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>
40 #include <net/tcp_states.h>
41 #include <asm/uaccess.h>
42 #include <asm/ioctls.h>
44 #include <linux/sunrpc/types.h>
45 #include <linux/sunrpc/clnt.h>
46 #include <linux/sunrpc/xdr.h>
47 #include <linux/sunrpc/svcsock.h>
48 #include <linux/sunrpc/stats.h>
50 /* SMP locking strategy:
52 * svc_pool->sp_lock protects most of the fields of that pool.
53 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
54 * when both need to be taken (rare), svc_serv->sv_lock is first.
55 * BKL protects svc_serv->sv_nrthread.
56 * svc_sock->sk_lock protects the svc_sock->sk_deferred list
57 * and the ->sk_info_authunix cache.
58 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply.
60 * Some flags can be set to certain values at any time
61 * providing that certain rules are followed:
63 * SK_CONN, SK_DATA, can be set or cleared at any time.
64 * after a set, svc_sock_enqueue must be called.
65 * after a clear, the socket must be read/accepted
66 * if this succeeds, it must be set again.
67 * SK_CLOSE can set at any time. It is never cleared.
68 * sk_inuse contains a bias of '1' until SK_DEAD is set.
69 * so when sk_inuse hits zero, we know the socket is dead
70 * and no-one is using it.
71 * SK_DEAD can only be set while SK_BUSY is held which ensures
72 * no other thread will be using the socket or will try to
77 #define RPCDBG_FACILITY RPCDBG_SVCSOCK
80 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
81 int *errp
, int flags
);
82 static void svc_delete_socket(struct svc_sock
*svsk
);
83 static void svc_udp_data_ready(struct sock
*, int);
84 static int svc_udp_recvfrom(struct svc_rqst
*);
85 static int svc_udp_sendto(struct svc_rqst
*);
86 static void svc_close_socket(struct svc_sock
*svsk
);
88 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_sock
*svsk
);
89 static int svc_deferred_recv(struct svc_rqst
*rqstp
);
90 static struct cache_deferred_req
*svc_defer(struct cache_req
*req
);
92 /* apparently the "standard" is that clients close
93 * idle connections after 5 minutes, servers after
95 * http://www.connectathon.org/talks96/nfstcp.pdf
97 static int svc_conn_age_period
= 6*60;
99 #ifdef CONFIG_DEBUG_LOCK_ALLOC
100 static struct lock_class_key svc_key
[2];
101 static struct lock_class_key svc_slock_key
[2];
103 static inline void svc_reclassify_socket(struct socket
*sock
)
105 struct sock
*sk
= sock
->sk
;
106 BUG_ON(sk
->sk_lock
.owner
!= NULL
);
107 switch (sk
->sk_family
) {
109 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
110 &svc_slock_key
[0], "sk_lock-AF_INET-NFSD", &svc_key
[0]);
114 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
115 &svc_slock_key
[1], "sk_lock-AF_INET6-NFSD", &svc_key
[1]);
123 static inline void svc_reclassify_socket(struct socket
*sock
)
128 static char *__svc_print_addr(struct sockaddr
*addr
, char *buf
, size_t len
)
130 switch (addr
->sa_family
) {
132 snprintf(buf
, len
, "%u.%u.%u.%u, port=%u",
133 NIPQUAD(((struct sockaddr_in
*) addr
)->sin_addr
),
134 ntohs(((struct sockaddr_in
*) addr
)->sin_port
));
138 snprintf(buf
, len
, "%x:%x:%x:%x:%x:%x:%x:%x, port=%u",
139 NIP6(((struct sockaddr_in6
*) addr
)->sin6_addr
),
140 ntohs(((struct sockaddr_in6
*) addr
)->sin6_port
));
144 snprintf(buf
, len
, "unknown address type: %d", addr
->sa_family
);
151 * svc_print_addr - Format rq_addr field for printing
152 * @rqstp: svc_rqst struct containing address to print
153 * @buf: target buffer for formatted address
154 * @len: length of target buffer
157 char *svc_print_addr(struct svc_rqst
*rqstp
, char *buf
, size_t len
)
159 return __svc_print_addr(svc_addr(rqstp
), buf
, len
);
161 EXPORT_SYMBOL_GPL(svc_print_addr
);
164 * Queue up an idle server thread. Must have pool->sp_lock held.
165 * Note: this is really a stack rather than a queue, so that we only
166 * use as many different threads as we need, and the rest don't pollute
170 svc_thread_enqueue(struct svc_pool
*pool
, struct svc_rqst
*rqstp
)
172 list_add(&rqstp
->rq_list
, &pool
->sp_threads
);
176 * Dequeue an nfsd thread. Must have pool->sp_lock held.
179 svc_thread_dequeue(struct svc_pool
*pool
, struct svc_rqst
*rqstp
)
181 list_del(&rqstp
->rq_list
);
185 * Release an skbuff after use
188 svc_release_skb(struct svc_rqst
*rqstp
)
190 struct sk_buff
*skb
= rqstp
->rq_skbuff
;
191 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
194 rqstp
->rq_skbuff
= NULL
;
196 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
197 skb_free_datagram(rqstp
->rq_sock
->sk_sk
, skb
);
200 rqstp
->rq_deferred
= NULL
;
206 * Any space to write?
208 static inline unsigned long
209 svc_sock_wspace(struct svc_sock
*svsk
)
213 if (svsk
->sk_sock
->type
== SOCK_STREAM
)
214 wspace
= sk_stream_wspace(svsk
->sk_sk
);
216 wspace
= sock_wspace(svsk
->sk_sk
);
222 * Queue up a socket with data pending. If there are idle nfsd
223 * processes, wake 'em up.
227 svc_sock_enqueue(struct svc_sock
*svsk
)
229 struct svc_serv
*serv
= svsk
->sk_server
;
230 struct svc_pool
*pool
;
231 struct svc_rqst
*rqstp
;
234 if (!(svsk
->sk_flags
&
235 ( (1<<SK_CONN
)|(1<<SK_DATA
)|(1<<SK_CLOSE
)|(1<<SK_DEFERRED
)) ))
237 if (test_bit(SK_DEAD
, &svsk
->sk_flags
))
241 pool
= svc_pool_for_cpu(svsk
->sk_server
, cpu
);
244 spin_lock_bh(&pool
->sp_lock
);
246 if (!list_empty(&pool
->sp_threads
) &&
247 !list_empty(&pool
->sp_sockets
))
249 "svc_sock_enqueue: threads and sockets both waiting??\n");
251 if (test_bit(SK_DEAD
, &svsk
->sk_flags
)) {
252 /* Don't enqueue dead sockets */
253 dprintk("svc: socket %p is dead, not enqueued\n", svsk
->sk_sk
);
257 /* Mark socket as busy. It will remain in this state until the
258 * server has processed all pending data and put the socket back
259 * on the idle list. We update SK_BUSY atomically because
260 * it also guards against trying to enqueue the svc_sock twice.
