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 <net/sock.h>
40#include <net/checksum.h>
41#include <net/ip.h>
42#include <net/ipv6.h>
43#include <net/udp.h>
44#include <net/tcp.h>
45#include <net/tcp_states.h>
46#include <linux/uaccess.h>
47#include <linux/highmem.h>
48#include <asm/ioctls.h>
49
50#include <linux/sunrpc/types.h>
51#include <linux/sunrpc/clnt.h>
52#include <linux/sunrpc/xdr.h>
53#include <linux/sunrpc/msg_prot.h>
54#include <linux/sunrpc/svcsock.h>
55#include <linux/sunrpc/stats.h>
56#include <linux/sunrpc/xprt.h>
57
58#include <trace/events/sunrpc.h>
59
60#include "socklib.h"
61#include "sunrpc.h"
62
63#define RPCDBG_FACILITY RPCDBG_SVCXPRT
64
65
66static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
67 int flags);
68static int svc_udp_recvfrom(struct svc_rqst *);
69static int svc_udp_sendto(struct svc_rqst *);
70static void svc_sock_detach(struct svc_xprt *);
71static void svc_tcp_sock_detach(struct svc_xprt *);
72static void svc_sock_free(struct svc_xprt *);
73
74static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77#ifdef CONFIG_DEBUG_LOCK_ALLOC
78static struct lock_class_key svc_key[2];
79static struct lock_class_key svc_slock_key[2];
80
81static void svc_reclassify_socket(struct socket *sock)
82{
83 struct sock *sk = sock->sk;
84
85 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
86 return;
87
88 switch (sk->sk_family) {
89 case AF_INET:
90 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
91 &svc_slock_key[0],
92 "sk_xprt.xpt_lock-AF_INET-NFSD",
93 &svc_key[0]);
94 break;
95
96 case AF_INET6:
97 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
98 &svc_slock_key[1],
99 "sk_xprt.xpt_lock-AF_INET6-NFSD",
100 &svc_key[1]);
101 break;
102
103 default:
104 BUG();
105 }
106}
107#else
108static void svc_reclassify_socket(struct socket *sock)
109{
110}
111#endif
112
113/**
114 * svc_tcp_release_rqst - Release transport-related resources
115 * @rqstp: request structure with resources to be released
116 *
117 */
118static void svc_tcp_release_rqst(struct svc_rqst *rqstp)
119{
120}
121
122/**
123 * svc_udp_release_rqst - Release transport-related resources
124 * @rqstp: request structure with resources to be released
125 *
126 */
127static void svc_udp_release_rqst(struct svc_rqst *rqstp)
128{
129 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
130
131 if (skb) {
132 rqstp->rq_xprt_ctxt = NULL;
133 consume_skb(skb);
134 }
135}
136
137union svc_pktinfo_u {
138 struct in_pktinfo pkti;
139 struct in6_pktinfo pkti6;
140};
141#define SVC_PKTINFO_SPACE \
142 CMSG_SPACE(sizeof(union svc_pktinfo_u))
143
144static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
145{
146 struct svc_sock *svsk =
147 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
148 switch (svsk->sk_sk->sk_family) {
149 case AF_INET: {
150 struct in_pktinfo *pki = CMSG_DATA(cmh);
151
152 cmh->cmsg_level = SOL_IP;
153 cmh->cmsg_type = IP_PKTINFO;
154 pki->ipi_ifindex = 0;
155 pki->ipi_spec_dst.s_addr =
156 svc_daddr_in(rqstp)->sin_addr.s_addr;
157 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
158 }
159 break;
160
161 case AF_INET6: {
162 struct in6_pktinfo *pki = CMSG_DATA(cmh);
163 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
164
165 cmh->cmsg_level = SOL_IPV6;
166 cmh->cmsg_type = IPV6_PKTINFO;
167 pki->ipi6_ifindex = daddr->sin6_scope_id;
168 pki->ipi6_addr = daddr->sin6_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
170 }
171 break;
172 }
173}
174
175static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
176 unsigned int length)
177{
178 return 0;
179}
180
181/*
182 * Report socket names for nfsdfs
183 */
184static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
185{
186 const struct sock *sk = svsk->sk_sk;
187 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
188 "udp" : "tcp";
189 int len;
190
191 switch (sk->sk_family) {
192 case PF_INET:
193 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
194 proto_name,
195 &inet_sk(sk)->inet_rcv_saddr,
196 inet_sk(sk)->inet_num);
197 break;
198#if IS_ENABLED(CONFIG_IPV6)
199 case PF_INET6:
200 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
201 proto_name,
202 &sk->sk_v6_rcv_saddr,
203 inet_sk(sk)->inet_num);
204 break;
205#endif
206 default:
207 len = snprintf(buf, remaining, "*unknown-%d*\n",
208 sk->sk_family);
209 }
210
211 if (len >= remaining) {
212 *buf = '\0';
213 return -ENAMETOOLONG;
214 }
215 return len;
216}
217
218#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
219static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
220{
221 struct bvec_iter bi = {
222 .bi_size = size + seek,
223 };
224 struct bio_vec bv;
225
226 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
227 for_each_bvec(bv, bvec, bi, bi)
228 flush_dcache_page(bv.bv_page);
229}
230#else
231static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
232 size_t seek)
233{
234}
235#endif
236
237/*
238 * Read from @rqstp's transport socket. The incoming message fills whole
239 * pages in @rqstp's rq_pages array until the last page of the message
240 * has been received into a partial page.
241 */
242static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
243 size_t seek)
244{
245 struct svc_sock *svsk =
246 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
247 struct bio_vec *bvec = rqstp->rq_bvec;
248 struct msghdr msg = { NULL };
249 unsigned int i;
250 ssize_t len;
251 size_t t;
252
253 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
254
255 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) {
256 bvec[i].bv_page = rqstp->rq_pages[i];
257 bvec[i].bv_len = PAGE_SIZE;
258 bvec[i].bv_offset = 0;
259 }
260 rqstp->rq_respages = &rqstp->rq_pages[i];
261 rqstp->rq_next_page = rqstp->rq_respages + 1;
262
263 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen);
264 if (seek) {
265 iov_iter_advance(&msg.msg_iter, seek);
266 buflen -= seek;
267 }
268 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
269 if (len > 0)
270 svc_flush_bvec(bvec, len, seek);
271
272 /* If we read a full record, then assume there may be more
273 * data to read (stream based sockets only!)
274 */
275 if (len == buflen)
276 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
277
278 return len;
279}
280
281/*
282 * Set socket snd and rcv buffer lengths
283 */
284static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
285{
286 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
287 struct socket *sock = svsk->sk_sock;
288
289 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
290
291 lock_sock(sock->sk);
292 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
293 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
294 sock->sk->sk_write_space(sock->sk);
295 release_sock(sock->sk);
296}
297
298static void svc_sock_secure_port(struct svc_rqst *rqstp)
299{
300 if (svc_port_is_privileged(svc_addr(rqstp)))
301 set_bit(RQ_SECURE, &rqstp->rq_flags);
302 else
303 clear_bit(RQ_SECURE, &rqstp->rq_flags);
304}
305
306/*
307 * INET callback when data has been received on the socket.
308 */
309static void svc_data_ready(struct sock *sk)
310{
311 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
312
313 if (svsk) {
314 /* Refer to svc_setup_socket() for details. */
315 rmb();
316 svsk->sk_odata(sk);
317 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
318 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
319 svc_xprt_enqueue(&svsk->sk_xprt);
320 }
321}
322
323/*
324 * INET callback when space is newly available on the socket.
325 */
326static void svc_write_space(struct sock *sk)
327{
328 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
329
330 if (svsk) {
331 /* Refer to svc_setup_socket() for details. */
332 rmb();
333 trace_svcsock_write_space(&svsk->sk_xprt, 0);
334 svsk->sk_owspace(sk);
335 svc_xprt_enqueue(&svsk->sk_xprt);
336 }
337}
338
339static int svc_tcp_has_wspace(struct svc_xprt *xprt)
340{
341 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
342
343 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
344 return 1;
345 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
346}
347
348static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
349{
350 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
351
352 sock_no_linger(svsk->sk_sock->sk);
353}
354
355/*
356 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
357 */
358static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
359 struct cmsghdr *cmh)
360{
361 struct in_pktinfo *pki = CMSG_DATA(cmh);
362 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
363
364 if (cmh->cmsg_type != IP_PKTINFO)
365 return 0;
366
367 daddr->sin_family = AF_INET;
368 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
369 return 1;
370}
371
372/*
373 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
374 */
375static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
376 struct cmsghdr *cmh)
377{
378 struct in6_pktinfo *pki = CMSG_DATA(cmh);
379 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
380
381 if (cmh->cmsg_type != IPV6_PKTINFO)
382 return 0;
383
384 daddr->sin6_family = AF_INET6;
385 daddr->sin6_addr = pki->ipi6_addr;
386 daddr->sin6_scope_id = pki->ipi6_ifindex;
387 return 1;
388}
389
390/*
391 * Copy the UDP datagram's destination address to the rqstp structure.
392 * The 'destination' address in this case is the address to which the
393 * peer sent the datagram, i.e. our local address. For multihomed
394 * hosts, this can change from msg to msg. Note that only the IP
395 * address changes, the port number should remain the same.
396 */
397static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
398 struct cmsghdr *cmh)
399{
400 switch (cmh->cmsg_level) {
401 case SOL_IP:
402 return svc_udp_get_dest_address4(rqstp, cmh);
403 case SOL_IPV6:
404 return svc_udp_get_dest_address6(rqstp, cmh);
405 }
406
407 return 0;
408}
409
410/**
411 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
412 * @rqstp: request structure into which to receive an RPC Call
413 *
414 * Called in a loop when XPT_DATA has been set.
415 *
416 * Returns:
417 * On success, the number of bytes in a received RPC Call, or
418 * %0 if a complete RPC Call message was not ready to return
419 */
420static int svc_udp_recvfrom(struct svc_rqst *rqstp)
421{
422 struct svc_sock *svsk =
423 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
424 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
425 struct sk_buff *skb;
426 union {
427 struct cmsghdr hdr;
428 long all[SVC_PKTINFO_SPACE / sizeof(long)];
429 } buffer;
430 struct cmsghdr *cmh = &buffer.hdr;
431 struct msghdr msg = {
432 .msg_name = svc_addr(rqstp),
433 .msg_control = cmh,
434 .msg_controllen = sizeof(buffer),
435 .msg_flags = MSG_DONTWAIT,
436 };
437 size_t len;
438 int err;
439
440 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
441 /* udp sockets need large rcvbuf as all pending
442 * requests are still in that buffer. sndbuf must
443 * also be large enough that there is enough space
444 * for one reply per thread. We count all threads
445 * rather than threads in a particular pool, which
446 * provides an upper bound on the number of threads
447 * which will access the socket.
