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 <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 smp_wmb();
1088 tcp_set_rcvlowat(svsk->sk_sk, 1);
1089}
1090
1091/**
1092 * svc_tcp_recvfrom - Receive data from a TCP socket
1093 * @rqstp: request structure into which to receive an RPC Call
1094 *
1095 * Called in a loop when XPT_DATA has been set.
1096 *
1097 * Read the 4-byte stream record marker, then use the record length
1098 * in that marker to set up exactly the resources needed to receive
1099 * the next RPC message into @rqstp.
1100 *
1101 * Returns:
1102 * On success, the number of bytes in a received RPC Call, or
1103 * %0 if a complete RPC Call message was not ready to return
1104 *
1105 * The zero return case handles partial receives and callback Replies.
1106 * The state of a partial receive is preserved in the svc_sock for
1107 * the next call to svc_tcp_recvfrom.
1108 */
1109static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1110{
1111 struct svc_sock *svsk =
1112 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1113 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1114 size_t want, base;
1115 ssize_t len;
1116 __be32 *p;
1117 __be32 calldir;
1118
1119 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1120 len = svc_tcp_read_marker(svsk, rqstp);
1121 if (len < 0)
1122 goto error;
1123
1124 base = svc_tcp_restore_pages(svsk, rqstp);
1125 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1126 len = svc_tcp_read_msg(rqstp, base + want, base);
1127 if (len >= 0) {
1128 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1129 svsk->sk_tcplen += len;
1130 svsk->sk_datalen += len;
1131 }
1132 if (len != want || !svc_sock_final_rec(svsk))
1133 goto err_incomplete;
1134 if (svsk->sk_datalen < 8)
1135 goto err_nuts;
1136
1137 rqstp->rq_arg.len = svsk->sk_datalen;
1138 rqstp->rq_arg.page_base = 0;
1139 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1140 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1141 rqstp->rq_arg.page_len = 0;
1142 } else
1143 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1144
1145 rqstp->rq_xprt_ctxt = NULL;
1146 rqstp->rq_prot = IPPROTO_TCP;
1147 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1148 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1149 else
1150 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1151
1152 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1153 calldir = p[1];
1154 if (calldir)
1155 len = receive_cb_reply(svsk, rqstp);
1156
1157 /* Reset TCP read info */
1158 svsk->sk_datalen = 0;
1159 svc_tcp_fragment_received(svsk);
1160
1161 if (len < 0)
1162 goto error;
1163
1164 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1165 if (serv->sv_stats)
1166 serv->sv_stats->nettcpcnt++;
1167
1168 svc_sock_secure_port(rqstp);
1169 svc_xprt_received(rqstp->rq_xprt);
1170 return rqstp->rq_arg.len;
1171
1172err_incomplete:
1173 svc_tcp_save_pages(svsk, rqstp);
1174 if (len < 0 && len != -EAGAIN)
1175 goto err_delete;
1176 if (len == want)
1177 svc_tcp_fragment_received(svsk);
1178 else {
1179 /* Avoid more ->sk_data_ready() calls until the rest
1180 * of the message has arrived. This reduces service
1181 * thread wake-ups on large incoming messages. */
1182 tcp_set_rcvlowat(svsk->sk_sk,
1183 svc_sock_reclen(svsk) - svsk->sk_tcplen);
1184
1185 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1186 svc_sock_reclen(svsk),
1187 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1188 }
1189 goto err_noclose;
1190error:
1191 if (len != -EAGAIN)
1192 goto err_delete;
1193 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1194 goto err_noclose;
1195err_nuts:
1196 svsk->sk_datalen = 0;
1197err_delete:
1198 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1199 svc_xprt_deferred_close(&svsk->sk_xprt);
1200err_noclose:
1201 svc_xprt_received(rqstp->rq_xprt);
1202 return 0; /* record not complete */
1203}
1204
1205/*
1206 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1207 * copy operations in this path. Therefore the caller must ensure
1208 * that the pages backing @xdr are unchanging.
1209 */
1210static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1211 rpc_fraghdr marker, unsigned int *sentp)
1212{
1213 struct msghdr msg = {
1214 .msg_flags = MSG_SPLICE_PAGES,
1215 };
1216 unsigned int count;
1217 void *buf;
1218 int ret;
1219
1220 *sentp = 0;
1221
1222 /* The stream record marker is copied into a temporary page
1223 * fragment buffer so that it can be included in rq_bvec.
