?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
xen mmu: fix a race window causing leave_mm BUG()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sunrpc / svcsock.c
blobd802e941d365e9b53bf7c6be04bd1555bc0d1144
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
424 if (svsk) {
425 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
426 svsk, sk, count,
427 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
428 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
429 svc_xprt_enqueue(&svsk->sk_xprt);
430 }
431 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
432 wake_up_interruptible(sk_sleep(sk));
433 }
434
435 /*
436 * INET callback when space is newly available on the socket.
437 */
438 static void svc_write_space(struct sock *sk)
439 {
440 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
441
442 if (svsk) {
443 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
444 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
445 svc_xprt_enqueue(&svsk->sk_xprt);
446 }
447
448 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk))) {
449 dprintk("RPC svc_write_space: someone sleeping on %p\n",
450 svsk);
451 wake_up_interruptible(sk_sleep(sk));
452 }
453 }
454
455 static void svc_tcp_write_space(struct sock *sk)
456 {
457 struct socket *sock = sk->sk_socket;
458
459 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
460 clear_bit(SOCK_NOSPACE, &sock->flags);
461 svc_write_space(sk);
462 }
463
464 /*
465 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
466 */
467 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
468 struct cmsghdr *cmh)
469 {
470 struct in_pktinfo *pki = CMSG_DATA(cmh);
471 if (cmh->cmsg_type != IP_PKTINFO)
472 return 0;
473 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
474 return 1;
475 }
476
477 /*
478 * See net/ipv6/datagram.c : datagram_recv_ctl
479 */
480 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
481 struct cmsghdr *cmh)
482 {
483 struct in6_pktinfo *pki = CMSG_DATA(cmh);
484 if (cmh->cmsg_type != IPV6_PKTINFO)
485 return 0;
486 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
487 return 1;
488 }
489
490 /*
491 * Copy the UDP datagram's destination address to the rqstp structure.
492 * The 'destination' address in this case is the address to which the
493 * peer sent the datagram, i.e. our local address. For multihomed
494 * hosts, this can change from msg to msg. Note that only the IP
495 * address changes, the port number should remain the same.
496 */
497 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
498 struct cmsghdr *cmh)
499 {
500 switch (cmh->cmsg_level) {
501 case SOL_IP:
502 return svc_udp_get_dest_address4(rqstp, cmh);
503 case SOL_IPV6:
504 return svc_udp_get_dest_address6(rqstp, cmh);
505 }
506
507 return 0;
508 }
509
510 /*
511 * Receive a datagram from a UDP socket.
512 */
513 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
514 {
515 struct svc_sock *svsk =
516 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
517 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
518 struct sk_buff *skb;
519 union {
520 struct cmsghdr hdr;
521 long all[SVC_PKTINFO_SPACE / sizeof(long)];
522 } buffer;
523 struct cmsghdr *cmh = &buffer.hdr;
524 struct msghdr msg = {
525 .msg_name = svc_addr(rqstp),
526 .msg_control = cmh,
527 .msg_controllen = sizeof(buffer),
528 .msg_flags = MSG_DONTWAIT,
529 };
530 size_t len;
531 int err;
532
533 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
534 /* udp sockets need large rcvbuf as all pending
535 * requests are still in that buffer. sndbuf must
536 * also be large enough that there is enough space
537 * for one reply per thread. We count all threads
538 * rather than threads in a particular pool, which
539 * provides an upper bound on the number of threads
540 * which will access the socket.
