Linux 2.6.25.20
[linux/fpc-iii.git] / net / sunrpc / svcsock.c
blobc475977de05aa13240292e7ec16ab2fe9789f94a
1 /*
2 * linux/net/sunrpc/svcsock.c
4 * These are the RPC server socket internals.
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...
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.
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
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_states.h>
42 #include <asm/uaccess.h>
43 #include <asm/ioctls.h>
45 #include <linux/sunrpc/types.h>
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/sunrpc/xdr.h>
48 #include <linux/sunrpc/svcsock.h>
49 #include <linux/sunrpc/stats.h>
51 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
54 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
55 int *errp, int flags);
56 static void svc_udp_data_ready(struct sock *, int);
57 static int svc_udp_recvfrom(struct svc_rqst *);
58 static int svc_udp_sendto(struct svc_rqst *);
59 static void svc_sock_detach(struct svc_xprt *);
60 static void svc_sock_free(struct svc_xprt *);
62 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
63 struct sockaddr *, int, int);
64 #ifdef CONFIG_DEBUG_LOCK_ALLOC
65 static struct lock_class_key svc_key[2];
66 static struct lock_class_key svc_slock_key[2];
68 static void svc_reclassify_socket(struct socket *sock)
70 struct sock *sk = sock->sk;
71 BUG_ON(sock_owned_by_user(sk));
72 switch (sk->sk_family) {
73 case AF_INET:
74 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
75 &svc_slock_key[0],
76 "sk_xprt.xpt_lock-AF_INET-NFSD",
77 &svc_key[0]);
78 break;
80 case AF_INET6:
81 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
82 &svc_slock_key[1],
83 "sk_xprt.xpt_lock-AF_INET6-NFSD",
84 &svc_key[1]);
85 break;
87 default:
88 BUG();
91 #else
92 static void svc_reclassify_socket(struct socket *sock)
95 #endif
98 * Release an skbuff after use
100 static void svc_release_skb(struct svc_rqst *rqstp)
102 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
103 struct svc_deferred_req *dr = rqstp->rq_deferred;
105 if (skb) {
106 struct svc_sock *svsk =
107 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
108 rqstp->rq_xprt_ctxt = NULL;
110 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
111 skb_free_datagram(svsk->sk_sk, skb);
113 if (dr) {
114 rqstp->rq_deferred = NULL;
115 kfree(dr);
119 union svc_pktinfo_u {
120 struct in_pktinfo pkti;
121 struct in6_pktinfo pkti6;
123 #define SVC_PKTINFO_SPACE \
124 CMSG_SPACE(sizeof(union svc_pktinfo_u))
126 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
128 struct svc_sock *svsk =
129 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
130 switch (svsk->sk_sk->sk_family) {
131 case AF_INET: {
132 struct in_pktinfo *pki = CMSG_DATA(cmh);
134 cmh->cmsg_level = SOL_IP;
135 cmh->cmsg_type = IP_PKTINFO;
136 pki->ipi_ifindex = 0;
137 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
138 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
140 break;
142 case AF_INET6: {
143 struct in6_pktinfo *pki = CMSG_DATA(cmh);
145 cmh->cmsg_level = SOL_IPV6;
146 cmh->cmsg_type = IPV6_PKTINFO;
147 pki->ipi6_ifindex = 0;
148 ipv6_addr_copy(&pki->ipi6_addr,
149 &rqstp->rq_daddr.addr6);
150 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
152 break;
154 return;
158 * Generic sendto routine
160 static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
162 struct svc_sock *svsk =
163 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
164 struct socket *sock = svsk->sk_sock;
165 int slen;
166 union {
167 struct cmsghdr hdr;
168 long all[SVC_PKTINFO_SPACE / sizeof(long)];
169 } buffer;
170 struct cmsghdr *cmh = &buffer.hdr;
171 int len = 0;
172 int result;
173 int size;
174 struct page **ppage = xdr->pages;
175 size_t base = xdr->page_base;
176 unsigned int pglen = xdr->page_len;
177 unsigned int flags = MSG_MORE;
178 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
180 slen = xdr->len;
182 if (rqstp->rq_prot == IPPROTO_UDP) {
183 struct msghdr msg = {
184 .msg_name = &rqstp->rq_addr,
185 .msg_namelen = rqstp->rq_addrlen,
186 .msg_control = cmh,
