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/clnt.h>
45 #include <linux/sunrpc/xdr.h>
46 #include <linux/sunrpc/svcsock.h>
47 #include <linux/sunrpc/stats.h>
49 /* SMP locking strategy:
51 * svc_pool->sp_lock protects most of the fields of that pool.
52 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
53 * when both need to be taken (rare), svc_serv->sv_lock is first.
54 * BKL protects svc_serv->sv_nrthread.
55 * svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list
56 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply.
58 * Some flags can be set to certain values at any time
59 * providing that certain rules are followed:
61 * SK_CONN, SK_DATA, can be set or cleared at any time.
62 * after a set, svc_sock_enqueue must be called.
63 * after a clear, the socket must be read/accepted
64 * if this succeeds, it must be set again.
65 * SK_CLOSE can set at any time. It is never cleared.
66 * sk_inuse contains a bias of '1' until SK_DEAD is set.
67 * so when sk_inuse hits zero, we know the socket is dead
68 * and no-one is using it.
69 * SK_DEAD can only be set while SK_BUSY is held which ensures
70 * no other thread will be using the socket or will try to
75 #define RPCDBG_FACILITY RPCDBG_SVCSOCK
78 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
79 int *errp, int flags);
80 static void svc_delete_socket(struct svc_sock *svsk);
81 static void svc_udp_data_ready(struct sock *, int);
82 static int svc_udp_recvfrom(struct svc_rqst *);
83 static int svc_udp_sendto(struct svc_rqst *);
85 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
86 static int svc_deferred_recv(struct svc_rqst *rqstp);
87 static struct cache_deferred_req *svc_defer(struct cache_req *req);
89 /* apparently the "standard" is that clients close
90 * idle connections after 5 minutes, servers after
92 * http://www.connectathon.org/talks96/nfstcp.pdf
94 static int svc_conn_age_period = 6*60;
96 #ifdef CONFIG_DEBUG_LOCK_ALLOC
97 static struct lock_class_key svc_key[2];
98 static struct lock_class_key svc_slock_key[2];
100 static inline void svc_reclassify_socket(struct socket *sock)
102 struct sock *sk = sock->sk;
103 BUG_ON(sk->sk_lock.owner != NULL);
104 switch (sk->sk_family) {
106 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
107 &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]);
111 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
112 &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]);
120 static inline void svc_reclassify_socket(struct socket *sock)
125 static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len)
127 switch (addr->sa_family) {
129 snprintf(buf, len, "%u.%u.%u.%u, port=%u",
130 NIPQUAD(((struct sockaddr_in *) addr)->sin_addr),
131 htons(((struct sockaddr_in *) addr)->sin_port));
133 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
135 snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u",
136 NIP6(((struct sockaddr_in6 *) addr)->sin6_addr),
137 htons(((struct sockaddr_in6 *) addr)->sin6_port));
141 snprintf(buf, len, "unknown address type: %d", addr->sa_family);
148 * svc_print_addr - Format rq_addr field for printing
149 * @rqstp: svc_rqst struct containing address to print
150 * @buf: target buffer for formatted address
151 * @len: length of target buffer
154 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
156 return __svc_print_addr(svc_addr(rqstp), buf, len);
158 EXPORT_SYMBOL_GPL(svc_print_addr);
161 * Queue up an idle server thread. Must have pool->sp_lock held.
162 * Note: this is really a stack rather than a queue, so that we only
163 * use as many different threads as we need, and the rest don't pollute
167 svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
169 list_add(&rqstp->rq_list, &pool->sp_threads);
173 * Dequeue an nfsd thread. Must have pool->sp_lock held.
176 svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
178 list_del(&rqstp->rq_list);
182 * Release an skbuff after use
185 svc_release_skb(struct svc_rqst *rqstp)
187 struct sk_buff *skb = rqstp->rq_skbuff;
188 struct svc_deferred_req *dr = rqstp->rq_deferred;
191 rqstp->rq_skbuff = NULL;
193 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
194 skb_free_datagram(rqstp->rq_sock->sk_sk, skb);
197 rqstp->rq_deferred = NULL;
203 * Any space to write?
205 static inline unsigned long
206 svc_sock_wspace(struct svc_sock *svsk)
210 if (svsk->sk_sock->type == SOCK_STREAM)
211 wspace = sk_stream_wspace(svsk->sk_sk);
213 wspace = sock_wspace(svsk->sk_sk);
219 * Queue up a socket with data pending. If there are idle nfsd
220 * processes, wake 'em up.
224 svc_sock_enqueue(struct svc_sock *svsk)
226 struct svc_serv *serv = svsk->sk_server;
227 struct svc_pool *pool;
228 struct svc_rqst *rqstp;
231 if (!(svsk->sk_flags &
232 ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) ))
234 if (test_bit(SK_DEAD, &svsk->sk_flags))
238 pool = svc_pool_for_cpu(svsk->sk_server, cpu);
241 spin_lock_bh(&pool->sp_lock);
243 if (!list_empty(&pool->sp_threads) &&
244 !list_empty(&pool->sp_sockets))
246 "svc_sock_enqueue: threads and sockets both waiting??\n");
248 if (test_bit(SK_DEAD, &svsk->sk_flags)) {
249 /* Don't enqueue dead sockets */
250 dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk);
254 /* Mark socket as busy. It will remain in this state until the
255 * server has processed all pending data and put the socket back
256 * on the idle list. We update SK_BUSY atomically because
257 * it also guards against trying to enqueue the svc_sock twice.
259 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) {
260 /* Don't enqueue socket while already enqueued */
261 dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk);
264 BUG_ON(svsk->sk_pool != NULL);
265 svsk->sk_pool = pool;
267 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
268 if (((atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg)*2
269 > svc_sock_wspace(svsk))
270 && !test_bit(SK_CLOSE, &svsk->sk_flags)
271 && !test_bit(SK_CONN, &svsk->sk_flags)) {
272 /* Don't enqueue while not enough space for reply */
273 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n",
274 svsk->sk_sk, atomic_read(&svsk->sk_reserved)+serv->sv_max_mesg,
275 svc_sock_wspace(svsk));
276 svsk->sk_pool = NULL;
277 clear_bit(SK_BUSY, &svsk->sk_flags);
280 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
283 if (!list_empty(&pool->sp_threads)) {
284 rqstp = list_entry(pool->sp_threads.next,
287 dprintk("svc: socket %p served by daemon %p\n",
289 svc_thread_dequeue(pool, rqstp);
292 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
293 rqstp, rqstp->rq_sock);
294 rqstp->rq_sock = svsk;
295 atomic_inc(&svsk->sk_inuse);
296 rqstp->rq_reserved = serv->sv_max_mesg;
297 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
298 BUG_ON(svsk->sk_pool != pool);
299 wake_up(&rqstp->rq_wait);
301 dprintk("svc: socket %p put into queue\n", svsk->sk_sk);
302 list_add_tail(&svsk->sk_ready, &pool->sp_sockets);
303 BUG_ON(svsk->sk_pool != pool);
307 spin_unlock_bh(&pool->sp_lock);
311 * Dequeue the first socket. Must be called with the pool->sp_lock held.
313 static inline struct svc_sock *
314 svc_sock_dequeue(struct svc_pool *pool)
316 struct svc_sock *svsk;
318 if (list_empty(&pool->sp_sockets))
321 svsk = list_entry(pool->sp_sockets.next,
322 struct svc_sock, sk_ready);
323 list_del_init(&svsk->sk_ready);
325 dprintk("svc: socket %p dequeued, inuse=%d\n",
326 svsk->sk_sk, atomic_read(&svsk->sk_inuse));
332 * Having read something from a socket, check whether it
333 * needs to be re-enqueued.
334 * Note: SK_DATA only gets cleared when a read-attempt finds
335 * no (or insufficient) data.
338 svc_sock_received(struct svc_sock *svsk)
340 svsk->sk_pool = NULL;
341 clear_bit(SK_BUSY, &svsk->sk_flags);
342 svc_sock_enqueue(svsk);
347 * svc_reserve - change the space reserved for the reply to a request.
348 * @rqstp: The request in question
349 * @space: new max space to reserve
351 * Each request reserves some space on the output queue of the socket
352 * to make sure the reply fits. This function reduces that reserved
353 * space to be the amount of space used already, plus @space.
356 void svc_reserve(struct svc_rqst *rqstp, int space)
358 space += rqstp->rq_res.head[0].iov_len;
360 if (space < rqstp->rq_reserved) {
361 struct svc_sock *svsk = rqstp->rq_sock;
362 atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved);
363 rqstp->rq_reserved = space;
365 svc_sock_enqueue(svsk);
370 * Release a socket after use.
373 svc_sock_put(struct svc_sock *svsk)
375 if (atomic_dec_and_test(&svsk->sk_inuse)) {
376 BUG_ON(! test_bit(SK_DEAD, &svsk->sk_flags));
378 dprintk("svc: releasing dead socket\n");
379 if (svsk->sk_sock->file)
380 sockfd_put(svsk->sk_sock);
382 sock_release(svsk->sk_sock);
383 if (svsk->sk_info_authunix != NULL)
384 svcauth_unix_info_release(svsk->sk_info_authunix);
390 svc_sock_release(struct svc_rqst *rqstp)
392 struct svc_sock *svsk = rqstp->rq_sock;
394 svc_release_skb(rqstp);
396 svc_free_res_pages(rqstp);
397 rqstp->rq_res.page_len = 0;
398 rqstp->rq_res.page_base = 0;
401 /* Reset response buffer and release
403 * But first, check that enough space was reserved
404 * for the reply, otherwise we have a bug!
406 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
407 printk(KERN_ERR "RPC request reserved %d but used %d\n",
411 rqstp->rq_res.head[0].iov_len = 0;
412 svc_reserve(rqstp, 0);
413 rqstp->rq_sock = NULL;
419 * External function to wake up a server waiting for data
420 * This really only makes sense for services like lockd
421 * which have exactly one thread anyway.
