xref: /linux/net/sunrpc/xprtsock.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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 
54 static void xs_close(struct rpc_xprt *xprt);
55 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
56 		struct socket *sock);
57 
58 /*
59  * xprtsock tunables
60  */
61 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
62 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
63 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
64 
65 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
66 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
67 
68 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
69 
70 #define XS_TCP_LINGER_TO	(15U * HZ)
71 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
72 
73 /*
74  * We can register our own files under /proc/sys/sunrpc by
75  * calling register_sysctl_table() again.  The files in that
76  * directory become the union of all files registered there.
77  *
78  * We simply need to make sure that we don't collide with
79  * someone else's file names!
80  */
81 
82 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
83 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
84 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
85 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
86 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
87 
88 static struct ctl_table_header *sunrpc_table_header;
89 
90 /*
91  * FIXME: changing the UDP slot table size should also resize the UDP
92  *        socket buffers for existing UDP transports
93  */
94 static struct ctl_table xs_tunables_table[] = {
95 	{
96 		.procname	= "udp_slot_table_entries",
97 		.data		= &xprt_udp_slot_table_entries,
98 		.maxlen		= sizeof(unsigned int),
99 		.mode		= 0644,
100 		.proc_handler	= proc_dointvec_minmax,
101 		.extra1		= &min_slot_table_size,
102 		.extra2		= &max_slot_table_size
103 	},
104 	{
105 		.procname	= "tcp_slot_table_entries",
106 		.data		= &xprt_tcp_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_max_slot_table_entries",
115 		.data		= &xprt_max_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_tcp_slot_table_limit
121 	},
122 	{
123 		.procname	= "min_resvport",
124 		.data		= &xprt_min_resvport,
125 		.maxlen		= sizeof(unsigned int),
126 		.mode		= 0644,
127 		.proc_handler	= proc_dointvec_minmax,
128 		.extra1		= &xprt_min_resvport_limit,
129 		.extra2		= &xprt_max_resvport
130 	},
131 	{
132 		.procname	= "max_resvport",
133 		.data		= &xprt_max_resvport,
134 		.maxlen		= sizeof(unsigned int),
135 		.mode		= 0644,
136 		.proc_handler	= proc_dointvec_minmax,
137 		.extra1		= &xprt_min_resvport,
138 		.extra2		= &xprt_max_resvport_limit
139 	},
140 	{
141 		.procname	= "tcp_fin_timeout",
142 		.data		= &xs_tcp_fin_timeout,
143 		.maxlen		= sizeof(xs_tcp_fin_timeout),
144 		.mode		= 0644,
145 		.proc_handler	= proc_dointvec_jiffies,
146 	},
147 	{ },
148 };
149 
150 static struct ctl_table sunrpc_table[] = {
151 	{
152 		.procname	= "sunrpc",
153 		.mode		= 0555,
154 		.child		= xs_tunables_table
155 	},
156 	{ },
157 };
158 
159 #endif
160 
161 /*
162  * Wait duration for a reply from the RPC portmapper.
163  */
164 #define XS_BIND_TO		(60U * HZ)
165 
166 /*
167  * Delay if a UDP socket connect error occurs.  This is most likely some
168  * kind of resource problem on the local host.
169  */
170 #define XS_UDP_REEST_TO		(2U * HZ)
171 
172 /*
173  * The reestablish timeout allows clients to delay for a bit before attempting
174  * to reconnect to a server that just dropped our connection.
175  *
176  * We implement an exponential backoff when trying to reestablish a TCP
177  * transport connection with the server.  Some servers like to drop a TCP
178  * connection when they are overworked, so we start with a short timeout and
179  * increase over time if the server is down or not responding.
180  */
181 #define XS_TCP_INIT_REEST_TO	(3U * HZ)
182 
183 /*
184  * TCP idle timeout; client drops the transport socket if it is idle
185  * for this long.  Note that we also timeout UDP sockets to prevent
186  * holding port numbers when there is no RPC traffic.
187  */
188 #define XS_IDLE_DISC_TO		(5U * 60 * HZ)
189 
190 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
191 # undef  RPC_DEBUG_DATA
192 # define RPCDBG_FACILITY	RPCDBG_TRANS
193 #endif
194 
195 #ifdef RPC_DEBUG_DATA
196 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
197 {
198 	u8 *buf = (u8 *) packet;
199 	int j;
200 
201 	dprintk("RPC:       %s\n", msg);
202 	for (j = 0; j < count && j < 128; j += 4) {
203 		if (!(j & 31)) {
204 			if (j)
205 				dprintk("\n");
206 			dprintk("0x%04x ", j);
207 		}
208 		dprintk("%02x%02x%02x%02x ",
209 			buf[j], buf[j+1], buf[j+2], buf[j+3]);
210 	}
211 	dprintk("\n");
212 }
213 #else
214 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
215 {
216 	/* NOP */
217 }
218 #endif
219 
220 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
221 {
222 	return (struct rpc_xprt *) sk->sk_user_data;
223 }
224 
225 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
226 {
227 	return (struct sockaddr *) &xprt->addr;
228 }
229 
230 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
231 {
232 	return (struct sockaddr_un *) &xprt->addr;
233 }
234 
235 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
236 {
237 	return (struct sockaddr_in *) &xprt->addr;
238 }
239 
240 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
241 {
242 	return (struct sockaddr_in6 *) &xprt->addr;
243 }
244 
245 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
246 {
247 	struct sockaddr *sap = xs_addr(xprt);
248 	struct sockaddr_in6 *sin6;
249 	struct sockaddr_in *sin;
250 	struct sockaddr_un *sun;
251 	char buf[128];
252 
253 	switch (sap->sa_family) {
254 	case AF_LOCAL:
255 		sun = xs_addr_un(xprt);
256 		strlcpy(buf, sun->sun_path, sizeof(buf));
257 		xprt->address_strings[RPC_DISPLAY_ADDR] =
258 						kstrdup(buf, GFP_KERNEL);
259 		break;
260 	case AF_INET:
261 		(void)rpc_ntop(sap, buf, sizeof(buf));
262 		xprt->address_strings[RPC_DISPLAY_ADDR] =
263 						kstrdup(buf, GFP_KERNEL);
264 		sin = xs_addr_in(xprt);
265 		snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
266 		break;
267 	case AF_INET6:
268 		(void)rpc_ntop(sap, buf, sizeof(buf));
269 		xprt->address_strings[RPC_DISPLAY_ADDR] =
270 						kstrdup(buf, GFP_KERNEL);
271 		sin6 = xs_addr_in6(xprt);
272 		snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
273 		break;
274 	default:
275 		BUG();
276 	}
277 
278 	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
279 }
280 
281 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
282 {
283 	struct sockaddr *sap = xs_addr(xprt);
284 	char buf[128];
285 
286 	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
287 	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
288 
289 	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
290 	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
291 }
292 
293 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
294 				     const char *protocol,
295 				     const char *netid)
296 {
297 	xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
298 	xprt->address_strings[RPC_DISPLAY_NETID] = netid;
299 	xs_format_common_peer_addresses(xprt);
300 	xs_format_common_peer_ports(xprt);
301 }
302 
303 static void xs_update_peer_port(struct rpc_xprt *xprt)
304 {
305 	kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
306 	kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
307 
308 	xs_format_common_peer_ports(xprt);
309 }
310 
311 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
312 {
313 	unsigned int i;
314 
315 	for (i = 0; i < RPC_DISPLAY_MAX; i++)
316 		switch (i) {
317 		case RPC_DISPLAY_PROTO:
318 		case RPC_DISPLAY_NETID:
319 			continue;
320 		default:
321 			kfree(xprt->address_strings[i]);
322 		}
323 }
324 
325 #define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
326 
327 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
328 {
329 	struct msghdr msg = {
330 		.msg_name	= addr,
331 		.msg_namelen	= addrlen,
332 		.msg_flags	= XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
333 	};
334 	struct kvec iov = {
335 		.iov_base	= vec->iov_base + base,
336 		.iov_len	= vec->iov_len - base,
337 	};
338 
339 	if (iov.iov_len != 0)
340 		return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
341 	return kernel_sendmsg(sock, &msg, NULL, 0, 0);
342 }
343 
344 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
345 {
346 	ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
347 			int offset, size_t size, int flags);
348 	struct page **ppage;
349 	unsigned int remainder;
350 	int err;
351 
352 	remainder = xdr->page_len - base;
353 	base += xdr->page_base;
354 	ppage = xdr->pages + (base >> PAGE_SHIFT);
355 	base &= ~PAGE_MASK;
356 	do_sendpage = sock->ops->sendpage;
357 	if (!zerocopy)
358 		do_sendpage = sock_no_sendpage;
359 	for(;;) {
360 		unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
361 		int flags = XS_SENDMSG_FLAGS;
362 
363 		remainder -= len;
364 		if (more)
365 			flags |= MSG_MORE;
366 		if (remainder != 0)
367 			flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
368 		err = do_sendpage(sock, *ppage, base, len, flags);
369 		if (remainder == 0 || err != len)
370 			break;
371 		*sent_p += err;
372 		ppage++;
373 		base = 0;
374 	}
375 	if (err > 0) {
376 		*sent_p += err;
377 		err = 0;
378 	}
379 	return err;
380 }
381 
382 /**
383  * xs_sendpages - write pages directly to a socket
384  * @sock: socket to send on
385  * @addr: UDP only -- address of destination
386  * @addrlen: UDP only -- length of destination address
387  * @xdr: buffer containing this request
388  * @base: starting position in the buffer
389  * @zerocopy: true if it is safe to use sendpage()
390  * @sent_p: return the total number of bytes successfully queued for sending
391  *
392  */
393 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
394 {
395 	unsigned int remainder = xdr->len - base;
396 	int err = 0;
397 	int sent = 0;
398 
399 	if (unlikely(!sock))
400 		return -ENOTSOCK;
401 
402 	if (base != 0) {
403 		addr = NULL;
404 		addrlen = 0;
405 	}
406 
407 	if (base < xdr->head[0].iov_len || addr != NULL) {
408 		unsigned int len = xdr->head[0].iov_len - base;
409 		remainder -= len;
410 		err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
411 		if (remainder == 0 || err != len)
412 			goto out;
413 		*sent_p += err;
414 		base = 0;
415 	} else
416 		base -= xdr->head[0].iov_len;
417 
418 	if (base < xdr->page_len) {
419 		unsigned int len = xdr->page_len - base;
420 		remainder -= len;
421 		err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
422 		*sent_p += sent;
423 		if (remainder == 0 || sent != len)
424 			goto out;
425 		base = 0;
426 	} else
427 		base -= xdr->page_len;
428 
429 	if (base >= xdr->tail[0].iov_len)
430 		return 0;
431 	err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
432 out:
433 	if (err > 0) {
434 		*sent_p += err;
435 		err = 0;
436 	}
437 	return err;
438 }
439 
440 static void xs_nospace_callback(struct rpc_task *task)
441 {
442 	struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
443 
444 	transport->inet->sk_write_pending--;
445 }
446 
447 /**
448  * xs_nospace - place task on wait queue if transmit was incomplete
449  * @task: task to put to sleep
450  *
451  */
452 static int xs_nospace(struct rpc_task *task)
453 {
454 	struct rpc_rqst *req = task->tk_rqstp;
455 	struct rpc_xprt *xprt = req->rq_xprt;
456 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
457 	struct sock *sk = transport->inet;
458 	int ret = -EAGAIN;
459 
460 	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
461 			task->tk_pid, req->rq_slen - req->rq_bytes_sent,
462 			req->rq_slen);
463 
464 	/* Protect against races with write_space */
465 	spin_lock_bh(&xprt->transport_lock);
466 
467 	/* Don't race with disconnect */
468 	if (xprt_connected(xprt)) {
469 		/* wait for more buffer space */
470 		sk->sk_write_pending++;
471 		xprt_wait_for_buffer_space(task, xs_nospace_callback);
472 	} else
473 		ret = -ENOTCONN;
474 
475 	spin_unlock_bh(&xprt->transport_lock);
476 
477 	/* Race breaker in case memory is freed before above code is called */
478 	if (ret == -EAGAIN) {
479 		struct socket_wq *wq;
480 
481 		rcu_read_lock();
482 		wq = rcu_dereference(sk->sk_wq);
483 		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
484 		rcu_read_unlock();
485 
486 		sk->sk_write_space(sk);
487 	}
488 	return ret;
489 }
490 
491 /*
492  * Construct a stream transport record marker in @buf.