187 .msg_controllen = sizeof(buffer),
188 .msg_flags = MSG_MORE,
191 svc_set_cmsg_data(rqstp, cmh);
193 if (sock_sendmsg(sock, &msg, 0) < 0)
194 goto out;
197 /* send head */
198 if (slen == xdr->head[0].iov_len)
199 flags = 0;
200 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
201 xdr->head[0].iov_len, flags);
202 if (len != xdr->head[0].iov_len)
203 goto out;
204 slen -= xdr->head[0].iov_len;
205 if (slen == 0)
206 goto out;
208 /* send page data */
209 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
210 while (pglen > 0) {
211 if (slen == size)
212 flags = 0;
213 result = kernel_sendpage(sock, *ppage, base, size, flags);
214 if (result > 0)
215 len += result;
216 if (result != size)
217 goto out;
218 slen -= size;
219 pglen -= size;
220 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
221 base = 0;
222 ppage++;
224 /* send tail */
225 if (xdr->tail[0].iov_len) {
226 result = kernel_sendpage(sock, rqstp->rq_respages[0],
227 ((unsigned long)xdr->tail[0].iov_base)
228 & (PAGE_SIZE-1),
229 xdr->tail[0].iov_len, 0);
231 if (result > 0)
232 len += result;
234 out:
235 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
236 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
237 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
239 return len;
243 * Report socket names for nfsdfs
245 static int one_sock_name(char *buf, struct svc_sock *svsk)
247 int len;
249 switch(svsk->sk_sk->sk_family) {
250 case AF_INET:
251 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
252 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
253 "udp" : "tcp",
254 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
255 inet_sk(svsk->sk_sk)->num);
256 break;
257 default:
258 len = sprintf(buf, "*unknown-%d*\n",
259 svsk->sk_sk->sk_family);
261 return len;
265 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
267 struct svc_sock *svsk, *closesk = NULL;
268 int len = 0;
270 if (!serv)
271 return 0;
272 spin_lock_bh(&serv->sv_lock);
273 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
274 int onelen = one_sock_name(buf+len, svsk);
275 if (toclose && strcmp(toclose, buf+len) == 0)
276 closesk = svsk;
277 else
278 len += onelen;
280 spin_unlock_bh(&serv->sv_lock);
281 if (closesk)
282 /* Should unregister with portmap, but you cannot
283 * unregister just one protocol...
285 svc_close_xprt(&closesk->sk_xprt);
286 else if (toclose)
287 return -ENOENT;
288 return len;
290 EXPORT_SYMBOL(svc_sock_names);
293 * Check input queue length
295 static int svc_recv_available(struct svc_sock *svsk)
297 struct socket *sock = svsk->sk_sock;
298 int avail, err;
300 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
302 return (err >= 0)? avail : err;
306 * Generic recvfrom routine.
308 static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
309 int buflen)
311 struct svc_sock *svsk =
312 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
313 struct msghdr msg = {
314 .msg_flags = MSG_DONTWAIT,
316 int len;
318 rqstp->rq_xprt_hlen = 0;
320 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
321 msg.msg_flags);
323 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
324 svsk, iov[0].iov_base, iov[0].iov_len, len);
325 return len;
329 * Set socket snd and rcv buffer lengths
331 static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
332 unsigned int rcv)
334 #if 0
335 mm_segment_t oldfs;
336 oldfs = get_fs(); set_fs(KERNEL_DS);
337 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
338 (char*)&snd, sizeof(snd));
339 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
340 (char*)&rcv, sizeof(rcv));
341 #else
342 /* sock_setsockopt limits use to sysctl_?mem_max,
343 * which isn't acceptable. Until that is made conditional
344 * on not having CAP_SYS_RESOURCE or similar, we go direct...
345 * DaveM said I could!
347 lock_sock(sock->sk);
348 sock->sk->sk_sndbuf = snd * 2;
349 sock->sk->sk_rcvbuf = rcv * 2;
350 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
351 release_sock(sock->sk);
352 #endif
355 * INET callback when data has been received on the socket.
