Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svcsock.c
4 *
5 * These are the RPC server socket internals.
6 *
7 * The server scheduling algorithm does not always distribute the load
8 * evenly when servicing a single client. May need to modify the
9 * svc_xprt_enqueue procedure...
10 *
11 * TCP support is largely untested and may be a little slow. The problem
12 * is that we currently do two separate recvfrom's, one for the 4-byte
13 * record length, and the second for the actual record. This could possibly
14 * be improved by always reading a minimum size of around 100 bytes and
15 * tucking any superfluous bytes away in a temporary store. Still, that
16 * leaves write requests out in the rain. An alternative may be to peek at
17 * the first skb in the queue, and if it matches the next TCP sequence
18 * number, to extract the record marker. Yuck.
19 *
20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
21 */
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/module.h>
26#include <linux/errno.h>
27#include <linux/fcntl.h>
28#include <linux/net.h>
29#include <linux/in.h>
30#include <linux/inet.h>
31#include <linux/udp.h>
32#include <linux/tcp.h>
33#include <linux/unistd.h>
34#include <linux/slab.h>
35#include <linux/netdevice.h>
36#include <linux/skbuff.h>
37#include <linux/file.h>
38#include <linux/freezer.h>
39#include <net/sock.h>
40#include <net/checksum.h>
41#include <net/ip.h>
42#include <net/ipv6.h>
43#include <net/udp.h>
44#include <net/tcp.h>
45#include <net/tcp_states.h>
46#include <linux/uaccess.h>
47#include <linux/highmem.h>
48#include <asm/ioctls.h>
49
50#include <linux/sunrpc/types.h>
51#include <linux/sunrpc/clnt.h>
52#include <linux/sunrpc/xdr.h>
53#include <linux/sunrpc/msg_prot.h>
54#include <linux/sunrpc/svcsock.h>
55#include <linux/sunrpc/stats.h>
56#include <linux/sunrpc/xprt.h>
57
58#include <trace/events/sunrpc.h>
59
60#include "socklib.h"
61#include "sunrpc.h"
62
63#define RPCDBG_FACILITY RPCDBG_SVCXPRT
64
65
66static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
67 int flags);
68static int svc_udp_recvfrom(struct svc_rqst *);
69static int svc_udp_sendto(struct svc_rqst *);
70static void svc_sock_detach(struct svc_xprt *);
71static void svc_tcp_sock_detach(struct svc_xprt *);
72static void svc_sock_free(struct svc_xprt *);
73
74static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77#ifdef CONFIG_DEBUG_LOCK_ALLOC
78static struct lock_class_key svc_key[2];
79static struct lock_class_key svc_slock_key[2];
80
81static void svc_reclassify_socket(struct socket *sock)
82{
83 struct sock *sk = sock->sk;
84
85 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
86 return;
87
88 switch (sk->sk_family) {
89 case AF_INET:
90 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
91 &svc_slock_key[0],
92 "sk_xprt.xpt_lock-AF_INET-NFSD",
93 &svc_key[0]);
94 break;
95
96 case AF_INET6:
97 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
98 &svc_slock_key[1],
99 "sk_xprt.xpt_lock-AF_INET6-NFSD",
100 &svc_key[1]);
101 break;
102
103 default:
104 BUG();
105 }
106}
107#else
108static void svc_reclassify_socket(struct socket *sock)
109{
110}
111#endif
112
113/**
114 * svc_tcp_release_rqst - Release transport-related resources
115 * @rqstp: request structure with resources to be released
116 *
117 */
118static void svc_tcp_release_rqst(struct svc_rqst *rqstp)
119{
120}
121
122/**
123 * svc_udp_release_rqst - Release transport-related resources
124 * @rqstp: request structure with resources to be released
125 *
126 */
127static void svc_udp_release_rqst(struct svc_rqst *rqstp)
128{
129 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
130
131 if (skb) {
132 rqstp->rq_xprt_ctxt = NULL;
133 consume_skb(skb);
134 }
135}
136
137union svc_pktinfo_u {
138 struct in_pktinfo pkti;
139 struct in6_pktinfo pkti6;
140};
141#define SVC_PKTINFO_SPACE \
142 CMSG_SPACE(sizeof(union svc_pktinfo_u))
143
144static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
145{
146 struct svc_sock *svsk =
147 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
148 switch (svsk->sk_sk->sk_family) {
149 case AF_INET: {
150 struct in_pktinfo *pki = CMSG_DATA(cmh);
151
152 cmh->cmsg_level = SOL_IP;
153 cmh->cmsg_type = IP_PKTINFO;
154 pki->ipi_ifindex = 0;
155 pki->ipi_spec_dst.s_addr =
156 svc_daddr_in(rqstp)->sin_addr.s_addr;
157 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
158 }
159 break;
160
161 case AF_INET6: {
162 struct in6_pktinfo *pki = CMSG_DATA(cmh);
163 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
164
165 cmh->cmsg_level = SOL_IPV6;
166 cmh->cmsg_type = IPV6_PKTINFO;
167 pki->ipi6_ifindex = daddr->sin6_scope_id;
168 pki->ipi6_addr = daddr->sin6_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
170 }
171 break;
172 }
173}
174
175static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
176 unsigned int length)
177{
178 return 0;
179}
180
181/*
182 * Report socket names for nfsdfs
183 */
184static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
185{
186 const struct sock *sk = svsk->sk_sk;
187 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
188 "udp" : "tcp";
189 int len;
190
191 switch (sk->sk_family) {
192 case PF_INET:
193 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
194 proto_name,
195 &inet_sk(sk)->inet_rcv_saddr,
196 inet_sk(sk)->inet_num);
197 break;
198#if IS_ENABLED(CONFIG_IPV6)
199 case PF_INET6:
200 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
201 proto_name,
202 &sk->sk_v6_rcv_saddr,
203 inet_sk(sk)->inet_num);
204 break;
205#endif
206 default:
207 len = snprintf(buf, remaining, "*unknown-%d*\n",
208 sk->sk_family);
209 }
210
211 if (len >= remaining) {
212 *buf = '\0';
213 return -ENAMETOOLONG;
214 }
215 return len;
216}
217
218#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
219static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
220{
221 struct bvec_iter bi = {
222 .bi_size = size + seek,
223 };
224 struct bio_vec bv;
225
226 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
227 for_each_bvec(bv, bvec, bi, bi)
228 flush_dcache_page(bv.bv_page);
229}
230#else
231static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
232 size_t seek)
233{
234}
235#endif
236
237/*
238 * Read from @rqstp's transport socket. The incoming message fills whole
239 * pages in @rqstp's rq_pages array until the last page of the message
240 * has been received into a partial page.
241 */
242static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
243 size_t seek)
244{
245 struct svc_sock *svsk =
246 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
247 struct bio_vec *bvec = rqstp->rq_bvec;
248 struct msghdr msg = { NULL };
249 unsigned int i;
250 ssize_t len;
251 size_t t;
252
253 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
254
255 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE) {
256 bvec[i].bv_page = rqstp->rq_pages[i];
257 bvec[i].bv_len = PAGE_SIZE;
258 bvec[i].bv_offset = 0;
259 }
260 rqstp->rq_respages = &rqstp->rq_pages[i];
261 rqstp->rq_next_page = rqstp->rq_respages + 1;
262
263 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen);
264 if (seek) {
265 iov_iter_advance(&msg.msg_iter, seek);
266 buflen -= seek;
267 }
268 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
269 if (len > 0)
270 svc_flush_bvec(bvec, len, seek);
271
272 /* If we read a full record, then assume there may be more
273 * data to read (stream based sockets only!)
274 */
275 if (len == buflen)
276 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
277
278 return len;
279}
280
281/*
282 * Set socket snd and rcv buffer lengths
283 */
284static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
285{
286 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
287 struct socket *sock = svsk->sk_sock;
288
289 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
290
291 lock_sock(sock->sk);
292 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
293 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
294 sock->sk->sk_write_space(sock->sk);
295 release_sock(sock->sk);
296}
297
298static void svc_sock_secure_port(struct svc_rqst *rqstp)
299{
300 if (svc_port_is_privileged(svc_addr(rqstp)))
301 set_bit(RQ_SECURE, &rqstp->rq_flags);
302 else
303 clear_bit(RQ_SECURE, &rqstp->rq_flags);
304}
305
306/*
307 * INET callback when data has been received on the socket.
308 */
309static void svc_data_ready(struct sock *sk)
310{
311 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
312
313 if (svsk) {
314 /* Refer to svc_setup_socket() for details. */
315 rmb();
316 svsk->sk_odata(sk);
317 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
318 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
319 svc_xprt_enqueue(&svsk->sk_xprt);
320 }
321}
322
323/*
324 * INET callback when space is newly available on the socket.
