xref: /openbmc/linux/net/sunrpc/xprtsock.c (revision 82e6fdd6)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/net/sunrpc/xprtsock.c
4  *
5  * Client-side transport implementation for sockets.
6  *
7  * TCP callback races fixes (C) 1998 Red Hat
8  * TCP send fixes (C) 1998 Red Hat
9  * TCP NFS related read + write fixes
10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11  *
12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
13  * Fix behaviour when socket buffer is full.
14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15  *
16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17  *
18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19  *   <gilles.quillard@bull.net>
20  */
21 
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
30 #include <linux/in.h>
31 #include <linux/net.h>
32 #include <linux/mm.h>
33 #include <linux/un.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
44 #endif
45 
46 #include <net/sock.h>
47 #include <net/checksum.h>
48 #include <net/udp.h>
49 #include <net/tcp.h>
50 
51 #include <trace/events/sunrpc.h>
52 
53 #include "sunrpc.h"
54 
55 #define RPC_TCP_READ_CHUNK_SZ	(3*512*1024)
56 
57 static void xs_close(struct rpc_xprt *xprt);
58 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
59 		struct socket *sock);
60 
61 /*
62  * xprtsock tunables
63  */
64 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
65 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
66 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
67 
68 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
69 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
70 
71 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
72 
73 #define XS_TCP_LINGER_TO	(15U * HZ)
74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
75 
76 /*
77  * We can register our own files under /proc/sys/sunrpc by
78  * calling register_sysctl_table() again.  The files in that
79  * directory become the union of all files registered there.
80  *
81  * We simply need to make sure that we don't collide with
82  * someone else's file names!
83  */
84 
85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
90 
91 static struct ctl_table_header *sunrpc_table_header;
92 
93 /*
94  * FIXME: changing the UDP slot table size should also resize the UDP
95  *        socket buffers for existing UDP transports
96  */
97 static struct ctl_table xs_tunables_table[] = {
98 	{
99 		.procname	= "udp_slot_table_entries",
100 		.data		= &xprt_udp_slot_table_entries,
101 		.maxlen		= sizeof(unsigned int),
102 		.mode		= 0644,
103 		.proc_handler	= proc_dointvec_minmax,
104 		.extra1		= &min_slot_table_size,
105 		.extra2		= &max_slot_table_size
106 	},
107 	{
108 		.procname	= "tcp_slot_table_entries",
109 		.data		= &xprt_tcp_slot_table_entries,
110 		.maxlen		= sizeof(unsigned int),
111 		.mode		= 0644,
112 		.proc_handler	= proc_dointvec_minmax,
113 		.extra1		= &min_slot_table_size,
114 		.extra2		= &max_slot_table_size
115 	},
116 	{
117 		.procname	= "tcp_max_slot_table_entries",
118 		.data		= &xprt_max_tcp_slot_table_entries,
119 		.maxlen		= sizeof(unsigned int),
120 		.mode		= 0644,
121 		.proc_handler	= proc_dointvec_minmax,
122 		.extra1		= &min_slot_table_size,
123 		.extra2		= &max_tcp_slot_table_limit
124 	},
125 	{
126 		.procname	= "min_resvport",
127 		.data		= &xprt_min_resvport,
128 		.maxlen		= sizeof(unsigned int),
129 		.mode		= 0644,
130 		.proc_handler	= proc_dointvec_minmax,
131 		.extra1		= &xprt_min_resvport_limit,
132 		.extra2		= &xprt_max_resvport
133 	},
134 	{
135 		.procname	= "max_resvport",
136 		.data		= &xprt_max_resvport,
137 		.maxlen		= sizeof(unsigned int),
138 		.mode		= 0644,
139 		.proc_handler	= proc_dointvec_minmax,
140 		.extra1		= &xprt_min_resvport,
141 		.extra2		= &xprt_max_resvport_limit
142 	},
143 	{
144 		.procname	= "tcp_fin_timeout",
145 		.data		= &xs_tcp_fin_timeout,
146 		.maxlen		= sizeof(xs_tcp_fin_timeout),
147 		.mode		= 0644,
148 		.proc_handler	= proc_dointvec_jiffies,
149 	},
150 	{ },
151 };
152 
153 static struct ctl_table sunrpc_table[] = {
154 	{
155 		.procname	= "sunrpc",
156 		.mode		= 0555,
157 		.child		= xs_tunables_table
158 	},
159 	{ },
160 };
161 
162 #endif
163 
164 /*
165  * Wait duration for a reply from the RPC portmapper.
166  */
167 #define XS_BIND_TO		(60U * HZ)
168 
169 /*
170  * Delay if a UDP socket connect error occurs.  This is most likely some
171  * kind of resource problem on the local host.
172  */
173 #define XS_UDP_REEST_TO		(2U * HZ)
174 
175 /*
176  * The reestablish timeout allows clients to delay for a bit before attempting
177  * to reconnect to a server that just dropped our connection.
178  *
179  * We implement an exponential backoff when trying to reestablish a TCP
180  * transport connection with the server.  Some servers like to drop a TCP
181  * connection when they are overworked, so we start with a short timeout and
182  * increase over time if the server is down or not responding.
183  */
184 #define XS_TCP_INIT_REEST_TO	(3U * HZ)
185 
186 /*
187  * TCP idle timeout; client drops the transport socket if it is idle
188  * for this long.  Note that we also timeout UDP sockets to prevent
189  * holding port numbers when there is no RPC traffic.
