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 */
74 enum {
75 SVC_HANDSHAKE_TO = 5U * HZ
76 };
77
78 static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
79 int flags);
80 static int svc_udp_recvfrom(struct svc_rqst *);
81 static int svc_udp_sendto(struct svc_rqst *);
82 static void svc_sock_detach(struct svc_xprt *);
83 static void svc_tcp_sock_detach(struct svc_xprt *);
84 static void svc_sock_free(struct svc_xprt *);
85
86 static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
87 struct net *, struct sockaddr *,
88 int, int);
89 #ifdef CONFIG_DEBUG_LOCK_ALLOC
90 static struct lock_class_key svc_key[2];
91 static struct lock_class_key svc_slock_key[2];
92
svc_reclassify_socket(struct socket * sock)93 static 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
svc_reclassify_socket(struct socket * sock)120 static 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 */
svc_tcp_release_ctxt(struct svc_xprt * xprt,void * ctxt)131 static 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 */
svc_udp_release_ctxt(struct svc_xprt * xprt,void * ctxt)141 static 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
149 union 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
svc_set_cmsg_data(struct svc_rqst * rqstp,struct cmsghdr * cmh)156 static 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
svc_sock_result_payload(struct svc_rqst * rqstp,unsigned int offset,unsigned int length)187 static 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 */
svc_one_sock_name(struct svc_sock * svsk,char * buf,int remaining)196 static 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
230 static int
svc_tcp_sock_process_cmsg(struct socket * sock,struct msghdr * msg,struct cmsghdr * cmsg,int ret)231 svc_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
259 static int
svc_tcp_sock_recv_cmsg(struct svc_sock * svsk,struct msghdr * msg)260 svc_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
svc_flush_bvec(const struct bio_vec * bvec,size_t size,size_t seek)278 static 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
svc_flush_bvec(const struct bio_vec * bvec,size_t size,size_t seek)290 static 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 */
svc_tcp_read_msg(struct svc_rqst * rqstp,size_t buflen,size_t seek)301 static 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 */
svc_sock_setbufsize(struct svc_sock * svsk,unsigned int nreqs)340 static 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
svc_sock_secure_port(struct svc_rqst * rqstp)354 static 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 */
svc_data_ready(struct sock * sk)365 static 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 */
svc_write_space(struct sock * sk)386 static 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
svc_tcp_has_wspace(struct svc_xprt * xprt)399 static 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
svc_tcp_kill_temp_xprt(struct svc_xprt * xprt)408 static 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 */
svc_tcp_handshake_done(void * data,int status,key_serial_t peerid)426 static 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 */
svc_tcp_handshake(struct svc_xprt * xprt)445 static 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
489 out_close:
490 set_bit(XPT_CLOSE, &xprt->xpt_flags);
491 out_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 */
svc_udp_get_dest_address4(struct svc_rqst * rqstp,struct cmsghdr * cmh)500 static 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 */
svc_udp_get_dest_address6(struct svc_rqst * rqstp,struct cmsghdr * cmh)517 static 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 */
svc_udp_get_dest_address(struct svc_rqst * rqstp,struct cmsghdr * cmh)539 static 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 */
svc_udp_recvfrom(struct svc_rqst * rqstp)562 static 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
657 out_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;
664 out_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;
668 out_bh_enable:
669 local_bh_enable();
670 out_free:
671 kfree_skb(skb);
672 out_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 */
svc_udp_sendto(struct svc_rqst * rqstp)686 static 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
734 out_notconn:
735 mutex_unlock(&xprt->xpt_mutex);
736 return -ENOTCONN;
737 }
738
svc_udp_has_wspace(struct svc_xprt * xprt)739 static 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
svc_udp_accept(struct svc_xprt * xprt)757 static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
758 {
759 BUG();
760 return NULL;
761 }
762
svc_udp_kill_temp_xprt(struct svc_xprt * xprt)763 static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
764 {
765 }
766
svc_udp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)767 static 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
775 static 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
788 static 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
svc_udp_init(struct svc_sock * svsk,struct svc_serv * serv)796 static 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 */
svc_tcp_listen_data_ready(struct sock * sk)831 static 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 */
svc_tcp_state_change(struct sock * sk)863 static 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 */
svc_tcp_accept(struct svc_xprt * xprt)880 static 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
943 failed:
944 sockfd_put(newsock);
945 return NULL;
946 }
947
svc_tcp_restore_pages(struct svc_sock * svsk,struct svc_rqst * rqstp)948 static 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
svc_tcp_save_pages(struct svc_sock * svsk,struct svc_rqst * rqstp)968 static 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
svc_tcp_clear_pages(struct svc_sock * svsk)982 static 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 }
998 out:
999 svsk->sk_tcplen = 0;
1000 svsk->sk_datalen = 0;
1001 }
1002
1003 /*
1004 * Receive fragment record header into sk_marker.
