Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svcsock.c
4 *
5 * These are the RPC server socket internals.
6 *
7 * The server scheduling algorithm does not always distribute the load
8 * evenly when servicing a single client. May need to modify the
9 * svc_xprt_enqueue procedure...
10 *
11 * TCP support is largely untested and may be a little slow. The problem
12 * is that we currently do two separate recvfrom's, one for the 4-byte
13 * record length, and the second for the actual record. This could possibly
14 * be improved by always reading a minimum size of around 100 bytes and
15 * tucking any superfluous bytes away in a temporary store. Still, that
16 * leaves write requests out in the rain. An alternative may be to peek at
17 * the first skb in the queue, and if it matches the next TCP sequence
18 * number, to extract the record marker. Yuck.
19 *
20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
21 */
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/module.h>
26#include <linux/errno.h>
27#include <linux/fcntl.h>
28#include <linux/net.h>
29#include <linux/in.h>
30#include <linux/inet.h>
31#include <linux/udp.h>
32#include <linux/tcp.h>
33#include <linux/unistd.h>
34#include <linux/slab.h>
35#include <linux/netdevice.h>
36#include <linux/skbuff.h>
37#include <linux/file.h>
38#include <linux/freezer.h>
39#include <net/sock.h>
40#include <net/checksum.h>
41#include <net/ip.h>
42#include <net/ipv6.h>
43#include <net/udp.h>
44#include <net/tcp.h>
45#include <net/tcp_states.h>
46#include <linux/uaccess.h>
47#include <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}
1/*
2 * linux/net/sunrpc/svcsock.c
3 *
4 * These are the RPC server socket internals.
5 *
6 * The server scheduling algorithm does not always distribute the load
7 * evenly when servicing a single client. May need to modify the
8 * svc_xprt_enqueue procedure...
9 *
10 * TCP support is largely untested and may be a little slow. The problem
11 * is that we currently do two separate recvfrom's, one for the 4-byte
12 * record length, and the second for the actual record. This could possibly
13 * be improved by always reading a minimum size of around 100 bytes and
14 * tucking any superfluous bytes away in a temporary store. Still, that
15 * leaves write requests out in the rain. An alternative may be to peek at
16 * the first skb in the queue, and if it matches the next TCP sequence
17 * number, to extract the record marker. Yuck.
18 *
19 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
20 */
21
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/module.h>
25#include <linux/errno.h>
26#include <linux/fcntl.h>
27#include <linux/net.h>
28#include <linux/in.h>
29#include <linux/inet.h>
30#include <linux/udp.h>
31#include <linux/tcp.h>
32#include <linux/unistd.h>
33#include <linux/slab.h>
34#include <linux/netdevice.h>
35#include <linux/skbuff.h>
36#include <linux/file.h>
37#include <linux/freezer.h>
38#include <net/sock.h>
39#include <net/checksum.h>
40#include <net/ip.h>
41#include <net/ipv6.h>
42#include <net/udp.h>
43#include <net/tcp.h>
44#include <net/tcp_states.h>
45#include <linux/uaccess.h>
46#include <asm/ioctls.h>
47#include <trace/events/skb.h>
48
49#include <linux/sunrpc/types.h>
50#include <linux/sunrpc/clnt.h>
51#include <linux/sunrpc/xdr.h>
52#include <linux/sunrpc/msg_prot.h>
53#include <linux/sunrpc/svcsock.h>
54#include <linux/sunrpc/stats.h>
55#include <linux/sunrpc/xprt.h>
56
57#include "sunrpc.h"
58
59#define RPCDBG_FACILITY RPCDBG_SVCXPRT
60
61
62static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
63 int flags);
64static int svc_udp_recvfrom(struct svc_rqst *);
65static int svc_udp_sendto(struct svc_rqst *);
66static void svc_sock_detach(struct svc_xprt *);
67static void svc_tcp_sock_detach(struct svc_xprt *);
68static void svc_sock_free(struct svc_xprt *);
69
70static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
71 struct net *, struct sockaddr *,
72 int, int);
73#if defined(CONFIG_SUNRPC_BACKCHANNEL)
74static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
75 struct net *, struct sockaddr *,
76 int, int);
77static void svc_bc_sock_free(struct svc_xprt *xprt);
78#endif /* CONFIG_SUNRPC_BACKCHANNEL */
79
80#ifdef CONFIG_DEBUG_LOCK_ALLOC
81static struct lock_class_key svc_key[2];
82static struct lock_class_key svc_slock_key[2];
83
84static void svc_reclassify_socket(struct socket *sock)
85{
86 struct sock *sk = sock->sk;
87
88 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
89 return;
90
91 switch (sk->sk_family) {
92 case AF_INET:
93 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
94 &svc_slock_key[0],
95 "sk_xprt.xpt_lock-AF_INET-NFSD",
96 &svc_key[0]);
97 break;
98
99 case AF_INET6:
100 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
101 &svc_slock_key[1],
102 "sk_xprt.xpt_lock-AF_INET6-NFSD",
103 &svc_key[1]);
104 break;
105
106 default:
107 BUG();
108 }
109}
110#else
111static void svc_reclassify_socket(struct socket *sock)
112{
113}
114#endif
115
116/*
117 * Release an skbuff after use
118 */
119static void svc_release_skb(struct svc_rqst *rqstp)
120{
121 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
122
123 if (skb) {
124 struct svc_sock *svsk =
125 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
126 rqstp->rq_xprt_ctxt = NULL;
127
128 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
129 skb_free_datagram_locked(svsk->sk_sk, skb);
130 }
131}
132
133static void svc_release_udp_skb(struct svc_rqst *rqstp)
134{
135 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
136
137 if (skb) {
138 rqstp->rq_xprt_ctxt = NULL;
139
140 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
141 consume_skb(skb);
142 }
143}
144
145union svc_pktinfo_u {
146 struct in_pktinfo pkti;
147 struct in6_pktinfo pkti6;
148};
149#define SVC_PKTINFO_SPACE \
150 CMSG_SPACE(sizeof(union svc_pktinfo_u))
151
152static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
153{
154 struct svc_sock *svsk =
155 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
156 switch (svsk->sk_sk->sk_family) {
157 case AF_INET: {
158 struct in_pktinfo *pki = CMSG_DATA(cmh);
159
160 cmh->cmsg_level = SOL_IP;
161 cmh->cmsg_type = IP_PKTINFO;
162 pki->ipi_ifindex = 0;