448 */
449 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
450
451 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
452 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
453 0, 0, MSG_PEEK | MSG_DONTWAIT);
454 if (err < 0)
455 goto out_recv_err;
456 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err);
457 if (!skb)
458 goto out_recv_err;
459
460 len = svc_addr_len(svc_addr(rqstp));
461 rqstp->rq_addrlen = len;
462 if (skb->tstamp == 0) {
463 skb->tstamp = ktime_get_real();
464 /* Don't enable netstamp, sunrpc doesn't
465 need that much accuracy */
466 }
467 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
468 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
469
470 len = skb->len;
471 rqstp->rq_arg.len = len;
472 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
473
474 rqstp->rq_prot = IPPROTO_UDP;
475
476 if (!svc_udp_get_dest_address(rqstp, cmh))
477 goto out_cmsg_err;
478 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
479
480 if (skb_is_nonlinear(skb)) {
481 /* we have to copy */
482 local_bh_disable();
483 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
484 goto out_bh_enable;
485 local_bh_enable();
486 consume_skb(skb);
487 } else {
488 /* we can use it in-place */
489 rqstp->rq_arg.head[0].iov_base = skb->data;
490 rqstp->rq_arg.head[0].iov_len = len;
491 if (skb_checksum_complete(skb))
492 goto out_free;
493 rqstp->rq_xprt_ctxt = skb;
494 }
495
496 rqstp->rq_arg.page_base = 0;
497 if (len <= rqstp->rq_arg.head[0].iov_len) {
498 rqstp->rq_arg.head[0].iov_len = len;
499 rqstp->rq_arg.page_len = 0;
500 rqstp->rq_respages = rqstp->rq_pages+1;
501 } else {
502 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
503 rqstp->rq_respages = rqstp->rq_pages + 1 +
504 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
505 }
506 rqstp->rq_next_page = rqstp->rq_respages+1;
507
508 if (serv->sv_stats)
509 serv->sv_stats->netudpcnt++;
510
511 svc_xprt_received(rqstp->rq_xprt);
512 return len;
513
514out_recv_err:
515 if (err != -EAGAIN) {
516 /* possibly an icmp error */
517 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
518 }
519 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
520 goto out_clear_busy;
521out_cmsg_err:
522 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
523 cmh->cmsg_level, cmh->cmsg_type);
524 goto out_free;
525out_bh_enable:
526 local_bh_enable();
527out_free:
528 kfree_skb(skb);
529out_clear_busy:
530 svc_xprt_received(rqstp->rq_xprt);
531 return 0;
532}
533
534/**
535 * svc_udp_sendto - Send out a reply on a UDP socket
536 * @rqstp: completed svc_rqst
537 *
538 * xpt_mutex ensures @rqstp's whole message is written to the socket
539 * without interruption.
540 *
541 * Returns the number of bytes sent, or a negative errno.
542 */
543static int svc_udp_sendto(struct svc_rqst *rqstp)
544{
545 struct svc_xprt *xprt = rqstp->rq_xprt;
546 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
547 struct xdr_buf *xdr = &rqstp->rq_res;
548 union {
549 struct cmsghdr hdr;
550 long all[SVC_PKTINFO_SPACE / sizeof(long)];
551 } buffer;
552 struct cmsghdr *cmh = &buffer.hdr;
553 struct msghdr msg = {
554 .msg_name = &rqstp->rq_addr,
555 .msg_namelen = rqstp->rq_addrlen,
556 .msg_control = cmh,
557 .msg_controllen = sizeof(buffer),
558 };
559 unsigned int sent;
560 int err;
561
562 svc_udp_release_rqst(rqstp);
563
564 svc_set_cmsg_data(rqstp, cmh);
565
566 mutex_lock(&xprt->xpt_mutex);
567
568 if (svc_xprt_is_dead(xprt))
569 goto out_notconn;
570
571 err = xdr_alloc_bvec(xdr, GFP_KERNEL);
572 if (err < 0)
573 goto out_unlock;
574
575 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
576 if (err == -ECONNREFUSED) {
577 /* ICMP error on earlier request. */
578 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
579 }
580 xdr_free_bvec(xdr);
581 trace_svcsock_udp_send(xprt, err);
582out_unlock:
583 mutex_unlock(&xprt->xpt_mutex);
584 if (err < 0)
585 return err;
586 return sent;
587
588out_notconn:
589 mutex_unlock(&xprt->xpt_mutex);
590 return -ENOTCONN;
591}
592
593static int svc_udp_has_wspace(struct svc_xprt *xprt)
594{
595 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
596 struct svc_serv *serv = xprt->xpt_server;
597 unsigned long required;
598
599 /*
600 * Set the SOCK_NOSPACE flag before checking the available
601 * sock space.
602 */
603 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
604 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
605 if (required*2 > sock_wspace(svsk->sk_sk))
606 return 0;
607 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
608 return 1;
609}
610
611static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
612{
613 BUG();
614 return NULL;
615}
616
617static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
618{
619}
620
621static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
622 struct net *net,
623 struct sockaddr *sa, int salen,
624 int flags)
625{
626 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
627}
628
629static const struct svc_xprt_ops svc_udp_ops = {
630 .xpo_create = svc_udp_create,
631 .xpo_recvfrom = svc_udp_recvfrom,
632 .xpo_sendto = svc_udp_sendto,
633 .xpo_result_payload = svc_sock_result_payload,
634 .xpo_release_rqst = svc_udp_release_rqst,
635 .xpo_detach = svc_sock_detach,
636 .xpo_free = svc_sock_free,
637 .xpo_has_wspace = svc_udp_has_wspace,
638 .xpo_accept = svc_udp_accept,
639 .xpo_secure_port = svc_sock_secure_port,
640 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
641};
642
643static struct svc_xprt_class svc_udp_class = {
644 .xcl_name = "udp",
645 .xcl_owner = THIS_MODULE,
646 .xcl_ops = &svc_udp_ops,
647 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
648 .xcl_ident = XPRT_TRANSPORT_UDP,
649};
650
651static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
652{
653 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
654 &svsk->sk_xprt, serv);
655 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
656 svsk->sk_sk->sk_data_ready = svc_data_ready;
657 svsk->sk_sk->sk_write_space = svc_write_space;
658
659 /* initialise setting must have enough space to
660 * receive and respond to one request.
661 * svc_udp_recvfrom will re-adjust if necessary
662 */
663 svc_sock_setbufsize(svsk, 3);
664
665 /* data might have come in before data_ready set up */
666 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
667 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
668
669 /* make sure we get destination address info */
670 switch (svsk->sk_sk->sk_family) {
671 case AF_INET:
672 ip_sock_set_pktinfo(svsk->sk_sock->sk);
673 break;
674 case AF_INET6:
675 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
676 break;
677 default:
678 BUG();
679 }
680}
681
682/*
683 * A data_ready event on a listening socket means there's a connection
684 * pending. Do not use state_change as a substitute for it.
685 */
686static void svc_tcp_listen_data_ready(struct sock *sk)
687{
688 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
689
690 if (svsk) {
691 /* Refer to svc_setup_socket() for details. */
692 rmb();
693 svsk->sk_odata(sk);
694 }
695
696 /*
697 * This callback may called twice when a new connection
698 * is established as a child socket inherits everything
699 * from a parent LISTEN socket.
700 * 1) data_ready method of the parent socket will be called
701 * when one of child sockets become ESTABLISHED.
702 * 2) data_ready method of the child socket may be called
703 * when it receives data before the socket is accepted.
704 * In case of 2, we should ignore it silently.
705 */
706 if (sk->sk_state == TCP_LISTEN) {
707 if (svsk) {
708 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
709 svc_xprt_enqueue(&svsk->sk_xprt);
710 }
711 }
712}
713
714/*
715 * A state change on a connected socket means it's dying or dead.
716 */
717static void svc_tcp_state_change(struct sock *sk)
718{
719 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
720
721 if (svsk) {
722 /* Refer to svc_setup_socket() for details. */
723 rmb();
724 svsk->sk_ostate(sk);
725 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
726 if (sk->sk_state != TCP_ESTABLISHED)
727 svc_xprt_deferred_close(&svsk->sk_xprt);
728 }
729}
730
731/*
732 * Accept a TCP connection
733 */
734static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
735{
736 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
737 struct sockaddr_storage addr;
738 struct sockaddr *sin = (struct sockaddr *) &addr;
739 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
740 struct socket *sock = svsk->sk_sock;
741 struct socket *newsock;
742 struct svc_sock *newsvsk;
743 int err, slen;
744
745 if (!sock)
746 return NULL;
747
748 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
749 err = kernel_accept(sock, &newsock, O_NONBLOCK);
750 if (err < 0) {
751 if (err == -ENOMEM)
752 printk(KERN_WARNING "%s: no more sockets!\n",
753 serv->sv_name);
754 else if (err != -EAGAIN)
755 net_warn_ratelimited("%s: accept failed (err %d)!\n",
756 serv->sv_name, -err);
757 trace_svcsock_accept_err(xprt, serv->sv_name, err);
758 return NULL;
759 }
760 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
761
762 err = kernel_getpeername(newsock, sin);
763 if (err < 0) {
764 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
765 goto failed; /* aborted connection or whatever */
766 }
767 slen = err;
768
769 /* Reset the inherited callbacks before calling svc_setup_socket */
770 newsock->sk->sk_state_change = svsk->sk_ostate;
771 newsock->sk->sk_data_ready = svsk->sk_odata;
772 newsock->sk->sk_write_space = svsk->sk_owspace;
773
774 /* make sure that a write doesn't block forever when
775 * low on memory
776 */
777 newsock->sk->sk_sndtimeo = HZ*30;
778
779 newsvsk = svc_setup_socket(serv, newsock,
780 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
781 if (IS_ERR(newsvsk))
782 goto failed;
783 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
784 err = kernel_getsockname(newsock, sin);
785 slen = err;
786 if (unlikely(err < 0))
787 slen = offsetof(struct sockaddr, sa_data);
788 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
789
790 if (sock_is_loopback(newsock->sk))
791 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
792 else
793 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
794 if (serv->sv_stats)
795 serv->sv_stats->nettcpconn++;
796
797 return &newsvsk->sk_xprt;
798
799failed:
800 sock_release(newsock);
801 return NULL;
802}
803
804static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
805 struct svc_rqst *rqstp)
806{
807 size_t len = svsk->sk_datalen;
808 unsigned int i, npages;
809
810 if (!len)
811 return 0;
812 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
813 for (i = 0; i < npages; i++) {
814 if (rqstp->rq_pages[i] != NULL)
815 put_page(rqstp->rq_pages[i]);
816 BUG_ON(svsk->sk_pages[i] == NULL);
817 rqstp->rq_pages[i] = svsk->sk_pages[i];
818 svsk->sk_pages[i] = NULL;
819 }
820 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
821 return len;
822}
823
824static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
825{
826 unsigned int i, len, npages;
827
828 if (svsk->sk_datalen == 0)
829 return;
830 len = svsk->sk_datalen;
831 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
832 for (i = 0; i < npages; i++) {
833 svsk->sk_pages[i] = rqstp->rq_pages[i];
834 rqstp->rq_pages[i] = NULL;
835 }
836}
837
838static void svc_tcp_clear_pages(struct svc_sock *svsk)
839{
840 unsigned int i, len, npages;
841
842 if (svsk->sk_datalen == 0)
843 goto out;
844 len = svsk->sk_datalen;
845 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
846 for (i = 0; i < npages; i++) {
847 if (svsk->sk_pages[i] == NULL) {
848 WARN_ON_ONCE(1);
849 continue;
850 }
851 put_page(svsk->sk_pages[i]);
852 svsk->sk_pages[i] = NULL;
853 }
854out:
855 svsk->sk_tcplen = 0;
856 svsk->sk_datalen = 0;
857}
858
859/*
860 * Receive fragment record header into sk_marker.
