xref: /openbmc/linux/net/sunrpc/xprtsock.c (revision 90a53e44)
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(&xprt->recv_lock);
973 	rovr = xprt_lookup_rqst(xprt, *xp);
974 	if (!rovr)
975 		goto out_unlock;
976 	xprt_pin_rqst(rovr);
977 	spin_unlock(&xprt->recv_lock);
978 	task = rovr->rq_task;
979 
980 	copied = rovr->rq_private_buf.buflen;
981 	if (copied > repsize)
982 		copied = repsize;
983 
984 	if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
985 		dprintk("RPC:       sk_buff copy failed\n");
986 		spin_lock(&xprt->recv_lock);
987 		goto out_unpin;
988 	}
989 
990 	spin_lock(&xprt->recv_lock);
991 	xprt_complete_rqst(task, copied);
992 out_unpin:
993 	xprt_unpin_rqst(rovr);
994  out_unlock:
995 	spin_unlock(&xprt->recv_lock);
996 }
997 
998 static void xs_local_data_receive(struct sock_xprt *transport)
999 {
1000 	struct sk_buff *skb;
1001 	struct sock *sk;
1002 	int err;
1003 
1004 	mutex_lock(&transport->recv_mutex);
1005 	sk = transport->inet;
1006 	if (sk == NULL)
1007 		goto out;
1008 	for (;;) {
1009 		skb = skb_recv_datagram(sk, 0, 1, &err);
1010 		if (skb != NULL) {
1011 			xs_local_data_read_skb(&transport->xprt, sk, skb);
1012 			skb_free_datagram(sk, skb);
1013 			continue;
1014 		}
1015 		if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1016 			break;
1017 	}
1018 out:
1019 	mutex_unlock(&transport->recv_mutex);
1020 }
1021 
1022 static void xs_local_data_receive_workfn(struct work_struct *work)
1023 {
1024 	struct sock_xprt *transport =
1025 		container_of(work, struct sock_xprt, recv_worker);
1026 	xs_local_data_receive(transport);
1027 }
1028 
1029 /**
1030  * xs_udp_data_read_skb - receive callback for UDP sockets
1031  * @xprt: transport
1032  * @sk: socket
1033  * @skb: skbuff
1034  *
1035  */
1036 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1037 		struct sock *sk,
1038 		struct sk_buff *skb)
1039 {
1040 	struct rpc_task *task;
1041 	struct rpc_rqst *rovr;
1042 	int repsize, copied;
1043 	u32 _xid;
1044 	__be32 *xp;
1045 
1046 	repsize = skb->len;
1047 	if (repsize < 4) {
1048 		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1049 		return;
1050 	}
1051 
1052 	/* Copy the XID from the skb... */
1053 	xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1054 	if (xp == NULL)
1055 		return;
1056 
1057 	/* Look up and lock the request corresponding to the given XID */
1058 	spin_lock(&xprt->recv_lock);
1059 	rovr = xprt_lookup_rqst(xprt, *xp);
1060 	if (!rovr)
1061 		goto out_unlock;
1062 	xprt_pin_rqst(rovr);
1063 	spin_unlock(&xprt->recv_lock);
1064 	task = rovr->rq_task;
1065 
1066 	if ((copied = rovr->rq_private_buf.buflen) > repsize)
1067 		copied = repsize;
1068 
1069 	/* Suck it into the iovec, verify checksum if not done by hw. */
1070 	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1071 		__UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1072 		spin_lock(&xprt->recv_lock);
1073 		goto out_unpin;
1074 	}
1075 
1076 	__UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1077 
1078 	spin_lock_bh(&xprt->transport_lock);
1079 	xprt_adjust_cwnd(xprt, task, copied);
1080 	spin_unlock_bh(&xprt->transport_lock);
1081 	spin_lock(&xprt->recv_lock);
1082 	xprt_complete_rqst(task, copied);
1083 out_unpin:
1084 	xprt_unpin_rqst(rovr);
1085  out_unlock:
1086 	spin_unlock(&xprt->recv_lock);
1087 }
1088 
1089 static void xs_udp_data_receive(struct sock_xprt *transport)
1090 {
1091 	struct sk_buff *skb;
1092 	struct sock *sk;
1093 	int err;
1094 
1095 	mutex_lock(&transport->recv_mutex);
1096 	sk = transport->inet;
1097 	if (sk == NULL)
1098 		goto out;
1099 	for (;;) {
1100 		skb = skb_recv_udp(sk, 0, 1, &err);
1101 		if (skb != NULL) {
1102 			xs_udp_data_read_skb(&transport->xprt, sk, skb);
1103 			consume_skb(skb);
1104 			continue;
1105 		}
1106 		if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1107 			break;
1108 	}
1109 out:
1110 	mutex_unlock(&transport->recv_mutex);
1111 }
1112 
1113 static void xs_udp_data_receive_workfn(struct work_struct *work)
1114 {
1115 	struct sock_xprt *transport =
1116 		container_of(work, struct sock_xprt, recv_worker);
1117 	xs_udp_data_receive(transport);
1118 }
1119 
1120 /**
1121  * xs_data_ready - "data ready" callback for UDP sockets
1122  * @sk: socket with data to read
1123  *
1124  */
1125 static void xs_data_ready(struct sock *sk)
1126 {
1127 	struct rpc_xprt *xprt;
1128 
1129 	read_lock_bh(&sk->sk_callback_lock);
1130 	dprintk("RPC:       xs_data_ready...\n");
1131 	xprt = xprt_from_sock(sk);
1132 	if (xprt != NULL) {
1133 		struct sock_xprt *transport = container_of(xprt,
1134 				struct sock_xprt, xprt);
1135 		transport->old_data_ready(sk);
1136 		/* Any data means we had a useful conversation, so
1137 		 * then we don't need to delay the next reconnect
1138 		 */
1139 		if (xprt->reestablish_timeout)
1140 			xprt->reestablish_timeout = 0;
1141 		if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1142 			queue_work(xprtiod_workqueue, &transport->recv_worker);
1143 	}
1144 	read_unlock_bh(&sk->sk_callback_lock);
1145 }
1146 
1147 /*
1148  * Helper function to force a TCP close if the server is sending
1149  * junk and/or it has put us in CLOSE_WAIT
1150  */
1151 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1152 {
1153 	xprt_force_disconnect(xprt);
1154 }
1155 
1156 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1157 {
1158 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1159 	size_t len, used;
1160 	char *p;
1161 
1162 	p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1163 	len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1164 	used = xdr_skb_read_bits(desc, p, len);
1165 	transport->tcp_offset += used;
1166 	if (used != len)
1167 		return;
1168 
1169 	transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1170 	if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1171 		transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1172 	else
1173 		transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1174 	transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1175 
1176 	transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1177 	transport->tcp_offset = 0;
1178 
1179 	/* Sanity check of the record length */
1180 	if (unlikely(transport->tcp_reclen < 8)) {
1181 		dprintk("RPC:       invalid TCP record fragment length\n");
1182 		xs_tcp_force_close(xprt);
1183 		return;
1184 	}
1185 	dprintk("RPC:       reading TCP record fragment of length %d\n",
1186 			transport->tcp_reclen);
1187 }
1188 
1189 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1190 {
1191 	if (transport->tcp_offset == transport->tcp_reclen) {
1192 		transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1193 		transport->tcp_offset = 0;
1194 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1195 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1196 			transport->tcp_flags |= TCP_RCV_COPY_XID;
1197 			transport->tcp_copied = 0;
1198 		}
1199 	}
1200 }
1201 
1202 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1203 {
1204 	size_t len, used;
1205 	char *p;
1206 
1207 	len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1208 	dprintk("RPC:       reading XID (%zu bytes)\n", len);
1209 	p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1210 	used = xdr_skb_read_bits(desc, p, len);
1211 	transport->tcp_offset += used;
1212 	if (used != len)
1213 		return;
1214 	transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1215 	transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1216 	transport->tcp_copied = 4;
1217 	dprintk("RPC:       reading %s XID %08x\n",
1218 			(transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1219 							      : "request with",
1220 			ntohl(transport->tcp_xid));
1221 	xs_tcp_check_fraghdr(transport);
1222 }
1223 
1224 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1225 				       struct xdr_skb_reader *desc)
1226 {
1227 	size_t len, used;
1228 	u32 offset;
1229 	char *p;
1230 
1231 	/*
1232 	 * We want transport->tcp_offset to be 8 at the end of this routine
1233 	 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1234 	 * When this function is called for the first time,
1235 	 * transport->tcp_offset is 4 (after having already read the xid).