163 pki->ipi_spec_dst.s_addr =
164 svc_daddr_in(rqstp)->sin_addr.s_addr;
165 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
166 }
167 break;
168
169 case AF_INET6: {
170 struct in6_pktinfo *pki = CMSG_DATA(cmh);
171 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
172
173 cmh->cmsg_level = SOL_IPV6;
174 cmh->cmsg_type = IPV6_PKTINFO;
175 pki->ipi6_ifindex = daddr->sin6_scope_id;
176 pki->ipi6_addr = daddr->sin6_addr;
177 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
178 }
179 break;
180 }
181}
182
183/*
184 * send routine intended to be shared by the fore- and back-channel
185 */
186int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
187 struct page *headpage, unsigned long headoffset,
188 struct page *tailpage, unsigned long tailoffset)
189{
190 int result;
191 int size;
192 struct page **ppage = xdr->pages;
193 size_t base = xdr->page_base;
194 unsigned int pglen = xdr->page_len;
195 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
196 int slen;
197 int len = 0;
198
199 slen = xdr->len;
200
201 /* send head */
202 if (slen == xdr->head[0].iov_len)
203 flags = 0;
204 len = kernel_sendpage(sock, headpage, headoffset,
205 xdr->head[0].iov_len, flags);
206 if (len != xdr->head[0].iov_len)
207 goto out;
208 slen -= xdr->head[0].iov_len;
209 if (slen == 0)
210 goto out;
211
212 /* send page data */
213 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
214 while (pglen > 0) {
215 if (slen == size)
216 flags = 0;
217 result = kernel_sendpage(sock, *ppage, base, size, flags);
218 if (result > 0)
219 len += result;
220 if (result != size)
221 goto out;
222 slen -= size;
223 pglen -= size;
224 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
225 base = 0;
226 ppage++;
227 }
228
229 /* send tail */
230 if (xdr->tail[0].iov_len) {
231 result = kernel_sendpage(sock, tailpage, tailoffset,
232 xdr->tail[0].iov_len, 0);
233 if (result > 0)
234 len += result;
235 }
236
237out:
238 return len;
239}
240
241
242/*
243 * Generic sendto routine
244 */
245static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
246{
247 struct svc_sock *svsk =
248 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
249 struct socket *sock = svsk->sk_sock;
250 union {
251 struct cmsghdr hdr;
252 long all[SVC_PKTINFO_SPACE / sizeof(long)];
253 } buffer;
254 struct cmsghdr *cmh = &buffer.hdr;
255 int len = 0;
256 unsigned long tailoff;
257 unsigned long headoff;
258 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
259
260 if (rqstp->rq_prot == IPPROTO_UDP) {
261 struct msghdr msg = {
262 .msg_name = &rqstp->rq_addr,
263 .msg_namelen = rqstp->rq_addrlen,
264 .msg_control = cmh,
265 .msg_controllen = sizeof(buffer),
266 .msg_flags = MSG_MORE,
267 };
268
269 svc_set_cmsg_data(rqstp, cmh);
270
271 if (sock_sendmsg(sock, &msg) < 0)
272 goto out;
273 }
274
275 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
276 headoff = 0;
277 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
278 rqstp->rq_respages[0], tailoff);
279
280out:
281 dprintk("svc: socket %p sendto([%p %zu... ], %d) = %d (addr %s)\n",
282 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
283 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
284
285 return len;
286}
287
288/*
289 * Report socket names for nfsdfs
290 */
291static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
292{
293 const struct sock *sk = svsk->sk_sk;
294 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
295 "udp" : "tcp";
296 int len;
297
298 switch (sk->sk_family) {
299 case PF_INET:
300 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
301 proto_name,
302 &inet_sk(sk)->inet_rcv_saddr,
303 inet_sk(sk)->inet_num);
304 break;
305#if IS_ENABLED(CONFIG_IPV6)
306 case PF_INET6:
307 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
308 proto_name,
309 &sk->sk_v6_rcv_saddr,
310 inet_sk(sk)->inet_num);
311 break;
312#endif
313 default:
314 len = snprintf(buf, remaining, "*unknown-%d*\n",
315 sk->sk_family);
316 }
317
318 if (len >= remaining) {
319 *buf = '\0';
320 return -ENAMETOOLONG;
321 }
322 return len;
323}
324
325/*
326 * Generic recvfrom routine.
327 */
328static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
329 int buflen)
330{
331 struct svc_sock *svsk =
332 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
333 struct msghdr msg = {
334 .msg_flags = MSG_DONTWAIT,
335 };
336 int len;
337
338 rqstp->rq_xprt_hlen = 0;
339
340 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
341 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, iov, nr, buflen);
342 len = sock_recvmsg(svsk->sk_sock, &msg, msg.msg_flags);
343 /* If we read a full record, then assume there may be more
344 * data to read (stream based sockets only!)
345 */
346 if (len == buflen)
347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
348
349 dprintk("svc: socket %p recvfrom(%p, %zu) = %d\n",
350 svsk, iov[0].iov_base, iov[0].iov_len, len);
351 return len;
352}
353
354static int svc_partial_recvfrom(struct svc_rqst *rqstp,
355 struct kvec *iov, int nr,
356 int buflen, unsigned int base)
357{
358 size_t save_iovlen;
359 void *save_iovbase;
360 unsigned int i;
361 int ret;
362
363 if (base == 0)
364 return svc_recvfrom(rqstp, iov, nr, buflen);
365
366 for (i = 0; i < nr; i++) {
367 if (iov[i].iov_len > base)
368 break;
369 base -= iov[i].iov_len;
370 }
371 save_iovlen = iov[i].iov_len;
372 save_iovbase = iov[i].iov_base;
373 iov[i].iov_len -= base;
374 iov[i].iov_base += base;
375 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
376 iov[i].iov_len = save_iovlen;
377 iov[i].iov_base = save_iovbase;
378 return ret;
379}
380
381/*
382 * Set socket snd and rcv buffer lengths
383 */
384static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
385 unsigned int rcv)
386{
387 lock_sock(sock->sk);
388 sock->sk->sk_sndbuf = snd * 2;
389 sock->sk->sk_rcvbuf = rcv * 2;
390 sock->sk->sk_write_space(sock->sk);
391 release_sock(sock->sk);
392}
393
394static void svc_sock_secure_port(struct svc_rqst *rqstp)
395{
396 if (svc_port_is_privileged(svc_addr(rqstp)))
397 set_bit(RQ_SECURE, &rqstp->rq_flags);
398 else
399 clear_bit(RQ_SECURE, &rqstp->rq_flags);
400}
401
402/*
403 * INET callback when data has been received on the socket.
