Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svcsock.c
4 *
5 * These are the RPC server socket internals.
6 *
7 * The server scheduling algorithm does not always distribute the load
8 * evenly when servicing a single client. May need to modify the
9 * svc_xprt_enqueue procedure...
10 *
11 * TCP support is largely untested and may be a little slow. The problem
12 * is that we currently do two separate recvfrom's, one for the 4-byte
13 * record length, and the second for the actual record. This could possibly
14 * be improved by always reading a minimum size of around 100 bytes and
15 * tucking any superfluous bytes away in a temporary store. Still, that
16 * leaves write requests out in the rain. An alternative may be to peek at
17 * the first skb in the queue, and if it matches the next TCP sequence
18 * number, to extract the record marker. Yuck.
19 *
20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
21 */
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/module.h>
26#include <linux/errno.h>
27#include <linux/fcntl.h>
28#include <linux/net.h>
29#include <linux/in.h>
30#include <linux/inet.h>
31#include <linux/udp.h>
32#include <linux/tcp.h>
33#include <linux/unistd.h>
34#include <linux/slab.h>
35#include <linux/netdevice.h>
36#include <linux/skbuff.h>
37#include <linux/file.h>
38#include <linux/freezer.h>
39#include <linux/bvec.h>
40
41#include <net/sock.h>
42#include <net/checksum.h>
43#include <net/ip.h>
44#include <net/ipv6.h>
45#include <net/udp.h>
46#include <net/tcp.h>
47#include <net/tcp_states.h>
48#include <net/tls_prot.h>
49#include <net/handshake.h>
50#include <linux/uaccess.h>
51#include <linux/highmem.h>
52#include <asm/ioctls.h>
53#include <linux/key.h>
54
55#include <linux/sunrpc/types.h>
56#include <linux/sunrpc/clnt.h>
57#include <linux/sunrpc/xdr.h>
58#include <linux/sunrpc/msg_prot.h>
59#include <linux/sunrpc/svcsock.h>
60#include <linux/sunrpc/stats.h>
61#include <linux/sunrpc/xprt.h>
62
63#include <trace/events/sock.h>
64#include <trace/events/sunrpc.h>
65
66#include "socklib.h"
67#include "sunrpc.h"
68
69#define RPCDBG_FACILITY RPCDBG_SVCXPRT
70
71/* To-do: to avoid tying up an nfsd thread while waiting for a
72 * handshake request, the request could instead be deferred.
73 */
74enum {
75 SVC_HANDSHAKE_TO = 5U * HZ
76};
77
78static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
79 int flags);
80static int svc_udp_recvfrom(struct svc_rqst *);
81static int svc_udp_sendto(struct svc_rqst *);
82static void svc_sock_detach(struct svc_xprt *);
83static void svc_tcp_sock_detach(struct svc_xprt *);
84static void svc_sock_free(struct svc_xprt *);
85
86static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
87 struct net *, struct sockaddr *,
88 int, int);
89#ifdef CONFIG_DEBUG_LOCK_ALLOC
90static struct lock_class_key svc_key[2];
91static struct lock_class_key svc_slock_key[2];
92
93static void svc_reclassify_socket(struct socket *sock)
94{
95 struct sock *sk = sock->sk;
96
97 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
98 return;
99
100 switch (sk->sk_family) {
101 case AF_INET:
102 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
103 &svc_slock_key[0],
104 "sk_xprt.xpt_lock-AF_INET-NFSD",
105 &svc_key[0]);
106 break;
107
108 case AF_INET6:
109 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
110 &svc_slock_key[1],
111 "sk_xprt.xpt_lock-AF_INET6-NFSD",
112 &svc_key[1]);
113 break;
114
115 default:
116 BUG();
117 }
118}
119#else
120static void svc_reclassify_socket(struct socket *sock)
121{
122}
123#endif
124
125/**
126 * svc_tcp_release_ctxt - Release transport-related resources
127 * @xprt: the transport which owned the context
128 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
129 *
130 */
131static void svc_tcp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
132{
133}
134
135/**
136 * svc_udp_release_ctxt - Release transport-related resources
137 * @xprt: the transport which owned the context
138 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
139 *
140 */
141static void svc_udp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
142{
143 struct sk_buff *skb = ctxt;
144
145 if (skb)
146 consume_skb(skb);
147}
148
149union svc_pktinfo_u {
150 struct in_pktinfo pkti;
151 struct in6_pktinfo pkti6;
152};
153#define SVC_PKTINFO_SPACE \
154 CMSG_SPACE(sizeof(union svc_pktinfo_u))
155
156static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
157{
158 struct svc_sock *svsk =
159 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
160 switch (svsk->sk_sk->sk_family) {
161 case AF_INET: {
162 struct in_pktinfo *pki = CMSG_DATA(cmh);
163
164 cmh->cmsg_level = SOL_IP;
165 cmh->cmsg_type = IP_PKTINFO;
166 pki->ipi_ifindex = 0;
167 pki->ipi_spec_dst.s_addr =
168 svc_daddr_in(rqstp)->sin_addr.s_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
170 }
171 break;
172
173 case AF_INET6: {
174 struct in6_pktinfo *pki = CMSG_DATA(cmh);
175 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
176
177 cmh->cmsg_level = SOL_IPV6;
178 cmh->cmsg_type = IPV6_PKTINFO;
179 pki->ipi6_ifindex = daddr->sin6_scope_id;
180 pki->ipi6_addr = daddr->sin6_addr;
181 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
182 }
183 break;
184 }
185}
186
187static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
188 unsigned int length)
189{
190 return 0;
191}
192
193/*
194 * Report socket names for nfsdfs
195 */
196static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
197{
198 const struct sock *sk = svsk->sk_sk;
199 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
200 "udp" : "tcp";
201 int len;
202
203 switch (sk->sk_family) {
204 case PF_INET:
205 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
206 proto_name,
207 &inet_sk(sk)->inet_rcv_saddr,
208 inet_sk(sk)->inet_num);
209 break;
210#if IS_ENABLED(CONFIG_IPV6)
211 case PF_INET6:
212 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
213 proto_name,
214 &sk->sk_v6_rcv_saddr,
215 inet_sk(sk)->inet_num);
216 break;
217#endif
218 default:
219 len = snprintf(buf, remaining, "*unknown-%d*\n",
220 sk->sk_family);
221 }
222
223 if (len >= remaining) {
224 *buf = '\0';
225 return -ENAMETOOLONG;
226 }
227 return len;
228}
229
230static int
231svc_tcp_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
232 struct cmsghdr *cmsg, int ret)
233{
234 u8 content_type = tls_get_record_type(sock->sk, cmsg);
235 u8 level, description;
236
237 switch (content_type) {
238 case 0:
239 break;
240 case TLS_RECORD_TYPE_DATA:
241 /* TLS sets EOR at the end of each application data
242 * record, even though there might be more frames
243 * waiting to be decrypted.
244 */
245 msg->msg_flags &= ~MSG_EOR;
246 break;
247 case TLS_RECORD_TYPE_ALERT:
248 tls_alert_recv(sock->sk, msg, &level, &description);
249 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
250 -ENOTCONN : -EAGAIN;
251 break;
252 default:
253 /* discard this record type */
254 ret = -EAGAIN;
255 }
256 return ret;
257}
258
259static int
260svc_tcp_sock_recv_cmsg(struct svc_sock *svsk, struct msghdr *msg)
261{
262 union {
263 struct cmsghdr cmsg;
264 u8 buf[CMSG_SPACE(sizeof(u8))];
265 } u;
266 struct socket *sock = svsk->sk_sock;
267 int ret;
268
269 msg->msg_control = &u;
270 msg->msg_controllen = sizeof(u);
271 ret = sock_recvmsg(sock, msg, MSG_DONTWAIT);
272 if (unlikely(msg->msg_controllen != sizeof(u)))
273 ret = svc_tcp_sock_process_cmsg(sock, msg, &u.cmsg, ret);
274 return ret;
275}
276
277#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
278static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
279{
280 struct bvec_iter bi = {
281 .bi_size = size + seek,
282 };
283 struct bio_vec bv;
284
285 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
286 for_each_bvec(bv, bvec, bi, bi)
287 flush_dcache_page(bv.bv_page);
288}
289#else
290static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
291 size_t seek)
292{
293}
294#endif
295
296/*
297 * Read from @rqstp's transport socket. The incoming message fills whole
298 * pages in @rqstp's rq_pages array until the last page of the message
299 * has been received into a partial page.
300 */
301static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
302 size_t seek)
303{
304 struct svc_sock *svsk =
305 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
306 struct bio_vec *bvec = rqstp->rq_bvec;
307 struct msghdr msg = { NULL };
308 unsigned int i;
309 ssize_t len;
310 size_t t;
311
312 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
313
314 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE)
315 bvec_set_page(&bvec[i], rqstp->rq_pages[i], PAGE_SIZE, 0);
316 rqstp->rq_respages = &rqstp->rq_pages[i];
317 rqstp->rq_next_page = rqstp->rq_respages + 1;
318
319 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen);
320 if (seek) {
321 iov_iter_advance(&msg.msg_iter, seek);
322 buflen -= seek;
323 }
324 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
325 if (len > 0)
326 svc_flush_bvec(bvec, len, seek);
327
328 /* If we read a full record, then assume there may be more
329 * data to read (stream based sockets only!)
330 */
331 if (len == buflen)
332 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
333
334 return len;
335}
336
337/*
338 * Set socket snd and rcv buffer lengths
339 */
340static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
341{
342 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
343 struct socket *sock = svsk->sk_sock;
344
345 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
346
347 lock_sock(sock->sk);
348 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
349 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
350 sock->sk->sk_write_space(sock->sk);
351 release_sock(sock->sk);
352}
353
354static void svc_sock_secure_port(struct svc_rqst *rqstp)
355{
356 if (svc_port_is_privileged(svc_addr(rqstp)))
357 set_bit(RQ_SECURE, &rqstp->rq_flags);
358 else
359 clear_bit(RQ_SECURE, &rqstp->rq_flags);
360}
361
362/*
363 * INET callback when data has been received on the socket.
364 */
365static void svc_data_ready(struct sock *sk)
366{
367 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
368
369 trace_sk_data_ready(sk);
370
371 if (svsk) {
372 /* Refer to svc_setup_socket() for details. */
373 rmb();
374 svsk->sk_odata(sk);
375 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
376 if (test_bit(XPT_HANDSHAKE, &svsk->sk_xprt.xpt_flags))
377 return;
378 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
379 svc_xprt_enqueue(&svsk->sk_xprt);
380 }
381}
382
383/*
384 * INET callback when space is newly available on the socket.
385 */
386static void svc_write_space(struct sock *sk)
387{
388 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
389
390 if (svsk) {
391 /* Refer to svc_setup_socket() for details. */
392 rmb();
393 trace_svcsock_write_space(&svsk->sk_xprt, 0);
394 svsk->sk_owspace(sk);
395 svc_xprt_enqueue(&svsk->sk_xprt);
396 }
397}
398
399static int svc_tcp_has_wspace(struct svc_xprt *xprt)
400{
401 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
402
403 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
404 return 1;
405 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
406}
407
408static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
409{
410 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
411
412 sock_no_linger(svsk->sk_sock->sk);
413}
414
415/**
416 * svc_tcp_handshake_done - Handshake completion handler
417 * @data: address of xprt to wake
418 * @status: status of handshake
419 * @peerid: serial number of key containing the remote peer's identity
420 *
421 * If a security policy is specified as an export option, we don't
422 * have a specific export here to check. So we set a "TLS session
423 * is present" flag on the xprt and let an upper layer enforce local
424 * security policy.
