?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nfsd4: typo logical vs bitwise negate for want_mask
[linux-btrfs-devel.git] / net / sunrpc / svcsock.c
blobdfd686eb0b7f2de3c75132e69194da8dfba7672a
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 #include "sunrpc.h"
55
56 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
57
58
59 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
60 int *errp, int flags);
61 static void svc_udp_data_ready(struct sock *, int);
62 static int svc_udp_recvfrom(struct svc_rqst *);
63 static int svc_udp_sendto(struct svc_rqst *);
64 static void svc_sock_detach(struct svc_xprt *);
65 static void svc_tcp_sock_detach(struct svc_xprt *);
66 static void svc_sock_free(struct svc_xprt *);
67
68 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
69 struct net *, struct sockaddr *,
70 int, int);
71 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
72 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
73 struct net *, struct sockaddr *,
74 int, int);
75 static void svc_bc_sock_free(struct svc_xprt *xprt);
76 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
77
78 #ifdef CONFIG_DEBUG_LOCK_ALLOC
79 static struct lock_class_key svc_key[2];
80 static struct lock_class_key svc_slock_key[2];
81
82 static void svc_reclassify_socket(struct socket *sock)
83 {
84 struct sock *sk = sock->sk;
85 BUG_ON(sock_owned_by_user(sk));
86 switch (sk->sk_family) {
87 case AF_INET:
88 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
89 &svc_slock_key[0],
90 "sk_xprt.xpt_lock-AF_INET-NFSD",
91 &svc_key[0]);
92 break;
93
94 case AF_INET6:
95 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
96 &svc_slock_key[1],
97 "sk_xprt.xpt_lock-AF_INET6-NFSD",
98 &svc_key[1]);
99 break;
100
101 default:
102 BUG();
103 }
104 }
105 #else
106 static void svc_reclassify_socket(struct socket *sock)
107 {
108 }
109 #endif
110
111 /*
112 * Release an skbuff after use
113 */
114 static void svc_release_skb(struct svc_rqst *rqstp)
115 {
116 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
117
118 if (skb) {
119 struct svc_sock *svsk =
120 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
121 rqstp->rq_xprt_ctxt = NULL;
122
123 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
124 skb_free_datagram_locked(svsk->sk_sk, skb);
125 }
126 }
127
128 union svc_pktinfo_u {
129 struct in_pktinfo pkti;
130 struct in6_pktinfo pkti6;
131 };
132 #define SVC_PKTINFO_SPACE \
133 CMSG_SPACE(sizeof(union svc_pktinfo_u))
134
135 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
136 {
137 struct svc_sock *svsk =
138 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
139 switch (svsk->sk_sk->sk_family) {
140 case AF_INET: {
141 struct in_pktinfo *pki = CMSG_DATA(cmh);
142
143 cmh->cmsg_level = SOL_IP;
144 cmh->cmsg_type = IP_PKTINFO;
145 pki->ipi_ifindex = 0;
146 pki->ipi_spec_dst.s_addr =
147 svc_daddr_in(rqstp)->sin_addr.s_addr;
148 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
149 }
150 break;
151
152 case AF_INET6: {
153 struct in6_pktinfo *pki = CMSG_DATA(cmh);
154 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
155
156 cmh->cmsg_level = SOL_IPV6;
157 cmh->cmsg_type = IPV6_PKTINFO;
158 pki->ipi6_ifindex = daddr->sin6_scope_id;
159 ipv6_addr_copy(&pki->ipi6_addr, &daddr->sin6_addr);
160 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
161 }
162 break;
163 }
164 }
165
166 /*
167 * send routine intended to be shared by the fore- and back-channel
168 */
169 int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
170 struct page *headpage, unsigned long headoffset,
171 struct page *tailpage, unsigned long tailoffset)
172 {
173 int result;
174 int size;
175 struct page **ppage = xdr->pages;
176 size_t base = xdr->page_base;
177 unsigned int pglen = xdr->page_len;
178 unsigned int flags = MSG_MORE;
179 int slen;
180 int len = 0;
181
182 slen = xdr->len;
183
184 /* send head */
185 if (slen == xdr->head[0].iov_len)
186 flags = 0;
187 len = kernel_sendpage(sock, headpage, headoffset,
188 xdr->head[0].iov_len, flags);
189 if (len != xdr->head[0].iov_len)
190 goto out;
191 slen -= xdr->head[0].iov_len;
192 if (slen == 0)
193 goto out;
194
195 /* send page data */
196 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
197 while (pglen > 0) {
198 if (slen == size)
199 flags = 0;
200 result = kernel_sendpage(sock, *ppage, base, size, flags);
201 if (result > 0)
202 len += result;
203 if (result != size)
204 goto out;
205 slen -= size;
206 pglen -= size;
207 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
208 base = 0;
209 ppage++;
210 }
211
212 /* send tail */
213 if (xdr->tail[0].iov_len) {
214 result = kernel_sendpage(sock, tailpage, tailoffset,
215 xdr->tail[0].iov_len, 0);
216 if (result > 0)
217 len += result;
218 }
219
220 out:
221 return len;
222 }
223
224
225 /*
226 * Generic sendto routine
227 */
228 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
229 {
230 struct svc_sock *svsk =
231 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
232 struct socket *sock = svsk->sk_sock;
233 union {
234 struct cmsghdr hdr;
235 long all[SVC_PKTINFO_SPACE / sizeof(long)];
236 } buffer;
237 struct cmsghdr *cmh = &buffer.hdr;
238 int len = 0;
239 unsigned long tailoff;
240 unsigned long headoff;
241 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
242
243 if (rqstp->rq_prot == IPPROTO_UDP) {
244 struct msghdr msg = {
245 .msg_name = &rqstp->rq_addr,
246 .msg_namelen = rqstp->rq_addrlen,
247 .msg_control = cmh,
248 .msg_controllen = sizeof(buffer),
249 .msg_flags = MSG_MORE,
250 };
251
252 svc_set_cmsg_data(rqstp, cmh);
253
254 if (sock_sendmsg(sock, &msg, 0) < 0)
255 goto out;
256 }
257
