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/suspend.h>
36 #include <net/checksum.h>
39 #include <asm/uaccess.h>
40 #include <asm/ioctls.h>
42 #include <linux/sunrpc/types.h>
43 #include <linux/sunrpc/xdr.h>
44 #include <linux/sunrpc/svcsock.h>
45 #include <linux/sunrpc/stats.h>
47 /* SMP locking strategy:
49 * svc_serv->sv_lock protects most stuff for that service.
51 * Some flags can be set to certain values at any time
52 * providing that certain rules are followed:
54 * SK_BUSY can be set to 0 at any time.
55 * svc_sock_enqueue must be called afterwards
56 * SK_CONN, SK_DATA, can be set or cleared at any time.
57 * after a set, svc_sock_enqueue must be called.
58 * after a clear, the socket must be read/accepted
59 * if this succeeds, it must be set again.
60 * SK_CLOSE can set at any time. It is never cleared.
64 #define RPCDBG_FACILITY RPCDBG_SVCSOCK
67 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
68 int *errp
, int pmap_reg
);
69 static void svc_udp_data_ready(struct sock
*, int);
70 static int svc_udp_recvfrom(struct svc_rqst
*);
71 static int svc_udp_sendto(struct svc_rqst
*);
73 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_sock
*svsk
);
74 static int svc_deferred_recv(struct svc_rqst
*rqstp
);
75 static struct cache_deferred_req
*svc_defer(struct cache_req
*req
);
78 * Queue up an idle server thread. Must have serv->sv_lock held.
79 * Note: this is really a stack rather than a queue, so that we only
80 * use as many different threads as we need, and the rest don't polute
84 svc_serv_enqueue(struct svc_serv
*serv
, struct svc_rqst
*rqstp
)
86 list_add(&rqstp
->rq_list
, &serv
->sv_threads
);
90 * Dequeue an nfsd thread. Must have serv->sv_lock held.
93 svc_serv_dequeue(struct svc_serv
*serv
, struct svc_rqst
*rqstp
)
95 list_del(&rqstp
->rq_list
);
99 * Release an skbuff after use
102 svc_release_skb(struct svc_rqst
*rqstp
)
104 struct sk_buff
*skb
= rqstp
->rq_skbuff
;
105 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
108 rqstp
->rq_skbuff
= NULL
;
110 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
111 skb_free_datagram(rqstp
->rq_sock
->sk_sk
, skb
);
114 rqstp
->rq_deferred
= NULL
;
120 * Any space to write?
122 static inline unsigned long
123 svc_sock_wspace(struct svc_sock
*svsk
)
127 if (svsk
->sk_sock
->type
== SOCK_STREAM
)
128 wspace
= sk_stream_wspace(svsk
->sk_sk
);
130 wspace
= sock_wspace(svsk
->sk_sk
);
136 * Queue up a socket with data pending. If there are idle nfsd
137 * processes, wake 'em up.
141 svc_sock_enqueue(struct svc_sock
*svsk
)
143 struct svc_serv
*serv
= svsk
->sk_server
;
144 struct svc_rqst
*rqstp
;
146 if (!(svsk
->sk_flags
&
147 ( (1<<SK_CONN
)|(1<<SK_DATA
)|(1<<SK_CLOSE
)|(1<<SK_DEFERRED
)) ))
149 if (test_bit(SK_DEAD
, &svsk
->sk_flags
))
152 spin_lock_bh(&serv
->sv_lock
);
154 if (!list_empty(&serv
->sv_threads
) &&
155 !list_empty(&serv
->sv_sockets
))
157 "svc_sock_enqueue: threads and sockets both waiting??\n");
159 if (test_bit(SK_DEAD
, &svsk
->sk_flags
)) {
160 /* Don't enqueue dead sockets */
161 dprintk("svc: socket %p is dead, not enqueued\n", svsk
->sk_sk
);
165 if (test_bit(SK_BUSY
, &svsk
->sk_flags
)) {
166 /* Don't enqueue socket while daemon is receiving */
167 dprintk("svc: socket %p busy, not enqueued\n", svsk
->sk_sk
);
171 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
172 if (((svsk
->sk_reserved
+ serv
->sv_bufsz
)*2
173 > svc_sock_wspace(svsk
))
174 && !test_bit(SK_CLOSE
, &svsk
->sk_flags
)
175 && !test_bit(SK_CONN
, &svsk
->sk_flags
)) {
176 /* Don't enqueue while not enough space for reply */
177 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n",
178 svsk
->sk_sk
, svsk
->sk_reserved
+serv
->sv_bufsz
,
179 svc_sock_wspace(svsk
));
182 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
184 /* Mark socket as busy. It will remain in this state until the
185 * server has processed all pending data and put the socket back
188 set_bit(SK_BUSY
, &svsk
->sk_flags
);
190 if (!list_empty(&serv
->sv_threads
)) {
191 rqstp
= list_entry(serv
->sv_threads
.next
,
194 dprintk("svc: socket %p served by daemon %p\n",
196 svc_serv_dequeue(serv
, rqstp
);
199 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
200 rqstp
, rqstp
->rq_sock
);
201 rqstp
->rq_sock
= svsk
;
203 rqstp
->rq_reserved
= serv
->sv_bufsz
;
204 svsk
->sk_reserved
+= rqstp
->rq_reserved
;
205 wake_up(&rqstp
->rq_wait
);
207 dprintk("svc: socket %p put into queue\n", svsk
->sk_sk
);
208 list_add_tail(&svsk
->sk_ready
, &serv
->sv_sockets
);
212 spin_unlock_bh(&serv
->sv_lock
);
216 * Dequeue the first socket. Must be called with the serv->sv_lock held.