1005 */
svc_tcp_read_marker(struct svc_sock * svsk,struct svc_rqst * rqstp)1006 static 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
1037 err_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);
1042 err_short:
1043 return -EAGAIN;
1044 }
1045
receive_cb_reply(struct svc_sock * svsk,struct svc_rqst * rqstp)1046 static 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;
1076 unlock_eagain:
1077 spin_unlock(&bc_xprt->queue_lock);
1078 return -EAGAIN;
1079 }
1080
svc_tcp_fragment_received(struct svc_sock * svsk)1081 static 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 */
svc_tcp_recvfrom(struct svc_rqst * rqstp)1109 static 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
1172 err_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;
1190 error:
1191 if (len != -EAGAIN)
1192 goto err_delete;
1193 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1194 goto err_noclose;
1195 err_nuts:
1196 svsk->sk_datalen = 0;
1197 err_delete:
1198 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1199 svc_xprt_deferred_close(&svsk->sk_xprt);
1200 err_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 */
svc_tcp_sendmsg(struct svc_sock * svsk,struct svc_rqst * rqstp,rpc_fraghdr marker,unsigned int * sentp)1210 static 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 */
svc_tcp_sendto(struct svc_rqst * rqstp)1254 static 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
1277 out_notconn:
1278 mutex_unlock(&xprt->xpt_mutex);
1279 return -ENOTCONN;
1280 out_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
svc_tcp_create(struct svc_serv * serv,struct net * net,struct sockaddr * sa,int salen,int flags)1290 static 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
1298 static 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
1312 static 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
svc_init_xprt_sock(void)1320 void svc_init_xprt_sock(void)
1321 {
1322 svc_reg_xprt_class(&svc_tcp_class);
1323 svc_reg_xprt_class(&svc_udp_class);
1324 }
1325
svc_cleanup_xprt_sock(void)1326 void svc_cleanup_xprt_sock(void)
1327 {
1328 svc_unreg_xprt_class(&svc_tcp_class);
1329 svc_unreg_xprt_class(&svc_udp_class);
1330 }
1331
svc_tcp_init(struct svc_sock * svsk,struct svc_serv * serv)1332 static 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
svc_sock_update_bufs(struct svc_serv * serv)1368 void 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 }
1381
1382 /*
1383 * Initialize socket for RPC use and create svc_sock struct
1384 */
svc_setup_socket(struct svc_serv * serv,struct socket * sock,int flags)1385 static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1386 struct socket *sock,
1387 int flags)
1388 {
1389 struct svc_sock *svsk;
1390 struct sock *inet;
1391 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1392
1393 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1394 if (!svsk)
1395 return ERR_PTR(-ENOMEM);
1396
1397 inet = sock->sk;
1398
1399 if (pmap_register) {
1400 int err;
1401
1402 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1403 inet->sk_protocol,
1404 ntohs(inet_sk(inet)->inet_sport));
1405 if (err < 0) {
1406 kfree(svsk);
1407 return ERR_PTR(err);
1408 }
1409 }
1410
1411 svsk->sk_sock = sock;
1412 svsk->sk_sk = inet;
1413 svsk->sk_ostate = inet->sk_state_change;
1414 svsk->sk_odata = inet->sk_data_ready;
1415 svsk->sk_owspace = inet->sk_write_space;
1416 /*
1417 * This barrier is necessary in order to prevent race condition
1418 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1419 * when calling callbacks above.
1420 */
1421 wmb();
1422 inet->sk_user_data = svsk;
1423
1424 /* Initialize the socket */
1425 if (sock->type == SOCK_DGRAM)
1426 svc_udp_init(svsk, serv);
1427 else
1428 svc_tcp_init(svsk, serv);
1429
1430 trace_svcsock_new(svsk, sock);
1431 return svsk;
1432 }
1433
1434 /**
1435 * svc_addsock - add a listener socket to an RPC service
1436 * @serv: pointer to RPC service to which to add a new listener
1437 * @net: caller's network namespace
1438 * @fd: file descriptor of the new listener
1439 * @name_return: pointer to buffer to fill in with name of listener
1440 * @len: size of the buffer
1441 * @cred: credential
1442 *
1443 * Fills in socket name and returns positive length of name if successful.
1444 * Name is terminated with '\n'. On error, returns a negative errno
1445 * value.
1446 */
svc_addsock(struct svc_serv * serv,struct net * net,const int fd,char * name_return,const size_t len,const struct cred * cred)1447 int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1448 char *name_return, const size_t len, const struct cred *cred)
1449 {
1450 int err = 0;
1451 struct socket *so = sockfd_lookup(fd, &err);
1452 struct svc_sock *svsk = NULL;
1453 struct sockaddr_storage addr;
1454 struct sockaddr *sin = (struct sockaddr *)&addr;
1455 int salen;
1456
1457 if (!so)
1458 return err;
1459 err = -EINVAL;
1460 if (sock_net(so->sk) != net)
1461 goto out;
1462 err = -EAFNOSUPPORT;
1463 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1464 goto out;
1465 err = -EPROTONOSUPPORT;
1466 if (so->sk->sk_protocol != IPPROTO_TCP &&
1467 so->sk->sk_protocol != IPPROTO_UDP)
1468 goto out;
1469 err = -EISCONN;
1470 if (so->state > SS_UNCONNECTED)
1471 goto out;
1472 err = -ENOENT;
1473 if (!try_module_get(THIS_MODULE))
1474 goto out;
1475 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1476 if (IS_ERR(svsk)) {
1477 module_put(THIS_MODULE);
1478 err = PTR_ERR(svsk);
1479 goto out;
1480 }
1481 salen = kernel_getsockname(svsk->sk_sock, sin);
1482 if (salen >= 0)
1483 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1484 svsk->sk_xprt.xpt_cred = get_cred(cred);
1485 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1486 return svc_one_sock_name(svsk, name_return, len);
1487 out:
1488 sockfd_put(so);
1489 return err;
1490 }
1491 EXPORT_SYMBOL_GPL(svc_addsock);
1492
1493 /*
1494 * Create socket for RPC service.
