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 <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 *
1210 * Note that the send is non-blocking. The caller has incremented
1211 * the reference count on each page backing the RPC message, and
1212 * the network layer will "put" these pages when transmission is
1213 * complete.
1214 *
1215 * This is safe for our RPC services because the memory backing
1216 * the head and tail components is never kmalloc'd. These always
1217 * come from pages in the svc_rqst::rq_pages array.
1218 */
1219static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1220 rpc_fraghdr marker, unsigned int *sentp)
1221{
1222 struct msghdr msg = {
1223 .msg_flags = MSG_SPLICE_PAGES,
1224 };
1225 unsigned int count;
1226 void *buf;
1227 int ret;
1228
1229 *sentp = 0;
1230
1231 /* The stream record marker is copied into a temporary page
1232 * fragment buffer so that it can be included in rq_bvec.
1233 */
1234 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1235 GFP_KERNEL);
1236 if (!buf)
1237 return -ENOMEM;
1238 memcpy(buf, &marker, sizeof(marker));
1239 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1240
1241 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1242 ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1243
1244 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1245 1 + count, sizeof(marker) + rqstp->rq_res.len);
1246 ret = sock_sendmsg(svsk->sk_sock, &msg);
1247 if (ret < 0)
1248 return ret;
1249 *sentp += ret;
1250 return 0;
1251}
1252
1253/**
1254 * svc_tcp_sendto - Send out a reply on a TCP socket
1255 * @rqstp: completed svc_rqst
1256 *
1257 * xpt_mutex ensures @rqstp's whole message is written to the socket
1258 * without interruption.
1259 *
1260 * Returns the number of bytes sent, or a negative errno.
1261 */
1262static int svc_tcp_sendto(struct svc_rqst *rqstp)
1263{
1264 struct svc_xprt *xprt = rqstp->rq_xprt;
1265 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1266 struct xdr_buf *xdr = &rqstp->rq_res;
1267 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1268 (u32)xdr->len);
1269 unsigned int sent;
1270 int err;
1271
1272 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1273 rqstp->rq_xprt_ctxt = NULL;
1274
1275 mutex_lock(&xprt->xpt_mutex);
1276 if (svc_xprt_is_dead(xprt))
1277 goto out_notconn;
1278 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1279 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1280 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1281 goto out_close;
1282 mutex_unlock(&xprt->xpt_mutex);
1283 return sent;
1284
1285out_notconn:
1286 mutex_unlock(&xprt->xpt_mutex);
1287 return -ENOTCONN;
1288out_close:
1289 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1290 xprt->xpt_server->sv_name,
1291 (err < 0) ? "got error" : "sent",
1292 (err < 0) ? err : sent, xdr->len);
1293 svc_xprt_deferred_close(xprt);
1294 mutex_unlock(&xprt->xpt_mutex);
1295 return -EAGAIN;
1296}
1297
1298static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1299 struct net *net,
1300 struct sockaddr *sa, int salen,
1301 int flags)
1302{
1303 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1304}
1305
1306static const struct svc_xprt_ops svc_tcp_ops = {
1307 .xpo_create = svc_tcp_create,
1308 .xpo_recvfrom = svc_tcp_recvfrom,
1309 .xpo_sendto = svc_tcp_sendto,
1310 .xpo_result_payload = svc_sock_result_payload,
1311 .xpo_release_ctxt = svc_tcp_release_ctxt,
1312 .xpo_detach = svc_tcp_sock_detach,
1313 .xpo_free = svc_sock_free,
1314 .xpo_has_wspace = svc_tcp_has_wspace,
1315 .xpo_accept = svc_tcp_accept,
1316 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1317 .xpo_handshake = svc_tcp_handshake,
1318};
1319
1320static struct svc_xprt_class svc_tcp_class = {
1321 .xcl_name = "tcp",
1322 .xcl_owner = THIS_MODULE,
1323 .xcl_ops = &svc_tcp_ops,
1324 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1325 .xcl_ident = XPRT_TRANSPORT_TCP,
1326};
1327
1328void svc_init_xprt_sock(void)
1329{
1330 svc_reg_xprt_class(&svc_tcp_class);
1331 svc_reg_xprt_class(&svc_udp_class);
1332}
1333
1334void svc_cleanup_xprt_sock(void)
1335{
1336 svc_unreg_xprt_class(&svc_tcp_class);
1337 svc_unreg_xprt_class(&svc_udp_class);
1338}
1339
1340static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1341{
1342 struct sock *sk = svsk->sk_sk;
1343
1344 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1345 &svsk->sk_xprt, serv);
1346 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1347 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1348 if (sk->sk_state == TCP_LISTEN) {
1349 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1350 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1351 sk->sk_data_ready = svc_tcp_listen_data_ready;
1352 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1353 } else {
1354 sk->sk_state_change = svc_tcp_state_change;
1355 sk->sk_data_ready = svc_data_ready;
1356 sk->sk_write_space = svc_write_space;
1357
1358 svsk->sk_marker = xdr_zero;
1359 svsk->sk_tcplen = 0;
1360 svsk->sk_datalen = 0;
1361 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1362
1363 tcp_sock_set_nodelay(sk);
1364
1365 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1366 switch (sk->sk_state) {
1367 case TCP_SYN_RECV:
1368 case TCP_ESTABLISHED:
1369 break;
1370 default:
1371 svc_xprt_deferred_close(&svsk->sk_xprt);
1372 }
1373 }
1374}
1375
1376void svc_sock_update_bufs(struct svc_serv *serv)
1377{
1378 /*
1379 * The number of server threads has changed. Update
1380 * rcvbuf and sndbuf accordingly on all sockets
1381 */
1382 struct svc_sock *svsk;
1383
1384 spin_lock_bh(&serv->sv_lock);
1385 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1386 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1387 spin_unlock_bh(&serv->sv_lock);
1388}
1389EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1390
1391/*
1392 * Initialize socket for RPC use and create svc_sock struct
1393 */
1394static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1395 struct socket *sock,
1396 int flags)
1397{
1398 struct svc_sock *svsk;
1399 struct sock *inet;
1400 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1401
1402 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1403 if (!svsk)
1404 return ERR_PTR(-ENOMEM);
1405
1406 inet = sock->sk;
1407
1408 if (pmap_register) {
1409 int err;
1410
1411 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1412 inet->sk_protocol,
1413 ntohs(inet_sk(inet)->inet_sport));
1414 if (err < 0) {
1415 kfree(svsk);
1416 return ERR_PTR(err);
1417 }
1418 }
1419
1420 svsk->sk_sock = sock;
1421 svsk->sk_sk = inet;
1422 svsk->sk_ostate = inet->sk_state_change;
1423 svsk->sk_odata = inet->sk_data_ready;
1424 svsk->sk_owspace = inet->sk_write_space;
1425 /*
1426 * This barrier is necessary in order to prevent race condition
1427 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1428 * when calling callbacks above.
