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/module.h>
25 #include <linux/errno.h>
26 #include <linux/fcntl.h>
27 #include <linux/net.h>
29 #include <linux/inet.h>
30 #include <linux/udp.h>
31 #include <linux/tcp.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/netdevice.h>
35 #include <linux/skbuff.h>
36 #include <linux/file.h>
37 #include <linux/freezer.h>
39 #include <net/checksum.h>
43 #include <net/tcp_states.h>
44 #include <asm/uaccess.h>
45 #include <asm/ioctls.h>
46 #include <trace/events/skb.h>
48 #include <linux/sunrpc/types.h>
49 #include <linux/sunrpc/clnt.h>
50 #include <linux/sunrpc/xdr.h>
51 #include <linux/sunrpc/msg_prot.h>
52 #include <linux/sunrpc/svcsock.h>
53 #include <linux/sunrpc/stats.h>
54 #include <linux/sunrpc/xprt.h>
58 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
61 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
63 static void svc_udp_data_ready(struct sock
*, int);
64 static int svc_udp_recvfrom(struct svc_rqst
*);
65 static int svc_udp_sendto(struct svc_rqst
*);
66 static void svc_sock_detach(struct svc_xprt
*);
67 static void svc_tcp_sock_detach(struct svc_xprt
*);
68 static void svc_sock_free(struct svc_xprt
*);
70 static struct svc_xprt
*svc_create_socket(struct svc_serv
*, int,
71 struct net
*, struct sockaddr
*,
73 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
74 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
75 struct net
*, struct sockaddr
*,
77 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
78 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
80 #ifdef CONFIG_DEBUG_LOCK_ALLOC
81 static struct lock_class_key svc_key
[2];
82 static struct lock_class_key svc_slock_key
[2];
84 static void svc_reclassify_socket(struct socket
*sock
)
86 struct sock
*sk
= sock
->sk
;
88 WARN_ON_ONCE(sock_owned_by_user(sk
));
89 if (sock_owned_by_user(sk
))
92 switch (sk
->sk_family
) {
94 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
101 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
112 static void svc_reclassify_socket(struct socket
*sock
)
118 * Release an skbuff after use
120 static void svc_release_skb(struct svc_rqst
*rqstp
)
122 struct sk_buff
*skb
= rqstp
->rq_xprt_ctxt
;
125 struct svc_sock
*svsk
=
126 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
127 rqstp
->rq_xprt_ctxt
= NULL
;
129 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
130 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
134 union svc_pktinfo_u
{
135 struct in_pktinfo pkti
;
136 struct in6_pktinfo pkti6
;
138 #define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
141 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
143 struct svc_sock
*svsk
=
144 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
145 switch (svsk
->sk_sk
->sk_family
) {
147 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
149 cmh
->cmsg_level
= SOL_IP
;
150 cmh
->cmsg_type
= IP_PKTINFO
;
151 pki
->ipi_ifindex
= 0;
152 pki
->ipi_spec_dst
.s_addr
=
153 svc_daddr_in(rqstp
)->sin_addr
.s_addr
;
154 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
159 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
160 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
162 cmh
->cmsg_level
= SOL_IPV6
;
163 cmh
->cmsg_type
= IPV6_PKTINFO
;
164 pki
->ipi6_ifindex
= daddr
->sin6_scope_id
;
165 pki
->ipi6_addr
= daddr
->sin6_addr
;
166 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
173 * send routine intended to be shared by the fore- and back-channel
175 int svc_send_common(struct socket
*sock
, struct xdr_buf
*xdr
,
176 struct page
*headpage
, unsigned long headoffset
,
177 struct page
*tailpage
, unsigned long tailoffset
)
181 struct page
**ppage
= xdr
->pages
;
182 size_t base
= xdr
->page_base
;
183 unsigned int pglen
= xdr
->page_len
;
184 unsigned int flags
= MSG_MORE
;
191 if (slen
== xdr
->head
[0].iov_len
)
193 len
= kernel_sendpage(sock
, headpage
, headoffset
,
194 xdr
->head
[0].iov_len
, flags
);
195 if (len
!= xdr
->head
[0].iov_len
)
197 slen
-= xdr
->head
[0].iov_len
;
202 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
206 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
213 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
219 if (xdr
->tail
[0].iov_len
) {
220 result
= kernel_sendpage(sock
, tailpage
, tailoffset
,
221 xdr
->tail
[0].iov_len
, 0);
232 * Generic sendto routine
234 static int svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
236 struct svc_sock
*svsk
=
237 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
238 struct socket
*sock
= svsk
->sk_sock
;
241 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
243 struct cmsghdr
*cmh
= &buffer
.hdr
;
245 unsigned long tailoff
;
246 unsigned long headoff
;
247 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
249 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
250 struct msghdr msg
= {
251 .msg_name
= &rqstp
->rq_addr
,
252 .msg_namelen
= rqstp
->rq_addrlen
,
254 .msg_controllen
= sizeof(buffer
),
255 .msg_flags
= MSG_MORE
,
258 svc_set_cmsg_data(rqstp
, cmh
);
260 if (sock_sendmsg(sock
, &msg
, 0) < 0)
264 tailoff
= ((unsigned long)xdr
->tail
[0].iov_base
) & (PAGE_SIZE
-1);
266 len
= svc_send_common(sock
, xdr
, rqstp
->rq_respages
[0], headoff
,
267 rqstp
->rq_respages
[0], tailoff
);
270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
271 svsk
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
272 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
278 * Report socket names for nfsdfs
280 static int svc_one_sock_name(struct svc_sock
*svsk
, char *buf
, int remaining
)
282 const struct sock
*sk
= svsk
->sk_sk
;
283 const char *proto_name
= sk
->sk_protocol
== IPPROTO_UDP
?