493  */
494 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
495 {
496 	u32 reclen = buf->len - sizeof(rpc_fraghdr);
497 	rpc_fraghdr *base = buf->head[0].iov_base;
498 	*base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
499 }
500 
501 /**
502  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
503  * @task: RPC task that manages the state of an RPC request
504  *
505  * Return values:
506  *        0:	The request has been sent
507  *   EAGAIN:	The socket was blocked, please call again later to
508  *		complete the request
509  * ENOTCONN:	Caller needs to invoke connect logic then call again
510  *    other:	Some other error occured, the request was not sent
511  */
512 static int xs_local_send_request(struct rpc_task *task)
513 {
514 	struct rpc_rqst *req = task->tk_rqstp;
515 	struct rpc_xprt *xprt = req->rq_xprt;
516 	struct sock_xprt *transport =
517 				container_of(xprt, struct sock_xprt, xprt);
518 	struct xdr_buf *xdr = &req->rq_snd_buf;
519 	int status;
520 	int sent = 0;
521 
522 	xs_encode_stream_record_marker(&req->rq_snd_buf);
523 
524 	xs_pktdump("packet data:",
525 			req->rq_svec->iov_base, req->rq_svec->iov_len);
526 
527 	status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
528 			      true, &sent);
529 	dprintk("RPC:       %s(%u) = %d\n",
530 			__func__, xdr->len - req->rq_bytes_sent, status);
531 
532 	if (status == -EAGAIN && sock_writeable(transport->inet))
533 		status = -ENOBUFS;
534 
535 	if (likely(sent > 0) || status == 0) {
536 		req->rq_bytes_sent += sent;
537 		req->rq_xmit_bytes_sent += sent;
538 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
539 			req->rq_bytes_sent = 0;
540 			return 0;
541 		}
542 		status = -EAGAIN;
543 	}
544 
545 	switch (status) {
546 	case -ENOBUFS:
547 		break;
548 	case -EAGAIN:
549 		status = xs_nospace(task);
550 		break;
551 	default:
552 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
553 			-status);
554 	case -EPIPE:
555 		xs_close(xprt);
556 		status = -ENOTCONN;
557 	}
558 
559 	return status;
560 }
561 
562 /**
563  * xs_udp_send_request - write an RPC request to a UDP socket
564  * @task: address of RPC task that manages the state of an RPC request
565  *
566  * Return values:
567  *        0:	The request has been sent
568  *   EAGAIN:	The socket was blocked, please call again later to
569  *		complete the request
570  * ENOTCONN:	Caller needs to invoke connect logic then call again
571  *    other:	Some other error occurred, the request was not sent
572  */
573 static int xs_udp_send_request(struct rpc_task *task)
574 {
575 	struct rpc_rqst *req = task->tk_rqstp;
576 	struct rpc_xprt *xprt = req->rq_xprt;
577 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
578 	struct xdr_buf *xdr = &req->rq_snd_buf;
579 	int sent = 0;
580 	int status;
581 
582 	xs_pktdump("packet data:",
583 				req->rq_svec->iov_base,
584 				req->rq_svec->iov_len);
585 
586 	if (!xprt_bound(xprt))
587 		return -ENOTCONN;
588 	status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
589 			      xdr, req->rq_bytes_sent, true, &sent);
590 
591 	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
592 			xdr->len - req->rq_bytes_sent, status);
593 
594 	/* firewall is blocking us, don't return -EAGAIN or we end up looping */
595 	if (status == -EPERM)
596 		goto process_status;
597 
598 	if (status == -EAGAIN && sock_writeable(transport->inet))
599 		status = -ENOBUFS;
600 
601 	if (sent > 0 || status == 0) {
602 		req->rq_xmit_bytes_sent += sent;
603 		if (sent >= req->rq_slen)
604 			return 0;
605 		/* Still some bytes left; set up for a retry later. */
606 		status = -EAGAIN;
607 	}
608 
609 process_status:
610 	switch (status) {
611 	case -ENOTSOCK:
612 		status = -ENOTCONN;
613 		/* Should we call xs_close() here? */
614 		break;
615 	case -EAGAIN:
616 		status = xs_nospace(task);
617 		break;
618 	case -ENETUNREACH:
619 	case -ENOBUFS:
620 	case -EPIPE:
621 	case -ECONNREFUSED:
622 	case -EPERM:
623 		/* When the server has died, an ICMP port unreachable message
624 		 * prompts ECONNREFUSED. */
625 		break;
626 	default:
627 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
628 			-status);
629 	}
630 
631 	return status;
632 }
633 
634 /**
635  * xs_tcp_send_request - write an RPC request to a TCP socket
636  * @task: address of RPC task that manages the state of an RPC request
637  *
638  * Return values:
639  *        0:	The request has been sent
640  *   EAGAIN:	The socket was blocked, please call again later to
641  *		complete the request
642  * ENOTCONN:	Caller needs to invoke connect logic then call again
643  *    other:	Some other error occurred, the request was not sent
644  *
645  * XXX: In the case of soft timeouts, should we eventually give up
646  *	if sendmsg is not able to make progress?
647  */
648 static int xs_tcp_send_request(struct rpc_task *task)
649 {
650 	struct rpc_rqst *req = task->tk_rqstp;
651 	struct rpc_xprt *xprt = req->rq_xprt;
652 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
653 	struct xdr_buf *xdr = &req->rq_snd_buf;
654 	bool zerocopy = true;
655 	bool vm_wait = false;
656 	int status;
657 	int sent;
658 
659 	xs_encode_stream_record_marker(&req->rq_snd_buf);
660 
661 	xs_pktdump("packet data:",
662 				req->rq_svec->iov_base,
663 				req->rq_svec->iov_len);
664 	/* Don't use zero copy if this is a resend. If the RPC call
665 	 * completes while the socket holds a reference to the pages,
666 	 * then we may end up resending corrupted data.
667 	 */
668 	if (task->tk_flags & RPC_TASK_SENT)
669 		zerocopy = false;
670 
671 	if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
672 		xs_tcp_set_socket_timeouts(xprt, transport->sock);
673 
674 	/* Continue transmitting the packet/record. We must be careful
675 	 * to cope with writespace callbacks arriving _after_ we have
676 	 * called sendmsg(). */
677 	while (1) {
678 		sent = 0;
679 		status = xs_sendpages(transport->sock, NULL, 0, xdr,
680 				      req->rq_bytes_sent, zerocopy, &sent);
681 
682 		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
683 				xdr->len - req->rq_bytes_sent, status);
684 
685 		/* If we've sent the entire packet, immediately
686 		 * reset the count of bytes sent. */
687 		req->rq_bytes_sent += sent;
688 		req->rq_xmit_bytes_sent += sent;
689 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
690 			req->rq_bytes_sent = 0;
691 			return 0;
692 		}
693 
694 		WARN_ON_ONCE(sent == 0 && status == 0);
695 
696 		if (status == -EAGAIN ) {
697 			/*
698 			 * Return EAGAIN if we're sure we're hitting the
699 			 * socket send buffer limits.
700 			 */
701 			if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
702 				break;
703 			/*
704 			 * Did we hit a memory allocation failure?
705 			 */
706 			if (sent == 0) {
707 				status = -ENOBUFS;
708 				if (vm_wait)
709 					break;
710 				/* Retry, knowing now that we're below the
711 				 * socket send buffer limit
712 				 */
713 				vm_wait = true;
714 			}
715 			continue;
716 		}
717 		if (status < 0)
718 			break;
719 		vm_wait = false;
720 	}
721 
722 	switch (status) {
723 	case -ENOTSOCK:
724 		status = -ENOTCONN;
725 		/* Should we call xs_close() here? */
726 		break;
727 	case -EAGAIN:
728 		status = xs_nospace(task);
729 		break;
730 	case -ECONNRESET:
731 	case -ECONNREFUSED:
732 	case -ENOTCONN:
733 	case -EADDRINUSE:
734 	case -ENOBUFS:
735 	case -EPIPE:
736 		break;
737 	default:
738 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
739 			-status);
740 	}
741 
742 	return status;
743 }
744 
745 /**
746  * xs_tcp_release_xprt - clean up after a tcp transmission
747  * @xprt: transport
748  * @task: rpc task
749  *
750  * This cleans up if an error causes us to abort the transmission of a request.
751  * In this case, the socket may need to be reset in order to avoid confusing
752  * the server.
753  */
754 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
755 {
756 	struct rpc_rqst *req;
757 
758 	if (task != xprt->snd_task)
759 		return;
760 	if (task == NULL)
761 		goto out_release;
762 	req = task->tk_rqstp;
763 	if (req == NULL)
764 		goto out_release;
765 	if (req->rq_bytes_sent == 0)
766 		goto out_release;
767 	if (req->rq_bytes_sent == req->rq_snd_buf.len)
768 		goto out_release;
769 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
770 out_release:
771 	xprt_release_xprt(xprt, task);
772 }
773 
774 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
775 {
776 	transport->old_data_ready = sk->sk_data_ready;
777 	transport->old_state_change = sk->sk_state_change;
778 	transport->old_write_space = sk->sk_write_space;
779 	transport->old_error_report = sk->sk_error_report;
780 }
781 
782 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
783 {
784 	sk->sk_data_ready = transport->old_data_ready;
785 	sk->sk_state_change = transport->old_state_change;
786 	sk->sk_write_space = transport->old_write_space;
787 	sk->sk_error_report = transport->old_error_report;
788 }
789 
790 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
791 {
792 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
793 
794 	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
795 }
796 
797 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
798 {
799 	smp_mb__before_atomic();
800 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
801 	clear_bit(XPRT_CLOSING, &xprt->state);
802 	xs_sock_reset_state_flags(xprt);
803 	smp_mb__after_atomic();
804 }
805 
806 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
807 {
808 	xs_sock_reset_connection_flags(xprt);
809 	/* Mark transport as closed and wake up all pending tasks */
810 	xprt_disconnect_done(xprt);
811 }
812 
813 /**
814  * xs_error_report - callback to handle TCP socket state errors
815  * @sk: socket
816  *
817  * Note: we don't call sock_error() since there may be a rpc_task
818  * using the socket, and so we don't want to clear sk->sk_err.
819  */
820 static void xs_error_report(struct sock *sk)
821 {
822 	struct rpc_xprt *xprt;
823 	int err;
824 
825 	read_lock_bh(&sk->sk_callback_lock);
826 	if (!(xprt = xprt_from_sock(sk)))
827 		goto out;
828 
829 	err = -sk->sk_err;
830 	if (err == 0)
831 		goto out;
832 	/* Is this a reset event? */
833 	if (sk->sk_state == TCP_CLOSE)
834 		xs_sock_mark_closed(xprt);
835 	dprintk("RPC:       xs_error_report client %p, error=%d...\n",
836 			xprt, -err);
837 	trace_rpc_socket_error(xprt, sk->sk_socket, err);
838 	xprt_wake_pending_tasks(xprt, err);
839  out:
840 	read_unlock_bh(&sk->sk_callback_lock);
841 }
842 
843 static void xs_reset_transport(struct sock_xprt *transport)
844 {
845 	struct socket *sock = transport->sock;
846 	struct sock *sk = transport->inet;
847 	struct rpc_xprt *xprt = &transport->xprt;
848 
849 	if (sk == NULL)
850 		return;
851 
852 	if (atomic_read(&transport->xprt.swapper))
853 		sk_clear_memalloc(sk);
854 
855 	kernel_sock_shutdown(sock, SHUT_RDWR);
856 
857 	mutex_lock(&transport->recv_mutex);
858 	write_lock_bh(&sk->sk_callback_lock);
859 	transport->inet = NULL;
860 	transport->sock = NULL;
861 
862 	sk->sk_user_data = NULL;
863 
864 	xs_restore_old_callbacks(transport, sk);
865 	xprt_clear_connected(xprt);
866 	write_unlock_bh(&sk->sk_callback_lock);
867 	xs_sock_reset_connection_flags(xprt);
868 	mutex_unlock(&transport->recv_mutex);
869 
870 	trace_rpc_socket_close(xprt, sock);
871 	sock_release(sock);
872 }
873 
874 /**
875  * xs_close - close a socket
876  * @xprt: transport
877  *
878  * This is used when all requests are complete; ie, no DRC state remains
879  * on the server we want to save.
880  *
881  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
882  * xs_reset_transport() zeroing the socket from underneath a writer.
883  */
884 static void xs_close(struct rpc_xprt *xprt)
885 {
886 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
887 
888 	dprintk("RPC:       xs_close xprt %p\n", xprt);
889 
890 	xs_reset_transport(transport);
891 	xprt->reestablish_timeout = 0;
892 
893 	xprt_disconnect_done(xprt);
894 }
895 
896 static void xs_inject_disconnect(struct rpc_xprt *xprt)
897 {
898 	dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
899 		xprt);
900 	xprt_disconnect_done(xprt);
901 }
902 
903 static void xs_xprt_free(struct rpc_xprt *xprt)
904 {
905 	xs_free_peer_addresses(xprt);
906 	xprt_free(xprt);
907 }
908 
909 /**
910  * xs_destroy - prepare to shutdown a transport
911  * @xprt: doomed transport
912  *
913  */
914 static void xs_destroy(struct rpc_xprt *xprt)
915 {
916 	struct sock_xprt *transport = container_of(xprt,
917 			struct sock_xprt, xprt);
918 	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
919 
920 	cancel_delayed_work_sync(&transport->connect_worker);
921 	xs_close(xprt);
922 	cancel_work_sync(&transport->recv_worker);
923 	xs_xprt_free(xprt);
924 	module_put(THIS_MODULE);
925 }
926 
927 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
928 {
929 	struct xdr_skb_reader desc = {
930 		.skb		= skb,
931 		.offset		= sizeof(rpc_fraghdr),
932 		.count		= skb->len - sizeof(rpc_fraghdr),
933 	};
934 
935 	if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
936 		return -1;
937 	if (desc.count)
938 		return -1;
939 	return 0;
940 }
941 
942 /**
943  * xs_local_data_read_skb
944  * @xprt: transport
945  * @sk: socket
946  * @skb: skbuff
947  *
948  * Currently this assumes we can read the whole reply in a single gulp.
