projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'probes-fixes-v6.16-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-block.git] / net / sunrpc / svcsock.c
... / ...
CommitLineData
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, rqstp->rq_maxpages, xdr);
717
718 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
719 count, rqstp->rq_res.len);
720 err = sock_sendmsg(svsk->sk_sock, &msg);
721 if (err == -ECONNREFUSED) {
722 /* ICMP error on earlier request. */
723 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
724 count, rqstp->rq_res.len);
725 err = sock_sendmsg(svsk->sk_sock, &msg);
726 }
727
728 trace_svcsock_udp_send(xprt, err);
729
730 mutex_unlock(&xprt->xpt_mutex);
731 return err;
732
733out_notconn:
734 mutex_unlock(&xprt->xpt_mutex);
735 return -ENOTCONN;
736}
737
738static int svc_udp_has_wspace(struct svc_xprt *xprt)
739{
740 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
741 struct svc_serv *serv = xprt->xpt_server;
742 unsigned long required;
743
744 /*
745 * Set the SOCK_NOSPACE flag before checking the available
746 * sock space.
747 */
748 set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
749 required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
750 if (required*2 > sock_wspace(svsk->sk_sk))
751 return 0;
752 clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
753 return 1;
754}
755
756static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
757{
758 BUG();
759 return NULL;
760}
761
762static void svc_udp_kill_temp_xprt(struct svc_xprt *xprt)
763{
764}
765
766static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
767 struct net *net,
768 struct sockaddr *sa, int salen,
769 int flags)
770{
771 return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
772}
773
774static const struct svc_xprt_ops svc_udp_ops = {
775 .xpo_create = svc_udp_create,
776 .xpo_recvfrom = svc_udp_recvfrom,
777 .xpo_sendto = svc_udp_sendto,
778 .xpo_result_payload = svc_sock_result_payload,
779 .xpo_release_ctxt = svc_udp_release_ctxt,
780 .xpo_detach = svc_sock_detach,
781 .xpo_free = svc_sock_free,
782 .xpo_has_wspace = svc_udp_has_wspace,
783 .xpo_accept = svc_udp_accept,
784 .xpo_kill_temp_xprt = svc_udp_kill_temp_xprt,
785};
786
787static struct svc_xprt_class svc_udp_class = {
788 .xcl_name = "udp",
789 .xcl_owner = THIS_MODULE,
790 .xcl_ops = &svc_udp_ops,
791 .xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
792 .xcl_ident = XPRT_TRANSPORT_UDP,
793};
794
795static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
796{
797 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
798 &svsk->sk_xprt, serv);
799 clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
800 svsk->sk_sk->sk_data_ready = svc_data_ready;
801 svsk->sk_sk->sk_write_space = svc_write_space;
802
803 /* initialise setting must have enough space to
804 * receive and respond to one request.
805 * svc_udp_recvfrom will re-adjust if necessary
806 */
807 svc_sock_setbufsize(svsk, 3);
808
809 /* data might have come in before data_ready set up */
810 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
811 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
812
813 /* make sure we get destination address info */
814 switch (svsk->sk_sk->sk_family) {
815 case AF_INET:
816 ip_sock_set_pktinfo(svsk->sk_sock->sk);
817 break;
818 case AF_INET6:
819 ip6_sock_set_recvpktinfo(svsk->sk_sock->sk);
820 break;
821 default:
822 BUG();
823 }
824}
825
826/*
827 * A data_ready event on a listening socket means there's a connection
828 * pending. Do not use state_change as a substitute for it.
829 */
830static void svc_tcp_listen_data_ready(struct sock *sk)
831{
832 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
833
834 trace_sk_data_ready(sk);
835
836 /*
837 * This callback may called twice when a new connection
838 * is established as a child socket inherits everything
839 * from a parent LISTEN socket.
840 * 1) data_ready method of the parent socket will be called
841 * when one of child sockets become ESTABLISHED.
842 * 2) data_ready method of the child socket may be called
843 * when it receives data before the socket is accepted.
844 * In case of 2, we should ignore it silently and DO NOT
845 * dereference svsk.