861 */
862static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
863 struct svc_rqst *rqstp)
864{
865 ssize_t want, len;
866
867 /* If we haven't gotten the record length yet,
868 * get the next four bytes.
869 */
870 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
871 struct msghdr msg = { NULL };
872 struct kvec iov;
873
874 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
875 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
876 iov.iov_len = want;
877 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
878 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
879 if (len < 0)
880 return len;
881 svsk->sk_tcplen += len;
882 if (len < want) {
883 /* call again to read the remaining bytes */
884 goto err_short;
885 }
886 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
887 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
888 svsk->sk_xprt.xpt_server->sv_max_mesg)
889 goto err_too_large;
890 }
891 return svc_sock_reclen(svsk);
892
893err_too_large:
894 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
895 __func__, svsk->sk_xprt.xpt_server->sv_name,
896 svc_sock_reclen(svsk));
897 svc_xprt_deferred_close(&svsk->sk_xprt);
898err_short:
899 return -EAGAIN;
900}
901
902static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
903{
904 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
905 struct rpc_rqst *req = NULL;
906 struct kvec *src, *dst;
907 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
908 __be32 xid;
909 __be32 calldir;
910
911 xid = *p++;
912 calldir = *p;
913
914 if (!bc_xprt)
915 return -EAGAIN;
916 spin_lock(&bc_xprt->queue_lock);
917 req = xprt_lookup_rqst(bc_xprt, xid);
918 if (!req)
919 goto unlock_notfound;
920
921 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
922 /*
923 * XXX!: cheating for now! Only copying HEAD.
924 * But we know this is good enough for now (in fact, for any
925 * callback reply in the forseeable future).
926 */
927 dst = &req->rq_private_buf.head[0];
928 src = &rqstp->rq_arg.head[0];
929 if (dst->iov_len < src->iov_len)
930 goto unlock_eagain; /* whatever; just giving up. */
931 memcpy(dst->iov_base, src->iov_base, src->iov_len);
932 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
933 rqstp->rq_arg.len = 0;
934 spin_unlock(&bc_xprt->queue_lock);
935 return 0;
936unlock_notfound:
937 printk(KERN_NOTICE
938 "%s: Got unrecognized reply: "
939 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
940 __func__, ntohl(calldir),
941 bc_xprt, ntohl(xid));
942unlock_eagain:
943 spin_unlock(&bc_xprt->queue_lock);
944 return -EAGAIN;
945}
946
947static void svc_tcp_fragment_received(struct svc_sock *svsk)
948{
949 /* If we have more data, signal svc_xprt_enqueue() to try again */
950 svsk->sk_tcplen = 0;
951 svsk->sk_marker = xdr_zero;
952}
953
954/**
955 * svc_tcp_recvfrom - Receive data from a TCP socket
956 * @rqstp: request structure into which to receive an RPC Call
957 *
958 * Called in a loop when XPT_DATA has been set.
959 *
960 * Read the 4-byte stream record marker, then use the record length
961 * in that marker to set up exactly the resources needed to receive
962 * the next RPC message into @rqstp.
963 *
964 * Returns:
965 * On success, the number of bytes in a received RPC Call, or
966 * %0 if a complete RPC Call message was not ready to return
967 *
968 * The zero return case handles partial receives and callback Replies.
969 * The state of a partial receive is preserved in the svc_sock for
970 * the next call to svc_tcp_recvfrom.
971 */
972static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
973{
974 struct svc_sock *svsk =
975 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
976 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
977 size_t want, base;
978 ssize_t len;
979 __be32 *p;
980 __be32 calldir;
981
982 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
983 len = svc_tcp_read_marker(svsk, rqstp);
984 if (len < 0)
985 goto error;
986
987 base = svc_tcp_restore_pages(svsk, rqstp);
988 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
989 len = svc_tcp_read_msg(rqstp, base + want, base);
990 if (len >= 0) {
991 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
992 svsk->sk_tcplen += len;
993 svsk->sk_datalen += len;
994 }
995 if (len != want || !svc_sock_final_rec(svsk))
996 goto err_incomplete;
997 if (svsk->sk_datalen < 8)
998 goto err_nuts;
999
1000 rqstp->rq_arg.len = svsk->sk_datalen;
1001 rqstp->rq_arg.page_base = 0;
1002 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1003 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1004 rqstp->rq_arg.page_len = 0;
1005 } else
1006 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1007
1008 rqstp->rq_xprt_ctxt = NULL;
1009 rqstp->rq_prot = IPPROTO_TCP;
1010 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1011 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1012 else
1013 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1014
1015 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1016 calldir = p[1];
1017 if (calldir)
1018 len = receive_cb_reply(svsk, rqstp);
1019
1020 /* Reset TCP read info */
1021 svsk->sk_datalen = 0;
1022 svc_tcp_fragment_received(svsk);
1023
1024 if (len < 0)
1025 goto error;
1026
1027 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1028 if (serv->sv_stats)
1029 serv->sv_stats->nettcpcnt++;
1030
1031 svc_xprt_received(rqstp->rq_xprt);
1032 return rqstp->rq_arg.len;
1033
1034err_incomplete:
1035 svc_tcp_save_pages(svsk, rqstp);
1036 if (len < 0 && len != -EAGAIN)
1037 goto err_delete;
1038 if (len == want)
1039 svc_tcp_fragment_received(svsk);
1040 else
1041 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1042 svc_sock_reclen(svsk),
1043 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1044 goto err_noclose;
1045error:
1046 if (len != -EAGAIN)
1047 goto err_delete;
1048 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1049 goto err_noclose;
1050err_nuts:
1051 svsk->sk_datalen = 0;
1052err_delete:
1053 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1054 svc_xprt_deferred_close(&svsk->sk_xprt);
1055err_noclose:
1056 svc_xprt_received(rqstp->rq_xprt);
1057 return 0; /* record not complete */
1058}
1059
1060static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1061 int flags)
1062{
1063 return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1064 offset_in_page(vec->iov_base),
1065 vec->iov_len, flags);
1066}
1067
1068/*
1069 * kernel_sendpage() is used exclusively to reduce the number of
1070 * copy operations in this path. Therefore the caller must ensure
1071 * that the pages backing @xdr are unchanging.
1072 *
1073 * In addition, the logic assumes that * .bv_len is never larger
1074 * than PAGE_SIZE.
1075 */
1076static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
1077 rpc_fraghdr marker, unsigned int *sentp)
1078{
1079 const struct kvec *head = xdr->head;
1080 const struct kvec *tail = xdr->tail;
1081 struct kvec rm = {
1082 .iov_base = &marker,
1083 .iov_len = sizeof(marker),
1084 };
1085 struct msghdr msg = {
1086 .msg_flags = 0,
1087 };
1088 int ret;
1089
1090 *sentp = 0;
1091 ret = xdr_alloc_bvec(xdr, GFP_KERNEL);
1092 if (ret < 0)
1093 return ret;
1094
1095 ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len);
1096 if (ret < 0)
1097 return ret;
1098 *sentp += ret;
1099 if (ret != rm.iov_len)
1100 return -EAGAIN;
1101
1102 ret = svc_tcp_send_kvec(sock, head, 0);
1103 if (ret < 0)
1104 return ret;
1105 *sentp += ret;
1106 if (ret != head->iov_len)
1107 goto out;
1108
1109 if (xdr->page_len) {
1110 unsigned int offset, len, remaining;
1111 struct bio_vec *bvec;
1112
1113 bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1114 offset = offset_in_page(xdr->page_base);
1115 remaining = xdr->page_len;
1116 while (remaining > 0) {
1117 len = min(remaining, bvec->bv_len - offset);
1118 ret = kernel_sendpage(sock, bvec->bv_page,
1119 bvec->bv_offset + offset,
1120 len, 0);
1121 if (ret < 0)
1122 return ret;
1123 *sentp += ret;
1124 if (ret != len)
1125 goto out;
1126 remaining -= len;
1127 offset = 0;
1128 bvec++;
1129 }
1130 }
1131
1132 if (tail->iov_len) {
1133 ret = svc_tcp_send_kvec(sock, tail, 0);
1134 if (ret < 0)
1135 return ret;
1136 *sentp += ret;
1137 }
1138
1139out:
1140 return 0;
1141}
1142
1143/**
1144 * svc_tcp_sendto - Send out a reply on a TCP socket
1145 * @rqstp: completed svc_rqst
1146 *
1147 * xpt_mutex ensures @rqstp's whole message is written to the socket
1148 * without interruption.
1149 *
1150 * Returns the number of bytes sent, or a negative errno.
