Linux Audio

Check our new training course

Loading...
v6.8
   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	smp_wmb();
1088	tcp_set_rcvlowat(svsk->sk_sk, 1);
1089}
1090
1091/**
1092 * svc_tcp_recvfrom - Receive data from a TCP socket
1093 * @rqstp: request structure into which to receive an RPC Call
1094 *
1095 * Called in a loop when XPT_DATA has been set.
1096 *
1097 * Read the 4-byte stream record marker, then use the record length
1098 * in that marker to set up exactly the resources needed to receive
1099 * the next RPC message into @rqstp.
1100 *
1101 * Returns:
1102 *   On success, the number of bytes in a received RPC Call, or
1103 *   %0 if a complete RPC Call message was not ready to return
1104 *
1105 * The zero return case handles partial receives and callback Replies.
1106 * The state of a partial receive is preserved in the svc_sock for
1107 * the next call to svc_tcp_recvfrom.
1108 */
1109static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1110{
1111	struct svc_sock	*svsk =
1112		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1113	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1114	size_t want, base;
1115	ssize_t len;
 
1116	__be32 *p;
1117	__be32 calldir;
 
1118
1119	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1120	len = svc_tcp_read_marker(svsk, rqstp);
 
 
 
 
1121	if (len < 0)
1122		goto error;
1123
1124	base = svc_tcp_restore_pages(svsk, rqstp);
1125	want = len - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1126	len = svc_tcp_read_msg(rqstp, base + want, base);
 
 
 
 
 
 
 
 
 
 
1127	if (len >= 0) {
1128		trace_svcsock_tcp_recv(&svsk->sk_xprt, len);
1129		svsk->sk_tcplen += len;
1130		svsk->sk_datalen += len;
1131	}
1132	if (len != want || !svc_sock_final_rec(svsk))
1133		goto err_incomplete;
1134	if (svsk->sk_datalen < 8)
1135		goto err_nuts;
 
 
 
 
 
 
 
 
 
 
 
 
 
1136
1137	rqstp->rq_arg.len = svsk->sk_datalen;
1138	rqstp->rq_arg.page_base = 0;
1139	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1140		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1141		rqstp->rq_arg.page_len = 0;
1142	} else
1143		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1144
1145	rqstp->rq_xprt_ctxt   = NULL;
1146	rqstp->rq_prot	      = IPPROTO_TCP;
1147	if (test_bit(XPT_LOCAL, &svsk->sk_xprt.xpt_flags))
1148		set_bit(RQ_LOCAL, &rqstp->rq_flags);
1149	else
1150		clear_bit(RQ_LOCAL, &rqstp->rq_flags);
1151
1152	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1153	calldir = p[1];
1154	if (calldir)
1155		len = receive_cb_reply(svsk, rqstp);
1156
1157	/* Reset TCP read info */
1158	svsk->sk_datalen = 0;
1159	svc_tcp_fragment_received(svsk);
1160
1161	if (len < 0)
1162		goto error;
1163
1164	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1165	if (serv->sv_stats)
1166		serv->sv_stats->nettcpcnt++;
1167
1168	svc_sock_secure_port(rqstp);
1169	svc_xprt_received(rqstp->rq_xprt);
1170	return rqstp->rq_arg.len;
1171
1172err_incomplete:
1173	svc_tcp_save_pages(svsk, rqstp);
1174	if (len < 0 && len != -EAGAIN)
1175		goto err_delete;
1176	if (len == want)
1177		svc_tcp_fragment_received(svsk);
1178	else {
1179		/* Avoid more ->sk_data_ready() calls until the rest
1180		 * of the message has arrived. This reduces service
1181		 * thread wake-ups on large incoming messages. */
1182		tcp_set_rcvlowat(svsk->sk_sk,
1183				 svc_sock_reclen(svsk) - svsk->sk_tcplen);
1184
1185		trace_svcsock_tcp_recv_short(&svsk->sk_xprt,
1186				svc_sock_reclen(svsk),
1187				svsk->sk_tcplen - sizeof(rpc_fraghdr));
1188	}
1189	goto err_noclose;
1190error:
1191	if (len != -EAGAIN)
1192		goto err_delete;
1193	trace_svcsock_tcp_recv_eagain(&svsk->sk_xprt, 0);
1194	goto err_noclose;
1195err_nuts:
1196	svsk->sk_datalen = 0;
1197err_delete:
1198	trace_svcsock_tcp_recv_err(&svsk->sk_xprt, len);
1199	svc_xprt_deferred_close(&svsk->sk_xprt);
 
1200err_noclose:
1201	svc_xprt_received(rqstp->rq_xprt);
1202	return 0;	/* record not complete */
1203}
1204
1205/*
1206 * MSG_SPLICE_PAGES is used exclusively to reduce the number of
1207 * copy operations in this path. Therefore the caller must ensure
1208 * that the pages backing @xdr are unchanging.
1209 *
1210 * Note that the send is non-blocking. The caller has incremented
1211 * the reference count on each page backing the RPC message, and
1212 * the network layer will "put" these pages when transmission is
1213 * complete.
1214 *
1215 * This is safe for our RPC services because the memory backing
1216 * the head and tail components is never kmalloc'd. These always
1217 * come from pages in the svc_rqst::rq_pages array.
1218 */
1219static int svc_tcp_sendmsg(struct svc_sock *svsk, struct svc_rqst *rqstp,
1220			   rpc_fraghdr marker, unsigned int *sentp)
1221{
1222	struct msghdr msg = {
1223		.msg_flags	= MSG_SPLICE_PAGES,
1224	};
1225	unsigned int count;
1226	void *buf;
1227	int ret;
1228
1229	*sentp = 0;
1230
1231	/* The stream record marker is copied into a temporary page
1232	 * fragment buffer so that it can be included in rq_bvec.
1233	 */
1234	buf = page_frag_alloc(&svsk->sk_frag_cache, sizeof(marker),
1235			      GFP_KERNEL);
1236	if (!buf)
1237		return -ENOMEM;
1238	memcpy(buf, &marker, sizeof(marker));
1239	bvec_set_virt(rqstp->rq_bvec, buf, sizeof(marker));
1240
1241	count = xdr_buf_to_bvec(rqstp->rq_bvec + 1,
1242				ARRAY_SIZE(rqstp->rq_bvec) - 1, &rqstp->rq_res);
1243
1244	iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, rqstp->rq_bvec,
1245		      1 + count, sizeof(marker) + rqstp->rq_res.len);
1246	ret = sock_sendmsg(svsk->sk_sock, &msg);
1247	if (ret < 0)
1248		return ret;
1249	*sentp += ret;
1250	return 0;
1251}
1252
1253/**
1254 * svc_tcp_sendto - Send out a reply on a TCP socket
1255 * @rqstp: completed svc_rqst
1256 *
1257 * xpt_mutex ensures @rqstp's whole message is written to the socket
1258 * without interruption.
1259 *
1260 * Returns the number of bytes sent, or a negative errno.
1261 */
1262static int svc_tcp_sendto(struct svc_rqst *rqstp)
1263{
1264	struct svc_xprt *xprt = rqstp->rq_xprt;
1265	struct svc_sock	*svsk = container_of(xprt, struct svc_sock, sk_xprt);
1266	struct xdr_buf *xdr = &rqstp->rq_res;
1267	rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
1268					 (u32)xdr->len);
1269	unsigned int sent;
1270	int err;
1271
1272	svc_tcp_release_ctxt(xprt, rqstp->rq_xprt_ctxt);
1273	rqstp->rq_xprt_ctxt = NULL;
 
1274
1275	mutex_lock(&xprt->xpt_mutex);
1276	if (svc_xprt_is_dead(xprt))
1277		goto out_notconn;
1278	err = svc_tcp_sendmsg(svsk, rqstp, marker, &sent);
1279	trace_svcsock_tcp_send(xprt, err < 0 ? (long)err : sent);
1280	if (err < 0 || sent != (xdr->len + sizeof(marker)))
1281		goto out_close;
1282	mutex_unlock(&xprt->xpt_mutex);
1283	return sent;
1284
1285out_notconn:
1286	mutex_unlock(&xprt->xpt_mutex);
1287	return -ENOTCONN;
1288out_close:
1289	pr_notice("rpc-srv/tcp: %s: %s %d when sending %d bytes - shutting down socket\n",
1290		  xprt->xpt_server->sv_name,
1291		  (err < 0) ? "got error" : "sent",
1292		  (err < 0) ? err : sent, xdr->len);
1293	svc_xprt_deferred_close(xprt);
1294	mutex_unlock(&xprt->xpt_mutex);
1295	return -EAGAIN;
1296}
1297
1298static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1299				       struct net *net,
1300				       struct sockaddr *sa, int salen,
1301				       int flags)
1302{
1303	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1304}
1305
1306static const struct svc_xprt_ops svc_tcp_ops = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1307	.xpo_create = svc_tcp_create,
1308	.xpo_recvfrom = svc_tcp_recvfrom,
1309	.xpo_sendto = svc_tcp_sendto,
1310	.xpo_result_payload = svc_sock_result_payload,
1311	.xpo_release_ctxt = svc_tcp_release_ctxt,
1312	.xpo_detach = svc_tcp_sock_detach,
1313	.xpo_free = svc_sock_free,
 