424 svc_wake_up(struct svc_serv *serv)
426 struct svc_rqst *rqstp;
428 struct svc_pool *pool;
430 for (i = 0; i < serv->sv_nrpools; i++) {
431 pool = &serv->sv_pools[i];
433 spin_lock_bh(&pool->sp_lock);
434 if (!list_empty(&pool->sp_threads)) {
435 rqstp = list_entry(pool->sp_threads.next,
438 dprintk("svc: daemon %p woken up.\n", rqstp);
440 svc_thread_dequeue(pool, rqstp);
441 rqstp->rq_sock = NULL;
443 wake_up(&rqstp->rq_wait);
445 spin_unlock_bh(&pool->sp_lock);
450 * Generic sendto routine
453 svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
455 struct svc_sock *svsk = rqstp->rq_sock;
456 struct socket *sock = svsk->sk_sock;
458 char buffer[CMSG_SPACE(sizeof(struct in_pktinfo))];
459 struct cmsghdr *cmh = (struct cmsghdr *)buffer;
460 struct in_pktinfo *pki = (struct in_pktinfo *)CMSG_DATA(cmh);
464 struct page **ppage = xdr->pages;
465 size_t base = xdr->page_base;
466 unsigned int pglen = xdr->page_len;
467 unsigned int flags = MSG_MORE;
468 char buf[RPC_MAX_ADDRBUFLEN];
472 if (rqstp->rq_prot == IPPROTO_UDP) {
473 /* set the source and destination */
475 msg.msg_name = &rqstp->rq_addr;
476 msg.msg_namelen = rqstp->rq_addrlen;
479 msg.msg_flags = MSG_MORE;
481 msg.msg_control = cmh;
482 msg.msg_controllen = sizeof(buffer);
483 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
484 cmh->cmsg_level = SOL_IP;
485 cmh->cmsg_type = IP_PKTINFO;
486 pki->ipi_ifindex = 0;
487 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
489 if (sock_sendmsg(sock, &msg, 0) < 0)
494 if (slen == xdr->head[0].iov_len)
496 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
497 xdr->head[0].iov_len, flags);
498 if (len != xdr->head[0].iov_len)
500 slen -= xdr->head[0].iov_len;
505 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
509 result = kernel_sendpage(sock, *ppage, base, size, flags);
516 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
521 if (xdr->tail[0].iov_len) {
522 result = kernel_sendpage(sock, rqstp->rq_respages[0],
523 ((unsigned long)xdr->tail[0].iov_base)
525 xdr->tail[0].iov_len, 0);
531 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
532 rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len,
533 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
539 * Report socket names for nfsdfs
541 static int one_sock_name(char *buf, struct svc_sock *svsk)
545 switch(svsk->sk_sk->sk_family) {
547 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
548 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
550 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
551 inet_sk(svsk->sk_sk)->num);
554 len = sprintf(buf, "*unknown-%d*\n",
555 svsk->sk_sk->sk_family);
561 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
563 struct svc_sock *svsk, *closesk = NULL;
568 spin_lock_bh(&serv->sv_lock);
569 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) {
570 int onelen = one_sock_name(buf+len, svsk);
571 if (toclose && strcmp(toclose, buf+len) == 0)
576 spin_unlock_bh(&serv->sv_lock);
578 /* Should unregister with portmap, but you cannot
579 * unregister just one protocol...
581 svc_close_socket(closesk);
586 EXPORT_SYMBOL(svc_sock_names);
589 * Check input queue length
592 svc_recv_available(struct svc_sock *svsk)
594 struct socket *sock = svsk->sk_sock;
597 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
599 return (err >= 0)? avail : err;
603 * Generic recvfrom routine.
606 svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen)
608 struct svc_sock *svsk = rqstp->rq_sock;
609 struct msghdr msg = {
610 .msg_flags = MSG_DONTWAIT,
614 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
617 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
619 memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen);
620 rqstp->rq_addrlen = svsk->sk_remotelen;
622 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
623 svsk, iov[0].iov_base, iov[0].iov_len, len);
629 * Set socket snd and rcv buffer lengths
632 svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
636 oldfs = get_fs(); set_fs(KERNEL_DS);
637 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
638 (char*)&snd, sizeof(snd));
639 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
640 (char*)&rcv, sizeof(rcv));
642 /* sock_setsockopt limits use to sysctl_?mem_max,
643 * which isn't acceptable. Until that is made conditional
644 * on not having CAP_SYS_RESOURCE or similar, we go direct...
645 * DaveM said I could!
648 sock->sk->sk_sndbuf = snd * 2;
649 sock->sk->sk_rcvbuf = rcv * 2;
650 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
651 release_sock(sock->sk);
655 * INET callback when data has been received on the socket.
658 svc_udp_data_ready(struct sock *sk, int count)
660 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
663 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
664 svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags));
665 set_bit(SK_DATA, &svsk->sk_flags);
666 svc_sock_enqueue(svsk);
668 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
669 wake_up_interruptible(sk->sk_sleep);
673 * INET callback when space is newly available on the socket.
676 svc_write_space(struct sock *sk)
678 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
681 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
682 svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags));
683 svc_sock_enqueue(svsk);
686 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
687 dprintk("RPC svc_write_space: someone sleeping on %p\n",
689 wake_up_interruptible(sk->sk_sleep);
694 * Receive a datagram from a UDP socket.
697 svc_udp_recvfrom(struct svc_rqst *rqstp)
699 struct sockaddr_in *sin = svc_addr_in(rqstp);
700 struct svc_sock *svsk = rqstp->rq_sock;
701 struct svc_serv *serv = svsk->sk_server;
705 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
706 /* udp sockets need large rcvbuf as all pending
707 * requests are still in that buffer. sndbuf must
708 * also be large enough that there is enough space
709 * for one reply per thread. We count all threads
710 * rather than threads in a particular pool, which
711 * provides an upper bound on the number of threads
712 * which will access the socket.
714 svc_sock_setbufsize(svsk->sk_sock,
715 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
716 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
718 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
719 svc_sock_received(svsk);
720 return svc_deferred_recv(rqstp);
723 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
724 svc_delete_socket(svsk);
728 clear_bit(SK_DATA, &svsk->sk_flags);
729 while ((skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err)) == NULL) {
730 if (err == -EAGAIN) {
731 svc_sock_received(svsk);
734 /* possibly an icmp error */
735 dprintk("svc: recvfrom returned error %d\n", -err);
737 if (skb->tstamp.off_sec == 0) {
740 tv.tv_sec = xtime.tv_sec;
741 tv.tv_usec = xtime.tv_nsec / NSEC_PER_USEC;
742 skb_set_timestamp(skb, &tv);
743 /* Don't enable netstamp, sunrpc doesn't
744 need that much accuracy */
746 skb_get_timestamp(skb, &svsk->sk_sk->sk_stamp);
747 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
750 * Maybe more packets - kick another thread ASAP.
