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_sock_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/sched.h>
23 #include <linux/errno.h>
24 #include <linux/fcntl.h>
25 #include <linux/net.h>
27 #include <linux/inet.h>
28 #include <linux/udp.h>
29 #include <linux/tcp.h>
30 #include <linux/unistd.h>
31 #include <linux/slab.h>
32 #include <linux/netdevice.h>
33 #include <linux/skbuff.h>
34 #include <linux/file.h>
35 #include <linux/freezer.h>
37 #include <net/checksum.h>
39 #include <net/tcp_states.h>
40 #include <asm/uaccess.h>
41 #include <asm/ioctls.h>
43 #include <linux/sunrpc/types.h>
44 #include <linux/sunrpc/xdr.h>
45 #include <linux/sunrpc/svcsock.h>
46 #include <linux/sunrpc/stats.h>
48 /* SMP locking strategy:
50 * svc_pool->sp_lock protects most of the fields of that pool.
51 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
52 * when both need to be taken (rare), svc_serv->sv_lock is first.
53 * BKL protects svc_serv->sv_nrthread.
54 * svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list
55 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply.
57 * Some flags can be set to certain values at any time
58 * providing that certain rules are followed:
60 * SK_CONN, SK_DATA, can be set or cleared at any time.
61 * after a set, svc_sock_enqueue must be called.
62 * after a clear, the socket must be read/accepted
63 * if this succeeds, it must be set again.
64 * SK_CLOSE can set at any time. It is never cleared.
68 #define RPCDBG_FACILITY RPCDBG_SVCSOCK
71 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
72 int *errp, int pmap_reg);
73 static void svc_udp_data_ready(struct sock *, int);
74 static int svc_udp_recvfrom(struct svc_rqst *);
75 static int svc_udp_sendto(struct svc_rqst *);
77 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
78 static int svc_deferred_recv(struct svc_rqst *rqstp);
79 static struct cache_deferred_req *svc_defer(struct cache_req *req);
81 /* apparently the "standard" is that clients close
82 * idle connections after 5 minutes, servers after
84 * http://www.connectathon.org/talks96/nfstcp.pdf
86 static int svc_conn_age_period = 6*60;
88 #ifdef CONFIG_DEBUG_LOCK_ALLOC
89 static struct lock_class_key svc_key[2];
90 static struct lock_class_key svc_slock_key[2];
92 static inline void svc_reclassify_socket(struct socket *sock)
94 struct sock *sk = sock->sk;
95 BUG_ON(sk->sk_lock.owner != NULL);
96 switch (sk->sk_family) {
98 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
99 &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]);
103 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
104 &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]);
112 static inline void svc_reclassify_socket(struct socket *sock)
118 * Queue up an idle server thread. Must have pool->sp_lock held.
119 * Note: this is really a stack rather than a queue, so that we only
120 * use as many different threads as we need, and the rest don't pollute
124 svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
126 list_add(&rqstp->rq_list, &pool->sp_threads);
130 * Dequeue an nfsd thread. Must have pool->sp_lock held.
133 svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
135 list_del(&rqstp->rq_list);
139 * Release an skbuff after use
142 svc_release_skb(struct svc_rqst *rqstp)
144 struct sk_buff *skb = rqstp->rq_skbuff;
145 struct svc_deferred_req *dr = rqstp->rq_deferred;
148 rqstp->rq_skbuff = NULL;
150 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
151 skb_free_datagram(rqstp->rq_sock->sk_sk, skb);
154 rqstp->rq_deferred = NULL;
160 * Any space to write?
162 static inline unsigned long
163 svc_sock_wspace(struct svc_sock *svsk)
167 if (svsk->sk_sock->type == SOCK_STREAM)
168 wspace = sk_stream_wspace(svsk->sk_sk);
170 wspace = sock_wspace(svsk->sk_sk);
176 * Queue up a socket with data pending. If there are idle nfsd
177 * processes, wake 'em up.
181 svc_sock_enqueue(struct svc_sock *svsk)
183 struct svc_serv *serv = svsk->sk_server;
184 struct svc_pool *pool;
185 struct svc_rqst *rqstp;
188 if (!(svsk->sk_flags &
189 ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) ))
191 if (test_bit(SK_DEAD, &svsk->sk_flags))
195 pool = svc_pool_for_cpu(svsk->sk_server, cpu);
198 spin_lock_bh(&pool->sp_lock);
200 if (!list_empty(&pool->sp_threads) &&
201 !list_empty(&pool->sp_sockets))
203 "svc_sock_enqueue: threads and sockets both waiting??\n");
205 if (test_bit(SK_DEAD, &svsk->sk_flags)) {
206 /* Don't enqueue dead sockets */
207 dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk);
211 /* Mark socket as busy. It will remain in this state until the
212 * server has processed all pending data and put the socket back
213 * on the idle list. We update SK_BUSY atomically because
214 * it also guards against trying to enqueue the svc_sock twice.