1224 */
1225 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1226 GFP_KERNEL);
1227 if (!buf)
1228 return -ENOMEM;
1229 memcpy(buf, &marker, sizeof(marker));
1230 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1231
1232 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1233 ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1234
1235 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1236 1 + count, sizeof(marker) + rqstp->rq_res.len);
1237 ret = sock_sendmsg(svsk->sk_sock, &msg);
1238 page_frag_free(buf);
1239 if (ret < 0)
1240 return ret;
1241 *sentp += ret;
1242 return 0;
1243}
1244
1245/**
1246 * svc_tcp_sendto - Send out a reply on a TCP socket
1247 * @rqstp: completed svc_rqst
1248 *
1249 * xpt_mutex ensures @rqstp's whole message is written to the socket
1250 * without interruption.
1251 *
1252 * Returns the number of bytes sent, or a negative errno.
1253 */
1254static int svc_tcp_sendto(struct svc_rqst *rqstp)
1255{
1256 struct svc_xprt *xprt = rqstp->rq_xprt;
1257 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1258 struct xdr_buf *xdr = &rqstp->rq_res;
1259 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1260 (u32)xdr->len);
1261 unsigned int sent;
1262 int err;
1263
1264 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1265 rqstp->rq_xprt_ctxt = NULL;
1266
1267 mutex_lock(&xprt->xpt_mutex);
1268 if (svc_xprt_is_dead(xprt))
1269 goto out_notconn;
1270 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1271 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1272 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1273 goto out_close;
1274 mutex_unlock(&xprt->xpt_mutex);
1275 return sent;
1276
1277out_notconn:
1278 mutex_unlock(&xprt->xpt_mutex);
1279 return -ENOTCONN;
1280out_close:
1281 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1282 xprt->xpt_server->sv_name,
1283 (err < 0) ? "got error" : "sent",
1284 (err < 0) ? err : sent, xdr->len);
1285 svc_xprt_deferred_close(xprt);
1286 mutex_unlock(&xprt->xpt_mutex);
1287 return -EAGAIN;
1288}
1289
1290static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1291 struct net *net,
1292 struct sockaddr *sa, int salen,
1293 int flags)
1294{
1295 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1296}
1297
1298static const struct svc_xprt_ops svc_tcp_ops = {
1299 .xpo_create = svc_tcp_create,
1300 .xpo_recvfrom = svc_tcp_recvfrom,
1301 .xpo_sendto = svc_tcp_sendto,
1302 .xpo_result_payload = svc_sock_result_payload,
1303 .xpo_release_ctxt = svc_tcp_release_ctxt,
1304 .xpo_detach = svc_tcp_sock_detach,
1305 .xpo_free = svc_sock_free,
1306 .xpo_has_wspace = svc_tcp_has_wspace,
1307 .xpo_accept = svc_tcp_accept,
1308 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1309 .xpo_handshake = svc_tcp_handshake,
1310};
1311
1312static struct svc_xprt_class svc_tcp_class = {
1313 .xcl_name = "tcp",
1314 .xcl_owner = THIS_MODULE,
1315 .xcl_ops = &svc_tcp_ops,
1316 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1317 .xcl_ident = XPRT_TRANSPORT_TCP,
1318};
1319
1320void svc_init_xprt_sock(void)
1321{
1322 svc_reg_xprt_class(&svc_tcp_class);
1323 svc_reg_xprt_class(&svc_udp_class);
1324}
1325
1326void svc_cleanup_xprt_sock(void)
1327{
1328 svc_unreg_xprt_class(&svc_tcp_class);
1329 svc_unreg_xprt_class(&svc_udp_class);
1330}
1331
1332static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1333{
1334 struct sock *sk = svsk->sk_sk;
1335
1336 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1337 &svsk->sk_xprt, serv);
1338 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1339 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1340 if (sk->sk_state == TCP_LISTEN) {
1341 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1342 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1343 sk->sk_data_ready = svc_tcp_listen_data_ready;
1344 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1345 } else {
1346 sk->sk_state_change = svc_tcp_state_change;
1347 sk->sk_data_ready = svc_data_ready;
1348 sk->sk_write_space = svc_write_space;
1349
1350 svsk->sk_marker = xdr_zero;
1351 svsk->sk_tcplen = 0;