1151 */
1152static int svc_tcp_sendto(struct svc_rqst *rqstp)
1153{
1154 struct svc_xprt *xprt = rqstp->rq_xprt;
1155 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1156 struct xdr_buf *xdr = &rqstp->rq_res;
1157 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1158 (u32)xdr->len);
1159 unsigned int sent;
1160 int err;
1161
1162 svc_tcp_release_rqst(rqstp);
1163
1164 atomic_inc(&svsk->sk_sendqlen);
1165 mutex_lock(&xprt->xpt_mutex);
1166 if (svc_xprt_is_dead(xprt))
1167 goto out_notconn;
1168 tcp_sock_set_cork(svsk->sk_sk, true);
1169 err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent);
1170 xdr_free_bvec(xdr);
1171 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1172 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1173 goto out_close;
1174 if (atomic_dec_and_test(&svsk->sk_sendqlen))
1175 tcp_sock_set_cork(svsk->sk_sk, false);
1176 mutex_unlock(&xprt->xpt_mutex);
1177 return sent;
1178
1179out_notconn:
1180 atomic_dec(&svsk->sk_sendqlen);
1181 mutex_unlock(&xprt->xpt_mutex);
1182 return -ENOTCONN;
1183out_close:
1184 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1185 xprt->xpt_server->sv_name,
1186 (err < 0) ? "got error" : "sent",
1187 (err < 0) ? err : sent, xdr->len);
1188 svc_xprt_deferred_close(xprt);
1189 atomic_dec(&svsk->sk_sendqlen);
1190 mutex_unlock(&xprt->xpt_mutex);
1191 return -EAGAIN;
1192}
1193
1194static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1195 struct net *net,
1196 struct sockaddr *sa, int salen,
1197 int flags)
1198{
1199 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1200}
1201
1202static const struct svc_xprt_ops svc_tcp_ops = {
1203 .xpo_create = svc_tcp_create,
1204 .xpo_recvfrom = svc_tcp_recvfrom,
1205 .xpo_sendto = svc_tcp_sendto,
1206 .xpo_result_payload = svc_sock_result_payload,
1207 .xpo_release_rqst = svc_tcp_release_rqst,
1208 .xpo_detach = svc_tcp_sock_detach,
1209 .xpo_free = svc_sock_free,
1210 .xpo_has_wspace = svc_tcp_has_wspace,
1211 .xpo_accept = svc_tcp_accept,
1212 .xpo_secure_port = svc_sock_secure_port,
1213 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1214};
1215
1216static struct svc_xprt_class svc_tcp_class = {
1217 .xcl_name = "tcp",
1218 .xcl_owner = THIS_MODULE,
1219 .xcl_ops = &svc_tcp_ops,
1220 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1221 .xcl_ident = XPRT_TRANSPORT_TCP,
1222};
1223
1224void svc_init_xprt_sock(void)
1225{
1226 svc_reg_xprt_class(&svc_tcp_class);
1227 svc_reg_xprt_class(&svc_udp_class);
1228}
1229
1230void svc_cleanup_xprt_sock(void)
1231{
1232 svc_unreg_xprt_class(&svc_tcp_class);
1233 svc_unreg_xprt_class(&svc_udp_class);
1234}
1235
1236static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1237{
1238 struct sock *sk = svsk->sk_sk;
1239
1240 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1241 &svsk->sk_xprt, serv);
1242 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1243 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1244 if (sk->sk_state == TCP_LISTEN) {
1245 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1246 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1247 sk->sk_data_ready = svc_tcp_listen_data_ready;
1248 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1249 } else {
1250 sk->sk_state_change = svc_tcp_state_change;
1251 sk->sk_data_ready = svc_data_ready;
1252 sk->sk_write_space = svc_write_space;
1253
1254 svsk->sk_marker = xdr_zero;
1255 svsk->sk_tcplen = 0;
1256 svsk->sk_datalen = 0;
1257 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1258
1259 tcp_sock_set_nodelay(sk);
1260
1261 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1262 switch (sk->sk_state) {
1263 case TCP_SYN_RECV:
1264 case TCP_ESTABLISHED:
1265 break;
1266 default:
1267 svc_xprt_deferred_close(&svsk->sk_xprt);
1268 }
1269 }
1270}
1271
1272void svc_sock_update_bufs(struct svc_serv *serv)
1273{
1274 /*
1275 * The number of server threads has changed. Update
1276 * rcvbuf and sndbuf accordingly on all sockets
1277 */
1278 struct svc_sock *svsk;
1279
1280 spin_lock_bh(&serv->sv_lock);
1281 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1282 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1283 spin_unlock_bh(&serv->sv_lock);
1284}
1285EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1286
1287/*
1288 * Initialize socket for RPC use and create svc_sock struct
1289 */
1290static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1291 struct socket *sock,
1292 int flags)
1293{
1294 struct svc_sock *svsk;
1295 struct sock *inet;
1296 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1297 int err = 0;
1298
1299 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1300 if (!svsk)
1301 return ERR_PTR(-ENOMEM);
1302
1303 inet = sock->sk;
1304
1305 /* Register socket with portmapper */
1306 if (pmap_register)
1307 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1308 inet->sk_protocol,
1309 ntohs(inet_sk(inet)->inet_sport));
1310
1311 if (err < 0) {
1312 kfree(svsk);
1313 return ERR_PTR(err);
1314 }
1315
1316 svsk->sk_sock = sock;
1317 svsk->sk_sk = inet;
1318 svsk->sk_ostate = inet->sk_state_change;
1319 svsk->sk_odata = inet->sk_data_ready;
1320 svsk->sk_owspace = inet->sk_write_space;
1321 /*
1322 * This barrier is necessary in order to prevent race condition
1323 * with svc_data_ready(), svc_listen_data_ready() and others
1324 * when calling callbacks above.
1325 */
1326 wmb();
1327 inet->sk_user_data = svsk;
1328
1329 /* Initialize the socket */
1330 if (sock->type == SOCK_DGRAM)
1331 svc_udp_init(svsk, serv);
1332 else
1333 svc_tcp_init(svsk, serv);
1334
1335 trace_svcsock_new_socket(sock);
1336 return svsk;
1337}
1338
1339bool svc_alien_sock(struct net *net, int fd)
1340{
1341 int err;
1342 struct socket *sock = sockfd_lookup(fd, &err);
1343 bool ret = false;
1344
1345 if (!sock)
1346 goto out;
1347 if (sock_net(sock->sk) != net)
1348 ret = true;
1349 sockfd_put(sock);
1350out:
1351 return ret;
1352}
1353EXPORT_SYMBOL_GPL(svc_alien_sock);
1354
1355/**
1356 * svc_addsock - add a listener socket to an RPC service
1357 * @serv: pointer to RPC service to which to add a new listener
1358 * @fd: file descriptor of the new listener
1359 * @name_return: pointer to buffer to fill in with name of listener
1360 * @len: size of the buffer
1361 * @cred: credential
1362 *
1363 * Fills in socket name and returns positive length of name if successful.
1364 * Name is terminated with '\n'. On error, returns a negative errno
1365 * value.
1366 */
1367int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1368 const size_t len, const struct cred *cred)
1369{
1370 int err = 0;
1371 struct socket *so = sockfd_lookup(fd, &err);
1372 struct svc_sock *svsk = NULL;
1373 struct sockaddr_storage addr;
1374 struct sockaddr *sin = (struct sockaddr *)&addr;
1375 int salen;
1376
1377 if (!so)
1378 return err;
1379 err = -EAFNOSUPPORT;
1380 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1381 goto out;
1382 err = -EPROTONOSUPPORT;
1383 if (so->sk->sk_protocol != IPPROTO_TCP &&
1384 so->sk->sk_protocol != IPPROTO_UDP)
1385 goto out;
1386 err = -EISCONN;
1387 if (so->state > SS_UNCONNECTED)
1388 goto out;
1389 err = -ENOENT;
1390 if (!try_module_get(THIS_MODULE))
1391 goto out;
1392 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1393 if (IS_ERR(svsk)) {
1394 module_put(THIS_MODULE);
1395 err = PTR_ERR(svsk);
1396 goto out;
1397 }
1398 salen = kernel_getsockname(svsk->sk_sock, sin);
1399 if (salen >= 0)
1400 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1401 svsk->sk_xprt.xpt_cred = get_cred(cred);
1402 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1403 return svc_one_sock_name(svsk, name_return, len);
1404out:
1405 sockfd_put(so);
1406 return err;
1407}
1408EXPORT_SYMBOL_GPL(svc_addsock);
1409
1410/*
1411 * Create socket for RPC service.
1412 */
1413static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1414 int protocol,
1415 struct net *net,
1416 struct sockaddr *sin, int len,
1417 int flags)
1418{
1419 struct svc_sock *svsk;
1420 struct socket *sock;
1421 int error;
1422 int type;
1423 struct sockaddr_storage addr;
1424 struct sockaddr *newsin = (struct sockaddr *)&addr;
1425 int newlen;
1426 int family;
1427
1428 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1429 printk(KERN_WARNING "svc: only UDP and TCP "
1430 "sockets supported\n");
1431 return ERR_PTR(-EINVAL);
1432 }
1433
1434 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1435 switch (sin->sa_family) {
1436 case AF_INET6:
1437 family = PF_INET6;
1438 break;
1439 case AF_INET:
1440 family = PF_INET;
1441 break;
1442 default:
1443 return ERR_PTR(-EINVAL);
1444 }
1445
1446 error = __sock_create(net, family, type, protocol, &sock, 1);
1447 if (error < 0)
1448 return ERR_PTR(error);
1449
1450 svc_reclassify_socket(sock);
1451
1452 /*
1453 * If this is an PF_INET6 listener, we want to avoid
1454 * getting requests from IPv4 remotes. Those should
1455 * be shunted to a PF_INET listener via rpcbind.
1456 */
1457 if (family == PF_INET6)
1458 ip6_sock_set_v6only(sock->sk);
1459 if (type == SOCK_STREAM)
1460 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1461 error = kernel_bind(sock, sin, len);
1462 if (error < 0)
1463 goto bummer;
1464
1465 error = kernel_getsockname(sock, newsin);
1466 if (error < 0)
1467 goto bummer;
1468 newlen = error;
1469
1470 if (protocol == IPPROTO_TCP) {
1471 if ((error = kernel_listen(sock, 64)) < 0)
1472 goto bummer;
1473 }
1474
1475 svsk = svc_setup_socket(serv, sock, flags);
1476 if (IS_ERR(svsk)) {
1477 error = PTR_ERR(svsk);
1478 goto bummer;
1479 }
1480 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1481 return (struct svc_xprt *)svsk;
1482bummer:
1483 sock_release(sock);
1484 return ERR_PTR(error);
1485}
1486
1487/*
1488 * Detach the svc_sock from the socket so that no
1489 * more callbacks occur.
1490 */
1491static void svc_sock_detach(struct svc_xprt *xprt)
1492{
1493 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1494 struct sock *sk = svsk->sk_sk;
1495
1496 /* put back the old socket callbacks */
1497 lock_sock(sk);
1498 sk->sk_state_change = svsk->sk_ostate;
1499 sk->sk_data_ready = svsk->sk_odata;
1500 sk->sk_write_space = svsk->sk_owspace;
1501 sk->sk_user_data = NULL;
1502 release_sock(sk);
1503}
1504
1505/*
1506 * Disconnect the socket, and reset the callbacks
1507 */
1508static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1509{
1510 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1511
1512 svc_sock_detach(xprt);
1513
1514 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1515 svc_tcp_clear_pages(svsk);
1516 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1517 }
1518}
1519
1520/*
1521 * Free the svc_sock's socket resources and the svc_sock itself.