752 svc_sock_received(svsk);
754 len = skb->len - sizeof(struct udphdr);
755 rqstp->rq_arg.len = len;
757 rqstp->rq_prot = IPPROTO_UDP;
759 /* Get sender address */
760 sin->sin_family = AF_INET;
761 sin->sin_port = skb->h.uh->source;
762 sin->sin_addr.s_addr = skb->nh.iph->saddr;
763 rqstp->rq_addrlen = sizeof(struct sockaddr_in);
765 /* Remember which interface received this request */
766 rqstp->rq_daddr.addr.s_addr = skb->nh.iph->daddr;
768 if (skb_is_nonlinear(skb)) {
769 /* we have to copy */
771 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
774 skb_free_datagram(svsk->sk_sk, skb);
778 skb_free_datagram(svsk->sk_sk, skb);
780 /* we can use it in-place */
781 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
782 rqstp->rq_arg.head[0].iov_len = len;
783 if (skb_checksum_complete(skb)) {
784 skb_free_datagram(svsk->sk_sk, skb);
787 rqstp->rq_skbuff = skb;
790 rqstp->rq_arg.page_base = 0;
791 if (len <= rqstp->rq_arg.head[0].iov_len) {
792 rqstp->rq_arg.head[0].iov_len = len;
793 rqstp->rq_arg.page_len = 0;
794 rqstp->rq_respages = rqstp->rq_pages+1;
796 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
797 rqstp->rq_respages = rqstp->rq_pages + 1 +
798 (rqstp->rq_arg.page_len + PAGE_SIZE - 1)/ PAGE_SIZE;
802 serv->sv_stats->netudpcnt++;
808 svc_udp_sendto(struct svc_rqst *rqstp)
812 error = svc_sendto(rqstp, &rqstp->rq_res);
813 if (error == -ECONNREFUSED)
814 /* ICMP error on earlier request. */
815 error = svc_sendto(rqstp, &rqstp->rq_res);
821 svc_udp_init(struct svc_sock *svsk)
823 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
824 svsk->sk_sk->sk_write_space = svc_write_space;
825 svsk->sk_recvfrom = svc_udp_recvfrom;
826 svsk->sk_sendto = svc_udp_sendto;
828 /* initialise setting must have enough space to
829 * receive and respond to one request.
830 * svc_udp_recvfrom will re-adjust if necessary
832 svc_sock_setbufsize(svsk->sk_sock,
833 3 * svsk->sk_server->sv_max_mesg,
834 3 * svsk->sk_server->sv_max_mesg);
836 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
837 set_bit(SK_CHNGBUF, &svsk->sk_flags);
841 * A data_ready event on a listening socket means there's a connection
842 * pending. Do not use state_change as a substitute for it.
845 svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
847 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
849 dprintk("svc: socket %p TCP (listen) state change %d\n",
853 * This callback may called twice when a new connection
854 * is established as a child socket inherits everything
855 * from a parent LISTEN socket.
856 * 1) data_ready method of the parent socket will be called
857 * when one of child sockets become ESTABLISHED.
858 * 2) data_ready method of the child socket may be called
859 * when it receives data before the socket is accepted.
860 * In case of 2, we should ignore it silently.
862 if (sk->sk_state == TCP_LISTEN) {
864 set_bit(SK_CONN, &svsk->sk_flags);
865 svc_sock_enqueue(svsk);
867 printk("svc: socket %p: no user data\n", sk);
870 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
871 wake_up_interruptible_all(sk->sk_sleep);
875 * A state change on a connected socket means it's dying or dead.
878 svc_tcp_state_change(struct sock *sk)
880 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
882 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
883 sk, sk->sk_state, sk->sk_user_data);
886 printk("svc: socket %p: no user data\n", sk);
888 set_bit(SK_CLOSE, &svsk->sk_flags);
889 svc_sock_enqueue(svsk);
891 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
892 wake_up_interruptible_all(sk->sk_sleep);
896 svc_tcp_data_ready(struct sock *sk, int count)
898 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
900 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
901 sk, sk->sk_user_data);
903 set_bit(SK_DATA, &svsk->sk_flags);
904 svc_sock_enqueue(svsk);
906 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
907 wake_up_interruptible(sk->sk_sleep);
911 * Accept a TCP connection
914 svc_tcp_accept(struct svc_sock *svsk)
916 struct sockaddr_in sin;
917 struct svc_serv *serv = svsk->sk_server;
918 struct socket *sock = svsk->sk_sock;
919 struct socket *newsock;
920 struct svc_sock *newsvsk;
922 char buf[RPC_MAX_ADDRBUFLEN];
924 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
928 clear_bit(SK_CONN, &svsk->sk_flags);
929 err = kernel_accept(sock, &newsock, O_NONBLOCK);
932 printk(KERN_WARNING "%s: no more sockets!\n",
934 else if (err != -EAGAIN && net_ratelimit())
935 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
936 serv->sv_name, -err);
940 set_bit(SK_CONN, &svsk->sk_flags);
941 svc_sock_enqueue(svsk);
944 err = kernel_getpeername(newsock, (struct sockaddr *) &sin, &slen);
947 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
948 serv->sv_name, -err);
949 goto failed; /* aborted connection or whatever */
952 /* Ideally, we would want to reject connections from unauthorized
953 * hosts here, but when we get encryption, the IP of the host won't
954 * tell us anything. For now just warn about unpriv connections.
956 if (ntohs(sin.sin_port) >= 1024) {
958 "%s: connect from unprivileged port: %s\n",
960 __svc_print_addr((struct sockaddr *) &sin, buf,
963 dprintk("%s: connect from %s\n", serv->sv_name,
964 __svc_print_addr((struct sockaddr *) &sin, buf,
967 /* make sure that a write doesn't block forever when
970 newsock->sk->sk_sndtimeo = HZ*30;
972 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
973 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
975 memcpy(&newsvsk->sk_remote, &sin, slen);
976 newsvsk->sk_remotelen = slen;
978 svc_sock_received(newsvsk);
980 /* make sure that we don't have too many active connections.
981 * If we have, something must be dropped.
983 * There's no point in trying to do random drop here for
984 * DoS prevention. The NFS clients does 1 reconnect in 15
985 * seconds. An attacker can easily beat that.
987 * The only somewhat efficient mechanism would be if drop
988 * old connections from the same IP first. But right now
989 * we don't even record the client IP in svc_sock.