404 */
405static void svc_data_ready(struct sock *sk)
406{
407 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
408
409 if (svsk) {
410 dprintk("svc: socket %p(inet %p), busy=%d\n",
411 svsk, sk,
412 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
413
414 /* Refer to svc_setup_socket() for details. */
415 rmb();
416 svsk->sk_odata(sk);
417 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
418 svc_xprt_enqueue(&svsk->sk_xprt);
419 }
420}
421
422/*
423 * INET callback when space is newly available on the socket.
424 */
425static void svc_write_space(struct sock *sk)
426{
427 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
428
429 if (svsk) {
430 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
431 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
432
433 /* Refer to svc_setup_socket() for details. */
434 rmb();
435 svsk->sk_owspace(sk);
436 svc_xprt_enqueue(&svsk->sk_xprt);
437 }
438}
439
440static int svc_tcp_has_wspace(struct svc_xprt *xprt)
441{
442 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
443
444 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
445 return 1;
446 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
447}
448
449static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
450{
451 struct svc_sock *svsk;
452 struct socket *sock;
453 struct linger no_linger = {
454 .l_onoff = 1,
455 .l_linger = 0,
456 };
457
458 svsk = container_of(xprt, struct svc_sock, sk_xprt);
459 sock = svsk->sk_sock;
460 kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
461 (char *)&no_linger, sizeof(no_linger));
462}
463
464/*
465 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
466 */
467static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
468 struct cmsghdr *cmh)
469{
470 struct in_pktinfo *pki = CMSG_DATA(cmh);
471 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
472
473 if (cmh->cmsg_type != IP_PKTINFO)
474 return 0;
475
476 daddr->sin_family = AF_INET;
477 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
478 return 1;
479}
480
481/*
482 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
483 */
484static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
485 struct cmsghdr *cmh)
486{
487 struct in6_pktinfo *pki = CMSG_DATA(cmh);
488 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
489
490 if (cmh->cmsg_type != IPV6_PKTINFO)
491 return 0;
492
493 daddr->sin6_family = AF_INET6;
494 daddr->sin6_addr = pki->ipi6_addr;
495 daddr->sin6_scope_id = pki->ipi6_ifindex;
496 return 1;
497}
498
499/*
500 * Copy the UDP datagram's destination address to the rqstp structure.
501 * The 'destination' address in this case is the address to which the
502 * peer sent the datagram, i.e. our local address. For multihomed
503 * hosts, this can change from msg to msg. Note that only the IP
504 * address changes, the port number should remain the same.
505 */
506static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
507 struct cmsghdr *cmh)
508{
509 switch (cmh->cmsg_level) {
510 case SOL_IP:
511 return svc_udp_get_dest_address4(rqstp, cmh);
512 case SOL_IPV6:
513 return svc_udp_get_dest_address6(rqstp, cmh);
514 }
515
516 return 0;
517}
518
519/*
520 * Receive a datagram from a UDP socket.
521 */
522static int svc_udp_recvfrom(struct svc_rqst *rqstp)
523{
524 struct svc_sock *svsk =
525 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
526 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
527 struct sk_buff *skb;
528 union {
529 struct cmsghdr hdr;
530 long all[SVC_PKTINFO_SPACE / sizeof(long)];
531 } buffer;
532 struct cmsghdr *cmh = &buffer.hdr;
533 struct msghdr msg = {
534 .msg_name = svc_addr(rqstp),
535 .msg_control = cmh,
536 .msg_controllen = sizeof(buffer),
537 .msg_flags = MSG_DONTWAIT,
538 };
539 size_t len;
540 int err;
541
542 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
543 /* udp sockets need large rcvbuf as all pending
544 * requests are still in that buffer. sndbuf must
545 * also be large enough that there is enough space
546 * for one reply per thread. We count all threads
547 * rather than threads in a particular pool, which
548 * provides an upper bound on the number of threads
549 * which will access the socket.