541 */
542 svc_sock_setbufsize(svsk->sk_sock,
543 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
544 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
545
546 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
547 skb = NULL;
548 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
549 0, 0, MSG_PEEK | MSG_DONTWAIT);
550 if (err >= 0)
551 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
552
553 if (skb == NULL) {
554 if (err != -EAGAIN) {
555 /* possibly an icmp error */
556 dprintk("svc: recvfrom returned error %d\n", -err);
557 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
558 }
559 return -EAGAIN;
560 }
561 len = svc_addr_len(svc_addr(rqstp));
562 if (len == 0)
563 return -EAFNOSUPPORT;
564 rqstp->rq_addrlen = len;
565 if (skb->tstamp.tv64 == 0) {
566 skb->tstamp = ktime_get_real();
567 /* Don't enable netstamp, sunrpc doesn't
568 need that much accuracy */
569 }
570 svsk->sk_sk->sk_stamp = skb->tstamp;
571 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
572
573 len = skb->len - sizeof(struct udphdr);
574 rqstp->rq_arg.len = len;
575
576 rqstp->rq_prot = IPPROTO_UDP;
577
578 if (!svc_udp_get_dest_address(rqstp, cmh)) {
579 if (net_ratelimit())
580 printk(KERN_WARNING
581 "svc: received unknown control message %d/%d; "
582 "dropping RPC reply datagram\n",
583 cmh->cmsg_level, cmh->cmsg_type);
584 skb_free_datagram_locked(svsk->sk_sk, skb);
585 return 0;
586 }
587
588 if (skb_is_nonlinear(skb)) {
589 /* we have to copy */
590 local_bh_disable();
591 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
592 local_bh_enable();
593 /* checksum error */
594 skb_free_datagram_locked(svsk->sk_sk, skb);
595 return 0;
596 }
597 local_bh_enable();
598 skb_free_datagram_locked(svsk->sk_sk, skb);
599 } else {
600 /* we can use it in-place */
601 rqstp->rq_arg.head[0].iov_base = skb->data +
602 sizeof(struct udphdr);
603 rqstp->rq_arg.head[0].iov_len = len;
604 if (skb_checksum_complete(skb)) {
605 skb_free_datagram_locked(svsk->sk_sk, skb);
606 return 0;
607 }
608 rqstp->rq_xprt_ctxt = skb;
609 }
610
611 rqstp->rq_arg.page_base = 0;
612 if (len <= rqstp->rq_arg.head[0].iov_len) {
613 rqstp->rq_arg.head[0].iov_len = len;
614 rqstp->rq_arg.page_len = 0;
615 rqstp->rq_respages = rqstp->rq_pages+1;
616 } else {
617 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
618 rqstp->rq_respages = rqstp->rq_pages + 1 +
619 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
620 }
621
622 if (serv->sv_stats)
623 serv->sv_stats->netudpcnt++;
624
625 return len;
626 }
627
628 static int
629 svc_udp_sendto(struct svc_rqst *rqstp)
630 {
631 int error;
632
633 error = svc_sendto(rqstp, &rqstp->rq_res);
634 if (error == -ECONNREFUSED)
635 /* ICMP error on earlier request. */
636 error = svc_sendto(rqstp, &rqstp->rq_res);
637
638 return error;
639 }
640
641 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
642 {
643 }
644
645 static int svc_udp_has_wspace(struct svc_xprt *xprt)
646 {
647 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
648 struct svc_serv *serv = xprt->xpt_server;
649 unsigned long required;
650
651 /*
652 * Set the SOCK_NOSPACE flag before checking the available
653 * sock space.
654 */
655 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
656 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
657 if (required*2 > sock_wspace(svsk->sk_sk))
658 return 0;
659 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
660 return 1;
661 }
662
663 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
664 {
665 BUG();
666 return NULL;
667 }
668
669 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
670 struct net *net,
671 struct sockaddr *sa, int salen,
672 int flags)
673 {
674 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
675 }
676
677 static struct svc_xprt_ops svc_udp_ops = {
678 .xpo_create = svc_udp_create,
679 .xpo_recvfrom = svc_udp_recvfrom,
680 .xpo_sendto = svc_udp_sendto,
681 .xpo_release_rqst = svc_release_skb,
682 .xpo_detach = svc_sock_detach,
683 .xpo_free = svc_sock_free,
684 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
685 .xpo_has_wspace = svc_udp_has_wspace,
686 .xpo_accept = svc_udp_accept,
687 };
688
689 static struct svc_xprt_class svc_udp_class = {
690 .xcl_name = "udp",
691 .xcl_owner = THIS_MODULE,
692 .xcl_ops = &svc_udp_ops,
693 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
694 };
695
696 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
697 {
698 int err, level, optname, one = 1;
699
700 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
701 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
702 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
703 svsk->sk_sk->sk_write_space = svc_write_space;
704
705 /* initialise setting must have enough space to
706 * receive and respond to one request.
707 * svc_udp_recvfrom will re-adjust if necessary
708 */
709 svc_sock_setbufsize(svsk->sk_sock,
710 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
711 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
712
713 /* data might have come in before data_ready set up */
714 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
715 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
716
717 /* make sure we get destination address info */
718 switch (svsk->sk_sk->sk_family) {
719 case AF_INET:
720 level = SOL_IP;
721 optname = IP_PKTINFO;
722 break;
723 case AF_INET6:
724 level = SOL_IPV6;
725 optname = IPV6_RECVPKTINFO;
726 break;
727 default:
728 BUG();
729 }
730 err = kernel_setsockopt(svsk->sk_sock, level, optname,
731 (char *)&one, sizeof(one));
732 dprintk("svc: kernel_setsockopt returned %d\n", err);
733 }
734
735 /*
736 * A data_ready event on a listening socket means there's a connection
737 * pending. Do not use state_change as a substitute for it.