287 switch (sk
->sk_family
) {
289 len
= snprintf(buf
, remaining
, "ipv4 %s %pI4 %d\n",
291 &inet_sk(sk
)->inet_rcv_saddr
,
292 inet_sk(sk
)->inet_num
);
295 len
= snprintf(buf
, remaining
, "ipv6 %s %pI6 %d\n",
297 &inet6_sk(sk
)->rcv_saddr
,
298 inet_sk(sk
)->inet_num
);
301 len
= snprintf(buf
, remaining
, "*unknown-%d*\n",
305 if (len
>= remaining
) {
307 return -ENAMETOOLONG
;
313 * Check input queue length
315 static int svc_recv_available(struct svc_sock
*svsk
)
317 struct socket
*sock
= svsk
->sk_sock
;
320 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
322 return (err
>= 0)? avail
: err
;
326 * Generic recvfrom routine.
328 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
331 struct svc_sock
*svsk
=
332 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
333 struct msghdr msg
= {
334 .msg_flags
= MSG_DONTWAIT
,
338 rqstp
->rq_xprt_hlen
= 0;
340 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
343 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
344 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
348 static int svc_partial_recvfrom(struct svc_rqst
*rqstp
,
349 struct kvec
*iov
, int nr
,
350 int buflen
, unsigned int base
)
358 return svc_recvfrom(rqstp
, iov
, nr
, buflen
);
360 for (i
= 0; i
< nr
; i
++) {
361 if (iov
[i
].iov_len
> base
)
363 base
-= iov
[i
].iov_len
;
365 save_iovlen
= iov
[i
].iov_len
;
366 save_iovbase
= iov
[i
].iov_base
;
367 iov
[i
].iov_len
-= base
;
368 iov
[i
].iov_base
+= base
;
369 ret
= svc_recvfrom(rqstp
, &iov
[i
], nr
- i
, buflen
);
370 iov
[i
].iov_len
= save_iovlen
;
371 iov
[i
].iov_base
= save_iovbase
;
376 * Set socket snd and rcv buffer lengths
378 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
383 oldfs
= get_fs(); set_fs(KERNEL_DS
);
384 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
385 (char*)&snd
, sizeof(snd
));
386 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
387 (char*)&rcv
, sizeof(rcv
));
389 /* sock_setsockopt limits use to sysctl_?mem_max,
390 * which isn't acceptable. Until that is made conditional
391 * on not having CAP_SYS_RESOURCE or similar, we go direct...
392 * DaveM said I could!
395 sock
->sk
->sk_sndbuf
= snd
* 2;
396 sock
->sk
->sk_rcvbuf
= rcv
* 2;
397 sock
->sk
->sk_write_space(sock
->sk
);
398 release_sock(sock
->sk
);
402 * INET callback when data has been received on the socket.
404 static void svc_udp_data_ready(struct sock
*sk
, int count
)
406 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
407 wait_queue_head_t
*wq
= sk_sleep(sk
);
410 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
412 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
413 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
414 svc_xprt_enqueue(&svsk
->sk_xprt
);
416 if (wq
&& waitqueue_active(wq
))
417 wake_up_interruptible(wq
);
421 * INET callback when space is newly available on the socket.