550 */
551 svc_sock_setbufsize(svsk->sk_sock,
552 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
553 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
554
555 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
556 skb = NULL;
557 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
558 0, 0, MSG_PEEK | MSG_DONTWAIT);
559 if (err >= 0)
560 skb = skb_recv_udp(svsk->sk_sk, 0, 1, &err);
561
562 if (skb == NULL) {
563 if (err != -EAGAIN) {
564 /* possibly an icmp error */
565 dprintk("svc: recvfrom returned error %d\n", -err);
566 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
567 }
568 return 0;
569 }
570 len = svc_addr_len(svc_addr(rqstp));
571 rqstp->rq_addrlen = len;
572 if (skb->tstamp == 0) {
573 skb->tstamp = ktime_get_real();
574 /* Don't enable netstamp, sunrpc doesn't
575 need that much accuracy */
576 }
577 svsk->sk_sk->sk_stamp = skb->tstamp;
578 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
579
580 len = skb->len;
581 rqstp->rq_arg.len = len;
582
583 rqstp->rq_prot = IPPROTO_UDP;
584
585 if (!svc_udp_get_dest_address(rqstp, cmh)) {
586 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
587 cmh->cmsg_level, cmh->cmsg_type);
588 goto out_free;
589 }
590 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
591
592 if (skb_is_nonlinear(skb)) {
593 /* we have to copy */
594 local_bh_disable();
595 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
596 local_bh_enable();
597 /* checksum error */
598 goto out_free;
599 }
600 local_bh_enable();
601 consume_skb(skb);
602 } else {
603 /* we can use it in-place */
604 rqstp->rq_arg.head[0].iov_base = skb->data;
605 rqstp->rq_arg.head[0].iov_len = len;
606 if (skb_checksum_complete(skb))
607 goto out_free;
608 rqstp->rq_xprt_ctxt = skb;
609 }
610
611 rqstp->rq_arg.page_base = 0;
612 if (len <= rqstp->rq_arg.head[0].iov_len) {
613 rqstp->rq_arg.head[0].iov_len = len;
614 rqstp->rq_arg.page_len = 0;
615 rqstp->rq_respages = rqstp->rq_pages+1;
616 } else {
617 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
618 rqstp->rq_respages = rqstp->rq_pages + 1 +
619 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
620 }
621 rqstp->rq_next_page = rqstp->rq_respages+1;
622
623 if (serv->sv_stats)
624 serv->sv_stats->netudpcnt++;
625
626 return len;
627out_free:
628 kfree_skb(skb);
629 return 0;
630}
631
632static int
633svc_udp_sendto(struct svc_rqst *rqstp)
634{
635 int error;
636
637 error = svc_sendto(rqstp, &rqstp->rq_res);
638 if (error == -ECONNREFUSED)
639 /* ICMP error on earlier request. */
640 error = svc_sendto(rqstp, &rqstp->rq_res);
641
642 return error;
643}
644
645static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
646{
647}
648
649static int svc_udp_has_wspace(struct svc_xprt *xprt)
650{
651 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
652 struct svc_serv *serv = xprt->xpt_server;
653 unsigned long required;
654
655 /*
656 * Set the SOCK_NOSPACE flag before checking the available
657 * sock space.
658 */
659 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
660 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
661 if (required*2 > sock_wspace(svsk->sk_sk))
662 return 0;
663 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
664 return 1;
665}
666
667static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
668{
669 BUG();
670 return NULL;
671}
672
673static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
674{
675}
676
677static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
678 struct net *net,
679 struct sockaddr *sa, int salen,
680 int flags)
681{
682 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
683}
684
685static const struct svc_xprt_ops svc_udp_ops = {
686 .xpo_create = svc_udp_create,
687 .xpo_recvfrom = svc_udp_recvfrom,
688 .xpo_sendto = svc_udp_sendto,
689 .xpo_release_rqst = svc_release_udp_skb,
690 .xpo_detach = svc_sock_detach,
691 .xpo_free = svc_sock_free,
692 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
693 .xpo_has_wspace = svc_udp_has_wspace,
694 .xpo_accept = svc_udp_accept,
695 .xpo_secure_port = svc_sock_secure_port,
696 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
697};
698
699static struct svc_xprt_class svc_udp_class = {
700 .xcl_name = "udp",
701 .xcl_owner = THIS_MODULE,
702 .xcl_ops = &svc_udp_ops,
703 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
704 .xcl_ident = XPRT_TRANSPORT_UDP,
705};
706
707static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
708{
709 int err, level, optname, one = 1;
710
711 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
712 &svsk->sk_xprt, serv);
713 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
714 svsk->sk_sk->sk_data_ready = svc_data_ready;
715 svsk->sk_sk->sk_write_space = svc_write_space;
716
717 /* initialise setting must have enough space to
718 * receive and respond to one request.
719 * svc_udp_recvfrom will re-adjust if necessary
720 */
721 svc_sock_setbufsize(svsk->sk_sock,
722 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
723 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
724
725 /* data might have come in before data_ready set up */
726 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
727 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
728
729 /* make sure we get destination address info */
730 switch (svsk->sk_sk->sk_family) {
731 case AF_INET:
732 level = SOL_IP;
733 optname = IP_PKTINFO;
734 break;
735 case AF_INET6:
736 level = SOL_IPV6;
737 optname = IPV6_RECVPKTINFO;
738 break;
739 default:
740 BUG();
741 }
742 err = kernel_setsockopt(svsk->sk_sock, level, optname,
743 (char *)&one, sizeof(one));
744 dprintk("svc: kernel_setsockopt returned %d\n", err);
745}
746
747/*
748 * A data_ready event on a listening socket means there's a connection
749 * pending. Do not use state_change as a substitute for it.
750 */
751static void svc_tcp_listen_data_ready(struct sock *sk)
752{
753 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
754
755 dprintk("svc: socket %p TCP (listen) state change %d\n",
756 sk, sk->sk_state);
757
758 if (svsk) {
759 /* Refer to svc_setup_socket() for details. */
760 rmb();
761 svsk->sk_odata(sk);
762 }
763
764 /*
765 * This callback may called twice when a new connection
766 * is established as a child socket inherits everything
767 * from a parent LISTEN socket.
768 * 1) data_ready method of the parent socket will be called
769 * when one of child sockets become ESTABLISHED.
770 * 2) data_ready method of the child socket may be called
771 * when it receives data before the socket is accepted.
772 * In case of 2, we should ignore it silently.
773 */
774 if (sk->sk_state == TCP_LISTEN) {
775 if (svsk) {
776 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
777 svc_xprt_enqueue(&svsk->sk_xprt);
778 } else
779 printk("svc: socket %p: no user data\n", sk);
780 }
781}
782
783/*
784 * A state change on a connected socket means it's dying or dead.