738 */
739 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
740 {
741 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
742
743 dprintk("svc: socket %p TCP (listen) state change %d\n",
744 sk, sk->sk_state);
745
746 /*
747 * This callback may called twice when a new connection
748 * is established as a child socket inherits everything
749 * from a parent LISTEN socket.
750 * 1) data_ready method of the parent socket will be called
751 * when one of child sockets become ESTABLISHED.
752 * 2) data_ready method of the child socket may be called
753 * when it receives data before the socket is accepted.
754 * In case of 2, we should ignore it silently.
755 */
756 if (sk->sk_state == TCP_LISTEN) {
757 if (svsk) {
758 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
759 svc_xprt_enqueue(&svsk->sk_xprt);
760 } else
761 printk("svc: socket %p: no user data\n", sk);
762 }
763
764 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
765 wake_up_interruptible_all(sk_sleep(sk));
766 }
767
768 /*
769 * A state change on a connected socket means it's dying or dead.
770 */
771 static void svc_tcp_state_change(struct sock *sk)
772 {
773 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
774
775 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
776 sk, sk->sk_state, sk->sk_user_data);
777
778 if (!svsk)
779 printk("svc: socket %p: no user data\n", sk);
780 else {
781 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
782 svc_xprt_enqueue(&svsk->sk_xprt);
783 }
784 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
785 wake_up_interruptible_all(sk_sleep(sk));
786 }
787
788 static void svc_tcp_data_ready(struct sock *sk, int count)
789 {
790 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
791
792 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
793 sk, sk->sk_user_data);
794 if (svsk) {
795 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
796 svc_xprt_enqueue(&svsk->sk_xprt);
797 }
798 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
799 wake_up_interruptible(sk_sleep(sk));
800 }
801
802 /*
803 * Accept a TCP connection
804 */
805 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
806 {
807 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
808 struct sockaddr_storage addr;
809 struct sockaddr *sin = (struct sockaddr *) &addr;
810 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
811 struct socket *sock = svsk->sk_sock;
812 struct socket *newsock;
813 struct svc_sock *newsvsk;
814 int err, slen;
815 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
816
817 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
818 if (!sock)
819 return NULL;
820
821 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
822 err = kernel_accept(sock, &newsock, O_NONBLOCK);
823 if (err < 0) {
824 if (err == -ENOMEM)
825 printk(KERN_WARNING "%s: no more sockets!\n",
826 serv->sv_name);
827 else if (err != -EAGAIN && net_ratelimit())
828 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
829 serv->sv_name, -err);
830 return NULL;
831 }
832 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
833
834 err = kernel_getpeername(newsock, sin, &slen);
835 if (err < 0) {
836 if (net_ratelimit())
837 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
838 serv->sv_name, -err);
839 goto failed; /* aborted connection or whatever */
840 }
841
842 /* Ideally, we would want to reject connections from unauthorized
843 * hosts here, but when we get encryption, the IP of the host won't
844 * tell us anything. For now just warn about unpriv connections.
845 */
846 if (!svc_port_is_privileged(sin)) {
847 dprintk(KERN_WARNING
848 "%s: connect from unprivileged port: %s\n",
849 serv->sv_name,
850 __svc_print_addr(sin, buf, sizeof(buf)));
851 }
852 dprintk("%s: connect from %s\n", serv->sv_name,
853 __svc_print_addr(sin, buf, sizeof(buf)));
854
855 /* make sure that a write doesn't block forever when
856 * low on memory
857 */
858 newsock->sk->sk_sndtimeo = HZ*30;
859
860 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
861 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
862 goto failed;
863 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
864 err = kernel_getsockname(newsock, sin, &slen);
865 if (unlikely(err < 0)) {
866 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
867 slen = offsetof(struct sockaddr, sa_data);
868 }
869 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
870
871 if (serv->sv_stats)
872 serv->sv_stats->nettcpconn++;
873
874 return &newsvsk->sk_xprt;
875
876 failed:
877 sock_release(newsock);
878 return NULL;
879 }
880
881 /*
882 * Receive data.
