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 <linux/bvec.h>
40
41#include <net/sock.h>
42#include <net/checksum.h>
43#include <net/ip.h>
44#include <net/ipv6.h>
45#include <net/udp.h>
46#include <net/tcp.h>
47#include <net/tcp_states.h>
48#include <net/tls_prot.h>
49#include <net/handshake.h>
50#include <linux/uaccess.h>
51#include <linux/highmem.h>
52#include <asm/ioctls.h>
53#include <linux/key.h>
54
55#include <linux/sunrpc/types.h>
56#include <linux/sunrpc/clnt.h>
57#include <linux/sunrpc/xdr.h>
58#include <linux/sunrpc/msg_prot.h>
59#include <linux/sunrpc/svcsock.h>
60#include <linux/sunrpc/stats.h>
61#include <linux/sunrpc/xprt.h>
62
63#include <trace/events/sock.h>
64#include <trace/events/sunrpc.h>
65
66#include "socklib.h"
67#include "sunrpc.h"
68
69#define RPCDBG_FACILITY RPCDBG_SVCXPRT
70
71/* To-do: to avoid tying up an nfsd thread while waiting for a
72 * handshake request, the request could instead be deferred.
73 */
74enum {
75 SVC_HANDSHAKE_TO = 5U * HZ
76};
77
78static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
79 int flags);
80static int svc_udp_recvfrom(struct svc_rqst *);
81static int svc_udp_sendto(struct svc_rqst *);
82static void svc_sock_detach(struct svc_xprt *);
83static void svc_tcp_sock_detach(struct svc_xprt *);
84static void svc_sock_free(struct svc_xprt *);
85
86static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
87 struct net *, struct sockaddr *,
88 int, int);
89#ifdef CONFIG_DEBUG_LOCK_ALLOC
90static struct lock_class_key svc_key[2];
91static struct lock_class_key svc_slock_key[2];
92
93static void svc_reclassify_socket(struct socket *sock)
94{
95 struct sock *sk = sock->sk;
96
97 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
98 return;
99
100 switch (sk->sk_family) {
101 case AF_INET:
102 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
103 &svc_slock_key[0],
104 "sk_xprt.xpt_lock-AF_INET-NFSD",
105 &svc_key[0]);
106 break;
107
108 case AF_INET6:
109 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
110 &svc_slock_key[1],
111 "sk_xprt.xpt_lock-AF_INET6-NFSD",
112 &svc_key[1]);
113 break;
114
115 default:
116 BUG();
117 }
118}
119#else
120static void svc_reclassify_socket(struct socket *sock)
121{
122}
123#endif
124
125/**
126 * svc_tcp_release_ctxt - Release transport-related resources
127 * @xprt: the transport which owned the context
128 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
129 *
130 */
131static void svc_tcp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
132{
133}
134
135/**
136 * svc_udp_release_ctxt - Release transport-related resources
137 * @xprt: the transport which owned the context
138 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
139 *
140 */
141static void svc_udp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
142{
143 struct sk_buff *skb = ctxt;
144
145 if (skb)
146 consume_skb(skb);
147}
148
149union svc_pktinfo_u {
150 struct in_pktinfo pkti;
151 struct in6_pktinfo pkti6;
152};
153#define SVC_PKTINFO_SPACE \
154 CMSG_SPACE(sizeof(union svc_pktinfo_u))
155
156static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
157{
158 struct svc_sock *svsk =
159 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
160 switch (svsk->sk_sk->sk_family) {
161 case AF_INET: {
162 struct in_pktinfo *pki = CMSG_DATA(cmh);
163
164 cmh->cmsg_level = SOL_IP;
165 cmh->cmsg_type = IP_PKTINFO;
166 pki->ipi_ifindex = 0;
167 pki->ipi_spec_dst.s_addr =
168 svc_daddr_in(rqstp)->sin_addr.s_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
170 }
171 break;
172
173 case AF_INET6: {
174 struct in6_pktinfo *pki = CMSG_DATA(cmh);
175 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
176
177 cmh->cmsg_level = SOL_IPV6;
178 cmh->cmsg_type = IPV6_PKTINFO;
179 pki->ipi6_ifindex = daddr->sin6_scope_id;
180 pki->ipi6_addr = daddr->sin6_addr;
181 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
182 }
183 break;
184 }
185}
186
187static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
188 unsigned int length)
189{
190 return 0;
191}
192
193/*
194 * Report socket names for nfsdfs
195 */
196static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
197{
198 const struct sock *sk = svsk->sk_sk;
199 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
200 "udp" : "tcp";
201 int len;
202
203 switch (sk->sk_family) {
204 case PF_INET:
205 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
206 proto_name,
207 &inet_sk(sk)->inet_rcv_saddr,
208 inet_sk(sk)->inet_num);
209 break;
210#if IS_ENABLED(CONFIG_IPV6)
211 case PF_INET6:
212 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
213 proto_name,
214 &sk->sk_v6_rcv_saddr,
215 inet_sk(sk)->inet_num);
216 break;
217#endif
218 default:
219 len = snprintf(buf, remaining, "*unknown-%d*\n",
220 sk->sk_family);
221 }
222
223 if (len >= remaining) {
224 *buf = '\0';
225 return -ENAMETOOLONG;
226 }
227 return len;
228}
229
230static int
231svc_tcp_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
232 struct cmsghdr *cmsg, int ret)
233{
234 u8 content_type = tls_get_record_type(sock->sk, cmsg);
235 u8 level, description;
236
237 switch (content_type) {
238 case 0:
239 break;
240 case TLS_RECORD_TYPE_DATA:
241 /* TLS sets EOR at the end of each application data
242 * record, even though there might be more frames
243 * waiting to be decrypted.
