2 * linux/net/sunrpc/svcsock.c
4 * These are the RPC server socket internals.
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_sock_enqueue procedure...
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
22 #include <linux/sched.h>
23 #include <linux/errno.h>
24 #include <linux/fcntl.h>
25 #include <linux/net.h>
27 #include <linux/inet.h>
28 #include <linux/udp.h>
29 #include <linux/tcp.h>
30 #include <linux/unistd.h>
31 #include <linux/slab.h>
32 #include <linux/netdevice.h>
33 #include <linux/skbuff.h>
34 #include <linux/file.h>
35 #include <linux/freezer.h>
37 #include <net/checksum.h>
39 #include <net/tcp_states.h>
40 #include <asm/uaccess.h>
41 #include <asm/ioctls.h>
43 #include <linux/sunrpc/types.h>
44 #include <linux/sunrpc/xdr.h>
45 #include <linux/sunrpc/svcsock.h>
46 #include <linux/sunrpc/stats.h>
48 /* SMP locking strategy:
50 * svc_pool->sp_lock protects most of the fields of that pool.
51 * svc_serv->sv_lock protects sv_tempsocks, sv_permsocks, sv_tmpcnt.
52 * when both need to be taken (rare), svc_serv->sv_lock is first.
53 * BKL protects svc_serv->sv_nrthread.
54 * svc_sock->sk_defer_lock protects the svc_sock->sk_deferred list
55 * svc_sock->sk_flags.SK_BUSY prevents a svc_sock being enqueued multiply.
57 * Some flags can be set to certain values at any time
58 * providing that certain rules are followed:
60 * SK_CONN, SK_DATA, can be set or cleared at any time.
61 * after a set, svc_sock_enqueue must be called.
62 * after a clear, the socket must be read/accepted
63 * if this succeeds, it must be set again.
64 * SK_CLOSE can set at any time. It is never cleared.
68 #define RPCDBG_FACILITY RPCDBG_SVCSOCK
71 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
72 int *errp
, int pmap_reg
);
73 static void svc_udp_data_ready(struct sock
*, int);
74 static int svc_udp_recvfrom(struct svc_rqst
*);
75 static int svc_udp_sendto(struct svc_rqst
*);
77 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_sock
*svsk
);
78 static int svc_deferred_recv(struct svc_rqst
*rqstp
);
79 static struct cache_deferred_req
*svc_defer(struct cache_req
*req
);
81 /* apparently the "standard" is that clients close
82 * idle connections after 5 minutes, servers after
84 * http://www.connectathon.org/talks96/nfstcp.pdf
86 static int svc_conn_age_period
= 6*60;
89 * Queue up an idle server thread. Must have pool->sp_lock held.
90 * Note: this is really a stack rather than a queue, so that we only
91 * use as many different threads as we need, and the rest don't pollute
95 svc_thread_enqueue(struct svc_pool
*pool
, struct svc_rqst
*rqstp
)
97 list_add(&rqstp
->rq_list
, &pool
->sp_threads
);
101 * Dequeue an nfsd thread. Must have pool->sp_lock held.
104 svc_thread_dequeue(struct svc_pool
*pool
, struct svc_rqst
*rqstp
)
106 list_del(&rqstp
->rq_list
);
110 * Release an skbuff after use
113 svc_release_skb(struct svc_rqst
*rqstp
)
115 struct sk_buff
*skb
= rqstp
->rq_skbuff
;
116 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
119 rqstp
->rq_skbuff
= NULL
;
121 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
122 skb_free_datagram(rqstp
->rq_sock
->sk_sk
, skb
);
125 rqstp
->rq_deferred
= NULL
;
131 * Any space to write?
133 static inline unsigned long
134 svc_sock_wspace(struct svc_sock
*svsk
)
138 if (svsk
->sk_sock
->type
== SOCK_STREAM
)
139 wspace
= sk_stream_wspace(svsk
->sk_sk
);
141 wspace
= sock_wspace(svsk
->sk_sk
);
147 * Queue up a socket with data pending. If there are idle nfsd
148 * processes, wake 'em up.
152 svc_sock_enqueue(struct svc_sock
*svsk
)
154 struct svc_serv
*serv
= svsk
->sk_server
;
155 struct svc_pool
*pool
;
156 struct svc_rqst
*rqstp
;
159 if (!(svsk
->sk_flags
&
160 ( (1<<SK_CONN
)|(1<<SK_DATA
)|(1<<SK_CLOSE
)|(1<<SK_DEFERRED
)) ))
162 if (test_bit(SK_DEAD
, &svsk
->sk_flags
))
166 pool
= svc_pool_for_cpu(svsk
->sk_server
, cpu
);
169 spin_lock_bh(&pool
->sp_lock
);
171 if (!list_empty(&pool
->sp_threads
) &&
172 !list_empty(&pool
->sp_sockets
))
174 "svc_sock_enqueue: threads and sockets both waiting??\n");
176 if (test_bit(SK_DEAD
, &svsk
->sk_flags
)) {
177 /* Don't enqueue dead sockets */
178 dprintk("svc: socket %p is dead, not enqueued\n", svsk
->sk_sk
);
182 /* Mark socket as busy. It will remain in this state until the
183 * server has processed all pending data and put the socket back
184 * on the idle list. We update SK_BUSY atomically because
185 * it also guards against trying to enqueue the svc_sock twice.
187 if (test_and_set_bit(SK_BUSY
, &svsk
->sk_flags
)) {
188 /* Don't enqueue socket while already enqueued */
189 dprintk("svc: socket %p busy, not enqueued\n", svsk
->sk_sk
);
192 BUG_ON(svsk
->sk_pool
!= NULL
);
193 svsk
->sk_pool
= pool
;
195 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
196 if (((atomic_read(&svsk
->sk_reserved
) + serv
->sv_max_mesg
)*2
197 > svc_sock_wspace(svsk
))
198 && !test_bit(SK_CLOSE
, &svsk
->sk_flags
)
199 && !test_bit(SK_CONN
, &svsk
->sk_flags
)) {
200 /* Don't enqueue while not enough space for reply */
201 dprintk("svc: socket %p no space, %d*2 > %ld, not enqueued\n",
202 svsk
->sk_sk
, atomic_read(&svsk
->sk_reserved
)+serv
->sv_max_mesg
,
203 svc_sock_wspace(svsk
));
204 svsk
->sk_pool
= NULL
;
205 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
208 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
211 if (!list_empty(&pool
->sp_threads
)) {
212 rqstp
= list_entry(pool
->sp_threads
.next
,
215 dprintk("svc: socket %p served by daemon %p\n",
217 svc_thread_dequeue(pool
, rqstp
);
220 "svc_sock_enqueue: server %p, rq_sock=%p!\n",
221 rqstp
, rqstp
->rq_sock
);
222 rqstp
->rq_sock
= svsk
;
223 atomic_inc(&svsk
->sk_inuse
);
224 rqstp
->rq_reserved
= serv
->sv_max_mesg
;
225 atomic_add(rqstp
->rq_reserved
, &svsk
->sk_reserved
);
226 BUG_ON(svsk
->sk_pool
!= pool
);
227 wake_up(&rqstp
->rq_wait
);
229 dprintk("svc: socket %p put into queue\n", svsk
->sk_sk
);
230 list_add_tail(&svsk
->sk_ready
, &pool
->sp_sockets
);
231 BUG_ON(svsk
->sk_pool
!= pool
);
235 spin_unlock_bh(&pool
->sp_lock
);
239 * Dequeue the first socket. Must be called with the pool->sp_lock held.