325 */
326static void svc_write_space(struct sock *sk)
327{
328 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
329
330 if (svsk) {
331 /* Refer to svc_setup_socket() for details. */
332 rmb();
333 trace_svcsock_write_space(&svsk->sk_xprt, 0);
334 svsk->sk_owspace(sk);
335 svc_xprt_enqueue(&svsk->sk_xprt);
336 }
337}
338
339static int svc_tcp_has_wspace(struct svc_xprt *xprt)
340{
341 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
342
343 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
344 return 1;
345 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
346}
347
348static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
349{
350 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
351
352 sock_no_linger(svsk->sk_sock->sk);
353}
354
355/*
356 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
357 */
358static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
359 struct cmsghdr *cmh)
360{
361 struct in_pktinfo *pki = CMSG_DATA(cmh);
362 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
363
364 if (cmh->cmsg_type != IP_PKTINFO)
365 return 0;
366
367 daddr->sin_family = AF_INET;
368 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
369 return 1;
370}
371
372/*
373 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
374 */
375static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
376 struct cmsghdr *cmh)
377{
378 struct in6_pktinfo *pki = CMSG_DATA(cmh);
379 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
380
381 if (cmh->cmsg_type != IPV6_PKTINFO)
382 return 0;
383
384 daddr->sin6_family = AF_INET6;
385 daddr->sin6_addr = pki->ipi6_addr;
386 daddr->sin6_scope_id = pki->ipi6_ifindex;
387 return 1;
388}
389
390/*
391 * Copy the UDP datagram's destination address to the rqstp structure.
392 * The 'destination' address in this case is the address to which the
393 * peer sent the datagram, i.e. our local address. For multihomed
394 * hosts, this can change from msg to msg. Note that only the IP
395 * address changes, the port number should remain the same.
396 */
397static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
398 struct cmsghdr *cmh)
399{
400 switch (cmh->cmsg_level) {
401 case SOL_IP:
402 return svc_udp_get_dest_address4(rqstp, cmh);
403 case SOL_IPV6:
404 return svc_udp_get_dest_address6(rqstp, cmh);
405 }
406
407 return 0;
408}
409
410/**
411 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
412 * @rqstp: request structure into which to receive an RPC Call
413 *
414 * Called in a loop when XPT_DATA has been set.
415 *
416 * Returns:
417 * On success, the number of bytes in a received RPC Call, or
418 * %0 if a complete RPC Call message was not ready to return
419 */
420static int svc_udp_recvfrom(struct svc_rqst *rqstp)
421{
422 struct svc_sock *svsk =
423 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
424 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
425 struct sk_buff *skb;
426 union {
427 struct cmsghdr hdr;
428 long all[SVC_PKTINFO_SPACE / sizeof(long)];
429 } buffer;
430 struct cmsghdr *cmh = &buffer.hdr;
431 struct msghdr msg = {
432 .msg_name = svc_addr(rqstp),
433 .msg_control = cmh,
434 .msg_controllen = sizeof(buffer),
435 .msg_flags = MSG_DONTWAIT,
436 };
437 size_t len;
438 int err;
439
440 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
441 /* udp sockets need large rcvbuf as all pending
442 * requests are still in that buffer. sndbuf must
443 * also be large enough that there is enough space
444 * for one reply per thread. We count all threads
445 * rather than threads in a particular pool, which
446 * provides an upper bound on the number of threads
447 * which will access the socket.
448 */
449 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
450
451 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
452 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
453 0, 0, MSG_PEEK | MSG_DONTWAIT);
454 if (err < 0)
455 goto out_recv_err;
456 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err);
457 if (!skb)
458 goto out_recv_err;
459
460 len = svc_addr_len(svc_addr(rqstp));
461 rqstp->rq_addrlen = len;
462 if (skb->tstamp == 0) {
463 skb->tstamp = ktime_get_real();
464 /* Don't enable netstamp, sunrpc doesn't
465 need that much accuracy */
466 }
467 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
468 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
469
470 len = skb->len;
471 rqstp->rq_arg.len = len;
472 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
473
474 rqstp->rq_prot = IPPROTO_UDP;
475
476 if (!svc_udp_get_dest_address(rqstp, cmh))
477 goto out_cmsg_err;
478 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
479
480 if (skb_is_nonlinear(skb)) {
481 /* we have to copy */
482 local_bh_disable();
483 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
484 goto out_bh_enable;
485 local_bh_enable();
486 consume_skb(skb);
487 } else {
488 /* we can use it in-place */
489 rqstp->rq_arg.head[0].iov_base = skb->data;
490 rqstp->rq_arg.head[0].iov_len = len;
491 if (skb_checksum_complete(skb))
492 goto out_free;
493 rqstp->rq_xprt_ctxt = skb;
494 }
495
496 rqstp->rq_arg.page_base = 0;
497 if (len <= rqstp->rq_arg.head[0].iov_len) {
498 rqstp->rq_arg.head[0].iov_len = len;
499 rqstp->rq_arg.page_len = 0;
500 rqstp->rq_respages = rqstp->rq_pages+1;
501 } else {
502 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
503 rqstp->rq_respages = rqstp->rq_pages + 1 +
504 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
505 }
506 rqstp->rq_next_page = rqstp->rq_respages+1;
507
508 if (serv->sv_stats)
509 serv->sv_stats->netudpcnt++;
510
511 svc_xprt_received(rqstp->rq_xprt);
512 return len;
513
514out_recv_err:
515 if (err != -EAGAIN) {
516 /* possibly an icmp error */
517 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
518 }
519 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
520 goto out_clear_busy;
521out_cmsg_err:
522 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
523 cmh->cmsg_level, cmh->cmsg_type);
524 goto out_free;
525out_bh_enable:
526 local_bh_enable();
527out_free:
528 kfree_skb(skb);
529out_clear_busy:
530 svc_xprt_received(rqstp->rq_xprt);
531 return 0;
532}
533
534/**
535 * svc_udp_sendto - Send out a reply on a UDP socket
536 * @rqstp: completed svc_rqst
537 *
538 * xpt_mutex ensures @rqstp's whole message is written to the socket
539 * without interruption.
540 *
541 * Returns the number of bytes sent, or a negative errno.
542 */
543static int svc_udp_sendto(struct svc_rqst *rqstp)
544{
545 struct svc_xprt *xprt = rqstp->rq_xprt;
546 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
547 struct xdr_buf *xdr = &rqstp->rq_res;
548 union {
549 struct cmsghdr hdr;
550 long all[SVC_PKTINFO_SPACE / sizeof(long)];
551 } buffer;
552 struct cmsghdr *cmh = &buffer.hdr;
553 struct msghdr msg = {
554 .msg_name = &rqstp->rq_addr,
555 .msg_namelen = rqstp->rq_addrlen,
556 .msg_control = cmh,
557 .msg_controllen = sizeof(buffer),
558 };
559 unsigned int sent;
560 int err;
561
562 svc_udp_release_rqst(rqstp);
563
564 svc_set_cmsg_data(rqstp, cmh);
565
566 mutex_lock(&xprt->xpt_mutex);
567
568 if (svc_xprt_is_dead(xprt))
569 goto out_notconn;
570
571 err = xdr_alloc_bvec(xdr, GFP_KERNEL);
572 if (err < 0)
573 goto out_unlock;
574
575 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
576 if (err == -ECONNREFUSED) {
577 /* ICMP error on earlier request. */
578 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
579 }
580 xdr_free_bvec(xdr);
581 trace_svcsock_udp_send(xprt, err);
582out_unlock:
583 mutex_unlock(&xprt->xpt_mutex);
584 if (err < 0)
585 return err;
586 return sent;
587
588out_notconn:
589 mutex_unlock(&xprt->xpt_mutex);
590 return -ENOTCONN;
591}
592
593static int svc_udp_has_wspace(struct svc_xprt *xprt)
594{
595 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
596 struct svc_serv *serv = xprt->xpt_server;
597 unsigned long required;
598
599 /*
600 * Set the SOCK_NOSPACE flag before checking the available
601 * sock space.
602 */
603 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
604 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
605 if (required*2 > sock_wspace(svsk->sk_sk))
606 return 0;
607 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
608 return 1;
609}
610
611static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
612{
613 BUG();
614 return NULL;
615}
616
617static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
618{
619}
620
621static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
622 struct net *net,
623 struct sockaddr *sa, int salen,
624 int flags)
625{
626 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
627}
628
629static const struct svc_xprt_ops svc_udp_ops = {
630 .xpo_create = svc_udp_create,
631 .xpo_recvfrom = svc_udp_recvfrom,
632 .xpo_sendto = svc_udp_sendto,
633 .xpo_result_payload = svc_sock_result_payload,
634 .xpo_release_rqst = svc_udp_release_rqst,
635 .xpo_detach = svc_sock_detach,
636 .xpo_free = svc_sock_free,
637 .xpo_has_wspace = svc_udp_has_wspace,
638 .xpo_accept = svc_udp_accept,
639 .xpo_secure_port = svc_sock_secure_port,
640 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
641};
642
643static struct svc_xprt_class svc_udp_class = {
644 .xcl_name = "udp",
645 .xcl_owner = THIS_MODULE,
646 .xcl_ops = &svc_udp_ops,
647 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
648 .xcl_ident = XPRT_TRANSPORT_UDP,
649};
650
651static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
652{
653 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
654 &svsk->sk_xprt, serv);
655 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
656 svsk->sk_sk->sk_data_ready = svc_data_ready;
657 svsk->sk_sk->sk_write_space = svc_write_space;
658
659 /* initialise setting must have enough space to
660 * receive and respond to one request.
661 * svc_udp_recvfrom will re-adjust if necessary
662 */
663 svc_sock_setbufsize(svsk, 3);
664
665 /* data might have come in before data_ready set up */
666 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
667 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
668
669 /* make sure we get destination address info */
670 switch (svsk->sk_sk->sk_family) {
671 case AF_INET:
672 ip_sock_set_pktinfo(svsk->sk_sock->sk);
673 break;
674 case AF_INET6:
675 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
676 break;
677 default:
678 BUG();
679 }
680}
681
682/*
683 * A data_ready event on a listening socket means there's a connection
684 * pending. Do not use state_change as a substitute for it.
