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