785 */
786static void svc_tcp_state_change(struct sock *sk)
787{
788 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
789
790 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
791 sk, sk->sk_state, sk->sk_user_data);
792
793 if (!svsk)
794 printk("svc: socket %p: no user data\n", sk);
795 else {
796 /* Refer to svc_setup_socket() for details. */
797 rmb();
798 svsk->sk_ostate(sk);
799 if (sk->sk_state != TCP_ESTABLISHED) {
800 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
801 svc_xprt_enqueue(&svsk->sk_xprt);
802 }
803 }
804}
805
806/*
807 * Accept a TCP connection
808 */
809static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
810{
811 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
812 struct sockaddr_storage addr;
813 struct sockaddr *sin = (struct sockaddr *) &addr;
814 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
815 struct socket *sock = svsk->sk_sock;
816 struct socket *newsock;
817 struct svc_sock *newsvsk;
818 int err, slen;
819 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
820
821 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
822 if (!sock)
823 return NULL;
824
825 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
826 err = kernel_accept(sock, &newsock, O_NONBLOCK);
827 if (err < 0) {
828 if (err == -ENOMEM)
829 printk(KERN_WARNING "%s: no more sockets!\n",
830 serv->sv_name);
831 else if (err != -EAGAIN)
832 net_warn_ratelimited("%s: accept failed (err %d)!\n",
833 serv->sv_name, -err);
834 return NULL;
835 }
836 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
837
838 err = kernel_getpeername(newsock, sin);
839 if (err < 0) {
840 net_warn_ratelimited("%s: peername failed (err %d)!\n",
841 serv->sv_name, -err);
842 goto failed; /* aborted connection or whatever */
843 }
844 slen = err;
845
846 /* Ideally, we would want to reject connections from unauthorized
847 * hosts here, but when we get encryption, the IP of the host won't
848 * tell us anything. For now just warn about unpriv connections.
849 */
850 if (!svc_port_is_privileged(sin)) {
851 dprintk("%s: connect from unprivileged port: %s\n",
852 serv->sv_name,
853 __svc_print_addr(sin, buf, sizeof(buf)));
854 }
855 dprintk("%s: connect from %s\n", serv->sv_name,
856 __svc_print_addr(sin, buf, sizeof(buf)));
857
858 /* Reset the inherited callbacks before calling svc_setup_socket */
859 newsock->sk->sk_state_change = svsk->sk_ostate;
860 newsock->sk->sk_data_ready = svsk->sk_odata;
861 newsock->sk->sk_write_space = svsk->sk_owspace;
862
863 /* make sure that a write doesn't block forever when
864 * low on memory
865 */
866 newsock->sk->sk_sndtimeo = HZ*30;
867
868 newsvsk = svc_setup_socket(serv, newsock,
869 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
870 if (IS_ERR(newsvsk))
871 goto failed;
872 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
873 err = kernel_getsockname(newsock, sin);
874 slen = err;
875 if (unlikely(err < 0)) {
876 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
877 slen = offsetof(struct sockaddr, sa_data);
878 }
879 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
880
881 if (sock_is_loopback(newsock->sk))
882 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
883 else
884 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
885 if (serv->sv_stats)
886 serv->sv_stats->nettcpconn++;
887
888 return &newsvsk->sk_xprt;
889
890failed:
891 sock_release(newsock);
892 return NULL;
893}
894
895static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
896{
897 unsigned int i, len, npages;
898
899 if (svsk->sk_datalen == 0)
900 return 0;
901 len = svsk->sk_datalen;
902 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
903 for (i = 0; i < npages; i++) {
904 if (rqstp->rq_pages[i] != NULL)
905 put_page(rqstp->rq_pages[i]);
906 BUG_ON(svsk->sk_pages[i] == NULL);
907 rqstp->rq_pages[i] = svsk->sk_pages[i];
908 svsk->sk_pages[i] = NULL;
909 }
910 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
911 return len;
912}
913
914static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
915{
916 unsigned int i, len, npages;
917
918 if (svsk->sk_datalen == 0)
919 return;
920 len = svsk->sk_datalen;
921 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
922 for (i = 0; i < npages; i++) {
923 svsk->sk_pages[i] = rqstp->rq_pages[i];
924 rqstp->rq_pages[i] = NULL;
925 }
926}
927
928static void svc_tcp_clear_pages(struct svc_sock *svsk)
929{
930 unsigned int i, len, npages;
931
932 if (svsk->sk_datalen == 0)
933 goto out;
934 len = svsk->sk_datalen;
935 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
936 for (i = 0; i < npages; i++) {
937 if (svsk->sk_pages[i] == NULL) {
938 WARN_ON_ONCE(1);
939 continue;
940 }
941 put_page(svsk->sk_pages[i]);
942 svsk->sk_pages[i] = NULL;
943 }
944out:
945 svsk->sk_tcplen = 0;
946 svsk->sk_datalen = 0;
947}
948
949/*
950 * Receive fragment record header.
951 * If we haven't gotten the record length yet, get the next four bytes.
952 */
953static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
954{
955 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
956 unsigned int want;
957 int len;
958
959 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
960 struct kvec iov;
961
962 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
963 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
964 iov.iov_len = want;
965 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
966 goto error;
967 svsk->sk_tcplen += len;
968
969 if (len < want) {
970 dprintk("svc: short recvfrom while reading record "
971 "length (%d of %d)\n", len, want);
972 return -EAGAIN;
973 }
974
975 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
976 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
977 serv->sv_max_mesg) {
978 net_notice_ratelimited("RPC: fragment too large: %d\n",
979 svc_sock_reclen(svsk));
980 goto err_delete;
981 }
982 }
983
984 return svc_sock_reclen(svsk);
985error:
986 dprintk("RPC: TCP recv_record got %d\n", len);
987 return len;
988err_delete:
989 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
990 return -EAGAIN;
991}
992
993static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
994{
995 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
996 struct rpc_rqst *req = NULL;
997 struct kvec *src, *dst;
998 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
999 __be32 xid;
1000 __be32 calldir;
1001
1002 xid = *p++;
1003 calldir = *p;
1004
1005 if (!bc_xprt)
1006 return -EAGAIN;
1007 spin_lock(&bc_xprt->recv_lock);
1008 req = xprt_lookup_rqst(bc_xprt, xid);
1009 if (!req)
1010 goto unlock_notfound;
1011
1012 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1013 /*
1014 * XXX!: cheating for now! Only copying HEAD.
1015 * But we know this is good enough for now (in fact, for any
1016 * callback reply in the forseeable future).