423 static void svc_write_space(struct sock
*sk
)
425 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
426 wait_queue_head_t
*wq
= sk_sleep(sk
);
429 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
430 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
431 svc_xprt_enqueue(&svsk
->sk_xprt
);
434 if (wq
&& waitqueue_active(wq
)) {
435 dprintk("RPC svc_write_space: someone sleeping on %p\n",
437 wake_up_interruptible(wq
);
441 static void svc_tcp_write_space(struct sock
*sk
)
443 struct socket
*sock
= sk
->sk_socket
;
445 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
) && sock
)
446 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
451 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
453 static int svc_udp_get_dest_address4(struct svc_rqst
*rqstp
,
456 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
457 struct sockaddr_in
*daddr
= svc_daddr_in(rqstp
);
459 if (cmh
->cmsg_type
!= IP_PKTINFO
)
462 daddr
->sin_family
= AF_INET
;
463 daddr
->sin_addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
468 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
470 static int svc_udp_get_dest_address6(struct svc_rqst
*rqstp
,
473 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
474 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
476 if (cmh
->cmsg_type
!= IPV6_PKTINFO
)
479 daddr
->sin6_family
= AF_INET6
;
480 daddr
->sin6_addr
= pki
->ipi6_addr
;
481 daddr
->sin6_scope_id
= pki
->ipi6_ifindex
;
486 * Copy the UDP datagram's destination address to the rqstp structure.
487 * The 'destination' address in this case is the address to which the
488 * peer sent the datagram, i.e. our local address. For multihomed
489 * hosts, this can change from msg to msg. Note that only the IP
490 * address changes, the port number should remain the same.
492 static int svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
495 switch (cmh
->cmsg_level
) {
497 return svc_udp_get_dest_address4(rqstp
, cmh
);
499 return svc_udp_get_dest_address6(rqstp
, cmh
);
506 * Receive a datagram from a UDP socket.
508 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
510 struct svc_sock
*svsk
=
511 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
512 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
516 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
518 struct cmsghdr
*cmh
= &buffer
.hdr
;
519 struct msghdr msg
= {
520 .msg_name
= svc_addr(rqstp
),
522 .msg_controllen
= sizeof(buffer
),
523 .msg_flags
= MSG_DONTWAIT
,
528 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
529 /* udp sockets need large rcvbuf as all pending
530 * requests are still in that buffer. sndbuf must
531 * also be large enough that there is enough space
532 * for one reply per thread. We count all threads
533 * rather than threads in a particular pool, which
534 * provides an upper bound on the number of threads
535 * which will access the socket.
537 svc_sock_setbufsize(svsk
->sk_sock
,
538 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
539 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
541 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
543 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
544 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
546 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
549 if (err
!= -EAGAIN
) {
550 /* possibly an icmp error */
551 dprintk("svc: recvfrom returned error %d\n", -err
);
552 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
556 len
= svc_addr_len(svc_addr(rqstp
));
557 rqstp
->rq_addrlen
= len
;
558 if (skb
->tstamp
.tv64
== 0) {
559 skb
->tstamp
= ktime_get_real();
560 /* Don't enable netstamp, sunrpc doesn't
561 need that much accuracy */
563 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
564 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
566 len
= skb
->len
- sizeof(struct udphdr
);
567 rqstp
->rq_arg
.len
= len
;
569 rqstp
->rq_prot
= IPPROTO_UDP
;
571 if (!svc_udp_get_dest_address(rqstp
, cmh
)) {
572 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
573 cmh
->cmsg_level
, cmh
->cmsg_type
);
576 rqstp
->rq_daddrlen
= svc_addr_len(svc_daddr(rqstp
));
578 if (skb_is_nonlinear(skb
)) {
579 /* we have to copy */
581 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
587 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
589 /* we can use it in-place */
590 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+
591 sizeof(struct udphdr
);
592 rqstp
->rq_arg
.head
[0].iov_len
= len
;
593 if (skb_checksum_complete(skb
))
595 rqstp
->rq_xprt_ctxt
= skb
;
598 rqstp
->rq_arg
.page_base
= 0;
599 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
600 rqstp
->rq_arg
.head
[0].iov_len
= len
;
601 rqstp
->rq_arg
.page_len
= 0;
602 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
604 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
605 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
606 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
608 rqstp
->rq_next_page
= rqstp
->rq_respages
+1;
611 serv
->sv_stats
->netudpcnt
++;
615 trace_kfree_skb(skb
, svc_udp_recvfrom
);
616 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
621 svc_udp_sendto(struct svc_rqst
*rqstp
)
625 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
626 if (error
== -ECONNREFUSED
)
627 /* ICMP error on earlier request. */
628 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
633 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
637 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
639 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
640 struct svc_serv
*serv
= xprt
->xpt_server
;
641 unsigned long required
;
644 * Set the SOCK_NOSPACE flag before checking the available
647 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
648 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
649 if (required
*2 > sock_wspace(svsk
->sk_sk
))
651 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
655 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
661 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
663 struct sockaddr
*sa
, int salen
,
666 return svc_create_socket(serv
, IPPROTO_UDP
, net
, sa
, salen
, flags
);
669 static struct svc_xprt_ops svc_udp_ops
= {
670 .xpo_create
= svc_udp_create
,
671 .xpo_recvfrom
= svc_udp_recvfrom
,
672 .xpo_sendto
= svc_udp_sendto
,
673 .xpo_release_rqst
= svc_release_skb
,
674 .xpo_detach
= svc_sock_detach
,
675 .xpo_free
= svc_sock_free
,
676 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
677 .xpo_has_wspace
= svc_udp_has_wspace
,
678 .xpo_accept
= svc_udp_accept
,
681 static struct svc_xprt_class svc_udp_class
= {
683 .xcl_owner
= THIS_MODULE
,
684 .xcl_ops
= &svc_udp_ops
,
685 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
688 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
690 int err
, level
, optname
, one
= 1;
692 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_udp_class
,
693 &svsk
->sk_xprt
, serv
);
694 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
695 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
696 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
698 /* initialise setting must have enough space to
699 * receive and respond to one request.
700 * svc_udp_recvfrom will re-adjust if necessary
702 svc_sock_setbufsize(svsk
->sk_sock
,
703 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
704 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
706 /* data might have come in before data_ready set up */
707 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
708 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
710 /* make sure we get destination address info */
711 switch (svsk
->sk_sk
->sk_family
) {
714 optname
= IP_PKTINFO
;
718 optname
= IPV6_RECVPKTINFO
;
723 err
= kernel_setsockopt(svsk
->sk_sock
, level
, optname
,
724 (char *)&one
, sizeof(one
));
725 dprintk("svc: kernel_setsockopt returned %d\n", err
);
729 * A data_ready event on a listening socket means there's a connection
730 * pending. Do not use state_change as a substitute for it.
732 static void svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
734 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
735 wait_queue_head_t
*wq
;
737 dprintk("svc: socket %p TCP (listen) state change %d\n",
741 * This callback may called twice when a new connection
742 * is established as a child socket inherits everything
743 * from a parent LISTEN socket.
744 * 1) data_ready method of the parent socket will be called
745 * when one of child sockets become ESTABLISHED.
746 * 2) data_ready method of the child socket may be called
747 * when it receives data before the socket is accepted.
748 * In case of 2, we should ignore it silently.
750 if (sk
->sk_state
== TCP_LISTEN
) {
752 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
753 svc_xprt_enqueue(&svsk
->sk_xprt
);
755 printk("svc: socket %p: no user data\n", sk
);
759 if (wq
&& waitqueue_active(wq
))
760 wake_up_interruptible_all(wq
);
764 * A state change on a connected socket means it's dying or dead.
766 static void svc_tcp_state_change(struct sock
*sk
)
768 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
769 wait_queue_head_t
*wq
= sk_sleep(sk
);
771 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
772 sk
, sk
->sk_state
, sk
->sk_user_data
);
775 printk("svc: socket %p: no user data\n", sk
);
777 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
778 svc_xprt_enqueue(&svsk
->sk_xprt
);
780 if (wq
&& waitqueue_active(wq
))
781 wake_up_interruptible_all(wq
);
784 static void svc_tcp_data_ready(struct sock
*sk
, int count
)
786 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
787 wait_queue_head_t
*wq
= sk_sleep(sk
);
789 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
790 sk
, sk
->sk_user_data
);
792 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
793 svc_xprt_enqueue(&svsk
->sk_xprt
);
795 if (wq
&& waitqueue_active(wq
))
796 wake_up_interruptible(wq
);
800 * Accept a TCP connection
802 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
804 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
805 struct sockaddr_storage addr
;
806 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
807 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
808 struct socket
*sock
= svsk
->sk_sock
;
809 struct socket
*newsock
;
810 struct svc_sock
*newsvsk
;
812 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
814 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
818 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
819 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
822 printk(KERN_WARNING
"%s: no more sockets!\n",
824 else if (err
!= -EAGAIN
)
825 net_warn_ratelimited("%s: accept failed (err %d)!\n",
826 serv
->sv_name
, -err
);
829 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
831 err
= kernel_getpeername(newsock
, sin
, &slen
);
833 net_warn_ratelimited("%s: peername failed (err %d)!\n",
834 serv
->sv_name
, -err
);
835 goto failed
; /* aborted connection or whatever */
838 /* Ideally, we would want to reject connections from unauthorized
839 * hosts here, but when we get encryption, the IP of the host won't
840 * tell us anything. For now just warn about unpriv connections.
842 if (!svc_port_is_privileged(sin
)) {
844 "%s: connect from unprivileged port: %s\n",
846 __svc_print_addr(sin
, buf
, sizeof(buf
)));
848 dprintk("%s: connect from %s\n", serv
->sv_name
,
849 __svc_print_addr(sin
, buf
, sizeof(buf
)));
851 /* make sure that a write doesn't block forever when
854 newsock
->sk
->sk_sndtimeo
= HZ
*30;
856 newsvsk
= svc_setup_socket(serv
, newsock
,
857 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
));
860 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
861 err
= kernel_getsockname(newsock
, sin
, &slen
);
862 if (unlikely(err
< 0)) {
863 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
864 slen
= offsetof(struct sockaddr
, sa_data
);
866 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
869 serv
->sv_stats
->nettcpconn
++;
871 return &newsvsk
->sk_xprt
;
874 sock_release(newsock
);
878 static unsigned int svc_tcp_restore_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
880 unsigned int i
, len
, npages
;
882 if (svsk
->sk_datalen
== 0)
884 len
= svsk
->sk_datalen
;
885 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
886 for (i
= 0; i
< npages
; i
++) {
887 if (rqstp
->rq_pages
[i
] != NULL
)
888 put_page(rqstp
->rq_pages
[i
]);
889 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
890 rqstp
->rq_pages
[i
] = svsk
->sk_pages
[i
];
891 svsk
->sk_pages
[i
] = NULL
;
893 rqstp
->rq_arg
.head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
897 static void svc_tcp_save_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
899 unsigned int i
, len
, npages
;
901 if (svsk
->sk_datalen
== 0)
903 len
= svsk
->sk_datalen
;
904 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
905 for (i
= 0; i
< npages
; i
++) {
906 svsk
->sk_pages
[i
] = rqstp
->rq_pages
[i
];
907 rqstp
->rq_pages
[i
] = NULL
;
911 static void svc_tcp_clear_pages(struct svc_sock
*svsk
)
913 unsigned int i
, len
, npages
;
915 if (svsk
->sk_datalen
== 0)
917 len
= svsk
->sk_datalen
;
918 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
919 for (i
= 0; i
< npages
; i
++) {
920 if (svsk
->sk_pages
[i
] == NULL
) {
924 put_page(svsk
->sk_pages
[i
]);
925 svsk
->sk_pages
[i
] = NULL
;
929 svsk
->sk_datalen
= 0;
933 * Receive fragment record header.
934 * If we haven't gotten the record length yet, get the next four bytes.
936 static int svc_tcp_recv_record(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
938 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
942 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
944 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
947 want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
948 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
950 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
952 svsk
->sk_tcplen
+= len
;
955 dprintk("svc: short recvfrom while reading record "
956 "length (%d of %d)\n", len
, want
);
960 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk
));
961 if (svc_sock_reclen(svsk
) + svsk
->sk_datalen
>
963 net_notice_ratelimited("RPC: fragment too large: %d\n",
964 svc_sock_reclen(svsk
));
969 return svc_sock_reclen(svsk
);
971 dprintk("RPC: TCP recv_record got %d\n", len
);
974 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
978 static int receive_cb_reply(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
980 struct rpc_xprt
*bc_xprt
= svsk
->sk_xprt
.xpt_bc_xprt
;
981 struct rpc_rqst
*req
= NULL
;
982 struct kvec
*src
, *dst
;
983 __be32
*p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
991 req
= xprt_lookup_rqst(bc_xprt
, xid
);
995 "%s: Got unrecognized reply: "
996 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
997 __func__
, ntohl(calldir
),
1002 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
, sizeof(struct xdr_buf
));
1004 * XXX!: cheating for now! Only copying HEAD.
1005 * But we know this is good enough for now (in fact, for any
1006 * callback reply in the forseeable future).
1008 dst
= &req
->rq_private_buf
.head
[0];
1009 src
= &rqstp
->rq_arg
.head
[0];
1010 if (dst
->iov_len
< src
->iov_len
)
1011 return -EAGAIN
; /* whatever; just giving up. */
1012 memcpy(dst
->iov_base
, src
->iov_base
, src
->iov_len
);
1013 xprt_complete_rqst(req
->rq_task
, rqstp
->rq_arg
.len
);
1014 rqstp
->rq_arg
.len
= 0;
1018 static int copy_pages_to_kvecs(struct kvec
*vec
, struct page
**pages
, int len
)
1024 vec
[i
].iov_base
= page_address(pages
[i
]);
1025 vec
[i
].iov_len
= PAGE_SIZE
;
1032 static void svc_tcp_fragment_received(struct svc_sock
*svsk
)
1034 /* If we have more data, signal svc_xprt_enqueue() to try again */
1035 if (svc_recv_available(svsk
) > sizeof(rpc_fraghdr
))
1036 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1037 dprintk("svc: TCP %s record (%d bytes)\n",
1038 svc_sock_final_rec(svsk
) ? "final" : "nonfinal",
1039 svc_sock_reclen(svsk
));
1040 svsk
->sk_tcplen
= 0;
1041 svsk
->sk_reclen
= 0;
1045 * Receive data from a TCP socket.
1047 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1049 struct svc_sock
*svsk
=
1050 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
1051 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1054 unsigned int want
, base
;
1059 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1060 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
1061 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
1062 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1064 len
= svc_tcp_recv_record(svsk
, rqstp
);
1068 base
= svc_tcp_restore_pages(svsk
, rqstp
);
1069 want
= svc_sock_reclen(svsk
) - (svsk
->sk_tcplen
- sizeof(rpc_fraghdr
));
1071 vec
= rqstp
->rq_vec
;
1073 pnum
= copy_pages_to_kvecs(&vec
[0], &rqstp
->rq_pages
[0],
1074 svsk
->sk_datalen
+ want
);
1076 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1077 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
1079 /* Now receive data */
1080 len
= svc_partial_recvfrom(rqstp
, vec
, pnum
, want
, base
);
1082 svsk
->sk_tcplen
+= len
;
1083 svsk
->sk_datalen
+= len
;
1085 if (len
!= want
|| !svc_sock_final_rec(svsk
)) {
1086 svc_tcp_save_pages(svsk
, rqstp
);
1087 if (len
< 0 && len
!= -EAGAIN
)
1090 svc_tcp_fragment_received(svsk
);
1092 dprintk("svc: incomplete TCP record (%d of %d)\n",
1093 (int)(svsk
->sk_tcplen
- sizeof(rpc_fraghdr
)),
1094 svc_sock_reclen(svsk
));
1098 if (svsk
->sk_datalen
< 8) {
1099 svsk
->sk_datalen
= 0;
1100 goto err_delete
; /* client is nuts. */
1103 rqstp
->rq_arg
.len
= svsk
->sk_datalen
;
1104 rqstp
->rq_arg
.page_base
= 0;
1105 if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1106 rqstp
->rq_arg
.head
[0].iov_len
= rqstp
->rq_arg
.len
;
1107 rqstp
->rq_arg
.page_len
= 0;
1109 rqstp
->rq_arg
.page_len
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1111 rqstp
->rq_xprt_ctxt
= NULL
;
1112 rqstp
->rq_prot
= IPPROTO_TCP
;
1114 p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1117 len
= receive_cb_reply(svsk
, rqstp
);
1119 /* Reset TCP read info */
1120 svsk
->sk_datalen
= 0;
1121 svc_tcp_fragment_received(svsk
);
1126 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1128 serv
->sv_stats
->nettcpcnt
++;
1130 return rqstp
->rq_arg
.len
;
1135 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1138 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1139 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1140 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1142 return 0; /* record not complete */
1146 * Send out data on TCP socket.
1148 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
1150 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1154 /* Set up the first element of the reply kvec.
1155 * Any other kvecs that may be in use have been taken
1156 * care of by the server implementation itself.
1158 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1159 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1161 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1162 if (sent
!= xbufp
->len
) {
1164 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1165 "- shutting down socket\n",
1166 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1167 (sent
<0)?"got error":"sent only",
1169 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1170 svc_xprt_enqueue(rqstp
->rq_xprt
);
1177 * Setup response header. TCP has a 4B record length field.
1179 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1181 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1183 /* tcp needs a space for the record length... */
1187 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
1189 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1190 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1193 if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
1195 required
= atomic_read(&xprt
->xpt_reserved
) + serv
->sv_max_mesg
;
1196 if (sk_stream_wspace(svsk
->sk_sk
) >= required
)
1198 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
1202 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1204 struct sockaddr
*sa
, int salen
,
1207 return svc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1210 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1211 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
1212 struct net
*, struct sockaddr
*,
1214 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
1216 static struct svc_xprt
*svc_bc_tcp_create(struct svc_serv
*serv
,
1218 struct sockaddr
*sa
, int salen
,
1221 return svc_bc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1224 static void svc_bc_tcp_sock_detach(struct svc_xprt
*xprt
)
1228 static struct svc_xprt_ops svc_tcp_bc_ops
= {
1229 .xpo_create
= svc_bc_tcp_create
,
1230 .xpo_detach
= svc_bc_tcp_sock_detach
,
1231 .xpo_free
= svc_bc_sock_free
,
1232 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1235 static struct svc_xprt_class svc_tcp_bc_class
= {
1236 .xcl_name
= "tcp-bc",
1237 .xcl_owner
= THIS_MODULE
,
1238 .xcl_ops
= &svc_tcp_bc_ops
,
1239 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1242 static void svc_init_bc_xprt_sock(void)
1244 svc_reg_xprt_class(&svc_tcp_bc_class
);
1247 static void svc_cleanup_bc_xprt_sock(void)
1249 svc_unreg_xprt_class(&svc_tcp_bc_class
);
1251 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1252 static void svc_init_bc_xprt_sock(void)
1256 static void svc_cleanup_bc_xprt_sock(void)
1259 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1261 static struct svc_xprt_ops svc_tcp_ops
= {
1262 .xpo_create
= svc_tcp_create
,
1263 .xpo_recvfrom
= svc_tcp_recvfrom
,
1264 .xpo_sendto
= svc_tcp_sendto
,
1265 .xpo_release_rqst
= svc_release_skb
,
1266 .xpo_detach
= svc_tcp_sock_detach
,
1267 .xpo_free
= svc_sock_free
,
1268 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1269 .xpo_has_wspace
= svc_tcp_has_wspace
,
1270 .xpo_accept
= svc_tcp_accept
,
1273 static struct svc_xprt_class svc_tcp_class
= {
1275 .xcl_owner
= THIS_MODULE
,
1276 .xcl_ops
= &svc_tcp_ops
,
1277 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1280 void svc_init_xprt_sock(void)
1282 svc_reg_xprt_class(&svc_tcp_class
);
1283 svc_reg_xprt_class(&svc_udp_class
);
1284 svc_init_bc_xprt_sock();
1287 void svc_cleanup_xprt_sock(void)
1289 svc_unreg_xprt_class(&svc_tcp_class
);
1290 svc_unreg_xprt_class(&svc_udp_class
);
1291 svc_cleanup_bc_xprt_sock();
1294 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1296 struct sock
*sk
= svsk
->sk_sk
;
1298 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_tcp_class
,
1299 &svsk
->sk_xprt
, serv
);
1300 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1301 if (sk
->sk_state
== TCP_LISTEN
) {
1302 dprintk("setting up TCP socket for listening\n");
1303 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1304 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1305 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1307 dprintk("setting up TCP socket for reading\n");
1308 sk
->sk_state_change
= svc_tcp_state_change
;
1309 sk
->sk_data_ready
= svc_tcp_data_ready
;
1310 sk
->sk_write_space
= svc_tcp_write_space
;
1312 svsk
->sk_reclen
= 0;
1313 svsk
->sk_tcplen
= 0;
1314 svsk
->sk_datalen
= 0;
1315 memset(&svsk
->sk_pages
[0], 0, sizeof(svsk
->sk_pages
));
1317 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1319 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1320 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1321 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1325 void svc_sock_update_bufs(struct svc_serv
*serv
)
1328 * The number of server threads has changed. Update
1329 * rcvbuf and sndbuf accordingly on all sockets
1331 struct svc_sock
*svsk
;
1333 spin_lock_bh(&serv
->sv_lock
);
1334 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
)
1335 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1336 spin_unlock_bh(&serv
->sv_lock
);
1338 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1341 * Initialize socket for RPC use and create svc_sock struct
1342 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1344 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1345 struct socket
*sock
,
1348 struct svc_sock
*svsk
;
1350 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1353 dprintk("svc: svc_setup_socket %p\n", sock
);
1354 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1356 return ERR_PTR(-ENOMEM
);
1360 /* Register socket with portmapper */
1362 err
= svc_register(serv
, sock_net(sock
->sk
), inet
->sk_family
,
1364 ntohs(inet_sk(inet
)->inet_sport
));
1368 return ERR_PTR(err
);
1371 inet
->sk_user_data
= svsk
;
1372 svsk
->sk_sock
= sock
;
1374 svsk
->sk_ostate
= inet
->sk_state_change
;
1375 svsk
->sk_odata
= inet
->sk_data_ready
;
1376 svsk
->sk_owspace
= inet
->sk_write_space
;
1378 /* Initialize the socket */
1379 if (sock
->type
== SOCK_DGRAM
)
1380 svc_udp_init(svsk
, serv
);
1382 /* initialise setting must have enough space to
1383 * receive and respond to one request.
1385 svc_sock_setbufsize(svsk
->sk_sock
, 4 * serv
->sv_max_mesg
,
1386 4 * serv
->sv_max_mesg
);
1387 svc_tcp_init(svsk
, serv
);
1390 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1397 * svc_addsock - add a listener socket to an RPC service
1398 * @serv: pointer to RPC service to which to add a new listener
1399 * @fd: file descriptor of the new listener
1400 * @name_return: pointer to buffer to fill in with name of listener
1401 * @len: size of the buffer
1403 * Fills in socket name and returns positive length of name if successful.
