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
;
87 BUG_ON(sock_owned_by_user(sk
));
88 switch (sk
->sk_family
) {
90 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
92 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
99 "sk_xprt.xpt_lock-AF_INET6-NFSD",
108 static void svc_reclassify_socket(struct socket
*sock
)
114 * Release an skbuff after use
116 static void svc_release_skb(struct svc_rqst
*rqstp
)
118 struct sk_buff
*skb
= rqstp
->rq_xprt_ctxt
;
121 struct svc_sock
*svsk
=
122 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
123 rqstp
->rq_xprt_ctxt
= NULL
;
125 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
126 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
130 union svc_pktinfo_u
{
131 struct in_pktinfo pkti
;
132 struct in6_pktinfo pkti6
;
134 #define SVC_PKTINFO_SPACE \
135 CMSG_SPACE(sizeof(union svc_pktinfo_u))
137 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
139 struct svc_sock
*svsk
=
140 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
141 switch (svsk
->sk_sk
->sk_family
) {
143 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
145 cmh
->cmsg_level
= SOL_IP
;
146 cmh
->cmsg_type
= IP_PKTINFO
;
147 pki
->ipi_ifindex
= 0;
148 pki
->ipi_spec_dst
.s_addr
=
149 svc_daddr_in(rqstp
)->sin_addr
.s_addr
;
150 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
155 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
156 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
158 cmh
->cmsg_level
= SOL_IPV6
;
159 cmh
->cmsg_type
= IPV6_PKTINFO
;
160 pki
->ipi6_ifindex
= daddr
->sin6_scope_id
;
161 pki
->ipi6_addr
= daddr
->sin6_addr
;
162 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
169 * send routine intended to be shared by the fore- and back-channel
171 int svc_send_common(struct socket
*sock
, struct xdr_buf
*xdr
,
172 struct page
*headpage
, unsigned long headoffset
,
173 struct page
*tailpage
, unsigned long tailoffset
)
177 struct page
**ppage
= xdr
->pages
;
178 size_t base
= xdr
->page_base
;
179 unsigned int pglen
= xdr
->page_len
;
180 unsigned int flags
= MSG_MORE
;
187 if (slen
== xdr
->head
[0].iov_len
)
189 len
= kernel_sendpage(sock
, headpage
, headoffset
,
190 xdr
->head
[0].iov_len
, flags
);
191 if (len
!= xdr
->head
[0].iov_len
)
193 slen
-= xdr
->head
[0].iov_len
;
198 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
202 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
209 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
215 if (xdr
->tail
[0].iov_len
) {
216 result
= kernel_sendpage(sock
, tailpage
, tailoffset
,
217 xdr
->tail
[0].iov_len
, 0);
228 * Generic sendto routine
230 static int svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
232 struct svc_sock
*svsk
=
233 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
234 struct socket
*sock
= svsk
->sk_sock
;
237 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
239 struct cmsghdr
*cmh
= &buffer
.hdr
;
241 unsigned long tailoff
;
242 unsigned long headoff
;
243 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
245 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
246 struct msghdr msg
= {
247 .msg_name
= &rqstp
->rq_addr
,
248 .msg_namelen
= rqstp
->rq_addrlen
,
250 .msg_controllen
= sizeof(buffer
),
251 .msg_flags
= MSG_MORE
,
254 svc_set_cmsg_data(rqstp
, cmh
);
256 if (sock_sendmsg(sock
, &msg
, 0) < 0)
260 tailoff
= ((unsigned long)xdr
->tail
[0].iov_base
) & (PAGE_SIZE
-1);
262 len
= svc_send_common(sock
, xdr
, rqstp
->rq_respages
[0], headoff
,
263 rqstp
->rq_respages
[0], tailoff
);
266 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
267 svsk
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
268 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
274 * Report socket names for nfsdfs
276 static int svc_one_sock_name(struct svc_sock
*svsk
, char *buf
, int remaining
)
278 const struct sock
*sk
= svsk
->sk_sk
;
279 const char *proto_name
= sk
->sk_protocol
== IPPROTO_UDP
?
283 switch (sk
->sk_family
) {
285 len
= snprintf(buf
, remaining
, "ipv4 %s %pI4 %d\n",
287 &inet_sk(sk
)->inet_rcv_saddr
,
288 inet_sk(sk
)->inet_num
);
291 len
= snprintf(buf
, remaining
, "ipv6 %s %pI6 %d\n",
293 &inet6_sk(sk
)->rcv_saddr
,
294 inet_sk(sk
)->inet_num
);
297 len
= snprintf(buf
, remaining
, "*unknown-%d*\n",
301 if (len
>= remaining
) {
303 return -ENAMETOOLONG
;
309 * Check input queue length
311 static int svc_recv_available(struct svc_sock
*svsk
)
313 struct socket
*sock
= svsk
->sk_sock
;
316 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
318 return (err
>= 0)? avail
: err
;
322 * Generic recvfrom routine.
324 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
327 struct svc_sock
*svsk
=
328 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
329 struct msghdr msg
= {
330 .msg_flags
= MSG_DONTWAIT
,
334 rqstp
->rq_xprt_hlen
= 0;
336 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
339 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
340 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
344 static int svc_partial_recvfrom(struct svc_rqst
*rqstp
,
345 struct kvec
*iov
, int nr
,
346 int buflen
, unsigned int base
)
354 return svc_recvfrom(rqstp
, iov
, nr
, buflen
);
356 for (i
= 0; i
< nr
; i
++) {
357 if (iov
[i
].iov_len
> base
)
359 base
-= iov
[i
].iov_len
;
361 save_iovlen
= iov
[i
].iov_len
;
362 save_iovbase
= iov
[i
].iov_base
;
363 iov
[i
].iov_len
-= base
;
364 iov
[i
].iov_base
+= base
;
365 ret
= svc_recvfrom(rqstp
, &iov
[i
], nr
- i
, buflen
);
366 iov
[i
].iov_len
= save_iovlen
;
367 iov
[i
].iov_base
= save_iovbase
;
372 * Set socket snd and rcv buffer lengths
374 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
379 oldfs
= get_fs(); set_fs(KERNEL_DS
);
380 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
381 (char*)&snd
, sizeof(snd
));
382 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
383 (char*)&rcv
, sizeof(rcv
));
385 /* sock_setsockopt limits use to sysctl_?mem_max,
386 * which isn't acceptable. Until that is made conditional
387 * on not having CAP_SYS_RESOURCE or similar, we go direct...
388 * DaveM said I could!
391 sock
->sk
->sk_sndbuf
= snd
* 2;
392 sock
->sk
->sk_rcvbuf
= rcv
* 2;
393 sock
->sk
->sk_write_space(sock
->sk
);
394 release_sock(sock
->sk
);
398 * INET callback when data has been received on the socket.
400 static void svc_udp_data_ready(struct sock
*sk
, int count
)
402 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
403 wait_queue_head_t
*wq
= sk_sleep(sk
);
406 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
408 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
409 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
410 svc_xprt_enqueue(&svsk
->sk_xprt
);
412 if (wq
&& waitqueue_active(wq
))
413 wake_up_interruptible(wq
);
417 * INET callback when space is newly available on the socket.
419 static void svc_write_space(struct sock
*sk
)
421 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
422 wait_queue_head_t
*wq
= sk_sleep(sk
);
425 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
426 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
427 svc_xprt_enqueue(&svsk
->sk_xprt
);
430 if (wq
&& waitqueue_active(wq
)) {
431 dprintk("RPC svc_write_space: someone sleeping on %p\n",
433 wake_up_interruptible(wq
);
437 static void svc_tcp_write_space(struct sock
*sk
)
439 struct socket
*sock
= sk
->sk_socket
;
441 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
) && sock
)
442 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
447 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
449 static int svc_udp_get_dest_address4(struct svc_rqst
*rqstp
,
452 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
453 struct sockaddr_in
*daddr
= svc_daddr_in(rqstp
);
455 if (cmh
->cmsg_type
!= IP_PKTINFO
)
458 daddr
->sin_family
= AF_INET
;
459 daddr
->sin_addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
464 * See net/ipv6/datagram.c : datagram_recv_ctl
466 static int svc_udp_get_dest_address6(struct svc_rqst
*rqstp
,
469 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
470 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
472 if (cmh
->cmsg_type
!= IPV6_PKTINFO
)
475 daddr
->sin6_family
= AF_INET6
;
476 daddr
->sin6_addr
= pki
->ipi6_addr
;
477 daddr
->sin6_scope_id
= pki
->ipi6_ifindex
;
482 * Copy the UDP datagram's destination address to the rqstp structure.
483 * The 'destination' address in this case is the address to which the
484 * peer sent the datagram, i.e. our local address. For multihomed
485 * hosts, this can change from msg to msg. Note that only the IP
486 * address changes, the port number should remain the same.
488 static int svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
491 switch (cmh
->cmsg_level
) {
493 return svc_udp_get_dest_address4(rqstp
, cmh
);
495 return svc_udp_get_dest_address6(rqstp
, cmh
);
502 * Receive a datagram from a UDP socket.
504 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
506 struct svc_sock
*svsk
=
507 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
508 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
512 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
514 struct cmsghdr
*cmh
= &buffer
.hdr
;
515 struct msghdr msg
= {
516 .msg_name
= svc_addr(rqstp
),
518 .msg_controllen
= sizeof(buffer
),
519 .msg_flags
= MSG_DONTWAIT
,
524 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
525 /* udp sockets need large rcvbuf as all pending
526 * requests are still in that buffer. sndbuf must
527 * also be large enough that there is enough space
528 * for one reply per thread. We count all threads
529 * rather than threads in a particular pool, which
530 * provides an upper bound on the number of threads
531 * which will access the socket.
533 svc_sock_setbufsize(svsk
->sk_sock
,
534 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
535 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
537 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
539 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
540 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
542 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
545 if (err
!= -EAGAIN
) {
546 /* possibly an icmp error */
547 dprintk("svc: recvfrom returned error %d\n", -err
);
548 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
552 len
= svc_addr_len(svc_addr(rqstp
));
553 rqstp
->rq_addrlen
= len
;
554 if (skb
->tstamp
.tv64
== 0) {
555 skb
->tstamp
= ktime_get_real();
556 /* Don't enable netstamp, sunrpc doesn't
557 need that much accuracy */
559 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
560 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
562 len
= skb
->len
- sizeof(struct udphdr
);
563 rqstp
->rq_arg
.len
= len
;
565 rqstp
->rq_prot
= IPPROTO_UDP
;
567 if (!svc_udp_get_dest_address(rqstp
, cmh
)) {
568 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
569 cmh
->cmsg_level
, cmh
->cmsg_type
);
572 rqstp
->rq_daddrlen
= svc_addr_len(svc_daddr(rqstp
));
574 if (skb_is_nonlinear(skb
)) {
575 /* we have to copy */
577 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
583 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
585 /* we can use it in-place */
586 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+
587 sizeof(struct udphdr
);
588 rqstp
->rq_arg
.head
[0].iov_len
= len
;
589 if (skb_checksum_complete(skb
))
591 rqstp
->rq_xprt_ctxt
= skb
;
594 rqstp
->rq_arg
.page_base
= 0;
595 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
596 rqstp
->rq_arg
.head
[0].iov_len
= len
;
597 rqstp
->rq_arg
.page_len
= 0;
598 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
600 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
601 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
602 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
606 serv
->sv_stats
->netudpcnt
++;
610 trace_kfree_skb(skb
, svc_udp_recvfrom
);
611 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
616 svc_udp_sendto(struct svc_rqst
*rqstp
)
620 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
621 if (error
== -ECONNREFUSED
)
622 /* ICMP error on earlier request. */
623 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
628 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
632 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
634 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
635 struct svc_serv
*serv
= xprt
->xpt_server
;
636 unsigned long required
;
639 * Set the SOCK_NOSPACE flag before checking the available
642 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
643 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
644 if (required
*2 > sock_wspace(svsk
->sk_sk
))
646 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
650 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
656 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
658 struct sockaddr
*sa
, int salen
,
661 return svc_create_socket(serv
, IPPROTO_UDP
, net
, sa
, salen
, flags
);
664 static struct svc_xprt_ops svc_udp_ops
= {
665 .xpo_create
= svc_udp_create
,
666 .xpo_recvfrom
= svc_udp_recvfrom
,
667 .xpo_sendto
= svc_udp_sendto
,
668 .xpo_release_rqst
= svc_release_skb
,
669 .xpo_detach
= svc_sock_detach
,
670 .xpo_free
= svc_sock_free
,
671 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
672 .xpo_has_wspace
= svc_udp_has_wspace
,
673 .xpo_accept
= svc_udp_accept
,
676 static struct svc_xprt_class svc_udp_class
= {
678 .xcl_owner
= THIS_MODULE
,
679 .xcl_ops
= &svc_udp_ops
,
680 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
683 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
685 int err
, level
, optname
, one
= 1;
687 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_udp_class
,
688 &svsk
->sk_xprt
, serv
);
689 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
690 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
691 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
693 /* initialise setting must have enough space to
694 * receive and respond to one request.
695 * svc_udp_recvfrom will re-adjust if necessary
697 svc_sock_setbufsize(svsk
->sk_sock
,
698 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
699 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
701 /* data might have come in before data_ready set up */
702 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
703 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
705 /* make sure we get destination address info */
706 switch (svsk
->sk_sk
->sk_family
) {
709 optname
= IP_PKTINFO
;
713 optname
= IPV6_RECVPKTINFO
;
718 err
= kernel_setsockopt(svsk
->sk_sock
, level
, optname
,
719 (char *)&one
, sizeof(one
));
720 dprintk("svc: kernel_setsockopt returned %d\n", err
);
724 * A data_ready event on a listening socket means there's a connection
725 * pending. Do not use state_change as a substitute for it.
727 static void svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
729 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
730 wait_queue_head_t
*wq
;
732 dprintk("svc: socket %p TCP (listen) state change %d\n",
736 * This callback may called twice when a new connection
737 * is established as a child socket inherits everything
738 * from a parent LISTEN socket.
739 * 1) data_ready method of the parent socket will be called
740 * when one of child sockets become ESTABLISHED.
741 * 2) data_ready method of the child socket may be called
742 * when it receives data before the socket is accepted.
743 * In case of 2, we should ignore it silently.
745 if (sk
->sk_state
== TCP_LISTEN
) {
747 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
748 svc_xprt_enqueue(&svsk
->sk_xprt
);
750 printk("svc: socket %p: no user data\n", sk
);
754 if (wq
&& waitqueue_active(wq
))
755 wake_up_interruptible_all(wq
);
759 * A state change on a connected socket means it's dying or dead.
761 static void svc_tcp_state_change(struct sock
*sk
)
763 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
764 wait_queue_head_t
*wq
= sk_sleep(sk
);
766 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
767 sk
, sk
->sk_state
, sk
->sk_user_data
);
770 printk("svc: socket %p: no user data\n", sk
);
772 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
773 svc_xprt_enqueue(&svsk
->sk_xprt
);
775 if (wq
&& waitqueue_active(wq
))
776 wake_up_interruptible_all(wq
);
779 static void svc_tcp_data_ready(struct sock
*sk
, int count
)
781 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
782 wait_queue_head_t
*wq
= sk_sleep(sk
);
784 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
785 sk
, sk
->sk_user_data
);
787 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
788 svc_xprt_enqueue(&svsk
->sk_xprt
);
790 if (wq
&& waitqueue_active(wq
))
791 wake_up_interruptible(wq
);
795 * Accept a TCP connection
797 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
799 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
800 struct sockaddr_storage addr
;
801 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
802 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
803 struct socket
*sock
= svsk
->sk_sock
;
804 struct socket
*newsock
;
805 struct svc_sock
*newsvsk
;
807 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
809 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
813 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
814 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
817 printk(KERN_WARNING
"%s: no more sockets!\n",
819 else if (err
!= -EAGAIN
)
820 net_warn_ratelimited("%s: accept failed (err %d)!\n",
821 serv
->sv_name
, -err
);
824 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
826 err
= kernel_getpeername(newsock
, sin
, &slen
);
828 net_warn_ratelimited("%s: peername failed (err %d)!\n",
829 serv
->sv_name
, -err
);
830 goto failed
; /* aborted connection or whatever */
833 /* Ideally, we would want to reject connections from unauthorized
834 * hosts here, but when we get encryption, the IP of the host won't
835 * tell us anything. For now just warn about unpriv connections.
837 if (!svc_port_is_privileged(sin
)) {
839 "%s: connect from unprivileged port: %s\n",
841 __svc_print_addr(sin
, buf
, sizeof(buf
)));
843 dprintk("%s: connect from %s\n", serv
->sv_name
,
844 __svc_print_addr(sin
, buf
, sizeof(buf
)));
846 /* make sure that a write doesn't block forever when
849 newsock
->sk
->sk_sndtimeo
= HZ
*30;
851 newsvsk
= svc_setup_socket(serv
, newsock
,
852 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
));
855 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
856 err
= kernel_getsockname(newsock
, sin
, &slen
);
857 if (unlikely(err
< 0)) {
858 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
859 slen
= offsetof(struct sockaddr
, sa_data
);
861 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
864 serv
->sv_stats
->nettcpconn
++;
866 return &newsvsk
->sk_xprt
;
869 sock_release(newsock
);
873 static unsigned int svc_tcp_restore_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
875 unsigned int i
, len
, npages
;
877 if (svsk
->sk_tcplen
<= sizeof(rpc_fraghdr
))
879 len
= svsk
->sk_tcplen
- sizeof(rpc_fraghdr
);
880 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
881 for (i
= 0; i
< npages
; i
++) {
882 if (rqstp
->rq_pages
[i
] != NULL
)
883 put_page(rqstp
->rq_pages
[i
]);
884 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
885 rqstp
->rq_pages
[i
] = svsk
->sk_pages
[i
];
886 svsk
->sk_pages
[i
] = NULL
;
888 rqstp
->rq_arg
.head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
892 static void svc_tcp_save_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
894 unsigned int i
, len
, npages
;
896 if (svsk
->sk_tcplen
<= sizeof(rpc_fraghdr
))
898 len
= svsk
->sk_tcplen
- sizeof(rpc_fraghdr
);
899 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
900 for (i
= 0; i
< npages
; i
++) {
901 svsk
->sk_pages
[i
] = rqstp
->rq_pages
[i
];
902 rqstp
->rq_pages
[i
] = NULL
;
906 static void svc_tcp_clear_pages(struct svc_sock
*svsk
)
908 unsigned int i
, len
, npages
;
910 if (svsk
->sk_tcplen
<= sizeof(rpc_fraghdr
))
912 len
= svsk
->sk_tcplen
- sizeof(rpc_fraghdr
);
913 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
914 for (i
= 0; i
< npages
; i
++) {
915 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
916 put_page(svsk
->sk_pages
[i
]);
917 svsk
->sk_pages
[i
] = NULL
;
925 * If we haven't gotten the record length yet, get the next four bytes.
926 * Otherwise try to gobble up as much as possible up to the complete
929 static int svc_tcp_recv_record(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
931 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
935 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
937 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
940 want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
941 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
943 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
945 svsk
->sk_tcplen
+= len
;
948 dprintk("svc: short recvfrom while reading record "
949 "length (%d of %d)\n", len
, want
);
953 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
954 if (!(svsk
->sk_reclen
& RPC_LAST_STREAM_FRAGMENT
)) {
955 /* FIXME: technically, a record can be fragmented,
956 * and non-terminal fragments will not have the top
957 * bit set in the fragment length header.
958 * But apparently no known nfs clients send fragmented
960 net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
964 svsk
->sk_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
965 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
966 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
967 net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
968 (unsigned long)svsk
->sk_reclen
);
973 if (svsk
->sk_reclen
< 8)
974 goto err_delete
; /* client is nuts. */
976 len
= svsk
->sk_reclen
;
980 dprintk("RPC: TCP recv_record got %d\n", len
);
983 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
987 static int receive_cb_reply(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
989 struct rpc_xprt
*bc_xprt
= svsk
->sk_xprt
.xpt_bc_xprt
;
990 struct rpc_rqst
*req
= NULL
;
991 struct kvec
*src
, *dst
;
992 __be32
*p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1000 req
= xprt_lookup_rqst(bc_xprt
, xid
);
1004 "%s: Got unrecognized reply: "
1005 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1006 __func__
, ntohl(calldir
),
1011 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
, sizeof(struct xdr_buf
));
1013 * XXX!: cheating for now! Only copying HEAD.
1014 * But we know this is good enough for now (in fact, for any
1015 * callback reply in the forseeable future).
1017 dst
= &req
->rq_private_buf
.head
[0];
1018 src
= &rqstp
->rq_arg
.head
[0];
1019 if (dst
->iov_len
< src
->iov_len
)
1020 return -EAGAIN
; /* whatever; just giving up. */
1021 memcpy(dst
->iov_base
, src
->iov_base
, src
->iov_len
);
1022 xprt_complete_rqst(req
->rq_task
, svsk
->sk_reclen
);
1023 rqstp
->rq_arg
.len
= 0;
1027 static int copy_pages_to_kvecs(struct kvec
*vec
, struct page
**pages
, int len
)
1033 vec
[i
].iov_base
= page_address(pages
[i
]);
1034 vec
[i
].iov_len
= PAGE_SIZE
;
1043 * Receive data from a TCP socket.
1045 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1047 struct svc_sock
*svsk
=
1048 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
1049 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1052 unsigned int want
, base
;
1057 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1058 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
1059 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
1060 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1062 len
= svc_tcp_recv_record(svsk
, rqstp
);
1066 base
= svc_tcp_restore_pages(svsk
, rqstp
);
1067 want
= svsk
->sk_reclen
- base
;
1069 vec
= rqstp
->rq_vec
;
1071 pnum
= copy_pages_to_kvecs(&vec
[0], &rqstp
->rq_pages
[0],
1074 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1076 /* Now receive data */
1077 len
= svc_partial_recvfrom(rqstp
, vec
, pnum
, want
, base
);
1079 svsk
->sk_tcplen
+= len
;
1081 svc_tcp_save_pages(svsk
, rqstp
);
1082 if (len
< 0 && len
!= -EAGAIN
)
1084 dprintk("svc: incomplete TCP record (%d of %d)\n",
1085 svsk
->sk_tcplen
, svsk
->sk_reclen
);
1089 rqstp
->rq_arg
.len
= svsk
->sk_reclen
;
1090 rqstp
->rq_arg
.page_base
= 0;
1091 if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1092 rqstp
->rq_arg
.head
[0].iov_len
= rqstp
->rq_arg
.len
;
1093 rqstp
->rq_arg
.page_len
= 0;
1095 rqstp
->rq_arg
.page_len
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1097 rqstp
->rq_xprt_ctxt
= NULL
;
1098 rqstp
->rq_prot
= IPPROTO_TCP
;
1100 p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1103 len
= receive_cb_reply(svsk
, rqstp
);
1105 /* Reset TCP read info */
1106 svsk
->sk_reclen
= 0;
1107 svsk
->sk_tcplen
= 0;
1108 /* If we have more data, signal svc_xprt_enqueue() to try again */
1109 if (svc_recv_available(svsk
) > sizeof(rpc_fraghdr
))
1110 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1115 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1117 serv
->sv_stats
->nettcpcnt
++;
1119 dprintk("svc: TCP complete record (%d bytes)\n", rqstp
->rq_arg
.len
);
1120 return rqstp
->rq_arg
.len
;
1125 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1128 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1129 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1130 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1132 return 0; /* record not complete */
1136 * Send out data on TCP socket.
1138 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
1140 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1144 /* Set up the first element of the reply kvec.
1145 * Any other kvecs that may be in use have been taken
1146 * care of by the server implementation itself.
1148 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1149 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1151 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1152 if (sent
!= xbufp
->len
) {
1154 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1155 "- shutting down socket\n",
1156 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1157 (sent
<0)?"got error":"sent only",
1159 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1160 svc_xprt_enqueue(rqstp
->rq_xprt
);
1167 * Setup response header. TCP has a 4B record length field.
1169 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1171 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1173 /* tcp needs a space for the record length... */
1177 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
1179 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1180 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1183 if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
1185 required
= atomic_read(&xprt
->xpt_reserved
) + serv
->sv_max_mesg
;
1186 if (sk_stream_wspace(svsk
->sk_sk
) >= required
)
1188 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
1192 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1194 struct sockaddr
*sa
, int salen
,
1197 return svc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1200 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1201 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
1202 struct net
*, struct sockaddr
*,
1204 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
1206 static struct svc_xprt
*svc_bc_tcp_create(struct svc_serv
*serv
,
1208 struct sockaddr
*sa
, int salen
,
1211 return svc_bc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1214 static void svc_bc_tcp_sock_detach(struct svc_xprt
*xprt
)
1218 static struct svc_xprt_ops svc_tcp_bc_ops
= {
1219 .xpo_create
= svc_bc_tcp_create
,
1220 .xpo_detach
= svc_bc_tcp_sock_detach
,
1221 .xpo_free
= svc_bc_sock_free
,
1222 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1225 static struct svc_xprt_class svc_tcp_bc_class
= {
1226 .xcl_name
= "tcp-bc",
1227 .xcl_owner
= THIS_MODULE
,
1228 .xcl_ops
= &svc_tcp_bc_ops
,
1229 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1232 static void svc_init_bc_xprt_sock(void)
1234 svc_reg_xprt_class(&svc_tcp_bc_class
);
1237 static void svc_cleanup_bc_xprt_sock(void)
1239 svc_unreg_xprt_class(&svc_tcp_bc_class
);
1241 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1242 static void svc_init_bc_xprt_sock(void)
1246 static void svc_cleanup_bc_xprt_sock(void)
1249 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1251 static struct svc_xprt_ops svc_tcp_ops
= {
1252 .xpo_create
= svc_tcp_create
,
1253 .xpo_recvfrom
= svc_tcp_recvfrom
,
1254 .xpo_sendto
= svc_tcp_sendto
,
1255 .xpo_release_rqst
= svc_release_skb
,
1256 .xpo_detach
= svc_tcp_sock_detach
,
1257 .xpo_free
= svc_sock_free
,
1258 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1259 .xpo_has_wspace
= svc_tcp_has_wspace
,
1260 .xpo_accept
= svc_tcp_accept
,
1263 static struct svc_xprt_class svc_tcp_class
= {
1265 .xcl_owner
= THIS_MODULE
,
1266 .xcl_ops
= &svc_tcp_ops
,
1267 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1270 void svc_init_xprt_sock(void)
1272 svc_reg_xprt_class(&svc_tcp_class
);
1273 svc_reg_xprt_class(&svc_udp_class
);
1274 svc_init_bc_xprt_sock();
1277 void svc_cleanup_xprt_sock(void)
1279 svc_unreg_xprt_class(&svc_tcp_class
);
1280 svc_unreg_xprt_class(&svc_udp_class
);
1281 svc_cleanup_bc_xprt_sock();
1284 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1286 struct sock
*sk
= svsk
->sk_sk
;
1288 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_tcp_class
,
1289 &svsk
->sk_xprt
, serv
);
1290 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1291 if (sk
->sk_state
== TCP_LISTEN
) {
1292 dprintk("setting up TCP socket for listening\n");
1293 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1294 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1295 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1297 dprintk("setting up TCP socket for reading\n");
1298 sk
->sk_state_change
= svc_tcp_state_change
;
1299 sk
->sk_data_ready
= svc_tcp_data_ready
;
1300 sk
->sk_write_space
= svc_tcp_write_space
;
1302 svsk
->sk_reclen
= 0;
1303 svsk
->sk_tcplen
= 0;
1304 memset(&svsk
->sk_pages
[0], 0, sizeof(svsk
->sk_pages
));
1306 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1308 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1309 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1310 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1314 void svc_sock_update_bufs(struct svc_serv
*serv
)
1317 * The number of server threads has changed. Update
1318 * rcvbuf and sndbuf accordingly on all sockets
1320 struct svc_sock
*svsk
;
1322 spin_lock_bh(&serv
->sv_lock
);
1323 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
)
1324 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1325 spin_unlock_bh(&serv
->sv_lock
);
1327 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1330 * Initialize socket for RPC use and create svc_sock struct
1331 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1333 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1334 struct socket
*sock
,
1337 struct svc_sock
*svsk
;
1339 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1342 dprintk("svc: svc_setup_socket %p\n", sock
);
1343 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1345 return ERR_PTR(-ENOMEM
);
1349 /* Register socket with portmapper */
1351 err
= svc_register(serv
, sock_net(sock
->sk
), inet
->sk_family
,
1353 ntohs(inet_sk(inet
)->inet_sport
));
1357 return ERR_PTR(err
);
1360 inet
->sk_user_data
= svsk
;
1361 svsk
->sk_sock
= sock
;
1363 svsk
->sk_ostate
= inet
->sk_state_change
;
1364 svsk
->sk_odata
= inet
->sk_data_ready
;
1365 svsk
->sk_owspace
= inet
->sk_write_space
;
1367 /* Initialize the socket */
1368 if (sock
->type
== SOCK_DGRAM
)
1369 svc_udp_init(svsk
, serv
);
1371 /* initialise setting must have enough space to
1372 * receive and respond to one request.
1374 svc_sock_setbufsize(svsk
->sk_sock
, 4 * serv
->sv_max_mesg
,
1375 4 * serv
->sv_max_mesg
);
1376 svc_tcp_init(svsk
, serv
);
1379 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1386 * svc_addsock - add a listener socket to an RPC service
1387 * @serv: pointer to RPC service to which to add a new listener
1388 * @fd: file descriptor of the new listener
1389 * @name_return: pointer to buffer to fill in with name of listener
1390 * @len: size of the buffer
1392 * Fills in socket name and returns positive length of name if successful.
1393 * Name is terminated with '\n'. On error, returns a negative errno
1396 int svc_addsock(struct svc_serv
*serv
, const int fd
, char *name_return
,
1400 struct socket
*so
= sockfd_lookup(fd
, &err
);
1401 struct svc_sock
*svsk
= NULL
;
1402 struct sockaddr_storage addr
;
1403 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1408 err
= -EAFNOSUPPORT
;
1409 if ((so
->sk
->sk_family
!= PF_INET
) && (so
->sk
->sk_family
!= PF_INET6
))
1411 err
= -EPROTONOSUPPORT
;
1412 if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1413 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1416 if (so
->state
> SS_UNCONNECTED
)
1419 if (!try_module_get(THIS_MODULE
))
1421 svsk
= svc_setup_socket(serv
, so
, SVC_SOCK_DEFAULTS
);
1423 module_put(THIS_MODULE
);
1424 err
= PTR_ERR(svsk
);
1427 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1428 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1429 svc_add_new_perm_xprt(serv
, &svsk
->sk_xprt
);
1430 return svc_one_sock_name(svsk
, name_return
, len
);
1435 EXPORT_SYMBOL_GPL(svc_addsock
);
1438 * Create socket for RPC service.
1440 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1443 struct sockaddr
*sin
, int len
,
1446 struct svc_sock
*svsk
;
1447 struct socket
*sock
;
1450 struct sockaddr_storage addr
;
1451 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1455 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1457 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1458 serv
->sv_program
->pg_name
, protocol
,
1459 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1461 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1462 printk(KERN_WARNING
"svc: only UDP and TCP "
1463 "sockets supported\n");
1464 return ERR_PTR(-EINVAL
);
1467 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1468 switch (sin
->sa_family
) {
1476 return ERR_PTR(-EINVAL
);
1479 error
= __sock_create(net
, family
, type
, protocol
, &sock
, 1);
1481 return ERR_PTR(error
);
1483 svc_reclassify_socket(sock
);
1486 * If this is an PF_INET6 listener, we want to avoid
1487 * getting requests from IPv4 remotes. Those should
1488 * be shunted to a PF_INET listener via rpcbind.
1491 if (family
== PF_INET6
)
1492 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1493 (char *)&val
, sizeof(val
));
1495 if (type
== SOCK_STREAM
)
1496 sock
->sk
->sk_reuse
= SK_CAN_REUSE
; /* allow address reuse */
1497 error
= kernel_bind(sock
, sin
, len
);
1502 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1506 if (protocol
== IPPROTO_TCP
) {
1507 if ((error
= kernel_listen(sock
, 64)) < 0)
1511 svsk
= svc_setup_socket(serv
, sock
, flags
);
1513 error
= PTR_ERR(svsk
);
1516 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1517 return (struct svc_xprt
*)svsk
;
1519 dprintk("svc: svc_create_socket error = %d\n", -error
);
1521 return ERR_PTR(error
);
1525 * Detach the svc_sock from the socket so that no
1526 * more callbacks occur.
1528 static void svc_sock_detach(struct svc_xprt
*xprt
)
1530 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1531 struct sock
*sk
= svsk
->sk_sk
;
1532 wait_queue_head_t
*wq
;
1534 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1536 /* put back the old socket callbacks */
1537 sk
->sk_state_change
= svsk
->sk_ostate
;
1538 sk
->sk_data_ready
= svsk
->sk_odata
;
1539 sk
->sk_write_space
= svsk
->sk_owspace
;
1542 if (wq
&& waitqueue_active(wq
))
1543 wake_up_interruptible(wq
);
1547 * Disconnect the socket, and reset the callbacks
1549 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1551 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1553 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1555 svc_sock_detach(xprt
);
1557 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
)) {
1558 svc_tcp_clear_pages(svsk
);
1559 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1564 * Free the svc_sock's socket resources and the svc_sock itself.
1566 static void svc_sock_free(struct svc_xprt
*xprt
)
1568 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1569 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1571 if (svsk
->sk_sock
->file
)
1572 sockfd_put(svsk
->sk_sock
);
1574 sock_release(svsk
->sk_sock
);
1578 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1580 * Create a back channel svc_xprt which shares the fore channel socket.
1582 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*serv
,
1585 struct sockaddr
*sin
, int len
,
1588 struct svc_sock
*svsk
;
1589 struct svc_xprt
*xprt
;
1591 if (protocol
!= IPPROTO_TCP
) {
1592 printk(KERN_WARNING
"svc: only TCP sockets"
1593 " supported on shared back channel\n");
1594 return ERR_PTR(-EINVAL
);
1597 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1599 return ERR_PTR(-ENOMEM
);
1601 xprt
= &svsk
->sk_xprt
;
1602 svc_xprt_init(net
, &svc_tcp_bc_class
, xprt
, serv
);
1604 serv
->sv_bc_xprt
= xprt
;
1610 * Free a back channel svc_sock.
1612 static void svc_bc_sock_free(struct svc_xprt
*xprt
)
1615 kfree(container_of(xprt
, struct svc_sock
, sk_xprt
));
1617 #endif /* CONFIG_SUNRPC_BACKCHANNEL */