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_xprt_enqueue procedure...
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/errno.h>
25 #include <linux/fcntl.h>
26 #include <linux/net.h>
28 #include <linux/inet.h>
29 #include <linux/udp.h>
30 #include <linux/tcp.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/netdevice.h>
34 #include <linux/skbuff.h>
35 #include <linux/file.h>
36 #include <linux/freezer.h>
38 #include <net/checksum.h>
42 #include <net/tcp_states.h>
43 #include <asm/uaccess.h>
44 #include <asm/ioctls.h>
46 #include <linux/sunrpc/types.h>
47 #include <linux/sunrpc/clnt.h>
48 #include <linux/sunrpc/xdr.h>
49 #include <linux/sunrpc/msg_prot.h>
50 #include <linux/sunrpc/svcsock.h>
51 #include <linux/sunrpc/stats.h>
53 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
56 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
57 int *errp
, int flags
);
58 static void svc_udp_data_ready(struct sock
*, int);
59 static int svc_udp_recvfrom(struct svc_rqst
*);
60 static int svc_udp_sendto(struct svc_rqst
*);
61 static void svc_sock_detach(struct svc_xprt
*);
62 static void svc_tcp_sock_detach(struct svc_xprt
*);
63 static void svc_sock_free(struct svc_xprt
*);
65 static struct svc_xprt
*svc_create_socket(struct svc_serv
*, int,
66 struct sockaddr
*, int, int);
67 #ifdef CONFIG_DEBUG_LOCK_ALLOC
68 static struct lock_class_key svc_key
[2];
69 static struct lock_class_key svc_slock_key
[2];
71 static void svc_reclassify_socket(struct socket
*sock
)
73 struct sock
*sk
= sock
->sk
;
74 BUG_ON(sock_owned_by_user(sk
));
75 switch (sk
->sk_family
) {
77 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
79 "sk_xprt.xpt_lock-AF_INET-NFSD",
84 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
86 "sk_xprt.xpt_lock-AF_INET6-NFSD",
95 static void svc_reclassify_socket(struct socket
*sock
)
101 * Release an skbuff after use
103 static void svc_release_skb(struct svc_rqst
*rqstp
)
105 struct sk_buff
*skb
= rqstp
->rq_xprt_ctxt
;
108 struct svc_sock
*svsk
=
109 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
110 rqstp
->rq_xprt_ctxt
= NULL
;
112 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
113 skb_free_datagram(svsk
->sk_sk
, skb
);
117 union svc_pktinfo_u
{
118 struct in_pktinfo pkti
;
119 struct in6_pktinfo pkti6
;
121 #define SVC_PKTINFO_SPACE \
122 CMSG_SPACE(sizeof(union svc_pktinfo_u))
124 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
126 struct svc_sock
*svsk
=
127 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
128 switch (svsk
->sk_sk
->sk_family
) {
130 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
132 cmh
->cmsg_level
= SOL_IP
;
133 cmh
->cmsg_type
= IP_PKTINFO
;
134 pki
->ipi_ifindex
= 0;
135 pki
->ipi_spec_dst
.s_addr
= rqstp
->rq_daddr
.addr
.s_addr
;
136 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
141 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
143 cmh
->cmsg_level
= SOL_IPV6
;
144 cmh
->cmsg_type
= IPV6_PKTINFO
;
145 pki
->ipi6_ifindex
= 0;
146 ipv6_addr_copy(&pki
->ipi6_addr
,
147 &rqstp
->rq_daddr
.addr6
);
148 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
156 * Generic sendto routine
158 static int svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
160 struct svc_sock
*svsk
=
161 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
162 struct socket
*sock
= svsk
->sk_sock
;
166 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
168 struct cmsghdr
*cmh
= &buffer
.hdr
;
172 struct page
**ppage
= xdr
->pages
;
173 size_t base
= xdr
->page_base
;
174 unsigned int pglen
= xdr
->page_len
;
175 unsigned int flags
= MSG_MORE
;
176 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
180 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
181 struct msghdr msg
= {
182 .msg_name
= &rqstp
->rq_addr
,
183 .msg_namelen
= rqstp
->rq_addrlen
,
185 .msg_controllen
= sizeof(buffer
),
186 .msg_flags
= MSG_MORE
,
189 svc_set_cmsg_data(rqstp
, cmh
);
191 if (sock_sendmsg(sock
, &msg
, 0) < 0)
196 if (slen
== xdr
->head
[0].iov_len
)
198 len
= kernel_sendpage(sock
, rqstp
->rq_respages
[0], 0,
199 xdr
->head
[0].iov_len
, flags
);
200 if (len
!= xdr
->head
[0].iov_len
)
202 slen
-= xdr
->head
[0].iov_len
;
207 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
211 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
218 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
223 if (xdr
->tail
[0].iov_len
) {
224 result
= kernel_sendpage(sock
, rqstp
->rq_respages
[0],
225 ((unsigned long)xdr
->tail
[0].iov_base
)
227 xdr
->tail
[0].iov_len
, 0);
233 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
234 svsk
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
235 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
241 * Report socket names for nfsdfs
243 static int svc_one_sock_name(struct svc_sock
*svsk
, char *buf
, int remaining
)
245 const struct sock
*sk
= svsk
->sk_sk
;
246 const char *proto_name
= sk
->sk_protocol
== IPPROTO_UDP
?
