Loading...
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 */
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/net/sunrpc/svcsock.c
4 *
5 * These are the RPC server socket internals.
6 *
7 * The server scheduling algorithm does not always distribute the load
8 * evenly when servicing a single client. May need to modify the
9 * svc_xprt_enqueue procedure...
10 *
11 * TCP support is largely untested and may be a little slow. The problem
12 * is that we currently do two separate recvfrom's, one for the 4-byte
13 * record length, and the second for the actual record. This could possibly
14 * be improved by always reading a minimum size of around 100 bytes and
15 * tucking any superfluous bytes away in a temporary store. Still, that
16 * leaves write requests out in the rain. An alternative may be to peek at
17 * the first skb in the queue, and if it matches the next TCP sequence
18 * number, to extract the record marker. Yuck.
19 *
20 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
21 */
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/module.h>
26#include <linux/errno.h>
27#include <linux/fcntl.h>
28#include <linux/net.h>
29#include <linux/in.h>
30#include <linux/inet.h>
31#include <linux/udp.h>
32#include <linux/tcp.h>
33#include <linux/unistd.h>
34#include <linux/slab.h>
35#include <linux/netdevice.h>
36#include <linux/skbuff.h>
37#include <linux/file.h>
38#include <linux/freezer.h>
39#include <linux/bvec.h>
40
41#include <net/sock.h>
42#include <net/checksum.h>
43#include <net/ip.h>
44#include <net/ipv6.h>
45#include <net/udp.h>
46#include <net/tcp.h>
47#include <net/tcp_states.h>
48#include <net/tls_prot.h>
49#include <net/handshake.h>
50#include <linux/uaccess.h>
51#include <linux/highmem.h>
52#include <asm/ioctls.h>
53#include <linux/key.h>
54
55#include <linux/sunrpc/types.h>
56#include <linux/sunrpc/clnt.h>
57#include <linux/sunrpc/xdr.h>
58#include <linux/sunrpc/msg_prot.h>
59#include <linux/sunrpc/svcsock.h>
60#include <linux/sunrpc/stats.h>
61#include <linux/sunrpc/xprt.h>
62
63#include <trace/events/sock.h>
64#include <trace/events/sunrpc.h>
65
66#include "socklib.h"
67#include "sunrpc.h"
68
69#define RPCDBG_FACILITY RPCDBG_SVCXPRT
70
71/* To-do: to avoid tying up an nfsd thread while waiting for a
72 * handshake request, the request could instead be deferred.
73 */
74enum {
75 SVC_HANDSHAKE_TO = 5U * HZ
76};
77
78static struct svc_sock *svc_setup_socket(struct svc_serv *, struct socket *,
79 int flags);
80static int svc_udp_recvfrom(struct svc_rqst *);
81static int svc_udp_sendto(struct svc_rqst *);
82static void svc_sock_detach(struct svc_xprt *);
83static void svc_tcp_sock_detach(struct svc_xprt *);
84static void svc_sock_free(struct svc_xprt *);
85
86static struct svc_xprt *svc_create_socket(struct svc_serv *, int,
87 struct net *, struct sockaddr *,
88 int, int);
89#ifdef CONFIG_DEBUG_LOCK_ALLOC
90static struct lock_class_key svc_key[2];
91static struct lock_class_key svc_slock_key[2];
92
93static void svc_reclassify_socket(struct socket *sock)
94{
95 struct sock *sk = sock->sk;
96
97 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
98 return;
99
100 switch (sk->sk_family) {
101 case AF_INET:
102 sock_lock_init_class_and_name(sk, "slock-AF_INET-NFSD",
103 &svc_slock_key[0],
104 "sk_xprt.xpt_lock-AF_INET-NFSD",
105 &svc_key[0]);
106 break;
107
108 case AF_INET6:
109 sock_lock_init_class_and_name(sk, "slock-AF_INET6-NFSD",
110 &svc_slock_key[1],
111 "sk_xprt.xpt_lock-AF_INET6-NFSD",
112 &svc_key[1]);
113 break;
114
115 default:
116 BUG();
117 }
118}
119#else
120static void svc_reclassify_socket(struct socket *sock)
121{
122}
123#endif
124
125/**
126 * svc_tcp_release_ctxt - Release transport-related resources
127 * @xprt: the transport which owned the context
128 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
129 *
130 */
131static void svc_tcp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
132{
133}
134
135/**
136 * svc_udp_release_ctxt - Release transport-related resources
137 * @xprt: the transport which owned the context
138 * @ctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt
139 *
140 */
141static void svc_udp_release_ctxt(struct svc_xprt *xprt, void *ctxt)
142{
143 struct sk_buff *skb = ctxt;
144
145 if (skb)
146 consume_skb(skb);
147}
148
149union svc_pktinfo_u {
150 struct in_pktinfo pkti;
151 struct in6_pktinfo pkti6;
152};
153#define SVC_PKTINFO_SPACE \
154 CMSG_SPACE(sizeof(union svc_pktinfo_u))
155
156static void svc_set_cmsg_data(struct svc_rqst *rqstp, struct cmsghdr *cmh)
157{
158 struct svc_sock *svsk =
159 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
160 switch (svsk->sk_sk->sk_family) {
161 case AF_INET: {
162 struct in_pktinfo *pki = CMSG_DATA(cmh);
163
164 cmh->cmsg_level = SOL_IP;
165 cmh->cmsg_type = IP_PKTINFO;
166 pki->ipi_ifindex = 0;
167 pki->ipi_spec_dst.s_addr =
168 svc_daddr_in(rqstp)->sin_addr.s_addr;
169 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
170 }
171 break;
172
173 case AF_INET6: {
174 struct in6_pktinfo *pki = CMSG_DATA(cmh);
175 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
176
177 cmh->cmsg_level = SOL_IPV6;
178 cmh->cmsg_type = IPV6_PKTINFO;
179 pki->ipi6_ifindex = daddr->sin6_scope_id;
180 pki->ipi6_addr = daddr->sin6_addr;
181 cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
182 }
183 break;
184 }
185}
186
187static int svc_sock_result_payload(struct svc_rqst *rqstp, unsigned int offset,
188 unsigned int length)
189{
190 return 0;
191}
192
193/*
194 * Report socket names for nfsdfs
195 */
196static int svc_one_sock_name(struct svc_sock *svsk, char *buf, int remaining)
197{
198 const struct sock *sk = svsk->sk_sk;
199 const char *proto_name = sk->sk_protocol == IPPROTO_UDP ?