425 */
426static void svc_tcp_handshake_done(void *data, int status, key_serial_t peerid)
427{
428 struct svc_xprt *xprt = data;
429 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
430
431 if (!status) {
432 if (peerid != TLS_NO_PEERID)
433 set_bit(XPT_PEER_AUTH, &xprt->xpt_flags);
434 set_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
435 }
436 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
437 complete_all(&svsk->sk_handshake_done);
438}
439
440/**
441 * svc_tcp_handshake - Perform a transport-layer security handshake
442 * @xprt: connected transport endpoint
443 *
444 */
445static void svc_tcp_handshake(struct svc_xprt *xprt)
446{
447 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
448 struct sock *sk = svsk->sk_sock->sk;
449 struct tls_handshake_args args = {
450 .ta_sock = svsk->sk_sock,
451 .ta_done = svc_tcp_handshake_done,
452 .ta_data = xprt,
453 };
454 int ret;
455
456 trace_svc_tls_upcall(xprt);
457
458 clear_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
459 init_completion(&svsk->sk_handshake_done);
460
461 ret = tls_server_hello_x509(&args, GFP_KERNEL);
462 if (ret) {
463 trace_svc_tls_not_started(xprt);
464 goto out_failed;
465 }
466
467 ret = wait_for_completion_interruptible_timeout(&svsk->sk_handshake_done,
468 SVC_HANDSHAKE_TO);
469 if (ret <= 0) {
470 if (tls_handshake_cancel(sk)) {
471 trace_svc_tls_timed_out(xprt);
472 goto out_close;
473 }
474 }
475
476 if (!test_bit(XPT_TLS_SESSION, &xprt->xpt_flags)) {
477 trace_svc_tls_unavailable(xprt);
478 goto out_close;
479 }
480
481 /* Mark the transport ready in case the remote sent RPC
482 * traffic before the kernel received the handshake
483 * completion downcall.
484 */
485 set_bit(XPT_DATA, &xprt->xpt_flags);
486 svc_xprt_enqueue(xprt);
487 return;
488
489out_close:
490 set_bit(XPT_CLOSE, &xprt->xpt_flags);
491out_failed:
492 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
493 set_bit(XPT_DATA, &xprt->xpt_flags);
494 svc_xprt_enqueue(xprt);
495}
496
497/*
498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
499 */
500static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
501 struct cmsghdr *cmh)
502{
503 struct in_pktinfo *pki = CMSG_DATA(cmh);
504 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
505
506 if (cmh->cmsg_type != IP_PKTINFO)
507 return 0;
508
509 daddr->sin_family = AF_INET;
510 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
511 return 1;
512}
513
514/*
515 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
516 */
517static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
518 struct cmsghdr *cmh)
519{
520 struct in6_pktinfo *pki = CMSG_DATA(cmh);
521 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
522
523 if (cmh->cmsg_type != IPV6_PKTINFO)
524 return 0;
525
526 daddr->sin6_family = AF_INET6;
527 daddr->sin6_addr = pki->ipi6_addr;
528 daddr->sin6_scope_id = pki->ipi6_ifindex;
529 return 1;
530}
531
532/*
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.
538 */
539static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
540 struct cmsghdr *cmh)
541{
542 switch (cmh->cmsg_level) {
543 case SOL_IP:
544 return svc_udp_get_dest_address4(rqstp, cmh);
545 case SOL_IPV6:
546 return svc_udp_get_dest_address6(rqstp, cmh);
547 }
548
549 return 0;
550}
551
552/**
553 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
554 * @rqstp: request structure into which to receive an RPC Call
555 *
556 * Called in a loop when XPT_DATA has been set.
557 *
558 * Returns:
559 * On success, the number of bytes in a received RPC Call, or
560 * %0 if a complete RPC Call message was not ready to return
561 */
562static int svc_udp_recvfrom(struct svc_rqst *rqstp)
563{
564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
567 struct sk_buff *skb;
568 union {
569 struct cmsghdr hdr;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
571 } buffer;
572 struct cmsghdr *cmh = &buffer.hdr;
573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
575 .msg_control = cmh,
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
578 };
579 size_t len;
580 int err;
581
582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
590 */
591 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
592
593 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
594 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
595 0, 0, MSG_PEEK | MSG_DONTWAIT);
596 if (err < 0)
597 goto out_recv_err;
598 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err);
599 if (!skb)
600 goto out_recv_err;
601
602 len = svc_addr_len(svc_addr(rqstp));
603 rqstp->rq_addrlen = len;
604 if (skb->tstamp == 0) {
605 skb->tstamp = ktime_get_real();
606 /* Don't enable netstamp, sunrpc doesn't
607 need that much accuracy */
608 }
609 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
610 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
611
612 len = skb->len;
613 rqstp->rq_arg.len = len;
614 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
615
616 rqstp->rq_prot = IPPROTO_UDP;
617
618 if (!svc_udp_get_dest_address(rqstp, cmh))
619 goto out_cmsg_err;
620 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
621
622 if (skb_is_nonlinear(skb)) {
623 /* we have to copy */
624 local_bh_disable();
625 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
626 goto out_bh_enable;
627 local_bh_enable();
628 consume_skb(skb);
629 } else {
630 /* we can use it in-place */
631 rqstp->rq_arg.head[0].iov_base = skb->data;
632 rqstp->rq_arg.head[0].iov_len = len;
633 if (skb_checksum_complete(skb))
634 goto out_free;
635 rqstp->rq_xprt_ctxt = skb;
636 }
637
638 rqstp->rq_arg.page_base = 0;
639 if (len <= rqstp->rq_arg.head[0].iov_len) {
640 rqstp->rq_arg.head[0].iov_len = len;
641 rqstp->rq_arg.page_len = 0;
642 rqstp->rq_respages = rqstp->rq_pages+1;
643 } else {
644 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
645 rqstp->rq_respages = rqstp->rq_pages + 1 +
646 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
647 }
648 rqstp->rq_next_page = rqstp->rq_respages+1;
649
650 if (serv->sv_stats)
651 serv->sv_stats->netudpcnt++;
652
653 svc_sock_secure_port(rqstp);
654 svc_xprt_received(rqstp->rq_xprt);
655 return len;
656
657out_recv_err:
658 if (err != -EAGAIN) {
659 /* possibly an icmp error */
660 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
661 }
662 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
663 goto out_clear_busy;
664out_cmsg_err:
665 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
666 cmh->cmsg_level, cmh->cmsg_type);
667 goto out_free;
668out_bh_enable:
669 local_bh_enable();
670out_free:
671 kfree_skb(skb);
672out_clear_busy:
673 svc_xprt_received(rqstp->rq_xprt);
674 return 0;
675}
676
677/**
678 * svc_udp_sendto - Send out a reply on a UDP socket
679 * @rqstp: completed svc_rqst
680 *
681 * xpt_mutex ensures @rqstp's whole message is written to the socket
682 * without interruption.
683 *
684 * Returns the number of bytes sent, or a negative errno.
685 */
686static int svc_udp_sendto(struct svc_rqst *rqstp)
687{
688 struct svc_xprt *xprt = rqstp->rq_xprt;
689 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
690 struct xdr_buf *xdr = &rqstp->rq_res;
691 union {
692 struct cmsghdr hdr;
693 long all[SVC_PKTINFO_SPACE / sizeof(long)];
694 } buffer;
695 struct cmsghdr *cmh = &buffer.hdr;
696 struct msghdr msg = {
697 .msg_name = &rqstp->rq_addr,
698 .msg_namelen = rqstp->rq_addrlen,
699 .msg_control = cmh,
700 .msg_flags = MSG_SPLICE_PAGES,
701 .msg_controllen = sizeof(buffer),
702 };
703 unsigned int count;
704 int err;
705
706 svc_udp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
707 rqstp->rq_xprt_ctxt = NULL;
708
709 svc_set_cmsg_data(rqstp, cmh);
710
711 mutex_lock(&xprt->xpt_mutex);
712
713 if (svc_xprt_is_dead(xprt))
714 goto out_notconn;
715
716 count = xdr_buf_to_bvec(rqstp->rq_bvec,
717 ARRAY_SIZE(rqstp->rq_bvec), xdr);
718
719 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
720 count, rqstp->rq_res.len);
721 err = sock_sendmsg(svsk->sk_sock, &msg);
722 if (err == -ECONNREFUSED) {
723 /* ICMP error on earlier request. */
724 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
725 count, rqstp->rq_res.len);
726 err = sock_sendmsg(svsk->sk_sock, &msg);
727 }
728
729 trace_svcsock_udp_send(xprt, err);
730
731 mutex_unlock(&xprt->xpt_mutex);
732 return err;
733
734out_notconn:
735 mutex_unlock(&xprt->xpt_mutex);
736 return -ENOTCONN;
737}
738
739static int svc_udp_has_wspace(struct svc_xprt *xprt)
740{
741 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
742 struct svc_serv *serv = xprt->xpt_server;
743 unsigned long required;
744
745 /*
746 * Set the SOCK_NOSPACE flag before checking the available
747 * sock space.
748 */
749 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
750 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
751 if (required*2 > sock_wspace(svsk->sk_sk))
752 return 0;
753 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
754 return 1;
755}
756
757static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
758{
759 BUG();
760 return NULL;
761}
762
763static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
764{
765}
766
767static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
768 struct net *net,
769 struct sockaddr *sa, int salen,
770 int flags)
771{
772 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
773}
774
775static const struct svc_xprt_ops svc_udp_ops = {
776 .xpo_create = svc_udp_create,
777 .xpo_recvfrom = svc_udp_recvfrom,
778 .xpo_sendto = svc_udp_sendto,
779 .xpo_result_payload = svc_sock_result_payload,
780 .xpo_release_ctxt = svc_udp_release_ctxt,
781 .xpo_detach = svc_sock_detach,
782 .xpo_free = svc_sock_free,
783 .xpo_has_wspace = svc_udp_has_wspace,
784 .xpo_accept = svc_udp_accept,
785 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
786};
787
788static struct svc_xprt_class svc_udp_class = {
789 .xcl_name = "udp",
790 .xcl_owner = THIS_MODULE,
791 .xcl_ops = &svc_udp_ops,
792 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
793 .xcl_ident = XPRT_TRANSPORT_UDP,
794};
795
796static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
797{
798 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
799 &svsk->sk_xprt, serv);
800 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
801 svsk->sk_sk->sk_data_ready = svc_data_ready;
802 svsk->sk_sk->sk_write_space = svc_write_space;
803
804 /* initialise setting must have enough space to
805 * receive and respond to one request.
806 * svc_udp_recvfrom will re-adjust if necessary
807 */
808 svc_sock_setbufsize(svsk, 3);
809
810 /* data might have come in before data_ready set up */
811 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
812 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
813
814 /* make sure we get destination address info */
815 switch (svsk->sk_sk->sk_family) {
816 case AF_INET:
817 ip_sock_set_pktinfo(svsk->sk_sock->sk);
818 break;
819 case AF_INET6:
820 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
821 break;
822 default:
823 BUG();
824 }
825}
826
827/*
828 * A data_ready event on a listening socket means there's a connection
829 * pending. Do not use state_change as a substitute for it.
830 */
831static void svc_tcp_listen_data_ready(struct sock *sk)
832{
833 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
834
835 trace_sk_data_ready(sk);
836
837 /*
838 * This callback may called twice when a new connection
839 * is established as a child socket inherits everything
840 * from a parent LISTEN socket.
841 * 1) data_ready method of the parent socket will be called
842 * when one of child sockets become ESTABLISHED.
843 * 2) data_ready method of the child socket may be called
844 * when it receives data before the socket is accepted.
845 * In case of 2, we should ignore it silently and DO NOT
846 * dereference svsk.