685 */
686static void svc_tcp_listen_data_ready(struct sock *sk)
687{
688 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
689
690 if (svsk) {
691 /* Refer to svc_setup_socket() for details. */
692 rmb();
693 svsk->sk_odata(sk);
694 }
695
696 /*
697 * This callback may called twice when a new connection
698 * is established as a child socket inherits everything
699 * from a parent LISTEN socket.
700 * 1) data_ready method of the parent socket will be called
701 * when one of child sockets become ESTABLISHED.
702 * 2) data_ready method of the child socket may be called
703 * when it receives data before the socket is accepted.
704 * In case of 2, we should ignore it silently.
705 */
706 if (sk->sk_state == TCP_LISTEN) {
707 if (svsk) {
708 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
709 svc_xprt_enqueue(&svsk->sk_xprt);
710 }
711 }
712}
713
714/*
715 * A state change on a connected socket means it's dying or dead.
716 */
717static void svc_tcp_state_change(struct sock *sk)
718{
719 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
720
721 if (svsk) {
722 /* Refer to svc_setup_socket() for details. */
723 rmb();
724 svsk->sk_ostate(sk);
725 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
726 if (sk->sk_state != TCP_ESTABLISHED)
727 svc_xprt_deferred_close(&svsk->sk_xprt);
728 }
729}
730
731/*
732 * Accept a TCP connection
733 */
734static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
735{
736 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
737 struct sockaddr_storage addr;
738 struct sockaddr *sin = (struct sockaddr *) &addr;
739 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
740 struct socket *sock = svsk->sk_sock;
741 struct socket *newsock;
742 struct svc_sock *newsvsk;
743 int err, slen;
744
745 if (!sock)
746 return NULL;
747
748 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
749 err = kernel_accept(sock, &newsock, O_NONBLOCK);
750 if (err < 0) {
751 if (err == -ENOMEM)
752 printk(KERN_WARNING "%s: no more sockets!\n",
753 serv->sv_name);
754 else if (err != -EAGAIN)
755 net_warn_ratelimited("%s: accept failed (err %d)!\n",
756 serv->sv_name, -err);
757 trace_svcsock_accept_err(xprt, serv->sv_name, err);
758 return NULL;
759 }
760 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
761
762 err = kernel_getpeername(newsock, sin);
763 if (err < 0) {
764 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
765 goto failed; /* aborted connection or whatever */
766 }
767 slen = err;
768
769 /* Reset the inherited callbacks before calling svc_setup_socket */
770 newsock->sk->sk_state_change = svsk->sk_ostate;
771 newsock->sk->sk_data_ready = svsk->sk_odata;
772 newsock->sk->sk_write_space = svsk->sk_owspace;
773
774 /* make sure that a write doesn't block forever when
775 * low on memory
776 */
777 newsock->sk->sk_sndtimeo = HZ*30;
778
779 newsvsk = svc_setup_socket(serv, newsock,
780 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
781 if (IS_ERR(newsvsk))
782 goto failed;
783 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
784 err = kernel_getsockname(newsock, sin);
785 slen = err;
786 if (unlikely(err < 0))
787 slen = offsetof(struct sockaddr, sa_data);
788 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
789
790 if (sock_is_loopback(newsock->sk))
791 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
792 else
793 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
794 if (serv->sv_stats)
795 serv->sv_stats->nettcpconn++;
796
797 return &newsvsk->sk_xprt;
798
799failed:
800 sock_release(newsock);
801 return NULL;
802}
803
804static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
805 struct svc_rqst *rqstp)
806{
807 size_t len = svsk->sk_datalen;
808 unsigned int i, npages;
809
810 if (!len)
811 return 0;
812 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
813 for (i = 0; i < npages; i++) {
814 if (rqstp->rq_pages[i] != NULL)
815 put_page(rqstp->rq_pages[i]);
816 BUG_ON(svsk->sk_pages[i] == NULL);
817 rqstp->rq_pages[i] = svsk->sk_pages[i];
818 svsk->sk_pages[i] = NULL;
819 }
820 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
821 return len;
822}
823
824static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
825{
826 unsigned int i, len, npages;
827
828 if (svsk->sk_datalen == 0)
829 return;
830 len = svsk->sk_datalen;
831 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
832 for (i = 0; i < npages; i++) {
833 svsk->sk_pages[i] = rqstp->rq_pages[i];
834 rqstp->rq_pages[i] = NULL;
835 }
836}
837
838static void svc_tcp_clear_pages(struct svc_sock *svsk)
839{
840 unsigned int i, len, npages;
841
842 if (svsk->sk_datalen == 0)
843 goto out;
844 len = svsk->sk_datalen;
845 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
846 for (i = 0; i < npages; i++) {
847 if (svsk->sk_pages[i] == NULL) {
848 WARN_ON_ONCE(1);
849 continue;
850 }
851 put_page(svsk->sk_pages[i]);
852 svsk->sk_pages[i] = NULL;
853 }
854out:
855 svsk->sk_tcplen = 0;
856 svsk->sk_datalen = 0;
857}
858
859/*
860 * Receive fragment record header into sk_marker.
861 */
862static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
863 struct svc_rqst *rqstp)
864{
865 ssize_t want, len;
866
867 /* If we haven't gotten the record length yet,
868 * get the next four bytes.
869 */
870 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
871 struct msghdr msg = { NULL };
872 struct kvec iov;
873
874 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
875 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
876 iov.iov_len = want;
877 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
878 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
879 if (len < 0)
880 return len;
881 svsk->sk_tcplen += len;
882 if (len < want) {
883 /* call again to read the remaining bytes */
884 goto err_short;
885 }
886 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
887 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
888 svsk->sk_xprt.xpt_server->sv_max_mesg)
889 goto err_too_large;
890 }
891 return svc_sock_reclen(svsk);
892
893err_too_large:
894 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
895 __func__, svsk->sk_xprt.xpt_server->sv_name,
896 svc_sock_reclen(svsk));
897 svc_xprt_deferred_close(&svsk->sk_xprt);
898err_short:
899 return -EAGAIN;
900}
901
902static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
903{
904 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
905 struct rpc_rqst *req = NULL;
906 struct kvec *src, *dst;
907 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
908 __be32 xid;
909 __be32 calldir;
910
911 xid = *p++;
912 calldir = *p;
913
914 if (!bc_xprt)
915 return -EAGAIN;
916 spin_lock(&bc_xprt->queue_lock);
917 req = xprt_lookup_rqst(bc_xprt, xid);
918 if (!req)
919 goto unlock_notfound;
920
921 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
922 /*
923 * XXX!: cheating for now! Only copying HEAD.
924 * But we know this is good enough for now (in fact, for any
925 * callback reply in the forseeable future).
926 */
927 dst = &req->rq_private_buf.head[0];
928 src = &rqstp->rq_arg.head[0];
929 if (dst->iov_len < src->iov_len)
930 goto unlock_eagain; /* whatever; just giving up. */
931 memcpy(dst->iov_base, src->iov_base, src->iov_len);
932 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
933 rqstp->rq_arg.len = 0;
934 spin_unlock(&bc_xprt->queue_lock);
935 return 0;
936unlock_notfound:
937 printk(KERN_NOTICE
938 "%s: Got unrecognized reply: "
939 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
940 __func__, ntohl(calldir),
941 bc_xprt, ntohl(xid));
942unlock_eagain:
943 spin_unlock(&bc_xprt->queue_lock);
944 return -EAGAIN;
945}
946
947static void svc_tcp_fragment_received(struct svc_sock *svsk)
948{
949 /* If we have more data, signal svc_xprt_enqueue() to try again */
950 svsk->sk_tcplen = 0;
951 svsk->sk_marker = xdr_zero;
952}
953
954/**
955 * svc_tcp_recvfrom - Receive data from a TCP socket
956 * @rqstp: request structure into which to receive an RPC Call
957 *
958 * Called in a loop when XPT_DATA has been set.
959 *
960 * Read the 4-byte stream record marker, then use the record length
961 * in that marker to set up exactly the resources needed to receive
962 * the next RPC message into @rqstp.
963 *
964 * Returns:
965 * On success, the number of bytes in a received RPC Call, or
966 * %0 if a complete RPC Call message was not ready to return
967 *
968 * The zero return case handles partial receives and callback Replies.
969 * The state of a partial receive is preserved in the svc_sock for
970 * the next call to svc_tcp_recvfrom.
