Linux Audio

Check our new training course

Loading...
v4.6
 
   1/*
   2 * linux/net/sunrpc/xprtsock.c
   3 *
   4 * Client-side transport implementation for sockets.
   5 *
   6 * TCP callback races fixes (C) 1998 Red Hat
   7 * TCP send fixes (C) 1998 Red Hat
   8 * TCP NFS related read + write fixes
   9 *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10 *
  11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12 * Fix behaviour when socket buffer is full.
  13 *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14 *
  15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16 *
  17 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18 *   <gilles.quillard@bull.net>
  19 */
  20
  21#include <linux/types.h>
  22#include <linux/string.h>
  23#include <linux/slab.h>
  24#include <linux/module.h>
  25#include <linux/capability.h>
  26#include <linux/pagemap.h>
  27#include <linux/errno.h>
  28#include <linux/socket.h>
  29#include <linux/in.h>
  30#include <linux/net.h>
  31#include <linux/mm.h>
  32#include <linux/un.h>
  33#include <linux/udp.h>
  34#include <linux/tcp.h>
  35#include <linux/sunrpc/clnt.h>
  36#include <linux/sunrpc/addr.h>
  37#include <linux/sunrpc/sched.h>
  38#include <linux/sunrpc/svcsock.h>
  39#include <linux/sunrpc/xprtsock.h>
  40#include <linux/file.h>
  41#ifdef CONFIG_SUNRPC_BACKCHANNEL
  42#include <linux/sunrpc/bc_xprt.h>
  43#endif
  44
  45#include <net/sock.h>
  46#include <net/checksum.h>
  47#include <net/udp.h>
  48#include <net/tcp.h>
 
 
 
 
  49
  50#include <trace/events/sunrpc.h>
  51
 
  52#include "sunrpc.h"
  53
  54static void xs_close(struct rpc_xprt *xprt);
 
 
  55
  56/*
  57 * xprtsock tunables
  58 */
  59static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  60static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  61static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  62
  63static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  64static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  65
  66#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  67
  68#define XS_TCP_LINGER_TO	(15U * HZ)
  69static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  70
  71/*
  72 * We can register our own files under /proc/sys/sunrpc by
  73 * calling register_sysctl_table() again.  The files in that
  74 * directory become the union of all files registered there.
  75 *
  76 * We simply need to make sure that we don't collide with
  77 * someone else's file names!
  78 */
  79
  80static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  81static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  82static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  83static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  84static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  85
  86static struct ctl_table_header *sunrpc_table_header;
  87
 
 
 
 
 
  88/*
  89 * FIXME: changing the UDP slot table size should also resize the UDP
  90 *        socket buffers for existing UDP transports
  91 */
  92static struct ctl_table xs_tunables_table[] = {
  93	{
  94		.procname	= "udp_slot_table_entries",
  95		.data		= &xprt_udp_slot_table_entries,
  96		.maxlen		= sizeof(unsigned int),
  97		.mode		= 0644,
  98		.proc_handler	= proc_dointvec_minmax,
  99		.extra1		= &min_slot_table_size,
 100		.extra2		= &max_slot_table_size
 101	},
 102	{
 103		.procname	= "tcp_slot_table_entries",
 104		.data		= &xprt_tcp_slot_table_entries,
 105		.maxlen		= sizeof(unsigned int),
 106		.mode		= 0644,
 107		.proc_handler	= proc_dointvec_minmax,
 108		.extra1		= &min_slot_table_size,
 109		.extra2		= &max_slot_table_size
 110	},
 111	{
 112		.procname	= "tcp_max_slot_table_entries",
 113		.data		= &xprt_max_tcp_slot_table_entries,
 114		.maxlen		= sizeof(unsigned int),
 115		.mode		= 0644,
 116		.proc_handler	= proc_dointvec_minmax,
 117		.extra1		= &min_slot_table_size,
 118		.extra2		= &max_tcp_slot_table_limit
 119	},
 120	{
 121		.procname	= "min_resvport",
 122		.data		= &xprt_min_resvport,
 123		.maxlen		= sizeof(unsigned int),
 124		.mode		= 0644,
 125		.proc_handler	= proc_dointvec_minmax,
 126		.extra1		= &xprt_min_resvport_limit,
 127		.extra2		= &xprt_max_resvport_limit
 128	},
 129	{
 130		.procname	= "max_resvport",
 131		.data		= &xprt_max_resvport,
 132		.maxlen		= sizeof(unsigned int),
 133		.mode		= 0644,
 134		.proc_handler	= proc_dointvec_minmax,
 135		.extra1		= &xprt_min_resvport_limit,
 136		.extra2		= &xprt_max_resvport_limit
 137	},
 138	{
 139		.procname	= "tcp_fin_timeout",
 140		.data		= &xs_tcp_fin_timeout,
 141		.maxlen		= sizeof(xs_tcp_fin_timeout),
 142		.mode		= 0644,
 143		.proc_handler	= proc_dointvec_jiffies,
 144	},
 145	{ },
 146};
 147
 148static struct ctl_table sunrpc_table[] = {
 149	{
 150		.procname	= "sunrpc",
 151		.mode		= 0555,
 152		.child		= xs_tunables_table
 153	},
 154	{ },
 155};
 156
 157#endif
 158
 159/*
 160 * Wait duration for a reply from the RPC portmapper.
 161 */
 162#define XS_BIND_TO		(60U * HZ)
 163
 164/*
 165 * Delay if a UDP socket connect error occurs.  This is most likely some
 166 * kind of resource problem on the local host.
 167 */
 168#define XS_UDP_REEST_TO		(2U * HZ)
 169
 170/*
 171 * The reestablish timeout allows clients to delay for a bit before attempting
 172 * to reconnect to a server that just dropped our connection.
 173 *
 174 * We implement an exponential backoff when trying to reestablish a TCP
 175 * transport connection with the server.  Some servers like to drop a TCP
 176 * connection when they are overworked, so we start with a short timeout and
 177 * increase over time if the server is down or not responding.
 178 */
 179#define XS_TCP_INIT_REEST_TO	(3U * HZ)
 180#define XS_TCP_MAX_REEST_TO	(5U * 60 * HZ)
 181
 182/*
 183 * TCP idle timeout; client drops the transport socket if it is idle
 184 * for this long.  Note that we also timeout UDP sockets to prevent
 185 * holding port numbers when there is no RPC traffic.
 186 */
 187#define XS_IDLE_DISC_TO		(5U * 60 * HZ)
 188
 189#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 190# undef  RPC_DEBUG_DATA
 191# define RPCDBG_FACILITY	RPCDBG_TRANS
 192#endif
 193
 194#ifdef RPC_DEBUG_DATA
 195static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 196{
 197	u8 *buf = (u8 *) packet;
 198	int j;
 199
 200	dprintk("RPC:       %s\n", msg);
 201	for (j = 0; j < count && j < 128; j += 4) {
 202		if (!(j & 31)) {
 203			if (j)
 204				dprintk("\n");
 205			dprintk("0x%04x ", j);
 206		}
 207		dprintk("%02x%02x%02x%02x ",
 208			buf[j], buf[j+1], buf[j+2], buf[j+3]);
 209	}
 210	dprintk("\n");
 211}
 212#else
 213static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 214{
 215	/* NOP */
 216}
 217#endif
 218
 219static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 220{
 221	return (struct rpc_xprt *) sk->sk_user_data;
 222}
 223
 224static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 225{
 226	return (struct sockaddr *) &xprt->addr;
 227}
 228
 229static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 230{
 231	return (struct sockaddr_un *) &xprt->addr;
 232}
 233
 234static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 235{
 236	return (struct sockaddr_in *) &xprt->addr;
 237}
 238
 239static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 240{
 241	return (struct sockaddr_in6 *) &xprt->addr;
 242}
 243
 244static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 245{
 246	struct sockaddr *sap = xs_addr(xprt);
 247	struct sockaddr_in6 *sin6;
 248	struct sockaddr_in *sin;
 249	struct sockaddr_un *sun;
 250	char buf[128];
 251
 252	switch (sap->sa_family) {
 253	case AF_LOCAL:
 254		sun = xs_addr_un(xprt);
 255		strlcpy(buf, sun->sun_path, sizeof(buf));
 256		xprt->address_strings[RPC_DISPLAY_ADDR] =
 257						kstrdup(buf, GFP_KERNEL);
 258		break;
 259	case AF_INET:
 260		(void)rpc_ntop(sap, buf, sizeof(buf));
 261		xprt->address_strings[RPC_DISPLAY_ADDR] =
 262						kstrdup(buf, GFP_KERNEL);
 263		sin = xs_addr_in(xprt);
 264		snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 265		break;
 266	case AF_INET6:
 267		(void)rpc_ntop(sap, buf, sizeof(buf));
 268		xprt->address_strings[RPC_DISPLAY_ADDR] =
 269						kstrdup(buf, GFP_KERNEL);
 270		sin6 = xs_addr_in6(xprt);
 271		snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 272		break;
 273	default:
 274		BUG();
 275	}
 276
 277	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 278}
 279
 280static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 281{
 282	struct sockaddr *sap = xs_addr(xprt);
 283	char buf[128];
 284
 285	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 286	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 287
 288	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 289	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 290}
 291
 292static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 293				     const char *protocol,
 294				     const char *netid)
 295{
 296	xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 297	xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 298	xs_format_common_peer_addresses(xprt);
 299	xs_format_common_peer_ports(xprt);
 300}
 301
 302static void xs_update_peer_port(struct rpc_xprt *xprt)
 303{
 304	kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 305	kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 306
 307	xs_format_common_peer_ports(xprt);
 308}
 309
 310static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 311{
 312	unsigned int i;
 313
 314	for (i = 0; i < RPC_DISPLAY_MAX; i++)
 315		switch (i) {
 316		case RPC_DISPLAY_PROTO:
 317		case RPC_DISPLAY_NETID:
 318			continue;
 319		default:
 320			kfree(xprt->address_strings[i]);
 321		}
 322}
 323
 324#define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
 
 
 
 325
 326static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 327{
 328	struct msghdr msg = {
 329		.msg_name	= addr,
 330		.msg_namelen	= addrlen,
 331		.msg_flags	= XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 332	};
 333	struct kvec iov = {
 334		.iov_base	= vec->iov_base + base,
 335		.iov_len	= vec->iov_len - base,
 336	};
 337
 338	if (iov.iov_len != 0)
 339		return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 340	return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 
 
 
 341}
 342
 343static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
 
 
 
 344{
 345	ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
 346			int offset, size_t size, int flags);
 347	struct page **ppage;
 348	unsigned int remainder;
 349	int err;
 350
 351	remainder = xdr->page_len - base;
 352	base += xdr->page_base;
 353	ppage = xdr->pages + (base >> PAGE_SHIFT);
 354	base &= ~PAGE_MASK;
 355	do_sendpage = sock->ops->sendpage;
 356	if (!zerocopy)
 357		do_sendpage = sock_no_sendpage;
 358	for(;;) {
 359		unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 360		int flags = XS_SENDMSG_FLAGS;
 361
 362		remainder -= len;
 363		if (more)
 364			flags |= MSG_MORE;
 365		if (remainder != 0)
 366			flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
 367		err = do_sendpage(sock, *ppage, base, len, flags);
 368		if (remainder == 0 || err != len)
 369			break;
 370		*sent_p += err;
 371		ppage++;
 372		base = 0;
 373	}
 374	if (err > 0) {
 375		*sent_p += err;
 376		err = 0;
 377	}
 378	return err;
 379}
 380
 381/**
 382 * xs_sendpages - write pages directly to a socket
 383 * @sock: socket to send on
 384 * @addr: UDP only -- address of destination
 385 * @addrlen: UDP only -- length of destination address
 386 * @xdr: buffer containing this request
 387 * @base: starting position in the buffer
 388 * @zerocopy: true if it is safe to use sendpage()
 389 * @sent_p: return the total number of bytes successfully queued for sending
 390 *
 391 */
 392static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
 393{
 394	unsigned int remainder = xdr->len - base;
 395	int err = 0;
 396	int sent = 0;
 397
 398	if (unlikely(!sock))
 399		return -ENOTSOCK;
 400
 401	if (base != 0) {
 402		addr = NULL;
 403		addrlen = 0;
 404	}
 405
 406	if (base < xdr->head[0].iov_len || addr != NULL) {
 407		unsigned int len = xdr->head[0].iov_len - base;
 408		remainder -= len;
 409		err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 410		if (remainder == 0 || err != len)
 
 
 
 
 411			goto out;
 412		*sent_p += err;
 413		base = 0;
 414	} else
 415		base -= xdr->head[0].iov_len;
 
 
 
 416
 417	if (base < xdr->page_len) {
 418		unsigned int len = xdr->page_len - base;
 419		remainder -= len;
 420		err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
 421		*sent_p += sent;
 422		if (remainder == 0 || sent != len)
 
 
 
 
 
 
 
 
 423			goto out;
 424		base = 0;
 425	} else
 426		base -= xdr->page_len;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 427
 428	if (base >= xdr->tail[0].iov_len)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 429		return 0;
 430	err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 431out:
 432	if (err > 0) {
 433		*sent_p += err;
 434		err = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 435	}
 436	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 437}
 438
 439static void xs_nospace_callback(struct rpc_task *task)
 440{
 441	struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 
 
 
 
 
 
 442
 443	transport->inet->sk_write_pending--;
 
 
 
 
 
 
 444}
 445
 
 
 
 
 
 
 
 
 
 446/**
 447 * xs_nospace - place task on wait queue if transmit was incomplete
 448 * @task: task to put to sleep
 449 *
 450 */
 451static int xs_nospace(struct rpc_task *task)
 452{
 453	struct rpc_rqst *req = task->tk_rqstp;
 454	struct rpc_xprt *xprt = req->rq_xprt;
 455	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 456	struct sock *sk = transport->inet;
 457	int ret = -EAGAIN;
 458
 459	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 460			task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 461			req->rq_slen);
 462
 463	/* Protect against races with write_space */
 464	spin_lock_bh(&xprt->transport_lock);
 465
 466	/* Don't race with disconnect */
 467	if (xprt_connected(xprt)) {
 468		/* wait for more buffer space */
 469		sk->sk_write_pending++;
 470		xprt_wait_for_buffer_space(task, xs_nospace_callback);
 471	} else
 472		ret = -ENOTCONN;
 473
 474	spin_unlock_bh(&xprt->transport_lock);
 475
 476	/* Race breaker in case memory is freed before above code is called */
 477	sk->sk_write_space(sk);
 
 
 
 
 
 
 
 
 
 478	return ret;
 479}
 480
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 481/*
 482 * Construct a stream transport record marker in @buf.
 483 */
 484static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
 
 485{
 486	u32 reclen = buf->len - sizeof(rpc_fraghdr);
 487	rpc_fraghdr *base = buf->head[0].iov_base;
 488	*base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
 489}
 490
 491/**
 492 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 493 * @task: RPC task that manages the state of an RPC request
 494 *
 495 * Return values:
 496 *        0:	The request has been sent
 497 *   EAGAIN:	The socket was blocked, please call again later to
 498 *		complete the request
 499 * ENOTCONN:	Caller needs to invoke connect logic then call again
 500 *    other:	Some other error occured, the request was not sent
 501 */
 502static int xs_local_send_request(struct rpc_task *task)
 503{
 504	struct rpc_rqst *req = task->tk_rqstp;
 505	struct rpc_xprt *xprt = req->rq_xprt;
 506	struct sock_xprt *transport =
 507				container_of(xprt, struct sock_xprt, xprt);
 508	struct xdr_buf *xdr = &req->rq_snd_buf;
 
 
 
 
 
 
 509	int status;
 510	int sent = 0;
 511
 512	xs_encode_stream_record_marker(&req->rq_snd_buf);
 
 
 
 
 513
 514	xs_pktdump("packet data:",
 515			req->rq_svec->iov_base, req->rq_svec->iov_len);
 516
 517	status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
 518			      true, &sent);
 
 519	dprintk("RPC:       %s(%u) = %d\n",
 520			__func__, xdr->len - req->rq_bytes_sent, status);
 521
 522	if (status == -EAGAIN && sock_writeable(transport->inet))
 523		status = -ENOBUFS;
 524
 525	if (likely(sent > 0) || status == 0) {
 526		req->rq_bytes_sent += sent;
 527		req->rq_xmit_bytes_sent += sent;
 528		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 529			req->rq_bytes_sent = 0;
 
 530			return 0;
 531		}
 532		status = -EAGAIN;
 533	}
 534
 535	switch (status) {
 536	case -ENOBUFS:
 537		break;
 538	case -EAGAIN:
 539		status = xs_nospace(task);
 540		break;
 541	default:
 542		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 543			-status);
 
 544	case -EPIPE:
 545		xs_close(xprt);
 546		status = -ENOTCONN;
 547	}
 548
 549	return status;
 550}
 551
 552/**
 553 * xs_udp_send_request - write an RPC request to a UDP socket
 554 * @task: address of RPC task that manages the state of an RPC request
 555 *
 556 * Return values:
 557 *        0:	The request has been sent
 558 *   EAGAIN:	The socket was blocked, please call again later to
 559 *		complete the request
 560 * ENOTCONN:	Caller needs to invoke connect logic then call again
 561 *    other:	Some other error occurred, the request was not sent
 562 */
 563static int xs_udp_send_request(struct rpc_task *task)
 564{
 565	struct rpc_rqst *req = task->tk_rqstp;
 566	struct rpc_xprt *xprt = req->rq_xprt;
 567	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 568	struct xdr_buf *xdr = &req->rq_snd_buf;
 569	int sent = 0;
 
 
 
 
 
 570	int status;
 571
 572	xs_pktdump("packet data:",
 573				req->rq_svec->iov_base,
 574				req->rq_svec->iov_len);
 575
 576	if (!xprt_bound(xprt))
 577		return -ENOTCONN;
 578	status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
 579			      xdr, req->rq_bytes_sent, true, &sent);
 
 
 
 
 580
 581	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 582			xdr->len - req->rq_bytes_sent, status);
 583
 584	/* firewall is blocking us, don't return -EAGAIN or we end up looping */
 585	if (status == -EPERM)
 586		goto process_status;
 587
 588	if (status == -EAGAIN && sock_writeable(transport->inet))
 589		status = -ENOBUFS;
 590
 591	if (sent > 0 || status == 0) {
 592		req->rq_xmit_bytes_sent += sent;
 593		if (sent >= req->rq_slen)
 594			return 0;
 595		/* Still some bytes left; set up for a retry later. */
 596		status = -EAGAIN;
 597	}
 598
 599process_status:
 600	switch (status) {
 601	case -ENOTSOCK:
 602		status = -ENOTCONN;
 603		/* Should we call xs_close() here? */
 604		break;
 605	case -EAGAIN:
 606		status = xs_nospace(task);
 607		break;
 608	case -ENETUNREACH:
 609	case -ENOBUFS:
 610	case -EPIPE:
 611	case -ECONNREFUSED:
 612	case -EPERM:
 613		/* When the server has died, an ICMP port unreachable message
 614		 * prompts ECONNREFUSED. */
 615		break;
 616	default:
 617		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 618			-status);
 619	}
 620
 621	return status;
 622}
 623
 624/**
 625 * xs_tcp_send_request - write an RPC request to a TCP socket
 626 * @task: address of RPC task that manages the state of an RPC request
 627 *
 628 * Return values:
 629 *        0:	The request has been sent
 630 *   EAGAIN:	The socket was blocked, please call again later to
 631 *		complete the request
 632 * ENOTCONN:	Caller needs to invoke connect logic then call again
 633 *    other:	Some other error occurred, the request was not sent
 634 *
 635 * XXX: In the case of soft timeouts, should we eventually give up
 636 *	if sendmsg is not able to make progress?
 637 */
 638static int xs_tcp_send_request(struct rpc_task *task)
 639{
 640	struct rpc_rqst *req = task->tk_rqstp;
 641	struct rpc_xprt *xprt = req->rq_xprt;
 642	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 643	struct xdr_buf *xdr = &req->rq_snd_buf;
 644	bool zerocopy = true;
 
 
 
 
 
 
 645	int status;
 646	int sent;
 647
 648	xs_encode_stream_record_marker(&req->rq_snd_buf);
 
 
 
 
 
 
 
 649
 650	xs_pktdump("packet data:",
 651				req->rq_svec->iov_base,
 652				req->rq_svec->iov_len);
 653	/* Don't use zero copy if this is a resend. If the RPC call
 654	 * completes while the socket holds a reference to the pages,
 655	 * then we may end up resending corrupted data.
 656	 */
 657	if (task->tk_flags & RPC_TASK_SENT)
 658		zerocopy = false;
 659
 660	/* Continue transmitting the packet/record. We must be careful
 661	 * to cope with writespace callbacks arriving _after_ we have
 662	 * called sendmsg(). */
 
 
 663	while (1) {
 664		sent = 0;
 665		status = xs_sendpages(transport->sock, NULL, 0, xdr,
 666				      req->rq_bytes_sent, zerocopy, &sent);
 667
 668		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 669				xdr->len - req->rq_bytes_sent, status);
 670
 671		/* If we've sent the entire packet, immediately
 672		 * reset the count of bytes sent. */
 673		req->rq_bytes_sent += sent;
 674		req->rq_xmit_bytes_sent += sent;
 675		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 676			req->rq_bytes_sent = 0;
 
 
 
 677			return 0;
 678		}
 679
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 680		if (status < 0)
 681			break;
 682		if (sent == 0) {
 683			status = -EAGAIN;
 684			break;
 685		}
 686	}
 687	if (status == -EAGAIN && sk_stream_is_writeable(transport->inet))
 688		status = -ENOBUFS;
 689
 690	switch (status) {
 691	case -ENOTSOCK:
 692		status = -ENOTCONN;
 693		/* Should we call xs_close() here? */
 694		break;
 695	case -EAGAIN:
 696		status = xs_nospace(task);
 697		break;
 698	case -ECONNRESET:
 699	case -ECONNREFUSED:
 700	case -ENOTCONN:
 701	case -EADDRINUSE:
 702	case -ENOBUFS:
 703	case -EPIPE:
 704		break;
 705	default:
 706		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 707			-status);
 708	}
 709
 710	return status;
 711}
 712
 713/**
 714 * xs_tcp_release_xprt - clean up after a tcp transmission
 715 * @xprt: transport
 716 * @task: rpc task
 717 *
 718 * This cleans up if an error causes us to abort the transmission of a request.
 719 * In this case, the socket may need to be reset in order to avoid confusing
 720 * the server.
 721 */
 722static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 723{
 724	struct rpc_rqst *req;
 725
 726	if (task != xprt->snd_task)
 727		return;
 728	if (task == NULL)
 729		goto out_release;
 730	req = task->tk_rqstp;
 731	if (req == NULL)
 732		goto out_release;
 733	if (req->rq_bytes_sent == 0)
 734		goto out_release;
 735	if (req->rq_bytes_sent == req->rq_snd_buf.len)
 736		goto out_release;
 737	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
 738out_release:
 739	xprt_release_xprt(xprt, task);
 740}
 741
 742static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 743{
 744	transport->old_data_ready = sk->sk_data_ready;
 745	transport->old_state_change = sk->sk_state_change;
 746	transport->old_write_space = sk->sk_write_space;
 747	transport->old_error_report = sk->sk_error_report;
 748}
 749
 750static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 751{
 752	sk->sk_data_ready = transport->old_data_ready;
 753	sk->sk_state_change = transport->old_state_change;
 754	sk->sk_write_space = transport->old_write_space;
 755	sk->sk_error_report = transport->old_error_report;
 756}
 757
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 758static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
 759{
 760	smp_mb__before_atomic();
 761	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 762	clear_bit(XPRT_CLOSING, &xprt->state);
 
 763	smp_mb__after_atomic();
 764}
 765
 766static void xs_sock_mark_closed(struct rpc_xprt *xprt)
 767{
 768	xs_sock_reset_connection_flags(xprt);
 769	/* Mark transport as closed and wake up all pending tasks */
 770	xprt_disconnect_done(xprt);
 771}
 772
 773/**
 774 * xs_error_report - callback to handle TCP socket state errors
 775 * @sk: socket
 776 *
 777 * Note: we don't call sock_error() since there may be a rpc_task
 778 * using the socket, and so we don't want to clear sk->sk_err.
 779 */
 780static void xs_error_report(struct sock *sk)
 781{
 
 782	struct rpc_xprt *xprt;
 783	int err;
 784
 785	read_lock_bh(&sk->sk_callback_lock);
 786	if (!(xprt = xprt_from_sock(sk)))
 787		goto out;
 788
 789	err = -sk->sk_err;
 790	if (err == 0)
 
 791		goto out;
 792	/* Is this a reset event? */
 793	if (sk->sk_state == TCP_CLOSE)
 794		xs_sock_mark_closed(xprt);
 795	dprintk("RPC:       xs_error_report client %p, error=%d...\n",
 796			xprt, -err);
 797	trace_rpc_socket_error(xprt, sk->sk_socket, err);
 798	xprt_wake_pending_tasks(xprt, err);
 