357 static void svc_udp_data_ready(struct sock *sk, int count)
359 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
361 if (svsk) {
362 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
363 svsk, sk, count,
364 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
365 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
366 svc_xprt_enqueue(&svsk->sk_xprt);
368 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
369 wake_up_interruptible(sk->sk_sleep);
373 * INET callback when space is newly available on the socket.
375 static void svc_write_space(struct sock *sk)
377 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
379 if (svsk) {
380 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
381 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
382 svc_xprt_enqueue(&svsk->sk_xprt);
385 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
386 dprintk("RPC svc_write_space: someone sleeping on %p\n",
387 svsk);
388 wake_up_interruptible(sk->sk_sleep);
393 * Copy the UDP datagram's destination address to the rqstp structure.
394 * The 'destination' address in this case is the address to which the
395 * peer sent the datagram, i.e. our local address. For multihomed
396 * hosts, this can change from msg to msg. Note that only the IP
397 * address changes, the port number should remain the same.
399 static void svc_udp_get_dest_address(struct svc_rqst *rqstp,
400 struct cmsghdr *cmh)
402 struct svc_sock *svsk =
403 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
404 switch (svsk->sk_sk->sk_family) {
405 case AF_INET: {
406 struct in_pktinfo *pki = CMSG_DATA(cmh);
407 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
408 break;
410 case AF_INET6: {
411 struct in6_pktinfo *pki = CMSG_DATA(cmh);
412 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
413 break;
419 * Receive a datagram from a UDP socket.
421 static int svc_udp_recvfrom(struct svc_rqst *rqstp)
423 struct svc_sock *svsk =
424 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
425 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
426 struct sk_buff *skb;
427 union {
428 struct cmsghdr hdr;
429 long all[SVC_PKTINFO_SPACE / sizeof(long)];
430 } buffer;
431 struct cmsghdr *cmh = &buffer.hdr;
432 int err, len;
433 struct msghdr msg = {
434 .msg_name = svc_addr(rqstp),
435 .msg_control = cmh,
436 .msg_controllen = sizeof(buffer),
437 .msg_flags = MSG_DONTWAIT,
440 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
441 /* udp sockets need large rcvbuf as all pending
442 * requests are still in that buffer. sndbuf must
443 * also be large enough that there is enough space
444 * for one reply per thread. We count all threads
445 * rather than threads in a particular pool, which
446 * provides an upper bound on the number of threads
447 * which will access the socket.
449 svc_sock_setbufsize(svsk->sk_sock,
450 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
451 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
453 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
454 skb = NULL;
455 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
456 0, 0, MSG_PEEK | MSG_DONTWAIT);
457 if (err >= 0)
458 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
460 if (skb == NULL) {
461 if (err != -EAGAIN) {
462 /* possibly an icmp error */
463 dprintk("svc: recvfrom returned error %d\n", -err);
464 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
466 svc_xprt_received(&svsk->sk_xprt);
467 return -EAGAIN;
469 len = svc_addr_len(svc_addr(rqstp));
470 if (len < 0)
471 return len;
472 rqstp->rq_addrlen = len;
473 if (skb->tstamp.tv64 == 0) {
474 skb->tstamp = ktime_get_real();
475 /* Don't enable netstamp, sunrpc doesn't
476 need that much accuracy */
478 svsk->sk_sk->sk_stamp = skb->tstamp;
479 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
482 * Maybe more packets - kick another thread ASAP.