1429 */
1430 wmb();
1431 inet->sk_user_data = svsk;
1432
1433 /* Initialize the socket */
1434 if (sock->type == SOCK_DGRAM)
1435 svc_udp_init(svsk, serv);
1436 else
1437 svc_tcp_init(svsk, serv);
1438
1439 trace_svcsock_new(svsk, sock);
1440 return svsk;
1441}
1442
1443/**
1444 * svc_addsock - add a listener socket to an RPC service
1445 * @serv: pointer to RPC service to which to add a new listener
1446 * @net: caller's network namespace
1447 * @fd: file descriptor of the new listener
1448 * @name_return: pointer to buffer to fill in with name of listener
1449 * @len: size of the buffer
1450 * @cred: credential
1451 *
1452 * Fills in socket name and returns positive length of name if successful.
1453 * Name is terminated with '\n'. On error, returns a negative errno
1454 * value.
1455 */
1456int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1457 char *name_return, const size_t len, const struct cred *cred)
1458{
1459 int err = 0;
1460 struct socket *so = sockfd_lookup(fd, &err);
1461 struct svc_sock *svsk = NULL;
1462 struct sockaddr_storage addr;
1463 struct sockaddr *sin = (struct sockaddr *)&addr;
1464 int salen;
1465
1466 if (!so)
1467 return err;
1468 err = -EINVAL;
1469 if (sock_net(so->sk) != net)
1470 goto out;
1471 err = -EAFNOSUPPORT;
1472 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1473 goto out;
1474 err = -EPROTONOSUPPORT;
1475 if (so->sk->sk_protocol != IPPROTO_TCP &&
1476 so->sk->sk_protocol != IPPROTO_UDP)
1477 goto out;
1478 err = -EISCONN;
1479 if (so->state > SS_UNCONNECTED)
1480 goto out;
1481 err = -ENOENT;
1482 if (!try_module_get(THIS_MODULE))
1483 goto out;
1484 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1485 if (IS_ERR(svsk)) {
1486 module_put(THIS_MODULE);
1487 err = PTR_ERR(svsk);
1488 goto out;
1489 }
1490 salen = kernel_getsockname(svsk->sk_sock, sin);
1491 if (salen >= 0)
1492 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1493 svsk->sk_xprt.xpt_cred = get_cred(cred);
1494 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1495 return svc_one_sock_name(svsk, name_return, len);
1496out:
1497 sockfd_put(so);
1498 return err;
1499}
1500EXPORT_SYMBOL_GPL(svc_addsock);
1501
1502/*
1503 * Create socket for RPC service.
1504 */
1505static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1506 int protocol,
1507 struct net *net,
1508 struct sockaddr *sin, int len,
1509 int flags)
1510{
1511 struct svc_sock *svsk;
1512 struct socket *sock;
1513 int error;
1514 int type;
1515 struct sockaddr_storage addr;
1516 struct sockaddr *newsin = (struct sockaddr *)&addr;
1517 int newlen;
1518 int family;
1519
1520 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1521 printk(KERN_WARNING "svc: only UDP and TCP "
1522 "sockets supported\n");
1523 return ERR_PTR(-EINVAL);
1524 }
1525
1526 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1527 switch (sin->sa_family) {
1528 case AF_INET6:
1529 family = PF_INET6;
1530 break;
1531 case AF_INET:
1532 family = PF_INET;
1533 break;
1534 default:
1535 return ERR_PTR(-EINVAL);
1536 }
1537
1538 error = __sock_create(net, family, type, protocol, &sock, 1);
1539 if (error < 0)
1540 return ERR_PTR(error);
1541
1542 svc_reclassify_socket(sock);
1543
1544 /*
1545 * If this is an PF_INET6 listener, we want to avoid
1546 * getting requests from IPv4 remotes. Those should
1547 * be shunted to a PF_INET listener via rpcbind.
1548 */
1549 if (family == PF_INET6)
1550 ip6_sock_set_v6only(sock->sk);
1551 if (type == SOCK_STREAM)
1552 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1553 error = kernel_bind(sock, sin, len);
1554 if (error < 0)
1555 goto bummer;
1556
1557 error = kernel_getsockname(sock, newsin);
1558 if (error < 0)
1559 goto bummer;
1560 newlen = error;
1561
1562 if (protocol == IPPROTO_TCP) {
1563 if ((error = kernel_listen(sock, 64)) < 0)
1564 goto bummer;
1565 }
1566
1567 svsk = svc_setup_socket(serv, sock, flags);
1568 if (IS_ERR(svsk)) {
1569 error = PTR_ERR(svsk);
1570 goto bummer;
1571 }
1572 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1573 return (struct svc_xprt *)svsk;
1574bummer:
1575 sock_release(sock);
1576 return ERR_PTR(error);
1577}
1578
1579/*
1580 * Detach the svc_sock from the socket so that no
1581 * more callbacks occur.
1582 */
1583static void svc_sock_detach(struct svc_xprt *xprt)
1584{
1585 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1586 struct sock *sk = svsk->sk_sk;
1587
1588 /* put back the old socket callbacks */
1589 lock_sock(sk);
1590 sk->sk_state_change = svsk->sk_ostate;
1591 sk->sk_data_ready = svsk->sk_odata;
1592 sk->sk_write_space = svsk->sk_owspace;
1593 sk->sk_user_data = NULL;
1594 release_sock(sk);
1595}
1596
1597/*
1598 * Disconnect the socket, and reset the callbacks
1599 */
1600static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1601{
1602 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1603
1604 tls_handshake_close(svsk->sk_sock);
1605
1606 svc_sock_detach(xprt);
1607
1608 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1609 svc_tcp_clear_pages(svsk);
1610 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1611 }
1612}
1613
1614/*
1615 * Free the svc_sock's socket resources and the svc_sock itself.
1616 */
1617static void svc_sock_free(struct svc_xprt *xprt)
1618{
1619 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1620 struct page_frag_cache *pfc = &svsk->sk_frag_cache;
1621 struct socket *sock = svsk->sk_sock;
1622
1623 trace_svcsock_free(svsk, sock);
1624
1625 tls_handshake_cancel(sock->sk);
1626 if (sock->file)
1627 sockfd_put(sock);
1628 else
1629 sock_release(sock);
1630 if (pfc->va)
1631 __page_frag_cache_drain(virt_to_head_page(pfc->va),
1632 pfc->pagecnt_bias);
1633 kfree(svsk);
1634}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svcsock.c
4 *
5 * These are the RPC server socket internals.
6 *
7 * The server scheduling algorithm does not always distribute the load
8 * evenly when servicing a single client. May need to modify the
9 * svc_xprt_enqueue procedure...
10 *
11 * TCP support is largely untested and may be a little slow. The problem
12 * is that we currently do two separate recvfrom's, one for the 4-byte
13 * record length, and the second for the actual record. This could possibly
14 * be improved by always reading a minimum size of around 100 bytes and
15 * tucking any superfluous bytes away in a temporary store. Still, that
16 * leaves write requests out in the rain. An alternative may be to peek at
17 * the first skb in the queue, and if it matches the next TCP sequence
18 * number, to extract the record marker. Yuck.