218 static inline struct svc_sock
*
219 svc_sock_dequeue(struct svc_serv
*serv
)
221 struct svc_sock
*svsk
;
223 if (list_empty(&serv
->sv_sockets
))
226 svsk
= list_entry(serv
->sv_sockets
.next
,
227 struct svc_sock
, sk_ready
);
228 list_del_init(&svsk
->sk_ready
);
230 dprintk("svc: socket %p dequeued, inuse=%d\n",
231 svsk
->sk_sk
, svsk
->sk_inuse
);
237 * Having read something from a socket, check whether it
238 * needs to be re-enqueued.
239 * Note: SK_DATA only gets cleared when a read-attempt finds
240 * no (or insufficient) data.
243 svc_sock_received(struct svc_sock
*svsk
)
245 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
246 svc_sock_enqueue(svsk
);
251 * svc_reserve - change the space reserved for the reply to a request.
252 * @rqstp: The request in question
253 * @space: new max space to reserve
255 * Each request reserves some space on the output queue of the socket
256 * to make sure the reply fits. This function reduces that reserved
257 * space to be the amount of space used already, plus @space.
260 void svc_reserve(struct svc_rqst
*rqstp
, int space
)
262 space
+= rqstp
->rq_res
.head
[0].iov_len
;
264 if (space
< rqstp
->rq_reserved
) {
265 struct svc_sock
*svsk
= rqstp
->rq_sock
;
266 spin_lock_bh(&svsk
->sk_server
->sv_lock
);
267 svsk
->sk_reserved
-= (rqstp
->rq_reserved
- space
);
268 rqstp
->rq_reserved
= space
;
269 spin_unlock_bh(&svsk
->sk_server
->sv_lock
);
271 svc_sock_enqueue(svsk
);
276 * Release a socket after use.
279 svc_sock_put(struct svc_sock
*svsk
)
281 struct svc_serv
*serv
= svsk
->sk_server
;
283 spin_lock_bh(&serv
->sv_lock
);
284 if (!--(svsk
->sk_inuse
) && test_bit(SK_DEAD
, &svsk
->sk_flags
)) {
285 spin_unlock_bh(&serv
->sv_lock
);
286 dprintk("svc: releasing dead socket\n");
287 sock_release(svsk
->sk_sock
);
291 spin_unlock_bh(&serv
->sv_lock
);
295 svc_sock_release(struct svc_rqst
*rqstp
)
297 struct svc_sock
*svsk
= rqstp
->rq_sock
;
299 svc_release_skb(rqstp
);
301 svc_free_allpages(rqstp
);
302 rqstp
->rq_res
.page_len
= 0;
303 rqstp
->rq_res
.page_base
= 0;
306 /* Reset response buffer and release
308 * But first, check that enough space was reserved
309 * for the reply, otherwise we have a bug!
311 if ((rqstp
->rq_res
.len
) > rqstp
->rq_reserved
)
312 printk(KERN_ERR
"RPC request reserved %d but used %d\n",
316 rqstp
->rq_res
.head
[0].iov_len
= 0;
317 svc_reserve(rqstp
, 0);
318 rqstp
->rq_sock
= NULL
;
324 * External function to wake up a server waiting for data
327 svc_wake_up(struct svc_serv
*serv
)
329 struct svc_rqst
*rqstp
;
331 spin_lock_bh(&serv
->sv_lock
);
332 if (!list_empty(&serv
->sv_threads
)) {
333 rqstp
= list_entry(serv
->sv_threads
.next
,
336 dprintk("svc: daemon %p woken up.\n", rqstp
);
338 svc_serv_dequeue(serv, rqstp);
339 rqstp->rq_sock = NULL;
341 wake_up(&rqstp
->rq_wait
);
343 spin_unlock_bh(&serv
->sv_lock
);
347 * Generic sendto routine
350 svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
352 struct svc_sock
*svsk
= rqstp
->rq_sock
;
353 struct socket
*sock
= svsk
->sk_sock
;
355 char buffer
[CMSG_SPACE(sizeof(struct in_pktinfo
))];
356 struct cmsghdr
*cmh
= (struct cmsghdr
*)buffer
;
357 struct in_pktinfo
*pki
= (struct in_pktinfo
*)CMSG_DATA(cmh
);
361 struct page
**ppage
= xdr
->pages
;
362 size_t base
= xdr
->page_base
;
363 unsigned int pglen
= xdr
->page_len
;
364 unsigned int flags
= MSG_MORE
;
368 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
369 /* set the source and destination */
371 msg
.msg_name
= &rqstp
->rq_addr
;
372 msg
.msg_namelen
= sizeof(rqstp
->rq_addr
);
375 msg
.msg_flags
= MSG_MORE
;
377 msg
.msg_control
= cmh
;
378 msg
.msg_controllen
= sizeof(buffer
);
379 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
380 cmh
->cmsg_level
= SOL_IP
;
381 cmh
->cmsg_type
= IP_PKTINFO
;
382 pki
->ipi_ifindex
= 0;
383 pki
->ipi_spec_dst
.s_addr
= rqstp
->rq_daddr
;
385 if (sock_sendmsg(sock
, &msg
, 0) < 0)
390 if (slen
== xdr
->head
[0].iov_len
)
392 len
= sock
->ops
->sendpage(sock
, rqstp
->rq_respages
[0], 0, xdr
->head
[0].iov_len
, flags
);
393 if (len
!= xdr
->head
[0].iov_len
)
395 slen
-= xdr
->head
[0].iov_len
;
400 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
404 result
= sock
->ops
->sendpage(sock
, *ppage
, base
, size
, flags
);
411 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
416 if (xdr
->tail
[0].iov_len
) {
417 result
= sock
->ops
->sendpage(sock
, rqstp
->rq_respages
[rqstp
->rq_restailpage
],
418 ((unsigned long)xdr
->tail
[0].iov_base
)& (PAGE_SIZE
-1),
419 xdr
->tail
[0].iov_len
, 0);
425 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %x)\n",
426 rqstp
->rq_sock
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
, xdr
->len
, len
,
427 rqstp
->rq_addr
.sin_addr
.s_addr
);
433 * Check input queue length
436 svc_recv_available(struct svc_sock
*svsk
)
439 struct socket
*sock
= svsk
->sk_sock
;
442 oldfs
= get_fs(); set_fs(KERNEL_DS
);
443 err
= sock
->ops
->ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
446 return (err
>= 0)? avail
: err
;
450 * Generic recvfrom routine.
