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/tcp.h>
43#include <net/tcp_states.h>
44#include <asm/uaccess.h>
45#include <asm/ioctls.h>
46#include <trace/events/skb.h>
47
48#include <linux/sunrpc/types.h>
49#include <linux/sunrpc/clnt.h>
50#include <linux/sunrpc/xdr.h>
51#include <linux/sunrpc/msg_prot.h>
52#include <linux/sunrpc/svcsock.h>
53#include <linux/sunrpc/stats.h>
54#include <linux/sunrpc/xprt.h>
55
56#include "sunrpc.h"
57
58#define RPCDBG_FACILITY RPCDBG_SVCXPRT
59
60
61static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
62 int flags);
63static void svc_udp_data_ready(struct sock *);
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 WARN_ON_ONCE(sock_owned_by_user(sk));
89 if (sock_owned_by_user(sk))
90 return;
91
92 switch (sk->sk_family) {
93 case AF_INET:
94 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
95 &svc_slock_key[0],
96 "sk_xprt.xpt_lock-AF_INET-NFSD",
97 &svc_key[0]);
98 break;
99
100 case AF_INET6:
101 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
102 &svc_slock_key[1],
103 "sk_xprt.xpt_lock-AF_INET6-NFSD",
104 &svc_key[1]);
105 break;
106
107 default:
108 BUG();
109 }
110}
111#else
112static void svc_reclassify_socket(struct socket *sock)
113{
114}
115#endif
116
117/*
118 * Release an skbuff after use
119 */
120static void svc_release_skb(struct svc_rqst *rqstp)
121{
122 struct sk_buff *skb = rqstp->rq_xprt_ctxt;
123
124 if (skb) {
125 struct svc_sock *svsk =
126 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
127 rqstp->rq_xprt_ctxt = NULL;
128
129 dprintk("svc: service %p, releasing skb %p\n", rqstp, skb);
130 skb_free_datagram_locked(svsk->sk_sk, skb);
131 }
132}
133
134union svc_pktinfo_u {
135 struct in_pktinfo pkti;
136 struct in6_pktinfo pkti6;
137};
138#define SVC_PKTINFO_SPACE \
139 CMSG_SPACE(sizeof(union svc_pktinfo_u))
140
141static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
142{
143 struct svc_sock *svsk =
144 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
145 switch (svsk->sk_sk->sk_family) {
146 case AF_INET: {
147 struct in_pktinfo *pki = CMSG_DATA(cmh);
148
149 cmh->cmsg_level = SOL_IP;
150 cmh->cmsg_type = IP_PKTINFO;
151 pki->ipi_ifindex = 0;
152 pki->ipi_spec_dst.s_addr =
153 svc_daddr_in(rqstp)->sin_addr.s_addr;
154 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
155 }
156 break;
157
158 case AF_INET6: {
159 struct in6_pktinfo *pki = CMSG_DATA(cmh);
160 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
161
162 cmh->cmsg_level = SOL_IPV6;
163 cmh->cmsg_type = IPV6_PKTINFO;
164 pki->ipi6_ifindex = daddr->sin6_scope_id;
165 pki->ipi6_addr = daddr->sin6_addr;
166 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
167 }
168 break;
169 }
170}
171
172/*
173 * send routine intended to be shared by the fore- and back-channel
174 */
175int svc_send_common(struct socket *sock, struct xdr_buf *xdr,
176 struct page *headpage, unsigned long headoffset,
177 struct page *tailpage, unsigned long tailoffset)
178{
179 int result;
180 int size;
181 struct page **ppage = xdr->pages;
182 size_t base = xdr->page_base;
183 unsigned int pglen = xdr->page_len;
184 unsigned int flags = MSG_MORE | MSG_SENDPAGE_NOTLAST;
185 int slen;
186 int len = 0;
187
188 slen = xdr->len;
189
190 /* send head */
191 if (slen == xdr->head[0].iov_len)
192 flags = 0;
193 len = kernel_sendpage(sock, headpage, headoffset,
194 xdr->head[0].iov_len, flags);
195 if (len != xdr->head[0].iov_len)
196 goto out;
197 slen -= xdr->head[0].iov_len;
198 if (slen == 0)
199 goto out;
200
201 /* send page data */
202 size = PAGE_SIZE - base < pglen ? PAGE_SIZE - base : pglen;
203 while (pglen > 0) {
204 if (slen == size)
205 flags = 0;
206 result = kernel_sendpage(sock, *ppage, base, size, flags);
207 if (result > 0)
208 len += result;
209 if (result != size)
210 goto out;
211 slen -= size;
212 pglen -= size;
213 size = PAGE_SIZE < pglen ? PAGE_SIZE : pglen;
214 base = 0;
215 ppage++;
216 }
217
218 /* send tail */
219 if (xdr->tail[0].iov_len) {
220 result = kernel_sendpage(sock, tailpage, tailoffset,
221 xdr->tail[0].iov_len, 0);
222 if (result > 0)
223 len += result;
224 }
225
226out:
227 return len;
228}
229
230
231/*
232 * Generic sendto routine
233 */
234static int svc_sendto(struct svc_rqst *rqstp, struct xdr_buf *xdr)
235{
236 struct svc_sock *svsk =
237 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
238 struct socket *sock = svsk->sk_sock;
239 union {
240 struct cmsghdr hdr;
241 long all[SVC_PKTINFO_SPACE / sizeof(long)];
242 } buffer;
243 struct cmsghdr *cmh = &buffer.hdr;
244 int len = 0;
245 unsigned long tailoff;
246 unsigned long headoff;
247 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
248
249 if (rqstp->rq_prot == IPPROTO_UDP) {
250 struct msghdr msg = {
251 .msg_name = &rqstp->rq_addr,
252 .msg_namelen = rqstp->rq_addrlen,
253 .msg_control = cmh,
254 .msg_controllen = sizeof(buffer),
255 .msg_flags = MSG_MORE,
256 };
257
258 svc_set_cmsg_data(rqstp, cmh);
259
260 if (sock_sendmsg(sock, &msg) < 0)
261 goto out;
262 }
263
264 tailoff = ((unsigned long)xdr->tail[0].iov_base) & (PAGE_SIZE-1);
265 headoff = 0;
266 len = svc_send_common(sock, xdr, rqstp->rq_respages[0], headoff,
267 rqstp->rq_respages[0], tailoff);
268
269out:
270 dprintk("svc: socket %p sendto([%p %Zu... ], %d) = %d (addr %s)\n",
271 svsk, xdr->head[0].iov_base, xdr->head[0].iov_len,
272 xdr->len, len, svc_print_addr(rqstp, buf, sizeof(buf)));
273
274 return len;
275}
276
277/*
278 * Report socket names for nfsdfs
279 */
280static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
281{
282 const struct sock *sk = svsk->sk_sk;
283 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
284 "udp" : "tcp";
285 int len;
286
287 switch (sk->sk_family) {
288 case PF_INET:
289 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
290 proto_name,
291 &inet_sk(sk)->inet_rcv_saddr,
292 inet_sk(sk)->inet_num);
293 break;
294#if IS_ENABLED(CONFIG_IPV6)
295 case PF_INET6:
296 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
297 proto_name,
298 &sk->sk_v6_rcv_saddr,
299 inet_sk(sk)->inet_num);
300 break;
301#endif
302 default:
303 len = snprintf(buf, remaining, "*unknown-%d*\n",
304 sk->sk_family);
305 }
306
307 if (len >= remaining) {
308 *buf = '\0';
309 return -ENAMETOOLONG;
310 }
311 return len;
312}
313
314/*
315 * Generic recvfrom routine.