1522 */
1523static void svc_sock_free(struct svc_xprt *xprt)
1524{
1525 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1526
1527 if (svsk->sk_sock->file)
1528 sockfd_put(svsk->sk_sock);
1529 else
1530 sock_release(svsk->sk_sock);
1531 kfree(svsk);
1532}
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/module.h>
25#include <linux/errno.h>
26#include <linux/fcntl.h>
27#include <linux/net.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/udp.h>
31#include <linux/tcp.h>
32#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/netdevice.h>
35#include <linux/skbuff.h>
36#include <linux/file.h>
37#include <linux/freezer.h>
38#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/ip.h>
41#include <net/ipv6.h>
42#include <net/tcp.h>
43#include <net/tcp_states.h>
44#include <asm/uaccess.h>
45#include <asm/ioctls.h>
46
47#include <linux/sunrpc/types.h>
48#include <linux/sunrpc/clnt.h>
49#include <linux/sunrpc/xdr.h>
50#include <linux/sunrpc/msg_prot.h>
51#include <linux/sunrpc/svcsock.h>
52#include <linux/sunrpc/stats.h>
53#include <linux/sunrpc/xprt.h>
54
55#include "sunrpc.h"
56
57#define RPCDBG_FACILITY RPCDBG_SVCXPRT
58
59
60static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
61 int *errp, int flags);
62static void svc_udp_data_ready(struct sock *, int);
63static int svc_udp_recvfrom(struct svc_rqst *);
64static int svc_udp_sendto(struct svc_rqst *);
65static void svc_sock_detach(struct svc_xprt *);
66static void svc_tcp_sock_detach(struct svc_xprt *);
67static void svc_sock_free(struct svc_xprt *);
68
69static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
70 struct net *, struct sockaddr *,
71 int, int);
72#if defined(CONFIG_SUNRPC_BACKCHANNEL)
73static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
74 struct net *, struct sockaddr *,
75 int, int);
76static void svc_bc_sock_free(struct svc_xprt *xprt);
77#endif /* CONFIG_SUNRPC_BACKCHANNEL */
78
79#ifdef CONFIG_DEBUG_LOCK_ALLOC
80static struct lock_class_key svc_key[2];
81static struct lock_class_key svc_slock_key[2];
82
83static void svc_reclassify_socket(struct socket *sock)
84{
85 struct sock *sk = sock->sk;
86 BUG_ON(sock_owned_by_user(sk));
87 switch (sk->sk_family) {
88 case AF_INET:
89 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
90 &svc_slock_key[0],
91 "sk_xprt.xpt_lock-AF_INET-NFSD",
92 &svc_key[0]);
93 break;
94
95 case AF_INET6:
96 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
97 &svc_slock_key[1],
98 "sk_xprt.xpt_lock-AF_INET6-NFSD",
99 &svc_key[1]);
100 break;
101
102 default:
103 BUG();
104 }
105}
106#else
107static void svc_reclassify_socket(struct socket *sock)
108{
109}
110#endif
111
112/*
113 * Release an skbuff after use
114 */
115static void svc_release_skb(struct svc_rqst *rqstp)
116{
117 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
118
119 if (skb) {
120 struct svc_sock *svsk =
121 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
122 rqstp->rq_xprt_ctxt = NULL;
123
124 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
125 skb_free_datagram_locked(svsk->sk_sk, skb);
126 }
127}
128
129union svc_pktinfo_u {
130 struct in_pktinfo pkti;
131 struct in6_pktinfo pkti6;
132};
133#define SVC_PKTINFO_SPACE \
134 CMSG_SPACE(sizeof(union svc_pktinfo_u))
135
136static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
137{
138 struct svc_sock *svsk =
139 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
140 switch (svsk->sk_sk->sk_family) {
141 case AF_INET: {
142 struct in_pktinfo *pki = CMSG_DATA(cmh);
143
144 cmh->cmsg_level = SOL_IP;
145 cmh->cmsg_type = IP_PKTINFO;
146 pki->ipi_ifindex = 0;
147 pki->ipi_spec_dst.s_addr =
148 svc_daddr_in(rqstp)->sin_addr.s_addr;
149 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
150 }
151 break;
152
153 case AF_INET6: {
154 struct in6_pktinfo *pki = CMSG_DATA(cmh);
155 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
156
157 cmh->cmsg_level = SOL_IPV6;
158 cmh->cmsg_type = IPV6_PKTINFO;
159 pki->ipi6_ifindex = daddr->sin6_scope_id;
160 pki->ipi6_addr = daddr->sin6_addr;
161 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
162 }
163 break;
164 }
165}
166
167/*
168 * send routine intended to be shared by the fore- and back-channel
169 */
170int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
171 struct page *headpage, unsigned long headoffset,
172 struct page *tailpage, unsigned long tailoffset)
173{
174 int result;
175 int size;
176 struct page **ppage = xdr->pages;
177 size_t base = xdr->page_base;
178 unsigned int pglen = xdr->page_len;
179 unsigned int flags = MSG_MORE;
180 int slen;
181 int len = 0;
182
183 slen = xdr->len;
184
185 /* send head */
186 if (slen == xdr->head[0].iov_len)
187 flags = 0;
188 len = kernel_sendpage(sock, headpage, headoffset,
189 xdr->head[0].iov_len, flags);
190 if (len != xdr->head[0].iov_len)
191 goto out;
192 slen -= xdr->head[0].iov_len;
193 if (slen == 0)
194 goto out;
195
196 /* send page data */
197 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
198 while (pglen > 0) {
199 if (slen == size)
200 flags = 0;
201 result = kernel_sendpage(sock, *ppage, base, size, flags);
202 if (result > 0)
203 len += result;
204 if (result != size)
205 goto out;
206 slen -= size;
207 pglen -= size;
208 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
209 base = 0;
210 ppage++;
211 }
212
213 /* send tail */
214 if (xdr->tail[0].iov_len) {
215 result = kernel_sendpage(sock, tailpage, tailoffset,
216 xdr->tail[0].iov_len, 0);
217 if (result > 0)
218 len += result;
219 }
220
221out:
222 return len;
223}
224
225
226/*
227 * Generic sendto routine
228 */
229static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
230{
231 struct svc_sock *svsk =
232 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
233 struct socket *sock = svsk->sk_sock;
234 union {
235 struct cmsghdr hdr;
236 long all[SVC_PKTINFO_SPACE / sizeof(long)];
237 } buffer;
238 struct cmsghdr *cmh = &buffer.hdr;
239 int len = 0;
240 unsigned long tailoff;
241 unsigned long headoff;
242 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
243
244 if (rqstp->rq_prot == IPPROTO_UDP) {
245 struct msghdr msg = {
246 .msg_name = &rqstp->rq_addr,
247 .msg_namelen = rqstp->rq_addrlen,
248 .msg_control = cmh,
249 .msg_controllen = sizeof(buffer),
250 .msg_flags = MSG_MORE,
251 };
252
253 svc_set_cmsg_data(rqstp, cmh);
254
255 if (sock_sendmsg(sock, &msg, 0) < 0)
256 goto out;
257 }
258
259 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
260 headoff = 0;
261 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
262 rqstp->rq_respages[0], tailoff);
263
264out:
265 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
266 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
267 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
268
269 return len;
270}
271
272/*
273 * Report socket names for nfsdfs
274 */
275static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
276{
277 const struct sock *sk = svsk->sk_sk;
278 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
279 "udp" : "tcp";
280 int len;
281
282 switch (sk->sk_family) {
283 case PF_INET:
284 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
285 proto_name,
286 &inet_sk(sk)->inet_rcv_saddr,
287 inet_sk(sk)->inet_num);
288 break;
289 case PF_INET6:
290 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
291 proto_name,
292 &inet6_sk(sk)->rcv_saddr,
293 inet_sk(sk)->inet_num);
294 break;
295 default:
296 len = snprintf(buf, remaining, "*unknown-%d*\n",
297 sk->sk_family);
298 }
299
300 if (len >= remaining) {
301 *buf = '\0';
302 return -ENAMETOOLONG;
303 }
304 return len;
305}
306
307/**
308 * svc_sock_names - construct a list of listener names in a string
309 * @serv: pointer to RPC service
310 * @buf: pointer to a buffer to fill in with socket names
311 * @buflen: size of the buffer to be filled
312 * @toclose: pointer to '\0'-terminated C string containing the name
313 * of a listener to be closed
314 *
315 * Fills in @buf with a '\n'-separated list of names of listener
316 * sockets. If @toclose is not NULL, the socket named by @toclose
317 * is closed, and is not included in the output list.
318 *
319 * Returns positive length of the socket name string, or a negative
320 * errno value on error.
321 */
322int svc_sock_names(struct svc_serv *serv, char *buf, const size_t buflen,
323 const char *toclose)
324{
325 struct svc_sock *svsk, *closesk = NULL;
326 int len = 0;
327
328 if (!serv)
329 return 0;
330
331 spin_lock_bh(&serv->sv_lock);
332 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list) {
333 int onelen = svc_one_sock_name(svsk, buf + len, buflen - len);
334 if (onelen < 0) {
335 len = onelen;
336 break;
337 }
338 if (toclose && strcmp(toclose, buf + len) == 0) {
339 closesk = svsk;
340 svc_xprt_get(&closesk->sk_xprt);
341 } else
342 len += onelen;
343 }
344 spin_unlock_bh(&serv->sv_lock);
345
346 if (closesk) {
347 /* Should unregister with portmap, but you cannot
348 * unregister just one protocol...
349 */
350 svc_close_xprt(&closesk->sk_xprt);
351 svc_xprt_put(&closesk->sk_xprt);
352 } else if (toclose)
353 return -ENOENT;
354 return len;
355}
356EXPORT_SYMBOL_GPL(svc_sock_names);
357
358/*
359 * Check input queue length
360 */
361static int svc_recv_available(struct svc_sock *svsk)
362{
363 struct socket *sock = svsk->sk_sock;
364 int avail, err;
365
366 err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
367
368 return (err >= 0)? avail : err;
369}
370
371/*
372 * Generic recvfrom routine.
373 */
374static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
375 int buflen)
376{
377 struct svc_sock *svsk =
378 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
379 struct msghdr msg = {
380 .msg_flags = MSG_DONTWAIT,
381 };
382 int len;
383
384 rqstp->rq_xprt_hlen = 0;
385
386 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
387 msg.msg_flags);
388
389 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
390 svsk, iov[0].iov_base, iov[0].iov_len, len);
391 return len;
392}
393
394static int svc_partial_recvfrom(struct svc_rqst *rqstp,
395 struct kvec *iov, int nr,
396 int buflen, unsigned int base)
397{
398 size_t save_iovlen;
399 void *save_iovbase;
400 unsigned int i;
401 int ret;
402
403 if (base == 0)
404 return svc_recvfrom(rqstp, iov, nr, buflen);
405
406 for (i = 0; i < nr; i++) {
407 if (iov[i].iov_len > base)
408 break;
409 base -= iov[i].iov_len;
410 }
411 save_iovlen = iov[i].iov_len;
412 save_iovbase = iov[i].iov_base;
413 iov[i].iov_len -= base;
414 iov[i].iov_base += base;
415 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
416 iov[i].iov_len = save_iovlen;
417 iov[i].iov_base = save_iovbase;
418 return ret;
419}
420
421/*
422 * Set socket snd and rcv buffer lengths
423 */
424static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
425 unsigned int rcv)
426{
427#if 0
428 mm_segment_t oldfs;
429 oldfs = get_fs(); set_fs(KERNEL_DS);
430 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
431 (char*)&snd, sizeof(snd));
432 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
433 (char*)&rcv, sizeof(rcv));
434#else
435 /* sock_setsockopt limits use to sysctl_?mem_max,
436 * which isn't acceptable. Until that is made conditional
437 * on not having CAP_SYS_RESOURCE or similar, we go direct...
438 * DaveM said I could!
439 */
440 lock_sock(sock->sk);
441 sock->sk->sk_sndbuf = snd * 2;
442 sock->sk->sk_rcvbuf = rcv * 2;
443 sock->sk->sk_write_space(sock->sk);
444 release_sock(sock->sk);
445#endif
446}
447/*
448 * INET callback when data has been received on the socket.
449 */
450static void svc_udp_data_ready(struct sock *sk, int count)
451{
452 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
453 wait_queue_head_t *wq = sk_sleep(sk);
454
455 if (svsk) {
456 dprintk("svc: socket %p(inet %p), count=%d, busy=%d\n",
457 svsk, sk, count,
458 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
459 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
460 svc_xprt_enqueue(&svsk->sk_xprt);
461 }
462 if (wq && waitqueue_active(wq))
463 wake_up_interruptible(wq);
464}
465
466/*
467 * INET callback when space is newly available on the socket.
468 */
469static void svc_write_space(struct sock *sk)
470{
471 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
472 wait_queue_head_t *wq = sk_sleep(sk);
473
474 if (svsk) {
475 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
476 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
477 svc_xprt_enqueue(&svsk->sk_xprt);
478 }
479
480 if (wq && waitqueue_active(wq)) {
481 dprintk("RPC svc_write_space: someone sleeping on %p\n",
482 svsk);
483 wake_up_interruptible(wq);
484 }
485}
486
487static void svc_tcp_write_space(struct sock *sk)
488{
489 struct socket *sock = sk->sk_socket;
490
491 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && sock)
492 clear_bit(SOCK_NOSPACE, &sock->flags);
493 svc_write_space(sk);
494}
495
496/*
497 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
498 */
499static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
500 struct cmsghdr *cmh)
501{
502 struct in_pktinfo *pki = CMSG_DATA(cmh);
503 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
504
505 if (cmh->cmsg_type != IP_PKTINFO)
506 return 0;
507
508 daddr->sin_family = AF_INET;
509 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
510 return 1;
511}
512
513/*
514 * See net/ipv6/datagram.c : datagram_recv_ctl
515 */
516static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
517 struct cmsghdr *cmh)
518{
519 struct in6_pktinfo *pki = CMSG_DATA(cmh);
520 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
521
522 if (cmh->cmsg_type != IPV6_PKTINFO)
523 return 0;
524
525 daddr->sin6_family = AF_INET6;
526 daddr->sin6_addr = pki->ipi6_addr;
527 daddr->sin6_scope_id = pki->ipi6_ifindex;
528 return 1;
529}
530
531/*
532 * Copy the UDP datagram's destination address to the rqstp structure.