971 */
972static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
973{
974 struct svc_sock *svsk =
975 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
976 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
977 size_t want, base;
978 ssize_t len;
979 __be32 *p;
980 __be32 calldir;
981
982 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
983 len = svc_tcp_read_marker(svsk, rqstp);
984 if (len < 0)
985 goto error;
986
987 base = svc_tcp_restore_pages(svsk, rqstp);
988 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
989 len = svc_tcp_read_msg(rqstp, base + want, base);
990 if (len >= 0) {
991 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
992 svsk->sk_tcplen += len;
993 svsk->sk_datalen += len;
994 }
995 if (len != want || !svc_sock_final_rec(svsk))
996 goto err_incomplete;
997 if (svsk->sk_datalen < 8)
998 goto err_nuts;
999
1000 rqstp->rq_arg.len = svsk->sk_datalen;
1001 rqstp->rq_arg.page_base = 0;
1002 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1003 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1004 rqstp->rq_arg.page_len = 0;
1005 } else
1006 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1007
1008 rqstp->rq_xprt_ctxt = NULL;
1009 rqstp->rq_prot = IPPROTO_TCP;
1010 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1011 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1012 else
1013 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1014
1015 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1016 calldir = p[1];
1017 if (calldir)
1018 len = receive_cb_reply(svsk, rqstp);
1019
1020 /* Reset TCP read info */
1021 svsk->sk_datalen = 0;
1022 svc_tcp_fragment_received(svsk);
1023
1024 if (len < 0)
1025 goto error;
1026
1027 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1028 if (serv->sv_stats)
1029 serv->sv_stats->nettcpcnt++;
1030
1031 svc_xprt_received(rqstp->rq_xprt);
1032 return rqstp->rq_arg.len;
1033
1034err_incomplete:
1035 svc_tcp_save_pages(svsk, rqstp);
1036 if (len < 0 && len != -EAGAIN)
1037 goto err_delete;
1038 if (len == want)
1039 svc_tcp_fragment_received(svsk);
1040 else
1041 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1042 svc_sock_reclen(svsk),
1043 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1044 goto err_noclose;
1045error:
1046 if (len != -EAGAIN)
1047 goto err_delete;
1048 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1049 goto err_noclose;
1050err_nuts:
1051 svsk->sk_datalen = 0;
1052err_delete:
1053 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1054 svc_xprt_deferred_close(&svsk->sk_xprt);
1055err_noclose:
1056 svc_xprt_received(rqstp->rq_xprt);
1057 return 0; /* record not complete */
1058}
1059
1060static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1061 int flags)
1062{
1063 return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1064 offset_in_page(vec->iov_base),
1065 vec->iov_len, flags);
1066}
1067
1068/*
1069 * kernel_sendpage() is used exclusively to reduce the number of
1070 * copy operations in this path. Therefore the caller must ensure
1071 * that the pages backing @xdr are unchanging.
1072 *
1073 * In addition, the logic assumes that * .bv_len is never larger
1074 * than PAGE_SIZE.
1075 */
1076static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
1077 rpc_fraghdr marker, unsigned int *sentp)
1078{
1079 const struct kvec *head = xdr->head;
1080 const struct kvec *tail = xdr->tail;
1081 struct kvec rm = {
1082 .iov_base = &marker,
1083 .iov_len = sizeof(marker),
1084 };
1085 struct msghdr msg = {
1086 .msg_flags = 0,
1087 };
1088 int ret;
1089
1090 *sentp = 0;
1091 ret = xdr_alloc_bvec(xdr, GFP_KERNEL);
1092 if (ret < 0)
1093 return ret;
1094
1095 ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len);
1096 if (ret < 0)
1097 return ret;
1098 *sentp += ret;
1099 if (ret != rm.iov_len)
1100 return -EAGAIN;
1101
1102 ret = svc_tcp_send_kvec(sock, head, 0);
1103 if (ret < 0)
1104 return ret;
1105 *sentp += ret;
1106 if (ret != head->iov_len)
1107 goto out;
1108
1109 if (xdr->page_len) {
1110 unsigned int offset, len, remaining;
1111 struct bio_vec *bvec;
1112
1113 bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1114 offset = offset_in_page(xdr->page_base);
1115 remaining = xdr->page_len;
1116 while (remaining > 0) {
1117 len = min(remaining, bvec->bv_len - offset);
1118 ret = kernel_sendpage(sock, bvec->bv_page,
1119 bvec->bv_offset + offset,
1120 len, 0);
1121 if (ret < 0)
1122 return ret;
1123 *sentp += ret;
1124 if (ret != len)
1125 goto out;
1126 remaining -= len;
1127 offset = 0;
1128 bvec++;
1129 }
1130 }
1131
1132 if (tail->iov_len) {
1133 ret = svc_tcp_send_kvec(sock, tail, 0);
1134 if (ret < 0)
1135 return ret;
1136 *sentp += ret;
1137 }
1138
1139out:
1140 return 0;
1141}
1142
1143/**
1144 * svc_tcp_sendto - Send out a reply on a TCP socket
1145 * @rqstp: completed svc_rqst
1146 *
1147 * xpt_mutex ensures @rqstp's whole message is written to the socket
1148 * without interruption.
1149 *
1150 * Returns the number of bytes sent, or a negative errno.
1151 */
1152static int svc_tcp_sendto(struct svc_rqst *rqstp)
1153{
1154 struct svc_xprt *xprt = rqstp->rq_xprt;
1155 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1156 struct xdr_buf *xdr = &rqstp->rq_res;
1157 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1158 (u32)xdr->len);
1159 unsigned int sent;
1160 int err;
1161
1162 svc_tcp_release_rqst(rqstp);
1163
1164 atomic_inc(&svsk->sk_sendqlen);
1165 mutex_lock(&xprt->xpt_mutex);
1166 if (svc_xprt_is_dead(xprt))
1167 goto out_notconn;
1168 tcp_sock_set_cork(svsk->sk_sk, true);
1169 err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent);
1170 xdr_free_bvec(xdr);
1171 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1172 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1173 goto out_close;
1174 if (atomic_dec_and_test(&svsk->sk_sendqlen))
1175 tcp_sock_set_cork(svsk->sk_sk, false);
1176 mutex_unlock(&xprt->xpt_mutex);
1177 return sent;
1178
1179out_notconn:
1180 atomic_dec(&svsk->sk_sendqlen);
1181 mutex_unlock(&xprt->xpt_mutex);
1182 return -ENOTCONN;
1183out_close:
1184 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1185 xprt->xpt_server->sv_name,
1186 (err < 0) ? "got error" : "sent",
1187 (err < 0) ? err : sent, xdr->len);
1188 svc_xprt_deferred_close(xprt);
1189 atomic_dec(&svsk->sk_sendqlen);
1190 mutex_unlock(&xprt->xpt_mutex);
1191 return -EAGAIN;
1192}
1193
1194static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1195 struct net *net,
1196 struct sockaddr *sa, int salen,
1197 int flags)
1198{
1199 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1200}
1201
1202static const struct svc_xprt_ops svc_tcp_ops = {
1203 .xpo_create = svc_tcp_create,
1204 .xpo_recvfrom = svc_tcp_recvfrom,
1205 .xpo_sendto = svc_tcp_sendto,
1206 .xpo_result_payload = svc_sock_result_payload,
1207 .xpo_release_rqst = svc_tcp_release_rqst,
1208 .xpo_detach = svc_tcp_sock_detach,
1209 .xpo_free = svc_sock_free,
1210 .xpo_has_wspace = svc_tcp_has_wspace,
1211 .xpo_accept = svc_tcp_accept,
1212 .xpo_secure_port = svc_sock_secure_port,
1213 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1214};
1215
1216static struct svc_xprt_class svc_tcp_class = {
1217 .xcl_name = "tcp",
1218 .xcl_owner = THIS_MODULE,
1219 .xcl_ops = &svc_tcp_ops,
1220 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1221 .xcl_ident = XPRT_TRANSPORT_TCP,
1222};
1223
1224void svc_init_xprt_sock(void)
1225{
1226 svc_reg_xprt_class(&svc_tcp_class);
1227 svc_reg_xprt_class(&svc_udp_class);
1228}
1229
1230void svc_cleanup_xprt_sock(void)
1231{
1232 svc_unreg_xprt_class(&svc_tcp_class);
1233 svc_unreg_xprt_class(&svc_udp_class);
1234}
1235
1236static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1237{
1238 struct sock *sk = svsk->sk_sk;
1239
1240 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1241 &svsk->sk_xprt, serv);
1242 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1243 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1244 if (sk->sk_state == TCP_LISTEN) {
1245 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1246 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1247 sk->sk_data_ready = svc_tcp_listen_data_ready;
1248 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1249 } else {
1250 sk->sk_state_change = svc_tcp_state_change;
1251 sk->sk_data_ready = svc_data_ready;
1252 sk->sk_write_space = svc_write_space;
1253
1254 svsk->sk_marker = xdr_zero;
1255 svsk->sk_tcplen = 0;
1256 svsk->sk_datalen = 0;
1257 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1258
1259 tcp_sock_set_nodelay(sk);
1260
1261 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1262 switch (sk->sk_state) {
1263 case TCP_SYN_RECV:
1264 case TCP_ESTABLISHED:
1265 break;
1266 default:
1267 svc_xprt_deferred_close(&svsk->sk_xprt);
1268 }
1269 }
1270}
1271
1272void svc_sock_update_bufs(struct svc_serv *serv)
1273{
1274 /*
1275 * The number of server threads has changed. Update
1276 * rcvbuf and sndbuf accordingly on all sockets
1277 */
1278 struct svc_sock *svsk;
1279
1280 spin_lock_bh(&serv->sv_lock);
1281 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1282 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1283 spin_unlock_bh(&serv->sv_lock);
1284}
1285EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1286
1287/*
1288 * Initialize socket for RPC use and create svc_sock struct
1289 */
1290static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1291 struct socket *sock,
1292 int flags)
1293{
1294 struct svc_sock *svsk;
1295 struct sock *inet;
1296 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1297 int err = 0;
1298
1299 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1300 if (!svsk)
1301 return ERR_PTR(-ENOMEM);
1302
1303 inet = sock->sk;
1304
1305 /* Register socket with portmapper */
1306 if (pmap_register)
1307 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1308 inet->sk_protocol,
1309 ntohs(inet_sk(inet)->inet_sport));
1310
1311 if (err < 0) {
1312 kfree(svsk);
1313 return ERR_PTR(err);
1314 }
1315
1316 svsk->sk_sock = sock;
1317 svsk->sk_sk = inet;
1318 svsk->sk_ostate = inet->sk_state_change;
1319 svsk->sk_odata = inet->sk_data_ready;
1320 svsk->sk_owspace = inet->sk_write_space;
1321 /*
1322 * This barrier is necessary in order to prevent race condition
1323 * with svc_data_ready(), svc_listen_data_ready() and others
1324 * when calling callbacks above.