19 *
20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
21 */
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/module.h>
26#include <linux/errno.h>
27#include <linux/fcntl.h>
28#include <linux/net.h>
29#include <linux/in.h>
30#include <linux/inet.h>
31#include <linux/udp.h>
32#include <linux/tcp.h>
33#include <linux/unistd.h>
34#include <linux/slab.h>
35#include <linux/netdevice.h>
36#include <linux/skbuff.h>
37#include <linux/file.h>
38#include <linux/freezer.h>
39#include <net/sock.h>
40#include <net/checksum.h>
41#include <net/ip.h>
42#include <net/ipv6.h>
43#include <net/udp.h>
44#include <net/tcp.h>
45#include <net/tcp_states.h>
46#include <linux/uaccess.h>
47#include <asm/ioctls.h>
48#include <trace/events/skb.h>
49
50#include <linux/sunrpc/types.h>
51#include <linux/sunrpc/clnt.h>
52#include <linux/sunrpc/xdr.h>
53#include <linux/sunrpc/msg_prot.h>
54#include <linux/sunrpc/svcsock.h>
55#include <linux/sunrpc/stats.h>
56#include <linux/sunrpc/xprt.h>
57
58#include "sunrpc.h"
59
60#define RPCDBG_FACILITY RPCDBG_SVCXPRT
61
62
63static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
64 int flags);
65static int svc_udp_recvfrom(struct svc_rqst *);
66static int svc_udp_sendto(struct svc_rqst *);
67static void svc_sock_detach(struct svc_xprt *);
68static void svc_tcp_sock_detach(struct svc_xprt *);
69static void svc_sock_free(struct svc_xprt *);
70
71static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
72 struct net *, struct sockaddr *,
73 int, int);
74#ifdef CONFIG_DEBUG_LOCK_ALLOC
75static struct lock_class_key svc_key[2];
76static struct lock_class_key svc_slock_key[2];
77
78static void svc_reclassify_socket(struct socket *sock)
79{
80 struct sock *sk = sock->sk;
81
82 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
83 return;
84
85 switch (sk->sk_family) {
86 case AF_INET:
87 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
88 &svc_slock_key[0],
89 "sk_xprt.xpt_lock-AF_INET-NFSD",
90 &svc_key[0]);
91 break;
92
93 case AF_INET6:
94 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
95 &svc_slock_key[1],
96 "sk_xprt.xpt_lock-AF_INET6-NFSD",
97 &svc_key[1]);
98 break;
99
100 default:
101 BUG();
102 }
103}
104#else
105static void svc_reclassify_socket(struct socket *sock)
106{
107}
108#endif
109
110/*
111 * Release an skbuff after use
112 */
113static void svc_release_skb(struct svc_rqst *rqstp)
114{
115 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
116
117 if (skb) {
118 struct svc_sock *svsk =
119 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
120 rqstp->rq_xprt_ctxt = NULL;
121
122 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
123 skb_free_datagram_locked(svsk->sk_sk, skb);
124 }
125}
126
127static void svc_release_udp_skb(struct svc_rqst *rqstp)
128{
129 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
130
131 if (skb) {
132 rqstp->rq_xprt_ctxt = NULL;
133
134 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
135 consume_skb(skb);
136 }
137}
138
139union svc_pktinfo_u {
140 struct in_pktinfo pkti;
141 struct in6_pktinfo pkti6;
142};
143#define SVC_PKTINFO_SPACE \
144 CMSG_SPACE(sizeof(union svc_pktinfo_u))
145
146static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
147{
148 struct svc_sock *svsk =
149 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
150 switch (svsk->sk_sk->sk_family) {
151 case AF_INET: {
152 struct in_pktinfo *pki = CMSG_DATA(cmh);
153
154 cmh->cmsg_level = SOL_IP;
155 cmh->cmsg_type = IP_PKTINFO;
156 pki->ipi_ifindex = 0;
157 pki->ipi_spec_dst.s_addr =
158 svc_daddr_in(rqstp)->sin_addr.s_addr;
159 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
160 }
161 break;
162
163 case AF_INET6: {
164 struct in6_pktinfo *pki = CMSG_DATA(cmh);
165 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
166
167 cmh->cmsg_level = SOL_IPV6;
168 cmh->cmsg_type = IPV6_PKTINFO;
169 pki->ipi6_ifindex = daddr->sin6_scope_id;
170 pki->ipi6_addr = daddr->sin6_addr;
171 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
172 }
173 break;
174 }
175}
176
177/*
178 * send routine intended to be shared by the fore- and back-channel
179 */
180int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
181 struct page *headpage, unsigned long headoffset,
182 struct page *tailpage, unsigned long tailoffset)
183{
184 int result;
185 int size;
186 struct page **ppage = xdr->pages;
187 size_t base = xdr->page_base;
188 unsigned int pglen = xdr->page_len;
189 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
190 int slen;
191 int len = 0;
192
193 slen = xdr->len;
194
195 /* send head */
196 if (slen == xdr->head[0].iov_len)
197 flags = 0;
198 len = kernel_sendpage(sock, headpage, headoffset,
199 xdr->head[0].iov_len, flags);
200 if (len != xdr->head[0].iov_len)
201 goto out;
202 slen -= xdr->head[0].iov_len;
203 if (slen == 0)
204 goto out;
205
206 /* send page data */
207 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
208 while (pglen > 0) {
209 if (slen == size)
210 flags = 0;
211 result = kernel_sendpage(sock, *ppage, base, size, flags);
212 if (result > 0)
213 len += result;
214 if (result != size)
215 goto out;
216 slen -= size;
217 pglen -= size;
218 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
219 base = 0;
220 ppage++;
221 }
222
223 /* send tail */
224 if (xdr->tail[0].iov_len) {
225 result = kernel_sendpage(sock, tailpage, tailoffset,
226 xdr->tail[0].iov_len, 0);
227 if (result > 0)
228 len += result;
229 }
230
231out:
232 return len;
233}
234
235
236/*
237 * Generic sendto routine
238 */
239static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
240{
241 struct svc_sock *svsk =
242 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
243 struct socket *sock = svsk->sk_sock;
244 union {
245 struct cmsghdr hdr;
246 long all[SVC_PKTINFO_SPACE / sizeof(long)];
247 } buffer;
248 struct cmsghdr *cmh = &buffer.hdr;
249 int len = 0;
250 unsigned long tailoff;
251 unsigned long headoff;
252 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
253
254 if (rqstp->rq_prot == IPPROTO_UDP) {
255 struct msghdr msg = {
256 .msg_name = &rqstp->rq_addr,
257 .msg_namelen = rqstp->rq_addrlen,
258 .msg_control = cmh,
259 .msg_controllen = sizeof(buffer),
260 .msg_flags = MSG_MORE,
261 };
262
263 svc_set_cmsg_data(rqstp, cmh);
264
265 if (sock_sendmsg(sock, &msg) < 0)
266 goto out;
267 }
268
269 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
270 headoff = 0;
271 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
272 rqstp->rq_respages[0], tailoff);
273
274out:
275 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
276 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
277 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
278
279 return len;
280}
281
282/*
283 * Report socket names for nfsdfs
284 */
285static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
286{
287 const struct sock *sk = svsk->sk_sk;
288 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
289 "udp" : "tcp";
290 int len;
291
292 switch (sk->sk_family) {
293 case PF_INET:
294 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
295 proto_name,
296 &inet_sk(sk)->inet_rcv_saddr,
297 inet_sk(sk)->inet_num);
298 break;
299#if IS_ENABLED(CONFIG_IPV6)
300 case PF_INET6:
301 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
302 proto_name,
303 &sk->sk_v6_rcv_saddr,
304 inet_sk(sk)->inet_num);
305 break;
306#endif
307 default:
308 len = snprintf(buf, remaining, "*unknown-%d*\n",
309 sk->sk_family);
310 }
311
312 if (len >= remaining) {
313 *buf = '\0';
314 return -ENAMETOOLONG;
315 }
316 return len;
317}
318
319/*
320 * Generic recvfrom routine.