847 */
848 if (sk->sk_state != TCP_LISTEN)
849 return;
850
851 if (svsk) {
852 /* Refer to svc_setup_socket() for details. */
853 rmb();
854 svsk->sk_odata(sk);
855 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
856 svc_xprt_enqueue(&svsk->sk_xprt);
857 }
858}
859
860/*
861 * A state change on a connected socket means it's dying or dead.
862 */
863static void svc_tcp_state_change(struct sock *sk)
864{
865 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
866
867 if (svsk) {
868 /* Refer to svc_setup_socket() for details. */
869 rmb();
870 svsk->sk_ostate(sk);
871 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
872 if (sk->sk_state != TCP_ESTABLISHED)
873 svc_xprt_deferred_close(&svsk->sk_xprt);
874 }
875}
876
877/*
878 * Accept a TCP connection
879 */
880static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
881{
882 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
883 struct sockaddr_storage addr;
884 struct sockaddr *sin = (struct sockaddr *) &addr;
885 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
886 struct socket *sock = svsk->sk_sock;
887 struct socket *newsock;
888 struct svc_sock *newsvsk;
889 int err, slen;
890
891 if (!sock)
892 return NULL;
893
894 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
895 err = kernel_accept(sock, &newsock, O_NONBLOCK);
896 if (err < 0) {
897 if (err != -EAGAIN)
898 trace_svcsock_accept_err(xprt, serv->sv_name, err);
899 return NULL;
900 }
901 if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL)))
902 return NULL;
903
904 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
905
906 err = kernel_getpeername(newsock, sin);
907 if (err < 0) {
908 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
909 goto failed; /* aborted connection or whatever */
910 }
911 slen = err;
912
913 /* Reset the inherited callbacks before calling svc_setup_socket */
914 newsock->sk->sk_state_change = svsk->sk_ostate;
915 newsock->sk->sk_data_ready = svsk->sk_odata;
916 newsock->sk->sk_write_space = svsk->sk_owspace;
917
918 /* make sure that a write doesn't block forever when
919 * low on memory
920 */
921 newsock->sk->sk_sndtimeo = HZ*30;
922
923 newsvsk = svc_setup_socket(serv, newsock,
924 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
925 if (IS_ERR(newsvsk))
926 goto failed;
927 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
928 err = kernel_getsockname(newsock, sin);
929 slen = err;
930 if (unlikely(err < 0))
931 slen = offsetof(struct sockaddr, sa_data);
932 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
933
934 if (sock_is_loopback(newsock->sk))
935 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
936 else
937 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
938 if (serv->sv_stats)
939 serv->sv_stats->nettcpconn++;
940
941 return &newsvsk->sk_xprt;
942
943failed:
944 sockfd_put(newsock);
945 return NULL;
946}
947
948static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
949 struct svc_rqst *rqstp)
950{
951 size_t len = svsk->sk_datalen;
952 unsigned int i, npages;
953
954 if (!len)
955 return 0;
956 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
957 for (i = 0; i < npages; i++) {
958 if (rqstp->rq_pages[i] != NULL)
959 put_page(rqstp->rq_pages[i]);
960 BUG_ON(svsk->sk_pages[i] == NULL);
961 rqstp->rq_pages[i] = svsk->sk_pages[i];
962 svsk->sk_pages[i] = NULL;
963 }
964 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
965 return len;
966}
967
968static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
969{
970 unsigned int i, len, npages;
971
972 if (svsk->sk_datalen == 0)
973 return;
974 len = svsk->sk_datalen;
975 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 for (i = 0; i < npages; i++) {
977 svsk->sk_pages[i] = rqstp->rq_pages[i];
978 rqstp->rq_pages[i] = NULL;
979 }
980}
981
982static void svc_tcp_clear_pages(struct svc_sock *svsk)
983{
984 unsigned int i, len, npages;
985
986 if (svsk->sk_datalen == 0)
987 goto out;
988 len = svsk->sk_datalen;
989 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
990 for (i = 0; i < npages; i++) {
991 if (svsk->sk_pages[i] == NULL) {
992 WARN_ON_ONCE(1);
993 continue;
994 }
995 put_page(svsk->sk_pages[i]);
996 svsk->sk_pages[i] = NULL;
997 }
998out:
999 svsk->sk_tcplen = 0;
1000 svsk->sk_datalen = 0;
1001}
1002
1003/*
1004 * Receive fragment record header into sk_marker.
1005 */
1006static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
1007 struct svc_rqst *rqstp)
1008{
1009 ssize_t want, len;
1010
1011 /* If we haven't gotten the record length yet,
1012 * get the next four bytes.
1013 */
1014 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1015 struct msghdr msg = { NULL };
1016 struct kvec iov;
1017
1018 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1019 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
1020 iov.iov_len = want;
1021 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
1022 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
1023 if (len < 0)
1024 return len;
1025 svsk->sk_tcplen += len;
1026 if (len < want) {
1027 /* call again to read the remaining bytes */
1028 goto err_short;
1029 }
1030 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
1031 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1032 svsk->sk_xprt.xpt_server->sv_max_mesg)
1033 goto err_too_large;
1034 }
1035 return svc_sock_reclen(svsk);
1036
1037err_too_large:
1038 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
1039 __func__, svsk->sk_xprt.xpt_server->sv_name,
1040 svc_sock_reclen(svsk));
1041 svc_xprt_deferred_close(&svsk->sk_xprt);
1042err_short:
1043 return -EAGAIN;
1044}
1045
1046static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1047{
1048 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1049 struct rpc_rqst *req = NULL;
1050 struct kvec *src, *dst;
1051 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1052 __be32 xid = *p;
1053
1054 if (!bc_xprt)
1055 return -EAGAIN;
1056 spin_lock(&bc_xprt->queue_lock);
1057 req = xprt_lookup_rqst(bc_xprt, xid);
1058 if (!req)
1059 goto unlock_eagain;
1060
1061 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1062 /*
1063 * XXX!: cheating for now! Only copying HEAD.
1064 * But we know this is good enough for now (in fact, for any
1065 * callback reply in the forseeable future).
1066 */
1067 dst = &req->rq_private_buf.head[0];
1068 src = &rqstp->rq_arg.head[0];
1069 if (dst->iov_len < src->iov_len)
1070 goto unlock_eagain; /* whatever; just giving up. */
1071 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1072 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1073 rqstp->rq_arg.len = 0;
1074 spin_unlock(&bc_xprt->queue_lock);
1075 return 0;
1076unlock_eagain:
1077 spin_unlock(&bc_xprt->queue_lock);
1078 return -EAGAIN;
1079}
1080
1081static void svc_tcp_fragment_received(struct svc_sock *svsk)
1082{
1083 /* If we have more data, signal svc_xprt_enqueue() to try again */
1084 svsk->sk_tcplen = 0;
1085 svsk->sk_marker = xdr_zero;
1086
1087 smp_wmb();
1088 tcp_set_rcvlowat(svsk->sk_sk, 1);
1089}
1090
1091/**
1092 * svc_tcp_recvfrom - Receive data from a TCP socket
1093 * @rqstp: request structure into which to receive an RPC Call
1094 *
1095 * Called in a loop when XPT_DATA has been set.
1096 *
1097 * Read the 4-byte stream record marker, then use the record length
1098 * in that marker to set up exactly the resources needed to receive
1099 * the next RPC message into @rqstp.
1100 *
1101 * Returns:
1102 * On success, the number of bytes in a received RPC Call, or
1103 * %0 if a complete RPC Call message was not ready to return
1104 *
1105 * The zero return case handles partial receives and callback Replies.
1106 * The state of a partial receive is preserved in the svc_sock for
1107 * the next call to svc_tcp_recvfrom.
1108 */
1109static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1110{
1111 struct svc_sock *svsk =
1112 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1113 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1114 size_t want, base;
1115 ssize_t len;
1116 __be32 *p;
1117 __be32 calldir;
1118
1119 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1120 len = svc_tcp_read_marker(svsk, rqstp);
1121 if (len < 0)
1122 goto error;
1123
1124 base = svc_tcp_restore_pages(svsk, rqstp);
1125 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1126 len = svc_tcp_read_msg(rqstp, base + want, base);
1127 if (len >= 0) {
1128 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1129 svsk->sk_tcplen += len;
1130 svsk->sk_datalen += len;
1131 }
1132 if (len != want || !svc_sock_final_rec(svsk))
1133 goto err_incomplete;
1134 if (svsk->sk_datalen < 8)
1135 goto err_nuts;
1136
1137 rqstp->rq_arg.len = svsk->sk_datalen;
1138 rqstp->rq_arg.page_base = 0;
1139 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1140 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1141 rqstp->rq_arg.page_len = 0;
1142 } else
1143 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1144
1145 rqstp->rq_xprt_ctxt = NULL;
1146 rqstp->rq_prot = IPPROTO_TCP;
1147 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1148 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1149 else
1150 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1151
1152 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1153 calldir = p[1];
1154 if (calldir)
1155 len = receive_cb_reply(svsk, rqstp);
1156
1157 /* Reset TCP read info */
1158 svsk->sk_datalen = 0;
1159 svc_tcp_fragment_received(svsk);
1160
1161 if (len < 0)
1162 goto error;
1163
1164 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1165 if (serv->sv_stats)
1166 serv->sv_stats->nettcpcnt++;
1167
1168 svc_sock_secure_port(rqstp);
1169 svc_xprt_received(rqstp->rq_xprt);
1170 return rqstp->rq_arg.len;
1171
1172err_incomplete:
1173 svc_tcp_save_pages(svsk, rqstp);
1174 if (len < 0 && len != -EAGAIN)
1175 goto err_delete;
1176 if (len == want)
1177 svc_tcp_fragment_received(svsk);
1178 else {
1179 /* Avoid more ->sk_data_ready() calls until the rest
1180 * of the message has arrived. This reduces service
1181 * thread wake-ups on large incoming messages. */
1182 tcp_set_rcvlowat(svsk->sk_sk,
1183 svc_sock_reclen(svsk) - svsk->sk_tcplen);
1184
1185 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1186 svc_sock_reclen(svsk),
1187 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1188 }
1189 goto err_noclose;
1190error:
1191 if (len != -EAGAIN)
1192 goto err_delete;
1193 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1194 goto err_noclose;
1195err_nuts:
1196 svsk->sk_datalen = 0;
1197err_delete:
1198 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1199 svc_xprt_deferred_close(&svsk->sk_xprt);
1200err_noclose:
1201 svc_xprt_received(rqstp->rq_xprt);
1202 return 0; /* record not complete */
1203}
1204
1205/*
1206 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1207 * copy operations in this path. Therefore the caller must ensure
1208 * that the pages backing @xdr are unchanging.
1209 *
1210 * Note that the send is non-blocking. The caller has incremented
1211 * the reference count on each page backing the RPC message, and
1212 * the network layer will "put" these pages when transmission is
1213 * complete.
1214 *
1215 * This is safe for our RPC services because the memory backing
1216 * the head and tail components is never kmalloc'd. These always
1217 * come from pages in the svc_rqst::rq_pages array.
1218 */
1219static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1220 rpc_fraghdr marker, unsigned int *sentp)
1221{
1222 struct msghdr msg = {
1223 .msg_flags = MSG_SPLICE_PAGES,
1224 };
1225 unsigned int count;
1226 void *buf;
1227 int ret;
1228
1229 *sentp = 0;
1230
1231 /* The stream record marker is copied into a temporary page
1232 * fragment buffer so that it can be included in rq_bvec.