1236 	 */
1237 	offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1238 	len = sizeof(transport->tcp_calldir) - offset;
1239 	dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1240 	p = ((char *) &transport->tcp_calldir) + offset;
1241 	used = xdr_skb_read_bits(desc, p, len);
1242 	transport->tcp_offset += used;
1243 	if (used != len)
1244 		return;
1245 	transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1246 	/*
1247 	 * We don't yet have the XDR buffer, so we will write the calldir
1248 	 * out after we get the buffer from the 'struct rpc_rqst'
1249 	 */
1250 	switch (ntohl(transport->tcp_calldir)) {
1251 	case RPC_REPLY:
1252 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1253 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1254 		transport->tcp_flags |= TCP_RPC_REPLY;
1255 		break;
1256 	case RPC_CALL:
1257 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1258 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1259 		transport->tcp_flags &= ~TCP_RPC_REPLY;
1260 		break;
1261 	default:
1262 		dprintk("RPC:       invalid request message type\n");
1263 		xs_tcp_force_close(&transport->xprt);
1264 	}
1265 	xs_tcp_check_fraghdr(transport);
1266 }
1267 
1268 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1269 				     struct xdr_skb_reader *desc,
1270 				     struct rpc_rqst *req)
1271 {
1272 	struct sock_xprt *transport =
1273 				container_of(xprt, struct sock_xprt, xprt);
1274 	struct xdr_buf *rcvbuf;
1275 	size_t len;
1276 	ssize_t r;
1277 
1278 	rcvbuf = &req->rq_private_buf;
1279 
1280 	if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1281 		/*
1282 		 * Save the RPC direction in the XDR buffer
1283 		 */
1284 		memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1285 			&transport->tcp_calldir,
1286 			sizeof(transport->tcp_calldir));
1287 		transport->tcp_copied += sizeof(transport->tcp_calldir);
1288 		transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1289 	}
1290 
1291 	len = desc->count;
1292 	if (len > transport->tcp_reclen - transport->tcp_offset)
1293 		desc->count = transport->tcp_reclen - transport->tcp_offset;
1294 	r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1295 					  desc, xdr_skb_read_bits);
1296 
1297 	if (desc->count) {
1298 		/* Error when copying to the receive buffer,
1299 		 * usually because we weren't able to allocate
1300 		 * additional buffer pages. All we can do now
1301 		 * is turn off TCP_RCV_COPY_DATA, so the request
1302 		 * will not receive any additional updates,
1303 		 * and time out.
1304 		 * Any remaining data from this record will
1305 		 * be discarded.
1306 		 */
1307 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1308 		dprintk("RPC:       XID %08x truncated request\n",
1309 				ntohl(transport->tcp_xid));
1310 		dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1311 				"tcp_offset = %u, tcp_reclen = %u\n",
1312 				xprt, transport->tcp_copied,
1313 				transport->tcp_offset, transport->tcp_reclen);
1314 		return;
1315 	}
1316 
1317 	transport->tcp_copied += r;
1318 	transport->tcp_offset += r;
1319 	desc->count = len - r;
1320 
1321 	dprintk("RPC:       XID %08x read %zd bytes\n",
1322 			ntohl(transport->tcp_xid), r);
1323 	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1324 			"tcp_reclen = %u\n", xprt, transport->tcp_copied,
1325 			transport->tcp_offset, transport->tcp_reclen);
1326 
1327 	if (transport->tcp_copied == req->rq_private_buf.buflen)
1328 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1329 	else if (transport->tcp_offset == transport->tcp_reclen) {
1330 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1331 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1332 	}
1333 }
1334 
1335 /*
1336  * Finds the request corresponding to the RPC xid and invokes the common
1337  * tcp read code to read the data.
1338  */
1339 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1340 				    struct xdr_skb_reader *desc)
1341 {
1342 	struct sock_xprt *transport =
1343 				container_of(xprt, struct sock_xprt, xprt);
1344 	struct rpc_rqst *req;
1345 
1346 	dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1347 
1348 	/* Find and lock the request corresponding to this xid */
1349 	spin_lock(&xprt->recv_lock);
1350 	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1351 	if (!req) {
1352 		dprintk("RPC:       XID %08x request not found!\n",
1353 				ntohl(transport->tcp_xid));
1354 		spin_unlock(&xprt->recv_lock);
1355 		return -1;
1356 	}
1357 	xprt_pin_rqst(req);
1358 	spin_unlock(&xprt->recv_lock);
1359 
1360 	xs_tcp_read_common(xprt, desc, req);
1361 
1362 	spin_lock(&xprt->recv_lock);
1363 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1364 		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1365 	xprt_unpin_rqst(req);
1366 	spin_unlock(&xprt->recv_lock);
1367 	return 0;
1368 }
1369 
1370 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1371 /*
1372  * Obtains an rpc_rqst previously allocated and invokes the common
1373  * tcp read code to read the data.  The result is placed in the callback
1374  * queue.
1375  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1376  * connection and return -1.
1377  */
1378 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1379 				       struct xdr_skb_reader *desc)
1380 {
1381 	struct sock_xprt *transport =
1382 				container_of(xprt, struct sock_xprt, xprt);
1383 	struct rpc_rqst *req;
1384 
1385 	/* Look up the request corresponding to the given XID */
1386 	req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1387 	if (req == NULL) {
1388 		printk(KERN_WARNING "Callback slot table overflowed\n");
1389 		xprt_force_disconnect(xprt);
1390 		return -1;
1391 	}
1392 
1393 	dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1394 	xs_tcp_read_common(xprt, desc, req);
1395 
1396 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1397 		xprt_complete_bc_request(req, transport->tcp_copied);
1398 
1399 	return 0;
1400 }
1401 
1402 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1403 					struct xdr_skb_reader *desc)
1404 {
1405 	struct sock_xprt *transport =
1406 				container_of(xprt, struct sock_xprt, xprt);
1407 
1408 	return (transport->tcp_flags & TCP_RPC_REPLY) ?
1409 		xs_tcp_read_reply(xprt, desc) :
1410 		xs_tcp_read_callback(xprt, desc);
1411 }
1412 
1413 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1414 {
1415 	int ret;
1416 
1417 	ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1418 			      SVC_SOCK_ANONYMOUS);
1419 	if (ret < 0)
1420 		return ret;
1421 	return 0;
1422 }
1423 
1424 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1425 {
1426 	return PAGE_SIZE;
1427 }
1428 #else
1429 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1430 					struct xdr_skb_reader *desc)
1431 {
1432 	return xs_tcp_read_reply(xprt, desc);
1433 }
1434 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1435 
1436 /*
1437  * Read data off the transport.  This can be either an RPC_CALL or an
1438  * RPC_REPLY.  Relay the processing to helper functions.
1439  */
1440 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1441 				    struct xdr_skb_reader *desc)
1442 {
1443 	struct sock_xprt *transport =
1444 				container_of(xprt, struct sock_xprt, xprt);
1445 
1446 	if (_xs_tcp_read_data(xprt, desc) == 0)
1447 		xs_tcp_check_fraghdr(transport);
1448 	else {
1449 		/*
1450 		 * The transport_lock protects the request handling.
1451 		 * There's no need to hold it to update the tcp_flags.
1452 		 */
1453 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1454 	}
1455 }
1456 
1457 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1458 {
1459 	size_t len;
1460 
1461 	len = transport->tcp_reclen - transport->tcp_offset;
1462 	if (len > desc->count)
1463 		len = desc->count;
1464 	desc->count -= len;
1465 	desc->offset += len;
1466 	transport->tcp_offset += len;
1467 	dprintk("RPC:       discarded %zu bytes\n", len);
1468 	xs_tcp_check_fraghdr(transport);
1469 }
1470 
1471 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1472 {
1473 	struct rpc_xprt *xprt = rd_desc->arg.data;
1474 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1475 	struct xdr_skb_reader desc = {
1476 		.skb	= skb,
1477 		.offset	= offset,
1478 		.count	= len,
1479 	};
1480 
1481 	dprintk("RPC:       xs_tcp_data_recv started\n");
1482 	do {
1483 		trace_xs_tcp_data_recv(transport);
1484 		/* Read in a new fragment marker if necessary */
1485 		/* Can we ever really expect to get completely empty fragments? */
1486 		if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1487 			xs_tcp_read_fraghdr(xprt, &desc);
1488 			continue;
1489 		}
1490 		/* Read in the xid if necessary */
1491 		if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1492 			xs_tcp_read_xid(transport, &desc);
1493 			continue;
1494 		}
1495 		/* Read in the call/reply flag */
1496 		if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1497 			xs_tcp_read_calldir(transport, &desc);
1498 			continue;
1499 		}
1500 		/* Read in the request data */
1501 		if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1502 			xs_tcp_read_data(xprt, &desc);
1503 			continue;
1504 		}
1505 		/* Skip over any trailing bytes on short reads */
1506 		xs_tcp_read_discard(transport, &desc);
1507 	} while (desc.count);
1508 	trace_xs_tcp_data_recv(transport);
1509 	dprintk("RPC:       xs_tcp_data_recv done\n");
1510 	return len - desc.count;
1511 }
1512 
1513 static void xs_tcp_data_receive(struct sock_xprt *transport)
1514 {
1515 	struct rpc_xprt *xprt = &transport->xprt;
1516 	struct sock *sk;
1517 	read_descriptor_t rd_desc = {
1518 		.count = 2*1024*1024,
1519 		.arg.data = xprt,
1520 	};
1521 	unsigned long total = 0;
1522 	int loop;
1523 	int read = 0;
1524 
1525 	mutex_lock(&transport->recv_mutex);
1526 	sk = transport->inet;
1527 	if (sk == NULL)
1528 		goto out;
1529 
1530 	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1531 	for (loop = 0; loop < 64; loop++) {
1532 		lock_sock(sk);
1533 		read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1534 		if (read <= 0) {
1535 			clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1536 			release_sock(sk);
1537 			break;
1538 		}