216 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) {
217 /* Don't enqueue socket while already enqueued */
218 dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk);
221 BUG_ON(svsk->sk_pool != NULL);
222 svsk->sk_pool = pool;
224 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
225 if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2
226 > svc_sock_wspace(svsk))
227 && !test_bit(SK_CLOSE, &svsk->sk_flags)
228 && !test_bit(SK_CONN, &svsk->sk_flags)) {
229 /* Don't enqueue while not enough space for reply */
230 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n",
231 svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg,
232 svc_sock_wspace(svsk));
233 svsk->sk_pool = NULL;
234 clear_bit(SK_BUSY, &svsk->sk_flags);
237 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
240 if (!list_empty(&pool->sp_threads)) {
241 rqstp = list_entry(pool->sp_threads.next,
244 dprintk("svc: socket %p served by daemon %p\n",
246 svc_thread_dequeue(pool, rqstp);
249 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
250 rqstp, rqstp->rq_sock);
251 rqstp->rq_sock = svsk;
252 atomic_inc(&svsk->sk_inuse);
253 rqstp->rq_reserved = serv->sv_max_mesg;
254 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
255 BUG_ON(svsk->sk_pool != pool);
256 wake_up(&rqstp->rq_wait);
258 dprintk("svc: socket %p put into queue\n", svsk->sk_sk);
259 list_add_tail(&svsk->sk_ready, &pool->sp_sockets);
260 BUG_ON(svsk->sk_pool != pool);
264 spin_unlock_bh(&pool->sp_lock);
268 * Dequeue the first socket. Must be called with the pool->sp_lock held.
270 static inline struct svc_sock *
271 svc_sock_dequeue(struct svc_pool *pool)
273 struct svc_sock *svsk;
275 if (list_empty(&pool->sp_sockets))
278 svsk = list_entry(pool->sp_sockets.next,
279 struct svc_sock, sk_ready);
280 list_del_init(&svsk->sk_ready);
282 dprintk("svc: socket %p dequeued, inuse=%d\n",
283 svsk->sk_sk, atomic_read(&svsk->sk_inuse));
289 * Having read something from a socket, check whether it
290 * needs to be re-enqueued.
291 * Note: SK_DATA only gets cleared when a read-attempt finds
292 * no (or insufficient) data.
295 svc_sock_received(struct svc_sock *svsk)
297 svsk->sk_pool = NULL;
298 clear_bit(SK_BUSY, &svsk->sk_flags);
299 svc_sock_enqueue(svsk);
304 * svc_reserve - change the space reserved for the reply to a request.
305 * @rqstp: The request in question
306 * @space: new max space to reserve
308 * Each request reserves some space on the output queue of the socket
309 * to make sure the reply fits. This function reduces that reserved
310 * space to be the amount of space used already, plus @space.
313 void svc_reserve(struct svc_rqst *rqstp, int space)
315 space += rqstp->rq_res.head[0].iov_len;
317 if (space < rqstp->rq_reserved) {
318 struct svc_sock *svsk = rqstp->rq_sock;
319 atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved);
320 rqstp->rq_reserved = space;
322 svc_sock_enqueue(svsk);
327 * Release a socket after use.
330 svc_sock_put(struct svc_sock *svsk)
332 if (atomic_dec_and_test(&svsk->sk_inuse) &&
333 test_bit(SK_DEAD, &svsk->sk_flags)) {
334 dprintk("svc: releasing dead socket\n");
335 if (svsk->sk_sock->file)
336 sockfd_put(svsk->sk_sock);
338 sock_release(svsk->sk_sock);
339 if (svsk->sk_info_authunix != NULL)
340 svcauth_unix_info_release(svsk->sk_info_authunix);
346 svc_sock_release(struct svc_rqst *rqstp)
348 struct svc_sock *svsk = rqstp->rq_sock;
350 svc_release_skb(rqstp);
352 svc_free_res_pages(rqstp);
353 rqstp->rq_res.page_len = 0;
354 rqstp->rq_res.page_base = 0;
357 /* Reset response buffer and release
359 * But first, check that enough space was reserved
360 * for the reply, otherwise we have a bug!
362 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
363 printk(KERN_ERR "RPC request reserved %d but used %d\n",
367 rqstp->rq_res.head[0].iov_len = 0;
368 svc_reserve(rqstp, 0);
369 rqstp->rq_sock = NULL;
375 * External function to wake up a server waiting for data
376 * This really only makes sense for services like lockd
377 * which have exactly one thread anyway.
380 svc_wake_up(struct svc_serv *serv)
382 struct svc_rqst *rqstp;
384 struct svc_pool *pool;
386 for (i = 0; i < serv->sv_nrpools; i++) {
387 pool = &serv->sv_pools[i];
389 spin_lock_bh(&pool->sp_lock);
390 if (!list_empty(&pool->sp_threads)) {
391 rqstp = list_entry(pool->sp_threads.next,
394 dprintk("svc: daemon %p woken up.\n", rqstp);
396 svc_thread_dequeue(pool, rqstp);
397 rqstp->rq_sock = NULL;
399 wake_up(&rqstp->rq_wait);
401 spin_unlock_bh(&pool->sp_lock);
406 * Generic sendto routine
409 svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
411 struct svc_sock *svsk = rqstp->rq_sock;
412 struct socket *sock = svsk->sk_sock;
414 char buffer[CMSG_SPACE(sizeof(struct in_pktinfo))];
415 struct cmsghdr *cmh = (struct cmsghdr *)buffer;
416 struct in_pktinfo *pki = (struct in_pktinfo *)CMSG_DATA(cmh);
420 struct page **ppage = xdr->pages;
421 size_t base = xdr->page_base;
422 unsigned int pglen = xdr->page_len;
423 unsigned int flags = MSG_MORE;
427 if (rqstp->rq_prot == IPPROTO_UDP) {
428 /* set the source and destination */
430 msg.msg_name = &rqstp->rq_addr;
431 msg.msg_namelen = sizeof(rqstp->rq_addr);
434 msg.msg_flags = MSG_MORE;
436 msg.msg_control = cmh;
437 msg.msg_controllen = sizeof(buffer);
438 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
439 cmh->cmsg_level = SOL_IP;
440 cmh->cmsg_type = IP_PKTINFO;
441 pki->ipi_ifindex = 0;
442 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr;
444 if (sock_sendmsg(sock, &msg, 0) < 0)
449 if (slen == xdr->head[0].iov_len)
451 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
452 xdr->head[0].iov_len, flags);
453 if (len != xdr->head[0].iov_len)
455 slen -= xdr->head[0].iov_len;
460 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
464 result = kernel_sendpage(sock, *ppage, base, size, flags);
471 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
476 if (xdr->tail[0].iov_len) {
477 result = kernel_sendpage(sock, rqstp->rq_respages[0],
478 ((unsigned long)xdr->tail[0].iov_base)
480 xdr->tail[0].iov_len, 0);
486 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %x)\n",
487 rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len, xdr->len, len,
488 rqstp->rq_addr.sin_addr.s_addr);
494 * Report socket names for nfsdfs
496 static int one_sock_name(char *buf, struct svc_sock *svsk)
500 switch(svsk->sk_sk->sk_family) {
502 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
503 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
505 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
506 inet_sk(svsk->sk_sk)->num);
509 len = sprintf(buf, "*unknown-%d*\n",
510 svsk->sk_sk->sk_family);
516 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
518 struct svc_sock *svsk, *closesk = NULL;
523 spin_lock(&serv->sv_lock);
524 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) {
525 int onelen = one_sock_name(buf+len, svsk);
526 if (toclose && strcmp(toclose, buf+len) == 0)
531 spin_unlock(&serv->sv_lock);
533 /* Should unregister with portmap, but you cannot
534 * unregister just one protocol...