883 * If we haven't gotten the record length yet, get the next four bytes.
884 * Otherwise try to gobble up as much as possible up to the complete
885 * record length.
886 */
887 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
888 {
889 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
890 int len;
891
892 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
893 /* sndbuf needs to have room for one request
894 * per thread, otherwise we can stall even when the
895 * network isn't a bottleneck.
896 *
897 * We count all threads rather than threads in a
898 * particular pool, which provides an upper bound
899 * on the number of threads which will access the socket.
900 *
901 * rcvbuf just needs to be able to hold a few requests.
902 * Normally they will be removed from the queue
903 * as soon a a complete request arrives.
904 */
905 svc_sock_setbufsize(svsk->sk_sock,
906 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
907 3 * serv->sv_max_mesg);
908
909 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
910
911 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
912 int want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
913 struct kvec iov;
914
915 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
916 iov.iov_len = want;
917 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
918 goto error;
919 svsk->sk_tcplen += len;
920
921 if (len < want) {
922 dprintk("svc: short recvfrom while reading record "
923 "length (%d of %d)\n", len, want);
924 goto err_again; /* record header not complete */
925 }
926
927 svsk->sk_reclen = ntohl(svsk->sk_reclen);
928 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
929 /* FIXME: technically, a record can be fragmented,
930 * and non-terminal fragments will not have the top
931 * bit set in the fragment length header.
932 * But apparently no known nfs clients send fragmented
933 * records. */
934 if (net_ratelimit())
935 printk(KERN_NOTICE "RPC: multiple fragments "
936 "per record not supported\n");
937 goto err_delete;
938 }
939
940 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
941 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
942 if (svsk->sk_reclen > serv->sv_max_mesg) {
943 if (net_ratelimit())
944 printk(KERN_NOTICE "RPC: "
945 "fragment too large: 0x%08lx\n",
946 (unsigned long)svsk->sk_reclen);
947 goto err_delete;
948 }
949 }
950
951 /* Check whether enough data is available */
952 len = svc_recv_available(svsk);
953 if (len < 0)
954 goto error;
955
956 if (len < svsk->sk_reclen) {
957 dprintk("svc: incomplete TCP record (%d of %d)\n",
958 len, svsk->sk_reclen);
959 goto err_again; /* record not complete */
960 }
961 len = svsk->sk_reclen;
962 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
963
964 return len;
965 error:
966 if (len == -EAGAIN)
967 dprintk("RPC: TCP recv_record got EAGAIN\n");
968 return len;
969 err_delete:
970 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
971 err_again:
972 return -EAGAIN;
973 }
974
975 static int svc_process_calldir(struct svc_sock *svsk, struct svc_rqst *rqstp,
976 struct rpc_rqst **reqpp, struct kvec *vec)
977 {
978 struct rpc_rqst *req = NULL;
979 u32 *p;
980 u32 xid;
981 u32 calldir;
982 int len;
983
984 len = svc_recvfrom(rqstp, vec, 1, 8);
985 if (len < 0)
986 goto error;
987
988 p = (u32 *)rqstp->rq_arg.head[0].iov_base;
989 xid = *p++;
990 calldir = *p;
991
992 if (calldir == 0) {
993 /* REQUEST is the most common case */
994 vec[0] = rqstp->rq_arg.head[0];
995 } else {
996 /* REPLY */
997 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
998
999 if (bc_xprt)
1000 req = xprt_lookup_rqst(bc_xprt, xid);
1001
1002 if (!req) {
1003 printk(KERN_NOTICE
1004 "%s: Got unrecognized reply: "
1005 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1006 __func__, ntohl(calldir),
1007 bc_xprt, xid);
1008 vec[0] = rqstp->rq_arg.head[0];
1009 goto out;
1010 }
1011
1012 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1013 sizeof(struct xdr_buf));
1014 /* copy the xid and call direction */
1015 memcpy(req->rq_private_buf.head[0].iov_base,
1016 rqstp->rq_arg.head[0].iov_base, 8);
1017 vec[0] = req->rq_private_buf.head[0];
1018 }
1019 out:
1020 vec[0].iov_base += 8;
1021 vec[0].iov_len -= 8;
1022 len = svsk->sk_reclen - 8;
1023 error:
1024 *reqpp = req;
1025 return len;
1026 }
1027
1028 /*
1029 * Receive data from a TCP socket.