 
 
 799 out:
 800	read_unlock_bh(&sk->sk_callback_lock);
 801}
 802
 803static void xs_reset_transport(struct sock_xprt *transport)
 804{
 805	struct socket *sock = transport->sock;
 806	struct sock *sk = transport->inet;
 807	struct rpc_xprt *xprt = &transport->xprt;
 
 808
 809	if (sk == NULL)
 810		return;
 811
 812	if (atomic_read(&transport->xprt.swapper))
 813		sk_clear_memalloc(sk);
 814
 815	kernel_sock_shutdown(sock, SHUT_RDWR);
 816
 817	mutex_lock(&transport->recv_mutex);
 818	write_lock_bh(&sk->sk_callback_lock);
 819	transport->inet = NULL;
 820	transport->sock = NULL;
 
 821
 822	sk->sk_user_data = NULL;
 823
 824	xs_restore_old_callbacks(transport, sk);
 825	xprt_clear_connected(xprt);
 826	write_unlock_bh(&sk->sk_callback_lock);
 827	xs_sock_reset_connection_flags(xprt);
 
 
 828	mutex_unlock(&transport->recv_mutex);
 829
 830	trace_rpc_socket_close(xprt, sock);
 831	sock_release(sock);
 
 
 832}
 833
 834/**
 835 * xs_close - close a socket
 836 * @xprt: transport
 837 *
 838 * This is used when all requests are complete; ie, no DRC state remains
 839 * on the server we want to save.
 840 *
 841 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 842 * xs_reset_transport() zeroing the socket from underneath a writer.
 843 */
 844static void xs_close(struct rpc_xprt *xprt)
 845{
 846	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 847
 848	dprintk("RPC:       xs_close xprt %p\n", xprt);
 849
 850	xs_reset_transport(transport);
 851	xprt->reestablish_timeout = 0;
 852
 853	xprt_disconnect_done(xprt);
 854}
 855
 856static void xs_inject_disconnect(struct rpc_xprt *xprt)
 857{
 858	dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
 859		xprt);
 860	xprt_disconnect_done(xprt);
 861}
 862
 863static void xs_xprt_free(struct rpc_xprt *xprt)
 864{
 865	xs_free_peer_addresses(xprt);
 866	xprt_free(xprt);
 867}
 868
 869/**
 870 * xs_destroy - prepare to shutdown a transport
 871 * @xprt: doomed transport
 872 *
 873 */
 874static void xs_destroy(struct rpc_xprt *xprt)
 875{
 876	struct sock_xprt *transport = container_of(xprt,
 877			struct sock_xprt, xprt);
 878	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 879
 880	cancel_delayed_work_sync(&transport->connect_worker);
 881	xs_close(xprt);
 882	cancel_work_sync(&transport->recv_worker);
 
 883	xs_xprt_free(xprt);
 884	module_put(THIS_MODULE);
 885}
 886
 887static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
 888{
 889	struct xdr_skb_reader desc = {
 890		.skb		= skb,
 891		.offset		= sizeof(rpc_fraghdr),
 892		.count		= skb->len - sizeof(rpc_fraghdr),
 893	};
 894
 895	if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
 896		return -1;
 897	if (desc.count)
 898		return -1;
 899	return 0;
 900}
 901
 902/**
 903 * xs_local_data_read_skb
 904 * @xprt: transport
 905 * @sk: socket
 906 * @skb: skbuff
 907 *
 908 * Currently this assumes we can read the whole reply in a single gulp.
 909 */
 910static void xs_local_data_read_skb(struct rpc_xprt *xprt,
 911		struct sock *sk,
 912		struct sk_buff *skb)
 913{
 914	struct rpc_task *task;
 915	struct rpc_rqst *rovr;
 916	int repsize, copied;
 917	u32 _xid;
 918	__be32 *xp;
 919
 920	repsize = skb->len - sizeof(rpc_fraghdr);
 921	if (repsize < 4) {
 922		dprintk("RPC:       impossible RPC reply size %d\n", repsize);
 923		return;
 924	}
 925
 926	/* Copy the XID from the skb... */
 927	xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
 928	if (xp == NULL)
 929		return;
 930
 931	/* Look up and lock the request corresponding to the given XID */
 932	spin_lock_bh(&xprt->transport_lock);
 933	rovr = xprt_lookup_rqst(xprt, *xp);
 934	if (!rovr)
 935		goto out_unlock;
 936	task = rovr->rq_task;
 937
 938	copied = rovr->rq_private_buf.buflen;
 939	if (copied > repsize)
 940		copied = repsize;
 941
 942	if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 943		dprintk("RPC:       sk_buff copy failed\n");
 944		goto out_unlock;
 945	}
 946
 947	xprt_complete_rqst(task, copied);
 948
 949 out_unlock:
 950	spin_unlock_bh(&xprt->transport_lock);
 951}
 952
 953static void xs_local_data_receive(struct sock_xprt *transport)
 954{
 955	struct sk_buff *skb;
 956	struct sock *sk;
 957	int err;
 958
 959	mutex_lock(&transport->recv_mutex);
 960	sk = transport->inet;
 961	if (sk == NULL)
 962		goto out;
 963	for (;;) {
 964		skb = skb_recv_datagram(sk, 0, 1, &err);
 965		if (skb == NULL)
 966			break;
 967		xs_local_data_read_skb(&transport->xprt, sk, skb);
 968		skb_free_datagram(sk, skb);
 969	}
 970out:
 971	mutex_unlock(&transport->recv_mutex);
 972}
 973
 974static void xs_local_data_receive_workfn(struct work_struct *work)
 975{
 976	struct sock_xprt *transport =
 977		container_of(work, struct sock_xprt, recv_worker);
 978	xs_local_data_receive(transport);
 979}
 980
 981/**
 982 * xs_udp_data_read_skb - receive callback for UDP sockets
 983 * @xprt: transport
 984 * @sk: socket
 985 * @skb: skbuff
 986 *
 987 */
 988static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
 989		struct sock *sk,
 990		struct sk_buff *skb)
 991{
 992	struct rpc_task *task;
 993	struct rpc_rqst *rovr;
 994	int repsize, copied;
 995	u32 _xid;
 996	__be32 *xp;
 997
 998	repsize = skb->len - sizeof(struct udphdr);
 999	if (repsize < 4) {
1000		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1001		return;
1002	}
1003
1004	/* Copy the XID from the skb... */
1005	xp = skb_header_pointer(skb, sizeof(struct udphdr),
1006				sizeof(_xid), &_xid);
1007	if (xp == NULL)
1008		return;
1009
1010	/* Look up and lock the request corresponding to the given XID */
1011	spin_lock_bh(&xprt->transport_lock);
1012	rovr = xprt_lookup_rqst(xprt, *xp);
1013	if (!rovr)
1014		goto out_unlock;
 
 
 
1015	task = rovr->rq_task;
1016
1017	if ((copied = rovr->rq_private_buf.buflen) > repsize)
1018		copied = repsize;
1019
1020	/* Suck it into the iovec, verify checksum if not done by hw. */
1021	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1022		UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1023		goto out_unlock;
 
1024	}
1025
1026	UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1027
 
1028	xprt_adjust_cwnd(xprt, task, copied);
 
 
1029	xprt_complete_rqst(task, copied);
1030
 
 
1031 out_unlock:
1032	spin_unlock_bh(&xprt->transport_lock);
1033}
1034
1035static void xs_udp_data_receive(struct sock_xprt *transport)
1036{
1037	struct sk_buff *skb;
1038	struct sock *sk;
1039	int err;
1040
1041	mutex_lock(&transport->recv_mutex);
1042	sk = transport->inet;
1043	if (sk == NULL)
1044		goto out;
1045	for (;;) {
1046		skb = skb_recv_datagram(sk, 0, 1, &err);
1047		if (skb == NULL)
1048			break;
1049		xs_udp_data_read_skb(&transport->xprt, sk, skb);
1050		skb_free_datagram(sk, skb);
 
1051	}
 
1052out:
1053	mutex_unlock(&transport->recv_mutex);
1054}
1055
1056static void xs_udp_data_receive_workfn(struct work_struct *work)
1057{
1058	struct sock_xprt *transport =
1059		container_of(work, struct sock_xprt, recv_worker);
 
 
1060	xs_udp_data_receive(transport);
 
1061}
1062
1063/**
1064 * xs_data_ready - "data ready" callback for UDP sockets
1065 * @sk: socket with data to read
1066 *
1067 */
1068static void xs_data_ready(struct sock *sk)
1069{
1070	struct rpc_xprt *xprt;
1071
1072	read_lock_bh(&sk->sk_callback_lock);
1073	dprintk("RPC:       xs_data_ready...\n");
1074	xprt = xprt_from_sock(sk);
1075	if (xprt != NULL) {
1076		struct sock_xprt *transport = container_of(xprt,
1077				struct sock_xprt, xprt);
1078		queue_work(rpciod_workqueue, &transport->recv_worker);
 
 
 
 
 
 
 
1079	}
1080	read_unlock_bh(&sk->sk_callback_lock);
1081}
1082
1083/*
1084 * Helper function to force a TCP close if the server is sending
1085 * junk and/or it has put us in CLOSE_WAIT
1086 */
1087static void xs_tcp_force_close(struct rpc_xprt *xprt)
1088{
1089	xprt_force_disconnect(xprt);
1090}
1091
1092static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1093{
1094	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1095	size_t len, used;
1096	char *p;
1097
1098	p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1099	len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1100	used = xdr_skb_read_bits(desc, p, len);
1101	transport->tcp_offset += used;
1102	if (used != len)
1103		return;
1104
1105	transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1106	if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1107		transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1108	else
1109		transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1110	transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1111
1112	transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1113	transport->tcp_offset = 0;
1114
1115	/* Sanity check of the record length */
1116	if (unlikely(transport->tcp_reclen < 8)) {
1117		dprintk("RPC:       invalid TCP record fragment length\n");
1118		xs_tcp_force_close(xprt);
1119		return;
1120	}
1121	dprintk("RPC:       reading TCP record fragment of length %d\n",
1122			transport->tcp_reclen);
1123}
1124
1125static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1126{
1127	if (transport->tcp_offset == transport->tcp_reclen) {
1128		transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1129		transport->tcp_offset = 0;
1130		if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1131			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1132			transport->tcp_flags |= TCP_RCV_COPY_XID;
1133			transport->tcp_copied = 0;
1134		}
1135	}
1136}
1137
1138static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1139{
1140	size_t len, used;
1141	char *p;
1142
1143	len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1144	dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1145	p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1146	used = xdr_skb_read_bits(desc, p, len);
1147	transport->tcp_offset += used;
1148	if (used != len)
1149		return;
1150	transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1151	transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1152	transport->tcp_copied = 4;
1153	dprintk("RPC:       reading %s XID %08x\n",
1154			(transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1155							      : "request with",
1156			ntohl(transport->tcp_xid));
1157	xs_tcp_check_fraghdr(transport);
1158}
1159
1160static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1161				       struct xdr_skb_reader *desc)
1162{
1163	size_t len, used;
1164	u32 offset;
1165	char *p;
1166
1167	/*
1168	 * We want transport->tcp_offset to be 8 at the end of this routine
1169	 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1170	 * When this function is called for the first time,
1171	 * transport->tcp_offset is 4 (after having already read the xid).
1172	 */
1173	offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1174	len = sizeof(transport->tcp_calldir) - offset;
1175	dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1176	p = ((char *) &transport->tcp_calldir) + offset;
1177	used = xdr_skb_read_bits(desc, p, len);
1178	transport->tcp_offset += used;
1179	if (used != len)
1180		return;
1181	transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1182	/*
1183	 * We don't yet have the XDR buffer, so we will write the calldir
1184	 * out after we get the buffer from the 'struct rpc_rqst'
1185	 */
1186	switch (ntohl(transport->tcp_calldir)) {
1187	case RPC_REPLY:
1188		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1189		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1190		transport->tcp_flags |= TCP_RPC_REPLY;
1191		break;
1192	case RPC_CALL:
1193		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1194		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1195		transport->tcp_flags &= ~TCP_RPC_REPLY;
1196		break;
1197	default:
1198		dprintk("RPC:       invalid request message type\n");
1199		xs_tcp_force_close(&transport->xprt);
1200	}
1201	xs_tcp_check_fraghdr(transport);
1202}
1203
1204static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1205				     struct xdr_skb_reader *desc,
1206				     struct rpc_rqst *req)
1207{
1208	struct sock_xprt *transport =
1209				container_of(xprt, struct sock_xprt, xprt);
1210	struct xdr_buf *rcvbuf;
1211	size_t len;
1212	ssize_t r;
1213
1214	rcvbuf = &req->rq_private_buf;
1215
1216	if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1217		/*
1218		 * Save the RPC direction in the XDR buffer
1219		 */
1220		memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1221			&transport->tcp_calldir,
1222			sizeof(transport->tcp_calldir));
1223		transport->tcp_copied += sizeof(transport->tcp_calldir);
1224		transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1225	}
1226
1227	len = desc->count;
1228	if (len > transport->tcp_reclen - transport->tcp_offset) {
1229		struct xdr_skb_reader my_desc;
1230
1231		len = transport->tcp_reclen - transport->tcp_offset;
1232		memcpy(&my_desc, desc, sizeof(my_desc));
1233		my_desc.count = len;
1234		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1235					  &my_desc, xdr_skb_read_bits);
1236		desc->count -= r;
1237		desc->offset += r;
1238	} else
1239		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1240					  desc, xdr_skb_read_bits);
1241
1242	if (r > 0) {
1243		transport->tcp_copied += r;
1244		transport->tcp_offset += r;
1245	}
1246	if (r != len) {
1247		/* Error when copying to the receive buffer,
1248		 * usually because we weren't able to allocate
1249		 * additional buffer pages. All we can do now
1250		 * is turn off TCP_RCV_COPY_DATA, so the request
1251		 * will not receive any additional updates,
1252		 * and time out.
1253		 * Any remaining data from this record will
1254		 * be discarded.
1255		 */
1256		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1257		dprintk("RPC:       XID %08x truncated request\n",
1258				ntohl(transport->tcp_xid));
1259		dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1260				"tcp_offset = %u, tcp_reclen = %u\n",
1261				xprt, transport->tcp_copied,
1262				transport->tcp_offset, transport->tcp_reclen);
1263		return;
1264	}
1265
1266	dprintk("RPC:       XID %08x read %Zd bytes\n",
1267			ntohl(transport->tcp_xid), r);
1268	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1269			"tcp_reclen = %u\n", xprt, transport->tcp_copied,
1270			transport->tcp_offset, transport->tcp_reclen);
1271
1272	if (transport->tcp_copied == req->rq_private_buf.buflen)
1273		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1274	else if (transport->tcp_offset == transport->tcp_reclen) {
1275		if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1276			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1277	}
1278}
1279
1280/*
1281 * Finds the request corresponding to the RPC xid and invokes the common
1282 * tcp read code to read the data.
1283 */
1284static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1285				    struct xdr_skb_reader *desc)
1286{
1287	struct sock_xprt *transport =
1288				container_of(xprt, struct sock_xprt, xprt);
1289	struct rpc_rqst *req;
1290
1291	dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1292
1293	/* Find and lock the request corresponding to this xid */
1294	spin_lock_bh(&xprt->transport_lock);
1295	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1296	if (!req) {
1297		dprintk("RPC:       XID %08x request not found!\n",
1298				ntohl(transport->tcp_xid));
1299		spin_unlock_bh(&xprt->transport_lock);
1300		return -1;
1301	}
1302
1303	xs_tcp_read_common(xprt, desc, req);
1304
1305	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1306		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1307
1308	spin_unlock_bh(&xprt->transport_lock);
1309	return 0;
1310}
1311
1312#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1313/*
1314 * Obtains an rpc_rqst previously allocated and invokes the common
1315 * tcp read code to read the data.  The result is placed in the callback
1316 * queue.
1317 * If we're unable to obtain the rpc_rqst we schedule the closing of the
1318 * connection and return -1.
1319 */
1320static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1321				       struct xdr_skb_reader *desc)
1322{
1323	struct sock_xprt *transport =
1324				container_of(xprt, struct sock_xprt, xprt);
1325	struct rpc_rqst *req;
1326
1327	/* Look up and lock the request corresponding to the given XID */
1328	spin_lock_bh(&xprt->transport_lock);
1329	req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1330	if (req == NULL) {
1331		spin_unlock_bh(&xprt->transport_lock);
1332		printk(KERN_WARNING "Callback slot table overflowed\n");
1333		xprt_force_disconnect(xprt);
1334		return -1;
1335	}
1336
1337	dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1338	xs_tcp_read_common(xprt, desc, req);
1339
1340	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1341		xprt_complete_bc_request(req, transport->tcp_copied);
1342	spin_unlock_bh(&xprt->transport_lock);
1343
1344	return 0;
1345}
1346
1347static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1348					struct xdr_skb_reader *desc)
1349{
1350	struct sock_xprt *transport =
1351				container_of(xprt, struct sock_xprt, xprt);
1352
1353	return (transport->tcp_flags & TCP_RPC_REPLY) ?
1354		xs_tcp_read_reply(xprt, desc) :
1355		xs_tcp_read_callback(xprt, desc);
1356}
1357
1358static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1359{
1360	int ret;
1361
1362	ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1363			      SVC_SOCK_ANONYMOUS);
1364	if (ret < 0)
1365		return ret;
1366	return 0;
1367}
1368#else
1369static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1370					struct xdr_skb_reader *desc)
1371{
1372	return xs_tcp_read_reply(xprt, desc);
1373}
1374#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1375
1376/*
1377 * Read data off the transport.  This can be either an RPC_CALL or an
1378 * RPC_REPLY.  Relay the processing to helper functions.
1379 */
1380static void xs_tcp_read_data(struct rpc_xprt *xprt,
1381				    struct xdr_skb_reader *desc)
1382{
1383	struct sock_xprt *transport =
1384				container_of(xprt, struct sock_xprt, xprt);
1385
1386	if (_xs_tcp_read_data(xprt, desc) == 0)
1387		xs_tcp_check_fraghdr(transport);
1388	else {
1389		/*
1390		 * The transport_lock protects the request handling.
1391		 * There's no need to hold it to update the tcp_flags.
1392		 */
1393		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1394	}
1395}
1396
1397static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1398{
1399	size_t len;
1400
1401	len = transport->tcp_reclen - transport->tcp_offset;
1402	if (len > desc->count)
1403		len = desc->count;
1404	desc->count -= len;
1405	desc->offset += len;
1406	transport->tcp_offset += len;
1407	dprintk("RPC:       discarded %Zu bytes\n", len);
1408	xs_tcp_check_fraghdr(transport);
1409}
1410
1411static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1412{
1413	struct rpc_xprt *xprt = rd_desc->arg.data;
1414	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1415	struct xdr_skb_reader desc = {
1416		.skb	= skb,
1417		.offset	= offset,
1418		.count	= len,
1419	};
1420
1421	dprintk("RPC:       xs_tcp_data_recv started\n");
1422	do {
1423		trace_xs_tcp_data_recv(transport);
1424		/* Read in a new fragment marker if necessary */
1425		/* Can we ever really expect to get completely empty fragments? */
1426		if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1427			xs_tcp_read_fraghdr(xprt, &desc);
1428			continue;
1429		}
1430		/* Read in the xid if necessary */
1431		if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1432			xs_tcp_read_xid(transport, &desc);
1433			continue;
1434		}
1435		/* Read in the call/reply flag */
1436		if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1437			xs_tcp_read_calldir(transport, &desc);
1438			continue;
1439		}
1440		/* Read in the request data */
1441		if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1442			xs_tcp_read_data(xprt, &desc);
1443			continue;
1444		}
1445		/* Skip over any trailing bytes on short reads */
1446		xs_tcp_read_discard(transport, &desc);
1447	} while (desc.count);
1448	trace_xs_tcp_data_recv(transport);
1449	dprintk("RPC:       xs_tcp_data_recv done\n");
1450	return len - desc.count;
1451}
1452
1453static void xs_tcp_data_receive(struct sock_xprt *transport)
1454{
1455	struct rpc_xprt *xprt = &transport->xprt;
1456	struct sock *sk;
1457	read_descriptor_t rd_desc = {
1458		.count = 2*1024*1024,
1459		.arg.data = xprt,
1460	};
1461	unsigned long total = 0;
1462	int read = 0;
1463
1464	mutex_lock(&transport->recv_mutex);
1465	sk = transport->inet;
1466	if (sk == NULL)
1467		goto out;
1468
1469	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1470	for (;;) {
1471		lock_sock(sk);
1472		read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1473		release_sock(sk);
1474		if (read <= 0)
1475			break;
1476		total += read;
1477		rd_desc.count = 65536;
1478	}
1479out:
1480	mutex_unlock(&transport->recv_mutex);
1481	trace_xs_tcp_data_ready(xprt, read, total);
1482}
1483
1484static void xs_tcp_data_receive_workfn(struct work_struct *work)
1485{
1486	struct sock_xprt *transport =
1487		container_of(work, struct sock_xprt, recv_worker);
1488	xs_tcp_data_receive(transport);
1489}
1490
1491/**
1492 * xs_tcp_data_ready - "data ready" callback for TCP sockets
1493 * @sk: socket with data to read
1494 *
1495 */
1496static void xs_tcp_data_ready(struct sock *sk)
1497{
1498	struct sock_xprt *transport;
1499	struct rpc_xprt *xprt;
1500
1501	dprintk("RPC:       xs_tcp_data_ready...\n");
1502
1503	read_lock_bh(&sk->sk_callback_lock);
1504	if (!(xprt = xprt_from_sock(sk)))
1505		goto out;
1506	transport = container_of(xprt, struct sock_xprt, xprt);
1507
1508	/* Any data means we had a useful conversation, so
1509	 * the we don't need to delay the next reconnect
1510	 */
1511	if (xprt->reestablish_timeout)
1512		xprt->reestablish_timeout = 0;
1513	queue_work(rpciod_workqueue, &transport->recv_worker);
1514
1515out:
1516	read_unlock_bh(&sk->sk_callback_lock);
1517}
1518
1519/**
1520 * xs_tcp_state_change - callback to handle TCP socket state changes
1521 * @sk: socket whose state has changed
1522 *
1523 */
1524static void xs_tcp_state_change(struct sock *sk)
1525{
1526	struct rpc_xprt *xprt;
1527	struct sock_xprt *transport;
1528
1529	read_lock_bh(&sk->sk_callback_lock);
1530	if (!(xprt = xprt_from_sock(sk)))
1531		goto out;
1532	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1533	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1534			sk->sk_state, xprt_connected(xprt),
1535			sock_flag(sk, SOCK_DEAD),
1536			sock_flag(sk, SOCK_ZAPPED),
1537			sk->sk_shutdown);
1538
1539	transport = container_of(xprt, struct sock_xprt, xprt);
1540	trace_rpc_socket_state_change(xprt, sk->sk_socket);
1541	switch (sk->sk_state) {
1542	case TCP_ESTABLISHED:
1543		spin_lock(&xprt->transport_lock);
1544		if (!xprt_test_and_set_connected(xprt)) {
1545
1546			/* Reset TCP record info */
1547			transport->tcp_offset = 0;
1548			transport->tcp_reclen = 0;
1549			transport->tcp_copied = 0;
1550			transport->tcp_flags =
1551				TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1552			xprt->connect_cookie++;
1553			clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1554			xprt_clear_connecting(xprt);
1555
1556			xprt_wake_pending_tasks(xprt, -EAGAIN);
 
 
 
1557		}
1558		spin_unlock(&xprt->transport_lock);
1559		break;
1560	case TCP_FIN_WAIT1:
1561		/* The client initiated a shutdown of the socket */
1562		xprt->connect_cookie++;
1563		xprt->reestablish_timeout = 0;
1564		set_bit(XPRT_CLOSING, &xprt->state);
1565		smp_mb__before_atomic();
1566		clear_bit(XPRT_CONNECTED, &xprt->state);
1567		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1568		smp_mb__after_atomic();
1569		break;
1570	case TCP_CLOSE_WAIT:
1571		/* The server initiated a shutdown of the socket */
1572		xprt->connect_cookie++;
1573		clear_bit(XPRT_CONNECTED, &xprt->state);
1574		xs_tcp_force_close(xprt);
 
1575	case TCP_CLOSING:
1576		/*
1577		 * If the server closed down the connection, make sure that
1578		 * we back off before reconnecting
1579		 */
1580		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1581			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1582		break;
1583	case TCP_LAST_ACK:
1584		set_bit(XPRT_CLOSING, &xprt->state);
1585		smp_mb__before_atomic();
1586		clear_bit(XPRT_CONNECTED, &xprt->state);
1587		smp_mb__after_atomic();
1588		break;
1589	case TCP_CLOSE:
1590		if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1591					&transport->sock_state))
1592			xprt_clear_connecting(xprt);
1593		xs_sock_mark_closed(xprt);
 