536 svc_delete_socket(closesk);
541 EXPORT_SYMBOL(svc_sock_names);
544 * Check input queue length
547 svc_recv_available(struct svc_sock *svsk)
549 struct socket *sock = svsk->sk_sock;
552 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
554 return (err >= 0)? avail : err;
558 * Generic recvfrom routine.
561 svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen)
567 rqstp->rq_addrlen = sizeof(rqstp->rq_addr);
568 sock = rqstp->rq_sock->sk_sock;
570 msg.msg_name = &rqstp->rq_addr;
571 msg.msg_namelen = sizeof(rqstp->rq_addr);
572 msg.msg_control = NULL;
573 msg.msg_controllen = 0;
575 msg.msg_flags = MSG_DONTWAIT;
577 len = kernel_recvmsg(sock, &msg, iov, nr, buflen, MSG_DONTWAIT);
579 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
580 * possibly we should cache this in the svc_sock structure
581 * at accept time. FIXME
583 alen = sizeof(rqstp->rq_addr);
584 kernel_getpeername(sock, (struct sockaddr *)&rqstp->rq_addr, &alen);
586 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
587 rqstp->rq_sock, iov[0].iov_base, iov[0].iov_len, len);
593 * Set socket snd and rcv buffer lengths
596 svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
600 oldfs = get_fs(); set_fs(KERNEL_DS);
601 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
602 (char*)&snd, sizeof(snd));
603 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
604 (char*)&rcv, sizeof(rcv));
606 /* sock_setsockopt limits use to sysctl_?mem_max,
607 * which isn't acceptable. Until that is made conditional
608 * on not having CAP_SYS_RESOURCE or similar, we go direct...
609 * DaveM said I could!
612 sock->sk->sk_sndbuf = snd * 2;
613 sock->sk->sk_rcvbuf = rcv * 2;
614 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
615 release_sock(sock->sk);
619 * INET callback when data has been received on the socket.
622 svc_udp_data_ready(struct sock *sk, int count)
624 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
627 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
628 svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags));
629 set_bit(SK_DATA, &svsk->sk_flags);
630 svc_sock_enqueue(svsk);
632 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
633 wake_up_interruptible(sk->sk_sleep);
637 * INET callback when space is newly available on the socket.
640 svc_write_space(struct sock *sk)
642 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
645 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
646 svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags));
647 svc_sock_enqueue(svsk);
650 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
651 dprintk("RPC svc_write_space: someone sleeping on %p\n",
653 wake_up_interruptible(sk->sk_sleep);
658 * Receive a datagram from a UDP socket.
661 svc_udp_recvfrom(struct svc_rqst *rqstp)
663 struct svc_sock *svsk = rqstp->rq_sock;
664 struct svc_serv *serv = svsk->sk_server;
668 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
669 /* udp sockets need large rcvbuf as all pending
670 * requests are still in that buffer. sndbuf must
671 * also be large enough that there is enough space
672 * for one reply per thread. We count all threads
673 * rather than threads in a particular pool, which
674 * provides an upper bound on the number of threads
675 * which will access the socket.
677 svc_sock_setbufsize(svsk->sk_sock,
678 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
679 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
681 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
682 svc_sock_received(svsk);
683 return svc_deferred_recv(rqstp);
686 clear_bit(SK_DATA, &svsk->sk_flags);
687 while ((skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) {
688 if (err == -EAGAIN) {
689 svc_sock_received(svsk);
692 /* possibly an icmp error */
693 dprintk("svc: recvfrom returned error %d\n", -err);
695 if (skb->tstamp.off_sec == 0) {
698 tv.tv_sec = xtime.tv_sec;
699 tv.tv_usec = xtime.tv_nsec / NSEC_PER_USEC;
700 skb_set_timestamp(skb, &tv);
701 /* Don't enable netstamp, sunrpc doesn't
702 need that much accuracy */
704 skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp);
705 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
708 * Maybe more packets - kick another thread ASAP.