846 */
847 if (sk->sk_state != TCP_LISTEN)
848 return;
849
850 if (svsk) {
851 /* Refer to svc_setup_socket() for details. */
852 rmb();
853 svsk->sk_odata(sk);
854 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
855 svc_xprt_enqueue(&svsk->sk_xprt);
856 }
857}
858
859/*
860 * A state change on a connected socket means it's dying or dead.
861 */
862static void svc_tcp_state_change(struct sock *sk)
863{
864 struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
865
866 if (svsk) {
867 /* Refer to svc_setup_socket() for details. */
868 rmb();
869 svsk->sk_ostate(sk);
870 trace_svcsock_tcp_state(&svsk->sk_xprt, svsk->sk_sock);
871 if (sk->sk_state != TCP_ESTABLISHED)
872 svc_xprt_deferred_close(&svsk->sk_xprt);
873 }
874}
875
876/*
877 * Accept a TCP connection
878 */
879static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
880{
881 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
882 struct sockaddr_storage addr;
883 struct sockaddr *sin = (struct sockaddr *) &addr;
884 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
885 struct socket *sock = svsk->sk_sock;
886 struct socket *newsock;
887 struct svc_sock *newsvsk;
888 int err, slen;
889
890 if (!sock)
891 return NULL;
892
893 clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
894 err = kernel_accept(sock, &newsock, O_NONBLOCK);
895 if (err < 0) {
896 if (err != -EAGAIN)
897 trace_svcsock_accept_err(xprt, serv->sv_name, err);
898 return NULL;
899 }
900 if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL)))
901 return NULL;
902
903 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
904
905 err = kernel_getpeername(newsock, sin);
906 if (err < 0) {
907 trace_svcsock_getpeername_err(xprt, serv->sv_name, err);
908 goto failed; /* aborted connection or whatever */
909 }
910 slen = err;
911
912 /* Reset the inherited callbacks before calling svc_setup_socket */
913 newsock->sk->sk_state_change = svsk->sk_ostate;
914 newsock->sk->sk_data_ready = svsk->sk_odata;
915 newsock->sk->sk_write_space = svsk->sk_owspace;
916
917 /* make sure that a write doesn't block forever when
918 * low on memory
919 */
920 newsock->sk->sk_sndtimeo = HZ*30;
921
922 newsvsk = svc_setup_socket(serv, newsock,
923 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
924 if (IS_ERR(newsvsk))
925 goto failed;
926 svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
927 err = kernel_getsockname(newsock, sin);
928 slen = err;
929 if (unlikely(err < 0))
930 slen = offsetof(struct sockaddr, sa_data);
931 svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
932
933 if (sock_is_loopback(newsock->sk))
934 set_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
935 else
936 clear_bit(XPT_LOCAL, &newsvsk->sk_xprt.xpt_flags);
937 if (serv->sv_stats)
938 serv->sv_stats->nettcpconn++;
939
940 return &newsvsk->sk_xprt;
941
942failed:
943 sockfd_put(newsock);
944 return NULL;
945}
946
947static size_t svc_tcp_restore_pages(struct svc_sock *svsk,
948 struct svc_rqst *rqstp)
949{
950 size_t len = svsk->sk_datalen;
951 unsigned int i, npages;
952
953 if (!len)
954 return 0;
955 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
956 for (i = 0; i < npages; i++) {
957 if (rqstp->rq_pages[i] != NULL)
958 put_page(rqstp->rq_pages[i]);
959 BUG_ON(svsk->sk_pages[i] == NULL);
960 rqstp->rq_pages[i] = svsk->sk_pages[i];
961 svsk->sk_pages[i] = NULL;
962 }
963 rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
964 return len;
965}
966
967static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
968{
969 unsigned int i, len, npages;
970
971 if (svsk->sk_datalen == 0)
972 return;
973 len = svsk->sk_datalen;
974 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
975 for (i = 0; i < npages; i++) {
976 svsk->sk_pages[i] = rqstp->rq_pages[i];
977 rqstp->rq_pages[i] = NULL;
978 }
979}
980
981static void svc_tcp_clear_pages(struct svc_sock *svsk)
982{
983 unsigned int i, len, npages;
984
985 if (svsk->sk_datalen == 0)
986 goto out;
987 len = svsk->sk_datalen;
988 npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
989 for (i = 0; i < npages; i++) {
990 if (svsk->sk_pages[i] == NULL) {
991 WARN_ON_ONCE(1);
992 continue;
993 }
994 put_page(svsk->sk_pages[i]);
995 svsk->sk_pages[i] = NULL;
996 }
997out:
998 svsk->sk_tcplen = 0;
999 svsk->sk_datalen = 0;
1000}
1001
1002/*
1003 * Receive fragment record header into sk_marker.
