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 * svc_sock_names - construct a list of listener names in a string
310 * @serv: pointer to RPC service
311 * @buf: pointer to a buffer to fill in with socket names
312 * @buflen: size of the buffer to be filled
313 * @toclose: pointer to '\0'-terminated C string containing the name
314 * of a listener to be closed
316 * Fills in @buf with a '\n'-separated list of names of listener
317 * sockets. If @toclose is not NULL, the socket named by @toclose
318 * is closed, and is not included in the output list.
320 * Returns positive length of the socket name string, or a negative
321 * errno value on error.
323 int svc_sock_names(struct svc_serv
*serv
, char *buf
, const size_t buflen
,
326 struct svc_sock
*svsk
, *closesk
= NULL
;
332 spin_lock_bh(&serv
->sv_lock
);
333 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
) {
334 int onelen
= svc_one_sock_name(svsk
, buf
+ len
, buflen
- len
);
339 if (toclose
&& strcmp(toclose
, buf
+ len
) == 0) {
341 svc_xprt_get(&closesk
->sk_xprt
);
345 spin_unlock_bh(&serv
->sv_lock
);
348 /* Should unregister with portmap, but you cannot
349 * unregister just one protocol...
351 svc_close_xprt(&closesk
->sk_xprt
);
352 svc_xprt_put(&closesk
->sk_xprt
);
357 EXPORT_SYMBOL_GPL(svc_sock_names
);
360 * Check input queue length
362 static int svc_recv_available(struct svc_sock
*svsk
)
364 struct socket
*sock
= svsk
->sk_sock
;
367 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
369 return (err
>= 0)? avail
: err
;
373 * Generic recvfrom routine.
375 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
378 struct svc_sock
*svsk
=
379 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
380 struct msghdr msg
= {
381 .msg_flags
= MSG_DONTWAIT
,
385 rqstp
->rq_xprt_hlen
= 0;
387 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
390 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
391 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
395 static int svc_partial_recvfrom(struct svc_rqst
*rqstp
,
396 struct kvec
*iov
, int nr
,
397 int buflen
, unsigned int base
)
405 return svc_recvfrom(rqstp
, iov
, nr
, buflen
);
407 for (i
= 0; i
< nr
; i
++) {
408 if (iov
[i
].iov_len
> base
)
410 base
-= iov
[i
].iov_len
;
412 save_iovlen
= iov
[i
].iov_len
;
413 save_iovbase
= iov
[i
].iov_base
;
414 iov
[i
].iov_len
-= base
;
415 iov
[i
].iov_base
+= base
;
416 ret
= svc_recvfrom(rqstp
, &iov
[i
], nr
- i
, buflen
);
417 iov
[i
].iov_len
= save_iovlen
;
418 iov
[i
].iov_base
= save_iovbase
;
423 * Set socket snd and rcv buffer lengths
425 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
430 oldfs
= get_fs(); set_fs(KERNEL_DS
);
431 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
432 (char*)&snd
, sizeof(snd
));
433 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
434 (char*)&rcv
, sizeof(rcv
));
436 /* sock_setsockopt limits use to sysctl_?mem_max,
437 * which isn't acceptable. Until that is made conditional
438 * on not having CAP_SYS_RESOURCE or similar, we go direct...
439 * DaveM said I could!
442 sock
->sk
->sk_sndbuf
= snd
* 2;
443 sock
->sk
->sk_rcvbuf
= rcv
* 2;
444 sock
->sk
->sk_write_space(sock
->sk
);
445 release_sock(sock
->sk
);
449 * INET callback when data has been received on the socket.
451 static void svc_udp_data_ready(struct sock
*sk
, int count
)
453 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
454 wait_queue_head_t
*wq
= sk_sleep(sk
);
457 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
459 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
460 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
461 svc_xprt_enqueue(&svsk
->sk_xprt
);
463 if (wq
&& waitqueue_active(wq
))
464 wake_up_interruptible(wq
);
468 * INET callback when space is newly available on the socket.
470 static void svc_write_space(struct sock
*sk
)
472 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
473 wait_queue_head_t
*wq
= sk_sleep(sk
);
476 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
477 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
478 svc_xprt_enqueue(&svsk
->sk_xprt
);
481 if (wq
&& waitqueue_active(wq
)) {
482 dprintk("RPC svc_write_space: someone sleeping on %p\n",
484 wake_up_interruptible(wq
);
488 static void svc_tcp_write_space(struct sock
*sk
)
490 struct socket
*sock
= sk
->sk_socket
;
492 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
) && sock
)
493 clear_bit(SOCK_NOSPACE
, &sock
->flags
);
498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
500 static int svc_udp_get_dest_address4(struct svc_rqst
*rqstp
,
503 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
504 struct sockaddr_in
*daddr
= svc_daddr_in(rqstp
);
506 if (cmh
->cmsg_type
!= IP_PKTINFO
)
509 daddr
->sin_family
= AF_INET
;
510 daddr
->sin_addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
515 * See net/ipv6/datagram.c : datagram_recv_ctl
517 static int svc_udp_get_dest_address6(struct svc_rqst
*rqstp
,
520 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
521 struct sockaddr_in6
*daddr
= svc_daddr_in6(rqstp
);
523 if (cmh
->cmsg_type
!= IPV6_PKTINFO
)
526 daddr
->sin6_family
= AF_INET6
;
527 daddr
->sin6_addr
= pki
->ipi6_addr
;
528 daddr
->sin6_scope_id
= pki
->ipi6_ifindex
;
533 * Copy the UDP datagram's destination address to the rqstp structure.
534 * The 'destination' address in this case is the address to which the
535 * peer sent the datagram, i.e. our local address. For multihomed
536 * hosts, this can change from msg to msg. Note that only the IP
537 * address changes, the port number should remain the same.
539 static int svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
542 switch (cmh
->cmsg_level
) {
544 return svc_udp_get_dest_address4(rqstp
, cmh
);
546 return svc_udp_get_dest_address6(rqstp
, cmh
);
553 * Receive a datagram from a UDP socket.
555 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
557 struct svc_sock
*svsk
=
558 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
559 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
563 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
565 struct cmsghdr
*cmh
= &buffer
.hdr
;
566 struct msghdr msg
= {
567 .msg_name
= svc_addr(rqstp
),
569 .msg_controllen
= sizeof(buffer
),
570 .msg_flags
= MSG_DONTWAIT
,
575 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
576 /* udp sockets need large rcvbuf as all pending
577 * requests are still in that buffer. sndbuf must
578 * also be large enough that there is enough space
579 * for one reply per thread. We count all threads
580 * rather than threads in a particular pool, which
581 * provides an upper bound on the number of threads
582 * which will access the socket.
584 svc_sock_setbufsize(svsk
->sk_sock
,
585 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
586 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
588 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
590 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
591 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
593 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
596 if (err
!= -EAGAIN
) {
597 /* possibly an icmp error */
598 dprintk("svc: recvfrom returned error %d\n", -err
);
599 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
603 len
= svc_addr_len(svc_addr(rqstp
));
605 return -EAFNOSUPPORT
;
606 rqstp
->rq_addrlen
= len
;
607 if (skb
->tstamp
.tv64
== 0) {
608 skb
->tstamp
= ktime_get_real();
609 /* Don't enable netstamp, sunrpc doesn't
610 need that much accuracy */
612 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
613 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
615 len
= skb
->len
- sizeof(struct udphdr
);
616 rqstp
->rq_arg
.len
= len
;
618 rqstp
->rq_prot
= IPPROTO_UDP
;
620 if (!svc_udp_get_dest_address(rqstp
, cmh
)) {
621 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
622 cmh
->cmsg_level
, cmh
->cmsg_type
);
625 rqstp
->rq_daddrlen
= svc_addr_len(svc_daddr(rqstp
));
627 if (skb_is_nonlinear(skb
)) {
628 /* we have to copy */
630 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
636 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
638 /* we can use it in-place */
639 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+
640 sizeof(struct udphdr
);
641 rqstp
->rq_arg
.head
[0].iov_len
= len
;
642 if (skb_checksum_complete(skb
))
644 rqstp
->rq_xprt_ctxt
= skb
;
647 rqstp
->rq_arg
.page_base
= 0;
648 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
649 rqstp
->rq_arg
.head
[0].iov_len
= len
;
650 rqstp
->rq_arg
.page_len
= 0;
651 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
653 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
654 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
655 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
659 serv
->sv_stats
->netudpcnt
++;
663 trace_kfree_skb(skb
, svc_udp_recvfrom
);
664 skb_free_datagram_locked(svsk
->sk_sk
, skb
);
669 svc_udp_sendto(struct svc_rqst
*rqstp
)
673 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
674 if (error
== -ECONNREFUSED
)
675 /* ICMP error on earlier request. */
676 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
681 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
685 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
687 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
688 struct svc_serv
*serv
= xprt
->xpt_server
;
689 unsigned long required
;
692 * Set the SOCK_NOSPACE flag before checking the available
695 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
696 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
697 if (required
*2 > sock_wspace(svsk
->sk_sk
))
699 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
703 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
709 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
711 struct sockaddr
*sa
, int salen
,
714 return svc_create_socket(serv
, IPPROTO_UDP
, net
, sa
, salen
, flags
);
717 static struct svc_xprt_ops svc_udp_ops
= {
718 .xpo_create
= svc_udp_create
,
719 .xpo_recvfrom
= svc_udp_recvfrom
,
720 .xpo_sendto
= svc_udp_sendto
,
721 .xpo_release_rqst
= svc_release_skb
,
722 .xpo_detach
= svc_sock_detach
,
723 .xpo_free
= svc_sock_free
,
724 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
725 .xpo_has_wspace
= svc_udp_has_wspace
,
726 .xpo_accept
= svc_udp_accept
,
729 static struct svc_xprt_class svc_udp_class
= {
731 .xcl_owner
= THIS_MODULE
,
732 .xcl_ops
= &svc_udp_ops
,
733 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
736 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
738 int err
, level
, optname
, one
= 1;
740 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_udp_class
,
741 &svsk
->sk_xprt
, serv
);
742 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
743 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
744 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
746 /* initialise setting must have enough space to
747 * receive and respond to one request.
748 * svc_udp_recvfrom will re-adjust if necessary
750 svc_sock_setbufsize(svsk
->sk_sock
,
751 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
752 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
754 /* data might have come in before data_ready set up */
755 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
756 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
758 /* make sure we get destination address info */
759 switch (svsk
->sk_sk
->sk_family
) {
762 optname
= IP_PKTINFO
;
766 optname
= IPV6_RECVPKTINFO
;
771 err
= kernel_setsockopt(svsk
->sk_sock
, level
, optname
,
772 (char *)&one
, sizeof(one
));
773 dprintk("svc: kernel_setsockopt returned %d\n", err
);
777 * A data_ready event on a listening socket means there's a connection
778 * pending. Do not use state_change as a substitute for it.
780 static void svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
782 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
783 wait_queue_head_t
*wq
;
785 dprintk("svc: socket %p TCP (listen) state change %d\n",
789 * This callback may called twice when a new connection
790 * is established as a child socket inherits everything
791 * from a parent LISTEN socket.
792 * 1) data_ready method of the parent socket will be called
793 * when one of child sockets become ESTABLISHED.
794 * 2) data_ready method of the child socket may be called
795 * when it receives data before the socket is accepted.
796 * In case of 2, we should ignore it silently.
798 if (sk
->sk_state
== TCP_LISTEN
) {
800 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
801 svc_xprt_enqueue(&svsk
->sk_xprt
);
803 printk("svc: socket %p: no user data\n", sk
);
807 if (wq
&& waitqueue_active(wq
))
808 wake_up_interruptible_all(wq
);
812 * A state change on a connected socket means it's dying or dead.
814 static void svc_tcp_state_change(struct sock
*sk
)
816 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
817 wait_queue_head_t
*wq
= sk_sleep(sk
);
819 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
820 sk
, sk
->sk_state
, sk
->sk_user_data
);
823 printk("svc: socket %p: no user data\n", sk
);
825 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
826 svc_xprt_enqueue(&svsk
->sk_xprt
);
828 if (wq
&& waitqueue_active(wq
))
829 wake_up_interruptible_all(wq
);
832 static void svc_tcp_data_ready(struct sock
*sk
, int count
)
834 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
835 wait_queue_head_t
*wq
= sk_sleep(sk
);
837 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
838 sk
, sk
->sk_user_data
);
840 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
841 svc_xprt_enqueue(&svsk
->sk_xprt
);
843 if (wq
&& waitqueue_active(wq
))
844 wake_up_interruptible(wq
);
848 * Accept a TCP connection
850 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
852 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
853 struct sockaddr_storage addr
;
854 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
855 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
856 struct socket
*sock
= svsk
->sk_sock
;
857 struct socket
*newsock
;
858 struct svc_sock
*newsvsk
;
860 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
862 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
866 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
867 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
870 printk(KERN_WARNING
"%s: no more sockets!\n",
872 else if (err
!= -EAGAIN
)
873 net_warn_ratelimited("%s: accept failed (err %d)!\n",
874 serv
->sv_name
, -err
);
877 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
879 err
= kernel_getpeername(newsock
, sin
, &slen
);
881 net_warn_ratelimited("%s: peername failed (err %d)!\n",
882 serv
->sv_name
, -err
);
883 goto failed
; /* aborted connection or whatever */
886 /* Ideally, we would want to reject connections from unauthorized
887 * hosts here, but when we get encryption, the IP of the host won't
888 * tell us anything. For now just warn about unpriv connections.
890 if (!svc_port_is_privileged(sin
)) {
892 "%s: connect from unprivileged port: %s\n",
894 __svc_print_addr(sin
, buf
, sizeof(buf
)));
896 dprintk("%s: connect from %s\n", serv
->sv_name
,
897 __svc_print_addr(sin
, buf
, sizeof(buf
)));
899 /* make sure that a write doesn't block forever when
902 newsock
->sk
->sk_sndtimeo
= HZ
*30;
904 newsvsk
= svc_setup_socket(serv
, newsock
,
905 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
));
908 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
909 err
= kernel_getsockname(newsock
, sin
, &slen
);
910 if (unlikely(err
< 0)) {
911 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
912 slen
= offsetof(struct sockaddr
, sa_data
);
914 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
917 serv
->sv_stats
->nettcpconn
++;
919 return &newsvsk
->sk_xprt
;
922 sock_release(newsock
);
926 static unsigned int svc_tcp_restore_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
928 unsigned int i
, len
, npages
;
930 if (svsk
->sk_tcplen
<= sizeof(rpc_fraghdr
))
932 len
= svsk
->sk_tcplen
- sizeof(rpc_fraghdr
);
933 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
934 for (i
= 0; i
< npages
; i
++) {
935 if (rqstp
->rq_pages
[i
] != NULL
)
936 put_page(rqstp
->rq_pages
[i
]);
937 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
938 rqstp
->rq_pages
[i
] = svsk
->sk_pages
[i
];
939 svsk
->sk_pages
[i
] = NULL
;
941 rqstp
->rq_arg
.head
[0].iov_base
= page_address(rqstp
->rq_pages
[0]);
945 static void svc_tcp_save_pages(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
947 unsigned int i
, len
, npages
;
949 if (svsk
->sk_tcplen
<= sizeof(rpc_fraghdr
))
951 len
= svsk
->sk_tcplen
- sizeof(rpc_fraghdr
);
952 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
953 for (i
= 0; i
< npages
; i
++) {
954 svsk
->sk_pages
[i
] = rqstp
->rq_pages
[i
];
955 rqstp
->rq_pages
[i
] = NULL
;
959 static void svc_tcp_clear_pages(struct svc_sock
*svsk
)
961 unsigned int i
, len
, npages
;
963 if (svsk
->sk_tcplen
<= sizeof(rpc_fraghdr
))
965 len
= svsk
->sk_tcplen
- sizeof(rpc_fraghdr
);
966 npages
= (len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
967 for (i
= 0; i
< npages
; i
++) {
968 BUG_ON(svsk
->sk_pages
[i
] == NULL
);
969 put_page(svsk
->sk_pages
[i
]);
970 svsk
->sk_pages
[i
] = NULL
;
978 * If we haven't gotten the record length yet, get the next four bytes.
979 * Otherwise try to gobble up as much as possible up to the complete
982 static int svc_tcp_recv_record(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
984 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
988 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
990 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
993 want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
994 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
996 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
998 svsk
->sk_tcplen
+= len
;
1001 dprintk("svc: short recvfrom while reading record "
1002 "length (%d of %d)\n", len
, want
);
1006 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
1007 if (!(svsk
->sk_reclen
& RPC_LAST_STREAM_FRAGMENT
)) {
1008 /* FIXME: technically, a record can be fragmented,
1009 * and non-terminal fragments will not have the top
1010 * bit set in the fragment length header.
1011 * But apparently no known nfs clients send fragmented
1013 net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
1017 svsk
->sk_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
1018 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
1019 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
1020 net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
1021 (unsigned long)svsk
->sk_reclen
);
1026 if (svsk
->sk_reclen
< 8)
1027 goto err_delete
; /* client is nuts. */
1029 len
= svsk
->sk_reclen
;
1033 dprintk("RPC: TCP recv_record got %d\n", len
);
1036 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1040 static int receive_cb_reply(struct svc_sock
*svsk
, struct svc_rqst
*rqstp
)
1042 struct rpc_xprt
*bc_xprt
= svsk
->sk_xprt
.xpt_bc_xprt
;
1043 struct rpc_rqst
*req
= NULL
;
1044 struct kvec
*src
, *dst
;
1045 __be32
*p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1053 req
= xprt_lookup_rqst(bc_xprt
, xid
);
1057 "%s: Got unrecognized reply: "
1058 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1059 __func__
, ntohl(calldir
),
1064 memcpy(&req
->rq_private_buf
, &req
->rq_rcv_buf
, sizeof(struct xdr_buf
));
1066 * XXX!: cheating for now! Only copying HEAD.
1067 * But we know this is good enough for now (in fact, for any
1068 * callback reply in the forseeable future).
1070 dst
= &req
->rq_private_buf
.head
[0];
1071 src
= &rqstp
->rq_arg
.head
[0];
1072 if (dst
->iov_len
< src
->iov_len
)
1073 return -EAGAIN
; /* whatever; just giving up. */
1074 memcpy(dst
->iov_base
, src
->iov_base
, src
->iov_len
);
1075 xprt_complete_rqst(req
->rq_task
, svsk
->sk_reclen
);
1076 rqstp
->rq_arg
.len
= 0;
1080 static int copy_pages_to_kvecs(struct kvec
*vec
, struct page
**pages
, int len
)
1086 vec
[i
].iov_base
= page_address(pages
[i
]);
1087 vec
[i
].iov_len
= PAGE_SIZE
;
1096 * Receive data from a TCP socket.
1098 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
1100 struct svc_sock
*svsk
=
1101 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
1102 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1105 unsigned int want
, base
;
1110 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1111 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
1112 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
1113 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
1115 len
= svc_tcp_recv_record(svsk
, rqstp
);
1119 base
= svc_tcp_restore_pages(svsk
, rqstp
);
1120 want
= svsk
->sk_reclen
- base
;
1122 vec
= rqstp
->rq_vec
;
1124 pnum
= copy_pages_to_kvecs(&vec
[0], &rqstp
->rq_pages
[0],
1127 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
1129 /* Now receive data */
1130 len
= svc_partial_recvfrom(rqstp
, vec
, pnum
, want
, base
);
1132 svsk
->sk_tcplen
+= len
;
1134 svc_tcp_save_pages(svsk
, rqstp
);
1135 if (len
< 0 && len
!= -EAGAIN
)
1137 dprintk("svc: incomplete TCP record (%d of %d)\n",
1138 svsk
->sk_tcplen
, svsk
->sk_reclen
);
1142 rqstp
->rq_arg
.len
= svsk
->sk_reclen
;
1143 rqstp
->rq_arg
.page_base
= 0;
1144 if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1145 rqstp
->rq_arg
.head
[0].iov_len
= rqstp
->rq_arg
.len
;
1146 rqstp
->rq_arg
.page_len
= 0;
1148 rqstp
->rq_arg
.page_len
= rqstp
->rq_arg
.len
- rqstp
->rq_arg
.head
[0].iov_len
;
1150 rqstp
->rq_xprt_ctxt
= NULL
;
1151 rqstp
->rq_prot
= IPPROTO_TCP
;
1153 p
= (__be32
*)rqstp
->rq_arg
.head
[0].iov_base
;
1156 len
= receive_cb_reply(svsk
, rqstp
);
1158 /* Reset TCP read info */
1159 svsk
->sk_reclen
= 0;
1160 svsk
->sk_tcplen
= 0;
1161 /* If we have more data, signal svc_xprt_enqueue() to try again */
1162 if (svc_recv_available(svsk
) > sizeof(rpc_fraghdr
))
1163 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1168 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
1170 serv
->sv_stats
->nettcpcnt
++;
1172 dprintk("svc: TCP complete record (%d bytes)\n", rqstp
->rq_arg
.len
);
1173 return rqstp
->rq_arg
.len
;
1178 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1181 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
1182 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
1183 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1185 return -EAGAIN
; /* record not complete */
1189 * Send out data on TCP socket.
1191 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
1193 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
1197 /* Set up the first element of the reply kvec.
1198 * Any other kvecs that may be in use have been taken
1199 * care of by the server implementation itself.
1201 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
1202 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
1204 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
1205 if (sent
!= xbufp
->len
) {
1207 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1208 "- shutting down socket\n",
1209 rqstp
->rq_xprt
->xpt_server
->sv_name
,
1210 (sent
<0)?"got error":"sent only",
1212 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
1213 svc_xprt_enqueue(rqstp
->rq_xprt
);
1220 * Setup response header. TCP has a 4B record length field.
1222 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
1224 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1226 /* tcp needs a space for the record length... */
1230 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
1232 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1233 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
1236 if (test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
1238 required
= atomic_read(&xprt
->xpt_reserved
) + serv
->sv_max_mesg
;
1239 if (sk_stream_wspace(svsk
->sk_sk
) >= required
)
1241 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
1245 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
1247 struct sockaddr
*sa
, int salen
,
1250 return svc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1253 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1254 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*, int,
1255 struct net
*, struct sockaddr
*,
1257 static void svc_bc_sock_free(struct svc_xprt
*xprt
);
1259 static struct svc_xprt
*svc_bc_tcp_create(struct svc_serv
*serv
,
1261 struct sockaddr
*sa
, int salen
,
1264 return svc_bc_create_socket(serv
, IPPROTO_TCP
, net
, sa
, salen
, flags
);
1267 static void svc_bc_tcp_sock_detach(struct svc_xprt
*xprt
)
1271 static struct svc_xprt_ops svc_tcp_bc_ops
= {
1272 .xpo_create
= svc_bc_tcp_create
,
1273 .xpo_detach
= svc_bc_tcp_sock_detach
,
1274 .xpo_free
= svc_bc_sock_free
,
1275 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1278 static struct svc_xprt_class svc_tcp_bc_class
= {
1279 .xcl_name
= "tcp-bc",
1280 .xcl_owner
= THIS_MODULE
,
1281 .xcl_ops
= &svc_tcp_bc_ops
,
1282 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1285 static void svc_init_bc_xprt_sock(void)
1287 svc_reg_xprt_class(&svc_tcp_bc_class
);
1290 static void svc_cleanup_bc_xprt_sock(void)
1292 svc_unreg_xprt_class(&svc_tcp_bc_class
);
1294 #else /* CONFIG_SUNRPC_BACKCHANNEL */
1295 static void svc_init_bc_xprt_sock(void)
1299 static void svc_cleanup_bc_xprt_sock(void)
1302 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1304 static struct svc_xprt_ops svc_tcp_ops
= {
1305 .xpo_create
= svc_tcp_create
,
1306 .xpo_recvfrom
= svc_tcp_recvfrom
,
1307 .xpo_sendto
= svc_tcp_sendto
,
1308 .xpo_release_rqst
= svc_release_skb
,
1309 .xpo_detach
= svc_tcp_sock_detach
,
1310 .xpo_free
= svc_sock_free
,
1311 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1312 .xpo_has_wspace
= svc_tcp_has_wspace
,
1313 .xpo_accept
= svc_tcp_accept
,
1316 static struct svc_xprt_class svc_tcp_class
= {
1318 .xcl_owner
= THIS_MODULE
,
1319 .xcl_ops
= &svc_tcp_ops
,
1320 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1323 void svc_init_xprt_sock(void)
1325 svc_reg_xprt_class(&svc_tcp_class
);
1326 svc_reg_xprt_class(&svc_udp_class
);
1327 svc_init_bc_xprt_sock();
1330 void svc_cleanup_xprt_sock(void)
1332 svc_unreg_xprt_class(&svc_tcp_class
);
1333 svc_unreg_xprt_class(&svc_udp_class
);
1334 svc_cleanup_bc_xprt_sock();
1337 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1339 struct sock
*sk
= svsk
->sk_sk
;
1341 svc_xprt_init(sock_net(svsk
->sk_sock
->sk
), &svc_tcp_class
,
1342 &svsk
->sk_xprt
, serv
);