710 svc_sock_received(svsk);
712 len = skb->len - sizeof(struct udphdr);
713 rqstp->rq_arg.len = len;
715 rqstp->rq_prot = IPPROTO_UDP;
717 /* Get sender address */
718 rqstp->rq_addr.sin_family = AF_INET;
719 rqstp->rq_addr.sin_port = skb->h.uh->source;
720 rqstp->rq_addr.sin_addr.s_addr = skb->nh.iph->saddr;
721 rqstp->rq_daddr = skb->nh.iph->daddr;
723 if (skb_is_nonlinear(skb)) {
724 /* we have to copy */
726 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
729 skb_free_datagram(svsk->sk_sk, skb);
733 skb_free_datagram(svsk->sk_sk, skb);
735 /* we can use it in-place */
736 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
737 rqstp->rq_arg.head[0].iov_len = len;
738 if (skb_checksum_complete(skb)) {
739 skb_free_datagram(svsk->sk_sk, skb);
742 rqstp->rq_skbuff = skb;
745 rqstp->rq_arg.page_base = 0;
746 if (len <= rqstp->rq_arg.head[0].iov_len) {
747 rqstp->rq_arg.head[0].iov_len = len;
748 rqstp->rq_arg.page_len = 0;
749 rqstp->rq_respages = rqstp->rq_pages+1;
751 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
752 rqstp->rq_respages = rqstp->rq_pages + 1 +
753 (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE;
757 serv->sv_stats->netudpcnt++;
763 svc_udp_sendto(struct svc_rqst *rqstp)
767 error = svc_sendto(rqstp, &rqstp->rq_res);
768 if (error == -ECONNREFUSED)
769 /* ICMP error on earlier request. */
770 error = svc_sendto(rqstp, &rqstp->rq_res);
776 svc_udp_init(struct svc_sock *svsk)
778 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
779 svsk->sk_sk->sk_write_space = svc_write_space;
780 svsk->sk_recvfrom = svc_udp_recvfrom;
781 svsk->sk_sendto = svc_udp_sendto;
783 /* initialise setting must have enough space to
784 * receive and respond to one request.
785 * svc_udp_recvfrom will re-adjust if necessary
787 svc_sock_setbufsize(svsk->sk_sock,
788 3 * svsk->sk_server->sv_max_mesg,
789 3 * svsk->sk_server->sv_max_mesg);
791 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
792 set_bit(SK_CHNGBUF, &svsk->sk_flags);
796 * A data_ready event on a listening socket means there's a connection
797 * pending. Do not use state_change as a substitute for it.
800 svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
802 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
804 dprintk("svc: socket %p TCP (listen) state change %d\n",
808 * This callback may called twice when a new connection
809 * is established as a child socket inherits everything
810 * from a parent LISTEN socket.
811 * 1) data_ready method of the parent socket will be called
812 * when one of child sockets become ESTABLISHED.
813 * 2) data_ready method of the child socket may be called
814 * when it receives data before the socket is accepted.
815 * In case of 2, we should ignore it silently.
817 if (sk->sk_state == TCP_LISTEN) {
819 set_bit(SK_CONN, &svsk->sk_flags);
820 svc_sock_enqueue(svsk);
822 printk("svc: socket %p: no user data\n", sk);
825 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
826 wake_up_interruptible_all(sk->sk_sleep);
830 * A state change on a connected socket means it's dying or dead.
833 svc_tcp_state_change(struct sock *sk)
835 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
837 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
838 sk, sk->sk_state, sk->sk_user_data);
841 printk("svc: socket %p: no user data\n", sk);
843 set_bit(SK_CLOSE, &svsk->sk_flags);
844 svc_sock_enqueue(svsk);
846 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
847 wake_up_interruptible_all(sk->sk_sleep);
851 svc_tcp_data_ready(struct sock *sk, int count)
853 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
855 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
856 sk, sk->sk_user_data);
858 set_bit(SK_DATA, &svsk->sk_flags);
859 svc_sock_enqueue(svsk);
861 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
862 wake_up_interruptible(sk->sk_sleep);
866 * Accept a TCP connection
869 svc_tcp_accept(struct svc_sock *svsk)
871 struct sockaddr_in sin;
872 struct svc_serv *serv = svsk->sk_server;
873 struct socket *sock = svsk->sk_sock;
874 struct socket *newsock;
875 struct svc_sock *newsvsk;
878 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
882 clear_bit(SK_CONN, &svsk->sk_flags);
883 err = kernel_accept(sock, &newsock, O_NONBLOCK);
886 printk(KERN_WARNING "%s: no more sockets!\n",
888 else if (err != -EAGAIN && net_ratelimit())
889 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
890 serv->sv_name, -err);
894 set_bit(SK_CONN, &svsk->sk_flags);
895 svc_sock_enqueue(svsk);
898 err = kernel_getpeername(newsock, (struct sockaddr *) &sin, &slen);
901 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
902 serv->sv_name, -err);
903 goto failed; /* aborted connection or whatever */
906 /* Ideally, we would want to reject connections from unauthorized
907 * hosts here, but when we get encription, the IP of the host won't
908 * tell us anything. For now just warn about unpriv connections.
910 if (ntohs(sin.sin_port) >= 1024) {
912 "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n",
914 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
917 dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv->sv_name,
918 NIPQUAD(sin.sin_addr.s_addr), ntohs(sin.sin_port));
920 /* make sure that a write doesn't block forever when
923 newsock->sk->sk_sndtimeo = HZ*30;
925 if (!(newsvsk = svc_setup_socket(serv, newsock, &err, 0)))
929 /* make sure that we don't have too many active connections.
930 * If we have, something must be dropped.
932 * There's no point in trying to do random drop here for
933 * DoS prevention. The NFS clients does 1 reconnect in 15
934 * seconds. An attacker can easily beat that.
936 * The only somewhat efficient mechanism would be if drop
937 * old connections from the same IP first. But right now
938 * we don't even record the client IP in svc_sock.