1004 */
1005static ssize_t svc_tcp_read_marker(struct svc_sock *svsk,
1006 struct svc_rqst *rqstp)
1007{
1008 ssize_t want, len;
1009
1010 /* If we haven't gotten the record length yet,
1011 * get the next four bytes.
1012 */
1013 if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
1014 struct msghdr msg = { NULL };
1015 struct kvec iov;
1016
1017 want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
1018 iov.iov_base = ((char *)&svsk->sk_marker) + svsk->sk_tcplen;
1019 iov.iov_len = want;
1020 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, want);
1021 len = svc_tcp_sock_recv_cmsg(svsk, &msg);
1022 if (len < 0)
1023 return len;
1024 svsk->sk_tcplen += len;
1025 if (len < want) {
1026 /* call again to read the remaining bytes */
1027 goto err_short;
1028 }
1029 trace_svcsock_marker(&svsk->sk_xprt, svsk->sk_marker);
1030 if (svc_sock_reclen(svsk) + svsk->sk_datalen >
1031 svsk->sk_xprt.xpt_server->sv_max_mesg)
1032 goto err_too_large;
1033 }
1034 return svc_sock_reclen(svsk);
1035
1036err_too_large:
1037 net_notice_ratelimited("svc: %s %s RPC fragment too large: %d\n",
1038 __func__, svsk->sk_xprt.xpt_server->sv_name,
1039 svc_sock_reclen(svsk));
1040 svc_xprt_deferred_close(&svsk->sk_xprt);
1041err_short:
1042 return -EAGAIN;
1043}
1044
1045static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
1046{
1047 struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
1048 struct rpc_rqst *req = NULL;
1049 struct kvec *src, *dst;
1050 __be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1051 __be32 xid = *p;
1052
1053 if (!bc_xprt)
1054 return -EAGAIN;
1055 spin_lock(&bc_xprt->queue_lock);
1056 req = xprt_lookup_rqst(bc_xprt, xid);
1057 if (!req)
1058 goto unlock_eagain;
1059
1060 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1061 /*
1062 * XXX!: cheating for now! Only copying HEAD.
1063 * But we know this is good enough for now (in fact, for any
1064 * callback reply in the forseeable future).
1065 */
1066 dst = &req->rq_private_buf.head[0];
1067 src = &rqstp->rq_arg.head[0];
1068 if (dst->iov_len < src->iov_len)
1069 goto unlock_eagain; /* whatever; just giving up. */
1070 memcpy(dst->iov_base, src->iov_base, src->iov_len);
1071 xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1072 rqstp->rq_arg.len = 0;
1073 spin_unlock(&bc_xprt->queue_lock);
1074 return 0;
1075unlock_eagain:
1076 spin_unlock(&bc_xprt->queue_lock);
1077 return -EAGAIN;
1078}
1079
1080static void svc_tcp_fragment_received(struct svc_sock *svsk)
1081{
1082 /* If we have more data, signal svc_xprt_enqueue() to try again */
1083 svsk->sk_tcplen = 0;
1084 svsk->sk_marker = xdr_zero;
1085}
1086
1087/**
1088 * svc_tcp_recvfrom - Receive data from a TCP socket
1089 * @rqstp: request structure into which to receive an RPC Call
1090 *
1091 * Called in a loop when XPT_DATA has been set.