1030 */
1031 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1032 {
1033 struct svc_sock *svsk =
1034 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1035 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1036 int len;
1037 struct kvec *vec;
1038 int pnum, vlen;
1039 struct rpc_rqst *req = NULL;
1040
1041 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1042 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1043 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1044 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1045
1046 len = svc_tcp_recv_record(svsk, rqstp);
1047 if (len < 0)
1048 goto error;
1049
1050 vec = rqstp->rq_vec;
1051 vec[0] = rqstp->rq_arg.head[0];
1052 vlen = PAGE_SIZE;
1053
1054 /*
1055 * We have enough data for the whole tcp record. Let's try and read the
1056 * first 8 bytes to get the xid and the call direction. We can use this
1057 * to figure out if this is a call or a reply to a callback. If
1058 * sk_reclen is < 8 (xid and calldir), then this is a malformed packet.
1059 * In that case, don't bother with the calldir and just read the data.
1060 * It will be rejected in svc_process.
1061 */
1062 if (len >= 8) {
1063 len = svc_process_calldir(svsk, rqstp, &req, vec);
1064 if (len < 0)
1065 goto err_again;
1066 vlen -= 8;
1067 }
1068
1069 pnum = 1;
1070 while (vlen < len) {
1071 vec[pnum].iov_base = (req) ?
1072 page_address(req->rq_private_buf.pages[pnum - 1]) :
1073 page_address(rqstp->rq_pages[pnum]);
1074 vec[pnum].iov_len = PAGE_SIZE;
1075 pnum++;
1076 vlen += PAGE_SIZE;
1077 }
1078 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1079
1080 /* Now receive data */
1081 len = svc_recvfrom(rqstp, vec, pnum, len);
1082 if (len < 0)
1083 goto err_again;
1084
1085 /*
1086 * Account for the 8 bytes we read earlier
1087 */
1088 len += 8;
1089
1090 if (req) {
1091 xprt_complete_rqst(req->rq_task, len);
1092 len = 0;
1093 goto out;
1094 }
1095 dprintk("svc: TCP complete record (%d bytes)\n", len);
1096 rqstp->rq_arg.len = len;
1097 rqstp->rq_arg.page_base = 0;
1098 if (len <= rqstp->rq_arg.head[0].iov_len) {
1099 rqstp->rq_arg.head[0].iov_len = len;
1100 rqstp->rq_arg.page_len = 0;
1101 } else {
1102 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1103 }
1104
1105 rqstp->rq_xprt_ctxt = NULL;
1106 rqstp->rq_prot = IPPROTO_TCP;
1107
1108 out:
1109 /* Reset TCP read info */
1110 svsk->sk_reclen = 0;
1111 svsk->sk_tcplen = 0;
1112
1113 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1114 if (serv->sv_stats)
1115 serv->sv_stats->nettcpcnt++;
1116
1117 return len;
1118
1119 err_again:
1120 if (len == -EAGAIN) {
1121 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1122 return len;
1123 }
1124 error:
1125 if (len != -EAGAIN) {
1126 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1127 svsk->sk_xprt.xpt_server->sv_name, -len);
1128 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1129 }
1130 return -EAGAIN;
1131 }
1132
1133 /*
1134 * Send out data on TCP socket.
1135 */
1136 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1137 {
1138 struct xdr_buf *xbufp = &rqstp->rq_res;
1139 int sent;
1140 __be32 reclen;
1141
1142 /* Set up the first element of the reply kvec.
1143 * Any other kvecs that may be in use have been taken
1144 * care of by the server implementation itself.
1145 */
1146 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1147 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1148
1149 sent = svc_sendto(rqstp, &rqstp->rq_res);
1150 if (sent != xbufp->len) {
1151 printk(KERN_NOTICE
1152 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1153 "- shutting down socket\n",
1154 rqstp->rq_xprt->xpt_server->sv_name,
1155 (sent<0)?"got error":"sent only",
1156 sent, xbufp->len);
1157 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1158 svc_xprt_enqueue(rqstp->rq_xprt);
1159 sent = -EAGAIN;
1160 }
1161 return sent;
1162 }
1163
1164 /*
1165 * Setup response header. TCP has a 4B record length field.