940 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
941 struct svc_sock *svsk = NULL;
942 spin_lock_bh(&serv->sv_lock);
943 if (!list_empty(&serv->sv_tempsocks)) {
944 if (net_ratelimit()) {
945 /* Try to help the admin */
946 printk(KERN_NOTICE "%s: too many open TCP "
947 "sockets, consider increasing the "
948 "number of nfsd threads\n",
950 printk(KERN_NOTICE "%s: last TCP connect from "
953 NIPQUAD(sin.sin_addr.s_addr),
954 ntohs(sin.sin_port));
957 * Always select the oldest socket. It's not fair,
960 svsk = list_entry(serv->sv_tempsocks.prev,
963 set_bit(SK_CLOSE, &svsk->sk_flags);
964 atomic_inc(&svsk->sk_inuse);
966 spin_unlock_bh(&serv->sv_lock);
969 svc_sock_enqueue(svsk);
976 serv->sv_stats->nettcpconn++;
981 sock_release(newsock);
986 * Receive data from a TCP socket.
989 svc_tcp_recvfrom(struct svc_rqst *rqstp)
991 struct svc_sock *svsk = rqstp->rq_sock;
992 struct svc_serv *serv = svsk->sk_server;
997 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
998 svsk, test_bit(SK_DATA, &svsk->sk_flags),
999 test_bit(SK_CONN, &svsk->sk_flags),
1000 test_bit(SK_CLOSE, &svsk->sk_flags));
1002 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
1003 svc_sock_received(svsk);
1004 return svc_deferred_recv(rqstp);
1007 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
1008 svc_delete_socket(svsk);
1012 if (svsk->sk_sk->sk_state == TCP_LISTEN) {
1013 svc_tcp_accept(svsk);
1014 svc_sock_received(svsk);
1018 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
1019 /* sndbuf needs to have room for one request
1020 * per thread, otherwise we can stall even when the
1021 * network isn't a bottleneck.
1023 * We count all threads rather than threads in a
1024 * particular pool, which provides an upper bound
1025 * on the number of threads which will access the socket.
1027 * rcvbuf just needs to be able to hold a few requests.
1028 * Normally they will be removed from the queue
1029 * as soon a a complete request arrives.
1031 svc_sock_setbufsize(svsk->sk_sock,
1032 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
1033 3 * serv->sv_max_mesg);
1035 clear_bit(SK_DATA, &svsk->sk_flags);
1037 /* Receive data. If we haven't got the record length yet, get
1038 * the next four bytes. Otherwise try to gobble up as much as
1039 * possible up to the complete record length.
1041 if (svsk->sk_tcplen < 4) {
1042 unsigned long want = 4 - svsk->sk_tcplen;
1045 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1047 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1049 svsk->sk_tcplen += len;
1052 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
1054 svc_sock_received(svsk);
1055 return -EAGAIN; /* record header not complete */
1058 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1059 if (!(svsk->sk_reclen & 0x80000000)) {
1060 /* FIXME: technically, a record can be fragmented,
1061 * and non-terminal fragments will not have the top
1062 * bit set in the fragment length header.
1063 * But apparently no known nfs clients send fragmented
1065 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (non-terminal)\n",
1066 (unsigned long) svsk->sk_reclen);
1069 svsk->sk_reclen &= 0x7fffffff;
1070 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1071 if (svsk->sk_reclen > serv->sv_max_mesg) {
1072 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx (large)\n",
1073 (unsigned long) svsk->sk_reclen);
1078 /* Check whether enough data is available */
1079 len = svc_recv_available(svsk);
1083 if (len < svsk->sk_reclen) {
1084 dprintk("svc: incomplete TCP record (%d of %d)\n",
1085 len, svsk->sk_reclen);
1086 svc_sock_received(svsk);
1087 return -EAGAIN; /* record not complete */
1089 len = svsk->sk_reclen;
1090 set_bit(SK_DATA, &svsk->sk_flags);
1092 vec = rqstp->rq_vec;
1093 vec[0] = rqstp->rq_arg.head[0];
1096 while (vlen < len) {
1097 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
1098 vec[pnum].iov_len = PAGE_SIZE;
1102 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1104 /* Now receive data */
1105 len = svc_recvfrom(rqstp, vec, pnum, len);
1109 dprintk("svc: TCP complete record (%d bytes)\n", len);
1110 rqstp->rq_arg.len = len;
1111 rqstp->rq_arg.page_base = 0;
1112 if (len <= rqstp->rq_arg.head[0].iov_len) {
1113 rqstp->rq_arg.head[0].iov_len = len;
1114 rqstp->rq_arg.page_len = 0;
1116 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1119 rqstp->rq_skbuff = NULL;
1120 rqstp->rq_prot = IPPROTO_TCP;
1122 /* Reset TCP read info */
1123 svsk->sk_reclen = 0;
1124 svsk->sk_tcplen = 0;
1126 svc_sock_received(svsk);
1128 serv->sv_stats->nettcpcnt++;
1133 svc_delete_socket(svsk);
1137 if (len == -EAGAIN) {
1138 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1139 svc_sock_received(svsk);
1141 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1142 svsk->sk_server->sv_name, -len);
1150 * Send out data on TCP socket.
1153 svc_tcp_sendto(struct svc_rqst *rqstp)
1155 struct xdr_buf *xbufp = &rqstp->rq_res;
1159 /* Set up the first element of the reply kvec.
1160 * Any other kvecs that may be in use have been taken
1161 * care of by the server implementation itself.