321 */
322static ssize_t svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov,
323 unsigned int nr, size_t buflen, unsigned int base)
324{
325 struct svc_sock *svsk =
326 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
327 struct msghdr msg = { NULL };
328 ssize_t len;
329
330 rqstp->rq_xprt_hlen = 0;
331
332 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
333 iov_iter_kvec(&msg.msg_iter, READ, iov, nr, buflen);
334 if (base != 0) {
335 iov_iter_advance(&msg.msg_iter, base);
336 buflen -= base;
337 }
338 len = sock_recvmsg(svsk->sk_sock, &msg, MSG_DONTWAIT);
339 /* If we read a full record, then assume there may be more
340 * data to read (stream based sockets only!)
341 */
342 if (len == buflen)
343 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
344
345 dprintk("svc: socket %p recvfrom(%p, %zu) = %zd\n",
346 svsk, iov[0].iov_base, iov[0].iov_len, len);
347 return len;
348}
349
350/*
351 * Set socket snd and rcv buffer lengths
352 */
353static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
354{
355 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
356 struct socket *sock = svsk->sk_sock;
357
358 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
359
360 lock_sock(sock->sk);
361 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
362 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
363 sock->sk->sk_write_space(sock->sk);
364 release_sock(sock->sk);
365}
366
367static void svc_sock_secure_port(struct svc_rqst *rqstp)
368{
369 if (svc_port_is_privileged(svc_addr(rqstp)))
370 set_bit(RQ_SECURE, &rqstp->rq_flags);
371 else
372 clear_bit(RQ_SECURE, &rqstp->rq_flags);
373}
374
375/*
376 * INET callback when data has been received on the socket.
377 */
378static void svc_data_ready(struct sock *sk)
379{
380 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
381
382 if (svsk) {
383 dprintk("svc: socket %p(inet %p), busy=%d\n",
384 svsk, sk,
385 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
386
387 /* Refer to svc_setup_socket() for details. */
388 rmb();
389 svsk->sk_odata(sk);
390 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
391 svc_xprt_enqueue(&svsk->sk_xprt);
392 }
393}
394
395/*
396 * INET callback when space is newly available on the socket.
397 */
398static void svc_write_space(struct sock *sk)
399{
400 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
401
402 if (svsk) {
403 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
404 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
405
406 /* Refer to svc_setup_socket() for details. */
407 rmb();
408 svsk->sk_owspace(sk);
409 svc_xprt_enqueue(&svsk->sk_xprt);
410 }
411}
412
413static int svc_tcp_has_wspace(struct svc_xprt *xprt)
414{
415 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
416
417 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
418 return 1;
419 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
420}
421
422static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
423{
424 struct svc_sock *svsk;
425 struct socket *sock;
426 struct linger no_linger = {
427 .l_onoff = 1,
428 .l_linger = 0,
429 };
430
431 svsk = container_of(xprt, struct svc_sock, sk_xprt);
432 sock = svsk->sk_sock;
433 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
434 (char *)&no_linger, sizeof(no_linger));
435}
436
437/*
438 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
439 */
440static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
441 struct cmsghdr *cmh)
442{
443 struct in_pktinfo *pki = CMSG_DATA(cmh);
444 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
445
446 if (cmh->cmsg_type != IP_PKTINFO)
447 return 0;
448
449 daddr->sin_family = AF_INET;
450 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
451 return 1;
452}
453
454/*
455 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
456 */
457static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
458 struct cmsghdr *cmh)
459{
460 struct in6_pktinfo *pki = CMSG_DATA(cmh);
461 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
462
463 if (cmh->cmsg_type != IPV6_PKTINFO)
464 return 0;
465
466 daddr->sin6_family = AF_INET6;
467 daddr->sin6_addr = pki->ipi6_addr;
468 daddr->sin6_scope_id = pki->ipi6_ifindex;
469 return 1;
470}
471
472/*
473 * Copy the UDP datagram's destination address to the rqstp structure.
474 * The 'destination' address in this case is the address to which the
475 * peer sent the datagram, i.e. our local address. For multihomed
476 * hosts, this can change from msg to msg. Note that only the IP
477 * address changes, the port number should remain the same.
478 */
479static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
480 struct cmsghdr *cmh)
481{
482 switch (cmh->cmsg_level) {
483 case SOL_IP:
484 return svc_udp_get_dest_address4(rqstp, cmh);
485 case SOL_IPV6:
486 return svc_udp_get_dest_address6(rqstp, cmh);
487 }
488
489 return 0;
490}
491
492/*
493 * Receive a datagram from a UDP socket.
494 */
495static int svc_udp_recvfrom(struct svc_rqst *rqstp)
496{
497 struct svc_sock *svsk =
498 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
499 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
500 struct sk_buff *skb;
501 union {
502 struct cmsghdr hdr;
503 long all[SVC_PKTINFO_SPACE / sizeof(long)];
504 } buffer;
505 struct cmsghdr *cmh = &buffer.hdr;
506 struct msghdr msg = {
507 .msg_name = svc_addr(rqstp),
508 .msg_control = cmh,
509 .msg_controllen = sizeof(buffer),
510 .msg_flags = MSG_DONTWAIT,
511 };
512 size_t len;
513 int err;
514
515 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
516 /* udp sockets need large rcvbuf as all pending
517 * requests are still in that buffer. sndbuf must
518 * also be large enough that there is enough space
519 * for one reply per thread. We count all threads
520 * rather than threads in a particular pool, which
521 * provides an upper bound on the number of threads
522 * which will access the socket.
