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 int svc_udp_recvfrom(struct svc_rqst
*);
64 static int svc_udp_sendto(struct svc_rqst
*);
65 static void svc_sock_detach(struct svc_xprt
*);
66 static void svc_tcp_sock_detach(struct svc_xprt
*);
67 static void svc_sock_free(struct svc_xprt
*);
69 static struct svc_xprt
*svc_create_socket(struct svc_serv
*, int,
70 struct net
*, struct sockaddr
*,
72 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
73 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
74 struct net
*, struct sockaddr
*,
76 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
77 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
79 #ifdef CONFIG_DEBUG_LOCK_ALLOC
80 static struct lock_class_key svc_key
[2];
81 static struct lock_class_key svc_slock_key
[2];
83 static void svc_reclassify_socket(struct socket
*sock
)
85 struct sock
*sk
= sock
->sk
;
87 if (WARN_ON_ONCE(!sock_allow_reclassification(sk
)))
90 switch (sk
->sk_family
) {
92 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
94 "sk_xprt.xpt_lock-AF_INET-NFSD",
99 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
101 "sk_xprt.xpt_lock-AF_INET6-NFSD",
110 static void svc_reclassify_socket(struct socket
*sock
)
116 * Release an skbuff after use
118 static void svc_release_skb(struct svc_rqst
*rqstp
)
120 struct sk_buff
*skb
= rqstp
->rq_xprt_ctxt
;
123 struct svc_sock
*svsk
=
124 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
125 rqstp
->rq_xprt_ctxt
= NULL
;
127 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
128 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
132 union svc_pktinfo_u
{
133 struct in_pktinfo pkti
;
134 struct in6_pktinfo pkti6
;
136 #define SVC_PKTINFO_SPACE \
137 CMSG_SPACE(sizeof(union svc_pktinfo_u))
139 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
141 struct svc_sock
*svsk
=
142 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
143 switch (svsk
->sk_sk
->sk_family
) {
145 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
147 cmh
->cmsg_level
= SOL_IP
;
148 cmh
->cmsg_type
= IP_PKTINFO
;
149 pki
->ipi_ifindex
= 0;
150 pki
->ipi_spec_dst
.s_addr
=
151 svc_daddr_in(rqstp
)->sin_addr
.s_addr
;
152 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
157 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
158 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
160 cmh
->cmsg_level
= SOL_IPV6
;
161 cmh
->cmsg_type
= IPV6_PKTINFO
;
162 pki
->ipi6_ifindex
= daddr
->sin6_scope_id
;
163 pki
->ipi6_addr
= daddr
->sin6_addr
;
164 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
171 * send routine intended to be shared by the fore- and back-channel
173 int svc_send_common(struct socket
*sock
, struct xdr_buf
*xdr
,
174 struct page
*headpage
, unsigned long headoffset
,
175 struct page
*tailpage
, unsigned long tailoffset
)
179 struct page
**ppage
= xdr
->pages
;
180 size_t base
= xdr
->page_base
;
181 unsigned int pglen
= xdr
->page_len
;
182 unsigned int flags
= MSG_MORE
| MSG_SENDPAGE_NOTLAST
;
189 if (slen
== xdr
->head
[0].iov_len
)
191 len
= kernel_sendpage(sock
, headpage
, headoffset
,
192 xdr
->head
[0].iov_len
, flags
);
193 if (len
!= xdr
->head
[0].iov_len
)
195 slen
-= xdr
->head
[0].iov_len
;
200 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
204 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
211 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
217 if (xdr
->tail
[0].iov_len
) {
218 result
= kernel_sendpage(sock
, tailpage
, tailoffset
,
219 xdr
->tail
[0].iov_len
, 0);
230 * Generic sendto routine
232 static int svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
234 struct svc_sock
*svsk
=
235 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
236 struct socket
*sock
= svsk
->sk_sock
;
239 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
241 struct cmsghdr
*cmh
= &buffer
.hdr
;
243 unsigned long tailoff
;
244 unsigned long headoff
;
245 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
247 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
248 struct msghdr msg
= {
249 .msg_name
= &rqstp
->rq_addr
,
250 .msg_namelen
= rqstp
->rq_addrlen
,
252 .msg_controllen
= sizeof(buffer
),
253 .msg_flags
= MSG_MORE
,
256 svc_set_cmsg_data(rqstp
, cmh
);
258 if (sock_sendmsg(sock
, &msg
) < 0)
262 tailoff
= ((unsigned long)xdr
->tail
[0].iov_base
) & (PAGE_SIZE
-1);
264 len
= svc_send_common(sock
, xdr
, rqstp
->rq_respages
[0], headoff
,
265 rqstp
->rq_respages
[0], tailoff
);
268 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
269 svsk
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
270 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
276 * Report socket names for nfsdfs
278 static int svc_one_sock_name(struct svc_sock
*svsk
, char *buf
, int remaining
)
280 const struct sock
*sk
= svsk
->sk_sk
;
281 const char *proto_name
= sk
->sk_protocol
== IPPROTO_UDP
?