453 svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
, int buflen
)
459 rqstp
->rq_addrlen
= sizeof(rqstp
->rq_addr
);
460 sock
= rqstp
->rq_sock
->sk_sock
;
462 msg
.msg_name
= &rqstp
->rq_addr
;
463 msg
.msg_namelen
= sizeof(rqstp
->rq_addr
);
464 msg
.msg_control
= NULL
;
465 msg
.msg_controllen
= 0;
467 msg
.msg_flags
= MSG_DONTWAIT
;
469 len
= kernel_recvmsg(sock
, &msg
, iov
, nr
, buflen
, MSG_DONTWAIT
);
471 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
472 * possibly we should cache this in the svc_sock structure
473 * at accept time. FIXME
475 alen
= sizeof(rqstp
->rq_addr
);
476 sock
->ops
->getname(sock
, (struct sockaddr
*)&rqstp
->rq_addr
, &alen
, 1);
478 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
479 rqstp
->rq_sock
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
485 * Set socket snd and rcv buffer lengths
488 svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
, unsigned int rcv
)
492 oldfs
= get_fs(); set_fs(KERNEL_DS
);
493 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
494 (char*)&snd
, sizeof(snd
));
495 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
496 (char*)&rcv
, sizeof(rcv
));
498 /* sock_setsockopt limits use to sysctl_?mem_max,
499 * which isn't acceptable. Until that is made conditional
500 * on not having CAP_SYS_RESOURCE or similar, we go direct...
501 * DaveM said I could!
504 sock
->sk
->sk_sndbuf
= snd
* 2;
505 sock
->sk
->sk_rcvbuf
= rcv
* 2;
506 sock
->sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
|SOCK_RCVBUF_LOCK
;
507 release_sock(sock
->sk
);
511 * INET callback when data has been received on the socket.
514 svc_udp_data_ready(struct sock
*sk
, int count
)
516 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
520 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
521 svsk
, sk
, count
, test_bit(SK_BUSY
, &svsk
->sk_flags
));
522 set_bit(SK_DATA
, &svsk
->sk_flags
);
523 svc_sock_enqueue(svsk
);
525 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
526 wake_up_interruptible(sk
->sk_sleep
);
530 * INET callback when space is newly available on the socket.
533 svc_write_space(struct sock
*sk
)
535 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
538 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
539 svsk
, sk
, test_bit(SK_BUSY
, &svsk
->sk_flags
));
540 svc_sock_enqueue(svsk
);
543 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
)) {
544 printk(KERN_WARNING
"RPC svc_write_space: some sleeping on %p\n",
546 wake_up_interruptible(sk
->sk_sleep
);
551 * Receive a datagram from a UDP socket.
554 csum_partial_copy_to_xdr(struct xdr_buf
*xdr
, struct sk_buff
*skb
);
557 svc_udp_recvfrom(struct svc_rqst
*rqstp
)
559 struct svc_sock
*svsk
= rqstp
->rq_sock
;
560 struct svc_serv
*serv
= svsk
->sk_server
;
564 if (test_and_clear_bit(SK_CHNGBUF
, &svsk
->sk_flags
))
565 /* udp sockets need large rcvbuf as all pending
566 * requests are still in that buffer. sndbuf must
567 * also be large enough that there is enough space
568 * for one reply per thread.
570 svc_sock_setbufsize(svsk
->sk_sock
,
571 (serv
->sv_nrthreads
+3) * serv
->sv_bufsz
,
572 (serv
->sv_nrthreads
+3) * serv
->sv_bufsz
);
574 if ((rqstp
->rq_deferred
= svc_deferred_dequeue(svsk
))) {
575 svc_sock_received(svsk
);
576 return svc_deferred_recv(rqstp
);
579 clear_bit(SK_DATA
, &svsk
->sk_flags
);
580 while ((skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
)) == NULL
) {
581 if (err
== -EAGAIN
) {
582 svc_sock_received(svsk
);
585 /* possibly an icmp error */
586 dprintk("svc: recvfrom returned error %d\n", -err
);
588 if (skb
->stamp
.tv_sec
== 0) {
589 skb
->stamp
.tv_sec
= xtime
.tv_sec
;
590 skb
->stamp
.tv_usec
= xtime
.tv_nsec
* 1000;
591 /* Don't enable netstamp, sunrpc doesn't
592 need that much accuracy */
594 svsk
->sk_sk
->sk_stamp
= skb
->stamp
;
595 set_bit(SK_DATA
, &svsk
->sk_flags
); /* there may be more data... */
598 * Maybe more packets - kick another thread ASAP.
600 svc_sock_received(svsk
);
602 len
= skb
->len
- sizeof(struct udphdr
);
603 rqstp
->rq_arg
.len
= len
;
605 rqstp
->rq_prot
= IPPROTO_UDP
;
607 /* Get sender address */
608 rqstp
->rq_addr
.sin_family
= AF_INET
;
609 rqstp
->rq_addr
.sin_port
= skb
->h
.uh
->source
;
610 rqstp
->rq_addr
.sin_addr
.s_addr
= skb
->nh
.iph
->saddr
;
611 rqstp
->rq_daddr
= skb
->nh
.iph
->daddr
;
613 if (skb_is_nonlinear(skb
)) {
614 /* we have to copy */
616 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
619 skb_free_datagram(svsk
->sk_sk
, skb
);
623 skb_free_datagram(svsk
->sk_sk
, skb
);
625 /* we can use it in-place */
626 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+ sizeof(struct udphdr
);
627 rqstp
->rq_arg
.head
[0].iov_len
= len
;
628 if (skb
->ip_summed
!= CHECKSUM_UNNECESSARY
) {
629 if ((unsigned short)csum_fold(skb_checksum(skb
, 0, skb
->len
, skb
->csum
))) {
630 skb_free_datagram(svsk
->sk_sk
, skb
);
633 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
635 rqstp
->rq_skbuff
= skb
;
638 rqstp
->rq_arg
.page_base
= 0;
639 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
640 rqstp
->rq_arg
.head
[0].iov_len
= len
;
641 rqstp
->rq_arg
.page_len
= 0;
643 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
644 rqstp
->rq_argused
+= (rqstp
->rq_arg
.page_len
+ PAGE_SIZE
- 1)/ PAGE_SIZE
;
648 serv
->sv_stats
->netudpcnt
++;
654 svc_udp_sendto(struct svc_rqst
*rqstp
)
658 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
659 if (error
== -ECONNREFUSED
)
660 /* ICMP error on earlier request. */
661 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
667 svc_udp_init(struct svc_sock
*svsk
)
669 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
670 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
671 svsk
->sk_recvfrom
= svc_udp_recvfrom
;
672 svsk
->sk_sendto
= svc_udp_sendto
;
674 /* initialise setting must have enough space to
675 * receive and respond to one request.