523 */
524 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
525
526 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
527 skb = NULL;
528 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
529 0, 0, MSG_PEEK | MSG_DONTWAIT);
530 if (err >= 0)
531 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
532
533 if (skb == NULL) {
534 if (err != -EAGAIN) {
535 /* possibly an icmp error */
536 dprintk("svc: recvfrom returned error %d\n", -err);
537 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
538 }
539 return 0;
540 }
541 len = svc_addr_len(svc_addr(rqstp));
542 rqstp->rq_addrlen = len;
543 if (skb->tstamp == 0) {
544 skb->tstamp = ktime_get_real();
545 /* Don't enable netstamp, sunrpc doesn't
546 need that much accuracy */
547 }
548 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
549 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
550
551 len = skb->len;
552 rqstp->rq_arg.len = len;
553
554 rqstp->rq_prot = IPPROTO_UDP;
555
556 if (!svc_udp_get_dest_address(rqstp, cmh)) {
557 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
558 cmh->cmsg_level, cmh->cmsg_type);
559 goto out_free;
560 }
561 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
562
563 if (skb_is_nonlinear(skb)) {
564 /* we have to copy */
565 local_bh_disable();
566 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
567 local_bh_enable();
568 /* checksum error */
569 goto out_free;
570 }
571 local_bh_enable();
572 consume_skb(skb);
573 } else {
574 /* we can use it in-place */
575 rqstp->rq_arg.head[0].iov_base = skb->data;
576 rqstp->rq_arg.head[0].iov_len = len;
577 if (skb_checksum_complete(skb))
578 goto out_free;
579 rqstp->rq_xprt_ctxt = skb;
580 }
581
582 rqstp->rq_arg.page_base = 0;
583 if (len <= rqstp->rq_arg.head[0].iov_len) {
584 rqstp->rq_arg.head[0].iov_len = len;
585 rqstp->rq_arg.page_len = 0;
586 rqstp->rq_respages = rqstp->rq_pages+1;
587 } else {
588 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
589 rqstp->rq_respages = rqstp->rq_pages + 1 +
590 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
591 }
592 rqstp->rq_next_page = rqstp->rq_respages+1;
593
594 if (serv->sv_stats)
595 serv->sv_stats->netudpcnt++;
596
597 return len;
598out_free:
599 kfree_skb(skb);
600 return 0;
601}
602
603static int
604svc_udp_sendto(struct svc_rqst *rqstp)
605{
606 int error;
607
608 error = svc_sendto(rqstp, &rqstp->rq_res);
609 if (error == -ECONNREFUSED)
610 /* ICMP error on earlier request. */
611 error = svc_sendto(rqstp, &rqstp->rq_res);
612
613 return error;
614}
615
616static int svc_udp_has_wspace(struct svc_xprt *xprt)
617{
618 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
619 struct svc_serv *serv = xprt->xpt_server;
620 unsigned long required;
621
622 /*
623 * Set the SOCK_NOSPACE flag before checking the available
624 * sock space.
625 */
626 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
627 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
628 if (required*2 > sock_wspace(svsk->sk_sk))
629 return 0;
630 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
631 return 1;
632}
633
634static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
635{
636 BUG();
637 return NULL;
638}
639
640static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
641{
642}
643
644static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
645 struct net *net,
646 struct sockaddr *sa, int salen,
647 int flags)
648{
649 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
650}
651
652static const struct svc_xprt_ops svc_udp_ops = {
653 .xpo_create = svc_udp_create,
654 .xpo_recvfrom = svc_udp_recvfrom,
655 .xpo_sendto = svc_udp_sendto,
656 .xpo_release_rqst = svc_release_udp_skb,
657 .xpo_detach = svc_sock_detach,
658 .xpo_free = svc_sock_free,
659 .xpo_has_wspace = svc_udp_has_wspace,
660 .xpo_accept = svc_udp_accept,
661 .xpo_secure_port = svc_sock_secure_port,
662 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
663};
664
665static struct svc_xprt_class svc_udp_class = {
666 .xcl_name = "udp",
667 .xcl_owner = THIS_MODULE,
668 .xcl_ops = &svc_udp_ops,
669 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
670 .xcl_ident = XPRT_TRANSPORT_UDP,
671};
672
673static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
674{
675 int err, level, optname, one = 1;
676
677 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
678 &svsk->sk_xprt, serv);
679 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
680 svsk->sk_sk->sk_data_ready = svc_data_ready;
681 svsk->sk_sk->sk_write_space = svc_write_space;
682
683 /* initialise setting must have enough space to
684 * receive and respond to one request.
685 * svc_udp_recvfrom will re-adjust if necessary
686 */
687 svc_sock_setbufsize(svsk, 3);
688
689 /* data might have come in before data_ready set up */
690 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
691 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
692
693 /* make sure we get destination address info */
694 switch (svsk->sk_sk->sk_family) {
695 case AF_INET:
696 level = SOL_IP;
697 optname = IP_PKTINFO;
698 break;
699 case AF_INET6:
700 level = SOL_IPV6;
701 optname = IPV6_RECVPKTINFO;
702 break;
703 default:
704 BUG();
705 }
706 err = kernel_setsockopt(svsk->sk_sock, level, optname,
707 (char *)&one, sizeof(one));
708 dprintk("svc: kernel_setsockopt returned %d\n", err);
709}
710
711/*
712 * A data_ready event on a listening socket means there's a connection
713 * pending. Do not use state_change as a substitute for it.
714 */
715static void svc_tcp_listen_data_ready(struct sock *sk)
716{
717 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
718
719 dprintk("svc: socket %p TCP (listen) state change %d\n",
720 sk, sk->sk_state);
721
722 if (svsk) {
723 /* Refer to svc_setup_socket() for details. */
724 rmb();
725 svsk->sk_odata(sk);
726 }
727
728 /*
729 * This callback may called twice when a new connection
730 * is established as a child socket inherits everything
731 * from a parent LISTEN socket.
732 * 1) data_ready method of the parent socket will be called
733 * when one of child sockets become ESTABLISHED.
734 * 2) data_ready method of the child socket may be called
735 * when it receives data before the socket is accepted.
736 * In case of 2, we should ignore it silently.
737 */
738 if (sk->sk_state == TCP_LISTEN) {
739 if (svsk) {
740 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
741 svc_xprt_enqueue(&svsk->sk_xprt);
742 } else
743 printk("svc: socket %p: no user data\n", sk);
744 }
745}
746
747/*
748 * A state change on a connected socket means it's dying or dead.
749 */
750static void svc_tcp_state_change(struct sock *sk)
751{
752 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
753
754 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
755 sk, sk->sk_state, sk->sk_user_data);
756
757 if (!svsk)
758 printk("svc: socket %p: no user data\n", sk);
759 else {
760 /* Refer to svc_setup_socket() for details. */
761 rmb();
762 svsk->sk_ostate(sk);
763 if (sk->sk_state != TCP_ESTABLISHED) {
764 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
765 svc_xprt_enqueue(&svsk->sk_xprt);
766 }
767 }
768}
769
770/*
771 * Accept a TCP connection
772 */
773static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
774{
775 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
776 struct sockaddr_storage addr;
777 struct sockaddr *sin = (struct sockaddr *) &addr;
778 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
779 struct socket *sock = svsk->sk_sock;
780 struct socket *newsock;
781 struct svc_sock *newsvsk;
782 int err, slen;
783 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
784
785 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
786 if (!sock)
787 return NULL;
788
789 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
790 err = kernel_accept(sock, &newsock, O_NONBLOCK);
791 if (err < 0) {
792 if (err == -ENOMEM)
793 printk(KERN_WARNING "%s: no more sockets!\n",
794 serv->sv_name);
795 else if (err != -EAGAIN)
796 net_warn_ratelimited("%s: accept failed (err %d)!\n",
797 serv->sv_name, -err);
798 return NULL;
799 }
800 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
801
802 err = kernel_getpeername(newsock, sin);
803 if (err < 0) {
804 net_warn_ratelimited("%s: peername failed (err %d)!\n",
805 serv->sv_name, -err);
806 goto failed; /* aborted connection or whatever */
807 }
808 slen = err;
809
810 /* Ideally, we would want to reject connections from unauthorized
811 * hosts here, but when we get encryption, the IP of the host won't
812 * tell us anything. For now just warn about unpriv connections.