1166 */
1167 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1168 {
1169 struct kvec *resv = &rqstp->rq_res.head[0];
1170
1171 /* tcp needs a space for the record length... */
1172 svc_putnl(resv, 0);
1173 }
1174
1175 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1176 {
1177 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1178 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1179 int required;
1180
1181 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1182 return 1;
1183 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1184 if (sk_stream_wspace(svsk->sk_sk) >= required)
1185 return 1;
1186 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1187 return 0;
1188 }
1189
1190 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1191 struct net *net,
1192 struct sockaddr *sa, int salen,
1193 int flags)
1194 {
1195 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1196 }
1197
1198 #if defined(CONFIG_NFS_V4_1)
1199 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1200 struct net *, struct sockaddr *,
1201 int, int);
1202 static void svc_bc_sock_free(struct svc_xprt *xprt);
1203
1204 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1205 struct net *net,
1206 struct sockaddr *sa, int salen,
1207 int flags)
1208 {
1209 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1210 }
1211
1212 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1213 {
1214 }
1215
1216 static struct svc_xprt_ops svc_tcp_bc_ops = {
1217 .xpo_create = svc_bc_tcp_create,
1218 .xpo_detach = svc_bc_tcp_sock_detach,
1219 .xpo_free = svc_bc_sock_free,
1220 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1221 };
1222
1223 static struct svc_xprt_class svc_tcp_bc_class = {
1224 .xcl_name = "tcp-bc",
1225 .xcl_owner = THIS_MODULE,
1226 .xcl_ops = &svc_tcp_bc_ops,
1227 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1228 };
1229
1230 static void svc_init_bc_xprt_sock(void)
1231 {
1232 svc_reg_xprt_class(&svc_tcp_bc_class);
1233 }
1234
1235 static void svc_cleanup_bc_xprt_sock(void)
1236 {
1237 svc_unreg_xprt_class(&svc_tcp_bc_class);
1238 }
1239 #else /* CONFIG_NFS_V4_1 */
1240 static void svc_init_bc_xprt_sock(void)
1241 {
1242 }
1243
1244 static void svc_cleanup_bc_xprt_sock(void)
1245 {
1246 }
1247 #endif /* CONFIG_NFS_V4_1 */
1248
1249 static struct svc_xprt_ops svc_tcp_ops = {
1250 .xpo_create = svc_tcp_create,
1251 .xpo_recvfrom = svc_tcp_recvfrom,
1252 .xpo_sendto = svc_tcp_sendto,
1253 .xpo_release_rqst = svc_release_skb,
1254 .xpo_detach = svc_tcp_sock_detach,
1255 .xpo_free = svc_sock_free,
1256 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1257 .xpo_has_wspace = svc_tcp_has_wspace,
1258 .xpo_accept = svc_tcp_accept,
1259 };
1260
1261 static struct svc_xprt_class svc_tcp_class = {
1262 .xcl_name = "tcp",
1263 .xcl_owner = THIS_MODULE,
1264 .xcl_ops = &svc_tcp_ops,
1265 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1266 };
1267
1268 void svc_init_xprt_sock(void)
1269 {
1270 svc_reg_xprt_class(&svc_tcp_class);
1271 svc_reg_xprt_class(&svc_udp_class);
1272 svc_init_bc_xprt_sock();
1273 }
1274
1275 void svc_cleanup_xprt_sock(void)
1276 {
1277 svc_unreg_xprt_class(&svc_tcp_class);
1278 svc_unreg_xprt_class(&svc_udp_class);
1279 svc_cleanup_bc_xprt_sock();
1280 }
1281
1282 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1283 {
1284 struct sock *sk = svsk->sk_sk;
1285
1286 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1287 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1288 if (sk->sk_state == TCP_LISTEN) {
1289 dprintk("setting up TCP socket for listening\n");
1290 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1291 sk->sk_data_ready = svc_tcp_listen_data_ready;
1292 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1293 } else {
1294 dprintk("setting up TCP socket for reading\n");
1295 sk->sk_state_change = svc_tcp_state_change;
1296 sk->sk_data_ready = svc_tcp_data_ready;
1297 sk->sk_write_space = svc_tcp_write_space;
1298
1299 svsk->sk_reclen = 0;
1300 svsk->sk_tcplen = 0;
1301
1302 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1303
1304 /* initialise setting must have enough space to
1305 * receive and respond to one request.