1325 */
1326 wmb();
1327 inet->sk_user_data = svsk;
1328
1329 /* Initialize the socket */
1330 if (sock->type == SOCK_DGRAM)
1331 svc_udp_init(svsk, serv);
1332 else
1333 svc_tcp_init(svsk, serv);
1334
1335 trace_svcsock_new_socket(sock);
1336 return svsk;
1337}
1338
1339bool svc_alien_sock(struct net *net, int fd)
1340{
1341 int err;
1342 struct socket *sock = sockfd_lookup(fd, &err);
1343 bool ret = false;
1344
1345 if (!sock)
1346 goto out;
1347 if (sock_net(sock->sk) != net)
1348 ret = true;
1349 sockfd_put(sock);
1350out:
1351 return ret;
1352}
1353EXPORT_SYMBOL_GPL(svc_alien_sock);
1354
1355/**
1356 * svc_addsock - add a listener socket to an RPC service
1357 * @serv: pointer to RPC service to which to add a new listener
1358 * @fd: file descriptor of the new listener
1359 * @name_return: pointer to buffer to fill in with name of listener
1360 * @len: size of the buffer
1361 * @cred: credential
1362 *
1363 * Fills in socket name and returns positive length of name if successful.
1364 * Name is terminated with '\n'. On error, returns a negative errno
1365 * value.
1366 */
1367int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1368 const size_t len, const struct cred *cred)
1369{
1370 int err = 0;
1371 struct socket *so = sockfd_lookup(fd, &err);
1372 struct svc_sock *svsk = NULL;
1373 struct sockaddr_storage addr;
1374 struct sockaddr *sin = (struct sockaddr *)&addr;
1375 int salen;
1376
1377 if (!so)
1378 return err;
1379 err = -EAFNOSUPPORT;
1380 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1381 goto out;
1382 err = -EPROTONOSUPPORT;
1383 if (so->sk->sk_protocol != IPPROTO_TCP &&
1384 so->sk->sk_protocol != IPPROTO_UDP)
1385 goto out;
1386 err = -EISCONN;
1387 if (so->state > SS_UNCONNECTED)
1388 goto out;
1389 err = -ENOENT;
1390 if (!try_module_get(THIS_MODULE))
1391 goto out;
1392 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1393 if (IS_ERR(svsk)) {
1394 module_put(THIS_MODULE);
1395 err = PTR_ERR(svsk);
1396 goto out;
1397 }
1398 salen = kernel_getsockname(svsk->sk_sock, sin);
1399 if (salen >= 0)
1400 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1401 svsk->sk_xprt.xpt_cred = get_cred(cred);
1402 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1403 return svc_one_sock_name(svsk, name_return, len);
1404out:
1405 sockfd_put(so);
1406 return err;
1407}
1408EXPORT_SYMBOL_GPL(svc_addsock);
1409
1410/*
1411 * Create socket for RPC service.
1412 */
1413static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1414 int protocol,
1415 struct net *net,
1416 struct sockaddr *sin, int len,
1417 int flags)
1418{
1419 struct svc_sock *svsk;
1420 struct socket *sock;
1421 int error;
1422 int type;
1423 struct sockaddr_storage addr;
1424 struct sockaddr *newsin = (struct sockaddr *)&addr;
1425 int newlen;
1426 int family;
1427
1428 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1429 printk(KERN_WARNING "svc: only UDP and TCP "
1430 "sockets supported\n");
1431 return ERR_PTR(-EINVAL);
1432 }
1433
1434 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1435 switch (sin->sa_family) {
1436 case AF_INET6:
1437 family = PF_INET6;
1438 break;
1439 case AF_INET:
1440 family = PF_INET;
1441 break;
1442 default:
1443 return ERR_PTR(-EINVAL);
1444 }
1445
1446 error = __sock_create(net, family, type, protocol, &sock, 1);
1447 if (error < 0)
1448 return ERR_PTR(error);
1449
1450 svc_reclassify_socket(sock);
1451
1452 /*
1453 * If this is an PF_INET6 listener, we want to avoid
1454 * getting requests from IPv4 remotes. Those should
1455 * be shunted to a PF_INET listener via rpcbind.
1456 */
1457 if (family == PF_INET6)
1458 ip6_sock_set_v6only(sock->sk);
1459 if (type == SOCK_STREAM)
1460 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1461 error = kernel_bind(sock, sin, len);
1462 if (error < 0)
1463 goto bummer;
1464
1465 error = kernel_getsockname(sock, newsin);
1466 if (error < 0)
1467 goto bummer;
1468 newlen = error;
1469
1470 if (protocol == IPPROTO_TCP) {
1471 if ((error = kernel_listen(sock, 64)) < 0)
1472 goto bummer;
1473 }
1474
1475 svsk = svc_setup_socket(serv, sock, flags);
1476 if (IS_ERR(svsk)) {
1477 error = PTR_ERR(svsk);
1478 goto bummer;
1479 }
1480 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1481 return (struct svc_xprt *)svsk;
1482bummer:
1483 sock_release(sock);
1484 return ERR_PTR(error);
1485}
1486
1487/*
1488 * Detach the svc_sock from the socket so that no
1489 * more callbacks occur.
1490 */
1491static void svc_sock_detach(struct svc_xprt *xprt)
1492{
1493 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1494 struct sock *sk = svsk->sk_sk;
1495
1496 /* put back the old socket callbacks */
1497 lock_sock(sk);
1498 sk->sk_state_change = svsk->sk_ostate;
1499 sk->sk_data_ready = svsk->sk_odata;
1500 sk->sk_write_space = svsk->sk_owspace;
1501 sk->sk_user_data = NULL;
1502 release_sock(sk);
1503}
1504
1505/*
1506 * Disconnect the socket, and reset the callbacks
1507 */
1508static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1509{
1510 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1511
1512 svc_sock_detach(xprt);
1513
1514 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1515 svc_tcp_clear_pages(svsk);
1516 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1517 }
1518}
1519
1520/*
1521 * Free the svc_sock's socket resources and the svc_sock itself.
1522 */
1523static void svc_sock_free(struct svc_xprt *xprt)
1524{
1525 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1526
1527 if (svsk->sk_sock->file)
1528 sockfd_put(svsk->sk_sock);
1529 else
1530 sock_release(svsk->sk_sock);
1531 kfree(svsk);
1532}
1// 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_result_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 svc_xprt_received(rqstp->rq_xprt);
523 return len;
524
525out_recv_err:
526 if (err != -EAGAIN) {
527 /* possibly an icmp error */
528 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
529 }
530 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
531 goto out_clear_busy;
532out_cmsg_err:
533 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
534 cmh->cmsg_level, cmh->cmsg_type);
535 goto out_free;
536out_bh_enable:
537 local_bh_enable();
538out_free:
539 kfree_skb(skb);
540out_clear_busy:
541 svc_xprt_received(rqstp->rq_xprt);
542 return 0;
543}
544
545/**
546 * svc_udp_sendto - Send out a reply on a UDP socket
547 * @rqstp: completed svc_rqst
548 *
549 * xpt_mutex ensures @rqstp's whole message is written to the socket
550 * without interruption.
551 *
552 * Returns the number of bytes sent, or a negative errno.
553 */
554static int svc_udp_sendto(struct svc_rqst *rqstp)
555{
556 struct svc_xprt *xprt = rqstp->rq_xprt;
557 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
558 struct xdr_buf *xdr = &rqstp->rq_res;
559 union {
560 struct cmsghdr hdr;
561 long all[SVC_PKTINFO_SPACE / sizeof(long)];
562 } buffer;
563 struct cmsghdr *cmh = &buffer.hdr;
564 struct msghdr msg = {
565 .msg_name = &rqstp->rq_addr,
566 .msg_namelen = rqstp->rq_addrlen,
567 .msg_control = cmh,
568 .msg_controllen = sizeof(buffer),
569 };
570 unsigned int sent;
571 int err;
572
573 svc_udp_release_rqst(rqstp);
574
575 svc_set_cmsg_data(rqstp, cmh);
576
577 mutex_lock(&xprt->xpt_mutex);
578
579 if (svc_xprt_is_dead(xprt))
580 goto out_notconn;
581
582 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
583 xdr_free_bvec(xdr);
584 if (err == -ECONNREFUSED) {
585 /* ICMP error on earlier request. */
586 err = xprt_sock_sendmsg(svsk->sk_sock, &msg, xdr, 0, 0, &sent);
587 xdr_free_bvec(xdr);
588 }
589 trace_svcsock_udp_send(xprt, err);
590
591 mutex_unlock(&xprt->xpt_mutex);
592 if (err < 0)
593 return err;
594 return sent;
595
596out_notconn:
597 mutex_unlock(&xprt->xpt_mutex);
598 return -ENOTCONN;
599}
600
601static int svc_udp_has_wspace(struct svc_xprt *xprt)
602{
603 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
604 struct svc_serv *serv = xprt->xpt_server;
605 unsigned long required;
606
607 /*
608 * Set the SOCK_NOSPACE flag before checking the available
609 * sock space.