1233 */
1234 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1235 GFP_KERNEL);
1236 if (!buf)
1237 return -ENOMEM;
1238 memcpy(buf, &marker, sizeof(marker));
1239 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1240
1241 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1242 ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1243
1244 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1245 1 + count, sizeof(marker) + rqstp->rq_res.len);
1246 ret = sock_sendmsg(svsk->sk_sock, &msg);
1247 if (ret < 0)
1248 return ret;
1249 *sentp += ret;
1250 return 0;
1251}
1252
1253/**
1254 * svc_tcp_sendto - Send out a reply on a TCP socket
1255 * @rqstp: completed svc_rqst
1256 *
1257 * xpt_mutex ensures @rqstp's whole message is written to the socket
1258 * without interruption.
1259 *
1260 * Returns the number of bytes sent, or a negative errno.
1261 */
1262static int svc_tcp_sendto(struct svc_rqst *rqstp)
1263{
1264 struct svc_xprt *xprt = rqstp->rq_xprt;
1265 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1266 struct xdr_buf *xdr = &rqstp->rq_res;
1267 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1268 (u32)xdr->len);
1269 unsigned int sent;
1270 int err;
1271
1272 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1273 rqstp->rq_xprt_ctxt = NULL;
1274
1275 mutex_lock(&xprt->xpt_mutex);
1276 if (svc_xprt_is_dead(xprt))
1277 goto out_notconn;
1278 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1279 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1280 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1281 goto out_close;
1282 mutex_unlock(&xprt->xpt_mutex);
1283 return sent;
1284
1285out_notconn:
1286 mutex_unlock(&xprt->xpt_mutex);
1287 return -ENOTCONN;
1288out_close:
1289 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1290 xprt->xpt_server->sv_name,
1291 (err < 0) ? "got error" : "sent",
1292 (err < 0) ? err : sent, xdr->len);
1293 svc_xprt_deferred_close(xprt);
1294 mutex_unlock(&xprt->xpt_mutex);
1295 return -EAGAIN;
1296}
1297
1298static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1299 struct net *net,
1300 struct sockaddr *sa, int salen,
1301 int flags)
1302{
1303 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1304}
1305
1306static const struct svc_xprt_ops svc_tcp_ops = {
1307 .xpo_create = svc_tcp_create,
1308 .xpo_recvfrom = svc_tcp_recvfrom,
1309 .xpo_sendto = svc_tcp_sendto,
1310 .xpo_result_payload = svc_sock_result_payload,
1311 .xpo_release_ctxt = svc_tcp_release_ctxt,
1312 .xpo_detach = svc_tcp_sock_detach,
1313 .xpo_free = svc_sock_free,
1314 .xpo_has_wspace = svc_tcp_has_wspace,
1315 .xpo_accept = svc_tcp_accept,
1316 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1317 .xpo_handshake = svc_tcp_handshake,
1318};
1319
1320static struct svc_xprt_class svc_tcp_class = {
1321 .xcl_name = "tcp",
1322 .xcl_owner = THIS_MODULE,
1323 .xcl_ops = &svc_tcp_ops,
1324 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1325 .xcl_ident = XPRT_TRANSPORT_TCP,
1326};
1327
1328void svc_init_xprt_sock(void)
1329{
1330 svc_reg_xprt_class(&svc_tcp_class);
1331 svc_reg_xprt_class(&svc_udp_class);
1332}
1333
1334void svc_cleanup_xprt_sock(void)
1335{
1336 svc_unreg_xprt_class(&svc_tcp_class);
1337 svc_unreg_xprt_class(&svc_udp_class);
1338}
1339
1340static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1341{
1342 struct sock *sk = svsk->sk_sk;
1343
1344 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1345 &svsk->sk_xprt, serv);
1346 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1347 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1348 if (sk->sk_state == TCP_LISTEN) {
1349 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1350 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1351 sk->sk_data_ready = svc_tcp_listen_data_ready;
1352 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1353 } else {
1354 sk->sk_state_change = svc_tcp_state_change;
1355 sk->sk_data_ready = svc_data_ready;
1356 sk->sk_write_space = svc_write_space;
1357
1358 svsk->sk_marker = xdr_zero;
1359 svsk->sk_tcplen = 0;
1360 svsk->sk_datalen = 0;
1361 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1362
1363 tcp_sock_set_nodelay(sk);
1364
1365 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1366 switch (sk->sk_state) {
1367 case TCP_SYN_RECV:
1368 case TCP_ESTABLISHED:
1369 break;
1370 default:
1371 svc_xprt_deferred_close(&svsk->sk_xprt);
1372 }
1373 }
1374}
1375
1376void svc_sock_update_bufs(struct svc_serv *serv)
1377{
1378 /*
1379 * The number of server threads has changed. Update
1380 * rcvbuf and sndbuf accordingly on all sockets
1381 */
1382 struct svc_sock *svsk;
1383
1384 spin_lock_bh(&serv->sv_lock);
1385 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1386 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1387 spin_unlock_bh(&serv->sv_lock);
1388}
1389EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1390
1391/*
1392 * Initialize socket for RPC use and create svc_sock struct
1393 */
1394static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1395 struct socket *sock,
1396 int flags)
1397{
1398 struct svc_sock *svsk;
1399 struct sock *inet;
1400 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1401
1402 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1403 if (!svsk)
1404 return ERR_PTR(-ENOMEM);
1405
1406 inet = sock->sk;
1407
1408 if (pmap_register) {
1409 int err;
1410
1411 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1412 inet->sk_protocol,
1413 ntohs(inet_sk(inet)->inet_sport));
1414 if (err < 0) {
1415 kfree(svsk);
1416 return ERR_PTR(err);
1417 }
1418 }
1419
1420 svsk->sk_sock = sock;
1421 svsk->sk_sk = inet;
1422 svsk->sk_ostate = inet->sk_state_change;
1423 svsk->sk_odata = inet->sk_data_ready;
1424 svsk->sk_owspace = inet->sk_write_space;
1425 /*
1426 * This barrier is necessary in order to prevent race condition
1427 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1428 * when calling callbacks above.
1429 */
1430 wmb();
1431 inet->sk_user_data = svsk;
1432
1433 /* Initialize the socket */
1434 if (sock->type == SOCK_DGRAM)
1435 svc_udp_init(svsk, serv);
1436 else
1437 svc_tcp_init(svsk, serv);
1438
1439 trace_svcsock_new(svsk, sock);
1440 return svsk;
1441}
1442
1443/**
1444 * svc_addsock - add a listener socket to an RPC service
1445 * @serv: pointer to RPC service to which to add a new listener
1446 * @net: caller's network namespace
1447 * @fd: file descriptor of the new listener
1448 * @name_return: pointer to buffer to fill in with name of listener
1449 * @len: size of the buffer
1450 * @cred: credential
1451 *
1452 * Fills in socket name and returns positive length of name if successful.
1453 * Name is terminated with '\n'. On error, returns a negative errno
1454 * value.
1455 */
1456int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1457 char *name_return, const size_t len, const struct cred *cred)
1458{
1459 int err = 0;
1460 struct socket *so = sockfd_lookup(fd, &err);
1461 struct svc_sock *svsk = NULL;
1462 struct sockaddr_storage addr;
1463 struct sockaddr *sin = (struct sockaddr *)&addr;
1464 int salen;
1465
1466 if (!so)
1467 return err;
1468 err = -EINVAL;
1469 if (sock_net(so->sk) != net)
1470 goto out;
1471 err = -EAFNOSUPPORT;
1472 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1473 goto out;
1474 err = -EPROTONOSUPPORT;
1475 if (so->sk->sk_protocol != IPPROTO_TCP &&
1476 so->sk->sk_protocol != IPPROTO_UDP)
1477 goto out;
1478 err = -EISCONN;
1479 if (so->state > SS_UNCONNECTED)
1480 goto out;
1481 err = -ENOENT;
1482 if (!try_module_get(THIS_MODULE))
1483 goto out;
1484 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1485 if (IS_ERR(svsk)) {
1486 module_put(THIS_MODULE);
1487 err = PTR_ERR(svsk);
1488 goto out;
1489 }
1490 salen = kernel_getsockname(svsk->sk_sock, sin);
1491 if (salen >= 0)
1492 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1493 svsk->sk_xprt.xpt_cred = get_cred(cred);
1494 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1495 return svc_one_sock_name(svsk, name_return, len);
1496out:
1497 sockfd_put(so);
1498 return err;
1499}
1500EXPORT_SYMBOL_GPL(svc_addsock);
1501
1502/*
1503 * Create socket for RPC service.
1504 */
1505static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1506 int protocol,
1507 struct net *net,
1508 struct sockaddr *sin, int len,
1509 int flags)
1510{
1511 struct svc_sock *svsk;
1512 struct socket *sock;
1513 int error;
1514 int type;
1515 struct sockaddr_storage addr;
1516 struct sockaddr *newsin = (struct sockaddr *)&addr;
1517 int newlen;
1518 int family;
1519
1520 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1521 printk(KERN_WARNING "svc: only UDP and TCP "
1522 "sockets supported\n");
1523 return ERR_PTR(-EINVAL);
1524 }
1525
1526 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1527 switch (sin->sa_family) {
1528 case AF_INET6:
1529 family = PF_INET6;
1530 break;
1531 case AF_INET:
1532 family = PF_INET;
1533 break;
1534 default:
1535 return ERR_PTR(-EINVAL);
1536 }
1537
1538 error = __sock_create(net, family, type, protocol, &sock, 1);
1539 if (error < 0)
1540 return ERR_PTR(error);
1541
1542 svc_reclassify_socket(sock);
1543
1544 /*
1545 * If this is an PF_INET6 listener, we want to avoid
1546 * getting requests from IPv4 remotes. Those should
1547 * be shunted to a PF_INET listener via rpcbind.
1548 */
1549 if (family == PF_INET6)
1550 ip6_sock_set_v6only(sock->sk);
1551 if (type == SOCK_STREAM)
1552 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1553 error = kernel_bind(sock, sin, len);
1554 if (error < 0)
1555 goto bummer;
1556
1557 error = kernel_getsockname(sock, newsin);
1558 if (error < 0)
1559 goto bummer;
1560 newlen = error;
1561
1562 if (protocol == IPPROTO_TCP) {
1563 if ((error = kernel_listen(sock, 64)) < 0)
1564 goto bummer;
1565 }
1566
1567 svsk = svc_setup_socket(serv, sock, flags);
1568 if (IS_ERR(svsk)) {
1569 error = PTR_ERR(svsk);
1570 goto bummer;
1571 }
1572 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1573 return (struct svc_xprt *)svsk;
1574bummer:
1575 sock_release(sock);
1576 return ERR_PTR(error);
1577}
1578
1579/*
1580 * Detach the svc_sock from the socket so that no
1581 * more callbacks occur.
1582 */
1583static void svc_sock_detach(struct svc_xprt *xprt)
1584{
1585 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1586 struct sock *sk = svsk->sk_sk;
1587
1588 /* put back the old socket callbacks */
1589 lock_sock(sk);
1590 sk->sk_state_change = svsk->sk_ostate;
1591 sk->sk_data_ready = svsk->sk_odata;
1592 sk->sk_write_space = svsk->sk_owspace;
1593 sk->sk_user_data = NULL;
1594 release_sock(sk);
1595}
1596
1597/*
1598 * Disconnect the socket, and reset the callbacks
1599 */
1600static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1601{
1602 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1603
1604 tls_handshake_close(svsk->sk_sock);
1605
1606 svc_sock_detach(xprt);
1607
1608 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1609 svc_tcp_clear_pages(svsk);
1610 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1611 }
1612}
1613
1614/*
1615 * Free the svc_sock's socket resources and the svc_sock itself.