1306 * svc_tcp_recvfrom will re-adjust if necessary
1307 */
1308 svc_sock_setbufsize(svsk->sk_sock,
1309 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1310 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1311
1312 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1313 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1314 if (sk->sk_state != TCP_ESTABLISHED)
1315 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1316 }
1317 }
1318
1319 void svc_sock_update_bufs(struct svc_serv *serv)
1320 {
1321 /*
1322 * The number of server threads has changed. Update
1323 * rcvbuf and sndbuf accordingly on all sockets
1324 */
1325 struct svc_sock *svsk;
1326
1327 spin_lock_bh(&serv->sv_lock);
1328 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1329 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1330 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1331 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1332 spin_unlock_bh(&serv->sv_lock);
1333 }
1334 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1335
1336 /*
1337 * Initialize socket for RPC use and create svc_sock struct
1338 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1339 */
1340 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1341 struct socket *sock,
1342 int *errp, int flags)
1343 {
1344 struct svc_sock *svsk;
1345 struct sock *inet;
1346 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1347
1348 dprintk("svc: svc_setup_socket %p\n", sock);
1349 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1350 *errp = -ENOMEM;
1351 return NULL;
1352 }
1353
1354 inet = sock->sk;
1355
1356 /* Register socket with portmapper */
1357 if (*errp >= 0 && pmap_register)
1358 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1359 ntohs(inet_sk(inet)->inet_sport));
1360
1361 if (*errp < 0) {
1362 kfree(svsk);
1363 return NULL;
1364 }
1365
1366 inet->sk_user_data = svsk;
1367 svsk->sk_sock = sock;
1368 svsk->sk_sk = inet;
1369 svsk->sk_ostate = inet->sk_state_change;
1370 svsk->sk_odata = inet->sk_data_ready;
1371 svsk->sk_owspace = inet->sk_write_space;
1372
1373 /* Initialize the socket */
1374 if (sock->type == SOCK_DGRAM)
1375 svc_udp_init(svsk, serv);
1376 else
1377 svc_tcp_init(svsk, serv);
1378
1379 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1380 svsk, svsk->sk_sk);
1381
1382 return svsk;
1383 }
1384
1385 /**
1386 * svc_addsock - add a listener socket to an RPC service
1387 * @serv: pointer to RPC service to which to add a new listener
1388 * @fd: file descriptor of the new listener
1389 * @name_return: pointer to buffer to fill in with name of listener
1390 * @len: size of the buffer
1391 *
1392 * Fills in socket name and returns positive length of name if successful.
1393 * Name is terminated with '\n'. On error, returns a negative errno
1394 * value.
1395 */
1396 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1397 const size_t len)
1398 {
1399 int err = 0;
1400 struct socket *so = sockfd_lookup(fd, &err);
1401 struct svc_sock *svsk = NULL;
1402
1403 if (!so)
1404 return err;
1405 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1406 err = -EAFNOSUPPORT;
1407 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1408 so->sk->sk_protocol != IPPROTO_UDP)
1409 err = -EPROTONOSUPPORT;
1410 else if (so->state > SS_UNCONNECTED)
1411 err = -EISCONN;
1412 else {
1413 if (!try_module_get(THIS_MODULE))
1414 err = -ENOENT;
1415 else
1416 svsk = svc_setup_socket(serv, so, &err,
1417 SVC_SOCK_DEFAULTS);
1418 if (svsk) {
1419 struct sockaddr_storage addr;
1420 struct sockaddr *sin = (struct sockaddr *)&addr;
1421 int salen;
1422 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1423 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1424 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1425 spin_lock_bh(&serv->sv_lock);
1426 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1427 spin_unlock_bh(&serv->sv_lock);
1428 svc_xprt_received(&svsk->sk_xprt);
1429 err = 0;
1430 } else
1431 module_put(THIS_MODULE);
1432 }
1433 if (err) {
1434 sockfd_put(so);
1435 return err;
1436 }
1437 return svc_one_sock_name(svsk, name_return, len);
1438 }
1439 EXPORT_SYMBOL_GPL(svc_addsock);
1440
1441 /*
1442 * Create socket for RPC service.