533 * The 'destination' address in this case is the address to which the
534 * peer sent the datagram, i.e. our local address. For multihomed
535 * hosts, this can change from msg to msg. Note that only the IP
536 * address changes, the port number should remain the same.
537 */
538static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
539 struct cmsghdr *cmh)
540{
541 switch (cmh->cmsg_level) {
542 case SOL_IP:
543 return svc_udp_get_dest_address4(rqstp, cmh);
544 case SOL_IPV6:
545 return svc_udp_get_dest_address6(rqstp, cmh);
546 }
547
548 return 0;
549}
550
551/*
552 * Receive a datagram from a UDP socket.
553 */
554static int svc_udp_recvfrom(struct svc_rqst *rqstp)
555{
556 struct svc_sock *svsk =
557 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
558 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
559 struct sk_buff *skb;
560 union {
561 struct cmsghdr hdr;
562 long all[SVC_PKTINFO_SPACE / sizeof(long)];
563 } buffer;
564 struct cmsghdr *cmh = &buffer.hdr;
565 struct msghdr msg = {
566 .msg_name = svc_addr(rqstp),
567 .msg_control = cmh,
568 .msg_controllen = sizeof(buffer),
569 .msg_flags = MSG_DONTWAIT,
570 };
571 size_t len;
572 int err;
573
574 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
575 /* udp sockets need large rcvbuf as all pending
576 * requests are still in that buffer. sndbuf must
577 * also be large enough that there is enough space
578 * for one reply per thread. We count all threads
579 * rather than threads in a particular pool, which
580 * provides an upper bound on the number of threads
581 * which will access the socket.
582 */
583 svc_sock_setbufsize(svsk->sk_sock,
584 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
585 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
586
587 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
588 skb = NULL;
589 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
590 0, 0, MSG_PEEK | MSG_DONTWAIT);
591 if (err >= 0)
592 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
593
594 if (skb == NULL) {
595 if (err != -EAGAIN) {
596 /* possibly an icmp error */
597 dprintk("svc: recvfrom returned error %d\n", -err);
598 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
599 }
600 return -EAGAIN;
601 }
602 len = svc_addr_len(svc_addr(rqstp));
603 if (len == 0)
604 return -EAFNOSUPPORT;
605 rqstp->rq_addrlen = len;
606 if (skb->tstamp.tv64 == 0) {
607 skb->tstamp = ktime_get_real();
608 /* Don't enable netstamp, sunrpc doesn't
609 need that much accuracy */
610 }
611 svsk->sk_sk->sk_stamp = skb->tstamp;
612 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
613
614 len = skb->len - sizeof(struct udphdr);
615 rqstp->rq_arg.len = len;
616
617 rqstp->rq_prot = IPPROTO_UDP;
618
619 if (!svc_udp_get_dest_address(rqstp, cmh)) {
620 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
621 cmh->cmsg_level, cmh->cmsg_type);
622 skb_free_datagram_locked(svsk->sk_sk, skb);
623 return 0;
624 }
625 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
626
627 if (skb_is_nonlinear(skb)) {
628 /* we have to copy */
629 local_bh_disable();
630 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
631 local_bh_enable();
632 /* checksum error */
633 skb_free_datagram_locked(svsk->sk_sk, skb);
634 return 0;
635 }
636 local_bh_enable();
637 skb_free_datagram_locked(svsk->sk_sk, skb);
638 } else {
639 /* we can use it in-place */
640 rqstp->rq_arg.head[0].iov_base = skb->data +
641 sizeof(struct udphdr);
642 rqstp->rq_arg.head[0].iov_len = len;
643 if (skb_checksum_complete(skb)) {
644 skb_free_datagram_locked(svsk->sk_sk, skb);
645 return 0;
646 }
647 rqstp->rq_xprt_ctxt = skb;
648 }
649
650 rqstp->rq_arg.page_base = 0;
651 if (len <= rqstp->rq_arg.head[0].iov_len) {
652 rqstp->rq_arg.head[0].iov_len = len;
653 rqstp->rq_arg.page_len = 0;
654 rqstp->rq_respages = rqstp->rq_pages+1;
655 } else {
656 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
657 rqstp->rq_respages = rqstp->rq_pages + 1 +
658 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
659 }
660
661 if (serv->sv_stats)
662 serv->sv_stats->netudpcnt++;
663
664 return len;
665}
666
667static int
668svc_udp_sendto(struct svc_rqst *rqstp)
669{
670 int error;
671
672 error = svc_sendto(rqstp, &rqstp->rq_res);
673 if (error == -ECONNREFUSED)
674 /* ICMP error on earlier request. */
675 error = svc_sendto(rqstp, &rqstp->rq_res);
676
677 return error;
678}
679
680static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
681{
682}
683
684static int svc_udp_has_wspace(struct svc_xprt *xprt)
685{
686 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
687 struct svc_serv *serv = xprt->xpt_server;
688 unsigned long required;
689
690 /*
691 * Set the SOCK_NOSPACE flag before checking the available
692 * sock space.
693 */
694 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
695 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
696 if (required*2 > sock_wspace(svsk->sk_sk))
697 return 0;
698 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
699 return 1;
700}
701
702static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
703{
704 BUG();
705 return NULL;
706}
707
708static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
709 struct net *net,
710 struct sockaddr *sa, int salen,
711 int flags)
712{
713 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
714}
715
716static struct svc_xprt_ops svc_udp_ops = {
717 .xpo_create = svc_udp_create,
718 .xpo_recvfrom = svc_udp_recvfrom,
719 .xpo_sendto = svc_udp_sendto,
720 .xpo_release_rqst = svc_release_skb,
721 .xpo_detach = svc_sock_detach,
722 .xpo_free = svc_sock_free,
723 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
724 .xpo_has_wspace = svc_udp_has_wspace,
725 .xpo_accept = svc_udp_accept,
726};
727
728static struct svc_xprt_class svc_udp_class = {
729 .xcl_name = "udp",
730 .xcl_owner = THIS_MODULE,
731 .xcl_ops = &svc_udp_ops,
732 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
733};
734
735static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
736{
737 int err, level, optname, one = 1;
738
739 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
740 &svsk->sk_xprt, serv);
741 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
742 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
743 svsk->sk_sk->sk_write_space = svc_write_space;
744
745 /* initialise setting must have enough space to
746 * receive and respond to one request.
747 * svc_udp_recvfrom will re-adjust if necessary
748 */
749 svc_sock_setbufsize(svsk->sk_sock,
750 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
751 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
752
753 /* data might have come in before data_ready set up */
754 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
755 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
756
757 /* make sure we get destination address info */
758 switch (svsk->sk_sk->sk_family) {
759 case AF_INET:
760 level = SOL_IP;
761 optname = IP_PKTINFO;
762 break;
763 case AF_INET6:
764 level = SOL_IPV6;
765 optname = IPV6_RECVPKTINFO;
766 break;
767 default:
768 BUG();
769 }
770 err = kernel_setsockopt(svsk->sk_sock, level, optname,
771 (char *)&one, sizeof(one));
772 dprintk("svc: kernel_setsockopt returned %d\n", err);
773}
774
775/*
776 * A data_ready event on a listening socket means there's a connection
777 * pending. Do not use state_change as a substitute for it.
778 */
779static void svc_tcp_listen_data_ready(struct sock *sk, int count_unused)
780{
781 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
782 wait_queue_head_t *wq;
783
784 dprintk("svc: socket %p TCP (listen) state change %d\n",
785 sk, sk->sk_state);
786
787 /*
788 * This callback may called twice when a new connection
789 * is established as a child socket inherits everything
790 * from a parent LISTEN socket.
791 * 1) data_ready method of the parent socket will be called
792 * when one of child sockets become ESTABLISHED.
793 * 2) data_ready method of the child socket may be called
794 * when it receives data before the socket is accepted.
795 * In case of 2, we should ignore it silently.
796 */
797 if (sk->sk_state == TCP_LISTEN) {
798 if (svsk) {
799 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
800 svc_xprt_enqueue(&svsk->sk_xprt);
801 } else
802 printk("svc: socket %p: no user data\n", sk);
803 }
804
805 wq = sk_sleep(sk);
806 if (wq && waitqueue_active(wq))
807 wake_up_interruptible_all(wq);
808}
809
810/*
811 * A state change on a connected socket means it's dying or dead.
812 */
813static void svc_tcp_state_change(struct sock *sk)
814{
815 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
816 wait_queue_head_t *wq = sk_sleep(sk);
817
818 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
819 sk, sk->sk_state, sk->sk_user_data);
820
821 if (!svsk)
822 printk("svc: socket %p: no user data\n", sk);
823 else {
824 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
825 svc_xprt_enqueue(&svsk->sk_xprt);
826 }
827 if (wq && waitqueue_active(wq))
828 wake_up_interruptible_all(wq);
829}
830
831static void svc_tcp_data_ready(struct sock *sk, int count)
832{
833 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
834 wait_queue_head_t *wq = sk_sleep(sk);
835
836 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
837 sk, sk->sk_user_data);
838 if (svsk) {
839 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
840 svc_xprt_enqueue(&svsk->sk_xprt);
841 }
842 if (wq && waitqueue_active(wq))
843 wake_up_interruptible(wq);
844}
845
846/*
847 * Accept a TCP connection
848 */
849static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
850{
851 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
852 struct sockaddr_storage addr;
853 struct sockaddr *sin = (struct sockaddr *) &addr;
854 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
855 struct socket *sock = svsk->sk_sock;
856 struct socket *newsock;
857 struct svc_sock *newsvsk;
858 int err, slen;
859 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
860
861 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
862 if (!sock)
863 return NULL;
864
865 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
866 err = kernel_accept(sock, &newsock, O_NONBLOCK);
867 if (err < 0) {
868 if (err == -ENOMEM)
869 printk(KERN_WARNING "%s: no more sockets!\n",
870 serv->sv_name);
871 else if (err != -EAGAIN)
872 net_warn_ratelimited("%s: accept failed (err %d)!\n",
873 serv->sv_name, -err);
874 return NULL;
875 }
876 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
877
878 err = kernel_getpeername(newsock, sin, &slen);
879 if (err < 0) {
880 net_warn_ratelimited("%s: peername failed (err %d)!\n",
881 serv->sv_name, -err);
882 goto failed; /* aborted connection or whatever */
883 }
884
885 /* Ideally, we would want to reject connections from unauthorized
886 * hosts here, but when we get encryption, the IP of the host won't
887 * tell us anything. For now just warn about unpriv connections.