190  */
191 #define XS_IDLE_DISC_TO		(5U * 60 * HZ)
192 
193 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
194 # undef  RPC_DEBUG_DATA
195 # define RPCDBG_FACILITY	RPCDBG_TRANS
196 #endif
197 
198 #ifdef RPC_DEBUG_DATA
199 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
200 {
201 	u8 *buf = (u8 *) packet;
202 	int j;
203 
204 	dprintk("RPC:       %s\n", msg);
205 	for (j = 0; j < count && j < 128; j += 4) {
206 		if (!(j & 31)) {
207 			if (j)
208 				dprintk("\n");
209 			dprintk("0x%04x ", j);
210 		}
211 		dprintk("%02x%02x%02x%02x ",
212 			buf[j], buf[j+1], buf[j+2], buf[j+3]);
213 	}
214 	dprintk("\n");
215 }
216 #else
217 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
218 {
219 	/* NOP */
220 }
221 #endif
222 
223 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
224 {
225 	return (struct rpc_xprt *) sk->sk_user_data;
226 }
227 
228 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
229 {
230 	return (struct sockaddr *) &xprt->addr;
231 }
232 
233 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
234 {
235 	return (struct sockaddr_un *) &xprt->addr;
236 }
237 
238 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
239 {
240 	return (struct sockaddr_in *) &xprt->addr;
241 }
242 
243 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
244 {
245 	return (struct sockaddr_in6 *) &xprt->addr;
246 }
247 
248 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
249 {
250 	struct sockaddr *sap = xs_addr(xprt);
251 	struct sockaddr_in6 *sin6;
252 	struct sockaddr_in *sin;
253 	struct sockaddr_un *sun;
254 	char buf[128];
255 
256 	switch (sap->sa_family) {
257 	case AF_LOCAL:
258 		sun = xs_addr_un(xprt);
259 		strlcpy(buf, sun->sun_path, sizeof(buf));
260 		xprt->address_strings[RPC_DISPLAY_ADDR] =
261 						kstrdup(buf, GFP_KERNEL);
262 		break;
263 	case AF_INET:
264 		(void)rpc_ntop(sap, buf, sizeof(buf));
265 		xprt->address_strings[RPC_DISPLAY_ADDR] =
266 						kstrdup(buf, GFP_KERNEL);
267 		sin = xs_addr_in(xprt);
268 		snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
269 		break;
270 	case AF_INET6:
271 		(void)rpc_ntop(sap, buf, sizeof(buf));
272 		xprt->address_strings[RPC_DISPLAY_ADDR] =
273 						kstrdup(buf, GFP_KERNEL);
274 		sin6 = xs_addr_in6(xprt);
275 		snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
276 		break;
277 	default:
278 		BUG();
279 	}
280 
281 	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
282 }
283 
284 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
285 {
286 	struct sockaddr *sap = xs_addr(xprt);
287 	char buf[128];
288 
289 	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
290 	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
291 
292 	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
293 	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
294 }
295 
296 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
297 				     const char *protocol,
298 				     const char *netid)
299 {
300 	xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
301 	xprt->address_strings[RPC_DISPLAY_NETID] = netid;
302 	xs_format_common_peer_addresses(xprt);
303 	xs_format_common_peer_ports(xprt);
304 }
305 
306 static void xs_update_peer_port(struct rpc_xprt *xprt)
307 {
308 	kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
309 	kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
310 
311 	xs_format_common_peer_ports(xprt);
312 }
313 
314 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
315 {
316 	unsigned int i;
317 
318 	for (i = 0; i < RPC_DISPLAY_MAX; i++)
319 		switch (i) {
320 		case RPC_DISPLAY_PROTO:
321 		case RPC_DISPLAY_NETID:
322 			continue;
323 		default:
324 			kfree(xprt->address_strings[i]);
325 		}
326 }
327 
328 #define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
329 
330 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
331 {
332 	struct msghdr msg = {
333 		.msg_name	= addr,
334 		.msg_namelen	= addrlen,
335 		.msg_flags	= XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
336 	};
337 	struct kvec iov = {
338 		.iov_base	= vec->iov_base + base,
339 		.iov_len	= vec->iov_len - base,
340 	};
341 
342 	if (iov.iov_len != 0)
343 		return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
344 	return kernel_sendmsg(sock, &msg, NULL, 0, 0);
345 }
346 
347 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more, bool zerocopy, int *sent_p)
348 {
349 	ssize_t (*do_sendpage)(struct socket *sock, struct page *page,
350 			int offset, size_t size, int flags);
351 	struct page **ppage;
352 	unsigned int remainder;
353 	int err;
354 
355 	remainder = xdr->page_len - base;
356 	base += xdr->page_base;
357 	ppage = xdr->pages + (base >> PAGE_SHIFT);
358 	base &= ~PAGE_MASK;
359 	do_sendpage = sock->ops->sendpage;
360 	if (!zerocopy)
361 		do_sendpage = sock_no_sendpage;
362 	for(;;) {
363 		unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
364 		int flags = XS_SENDMSG_FLAGS;
365 
366 		remainder -= len;
367 		if (more)
368 			flags |= MSG_MORE;
369 		if (remainder != 0)
370 			flags |= MSG_SENDPAGE_NOTLAST | MSG_MORE;
371 		err = do_sendpage(sock, *ppage, base, len, flags);
372 		if (remainder == 0 || err != len)
373 			break;
374 		*sent_p += err;
375 		ppage++;
376 		base = 0;
377 	}
378 	if (err > 0) {
379 		*sent_p += err;
380 		err = 0;
381 	}
382 	return err;
383 }
384 
385 /**
386  * xs_sendpages - write pages directly to a socket
387  * @sock: socket to send on
388  * @addr: UDP only -- address of destination
389  * @addrlen: UDP only -- length of destination address
390  * @xdr: buffer containing this request
391  * @base: starting position in the buffer
392  * @zerocopy: true if it is safe to use sendpage()
393  * @sent_p: return the total number of bytes successfully queued for sending
394  *
395  */
396 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, bool zerocopy, int *sent_p)
397 {
398 	unsigned int remainder = xdr->len - base;
399 	int err = 0;
400 	int sent = 0;
401 
402 	if (unlikely(!sock))
403 		return -ENOTSOCK;
404 
405 	if (base != 0) {
406 		addr = NULL;
407 		addrlen = 0;
408 	}
409 
410 	if (base < xdr->head[0].iov_len || addr != NULL) {
411 		unsigned int len = xdr->head[0].iov_len - base;
412 		remainder -= len;
413 		err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
414 		if (remainder == 0 || err != len)
415 			goto out;
416 		*sent_p += err;
417 		base = 0;
418 	} else
419 		base -= xdr->head[0].iov_len;
420 
421 	if (base < xdr->page_len) {
422 		unsigned int len = xdr->page_len - base;
423 		remainder -= len;
424 		err = xs_send_pagedata(sock, xdr, base, remainder != 0, zerocopy, &sent);
425 		*sent_p += sent;
426 		if (remainder == 0 || sent != len)
427 			goto out;
428 		base = 0;
429 	} else
430 		base -= xdr->page_len;
431 
432 	if (base >= xdr->tail[0].iov_len)
433 		return 0;
434 	err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
435 out:
436 	if (err > 0) {
437 		*sent_p += err;
438 		err = 0;
439 	}
440 	return err;
441 }
442 
443 static void xs_nospace_callback(struct rpc_task *task)
444 {
445 	struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
446 
447 	transport->inet->sk_write_pending--;
448 }
449 
450 /**
451  * xs_nospace - place task on wait queue if transmit was incomplete
452  * @task: task to put to sleep
453  *
454  */
455 static int xs_nospace(struct rpc_task *task)
456 {
457 	struct rpc_rqst *req = task->tk_rqstp;
458 	struct rpc_xprt *xprt = req->rq_xprt;
459 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
460 	struct sock *sk = transport->inet;
461 	int ret = -EAGAIN;
462 
463 	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
464 			task->tk_pid, req->rq_slen - req->rq_bytes_sent,
465 			req->rq_slen);
466 
467 	/* Protect against races with write_space */
468 	spin_lock_bh(&xprt->transport_lock);
469 
470 	/* Don't race with disconnect */
471 	if (xprt_connected(xprt)) {
472 		/* wait for more buffer space */
473 		sk->sk_write_pending++;
474 		xprt_wait_for_buffer_space(task, xs_nospace_callback);
475 	} else
476 		ret = -ENOTCONN;
477 
478 	spin_unlock_bh(&xprt->transport_lock);
479 
480 	/* Race breaker in case memory is freed before above code is called */
481 	if (ret == -EAGAIN) {
482 		struct socket_wq *wq;
483 
484 		rcu_read_lock();
485 		wq = rcu_dereference(sk->sk_wq);
486 		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
487 		rcu_read_unlock();
488 
489 		sk->sk_write_space(sk);
490 	}
491 	return ret;
492 }
493 
494 /*
495  * Construct a stream transport record marker in @buf.