258 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
259 headoff = 0;
260 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
261 rqstp->rq_respages[0], tailoff);
262
263 out:
264 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
265 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
266 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
267
268 return len;
269 }
270
271 /*
272 * Report socket names for nfsdfs
273 */
274 static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
275 {
276 const struct sock *sk = svsk->sk_sk;
277 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
278 "udp" : "tcp";
279 int len;
280
281 switch (sk->sk_family) {
282 case PF_INET:
283 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
284 proto_name,
285 &inet_sk(sk)->inet_rcv_saddr,
286 inet_sk(sk)->inet_num);
287 break;
288 case PF_INET6:
289 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
290 proto_name,
291 &inet6_sk(sk)->rcv_saddr,
292 inet_sk(sk)->inet_num);
293 break;
294 default:
295 len = snprintf(buf, remaining, "*unknown-%d*\n",
296 sk->sk_family);
297 }
298
299 if (len >= remaining) {
300 *buf = '\0';
301 return -ENAMETOOLONG;
302 }
303 return len;
304 }
305
306 /**
307 * svc_sock_names - construct a list of listener names in a string
308 * @serv: pointer to RPC service
309 * @buf: pointer to a buffer to fill in with socket names
310 * @buflen: size of the buffer to be filled
311 * @toclose: pointer to '\0'-terminated C string containing the name
312 * of a listener to be closed
313 *
314 * Fills in @buf with a '\n'-separated list of names of listener
315 * sockets. If @toclose is not NULL, the socket named by @toclose
316 * is closed, and is not included in the output list.
317 *
318 * Returns positive length of the socket name string, or a negative
319 * errno value on error.
320 */
321 int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
322 const char *toclose)
323 {
324 struct svc_sock *svsk, *closesk = NULL;
325 int len = 0;
326
327 if (!serv)
328 return 0;
329
330 spin_lock_bh(&serv->sv_lock);
331 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
332 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
333 if (onelen < 0) {
334 len = onelen;
335 break;
336 }
337 if (toclose && strcmp(toclose, buf + len) == 0) {
338 closesk = svsk;
339 svc_xprt_get(&closesk->sk_xprt);
340 } else
341 len += onelen;
342 }
343 spin_unlock_bh(&serv->sv_lock);
344
345 if (closesk) {
346 /* Should unregister with portmap, but you cannot
347 * unregister just one protocol...
348 */
349 svc_close_xprt(&closesk->sk_xprt);
350 svc_xprt_put(&closesk->sk_xprt);
351 } else if (toclose)
352 return -ENOENT;
353 return len;
354 }
355 EXPORT_SYMBOL_GPL(svc_sock_names);
356
357 /*
358 * Check input queue length
359 */
360 static int svc_recv_available(struct svc_sock *svsk)
361 {
362 struct socket *sock = svsk->sk_sock;
363 int avail, err;
364
365 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
366
367 return (err >= 0)? avail : err;
368 }
369
370 /*
371 * Generic recvfrom routine.
372 */
373 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
374 int buflen)
375 {
376 struct svc_sock *svsk =
377 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
378 struct msghdr msg = {
379 .msg_flags = MSG_DONTWAIT,
380 };
381 int len;
382
383 rqstp->rq_xprt_hlen = 0;
384
385 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
386 msg.msg_flags);
387
388 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
389 svsk, iov[0].iov_base, iov[0].iov_len, len);
390 return len;
391 }
392
393 static int svc_partial_recvfrom(struct svc_rqst *rqstp,
394 struct kvec *iov, int nr,
395 int buflen, unsigned int base)
396 {
397 size_t save_iovlen;
398 void __user *save_iovbase;
399 unsigned int i;
400 int ret;
401
402 if (base == 0)
403 return svc_recvfrom(rqstp, iov, nr, buflen);
404
405 for (i = 0; i < nr; i++) {
406 if (iov[i].iov_len > base)
407 break;
408 base -= iov[i].iov_len;
409 }
410 save_iovlen = iov[i].iov_len;
411 save_iovbase = iov[i].iov_base;
412 iov[i].iov_len -= base;
413 iov[i].iov_base += base;
414 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
415 iov[i].iov_len = save_iovlen;
416 iov[i].iov_base = save_iovbase;
417 return ret;
418 }
419
420 /*
421 * Set socket snd and rcv buffer lengths
422 */
423 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
424 unsigned int rcv)
425 {
426 #if 0
427 mm_segment_t oldfs;
428 oldfs = get_fs(); set_fs(KERNEL_DS);
429 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
430 (char*)&snd, sizeof(snd));
431 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
432 (char*)&rcv, sizeof(rcv));
433 #else
434 /* sock_setsockopt limits use to sysctl_?mem_max,
435 * which isn't acceptable. Until that is made conditional
436 * on not having CAP_SYS_RESOURCE or similar, we go direct...
437 * DaveM said I could!
438 */
439 lock_sock(sock->sk);
440 sock->sk->sk_sndbuf = snd * 2;
441 sock->sk->sk_rcvbuf = rcv * 2;
442 sock->sk->sk_write_space(sock->sk);
443 release_sock(sock->sk);
444 #endif
445 }
446 /*
447 * INET callback when data has been received on the socket.
448 */
449 static void svc_udp_data_ready(struct sock *sk, int count)
450 {
451 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
452 wait_queue_head_t *wq = sk_sleep(sk);
453
454 if (svsk) {
455 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
456 svsk, sk, count,
457 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
458 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
459 svc_xprt_enqueue(&svsk->sk_xprt);
460 }
461 if (wq && waitqueue_active(wq))
462 wake_up_interruptible(wq);
463 }
464
465 /*
466 * INET callback when space is newly available on the socket.