1352 svsk->sk_datalen = 0;
1353 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1354
1355 tcp_sock_set_nodelay(sk);
1356
1357 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1358 switch (sk->sk_state) {
1359 case TCP_SYN_RECV:
1360 case TCP_ESTABLISHED:
1361 break;
1362 default:
1363 svc_xprt_deferred_close(&svsk->sk_xprt);
1364 }
1365 }
1366}
1367
1368void svc_sock_update_bufs(struct svc_serv *serv)
1369{
1370 /*
1371 * The number of server threads has changed. Update
1372 * rcvbuf and sndbuf accordingly on all sockets
1373 */
1374 struct svc_sock *svsk;
1375
1376 spin_lock_bh(&serv->sv_lock);
1377 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1378 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1379 spin_unlock_bh(&serv->sv_lock);
1380}
1381EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1382
1383/*
1384 * Initialize socket for RPC use and create svc_sock struct
1385 */
1386static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1387 struct socket *sock,
1388 int flags)
1389{
1390 struct svc_sock *svsk;
1391 struct sock *inet;
1392 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1393
1394 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1395 if (!svsk)
1396 return ERR_PTR(-ENOMEM);
1397
1398 inet = sock->sk;
1399
1400 if (pmap_register) {
1401 int err;
1402
1403 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1404 inet->sk_protocol,
1405 ntohs(inet_sk(inet)->inet_sport));
1406 if (err < 0) {
1407 kfree(svsk);
1408 return ERR_PTR(err);
1409 }
1410 }
1411
1412 svsk->sk_sock = sock;
1413 svsk->sk_sk = inet;
1414 svsk->sk_ostate = inet->sk_state_change;
1415 svsk->sk_odata = inet->sk_data_ready;
1416 svsk->sk_owspace = inet->sk_write_space;
1417 /*
1418 * This barrier is necessary in order to prevent race condition
1419 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1420 * when calling callbacks above.
1421 */
1422 wmb();
1423 inet->sk_user_data = svsk;
1424
1425 /* Initialize the socket */
1426 if (sock->type == SOCK_DGRAM)
1427 svc_udp_init(svsk, serv);
1428 else
1429 svc_tcp_init(svsk, serv);
1430
1431 trace_svcsock_new(svsk, sock);
1432 return svsk;
1433}
1434
1435/**
1436 * svc_addsock - add a listener socket to an RPC service
1437 * @serv: pointer to RPC service to which to add a new listener
1438 * @net: caller's network namespace
1439 * @fd: file descriptor of the new listener
1440 * @name_return: pointer to buffer to fill in with name of listener
1441 * @len: size of the buffer
1442 * @cred: credential
1443 *
1444 * Fills in socket name and returns positive length of name if successful.
1445 * Name is terminated with '\n'. On error, returns a negative errno
1446 * value.
1447 */
1448int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1449 char *name_return, const size_t len, const struct cred *cred)
1450{
1451 int err = 0;
1452 struct socket *so = sockfd_lookup(fd, &err);
1453 struct svc_sock *svsk = NULL;
1454 struct sockaddr_storage addr;
1455 struct sockaddr *sin = (struct sockaddr *)&addr;
1456 int salen;
1457
1458 if (!so)
1459 return err;
1460 err = -EINVAL;
1461 if (sock_net(so->sk) != net)
1462 goto out;
1463 err = -EAFNOSUPPORT;
1464 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1465 goto out;
1466 err = -EPROTONOSUPPORT;
1467 if (so->sk->sk_protocol != IPPROTO_TCP &&
1468 so->sk->sk_protocol != IPPROTO_UDP)
1469 goto out;
1470 err = -EISCONN;
1471 if (so->state > SS_UNCONNECTED)
1472 goto out;
1473 err = -ENOENT;
1474 if (!try_module_get(THIS_MODULE))
1475 goto out;
1476 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1477 if (IS_ERR(svsk)) {
1478 module_put(THIS_MODULE);
1479 err = PTR_ERR(svsk);
1480 goto out;
1481 }
1482 salen = kernel_getsockname(svsk->sk_sock, sin);
1483 if (salen >= 0)
1484 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1485 svsk->sk_xprt.xpt_cred = get_cred(cred);
1486 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1487 return svc_one_sock_name(svsk, name_return, len);
1488out:
1489 sockfd_put(so);
1490 return err;
1491}
1492EXPORT_SYMBOL_GPL(svc_addsock);
1493
1494/*
1495 * Create socket for RPC service.