285 switch (sk
->sk_family
) {
287 len
= snprintf(buf
, remaining
, "ipv4 %s %pI4 %d\n",
289 &inet_sk(sk
)->inet_rcv_saddr
,
290 inet_sk(sk
)->inet_num
);
292 #if IS_ENABLED(CONFIG_IPV6)
294 len
= snprintf(buf
, remaining
, "ipv6 %s %pI6 %d\n",
296 &sk
->sk_v6_rcv_saddr
,
297 inet_sk(sk
)->inet_num
);
301 len
= snprintf(buf
, remaining
, "*unknown-%d*\n",
305 if (len
>= remaining
) {
307 return -ENAMETOOLONG
;
313 * Generic recvfrom routine.
315 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
318 struct svc_sock
*svsk
=
319 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
320 struct msghdr msg
= {
321 .msg_flags
= MSG_DONTWAIT
,
325 rqstp
->rq_xprt_hlen
= 0;
327 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
328 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
330 /* If we read a full record, then assume there may be more
331 * data to read (stream based sockets only!)
334 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
336 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
337 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
341 static int svc_partial_recvfrom(struct svc_rqst
*rqstp
,
342 struct kvec
*iov
, int nr
,
343 int buflen
, unsigned int base
)
351 return svc_recvfrom(rqstp
, iov
, nr
, buflen
);
353 for (i
= 0; i
< nr
; i
++) {
354 if (iov
[i
].iov_len
> base
)
356 base
-= iov
[i
].iov_len
;
358 save_iovlen
= iov
[i
].iov_len
;
359 save_iovbase
= iov
[i
].iov_base
;
360 iov
[i
].iov_len
-= base
;
361 iov
[i
].iov_base
+= base
;
362 ret
= svc_recvfrom(rqstp
, &iov
[i
], nr
- i
, buflen
);
363 iov
[i
].iov_len
= save_iovlen
;
364 iov
[i
].iov_base
= save_iovbase
;
369 * Set socket snd and rcv buffer lengths
371 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
376 oldfs
= get_fs(); set_fs(KERNEL_DS
);
377 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
378 (char*)&snd
, sizeof(snd
));
379 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
380 (char*)&rcv
, sizeof(rcv
));
382 /* sock_setsockopt limits use to sysctl_?mem_max,
383 * which isn't acceptable. Until that is made conditional
384 * on not having CAP_SYS_RESOURCE or similar, we go direct...
385 * DaveM said I could!
388 sock
->sk
->sk_sndbuf
= snd
* 2;
389 sock
->sk
->sk_rcvbuf
= rcv
* 2;
390 sock
->sk
->sk_write_space(sock
->sk
);
391 release_sock(sock
->sk
);
395 static int svc_sock_secure_port(struct svc_rqst
*rqstp
)
397 return svc_port_is_privileged(svc_addr(rqstp
));
401 * INET callback when data has been received on the socket.
403 static void svc_data_ready(struct sock
*sk
)
405 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
408 dprintk("svc: socket %p(inet %p), busy=%d\n",
410 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
412 if (!test_and_set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
))
413 svc_xprt_enqueue(&svsk
->sk_xprt
);
418 * INET callback when space is newly available on the socket.
420 static void svc_write_space(struct sock
*sk
)
422 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
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 svsk
->sk_owspace(sk
);
428 svc_xprt_enqueue(&svsk
->sk_xprt
);
432 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
434 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
436 if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
438 return !test_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
442 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
444 static int svc_udp_get_dest_address4(struct svc_rqst
*rqstp
,
447 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
448 struct sockaddr_in
*daddr
= svc_daddr_in(rqstp
);
450 if (cmh
->cmsg_type
!= IP_PKTINFO
)
453 daddr
->sin_family
= AF_INET
;
454 daddr
->sin_addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
459 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
461 static int svc_udp_get_dest_address6(struct svc_rqst
*rqstp
,
464 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
465 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
467 if (cmh
->cmsg_type
!= IPV6_PKTINFO
)
470 daddr
->sin6_family
= AF_INET6
;
471 daddr
->sin6_addr
= pki
->ipi6_addr
;
472 daddr
->sin6_scope_id
= pki
->ipi6_ifindex
;
477 * Copy the UDP datagram's destination address to the rqstp structure.
