir = *p;
989
990 if (calldir == 0) {
991 /* REQUEST is the most common case */
992 vec[0] = rqstp->rq_arg.head[0];
993 } else {
994 /* REPLY */
995 if (svsk->sk_bc_xprt)
996 req = xprt_lookup_rqst(svsk->sk_bc_xprt, xid);
997
998 if (!req) {
999 printk(KERN_NOTICE
1000 "%s: Got unrecognized reply: "
1001 "calldir 0x%x sk_bc_xprt %p xid %08x\n",
1002 __func__, ntohl(calldir),
1003 svsk->sk_bc_xprt, xid);
1004 vec[0] = rqstp->rq_arg.head[0];
1005 goto out;
1006 }
1007
1008 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1009 sizeof(struct xdr_buf));
1010 /* copy the xid and call direction */
1011 memcpy(req->rq_private_buf.head[0].iov_base,
1012 rqstp->rq_arg.head[0].iov_base, 8);
1013 vec[0] = req->rq_private_buf.head[0];
1014 }
1015 out:
1016 vec[0].iov_base += 8;
1017 vec[0].iov_len -= 8;
1018 len = svsk->sk_reclen - 8;
1019 error:
1020 *reqpp = req;
1021 return len;
1022 }
1023
1024 /*
1025 * Receive data from a TCP socket.
1026 */
1027 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1028 {
1029 struct svc_sock *svsk =
1030 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1031 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1032 int len;
1033 struct kvec *vec;
1034 int pnum, vlen;
1035 struct rpc_rqst *req = NULL;
1036
1037 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1038 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1039 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1040 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1041
1042 len = svc_tcp_recv_record(svsk, rqstp);
1043 if (len < 0)
1044 goto error;
1045
1046 vec = rqstp->rq_vec;
1047 vec[0] = rqstp->rq_arg.head[0];
1048 vlen = PAGE_SIZE;
1049
1050 /*
1051 * We have enough data for the whole tcp record. Let's try and read the
1052 * first 8 bytes to get the xid and the call direction. We can use this
1053 * to figure out if this is a call or a reply to a callback. If
1054 * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
1055 * In that case, don't bother with the calldir and just read the data.
1056 * It will be rejected in svc_process.
1057 */
1058 if (len >= 8) {
1059 len = svc_process_calldir(svsk, rqstp, &req, vec);
1060 if (len < 0)
1061 goto err_again;
1062 vlen -= 8;
1063 }
1064
1065 pnum = 1;
1066 while (vlen < len) {
1067 vec[pnum].iov_base = (req) ?
1068 page_address(req->rq_private_buf.pages[pnum - 1]) :
1069 page_address(rqstp->rq_pages[pnum]);
1070 vec[pnum].iov_len = PAGE_SIZE;
1071 pnum++;
1072 vlen += PAGE_SIZE;
1073 }
1074 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1075
1076 /* Now receive data */
1077 len = svc_recvfrom(rqstp, vec, pnum, len);
1078 if (len < 0)
1079 goto err_again;
1080
1081 /*
1082 * Account for the 8 bytes we read earlier
1083 */
1084 len += 8;
1085
1086 if (req) {
1087 xprt_complete_rqst(req->rq_task, len);
1088 len = 0;
1089 goto out;
1090 }
1091 dprintk("svc: TCP complete record (%d bytes)\n", len);
1092 rqstp->rq_arg.len = len;
1093 rqstp->rq_arg.page_base = 0;
1094 if (len <= rqstp->rq_arg.head[0].iov_len) {
1095 rqstp->rq_arg.head[0].iov_len = len;
1096 rqstp->rq_arg.page_len = 0;
1097 } else {
1098 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1099 }
1100
1101 rqstp->rq_xprt_ctxt = NULL;
1102 rqstp->rq_prot = IPPROTO_TCP;
1103
1104 out:
1105 /* Reset TCP read info */
1106 svsk->sk_reclen = 0;
1107 svsk->sk_tcplen = 0;
1108
1109 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1110 if (serv->sv_stats)
1111 serv->sv_stats->nettcpcnt++;
1112
1113 return len;
1114
1115 err_again:
1116 if (len == -EAGAIN) {
1117 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1118 return len;
1119 }
1120 error:
1121 if (len != -EAGAIN) {
1122 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1123 svsk->sk_xprt.xpt_server->sv_name, -len);
1124 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1125 }
1126 return -EAGAIN;
1127 }
1128
1129 /*
1130 * Send out data on TCP socket.
1131 */
1132 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1133 {
1134 struct xdr_buf *xbufp = &rqstp->rq_res;
1135 int sent;
1136 __be32 reclen;
1137
1138 /* Set up the first element of the reply kvec.