1495 */
svc_create_socket(struct svc_serv * serv,int protocol,struct net * net,struct sockaddr * sin,int len,int flags)1496 static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1497 int protocol,
1498 struct net *net,
1499 struct sockaddr *sin, int len,
1500 int flags)
1501 {
1502 struct svc_sock *svsk;
1503 struct socket *sock;
1504 int error;
1505 int type;
1506 struct sockaddr_storage addr;
1507 struct sockaddr *newsin = (struct sockaddr *)&addr;
1508 int newlen;
1509 int family;
1510
1511 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1512 printk(KERN_WARNING "svc: only UDP and TCP "
1513 "sockets supported\n");
1514 return ERR_PTR(-EINVAL);
1515 }
1516
1517 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1518 switch (sin->sa_family) {
1519 case AF_INET6:
1520 family = PF_INET6;
1521 break;
1522 case AF_INET:
1523 family = PF_INET;
1524 break;
1525 default:
1526 return ERR_PTR(-EINVAL);
1527 }
1528
1529 error = __sock_create(net, family, type, protocol, &sock, 1);
1530 if (error < 0)
1531 return ERR_PTR(error);
1532
1533 svc_reclassify_socket(sock);
1534
1535 /*
1536 * If this is an PF_INET6 listener, we want to avoid
1537 * getting requests from IPv4 remotes. Those should
1538 * be shunted to a PF_INET listener via rpcbind.
1539 */
1540 if (family == PF_INET6)
1541 ip6_sock_set_v6only(sock->sk);
1542 if (type == SOCK_STREAM)
1543 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1544 error = kernel_bind(sock, sin, len);
1545 if (error < 0)
1546 goto bummer;
1547
1548 error = kernel_getsockname(sock, newsin);
1549 if (error < 0)
1550 goto bummer;
1551 newlen = error;
1552
1553 if (protocol == IPPROTO_TCP) {
1554 __netns_tracker_free(net, &sock->sk->ns_tracker, false);
1555 sock->sk->sk_net_refcnt = 1;
1556 get_net_track(net, &sock->sk->ns_tracker, GFP_KERNEL);
1557 sock_inuse_add(net, 1);
1558 if ((error = kernel_listen(sock, 64)) < 0)
1559 goto bummer;
1560 }
1561
1562 svsk = svc_setup_socket(serv, sock, flags);
1563 if (IS_ERR(svsk)) {
1564 error = PTR_ERR(svsk);
1565 goto bummer;
1566 }
1567 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1568 return (struct svc_xprt *)svsk;
1569 bummer:
1570 sock_release(sock);
1571 return ERR_PTR(error);
1572 }
1573
1574 /*
1575 * Detach the svc_sock from the socket so that no
1576 * more callbacks occur.
1577 */
svc_sock_detach(struct svc_xprt * xprt)1578 static void svc_sock_detach(struct svc_xprt *xprt)
1579 {
1580 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1581 struct sock *sk = svsk->sk_sk;
1582
1583 /* put back the old socket callbacks */
1584 lock_sock(sk);
1585 sk->sk_state_change = svsk->sk_ostate;
1586 sk->sk_data_ready = svsk->sk_odata;
1587 sk->sk_write_space = svsk->sk_owspace;
1588 sk->sk_user_data = NULL;
1589 release_sock(sk);
1590 }
1591
1592 /*
1593 * Disconnect the socket, and reset the callbacks
1594 */
svc_tcp_sock_detach(struct svc_xprt * xprt)1595 static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1596 {
1597 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1598
1599 tls_handshake_close(svsk->sk_sock);
1600
1601 svc_sock_detach(xprt);
1602
1603 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1604 svc_tcp_clear_pages(svsk);
1605 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1606 }
1607 }
1608
1609 /*
1610 * Free the svc_sock's socket resources and the svc_sock itself.
1611 */
svc_sock_free(struct svc_xprt * xprt)1612 static void svc_sock_free(struct svc_xprt *xprt)
1613 {
1614 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1615 struct socket *sock = svsk->sk_sock;
1616
1617 trace_svcsock_free(svsk, sock);
1618
1619 tls_handshake_cancel(sock->sk);
1620 if (sock->file)
1621 sockfd_put(sock);
1622 else
1623 sock_release(sock);
1624
1625 page_frag_cache_drain(&svsk->sk_frag_cache);
1626 kfree(svsk);
1627 }
1628