676 * svc_udp_recvfrom will re-adjust if necessary
678 svc_sock_setbufsize(svsk
->sk_sock
,
679 3 * svsk
->sk_server
->sv_bufsz
,
680 3 * svsk
->sk_server
->sv_bufsz
);
682 set_bit(SK_DATA
, &svsk
->sk_flags
); /* might have come in before data_ready set up */
683 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
687 * A data_ready event on a listening socket means there's a connection
688 * pending. Do not use state_change as a substitute for it.
691 svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
693 struct svc_sock
*svsk
;
695 dprintk("svc: socket %p TCP (listen) state change %d\n",
698 if (sk
->sk_state
!= TCP_LISTEN
) {
700 * This callback may called twice when a new connection
701 * is established as a child socket inherits everything
702 * from a parent LISTEN socket.
703 * 1) data_ready method of the parent socket will be called
704 * when one of child sockets become ESTABLISHED.
705 * 2) data_ready method of the child socket may be called
706 * when it receives data before the socket is accepted.
707 * In case of 2, we should ignore it silently.
711 if (!(svsk
= (struct svc_sock
*) sk
->sk_user_data
)) {
712 printk("svc: socket %p: no user data\n", sk
);
715 set_bit(SK_CONN
, &svsk
->sk_flags
);
716 svc_sock_enqueue(svsk
);
718 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
719 wake_up_interruptible_all(sk
->sk_sleep
);
723 * A state change on a connected socket means it's dying or dead.
726 svc_tcp_state_change(struct sock
*sk
)
728 struct svc_sock
*svsk
;
730 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
731 sk
, sk
->sk_state
, sk
->sk_user_data
);
733 if (!(svsk
= (struct svc_sock
*) sk
->sk_user_data
)) {
734 printk("svc: socket %p: no user data\n", sk
);
737 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
738 svc_sock_enqueue(svsk
);
740 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
741 wake_up_interruptible_all(sk
->sk_sleep
);
745 svc_tcp_data_ready(struct sock
*sk
, int count
)
747 struct svc_sock
* svsk
;
749 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
750 sk
, sk
->sk_user_data
);
751 if (!(svsk
= (struct svc_sock
*)(sk
->sk_user_data
)))
753 set_bit(SK_DATA
, &svsk
->sk_flags
);
754 svc_sock_enqueue(svsk
);
756 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
757 wake_up_interruptible(sk
->sk_sleep
);
761 * Accept a TCP connection
764 svc_tcp_accept(struct svc_sock
*svsk
)
766 struct sockaddr_in sin
;
767 struct svc_serv
*serv
= svsk
->sk_server
;
768 struct socket
*sock
= svsk
->sk_sock
;
769 struct socket
*newsock
;
770 struct proto_ops
*ops
;
771 struct svc_sock
*newsvsk
;
774 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
778 err
= sock_create_lite(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &newsock
);
781 printk(KERN_WARNING
"%s: no more sockets!\n",
786 dprintk("svc: tcp_accept %p allocated\n", newsock
);
787 newsock
->ops
= ops
= sock
->ops
;
789 clear_bit(SK_CONN
, &svsk
->sk_flags
);
790 if ((err
= ops
->accept(sock
, newsock
, O_NONBLOCK
)) < 0) {
791 if (err
!= -EAGAIN
&& net_ratelimit())
792 printk(KERN_WARNING
"%s: accept failed (err %d)!\n",
793 serv
->sv_name
, -err
);
794 goto failed
; /* aborted connection or whatever */
796 set_bit(SK_CONN
, &svsk
->sk_flags
);
797 svc_sock_enqueue(svsk
);
800 err
= ops
->getname(newsock
, (struct sockaddr
*) &sin
, &slen
, 1);
803 printk(KERN_WARNING
"%s: peername failed (err %d)!\n",
804 serv
->sv_name
, -err
);
805 goto failed
; /* aborted connection or whatever */
808 /* Ideally, we would want to reject connections from unauthorized
809 * hosts here, but when we get encription, the IP of the host won't
810 * tell us anything. For now just warn about unpriv connections.
812 if (ntohs(sin
.sin_port
) >= 1024) {
814 "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n",
816 NIPQUAD(sin
.sin_addr
.s_addr
), ntohs(sin
.sin_port
));
819 dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv
->sv_name
,
820 NIPQUAD(sin
.sin_addr
.s_addr
), ntohs(sin
.sin_port
));
822 /* make sure that a write doesn't block forever when
825 newsock
->sk
->sk_sndtimeo
= HZ
*30;
827 if (!(newsvsk
= svc_setup_socket(serv
, newsock
, &err
, 0)))
831 /* make sure that we don't have too many active connections.
832 * If we have, something must be dropped.
834 * There's no point in trying to do random drop here for
835 * DoS prevention. The NFS clients does 1 reconnect in 15
836 * seconds. An attacker can easily beat that.