1163 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1164 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1166 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1169 sent = svc_sendto(rqstp, &rqstp->rq_res);
1170 if (sent != xbufp->len) {
1171 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1172 rqstp->rq_sock->sk_server->sv_name,
1173 (sent<0)?"got error":"sent only",
1175 svc_delete_socket(rqstp->rq_sock);
1182 svc_tcp_init(struct svc_sock *svsk)
1184 struct sock *sk = svsk->sk_sk;
1185 struct tcp_sock *tp = tcp_sk(sk);
1187 svsk->sk_recvfrom = svc_tcp_recvfrom;
1188 svsk->sk_sendto = svc_tcp_sendto;
1190 if (sk->sk_state == TCP_LISTEN) {
1191 dprintk("setting up TCP socket for listening\n");
1192 sk->sk_data_ready = svc_tcp_listen_data_ready;
1193 set_bit(SK_CONN, &svsk->sk_flags);
1195 dprintk("setting up TCP socket for reading\n");
1196 sk->sk_state_change = svc_tcp_state_change;
1197 sk->sk_data_ready = svc_tcp_data_ready;
1198 sk->sk_write_space = svc_write_space;
1200 svsk->sk_reclen = 0;
1201 svsk->sk_tcplen = 0;
1203 tp->nonagle = 1; /* disable Nagle's algorithm */
1205 /* initialise setting must have enough space to
1206 * receive and respond to one request.
1207 * svc_tcp_recvfrom will re-adjust if necessary
1209 svc_sock_setbufsize(svsk->sk_sock,
1210 3 * svsk->sk_server->sv_max_mesg,
1211 3 * svsk->sk_server->sv_max_mesg);
1213 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1214 set_bit(SK_DATA, &svsk->sk_flags);
1215 if (sk->sk_state != TCP_ESTABLISHED)
1216 set_bit(SK_CLOSE, &svsk->sk_flags);
1221 svc_sock_update_bufs(struct svc_serv *serv)
1224 * The number of server threads has changed. Update
1225 * rcvbuf and sndbuf accordingly on all sockets
1227 struct list_head *le;
1229 spin_lock_bh(&serv->sv_lock);
1230 list_for_each(le, &serv->sv_permsocks) {
1231 struct svc_sock *svsk =
1232 list_entry(le, struct svc_sock, sk_list);
1233 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1235 list_for_each(le, &serv->sv_tempsocks) {
1236 struct svc_sock *svsk =
1237 list_entry(le, struct svc_sock, sk_list);
1238 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1240 spin_unlock_bh(&serv->sv_lock);
1244 * Receive the next request on any socket. This code is carefully
1245 * organised not to touch any cachelines in the shared svc_serv
1246 * structure, only cachelines in the local svc_pool.
1249 svc_recv(struct svc_rqst *rqstp, long timeout)
1251 struct svc_sock *svsk =NULL;
1252 struct svc_serv *serv = rqstp->rq_server;
1253 struct svc_pool *pool = rqstp->rq_pool;
1256 struct xdr_buf *arg;
1257 DECLARE_WAITQUEUE(wait, current);
1259 dprintk("svc: server %p waiting for data (to = %ld)\n",
1264 "svc_recv: service %p, socket not NULL!\n",
1266 if (waitqueue_active(&rqstp->rq_wait))
1268 "svc_recv: service %p, wait queue active!\n",
1272 /* now allocate needed pages. If we get a failure, sleep briefly */
1273 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
1274 for (i=0; i < pages ; i++)
1275 while (rqstp->rq_pages[i] == NULL) {
1276 struct page *p = alloc_page(GFP_KERNEL);
1278 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1279 rqstp->rq_pages[i] = p;
1281 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
1282 BUG_ON(pages >= RPCSVC_MAXPAGES);
1284 /* Make arg->head point to first page and arg->pages point to rest */
1285 arg = &rqstp->rq_arg;
1286 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
1287 arg->head[0].iov_len = PAGE_SIZE;
1288 arg->pages = rqstp->rq_pages + 1;
1290 /* save at least one page for response */
1291 arg->page_len = (pages-2)*PAGE_SIZE;
1292 arg->len = (pages-1)*PAGE_SIZE;
1293 arg->tail[0].iov_len = 0;
1300 spin_lock_bh(&pool->sp_lock);
1301 if ((svsk = svc_sock_dequeue(pool)) != NULL) {
1302 rqstp->rq_sock = svsk;
1303 atomic_inc(&svsk->sk_inuse);
1304 rqstp->rq_reserved = serv->sv_max_mesg;
1305 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
1307 /* No data pending. Go to sleep */
1308 svc_thread_enqueue(pool, rqstp);
1311 * We have to be able to interrupt this wait
1312 * to bring down the daemons ...
1314 set_current_state(TASK_INTERRUPTIBLE);
1315 add_wait_queue(&rqstp->rq_wait, &wait);
1316 spin_unlock_bh(&pool->sp_lock);
1318 schedule_timeout(timeout);
1322 spin_lock_bh(&pool->sp_lock);
1323 remove_wait_queue(&rqstp->rq_wait, &wait);
1325 if (!(svsk = rqstp->rq_sock)) {
1326 svc_thread_dequeue(pool, rqstp);
1327 spin_unlock_bh(&pool->sp_lock);
1328 dprintk("svc: server %p, no data yet\n", rqstp);
1329 return signalled()? -EINTR : -EAGAIN;
1332 spin_unlock_bh(&pool->sp_lock);
1334 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
1335 rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse));
1336 len = svsk->sk_recvfrom(rqstp);
1337 dprintk("svc: got len=%d\n", len);
1339 /* No data, incomplete (TCP) read, or accept() */
1340 if (len == 0 || len == -EAGAIN) {
1341 rqstp->rq_res.len = 0;
1342 svc_sock_release(rqstp);
1345 svsk->sk_lastrecv = get_seconds();
1346 clear_bit(SK_OLD, &svsk->sk_flags);
1348 rqstp->rq_secure = ntohs(rqstp->rq_addr.sin_port) < 1024;
1349 rqstp->rq_chandle.defer = svc_defer;
1352 serv->sv_stats->netcnt++;
1360 svc_drop(struct svc_rqst *rqstp)
1362 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1363 svc_sock_release(rqstp);
1367 * Return reply to client.