1139 * Any other kvecs that may be in use have been taken
1140 * care of by the server implementation itself.
1141 */
1142 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1143 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1144
1145 sent = svc_sendto(rqstp, &rqstp->rq_res);
1146 if (sent != xbufp->len) {
1147 printk(KERN_NOTICE
1148 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1149 "- shutting down socket\n",
1150 rqstp->rq_xprt->xpt_server->sv_name,
1151 (sent<0)?"got error":"sent only",
1152 sent, xbufp->len);
1153 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1154 svc_xprt_enqueue(rqstp->rq_xprt);
1155 sent = -EAGAIN;
1156 }
1157 return sent;
1158 }
1159
1160 /*
1161 * Setup response header. TCP has a 4B record length field.
1162 */
1163 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1164 {
1165 struct kvec *resv = &rqstp->rq_res.head[0];
1166
1167 /* tcp needs a space for the record length... */
1168 svc_putnl(resv, 0);
1169 }
1170
1171 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1172 {
1173 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1174 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1175 int required;
1176
1177 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1178 return 1;
1179 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1180 if (sk_stream_wspace(svsk->sk_sk) >= required)
1181 return 1;
1182 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1183 return 0;
1184 }
1185
1186 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1187 struct net *net,
1188 struct sockaddr *sa, int salen,
1189 int flags)
1190 {
1191 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1192 }
1193
1194 #if defined(CONFIG_NFS_V4_1)
1195 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1196 struct net *, struct sockaddr *,
1197 int, int);
1198 static void svc_bc_sock_free(struct svc_xprt *xprt);
1199
1200 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1201 struct net *net,
1202 struct sockaddr *sa, int salen,
1203 int flags)
1204 {
1205 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1206 }
1207
1208 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1209 {
1210 }
1211
1212 static struct svc_xprt_ops svc_tcp_bc_ops = {
1213 .xpo_create = svc_bc_tcp_create,
1214 .xpo_detach = svc_bc_tcp_sock_detach,
1215 .xpo_free = svc_bc_sock_free,
1216 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1217 };
1218
1219 static struct svc_xprt_class svc_tcp_bc_class = {
1220 .xcl_name = "tcp-bc",
1221 .xcl_owner = THIS_MODULE,
1222 .xcl_ops = &svc_tcp_bc_ops,
1223 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1224 };
1225
1226 static void svc_init_bc_xprt_sock(void)
1227 {
1228 svc_reg_xprt_class(&svc_tcp_bc_class);
1229 }
1230
1231 static void svc_cleanup_bc_xprt_sock(void)
1232 {
1233 svc_unreg_xprt_class(&svc_tcp_bc_class);
1234 }
1235 #else /* CONFIG_NFS_V4_1 */
1236 static void svc_init_bc_xprt_sock(void)
1237 {
1238 }
1239
1240 static void svc_cleanup_bc_xprt_sock(void)
1241 {
1242 }
1243 #endif /* CONFIG_NFS_V4_1 */
1244
1245 static struct svc_xprt_ops svc_tcp_ops = {
1246 .xpo_create = svc_tcp_create,
1247 .xpo_recvfrom = svc_tcp_recvfrom,
1248 .xpo_sendto = svc_tcp_sendto,
1249 .xpo_release_rqst = svc_release_skb,
1250 .xpo_detach = svc_tcp_sock_detach,
1251 .xpo_free = svc_sock_free,
1252 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1253 .xpo_has_wspace = svc_tcp_has_wspace,
1254 .xpo_accept = svc_tcp_accept,
1255 };
1256
1257 static struct svc_xprt_class svc_tcp_class = {
1258 .xcl_name = "tcp",
1259 .xcl_owner = THIS_MODULE,
1260 .xcl_ops = &svc_tcp_ops,
1261 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1262 };
1263
1264 void svc_init_xprt_sock(void)
1265 {
1266 svc_reg_xprt_class(&svc_tcp_class);
1267 svc_reg_xprt_class(&svc_udp_class);
1268 svc_init_bc_xprt_sock();
1269 }
1270
1271 void svc_cleanup_xprt_sock(void)
1272 {
1273 svc_unreg_xprt_class(&svc_tcp_class);
1274 svc_unreg_xprt_class(&svc_udp_class);
1275 svc_cleanup_bc_xprt_sock();
1276 }
1277
1278 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1279 {
1280 struct sock *sk = svsk->sk_sk;
1281
1282 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1283 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1284 if (sk->sk_state == TCP_LISTEN) {
1285 dprintk("setting up TCP socket for listening\n");
1286 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1287 sk->sk_data_ready = svc_tcp_listen_data_ready;
1288 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1289 } else {
1290 dprintk("setting up TCP socket for reading\n");
1291 sk->sk_state_change = svc_tcp_state_change;
1292 sk->sk_data_ready = svc_tcp_data_ready;
1293 sk->sk_write_space = svc_tcp_write_space;
1294
1295 svsk->sk_reclen = 0;
1296 svsk->sk_tcplen = 0;
1297
1298 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1299
1300 /* initialise setting must have enough space to
1301 * receive and respond to one request.