244 */
245 msg->msg_flags &= ~MSG_EOR;
246 break;
247 case TLS_RECORD_TYPE_ALERT:
248 tls_alert_recv(sock->sk, msg, &level, &description);
249 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
250 -ENOTCONN : -EAGAIN;
251 break;
252 default:
253 /* discard this record type */
254 ret = -EAGAIN;
255 }
256 return ret;
257}
258
259static int
260svc_tcp_sock_recv_cmsg(struct svc_sock *svsk, struct msghdr *msg)
261{
262 union {
263 struct cmsghdr cmsg;
264 u8 buf[CMSG_SPACE(sizeof(u8))];
265 } u;
266 struct socket *sock = svsk->sk_sock;
267 int ret;
268
269 msg->msg_control = &u;
270 msg->msg_controllen = sizeof(u);
271 ret = sock_recvmsg(sock, msg, MSG_DONTWAIT);
272 if (unlikely(msg->msg_controllen != sizeof(u)))
273 ret = svc_tcp_sock_process_cmsg(sock, msg, &u.cmsg, ret);
274 return ret;
275}
276
277#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
278static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
279{
280 struct bvec_iter bi = {
281 .bi_size = size + seek,
282 };
283 struct bio_vec bv;
284
285 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
286 for_each_bvec(bv, bvec, bi, bi)
287 flush_dcache_page(bv.bv_page);
288}
289#else
290static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
291 size_t seek)
292{
293}
294#endif
295
296/*
297 * Read from @rqstp's transport socket. The incoming message fills whole
298 * pages in @rqstp's rq_pages array until the last page of the message
299 * has been received into a partial page.
300 */
301static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
302 size_t seek)
303{
304 struct svc_sock *svsk =
305 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
306 struct bio_vec *bvec = rqstp->rq_bvec;
307 struct msghdr msg = { NULL };
308 unsigned int i;
309 ssize_t len;
310 size_t t;
311
312 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
313
314 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE)
315 bvec_set_page(&bvec[i], rqstp->rq_pages[i], PAGE_SIZE, 0);
316 rqstp->rq_respages = &rqstp->rq_pages[i];
317 rqstp->rq_next_page = rqstp->rq_respages + 1;
318
319 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen);
320 if (seek) {
321 iov_iter_advance(&msg.msg_iter, seek);
322 buflen -= seek;
323 }
324 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
325 if (len > 0)
326 svc_flush_bvec(bvec, len, seek);
327
328 /* If we read a full record, then assume there may be more
329 * data to read (stream based sockets only!)
330 */
331 if (len == buflen)
332 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
333
334 return len;
335}
336
337/*
338 * Set socket snd and rcv buffer lengths
339 */
340static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
341{
342 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
343 struct socket *sock = svsk->sk_sock;
344
345 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
346
347 lock_sock(sock->sk);
348 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
349 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
350 sock->sk->sk_write_space(sock->sk);
351 release_sock(sock->sk);
352}
353
354static void svc_sock_secure_port(struct svc_rqst *rqstp)
355{
356 if (svc_port_is_privileged(svc_addr(rqstp)))
357 set_bit(RQ_SECURE, &rqstp->rq_flags);
358 else
359 clear_bit(RQ_SECURE, &rqstp->rq_flags);
360}
361
362/*
363 * INET callback when data has been received on the socket.
364 */
365static void svc_data_ready(struct sock *sk)
366{
367 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
368
369 trace_sk_data_ready(sk);
370
371 if (svsk) {
372 /* Refer to svc_setup_socket() for details. */
373 rmb();
374 svsk->sk_odata(sk);
375 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
376 if (test_bit(XPT_HANDSHAKE, &svsk->sk_xprt.xpt_flags))
377 return;
378 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
379 svc_xprt_enqueue(&svsk->sk_xprt);
380 }
381}
382
383/*
384 * INET callback when space is newly available on the socket.
385 */
386static void svc_write_space(struct sock *sk)
387{
388 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
389
390 if (svsk) {
391 /* Refer to svc_setup_socket() for details. */
392 rmb();
393 trace_svcsock_write_space(&svsk->sk_xprt, 0);
394 svsk->sk_owspace(sk);
395 svc_xprt_enqueue(&svsk->sk_xprt);
396 }
397}
398
399static int svc_tcp_has_wspace(struct svc_xprt *xprt)
400{
401 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
402
403 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
404 return 1;
405 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
406}
407
408static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
409{
410 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
411
412 sock_no_linger(svsk->sk_sock->sk);
413}
414
415/**
416 * svc_tcp_handshake_done - Handshake completion handler
417 * @data: address of xprt to wake
418 * @status: status of handshake
419 * @peerid: serial number of key containing the remote peer's identity
420 *
421 * If a security policy is specified as an export option, we don't
422 * have a specific export here to check. So we set a "TLS session
423 * is present" flag on the xprt and let an upper layer enforce local
424 * security policy.