484 svc_xprt_received(&svsk->sk_xprt);
486 len = skb->len - sizeof(struct udphdr);
487 rqstp->rq_arg.len = len;
489 rqstp->rq_prot = IPPROTO_UDP;
491 if (cmh->cmsg_level != IPPROTO_IP ||
492 cmh->cmsg_type != IP_PKTINFO) {
493 if (net_ratelimit())
494 printk("rpcsvc: received unknown control message:"
495 "%d/%d\n",
496 cmh->cmsg_level, cmh->cmsg_type);
497 skb_free_datagram(svsk->sk_sk, skb);
498 return 0;
500 svc_udp_get_dest_address(rqstp, cmh);
502 if (skb_is_nonlinear(skb)) {
503 /* we have to copy */
504 local_bh_disable();
505 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
506 local_bh_enable();
507 /* checksum error */
508 skb_free_datagram(svsk->sk_sk, skb);
509 return 0;
511 local_bh_enable();
512 skb_free_datagram(svsk->sk_sk, skb);
513 } else {
514 /* we can use it in-place */
515 rqstp->rq_arg.head[0].iov_base = skb->data +
516 sizeof(struct udphdr);
517 rqstp->rq_arg.head[0].iov_len = len;
518 if (skb_checksum_complete(skb)) {
519 skb_free_datagram(svsk->sk_sk, skb);
520 return 0;
522 rqstp->rq_xprt_ctxt = skb;
525 rqstp->rq_arg.page_base = 0;
526 if (len <= rqstp->rq_arg.head[0].iov_len) {
527 rqstp->rq_arg.head[0].iov_len = len;
528 rqstp->rq_arg.page_len = 0;
529 rqstp->rq_respages = rqstp->rq_pages+1;
530 } else {
531 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
532 rqstp->rq_respages = rqstp->rq_pages + 1 +
533 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
536 if (serv->sv_stats)
537 serv->sv_stats->netudpcnt++;
539 return len;
542 static int
543 svc_udp_sendto(struct svc_rqst *rqstp)
545 int error;
547 error = svc_sendto(rqstp, &rqstp->rq_res);
548 if (error == -ECONNREFUSED)
549 /* ICMP error on earlier request. */
550 error = svc_sendto(rqstp, &rqstp->rq_res);
552 return error;
555 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
559 static int svc_udp_has_wspace(struct svc_xprt *xprt)
561 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
562 struct svc_serv *serv = xprt->xpt_server;
563 unsigned long required;
566 * Set the SOCK_NOSPACE flag before checking the available
567 * sock space.
569 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
570 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
571 if (required*2 > sock_wspace(svsk->sk_sk))
572 return 0;
573 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
574 return 1;
577 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
579 BUG();
580 return NULL;
583 static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
584 struct sockaddr *sa, int salen,
585 int flags)
587 return svc_create_socket(serv, IPPROTO_UDP, sa, salen, flags);
590 static struct svc_xprt_ops svc_udp_ops = {
591 .xpo_create = svc_udp_create,
592 .xpo_recvfrom = svc_udp_recvfrom,
593 .xpo_sendto = svc_udp_sendto,
594 .xpo_release_rqst = svc_release_skb,
595 .xpo_detach = svc_sock_detach,
596 .xpo_free = svc_sock_free,
597 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
598 .xpo_has_wspace = svc_udp_has_wspace,
599 .xpo_accept = svc_udp_accept,
602 static struct svc_xprt_class svc_udp_class = {
603 .xcl_name = "udp",
604 .xcl_owner = THIS_MODULE,
605 .xcl_ops = &svc_udp_ops,
606 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
609 static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
611 int one = 1;
612 mm_segment_t oldfs;
614 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
615 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
616 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
617 svsk->sk_sk->sk_write_space = svc_write_space;
619 /* initialise setting must have enough space to
620 * receive and respond to one request.
621 * svc_udp_recvfrom will re-adjust if necessary
623 svc_sock_setbufsize(svsk->sk_sock,
624 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
625 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
627 /* data might have come in before data_ready set up */
628 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
629 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
631 oldfs = get_fs();
632 set_fs(KERNEL_DS);
633 /* make sure we get destination address info */
634 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
635 (char __user *)&one, sizeof(one));
636 set_fs(oldfs);
640 * A data_ready event on a listening socket means there's a connection
641 * pending. Do not use state_change as a substitute for it.
643 static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
645 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
647 dprintk("svc: socket %p TCP (listen) state change %d\n",
648 sk, sk->sk_state);
651 * This callback may called twice when a new connection
652 * is established as a child socket inherits everything
653 * from a parent LISTEN socket.
654 * 1) data_ready method of the parent socket will be called
655 * when one of child sockets become ESTABLISHED.
656 * 2) data_ready method of the child socket may be called
657 * when it receives data before the socket is accepted.
658 * In case of 2, we should ignore it silently.
660 if (sk->sk_state == TCP_LISTEN) {
661 if (svsk) {
662 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
663 svc_xprt_enqueue(&svsk->sk_xprt);
664 } else
665 printk("svc: socket %p: no user data\n", sk);
668 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
669 wake_up_interruptible_all(sk->sk_sleep);
673 * A state change on a connected socket means it's dying or dead.