478 * The 'destination' address in this case is the address to which the
479 * peer sent the datagram, i.e. our local address. For multihomed
480 * hosts, this can change from msg to msg. Note that only the IP
481 * address changes, the port number should remain the same.
483 static int svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
486 switch (cmh
->cmsg_level
) {
488 return svc_udp_get_dest_address4(rqstp
, cmh
);
490 return svc_udp_get_dest_address6(rqstp
, cmh
);
497 * Receive a datagram from a UDP socket.
499 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
501 struct svc_sock
*svsk
=
502 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
503 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
507 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
509 struct cmsghdr
*cmh
= &buffer
.hdr
;
510 struct msghdr msg
= {
511 .msg_name
= svc_addr(rqstp
),
513 .msg_controllen
= sizeof(buffer
),
514 .msg_flags
= MSG_DONTWAIT
,
519 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
520 /* udp sockets need large rcvbuf as all pending
521 * requests are still in that buffer. sndbuf must
522 * also be large enough that there is enough space
523 * for one reply per thread. We count all threads
524 * rather than threads in a particular pool, which
525 * provides an upper bound on the number of threads
526 * which will access the socket.
528 svc_sock_setbufsize(svsk
->sk_sock
,
529 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
530 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
532 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
534 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
535 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
537 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
540 if (err
!= -EAGAIN
) {
541 /* possibly an icmp error */
542 dprintk("svc: recvfrom returned error %d\n", -err
);
543 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
547 len
= svc_addr_len(svc_addr(rqstp
));
548 rqstp
->rq_addrlen
= len
;
549 if (skb
->tstamp
.tv64
== 0) {
550 skb
->tstamp
= ktime_get_real();
551 /* Don't enable netstamp, sunrpc doesn't
552 need that much accuracy */
554 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
555 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
558 rqstp
->rq_arg
.len
= len
;
560 rqstp
->rq_prot
= IPPROTO_UDP
;
562 if (!svc_udp_get_dest_address(rqstp
, cmh
)) {
563 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
564 cmh
->cmsg_level
, cmh
->cmsg_type
);
567 rqstp
->rq_daddrlen
= svc_addr_len(svc_daddr(rqstp
));
569 if (skb_is_nonlinear(skb
)) {
570 /* we have to copy */
572 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
578 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
580 /* we can use it in-place */
581 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
;
582 rqstp
->rq_arg
.head
[0].iov_len
= len
;
583 if (skb_checksum_complete(skb
))
585 rqstp
->rq_xprt_ctxt
= skb
;
588 rqstp
->rq_arg
.page_base
= 0;
589 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
590 rqstp
->rq_arg
.head
[0].iov_len
= len
;
591 rqstp
->rq_arg
.page_len
= 0;
592 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
594 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
595 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
596 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
598 rqstp
->rq_next_page
= rqstp
->rq_respages
+1;
601 serv
->sv_stats
->netudpcnt
++;
605 trace_kfree_skb(skb
, svc_udp_recvfrom
);
606 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
611 svc_udp_sendto(struct svc_rqst
*rqstp
)
615 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
616 if (error
== -ECONNREFUSED
)
617 /* ICMP error on earlier request. */
618 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
623 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
627 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
629 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
630 struct svc_serv
*serv
= xprt
->xpt_server
;
631 unsigned long required
;
634 * Set the SOCK_NOSPACE flag before checking the available
637 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
638 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
639 if (required
*2 > sock_wspace(svsk
->sk_sk
))
641 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
645 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
651 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
653 struct sockaddr
*sa
, int salen
,
656 return svc_create_socket(serv
, IPPROTO_UDP
, net
, sa
, salen
, flags
);
659 static struct svc_xprt_ops svc_udp_ops
= {
660 .xpo_create
= svc_udp_create
,
661 .xpo_recvfrom
= svc_udp_recvfrom
,
662 .xpo_sendto
= svc_udp_sendto
,
663 .xpo_release_rqst
= svc_release_skb
,
664 .xpo_detach
= svc_sock_detach
,
665 .xpo_free
= svc_sock_free
,
666 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
667 .xpo_has_wspace
= svc_udp_has_wspace
,
668 .xpo_accept
= svc_udp_accept
,
669 .xpo_secure_port
= svc_sock_secure_port
,
672 static struct svc_xprt_class svc_udp_class
= {
674 .xcl_owner
= THIS_MODULE
,
675 .xcl_ops
= &svc_udp_ops
,
676 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
677 .xcl_ident
= XPRT_TRANSPORT_UDP
,
680 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
682 int err
, level
, optname
, one
= 1;
684 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_udp_class
,
685 &svsk
->sk_xprt
, serv
);
686 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
687 svsk
->sk_sk
->sk_data_ready
= svc_data_ready
;
688 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
690 /* initialise setting must have enough space to
691 * receive and respond to one request.