1302 * svc_tcp_recvfrom will re-adjust if necessary
1303 */
1304 svc_sock_setbufsize(svsk->sk_sock,
1305 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1306 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1307
1308 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1309 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1310 if (sk->sk_state != TCP_ESTABLISHED)
1311 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1312 }
1313 }
1314
1315 void svc_sock_update_bufs(struct svc_serv *serv)
1316 {
1317 /*
1318 * The number of server threads has changed. Update
1319 * rcvbuf and sndbuf accordingly on all sockets
1320 */
1321 struct svc_sock *svsk;
1322
1323 spin_lock_bh(&serv->sv_lock);
1324 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1325 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1326 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1327 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1328 spin_unlock_bh(&serv->sv_lock);
1329 }
1330 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1331
1332 /*
1333 * Initialize socket for RPC use and create svc_sock struct
1334 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1335 */
1336 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1337 struct socket *sock,
1338 int *errp, int flags)
1339 {
1340 struct svc_sock *svsk;
1341 struct sock *inet;
1342 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1343
1344 dprintk("svc: svc_setup_socket %p\n", sock);
1345 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1346 *errp = -ENOMEM;
1347 return NULL;
1348 }
1349
1350 inet = sock->sk;
1351
1352 /* Register socket with portmapper */
1353 if (*errp >= 0 && pmap_register)
1354 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1355 ntohs(inet_sk(inet)->inet_sport));
1356
1357 if (*errp < 0) {
1358 kfree(svsk);
1359 return NULL;
1360 }
1361
1362 inet->sk_user_data = svsk;
1363 svsk->sk_sock = sock;
1364 svsk->sk_sk = inet;
1365 svsk->sk_ostate = inet->sk_state_change;
1366 svsk->sk_odata = inet->sk_data_ready;
1367 svsk->sk_owspace = inet->sk_write_space;
1368
1369 /* Initialize the socket */
1370 if (sock->type == SOCK_DGRAM)
1371 svc_udp_init(svsk, serv);
1372 else
1373 svc_tcp_init(svsk, serv);
1374
1375 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1376 svsk, svsk->sk_sk);
1377
1378 return svsk;
1379 }
1380
1381 /**
1382 * svc_addsock - add a listener socket to an RPC service
1383 * @serv: pointer to RPC service to which to add a new listener
1384 * @fd: file descriptor of the new listener
1385 * @name_return: pointer to buffer to fill in with name of listener
1386 * @len: size of the buffer
1387 *
1388 * Fills in socket name and returns positive length of name if successful.
1389 * Name is terminated with '\n'. On error, returns a negative errno
1390 * value.
1391 */
1392 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1393 const size_t len)
1394 {
1395 int err = 0;
1396 struct socket *so = sockfd_lookup(fd, &err);
1397 struct svc_sock *svsk = NULL;
1398
1399 if (!so)
1400 return err;
1401 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1402 err = -EAFNOSUPPORT;
1403 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1404 so->sk->sk_protocol != IPPROTO_UDP)
1405 err = -EPROTONOSUPPORT;
1406 else if (so->state > SS_UNCONNECTED)
1407 err = -EISCONN;
1408 else {
1409 if (!try_module_get(THIS_MODULE))
1410 err = -ENOENT;
1411 else
1412 svsk = svc_setup_socket(serv, so, &err,
1413 SVC_SOCK_DEFAULTS);
1414 if (svsk) {
1415 struct sockaddr_storage addr;
1416 struct sockaddr *sin = (struct sockaddr *)&addr;
1417 int salen;
1418 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1419 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1420 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1421 spin_lock_bh(&serv->sv_lock);
1422 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1423 spin_unlock_bh(&serv->sv_lock);
1424 svc_xprt_received(&svsk->sk_xprt);
1425 err = 0;
1426 } else
1427 module_put(THIS_MODULE);
1428 }
1429 if (err) {
1430 sockfd_put(so);
1431 return err;
1432 }
1433 return svc_one_sock_name(svsk, name_return, len);
1434 }
1435 EXPORT_SYMBOL_GPL(svc_addsock);
1436
1437 /*
1438 * Create socket for RPC service.