610 */
611 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
612 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
613 if (required*2 > sock_wspace(svsk->sk_sk))
614 return 0;
615 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
616 return 1;
617}
618
619static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
620{
621 BUG();
622 return NULL;
623}
624
625static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
626{
627}
628
629static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
630 struct net *net,
631 struct sockaddr *sa, int salen,
632 int flags)
633{
634 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
635}
636
637static const struct svc_xprt_ops svc_udp_ops = {
638 .xpo_create = svc_udp_create,
639 .xpo_recvfrom = svc_udp_recvfrom,
640 .xpo_sendto = svc_udp_sendto,
641 .xpo_result_payload = svc_sock_result_payload,
642 .xpo_release_rqst = svc_udp_release_rqst,
643 .xpo_detach = svc_sock_detach,
644 .xpo_free = svc_sock_free,
645 .xpo_has_wspace = svc_udp_has_wspace,
646 .xpo_accept = svc_udp_accept,
647 .xpo_secure_port = svc_sock_secure_port,
648 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
649};
650
651static struct svc_xprt_class svc_udp_class = {
652 .xcl_name = "udp",
653 .xcl_owner = THIS_MODULE,
654 .xcl_ops = &svc_udp_ops,
655 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
656 .xcl_ident = XPRT_TRANSPORT_UDP,
657};
658
659static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
660{
661 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
662 &svsk->sk_xprt, serv);
663 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
664 svsk->sk_sk->sk_data_ready = svc_data_ready;
665 svsk->sk_sk->sk_write_space = svc_write_space;
666
667 /* initialise setting must have enough space to
668 * receive and respond to one request.
669 * svc_udp_recvfrom will re-adjust if necessary
670 */
671 svc_sock_setbufsize(svsk, 3);
672
673 /* data might have come in before data_ready set up */
674 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
675 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
676
677 /* make sure we get destination address info */
678 switch (svsk->sk_sk->sk_family) {
679 case AF_INET:
680 ip_sock_set_pktinfo(svsk->sk_sock->sk);
681 break;
682 case AF_INET6:
683 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
684 break;
685 default:
686 BUG();
687 }
688}
689
690/*
691 * A data_ready event on a listening socket means there's a connection
692 * pending. Do not use state_change as a substitute for it.
693 */
694static void svc_tcp_listen_data_ready(struct sock *sk)
695{
696 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
697
698 if (svsk) {
699 /* Refer to svc_setup_socket() for details. */
700 rmb();
701 svsk->sk_odata(sk);
702 }
703
704 /*
705 * This callback may called twice when a new connection
706 * is established as a child socket inherits everything
707 * from a parent LISTEN socket.
708 * 1) data_ready method of the parent socket will be called
709 * when one of child sockets become ESTABLISHED.
710 * 2) data_ready method of the child socket may be called
711 * when it receives data before the socket is accepted.
712 * In case of 2, we should ignore it silently.
713 */
714 if (sk->sk_state == TCP_LISTEN) {
715 if (svsk) {
716 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
717 svc_xprt_enqueue(&svsk->sk_xprt);
718 }
719 }
720}
721
722/*
723 * A state change on a connected socket means it's dying or dead.
724 */
725static void svc_tcp_state_change(struct sock *sk)
726{
727 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
728
729 if (svsk) {
730 /* Refer to svc_setup_socket() for details. */
731 rmb();
732 svsk->sk_ostate(sk);
733 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
734 if (sk->sk_state != TCP_ESTABLISHED)
735 svc_xprt_deferred_close(&svsk->sk_xprt);
736 }
737}
738
739/*
740 * Accept a TCP connection
741 */
742static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
743{
744 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
745 struct sockaddr_storage addr;
746 struct sockaddr *sin = (struct sockaddr *) &addr;
747 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
748 struct socket *sock = svsk->sk_sock;
749 struct socket *newsock;
750 struct svc_sock *newsvsk;
751 int err, slen;
752
753 if (!sock)
754 return NULL;
755
756 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
757 err = kernel_accept(sock, &newsock, O_NONBLOCK);
758 if (err < 0) {
759 if (err == -ENOMEM)
760 printk(KERN_WARNING "%s: no more sockets!\n",
761 serv->sv_name);
762 else if (err != -EAGAIN)
763 net_warn_ratelimited("%s: accept failed (err %d)!\n",
764 serv->sv_name, -err);
765 trace_svcsock_accept_err(xprt, serv->sv_name, err);
766 return NULL;
767 }
768 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
769
770 err = kernel_getpeername(newsock, sin);
771 if (err < 0) {
772 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
773 goto failed; /* aborted connection or whatever */
774 }
775 slen = err;
776
777 /* Reset the inherited callbacks before calling svc_setup_socket */
778 newsock->sk->sk_state_change = svsk->sk_ostate;
779 newsock->sk->sk_data_ready = svsk->sk_odata;
780 newsock->sk->sk_write_space = svsk->sk_owspace;
781
782 /* make sure that a write doesn't block forever when
783 * low on memory
784 */
785 newsock->sk->sk_sndtimeo = HZ*30;
786
787 newsvsk = svc_setup_socket(serv, newsock,
788 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
789 if (IS_ERR(newsvsk))
790 goto failed;
791 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
792 err = kernel_getsockname(newsock, sin);
793 slen = err;
794 if (unlikely(err < 0))
795 slen = offsetof(struct sockaddr, sa_data);
796 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
797
798 if (sock_is_loopback(newsock->sk))
799 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
800 else
801 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
802 if (serv->sv_stats)
803 serv->sv_stats->nettcpconn++;
804
805 return &newsvsk->sk_xprt;
806
807failed:
808 sock_release(newsock);
809 return NULL;
810}
811
812static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
813 struct svc_rqst *rqstp)
814{
815 size_t len = svsk->sk_datalen;
816 unsigned int i, npages;
817
818 if (!len)
819 return 0;
820 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
821 for (i = 0; i < npages; i++) {
822 if (rqstp->rq_pages[i] != NULL)
823 put_page(rqstp->rq_pages[i]);
824 BUG_ON(svsk->sk_pages[i] == NULL);
825 rqstp->rq_pages[i] = svsk->sk_pages[i];
826 svsk->sk_pages[i] = NULL;
827 }
828 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
829 return len;
830}
831
832static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
833{
834 unsigned int i, len, npages;
835
836 if (svsk->sk_datalen == 0)
837 return;
838 len = svsk->sk_datalen;
839 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
840 for (i = 0; i < npages; i++) {
841 svsk->sk_pages[i] = rqstp->rq_pages[i];
842 rqstp->rq_pages[i] = NULL;
843 }
844}
845
846static void svc_tcp_clear_pages(struct svc_sock *svsk)
847{
848 unsigned int i, len, npages;
849
850 if (svsk->sk_datalen == 0)
851 goto out;
852 len = svsk->sk_datalen;
853 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
854 for (i = 0; i < npages; i++) {
855 if (svsk->sk_pages[i] == NULL) {
856 WARN_ON_ONCE(1);
857 continue;
858 }
859 put_page(svsk->sk_pages[i]);
860 svsk->sk_pages[i] = NULL;
861 }
862out:
863 svsk->sk_tcplen = 0;
864 svsk->sk_datalen = 0;
865}
866
867/*
868 * Receive fragment record header into sk_marker.
869 */
870static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
871 struct svc_rqst *rqstp)
872{
873 ssize_t want, len;
874
875 /* If we haven't gotten the record length yet,
876 * get the next four bytes.
877 */
878 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
879 struct msghdr msg = { NULL };
880 struct kvec iov;
881
882 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
883 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
884 iov.iov_len = want;
885 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, want);
886 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
887 if (len < 0)
888 return len;
889 svsk->sk_tcplen += len;
890 if (len < want) {
891 /* call again to read the remaining bytes */
892 goto err_short;
893 }
894 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
895 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
896 svsk->sk_xprt.xpt_server->sv_max_mesg)
897 goto err_too_large;
898 }
899 return svc_sock_reclen(svsk);
900
901err_too_large:
902 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
903 __func__, svsk->sk_xprt.xpt_server->sv_name,
904 svc_sock_reclen(svsk));
905 svc_xprt_deferred_close(&svsk->sk_xprt);
906err_short:
907 return -EAGAIN;
908}
909
910static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
911{
912 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
913 struct rpc_rqst *req = NULL;
914 struct kvec *src, *dst;
915 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
916 __be32 xid;
917 __be32 calldir;
918
919 xid = *p++;
920 calldir = *p;
921
922 if (!bc_xprt)
923 return -EAGAIN;
924 spin_lock(&bc_xprt->queue_lock);
925 req = xprt_lookup_rqst(bc_xprt, xid);
926 if (!req)
927 goto unlock_notfound;
928
929 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
930 /*
931 * XXX!: cheating for now! Only copying HEAD.
932 * But we know this is good enough for now (in fact, for any
933 * callback reply in the forseeable future).