241 static inline struct svc_sock
*
242 svc_sock_dequeue(struct svc_pool
*pool
)
244 struct svc_sock
*svsk
;
246 if (list_empty(&pool
->sp_sockets
))
249 svsk
= list_entry(pool
->sp_sockets
.next
,
250 struct svc_sock
, sk_ready
);
251 list_del_init(&svsk
->sk_ready
);
253 dprintk("svc: socket %p dequeued, inuse=%d\n",
254 svsk
->sk_sk
, atomic_read(&svsk
->sk_inuse
));
260 * Having read something from a socket, check whether it
261 * needs to be re-enqueued.
262 * Note: SK_DATA only gets cleared when a read-attempt finds
263 * no (or insufficient) data.
266 svc_sock_received(struct svc_sock
*svsk
)
268 svsk
->sk_pool
= NULL
;
269 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
270 svc_sock_enqueue(svsk
);
275 * svc_reserve - change the space reserved for the reply to a request.
276 * @rqstp: The request in question
277 * @space: new max space to reserve
279 * Each request reserves some space on the output queue of the socket
280 * to make sure the reply fits. This function reduces that reserved
281 * space to be the amount of space used already, plus @space.
284 void svc_reserve(struct svc_rqst
*rqstp
, int space
)
286 space
+= rqstp
->rq_res
.head
[0].iov_len
;
288 if (space
< rqstp
->rq_reserved
) {
289 struct svc_sock
*svsk
= rqstp
->rq_sock
;
290 atomic_sub((rqstp
->rq_reserved
- space
), &svsk
->sk_reserved
);
291 rqstp
->rq_reserved
= space
;
293 svc_sock_enqueue(svsk
);
298 * Release a socket after use.
301 svc_sock_put(struct svc_sock
*svsk
)
303 if (atomic_dec_and_test(&svsk
->sk_inuse
) &&
304 test_bit(SK_DEAD
, &svsk
->sk_flags
)) {
305 dprintk("svc: releasing dead socket\n");
306 if (svsk
->sk_sock
->file
)
307 sockfd_put(svsk
->sk_sock
);
309 sock_release(svsk
->sk_sock
);
310 if (svsk
->sk_info_authunix
!= NULL
)
311 svcauth_unix_info_release(svsk
->sk_info_authunix
);
317 svc_sock_release(struct svc_rqst
*rqstp
)
319 struct svc_sock
*svsk
= rqstp
->rq_sock
;
321 svc_release_skb(rqstp
);
323 svc_free_res_pages(rqstp
);
324 rqstp
->rq_res
.page_len
= 0;
325 rqstp
->rq_res
.page_base
= 0;
328 /* Reset response buffer and release
330 * But first, check that enough space was reserved
331 * for the reply, otherwise we have a bug!
333 if ((rqstp
->rq_res
.len
) > rqstp
->rq_reserved
)
334 printk(KERN_ERR
"RPC request reserved %d but used %d\n",
338 rqstp
->rq_res
.head
[0].iov_len
= 0;
339 svc_reserve(rqstp
, 0);
340 rqstp
->rq_sock
= NULL
;
346 * External function to wake up a server waiting for data
347 * This really only makes sense for services like lockd
348 * which have exactly one thread anyway.
351 svc_wake_up(struct svc_serv
*serv
)
353 struct svc_rqst
*rqstp
;
355 struct svc_pool
*pool
;
357 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
358 pool
= &serv
->sv_pools
[i
];
360 spin_lock_bh(&pool
->sp_lock
);
361 if (!list_empty(&pool
->sp_threads
)) {
362 rqstp
= list_entry(pool
->sp_threads
.next
,
365 dprintk("svc: daemon %p woken up.\n", rqstp
);
367 svc_thread_dequeue(pool, rqstp);
368 rqstp->rq_sock = NULL;
370 wake_up(&rqstp
->rq_wait
);
372 spin_unlock_bh(&pool
->sp_lock
);
377 * Generic sendto routine
380 svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
382 struct svc_sock
*svsk
= rqstp
->rq_sock
;
383 struct socket
*sock
= svsk
->sk_sock
;
385 char buffer
[CMSG_SPACE(sizeof(struct in_pktinfo
))];
386 struct cmsghdr
*cmh
= (struct cmsghdr
*)buffer
;
387 struct in_pktinfo
*pki
= (struct in_pktinfo
*)CMSG_DATA(cmh
);
391 struct page
**ppage
= xdr
->pages
;
392 size_t base
= xdr
->page_base
;
393 unsigned int pglen
= xdr
->page_len
;
394 unsigned int flags
= MSG_MORE
;
398 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
399 /* set the source and destination */
401 msg
.msg_name
= &rqstp
->rq_addr
;
402 msg
.msg_namelen
= sizeof(rqstp
->rq_addr
);
405 msg
.msg_flags
= MSG_MORE
;
407 msg
.msg_control
= cmh
;
408 msg
.msg_controllen
= sizeof(buffer
);
409 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
410 cmh
->cmsg_level
= SOL_IP
;
411 cmh
->cmsg_type
= IP_PKTINFO
;
412 pki
->ipi_ifindex
= 0;
413 pki
->ipi_spec_dst
.s_addr
= rqstp
->rq_daddr
;
415 if (sock_sendmsg(sock
, &msg
, 0) < 0)
420 if (slen
== xdr
->head
[0].iov_len
)
422 len
= kernel_sendpage(sock
, rqstp
->rq_respages
[0], 0,
423 xdr
->head
[0].iov_len
, flags
);
424 if (len
!= xdr
->head
[0].iov_len
)
426 slen
-= xdr
->head
[0].iov_len
;
431 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
435 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
442 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
447 if (xdr
->tail
[0].iov_len
) {
448 result
= kernel_sendpage(sock
, rqstp
->rq_respages
[0],
449 ((unsigned long)xdr
->tail
[0].iov_base
)
451 xdr
->tail
[0].iov_len
, 0);
457 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %x)\n",
458 rqstp
->rq_sock
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
, xdr
->len
, len
,
459 rqstp
->rq_addr
.sin_addr
.s_addr
);
465 * Report socket names for nfsdfs
467 static int one_sock_name(char *buf
, struct svc_sock
*svsk
)
471 switch(svsk
->sk_sk
->sk_family
) {
473 len
= sprintf(buf
, "ipv4 %s %u.%u.%u.%u %d\n",
474 svsk
->sk_sk
->sk_protocol
==IPPROTO_UDP
?