838 * The only somewhat efficient mechanism would be if drop
839 * old connections from the same IP first. But right now
840 * we don't even record the client IP in svc_sock.
842 if (serv
->sv_tmpcnt
> (serv
->sv_nrthreads
+3)*20) {
843 struct svc_sock
*svsk
= NULL
;
844 spin_lock_bh(&serv
->sv_lock
);
845 if (!list_empty(&serv
->sv_tempsocks
)) {
846 if (net_ratelimit()) {
847 /* Try to help the admin */
848 printk(KERN_NOTICE
"%s: too many open TCP "
849 "sockets, consider increasing the "
850 "number of nfsd threads\n",
852 printk(KERN_NOTICE
"%s: last TCP connect from "
855 NIPQUAD(sin
.sin_addr
.s_addr
),
856 ntohs(sin
.sin_port
));
859 * Always select the oldest socket. It's not fair,
862 svsk
= list_entry(serv
->sv_tempsocks
.prev
,
865 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
868 spin_unlock_bh(&serv
->sv_lock
);
871 svc_sock_enqueue(svsk
);
878 serv
->sv_stats
->nettcpconn
++;
883 sock_release(newsock
);
888 * Receive data from a TCP socket.
891 svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
893 struct svc_sock
*svsk
= rqstp
->rq_sock
;
894 struct svc_serv
*serv
= svsk
->sk_server
;
896 struct kvec vec
[RPCSVC_MAXPAGES
];
899 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
900 svsk
, test_bit(SK_DATA
, &svsk
->sk_flags
),
901 test_bit(SK_CONN
, &svsk
->sk_flags
),
902 test_bit(SK_CLOSE
, &svsk
->sk_flags
));
904 if ((rqstp
->rq_deferred
= svc_deferred_dequeue(svsk
))) {
905 svc_sock_received(svsk
);
906 return svc_deferred_recv(rqstp
);
909 if (test_bit(SK_CLOSE
, &svsk
->sk_flags
)) {
910 svc_delete_socket(svsk
);
914 if (test_bit(SK_CONN
, &svsk
->sk_flags
)) {
915 svc_tcp_accept(svsk
);
916 svc_sock_received(svsk
);
920 if (test_and_clear_bit(SK_CHNGBUF
, &svsk
->sk_flags
))
921 /* sndbuf needs to have room for one request
922 * per thread, otherwise we can stall even when the
923 * network isn't a bottleneck.
924 * rcvbuf just needs to be able to hold a few requests.
925 * Normally they will be removed from the queue
926 * as soon a a complete request arrives.
928 svc_sock_setbufsize(svsk
->sk_sock
,
929 (serv
->sv_nrthreads
+3) * serv
->sv_bufsz
,
932 clear_bit(SK_DATA
, &svsk
->sk_flags
);
934 /* Receive data. If we haven't got the record length yet, get
935 * the next four bytes. Otherwise try to gobble up as much as
936 * possible up to the complete record length.
938 if (svsk
->sk_tcplen
< 4) {
939 unsigned long want
= 4 - svsk
->sk_tcplen
;
942 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
944 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
946 svsk
->sk_tcplen
+= len
;
949 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
951 svc_sock_received(svsk
);
952 return -EAGAIN
; /* record header not complete */
955 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
956 if (!(svsk
->sk_reclen
& 0x80000000)) {
957 /* FIXME: technically, a record can be fragmented,
958 * and non-terminal fragments will not have the top
959 * bit set in the fragment length header.
960 * But apparently no known nfs clients send fragmented
962 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx (non-terminal)\n",
963 (unsigned long) svsk
->sk_reclen
);
966 svsk
->sk_reclen
&= 0x7fffffff;
967 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
968 if (svsk
->sk_reclen
> serv
->sv_bufsz
) {
969 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx (large)\n",
970 (unsigned long) svsk
->sk_reclen
);
975 /* Check whether enough data is available */
976 len
= svc_recv_available(svsk
);
980 if (len
< svsk
->sk_reclen
) {
981 dprintk("svc: incomplete TCP record (%d of %d)\n",
982 len
, svsk
->sk_reclen
);
983 svc_sock_received(svsk
);
984 return -EAGAIN
; /* record not complete */
986 len
= svsk
->sk_reclen
;
987 set_bit(SK_DATA
, &svsk
->sk_flags
);
989 vec
[0] = rqstp
->rq_arg
.head
[0];
993 vec
[pnum
].iov_base
= page_address(rqstp
->rq_argpages
[rqstp
->rq_argused
++]);
994 vec
[pnum
].iov_len
= PAGE_SIZE
;
999 /* Now receive data */
1000 len
= svc_recvfrom(rqstp
, vec
, pnum
, len
);
1004 dprintk("svc: TCP complete record (%d bytes)\n", len
);
1005 rqstp
->rq_arg
.len
= len
;
1006 rqstp
->rq_arg
.page_base
= 0;
1007 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1008 rqstp
->rq_arg
.head
[0].iov_len
= len
;
1009 rqstp
->rq_arg
.page_len
= 0;
1011 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
1014 rqstp
->rq_skbuff
= NULL
;
1015 rqstp
->rq_prot
= IPPROTO_TCP
;
1017 /* Reset TCP read info */
1018 svsk
->sk_reclen
= 0;
1019 svsk
->sk_tcplen
= 0;
1021 svc_sock_received(svsk
);
1023 serv
->sv_stats
->nettcpcnt
++;
1028 svc_delete_socket(svsk
);
1032 if (len
== -EAGAIN
) {
1033 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1034 svc_sock_received(svsk
);
1036 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1037 svsk
->sk_server
->sv_name
, -len
);
1038 svc_sock_received(svsk
);
1045 * Send out data on TCP socket.
1048 svc_tcp_sendto(struct svc_rqst
*rqstp
)
1050 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1054 /* Set up the first element of the reply kvec.
1055 * Any other kvecs that may be in use have been taken
1056 * care of by the server implementation itself.
