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
*);
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
);
294 #if IS_ENABLED(CONFIG_IPV6)
296 len
= snprintf(buf
, remaining
, "ipv6 %s %pI6 %d\n",
298 &sk
->sk_v6_rcv_saddr
,
299 inet_sk(sk
)->inet_num
);
303 len
= snprintf(buf
, remaining
, "*unknown-%d*\n",
307 if (len
>= remaining
) {
309 return -ENAMETOOLONG
;
315 * Generic recvfrom routine.
317 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
320 struct svc_sock
*svsk
=
321 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
322 struct msghdr msg
= {
323 .msg_flags
= MSG_DONTWAIT
,
327 rqstp
->rq_xprt_hlen
= 0;
329 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
330 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
332 /* If we read a full record, then assume there may be more
333 * data to read (stream based sockets only!)
336 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
338 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
339 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
343 static int svc_partial_recvfrom(struct svc_rqst
*rqstp
,
344 struct kvec
*iov
, int nr
,
345 int buflen
, unsigned int base
)
353 return svc_recvfrom(rqstp
, iov
, nr
, buflen
);
355 for (i
= 0; i
< nr
; i
++) {
356 if (iov
[i
].iov_len
> base
)
358 base
-= iov
[i
].iov_len
;
360 save_iovlen
= iov
[i
].iov_len
;
361 save_iovbase
= iov
[i
].iov_base
;
362 iov
[i
].iov_len
-= base
;
363 iov
[i
].iov_base
+= base
;
364 ret
= svc_recvfrom(rqstp
, &iov
[i
], nr
- i
, buflen
);
365 iov
[i
].iov_len
= save_iovlen
;
366 iov
[i
].iov_base
= save_iovbase
;
371 * Set socket snd and rcv buffer lengths
373 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
378 oldfs
= get_fs(); set_fs(KERNEL_DS
);
379 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
380 (char*)&snd
, sizeof(snd
));
381 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
382 (char*)&rcv
, sizeof(rcv
));
384 /* sock_setsockopt limits use to sysctl_?mem_max,
385 * which isn't acceptable. Until that is made conditional
386 * on not having CAP_SYS_RESOURCE or similar, we go direct...
387 * DaveM said I could!
390 sock
->sk
->sk_sndbuf
= snd
* 2;
391 sock
->sk
->sk_rcvbuf
= rcv
* 2;
392 sock
->sk
->sk_write_space(sock
->sk
);
393 release_sock(sock
->sk
);
397 static int svc_sock_secure_port(struct svc_rqst
*rqstp
)
399 return svc_port_is_privileged(svc_addr(rqstp
));
403 * INET callback when data has been received on the socket.
405 static void svc_udp_data_ready(struct sock
*sk
)
407 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
408 wait_queue_head_t
*wq
= sk_sleep(sk
);
411 dprintk("svc: socket %p(inet %p), busy=%d\n",
413 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
414 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
415 svc_xprt_enqueue(&svsk
->sk_xprt
);
417 if (wq
&& waitqueue_active(wq
))
418 wake_up_interruptible(wq
);
422 * INET callback when space is newly available on the socket.
424 static void svc_write_space(struct sock
*sk
)
426 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
427 wait_queue_head_t
*wq
= sk_sleep(sk
);
430 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
431 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
432 svc_xprt_enqueue(&svsk
->sk_xprt
);
435 if (wq
&& waitqueue_active(wq
)) {
436 dprintk("RPC svc_write_space: someone sleeping on %p\n",
438 wake_up_interruptible(wq
);
442 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
444 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
445 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
448 if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
450 required
= atomic_read(&xprt
->xpt_reserved
) + serv
->sv_max_mesg
;
451 if (sk_stream_wspace(svsk
->sk_sk
) >= required
||
452 (sk_stream_min_wspace(svsk
->sk_sk
) == 0 &&
453 atomic_read(&xprt
->xpt_reserved
) == 0))
455 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
459 static void svc_tcp_write_space(struct sock
*sk
)
461 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
462 struct socket
*sock
= sk
->sk_socket
;
464 if (!sk_stream_is_writeable(sk
) || !sock
)
466 if (!svsk
|| svc_tcp_has_wspace(&svsk
->sk_xprt
))
467 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
471 static void svc_tcp_adjust_wspace(struct svc_xprt
*xprt
)
473 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
475 if (svc_tcp_has_wspace(xprt
))
476 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
480 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
482 static int svc_udp_get_dest_address4(struct svc_rqst
*rqstp
,
485 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
486 struct sockaddr_in
*daddr
= svc_daddr_in(rqstp
);
488 if (cmh
->cmsg_type
!= IP_PKTINFO
)
491 daddr
->sin_family
= AF_INET
;
492 daddr
->sin_addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
497 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
499 static int svc_udp_get_dest_address6(struct svc_rqst
*rqstp
,
502 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
503 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
505 if (cmh
->cmsg_type
!= IPV6_PKTINFO
)
508 daddr
->sin6_family
= AF_INET6
;
509 daddr
->sin6_addr
= pki
->ipi6_addr
;
510 daddr
->sin6_scope_id
= pki
->ipi6_ifindex
;
515 * Copy the UDP datagram's destination address to the rqstp structure.