476 NIPQUAD(inet_sk(svsk
->sk_sk
)->rcv_saddr
),
477 inet_sk(svsk
->sk_sk
)->num
);
480 len
= sprintf(buf
, "*unknown-%d*\n",
481 svsk
->sk_sk
->sk_family
);
487 svc_sock_names(char *buf
, struct svc_serv
*serv
, char *toclose
)
489 struct svc_sock
*svsk
, *closesk
= NULL
;
494 spin_lock(&serv
->sv_lock
);
495 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_list
) {
496 int onelen
= one_sock_name(buf
+len
, svsk
);
497 if (toclose
&& strcmp(toclose
, buf
+len
) == 0)
502 spin_unlock(&serv
->sv_lock
);
504 /* Should unregister with portmap, but you cannot
505 * unregister just one protocol...
507 svc_delete_socket(closesk
);
512 EXPORT_SYMBOL(svc_sock_names
);
515 * Check input queue length
518 svc_recv_available(struct svc_sock
*svsk
)
520 struct socket
*sock
= svsk
->sk_sock
;
523 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
525 return (err
>= 0)? avail
: err
;
529 * Generic recvfrom routine.
532 svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
, int buflen
)
538 rqstp
->rq_addrlen
= sizeof(rqstp
->rq_addr
);
539 sock
= rqstp
->rq_sock
->sk_sock
;
541 msg
.msg_name
= &rqstp
->rq_addr
;
542 msg
.msg_namelen
= sizeof(rqstp
->rq_addr
);
543 msg
.msg_control
= NULL
;
544 msg
.msg_controllen
= 0;
546 msg
.msg_flags
= MSG_DONTWAIT
;
548 len
= kernel_recvmsg(sock
, &msg
, iov
, nr
, buflen
, MSG_DONTWAIT
);
550 /* sock_recvmsg doesn't fill in the name/namelen, so we must..
551 * possibly we should cache this in the svc_sock structure
552 * at accept time. FIXME
554 alen
= sizeof(rqstp
->rq_addr
);
555 kernel_getpeername(sock
, (struct sockaddr
*)&rqstp
->rq_addr
, &alen
);
557 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
558 rqstp
->rq_sock
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
564 * Set socket snd and rcv buffer lengths
567 svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
, unsigned int rcv
)
571 oldfs
= get_fs(); set_fs(KERNEL_DS
);
572 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
573 (char*)&snd
, sizeof(snd
));
574 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
575 (char*)&rcv
, sizeof(rcv
));
577 /* sock_setsockopt limits use to sysctl_?mem_max,
578 * which isn't acceptable. Until that is made conditional
579 * on not having CAP_SYS_RESOURCE or similar, we go direct...
580 * DaveM said I could!
583 sock
->sk
->sk_sndbuf
= snd
* 2;
584 sock
->sk
->sk_rcvbuf
= rcv
* 2;
585 sock
->sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
|SOCK_RCVBUF_LOCK
;
586 release_sock(sock
->sk
);
590 * INET callback when data has been received on the socket.
593 svc_udp_data_ready(struct sock
*sk
, int count
)
595 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
598 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
599 svsk
, sk
, count
, test_bit(SK_BUSY
, &svsk
->sk_flags
));
600 set_bit(SK_DATA
, &svsk
->sk_flags
);
601 svc_sock_enqueue(svsk
);
603 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
604 wake_up_interruptible(sk
->sk_sleep
);
608 * INET callback when space is newly available on the socket.
611 svc_write_space(struct sock
*sk
)
613 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
616 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
617 svsk
, sk
, test_bit(SK_BUSY
, &svsk
->sk_flags
));
618 svc_sock_enqueue(svsk
);
621 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
)) {
622 dprintk("RPC svc_write_space: someone sleeping on %p\n",
624 wake_up_interruptible(sk
->sk_sleep
);
629 * Receive a datagram from a UDP socket.
632 svc_udp_recvfrom(struct svc_rqst
*rqstp
)
634 struct svc_sock
*svsk
= rqstp
->rq_sock
;
635 struct svc_serv
*serv
= svsk
->sk_server
;
639 if (test_and_clear_bit(SK_CHNGBUF
, &svsk
->sk_flags
))
640 /* udp sockets need large rcvbuf as all pending
641 * requests are still in that buffer. sndbuf must
642 * also be large enough that there is enough space
643 * for one reply per thread. We count all threads
644 * rather than threads in a particular pool, which
645 * provides an upper bound on the number of threads
646 * which will access the socket.
648 svc_sock_setbufsize(svsk
->sk_sock
,
649 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
650 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
652 if ((rqstp
->rq_deferred
= svc_deferred_dequeue(svsk
))) {
653 svc_sock_received(svsk
);
654 return svc_deferred_recv(rqstp
);
657 clear_bit(SK_DATA
, &svsk
->sk_flags
);
658 while ((skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
)) == NULL
) {
659 if (err
== -EAGAIN
) {
660 svc_sock_received(svsk
);
663 /* possibly an icmp error */
664 dprintk("svc: recvfrom returned error %d\n", -err
);
666 if (skb
->tstamp
.off_sec
== 0) {
669 tv
.tv_sec
= xtime
.tv_sec
;
670 tv
.tv_usec
= xtime
.tv_nsec
/ NSEC_PER_USEC
;
671 skb_set_timestamp(skb
, &tv
);
672 /* Don't enable netstamp, sunrpc doesn't
673 need that much accuracy */
675 skb_get_timestamp(skb
, &svsk
->sk_sk
->sk_stamp
);
676 set_bit(SK_DATA
, &svsk
->sk_flags
); /* there may be more data... */
679 * Maybe more packets - kick another thread ASAP.