262 if (test_and_set_bit(SK_BUSY
, &svsk
->sk_flags
)) {
263 /* Don't enqueue socket while already enqueued */
264 dprintk("svc: socket %p busy, not enqueued\n", svsk
->sk_sk
);
267 BUG_ON(svsk
->sk_pool
!= NULL
);
268 svsk
->sk_pool
= pool
;
270 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
271 if (((atomic_read(&svsk
->sk_reserved
) + serv
->sv_max_mesg
)*2
272 > svc_sock_wspace(svsk
))
273 && !test_bit(SK_CLOSE
, &svsk
->sk_flags
)
274 && !test_bit(SK_CONN
, &svsk
->sk_flags
)) {
275 /* Don't enqueue while not enough space for reply */
276 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n",
277 svsk
->sk_sk
, atomic_read(&svsk
->sk_reserved
)+serv
->sv_max_mesg
,
278 svc_sock_wspace(svsk
));
279 svsk
->sk_pool
= NULL
;
280 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
283 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
286 if (!list_empty(&pool
->sp_threads
)) {
287 rqstp
= list_entry(pool
->sp_threads
.next
,
290 dprintk("svc: socket %p served by daemon %p\n",
292 svc_thread_dequeue(pool
, rqstp
);
295 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
296 rqstp
, rqstp
->rq_sock
);
297 rqstp
->rq_sock
= svsk
;
298 atomic_inc(&svsk
->sk_inuse
);
299 rqstp
->rq_reserved
= serv
->sv_max_mesg
;
300 atomic_add(rqstp
->rq_reserved
, &svsk
->sk_reserved
);
301 BUG_ON(svsk
->sk_pool
!= pool
);
302 wake_up(&rqstp
->rq_wait
);
304 dprintk("svc: socket %p put into queue\n", svsk
->sk_sk
);
305 list_add_tail(&svsk
->sk_ready
, &pool
->sp_sockets
);
306 BUG_ON(svsk
->sk_pool
!= pool
);
310 spin_unlock_bh(&pool
->sp_lock
);
314 * Dequeue the first socket. Must be called with the pool->sp_lock held.
316 static inline struct svc_sock
*
317 svc_sock_dequeue(struct svc_pool
*pool
)
319 struct svc_sock
*svsk
;
321 if (list_empty(&pool
->sp_sockets
))
324 svsk
= list_entry(pool
->sp_sockets
.next
,
325 struct svc_sock
, sk_ready
);
326 list_del_init(&svsk
->sk_ready
);
328 dprintk("svc: socket %p dequeued, inuse=%d\n",
329 svsk
->sk_sk
, atomic_read(&svsk
->sk_inuse
));
335 * Having read something from a socket, check whether it
336 * needs to be re-enqueued.
337 * Note: SK_DATA only gets cleared when a read-attempt finds
338 * no (or insufficient) data.
341 svc_sock_received(struct svc_sock
*svsk
)
343 svsk
->sk_pool
= NULL
;
344 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
345 svc_sock_enqueue(svsk
);
350 * svc_reserve - change the space reserved for the reply to a request.
351 * @rqstp: The request in question
352 * @space: new max space to reserve
354 * Each request reserves some space on the output queue of the socket
355 * to make sure the reply fits. This function reduces that reserved
356 * space to be the amount of space used already, plus @space.
359 void svc_reserve(struct svc_rqst
*rqstp
, int space
)
361 space
+= rqstp
->rq_res
.head
[0].iov_len
;
363 if (space
< rqstp
->rq_reserved
) {
364 struct svc_sock
*svsk
= rqstp
->rq_sock
;
365 atomic_sub((rqstp
->rq_reserved
- space
), &svsk
->sk_reserved
);
366 rqstp
->rq_reserved
= space
;
368 svc_sock_enqueue(svsk
);
373 * Release a socket after use.
376 svc_sock_put(struct svc_sock
*svsk
)
378 if (atomic_dec_and_test(&svsk
->sk_inuse
)) {
379 BUG_ON(! test_bit(SK_DEAD
, &svsk
->sk_flags
));
381 dprintk("svc: releasing dead socket\n");
382 if (svsk
->sk_sock
->file
)
383 sockfd_put(svsk
->sk_sock
);
385 sock_release(svsk
->sk_sock
);
386 if (svsk
->sk_info_authunix
!= NULL
)
387 svcauth_unix_info_release(svsk
->sk_info_authunix
);
393 svc_sock_release(struct svc_rqst
*rqstp
)
395 struct svc_sock
*svsk
= rqstp
->rq_sock
;
397 svc_release_skb(rqstp
);
399 svc_free_res_pages(rqstp
);
400 rqstp
->rq_res
.page_len
= 0;
401 rqstp
->rq_res
.page_base
= 0;
404 /* Reset response buffer and release
406 * But first, check that enough space was reserved
407 * for the reply, otherwise we have a bug!
409 if ((rqstp
->rq_res
.len
) > rqstp
->rq_reserved
)
410 printk(KERN_ERR
"RPC request reserved %d but used %d\n",
414 rqstp
->rq_res
.head
[0].iov_len
= 0;
415 svc_reserve(rqstp
, 0);
416 rqstp
->rq_sock
= NULL
;
422 * External function to wake up a server waiting for data
423 * This really only makes sense for services like lockd
424 * which have exactly one thread anyway.
427 svc_wake_up(struct svc_serv
*serv
)
429 struct svc_rqst
*rqstp
;
431 struct svc_pool
*pool
;
433 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
434 pool
= &serv
->sv_pools
[i
];
436 spin_lock_bh(&pool
->sp_lock
);
437 if (!list_empty(&pool
->sp_threads
)) {
438 rqstp
= list_entry(pool
->sp_threads
.next
,
441 dprintk("svc: daemon %p woken up.\n", rqstp
);
443 svc_thread_dequeue(pool, rqstp);
444 rqstp->rq_sock = NULL;
446 wake_up(&rqstp
->rq_wait
);
448 spin_unlock_bh(&pool
->sp_lock
);
452 union svc_pktinfo_u
{
453 struct in_pktinfo pkti
;
454 struct in6_pktinfo pkti6
;
456 #define SVC_PKTINFO_SPACE \
457 CMSG_SPACE(sizeof(union svc_pktinfo_u))
459 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
461 switch (rqstp
->rq_sock
->sk_sk
->sk_family
) {
463 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
465 cmh
->cmsg_level
= SOL_IP
;
466 cmh
->cmsg_type
= IP_PKTINFO
;
467 pki
->ipi_ifindex
= 0;
468 pki
->ipi_spec_dst
.s_addr
= rqstp
->rq_daddr
.addr
.s_addr
;
469 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
474 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
476 cmh
->cmsg_level
= SOL_IPV6
;
477 cmh
->cmsg_type
= IPV6_PKTINFO
;
478 pki
->ipi6_ifindex
= 0;
479 ipv6_addr_copy(&pki
->ipi6_addr
,
480 &rqstp
->rq_daddr
.addr6
);
481 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
489 * Generic sendto routine
492 svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
494 struct svc_sock
*svsk
= rqstp
->rq_sock
;
495 struct socket
*sock
= svsk
->sk_sock
;
499 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
501 struct cmsghdr
*cmh
= &buffer
.hdr
;
505 struct page
**ppage
= xdr
->pages
;
506 size_t base
= xdr
->page_base
;
507 unsigned int pglen
= xdr
->page_len
;
508 unsigned int flags
= MSG_MORE
;
509 char buf
[RPC_MAX_ADDRBUFLEN
];
513 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
514 struct msghdr msg
= {
515 .msg_name
= &rqstp
->rq_addr
,
516 .msg_namelen
= rqstp
->rq_addrlen
,
518 .msg_controllen
= sizeof(buffer
),
519 .msg_flags
= MSG_MORE
,
522 svc_set_cmsg_data(rqstp
, cmh
);
524 if (sock_sendmsg(sock
, &msg
, 0) < 0)
529 if (slen
== xdr
->head
[0].iov_len
)
531 len
= kernel_sendpage(sock
, rqstp
->rq_respages
[0], 0,
532 xdr
->head
[0].iov_len
, flags
);
533 if (len
!= xdr
->head
[0].iov_len
)
535 slen
-= xdr
->head
[0].iov_len
;
540 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
544 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
551 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
556 if (xdr
->tail
[0].iov_len
) {
557 result
= kernel_sendpage(sock
, rqstp
->rq_respages
[0],
558 ((unsigned long)xdr
->tail
[0].iov_base
)
560 xdr
->tail
[0].iov_len
, 0);
566 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
567 rqstp
->rq_sock
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
568 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
574 * Report socket names for nfsdfs
576 static int one_sock_name(char *buf
, struct svc_sock
*svsk
)
580 switch(svsk
->sk_sk
->sk_family
) {
582 len
= sprintf(buf
, "ipv4 %s %u.%u.%u.%u %d\n",
583 svsk
->sk_sk
->sk_protocol
==IPPROTO_UDP
?