425 */
426static void svc_tcp_handshake_done(void *data, int status, key_serial_t peerid)
427{
428 struct svc_xprt *xprt = data;
429 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
430
431 if (!status) {
432 if (peerid != TLS_NO_PEERID)
433 set_bit(XPT_PEER_AUTH, &xprt->xpt_flags);
434 set_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
435 }
436 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
437 complete_all(&svsk->sk_handshake_done);
438}
439
440/**
441 * svc_tcp_handshake - Perform a transport-layer security handshake
442 * @xprt: connected transport endpoint
443 *
444 */
445static void svc_tcp_handshake(struct svc_xprt *xprt)
446{
447 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
448 struct sock *sk = svsk->sk_sock->sk;
449 struct tls_handshake_args args = {
450 .ta_sock = svsk->sk_sock,
451 .ta_done = svc_tcp_handshake_done,
452 .ta_data = xprt,
453 };
454 int ret;
455
456 trace_svc_tls_upcall(xprt);
457
458 clear_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
459 init_completion(&svsk->sk_handshake_done);
460
461 ret = tls_server_hello_x509(&args, GFP_KERNEL);
462 if (ret) {
463 trace_svc_tls_not_started(xprt);
464 goto out_failed;
465 }
466
467 ret = wait_for_completion_interruptible_timeout(&svsk->sk_handshake_done,
468 SVC_HANDSHAKE_TO);
469 if (ret <= 0) {
470 if (tls_handshake_cancel(sk)) {
471 trace_svc_tls_timed_out(xprt);
472 goto out_close;
473 }
474 }
475
476 if (!test_bit(XPT_TLS_SESSION, &xprt->xpt_flags)) {
477 trace_svc_tls_unavailable(xprt);
478 goto out_close;
479 }
480
481 /* Mark the transport ready in case the remote sent RPC
482 * traffic before the kernel received the handshake
483 * completion downcall.
484 */
485 set_bit(XPT_DATA, &xprt->xpt_flags);
486 svc_xprt_enqueue(xprt);
487 return;
488
489out_close:
490 set_bit(XPT_CLOSE, &xprt->xpt_flags);
491out_failed:
492 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
493 set_bit(XPT_DATA, &xprt->xpt_flags);
494 svc_xprt_enqueue(xprt);
495}
496
497/*
498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
499 */
500static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
501 struct cmsghdr *cmh)
502{
503 struct in_pktinfo *pki = CMSG_DATA(cmh);
504 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
505
506 if (cmh->cmsg_type != IP_PKTINFO)
507 return 0;
508
509 daddr->sin_family = AF_INET;
510 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
511 return 1;
512}
513
514/*
515 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
516 */
517static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
518 struct cmsghdr *cmh)
519{
520 struct in6_pktinfo *pki = CMSG_DATA(cmh);
521 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
522
523 if (cmh->cmsg_type != IPV6_PKTINFO)
524 return 0;
525
526 daddr->sin6_family = AF_INET6;
527 daddr->sin6_addr = pki->ipi6_addr;
528 daddr->sin6_scope_id = pki->ipi6_ifindex;
529 return 1;
530}
531
532/*
533 * Copy the UDP datagram's destination address to the rqstp structure.
534 * The 'destination' address in this case is the address to which the
535 * peer sent the datagram, i.e. our local address. For multihomed
536 * hosts, this can change from msg to msg. Note that only the IP
537 * address changes, the port number should remain the same.
538 */
539static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
540 struct cmsghdr *cmh)
541{
542 switch (cmh->cmsg_level) {
543 case SOL_IP:
544 return svc_udp_get_dest_address4(rqstp, cmh);
545 case SOL_IPV6:
546 return svc_udp_get_dest_address6(rqstp, cmh);
547 }
548
549 return 0;
550}
551
552/**
553 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
554 * @rqstp: request structure into which to receive an RPC Call
555 *
556 * Called in a loop when XPT_DATA has been set.
557 *
558 * Returns:
559 * On success, the number of bytes in a received RPC Call, or
560 * %0 if a complete RPC Call message was not ready to return
561 */
562static int svc_udp_recvfrom(struct svc_rqst *rqstp)
563{
564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
567 struct sk_buff *skb;
568 union {
569 struct cmsghdr hdr;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
571 } buffer;
572 struct cmsghdr *cmh = &buffer.hdr;
573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
575 .msg_control = cmh,
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
578 };
579 size_t len;
580 int err;
581
582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
590 */
591 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
592
593 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
594 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
595 0, 0, MSG_PEEK | MSG_DONTWAIT);
596 if (err < 0)
597 goto out_recv_err;
598 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err);
599 if (!skb)
600 goto out_recv_err;
601
602 len = svc_addr_len(svc_addr(rqstp));
603 rqstp->rq_addrlen = len;
604 if (skb->tstamp == 0) {
605 skb->tstamp = ktime_get_real();
606 /* Don't enable netstamp, sunrpc doesn't
607 need that much accuracy */
608 }
609 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
610 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
611
612 len = skb->len;
613 rqstp->rq_arg.len = len;
614 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
615
616 rqstp->rq_prot = IPPROTO_UDP;
617
618 if (!svc_udp_get_dest_address(rqstp, cmh))
619 goto out_cmsg_err;
620 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
621
622 if (skb_is_nonlinear(skb)) {
623 /* we have to copy */
624 local_bh_disable();
625 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
626 goto out_bh_enable;
627 local_bh_enable();
628 consume_skb(skb);
629 } else {
630 /* we can use it in-place */
631 rqstp->rq_arg.head[0].iov_base = skb->data;
632 rqstp->rq_arg.head[0].iov_len = len;
633 if (skb_checksum_complete(skb))
634 goto out_free;
635 rqstp->rq_xprt_ctxt = skb;
636 }
637
638 rqstp->rq_arg.page_base = 0;
639 if (len <= rqstp->rq_arg.head[0].iov_len) {
640 rqstp->rq_arg.head[0].iov_len = len;
641 rqstp->rq_arg.page_len = 0;
642 rqstp->rq_respages = rqstp->rq_pages+1;
643 } else {
644 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
645 rqstp->rq_respages = rqstp->rq_pages + 1 +
646 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
647 }
648 rqstp->rq_next_page = rqstp->rq_respages+1;
649
650 if (serv->sv_stats)
651 serv->sv_stats->netudpcnt++;
652
653 svc_sock_secure_port(rqstp);
654 svc_xprt_received(rqstp->rq_xprt);
655 return len;
656
657out_recv_err:
658 if (err != -EAGAIN) {
659 /* possibly an icmp error */
660 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
661 }
662 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
663 goto out_clear_busy;
664out_cmsg_err:
665 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
666 cmh->cmsg_level, cmh->cmsg_type);
667 goto out_free;
668out_bh_enable:
669 local_bh_enable();
670out_free:
671 kfree_skb(skb);
672out_clear_busy:
673 svc_xprt_received(rqstp->rq_xprt);
674 return 0;
675}
676
677/**
678 * svc_udp_sendto - Send out a reply on a UDP socket
679 * @rqstp: completed svc_rqst
680 *
681 * xpt_mutex ensures @rqstp's whole message is written to the socket
682 * without interruption.
683 *
684 * Returns the number of bytes sent, or a negative errno.
685 */
686static int svc_udp_sendto(struct svc_rqst *rqstp)
687{
688 struct svc_xprt *xprt = rqstp->rq_xprt;
689 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
690 struct xdr_buf *xdr = &rqstp->rq_res;
691 union {
692 struct cmsghdr hdr;
693 long all[SVC_PKTINFO_SPACE / sizeof(long)];
694 } buffer;
695 struct cmsghdr *cmh = &buffer.hdr;
696 struct msghdr msg = {
697 .msg_name = &rqstp->rq_addr,
698 .msg_namelen = rqstp->rq_addrlen,
699 .msg_control = cmh,
700 .msg_flags = MSG_SPLICE_PAGES,
701 .msg_controllen = sizeof(buffer),
702 };
703 unsigned int count;
704 int err;
705
706 svc_udp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
707 rqstp->rq_xprt_ctxt = NULL;
708
709 svc_set_cmsg_data(rqstp, cmh);
710
711 mutex_lock(&xprt->xpt_mutex);
712
713 if (svc_xprt_is_dead(xprt))
714 goto out_notconn;
715
716 count = xdr_buf_to_bvec(rqstp->rq_bvec,
717 ARRAY_SIZE(rqstp->rq_bvec), xdr);
718
719 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
720 count, rqstp->rq_res.len);
721 err = sock_sendmsg(svsk->sk_sock, &msg);
722 if (err == -ECONNREFUSED) {
723 /* ICMP error on earlier request. */
724 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
725 count, rqstp->rq_res.len);
726 err = sock_sendmsg(svsk->sk_sock, &msg);
727 }
728
729 trace_svcsock_udp_send(xprt, err);
730
731 mutex_unlock(&xprt->xpt_mutex);
732 return err;
733
734out_notconn:
735 mutex_unlock(&xprt->xpt_mutex);
736 return -ENOTCONN;
737}
738
739static int svc_udp_has_wspace(struct svc_xprt *xprt)
740{
741 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
742 struct svc_serv *serv = xprt->xpt_server;
743 unsigned long required;
744
745 /*
746 * Set the SOCK_NOSPACE flag before checking the available
747 * sock space.
748 */
749 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
750 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
751 if (required*2 > sock_wspace(svsk->sk_sk))
752 return 0;
753 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
754 return 1;
755}
756
757static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
758{
759 BUG();
760 return NULL;
761}
762
763static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
764{
765}
766
767static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
768 struct net *net,
769 struct sockaddr *sa, int salen,
770 int flags)
771{
772 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
773}
774
775static const struct svc_xprt_ops svc_udp_ops = {
776 .xpo_create = svc_udp_create,
777 .xpo_recvfrom = svc_udp_recvfrom,
778 .xpo_sendto = svc_udp_sendto,
779 .xpo_result_payload = svc_sock_result_payload,
780 .xpo_release_ctxt = svc_udp_release_ctxt,
781 .xpo_detach = svc_sock_detach,
782 .xpo_free = svc_sock_free,
783 .xpo_has_wspace = svc_udp_has_wspace,
784 .xpo_accept = svc_udp_accept,
785 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
786};
787
788static struct svc_xprt_class svc_udp_class = {
789 .xcl_name = "udp",
790 .xcl_owner = THIS_MODULE,
791 .xcl_ops = &svc_udp_ops,
792 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
793 .xcl_ident = XPRT_TRANSPORT_UDP,
794};
795
796static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
797{
798 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
799 &svsk->sk_xprt, serv);
800 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
801 svsk->sk_sk->sk_data_ready = svc_data_ready;
802 svsk->sk_sk->sk_write_space = svc_write_space;
803
804 /* initialise setting must have enough space to
805 * receive and respond to one request.
806 * svc_udp_recvfrom will re-adjust if necessary
807 */
808 svc_sock_setbufsize(svsk, 3);
809
810 /* data might have come in before data_ready set up */
811 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
812 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
813
814 /* make sure we get destination address info */
815 switch (svsk->sk_sk->sk_family) {
816 case AF_INET:
817 ip_sock_set_pktinfo(svsk->sk_sock->sk);
818 break;
819 case AF_INET6:
820 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
821 break;
822 default:
823 BUG();
824 }
825}
826
827/*
828 * A data_ready event on a listening socket means there's a connection
829 * pending. Do not use state_change as a substitute for it.
830 */
831static void svc_tcp_listen_data_ready(struct sock *sk)
832{
833 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
834
835 trace_sk_data_ready(sk);
836
837 /*
838 * This callback may called twice when a new connection
839 * is established as a child socket inherits everything
840 * from a parent LISTEN socket.
841 * 1) data_ready method of the parent socket will be called
842 * when one of child sockets become ESTABLISHED.
843 * 2) data_ready method of the child socket may be called
844 * when it receives data before the socket is accepted.
845 * In case of 2, we should ignore it silently and DO NOT
846 * dereference svsk.
847 */
848 if (sk->sk_state != TCP_LISTEN)
849 return;
850
851 if (svsk) {
852 /* Refer to svc_setup_socket() for details. */
853 rmb();
854 svsk->sk_odata(sk);
855 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
856 svc_xprt_enqueue(&svsk->sk_xprt);
857 }
858}
859
860/*
861 * A state change on a connected socket means it's dying or dead.