1058 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1059 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1061 if (test_bit(SK_DEAD
, &rqstp
->rq_sock
->sk_flags
))
1064 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1065 if (sent
!= xbufp
->len
) {
1066 printk(KERN_NOTICE
"rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1067 rqstp
->rq_sock
->sk_server
->sv_name
,
1068 (sent
<0)?"got error":"sent only",
1070 svc_delete_socket(rqstp
->rq_sock
);
1077 svc_tcp_init(struct svc_sock
*svsk
)
1079 struct sock
*sk
= svsk
->sk_sk
;
1080 struct tcp_opt
*tp
= tcp_sk(sk
);
1082 svsk
->sk_recvfrom
= svc_tcp_recvfrom
;
1083 svsk
->sk_sendto
= svc_tcp_sendto
;
1085 if (sk
->sk_state
== TCP_LISTEN
) {
1086 dprintk("setting up TCP socket for listening\n");
1087 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1088 set_bit(SK_CONN
, &svsk
->sk_flags
);
1090 dprintk("setting up TCP socket for reading\n");
1091 sk
->sk_state_change
= svc_tcp_state_change
;
1092 sk
->sk_data_ready
= svc_tcp_data_ready
;
1093 sk
->sk_write_space
= svc_write_space
;
1095 svsk
->sk_reclen
= 0;
1096 svsk
->sk_tcplen
= 0;
1098 tp
->nonagle
= 1; /* disable Nagle's algorithm */
1100 /* initialise setting must have enough space to
1101 * receive and respond to one request.
1102 * svc_tcp_recvfrom will re-adjust if necessary
1104 svc_sock_setbufsize(svsk
->sk_sock
,
1105 3 * svsk
->sk_server
->sv_bufsz
,
1106 3 * svsk
->sk_server
->sv_bufsz
);
1108 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1109 set_bit(SK_DATA
, &svsk
->sk_flags
);
1110 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1111 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1116 svc_sock_update_bufs(struct svc_serv
*serv
)
1119 * The number of server threads has changed. Update
1120 * rcvbuf and sndbuf accordingly on all sockets
1122 struct list_head
*le
;
1124 spin_lock_bh(&serv
->sv_lock
);
1125 list_for_each(le
, &serv
->sv_permsocks
) {
1126 struct svc_sock
*svsk
=
1127 list_entry(le
, struct svc_sock
, sk_list
);
1128 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1130 list_for_each(le
, &serv
->sv_tempsocks
) {
1131 struct svc_sock
*svsk
=
1132 list_entry(le
, struct svc_sock
, sk_list
);
1133 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1135 spin_unlock_bh(&serv
->sv_lock
);
1139 * Receive the next request on any socket.
1142 svc_recv(struct svc_serv
*serv
, struct svc_rqst
*rqstp
, long timeout
)
1144 struct svc_sock
*svsk
=NULL
;
1147 struct xdr_buf
*arg
;
1148 DECLARE_WAITQUEUE(wait
, current
);
1150 dprintk("svc: server %p waiting for data (to = %ld)\n",
1155 "svc_recv: service %p, socket not NULL!\n",
1157 if (waitqueue_active(&rqstp
->rq_wait
))
1159 "svc_recv: service %p, wait queue active!\n",
1162 /* Initialize the buffers */
1163 /* first reclaim pages that were moved to response list */
1164 svc_pushback_allpages(rqstp
);
1166 /* now allocate needed pages. If we get a failure, sleep briefly */
1167 pages
= 2 + (serv
->sv_bufsz
+ PAGE_SIZE
-1) / PAGE_SIZE
;
1168 while (rqstp
->rq_arghi
< pages
) {
1169 struct page
*p
= alloc_page(GFP_KERNEL
);
1171 set_current_state(TASK_UNINTERRUPTIBLE
);
1172 schedule_timeout(HZ
/2);
1175 rqstp
->rq_argpages
[rqstp
->rq_arghi
++] = p
;
1178 /* Make arg->head point to first page and arg->pages point to rest */
1179 arg
= &rqstp
->rq_arg
;
1180 arg
->head
[0].iov_base
= page_address(rqstp
->rq_argpages
[0]);
1181 arg
->head
[0].iov_len
= PAGE_SIZE
;
1182 rqstp
->rq_argused
= 1;
1183 arg
->pages
= rqstp
->rq_argpages
+ 1;
1185 /* save at least one page for response */
1186 arg
->page_len
= (pages
-2)*PAGE_SIZE
;
1187 arg
->len
= (pages
-1)*PAGE_SIZE
;
1188 arg
->tail
[0].iov_len
= 0;
1193 spin_lock_bh(&serv
->sv_lock
);
1194 if (!list_empty(&serv
->sv_tempsocks
)) {
1195 svsk
= list_entry(serv
->sv_tempsocks
.next
,
1196 struct svc_sock
, sk_list
);
1197 /* apparently the "standard" is that clients close
1198 * idle connections after 5 minutes, servers after
1200 * http://www.connectathon.org/talks96/nfstcp.pdf
1202 if (get_seconds() - svsk
->sk_lastrecv
< 6*60
1203 || test_bit(SK_BUSY
, &svsk
->sk_flags
))
1207 set_bit(SK_BUSY
, &svsk
->sk_flags
);
1208 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1209 rqstp
->rq_sock
= svsk
;
1211 } else if ((svsk
= svc_sock_dequeue(serv
)) != NULL
) {
1212 rqstp
->rq_sock
= svsk
;
1214 rqstp
->rq_reserved
= serv
->sv_bufsz
;
1215 svsk
->sk_reserved
+= rqstp
->rq_reserved
;
1217 /* No data pending. Go to sleep */
1218 svc_serv_enqueue(serv
, rqstp
);
1221 * We have to be able to interrupt this wait
1222 * to bring down the daemons ...