1314	.xpo_has_wspace = svc_tcp_has_wspace,
1315	.xpo_accept = svc_tcp_accept,
1316	.xpo_kill_temp_xprt = svc_tcp_kill_temp_xprt,
1317	.xpo_handshake = svc_tcp_handshake,
1318};
1319
1320static struct svc_xprt_class svc_tcp_class = {
1321	.xcl_name = "tcp",
1322	.xcl_owner = THIS_MODULE,
1323	.xcl_ops = &svc_tcp_ops,
1324	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1325	.xcl_ident = XPRT_TRANSPORT_TCP,
1326};
1327
1328void svc_init_xprt_sock(void)
1329{
1330	svc_reg_xprt_class(&svc_tcp_class);
1331	svc_reg_xprt_class(&svc_udp_class);
 
1332}
1333
1334void svc_cleanup_xprt_sock(void)
1335{
1336	svc_unreg_xprt_class(&svc_tcp_class);
1337	svc_unreg_xprt_class(&svc_udp_class);
 
1338}
1339
1340static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1341{
1342	struct sock	*sk = svsk->sk_sk;
1343
1344	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1345		      &svsk->sk_xprt, serv);
1346	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
1347	set_bit(XPT_CONG_CTRL, &svsk->sk_xprt.xpt_flags);
1348	if (sk->sk_state == TCP_LISTEN) {
1349		strcpy(svsk->sk_xprt.xpt_remotebuf, "listener");
1350		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1351		sk->sk_data_ready = svc_tcp_listen_data_ready;
1352		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1353	} else {
 
1354		sk->sk_state_change = svc_tcp_state_change;
1355		sk->sk_data_ready = svc_data_ready;
1356		sk->sk_write_space = svc_write_space;
1357
1358		svsk->sk_marker = xdr_zero;
1359		svsk->sk_tcplen = 0;
1360		svsk->sk_datalen = 0;
1361		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1362
1363		tcp_sock_set_nodelay(sk);
1364
1365		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1366		switch (sk->sk_state) {
1367		case TCP_SYN_RECV:
1368		case TCP_ESTABLISHED:
1369			break;
1370		default:
1371			svc_xprt_deferred_close(&svsk->sk_xprt);
1372		}
1373	}
1374}
1375
1376void svc_sock_update_bufs(struct svc_serv *serv)
1377{
1378	/*
1379	 * The number of server threads has changed. Update
1380	 * rcvbuf and sndbuf accordingly on all sockets
1381	 */
1382	struct svc_sock *svsk;
1383
1384	spin_lock_bh(&serv->sv_lock);
1385	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1386		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1387	spin_unlock_bh(&serv->sv_lock);
1388}
1389EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1390
1391/*
1392 * Initialize socket for RPC use and create svc_sock struct
 
1393 */
1394static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1395						struct socket *sock,
1396						int flags)
1397{
1398	struct svc_sock	*svsk;
1399	struct sock	*inet;
1400	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
 
1401
 
1402	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1403	if (!svsk)
1404		return ERR_PTR(-ENOMEM);
1405
1406	inet = sock->sk;
1407
1408	if (pmap_register) {
1409		int err;
1410
1411		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1412				     inet->sk_protocol,
1413				     ntohs(inet_sk(inet)->inet_sport));
1414		if (err < 0) {
1415			kfree(svsk);
1416			return ERR_PTR(err);
1417		}
1418	}
1419
 
1420	svsk->sk_sock = sock;
1421	svsk->sk_sk = inet;
1422	svsk->sk_ostate = inet->sk_state_change;
1423	svsk->sk_odata = inet->sk_data_ready;
1424	svsk->sk_owspace = inet->sk_write_space;
1425	/*
1426	 * This barrier is necessary in order to prevent race condition
1427	 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
1428	 * when calling callbacks above.
1429	 */
1430	wmb();
1431	inet->sk_user_data = svsk;
1432
1433	/* Initialize the socket */
1434	if (sock->type == SOCK_DGRAM)
1435		svc_udp_init(svsk, serv);
1436	else
 
 
 
 
 
1437		svc_tcp_init(svsk, serv);
 
 
 
 
1438
1439	trace_svcsock_new(svsk, sock);
1440	return svsk;
1441}
1442
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1443/**
1444 * svc_addsock - add a listener socket to an RPC service
1445 * @serv: pointer to RPC service to which to add a new listener
1446 * @net: caller's network namespace
1447 * @fd: file descriptor of the new listener
1448 * @name_return: pointer to buffer to fill in with name of listener
1449 * @len: size of the buffer
1450 * @cred: credential
1451 *
1452 * Fills in socket name and returns positive length of name if successful.
1453 * Name is terminated with '\n'.  On error, returns a negative errno
1454 * value.
1455 */
1456int svc_addsock(struct svc_serv *serv, struct net *net, const int fd,
1457		char *name_return, const size_t len, const struct cred *cred)
1458{
1459	int err = 0;
1460	struct socket *so = sockfd_lookup(fd, &err);
1461	struct svc_sock *svsk = NULL;
1462	struct sockaddr_storage addr;
1463	struct sockaddr *sin = (struct sockaddr *)&addr;
1464	int salen;
1465
1466	if (!so)
1467		return err;
1468	err = -EINVAL;
1469	if (sock_net(so->sk) != net)
1470		goto out;
1471	err = -EAFNOSUPPORT;
1472	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1473		goto out;
1474	err =  -EPROTONOSUPPORT;
1475	if (so->sk->sk_protocol != IPPROTO_TCP &&
1476	    so->sk->sk_protocol != IPPROTO_UDP)
1477		goto out;
1478	err = -EISCONN;
1479	if (so->state > SS_UNCONNECTED)
1480		goto out;
1481	err = -ENOENT;
1482	if (!try_module_get(THIS_MODULE))
1483		goto out;
1484	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1485	if (IS_ERR(svsk)) {
1486		module_put(THIS_MODULE);
1487		err = PTR_ERR(svsk);
1488		goto out;
1489	}
1490	salen = kernel_getsockname(svsk->sk_sock, sin);
1491	if (salen >= 0)
1492		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
1493	svsk->sk_xprt.xpt_cred = get_cred(cred);
1494	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1495	return svc_one_sock_name(svsk, name_return, len);
1496out:
1497	sockfd_put(so);
1498	return err;
1499}
1500EXPORT_SYMBOL_GPL(svc_addsock);
1501
1502/*
1503 * Create socket for RPC service.
1504 */
1505static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1506					  int protocol,
1507					  struct net *net,
1508					  struct sockaddr *sin, int len,
1509					  int flags)
1510{
1511	struct svc_sock	*svsk;
1512	struct socket	*sock;
1513	int		error;
1514	int		type;
1515	struct sockaddr_storage addr;
1516	struct sockaddr *newsin = (struct sockaddr *)&addr;
1517	int		newlen;
1518	int		family;
 
 
 
 
 
 
1519
1520	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1521		printk(KERN_WARNING "svc: only UDP and TCP "
1522				"sockets supported\n");
1523		return ERR_PTR(-EINVAL);
1524	}
1525
1526	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1527	switch (sin->sa_family) {
1528	case AF_INET6:
1529		family = PF_INET6;
1530		break;
1531	case AF_INET:
1532		family = PF_INET;
1533		break;
1534	default:
1535		return ERR_PTR(-EINVAL);
1536	}
1537
1538	error = __sock_create(net, family, type, protocol, &sock, 1);
1539	if (error < 0)
1540		return ERR_PTR(error);
1541
1542	svc_reclassify_socket(sock);
1543
1544	/*
1545	 * If this is an PF_INET6 listener, we want to avoid
1546	 * getting requests from IPv4 remotes.  Those should
1547	 * be shunted to a PF_INET listener via rpcbind.
1548	 */
 
1549	if (family == PF_INET6)
1550		ip6_sock_set_v6only(sock->sk);
 
 
1551	if (type == SOCK_STREAM)
1552		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1553	error = kernel_bind(sock, sin, len);
1554	if (error < 0)
1555		goto bummer;
1556
1557	error = kernel_getsockname(sock, newsin);
 
1558	if (error < 0)
1559		goto bummer;
1560	newlen = error;
1561
1562	if (protocol == IPPROTO_TCP) {
1563		if ((error = kernel_listen(sock, 64)) < 0)
1564			goto bummer;
1565	}
1566
1567	svsk = svc_setup_socket(serv, sock, flags);
1568	if (IS_ERR(svsk)) {
1569		error = PTR_ERR(svsk);
1570		goto bummer;
1571	}
1572	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1573	return (struct svc_xprt *)svsk;
1574bummer:
 