1443 */
1444 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1445 int protocol,
1446 struct net *net,
1447 struct sockaddr *sin, int len,
1448 int flags)
1449 {
1450 struct svc_sock *svsk;
1451 struct socket *sock;
1452 int error;
1453 int type;
1454 struct sockaddr_storage addr;
1455 struct sockaddr *newsin = (struct sockaddr *)&addr;
1456 int newlen;
1457 int family;
1458 int val;
1459 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1460
1461 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1462 serv->sv_program->pg_name, protocol,
1463 __svc_print_addr(sin, buf, sizeof(buf)));
1464
1465 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1466 printk(KERN_WARNING "svc: only UDP and TCP "
1467 "sockets supported\n");
1468 return ERR_PTR(-EINVAL);
1469 }
1470
1471 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1472 switch (sin->sa_family) {
1473 case AF_INET6:
1474 family = PF_INET6;
1475 break;
1476 case AF_INET:
1477 family = PF_INET;
1478 break;
1479 default:
1480 return ERR_PTR(-EINVAL);
1481 }
1482
1483 error = __sock_create(net, family, type, protocol, &sock, 1);
1484 if (error < 0)
1485 return ERR_PTR(error);
1486
1487 svc_reclassify_socket(sock);
1488
1489 /*
1490 * If this is an PF_INET6 listener, we want to avoid
1491 * getting requests from IPv4 remotes. Those should
1492 * be shunted to a PF_INET listener via rpcbind.
1493 */
1494 val = 1;
1495 if (family == PF_INET6)
1496 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1497 (char *)&val, sizeof(val));
1498
1499 if (type == SOCK_STREAM)
1500 sock->sk->sk_reuse = 1; /* allow address reuse */
1501 error = kernel_bind(sock, sin, len);
1502 if (error < 0)
1503 goto bummer;
1504
1505 newlen = len;
1506 error = kernel_getsockname(sock, newsin, &newlen);
1507 if (error < 0)
1508 goto bummer;
1509
1510 if (protocol == IPPROTO_TCP) {
1511 if ((error = kernel_listen(sock, 64)) < 0)
1512 goto bummer;
1513 }
1514
1515 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1516 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1517 return (struct svc_xprt *)svsk;
1518 }
1519
1520 bummer:
1521 dprintk("svc: svc_create_socket error = %d\n", -error);
1522 sock_release(sock);
1523 return ERR_PTR(error);
1524 }
1525
1526 /*
1527 * Detach the svc_sock from the socket so that no
1528 * more callbacks occur.
1529 */
1530 static void svc_sock_detach(struct svc_xprt *xprt)
1531 {
1532 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1533 struct sock *sk = svsk->sk_sk;
1534
1535 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1536
1537 /* put back the old socket callbacks */
1538 sk->sk_state_change = svsk->sk_ostate;
1539 sk->sk_data_ready = svsk->sk_odata;
1540 sk->sk_write_space = svsk->sk_owspace;
1541
1542 if (sk_sleep(sk) && waitqueue_active(sk_sleep(sk)))
1543 wake_up_interruptible(sk_sleep(sk));
1544 }
1545
1546 /*
1547 * Disconnect the socket, and reset the callbacks
1548 */
1549 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1550 {
1551 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1552
1553 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1554
1555 svc_sock_detach(xprt);
1556
1557 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags))
1558 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1559 }
1560
1561 /*
1562 * Free the svc_sock's socket resources and the svc_sock itself.
1563 */
1564 static void svc_sock_free(struct svc_xprt *xprt)
1565 {
1566 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1567 dprintk("svc: svc_sock_free(%p)\n", svsk);
1568
1569 if (svsk->sk_sock->file)
1570 sockfd_put(svsk->sk_sock);
1571 else
1572 sock_release(svsk->sk_sock);
1573 kfree(svsk);
1574 }
1575
1576 #if defined(CONFIG_NFS_V4_1)
1577 /*
1578 * Create a back channel svc_xprt which shares the fore channel socket.
1579 */
1580 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1581 int protocol,
1582 struct net *net,
1583 struct sockaddr *sin, int len,
1584 int flags)
1585 {
1586 struct svc_sock *svsk;
1587 struct svc_xprt *xprt;
1588
1589 if (protocol != IPPROTO_TCP) {
1590 printk(KERN_WARNING "svc: only TCP sockets"
1591 " supported on shared back channel\n");
1592 return ERR_PTR(-EINVAL);
1593 }
1594
1595 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1596 if (!svsk)
1597 return ERR_PTR(-ENOMEM);
1598
1599 xprt = &svsk->sk_xprt;
1600 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1601
1602 serv->sv_bc_xprt = xprt;
1603
1604 return xprt;
1605 }
1606
1607 /*
1608 * Free a back channel svc_sock.
1609 */
1610 static void svc_bc_sock_free(struct svc_xprt *xprt)
1611 {
1612 if (xprt)
1613 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1614 }
1615 #endif /* CONFIG_NFS_V4_1 */
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree