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/errno.h>
25 #include <linux/fcntl.h>
26 #include <linux/net.h>
28 #include <linux/inet.h>
29 #include <linux/udp.h>
30 #include <linux/tcp.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/netdevice.h>
34 #include <linux/skbuff.h>
35 #include <linux/file.h>
36 #include <linux/freezer.h>
38 #include <net/checksum.h>
42 #include <net/tcp_states.h>
43 #include <asm/uaccess.h>
44 #include <asm/ioctls.h>
46 #include <linux/sunrpc/types.h>
47 #include <linux/sunrpc/clnt.h>
48 #include <linux/sunrpc/xdr.h>
49 #include <linux/sunrpc/msg_prot.h>
50 #include <linux/sunrpc/svcsock.h>
51 #include <linux/sunrpc/stats.h>
53 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
56 static struct svc_sock
*svc_setup_socket(struct svc_serv
*, struct socket
*,
57 int *errp
, int flags
);
58 static void svc_udp_data_ready(struct sock
*, int);
59 static int svc_udp_recvfrom(struct svc_rqst
*);
60 static int svc_udp_sendto(struct svc_rqst
*);
61 static void svc_sock_detach(struct svc_xprt
*);
62 static void svc_tcp_sock_detach(struct svc_xprt
*);
63 static void svc_sock_free(struct svc_xprt
*);
65 static struct svc_xprt
*svc_create_socket(struct svc_serv
*, int,
66 struct sockaddr
*, int, int);
67 #ifdef CONFIG_DEBUG_LOCK_ALLOC
68 static struct lock_class_key svc_key
[2];
69 static struct lock_class_key svc_slock_key
[2];
71 static void svc_reclassify_socket(struct socket
*sock
)
73 struct sock
*sk
= sock
->sk
;
74 BUG_ON(sock_owned_by_user(sk
));
75 switch (sk
->sk_family
) {
77 sock_lock_init_class_and_name(sk
, "slock-AF_INET-NFSD",
79 "sk_xprt.xpt_lock-AF_INET-NFSD",
84 sock_lock_init_class_and_name(sk
, "slock-AF_INET6-NFSD",
86 "sk_xprt.xpt_lock-AF_INET6-NFSD",
95 static void svc_reclassify_socket(struct socket
*sock
)
101 * Release an skbuff after use
103 static void svc_release_skb(struct svc_rqst
*rqstp
)
105 struct sk_buff
*skb
= rqstp
->rq_xprt_ctxt
;
108 struct svc_sock
*svsk
=
109 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
110 rqstp
->rq_xprt_ctxt
= NULL
;
112 dprintk("svc: service %p, releasing skb %p\n", rqstp
, skb
);
113 skb_free_datagram(svsk
->sk_sk
, skb
);
117 union svc_pktinfo_u
{
118 struct in_pktinfo pkti
;
119 struct in6_pktinfo pkti6
;
121 #define SVC_PKTINFO_SPACE \
122 CMSG_SPACE(sizeof(union svc_pktinfo_u))
124 static void svc_set_cmsg_data(struct svc_rqst
*rqstp
, struct cmsghdr
*cmh
)
126 struct svc_sock
*svsk
=
127 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
128 switch (svsk
->sk_sk
->sk_family
) {
130 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
132 cmh
->cmsg_level
= SOL_IP
;
133 cmh
->cmsg_type
= IP_PKTINFO
;
134 pki
->ipi_ifindex
= 0;
135 pki
->ipi_spec_dst
.s_addr
= rqstp
->rq_daddr
.addr
.s_addr
;
136 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
141 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
143 cmh
->cmsg_level
= SOL_IPV6
;
144 cmh
->cmsg_type
= IPV6_PKTINFO
;
145 pki
->ipi6_ifindex
= 0;
146 ipv6_addr_copy(&pki
->ipi6_addr
,
147 &rqstp
->rq_daddr
.addr6
);
148 cmh
->cmsg_len
= CMSG_LEN(sizeof(*pki
));
156 * Generic sendto routine
158 static int svc_sendto(struct svc_rqst
*rqstp
, struct xdr_buf
*xdr
)
160 struct svc_sock
*svsk
=
161 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
162 struct socket
*sock
= svsk
->sk_sock
;
166 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
168 struct cmsghdr
*cmh
= &buffer
.hdr
;
172 struct page
**ppage
= xdr
->pages
;
173 size_t base
= xdr
->page_base
;
174 unsigned int pglen
= xdr
->page_len
;
175 unsigned int flags
= MSG_MORE
;
176 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
180 if (rqstp
->rq_prot
== IPPROTO_UDP
) {
181 struct msghdr msg
= {
182 .msg_name
= &rqstp
->rq_addr
,
183 .msg_namelen
= rqstp
->rq_addrlen
,
185 .msg_controllen
= sizeof(buffer
),
186 .msg_flags
= MSG_MORE
,
189 svc_set_cmsg_data(rqstp
, cmh
);
191 if (sock_sendmsg(sock
, &msg
, 0) < 0)
196 if (slen
== xdr
->head
[0].iov_len
)
198 len
= kernel_sendpage(sock
, rqstp
->rq_respages
[0], 0,
199 xdr
->head
[0].iov_len
, flags
);
200 if (len
!= xdr
->head
[0].iov_len
)
202 slen
-= xdr
->head
[0].iov_len
;
207 size
= PAGE_SIZE
- base
< pglen
? PAGE_SIZE
- base
: pglen
;
211 result
= kernel_sendpage(sock
, *ppage
, base
, size
, flags
);
218 size
= PAGE_SIZE
< pglen
? PAGE_SIZE
: pglen
;
223 if (xdr
->tail
[0].iov_len
) {
224 result
= kernel_sendpage(sock
, rqstp
->rq_respages
[0],
225 ((unsigned long)xdr
->tail
[0].iov_base
)
227 xdr
->tail
[0].iov_len
, 0);
233 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
234 svsk
, xdr
->head
[0].iov_base
, xdr
->head
[0].iov_len
,
235 xdr
->len
, len
, svc_print_addr(rqstp
, buf
, sizeof(buf
)));
241 * Report socket names for nfsdfs
243 static int one_sock_name(char *buf
, struct svc_sock
*svsk
)
247 switch(svsk
->sk_sk
->sk_family
) {
249 len
= sprintf(buf
, "ipv4 %s %pI4 %d\n",
250 svsk
->sk_sk
->sk_protocol
== IPPROTO_UDP
?
252 &inet_sk(svsk
->sk_sk
)->rcv_saddr
,
253 inet_sk(svsk
->sk_sk
)->num
);
256 len
= sprintf(buf
, "*unknown-%d*\n",
257 svsk
->sk_sk
->sk_family
);
263 svc_sock_names(char *buf
, struct svc_serv
*serv
, char *toclose
)
265 struct svc_sock
*svsk
, *closesk
= NULL
;
270 spin_lock_bh(&serv
->sv_lock
);
271 list_for_each_entry(svsk
, &serv
->sv_permsocks
, sk_xprt
.xpt_list
) {
272 int onelen
= one_sock_name(buf
+len
, svsk
);
273 if (toclose
&& strcmp(toclose
, buf
+len
) == 0)
278 spin_unlock_bh(&serv
->sv_lock
);
280 /* Should unregister with portmap, but you cannot
281 * unregister just one protocol...
283 svc_close_xprt(&closesk
->sk_xprt
);
288 EXPORT_SYMBOL_GPL(svc_sock_names
);
291 * Check input queue length
293 static int svc_recv_available(struct svc_sock
*svsk
)
295 struct socket
*sock
= svsk
->sk_sock
;
298 err
= kernel_sock_ioctl(sock
, TIOCINQ
, (unsigned long) &avail
);
300 return (err
>= 0)? avail
: err
;
304 * Generic recvfrom routine.
306 static int svc_recvfrom(struct svc_rqst
*rqstp
, struct kvec
*iov
, int nr
,
309 struct svc_sock
*svsk
=
310 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
311 struct msghdr msg
= {
312 .msg_flags
= MSG_DONTWAIT
,
316 rqstp
->rq_xprt_hlen
= 0;
318 len
= kernel_recvmsg(svsk
->sk_sock
, &msg
, iov
, nr
, buflen
,
321 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
322 svsk
, iov
[0].iov_base
, iov
[0].iov_len
, len
);
327 * Set socket snd and rcv buffer lengths
329 static void svc_sock_setbufsize(struct socket
*sock
, unsigned int snd
,
334 oldfs
= get_fs(); set_fs(KERNEL_DS
);
335 sock_setsockopt(sock
, SOL_SOCKET
, SO_SNDBUF
,
336 (char*)&snd
, sizeof(snd
));
337 sock_setsockopt(sock
, SOL_SOCKET
, SO_RCVBUF
,
338 (char*)&rcv
, sizeof(rcv
));
340 /* sock_setsockopt limits use to sysctl_?mem_max,
341 * which isn't acceptable. Until that is made conditional
342 * on not having CAP_SYS_RESOURCE or similar, we go direct...
343 * DaveM said I could!
346 sock
->sk
->sk_sndbuf
= snd
* 2;
347 sock
->sk
->sk_rcvbuf
= rcv
* 2;
348 release_sock(sock
->sk
);
352 * INET callback when data has been received on the socket.
354 static void svc_udp_data_ready(struct sock
*sk
, int count
)
356 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
359 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
361 test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
362 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
363 svc_xprt_enqueue(&svsk
->sk_xprt
);
365 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
366 wake_up_interruptible(sk
->sk_sleep
);
370 * INET callback when space is newly available on the socket.
372 static void svc_write_space(struct sock
*sk
)
374 struct svc_sock
*svsk
= (struct svc_sock
*)(sk
->sk_user_data
);
377 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
378 svsk
, sk
, test_bit(XPT_BUSY
, &svsk
->sk_xprt
.xpt_flags
));
379 svc_xprt_enqueue(&svsk
->sk_xprt
);
382 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
)) {
383 dprintk("RPC svc_write_space: someone sleeping on %p\n",
385 wake_up_interruptible(sk
->sk_sleep
);
390 * Copy the UDP datagram's destination address to the rqstp structure.
391 * The 'destination' address in this case is the address to which the
392 * peer sent the datagram, i.e. our local address. For multihomed
393 * hosts, this can change from msg to msg. Note that only the IP
394 * address changes, the port number should remain the same.
396 static void svc_udp_get_dest_address(struct svc_rqst
*rqstp
,
399 struct svc_sock
*svsk
=
400 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
401 switch (svsk
->sk_sk
->sk_family
) {
403 struct in_pktinfo
*pki
= CMSG_DATA(cmh
);
404 rqstp
->rq_daddr
.addr
.s_addr
= pki
->ipi_spec_dst
.s_addr
;
408 struct in6_pktinfo
*pki
= CMSG_DATA(cmh
);
409 ipv6_addr_copy(&rqstp
->rq_daddr
.addr6
, &pki
->ipi6_addr
);
416 * Receive a datagram from a UDP socket.
418 static int svc_udp_recvfrom(struct svc_rqst
*rqstp
)
420 struct svc_sock
*svsk
=
421 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
422 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
426 long all
[SVC_PKTINFO_SPACE
/ sizeof(long)];
428 struct cmsghdr
*cmh
= &buffer
.hdr
;
429 struct msghdr msg
= {
430 .msg_name
= svc_addr(rqstp
),
432 .msg_controllen
= sizeof(buffer
),
433 .msg_flags
= MSG_DONTWAIT
,
438 if (test_and_clear_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
))
439 /* udp sockets need large rcvbuf as all pending
440 * requests are still in that buffer. sndbuf must
441 * also be large enough that there is enough space
442 * for one reply per thread. We count all threads
443 * rather than threads in a particular pool, which
444 * provides an upper bound on the number of threads
445 * which will access the socket.
447 svc_sock_setbufsize(svsk
->sk_sock
,
448 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
,
449 (serv
->sv_nrthreads
+3) * serv
->sv_max_mesg
);
451 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
453 err
= kernel_recvmsg(svsk
->sk_sock
, &msg
, NULL
,
454 0, 0, MSG_PEEK
| MSG_DONTWAIT
);
456 skb
= skb_recv_datagram(svsk
->sk_sk
, 0, 1, &err
);
459 if (err
!= -EAGAIN
) {
460 /* possibly an icmp error */
461 dprintk("svc: recvfrom returned error %d\n", -err
);
462 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
464 svc_xprt_received(&svsk
->sk_xprt
);
467 len
= svc_addr_len(svc_addr(rqstp
));
469 return -EAFNOSUPPORT
;
470 rqstp
->rq_addrlen
= len
;
471 if (skb
->tstamp
.tv64
== 0) {
472 skb
->tstamp
= ktime_get_real();
473 /* Don't enable netstamp, sunrpc doesn't
474 need that much accuracy */
476 svsk
->sk_sk
->sk_stamp
= skb
->tstamp
;
477 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
); /* there may be more data... */
480 * Maybe more packets - kick another thread ASAP.
482 svc_xprt_received(&svsk
->sk_xprt
);
484 len
= skb
->len
- sizeof(struct udphdr
);
485 rqstp
->rq_arg
.len
= len
;
487 rqstp
->rq_prot
= IPPROTO_UDP
;
489 if (cmh
->cmsg_level
!= IPPROTO_IP
||
490 cmh
->cmsg_type
!= IP_PKTINFO
) {
492 printk("rpcsvc: received unknown control message:"
494 cmh
->cmsg_level
, cmh
->cmsg_type
);
495 skb_free_datagram(svsk
->sk_sk
, skb
);
498 svc_udp_get_dest_address(rqstp
, cmh
);
500 if (skb_is_nonlinear(skb
)) {
501 /* we have to copy */
503 if (csum_partial_copy_to_xdr(&rqstp
->rq_arg
, skb
)) {
506 skb_free_datagram(svsk
->sk_sk
, skb
);
510 skb_free_datagram(svsk
->sk_sk
, skb
);
512 /* we can use it in-place */
513 rqstp
->rq_arg
.head
[0].iov_base
= skb
->data
+
514 sizeof(struct udphdr
);
515 rqstp
->rq_arg
.head
[0].iov_len
= len
;
516 if (skb_checksum_complete(skb
)) {
517 skb_free_datagram(svsk
->sk_sk
, skb
);
520 rqstp
->rq_xprt_ctxt
= skb
;
523 rqstp
->rq_arg
.page_base
= 0;
524 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
525 rqstp
->rq_arg
.head
[0].iov_len
= len
;
526 rqstp
->rq_arg
.page_len
= 0;
527 rqstp
->rq_respages
= rqstp
->rq_pages
+1;
529 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
530 rqstp
->rq_respages
= rqstp
->rq_pages
+ 1 +
531 DIV_ROUND_UP(rqstp
->rq_arg
.page_len
, PAGE_SIZE
);
535 serv
->sv_stats
->netudpcnt
++;
541 svc_udp_sendto(struct svc_rqst
*rqstp
)
545 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
546 if (error
== -ECONNREFUSED
)
547 /* ICMP error on earlier request. */
548 error
= svc_sendto(rqstp
, &rqstp
->rq_res
);
553 static void svc_udp_prep_reply_hdr(struct svc_rqst
*rqstp
)
557 static int svc_udp_has_wspace(struct svc_xprt
*xprt
)
559 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
560 struct svc_serv
*serv
= xprt
->xpt_server
;
561 unsigned long required
;
564 * Set the SOCK_NOSPACE flag before checking the available
567 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
568 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
569 if (required
*2 > sock_wspace(svsk
->sk_sk
))
571 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
575 static struct svc_xprt
*svc_udp_accept(struct svc_xprt
*xprt
)
581 static struct svc_xprt
*svc_udp_create(struct svc_serv
*serv
,
582 struct sockaddr
*sa
, int salen
,
585 return svc_create_socket(serv
, IPPROTO_UDP
, sa
, salen
, flags
);
588 static struct svc_xprt_ops svc_udp_ops
= {
589 .xpo_create
= svc_udp_create
,
590 .xpo_recvfrom
= svc_udp_recvfrom
,
591 .xpo_sendto
= svc_udp_sendto
,
592 .xpo_release_rqst
= svc_release_skb
,
593 .xpo_detach
= svc_sock_detach
,
594 .xpo_free
= svc_sock_free
,
595 .xpo_prep_reply_hdr
= svc_udp_prep_reply_hdr
,
596 .xpo_has_wspace
= svc_udp_has_wspace
,
597 .xpo_accept
= svc_udp_accept
,
600 static struct svc_xprt_class svc_udp_class
= {
602 .xcl_owner
= THIS_MODULE
,
603 .xcl_ops
= &svc_udp_ops
,
604 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_UDP
,
607 static void svc_udp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
612 svc_xprt_init(&svc_udp_class
, &svsk
->sk_xprt
, serv
);
613 clear_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
614 svsk
->sk_sk
->sk_data_ready
= svc_udp_data_ready
;
615 svsk
->sk_sk
->sk_write_space
= svc_write_space
;
617 /* initialise setting must have enough space to
618 * receive and respond to one request.
619 * svc_udp_recvfrom will re-adjust if necessary
621 svc_sock_setbufsize(svsk
->sk_sock
,
622 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
,
623 3 * svsk
->sk_xprt
.xpt_server
->sv_max_mesg
);
625 /* data might have come in before data_ready set up */
626 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
627 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
631 /* make sure we get destination address info */
632 svsk
->sk_sock
->ops
->setsockopt(svsk
->sk_sock
, IPPROTO_IP
, IP_PKTINFO
,
633 (char __user
*)&one
, sizeof(one
));
638 * A data_ready event on a listening socket means there's a connection
639 * pending. Do not use state_change as a substitute for it.
641 static void svc_tcp_listen_data_ready(struct sock
*sk
, int count_unused
)
643 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
645 dprintk("svc: socket %p TCP (listen) state change %d\n",
649 * This callback may called twice when a new connection
650 * is established as a child socket inherits everything
651 * from a parent LISTEN socket.
652 * 1) data_ready method of the parent socket will be called
653 * when one of child sockets become ESTABLISHED.
654 * 2) data_ready method of the child socket may be called
655 * when it receives data before the socket is accepted.
656 * In case of 2, we should ignore it silently.
658 if (sk
->sk_state
== TCP_LISTEN
) {
660 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
661 svc_xprt_enqueue(&svsk
->sk_xprt
);
663 printk("svc: socket %p: no user data\n", sk
);
666 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
667 wake_up_interruptible_all(sk
->sk_sleep
);
671 * A state change on a connected socket means it's dying or dead.
673 static void svc_tcp_state_change(struct sock
*sk
)
675 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
677 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
678 sk
, sk
->sk_state
, sk
->sk_user_data
);
681 printk("svc: socket %p: no user data\n", sk
);
683 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
684 svc_xprt_enqueue(&svsk
->sk_xprt
);
686 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
687 wake_up_interruptible_all(sk
->sk_sleep
);
690 static void svc_tcp_data_ready(struct sock
*sk
, int count
)
692 struct svc_sock
*svsk
= (struct svc_sock
*)sk
->sk_user_data
;
694 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
695 sk
, sk
->sk_user_data
);
697 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
698 svc_xprt_enqueue(&svsk
->sk_xprt
);
700 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
701 wake_up_interruptible(sk
->sk_sleep
);
705 * Accept a TCP connection
707 static struct svc_xprt
*svc_tcp_accept(struct svc_xprt
*xprt
)
709 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
710 struct sockaddr_storage addr
;
711 struct sockaddr
*sin
= (struct sockaddr
*) &addr
;
712 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
713 struct socket
*sock
= svsk
->sk_sock
;
714 struct socket
*newsock
;
715 struct svc_sock
*newsvsk
;
717 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
719 dprintk("svc: tcp_accept %p sock %p\n", svsk
, sock
);
723 clear_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
724 err
= kernel_accept(sock
, &newsock
, O_NONBLOCK
);
727 printk(KERN_WARNING
"%s: no more sockets!\n",
729 else if (err
!= -EAGAIN
&& net_ratelimit())
730 printk(KERN_WARNING
"%s: accept failed (err %d)!\n",
731 serv
->sv_name
, -err
);
734 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
736 err
= kernel_getpeername(newsock
, sin
, &slen
);
739 printk(KERN_WARNING
"%s: peername failed (err %d)!\n",
740 serv
->sv_name
, -err
);
741 goto failed
; /* aborted connection or whatever */
744 /* Ideally, we would want to reject connections from unauthorized
745 * hosts here, but when we get encryption, the IP of the host won't
746 * tell us anything. For now just warn about unpriv connections.
748 if (!svc_port_is_privileged(sin
)) {
750 "%s: connect from unprivileged port: %s\n",
752 __svc_print_addr(sin
, buf
, sizeof(buf
)));
754 dprintk("%s: connect from %s\n", serv
->sv_name
,
755 __svc_print_addr(sin
, buf
, sizeof(buf
)));
757 /* make sure that a write doesn't block forever when
760 newsock
->sk
->sk_sndtimeo
= HZ
*30;
762 if (!(newsvsk
= svc_setup_socket(serv
, newsock
, &err
,
763 (SVC_SOCK_ANONYMOUS
| SVC_SOCK_TEMPORARY
))))
765 svc_xprt_set_remote(&newsvsk
->sk_xprt
, sin
, slen
);
766 err
= kernel_getsockname(newsock
, sin
, &slen
);
767 if (unlikely(err
< 0)) {
768 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err
);
769 slen
= offsetof(struct sockaddr
, sa_data
);
771 svc_xprt_set_local(&newsvsk
->sk_xprt
, sin
, slen
);
774 serv
->sv_stats
->nettcpconn
++;
776 return &newsvsk
->sk_xprt
;
779 sock_release(newsock
);
784 * Receive data from a TCP socket.
786 static int svc_tcp_recvfrom(struct svc_rqst
*rqstp
)
788 struct svc_sock
*svsk
=
789 container_of(rqstp
->rq_xprt
, struct svc_sock
, sk_xprt
);
790 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
795 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
796 svsk
, test_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
),
797 test_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
),
798 test_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
));
800 clear_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
802 /* Receive data. If we haven't got the record length yet, get
803 * the next four bytes. Otherwise try to gobble up as much as
804 * possible up to the complete record length.
806 if (svsk
->sk_tcplen
< sizeof(rpc_fraghdr
)) {
807 int want
= sizeof(rpc_fraghdr
) - svsk
->sk_tcplen
;
810 iov
.iov_base
= ((char *) &svsk
->sk_reclen
) + svsk
->sk_tcplen
;
812 if ((len
= svc_recvfrom(rqstp
, &iov
, 1, want
)) < 0)
814 svsk
->sk_tcplen
+= len
;
817 dprintk("svc: short recvfrom while reading record "
818 "length (%d of %d)\n", len
, want
);
819 svc_xprt_received(&svsk
->sk_xprt
);
820 return -EAGAIN
; /* record header not complete */
823 svsk
->sk_reclen
= ntohl(svsk
->sk_reclen
);
824 if (!(svsk
->sk_reclen
& RPC_LAST_STREAM_FRAGMENT
)) {
825 /* FIXME: technically, a record can be fragmented,
826 * and non-terminal fragments will not have the top
827 * bit set in the fragment length header.
828 * But apparently no known nfs clients send fragmented
831 printk(KERN_NOTICE
"RPC: multiple fragments "
832 "per record not supported\n");
835 svsk
->sk_reclen
&= RPC_FRAGMENT_SIZE_MASK
;
836 dprintk("svc: TCP record, %d bytes\n", svsk
->sk_reclen
);
837 if (svsk
->sk_reclen
> serv
->sv_max_mesg
) {
839 printk(KERN_NOTICE
"RPC: "
840 "fragment too large: 0x%08lx\n",
841 (unsigned long)svsk
->sk_reclen
);
846 /* Check whether enough data is available */
847 len
= svc_recv_available(svsk
);
851 if (len
< svsk
->sk_reclen
) {
852 dprintk("svc: incomplete TCP record (%d of %d)\n",
853 len
, svsk
->sk_reclen
);
854 svc_xprt_received(&svsk
->sk_xprt
);
855 return -EAGAIN
; /* record not complete */
857 len
= svsk
->sk_reclen
;
858 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
861 vec
[0] = rqstp
->rq_arg
.head
[0];
865 vec
[pnum
].iov_base
= page_address(rqstp
->rq_pages
[pnum
]);
866 vec
[pnum
].iov_len
= PAGE_SIZE
;
870 rqstp
->rq_respages
= &rqstp
->rq_pages
[pnum
];
872 /* Now receive data */
873 len
= svc_recvfrom(rqstp
, vec
, pnum
, len
);
877 dprintk("svc: TCP complete record (%d bytes)\n", len
);
878 rqstp
->rq_arg
.len
= len
;
879 rqstp
->rq_arg
.page_base
= 0;
880 if (len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
881 rqstp
->rq_arg
.head
[0].iov_len
= len
;
882 rqstp
->rq_arg
.page_len
= 0;
884 rqstp
->rq_arg
.page_len
= len
- rqstp
->rq_arg
.head
[0].iov_len
;
887 rqstp
->rq_xprt_ctxt
= NULL
;
888 rqstp
->rq_prot
= IPPROTO_TCP
;
890 /* Reset TCP read info */
894 svc_xprt_copy_addrs(rqstp
, &svsk
->sk_xprt
);
895 svc_xprt_received(&svsk
->sk_xprt
);
897 serv
->sv_stats
->nettcpcnt
++;
902 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
906 if (len
== -EAGAIN
) {
907 dprintk("RPC: TCP recvfrom got EAGAIN\n");
908 svc_xprt_received(&svsk
->sk_xprt
);
910 printk(KERN_NOTICE
"%s: recvfrom returned errno %d\n",
911 svsk
->sk_xprt
.xpt_server
->sv_name
, -len
);
919 * Send out data on TCP socket.
921 static int svc_tcp_sendto(struct svc_rqst
*rqstp
)
923 struct xdr_buf
*xbufp
= &rqstp
->rq_res
;
927 /* Set up the first element of the reply kvec.
928 * Any other kvecs that may be in use have been taken
929 * care of by the server implementation itself.
931 reclen
= htonl(0x80000000|((xbufp
->len
) - 4));
932 memcpy(xbufp
->head
[0].iov_base
, &reclen
, 4);
934 if (test_bit(XPT_DEAD
, &rqstp
->rq_xprt
->xpt_flags
))
937 sent
= svc_sendto(rqstp
, &rqstp
->rq_res
);
938 if (sent
!= xbufp
->len
) {
940 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
941 "- shutting down socket\n",
942 rqstp
->rq_xprt
->xpt_server
->sv_name
,
943 (sent
<0)?"got error":"sent only",
945 set_bit(XPT_CLOSE
, &rqstp
->rq_xprt
->xpt_flags
);
946 svc_xprt_enqueue(rqstp
->rq_xprt
);
953 * Setup response header. TCP has a 4B record length field.
955 static void svc_tcp_prep_reply_hdr(struct svc_rqst
*rqstp
)
957 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
959 /* tcp needs a space for the record length... */
963 static int svc_tcp_has_wspace(struct svc_xprt
*xprt
)
965 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
966 struct svc_serv
*serv
= svsk
->sk_xprt
.xpt_server
;
971 * Set the SOCK_NOSPACE flag before checking the available
974 set_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
975 required
= atomic_read(&svsk
->sk_xprt
.xpt_reserved
) + serv
->sv_max_mesg
;
976 wspace
= sk_stream_wspace(svsk
->sk_sk
);
978 if (wspace
< sk_stream_min_wspace(svsk
->sk_sk
))
980 if (required
* 2 > wspace
)
983 clear_bit(SOCK_NOSPACE
, &svsk
->sk_sock
->flags
);
987 static struct svc_xprt
*svc_tcp_create(struct svc_serv
*serv
,