862 */
863static void svc_tcp_state_change(struct sock *sk)
864{
865 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
866
867 if (svsk) {
868 /* Refer to svc_setup_socket() for details. */
869 rmb();
870 svsk->sk_ostate(sk);
871 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
872 if (sk->sk_state != TCP_ESTABLISHED)
873 svc_xprt_deferred_close(&svsk->sk_xprt);
874 }
875}
876
877/*
878 * Accept a TCP connection
879 */
880static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
881{
882 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
883 struct sockaddr_storage addr;
884 struct sockaddr *sin = (struct sockaddr *) &addr;
885 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
886 struct socket *sock = svsk->sk_sock;
887 struct socket *newsock;
888 struct svc_sock *newsvsk;
889 int err, slen;
890
891 if (!sock)
892 return NULL;
893
894 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
895 err = kernel_accept(sock, &newsock, O_NONBLOCK);
896 if (err < 0) {
897 if (err != -EAGAIN)
898 trace_svcsock_accept_err(xprt, serv->sv_name, err);
899 return NULL;
900 }
901 if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL)))
902 return NULL;
903
904 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
905
906 err = kernel_getpeername(newsock, sin);
907 if (err < 0) {
908 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
909 goto failed; /* aborted connection or whatever */
910 }
911 slen = err;
912
913 /* Reset the inherited callbacks before calling svc_setup_socket */
914 newsock->sk->sk_state_change = svsk->sk_ostate;
915 newsock->sk->sk_data_ready = svsk->sk_odata;
916 newsock->sk->sk_write_space = svsk->sk_owspace;
917
918 /* make sure that a write doesn't block forever when
919 * low on memory
920 */
921 newsock->sk->sk_sndtimeo = HZ*30;
922
923 newsvsk = svc_setup_socket(serv, newsock,
924 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
925 if (IS_ERR(newsvsk))
926 goto failed;
927 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
928 err = kernel_getsockname(newsock, sin);
929 slen = err;
930 if (unlikely(err < 0))
931 slen = offsetof(struct sockaddr, sa_data);
932 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
933
934 if (sock_is_loopback(newsock->sk))
935 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
936 else
937 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
938 if (serv->sv_stats)
939 serv->sv_stats->nettcpconn++;
940
941 return &newsvsk->sk_xprt;
942
943failed:
944 sockfd_put(newsock);
945 return NULL;
946}
947
948static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
949 struct svc_rqst *rqstp)
950{
951 size_t len = svsk->sk_datalen;
952 unsigned int i, npages;
953
954 if (!len)
955 return 0;
956 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
957 for (i = 0; i < npages; i++) {
958 if (rqstp->rq_pages[i] != NULL)
959 put_page(rqstp->rq_pages[i]);
960 BUG_ON(svsk->sk_pages[i] == NULL);
961 rqstp->rq_pages[i] = svsk->sk_pages[i];
962 svsk->sk_pages[i] = NULL;
963 }
964 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
965 return len;
966}
967
968static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
969{
970 unsigned int i, len, npages;
971
972 if (svsk->sk_datalen == 0)
973 return;
974 len = svsk->sk_datalen;
975 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 for (i = 0; i < npages; i++) {
977 svsk->sk_pages[i] = rqstp->rq_pages[i];
978 rqstp->rq_pages[i] = NULL;
979 }
980}
981
982static void svc_tcp_clear_pages(struct svc_sock *svsk)
983{
984 unsigned int i, len, npages;
985
986 if (svsk->sk_datalen == 0)
987 goto out;
988 len = svsk->sk_datalen;
989 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
990 for (i = 0; i < npages; i++) {
991 if (svsk->sk_pages[i] == NULL) {
992 WARN_ON_ONCE(1);
993 continue;
994 }
995 put_page(svsk->sk_pages[i]);
996 svsk->sk_pages[i] = NULL;
997 }
998out:
999 svsk->sk_tcplen = 0;
1000 svsk->sk_datalen = 0;
1001}
1002
1003/*
1004 * Receive fragment record header into sk_marker.
1005 */
1006static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
1007 struct svc_rqst *rqstp)
1008{
1009 ssize_t want, len;
1010
1011 /* If we haven't gotten the record length yet,
1012 * get the next four bytes.
1013 */
1014 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1015 struct msghdr msg = { NULL };
1016 struct kvec iov;
1017
1018 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1019 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
1020 iov.iov_len = want;
1021 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
1022 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
1023 if (len < 0)
1024 return len;
1025 svsk->sk_tcplen += len;
1026 if (len < want) {
1027 /* call again to read the remaining bytes */
1028 goto err_short;
1029 }
1030 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
1031 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1032 svsk->sk_xprt.xpt_server->sv_max_mesg)
1033 goto err_too_large;
1034 }
1035 return svc_sock_reclen(svsk);
1036
1037err_too_large:
1038 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
1039 __func__, svsk->sk_xprt.xpt_server->sv_name,
1040 svc_sock_reclen(svsk));
1041 svc_xprt_deferred_close(&svsk->sk_xprt);
1042err_short:
1043 return -EAGAIN;
1044}
1045
1046static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1047{
1048 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1049 struct rpc_rqst *req = NULL;
1050 struct kvec *src, *dst;
1051 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1052 __be32 xid = *p;
1053
1054 if (!bc_xprt)
1055 return -EAGAIN;
1056 spin_lock(&bc_xprt->queue_lock);
1057 req = xprt_lookup_rqst(bc_xprt, xid);
1058 if (!req)
1059 goto unlock_eagain;
1060
1061 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1062 /*
1063 * XXX!: cheating for now! Only copying HEAD.
1064 * But we know this is good enough for now (in fact, for any
1065 * callback reply in the forseeable future).
1066 */
1067 dst = &req->rq_private_buf.head[0];
1068 src = &rqstp->rq_arg.head[0];
1069 if (dst->iov_len < src->iov_len)
1070 goto unlock_eagain; /* whatever; just giving up. */
1071 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1072 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1073 rqstp->rq_arg.len = 0;
1074 spin_unlock(&bc_xprt->queue_lock);
1075 return 0;
1076unlock_eagain:
1077 spin_unlock(&bc_xprt->queue_lock);
1078 return -EAGAIN;
1079}
1080
1081static void svc_tcp_fragment_received(struct svc_sock *svsk)
1082{
1083 /* If we have more data, signal svc_xprt_enqueue() to try again */
1084 svsk->sk_tcplen = 0;
1085 svsk->sk_marker = xdr_zero;
1086}
1087
1088/**
1089 * svc_tcp_recvfrom - Receive data from a TCP socket
1090 * @rqstp: request structure into which to receive an RPC Call
1091 *
1092 * Called in a loop when XPT_DATA has been set.
1093 *
1094 * Read the 4-byte stream record marker, then use the record length
1095 * in that marker to set up exactly the resources needed to receive
1096 * the next RPC message into @rqstp.
1097 *
1098 * Returns:
1099 * On success, the number of bytes in a received RPC Call, or
1100 * %0 if a complete RPC Call message was not ready to return
1101 *
1102 * The zero return case handles partial receives and callback Replies.
1103 * The state of a partial receive is preserved in the svc_sock for
1104 * the next call to svc_tcp_recvfrom.
1105 */
1106static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1107{
1108 struct svc_sock *svsk =
1109 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1110 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1111 size_t want, base;
1112 ssize_t len;
1113 __be32 *p;
1114 __be32 calldir;
1115
1116 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1117 len = svc_tcp_read_marker(svsk, rqstp);
1118 if (len < 0)
1119 goto error;
1120
1121 base = svc_tcp_restore_pages(svsk, rqstp);
1122 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1123 len = svc_tcp_read_msg(rqstp, base + want, base);
1124 if (len >= 0) {
1125 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1126 svsk->sk_tcplen += len;
1127 svsk->sk_datalen += len;
1128 }
1129 if (len != want || !svc_sock_final_rec(svsk))
1130 goto err_incomplete;
1131 if (svsk->sk_datalen < 8)
1132 goto err_nuts;
1133
1134 rqstp->rq_arg.len = svsk->sk_datalen;
1135 rqstp->rq_arg.page_base = 0;
1136 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1137 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1138 rqstp->rq_arg.page_len = 0;
1139 } else
1140 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1141
1142 rqstp->rq_xprt_ctxt = NULL;
1143 rqstp->rq_prot = IPPROTO_TCP;
1144 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1145 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1146 else
1147 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1148
1149 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1150 calldir = p[1];
1151 if (calldir)
1152 len = receive_cb_reply(svsk, rqstp);
1153
1154 /* Reset TCP read info */
1155 svsk->sk_datalen = 0;
1156 svc_tcp_fragment_received(svsk);
1157
1158 if (len < 0)
1159 goto error;
1160
1161 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1162 if (serv->sv_stats)
1163 serv->sv_stats->nettcpcnt++;
1164
1165 svc_sock_secure_port(rqstp);
1166 svc_xprt_received(rqstp->rq_xprt);
1167 return rqstp->rq_arg.len;
1168
1169err_incomplete:
1170 svc_tcp_save_pages(svsk, rqstp);
1171 if (len < 0 && len != -EAGAIN)
1172 goto err_delete;
1173 if (len == want)
1174 svc_tcp_fragment_received(svsk);
1175 else
1176 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1177 svc_sock_reclen(svsk),
1178 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1179 goto err_noclose;
1180error:
1181 if (len != -EAGAIN)
1182 goto err_delete;
1183 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1184 goto err_noclose;
1185err_nuts:
1186 svsk->sk_datalen = 0;
1187err_delete:
1188 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1189 svc_xprt_deferred_close(&svsk->sk_xprt);
1190err_noclose:
1191 svc_xprt_received(rqstp->rq_xprt);
1192 return 0; /* record not complete */
1193}
1194
1195/*
1196 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1197 * copy operations in this path. Therefore the caller must ensure
1198 * that the pages backing @xdr are unchanging.
1199 */
1200static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1201 rpc_fraghdr marker, unsigned int *sentp)
1202{
1203 struct msghdr msg = {
1204 .msg_flags = MSG_SPLICE_PAGES,
1205 };
1206 unsigned int count;
1207 void *buf;
1208 int ret;
1209
1210 *sentp = 0;
1211
1212 /* The stream record marker is copied into a temporary page
1213 * fragment buffer so that it can be included in rq_bvec.
1214 */
1215 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1216 GFP_KERNEL);
1217 if (!buf)
1218 return -ENOMEM;
1219 memcpy(buf, &marker, sizeof(marker));
1220 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1221
1222 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1223 ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1224
1225 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1226 1 + count, sizeof(marker) + rqstp->rq_res.len);
1227 ret = sock_sendmsg(svsk->sk_sock, &msg);
1228 page_frag_free(buf);
1229 if (ret < 0)
1230 return ret;
1231 *sentp += ret;
1232 return 0;
1233}
1234
1235/**
1236 * svc_tcp_sendto - Send out a reply on a TCP socket
1237 * @rqstp: completed svc_rqst
1238 *
1239 * xpt_mutex ensures @rqstp's whole message is written to the socket
1240 * without interruption.
1241 *
1242 * Returns the number of bytes sent, or a negative errno.