949  */
950 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
951 		struct sock *sk,
952 		struct sk_buff *skb)
953 {
954 	struct rpc_task *task;
955 	struct rpc_rqst *rovr;
956 	int repsize, copied;
957 	u32 _xid;
958 	__be32 *xp;
959 
960 	repsize = skb->len - sizeof(rpc_fraghdr);
961 	if (repsize < 4) {
962 		dprintk("RPC:       impossible RPC reply size %d\n", repsize);
963 		return;
964 	}
965 
966 	/* Copy the XID from the skb... */
967 	xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
968 	if (xp == NULL)
969 		return;
970 
971 	/* Look up and lock the request corresponding to the given XID */
972 	spin_lock_bh(&xprt->transport_lock);
973 	rovr = xprt_lookup_rqst(xprt, *xp);
974 	if (!rovr)
975 		goto out_unlock;
976 	task = rovr->rq_task;
977 
978 	copied = rovr->rq_private_buf.buflen;
979 	if (copied > repsize)
980 		copied = repsize;
981 
982 	if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
983 		dprintk("RPC:       sk_buff copy failed\n");
984 		goto out_unlock;
985 	}
986 
987 	xprt_complete_rqst(task, copied);
988 
989  out_unlock:
990 	spin_unlock_bh(&xprt->transport_lock);
991 }
992 
993 static void xs_local_data_receive(struct sock_xprt *transport)
994 {
995 	struct sk_buff *skb;
996 	struct sock *sk;
997 	int err;
998 
999 	mutex_lock(&transport->recv_mutex);
1000 	sk = transport->inet;
1001 	if (sk == NULL)
1002 		goto out;
1003 	for (;;) {
1004 		skb = skb_recv_datagram(sk, 0, 1, &err);
1005 		if (skb != NULL) {
1006 			xs_local_data_read_skb(&transport->xprt, sk, skb);
1007 			skb_free_datagram(sk, skb);
1008 			continue;
1009 		}
1010 		if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1011 			break;
1012 	}
1013 out:
1014 	mutex_unlock(&transport->recv_mutex);
1015 }
1016 
1017 static void xs_local_data_receive_workfn(struct work_struct *work)
1018 {
1019 	struct sock_xprt *transport =
1020 		container_of(work, struct sock_xprt, recv_worker);
1021 	xs_local_data_receive(transport);
1022 }
1023 
1024 /**
1025  * xs_udp_data_read_skb - receive callback for UDP sockets
1026  * @xprt: transport
1027  * @sk: socket
1028  * @skb: skbuff
1029  *
1030  */
1031 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1032 		struct sock *sk,
1033 		struct sk_buff *skb)
1034 {
1035 	struct rpc_task *task;
1036 	struct rpc_rqst *rovr;
1037 	int repsize, copied;
1038 	u32 _xid;
1039 	__be32 *xp;
1040 
1041 	repsize = skb->len;
1042 	if (repsize < 4) {
1043 		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1044 		return;
1045 	}
1046 
1047 	/* Copy the XID from the skb... */
1048 	xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1049 	if (xp == NULL)
1050 		return;
1051 
1052 	/* Look up and lock the request corresponding to the given XID */
1053 	spin_lock_bh(&xprt->transport_lock);
1054 	rovr = xprt_lookup_rqst(xprt, *xp);
1055 	if (!rovr)
1056 		goto out_unlock;
1057 	task = rovr->rq_task;
1058 
1059 	if ((copied = rovr->rq_private_buf.buflen) > repsize)
1060 		copied = repsize;
1061 
1062 	/* Suck it into the iovec, verify checksum if not done by hw. */
1063 	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1064 		__UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1065 		goto out_unlock;
1066 	}
1067 
1068 	__UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1069 
1070 	xprt_adjust_cwnd(xprt, task, copied);
1071 	xprt_complete_rqst(task, copied);
1072 
1073  out_unlock:
1074 	spin_unlock_bh(&xprt->transport_lock);
1075 }
1076 
1077 static void xs_udp_data_receive(struct sock_xprt *transport)
1078 {
1079 	struct sk_buff *skb;
1080 	struct sock *sk;
1081 	int err;
1082 
1083 	mutex_lock(&transport->recv_mutex);
1084 	sk = transport->inet;
1085 	if (sk == NULL)
1086 		goto out;
1087 	for (;;) {
1088 		skb = skb_recv_udp(sk, 0, 1, &err);
1089 		if (skb != NULL) {
1090 			xs_udp_data_read_skb(&transport->xprt, sk, skb);
1091 			consume_skb(skb);
1092 			continue;
1093 		}
1094 		if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1095 			break;
1096 	}
1097 out:
1098 	mutex_unlock(&transport->recv_mutex);
1099 }
1100 
1101 static void xs_udp_data_receive_workfn(struct work_struct *work)
1102 {
1103 	struct sock_xprt *transport =
1104 		container_of(work, struct sock_xprt, recv_worker);
1105 	xs_udp_data_receive(transport);
1106 }
1107 
1108 /**
1109  * xs_data_ready - "data ready" callback for UDP sockets
1110  * @sk: socket with data to read
1111  *
1112  */
1113 static void xs_data_ready(struct sock *sk)
1114 {
1115 	struct rpc_xprt *xprt;
1116 
1117 	read_lock_bh(&sk->sk_callback_lock);
1118 	dprintk("RPC:       xs_data_ready...\n");
1119 	xprt = xprt_from_sock(sk);
1120 	if (xprt != NULL) {
1121 		struct sock_xprt *transport = container_of(xprt,
1122 				struct sock_xprt, xprt);
1123 		transport->old_data_ready(sk);
1124 		/* Any data means we had a useful conversation, so
1125 		 * then we don't need to delay the next reconnect
1126 		 */
1127 		if (xprt->reestablish_timeout)
1128 			xprt->reestablish_timeout = 0;
1129 		if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1130 			queue_work(xprtiod_workqueue, &transport->recv_worker);
1131 	}
1132 	read_unlock_bh(&sk->sk_callback_lock);
1133 }
1134 
1135 /*
1136  * Helper function to force a TCP close if the server is sending
1137  * junk and/or it has put us in CLOSE_WAIT
1138  */
1139 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1140 {
1141 	xprt_force_disconnect(xprt);
1142 }
1143 
1144 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1145 {
1146 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1147 	size_t len, used;
1148 	char *p;
1149 
1150 	p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1151 	len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1152 	used = xdr_skb_read_bits(desc, p, len);
1153 	transport->tcp_offset += used;
1154 	if (used != len)
1155 		return;
1156 
1157 	transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1158 	if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1159 		transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1160 	else
1161 		transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1162 	transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1163 
1164 	transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1165 	transport->tcp_offset = 0;
1166 
1167 	/* Sanity check of the record length */
1168 	if (unlikely(transport->tcp_reclen < 8)) {
1169 		dprintk("RPC:       invalid TCP record fragment length\n");
1170 		xs_tcp_force_close(xprt);
1171 		return;
1172 	}
1173 	dprintk("RPC:       reading TCP record fragment of length %d\n",
1174 			transport->tcp_reclen);
1175 }
1176 
1177 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1178 {
1179 	if (transport->tcp_offset == transport->tcp_reclen) {
1180 		transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1181 		transport->tcp_offset = 0;
1182 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1183 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1184 			transport->tcp_flags |= TCP_RCV_COPY_XID;
1185 			transport->tcp_copied = 0;
1186 		}
1187 	}
1188 }
1189 
1190 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1191 {
1192 	size_t len, used;
1193 	char *p;
1194 
1195 	len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1196 	dprintk("RPC:       reading XID (%zu bytes)\n", len);
1197 	p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1198 	used = xdr_skb_read_bits(desc, p, len);
1199 	transport->tcp_offset += used;
1200 	if (used != len)
1201 		return;
1202 	transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1203 	transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1204 	transport->tcp_copied = 4;
1205 	dprintk("RPC:       reading %s XID %08x\n",
1206 			(transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1207 							      : "request with",
1208 			ntohl(transport->tcp_xid));
1209 	xs_tcp_check_fraghdr(transport);
1210 }
1211 
1212 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1213 				       struct xdr_skb_reader *desc)
1214 {
1215 	size_t len, used;
1216 	u32 offset;
1217 	char *p;
1218 
1219 	/*
1220 	 * We want transport->tcp_offset to be 8 at the end of this routine
1221 	 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1222 	 * When this function is called for the first time,
1223 	 * transport->tcp_offset is 4 (after having already read the xid).
1224 	 */
1225 	offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1226 	len = sizeof(transport->tcp_calldir) - offset;
1227 	dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1228 	p = ((char *) &transport->tcp_calldir) + offset;
1229 	used = xdr_skb_read_bits(desc, p, len);
1230 	transport->tcp_offset += used;
1231 	if (used != len)
1232 		return;
1233 	transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1234 	/*
1235 	 * We don't yet have the XDR buffer, so we will write the calldir
1236 	 * out after we get the buffer from the 'struct rpc_rqst'
1237 	 */
1238 	switch (ntohl(transport->tcp_calldir)) {
1239 	case RPC_REPLY:
1240 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1241 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1242 		transport->tcp_flags |= TCP_RPC_REPLY;
1243 		break;
1244 	case RPC_CALL:
1245 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1246 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1247 		transport->tcp_flags &= ~TCP_RPC_REPLY;
1248 		break;
1249 	default:
1250 		dprintk("RPC:       invalid request message type\n");
1251 		xs_tcp_force_close(&transport->xprt);
1252 	}
1253 	xs_tcp_check_fraghdr(transport);
1254 }
1255 
1256 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1257 				     struct xdr_skb_reader *desc,
1258 				     struct rpc_rqst *req)
1259 {
1260 	struct sock_xprt *transport =
1261 				container_of(xprt, struct sock_xprt, xprt);
1262 	struct xdr_buf *rcvbuf;
1263 	size_t len;
1264 	ssize_t r;
1265 
1266 	rcvbuf = &req->rq_private_buf;
1267 
1268 	if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1269 		/*
1270 		 * Save the RPC direction in the XDR buffer
1271 		 */
1272 		memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1273 			&transport->tcp_calldir,
1274 			sizeof(transport->tcp_calldir));
1275 		transport->tcp_copied += sizeof(transport->tcp_calldir);
1276 		transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1277 	}
1278 
1279 	len = desc->count;
1280 	if (len > transport->tcp_reclen - transport->tcp_offset) {
1281 		struct xdr_skb_reader my_desc;
1282 
1283 		len = transport->tcp_reclen - transport->tcp_offset;
1284 		memcpy(&my_desc, desc, sizeof(my_desc));
1285 		my_desc.count = len;
1286 		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1287 					  &my_desc, xdr_skb_read_bits);
1288 		desc->count -= r;
1289 		desc->offset += r;
1290 	} else
1291 		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1292 					  desc, xdr_skb_read_bits);
1293 
1294 	if (r > 0) {
1295 		transport->tcp_copied += r;
1296 		transport->tcp_offset += r;
1297 	}
1298 	if (r != len) {
1299 		/* Error when copying to the receive buffer,
1300 		 * usually because we weren't able to allocate
1301 		 * additional buffer pages. All we can do now
1302 		 * is turn off TCP_RCV_COPY_DATA, so the request
1303 		 * will not receive any additional updates,
1304 		 * and time out.
1305 		 * Any remaining data from this record will
1306 		 * be discarded.
1307 		 */
1308 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1309 		dprintk("RPC:       XID %08x truncated request\n",
1310 				ntohl(transport->tcp_xid));
1311 		dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1312 				"tcp_offset = %u, tcp_reclen = %u\n",
1313 				xprt, transport->tcp_copied,
1314 				transport->tcp_offset, transport->tcp_reclen);
1315 		return;
1316 	}
1317 
1318 	dprintk("RPC:       XID %08x read %zd bytes\n",
1319 			ntohl(transport->tcp_xid), r);
1320 	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1321 			"tcp_reclen = %u\n", xprt, transport->tcp_copied,
1322 			transport->tcp_offset, transport->tcp_reclen);
1323 
1324 	if (transport->tcp_copied == req->rq_private_buf.buflen)
1325 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1326 	else if (transport->tcp_offset == transport->tcp_reclen) {
1327 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1328 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1329 	}
1330 }
1331 
1332 /*
1333  * Finds the request corresponding to the RPC xid and invokes the common
1334  * tcp read code to read the data.