 
1594	}
1595 out:
1596	read_unlock_bh(&sk->sk_callback_lock);
1597}
1598
1599static void xs_write_space(struct sock *sk)
1600{
1601	struct socket_wq *wq;
 
1602	struct rpc_xprt *xprt;
1603
1604	if (!sk->sk_socket)
1605		return;
1606	clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1607
1608	if (unlikely(!(xprt = xprt_from_sock(sk))))
1609		return;
 
1610	rcu_read_lock();
1611	wq = rcu_dereference(sk->sk_wq);
1612	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1613		goto out;
1614
1615	xprt_write_space(xprt);
 
1616out:
1617	rcu_read_unlock();
1618}
1619
1620/**
1621 * xs_udp_write_space - callback invoked when socket buffer space
1622 *                             becomes available
1623 * @sk: socket whose state has changed
1624 *
1625 * Called when more output buffer space is available for this socket.
1626 * We try not to wake our writers until they can make "significant"
1627 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1628 * with a bunch of small requests.
1629 */
1630static void xs_udp_write_space(struct sock *sk)
1631{
1632	read_lock_bh(&sk->sk_callback_lock);
1633
1634	/* from net/core/sock.c:sock_def_write_space */
1635	if (sock_writeable(sk))
1636		xs_write_space(sk);
1637
1638	read_unlock_bh(&sk->sk_callback_lock);
1639}
1640
1641/**
1642 * xs_tcp_write_space - callback invoked when socket buffer space
1643 *                             becomes available
1644 * @sk: socket whose state has changed
1645 *
1646 * Called when more output buffer space is available for this socket.
1647 * We try not to wake our writers until they can make "significant"
1648 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1649 * with a bunch of small requests.
1650 */
1651static void xs_tcp_write_space(struct sock *sk)
1652{
1653	read_lock_bh(&sk->sk_callback_lock);
1654
1655	/* from net/core/stream.c:sk_stream_write_space */
1656	if (sk_stream_is_writeable(sk))
1657		xs_write_space(sk);
1658
1659	read_unlock_bh(&sk->sk_callback_lock);
1660}
1661
1662static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1663{
1664	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1665	struct sock *sk = transport->inet;
1666
1667	if (transport->rcvsize) {
1668		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1669		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1670	}
1671	if (transport->sndsize) {
1672		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1673		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1674		sk->sk_write_space(sk);
1675	}
1676}
1677
1678/**
1679 * xs_udp_set_buffer_size - set send and receive limits
1680 * @xprt: generic transport
1681 * @sndsize: requested size of send buffer, in bytes
1682 * @rcvsize: requested size of receive buffer, in bytes
1683 *
1684 * Set socket send and receive buffer size limits.
1685 */
1686static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1687{
1688	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1689
1690	transport->sndsize = 0;
1691	if (sndsize)
1692		transport->sndsize = sndsize + 1024;
1693	transport->rcvsize = 0;
1694	if (rcvsize)
1695		transport->rcvsize = rcvsize + 1024;
1696
1697	xs_udp_do_set_buffer_size(xprt);
1698}
1699
1700/**
1701 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
 
1702 * @task: task that timed out
1703 *
1704 * Adjust the congestion window after a retransmit timeout has occurred.
1705 */
1706static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1707{
 
1708	xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
 
1709}
1710
1711static unsigned short xs_get_random_port(void)
1712{
1713	unsigned short range = xprt_max_resvport - xprt_min_resvport;
1714	unsigned short rand = (unsigned short) prandom_u32() % range;
1715	return rand + xprt_min_resvport;
1716}
1717
1718/**
1719 * xs_set_reuseaddr_port - set the socket's port and address reuse options
1720 * @sock: socket
1721 *
1722 * Note that this function has to be called on all sockets that share the
1723 * same port, and it must be called before binding.
1724 */
1725static void xs_sock_set_reuseport(struct socket *sock)
1726{
1727	int opt = 1;
1728
1729	kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1730			(char *)&opt, sizeof(opt));
1731}
1732
1733static unsigned short xs_sock_getport(struct socket *sock)
1734{
1735	struct sockaddr_storage buf;
1736	int buflen;
1737	unsigned short port = 0;
1738
1739	if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1740		goto out;
1741	switch (buf.ss_family) {
1742	case AF_INET6:
1743		port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1744		break;
1745	case AF_INET:
1746		port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1747	}
1748out:
1749	return port;
1750}
1751
1752/**
1753 * xs_set_port - reset the port number in the remote endpoint address
1754 * @xprt: generic transport
1755 * @port: new port number
1756 *
1757 */
1758static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1759{
1760	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1761
1762	rpc_set_port(xs_addr(xprt), port);
1763	xs_update_peer_port(xprt);
1764}
1765
1766static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1767{
1768	if (transport->srcport == 0)
1769		transport->srcport = xs_sock_getport(sock);
1770}
1771
1772static unsigned short xs_get_srcport(struct sock_xprt *transport)
1773{
1774	unsigned short port = transport->srcport;
1775
1776	if (port == 0 && transport->xprt.resvport)
1777		port = xs_get_random_port();
1778	return port;
1779}
1780
 
 
 
 
 
 
 
1781static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1782{
1783	if (transport->srcport != 0)
1784		transport->srcport = 0;
1785	if (!transport->xprt.resvport)
1786		return 0;
1787	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1788		return xprt_max_resvport;
1789	return --port;
1790}
1791static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1792{
1793	struct sockaddr_storage myaddr;
1794	int err, nloop = 0;
1795	unsigned short port = xs_get_srcport(transport);
1796	unsigned short last;
1797
1798	/*
1799	 * If we are asking for any ephemeral port (i.e. port == 0 &&
1800	 * transport->xprt.resvport == 0), don't bind.  Let the local
1801	 * port selection happen implicitly when the socket is used
1802	 * (for example at connect time).
1803	 *
1804	 * This ensures that we can continue to establish TCP
1805	 * connections even when all local ephemeral ports are already
1806	 * a part of some TCP connection.  This makes no difference
1807	 * for UDP sockets, but also doens't harm them.
1808	 *
1809	 * If we're asking for any reserved port (i.e. port == 0 &&
1810	 * transport->xprt.resvport == 1) xs_get_srcport above will
1811	 * ensure that port is non-zero and we will bind as needed.
1812	 */
1813	if (port == 0)
1814		return 0;
1815
1816	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1817	do {
1818		rpc_set_port((struct sockaddr *)&myaddr, port);
1819		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1820				transport->xprt.addrlen);
1821		if (err == 0) {
1822			transport->srcport = port;
 
1823			break;
1824		}
1825		last = port;
1826		port = xs_next_srcport(transport, port);
1827		if (port > last)
1828			nloop++;
1829	} while (err == -EADDRINUSE && nloop != 2);
1830
1831	if (myaddr.ss_family == AF_INET)
1832		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1833				&((struct sockaddr_in *)&myaddr)->sin_addr,
1834				port, err ? "failed" : "ok", err);
1835	else
1836		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1837				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1838				port, err ? "failed" : "ok", err);
1839	return err;
1840}
1841
1842/*
1843 * We don't support autobind on AF_LOCAL sockets
1844 */
1845static void xs_local_rpcbind(struct rpc_task *task)
1846{
1847	xprt_set_bound(task->tk_xprt);
1848}
1849
1850static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1851{
1852}
1853
1854#ifdef CONFIG_DEBUG_LOCK_ALLOC
1855static struct lock_class_key xs_key[2];
1856static struct lock_class_key xs_slock_key[2];
1857
1858static inline void xs_reclassify_socketu(struct socket *sock)
1859{
1860	struct sock *sk = sock->sk;
1861
1862	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1863		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1864}
1865
1866static inline void xs_reclassify_socket4(struct socket *sock)
1867{
1868	struct sock *sk = sock->sk;
1869
1870	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1871		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1872}
1873
1874static inline void xs_reclassify_socket6(struct socket *sock)
1875{
1876	struct sock *sk = sock->sk;
1877
1878	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1879		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1880}
1881
1882static inline void xs_reclassify_socket(int family, struct socket *sock)
1883{
1884	WARN_ON_ONCE(sock_owned_by_user(sock->sk));
1885	if (sock_owned_by_user(sock->sk))
1886		return;
1887
1888	switch (family) {
1889	case AF_LOCAL:
1890		xs_reclassify_socketu(sock);
1891		break;
1892	case AF_INET:
1893		xs_reclassify_socket4(sock);
1894		break;
1895	case AF_INET6:
1896		xs_reclassify_socket6(sock);
1897		break;
1898	}
1899}
1900#else
1901static inline void xs_reclassify_socket(int family, struct socket *sock)
1902{
1903}
1904#endif
1905
1906static void xs_dummy_setup_socket(struct work_struct *work)
1907{
1908}
1909
1910static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1911		struct sock_xprt *transport, int family, int type,
1912		int protocol, bool reuseport)
1913{
 
1914	struct socket *sock;
1915	int err;
1916
1917	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1918	if (err < 0) {
1919		dprintk("RPC:       can't create %d transport socket (%d).\n",
1920				protocol, -err);
1921		goto out;
1922	}
1923	xs_reclassify_socket(family, sock);
1924
1925	if (reuseport)
1926		xs_sock_set_reuseport(sock);
1927
1928	err = xs_bind(transport, sock);
1929	if (err) {
1930		sock_release(sock);
1931		goto out;
1932	}
1933
 
 
 
 
 
1934	return sock;
1935out:
1936	return ERR_PTR(err);
1937}
1938
1939static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1940				      struct socket *sock)
1941{
1942	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1943									xprt);
1944
1945	if (!transport->inet) {
1946		struct sock *sk = sock->sk;
1947
1948		write_lock_bh(&sk->sk_callback_lock);
1949
1950		xs_save_old_callbacks(transport, sk);
1951
1952		sk->sk_user_data = xprt;
1953		sk->sk_data_ready = xs_data_ready;
1954		sk->sk_write_space = xs_udp_write_space;
 
1955		sk->sk_error_report = xs_error_report;
1956		sk->sk_allocation = GFP_NOIO;
1957
1958		xprt_clear_connected(xprt);
1959
1960		/* Reset to new socket */
1961		transport->sock = sock;
1962		transport->inet = sk;
1963
1964		write_unlock_bh(&sk->sk_callback_lock);
1965	}
1966
1967	/* Tell the socket layer to start connecting... */
1968	xprt->stat.connect_count++;
1969	xprt->stat.connect_start = jiffies;
1970	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1971}
1972
1973/**
1974 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1975 * @transport: socket transport to connect
1976 */
1977static int xs_local_setup_socket(struct sock_xprt *transport)
1978{
1979	struct rpc_xprt *xprt = &transport->xprt;
 
1980	struct socket *sock;
1981	int status = -EIO;
1982
1983	status = __sock_create(xprt->xprt_net, AF_LOCAL,
1984					SOCK_STREAM, 0, &sock, 1);
1985	if (status < 0) {
1986		dprintk("RPC:       can't create AF_LOCAL "
1987			"transport socket (%d).\n", -status);
1988		goto out;
1989	}
1990	xs_reclassify_socket(AF_LOCAL, sock);
1991
 
 
 
 
 
 
 
1992	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1993			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1994
1995	status = xs_local_finish_connecting(xprt, sock);
1996	trace_rpc_socket_connect(xprt, sock, status);
1997	switch (status) {
1998	case 0:
1999		dprintk("RPC:       xprt %p connected to %s\n",
2000				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
 
 
 
2001		xprt_set_connected(xprt);
 
2002	case -ENOBUFS:
2003		break;
2004	case -ENOENT:
2005		dprintk("RPC:       xprt %p: socket %s does not exist\n",
2006				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2007		break;
2008	case -ECONNREFUSED:
2009		dprintk("RPC:       xprt %p: connection refused for %s\n",
2010				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2011		break;
2012	default:
2013		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2014				__func__, -status,
2015				xprt->address_strings[RPC_DISPLAY_ADDR]);
2016	}
2017
2018out:
2019	xprt_clear_connecting(xprt);
2020	xprt_wake_pending_tasks(xprt, status);
2021	return status;
2022}
2023
2024static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2025{
2026	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2027	int ret;
2028
2029	 if (RPC_IS_ASYNC(task)) {
2030		/*
2031		 * We want the AF_LOCAL connect to be resolved in the
2032		 * filesystem namespace of the process making the rpc
2033		 * call.  Thus we connect synchronously.
2034		 *
2035		 * If we want to support asynchronous AF_LOCAL calls,
2036		 * we'll need to figure out how to pass a namespace to
2037		 * connect.
2038		 */
 
2039		rpc_exit(task, -ENOTCONN);
2040		return;
2041	}
2042	ret = xs_local_setup_socket(transport);
2043	if (ret && !RPC_IS_SOFTCONN(task))
2044		msleep_interruptible(15000);
2045}
2046
2047#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2048/*
2049 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2050 * know that we have exclusive access to the socket), to guard against
2051 * races with xs_reset_transport.
2052 */
2053static void xs_set_memalloc(struct rpc_xprt *xprt)
2054{
2055	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2056			xprt);
2057
2058	/*
2059	 * If there's no sock, then we have nothing to set. The
2060	 * reconnecting process will get it for us.
2061	 */
2062	if (!transport->inet)
2063		return;
2064	if (atomic_read(&xprt->swapper))
2065		sk_set_memalloc(transport->inet);
2066}
2067
2068/**
2069 * xs_enable_swap - Tag this transport as being used for swap.
2070 * @xprt: transport to tag
2071 *
2072 * Take a reference to this transport on behalf of the rpc_clnt, and
2073 * optionally mark it for swapping if it wasn't already.
2074 */
2075static int
2076xs_enable_swap(struct rpc_xprt *xprt)
2077{
2078	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2079
2080	if (atomic_inc_return(&xprt->swapper) != 1)
2081		return 0;
2082	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2083		return -ERESTARTSYS;
2084	if (xs->inet)
2085		sk_set_memalloc(xs->inet);
2086	xprt_release_xprt(xprt, NULL);
2087	return 0;
2088}
2089
2090/**
2091 * xs_disable_swap - Untag this transport as being used for swap.
2092 * @xprt: transport to tag
2093 *
2094 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2095 * swapper refcount goes to 0, untag the socket as a memalloc socket.
2096 */
2097static void
2098xs_disable_swap(struct rpc_xprt *xprt)
2099{
2100	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2101
2102	if (!atomic_dec_and_test(&xprt->swapper))
2103		return;
2104	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2105		return;
2106	if (xs->inet)
2107		sk_clear_memalloc(xs->inet);
2108	xprt_release_xprt(xprt, NULL);
2109}
2110#else
2111static void xs_set_memalloc(struct rpc_xprt *xprt)
2112{
2113}
2114
2115static int
2116xs_enable_swap(struct rpc_xprt *xprt)
2117{
2118	return -EINVAL;
2119}
2120
2121static void
2122xs_disable_swap(struct rpc_xprt *xprt)
2123{
2124}
2125#endif
2126
2127static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2128{
2129	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2130
2131	if (!transport->inet) {
2132		struct sock *sk = sock->sk;
2133
2134		write_lock_bh(&sk->sk_callback_lock);
2135
2136		xs_save_old_callbacks(transport, sk);
2137
2138		sk->sk_user_data = xprt;
2139		sk->sk_data_ready = xs_data_ready;
2140		sk->sk_write_space = xs_udp_write_space;
2141		sk->sk_allocation = GFP_NOIO;
2142
2143		xprt_set_connected(xprt);
2144
2145		/* Reset to new socket */
2146		transport->sock = sock;
2147		transport->inet = sk;
2148
2149		xs_set_memalloc(xprt);
2150
2151		write_unlock_bh(&sk->sk_callback_lock);
2152	}
2153	xs_udp_do_set_buffer_size(xprt);
 
 
2154}
2155
2156static void xs_udp_setup_socket(struct work_struct *work)
2157{
2158	struct sock_xprt *transport =
2159		container_of(work, struct sock_xprt, connect_worker.work);
2160	struct rpc_xprt *xprt = &transport->xprt;
2161	struct socket *sock = transport->sock;
2162	int status = -EIO;
2163
2164	sock = xs_create_sock(xprt, transport,
2165			xs_addr(xprt)->sa_family, SOCK_DGRAM,
2166			IPPROTO_UDP, false);
2167	if (IS_ERR(sock))
2168		goto out;
2169
2170	dprintk("RPC:       worker connecting xprt %p via %s to "
2171				"%s (port %s)\n", xprt,
2172			xprt->address_strings[RPC_DISPLAY_PROTO],
2173			xprt->address_strings[RPC_DISPLAY_ADDR],
2174			xprt->address_strings[RPC_DISPLAY_PORT]);
2175
2176	xs_udp_finish_connecting(xprt, sock);
2177	trace_rpc_socket_connect(xprt, sock, 0);
2178	status = 0;
2179out:
2180	xprt_unlock_connect(xprt, transport);
2181	xprt_clear_connecting(xprt);
 
2182	xprt_wake_pending_tasks(xprt, status);
2183}
2184
2185/**
2186 * xs_tcp_shutdown - gracefully shut down a TCP socket
2187 * @xprt: transport
2188 *
2189 * Initiates a graceful shutdown of the TCP socket by calling the
2190 * equivalent of shutdown(SHUT_RDWR);
2191 */
2192static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2193{
2194	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2195	struct socket *sock = transport->sock;
 
2196
2197	if (sock == NULL)
2198		return;
2199	if (xprt_connected(xprt)) {
 
2200		kernel_sock_shutdown(sock, SHUT_RDWR);
2201		trace_rpc_socket_shutdown(xprt, sock);
2202	} else
 
 
2203		xs_reset_transport(transport);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2204}
2205
2206static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2207{
2208	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2209	int ret = -ENOTCONN;
2210
2211	if (!transport->inet) {
2212		struct sock *sk = sock->sk;
2213		unsigned int keepidle = xprt->timeout->to_initval / HZ;
2214		unsigned int keepcnt = xprt->timeout->to_retries + 1;
2215		unsigned int opt_on = 1;
2216		unsigned int timeo;
2217
2218		/* TCP Keepalive options */
2219		kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2220				(char *)&opt_on, sizeof(opt_on));
2221		kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2222				(char *)&keepidle, sizeof(keepidle));
2223		kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2224				(char *)&keepidle, sizeof(keepidle));
2225		kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2226				(char *)&keepcnt, sizeof(keepcnt));
2227
2228		/* TCP user timeout (see RFC5482) */
2229		timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2230			(xprt->timeout->to_retries + 1);
2231		kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2232				(char *)&timeo, sizeof(timeo));
2233
2234		write_lock_bh(&sk->sk_callback_lock);
2235
2236		xs_save_old_callbacks(transport, sk);
2237
2238		sk->sk_user_data = xprt;
2239		sk->sk_data_ready = xs_tcp_data_ready;
2240		sk->sk_state_change = xs_tcp_state_change;
2241		sk->sk_write_space = xs_tcp_write_space;
 
2242		sk->sk_error_report = xs_error_report;
2243		sk->sk_allocation = GFP_NOIO;
2244
2245		/* socket options */
2246		sock_reset_flag(sk, SOCK_LINGER);
2247		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2248
2249		xprt_clear_connected(xprt);
2250
2251		/* Reset to new socket */
2252		transport->sock = sock;
2253		transport->inet = sk;
2254
2255		write_unlock_bh(&sk->sk_callback_lock);
2256	}
2257
2258	if (!xprt_bound(xprt))
2259		goto out;
2260
2261	xs_set_memalloc(xprt);
2262
 
 
2263	/* Tell the socket layer to start connecting... */
2264	xprt->stat.connect_count++;
2265	xprt->stat.connect_start = jiffies;
2266	set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2267	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2268	switch (ret) {
2269	case 0:
2270		xs_set_srcport(transport, sock);
 
2271	case -EINPROGRESS:
2272		/* SYN_SENT! */
2273		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2274			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
 
 
 
 
2275	}
2276out:
2277	return ret;
2278}
2279
2280/**
2281 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
 
2282 *
2283 * Invoked by a work queue tasklet.
2284 */
2285static void xs_tcp_setup_socket(struct work_struct *work)
2286{
2287	struct sock_xprt *transport =
2288		container_of(work, struct sock_xprt, connect_worker.work);
2289	struct socket *sock = transport->sock;
2290	struct rpc_xprt *xprt = &transport->xprt;
2291	int status = -EIO;
2292
2293	if (!sock) {
2294		sock = xs_create_sock(xprt, transport,
2295				xs_addr(xprt)->sa_family, SOCK_STREAM,
2296				IPPROTO_TCP, true);
2297		if (IS_ERR(sock)) {
2298			status = PTR_ERR(sock);
2299			goto out;
2300		}
2301	}
2302
2303	dprintk("RPC:       worker connecting xprt %p via %s to "
2304				"%s (port %s)\n", xprt,
2305			xprt->address_strings[RPC_DISPLAY_PROTO],
2306			xprt->address_strings[RPC_DISPLAY_ADDR],
2307			xprt->address_strings[RPC_DISPLAY_PORT]);
2308
2309	status = xs_tcp_finish_connecting(xprt, sock);
2310	trace_rpc_socket_connect(xprt, sock, status);
2311	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2312			xprt, -status, xprt_connected(xprt),
2313			sock->sk->sk_state);
2314	switch (status) {
2315	default:
2316		printk("%s: connect returned unhandled error %d\n",
2317			__func__, status);
 
2318	case -EADDRNOTAVAIL:
2319		/* We're probably in TIME_WAIT. Get rid of existing socket,
2320		 * and retry
2321		 */
2322		xs_tcp_force_close(xprt);
2323		break;
2324	case 0:
2325	case -EINPROGRESS:
2326	case -EALREADY:
2327		xprt_unlock_connect(xprt, transport);
2328		return;
2329	case -EINVAL:
2330		/* Happens, for instance, if the user specified a link
2331		 * local IPv6 address without a scope-id.
2332		 */
2333	case -ECONNREFUSED:
2334	case -ECONNRESET:
 
2335	case -ENETUNREACH:
 
2336	case -EADDRINUSE:
2337	case -ENOBUFS:
2338		/* retry with existing socket, after a delay */
 
 
 
2339		xs_tcp_force_close(xprt);
2340		goto out;
2341	}
2342	status = -EAGAIN;
2343out:
2344	xprt_unlock_connect(xprt, transport);
2345	xprt_clear_connecting(xprt);
 
2346	xprt_wake_pending_tasks(xprt, status);
2347}
2348
2349/**
2350 * xs_connect - connect a socket to a remote endpoint
2351 * @xprt: pointer to transport structure
2352 * @task: address of RPC task that manages state of connect request
2353 *
2354 * TCP: If the remote end dropped the connection, delay reconnecting.
2355 *
2356 * UDP socket connects are synchronous, but we use a work queue anyway
2357 * to guarantee that even unprivileged user processes can set up a
2358 * socket on a privileged port.
2359 *
2360 * If a UDP socket connect fails, the delay behavior here prevents
2361 * retry floods (hard mounts).
2362 */
2363static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2364{
2365	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 
2366
2367	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2368
2369	if (transport->sock != NULL) {
2370		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2371				"seconds\n",
2372				xprt, xprt->reestablish_timeout / HZ);
2373
2374		/* Start by resetting any existing state */
2375		xs_reset_transport(transport);
2376
2377		queue_delayed_work(rpciod_workqueue,
2378				   &transport->connect_worker,
2379				   xprt->reestablish_timeout);
2380		xprt->reestablish_timeout <<= 1;
2381		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2382			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2383		if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2384			xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2385	} else {
2386		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2387		queue_delayed_work(rpciod_workqueue,
2388				   &transport->connect_worker, 0);
2389	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2390}
2391
2392/**
2393 * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2394 * @xprt: rpc_xprt struct containing statistics
2395 * @seq: output file
2396 *
2397 */
2398static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2399{
2400	long idle_time = 0;
2401
2402	if (xprt_connected(xprt))
2403		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2404
2405	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2406			"%llu %llu %lu %llu %llu\n",
2407			xprt->stat.bind_count,
2408			xprt->stat.connect_count,
2409			xprt->stat.connect_time,
2410			idle_time,
2411			xprt->stat.sends,
2412			xprt->stat.recvs,
2413			xprt->stat.bad_xids,
2414			xprt->stat.req_u,
2415			xprt->stat.bklog_u,
2416			xprt->stat.max_slots,
2417			xprt->stat.sending_u,
2418			xprt->stat.pending_u);
2419}
2420
2421/**
2422 * xs_udp_print_stats - display UDP socket-specifc stats
2423 * @xprt: rpc_xprt struct containing statistics
2424 * @seq: output file
2425 *
2426 */
2427static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2428{
2429	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2430
2431	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2432			"%lu %llu %llu\n",
2433			transport->srcport,
2434			xprt->stat.bind_count,
2435			xprt->stat.sends,
2436			xprt->stat.recvs,
2437			xprt->stat.bad_xids,
2438			xprt->stat.req_u,
2439			xprt->stat.bklog_u,
2440			xprt->stat.max_slots,
2441			xprt->stat.sending_u,
2442			xprt->stat.pending_u);
2443}
2444
2445/**
2446 * xs_tcp_print_stats - display TCP socket-specifc stats
2447 * @xprt: rpc_xprt struct containing statistics
2448 * @seq: output file
2449 *
2450 */
2451static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2452{
2453	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2454	long idle_time = 0;
2455
2456	if (xprt_connected(xprt))
2457		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2458
2459	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2460			"%llu %llu %lu %llu %llu\n",
2461			transport->srcport,
2462			xprt->stat.bind_count,
2463			xprt->stat.connect_count,
2464			xprt->stat.connect_time,
2465			idle_time,
2466			xprt->stat.sends,
2467			xprt->stat.recvs,
2468			xprt->stat.bad_xids,
2469			xprt->stat.req_u,
2470			xprt->stat.bklog_u,
2471			xprt->stat.max_slots,
2472			xprt->stat.sending_u,
2473			xprt->stat.pending_u);
2474}
2475
2476/*
2477 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2478 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2479 * to use the server side send routines.
2480 */
2481static void *bc_malloc(struct rpc_task *task, size_t size)
2482{
 