467 */
468 static void svc_write_space(struct sock *sk)
469 {
470 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
471 wait_queue_head_t *wq = sk_sleep(sk);
472
473 if (svsk) {
474 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
475 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
476 svc_xprt_enqueue(&svsk->sk_xprt);
477 }
478
479 if (wq && waitqueue_active(wq)) {
480 dprintk("RPC svc_write_space: someone sleeping on %p\n",
481 svsk);
482 wake_up_interruptible(wq);
483 }
484 }
485
486 static void svc_tcp_write_space(struct sock *sk)
487 {
488 struct socket *sock = sk->sk_socket;
489
490 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
491 clear_bit(SOCK_NOSPACE, &sock->flags);
492 svc_write_space(sk);
493 }
494
495 /*
496 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
497 */
498 static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
499 struct cmsghdr *cmh)
500 {
501 struct in_pktinfo *pki = CMSG_DATA(cmh);
502 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
503
504 if (cmh->cmsg_type != IP_PKTINFO)
505 return 0;
506
507 daddr->sin_family = AF_INET;
508 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
509 return 1;
510 }
511
512 /*
513 * See net/ipv6/datagram.c : datagram_recv_ctl
514 */
515 static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
516 struct cmsghdr *cmh)
517 {
518 struct in6_pktinfo *pki = CMSG_DATA(cmh);
519 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
520
521 if (cmh->cmsg_type != IPV6_PKTINFO)
522 return 0;
523
524 daddr->sin6_family = AF_INET6;
525 ipv6_addr_copy(&daddr->sin6_addr, &pki->ipi6_addr);
526 daddr->sin6_scope_id = pki->ipi6_ifindex;
527 return 1;
528 }
529
530 /*
531 * Copy the UDP datagram's destination address to the rqstp structure.
532 * The 'destination' address in this case is the address to which the
533 * peer sent the datagram, i.e. our local address. For multihomed
534 * hosts, this can change from msg to msg. Note that only the IP
535 * address changes, the port number should remain the same.
536 */
537 static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
538 struct cmsghdr *cmh)
539 {
540 switch (cmh->cmsg_level) {
541 case SOL_IP:
542 return svc_udp_get_dest_address4(rqstp, cmh);
543 case SOL_IPV6:
544 return svc_udp_get_dest_address6(rqstp, cmh);
545 }
546
547 return 0;
548 }
549
550 /*
551 * Receive a datagram from a UDP socket.
552 */
553 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
554 {
555 struct svc_sock *svsk =
556 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
557 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
558 struct sk_buff *skb;
559 union {
560 struct cmsghdr hdr;
561 long all[SVC_PKTINFO_SPACE / sizeof(long)];
562 } buffer;
563 struct cmsghdr *cmh = &buffer.hdr;
564 struct msghdr msg = {
565 .msg_name = svc_addr(rqstp),
566 .msg_control = cmh,
567 .msg_controllen = sizeof(buffer),
568 .msg_flags = MSG_DONTWAIT,
569 };
570 size_t len;
571 int err;
572
573 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
574 /* udp sockets need large rcvbuf as all pending
575 * requests are still in that buffer. sndbuf must
576 * also be large enough that there is enough space
577 * for one reply per thread. We count all threads
578 * rather than threads in a particular pool, which
579 * provides an upper bound on the number of threads
580 * which will access the socket.
581 */
582 svc_sock_setbufsize(svsk->sk_sock,
583 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
584 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
585
586 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
587 skb = NULL;
588 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
589 0, 0, MSG_PEEK | MSG_DONTWAIT);
590 if (err >= 0)
591 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
592
593 if (skb == NULL) {
594 if (err != -EAGAIN) {
595 /* possibly an icmp error */
596 dprintk("svc: recvfrom returned error %d\n", -err);
597 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
598 }
599 return -EAGAIN;
600 }
601 len = svc_addr_len(svc_addr(rqstp));
602 if (len == 0)
603 return -EAFNOSUPPORT;
604 rqstp->rq_addrlen = len;
605 if (skb->tstamp.tv64 == 0) {
606 skb->tstamp = ktime_get_real();
607 /* Don't enable netstamp, sunrpc doesn't
608 need that much accuracy */
609 }
610 svsk->sk_sk->sk_stamp = skb->tstamp;
611 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
612
613 len = skb->len - sizeof(struct udphdr);
614 rqstp->rq_arg.len = len;
615
616 rqstp->rq_prot = IPPROTO_UDP;
617
618 if (!svc_udp_get_dest_address(rqstp, cmh)) {
619 if (net_ratelimit())
620 printk(KERN_WARNING
621 "svc: received unknown control message %d/%d; "
622 "dropping RPC reply datagram\n",
623 cmh->cmsg_level, cmh->cmsg_type);
624 skb_free_datagram_locked(svsk->sk_sk, skb);
625 return 0;
626 }
627 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
628
629 if (skb_is_nonlinear(skb)) {
630 /* we have to copy */
631 local_bh_disable();
632 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
633 local_bh_enable();
634 /* checksum error */
635 skb_free_datagram_locked(svsk->sk_sk, skb);
636 return 0;
637 }
638 local_bh_enable();
639 skb_free_datagram_locked(svsk->sk_sk, skb);
640 } else {
641 /* we can use it in-place */
642 rqstp->rq_arg.head[0].iov_base = skb->data +
643 sizeof(struct udphdr);
644 rqstp->rq_arg.head[0].iov_len = len;
645 if (skb_checksum_complete(skb)) {
646 skb_free_datagram_locked(svsk->sk_sk, skb);
647 return 0;
648 }
649 rqstp->rq_xprt_ctxt = skb;
650 }
651
652 rqstp->rq_arg.page_base = 0;
653 if (len <= rqstp->rq_arg.head[0].iov_len) {
654 rqstp->rq_arg.head[0].iov_len = len;
655 rqstp->rq_arg.page_len = 0;
656 rqstp->rq_respages = rqstp->rq_pages+1;
657 } else {
658 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
659 rqstp->rq_respages = rqstp->rq_pages + 1 +
660 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
661 }
662
663 if (serv->sv_stats)
664 serv->sv_stats->netudpcnt++;
665
666 return len;
667 }
668
669 static int
670 svc_udp_sendto(struct svc_rqst *rqstp)
671 {
672 int error;
673
674 error = svc_sendto(rqstp, &rqstp->rq_res);
675 if (error == -ECONNREFUSED)
676 /* ICMP error on earlier request. */
677 error = svc_sendto(rqstp, &rqstp->rq_res);
678
679 return error;
680 }
681
682 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
683 {
684 }
685
686 static int svc_udp_has_wspace(struct svc_xprt *xprt)
687 {
688 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
689 struct svc_serv *serv = xprt->xpt_server;
690 unsigned long required;
691
692 /*
693 * Set the SOCK_NOSPACE flag before checking the available
694 * sock space.