1335  */
1336 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1337 				    struct xdr_skb_reader *desc)
1338 {
1339 	struct sock_xprt *transport =
1340 				container_of(xprt, struct sock_xprt, xprt);
1341 	struct rpc_rqst *req;
1342 
1343 	dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1344 
1345 	/* Find and lock the request corresponding to this xid */
1346 	spin_lock_bh(&xprt->transport_lock);
1347 	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1348 	if (!req) {
1349 		dprintk("RPC:       XID %08x request not found!\n",
1350 				ntohl(transport->tcp_xid));
1351 		spin_unlock_bh(&xprt->transport_lock);
1352 		return -1;
1353 	}
1354 
1355 	xs_tcp_read_common(xprt, desc, req);
1356 
1357 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1358 		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1359 
1360 	spin_unlock_bh(&xprt->transport_lock);
1361 	return 0;
1362 }
1363 
1364 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1365 /*
1366  * Obtains an rpc_rqst previously allocated and invokes the common
1367  * tcp read code to read the data.  The result is placed in the callback
1368  * queue.
1369  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1370  * connection and return -1.
1371  */
1372 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1373 				       struct xdr_skb_reader *desc)
1374 {
1375 	struct sock_xprt *transport =
1376 				container_of(xprt, struct sock_xprt, xprt);
1377 	struct rpc_rqst *req;
1378 
1379 	/* Look up and lock the request corresponding to the given XID */
1380 	spin_lock_bh(&xprt->transport_lock);
1381 	req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1382 	if (req == NULL) {
1383 		spin_unlock_bh(&xprt->transport_lock);
1384 		printk(KERN_WARNING "Callback slot table overflowed\n");
1385 		xprt_force_disconnect(xprt);
1386 		return -1;
1387 	}
1388 
1389 	dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1390 	xs_tcp_read_common(xprt, desc, req);
1391 
1392 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1393 		xprt_complete_bc_request(req, transport->tcp_copied);
1394 	spin_unlock_bh(&xprt->transport_lock);
1395 
1396 	return 0;
1397 }
1398 
1399 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1400 					struct xdr_skb_reader *desc)
1401 {
1402 	struct sock_xprt *transport =
1403 				container_of(xprt, struct sock_xprt, xprt);
1404 
1405 	return (transport->tcp_flags & TCP_RPC_REPLY) ?
1406 		xs_tcp_read_reply(xprt, desc) :
1407 		xs_tcp_read_callback(xprt, desc);
1408 }
1409 
1410 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1411 {
1412 	int ret;
1413 
1414 	ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1415 			      SVC_SOCK_ANONYMOUS);
1416 	if (ret < 0)
1417 		return ret;
1418 	return 0;
1419 }
1420 
1421 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1422 {
1423 	return PAGE_SIZE;
1424 }
1425 #else
1426 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1427 					struct xdr_skb_reader *desc)
1428 {
1429 	return xs_tcp_read_reply(xprt, desc);
1430 }
1431 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1432 
1433 /*
1434  * Read data off the transport.  This can be either an RPC_CALL or an
1435  * RPC_REPLY.  Relay the processing to helper functions.
1436  */
1437 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1438 				    struct xdr_skb_reader *desc)
1439 {
1440 	struct sock_xprt *transport =
1441 				container_of(xprt, struct sock_xprt, xprt);
1442 
1443 	if (_xs_tcp_read_data(xprt, desc) == 0)
1444 		xs_tcp_check_fraghdr(transport);
1445 	else {
1446 		/*
1447 		 * The transport_lock protects the request handling.
1448 		 * There's no need to hold it to update the tcp_flags.
1449 		 */
1450 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1451 	}
1452 }
1453 
1454 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1455 {
1456 	size_t len;
1457 
1458 	len = transport->tcp_reclen - transport->tcp_offset;
1459 	if (len > desc->count)
1460 		len = desc->count;
1461 	desc->count -= len;
1462 	desc->offset += len;
1463 	transport->tcp_offset += len;
1464 	dprintk("RPC:       discarded %zu bytes\n", len);
1465 	xs_tcp_check_fraghdr(transport);
1466 }
1467 
1468 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1469 {
1470 	struct rpc_xprt *xprt = rd_desc->arg.data;
1471 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1472 	struct xdr_skb_reader desc = {
1473 		.skb	= skb,
1474 		.offset	= offset,
1475 		.count	= len,
1476 	};
1477 
1478 	dprintk("RPC:       xs_tcp_data_recv started\n");
1479 	do {
1480 		trace_xs_tcp_data_recv(transport);
1481 		/* Read in a new fragment marker if necessary */
1482 		/* Can we ever really expect to get completely empty fragments? */
1483 		if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1484 			xs_tcp_read_fraghdr(xprt, &desc);
1485 			continue;
1486 		}
1487 		/* Read in the xid if necessary */
1488 		if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1489 			xs_tcp_read_xid(transport, &desc);
1490 			continue;
1491 		}
1492 		/* Read in the call/reply flag */
1493 		if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1494 			xs_tcp_read_calldir(transport, &desc);
1495 			continue;
1496 		}
1497 		/* Read in the request data */
1498 		if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1499 			xs_tcp_read_data(xprt, &desc);
1500 			continue;
1501 		}
1502 		/* Skip over any trailing bytes on short reads */
1503 		xs_tcp_read_discard(transport, &desc);
1504 	} while (desc.count);
1505 	trace_xs_tcp_data_recv(transport);
1506 	dprintk("RPC:       xs_tcp_data_recv done\n");
1507 	return len - desc.count;
1508 }
1509 
1510 static void xs_tcp_data_receive(struct sock_xprt *transport)
1511 {
1512 	struct rpc_xprt *xprt = &transport->xprt;
1513 	struct sock *sk;
1514 	read_descriptor_t rd_desc = {
1515 		.count = 2*1024*1024,
1516 		.arg.data = xprt,
1517 	};
1518 	unsigned long total = 0;
1519 	int read = 0;
1520 
1521 	mutex_lock(&transport->recv_mutex);
1522 	sk = transport->inet;
1523 	if (sk == NULL)
1524 		goto out;
1525 
1526 	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1527 	for (;;) {
1528 		lock_sock(sk);
1529 		read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1530 		if (read <= 0) {
1531 			clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1532 			release_sock(sk);
1533 			if (!test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1534 				break;
1535 		} else {
1536 			release_sock(sk);
1537 			total += read;
1538 		}
1539 		rd_desc.count = 65536;
1540 	}
1541 out:
1542 	mutex_unlock(&transport->recv_mutex);
1543 	trace_xs_tcp_data_ready(xprt, read, total);
1544 }
1545 
1546 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1547 {
1548 	struct sock_xprt *transport =
1549 		container_of(work, struct sock_xprt, recv_worker);
1550 	xs_tcp_data_receive(transport);
1551 }
1552 
1553 /**
1554  * xs_tcp_state_change - callback to handle TCP socket state changes
1555  * @sk: socket whose state has changed
1556  *
1557  */
1558 static void xs_tcp_state_change(struct sock *sk)
1559 {
1560 	struct rpc_xprt *xprt;
1561 	struct sock_xprt *transport;
1562 
1563 	read_lock_bh(&sk->sk_callback_lock);
1564 	if (!(xprt = xprt_from_sock(sk)))
1565 		goto out;
1566 	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1567 	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1568 			sk->sk_state, xprt_connected(xprt),
1569 			sock_flag(sk, SOCK_DEAD),
1570 			sock_flag(sk, SOCK_ZAPPED),
1571 			sk->sk_shutdown);
1572 
1573 	transport = container_of(xprt, struct sock_xprt, xprt);
1574 	trace_rpc_socket_state_change(xprt, sk->sk_socket);
1575 	switch (sk->sk_state) {
1576 	case TCP_ESTABLISHED:
1577 		spin_lock(&xprt->transport_lock);
1578 		if (!xprt_test_and_set_connected(xprt)) {
1579 
1580 			/* Reset TCP record info */
1581 			transport->tcp_offset = 0;
1582 			transport->tcp_reclen = 0;
1583 			transport->tcp_copied = 0;
1584 			transport->tcp_flags =
1585 				TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1586 			xprt->connect_cookie++;
1587 			clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1588 			xprt_clear_connecting(xprt);
1589 
1590 			xprt_wake_pending_tasks(xprt, -EAGAIN);
1591 		}
1592 		spin_unlock(&xprt->transport_lock);
1593 		break;
1594 	case TCP_FIN_WAIT1:
1595 		/* The client initiated a shutdown of the socket */
1596 		xprt->connect_cookie++;
1597 		xprt->reestablish_timeout = 0;
1598 		set_bit(XPRT_CLOSING, &xprt->state);
1599 		smp_mb__before_atomic();
1600 		clear_bit(XPRT_CONNECTED, &xprt->state);
1601 		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1602 		smp_mb__after_atomic();
1603 		break;
1604 	case TCP_CLOSE_WAIT:
1605 		/* The server initiated a shutdown of the socket */
1606 		xprt->connect_cookie++;
1607 		clear_bit(XPRT_CONNECTED, &xprt->state);
1608 		xs_tcp_force_close(xprt);
1609 	case TCP_CLOSING:
1610 		/*
1611 		 * If the server closed down the connection, make sure that
1612 		 * we back off before reconnecting
1613 		 */
1614 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1615 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1616 		break;
1617 	case TCP_LAST_ACK:
1618 		set_bit(XPRT_CLOSING, &xprt->state);
1619 		smp_mb__before_atomic();
1620 		clear_bit(XPRT_CONNECTED, &xprt->state);
1621 		smp_mb__after_atomic();
1622 		break;
1623 	case TCP_CLOSE:
1624 		if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1625 					&transport->sock_state))
1626 			xprt_clear_connecting(xprt);
1627 		xs_sock_mark_closed(xprt);
1628 	}
1629  out:
1630 	read_unlock_bh(&sk->sk_callback_lock);
1631 }
1632 
1633 static void xs_write_space(struct sock *sk)
1634 {
1635 	struct socket_wq *wq;
1636 	struct rpc_xprt *xprt;
1637 
1638 	if (!sk->sk_socket)
1639 		return;
1640 	clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1641 
1642 	if (unlikely(!(xprt = xprt_from_sock(sk))))
1643 		return;
1644 	rcu_read_lock();
1645 	wq = rcu_dereference(sk->sk_wq);
1646 	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1647 		goto out;
1648 
1649 	xprt_write_space(xprt);
1650 out:
1651 	rcu_read_unlock();
1652 }
1653 
1654 /**
1655  * xs_udp_write_space - callback invoked when socket buffer space
1656  *                             becomes available
1657  * @sk: socket whose state has changed
1658  *
1659  * Called when more output buffer space is available for this socket.
1660  * We try not to wake our writers until they can make "significant"
1661  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1662  * with a bunch of small requests.
1663  */
1664 static void xs_udp_write_space(struct sock *sk)
1665 {
1666 	read_lock_bh(&sk->sk_callback_lock);
1667 
1668 	/* from net/core/sock.c:sock_def_write_space */
1669 	if (sock_writeable(sk))
1670 		xs_write_space(sk);
1671 
1672 	read_unlock_bh(&sk->sk_callback_lock);
1673 }
1674 
1675 /**
1676  * xs_tcp_write_space - callback invoked when socket buffer space
1677  *                             becomes available
1678  * @sk: socket whose state has changed
1679  *
1680  * Called when more output buffer space is available for this socket.
1681  * We try not to wake our writers until they can make "significant"
1682  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1683  * with a bunch of small requests.
1684  */
1685 static void xs_tcp_write_space(struct sock *sk)
1686 {
1687 	read_lock_bh(&sk->sk_callback_lock);
1688 
1689 	/* from net/core/stream.c:sk_stream_write_space */
1690 	if (sk_stream_is_writeable(sk))
1691 		xs_write_space(sk);
1692 
1693 	read_unlock_bh(&sk->sk_callback_lock);
1694 }
1695 
1696 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1697 {
1698 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1699 	struct sock *sk = transport->inet;
1700 
1701 	if (transport->rcvsize) {
1702 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1703 		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1704 	}
1705 	if (transport->sndsize) {
1706 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1707 		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1708 		sk->sk_write_space(sk);
1709 	}
1710 }
1711 
1712 /**
1713  * xs_udp_set_buffer_size - set send and receive limits
1714  * @xprt: generic transport
1715  * @sndsize: requested size of send buffer, in bytes
1716  * @rcvsize: requested size of receive buffer, in bytes
1717  *
1718  * Set socket send and receive buffer size limits.
1719  */
1720 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1721 {
1722 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1723 
1724 	transport->sndsize = 0;
1725 	if (sndsize)
1726 		transport->sndsize = sndsize + 1024;
1727 	transport->rcvsize = 0;
1728 	if (rcvsize)
1729 		transport->rcvsize = rcvsize + 1024;
1730 
1731 	xs_udp_do_set_buffer_size(xprt);
1732 }
1733 
1734 /**
1735  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1736  * @task: task that timed out
1737  *
1738  * Adjust the congestion window after a retransmit timeout has occurred.