1092 *
1093 * Read the 4-byte stream record marker, then use the record length
1094 * in that marker to set up exactly the resources needed to receive
1095 * the next RPC message into @rqstp.
1096 *
1097 * Returns:
1098 * On success, the number of bytes in a received RPC Call, or
1099 * %0 if a complete RPC Call message was not ready to return
1100 *
1101 * The zero return case handles partial receives and callback Replies.
1102 * The state of a partial receive is preserved in the svc_sock for
1103 * the next call to svc_tcp_recvfrom.
1104 */
1105static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1106{
1107 struct svc_sock *svsk =
1108 container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1109 struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1110 size_t want, base;
1111 ssize_t len;
1112 __be32 *p;
1113 __be32 calldir;
1114
1115 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1116 len = svc_tcp_read_marker(svsk, rqstp);
1117 if (len < 0)
1118 goto error;
1119
1120 base = svc_tcp_restore_pages(svsk, rqstp);
1121 want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1122 len = svc_tcp_read_msg(rqstp, base + want, base);
1123 if (len >= 0) {
1124 trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1125 svsk->sk_tcplen += len;
1126 svsk->sk_datalen += len;
1127 }
1128 if (len != want || !svc_sock_final_rec(svsk))
1129 goto err_incomplete;
1130 if (svsk->sk_datalen < 8)
1131 goto err_nuts;
1132
1133 rqstp->rq_arg.len = svsk->sk_datalen;
1134 rqstp->rq_arg.page_base = 0;
1135 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1136 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1137 rqstp->rq_arg.page_len = 0;
1138 } else
1139 rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1140
1141 rqstp->rq_xprt_ctxt = NULL;
1142 rqstp->rq_prot = IPPROTO_TCP;
1143 if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1144 set_bit(RQ_LOCAL, &rqstp->rq_flags);
1145 else
1146 clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1147
1148 p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1149 calldir = p[1];
1150 if (calldir)
1151 len = receive_cb_reply(svsk, rqstp);
1152
1153 /* Reset TCP read info */
1154 svsk->sk_datalen = 0;
1155 svc_tcp_fragment_received(svsk);
1156
1157 if (len < 0)
1158 goto error;
1159
1160 svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1161 if (serv->sv_stats)
1162 serv->sv_stats->nettcpcnt++;
1163
1164 svc_sock_secure_port(rqstp);
1165 svc_xprt_received(rqstp->rq_xprt);
1166 return rqstp->rq_arg.len;
1167
1168err_incomplete:
1169 svc_tcp_save_pages(svsk, rqstp);
1170 if (len < 0 && len != -EAGAIN)
1171 goto err_delete;
1172 if (len == want)
1173 svc_tcp_fragment_received(svsk);
1174 else
1175 trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1176 svc_sock_reclen(svsk),
1177 svsk->sk_tcplen - sizeof(rpc_fraghdr));
1178 goto err_noclose;
1179error:
1180 if (len != -EAGAIN)
1181 goto err_delete;
1182 trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1183 goto err_noclose;
1184err_nuts:
1185 svsk->sk_datalen = 0;
1186err_delete:
1187 trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1188 svc_xprt_deferred_close(&svsk->sk_xprt);
1189err_noclose:
1190 svc_xprt_received(rqstp->rq_xprt);
1191 return 0; /* record not complete */
1192}
1193
1194/*
1195 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1196 * copy operations in this path. Therefore the caller must ensure
1197 * that the pages backing @xdr are unchanging.
1198 */
1199static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1200 rpc_fraghdr marker, unsigned int *sentp)
1201{
1202 struct msghdr msg = {
1203 .msg_flags = MSG_SPLICE_PAGES,
1204 };
1205 unsigned int count;
1206 void *buf;
1207 int ret;
1208
1209 *sentp = 0;
1210
1211 /* The stream record marker is copied into a temporary page
1212 * fragment buffer so that it can be included in rq_bvec.