516 * The 'destination' address in this case is the address to which the
517 * peer sent the datagram, i.e. our local address. For multihomed
518 * hosts, this can change from msg to msg. Note that only the IP
519 * address changes, the port number should remain the same.
521 static int svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
524 switch (cmh
->cmsg_level
) {
526 return svc_udp_get_dest_address4(rqstp
, cmh
);
528 return svc_udp_get_dest_address6(rqstp
, cmh
);
535 * Receive a datagram from a UDP socket.
537 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
539 struct svc_sock
*svsk
=
540 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
541 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
545 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
547 struct cmsghdr
*cmh
= &buffer
.hdr
;
548 struct msghdr msg
= {
549 .msg_name
= svc_addr(rqstp
),
551 .msg_controllen
= sizeof(buffer
),
552 .msg_flags
= MSG_DONTWAIT
,
557 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
558 /* udp sockets need large rcvbuf as all pending
559 * requests are still in that buffer. sndbuf must
560 * also be large enough that there is enough space
561 * for one reply per thread. We count all threads
562 * rather than threads in a particular pool, which
563 * provides an upper bound on the number of threads
564 * which will access the socket.
566 svc_sock_setbufsize(svsk
->sk_sock
,
567 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
568 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
570 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
572 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
573 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
575 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
578 if (err
!= -EAGAIN
) {
579 /* possibly an icmp error */
580 dprintk("svc: recvfrom returned error %d\n", -err
);
581 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
585 len
= svc_addr_len(svc_addr(rqstp
));
586 rqstp
->rq_addrlen
= len
;
587 if (skb
->tstamp
.tv64
== 0) {
588 skb
->tstamp
= ktime_get_real();
589 /* Don't enable netstamp, sunrpc doesn't
590 need that much accuracy */
592 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
593 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
595 len
= skb
->len
- sizeof(struct udphdr
);
596 rqstp
->rq_arg
.len
= len
;
598 rqstp
->rq_prot
= IPPROTO_UDP
;
600 if (!svc_udp_get_dest_address(rqstp
, cmh
)) {
601 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
602 cmh
->cmsg_level
, cmh
->cmsg_type
);
605 rqstp
->rq_daddrlen
= svc_addr_len(svc_daddr(rqstp
));
607 if (skb_is_nonlinear(skb
)) {
608 /* we have to copy */
610 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
616 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
618 /* we can use it in-place */
619 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+
620 sizeof(struct udphdr
);
621 rqstp
->rq_arg
.head
[0].iov_len
= len
;
622 if (skb_checksum_complete(skb
))
624 rqstp
->rq_xprt_ctxt
= skb
;
627 rqstp
->rq_arg
.page_base
= 0;
628 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
629 rqstp
->rq_arg
.head
[0].iov_len
= len
;
630 rqstp
->rq_arg
.page_len
= 0;
631 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
633 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
634 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
635 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
637 rqstp
->rq_next_page
= rqstp
->rq_respages
+1;
640 serv
->sv_stats
->netudpcnt
++;
644 trace_kfree_skb(skb
, svc_udp_recvfrom
);
645 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
650 svc_udp_sendto(struct svc_rqst
*rqstp
)
654 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
655 if (error
== -ECONNREFUSED
)
656 /* ICMP error on earlier request. */