675 static void svc_tcp_state_change(struct sock *sk)
677 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
679 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
680 sk, sk->sk_state, sk->sk_user_data);
682 if (!svsk)
683 printk("svc: socket %p: no user data\n", sk);
684 else {
685 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
686 svc_xprt_enqueue(&svsk->sk_xprt);
688 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
689 wake_up_interruptible_all(sk->sk_sleep);
692 static void svc_tcp_data_ready(struct sock *sk, int count)
694 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
696 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
697 sk, sk->sk_user_data);
698 if (svsk) {
699 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
700 svc_xprt_enqueue(&svsk->sk_xprt);
702 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
703 wake_up_interruptible(sk->sk_sleep);
707 * Accept a TCP connection
709 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
711 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
712 struct sockaddr_storage addr;
713 struct sockaddr *sin = (struct sockaddr *) &addr;
714 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
715 struct socket *sock = svsk->sk_sock;
716 struct socket *newsock;
717 struct svc_sock *newsvsk;
718 int err, slen;
719 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
721 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
722 if (!sock)
723 return NULL;
725 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
726 err = kernel_accept(sock, &newsock, O_NONBLOCK);
727 if (err < 0) {
728 if (err == -ENOMEM)
729 printk(KERN_WARNING "%s: no more sockets!\n",
730 serv->sv_name);
731 else if (err != -EAGAIN && net_ratelimit())
732 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
733 serv->sv_name, -err);
734 return NULL;
736 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
738 err = kernel_getpeername(newsock, sin, &slen);
739 if (err < 0) {
740 if (net_ratelimit())
741 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
742 serv->sv_name, -err);
743 goto failed; /* aborted connection or whatever */
746 /* Ideally, we would want to reject connections from unauthorized
747 * hosts here, but when we get encryption, the IP of the host won't
748 * tell us anything. For now just warn about unpriv connections.
750 if (!svc_port_is_privileged(sin)) {
751 dprintk(KERN_WARNING
752 "%s: connect from unprivileged port: %s\n",
753 serv->sv_name,
754 __svc_print_addr(sin, buf, sizeof(buf)));
756 dprintk("%s: connect from %s\n", serv->sv_name,
757 __svc_print_addr(sin, buf, sizeof(buf)));
759 /* make sure that a write doesn't block forever when
760 * low on memory
762 newsock->sk->sk_sndtimeo = HZ*30;
764 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
765 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
766 goto failed;
767 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
768 err = kernel_getsockname(newsock, sin, &slen);
769 if (unlikely(err < 0)) {
770 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
771 slen = offsetof(struct sockaddr, sa_data);
773 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
775 if (serv->sv_stats)
776 serv->sv_stats->nettcpconn++;
778 return &newsvsk->sk_xprt;
780 failed:
781 sock_release(newsock);
782 return NULL;
786 * Receive data from a TCP socket.
788 static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
790 struct svc_sock *svsk =
791 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
792 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
793 int len;
794 struct kvec *vec;
795 int pnum, vlen;
797 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
798 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
799 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
800 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
802 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
803 /* sndbuf needs to have room for one request
804 * per thread, otherwise we can stall even when the
805 * network isn't a bottleneck.
807 * We count all threads rather than threads in a
808 * particular pool, which provides an upper bound
809 * on the number of threads which will access the socket.
811 * rcvbuf just needs to be able to hold a few requests.
812 * Normally they will be removed from the queue
813 * as soon a a complete request arrives.
815 svc_sock_setbufsize(svsk->sk_sock,
816 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
817 3 * serv->sv_max_mesg);
819 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
821 /* Receive data. If we haven't got the record length yet, get
822 * the next four bytes. Otherwise try to gobble up as much as
823 * possible up to the complete record length.
825 if (svsk->sk_tcplen < 4) {
826 unsigned long want = 4 - svsk->sk_tcplen;
827 struct kvec iov;
829 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
830 iov.iov_len = want;
831 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
832 goto error;
833 svsk->sk_tcplen += len;
835 if (len < want) {
836 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
837 len, want);
838 svc_xprt_received(&svsk->sk_xprt);
839 return -EAGAIN; /* record header not complete */
842 svsk->sk_reclen = ntohl(svsk->sk_reclen);
843 if (!(svsk->sk_reclen & 0x80000000)) {
844 /* FIXME: technically, a record can be fragmented,
845 * and non-terminal fragments will not have the top
846 * bit set in the fragment length header.