585 NIPQUAD(inet_sk(svsk
->sk_sk
)->rcv_saddr
),
586 inet_sk(svsk
->sk_sk
)->num
);
589 len
= sprintf(buf
, "*unknown-%d*\n",
590 svsk
->sk_sk
->sk_family
);
596 svc_sock_names(char *buf
, struct svc_serv
*serv
, char *toclose
)
598 struct svc_sock
*svsk
, *closesk
= NULL
;
603 spin_lock_bh(&serv
->sv_lock
);
604 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_list
) {
605 int onelen
= one_sock_name(buf
+len
, svsk
);
606 if (toclose
&& strcmp(toclose
, buf
+len
) == 0)
611 spin_unlock_bh(&serv
->sv_lock
);
613 /* Should unregister with portmap, but you cannot
614 * unregister just one protocol...
616 svc_close_socket(closesk
);
621 EXPORT_SYMBOL(svc_sock_names
);
624 * Check input queue length
627 svc_recv_available(struct svc_sock
*svsk
)
629 struct socket
*sock
= svsk
->sk_sock
;
632 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
634 return (err
>= 0)? avail
: err
;
638 * Generic recvfrom routine.
641 svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
, int buflen
)
643 struct svc_sock
*svsk
= rqstp
->rq_sock
;
644 struct msghdr msg
= {
645 .msg_flags
= MSG_DONTWAIT
,
647 struct sockaddr
*sin
;
650 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
653 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
655 memcpy(&rqstp
->rq_addr
, &svsk
->sk_remote
, svsk
->sk_remotelen
);
656 rqstp
->rq_addrlen
= svsk
->sk_remotelen
;
658 /* Destination address in request is needed for binding the
659 * source address in RPC callbacks later.
661 sin
= (struct sockaddr
*)&svsk
->sk_local
;
662 switch (sin
->sa_family
) {
664 rqstp
->rq_daddr
.addr
= ((struct sockaddr_in
*)sin
)->sin_addr
;
667 rqstp
->rq_daddr
.addr6
= ((struct sockaddr_in6
*)sin
)->sin6_addr
;
671 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
672 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
678 * Set socket snd and rcv buffer lengths
681 svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
, unsigned int rcv
)
685 oldfs
= get_fs(); set_fs(KERNEL_DS
);
686 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
687 (char*)&snd
, sizeof(snd
));
688 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
689 (char*)&rcv
, sizeof(rcv
));
691 /* sock_setsockopt limits use to sysctl_?mem_max,
692 * which isn't acceptable. Until that is made conditional
693 * on not having CAP_SYS_RESOURCE or similar, we go direct...
694 * DaveM said I could!
697 sock
->sk
->sk_sndbuf
= snd
* 2;
698 sock
->sk
->sk_rcvbuf
= rcv
* 2;
699 sock
->sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
|SOCK_RCVBUF_LOCK
;
700 release_sock(sock
->sk
);
704 * INET callback when data has been received on the socket.
707 svc_udp_data_ready(struct sock
*sk
, int count
)
709 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
712 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
713 svsk
, sk
, count
, test_bit(SK_BUSY
, &svsk
->sk_flags
));
714 set_bit(SK_DATA
, &svsk
->sk_flags
);
715 svc_sock_enqueue(svsk
);
717 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
718 wake_up_interruptible(sk
->sk_sleep
);
722 * INET callback when space is newly available on the socket.
725 svc_write_space(struct sock
*sk
)
727 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
730 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
731 svsk
, sk
, test_bit(SK_BUSY
, &svsk
->sk_flags
));
732 svc_sock_enqueue(svsk
);
735 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
)) {
736 dprintk("RPC svc_write_space: someone sleeping on %p\n",
738 wake_up_interruptible(sk
->sk_sleep
);
742 static inline void svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
745 switch (rqstp
->rq_sock
->sk_sk
->sk_family
) {
747 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
748 rqstp
->rq_daddr
.addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
752 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
753 ipv6_addr_copy(&rqstp
->rq_daddr
.addr6
, &pki
->ipi6_addr
);
760 * Receive a datagram from a UDP socket.
763 svc_udp_recvfrom(struct svc_rqst
*rqstp
)
765 struct svc_sock
*svsk
= rqstp
->rq_sock
;
766 struct svc_serv
*serv
= svsk
->sk_server
;
770 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
772 struct cmsghdr
*cmh
= &buffer
.hdr
;
774 struct msghdr msg
= {
775 .msg_name
= svc_addr(rqstp
),
777 .msg_controllen
= sizeof(buffer
),
778 .msg_flags
= MSG_DONTWAIT
,
781 if (test_and_clear_bit(SK_CHNGBUF
, &svsk
->sk_flags
))
782 /* udp sockets need large rcvbuf as all pending
783 * requests are still in that buffer. sndbuf must
784 * also be large enough that there is enough space
785 * for one reply per thread. We count all threads
786 * rather than threads in a particular pool, which
787 * provides an upper bound on the number of threads
788 * which will access the socket.