695 */
696 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
697 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
698 if (required*2 > sock_wspace(svsk->sk_sk))
699 return 0;
700 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
701 return 1;
702 }
703
704 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
705 {
706 BUG();
707 return NULL;
708 }
709
710 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
711 struct net *net,
712 struct sockaddr *sa, int salen,
713 int flags)
714 {
715 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
716 }
717
718 static struct svc_xprt_ops svc_udp_ops = {
719 .xpo_create = svc_udp_create,
720 .xpo_recvfrom = svc_udp_recvfrom,
721 .xpo_sendto = svc_udp_sendto,
722 .xpo_release_rqst = svc_release_skb,
723 .xpo_detach = svc_sock_detach,
724 .xpo_free = svc_sock_free,
725 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
726 .xpo_has_wspace = svc_udp_has_wspace,
727 .xpo_accept = svc_udp_accept,
728 };
729
730 static struct svc_xprt_class svc_udp_class = {
731 .xcl_name = "udp",
732 .xcl_owner = THIS_MODULE,
733 .xcl_ops = &svc_udp_ops,
734 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
735 };
736
737 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
738 {
739 int err, level, optname, one = 1;
740
741 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
742 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
743 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
744 svsk->sk_sk->sk_write_space = svc_write_space;
745
746 /* initialise setting must have enough space to
747 * receive and respond to one request.
748 * svc_udp_recvfrom will re-adjust if necessary
749 */
750 svc_sock_setbufsize(svsk->sk_sock,
751 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
752 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
753
754 /* data might have come in before data_ready set up */
755 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
756 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
757
758 /* make sure we get destination address info */
759 switch (svsk->sk_sk->sk_family) {
760 case AF_INET:
761 level = SOL_IP;
762 optname = IP_PKTINFO;
763 break;
764 case AF_INET6:
765 level = SOL_IPV6;
766 optname = IPV6_RECVPKTINFO;
767 break;
768 default:
769 BUG();
770 }
771 err = kernel_setsockopt(svsk->sk_sock, level, optname,
772 (char *)&one, sizeof(one));
773 dprintk("svc: kernel_setsockopt returned %d\n", err);
774 }
775
776 /*
777 * A data_ready event on a listening socket means there's a connection
778 * pending. Do not use state_change as a substitute for it.
779 */
780 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
781 {
782 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
783 wait_queue_head_t *wq;
784
785 dprintk("svc: socket %p TCP (listen) state change %d\n",
786 sk, sk->sk_state);
787
788 /*
789 * This callback may called twice when a new connection
790 * is established as a child socket inherits everything
791 * from a parent LISTEN socket.
792 * 1) data_ready method of the parent socket will be called
793 * when one of child sockets become ESTABLISHED.
794 * 2) data_ready method of the child socket may be called
795 * when it receives data before the socket is accepted.
796 * In case of 2, we should ignore it silently.
797 */
798 if (sk->sk_state == TCP_LISTEN) {
799 if (svsk) {
800 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
801 svc_xprt_enqueue(&svsk->sk_xprt);
802 } else
803 printk("svc: socket %p: no user data\n", sk);
804 }
805
806 wq = sk_sleep(sk);
807 if (wq && waitqueue_active(wq))
808 wake_up_interruptible_all(wq);
809 }
810
811 /*
812 * A state change on a connected socket means it's dying or dead.
813 */
814 static void svc_tcp_state_change(struct sock *sk)
815 {
816 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
817 wait_queue_head_t *wq = sk_sleep(sk);
818
819 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
820 sk, sk->sk_state, sk->sk_user_data);
821
822 if (!svsk)
823 printk("svc: socket %p: no user data\n", sk);
824 else {
825 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
826 svc_xprt_enqueue(&svsk->sk_xprt);
827 }
828 if (wq && waitqueue_active(wq))
829 wake_up_interruptible_all(wq);
830 }
831
832 static void svc_tcp_data_ready(struct sock *sk, int count)
833 {
834 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
835 wait_queue_head_t *wq = sk_sleep(sk);
836
837 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
838 sk, sk->sk_user_data);
839 if (svsk) {
840 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
841 svc_xprt_enqueue(&svsk->sk_xprt);
842 }
843 if (wq && waitqueue_active(wq))
844 wake_up_interruptible(wq);
845 }
846
847 /*
848 * Accept a TCP connection
849 */
850 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
851 {
852 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
853 struct sockaddr_storage addr;
854 struct sockaddr *sin = (struct sockaddr *) &addr;
855 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
856 struct socket *sock = svsk->sk_sock;
857 struct socket *newsock;
858 struct svc_sock *newsvsk;
859 int err, slen;
860 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
861
862 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
863 if (!sock)
864 return NULL;
865
866 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
867 err = kernel_accept(sock, &newsock, O_NONBLOCK);
868 if (err < 0) {
869 if (err == -ENOMEM)
870 printk(KERN_WARNING "%s: no more sockets!\n",
871 serv->sv_name);
872 else if (err != -EAGAIN && net_ratelimit())
873 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
874 serv->sv_name, -err);
875 return NULL;
876 }
877 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
878
879 err = kernel_getpeername(newsock, sin, &slen);
880 if (err < 0) {
881 if (net_ratelimit())
882 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
883 serv->sv_name, -err);
884 goto failed; /* aborted connection or whatever */
885 }
886
887 /* Ideally, we would want to reject connections from unauthorized
888 * hosts here, but when we get encryption, the IP of the host won't
889 * tell us anything. For now just warn about unpriv connections.