813 */
814 if (!svc_port_is_privileged(sin)) {
815 dprintk("%s: connect from unprivileged port: %s\n",
816 serv->sv_name,
817 __svc_print_addr(sin, buf, sizeof(buf)));
818 }
819 dprintk("%s: connect from %s\n", serv->sv_name,
820 __svc_print_addr(sin, buf, sizeof(buf)));
821
822 /* Reset the inherited callbacks before calling svc_setup_socket */
823 newsock->sk->sk_state_change = svsk->sk_ostate;
824 newsock->sk->sk_data_ready = svsk->sk_odata;
825 newsock->sk->sk_write_space = svsk->sk_owspace;
826
827 /* make sure that a write doesn't block forever when
828 * low on memory
829 */
830 newsock->sk->sk_sndtimeo = HZ*30;
831
832 newsvsk = svc_setup_socket(serv, newsock,
833 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
834 if (IS_ERR(newsvsk))
835 goto failed;
836 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
837 err = kernel_getsockname(newsock, sin);
838 slen = err;
839 if (unlikely(err < 0)) {
840 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
841 slen = offsetof(struct sockaddr, sa_data);
842 }
843 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
844
845 if (sock_is_loopback(newsock->sk))
846 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
847 else
848 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
849 if (serv->sv_stats)
850 serv->sv_stats->nettcpconn++;
851
852 return &newsvsk->sk_xprt;
853
854failed:
855 sock_release(newsock);
856 return NULL;
857}
858
859static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
860{
861 unsigned int i, len, npages;
862
863 if (svsk->sk_datalen == 0)
864 return 0;
865 len = svsk->sk_datalen;
866 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
867 for (i = 0; i < npages; i++) {
868 if (rqstp->rq_pages[i] != NULL)
869 put_page(rqstp->rq_pages[i]);
870 BUG_ON(svsk->sk_pages[i] == NULL);
871 rqstp->rq_pages[i] = svsk->sk_pages[i];
872 svsk->sk_pages[i] = NULL;
873 }
874 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
875 return len;
876}
877
878static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
879{
880 unsigned int i, len, npages;
881
882 if (svsk->sk_datalen == 0)
883 return;
884 len = svsk->sk_datalen;
885 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
886 for (i = 0; i < npages; i++) {
887 svsk->sk_pages[i] = rqstp->rq_pages[i];
888 rqstp->rq_pages[i] = NULL;
889 }
890}
891
892static void svc_tcp_clear_pages(struct svc_sock *svsk)
893{
894 unsigned int i, len, npages;
895
896 if (svsk->sk_datalen == 0)
897 goto out;
898 len = svsk->sk_datalen;
899 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
900 for (i = 0; i < npages; i++) {
901 if (svsk->sk_pages[i] == NULL) {
902 WARN_ON_ONCE(1);
903 continue;
904 }
905 put_page(svsk->sk_pages[i]);
906 svsk->sk_pages[i] = NULL;
907 }
908out:
909 svsk->sk_tcplen = 0;
910 svsk->sk_datalen = 0;
911}
912
913/*
914 * Receive fragment record header.
915 * If we haven't gotten the record length yet, get the next four bytes.
916 */
917static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
918{
919 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
920 unsigned int want;
921 int len;
922
923 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
924 struct kvec iov;
925
926 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
927 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
928 iov.iov_len = want;
929 len = svc_recvfrom(rqstp, &iov, 1, want, 0);
930 if (len < 0)
931 goto error;
932 svsk->sk_tcplen += len;
933
934 if (len < want) {
935 dprintk("svc: short recvfrom while reading record "
936 "length (%d of %d)\n", len, want);
937 return -EAGAIN;
938 }
939
940 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
941 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
942 serv->sv_max_mesg) {
943 net_notice_ratelimited("RPC: fragment too large: %d\n",
944 svc_sock_reclen(svsk));
945 goto err_delete;
946 }
947 }
948
949 return svc_sock_reclen(svsk);
950error:
951 dprintk("RPC: TCP recv_record got %d\n", len);
952 return len;
953err_delete:
954 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
955 return -EAGAIN;
956}
957
958static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
959{
960 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
961 struct rpc_rqst *req = NULL;
962 struct kvec *src, *dst;
963 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
964 __be32 xid;
965 __be32 calldir;
966
967 xid = *p++;
968 calldir = *p;
969
970 if (!bc_xprt)
971 return -EAGAIN;
972 spin_lock(&bc_xprt->queue_lock);
973 req = xprt_lookup_rqst(bc_xprt, xid);
974 if (!req)
975 goto unlock_notfound;
976
977 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
978 /*
979 * XXX!: cheating for now! Only copying HEAD.
980 * But we know this is good enough for now (in fact, for any
981 * callback reply in the forseeable future).
982 */
983 dst = &req->rq_private_buf.head[0];
984 src = &rqstp->rq_arg.head[0];
985 if (dst->iov_len < src->iov_len)
986 goto unlock_eagain; /* whatever; just giving up. */
987 memcpy(dst->iov_base, src->iov_base, src->iov_len);
988 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
989 rqstp->rq_arg.len = 0;
990 spin_unlock(&bc_xprt->queue_lock);
991 return 0;
992unlock_notfound:
993 printk(KERN_NOTICE
994 "%s: Got unrecognized reply: "
995 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
996 __func__, ntohl(calldir),
997 bc_xprt, ntohl(xid));
998unlock_eagain:
999 spin_unlock(&bc_xprt->queue_lock);
1000 return -EAGAIN;
1001}
1002
1003static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1004{
1005 int i = 0;
1006 int t = 0;
1007
1008 while (t < len) {
1009 vec[i].iov_base = page_address(pages[i]);
1010 vec[i].iov_len = PAGE_SIZE;
1011 i++;
1012 t += PAGE_SIZE;
1013 }
1014 return i;
1015}
1016
1017static void svc_tcp_fragment_received(struct svc_sock *svsk)
1018{
1019 /* If we have more data, signal svc_xprt_enqueue() to try again */
1020 dprintk("svc: TCP %s record (%d bytes)\n",
1021 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1022 svc_sock_reclen(svsk));
1023 svsk->sk_tcplen = 0;
1024 svsk->sk_reclen = 0;
1025}
1026
1027/*
1028 * Receive data from a TCP socket.