657 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
662 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
666 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
668 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
669 struct svc_serv
*serv
= xprt
->xpt_server
;
670 unsigned long required
;
673 * Set the SOCK_NOSPACE flag before checking the available
676 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
677 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
678 if (required
*2 > sock_wspace(svsk
->sk_sk
))
680 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
684 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
690 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
692 struct sockaddr
*sa
, int salen
,
695 return svc_create_socket(serv
, IPPROTO_UDP
, net
, sa
, salen
, flags
);
698 static struct svc_xprt_ops svc_udp_ops
= {
699 .xpo_create
= svc_udp_create
,
700 .xpo_recvfrom
= svc_udp_recvfrom
,
701 .xpo_sendto
= svc_udp_sendto
,
702 .xpo_release_rqst
= svc_release_skb
,
703 .xpo_detach
= svc_sock_detach
,
704 .xpo_free
= svc_sock_free
,
705 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
706 .xpo_has_wspace
= svc_udp_has_wspace
,
707 .xpo_accept
= svc_udp_accept
,
708 .xpo_secure_port
= svc_sock_secure_port
,
711 static struct svc_xprt_class svc_udp_class
= {
713 .xcl_owner
= THIS_MODULE
,
714 .xcl_ops
= &svc_udp_ops
,
715 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
716 .xcl_ident
= XPRT_TRANSPORT_UDP
,
719 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
721 int err
, level
, optname
, one
= 1;
723 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_udp_class
,
724 &svsk
->sk_xprt
, serv
);
725 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
726 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
727 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
729 /* initialise setting must have enough space to
730 * receive and respond to one request.
731 * svc_udp_recvfrom will re-adjust if necessary
733 svc_sock_setbufsize(svsk
->sk_sock
,
734 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
735 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
737 /* data might have come in before data_ready set up */
738 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
739 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
741 /* make sure we get destination address info */
742 switch (svsk
->sk_sk
->sk_family
) {
745 optname
= IP_PKTINFO
;
749 optname
= IPV6_RECVPKTINFO
;
754 err
= kernel_setsockopt(svsk
->sk_sock
, level
, optname
,
755 (char *)&one
, sizeof(one
));
756 dprintk("svc: kernel_setsockopt returned %d\n", err
);
760 * A data_ready event on a listening socket means there's a connection
761 * pending. Do not use state_change as a substitute for it.
763 static void svc_tcp_listen_data_ready(struct sock
*sk
)
765 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
766 wait_queue_head_t
*wq
;
768 dprintk("svc: socket %p TCP (listen) state change %d\n",
772 * This callback may called twice when a new connection
773 * is established as a child socket inherits everything
774 * from a parent LISTEN socket.
775 * 1) data_ready method of the parent socket will be called
776 * when one of child sockets become ESTABLISHED.
777 * 2) data_ready method of the child socket may be called
778 * when it receives data before the socket is accepted.
779 * In case of 2, we should ignore it silently.
781 if (sk
->sk_state
== TCP_LISTEN
) {
783 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
784 svc_xprt_enqueue(&svsk
->sk_xprt
);
786 printk("svc: socket %p: no user data\n", sk
);
790 if (wq
&& waitqueue_active(wq
))
791 wake_up_interruptible_all(wq
);
795 * A state change on a connected socket means it's dying or dead.