1616 */
1617static void svc_sock_free(struct svc_xprt *xprt)
1618{
1619 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1620 struct page_frag_cache *pfc = &svsk->sk_frag_cache;
1621 struct socket *sock = svsk->sk_sock;
1622
1623 trace_svcsock_free(svsk, sock);
1624
1625 tls_handshake_cancel(sock->sk);
1626 if (sock->file)
1627 sockfd_put(sock);
1628 else
1629 sock_release(sock);
1630 if (pfc->va)
1631 __page_frag_cache_drain(virt_to_head_page(pfc->va),
1632 pfc->pagecnt_bias);
1633 kfree(svsk);
1634}
1/*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
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...
9 *
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.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
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>
27#include <linux/in.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>
37#include <net/sock.h>
38#include <net/checksum.h>
39#include <net/ip.h>
40#include <net/ipv6.h>
41#include <net/tcp.h>
42#include <net/tcp_states.h>
43#include <asm/uaccess.h>
44#include <asm/ioctls.h>
45
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>
52#include <linux/sunrpc/xprt.h>
53
54#include "sunrpc.h"
55
56#define RPCDBG_FACILITY RPCDBG_SVCXPRT
57
58
59static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
60 int *errp, int flags);
61static void svc_udp_data_ready(struct sock *, int);
62static int svc_udp_recvfrom(struct svc_rqst *);
63static int svc_udp_sendto(struct svc_rqst *);
64static void svc_sock_detach(struct svc_xprt *);
65static void svc_tcp_sock_detach(struct svc_xprt *);
66static void svc_sock_free(struct svc_xprt *);
67
68static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
69 struct net *, struct sockaddr *,
70 int, int);
71#if defined(CONFIG_SUNRPC_BACKCHANNEL)
72static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
73 struct net *, struct sockaddr *,
74 int, int);
75static void svc_bc_sock_free(struct svc_xprt *xprt);
76#endif /* CONFIG_SUNRPC_BACKCHANNEL */
77
78#ifdef CONFIG_DEBUG_LOCK_ALLOC
79static struct lock_class_key svc_key[2];
80static struct lock_class_key svc_slock_key[2];
81
82static void svc_reclassify_socket(struct socket *sock)
83{
84 struct sock *sk = sock->sk;
85 BUG_ON(sock_owned_by_user(sk));
86 switch (sk->sk_family) {
87 case AF_INET:
88 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
89 &svc_slock_key[0],
90 "sk_xprt.xpt_lock-AF_INET-NFSD",
91 &svc_key[0]);
92 break;
93
94 case AF_INET6:
95 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
96 &svc_slock_key[1],
97 "sk_xprt.xpt_lock-AF_INET6-NFSD",
98 &svc_key[1]);
99 break;
100
101 default:
102 BUG();
103 }
104}
105#else
106static void svc_reclassify_socket(struct socket *sock)
107{
108}
109#endif
110
111/*
112 * Release an skbuff after use
113 */
114static void svc_release_skb(struct svc_rqst *rqstp)
115{
116 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
117
118 if (skb) {
119 struct svc_sock *svsk =
120 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
121 rqstp->rq_xprt_ctxt = NULL;
122
123 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
124 skb_free_datagram_locked(svsk->sk_sk, skb);
125 }
126}
127
128union svc_pktinfo_u {
129 struct in_pktinfo pkti;
130 struct in6_pktinfo pkti6;
131};
132#define SVC_PKTINFO_SPACE \
133 CMSG_SPACE(sizeof(union svc_pktinfo_u))
134
135static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
136{
137 struct svc_sock *svsk =
138 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
139 switch (svsk->sk_sk->sk_family) {
140 case AF_INET: {
141 struct in_pktinfo *pki = CMSG_DATA(cmh);
142
143 cmh->cmsg_level = SOL_IP;
144 cmh->cmsg_type = IP_PKTINFO;
145 pki->ipi_ifindex = 0;
146 pki->ipi_spec_dst.s_addr = rqstp->rq_daddr.addr.s_addr;
147 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
148 }
149 break;
150
151 case AF_INET6: {
152 struct in6_pktinfo *pki = CMSG_DATA(cmh);
153
154 cmh->cmsg_level = SOL_IPV6;
155 cmh->cmsg_type = IPV6_PKTINFO;
156 pki->ipi6_ifindex = 0;
157 ipv6_addr_copy(&pki->ipi6_addr,
158 &rqstp->rq_daddr.addr6);
159 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
160 }
161 break;
162 }
163}
164
165/*
166 * send routine intended to be shared by the fore- and back-channel
167 */
168int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
169 struct page *headpage, unsigned long headoffset,
170 struct page *tailpage, unsigned long tailoffset)
171{
172 int result;
173 int size;
174 struct page **ppage = xdr->pages;
175 size_t base = xdr->page_base;
176 unsigned int pglen = xdr->page_len;
177 unsigned int flags = MSG_MORE;
178 int slen;
179 int len = 0;
180
181 slen = xdr->len;
182
183 /* send head */
184 if (slen == xdr->head[0].iov_len)
185 flags = 0;
186 len = kernel_sendpage(sock, headpage, headoffset,
187 xdr->head[0].iov_len, flags);
188 if (len != xdr->head[0].iov_len)
189 goto out;
190 slen -= xdr->head[0].iov_len;
191 if (slen == 0)
192 goto out;
193
194 /* send page data */
195 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
196 while (pglen > 0) {
197 if (slen == size)
198 flags = 0;
199 result = kernel_sendpage(sock, *ppage, base, size, flags);
200 if (result > 0)
201 len += result;
202 if (result != size)
203 goto out;
204 slen -= size;
205 pglen -= size;
206 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
207 base = 0;
208 ppage++;
209 }
210
211 /* send tail */
212 if (xdr->tail[0].iov_len) {
213 result = kernel_sendpage(sock, tailpage, tailoffset,
214 xdr->tail[0].iov_len, 0);
215 if (result > 0)
216 len += result;
217 }
218
219out:
220 return len;
221}
222
223
224/*
225 * Generic sendto routine
226 */
227static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
228{
229 struct svc_sock *svsk =
230 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
231 struct socket *sock = svsk->sk_sock;
232 union {
233 struct cmsghdr hdr;
234 long all[SVC_PKTINFO_SPACE / sizeof(long)];
235 } buffer;
236 struct cmsghdr *cmh = &buffer.hdr;
237 int len = 0;
238 unsigned long tailoff;
239 unsigned long headoff;
240 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
241
242 if (rqstp->rq_prot == IPPROTO_UDP) {
243 struct msghdr msg = {
244 .msg_name = &rqstp->rq_addr,
245 .msg_namelen = rqstp->rq_addrlen,
246 .msg_control = cmh,
247 .msg_controllen = sizeof(buffer),
248 .msg_flags = MSG_MORE,
249 };
250
251 svc_set_cmsg_data(rqstp, cmh);
252
253 if (sock_sendmsg(sock, &msg, 0) < 0)
254 goto out;
255 }
256
257 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
258 headoff = 0;
259 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
260 rqstp->rq_respages[0], tailoff);
261
262out:
263 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
264 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
265 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
266
267 return len;
268}
269
270/*
271 * Report socket names for nfsdfs
272 */
273static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
274{
275 const struct sock *sk = svsk->sk_sk;
276 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
277 "udp" : "tcp";
278 int len;
279
280 switch (sk->sk_family) {
281 case PF_INET:
282 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
283 proto_name,
284 &inet_sk(sk)->inet_rcv_saddr,
285 inet_sk(sk)->inet_num);
286 break;
287 case PF_INET6:
288 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
289 proto_name,
290 &inet6_sk(sk)->rcv_saddr,
291 inet_sk(sk)->inet_num);
292 break;
293 default:
294 len = snprintf(buf, remaining, "*unknown-%d*\n",
295 sk->sk_family);
296 }
297
298 if (len >= remaining) {
299 *buf = '\0';
300 return -ENAMETOOLONG;
301 }
302 return len;
303}
304
305/**
306 * svc_sock_names - construct a list of listener names in a string
307 * @serv: pointer to RPC service
308 * @buf: pointer to a buffer to fill in with socket names
309 * @buflen: size of the buffer to be filled
310 * @toclose: pointer to '\0'-terminated C string containing the name
311 * of a listener to be closed
312 *
313 * Fills in @buf with a '\n'-separated list of names of listener
314 * sockets. If @toclose is not NULL, the socket named by @toclose
315 * is closed, and is not included in the output list.
316 *
317 * Returns positive length of the socket name string, or a negative
318 * errno value on error.
319 */
320int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
321 const char *toclose)
322{
323 struct svc_sock *svsk, *closesk = NULL;
324 int len = 0;
325
326 if (!serv)
327 return 0;
328
329 spin_lock_bh(&serv->sv_lock);
330 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
331 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
332 if (onelen < 0) {
333 len = onelen;
334 break;
335 }
336 if (toclose && strcmp(toclose, buf + len) == 0) {
337 closesk = svsk;
338 svc_xprt_get(&closesk->sk_xprt);
339 } else
340 len += onelen;
341 }
342 spin_unlock_bh(&serv->sv_lock);
343
344 if (closesk) {
345 /* Should unregister with portmap, but you cannot
346 * unregister just one protocol...
347 */
348 svc_close_xprt(&closesk->sk_xprt);
349 svc_xprt_put(&closesk->sk_xprt);
350 } else if (toclose)
351 return -ENOENT;
352 return len;
353}
354EXPORT_SYMBOL_GPL(svc_sock_names);
355
356/*
357 * Check input queue length
358 */
359static int svc_recv_available(struct svc_sock *svsk)
360{
361 struct socket *sock = svsk->sk_sock;
362 int avail, err;
363
364 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
365
366 return (err >= 0)? avail : err;
367}
368
369/*
370 * Generic recvfrom routine.
371 */
372static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
373 int buflen)
374{
375 struct svc_sock *svsk =
376 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
377 struct msghdr msg = {
378 .msg_flags = MSG_DONTWAIT,
379 };
380 int len;
381
382 rqstp->rq_xprt_hlen = 0;
383
384 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
385 msg.msg_flags);
386
387 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
388 svsk, iov[0].iov_base, iov[0].iov_len, len);
389 return len;
390}
391
392static int svc_partial_recvfrom(struct svc_rqst *rqstp,
393 struct kvec *iov, int nr,
394 int buflen, unsigned int base)
395{
396 size_t save_iovlen;
397 void __user *save_iovbase;
398 unsigned int i;
399 int ret;
400
401 if (base == 0)
402 return svc_recvfrom(rqstp, iov, nr, buflen);
403
404 for (i = 0; i < nr; i++) {
405 if (iov[i].iov_len > base)
406 break;
407 base -= iov[i].iov_len;
408 }
409 save_iovlen = iov[i].iov_len;
410 save_iovbase = iov[i].iov_base;
411 iov[i].iov_len -= base;
412 iov[i].iov_base += base;
413 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
414 iov[i].iov_len = save_iovlen;
415 iov[i].iov_base = save_iovbase;
416 return ret;
417}
418
419/*
420 * Set socket snd and rcv buffer lengths
421 */
422static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
423 unsigned int rcv)
424{
425#if 0
426 mm_segment_t oldfs;
427 oldfs = get_fs(); set_fs(KERNEL_DS);
428 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
429 (char*)&snd, sizeof(snd));
430 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
431 (char*)&rcv, sizeof(rcv));
432#else
433 /* sock_setsockopt limits use to sysctl_?mem_max,
434 * which isn't acceptable. Until that is made conditional
435 * on not having CAP_SYS_RESOURCE or similar, we go direct...
436 * DaveM said I could!
437 */
438 lock_sock(sock->sk);
439 sock->sk->sk_sndbuf = snd * 2;
440 sock->sk->sk_rcvbuf = rcv * 2;
441 sock->sk->sk_write_space(sock->sk);
442 release_sock(sock->sk);
443#endif
444}
445/*
446 * INET callback when data has been received on the socket.