692 * svc_udp_recvfrom will re-adjust if necessary
694 svc_sock_setbufsize(svsk
->sk_sock
,
695 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
696 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
698 /* data might have come in before data_ready set up */
699 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
700 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
702 /* make sure we get destination address info */
703 switch (svsk
->sk_sk
->sk_family
) {
706 optname
= IP_PKTINFO
;
710 optname
= IPV6_RECVPKTINFO
;
715 err
= kernel_setsockopt(svsk
->sk_sock
, level
, optname
,
716 (char *)&one
, sizeof(one
));
717 dprintk("svc: kernel_setsockopt returned %d\n", err
);
721 * A data_ready event on a listening socket means there's a connection
722 * pending. Do not use state_change as a substitute for it.
724 static void svc_tcp_listen_data_ready(struct sock
*sk
)
726 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
728 dprintk("svc: socket %p TCP (listen) state change %d\n",
734 * This callback may called twice when a new connection
735 * is established as a child socket inherits everything
736 * from a parent LISTEN socket.
737 * 1) data_ready method of the parent socket will be called
738 * when one of child sockets become ESTABLISHED.
739 * 2) data_ready method of the child socket may be called
740 * when it receives data before the socket is accepted.
741 * In case of 2, we should ignore it silently.
743 if (sk
->sk_state
== TCP_LISTEN
) {
745 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
746 svc_xprt_enqueue(&svsk
->sk_xprt
);
748 printk("svc: socket %p: no user data\n", sk
);
753 * A state change on a connected socket means it's dying or dead.
755 static void svc_tcp_state_change(struct sock
*sk
)
757 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
759 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
760 sk
, sk
->sk_state
, sk
->sk_user_data
);
763 printk("svc: socket %p: no user data\n", sk
);
766 if (sk
->sk_state
!= TCP_ESTABLISHED
) {
767 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
768 svc_xprt_enqueue(&svsk
->sk_xprt
);
774 * Accept a TCP connection
776 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
778 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
779 struct sockaddr_storage addr
;
780 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
781 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
782 struct socket
*sock
= svsk
->sk_sock
;
783 struct socket
*newsock
;
784 struct svc_sock
*newsvsk
;
786 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
788 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
792 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
793 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
796 printk(KERN_WARNING
"%s: no more sockets!\n",
798 else if (err
!= -EAGAIN
)
799 net_warn_ratelimited("%s: accept failed (err %d)!\n",
800 serv
->sv_name
, -err
);
803 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
805 err
= kernel_getpeername(newsock
, sin
, &slen
);
807 net_warn_ratelimited("%s: peername failed (err %d)!\n",
808 serv
->sv_name
, -err
);
809 goto failed
; /* aborted connection or whatever */
812 /* Ideally, we would want to reject connections from unauthorized
813 * hosts here, but when we get encryption, the IP of the host won't
814 * tell us anything. For now just warn about unpriv connections.
816 if (!svc_port_is_privileged(sin
)) {
817 dprintk("%s: connect from unprivileged port: %s\n",
819 __svc_print_addr(sin
, buf
, sizeof(buf
)));
821 dprintk("%s: connect from %s\n", serv
->sv_name
,
822 __svc_print_addr(sin
, buf
, sizeof(buf
)));
824 /* Reset the inherited callbacks before calling svc_setup_socket */
825 newsock
->sk
->sk_state_change
= svsk
->sk_ostate
;
826 newsock
->sk
->sk_data_ready
= svsk
->sk_odata
;
827 newsock
->sk
->sk_write_space
= svsk
->sk_owspace
;
829 /* make sure that a write doesn't block forever when
832 newsock
->sk
->sk_sndtimeo
= HZ
*30;
834 newsvsk
= svc_setup_socket(serv
, newsock
,
835 (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
);
846 if (sock_is_loopback(newsock
->sk
))
847 set_bit(XPT_LOCAL
, &newsvsk
->sk_xprt
.xpt_flags
);
849 clear_bit(XPT_LOCAL
, &newsvsk
->sk_xprt
.xpt_flags
);
851 serv
->sv_stats
->nettcpconn
++;
853 return &newsvsk
->sk_xprt
;
856 sock_release(newsock
);
860 static unsigned int svc_tcp_restore_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
862 unsigned int i
, len
, npages
;
864 if (svsk
->sk_datalen
== 0)
866 len
= svsk
->sk_datalen
;
867 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
868 for (i
= 0; i
< npages
; i
++) {
869 if (rqstp
->rq_pages
[i
] != NULL
)
870 put_page(rqstp
->rq_pages
[i
]);
871 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
872 rqstp
->rq_pages
[i
] = svsk
->sk_pages
[i
];
873 svsk
->sk_pages
[i
] = NULL
;
875 rqstp
->rq_arg
.head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
879 static void svc_tcp_save_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
881 unsigned int i
, len
, npages
;
883 if (svsk
->sk_datalen
== 0)
885 len
= svsk
->sk_datalen
;
886 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
887 for (i
= 0; i
< npages
; i
++) {
888 svsk
->sk_pages
[i
] = rqstp
->rq_pages
[i
];
889 rqstp
->rq_pages
[i
] = NULL
;
893 static void svc_tcp_clear_pages(struct svc_sock
*svsk
)
895 unsigned int i
, len
, npages
;
897 if (svsk
->sk_datalen
== 0)
899 len
= svsk
->sk_datalen
;
900 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
901 for (i
= 0; i
< npages
; i
++) {
902 if (svsk
->sk_pages
[i
] == NULL
) {
906 put_page(svsk
->sk_pages
[i
]);
907 svsk
->sk_pages
[i
] = NULL
;
911 svsk
->sk_datalen
= 0;
915 * Receive fragment record header.