1539 		release_sock(sk);
1540 		total += read;
1541 		rd_desc.count = 65536;
1542 	}
1543 	if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1544 		queue_work(xprtiod_workqueue, &transport->recv_worker);
1545 out:
1546 	mutex_unlock(&transport->recv_mutex);
1547 	trace_xs_tcp_data_ready(xprt, read, total);
1548 }
1549 
1550 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1551 {
1552 	struct sock_xprt *transport =
1553 		container_of(work, struct sock_xprt, recv_worker);
1554 	xs_tcp_data_receive(transport);
1555 }
1556 
1557 /**
1558  * xs_tcp_state_change - callback to handle TCP socket state changes
1559  * @sk: socket whose state has changed
1560  *
1561  */
1562 static void xs_tcp_state_change(struct sock *sk)
1563 {
1564 	struct rpc_xprt *xprt;
1565 	struct sock_xprt *transport;
1566 
1567 	read_lock_bh(&sk->sk_callback_lock);
1568 	if (!(xprt = xprt_from_sock(sk)))
1569 		goto out;
1570 	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1571 	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1572 			sk->sk_state, xprt_connected(xprt),
1573 			sock_flag(sk, SOCK_DEAD),
1574 			sock_flag(sk, SOCK_ZAPPED),
1575 			sk->sk_shutdown);
1576 
1577 	transport = container_of(xprt, struct sock_xprt, xprt);
1578 	trace_rpc_socket_state_change(xprt, sk->sk_socket);
1579 	switch (sk->sk_state) {
1580 	case TCP_ESTABLISHED:
1581 		spin_lock(&xprt->transport_lock);
1582 		if (!xprt_test_and_set_connected(xprt)) {
1583 
1584 			/* Reset TCP record info */
1585 			transport->tcp_offset = 0;
1586 			transport->tcp_reclen = 0;
1587 			transport->tcp_copied = 0;
1588 			transport->tcp_flags =
1589 				TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1590 			xprt->connect_cookie++;
1591 			clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1592 			xprt_clear_connecting(xprt);
1593 
1594 			xprt_wake_pending_tasks(xprt, -EAGAIN);
1595 		}
1596 		spin_unlock(&xprt->transport_lock);
1597 		break;
1598 	case TCP_FIN_WAIT1:
1599 		/* The client initiated a shutdown of the socket */
1600 		xprt->connect_cookie++;
1601 		xprt->reestablish_timeout = 0;
1602 		set_bit(XPRT_CLOSING, &xprt->state);
1603 		smp_mb__before_atomic();
1604 		clear_bit(XPRT_CONNECTED, &xprt->state);
1605 		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1606 		smp_mb__after_atomic();
1607 		break;
1608 	case TCP_CLOSE_WAIT:
1609 		/* The server initiated a shutdown of the socket */
1610 		xprt->connect_cookie++;
1611 		clear_bit(XPRT_CONNECTED, &xprt->state);
1612 		xs_tcp_force_close(xprt);
1613 	case TCP_CLOSING:
1614 		/*
1615 		 * If the server closed down the connection, make sure that
1616 		 * we back off before reconnecting
1617 		 */
1618 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1619 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1620 		break;
1621 	case TCP_LAST_ACK:
1622 		set_bit(XPRT_CLOSING, &xprt->state);
1623 		smp_mb__before_atomic();
1624 		clear_bit(XPRT_CONNECTED, &xprt->state);
1625 		smp_mb__after_atomic();
1626 		break;
1627 	case TCP_CLOSE:
1628 		if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1629 					&transport->sock_state))
1630 			xprt_clear_connecting(xprt);
1631 		if (sk->sk_err)
1632 			xprt_wake_pending_tasks(xprt, -sk->sk_err);
1633 		xs_sock_mark_closed(xprt);
1634 	}
1635  out:
1636 	read_unlock_bh(&sk->sk_callback_lock);
1637 }
1638 
1639 static void xs_write_space(struct sock *sk)
1640 {
1641 	struct socket_wq *wq;
1642 	struct rpc_xprt *xprt;
1643 
1644 	if (!sk->sk_socket)
1645 		return;
1646 	clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1647 
1648 	if (unlikely(!(xprt = xprt_from_sock(sk))))
1649 		return;
1650 	rcu_read_lock();
1651 	wq = rcu_dereference(sk->sk_wq);
1652 	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1653 		goto out;
1654 
1655 	xprt_write_space(xprt);
1656 out:
1657 	rcu_read_unlock();
1658 }
1659 
1660 /**
1661  * xs_udp_write_space - callback invoked when socket buffer space
1662  *                             becomes available
1663  * @sk: socket whose state has changed
1664  *
1665  * Called when more output buffer space is available for this socket.
1666  * We try not to wake our writers until they can make "significant"
1667  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1668  * with a bunch of small requests.
1669  */
1670 static void xs_udp_write_space(struct sock *sk)
1671 {
1672 	read_lock_bh(&sk->sk_callback_lock);
1673 
1674 	/* from net/core/sock.c:sock_def_write_space */
1675 	if (sock_writeable(sk))
1676 		xs_write_space(sk);
1677 
1678 	read_unlock_bh(&sk->sk_callback_lock);
1679 }
1680 
1681 /**
1682  * xs_tcp_write_space - callback invoked when socket buffer space
1683  *                             becomes available
1684  * @sk: socket whose state has changed
1685  *
1686  * Called when more output buffer space is available for this socket.
1687  * We try not to wake our writers until they can make "significant"
1688  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1689  * with a bunch of small requests.
1690  */
1691 static void xs_tcp_write_space(struct sock *sk)
1692 {
1693 	read_lock_bh(&sk->sk_callback_lock);
1694 
1695 	/* from net/core/stream.c:sk_stream_write_space */
1696 	if (sk_stream_is_writeable(sk))
1697 		xs_write_space(sk);
1698 
1699 	read_unlock_bh(&sk->sk_callback_lock);
1700 }
1701 
1702 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1703 {
1704 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1705 	struct sock *sk = transport->inet;
1706 
1707 	if (transport->rcvsize) {
1708 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1709 		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1710 	}
1711 	if (transport->sndsize) {
1712 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1713 		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1714 		sk->sk_write_space(sk);
1715 	}
1716 }
1717 
1718 /**
1719  * xs_udp_set_buffer_size - set send and receive limits
1720  * @xprt: generic transport
1721  * @sndsize: requested size of send buffer, in bytes
1722  * @rcvsize: requested size of receive buffer, in bytes
1723  *
1724  * Set socket send and receive buffer size limits.
1725  */
1726 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1727 {
1728 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1729 
1730 	transport->sndsize = 0;
1731 	if (sndsize)
1732 		transport->sndsize = sndsize + 1024;
1733 	transport->rcvsize = 0;
1734 	if (rcvsize)
1735 		transport->rcvsize = rcvsize + 1024;
1736 
1737 	xs_udp_do_set_buffer_size(xprt);
1738 }
1739 
1740 /**
1741  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1742  * @task: task that timed out
1743  *
1744  * Adjust the congestion window after a retransmit timeout has occurred.
1745  */
1746 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1747 {
1748 	spin_lock_bh(&xprt->transport_lock);
1749 	xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1750 	spin_unlock_bh(&xprt->transport_lock);
1751 }
1752 
1753 static unsigned short xs_get_random_port(void)
1754 {
1755 	unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1756 	unsigned short rand = (unsigned short) prandom_u32() % range;
1757 	return rand + xprt_min_resvport;
1758 }
1759 
1760 /**
1761  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1762  * @sock: socket
1763  *
1764  * Note that this function has to be called on all sockets that share the
1765  * same port, and it must be called before binding.
1766  */
1767 static void xs_sock_set_reuseport(struct socket *sock)
1768 {
1769 	int opt = 1;
1770 
1771 	kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1772 			(char *)&opt, sizeof(opt));
1773 }
1774 
1775 static unsigned short xs_sock_getport(struct socket *sock)
1776 {
1777 	struct sockaddr_storage buf;
1778 	int buflen;
1779 	unsigned short port = 0;
1780 
1781 	if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1782 		goto out;
1783 	switch (buf.ss_family) {
1784 	case AF_INET6:
1785 		port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1786 		break;
1787 	case AF_INET:
1788 		port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1789 	}
1790 out:
1791 	return port;
1792 }
1793 
1794 /**
1795  * xs_set_port - reset the port number in the remote endpoint address
1796  * @xprt: generic transport
1797  * @port: new port number
1798  *
1799  */
1800 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1801 {
1802 	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1803 
1804 	rpc_set_port(xs_addr(xprt), port);
1805 	xs_update_peer_port(xprt);
1806 }
1807 
1808 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1809 {
1810 	if (transport->srcport == 0)
1811 		transport->srcport = xs_sock_getport(sock);
1812 }
1813 
1814 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1815 {
1816 	unsigned short port = transport->srcport;
1817 
1818 	if (port == 0 && transport->xprt.resvport)
1819 		port = xs_get_random_port();
1820 	return port;
1821 }
1822 
1823 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1824 {
1825 	if (transport->srcport != 0)
1826 		transport->srcport = 0;
1827 	if (!transport->xprt.resvport)
1828 		return 0;
1829 	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1830 		return xprt_max_resvport;
1831 	return --port;
1832 }
1833 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1834 {
1835 	struct sockaddr_storage myaddr;
1836 	int err, nloop = 0;
1837 	unsigned short port = xs_get_srcport(transport);
1838 	unsigned short last;
1839 
1840 	/*
1841 	 * If we are asking for any ephemeral port (i.e. port == 0 &&
1842 	 * transport->xprt.resvport == 0), don't bind.  Let the local
1843 	 * port selection happen implicitly when the socket is used
1844 	 * (for example at connect time).
1845 	 *
1846 	 * This ensures that we can continue to establish TCP
1847 	 * connections even when all local ephemeral ports are already
1848 	 * a part of some TCP connection.  This makes no difference
1849 	 * for UDP sockets, but also doens't harm them.
1850 	 *
1851 	 * If we're asking for any reserved port (i.e. port == 0 &&
1852 	 * transport->xprt.resvport == 1) xs_get_srcport above will
1853 	 * ensure that port is non-zero and we will bind as needed.