200 "udp" : "tcp";
201 int len;
202
203 switch (sk->sk_family) {
204 case PF_INET:
205 len = snprintf(buf, remaining, "ipv4 %s %pI4 %d\n",
206 proto_name,
207 &inet_sk(sk)->inet_rcv_saddr,
208 inet_sk(sk)->inet_num);
209 break;
210#if IS_ENABLED(CONFIG_IPV6)
211 case PF_INET6:
212 len = snprintf(buf, remaining, "ipv6 %s %pI6 %d\n",
213 proto_name,
214 &sk->sk_v6_rcv_saddr,
215 inet_sk(sk)->inet_num);
216 break;
217#endif
218 default:
219 len = snprintf(buf, remaining, "*unknown-%d*\n",
220 sk->sk_family);
221 }
222
223 if (len >= remaining) {
224 *buf = '\0';
225 return -ENAMETOOLONG;
226 }
227 return len;
228}
229
230static int
231svc_tcp_sock_process_cmsg(struct socket *sock, struct msghdr *msg,
232 struct cmsghdr *cmsg, int ret)
233{
234 u8 content_type = tls_get_record_type(sock->sk, cmsg);
235 u8 level, description;
236
237 switch (content_type) {
238 case 0:
239 break;
240 case TLS_RECORD_TYPE_DATA:
241 /* TLS sets EOR at the end of each application data
242 * record, even though there might be more frames
243 * waiting to be decrypted.
244 */
245 msg->msg_flags &= ~MSG_EOR;
246 break;
247 case TLS_RECORD_TYPE_ALERT:
248 tls_alert_recv(sock->sk, msg, &level, &description);
249 ret = (level == TLS_ALERT_LEVEL_FATAL) ?
250 -ENOTCONN : -EAGAIN;
251 break;
252 default:
253 /* discard this record type */
254 ret = -EAGAIN;
255 }
256 return ret;
257}
258
259static int
260svc_tcp_sock_recv_cmsg(struct svc_sock *svsk, struct msghdr *msg)
261{
262 union {
263 struct cmsghdr cmsg;
264 u8 buf[CMSG_SPACE(sizeof(u8))];
265 } u;
266 struct socket *sock = svsk->sk_sock;
267 int ret;
268
269 msg->msg_control = &u;
270 msg->msg_controllen = sizeof(u);
271 ret = sock_recvmsg(sock, msg, MSG_DONTWAIT);
272 if (unlikely(msg->msg_controllen != sizeof(u)))
273 ret = svc_tcp_sock_process_cmsg(sock, msg, &u.cmsg, ret);
274 return ret;
275}
276
277#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
278static void svc_flush_bvec(const struct bio_vec *bvec, size_t size, size_t seek)
279{
280 struct bvec_iter bi = {
281 .bi_size = size + seek,
282 };
283 struct bio_vec bv;
284
285 bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
286 for_each_bvec(bv, bvec, bi, bi)
287 flush_dcache_page(bv.bv_page);
288}
289#else
290static inline void svc_flush_bvec(const struct bio_vec *bvec, size_t size,
291 size_t seek)
292{
293}
294#endif
295
296/*
297 * Read from @rqstp's transport socket. The incoming message fills whole
298 * pages in @rqstp's rq_pages array until the last page of the message
299 * has been received into a partial page.
300 */
301static ssize_t svc_tcp_read_msg(struct svc_rqst *rqstp, size_t buflen,
302 size_t seek)
303{
304 struct svc_sock *svsk =
305 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
306 struct bio_vec *bvec = rqstp->rq_bvec;
307 struct msghdr msg = { NULL };
308 unsigned int i;
309 ssize_t len;
310 size_t t;
311
312 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
313
314 for (i = 0, t = 0; t < buflen; i++, t += PAGE_SIZE)
315 bvec_set_page(&bvec[i], rqstp->rq_pages[i], PAGE_SIZE, 0);
316 rqstp->rq_respages = &rqstp->rq_pages[i];
317 rqstp->rq_next_page = rqstp->rq_respages + 1;
318
319 iov_iter_bvec(&msg.msg_iter, ITER_DEST, bvec, i, buflen);
320 if (seek) {
321 iov_iter_advance(&msg.msg_iter, seek);
322 buflen -= seek;
323 }
324 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
325 if (len > 0)
326 svc_flush_bvec(bvec, len, seek);
327
328 /* If we read a full record, then assume there may be more
329 * data to read (stream based sockets only!)
330 */
331 if (len == buflen)
332 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
333
334 return len;
335}
336
337/*
338 * Set socket snd and rcv buffer lengths
339 */
340static void svc_sock_setbufsize(struct svc_sock *svsk, unsigned int nreqs)
341{
342 unsigned int max_mesg = svsk->sk_xprt.xpt_server->sv_max_mesg;
343 struct socket *sock = svsk->sk_sock;
344
345 nreqs = min(nreqs, INT_MAX / 2 / max_mesg);
346
347 lock_sock(sock->sk);
348 sock->sk->sk_sndbuf = nreqs * max_mesg * 2;
349 sock->sk->sk_rcvbuf = nreqs * max_mesg * 2;
350 sock->sk->sk_write_space(sock->sk);
351 release_sock(sock->sk);
352}
353
354static void svc_sock_secure_port(struct svc_rqst *rqstp)
355{
356 if (svc_port_is_privileged(svc_addr(rqstp)))
357 set_bit(RQ_SECURE, &rqstp->rq_flags);
358 else
359 clear_bit(RQ_SECURE, &rqstp->rq_flags);
360}
361
362/*
363 * INET callback when data has been received on the socket.
364 */
365static void svc_data_ready(struct sock *sk)
366{
367 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
368
369 trace_sk_data_ready(sk);
370
371 if (svsk) {
372 /* Refer to svc_setup_socket() for details. */
373 rmb();
374 svsk->sk_odata(sk);
375 trace_svcsock_data_ready(&svsk->sk_xprt, 0);
376 if (test_bit(XPT_HANDSHAKE, &svsk->sk_xprt.xpt_flags))
377 return;
378 if (!test_and_set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags))
379 svc_xprt_enqueue(&svsk->sk_xprt);
380 }
381}
382
383/*
384 * INET callback when space is newly available on the socket.