847 * But apparently no known nfs clients send fragmented
848 * records. */
849 if (net_ratelimit())
850 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
851 " (non-terminal)\n",
852 (unsigned long) svsk->sk_reclen);
853 goto err_delete;
855 svsk->sk_reclen &= 0x7fffffff;
856 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
857 if (svsk->sk_reclen > serv->sv_max_mesg) {
858 if (net_ratelimit())
859 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
860 " (large)\n",
861 (unsigned long) svsk->sk_reclen);
862 goto err_delete;
866 /* Check whether enough data is available */
867 len = svc_recv_available(svsk);
868 if (len < 0)
869 goto error;
871 if (len < svsk->sk_reclen) {
872 dprintk("svc: incomplete TCP record (%d of %d)\n",
873 len, svsk->sk_reclen);
874 svc_xprt_received(&svsk->sk_xprt);
875 return -EAGAIN; /* record not complete */
877 len = svsk->sk_reclen;
878 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
880 vec = rqstp->rq_vec;
881 vec[0] = rqstp->rq_arg.head[0];
882 vlen = PAGE_SIZE;
883 pnum = 1;
884 while (vlen < len) {
885 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
886 vec[pnum].iov_len = PAGE_SIZE;
887 pnum++;
888 vlen += PAGE_SIZE;
890 rqstp->rq_respages = &rqstp->rq_pages[pnum];
892 /* Now receive data */
893 len = svc_recvfrom(rqstp, vec, pnum, len);
894 if (len < 0)
895 goto error;
897 dprintk("svc: TCP complete record (%d bytes)\n", len);
898 rqstp->rq_arg.len = len;
899 rqstp->rq_arg.page_base = 0;
900 if (len <= rqstp->rq_arg.head[0].iov_len) {
901 rqstp->rq_arg.head[0].iov_len = len;
902 rqstp->rq_arg.page_len = 0;
903 } else {
904 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
907 rqstp->rq_xprt_ctxt = NULL;
908 rqstp->rq_prot = IPPROTO_TCP;
910 /* Reset TCP read info */
911 svsk->sk_reclen = 0;
912 svsk->sk_tcplen = 0;
914 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
915 svc_xprt_received(&svsk->sk_xprt);
916 if (serv->sv_stats)
917 serv->sv_stats->nettcpcnt++;
919 return len;
921 err_delete:
922 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
923 return -EAGAIN;
925 error:
926 if (len == -EAGAIN) {
927 dprintk("RPC: TCP recvfrom got EAGAIN\n");
928 svc_xprt_received(&svsk->sk_xprt);
929 } else {
930 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
931 svsk->sk_xprt.xpt_server->sv_name, -len);
932 goto err_delete;
935 return len;
939 * Send out data on TCP socket.
941 static int svc_tcp_sendto(struct svc_rqst *rqstp)
943 struct xdr_buf *xbufp = &rqstp->rq_res;
944 int sent;
945 __be32 reclen;
947 /* Set up the first element of the reply kvec.
948 * Any other kvecs that may be in use have been taken
949 * care of by the server implementation itself.
951 reclen = htonl(0x80000000|((xbufp->len ) - 4));
952 memcpy(xbufp->head[0].iov_base, &reclen, 4);
954 if (test_bit(XPT_DEAD, &rqstp->rq_xprt->xpt_flags))
955 return -ENOTCONN;
957 sent = svc_sendto(rqstp, &rqstp->rq_res);
958 if (sent != xbufp->len) {
959 printk(KERN_NOTICE
960 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
961 "- shutting down socket\n",
962 rqstp->rq_xprt->xpt_server->sv_name,
963 (sent<0)?"got error":"sent only",
964 sent, xbufp->len);
965 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
966 svc_xprt_enqueue(rqstp->rq_xprt);
967 sent = -EAGAIN;
969 return sent;
973 * Setup response header. TCP has a 4B record length field.
975 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
977 struct kvec *resv = &rqstp->rq_res.head[0];
979 /* tcp needs a space for the record length... */
980 svc_putnl(resv, 0);
983 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
985 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
986 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
987 int required;
988 int wspace;
991 * Set the SOCK_NOSPACE flag before checking the available
992 * sock space.