1575	sock_release(sock);
1576	return ERR_PTR(error);
1577}
1578
1579/*
1580 * Detach the svc_sock from the socket so that no
1581 * more callbacks occur.
1582 */
1583static void svc_sock_detach(struct svc_xprt *xprt)
1584{
1585	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1586	struct sock *sk = svsk->sk_sk;
 
 
 
1587
1588	/* put back the old socket callbacks */
1589	lock_sock(sk);
1590	sk->sk_state_change = svsk->sk_ostate;
1591	sk->sk_data_ready = svsk->sk_odata;
1592	sk->sk_write_space = svsk->sk_owspace;
1593	sk->sk_user_data = NULL;
1594	release_sock(sk);
 
 
1595}
1596
1597/*
1598 * Disconnect the socket, and reset the callbacks
1599 */
1600static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1601{
1602	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1603
1604	tls_handshake_close(svsk->sk_sock);
1605
1606	svc_sock_detach(xprt);
1607
1608	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1609		svc_tcp_clear_pages(svsk);
1610		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1611	}
1612}
1613
1614/*
1615 * Free the svc_sock's socket resources and the svc_sock itself.
1616 */
1617static void svc_sock_free(struct svc_xprt *xprt)
1618{
1619	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1620	struct page_frag_cache *pfc = &svsk->sk_frag_cache;
1621	struct socket *sock = svsk->sk_sock;
1622
1623	trace_svcsock_free(svsk, sock);
1624
1625	tls_handshake_cancel(sock->sk);
1626	if (sock->file)
1627		sockfd_put(sock);
1628	else
1629		sock_release(sock);
1630	if (pfc->va)
1631		__page_frag_cache_drain(virt_to_head_page(pfc->va),
1632					pfc->pagecnt_bias);
1633	kfree(svsk);
1634}
v3.15
 
   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;
 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) < 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 * Check input queue length
 316 */
 317static int svc_recv_available(struct svc_sock *svsk)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 318{
 319	struct socket	*sock = svsk->sk_sock;
 320	int		avail, err;
 
 
 
 
 321
 322	err = kernel_sock_ioctl(sock, TIOCINQ, (unsigned long) &avail);
 
 
 
 
 
 
 323
 324	return (err >= 0)? avail : err;
 
 
 
 
 
 
 
 
 
 
 325}
 
 
 
 
 
 
 326
 327/*
 328 * Generic recvfrom routine.
 
 
 329 */
 330static int svc_recvfrom(struct svc_rqst *rqstp, struct kvec *iov, int nr,
 331			int buflen)
 332{
 333	struct svc_sock *svsk =
 334		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 335	struct msghdr msg = {
 336		.msg_flags	= MSG_DONTWAIT,
 337	};
 338	int len;
 
 339
 340	rqstp->rq_xprt_hlen = 0;
 341
 342	len = kernel_recvmsg(svsk->sk_sock, &msg, iov, nr, buflen,
 343				msg.msg_flags);
 
 
 344
 345	dprintk("svc: socket %p recvfrom(%p, %Zu) = %d\n",
 346		svsk, iov[0].iov_base, iov[0].iov_len, len);
 347	return len;
 348}
 
 
 
 
 349
 350static int svc_partial_recvfrom(struct svc_rqst *rqstp,
 351				struct kvec *iov, int nr,
 352				int buflen, unsigned int base)
 353{
 354	size_t save_iovlen;
 355	void *save_iovbase;
 356	unsigned int i;
 357	int ret;
 358
 359	if (base == 0)
 360		return svc_recvfrom(rqstp, iov, nr, buflen);
 361
 362	for (i = 0; i < nr; i++) {
 363		if (iov[i].iov_len > base)
 364			break;
 365		base -= iov[i].iov_len;
 366	}
 367	save_iovlen = iov[i].iov_len;
 368	save_iovbase = iov[i].iov_base;
 369	iov[i].iov_len -= base;
 370	iov[i].iov_base += base;
 371	ret = svc_recvfrom(rqstp, &iov[i], nr - i, buflen);
 372	iov[i].iov_len = save_iovlen;
 373	iov[i].iov_base = save_iovbase;
 374	return ret;
 375}
 376
 377/*
 378 * Set socket snd and rcv buffer lengths
 379 */
 380static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
 381				unsigned int rcv)
 382{
 383#if 0
 384	mm_segment_t	oldfs;
 385	oldfs = get_fs(); set_fs(KERNEL_DS);
 386	sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
 387			(char*)&snd, sizeof(snd));
 388	sock_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
 389			(char*)&rcv, sizeof(rcv));
 390#else
 391	/* sock_setsockopt limits use to sysctl_?mem_max,
 392	 * which isn't acceptable.  Until that is made conditional
 393	 * on not having CAP_SYS_RESOURCE or similar, we go direct...
 394	 * DaveM said I could!
 395	 */
 396	lock_sock(sock->sk);
 397	sock->sk->sk_sndbuf = snd * 2;
 398	sock->sk->sk_rcvbuf = rcv * 2;
 399	sock->sk->sk_write_space(sock->sk);
 400	release_sock(sock->sk);
 401#endif
 402}
 
 
 
 
 
 
 
 
 
 403/*
 404 * INET callback when data has been received on the socket.
 405 */
 406static void svc_udp_data_ready(struct sock *sk)
 407{
 408	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
 409	wait_queue_head_t *wq = sk_sleep(sk);
 
 410
 411	if (svsk) {
 412		dprintk("svc: socket %p(inet %p), busy=%d\n",
 413			svsk, sk,
 414			test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
 415		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 416		svc_xprt_enqueue(&svsk->sk_xprt);
 
 
 
 417	}
 418	if (wq && waitqueue_active(wq))
 419		wake_up_interruptible(wq);
 420}
 421
 422/*
 423 * INET callback when space is newly available on the socket.
 424 */
 425static void svc_write_space(struct sock *sk)
 426{
 427	struct svc_sock	*svsk = (struct svc_sock *)(sk->sk_user_data);
 428	wait_queue_head_t *wq = sk_sleep(sk);
 429
 430	if (svsk) {
 431		dprintk("svc: socket %p(inet %p), write_space busy=%d\n",
 432			svsk, sk, test_bit(XPT_BUSY, &svsk->sk_xprt.xpt_flags));
 
 
 433		svc_xprt_enqueue(&svsk->sk_xprt);
 434	}
 
 435
 436	if (wq && waitqueue_active(wq)) {
 437		dprintk("RPC svc_write_space: someone sleeping on %p\n",
 438		       svsk);
 439		wake_up_interruptible(wq);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 440	}
 
 
 441}
 442
 443static void svc_tcp_write_space(struct sock *sk)
 
 
 
 
 
 444{
 445	struct socket *sock = sk->sk_socket;
 
 
 
 
 
 
 
 
 
 446
 447	if (sk_stream_is_writeable(sk) && sock)
 448		clear_bit(SOCK_NOSPACE, &sock->flags);
 449	svc_write_space(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 450}
 451
 452/*
 453 * See net/ipv6/ip_sockglue.c : ip_cmsg_recv_pktinfo
 454 */
 455static int svc_udp_get_dest_address4(struct svc_rqst *rqstp,
 456				     struct cmsghdr *cmh)
 457{
 458	struct in_pktinfo *pki = CMSG_DATA(cmh);
 459	struct sockaddr_in *daddr = svc_daddr_in(rqstp);
 460
 461	if (cmh->cmsg_type != IP_PKTINFO)
 462		return 0;
 463
 464	daddr->sin_family = AF_INET;
 465	daddr->sin_addr.s_addr = pki->ipi_spec_dst.s_addr;
 466	return 1;
 467}
 468
 469/*
 470 * See net/ipv6/datagram.c : ip6_datagram_recv_ctl
 471 */
 472static int svc_udp_get_dest_address6(struct svc_rqst *rqstp,
 473				     struct cmsghdr *cmh)
 474{
 475	struct in6_pktinfo *pki = CMSG_DATA(cmh);
 476	struct sockaddr_in6 *daddr = svc_daddr_in6(rqstp);
 477
 478	if (cmh->cmsg_type != IPV6_PKTINFO)
 479		return 0;
 480
 481	daddr->sin6_family = AF_INET6;
 482	daddr->sin6_addr = pki->ipi6_addr;
 483	daddr->sin6_scope_id = pki->ipi6_ifindex;
 484	return 1;
 485}
 486
 487/*
 488 * Copy the UDP datagram's destination address to the rqstp structure.
 489 * The 'destination' address in this case is the address to which the
 490 * peer sent the datagram, i.e. our local address. For multihomed
 491 * hosts, this can change from msg to msg. Note that only the IP
 492 * address changes, the port number should remain the same.
 493 */
 494static int svc_udp_get_dest_address(struct svc_rqst *rqstp,
 495				    struct cmsghdr *cmh)
 496{
 497	switch (cmh->cmsg_level) {
 498	case SOL_IP:
 499		return svc_udp_get_dest_address4(rqstp, cmh);
 500	case SOL_IPV6:
 501		return svc_udp_get_dest_address6(rqstp, cmh);
 502	}
 503
 504	return 0;
 505}
 506
 507/*
 508 * Receive a datagram from a UDP socket.
 