916 * If we haven't gotten the record length yet, get the next four bytes.
918 static int svc_tcp_recv_record(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
920 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
924 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
927 want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
928 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
930 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
932 svsk
->sk_tcplen
+= len
;
935 dprintk("svc: short recvfrom while reading record "
936 "length (%d of %d)\n", len
, want
);
940 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk
));
941 if (svc_sock_reclen(svsk
) + svsk
->sk_datalen
>
943 net_notice_ratelimited("RPC: fragment too large: %d\n",
944 svc_sock_reclen(svsk
));
949 return svc_sock_reclen(svsk
);
951 dprintk("RPC: TCP recv_record got %d\n", len
);
954 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
958 static int receive_cb_reply(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
960 struct rpc_xprt
*bc_xprt
= svsk
->sk_xprt
.xpt_bc_xprt
;
961 struct rpc_rqst
*req
= NULL
;
962 struct kvec
*src
, *dst
;
963 __be32
*p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
972 spin_lock_bh(&bc_xprt
->transport_lock
);
973 req
= xprt_lookup_rqst(bc_xprt
, xid
);
975 goto unlock_notfound
;
977 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
, sizeof(struct xdr_buf
));
979 * XXX!: cheating for now! Only copying HEAD.
980 * But we know this is good enough for now (in fact, for any
981 * callback reply in the forseeable future).
983 dst
= &req
->rq_private_buf
.head
[0];
984 src
= &rqstp
->rq_arg
.head
[0];
985 if (dst
->iov_len
< src
->iov_len
)
986 goto unlock_eagain
; /* whatever; just giving up. */
987 memcpy(dst
->iov_base
, src
->iov_base
, src
->iov_len
);
988 xprt_complete_rqst(req
->rq_task
, rqstp
->rq_arg
.len
);
989 rqstp
->rq_arg
.len
= 0;
990 spin_unlock_bh(&bc_xprt
->transport_lock
);
994 "%s: Got unrecognized reply: "
995 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
996 __func__
, ntohl(calldir
),
997 bc_xprt
, ntohl(xid
));
999 spin_unlock_bh(&bc_xprt
->transport_lock
);
1003 static int copy_pages_to_kvecs(struct kvec
*vec
, struct page
**pages
, int len
)
1009 vec
[i
].iov_base
= page_address(pages
[i
]);
1010 vec
[i
].iov_len
= PAGE_SIZE
;
1017 static void svc_tcp_fragment_received(struct svc_sock
*svsk
)
1019 /* If we have more data, signal svc_xprt_enqueue() to try again */
1020 dprintk("svc: TCP %s record (%d bytes)\n",
1021 svc_sock_final_rec(svsk
) ? "final" : "nonfinal",
1022 svc_sock_reclen(svsk
));
1023 svsk
->sk_tcplen
= 0;
1024 svsk
->sk_reclen
= 0;
1028 * Receive data from a TCP socket.