447 */
448static void svc_udp_data_ready(struct sock *sk, int count)
449{
450 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
451 wait_queue_head_t *wq = sk_sleep(sk);
452
453 if (svsk) {
454 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
455 svsk, sk, count,
456 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
457 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
458 svc_xprt_enqueue(&svsk->sk_xprt);
459 }
460 if (wq && waitqueue_active(wq))
461 wake_up_interruptible(wq);
462}
463
464/*
465 * INET callback when space is newly available on the socket.
466 */
467static void svc_write_space(struct sock *sk)
468{
469 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
470 wait_queue_head_t *wq = sk_sleep(sk);
471
472 if (svsk) {
473 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
474 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
475 svc_xprt_enqueue(&svsk->sk_xprt);
476 }
477
478 if (wq && waitqueue_active(wq)) {
479 dprintk("RPC svc_write_space: someone sleeping on %p\n",
480 svsk);
481 wake_up_interruptible(wq);
482 }
483}
484
485static void svc_tcp_write_space(struct sock *sk)
486{
487 struct socket *sock = sk->sk_socket;
488
489 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
490 clear_bit(SOCK_NOSPACE, &sock->flags);
491 svc_write_space(sk);
492}
493
494/*
495 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
496 */
497static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
498 struct cmsghdr *cmh)
499{
500 struct in_pktinfo *pki = CMSG_DATA(cmh);
501 if (cmh->cmsg_type != IP_PKTINFO)
502 return 0;
503 rqstp->rq_daddr.addr.s_addr = pki->ipi_spec_dst.s_addr;
504 return 1;
505}
506
507/*
508 * See net/ipv6/datagram.c : datagram_recv_ctl
509 */
510static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
511 struct cmsghdr *cmh)
512{
513 struct in6_pktinfo *pki = CMSG_DATA(cmh);
514 if (cmh->cmsg_type != IPV6_PKTINFO)
515 return 0;
516 ipv6_addr_copy(&rqstp->rq_daddr.addr6, &pki->ipi6_addr);
517 return 1;
518}
519
520/*
521 * Copy the UDP datagram's destination address to the rqstp structure.
522 * The 'destination' address in this case is the address to which the
523 * peer sent the datagram, i.e. our local address. For multihomed
524 * hosts, this can change from msg to msg. Note that only the IP
525 * address changes, the port number should remain the same.
526 */
527static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
528 struct cmsghdr *cmh)
529{
530 switch (cmh->cmsg_level) {
531 case SOL_IP:
532 return svc_udp_get_dest_address4(rqstp, cmh);
533 case SOL_IPV6:
534 return svc_udp_get_dest_address6(rqstp, cmh);
535 }
536
537 return 0;
538}
539
540/*
541 * Receive a datagram from a UDP socket.
542 */
543static int svc_udp_recvfrom(struct svc_rqst *rqstp)
544{
545 struct svc_sock *svsk =
546 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
547 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
548 struct sk_buff *skb;
549 union {
550 struct cmsghdr hdr;
551 long all[SVC_PKTINFO_SPACE / sizeof(long)];
552 } buffer;
553 struct cmsghdr *cmh = &buffer.hdr;
554 struct msghdr msg = {
555 .msg_name = svc_addr(rqstp),
556 .msg_control = cmh,
557 .msg_controllen = sizeof(buffer),
558 .msg_flags = MSG_DONTWAIT,
559 };
560 size_t len;
561 int err;
562
563 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
564 /* udp sockets need large rcvbuf as all pending
565 * requests are still in that buffer. sndbuf must
566 * also be large enough that there is enough space
567 * for one reply per thread. We count all threads
568 * rather than threads in a particular pool, which
569 * provides an upper bound on the number of threads
570 * which will access the socket.
571 */
572 svc_sock_setbufsize(svsk->sk_sock,
573 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
574 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
575
576 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
577 skb = NULL;
578 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
579 0, 0, MSG_PEEK | MSG_DONTWAIT);
580 if (err >= 0)
581 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
582
583 if (skb == NULL) {
584 if (err != -EAGAIN) {
585 /* possibly an icmp error */
586 dprintk("svc: recvfrom returned error %d\n", -err);
587 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
588 }
589 return -EAGAIN;
590 }
591 len = svc_addr_len(svc_addr(rqstp));
592 if (len == 0)
593 return -EAFNOSUPPORT;
594 rqstp->rq_addrlen = len;
595 if (skb->tstamp.tv64 == 0) {
596 skb->tstamp = ktime_get_real();
597 /* Don't enable netstamp, sunrpc doesn't
598 need that much accuracy */
599 }
600 svsk->sk_sk->sk_stamp = skb->tstamp;
601 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
602
603 len = skb->len - sizeof(struct udphdr);
604 rqstp->rq_arg.len = len;
605
606 rqstp->rq_prot = IPPROTO_UDP;
607
608 if (!svc_udp_get_dest_address(rqstp, cmh)) {
609 if (net_ratelimit())
610 printk(KERN_WARNING
611 "svc: received unknown control message %d/%d; "
612 "dropping RPC reply datagram\n",
613 cmh->cmsg_level, cmh->cmsg_type);
614 skb_free_datagram_locked(svsk->sk_sk, skb);
615 return 0;
616 }
617
618 if (skb_is_nonlinear(skb)) {
619 /* we have to copy */
620 local_bh_disable();
621 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
622 local_bh_enable();
623 /* checksum error */
624 skb_free_datagram_locked(svsk->sk_sk, skb);
625 return 0;
626 }
627 local_bh_enable();
628 skb_free_datagram_locked(svsk->sk_sk, skb);
629 } else {
630 /* we can use it in-place */
631 rqstp->rq_arg.head[0].iov_base = skb->data +
632 sizeof(struct udphdr);
633 rqstp->rq_arg.head[0].iov_len = len;
634 if (skb_checksum_complete(skb)) {
635 skb_free_datagram_locked(svsk->sk_sk, skb);
636 return 0;
637 }
638 rqstp->rq_xprt_ctxt = skb;
639 }
640
641 rqstp->rq_arg.page_base = 0;
642 if (len <= rqstp->rq_arg.head[0].iov_len) {
643 rqstp->rq_arg.head[0].iov_len = len;
644 rqstp->rq_arg.page_len = 0;
645 rqstp->rq_respages = rqstp->rq_pages+1;
646 } else {
647 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
648 rqstp->rq_respages = rqstp->rq_pages + 1 +
649 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
650 }
651
652 if (serv->sv_stats)
653 serv->sv_stats->netudpcnt++;
654
655 return len;
656}
657
658static int
659svc_udp_sendto(struct svc_rqst *rqstp)
660{
661 int error;
662
663 error = svc_sendto(rqstp, &rqstp->rq_res);
664 if (error == -ECONNREFUSED)
665 /* ICMP error on earlier request. */
666 error = svc_sendto(rqstp, &rqstp->rq_res);
667
668 return error;
669}
670
671static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
672{
673}
674
675static int svc_udp_has_wspace(struct svc_xprt *xprt)
676{
677 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
678 struct svc_serv *serv = xprt->xpt_server;
679 unsigned long required;
680
681 /*
682 * Set the SOCK_NOSPACE flag before checking the available
683 * sock space.
684 */
685 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
686 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
687 if (required*2 > sock_wspace(svsk->sk_sk))
688 return 0;
689 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
690 return 1;
691}
692
693static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
694{
695 BUG();
696 return NULL;
697}
698
699static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
700 struct net *net,
701 struct sockaddr *sa, int salen,
702 int flags)
703{
704 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
705}
706
707static struct svc_xprt_ops svc_udp_ops = {
708 .xpo_create = svc_udp_create,
709 .xpo_recvfrom = svc_udp_recvfrom,
710 .xpo_sendto = svc_udp_sendto,
711 .xpo_release_rqst = svc_release_skb,
712 .xpo_detach = svc_sock_detach,
713 .xpo_free = svc_sock_free,
714 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
715 .xpo_has_wspace = svc_udp_has_wspace,
716 .xpo_accept = svc_udp_accept,
717};
718
719static struct svc_xprt_class svc_udp_class = {
720 .xcl_name = "udp",
721 .xcl_owner = THIS_MODULE,
722 .xcl_ops = &svc_udp_ops,
723 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
724};
725
726static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
727{
728 int err, level, optname, one = 1;
729
730 svc_xprt_init(&svc_udp_class, &svsk->sk_xprt, serv);
731 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
732 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
733 svsk->sk_sk->sk_write_space = svc_write_space;
734
735 /* initialise setting must have enough space to
736 * receive and respond to one request.
737 * svc_udp_recvfrom will re-adjust if necessary
738 */
739 svc_sock_setbufsize(svsk->sk_sock,
740 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
741 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
742
743 /* data might have come in before data_ready set up */
744 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
745 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
746
747 /* make sure we get destination address info */
748 switch (svsk->sk_sk->sk_family) {
749 case AF_INET:
750 level = SOL_IP;
751 optname = IP_PKTINFO;
752 break;
753 case AF_INET6:
754 level = SOL_IPV6;
755 optname = IPV6_RECVPKTINFO;
756 break;
757 default:
758 BUG();
759 }
760 err = kernel_setsockopt(svsk->sk_sock, level, optname,
761 (char *)&one, sizeof(one));
762 dprintk("svc: kernel_setsockopt returned %d\n", err);
763}
764
765/*
766 * A data_ready event on a listening socket means there's a connection
767 * pending. Do not use state_change as a substitute for it.
768 */
769static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
770{
771 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
772 wait_queue_head_t *wq;
773
774 dprintk("svc: socket %p TCP (listen) state change %d\n",
775 sk, sk->sk_state);
776
777 /*
778 * This callback may called twice when a new connection
779 * is established as a child socket inherits everything
780 * from a parent LISTEN socket.
781 * 1) data_ready method of the parent socket will be called
782 * when one of child sockets become ESTABLISHED.
783 * 2) data_ready method of the child socket may be called
784 * when it receives data before the socket is accepted.
785 * In case of 2, we should ignore it silently.
786 */
787 if (sk->sk_state == TCP_LISTEN) {
788 if (svsk) {
789 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
790 svc_xprt_enqueue(&svsk->sk_xprt);
791 } else
792 printk("svc: socket %p: no user data\n", sk);
793 }
794
795 wq = sk_sleep(sk);
796 if (wq && waitqueue_active(wq))
797 wake_up_interruptible_all(wq);
798}
799
800/*
801 * A state change on a connected socket means it's dying or dead.