988 struct sockaddr
*sa
, int salen
,
991 return svc_create_socket(serv
, IPPROTO_TCP
, sa
, salen
, flags
);
994 static struct svc_xprt_ops svc_tcp_ops
= {
995 .xpo_create
= svc_tcp_create
,
996 .xpo_recvfrom
= svc_tcp_recvfrom
,
997 .xpo_sendto
= svc_tcp_sendto
,
998 .xpo_release_rqst
= svc_release_skb
,
999 .xpo_detach
= svc_tcp_sock_detach
,
1000 .xpo_free
= svc_sock_free
,
1001 .xpo_prep_reply_hdr
= svc_tcp_prep_reply_hdr
,
1002 .xpo_has_wspace
= svc_tcp_has_wspace
,
1003 .xpo_accept
= svc_tcp_accept
,
1006 static struct svc_xprt_class svc_tcp_class
= {
1008 .xcl_owner
= THIS_MODULE
,
1009 .xcl_ops
= &svc_tcp_ops
,
1010 .xcl_max_payload
= RPCSVC_MAXPAYLOAD_TCP
,
1013 void svc_init_xprt_sock(void)
1015 svc_reg_xprt_class(&svc_tcp_class
);
1016 svc_reg_xprt_class(&svc_udp_class
);
1019 void svc_cleanup_xprt_sock(void)
1021 svc_unreg_xprt_class(&svc_tcp_class
);
1022 svc_unreg_xprt_class(&svc_udp_class
);
1025 static void svc_tcp_init(struct svc_sock
*svsk
, struct svc_serv
*serv
)
1027 struct sock
*sk
= svsk
->sk_sk
;
1029 svc_xprt_init(&svc_tcp_class
, &svsk
->sk_xprt
, serv
);
1030 set_bit(XPT_CACHE_AUTH
, &svsk
->sk_xprt
.xpt_flags
);
1031 if (sk
->sk_state
== TCP_LISTEN
) {
1032 dprintk("setting up TCP socket for listening\n");
1033 set_bit(XPT_LISTENER
, &svsk
->sk_xprt
.xpt_flags
);
1034 sk
->sk_data_ready
= svc_tcp_listen_data_ready
;
1035 set_bit(XPT_CONN
, &svsk
->sk_xprt
.xpt_flags
);
1037 dprintk("setting up TCP socket for reading\n");
1038 sk
->sk_state_change
= svc_tcp_state_change
;
1039 sk
->sk_data_ready
= svc_tcp_data_ready
;
1040 sk
->sk_write_space
= svc_write_space
;
1042 svsk
->sk_reclen
= 0;
1043 svsk
->sk_tcplen
= 0;
1045 tcp_sk(sk
)->nonagle
|= TCP_NAGLE_OFF
;
1047 set_bit(XPT_DATA
, &svsk
->sk_xprt
.xpt_flags
);
1048 if (sk
->sk_state
!= TCP_ESTABLISHED
)
1049 set_bit(XPT_CLOSE
, &svsk
->sk_xprt
.xpt_flags
);
1053 void svc_sock_update_bufs(struct svc_serv
*serv
)
1056 * The number of server threads has changed. Update
1057 * rcvbuf and sndbuf accordingly on all sockets
1059 struct list_head
*le
;
1061 spin_lock_bh(&serv
->sv_lock
);
1062 list_for_each(le
, &serv
->sv_permsocks
) {
1063 struct svc_sock
*svsk
=
1064 list_entry(le
, struct svc_sock
, sk_xprt
.xpt_list
);
1065 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1067 list_for_each(le
, &serv
->sv_tempsocks
) {
1068 struct svc_sock
*svsk
=
1069 list_entry(le
, struct svc_sock
, sk_xprt
.xpt_list
);
1070 set_bit(XPT_CHNGBUF
, &svsk
->sk_xprt
.xpt_flags
);
1072 spin_unlock_bh(&serv
->sv_lock
);
1074 EXPORT_SYMBOL_GPL(svc_sock_update_bufs
);
1077 * Initialize socket for RPC use and create svc_sock struct
1078 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1080 static struct svc_sock
*svc_setup_socket(struct svc_serv
*serv
,
1081 struct socket
*sock
,
1082 int *errp
, int flags
)
1084 struct svc_sock
*svsk
;
1086 int pmap_register
= !(flags
& SVC_SOCK_ANONYMOUS
);
1088 dprintk("svc: svc_setup_socket %p\n", sock
);
1089 if (!(svsk
= kzalloc(sizeof(*svsk
), GFP_KERNEL
))) {
1096 /* Register socket with portmapper */
1097 if (*errp
>= 0 && pmap_register
)
1098 *errp
= svc_register(serv
, inet
->sk_family
, inet
->sk_protocol
,
1099 ntohs(inet_sk(inet
)->sport
));
1106 inet
->sk_user_data
= svsk
;
1107 svsk
->sk_sock
= sock
;
1109 svsk
->sk_ostate
= inet
->sk_state_change
;
1110 svsk
->sk_odata
= inet
->sk_data_ready
;
1111 svsk
->sk_owspace
= inet
->sk_write_space
;
1113 /* Initialize the socket */
1114 if (sock
->type
== SOCK_DGRAM
)
1115 svc_udp_init(svsk
, serv
);
1117 /* initialise setting must have enough space to
1118 * receive and respond to one request.