1030 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1032 struct svc_sock
*svsk
=
1033 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
1034 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1037 unsigned int want
, base
;
1042 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1043 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
1044 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
1045 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1047 len
= svc_tcp_recv_record(svsk
, rqstp
);
1051 base
= svc_tcp_restore_pages(svsk
, rqstp
);
1052 want
= svc_sock_reclen(svsk
) - (svsk
->sk_tcplen
- sizeof(rpc_fraghdr
));
1054 vec
= rqstp
->rq_vec
;
1056 pnum
= copy_pages_to_kvecs(&vec
[0], &rqstp
->rq_pages
[0],
1057 svsk
->sk_datalen
+ want
);
1059 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1060 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
1062 /* Now receive data */
1063 len
= svc_partial_recvfrom(rqstp
, vec
, pnum
, want
, base
);
1065 svsk
->sk_tcplen
+= len
;
1066 svsk
->sk_datalen
+= len
;
1068 if (len
!= want
|| !svc_sock_final_rec(svsk
)) {
1069 svc_tcp_save_pages(svsk
, rqstp
);
1070 if (len
< 0 && len
!= -EAGAIN
)
1073 svc_tcp_fragment_received(svsk
);
1075 dprintk("svc: incomplete TCP record (%d of %d)\n",
1076 (int)(svsk
->sk_tcplen
- sizeof(rpc_fraghdr
)),
1077 svc_sock_reclen(svsk
));
1081 if (svsk
->sk_datalen
< 8) {
1082 svsk
->sk_datalen
= 0;
1083 goto err_delete
; /* client is nuts. */
1086 rqstp
->rq_arg
.len
= svsk
->sk_datalen
;
1087 rqstp
->rq_arg
.page_base
= 0;
1088 if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1089 rqstp
->rq_arg
.head
[0].iov_len
= rqstp
->rq_arg
.len
;
1090 rqstp
->rq_arg
.page_len
= 0;
1092 rqstp
->rq_arg
.page_len
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1094 rqstp
->rq_xprt_ctxt
= NULL
;
1095 rqstp
->rq_prot
= IPPROTO_TCP
;
1096 if (test_bit(XPT_LOCAL
, &svsk
->sk_xprt
.xpt_flags
))
1097 set_bit(RQ_LOCAL
, &rqstp
->rq_flags
);
1099 clear_bit(RQ_LOCAL
, &rqstp
->rq_flags
);
1101 p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1104 len
= receive_cb_reply(svsk
, rqstp
);
1106 /* Reset TCP read info */
1107 svsk
->sk_datalen
= 0;
1108 svc_tcp_fragment_received(svsk
);
1113 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1115 serv
->sv_stats
->nettcpcnt
++;
1117 return rqstp
->rq_arg
.len
;
1122 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1125 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1126 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1127 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1129 return 0; /* record not complete */
1133 * Send out data on TCP socket.
1135 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
1137 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1141 /* Set up the first element of the reply kvec.
1142 * Any other kvecs that may be in use have been taken
1143 * care of by the server implementation itself.
1145 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1146 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1148 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1149 if (sent
!= xbufp
->len
) {
1151 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1152 "- shutting down socket\n",
1153 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1154 (sent
<0)?"got error":"sent only",
1156 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1157 svc_xprt_enqueue(rqstp
->rq_xprt
);
1164 * Setup response header. TCP has a 4B record length field.
1166 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1168 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1170 /* tcp needs a space for the record length... */
1174 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1176 struct sockaddr
*sa
, int salen
,
1179 return svc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1182 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1183 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
1184 struct net
*, struct sockaddr
*,
1186 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
1188 static struct svc_xprt
*svc_bc_tcp_create(struct svc_serv
*serv
,
1190 struct sockaddr
*sa
, int salen
,
1193 return svc_bc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1196 static void svc_bc_tcp_sock_detach(struct svc_xprt
*xprt
)
1200 static struct svc_xprt_ops svc_tcp_bc_ops
= {
1201 .xpo_create
= svc_bc_tcp_create
,
1202 .xpo_detach
= svc_bc_tcp_sock_detach
,
1203 .xpo_free
= svc_bc_sock_free
,
1204 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1205 .xpo_secure_port
= svc_sock_secure_port
,
1208 static struct svc_xprt_class svc_tcp_bc_class
= {
1209 .xcl_name
= "tcp-bc",
1210 .xcl_owner
= THIS_MODULE
,