681 svc_sock_received(svsk
);
683 len
= skb
->len
- sizeof(struct udphdr
);
684 rqstp
->rq_arg
.len
= len
;
686 rqstp
->rq_prot
= IPPROTO_UDP
;
688 /* Get sender address */
689 rqstp
->rq_addr
.sin_family
= AF_INET
;
690 rqstp
->rq_addr
.sin_port
= skb
->h
.uh
->source
;
691 rqstp
->rq_addr
.sin_addr
.s_addr
= skb
->nh
.iph
->saddr
;
692 rqstp
->rq_daddr
= skb
->nh
.iph
->daddr
;
694 if (skb_is_nonlinear(skb
)) {
695 /* we have to copy */
697 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
700 skb_free_datagram(svsk
->sk_sk
, skb
);
704 skb_free_datagram(svsk
->sk_sk
, skb
);
706 /* we can use it in-place */
707 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+ sizeof(struct udphdr
);
708 rqstp
->rq_arg
.head
[0].iov_len
= len
;
709 if (skb_checksum_complete(skb
)) {
710 skb_free_datagram(svsk
->sk_sk
, skb
);
713 rqstp
->rq_skbuff
= skb
;
716 rqstp
->rq_arg
.page_base
= 0;
717 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
718 rqstp
->rq_arg
.head
[0].iov_len
= len
;
719 rqstp
->rq_arg
.page_len
= 0;
720 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
722 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
723 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
724 (rqstp
->rq_arg
.page_len
+ PAGE_SIZE
- 1)/ PAGE_SIZE
;
728 serv
->sv_stats
->netudpcnt
++;
734 svc_udp_sendto(struct svc_rqst
*rqstp
)
738 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
739 if (error
== -ECONNREFUSED
)
740 /* ICMP error on earlier request. */
741 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
747 svc_udp_init(struct svc_sock
*svsk
)
749 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
750 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
751 svsk
->sk_recvfrom
= svc_udp_recvfrom
;
752 svsk
->sk_sendto
= svc_udp_sendto
;
754 /* initialise setting must have enough space to
755 * receive and respond to one request.
756 * svc_udp_recvfrom will re-adjust if necessary
758 svc_sock_setbufsize(svsk
->sk_sock
,
759 3 * svsk
->sk_server
->sv_max_mesg
,
760 3 * svsk
->sk_server
->sv_max_mesg
);
762 set_bit(SK_DATA
, &svsk
->sk_flags
); /* might have come in before data_ready set up */
763 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
767 * A data_ready event on a listening socket means there's a connection
768 * pending. Do not use state_change as a substitute for it.
771 svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
773 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
775 dprintk("svc: socket %p TCP (listen) state change %d\n",
779 * This callback may called twice when a new connection
780 * is established as a child socket inherits everything
781 * from a parent LISTEN socket.
782 * 1) data_ready method of the parent socket will be called
783 * when one of child sockets become ESTABLISHED.
784 * 2) data_ready method of the child socket may be called
785 * when it receives data before the socket is accepted.
786 * In case of 2, we should ignore it silently.
788 if (sk
->sk_state
== TCP_LISTEN
) {
790 set_bit(SK_CONN
, &svsk
->sk_flags
);
791 svc_sock_enqueue(svsk
);
793 printk("svc: socket %p: no user data\n", sk
);
796 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
797 wake_up_interruptible_all(sk
->sk_sleep
);
801 * A state change on a connected socket means it's dying or dead.
804 svc_tcp_state_change(struct sock
*sk
)
806 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
808 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
809 sk
, sk
->sk_state
, sk
->sk_user_data
);
812 printk("svc: socket %p: no user data\n", sk
);
814 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
815 svc_sock_enqueue(svsk
);
817 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
818 wake_up_interruptible_all(sk
->sk_sleep
);
822 svc_tcp_data_ready(struct sock
*sk
, int count
)
824 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
826 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
827 sk
, sk
->sk_user_data
);
829 set_bit(SK_DATA
, &svsk
->sk_flags
);
830 svc_sock_enqueue(svsk
);
832 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
833 wake_up_interruptible(sk
->sk_sleep
);
837 * Accept a TCP connection
840 svc_tcp_accept(struct svc_sock
*svsk
)
842 struct sockaddr_in sin
;
843 struct svc_serv
*serv
= svsk
->sk_server
;
844 struct socket
*sock
= svsk
->sk_sock
;
845 struct socket
*newsock
;
846 struct svc_sock
*newsvsk
;
849 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
853 clear_bit(SK_CONN
, &svsk
->sk_flags
);
854 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
857 printk(KERN_WARNING
"%s: no more sockets!\n",
859 else if (err
!= -EAGAIN
&& net_ratelimit())
860 printk(KERN_WARNING
"%s: accept failed (err %d)!\n",
861 serv
->sv_name
, -err
);
865 set_bit(SK_CONN
, &svsk
->sk_flags
);
866 svc_sock_enqueue(svsk
);
869 err
= kernel_getpeername(newsock
, (struct sockaddr
*) &sin
, &slen
);
872 printk(KERN_WARNING
"%s: peername failed (err %d)!\n",
873 serv
->sv_name
, -err
);
874 goto failed
; /* aborted connection or whatever */
877 /* Ideally, we would want to reject connections from unauthorized
878 * hosts here, but when we get encription, the IP of the host won't
879 * tell us anything. For now just warn about unpriv connections.
881 if (ntohs(sin
.sin_port
) >= 1024) {
883 "%s: connect from unprivileged port: %u.%u.%u.%u:%d\n",
885 NIPQUAD(sin
.sin_addr
.s_addr
), ntohs(sin
.sin_port
));
888 dprintk("%s: connect from %u.%u.%u.%u:%04x\n", serv
->sv_name
,
889 NIPQUAD(sin
.sin_addr
.s_addr
), ntohs(sin
.sin_port
));
891 /* make sure that a write doesn't block forever when
894 newsock
->sk
->sk_sndtimeo
= HZ
*30;
896 if (!(newsvsk
= svc_setup_socket(serv
, newsock
, &err
, 0)))
900 /* make sure that we don't have too many active connections.
901 * If we have, something must be dropped.
903 * There's no point in trying to do random drop here for
904 * DoS prevention. The NFS clients does 1 reconnect in 15
905 * seconds. An attacker can easily beat that.
907 * The only somewhat efficient mechanism would be if drop
908 * old connections from the same IP first. But right now
909 * we don't even record the client IP in svc_sock.