1213 */
1214 buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1215 GFP_KERNEL);
1216 if (!buf)
1217 return -ENOMEM;
1218 memcpy(buf, &marker, sizeof(marker));
1219 bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1220
1221 count = xdr_buf_to_bvec(rqstp->rq_bvec + 1, rqstp->rq_maxpages,
1222 &rqstp->rq_res);
1223
1224 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1225 1 + count, sizeof(marker) + rqstp->rq_res.len);
1226 ret = sock_sendmsg(svsk->sk_sock, &msg);
1227 page_frag_free(buf);
1228 if (ret < 0)
1229 return ret;
1230 *sentp += ret;
1231 return 0;
1232}
1233
1234/**
1235 * svc_tcp_sendto - Send out a reply on a TCP socket
1236 * @rqstp: completed svc_rqst
1237 *
1238 * xpt_mutex ensures @rqstp's whole message is written to the socket
1239 * without interruption.
1240 *
1241 * Returns the number of bytes sent, or a negative errno.
1242 */
1243static int svc_tcp_sendto(struct svc_rqst *rqstp)
1244{
1245 struct svc_xprt *xprt = rqstp->rq_xprt;
1246 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1247 struct xdr_buf *xdr = &rqstp->rq_res;
1248 rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1249 (u32)xdr->len);
1250 unsigned int sent;
1251 int err;
1252
1253 svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1254 rqstp->rq_xprt_ctxt = NULL;
1255
1256 mutex_lock(&xprt->xpt_mutex);
1257 if (svc_xprt_is_dead(xprt))
1258 goto out_notconn;
1259 err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1260 trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1261 if (err < 0 || sent != (xdr->len + sizeof(marker)))
1262 goto out_close;
1263 mutex_unlock(&xprt->xpt_mutex);
1264 return sent;
1265
1266out_notconn:
1267 mutex_unlock(&xprt->xpt_mutex);
1268 return -ENOTCONN;
1269out_close:
1270 pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1271 xprt->xpt_server->sv_name,
1272 (err < 0) ? "got error" : "sent",
1273 (err < 0) ? err : sent, xdr->len);
1274 svc_xprt_deferred_close(xprt);
1275 mutex_unlock(&xprt->xpt_mutex);
1276 return -EAGAIN;
1277}
1278
1279static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1280 struct net *net,
1281 struct sockaddr *sa, int salen,
1282 int flags)
1283{
1284 return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1285}
1286
1287static const struct svc_xprt_ops svc_tcp_ops = {
1288 .xpo_create = svc_tcp_create,
1289 .xpo_recvfrom = svc_tcp_recvfrom,
1290 .xpo_sendto = svc_tcp_sendto,
1291 .xpo_result_payload = svc_sock_result_payload,
1292 .xpo_release_ctxt = svc_tcp_release_ctxt,
1293 .xpo_detach = svc_tcp_sock_detach,
1294 .xpo_free = svc_sock_free,
1295 .xpo_has_wspace = svc_tcp_has_wspace,
1296 .xpo_accept = svc_tcp_accept,
1297 .xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1298 .xpo_handshake = svc_tcp_handshake,
1299};
1300
1301static struct svc_xprt_class svc_tcp_class = {
1302 .xcl_name = "tcp",
1303 .xcl_owner = THIS_MODULE,
1304 .xcl_ops = &svc_tcp_ops,
1305 .xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1306 .xcl_ident = XPRT_TRANSPORT_TCP,
1307};
1308
1309void svc_init_xprt_sock(void)
1310{
1311 svc_reg_xprt_class(&svc_tcp_class);
1312 svc_reg_xprt_class(&svc_udp_class);
1313}
1314
1315void svc_cleanup_xprt_sock(void)
1316{
1317 svc_unreg_xprt_class(&svc_tcp_class);
1318 svc_unreg_xprt_class(&svc_udp_class);
1319}
1320
1321static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1322{
1323 struct sock *sk = svsk->sk_sk;
1324
1325 svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1326 &svsk->sk_xprt, serv);
1327 set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1328 set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1329 if (sk->sk_state == TCP_LISTEN) {