250 switch (sk
->sk_family
) {
252 len
= snprintf(buf
, remaining
, "ipv4 %s %pI4 %d\n",
254 &inet_sk(sk
)->rcv_saddr
,
258 len
= snprintf(buf
, remaining
, "ipv6 %s %pI6 %d\n",
260 &inet6_sk(sk
)->rcv_saddr
,
264 len
= snprintf(buf
, remaining
, "*unknown-%d*\n",
268 if (len
>= remaining
) {
270 return -ENAMETOOLONG
;
276 * svc_sock_names - construct a list of listener names in a string
277 * @serv: pointer to RPC service
278 * @buf: pointer to a buffer to fill in with socket names
279 * @buflen: size of the buffer to be filled
280 * @toclose: pointer to '\0'-terminated C string containing the name
281 * of a listener to be closed
283 * Fills in @buf with a '\n'-separated list of names of listener
284 * sockets. If @toclose is not NULL, the socket named by @toclose
285 * is closed, and is not included in the output list.
287 * Returns positive length of the socket name string, or a negative
288 * errno value on error.
290 int svc_sock_names(struct svc_serv
*serv
, char *buf
, const size_t buflen
,
293 struct svc_sock
*svsk
, *closesk
= NULL
;
299 spin_lock_bh(&serv
->sv_lock
);
300 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
) {
301 int onelen
= svc_one_sock_name(svsk
, buf
+ len
, buflen
- len
);
306 if (toclose
&& strcmp(toclose
, buf
+ len
) == 0)
311 spin_unlock_bh(&serv
->sv_lock
);
314 /* Should unregister with portmap, but you cannot
315 * unregister just one protocol...
317 svc_close_xprt(&closesk
->sk_xprt
);
322 EXPORT_SYMBOL_GPL(svc_sock_names
);
325 * Check input queue length
327 static int svc_recv_available(struct svc_sock
*svsk
)
329 struct socket
*sock
= svsk
->sk_sock
;
332 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
334 return (err
>= 0)? avail
: err
;
338 * Generic recvfrom routine.
340 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
343 struct svc_sock
*svsk
=
344 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
345 struct msghdr msg
= {
346 .msg_flags
= MSG_DONTWAIT
,
350 rqstp
->rq_xprt_hlen
= 0;
352 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
355 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
356 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
361 * Set socket snd and rcv buffer lengths
363 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
368 oldfs
= get_fs(); set_fs(KERNEL_DS
);
369 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
370 (char*)&snd
, sizeof(snd
));
371 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
372 (char*)&rcv
, sizeof(rcv
));
374 /* sock_setsockopt limits use to sysctl_?mem_max,
375 * which isn't acceptable. Until that is made conditional
376 * on not having CAP_SYS_RESOURCE or similar, we go direct...
377 * DaveM said I could!
380 sock
->sk
->sk_sndbuf
= snd
* 2;
381 sock
->sk
->sk_rcvbuf
= rcv
* 2;
382 sock
->sk
->sk_userlocks
|= SOCK_SNDBUF_LOCK
|SOCK_RCVBUF_LOCK
;
383 sock
->sk
->sk_write_space(sock
->sk
);
384 release_sock(sock
->sk
);
388 * INET callback when data has been received on the socket.