1224 set_current_state(TASK_INTERRUPTIBLE
);
1225 add_wait_queue(&rqstp
->rq_wait
, &wait
);
1226 spin_unlock_bh(&serv
->sv_lock
);
1228 schedule_timeout(timeout
);
1230 if (current
->flags
& PF_FREEZE
)
1231 refrigerator(PF_FREEZE
);
1233 spin_lock_bh(&serv
->sv_lock
);
1234 remove_wait_queue(&rqstp
->rq_wait
, &wait
);
1236 if (!(svsk
= rqstp
->rq_sock
)) {
1237 svc_serv_dequeue(serv
, rqstp
);
1238 spin_unlock_bh(&serv
->sv_lock
);
1239 dprintk("svc: server %p, no data yet\n", rqstp
);
1240 return signalled()? -EINTR
: -EAGAIN
;
1243 spin_unlock_bh(&serv
->sv_lock
);
1245 dprintk("svc: server %p, socket %p, inuse=%d\n",
1246 rqstp
, svsk
, svsk
->sk_inuse
);
1247 len
= svsk
->sk_recvfrom(rqstp
);
1248 dprintk("svc: got len=%d\n", len
);
1250 /* No data, incomplete (TCP) read, or accept() */
1251 if (len
== 0 || len
== -EAGAIN
) {
1252 rqstp
->rq_res
.len
= 0;
1253 svc_sock_release(rqstp
);
1256 svsk
->sk_lastrecv
= get_seconds();
1257 if (test_bit(SK_TEMP
, &svsk
->sk_flags
)) {
1258 /* push active sockets to end of list */
1259 spin_lock_bh(&serv
->sv_lock
);
1260 if (!list_empty(&svsk
->sk_list
))
1261 list_move_tail(&svsk
->sk_list
, &serv
->sv_tempsocks
);
1262 spin_unlock_bh(&serv
->sv_lock
);
1265 rqstp
->rq_secure
= ntohs(rqstp
->rq_addr
.sin_port
) < 1024;
1266 rqstp
->rq_chandle
.defer
= svc_defer
;
1269 serv
->sv_stats
->netcnt
++;
1277 svc_drop(struct svc_rqst
*rqstp
)
1279 dprintk("svc: socket %p dropped request\n", rqstp
->rq_sock
);
1280 svc_sock_release(rqstp
);
1284 * Return reply to client.
1287 svc_send(struct svc_rqst
*rqstp
)
1289 struct svc_sock
*svsk
;
1293 if ((svsk
= rqstp
->rq_sock
) == NULL
) {
1294 printk(KERN_WARNING
"NULL socket pointer in %s:%d\n",
1295 __FILE__
, __LINE__
);
1299 /* release the receive skb before sending the reply */
1300 svc_release_skb(rqstp
);
1302 /* calculate over-all length */
1303 xb
= & rqstp
->rq_res
;
1304 xb
->len
= xb
->head
[0].iov_len
+
1306 xb
->tail
[0].iov_len
;
1308 /* Grab svsk->sk_sem to serialize outgoing data. */
1309 down(&svsk
->sk_sem
);
1310 if (test_bit(SK_DEAD
, &svsk
->sk_flags
))
1313 len
= svsk
->sk_sendto(rqstp
);
1315 svc_sock_release(rqstp
);
1317 if (len
== -ECONNREFUSED
|| len
== -ENOTCONN
|| len
== -EAGAIN
)
1323 * Initialize socket for RPC use and create svc_sock struct
1324 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1326 static struct svc_sock
*
1327 svc_setup_socket(struct svc_serv
*serv
, struct socket
*sock
,
1328 int *errp
, int pmap_register
)
1330 struct svc_sock
*svsk
;
1333 dprintk("svc: svc_setup_socket %p\n", sock
);
1334 if (!(svsk
= kmalloc(sizeof(*svsk
), GFP_KERNEL
))) {
1338 memset(svsk
, 0, sizeof(*svsk
));
1342 /* Register socket with portmapper */
1343 if (*errp
>= 0 && pmap_register
)
1344 *errp
= svc_register(serv
, inet
->sk_protocol
,
1345 ntohs(inet_sk(inet
)->sport
));
1352 set_bit(SK_BUSY
, &svsk
->sk_flags
);
1353 inet
->sk_user_data
= svsk
;
1354 svsk
->sk_sock
= sock
;
1356 svsk
->sk_ostate
= inet
->sk_state_change
;
1357 svsk
->sk_odata
= inet
->sk_data_ready
;
1358 svsk
->sk_owspace
= inet
->sk_write_space
;
1359 svsk
->sk_server
= serv
;
1360 svsk
->sk_lastrecv
= get_seconds();
1361 INIT_LIST_HEAD(&svsk
->sk_deferred
);
1362 INIT_LIST_HEAD(&svsk
->sk_ready
);
1363 sema_init(&svsk
->sk_sem
, 1);
1365 /* Initialize the socket */
1366 if (sock
->type
== SOCK_DGRAM
)
1371 spin_lock_bh(&serv
->sv_lock
);
1372 if (!pmap_register
) {
1373 set_bit(SK_TEMP
, &svsk
->sk_flags
);
1374 list_add(&svsk
->sk_list
, &serv
->sv_tempsocks
);
1377 clear_bit(SK_TEMP
, &svsk
->sk_flags
);
1378 list_add(&svsk
->sk_list
, &serv
->sv_permsocks
);
1380 spin_unlock_bh(&serv
->sv_lock
);
1382 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1385 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
1386 svc_sock_enqueue(svsk
);
1391 * Create socket for RPC service.