1739  */
1740 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1741 {
1742 	spin_lock_bh(&xprt->transport_lock);
1743 	xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1744 	spin_unlock_bh(&xprt->transport_lock);
1745 }
1746 
1747 static unsigned short xs_get_random_port(void)
1748 {
1749 	unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1750 	unsigned short rand = (unsigned short) prandom_u32() % range;
1751 	return rand + xprt_min_resvport;
1752 }
1753 
1754 /**
1755  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1756  * @sock: socket
1757  *
1758  * Note that this function has to be called on all sockets that share the
1759  * same port, and it must be called before binding.
1760  */
1761 static void xs_sock_set_reuseport(struct socket *sock)
1762 {
1763 	int opt = 1;
1764 
1765 	kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1766 			(char *)&opt, sizeof(opt));
1767 }
1768 
1769 static unsigned short xs_sock_getport(struct socket *sock)
1770 {
1771 	struct sockaddr_storage buf;
1772 	int buflen;
1773 	unsigned short port = 0;
1774 
1775 	if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1776 		goto out;
1777 	switch (buf.ss_family) {
1778 	case AF_INET6:
1779 		port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1780 		break;
1781 	case AF_INET:
1782 		port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1783 	}
1784 out:
1785 	return port;
1786 }
1787 
1788 /**
1789  * xs_set_port - reset the port number in the remote endpoint address
1790  * @xprt: generic transport
1791  * @port: new port number
1792  *
1793  */
1794 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1795 {
1796 	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1797 
1798 	rpc_set_port(xs_addr(xprt), port);
1799 	xs_update_peer_port(xprt);
1800 }
1801 
1802 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1803 {
1804 	if (transport->srcport == 0)
1805 		transport->srcport = xs_sock_getport(sock);
1806 }
1807 
1808 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1809 {
1810 	unsigned short port = transport->srcport;
1811 
1812 	if (port == 0 && transport->xprt.resvport)
1813 		port = xs_get_random_port();
1814 	return port;
1815 }
1816 
1817 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1818 {
1819 	if (transport->srcport != 0)
1820 		transport->srcport = 0;
1821 	if (!transport->xprt.resvport)
1822 		return 0;
1823 	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1824 		return xprt_max_resvport;
1825 	return --port;
1826 }
1827 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1828 {
1829 	struct sockaddr_storage myaddr;
1830 	int err, nloop = 0;
1831 	unsigned short port = xs_get_srcport(transport);
1832 	unsigned short last;
1833 
1834 	/*
1835 	 * If we are asking for any ephemeral port (i.e. port == 0 &&
1836 	 * transport->xprt.resvport == 0), don't bind.  Let the local
1837 	 * port selection happen implicitly when the socket is used
1838 	 * (for example at connect time).
1839 	 *
1840 	 * This ensures that we can continue to establish TCP
1841 	 * connections even when all local ephemeral ports are already
1842 	 * a part of some TCP connection.  This makes no difference
1843 	 * for UDP sockets, but also doens't harm them.
1844 	 *
1845 	 * If we're asking for any reserved port (i.e. port == 0 &&
1846 	 * transport->xprt.resvport == 1) xs_get_srcport above will
1847 	 * ensure that port is non-zero and we will bind as needed.
1848 	 */
1849 	if (port == 0)
1850 		return 0;
1851 
1852 	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1853 	do {
1854 		rpc_set_port((struct sockaddr *)&myaddr, port);
1855 		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1856 				transport->xprt.addrlen);
1857 		if (err == 0) {
1858 			transport->srcport = port;
1859 			break;
1860 		}
1861 		last = port;
1862 		port = xs_next_srcport(transport, port);
1863 		if (port > last)
1864 			nloop++;
1865 	} while (err == -EADDRINUSE && nloop != 2);
1866 
1867 	if (myaddr.ss_family == AF_INET)
1868 		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1869 				&((struct sockaddr_in *)&myaddr)->sin_addr,
1870 				port, err ? "failed" : "ok", err);
1871 	else
1872 		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1873 				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1874 				port, err ? "failed" : "ok", err);
1875 	return err;
1876 }
1877 
1878 /*
1879  * We don't support autobind on AF_LOCAL sockets
1880  */
1881 static void xs_local_rpcbind(struct rpc_task *task)
1882 {
1883 	xprt_set_bound(task->tk_xprt);
1884 }
1885 
1886 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1887 {
1888 }
1889 
1890 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1891 static struct lock_class_key xs_key[2];
1892 static struct lock_class_key xs_slock_key[2];
1893 
1894 static inline void xs_reclassify_socketu(struct socket *sock)
1895 {
1896 	struct sock *sk = sock->sk;
1897 
1898 	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1899 		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1900 }
1901 
1902 static inline void xs_reclassify_socket4(struct socket *sock)
1903 {
1904 	struct sock *sk = sock->sk;
1905 
1906 	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1907 		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1908 }
1909 
1910 static inline void xs_reclassify_socket6(struct socket *sock)
1911 {
1912 	struct sock *sk = sock->sk;
1913 
1914 	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1915 		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1916 }
1917 
1918 static inline void xs_reclassify_socket(int family, struct socket *sock)
1919 {
1920 	if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1921 		return;
1922 
1923 	switch (family) {
1924 	case AF_LOCAL:
1925 		xs_reclassify_socketu(sock);
1926 		break;
1927 	case AF_INET:
1928 		xs_reclassify_socket4(sock);
1929 		break;
1930 	case AF_INET6:
1931 		xs_reclassify_socket6(sock);
1932 		break;
1933 	}
1934 }
1935 #else
1936 static inline void xs_reclassify_socket(int family, struct socket *sock)
1937 {
1938 }
1939 #endif
1940 
1941 static void xs_dummy_setup_socket(struct work_struct *work)
1942 {
1943 }
1944 
1945 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1946 		struct sock_xprt *transport, int family, int type,
1947 		int protocol, bool reuseport)
1948 {
1949 	struct socket *sock;
1950 	int err;
1951 
1952 	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1953 	if (err < 0) {
1954 		dprintk("RPC:       can't create %d transport socket (%d).\n",
1955 				protocol, -err);
1956 		goto out;
1957 	}
1958 	xs_reclassify_socket(family, sock);
1959 
1960 	if (reuseport)
1961 		xs_sock_set_reuseport(sock);
1962 
1963 	err = xs_bind(transport, sock);
1964 	if (err) {
1965 		sock_release(sock);
1966 		goto out;
1967 	}
1968 
1969 	return sock;
1970 out:
1971 	return ERR_PTR(err);
1972 }
1973 
1974 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1975 				      struct socket *sock)
1976 {
1977 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1978 									xprt);
1979 
1980 	if (!transport->inet) {
1981 		struct sock *sk = sock->sk;
1982 
1983 		write_lock_bh(&sk->sk_callback_lock);
1984 
1985 		xs_save_old_callbacks(transport, sk);
1986 
1987 		sk->sk_user_data = xprt;
1988 		sk->sk_data_ready = xs_data_ready;
1989 		sk->sk_write_space = xs_udp_write_space;
1990 		sock_set_flag(sk, SOCK_FASYNC);
1991 		sk->sk_error_report = xs_error_report;
1992 		sk->sk_allocation = GFP_NOIO;
1993 
1994 		xprt_clear_connected(xprt);
1995 
1996 		/* Reset to new socket */
1997 		transport->sock = sock;
1998 		transport->inet = sk;
1999 
2000 		write_unlock_bh(&sk->sk_callback_lock);
2001 	}
2002 
2003 	/* Tell the socket layer to start connecting... */
2004 	xprt->stat.connect_count++;
2005 	xprt->stat.connect_start = jiffies;
2006 	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2007 }
2008 
2009 /**
2010  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2011  * @transport: socket transport to connect
2012  */
2013 static int xs_local_setup_socket(struct sock_xprt *transport)
2014 {
2015 	struct rpc_xprt *xprt = &transport->xprt;
2016 	struct socket *sock;
2017 	int status = -EIO;
2018 
2019 	status = __sock_create(xprt->xprt_net, AF_LOCAL,
2020 					SOCK_STREAM, 0, &sock, 1);
2021 	if (status < 0) {
2022 		dprintk("RPC:       can't create AF_LOCAL "
2023 			"transport socket (%d).\n", -status);
2024 		goto out;
2025 	}
2026 	xs_reclassify_socket(AF_LOCAL, sock);
2027 
2028 	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2029 			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2030 
2031 	status = xs_local_finish_connecting(xprt, sock);
2032 	trace_rpc_socket_connect(xprt, sock, status);
2033 	switch (status) {
2034 	case 0:
2035 		dprintk("RPC:       xprt %p connected to %s\n",
2036 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2037 		xprt_set_connected(xprt);
2038 	case -ENOBUFS:
2039 		break;
2040 	case -ENOENT:
2041 		dprintk("RPC:       xprt %p: socket %s does not exist\n",
2042 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2043 		break;
2044 	case -ECONNREFUSED:
2045 		dprintk("RPC:       xprt %p: connection refused for %s\n",
2046 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2047 		break;
2048 	default:
2049 		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2050 				__func__, -status,
2051 				xprt->address_strings[RPC_DISPLAY_ADDR]);
2052 	}
2053 
2054 out:
2055 	xprt_clear_connecting(xprt);
2056 	xprt_wake_pending_tasks(xprt, status);
2057 	return status;
2058 }
2059 
2060 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2061 {
2062 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2063 	int ret;
2064 
2065 	 if (RPC_IS_ASYNC(task)) {
2066 		/*
2067 		 * We want the AF_LOCAL connect to be resolved in the
2068 		 * filesystem namespace of the process making the rpc
2069 		 * call.  Thus we connect synchronously.
2070 		 *
2071 		 * If we want to support asynchronous AF_LOCAL calls,
2072 		 * we'll need to figure out how to pass a namespace to
2073 		 * connect.
2074 		 */
2075 		rpc_exit(task, -ENOTCONN);
2076 		return;
2077 	}
2078 	ret = xs_local_setup_socket(transport);
2079 	if (ret && !RPC_IS_SOFTCONN(task))
2080 		msleep_interruptible(15000);
2081 }
2082 
2083 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2084 /*
2085  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2086  * know that we have exclusive access to the socket), to guard against
2087  * races with xs_reset_transport.
2088  */
2089 static void xs_set_memalloc(struct rpc_xprt *xprt)
2090 {
2091 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2092 			xprt);
2093 
2094 	/*
2095 	 * If there's no sock, then we have nothing to set. The
2096 	 * reconnecting process will get it for us.
2097 	 */
2098 	if (!transport->inet)
2099 		return;
2100 	if (atomic_read(&xprt->swapper))
2101 		sk_set_memalloc(transport->inet);
2102 }
2103 
2104 /**
2105  * xs_enable_swap - Tag this transport as being used for swap.
2106  * @xprt: transport to tag
2107  *
2108  * Take a reference to this transport on behalf of the rpc_clnt, and
2109  * optionally mark it for swapping if it wasn't already.
2110  */
2111 static int
2112 xs_enable_swap(struct rpc_xprt *xprt)
2113 {
2114 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2115 
2116 	if (atomic_inc_return(&xprt->swapper) != 1)
2117 		return 0;
2118 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2119 		return -ERESTARTSYS;
2120 	if (xs->inet)
2121 		sk_set_memalloc(xs->inet);
2122 	xprt_release_xprt(xprt, NULL);
2123 	return 0;
2124 }
2125 
2126 /**
2127  * xs_disable_swap - Untag this transport as being used for swap.