385 */
386static void svc_write_space(struct sock *sk)
387{
388 struct svc_sock *svsk = (struct svc_sock *)(sk->sk_user_data);
389
390 if (svsk) {
391 /* Refer to svc_setup_socket() for details. */
392 rmb();
393 trace_svcsock_write_space(&svsk->sk_xprt, 0);
394 svsk->sk_owspace(sk);
395 svc_xprt_enqueue(&svsk->sk_xprt);
396 }
397}
398
399static int svc_tcp_has_wspace(struct svc_xprt *xprt)
400{
401 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
402
403 if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
404 return 1;
405 return !test_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
406}
407
408static void svc_tcp_kill_temp_xprt(struct svc_xprt *xprt)
409{
410 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
411
412 sock_no_linger(svsk->sk_sock->sk);
413}
414
415/**
416 * svc_tcp_handshake_done - Handshake completion handler
417 * @data: address of xprt to wake
418 * @status: status of handshake
419 * @peerid: serial number of key containing the remote peer's identity
420 *
421 * If a security policy is specified as an export option, we don't
422 * have a specific export here to check. So we set a "TLS session
423 * is present" flag on the xprt and let an upper layer enforce local
424 * security policy.
425 */
426static void svc_tcp_handshake_done(void *data, int status, key_serial_t peerid)
427{
428 struct svc_xprt *xprt = data;
429 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
430
431 if (!status) {
432 if (peerid != TLS_NO_PEERID)
433 set_bit(XPT_PEER_AUTH, &xprt->xpt_flags);
434 set_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
435 }
436 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
437 complete_all(&svsk->sk_handshake_done);
438}
439
440/**
441 * svc_tcp_handshake - Perform a transport-layer security handshake
442 * @xprt: connected transport endpoint
443 *
444 */
445static void svc_tcp_handshake(struct svc_xprt *xprt)
446{
447 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
448 struct sock *sk = svsk->sk_sock->sk;
449 struct tls_handshake_args args = {
450 .ta_sock = svsk->sk_sock,
451 .ta_done = svc_tcp_handshake_done,
452 .ta_data = xprt,
453 };
454 int ret;
455
456 trace_svc_tls_upcall(xprt);
457
458 clear_bit(XPT_TLS_SESSION, &xprt->xpt_flags);
459 init_completion(&svsk->sk_handshake_done);
460
461 ret = tls_server_hello_x509(&args, GFP_KERNEL);
462 if (ret) {
463 trace_svc_tls_not_started(xprt);
464 goto out_failed;
465 }
466
467 ret = wait_for_completion_interruptible_timeout(&svsk->sk_handshake_done,
468 SVC_HANDSHAKE_TO);
469 if (ret <= 0) {
470 if (tls_handshake_cancel(sk)) {
471 trace_svc_tls_timed_out(xprt);
472 goto out_close;
473 }
474 }
475
476 if (!test_bit(XPT_TLS_SESSION, &xprt->xpt_flags)) {
477 trace_svc_tls_unavailable(xprt);
478 goto out_close;
479 }
480
481 /* Mark the transport ready in case the remote sent RPC
482 * traffic before the kernel received the handshake
483 * completion downcall.
484 */
485 set_bit(XPT_DATA, &xprt->xpt_flags);
486 svc_xprt_enqueue(xprt);
487 return;
488
489out_close:
490 set_bit(XPT_CLOSE, &xprt->xpt_flags);
491out_failed:
492 clear_bit(XPT_HANDSHAKE, &xprt->xpt_flags);
493 set_bit(XPT_DATA, &xprt->xpt_flags);
494 svc_xprt_enqueue(xprt);
495}
496
497/*
498 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
499 */
500static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
501 struct cmsghdr *cmh)
502{
503 struct in_pktinfo *pki = CMSG_DATA(cmh);
504 struct sockaddr_in *daddr = svc_daddr_in(rqstp);
505
506 if (cmh->cmsg_type != IP_PKTINFO)
507 return 0;
508
509 daddr->sin_family = AF_INET;
510 daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
511 return 1;
512}
513
514/*
515 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
516 */
517static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
518 struct cmsghdr *cmh)
519{
520 struct in6_pktinfo *pki = CMSG_DATA(cmh);
521 struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
522
523 if (cmh->cmsg_type != IPV6_PKTINFO)
524 return 0;
525
526 daddr->sin6_family = AF_INET6;
527 daddr->sin6_addr = pki->ipi6_addr;
528 daddr->sin6_scope_id = pki->ipi6_ifindex;
529 return 1;
530}
531
532/*
533 * Copy the UDP datagram's destination address to the rqstp structure.
534 * The 'destination' address in this case is the address to which the
535 * peer sent the datagram, i.e. our local address. For multihomed
536 * hosts, this can change from msg to msg. Note that only the IP
537 * address changes, the port number should remain the same.
538 */
539static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
540 struct cmsghdr *cmh)
541{
542 switch (cmh->cmsg_level) {
543 case SOL_IP:
544 return svc_udp_get_dest_address4(rqstp, cmh);
545 case SOL_IPV6:
546 return svc_udp_get_dest_address6(rqstp, cmh);
547 }
548
549 return 0;
550}
551
552/**
553 * svc_udp_recvfrom - Receive a datagram from a UDP socket.
554 * @rqstp: request structure into which to receive an RPC Call
555 *
556 * Called in a loop when XPT_DATA has been set.
557 *
558 * Returns:
559 * On success, the number of bytes in a received RPC Call, or
560 * %0 if a complete RPC Call message was not ready to return
561 */
562static int svc_udp_recvfrom(struct svc_rqst *rqstp)
563{
564 struct svc_sock *svsk =
565 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
566 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
567 struct sk_buff *skb;
568 union {
569 struct cmsghdr hdr;
570 long all[SVC_PKTINFO_SPACE / sizeof(long)];
571 } buffer;
572 struct cmsghdr *cmh = &buffer.hdr;
573 struct msghdr msg = {
574 .msg_name = svc_addr(rqstp),
575 .msg_control = cmh,
576 .msg_controllen = sizeof(buffer),
577 .msg_flags = MSG_DONTWAIT,
578 };
579 size_t len;
580 int err;
581
582 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
583 /* udp sockets need large rcvbuf as all pending
584 * requests are still in that buffer. sndbuf must
585 * also be large enough that there is enough space
586 * for one reply per thread. We count all threads
587 * rather than threads in a particular pool, which
588 * provides an upper bound on the number of threads
589 * which will access the socket.