890 */
891 if (!svc_port_is_privileged(sin)) {
892 dprintk(KERN_WARNING
893 "%s: connect from unprivileged port: %s\n",
894 serv->sv_name,
895 __svc_print_addr(sin, buf, sizeof(buf)));
896 }
897 dprintk("%s: connect from %s\n", serv->sv_name,
898 __svc_print_addr(sin, buf, sizeof(buf)));
899
900 /* make sure that a write doesn't block forever when
901 * low on memory
902 */
903 newsock->sk->sk_sndtimeo = HZ*30;
904
905 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
906 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
907 goto failed;
908 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
909 err = kernel_getsockname(newsock, sin, &slen);
910 if (unlikely(err < 0)) {
911 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
912 slen = offsetof(struct sockaddr, sa_data);
913 }
914 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
915
916 if (serv->sv_stats)
917 serv->sv_stats->nettcpconn++;
918
919 return &newsvsk->sk_xprt;
920
921 failed:
922 sock_release(newsock);
923 return NULL;
924 }
925
926 static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
927 {
928 unsigned int i, len, npages;
929
930 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
931 return 0;
932 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
933 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
934 for (i = 0; i < npages; i++) {
935 if (rqstp->rq_pages[i] != NULL)
936 put_page(rqstp->rq_pages[i]);
937 BUG_ON(svsk->sk_pages[i] == NULL);
938 rqstp->rq_pages[i] = svsk->sk_pages[i];
939 svsk->sk_pages[i] = NULL;
940 }
941 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
942 return len;
943 }
944
945 static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
946 {
947 unsigned int i, len, npages;
948
949 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
950 return;
951 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
952 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
953 for (i = 0; i < npages; i++) {
954 svsk->sk_pages[i] = rqstp->rq_pages[i];
955 rqstp->rq_pages[i] = NULL;
956 }
957 }
958
959 static void svc_tcp_clear_pages(struct svc_sock *svsk)
960 {
961 unsigned int i, len, npages;
962
963 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
964 goto out;
965 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
966 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
967 for (i = 0; i < npages; i++) {
968 BUG_ON(svsk->sk_pages[i] == NULL);
969 put_page(svsk->sk_pages[i]);
970 svsk->sk_pages[i] = NULL;
971 }
972 out:
973 svsk->sk_tcplen = 0;
974 }
975
976 /*
977 * Receive data.
978 * If we haven't gotten the record length yet, get the next four bytes.
979 * Otherwise try to gobble up as much as possible up to the complete
980 * record length.
981 */
982 static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
983 {
984 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
985 unsigned int want;
986 int len;
987
988 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
989
990 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
991 struct kvec iov;
992
993 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
994 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
995 iov.iov_len = want;
996 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
997 goto error;
998 svsk->sk_tcplen += len;
999
1000 if (len < want) {
1001 dprintk("svc: short recvfrom while reading record "
1002 "length (%d of %d)\n", len, want);
1003 return -EAGAIN;
1004 }
1005
1006 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1007 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1008 /* FIXME: technically, a record can be fragmented,
1009 * and non-terminal fragments will not have the top
1010 * bit set in the fragment length header.
1011 * But apparently no known nfs clients send fragmented
1012 * records. */
1013 if (net_ratelimit())
1014 printk(KERN_NOTICE "RPC: multiple fragments "
1015 "per record not supported\n");
1016 goto err_delete;
1017 }
1018
1019 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1020 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1021 if (svsk->sk_reclen > serv->sv_max_mesg) {
1022 if (net_ratelimit())
1023 printk(KERN_NOTICE "RPC: "
1024 "fragment too large: 0x%08lx\n",
1025 (unsigned long)svsk->sk_reclen);
1026 goto err_delete;
1027 }
1028 }
1029
1030 if (svsk->sk_reclen < 8)
1031 goto err_delete; /* client is nuts. */
1032
1033 len = svsk->sk_reclen;
1034
1035 return len;
1036 error:
1037 dprintk("RPC: TCP recv_record got %d\n", len);
1038 return len;
1039 err_delete:
1040 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1041 return -EAGAIN;
1042 }
1043
1044 static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1045 {
1046 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1047 struct rpc_rqst *req = NULL;
1048 struct kvec *src, *dst;
1049 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1050 __be32 xid;
1051 __be32 calldir;
1052
1053 xid = *p++;
1054 calldir = *p;
1055
1056 if (bc_xprt)
1057 req = xprt_lookup_rqst(bc_xprt, xid);
1058
1059 if (!req) {
1060 printk(KERN_NOTICE
1061 "%s: Got unrecognized reply: "
1062 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1063 __func__, ntohl(calldir),
1064 bc_xprt, xid);
1065 return -EAGAIN;
1066 }
1067
1068 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1069 /*
1070 * XXX!: cheating for now! Only copying HEAD.
1071 * But we know this is good enough for now (in fact, for any
1072 * callback reply in the forseeable future).