1243 */
1244static int svc_tcp_sendto(struct svc_rqst *rqstp)
1245{
1246 struct svc_xprt *xprt = rqstp->rq_xprt;
1247 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1248 struct xdr_buf *xdr = &rqstp->rq_res;
1249 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1250 (u32)xdr->len);
1251 unsigned int sent;
1252 int err;
1253
1254 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1255 rqstp->rq_xprt_ctxt = NULL;
1256
1257 mutex_lock(&xprt->xpt_mutex);
1258 if (svc_xprt_is_dead(xprt))
1259 goto out_notconn;
1260 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1261 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1262 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1263 goto out_close;
1264 mutex_unlock(&xprt->xpt_mutex);
1265 return sent;
1266
1267out_notconn:
1268 mutex_unlock(&xprt->xpt_mutex);
1269 return -ENOTCONN;
1270out_close:
1271 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1272 xprt->xpt_server->sv_name,
1273 (err < 0) ? "got error" : "sent",
1274 (err < 0) ? err : sent, xdr->len);
1275 svc_xprt_deferred_close(xprt);
1276 mutex_unlock(&xprt->xpt_mutex);
1277 return -EAGAIN;
1278}
1279
1280static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1281 struct net *net,
1282 struct sockaddr *sa, int salen,
1283 int flags)
1284{
1285 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1286}
1287
1288static const struct svc_xprt_ops svc_tcp_ops = {
1289 .xpo_create = svc_tcp_create,
1290 .xpo_recvfrom = svc_tcp_recvfrom,
1291 .xpo_sendto = svc_tcp_sendto,
1292 .xpo_result_payload = svc_sock_result_payload,
1293 .xpo_release_ctxt = svc_tcp_release_ctxt,
1294 .xpo_detach = svc_tcp_sock_detach,
1295 .xpo_free = svc_sock_free,
1296 .xpo_has_wspace = svc_tcp_has_wspace,
1297 .xpo_accept = svc_tcp_accept,
1298 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1299 .xpo_handshake = svc_tcp_handshake,
1300};
1301
1302static struct svc_xprt_class svc_tcp_class = {
1303 .xcl_name = "tcp",
1304 .xcl_owner = THIS_MODULE,
1305 .xcl_ops = &svc_tcp_ops,
1306 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1307 .xcl_ident = XPRT_TRANSPORT_TCP,
1308};
1309
1310void svc_init_xprt_sock(void)
1311{
1312 svc_reg_xprt_class(&svc_tcp_class);
1313 svc_reg_xprt_class(&svc_udp_class);
1314}
1315
1316void svc_cleanup_xprt_sock(void)
1317{
1318 svc_unreg_xprt_class(&svc_tcp_class);
1319 svc_unreg_xprt_class(&svc_udp_class);
1320}
1321
1322static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1323{
1324 struct sock *sk = svsk->sk_sk;
1325
1326 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1327 &svsk->sk_xprt, serv);
1328 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1329 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1330 if (sk->sk_state == TCP_LISTEN) {
1331 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1332 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1333 sk->sk_data_ready = svc_tcp_listen_data_ready;
1334 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1335 } else {
1336 sk->sk_state_change = svc_tcp_state_change;
1337 sk->sk_data_ready = svc_data_ready;
1338 sk->sk_write_space = svc_write_space;
1339
1340 svsk->sk_marker = xdr_zero;
1341 svsk->sk_tcplen = 0;
1342 svsk->sk_datalen = 0;
1343 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1344
1345 tcp_sock_set_nodelay(sk);
1346
1347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1348 switch (sk->sk_state) {
1349 case TCP_SYN_RECV:
1350 case TCP_ESTABLISHED:
1351 break;
1352 default:
1353 svc_xprt_deferred_close(&svsk->sk_xprt);
1354 }
1355 }
1356}
1357
1358void svc_sock_update_bufs(struct svc_serv *serv)
1359{
1360 /*
1361 * The number of server threads has changed. Update
1362 * rcvbuf and sndbuf accordingly on all sockets
1363 */
1364 struct svc_sock *svsk;
1365
1366 spin_lock_bh(&serv->sv_lock);
1367 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1368 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1369 spin_unlock_bh(&serv->sv_lock);
1370}
1371
1372/*
1373 * Initialize socket for RPC use and create svc_sock struct
1374 */
1375static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1376 struct socket *sock,
1377 int flags)
1378{
1379 struct svc_sock *svsk;
1380 struct sock *inet;
1381 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1382
1383 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1384 if (!svsk)
1385 return ERR_PTR(-ENOMEM);
1386
1387 inet = sock->sk;
1388
1389 if (pmap_register) {
1390 int err;
1391
1392 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1393 inet->sk_protocol,
1394 ntohs(inet_sk(inet)->inet_sport));
1395 if (err < 0) {
1396 kfree(svsk);
1397 return ERR_PTR(err);
1398 }
1399 }
1400
1401 svsk->sk_sock = sock;
1402 svsk->sk_sk = inet;
1403 svsk->sk_ostate = inet->sk_state_change;
1404 svsk->sk_odata = inet->sk_data_ready;
1405 svsk->sk_owspace = inet->sk_write_space;
1406 /*
1407 * This barrier is necessary in order to prevent race condition
1408 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1409 * when calling callbacks above.
1410 */
1411 wmb();
1412 inet->sk_user_data = svsk;
1413
1414 /* Initialize the socket */
1415 if (sock->type == SOCK_DGRAM)
1416 svc_udp_init(svsk, serv);
1417 else
1418 svc_tcp_init(svsk, serv);
1419
1420 trace_svcsock_new(svsk, sock);
1421 return svsk;
1422}
1423
1424/**
1425 * svc_addsock - add a listener socket to an RPC service
1426 * @serv: pointer to RPC service to which to add a new listener
1427 * @net: caller's network namespace
1428 * @fd: file descriptor of the new listener
1429 * @name_return: pointer to buffer to fill in with name of listener
1430 * @len: size of the buffer
1431 * @cred: credential
1432 *
1433 * Fills in socket name and returns positive length of name if successful.