790 svc_sock_setbufsize(svsk
->sk_sock
,
791 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
792 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
794 if ((rqstp
->rq_deferred
= svc_deferred_dequeue(svsk
))) {
795 svc_sock_received(svsk
);
796 return svc_deferred_recv(rqstp
);
799 if (test_bit(SK_CLOSE
, &svsk
->sk_flags
)) {
800 svc_delete_socket(svsk
);
804 clear_bit(SK_DATA
, &svsk
->sk_flags
);
806 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
807 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
809 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
812 if (err
!= -EAGAIN
) {
813 /* possibly an icmp error */
814 dprintk("svc: recvfrom returned error %d\n", -err
);
815 set_bit(SK_DATA
, &svsk
->sk_flags
);
817 svc_sock_received(svsk
);
820 rqstp
->rq_addrlen
= sizeof(rqstp
->rq_addr
);
821 if (skb
->tstamp
.tv64
== 0) {
822 skb
->tstamp
= ktime_get_real();
823 /* Don't enable netstamp, sunrpc doesn't
824 need that much accuracy */
826 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
827 set_bit(SK_DATA
, &svsk
->sk_flags
); /* there may be more data... */
830 * Maybe more packets - kick another thread ASAP.
832 svc_sock_received(svsk
);
834 len
= skb
->len
- sizeof(struct udphdr
);
835 rqstp
->rq_arg
.len
= len
;
837 rqstp
->rq_prot
= IPPROTO_UDP
;
839 if (cmh
->cmsg_level
!= IPPROTO_IP
||
840 cmh
->cmsg_type
!= IP_PKTINFO
) {
842 printk("rpcsvc: received unknown control message:"
844 cmh
->cmsg_level
, cmh
->cmsg_type
);
845 skb_free_datagram(svsk
->sk_sk
, skb
);
848 svc_udp_get_dest_address(rqstp
, cmh
);
850 if (skb_is_nonlinear(skb
)) {
851 /* we have to copy */
853 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
856 skb_free_datagram(svsk
->sk_sk
, skb
);
860 skb_free_datagram(svsk
->sk_sk
, skb
);
862 /* we can use it in-place */
863 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+ sizeof(struct udphdr
);
864 rqstp
->rq_arg
.head
[0].iov_len
= len
;
865 if (skb_checksum_complete(skb
)) {
866 skb_free_datagram(svsk
->sk_sk
, skb
);
869 rqstp
->rq_skbuff
= skb
;
872 rqstp
->rq_arg
.page_base
= 0;
873 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
874 rqstp
->rq_arg
.head
[0].iov_len
= len
;
875 rqstp
->rq_arg
.page_len
= 0;
876 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
878 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
879 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
880 (rqstp
->rq_arg
.page_len
+ PAGE_SIZE
- 1)/ PAGE_SIZE
;
884 serv
->sv_stats
->netudpcnt
++;
890 svc_udp_sendto(struct svc_rqst
*rqstp
)
894 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
895 if (error
== -ECONNREFUSED
)
896 /* ICMP error on earlier request. */
897 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
903 svc_udp_init(struct svc_sock
*svsk
)
908 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
909 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
910 svsk
->sk_recvfrom
= svc_udp_recvfrom
;
911 svsk
->sk_sendto
= svc_udp_sendto
;
913 /* initialise setting must have enough space to
914 * receive and respond to one request.
915 * svc_udp_recvfrom will re-adjust if necessary
917 svc_sock_setbufsize(svsk
->sk_sock
,
918 3 * svsk
->sk_server
->sv_max_mesg
,
919 3 * svsk
->sk_server
->sv_max_mesg
);
921 set_bit(SK_DATA
, &svsk
->sk_flags
); /* might have come in before data_ready set up */
922 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
926 /* make sure we get destination address info */
927 svsk
->sk_sock
->ops
->setsockopt(svsk
->sk_sock
, IPPROTO_IP
, IP_PKTINFO
,
928 (char __user
*)&one
, sizeof(one
));
933 * A data_ready event on a listening socket means there's a connection
934 * pending. Do not use state_change as a substitute for it.
937 svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
939 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
941 dprintk("svc: socket %p TCP (listen) state change %d\n",
945 * This callback may called twice when a new connection
946 * is established as a child socket inherits everything
947 * from a parent LISTEN socket.
948 * 1) data_ready method of the parent socket will be called
949 * when one of child sockets become ESTABLISHED.
950 * 2) data_ready method of the child socket may be called
951 * when it receives data before the socket is accepted.
952 * In case of 2, we should ignore it silently.
954 if (sk
->sk_state
== TCP_LISTEN
) {
956 set_bit(SK_CONN
, &svsk
->sk_flags
);
957 svc_sock_enqueue(svsk
);
959 printk("svc: socket %p: no user data\n", sk
);
962 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
963 wake_up_interruptible_all(sk
->sk_sleep
);
967 * A state change on a connected socket means it's dying or dead.
970 svc_tcp_state_change(struct sock
*sk
)
972 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
974 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
975 sk
, sk
->sk_state
, sk
->sk_user_data
);
978 printk("svc: socket %p: no user data\n", sk
);
980 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
981 svc_sock_enqueue(svsk
);
983 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
984 wake_up_interruptible_all(sk
->sk_sleep
);
988 svc_tcp_data_ready(struct sock
*sk
, int count
)
990 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
992 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
993 sk
, sk
->sk_user_data
);
995 set_bit(SK_DATA
, &svsk
->sk_flags
);
996 svc_sock_enqueue(svsk
);
998 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
999 wake_up_interruptible(sk
->sk_sleep
);
1002 static inline int svc_port_is_privileged(struct sockaddr
*sin
)
1004 switch (sin
->sa_family
) {
1006 return ntohs(((struct sockaddr_in
*)sin
)->sin_port
)
1009 return ntohs(((struct sockaddr_in6
*)sin
)->sin6_port
)
1017 * Accept a TCP connection
1020 svc_tcp_accept(struct svc_sock
*svsk
)
1022 struct sockaddr_storage addr
;
1023 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
1024 struct svc_serv
*serv
= svsk
->sk_server
;
1025 struct socket
*sock
= svsk
->sk_sock
;
1026 struct socket
*newsock
;
1027 struct svc_sock
*newsvsk
;
1029 char buf
[RPC_MAX_ADDRBUFLEN
];
1031 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
1035 clear_bit(SK_CONN
, &svsk
->sk_flags
);
1036 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
1039 printk(KERN_WARNING
"%s: no more sockets!\n",
1041 else if (err
!= -EAGAIN
&& net_ratelimit())
1042 printk(KERN_WARNING
"%s: accept failed (err %d)!\n",
1043 serv
->sv_name
, -err
);
1047 set_bit(SK_CONN
, &svsk
->sk_flags
);
1048 svc_sock_enqueue(svsk
);
1050 err
= kernel_getpeername(newsock
, sin
, &slen
);
1052 if (net_ratelimit())
1053 printk(KERN_WARNING
"%s: peername failed (err %d)!\n",
1054 serv
->sv_name
, -err
);
1055 goto failed
; /* aborted connection or whatever */
1058 /* Ideally, we would want to reject connections from unauthorized
1059 * hosts here, but when we get encryption, the IP of the host won't
1060 * tell us anything. For now just warn about unpriv connections.