590 */
591 svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
592
593 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
594 err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
595 0, 0, MSG_PEEK | MSG_DONTWAIT);
596 if (err < 0)
597 goto out_recv_err;
598 skb = skb_recv_udp(svsk->sk_sk, MSG_DONTWAIT, &err);
599 if (!skb)
600 goto out_recv_err;
601
602 len = svc_addr_len(svc_addr(rqstp));
603 rqstp->rq_addrlen = len;
604 if (skb->tstamp == 0) {
605 skb->tstamp = ktime_get_real();
606 /* Don't enable netstamp, sunrpc doesn't
607 need that much accuracy */
608 }
609 sock_write_timestamp(svsk->sk_sk, skb->tstamp);
610 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
611
612 len = skb->len;
613 rqstp->rq_arg.len = len;
614 trace_svcsock_udp_recv(&svsk->sk_xprt, len);
615
616 rqstp->rq_prot = IPPROTO_UDP;
617
618 if (!svc_udp_get_dest_address(rqstp, cmh))
619 goto out_cmsg_err;
620 rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
621
622 if (skb_is_nonlinear(skb)) {
623 /* we have to copy */
624 local_bh_disable();
625 if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb))
626 goto out_bh_enable;
627 local_bh_enable();
628 consume_skb(skb);
629 } else {
630 /* we can use it in-place */
631 rqstp->rq_arg.head[0].iov_base = skb->data;
632 rqstp->rq_arg.head[0].iov_len = len;
633 if (skb_checksum_complete(skb))
634 goto out_free;
635 rqstp->rq_xprt_ctxt = skb;
636 }
637
638 rqstp->rq_arg.page_base = 0;
639 if (len <= rqstp->rq_arg.head[0].iov_len) {
640 rqstp->rq_arg.head[0].iov_len = len;
641 rqstp->rq_arg.page_len = 0;
642 rqstp->rq_respages = rqstp->rq_pages+1;
643 } else {
644 rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
645 rqstp->rq_respages = rqstp->rq_pages + 1 +
646 DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
647 }
648 rqstp->rq_next_page = rqstp->rq_respages+1;
649
650 if (serv->sv_stats)
651 serv->sv_stats->netudpcnt++;
652
653 svc_sock_secure_port(rqstp);
654 svc_xprt_received(rqstp->rq_xprt);
655 return len;
656
657out_recv_err:
658 if (err != -EAGAIN) {
659 /* possibly an icmp error */
660 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
661 }
662 trace_svcsock_udp_recv_err(&svsk->sk_xprt, err);
663 goto out_clear_busy;
664out_cmsg_err:
665 net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
666 cmh->cmsg_level, cmh->cmsg_type);
667 goto out_free;
668out_bh_enable:
669 local_bh_enable();
670out_free:
671 kfree_skb(skb);
672out_clear_busy:
673 svc_xprt_received(rqstp->rq_xprt);
674 return 0;
675}
676
677/**
678 * svc_udp_sendto - Send out a reply on a UDP socket
679 * @rqstp: completed svc_rqst
680 *
681 * xpt_mutex ensures @rqstp's whole message is written to the socket
682 * without interruption.
683 *
684 * Returns the number of bytes sent, or a negative errno.
685 */
686static int svc_udp_sendto(struct svc_rqst *rqstp)
687{
688 struct svc_xprt *xprt = rqstp->rq_xprt;
689 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
690 struct xdr_buf *xdr = &rqstp->rq_res;
691 union {
692 struct cmsghdr hdr;
693 long all[SVC_PKTINFO_SPACE / sizeof(long)];
694 } buffer;
695 struct cmsghdr *cmh = &buffer.hdr;
696 struct msghdr msg = {
697 .msg_name = &rqstp->rq_addr,
698 .msg_namelen = rqstp->rq_addrlen,
699 .msg_control = cmh,
700 .msg_flags = MSG_SPLICE_PAGES,
701 .msg_controllen = sizeof(buffer),
702 };
703 unsigned int count;
704 int err;
705
706 svc_udp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
707 rqstp->rq_xprt_ctxt = NULL;
708
709 svc_set_cmsg_data(rqstp, cmh);
710
711 mutex_lock(&xprt->xpt_mutex);
712
713 if (svc_xprt_is_dead(xprt))
714 goto out_notconn;
715
716 count = xdr_buf_to_bvec(rqstp->rq_bvec,
717 ARRAY_SIZE(rqstp->rq_bvec), xdr);
718
719 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
720 count, rqstp->rq_res.len);
721 err = sock_sendmsg(svsk->sk_sock, &msg);
722 if (err == -ECONNREFUSED) {
723 /* ICMP error on earlier request. */
724 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
725 count, rqstp->rq_res.len);
726 err = sock_sendmsg(svsk->sk_sock, &msg);
727 }
728
729 trace_svcsock_udp_send(xprt, err);
730
731 mutex_unlock(&xprt->xpt_mutex);
732 return err;
733
734out_notconn:
735 mutex_unlock(&xprt->xpt_mutex);
736 return -ENOTCONN;
737}
738
739static int svc_udp_has_wspace(struct svc_xprt *xprt)
740{
741 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
742 struct svc_serv *serv = xprt->xpt_server;
743 unsigned long required;
744
745 /*
746 * Set the SOCK_NOSPACE flag before checking the available
747 * sock space.