1370 svc_send(struct svc_rqst *rqstp)
1372 struct svc_sock *svsk;
1376 if ((svsk = rqstp->rq_sock) == NULL) {
1377 printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1378 __FILE__, __LINE__);
1382 /* release the receive skb before sending the reply */
1383 svc_release_skb(rqstp);
1385 /* calculate over-all length */
1386 xb = & rqstp->rq_res;
1387 xb->len = xb->head[0].iov_len +
1389 xb->tail[0].iov_len;
1391 /* Grab svsk->sk_mutex to serialize outgoing data. */
1392 mutex_lock(&svsk->sk_mutex);
1393 if (test_bit(SK_DEAD, &svsk->sk_flags))
1396 len = svsk->sk_sendto(rqstp);
1397 mutex_unlock(&svsk->sk_mutex);
1398 svc_sock_release(rqstp);
1400 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1406 * Timer function to close old temporary sockets, using
1407 * a mark-and-sweep algorithm.
1410 svc_age_temp_sockets(unsigned long closure)
1412 struct svc_serv *serv = (struct svc_serv *)closure;
1413 struct svc_sock *svsk;
1414 struct list_head *le, *next;
1415 LIST_HEAD(to_be_aged);
1417 dprintk("svc_age_temp_sockets\n");
1419 if (!spin_trylock_bh(&serv->sv_lock)) {
1420 /* busy, try again 1 sec later */
1421 dprintk("svc_age_temp_sockets: busy\n");
1422 mod_timer(&serv->sv_temptimer, jiffies + HZ);
1426 list_for_each_safe(le, next, &serv->sv_tempsocks) {
1427 svsk = list_entry(le, struct svc_sock, sk_list);
1429 if (!test_and_set_bit(SK_OLD, &svsk->sk_flags))
1431 if (atomic_read(&svsk->sk_inuse) || test_bit(SK_BUSY, &svsk->sk_flags))
1433 atomic_inc(&svsk->sk_inuse);
1434 list_move(le, &to_be_aged);
1435 set_bit(SK_CLOSE, &svsk->sk_flags);
1436 set_bit(SK_DETACHED, &svsk->sk_flags);
1438 spin_unlock_bh(&serv->sv_lock);
1440 while (!list_empty(&to_be_aged)) {
1441 le = to_be_aged.next;
1442 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */
1444 svsk = list_entry(le, struct svc_sock, sk_list);
1446 dprintk("queuing svsk %p for closing, %lu seconds old\n",
1447 svsk, get_seconds() - svsk->sk_lastrecv);
1449 /* a thread will dequeue and close it soon */
1450 svc_sock_enqueue(svsk);
1454 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
1458 * Initialize socket for RPC use and create svc_sock struct
1459 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1461 static struct svc_sock *
1462 svc_setup_socket(struct svc_serv *serv, struct socket *sock,
1463 int *errp, int pmap_register)
1465 struct svc_sock *svsk;
1468 dprintk("svc: svc_setup_socket %p\n", sock);
1469 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1476 /* Register socket with portmapper */
1477 if (*errp >= 0 && pmap_register)
1478 *errp = svc_register(serv, inet->sk_protocol,
1479 ntohs(inet_sk(inet)->sport));
1486 set_bit(SK_BUSY, &svsk->sk_flags);
1487 inet->sk_user_data = svsk;
1488 svsk->sk_sock = sock;
1490 svsk->sk_ostate = inet->sk_state_change;
1491 svsk->sk_odata = inet->sk_data_ready;
1492 svsk->sk_owspace = inet->sk_write_space;
1493 svsk->sk_server = serv;
1494 atomic_set(&svsk->sk_inuse, 0);
1495 svsk->sk_lastrecv = get_seconds();
1496 spin_lock_init(&svsk->sk_defer_lock);
1497 INIT_LIST_HEAD(&svsk->sk_deferred);
1498 INIT_LIST_HEAD(&svsk->sk_ready);
1499 mutex_init(&svsk->sk_mutex);
1501 /* Initialize the socket */
1502 if (sock->type == SOCK_DGRAM)
1507 spin_lock_bh(&serv->sv_lock);
1508 if (!pmap_register) {
1509 set_bit(SK_TEMP, &svsk->sk_flags);
1510 list_add(&svsk->sk_list, &serv->sv_tempsocks);
1512 if (serv->sv_temptimer.function == NULL) {
1513 /* setup timer to age temp sockets */
1514 setup_timer(&serv->sv_temptimer, svc_age_temp_sockets,
1515 (unsigned long)serv);
1516 mod_timer(&serv->sv_temptimer,
1517 jiffies + svc_conn_age_period * HZ);
1520 clear_bit(SK_TEMP, &svsk->sk_flags);
1521 list_add(&svsk->sk_list, &serv->sv_permsocks);
1523 spin_unlock_bh(&serv->sv_lock);
1525 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1528 clear_bit(SK_BUSY, &svsk->sk_flags);
1529 svc_sock_enqueue(svsk);
1533 int svc_addsock(struct svc_serv *serv,
1539 struct socket *so = sockfd_lookup(fd, &err);
1540 struct svc_sock *svsk = NULL;
1544 if (so->sk->sk_family != AF_INET)
1545 err = -EAFNOSUPPORT;
1546 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1547 so->sk->sk_protocol != IPPROTO_UDP)
1548 err = -EPROTONOSUPPORT;
1549 else if (so->state > SS_UNCONNECTED)
1552 svsk = svc_setup_socket(serv, so, &err, 1);
1560 if (proto) *proto = so->sk->sk_protocol;
1561 return one_sock_name(name_return, svsk);
1563 EXPORT_SYMBOL_GPL(svc_addsock);
1566 * Create socket for RPC service.