1062 if (!svc_port_is_privileged(sin
)) {
1063 dprintk(KERN_WARNING
1064 "%s: connect from unprivileged port: %s\n",
1066 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1068 dprintk("%s: connect from %s\n", serv
->sv_name
,
1069 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1071 /* make sure that a write doesn't block forever when
1074 newsock
->sk
->sk_sndtimeo
= HZ
*30;
1076 if (!(newsvsk
= svc_setup_socket(serv
, newsock
, &err
,
1077 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
))))
1079 memcpy(&newsvsk
->sk_remote
, sin
, slen
);
1080 newsvsk
->sk_remotelen
= slen
;
1081 err
= kernel_getsockname(newsock
, sin
, &slen
);
1082 if (unlikely(err
< 0)) {
1083 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
1084 slen
= offsetof(struct sockaddr
, sa_data
);
1086 memcpy(&newsvsk
->sk_local
, sin
, slen
);
1088 svc_sock_received(newsvsk
);
1090 /* make sure that we don't have too many active connections.
1091 * If we have, something must be dropped.
1093 * There's no point in trying to do random drop here for
1094 * DoS prevention. The NFS clients does 1 reconnect in 15
1095 * seconds. An attacker can easily beat that.
1097 * The only somewhat efficient mechanism would be if drop
1098 * old connections from the same IP first. But right now
1099 * we don't even record the client IP in svc_sock.
1101 if (serv
->sv_tmpcnt
> (serv
->sv_nrthreads
+3)*20) {
1102 struct svc_sock
*svsk
= NULL
;
1103 spin_lock_bh(&serv
->sv_lock
);
1104 if (!list_empty(&serv
->sv_tempsocks
)) {
1105 if (net_ratelimit()) {
1106 /* Try to help the admin */
1107 printk(KERN_NOTICE
"%s: too many open TCP "
1108 "sockets, consider increasing the "
1109 "number of nfsd threads\n",
1112 "%s: last TCP connect from %s\n",
1113 serv
->sv_name
, __svc_print_addr(sin
,
1117 * Always select the oldest socket. It's not fair,
1120 svsk
= list_entry(serv
->sv_tempsocks
.prev
,
1123 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1124 atomic_inc(&svsk
->sk_inuse
);
1126 spin_unlock_bh(&serv
->sv_lock
);
1129 svc_sock_enqueue(svsk
);
1136 serv
->sv_stats
->nettcpconn
++;
1141 sock_release(newsock
);
1146 * Receive data from a TCP socket.
1149 svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1151 struct svc_sock
*svsk
= rqstp
->rq_sock
;
1152 struct svc_serv
*serv
= svsk
->sk_server
;
1157 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1158 svsk
, test_bit(SK_DATA
, &svsk
->sk_flags
),
1159 test_bit(SK_CONN
, &svsk
->sk_flags
),
1160 test_bit(SK_CLOSE
, &svsk
->sk_flags
));
1162 if ((rqstp
->rq_deferred
= svc_deferred_dequeue(svsk
))) {
1163 svc_sock_received(svsk
);
1164 return svc_deferred_recv(rqstp
);
1167 if (test_bit(SK_CLOSE
, &svsk
->sk_flags
)) {
1168 svc_delete_socket(svsk
);
1172 if (svsk
->sk_sk
->sk_state
== TCP_LISTEN
) {
1173 svc_tcp_accept(svsk
);
1174 svc_sock_received(svsk
);
1178 if (test_and_clear_bit(SK_CHNGBUF
, &svsk
->sk_flags
))
1179 /* sndbuf needs to have room for one request
1180 * per thread, otherwise we can stall even when the
1181 * network isn't a bottleneck.
1183 * We count all threads rather than threads in a
1184 * particular pool, which provides an upper bound
1185 * on the number of threads which will access the socket.
1187 * rcvbuf just needs to be able to hold a few requests.
1188 * Normally they will be removed from the queue
1189 * as soon a a complete request arrives.
1191 svc_sock_setbufsize(svsk
->sk_sock
,
1192 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
1193 3 * serv
->sv_max_mesg
);
1195 clear_bit(SK_DATA
, &svsk
->sk_flags
);
1197 /* Receive data. If we haven't got the record length yet, get
1198 * the next four bytes. Otherwise try to gobble up as much as
1199 * possible up to the complete record length.
1201 if (svsk
->sk_tcplen
< 4) {
1202 unsigned long want
= 4 - svsk
->sk_tcplen
;
1205 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
1207 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
1209 svsk
->sk_tcplen
+= len
;
1212 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
1214 svc_sock_received(svsk
);
1215 return -EAGAIN
; /* record header not complete */
1218 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
1219 if (!(svsk
->sk_reclen
& 0x80000000)) {
1220 /* FIXME: technically, a record can be fragmented,
1221 * and non-terminal fragments will not have the top
1222 * bit set in the fragment length header.
1223 * But apparently no known nfs clients send fragmented
1225 if (net_ratelimit())
1226 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx"
1227 " (non-terminal)\n",
1228 (unsigned long) svsk
->sk_reclen
);
1231 svsk
->sk_reclen
&= 0x7fffffff;
1232 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
1233 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
1234 if (net_ratelimit())
1235 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx"
1237 (unsigned long) svsk
->sk_reclen
);
1242 /* Check whether enough data is available */
1243 len
= svc_recv_available(svsk
);
1247 if (len
< svsk
->sk_reclen
) {
1248 dprintk("svc: incomplete TCP record (%d of %d)\n",
1249 len
, svsk
->sk_reclen
);
1250 svc_sock_received(svsk
);
1251 return -EAGAIN
; /* record not complete */
1253 len
= svsk
->sk_reclen
;
1254 set_bit(SK_DATA
, &svsk
->sk_flags
);
1256 vec
= rqstp
->rq_vec
;
1257 vec
[0] = rqstp
->rq_arg
.head
[0];
1260 while (vlen
< len
) {
1261 vec
[pnum
].iov_base
= page_address(rqstp
->rq_pages
[pnum
]);
1262 vec
[pnum
].iov_len
= PAGE_SIZE
;
1266 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1268 /* Now receive data */
1269 len
= svc_recvfrom(rqstp
, vec
, pnum
, len
);
1273 dprintk("svc: TCP complete record (%d bytes)\n", len
);
1274 rqstp
->rq_arg
.len
= len
;
1275 rqstp
->rq_arg
.page_base
= 0;
1276 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1277 rqstp
->rq_arg
.head
[0].iov_len
= len
;
1278 rqstp
->rq_arg
.page_len
= 0;
1280 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
1283 rqstp
->rq_skbuff
= NULL
;
1284 rqstp
->rq_prot
= IPPROTO_TCP
;
1286 /* Reset TCP read info */
1287 svsk
->sk_reclen
= 0;
1288 svsk
->sk_tcplen
= 0;
1290 svc_sock_received(svsk
);
1292 serv
->sv_stats
->nettcpcnt
++;
1297 svc_delete_socket(svsk
);
1301 if (len
== -EAGAIN
) {
1302 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1303 svc_sock_received(svsk
);
1305 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1306 svsk
->sk_server
->sv_name
, -len
);
1314 * Send out data on TCP socket.
1317 svc_tcp_sendto(struct svc_rqst
*rqstp
)
1319 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1323 /* Set up the first element of the reply kvec.
1324 * Any other kvecs that may be in use have been taken
1325 * care of by the server implementation itself.