316 */
317static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
318 int buflen)
319{
320 struct svc_sock *svsk =
321 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
322 struct msghdr msg = {
323 .msg_flags = MSG_DONTWAIT,
324 };
325 int len;
326
327 rqstp->rq_xprt_hlen = 0;
328
329 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
330 len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
331 msg.msg_flags);
332 /* If we read a full record, then assume there may be more
333 * data to read (stream based sockets only!)
334 */
335 if (len == buflen)
336 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
337
338 dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
339 svsk, iov[0].iov_base, iov[0].iov_len, len);
340 return len;
341}
342
343static int svc_partial_recvfrom(struct svc_rqst *rqstp,
344 struct kvec *iov, int nr,
345 int buflen, unsigned int base)
346{
347 size_t save_iovlen;
348 void *save_iovbase;
349 unsigned int i;
350 int ret;
351
352 if (base == 0)
353 return svc_recvfrom(rqstp, iov, nr, buflen);
354
355 for (i = 0; i < nr; i++) {
356 if (iov[i].iov_len > base)
357 break;
358 base -= iov[i].iov_len;
359 }
360 save_iovlen = iov[i].iov_len;
361 save_iovbase = iov[i].iov_base;
362 iov[i].iov_len -= base;
363 iov[i].iov_base += base;
364 ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
365 iov[i].iov_len = save_iovlen;
366 iov[i].iov_base = save_iovbase;
367 return ret;
368}
369
370/*
371 * Set socket snd and rcv buffer lengths
372 */
373static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
374 unsigned int rcv)
375{
376#if 0
377 mm_segment_t oldfs;
378 oldfs = get_fs(); set_fs(KERNEL_DS);
379 sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
380 (char*)&snd, sizeof(snd));
381 sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
382 (char*)&rcv, sizeof(rcv));
383#else
384 /* sock_setsockopt limits use to sysctl_?mem_max,
385 * which isn't acceptable. Until that is made conditional
386 * on not having CAP_SYS_RESOURCE or similar, we go direct...
387 * DaveM said I could!
388 */
389 lock_sock(sock->sk);
390 sock->sk->sk_sndbuf = snd * 2;
391 sock->sk->sk_rcvbuf = rcv * 2;
392 sock->sk->sk_write_space(sock->sk);
393 release_sock(sock->sk);
394#endif
395}
396
397static int svc_sock_secure_port(struct svc_rqst *rqstp)
398{
399 return svc_port_is_privileged(svc_addr(rqstp));
400}
401
402static bool sunrpc_waitqueue_active(wait_queue_head_t *wq)
403{
404 if (!wq)
405 return false;
406 /*
407 * There should normally be a memory * barrier here--see
408 * wq_has_sleeper().
409 *
410 * It appears that isn't currently necessary, though, basically
411 * because callers all appear to have sufficient memory barriers
412 * between the time the relevant change is made and the
413 * time they call these callbacks.
414 *
415 * The nfsd code itself doesn't actually explicitly wait on
416 * these waitqueues, but it may wait on them for example in
417 * sendpage() or sendmsg() calls. (And those may be the only
418 * places, since it it uses nonblocking reads.)
419 *
420 * Maybe we should add the memory barriers anyway, but these are
421 * hot paths so we'd need to be convinced there's no sigificant
422 * penalty.
423 */
424 return waitqueue_active(wq);
425}
426
427/*
428 * INET callback when data has been received on the socket.
429 */
430static void svc_udp_data_ready(struct sock *sk)
431{
432 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
433 wait_queue_head_t *wq = sk_sleep(sk);
434
435 if (svsk) {
436 dprintk("svc: socket %p(inet %p), busy=%d\n",
437 svsk, sk,
438 test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
439 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
440 svc_xprt_enqueue(&svsk->sk_xprt);
441 }
442 if (sunrpc_waitqueue_active(wq))
443 wake_up_interruptible(wq);
444}
445
446/*
447 * INET callback when space is newly available on the socket.