934 */
935 dst = &req->rq_private_buf.head[0];
936 src = &rqstp->rq_arg.head[0];
937 if (dst->iov_len < src->iov_len)
938 goto unlock_eagain; /* whatever; just giving up. */
939 memcpy(dst->iov_base, src->iov_base, src->iov_len);
940 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
941 rqstp->rq_arg.len = 0;
942 spin_unlock(&bc_xprt->queue_lock);
943 return 0;
944unlock_notfound:
945 printk(KERN_NOTICE
946 "%s: Got unrecognized reply: "
947 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
948 __func__, ntohl(calldir),
949 bc_xprt, ntohl(xid));
950unlock_eagain:
951 spin_unlock(&bc_xprt->queue_lock);
952 return -EAGAIN;
953}
954
955static void svc_tcp_fragment_received(struct svc_sock *svsk)
956{
957 /* If we have more data, signal svc_xprt_enqueue() to try again */
958 svsk->sk_tcplen = 0;
959 svsk->sk_marker = xdr_zero;
960}
961
962/**
963 * svc_tcp_recvfrom - Receive data from a TCP socket
964 * @rqstp: request structure into which to receive an RPC Call
965 *
966 * Called in a loop when XPT_DATA has been set.
967 *
968 * Read the 4-byte stream record marker, then use the record length
969 * in that marker to set up exactly the resources needed to receive
970 * the next RPC message into @rqstp.
971 *
972 * Returns:
973 * On success, the number of bytes in a received RPC Call, or
974 * %0 if a complete RPC Call message was not ready to return
975 *
976 * The zero return case handles partial receives and callback Replies.
977 * The state of a partial receive is preserved in the svc_sock for
978 * the next call to svc_tcp_recvfrom.
979 */
980static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
981{
982 struct svc_sock *svsk =
983 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
984 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
985 size_t want, base;
986 ssize_t len;
987 __be32 *p;
988 __be32 calldir;
989
990 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
991 len = svc_tcp_read_marker(svsk, rqstp);
992 if (len < 0)
993 goto error;
994
995 base = svc_tcp_restore_pages(svsk, rqstp);
996 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
997 len = svc_tcp_read_msg(rqstp, base + want, base);
998 if (len >= 0) {
999 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1000 svsk->sk_tcplen += len;
1001 svsk->sk_datalen += len;
1002 }
1003 if (len != want || !svc_sock_final_rec(svsk))
1004 goto err_incomplete;
1005 if (svsk->sk_datalen < 8)
1006 goto err_nuts;
1007
1008 rqstp->rq_arg.len = svsk->sk_datalen;
1009 rqstp->rq_arg.page_base = 0;
1010 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1011 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1012 rqstp->rq_arg.page_len = 0;
1013 } else
1014 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1015
1016 rqstp->rq_xprt_ctxt = NULL;
1017 rqstp->rq_prot = IPPROTO_TCP;
1018 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1019 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1020 else
1021 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1022
1023 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1024 calldir = p[1];
1025 if (calldir)
1026 len = receive_cb_reply(svsk, rqstp);
1027
1028 /* Reset TCP read info */
1029 svsk->sk_datalen = 0;
1030 svc_tcp_fragment_received(svsk);
1031
1032 if (len < 0)
1033 goto error;
1034
1035 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1036 if (serv->sv_stats)
1037 serv->sv_stats->nettcpcnt++;
1038
1039 svc_xprt_received(rqstp->rq_xprt);
1040 return rqstp->rq_arg.len;
1041
1042err_incomplete:
1043 svc_tcp_save_pages(svsk, rqstp);
1044 if (len < 0 && len != -EAGAIN)
1045 goto err_delete;
1046 if (len == want)
1047 svc_tcp_fragment_received(svsk);
1048 else
1049 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1050 svc_sock_reclen(svsk),
1051 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1052 goto err_noclose;
1053error:
1054 if (len != -EAGAIN)
1055 goto err_delete;
1056 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1057 goto err_noclose;
1058err_nuts:
1059 svsk->sk_datalen = 0;
1060err_delete:
1061 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1062 svc_xprt_deferred_close(&svsk->sk_xprt);
1063err_noclose:
1064 svc_xprt_received(rqstp->rq_xprt);
1065 return 0; /* record not complete */
1066}
1067
1068static int svc_tcp_send_kvec(struct socket *sock, const struct kvec *vec,
1069 int flags)
1070{
1071 return kernel_sendpage(sock, virt_to_page(vec->iov_base),
1072 offset_in_page(vec->iov_base),
1073 vec->iov_len, flags);
1074}
1075
1076/*
1077 * kernel_sendpage() is used exclusively to reduce the number of
1078 * copy operations in this path. Therefore the caller must ensure
1079 * that the pages backing @xdr are unchanging.
1080 *
1081 * In addition, the logic assumes that * .bv_len is never larger
1082 * than PAGE_SIZE.
1083 */
1084static int svc_tcp_sendmsg(struct socket *sock, struct xdr_buf *xdr,
1085 rpc_fraghdr marker, unsigned int *sentp)
1086{
1087 const struct kvec *head = xdr->head;
1088 const struct kvec *tail = xdr->tail;
1089 struct kvec rm = {
1090 .iov_base = &marker,
1091 .iov_len = sizeof(marker),
1092 };
1093 struct msghdr msg = {
1094 .msg_flags = 0,
1095 };
1096 int ret;
1097
1098 *sentp = 0;
1099 xdr_alloc_bvec(xdr, GFP_KERNEL);
1100
1101 ret = kernel_sendmsg(sock, &msg, &rm, 1, rm.iov_len);
1102 if (ret < 0)
1103 return ret;
1104 *sentp += ret;
1105 if (ret != rm.iov_len)
1106 return -EAGAIN;
1107
1108 ret = svc_tcp_send_kvec(sock, head, 0);
1109 if (ret < 0)
1110 return ret;
1111 *sentp += ret;
1112 if (ret != head->iov_len)
1113 goto out;
1114
1115 if (xdr->page_len) {
1116 unsigned int offset, len, remaining;
1117 struct bio_vec *bvec;
1118
1119 bvec = xdr->bvec + (xdr->page_base >> PAGE_SHIFT);
1120 offset = offset_in_page(xdr->page_base);
1121 remaining = xdr->page_len;
1122 while (remaining > 0) {
1123 len = min(remaining, bvec->bv_len - offset);
1124 ret = kernel_sendpage(sock, bvec->bv_page,
1125 bvec->bv_offset + offset,
1126 len, 0);
1127 if (ret < 0)
1128 return ret;
1129 *sentp += ret;
1130 if (ret != len)
1131 goto out;
1132 remaining -= len;
1133 offset = 0;
1134 bvec++;
1135 }
1136 }
1137
1138 if (tail->iov_len) {
1139 ret = svc_tcp_send_kvec(sock, tail, 0);
1140 if (ret < 0)
1141 return ret;
1142 *sentp += ret;
1143 }
1144
1145out:
1146 return 0;
1147}
1148
1149/**
1150 * svc_tcp_sendto - Send out a reply on a TCP socket
1151 * @rqstp: completed svc_rqst
1152 *
1153 * xpt_mutex ensures @rqstp's whole message is written to the socket
1154 * without interruption.
1155 *
1156 * Returns the number of bytes sent, or a negative errno.