797 static void svc_tcp_state_change(struct sock
*sk
)
799 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
800 wait_queue_head_t
*wq
= sk_sleep(sk
);
802 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
803 sk
, sk
->sk_state
, sk
->sk_user_data
);
806 printk("svc: socket %p: no user data\n", sk
);
808 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
809 svc_xprt_enqueue(&svsk
->sk_xprt
);
811 if (wq
&& waitqueue_active(wq
))
812 wake_up_interruptible_all(wq
);
815 static void svc_tcp_data_ready(struct sock
*sk
)
817 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
818 wait_queue_head_t
*wq
= sk_sleep(sk
);
820 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
821 sk
, sk
->sk_user_data
);
823 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
824 svc_xprt_enqueue(&svsk
->sk_xprt
);
826 if (wq
&& waitqueue_active(wq
))
827 wake_up_interruptible(wq
);
831 * Accept a TCP connection
833 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
835 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
836 struct sockaddr_storage addr
;
837 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
838 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
839 struct socket
*sock
= svsk
->sk_sock
;
840 struct socket
*newsock
;
841 struct svc_sock
*newsvsk
;
843 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
845 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
849 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
850 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
853 printk(KERN_WARNING
"%s: no more sockets!\n",
855 else if (err
!= -EAGAIN
)
856 net_warn_ratelimited("%s: accept failed (err %d)!\n",
857 serv
->sv_name
, -err
);
860 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
862 err
= kernel_getpeername(newsock
, sin
, &slen
);
864 net_warn_ratelimited("%s: peername failed (err %d)!\n",
865 serv
->sv_name
, -err
);
866 goto failed
; /* aborted connection or whatever */
869 /* Ideally, we would want to reject connections from unauthorized
870 * hosts here, but when we get encryption, the IP of the host won't
871 * tell us anything. For now just warn about unpriv connections.
873 if (!svc_port_is_privileged(sin
)) {
874 dprintk("%s: connect from unprivileged port: %s\n",
876 __svc_print_addr(sin
, buf
, sizeof(buf
)));
878 dprintk("%s: connect from %s\n", serv
->sv_name
,
879 __svc_print_addr(sin
, buf
, sizeof(buf
)));
881 /* make sure that a write doesn't block forever when
884 newsock
->sk
->sk_sndtimeo
= HZ
*30;
886 newsvsk
= svc_setup_socket(serv
, newsock
,
887 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
));
890 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
891 err
= kernel_getsockname(newsock
, sin
, &slen
);
892 if (unlikely(err
< 0)) {
893 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
894 slen
= offsetof(struct sockaddr
, sa_data
);
896 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
898 if (sock_is_loopback(newsock
->sk
))
899 set_bit(XPT_LOCAL
, &newsvsk
->sk_xprt
.xpt_flags
);
901 clear_bit(XPT_LOCAL
, &newsvsk
->sk_xprt
.xpt_flags
);
903 serv
->sv_stats
->nettcpconn
++;
905 return &newsvsk
->sk_xprt
;
908 sock_release(newsock
);
912 static unsigned int svc_tcp_restore_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
914 unsigned int i
, len
, npages
;
916 if (svsk
->sk_datalen
== 0)
918 len
= svsk
->sk_datalen
;
919 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
920 for (i
= 0; i
< npages
; i
++) {
921 if (rqstp
->rq_pages
[i
] != NULL
)
922 put_page(rqstp
->rq_pages
[i
]);
923 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
924 rqstp
->rq_pages
[i
] = svsk
->sk_pages
[i
];
925 svsk
->sk_pages
[i
] = NULL
;
927 rqstp
->rq_arg
.head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
931 static void svc_tcp_save_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
933 unsigned int i
, len
, npages
;
935 if (svsk
->sk_datalen
== 0)
937 len
= svsk
->sk_datalen
;
938 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
939 for (i
= 0; i
< npages
; i
++) {
940 svsk
->sk_pages
[i
] = rqstp
->rq_pages
[i
];
941 rqstp
->rq_pages
[i
] = NULL
;
945 static void svc_tcp_clear_pages(struct svc_sock
*svsk
)
947 unsigned int i
, len
, npages
;
949 if (svsk
->sk_datalen
== 0)
951 len
= svsk
->sk_datalen
;
952 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
953 for (i
= 0; i
< npages
; i
++) {
954 if (svsk
->sk_pages
[i
] == NULL
) {
958 put_page(svsk
->sk_pages
[i
]);
959 svsk
->sk_pages
[i
] = NULL
;
963 svsk
->sk_datalen
= 0;
967 * Receive fragment record header.
968 * If we haven't gotten the record length yet, get the next four bytes.