1854 	 */
1855 	if (port == 0)
1856 		return 0;
1857 
1858 	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1859 	do {
1860 		rpc_set_port((struct sockaddr *)&myaddr, port);
1861 		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1862 				transport->xprt.addrlen);
1863 		if (err == 0) {
1864 			transport->srcport = port;
1865 			break;
1866 		}
1867 		last = port;
1868 		port = xs_next_srcport(transport, port);
1869 		if (port > last)
1870 			nloop++;
1871 	} while (err == -EADDRINUSE && nloop != 2);
1872 
1873 	if (myaddr.ss_family == AF_INET)
1874 		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1875 				&((struct sockaddr_in *)&myaddr)->sin_addr,
1876 				port, err ? "failed" : "ok", err);
1877 	else
1878 		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1879 				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1880 				port, err ? "failed" : "ok", err);
1881 	return err;
1882 }
1883 
1884 /*
1885  * We don't support autobind on AF_LOCAL sockets
1886  */
1887 static void xs_local_rpcbind(struct rpc_task *task)
1888 {
1889 	xprt_set_bound(task->tk_xprt);
1890 }
1891 
1892 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1893 {
1894 }
1895 
1896 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1897 static struct lock_class_key xs_key[2];
1898 static struct lock_class_key xs_slock_key[2];
1899 
1900 static inline void xs_reclassify_socketu(struct socket *sock)
1901 {
1902 	struct sock *sk = sock->sk;
1903 
1904 	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1905 		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1906 }
1907 
1908 static inline void xs_reclassify_socket4(struct socket *sock)
1909 {
1910 	struct sock *sk = sock->sk;
1911 
1912 	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1913 		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1914 }
1915 
1916 static inline void xs_reclassify_socket6(struct socket *sock)
1917 {
1918 	struct sock *sk = sock->sk;
1919 
1920 	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1921 		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1922 }
1923 
1924 static inline void xs_reclassify_socket(int family, struct socket *sock)
1925 {
1926 	if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1927 		return;
1928 
1929 	switch (family) {
1930 	case AF_LOCAL:
1931 		xs_reclassify_socketu(sock);
1932 		break;
1933 	case AF_INET:
1934 		xs_reclassify_socket4(sock);
1935 		break;
1936 	case AF_INET6:
1937 		xs_reclassify_socket6(sock);
1938 		break;
1939 	}
1940 }
1941 #else
1942 static inline void xs_reclassify_socket(int family, struct socket *sock)
1943 {
1944 }
1945 #endif
1946 
1947 static void xs_dummy_setup_socket(struct work_struct *work)
1948 {
1949 }
1950 
1951 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1952 		struct sock_xprt *transport, int family, int type,
1953 		int protocol, bool reuseport)
1954 {
1955 	struct socket *sock;
1956 	int err;
1957 
1958 	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1959 	if (err < 0) {
1960 		dprintk("RPC:       can't create %d transport socket (%d).\n",
1961 				protocol, -err);
1962 		goto out;
1963 	}
1964 	xs_reclassify_socket(family, sock);
1965 
1966 	if (reuseport)
1967 		xs_sock_set_reuseport(sock);
1968 
1969 	err = xs_bind(transport, sock);
1970 	if (err) {
1971 		sock_release(sock);
1972 		goto out;
1973 	}
1974 
1975 	return sock;
1976 out:
1977 	return ERR_PTR(err);
1978 }
1979 
1980 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1981 				      struct socket *sock)
1982 {
1983 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1984 									xprt);
1985 
1986 	if (!transport->inet) {
1987 		struct sock *sk = sock->sk;
1988 
1989 		write_lock_bh(&sk->sk_callback_lock);
1990 
1991 		xs_save_old_callbacks(transport, sk);
1992 
1993 		sk->sk_user_data = xprt;
1994 		sk->sk_data_ready = xs_data_ready;
1995 		sk->sk_write_space = xs_udp_write_space;
1996 		sock_set_flag(sk, SOCK_FASYNC);
1997 		sk->sk_error_report = xs_error_report;
1998 		sk->sk_allocation = GFP_NOIO;
1999 
2000 		xprt_clear_connected(xprt);
2001 
2002 		/* Reset to new socket */
2003 		transport->sock = sock;
2004 		transport->inet = sk;
2005 
2006 		write_unlock_bh(&sk->sk_callback_lock);
2007 	}
2008 
2009 	/* Tell the socket layer to start connecting... */
2010 	xprt->stat.connect_count++;
2011 	xprt->stat.connect_start = jiffies;
2012 	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2013 }
2014 
2015 /**
2016  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2017  * @transport: socket transport to connect
2018  */
2019 static int xs_local_setup_socket(struct sock_xprt *transport)
2020 {
2021 	struct rpc_xprt *xprt = &transport->xprt;
2022 	struct socket *sock;
2023 	int status = -EIO;
2024 
2025 	status = __sock_create(xprt->xprt_net, AF_LOCAL,
2026 					SOCK_STREAM, 0, &sock, 1);
2027 	if (status < 0) {
2028 		dprintk("RPC:       can't create AF_LOCAL "
2029 			"transport socket (%d).\n", -status);
2030 		goto out;
2031 	}
2032 	xs_reclassify_socket(AF_LOCAL, sock);
2033 
2034 	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2035 			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2036 
2037 	status = xs_local_finish_connecting(xprt, sock);
2038 	trace_rpc_socket_connect(xprt, sock, status);
2039 	switch (status) {
2040 	case 0:
2041 		dprintk("RPC:       xprt %p connected to %s\n",
2042 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2043 		xprt_set_connected(xprt);
2044 	case -ENOBUFS:
2045 		break;
2046 	case -ENOENT:
2047 		dprintk("RPC:       xprt %p: socket %s does not exist\n",
2048 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2049 		break;
2050 	case -ECONNREFUSED:
2051 		dprintk("RPC:       xprt %p: connection refused for %s\n",
2052 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2053 		break;
2054 	default:
2055 		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2056 				__func__, -status,
2057 				xprt->address_strings[RPC_DISPLAY_ADDR]);
2058 	}
2059 
2060 out:
2061 	xprt_clear_connecting(xprt);
2062 	xprt_wake_pending_tasks(xprt, status);
2063 	return status;
2064 }
2065 
2066 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2067 {
2068 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2069 	int ret;
2070 
2071 	 if (RPC_IS_ASYNC(task)) {
2072 		/*
2073 		 * We want the AF_LOCAL connect to be resolved in the
2074 		 * filesystem namespace of the process making the rpc
2075 		 * call.  Thus we connect synchronously.
2076 		 *
2077 		 * If we want to support asynchronous AF_LOCAL calls,
2078 		 * we'll need to figure out how to pass a namespace to
2079 		 * connect.
2080 		 */
2081 		rpc_exit(task, -ENOTCONN);
2082 		return;
2083 	}
2084 	ret = xs_local_setup_socket(transport);
2085 	if (ret && !RPC_IS_SOFTCONN(task))
2086 		msleep_interruptible(15000);
2087 }
2088 
2089 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2090 /*
2091  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2092  * know that we have exclusive access to the socket), to guard against
2093  * races with xs_reset_transport.
2094  */
2095 static void xs_set_memalloc(struct rpc_xprt *xprt)
2096 {
2097 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2098 			xprt);
2099 
2100 	/*
2101 	 * If there's no sock, then we have nothing to set. The
2102 	 * reconnecting process will get it for us.
2103 	 */
2104 	if (!transport->inet)
2105 		return;
2106 	if (atomic_read(&xprt->swapper))
2107 		sk_set_memalloc(transport->inet);
2108 }
2109 
2110 /**
2111  * xs_enable_swap - Tag this transport as being used for swap.
2112  * @xprt: transport to tag
2113  *
2114  * Take a reference to this transport on behalf of the rpc_clnt, and
2115  * optionally mark it for swapping if it wasn't already.
2116  */
2117 static int
2118 xs_enable_swap(struct rpc_xprt *xprt)
2119 {
2120 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2121 
2122 	if (atomic_inc_return(&xprt->swapper) != 1)
2123 		return 0;
2124 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2125 		return -ERESTARTSYS;
2126 	if (xs->inet)
2127 		sk_set_memalloc(xs->inet);
2128 	xprt_release_xprt(xprt, NULL);
2129 	return 0;
2130 }
2131 
2132 /**
2133  * xs_disable_swap - Untag this transport as being used for swap.
