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