748 */
749 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
750 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
751 if (required*2 > sock_wspace(svsk->sk_sk))
752 return 0;
753 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
754 return 1;
755}
756
757static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
758{
759 BUG();
760 return NULL;
761}
762
763static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
764{
765}
766
767static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
768 struct net *net,
769 struct sockaddr *sa, int salen,
770 int flags)
771{
772 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
773}
774
775static const struct svc_xprt_ops svc_udp_ops = {
776 .xpo_create = svc_udp_create,
777 .xpo_recvfrom = svc_udp_recvfrom,
778 .xpo_sendto = svc_udp_sendto,
779 .xpo_result_payload = svc_sock_result_payload,
780 .xpo_release_ctxt = svc_udp_release_ctxt,
781 .xpo_detach = svc_sock_detach,
782 .xpo_free = svc_sock_free,
783 .xpo_has_wspace = svc_udp_has_wspace,
784 .xpo_accept = svc_udp_accept,
785 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
786};
787
788static struct svc_xprt_class svc_udp_class = {
789 .xcl_name = "udp",
790 .xcl_owner = THIS_MODULE,
791 .xcl_ops = &svc_udp_ops,
792 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
793 .xcl_ident = XPRT_TRANSPORT_UDP,
794};
795
796static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
797{
798 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
799 &svsk->sk_xprt, serv);
800 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
801 svsk->sk_sk->sk_data_ready = svc_data_ready;
802 svsk->sk_sk->sk_write_space = svc_write_space;
803
804 /* initialise setting must have enough space to
805 * receive and respond to one request.
806 * svc_udp_recvfrom will re-adjust if necessary
807 */
808 svc_sock_setbufsize(svsk, 3);
809
810 /* data might have come in before data_ready set up */
811 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
812 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
813
814 /* make sure we get destination address info */
815 switch (svsk->sk_sk->sk_family) {
816 case AF_INET:
817 ip_sock_set_pktinfo(svsk->sk_sock->sk);
818 break;
819 case AF_INET6:
820 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
821 break;
822 default:
823 BUG();
824 }
825}
826
827/*
828 * A data_ready event on a listening socket means there's a connection
829 * pending. Do not use state_change as a substitute for it.
830 */
831static void svc_tcp_listen_data_ready(struct sock *sk)
832{
833 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
834
835 trace_sk_data_ready(sk);
836
837 /*
838 * This callback may called twice when a new connection
839 * is established as a child socket inherits everything
840 * from a parent LISTEN socket.
841 * 1) data_ready method of the parent socket will be called
842 * when one of child sockets become ESTABLISHED.
843 * 2) data_ready method of the child socket may be called
844 * when it receives data before the socket is accepted.
845 * In case of 2, we should ignore it silently and DO NOT
846 * dereference svsk.
847 */
848 if (sk->sk_state != TCP_LISTEN)
849 return;
850
851 if (svsk) {
852 /* Refer to svc_setup_socket() for details. */
853 rmb();
854 svsk->sk_odata(sk);
855 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
856 svc_xprt_enqueue(&svsk->sk_xprt);
857 }
858}
859
860/*
861 * A state change on a connected socket means it's dying or dead.
862 */
863static void svc_tcp_state_change(struct sock *sk)
864{
865 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
866
867 if (svsk) {
868 /* Refer to svc_setup_socket() for details. */
869 rmb();
870 svsk->sk_ostate(sk);
871 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
872 if (sk->sk_state != TCP_ESTABLISHED)
873 svc_xprt_deferred_close(&svsk->sk_xprt);
874 }
875}
876
877/*
878 * Accept a TCP connection
879 */
880static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
881{
882 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
883 struct sockaddr_storage addr;
884 struct sockaddr *sin = (struct sockaddr *) &addr;
885 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
886 struct socket *sock = svsk->sk_sock;
887 struct socket *newsock;
888 struct svc_sock *newsvsk;
889 int err, slen;
890
891 if (!sock)
892 return NULL;
893
894 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
895 err = kernel_accept(sock, &newsock, O_NONBLOCK);
896 if (err < 0) {
897 if (err != -EAGAIN)
898 trace_svcsock_accept_err(xprt, serv->sv_name, err);
899 return NULL;
900 }
901 if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL)))
902 return NULL;
903
904 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
905
906 err = kernel_getpeername(newsock, sin);
907 if (err < 0) {
908 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
909 goto failed; /* aborted connection or whatever */
910 }
911 slen = err;
912
913 /* Reset the inherited callbacks before calling svc_setup_socket */
914 newsock->sk->sk_state_change = svsk->sk_ostate;
915 newsock->sk->sk_data_ready = svsk->sk_odata;
916 newsock->sk->sk_write_space = svsk->sk_owspace;
917
918 /* make sure that a write doesn't block forever when
919 * low on memory
920 */
921 newsock->sk->sk_sndtimeo = HZ*30;
922
923 newsvsk = svc_setup_socket(serv, newsock,
924 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
925 if (IS_ERR(newsvsk))
926 goto failed;
927 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
928 err = kernel_getsockname(newsock, sin);
929 slen = err;
930 if (unlikely(err < 0))
931 slen = offsetof(struct sockaddr, sa_data);
932 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
933
934 if (sock_is_loopback(newsock->sk))
935 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
936 else
937 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
938 if (serv->sv_stats)
939 serv->sv_stats->nettcpconn++;
940
941 return &newsvsk->sk_xprt;
942
943failed:
944 sockfd_put(newsock);
945 return NULL;
946}
947
948static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
949 struct svc_rqst *rqstp)
950{
951 size_t len = svsk->sk_datalen;
952 unsigned int i, npages;
953
954 if (!len)
955 return 0;
956 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
957 for (i = 0; i < npages; i++) {
958 if (rqstp->rq_pages[i] != NULL)
959 put_page(rqstp->rq_pages[i]);
960 BUG_ON(svsk->sk_pages[i] == NULL);
961 rqstp->rq_pages[i] = svsk->sk_pages[i];
962 svsk->sk_pages[i] = NULL;
963 }
964 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
965 return len;
966}
967
968static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
969{
970 unsigned int i, len, npages;
971
972 if (svsk->sk_datalen == 0)
973 return;
974 len = svsk->sk_datalen;
975 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
976 for (i = 0; i < npages; i++) {
977 svsk->sk_pages[i] = rqstp->rq_pages[i];
978 rqstp->rq_pages[i] = NULL;
979 }
980}
981
982static void svc_tcp_clear_pages(struct svc_sock *svsk)
983{
984 unsigned int i, len, npages;
985
986 if (svsk->sk_datalen == 0)
987 goto out;
988 len = svsk->sk_datalen;
989 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
990 for (i = 0; i < npages; i++) {
991 if (svsk->sk_pages[i] == NULL) {
992 WARN_ON_ONCE(1);
993 continue;
994 }
995 put_page(svsk->sk_pages[i]);
996 svsk->sk_pages[i] = NULL;
997 }
998out:
999 svsk->sk_tcplen = 0;
1000 svsk->sk_datalen = 0;
1001}
1002
1003/*
1004 * Receive fragment record header into sk_marker.