448 */
449static void svc_write_space(struct sock *sk)
450{
451 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
452 wait_queue_head_t *wq = sk_sleep(sk);
453
454 if (svsk) {
455 dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
456 svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
457 svc_xprt_enqueue(&svsk->sk_xprt);
458 }
459
460 if (sunrpc_waitqueue_active(wq)) {
461 dprintk("RPC svc_write_space: someone sleeping on %p\n",
462 svsk);
463 wake_up_interruptible(wq);
464 }
465}
466
467static int svc_tcp_has_wspace(struct svc_xprt *xprt)
468{
469 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
470 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
471 int required;
472
473 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
474 return 1;
475 required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
476 if (sk_stream_wspace(svsk->sk_sk) >= required ||
477 (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
478 atomic_read(&xprt->xpt_reserved) == 0))
479 return 1;
480 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
481 return 0;
482}
483
484static void svc_tcp_write_space(struct sock *sk)
485{
486 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
487 struct socket *sock = sk->sk_socket;
488
489 if (!sk_stream_is_writeable(sk) || !sock)
490 return;
491 if (!svsk || svc_tcp_has_wspace(&svsk->sk_xprt))
492 clear_bit(SOCK_NOSPACE, &sock->flags);
493 svc_write_space(sk);
494}
495
496static void svc_tcp_adjust_wspace(struct svc_xprt *xprt)
497{
498 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
499
500 if (svc_tcp_has_wspace(xprt))
501 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
502}
503
504/*
505 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
506 */
507static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
508 struct cmsghdr *cmh)
509{
510 struct in_pktinfo *pki = CMSG_DATA(cmh);
511 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
512
513 if (cmh->cmsg_type != IP_PKTINFO)
514 return 0;
515
516 daddr->sin_family = AF_INET;
517 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
518 return 1;
519}
520
521/*
522 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
523 */
524static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
525 struct cmsghdr *cmh)
526{
527 struct in6_pktinfo *pki = CMSG_DATA(cmh);
528 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
529
530 if (cmh->cmsg_type != IPV6_PKTINFO)
531 return 0;
532
533 daddr->sin6_family = AF_INET6;
534 daddr->sin6_addr = pki->ipi6_addr;
535 daddr->sin6_scope_id = pki->ipi6_ifindex;
536 return 1;
537}
538
539/*
540 * Copy the UDP datagram's destination address to the rqstp structure.
541 * The 'destination' address in this case is the address to which the
542 * peer sent the datagram, i.e. our local address. For multihomed
543 * hosts, this can change from msg to msg. Note that only the IP
544 * address changes, the port number should remain the same.
545 */
546static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
547 struct cmsghdr *cmh)
548{
549 switch (cmh->cmsg_level) {
550 case SOL_IP:
551 return svc_udp_get_dest_address4(rqstp, cmh);
552 case SOL_IPV6:
553 return svc_udp_get_dest_address6(rqstp, cmh);
554 }
555
556 return 0;
557}
558
559/*
560 * Receive a datagram from a UDP socket.
561 */
562static int svc_udp_recvfrom(struct svc_rqst *rqstp)
563{
564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
567 struct sk_buff *skb;
568 union {
569 struct cmsghdr hdr;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
571 } buffer;
572 struct cmsghdr *cmh = &buffer.hdr;
573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
575 .msg_control = cmh,
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
578 };
579 size_t len;
580 int err;
581
582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
590 */
591 svc_sock_setbufsize(svsk->sk_sock,
592 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
593 (serv->sv_nrthreads+3) * serv->sv_max_mesg);
594
595 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
596 skb = NULL;
597 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
598 0, 0, MSG_PEEK | MSG_DONTWAIT);
599 if (err >= 0)
600 skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
601
602 if (skb == NULL) {
603 if (err != -EAGAIN) {
604 /* possibly an icmp error */
605 dprintk("svc: recvfrom returned error %d\n", -err);
606 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
607 }
608 return 0;
609 }
610 len = svc_addr_len(svc_addr(rqstp));
611 rqstp->rq_addrlen = len;
612 if (skb->tstamp.tv64 == 0) {
613 skb->tstamp = ktime_get_real();
614 /* Don't enable netstamp, sunrpc doesn't
615 need that much accuracy */
616 }
617 svsk->sk_sk->sk_stamp = skb->tstamp;
618 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
619
620 len = skb->len - sizeof(struct udphdr);
621 rqstp->rq_arg.len = len;
622
623 rqstp->rq_prot = IPPROTO_UDP;
624
625 if (!svc_udp_get_dest_address(rqstp, cmh)) {
626 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
627 cmh->cmsg_level, cmh->cmsg_type);
628 goto out_free;
629 }
630 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
631
632 if (skb_is_nonlinear(skb)) {
633 /* we have to copy */
634 local_bh_disable();
635 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
636 local_bh_enable();
637 /* checksum error */
638 goto out_free;
639 }
640 local_bh_enable();
641 skb_free_datagram_locked(svsk->sk_sk, skb);
642 } else {
643 /* we can use it in-place */
644 rqstp->rq_arg.head[0].iov_base = skb->data +
645 sizeof(struct udphdr);
646 rqstp->rq_arg.head[0].iov_len = len;
647 if (skb_checksum_complete(skb))
648 goto out_free;
649 rqstp->rq_xprt_ctxt = skb;
650 }
651
652 rqstp->rq_arg.page_base = 0;
653 if (len <= rqstp->rq_arg.head[0].iov_len) {
654 rqstp->rq_arg.head[0].iov_len = len;
655 rqstp->rq_arg.page_len = 0;
656 rqstp->rq_respages = rqstp->rq_pages+1;
657 } else {
658 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
659 rqstp->rq_respages = rqstp->rq_pages + 1 +
660 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
661 }
662 rqstp->rq_next_page = rqstp->rq_respages+1;
663
664 if (serv->sv_stats)
665 serv->sv_stats->netudpcnt++;
666
667 return len;
668out_free:
669 trace_kfree_skb(skb, svc_udp_recvfrom);
670 skb_free_datagram_locked(svsk->sk_sk, skb);
671 return 0;
672}
673
674static int
675svc_udp_sendto(struct svc_rqst *rqstp)
676{
677 int error;
678
679 error = svc_sendto(rqstp, &rqstp->rq_res);
680 if (error == -ECONNREFUSED)
681 /* ICMP error on earlier request. */
682 error = svc_sendto(rqstp, &rqstp->rq_res);
683
684 return error;
685}
686
687static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
688{
689}
690
691static int svc_udp_has_wspace(struct svc_xprt *xprt)
692{
693 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
694 struct svc_serv *serv = xprt->xpt_server;
695 unsigned long required;
696
697 /*
698 * Set the SOCK_NOSPACE flag before checking the available
699 * sock space.