390 static void svc_udp_data_ready(struct sock
*sk
, int count
)
392 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
395 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
397 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
398 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
399 svc_xprt_enqueue(&svsk
->sk_xprt
);
401 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
402 wake_up_interruptible(sk
->sk_sleep
);
406 * INET callback when space is newly available on the socket.
408 static void svc_write_space(struct sock
*sk
)
410 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
413 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
414 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
415 svc_xprt_enqueue(&svsk
->sk_xprt
);
418 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
)) {
419 dprintk("RPC svc_write_space: someone sleeping on %p\n",
421 wake_up_interruptible(sk
->sk_sleep
);
425 static void svc_tcp_write_space(struct sock
*sk
)
427 struct socket
*sock
= sk
->sk_socket
;
429 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
) && sock
)
430 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
435 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
437 static int svc_udp_get_dest_address4(struct svc_rqst
*rqstp
,
440 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
441 if (cmh
->cmsg_type
!= IP_PKTINFO
)
443 rqstp
->rq_daddr
.addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
448 * See net/ipv6/datagram.c : datagram_recv_ctl
450 static int svc_udp_get_dest_address6(struct svc_rqst
*rqstp
,
453 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
454 if (cmh
->cmsg_type
!= IPV6_PKTINFO
)
456 ipv6_addr_copy(&rqstp
->rq_daddr
.addr6
, &pki
->ipi6_addr
);
461 * Copy the UDP datagram's destination address to the rqstp structure.
462 * The 'destination' address in this case is the address to which the
463 * peer sent the datagram, i.e. our local address. For multihomed
464 * hosts, this can change from msg to msg. Note that only the IP
465 * address changes, the port number should remain the same.
467 static int svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
470 switch (cmh
->cmsg_level
) {
472 return svc_udp_get_dest_address4(rqstp
, cmh
);
474 return svc_udp_get_dest_address6(rqstp
, cmh
);
481 * Receive a datagram from a UDP socket.
483 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
485 struct svc_sock
*svsk
=
486 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
487 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
491 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
493 struct cmsghdr
*cmh
= &buffer
.hdr
;
494 struct msghdr msg
= {
495 .msg_name
= svc_addr(rqstp
),
497 .msg_controllen
= sizeof(buffer
),
498 .msg_flags
= MSG_DONTWAIT
,
503 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
504 /* udp sockets need large rcvbuf as all pending
505 * requests are still in that buffer. sndbuf must
506 * also be large enough that there is enough space
507 * for one reply per thread. We count all threads
508 * rather than threads in a particular pool, which
509 * provides an upper bound on the number of threads
510 * which will access the socket.
512 svc_sock_setbufsize(svsk
->sk_sock
,
513 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
514 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
516 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
518 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
519 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
521 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
524 if (err
!= -EAGAIN
) {
525 /* possibly an icmp error */
526 dprintk("svc: recvfrom returned error %d\n", -err
);
527 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
529 svc_xprt_received(&svsk
->sk_xprt
);
532 len
= svc_addr_len(svc_addr(rqstp
));
534 return -EAFNOSUPPORT
;
535 rqstp
->rq_addrlen
= len
;
536 if (skb
->tstamp
.tv64
== 0) {
537 skb
->tstamp
= ktime_get_real();
538 /* Don't enable netstamp, sunrpc doesn't
539 need that much accuracy */
541 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
542 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
545 * Maybe more packets - kick another thread ASAP.
547 svc_xprt_received(&svsk
->sk_xprt
);
549 len
= skb
->len
- sizeof(struct udphdr
);
550 rqstp
->rq_arg
.len
= len
;
552 rqstp
->rq_prot
= IPPROTO_UDP
;
554 if (!svc_udp_get_dest_address(rqstp
, cmh
)) {
557 "svc: received unknown control message %d/%d; "
558 "dropping RPC reply datagram\n",
559 cmh
->cmsg_level
, cmh
->cmsg_type
);
560 skb_free_datagram(svsk
->sk_sk
, skb
);
564 if (skb_is_nonlinear(skb
)) {
565 /* we have to copy */
567 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
570 skb_free_datagram(svsk
->sk_sk
, skb
);
574 skb_free_datagram(svsk
->sk_sk
, skb
);
576 /* we can use it in-place */
577 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+
578 sizeof(struct udphdr
);
579 rqstp
->rq_arg
.head
[0].iov_len
= len
;
580 if (skb_checksum_complete(skb
)) {
581 skb_free_datagram(svsk
->sk_sk
, skb
);
584 rqstp
->rq_xprt_ctxt
= skb
;
587 rqstp
->rq_arg
.page_base
= 0;
588 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
589 rqstp
->rq_arg
.head
[0].iov_len
= len
;
590 rqstp
->rq_arg
.page_len
= 0;
591 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
593 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
594 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
595 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
599 serv
->sv_stats
->netudpcnt
++;
605 svc_udp_sendto(struct svc_rqst
*rqstp
)
609 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
610 if (error
== -ECONNREFUSED
)
611 /* ICMP error on earlier request. */
612 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
617 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
621 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
623 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
624 struct svc_serv
*serv
= xprt
->xpt_server
;
625 unsigned long required
;
628 * Set the SOCK_NOSPACE flag before checking the available
631 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
632 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
633 if (required
*2 > sock_wspace(svsk
->sk_sk
))
635 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
639 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
645 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
646 struct sockaddr
*sa
, int salen
,
649 return svc_create_socket(serv
, IPPROTO_UDP
, sa
, salen
, flags
);
652 static struct svc_xprt_ops svc_udp_ops
= {
653 .xpo_create
= svc_udp_create
,
654 .xpo_recvfrom
= svc_udp_recvfrom
,
655 .xpo_sendto
= svc_udp_sendto
,
656 .xpo_release_rqst
= svc_release_skb
,
657 .xpo_detach
= svc_sock_detach
,
658 .xpo_free
= svc_sock_free
,
659 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
660 .xpo_has_wspace
= svc_udp_has_wspace
,
661 .xpo_accept
= svc_udp_accept
,
664 static struct svc_xprt_class svc_udp_class
= {
666 .xcl_owner
= THIS_MODULE
,
667 .xcl_ops
= &svc_udp_ops
,
668 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
671 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
673 int err
, level
, optname
, one
= 1;
675 svc_xprt_init(&svc_udp_class
, &svsk
->sk_xprt
, serv
);
676 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
677 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
678 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
680 /* initialise setting must have enough space to
681 * receive and respond to one request.
682 * svc_udp_recvfrom will re-adjust if necessary
684 svc_sock_setbufsize(svsk
->sk_sock
,
685 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
686 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
688 /* data might have come in before data_ready set up */
689 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
690 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
692 /* make sure we get destination address info */
693 switch (svsk
->sk_sk
->sk_family
) {
696 optname
= IP_PKTINFO
;
700 optname
= IPV6_RECVPKTINFO
;
705 err
= kernel_setsockopt(svsk
->sk_sock
, level
, optname
,
706 (char *)&one
, sizeof(one
));
707 dprintk("svc: kernel_setsockopt returned %d\n", err
);
711 * A data_ready event on a listening socket means there's a connection
712 * pending. Do not use state_change as a substitute for it.
714 static void svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
716 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
718 dprintk("svc: socket %p TCP (listen) state change %d\n",
722 * This callback may called twice when a new connection
723 * is established as a child socket inherits everything
724 * from a parent LISTEN socket.
725 * 1) data_ready method of the parent socket will be called
726 * when one of child sockets become ESTABLISHED.
727 * 2) data_ready method of the child socket may be called
728 * when it receives data before the socket is accepted.
729 * In case of 2, we should ignore it silently.
731 if (sk
->sk_state
== TCP_LISTEN
) {
733 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
734 svc_xprt_enqueue(&svsk
->sk_xprt
);
736 printk("svc: socket %p: no user data\n", sk
);
739 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
740 wake_up_interruptible_all(sk
->sk_sleep
);
744 * A state change on a connected socket means it's dying or dead.
746 static void svc_tcp_state_change(struct sock
*sk
)
748 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
750 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
751 sk
, sk
->sk_state
, sk
->sk_user_data
);
754 printk("svc: socket %p: no user data\n", sk
);
756 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
757 svc_xprt_enqueue(&svsk
->sk_xprt
);
759 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
760 wake_up_interruptible_all(sk
->sk_sleep
);
763 static void svc_tcp_data_ready(struct sock
*sk
, int count
)
765 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
767 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
768 sk
, sk
->sk_user_data
);
770 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
771 svc_xprt_enqueue(&svsk
->sk_xprt
);
773 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
774 wake_up_interruptible(sk
->sk_sleep
);
778 * Accept a TCP connection
780 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
782 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
783 struct sockaddr_storage addr
;
784 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
785 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
786 struct socket
*sock
= svsk
->sk_sock
;
787 struct socket
*newsock
;
788 struct svc_sock
*newsvsk
;
790 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
792 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
796 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
797 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
800 printk(KERN_WARNING
"%s: no more sockets!\n",
802 else if (err
!= -EAGAIN
&& net_ratelimit())
803 printk(KERN_WARNING
"%s: accept failed (err %d)!\n",
804 serv
->sv_name
, -err
);
807 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
809 err
= kernel_getpeername(newsock
, sin
, &slen
);
812 printk(KERN_WARNING
"%s: peername failed (err %d)!\n",
813 serv
->sv_name
, -err
);
814 goto failed
; /* aborted connection or whatever */
817 /* Ideally, we would want to reject connections from unauthorized
818 * hosts here, but when we get encryption, the IP of the host won't
819 * tell us anything. For now just warn about unpriv connections.
821 if (!svc_port_is_privileged(sin
)) {
823 "%s: connect from unprivileged port: %s\n",
825 __svc_print_addr(sin
, buf
, sizeof(buf
)));
827 dprintk("%s: connect from %s\n", serv
->sv_name
,
828 __svc_print_addr(sin
, buf
, sizeof(buf
)));
830 /* make sure that a write doesn't block forever when
833 newsock
->sk
->sk_sndtimeo
= HZ
*30;
835 if (!(newsvsk
= svc_setup_socket(serv
, newsock
, &err
,
836 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
))))
838 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
839 err
= kernel_getsockname(newsock
, sin
, &slen
);
840 if (unlikely(err
< 0)) {
841 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
842 slen
= offsetof(struct sockaddr
, sa_data
);
844 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
847 serv
->sv_stats
->nettcpconn
++;
849 return &newsvsk
->sk_xprt
;
852 sock_release(newsock
);
858 * If we haven't gotten the record length yet, get the next four bytes.
859 * Otherwise try to gobble up as much as possible up to the complete
862 static int svc_tcp_recv_record(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
864 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
867 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
868 /* sndbuf needs to have room for one request
869 * per thread, otherwise we can stall even when the
870 * network isn't a bottleneck.
872 * We count all threads rather than threads in a
873 * particular pool, which provides an upper bound
874 * on the number of threads which will access the socket.
876 * rcvbuf just needs to be able to hold a few requests.
877 * Normally they will be removed from the queue
878 * as soon a a complete request arrives.
880 svc_sock_setbufsize(svsk
->sk_sock
,
881 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
882 3 * serv
->sv_max_mesg
);
884 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
886 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
887 int want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
890 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
892 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
894 svsk
->sk_tcplen
+= len
;
897 dprintk("svc: short recvfrom while reading record "
898 "length (%d of %d)\n", len
, want
);
899 svc_xprt_received(&svsk
->sk_xprt
);
900 goto err_again
; /* record header not complete */
903 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
904 if (!(svsk
->sk_reclen
& RPC_LAST_STREAM_FRAGMENT
)) {
905 /* FIXME: technically, a record can be fragmented,
906 * and non-terminal fragments will not have the top
907 * bit set in the fragment length header.
908 * But apparently no known nfs clients send fragmented
911 printk(KERN_NOTICE
"RPC: multiple fragments "
912 "per record not supported\n");
916 svsk
->sk_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
917 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
918 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
920 printk(KERN_NOTICE
"RPC: "
921 "fragment too large: 0x%08lx\n",
922 (unsigned long)svsk
->sk_reclen
);
927 /* Check whether enough data is available */
928 len
= svc_recv_available(svsk
);
932 if (len
< svsk
->sk_reclen
) {
933 dprintk("svc: incomplete TCP record (%d of %d)\n",
934 len
, svsk
->sk_reclen
);
935 svc_xprt_received(&svsk
->sk_xprt
);
936 goto err_again
; /* record not complete */
938 len
= svsk
->sk_reclen
;
939 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
943 if (len
== -EAGAIN
) {
944 dprintk("RPC: TCP recv_record got EAGAIN\n");
945 svc_xprt_received(&svsk
->sk_xprt
);
949 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
955 * Receive data from a TCP socket.
957 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
959 struct svc_sock
*svsk
=
960 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
961 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
966 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
967 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
968 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
969 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
971 len
= svc_tcp_recv_record(svsk
, rqstp
);
976 vec
[0] = rqstp
->rq_arg
.head
[0];
980 vec
[pnum
].iov_base
= page_address(rqstp
->rq_pages
[pnum
]);
981 vec
[pnum
].iov_len
= PAGE_SIZE
;
985 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
987 /* Now receive data */
988 len
= svc_recvfrom(rqstp
, vec
, pnum
, len
);
992 dprintk("svc: TCP complete record (%d bytes)\n", len
);
993 rqstp
->rq_arg
.len
= len
;
994 rqstp
->rq_arg
.page_base
= 0;
995 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
996 rqstp
->rq_arg
.head
[0].iov_len
= len
;
997 rqstp
->rq_arg
.page_len
= 0;
999 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
1002 rqstp
->rq_xprt_ctxt
= NULL
;
1003 rqstp
->rq_prot
= IPPROTO_TCP
;
1005 /* Reset TCP read info */
1006 svsk
->sk_reclen
= 0;
1007 svsk
->sk_tcplen
= 0;
1009 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1010 svc_xprt_received(&svsk
->sk_xprt
);
1012 serv
->sv_stats
->nettcpcnt
++;
1017 if (len
== -EAGAIN
) {
1018 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1019 svc_xprt_received(&svsk
->sk_xprt
);
1023 if (len
!= -EAGAIN
) {
1024 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1025 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1026 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1032 * Send out data on TCP socket.
1034 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
1036 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1040 /* Set up the first element of the reply kvec.
1041 * Any other kvecs that may be in use have been taken
1042 * care of by the server implementation itself.
1044 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1045 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1047 if (test_bit(XPT_DEAD
, &rqstp
->rq_xprt
->xpt_flags
))
1050 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1051 if (sent
!= xbufp
->len
) {
1053 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1054 "- shutting down socket\n",
1055 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1056 (sent
<0)?"got error":"sent only",
1058 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1059 svc_xprt_enqueue(rqstp
->rq_xprt
);
1066 * Setup response header. TCP has a 4B record length field.
1068 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1070 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1072 /* tcp needs a space for the record length... */
1076 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
1078 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1079 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1082 if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
1084 required
= atomic_read(&xprt
->xpt_reserved
) + serv
->sv_max_mesg
;
1085 if (sk_stream_wspace(svsk
->sk_sk
) >= required
)
1087 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
1091 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1092 struct sockaddr
*sa
, int salen
,
1095 return svc_create_socket(serv
, IPPROTO_TCP
, sa
, salen
, flags
);
1098 static struct svc_xprt_ops svc_tcp_ops
= {
1099 .xpo_create
= svc_tcp_create
,
1100 .xpo_recvfrom
= svc_tcp_recvfrom
,
1101 .xpo_sendto
= svc_tcp_sendto
,
1102 .xpo_release_rqst
= svc_release_skb
,
1103 .xpo_detach
= svc_tcp_sock_detach
,
1104 .xpo_free
= svc_sock_free
,
1105 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1106 .xpo_has_wspace
= svc_tcp_has_wspace
,
1107 .xpo_accept
= svc_tcp_accept
,
1110 static struct svc_xprt_class svc_tcp_class
= {
1112 .xcl_owner
= THIS_MODULE
,
1113 .xcl_ops
= &svc_tcp_ops
,
1114 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1117 void svc_init_xprt_sock(void)
1119 svc_reg_xprt_class(&svc_tcp_class
);
1120 svc_reg_xprt_class(&svc_udp_class
);
1123 void svc_cleanup_xprt_sock(void)
1125 svc_unreg_xprt_class(&svc_tcp_class
);
1126 svc_unreg_xprt_class(&svc_udp_class
);
1129 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1131 struct sock
*sk
= svsk
->sk_sk
;
1133 svc_xprt_init(&svc_tcp_class
, &svsk
->sk_xprt
, serv
);
1134 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1135 if (sk
->sk_state
== TCP_LISTEN
) {
1136 dprintk("setting up TCP socket for listening\n");
1137 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1138 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1139 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1141 dprintk("setting up TCP socket for reading\n");
1142 sk
->sk_state_change
= svc_tcp_state_change
;
1143 sk
->sk_data_ready
= svc_tcp_data_ready
;
1144 sk
->sk_write_space
= svc_tcp_write_space
;
1146 svsk
->sk_reclen
= 0;
1147 svsk
->sk_tcplen
= 0;
1149 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1151 /* initialise setting must have enough space to
1152 * receive and respond to one request.
1153 * svc_tcp_recvfrom will re-adjust if necessary
1155 svc_sock_setbufsize(svsk
->sk_sock
,
1156 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
1157 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
1159 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1160 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1161 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1162 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1166 void svc_sock_update_bufs(struct svc_serv
*serv
)
1169 * The number of server threads has changed. Update
1170 * rcvbuf and sndbuf accordingly on all sockets
1172 struct list_head
*le
;
1174 spin_lock_bh(&serv
->sv_lock
);
1175 list_for_each(le
, &serv
->sv_permsocks
) {
1176 struct svc_sock
*svsk
=
1177 list_entry(le
, struct svc_sock
, sk_xprt
.xpt_list
);
1178 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1180 list_for_each(le
, &serv
->sv_tempsocks
) {
1181 struct svc_sock
*svsk
=
1182 list_entry(le
, struct svc_sock
, sk_xprt
.xpt_list
);
1183 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1185 spin_unlock_bh(&serv
->sv_lock
);
1187 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1190 * Initialize socket for RPC use and create svc_sock struct
1191 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1193 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1194 struct socket
*sock
,
1195 int *errp
, int flags
)
1197 struct svc_sock
*svsk
;
1199 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1201 dprintk("svc: svc_setup_socket %p\n", sock
);
1202 if (!(svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
))) {
1209 /* Register socket with portmapper */
1210 if (*errp
>= 0 && pmap_register
)
1211 *errp
= svc_register(serv
, inet
->sk_family
, inet
->sk_protocol
,
1212 ntohs(inet_sk(inet
)->sport
));
1219 inet
->sk_user_data
= svsk
;
1220 svsk
->sk_sock
= sock
;
1222 svsk
->sk_ostate
= inet
->sk_state_change
;
1223 svsk
->sk_odata
= inet
->sk_data_ready
;
1224 svsk
->sk_owspace
= inet
->sk_write_space
;
1226 /* Initialize the socket */
1227 if (sock
->type
== SOCK_DGRAM
)
1228 svc_udp_init(svsk
, serv
);
1230 svc_tcp_init(svsk
, serv
);
1232 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1239 * svc_addsock - add a listener socket to an RPC service
1240 * @serv: pointer to RPC service to which to add a new listener
1241 * @fd: file descriptor of the new listener
1242 * @name_return: pointer to buffer to fill in with name of listener
1243 * @len: size of the buffer
1245 * Fills in socket name and returns positive length of name if successful.
1246 * Name is terminated with '\n'. On error, returns a negative errno
1249 int svc_addsock(struct svc_serv
*serv
, const int fd
, char *name_return
,
1253 struct socket
*so
= sockfd_lookup(fd
, &err
);
1254 struct svc_sock
*svsk
= NULL
;
1258 if (so
->sk
->sk_family
!= AF_INET
)
1259 err
= -EAFNOSUPPORT
;
1260 else if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1261 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1262 err
= -EPROTONOSUPPORT
;
1263 else if (so
->state
> SS_UNCONNECTED
)
1266 if (!try_module_get(THIS_MODULE
))
1269 svsk
= svc_setup_socket(serv
, so
, &err
,
1272 struct sockaddr_storage addr
;
1273 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1275 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1276 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1277 clear_bit(XPT_TEMP
, &svsk
->sk_xprt
.xpt_flags
);
1278 spin_lock_bh(&serv
->sv_lock
);
1279 list_add(&svsk
->sk_xprt
.xpt_list
, &serv
->sv_permsocks
);
1280 spin_unlock_bh(&serv
->sv_lock
);
1281 svc_xprt_received(&svsk
->sk_xprt
);
1284 module_put(THIS_MODULE
);
1290 return svc_one_sock_name(svsk
, name_return
, len
);
1292 EXPORT_SYMBOL_GPL(svc_addsock
);
1295 * Create socket for RPC service.