970 static int svc_tcp_recv_record(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
972 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
976 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
979 want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
980 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
982 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
984 svsk
->sk_tcplen
+= len
;
987 dprintk("svc: short recvfrom while reading record "
988 "length (%d of %d)\n", len
, want
);
992 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk
));
993 if (svc_sock_reclen(svsk
) + svsk
->sk_datalen
>
995 net_notice_ratelimited("RPC: fragment too large: %d\n",
996 svc_sock_reclen(svsk
));
1001 return svc_sock_reclen(svsk
);
1003 dprintk("RPC: TCP recv_record got %d\n", len
);
1006 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1010 static int receive_cb_reply(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
1012 struct rpc_xprt
*bc_xprt
= svsk
->sk_xprt
.xpt_bc_xprt
;
1013 struct rpc_rqst
*req
= NULL
;
1014 struct kvec
*src
, *dst
;
1015 __be32
*p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1023 req
= xprt_lookup_rqst(bc_xprt
, xid
);
1027 "%s: Got unrecognized reply: "
1028 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1029 __func__
, ntohl(calldir
),
1030 bc_xprt
, ntohl(xid
));
1034 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
, sizeof(struct xdr_buf
));
1036 * XXX!: cheating for now! Only copying HEAD.
1037 * But we know this is good enough for now (in fact, for any
1038 * callback reply in the forseeable future).
1040 dst
= &req
->rq_private_buf
.head
[0];
1041 src
= &rqstp
->rq_arg
.head
[0];
1042 if (dst
->iov_len
< src
->iov_len
)
1043 return -EAGAIN
; /* whatever; just giving up. */
1044 memcpy(dst
->iov_base
, src
->iov_base
, src
->iov_len
);
1045 xprt_complete_rqst(req
->rq_task
, rqstp
->rq_arg
.len
);
1046 rqstp
->rq_arg
.len
= 0;
1050 static int copy_pages_to_kvecs(struct kvec
*vec
, struct page
**pages
, int len
)
1056 vec
[i
].iov_base
= page_address(pages
[i
]);
1057 vec
[i
].iov_len
= PAGE_SIZE
;
1064 static void svc_tcp_fragment_received(struct svc_sock
*svsk
)
1066 /* If we have more data, signal svc_xprt_enqueue() to try again */
1067 dprintk("svc: TCP %s record (%d bytes)\n",
1068 svc_sock_final_rec(svsk
) ? "final" : "nonfinal",
1069 svc_sock_reclen(svsk
));
1070 svsk
->sk_tcplen
= 0;
1071 svsk
->sk_reclen
= 0;
1075 * Receive data from a TCP socket.
1077 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1079 struct svc_sock
*svsk
=
1080 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
1081 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1084 unsigned int want
, base
;
1089 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1090 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
1091 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
1092 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1094 len
= svc_tcp_recv_record(svsk
, rqstp
);
1098 base
= svc_tcp_restore_pages(svsk
, rqstp
);
1099 want
= svc_sock_reclen(svsk
) - (svsk
->sk_tcplen
- sizeof(rpc_fraghdr
));
1101 vec
= rqstp
->rq_vec
;
1103 pnum
= copy_pages_to_kvecs(&vec
[0], &rqstp
->rq_pages
[0],
1104 svsk
->sk_datalen
+ want
);
1106 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1107 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
1109 /* Now receive data */
1110 len
= svc_partial_recvfrom(rqstp
, vec
, pnum
, want
, base
);
1112 svsk
->sk_tcplen
+= len
;
1113 svsk
->sk_datalen
+= len
;
1115 if (len
!= want
|| !svc_sock_final_rec(svsk
)) {
1116 svc_tcp_save_pages(svsk
, rqstp
);
1117 if (len
< 0 && len
!= -EAGAIN
)
1120 svc_tcp_fragment_received(svsk
);
1122 dprintk("svc: incomplete TCP record (%d of %d)\n",
1123 (int)(svsk
->sk_tcplen
- sizeof(rpc_fraghdr
)),
1124 svc_sock_reclen(svsk
));
1128 if (svsk
->sk_datalen
< 8) {
1129 svsk
->sk_datalen
= 0;
1130 goto err_delete
; /* client is nuts. */
1133 rqstp
->rq_arg
.len
= svsk
->sk_datalen
;
1134 rqstp
->rq_arg
.page_base
= 0;
1135 if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1136 rqstp
->rq_arg
.head
[0].iov_len
= rqstp
->rq_arg
.len
;
1137 rqstp
->rq_arg
.page_len
= 0;
1139 rqstp
->rq_arg
.page_len
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1141 rqstp
->rq_xprt_ctxt
= NULL
;
1142 rqstp
->rq_prot
= IPPROTO_TCP
;
1143 rqstp
->rq_local
= !!test_bit(XPT_LOCAL
, &svsk
->sk_xprt
.xpt_flags
);
1145 p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1148 len
= receive_cb_reply(svsk
, rqstp
);
1150 /* Reset TCP read info */
1151 svsk
->sk_datalen
= 0;
1152 svc_tcp_fragment_received(svsk
);
1157 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1159 serv
->sv_stats
->nettcpcnt
++;
1161 return rqstp
->rq_arg
.len
;
1166 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1169 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1170 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1171 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1173 return 0; /* record not complete */
1177 * Send out data on TCP socket.