1157 */
1158static int svc_tcp_sendto(struct svc_rqst *rqstp)
1159{
1160 struct svc_xprt *xprt = rqstp->rq_xprt;
1161 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1162 struct xdr_buf *xdr = &rqstp->rq_res;
1163 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1164 (u32)xdr->len);
1165 unsigned int sent;
1166 int err;
1167
1168 svc_tcp_release_rqst(rqstp);
1169
1170 atomic_inc(&svsk->sk_sendqlen);
1171 mutex_lock(&xprt->xpt_mutex);
1172 if (svc_xprt_is_dead(xprt))
1173 goto out_notconn;
1174 tcp_sock_set_cork(svsk->sk_sk, true);
1175 err = svc_tcp_sendmsg(svsk->sk_sock, xdr, marker, &sent);
1176 xdr_free_bvec(xdr);
1177 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1178 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1179 goto out_close;
1180 if (atomic_dec_and_test(&svsk->sk_sendqlen))
1181 tcp_sock_set_cork(svsk->sk_sk, false);
1182 mutex_unlock(&xprt->xpt_mutex);
1183 return sent;
1184
1185out_notconn:
1186 atomic_dec(&svsk->sk_sendqlen);
1187 mutex_unlock(&xprt->xpt_mutex);
1188 return -ENOTCONN;
1189out_close:
1190 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1191 xprt->xpt_server->sv_name,
1192 (err < 0) ? "got error" : "sent",
1193 (err < 0) ? err : sent, xdr->len);
1194 svc_xprt_deferred_close(xprt);
1195 atomic_dec(&svsk->sk_sendqlen);
1196 mutex_unlock(&xprt->xpt_mutex);
1197 return -EAGAIN;
1198}
1199
1200static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1201 struct net *net,
1202 struct sockaddr *sa, int salen,
1203 int flags)
1204{
1205 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1206}
1207
1208static const struct svc_xprt_ops svc_tcp_ops = {
1209 .xpo_create = svc_tcp_create,
1210 .xpo_recvfrom = svc_tcp_recvfrom,
1211 .xpo_sendto = svc_tcp_sendto,
1212 .xpo_result_payload = svc_sock_result_payload,
1213 .xpo_release_rqst = svc_tcp_release_rqst,
1214 .xpo_detach = svc_tcp_sock_detach,
1215 .xpo_free = svc_sock_free,
1216 .xpo_has_wspace = svc_tcp_has_wspace,
1217 .xpo_accept = svc_tcp_accept,
1218 .xpo_secure_port = svc_sock_secure_port,
1219 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1220};
1221
1222static struct svc_xprt_class svc_tcp_class = {
1223 .xcl_name = "tcp",
1224 .xcl_owner = THIS_MODULE,
1225 .xcl_ops = &svc_tcp_ops,
1226 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1227 .xcl_ident = XPRT_TRANSPORT_TCP,
1228};
1229
1230void svc_init_xprt_sock(void)
1231{
1232 svc_reg_xprt_class(&svc_tcp_class);
1233 svc_reg_xprt_class(&svc_udp_class);
1234}
1235
1236void svc_cleanup_xprt_sock(void)
1237{
1238 svc_unreg_xprt_class(&svc_tcp_class);
1239 svc_unreg_xprt_class(&svc_udp_class);
1240}
1241
1242static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1243{
1244 struct sock *sk = svsk->sk_sk;
1245
1246 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1247 &svsk->sk_xprt, serv);
1248 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1249 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1250 if (sk->sk_state == TCP_LISTEN) {
1251 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1252 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1253 sk->sk_data_ready = svc_tcp_listen_data_ready;
1254 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1255 } else {
1256 sk->sk_state_change = svc_tcp_state_change;
1257 sk->sk_data_ready = svc_data_ready;
1258 sk->sk_write_space = svc_write_space;
1259
1260 svsk->sk_marker = xdr_zero;
1261 svsk->sk_tcplen = 0;
1262 svsk->sk_datalen = 0;
1263 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1264
1265 tcp_sock_set_nodelay(sk);
1266
1267 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1268 switch (sk->sk_state) {
1269 case TCP_SYN_RECV:
1270 case TCP_ESTABLISHED:
1271 break;
1272 default:
1273 svc_xprt_deferred_close(&svsk->sk_xprt);
1274 }
1275 }
1276}
1277
1278void svc_sock_update_bufs(struct svc_serv *serv)
1279{
1280 /*
1281 * The number of server threads has changed. Update
1282 * rcvbuf and sndbuf accordingly on all sockets
1283 */
1284 struct svc_sock *svsk;
1285
1286 spin_lock_bh(&serv->sv_lock);
1287 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1288 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1289 spin_unlock_bh(&serv->sv_lock);
1290}
1291EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1292
1293/*
1294 * Initialize socket for RPC use and create svc_sock struct
1295 */
1296static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1297 struct socket *sock,
1298 int flags)
1299{
1300 struct svc_sock *svsk;
1301 struct sock *inet;
1302 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1303 int err = 0;
1304
1305 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1306 if (!svsk)
1307 return ERR_PTR(-ENOMEM);
1308
1309 inet = sock->sk;
1310
1311 /* Register socket with portmapper */
1312 if (pmap_register)
1313 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1314 inet->sk_protocol,
1315 ntohs(inet_sk(inet)->inet_sport));
1316
1317 if (err < 0) {
1318 kfree(svsk);
1319 return ERR_PTR(err);
1320 }
1321
1322 svsk->sk_sock = sock;
1323 svsk->sk_sk = inet;
1324 svsk->sk_ostate = inet->sk_state_change;
1325 svsk->sk_odata = inet->sk_data_ready;
1326 svsk->sk_owspace = inet->sk_write_space;
1327 /*
1328 * This barrier is necessary in order to prevent race condition
1329 * with svc_data_ready(), svc_listen_data_ready() and others
1330 * when calling callbacks above.
1331 */
1332 wmb();
1333 inet->sk_user_data = svsk;
1334
1335 /* Initialize the socket */
1336 if (sock->type == SOCK_DGRAM)
1337 svc_udp_init(svsk, serv);
1338 else
1339 svc_tcp_init(svsk, serv);
1340
1341 trace_svcsock_new_socket(sock);
1342 return svsk;
1343}
1344
1345bool svc_alien_sock(struct net *net, int fd)
1346{
1347 int err;
1348 struct socket *sock = sockfd_lookup(fd, &err);
1349 bool ret = false;
1350
1351 if (!sock)
1352 goto out;
1353 if (sock_net(sock->sk) != net)
1354 ret = true;
1355 sockfd_put(sock);
1356out:
1357 return ret;
1358}
1359EXPORT_SYMBOL_GPL(svc_alien_sock);
1360
1361/**
1362 * svc_addsock - add a listener socket to an RPC service
1363 * @serv: pointer to RPC service to which to add a new listener
1364 * @fd: file descriptor of the new listener
1365 * @name_return: pointer to buffer to fill in with name of listener
1366 * @len: size of the buffer
1367 * @cred: credential
1368 *
1369 * Fills in socket name and returns positive length of name if successful.
1370 * Name is terminated with '\n'. On error, returns a negative errno
1371 * value.
1372 */
1373int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1374 const size_t len, const struct cred *cred)
1375{
1376 int err = 0;
1377 struct socket *so = sockfd_lookup(fd, &err);
1378 struct svc_sock *svsk = NULL;
1379 struct sockaddr_storage addr;
1380 struct sockaddr *sin = (struct sockaddr *)&addr;
1381 int salen;
1382
1383 if (!so)
1384 return err;
1385 err = -EAFNOSUPPORT;
1386 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1387 goto out;
1388 err = -EPROTONOSUPPORT;
1389 if (so->sk->sk_protocol != IPPROTO_TCP &&
1390 so->sk->sk_protocol != IPPROTO_UDP)
1391 goto out;
1392 err = -EISCONN;
1393 if (so->state > SS_UNCONNECTED)
1394 goto out;
1395 err = -ENOENT;
1396 if (!try_module_get(THIS_MODULE))
1397 goto out;
1398 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1399 if (IS_ERR(svsk)) {
1400 module_put(THIS_MODULE);
1401 err = PTR_ERR(svsk);
1402 goto out;
1403 }
1404 salen = kernel_getsockname(svsk->sk_sock, sin);
1405 if (salen >= 0)
1406 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1407 svsk->sk_xprt.xpt_cred = get_cred(cred);
1408 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1409 return svc_one_sock_name(svsk, name_return, len);
1410out:
1411 sockfd_put(so);
1412 return err;
1413}
1414EXPORT_SYMBOL_GPL(svc_addsock);
1415
1416/*
1417 * Create socket for RPC service.
1418 */
1419static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1420 int protocol,
1421 struct net *net,
1422 struct sockaddr *sin, int len,
1423 int flags)
1424{
1425 struct svc_sock *svsk;
1426 struct socket *sock;
1427 int error;
1428 int type;
1429 struct sockaddr_storage addr;
1430 struct sockaddr *newsin = (struct sockaddr *)&addr;
1431 int newlen;
1432 int family;
1433
1434 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1435 printk(KERN_WARNING "svc: only UDP and TCP "
1436 "sockets supported\n");
1437 return ERR_PTR(-EINVAL);
1438 }
1439
1440 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1441 switch (sin->sa_family) {
1442 case AF_INET6:
1443 family = PF_INET6;
1444 break;
1445 case AF_INET:
1446 family = PF_INET;
1447 break;
1448 default:
1449 return ERR_PTR(-EINVAL);
1450 }
1451
1452 error = __sock_create(net, family, type, protocol, &sock, 1);
1453 if (error < 0)
1454 return ERR_PTR(error);
1455
1456 svc_reclassify_socket(sock);
1457
1458 /*
1459 * If this is an PF_INET6 listener, we want to avoid
1460 * getting requests from IPv4 remotes. Those should
1461 * be shunted to a PF_INET listener via rpcbind.
1462 */
1463 if (family == PF_INET6)
1464 ip6_sock_set_v6only(sock->sk);
1465 if (type == SOCK_STREAM)
1466 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1467 error = kernel_bind(sock, sin, len);
1468 if (error < 0)
1469 goto bummer;
1470
1471 error = kernel_getsockname(sock, newsin);
1472 if (error < 0)
1473 goto bummer;
1474 newlen = error;
1475
1476 if (protocol == IPPROTO_TCP) {
1477 if ((error = kernel_listen(sock, 64)) < 0)
1478 goto bummer;
1479 }
1480
1481 svsk = svc_setup_socket(serv, sock, flags);
1482 if (IS_ERR(svsk)) {
1483 error = PTR_ERR(svsk);
1484 goto bummer;
1485 }
1486 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1487 return (struct svc_xprt *)svsk;
1488bummer:
1489 sock_release(sock);
1490 return ERR_PTR(error);
1491}
1492
1493/*
1494 * Detach the svc_sock from the socket so that no
1495 * more callbacks occur.
1496 */
1497static void svc_sock_detach(struct svc_xprt *xprt)
1498{
1499 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1500 struct sock *sk = svsk->sk_sk;
1501
1502 /* put back the old socket callbacks */
1503 lock_sock(sk);
1504 sk->sk_state_change = svsk->sk_ostate;
1505 sk->sk_data_ready = svsk->sk_odata;
1506 sk->sk_write_space = svsk->sk_owspace;
1507 sk->sk_user_data = NULL;
1508 release_sock(sk);
1509}
1510
1511/*
1512 * Disconnect the socket, and reset the callbacks
1513 */
1514static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1515{
1516 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1517
1518 svc_sock_detach(xprt);
1519
1520 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1521 svc_tcp_clear_pages(svsk);
1522 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1523 }
1524}
1525
1526/*
1527 * Free the svc_sock's socket resources and the svc_sock itself.
1528 */
1529static void svc_sock_free(struct svc_xprt *xprt)
1530{
1531 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1532
1533 if (svsk->sk_sock->file)
1534 sockfd_put(svsk->sk_sock);
1535 else
1536 sock_release(svsk->sk_sock);
1537 kfree(svsk);
1538}