1029 */
1030static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1031{
1032 struct svc_sock *svsk =
1033 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1034 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1035 int len;
1036 struct kvec *vec;
1037 unsigned int want, base;
1038 __be32 *p;
1039 __be32 calldir;
1040 int pnum;
1041
1042 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1043 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1044 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1045 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1046
1047 len = svc_tcp_recv_record(svsk, rqstp);
1048 if (len < 0)
1049 goto error;
1050
1051 base = svc_tcp_restore_pages(svsk, rqstp);
1052 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1053
1054 vec = rqstp->rq_vec;
1055
1056 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0], base + want);
1057
1058 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1059 rqstp->rq_next_page = rqstp->rq_respages + 1;
1060
1061 /* Now receive data */
1062 len = svc_recvfrom(rqstp, vec, pnum, base + want, base);
1063 if (len >= 0) {
1064 svsk->sk_tcplen += len;
1065 svsk->sk_datalen += len;
1066 }
1067 if (len != want || !svc_sock_final_rec(svsk)) {
1068 svc_tcp_save_pages(svsk, rqstp);
1069 if (len < 0 && len != -EAGAIN)
1070 goto err_delete;
1071 if (len == want)
1072 svc_tcp_fragment_received(svsk);
1073 else
1074 dprintk("svc: incomplete TCP record (%d of %d)\n",
1075 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1076 svc_sock_reclen(svsk));
1077 goto err_noclose;
1078 }
1079
1080 if (svsk->sk_datalen < 8) {
1081 svsk->sk_datalen = 0;
1082 goto err_delete; /* client is nuts. */
1083 }
1084
1085 rqstp->rq_arg.len = svsk->sk_datalen;
1086 rqstp->rq_arg.page_base = 0;
1087 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1088 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1089 rqstp->rq_arg.page_len = 0;
1090 } else
1091 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1092
1093 rqstp->rq_xprt_ctxt = NULL;
1094 rqstp->rq_prot = IPPROTO_TCP;
1095 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1096 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1097 else
1098 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1099
1100 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1101 calldir = p[1];
1102 if (calldir)
1103 len = receive_cb_reply(svsk, rqstp);
1104
1105 /* Reset TCP read info */
1106 svsk->sk_datalen = 0;
1107 svc_tcp_fragment_received(svsk);
1108
1109 if (len < 0)
1110 goto error;
1111
1112 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1113 if (serv->sv_stats)
1114 serv->sv_stats->nettcpcnt++;
1115
1116 return rqstp->rq_arg.len;
1117
1118error:
1119 if (len != -EAGAIN)
1120 goto err_delete;
1121 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1122 return 0;
1123err_delete:
1124 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1125 svsk->sk_xprt.xpt_server->sv_name, -len);
1126 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1127err_noclose:
1128 return 0; /* record not complete */
1129}
1130
1131/*
1132 * Send out data on TCP socket.
1133 */
1134static int svc_tcp_sendto(struct svc_rqst *rqstp)
1135{
1136 struct xdr_buf *xbufp = &rqstp->rq_res;
1137 int sent;
1138 __be32 reclen;
1139
1140 /* Set up the first element of the reply kvec.
1141 * Any other kvecs that may be in use have been taken
1142 * care of by the server implementation itself.
1143 */
1144 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1145 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1146
1147 sent = svc_sendto(rqstp, &rqstp->rq_res);
1148 if (sent != xbufp->len) {
1149 printk(KERN_NOTICE
1150 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1151 "- shutting down socket\n",
1152 rqstp->rq_xprt->xpt_server->sv_name,
1153 (sent<0)?"got error":"sent only",
1154 sent, xbufp->len);
1155 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1156 svc_xprt_enqueue(rqstp->rq_xprt);
1157 sent = -EAGAIN;
1158 }
1159 return sent;
1160}
1161
1162static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1163 struct net *net,
1164 struct sockaddr *sa, int salen,
1165 int flags)
1166{
1167 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1168}
1169
1170static const struct svc_xprt_ops svc_tcp_ops = {
1171 .xpo_create = svc_tcp_create,
1172 .xpo_recvfrom = svc_tcp_recvfrom,
1173 .xpo_sendto = svc_tcp_sendto,
1174 .xpo_release_rqst = svc_release_skb,
1175 .xpo_detach = svc_tcp_sock_detach,
1176 .xpo_free = svc_sock_free,
1177 .xpo_has_wspace = svc_tcp_has_wspace,
1178 .xpo_accept = svc_tcp_accept,
1179 .xpo_secure_port = svc_sock_secure_port,
1180 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1181};
1182
1183static struct svc_xprt_class svc_tcp_class = {
1184 .xcl_name = "tcp",
1185 .xcl_owner = THIS_MODULE,
1186 .xcl_ops = &svc_tcp_ops,
1187 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1188 .xcl_ident = XPRT_TRANSPORT_TCP,
1189};
1190
1191void svc_init_xprt_sock(void)
1192{
1193 svc_reg_xprt_class(&svc_tcp_class);
1194 svc_reg_xprt_class(&svc_udp_class);
1195}
1196
1197void svc_cleanup_xprt_sock(void)
1198{
1199 svc_unreg_xprt_class(&svc_tcp_class);
1200 svc_unreg_xprt_class(&svc_udp_class);
1201}
1202
1203static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1204{
1205 struct sock *sk = svsk->sk_sk;
1206
1207 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1208 &svsk->sk_xprt, serv);
1209 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1210 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1211 if (sk->sk_state == TCP_LISTEN) {
1212 dprintk("setting up TCP socket for listening\n");
1213 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1214 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1215 sk->sk_data_ready = svc_tcp_listen_data_ready;
1216 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1217 } else {
1218 dprintk("setting up TCP socket for reading\n");
1219 sk->sk_state_change = svc_tcp_state_change;
1220 sk->sk_data_ready = svc_data_ready;
1221 sk->sk_write_space = svc_write_space;
1222
1223 svsk->sk_reclen = 0;
1224 svsk->sk_tcplen = 0;
1225 svsk->sk_datalen = 0;
1226 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1227
1228 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1229
1230 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1231 switch (sk->sk_state) {
1232 case TCP_SYN_RECV:
1233 case TCP_ESTABLISHED:
1234 break;
1235 default:
1236 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1237 }
1238 }
1239}
1240
1241void svc_sock_update_bufs(struct svc_serv *serv)
1242{
1243 /*
1244 * The number of server threads has changed. Update
1245 * rcvbuf and sndbuf accordingly on all sockets
1246 */
1247 struct svc_sock *svsk;
1248
1249 spin_lock_bh(&serv->sv_lock);
1250 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1251 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1252 spin_unlock_bh(&serv->sv_lock);
1253}
1254EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1255
1256/*
1257 * Initialize socket for RPC use and create svc_sock struct
1258 */
1259static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1260 struct socket *sock,
1261 int flags)
1262{
1263 struct svc_sock *svsk;
1264 struct sock *inet;
1265 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1266 int err = 0;
1267
1268 dprintk("svc: svc_setup_socket %p\n", sock);
1269 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1270 if (!svsk)
1271 return ERR_PTR(-ENOMEM);
1272
1273 inet = sock->sk;
1274
1275 /* Register socket with portmapper */
1276 if (pmap_register)
1277 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1278 inet->sk_protocol,
1279 ntohs(inet_sk(inet)->inet_sport));
1280
1281 if (err < 0) {
1282 kfree(svsk);
1283 return ERR_PTR(err);
1284 }
1285
1286 svsk->sk_sock = sock;
1287 svsk->sk_sk = inet;
1288 svsk->sk_ostate = inet->sk_state_change;
1289 svsk->sk_odata = inet->sk_data_ready;
1290 svsk->sk_owspace = inet->sk_write_space;
1291 /*
1292 * This barrier is necessary in order to prevent race condition
1293 * with svc_data_ready(), svc_listen_data_ready() and others
1294 * when calling callbacks above.