994 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
995 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
996 wspace = sk_stream_wspace(svsk->sk_sk);
998 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
999 return 0;
1000 if (required * 2 > wspace)
1001 return 0;
1003 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1004 return 1;
1007 static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1008 struct sockaddr *sa, int salen,
1009 int flags)
1011 return svc_create_socket(serv, IPPROTO_TCP, sa, salen, flags);
1014 static struct svc_xprt_ops svc_tcp_ops = {
1015 .xpo_create = svc_tcp_create,
1016 .xpo_recvfrom = svc_tcp_recvfrom,
1017 .xpo_sendto = svc_tcp_sendto,
1018 .xpo_release_rqst = svc_release_skb,
1019 .xpo_detach = svc_sock_detach,
1020 .xpo_free = svc_sock_free,
1021 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1022 .xpo_has_wspace = svc_tcp_has_wspace,
1023 .xpo_accept = svc_tcp_accept,
1026 static struct svc_xprt_class svc_tcp_class = {
1027 .xcl_name = "tcp",
1028 .xcl_owner = THIS_MODULE,
1029 .xcl_ops = &svc_tcp_ops,
1030 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1033 void svc_init_xprt_sock(void)
1035 svc_reg_xprt_class(&svc_tcp_class);
1036 svc_reg_xprt_class(&svc_udp_class);
1039 void svc_cleanup_xprt_sock(void)
1041 svc_unreg_xprt_class(&svc_tcp_class);
1042 svc_unreg_xprt_class(&svc_udp_class);
1045 static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1047 struct sock *sk = svsk->sk_sk;
1048 struct tcp_sock *tp = tcp_sk(sk);
1050 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1051 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1052 if (sk->sk_state == TCP_LISTEN) {
1053 dprintk("setting up TCP socket for listening\n");
1054 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1055 sk->sk_data_ready = svc_tcp_listen_data_ready;
1056 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1057 } else {
1058 dprintk("setting up TCP socket for reading\n");
1059 sk->sk_state_change = svc_tcp_state_change;
1060 sk->sk_data_ready = svc_tcp_data_ready;
1061 sk->sk_write_space = svc_write_space;
1063 svsk->sk_reclen = 0;
1064 svsk->sk_tcplen = 0;
1066 tp->nonagle = 1; /* disable Nagle's algorithm */
1068 /* initialise setting must have enough space to
1069 * receive and respond to one request.
1070 * svc_tcp_recvfrom will re-adjust if necessary
1072 svc_sock_setbufsize(svsk->sk_sock,
1073 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1074 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1076 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1077 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1078 if (sk->sk_state != TCP_ESTABLISHED)
1079 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1083 void svc_sock_update_bufs(struct svc_serv *serv)
1086 * The number of server threads has changed. Update
1087 * rcvbuf and sndbuf accordingly on all sockets
1089 struct list_head *le;
1091 spin_lock_bh(&serv->sv_lock);
1092 list_for_each(le, &serv->sv_permsocks) {
1093 struct svc_sock *svsk =
1094 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1095 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1097 list_for_each(le, &serv->sv_tempsocks) {
1098 struct svc_sock *svsk =
1099 list_entry(le, struct svc_sock, sk_xprt.xpt_list);
1100 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1102 spin_unlock_bh(&serv->sv_lock);
1106 * Initialize socket for RPC use and create svc_sock struct
1107 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1109 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1110 struct socket *sock,
1111 int *errp, int flags)
1113 struct svc_sock *svsk;
1114 struct sock *inet;
1115 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1117 dprintk("svc: svc_setup_socket %p\n", sock);
1118 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1119 *errp = -ENOMEM;
1120 return NULL;
1123 inet = sock->sk;
1125 /* Register socket with portmapper */
1126 if (*errp >= 0 && pmap_register)
1127 *errp = svc_register(serv, inet->sk_protocol,
1128 ntohs(inet_sk(inet)->sport));
1130 if (*errp < 0) {
1131 kfree(svsk);
1132 return NULL;