2134  * @xprt: transport to tag
2135  *
2136  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2137  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2138  */
2139 static void
2140 xs_disable_swap(struct rpc_xprt *xprt)
2141 {
2142 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2143 
2144 	if (!atomic_dec_and_test(&xprt->swapper))
2145 		return;
2146 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2147 		return;
2148 	if (xs->inet)
2149 		sk_clear_memalloc(xs->inet);
2150 	xprt_release_xprt(xprt, NULL);
2151 }
2152 #else
2153 static void xs_set_memalloc(struct rpc_xprt *xprt)
2154 {
2155 }
2156 
2157 static int
2158 xs_enable_swap(struct rpc_xprt *xprt)
2159 {
2160 	return -EINVAL;
2161 }
2162 
2163 static void
2164 xs_disable_swap(struct rpc_xprt *xprt)
2165 {
2166 }
2167 #endif
2168 
2169 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2170 {
2171 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2172 
2173 	if (!transport->inet) {
2174 		struct sock *sk = sock->sk;
2175 
2176 		write_lock_bh(&sk->sk_callback_lock);
2177 
2178 		xs_save_old_callbacks(transport, sk);
2179 
2180 		sk->sk_user_data = xprt;
2181 		sk->sk_data_ready = xs_data_ready;
2182 		sk->sk_write_space = xs_udp_write_space;
2183 		sock_set_flag(sk, SOCK_FASYNC);
2184 		sk->sk_allocation = GFP_NOIO;
2185 
2186 		xprt_set_connected(xprt);
2187 
2188 		/* Reset to new socket */
2189 		transport->sock = sock;
2190 		transport->inet = sk;
2191 
2192 		xs_set_memalloc(xprt);
2193 
2194 		write_unlock_bh(&sk->sk_callback_lock);
2195 	}
2196 	xs_udp_do_set_buffer_size(xprt);
2197 
2198 	xprt->stat.connect_start = jiffies;
2199 }
2200 
2201 static void xs_udp_setup_socket(struct work_struct *work)
2202 {
2203 	struct sock_xprt *transport =
2204 		container_of(work, struct sock_xprt, connect_worker.work);
2205 	struct rpc_xprt *xprt = &transport->xprt;
2206 	struct socket *sock = transport->sock;
2207 	int status = -EIO;
2208 
2209 	sock = xs_create_sock(xprt, transport,
2210 			xs_addr(xprt)->sa_family, SOCK_DGRAM,
2211 			IPPROTO_UDP, false);
2212 	if (IS_ERR(sock))
2213 		goto out;
2214 
2215 	dprintk("RPC:       worker connecting xprt %p via %s to "
2216 				"%s (port %s)\n", xprt,
2217 			xprt->address_strings[RPC_DISPLAY_PROTO],
2218 			xprt->address_strings[RPC_DISPLAY_ADDR],
2219 			xprt->address_strings[RPC_DISPLAY_PORT]);
2220 
2221 	xs_udp_finish_connecting(xprt, sock);
2222 	trace_rpc_socket_connect(xprt, sock, 0);
2223 	status = 0;
2224 out:
2225 	xprt_unlock_connect(xprt, transport);
2226 	xprt_clear_connecting(xprt);
2227 	xprt_wake_pending_tasks(xprt, status);
2228 }
2229 
2230 /**
2231  * xs_tcp_shutdown - gracefully shut down a TCP socket
2232  * @xprt: transport
2233  *
2234  * Initiates a graceful shutdown of the TCP socket by calling the
2235  * equivalent of shutdown(SHUT_RDWR);
2236  */
2237 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2238 {
2239 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2240 	struct socket *sock = transport->sock;
2241 
2242 	if (sock == NULL)
2243 		return;
2244 	if (xprt_connected(xprt)) {
2245 		kernel_sock_shutdown(sock, SHUT_RDWR);
2246 		trace_rpc_socket_shutdown(xprt, sock);
2247 	} else
2248 		xs_reset_transport(transport);
2249 }
2250 
2251 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2252 		struct socket *sock)
2253 {
2254 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2255 	unsigned int keepidle;
2256 	unsigned int keepcnt;
2257 	unsigned int opt_on = 1;
2258 	unsigned int timeo;
2259 
2260 	spin_lock_bh(&xprt->transport_lock);
2261 	keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2262 	keepcnt = xprt->timeout->to_retries + 1;
2263 	timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2264 		(xprt->timeout->to_retries + 1);
2265 	clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2266 	spin_unlock_bh(&xprt->transport_lock);
2267 
2268 	/* TCP Keepalive options */
2269 	kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2270 			(char *)&opt_on, sizeof(opt_on));
2271 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2272 			(char *)&keepidle, sizeof(keepidle));
2273 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2274 			(char *)&keepidle, sizeof(keepidle));
2275 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2276 			(char *)&keepcnt, sizeof(keepcnt));
2277 
2278 	/* TCP user timeout (see RFC5482) */
2279 	kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2280 			(char *)&timeo, sizeof(timeo));
2281 }
2282 
2283 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2284 		unsigned long connect_timeout,
2285 		unsigned long reconnect_timeout)
2286 {
2287 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2288 	struct rpc_timeout to;
2289 	unsigned long initval;
2290 
2291 	spin_lock_bh(&xprt->transport_lock);
2292 	if (reconnect_timeout < xprt->max_reconnect_timeout)
2293 		xprt->max_reconnect_timeout = reconnect_timeout;
2294 	if (connect_timeout < xprt->connect_timeout) {
2295 		memcpy(&to, xprt->timeout, sizeof(to));
2296 		initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2297 		/* Arbitrary lower limit */
2298 		if (initval <  XS_TCP_INIT_REEST_TO << 1)
2299 			initval = XS_TCP_INIT_REEST_TO << 1;
2300 		to.to_initval = initval;
2301 		to.to_maxval = initval;
2302 		memcpy(&transport->tcp_timeout, &to,
2303 				sizeof(transport->tcp_timeout));
2304 		xprt->timeout = &transport->tcp_timeout;
2305 		xprt->connect_timeout = connect_timeout;
2306 	}
2307 	set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2308 	spin_unlock_bh(&xprt->transport_lock);
2309 }
2310 
2311 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2312 {
2313 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2314 	int ret = -ENOTCONN;
2315 
2316 	if (!transport->inet) {
2317 		struct sock *sk = sock->sk;
2318 		unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2319 
2320 		/* Avoid temporary address, they are bad for long-lived
2321 		 * connections such as NFS mounts.
2322 		 * RFC4941, section 3.6 suggests that:
2323 		 *    Individual applications, which have specific
2324 		 *    knowledge about the normal duration of connections,
2325 		 *    MAY override this as appropriate.
2326 		 */
2327 		kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2328 				(char *)&addr_pref, sizeof(addr_pref));
2329 
2330 		xs_tcp_set_socket_timeouts(xprt, sock);
2331 
2332 		write_lock_bh(&sk->sk_callback_lock);
2333 
2334 		xs_save_old_callbacks(transport, sk);
2335 
2336 		sk->sk_user_data = xprt;
2337 		sk->sk_data_ready = xs_data_ready;
2338 		sk->sk_state_change = xs_tcp_state_change;
2339 		sk->sk_write_space = xs_tcp_write_space;
2340 		sock_set_flag(sk, SOCK_FASYNC);
2341 		sk->sk_error_report = xs_error_report;
2342 		sk->sk_allocation = GFP_NOIO;
2343 
2344 		/* socket options */
2345 		sock_reset_flag(sk, SOCK_LINGER);
2346 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2347 
2348 		xprt_clear_connected(xprt);
2349 
2350 		/* Reset to new socket */
2351 		transport->sock = sock;
2352 		transport->inet = sk;
2353 
2354 		write_unlock_bh(&sk->sk_callback_lock);
2355 	}
2356 
2357 	if (!xprt_bound(xprt))
2358 		goto out;
2359 
2360 	xs_set_memalloc(xprt);
2361 
2362 	/* Tell the socket layer to start connecting... */
2363 	xprt->stat.connect_count++;
2364 	xprt->stat.connect_start = jiffies;
2365 	set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2366 	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2367 	switch (ret) {
2368 	case 0:
2369 		xs_set_srcport(transport, sock);
2370 	case -EINPROGRESS:
2371 		/* SYN_SENT! */
2372 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2373 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2374 		break;
2375 	case -EADDRNOTAVAIL:
2376 		/* Source port number is unavailable. Try a new one! */
2377 		transport->srcport = 0;
2378 	}
2379 out:
2380 	return ret;
2381 }
2382 
2383 /**
2384  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2385  *
2386  * Invoked by a work queue tasklet.
2387  */
2388 static void xs_tcp_setup_socket(struct work_struct *work)
2389 {
2390 	struct sock_xprt *transport =
2391 		container_of(work, struct sock_xprt, connect_worker.work);
2392 	struct socket *sock = transport->sock;
2393 	struct rpc_xprt *xprt = &transport->xprt;
2394 	int status = -EIO;
2395 
2396 	if (!sock) {
2397 		sock = xs_create_sock(xprt, transport,
2398 				xs_addr(xprt)->sa_family, SOCK_STREAM,
2399 				IPPROTO_TCP, true);
2400 		if (IS_ERR(sock)) {
2401 			status = PTR_ERR(sock);
2402 			goto out;
2403 		}
2404 	}
2405 
2406 	dprintk("RPC:       worker connecting xprt %p via %s to "
2407 				"%s (port %s)\n", xprt,
2408 			xprt->address_strings[RPC_DISPLAY_PROTO],
2409 			xprt->address_strings[RPC_DISPLAY_ADDR],
2410 			xprt->address_strings[RPC_DISPLAY_PORT]);
2411 
2412 	status = xs_tcp_finish_connecting(xprt, sock);
2413 	trace_rpc_socket_connect(xprt, sock, status);
2414 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2415 			xprt, -status, xprt_connected(xprt),
2416 			sock->sk->sk_state);
2417 	switch (status) {
2418 	default:
2419 		printk("%s: connect returned unhandled error %d\n",
2420 			__func__, status);
2421 	case -EADDRNOTAVAIL:
2422 		/* We're probably in TIME_WAIT. Get rid of existing socket,
2423 		 * and retry
2424 		 */
2425 		xs_tcp_force_close(xprt);
2426 		break;
2427 	case 0:
2428 	case -EINPROGRESS:
2429 	case -EALREADY:
2430 		xprt_unlock_connect(xprt, transport);
2431 		return;
2432 	case -EINVAL:
2433 		/* Happens, for instance, if the user specified a link
2434 		 * local IPv6 address without a scope-id.
2435 		 */
2436 	case -ECONNREFUSED:
2437 	case -ECONNRESET:
2438 	case -ENETUNREACH:
2439 	case -EADDRINUSE:
2440 	case -ENOBUFS:
2441 		/*
2442 		 * xs_tcp_force_close() wakes tasks with -EIO.
2443 		 * We need to wake them first to ensure the
2444 		 * correct error code.