1343 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1344 if (sk
->sk_state
== TCP_LISTEN
) {
1345 dprintk("setting up TCP socket for listening\n");
1346 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1347 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1348 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1350 dprintk("setting up TCP socket for reading\n");
1351 sk
->sk_state_change
= svc_tcp_state_change
;
1352 sk
->sk_data_ready
= svc_tcp_data_ready
;
1353 sk
->sk_write_space
= svc_tcp_write_space
;
1355 svsk
->sk_reclen
= 0;
1356 svsk
->sk_tcplen
= 0;
1357 memset(&svsk
->sk_pages
[0], 0, sizeof(svsk
->sk_pages
));
1359 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1361 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1362 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1363 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1367 void svc_sock_update_bufs(struct svc_serv
*serv
)
1370 * The number of server threads has changed. Update
1371 * rcvbuf and sndbuf accordingly on all sockets
1373 struct svc_sock
*svsk
;
1375 spin_lock_bh(&serv
->sv_lock
);
1376 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
)
1377 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1378 spin_unlock_bh(&serv
->sv_lock
);
1380 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1383 * Initialize socket for RPC use and create svc_sock struct
1384 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1386 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1387 struct socket
*sock
,
1390 struct svc_sock
*svsk
;
1392 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1395 dprintk("svc: svc_setup_socket %p\n", sock
);
1396 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1398 return ERR_PTR(-ENOMEM
);
1402 /* Register socket with portmapper */
1404 err
= svc_register(serv
, sock_net(sock
->sk
), inet
->sk_family
,
1406 ntohs(inet_sk(inet
)->inet_sport
));
1410 return ERR_PTR(err
);
1413 inet
->sk_user_data
= svsk
;
1414 svsk
->sk_sock
= sock
;
1416 svsk
->sk_ostate
= inet
->sk_state_change
;
1417 svsk
->sk_odata
= inet
->sk_data_ready
;
1418 svsk
->sk_owspace
= inet
->sk_write_space
;
1420 /* Initialize the socket */
1421 if (sock
->type
== SOCK_DGRAM
)
1422 svc_udp_init(svsk
, serv
);
1424 /* initialise setting must have enough space to
1425 * receive and respond to one request.
1427 svc_sock_setbufsize(svsk
->sk_sock
, 4 * serv
->sv_max_mesg
,
1428 4 * serv
->sv_max_mesg
);
1429 svc_tcp_init(svsk
, serv
);
1432 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1439 * svc_addsock - add a listener socket to an RPC service
1440 * @serv: pointer to RPC service to which to add a new listener
1441 * @fd: file descriptor of the new listener
1442 * @name_return: pointer to buffer to fill in with name of listener
1443 * @len: size of the buffer
1445 * Fills in socket name and returns positive length of name if successful.
1446 * Name is terminated with '\n'. On error, returns a negative errno
1449 int svc_addsock(struct svc_serv
*serv
, const int fd
, char *name_return
,
1453 struct socket
*so
= sockfd_lookup(fd
, &err
);
1454 struct svc_sock
*svsk
= NULL
;
1455 struct sockaddr_storage addr
;
1456 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1461 err
= -EAFNOSUPPORT
;
1462 if ((so
->sk
->sk_family
!= PF_INET
) && (so
->sk
->sk_family
!= PF_INET6
))
1464 err
= -EPROTONOSUPPORT
;
1465 if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1466 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1469 if (so
->state
> SS_UNCONNECTED
)
1472 if (!try_module_get(THIS_MODULE
))
1474 svsk
= svc_setup_socket(serv
, so
, SVC_SOCK_DEFAULTS
);
1476 module_put(THIS_MODULE
);
1477 err
= PTR_ERR(svsk
);
1480 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1481 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1482 svc_add_new_perm_xprt(serv
, &svsk
->sk_xprt
);
1483 return svc_one_sock_name(svsk
, name_return
, len
);
1488 EXPORT_SYMBOL_GPL(svc_addsock
);
1491 * Create socket for RPC service.