1496 */
1497static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1498 int protocol,
1499 struct net *net,
1500 struct sockaddr *sin, int len,
1501 int flags)
1502{
1503 struct svc_sock *svsk;
1504 struct socket *sock;
1505 int error;
1506 int type;
1507 struct sockaddr_storage addr;
1508 struct sockaddr *newsin = (struct sockaddr *)&addr;
1509 int newlen;
1510 int family;
1511
1512 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1513 printk(KERN_WARNING "svc: only UDP and TCP "
1514 "sockets supported\n");
1515 return ERR_PTR(-EINVAL);
1516 }
1517
1518 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1519 switch (sin->sa_family) {
1520 case AF_INET6:
1521 family = PF_INET6;
1522 break;
1523 case AF_INET:
1524 family = PF_INET;
1525 break;
1526 default:
1527 return ERR_PTR(-EINVAL);
1528 }
1529
1530 error = __sock_create(net, family, type, protocol, &sock, 1);
1531 if (error < 0)
1532 return ERR_PTR(error);
1533
1534 svc_reclassify_socket(sock);
1535
1536 /*
1537 * If this is an PF_INET6 listener, we want to avoid
1538 * getting requests from IPv4 remotes. Those should
1539 * be shunted to a PF_INET listener via rpcbind.
1540 */
1541 if (family == PF_INET6)
1542 ip6_sock_set_v6only(sock->sk);
1543 if (type == SOCK_STREAM)
1544 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1545 error = kernel_bind(sock, sin, len);
1546 if (error < 0)
1547 goto bummer;
1548
1549 error = kernel_getsockname(sock, newsin);
1550 if (error < 0)
1551 goto bummer;
1552 newlen = error;
1553
1554 if (protocol == IPPROTO_TCP) {
1555 if ((error = kernel_listen(sock, 64)) < 0)
1556 goto bummer;
1557 }
1558
1559 svsk = svc_setup_socket(serv, sock, flags);
1560 if (IS_ERR(svsk)) {
1561 error = PTR_ERR(svsk);
1562 goto bummer;
1563 }
1564 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1565 return (struct svc_xprt *)svsk;
1566bummer:
1567 sock_release(sock);
1568 return ERR_PTR(error);
1569}
1570
1571/*
1572 * Detach the svc_sock from the socket so that no
1573 * more callbacks occur.
1574 */
1575static void svc_sock_detach(struct svc_xprt *xprt)
1576{
1577 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1578 struct sock *sk = svsk->sk_sk;
1579
1580 /* put back the old socket callbacks */
1581 lock_sock(sk);
1582 sk->sk_state_change = svsk->sk_ostate;
1583 sk->sk_data_ready = svsk->sk_odata;
1584 sk->sk_write_space = svsk->sk_owspace;
1585 sk->sk_user_data = NULL;
1586 release_sock(sk);
1587}
1588
1589/*
1590 * Disconnect the socket, and reset the callbacks
1591 */
1592static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1593{
1594 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1595
1596 tls_handshake_close(svsk->sk_sock);
1597
1598 svc_sock_detach(xprt);
1599
1600 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1601 svc_tcp_clear_pages(svsk);
1602 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1603 }
1604}
1605
1606/*
1607 * Free the svc_sock's socket resources and the svc_sock itself.
1608 */
1609static void svc_sock_free(struct svc_xprt *xprt)
1610{
1611 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1612 struct page_frag_cache *pfc = &svsk->sk_frag_cache;
1613 struct socket *sock = svsk->sk_sock;
1614
1615 trace_svcsock_free(svsk, sock);
1616
1617 tls_handshake_cancel(sock->sk);
1618 if (sock->file)
1619 sockfd_put(sock);
1620 else
1621 sock_release(sock);
1622 if (pfc->va)
1623 __page_frag_cache_drain(virt_to_head_page(pfc->va),
1624 pfc->pagecnt_bias);
1625 kfree(svsk);
1626}