991 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
992 struct svc_sock *svsk = NULL;
993 spin_lock_bh(&serv->sv_lock);
994 if (!list_empty(&serv->sv_tempsocks)) {
995 if (net_ratelimit()) {
996 /* Try to help the admin */
997 printk(KERN_NOTICE "%s: too many open TCP "
998 "sockets, consider increasing the "
999 "number of nfsd threads\n",
1002 "%s: last TCP connect from %s\n",
1003 serv->sv_name, buf);
1006 * Always select the oldest socket. It's not fair,
1009 svsk = list_entry(serv->sv_tempsocks.prev,
1012 set_bit(SK_CLOSE, &svsk->sk_flags);
1013 atomic_inc(&svsk->sk_inuse);
1015 spin_unlock_bh(&serv->sv_lock);
1018 svc_sock_enqueue(svsk);
1025 serv->sv_stats->nettcpconn++;
1030 sock_release(newsock);
1035 * Receive data from a TCP socket.
1038 svc_tcp_recvfrom(struct svc_rqst *rqstp)
1040 struct svc_sock *svsk = rqstp->rq_sock;
1041 struct svc_serv *serv = svsk->sk_server;
1046 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1047 svsk, test_bit(SK_DATA, &svsk->sk_flags),
1048 test_bit(SK_CONN, &svsk->sk_flags),
1049 test_bit(SK_CLOSE, &svsk->sk_flags));
1051 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
1052 svc_sock_received(svsk);
1053 return svc_deferred_recv(rqstp);
1056 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
1057 svc_delete_socket(svsk);
1061 if (svsk->sk_sk->sk_state == TCP_LISTEN) {
1062 svc_tcp_accept(svsk);
1063 svc_sock_received(svsk);
1067 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
1068 /* sndbuf needs to have room for one request
1069 * per thread, otherwise we can stall even when the
1070 * network isn't a bottleneck.
1072 * We count all threads rather than threads in a
1073 * particular pool, which provides an upper bound
1074 * on the number of threads which will access the socket.
1076 * rcvbuf just needs to be able to hold a few requests.
1077 * Normally they will be removed from the queue
1078 * as soon a a complete request arrives.
1080 svc_sock_setbufsize(svsk->sk_sock,
1081 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
1082 3 * serv->sv_max_mesg);
1084 clear_bit(SK_DATA, &svsk->sk_flags);
1086 /* Receive data. If we haven't got the record length yet, get
1087 * the next four bytes. Otherwise try to gobble up as much as
1088 * possible up to the complete record length.
1090 if (svsk->sk_tcplen < 4) {
1091 unsigned long want = 4 - svsk->sk_tcplen;
1094 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1096 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1098 svsk->sk_tcplen += len;
1101 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
1103 svc_sock_received(svsk);
1104 return -EAGAIN; /* record header not complete */
1107 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1108 if (!(svsk->sk_reclen & 0x80000000)) {
1109 /* FIXME: technically, a record can be fragmented,
1110 * and non-terminal fragments will not have the top
1111 * bit set in the fragment length header.
1112 * But apparently no known nfs clients send fragmented
1114 if (net_ratelimit())
1115 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
1116 " (non-terminal)\n",
1117 (unsigned long) svsk->sk_reclen);
1120 svsk->sk_reclen &= 0x7fffffff;
1121 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1122 if (svsk->sk_reclen > serv->sv_max_mesg) {
1123 if (net_ratelimit())
1124 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
1126 (unsigned long) svsk->sk_reclen);
1131 /* Check whether enough data is available */
1132 len = svc_recv_available(svsk);
1136 if (len < svsk->sk_reclen) {
1137 dprintk("svc: incomplete TCP record (%d of %d)\n",
1138 len, svsk->sk_reclen);
1139 svc_sock_received(svsk);
1140 return -EAGAIN; /* record not complete */
1142 len = svsk->sk_reclen;
1143 set_bit(SK_DATA, &svsk->sk_flags);
1145 vec = rqstp->rq_vec;
1146 vec[0] = rqstp->rq_arg.head[0];
1149 while (vlen < len) {
1150 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
1151 vec[pnum].iov_len = PAGE_SIZE;
1155 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1157 /* Now receive data */
1158 len = svc_recvfrom(rqstp, vec, pnum, len);
1162 dprintk("svc: TCP complete record (%d bytes)\n", len);
1163 rqstp->rq_arg.len = len;
1164 rqstp->rq_arg.page_base = 0;
1165 if (len <= rqstp->rq_arg.head[0].iov_len) {
1166 rqstp->rq_arg.head[0].iov_len = len;
1167 rqstp->rq_arg.page_len = 0;
1169 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1172 rqstp->rq_skbuff = NULL;
1173 rqstp->rq_prot = IPPROTO_TCP;
1175 /* Reset TCP read info */
1176 svsk->sk_reclen = 0;
1177 svsk->sk_tcplen = 0;
1179 svc_sock_received(svsk);
1181 serv->sv_stats->nettcpcnt++;
1186 svc_delete_socket(svsk);
1190 if (len == -EAGAIN) {
1191 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1192 svc_sock_received(svsk);
1194 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1195 svsk->sk_server->sv_name, -len);
1203 * Send out data on TCP socket.
1206 svc_tcp_sendto(struct svc_rqst *rqstp)
1208 struct xdr_buf *xbufp = &rqstp->rq_res;
1212 /* Set up the first element of the reply kvec.
1213 * Any other kvecs that may be in use have been taken
1214 * care of by the server implementation itself.