700 */
701 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
702 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
703 if (required*2 > sock_wspace(svsk->sk_sk))
704 return 0;
705 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
706 return 1;
707}
708
709static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
710{
711 BUG();
712 return NULL;
713}
714
715static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
716 struct net *net,
717 struct sockaddr *sa, int salen,
718 int flags)
719{
720 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
721}
722
723static struct svc_xprt_ops svc_udp_ops = {
724 .xpo_create = svc_udp_create,
725 .xpo_recvfrom = svc_udp_recvfrom,
726 .xpo_sendto = svc_udp_sendto,
727 .xpo_release_rqst = svc_release_skb,
728 .xpo_detach = svc_sock_detach,
729 .xpo_free = svc_sock_free,
730 .xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
731 .xpo_has_wspace = svc_udp_has_wspace,
732 .xpo_accept = svc_udp_accept,
733 .xpo_secure_port = svc_sock_secure_port,
734};
735
736static struct svc_xprt_class svc_udp_class = {
737 .xcl_name = "udp",
738 .xcl_owner = THIS_MODULE,
739 .xcl_ops = &svc_udp_ops,
740 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
741 .xcl_ident = XPRT_TRANSPORT_UDP,
742};
743
744static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
745{
746 int err, level, optname, one = 1;
747
748 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
749 &svsk->sk_xprt, serv);
750 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
751 svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
752 svsk->sk_sk->sk_write_space = svc_write_space;
753
754 /* initialise setting must have enough space to
755 * receive and respond to one request.
756 * svc_udp_recvfrom will re-adjust if necessary
757 */
758 svc_sock_setbufsize(svsk->sk_sock,
759 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
760 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
761
762 /* data might have come in before data_ready set up */
763 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
764 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
765
766 /* make sure we get destination address info */
767 switch (svsk->sk_sk->sk_family) {
768 case AF_INET:
769 level = SOL_IP;
770 optname = IP_PKTINFO;
771 break;
772 case AF_INET6:
773 level = SOL_IPV6;
774 optname = IPV6_RECVPKTINFO;
775 break;
776 default:
777 BUG();
778 }
779 err = kernel_setsockopt(svsk->sk_sock, level, optname,
780 (char *)&one, sizeof(one));
781 dprintk("svc: kernel_setsockopt returned %d\n", err);
782}
783
784/*
785 * A data_ready event on a listening socket means there's a connection
786 * pending. Do not use state_change as a substitute for it.
787 */
788static void svc_tcp_listen_data_ready(struct sock *sk)
789{
790 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
791 wait_queue_head_t *wq;
792
793 dprintk("svc: socket %p TCP (listen) state change %d\n",
794 sk, sk->sk_state);
795
796 /*
797 * This callback may called twice when a new connection
798 * is established as a child socket inherits everything
799 * from a parent LISTEN socket.
800 * 1) data_ready method of the parent socket will be called
801 * when one of child sockets become ESTABLISHED.
802 * 2) data_ready method of the child socket may be called
803 * when it receives data before the socket is accepted.
804 * In case of 2, we should ignore it silently.
805 */
806 if (sk->sk_state == TCP_LISTEN) {
807 if (svsk) {
808 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
809 svc_xprt_enqueue(&svsk->sk_xprt);
810 } else
811 printk("svc: socket %p: no user data\n", sk);
812 }
813
814 wq = sk_sleep(sk);
815 if (sunrpc_waitqueue_active(wq))
816 wake_up_interruptible_all(wq);
817}
818
819/*
820 * A state change on a connected socket means it's dying or dead.
821 */
822static void svc_tcp_state_change(struct sock *sk)
823{
824 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
825 wait_queue_head_t *wq = sk_sleep(sk);
826
827 dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
828 sk, sk->sk_state, sk->sk_user_data);
829
830 if (!svsk)
831 printk("svc: socket %p: no user data\n", sk);
832 else {
833 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
834 svc_xprt_enqueue(&svsk->sk_xprt);
835 }
836 if (sunrpc_waitqueue_active(wq))
837 wake_up_interruptible_all(wq);
838}
839
840static void svc_tcp_data_ready(struct sock *sk)
841{
842 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
843 wait_queue_head_t *wq = sk_sleep(sk);
844
845 dprintk("svc: socket %p TCP data ready (svsk %p)\n",
846 sk, sk->sk_user_data);
847 if (svsk) {
848 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
849 svc_xprt_enqueue(&svsk->sk_xprt);
850 }
851 if (sunrpc_waitqueue_active(wq))
852 wake_up_interruptible(wq);
853}
854
855/*
856 * Accept a TCP connection
857 */
858static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
859{
860 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
861 struct sockaddr_storage addr;
862 struct sockaddr *sin = (struct sockaddr *) &addr;
863 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
864 struct socket *sock = svsk->sk_sock;
865 struct socket *newsock;
866 struct svc_sock *newsvsk;
867 int err, slen;
868 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
869
870 dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
871 if (!sock)
872 return NULL;
873
874 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
875 err = kernel_accept(sock, &newsock, O_NONBLOCK);
876 if (err < 0) {
877 if (err == -ENOMEM)
878 printk(KERN_WARNING "%s: no more sockets!\n",
879 serv->sv_name);
880 else if (err != -EAGAIN)
881 net_warn_ratelimited("%s: accept failed (err %d)!\n",
882 serv->sv_name, -err);
883 return NULL;
884 }
885 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
886
887 err = kernel_getpeername(newsock, sin, &slen);
888 if (err < 0) {
889 net_warn_ratelimited("%s: peername failed (err %d)!\n",
890 serv->sv_name, -err);
891 goto failed; /* aborted connection or whatever */
892 }
893
894 /* Ideally, we would want to reject connections from unauthorized
895 * hosts here, but when we get encryption, the IP of the host won't
896 * tell us anything. For now just warn about unpriv connections.