1017 */
1018 dst = &req->rq_private_buf.head[0];
1019 src = &rqstp->rq_arg.head[0];
1020 if (dst->iov_len < src->iov_len)
1021 goto unlock_eagain; /* whatever; just giving up. */
1022 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1023 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1024 rqstp->rq_arg.len = 0;
1025 spin_unlock(&bc_xprt->recv_lock);
1026 return 0;
1027unlock_notfound:
1028 printk(KERN_NOTICE
1029 "%s: Got unrecognized reply: "
1030 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1031 __func__, ntohl(calldir),
1032 bc_xprt, ntohl(xid));
1033unlock_eagain:
1034 spin_unlock(&bc_xprt->recv_lock);
1035 return -EAGAIN;
1036}
1037
1038static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1039{
1040 int i = 0;
1041 int t = 0;
1042
1043 while (t < len) {
1044 vec[i].iov_base = page_address(pages[i]);
1045 vec[i].iov_len = PAGE_SIZE;
1046 i++;
1047 t += PAGE_SIZE;
1048 }
1049 return i;
1050}
1051
1052static void svc_tcp_fragment_received(struct svc_sock *svsk)
1053{
1054 /* If we have more data, signal svc_xprt_enqueue() to try again */
1055 dprintk("svc: TCP %s record (%d bytes)\n",
1056 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1057 svc_sock_reclen(svsk));
1058 svsk->sk_tcplen = 0;
1059 svsk->sk_reclen = 0;
1060}
1061
1062/*
1063 * Receive data from a TCP socket.
1064 */
1065static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1066{
1067 struct svc_sock *svsk =
1068 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1069 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1070 int len;
1071 struct kvec *vec;
1072 unsigned int want, base;
1073 __be32 *p;
1074 __be32 calldir;
1075 int pnum;
1076
1077 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1078 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1079 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1080 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1081
1082 len = svc_tcp_recv_record(svsk, rqstp);
1083 if (len < 0)
1084 goto error;
1085
1086 base = svc_tcp_restore_pages(svsk, rqstp);
1087 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1088
1089 vec = rqstp->rq_vec;
1090
1091 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1092 svsk->sk_datalen + want);
1093
1094 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1095 rqstp->rq_next_page = rqstp->rq_respages + 1;
1096
1097 /* Now receive data */
1098 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1099 if (len >= 0) {
1100 svsk->sk_tcplen += len;
1101 svsk->sk_datalen += len;
1102 }
1103 if (len != want || !svc_sock_final_rec(svsk)) {
1104 svc_tcp_save_pages(svsk, rqstp);
1105 if (len < 0 && len != -EAGAIN)
1106 goto err_delete;
1107 if (len == want)
1108 svc_tcp_fragment_received(svsk);
1109 else
1110 dprintk("svc: incomplete TCP record (%d of %d)\n",
1111 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1112 svc_sock_reclen(svsk));
1113 goto err_noclose;
1114 }
1115
1116 if (svsk->sk_datalen < 8) {
1117 svsk->sk_datalen = 0;
1118 goto err_delete; /* client is nuts. */
1119 }
1120
1121 rqstp->rq_arg.len = svsk->sk_datalen;
1122 rqstp->rq_arg.page_base = 0;
1123 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1124 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1125 rqstp->rq_arg.page_len = 0;
1126 } else
1127 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1128
1129 rqstp->rq_xprt_ctxt = NULL;
1130 rqstp->rq_prot = IPPROTO_TCP;
1131 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1132 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1133 else
1134 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1135
1136 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1137 calldir = p[1];
1138 if (calldir)
1139 len = receive_cb_reply(svsk, rqstp);
1140
1141 /* Reset TCP read info */
1142 svsk->sk_datalen = 0;
1143 svc_tcp_fragment_received(svsk);
1144
1145 if (len < 0)
1146 goto error;
1147
1148 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1149 if (serv->sv_stats)
1150 serv->sv_stats->nettcpcnt++;
1151
1152 return rqstp->rq_arg.len;
1153
1154error:
1155 if (len != -EAGAIN)
1156 goto err_delete;
1157 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1158 return 0;
1159err_delete:
1160 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1161 svsk->sk_xprt.xpt_server->sv_name, -len);
1162 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1163err_noclose:
1164 return 0; /* record not complete */
1165}
1166
1167/*
1168 * Send out data on TCP socket.
1169 */
1170static int svc_tcp_sendto(struct svc_rqst *rqstp)
1171{
1172 struct xdr_buf *xbufp = &rqstp->rq_res;
1173 int sent;
1174 __be32 reclen;
1175
1176 /* Set up the first element of the reply kvec.
1177 * Any other kvecs that may be in use have been taken
1178 * care of by the server implementation itself.
1179 */
1180 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1181 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1182
1183 sent = svc_sendto(rqstp, &rqstp->rq_res);
1184 if (sent != xbufp->len) {
1185 printk(KERN_NOTICE
1186 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1187 "- shutting down socket\n",
1188 rqstp->rq_xprt->xpt_server->sv_name,
1189 (sent<0)?"got error":"sent only",
1190 sent, xbufp->len);
1191 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1192 svc_xprt_enqueue(rqstp->rq_xprt);
1193 sent = -EAGAIN;
1194 }
1195 return sent;
1196}
1197
1198/*
1199 * Setup response header. TCP has a 4B record length field.