1297 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1299 struct sockaddr
*sin
, int len
,
1302 struct svc_sock
*svsk
;
1303 struct socket
*sock
;
1306 struct sockaddr_storage addr
;
1307 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1311 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1313 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1314 serv
->sv_program
->pg_name
, protocol
,
1315 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1317 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1318 printk(KERN_WARNING
"svc: only UDP and TCP "
1319 "sockets supported\n");
1320 return ERR_PTR(-EINVAL
);
1323 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1324 switch (sin
->sa_family
) {
1332 return ERR_PTR(-EINVAL
);
1335 error
= sock_create_kern(family
, type
, protocol
, &sock
);
1337 return ERR_PTR(error
);
1339 svc_reclassify_socket(sock
);
1342 * If this is an PF_INET6 listener, we want to avoid
1343 * getting requests from IPv4 remotes. Those should
1344 * be shunted to a PF_INET listener via rpcbind.
1347 if (family
== PF_INET6
)
1348 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1349 (char *)&val
, sizeof(val
));
1351 if (type
== SOCK_STREAM
)
1352 sock
->sk
->sk_reuse
= 1; /* allow address reuse */
1353 error
= kernel_bind(sock
, sin
, len
);
1358 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1362 if (protocol
== IPPROTO_TCP
) {
1363 if ((error
= kernel_listen(sock
, 64)) < 0)
1367 if ((svsk
= svc_setup_socket(serv
, sock
, &error
, flags
)) != NULL
) {
1368 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1369 return (struct svc_xprt
*)svsk
;
1373 dprintk("svc: svc_create_socket error = %d\n", -error
);
1375 return ERR_PTR(error
);
1379 * Detach the svc_sock from the socket so that no
1380 * more callbacks occur.
1382 static void svc_sock_detach(struct svc_xprt
*xprt
)
1384 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1385 struct sock
*sk
= svsk
->sk_sk
;
1387 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1389 /* put back the old socket callbacks */
1390 sk
->sk_state_change
= svsk
->sk_ostate
;
1391 sk
->sk_data_ready
= svsk
->sk_odata
;
1392 sk
->sk_write_space
= svsk
->sk_owspace
;
1394 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1395 wake_up_interruptible(sk
->sk_sleep
);
1399 * Disconnect the socket, and reset the callbacks
1401 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1403 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1405 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1407 svc_sock_detach(xprt
);
1409 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
1410 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1414 * Free the svc_sock's socket resources and the svc_sock itself.
1416 static void svc_sock_free(struct svc_xprt
*xprt
)
1418 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1419 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1421 if (svsk
->sk_sock
->file
)
1422 sockfd_put(svsk
->sk_sock
);
1424 sock_release(svsk
->sk_sock
);
1429 * Create a svc_xprt.
1431 * For internal use only (e.g. nfsv4.1 backchannel).
1432 * Callers should typically use the xpo_create() method.
1434 struct svc_xprt
*svc_sock_create(struct svc_serv
*serv
, int prot
)
1436 struct svc_sock
*svsk
;
1437 struct svc_xprt
*xprt
= NULL
;
1439 dprintk("svc: %s\n", __func__
);
1440 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1444 xprt
= &svsk
->sk_xprt
;
1445 if (prot
== IPPROTO_TCP
)
1446 svc_xprt_init(&svc_tcp_class
, xprt
, serv
);
1447 else if (prot
== IPPROTO_UDP
)
1448 svc_xprt_init(&svc_udp_class
, xprt
, serv
);
1452 dprintk("svc: %s return %p\n", __func__
, xprt
);
1455 EXPORT_SYMBOL_GPL(svc_sock_create
);
1458 * Destroy a svc_sock.
1460 void svc_sock_destroy(struct svc_xprt
*xprt
)
1463 kfree(container_of(xprt
, struct svc_sock
, sk_xprt
));
1465 EXPORT_SYMBOL_GPL(svc_sock_destroy
);