802 */
803static void svc_tcp_state_change(struct sock *sk)
804{
805 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
806 wait_queue_head_t *wq = sk_sleep(sk);
807
808 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
809 sk, sk->sk_state, sk->sk_user_data);
810
811 if (!svsk)
812 printk("svc: socket %p: no user data\n", sk);
813 else {
814 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
815 svc_xprt_enqueue(&svsk->sk_xprt);
816 }
817 if (wq && waitqueue_active(wq))
818 wake_up_interruptible_all(wq);
819}
820
821static void svc_tcp_data_ready(struct sock *sk, int count)
822{
823 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
824 wait_queue_head_t *wq = sk_sleep(sk);
825
826 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
827 sk, sk->sk_user_data);
828 if (svsk) {
829 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
830 svc_xprt_enqueue(&svsk->sk_xprt);
831 }
832 if (wq && waitqueue_active(wq))
833 wake_up_interruptible(wq);
834}
835
836/*
837 * Accept a TCP connection
838 */
839static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
840{
841 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
842 struct sockaddr_storage addr;
843 struct sockaddr *sin = (struct sockaddr *) &addr;
844 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
845 struct socket *sock = svsk->sk_sock;
846 struct socket *newsock;
847 struct svc_sock *newsvsk;
848 int err, slen;
849 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
850
851 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
852 if (!sock)
853 return NULL;
854
855 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
856 err = kernel_accept(sock, &newsock, O_NONBLOCK);
857 if (err < 0) {
858 if (err == -ENOMEM)
859 printk(KERN_WARNING "%s: no more sockets!\n",
860 serv->sv_name);
861 else if (err != -EAGAIN && net_ratelimit())
862 printk(KERN_WARNING "%s: accept failed (err %d)!\n",
863 serv->sv_name, -err);
864 return NULL;
865 }
866 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
867
868 err = kernel_getpeername(newsock, sin, &slen);
869 if (err < 0) {
870 if (net_ratelimit())
871 printk(KERN_WARNING "%s: peername failed (err %d)!\n",
872 serv->sv_name, -err);
873 goto failed; /* aborted connection or whatever */
874 }
875
876 /* Ideally, we would want to reject connections from unauthorized
877 * hosts here, but when we get encryption, the IP of the host won't
878 * tell us anything. For now just warn about unpriv connections.
879 */
880 if (!svc_port_is_privileged(sin)) {
881 dprintk(KERN_WARNING
882 "%s: connect from unprivileged port: %s\n",
883 serv->sv_name,
884 __svc_print_addr(sin, buf, sizeof(buf)));
885 }
886 dprintk("%s: connect from %s\n", serv->sv_name,
887 __svc_print_addr(sin, buf, sizeof(buf)));
888
889 /* make sure that a write doesn't block forever when
890 * low on memory
891 */
892 newsock->sk->sk_sndtimeo = HZ*30;
893
894 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
895 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
896 goto failed;
897 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
898 err = kernel_getsockname(newsock, sin, &slen);
899 if (unlikely(err < 0)) {
900 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
901 slen = offsetof(struct sockaddr, sa_data);
902 }
903 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
904
905 if (serv->sv_stats)
906 serv->sv_stats->nettcpconn++;
907
908 return &newsvsk->sk_xprt;
909
910failed:
911 sock_release(newsock);
912 return NULL;
913}
914
915static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
916{
917 unsigned int i, len, npages;
918
919 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
920 return 0;
921 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
922 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
923 for (i = 0; i < npages; i++) {
924 if (rqstp->rq_pages[i] != NULL)
925 put_page(rqstp->rq_pages[i]);
926 BUG_ON(svsk->sk_pages[i] == NULL);
927 rqstp->rq_pages[i] = svsk->sk_pages[i];
928 svsk->sk_pages[i] = NULL;
929 }
930 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
931 return len;
932}
933
934static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
935{
936 unsigned int i, len, npages;
937
938 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
939 return;
940 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
941 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
942 for (i = 0; i < npages; i++) {
943 svsk->sk_pages[i] = rqstp->rq_pages[i];
944 rqstp->rq_pages[i] = NULL;
945 }
946}
947
948static void svc_tcp_clear_pages(struct svc_sock *svsk)
949{
950 unsigned int i, len, npages;
951
952 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
953 goto out;
954 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
955 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
956 for (i = 0; i < npages; i++) {
957 BUG_ON(svsk->sk_pages[i] == NULL);
958 put_page(svsk->sk_pages[i]);
959 svsk->sk_pages[i] = NULL;
960 }
961out:
962 svsk->sk_tcplen = 0;
963}
964
965/*
966 * Receive data.
967 * If we haven't gotten the record length yet, get the next four bytes.
968 * Otherwise try to gobble up as much as possible up to the complete
969 * record length.
970 */
971static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
972{
973 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
974 unsigned int want;
975 int len;
976
977 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
978
979 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
980 struct kvec iov;
981
982 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
983 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
984 iov.iov_len = want;
985 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
986 goto error;
987 svsk->sk_tcplen += len;
988
989 if (len < want) {
990 dprintk("svc: short recvfrom while reading record "
991 "length (%d of %d)\n", len, want);
992 return -EAGAIN;
993 }
994
995 svsk->sk_reclen = ntohl(svsk->sk_reclen);
996 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
997 /* FIXME: technically, a record can be fragmented,
998 * and non-terminal fragments will not have the top
999 * bit set in the fragment length header.
1000 * But apparently no known nfs clients send fragmented
1001 * records. */
1002 if (net_ratelimit())
1003 printk(KERN_NOTICE "RPC: multiple fragments "
1004 "per record not supported\n");
1005 goto err_delete;
1006 }
1007
1008 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1009 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1010 if (svsk->sk_reclen > serv->sv_max_mesg) {
1011 if (net_ratelimit())
1012 printk(KERN_NOTICE "RPC: "
1013 "fragment too large: 0x%08lx\n",
1014 (unsigned long)svsk->sk_reclen);
1015 goto err_delete;
1016 }
1017 }
1018
1019 if (svsk->sk_reclen < 8)
1020 goto err_delete; /* client is nuts. */
1021
1022 len = svsk->sk_reclen;
1023
1024 return len;
1025error:
1026 dprintk("RPC: TCP recv_record got %d\n", len);
1027 return len;
1028err_delete:
1029 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1030 return -EAGAIN;
1031}
1032
1033static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1034{
1035 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1036 struct rpc_rqst *req = NULL;
1037 struct kvec *src, *dst;
1038 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1039 __be32 xid;
1040 __be32 calldir;
1041
1042 xid = *p++;
1043 calldir = *p;
1044
1045 if (bc_xprt)
1046 req = xprt_lookup_rqst(bc_xprt, xid);
1047
1048 if (!req) {
1049 printk(KERN_NOTICE
1050 "%s: Got unrecognized reply: "
1051 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1052 __func__, ntohl(calldir),
1053 bc_xprt, xid);
1054 return -EAGAIN;
1055 }
1056
1057 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1058 /*
1059 * XXX!: cheating for now! Only copying HEAD.
1060 * But we know this is good enough for now (in fact, for any
1061 * callback reply in the forseeable future).
1062 */
1063 dst = &req->rq_private_buf.head[0];
1064 src = &rqstp->rq_arg.head[0];
1065 if (dst->iov_len < src->iov_len)
1066 return -EAGAIN; /* whatever; just giving up. */
1067 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1068 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1069 rqstp->rq_arg.len = 0;
1070 return 0;
1071}
1072
1073static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1074{
1075 int i = 0;
1076 int t = 0;
1077
1078 while (t < len) {
1079 vec[i].iov_base = page_address(pages[i]);
1080 vec[i].iov_len = PAGE_SIZE;
1081 i++;
1082 t += PAGE_SIZE;
1083 }
1084 return i;
1085}
1086
1087
1088/*
1089 * Receive data from a TCP socket.
1090 */
1091static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1092{
1093 struct svc_sock *svsk =
1094 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1095 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1096 int len;
1097 struct kvec *vec;
1098 unsigned int want, base;
1099 __be32 *p;
1100 __be32 calldir;
1101 int pnum;
1102
1103 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1104 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1105 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1106 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1107
1108 len = svc_tcp_recv_record(svsk, rqstp);
1109 if (len < 0)
1110 goto error;
1111
1112 base = svc_tcp_restore_pages(svsk, rqstp);
1113 want = svsk->sk_reclen - base;
1114
1115 vec = rqstp->rq_vec;
1116
1117 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1118 svsk->sk_reclen);
1119
1120 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1121
1122 /* Now receive data */
1123 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1124 if (len >= 0)
1125 svsk->sk_tcplen += len;
1126 if (len != want) {
1127 if (len < 0 && len != -EAGAIN)
1128 goto err_other;
1129 svc_tcp_save_pages(svsk, rqstp);
1130 dprintk("svc: incomplete TCP record (%d of %d)\n",
1131 svsk->sk_tcplen, svsk->sk_reclen);
1132 goto err_noclose;
1133 }
1134
1135 rqstp->rq_arg.len = svsk->sk_reclen;
1136 rqstp->rq_arg.page_base = 0;
1137 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1138 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1139 rqstp->rq_arg.page_len = 0;
1140 } else
1141 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1142
1143 rqstp->rq_xprt_ctxt = NULL;
1144 rqstp->rq_prot = IPPROTO_TCP;
1145
1146 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1147 calldir = p[1];
1148 if (calldir)
1149 len = receive_cb_reply(svsk, rqstp);
1150
1151 /* Reset TCP read info */
1152 svsk->sk_reclen = 0;
1153 svsk->sk_tcplen = 0;
1154 /* If we have more data, signal svc_xprt_enqueue() to try again */
1155 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1156 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1157
1158 if (len < 0)
1159 goto error;
1160
1161 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1162 if (serv->sv_stats)
1163 serv->sv_stats->nettcpcnt++;
1164
1165 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1166 return rqstp->rq_arg.len;
1167
1168error:
1169 if (len != -EAGAIN)
1170 goto err_other;
1171 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1172 return -EAGAIN;
1173err_other:
1174 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1175 svsk->sk_xprt.xpt_server->sv_name, -len);
1176 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1177err_noclose:
1178 return -EAGAIN; /* record not complete */
1179}
1180
1181/*
1182 * Send out data on TCP socket.
1183 */
1184static int svc_tcp_sendto(struct svc_rqst *rqstp)
1185{
1186 struct xdr_buf *xbufp = &rqstp->rq_res;
1187 int sent;
1188 __be32 reclen;
1189
1190 /* Set up the first element of the reply kvec.
1191 * Any other kvecs that may be in use have been taken
1192 * care of by the server implementation itself.
1193 */
1194 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1195 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1196
1197 sent = svc_sendto(rqstp, &rqstp->rq_res);
1198 if (sent != xbufp->len) {
1199 printk(KERN_NOTICE
1200 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1201 "- shutting down socket\n",
1202 rqstp->rq_xprt->xpt_server->sv_name,
1203 (sent<0)?"got error":"sent only",
1204 sent, xbufp->len);
1205 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1206 svc_xprt_enqueue(rqstp->rq_xprt);
1207 sent = -EAGAIN;
1208 }
1209 return sent;
1210}
1211
1212/*
1213 * Setup response header. TCP has a 4B record length field.