1569 svc_create_socket(struct svc_serv *serv, int protocol, struct sockaddr_in *sin)
1571 struct svc_sock *svsk;
1572 struct socket *sock;
1576 dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n",
1577 serv->sv_program->pg_name, protocol,
1578 NIPQUAD(sin->sin_addr.s_addr),
1579 ntohs(sin->sin_port));
1581 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1582 printk(KERN_WARNING "svc: only UDP and TCP "
1583 "sockets supported\n");
1586 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1588 if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0)
1591 svc_reclassify_socket(sock);
1593 if (type == SOCK_STREAM)
1594 sock->sk->sk_reuse = 1; /* allow address reuse */
1595 error = kernel_bind(sock, (struct sockaddr *) sin,
1600 if (protocol == IPPROTO_TCP) {
1601 if ((error = kernel_listen(sock, 64)) < 0)
1605 if ((svsk = svc_setup_socket(serv, sock, &error, 1)) != NULL)
1609 dprintk("svc: svc_create_socket error = %d\n", -error);
1615 * Remove a dead socket
1618 svc_delete_socket(struct svc_sock *svsk)
1620 struct svc_serv *serv;
1623 dprintk("svc: svc_delete_socket(%p)\n", svsk);
1625 serv = svsk->sk_server;
1628 sk->sk_state_change = svsk->sk_ostate;
1629 sk->sk_data_ready = svsk->sk_odata;
1630 sk->sk_write_space = svsk->sk_owspace;
1632 spin_lock_bh(&serv->sv_lock);
1634 if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags))
1635 list_del_init(&svsk->sk_list);
1637 * We used to delete the svc_sock from whichever list
1638 * it's sk_ready node was on, but we don't actually
1639 * need to. This is because the only time we're called
1640 * while still attached to a queue, the queue itself
1641 * is about to be destroyed (in svc_destroy).
1643 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags))
1644 if (test_bit(SK_TEMP, &svsk->sk_flags))
1647 /* This atomic_inc should be needed - svc_delete_socket
1648 * should have the semantic of dropping a reference.
1649 * But it doesn't yet....
1651 atomic_inc(&svsk->sk_inuse);
1652 spin_unlock_bh(&serv->sv_lock);
1657 * Make a socket for nfsd and lockd
1660 svc_makesock(struct svc_serv *serv, int protocol, unsigned short port)
1662 struct sockaddr_in sin;
1664 dprintk("svc: creating socket proto = %d\n", protocol);
1665 sin.sin_family = AF_INET;
1666 sin.sin_addr.s_addr = INADDR_ANY;
1667 sin.sin_port = htons(port);
1668 return svc_create_socket(serv, protocol, &sin);
1672 * Handle defer and revisit of requests
1675 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1677 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
1678 struct svc_sock *svsk;
1681 svc_sock_put(dr->svsk);
1685 dprintk("revisit queued\n");
1688 spin_lock_bh(&svsk->sk_defer_lock);
1689 list_add(&dr->handle.recent, &svsk->sk_deferred);
1690 spin_unlock_bh(&svsk->sk_defer_lock);
1691 set_bit(SK_DEFERRED, &svsk->sk_flags);
1692 svc_sock_enqueue(svsk);
1696 static struct cache_deferred_req *
1697 svc_defer(struct cache_req *req)
1699 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1700 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
1701 struct svc_deferred_req *dr;
1703 if (rqstp->rq_arg.page_len)
1704 return NULL; /* if more than a page, give up FIXME */
1705 if (rqstp->rq_deferred) {
1706 dr = rqstp->rq_deferred;
1707 rqstp->rq_deferred = NULL;
1709 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1710 /* FIXME maybe discard if size too large */
1711 dr = kmalloc(size, GFP_KERNEL);
1715 dr->handle.owner = rqstp->rq_server;
1716 dr->prot = rqstp->rq_prot;
1717 dr->addr = rqstp->rq_addr;
1718 dr->daddr = rqstp->rq_daddr;
1719 dr->argslen = rqstp->rq_arg.len >> 2;
1720 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
1722 atomic_inc(&rqstp->rq_sock->sk_inuse);
1723 dr->svsk = rqstp->rq_sock;
1725 dr->handle.revisit = svc_revisit;
1730 * recv data from a deferred request into an active one
1732 static int svc_deferred_recv(struct svc_rqst *rqstp)
1734 struct svc_deferred_req *dr = rqstp->rq_deferred;
1736 rqstp->rq_arg.head[0].iov_base = dr->args;
1737 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
1738 rqstp->rq_arg.page_len = 0;
1739 rqstp->rq_arg.len = dr->argslen<<2;
1740 rqstp->rq_prot = dr->prot;
1741 rqstp->rq_addr = dr->addr;
1742 rqstp->rq_daddr = dr->daddr;
1743 rqstp->rq_respages = rqstp->rq_pages;
1744 return dr->argslen<<2;
1748 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
1750 struct svc_deferred_req *dr = NULL;
1752 if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
1754 spin_lock_bh(&svsk->sk_defer_lock);
1755 clear_bit(SK_DEFERRED, &svsk->sk_flags);
1756 if (!list_empty(&svsk->sk_deferred)) {
1757 dr = list_entry(svsk->sk_deferred.next,
1758 struct svc_deferred_req,
1760 list_del_init(&dr->handle.recent);
1761 set_bit(SK_DEFERRED, &svsk->sk_flags);
1763 spin_unlock_bh(&svsk->sk_defer_lock);