897 */
898 if (!svc_port_is_privileged(sin)) {
899 dprintk("%s: connect from unprivileged port: %s\n",
900 serv->sv_name,
901 __svc_print_addr(sin, buf, sizeof(buf)));
902 }
903 dprintk("%s: connect from %s\n", serv->sv_name,
904 __svc_print_addr(sin, buf, sizeof(buf)));
905
906 /* make sure that a write doesn't block forever when
907 * low on memory
908 */
909 newsock->sk->sk_sndtimeo = HZ*30;
910
911 newsvsk = svc_setup_socket(serv, newsock,
912 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
913 if (IS_ERR(newsvsk))
914 goto failed;
915 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
916 err = kernel_getsockname(newsock, sin, &slen);
917 if (unlikely(err < 0)) {
918 dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
919 slen = offsetof(struct sockaddr, sa_data);
920 }
921 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
922
923 if (sock_is_loopback(newsock->sk))
924 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
925 else
926 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
927 if (serv->sv_stats)
928 serv->sv_stats->nettcpconn++;
929
930 return &newsvsk->sk_xprt;
931
932failed:
933 sock_release(newsock);
934 return NULL;
935}
936
937static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
938{
939 unsigned int i, len, npages;
940
941 if (svsk->sk_datalen == 0)
942 return 0;
943 len = svsk->sk_datalen;
944 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
945 for (i = 0; i < npages; i++) {
946 if (rqstp->rq_pages[i] != NULL)
947 put_page(rqstp->rq_pages[i]);
948 BUG_ON(svsk->sk_pages[i] == NULL);
949 rqstp->rq_pages[i] = svsk->sk_pages[i];
950 svsk->sk_pages[i] = NULL;
951 }
952 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
953 return len;
954}
955
956static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
957{
958 unsigned int i, len, npages;
959
960 if (svsk->sk_datalen == 0)
961 return;
962 len = svsk->sk_datalen;
963 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
964 for (i = 0; i < npages; i++) {
965 svsk->sk_pages[i] = rqstp->rq_pages[i];
966 rqstp->rq_pages[i] = NULL;
967 }
968}
969
970static void svc_tcp_clear_pages(struct svc_sock *svsk)
971{
972 unsigned int i, len, npages;
973
974 if (svsk->sk_datalen == 0)
975 goto out;
976 len = svsk->sk_datalen;
977 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
978 for (i = 0; i < npages; i++) {
979 if (svsk->sk_pages[i] == NULL) {
980 WARN_ON_ONCE(1);
981 continue;
982 }
983 put_page(svsk->sk_pages[i]);
984 svsk->sk_pages[i] = NULL;
985 }
986out:
987 svsk->sk_tcplen = 0;
988 svsk->sk_datalen = 0;
989}
990
991/*
992 * Receive fragment record header.
993 * If we haven't gotten the record length yet, get the next four bytes.
994 */
995static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
996{
997 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
998 unsigned int want;
999 int len;
1000
1001 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1002 struct kvec iov;
1003
1004 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1005 iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
1006 iov.iov_len = want;
1007 if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
1008 goto error;
1009 svsk->sk_tcplen += len;
1010
1011 if (len < want) {
1012 dprintk("svc: short recvfrom while reading record "
1013 "length (%d of %d)\n", len, want);
1014 return -EAGAIN;
1015 }
1016
1017 dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
1018 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1019 serv->sv_max_mesg) {
1020 net_notice_ratelimited("RPC: fragment too large: %d\n",
1021 svc_sock_reclen(svsk));
1022 goto err_delete;
1023 }
1024 }
1025
1026 return svc_sock_reclen(svsk);
1027error:
1028 dprintk("RPC: TCP recv_record got %d\n", len);
1029 return len;
1030err_delete:
1031 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1032 return -EAGAIN;
1033}
1034
1035static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1036{
1037 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1038 struct rpc_rqst *req = NULL;
1039 struct kvec *src, *dst;
1040 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1041 __be32 xid;
1042 __be32 calldir;
1043
1044 xid = *p++;
1045 calldir = *p;
1046
1047 if (!bc_xprt)
1048 return -EAGAIN;
1049 spin_lock_bh(&bc_xprt->transport_lock);
1050 req = xprt_lookup_rqst(bc_xprt, xid);
1051 if (!req)
1052 goto unlock_notfound;
1053
1054 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1055 /*
1056 * XXX!: cheating for now! Only copying HEAD.
1057 * But we know this is good enough for now (in fact, for any
1058 * callback reply in the forseeable future).
1059 */
1060 dst = &req->rq_private_buf.head[0];
1061 src = &rqstp->rq_arg.head[0];
1062 if (dst->iov_len < src->iov_len)
1063 goto unlock_eagain; /* whatever; just giving up. */
1064 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1065 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1066 rqstp->rq_arg.len = 0;
1067 spin_unlock_bh(&bc_xprt->transport_lock);
1068 return 0;
1069unlock_notfound:
1070 printk(KERN_NOTICE
1071 "%s: Got unrecognized reply: "
1072 "calldir 0x%x xpt_bc_xprt %p xid %08x\n",
1073 __func__, ntohl(calldir),
1074 bc_xprt, ntohl(xid));
1075unlock_eagain:
1076 spin_unlock_bh(&bc_xprt->transport_lock);
1077 return -EAGAIN;
1078}
1079
1080static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1081{
1082 int i = 0;
1083 int t = 0;
1084
1085 while (t < len) {
1086 vec[i].iov_base = page_address(pages[i]);
1087 vec[i].iov_len = PAGE_SIZE;
1088 i++;
1089 t += PAGE_SIZE;
1090 }
1091 return i;
1092}
1093
1094static void svc_tcp_fragment_received(struct svc_sock *svsk)
1095{
1096 /* If we have more data, signal svc_xprt_enqueue() to try again */
1097 dprintk("svc: TCP %s record (%d bytes)\n",
1098 svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1099 svc_sock_reclen(svsk));
1100 svsk->sk_tcplen = 0;
1101 svsk->sk_reclen = 0;
1102}
1103
1104/*
1105 * Receive data from a TCP socket.