888 */
889 if (!svc_port_is_privileged(sin)) {
890 dprintk(KERN_WARNING
891 "%s: connect from unprivileged port: %s\n",
892 serv->sv_name,
893 __svc_print_addr(sin, buf, sizeof(buf)));
894 }
895 dprintk("%s: connect from %s\n", serv->sv_name,
896 __svc_print_addr(sin, buf, sizeof(buf)));
897
898 /* make sure that a write doesn't block forever when
899 * low on memory
900 */
901 newsock->sk->sk_sndtimeo = HZ*30;
902
903 if (!(newsvsk = svc_setup_socket(serv, newsock, &err,
904 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY))))
905 goto failed;
906 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
907 err = kernel_getsockname(newsock, sin, &slen);
908 if (unlikely(err < 0)) {
909 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
910 slen = offsetof(struct sockaddr, sa_data);
911 }
912 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
913
914 if (serv->sv_stats)
915 serv->sv_stats->nettcpconn++;
916
917 return &newsvsk->sk_xprt;
918
919failed:
920 sock_release(newsock);
921 return NULL;
922}
923
924static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
925{
926 unsigned int i, len, npages;
927
928 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
929 return 0;
930 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
931 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
932 for (i = 0; i < npages; i++) {
933 if (rqstp->rq_pages[i] != NULL)
934 put_page(rqstp->rq_pages[i]);
935 BUG_ON(svsk->sk_pages[i] == NULL);
936 rqstp->rq_pages[i] = svsk->sk_pages[i];
937 svsk->sk_pages[i] = NULL;
938 }
939 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
940 return len;
941}
942
943static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
944{
945 unsigned int i, len, npages;
946
947 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
948 return;
949 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
950 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 for (i = 0; i < npages; i++) {
952 svsk->sk_pages[i] = rqstp->rq_pages[i];
953 rqstp->rq_pages[i] = NULL;
954 }
955}
956
957static void svc_tcp_clear_pages(struct svc_sock *svsk)
958{
959 unsigned int i, len, npages;
960
961 if (svsk->sk_tcplen <= sizeof(rpc_fraghdr))
962 goto out;
963 len = svsk->sk_tcplen - sizeof(rpc_fraghdr);
964 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
965 for (i = 0; i < npages; i++) {
966 BUG_ON(svsk->sk_pages[i] == NULL);
967 put_page(svsk->sk_pages[i]);
968 svsk->sk_pages[i] = NULL;
969 }
970out:
971 svsk->sk_tcplen = 0;
972}
973
974/*
975 * Receive data.
976 * If we haven't gotten the record length yet, get the next four bytes.
977 * Otherwise try to gobble up as much as possible up to the complete
978 * record length.
979 */
980static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
981{
982 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
983 unsigned int want;
984 int len;
985
986 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
987
988 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
989 struct kvec iov;
990
991 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
992 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
993 iov.iov_len = want;
994 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
995 goto error;
996 svsk->sk_tcplen += len;
997
998 if (len < want) {
999 dprintk("svc: short recvfrom while reading record "
1000 "length (%d of %d)\n", len, want);
1001 return -EAGAIN;
1002 }
1003
1004 svsk->sk_reclen = ntohl(svsk->sk_reclen);
1005 if (!(svsk->sk_reclen & RPC_LAST_STREAM_FRAGMENT)) {
1006 /* FIXME: technically, a record can be fragmented,
1007 * and non-terminal fragments will not have the top
1008 * bit set in the fragment length header.
1009 * But apparently no known nfs clients send fragmented
1010 * records. */
1011 net_notice_ratelimited("RPC: multiple fragments per record not supported\n");
1012 goto err_delete;
1013 }
1014
1015 svsk->sk_reclen &= RPC_FRAGMENT_SIZE_MASK;
1016 dprintk("svc: TCP record, %d bytes\n", svsk->sk_reclen);
1017 if (svsk->sk_reclen > serv->sv_max_mesg) {
1018 net_notice_ratelimited("RPC: fragment too large: 0x%08lx\n",
1019 (unsigned long)svsk->sk_reclen);
1020 goto err_delete;
1021 }
1022 }
1023
1024 if (svsk->sk_reclen < 8)
1025 goto err_delete; /* client is nuts. */
1026
1027 len = svsk->sk_reclen;
1028
1029 return len;
1030error:
1031 dprintk("RPC: TCP recv_record got %d\n", len);
1032 return len;
1033err_delete:
1034 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1035 return -EAGAIN;
1036}
1037
1038static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1039{
1040 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1041 struct rpc_rqst *req = NULL;
1042 struct kvec *src, *dst;
1043 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1044 __be32 xid;
1045 __be32 calldir;
1046
1047 xid = *p++;
1048 calldir = *p;
1049
1050 if (bc_xprt)
1051 req = xprt_lookup_rqst(bc_xprt, xid);
1052
1053 if (!req) {
1054 printk(KERN_NOTICE
1055 "%s: Got unrecognized reply: "
1056 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1057 __func__, ntohl(calldir),
1058 bc_xprt, xid);
1059 return -EAGAIN;
1060 }
1061
1062 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1063 /*
1064 * XXX!: cheating for now! Only copying HEAD.
1065 * But we know this is good enough for now (in fact, for any
1066 * callback reply in the forseeable future).
1067 */
1068 dst = &req->rq_private_buf.head[0];
1069 src = &rqstp->rq_arg.head[0];
1070 if (dst->iov_len < src->iov_len)
1071 return -EAGAIN; /* whatever; just giving up. */
1072 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1073 xprt_complete_rqst(req->rq_task, svsk->sk_reclen);
1074 rqstp->rq_arg.len = 0;
1075 return 0;
1076}
1077
1078static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1079{
1080 int i = 0;
1081 int t = 0;
1082
1083 while (t < len) {
1084 vec[i].iov_base = page_address(pages[i]);
1085 vec[i].iov_len = PAGE_SIZE;
1086 i++;
1087 t += PAGE_SIZE;
1088 }
1089 return i;
1090}
1091
1092
1093/*
1094 * Receive data from a TCP socket.
1095 */
1096static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1097{
1098 struct svc_sock *svsk =
1099 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1100 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1101 int len;
1102 struct kvec *vec;
1103 unsigned int want, base;
1104 __be32 *p;
1105 __be32 calldir;
1106 int pnum;
1107
1108 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1109 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1110 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1111 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1112
1113 len = svc_tcp_recv_record(svsk, rqstp);
1114 if (len < 0)
1115 goto error;
1116
1117 base = svc_tcp_restore_pages(svsk, rqstp);
1118 want = svsk->sk_reclen - base;
1119
1120 vec = rqstp->rq_vec;
1121
1122 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1123 svsk->sk_reclen);
1124
1125 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1126
1127 /* Now receive data */
1128 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1129 if (len >= 0)
1130 svsk->sk_tcplen += len;
1131 if (len != want) {
1132 svc_tcp_save_pages(svsk, rqstp);
1133 if (len < 0 && len != -EAGAIN)
1134 goto err_other;
1135 dprintk("svc: incomplete TCP record (%d of %d)\n",
1136 svsk->sk_tcplen, svsk->sk_reclen);
1137 goto err_noclose;
1138 }
1139
1140 rqstp->rq_arg.len = svsk->sk_reclen;
1141 rqstp->rq_arg.page_base = 0;
1142 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1143 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1144 rqstp->rq_arg.page_len = 0;
1145 } else
1146 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1147
1148 rqstp->rq_xprt_ctxt = NULL;
1149 rqstp->rq_prot = IPPROTO_TCP;
1150
1151 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1152 calldir = p[1];
1153 if (calldir)
1154 len = receive_cb_reply(svsk, rqstp);
1155
1156 /* Reset TCP read info */
1157 svsk->sk_reclen = 0;
1158 svsk->sk_tcplen = 0;
1159 /* If we have more data, signal svc_xprt_enqueue() to try again */
1160 if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1161 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1162
1163 if (len < 0)
1164 goto error;
1165
1166 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1167 if (serv->sv_stats)
1168 serv->sv_stats->nettcpcnt++;
1169
1170 dprintk("svc: TCP complete record (%d bytes)\n", rqstp->rq_arg.len);
1171 return rqstp->rq_arg.len;
1172
1173error:
1174 if (len != -EAGAIN)
1175 goto err_other;
1176 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1177 return -EAGAIN;
1178err_other:
1179 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1180 svsk->sk_xprt.xpt_server->sv_name, -len);
1181 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1182err_noclose:
1183 return -EAGAIN; /* record not complete */
1184}
1185
1186/*
1187 * Send out data on TCP socket.
1188 */
1189static int svc_tcp_sendto(struct svc_rqst *rqstp)
1190{
1191 struct xdr_buf *xbufp = &rqstp->rq_res;
1192 int sent;
1193 __be32 reclen;
1194
1195 /* Set up the first element of the reply kvec.
1196 * Any other kvecs that may be in use have been taken
1197 * care of by the server implementation itself.
1198 */
1199 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1200 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1201
1202 sent = svc_sendto(rqstp, &rqstp->rq_res);
1203 if (sent != xbufp->len) {
1204 printk(KERN_NOTICE
1205 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1206 "- shutting down socket\n",
1207 rqstp->rq_xprt->xpt_server->sv_name,
1208 (sent<0)?"got error":"sent only",
1209 sent, xbufp->len);
1210 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1211 svc_xprt_enqueue(rqstp->rq_xprt);
1212 sent = -EAGAIN;
1213 }
1214 return sent;
1215}
1216
1217/*
1218 * Setup response header. TCP has a 4B record length field.