2128  * @xprt: transport to tag
2129  *
2130  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2131  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2132  */
2133 static void
2134 xs_disable_swap(struct rpc_xprt *xprt)
2135 {
2136 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2137 
2138 	if (!atomic_dec_and_test(&xprt->swapper))
2139 		return;
2140 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2141 		return;
2142 	if (xs->inet)
2143 		sk_clear_memalloc(xs->inet);
2144 	xprt_release_xprt(xprt, NULL);
2145 }
2146 #else
2147 static void xs_set_memalloc(struct rpc_xprt *xprt)
2148 {
2149 }
2150 
2151 static int
2152 xs_enable_swap(struct rpc_xprt *xprt)
2153 {
2154 	return -EINVAL;
2155 }
2156 
2157 static void
2158 xs_disable_swap(struct rpc_xprt *xprt)
2159 {
2160 }
2161 #endif
2162 
2163 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2164 {
2165 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2166 
2167 	if (!transport->inet) {
2168 		struct sock *sk = sock->sk;
2169 
2170 		write_lock_bh(&sk->sk_callback_lock);
2171 
2172 		xs_save_old_callbacks(transport, sk);
2173 
2174 		sk->sk_user_data = xprt;
2175 		sk->sk_data_ready = xs_data_ready;
2176 		sk->sk_write_space = xs_udp_write_space;
2177 		sock_set_flag(sk, SOCK_FASYNC);
2178 		sk->sk_allocation = GFP_NOIO;
2179 
2180 		xprt_set_connected(xprt);
2181 
2182 		/* Reset to new socket */
2183 		transport->sock = sock;
2184 		transport->inet = sk;
2185 
2186 		xs_set_memalloc(xprt);
2187 
2188 		write_unlock_bh(&sk->sk_callback_lock);
2189 	}
2190 	xs_udp_do_set_buffer_size(xprt);
2191 
2192 	xprt->stat.connect_start = jiffies;
2193 }
2194 
2195 static void xs_udp_setup_socket(struct work_struct *work)
2196 {
2197 	struct sock_xprt *transport =
2198 		container_of(work, struct sock_xprt, connect_worker.work);
2199 	struct rpc_xprt *xprt = &transport->xprt;
2200 	struct socket *sock = transport->sock;
2201 	int status = -EIO;
2202 
2203 	sock = xs_create_sock(xprt, transport,
2204 			xs_addr(xprt)->sa_family, SOCK_DGRAM,
2205 			IPPROTO_UDP, false);
2206 	if (IS_ERR(sock))
2207 		goto out;
2208 
2209 	dprintk("RPC:       worker connecting xprt %p via %s to "
2210 				"%s (port %s)\n", xprt,
2211 			xprt->address_strings[RPC_DISPLAY_PROTO],
2212 			xprt->address_strings[RPC_DISPLAY_ADDR],
2213 			xprt->address_strings[RPC_DISPLAY_PORT]);
2214 
2215 	xs_udp_finish_connecting(xprt, sock);
2216 	trace_rpc_socket_connect(xprt, sock, 0);
2217 	status = 0;
2218 out:
2219 	xprt_unlock_connect(xprt, transport);
2220 	xprt_clear_connecting(xprt);
2221 	xprt_wake_pending_tasks(xprt, status);
2222 }
2223 
2224 /**
2225  * xs_tcp_shutdown - gracefully shut down a TCP socket
2226  * @xprt: transport
2227  *
2228  * Initiates a graceful shutdown of the TCP socket by calling the
2229  * equivalent of shutdown(SHUT_RDWR);
2230  */
2231 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2232 {
2233 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2234 	struct socket *sock = transport->sock;
2235 
2236 	if (sock == NULL)
2237 		return;
2238 	if (xprt_connected(xprt)) {
2239 		kernel_sock_shutdown(sock, SHUT_RDWR);
2240 		trace_rpc_socket_shutdown(xprt, sock);
2241 	} else
2242 		xs_reset_transport(transport);
2243 }
2244 
2245 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2246 		struct socket *sock)
2247 {
2248 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2249 	unsigned int keepidle;
2250 	unsigned int keepcnt;
2251 	unsigned int opt_on = 1;
2252 	unsigned int timeo;
2253 
2254 	spin_lock_bh(&xprt->transport_lock);
2255 	keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2256 	keepcnt = xprt->timeout->to_retries + 1;
2257 	timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2258 		(xprt->timeout->to_retries + 1);
2259 	clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2260 	spin_unlock_bh(&xprt->transport_lock);
2261 
2262 	/* TCP Keepalive options */
2263 	kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2264 			(char *)&opt_on, sizeof(opt_on));
2265 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2266 			(char *)&keepidle, sizeof(keepidle));
2267 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2268 			(char *)&keepidle, sizeof(keepidle));
2269 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2270 			(char *)&keepcnt, sizeof(keepcnt));
2271 
2272 	/* TCP user timeout (see RFC5482) */
2273 	kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2274 			(char *)&timeo, sizeof(timeo));
2275 }
2276 
2277 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2278 		unsigned long connect_timeout,
2279 		unsigned long reconnect_timeout)
2280 {
2281 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2282 	struct rpc_timeout to;
2283 	unsigned long initval;
2284 
2285 	spin_lock_bh(&xprt->transport_lock);
2286 	if (reconnect_timeout < xprt->max_reconnect_timeout)
2287 		xprt->max_reconnect_timeout = reconnect_timeout;
2288 	if (connect_timeout < xprt->connect_timeout) {
2289 		memcpy(&to, xprt->timeout, sizeof(to));
2290 		initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2291 		/* Arbitrary lower limit */
2292 		if (initval <  XS_TCP_INIT_REEST_TO << 1)
2293 			initval = XS_TCP_INIT_REEST_TO << 1;
2294 		to.to_initval = initval;
2295 		to.to_maxval = initval;
2296 		memcpy(&transport->tcp_timeout, &to,
2297 				sizeof(transport->tcp_timeout));
2298 		xprt->timeout = &transport->tcp_timeout;
2299 		xprt->connect_timeout = connect_timeout;
2300 	}
2301 	set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2302 	spin_unlock_bh(&xprt->transport_lock);
2303 }
2304 
2305 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2306 {
2307 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2308 	int ret = -ENOTCONN;
2309 
2310 	if (!transport->inet) {
2311 		struct sock *sk = sock->sk;
2312 		unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2313 
2314 		/* Avoid temporary address, they are bad for long-lived
2315 		 * connections such as NFS mounts.
2316 		 * RFC4941, section 3.6 suggests that:
2317 		 *    Individual applications, which have specific
2318 		 *    knowledge about the normal duration of connections,
2319 		 *    MAY override this as appropriate.
2320 		 */
2321 		kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2322 				(char *)&addr_pref, sizeof(addr_pref));
2323 
2324 		xs_tcp_set_socket_timeouts(xprt, sock);
2325 
2326 		write_lock_bh(&sk->sk_callback_lock);
2327 
2328 		xs_save_old_callbacks(transport, sk);
2329 
2330 		sk->sk_user_data = xprt;
2331 		sk->sk_data_ready = xs_data_ready;
2332 		sk->sk_state_change = xs_tcp_state_change;
2333 		sk->sk_write_space = xs_tcp_write_space;
2334 		sock_set_flag(sk, SOCK_FASYNC);
2335 		sk->sk_error_report = xs_error_report;
2336 		sk->sk_allocation = GFP_NOIO;
2337 
2338 		/* socket options */
2339 		sock_reset_flag(sk, SOCK_LINGER);
2340 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2341 
2342 		xprt_clear_connected(xprt);
2343 
2344 		/* Reset to new socket */
2345 		transport->sock = sock;
2346 		transport->inet = sk;
2347 
2348 		write_unlock_bh(&sk->sk_callback_lock);
2349 	}
2350 
2351 	if (!xprt_bound(xprt))
2352 		goto out;
2353 
2354 	xs_set_memalloc(xprt);
2355 
2356 	/* Tell the socket layer to start connecting... */
2357 	xprt->stat.connect_count++;
2358 	xprt->stat.connect_start = jiffies;
2359 	set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2360 	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2361 	switch (ret) {
2362 	case 0:
2363 		xs_set_srcport(transport, sock);
2364 	case -EINPROGRESS:
2365 		/* SYN_SENT! */
2366 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2367 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2368 		break;
2369 	case -EADDRNOTAVAIL:
2370 		/* Source port number is unavailable. Try a new one! */
2371 		transport->srcport = 0;
2372 	}
2373 out:
2374 	return ret;
2375 }
2376 
2377 /**
2378  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2379  *
2380  * Invoked by a work queue tasklet.
2381  */
2382 static void xs_tcp_setup_socket(struct work_struct *work)
2383 {
2384 	struct sock_xprt *transport =
2385 		container_of(work, struct sock_xprt, connect_worker.work);
2386 	struct socket *sock = transport->sock;
2387 	struct rpc_xprt *xprt = &transport->xprt;
2388 	int status = -EIO;
2389 
2390 	if (!sock) {
2391 		sock = xs_create_sock(xprt, transport,
2392 				xs_addr(xprt)->sa_family, SOCK_STREAM,
2393 				IPPROTO_TCP, true);
2394 		if (IS_ERR(sock)) {
2395 			status = PTR_ERR(sock);
2396 			goto out;
2397 		}
2398 	}
2399 
2400 	dprintk("RPC:       worker connecting xprt %p via %s to "
2401 				"%s (port %s)\n", xprt,
2402 			xprt->address_strings[RPC_DISPLAY_PROTO],
2403 			xprt->address_strings[RPC_DISPLAY_ADDR],
2404 			xprt->address_strings[RPC_DISPLAY_PORT]);
2405 
2406 	status = xs_tcp_finish_connecting(xprt, sock);
2407 	trace_rpc_socket_connect(xprt, sock, status);
2408 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2409 			xprt, -status, xprt_connected(xprt),
2410 			sock->sk->sk_state);
2411 	switch (status) {
2412 	default:
2413 		printk("%s: connect returned unhandled error %d\n",
2414 			__func__, status);
2415 	case -EADDRNOTAVAIL:
2416 		/* We're probably in TIME_WAIT. Get rid of existing socket,
2417 		 * and retry
2418 		 */
2419 		xs_tcp_force_close(xprt);
2420 		break;
2421 	case 0:
2422 	case -EINPROGRESS:
2423 	case -EALREADY:
2424 		xprt_unlock_connect(xprt, transport);
2425 		return;
2426 	case -EINVAL:
2427 		/* Happens, for instance, if the user specified a link
2428 		 * local IPv6 address without a scope-id.
2429 		 */
2430 	case -ECONNREFUSED:
2431 	case -ECONNRESET:
2432 	case -ENETUNREACH:
2433 	case -EADDRINUSE:
2434 	case -ENOBUFS:
2435 		/* retry with existing socket, after a delay */
2436 		xs_tcp_force_close(xprt);
2437 		goto out;
2438 	}
2439 	status = -EAGAIN;
2440 out:
2441 	xprt_unlock_connect(xprt, transport);
2442 	xprt_clear_connecting(xprt);
2443 	xprt_wake_pending_tasks(xprt, status);
2444 }
2445 
2446 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2447 {
2448 	unsigned long start, now = jiffies;
2449 
2450 	start = xprt->stat.connect_start + xprt->reestablish_timeout;
2451 	if (time_after(start, now))
2452 		return start - now;
2453 	return 0;
2454 }
2455 
2456 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2457 {
2458 	xprt->reestablish_timeout <<= 1;
2459 	if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2460 		xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2461 	if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2462 		xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2463 }
2464 
2465 /**
2466  * xs_connect - connect a socket to a remote endpoint
2467  * @xprt: pointer to transport structure
2468  * @task: address of RPC task that manages state of connect request
2469  *
2470  * TCP: If the remote end dropped the connection, delay reconnecting.
2471  *
2472  * UDP socket connects are synchronous, but we use a work queue anyway
2473  * to guarantee that even unprivileged user processes can set up a
2474  * socket on a privileged port.
2475  *
2476  * If a UDP socket connect fails, the delay behavior here prevents
2477  * retry floods (hard mounts).
2478  */
2479 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2480 {
2481 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2482 	unsigned long delay = 0;
2483 
2484 	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2485 
2486 	if (transport->sock != NULL) {
2487 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2488 				"seconds\n",
2489 				xprt, xprt->reestablish_timeout / HZ);
2490 
2491 		/* Start by resetting any existing state */
2492 		xs_reset_transport(transport);
2493 
2494 		delay = xs_reconnect_delay(xprt);
2495 		xs_reconnect_backoff(xprt);
2496 
2497 	} else
2498 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2499 
2500 	queue_delayed_work(xprtiod_workqueue,
2501 			&transport->connect_worker,
2502 			delay);
2503 }
2504 
2505 /**
2506  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2507  * @xprt: rpc_xprt struct containing statistics
2508  * @seq: output file
2509  *
2510  */
2511 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2512 {
2513 	long idle_time = 0;
2514 
2515 	if (xprt_connected(xprt))
2516 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2517 
2518 	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2519 			"%llu %llu %lu %llu %llu\n",
2520 			xprt->stat.bind_count,
2521 			xprt->stat.connect_count,
2522 			xprt->stat.connect_time,
2523 			idle_time,
2524 			xprt->stat.sends,
2525 			xprt->stat.recvs,
2526 			xprt->stat.bad_xids,
2527 			xprt->stat.req_u,
2528 			xprt->stat.bklog_u,
2529 			xprt->stat.max_slots,
2530 			xprt->stat.sending_u,
2531 			xprt->stat.pending_u);
2532 }
2533 
2534 /**
2535  * xs_udp_print_stats - display UDP socket-specifc stats
2536  * @xprt: rpc_xprt struct containing statistics
2537  * @seq: output file
2538  *
2539  */
2540 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2541 {
2542 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2543 
2544 	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2545 			"%lu %llu %llu\n",
2546 			transport->srcport,
2547 			xprt->stat.bind_count,
2548 			xprt->stat.sends,
2549 			xprt->stat.recvs,
2550 			xprt->stat.bad_xids,
2551 			xprt->stat.req_u,
2552 			xprt->stat.bklog_u,
2553 			xprt->stat.max_slots,
2554 			xprt->stat.sending_u,
2555 			xprt->stat.pending_u);
2556 }
2557 
2558 /**
2559  * xs_tcp_print_stats - display TCP socket-specifc stats
2560  * @xprt: rpc_xprt struct containing statistics
2561  * @seq: output file
2562  *
2563  */
2564 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2565 {
2566 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2567 	long idle_time = 0;
2568 
2569 	if (xprt_connected(xprt))
2570 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2571 
2572 	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2573 			"%llu %llu %lu %llu %llu\n",
2574 			transport->srcport,
2575 			xprt->stat.bind_count,
2576 			xprt->stat.connect_count,
2577 			xprt->stat.connect_time,
2578 			idle_time,
2579 			xprt->stat.sends,
2580 			xprt->stat.recvs,
2581 			xprt->stat.bad_xids,
2582 			xprt->stat.req_u,
2583 			xprt->stat.bklog_u,
2584 			xprt->stat.max_slots,
2585 			xprt->stat.sending_u,
2586 			xprt->stat.pending_u);
2587 }
2588 
2589 /*
2590  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2591  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2592  * to use the server side send routines.