1216 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1217 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1219 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1222 sent = svc_sendto(rqstp, &rqstp->rq_res);
1223 if (sent != xbufp->len) {
1224 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1225 rqstp->rq_sock->sk_server->sv_name,
1226 (sent<0)?"got error":"sent only",
1228 set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags);
1229 svc_sock_enqueue(rqstp->rq_sock);
1236 svc_tcp_init(struct svc_sock *svsk)
1238 struct sock *sk = svsk->sk_sk;
1239 struct tcp_sock *tp = tcp_sk(sk);
1241 svsk->sk_recvfrom = svc_tcp_recvfrom;
1242 svsk->sk_sendto = svc_tcp_sendto;
1244 if (sk->sk_state == TCP_LISTEN) {
1245 dprintk("setting up TCP socket for listening\n");
1246 sk->sk_data_ready = svc_tcp_listen_data_ready;
1247 set_bit(SK_CONN, &svsk->sk_flags);
1249 dprintk("setting up TCP socket for reading\n");
1250 sk->sk_state_change = svc_tcp_state_change;
1251 sk->sk_data_ready = svc_tcp_data_ready;
1252 sk->sk_write_space = svc_write_space;
1254 svsk->sk_reclen = 0;
1255 svsk->sk_tcplen = 0;
1257 tp->nonagle = 1; /* disable Nagle's algorithm */
1259 /* initialise setting must have enough space to
1260 * receive and respond to one request.
1261 * svc_tcp_recvfrom will re-adjust if necessary
1263 svc_sock_setbufsize(svsk->sk_sock,
1264 3 * svsk->sk_server->sv_max_mesg,
1265 3 * svsk->sk_server->sv_max_mesg);
1267 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1268 set_bit(SK_DATA, &svsk->sk_flags);
1269 if (sk->sk_state != TCP_ESTABLISHED)
1270 set_bit(SK_CLOSE, &svsk->sk_flags);
1275 svc_sock_update_bufs(struct svc_serv *serv)
1278 * The number of server threads has changed. Update
1279 * rcvbuf and sndbuf accordingly on all sockets
1281 struct list_head *le;
1283 spin_lock_bh(&serv->sv_lock);
1284 list_for_each(le, &serv->sv_permsocks) {
1285 struct svc_sock *svsk =
1286 list_entry(le, struct svc_sock, sk_list);
1287 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1289 list_for_each(le, &serv->sv_tempsocks) {
1290 struct svc_sock *svsk =
1291 list_entry(le, struct svc_sock, sk_list);
1292 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1294 spin_unlock_bh(&serv->sv_lock);
1298 * Receive the next request on any socket. This code is carefully
1299 * organised not to touch any cachelines in the shared svc_serv
1300 * structure, only cachelines in the local svc_pool.
1303 svc_recv(struct svc_rqst *rqstp, long timeout)
1305 struct svc_sock *svsk = NULL;
1306 struct sockaddr_in *sin = svc_addr_in(rqstp);
1307 struct svc_serv *serv = rqstp->rq_server;
1308 struct svc_pool *pool = rqstp->rq_pool;
1311 struct xdr_buf *arg;
1312 DECLARE_WAITQUEUE(wait, current);
1314 dprintk("svc: server %p waiting for data (to = %ld)\n",
1319 "svc_recv: service %p, socket not NULL!\n",
1321 if (waitqueue_active(&rqstp->rq_wait))
1323 "svc_recv: service %p, wait queue active!\n",
1327 /* now allocate needed pages. If we get a failure, sleep briefly */
1328 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
1329 for (i=0; i < pages ; i++)
1330 while (rqstp->rq_pages[i] == NULL) {
1331 struct page *p = alloc_page(GFP_KERNEL);
1333 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1334 rqstp->rq_pages[i] = p;
1336 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
1337 BUG_ON(pages >= RPCSVC_MAXPAGES);
1339 /* Make arg->head point to first page and arg->pages point to rest */
1340 arg = &rqstp->rq_arg;
1341 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
1342 arg->head[0].iov_len = PAGE_SIZE;
1343 arg->pages = rqstp->rq_pages + 1;
1345 /* save at least one page for response */
1346 arg->page_len = (pages-2)*PAGE_SIZE;
1347 arg->len = (pages-1)*PAGE_SIZE;
1348 arg->tail[0].iov_len = 0;
1355 spin_lock_bh(&pool->sp_lock);
1356 if ((svsk = svc_sock_dequeue(pool)) != NULL) {
1357 rqstp->rq_sock = svsk;
1358 atomic_inc(&svsk->sk_inuse);
1359 rqstp->rq_reserved = serv->sv_max_mesg;
1360 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
1362 /* No data pending. Go to sleep */
1363 svc_thread_enqueue(pool, rqstp);
1366 * We have to be able to interrupt this wait
1367 * to bring down the daemons ...
1369 set_current_state(TASK_INTERRUPTIBLE);
1370 add_wait_queue(&rqstp->rq_wait, &wait);
1371 spin_unlock_bh(&pool->sp_lock);
1373 schedule_timeout(timeout);
1377 spin_lock_bh(&pool->sp_lock);
1378 remove_wait_queue(&rqstp->rq_wait, &wait);
1380 if (!(svsk = rqstp->rq_sock)) {
1381 svc_thread_dequeue(pool, rqstp);
1382 spin_unlock_bh(&pool->sp_lock);
1383 dprintk("svc: server %p, no data yet\n", rqstp);
1384 return signalled()? -EINTR : -EAGAIN;
1387 spin_unlock_bh(&pool->sp_lock);
1389 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
1390 rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse));
1391 len = svsk->sk_recvfrom(rqstp);
1392 dprintk("svc: got len=%d\n", len);
1394 /* No data, incomplete (TCP) read, or accept() */
1395 if (len == 0 || len == -EAGAIN) {
1396 rqstp->rq_res.len = 0;
1397 svc_sock_release(rqstp);
1400 svsk->sk_lastrecv = get_seconds();
1401 clear_bit(SK_OLD, &svsk->sk_flags);
1403 rqstp->rq_secure = ntohs(sin->sin_port) < PROT_SOCK;
1404 rqstp->rq_chandle.defer = svc_defer;
1407 serv->sv_stats->netcnt++;
1415 svc_drop(struct svc_rqst *rqstp)
1417 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1418 svc_sock_release(rqstp);
1422 * Return reply to client.
1425 svc_send(struct svc_rqst *rqstp)
1427 struct svc_sock *svsk;
1431 if ((svsk = rqstp->rq_sock) == NULL) {
1432 printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1433 __FILE__, __LINE__);
1437 /* release the receive skb before sending the reply */
1438 svc_release_skb(rqstp);
1440 /* calculate over-all length */
1441 xb = & rqstp->rq_res;
1442 xb->len = xb->head[0].iov_len +
1444 xb->tail[0].iov_len;
1446 /* Grab svsk->sk_mutex to serialize outgoing data. */
1447 mutex_lock(&svsk->sk_mutex);
1448 if (test_bit(SK_DEAD, &svsk->sk_flags))
1451 len = svsk->sk_sendto(rqstp);
1452 mutex_unlock(&svsk->sk_mutex);
1453 svc_sock_release(rqstp);
1455 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1461 * Timer function to close old temporary sockets, using
1462 * a mark-and-sweep algorithm.
1465 svc_age_temp_sockets(unsigned long closure)
1467 struct svc_serv *serv = (struct svc_serv *)closure;
1468 struct svc_sock *svsk;
1469 struct list_head *le, *next;
1470 LIST_HEAD(to_be_aged);
1472 dprintk("svc_age_temp_sockets\n");
1474 if (!spin_trylock_bh(&serv->sv_lock)) {
1475 /* busy, try again 1 sec later */
1476 dprintk("svc_age_temp_sockets: busy\n");
1477 mod_timer(&serv->sv_temptimer, jiffies + HZ);
1481 list_for_each_safe(le, next, &serv->sv_tempsocks) {
1482 svsk = list_entry(le, struct svc_sock, sk_list);
1484 if (!test_and_set_bit(SK_OLD, &svsk->sk_flags))
1486 if (atomic_read(&svsk->sk_inuse) || test_bit(SK_BUSY, &svsk->sk_flags))
1488 atomic_inc(&svsk->sk_inuse);
1489 list_move(le, &to_be_aged);
1490 set_bit(SK_CLOSE, &svsk->sk_flags);
1491 set_bit(SK_DETACHED, &svsk->sk_flags);
1493 spin_unlock_bh(&serv->sv_lock);
1495 while (!list_empty(&to_be_aged)) {
1496 le = to_be_aged.next;
1497 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */
1499 svsk = list_entry(le, struct svc_sock, sk_list);
1501 dprintk("queuing svsk %p for closing, %lu seconds old\n",
1502 svsk, get_seconds() - svsk->sk_lastrecv);
1504 /* a thread will dequeue and close it soon */
1505 svc_sock_enqueue(svsk);
1509 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
1513 * Initialize socket for RPC use and create svc_sock struct
1514 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1516 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1517 struct socket *sock,
1518 int *errp, int flags)
1520 struct svc_sock *svsk;
1522 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1523 int is_temporary = flags & SVC_SOCK_TEMPORARY;
1525 dprintk("svc: svc_setup_socket %p\n", sock);
1526 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1533 /* Register socket with portmapper */
1534 if (*errp >= 0 && pmap_register)
1535 *errp = svc_register(serv, inet->sk_protocol,
1536 ntohs(inet_sk(inet)->sport));
1543 set_bit(SK_BUSY, &svsk->sk_flags);
1544 inet->sk_user_data = svsk;
1545 svsk->sk_sock = sock;
1547 svsk->sk_ostate = inet->sk_state_change;
1548 svsk->sk_odata = inet->sk_data_ready;
1549 svsk->sk_owspace = inet->sk_write_space;
1550 svsk->sk_server = serv;
1551 atomic_set(&svsk->sk_inuse, 1);
1552 svsk->sk_lastrecv = get_seconds();
1553 spin_lock_init(&svsk->sk_defer_lock);
1554 INIT_LIST_HEAD(&svsk->sk_deferred);
1555 INIT_LIST_HEAD(&svsk->sk_ready);
1556 mutex_init(&svsk->sk_mutex);
1558 /* Initialize the socket */
1559 if (sock->type == SOCK_DGRAM)
1564 spin_lock_bh(&serv->sv_lock);
1566 set_bit(SK_TEMP, &svsk->sk_flags);
1567 list_add(&svsk->sk_list, &serv->sv_tempsocks);
1569 if (serv->sv_temptimer.function == NULL) {
1570 /* setup timer to age temp sockets */
1571 setup_timer(&serv->sv_temptimer, svc_age_temp_sockets,
1572 (unsigned long)serv);
1573 mod_timer(&serv->sv_temptimer,
1574 jiffies + svc_conn_age_period * HZ);
1577 clear_bit(SK_TEMP, &svsk->sk_flags);
1578 list_add(&svsk->sk_list, &serv->sv_permsocks);
1580 spin_unlock_bh(&serv->sv_lock);
1582 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1588 int svc_addsock(struct svc_serv *serv,
1594 struct socket *so = sockfd_lookup(fd, &err);
1595 struct svc_sock *svsk = NULL;
1599 if (so->sk->sk_family != AF_INET)
1600 err = -EAFNOSUPPORT;
1601 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1602 so->sk->sk_protocol != IPPROTO_UDP)
1603 err = -EPROTONOSUPPORT;
1604 else if (so->state > SS_UNCONNECTED)
1607 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
1609 svc_sock_received(svsk);
1617 if (proto) *proto = so->sk->sk_protocol;
1618 return one_sock_name(name_return, svsk);
1620 EXPORT_SYMBOL_GPL(svc_addsock);
1623 * Create socket for RPC service.