1434 * Name is terminated with '\n'. On error, returns a negative errno
1435 * value.
1436 */
1437int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1438 char *name_return, const size_t len, const struct cred *cred)
1439{
1440 int err = 0;
1441 struct socket *so = sockfd_lookup(fd, &err);
1442 struct svc_sock *svsk = NULL;
1443 struct sockaddr_storage addr;
1444 struct sockaddr *sin = (struct sockaddr *)&addr;
1445 int salen;
1446
1447 if (!so)
1448 return err;
1449 err = -EINVAL;
1450 if (sock_net(so->sk) != net)
1451 goto out;
1452 err = -EAFNOSUPPORT;
1453 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1454 goto out;
1455 err = -EPROTONOSUPPORT;
1456 if (so->sk->sk_protocol != IPPROTO_TCP &&
1457 so->sk->sk_protocol != IPPROTO_UDP)
1458 goto out;
1459 err = -EISCONN;
1460 if (so->state > SS_UNCONNECTED)
1461 goto out;
1462 err = -ENOENT;
1463 if (!try_module_get(THIS_MODULE))
1464 goto out;
1465 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1466 if (IS_ERR(svsk)) {
1467 module_put(THIS_MODULE);
1468 err = PTR_ERR(svsk);
1469 goto out;
1470 }
1471 salen = kernel_getsockname(svsk->sk_sock, sin);
1472 if (salen >= 0)
1473 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1474 svsk->sk_xprt.xpt_cred = get_cred(cred);
1475 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1476 return svc_one_sock_name(svsk, name_return, len);
1477out:
1478 sockfd_put(so);
1479 return err;
1480}
1481EXPORT_SYMBOL_GPL(svc_addsock);
1482
1483/*
1484 * Create socket for RPC service.
1485 */
1486static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1487 int protocol,
1488 struct net *net,
1489 struct sockaddr *sin, int len,
1490 int flags)
1491{
1492 struct svc_sock *svsk;
1493 struct socket *sock;
1494 int error;
1495 int type;
1496 struct sockaddr_storage addr;
1497 struct sockaddr *newsin = (struct sockaddr *)&addr;
1498 int newlen;
1499 int family;
1500
1501 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1502 printk(KERN_WARNING "svc: only UDP and TCP "
1503 "sockets supported\n");
1504 return ERR_PTR(-EINVAL);
1505 }
1506
1507 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1508 switch (sin->sa_family) {
1509 case AF_INET6:
1510 family = PF_INET6;
1511 break;
1512 case AF_INET:
1513 family = PF_INET;
1514 break;
1515 default:
1516 return ERR_PTR(-EINVAL);
1517 }
1518
1519 error = __sock_create(net, family, type, protocol, &sock, 1);
1520 if (error < 0)
1521 return ERR_PTR(error);
1522
1523 svc_reclassify_socket(sock);
1524
1525 /*
1526 * If this is an PF_INET6 listener, we want to avoid
1527 * getting requests from IPv4 remotes. Those should
1528 * be shunted to a PF_INET listener via rpcbind.
1529 */
1530 if (family == PF_INET6)
1531 ip6_sock_set_v6only(sock->sk);
1532 if (type == SOCK_STREAM)
1533 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1534 error = kernel_bind(sock, sin, len);
1535 if (error < 0)
1536 goto bummer;
1537
1538 error = kernel_getsockname(sock, newsin);
1539 if (error < 0)
1540 goto bummer;
1541 newlen = error;
1542
1543 if (protocol == IPPROTO_TCP) {
1544 sk_net_refcnt_upgrade(sock->sk);
1545 if ((error = kernel_listen(sock, 64)) < 0)
1546 goto bummer;
1547 }
1548
1549 svsk = svc_setup_socket(serv, sock, flags);
1550 if (IS_ERR(svsk)) {
1551 error = PTR_ERR(svsk);
1552 goto bummer;
1553 }
1554 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1555 return (struct svc_xprt *)svsk;
1556bummer:
1557 sock_release(sock);
1558 return ERR_PTR(error);
1559}
1560
1561/*
1562 * Detach the svc_sock from the socket so that no
1563 * more callbacks occur.
1564 */
1565static void svc_sock_detach(struct svc_xprt *xprt)
1566{
1567 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1568 struct sock *sk = svsk->sk_sk;
1569
1570 /* put back the old socket callbacks */
1571 lock_sock(sk);
1572 sk->sk_state_change = svsk->sk_ostate;
1573 sk->sk_data_ready = svsk->sk_odata;
1574 sk->sk_write_space = svsk->sk_owspace;
1575 sk->sk_user_data = NULL;
1576 release_sock(sk);
1577}
1578
1579/*
1580 * Disconnect the socket, and reset the callbacks
1581 */
1582static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1583{
1584 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1585
1586 tls_handshake_close(svsk->sk_sock);
1587
1588 svc_sock_detach(xprt);
1589
1590 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1591 svc_tcp_clear_pages(svsk);
1592 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1593 }
1594}
1595
1596/*
1597 * Free the svc_sock's socket resources and the svc_sock itself.
1598 */
1599static void svc_sock_free(struct svc_xprt *xprt)
1600{
1601 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1602 struct socket *sock = svsk->sk_sock;
1603
1604 trace_svcsock_free(svsk, sock);
1605
1606 tls_handshake_cancel(sock->sk);
1607 if (sock->file)
1608 sockfd_put(sock);
1609 else
1610 sock_release(sock);
1611
1612 page_frag_cache_drain(&svsk->sk_frag_cache);
1613 kfree(svsk);
1614}