 
2483	struct page *page;
2484	struct rpc_buffer *buf;
2485
2486	WARN_ON_ONCE(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2487	if (size > PAGE_SIZE - sizeof(struct rpc_buffer))
2488		return NULL;
 
 
2489
2490	page = alloc_page(GFP_KERNEL);
2491	if (!page)
2492		return NULL;
2493
2494	buf = page_address(page);
2495	buf->len = PAGE_SIZE;
2496
2497	return buf->data;
 
 
2498}
2499
2500/*
2501 * Free the space allocated in the bc_alloc routine
2502 */
2503static void bc_free(void *buffer)
2504{
 
2505	struct rpc_buffer *buf;
2506
2507	if (!buffer)
2508		return;
2509
2510	buf = container_of(buffer, struct rpc_buffer, data);
2511	free_page((unsigned long)buf);
2512}
2513
2514/*
2515 * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2516 * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2517 */
2518static int bc_sendto(struct rpc_rqst *req)
2519{
2520	int len;
2521	struct xdr_buf *xbufp = &req->rq_snd_buf;
2522	struct rpc_xprt *xprt = req->rq_xprt;
2523	struct sock_xprt *transport =
2524				container_of(xprt, struct sock_xprt, xprt);
2525	struct socket *sock = transport->sock;
2526	unsigned long headoff;
2527	unsigned long tailoff;
2528
2529	xs_encode_stream_record_marker(xbufp);
2530
2531	tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2532	headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2533	len = svc_send_common(sock, xbufp,
2534			      virt_to_page(xbufp->head[0].iov_base), headoff,
2535			      xbufp->tail[0].iov_base, tailoff);
2536
2537	if (len != xbufp->len) {
2538		printk(KERN_NOTICE "Error sending entire callback!\n");
2539		len = -EAGAIN;
2540	}
2541
2542	return len;
 
 
 
 
 
2543}
2544
2545/*
2546 * The send routine. Borrows from svc_send
 
 
 
 
 
 
 
 
2547 */
2548static int bc_send_request(struct rpc_task *task)
2549{
2550	struct rpc_rqst *req = task->tk_rqstp;
2551	struct svc_xprt	*xprt;
2552	int len;
2553
2554	dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2555	/*
2556	 * Get the server socket associated with this callback xprt
2557	 */
2558	xprt = req->rq_xprt->bc_xprt;
2559
2560	/*
2561	 * Grab the mutex to serialize data as the connection is shared
2562	 * with the fore channel
2563	 */
2564	if (!mutex_trylock(&xprt->xpt_mutex)) {
2565		rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2566		if (!mutex_trylock(&xprt->xpt_mutex))
2567			return -EAGAIN;
2568		rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2569	}
2570	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2571		len = -ENOTCONN;
2572	else
2573		len = bc_sendto(req);
2574	mutex_unlock(&xprt->xpt_mutex);
2575
2576	if (len > 0)
2577		len = 0;
2578
2579	return len;
2580}
2581
2582/*
2583 * The close routine. Since this is client initiated, we do nothing
2584 */
2585
2586static void bc_close(struct rpc_xprt *xprt)
2587{
 
2588}
2589
2590/*
2591 * The xprt destroy routine. Again, because this connection is client
2592 * initiated, we do nothing
2593 */
2594
2595static void bc_destroy(struct rpc_xprt *xprt)
2596{
2597	dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2598
2599	xs_xprt_free(xprt);
2600	module_put(THIS_MODULE);
2601}
2602
2603static struct rpc_xprt_ops xs_local_ops = {
2604	.reserve_xprt		= xprt_reserve_xprt,
2605	.release_xprt		= xs_tcp_release_xprt,
2606	.alloc_slot		= xprt_alloc_slot,
 
2607	.rpcbind		= xs_local_rpcbind,
2608	.set_port		= xs_local_set_port,
2609	.connect		= xs_local_connect,
2610	.buf_alloc		= rpc_malloc,
2611	.buf_free		= rpc_free,
 
2612	.send_request		= xs_local_send_request,
2613	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2614	.close			= xs_close,
2615	.destroy		= xs_destroy,
2616	.print_stats		= xs_local_print_stats,
2617	.enable_swap		= xs_enable_swap,
2618	.disable_swap		= xs_disable_swap,
2619};
2620
2621static struct rpc_xprt_ops xs_udp_ops = {
2622	.set_buffer_size	= xs_udp_set_buffer_size,
2623	.reserve_xprt		= xprt_reserve_xprt_cong,
2624	.release_xprt		= xprt_release_xprt_cong,
2625	.alloc_slot		= xprt_alloc_slot,
 
2626	.rpcbind		= rpcb_getport_async,
2627	.set_port		= xs_set_port,
2628	.connect		= xs_connect,
2629	.buf_alloc		= rpc_malloc,
2630	.buf_free		= rpc_free,
2631	.send_request		= xs_udp_send_request,
2632	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
2633	.timer			= xs_udp_timer,
2634	.release_request	= xprt_release_rqst_cong,
2635	.close			= xs_close,
2636	.destroy		= xs_destroy,
2637	.print_stats		= xs_udp_print_stats,
2638	.enable_swap		= xs_enable_swap,
2639	.disable_swap		= xs_disable_swap,
2640	.inject_disconnect	= xs_inject_disconnect,
2641};
2642
2643static struct rpc_xprt_ops xs_tcp_ops = {
2644	.reserve_xprt		= xprt_reserve_xprt,
2645	.release_xprt		= xs_tcp_release_xprt,
2646	.alloc_slot		= xprt_lock_and_alloc_slot,
 
2647	.rpcbind		= rpcb_getport_async,
2648	.set_port		= xs_set_port,
2649	.connect		= xs_connect,
2650	.buf_alloc		= rpc_malloc,
2651	.buf_free		= rpc_free,
 
2652	.send_request		= xs_tcp_send_request,
2653	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2654	.close			= xs_tcp_shutdown,
2655	.destroy		= xs_destroy,
 
2656	.print_stats		= xs_tcp_print_stats,
2657	.enable_swap		= xs_enable_swap,
2658	.disable_swap		= xs_disable_swap,
2659	.inject_disconnect	= xs_inject_disconnect,
2660#ifdef CONFIG_SUNRPC_BACKCHANNEL
2661	.bc_setup		= xprt_setup_bc,
2662	.bc_up			= xs_tcp_bc_up,
 
2663	.bc_free_rqst		= xprt_free_bc_rqst,
2664	.bc_destroy		= xprt_destroy_bc,
2665#endif
2666};
2667
2668/*
2669 * The rpc_xprt_ops for the server backchannel
2670 */
2671
2672static struct rpc_xprt_ops bc_tcp_ops = {
2673	.reserve_xprt		= xprt_reserve_xprt,
2674	.release_xprt		= xprt_release_xprt,
2675	.alloc_slot		= xprt_alloc_slot,
 
2676	.buf_alloc		= bc_malloc,
2677	.buf_free		= bc_free,
2678	.send_request		= bc_send_request,
2679	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2680	.close			= bc_close,
2681	.destroy		= bc_destroy,
2682	.print_stats		= xs_tcp_print_stats,
2683	.enable_swap		= xs_enable_swap,
2684	.disable_swap		= xs_disable_swap,
2685	.inject_disconnect	= xs_inject_disconnect,
2686};
2687
2688static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2689{
2690	static const struct sockaddr_in sin = {
2691		.sin_family		= AF_INET,
2692		.sin_addr.s_addr	= htonl(INADDR_ANY),
2693	};
2694	static const struct sockaddr_in6 sin6 = {
2695		.sin6_family		= AF_INET6,
2696		.sin6_addr		= IN6ADDR_ANY_INIT,
2697	};
2698
2699	switch (family) {
2700	case AF_LOCAL:
2701		break;
2702	case AF_INET:
2703		memcpy(sap, &sin, sizeof(sin));
2704		break;
2705	case AF_INET6:
2706		memcpy(sap, &sin6, sizeof(sin6));
2707		break;
2708	default:
2709		dprintk("RPC:       %s: Bad address family\n", __func__);
2710		return -EAFNOSUPPORT;
2711	}
2712	return 0;
2713}
2714
2715static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2716				      unsigned int slot_table_size,
2717				      unsigned int max_slot_table_size)
2718{
2719	struct rpc_xprt *xprt;
2720	struct sock_xprt *new;
2721
2722	if (args->addrlen > sizeof(xprt->addr)) {
2723		dprintk("RPC:       xs_setup_xprt: address too large\n");
2724		return ERR_PTR(-EBADF);
2725	}
2726
2727	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2728			max_slot_table_size);
2729	if (xprt == NULL) {
2730		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2731				"rpc_xprt\n");
2732		return ERR_PTR(-ENOMEM);
2733	}
2734
2735	new = container_of(xprt, struct sock_xprt, xprt);
2736	mutex_init(&new->recv_mutex);
2737	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2738	xprt->addrlen = args->addrlen;
2739	if (args->srcaddr)
2740		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2741	else {
2742		int err;
2743		err = xs_init_anyaddr(args->dstaddr->sa_family,
2744					(struct sockaddr *)&new->srcaddr);
2745		if (err != 0) {
2746			xprt_free(xprt);
2747			return ERR_PTR(err);
2748		}
2749	}
2750
2751	return xprt;
2752}
2753
2754static const struct rpc_timeout xs_local_default_timeout = {
2755	.to_initval = 10 * HZ,
2756	.to_maxval = 10 * HZ,
2757	.to_retries = 2,
2758};
2759
2760/**
2761 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2762 * @args: rpc transport creation arguments
2763 *
2764 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2765 */
2766static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2767{
2768	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2769	struct sock_xprt *transport;
2770	struct rpc_xprt *xprt;
2771	struct rpc_xprt *ret;
2772
2773	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2774			xprt_max_tcp_slot_table_entries);
2775	if (IS_ERR(xprt))
2776		return xprt;
2777	transport = container_of(xprt, struct sock_xprt, xprt);
2778
2779	xprt->prot = 0;
2780	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2781	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2782
2783	xprt->bind_timeout = XS_BIND_TO;
2784	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2785	xprt->idle_timeout = XS_IDLE_DISC_TO;
2786
2787	xprt->ops = &xs_local_ops;
2788	xprt->timeout = &xs_local_default_timeout;
2789
2790	INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2791	INIT_DELAYED_WORK(&transport->connect_worker,
2792			xs_dummy_setup_socket);
2793
2794	switch (sun->sun_family) {
2795	case AF_LOCAL:
2796		if (sun->sun_path[0] != '/') {
2797			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2798					sun->sun_path);
2799			ret = ERR_PTR(-EINVAL);
2800			goto out_err;
2801		}
2802		xprt_set_bound(xprt);
2803		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2804		ret = ERR_PTR(xs_local_setup_socket(transport));
2805		if (ret)
2806			goto out_err;
2807		break;
2808	default:
2809		ret = ERR_PTR(-EAFNOSUPPORT);
2810		goto out_err;
2811	}
2812
2813	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2814			xprt->address_strings[RPC_DISPLAY_ADDR]);
2815
2816	if (try_module_get(THIS_MODULE))
2817		return xprt;
2818	ret = ERR_PTR(-EINVAL);
2819out_err:
2820	xs_xprt_free(xprt);
2821	return ret;
2822}
2823
2824static const struct rpc_timeout xs_udp_default_timeout = {
2825	.to_initval = 5 * HZ,
2826	.to_maxval = 30 * HZ,
2827	.to_increment = 5 * HZ,
2828	.to_retries = 5,
2829};
2830
2831/**
2832 * xs_setup_udp - Set up transport to use a UDP socket
2833 * @args: rpc transport creation arguments
2834 *
2835 */
2836static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2837{
2838	struct sockaddr *addr = args->dstaddr;
2839	struct rpc_xprt *xprt;
2840	struct sock_xprt *transport;
2841	struct rpc_xprt *ret;
2842
2843	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2844			xprt_udp_slot_table_entries);
2845	if (IS_ERR(xprt))
2846		return xprt;
2847	transport = container_of(xprt, struct sock_xprt, xprt);
2848
2849	xprt->prot = IPPROTO_UDP;
2850	xprt->tsh_size = 0;
2851	/* XXX: header size can vary due to auth type, IPv6, etc. */
2852	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2853
2854	xprt->bind_timeout = XS_BIND_TO;
2855	xprt->reestablish_timeout = XS_UDP_REEST_TO;
2856	xprt->idle_timeout = XS_IDLE_DISC_TO;
2857
2858	xprt->ops = &xs_udp_ops;
2859
2860	xprt->timeout = &xs_udp_default_timeout;
2861
2862	INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
 
2863	INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2864
2865	switch (addr->sa_family) {
2866	case AF_INET:
2867		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2868			xprt_set_bound(xprt);
2869
2870		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2871		break;
2872	case AF_INET6:
2873		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2874			xprt_set_bound(xprt);
2875
2876		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2877		break;
2878	default:
2879		ret = ERR_PTR(-EAFNOSUPPORT);
2880		goto out_err;
2881	}
2882
2883	if (xprt_bound(xprt))
2884		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2885				xprt->address_strings[RPC_DISPLAY_ADDR],
2886				xprt->address_strings[RPC_DISPLAY_PORT],
2887				xprt->address_strings[RPC_DISPLAY_PROTO]);
2888	else
2889		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2890				xprt->address_strings[RPC_DISPLAY_ADDR],
2891				xprt->address_strings[RPC_DISPLAY_PROTO]);
2892
2893	if (try_module_get(THIS_MODULE))
2894		return xprt;
2895	ret = ERR_PTR(-EINVAL);
2896out_err:
2897	xs_xprt_free(xprt);
2898	return ret;
2899}
2900
2901static const struct rpc_timeout xs_tcp_default_timeout = {
2902	.to_initval = 60 * HZ,
2903	.to_maxval = 60 * HZ,
2904	.to_retries = 2,
2905};
2906
2907/**
2908 * xs_setup_tcp - Set up transport to use a TCP socket
2909 * @args: rpc transport creation arguments
2910 *
2911 */
2912static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2913{
2914	struct sockaddr *addr = args->dstaddr;
2915	struct rpc_xprt *xprt;
2916	struct sock_xprt *transport;
2917	struct rpc_xprt *ret;
2918	unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2919
2920	if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2921		max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2922
2923	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2924			max_slot_table_size);
2925	if (IS_ERR(xprt))
2926		return xprt;
2927	transport = container_of(xprt, struct sock_xprt, xprt);
2928
2929	xprt->prot = IPPROTO_TCP;
2930	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2931	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2932
2933	xprt->bind_timeout = XS_BIND_TO;
2934	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2935	xprt->idle_timeout = XS_IDLE_DISC_TO;
2936
2937	xprt->ops = &xs_tcp_ops;
2938	xprt->timeout = &xs_tcp_default_timeout;
2939
2940	INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
 
 
 
 
 
2941	INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2942
2943	switch (addr->sa_family) {
2944	case AF_INET:
2945		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2946			xprt_set_bound(xprt);
2947
2948		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2949		break;
2950	case AF_INET6:
2951		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2952			xprt_set_bound(xprt);
2953
2954		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2955		break;
2956	default:
2957		ret = ERR_PTR(-EAFNOSUPPORT);
2958		goto out_err;
2959	}
2960
2961	if (xprt_bound(xprt))
2962		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2963				xprt->address_strings[RPC_DISPLAY_ADDR],
2964				xprt->address_strings[RPC_DISPLAY_PORT],
2965				xprt->address_strings[RPC_DISPLAY_PROTO]);
2966	else
2967		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2968				xprt->address_strings[RPC_DISPLAY_ADDR],
2969				xprt->address_strings[RPC_DISPLAY_PROTO]);
2970
2971	if (try_module_get(THIS_MODULE))
2972		return xprt;
2973	ret = ERR_PTR(-EINVAL);
2974out_err:
2975	xs_xprt_free(xprt);
2976	return ret;
2977}
2978
2979/**
2980 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2981 * @args: rpc transport creation arguments
2982 *
2983 */
2984static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2985{
2986	struct sockaddr *addr = args->dstaddr;
2987	struct rpc_xprt *xprt;
2988	struct sock_xprt *transport;
2989	struct svc_sock *bc_sock;
2990	struct rpc_xprt *ret;
2991
2992	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2993			xprt_tcp_slot_table_entries);
2994	if (IS_ERR(xprt))
2995		return xprt;
2996	transport = container_of(xprt, struct sock_xprt, xprt);
2997
2998	xprt->prot = IPPROTO_TCP;
2999	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3000	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3001	xprt->timeout = &xs_tcp_default_timeout;
3002
3003	/* backchannel */
3004	xprt_set_bound(xprt);
3005	xprt->bind_timeout = 0;
3006	xprt->reestablish_timeout = 0;
3007	xprt->idle_timeout = 0;
3008
3009	xprt->ops = &bc_tcp_ops;
3010
3011	switch (addr->sa_family) {
3012	case AF_INET:
3013		xs_format_peer_addresses(xprt, "tcp",
3014					 RPCBIND_NETID_TCP);
3015		break;
3016	case AF_INET6:
3017		xs_format_peer_addresses(xprt, "tcp",
3018				   RPCBIND_NETID_TCP6);
3019		break;
3020	default:
3021		ret = ERR_PTR(-EAFNOSUPPORT);
3022		goto out_err;
3023	}
3024
3025	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3026			xprt->address_strings[RPC_DISPLAY_ADDR],
3027			xprt->address_strings[RPC_DISPLAY_PORT],
3028			xprt->address_strings[RPC_DISPLAY_PROTO]);
3029
3030	/*
3031	 * Once we've associated a backchannel xprt with a connection,
3032	 * we want to keep it around as long as the connection lasts,
3033	 * in case we need to start using it for a backchannel again;
3034	 * this reference won't be dropped until bc_xprt is destroyed.
3035	 */
3036	xprt_get(xprt);
3037	args->bc_xprt->xpt_bc_xprt = xprt;
3038	xprt->bc_xprt = args->bc_xprt;
3039	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3040	transport->sock = bc_sock->sk_sock;
3041	transport->inet = bc_sock->sk_sk;
3042
3043	/*
3044	 * Since we don't want connections for the backchannel, we set
3045	 * the xprt status to connected
3046	 */
3047	xprt_set_connected(xprt);
3048
3049	if (try_module_get(THIS_MODULE))
3050		return xprt;
3051
3052	args->bc_xprt->xpt_bc_xprt = NULL;
 
3053	xprt_put(xprt);
3054	ret = ERR_PTR(-EINVAL);
3055out_err:
3056	xs_xprt_free(xprt);
3057	return ret;
3058}
3059
3060static struct xprt_class	xs_local_transport = {
3061	.list		= LIST_HEAD_INIT(xs_local_transport.list),
3062	.name		= "named UNIX socket",
3063	.owner		= THIS_MODULE,
3064	.ident		= XPRT_TRANSPORT_LOCAL,
3065	.setup		= xs_setup_local,
 
3066};
3067
3068static struct xprt_class	xs_udp_transport = {
3069	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
3070	.name		= "udp",
3071	.owner		= THIS_MODULE,
3072	.ident		= XPRT_TRANSPORT_UDP,
3073	.setup		= xs_setup_udp,
 
3074};
3075
3076static struct xprt_class	xs_tcp_transport = {
3077	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
3078	.name		= "tcp",
3079	.owner		= THIS_MODULE,
3080	.ident		= XPRT_TRANSPORT_TCP,
3081	.setup		= xs_setup_tcp,
 
3082};
3083
3084static struct xprt_class	xs_bc_tcp_transport = {
3085	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3086	.name		= "tcp NFSv4.1 backchannel",
3087	.owner		= THIS_MODULE,
3088	.ident		= XPRT_TRANSPORT_BC_TCP,
3089	.setup		= xs_setup_bc_tcp,
 
3090};
3091
3092/**
3093 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3094 *
3095 */
3096int init_socket_xprt(void)
3097{
3098#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3099	if (!sunrpc_table_header)
3100		sunrpc_table_header = register_sysctl_table(sunrpc_table);
3101#endif
3102
3103	xprt_register_transport(&xs_local_transport);
3104	xprt_register_transport(&xs_udp_transport);
3105	xprt_register_transport(&xs_tcp_transport);
3106	xprt_register_transport(&xs_bc_tcp_transport);
3107
3108	return 0;
3109}
3110
3111/**
3112 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3113 *
3114 */
3115void cleanup_socket_xprt(void)
3116{
3117#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3118	if (sunrpc_table_header) {
3119		unregister_sysctl_table(sunrpc_table_header);
3120		sunrpc_table_header = NULL;
3121	}
3122#endif
3123
3124	xprt_unregister_transport(&xs_local_transport);
3125	xprt_unregister_transport(&xs_udp_transport);
3126	xprt_unregister_transport(&xs_tcp_transport);
3127	xprt_unregister_transport(&xs_bc_tcp_transport);
3128}
3129
3130static int param_set_uint_minmax(const char *val,
3131		const struct kernel_param *kp,
3132		unsigned int min, unsigned int max)
3133{
3134	unsigned int num;
3135	int ret;
3136
3137	if (!val)
3138		return -EINVAL;
3139	ret = kstrtouint(val, 0, &num);
3140	if (ret == -EINVAL || num < min || num > max)
 
 
3141		return -EINVAL;
3142	*((unsigned int *)kp->arg) = num;
3143	return 0;
3144}
3145
3146static int param_set_portnr(const char *val, const struct kernel_param *kp)
3147{
3148	return param_set_uint_minmax(val, kp,
3149			RPC_MIN_RESVPORT,
3150			RPC_MAX_RESVPORT);
3151}
3152
3153static const struct kernel_param_ops param_ops_portnr = {
3154	.set = param_set_portnr,
3155	.get = param_get_uint,
3156};
3157
3158#define param_check_portnr(name, p) \
3159	__param_check(name, p, unsigned int);
3160
3161module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3162module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3163
3164static int param_set_slot_table_size(const char *val,
3165				     const struct kernel_param *kp)
3166{
3167	return param_set_uint_minmax(val, kp,
3168			RPC_MIN_SLOT_TABLE,
3169			RPC_MAX_SLOT_TABLE);
3170}
3171
3172static const struct kernel_param_ops param_ops_slot_table_size = {
3173	.set = param_set_slot_table_size,
3174	.get = param_get_uint,
3175};
3176
3177#define param_check_slot_table_size(name, p) \
3178	__param_check(name, p, unsigned int);
3179
3180static int param_set_max_slot_table_size(const char *val,
3181				     const struct kernel_param *kp)
3182{
3183	return param_set_uint_minmax(val, kp,
3184			RPC_MIN_SLOT_TABLE,
3185			RPC_MAX_SLOT_TABLE_LIMIT);
3186}
3187
3188static const struct kernel_param_ops param_ops_max_slot_table_size = {
3189	.set = param_set_max_slot_table_size,
3190	.get = param_get_uint,
3191};
3192
3193#define param_check_max_slot_table_size(name, p) \
3194	__param_check(name, p, unsigned int);
3195
3196module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3197		   slot_table_size, 0644);
3198module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3199		   max_slot_table_size, 0644);
3200module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3201		   slot_table_size, 0644);
3202
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/net/sunrpc/xprtsock.c
   4 *
   5 * Client-side transport implementation for sockets.
   6 *
   7 * TCP callback races fixes (C) 1998 Red Hat
   8 * TCP send fixes (C) 1998 Red Hat
   9 * TCP NFS related read + write fixes
  10 *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  11 *
  12 * Rewrite of larges part of the code in order to stabilize TCP stuff.
  13 * Fix behaviour when socket buffer is full.
  14 *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  15 *
  16 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  17 *
  18 * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  19 *   <gilles.quillard@bull.net>
  20 */
  21
  22#include <linux/types.h>
  23#include <linux/string.h>
  24#include <linux/slab.h>
  25#include <linux/module.h>
  26#include <linux/capability.h>
  27#include <linux/pagemap.h>
  28#include <linux/errno.h>
  29#include <linux/socket.h>
  30#include <linux/in.h>
  31#include <linux/net.h>
  32#include <linux/mm.h>
  33#include <linux/un.h>
  34#include <linux/udp.h>
  35#include <linux/tcp.h>
  36#include <linux/sunrpc/clnt.h>
  37#include <linux/sunrpc/addr.h>
  38#include <linux/sunrpc/sched.h>
  39#include <linux/sunrpc/svcsock.h>
  40#include <linux/sunrpc/xprtsock.h>
  41#include <linux/file.h>
  42#ifdef CONFIG_SUNRPC_BACKCHANNEL
  43#include <linux/sunrpc/bc_xprt.h>
  44#endif
  45
  46#include <net/sock.h>
  47#include <net/checksum.h>
  48#include <net/udp.h>
  49#include <net/tcp.h>
  50#include <linux/bvec.h>
  51#include <linux/highmem.h>
  52#include <linux/uio.h>
  53#include <linux/sched/mm.h>
  54
  55#include <trace/events/sunrpc.h>
  56
  57#include "socklib.h"
  58#include "sunrpc.h"
  59
  60static void xs_close(struct rpc_xprt *xprt);
  61static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
  62		struct socket *sock);
  63
  64/*
  65 * xprtsock tunables
  66 */
  67static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  68static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
  69static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
  70
  71static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  72static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  73
 