1073 */
1074 dst = &req->rq_private_buf.head[0];
1075 src = &rqstp->rq_arg.head[0];
1076 if (dst->iov_len < src->iov_len)
1077 return -EAGAIN; /* whatever; just giving up. */
1078 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1079 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1080 rqstp->rq_arg.len = 0;
1081 return 0;
1082 }
1083
1084 static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1085 {
1086 int i = 0;
1087 int t = 0;
1088
1089 while (t < len) {
1090 vec[i].iov_base = page_address(pages[i]);
1091 vec[i].iov_len = PAGE_SIZE;
1092 i++;
1093 t += PAGE_SIZE;
1094 }
1095 return i;
1096 }
1097
1098
1099 /*
1100 * Receive data from a TCP socket.
1101 */
1102 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1103 {
1104 struct svc_sock *svsk =
1105 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1106 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1107 int len;
1108 struct kvec *vec;
1109 unsigned int want, base;
1110 __be32 *p;
1111 __be32 calldir;
1112 int pnum;
1113
1114 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1115 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1116 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1117 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1118
1119 len = svc_tcp_recv_record(svsk, rqstp);
1120 if (len < 0)
1121 goto error;
1122
1123 base = svc_tcp_restore_pages(svsk, rqstp);
1124 want = svsk->sk_reclen - base;
1125
1126 vec = rqstp->rq_vec;
1127
1128 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1129 svsk->sk_reclen);
1130
1131 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1132
1133 /* Now receive data */
1134 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1135 if (len >= 0)
1136 svsk->sk_tcplen += len;
1137 if (len != want) {
1138 if (len < 0 && len != -EAGAIN)
1139 goto err_other;
1140 svc_tcp_save_pages(svsk, rqstp);
1141 dprintk("svc: incomplete TCP record (%d of %d)\n",
1142 svsk->sk_tcplen, svsk->sk_reclen);
1143 goto err_noclose;
1144 }
1145
1146 rqstp->rq_arg.len = svsk->sk_reclen;
1147 rqstp->rq_arg.page_base = 0;
1148 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1149 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1150 rqstp->rq_arg.page_len = 0;
1151 } else
1152 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1153
1154 rqstp->rq_xprt_ctxt = NULL;
1155 rqstp->rq_prot = IPPROTO_TCP;
1156
1157 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1158 calldir = p[1];
1159 if (calldir)
1160 len = receive_cb_reply(svsk, rqstp);
1161
1162 /* Reset TCP read info */
1163 svsk->sk_reclen = 0;
1164 svsk->sk_tcplen = 0;
1165 /* If we have more data, signal svc_xprt_enqueue() to try again */
1166 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1167 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1168
1169 if (len < 0)
1170 goto error;
1171
1172 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1173 if (serv->sv_stats)
1174 serv->sv_stats->nettcpcnt++;
1175
1176 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1177 return rqstp->rq_arg.len;
1178
1179 error:
1180 if (len != -EAGAIN)
1181 goto err_other;
1182 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1183 return -EAGAIN;
1184 err_other:
1185 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1186 svsk->sk_xprt.xpt_server->sv_name, -len);
1187 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1188 err_noclose:
1189 return -EAGAIN; /* record not complete */
1190 }
1191
1192 /*
1193 * Send out data on TCP socket.
1194 */
1195 static int svc_tcp_sendto(struct svc_rqst *rqstp)
1196 {
1197 struct xdr_buf *xbufp = &rqstp->rq_res;
1198 int sent;
1199 __be32 reclen;
1200
1201 /* Set up the first element of the reply kvec.
1202 * Any other kvecs that may be in use have been taken
1203 * care of by the server implementation itself.
1204 */
1205 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1206 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1207
1208 sent = svc_sendto(rqstp, &rqstp->rq_res);
1209 if (sent != xbufp->len) {
1210 printk(KERN_NOTICE
1211 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1212 "- shutting down socket\n",
1213 rqstp->rq_xprt->xpt_server->sv_name,
1214 (sent<0)?"got error":"sent only",
1215 sent, xbufp->len);
1216 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1217 svc_xprt_enqueue(rqstp->rq_xprt);
1218 sent = -EAGAIN;
1219 }
1220 return sent;
1221 }
1222
1223 /*
1224 * Setup response header. TCP has a 4B record length field.