1200 */
1201static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1202{
1203 struct kvec *resv = &rqstp->rq_res.head[0];
1204
1205 /* tcp needs a space for the record length... */
1206 svc_putnl(resv, 0);
1207}
1208
1209static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1210 struct net *net,
1211 struct sockaddr *sa, int salen,
1212 int flags)
1213{
1214 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1215}
1216
1217#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1218static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1219 struct net *, struct sockaddr *,
1220 int, int);
1221static void svc_bc_sock_free(struct svc_xprt *xprt);
1222
1223static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1224 struct net *net,
1225 struct sockaddr *sa, int salen,
1226 int flags)
1227{
1228 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1229}
1230
1231static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1232{
1233}
1234
1235static const struct svc_xprt_ops svc_tcp_bc_ops = {
1236 .xpo_create = svc_bc_tcp_create,
1237 .xpo_detach = svc_bc_tcp_sock_detach,
1238 .xpo_free = svc_bc_sock_free,
1239 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1240 .xpo_secure_port = svc_sock_secure_port,
1241};
1242
1243static struct svc_xprt_class svc_tcp_bc_class = {
1244 .xcl_name = "tcp-bc",
1245 .xcl_owner = THIS_MODULE,
1246 .xcl_ops = &svc_tcp_bc_ops,
1247 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1248};
1249
1250static void svc_init_bc_xprt_sock(void)
1251{
1252 svc_reg_xprt_class(&svc_tcp_bc_class);
1253}
1254
1255static void svc_cleanup_bc_xprt_sock(void)
1256{
1257 svc_unreg_xprt_class(&svc_tcp_bc_class);
1258}
1259#else /* CONFIG_SUNRPC_BACKCHANNEL */
1260static void svc_init_bc_xprt_sock(void)
1261{
1262}
1263
1264static void svc_cleanup_bc_xprt_sock(void)
1265{
1266}
1267#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1268
1269static const struct svc_xprt_ops svc_tcp_ops = {
1270 .xpo_create = svc_tcp_create,
1271 .xpo_recvfrom = svc_tcp_recvfrom,
1272 .xpo_sendto = svc_tcp_sendto,
1273 .xpo_release_rqst = svc_release_skb,
1274 .xpo_detach = svc_tcp_sock_detach,
1275 .xpo_free = svc_sock_free,
1276 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1277 .xpo_has_wspace = svc_tcp_has_wspace,
1278 .xpo_accept = svc_tcp_accept,
1279 .xpo_secure_port = svc_sock_secure_port,
1280 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1281};
1282
1283static struct svc_xprt_class svc_tcp_class = {
1284 .xcl_name = "tcp",
1285 .xcl_owner = THIS_MODULE,
1286 .xcl_ops = &svc_tcp_ops,
1287 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1288 .xcl_ident = XPRT_TRANSPORT_TCP,
1289};
1290
1291void svc_init_xprt_sock(void)
1292{
1293 svc_reg_xprt_class(&svc_tcp_class);
1294 svc_reg_xprt_class(&svc_udp_class);
1295 svc_init_bc_xprt_sock();
1296}
1297
1298void svc_cleanup_xprt_sock(void)
1299{
1300 svc_unreg_xprt_class(&svc_tcp_class);
1301 svc_unreg_xprt_class(&svc_udp_class);
1302 svc_cleanup_bc_xprt_sock();
1303}
1304
1305static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1306{
1307 struct sock *sk = svsk->sk_sk;
1308
1309 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1310 &svsk->sk_xprt, serv);
1311 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1312 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1313 if (sk->sk_state == TCP_LISTEN) {
1314 dprintk("setting up TCP socket for listening\n");
1315 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1316 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1317 sk->sk_data_ready = svc_tcp_listen_data_ready;
1318 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1319 } else {
1320 dprintk("setting up TCP socket for reading\n");
1321 sk->sk_state_change = svc_tcp_state_change;
1322 sk->sk_data_ready = svc_data_ready;
1323 sk->sk_write_space = svc_write_space;
1324
1325 svsk->sk_reclen = 0;
1326 svsk->sk_tcplen = 0;
1327 svsk->sk_datalen = 0;
1328 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1329
1330 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1331
1332 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1333 switch (sk->sk_state) {
1334 case TCP_SYN_RECV:
1335 case TCP_ESTABLISHED:
1336 break;
1337 default:
1338 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1339 }
1340 }
1341}
1342
1343void svc_sock_update_bufs(struct svc_serv *serv)
1344{
1345 /*
1346 * The number of server threads has changed. Update
1347 * rcvbuf and sndbuf accordingly on all sockets
1348 */
1349 struct svc_sock *svsk;
1350
1351 spin_lock_bh(&serv->sv_lock);
1352 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1353 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1354 spin_unlock_bh(&serv->sv_lock);
1355}
1356EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1357
1358/*
1359 * Initialize socket for RPC use and create svc_sock struct
1360 */
1361static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1362 struct socket *sock,
1363 int flags)
1364{
1365 struct svc_sock *svsk;
1366 struct sock *inet;
1367 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1368 int err = 0;
1369
1370 dprintk("svc: svc_setup_socket %p\n", sock);
1371 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1372 if (!svsk)
1373 return ERR_PTR(-ENOMEM);
1374
1375 inet = sock->sk;
1376
1377 /* Register socket with portmapper */
1378 if (pmap_register)
1379 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1380 inet->sk_protocol,
1381 ntohs(inet_sk(inet)->inet_sport));
1382
1383 if (err < 0) {
1384 kfree(svsk);
1385 return ERR_PTR(err);
1386 }
1387
1388 svsk->sk_sock = sock;
1389 svsk->sk_sk = inet;
1390 svsk->sk_ostate = inet->sk_state_change;
1391 svsk->sk_odata = inet->sk_data_ready;
1392 svsk->sk_owspace = inet->sk_write_space;
1393 /*
1394 * This barrier is necessary in order to prevent race condition
1395 * with svc_data_ready(), svc_listen_data_ready() and others
1396 * when calling callbacks above.
1397 */
1398 wmb();
1399 inet->sk_user_data = svsk;
1400
1401 /* Initialize the socket */
1402 if (sock->type == SOCK_DGRAM)
1403 svc_udp_init(svsk, serv);
1404 else
1405 svc_tcp_init(svsk, serv);
1406
1407 dprintk("svc: svc_setup_socket created %p (inet %p), "
1408 "listen %d close %d\n",
1409 svsk, svsk->sk_sk,
1410 test_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags),
1411 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1412
1413 return svsk;
1414}
1415
1416bool svc_alien_sock(struct net *net, int fd)
1417{
1418 int err;
1419 struct socket *sock = sockfd_lookup(fd, &err);
1420 bool ret = false;
1421
1422 if (!sock)
1423 goto out;
1424 if (sock_net(sock->sk) != net)
1425 ret = true;
1426 sockfd_put(sock);
1427out:
1428 return ret;
1429}
1430EXPORT_SYMBOL_GPL(svc_alien_sock);
1431
1432/**
1433 * svc_addsock - add a listener socket to an RPC service
1434 * @serv: pointer to RPC service to which to add a new listener
1435 * @fd: file descriptor of the new listener
1436 * @name_return: pointer to buffer to fill in with name of listener
1437 * @len: size of the buffer
1438 *
1439 * Fills in socket name and returns positive length of name if successful.