1493 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1496 struct sockaddr
*sin
, int len
,
1499 struct svc_sock
*svsk
;
1500 struct socket
*sock
;
1503 struct sockaddr_storage addr
;
1504 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1508 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1510 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1511 serv
->sv_program
->pg_name
, protocol
,
1512 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1514 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1515 printk(KERN_WARNING
"svc: only UDP and TCP "
1516 "sockets supported\n");
1517 return ERR_PTR(-EINVAL
);
1520 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1521 switch (sin
->sa_family
) {
1529 return ERR_PTR(-EINVAL
);
1532 error
= __sock_create(net
, family
, type
, protocol
, &sock
, 1);
1534 return ERR_PTR(error
);
1536 svc_reclassify_socket(sock
);
1539 * If this is an PF_INET6 listener, we want to avoid
1540 * getting requests from IPv4 remotes. Those should
1541 * be shunted to a PF_INET listener via rpcbind.
1544 if (family
== PF_INET6
)
1545 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1546 (char *)&val
, sizeof(val
));
1548 if (type
== SOCK_STREAM
)
1549 sock
->sk
->sk_reuse
= SK_CAN_REUSE
; /* allow address reuse */
1550 error
= kernel_bind(sock
, sin
, len
);
1555 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1559 if (protocol
== IPPROTO_TCP
) {
1560 if ((error
= kernel_listen(sock
, 64)) < 0)
1564 svsk
= svc_setup_socket(serv
, sock
, flags
);
1566 error
= PTR_ERR(svsk
);
1569 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1570 return (struct svc_xprt
*)svsk
;
1572 dprintk("svc: svc_create_socket error = %d\n", -error
);
1574 return ERR_PTR(error
);
1578 * Detach the svc_sock from the socket so that no
1579 * more callbacks occur.
1581 static void svc_sock_detach(struct svc_xprt
*xprt
)
1583 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1584 struct sock
*sk
= svsk
->sk_sk
;
1585 wait_queue_head_t
*wq
;
1587 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1589 /* put back the old socket callbacks */
1590 sk
->sk_state_change
= svsk
->sk_ostate
;
1591 sk
->sk_data_ready
= svsk
->sk_odata
;
1592 sk
->sk_write_space
= svsk
->sk_owspace
;
1595 if (wq
&& waitqueue_active(wq
))
1596 wake_up_interruptible(wq
);
1600 * Disconnect the socket, and reset the callbacks
1602 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1604 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1606 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1608 svc_sock_detach(xprt
);
1610 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
)) {
1611 svc_tcp_clear_pages(svsk
);
1612 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1617 * Free the svc_sock's socket resources and the svc_sock itself.
1619 static void svc_sock_free(struct svc_xprt
*xprt
)
1621 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1622 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1624 if (svsk
->sk_sock
->file
)
1625 sockfd_put(svsk
->sk_sock
);
1627 sock_release(svsk
->sk_sock
);
1631 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1633 * Create a back channel svc_xprt which shares the fore channel socket.
1635 static struct svc_xprt
*svc_bc_create_socket(struct svc_serv
*serv
,
1638 struct sockaddr
*sin
, int len
,
1641 struct svc_sock
*svsk
;
1642 struct svc_xprt
*xprt
;
1644 if (protocol
!= IPPROTO_TCP
) {
1645 printk(KERN_WARNING
"svc: only TCP sockets"
1646 " supported on shared back channel\n");
1647 return ERR_PTR(-EINVAL
);
1650 svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
);
1652 return ERR_PTR(-ENOMEM
);
1654 xprt
= &svsk
->sk_xprt
;
1655 svc_xprt_init(net
, &svc_tcp_bc_class
, xprt
, serv
);
1657 serv
->sv_bc_xprt
= xprt
;
1663 * Free a back channel svc_sock.
1665 static void svc_bc_sock_free(struct svc_xprt
*xprt
)
1668 kfree(container_of(xprt
, struct svc_sock
, sk_xprt
));
1670 #endif /* CONFIG_SUNRPC_BACKCHANNEL */