1327 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1328 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1330 if (test_bit(SK_DEAD
, &rqstp
->rq_sock
->sk_flags
))
1333 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1334 if (sent
!= xbufp
->len
) {
1335 printk(KERN_NOTICE
"rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1336 rqstp
->rq_sock
->sk_server
->sv_name
,
1337 (sent
<0)?"got error":"sent only",
1339 set_bit(SK_CLOSE
, &rqstp
->rq_sock
->sk_flags
);
1340 svc_sock_enqueue(rqstp
->rq_sock
);
1347 svc_tcp_init(struct svc_sock
*svsk
)
1349 struct sock
*sk
= svsk
->sk_sk
;
1350 struct tcp_sock
*tp
= tcp_sk(sk
);
1352 svsk
->sk_recvfrom
= svc_tcp_recvfrom
;
1353 svsk
->sk_sendto
= svc_tcp_sendto
;
1355 if (sk
->sk_state
== TCP_LISTEN
) {
1356 dprintk("setting up TCP socket for listening\n");
1357 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1358 set_bit(SK_CONN
, &svsk
->sk_flags
);
1360 dprintk("setting up TCP socket for reading\n");
1361 sk
->sk_state_change
= svc_tcp_state_change
;
1362 sk
->sk_data_ready
= svc_tcp_data_ready
;
1363 sk
->sk_write_space
= svc_write_space
;
1365 svsk
->sk_reclen
= 0;
1366 svsk
->sk_tcplen
= 0;
1368 tp
->nonagle
= 1; /* disable Nagle's algorithm */
1370 /* initialise setting must have enough space to
1371 * receive and respond to one request.
1372 * svc_tcp_recvfrom will re-adjust if necessary
1374 svc_sock_setbufsize(svsk
->sk_sock
,
1375 3 * svsk
->sk_server
->sv_max_mesg
,
1376 3 * svsk
->sk_server
->sv_max_mesg
);
1378 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1379 set_bit(SK_DATA
, &svsk
->sk_flags
);
1380 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1381 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1386 svc_sock_update_bufs(struct svc_serv
*serv
)
1389 * The number of server threads has changed. Update
1390 * rcvbuf and sndbuf accordingly on all sockets
1392 struct list_head
*le
;
1394 spin_lock_bh(&serv
->sv_lock
);
1395 list_for_each(le
, &serv
->sv_permsocks
) {
1396 struct svc_sock
*svsk
=
1397 list_entry(le
, struct svc_sock
, sk_list
);
1398 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1400 list_for_each(le
, &serv
->sv_tempsocks
) {
1401 struct svc_sock
*svsk
=
1402 list_entry(le
, struct svc_sock
, sk_list
);
1403 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1405 spin_unlock_bh(&serv
->sv_lock
);
1409 * Receive the next request on any socket. This code is carefully
1410 * organised not to touch any cachelines in the shared svc_serv
1411 * structure, only cachelines in the local svc_pool.
1414 svc_recv(struct svc_rqst
*rqstp
, long timeout
)
1416 struct svc_sock
*svsk
= NULL
;
1417 struct svc_serv
*serv
= rqstp
->rq_server
;
1418 struct svc_pool
*pool
= rqstp
->rq_pool
;
1421 struct xdr_buf
*arg
;
1422 DECLARE_WAITQUEUE(wait
, current
);
1424 dprintk("svc: server %p waiting for data (to = %ld)\n",
1429 "svc_recv: service %p, socket not NULL!\n",
1431 if (waitqueue_active(&rqstp
->rq_wait
))
1433 "svc_recv: service %p, wait queue active!\n",
1437 /* now allocate needed pages. If we get a failure, sleep briefly */
1438 pages
= (serv
->sv_max_mesg
+ PAGE_SIZE
) / PAGE_SIZE
;
1439 for (i
=0; i
< pages
; i
++)
1440 while (rqstp
->rq_pages
[i
] == NULL
) {
1441 struct page
*p
= alloc_page(GFP_KERNEL
);
1443 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1444 rqstp
->rq_pages
[i
] = p
;
1446 rqstp
->rq_pages
[i
++] = NULL
; /* this might be seen in nfs_read_actor */
1447 BUG_ON(pages
>= RPCSVC_MAXPAGES
);
1449 /* Make arg->head point to first page and arg->pages point to rest */
1450 arg
= &rqstp
->rq_arg
;
1451 arg
->head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
1452 arg
->head
[0].iov_len
= PAGE_SIZE
;
1453 arg
->pages
= rqstp
->rq_pages
+ 1;
1455 /* save at least one page for response */
1456 arg
->page_len
= (pages
-2)*PAGE_SIZE
;
1457 arg
->len
= (pages
-1)*PAGE_SIZE
;
1458 arg
->tail
[0].iov_len
= 0;
1465 spin_lock_bh(&pool
->sp_lock
);
1466 if ((svsk
= svc_sock_dequeue(pool
)) != NULL
) {
1467 rqstp
->rq_sock
= svsk
;
1468 atomic_inc(&svsk
->sk_inuse
);
1469 rqstp
->rq_reserved
= serv
->sv_max_mesg
;
1470 atomic_add(rqstp
->rq_reserved
, &svsk
->sk_reserved
);
1472 /* No data pending. Go to sleep */
1473 svc_thread_enqueue(pool
, rqstp
);
1476 * We have to be able to interrupt this wait
1477 * to bring down the daemons ...
1479 set_current_state(TASK_INTERRUPTIBLE
);
1480 add_wait_queue(&rqstp
->rq_wait
, &wait
);
1481 spin_unlock_bh(&pool
->sp_lock
);
1483 schedule_timeout(timeout
);
1487 spin_lock_bh(&pool
->sp_lock
);
1488 remove_wait_queue(&rqstp
->rq_wait
, &wait
);
1490 if (!(svsk
= rqstp
->rq_sock
)) {
1491 svc_thread_dequeue(pool
, rqstp
);
1492 spin_unlock_bh(&pool
->sp_lock
);
1493 dprintk("svc: server %p, no data yet\n", rqstp
);
1494 return signalled()? -EINTR
: -EAGAIN
;
1497 spin_unlock_bh(&pool
->sp_lock
);
1499 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
1500 rqstp
, pool
->sp_id
, svsk
, atomic_read(&svsk
->sk_inuse
));
1501 len
= svsk
->sk_recvfrom(rqstp
);
1502 dprintk("svc: got len=%d\n", len
);
1504 /* No data, incomplete (TCP) read, or accept() */
1505 if (len
== 0 || len
== -EAGAIN
) {
1506 rqstp
->rq_res
.len
= 0;
1507 svc_sock_release(rqstp
);
1510 svsk
->sk_lastrecv
= get_seconds();
1511 clear_bit(SK_OLD
, &svsk
->sk_flags
);
1513 rqstp
->rq_secure
= svc_port_is_privileged(svc_addr(rqstp
));
1514 rqstp
->rq_chandle
.defer
= svc_defer
;
1517 serv
->sv_stats
->netcnt
++;
1525 svc_drop(struct svc_rqst
*rqstp
)
1527 dprintk("svc: socket %p dropped request\n", rqstp
->rq_sock
);
1528 svc_sock_release(rqstp
);
1532 * Return reply to client.