2593  */
2594 static int bc_malloc(struct rpc_task *task)
2595 {
2596 	struct rpc_rqst *rqst = task->tk_rqstp;
2597 	size_t size = rqst->rq_callsize;
2598 	struct page *page;
2599 	struct rpc_buffer *buf;
2600 
2601 	if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2602 		WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2603 			  size);
2604 		return -EINVAL;
2605 	}
2606 
2607 	page = alloc_page(GFP_KERNEL);
2608 	if (!page)
2609 		return -ENOMEM;
2610 
2611 	buf = page_address(page);
2612 	buf->len = PAGE_SIZE;
2613 
2614 	rqst->rq_buffer = buf->data;
2615 	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2616 	return 0;
2617 }
2618 
2619 /*
2620  * Free the space allocated in the bc_alloc routine
2621  */
2622 static void bc_free(struct rpc_task *task)
2623 {
2624 	void *buffer = task->tk_rqstp->rq_buffer;
2625 	struct rpc_buffer *buf;
2626 
2627 	buf = container_of(buffer, struct rpc_buffer, data);
2628 	free_page((unsigned long)buf);
2629 }
2630 
2631 /*
2632  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2633  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2634  */
2635 static int bc_sendto(struct rpc_rqst *req)
2636 {
2637 	int len;
2638 	struct xdr_buf *xbufp = &req->rq_snd_buf;
2639 	struct rpc_xprt *xprt = req->rq_xprt;
2640 	struct sock_xprt *transport =
2641 				container_of(xprt, struct sock_xprt, xprt);
2642 	struct socket *sock = transport->sock;
2643 	unsigned long headoff;
2644 	unsigned long tailoff;
2645 
2646 	xs_encode_stream_record_marker(xbufp);
2647 
2648 	tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2649 	headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2650 	len = svc_send_common(sock, xbufp,
2651 			      virt_to_page(xbufp->head[0].iov_base), headoff,
2652 			      xbufp->tail[0].iov_base, tailoff);
2653 
2654 	if (len != xbufp->len) {
2655 		printk(KERN_NOTICE "Error sending entire callback!\n");
2656 		len = -EAGAIN;
2657 	}
2658 
2659 	return len;
2660 }
2661 
2662 /*
2663  * The send routine. Borrows from svc_send
2664  */
2665 static int bc_send_request(struct rpc_task *task)
2666 {
2667 	struct rpc_rqst *req = task->tk_rqstp;
2668 	struct svc_xprt	*xprt;
2669 	int len;
2670 
2671 	dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2672 	/*
2673 	 * Get the server socket associated with this callback xprt
2674 	 */
2675 	xprt = req->rq_xprt->bc_xprt;
2676 
2677 	/*
2678 	 * Grab the mutex to serialize data as the connection is shared
2679 	 * with the fore channel
2680 	 */
2681 	if (!mutex_trylock(&xprt->xpt_mutex)) {
2682 		rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2683 		if (!mutex_trylock(&xprt->xpt_mutex))
2684 			return -EAGAIN;
2685 		rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2686 	}
2687 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2688 		len = -ENOTCONN;
2689 	else
2690 		len = bc_sendto(req);
2691 	mutex_unlock(&xprt->xpt_mutex);
2692 
2693 	if (len > 0)
2694 		len = 0;
2695 
2696 	return len;
2697 }
2698 
2699 /*
2700  * The close routine. Since this is client initiated, we do nothing
2701  */
2702 
2703 static void bc_close(struct rpc_xprt *xprt)
2704 {
2705 }
2706 
2707 /*
2708  * The xprt destroy routine. Again, because this connection is client
2709  * initiated, we do nothing
2710  */
2711 
2712 static void bc_destroy(struct rpc_xprt *xprt)
2713 {
2714 	dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2715 
2716 	xs_xprt_free(xprt);
2717 	module_put(THIS_MODULE);
2718 }
2719 
2720 static struct rpc_xprt_ops xs_local_ops = {
2721 	.reserve_xprt		= xprt_reserve_xprt,
2722 	.release_xprt		= xs_tcp_release_xprt,
2723 	.alloc_slot		= xprt_alloc_slot,
2724 	.rpcbind		= xs_local_rpcbind,
2725 	.set_port		= xs_local_set_port,
2726 	.connect		= xs_local_connect,
2727 	.buf_alloc		= rpc_malloc,
2728 	.buf_free		= rpc_free,
2729 	.send_request		= xs_local_send_request,
2730 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2731 	.close			= xs_close,
2732 	.destroy		= xs_destroy,
2733 	.print_stats		= xs_local_print_stats,
2734 	.enable_swap		= xs_enable_swap,
2735 	.disable_swap		= xs_disable_swap,
2736 };
2737 
2738 static struct rpc_xprt_ops xs_udp_ops = {
2739 	.set_buffer_size	= xs_udp_set_buffer_size,
2740 	.reserve_xprt		= xprt_reserve_xprt_cong,
2741 	.release_xprt		= xprt_release_xprt_cong,
2742 	.alloc_slot		= xprt_alloc_slot,
2743 	.rpcbind		= rpcb_getport_async,
2744 	.set_port		= xs_set_port,
2745 	.connect		= xs_connect,
2746 	.buf_alloc		= rpc_malloc,
2747 	.buf_free		= rpc_free,
2748 	.send_request		= xs_udp_send_request,
2749 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
2750 	.timer			= xs_udp_timer,
2751 	.release_request	= xprt_release_rqst_cong,
2752 	.close			= xs_close,
2753 	.destroy		= xs_destroy,
2754 	.print_stats		= xs_udp_print_stats,
2755 	.enable_swap		= xs_enable_swap,
2756 	.disable_swap		= xs_disable_swap,
2757 	.inject_disconnect	= xs_inject_disconnect,
2758 };
2759 
2760 static struct rpc_xprt_ops xs_tcp_ops = {
2761 	.reserve_xprt		= xprt_reserve_xprt,
2762 	.release_xprt		= xs_tcp_release_xprt,
2763 	.alloc_slot		= xprt_lock_and_alloc_slot,
2764 	.rpcbind		= rpcb_getport_async,
2765 	.set_port		= xs_set_port,
2766 	.connect		= xs_connect,
2767 	.buf_alloc		= rpc_malloc,
2768 	.buf_free		= rpc_free,
2769 	.send_request		= xs_tcp_send_request,
2770 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2771 	.close			= xs_tcp_shutdown,
2772 	.destroy		= xs_destroy,
2773 	.set_connect_timeout	= xs_tcp_set_connect_timeout,
2774 	.print_stats		= xs_tcp_print_stats,
2775 	.enable_swap		= xs_enable_swap,
2776 	.disable_swap		= xs_disable_swap,
2777 	.inject_disconnect	= xs_inject_disconnect,
2778 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2779 	.bc_setup		= xprt_setup_bc,
2780 	.bc_up			= xs_tcp_bc_up,
2781 	.bc_maxpayload		= xs_tcp_bc_maxpayload,
2782 	.bc_free_rqst		= xprt_free_bc_rqst,
2783 	.bc_destroy		= xprt_destroy_bc,
2784 #endif
2785 };
2786 
2787 /*
2788  * The rpc_xprt_ops for the server backchannel
2789  */
2790 
2791 static struct rpc_xprt_ops bc_tcp_ops = {
2792 	.reserve_xprt		= xprt_reserve_xprt,
2793 	.release_xprt		= xprt_release_xprt,
2794 	.alloc_slot		= xprt_alloc_slot,
2795 	.buf_alloc		= bc_malloc,
2796 	.buf_free		= bc_free,
2797 	.send_request		= bc_send_request,
2798 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2799 	.close			= bc_close,
2800 	.destroy		= bc_destroy,
2801 	.print_stats		= xs_tcp_print_stats,
2802 	.enable_swap		= xs_enable_swap,
2803 	.disable_swap		= xs_disable_swap,
2804 	.inject_disconnect	= xs_inject_disconnect,
2805 };
2806 
2807 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2808 {
2809 	static const struct sockaddr_in sin = {
2810 		.sin_family		= AF_INET,
2811 		.sin_addr.s_addr	= htonl(INADDR_ANY),
2812 	};
2813 	static const struct sockaddr_in6 sin6 = {
2814 		.sin6_family		= AF_INET6,
2815 		.sin6_addr		= IN6ADDR_ANY_INIT,
2816 	};
2817 
2818 	switch (family) {
2819 	case AF_LOCAL:
2820 		break;
2821 	case AF_INET:
2822 		memcpy(sap, &sin, sizeof(sin));
2823 		break;
2824 	case AF_INET6:
2825 		memcpy(sap, &sin6, sizeof(sin6));
2826 		break;
2827 	default:
2828 		dprintk("RPC:       %s: Bad address family\n", __func__);
2829 		return -EAFNOSUPPORT;
2830 	}
2831 	return 0;
2832 }
2833 
2834 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2835 				      unsigned int slot_table_size,
2836 				      unsigned int max_slot_table_size)
2837 {
2838 	struct rpc_xprt *xprt;
2839 	struct sock_xprt *new;
2840 
2841 	if (args->addrlen > sizeof(xprt->addr)) {
2842 		dprintk("RPC:       xs_setup_xprt: address too large\n");
2843 		return ERR_PTR(-EBADF);
2844 	}
2845 
2846 	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2847 			max_slot_table_size);
2848 	if (xprt == NULL) {
2849 		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2850 				"rpc_xprt\n");
2851 		return ERR_PTR(-ENOMEM);
2852 	}
2853 
2854 	new = container_of(xprt, struct sock_xprt, xprt);
2855 	mutex_init(&new->recv_mutex);
2856 	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2857 	xprt->addrlen = args->addrlen;
2858 	if (args->srcaddr)
2859 		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2860 	else {
2861 		int err;
2862 		err = xs_init_anyaddr(args->dstaddr->sa_family,
2863 					(struct sockaddr *)&new->srcaddr);
2864 		if (err != 0) {
2865 			xprt_free(xprt);
2866 			return ERR_PTR(err);
2867 		}
2868 	}
2869 
2870 	return xprt;
2871 }
2872 
2873 static const struct rpc_timeout xs_local_default_timeout = {
2874 	.to_initval = 10 * HZ,
2875 	.to_maxval = 10 * HZ,
2876 	.to_retries = 2,
2877 };
2878 
2879 /**
2880  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2881  * @args: rpc transport creation arguments
2882  *
2883  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2884  */
2885 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2886 {
2887 	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2888 	struct sock_xprt *transport;
2889 	struct rpc_xprt *xprt;
2890 	struct rpc_xprt *ret;
2891 
2892 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2893 			xprt_max_tcp_slot_table_entries);
2894 	if (IS_ERR(xprt))
2895 		return xprt;
2896 	transport = container_of(xprt, struct sock_xprt, xprt);
2897 
2898 	xprt->prot = 0;
2899 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2900 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2901 
2902 	xprt->bind_timeout = XS_BIND_TO;
2903 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2904 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2905 
2906 	xprt->ops = &xs_local_ops;
2907 	xprt->timeout = &xs_local_default_timeout;
2908 
2909 	INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2910 	INIT_DELAYED_WORK(&transport->connect_worker,
2911 			xs_dummy_setup_socket);
2912 
2913 	switch (sun->sun_family) {
2914 	case AF_LOCAL:
2915 		if (sun->sun_path[0] != '/') {
2916 			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2917 					sun->sun_path);
2918 			ret = ERR_PTR(-EINVAL);
2919 			goto out_err;
2920 		}
2921 		xprt_set_bound(xprt);
2922 		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2923 		ret = ERR_PTR(xs_local_setup_socket(transport));
2924 		if (ret)
2925 			goto out_err;
2926 		break;
2927 	default:
2928 		ret = ERR_PTR(-EAFNOSUPPORT);
2929 		goto out_err;
2930 	}
2931 
2932 	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2933 			xprt->address_strings[RPC_DISPLAY_ADDR]);
2934 
2935 	if (try_module_get(THIS_MODULE))
2936 		return xprt;
2937 	ret = ERR_PTR(-EINVAL);
2938 out_err:
2939 	xs_xprt_free(xprt);
2940 	return ret;
2941 }
2942 
2943 static const struct rpc_timeout xs_udp_default_timeout = {
2944 	.to_initval = 5 * HZ,
2945 	.to_maxval = 30 * HZ,
2946 	.to_increment = 5 * HZ,
2947 	.to_retries = 5,
2948 };
2949 
2950 /**
2951  * xs_setup_udp - Set up transport to use a UDP socket
2952  * @args: rpc transport creation arguments
2953  *
2954  */
2955 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2956 {
2957 	struct sockaddr *addr = args->dstaddr;
2958 	struct rpc_xprt *xprt;
2959 	struct sock_xprt *transport;
2960 	struct rpc_xprt *ret;
2961 
2962 	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2963 			xprt_udp_slot_table_entries);
2964 	if (IS_ERR(xprt))
2965 		return xprt;
2966 	transport = container_of(xprt, struct sock_xprt, xprt);