1106 */
1107static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1108{
1109 struct svc_sock *svsk =
1110 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1111 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1112 int len;
1113 struct kvec *vec;
1114 unsigned int want, base;
1115 __be32 *p;
1116 __be32 calldir;
1117 int pnum;
1118
1119 dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1120 svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1121 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1122 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1123
1124 len = svc_tcp_recv_record(svsk, rqstp);
1125 if (len < 0)
1126 goto error;
1127
1128 base = svc_tcp_restore_pages(svsk, rqstp);
1129 want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1130
1131 vec = rqstp->rq_vec;
1132
1133 pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1134 svsk->sk_datalen + want);
1135
1136 rqstp->rq_respages = &rqstp->rq_pages[pnum];
1137 rqstp->rq_next_page = rqstp->rq_respages + 1;
1138
1139 /* Now receive data */
1140 len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1141 if (len >= 0) {
1142 svsk->sk_tcplen += len;
1143 svsk->sk_datalen += len;
1144 }
1145 if (len != want || !svc_sock_final_rec(svsk)) {
1146 svc_tcp_save_pages(svsk, rqstp);
1147 if (len < 0 && len != -EAGAIN)
1148 goto err_delete;
1149 if (len == want)
1150 svc_tcp_fragment_received(svsk);
1151 else
1152 dprintk("svc: incomplete TCP record (%d of %d)\n",
1153 (int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1154 svc_sock_reclen(svsk));
1155 goto err_noclose;
1156 }
1157
1158 if (svsk->sk_datalen < 8) {
1159 svsk->sk_datalen = 0;
1160 goto err_delete; /* client is nuts. */
1161 }
1162
1163 rqstp->rq_arg.len = svsk->sk_datalen;
1164 rqstp->rq_arg.page_base = 0;
1165 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1166 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1167 rqstp->rq_arg.page_len = 0;
1168 } else
1169 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1170
1171 rqstp->rq_xprt_ctxt = NULL;
1172 rqstp->rq_prot = IPPROTO_TCP;
1173 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1174 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1175 else
1176 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1177
1178 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1179 calldir = p[1];
1180 if (calldir)
1181 len = receive_cb_reply(svsk, rqstp);
1182
1183 /* Reset TCP read info */
1184 svsk->sk_datalen = 0;
1185 svc_tcp_fragment_received(svsk);
1186
1187 if (len < 0)
1188 goto error;
1189
1190 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1191 if (serv->sv_stats)
1192 serv->sv_stats->nettcpcnt++;
1193
1194 return rqstp->rq_arg.len;
1195
1196error:
1197 if (len != -EAGAIN)
1198 goto err_delete;
1199 dprintk("RPC: TCP recvfrom got EAGAIN\n");
1200 return 0;
1201err_delete:
1202 printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1203 svsk->sk_xprt.xpt_server->sv_name, -len);
1204 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1205err_noclose:
1206 return 0; /* record not complete */
1207}
1208
1209/*
1210 * Send out data on TCP socket.
1211 */
1212static int svc_tcp_sendto(struct svc_rqst *rqstp)
1213{
1214 struct xdr_buf *xbufp = &rqstp->rq_res;
1215 int sent;
1216 __be32 reclen;
1217
1218 /* Set up the first element of the reply kvec.
1219 * Any other kvecs that may be in use have been taken
1220 * care of by the server implementation itself.
1221 */
1222 reclen = htonl(0x80000000|((xbufp->len ) - 4));
1223 memcpy(xbufp->head[0].iov_base, &reclen, 4);
1224
1225 sent = svc_sendto(rqstp, &rqstp->rq_res);
1226 if (sent != xbufp->len) {
1227 printk(KERN_NOTICE
1228 "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1229 "- shutting down socket\n",
1230 rqstp->rq_xprt->xpt_server->sv_name,
1231 (sent<0)?"got error":"sent only",
1232 sent, xbufp->len);
1233 set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1234 svc_xprt_enqueue(rqstp->rq_xprt);
1235 sent = -EAGAIN;
1236 }
1237 return sent;
1238}
1239
1240/*
1241 * Setup response header. TCP has a 4B record length field.