911 if (serv
->sv_tmpcnt
> (serv
->sv_nrthreads
+3)*20) {
912 struct svc_sock
*svsk
= NULL
;
913 spin_lock_bh(&serv
->sv_lock
);
914 if (!list_empty(&serv
->sv_tempsocks
)) {
915 if (net_ratelimit()) {
916 /* Try to help the admin */
917 printk(KERN_NOTICE
"%s: too many open TCP "
918 "sockets, consider increasing the "
919 "number of nfsd threads\n",
921 printk(KERN_NOTICE
"%s: last TCP connect from "
924 NIPQUAD(sin
.sin_addr
.s_addr
),
925 ntohs(sin
.sin_port
));
928 * Always select the oldest socket. It's not fair,
931 svsk
= list_entry(serv
->sv_tempsocks
.prev
,
934 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
935 atomic_inc(&svsk
->sk_inuse
);
937 spin_unlock_bh(&serv
->sv_lock
);
940 svc_sock_enqueue(svsk
);
947 serv
->sv_stats
->nettcpconn
++;
952 sock_release(newsock
);
957 * Receive data from a TCP socket.
960 svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
962 struct svc_sock
*svsk
= rqstp
->rq_sock
;
963 struct svc_serv
*serv
= svsk
->sk_server
;
968 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
969 svsk
, test_bit(SK_DATA
, &svsk
->sk_flags
),
970 test_bit(SK_CONN
, &svsk
->sk_flags
),
971 test_bit(SK_CLOSE
, &svsk
->sk_flags
));
973 if ((rqstp
->rq_deferred
= svc_deferred_dequeue(svsk
))) {
974 svc_sock_received(svsk
);
975 return svc_deferred_recv(rqstp
);
978 if (test_bit(SK_CLOSE
, &svsk
->sk_flags
)) {
979 svc_delete_socket(svsk
);
983 if (svsk
->sk_sk
->sk_state
== TCP_LISTEN
) {
984 svc_tcp_accept(svsk
);
985 svc_sock_received(svsk
);
989 if (test_and_clear_bit(SK_CHNGBUF
, &svsk
->sk_flags
))
990 /* sndbuf needs to have room for one request
991 * per thread, otherwise we can stall even when the
992 * network isn't a bottleneck.
994 * We count all threads rather than threads in a
995 * particular pool, which provides an upper bound
996 * on the number of threads which will access the socket.
998 * rcvbuf just needs to be able to hold a few requests.
999 * Normally they will be removed from the queue
1000 * as soon a a complete request arrives.
1002 svc_sock_setbufsize(svsk
->sk_sock
,
1003 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
1004 3 * serv
->sv_max_mesg
);
1006 clear_bit(SK_DATA
, &svsk
->sk_flags
);
1008 /* Receive data. If we haven't got the record length yet, get
1009 * the next four bytes. Otherwise try to gobble up as much as
1010 * possible up to the complete record length.
1012 if (svsk
->sk_tcplen
< 4) {
1013 unsigned long want
= 4 - svsk
->sk_tcplen
;
1016 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
1018 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
1020 svsk
->sk_tcplen
+= len
;
1023 dprintk("svc: short recvfrom while reading record length (%d of %lu)\n",
1025 svc_sock_received(svsk
);
1026 return -EAGAIN
; /* record header not complete */
1029 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
1030 if (!(svsk
->sk_reclen
& 0x80000000)) {
1031 /* FIXME: technically, a record can be fragmented,
1032 * and non-terminal fragments will not have the top
1033 * bit set in the fragment length header.
1034 * But apparently no known nfs clients send fragmented
1036 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx (non-terminal)\n",
1037 (unsigned long) svsk
->sk_reclen
);
1040 svsk
->sk_reclen
&= 0x7fffffff;
1041 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
1042 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
1043 printk(KERN_NOTICE
"RPC: bad TCP reclen 0x%08lx (large)\n",
1044 (unsigned long) svsk
->sk_reclen
);
1049 /* Check whether enough data is available */
1050 len
= svc_recv_available(svsk
);
1054 if (len
< svsk
->sk_reclen
) {
1055 dprintk("svc: incomplete TCP record (%d of %d)\n",
1056 len
, svsk
->sk_reclen
);
1057 svc_sock_received(svsk
);
1058 return -EAGAIN
; /* record not complete */
1060 len
= svsk
->sk_reclen
;
1061 set_bit(SK_DATA
, &svsk
->sk_flags
);
1063 vec
= rqstp
->rq_vec
;
1064 vec
[0] = rqstp
->rq_arg
.head
[0];
1067 while (vlen
< len
) {
1068 vec
[pnum
].iov_base
= page_address(rqstp
->rq_pages
[pnum
]);
1069 vec
[pnum
].iov_len
= PAGE_SIZE
;
1073 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1075 /* Now receive data */
1076 len
= svc_recvfrom(rqstp
, vec
, pnum
, len
);
1080 dprintk("svc: TCP complete record (%d bytes)\n", len
);
1081 rqstp
->rq_arg
.len
= len
;
1082 rqstp
->rq_arg
.page_base
= 0;
1083 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1084 rqstp
->rq_arg
.head
[0].iov_len
= len
;
1085 rqstp
->rq_arg
.page_len
= 0;
1087 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
1090 rqstp
->rq_skbuff
= NULL
;
1091 rqstp
->rq_prot
= IPPROTO_TCP
;
1093 /* Reset TCP read info */
1094 svsk
->sk_reclen
= 0;
1095 svsk
->sk_tcplen
= 0;
1097 svc_sock_received(svsk
);
1099 serv
->sv_stats
->nettcpcnt
++;
1104 svc_delete_socket(svsk
);
1108 if (len
== -EAGAIN
) {
1109 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1110 svc_sock_received(svsk
);
1112 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1113 svsk
->sk_server
->sv_name
, -len
);
1121 * Send out data on TCP socket.
1124 svc_tcp_sendto(struct svc_rqst
*rqstp
)
1126 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1130 /* Set up the first element of the reply kvec.
1131 * Any other kvecs that may be in use have been taken
1132 * care of by the server implementation itself.
1134 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1135 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1137 if (test_bit(SK_DEAD
, &rqstp
->rq_sock
->sk_flags
))
1140 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1141 if (sent
!= xbufp
->len
) {
1142 printk(KERN_NOTICE
"rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1143 rqstp
->rq_sock
->sk_server
->sv_name
,
1144 (sent
<0)?"got error":"sent only",
1146 svc_delete_socket(rqstp
->rq_sock
);
1153 svc_tcp_init(struct svc_sock
*svsk
)
1155 struct sock
*sk
= svsk
->sk_sk
;
1156 struct tcp_sock
*tp
= tcp_sk(sk
);
1158 svsk
->sk_recvfrom
= svc_tcp_recvfrom
;
1159 svsk
->sk_sendto
= svc_tcp_sendto
;
1161 if (sk
->sk_state
== TCP_LISTEN
) {
1162 dprintk("setting up TCP socket for listening\n");
1163 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1164 set_bit(SK_CONN
, &svsk
->sk_flags
);
1166 dprintk("setting up TCP socket for reading\n");
1167 sk
->sk_state_change
= svc_tcp_state_change
;
1168 sk
->sk_data_ready
= svc_tcp_data_ready
;
1169 sk
->sk_write_space
= svc_write_space
;
1171 svsk
->sk_reclen
= 0;
1172 svsk
->sk_tcplen
= 0;
1174 tp
->nonagle
= 1; /* disable Nagle's algorithm */
1176 /* initialise setting must have enough space to
1177 * receive and respond to one request.