 
  74#define XS_TCP_LINGER_TO	(15U * HZ)
  75static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  76
  77/*
  78 * We can register our own files under /proc/sys/sunrpc by
  79 * calling register_sysctl_table() again.  The files in that
  80 * directory become the union of all files registered there.
  81 *
  82 * We simply need to make sure that we don't collide with
  83 * someone else's file names!
  84 */
  85
  86static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  87static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  88static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
  89static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  90static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  91
  92static struct ctl_table_header *sunrpc_table_header;
  93
  94static struct xprt_class xs_local_transport;
  95static struct xprt_class xs_udp_transport;
  96static struct xprt_class xs_tcp_transport;
  97static struct xprt_class xs_bc_tcp_transport;
  98
  99/*
 100 * FIXME: changing the UDP slot table size should also resize the UDP
 101 *        socket buffers for existing UDP transports
 102 */
 103static struct ctl_table xs_tunables_table[] = {
 104	{
 105		.procname	= "udp_slot_table_entries",
 106		.data		= &xprt_udp_slot_table_entries,
 107		.maxlen		= sizeof(unsigned int),
 108		.mode		= 0644,
 109		.proc_handler	= proc_dointvec_minmax,
 110		.extra1		= &min_slot_table_size,
 111		.extra2		= &max_slot_table_size
 112	},
 113	{
 114		.procname	= "tcp_slot_table_entries",
 115		.data		= &xprt_tcp_slot_table_entries,
 116		.maxlen		= sizeof(unsigned int),
 117		.mode		= 0644,
 118		.proc_handler	= proc_dointvec_minmax,
 119		.extra1		= &min_slot_table_size,
 120		.extra2		= &max_slot_table_size
 121	},
 122	{
 123		.procname	= "tcp_max_slot_table_entries",
 124		.data		= &xprt_max_tcp_slot_table_entries,
 125		.maxlen		= sizeof(unsigned int),
 126		.mode		= 0644,
 127		.proc_handler	= proc_dointvec_minmax,
 128		.extra1		= &min_slot_table_size,
 129		.extra2		= &max_tcp_slot_table_limit
 130	},
 131	{
 132		.procname	= "min_resvport",
 133		.data		= &xprt_min_resvport,
 134		.maxlen		= sizeof(unsigned int),
 135		.mode		= 0644,
 136		.proc_handler	= proc_dointvec_minmax,
 137		.extra1		= &xprt_min_resvport_limit,
 138		.extra2		= &xprt_max_resvport_limit
 139	},
 140	{
 141		.procname	= "max_resvport",
 142		.data		= &xprt_max_resvport,
 143		.maxlen		= sizeof(unsigned int),
 144		.mode		= 0644,
 145		.proc_handler	= proc_dointvec_minmax,
 146		.extra1		= &xprt_min_resvport_limit,
 147		.extra2		= &xprt_max_resvport_limit
 148	},
 149	{
 150		.procname	= "tcp_fin_timeout",
 151		.data		= &xs_tcp_fin_timeout,
 152		.maxlen		= sizeof(xs_tcp_fin_timeout),
 153		.mode		= 0644,
 154		.proc_handler	= proc_dointvec_jiffies,
 155	},
 156	{ },
 157};
 158
 159static struct ctl_table sunrpc_table[] = {
 160	{
 161		.procname	= "sunrpc",
 162		.mode		= 0555,
 163		.child		= xs_tunables_table
 164	},
 165	{ },
 166};
 167
 
 
 168/*
 169 * Wait duration for a reply from the RPC portmapper.
 170 */
 171#define XS_BIND_TO		(60U * HZ)
 172
 173/*
 174 * Delay if a UDP socket connect error occurs.  This is most likely some
 175 * kind of resource problem on the local host.
 176 */
 177#define XS_UDP_REEST_TO		(2U * HZ)
 178
 179/*
 180 * The reestablish timeout allows clients to delay for a bit before attempting
 181 * to reconnect to a server that just dropped our connection.
 182 *
 183 * We implement an exponential backoff when trying to reestablish a TCP
 184 * transport connection with the server.  Some servers like to drop a TCP
 185 * connection when they are overworked, so we start with a short timeout and
 186 * increase over time if the server is down or not responding.
 187 */
 188#define XS_TCP_INIT_REEST_TO	(3U * HZ)
 
 189
 190/*
 191 * TCP idle timeout; client drops the transport socket if it is idle
 192 * for this long.  Note that we also timeout UDP sockets to prevent
 193 * holding port numbers when there is no RPC traffic.
 194 */
 195#define XS_IDLE_DISC_TO		(5U * 60 * HZ)
 196
 197#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 198# undef  RPC_DEBUG_DATA
 199# define RPCDBG_FACILITY	RPCDBG_TRANS
 200#endif
 201
 202#ifdef RPC_DEBUG_DATA
 203static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 204{
 205	u8 *buf = (u8 *) packet;
 206	int j;
 207
 208	dprintk("RPC:       %s\n", msg);
 209	for (j = 0; j < count && j < 128; j += 4) {
 210		if (!(j & 31)) {
 211			if (j)
 212				dprintk("\n");
 213			dprintk("0x%04x ", j);
 214		}
 215		dprintk("%02x%02x%02x%02x ",
 216			buf[j], buf[j+1], buf[j+2], buf[j+3]);
 217	}
 218	dprintk("\n");
 219}
 220#else
 221static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 222{
 223	/* NOP */
 224}
 225#endif
 226
 227static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 228{
 229	return (struct rpc_xprt *) sk->sk_user_data;
 230}
 231
 232static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 233{
 234	return (struct sockaddr *) &xprt->addr;
 235}
 236
 237static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
 238{
 239	return (struct sockaddr_un *) &xprt->addr;
 240}
 241
 242static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 243{
 244	return (struct sockaddr_in *) &xprt->addr;
 245}
 246
 247static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 248{
 249	return (struct sockaddr_in6 *) &xprt->addr;
 250}
 251
 252static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 253{
 254	struct sockaddr *sap = xs_addr(xprt);
 255	struct sockaddr_in6 *sin6;
 256	struct sockaddr_in *sin;
 257	struct sockaddr_un *sun;
 258	char buf[128];
 259
 260	switch (sap->sa_family) {
 261	case AF_LOCAL:
 262		sun = xs_addr_un(xprt);
 263		strlcpy(buf, sun->sun_path, sizeof(buf));
 264		xprt->address_strings[RPC_DISPLAY_ADDR] =
 265						kstrdup(buf, GFP_KERNEL);
 266		break;
 267	case AF_INET:
 268		(void)rpc_ntop(sap, buf, sizeof(buf));
 269		xprt->address_strings[RPC_DISPLAY_ADDR] =
 270						kstrdup(buf, GFP_KERNEL);
 271		sin = xs_addr_in(xprt);
 272		snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 273		break;
 274	case AF_INET6:
 275		(void)rpc_ntop(sap, buf, sizeof(buf));
 276		xprt->address_strings[RPC_DISPLAY_ADDR] =
 277						kstrdup(buf, GFP_KERNEL);
 278		sin6 = xs_addr_in6(xprt);
 279		snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 280		break;
 281	default:
 282		BUG();
 283	}
 284
 285	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 286}
 287
 288static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 289{
 290	struct sockaddr *sap = xs_addr(xprt);
 291	char buf[128];
 292
 293	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 294	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 295
 296	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 297	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 298}
 299
 300static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 301				     const char *protocol,
 302				     const char *netid)
 303{
 304	xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 305	xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 306	xs_format_common_peer_addresses(xprt);
 307	xs_format_common_peer_ports(xprt);
 308}
 309
 310static void xs_update_peer_port(struct rpc_xprt *xprt)
 311{
 312	kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 313	kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 314
 315	xs_format_common_peer_ports(xprt);
 316}
 317
 318static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 319{
 320	unsigned int i;
 321
 322	for (i = 0; i < RPC_DISPLAY_MAX; i++)
 323		switch (i) {
 324		case RPC_DISPLAY_PROTO:
 325		case RPC_DISPLAY_NETID:
 326			continue;
 327		default:
 328			kfree(xprt->address_strings[i]);
 329		}
 330}
 331
 332static size_t
 333xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
 334{
 335	size_t i,n;
 336
 337	if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
 338		return want;
 339	n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
 340	for (i = 0; i < n; i++) {
 341		if (buf->pages[i])
 342			continue;
 343		buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
 344		if (!buf->pages[i]) {
 345			i *= PAGE_SIZE;
 346			return i > buf->page_base ? i - buf->page_base : 0;
 347		}
 348	}
 349	return want;
 350}
 351
 352static ssize_t
 353xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
 354{
 355	ssize_t ret;
 356	if (seek != 0)
 357		iov_iter_advance(&msg->msg_iter, seek);
 358	ret = sock_recvmsg(sock, msg, flags);
 359	return ret > 0 ? ret + seek : ret;
 360}
 
 
 
 361
 362static ssize_t
 363xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
 364		struct kvec *kvec, size_t count, size_t seek)
 365{
 366	iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
 367	return xs_sock_recvmsg(sock, msg, flags, seek);
 368}
 369
 370static ssize_t
 371xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
 372		struct bio_vec *bvec, unsigned long nr, size_t count,
 373		size_t seek)
 374{
 375	iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
 376	return xs_sock_recvmsg(sock, msg, flags, seek);
 377}
 
 
 378
 379static ssize_t
 380xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
 381		size_t count)
 382{
 383	iov_iter_discard(&msg->msg_iter, READ, count);
 384	return sock_recvmsg(sock, msg, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 385}
 386
 387#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
 388static void
 389xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 390{
 391	struct bvec_iter bi = {
 392		.bi_size = count,
 393	};
 394	struct bio_vec bv;
 395
 396	bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
 397	for_each_bvec(bv, bvec, bi, bi)
 398		flush_dcache_page(bv.bv_page);
 399}
 400#else
 401static inline void
 402xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
 403{
 404}
 405#endif
 406
 407static ssize_t
 408xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 409		struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
 410{
 411	size_t want, seek_init = seek, offset = 0;
 412	ssize_t ret;
 413
 414	want = min_t(size_t, count, buf->head[0].iov_len);
 415	if (seek < want) {
 416		ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
 417		if (ret <= 0)
 418			goto sock_err;
 419		offset += ret;
 420		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 421			goto out;
 422		if (ret != want)
 423			goto out;
 424		seek = 0;
 425	} else {
 426		seek -= want;
 427		offset += want;
 428	}
 429
 430	want = xs_alloc_sparse_pages(buf,
 431			min_t(size_t, count - offset, buf->page_len),
 432			GFP_KERNEL);
 433	if (seek < want) {
 434		ret = xs_read_bvec(sock, msg, flags, buf->bvec,
 435				xdr_buf_pagecount(buf),
 436				want + buf->page_base,
 437				seek + buf->page_base);
 438		if (ret <= 0)
 439			goto sock_err;
 440		xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
 441		ret -= buf->page_base;
 442		offset += ret;
 443		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 444			goto out;
 445		if (ret != want)
 446			goto out;
 447		seek = 0;
 448	} else {
 449		seek -= want;
 450		offset += want;
 451	}
 452
 453	want = min_t(size_t, count - offset, buf->tail[0].iov_len);
 454	if (seek < want) {
 455		ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
 456		if (ret <= 0)
 457			goto sock_err;
 458		offset += ret;
 459		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 460			goto out;
 461		if (ret != want)
 462			goto out;
 463	} else if (offset < seek_init)
 464		offset = seek_init;
 465	ret = -EMSGSIZE;
 466out:
 467	*read = offset - seek_init;
 468	return ret;
 469sock_err:
 470	offset += seek;
 471	goto out;
 472}
 473
 474static void
 475xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
 476{
 477	if (!transport->recv.copied) {
 478		if (buf->head[0].iov_len >= transport->recv.offset)
 479			memcpy(buf->head[0].iov_base,
 480					&transport->recv.xid,
 481					transport->recv.offset);
 482		transport->recv.copied = transport->recv.offset;
 483	}
 484}
 485
 486static bool
 487xs_read_stream_request_done(struct sock_xprt *transport)
 488{
 489	return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
 490}
 491
 492static void
 493xs_read_stream_check_eor(struct sock_xprt *transport,
 494		struct msghdr *msg)
 495{
 496	if (xs_read_stream_request_done(transport))
 497		msg->msg_flags |= MSG_EOR;
 498}
 499
 500static ssize_t
 501xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 502		int flags, struct rpc_rqst *req)
 503{
 504	struct xdr_buf *buf = &req->rq_private_buf;
 505	size_t want, read;
 506	ssize_t ret;
 507
 508	xs_read_header(transport, buf);
 509
 510	want = transport->recv.len - transport->recv.offset;
 511	if (want != 0) {
 512		ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
 513				transport->recv.copied + want,
 514				transport->recv.copied,
 515				&read);
 516		transport->recv.offset += read;
 517		transport->recv.copied += read;
 518	}
 519
 520	if (transport->recv.offset == transport->recv.len)
 521		xs_read_stream_check_eor(transport, msg);
 522
 523	if (want == 0)
 524		return 0;
 525
 526	switch (ret) {
 527	default:
 528		break;
 529	case -EFAULT:
 530	case -EMSGSIZE:
 531		msg->msg_flags |= MSG_TRUNC;
 532		return read;
 533	case 0:
 534		return -ESHUTDOWN;
 535	}
 536	return ret < 0 ? ret : read;
 537}
 538
 539static size_t
 540xs_read_stream_headersize(bool isfrag)
 541{
 542	if (isfrag)
 543		return sizeof(__be32);
 544	return 3 * sizeof(__be32);
 545}
 546
 547static ssize_t
 548xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
 549		int flags, size_t want, size_t seek)
 550{
 551	struct kvec kvec = {
 552		.iov_base = &transport->recv.fraghdr,
 553		.iov_len = want,
 554	};
 555	return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
 556}
 557
 558#if defined(CONFIG_SUNRPC_BACKCHANNEL)
 559static ssize_t
 560xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 561{
 562	struct rpc_xprt *xprt = &transport->xprt;
 563	struct rpc_rqst *req;
 564	ssize_t ret;
 565
 566	/* Is this transport associated with the backchannel? */
 567	if (!xprt->bc_serv)
 568		return -ESHUTDOWN;
 569
 570	/* Look up and lock the request corresponding to the given XID */
 571	req = xprt_lookup_bc_request(xprt, transport->recv.xid);
 572	if (!req) {
 573		printk(KERN_WARNING "Callback slot table overflowed\n");
 574		return -ESHUTDOWN;
 575	}
 576	if (transport->recv.copied && !req->rq_private_buf.len)
 577		return -ESHUTDOWN;
 578
 579	ret = xs_read_stream_request(transport, msg, flags, req);
 580	if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 581		xprt_complete_bc_request(req, transport->recv.copied);
 582	else
 583		req->rq_private_buf.len = transport->recv.copied;
 584
 585	return ret;
 586}
 587#else /* CONFIG_SUNRPC_BACKCHANNEL */
 588static ssize_t
 589xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 590{
 591	return -ESHUTDOWN;
 592}
 593#endif /* CONFIG_SUNRPC_BACKCHANNEL */
 594
 595static ssize_t
 596xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 597{
 598	struct rpc_xprt *xprt = &transport->xprt;
 599	struct rpc_rqst *req;
 600	ssize_t ret = 0;
 601
 602	/* Look up and lock the request corresponding to the given XID */
 603	spin_lock(&xprt->queue_lock);
 604	req = xprt_lookup_rqst(xprt, transport->recv.xid);
 605	if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
 606		msg->msg_flags |= MSG_TRUNC;
 607		goto out;
 608	}
 609	xprt_pin_rqst(req);
 610	spin_unlock(&xprt->queue_lock);
 611
 612	ret = xs_read_stream_request(transport, msg, flags, req);
 613
 614	spin_lock(&xprt->queue_lock);
 615	if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 616		xprt_complete_rqst(req->rq_task, transport->recv.copied);
 617	else
 618		req->rq_private_buf.len = transport->recv.copied;
 619	xprt_unpin_rqst(req);
 620out:
 621	spin_unlock(&xprt->queue_lock);
 622	return ret;
 623}
 624
 625static ssize_t
 626xs_read_stream(struct sock_xprt *transport, int flags)
 627{
 628	struct msghdr msg = { 0 };
 629	size_t want, read = 0;
 630	ssize_t ret = 0;
 631
 632	if (transport->recv.len == 0) {
 633		want = xs_read_stream_headersize(transport->recv.copied != 0);
 634		ret = xs_read_stream_header(transport, &msg, flags, want,
 635				transport->recv.offset);
 636		if (ret <= 0)
 637			goto out_err;
 638		transport->recv.offset = ret;
 639		if (transport->recv.offset != want)
 640			return transport->recv.offset;
 641		transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 642			RPC_FRAGMENT_SIZE_MASK;
 643		transport->recv.offset -= sizeof(transport->recv.fraghdr);
 644		read = ret;
 645	}
 646
 647	switch (be32_to_cpu(transport->recv.calldir)) {
 648	default:
 649		msg.msg_flags |= MSG_TRUNC;
 650		break;
 651	case RPC_CALL:
 652		ret = xs_read_stream_call(transport, &msg, flags);
 653		break;
 654	case RPC_REPLY:
 655		ret = xs_read_stream_reply(transport, &msg, flags);
 656	}
 657	if (msg.msg_flags & MSG_TRUNC) {
 658		transport->recv.calldir = cpu_to_be32(-1);
 659		transport->recv.copied = -1;
 660	}
 661	if (ret < 0)
 662		goto out_err;
 663	read += ret;
 664	if (transport->recv.offset < transport->recv.len) {
 665		if (!(msg.msg_flags & MSG_TRUNC))
 666			return read;
 667		msg.msg_flags = 0;
 668		ret = xs_read_discard(transport->sock, &msg, flags,
 669				transport->recv.len - transport->recv.offset);
 670		if (ret <= 0)
 671			goto out_err;
 672		transport->recv.offset += ret;
 673		read += ret;
 674		if (transport->recv.offset != transport->recv.len)
 675			return read;
 676	}
 677	if (xs_read_stream_request_done(transport)) {
 678		trace_xs_stream_read_request(transport);
 679		transport->recv.copied = 0;
 680	}
 681	transport->recv.offset = 0;
 682	transport->recv.len = 0;
 683	return read;
 684out_err:
 685	return ret != 0 ? ret : -ESHUTDOWN;
 686}
 687
 688static __poll_t xs_poll_socket(struct sock_xprt *transport)
 689{
 690	return transport->sock->ops->poll(transport->file, transport->sock,
 691			NULL);
 692}
 693
 694static bool xs_poll_socket_readable(struct sock_xprt *transport)
 695{
 696	__poll_t events = xs_poll_socket(transport);
 697
 698	return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
 699}
 700
 701static void xs_poll_check_readable(struct sock_xprt *transport)
 702{
 703
 704	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 705	if (!xs_poll_socket_readable(transport))
 706		return;
 707	if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
 708		queue_work(xprtiod_workqueue, &transport->recv_worker);
 709}
 710
 711static void xs_stream_data_receive(struct sock_xprt *transport)
 712{
 713	size_t read = 0;
 714	ssize_t ret = 0;
 715
 716	mutex_lock(&transport->recv_mutex);
 717	if (transport->sock == NULL)
 718		goto out;
 719	for (;;) {
 720		ret = xs_read_stream(transport, MSG_DONTWAIT);
 721		if (ret < 0)
 722			break;
 723		read += ret;
 724		cond_resched();
 725	}
 726	if (ret == -ESHUTDOWN)
 727		kernel_sock_shutdown(transport->sock, SHUT_RDWR);
 728	else
 729		xs_poll_check_readable(transport);
 730out:
 731	mutex_unlock(&transport->recv_mutex);
 732	trace_xs_stream_read_data(&transport->xprt, ret, read);
 733}
 734
 735static void xs_stream_data_receive_workfn(struct work_struct *work)
 736{
 737	struct sock_xprt *transport =
 738		container_of(work, struct sock_xprt, recv_worker);
 739	unsigned int pflags = memalloc_nofs_save();
 740
 741	xs_stream_data_receive(transport);
 742	memalloc_nofs_restore(pflags);
 743}
 744
 745static void
 746xs_stream_reset_connect(struct sock_xprt *transport)
 747{
 748	transport->recv.offset = 0;
 749	transport->recv.len = 0;
 750	transport->recv.copied = 0;
 751	transport->xmit.offset = 0;
 752}
 753
 754static void
 755xs_stream_start_connect(struct sock_xprt *transport)
 756{
 757	transport->xprt.stat.connect_count++;
 758	transport->xprt.stat.connect_start = jiffies;
 759}
 760
 761#define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
 762
 763/**
 764 * xs_nospace - handle transmit was incomplete
 765 * @req: pointer to RPC request
 766 *
 767 */
 768static int xs_nospace(struct rpc_rqst *req)
 769{
 
 770	struct rpc_xprt *xprt = req->rq_xprt;
 771	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 772	struct sock *sk = transport->inet;
 773	int ret = -EAGAIN;
 774
 775	trace_rpc_socket_nospace(req, transport);
 
 
 776
 777	/* Protect against races with write_space */
 778	spin_lock(&xprt->transport_lock);
 779
 780	/* Don't race with disconnect */
 781	if (xprt_connected(xprt)) {
 782		/* wait for more buffer space */
 783		sk->sk_write_pending++;
 784		xprt_wait_for_buffer_space(xprt);
 785	} else
 786		ret = -ENOTCONN;
 787
 788	spin_unlock(&xprt->transport_lock);
 789
 790	/* Race breaker in case memory is freed before above code is called */
 791	if (ret == -EAGAIN) {
 792		struct socket_wq *wq;
 793
 794		rcu_read_lock();
 795		wq = rcu_dereference(sk->sk_wq);
 796		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
 797		rcu_read_unlock();
 798
 799		sk->sk_write_space(sk);
 800	}
 801	return ret;
 802}
 803
 804static void
 805xs_stream_prepare_request(struct rpc_rqst *req)
 806{
 807	xdr_free_bvec(&req->rq_rcv_buf);
 808	req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
 809}
 810
 811/*
 812 * Determine if the previous message in the stream was aborted before it
 813 * could complete transmission.
 814 */
 815static bool
 816xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
 817{
 818	return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
 819}
 820
 821/*
 822 * Return the stream record marker field for a record of length < 2^31-1
 823 */
 824static rpc_fraghdr
 825xs_stream_record_marker(struct xdr_buf *xdr)
 826{
 827	if (!xdr->len)
 828		return 0;
 829	return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
 830}
 831
 832/**
 833 * xs_local_send_request - write an RPC request to an AF_LOCAL socket
 834 * @req: pointer to RPC request
 835 *
 836 * Return values:
 837 *        0:	The request has been sent
 838 *   EAGAIN:	The socket was blocked, please call again later to
 839 *		complete the request
 840 * ENOTCONN:	Caller needs to invoke connect logic then call again
 841 *    other:	Some other error occurred, the request was not sent
 842 */
 843static int xs_local_send_request(struct rpc_rqst *req)
 844{
 
 845	struct rpc_xprt *xprt = req->rq_xprt;
 846	struct sock_xprt *transport =
 847				container_of(xprt, struct sock_xprt, xprt);
 848	struct xdr_buf *xdr = &req->rq_snd_buf;
 849	rpc_fraghdr rm = xs_stream_record_marker(xdr);
 850	unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
 851	struct msghdr msg = {
 852		.msg_flags	= XS_SENDMSG_FLAGS,
 853	};
 854	unsigned int sent;
 855	int status;
 