1330 strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1331 set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1332 sk->sk_data_ready = svc_tcp_listen_data_ready;
1333 set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1334 } else {
1335 sk->sk_state_change = svc_tcp_state_change;
1336 sk->sk_data_ready = svc_data_ready;
1337 sk->sk_write_space = svc_write_space;
1338
1339 svsk->sk_marker = xdr_zero;
1340 svsk->sk_tcplen = 0;
1341 svsk->sk_datalen = 0;
1342 memset(&svsk->sk_pages[0], 0,
1343 svsk->sk_maxpages * sizeof(struct page *));
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 unsigned long pages;
1383
1384 pages = svc_serv_maxpages(serv);
1385 svsk = kzalloc(struct_size(svsk, sk_pages, pages), GFP_KERNEL);
1386 if (!svsk)
1387 return ERR_PTR(-ENOMEM);
1388 svsk->sk_maxpages = pages;
1389
1390 inet = sock->sk;
1391
1392 if (pmap_register) {
1393 int err;
1394
1395 err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1396 inet->sk_protocol,
1397 ntohs(inet_sk(inet)->inet_sport));
1398 if (err < 0) {
1399 kfree(svsk);
1400 return ERR_PTR(err);
1401 }
1402 }
1403
1404 svsk->sk_sock = sock;
1405 svsk->sk_sk = inet;
1406 svsk->sk_ostate = inet->sk_state_change;
1407 svsk->sk_odata = inet->sk_data_ready;
1408 svsk->sk_owspace = inet->sk_write_space;
1409 /*
1410 * This barrier is necessary in order to prevent race condition
1411 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1412 * when calling callbacks above.
1413 */
1414 wmb();
1415 inet->sk_user_data = svsk;
1416
1417 /* Initialize the socket */
1418 if (sock->type == SOCK_DGRAM)
1419 svc_udp_init(svsk, serv);
1420 else
1421 svc_tcp_init(svsk, serv);
1422
1423 trace_svcsock_new(svsk, sock);
1424 return svsk;
1425}
1426
1427/**
1428 * svc_addsock - add a listener socket to an RPC service
1429 * @serv: pointer to RPC service to which to add a new listener
1430 * @net: caller's network namespace
1431 * @fd: file descriptor of the new listener
1432 * @name_return: pointer to buffer to fill in with name of listener
1433 * @len: size of the buffer
1434 * @cred: credential
1435 *
1436 * Fills in socket name and returns positive length of name if successful.
1437 * Name is terminated with '\n'. On error, returns a negative errno
1438 * value.
1439 */
1440int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1441 char *name_return, const size_t len, const struct cred *cred)
1442{
1443 int err = 0;
1444 struct socket *so = sockfd_lookup(fd, &err);
1445 struct svc_sock *svsk = NULL;
1446 struct sockaddr_storage addr;
1447 struct sockaddr *sin = (struct sockaddr *)&addr;
1448 int salen;
1449
1450 if (!so)
1451 return err;
1452 err = -EINVAL;
1453 if (sock_net(so->sk) != net)
1454 goto out;
1455 err = -EAFNOSUPPORT;
1456 if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1457 goto out;
1458 err = -EPROTONOSUPPORT;
1459 if (so->sk->sk_protocol != IPPROTO_TCP &&
1460 so->sk->sk_protocol != IPPROTO_UDP)
1461 goto out;
1462 err = -EISCONN;
1463 if (so->state > SS_UNCONNECTED)
1464 goto out;
1465 err = -ENOENT;
1466 if (!try_module_get(THIS_MODULE))
1467 goto out;
1468 svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1469 if (IS_ERR(svsk)) {
1470 module_put(THIS_MODULE);
1471 err = PTR_ERR(svsk);
1472 goto out;
1473 }
1474 salen = kernel_getsockname(svsk->sk_sock, sin);
1475 if (salen >= 0)
1476 svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1477 svsk->sk_xprt.xpt_cred = get_cred(cred);
1478 svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1479 return svc_one_sock_name(svsk, name_return, len);
1480out:
1481 sockfd_put(so);
1482 return err;
1483}
1484EXPORT_SYMBOL_GPL(svc_addsock);
1485
1486/*
1487 * Create socket for RPC service.