1394 svc_create_socket(struct svc_serv
*serv
, int protocol
, struct sockaddr_in
*sin
)
1396 struct svc_sock
*svsk
;
1397 struct socket
*sock
;
1401 dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n",
1402 serv
->sv_program
->pg_name
, protocol
,
1403 NIPQUAD(sin
->sin_addr
.s_addr
),
1404 ntohs(sin
->sin_port
));
1406 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1407 printk(KERN_WARNING
"svc: only UDP and TCP "
1408 "sockets supported\n");
1411 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1413 if ((error
= sock_create_kern(PF_INET
, type
, protocol
, &sock
)) < 0)
1417 if (type
== SOCK_STREAM
)
1418 sock
->sk
->sk_reuse
= 1; /* allow address reuse */
1419 error
= sock
->ops
->bind(sock
, (struct sockaddr
*) sin
,
1425 if (protocol
== IPPROTO_TCP
) {
1426 if ((error
= sock
->ops
->listen(sock
, 64)) < 0)
1430 if ((svsk
= svc_setup_socket(serv
, sock
, &error
, 1)) != NULL
)
1434 dprintk("svc: svc_create_socket error = %d\n", -error
);
1440 * Remove a dead socket
1443 svc_delete_socket(struct svc_sock
*svsk
)
1445 struct svc_serv
*serv
;
1448 dprintk("svc: svc_delete_socket(%p)\n", svsk
);
1450 serv
= svsk
->sk_server
;
1453 sk
->sk_state_change
= svsk
->sk_ostate
;
1454 sk
->sk_data_ready
= svsk
->sk_odata
;
1455 sk
->sk_write_space
= svsk
->sk_owspace
;
1457 spin_lock_bh(&serv
->sv_lock
);
1459 list_del_init(&svsk
->sk_list
);
1460 list_del_init(&svsk
->sk_ready
);
1461 if (!test_and_set_bit(SK_DEAD
, &svsk
->sk_flags
))
1462 if (test_bit(SK_TEMP
, &svsk
->sk_flags
))
1465 if (!svsk
->sk_inuse
) {
1466 spin_unlock_bh(&serv
->sv_lock
);
1467 sock_release(svsk
->sk_sock
);
1470 spin_unlock_bh(&serv
->sv_lock
);
1471 dprintk(KERN_NOTICE
"svc: server socket destroy delayed\n");
1472 /* svsk->sk_server = NULL; */
1477 * Make a socket for nfsd and lockd
1480 svc_makesock(struct svc_serv
*serv
, int protocol
, unsigned short port
)
1482 struct sockaddr_in sin
;
1484 dprintk("svc: creating socket proto = %d\n", protocol
);
1485 sin
.sin_family
= AF_INET
;
1486 sin
.sin_addr
.s_addr
= INADDR_ANY
;
1487 sin
.sin_port
= htons(port
);
1488 return svc_create_socket(serv
, protocol
, &sin
);
1492 * Handle defer and revisit of requests
1495 static void svc_revisit(struct cache_deferred_req
*dreq
, int too_many
)
1497 struct svc_deferred_req
*dr
= container_of(dreq
, struct svc_deferred_req
, handle
);
1498 struct svc_serv
*serv
= dreq
->owner
;
1499 struct svc_sock
*svsk
;
1502 svc_sock_put(dr
->svsk
);
1506 dprintk("revisit queued\n");
1509 spin_lock_bh(&serv
->sv_lock
);
1510 list_add(&dr
->handle
.recent
, &svsk
->sk_deferred
);
1511 spin_unlock_bh(&serv
->sv_lock
);
1512 set_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1513 svc_sock_enqueue(svsk
);
1517 static struct cache_deferred_req
*
1518 svc_defer(struct cache_req
*req
)
1520 struct svc_rqst
*rqstp
= container_of(req
, struct svc_rqst
, rq_chandle
);
1521 int size
= sizeof(struct svc_deferred_req
) + (rqstp
->rq_arg
.len
);
1522 struct svc_deferred_req
*dr
;
1524 if (rqstp
->rq_arg
.page_len
)
1525 return NULL
; /* if more than a page, give up FIXME */
1526 if (rqstp
->rq_deferred
) {
1527 dr
= rqstp
->rq_deferred
;
1528 rqstp
->rq_deferred
= NULL
;
1530 int skip
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1531 /* FIXME maybe discard if size too large */
1532 dr
= kmalloc(size
, GFP_KERNEL
);
1536 dr
->handle
.owner
= rqstp
->rq_server
;
1537 dr
->prot
= rqstp
->rq_prot
;
1538 dr
->addr
= rqstp
->rq_addr
;
1539 dr
->argslen
= rqstp
->rq_arg
.len
>> 2;
1540 memcpy(dr
->args
, rqstp
->rq_arg
.head
[0].iov_base
-skip
, dr
->argslen
<<2);
1542 spin_lock_bh(&rqstp
->rq_server
->sv_lock
);
1543 rqstp
->rq_sock
->sk_inuse
++;
1544 dr
->svsk
= rqstp
->rq_sock
;
1545 spin_unlock_bh(&rqstp
->rq_server
->sv_lock
);
1547 dr
->handle
.revisit
= svc_revisit
;
1552 * recv data from a deferred request into an active one
1554 static int svc_deferred_recv(struct svc_rqst
*rqstp
)
1556 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
1558 rqstp
->rq_arg
.head
[0].iov_base
= dr
->args
;
1559 rqstp
->rq_arg
.head
[0].iov_len
= dr
->argslen
<<2;
1560 rqstp
->rq_arg
.page_len
= 0;
1561 rqstp
->rq_arg
.len
= dr
->argslen
<<2;
1562 rqstp
->rq_prot
= dr
->prot
;
1563 rqstp
->rq_addr
= dr
->addr
;
1564 return dr
->argslen
<<2;
1568 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_sock
*svsk
)
1570 struct svc_deferred_req
*dr
= NULL
;
1571 struct svc_serv
*serv
= svsk
->sk_server
;
1573 if (!test_bit(SK_DEFERRED
, &svsk
->sk_flags
))
1575 spin_lock_bh(&serv
->sv_lock
);
1576 clear_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1577 if (!list_empty(&svsk
->sk_deferred
)) {
1578 dr
= list_entry(svsk
->sk_deferred
.next
,
1579 struct svc_deferred_req
,
1581 list_del_init(&dr
->handle
.recent
);
1582 set_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1584 spin_unlock_bh(&serv
->sv_lock
);