1535 svc_send(struct svc_rqst
*rqstp
)
1537 struct svc_sock
*svsk
;
1541 if ((svsk
= rqstp
->rq_sock
) == NULL
) {
1542 printk(KERN_WARNING
"NULL socket pointer in %s:%d\n",
1543 __FILE__
, __LINE__
);
1547 /* release the receive skb before sending the reply */
1548 svc_release_skb(rqstp
);
1550 /* calculate over-all length */
1551 xb
= & rqstp
->rq_res
;
1552 xb
->len
= xb
->head
[0].iov_len
+
1554 xb
->tail
[0].iov_len
;
1556 /* Grab svsk->sk_mutex to serialize outgoing data. */
1557 mutex_lock(&svsk
->sk_mutex
);
1558 if (test_bit(SK_DEAD
, &svsk
->sk_flags
))
1561 len
= svsk
->sk_sendto(rqstp
);
1562 mutex_unlock(&svsk
->sk_mutex
);
1563 svc_sock_release(rqstp
);
1565 if (len
== -ECONNREFUSED
|| len
== -ENOTCONN
|| len
== -EAGAIN
)
1571 * Timer function to close old temporary sockets, using
1572 * a mark-and-sweep algorithm.
1575 svc_age_temp_sockets(unsigned long closure
)
1577 struct svc_serv
*serv
= (struct svc_serv
*)closure
;
1578 struct svc_sock
*svsk
;
1579 struct list_head
*le
, *next
;
1580 LIST_HEAD(to_be_aged
);
1582 dprintk("svc_age_temp_sockets\n");
1584 if (!spin_trylock_bh(&serv
->sv_lock
)) {
1585 /* busy, try again 1 sec later */
1586 dprintk("svc_age_temp_sockets: busy\n");
1587 mod_timer(&serv
->sv_temptimer
, jiffies
+ HZ
);
1591 list_for_each_safe(le
, next
, &serv
->sv_tempsocks
) {
1592 svsk
= list_entry(le
, struct svc_sock
, sk_list
);
1594 if (!test_and_set_bit(SK_OLD
, &svsk
->sk_flags
))
1596 if (atomic_read(&svsk
->sk_inuse
) > 1 || test_bit(SK_BUSY
, &svsk
->sk_flags
))
1598 atomic_inc(&svsk
->sk_inuse
);
1599 list_move(le
, &to_be_aged
);
1600 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1601 set_bit(SK_DETACHED
, &svsk
->sk_flags
);
1603 spin_unlock_bh(&serv
->sv_lock
);
1605 while (!list_empty(&to_be_aged
)) {
1606 le
= to_be_aged
.next
;
1607 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */
1609 svsk
= list_entry(le
, struct svc_sock
, sk_list
);
1611 dprintk("queuing svsk %p for closing, %lu seconds old\n",
1612 svsk
, get_seconds() - svsk
->sk_lastrecv
);
1614 /* a thread will dequeue and close it soon */
1615 svc_sock_enqueue(svsk
);
1619 mod_timer(&serv
->sv_temptimer
, jiffies
+ svc_conn_age_period
* HZ
);
1623 * Initialize socket for RPC use and create svc_sock struct
1624 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1626 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1627 struct socket
*sock
,
1628 int *errp
, int flags
)
1630 struct svc_sock
*svsk
;
1632 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1633 int is_temporary
= flags
& SVC_SOCK_TEMPORARY
;
1635 dprintk("svc: svc_setup_socket %p\n", sock
);
1636 if (!(svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
))) {
1643 /* Register socket with portmapper */
1644 if (*errp
>= 0 && pmap_register
)
1645 *errp
= svc_register(serv
, inet
->sk_protocol
,
1646 ntohs(inet_sk(inet
)->sport
));
1653 set_bit(SK_BUSY
, &svsk
->sk_flags
);
1654 inet
->sk_user_data
= svsk
;
1655 svsk
->sk_sock
= sock
;
1657 svsk
->sk_ostate
= inet
->sk_state_change
;
1658 svsk
->sk_odata
= inet
->sk_data_ready
;
1659 svsk
->sk_owspace
= inet
->sk_write_space
;
1660 svsk
->sk_server
= serv
;
1661 atomic_set(&svsk
->sk_inuse
, 1);
1662 svsk
->sk_lastrecv
= get_seconds();
1663 spin_lock_init(&svsk
->sk_lock
);
1664 INIT_LIST_HEAD(&svsk
->sk_deferred
);
1665 INIT_LIST_HEAD(&svsk
->sk_ready
);
1666 mutex_init(&svsk
->sk_mutex
);
1668 /* Initialize the socket */
1669 if (sock
->type
== SOCK_DGRAM
)
1674 spin_lock_bh(&serv
->sv_lock
);
1676 set_bit(SK_TEMP
, &svsk
->sk_flags
);
1677 list_add(&svsk
->sk_list
, &serv
->sv_tempsocks
);
1679 if (serv
->sv_temptimer
.function
== NULL
) {
1680 /* setup timer to age temp sockets */
1681 setup_timer(&serv
->sv_temptimer
, svc_age_temp_sockets
,
1682 (unsigned long)serv
);
1683 mod_timer(&serv
->sv_temptimer
,
1684 jiffies
+ svc_conn_age_period
* HZ
);
1687 clear_bit(SK_TEMP
, &svsk
->sk_flags
);
1688 list_add(&svsk
->sk_list
, &serv
->sv_permsocks
);
1690 spin_unlock_bh(&serv
->sv_lock
);
1692 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1698 int svc_addsock(struct svc_serv
*serv
,
1704 struct socket
*so
= sockfd_lookup(fd
, &err
);
1705 struct svc_sock
*svsk
= NULL
;
1709 if (so
->sk
->sk_family
!= AF_INET
)
1710 err
= -EAFNOSUPPORT
;
1711 else if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1712 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1713 err
= -EPROTONOSUPPORT
;
1714 else if (so
->state
> SS_UNCONNECTED
)
1717 svsk
= svc_setup_socket(serv
, so
, &err
, SVC_SOCK_DEFAULTS
);
1719 svc_sock_received(svsk
);
1727 if (proto
) *proto
= so
->sk
->sk_protocol
;
1728 return one_sock_name(name_return
, svsk
);
1730 EXPORT_SYMBOL_GPL(svc_addsock
);
1733 * Create socket for RPC service.