1135 inet->sk_user_data = svsk;
1136 svsk->sk_sock = sock;
1137 svsk->sk_sk = inet;
1138 svsk->sk_ostate = inet->sk_state_change;
1139 svsk->sk_odata = inet->sk_data_ready;
1140 svsk->sk_owspace = inet->sk_write_space;
1142 /* Initialize the socket */
1143 if (sock->type == SOCK_DGRAM)
1144 svc_udp_init(svsk, serv);
1145 else
1146 svc_tcp_init(svsk, serv);
1148 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1149 svsk, svsk->sk_sk);
1151 return svsk;
1154 int svc_addsock(struct svc_serv *serv,
1155 int fd,
1156 char *name_return,
1157 int *proto)
1159 int err = 0;
1160 struct socket *so = sockfd_lookup(fd, &err);
1161 struct svc_sock *svsk = NULL;
1163 if (!so)
1164 return err;
1165 if (so->sk->sk_family != AF_INET)
1166 err = -EAFNOSUPPORT;
1167 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1168 so->sk->sk_protocol != IPPROTO_UDP)
1169 err = -EPROTONOSUPPORT;
1170 else if (so->state > SS_UNCONNECTED)
1171 err = -EISCONN;
1172 else {
1173 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
1174 if (svsk) {
1175 struct sockaddr_storage addr;
1176 struct sockaddr *sin = (struct sockaddr *)&addr;
1177 int salen;
1178 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1179 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1180 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1181 spin_lock_bh(&serv->sv_lock);
1182 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1183 spin_unlock_bh(&serv->sv_lock);
1184 svc_xprt_received(&svsk->sk_xprt);
1185 err = 0;
1188 if (err) {
1189 sockfd_put(so);
1190 return err;
1192 if (proto) *proto = so->sk->sk_protocol;
1193 return one_sock_name(name_return, svsk);
1195 EXPORT_SYMBOL_GPL(svc_addsock);
1198 * Create socket for RPC service.
1200 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1201 int protocol,
1202 struct sockaddr *sin, int len,
1203 int flags)
1205 struct svc_sock *svsk;
1206 struct socket *sock;
1207 int error;
1208 int type;
1209 struct sockaddr_storage addr;
1210 struct sockaddr *newsin = (struct sockaddr *)&addr;
1211 int newlen;
1212 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1214 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1215 serv->sv_program->pg_name, protocol,
1216 __svc_print_addr(sin, buf, sizeof(buf)));
1218 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1219 printk(KERN_WARNING "svc: only UDP and TCP "
1220 "sockets supported\n");
1221 return ERR_PTR(-EINVAL);
1223 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1225 error = sock_create_kern(sin->sa_family, type, protocol, &sock);
1226 if (error < 0)
1227 return ERR_PTR(error);
1229 svc_reclassify_socket(sock);
1231 if (type == SOCK_STREAM)
1232 sock->sk->sk_reuse = 1; /* allow address reuse */
1233 error = kernel_bind(sock, sin, len);
1234 if (error < 0)
1235 goto bummer;
1237 newlen = len;
1238 error = kernel_getsockname(sock, newsin, &newlen);
1239 if (error < 0)
1240 goto bummer;
1242 if (protocol == IPPROTO_TCP) {
1243 if ((error = kernel_listen(sock, 64)) < 0)
1244 goto bummer;
1247 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1248 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1249 return (struct svc_xprt *)svsk;
1252 bummer:
1253 dprintk("svc: svc_create_socket error = %d\n", -error);
1254 sock_release(sock);
1255 return ERR_PTR(error);
1259 * Detach the svc_sock from the socket so that no
1260 * more callbacks occur.
1262 static void svc_sock_detach(struct svc_xprt *xprt)
1264 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1265 struct sock *sk = svsk->sk_sk;
1267 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1269 /* put back the old socket callbacks */
1270 sk->sk_state_change = svsk->sk_ostate;
1271 sk->sk_data_ready = svsk->sk_odata;
1272 sk->sk_write_space = svsk->sk_owspace;
1276 * Free the svc_sock's socket resources and the svc_sock itself.
1278 static void svc_sock_free(struct svc_xprt *xprt)
1280 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1281 dprintk("svc: svc_sock_free(%p)\n", svsk);
1283 if (svsk->sk_sock->file)
1284 sockfd_put(svsk->sk_sock);
1285 else
1286 sock_release(svsk->sk_sock);
1287 kfree(svsk);