 
 
 
 
 
 
 509 */
 510static int svc_udp_recvfrom(struct svc_rqst *rqstp)
 511{
 512	struct svc_sock	*svsk =
 513		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
 514	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
 515	struct sk_buff	*skb;
 516	union {
 517		struct cmsghdr	hdr;
 518		long		all[SVC_PKTINFO_SPACE / sizeof(long)];
 519	} buffer;
 520	struct cmsghdr *cmh = &buffer.hdr;
 521	struct msghdr msg = {
 522		.msg_name = svc_addr(rqstp),
 523		.msg_control = cmh,
 524		.msg_controllen = sizeof(buffer),
 525		.msg_flags = MSG_DONTWAIT,
 526	};
 527	size_t len;
 528	int err;
 529
 530	if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
 531	    /* udp sockets need large rcvbuf as all pending
 532	     * requests are still in that buffer.  sndbuf must
 533	     * also be large enough that there is enough space
 534	     * for one reply per thread.  We count all threads
 535	     * rather than threads in a particular pool, which
 536	     * provides an upper bound on the number of threads
 537	     * which will access the socket.
 538	     */
 539	    svc_sock_setbufsize(svsk->sk_sock,
 540				(serv->sv_nrthreads+3) * serv->sv_max_mesg,
 541				(serv->sv_nrthreads+3) * serv->sv_max_mesg);
 542
 543	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 544	skb = NULL;
 545	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,
 546			     0, 0, MSG_PEEK | MSG_DONTWAIT);
 547	if (err >= 0)
 548		skb = skb_recv_datagram(svsk->sk_sk, 0, 1, &err);
 
 
 
 549
 550	if (skb == NULL) {
 551		if (err != -EAGAIN) {
 552			/* possibly an icmp error */
 553			dprintk("svc: recvfrom returned error %d\n", -err);
 554			set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 555		}
 556		return 0;
 557	}
 558	len = svc_addr_len(svc_addr(rqstp));
 559	rqstp->rq_addrlen = len;
 560	if (skb->tstamp.tv64 == 0) {
 561		skb->tstamp = ktime_get_real();
 562		/* Don't enable netstamp, sunrpc doesn't
 563		   need that much accuracy */
 564	}
 565	svsk->sk_sk->sk_stamp = skb->tstamp;
 566	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* there may be more data... */
 567
 568	len  = skb->len - sizeof(struct udphdr);
 569	rqstp->rq_arg.len = len;
 
 570
 571	rqstp->rq_prot = IPPROTO_UDP;
 572
 573	if (!svc_udp_get_dest_address(rqstp, cmh)) {
 574		net_warn_ratelimited("svc: received unknown control message %d/%d; dropping RPC reply datagram\n",
 575				     cmh->cmsg_level, cmh->cmsg_type);
 576		goto out_free;
 577	}
 578	rqstp->rq_daddrlen = svc_addr_len(svc_daddr(rqstp));
 579
 580	if (skb_is_nonlinear(skb)) {
 581		/* we have to copy */
 582		local_bh_disable();
 583		if (csum_partial_copy_to_xdr(&rqstp->rq_arg, skb)) {
 584			local_bh_enable();
 585			/* checksum error */
 586			goto out_free;
 587		}
 588		local_bh_enable();
 589		skb_free_datagram_locked(svsk->sk_sk, skb);
 590	} else {
 591		/* we can use it in-place */
 592		rqstp->rq_arg.head[0].iov_base = skb->data +
 593			sizeof(struct udphdr);
 594		rqstp->rq_arg.head[0].iov_len = len;
 595		if (skb_checksum_complete(skb))
 596			goto out_free;
 597		rqstp->rq_xprt_ctxt = skb;
 598	}
 599
 600	rqstp->rq_arg.page_base = 0;
 601	if (len <= rqstp->rq_arg.head[0].iov_len) {
 602		rqstp->rq_arg.head[0].iov_len = len;
 603		rqstp->rq_arg.page_len = 0;
 604		rqstp->rq_respages = rqstp->rq_pages+1;
 605	} else {
 606		rqstp->rq_arg.page_len = len - rqstp->rq_arg.head[0].iov_len;
 607		rqstp->rq_respages = rqstp->rq_pages + 1 +
 608			DIV_ROUND_UP(rqstp->rq_arg.page_len, PAGE_SIZE);
 609	}
 610	rqstp->rq_next_page = rqstp->rq_respages+1;
 611
 612	if (serv->sv_stats)
 613		serv->sv_stats->netudpcnt++;
 614
 
 
 615	return len;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 616out_free:
 617	trace_kfree_skb(skb, svc_udp_recvfrom);
 618	skb_free_datagram_locked(svsk->sk_sk, skb);
 
 619	return 0;
 620}
 621
 622static int
 623svc_udp_sendto(struct svc_rqst *rqstp)
 
 
 
 
 
 
 
 
 624{
 625	int		error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 626
 627	error = svc_sendto(rqstp, &rqstp->rq_res);
 628	if (error == -ECONNREFUSED)
 
 
 629		/* ICMP error on earlier request. */
 630		error = svc_sendto(rqstp, &rqstp->rq_res);
 
 
 
 631
 632	return error;
 633}
 
 
 634
 635static void svc_udp_prep_reply_hdr(struct svc_rqst *rqstp)
 636{
 
 637}
 638
 639static int svc_udp_has_wspace(struct svc_xprt *xprt)
 640{
 641	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 642	struct svc_serv	*serv = xprt->xpt_server;
 643	unsigned long required;
 644
 645	/*
 646	 * Set the SOCK_NOSPACE flag before checking the available
 647	 * sock space.
 648	 */
 649	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
 650	required = atomic_read(&svsk->sk_xprt.xpt_reserved) + serv->sv_max_mesg;
 651	if (required*2 > sock_wspace(svsk->sk_sk))
 652		return 0;
 653	clear_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
 654	return 1;
 655}
 656
 657static struct svc_xprt *svc_udp_accept(struct svc_xprt *xprt)
 658{
 659	BUG();
 660	return NULL;
 661}
 662
 
 
 
 
 663static struct svc_xprt *svc_udp_create(struct svc_serv *serv,
 664				       struct net *net,
 665				       struct sockaddr *sa, int salen,
 666				       int flags)
 667{
 668	return svc_create_socket(serv, IPPROTO_UDP, net, sa, salen, flags);
 669}
 670
 671static struct svc_xprt_ops svc_udp_ops = {
 672	.xpo_create = svc_udp_create,
 673	.xpo_recvfrom = svc_udp_recvfrom,
 674	.xpo_sendto = svc_udp_sendto,
 675	.xpo_release_rqst = svc_release_skb,
 
 676	.xpo_detach = svc_sock_detach,
 677	.xpo_free = svc_sock_free,
 678	.xpo_prep_reply_hdr = svc_udp_prep_reply_hdr,
 679	.xpo_has_wspace = svc_udp_has_wspace,
 680	.xpo_accept = svc_udp_accept,
 
 681};
 682
 683static struct svc_xprt_class svc_udp_class = {
 684	.xcl_name = "udp",
 685	.xcl_owner = THIS_MODULE,
 686	.xcl_ops = &svc_udp_ops,
 687	.xcl_max_payload = RPCSVC_MAXPAYLOAD_UDP,
 
 688};
 689
 690static void svc_udp_init(struct svc_sock *svsk, struct svc_serv *serv)
 691{
 692	int err, level, optname, one = 1;
 693
 694	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_udp_class,
 695		      &svsk->sk_xprt, serv);
 696	clear_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
 697	svsk->sk_sk->sk_data_ready = svc_udp_data_ready;
 698	svsk->sk_sk->sk_write_space = svc_write_space;
 699
 700	/* initialise setting must have enough space to
 701	 * receive and respond to one request.
 702	 * svc_udp_recvfrom will re-adjust if necessary
 703	 */
 704	svc_sock_setbufsize(svsk->sk_sock,
 705			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
 706			    3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
 707
 708	/* data might have come in before data_ready set up */
 709	set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 710	set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
 711
 712	/* make sure we get destination address info */
 713	switch (svsk->sk_sk->sk_family) {
 714	case AF_INET:
 715		level = SOL_IP;
 716		optname = IP_PKTINFO;
 717		break;
 718	case AF_INET6:
 719		level = SOL_IPV6;
 720		optname = IPV6_RECVPKTINFO;
 721		break;
 722	default:
 723		BUG();
 724	}
 725	err = kernel_setsockopt(svsk->sk_sock, level, optname,
 726					(char *)&one, sizeof(one));
 727	dprintk("svc: kernel_setsockopt returned %d\n", err);
 728}
 729
 730/*
 731 * A data_ready event on a listening socket means there's a connection
 732 * pending. Do not use state_change as a substitute for it.
 733 */
 734static void svc_tcp_listen_data_ready(struct sock *sk)
 735{
 736	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
 737	wait_queue_head_t *wq;
 738
 739	dprintk("svc: socket %p TCP (listen) state change %d\n",
 740		sk, sk->sk_state);
 741
 742	/*
 743	 * This callback may called twice when a new connection
 744	 * is established as a child socket inherits everything
 745	 * from a parent LISTEN socket.
 746	 * 1) data_ready method of the parent socket will be called
 747	 *    when one of child sockets become ESTABLISHED.
 748	 * 2) data_ready method of the child socket may be called
 749	 *    when it receives data before the socket is accepted.
 750	 * In case of 2, we should ignore it silently.
 