496  */
497 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
498 {
499 	u32 reclen = buf->len - sizeof(rpc_fraghdr);
500 	rpc_fraghdr *base = buf->head[0].iov_base;
501 	*base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
502 }
503 
504 /**
505  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
506  * @task: RPC task that manages the state of an RPC request
507  *
508  * Return values:
509  *        0:	The request has been sent
510  *   EAGAIN:	The socket was blocked, please call again later to
511  *		complete the request
512  * ENOTCONN:	Caller needs to invoke connect logic then call again
513  *    other:	Some other error occured, the request was not sent
514  */
515 static int xs_local_send_request(struct rpc_task *task)
516 {
517 	struct rpc_rqst *req = task->tk_rqstp;
518 	struct rpc_xprt *xprt = req->rq_xprt;
519 	struct sock_xprt *transport =
520 				container_of(xprt, struct sock_xprt, xprt);
521 	struct xdr_buf *xdr = &req->rq_snd_buf;
522 	int status;
523 	int sent = 0;
524 
525 	xs_encode_stream_record_marker(&req->rq_snd_buf);
526 
527 	xs_pktdump("packet data:",
528 			req->rq_svec->iov_base, req->rq_svec->iov_len);
529 
530 	status = xs_sendpages(transport->sock, NULL, 0, xdr, req->rq_bytes_sent,
531 			      true, &sent);
532 	dprintk("RPC:       %s(%u) = %d\n",
533 			__func__, xdr->len - req->rq_bytes_sent, status);
534 
535 	if (status == -EAGAIN && sock_writeable(transport->inet))
536 		status = -ENOBUFS;
537 
538 	if (likely(sent > 0) || status == 0) {
539 		req->rq_bytes_sent += sent;
540 		req->rq_xmit_bytes_sent += sent;
541 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
542 			req->rq_bytes_sent = 0;
543 			return 0;
544 		}
545 		status = -EAGAIN;
546 	}
547 
548 	switch (status) {
549 	case -ENOBUFS:
550 		break;
551 	case -EAGAIN:
552 		status = xs_nospace(task);
553 		break;
554 	default:
555 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
556 			-status);
557 		/* fall through */
558 	case -EPIPE:
559 		xs_close(xprt);
560 		status = -ENOTCONN;
561 	}
562 
563 	return status;
564 }
565 
566 /**
567  * xs_udp_send_request - write an RPC request to a UDP socket
568  * @task: address of RPC task that manages the state of an RPC request
569  *
570  * Return values:
571  *        0:	The request has been sent
572  *   EAGAIN:	The socket was blocked, please call again later to
573  *		complete the request
574  * ENOTCONN:	Caller needs to invoke connect logic then call again
575  *    other:	Some other error occurred, the request was not sent
576  */
577 static int xs_udp_send_request(struct rpc_task *task)
578 {
579 	struct rpc_rqst *req = task->tk_rqstp;
580 	struct rpc_xprt *xprt = req->rq_xprt;
581 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
582 	struct xdr_buf *xdr = &req->rq_snd_buf;
583 	int sent = 0;
584 	int status;
585 
586 	xs_pktdump("packet data:",
587 				req->rq_svec->iov_base,
588 				req->rq_svec->iov_len);
589 
590 	if (!xprt_bound(xprt))
591 		return -ENOTCONN;
592 	status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
593 			      xdr, req->rq_bytes_sent, true, &sent);
594 
595 	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
596 			xdr->len - req->rq_bytes_sent, status);
597 
598 	/* firewall is blocking us, don't return -EAGAIN or we end up looping */
599 	if (status == -EPERM)
600 		goto process_status;
601 
602 	if (status == -EAGAIN && sock_writeable(transport->inet))
603 		status = -ENOBUFS;
604 
605 	if (sent > 0 || status == 0) {
606 		req->rq_xmit_bytes_sent += sent;
607 		if (sent >= req->rq_slen)
608 			return 0;
609 		/* Still some bytes left; set up for a retry later. */
610 		status = -EAGAIN;
611 	}
612 
613 process_status:
614 	switch (status) {
615 	case -ENOTSOCK:
616 		status = -ENOTCONN;
617 		/* Should we call xs_close() here? */
618 		break;
619 	case -EAGAIN:
620 		status = xs_nospace(task);
621 		break;
622 	case -ENETUNREACH:
623 	case -ENOBUFS:
624 	case -EPIPE:
625 	case -ECONNREFUSED:
626 	case -EPERM:
627 		/* When the server has died, an ICMP port unreachable message
628 		 * prompts ECONNREFUSED. */
629 		break;
630 	default:
631 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
632 			-status);
633 	}
634 
635 	return status;
636 }
637 
638 /**
639  * xs_tcp_send_request - write an RPC request to a TCP socket
640  * @task: address of RPC task that manages the state of an RPC request
641  *
642  * Return values:
643  *        0:	The request has been sent
644  *   EAGAIN:	The socket was blocked, please call again later to
645  *		complete the request
646  * ENOTCONN:	Caller needs to invoke connect logic then call again
647  *    other:	Some other error occurred, the request was not sent
648  *
649  * XXX: In the case of soft timeouts, should we eventually give up
650  *	if sendmsg is not able to make progress?
651  */
652 static int xs_tcp_send_request(struct rpc_task *task)
653 {
654 	struct rpc_rqst *req = task->tk_rqstp;
655 	struct rpc_xprt *xprt = req->rq_xprt;
656 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
657 	struct xdr_buf *xdr = &req->rq_snd_buf;
658 	bool zerocopy = true;
659 	bool vm_wait = false;
660 	int status;
661 	int sent;
662 
663 	xs_encode_stream_record_marker(&req->rq_snd_buf);
664 
665 	xs_pktdump("packet data:",
666 				req->rq_svec->iov_base,
667 				req->rq_svec->iov_len);
668 	/* Don't use zero copy if this is a resend. If the RPC call
669 	 * completes while the socket holds a reference to the pages,
670 	 * then we may end up resending corrupted data.
671 	 */
672 	if (task->tk_flags & RPC_TASK_SENT)
673 		zerocopy = false;
674 
675 	if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
676 		xs_tcp_set_socket_timeouts(xprt, transport->sock);
677 
678 	/* Continue transmitting the packet/record. We must be careful
679 	 * to cope with writespace callbacks arriving _after_ we have
680 	 * called sendmsg(). */
681 	while (1) {
682 		sent = 0;
683 		status = xs_sendpages(transport->sock, NULL, 0, xdr,
684 				      req->rq_bytes_sent, zerocopy, &sent);
685 
686 		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
687 				xdr->len - req->rq_bytes_sent, status);
688 
689 		/* If we've sent the entire packet, immediately
690 		 * reset the count of bytes sent. */
691 		req->rq_bytes_sent += sent;
692 		req->rq_xmit_bytes_sent += sent;
693 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
694 			req->rq_bytes_sent = 0;
695 			return 0;
696 		}
697 
698 		WARN_ON_ONCE(sent == 0 && status == 0);
699 
700 		if (status == -EAGAIN ) {
701 			/*
702 			 * Return EAGAIN if we're sure we're hitting the
703 			 * socket send buffer limits.
704 			 */
705 			if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
706 				break;
707 			/*
708 			 * Did we hit a memory allocation failure?