1404 * Name is terminated with '\n'. On error, returns a negative errno
1407 int svc_addsock(struct svc_serv
*serv
, const int fd
, char *name_return
,
1411 struct socket
*so
= sockfd_lookup(fd
, &err
);
1412 struct svc_sock
*svsk
= NULL
;
1413 struct sockaddr_storage addr
;
1414 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1419 err
= -EAFNOSUPPORT
;
1420 if ((so
->sk
->sk_family
!= PF_INET
) && (so
->sk
->sk_family
!= PF_INET6
))
1422 err
= -EPROTONOSUPPORT
;
1423 if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1424 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1427 if (so
->state
> SS_UNCONNECTED
)
1430 if (!try_module_get(THIS_MODULE
))
1432 svsk
= svc_setup_socket(serv
, so
, SVC_SOCK_DEFAULTS
);
1434 module_put(THIS_MODULE
);
1435 err
= PTR_ERR(svsk
);
1438 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1439 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1440 svc_add_new_perm_xprt(serv
, &svsk
->sk_xprt
);
1441 return svc_one_sock_name(svsk
, name_return
, len
);
1446 EXPORT_SYMBOL_GPL(svc_addsock
);
1449 * Create socket for RPC service.
1451 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1454 struct sockaddr
*sin
, int len
,
1457 struct svc_sock
*svsk
;
1458 struct socket
*sock
;
1461 struct sockaddr_storage addr
;
1462 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1466 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1468 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1469 serv
->sv_program
->pg_name
, protocol
,
1470 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1472 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1473 printk(KERN_WARNING
"svc: only UDP and TCP "
1474 "sockets supported\n");
1475 return ERR_PTR(-EINVAL
);
1478 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1479 switch (sin
->sa_family
) {
1487 return ERR_PTR(-EINVAL
);
1490 error
= __sock_create(net
, family
, type
, protocol
, &sock
, 1);
1492 return ERR_PTR(error
);
1494 svc_reclassify_socket(sock
);
1497 * If this is an PF_INET6 listener, we want to avoid
1498 * getting requests from IPv4 remotes. Those should
1499 * be shunted to a PF_INET listener via rpcbind.
1502 if (family
== PF_INET6
)
1503 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1504 (char *)&val
, sizeof(val
));
1506 if (type
== SOCK_STREAM
)
1507 sock
->sk
->sk_reuse
= SK_CAN_REUSE
; /* allow address reuse */
1508 error
= kernel_bind(sock
, sin
, len
);
1513 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1517 if (protocol
== IPPROTO_TCP
) {
1518 if ((error
= kernel_listen(sock
, 64)) < 0)
1522 svsk
= svc_setup_socket(serv
, sock
, flags
);
1524 error
= PTR_ERR(svsk
);
1527 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1528 return (struct svc_xprt
*)svsk
;
1530 dprintk("svc: svc_create_socket error = %d\n", -error
);
1532 return ERR_PTR(error
);
1536 * Detach the svc_sock from the socket so that no
1537 * more callbacks occur.
1539 static void svc_sock_detach(struct svc_xprt
*xprt
)
1541 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1542 struct sock
*sk
= svsk
->sk_sk
;
1543 wait_queue_head_t
*wq
;
1545 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1547 /* put back the old socket callbacks */
1548 sk
->sk_state_change
= svsk
->sk_ostate
;
1549 sk
->sk_data_ready
= svsk
->sk_odata
;
1550 sk
->sk_write_space
= svsk
->sk_owspace
;
1553 if (wq
&& waitqueue_active(wq
))
1554 wake_up_interruptible(wq
);
1558 * Disconnect the socket, and reset the callbacks
1560 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1562 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1564 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1566 svc_sock_detach(xprt
);
1568 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
)) {
1569 svc_tcp_clear_pages(svsk
);
1570 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1575 * Free the svc_sock's socket resources and the svc_sock itself.
1577 static void svc_sock_free(struct svc_xprt
*xprt
)
1579 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1580 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1582 if (svsk
->sk_sock
->file
)
1583 sockfd_put(svsk
->sk_sock
);
1585 sock_release(svsk
->sk_sock
);
1589 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1591 * Create a back channel svc_xprt which shares the fore channel socket.
1593 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*serv
,
1596 struct sockaddr
*sin
, int len
,
1599 struct svc_sock
*svsk
;
1600 struct svc_xprt
*xprt
;
1602 if (protocol
!= IPPROTO_TCP
) {
1603 printk(KERN_WARNING
"svc: only TCP sockets"
1604 " supported on shared back channel\n");
1605 return ERR_PTR(-EINVAL
);
1608 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1610 return ERR_PTR(-ENOMEM
);
1612 xprt
= &svsk
->sk_xprt
;
1613 svc_xprt_init(net
, &svc_tcp_bc_class
, xprt
, serv
);
1615 serv
->sv_bc_xprt
= xprt
;
1621 * Free a back channel svc_sock.
1623 static void svc_bc_sock_free(struct svc_xprt
*xprt
)
1626 kfree(container_of(xprt
, struct svc_sock
, sk_xprt
));
1628 #endif /* CONFIG_SUNRPC_BACKCHANNEL */