1178 * svc_tcp_recvfrom will re-adjust if necessary
1180 svc_sock_setbufsize(svsk
->sk_sock
,
1181 3 * svsk
->sk_server
->sv_max_mesg
,
1182 3 * svsk
->sk_server
->sv_max_mesg
);
1184 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1185 set_bit(SK_DATA
, &svsk
->sk_flags
);
1186 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1187 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1192 svc_sock_update_bufs(struct svc_serv
*serv
)
1195 * The number of server threads has changed. Update
1196 * rcvbuf and sndbuf accordingly on all sockets
1198 struct list_head
*le
;
1200 spin_lock_bh(&serv
->sv_lock
);
1201 list_for_each(le
, &serv
->sv_permsocks
) {
1202 struct svc_sock
*svsk
=
1203 list_entry(le
, struct svc_sock
, sk_list
);
1204 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1206 list_for_each(le
, &serv
->sv_tempsocks
) {
1207 struct svc_sock
*svsk
=
1208 list_entry(le
, struct svc_sock
, sk_list
);
1209 set_bit(SK_CHNGBUF
, &svsk
->sk_flags
);
1211 spin_unlock_bh(&serv
->sv_lock
);
1215 * Receive the next request on any socket. This code is carefully
1216 * organised not to touch any cachelines in the shared svc_serv
1217 * structure, only cachelines in the local svc_pool.
1220 svc_recv(struct svc_rqst
*rqstp
, long timeout
)
1222 struct svc_sock
*svsk
=NULL
;
1223 struct svc_serv
*serv
= rqstp
->rq_server
;
1224 struct svc_pool
*pool
= rqstp
->rq_pool
;
1227 struct xdr_buf
*arg
;
1228 DECLARE_WAITQUEUE(wait
, current
);
1230 dprintk("svc: server %p waiting for data (to = %ld)\n",
1235 "svc_recv: service %p, socket not NULL!\n",
1237 if (waitqueue_active(&rqstp
->rq_wait
))
1239 "svc_recv: service %p, wait queue active!\n",
1243 /* now allocate needed pages. If we get a failure, sleep briefly */
1244 pages
= (serv
->sv_max_mesg
+ PAGE_SIZE
) / PAGE_SIZE
;
1245 for (i
=0; i
< pages
; i
++)
1246 while (rqstp
->rq_pages
[i
] == NULL
) {
1247 struct page
*p
= alloc_page(GFP_KERNEL
);
1249 schedule_timeout_uninterruptible(msecs_to_jiffies(500));
1250 rqstp
->rq_pages
[i
] = p
;
1253 /* Make arg->head point to first page and arg->pages point to rest */
1254 arg
= &rqstp
->rq_arg
;
1255 arg
->head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
1256 arg
->head
[0].iov_len
= PAGE_SIZE
;
1257 arg
->pages
= rqstp
->rq_pages
+ 1;
1259 /* save at least one page for response */
1260 arg
->page_len
= (pages
-2)*PAGE_SIZE
;
1261 arg
->len
= (pages
-1)*PAGE_SIZE
;
1262 arg
->tail
[0].iov_len
= 0;
1269 spin_lock_bh(&pool
->sp_lock
);
1270 if ((svsk
= svc_sock_dequeue(pool
)) != NULL
) {
1271 rqstp
->rq_sock
= svsk
;
1272 atomic_inc(&svsk
->sk_inuse
);
1273 rqstp
->rq_reserved
= serv
->sv_max_mesg
;
1274 atomic_add(rqstp
->rq_reserved
, &svsk
->sk_reserved
);
1276 /* No data pending. Go to sleep */
1277 svc_thread_enqueue(pool
, rqstp
);
1280 * We have to be able to interrupt this wait
1281 * to bring down the daemons ...
1283 set_current_state(TASK_INTERRUPTIBLE
);
1284 add_wait_queue(&rqstp
->rq_wait
, &wait
);
1285 spin_unlock_bh(&pool
->sp_lock
);
1287 schedule_timeout(timeout
);
1291 spin_lock_bh(&pool
->sp_lock
);
1292 remove_wait_queue(&rqstp
->rq_wait
, &wait
);
1294 if (!(svsk
= rqstp
->rq_sock
)) {
1295 svc_thread_dequeue(pool
, rqstp
);
1296 spin_unlock_bh(&pool
->sp_lock
);
1297 dprintk("svc: server %p, no data yet\n", rqstp
);
1298 return signalled()? -EINTR
: -EAGAIN
;
1301 spin_unlock_bh(&pool
->sp_lock
);
1303 dprintk("svc: server %p, pool %u, socket %p, inuse=%d\n",
1304 rqstp
, pool
->sp_id
, svsk
, atomic_read(&svsk
->sk_inuse
));
1305 len
= svsk
->sk_recvfrom(rqstp
);
1306 dprintk("svc: got len=%d\n", len
);
1308 /* No data, incomplete (TCP) read, or accept() */
1309 if (len
== 0 || len
== -EAGAIN
) {
1310 rqstp
->rq_res
.len
= 0;
1311 svc_sock_release(rqstp
);
1314 svsk
->sk_lastrecv
= get_seconds();
1315 clear_bit(SK_OLD
, &svsk
->sk_flags
);
1317 rqstp
->rq_secure
= ntohs(rqstp
->rq_addr
.sin_port
) < 1024;
1318 rqstp
->rq_chandle
.defer
= svc_defer
;
1321 serv
->sv_stats
->netcnt
++;
1329 svc_drop(struct svc_rqst
*rqstp
)
1331 dprintk("svc: socket %p dropped request\n", rqstp
->rq_sock
);
1332 svc_sock_release(rqstp
);
1336 * Return reply to client.