709 			 */
710 			if (sent == 0) {
711 				status = -ENOBUFS;
712 				if (vm_wait)
713 					break;
714 				/* Retry, knowing now that we're below the
715 				 * socket send buffer limit
716 				 */
717 				vm_wait = true;
718 			}
719 			continue;
720 		}
721 		if (status < 0)
722 			break;
723 		vm_wait = false;
724 	}
725 
726 	switch (status) {
727 	case -ENOTSOCK:
728 		status = -ENOTCONN;
729 		/* Should we call xs_close() here? */
730 		break;
731 	case -EAGAIN:
732 		status = xs_nospace(task);
733 		break;
734 	case -ECONNRESET:
735 	case -ECONNREFUSED:
736 	case -ENOTCONN:
737 	case -EADDRINUSE:
738 	case -ENOBUFS:
739 	case -EPIPE:
740 		break;
741 	default:
742 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
743 			-status);
744 	}
745 
746 	return status;
747 }
748 
749 /**
750  * xs_tcp_release_xprt - clean up after a tcp transmission
751  * @xprt: transport
752  * @task: rpc task
753  *
754  * This cleans up if an error causes us to abort the transmission of a request.
755  * In this case, the socket may need to be reset in order to avoid confusing
756  * the server.
757  */
758 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
759 {
760 	struct rpc_rqst *req;
761 
762 	if (task != xprt->snd_task)
763 		return;
764 	if (task == NULL)
765 		goto out_release;
766 	req = task->tk_rqstp;
767 	if (req == NULL)
768 		goto out_release;
769 	if (req->rq_bytes_sent == 0)
770 		goto out_release;
771 	if (req->rq_bytes_sent == req->rq_snd_buf.len)
772 		goto out_release;
773 	set_bit(XPRT_CLOSE_WAIT, &xprt->state);
774 out_release:
775 	xprt_release_xprt(xprt, task);
776 }
777 
778 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
779 {
780 	transport->old_data_ready = sk->sk_data_ready;
781 	transport->old_state_change = sk->sk_state_change;
782 	transport->old_write_space = sk->sk_write_space;
783 	transport->old_error_report = sk->sk_error_report;
784 }
785 
786 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
787 {
788 	sk->sk_data_ready = transport->old_data_ready;
789 	sk->sk_state_change = transport->old_state_change;
790 	sk->sk_write_space = transport->old_write_space;
791 	sk->sk_error_report = transport->old_error_report;
792 }
793 
794 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
795 {
796 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
797 
798 	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
799 }
800 
801 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
802 {
803 	smp_mb__before_atomic();
804 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
805 	clear_bit(XPRT_CLOSING, &xprt->state);
806 	xs_sock_reset_state_flags(xprt);
807 	smp_mb__after_atomic();
808 }
809 
810 /**
811  * xs_error_report - callback to handle TCP socket state errors
812  * @sk: socket
813  *
814  * Note: we don't call sock_error() since there may be a rpc_task
815  * using the socket, and so we don't want to clear sk->sk_err.
816  */
817 static void xs_error_report(struct sock *sk)
818 {
819 	struct rpc_xprt *xprt;
820 	int err;
821 
822 	read_lock_bh(&sk->sk_callback_lock);
823 	if (!(xprt = xprt_from_sock(sk)))
824 		goto out;
825 
826 	err = -sk->sk_err;
827 	if (err == 0)
828 		goto out;
829 	dprintk("RPC:       xs_error_report client %p, error=%d...\n",
830 			xprt, -err);
831 	trace_rpc_socket_error(xprt, sk->sk_socket, err);
832 	xprt_wake_pending_tasks(xprt, err);
833  out:
834 	read_unlock_bh(&sk->sk_callback_lock);
835 }
836 
837 static void xs_reset_transport(struct sock_xprt *transport)
838 {
839 	struct socket *sock = transport->sock;
840 	struct sock *sk = transport->inet;
841 	struct rpc_xprt *xprt = &transport->xprt;
842 
843 	if (sk == NULL)
844 		return;
845 
846 	if (atomic_read(&transport->xprt.swapper))
847 		sk_clear_memalloc(sk);
848 
849 	kernel_sock_shutdown(sock, SHUT_RDWR);
850 
851 	mutex_lock(&transport->recv_mutex);
852 	write_lock_bh(&sk->sk_callback_lock);
853 	transport->inet = NULL;
854 	transport->sock = NULL;
855 
856 	sk->sk_user_data = NULL;
857 
858 	xs_restore_old_callbacks(transport, sk);
859 	xprt_clear_connected(xprt);
860 	write_unlock_bh(&sk->sk_callback_lock);
861 	xs_sock_reset_connection_flags(xprt);
862 	mutex_unlock(&transport->recv_mutex);
863 
864 	trace_rpc_socket_close(xprt, sock);
865 	sock_release(sock);
866 }
867 
868 /**
869  * xs_close - close a socket
870  * @xprt: transport
871  *
872  * This is used when all requests are complete; ie, no DRC state remains
873  * on the server we want to save.
874  *
875  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
876  * xs_reset_transport() zeroing the socket from underneath a writer.
877  */
878 static void xs_close(struct rpc_xprt *xprt)
879 {
880 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
881 
882 	dprintk("RPC:       xs_close xprt %p\n", xprt);
883 
884 	xs_reset_transport(transport);
885 	xprt->reestablish_timeout = 0;
886 
887 	xprt_disconnect_done(xprt);
888 }
889 
890 static void xs_inject_disconnect(struct rpc_xprt *xprt)
891 {
892 	dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
893 		xprt);
894 	xprt_disconnect_done(xprt);
895 }
896 
897 static void xs_xprt_free(struct rpc_xprt *xprt)
898 {
899 	xs_free_peer_addresses(xprt);
900 	xprt_free(xprt);
901 }
902 
903 /**
904  * xs_destroy - prepare to shutdown a transport
905  * @xprt: doomed transport
906  *
907  */
908 static void xs_destroy(struct rpc_xprt *xprt)
909 {
910 	struct sock_xprt *transport = container_of(xprt,
911 			struct sock_xprt, xprt);
912 	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
913 
914 	cancel_delayed_work_sync(&transport->connect_worker);
915 	xs_close(xprt);
916 	cancel_work_sync(&transport->recv_worker);
917 	xs_xprt_free(xprt);
918 	module_put(THIS_MODULE);
919 }
920 
921 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
922 {
923 	struct xdr_skb_reader desc = {
924 		.skb		= skb,
925 		.offset		= sizeof(rpc_fraghdr),
926 		.count		= skb->len - sizeof(rpc_fraghdr),
927 	};
928 
929 	if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
930 		return -1;
931 	if (desc.count)
932 		return -1;
933 	return 0;
934 }
935 
936 /**
937  * xs_local_data_read_skb
938  * @xprt: transport
939  * @sk: socket
940  * @skb: skbuff
941  *
942  * Currently this assumes we can read the whole reply in a single gulp.