1242 */
1243static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1244{
1245 struct kvec *resv = &rqstp->rq_res.head[0];
1246
1247 /* tcp needs a space for the record length... */
1248 svc_putnl(resv, 0);
1249}
1250
1251static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1252 struct net *net,
1253 struct sockaddr *sa, int salen,
1254 int flags)
1255{
1256 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1257}
1258
1259#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1260static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1261 struct net *, struct sockaddr *,
1262 int, int);
1263static void svc_bc_sock_free(struct svc_xprt *xprt);
1264
1265static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1266 struct net *net,
1267 struct sockaddr *sa, int salen,
1268 int flags)
1269{
1270 return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1271}
1272
1273static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1274{
1275}
1276
1277static struct svc_xprt_ops svc_tcp_bc_ops = {
1278 .xpo_create = svc_bc_tcp_create,
1279 .xpo_detach = svc_bc_tcp_sock_detach,
1280 .xpo_free = svc_bc_sock_free,
1281 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1282 .xpo_secure_port = svc_sock_secure_port,
1283};
1284
1285static struct svc_xprt_class svc_tcp_bc_class = {
1286 .xcl_name = "tcp-bc",
1287 .xcl_owner = THIS_MODULE,
1288 .xcl_ops = &svc_tcp_bc_ops,
1289 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1290};
1291
1292static void svc_init_bc_xprt_sock(void)
1293{
1294 svc_reg_xprt_class(&svc_tcp_bc_class);
1295}
1296
1297static void svc_cleanup_bc_xprt_sock(void)
1298{
1299 svc_unreg_xprt_class(&svc_tcp_bc_class);
1300}
1301#else /* CONFIG_SUNRPC_BACKCHANNEL */
1302static void svc_init_bc_xprt_sock(void)
1303{
1304}
1305
1306static void svc_cleanup_bc_xprt_sock(void)
1307{
1308}
1309#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1310
1311static struct svc_xprt_ops svc_tcp_ops = {
1312 .xpo_create = svc_tcp_create,
1313 .xpo_recvfrom = svc_tcp_recvfrom,
1314 .xpo_sendto = svc_tcp_sendto,
1315 .xpo_release_rqst = svc_release_skb,
1316 .xpo_detach = svc_tcp_sock_detach,
1317 .xpo_free = svc_sock_free,
1318 .xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1319 .xpo_has_wspace = svc_tcp_has_wspace,
1320 .xpo_accept = svc_tcp_accept,
1321 .xpo_secure_port = svc_sock_secure_port,
1322 .xpo_adjust_wspace = svc_tcp_adjust_wspace,
1323};
1324
1325static struct svc_xprt_class svc_tcp_class = {
1326 .xcl_name = "tcp",
1327 .xcl_owner = THIS_MODULE,
1328 .xcl_ops = &svc_tcp_ops,
1329 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1330 .xcl_ident = XPRT_TRANSPORT_TCP,
1331};
1332
1333void svc_init_xprt_sock(void)
1334{
1335 svc_reg_xprt_class(&svc_tcp_class);
1336 svc_reg_xprt_class(&svc_udp_class);
1337 svc_init_bc_xprt_sock();
1338}
1339
1340void svc_cleanup_xprt_sock(void)
1341{
1342 svc_unreg_xprt_class(&svc_tcp_class);
1343 svc_unreg_xprt_class(&svc_udp_class);
1344 svc_cleanup_bc_xprt_sock();
1345}
1346
1347static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1348{
1349 struct sock *sk = svsk->sk_sk;
1350
1351 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1352 &svsk->sk_xprt, serv);
1353 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1354 if (sk->sk_state == TCP_LISTEN) {
1355 dprintk("setting up TCP socket for listening\n");
1356 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1357 sk->sk_data_ready = svc_tcp_listen_data_ready;
1358 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1359 } else {
1360 dprintk("setting up TCP socket for reading\n");
1361 sk->sk_state_change = svc_tcp_state_change;
1362 sk->sk_data_ready = svc_tcp_data_ready;
1363 sk->sk_write_space = svc_tcp_write_space;
1364
1365 svsk->sk_reclen = 0;
1366 svsk->sk_tcplen = 0;
1367 svsk->sk_datalen = 0;
1368 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1369
1370 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1371
1372 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1373 if (sk->sk_state != TCP_ESTABLISHED)
1374 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1375 }
1376}
1377
1378void svc_sock_update_bufs(struct svc_serv *serv)
1379{
1380 /*
1381 * The number of server threads has changed. Update
1382 * rcvbuf and sndbuf accordingly on all sockets
1383 */
1384 struct svc_sock *svsk;
1385
1386 spin_lock_bh(&serv->sv_lock);
1387 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1388 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1389 spin_unlock_bh(&serv->sv_lock);
1390}
1391EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1392
1393/*
1394 * Initialize socket for RPC use and create svc_sock struct
1395 */
1396static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1397 struct socket *sock,
1398 int flags)
1399{
1400 struct svc_sock *svsk;
1401 struct sock *inet;
1402 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1403 int err = 0;
1404
1405 dprintk("svc: svc_setup_socket %p\n", sock);
1406 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1407 if (!svsk)
1408 return ERR_PTR(-ENOMEM);
1409
1410 inet = sock->sk;
1411
1412 /* Register socket with portmapper */
1413 if (pmap_register)
1414 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1415 inet->sk_protocol,
1416 ntohs(inet_sk(inet)->inet_sport));
1417
1418 if (err < 0) {
1419 kfree(svsk);
1420 return ERR_PTR(err);
1421 }
1422
1423 inet->sk_user_data = svsk;
1424 svsk->sk_sock = sock;
1425 svsk->sk_sk = inet;
1426 svsk->sk_ostate = inet->sk_state_change;
1427 svsk->sk_odata = inet->sk_data_ready;
1428 svsk->sk_owspace = inet->sk_write_space;
1429
1430 /* Initialize the socket */
1431 if (sock->type == SOCK_DGRAM)
1432 svc_udp_init(svsk, serv);
1433 else {
1434 /* initialise setting must have enough space to
1435 * receive and respond to one request.
1436 */
1437 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1438 4 * serv->sv_max_mesg);
1439 svc_tcp_init(svsk, serv);
1440 }
1441
1442 dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1443 svsk, svsk->sk_sk);
1444
1445 return svsk;
1446}
1447
1448bool svc_alien_sock(struct net *net, int fd)
1449{
1450 int err;
1451 struct socket *sock = sockfd_lookup(fd, &err);
1452 bool ret = false;
1453
1454 if (!sock)
1455 goto out;
1456 if (sock_net(sock->sk) != net)
1457 ret = true;
1458 sockfd_put(sock);
1459out:
1460 return ret;
1461}
1462EXPORT_SYMBOL_GPL(svc_alien_sock);
1463
1464/**
1465 * svc_addsock - add a listener socket to an RPC service
1466 * @serv: pointer to RPC service to which to add a new listener
1467 * @fd: file descriptor of the new listener
1468 * @name_return: pointer to buffer to fill in with name of listener
1469 * @len: size of the buffer
1470 *
1471 * Fills in socket name and returns positive length of name if successful.
1472 * Name is terminated with '\n'. On error, returns a negative errno
1473 * value.