1211 .xcl_ops
= &svc_tcp_bc_ops
,
1212 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1215 static void svc_init_bc_xprt_sock(void)
1217 svc_reg_xprt_class(&svc_tcp_bc_class
);
1220 static void svc_cleanup_bc_xprt_sock(void)
1222 svc_unreg_xprt_class(&svc_tcp_bc_class
);
1224 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1225 static void svc_init_bc_xprt_sock(void)
1229 static void svc_cleanup_bc_xprt_sock(void)
1232 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1234 static struct svc_xprt_ops svc_tcp_ops
= {
1235 .xpo_create
= svc_tcp_create
,
1236 .xpo_recvfrom
= svc_tcp_recvfrom
,
1237 .xpo_sendto
= svc_tcp_sendto
,
1238 .xpo_release_rqst
= svc_release_skb
,
1239 .xpo_detach
= svc_tcp_sock_detach
,
1240 .xpo_free
= svc_sock_free
,
1241 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1242 .xpo_has_wspace
= svc_tcp_has_wspace
,
1243 .xpo_accept
= svc_tcp_accept
,
1244 .xpo_secure_port
= svc_sock_secure_port
,
1247 static struct svc_xprt_class svc_tcp_class
= {
1249 .xcl_owner
= THIS_MODULE
,
1250 .xcl_ops
= &svc_tcp_ops
,
1251 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1252 .xcl_ident
= XPRT_TRANSPORT_TCP
,
1255 void svc_init_xprt_sock(void)
1257 svc_reg_xprt_class(&svc_tcp_class
);
1258 svc_reg_xprt_class(&svc_udp_class
);
1259 svc_init_bc_xprt_sock();
1262 void svc_cleanup_xprt_sock(void)
1264 svc_unreg_xprt_class(&svc_tcp_class
);
1265 svc_unreg_xprt_class(&svc_udp_class
);
1266 svc_cleanup_bc_xprt_sock();
1269 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1271 struct sock
*sk
= svsk
->sk_sk
;
1273 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_tcp_class
,
1274 &svsk
->sk_xprt
, serv
);
1275 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1276 if (sk
->sk_state
== TCP_LISTEN
) {
1277 dprintk("setting up TCP socket for listening\n");
1278 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1279 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1280 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1282 dprintk("setting up TCP socket for reading\n");
1283 sk
->sk_state_change
= svc_tcp_state_change
;
1284 sk
->sk_data_ready
= svc_data_ready
;
1285 sk
->sk_write_space
= svc_write_space
;
1287 svsk
->sk_reclen
= 0;
1288 svsk
->sk_tcplen
= 0;
1289 svsk
->sk_datalen
= 0;
1290 memset(&svsk
->sk_pages
[0], 0, sizeof(svsk
->sk_pages
));
1292 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1294 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1295 switch (sk
->sk_state
) {
1297 case TCP_ESTABLISHED
:
1300 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1305 void svc_sock_update_bufs(struct svc_serv
*serv
)
1308 * The number of server threads has changed. Update
1309 * rcvbuf and sndbuf accordingly on all sockets
1311 struct svc_sock
*svsk
;
1313 spin_lock_bh(&serv
->sv_lock
);
1314 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
)
1315 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1316 spin_unlock_bh(&serv
->sv_lock
);
1318 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1321 * Initialize socket for RPC use and create svc_sock struct
1323 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1324 struct socket
*sock
,
1327 struct svc_sock
*svsk
;
1329 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1332 dprintk("svc: svc_setup_socket %p\n", sock
);
1333 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1335 return ERR_PTR(-ENOMEM
);
1339 /* Register socket with portmapper */
1341 err
= svc_register(serv
, sock_net(sock
->sk
), inet
->sk_family
,
1343 ntohs(inet_sk(inet
)->inet_sport
));
1347 return ERR_PTR(err
);
1350 inet
->sk_user_data
= svsk
;
1351 svsk
->sk_sock
= sock
;
1353 svsk
->sk_ostate
= inet
->sk_state_change
;
1354 svsk
->sk_odata
= inet
->sk_data_ready
;
1355 svsk
->sk_owspace
= inet
->sk_write_space
;
1357 /* Initialize the socket */
1358 if (sock
->type
== SOCK_DGRAM
)
1359 svc_udp_init(svsk
, serv
);
1361 svc_tcp_init(svsk
, serv
);
1363 dprintk("svc: svc_setup_socket created %p (inet %p), "
1364 "listen %d close %d\n",
1366 test_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
),
1367 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1372 bool svc_alien_sock(struct net
*net
, int fd
)
1375 struct socket
*sock
= sockfd_lookup(fd
, &err
);
1380 if (sock_net(sock
->sk
) != net
)
1386 EXPORT_SYMBOL_GPL(svc_alien_sock
);
1389 * svc_addsock - add a listener socket to an RPC service
1390 * @serv: pointer to RPC service to which to add a new listener
1391 * @fd: file descriptor of the new listener
1392 * @name_return: pointer to buffer to fill in with name of listener
1393 * @len: size of the buffer
1395 * Fills in socket name and returns positive length of name if successful.