1214 */
1215static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1216{
1217 struct kvec *resv = &rqstp->rq_res.head[0];
1218
1219 /* tcp needs a space for the record length... */
1220 svc_putnl(resv, 0);
1221}
1222
1223static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1224{
1225 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1226 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1227 int required;
1228
1229 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1230 return 1;
1231 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1232 if (sk_stream_wspace(svsk->sk_sk) >= required)
1233 return 1;
1234 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1235 return 0;
1236}
1237
1238static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1239 struct net *net,
1240 struct sockaddr *sa, int salen,
1241 int flags)
1242{
1243 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1244}
1245
1246#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1247static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1248 struct net *, struct sockaddr *,
1249 int, int);
1250static void svc_bc_sock_free(struct svc_xprt *xprt);
1251
1252static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1253 struct net *net,
1254 struct sockaddr *sa, int salen,
1255 int flags)
1256{
1257 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1258}
1259
1260static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1261{
1262}
1263
1264static struct svc_xprt_ops svc_tcp_bc_ops = {
1265 .xpo_create = svc_bc_tcp_create,
1266 .xpo_detach = svc_bc_tcp_sock_detach,
1267 .xpo_free = svc_bc_sock_free,
1268 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1269};
1270
1271static struct svc_xprt_class svc_tcp_bc_class = {
1272 .xcl_name = "tcp-bc",
1273 .xcl_owner = THIS_MODULE,
1274 .xcl_ops = &svc_tcp_bc_ops,
1275 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1276};
1277
1278static void svc_init_bc_xprt_sock(void)
1279{
1280 svc_reg_xprt_class(&svc_tcp_bc_class);
1281}
1282
1283static void svc_cleanup_bc_xprt_sock(void)
1284{
1285 svc_unreg_xprt_class(&svc_tcp_bc_class);
1286}
1287#else /* CONFIG_SUNRPC_BACKCHANNEL */
1288static void svc_init_bc_xprt_sock(void)
1289{
1290}
1291
1292static void svc_cleanup_bc_xprt_sock(void)
1293{
1294}
1295#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1296
1297static struct svc_xprt_ops svc_tcp_ops = {
1298 .xpo_create = svc_tcp_create,
1299 .xpo_recvfrom = svc_tcp_recvfrom,
1300 .xpo_sendto = svc_tcp_sendto,
1301 .xpo_release_rqst = svc_release_skb,
1302 .xpo_detach = svc_tcp_sock_detach,
1303 .xpo_free = svc_sock_free,
1304 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1305 .xpo_has_wspace = svc_tcp_has_wspace,
1306 .xpo_accept = svc_tcp_accept,
1307};
1308
1309static struct svc_xprt_class svc_tcp_class = {
1310 .xcl_name = "tcp",
1311 .xcl_owner = THIS_MODULE,
1312 .xcl_ops = &svc_tcp_ops,
1313 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1314};
1315
1316void svc_init_xprt_sock(void)
1317{
1318 svc_reg_xprt_class(&svc_tcp_class);
1319 svc_reg_xprt_class(&svc_udp_class);
1320 svc_init_bc_xprt_sock();
1321}
1322
1323void svc_cleanup_xprt_sock(void)
1324{
1325 svc_unreg_xprt_class(&svc_tcp_class);
1326 svc_unreg_xprt_class(&svc_udp_class);
1327 svc_cleanup_bc_xprt_sock();
1328}
1329
1330static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1331{
1332 struct sock *sk = svsk->sk_sk;
1333
1334 svc_xprt_init(&svc_tcp_class, &svsk->sk_xprt, serv);
1335 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1336 if (sk->sk_state == TCP_LISTEN) {
1337 dprintk("setting up TCP socket for listening\n");
1338 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1339 sk->sk_data_ready = svc_tcp_listen_data_ready;
1340 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1341 } else {
1342 dprintk("setting up TCP socket for reading\n");
1343 sk->sk_state_change = svc_tcp_state_change;
1344 sk->sk_data_ready = svc_tcp_data_ready;
1345 sk->sk_write_space = svc_tcp_write_space;
1346
1347 svsk->sk_reclen = 0;
1348 svsk->sk_tcplen = 0;
1349 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1350
1351 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1352
1353 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1354 if (sk->sk_state != TCP_ESTABLISHED)
1355 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1356 }
1357}
1358
1359void svc_sock_update_bufs(struct svc_serv *serv)
1360{
1361 /*
1362 * The number of server threads has changed. Update
1363 * rcvbuf and sndbuf accordingly on all sockets
1364 */
1365 struct svc_sock *svsk;
1366
1367 spin_lock_bh(&serv->sv_lock);
1368 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1369 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1370 list_for_each_entry(svsk, &serv->sv_tempsocks, sk_xprt.xpt_list)
1371 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1372 spin_unlock_bh(&serv->sv_lock);
1373}
1374EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1375
1376/*
1377 * Initialize socket for RPC use and create svc_sock struct
1378 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1379 */
1380static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1381 struct socket *sock,
1382 int *errp, int flags)
1383{
1384 struct svc_sock *svsk;
1385 struct sock *inet;
1386 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1387
1388 dprintk("svc: svc_setup_socket %p\n", sock);
1389 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1390 *errp = -ENOMEM;
1391 return NULL;
1392 }
1393
1394 inet = sock->sk;
1395
1396 /* Register socket with portmapper */
1397 if (*errp >= 0 && pmap_register)
1398 *errp = svc_register(serv, inet->sk_family, inet->sk_protocol,
1399 ntohs(inet_sk(inet)->inet_sport));
1400
1401 if (*errp < 0) {
1402 kfree(svsk);
1403 return NULL;
1404 }
1405
1406 inet->sk_user_data = svsk;
1407 svsk->sk_sock = sock;
1408 svsk->sk_sk = inet;
1409 svsk->sk_ostate = inet->sk_state_change;
1410 svsk->sk_odata = inet->sk_data_ready;
1411 svsk->sk_owspace = inet->sk_write_space;
1412
1413 /* Initialize the socket */
1414 if (sock->type == SOCK_DGRAM)
1415 svc_udp_init(svsk, serv);
1416 else {
1417 /* initialise setting must have enough space to
1418 * receive and respond to one request.
1419 */
1420 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1421 4 * serv->sv_max_mesg);
1422 svc_tcp_init(svsk, serv);
1423 }
1424
1425 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1426 svsk, svsk->sk_sk);
1427
1428 return svsk;
1429}
1430
1431/**
1432 * svc_addsock - add a listener socket to an RPC service
1433 * @serv: pointer to RPC service to which to add a new listener
1434 * @fd: file descriptor of the new listener
1435 * @name_return: pointer to buffer to fill in with name of listener
1436 * @len: size of the buffer
1437 *
1438 * Fills in socket name and returns positive length of name if successful.
1439 * Name is terminated with '\n'. On error, returns a negative errno
1440 * value.
1441 */
1442int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1443 const size_t len)
1444{
1445 int err = 0;
1446 struct socket *so = sockfd_lookup(fd, &err);
1447 struct svc_sock *svsk = NULL;
1448
1449 if (!so)
1450 return err;
1451 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1452 err = -EAFNOSUPPORT;
1453 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1454 so->sk->sk_protocol != IPPROTO_UDP)
1455 err = -EPROTONOSUPPORT;
1456 else if (so->state > SS_UNCONNECTED)
1457 err = -EISCONN;
1458 else {
1459 if (!try_module_get(THIS_MODULE))
1460 err = -ENOENT;
1461 else
1462 svsk = svc_setup_socket(serv, so, &err,
1463 SVC_SOCK_DEFAULTS);
1464 if (svsk) {
1465 struct sockaddr_storage addr;
1466 struct sockaddr *sin = (struct sockaddr *)&addr;
1467 int salen;
1468 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1469 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1470 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1471 spin_lock_bh(&serv->sv_lock);
1472 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1473 spin_unlock_bh(&serv->sv_lock);
1474 svc_xprt_received(&svsk->sk_xprt);
1475 err = 0;
1476 } else
1477 module_put(THIS_MODULE);
1478 }
1479 if (err) {
1480 sockfd_put(so);
1481 return err;
1482 }
1483 return svc_one_sock_name(svsk, name_return, len);
1484}
1485EXPORT_SYMBOL_GPL(svc_addsock);
1486
1487/*
1488 * Create socket for RPC service.
1489 */
1490static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1491 int protocol,
1492 struct net *net,
1493 struct sockaddr *sin, int len,
1494 int flags)
1495{
1496 struct svc_sock *svsk;
1497 struct socket *sock;
1498 int error;
1499 int type;
1500 struct sockaddr_storage addr;
1501 struct sockaddr *newsin = (struct sockaddr *)&addr;
1502 int newlen;
1503 int family;
1504 int val;
1505 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1506
1507 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1508 serv->sv_program->pg_name, protocol,
1509 __svc_print_addr(sin, buf, sizeof(buf)));
1510
1511 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1512 printk(KERN_WARNING "svc: only UDP and TCP "
1513 "sockets supported\n");
1514 return ERR_PTR(-EINVAL);
1515 }
1516
1517 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1518 switch (sin->sa_family) {
1519 case AF_INET6:
1520 family = PF_INET6;
1521 break;
1522 case AF_INET:
1523 family = PF_INET;
1524 break;
1525 default:
1526 return ERR_PTR(-EINVAL);
1527 }
1528
1529 error = __sock_create(net, family, type, protocol, &sock, 1);
1530 if (error < 0)
1531 return ERR_PTR(error);
1532
1533 svc_reclassify_socket(sock);
1534
1535 /*
1536 * If this is an PF_INET6 listener, we want to avoid
1537 * getting requests from IPv4 remotes. Those should
1538 * be shunted to a PF_INET listener via rpcbind.
1539 */
1540 val = 1;
1541 if (family == PF_INET6)
1542 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1543 (char *)&val, sizeof(val));
1544
1545 if (type == SOCK_STREAM)
1546 sock->sk->sk_reuse = 1; /* allow address reuse */
1547 error = kernel_bind(sock, sin, len);
1548 if (error < 0)
1549 goto bummer;
1550
1551 newlen = len;
1552 error = kernel_getsockname(sock, newsin, &newlen);
1553 if (error < 0)
1554 goto bummer;
1555
1556 if (protocol == IPPROTO_TCP) {
1557 if ((error = kernel_listen(sock, 64)) < 0)
1558 goto bummer;
1559 }
1560
1561 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1562 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1563 return (struct svc_xprt *)svsk;
1564 }
1565
1566bummer:
1567 dprintk("svc: svc_create_socket error = %d\n", -error);
1568 sock_release(sock);
1569 return ERR_PTR(error);
1570}
1571
1572/*
1573 * Detach the svc_sock from the socket so that no
1574 * more callbacks occur.
1575 */
1576static void svc_sock_detach(struct svc_xprt *xprt)
1577{
1578 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1579 struct sock *sk = svsk->sk_sk;
1580 wait_queue_head_t *wq;
1581
1582 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1583
1584 /* put back the old socket callbacks */
1585 sk->sk_state_change = svsk->sk_ostate;
1586 sk->sk_data_ready = svsk->sk_odata;
1587 sk->sk_write_space = svsk->sk_owspace;
1588
1589 wq = sk_sleep(sk);
1590 if (wq && waitqueue_active(wq))
1591 wake_up_interruptible(wq);
1592}
1593
1594/*
1595 * Disconnect the socket, and reset the callbacks
1596 */
1597static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1598{
1599 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1600
1601 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1602
1603 svc_sock_detach(xprt);
1604
1605 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1606 svc_tcp_clear_pages(svsk);
1607 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1608 }
1609}
1610
1611/*
1612 * Free the svc_sock's socket resources and the svc_sock itself.
1613 */
1614static void svc_sock_free(struct svc_xprt *xprt)
1615{
1616 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1617 dprintk("svc: svc_sock_free(%p)\n", svsk);
1618
1619 if (svsk->sk_sock->file)
1620 sockfd_put(svsk->sk_sock);
1621 else
1622 sock_release(svsk->sk_sock);
1623 kfree(svsk);
1624}
1625
1626#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1627/*
1628 * Create a back channel svc_xprt which shares the fore channel socket.
1629 */
1630static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1631 int protocol,
1632 struct net *net,
1633 struct sockaddr *sin, int len,
1634 int flags)
1635{
1636 struct svc_sock *svsk;
1637 struct svc_xprt *xprt;
1638
1639 if (protocol != IPPROTO_TCP) {
1640 printk(KERN_WARNING "svc: only TCP sockets"
1641 " supported on shared back channel\n");
1642 return ERR_PTR(-EINVAL);
1643 }
1644
1645 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1646 if (!svsk)
1647 return ERR_PTR(-ENOMEM);
1648
1649 xprt = &svsk->sk_xprt;
1650 svc_xprt_init(&svc_tcp_bc_class, xprt, serv);
1651
1652 serv->sv_bc_xprt = xprt;
1653
1654 return xprt;
1655}
1656
1657/*
1658 * Free a back channel svc_sock.
1659 */
1660static void svc_bc_sock_free(struct svc_xprt *xprt)
1661{
1662 if (xprt)
1663 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1664}
1665#endif /* CONFIG_SUNRPC_BACKCHANNEL */