1440 * Name is terminated with '\n'. On error, returns a negative errno
1441 * value.
1442 */
1443int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1444 const size_t len)
1445{
1446 int err = 0;
1447 struct socket *so = sockfd_lookup(fd, &err);
1448 struct svc_sock *svsk = NULL;
1449 struct sockaddr_storage addr;
1450 struct sockaddr *sin = (struct sockaddr *)&addr;
1451 int salen;
1452
1453 if (!so)
1454 return err;
1455 err = -EAFNOSUPPORT;
1456 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1457 goto out;
1458 err = -EPROTONOSUPPORT;
1459 if (so->sk->sk_protocol != IPPROTO_TCP &&
1460 so->sk->sk_protocol != IPPROTO_UDP)
1461 goto out;
1462 err = -EISCONN;
1463 if (so->state > SS_UNCONNECTED)
1464 goto out;
1465 err = -ENOENT;
1466 if (!try_module_get(THIS_MODULE))
1467 goto out;
1468 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1469 if (IS_ERR(svsk)) {
1470 module_put(THIS_MODULE);
1471 err = PTR_ERR(svsk);
1472 goto out;
1473 }
1474 salen = kernel_getsockname(svsk->sk_sock, sin);
1475 if (salen >= 0)
1476 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1477 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1478 return svc_one_sock_name(svsk, name_return, len);
1479out:
1480 sockfd_put(so);
1481 return err;
1482}
1483EXPORT_SYMBOL_GPL(svc_addsock);
1484
1485/*
1486 * Create socket for RPC service.
1487 */
1488static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1489 int protocol,
1490 struct net *net,
1491 struct sockaddr *sin, int len,
1492 int flags)
1493{
1494 struct svc_sock *svsk;
1495 struct socket *sock;
1496 int error;
1497 int type;
1498 struct sockaddr_storage addr;
1499 struct sockaddr *newsin = (struct sockaddr *)&addr;
1500 int newlen;
1501 int family;
1502 int val;
1503 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1504
1505 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1506 serv->sv_program->pg_name, protocol,
1507 __svc_print_addr(sin, buf, sizeof(buf)));
1508
1509 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1510 printk(KERN_WARNING "svc: only UDP and TCP "
1511 "sockets supported\n");
1512 return ERR_PTR(-EINVAL);
1513 }
1514
1515 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1516 switch (sin->sa_family) {
1517 case AF_INET6:
1518 family = PF_INET6;
1519 break;
1520 case AF_INET:
1521 family = PF_INET;
1522 break;
1523 default:
1524 return ERR_PTR(-EINVAL);
1525 }
1526
1527 error = __sock_create(net, family, type, protocol, &sock, 1);
1528 if (error < 0)
1529 return ERR_PTR(error);
1530
1531 svc_reclassify_socket(sock);
1532
1533 /*
1534 * If this is an PF_INET6 listener, we want to avoid
1535 * getting requests from IPv4 remotes. Those should
1536 * be shunted to a PF_INET listener via rpcbind.
1537 */
1538 val = 1;
1539 if (family == PF_INET6)
1540 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1541 (char *)&val, sizeof(val));
1542
1543 if (type == SOCK_STREAM)
1544 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1545 error = kernel_bind(sock, sin, len);
1546 if (error < 0)
1547 goto bummer;
1548
1549 error = kernel_getsockname(sock, newsin);
1550 if (error < 0)
1551 goto bummer;
1552 newlen = error;
1553
1554 if (protocol == IPPROTO_TCP) {
1555 if ((error = kernel_listen(sock, 64)) < 0)
1556 goto bummer;
1557 }
1558
1559 svsk = svc_setup_socket(serv, sock, flags);
1560 if (IS_ERR(svsk)) {
1561 error = PTR_ERR(svsk);
1562 goto bummer;
1563 }
1564 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1565 return (struct svc_xprt *)svsk;
1566bummer:
1567 dprintk("svc: svc_create_socket error = %d\n", -error);
1568 sock_release(sock);
1569 return ERR_PTR(error);
1570}
1571
1572/*
1573 * Detach the svc_sock from the socket so that no
1574 * more callbacks occur.
1575 */
1576static void svc_sock_detach(struct svc_xprt *xprt)
1577{
1578 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1579 struct sock *sk = svsk->sk_sk;
1580
1581 dprintk("svc: svc_sock_detach(%p)\n", svsk);
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 */
1595static 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 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
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 */
1612static void svc_sock_free(struct svc_xprt *xprt)
1613{
1614 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1615 dprintk("svc: svc_sock_free(%p)\n", svsk);
1616
1617 if (svsk->sk_sock->file)
1618 sockfd_put(svsk->sk_sock);
1619 else
1620 sock_release(svsk->sk_sock);
1621 kfree(svsk);
1622}
1623
1624#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1625/*
1626 * Create a back channel svc_xprt which shares the fore channel socket.
1627 */
1628static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1629 int protocol,
1630 struct net *net,
1631 struct sockaddr *sin, int len,
1632 int flags)
1633{
1634 struct svc_sock *svsk;
1635 struct svc_xprt *xprt;
1636
1637 if (protocol != IPPROTO_TCP) {
1638 printk(KERN_WARNING "svc: only TCP sockets"
1639 " supported on shared back channel\n");
1640 return ERR_PTR(-EINVAL);
1641 }
1642
1643 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1644 if (!svsk)
1645 return ERR_PTR(-ENOMEM);
1646
1647 xprt = &svsk->sk_xprt;
1648 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1649 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1650
1651 serv->sv_bc_xprt = xprt;
1652
1653 return xprt;
1654}
1655
1656/*
1657 * Free a back channel svc_sock.
1658 */
1659static void svc_bc_sock_free(struct svc_xprt *xprt)
1660{
1661 if (xprt)
1662 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1663}
1664#endif /* CONFIG_SUNRPC_BACKCHANNEL */