 751	 */
 752	if (sk->sk_state == TCP_LISTEN) {
 753		if (svsk) {
 754			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 755			svc_xprt_enqueue(&svsk->sk_xprt);
 756		} else
 757			printk("svc: socket %p: no user data\n", sk);
 
 
 
 758	}
 759
 760	wq = sk_sleep(sk);
 761	if (wq && waitqueue_active(wq))
 762		wake_up_interruptible_all(wq);
 763}
 764
 765/*
 766 * A state change on a connected socket means it's dying or dead.
 767 */
 768static void svc_tcp_state_change(struct sock *sk)
 769{
 770	struct svc_sock	*svsk = (struct svc_sock *)sk->sk_user_data;
 771	wait_queue_head_t *wq = sk_sleep(sk);
 772
 773	dprintk("svc: socket %p TCP (connected) state change %d (svsk %p)\n",
 774		sk, sk->sk_state, sk->sk_user_data);
 775
 776	if (!svsk)
 777		printk("svc: socket %p: no user data\n", sk);
 778	else {
 779		set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
 780		svc_xprt_enqueue(&svsk->sk_xprt);
 781	}
 782	if (wq && waitqueue_active(wq))
 783		wake_up_interruptible_all(wq);
 784}
 785
 786static void svc_tcp_data_ready(struct sock *sk)
 787{
 788	struct svc_sock *svsk = (struct svc_sock *)sk->sk_user_data;
 789	wait_queue_head_t *wq = sk_sleep(sk);
 790
 791	dprintk("svc: socket %p TCP data ready (svsk %p)\n",
 792		sk, sk->sk_user_data);
 793	if (svsk) {
 794		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 795		svc_xprt_enqueue(&svsk->sk_xprt);
 
 
 
 
 796	}
 797	if (wq && waitqueue_active(wq))
 798		wake_up_interruptible(wq);
 799}
 800
 801/*
 802 * Accept a TCP connection
 803 */
 804static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
 805{
 806	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 807	struct sockaddr_storage addr;
 808	struct sockaddr	*sin = (struct sockaddr *) &addr;
 809	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
 810	struct socket	*sock = svsk->sk_sock;
 811	struct socket	*newsock;
 812	struct svc_sock	*newsvsk;
 813	int		err, slen;
 814	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
 815
 816	dprintk("svc: tcp_accept %p sock %p\n", svsk, sock);
 817	if (!sock)
 818		return NULL;
 819
 820	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 821	err = kernel_accept(sock, &newsock, O_NONBLOCK);
 822	if (err < 0) {
 823		if (err == -ENOMEM)
 824			printk(KERN_WARNING "%s: no more sockets!\n",
 825			       serv->sv_name);
 826		else if (err != -EAGAIN)
 827			net_warn_ratelimited("%s: accept failed (err %d)!\n",
 828					     serv->sv_name, -err);
 829		return NULL;
 830	}
 
 
 
 831	set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 832
 833	err = kernel_getpeername(newsock, sin, &slen);
 834	if (err < 0) {
 835		net_warn_ratelimited("%s: peername failed (err %d)!\n",
 836				     serv->sv_name, -err);
 837		goto failed;		/* aborted connection or whatever */
 838	}
 
 839
 840	/* Ideally, we would want to reject connections from unauthorized
 841	 * hosts here, but when we get encryption, the IP of the host won't
 842	 * tell us anything.  For now just warn about unpriv connections.
 843	 */
 844	if (!svc_port_is_privileged(sin)) {
 845		dprintk(KERN_WARNING
 846			"%s: connect from unprivileged port: %s\n",
 847			serv->sv_name,
 848			__svc_print_addr(sin, buf, sizeof(buf)));
 849	}
 850	dprintk("%s: connect from %s\n", serv->sv_name,
 851		__svc_print_addr(sin, buf, sizeof(buf)));
 852
 853	/* make sure that a write doesn't block forever when
 854	 * low on memory
 855	 */
 856	newsock->sk->sk_sndtimeo = HZ*30;
 857
 858	newsvsk = svc_setup_socket(serv, newsock,
 859				 (SVC_SOCK_ANONYMOUS | SVC_SOCK_TEMPORARY));
 860	if (IS_ERR(newsvsk))
 861		goto failed;
 862	svc_xprt_set_remote(&newsvsk->sk_xprt, sin, slen);
 863	err = kernel_getsockname(newsock, sin, &slen);
 864	if (unlikely(err < 0)) {
 865		dprintk("svc_tcp_accept: kernel_getsockname error %d\n", -err);
 866		slen = offsetof(struct sockaddr, sa_data);
 867	}
 868	svc_xprt_set_local(&newsvsk->sk_xprt, sin, slen);
 869
 
 
 
 
 870	if (serv->sv_stats)
 871		serv->sv_stats->nettcpconn++;
 872
 873	return &newsvsk->sk_xprt;
 874
 875failed:
 876	sock_release(newsock);
 877	return NULL;
 878}
 879
 880static unsigned int svc_tcp_restore_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
 
 881{
 882	unsigned int i, len, npages;
 
 883
 884	if (svsk->sk_datalen == 0)
 885		return 0;
 886	len = svsk->sk_datalen;
 887	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 888	for (i = 0; i < npages; i++) {
 889		if (rqstp->rq_pages[i] != NULL)
 890			put_page(rqstp->rq_pages[i]);
 891		BUG_ON(svsk->sk_pages[i] == NULL);
 892		rqstp->rq_pages[i] = svsk->sk_pages[i];
 893		svsk->sk_pages[i] = NULL;
 894	}
 895	rqstp->rq_arg.head[0].iov_base = page_address(rqstp->rq_pages[0]);
 896	return len;
 897}
 898
 899static void svc_tcp_save_pages(struct svc_sock *svsk, struct svc_rqst *rqstp)
 900{
 901	unsigned int i, len, npages;
 902
 903	if (svsk->sk_datalen == 0)
 904		return;
 905	len = svsk->sk_datalen;
 906	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 907	for (i = 0; i < npages; i++) {
 908		svsk->sk_pages[i] = rqstp->rq_pages[i];
 909		rqstp->rq_pages[i] = NULL;
 910	}
 911}
 912
 913static void svc_tcp_clear_pages(struct svc_sock *svsk)
 914{
 915	unsigned int i, len, npages;
 916
 917	if (svsk->sk_datalen == 0)
 918		goto out;
 919	len = svsk->sk_datalen;
 920	npages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 921	for (i = 0; i < npages; i++) {
 922		if (svsk->sk_pages[i] == NULL) {
 923			WARN_ON_ONCE(1);
 924			continue;
 925		}
 926		put_page(svsk->sk_pages[i]);
 927		svsk->sk_pages[i] = NULL;
 928	}
 929out:
 930	svsk->sk_tcplen = 0;
 931	svsk->sk_datalen = 0;
 932}
 933
 934/*
 935 * Receive fragment record header.
 936 * If we haven't gotten the record length yet, get the next four bytes.
 937 */
 938static int svc_tcp_recv_record(struct svc_sock *svsk, struct svc_rqst *rqstp)
 
 939{
 940	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
 941	unsigned int want;
 942	int len;
 943
 944	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
 945
 
 
 
 946	if (svsk->sk_tcplen < sizeof(rpc_fraghdr)) {
 
 947		struct kvec	iov;
 948
 949		want = sizeof(rpc_fraghdr) - svsk->sk_tcplen;
 950		iov.iov_base = ((char *) &svsk->sk_reclen) + svsk->sk_tcplen;
 951		iov.iov_len  = want;
 952		if ((len = svc_recvfrom(rqstp, &iov, 1, want)) < 0)
 953			goto error;
 