1219 */
1220static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1221{
1222 struct kvec *resv = &rqstp->rq_res.head[0];
1223
1224 /* tcp needs a space for the record length... */
1225 svc_putnl(resv, 0);
1226}
1227
1228static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1229{
1230 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1231 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1232 int required;
1233
1234 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1235 return 1;
1236 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1237 if (sk_stream_wspace(svsk->sk_sk) >= required)
1238 return 1;
1239 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1240 return 0;
1241}
1242
1243static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1244 struct net *net,
1245 struct sockaddr *sa, int salen,
1246 int flags)
1247{
1248 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1249}
1250
1251#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1252static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1253 struct net *, struct sockaddr *,
1254 int, int);
1255static void svc_bc_sock_free(struct svc_xprt *xprt);
1256
1257static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1258 struct net *net,
1259 struct sockaddr *sa, int salen,
1260 int flags)
1261{
1262 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1263}
1264
1265static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1266{
1267}
1268
1269static struct svc_xprt_ops svc_tcp_bc_ops = {
1270 .xpo_create = svc_bc_tcp_create,
1271 .xpo_detach = svc_bc_tcp_sock_detach,
1272 .xpo_free = svc_bc_sock_free,
1273 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1274};
1275
1276static struct svc_xprt_class svc_tcp_bc_class = {
1277 .xcl_name = "tcp-bc",
1278 .xcl_owner = THIS_MODULE,
1279 .xcl_ops = &svc_tcp_bc_ops,
1280 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1281};
1282
1283static void svc_init_bc_xprt_sock(void)
1284{
1285 svc_reg_xprt_class(&svc_tcp_bc_class);
1286}
1287
1288static void svc_cleanup_bc_xprt_sock(void)
1289{
1290 svc_unreg_xprt_class(&svc_tcp_bc_class);
1291}
1292#else /* CONFIG_SUNRPC_BACKCHANNEL */
1293static void svc_init_bc_xprt_sock(void)
1294{
1295}
1296
1297static void svc_cleanup_bc_xprt_sock(void)
1298{
1299}
1300#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1301
1302static struct svc_xprt_ops svc_tcp_ops = {
1303 .xpo_create = svc_tcp_create,
1304 .xpo_recvfrom = svc_tcp_recvfrom,
1305 .xpo_sendto = svc_tcp_sendto,
1306 .xpo_release_rqst = svc_release_skb,
1307 .xpo_detach = svc_tcp_sock_detach,
1308 .xpo_free = svc_sock_free,
1309 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1310 .xpo_has_wspace = svc_tcp_has_wspace,
1311 .xpo_accept = svc_tcp_accept,
1312};
1313
1314static struct svc_xprt_class svc_tcp_class = {
1315 .xcl_name = "tcp",
1316 .xcl_owner = THIS_MODULE,
1317 .xcl_ops = &svc_tcp_ops,
1318 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1319};
1320
1321void svc_init_xprt_sock(void)
1322{
1323 svc_reg_xprt_class(&svc_tcp_class);
1324 svc_reg_xprt_class(&svc_udp_class);
1325 svc_init_bc_xprt_sock();
1326}
1327
1328void svc_cleanup_xprt_sock(void)
1329{
1330 svc_unreg_xprt_class(&svc_tcp_class);
1331 svc_unreg_xprt_class(&svc_udp_class);
1332 svc_cleanup_bc_xprt_sock();
1333}
1334
1335static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1336{
1337 struct sock *sk = svsk->sk_sk;
1338
1339 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1340 &svsk->sk_xprt, serv);
1341 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1342 if (sk->sk_state == TCP_LISTEN) {
1343 dprintk("setting up TCP socket for listening\n");
1344 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1345 sk->sk_data_ready = svc_tcp_listen_data_ready;
1346 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1347 } else {
1348 dprintk("setting up TCP socket for reading\n");
1349 sk->sk_state_change = svc_tcp_state_change;
1350 sk->sk_data_ready = svc_tcp_data_ready;
1351 sk->sk_write_space = svc_tcp_write_space;
1352
1353 svsk->sk_reclen = 0;
1354 svsk->sk_tcplen = 0;
1355 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1356
1357 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1358
1359 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1360 if (sk->sk_state != TCP_ESTABLISHED)
1361 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1362 }
1363}
1364
1365void svc_sock_update_bufs(struct svc_serv *serv)
1366{
1367 /*
1368 * The number of server threads has changed. Update
1369 * rcvbuf and sndbuf accordingly on all sockets
1370 */
1371 struct svc_sock *svsk;
1372
1373 spin_lock_bh(&serv->sv_lock);
1374 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1375 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1376 spin_unlock_bh(&serv->sv_lock);
1377}
1378EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1379
1380/*
1381 * Initialize socket for RPC use and create svc_sock struct
1382 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1383 */
1384static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1385 struct socket *sock,
1386 int *errp, int flags)
1387{
1388 struct svc_sock *svsk;
1389 struct sock *inet;
1390 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1391
1392 dprintk("svc: svc_setup_socket %p\n", sock);
1393 if (!(svsk = kzalloc(sizeof(*svsk), GFP_KERNEL))) {
1394 *errp = -ENOMEM;
1395 return NULL;
1396 }
1397
1398 inet = sock->sk;
1399
1400 /* Register socket with portmapper */
1401 if (*errp >= 0 && pmap_register)
1402 *errp = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1403 inet->sk_protocol,
1404 ntohs(inet_sk(inet)->inet_sport));
1405
1406 if (*errp < 0) {
1407 kfree(svsk);
1408 return NULL;
1409 }
1410
1411 inet->sk_user_data = svsk;
1412 svsk->sk_sock = sock;
1413 svsk->sk_sk = inet;
1414 svsk->sk_ostate = inet->sk_state_change;
1415 svsk->sk_odata = inet->sk_data_ready;
1416 svsk->sk_owspace = inet->sk_write_space;
1417
1418 /* Initialize the socket */
1419 if (sock->type == SOCK_DGRAM)
1420 svc_udp_init(svsk, serv);
1421 else {
1422 /* initialise setting must have enough space to
1423 * receive and respond to one request.
1424 */
1425 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1426 4 * serv->sv_max_mesg);
1427 svc_tcp_init(svsk, serv);
1428 }
1429
1430 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1431 svsk, svsk->sk_sk);
1432
1433 return svsk;
1434}
1435
1436/**
1437 * svc_addsock - add a listener socket to an RPC service
1438 * @serv: pointer to RPC service to which to add a new listener
1439 * @fd: file descriptor of the new listener
1440 * @name_return: pointer to buffer to fill in with name of listener
1441 * @len: size of the buffer
1442 *
1443 * Fills in socket name and returns positive length of name if successful.
1444 * Name is terminated with '\n'. On error, returns a negative errno
1445 * value.
1446 */
1447int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1448 const size_t len)
1449{
1450 int err = 0;
1451 struct socket *so = sockfd_lookup(fd, &err);
1452 struct svc_sock *svsk = NULL;
1453
1454 if (!so)
1455 return err;
1456 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1457 err = -EAFNOSUPPORT;
1458 else if (so->sk->sk_protocol != IPPROTO_TCP &&
1459 so->sk->sk_protocol != IPPROTO_UDP)
1460 err = -EPROTONOSUPPORT;
1461 else if (so->state > SS_UNCONNECTED)
1462 err = -EISCONN;
1463 else {
1464 if (!try_module_get(THIS_MODULE))
1465 err = -ENOENT;
1466 else
1467 svsk = svc_setup_socket(serv, so, &err,
1468 SVC_SOCK_DEFAULTS);
1469 if (svsk) {
1470 struct sockaddr_storage addr;
1471 struct sockaddr *sin = (struct sockaddr *)&addr;
1472 int salen;
1473 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1474 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1475 clear_bit(XPT_TEMP, &svsk->sk_xprt.xpt_flags);
1476 spin_lock_bh(&serv->sv_lock);
1477 list_add(&svsk->sk_xprt.xpt_list, &serv->sv_permsocks);
1478 spin_unlock_bh(&serv->sv_lock);
1479 svc_xprt_received(&svsk->sk_xprt);
1480 err = 0;
1481 } else
1482 module_put(THIS_MODULE);
1483 }
1484 if (err) {
1485 sockfd_put(so);
1486 return err;
1487 }
1488 return svc_one_sock_name(svsk, name_return, len);
1489}
1490EXPORT_SYMBOL_GPL(svc_addsock);
1491
1492/*
1493 * Create socket for RPC service.
1494 */
1495static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1496 int protocol,
1497 struct net *net,
1498 struct sockaddr *sin, int len,
1499 int flags)
1500{
1501 struct svc_sock *svsk;
1502 struct socket *sock;
1503 int error;
1504 int type;
1505 struct sockaddr_storage addr;
1506 struct sockaddr *newsin = (struct sockaddr *)&addr;
1507 int newlen;
1508 int family;
1509 int val;
1510 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1511
1512 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1513 serv->sv_program->pg_name, protocol,
1514 __svc_print_addr(sin, buf, sizeof(buf)));
1515
1516 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1517 printk(KERN_WARNING "svc: only UDP and TCP "
1518 "sockets supported\n");
1519 return ERR_PTR(-EINVAL);
1520 }
1521
1522 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1523 switch (sin->sa_family) {
1524 case AF_INET6:
1525 family = PF_INET6;
1526 break;
1527 case AF_INET:
1528 family = PF_INET;
1529 break;
1530 default:
1531 return ERR_PTR(-EINVAL);
1532 }
1533
1534 error = __sock_create(net, family, type, protocol, &sock, 1);
1535 if (error < 0)
1536 return ERR_PTR(error);
1537
1538 svc_reclassify_socket(sock);
1539
1540 /*
1541 * If this is an PF_INET6 listener, we want to avoid
1542 * getting requests from IPv4 remotes. Those should
1543 * be shunted to a PF_INET listener via rpcbind.
1544 */
1545 val = 1;
1546 if (family == PF_INET6)
1547 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1548 (char *)&val, sizeof(val));
1549
1550 if (type == SOCK_STREAM)
1551 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1552 error = kernel_bind(sock, sin, len);
1553 if (error < 0)
1554 goto bummer;
1555
1556 newlen = len;
1557 error = kernel_getsockname(sock, newsin, &newlen);
1558 if (error < 0)
1559 goto bummer;
1560
1561 if (protocol == IPPROTO_TCP) {
1562 if ((error = kernel_listen(sock, 64)) < 0)
1563 goto bummer;
1564 }
1565
1566 if ((svsk = svc_setup_socket(serv, sock, &error, flags)) != NULL) {
1567 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1568 return (struct svc_xprt *)svsk;
1569 }
1570
1571bummer:
1572 dprintk("svc: svc_create_socket error = %d\n", -error);
1573 sock_release(sock);
1574 return ERR_PTR(error);
1575}
1576
1577/*
1578 * Detach the svc_sock from the socket so that no
1579 * more callbacks occur.
1580 */
1581static void svc_sock_detach(struct svc_xprt *xprt)
1582{
1583 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1584 struct sock *sk = svsk->sk_sk;
1585 wait_queue_head_t *wq;
1586
1587 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1588
1589 /* put back the old socket callbacks */
1590 sk->sk_state_change = svsk->sk_ostate;
1591 sk->sk_data_ready = svsk->sk_odata;
1592 sk->sk_write_space = svsk->sk_owspace;
1593
1594 wq = sk_sleep(sk);
1595 if (wq && waitqueue_active(wq))
1596 wake_up_interruptible(wq);
1597}
1598
1599/*
1600 * Disconnect the socket, and reset the callbacks
1601 */
1602static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1603{
1604 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1605
1606 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1607
1608 svc_sock_detach(xprt);
1609
1610 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1611 svc_tcp_clear_pages(svsk);
1612 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1613 }
1614}
1615
1616/*
1617 * Free the svc_sock's socket resources and the svc_sock itself.
1618 */
1619static void svc_sock_free(struct svc_xprt *xprt)
1620{
1621 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1622 dprintk("svc: svc_sock_free(%p)\n", svsk);
1623
1624 if (svsk->sk_sock->file)
1625 sockfd_put(svsk->sk_sock);
1626 else
1627 sock_release(svsk->sk_sock);
1628 kfree(svsk);
1629}
1630
1631#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1632/*
1633 * Create a back channel svc_xprt which shares the fore channel socket.
1634 */
1635static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1636 int protocol,
1637 struct net *net,
1638 struct sockaddr *sin, int len,
1639 int flags)
1640{
1641 struct svc_sock *svsk;
1642 struct svc_xprt *xprt;
1643
1644 if (protocol != IPPROTO_TCP) {
1645 printk(KERN_WARNING "svc: only TCP sockets"
1646 " supported on shared back channel\n");
1647 return ERR_PTR(-EINVAL);
1648 }
1649
1650 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1651 if (!svsk)
1652 return ERR_PTR(-ENOMEM);
1653
1654 xprt = &svsk->sk_xprt;
1655 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1656
1657 serv->sv_bc_xprt = xprt;
1658
1659 return xprt;
1660}
1661
1662/*
1663 * Free a back channel svc_sock.
1664 */
1665static void svc_bc_sock_free(struct svc_xprt *xprt)
1666{
1667 if (xprt)
1668 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1669}
1670#endif /* CONFIG_SUNRPC_BACKCHANNEL */