943  */
944 static void xs_local_data_read_skb(struct rpc_xprt *xprt,
945 		struct sock *sk,
946 		struct sk_buff *skb)
947 {
948 	struct rpc_task *task;
949 	struct rpc_rqst *rovr;
950 	int repsize, copied;
951 	u32 _xid;
952 	__be32 *xp;
953 
954 	repsize = skb->len - sizeof(rpc_fraghdr);
955 	if (repsize < 4) {
956 		dprintk("RPC:       impossible RPC reply size %d\n", repsize);
957 		return;
958 	}
959 
960 	/* Copy the XID from the skb... */
961 	xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
962 	if (xp == NULL)
963 		return;
964 
965 	/* Look up and lock the request corresponding to the given XID */
966 	spin_lock(&xprt->recv_lock);
967 	rovr = xprt_lookup_rqst(xprt, *xp);
968 	if (!rovr)
969 		goto out_unlock;
970 	xprt_pin_rqst(rovr);
971 	spin_unlock(&xprt->recv_lock);
972 	task = rovr->rq_task;
973 
974 	copied = rovr->rq_private_buf.buflen;
975 	if (copied > repsize)
976 		copied = repsize;
977 
978 	if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
979 		dprintk("RPC:       sk_buff copy failed\n");
980 		spin_lock(&xprt->recv_lock);
981 		goto out_unpin;
982 	}
983 
984 	spin_lock(&xprt->recv_lock);
985 	xprt_complete_rqst(task, copied);
986 out_unpin:
987 	xprt_unpin_rqst(rovr);
988  out_unlock:
989 	spin_unlock(&xprt->recv_lock);
990 }
991 
992 static void xs_local_data_receive(struct sock_xprt *transport)
993 {
994 	struct sk_buff *skb;
995 	struct sock *sk;
996 	int err;
997 
998 restart:
999 	mutex_lock(&transport->recv_mutex);
1000 	sk = transport->inet;
1001 	if (sk == NULL)
1002 		goto out;
1003 	for (;;) {
1004 		skb = skb_recv_datagram(sk, 0, 1, &err);
1005 		if (skb != NULL) {
1006 			xs_local_data_read_skb(&transport->xprt, sk, skb);
1007 			skb_free_datagram(sk, skb);
1008 			continue;
1009 		}
1010 		if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1011 			break;
1012 		if (need_resched()) {
1013 			mutex_unlock(&transport->recv_mutex);
1014 			cond_resched();
1015 			goto restart;
1016 		}
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 		spin_lock(&xprt->recv_lock);
1072 		__UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1073 		goto out_unpin;
1074 	}
1075 
1076 
1077 	spin_lock_bh(&xprt->transport_lock);
1078 	xprt_adjust_cwnd(xprt, task, copied);
1079 	spin_unlock_bh(&xprt->transport_lock);
1080 	spin_lock(&xprt->recv_lock);
1081 	xprt_complete_rqst(task, copied);
1082 	__UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
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 restart:
1096 	mutex_lock(&transport->recv_mutex);
1097 	sk = transport->inet;
1098 	if (sk == NULL)
1099 		goto out;
1100 	for (;;) {
1101 		skb = skb_recv_udp(sk, 0, 1, &err);
1102 		if (skb != NULL) {
1103 			xs_udp_data_read_skb(&transport->xprt, sk, skb);
1104 			consume_skb(skb);
1105 			continue;
1106 		}
1107 		if (!test_and_clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1108 			break;
1109 		if (need_resched()) {
1110 			mutex_unlock(&transport->recv_mutex);
1111 			cond_resched();
1112 			goto restart;
1113 		}
1114 	}
1115 out:
1116 	mutex_unlock(&transport->recv_mutex);
1117 }
1118 
1119 static void xs_udp_data_receive_workfn(struct work_struct *work)
1120 {
1121 	struct sock_xprt *transport =
1122 		container_of(work, struct sock_xprt, recv_worker);
1123 	xs_udp_data_receive(transport);
1124 }
1125 
1126 /**
1127  * xs_data_ready - "data ready" callback for UDP sockets
1128  * @sk: socket with data to read
1129  *
1130  */
1131 static void xs_data_ready(struct sock *sk)
1132 {
1133 	struct rpc_xprt *xprt;
1134 
1135 	read_lock_bh(&sk->sk_callback_lock);
1136 	dprintk("RPC:       xs_data_ready...\n");
1137 	xprt = xprt_from_sock(sk);
1138 	if (xprt != NULL) {
1139 		struct sock_xprt *transport = container_of(xprt,
1140 				struct sock_xprt, xprt);
1141 		transport->old_data_ready(sk);
1142 		/* Any data means we had a useful conversation, so
1143 		 * then we don't need to delay the next reconnect
1144 		 */
1145 		if (xprt->reestablish_timeout)
1146 			xprt->reestablish_timeout = 0;
1147 		if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1148 			queue_work(xprtiod_workqueue, &transport->recv_worker);
1149 	}
1150 	read_unlock_bh(&sk->sk_callback_lock);
1151 }
1152 
1153 /*
1154  * Helper function to force a TCP close if the server is sending
1155  * junk and/or it has put us in CLOSE_WAIT
1156  */
1157 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1158 {
1159 	xprt_force_disconnect(xprt);
1160 }
1161 
1162 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1163 {
1164 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1165 	size_t len, used;
1166 	char *p;
1167 
1168 	p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1169 	len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1170 	used = xdr_skb_read_bits(desc, p, len);
1171 	transport->tcp_offset += used;
1172 	if (used != len)
1173 		return;
1174 
1175 	transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1176 	if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1177 		transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1178 	else
1179 		transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1180 	transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1181 
1182 	transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1183 	transport->tcp_offset = 0;
1184 
1185 	/* Sanity check of the record length */
1186 	if (unlikely(transport->tcp_reclen < 8)) {
1187 		dprintk("RPC:       invalid TCP record fragment length\n");
1188 		xs_tcp_force_close(xprt);
1189 		return;
1190 	}
1191 	dprintk("RPC:       reading TCP record fragment of length %d\n",
1192 			transport->tcp_reclen);
1193 }
1194 
1195 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1196 {
1197 	if (transport->tcp_offset == transport->tcp_reclen) {
1198 		transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1199 		transport->tcp_offset = 0;
1200 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1201 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1202 			transport->tcp_flags |= TCP_RCV_COPY_XID;
1203 			transport->tcp_copied = 0;
1204 		}
1205 	}
1206 }
1207 
1208 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1209 {
1210 	size_t len, used;
1211 	char *p;
1212 
1213 	len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1214 	dprintk("RPC:       reading XID (%zu bytes)\n", len);
1215 	p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1216 	used = xdr_skb_read_bits(desc, p, len);
1217 	transport->tcp_offset += used;
1218 	if (used != len)
1219 		return;
1220 	transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1221 	transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1222 	transport->tcp_copied = 4;
1223 	dprintk("RPC:       reading %s XID %08x\n",
1224 			(transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1225 							      : "request with",
1226 			ntohl(transport->tcp_xid));
1227 	xs_tcp_check_fraghdr(transport);
1228 }
1229 
1230 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1231 				       struct xdr_skb_reader *desc)
1232 {
1233 	size_t len, used;
1234 	u32 offset;
1235 	char *p;
1236 
1237 	/*
1238 	 * We want transport->tcp_offset to be 8 at the end of this routine
1239 	 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1240 	 * When this function is called for the first time,
1241 	 * transport->tcp_offset is 4 (after having already read the xid).