1339 svc_send(struct svc_rqst
*rqstp
)
1341 struct svc_sock
*svsk
;
1345 if ((svsk
= rqstp
->rq_sock
) == NULL
) {
1346 printk(KERN_WARNING
"NULL socket pointer in %s:%d\n",
1347 __FILE__
, __LINE__
);
1351 /* release the receive skb before sending the reply */
1352 svc_release_skb(rqstp
);
1354 /* calculate over-all length */
1355 xb
= & rqstp
->rq_res
;
1356 xb
->len
= xb
->head
[0].iov_len
+
1358 xb
->tail
[0].iov_len
;
1360 /* Grab svsk->sk_mutex to serialize outgoing data. */
1361 mutex_lock(&svsk
->sk_mutex
);
1362 if (test_bit(SK_DEAD
, &svsk
->sk_flags
))
1365 len
= svsk
->sk_sendto(rqstp
);
1366 mutex_unlock(&svsk
->sk_mutex
);
1367 svc_sock_release(rqstp
);
1369 if (len
== -ECONNREFUSED
|| len
== -ENOTCONN
|| len
== -EAGAIN
)
1375 * Timer function to close old temporary sockets, using
1376 * a mark-and-sweep algorithm.
1379 svc_age_temp_sockets(unsigned long closure
)
1381 struct svc_serv
*serv
= (struct svc_serv
*)closure
;
1382 struct svc_sock
*svsk
;
1383 struct list_head
*le
, *next
;
1384 LIST_HEAD(to_be_aged
);
1386 dprintk("svc_age_temp_sockets\n");
1388 if (!spin_trylock_bh(&serv
->sv_lock
)) {
1389 /* busy, try again 1 sec later */
1390 dprintk("svc_age_temp_sockets: busy\n");
1391 mod_timer(&serv
->sv_temptimer
, jiffies
+ HZ
);
1395 list_for_each_safe(le
, next
, &serv
->sv_tempsocks
) {
1396 svsk
= list_entry(le
, struct svc_sock
, sk_list
);
1398 if (!test_and_set_bit(SK_OLD
, &svsk
->sk_flags
))
1400 if (atomic_read(&svsk
->sk_inuse
) || test_bit(SK_BUSY
, &svsk
->sk_flags
))
1402 atomic_inc(&svsk
->sk_inuse
);
1403 list_move(le
, &to_be_aged
);
1404 set_bit(SK_CLOSE
, &svsk
->sk_flags
);
1405 set_bit(SK_DETACHED
, &svsk
->sk_flags
);
1407 spin_unlock_bh(&serv
->sv_lock
);
1409 while (!list_empty(&to_be_aged
)) {
1410 le
= to_be_aged
.next
;
1411 /* fiddling the sk_list node is safe 'cos we're SK_DETACHED */
1413 svsk
= list_entry(le
, struct svc_sock
, sk_list
);
1415 dprintk("queuing svsk %p for closing, %lu seconds old\n",
1416 svsk
, get_seconds() - svsk
->sk_lastrecv
);
1418 /* a thread will dequeue and close it soon */
1419 svc_sock_enqueue(svsk
);
1423 mod_timer(&serv
->sv_temptimer
, jiffies
+ svc_conn_age_period
* HZ
);
1427 * Initialize socket for RPC use and create svc_sock struct
1428 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1430 static struct svc_sock
*
1431 svc_setup_socket(struct svc_serv
*serv
, struct socket
*sock
,
1432 int *errp
, int pmap_register
)
1434 struct svc_sock
*svsk
;
1437 dprintk("svc: svc_setup_socket %p\n", sock
);
1438 if (!(svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
))) {
1445 /* Register socket with portmapper */
1446 if (*errp
>= 0 && pmap_register
)
1447 *errp
= svc_register(serv
, inet
->sk_protocol
,
1448 ntohs(inet_sk(inet
)->sport
));
1455 set_bit(SK_BUSY
, &svsk
->sk_flags
);
1456 inet
->sk_user_data
= svsk
;
1457 svsk
->sk_sock
= sock
;
1459 svsk
->sk_ostate
= inet
->sk_state_change
;
1460 svsk
->sk_odata
= inet
->sk_data_ready
;
1461 svsk
->sk_owspace
= inet
->sk_write_space
;
1462 svsk
->sk_server
= serv
;
1463 atomic_set(&svsk
->sk_inuse
, 0);
1464 svsk
->sk_lastrecv
= get_seconds();
1465 spin_lock_init(&svsk
->sk_defer_lock
);
1466 INIT_LIST_HEAD(&svsk
->sk_deferred
);
1467 INIT_LIST_HEAD(&svsk
->sk_ready
);
1468 mutex_init(&svsk
->sk_mutex
);
1470 /* Initialize the socket */
1471 if (sock
->type
== SOCK_DGRAM
)
1476 spin_lock_bh(&serv
->sv_lock
);
1477 if (!pmap_register
) {
1478 set_bit(SK_TEMP
, &svsk
->sk_flags
);
1479 list_add(&svsk
->sk_list
, &serv
->sv_tempsocks
);
1481 if (serv
->sv_temptimer
.function
== NULL
) {
1482 /* setup timer to age temp sockets */
1483 setup_timer(&serv
->sv_temptimer
, svc_age_temp_sockets
,
1484 (unsigned long)serv
);
1485 mod_timer(&serv
->sv_temptimer
,
1486 jiffies
+ svc_conn_age_period
* HZ
);
1489 clear_bit(SK_TEMP
, &svsk
->sk_flags
);
1490 list_add(&svsk
->sk_list
, &serv
->sv_permsocks
);
1492 spin_unlock_bh(&serv
->sv_lock
);
1494 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1497 clear_bit(SK_BUSY
, &svsk
->sk_flags
);
1498 svc_sock_enqueue(svsk
);
1502 int svc_addsock(struct svc_serv
*serv
,
1508 struct socket
*so
= sockfd_lookup(fd
, &err
);
1509 struct svc_sock
*svsk
= NULL
;
1513 if (so
->sk
->sk_family
!= AF_INET
)
1514 err
= -EAFNOSUPPORT
;
1515 else if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1516 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1517 err
= -EPROTONOSUPPORT
;
1518 else if (so
->state
> SS_UNCONNECTED
)
1521 svsk
= svc_setup_socket(serv
, so
, &err
, 1);
1529 if (proto
) *proto
= so
->sk
->sk_protocol
;
1530 return one_sock_name(name_return
, svsk
);
1532 EXPORT_SYMBOL_GPL(svc_addsock
);
1535 * Create socket for RPC service.
