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