1242 	 */
1243 	offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1244 	len = sizeof(transport->tcp_calldir) - offset;
1245 	dprintk("RPC:       reading CALL/REPLY flag (%zu bytes)\n", len);
1246 	p = ((char *) &transport->tcp_calldir) + offset;
1247 	used = xdr_skb_read_bits(desc, p, len);
1248 	transport->tcp_offset += used;
1249 	if (used != len)
1250 		return;
1251 	transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1252 	/*
1253 	 * We don't yet have the XDR buffer, so we will write the calldir
1254 	 * out after we get the buffer from the 'struct rpc_rqst'
1255 	 */
1256 	switch (ntohl(transport->tcp_calldir)) {
1257 	case RPC_REPLY:
1258 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1259 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1260 		transport->tcp_flags |= TCP_RPC_REPLY;
1261 		break;
1262 	case RPC_CALL:
1263 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1264 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1265 		transport->tcp_flags &= ~TCP_RPC_REPLY;
1266 		break;
1267 	default:
1268 		dprintk("RPC:       invalid request message type\n");
1269 		xs_tcp_force_close(&transport->xprt);
1270 	}
1271 	xs_tcp_check_fraghdr(transport);
1272 }
1273 
1274 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1275 				     struct xdr_skb_reader *desc,
1276 				     struct rpc_rqst *req)
1277 {
1278 	struct sock_xprt *transport =
1279 				container_of(xprt, struct sock_xprt, xprt);
1280 	struct xdr_buf *rcvbuf;
1281 	size_t len;
1282 	ssize_t r;
1283 
1284 	rcvbuf = &req->rq_private_buf;
1285 
1286 	if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1287 		/*
1288 		 * Save the RPC direction in the XDR buffer
1289 		 */
1290 		memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1291 			&transport->tcp_calldir,
1292 			sizeof(transport->tcp_calldir));
1293 		transport->tcp_copied += sizeof(transport->tcp_calldir);
1294 		transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1295 	}
1296 
1297 	len = desc->count;
1298 	if (len > transport->tcp_reclen - transport->tcp_offset)
1299 		desc->count = transport->tcp_reclen - transport->tcp_offset;
1300 	r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1301 					  desc, xdr_skb_read_bits);
1302 
1303 	if (desc->count) {
1304 		/* Error when copying to the receive buffer,
1305 		 * usually because we weren't able to allocate
1306 		 * additional buffer pages. All we can do now
1307 		 * is turn off TCP_RCV_COPY_DATA, so the request
1308 		 * will not receive any additional updates,
1309 		 * and time out.
1310 		 * Any remaining data from this record will
1311 		 * be discarded.
1312 		 */
1313 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1314 		dprintk("RPC:       XID %08x truncated request\n",
1315 				ntohl(transport->tcp_xid));
1316 		dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1317 				"tcp_offset = %u, tcp_reclen = %u\n",
1318 				xprt, transport->tcp_copied,
1319 				transport->tcp_offset, transport->tcp_reclen);
1320 		return;
1321 	}
1322 
1323 	transport->tcp_copied += r;
1324 	transport->tcp_offset += r;
1325 	desc->count = len - r;
1326 
1327 	dprintk("RPC:       XID %08x read %zd bytes\n",
1328 			ntohl(transport->tcp_xid), r);
1329 	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1330 			"tcp_reclen = %u\n", xprt, transport->tcp_copied,
1331 			transport->tcp_offset, transport->tcp_reclen);
1332 
1333 	if (transport->tcp_copied == req->rq_private_buf.buflen)
1334 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1335 	else if (transport->tcp_offset == transport->tcp_reclen) {
1336 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1337 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1338 	}
1339 }
1340 
1341 /*
1342  * Finds the request corresponding to the RPC xid and invokes the common
1343  * tcp read code to read the data.
1344  */
1345 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1346 				    struct xdr_skb_reader *desc)
1347 {
1348 	struct sock_xprt *transport =
1349 				container_of(xprt, struct sock_xprt, xprt);
1350 	struct rpc_rqst *req;
1351 
1352 	dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1353 
1354 	/* Find and lock the request corresponding to this xid */
1355 	spin_lock(&xprt->recv_lock);
1356 	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1357 	if (!req) {
1358 		dprintk("RPC:       XID %08x request not found!\n",
1359 				ntohl(transport->tcp_xid));
1360 		spin_unlock(&xprt->recv_lock);
1361 		return -1;
1362 	}
1363 	xprt_pin_rqst(req);
1364 	spin_unlock(&xprt->recv_lock);
1365 
1366 	xs_tcp_read_common(xprt, desc, req);
1367 
1368 	spin_lock(&xprt->recv_lock);
1369 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1370 		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1371 	xprt_unpin_rqst(req);
1372 	spin_unlock(&xprt->recv_lock);
1373 	return 0;
1374 }
1375 
1376 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1377 /*
1378  * Obtains an rpc_rqst previously allocated and invokes the common
1379  * tcp read code to read the data.  The result is placed in the callback
1380  * queue.
1381  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1382  * connection and return -1.
1383  */
1384 static int xs_tcp_read_callback(struct rpc_xprt *xprt,
1385 				       struct xdr_skb_reader *desc)
1386 {
1387 	struct sock_xprt *transport =
1388 				container_of(xprt, struct sock_xprt, xprt);
1389 	struct rpc_rqst *req;
1390 
1391 	/* Look up the request corresponding to the given XID */
1392 	req = xprt_lookup_bc_request(xprt, transport->tcp_xid);
1393 	if (req == NULL) {
1394 		printk(KERN_WARNING "Callback slot table overflowed\n");
1395 		xprt_force_disconnect(xprt);
1396 		return -1;
1397 	}
1398 
1399 	dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1400 	xs_tcp_read_common(xprt, desc, req);
1401 
1402 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1403 		xprt_complete_bc_request(req, transport->tcp_copied);
1404 
1405 	return 0;
1406 }
1407 
1408 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1409 					struct xdr_skb_reader *desc)
1410 {
1411 	struct sock_xprt *transport =
1412 				container_of(xprt, struct sock_xprt, xprt);
1413 
1414 	return (transport->tcp_flags & TCP_RPC_REPLY) ?
1415 		xs_tcp_read_reply(xprt, desc) :
1416 		xs_tcp_read_callback(xprt, desc);
1417 }
1418 
1419 static int xs_tcp_bc_up(struct svc_serv *serv, struct net *net)
1420 {
1421 	int ret;
1422 
1423 	ret = svc_create_xprt(serv, "tcp-bc", net, PF_INET, 0,
1424 			      SVC_SOCK_ANONYMOUS);
1425 	if (ret < 0)
1426 		return ret;
1427 	return 0;
1428 }
1429 
1430 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1431 {
1432 	return PAGE_SIZE;
1433 }
1434 #else
1435 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1436 					struct xdr_skb_reader *desc)
1437 {
1438 	return xs_tcp_read_reply(xprt, desc);
1439 }
1440 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1441 
1442 /*
1443  * Read data off the transport.  This can be either an RPC_CALL or an
1444  * RPC_REPLY.  Relay the processing to helper functions.
1445  */
1446 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1447 				    struct xdr_skb_reader *desc)
1448 {
1449 	struct sock_xprt *transport =
1450 				container_of(xprt, struct sock_xprt, xprt);
1451 
1452 	if (_xs_tcp_read_data(xprt, desc) == 0)
1453 		xs_tcp_check_fraghdr(transport);
1454 	else {
1455 		/*
1456 		 * The transport_lock protects the request handling.
1457 		 * There's no need to hold it to update the tcp_flags.
