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