1005 */
1006static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
1007 struct svc_rqst *rqstp)
1008{
1009 ssize_t want, len;
1010
1011 /* If we haven't gotten the record length yet,
1012 * get the next four bytes.
1013 */
1014 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1015 struct msghdr msg = { NULL };
1016 struct kvec iov;
1017
1018 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1019 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
1020 iov.iov_len = want;
1021 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
1022 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
1023 if (len < 0)
1024 return len;
1025 svsk->sk_tcplen += len;
1026 if (len < want) {
1027 /* call again to read the remaining bytes */
1028 goto err_short;
1029 }
1030 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
1031 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1032 svsk->sk_xprt.xpt_server->sv_max_mesg)
1033 goto err_too_large;
1034 }
1035 return svc_sock_reclen(svsk);
1036
1037err_too_large:
1038 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
1039 __func__, svsk->sk_xprt.xpt_server->sv_name,
1040 svc_sock_reclen(svsk));
1041 svc_xprt_deferred_close(&svsk->sk_xprt);
1042err_short:
1043 return -EAGAIN;
1044}
1045
1046static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1047{
1048 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1049 struct rpc_rqst *req = NULL;
1050 struct kvec *src, *dst;
1051 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1052 __be32 xid = *p;
1053
1054 if (!bc_xprt)
1055 return -EAGAIN;
1056 spin_lock(&bc_xprt->queue_lock);
1057 req = xprt_lookup_rqst(bc_xprt, xid);
1058 if (!req)
1059 goto unlock_eagain;
1060
1061 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1062 /*
1063 * XXX!: cheating for now! Only copying HEAD.
1064 * But we know this is good enough for now (in fact, for any
1065 * callback reply in the forseeable future).
1066 */
1067 dst = &req->rq_private_buf.head[0];
1068 src = &rqstp->rq_arg.head[0];
1069 if (dst->iov_len < src->iov_len)
1070 goto unlock_eagain; /* whatever; just giving up. */
1071 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1072 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1073 rqstp->rq_arg.len = 0;
1074 spin_unlock(&bc_xprt->queue_lock);
1075 return 0;
1076unlock_eagain:
1077 spin_unlock(&bc_xprt->queue_lock);
1078 return -EAGAIN;
1079}
1080
1081static void svc_tcp_fragment_received(struct svc_sock *svsk)
1082{
1083 /* If we have more data, signal svc_xprt_enqueue() to try again */
1084 svsk->sk_tcplen = 0;
1085 svsk->sk_marker = xdr_zero;
1086}
1087
1088/**
1089 * svc_tcp_recvfrom - Receive data from a TCP socket
1090 * @rqstp: request structure into which to receive an RPC Call
1091 *
1092 * Called in a loop when XPT_DATA has been set.
1093 *
1094 * Read the 4-byte stream record marker, then use the record length
1095 * in that marker to set up exactly the resources needed to receive
1096 * the next RPC message into @rqstp.
1097 *
1098 * Returns:
1099 * On success, the number of bytes in a received RPC Call, or
1100 * %0 if a complete RPC Call message was not ready to return
1101 *
1102 * The zero return case handles partial receives and callback Replies.
1103 * The state of a partial receive is preserved in the svc_sock for
1104 * the next call to svc_tcp_recvfrom.
1105 */
1106static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1107{
1108 struct svc_sock *svsk =
1109 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1110 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1111 size_t want, base;
1112 ssize_t len;
1113 __be32 *p;
1114 __be32 calldir;
1115
1116 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1117 len = svc_tcp_read_marker(svsk, rqstp);
1118 if (len < 0)
1119 goto error;
1120
1121 base = svc_tcp_restore_pages(svsk, rqstp);
1122 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1123 len = svc_tcp_read_msg(rqstp, base + want, base);
1124 if (len >= 0) {
1125 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1126 svsk->sk_tcplen += len;
1127 svsk->sk_datalen += len;
1128 }
1129 if (len != want || !svc_sock_final_rec(svsk))
1130 goto err_incomplete;
1131 if (svsk->sk_datalen < 8)
1132 goto err_nuts;
1133
1134 rqstp->rq_arg.len = svsk->sk_datalen;
1135 rqstp->rq_arg.page_base = 0;
1136 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1137 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1138 rqstp->rq_arg.page_len = 0;
1139 } else
1140 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1141
1142 rqstp->rq_xprt_ctxt = NULL;
1143 rqstp->rq_prot = IPPROTO_TCP;
1144 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1145 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1146 else
1147 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1148
1149 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1150 calldir = p[1];
1151 if (calldir)
1152 len = receive_cb_reply(svsk, rqstp);
1153
1154 /* Reset TCP read info */
1155 svsk->sk_datalen = 0;
1156 svc_tcp_fragment_received(svsk);
1157
1158 if (len < 0)
1159 goto error;
1160
1161 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1162 if (serv->sv_stats)
1163 serv->sv_stats->nettcpcnt++;
1164
1165 svc_sock_secure_port(rqstp);
1166 svc_xprt_received(rqstp->rq_xprt);
1167 return rqstp->rq_arg.len;
1168
1169err_incomplete:
1170 svc_tcp_save_pages(svsk, rqstp);
1171 if (len < 0 && len != -EAGAIN)
1172 goto err_delete;
1173 if (len == want)
1174 svc_tcp_fragment_received(svsk);
1175 else
1176 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1177 svc_sock_reclen(svsk),
1178 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1179 goto err_noclose;
1180error:
1181 if (len != -EAGAIN)
1182 goto err_delete;
1183 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1184 goto err_noclose;
1185err_nuts:
1186 svsk->sk_datalen = 0;
1187err_delete:
1188 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1189 svc_xprt_deferred_close(&svsk->sk_xprt);
1190err_noclose:
1191 svc_xprt_received(rqstp->rq_xprt);
1192 return 0; /* record not complete */
1193}
1194
1195/*
1196 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1197 * copy operations in this path. Therefore the caller must ensure
1198 * that the pages backing @xdr are unchanging.
1199 */
1200static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1201 rpc_fraghdr marker, unsigned int *sentp)
1202{
1203 struct msghdr msg = {
1204 .msg_flags = MSG_SPLICE_PAGES,
1205 };
1206 unsigned int count;
1207 void *buf;
1208 int ret;
1209
1210 *sentp = 0;
1211
1212 /* The stream record marker is copied into a temporary page
1213 * fragment buffer so that it can be included in rq_bvec.
1214 */
1215 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1216 GFP_KERNEL);
1217 if (!buf)
1218 return -ENOMEM;
1219 memcpy(buf, &marker, sizeof(marker));
1220 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1221
1222 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1223 ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1224
1225 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1226 1 + count, sizeof(marker) + rqstp->rq_res.len);
1227 ret = sock_sendmsg(svsk->sk_sock, &msg);
1228 page_frag_free(buf);
1229 if (ret < 0)
1230 return ret;
1231 *sentp += ret;
1232 return 0;
1233}
1234
1235/**
1236 * svc_tcp_sendto - Send out a reply on a TCP socket
1237 * @rqstp: completed svc_rqst
1238 *
1239 * xpt_mutex ensures @rqstp's whole message is written to the socket
1240 * without interruption.
1241 *
1242 * Returns the number of bytes sent, or a negative errno.
1243 */
1244static int svc_tcp_sendto(struct svc_rqst *rqstp)
1245{
1246 struct svc_xprt *xprt = rqstp->rq_xprt;
1247 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1248 struct xdr_buf *xdr = &rqstp->rq_res;
1249 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1250 (u32)xdr->len);
1251 unsigned int sent;
1252 int err;
1253
1254 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1255 rqstp->rq_xprt_ctxt = NULL;
1256
1257 mutex_lock(&xprt->xpt_mutex);
1258 if (svc_xprt_is_dead(xprt))
1259 goto out_notconn;
1260 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1261 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1262 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1263 goto out_close;
1264 mutex_unlock(&xprt->xpt_mutex);
1265 return sent;
1266
1267out_notconn:
1268 mutex_unlock(&xprt->xpt_mutex);
1269 return -ENOTCONN;
1270out_close:
1271 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1272 xprt->xpt_server->sv_name,
1273 (err < 0) ? "got error" : "sent",
1274 (err < 0) ? err : sent, xdr->len);
1275 svc_xprt_deferred_close(xprt);
1276 mutex_unlock(&xprt->xpt_mutex);
1277 return -EAGAIN;
1278}
1279
1280static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1281 struct net *net,
1282 struct sockaddr *sa, int salen,
1283 int flags)
1284{
1285 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1286}
1287
1288static const struct svc_xprt_ops svc_tcp_ops = {
1289 .xpo_create = svc_tcp_create,
1290 .xpo_recvfrom = svc_tcp_recvfrom,
1291 .xpo_sendto = svc_tcp_sendto,
1292 .xpo_result_payload = svc_sock_result_payload,
1293 .xpo_release_ctxt = svc_tcp_release_ctxt,
1294 .xpo_detach = svc_tcp_sock_detach,
1295 .xpo_free = svc_sock_free,
1296 .xpo_has_wspace = svc_tcp_has_wspace,
1297 .xpo_accept = svc_tcp_accept,
1298 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1299 .xpo_handshake = svc_tcp_handshake,
1300};
1301
1302static struct svc_xprt_class svc_tcp_class = {
1303 .xcl_name = "tcp",
1304 .xcl_owner = THIS_MODULE,
1305 .xcl_ops = &svc_tcp_ops,
1306 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1307 .xcl_ident = XPRT_TRANSPORT_TCP,
1308};
1309
1310void svc_init_xprt_sock(void)
1311{
1312 svc_reg_xprt_class(&svc_tcp_class);
1313 svc_reg_xprt_class(&svc_udp_class);
1314}
1315
1316void svc_cleanup_xprt_sock(void)
1317{
1318 svc_unreg_xprt_class(&svc_tcp_class);
1319 svc_unreg_xprt_class(&svc_udp_class);
1320}
1321
1322static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1323{
1324 struct sock *sk = svsk->sk_sk;
1325
1326 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1327 &svsk->sk_xprt, serv);
1328 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1329 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1330 if (sk->sk_state == TCP_LISTEN) {
1331 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1332 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1333 sk->sk_data_ready = svc_tcp_listen_data_ready;
1334 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1335 } else {
1336 sk->sk_state_change = svc_tcp_state_change;
1337 sk->sk_data_ready = svc_data_ready;
1338 sk->sk_write_space = svc_write_space;
1339
1340 svsk->sk_marker = xdr_zero;
1341 svsk->sk_tcplen = 0;
1342 svsk->sk_datalen = 0;
1343 memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1344
1345 tcp_sock_set_nodelay(sk);
1346
1347 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1348 switch (sk->sk_state) {
1349 case TCP_SYN_RECV:
1350 case TCP_ESTABLISHED:
1351 break;
1352 default:
1353 svc_xprt_deferred_close(&svsk->sk_xprt);
1354 }
1355 }
1356}
1357
1358void svc_sock_update_bufs(struct svc_serv *serv)
1359{
1360 /*
1361 * The number of server threads has changed. Update
1362 * rcvbuf and sndbuf accordingly on all sockets
1363 */
1364 struct svc_sock *svsk;
1365
1366 spin_lock_bh(&serv->sv_lock);
1367 list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1368 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1369 spin_unlock_bh(&serv->sv_lock);
1370}
1371
1372/*
1373 * Initialize socket for RPC use and create svc_sock struct
1374 */
1375static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1376 struct socket *sock,
1377 int flags)
1378{
1379 struct svc_sock *svsk;
1380 struct sock *inet;
1381 int pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1382
1383 svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1384 if (!svsk)
1385 return ERR_PTR(-ENOMEM);
1386
1387 inet = sock->sk;
1388
1389 if (pmap_register) {
1390 int err;
1391
1392 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1393 inet->sk_protocol,
1394 ntohs(inet_sk(inet)->inet_sport));
1395 if (err < 0) {
1396 kfree(svsk);
1397 return ERR_PTR(err);
1398 }
1399 }
1400
1401 svsk->sk_sock = sock;
1402 svsk->sk_sk = inet;
1403 svsk->sk_ostate = inet->sk_state_change;
1404 svsk->sk_odata = inet->sk_data_ready;
1405 svsk->sk_owspace = inet->sk_write_space;
1406 /*
1407 * This barrier is necessary in order to prevent race condition
1408 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1409 * when calling callbacks above.