2445 		 */
2446 		xprt_wake_pending_tasks(xprt, status);
2447 		xs_tcp_force_close(xprt);
2448 		goto out;
2449 	}
2450 	status = -EAGAIN;
2451 out:
2452 	xprt_unlock_connect(xprt, transport);
2453 	xprt_clear_connecting(xprt);
2454 	xprt_wake_pending_tasks(xprt, status);
2455 }
2456 
2457 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2458 {
2459 	unsigned long start, now = jiffies;
2460 
2461 	start = xprt->stat.connect_start + xprt->reestablish_timeout;
2462 	if (time_after(start, now))
2463 		return start - now;
2464 	return 0;
2465 }
2466 
2467 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2468 {
2469 	xprt->reestablish_timeout <<= 1;
2470 	if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2471 		xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2472 	if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2473 		xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2474 }
2475 
2476 /**
2477  * xs_connect - connect a socket to a remote endpoint
2478  * @xprt: pointer to transport structure
2479  * @task: address of RPC task that manages state of connect request
2480  *
2481  * TCP: If the remote end dropped the connection, delay reconnecting.
2482  *
2483  * UDP socket connects are synchronous, but we use a work queue anyway
2484  * to guarantee that even unprivileged user processes can set up a
2485  * socket on a privileged port.
2486  *
2487  * If a UDP socket connect fails, the delay behavior here prevents
2488  * retry floods (hard mounts).
2489  */
2490 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2491 {
2492 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2493 	unsigned long delay = 0;
2494 
2495 	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2496 
2497 	if (transport->sock != NULL) {
2498 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2499 				"seconds\n",
2500 				xprt, xprt->reestablish_timeout / HZ);
2501 
2502 		/* Start by resetting any existing state */
2503 		xs_reset_transport(transport);
2504 
2505 		delay = xs_reconnect_delay(xprt);
2506 		xs_reconnect_backoff(xprt);
2507 
2508 	} else
2509 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2510 
2511 	queue_delayed_work(xprtiod_workqueue,
2512 			&transport->connect_worker,
2513 			delay);
2514 }
2515 
2516 /**
2517  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2518  * @xprt: rpc_xprt struct containing statistics
2519  * @seq: output file
2520  *
2521  */
2522 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2523 {
2524 	long idle_time = 0;
2525 
2526 	if (xprt_connected(xprt))
2527 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2528 
2529 	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2530 			"%llu %llu %lu %llu %llu\n",
2531 			xprt->stat.bind_count,
2532 			xprt->stat.connect_count,
2533 			xprt->stat.connect_time,
2534 			idle_time,
2535 			xprt->stat.sends,
2536 			xprt->stat.recvs,
2537 			xprt->stat.bad_xids,
2538 			xprt->stat.req_u,
2539 			xprt->stat.bklog_u,
2540 			xprt->stat.max_slots,
2541 			xprt->stat.sending_u,
2542 			xprt->stat.pending_u);
2543 }
2544 
2545 /**
2546  * xs_udp_print_stats - display UDP socket-specifc stats
2547  * @xprt: rpc_xprt struct containing statistics
2548  * @seq: output file
2549  *
2550  */
2551 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2552 {
2553 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2554 
2555 	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2556 			"%lu %llu %llu\n",
2557 			transport->srcport,
2558 			xprt->stat.bind_count,
2559 			xprt->stat.sends,
2560 			xprt->stat.recvs,
2561 			xprt->stat.bad_xids,
2562 			xprt->stat.req_u,
2563 			xprt->stat.bklog_u,
2564 			xprt->stat.max_slots,
2565 			xprt->stat.sending_u,
2566 			xprt->stat.pending_u);
2567 }
2568 
2569 /**
2570  * xs_tcp_print_stats - display TCP socket-specifc stats
2571  * @xprt: rpc_xprt struct containing statistics
2572  * @seq: output file
2573  *
2574  */
2575 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2576 {
2577 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2578 	long idle_time = 0;
2579 
2580 	if (xprt_connected(xprt))
2581 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2582 
2583 	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2584 			"%llu %llu %lu %llu %llu\n",
2585 			transport->srcport,
2586 			xprt->stat.bind_count,
2587 			xprt->stat.connect_count,
2588 			xprt->stat.connect_time,
2589 			idle_time,
2590 			xprt->stat.sends,
2591 			xprt->stat.recvs,
2592 			xprt->stat.bad_xids,
2593 			xprt->stat.req_u,
2594 			xprt->stat.bklog_u,
2595 			xprt->stat.max_slots,
2596 			xprt->stat.sending_u,
2597 			xprt->stat.pending_u);
2598 }
2599 
2600 /*
2601  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2602  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2603  * to use the server side send routines.
2604  */
2605 static int bc_malloc(struct rpc_task *task)
2606 {
2607 	struct rpc_rqst *rqst = task->tk_rqstp;
2608 	size_t size = rqst->rq_callsize;
2609 	struct page *page;
2610 	struct rpc_buffer *buf;
2611 
2612 	if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2613 		WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2614 			  size);
2615 		return -EINVAL;
2616 	}
2617 
2618 	page = alloc_page(GFP_KERNEL);
2619 	if (!page)
2620 		return -ENOMEM;
2621 
2622 	buf = page_address(page);
2623 	buf->len = PAGE_SIZE;
2624 
2625 	rqst->rq_buffer = buf->data;
2626 	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2627 	return 0;
2628 }
2629 
2630 /*
2631  * Free the space allocated in the bc_alloc routine
2632  */
2633 static void bc_free(struct rpc_task *task)
2634 {
2635 	void *buffer = task->tk_rqstp->rq_buffer;
2636 	struct rpc_buffer *buf;
2637 
2638 	buf = container_of(buffer, struct rpc_buffer, data);
2639 	free_page((unsigned long)buf);
2640 }
2641 
2642 /*
2643  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2644  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2645  */
2646 static int bc_sendto(struct rpc_rqst *req)
2647 {
2648 	int len;
2649 	struct xdr_buf *xbufp = &req->rq_snd_buf;
2650 	struct rpc_xprt *xprt = req->rq_xprt;
2651 	struct sock_xprt *transport =
2652 				container_of(xprt, struct sock_xprt, xprt);
2653 	struct socket *sock = transport->sock;
2654 	unsigned long headoff;
2655 	unsigned long tailoff;
2656 
2657 	xs_encode_stream_record_marker(xbufp);
2658 
2659 	tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2660 	headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2661 	len = svc_send_common(sock, xbufp,
2662 			      virt_to_page(xbufp->head[0].iov_base), headoff,
2663 			      xbufp->tail[0].iov_base, tailoff);
2664 
2665 	if (len != xbufp->len) {
2666 		printk(KERN_NOTICE "Error sending entire callback!\n");
2667 		len = -EAGAIN;
2668 	}
2669 
2670 	return len;
2671 }
2672 
2673 /*
2674  * The send routine. Borrows from svc_send
2675  */
2676 static int bc_send_request(struct rpc_task *task)
2677 {
2678 	struct rpc_rqst *req = task->tk_rqstp;
2679 	struct svc_xprt	*xprt;
2680 	int len;
2681 
2682 	dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2683 	/*
2684 	 * Get the server socket associated with this callback xprt
2685 	 */
2686 	xprt = req->rq_xprt->bc_xprt;
2687 
2688 	/*
2689 	 * Grab the mutex to serialize data as the connection is shared
2690 	 * with the fore channel
2691 	 */
2692 	if (!mutex_trylock(&xprt->xpt_mutex)) {
2693 		rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2694 		if (!mutex_trylock(&xprt->xpt_mutex))
2695 			return -EAGAIN;
2696 		rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2697 	}
2698 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2699 		len = -ENOTCONN;
2700 	else
2701 		len = bc_sendto(req);
2702 	mutex_unlock(&xprt->xpt_mutex);
2703 
2704 	if (len > 0)
2705 		len = 0;
2706 
2707 	return len;
2708 }
2709 
2710 /*
2711  * The close routine. Since this is client initiated, we do nothing
2712  */
2713 
2714 static void bc_close(struct rpc_xprt *xprt)
2715 {
2716 }
2717 
2718 /*
2719  * The xprt destroy routine. Again, because this connection is client
2720  * initiated, we do nothing
2721  */
2722 
2723 static void bc_destroy(struct rpc_xprt *xprt)
2724 {
2725 	dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2726 
2727 	xs_xprt_free(xprt);
2728 	module_put(THIS_MODULE);
2729 }
2730 
2731 static const struct rpc_xprt_ops xs_local_ops = {
2732 	.reserve_xprt		= xprt_reserve_xprt,
2733 	.release_xprt		= xs_tcp_release_xprt,
2734 	.alloc_slot		= xprt_alloc_slot,
2735 	.rpcbind		= xs_local_rpcbind,
2736 	.set_port		= xs_local_set_port,
2737 	.connect		= xs_local_connect,
2738 	.buf_alloc		= rpc_malloc,
2739 	.buf_free		= rpc_free,
2740 	.send_request		= xs_local_send_request,
2741 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2742 	.close			= xs_close,
2743 	.destroy		= xs_destroy,
2744 	.print_stats		= xs_local_print_stats,
2745 	.enable_swap		= xs_enable_swap,
2746 	.disable_swap		= xs_disable_swap,
2747 };
2748 
2749 static const struct rpc_xprt_ops xs_udp_ops = {
2750 	.set_buffer_size	= xs_udp_set_buffer_size,
2751 	.reserve_xprt		= xprt_reserve_xprt_cong,
2752 	.release_xprt		= xprt_release_xprt_cong,