 856
 857	/* Close the stream if the previous transmission was incomplete */
 858	if (xs_send_request_was_aborted(transport, req)) {
 859		xs_close(xprt);
 860		return -ENOTCONN;
 861	}
 862
 863	xs_pktdump("packet data:",
 864			req->rq_svec->iov_base, req->rq_svec->iov_len);
 865
 866	req->rq_xtime = ktime_get();
 867	status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 868				   transport->xmit.offset, rm, &sent);
 869	dprintk("RPC:       %s(%u) = %d\n",
 870			__func__, xdr->len - transport->xmit.offset, status);
 871
 872	if (status == -EAGAIN && sock_writeable(transport->inet))
 873		status = -ENOBUFS;
 874
 875	if (likely(sent > 0) || status == 0) {
 876		transport->xmit.offset += sent;
 877		req->rq_bytes_sent = transport->xmit.offset;
 878		if (likely(req->rq_bytes_sent >= msglen)) {
 879			req->rq_xmit_bytes_sent += transport->xmit.offset;
 880			transport->xmit.offset = 0;
 881			return 0;
 882		}
 883		status = -EAGAIN;
 884	}
 885
 886	switch (status) {
 887	case -ENOBUFS:
 888		break;
 889	case -EAGAIN:
 890		status = xs_nospace(req);
 891		break;
 892	default:
 893		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 894			-status);
 895		fallthrough;
 896	case -EPIPE:
 897		xs_close(xprt);
 898		status = -ENOTCONN;
 899	}
 900
 901	return status;
 902}
 903
 904/**
 905 * xs_udp_send_request - write an RPC request to a UDP socket
 906 * @req: pointer to RPC request
 907 *
 908 * Return values:
 909 *        0:	The request has been sent
 910 *   EAGAIN:	The socket was blocked, please call again later to
 911 *		complete the request
 912 * ENOTCONN:	Caller needs to invoke connect logic then call again
 913 *    other:	Some other error occurred, the request was not sent
 914 */
 915static int xs_udp_send_request(struct rpc_rqst *req)
 916{
 
 917	struct rpc_xprt *xprt = req->rq_xprt;
 918	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 919	struct xdr_buf *xdr = &req->rq_snd_buf;
 920	struct msghdr msg = {
 921		.msg_name	= xs_addr(xprt),
 922		.msg_namelen	= xprt->addrlen,
 923		.msg_flags	= XS_SENDMSG_FLAGS,
 924	};
 925	unsigned int sent;
 926	int status;
 927
 928	xs_pktdump("packet data:",
 929				req->rq_svec->iov_base,
 930				req->rq_svec->iov_len);
 931
 932	if (!xprt_bound(xprt))
 933		return -ENOTCONN;
 934
 935	if (!xprt_request_get_cong(xprt, req))
 936		return -EBADSLT;
 937
 938	req->rq_xtime = ktime_get();
 939	status = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, 0, &sent);
 940
 941	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 942			xdr->len, status);
 943
 944	/* firewall is blocking us, don't return -EAGAIN or we end up looping */
 945	if (status == -EPERM)
 946		goto process_status;
 947
 948	if (status == -EAGAIN && sock_writeable(transport->inet))
 949		status = -ENOBUFS;
 950
 951	if (sent > 0 || status == 0) {
 952		req->rq_xmit_bytes_sent += sent;
 953		if (sent >= req->rq_slen)
 954			return 0;
 955		/* Still some bytes left; set up for a retry later. */
 956		status = -EAGAIN;
 957	}
 958
 959process_status:
 960	switch (status) {
 961	case -ENOTSOCK:
 962		status = -ENOTCONN;
 963		/* Should we call xs_close() here? */
 964		break;
 965	case -EAGAIN:
 966		status = xs_nospace(req);
 967		break;
 968	case -ENETUNREACH:
 969	case -ENOBUFS:
 970	case -EPIPE:
 971	case -ECONNREFUSED:
 972	case -EPERM:
 973		/* When the server has died, an ICMP port unreachable message
 974		 * prompts ECONNREFUSED. */
 975		break;
 976	default:
 977		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 978			-status);
 979	}
 980
 981	return status;
 982}
 983
 984/**
 985 * xs_tcp_send_request - write an RPC request to a TCP socket
 986 * @req: pointer to RPC request
 987 *
 988 * Return values:
 989 *        0:	The request has been sent
 990 *   EAGAIN:	The socket was blocked, please call again later to
 991 *		complete the request
 992 * ENOTCONN:	Caller needs to invoke connect logic then call again
 993 *    other:	Some other error occurred, the request was not sent
 994 *
 995 * XXX: In the case of soft timeouts, should we eventually give up
 996 *	if sendmsg is not able to make progress?
 997 */
 998static int xs_tcp_send_request(struct rpc_rqst *req)
 999{
 
1000	struct rpc_xprt *xprt = req->rq_xprt;
1001	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1002	struct xdr_buf *xdr = &req->rq_snd_buf;
1003	rpc_fraghdr rm = xs_stream_record_marker(xdr);
1004	unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1005	struct msghdr msg = {
1006		.msg_flags	= XS_SENDMSG_FLAGS,
1007	};
1008	bool vm_wait = false;
1009	unsigned int sent;
1010	int status;
 
1011
1012	/* Close the stream if the previous transmission was incomplete */
1013	if (xs_send_request_was_aborted(transport, req)) {
1014		if (transport->sock != NULL)
1015			kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1016		return -ENOTCONN;
1017	}
1018	if (!transport->inet)
1019		return -ENOTCONN;
1020
1021	xs_pktdump("packet data:",
1022				req->rq_svec->iov_base,
1023				req->rq_svec->iov_len);
1024
1025	if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1026		xs_tcp_set_socket_timeouts(xprt, transport->sock);
 
 
 
1027
1028	/* Continue transmitting the packet/record. We must be careful
1029	 * to cope with writespace callbacks arriving _after_ we have
1030	 * called sendmsg(). */
1031	req->rq_xtime = ktime_get();
1032	tcp_sock_set_cork(transport->inet, true);
1033	while (1) {
1034		status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
1035					   transport->xmit.offset, rm, &sent);
 
1036
1037		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1038				xdr->len - transport->xmit.offset, status);
1039
1040		/* If we've sent the entire packet, immediately
1041		 * reset the count of bytes sent. */
1042		transport->xmit.offset += sent;
1043		req->rq_bytes_sent = transport->xmit.offset;
1044		if (likely(req->rq_bytes_sent >= msglen)) {
1045			req->rq_xmit_bytes_sent += transport->xmit.offset;
1046			transport->xmit.offset = 0;
1047			if (atomic_long_read(&xprt->xmit_queuelen) == 1)
1048				tcp_sock_set_cork(transport->inet, false);
1049			return 0;
1050		}
1051
1052		WARN_ON_ONCE(sent == 0 && status == 0);
1053
1054		if (status == -EAGAIN ) {
1055			/*
1056			 * Return EAGAIN if we're sure we're hitting the
1057			 * socket send buffer limits.
1058			 */
1059			if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1060				break;
1061			/*
1062			 * Did we hit a memory allocation failure?
1063			 */
1064			if (sent == 0) {
1065				status = -ENOBUFS;
1066				if (vm_wait)
1067					break;
1068				/* Retry, knowing now that we're below the
1069				 * socket send buffer limit
1070				 */
1071				vm_wait = true;
1072			}
1073			continue;
1074		}
1075		if (status < 0)
1076			break;
1077		vm_wait = false;
 
 
 
1078	}
 
 
1079
1080	switch (status) {
1081	case -ENOTSOCK:
1082		status = -ENOTCONN;
1083		/* Should we call xs_close() here? */
1084		break;
1085	case -EAGAIN:
1086		status = xs_nospace(req);
1087		break;
1088	case -ECONNRESET:
1089	case -ECONNREFUSED:
1090	case -ENOTCONN:
1091	case -EADDRINUSE:
1092	case -ENOBUFS:
1093	case -EPIPE:
1094		break;
1095	default:
1096		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1097			-status);
1098	}
1099
1100	return status;
1101}
1102
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1103static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1104{
1105	transport->old_data_ready = sk->sk_data_ready;
1106	transport->old_state_change = sk->sk_state_change;
1107	transport->old_write_space = sk->sk_write_space;
1108	transport->old_error_report = sk->sk_error_report;
1109}
1110
1111static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1112{
1113	sk->sk_data_ready = transport->old_data_ready;
1114	sk->sk_state_change = transport->old_state_change;
1115	sk->sk_write_space = transport->old_write_space;
1116	sk->sk_error_report = transport->old_error_report;
1117}
1118
1119static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1120{
1121	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1122
1123	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1124	clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1125	clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1126	clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1127}
1128
1129static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1130{
1131	set_bit(nr, &transport->sock_state);
1132	queue_work(xprtiod_workqueue, &transport->error_worker);
1133}
1134
1135static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1136{
1137	smp_mb__before_atomic();
1138	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1139	clear_bit(XPRT_CLOSING, &xprt->state);
1140	xs_sock_reset_state_flags(xprt);
1141	smp_mb__after_atomic();
1142}
1143
 
 
 
 
 
 
 
1144/**
1145 * xs_error_report - callback to handle TCP socket state errors
1146 * @sk: socket
1147 *
1148 * Note: we don't call sock_error() since there may be a rpc_task
1149 * using the socket, and so we don't want to clear sk->sk_err.
1150 */
1151static void xs_error_report(struct sock *sk)
1152{
1153	struct sock_xprt *transport;
1154	struct rpc_xprt *xprt;
 
1155
1156	read_lock_bh(&sk->sk_callback_lock);
1157	if (!(xprt = xprt_from_sock(sk)))
1158		goto out;
1159
1160	transport = container_of(xprt, struct sock_xprt, xprt);
1161	transport->xprt_err = -sk->sk_err;
1162	if (transport->xprt_err == 0)
1163		goto out;
 
 
 
1164	dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1165			xprt, -transport->xprt_err);
1166	trace_rpc_socket_error(xprt, sk->sk_socket, transport->xprt_err);
1167
1168	/* barrier ensures xprt_err is set before XPRT_SOCK_WAKE_ERROR */
1169	smp_mb__before_atomic();
1170	xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1171 out:
1172	read_unlock_bh(&sk->sk_callback_lock);
1173}
1174
1175static void xs_reset_transport(struct sock_xprt *transport)
1176{
1177	struct socket *sock = transport->sock;
1178	struct sock *sk = transport->inet;
1179	struct rpc_xprt *xprt = &transport->xprt;
1180	struct file *filp = transport->file;
1181
1182	if (sk == NULL)
1183		return;
1184
1185	if (atomic_read(&transport->xprt.swapper))
1186		sk_clear_memalloc(sk);
1187
1188	kernel_sock_shutdown(sock, SHUT_RDWR);
1189
1190	mutex_lock(&transport->recv_mutex);
1191	write_lock_bh(&sk->sk_callback_lock);
1192	transport->inet = NULL;
1193	transport->sock = NULL;
1194	transport->file = NULL;
1195
1196	sk->sk_user_data = NULL;
1197
1198	xs_restore_old_callbacks(transport, sk);
1199	xprt_clear_connected(xprt);
1200	write_unlock_bh(&sk->sk_callback_lock);
1201	xs_sock_reset_connection_flags(xprt);
1202	/* Reset stream record info */
1203	xs_stream_reset_connect(transport);
1204	mutex_unlock(&transport->recv_mutex);
1205
1206	trace_rpc_socket_close(xprt, sock);
1207	fput(filp);
1208
1209	xprt_disconnect_done(xprt);
1210}
1211
1212/**
1213 * xs_close - close a socket
1214 * @xprt: transport
1215 *
1216 * This is used when all requests are complete; ie, no DRC state remains
1217 * on the server we want to save.
1218 *
1219 * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1220 * xs_reset_transport() zeroing the socket from underneath a writer.
1221 */
1222static void xs_close(struct rpc_xprt *xprt)
1223{
1224	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1225
1226	dprintk("RPC:       xs_close xprt %p\n", xprt);
1227
1228	xs_reset_transport(transport);
1229	xprt->reestablish_timeout = 0;
 
 
1230}
1231
1232static void xs_inject_disconnect(struct rpc_xprt *xprt)
1233{
1234	dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1235		xprt);
1236	xprt_disconnect_done(xprt);
1237}
1238
1239static void xs_xprt_free(struct rpc_xprt *xprt)
1240{
1241	xs_free_peer_addresses(xprt);
1242	xprt_free(xprt);
1243}
1244
1245/**
1246 * xs_destroy - prepare to shutdown a transport
1247 * @xprt: doomed transport
1248 *
1249 */
1250static void xs_destroy(struct rpc_xprt *xprt)
1251{
1252	struct sock_xprt *transport = container_of(xprt,
1253			struct sock_xprt, xprt);
1254	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1255
1256	cancel_delayed_work_sync(&transport->connect_worker);
1257	xs_close(xprt);
1258	cancel_work_sync(&transport->recv_worker);
1259	cancel_work_sync(&transport->error_worker);
1260	xs_xprt_free(xprt);
1261	module_put(THIS_MODULE);
1262}
1263
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1264/**
1265 * xs_udp_data_read_skb - receive callback for UDP sockets
1266 * @xprt: transport
1267 * @sk: socket
1268 * @skb: skbuff
1269 *
1270 */
1271static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1272		struct sock *sk,
1273		struct sk_buff *skb)
1274{
1275	struct rpc_task *task;
1276	struct rpc_rqst *rovr;
1277	int repsize, copied;
1278	u32 _xid;
1279	__be32 *xp;
1280
1281	repsize = skb->len;
1282	if (repsize < 4) {
1283		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1284		return;
1285	}
1286
1287	/* Copy the XID from the skb... */
1288	xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
 
1289	if (xp == NULL)
1290		return;
1291
1292	/* Look up and lock the request corresponding to the given XID */
1293	spin_lock(&xprt->queue_lock);
1294	rovr = xprt_lookup_rqst(xprt, *xp);
1295	if (!rovr)
1296		goto out_unlock;
1297	xprt_pin_rqst(rovr);
1298	xprt_update_rtt(rovr->rq_task);
1299	spin_unlock(&xprt->queue_lock);
1300	task = rovr->rq_task;
1301
1302	if ((copied = rovr->rq_private_buf.buflen) > repsize)
1303		copied = repsize;
1304
1305	/* Suck it into the iovec, verify checksum if not done by hw. */
1306	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1307		spin_lock(&xprt->queue_lock);
1308		__UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1309		goto out_unpin;
1310	}
1311
 
1312
1313	spin_lock(&xprt->transport_lock);
1314	xprt_adjust_cwnd(xprt, task, copied);
1315	spin_unlock(&xprt->transport_lock);
1316	spin_lock(&xprt->queue_lock);
1317	xprt_complete_rqst(task, copied);
1318	__UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1319out_unpin:
1320	xprt_unpin_rqst(rovr);
1321 out_unlock:
1322	spin_unlock(&xprt->queue_lock);
1323}
1324
1325static void xs_udp_data_receive(struct sock_xprt *transport)
1326{
1327	struct sk_buff *skb;
1328	struct sock *sk;
1329	int err;
1330
1331	mutex_lock(&transport->recv_mutex);
1332	sk = transport->inet;
1333	if (sk == NULL)
1334		goto out;
1335	for (;;) {
1336		skb = skb_recv_udp(sk, 0, 1, &err);
1337		if (skb == NULL)
1338			break;
1339		xs_udp_data_read_skb(&transport->xprt, sk, skb);
1340		consume_skb(skb);
1341		cond_resched();
1342	}
1343	xs_poll_check_readable(transport);
1344out:
1345	mutex_unlock(&transport->recv_mutex);
1346}
1347
1348static void xs_udp_data_receive_workfn(struct work_struct *work)
1349{
1350	struct sock_xprt *transport =
1351		container_of(work, struct sock_xprt, recv_worker);
1352	unsigned int pflags = memalloc_nofs_save();
1353
1354	xs_udp_data_receive(transport);
1355	memalloc_nofs_restore(pflags);
1356}
1357
1358/**
1359 * xs_data_ready - "data ready" callback for UDP sockets
1360 * @sk: socket with data to read
1361 *
1362 */
1363static void xs_data_ready(struct sock *sk)
1364{
1365	struct rpc_xprt *xprt;
1366
1367	read_lock_bh(&sk->sk_callback_lock);
1368	dprintk("RPC:       xs_data_ready...\n");
1369	xprt = xprt_from_sock(sk);
1370	if (xprt != NULL) {
1371		struct sock_xprt *transport = container_of(xprt,
1372				struct sock_xprt, xprt);
1373		transport->old_data_ready(sk);
1374		/* Any data means we had a useful conversation, so
1375		 * then we don't need to delay the next reconnect
1376		 */
1377		if (xprt->reestablish_timeout)
1378			xprt->reestablish_timeout = 0;
1379		if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1380			queue_work(xprtiod_workqueue, &transport->recv_worker);
1381	}
1382	read_unlock_bh(&sk->sk_callback_lock);
1383}
1384
1385/*
1386 * Helper function to force a TCP close if the server is sending
1387 * junk and/or it has put us in CLOSE_WAIT
1388 */
1389static void xs_tcp_force_close(struct rpc_xprt *xprt)
1390{
1391	xprt_force_disconnect(xprt);
1392}
1393
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1394#if defined(CONFIG_SUNRPC_BACKCHANNEL)
1395static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
 
 
 
 
 
 
 
 
1396{
1397	return PAGE_SIZE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1398}
1399#endif /* CONFIG_SUNRPC_BACKCHANNEL */
1400
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1401/**
1402 * xs_tcp_state_change - callback to handle TCP socket state changes
1403 * @sk: socket whose state has changed
1404 *
1405 */
1406static void xs_tcp_state_change(struct sock *sk)
1407{
1408	struct rpc_xprt *xprt;
1409	struct sock_xprt *transport;
1410
1411	read_lock_bh(&sk->sk_callback_lock);
1412	if (!(xprt = xprt_from_sock(sk)))
1413		goto out;
1414	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1415	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1416			sk->sk_state, xprt_connected(xprt),
1417			sock_flag(sk, SOCK_DEAD),
1418			sock_flag(sk, SOCK_ZAPPED),
1419			sk->sk_shutdown);
1420
1421	transport = container_of(xprt, struct sock_xprt, xprt);
1422	trace_rpc_socket_state_change(xprt, sk->sk_socket);
1423	switch (sk->sk_state) {
1424	case TCP_ESTABLISHED:
 
1425		if (!xprt_test_and_set_connected(xprt)) {
 
 
 
 
 
 
 
1426			xprt->connect_cookie++;
1427			clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1428			xprt_clear_connecting(xprt);
1429
1430			xprt->stat.connect_count++;
1431			xprt->stat.connect_time += (long)jiffies -
1432						   xprt->stat.connect_start;
1433			xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1434		}
 
1435		break;
1436	case TCP_FIN_WAIT1:
1437		/* The client initiated a shutdown of the socket */
1438		xprt->connect_cookie++;
1439		xprt->reestablish_timeout = 0;
1440		set_bit(XPRT_CLOSING, &xprt->state);
1441		smp_mb__before_atomic();
1442		clear_bit(XPRT_CONNECTED, &xprt->state);
1443		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1444		smp_mb__after_atomic();
1445		break;
1446	case TCP_CLOSE_WAIT:
1447		/* The server initiated a shutdown of the socket */
1448		xprt->connect_cookie++;
1449		clear_bit(XPRT_CONNECTED, &xprt->state);
1450		xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1451		fallthrough;
1452	case TCP_CLOSING:
1453		/*
1454		 * If the server closed down the connection, make sure that
1455		 * we back off before reconnecting
1456		 */
1457		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1458			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1459		break;
1460	case TCP_LAST_ACK:
1461		set_bit(XPRT_CLOSING, &xprt->state);
1462		smp_mb__before_atomic();
1463		clear_bit(XPRT_CONNECTED, &xprt->state);
1464		smp_mb__after_atomic();
1465		break;
1466	case TCP_CLOSE:
1467		if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1468					&transport->sock_state))
1469			xprt_clear_connecting(xprt);
1470		clear_bit(XPRT_CLOSING, &xprt->state);
1471		/* Trigger the socket release */
1472		xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1473	}
1474 out:
1475	read_unlock_bh(&sk->sk_callback_lock);
1476}
1477
1478static void xs_write_space(struct sock *sk)
1479{
1480	struct socket_wq *wq;
1481	struct sock_xprt *transport;
1482	struct rpc_xprt *xprt;
1483
1484	if (!sk->sk_socket)
1485		return;
1486	clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1487
1488	if (unlikely(!(xprt = xprt_from_sock(sk))))
1489		return;
1490	transport = container_of(xprt, struct sock_xprt, xprt);
1491	rcu_read_lock();
1492	wq = rcu_dereference(sk->sk_wq);
1493	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1494		goto out;
1495
1496	xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1497	sk->sk_write_pending--;
1498out:
1499	rcu_read_unlock();
1500}
1501
1502/**
1503 * xs_udp_write_space - callback invoked when socket buffer space
1504 *                             becomes available
1505 * @sk: socket whose state has changed
1506 *
1507 * Called when more output buffer space is available for this socket.
1508 * We try not to wake our writers until they can make "significant"
1509 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1510 * with a bunch of small requests.
1511 */
1512static void xs_udp_write_space(struct sock *sk)
1513{
1514	read_lock_bh(&sk->sk_callback_lock);
1515
1516	/* from net/core/sock.c:sock_def_write_space */
1517	if (sock_writeable(sk))
1518		xs_write_space(sk);
1519
1520	read_unlock_bh(&sk->sk_callback_lock);
1521}
1522
1523/**
1524 * xs_tcp_write_space - callback invoked when socket buffer space
1525 *                             becomes available
1526 * @sk: socket whose state has changed
1527 *
1528 * Called when more output buffer space is available for this socket.
1529 * We try not to wake our writers until they can make "significant"
1530 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1531 * with a bunch of small requests.
1532 */
1533static void xs_tcp_write_space(struct sock *sk)
1534{
1535	read_lock_bh(&sk->sk_callback_lock);
1536
1537	/* from net/core/stream.c:sk_stream_write_space */
1538	if (sk_stream_is_writeable(sk))
1539		xs_write_space(sk);
1540
1541	read_unlock_bh(&sk->sk_callback_lock);
1542}
1543
1544static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1545{
1546	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1547	struct sock *sk = transport->inet;
1548
1549	if (transport->rcvsize) {
1550		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1551		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1552	}
1553	if (transport->sndsize) {
1554		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1555		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1556		sk->sk_write_space(sk);
1557	}
1558}
1559
1560/**
1561 * xs_udp_set_buffer_size - set send and receive limits
1562 * @xprt: generic transport
1563 * @sndsize: requested size of send buffer, in bytes
1564 * @rcvsize: requested size of receive buffer, in bytes
1565 *
1566 * Set socket send and receive buffer size limits.
1567 */
1568static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1569{
1570	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1571
1572	transport->sndsize = 0;
1573	if (sndsize)
1574		transport->sndsize = sndsize + 1024;
1575	transport->rcvsize = 0;
1576	if (rcvsize)
1577		transport->rcvsize = rcvsize + 1024;
1578
1579	xs_udp_do_set_buffer_size(xprt);
1580}
1581
1582/**
1583 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1584 * @xprt: controlling transport
1585 * @task: task that timed out
1586 *
1587 * Adjust the congestion window after a retransmit timeout has occurred.
1588 */
1589static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1590{
1591	spin_lock(&xprt->transport_lock);
1592	xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1593	spin_unlock(&xprt->transport_lock);
1594}
1595
1596static int xs_get_random_port(void)
1597{
1598	unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1599	unsigned short range;
1600	unsigned short rand;
 
1601
1602	if (max < min)
1603		return -EADDRINUSE;
1604	range = max - min + 1;
1605	rand = (unsigned short) prandom_u32() % range;
1606	return rand + min;
 
 
 
 
 
 
 
 
1607}
1608
1609static unsigned short xs_sock_getport(struct socket *sock)
1610{
1611	struct sockaddr_storage buf;
 