1439 */
1440 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1441 int protocol,
1442 struct net *net,
1443 struct sockaddr *sin, int len,
1444 int flags)
1445 {
1446 struct svc_sock *svsk;
1447 struct socket *sock;
1448 int error;
1449 int type;
1450 struct sockaddr_storage addr;
1451 struct sockaddr *newsin = (struct sockaddr *)&addr;
1452 int newlen;
1453 int family;
1454 int val;
1455 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1456
1457 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1458 serv->sv_program->pg_name, protocol,
1459 __svc_print_addr(sin, buf, sizeof(buf)));
1460
1461 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1462 printk(KERN_WARNING "svc: only UDP and TCP "
1463 "sockets supported\n");
1464 return ERR_PTR(-EINVAL);
1465 }
1466
1467 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1468 switch (sin->sa_family) {
1469 case AF_INET6:
1470 family = PF_INET6;
1471 break;
1472 case AF_INET:
1473 family = PF_INET;
1474 break;
1475 default:
1476 return ERR_PTR(-EINVAL);
1477 }
1478
1479 error = __sock_create(net, family, type, protocol, &sock, 1);
1480 if (error < 0)
1481 return ERR_PTR(error);
1482
1483 svc_reclassify_socket(sock);
1484
1485 /*
1486 * If this is an PF_INET6 listener, we want to avoid
1487 * getting requests from IPv4 remotes. Those should
1488 * be shunted to a PF_INET listener via rpcbind.
1489 */
1490 val = 1;
1491 if (family == PF_INET6)
1492 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1493 (char *)&val, sizeof(val));
1494
1495 if (type == SOCK_STREAM)
1496 sock->sk->sk_reuse = 1; /* allow address reuse */
1497 error = kernel_bind(sock, sin, len);
1498 if (error < 0)
1499 goto bummer;
1500
1501 newlen = len;
1502 error = kernel_getsockname(sock, newsin, &newlen);
1503 if (error < 0)
1504 goto bummer;
1505
1506 if (protocol == IPPROTO_TCP) {
1507 if ((error = kernel_listen(sock, 64)) < 0)
1508 goto bummer;
1509 }
1510
1511 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1512 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1513 return (struct svc_xprt *)svsk;
1514 }
1515
1516 bummer:
1517 dprintk("svc: svc_create_socket error = %d\n", -error);
1518 sock_release(sock);
1519 return ERR_PTR(error);
1520 }
1521
1522 /*
1523 * Detach the svc_sock from the socket so that no
1524 * more callbacks occur.
1525 */
1526 static void svc_sock_detach(struct svc_xprt *xprt)
1527 {
1528 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1529 struct sock *sk = svsk->sk_sk;
1530
1531 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1532
1533 /* put back the old socket callbacks */
1534 sk->sk_state_change = svsk->sk_ostate;
1535 sk->sk_data_ready = svsk->sk_odata;
1536 sk->sk_write_space = svsk->sk_owspace;
1537
1538 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
1539 wake_up_interruptible(sk_sleep(sk));
1540 }
1541
1542 /*
1543 * Disconnect the socket, and reset the callbacks
1544 */
1545 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1546 {
1547 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1548
1549 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1550
1551 svc_sock_detach(xprt);
1552
1553 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1554 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1555 }
1556
1557 /*
1558 * Free the svc_sock's socket resources and the svc_sock itself.
1559 */
1560 static void svc_sock_free(struct svc_xprt *xprt)
1561 {
1562 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1563 dprintk("svc: svc_sock_free(%p)\n", svsk);
1564
1565 if (svsk->sk_sock->file)
1566 sockfd_put(svsk->sk_sock);
1567 else
1568 sock_release(svsk->sk_sock);
1569 kfree(svsk);
1570 }
1571
1572 #if defined(CONFIG_NFS_V4_1)
1573 /*
1574 * Create a back channel svc_xprt which shares the fore channel socket.
1575 */
1576 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1577 int protocol,
1578 struct net *net,
1579 struct sockaddr *sin, int len,
1580 int flags)
1581 {
1582 struct svc_sock *svsk;
1583 struct svc_xprt *xprt;
1584
1585 if (protocol != IPPROTO_TCP) {
1586 printk(KERN_WARNING "svc: only TCP sockets"
1587 " supported on shared back channel\n");
1588 return ERR_PTR(-EINVAL);
1589 }
1590
1591 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1592 if (!svsk)
1593 return ERR_PTR(-ENOMEM);
1594
1595 xprt = &svsk->sk_xprt;
1596 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1597
1598 serv->sv_bc_xprt = xprt;
1599
1600 return xprt;
1601 }
1602
1603 /*
1604 * Free a back channel svc_sock.
1605 */
1606 static void svc_bc_sock_free(struct svc_xprt *xprt)
1607 {
1608 if (xprt) {
1609 kfree(xprt->xpt_bc_sid);
1610 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1611 }
1612 }
1613 #endif /* CONFIG_NFS_V4_1 */
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree