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/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/errno.h>
25 #include <linux/fcntl.h>
26 #include <linux/net.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>
38 #include <net/checksum.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 /* SMP locking strategy:
53 * svc_pool->sp_lock protects most of the fields of that pool.
54 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
55 * when both need to be taken (rare), svc_serv->sv_lock is first.
56 * BKL protects svc_serv->sv_nrthread.
57 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
58 * and the ->sk_info_authunix cache.
59 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply.
61 * Some flags can be set to certain values at any time
62 * providing that certain rules are followed:
64 * SK_CONN, SK_DATA, can be set or cleared at any time.
65 * after a set, svc_sock_enqueue must be called.
66 * after a clear, the socket must be read/accepted
67 * if this succeeds, it must be set again.
68 * SK_CLOSE can set at any time. It is never cleared.
69 * sk_inuse contains a bias of '1' until SK_DEAD is set.
70 * so when sk_inuse hits zero, we know the socket is dead
71 * and no-one is using it.
72 * SK_DEAD can only be set while SK_BUSY is held which ensures
73 * no other thread will be using the socket or will try to
78 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
81 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
82 int *errp, int flags);
83 static void svc_delete_socket(struct svc_sock *svsk);
84 static void svc_udp_data_ready(struct sock *, int);
85 static int svc_udp_recvfrom(struct svc_rqst *);
86 static int svc_udp_sendto(struct svc_rqst *);
87 static void svc_close_socket(struct svc_sock *svsk);
88 static void svc_sock_detach(struct svc_xprt *);
89 static void svc_sock_free(struct svc_xprt *);
91 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk);
92 static int svc_deferred_recv(struct svc_rqst *rqstp);
93 static struct cache_deferred_req *svc_defer(struct cache_req *req);
95 /* apparently the "standard" is that clients close
96 * idle connections after 5 minutes, servers after
98 * http://www.connectathon.org/talks96/nfstcp.pdf
100 static int svc_conn_age_period = 6*60;
102 #ifdef CONFIG_DEBUG_LOCK_ALLOC
103 static struct lock_class_key svc_key[2];
104 static struct lock_class_key svc_slock_key[2];
106 static inline void svc_reclassify_socket(struct socket *sock)
108 struct sock *sk = sock->sk;
109 BUG_ON(sock_owned_by_user(sk));
110 switch (sk->sk_family) {
112 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
113 &svc_slock_key[0], "sk_lock-AF_INET-NFSD", &svc_key[0]);
117 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
118 &svc_slock_key[1], "sk_lock-AF_INET6-NFSD", &svc_key[1]);
126 static inline void svc_reclassify_socket(struct socket *sock)
131 static char *__svc_print_addr(struct sockaddr *addr, char *buf, size_t len)
133 switch (addr->sa_family) {
135 snprintf(buf, len, "%u.%u.%u.%u, port=%u",
136 NIPQUAD(((struct sockaddr_in *) addr)->sin_addr),
137 ntohs(((struct sockaddr_in *) addr)->sin_port));
141 snprintf(buf, len, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u",
142 NIP6(((struct sockaddr_in6 *) addr)->sin6_addr),
143 ntohs(((struct sockaddr_in6 *) addr)->sin6_port));
147 snprintf(buf, len, "unknown address type: %d", addr->sa_family);
154 * svc_print_addr - Format rq_addr field for printing
155 * @rqstp: svc_rqst struct containing address to print
156 * @buf: target buffer for formatted address
157 * @len: length of target buffer
160 char *svc_print_addr(struct svc_rqst *rqstp, char *buf, size_t len)
162 return __svc_print_addr(svc_addr(rqstp), buf, len);
164 EXPORT_SYMBOL_GPL(svc_print_addr);
167 * Queue up an idle server thread. Must have pool->sp_lock held.
168 * Note: this is really a stack rather than a queue, so that we only
169 * use as many different threads as we need, and the rest don't pollute
173 svc_thread_enqueue(struct svc_pool *pool, struct svc_rqst *rqstp)
175 list_add(&rqstp->rq_list, &pool->sp_threads);
179 * Dequeue an nfsd thread. Must have pool->sp_lock held.
182 svc_thread_dequeue(struct svc_pool *pool, struct svc_rqst *rqstp)
184 list_del(&rqstp->rq_list);
188 * Release an skbuff after use
190 static void svc_release_skb(struct svc_rqst *rqstp)
192 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
193 struct svc_deferred_req *dr = rqstp->rq_deferred;
196 rqstp->rq_xprt_ctxt = NULL;
198 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
199 skb_free_datagram(rqstp->rq_sock->sk_sk, skb);
202 rqstp->rq_deferred = NULL;
208 * Queue up a socket with data pending. If there are idle nfsd
209 * processes, wake 'em up.
213 svc_sock_enqueue(struct svc_sock *svsk)
215 struct svc_serv *serv = svsk->sk_server;
216 struct svc_pool *pool;
217 struct svc_rqst *rqstp;
220 if (!(svsk->sk_flags &
221 ( (1<<SK_CONN)|(1<<SK_DATA)|(1<<SK_CLOSE)|(1<<SK_DEFERRED)) ))
223 if (test_bit(SK_DEAD, &svsk->sk_flags))
227 pool = svc_pool_for_cpu(svsk->sk_server, cpu);
230 spin_lock_bh(&pool->sp_lock);
232 if (!list_empty(&pool->sp_threads) &&
233 !list_empty(&pool->sp_sockets))
235 "svc_sock_enqueue: threads and sockets both waiting??\n");
237 if (test_bit(SK_DEAD, &svsk->sk_flags)) {
238 /* Don't enqueue dead sockets */
239 dprintk("svc: socket %p is dead, not enqueued\n", svsk->sk_sk);
243 /* Mark socket as busy. It will remain in this state until the
244 * server has processed all pending data and put the socket back
245 * on the idle list. We update SK_BUSY atomically because
246 * it also guards against trying to enqueue the svc_sock twice.
248 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags)) {
249 /* Don't enqueue socket while already enqueued */
250 dprintk("svc: socket %p busy, not enqueued\n", svsk->sk_sk);
253 BUG_ON(svsk->sk_pool != NULL);
254 svsk->sk_pool = pool;
256 /* Handle pending connection */
257 if (test_bit(SK_CONN, &svsk->sk_flags))
260 /* Handle close in-progress */
261 if (test_bit(SK_CLOSE, &svsk->sk_flags))
264 /* Check if we have space to reply to a request */
265 if (!svsk->sk_xprt.xpt_ops->xpo_has_wspace(&svsk->sk_xprt)) {
266 /* Don't enqueue while not enough space for reply */
267 dprintk("svc: no write space, socket %p not enqueued\n", svsk);
268 svsk->sk_pool = NULL;
269 clear_bit(SK_BUSY, &svsk->sk_flags);
274 if (!list_empty(&pool->sp_threads)) {
275 rqstp = list_entry(pool->sp_threads.next,
278 dprintk("svc: socket %p served by daemon %p\n",
280 svc_thread_dequeue(pool, rqstp);
283 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
284 rqstp, rqstp->rq_sock);
285 rqstp->rq_sock = svsk;
286 atomic_inc(&svsk->sk_inuse);
287 rqstp->rq_reserved = serv->sv_max_mesg;
288 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
289 BUG_ON(svsk->sk_pool != pool);
290 wake_up(&rqstp->rq_wait);
292 dprintk("svc: socket %p put into queue\n", svsk->sk_sk);
293 list_add_tail(&svsk->sk_ready, &pool->sp_sockets);
294 BUG_ON(svsk->sk_pool != pool);
298 spin_unlock_bh(&pool->sp_lock);
302 * Dequeue the first socket. Must be called with the pool->sp_lock held.
304 static inline struct svc_sock *
305 svc_sock_dequeue(struct svc_pool *pool)
307 struct svc_sock *svsk;
309 if (list_empty(&pool->sp_sockets))
312 svsk = list_entry(pool->sp_sockets.next,
313 struct svc_sock, sk_ready);
314 list_del_init(&svsk->sk_ready);
316 dprintk("svc: socket %p dequeued, inuse=%d\n",
317 svsk->sk_sk, atomic_read(&svsk->sk_inuse));
323 * Having read something from a socket, check whether it
324 * needs to be re-enqueued.
325 * Note: SK_DATA only gets cleared when a read-attempt finds
326 * no (or insufficient) data.
329 svc_sock_received(struct svc_sock *svsk)
331 svsk->sk_pool = NULL;
332 clear_bit(SK_BUSY, &svsk->sk_flags);
333 svc_sock_enqueue(svsk);
338 * svc_reserve - change the space reserved for the reply to a request.
339 * @rqstp: The request in question
340 * @space: new max space to reserve
342 * Each request reserves some space on the output queue of the socket
343 * to make sure the reply fits. This function reduces that reserved
344 * space to be the amount of space used already, plus @space.
347 void svc_reserve(struct svc_rqst *rqstp, int space)
349 space += rqstp->rq_res.head[0].iov_len;
351 if (space < rqstp->rq_reserved) {
352 struct svc_sock *svsk = rqstp->rq_sock;
353 atomic_sub((rqstp->rq_reserved - space), &svsk->sk_reserved);
354 rqstp->rq_reserved = space;
356 svc_sock_enqueue(svsk);
361 * Release a socket after use.
364 svc_sock_put(struct svc_sock *svsk)
366 if (atomic_dec_and_test(&svsk->sk_inuse)) {
367 BUG_ON(!test_bit(SK_DEAD, &svsk->sk_flags));
368 svsk->sk_xprt.xpt_ops->xpo_free(&svsk->sk_xprt);
373 svc_sock_release(struct svc_rqst *rqstp)
375 struct svc_sock *svsk = rqstp->rq_sock;
377 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
379 svc_free_res_pages(rqstp);
380 rqstp->rq_res.page_len = 0;
381 rqstp->rq_res.page_base = 0;
384 /* Reset response buffer and release
386 * But first, check that enough space was reserved
387 * for the reply, otherwise we have a bug!
389 if ((rqstp->rq_res.len) > rqstp->rq_reserved)
390 printk(KERN_ERR "RPC request reserved %d but used %d\n",
394 rqstp->rq_res.head[0].iov_len = 0;
395 svc_reserve(rqstp, 0);
396 rqstp->rq_sock = NULL;
402 * External function to wake up a server waiting for data
403 * This really only makes sense for services like lockd
404 * which have exactly one thread anyway.
407 svc_wake_up(struct svc_serv *serv)
409 struct svc_rqst *rqstp;
411 struct svc_pool *pool;
413 for (i = 0; i < serv->sv_nrpools; i++) {
414 pool = &serv->sv_pools[i];
416 spin_lock_bh(&pool->sp_lock);
417 if (!list_empty(&pool->sp_threads)) {
418 rqstp = list_entry(pool->sp_threads.next,
421 dprintk("svc: daemon %p woken up.\n", rqstp);
423 svc_thread_dequeue(pool, rqstp);
424 rqstp->rq_sock = NULL;
426 wake_up(&rqstp->rq_wait);
428 spin_unlock_bh(&pool->sp_lock);
432 union svc_pktinfo_u {
433 struct in_pktinfo pkti;
434 struct in6_pktinfo pkti6;
436 #define SVC_PKTINFO_SPACE \
437 CMSG_SPACE(sizeof(union svc_pktinfo_u))
439 static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
441 switch (rqstp->rq_sock->sk_sk->sk_family) {
443 struct in_pktinfo *pki = CMSG_DATA(cmh);
445 cmh->cmsg_level = SOL_IP;
446 cmh->cmsg_type = IP_PKTINFO;
447 pki->ipi_ifindex = 0;
448 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
449 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
454 struct in6_pktinfo *pki = CMSG_DATA(cmh);
456 cmh->cmsg_level = SOL_IPV6;
457 cmh->cmsg_type = IPV6_PKTINFO;
458 pki->ipi6_ifindex = 0;
459 ipv6_addr_copy(&pki->ipi6_addr,
460 &rqstp->rq_daddr.addr6);
461 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
469 * Generic sendto routine
472 svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
474 struct svc_sock *svsk = rqstp->rq_sock;
475 struct socket *sock = svsk->sk_sock;
479 long all[SVC_PKTINFO_SPACE / sizeof(long)];
481 struct cmsghdr *cmh = &buffer.hdr;
485 struct page **ppage = xdr->pages;
486 size_t base = xdr->page_base;
487 unsigned int pglen = xdr->page_len;
488 unsigned int flags = MSG_MORE;
489 char buf[RPC_MAX_ADDRBUFLEN];
493 if (rqstp->rq_prot == IPPROTO_UDP) {
494 struct msghdr msg = {
495 .msg_name = &rqstp->rq_addr,
496 .msg_namelen = rqstp->rq_addrlen,
498 .msg_controllen = sizeof(buffer),
499 .msg_flags = MSG_MORE,
502 svc_set_cmsg_data(rqstp, cmh);
504 if (sock_sendmsg(sock, &msg, 0) < 0)
509 if (slen == xdr->head[0].iov_len)
511 len = kernel_sendpage(sock, rqstp->rq_respages[0], 0,
512 xdr->head[0].iov_len, flags);
513 if (len != xdr->head[0].iov_len)
515 slen -= xdr->head[0].iov_len;
520 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
524 result = kernel_sendpage(sock, *ppage, base, size, flags);
531 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
536 if (xdr->tail[0].iov_len) {
537 result = kernel_sendpage(sock, rqstp->rq_respages[0],
538 ((unsigned long)xdr->tail[0].iov_base)
540 xdr->tail[0].iov_len, 0);
546 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
547 rqstp->rq_sock, xdr->head[0].iov_base, xdr->head[0].iov_len,
548 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
554 * Report socket names for nfsdfs
556 static int one_sock_name(char *buf, struct svc_sock *svsk)
560 switch(svsk->sk_sk->sk_family) {
562 len = sprintf(buf, "ipv4 %s %u.%u.%u.%u %d\n",
563 svsk->sk_sk->sk_protocol==IPPROTO_UDP?
565 NIPQUAD(inet_sk(svsk->sk_sk)->rcv_saddr),
566 inet_sk(svsk->sk_sk)->num);
569 len = sprintf(buf, "*unknown-%d*\n",
570 svsk->sk_sk->sk_family);
576 svc_sock_names(char *buf, struct svc_serv *serv, char *toclose)
578 struct svc_sock *svsk, *closesk = NULL;
583 spin_lock_bh(&serv->sv_lock);
584 list_for_each_entry(svsk, &serv->sv_permsocks, sk_list) {
585 int onelen = one_sock_name(buf+len, svsk);
586 if (toclose && strcmp(toclose, buf+len) == 0)
591 spin_unlock_bh(&serv->sv_lock);
593 /* Should unregister with portmap, but you cannot
594 * unregister just one protocol...
596 svc_close_socket(closesk);
601 EXPORT_SYMBOL(svc_sock_names);
604 * Check input queue length
607 svc_recv_available(struct svc_sock *svsk)
609 struct socket *sock = svsk->sk_sock;
612 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
614 return (err >= 0)? avail : err;
618 * Generic recvfrom routine.
621 svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr, int buflen)
623 struct svc_sock *svsk = rqstp->rq_sock;
624 struct msghdr msg = {
625 .msg_flags = MSG_DONTWAIT,
627 struct sockaddr *sin;
630 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
633 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
635 memcpy(&rqstp->rq_addr, &svsk->sk_remote, svsk->sk_remotelen);
636 rqstp->rq_addrlen = svsk->sk_remotelen;
638 /* Destination address in request is needed for binding the
639 * source address in RPC callbacks later.
641 sin = (struct sockaddr *)&svsk->sk_local;
642 switch (sin->sa_family) {
644 rqstp->rq_daddr.addr = ((struct sockaddr_in *)sin)->sin_addr;
647 rqstp->rq_daddr.addr6 = ((struct sockaddr_in6 *)sin)->sin6_addr;
651 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
652 svsk, iov[0].iov_base, iov[0].iov_len, len);
658 * Set socket snd and rcv buffer lengths
661 svc_sock_setbufsize(struct socket *sock, unsigned int snd, unsigned int rcv)
665 oldfs = get_fs(); set_fs(KERNEL_DS);
666 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
667 (char*)&snd, sizeof(snd));
668 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
669 (char*)&rcv, sizeof(rcv));
671 /* sock_setsockopt limits use to sysctl_?mem_max,
672 * which isn't acceptable. Until that is made conditional
673 * on not having CAP_SYS_RESOURCE or similar, we go direct...
674 * DaveM said I could!
677 sock->sk->sk_sndbuf = snd * 2;
678 sock->sk->sk_rcvbuf = rcv * 2;
679 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
680 release_sock(sock->sk);
684 * INET callback when data has been received on the socket.
687 svc_udp_data_ready(struct sock *sk, int count)
689 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
692 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
693 svsk, sk, count, test_bit(SK_BUSY, &svsk->sk_flags));
694 set_bit(SK_DATA, &svsk->sk_flags);
695 svc_sock_enqueue(svsk);
697 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
698 wake_up_interruptible(sk->sk_sleep);
702 * INET callback when space is newly available on the socket.
705 svc_write_space(struct sock *sk)
707 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
710 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
711 svsk, sk, test_bit(SK_BUSY, &svsk->sk_flags));
712 svc_sock_enqueue(svsk);
715 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) {
716 dprintk("RPC svc_write_space: someone sleeping on %p\n",
718 wake_up_interruptible(sk->sk_sleep);
722 static inline void svc_udp_get_dest_address(struct svc_rqst *rqstp,
725 switch (rqstp->rq_sock->sk_sk->sk_family) {
727 struct in_pktinfo *pki = CMSG_DATA(cmh);
728 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
732 struct in6_pktinfo *pki = CMSG_DATA(cmh);
733 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
740 * Receive a datagram from a UDP socket.
743 svc_udp_recvfrom(struct svc_rqst *rqstp)
745 struct svc_sock *svsk = rqstp->rq_sock;
746 struct svc_serv *serv = svsk->sk_server;
750 long all[SVC_PKTINFO_SPACE / sizeof(long)];
752 struct cmsghdr *cmh = &buffer.hdr;
754 struct msghdr msg = {
755 .msg_name = svc_addr(rqstp),
757 .msg_controllen = sizeof(buffer),
758 .msg_flags = MSG_DONTWAIT,
761 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
762 /* udp sockets need large rcvbuf as all pending
763 * requests are still in that buffer. sndbuf must
764 * also be large enough that there is enough space
765 * for one reply per thread. We count all threads
766 * rather than threads in a particular pool, which
767 * provides an upper bound on the number of threads
768 * which will access the socket.
770 svc_sock_setbufsize(svsk->sk_sock,
771 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
772 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
774 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
775 svc_sock_received(svsk);
776 return svc_deferred_recv(rqstp);
779 clear_bit(SK_DATA, &svsk->sk_flags);
781 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
782 0, 0, MSG_PEEK | MSG_DONTWAIT);
784 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
787 if (err != -EAGAIN) {
788 /* possibly an icmp error */
789 dprintk("svc: recvfrom returned error %d\n", -err);
790 set_bit(SK_DATA, &svsk->sk_flags);
792 svc_sock_received(svsk);
795 rqstp->rq_addrlen = sizeof(rqstp->rq_addr);
796 if (skb->tstamp.tv64 == 0) {
797 skb->tstamp = ktime_get_real();
798 /* Don't enable netstamp, sunrpc doesn't
799 need that much accuracy */
801 svsk->sk_sk->sk_stamp = skb->tstamp;
802 set_bit(SK_DATA, &svsk->sk_flags); /* there may be more data... */
805 * Maybe more packets - kick another thread ASAP.
807 svc_sock_received(svsk);
809 len = skb->len - sizeof(struct udphdr);
810 rqstp->rq_arg.len = len;
812 rqstp->rq_prot = IPPROTO_UDP;
814 if (cmh->cmsg_level != IPPROTO_IP ||
815 cmh->cmsg_type != IP_PKTINFO) {
817 printk("rpcsvc: received unknown control message:"
819 cmh->cmsg_level, cmh->cmsg_type);
820 skb_free_datagram(svsk->sk_sk, skb);
823 svc_udp_get_dest_address(rqstp, cmh);
825 if (skb_is_nonlinear(skb)) {
826 /* we have to copy */
828 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
831 skb_free_datagram(svsk->sk_sk, skb);
835 skb_free_datagram(svsk->sk_sk, skb);
837 /* we can use it in-place */
838 rqstp->rq_arg.head[0].iov_base = skb->data + sizeof(struct udphdr);
839 rqstp->rq_arg.head[0].iov_len = len;
840 if (skb_checksum_complete(skb)) {
841 skb_free_datagram(svsk->sk_sk, skb);
844 rqstp->rq_xprt_ctxt = skb;
847 rqstp->rq_arg.page_base = 0;
848 if (len <= rqstp->rq_arg.head[0].iov_len) {
849 rqstp->rq_arg.head[0].iov_len = len;
850 rqstp->rq_arg.page_len = 0;
851 rqstp->rq_respages = rqstp->rq_pages+1;
853 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
854 rqstp->rq_respages = rqstp->rq_pages + 1 +
855 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
859 serv->sv_stats->netudpcnt++;
865 svc_udp_sendto(struct svc_rqst *rqstp)
869 error = svc_sendto(rqstp, &rqstp->rq_res);
870 if (error == -ECONNREFUSED)
871 /* ICMP error on earlier request. */
872 error = svc_sendto(rqstp, &rqstp->rq_res);
877 static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
881 static int svc_udp_has_wspace(struct svc_xprt *xprt)
883 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
884 struct svc_serv *serv = svsk->sk_server;
885 unsigned long required;
888 * Set the SOCK_NOSPACE flag before checking the available
891 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
892 required = atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg;
893 if (required*2 > sock_wspace(svsk->sk_sk))
895 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
899 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
905 static struct svc_xprt_ops svc_udp_ops = {
906 .xpo_recvfrom = svc_udp_recvfrom,
907 .xpo_sendto = svc_udp_sendto,
908 .xpo_release_rqst = svc_release_skb,
909 .xpo_detach = svc_sock_detach,
910 .xpo_free = svc_sock_free,
911 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
912 .xpo_has_wspace = svc_udp_has_wspace,
913 .xpo_accept = svc_udp_accept,
916 static struct svc_xprt_class svc_udp_class = {
918 .xcl_ops = &svc_udp_ops,
919 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
923 svc_udp_init(struct svc_sock *svsk)
928 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt);
929 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
930 svsk->sk_sk->sk_write_space = svc_write_space;
932 /* initialise setting must have enough space to
933 * receive and respond to one request.
934 * svc_udp_recvfrom will re-adjust if necessary
936 svc_sock_setbufsize(svsk->sk_sock,
937 3 * svsk->sk_server->sv_max_mesg,
938 3 * svsk->sk_server->sv_max_mesg);
940 set_bit(SK_DATA, &svsk->sk_flags); /* might have come in before data_ready set up */
941 set_bit(SK_CHNGBUF, &svsk->sk_flags);
945 /* make sure we get destination address info */
946 svsk->sk_sock->ops->setsockopt(svsk->sk_sock, IPPROTO_IP, IP_PKTINFO,
947 (char __user *)&one, sizeof(one));
952 * A data_ready event on a listening socket means there's a connection
953 * pending. Do not use state_change as a substitute for it.
956 svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
958 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
960 dprintk("svc: socket %p TCP (listen) state change %d\n",
964 * This callback may called twice when a new connection
965 * is established as a child socket inherits everything
966 * from a parent LISTEN socket.
967 * 1) data_ready method of the parent socket will be called
968 * when one of child sockets become ESTABLISHED.
969 * 2) data_ready method of the child socket may be called
970 * when it receives data before the socket is accepted.
971 * In case of 2, we should ignore it silently.
973 if (sk->sk_state == TCP_LISTEN) {
975 set_bit(SK_CONN, &svsk->sk_flags);
976 svc_sock_enqueue(svsk);
978 printk("svc: socket %p: no user data\n", sk);
981 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
982 wake_up_interruptible_all(sk->sk_sleep);
986 * A state change on a connected socket means it's dying or dead.
989 svc_tcp_state_change(struct sock *sk)
991 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
993 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
994 sk, sk->sk_state, sk->sk_user_data);
997 printk("svc: socket %p: no user data\n", sk);
999 set_bit(SK_CLOSE, &svsk->sk_flags);
1000 svc_sock_enqueue(svsk);
1002 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1003 wake_up_interruptible_all(sk->sk_sleep);
1007 svc_tcp_data_ready(struct sock *sk, int count)
1009 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
1011 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
1012 sk, sk->sk_user_data);
1014 set_bit(SK_DATA, &svsk->sk_flags);
1015 svc_sock_enqueue(svsk);
1017 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1018 wake_up_interruptible(sk->sk_sleep);
1021 static inline int svc_port_is_privileged(struct sockaddr *sin)
1023 switch (sin->sa_family) {
1025 return ntohs(((struct sockaddr_in *)sin)->sin_port)
1028 return ntohs(((struct sockaddr_in6 *)sin)->sin6_port)
1036 * Accept a TCP connection
1038 static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
1040 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1041 struct sockaddr_storage addr;
1042 struct sockaddr *sin = (struct sockaddr *) &addr;
1043 struct svc_serv *serv = svsk->sk_server;
1044 struct socket *sock = svsk->sk_sock;
1045 struct socket *newsock;
1046 struct svc_sock *newsvsk;
1048 char buf[RPC_MAX_ADDRBUFLEN];
1050 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
1054 clear_bit(SK_CONN, &svsk->sk_flags);
1055 err = kernel_accept(sock, &newsock, O_NONBLOCK);
1058 printk(KERN_WARNING "%s: no more sockets!\n",
1060 else if (err != -EAGAIN && net_ratelimit())
1061 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
1062 serv->sv_name, -err);
1066 set_bit(SK_CONN, &svsk->sk_flags);
1068 err = kernel_getpeername(newsock, sin, &slen);
1070 if (net_ratelimit())
1071 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
1072 serv->sv_name, -err);
1073 goto failed; /* aborted connection or whatever */
1076 /* Ideally, we would want to reject connections from unauthorized
1077 * hosts here, but when we get encryption, the IP of the host won't
1078 * tell us anything. For now just warn about unpriv connections.
1080 if (!svc_port_is_privileged(sin)) {
1081 dprintk(KERN_WARNING
1082 "%s: connect from unprivileged port: %s\n",
1084 __svc_print_addr(sin, buf, sizeof(buf)));
1086 dprintk("%s: connect from %s\n", serv->sv_name,
1087 __svc_print_addr(sin, buf, sizeof(buf)));
1089 /* make sure that a write doesn't block forever when
1092 newsock->sk->sk_sndtimeo = HZ*30;
1094 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
1095 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
1097 memcpy(&newsvsk->sk_remote, sin, slen);
1098 newsvsk->sk_remotelen = slen;
1099 err = kernel_getsockname(newsock, sin, &slen);
1100 if (unlikely(err < 0)) {
1101 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
1102 slen = offsetof(struct sockaddr, sa_data);
1104 memcpy(&newsvsk->sk_local, sin, slen);
1106 svc_sock_received(newsvsk);
1109 serv->sv_stats->nettcpconn++;
1111 return &newsvsk->sk_xprt;
1114 sock_release(newsock);
1119 * Make sure that we don't have too many active connections. If we
1120 * have, something must be dropped.
1122 * There's no point in trying to do random drop here for DoS
1123 * prevention. The NFS clients does 1 reconnect in 15 seconds. An
1124 * attacker can easily beat that.
1126 * The only somewhat efficient mechanism would be if drop old
1127 * connections from the same IP first. But right now we don't even
1128 * record the client IP in svc_sock.
1130 static void svc_check_conn_limits(struct svc_serv *serv)
1132 if (serv->sv_tmpcnt > (serv->sv_nrthreads+3)*20) {
1133 struct svc_sock *svsk = NULL;
1134 spin_lock_bh(&serv->sv_lock);
1135 if (!list_empty(&serv->sv_tempsocks)) {
1136 if (net_ratelimit()) {
1137 /* Try to help the admin */
1138 printk(KERN_NOTICE "%s: too many open TCP "
1139 "sockets, consider increasing the "
1140 "number of nfsd threads\n",
1144 * Always select the oldest socket. It's not fair,
1147 svsk = list_entry(serv->sv_tempsocks.prev,
1150 set_bit(SK_CLOSE, &svsk->sk_flags);
1151 atomic_inc(&svsk->sk_inuse);
1153 spin_unlock_bh(&serv->sv_lock);
1156 svc_sock_enqueue(svsk);
1163 * Receive data from a TCP socket.
1166 svc_tcp_recvfrom(struct svc_rqst *rqstp)
1168 struct svc_sock *svsk = rqstp->rq_sock;
1169 struct svc_serv *serv = svsk->sk_server;
1174 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1175 svsk, test_bit(SK_DATA, &svsk->sk_flags),
1176 test_bit(SK_CONN, &svsk->sk_flags),
1177 test_bit(SK_CLOSE, &svsk->sk_flags));
1179 if ((rqstp->rq_deferred = svc_deferred_dequeue(svsk))) {
1180 svc_sock_received(svsk);
1181 return svc_deferred_recv(rqstp);
1184 if (test_and_clear_bit(SK_CHNGBUF, &svsk->sk_flags))
1185 /* sndbuf needs to have room for one request
1186 * per thread, otherwise we can stall even when the
1187 * network isn't a bottleneck.
1189 * We count all threads rather than threads in a
1190 * particular pool, which provides an upper bound
1191 * on the number of threads which will access the socket.
1193 * rcvbuf just needs to be able to hold a few requests.
1194 * Normally they will be removed from the queue
1195 * as soon a a complete request arrives.
1197 svc_sock_setbufsize(svsk->sk_sock,
1198 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
1199 3 * serv->sv_max_mesg);
1201 clear_bit(SK_DATA, &svsk->sk_flags);
1203 /* Receive data. If we haven't got the record length yet, get
1204 * the next four bytes. Otherwise try to gobble up as much as
1205 * possible up to the complete record length.
1207 if (svsk->sk_tcplen < 4) {
1208 unsigned long want = 4 - svsk->sk_tcplen;
1211 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1213 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1215 svsk->sk_tcplen += len;
1218 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
1220 svc_sock_received(svsk);
1221 return -EAGAIN; /* record header not complete */
1224 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1225 if (!(svsk->sk_reclen & 0x80000000)) {
1226 /* FIXME: technically, a record can be fragmented,
1227 * and non-terminal fragments will not have the top
1228 * bit set in the fragment length header.
1229 * But apparently no known nfs clients send fragmented
1231 if (net_ratelimit())
1232 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
1233 " (non-terminal)\n",
1234 (unsigned long) svsk->sk_reclen);
1237 svsk->sk_reclen &= 0x7fffffff;
1238 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1239 if (svsk->sk_reclen > serv->sv_max_mesg) {
1240 if (net_ratelimit())
1241 printk(KERN_NOTICE "RPC: bad TCP reclen 0x%08lx"
1243 (unsigned long) svsk->sk_reclen);
1248 /* Check whether enough data is available */
1249 len = svc_recv_available(svsk);
1253 if (len < svsk->sk_reclen) {
1254 dprintk("svc: incomplete TCP record (%d of %d)\n",
1255 len, svsk->sk_reclen);
1256 svc_sock_received(svsk);
1257 return -EAGAIN; /* record not complete */
1259 len = svsk->sk_reclen;
1260 set_bit(SK_DATA, &svsk->sk_flags);
1262 vec = rqstp->rq_vec;
1263 vec[0] = rqstp->rq_arg.head[0];
1266 while (vlen < len) {
1267 vec[pnum].iov_base = page_address(rqstp->rq_pages[pnum]);
1268 vec[pnum].iov_len = PAGE_SIZE;
1272 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1274 /* Now receive data */
1275 len = svc_recvfrom(rqstp, vec, pnum, len);
1279 dprintk("svc: TCP complete record (%d bytes)\n", len);
1280 rqstp->rq_arg.len = len;
1281 rqstp->rq_arg.page_base = 0;
1282 if (len <= rqstp->rq_arg.head[0].iov_len) {
1283 rqstp->rq_arg.head[0].iov_len = len;
1284 rqstp->rq_arg.page_len = 0;
1286 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
1289 rqstp->rq_xprt_ctxt = NULL;
1290 rqstp->rq_prot = IPPROTO_TCP;
1292 /* Reset TCP read info */
1293 svsk->sk_reclen = 0;
1294 svsk->sk_tcplen = 0;
1296 svc_sock_received(svsk);
1298 serv->sv_stats->nettcpcnt++;
1303 set_bit(SK_CLOSE, &svsk->sk_flags);
1307 if (len == -EAGAIN) {
1308 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1309 svc_sock_received(svsk);
1311 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1312 svsk->sk_server->sv_name, -len);
1320 * Send out data on TCP socket.
1323 svc_tcp_sendto(struct svc_rqst *rqstp)
1325 struct xdr_buf *xbufp = &rqstp->rq_res;
1329 /* Set up the first element of the reply kvec.
1330 * Any other kvecs that may be in use have been taken
1331 * care of by the server implementation itself.
1333 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1334 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1336 if (test_bit(SK_DEAD, &rqstp->rq_sock->sk_flags))
1339 sent = svc_sendto(rqstp, &rqstp->rq_res);
1340 if (sent != xbufp->len) {
1341 printk(KERN_NOTICE "rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1342 rqstp->rq_sock->sk_server->sv_name,
1343 (sent<0)?"got error":"sent only",
1345 set_bit(SK_CLOSE, &rqstp->rq_sock->sk_flags);
1346 svc_sock_enqueue(rqstp->rq_sock);
1353 * Setup response header. TCP has a 4B record length field.
1355 static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1357 struct kvec *resv = &rqstp->rq_res.head[0];
1359 /* tcp needs a space for the record length... */
1363 static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1365 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1366 struct svc_serv *serv = svsk->sk_server;
1371 * Set the SOCK_NOSPACE flag before checking the available
1374 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1375 required = atomic_read(&svsk->sk_reserved) + serv->sv_max_mesg;
1376 wspace = sk_stream_wspace(svsk->sk_sk);
1378 if (wspace < sk_stream_min_wspace(svsk->sk_sk))
1380 if (required * 2 > wspace)
1383 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1387 static struct svc_xprt_ops svc_tcp_ops = {
1388 .xpo_recvfrom = svc_tcp_recvfrom,
1389 .xpo_sendto = svc_tcp_sendto,
1390 .xpo_release_rqst = svc_release_skb,
1391 .xpo_detach = svc_sock_detach,
1392 .xpo_free = svc_sock_free,
1393 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1394 .xpo_has_wspace = svc_tcp_has_wspace,
1395 .xpo_accept = svc_tcp_accept,
1398 static struct svc_xprt_class svc_tcp_class = {
1400 .xcl_ops = &svc_tcp_ops,
1401 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1404 void svc_init_xprt_sock(void)
1406 svc_reg_xprt_class(&svc_tcp_class);
1407 svc_reg_xprt_class(&svc_udp_class);
1410 void svc_cleanup_xprt_sock(void)
1412 svc_unreg_xprt_class(&svc_tcp_class);
1413 svc_unreg_xprt_class(&svc_udp_class);
1417 svc_tcp_init(struct svc_sock *svsk)
1419 struct sock *sk = svsk->sk_sk;
1420 struct tcp_sock *tp = tcp_sk(sk);
1422 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt);
1424 if (sk->sk_state == TCP_LISTEN) {
1425 dprintk("setting up TCP socket for listening\n");
1426 set_bit(SK_LISTENER, &svsk->sk_flags);
1427 sk->sk_data_ready = svc_tcp_listen_data_ready;
1428 set_bit(SK_CONN, &svsk->sk_flags);
1430 dprintk("setting up TCP socket for reading\n");
1431 sk->sk_state_change = svc_tcp_state_change;
1432 sk->sk_data_ready = svc_tcp_data_ready;
1433 sk->sk_write_space = svc_write_space;
1435 svsk->sk_reclen = 0;
1436 svsk->sk_tcplen = 0;
1438 tp->nonagle = 1; /* disable Nagle's algorithm */
1440 /* initialise setting must have enough space to
1441 * receive and respond to one request.
1442 * svc_tcp_recvfrom will re-adjust if necessary
1444 svc_sock_setbufsize(svsk->sk_sock,
1445 3 * svsk->sk_server->sv_max_mesg,
1446 3 * svsk->sk_server->sv_max_mesg);
1448 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1449 set_bit(SK_DATA, &svsk->sk_flags);
1450 if (sk->sk_state != TCP_ESTABLISHED)
1451 set_bit(SK_CLOSE, &svsk->sk_flags);
1456 svc_sock_update_bufs(struct svc_serv *serv)
1459 * The number of server threads has changed. Update
1460 * rcvbuf and sndbuf accordingly on all sockets
1462 struct list_head *le;
1464 spin_lock_bh(&serv->sv_lock);
1465 list_for_each(le, &serv->sv_permsocks) {
1466 struct svc_sock *svsk =
1467 list_entry(le, struct svc_sock, sk_list);
1468 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1470 list_for_each(le, &serv->sv_tempsocks) {
1471 struct svc_sock *svsk =
1472 list_entry(le, struct svc_sock, sk_list);
1473 set_bit(SK_CHNGBUF, &svsk->sk_flags);
1475 spin_unlock_bh(&serv->sv_lock);
1479 * Receive the next request on any socket. This code is carefully
1480 * organised not to touch any cachelines in the shared svc_serv
1481 * structure, only cachelines in the local svc_pool.
1484 svc_recv(struct svc_rqst *rqstp, long timeout)
1486 struct svc_sock *svsk = NULL;
1487 struct svc_serv *serv = rqstp->rq_server;
1488 struct svc_pool *pool = rqstp->rq_pool;
1491 struct xdr_buf *arg;
1492 DECLARE_WAITQUEUE(wait, current);
1494 dprintk("svc: server %p waiting for data (to = %ld)\n",
1499 "svc_recv: service %p, socket not NULL!\n",
1501 if (waitqueue_active(&rqstp->rq_wait))
1503 "svc_recv: service %p, wait queue active!\n",
1507 /* now allocate needed pages. If we get a failure, sleep briefly */
1508 pages = (serv->sv_max_mesg + PAGE_SIZE) / PAGE_SIZE;
1509 for (i=0; i < pages ; i++)
1510 while (rqstp->rq_pages[i] == NULL) {
1511 struct page *p = alloc_page(GFP_KERNEL);
1513 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1514 rqstp->rq_pages[i] = p;
1516 rqstp->rq_pages[i++] = NULL; /* this might be seen in nfs_read_actor */
1517 BUG_ON(pages >= RPCSVC_MAXPAGES);
1519 /* Make arg->head point to first page and arg->pages point to rest */
1520 arg = &rqstp->rq_arg;
1521 arg->head[0].iov_base = page_address(rqstp->rq_pages[0]);
1522 arg->head[0].iov_len = PAGE_SIZE;
1523 arg->pages = rqstp->rq_pages + 1;
1525 /* save at least one page for response */
1526 arg->page_len = (pages-2)*PAGE_SIZE;
1527 arg->len = (pages-1)*PAGE_SIZE;
1528 arg->tail[0].iov_len = 0;
1535 spin_lock_bh(&pool->sp_lock);
1536 if ((svsk = svc_sock_dequeue(pool)) != NULL) {
1537 rqstp->rq_sock = svsk;
1538 atomic_inc(&svsk->sk_inuse);
1539 rqstp->rq_reserved = serv->sv_max_mesg;
1540 atomic_add(rqstp->rq_reserved, &svsk->sk_reserved);
1542 /* No data pending. Go to sleep */
1543 svc_thread_enqueue(pool, rqstp);
1546 * We have to be able to interrupt this wait
1547 * to bring down the daemons ...
1549 set_current_state(TASK_INTERRUPTIBLE);
1550 add_wait_queue(&rqstp->rq_wait, &wait);
1551 spin_unlock_bh(&pool->sp_lock);
1553 schedule_timeout(timeout);
1557 spin_lock_bh(&pool->sp_lock);
1558 remove_wait_queue(&rqstp->rq_wait, &wait);
1560 if (!(svsk = rqstp->rq_sock)) {
1561 svc_thread_dequeue(pool, rqstp);
1562 spin_unlock_bh(&pool->sp_lock);
1563 dprintk("svc: server %p, no data yet\n", rqstp);
1564 return signalled()? -EINTR : -EAGAIN;
1567 spin_unlock_bh(&pool->sp_lock);
1570 if (test_bit(SK_CLOSE, &svsk->sk_flags)) {
1571 dprintk("svc_recv: found SK_CLOSE\n");
1572 svc_delete_socket(svsk);
1573 } else if (test_bit(SK_LISTENER, &svsk->sk_flags)) {
1574 struct svc_xprt *newxpt;
1575 newxpt = svsk->sk_xprt.xpt_ops->xpo_accept(&svsk->sk_xprt);
1577 svc_check_conn_limits(svsk->sk_server);
1578 svc_sock_received(svsk);
1580 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
1581 rqstp, pool->sp_id, svsk, atomic_read(&svsk->sk_inuse));
1582 len = svsk->sk_xprt.xpt_ops->xpo_recvfrom(rqstp);
1583 dprintk("svc: got len=%d\n", len);
1586 /* No data, incomplete (TCP) read, or accept() */
1587 if (len == 0 || len == -EAGAIN) {
1588 rqstp->rq_res.len = 0;
1589 svc_sock_release(rqstp);
1592 svsk->sk_lastrecv = get_seconds();
1593 clear_bit(SK_OLD, &svsk->sk_flags);
1595 rqstp->rq_secure = svc_port_is_privileged(svc_addr(rqstp));
1596 rqstp->rq_chandle.defer = svc_defer;
1599 serv->sv_stats->netcnt++;
1607 svc_drop(struct svc_rqst *rqstp)
1609 dprintk("svc: socket %p dropped request\n", rqstp->rq_sock);
1610 svc_sock_release(rqstp);
1614 * Return reply to client.
1617 svc_send(struct svc_rqst *rqstp)
1619 struct svc_sock *svsk;
1623 if ((svsk = rqstp->rq_sock) == NULL) {
1624 printk(KERN_WARNING "NULL socket pointer in %s:%d\n",
1625 __FILE__, __LINE__);
1629 /* release the receive skb before sending the reply */
1630 rqstp->rq_xprt->xpt_ops->xpo_release_rqst(rqstp);
1632 /* calculate over-all length */
1633 xb = & rqstp->rq_res;
1634 xb->len = xb->head[0].iov_len +
1636 xb->tail[0].iov_len;
1638 /* Grab svsk->sk_mutex to serialize outgoing data. */
1639 mutex_lock(&svsk->sk_mutex);
1640 if (test_bit(SK_DEAD, &svsk->sk_flags))
1643 len = svsk->sk_xprt.xpt_ops->xpo_sendto(rqstp);
1644 mutex_unlock(&svsk->sk_mutex);
1645 svc_sock_release(rqstp);
1647 if (len == -ECONNREFUSED || len == -ENOTCONN || len == -EAGAIN)
1653 * Timer function to close old temporary sockets, using
1654 * a mark-and-sweep algorithm.
1657 svc_age_temp_sockets(unsigned long closure)
1659 struct svc_serv *serv = (struct svc_serv *)closure;
1660 struct svc_sock *svsk;
1661 struct list_head *le, *next;
1662 LIST_HEAD(to_be_aged);
1664 dprintk("svc_age_temp_sockets\n");
1666 if (!spin_trylock_bh(&serv->sv_lock)) {
1667 /* busy, try again 1 sec later */
1668 dprintk("svc_age_temp_sockets: busy\n");
1669 mod_timer(&serv->sv_temptimer, jiffies + HZ);
1673 list_for_each_safe(le, next, &serv->sv_tempsocks) {
1674 svsk = list_entry(le, struct svc_sock, sk_list);
1676 if (!test_and_set_bit(SK_OLD, &svsk->sk_flags))
1678 if (atomic_read(&svsk->sk_inuse) > 1 || test_bit(SK_BUSY, &svsk->sk_flags))
1680 atomic_inc(&svsk->sk_inuse);
1681 list_move(le, &to_be_aged);
1682 set_bit(SK_CLOSE, &svsk->sk_flags);
1683 set_bit(SK_DETACHED, &svsk->sk_flags);
1685 spin_unlock_bh(&serv->sv_lock);
1687 while (!list_empty(&to_be_aged)) {
1688 le = to_be_aged.next;
1689 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */
1691 svsk = list_entry(le, struct svc_sock, sk_list);
1693 dprintk("queuing svsk %p for closing, %lu seconds old\n",
1694 svsk, get_seconds() - svsk->sk_lastrecv);
1696 /* a thread will dequeue and close it soon */
1697 svc_sock_enqueue(svsk);
1701 mod_timer(&serv->sv_temptimer, jiffies + svc_conn_age_period * HZ);
1705 * Initialize socket for RPC use and create svc_sock struct
1706 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1708 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1709 struct socket *sock,
1710 int *errp, int flags)
1712 struct svc_sock *svsk;
1714 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1715 int is_temporary = flags & SVC_SOCK_TEMPORARY;
1717 dprintk("svc: svc_setup_socket %p\n", sock);
1718 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1725 /* Register socket with portmapper */
1726 if (*errp >= 0 && pmap_register)
1727 *errp = svc_register(serv, inet->sk_protocol,
1728 ntohs(inet_sk(inet)->sport));
1735 set_bit(SK_BUSY, &svsk->sk_flags);
1736 inet->sk_user_data = svsk;
1737 svsk->sk_sock = sock;
1739 svsk->sk_ostate = inet->sk_state_change;
1740 svsk->sk_odata = inet->sk_data_ready;
1741 svsk->sk_owspace = inet->sk_write_space;
1742 svsk->sk_server = serv;
1743 atomic_set(&svsk->sk_inuse, 1);
1744 svsk->sk_lastrecv = get_seconds();
1745 spin_lock_init(&svsk->sk_lock);
1746 INIT_LIST_HEAD(&svsk->sk_deferred);
1747 INIT_LIST_HEAD(&svsk->sk_ready);
1748 mutex_init(&svsk->sk_mutex);
1750 /* Initialize the socket */
1751 if (sock->type == SOCK_DGRAM)
1756 spin_lock_bh(&serv->sv_lock);
1758 set_bit(SK_TEMP, &svsk->sk_flags);
1759 list_add(&svsk->sk_list, &serv->sv_tempsocks);
1761 if (serv->sv_temptimer.function == NULL) {
1762 /* setup timer to age temp sockets */
1763 setup_timer(&serv->sv_temptimer, svc_age_temp_sockets,
1764 (unsigned long)serv);
1765 mod_timer(&serv->sv_temptimer,
1766 jiffies + svc_conn_age_period * HZ);
1769 clear_bit(SK_TEMP, &svsk->sk_flags);
1770 list_add(&svsk->sk_list, &serv->sv_permsocks);
1772 spin_unlock_bh(&serv->sv_lock);
1774 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1780 int svc_addsock(struct svc_serv *serv,
1786 struct socket *so = sockfd_lookup(fd, &err);
1787 struct svc_sock *svsk = NULL;
1791 if (so->sk->sk_family != AF_INET)
1792 err = -EAFNOSUPPORT;
1793 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1794 so->sk->sk_protocol != IPPROTO_UDP)
1795 err = -EPROTONOSUPPORT;
1796 else if (so->state > SS_UNCONNECTED)
1799 svsk = svc_setup_socket(serv, so, &err, SVC_SOCK_DEFAULTS);
1801 svc_sock_received(svsk);
1809 if (proto) *proto = so->sk->sk_protocol;
1810 return one_sock_name(name_return, svsk);
1812 EXPORT_SYMBOL_GPL(svc_addsock);
1815 * Create socket for RPC service.
1817 static int svc_create_socket(struct svc_serv *serv, int protocol,
1818 struct sockaddr *sin, int len, int flags)
1820 struct svc_sock *svsk;
1821 struct socket *sock;
1824 char buf[RPC_MAX_ADDRBUFLEN];
1826 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1827 serv->sv_program->pg_name, protocol,
1828 __svc_print_addr(sin, buf, sizeof(buf)));
1830 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1831 printk(KERN_WARNING "svc: only UDP and TCP "
1832 "sockets supported\n");
1835 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1837 error = sock_create_kern(sin->sa_family, type, protocol, &sock);
1841 svc_reclassify_socket(sock);
1843 if (type == SOCK_STREAM)
1844 sock->sk->sk_reuse = 1; /* allow address reuse */
1845 error = kernel_bind(sock, sin, len);
1849 if (protocol == IPPROTO_TCP) {
1850 if ((error = kernel_listen(sock, 64)) < 0)
1854 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1855 svc_sock_received(svsk);
1856 return ntohs(inet_sk(svsk->sk_sk)->sport);
1860 dprintk("svc: svc_create_socket error = %d\n", -error);
1866 * Detach the svc_sock from the socket so that no
1867 * more callbacks occur.
1869 static void svc_sock_detach(struct svc_xprt *xprt)
1871 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1872 struct sock *sk = svsk->sk_sk;
1874 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1876 /* put back the old socket callbacks */
1877 sk->sk_state_change = svsk->sk_ostate;
1878 sk->sk_data_ready = svsk->sk_odata;
1879 sk->sk_write_space = svsk->sk_owspace;
1883 * Free the svc_sock's socket resources and the svc_sock itself.
1885 static void svc_sock_free(struct svc_xprt *xprt)
1887 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1888 dprintk("svc: svc_sock_free(%p)\n", svsk);
1890 if (svsk->sk_info_authunix != NULL)
1891 svcauth_unix_info_release(svsk->sk_info_authunix);
1892 if (svsk->sk_sock->file)
1893 sockfd_put(svsk->sk_sock);
1895 sock_release(svsk->sk_sock);
1900 * Remove a dead socket
1903 svc_delete_socket(struct svc_sock *svsk)
1905 struct svc_serv *serv;
1908 dprintk("svc: svc_delete_socket(%p)\n", svsk);
1910 serv = svsk->sk_server;
1913 svsk->sk_xprt.xpt_ops->xpo_detach(&svsk->sk_xprt);
1915 spin_lock_bh(&serv->sv_lock);
1917 if (!test_and_set_bit(SK_DETACHED, &svsk->sk_flags))
1918 list_del_init(&svsk->sk_list);
1920 * We used to delete the svc_sock from whichever list
1921 * it's sk_ready node was on, but we don't actually
1922 * need to. This is because the only time we're called
1923 * while still attached to a queue, the queue itself
1924 * is about to be destroyed (in svc_destroy).
1926 if (!test_and_set_bit(SK_DEAD, &svsk->sk_flags)) {
1927 BUG_ON(atomic_read(&svsk->sk_inuse)<2);
1928 atomic_dec(&svsk->sk_inuse);
1929 if (test_bit(SK_TEMP, &svsk->sk_flags))
1933 spin_unlock_bh(&serv->sv_lock);
1936 static void svc_close_socket(struct svc_sock *svsk)
1938 set_bit(SK_CLOSE, &svsk->sk_flags);
1939 if (test_and_set_bit(SK_BUSY, &svsk->sk_flags))
1940 /* someone else will have to effect the close */
1943 atomic_inc(&svsk->sk_inuse);
1944 svc_delete_socket(svsk);
1945 clear_bit(SK_BUSY, &svsk->sk_flags);
1949 void svc_force_close_socket(struct svc_sock *svsk)
1951 set_bit(SK_CLOSE, &svsk->sk_flags);
1952 if (test_bit(SK_BUSY, &svsk->sk_flags)) {
1953 /* Waiting to be processed, but no threads left,
1954 * So just remove it from the waiting list
1956 list_del_init(&svsk->sk_ready);
1957 clear_bit(SK_BUSY, &svsk->sk_flags);
1959 svc_close_socket(svsk);
1963 * svc_makesock - Make a socket for nfsd and lockd
1964 * @serv: RPC server structure
1965 * @protocol: transport protocol to use
1966 * @port: port to use
1967 * @flags: requested socket characteristics
1970 int svc_makesock(struct svc_serv *serv, int protocol, unsigned short port,
1973 struct sockaddr_in sin = {
1974 .sin_family = AF_INET,
1975 .sin_addr.s_addr = INADDR_ANY,
1976 .sin_port = htons(port),
1979 dprintk("svc: creating socket proto = %d\n", protocol);
1980 return svc_create_socket(serv, protocol, (struct sockaddr *) &sin,
1981 sizeof(sin), flags);
1985 * Handle defer and revisit of requests
1988 static void svc_revisit(struct cache_deferred_req *dreq, int too_many)
1990 struct svc_deferred_req *dr = container_of(dreq, struct svc_deferred_req, handle);
1991 struct svc_sock *svsk;
1994 svc_sock_put(dr->svsk);
1998 dprintk("revisit queued\n");
2001 spin_lock(&svsk->sk_lock);
2002 list_add(&dr->handle.recent, &svsk->sk_deferred);
2003 spin_unlock(&svsk->sk_lock);
2004 set_bit(SK_DEFERRED, &svsk->sk_flags);
2005 svc_sock_enqueue(svsk);
2009 static struct cache_deferred_req *
2010 svc_defer(struct cache_req *req)
2012 struct svc_rqst *rqstp = container_of(req, struct svc_rqst, rq_chandle);
2013 int size = sizeof(struct svc_deferred_req) + (rqstp->rq_arg.len);
2014 struct svc_deferred_req *dr;
2016 if (rqstp->rq_arg.page_len)
2017 return NULL; /* if more than a page, give up FIXME */
2018 if (rqstp->rq_deferred) {
2019 dr = rqstp->rq_deferred;
2020 rqstp->rq_deferred = NULL;
2022 int skip = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
2023 /* FIXME maybe discard if size too large */
2024 dr = kmalloc(size, GFP_KERNEL);
2028 dr->handle.owner = rqstp->rq_server;
2029 dr->prot = rqstp->rq_prot;
2030 memcpy(&dr->addr, &rqstp->rq_addr, rqstp->rq_addrlen);
2031 dr->addrlen = rqstp->rq_addrlen;
2032 dr->daddr = rqstp->rq_daddr;
2033 dr->argslen = rqstp->rq_arg.len >> 2;
2034 memcpy(dr->args, rqstp->rq_arg.head[0].iov_base-skip, dr->argslen<<2);
2036 atomic_inc(&rqstp->rq_sock->sk_inuse);
2037 dr->svsk = rqstp->rq_sock;
2039 dr->handle.revisit = svc_revisit;
2044 * recv data from a deferred request into an active one
2046 static int svc_deferred_recv(struct svc_rqst *rqstp)
2048 struct svc_deferred_req *dr = rqstp->rq_deferred;
2050 rqstp->rq_arg.head[0].iov_base = dr->args;
2051 rqstp->rq_arg.head[0].iov_len = dr->argslen<<2;
2052 rqstp->rq_arg.page_len = 0;
2053 rqstp->rq_arg.len = dr->argslen<<2;
2054 rqstp->rq_prot = dr->prot;
2055 memcpy(&rqstp->rq_addr, &dr->addr, dr->addrlen);
2056 rqstp->rq_addrlen = dr->addrlen;
2057 rqstp->rq_daddr = dr->daddr;
2058 rqstp->rq_respages = rqstp->rq_pages;
2059 return dr->argslen<<2;
2063 static struct svc_deferred_req *svc_deferred_dequeue(struct svc_sock *svsk)
2065 struct svc_deferred_req *dr = NULL;
2067 if (!test_bit(SK_DEFERRED, &svsk->sk_flags))
2069 spin_lock(&svsk->sk_lock);
2070 clear_bit(SK_DEFERRED, &svsk->sk_flags);
2071 if (!list_empty(&svsk->sk_deferred)) {
2072 dr = list_entry(svsk->sk_deferred.next,
2073 struct svc_deferred_req,
2075 list_del_init(&dr->handle.recent);
2076 set_bit(SK_DEFERRED, &svsk->sk_flags);
2078 spin_unlock(&svsk->sk_lock);