1488 */
1489static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1490 int protocol,
1491 struct net *net,
1492 struct sockaddr *sin, int len,
1493 int flags)
1494{
1495 struct svc_sock *svsk;
1496 struct socket *sock;
1497 int error;
1498 int type;
1499 struct sockaddr_storage addr;
1500 struct sockaddr *newsin = (struct sockaddr *)&addr;
1501 int newlen;
1502 int family;
1503
1504 if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1505 printk(KERN_WARNING "svc: only UDP and TCP "
1506 "sockets supported\n");
1507 return ERR_PTR(-EINVAL);
1508 }
1509
1510 type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1511 switch (sin->sa_family) {
1512 case AF_INET6:
1513 family = PF_INET6;
1514 break;
1515 case AF_INET:
1516 family = PF_INET;
1517 break;
1518 default:
1519 return ERR_PTR(-EINVAL);
1520 }
1521
1522 error = __sock_create(net, family, type, protocol, &sock, 1);
1523 if (error < 0)
1524 return ERR_PTR(error);
1525
1526 svc_reclassify_socket(sock);
1527
1528 /*
1529 * If this is an PF_INET6 listener, we want to avoid
1530 * getting requests from IPv4 remotes. Those should
1531 * be shunted to a PF_INET listener via rpcbind.
1532 */
1533 if (family == PF_INET6)
1534 ip6_sock_set_v6only(sock->sk);
1535 if (type == SOCK_STREAM)
1536 sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1537 error = kernel_bind(sock, sin, len);
1538 if (error < 0)
1539 goto bummer;
1540
1541 error = kernel_getsockname(sock, newsin);
1542 if (error < 0)
1543 goto bummer;
1544 newlen = error;
1545
1546 if (protocol == IPPROTO_TCP) {
1547 sk_net_refcnt_upgrade(sock->sk);
1548 if ((error = kernel_listen(sock, SOMAXCONN)) < 0)
1549 goto bummer;
1550 }
1551
1552 svsk = svc_setup_socket(serv, sock, flags);
1553 if (IS_ERR(svsk)) {
1554 error = PTR_ERR(svsk);
1555 goto bummer;
1556 }
1557 svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1558 return (struct svc_xprt *)svsk;
1559bummer:
1560 sock_release(sock);
1561 return ERR_PTR(error);
1562}
1563
1564/*
1565 * Detach the svc_sock from the socket so that no
1566 * more callbacks occur.
1567 */
1568static void svc_sock_detach(struct svc_xprt *xprt)
1569{
1570 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1571 struct sock *sk = svsk->sk_sk;
1572
1573 /* put back the old socket callbacks */
1574 lock_sock(sk);
1575 sk->sk_state_change = svsk->sk_ostate;
1576 sk->sk_data_ready = svsk->sk_odata;
1577 sk->sk_write_space = svsk->sk_owspace;
1578 sk->sk_user_data = NULL;
1579 release_sock(sk);
1580}
1581
1582/*
1583 * Disconnect the socket, and reset the callbacks
1584 */
1585static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1586{
1587 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1588
1589 tls_handshake_close(svsk->sk_sock);
1590
1591 svc_sock_detach(xprt);
1592
1593 if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1594 svc_tcp_clear_pages(svsk);
1595 kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1596 }
1597}
1598
1599/*
1600 * Free the svc_sock's socket resources and the svc_sock itself.
1601 */
1602static void svc_sock_free(struct svc_xprt *xprt)
1603{
1604 struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1605 struct socket *sock = svsk->sk_sock;
1606
1607 trace_svcsock_free(svsk, sock);
1608
1609 tls_handshake_cancel(sock->sk);
1610 if (sock->file)
1611 sockfd_put(sock);
1612 else
1613 sock_release(sock);
1614
1615 page_frag_cache_drain(&svsk->sk_frag_cache);
1616 kfree(svsk);
1617}
Linux 6.x block layer and io_uring tree(s)