1735 static int svc_create_socket(struct svc_serv
*serv
, int protocol
,
1736 struct sockaddr
*sin
, int len
, int flags
)
1738 struct svc_sock
*svsk
;
1739 struct socket
*sock
;
1742 char buf
[RPC_MAX_ADDRBUFLEN
];
1744 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1745 serv
->sv_program
->pg_name
, protocol
,
1746 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1748 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1749 printk(KERN_WARNING
"svc: only UDP and TCP "
1750 "sockets supported\n");
1753 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1755 error
= sock_create_kern(sin
->sa_family
, type
, protocol
, &sock
);
1759 svc_reclassify_socket(sock
);
1761 if (type
== SOCK_STREAM
)
1762 sock
->sk
->sk_reuse
= 1; /* allow address reuse */
1763 error
= kernel_bind(sock
, sin
, len
);
1767 if (protocol
== IPPROTO_TCP
) {
1768 if ((error
= kernel_listen(sock
, 64)) < 0)
1772 if ((svsk
= svc_setup_socket(serv
, sock
, &error
, flags
)) != NULL
) {
1773 svc_sock_received(svsk
);
1774 return ntohs(inet_sk(svsk
->sk_sk
)->sport
);
1778 dprintk("svc: svc_create_socket error = %d\n", -error
);
1784 * Remove a dead socket
1787 svc_delete_socket(struct svc_sock
*svsk
)
1789 struct svc_serv
*serv
;
1792 dprintk("svc: svc_delete_socket(%p)\n", svsk
);
1794 serv
= svsk
->sk_server
;
1797 sk
->sk_state_change
= svsk
->sk_ostate
;
1798 sk
->sk_data_ready
= svsk
->sk_odata
;
1799 sk
->sk_write_space
= svsk
->sk_owspace
;
1801 spin_lock_bh(&serv
->sv_lock
);
1803 if (!test_and_set_bit(SK_DETACHED
, &svsk
->sk_flags
))
1804 list_del_init(&svsk
->sk_list
);
1806 * We used to delete the svc_sock from whichever list
1807 * it's sk_ready node was on, but we don't actually
1808 * need to. This is because the only time we're called
1809 * while still attached to a queue, the queue itself
1810 * is about to be destroyed (in svc_destroy).
1812 if (!test_and_set_bit(SK_DEAD
, &svsk
->sk_flags
)) {
1813 BUG_ON(atomic_read(&svsk
->sk_inuse
)<2);
1814 atomic_dec(&svsk
->sk_inuse
);
1815 if (test_bit(SK_TEMP
, &svsk
->sk_flags
))
1819 spin_unlock_bh(&serv
->sv_lock
);
1822 static void svc_close_socket(struct svc_sock
*svsk
)
1824 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1825 if (test_and_set_bit(SK_BUSY
, &svsk
->sk_flags
))
1826 /* someone else will have to effect the close */
1829 atomic_inc(&svsk
->sk_inuse
);
1830 svc_delete_socket(svsk
);
1831 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
1835 void svc_force_close_socket(struct svc_sock
*svsk
)
1837 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1838 if (test_bit(SK_BUSY
, &svsk
->sk_flags
)) {
1839 /* Waiting to be processed, but no threads left,
1840 * So just remove it from the waiting list
1842 list_del_init(&svsk
->sk_ready
);
1843 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
1845 svc_close_socket(svsk
);
1849 * svc_makesock - Make a socket for nfsd and lockd
1850 * @serv: RPC server structure
1851 * @protocol: transport protocol to use
1852 * @port: port to use
1853 * @flags: requested socket characteristics
1856 int svc_makesock(struct svc_serv
*serv
, int protocol
, unsigned short port
,
1859 struct sockaddr_in sin
= {
1860 .sin_family
= AF_INET
,
1861 .sin_addr
.s_addr
= INADDR_ANY
,
1862 .sin_port
= htons(port
),
1865 dprintk("svc: creating socket proto = %d\n", protocol
);
1866 return svc_create_socket(serv
, protocol
, (struct sockaddr
*) &sin
,
1867 sizeof(sin
), flags
);
1871 * Handle defer and revisit of requests
1874 static void svc_revisit(struct cache_deferred_req
*dreq
, int too_many
)
1876 struct svc_deferred_req
*dr
= container_of(dreq
, struct svc_deferred_req
, handle
);
1877 struct svc_sock
*svsk
;
1880 svc_sock_put(dr
->svsk
);
1884 dprintk("revisit queued\n");
1887 spin_lock(&svsk
->sk_lock
);
1888 list_add(&dr
->handle
.recent
, &svsk
->sk_deferred
);
1889 spin_unlock(&svsk
->sk_lock
);
1890 set_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1891 svc_sock_enqueue(svsk
);
1895 static struct cache_deferred_req
*
1896 svc_defer(struct cache_req
*req
)
1898 struct svc_rqst
*rqstp
= container_of(req
, struct svc_rqst
, rq_chandle
);
1899 int size
= sizeof(struct svc_deferred_req
) + (rqstp
->rq_arg
.len
);
1900 struct svc_deferred_req
*dr
;
1902 if (rqstp
->rq_arg
.page_len
)
1903 return NULL
; /* if more than a page, give up FIXME */
1904 if (rqstp
->rq_deferred
) {
1905 dr
= rqstp
->rq_deferred
;
1906 rqstp
->rq_deferred
= NULL
;
1908 int skip
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1909 /* FIXME maybe discard if size too large */
1910 dr
= kmalloc(size
, GFP_KERNEL
);
1914 dr
->handle
.owner
= rqstp
->rq_server
;
1915 dr
->prot
= rqstp
->rq_prot
;
1916 memcpy(&dr
->addr
, &rqstp
->rq_addr
, rqstp
->rq_addrlen
);
1917 dr
->addrlen
= rqstp
->rq_addrlen
;
1918 dr
->daddr
= rqstp
->rq_daddr
;
1919 dr
->argslen
= rqstp
->rq_arg
.len
>> 2;
1920 memcpy(dr
->args
, rqstp
->rq_arg
.head
[0].iov_base
-skip
, dr
->argslen
<<2);
1922 atomic_inc(&rqstp
->rq_sock
->sk_inuse
);
1923 dr
->svsk
= rqstp
->rq_sock
;
1925 dr
->handle
.revisit
= svc_revisit
;
1930 * recv data from a deferred request into an active one
1932 static int svc_deferred_recv(struct svc_rqst
*rqstp
)
1934 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
1936 rqstp
->rq_arg
.head
[0].iov_base
= dr
->args
;
1937 rqstp
->rq_arg
.head
[0].iov_len
= dr
->argslen
<<2;
1938 rqstp
->rq_arg
.page_len
= 0;
1939 rqstp
->rq_arg
.len
= dr
->argslen
<<2;
1940 rqstp
->rq_prot
= dr
->prot
;
1941 memcpy(&rqstp
->rq_addr
, &dr
->addr
, dr
->addrlen
);
1942 rqstp
->rq_addrlen
= dr
->addrlen
;
1943 rqstp
->rq_daddr
= dr
->daddr
;
1944 rqstp
->rq_respages
= rqstp
->rq_pages
;
1945 return dr
->argslen
<<2;
1949 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_sock
*svsk
)
1951 struct svc_deferred_req
*dr
= NULL
;
1953 if (!test_bit(SK_DEFERRED
, &svsk
->sk_flags
))
1955 spin_lock(&svsk
->sk_lock
);
1956 clear_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1957 if (!list_empty(&svsk
->sk_deferred
)) {
1958 dr
= list_entry(svsk
->sk_deferred
.next
,
1959 struct svc_deferred_req
,
1961 list_del_init(&dr
->handle
.recent
);
1962 set_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1964 spin_unlock(&svsk
->sk_lock
);