1225 */
1226 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1227 {
1228 struct kvec *resv = &rqstp->rq_res.head[0];
1229
1230 /* tcp needs a space for the record length... */
1231 svc_putnl(resv, 0);
1232 }
1233
1234 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1235 {
1236 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1237 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1238 int required;
1239
1240 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1241 return 1;
1242 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1243 if (sk_stream_wspace(svsk->sk_sk) >= required)
1244 return 1;
1245 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1246 return 0;
1247 }
1248
1249 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1250 struct net *net,
1251 struct sockaddr *sa, int salen,
1252 int flags)
1253 {
1254 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1255 }
1256
1257 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1258 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1259 struct net *, struct sockaddr *,
1260 int, int);
1261 static void svc_bc_sock_free(struct svc_xprt *xprt);
1262
1263 static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1264 struct net *net,
1265 struct sockaddr *sa, int salen,
1266 int flags)
1267 {
1268 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1269 }
1270
1271 static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1272 {
1273 }
1274
1275 static struct svc_xprt_ops svc_tcp_bc_ops = {
1276 .xpo_create = svc_bc_tcp_create,
1277 .xpo_detach = svc_bc_tcp_sock_detach,
1278 .xpo_free = svc_bc_sock_free,
1279 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1280 };
1281
1282 static struct svc_xprt_class svc_tcp_bc_class = {
1283 .xcl_name = "tcp-bc",
1284 .xcl_owner = THIS_MODULE,
1285 .xcl_ops = &svc_tcp_bc_ops,
1286 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1287 };
1288
1289 static void svc_init_bc_xprt_sock(void)
1290 {
1291 svc_reg_xprt_class(&svc_tcp_bc_class);
1292 }
1293
1294 static void svc_cleanup_bc_xprt_sock(void)
1295 {
1296 svc_unreg_xprt_class(&svc_tcp_bc_class);
1297 }
1298 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1299 static void svc_init_bc_xprt_sock(void)
1300 {
1301 }
1302
1303 static void svc_cleanup_bc_xprt_sock(void)
1304 {
1305 }
1306 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1307
1308 static struct svc_xprt_ops svc_tcp_ops = {
1309 .xpo_create = svc_tcp_create,
1310 .xpo_recvfrom = svc_tcp_recvfrom,
1311 .xpo_sendto = svc_tcp_sendto,
1312 .xpo_release_rqst = svc_release_skb,
1313 .xpo_detach = svc_tcp_sock_detach,
1314 .xpo_free = svc_sock_free,
1315 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1316 .xpo_has_wspace = svc_tcp_has_wspace,
1317 .xpo_accept = svc_tcp_accept,
1318 };
1319
1320 static struct svc_xprt_class svc_tcp_class = {
1321 .xcl_name = "tcp",
1322 .xcl_owner = THIS_MODULE,
1323 .xcl_ops = &svc_tcp_ops,
1324 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1325 };
1326
1327 void svc_init_xprt_sock(void)
1328 {
1329 svc_reg_xprt_class(&svc_tcp_class);
1330 svc_reg_xprt_class(&svc_udp_class);
1331 svc_init_bc_xprt_sock();
1332 }
1333
1334 void svc_cleanup_xprt_sock(void)
1335 {
1336 svc_unreg_xprt_class(&svc_tcp_class);
1337 svc_unreg_xprt_class(&svc_udp_class);
1338 svc_cleanup_bc_xprt_sock();
1339 }
1340
1341 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1342 {
1343 struct sock *sk = svsk->sk_sk;
1344
1345 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1346 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1347 if (sk->sk_state == TCP_LISTEN) {
1348 dprintk("setting up TCP socket for listening\n");
1349 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1350 sk->sk_data_ready = svc_tcp_listen_data_ready;
1351 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1352 } else {
1353 dprintk("setting up TCP socket for reading\n");
1354 sk->sk_state_change = svc_tcp_state_change;
1355 sk->sk_data_ready = svc_tcp_data_ready;
1356 sk->sk_write_space = svc_tcp_write_space;
1357
1358 svsk->sk_reclen = 0;
1359 svsk->sk_tcplen = 0;
1360 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1361
1362 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1363
1364 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1365 if (sk->sk_state != TCP_ESTABLISHED)
1366 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1367 }
1368 }
1369
1370 void svc_sock_update_bufs(struct svc_serv *serv)
1371 {
1372 /*
1373 * The number of server threads has changed. Update
1374 * rcvbuf and sndbuf accordingly on all sockets
1375 */
1376 struct svc_sock *svsk;
1377
1378 spin_lock_bh(&serv->sv_lock);
1379 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1380 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1381 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1382 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1383 spin_unlock_bh(&serv->sv_lock);
1384 }
1385 EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1386
1387 /*
1388 * Initialize socket for RPC use and create svc_sock struct
1389 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1390 */
1391 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1392 struct socket *sock,
1393 int *errp, int flags)
1394 {
1395 struct svc_sock *svsk;
1396 struct sock *inet;
1397 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1398
1399 dprintk("svc: svc_setup_socket %p\n", sock);
1400 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1401 *errp = -ENOMEM;
1402 return NULL;
1403 }
1404
1405 inet = sock->sk;
1406
1407 /* Register socket with portmapper */
1408 if (*errp >= 0 && pmap_register)
1409 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1410 ntohs(inet_sk(inet)->inet_sport));
1411
1412 if (*errp < 0) {
1413 kfree(svsk);
1414 return NULL;
1415 }
1416
1417 inet->sk_user_data = svsk;
1418 svsk->sk_sock = sock;
1419 svsk->sk_sk = inet;
1420 svsk->sk_ostate = inet->sk_state_change;
1421 svsk->sk_odata = inet->sk_data_ready;
1422 svsk->sk_owspace = inet->sk_write_space;
1423
1424 /* Initialize the socket */
1425 if (sock->type == SOCK_DGRAM)
1426 svc_udp_init(svsk, serv);
1427 else {
1428 /* initialise setting must have enough space to
1429 * receive and respond to one request.
1430 */
1431 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1432 4 * serv->sv_max_mesg);
1433 svc_tcp_init(svsk, serv);
1434 }
1435
1436 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1437 svsk, svsk->sk_sk);
1438
1439 return svsk;
1440 }
1441
1442 /**
1443 * svc_addsock - add a listener socket to an RPC service
1444 * @serv: pointer to RPC service to which to add a new listener
1445 * @fd: file descriptor of the new listener
1446 * @name_return: pointer to buffer to fill in with name of listener
1447 * @len: size of the buffer
1448 *
1449 * Fills in socket name and returns positive length of name if successful.
1450 * Name is terminated with '\n'. On error, returns a negative errno
1451 * value.