2967 
2968 	xprt->prot = IPPROTO_UDP;
2969 	xprt->tsh_size = 0;
2970 	/* XXX: header size can vary due to auth type, IPv6, etc. */
2971 	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2972 
2973 	xprt->bind_timeout = XS_BIND_TO;
2974 	xprt->reestablish_timeout = XS_UDP_REEST_TO;
2975 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2976 
2977 	xprt->ops = &xs_udp_ops;
2978 
2979 	xprt->timeout = &xs_udp_default_timeout;
2980 
2981 	INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2982 	INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2983 
2984 	switch (addr->sa_family) {
2985 	case AF_INET:
2986 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2987 			xprt_set_bound(xprt);
2988 
2989 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2990 		break;
2991 	case AF_INET6:
2992 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2993 			xprt_set_bound(xprt);
2994 
2995 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2996 		break;
2997 	default:
2998 		ret = ERR_PTR(-EAFNOSUPPORT);
2999 		goto out_err;
3000 	}
3001 
3002 	if (xprt_bound(xprt))
3003 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3004 				xprt->address_strings[RPC_DISPLAY_ADDR],
3005 				xprt->address_strings[RPC_DISPLAY_PORT],
3006 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3007 	else
3008 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3009 				xprt->address_strings[RPC_DISPLAY_ADDR],
3010 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3011 
3012 	if (try_module_get(THIS_MODULE))
3013 		return xprt;
3014 	ret = ERR_PTR(-EINVAL);
3015 out_err:
3016 	xs_xprt_free(xprt);
3017 	return ret;
3018 }
3019 
3020 static const struct rpc_timeout xs_tcp_default_timeout = {
3021 	.to_initval = 60 * HZ,
3022 	.to_maxval = 60 * HZ,
3023 	.to_retries = 2,
3024 };
3025 
3026 /**
3027  * xs_setup_tcp - Set up transport to use a TCP socket
3028  * @args: rpc transport creation arguments
3029  *
3030  */
3031 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3032 {
3033 	struct sockaddr *addr = args->dstaddr;
3034 	struct rpc_xprt *xprt;
3035 	struct sock_xprt *transport;
3036 	struct rpc_xprt *ret;
3037 	unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3038 
3039 	if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3040 		max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3041 
3042 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3043 			max_slot_table_size);
3044 	if (IS_ERR(xprt))
3045 		return xprt;
3046 	transport = container_of(xprt, struct sock_xprt, xprt);
3047 
3048 	xprt->prot = IPPROTO_TCP;
3049 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3050 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3051 
3052 	xprt->bind_timeout = XS_BIND_TO;
3053 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3054 	xprt->idle_timeout = XS_IDLE_DISC_TO;
3055 
3056 	xprt->ops = &xs_tcp_ops;
3057 	xprt->timeout = &xs_tcp_default_timeout;
3058 
3059 	xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3060 	xprt->connect_timeout = xprt->timeout->to_initval *
3061 		(xprt->timeout->to_retries + 1);
3062 
3063 	INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3064 	INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3065 
3066 	switch (addr->sa_family) {
3067 	case AF_INET:
3068 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3069 			xprt_set_bound(xprt);
3070 
3071 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3072 		break;
3073 	case AF_INET6:
3074 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3075 			xprt_set_bound(xprt);
3076 
3077 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3078 		break;
3079 	default:
3080 		ret = ERR_PTR(-EAFNOSUPPORT);
3081 		goto out_err;
3082 	}
3083 
3084 	if (xprt_bound(xprt))
3085 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3086 				xprt->address_strings[RPC_DISPLAY_ADDR],
3087 				xprt->address_strings[RPC_DISPLAY_PORT],
3088 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3089 	else
3090 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3091 				xprt->address_strings[RPC_DISPLAY_ADDR],
3092 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3093 
3094 	if (try_module_get(THIS_MODULE))
3095 		return xprt;
3096 	ret = ERR_PTR(-EINVAL);
3097 out_err:
3098 	xs_xprt_free(xprt);
3099 	return ret;
3100 }
3101 
3102 /**
3103  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3104  * @args: rpc transport creation arguments
3105  *
3106  */
3107 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3108 {
3109 	struct sockaddr *addr = args->dstaddr;
3110 	struct rpc_xprt *xprt;
3111 	struct sock_xprt *transport;
3112 	struct svc_sock *bc_sock;
3113 	struct rpc_xprt *ret;
3114 
3115 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3116 			xprt_tcp_slot_table_entries);
3117 	if (IS_ERR(xprt))
3118 		return xprt;
3119 	transport = container_of(xprt, struct sock_xprt, xprt);
3120 
3121 	xprt->prot = IPPROTO_TCP;
3122 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3123 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3124 	xprt->timeout = &xs_tcp_default_timeout;
3125 
3126 	/* backchannel */
3127 	xprt_set_bound(xprt);
3128 	xprt->bind_timeout = 0;
3129 	xprt->reestablish_timeout = 0;
3130 	xprt->idle_timeout = 0;
3131 
3132 	xprt->ops = &bc_tcp_ops;
3133 
3134 	switch (addr->sa_family) {
3135 	case AF_INET:
3136 		xs_format_peer_addresses(xprt, "tcp",
3137 					 RPCBIND_NETID_TCP);
3138 		break;
3139 	case AF_INET6:
3140 		xs_format_peer_addresses(xprt, "tcp",
3141 				   RPCBIND_NETID_TCP6);
3142 		break;
3143 	default:
3144 		ret = ERR_PTR(-EAFNOSUPPORT);
3145 		goto out_err;
3146 	}
3147 
3148 	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3149 			xprt->address_strings[RPC_DISPLAY_ADDR],
3150 			xprt->address_strings[RPC_DISPLAY_PORT],
3151 			xprt->address_strings[RPC_DISPLAY_PROTO]);
3152 
3153 	/*
3154 	 * Once we've associated a backchannel xprt with a connection,
3155 	 * we want to keep it around as long as the connection lasts,
3156 	 * in case we need to start using it for a backchannel again;
3157 	 * this reference won't be dropped until bc_xprt is destroyed.
3158 	 */
3159 	xprt_get(xprt);
3160 	args->bc_xprt->xpt_bc_xprt = xprt;
3161 	xprt->bc_xprt = args->bc_xprt;
3162 	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3163 	transport->sock = bc_sock->sk_sock;
3164 	transport->inet = bc_sock->sk_sk;
3165 
3166 	/*
3167 	 * Since we don't want connections for the backchannel, we set
3168 	 * the xprt status to connected
3169 	 */
3170 	xprt_set_connected(xprt);
3171 
3172 	if (try_module_get(THIS_MODULE))
3173 		return xprt;
3174 
3175 	args->bc_xprt->xpt_bc_xprt = NULL;
3176 	args->bc_xprt->xpt_bc_xps = NULL;
3177 	xprt_put(xprt);
3178 	ret = ERR_PTR(-EINVAL);
3179 out_err:
3180 	xs_xprt_free(xprt);
3181 	return ret;
3182 }
3183 
3184 static struct xprt_class	xs_local_transport = {
3185 	.list		= LIST_HEAD_INIT(xs_local_transport.list),
3186 	.name		= "named UNIX socket",
3187 	.owner		= THIS_MODULE,
3188 	.ident		= XPRT_TRANSPORT_LOCAL,
3189 	.setup		= xs_setup_local,
3190 };
3191 
3192 static struct xprt_class	xs_udp_transport = {
3193 	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
3194 	.name		= "udp",
3195 	.owner		= THIS_MODULE,
3196 	.ident		= XPRT_TRANSPORT_UDP,
3197 	.setup		= xs_setup_udp,
3198 };
3199 
3200 static struct xprt_class	xs_tcp_transport = {
3201 	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
3202 	.name		= "tcp",
3203 	.owner		= THIS_MODULE,
3204 	.ident		= XPRT_TRANSPORT_TCP,
3205 	.setup		= xs_setup_tcp,
3206 };
3207 
3208 static struct xprt_class	xs_bc_tcp_transport = {
3209 	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3210 	.name		= "tcp NFSv4.1 backchannel",
3211 	.owner		= THIS_MODULE,
3212 	.ident		= XPRT_TRANSPORT_BC_TCP,
3213 	.setup		= xs_setup_bc_tcp,
3214 };
3215 
3216 /**
3217  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3218  *
3219  */
3220 int init_socket_xprt(void)
3221 {
3222 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3223 	if (!sunrpc_table_header)
3224 		sunrpc_table_header = register_sysctl_table(sunrpc_table);
3225 #endif
3226 
3227 	xprt_register_transport(&xs_local_transport);
3228 	xprt_register_transport(&xs_udp_transport);
3229 	xprt_register_transport(&xs_tcp_transport);
3230 	xprt_register_transport(&xs_bc_tcp_transport);
3231 
3232 	return 0;
3233 }
3234 
3235 /**
3236  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3237  *
3238  */
3239 void cleanup_socket_xprt(void)
3240 {
3241 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3242 	if (sunrpc_table_header) {
3243 		unregister_sysctl_table(sunrpc_table_header);
3244 		sunrpc_table_header = NULL;
3245 	}
3246 #endif
3247 
3248 	xprt_unregister_transport(&xs_local_transport);
3249 	xprt_unregister_transport(&xs_udp_transport);
3250 	xprt_unregister_transport(&xs_tcp_transport);
3251 	xprt_unregister_transport(&xs_bc_tcp_transport);
3252 }
3253 
3254 static int param_set_uint_minmax(const char *val,
3255 		const struct kernel_param *kp,
3256 		unsigned int min, unsigned int max)
3257 {
3258 	unsigned int num;
3259 	int ret;
3260 
3261 	if (!val)
3262 		return -EINVAL;
3263 	ret = kstrtouint(val, 0, &num);
3264 	if (ret)
3265 		return ret;
3266 	if (num < min || num > max)
3267 		return -EINVAL;
3268 	*((unsigned int *)kp->arg) = num;
3269 	return 0;
3270 }
3271 
3272 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3273 {
3274 	if (kp->arg == &xprt_min_resvport)
3275 		return param_set_uint_minmax(val, kp,
3276 			RPC_MIN_RESVPORT,
3277 			xprt_max_resvport);
3278 	return param_set_uint_minmax(val, kp,
3279 			xprt_min_resvport,
3280 			RPC_MAX_RESVPORT);
3281 }
3282 
3283 static const struct kernel_param_ops param_ops_portnr = {
3284 	.set = param_set_portnr,
3285 	.get = param_get_uint,
3286 };
3287 
3288 #define param_check_portnr(name, p) \
3289 	__param_check(name, p, unsigned int);
3290 
3291 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3292 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3293 
3294 static int param_set_slot_table_size(const char *val,
3295 				     const struct kernel_param *kp)
3296 {
3297 	return param_set_uint_minmax(val, kp,
3298 			RPC_MIN_SLOT_TABLE,
3299 			RPC_MAX_SLOT_TABLE);
3300 }
3301 
3302 static const struct kernel_param_ops param_ops_slot_table_size = {
3303 	.set = param_set_slot_table_size,
3304 	.get = param_get_uint,
3305 };
3306 
3307 #define param_check_slot_table_size(name, p) \
3308 	__param_check(name, p, unsigned int);
3309 
3310 static int param_set_max_slot_table_size(const char *val,
3311 				     const struct kernel_param *kp)
3312 {
3313 	return param_set_uint_minmax(val, kp,
3314 			RPC_MIN_SLOT_TABLE,
3315 			RPC_MAX_SLOT_TABLE_LIMIT);
3316 }
3317 
3318 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3319 	.set = param_set_max_slot_table_size,
3320 	.get = param_get_uint,
3321 };
3322 
3323 #define param_check_max_slot_table_size(name, p) \
3324 	__param_check(name, p, unsigned int);
3325 
3326 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3327 		   slot_table_size, 0644);
3328 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3329 		   max_slot_table_size, 0644);
3330 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3331 		   slot_table_size, 0644);
3332 
3333