1474 */
1475int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1476 const size_t len)
1477{
1478 int err = 0;
1479 struct socket *so = sockfd_lookup(fd, &err);
1480 struct svc_sock *svsk = NULL;
1481 struct sockaddr_storage addr;
1482 struct sockaddr *sin = (struct sockaddr *)&addr;
1483 int salen;
1484
1485 if (!so)
1486 return err;
1487 err = -EAFNOSUPPORT;
1488 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1489 goto out;
1490 err = -EPROTONOSUPPORT;
1491 if (so->sk->sk_protocol != IPPROTO_TCP &&
1492 so->sk->sk_protocol != IPPROTO_UDP)
1493 goto out;
1494 err = -EISCONN;
1495 if (so->state > SS_UNCONNECTED)
1496 goto out;
1497 err = -ENOENT;
1498 if (!try_module_get(THIS_MODULE))
1499 goto out;
1500 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1501 if (IS_ERR(svsk)) {
1502 module_put(THIS_MODULE);
1503 err = PTR_ERR(svsk);
1504 goto out;
1505 }
1506 if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
1507 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1508 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1509 return svc_one_sock_name(svsk, name_return, len);
1510out:
1511 sockfd_put(so);
1512 return err;
1513}
1514EXPORT_SYMBOL_GPL(svc_addsock);
1515
1516/*
1517 * Create socket for RPC service.
1518 */
1519static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1520 int protocol,
1521 struct net *net,
1522 struct sockaddr *sin, int len,
1523 int flags)
1524{
1525 struct svc_sock *svsk;
1526 struct socket *sock;
1527 int error;
1528 int type;
1529 struct sockaddr_storage addr;
1530 struct sockaddr *newsin = (struct sockaddr *)&addr;
1531 int newlen;
1532 int family;
1533 int val;
1534 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1535
1536 dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1537 serv->sv_program->pg_name, protocol,
1538 __svc_print_addr(sin, buf, sizeof(buf)));
1539
1540 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1541 printk(KERN_WARNING "svc: only UDP and TCP "
1542 "sockets supported\n");
1543 return ERR_PTR(-EINVAL);
1544 }
1545
1546 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1547 switch (sin->sa_family) {
1548 case AF_INET6:
1549 family = PF_INET6;
1550 break;
1551 case AF_INET:
1552 family = PF_INET;
1553 break;
1554 default:
1555 return ERR_PTR(-EINVAL);
1556 }
1557
1558 error = __sock_create(net, family, type, protocol, &sock, 1);
1559 if (error < 0)
1560 return ERR_PTR(error);
1561
1562 svc_reclassify_socket(sock);
1563
1564 /*
1565 * If this is an PF_INET6 listener, we want to avoid
1566 * getting requests from IPv4 remotes. Those should
1567 * be shunted to a PF_INET listener via rpcbind.
1568 */
1569 val = 1;
1570 if (family == PF_INET6)
1571 kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1572 (char *)&val, sizeof(val));
1573
1574 if (type == SOCK_STREAM)
1575 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1576 error = kernel_bind(sock, sin, len);
1577 if (error < 0)
1578 goto bummer;
1579
1580 newlen = len;
1581 error = kernel_getsockname(sock, newsin, &newlen);
1582 if (error < 0)
1583 goto bummer;
1584
1585 if (protocol == IPPROTO_TCP) {
1586 if ((error = kernel_listen(sock, 64)) < 0)
1587 goto bummer;
1588 }
1589
1590 svsk = svc_setup_socket(serv, sock, flags);
1591 if (IS_ERR(svsk)) {
1592 error = PTR_ERR(svsk);
1593 goto bummer;
1594 }
1595 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1596 return (struct svc_xprt *)svsk;
1597bummer:
1598 dprintk("svc: svc_create_socket error = %d\n", -error);
1599 sock_release(sock);
1600 return ERR_PTR(error);
1601}
1602
1603/*
1604 * Detach the svc_sock from the socket so that no
1605 * more callbacks occur.
1606 */
1607static void svc_sock_detach(struct svc_xprt *xprt)
1608{
1609 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1610 struct sock *sk = svsk->sk_sk;
1611 wait_queue_head_t *wq;
1612
1613 dprintk("svc: svc_sock_detach(%p)\n", svsk);
1614
1615 /* put back the old socket callbacks */
1616 sk->sk_state_change = svsk->sk_ostate;
1617 sk->sk_data_ready = svsk->sk_odata;
1618 sk->sk_write_space = svsk->sk_owspace;
1619
1620 wq = sk_sleep(sk);
1621 if (sunrpc_waitqueue_active(wq))
1622 wake_up_interruptible(wq);
1623}
1624
1625/*
1626 * Disconnect the socket, and reset the callbacks
1627 */
1628static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1629{
1630 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1631
1632 dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1633
1634 svc_sock_detach(xprt);
1635
1636 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1637 svc_tcp_clear_pages(svsk);
1638 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1639 }
1640}
1641
1642/*
1643 * Free the svc_sock's socket resources and the svc_sock itself.
1644 */
1645static void svc_sock_free(struct svc_xprt *xprt)
1646{
1647 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1648 dprintk("svc: svc_sock_free(%p)\n", svsk);
1649
1650 if (svsk->sk_sock->file)
1651 sockfd_put(svsk->sk_sock);
1652 else
1653 sock_release(svsk->sk_sock);
1654 kfree(svsk);
1655}
1656
1657#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1658/*
1659 * Create a back channel svc_xprt which shares the fore channel socket.
1660 */
1661static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1662 int protocol,
1663 struct net *net,
1664 struct sockaddr *sin, int len,
1665 int flags)
1666{
1667 struct svc_sock *svsk;
1668 struct svc_xprt *xprt;
1669
1670 if (protocol != IPPROTO_TCP) {
1671 printk(KERN_WARNING "svc: only TCP sockets"
1672 " supported on shared back channel\n");
1673 return ERR_PTR(-EINVAL);
1674 }
1675
1676 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1677 if (!svsk)
1678 return ERR_PTR(-ENOMEM);
1679
1680 xprt = &svsk->sk_xprt;
1681 svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1682
1683 serv->sv_bc_xprt = xprt;
1684
1685 return xprt;
1686}
1687
1688/*
1689 * Free a back channel svc_sock.
1690 */
1691static void svc_bc_sock_free(struct svc_xprt *xprt)
1692{
1693 if (xprt)
1694 kfree(container_of(xprt, struct svc_sock, sk_xprt));
1695}
1696#endif /* CONFIG_SUNRPC_BACKCHANNEL */