1452 */
1453 int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1454 const size_t len)
1455 {
1456 int err = 0;
1457 struct socket *so = sockfd_lookup(fd, &err);
1458 struct svc_sock *svsk = NULL;
1459
1460 if (!so)
1461 return err;
1462 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1463 err = -EAFNOSUPPORT;
1464 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1465 so->sk->sk_protocol != IPPROTO_UDP)
1466 err = -EPROTONOSUPPORT;
1467 else if (so->state > SS_UNCONNECTED)
1468 err = -EISCONN;
1469 else {
1470 if (!try_module_get(THIS_MODULE))
1471 err = -ENOENT;
1472 else
1473 svsk = svc_setup_socket(serv, so, &err,
1474 SVC_SOCK_DEFAULTS);
1475 if (svsk) {
1476 struct sockaddr_storage addr;
1477 struct sockaddr *sin = (struct sockaddr *)&addr;
1478 int salen;
1479 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1480 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1481 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1482 spin_lock_bh(&serv->sv_lock);
1483 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1484 spin_unlock_bh(&serv->sv_lock);
1485 svc_xprt_received(&svsk->sk_xprt);
1486 err = 0;
1487 } else
1488 module_put(THIS_MODULE);
1489 }
1490 if (err) {
1491 sockfd_put(so);
1492 return err;
1493 }
1494 return svc_one_sock_name(svsk, name_return, len);
1495 }
1496 EXPORT_SYMBOL_GPL(svc_addsock);
1497
1498 /*
1499 * Create socket for RPC service.
1500 */
1501 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1502 int protocol,
1503 struct net *net,
1504 struct sockaddr *sin, int len,
1505 int flags)
1506 {
1507 struct svc_sock *svsk;
1508 struct socket *sock;
1509 int error;
1510 int type;
1511 struct sockaddr_storage addr;
1512 struct sockaddr *newsin = (struct sockaddr *)&addr;
1513 int newlen;
1514 int family;
1515 int val;
1516 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1517
1518 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1519 serv->sv_program->pg_name, protocol,
1520 __svc_print_addr(sin, buf, sizeof(buf)));
1521
1522 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1523 printk(KERN_WARNING "svc: only UDP and TCP "
1524 "sockets supported\n");
1525 return ERR_PTR(-EINVAL);
1526 }
1527
1528 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1529 switch (sin->sa_family) {
1530 case AF_INET6:
1531 family = PF_INET6;
1532 break;
1533 case AF_INET:
1534 family = PF_INET;
1535 break;
1536 default:
1537 return ERR_PTR(-EINVAL);
1538 }
1539
1540 error = __sock_create(net, family, type, protocol, &sock, 1);
1541 if (error < 0)
1542 return ERR_PTR(error);
1543
1544 svc_reclassify_socket(sock);
1545
1546 /*
1547 * If this is an PF_INET6 listener, we want to avoid
1548 * getting requests from IPv4 remotes. Those should
1549 * be shunted to a PF_INET listener via rpcbind.
1550 */
1551 val = 1;
1552 if (family == PF_INET6)
1553 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1554 (char *)&val, sizeof(val));
1555
1556 if (type == SOCK_STREAM)
1557 sock->sk->sk_reuse = 1; /* allow address reuse */
1558 error = kernel_bind(sock, sin, len);
1559 if (error < 0)
1560 goto bummer;
1561
1562 newlen = len;
1563 error = kernel_getsockname(sock, newsin, &newlen);
1564 if (error < 0)
1565 goto bummer;
1566
1567 if (protocol == IPPROTO_TCP) {
1568 if ((error = kernel_listen(sock, 64)) < 0)
1569 goto bummer;
1570 }
1571
1572 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1573 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1574 return (struct svc_xprt *)svsk;
1575 }
1576
1577 bummer:
1578 dprintk("svc: svc_create_socket error = %d\n", -error);
1579 sock_release(sock);
1580 return ERR_PTR(error);
1581 }
1582
1583 /*
1584 * Detach the svc_sock from the socket so that no
1585 * more callbacks occur.
1586 */
1587 static void svc_sock_detach(struct svc_xprt *xprt)
1588 {
1589 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1590 struct sock *sk = svsk->sk_sk;
1591 wait_queue_head_t *wq;
1592
1593 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1594
1595 /* put back the old socket callbacks */
1596 sk->sk_state_change = svsk->sk_ostate;
1597 sk->sk_data_ready = svsk->sk_odata;
1598 sk->sk_write_space = svsk->sk_owspace;
1599
1600 wq = sk_sleep(sk);
1601 if (wq && waitqueue_active(wq))
1602 wake_up_interruptible(wq);
1603 }
1604
1605 /*
1606 * Disconnect the socket, and reset the callbacks
1607 */
1608 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1609 {
1610 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1611
1612 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1613
1614 svc_sock_detach(xprt);
1615
1616 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1617 svc_tcp_clear_pages(svsk);
1618 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1619 }
1620 }
1621
1622 /*
1623 * Free the svc_sock's socket resources and the svc_sock itself.
1624 */
1625 static void svc_sock_free(struct svc_xprt *xprt)
1626 {
1627 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1628 dprintk("svc: svc_sock_free(%p)\n", svsk);
1629
1630 if (svsk->sk_sock->file)
1631 sockfd_put(svsk->sk_sock);
1632 else
1633 sock_release(svsk->sk_sock);
1634 kfree(svsk);
1635 }
1636
1637 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1638 /*
1639 * Create a back channel svc_xprt which shares the fore channel socket.
1640 */
1641 static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1642 int protocol,
1643 struct net *net,
1644 struct sockaddr *sin, int len,
1645 int flags)
1646 {
1647 struct svc_sock *svsk;
1648 struct svc_xprt *xprt;
1649
1650 if (protocol != IPPROTO_TCP) {
1651 printk(KERN_WARNING "svc: only TCP sockets"
1652 " supported on shared back channel\n");
1653 return ERR_PTR(-EINVAL);
1654 }
1655
1656 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1657 if (!svsk)
1658 return ERR_PTR(-ENOMEM);
1659
1660 xprt = &svsk->sk_xprt;
1661 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1662
1663 serv->sv_bc_xprt = xprt;
1664
1665 return xprt;
1666 }
1667
1668 /*
1669 * Free a back channel svc_sock.
1670 */
1671 static void svc_bc_sock_free(struct svc_xprt *xprt)
1672 {
1673 if (xprt)
1674 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1675 }
1676 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
Linux Btrfs Development