2753 	.alloc_slot		= xprt_alloc_slot,
2754 	.rpcbind		= rpcb_getport_async,
2755 	.set_port		= xs_set_port,
2756 	.connect		= xs_connect,
2757 	.buf_alloc		= rpc_malloc,
2758 	.buf_free		= rpc_free,
2759 	.send_request		= xs_udp_send_request,
2760 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
2761 	.timer			= xs_udp_timer,
2762 	.release_request	= xprt_release_rqst_cong,
2763 	.close			= xs_close,
2764 	.destroy		= xs_destroy,
2765 	.print_stats		= xs_udp_print_stats,
2766 	.enable_swap		= xs_enable_swap,
2767 	.disable_swap		= xs_disable_swap,
2768 	.inject_disconnect	= xs_inject_disconnect,
2769 };
2770 
2771 static const struct rpc_xprt_ops xs_tcp_ops = {
2772 	.reserve_xprt		= xprt_reserve_xprt,
2773 	.release_xprt		= xs_tcp_release_xprt,
2774 	.alloc_slot		= xprt_lock_and_alloc_slot,
2775 	.rpcbind		= rpcb_getport_async,
2776 	.set_port		= xs_set_port,
2777 	.connect		= xs_connect,
2778 	.buf_alloc		= rpc_malloc,
2779 	.buf_free		= rpc_free,
2780 	.send_request		= xs_tcp_send_request,
2781 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2782 	.close			= xs_tcp_shutdown,
2783 	.destroy		= xs_destroy,
2784 	.set_connect_timeout	= xs_tcp_set_connect_timeout,
2785 	.print_stats		= xs_tcp_print_stats,
2786 	.enable_swap		= xs_enable_swap,
2787 	.disable_swap		= xs_disable_swap,
2788 	.inject_disconnect	= xs_inject_disconnect,
2789 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2790 	.bc_setup		= xprt_setup_bc,
2791 	.bc_up			= xs_tcp_bc_up,
2792 	.bc_maxpayload		= xs_tcp_bc_maxpayload,
2793 	.bc_free_rqst		= xprt_free_bc_rqst,
2794 	.bc_destroy		= xprt_destroy_bc,
2795 #endif
2796 };
2797 
2798 /*
2799  * The rpc_xprt_ops for the server backchannel
2800  */
2801 
2802 static const struct rpc_xprt_ops bc_tcp_ops = {
2803 	.reserve_xprt		= xprt_reserve_xprt,
2804 	.release_xprt		= xprt_release_xprt,
2805 	.alloc_slot		= xprt_alloc_slot,
2806 	.buf_alloc		= bc_malloc,
2807 	.buf_free		= bc_free,
2808 	.send_request		= bc_send_request,
2809 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2810 	.close			= bc_close,
2811 	.destroy		= bc_destroy,
2812 	.print_stats		= xs_tcp_print_stats,
2813 	.enable_swap		= xs_enable_swap,
2814 	.disable_swap		= xs_disable_swap,
2815 	.inject_disconnect	= xs_inject_disconnect,
2816 };
2817 
2818 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2819 {
2820 	static const struct sockaddr_in sin = {
2821 		.sin_family		= AF_INET,
2822 		.sin_addr.s_addr	= htonl(INADDR_ANY),
2823 	};
2824 	static const struct sockaddr_in6 sin6 = {
2825 		.sin6_family		= AF_INET6,
2826 		.sin6_addr		= IN6ADDR_ANY_INIT,
2827 	};
2828 
2829 	switch (family) {
2830 	case AF_LOCAL:
2831 		break;
2832 	case AF_INET:
2833 		memcpy(sap, &sin, sizeof(sin));
2834 		break;
2835 	case AF_INET6:
2836 		memcpy(sap, &sin6, sizeof(sin6));
2837 		break;
2838 	default:
2839 		dprintk("RPC:       %s: Bad address family\n", __func__);
2840 		return -EAFNOSUPPORT;
2841 	}
2842 	return 0;
2843 }
2844 
2845 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2846 				      unsigned int slot_table_size,
2847 				      unsigned int max_slot_table_size)
2848 {
2849 	struct rpc_xprt *xprt;
2850 	struct sock_xprt *new;
2851 
2852 	if (args->addrlen > sizeof(xprt->addr)) {
2853 		dprintk("RPC:       xs_setup_xprt: address too large\n");
2854 		return ERR_PTR(-EBADF);
2855 	}
2856 
2857 	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2858 			max_slot_table_size);
2859 	if (xprt == NULL) {
2860 		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2861 				"rpc_xprt\n");
2862 		return ERR_PTR(-ENOMEM);
2863 	}
2864 
2865 	new = container_of(xprt, struct sock_xprt, xprt);
2866 	mutex_init(&new->recv_mutex);
2867 	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2868 	xprt->addrlen = args->addrlen;
2869 	if (args->srcaddr)
2870 		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2871 	else {
2872 		int err;
2873 		err = xs_init_anyaddr(args->dstaddr->sa_family,
2874 					(struct sockaddr *)&new->srcaddr);
2875 		if (err != 0) {
2876 			xprt_free(xprt);
2877 			return ERR_PTR(err);
2878 		}
2879 	}
2880 
2881 	return xprt;
2882 }
2883 
2884 static const struct rpc_timeout xs_local_default_timeout = {
2885 	.to_initval = 10 * HZ,
2886 	.to_maxval = 10 * HZ,
2887 	.to_retries = 2,
2888 };
2889 
2890 /**
2891  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2892  * @args: rpc transport creation arguments
2893  *
2894  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2895  */
2896 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2897 {
2898 	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2899 	struct sock_xprt *transport;
2900 	struct rpc_xprt *xprt;
2901 	struct rpc_xprt *ret;
2902 
2903 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2904 			xprt_max_tcp_slot_table_entries);
2905 	if (IS_ERR(xprt))
2906 		return xprt;
2907 	transport = container_of(xprt, struct sock_xprt, xprt);
2908 
2909 	xprt->prot = 0;
2910 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2911 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2912 
2913 	xprt->bind_timeout = XS_BIND_TO;
2914 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2915 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2916 
2917 	xprt->ops = &xs_local_ops;
2918 	xprt->timeout = &xs_local_default_timeout;
2919 
2920 	INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2921 	INIT_DELAYED_WORK(&transport->connect_worker,
2922 			xs_dummy_setup_socket);
2923 
2924 	switch (sun->sun_family) {
2925 	case AF_LOCAL:
2926 		if (sun->sun_path[0] != '/') {
2927 			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2928 					sun->sun_path);
2929 			ret = ERR_PTR(-EINVAL);
2930 			goto out_err;
2931 		}
2932 		xprt_set_bound(xprt);
2933 		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2934 		ret = ERR_PTR(xs_local_setup_socket(transport));
2935 		if (ret)
2936 			goto out_err;
2937 		break;
2938 	default:
2939 		ret = ERR_PTR(-EAFNOSUPPORT);
2940 		goto out_err;
2941 	}
2942 
2943 	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2944 			xprt->address_strings[RPC_DISPLAY_ADDR]);
2945 
2946 	if (try_module_get(THIS_MODULE))
2947 		return xprt;
2948 	ret = ERR_PTR(-EINVAL);
2949 out_err:
2950 	xs_xprt_free(xprt);
2951 	return ret;
2952 }
2953 
2954 static const struct rpc_timeout xs_udp_default_timeout = {
2955 	.to_initval = 5 * HZ,
2956 	.to_maxval = 30 * HZ,
2957 	.to_increment = 5 * HZ,
2958 	.to_retries = 5,
2959 };
2960 
2961 /**
2962  * xs_setup_udp - Set up transport to use a UDP socket
2963  * @args: rpc transport creation arguments
2964  *
2965  */
2966 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2967 {
2968 	struct sockaddr *addr = args->dstaddr;
2969 	struct rpc_xprt *xprt;
2970 	struct sock_xprt *transport;
2971 	struct rpc_xprt *ret;
2972 
2973 	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2974 			xprt_udp_slot_table_entries);
2975 	if (IS_ERR(xprt))
2976 		return xprt;
2977 	transport = container_of(xprt, struct sock_xprt, xprt);
2978 
2979 	xprt->prot = IPPROTO_UDP;
2980 	xprt->tsh_size = 0;
2981 	/* XXX: header size can vary due to auth type, IPv6, etc. */
2982 	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2983 
2984 	xprt->bind_timeout = XS_BIND_TO;
2985 	xprt->reestablish_timeout = XS_UDP_REEST_TO;
2986 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2987 
2988 	xprt->ops = &xs_udp_ops;
2989 
2990 	xprt->timeout = &xs_udp_default_timeout;
2991 
2992 	INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2993 	INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
2994 
2995 	switch (addr->sa_family) {
2996 	case AF_INET:
2997 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2998 			xprt_set_bound(xprt);
2999 
3000 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3001 		break;
3002 	case AF_INET6:
3003 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3004 			xprt_set_bound(xprt);
3005 
3006 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3007 		break;
3008 	default:
3009 		ret = ERR_PTR(-EAFNOSUPPORT);
3010 		goto out_err;
3011 	}
3012 
3013 	if (xprt_bound(xprt))
3014 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3015 				xprt->address_strings[RPC_DISPLAY_ADDR],
3016 				xprt->address_strings[RPC_DISPLAY_PORT],
3017 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3018 	else
3019 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3020 				xprt->address_strings[RPC_DISPLAY_ADDR],
3021 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3022 
3023 	if (try_module_get(THIS_MODULE))
3024 		return xprt;
3025 	ret = ERR_PTR(-EINVAL);
3026 out_err:
3027 	xs_xprt_free(xprt);
3028 	return ret;
3029 }
3030 
3031 static const struct rpc_timeout xs_tcp_default_timeout = {
3032 	.to_initval = 60 * HZ,
3033 	.to_maxval = 60 * HZ,
3034 	.to_retries = 2,
3035 };
3036 
3037 /**
3038  * xs_setup_tcp - Set up transport to use a TCP socket
3039  * @args: rpc transport creation arguments
3040  *
3041  */
3042 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3043 {
3044 	struct sockaddr *addr = args->dstaddr;
3045 	struct rpc_xprt *xprt;
3046 	struct sock_xprt *transport;
3047 	struct rpc_xprt *ret;