1458 		 */
1459 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1460 	}
1461 }
1462 
1463 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1464 {
1465 	size_t len;
1466 
1467 	len = transport->tcp_reclen - transport->tcp_offset;
1468 	if (len > desc->count)
1469 		len = desc->count;
1470 	desc->count -= len;
1471 	desc->offset += len;
1472 	transport->tcp_offset += len;
1473 	dprintk("RPC:       discarded %zu bytes\n", len);
1474 	xs_tcp_check_fraghdr(transport);
1475 }
1476 
1477 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1478 {
1479 	struct rpc_xprt *xprt = rd_desc->arg.data;
1480 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1481 	struct xdr_skb_reader desc = {
1482 		.skb	= skb,
1483 		.offset	= offset,
1484 		.count	= len,
1485 	};
1486 	size_t ret;
1487 
1488 	dprintk("RPC:       xs_tcp_data_recv started\n");
1489 	do {
1490 		trace_xs_tcp_data_recv(transport);
1491 		/* Read in a new fragment marker if necessary */
1492 		/* Can we ever really expect to get completely empty fragments? */
1493 		if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1494 			xs_tcp_read_fraghdr(xprt, &desc);
1495 			continue;
1496 		}
1497 		/* Read in the xid if necessary */
1498 		if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1499 			xs_tcp_read_xid(transport, &desc);
1500 			continue;
1501 		}
1502 		/* Read in the call/reply flag */
1503 		if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1504 			xs_tcp_read_calldir(transport, &desc);
1505 			continue;
1506 		}
1507 		/* Read in the request data */
1508 		if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1509 			xs_tcp_read_data(xprt, &desc);
1510 			continue;
1511 		}
1512 		/* Skip over any trailing bytes on short reads */
1513 		xs_tcp_read_discard(transport, &desc);
1514 	} while (desc.count);
1515 	ret = len - desc.count;
1516 	if (ret < rd_desc->count)
1517 		rd_desc->count -= ret;
1518 	else
1519 		rd_desc->count = 0;
1520 	trace_xs_tcp_data_recv(transport);
1521 	dprintk("RPC:       xs_tcp_data_recv done\n");
1522 	return ret;
1523 }
1524 
1525 static void xs_tcp_data_receive(struct sock_xprt *transport)
1526 {
1527 	struct rpc_xprt *xprt = &transport->xprt;
1528 	struct sock *sk;
1529 	read_descriptor_t rd_desc = {
1530 		.arg.data = xprt,
1531 	};
1532 	unsigned long total = 0;
1533 	int read = 0;
1534 
1535 restart:
1536 	mutex_lock(&transport->recv_mutex);
1537 	sk = transport->inet;
1538 	if (sk == NULL)
1539 		goto out;
1540 
1541 	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1542 	for (;;) {
1543 		rd_desc.count = RPC_TCP_READ_CHUNK_SZ;
1544 		lock_sock(sk);
1545 		read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1546 		if (rd_desc.count != 0 || read < 0) {
1547 			clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1548 			release_sock(sk);
1549 			break;
1550 		}
1551 		release_sock(sk);
1552 		total += read;
1553 		if (need_resched()) {
1554 			mutex_unlock(&transport->recv_mutex);
1555 			cond_resched();
1556 			goto restart;
1557 		}
1558 	}
1559 	if (test_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1560 		queue_work(xprtiod_workqueue, &transport->recv_worker);
1561 out:
1562 	mutex_unlock(&transport->recv_mutex);
1563 	trace_xs_tcp_data_ready(xprt, read, total);
1564 }
1565 
1566 static void xs_tcp_data_receive_workfn(struct work_struct *work)
1567 {
1568 	struct sock_xprt *transport =
1569 		container_of(work, struct sock_xprt, recv_worker);
1570 	xs_tcp_data_receive(transport);
1571 }
1572 
1573 /**
1574  * xs_tcp_state_change - callback to handle TCP socket state changes
1575  * @sk: socket whose state has changed
1576  *
1577  */
1578 static void xs_tcp_state_change(struct sock *sk)
1579 {
1580 	struct rpc_xprt *xprt;
1581 	struct sock_xprt *transport;
1582 
1583 	read_lock_bh(&sk->sk_callback_lock);
1584 	if (!(xprt = xprt_from_sock(sk)))
1585 		goto out;
1586 	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1587 	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1588 			sk->sk_state, xprt_connected(xprt),
1589 			sock_flag(sk, SOCK_DEAD),
1590 			sock_flag(sk, SOCK_ZAPPED),
1591 			sk->sk_shutdown);
1592 
1593 	transport = container_of(xprt, struct sock_xprt, xprt);
1594 	trace_rpc_socket_state_change(xprt, sk->sk_socket);
1595 	switch (sk->sk_state) {
1596 	case TCP_ESTABLISHED:
1597 		spin_lock(&xprt->transport_lock);
1598 		if (!xprt_test_and_set_connected(xprt)) {
1599 
1600 			/* Reset TCP record info */
1601 			transport->tcp_offset = 0;
1602 			transport->tcp_reclen = 0;
1603 			transport->tcp_copied = 0;
1604 			transport->tcp_flags =
1605 				TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1606 			xprt->connect_cookie++;
1607 			clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1608 			xprt_clear_connecting(xprt);
1609 
1610 			xprt_wake_pending_tasks(xprt, -EAGAIN);
1611 		}
1612 		spin_unlock(&xprt->transport_lock);
1613 		break;
1614 	case TCP_FIN_WAIT1:
1615 		/* The client initiated a shutdown of the socket */
1616 		xprt->connect_cookie++;
1617 		xprt->reestablish_timeout = 0;
1618 		set_bit(XPRT_CLOSING, &xprt->state);
1619 		smp_mb__before_atomic();
1620 		clear_bit(XPRT_CONNECTED, &xprt->state);
1621 		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1622 		smp_mb__after_atomic();
1623 		break;
1624 	case TCP_CLOSE_WAIT:
1625 		/* The server initiated a shutdown of the socket */
1626 		xprt->connect_cookie++;
1627 		clear_bit(XPRT_CONNECTED, &xprt->state);
1628 		xs_tcp_force_close(xprt);
1629 		/* fall through */
1630 	case TCP_CLOSING:
1631 		/*
1632 		 * If the server closed down the connection, make sure that
1633 		 * we back off before reconnecting
1634 		 */
1635 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1636 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1637 		break;
1638 	case TCP_LAST_ACK:
1639 		set_bit(XPRT_CLOSING, &xprt->state);
1640 		smp_mb__before_atomic();
1641 		clear_bit(XPRT_CONNECTED, &xprt->state);
1642 		smp_mb__after_atomic();
1643 		break;
1644 	case TCP_CLOSE:
1645 		if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1646 					&transport->sock_state))
1647 			xprt_clear_connecting(xprt);
1648 		clear_bit(XPRT_CLOSING, &xprt->state);
1649 		if (sk->sk_err)
1650 			xprt_wake_pending_tasks(xprt, -sk->sk_err);
1651 		/* Trigger the socket release */
1652 		xs_tcp_force_close(xprt);
1653 	}
1654  out:
1655 	read_unlock_bh(&sk->sk_callback_lock);
1656 }
1657 
1658 static void xs_write_space(struct sock *sk)
1659 {
1660 	struct socket_wq *wq;
1661 	struct rpc_xprt *xprt;
1662 
1663 	if (!sk->sk_socket)
1664 		return;
1665 	clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1666 
1667 	if (unlikely(!(xprt = xprt_from_sock(sk))))
1668 		return;
1669 	rcu_read_lock();
1670 	wq = rcu_dereference(sk->sk_wq);
1671 	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1672 		goto out;
1673 
1674 	xprt_write_space(xprt);
1675 out:
1676 	rcu_read_unlock();
1677 }
1678 
1679 /**
1680  * xs_udp_write_space - callback invoked when socket buffer space
1681  *                             becomes available
1682  * @sk: socket whose state has changed
1683  *
1684  * Called when more output buffer space is available for this socket.