1612	unsigned short port = 0;
1613
1614	if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1615		goto out;
1616	switch (buf.ss_family) {
1617	case AF_INET6:
1618		port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1619		break;
1620	case AF_INET:
1621		port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1622	}
1623out:
1624	return port;
1625}
1626
1627/**
1628 * xs_set_port - reset the port number in the remote endpoint address
1629 * @xprt: generic transport
1630 * @port: new port number
1631 *
1632 */
1633static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1634{
1635	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1636
1637	rpc_set_port(xs_addr(xprt), port);
1638	xs_update_peer_port(xprt);
1639}
1640
1641static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1642{
1643	if (transport->srcport == 0 && transport->xprt.reuseport)
1644		transport->srcport = xs_sock_getport(sock);
1645}
1646
1647static int xs_get_srcport(struct sock_xprt *transport)
1648{
1649	int port = transport->srcport;
1650
1651	if (port == 0 && transport->xprt.resvport)
1652		port = xs_get_random_port();
1653	return port;
1654}
1655
1656unsigned short get_srcport(struct rpc_xprt *xprt)
1657{
1658	struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
1659	return xs_sock_getport(sock->sock);
1660}
1661EXPORT_SYMBOL(get_srcport);
1662
1663static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1664{
1665	if (transport->srcport != 0)
1666		transport->srcport = 0;
1667	if (!transport->xprt.resvport)
1668		return 0;
1669	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1670		return xprt_max_resvport;
1671	return --port;
1672}
1673static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1674{
1675	struct sockaddr_storage myaddr;
1676	int err, nloop = 0;
1677	int port = xs_get_srcport(transport);
1678	unsigned short last;
1679
1680	/*
1681	 * If we are asking for any ephemeral port (i.e. port == 0 &&
1682	 * transport->xprt.resvport == 0), don't bind.  Let the local
1683	 * port selection happen implicitly when the socket is used
1684	 * (for example at connect time).
1685	 *
1686	 * This ensures that we can continue to establish TCP
1687	 * connections even when all local ephemeral ports are already
1688	 * a part of some TCP connection.  This makes no difference
1689	 * for UDP sockets, but also doesn't harm them.
1690	 *
1691	 * If we're asking for any reserved port (i.e. port == 0 &&
1692	 * transport->xprt.resvport == 1) xs_get_srcport above will
1693	 * ensure that port is non-zero and we will bind as needed.
1694	 */
1695	if (port <= 0)
1696		return port;
1697
1698	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1699	do {
1700		rpc_set_port((struct sockaddr *)&myaddr, port);
1701		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1702				transport->xprt.addrlen);
1703		if (err == 0) {
1704			if (transport->xprt.reuseport)
1705				transport->srcport = port;
1706			break;
1707		}
1708		last = port;
1709		port = xs_next_srcport(transport, port);
1710		if (port > last)
1711			nloop++;
1712	} while (err == -EADDRINUSE && nloop != 2);
1713
1714	if (myaddr.ss_family == AF_INET)
1715		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1716				&((struct sockaddr_in *)&myaddr)->sin_addr,
1717				port, err ? "failed" : "ok", err);
1718	else
1719		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1720				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1721				port, err ? "failed" : "ok", err);
1722	return err;
1723}
1724
1725/*
1726 * We don't support autobind on AF_LOCAL sockets
1727 */
1728static void xs_local_rpcbind(struct rpc_task *task)
1729{
1730	xprt_set_bound(task->tk_xprt);
1731}
1732
1733static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1734{
1735}
1736
1737#ifdef CONFIG_DEBUG_LOCK_ALLOC
1738static struct lock_class_key xs_key[2];
1739static struct lock_class_key xs_slock_key[2];
1740
1741static inline void xs_reclassify_socketu(struct socket *sock)
1742{
1743	struct sock *sk = sock->sk;
1744
1745	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1746		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1747}
1748
1749static inline void xs_reclassify_socket4(struct socket *sock)
1750{
1751	struct sock *sk = sock->sk;
1752
1753	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1754		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1755}
1756
1757static inline void xs_reclassify_socket6(struct socket *sock)
1758{
1759	struct sock *sk = sock->sk;
1760
1761	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1762		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1763}
1764
1765static inline void xs_reclassify_socket(int family, struct socket *sock)
1766{
1767	if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
 
1768		return;
1769
1770	switch (family) {
1771	case AF_LOCAL:
1772		xs_reclassify_socketu(sock);
1773		break;
1774	case AF_INET:
1775		xs_reclassify_socket4(sock);
1776		break;
1777	case AF_INET6:
1778		xs_reclassify_socket6(sock);
1779		break;
1780	}
1781}
1782#else
1783static inline void xs_reclassify_socket(int family, struct socket *sock)
1784{
1785}
1786#endif
1787
1788static void xs_dummy_setup_socket(struct work_struct *work)
1789{
1790}
1791
1792static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1793		struct sock_xprt *transport, int family, int type,
1794		int protocol, bool reuseport)
1795{
1796	struct file *filp;
1797	struct socket *sock;
1798	int err;
1799
1800	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1801	if (err < 0) {
1802		dprintk("RPC:       can't create %d transport socket (%d).\n",
1803				protocol, -err);
1804		goto out;
1805	}
1806	xs_reclassify_socket(family, sock);
1807
1808	if (reuseport)
1809		sock_set_reuseport(sock->sk);
1810
1811	err = xs_bind(transport, sock);
1812	if (err) {
1813		sock_release(sock);
1814		goto out;
1815	}
1816
1817	filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1818	if (IS_ERR(filp))
1819		return ERR_CAST(filp);
1820	transport->file = filp;
1821
1822	return sock;
1823out:
1824	return ERR_PTR(err);
1825}
1826
1827static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1828				      struct socket *sock)
1829{
1830	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1831									xprt);
1832
1833	if (!transport->inet) {
1834		struct sock *sk = sock->sk;
1835
1836		write_lock_bh(&sk->sk_callback_lock);
1837
1838		xs_save_old_callbacks(transport, sk);
1839
1840		sk->sk_user_data = xprt;
1841		sk->sk_data_ready = xs_data_ready;
1842		sk->sk_write_space = xs_udp_write_space;
1843		sock_set_flag(sk, SOCK_FASYNC);
1844		sk->sk_error_report = xs_error_report;
 
1845
1846		xprt_clear_connected(xprt);
1847
1848		/* Reset to new socket */
1849		transport->sock = sock;
1850		transport->inet = sk;
1851
1852		write_unlock_bh(&sk->sk_callback_lock);
1853	}
1854
1855	xs_stream_start_connect(transport);
1856
 
1857	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1858}
1859
1860/**
1861 * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1862 * @transport: socket transport to connect
1863 */
1864static int xs_local_setup_socket(struct sock_xprt *transport)
1865{
1866	struct rpc_xprt *xprt = &transport->xprt;
1867	struct file *filp;
1868	struct socket *sock;
1869	int status;
1870
1871	status = __sock_create(xprt->xprt_net, AF_LOCAL,
1872					SOCK_STREAM, 0, &sock, 1);
1873	if (status < 0) {
1874		dprintk("RPC:       can't create AF_LOCAL "
1875			"transport socket (%d).\n", -status);
1876		goto out;
1877	}
1878	xs_reclassify_socket(AF_LOCAL, sock);
1879
1880	filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1881	if (IS_ERR(filp)) {
1882		status = PTR_ERR(filp);
1883		goto out;
1884	}
1885	transport->file = filp;
1886
1887	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1888			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1889
1890	status = xs_local_finish_connecting(xprt, sock);
1891	trace_rpc_socket_connect(xprt, sock, status);
1892	switch (status) {
1893	case 0:
1894		dprintk("RPC:       xprt %p connected to %s\n",
1895				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1896		xprt->stat.connect_count++;
1897		xprt->stat.connect_time += (long)jiffies -
1898					   xprt->stat.connect_start;
1899		xprt_set_connected(xprt);
1900		break;
1901	case -ENOBUFS:
1902		break;
1903	case -ENOENT:
1904		dprintk("RPC:       xprt %p: socket %s does not exist\n",
1905				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1906		break;
1907	case -ECONNREFUSED:
1908		dprintk("RPC:       xprt %p: connection refused for %s\n",
1909				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1910		break;
1911	default:
1912		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1913				__func__, -status,
1914				xprt->address_strings[RPC_DISPLAY_ADDR]);
1915	}
1916
1917out:
1918	xprt_clear_connecting(xprt);
1919	xprt_wake_pending_tasks(xprt, status);
1920	return status;
1921}
1922
1923static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
1924{
1925	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1926	int ret;
1927
1928	 if (RPC_IS_ASYNC(task)) {
1929		/*
1930		 * We want the AF_LOCAL connect to be resolved in the
1931		 * filesystem namespace of the process making the rpc
1932		 * call.  Thus we connect synchronously.
1933		 *
1934		 * If we want to support asynchronous AF_LOCAL calls,
1935		 * we'll need to figure out how to pass a namespace to
1936		 * connect.
1937		 */
1938		task->tk_rpc_status = -ENOTCONN;
1939		rpc_exit(task, -ENOTCONN);
1940		return;
1941	}
1942	ret = xs_local_setup_socket(transport);
1943	if (ret && !RPC_IS_SOFTCONN(task))
1944		msleep_interruptible(15000);
1945}
1946
1947#if IS_ENABLED(CONFIG_SUNRPC_SWAP)
1948/*
1949 * Note that this should be called with XPRT_LOCKED held (or when we otherwise
1950 * know that we have exclusive access to the socket), to guard against
1951 * races with xs_reset_transport.
1952 */
1953static void xs_set_memalloc(struct rpc_xprt *xprt)
1954{
1955	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1956			xprt);
1957
1958	/*
1959	 * If there's no sock, then we have nothing to set. The
1960	 * reconnecting process will get it for us.
1961	 */
1962	if (!transport->inet)
1963		return;
1964	if (atomic_read(&xprt->swapper))
1965		sk_set_memalloc(transport->inet);
1966}
1967
1968/**
1969 * xs_enable_swap - Tag this transport as being used for swap.
1970 * @xprt: transport to tag
1971 *
1972 * Take a reference to this transport on behalf of the rpc_clnt, and
1973 * optionally mark it for swapping if it wasn't already.
1974 */
1975static int
1976xs_enable_swap(struct rpc_xprt *xprt)
1977{
1978	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
1979
1980	if (atomic_inc_return(&xprt->swapper) != 1)
1981		return 0;
1982	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
1983		return -ERESTARTSYS;
1984	if (xs->inet)
1985		sk_set_memalloc(xs->inet);
1986	xprt_release_xprt(xprt, NULL);
1987	return 0;
1988}
1989
1990/**
1991 * xs_disable_swap - Untag this transport as being used for swap.
1992 * @xprt: transport to tag
1993 *
1994 * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
1995 * swapper refcount goes to 0, untag the socket as a memalloc socket.
1996 */
1997static void
1998xs_disable_swap(struct rpc_xprt *xprt)
1999{
2000	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2001
2002	if (!atomic_dec_and_test(&xprt->swapper))
2003		return;
2004	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2005		return;
2006	if (xs->inet)
2007		sk_clear_memalloc(xs->inet);
2008	xprt_release_xprt(xprt, NULL);
2009}
2010#else
2011static void xs_set_memalloc(struct rpc_xprt *xprt)
2012{
2013}
2014
2015static int
2016xs_enable_swap(struct rpc_xprt *xprt)
2017{
2018	return -EINVAL;
2019}
2020
2021static void
2022xs_disable_swap(struct rpc_xprt *xprt)
2023{
2024}
2025#endif
2026
2027static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2028{
2029	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2030
2031	if (!transport->inet) {
2032		struct sock *sk = sock->sk;
2033
2034		write_lock_bh(&sk->sk_callback_lock);
2035
2036		xs_save_old_callbacks(transport, sk);
2037
2038		sk->sk_user_data = xprt;
2039		sk->sk_data_ready = xs_data_ready;
2040		sk->sk_write_space = xs_udp_write_space;
2041		sock_set_flag(sk, SOCK_FASYNC);
2042
2043		xprt_set_connected(xprt);
2044
2045		/* Reset to new socket */
2046		transport->sock = sock;
2047		transport->inet = sk;
2048
2049		xs_set_memalloc(xprt);
2050
2051		write_unlock_bh(&sk->sk_callback_lock);
2052	}
2053	xs_udp_do_set_buffer_size(xprt);
2054
2055	xprt->stat.connect_start = jiffies;
2056}
2057
2058static void xs_udp_setup_socket(struct work_struct *work)
2059{
2060	struct sock_xprt *transport =
2061		container_of(work, struct sock_xprt, connect_worker.work);
2062	struct rpc_xprt *xprt = &transport->xprt;
2063	struct socket *sock;
2064	int status = -EIO;
2065
2066	sock = xs_create_sock(xprt, transport,
2067			xs_addr(xprt)->sa_family, SOCK_DGRAM,
2068			IPPROTO_UDP, false);
2069	if (IS_ERR(sock))
2070		goto out;
2071
2072	dprintk("RPC:       worker connecting xprt %p via %s to "
2073				"%s (port %s)\n", xprt,
2074			xprt->address_strings[RPC_DISPLAY_PROTO],
2075			xprt->address_strings[RPC_DISPLAY_ADDR],
2076			xprt->address_strings[RPC_DISPLAY_PORT]);
2077
2078	xs_udp_finish_connecting(xprt, sock);
2079	trace_rpc_socket_connect(xprt, sock, 0);
2080	status = 0;
2081out:
 
2082	xprt_clear_connecting(xprt);
2083	xprt_unlock_connect(xprt, transport);
2084	xprt_wake_pending_tasks(xprt, status);
2085}
2086
2087/**
2088 * xs_tcp_shutdown - gracefully shut down a TCP socket
2089 * @xprt: transport
2090 *
2091 * Initiates a graceful shutdown of the TCP socket by calling the
2092 * equivalent of shutdown(SHUT_RDWR);
2093 */
2094static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2095{
2096	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2097	struct socket *sock = transport->sock;
2098	int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2099
2100	if (sock == NULL)
2101		return;
2102	switch (skst) {
2103	default:
2104		kernel_sock_shutdown(sock, SHUT_RDWR);
2105		trace_rpc_socket_shutdown(xprt, sock);
2106		break;
2107	case TCP_CLOSE:
2108	case TCP_TIME_WAIT:
2109		xs_reset_transport(transport);
2110	}
2111}
2112
2113static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2114		struct socket *sock)
2115{
2116	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2117	unsigned int keepidle;
2118	unsigned int keepcnt;
2119	unsigned int timeo;
2120
2121	spin_lock(&xprt->transport_lock);
2122	keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2123	keepcnt = xprt->timeout->to_retries + 1;
2124	timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2125		(xprt->timeout->to_retries + 1);
2126	clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2127	spin_unlock(&xprt->transport_lock);
2128
2129	/* TCP Keepalive options */
2130	sock_set_keepalive(sock->sk);
2131	tcp_sock_set_keepidle(sock->sk, keepidle);
2132	tcp_sock_set_keepintvl(sock->sk, keepidle);
2133	tcp_sock_set_keepcnt(sock->sk, keepcnt);
2134
2135	/* TCP user timeout (see RFC5482) */
2136	tcp_sock_set_user_timeout(sock->sk, timeo);
2137}
2138
2139static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2140		unsigned long connect_timeout,
2141		unsigned long reconnect_timeout)
2142{
2143	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2144	struct rpc_timeout to;
2145	unsigned long initval;
2146
2147	spin_lock(&xprt->transport_lock);
2148	if (reconnect_timeout < xprt->max_reconnect_timeout)
2149		xprt->max_reconnect_timeout = reconnect_timeout;
2150	if (connect_timeout < xprt->connect_timeout) {
2151		memcpy(&to, xprt->timeout, sizeof(to));
2152		initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2153		/* Arbitrary lower limit */
2154		if (initval <  XS_TCP_INIT_REEST_TO << 1)
2155			initval = XS_TCP_INIT_REEST_TO << 1;
2156		to.to_initval = initval;
2157		to.to_maxval = initval;
2158		memcpy(&transport->tcp_timeout, &to,
2159				sizeof(transport->tcp_timeout));
2160		xprt->timeout = &transport->tcp_timeout;
2161		xprt->connect_timeout = connect_timeout;
2162	}
2163	set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2164	spin_unlock(&xprt->transport_lock);
2165}
2166
2167static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2168{
2169	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2170	int ret = -ENOTCONN;
2171
2172	if (!transport->inet) {
2173		struct sock *sk = sock->sk;
2174
2175		/* Avoid temporary address, they are bad for long-lived
2176		 * connections such as NFS mounts.
2177		 * RFC4941, section 3.6 suggests that:
2178		 *    Individual applications, which have specific
2179		 *    knowledge about the normal duration of connections,
2180		 *    MAY override this as appropriate.
2181		 */
2182		if (xs_addr(xprt)->sa_family == PF_INET6) {
2183			ip6_sock_set_addr_preferences(sk,
2184				IPV6_PREFER_SRC_PUBLIC);
2185		}
2186
2187		xs_tcp_set_socket_timeouts(xprt, sock);
2188		tcp_sock_set_nodelay(sk);
 
 
 
 
 
2189
2190		write_lock_bh(&sk->sk_callback_lock);
2191
2192		xs_save_old_callbacks(transport, sk);
2193
2194		sk->sk_user_data = xprt;
2195		sk->sk_data_ready = xs_data_ready;
2196		sk->sk_state_change = xs_tcp_state_change;
2197		sk->sk_write_space = xs_tcp_write_space;
2198		sock_set_flag(sk, SOCK_FASYNC);
2199		sk->sk_error_report = xs_error_report;
 
2200
2201		/* socket options */
2202		sock_reset_flag(sk, SOCK_LINGER);
 
2203
2204		xprt_clear_connected(xprt);
2205
2206		/* Reset to new socket */
2207		transport->sock = sock;
2208		transport->inet = sk;
2209
2210		write_unlock_bh(&sk->sk_callback_lock);
2211	}
2212
2213	if (!xprt_bound(xprt))
2214		goto out;
2215
2216	xs_set_memalloc(xprt);
2217
2218	xs_stream_start_connect(transport);
2219
2220	/* Tell the socket layer to start connecting... */
 
 
2221	set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2222	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2223	switch (ret) {
2224	case 0:
2225		xs_set_srcport(transport, sock);
2226		fallthrough;
2227	case -EINPROGRESS:
2228		/* SYN_SENT! */
2229		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2230			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2231		break;
2232	case -EADDRNOTAVAIL:
2233		/* Source port number is unavailable. Try a new one! */
2234		transport->srcport = 0;
2235	}
2236out:
2237	return ret;
2238}
2239
2240/**
2241 * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2242 * @work: queued work item
2243 *
2244 * Invoked by a work queue tasklet.
2245 */
2246static void xs_tcp_setup_socket(struct work_struct *work)
2247{
2248	struct sock_xprt *transport =
2249		container_of(work, struct sock_xprt, connect_worker.work);
2250	struct socket *sock = transport->sock;
2251	struct rpc_xprt *xprt = &transport->xprt;
2252	int status = -EIO;
2253
2254	if (!sock) {
2255		sock = xs_create_sock(xprt, transport,
2256				xs_addr(xprt)->sa_family, SOCK_STREAM,
2257				IPPROTO_TCP, true);
2258		if (IS_ERR(sock)) {
2259			status = PTR_ERR(sock);
2260			goto out;
2261		}
2262	}
2263
2264	dprintk("RPC:       worker connecting xprt %p via %s to "
2265				"%s (port %s)\n", xprt,
2266			xprt->address_strings[RPC_DISPLAY_PROTO],
2267			xprt->address_strings[RPC_DISPLAY_ADDR],
2268			xprt->address_strings[RPC_DISPLAY_PORT]);
2269
2270	status = xs_tcp_finish_connecting(xprt, sock);
2271	trace_rpc_socket_connect(xprt, sock, status);
2272	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2273			xprt, -status, xprt_connected(xprt),
2274			sock->sk->sk_state);
2275	switch (status) {
2276	default:
2277		printk("%s: connect returned unhandled error %d\n",
2278			__func__, status);
2279		fallthrough;
2280	case -EADDRNOTAVAIL:
2281		/* We're probably in TIME_WAIT. Get rid of existing socket,
2282		 * and retry
2283		 */
2284		xs_tcp_force_close(xprt);
2285		break;
2286	case 0:
2287	case -EINPROGRESS:
2288	case -EALREADY:
2289		xprt_unlock_connect(xprt, transport);
2290		return;
2291	case -EINVAL:
2292		/* Happens, for instance, if the user specified a link
2293		 * local IPv6 address without a scope-id.
2294		 */
2295	case -ECONNREFUSED:
2296	case -ECONNRESET:
2297	case -ENETDOWN:
2298	case -ENETUNREACH:
2299	case -EHOSTUNREACH:
2300	case -EADDRINUSE:
2301	case -ENOBUFS:
2302		/* xs_tcp_force_close() wakes tasks with a fixed error code.
2303		 * We need to wake them first to ensure the correct error code.
2304		 */
2305		xprt_wake_pending_tasks(xprt, status);
2306		xs_tcp_force_close(xprt);
2307		goto out;
2308	}
2309	status = -EAGAIN;
2310out:
 
2311	xprt_clear_connecting(xprt);
2312	xprt_unlock_connect(xprt, transport);
2313	xprt_wake_pending_tasks(xprt, status);
2314}
2315
2316/**
2317 * xs_connect - connect a socket to a remote endpoint
2318 * @xprt: pointer to transport structure
2319 * @task: address of RPC task that manages state of connect request
2320 *
2321 * TCP: If the remote end dropped the connection, delay reconnecting.
2322 *
2323 * UDP socket connects are synchronous, but we use a work queue anyway
2324 * to guarantee that even unprivileged user processes can set up a
2325 * socket on a privileged port.
2326 *
2327 * If a UDP socket connect fails, the delay behavior here prevents
2328 * retry floods (hard mounts).
2329 */
2330static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2331{
2332	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2333	unsigned long delay = 0;
2334
2335	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2336
2337	if (transport->sock != NULL) {
2338		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2339				"seconds\n",
2340				xprt, xprt->reestablish_timeout / HZ);
2341
2342		/* Start by resetting any existing state */
2343		xs_reset_transport(transport);
2344
2345		delay = xprt_reconnect_delay(xprt);
2346		xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2347
2348	} else
 
 
 
 
 
2349		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2350
2351	queue_delayed_work(xprtiod_workqueue,
2352			&transport->connect_worker,
2353			delay);
2354}
2355
2356static void xs_wake_disconnect(struct sock_xprt *transport)
2357{
2358	if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2359		xs_tcp_force_close(&transport->xprt);
2360}
2361
2362static void xs_wake_write(struct sock_xprt *transport)
2363{
2364	if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2365		xprt_write_space(&transport->xprt);
2366}
2367
2368static void xs_wake_error(struct sock_xprt *transport)
2369{
2370	int sockerr;
2371
2372	if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2373		return;
2374	mutex_lock(&transport->recv_mutex);
2375	if (transport->sock == NULL)
2376		goto out;
2377	if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2378		goto out;
2379	sockerr = xchg(&transport->xprt_err, 0);
2380	if (sockerr < 0)
2381		xprt_wake_pending_tasks(&transport->xprt, sockerr);
2382out:
2383	mutex_unlock(&transport->recv_mutex);
2384}
2385
2386static void xs_wake_pending(struct sock_xprt *transport)
2387{
2388	if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2389		xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2390}
2391
2392static void xs_error_handle(struct work_struct *work)
2393{
2394	struct sock_xprt *transport = container_of(work,
2395			struct sock_xprt, error_worker);
2396
2397	xs_wake_disconnect(transport);
2398	xs_wake_write(transport);
2399	xs_wake_error(transport);
2400	xs_wake_pending(transport);
2401}
2402
2403/**
2404 * xs_local_print_stats - display AF_LOCAL socket-specific stats
2405 * @xprt: rpc_xprt struct containing statistics
2406 * @seq: output file
2407 *
2408 */
2409static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2410{
2411	long idle_time = 0;
2412
2413	if (xprt_connected(xprt))
2414		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2415
2416	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2417			"%llu %llu %lu %llu %llu\n",
2418			xprt->stat.bind_count,
2419			xprt->stat.connect_count,
2420			xprt->stat.connect_time / HZ,
2421			idle_time,
2422			xprt->stat.sends,
2423			xprt->stat.recvs,
2424			xprt->stat.bad_xids,
2425			xprt->stat.req_u,
2426			xprt->stat.bklog_u,
2427			xprt->stat.max_slots,
2428			xprt->stat.sending_u,
2429			xprt->stat.pending_u);
2430}
2431
2432/**
2433 * xs_udp_print_stats - display UDP socket-specific stats
2434 * @xprt: rpc_xprt struct containing statistics
2435 * @seq: output file
2436 *
2437 */
2438static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2439{
2440	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2441
2442	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2443			"%lu %llu %llu\n",
2444			transport->srcport,
2445			xprt->stat.bind_count,
2446			xprt->stat.sends,
2447			xprt->stat.recvs,
2448			xprt->stat.bad_xids,
2449			xprt->stat.req_u,
2450			xprt->stat.bklog_u,
2451			xprt->stat.max_slots,
2452			xprt->stat.sending_u,
2453			xprt->stat.pending_u);
2454}
2455
2456/**
2457 * xs_tcp_print_stats - display TCP socket-specific stats
2458 * @xprt: rpc_xprt struct containing statistics
2459 * @seq: output file
2460 *
2461 */
2462static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2463{
2464	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2465	long idle_time = 0;
2466
2467	if (xprt_connected(xprt))
2468		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2469
2470	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2471			"%llu %llu %lu %llu %llu\n",
2472			transport->srcport,
2473			xprt->stat.bind_count,
2474			xprt->stat.connect_count,
2475			xprt->stat.connect_time / HZ,
2476			idle_time,
2477			xprt->stat.sends,
2478			xprt->stat.recvs,
2479			xprt->stat.bad_xids,
2480			xprt->stat.req_u,
2481			xprt->stat.bklog_u,
2482			xprt->stat.max_slots,
2483			xprt->stat.sending_u,
2484			xprt->stat.pending_u);
2485}
2486
2487/*
2488 * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2489 * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2490 * to use the server side send routines.
2491 */
2492static int bc_malloc(struct rpc_task *task)
2493{
2494	struct rpc_rqst *rqst = task->tk_rqstp;
2495	size_t size = rqst->rq_callsize;
2496	struct page *page;
2497	struct rpc_buffer *buf;
2498
2499	if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2500		WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2501			  size);
2502		return -EINVAL;
2503	}
2504
2505	page = alloc_page(GFP_KERNEL);
2506	if (!page)
2507		return -ENOMEM;
2508
2509	buf = page_address(page);
2510	buf->len = PAGE_SIZE;
2511
2512	rqst->rq_buffer = buf->data;
2513	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2514	return 0;
2515}
2516
2517/*
2518 * Free the space allocated in the bc_alloc routine
2519 */
2520static void bc_free(struct rpc_task *task)
2521{
2522	void *buffer = task->tk_rqstp->rq_buffer;
2523	struct rpc_buffer *buf;
2524
 