1538 svc_create_socket(struct svc_serv
*serv
, int protocol
, struct sockaddr_in
*sin
)
1540 struct svc_sock
*svsk
;
1541 struct socket
*sock
;
1545 dprintk("svc: svc_create_socket(%s, %d, %u.%u.%u.%u:%d)\n",
1546 serv
->sv_program
->pg_name
, protocol
,
1547 NIPQUAD(sin
->sin_addr
.s_addr
),
1548 ntohs(sin
->sin_port
));
1550 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1551 printk(KERN_WARNING
"svc: only UDP and TCP "
1552 "sockets supported\n");
1555 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1557 if ((error
= sock_create_kern(PF_INET
, type
, protocol
, &sock
)) < 0)
1560 if (type
== SOCK_STREAM
)
1561 sock
->sk
->sk_reuse
= 1; /* allow address reuse */
1562 error
= kernel_bind(sock
, (struct sockaddr
*) sin
,
1567 if (protocol
== IPPROTO_TCP
) {
1568 if ((error
= kernel_listen(sock
, 64)) < 0)
1572 if ((svsk
= svc_setup_socket(serv
, sock
, &error
, 1)) != NULL
)
1576 dprintk("svc: svc_create_socket error = %d\n", -error
);
1582 * Remove a dead socket
1585 svc_delete_socket(struct svc_sock
*svsk
)
1587 struct svc_serv
*serv
;
1590 dprintk("svc: svc_delete_socket(%p)\n", svsk
);
1592 serv
= svsk
->sk_server
;
1595 sk
->sk_state_change
= svsk
->sk_ostate
;
1596 sk
->sk_data_ready
= svsk
->sk_odata
;
1597 sk
->sk_write_space
= svsk
->sk_owspace
;
1599 spin_lock_bh(&serv
->sv_lock
);
1601 if (!test_and_set_bit(SK_DETACHED
, &svsk
->sk_flags
))
1602 list_del_init(&svsk
->sk_list
);
1604 * We used to delete the svc_sock from whichever list
1605 * it's sk_ready node was on, but we don't actually
1606 * need to. This is because the only time we're called
1607 * while still attached to a queue, the queue itself
1608 * is about to be destroyed (in svc_destroy).
1610 if (!test_and_set_bit(SK_DEAD
, &svsk
->sk_flags
))
1611 if (test_bit(SK_TEMP
, &svsk
->sk_flags
))
1614 /* This atomic_inc should be needed - svc_delete_socket
1615 * should have the semantic of dropping a reference.
1616 * But it doesn't yet....
1618 atomic_inc(&svsk
->sk_inuse
);
1619 spin_unlock_bh(&serv
->sv_lock
);
1624 * Make a socket for nfsd and lockd
1627 svc_makesock(struct svc_serv
*serv
, int protocol
, unsigned short port
)
1629 struct sockaddr_in sin
;
1631 dprintk("svc: creating socket proto = %d\n", protocol
);
1632 sin
.sin_family
= AF_INET
;
1633 sin
.sin_addr
.s_addr
= INADDR_ANY
;
1634 sin
.sin_port
= htons(port
);
1635 return svc_create_socket(serv
, protocol
, &sin
);
1639 * Handle defer and revisit of requests
1642 static void svc_revisit(struct cache_deferred_req
*dreq
, int too_many
)
1644 struct svc_deferred_req
*dr
= container_of(dreq
, struct svc_deferred_req
, handle
);
1645 struct svc_sock
*svsk
;
1648 svc_sock_put(dr
->svsk
);
1652 dprintk("revisit queued\n");
1655 spin_lock_bh(&svsk
->sk_defer_lock
);
1656 list_add(&dr
->handle
.recent
, &svsk
->sk_deferred
);
1657 spin_unlock_bh(&svsk
->sk_defer_lock
);
1658 set_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1659 svc_sock_enqueue(svsk
);
1663 static struct cache_deferred_req
*
1664 svc_defer(struct cache_req
*req
)
1666 struct svc_rqst
*rqstp
= container_of(req
, struct svc_rqst
, rq_chandle
);
1667 int size
= sizeof(struct svc_deferred_req
) + (rqstp
->rq_arg
.len
);
1668 struct svc_deferred_req
*dr
;
1670 if (rqstp
->rq_arg
.page_len
)
1671 return NULL
; /* if more than a page, give up FIXME */
1672 if (rqstp
->rq_deferred
) {
1673 dr
= rqstp
->rq_deferred
;
1674 rqstp
->rq_deferred
= NULL
;
1676 int skip
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1677 /* FIXME maybe discard if size too large */
1678 dr
= kmalloc(size
, GFP_KERNEL
);
1682 dr
->handle
.owner
= rqstp
->rq_server
;
1683 dr
->prot
= rqstp
->rq_prot
;
1684 dr
->addr
= rqstp
->rq_addr
;
1685 dr
->daddr
= rqstp
->rq_daddr
;
1686 dr
->argslen
= rqstp
->rq_arg
.len
>> 2;
1687 memcpy(dr
->args
, rqstp
->rq_arg
.head
[0].iov_base
-skip
, dr
->argslen
<<2);
1689 atomic_inc(&rqstp
->rq_sock
->sk_inuse
);
1690 dr
->svsk
= rqstp
->rq_sock
;
1692 dr
->handle
.revisit
= svc_revisit
;
1697 * recv data from a deferred request into an active one
1699 static int svc_deferred_recv(struct svc_rqst
*rqstp
)
1701 struct svc_deferred_req
*dr
= rqstp
->rq_deferred
;
1703 rqstp
->rq_arg
.head
[0].iov_base
= dr
->args
;
1704 rqstp
->rq_arg
.head
[0].iov_len
= dr
->argslen
<<2;
1705 rqstp
->rq_arg
.page_len
= 0;
1706 rqstp
->rq_arg
.len
= dr
->argslen
<<2;
1707 rqstp
->rq_prot
= dr
->prot
;
1708 rqstp
->rq_addr
= dr
->addr
;
1709 rqstp
->rq_daddr
= dr
->daddr
;
1710 rqstp
->rq_respages
= rqstp
->rq_pages
;
1711 return dr
->argslen
<<2;
1715 static struct svc_deferred_req
*svc_deferred_dequeue(struct svc_sock
*svsk
)
1717 struct svc_deferred_req
*dr
= NULL
;
1719 if (!test_bit(SK_DEFERRED
, &svsk
->sk_flags
))
1721 spin_lock_bh(&svsk
->sk_defer_lock
);
1722 clear_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1723 if (!list_empty(&svsk
->sk_deferred
)) {
1724 dr
= list_entry(svsk
->sk_deferred
.next
,
1725 struct svc_deferred_req
,
1727 list_del_init(&dr
->handle
.recent
);
1728 set_bit(SK_DEFERRED
, &svsk
->sk_flags
);
1730 spin_unlock_bh(&svsk
->sk_defer_lock
);