1179 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
1181 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1185 /* Set up the first element of the reply kvec.
1186 * Any other kvecs that may be in use have been taken
1187 * care of by the server implementation itself.
1189 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1190 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1192 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1193 if (sent
!= xbufp
->len
) {
1195 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1196 "- shutting down socket\n",
1197 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1198 (sent
<0)?"got error":"sent only",
1200 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1201 svc_xprt_enqueue(rqstp
->rq_xprt
);
1208 * Setup response header. TCP has a 4B record length field.
1210 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1212 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1214 /* tcp needs a space for the record length... */
1218 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1220 struct sockaddr
*sa
, int salen
,
1223 return svc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1226 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1227 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
1228 struct net
*, struct sockaddr
*,
1230 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
1232 static struct svc_xprt
*svc_bc_tcp_create(struct svc_serv
*serv
,
1234 struct sockaddr
*sa
, int salen
,
1237 return svc_bc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1240 static void svc_bc_tcp_sock_detach(struct svc_xprt
*xprt
)
1244 static struct svc_xprt_ops svc_tcp_bc_ops
= {
1245 .xpo_create
= svc_bc_tcp_create
,
1246 .xpo_detach
= svc_bc_tcp_sock_detach
,
1247 .xpo_free
= svc_bc_sock_free
,
1248 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1249 .xpo_secure_port
= svc_sock_secure_port
,
1252 static struct svc_xprt_class svc_tcp_bc_class
= {
1253 .xcl_name
= "tcp-bc",
1254 .xcl_owner
= THIS_MODULE
,
1255 .xcl_ops
= &svc_tcp_bc_ops
,
1256 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1259 static void svc_init_bc_xprt_sock(void)
1261 svc_reg_xprt_class(&svc_tcp_bc_class
);
1264 static void svc_cleanup_bc_xprt_sock(void)
1266 svc_unreg_xprt_class(&svc_tcp_bc_class
);
1268 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1269 static void svc_init_bc_xprt_sock(void)
1273 static void svc_cleanup_bc_xprt_sock(void)
1276 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1278 static struct svc_xprt_ops svc_tcp_ops
= {
1279 .xpo_create
= svc_tcp_create
,
1280 .xpo_recvfrom
= svc_tcp_recvfrom
,
1281 .xpo_sendto
= svc_tcp_sendto
,
1282 .xpo_release_rqst
= svc_release_skb
,
1283 .xpo_detach
= svc_tcp_sock_detach
,
1284 .xpo_free
= svc_sock_free
,
1285 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1286 .xpo_has_wspace
= svc_tcp_has_wspace
,
1287 .xpo_accept
= svc_tcp_accept
,
1288 .xpo_secure_port
= svc_sock_secure_port
,
1289 .xpo_adjust_wspace
= svc_tcp_adjust_wspace
,
1292 static struct svc_xprt_class svc_tcp_class
= {
1294 .xcl_owner
= THIS_MODULE
,
1295 .xcl_ops
= &svc_tcp_ops
,
1296 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1297 .xcl_ident
= XPRT_TRANSPORT_TCP
,
1300 void svc_init_xprt_sock(void)
1302 svc_reg_xprt_class(&svc_tcp_class
);
1303 svc_reg_xprt_class(&svc_udp_class
);
1304 svc_init_bc_xprt_sock();
1307 void svc_cleanup_xprt_sock(void)
1309 svc_unreg_xprt_class(&svc_tcp_class
);
1310 svc_unreg_xprt_class(&svc_udp_class
);
1311 svc_cleanup_bc_xprt_sock();
1314 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1316 struct sock
*sk
= svsk
->sk_sk
;
1318 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_tcp_class
,
1319 &svsk
->sk_xprt
, serv
);
1320 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1321 if (sk
->sk_state
== TCP_LISTEN
) {
1322 dprintk("setting up TCP socket for listening\n");