 
 
2525	buf = container_of(buffer, struct rpc_buffer, data);
2526	free_page((unsigned long)buf);
2527}
2528
 
 
 
 
2529static int bc_sendto(struct rpc_rqst *req)
2530{
2531	struct xdr_buf *xdr = &req->rq_snd_buf;
 
 
2532	struct sock_xprt *transport =
2533			container_of(req->rq_xprt, struct sock_xprt, xprt);
2534	struct msghdr msg = {
2535		.msg_flags	= 0,
2536	};
2537	rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2538					 (u32)xdr->len);
2539	unsigned int sent = 0;
2540	int err;
 
 
 
 
 
 
 
 
 
2541
2542	req->rq_xtime = ktime_get();
2543	err = xprt_sock_sendmsg(transport->sock, &msg, xdr, 0, marker, &sent);
2544	xdr_free_bvec(xdr);
2545	if (err < 0 || sent != (xdr->len + sizeof(marker)))
2546		return -EAGAIN;
2547	return sent;
2548}
2549
2550/**
2551 * bc_send_request - Send a backchannel Call on a TCP socket
2552 * @req: rpc_rqst containing Call message to be sent
2553 *
2554 * xpt_mutex ensures @rqstp's whole message is written to the socket
2555 * without interruption.
2556 *
2557 * Return values:
2558 *   %0 if the message was sent successfully
2559 *   %ENOTCONN if the message was not sent
2560 */
2561static int bc_send_request(struct rpc_rqst *req)
2562{
 
2563	struct svc_xprt	*xprt;
2564	int len;
2565
 
2566	/*
2567	 * Get the server socket associated with this callback xprt
2568	 */
2569	xprt = req->rq_xprt->bc_xprt;
2570
2571	/*
2572	 * Grab the mutex to serialize data as the connection is shared
2573	 * with the fore channel
2574	 */
2575	mutex_lock(&xprt->xpt_mutex);
 
 
 
 
 
2576	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2577		len = -ENOTCONN;
2578	else
2579		len = bc_sendto(req);
2580	mutex_unlock(&xprt->xpt_mutex);
2581
2582	if (len > 0)
2583		len = 0;
2584
2585	return len;
2586}
2587
2588/*
2589 * The close routine. Since this is client initiated, we do nothing
2590 */
2591
2592static void bc_close(struct rpc_xprt *xprt)
2593{
2594	xprt_disconnect_done(xprt);
2595}
2596
2597/*
2598 * The xprt destroy routine. Again, because this connection is client
2599 * initiated, we do nothing
2600 */
2601
2602static void bc_destroy(struct rpc_xprt *xprt)
2603{
2604	dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2605
2606	xs_xprt_free(xprt);
2607	module_put(THIS_MODULE);
2608}
2609
2610static const struct rpc_xprt_ops xs_local_ops = {
2611	.reserve_xprt		= xprt_reserve_xprt,
2612	.release_xprt		= xprt_release_xprt,
2613	.alloc_slot		= xprt_alloc_slot,
2614	.free_slot		= xprt_free_slot,
2615	.rpcbind		= xs_local_rpcbind,
2616	.set_port		= xs_local_set_port,
2617	.connect		= xs_local_connect,
2618	.buf_alloc		= rpc_malloc,
2619	.buf_free		= rpc_free,
2620	.prepare_request	= xs_stream_prepare_request,
2621	.send_request		= xs_local_send_request,
2622	.wait_for_reply_request	= xprt_wait_for_reply_request_def,
2623	.close			= xs_close,
2624	.destroy		= xs_destroy,
2625	.print_stats		= xs_local_print_stats,
2626	.enable_swap		= xs_enable_swap,
2627	.disable_swap		= xs_disable_swap,
2628};
2629
2630static const struct rpc_xprt_ops xs_udp_ops = {
2631	.set_buffer_size	= xs_udp_set_buffer_size,
2632	.reserve_xprt		= xprt_reserve_xprt_cong,
2633	.release_xprt		= xprt_release_xprt_cong,
2634	.alloc_slot		= xprt_alloc_slot,
2635	.free_slot		= xprt_free_slot,
2636	.rpcbind		= rpcb_getport_async,
2637	.set_port		= xs_set_port,
2638	.connect		= xs_connect,
2639	.buf_alloc		= rpc_malloc,
2640	.buf_free		= rpc_free,
2641	.send_request		= xs_udp_send_request,
2642	.wait_for_reply_request	= xprt_wait_for_reply_request_rtt,
2643	.timer			= xs_udp_timer,
2644	.release_request	= xprt_release_rqst_cong,
2645	.close			= xs_close,
2646	.destroy		= xs_destroy,
2647	.print_stats		= xs_udp_print_stats,
2648	.enable_swap		= xs_enable_swap,
2649	.disable_swap		= xs_disable_swap,
2650	.inject_disconnect	= xs_inject_disconnect,
2651};
2652
2653static const struct rpc_xprt_ops xs_tcp_ops = {
2654	.reserve_xprt		= xprt_reserve_xprt,
2655	.release_xprt		= xprt_release_xprt,
2656	.alloc_slot		= xprt_alloc_slot,
2657	.free_slot		= xprt_free_slot,
2658	.rpcbind		= rpcb_getport_async,
2659	.set_port		= xs_set_port,
2660	.connect		= xs_connect,
2661	.buf_alloc		= rpc_malloc,
2662	.buf_free		= rpc_free,
2663	.prepare_request	= xs_stream_prepare_request,
2664	.send_request		= xs_tcp_send_request,
2665	.wait_for_reply_request	= xprt_wait_for_reply_request_def,
2666	.close			= xs_tcp_shutdown,
2667	.destroy		= xs_destroy,
2668	.set_connect_timeout	= xs_tcp_set_connect_timeout,
2669	.print_stats		= xs_tcp_print_stats,
2670	.enable_swap		= xs_enable_swap,
2671	.disable_swap		= xs_disable_swap,
2672	.inject_disconnect	= xs_inject_disconnect,
2673#ifdef CONFIG_SUNRPC_BACKCHANNEL
2674	.bc_setup		= xprt_setup_bc,
2675	.bc_maxpayload		= xs_tcp_bc_maxpayload,
2676	.bc_num_slots		= xprt_bc_max_slots,
2677	.bc_free_rqst		= xprt_free_bc_rqst,
2678	.bc_destroy		= xprt_destroy_bc,
2679#endif
2680};
2681
2682/*
2683 * The rpc_xprt_ops for the server backchannel
2684 */
2685
2686static const struct rpc_xprt_ops bc_tcp_ops = {
2687	.reserve_xprt		= xprt_reserve_xprt,
2688	.release_xprt		= xprt_release_xprt,
2689	.alloc_slot		= xprt_alloc_slot,
2690	.free_slot		= xprt_free_slot,
2691	.buf_alloc		= bc_malloc,
2692	.buf_free		= bc_free,
2693	.send_request		= bc_send_request,
2694	.wait_for_reply_request	= xprt_wait_for_reply_request_def,
2695	.close			= bc_close,
2696	.destroy		= bc_destroy,
2697	.print_stats		= xs_tcp_print_stats,
2698	.enable_swap		= xs_enable_swap,
2699	.disable_swap		= xs_disable_swap,
2700	.inject_disconnect	= xs_inject_disconnect,
2701};
2702
2703static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2704{
2705	static const struct sockaddr_in sin = {
2706		.sin_family		= AF_INET,
2707		.sin_addr.s_addr	= htonl(INADDR_ANY),
2708	};
2709	static const struct sockaddr_in6 sin6 = {
2710		.sin6_family		= AF_INET6,
2711		.sin6_addr		= IN6ADDR_ANY_INIT,
2712	};
2713
2714	switch (family) {
2715	case AF_LOCAL:
2716		break;
2717	case AF_INET:
2718		memcpy(sap, &sin, sizeof(sin));
2719		break;
2720	case AF_INET6:
2721		memcpy(sap, &sin6, sizeof(sin6));
2722		break;
2723	default:
2724		dprintk("RPC:       %s: Bad address family\n", __func__);
2725		return -EAFNOSUPPORT;
2726	}
2727	return 0;
2728}
2729
2730static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2731				      unsigned int slot_table_size,
2732				      unsigned int max_slot_table_size)
2733{
2734	struct rpc_xprt *xprt;
2735	struct sock_xprt *new;
2736
2737	if (args->addrlen > sizeof(xprt->addr)) {
2738		dprintk("RPC:       xs_setup_xprt: address too large\n");
2739		return ERR_PTR(-EBADF);
2740	}
2741
2742	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2743			max_slot_table_size);
2744	if (xprt == NULL) {
2745		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2746				"rpc_xprt\n");
2747		return ERR_PTR(-ENOMEM);
2748	}
2749
2750	new = container_of(xprt, struct sock_xprt, xprt);
2751	mutex_init(&new->recv_mutex);
2752	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2753	xprt->addrlen = args->addrlen;
2754	if (args->srcaddr)
2755		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2756	else {
2757		int err;
2758		err = xs_init_anyaddr(args->dstaddr->sa_family,
2759					(struct sockaddr *)&new->srcaddr);
2760		if (err != 0) {
2761			xprt_free(xprt);
2762			return ERR_PTR(err);
2763		}
2764	}
2765
2766	return xprt;
2767}
2768
2769static const struct rpc_timeout xs_local_default_timeout = {
2770	.to_initval = 10 * HZ,
2771	.to_maxval = 10 * HZ,
2772	.to_retries = 2,
2773};
2774
2775/**
2776 * xs_setup_local - Set up transport to use an AF_LOCAL socket
2777 * @args: rpc transport creation arguments
2778 *
2779 * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2780 */
2781static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2782{
2783	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2784	struct sock_xprt *transport;
2785	struct rpc_xprt *xprt;
2786	struct rpc_xprt *ret;
2787
2788	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2789			xprt_max_tcp_slot_table_entries);
2790	if (IS_ERR(xprt))
2791		return xprt;
2792	transport = container_of(xprt, struct sock_xprt, xprt);
2793
2794	xprt->prot = 0;
2795	xprt->xprt_class = &xs_local_transport;
2796	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2797
2798	xprt->bind_timeout = XS_BIND_TO;
2799	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2800	xprt->idle_timeout = XS_IDLE_DISC_TO;
2801
2802	xprt->ops = &xs_local_ops;
2803	xprt->timeout = &xs_local_default_timeout;
2804
2805	INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2806	INIT_WORK(&transport->error_worker, xs_error_handle);
2807	INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2808
2809	switch (sun->sun_family) {
2810	case AF_LOCAL:
2811		if (sun->sun_path[0] != '/') {
2812			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2813					sun->sun_path);
2814			ret = ERR_PTR(-EINVAL);
2815			goto out_err;
2816		}
2817		xprt_set_bound(xprt);
2818		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2819		ret = ERR_PTR(xs_local_setup_socket(transport));
2820		if (ret)
2821			goto out_err;
2822		break;
2823	default:
2824		ret = ERR_PTR(-EAFNOSUPPORT);
2825		goto out_err;
2826	}
2827
2828	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2829			xprt->address_strings[RPC_DISPLAY_ADDR]);
2830
2831	if (try_module_get(THIS_MODULE))
2832		return xprt;
2833	ret = ERR_PTR(-EINVAL);
2834out_err:
2835	xs_xprt_free(xprt);
2836	return ret;
2837}
2838
2839static const struct rpc_timeout xs_udp_default_timeout = {
2840	.to_initval = 5 * HZ,
2841	.to_maxval = 30 * HZ,
2842	.to_increment = 5 * HZ,
2843	.to_retries = 5,
2844};
2845
2846/**
2847 * xs_setup_udp - Set up transport to use a UDP socket
2848 * @args: rpc transport creation arguments
2849 *
2850 */
2851static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2852{
2853	struct sockaddr *addr = args->dstaddr;
2854	struct rpc_xprt *xprt;
2855	struct sock_xprt *transport;
2856	struct rpc_xprt *ret;
2857
2858	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2859			xprt_udp_slot_table_entries);
2860	if (IS_ERR(xprt))
2861		return xprt;
2862	transport = container_of(xprt, struct sock_xprt, xprt);
2863
2864	xprt->prot = IPPROTO_UDP;
2865	xprt->xprt_class = &xs_udp_transport;
2866	/* XXX: header size can vary due to auth type, IPv6, etc. */
2867	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2868
2869	xprt->bind_timeout = XS_BIND_TO;
2870	xprt->reestablish_timeout = XS_UDP_REEST_TO;
2871	xprt->idle_timeout = XS_IDLE_DISC_TO;
2872
2873	xprt->ops = &xs_udp_ops;
2874
2875	xprt->timeout = &xs_udp_default_timeout;
2876
2877	INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2878	INIT_WORK(&transport->error_worker, xs_error_handle);
2879	INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2880
2881	switch (addr->sa_family) {
2882	case AF_INET:
2883		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2884			xprt_set_bound(xprt);
2885
2886		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2887		break;
2888	case AF_INET6:
2889		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2890			xprt_set_bound(xprt);
2891
2892		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2893		break;
2894	default:
2895		ret = ERR_PTR(-EAFNOSUPPORT);
2896		goto out_err;
2897	}
2898
2899	if (xprt_bound(xprt))
2900		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2901				xprt->address_strings[RPC_DISPLAY_ADDR],
2902				xprt->address_strings[RPC_DISPLAY_PORT],
2903				xprt->address_strings[RPC_DISPLAY_PROTO]);
2904	else
2905		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2906				xprt->address_strings[RPC_DISPLAY_ADDR],
2907				xprt->address_strings[RPC_DISPLAY_PROTO]);
2908
2909	if (try_module_get(THIS_MODULE))
2910		return xprt;
2911	ret = ERR_PTR(-EINVAL);
2912out_err:
2913	xs_xprt_free(xprt);
2914	return ret;
2915}
2916
2917static const struct rpc_timeout xs_tcp_default_timeout = {
2918	.to_initval = 60 * HZ,
2919	.to_maxval = 60 * HZ,
2920	.to_retries = 2,
2921};
2922
2923/**
2924 * xs_setup_tcp - Set up transport to use a TCP socket
2925 * @args: rpc transport creation arguments
2926 *
2927 */
2928static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2929{
2930	struct sockaddr *addr = args->dstaddr;
2931	struct rpc_xprt *xprt;
2932	struct sock_xprt *transport;
2933	struct rpc_xprt *ret;
2934	unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
2935
2936	if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
2937		max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
2938
2939	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2940			max_slot_table_size);
2941	if (IS_ERR(xprt))
2942		return xprt;
2943	transport = container_of(xprt, struct sock_xprt, xprt);
2944
2945	xprt->prot = IPPROTO_TCP;
2946	xprt->xprt_class = &xs_tcp_transport;
2947	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2948
2949	xprt->bind_timeout = XS_BIND_TO;
2950	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2951	xprt->idle_timeout = XS_IDLE_DISC_TO;
2952
2953	xprt->ops = &xs_tcp_ops;
2954	xprt->timeout = &xs_tcp_default_timeout;
2955
2956	xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
2957	xprt->connect_timeout = xprt->timeout->to_initval *
2958		(xprt->timeout->to_retries + 1);
2959
2960	INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2961	INIT_WORK(&transport->error_worker, xs_error_handle);
2962	INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
2963
2964	switch (addr->sa_family) {
2965	case AF_INET:
2966		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2967			xprt_set_bound(xprt);
2968
2969		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2970		break;
2971	case AF_INET6:
2972		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2973			xprt_set_bound(xprt);
2974
2975		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2976		break;
2977	default:
2978		ret = ERR_PTR(-EAFNOSUPPORT);
2979		goto out_err;
2980	}
2981
2982	if (xprt_bound(xprt))
2983		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2984				xprt->address_strings[RPC_DISPLAY_ADDR],
2985				xprt->address_strings[RPC_DISPLAY_PORT],
2986				xprt->address_strings[RPC_DISPLAY_PROTO]);
2987	else
2988		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2989				xprt->address_strings[RPC_DISPLAY_ADDR],
2990				xprt->address_strings[RPC_DISPLAY_PROTO]);
2991
2992	if (try_module_get(THIS_MODULE))
2993		return xprt;
2994	ret = ERR_PTR(-EINVAL);
2995out_err:
2996	xs_xprt_free(xprt);
2997	return ret;
2998}
2999
3000/**
3001 * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3002 * @args: rpc transport creation arguments
3003 *
3004 */
3005static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3006{
3007	struct sockaddr *addr = args->dstaddr;
3008	struct rpc_xprt *xprt;
3009	struct sock_xprt *transport;
3010	struct svc_sock *bc_sock;
3011	struct rpc_xprt *ret;
3012
3013	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3014			xprt_tcp_slot_table_entries);
3015	if (IS_ERR(xprt))
3016		return xprt;
3017	transport = container_of(xprt, struct sock_xprt, xprt);
3018
3019	xprt->prot = IPPROTO_TCP;
3020	xprt->xprt_class = &xs_bc_tcp_transport;
3021	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3022	xprt->timeout = &xs_tcp_default_timeout;
3023
3024	/* backchannel */
3025	xprt_set_bound(xprt);
3026	xprt->bind_timeout = 0;
3027	xprt->reestablish_timeout = 0;
3028	xprt->idle_timeout = 0;
3029
3030	xprt->ops = &bc_tcp_ops;
3031
3032	switch (addr->sa_family) {
3033	case AF_INET:
3034		xs_format_peer_addresses(xprt, "tcp",
3035					 RPCBIND_NETID_TCP);
3036		break;
3037	case AF_INET6:
3038		xs_format_peer_addresses(xprt, "tcp",
3039				   RPCBIND_NETID_TCP6);
3040		break;
3041	default:
3042		ret = ERR_PTR(-EAFNOSUPPORT);
3043		goto out_err;
3044	}
3045
3046	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3047			xprt->address_strings[RPC_DISPLAY_ADDR],
3048			xprt->address_strings[RPC_DISPLAY_PORT],
3049			xprt->address_strings[RPC_DISPLAY_PROTO]);
3050
3051	/*
3052	 * Once we've associated a backchannel xprt with a connection,
3053	 * we want to keep it around as long as the connection lasts,
3054	 * in case we need to start using it for a backchannel again;
3055	 * this reference won't be dropped until bc_xprt is destroyed.
3056	 */
3057	xprt_get(xprt);
3058	args->bc_xprt->xpt_bc_xprt = xprt;
3059	xprt->bc_xprt = args->bc_xprt;
3060	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3061	transport->sock = bc_sock->sk_sock;
3062	transport->inet = bc_sock->sk_sk;
3063
3064	/*
3065	 * Since we don't want connections for the backchannel, we set
3066	 * the xprt status to connected
3067	 */
3068	xprt_set_connected(xprt);
3069
3070	if (try_module_get(THIS_MODULE))
3071		return xprt;
3072
3073	args->bc_xprt->xpt_bc_xprt = NULL;
3074	args->bc_xprt->xpt_bc_xps = NULL;
3075	xprt_put(xprt);
3076	ret = ERR_PTR(-EINVAL);
3077out_err:
3078	xs_xprt_free(xprt);
3079	return ret;
3080}
3081
3082static struct xprt_class	xs_local_transport = {
3083	.list		= LIST_HEAD_INIT(xs_local_transport.list),
3084	.name		= "named UNIX socket",
3085	.owner		= THIS_MODULE,
3086	.ident		= XPRT_TRANSPORT_LOCAL,
3087	.setup		= xs_setup_local,
3088	.netid		= { "" },
3089};
3090
3091static struct xprt_class	xs_udp_transport = {
3092	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
3093	.name		= "udp",
3094	.owner		= THIS_MODULE,
3095	.ident		= XPRT_TRANSPORT_UDP,
3096	.setup		= xs_setup_udp,
3097	.netid		= { "udp", "udp6", "" },
3098};
3099
3100static struct xprt_class	xs_tcp_transport = {
3101	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
3102	.name		= "tcp",
3103	.owner		= THIS_MODULE,
3104	.ident		= XPRT_TRANSPORT_TCP,
3105	.setup		= xs_setup_tcp,
3106	.netid		= { "tcp", "tcp6", "" },
3107};
3108
3109static struct xprt_class	xs_bc_tcp_transport = {
3110	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3111	.name		= "tcp NFSv4.1 backchannel",
3112	.owner		= THIS_MODULE,
3113	.ident		= XPRT_TRANSPORT_BC_TCP,
3114	.setup		= xs_setup_bc_tcp,
3115	.netid		= { "" },
3116};
3117
3118/**
3119 * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3120 *
3121 */
3122int init_socket_xprt(void)
3123{
 
3124	if (!sunrpc_table_header)
3125		sunrpc_table_header = register_sysctl_table(sunrpc_table);
 
3126
3127	xprt_register_transport(&xs_local_transport);
3128	xprt_register_transport(&xs_udp_transport);
3129	xprt_register_transport(&xs_tcp_transport);
3130	xprt_register_transport(&xs_bc_tcp_transport);
3131
3132	return 0;
3133}
3134
3135/**
3136 * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3137 *
3138 */
3139void cleanup_socket_xprt(void)
3140{
 
3141	if (sunrpc_table_header) {
3142		unregister_sysctl_table(sunrpc_table_header);
3143		sunrpc_table_header = NULL;
3144	}
 
3145
3146	xprt_unregister_transport(&xs_local_transport);
3147	xprt_unregister_transport(&xs_udp_transport);
3148	xprt_unregister_transport(&xs_tcp_transport);
3149	xprt_unregister_transport(&xs_bc_tcp_transport);
3150}
3151
3152static int param_set_uint_minmax(const char *val,
3153		const struct kernel_param *kp,
3154		unsigned int min, unsigned int max)
3155{
3156	unsigned int num;
3157	int ret;
3158
3159	if (!val)
3160		return -EINVAL;
3161	ret = kstrtouint(val, 0, &num);
3162	if (ret)
3163		return ret;
3164	if (num < min || num > max)
3165		return -EINVAL;
3166	*((unsigned int *)kp->arg) = num;
3167	return 0;
3168}
3169
3170static int param_set_portnr(const char *val, const struct kernel_param *kp)
3171{
3172	return param_set_uint_minmax(val, kp,
3173			RPC_MIN_RESVPORT,
3174			RPC_MAX_RESVPORT);
3175}
3176
3177static const struct kernel_param_ops param_ops_portnr = {
3178	.set = param_set_portnr,
3179	.get = param_get_uint,
3180};
3181
3182#define param_check_portnr(name, p) \
3183	__param_check(name, p, unsigned int);
3184
3185module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3186module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3187
3188static int param_set_slot_table_size(const char *val,
3189				     const struct kernel_param *kp)
3190{
3191	return param_set_uint_minmax(val, kp,
3192			RPC_MIN_SLOT_TABLE,
3193			RPC_MAX_SLOT_TABLE);
3194}
3195
3196static const struct kernel_param_ops param_ops_slot_table_size = {
3197	.set = param_set_slot_table_size,
3198	.get = param_get_uint,
3199};
3200
3201#define param_check_slot_table_size(name, p) \
3202	__param_check(name, p, unsigned int);
3203
3204static int param_set_max_slot_table_size(const char *val,
3205				     const struct kernel_param *kp)
3206{
3207	return param_set_uint_minmax(val, kp,
3208			RPC_MIN_SLOT_TABLE,
3209			RPC_MAX_SLOT_TABLE_LIMIT);
3210}
3211
3212static const struct kernel_param_ops param_ops_max_slot_table_size = {
3213	.set = param_set_max_slot_table_size,
3214	.get = param_get_uint,
3215};
3216
3217#define param_check_max_slot_table_size(name, p) \
3218	__param_check(name, p, unsigned int);
3219
3220module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3221		   slot_table_size, 0644);
3222module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3223		   max_slot_table_size, 0644);
3224module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3225		   slot_table_size, 0644);