 
 954		svsk->sk_tcplen += len;
 955
 956		if (len < want) {
 957			dprintk("svc: short recvfrom while reading record "
 958				"length (%d of %d)\n", len, want);
 959			return -EAGAIN;
 960		}
 961
 962		dprintk("svc: TCP record, %d bytes\n", svc_sock_reclen(svsk));
 963		if (svc_sock_reclen(svsk) + svsk->sk_datalen >
 964							serv->sv_max_mesg) {
 965			net_notice_ratelimited("RPC: fragment too large: %d\n",
 966					svc_sock_reclen(svsk));
 967			goto err_delete;
 968		}
 969	}
 
 970
 971	return svc_sock_reclen(svsk);
 972error:
 973	dprintk("RPC: TCP recv_record got %d\n", len);
 974	return len;
 975err_delete:
 976	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
 977	return -EAGAIN;
 978}
 979
 980static int receive_cb_reply(struct svc_sock *svsk, struct svc_rqst *rqstp)
 981{
 982	struct rpc_xprt *bc_xprt = svsk->sk_xprt.xpt_bc_xprt;
 983	struct rpc_rqst *req = NULL;
 984	struct kvec *src, *dst;
 985	__be32 *p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
 986	__be32 xid;
 987	__be32 calldir;
 988
 989	xid = *p++;
 990	calldir = *p;
 991
 992	if (bc_xprt)
 993		req = xprt_lookup_rqst(bc_xprt, xid);
 994
 995	if (!req) {
 996		printk(KERN_NOTICE
 997			"%s: Got unrecognized reply: "
 998			"calldir 0x%x xpt_bc_xprt %p xid %08x\n",
 999			__func__, ntohl(calldir),
1000			bc_xprt, xid);
1001		return -EAGAIN;
1002	}
 
 
 
1003
1004	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
1005	/*
1006	 * XXX!: cheating for now!  Only copying HEAD.
1007	 * But we know this is good enough for now (in fact, for any
1008	 * callback reply in the forseeable future).
1009	 */
1010	dst = &req->rq_private_buf.head[0];
1011	src = &rqstp->rq_arg.head[0];
1012	if (dst->iov_len < src->iov_len)
1013		return -EAGAIN; /* whatever; just giving up. */
1014	memcpy(dst->iov_base, src->iov_base, src->iov_len);
1015	xprt_complete_rqst(req->rq_task, rqstp->rq_arg.len);
1016	rqstp->rq_arg.len = 0;
 
1017	return 0;
1018}
1019
1020static int copy_pages_to_kvecs(struct kvec *vec, struct page **pages, int len)
1021{
1022	int i = 0;
1023	int t = 0;
1024
1025	while (t < len) {
1026		vec[i].iov_base = page_address(pages[i]);
1027		vec[i].iov_len = PAGE_SIZE;
1028		i++;
1029		t += PAGE_SIZE;
1030	}
1031	return i;
1032}
1033
1034static void svc_tcp_fragment_received(struct svc_sock *svsk)
1035{
1036	/* If we have more data, signal svc_xprt_enqueue() to try again */
1037	if (svc_recv_available(svsk) > sizeof(rpc_fraghdr))
1038		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1039	dprintk("svc: TCP %s record (%d bytes)\n",
1040		svc_sock_final_rec(svsk) ? "final" : "nonfinal",
1041		svc_sock_reclen(svsk));
1042	svsk->sk_tcplen = 0;
1043	svsk->sk_reclen = 0;
 
 
 
1044}
1045
1046/*
1047 * Receive data from a TCP socket.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1048 */
1049static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
1050{
1051	struct svc_sock	*svsk =
1052		container_of(rqstp->rq_xprt, struct svc_sock, sk_xprt);
1053	struct svc_serv	*serv = svsk->sk_xprt.xpt_server;
1054	int		len;
1055	struct kvec *vec;
1056	unsigned int want, base;
1057	__be32 *p;
1058	__be32 calldir;
1059	int pnum;
1060
1061	dprintk("svc: tcp_recv %p data %d conn %d close %d\n",
1062		svsk, test_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags),
1063		test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
1064		test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
1065
1066	len = svc_tcp_recv_record(svsk, rqstp);
1067	if (len < 0)
1068		goto error;
1069
1070	base = svc_tcp_restore_pages(svsk, rqstp);
1071	want = svc_sock_reclen(svsk) - (svsk->sk_tcplen - sizeof(rpc_fraghdr));
1072
1073	vec = rqstp->rq_vec;
1074
1075	pnum = copy_pages_to_kvecs(&vec[0], &rqstp->rq_pages[0],
1076						svsk->sk_datalen + want);
1077
1078	rqstp->rq_respages = &rqstp->rq_pages[pnum];
1079	rqstp->rq_next_page = rqstp->rq_respages + 1;
1080
1081	/* Now receive data */
1082	len = svc_partial_recvfrom(rqstp, vec, pnum, want, base);
1083	if (len >= 0) {
 
1084		svsk->sk_tcplen += len;
1085		svsk->sk_datalen += len;
1086	}
1087	if (len != want || !svc_sock_final_rec(svsk)) {
1088		svc_tcp_save_pages(svsk, rqstp);
1089		if (len < 0 && len != -EAGAIN)
1090			goto err_delete;
1091		if (len == want)
1092			svc_tcp_fragment_received(svsk);
1093		else
1094			dprintk("svc: incomplete TCP record (%d of %d)\n",
1095				(int)(svsk->sk_tcplen - sizeof(rpc_fraghdr)),
1096				svc_sock_reclen(svsk));
1097		goto err_noclose;
1098	}
1099
1100	if (svsk->sk_datalen < 8) {
1101		svsk->sk_datalen = 0;
1102		goto err_delete; /* client is nuts. */
1103	}
1104
1105	rqstp->rq_arg.len = svsk->sk_datalen;
1106	rqstp->rq_arg.page_base = 0;
1107	if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) {
1108		rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len;
1109		rqstp->rq_arg.page_len = 0;
1110	} else
1111		rqstp->rq_arg.page_len = rqstp->rq_arg.len - rqstp->rq_arg.head[0].iov_len;
1112
1113	rqstp->rq_xprt_ctxt   = NULL;
1114	rqstp->rq_prot	      = IPPROTO_TCP;
 
 
 
 
1115
1116	p = (__be32 *)rqstp->rq_arg.head[0].iov_base;
1117	calldir = p[1];
1118	if (calldir)
1119		len = receive_cb_reply(svsk, rqstp);
1120
1121	/* Reset TCP read info */
1122	svsk->sk_datalen = 0;
1123	svc_tcp_fragment_received(svsk);
1124
1125	if (len < 0)
1126		goto error;
1127
1128	svc_xprt_copy_addrs(rqstp, &svsk->sk_xprt);
1129	if (serv->sv_stats)
1130		serv->sv_stats->nettcpcnt++;
1131
 
 
1132	return rqstp->rq_arg.len;
1133
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1134error:
1135	if (len != -EAGAIN)
1136		goto err_delete;
1137	dprintk("RPC: TCP recvfrom got EAGAIN\n");
1138	return 0;
 
 
1139err_delete:
1140	printk(KERN_NOTICE "%s: recvfrom returned errno %d\n",
1141	       svsk->sk_xprt.xpt_server->sv_name, -len);
1142	set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
1143err_noclose:
 
1144	return 0;	/* record not complete */
1145}
1146
1147/*
1148 * Send out data on TCP socket.
 
 
 
 
 
 
 
 
 
 
 
1149 */
1150static int svc_tcp_sendto(struct svc_rqst *rqstp)
 
1151{
1152	struct xdr_buf	*xbufp = &rqstp->rq_res;
1153	int sent;
1154	__be32 reclen;
1155
1156	/* Set up the first element of the reply kvec.
1157	 * Any other kvecs that may be in use have been taken
1158	 * care of by the server implementation itself.
 
 
 
 
1159	 */
1160	reclen = htonl(0x80000000|((xbufp->len ) - 4));
1161	memcpy(xbufp->head[0].iov_base, &reclen, 4);
1162
1163	sent = svc_sendto(rqstp, &rqstp->rq_res);
1164	if (sent != xbufp->len) {
1165		printk(KERN_NOTICE
1166		       "rpc-srv/tcp: %s: %s %d when sending %d bytes "
1167		       "- shutting down socket\n",
1168		       rqstp->rq_xprt->xpt_server->sv_name,
1169		       (sent<0)?"got error":"sent only",
1170		       sent, xbufp->len);
1171		set_bit(XPT_CLOSE, &rqstp->rq_xprt->xpt_flags);
1172		svc_xprt_enqueue(rqstp->rq_xprt);
1173		sent = -EAGAIN;
1174	}
1175	return sent;
 
1176}
1177
1178/*
1179 * Setup response header. TCP has a 4B record length field.
 