1323 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1324 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1325 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1327 dprintk("setting up TCP socket for reading\n");
1328 sk
->sk_state_change
= svc_tcp_state_change
;
1329 sk
->sk_data_ready
= svc_tcp_data_ready
;
1330 sk
->sk_write_space
= svc_tcp_write_space
;
1332 svsk
->sk_reclen
= 0;
1333 svsk
->sk_tcplen
= 0;
1334 svsk
->sk_datalen
= 0;
1335 memset(&svsk
->sk_pages
[0], 0, sizeof(svsk
->sk_pages
));
1337 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1339 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1340 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1341 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1345 void svc_sock_update_bufs(struct svc_serv
*serv
)
1348 * The number of server threads has changed. Update
1349 * rcvbuf and sndbuf accordingly on all sockets
1351 struct svc_sock
*svsk
;
1353 spin_lock_bh(&serv
->sv_lock
);
1354 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
)
1355 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1356 spin_unlock_bh(&serv
->sv_lock
);
1358 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1361 * Initialize socket for RPC use and create svc_sock struct
1362 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1364 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1365 struct socket
*sock
,
1368 struct svc_sock
*svsk
;
1370 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1373 dprintk("svc: svc_setup_socket %p\n", sock
);
1374 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1376 return ERR_PTR(-ENOMEM
);
1380 /* Register socket with portmapper */
1382 err
= svc_register(serv
, sock_net(sock
->sk
), inet
->sk_family
,
1384 ntohs(inet_sk(inet
)->inet_sport
));
1388 return ERR_PTR(err
);
1391 inet
->sk_user_data
= svsk
;
1392 svsk
->sk_sock
= sock
;
1394 svsk
->sk_ostate
= inet
->sk_state_change
;
1395 svsk
->sk_odata
= inet
->sk_data_ready
;
1396 svsk
->sk_owspace
= inet
->sk_write_space
;
1398 /* Initialize the socket */
1399 if (sock
->type
== SOCK_DGRAM
)
1400 svc_udp_init(svsk
, serv
);
1402 /* initialise setting must have enough space to
1403 * receive and respond to one request.
1405 svc_sock_setbufsize(svsk
->sk_sock
, 4 * serv
->sv_max_mesg
,
1406 4 * serv
->sv_max_mesg
);
1407 svc_tcp_init(svsk
, serv
);
1410 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1416 bool svc_alien_sock(struct net
*net
, int fd
)
1419 struct socket
*sock
= sockfd_lookup(fd
, &err
);
1424 if (sock_net(sock
->sk
) != net
)
1430 EXPORT_SYMBOL_GPL(svc_alien_sock
);
1433 * svc_addsock - add a listener socket to an RPC service
1434 * @serv: pointer to RPC service to which to add a new listener
1435 * @fd: file descriptor of the new listener
1436 * @name_return: pointer to buffer to fill in with name of listener
1437 * @len: size of the buffer
1439 * Fills in socket name and returns positive length of name if successful.
1440 * Name is terminated with '\n'. On error, returns a negative errno
1443 int svc_addsock(struct svc_serv
*serv
, const int fd
, char *name_return
,
1447 struct socket
*so
= sockfd_lookup(fd
, &err
);
1448 struct svc_sock
*svsk
= NULL
;
1449 struct sockaddr_storage addr
;
1450 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1455 err
= -EAFNOSUPPORT
;
1456 if ((so
->sk
->sk_family
!= PF_INET
) && (so
->sk
->sk_family
!= PF_INET6
))
1458 err
= -EPROTONOSUPPORT
;
1459 if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1460 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1463 if (so
->state
> SS_UNCONNECTED
)
1466 if (!try_module_get(THIS_MODULE
))
1468 svsk
= svc_setup_socket(serv
, so
, SVC_SOCK_DEFAULTS
);
1470 module_put(THIS_MODULE
);
1471 err
= PTR_ERR(svsk
);
1474 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1475 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1476 svc_add_new_perm_xprt(serv
, &svsk
->sk_xprt
);
1477 return svc_one_sock_name(svsk
, name_return
, len
);
1482 EXPORT_SYMBOL_GPL(svc_addsock
);
1485 * Create socket for RPC service.