1295 */
1296 wmb();
1297 inet->sk_user_data = svsk;
1298
1299 /* Initialize the socket */
1300 if (sock->type == SOCK_DGRAM)
1301 svc_udp_init(svsk, serv);
1302 else
1303 svc_tcp_init(svsk, serv);
1304
1305 dprintk("svc: svc_setup_socket created %p (inet %p), "
1306 "listen %d close %d\n",
1307 svsk, svsk->sk_sk,
1308 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1309 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1310
1311 return svsk;
1312}
1313
1314bool svc_alien_sock(struct net *net, int fd)
1315{
1316 int err;
1317 struct socket *sock = sockfd_lookup(fd, &err);
1318 bool ret = false;
1319
1320 if (!sock)
1321 goto out;
1322 if (sock_net(sock->sk) != net)
1323 ret = true;
1324 sockfd_put(sock);
1325out:
1326 return ret;
1327}
1328EXPORT_SYMBOL_GPL(svc_alien_sock);
1329
1330/**
1331 * svc_addsock - add a listener socket to an RPC service
1332 * @serv: pointer to RPC service to which to add a new listener
1333 * @fd: file descriptor of the new listener
1334 * @name_return: pointer to buffer to fill in with name of listener
1335 * @len: size of the buffer
1336 * @cred: credential
1337 *
1338 * Fills in socket name and returns positive length of name if successful.
1339 * Name is terminated with '\n'. On error, returns a negative errno
1340 * value.
1341 */
1342int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1343 const size_t len, const struct cred *cred)
1344{
1345 int err = 0;
1346 struct socket *so = sockfd_lookup(fd, &err);
1347 struct svc_sock *svsk = NULL;
1348 struct sockaddr_storage addr;
1349 struct sockaddr *sin = (struct sockaddr *)&addr;
1350 int salen;
1351
1352 if (!so)
1353 return err;
1354 err = -EAFNOSUPPORT;
1355 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1356 goto out;
1357 err = -EPROTONOSUPPORT;
1358 if (so->sk->sk_protocol != IPPROTO_TCP &&
1359 so->sk->sk_protocol != IPPROTO_UDP)
1360 goto out;
1361 err = -EISCONN;
1362 if (so->state > SS_UNCONNECTED)
1363 goto out;
1364 err = -ENOENT;
1365 if (!try_module_get(THIS_MODULE))
1366 goto out;
1367 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1368 if (IS_ERR(svsk)) {
1369 module_put(THIS_MODULE);
1370 err = PTR_ERR(svsk);
1371 goto out;
1372 }
1373 salen = kernel_getsockname(svsk->sk_sock, sin);
1374 if (salen >= 0)
1375 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1376 svsk->sk_xprt.xpt_cred = get_cred(cred);
1377 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1378 return svc_one_sock_name(svsk, name_return, len);
1379out:
1380 sockfd_put(so);
1381 return err;
1382}
1383EXPORT_SYMBOL_GPL(svc_addsock);
1384
1385/*
1386 * Create socket for RPC service.
1387 */
1388static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1389 int protocol,
1390 struct net *net,
1391 struct sockaddr *sin, int len,
1392 int flags)
1393{
1394 struct svc_sock *svsk;
1395 struct socket *sock;
1396 int error;
1397 int type;
1398 struct sockaddr_storage addr;
1399 struct sockaddr *newsin = (struct sockaddr *)&addr;
1400 int newlen;
1401 int family;
1402 int val;
1403 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1404
1405 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1406 serv->sv_program->pg_name, protocol,
1407 __svc_print_addr(sin, buf, sizeof(buf)));
1408
1409 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1410 printk(KERN_WARNING "svc: only UDP and TCP "
1411 "sockets supported\n");
1412 return ERR_PTR(-EINVAL);
1413 }
1414
1415 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1416 switch (sin->sa_family) {
1417 case AF_INET6:
1418 family = PF_INET6;
1419 break;
1420 case AF_INET:
1421 family = PF_INET;
1422 break;
1423 default:
1424 return ERR_PTR(-EINVAL);
1425 }
1426
1427 error = __sock_create(net, family, type, protocol, &sock, 1);
1428 if (error < 0)
1429 return ERR_PTR(error);
1430
1431 svc_reclassify_socket(sock);
1432
1433 /*
1434 * If this is an PF_INET6 listener, we want to avoid
1435 * getting requests from IPv4 remotes. Those should
1436 * be shunted to a PF_INET listener via rpcbind.
1437 */
1438 val = 1;
1439 if (family == PF_INET6)
1440 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1441 (char *)&val, sizeof(val));
1442
1443 if (type == SOCK_STREAM)
1444 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1445 error = kernel_bind(sock, sin, len);
1446 if (error < 0)
1447 goto bummer;
1448
1449 error = kernel_getsockname(sock, newsin);
1450 if (error < 0)
1451 goto bummer;
1452 newlen = error;
1453
1454 if (protocol == IPPROTO_TCP) {
1455 if ((error = kernel_listen(sock, 64)) < 0)
1456 goto bummer;
1457 }
1458
1459 svsk = svc_setup_socket(serv, sock, flags);
1460 if (IS_ERR(svsk)) {
1461 error = PTR_ERR(svsk);
1462 goto bummer;
1463 }
1464 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1465 return (struct svc_xprt *)svsk;
1466bummer:
1467 dprintk("svc: svc_create_socket error = %d\n", -error);
1468 sock_release(sock);
1469 return ERR_PTR(error);
1470}
1471
1472/*
1473 * Detach the svc_sock from the socket so that no
1474 * more callbacks occur.
1475 */
1476static void svc_sock_detach(struct svc_xprt *xprt)
1477{
1478 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1479 struct sock *sk = svsk->sk_sk;
1480
1481 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1482
1483 /* put back the old socket callbacks */
1484 lock_sock(sk);
1485 sk->sk_state_change = svsk->sk_ostate;
1486 sk->sk_data_ready = svsk->sk_odata;
1487 sk->sk_write_space = svsk->sk_owspace;
1488 sk->sk_user_data = NULL;
1489 release_sock(sk);
1490}
1491
1492/*
1493 * Disconnect the socket, and reset the callbacks
1494 */
1495static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1496{
1497 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1498
1499 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1500
1501 svc_sock_detach(xprt);
1502
1503 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1504 svc_tcp_clear_pages(svsk);
1505 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1506 }
1507}
1508
1509/*
1510 * Free the svc_sock's socket resources and the svc_sock itself.
1511 */
1512static void svc_sock_free(struct svc_xprt *xprt)
1513{
1514 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1515 dprintk("svc: svc_sock_free(%p)\n", svsk);
1516
1517 if (svsk->sk_sock->file)
1518 sockfd_put(svsk->sk_sock);
1519 else
1520 sock_release(svsk->sk_sock);
1521 kfree(svsk);
1522}