 
 
 
 
 
1180 */
1181static void svc_tcp_prep_reply_hdr(struct svc_rqst *rqstp)
1182{
1183	struct kvec *resv = &rqstp->rq_res.head[0];
 
 
 
 
 
 
1184
1185	/* tcp needs a space for the record length... */
1186	svc_putnl(resv, 0);
1187}
1188
1189static int svc_tcp_has_wspace(struct svc_xprt *xprt)
1190{
1191	struct svc_sock *svsk =	container_of(xprt, struct svc_sock, sk_xprt);
1192	struct svc_serv *serv = svsk->sk_xprt.xpt_server;
1193	int required;
 
 
 
 
1194
1195	if (test_bit(XPT_LISTENER, &xprt->xpt_flags))
1196		return 1;
1197	required = atomic_read(&xprt->xpt_reserved) + serv->sv_max_mesg;
1198	if (sk_stream_wspace(svsk->sk_sk) >= required ||
1199	    (sk_stream_min_wspace(svsk->sk_sk) == 0 &&
1200	     atomic_read(&xprt->xpt_reserved) == 0))
1201		return 1;
1202	set_bit(SOCK_NOSPACE, &svsk->sk_sock->flags);
1203	return 0;
 
 
1204}
1205
1206static struct svc_xprt *svc_tcp_create(struct svc_serv *serv,
1207				       struct net *net,
1208				       struct sockaddr *sa, int salen,
1209				       int flags)
1210{
1211	return svc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1212}
1213
1214#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1215static struct svc_xprt *svc_bc_create_socket(struct svc_serv *, int,
1216					     struct net *, struct sockaddr *,
1217					     int, int);
1218static void svc_bc_sock_free(struct svc_xprt *xprt);
1219
1220static struct svc_xprt *svc_bc_tcp_create(struct svc_serv *serv,
1221				       struct net *net,
1222				       struct sockaddr *sa, int salen,
1223				       int flags)
1224{
1225	return svc_bc_create_socket(serv, IPPROTO_TCP, net, sa, salen, flags);
1226}
1227
1228static void svc_bc_tcp_sock_detach(struct svc_xprt *xprt)
1229{
1230}
1231
1232static struct svc_xprt_ops svc_tcp_bc_ops = {
1233	.xpo_create = svc_bc_tcp_create,
1234	.xpo_detach = svc_bc_tcp_sock_detach,
1235	.xpo_free = svc_bc_sock_free,
1236	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1237};
1238
1239static struct svc_xprt_class svc_tcp_bc_class = {
1240	.xcl_name = "tcp-bc",
1241	.xcl_owner = THIS_MODULE,
1242	.xcl_ops = &svc_tcp_bc_ops,
1243	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
1244};
1245
1246static void svc_init_bc_xprt_sock(void)
1247{
1248	svc_reg_xprt_class(&svc_tcp_bc_class);
1249}
1250
1251static void svc_cleanup_bc_xprt_sock(void)
1252{
1253	svc_unreg_xprt_class(&svc_tcp_bc_class);
1254}
1255#else /* CONFIG_SUNRPC_BACKCHANNEL */
1256static void svc_init_bc_xprt_sock(void)
1257{
1258}
1259
1260static void svc_cleanup_bc_xprt_sock(void)
1261{
1262}
1263#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1264
1265static struct svc_xprt_ops svc_tcp_ops = {
1266	.xpo_create = svc_tcp_create,
1267	.xpo_recvfrom = svc_tcp_recvfrom,
1268	.xpo_sendto = svc_tcp_sendto,
1269	.xpo_release_rqst = svc_release_skb,
 
1270	.xpo_detach = svc_tcp_sock_detach,
1271	.xpo_free = svc_sock_free,
1272	.xpo_prep_reply_hdr = svc_tcp_prep_reply_hdr,
1273	.xpo_has_wspace = svc_tcp_has_wspace,
1274	.xpo_accept = svc_tcp_accept,
 
 
1275};
1276
1277static struct svc_xprt_class svc_tcp_class = {
1278	.xcl_name = "tcp",
1279	.xcl_owner = THIS_MODULE,
1280	.xcl_ops = &svc_tcp_ops,
1281	.xcl_max_payload = RPCSVC_MAXPAYLOAD_TCP,
 
1282};
1283
1284void svc_init_xprt_sock(void)
1285{
1286	svc_reg_xprt_class(&svc_tcp_class);
1287	svc_reg_xprt_class(&svc_udp_class);
1288	svc_init_bc_xprt_sock();
1289}
1290
1291void svc_cleanup_xprt_sock(void)
1292{
1293	svc_unreg_xprt_class(&svc_tcp_class);
1294	svc_unreg_xprt_class(&svc_udp_class);
1295	svc_cleanup_bc_xprt_sock();
1296}
1297
1298static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1299{
1300	struct sock	*sk = svsk->sk_sk;
1301
1302	svc_xprt_init(sock_net(svsk->sk_sock->sk), &svc_tcp_class,
1303		      &svsk->sk_xprt, serv);
1304	set_bit(XPT_CACHE_AUTH, &svsk->sk_xprt.xpt_flags);
 
1305	if (sk->sk_state == TCP_LISTEN) {
1306		dprintk("setting up TCP socket for listening\n");
1307		set_bit(XPT_LISTENER, &svsk->sk_xprt.xpt_flags);
1308		sk->sk_data_ready = svc_tcp_listen_data_ready;
1309		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
1310	} else {
1311		dprintk("setting up TCP socket for reading\n");
1312		sk->sk_state_change = svc_tcp_state_change;
1313		sk->sk_data_ready = svc_tcp_data_ready;
1314		sk->sk_write_space = svc_tcp_write_space;
1315
1316		svsk->sk_reclen = 0;
1317		svsk->sk_tcplen = 0;
1318		svsk->sk_datalen = 0;
1319		memset(&svsk->sk_pages[0], 0, sizeof(svsk->sk_pages));
1320
1321		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1322
1323		set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1324		if (sk->sk_state != TCP_ESTABLISHED)
1325			set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
 
 
 
 
 
1326	}
1327}
1328
1329void svc_sock_update_bufs(struct svc_serv *serv)
1330{
1331	/*
1332	 * The number of server threads has changed. Update
1333	 * rcvbuf and sndbuf accordingly on all sockets
1334	 */
1335	struct svc_sock *svsk;
1336
1337	spin_lock_bh(&serv->sv_lock);
1338	list_for_each_entry(svsk, &serv->sv_permsocks, sk_xprt.xpt_list)
1339		set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1340	spin_unlock_bh(&serv->sv_lock);
1341}
1342EXPORT_SYMBOL_GPL(svc_sock_update_bufs);
1343
1344/*
1345 * Initialize socket for RPC use and create svc_sock struct
1346 * XXX: May want to setsockopt SO_SNDBUF and SO_RCVBUF.
1347 */
1348static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1349						struct socket *sock,
1350						int flags)
1351{
1352	struct svc_sock	*svsk;
1353	struct sock	*inet;
1354	int		pmap_register = !(flags & SVC_SOCK_ANONYMOUS);
1355	int		err = 0;
1356
1357	dprintk("svc: svc_setup_socket %p\n", sock);
1358	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1359	if (!svsk)
1360		return ERR_PTR(-ENOMEM);
1361
1362	inet = sock->sk;
1363
1364	/* Register socket with portmapper */
1365	if (pmap_register)
 
1366		err = svc_register(serv, sock_net(sock->sk), inet->sk_family,
1367				     inet->sk_protocol,
1368				     ntohs(inet_sk(inet)->inet_sport));
1369
1370	if (err < 0) {
1371		kfree(svsk);
1372		return ERR_PTR(err);
1373	}
1374
1375	inet->sk_user_data = svsk;
1376	svsk->sk_sock = sock;
1377	svsk->sk_sk = inet;
1378	svsk->sk_ostate = inet->sk_state_change;
1379	svsk->sk_odata = inet->sk_data_ready;
1380	svsk->sk_owspace = inet->sk_write_space;
 
 
 
 
 
 
 
1381
1382	/* Initialize the socket */
1383	if (sock->type == SOCK_DGRAM)
1384		svc_udp_init(svsk, serv);
1385	else {
1386		/* initialise setting must have enough space to
1387		 * receive and respond to one request.
1388		 */
1389		svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1390					4 * serv->sv_max_mesg);
1391		svc_tcp_init(svsk, serv);
1392	}
1393
1394	dprintk("svc: svc_setup_socket created %p (inet %p)\n",
1395				svsk, svsk->sk_sk);
1396
 
1397	return svsk;
1398}
1399
1400bool svc_alien_sock(struct net *net, int fd)
1401{
1402	int err;
1403	struct socket *sock = sockfd_lookup(fd, &err);
1404	bool ret = false;
1405
1406	if (!sock)
1407		goto out;
1408	if (sock_net(sock->sk) != net)
1409		ret = true;
1410	sockfd_put(sock);
1411out:
1412	return ret;
1413}
1414EXPORT_SYMBOL_GPL(svc_alien_sock);
1415
1416/**
1417 * svc_addsock - add a listener socket to an RPC service
1418 * @serv: pointer to RPC service to which to add a new listener
 
1419 * @fd: file descriptor of the new listener
1420 * @name_return: pointer to buffer to fill in with name of listener
1421 * @len: size of the buffer
 
1422 *
1423 * Fills in socket name and returns positive length of name if successful.
1424 * Name is terminated with '\n'.  On error, returns a negative errno
1425 * value.
1426 */
1427int svc_addsock(struct svc_serv *serv, const int fd, char *name_return,
1428		const size_t len)
1429{
1430	int err = 0;
1431	struct socket *so = sockfd_lookup(fd, &err);
1432	struct svc_sock *svsk = NULL;
1433	struct sockaddr_storage addr;
1434	struct sockaddr *sin = (struct sockaddr *)&addr;
1435	int salen;
1436
1437	if (!so)
1438		return err;
 
 
 
1439	err = -EAFNOSUPPORT;
1440	if ((so->sk->sk_family != PF_INET) && (so->sk->sk_family != PF_INET6))
1441		goto out;
1442	err =  -EPROTONOSUPPORT;
1443	if (so->sk->sk_protocol != IPPROTO_TCP &&
1444	    so->sk->sk_protocol != IPPROTO_UDP)
1445		goto out;
1446	err = -EISCONN;
1447	if (so->state > SS_UNCONNECTED)
1448		goto out;
1449	err = -ENOENT;
1450	if (!try_module_get(THIS_MODULE))
1451		goto out;
1452	svsk = svc_setup_socket(serv, so, SVC_SOCK_DEFAULTS);
1453	if (IS_ERR(svsk)) {
1454		module_put(THIS_MODULE);
1455		err = PTR_ERR(svsk);
1456		goto out;
1457	}
1458	if (kernel_getsockname(svsk->sk_sock, sin, &salen) == 0)
 
1459		svc_xprt_set_local(&svsk->sk_xprt, sin, salen);
 
1460	svc_add_new_perm_xprt(serv, &svsk->sk_xprt);
1461	return svc_one_sock_name(svsk, name_return, len);
1462out:
1463	sockfd_put(so);
1464	return err;
1465}
1466EXPORT_SYMBOL_GPL(svc_addsock);
1467
1468/*
1469 * Create socket for RPC service.
1470 */
1471static struct svc_xprt *svc_create_socket(struct svc_serv *serv,
1472					  int protocol,
1473					  struct net *net,
1474					  struct sockaddr *sin, int len,
1475					  int flags)
1476{
1477	struct svc_sock	*svsk;
1478	struct socket	*sock;
1479	int		error;
1480	int		type;
1481	struct sockaddr_storage addr;
1482	struct sockaddr *newsin = (struct sockaddr *)&addr;
1483	int		newlen;
1484	int		family;
1485	int		val;
1486	RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
1487
1488	dprintk("svc: svc_create_socket(%s, %d, %s)\n",
1489			serv->sv_program->pg_name, protocol,
1490			__svc_print_addr(sin, buf, sizeof(buf)));
1491
1492	if (protocol != IPPROTO_UDP && protocol != IPPROTO_TCP) {
1493		printk(KERN_WARNING "svc: only UDP and TCP "
1494				"sockets supported\n");
1495		return ERR_PTR(-EINVAL);
1496	}
1497
1498	type = (protocol == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1499	switch (sin->sa_family) {
1500	case AF_INET6:
1501		family = PF_INET6;
1502		break;
1503	case AF_INET:
1504		family = PF_INET;
1505		break;
1506	default:
1507		return ERR_PTR(-EINVAL);
1508	}
1509
1510	error = __sock_create(net, family, type, protocol, &sock, 1);
1511	if (error < 0)
1512		return ERR_PTR(error);
1513
1514	svc_reclassify_socket(sock);
1515
1516	/*
1517	 * If this is an PF_INET6 listener, we want to avoid
1518	 * getting requests from IPv4 remotes.  Those should
1519	 * be shunted to a PF_INET listener via rpcbind.
1520	 */
1521	val = 1;
1522	if (family == PF_INET6)
1523		kernel_setsockopt(sock, SOL_IPV6, IPV6_V6ONLY,
1524					(char *)&val, sizeof(val));
1525
1526	if (type == SOCK_STREAM)
1527		sock->sk->sk_reuse = SK_CAN_REUSE; /* allow address reuse */
1528	error = kernel_bind(sock, sin, len);
1529	if (error < 0)
1530		goto bummer;
1531
1532	newlen = len;
1533	error = kernel_getsockname(sock, newsin, &newlen);
1534	if (error < 0)
1535		goto bummer;
 
1536
1537	if (protocol == IPPROTO_TCP) {
1538		if ((error = kernel_listen(sock, 64)) < 0)
1539			goto bummer;
1540	}
1541
1542	svsk = svc_setup_socket(serv, sock, flags);
1543	if (IS_ERR(svsk)) {
1544		error = PTR_ERR(svsk);
1545		goto bummer;
1546	}
1547	svc_xprt_set_local(&svsk->sk_xprt, newsin, newlen);
1548	return (struct svc_xprt *)svsk;
1549bummer:
1550	dprintk("svc: svc_create_socket error = %d\n", -error);
1551	sock_release(sock);
1552	return ERR_PTR(error);
1553}
1554
1555/*
1556 * Detach the svc_sock from the socket so that no
1557 * more callbacks occur.
1558 */
1559static void svc_sock_detach(struct svc_xprt *xprt)
1560{
1561	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1562	struct sock *sk = svsk->sk_sk;
1563	wait_queue_head_t *wq;
1564
1565	dprintk("svc: svc_sock_detach(%p)\n", svsk);
1566
1567	/* put back the old socket callbacks */
 
1568	sk->sk_state_change = svsk->sk_ostate;
1569	sk->sk_data_ready = svsk->sk_odata;
1570	sk->sk_write_space = svsk->sk_owspace;
1571
1572	wq = sk_sleep(sk);
1573	if (wq && waitqueue_active(wq))
1574		wake_up_interruptible(wq);
1575}
1576
1577/*
1578 * Disconnect the socket, and reset the callbacks
1579 */
1580static void svc_tcp_sock_detach(struct svc_xprt *xprt)
1581{
1582	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1583
1584	dprintk("svc: svc_tcp_sock_detach(%p)\n", svsk);
1585
1586	svc_sock_detach(xprt);
1587
1588	if (!test_bit(XPT_LISTENER, &xprt->xpt_flags)) {
1589		svc_tcp_clear_pages(svsk);
1590		kernel_sock_shutdown(svsk->sk_sock, SHUT_RDWR);
1591	}
1592}
1593
1594/*
1595 * Free the svc_sock's socket resources and the svc_sock itself.
1596 */
1597static void svc_sock_free(struct svc_xprt *xprt)
1598{
1599	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
1600	dprintk("svc: svc_sock_free(%p)\n", svsk);
 
1601
1602	if (svsk->sk_sock->file)
1603		sockfd_put(svsk->sk_sock);
 
 
 
1604	else
1605		sock_release(svsk->sk_sock);
 
 
 
1606	kfree(svsk);
1607}
1608
1609#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1610/*
1611 * Create a back channel svc_xprt which shares the fore channel socket.
1612 */
1613static struct svc_xprt *svc_bc_create_socket(struct svc_serv *serv,
1614					     int protocol,
1615					     struct net *net,
1616					     struct sockaddr *sin, int len,
1617					     int flags)
1618{
1619	struct svc_sock *svsk;
1620	struct svc_xprt *xprt;
1621
1622	if (protocol != IPPROTO_TCP) {
1623		printk(KERN_WARNING "svc: only TCP sockets"
1624			" supported on shared back channel\n");
1625		return ERR_PTR(-EINVAL);
1626	}
1627
1628	svsk = kzalloc(sizeof(*svsk), GFP_KERNEL);
1629	if (!svsk)
1630		return ERR_PTR(-ENOMEM);
1631
1632	xprt = &svsk->sk_xprt;
1633	svc_xprt_init(net, &svc_tcp_bc_class, xprt, serv);
1634
1635	serv->sv_bc_xprt = xprt;
1636
1637	return xprt;
1638}
1639
1640/*
1641 * Free a back channel svc_sock.
1642 */
1643static void svc_bc_sock_free(struct svc_xprt *xprt)
1644{
1645	if (xprt)
1646		kfree(container_of(xprt, struct svc_sock, sk_xprt));
1647}
1648#endif /* CONFIG_SUNRPC_BACKCHANNEL */