1487 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1490 struct sockaddr
*sin
, int len
,
1493 struct svc_sock
*svsk
;
1494 struct socket
*sock
;
1497 struct sockaddr_storage addr
;
1498 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1502 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1504 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1505 serv
->sv_program
->pg_name
, protocol
,
1506 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1508 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1509 printk(KERN_WARNING
"svc: only UDP and TCP "
1510 "sockets supported\n");
1511 return ERR_PTR(-EINVAL
);
1514 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1515 switch (sin
->sa_family
) {
1523 return ERR_PTR(-EINVAL
);
1526 error
= __sock_create(net
, family
, type
, protocol
, &sock
, 1);
1528 return ERR_PTR(error
);
1530 svc_reclassify_socket(sock
);
1533 * If this is an PF_INET6 listener, we want to avoid
1534 * getting requests from IPv4 remotes. Those should
1535 * be shunted to a PF_INET listener via rpcbind.
1538 if (family
== PF_INET6
)
1539 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1540 (char *)&val
, sizeof(val
));
1542 if (type
== SOCK_STREAM
)
1543 sock
->sk
->sk_reuse
= SK_CAN_REUSE
; /* allow address reuse */
1544 error
= kernel_bind(sock
, sin
, len
);
1549 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1553 if (protocol
== IPPROTO_TCP
) {
1554 if ((error
= kernel_listen(sock
, 64)) < 0)
1558 svsk
= svc_setup_socket(serv
, sock
, flags
);
1560 error
= PTR_ERR(svsk
);
1563 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1564 return (struct svc_xprt
*)svsk
;
1566 dprintk("svc: svc_create_socket error = %d\n", -error
);
1568 return ERR_PTR(error
);
1572 * Detach the svc_sock from the socket so that no
1573 * more callbacks occur.
1575 static void svc_sock_detach(struct svc_xprt
*xprt
)
1577 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1578 struct sock
*sk
= svsk
->sk_sk
;
1579 wait_queue_head_t
*wq
;
1581 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1583 /* put back the old socket callbacks */
1584 sk
->sk_state_change
= svsk
->sk_ostate
;
1585 sk
->sk_data_ready
= svsk
->sk_odata
;
1586 sk
->sk_write_space
= svsk
->sk_owspace
;
1589 if (wq
&& waitqueue_active(wq
))
1590 wake_up_interruptible(wq
);
1594 * Disconnect the socket, and reset the callbacks
1596 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1598 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1600 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1602 svc_sock_detach(xprt
);
1604 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
)) {
1605 svc_tcp_clear_pages(svsk
);
1606 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1611 * Free the svc_sock's socket resources and the svc_sock itself.
1613 static void svc_sock_free(struct svc_xprt
*xprt
)
1615 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1616 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1618 if (svsk
->sk_sock
->file
)
1619 sockfd_put(svsk
->sk_sock
);
1621 sock_release(svsk
->sk_sock
);
1625 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1627 * Create a back channel svc_xprt which shares the fore channel socket.
1629 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*serv
,
1632 struct sockaddr
*sin
, int len
,
1635 struct svc_sock
*svsk
;
1636 struct svc_xprt
*xprt
;
1638 if (protocol
!= IPPROTO_TCP
) {
1639 printk(KERN_WARNING
"svc: only TCP sockets"
1640 " supported on shared back channel\n");
1641 return ERR_PTR(-EINVAL
);
1644 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1646 return ERR_PTR(-ENOMEM
);
1648 xprt
= &svsk
->sk_xprt
;
1649 svc_xprt_init(net
, &svc_tcp_bc_class
, xprt
, serv
);
1651 serv
->sv_bc_xprt
= xprt
;
1657 * Free a back channel svc_sock.
1659 static void svc_bc_sock_free(struct svc_xprt
*xprt
)
1662 kfree(container_of(xprt
, struct svc_sock
, sk_xprt
));
1664 #endif /* CONFIG_SUNRPC_BACKCHANNEL */