1396 * Name is terminated with '\n'. On error, returns a negative errno
1399 int svc_addsock(struct svc_serv
*serv
, const int fd
, char *name_return
,
1403 struct socket
*so
= sockfd_lookup(fd
, &err
);
1404 struct svc_sock
*svsk
= NULL
;
1405 struct sockaddr_storage addr
;
1406 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1411 err
= -EAFNOSUPPORT
;
1412 if ((so
->sk
->sk_family
!= PF_INET
) && (so
->sk
->sk_family
!= PF_INET6
))
1414 err
= -EPROTONOSUPPORT
;
1415 if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1416 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1419 if (so
->state
> SS_UNCONNECTED
)
1422 if (!try_module_get(THIS_MODULE
))
1424 svsk
= svc_setup_socket(serv
, so
, SVC_SOCK_DEFAULTS
);
1426 module_put(THIS_MODULE
);
1427 err
= PTR_ERR(svsk
);
1430 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1431 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1432 svc_add_new_perm_xprt(serv
, &svsk
->sk_xprt
);
1433 return svc_one_sock_name(svsk
, name_return
, len
);
1438 EXPORT_SYMBOL_GPL(svc_addsock
);
1441 * Create socket for RPC service.
1443 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1446 struct sockaddr
*sin
, int len
,
1449 struct svc_sock
*svsk
;
1450 struct socket
*sock
;
1453 struct sockaddr_storage addr
;
1454 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1458 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1460 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1461 serv
->sv_program
->pg_name
, protocol
,
1462 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1464 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1465 printk(KERN_WARNING
"svc: only UDP and TCP "
1466 "sockets supported\n");
1467 return ERR_PTR(-EINVAL
);
1470 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1471 switch (sin
->sa_family
) {
1479 return ERR_PTR(-EINVAL
);
1482 error
= __sock_create(net
, family
, type
, protocol
, &sock
, 1);
1484 return ERR_PTR(error
);
1486 svc_reclassify_socket(sock
);
1489 * If this is an PF_INET6 listener, we want to avoid
1490 * getting requests from IPv4 remotes. Those should
1491 * be shunted to a PF_INET listener via rpcbind.
1494 if (family
== PF_INET6
)
1495 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1496 (char *)&val
, sizeof(val
));
1498 if (type
== SOCK_STREAM
)
1499 sock
->sk
->sk_reuse
= SK_CAN_REUSE
; /* allow address reuse */
1500 error
= kernel_bind(sock
, sin
, len
);
1505 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1509 if (protocol
== IPPROTO_TCP
) {
1510 if ((error
= kernel_listen(sock
, 64)) < 0)
1514 svsk
= svc_setup_socket(serv
, sock
, flags
);
1516 error
= PTR_ERR(svsk
);
1519 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1520 return (struct svc_xprt
*)svsk
;
1522 dprintk("svc: svc_create_socket error = %d\n", -error
);
1524 return ERR_PTR(error
);
1528 * Detach the svc_sock from the socket so that no
1529 * more callbacks occur.
1531 static void svc_sock_detach(struct svc_xprt
*xprt
)
1533 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1534 struct sock
*sk
= svsk
->sk_sk
;
1536 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1538 /* put back the old socket callbacks */
1540 sk
->sk_state_change
= svsk
->sk_ostate
;
1541 sk
->sk_data_ready
= svsk
->sk_odata
;
1542 sk
->sk_write_space
= svsk
->sk_owspace
;
1543 sk
->sk_user_data
= NULL
;
1548 * Disconnect the socket, and reset the callbacks
1550 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1552 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1554 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1556 svc_sock_detach(xprt
);
1558 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
)) {
1559 svc_tcp_clear_pages(svsk
);
1560 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1565 * Free the svc_sock's socket resources and the svc_sock itself.
1567 static void svc_sock_free(struct svc_xprt
*xprt
)
1569 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1570 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1572 if (svsk
->sk_sock
->file
)
1573 sockfd_put(svsk
->sk_sock
);
1575 sock_release(svsk
->sk_sock
);
1579 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1581 * Create a back channel svc_xprt which shares the fore channel socket.
1583 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*serv
,
1586 struct sockaddr
*sin
, int len
,
1589 struct svc_sock
*svsk
;
1590 struct svc_xprt
*xprt
;
1592 if (protocol
!= IPPROTO_TCP
) {
1593 printk(KERN_WARNING
"svc: only TCP sockets"
1594 " supported on shared back channel\n");
1595 return ERR_PTR(-EINVAL
);
1598 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1600 return ERR_PTR(-ENOMEM
);
1602 xprt
= &svsk
->sk_xprt
;
1603 svc_xprt_init(net
, &svc_tcp_bc_class
, xprt
, serv
);
1605 serv
->sv_bc_xprt
= xprt
;
1611 * Free a back channel svc_sock.
1613 static void svc_bc_sock_free(struct svc_xprt
*xprt
)
1616 kfree(container_of(xprt
, struct svc_sock
, sk_xprt
));
1618 #endif /* CONFIG_SUNRPC_BACKCHANNEL */