1120 svc_sock_setbufsize(svsk
->sk_sock
, 4 * serv
->sv_max_mesg
,
1121 4 * serv
->sv_max_mesg
);
1122 svc_tcp_init(svsk
, serv
);
1125 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1131 int svc_addsock(struct svc_serv
*serv
,
1136 struct socket
*so
= sockfd_lookup(fd
, &err
);
1137 struct svc_sock
*svsk
= NULL
;
1141 if (so
->sk
->sk_family
!= AF_INET
)
1142 err
= -EAFNOSUPPORT
;
1143 else if (so
->sk
->sk_protocol
!= IPPROTO_TCP
&&
1144 so
->sk
->sk_protocol
!= IPPROTO_UDP
)
1145 err
= -EPROTONOSUPPORT
;
1146 else if (so
->state
> SS_UNCONNECTED
)
1149 if (!try_module_get(THIS_MODULE
))
1152 svsk
= svc_setup_socket(serv
, so
, &err
,
1155 struct sockaddr_storage addr
;
1156 struct sockaddr
*sin
= (struct sockaddr
*)&addr
;
1158 if (kernel_getsockname(svsk
->sk_sock
, sin
, &salen
) == 0)
1159 svc_xprt_set_local(&svsk
->sk_xprt
, sin
, salen
);
1160 clear_bit(XPT_TEMP
, &svsk
->sk_xprt
.xpt_flags
);
1161 spin_lock_bh(&serv
->sv_lock
);
1162 list_add(&svsk
->sk_xprt
.xpt_list
, &serv
->sv_permsocks
);
1163 spin_unlock_bh(&serv
->sv_lock
);
1164 svc_xprt_received(&svsk
->sk_xprt
);
1167 module_put(THIS_MODULE
);
1173 return one_sock_name(name_return
, svsk
);
1175 EXPORT_SYMBOL_GPL(svc_addsock
);
1178 * Create socket for RPC service.
1180 static struct svc_xprt
*svc_create_socket(struct svc_serv
*serv
,
1182 struct sockaddr
*sin
, int len
,
1185 struct svc_sock
*svsk
;
1186 struct socket
*sock
;
1189 struct sockaddr_storage addr
;
1190 struct sockaddr
*newsin
= (struct sockaddr
*)&addr
;
1194 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
1196 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1197 serv
->sv_program
->pg_name
, protocol
,
1198 __svc_print_addr(sin
, buf
, sizeof(buf
)));
1200 if (protocol
!= IPPROTO_UDP
&& protocol
!= IPPROTO_TCP
) {
1201 printk(KERN_WARNING
"svc: only UDP and TCP "
1202 "sockets supported\n");
1203 return ERR_PTR(-EINVAL
);
1206 type
= (protocol
== IPPROTO_UDP
)? SOCK_DGRAM
: SOCK_STREAM
;
1207 switch (sin
->sa_family
) {
1215 return ERR_PTR(-EINVAL
);
1218 error
= sock_create_kern(family
, type
, protocol
, &sock
);
1220 return ERR_PTR(error
);
1222 svc_reclassify_socket(sock
);
1225 * If this is an PF_INET6 listener, we want to avoid
1226 * getting requests from IPv4 remotes. Those should
1227 * be shunted to a PF_INET listener via rpcbind.
1230 if (family
== PF_INET6
)
1231 kernel_setsockopt(sock
, SOL_IPV6
, IPV6_V6ONLY
,
1232 (char *)&val
, sizeof(val
));
1234 if (type
== SOCK_STREAM
)
1235 sock
->sk
->sk_reuse
= 1; /* allow address reuse */
1236 error
= kernel_bind(sock
, sin
, len
);
1241 error
= kernel_getsockname(sock
, newsin
, &newlen
);
1245 if (protocol
== IPPROTO_TCP
) {
1246 if ((error
= kernel_listen(sock
, 64)) < 0)
1250 if ((svsk
= svc_setup_socket(serv
, sock
, &error
, flags
)) != NULL
) {
1251 svc_xprt_set_local(&svsk
->sk_xprt
, newsin
, newlen
);
1252 return (struct svc_xprt
*)svsk
;
1256 dprintk("svc: svc_create_socket error = %d\n", -error
);
1258 return ERR_PTR(error
);
1262 * Detach the svc_sock from the socket so that no
1263 * more callbacks occur.
1265 static void svc_sock_detach(struct svc_xprt
*xprt
)
1267 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1268 struct sock
*sk
= svsk
->sk_sk
;
1270 dprintk("svc: svc_sock_detach(%p)\n", svsk
);
1272 /* put back the old socket callbacks */
1273 sk
->sk_state_change
= svsk
->sk_ostate
;
1274 sk
->sk_data_ready
= svsk
->sk_odata
;
1275 sk
->sk_write_space
= svsk
->sk_owspace
;
1277 if (sk
->sk_sleep
&& waitqueue_active(sk
->sk_sleep
))
1278 wake_up_interruptible(sk
->sk_sleep
);
1282 * Disconnect the socket, and reset the callbacks
1284 static void svc_tcp_sock_detach(struct svc_xprt
*xprt
)
1286 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1288 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk
);
1290 svc_sock_detach(xprt
);
1292 if (!test_bit(XPT_LISTENER
, &xprt
->xpt_flags
))
1293 kernel_sock_shutdown(svsk
->sk_sock
, SHUT_RDWR
);
1297 * Free the svc_sock's socket resources and the svc_sock itself.
1299 static void svc_sock_free(struct svc_xprt
*xprt
)
1301 struct svc_sock
*svsk
= container_of(xprt
, struct svc_sock
, sk_xprt
);
1302 dprintk("svc: svc_sock_free(%p)\n", svsk
);
1304 if (svsk
->sk_sock
->file
)
1305 sockfd_put(svsk
->sk_sock
);
1307 sock_release(svsk
->sk_sock
);