3048 	unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3049 
3050 	if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3051 		max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3052 
3053 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3054 			max_slot_table_size);
3055 	if (IS_ERR(xprt))
3056 		return xprt;
3057 	transport = container_of(xprt, struct sock_xprt, xprt);
3058 
3059 	xprt->prot = IPPROTO_TCP;
3060 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3061 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3062 
3063 	xprt->bind_timeout = XS_BIND_TO;
3064 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3065 	xprt->idle_timeout = XS_IDLE_DISC_TO;
3066 
3067 	xprt->ops = &xs_tcp_ops;
3068 	xprt->timeout = &xs_tcp_default_timeout;
3069 
3070 	xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3071 	xprt->connect_timeout = xprt->timeout->to_initval *
3072 		(xprt->timeout->to_retries + 1);
3073 
3074 	INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3075 	INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3076 
3077 	switch (addr->sa_family) {
3078 	case AF_INET:
3079 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3080 			xprt_set_bound(xprt);
3081 
3082 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3083 		break;
3084 	case AF_INET6:
3085 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3086 			xprt_set_bound(xprt);
3087 
3088 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3089 		break;
3090 	default:
3091 		ret = ERR_PTR(-EAFNOSUPPORT);
3092 		goto out_err;
3093 	}
3094 
3095 	if (xprt_bound(xprt))
3096 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3097 				xprt->address_strings[RPC_DISPLAY_ADDR],
3098 				xprt->address_strings[RPC_DISPLAY_PORT],
3099 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3100 	else
3101 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3102 				xprt->address_strings[RPC_DISPLAY_ADDR],
3103 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3104 
3105 	if (try_module_get(THIS_MODULE))
3106 		return xprt;
3107 	ret = ERR_PTR(-EINVAL);
3108 out_err:
3109 	xs_xprt_free(xprt);
3110 	return ret;
3111 }
3112 
3113 /**
3114  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3115  * @args: rpc transport creation arguments
3116  *
3117  */
3118 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3119 {
3120 	struct sockaddr *addr = args->dstaddr;
3121 	struct rpc_xprt *xprt;
3122 	struct sock_xprt *transport;
3123 	struct svc_sock *bc_sock;
3124 	struct rpc_xprt *ret;
3125 
3126 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3127 			xprt_tcp_slot_table_entries);
3128 	if (IS_ERR(xprt))
3129 		return xprt;
3130 	transport = container_of(xprt, struct sock_xprt, xprt);
3131 
3132 	xprt->prot = IPPROTO_TCP;
3133 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3134 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3135 	xprt->timeout = &xs_tcp_default_timeout;
3136 
3137 	/* backchannel */
3138 	xprt_set_bound(xprt);
3139 	xprt->bind_timeout = 0;
3140 	xprt->reestablish_timeout = 0;
3141 	xprt->idle_timeout = 0;
3142 
3143 	xprt->ops = &bc_tcp_ops;
3144 
3145 	switch (addr->sa_family) {
3146 	case AF_INET:
3147 		xs_format_peer_addresses(xprt, "tcp",
3148 					 RPCBIND_NETID_TCP);
3149 		break;
3150 	case AF_INET6:
3151 		xs_format_peer_addresses(xprt, "tcp",
3152 				   RPCBIND_NETID_TCP6);
3153 		break;
3154 	default:
3155 		ret = ERR_PTR(-EAFNOSUPPORT);
3156 		goto out_err;
3157 	}
3158 
3159 	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3160 			xprt->address_strings[RPC_DISPLAY_ADDR],
3161 			xprt->address_strings[RPC_DISPLAY_PORT],
3162 			xprt->address_strings[RPC_DISPLAY_PROTO]);
3163 
3164 	/*
3165 	 * Once we've associated a backchannel xprt with a connection,
3166 	 * we want to keep it around as long as the connection lasts,
3167 	 * in case we need to start using it for a backchannel again;
3168 	 * this reference won't be dropped until bc_xprt is destroyed.
3169 	 */
3170 	xprt_get(xprt);
3171 	args->bc_xprt->xpt_bc_xprt = xprt;
3172 	xprt->bc_xprt = args->bc_xprt;
3173 	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3174 	transport->sock = bc_sock->sk_sock;
3175 	transport->inet = bc_sock->sk_sk;
3176 
3177 	/*
3178 	 * Since we don't want connections for the backchannel, we set
3179 	 * the xprt status to connected
3180 	 */
3181 	xprt_set_connected(xprt);
3182 
3183 	if (try_module_get(THIS_MODULE))
3184 		return xprt;
3185 
3186 	args->bc_xprt->xpt_bc_xprt = NULL;
3187 	args->bc_xprt->xpt_bc_xps = NULL;
3188 	xprt_put(xprt);
3189 	ret = ERR_PTR(-EINVAL);
3190 out_err:
3191 	xs_xprt_free(xprt);
3192 	return ret;
3193 }
3194 
3195 static struct xprt_class	xs_local_transport = {
3196 	.list		= LIST_HEAD_INIT(xs_local_transport.list),
3197 	.name		= "named UNIX socket",
3198 	.owner		= THIS_MODULE,
3199 	.ident		= XPRT_TRANSPORT_LOCAL,
3200 	.setup		= xs_setup_local,
3201 };
3202 
3203 static struct xprt_class	xs_udp_transport = {
3204 	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
3205 	.name		= "udp",
3206 	.owner		= THIS_MODULE,
3207 	.ident		= XPRT_TRANSPORT_UDP,
3208 	.setup		= xs_setup_udp,
3209 };
3210 
3211 static struct xprt_class	xs_tcp_transport = {
3212 	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
3213 	.name		= "tcp",
3214 	.owner		= THIS_MODULE,
3215 	.ident		= XPRT_TRANSPORT_TCP,
3216 	.setup		= xs_setup_tcp,
3217 };
3218 
3219 static struct xprt_class	xs_bc_tcp_transport = {
3220 	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3221 	.name		= "tcp NFSv4.1 backchannel",
3222 	.owner		= THIS_MODULE,
3223 	.ident		= XPRT_TRANSPORT_BC_TCP,
3224 	.setup		= xs_setup_bc_tcp,
3225 };
3226 
3227 /**
3228  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3229  *
3230  */
3231 int init_socket_xprt(void)
3232 {
3233 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3234 	if (!sunrpc_table_header)
3235 		sunrpc_table_header = register_sysctl_table(sunrpc_table);
3236 #endif
3237 
3238 	xprt_register_transport(&xs_local_transport);
3239 	xprt_register_transport(&xs_udp_transport);
3240 	xprt_register_transport(&xs_tcp_transport);
3241 	xprt_register_transport(&xs_bc_tcp_transport);
3242 
3243 	return 0;
3244 }
3245 
3246 /**
3247  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3248  *
3249  */
3250 void cleanup_socket_xprt(void)
3251 {
3252 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3253 	if (sunrpc_table_header) {
3254 		unregister_sysctl_table(sunrpc_table_header);
3255 		sunrpc_table_header = NULL;
3256 	}
3257 #endif
3258 
3259 	xprt_unregister_transport(&xs_local_transport);
3260 	xprt_unregister_transport(&xs_udp_transport);
3261 	xprt_unregister_transport(&xs_tcp_transport);
3262 	xprt_unregister_transport(&xs_bc_tcp_transport);
3263 }
3264 
3265 static int param_set_uint_minmax(const char *val,
3266 		const struct kernel_param *kp,
3267 		unsigned int min, unsigned int max)
3268 {
3269 	unsigned int num;
3270 	int ret;
3271 
3272 	if (!val)
3273 		return -EINVAL;
3274 	ret = kstrtouint(val, 0, &num);
3275 	if (ret)
3276 		return ret;
3277 	if (num < min || num > max)
3278 		return -EINVAL;
3279 	*((unsigned int *)kp->arg) = num;
3280 	return 0;
3281 }
3282 
3283 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3284 {
3285 	if (kp->arg == &xprt_min_resvport)
3286 		return param_set_uint_minmax(val, kp,
3287 			RPC_MIN_RESVPORT,
3288 			xprt_max_resvport);
3289 	return param_set_uint_minmax(val, kp,
3290 			xprt_min_resvport,
3291 			RPC_MAX_RESVPORT);
3292 }
3293 
3294 static const struct kernel_param_ops param_ops_portnr = {
3295 	.set = param_set_portnr,
3296 	.get = param_get_uint,
3297 };
3298 
3299 #define param_check_portnr(name, p) \
3300 	__param_check(name, p, unsigned int);
3301 
3302 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3303 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3304 
3305 static int param_set_slot_table_size(const char *val,
3306 				     const struct kernel_param *kp)
3307 {
3308 	return param_set_uint_minmax(val, kp,
3309 			RPC_MIN_SLOT_TABLE,
3310 			RPC_MAX_SLOT_TABLE);
3311 }
3312 
3313 static const struct kernel_param_ops param_ops_slot_table_size = {
3314 	.set = param_set_slot_table_size,
3315 	.get = param_get_uint,
3316 };
3317 
3318 #define param_check_slot_table_size(name, p) \
3319 	__param_check(name, p, unsigned int);
3320 
3321 static int param_set_max_slot_table_size(const char *val,
3322 				     const struct kernel_param *kp)
3323 {
3324 	return param_set_uint_minmax(val, kp,
3325 			RPC_MIN_SLOT_TABLE,
3326 			RPC_MAX_SLOT_TABLE_LIMIT);
3327 }
3328 
3329 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3330 	.set = param_set_max_slot_table_size,
3331 	.get = param_get_uint,
3332 };
3333 
3334 #define param_check_max_slot_table_size(name, p) \
3335 	__param_check(name, p, unsigned int);
3336 
3337 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3338 		   slot_table_size, 0644);
3339 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3340 		   max_slot_table_size, 0644);
3341 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3342 		   slot_table_size, 0644);
3343 
3344