1410 */
1411 wmb();
1412 inet->sk_user_data = svsk;
1413
1414 /* Initialize the socket */
1415 if (sock->type == SOCK_DGRAM)
1416 svc_udp_init(svsk, serv);
1417 else
1418 svc_tcp_init(svsk, serv);
1419
1420 trace_svcsock_new(svsk, sock);
1421 return svsk;
1422}
1423
1424/**
1425 * svc_addsock - add a listener socket to an RPC service
1426 * @serv: pointer to RPC service to which to add a new listener
1427 * @net: caller's network namespace
1428 * @fd: file descriptor of the new listener
1429 * @name_return: pointer to buffer to fill in with name of listener
1430 * @len: size of the buffer
1431 * @cred: credential
1432 *
1433 * Fills in socket name and returns positive length of name if successful.
1434 * Name is terminated with '\n'. On error, returns a negative errno
1435 * value.
1436 */
1437int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1438 char *name_return, const size_t len, const struct cred *cred)
1439{
1440 int err = 0;
1441 struct socket *so = sockfd_lookup(fd, &err);
1442 struct svc_sock *svsk = NULL;
1443 struct sockaddr_storage addr;
1444 struct sockaddr *sin = (struct sockaddr *)&addr;
1445 int salen;
1446
1447 if (!so)
1448 return err;
1449 err = -EINVAL;
1450 if (sock_net(so->sk) != net)
1451 goto out;
1452 err = -EAFNOSUPPORT;
1453 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1454 goto out;
1455 err = -EPROTONOSUPPORT;
1456 if (so->sk->sk_protocol != IPPROTO_TCP &&
1457 so->sk->sk_protocol != IPPROTO_UDP)
1458 goto out;
1459 err = -EISCONN;
1460 if (so->state > SS_UNCONNECTED)
1461 goto out;
1462 err = -ENOENT;
1463 if (!try_module_get(THIS_MODULE))
1464 goto out;
1465 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1466 if (IS_ERR(svsk)) {
1467 module_put(THIS_MODULE);
1468 err = PTR_ERR(svsk);
1469 goto out;
1470 }
1471 salen = kernel_getsockname(svsk->sk_sock, sin);
1472 if (salen >= 0)
1473 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1474 svsk->sk_xprt.xpt_cred = get_cred(cred);
1475 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1476 return svc_one_sock_name(svsk, name_return, len);
1477out:
1478 sockfd_put(so);
1479 return err;
1480}
1481EXPORT_SYMBOL_GPL(svc_addsock);
1482
1483/*
1484 * Create socket for RPC service.
1485 */
1486static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1487 int protocol,
1488 struct net *net,
1489 struct sockaddr *sin, int len,
1490 int flags)
1491{
1492 struct svc_sock *svsk;
1493 struct socket *sock;
1494 int error;
1495 int type;
1496 struct sockaddr_storage addr;
1497 struct sockaddr *newsin = (struct sockaddr *)&addr;
1498 int newlen;
1499 int family;
1500
1501 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1502 printk(KERN_WARNING "svc: only UDP and TCP "
1503 "sockets supported\n");
1504 return ERR_PTR(-EINVAL);
1505 }
1506
1507 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1508 switch (sin->sa_family) {
1509 case AF_INET6:
1510 family = PF_INET6;
1511 break;
1512 case AF_INET:
1513 family = PF_INET;
1514 break;
1515 default:
1516 return ERR_PTR(-EINVAL);
1517 }
1518
1519 error = __sock_create(net, family, type, protocol, &sock, 1);
1520 if (error < 0)
1521 return ERR_PTR(error);
1522
1523 svc_reclassify_socket(sock);
1524
1525 /*
1526 * If this is an PF_INET6 listener, we want to avoid
1527 * getting requests from IPv4 remotes. Those should
1528 * be shunted to a PF_INET listener via rpcbind.
1529 */
1530 if (family == PF_INET6)
1531 ip6_sock_set_v6only(sock->sk);
1532 if (type == SOCK_STREAM)
1533 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1534 error = kernel_bind(sock, sin, len);
1535 if (error < 0)
1536 goto bummer;
1537
1538 error = kernel_getsockname(sock, newsin);
1539 if (error < 0)
1540 goto bummer;
1541 newlen = error;
1542
1543 if (protocol == IPPROTO_TCP) {
1544 sk_net_refcnt_upgrade(sock->sk);
1545 if ((error = kernel_listen(sock, 64)) < 0)
1546 goto bummer;
1547 }
1548
1549 svsk = svc_setup_socket(serv, sock, flags);
1550 if (IS_ERR(svsk)) {
1551 error = PTR_ERR(svsk);
1552 goto bummer;
1553 }
1554 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1555 return (struct svc_xprt *)svsk;
1556bummer:
1557 sock_release(sock);
1558 return ERR_PTR(error);
1559}
1560
1561/*
1562 * Detach the svc_sock from the socket so that no
1563 * more callbacks occur.
1564 */
1565static void svc_sock_detach(struct svc_xprt *xprt)
1566{
1567 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1568 struct sock *sk = svsk->sk_sk;
1569
1570 /* put back the old socket callbacks */
1571 lock_sock(sk);
1572 sk->sk_state_change = svsk->sk_ostate;
1573 sk->sk_data_ready = svsk->sk_odata;
1574 sk->sk_write_space = svsk->sk_owspace;
1575 sk->sk_user_data = NULL;
1576 release_sock(sk);
1577}
1578
1579/*
1580 * Disconnect the socket, and reset the callbacks
1581 */
1582static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1583{
1584 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1585
1586 tls_handshake_close(svsk->sk_sock);
1587
1588 svc_sock_detach(xprt);
1589
1590 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1591 svc_tcp_clear_pages(svsk);
1592 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1593 }
1594}
1595
1596/*
1597 * Free the svc_sock's socket resources and the svc_sock itself.
1598 */
1599static void svc_sock_free(struct svc_xprt *xprt)
1600{
1601 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1602 struct socket *sock = svsk->sk_sock;
1603
1604 trace_svcsock_free(svsk, sock);
1605
1606 tls_handshake_cancel(sock->sk);
1607 if (sock->file)
1608 sockfd_put(sock);
1609 else
1610 sock_release(sock);
1611
1612 page_frag_cache_drain(&svsk->sk_frag_cache);
1613 kfree(svsk);
1614}