1625 static int svc_create_socket(struct svc_serv *serv, int protocol,
1626 struct sockaddr_in *sin, int flags)
1628 struct svc_sock *svsk;
1629 struct socket *sock;
1632 char buf[RPC_MAX_ADDRBUFLEN];
1634 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1635 serv->sv_program->pg_name, protocol,
1636 __svc_print_addr((struct sockaddr *) sin, buf,
1639 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1640 printk(KERN_WARNING "svc: only UDP and TCP "
1641 "sockets supported\n");
1644 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1646 if ((error = sock_create_kern(PF_INET, type, protocol, &sock)) < 0)
1649 svc_reclassify_socket(sock);
1651 if (type == SOCK_STREAM)
1652 sock->sk->sk_reuse = 1; /* allow address reuse */
1653 error = kernel_bind(sock, (struct sockaddr *) sin,
1658 if (protocol == IPPROTO_TCP) {
1659 if ((error = kernel_listen(sock, 64)) < 0)
1663 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1664 svc_sock_received(svsk);
1665 return ntohs(inet_sk(svsk->sk_sk)->sport);
1669 dprintk("svc: svc_create_socket error = %d\n", -error);
1675 * Remove a dead socket
1678 svc_delete_socket(struct svc_sock *svsk)
1680 struct svc_serv *serv;
1683 dprintk("svc: svc_delete_socket(%p)\n", svsk);
1685 serv = svsk->sk_server;
1688 sk->sk_state_change = svsk->sk_ostate;
1689 sk->sk_data_ready = svsk->sk_odata;
1690 sk->sk_write_space = svsk->sk_owspace;
1692 spin_lock_bh(&serv->sv_lock);
1694 if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags))
1695 list_del_init(&svsk->sk_list);
1697 * We used to delete the svc_sock from whichever list
1698 * it's sk_ready node was on, but we don't actually
1699 * need to. This is because the only time we're called
1700 * while still attached to a queue, the queue itself
1701 * is about to be destroyed (in svc_destroy).
1703 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) {
1704 BUG_ON(atomic_read(&svsk->sk_inuse)<2);
1705 atomic_dec(&svsk->sk_inuse);
1706 if (test_bit(SK_TEMP, &svsk->sk_flags))
1710 spin_unlock_bh(&serv->sv_lock);
1713 void svc_close_socket(struct svc_sock *svsk)
1715 set_bit(SK_CLOSE, &svsk->sk_flags);
1716 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags))
1717 /* someone else will have to effect the close */
1720 atomic_inc(&svsk->sk_inuse);
1721 svc_delete_socket(svsk);
1722 clear_bit(SK_BUSY, &svsk->sk_flags);
1727 * svc_makesock - Make a socket for nfsd and lockd
1728 * @serv: RPC server structure
1729 * @protocol: transport protocol to use
1730 * @port: port to use
1731 * @flags: requested socket characteristics
1734 int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port,
1737 struct sockaddr_in sin = {
1738 .sin_family = AF_INET,
1739 .sin_addr.s_addr = INADDR_ANY,
1740 .sin_port = htons(port),
1743 dprintk("svc: creating socket proto = %d\n", protocol);
1744 return svc_create_socket(serv, protocol, &sin, flags);
1748 * Handle defer and revisit of requests
1751 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1753 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
1754 struct svc_sock *svsk;
1757 svc_sock_put(dr->svsk);
1761 dprintk("revisit queued\n");
1764 spin_lock_bh(&svsk->sk_defer_lock);
1765 list_add(&dr->handle.recent, &svsk->sk_deferred);
1766 spin_unlock_bh(&svsk->sk_defer_lock);
1767 set_bit(SK_DEFERRED, &svsk->sk_flags);
1768 svc_sock_enqueue(svsk);
1772 static struct cache_deferred_req *
1773 svc_defer(struct cache_req *req)
1775 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
1776 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
1777 struct svc_deferred_req *dr;
1779 if (rqstp->rq_arg.page_len)
1780 return NULL; /* if more than a page, give up FIXME */
1781 if (rqstp->rq_deferred) {
1782 dr = rqstp->rq_deferred;
1783 rqstp->rq_deferred = NULL;
1785 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1786 /* FIXME maybe discard if size too large */
1787 dr = kmalloc(size, GFP_KERNEL);
1791 dr->handle.owner = rqstp->rq_server;
1792 dr->prot = rqstp->rq_prot;
1793 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
1794 dr->addrlen = rqstp->rq_addrlen;
1795 dr->daddr = rqstp->rq_daddr;
1796 dr->argslen = rqstp->rq_arg.len >> 2;
1797 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
1799 atomic_inc(&rqstp->rq_sock->sk_inuse);
1800 dr->svsk = rqstp->rq_sock;
1802 dr->handle.revisit = svc_revisit;
1807 * recv data from a deferred request into an active one
1809 static int svc_deferred_recv(struct svc_rqst *rqstp)
1811 struct svc_deferred_req *dr = rqstp->rq_deferred;
1813 rqstp->rq_arg.head[0].iov_base = dr->args;
1814 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
1815 rqstp->rq_arg.page_len = 0;
1816 rqstp->rq_arg.len = dr->argslen<<2;
1817 rqstp->rq_prot = dr->prot;
1818 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
1819 rqstp->rq_addrlen = dr->addrlen;
1820 rqstp->rq_daddr = dr->daddr;
1821 rqstp->rq_respages = rqstp->rq_pages;
1822 return dr->argslen<<2;
1826 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
1828 struct svc_deferred_req *dr = NULL;
1830 if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
1832 spin_lock_bh(&svsk->sk_defer_lock);
1833 clear_bit(SK_DEFERRED, &svsk->sk_flags);
1834 if (!list_empty(&svsk->sk_deferred)) {
1835 dr = list_entry(svsk->sk_deferred.next,
1836 struct svc_deferred_req,
1838 list_del_init(&dr->handle.recent);
1839 set_bit(SK_DEFERRED, &svsk->sk_flags);
1841 spin_unlock_bh(&svsk->sk_defer_lock);