1685  * We try not to wake our writers until they can make "significant"
1686  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1687  * with a bunch of small requests.
1688  */
1689 static void xs_udp_write_space(struct sock *sk)
1690 {
1691 	read_lock_bh(&sk->sk_callback_lock);
1692 
1693 	/* from net/core/sock.c:sock_def_write_space */
1694 	if (sock_writeable(sk))
1695 		xs_write_space(sk);
1696 
1697 	read_unlock_bh(&sk->sk_callback_lock);
1698 }
1699 
1700 /**
1701  * xs_tcp_write_space - callback invoked when socket buffer space
1702  *                             becomes available
1703  * @sk: socket whose state has changed
1704  *
1705  * Called when more output buffer space is available for this socket.
1706  * We try not to wake our writers until they can make "significant"
1707  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1708  * with a bunch of small requests.
1709  */
1710 static void xs_tcp_write_space(struct sock *sk)
1711 {
1712 	read_lock_bh(&sk->sk_callback_lock);
1713 
1714 	/* from net/core/stream.c:sk_stream_write_space */
1715 	if (sk_stream_is_writeable(sk))
1716 		xs_write_space(sk);
1717 
1718 	read_unlock_bh(&sk->sk_callback_lock);
1719 }
1720 
1721 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1722 {
1723 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1724 	struct sock *sk = transport->inet;
1725 
1726 	if (transport->rcvsize) {
1727 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1728 		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1729 	}
1730 	if (transport->sndsize) {
1731 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1732 		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1733 		sk->sk_write_space(sk);
1734 	}
1735 }
1736 
1737 /**
1738  * xs_udp_set_buffer_size - set send and receive limits
1739  * @xprt: generic transport
1740  * @sndsize: requested size of send buffer, in bytes
1741  * @rcvsize: requested size of receive buffer, in bytes
1742  *
1743  * Set socket send and receive buffer size limits.
1744  */
1745 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1746 {
1747 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1748 
1749 	transport->sndsize = 0;
1750 	if (sndsize)
1751 		transport->sndsize = sndsize + 1024;
1752 	transport->rcvsize = 0;
1753 	if (rcvsize)
1754 		transport->rcvsize = rcvsize + 1024;
1755 
1756 	xs_udp_do_set_buffer_size(xprt);
1757 }
1758 
1759 /**
1760  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1761  * @task: task that timed out
1762  *
1763  * Adjust the congestion window after a retransmit timeout has occurred.
1764  */
1765 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1766 {
1767 	spin_lock_bh(&xprt->transport_lock);
1768 	xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1769 	spin_unlock_bh(&xprt->transport_lock);
1770 }
1771 
1772 static unsigned short xs_get_random_port(void)
1773 {
1774 	unsigned short range = xprt_max_resvport - xprt_min_resvport + 1;
1775 	unsigned short rand = (unsigned short) prandom_u32() % range;
1776 	return rand + xprt_min_resvport;
1777 }
1778 
1779 /**
1780  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1781  * @sock: socket
1782  *
1783  * Note that this function has to be called on all sockets that share the
1784  * same port, and it must be called before binding.
1785  */
1786 static void xs_sock_set_reuseport(struct socket *sock)
1787 {
1788 	int opt = 1;
1789 
1790 	kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1791 			(char *)&opt, sizeof(opt));
1792 }
1793 
1794 static unsigned short xs_sock_getport(struct socket *sock)
1795 {
1796 	struct sockaddr_storage buf;
1797 	int buflen;
1798 	unsigned short port = 0;
1799 
1800 	if (kernel_getsockname(sock, (struct sockaddr *)&buf, &buflen) < 0)
1801 		goto out;
1802 	switch (buf.ss_family) {
1803 	case AF_INET6:
1804 		port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1805 		break;
1806 	case AF_INET:
1807 		port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1808 	}
1809 out:
1810 	return port;
1811 }
1812 
1813 /**
1814  * xs_set_port - reset the port number in the remote endpoint address
1815  * @xprt: generic transport
1816  * @port: new port number
1817  *
1818  */
1819 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1820 {
1821 	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1822 
1823 	rpc_set_port(xs_addr(xprt), port);
1824 	xs_update_peer_port(xprt);
1825 }
1826 
1827 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1828 {
1829 	if (transport->srcport == 0)
1830 		transport->srcport = xs_sock_getport(sock);
1831 }
1832 
1833 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1834 {
1835 	unsigned short port = transport->srcport;
1836 
1837 	if (port == 0 && transport->xprt.resvport)
1838 		port = xs_get_random_port();
1839 	return port;
1840 }
1841 
1842 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1843 {
1844 	if (transport->srcport != 0)
1845 		transport->srcport = 0;
1846 	if (!transport->xprt.resvport)
1847 		return 0;
1848 	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1849 		return xprt_max_resvport;
1850 	return --port;
1851 }
1852 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1853 {
1854 	struct sockaddr_storage myaddr;
1855 	int err, nloop = 0;
1856 	unsigned short port = xs_get_srcport(transport);
1857 	unsigned short last;
1858 
1859 	/*
1860 	 * If we are asking for any ephemeral port (i.e. port == 0 &&
1861 	 * transport->xprt.resvport == 0), don't bind.  Let the local
1862 	 * port selection happen implicitly when the socket is used
1863 	 * (for example at connect time).
1864 	 *
1865 	 * This ensures that we can continue to establish TCP
1866 	 * connections even when all local ephemeral ports are already
1867 	 * a part of some TCP connection.  This makes no difference
1868 	 * for UDP sockets, but also doens't harm them.
1869 	 *
1870 	 * If we're asking for any reserved port (i.e. port == 0 &&
1871 	 * transport->xprt.resvport == 1) xs_get_srcport above will
1872 	 * ensure that port is non-zero and we will bind as needed.
1873 	 */
1874 	if (port == 0)
1875 		return 0;
1876 
1877 	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1878 	do {
1879 		rpc_set_port((struct sockaddr *)&myaddr, port);
1880 		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1881 				transport->xprt.addrlen);
1882 		if (err == 0) {
1883 			transport->srcport = port;
1884 			break;
1885 		}
1886 		last = port;
1887 		port = xs_next_srcport(transport, port);
1888 		if (port > last)
1889 			nloop++;
1890 	} while (err == -EADDRINUSE && nloop != 2);
1891 
1892 	if (myaddr.ss_family == AF_INET)
1893 		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1894 				&((struct sockaddr_in *)&myaddr)->sin_addr,
1895 				port, err ? "failed" : "ok", err);
1896 	else
1897 		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1898 				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1899 				port, err ? "failed" : "ok", err);
1900 	return err;
1901 }
1902 
1903 /*
1904  * We don't support autobind on AF_LOCAL sockets
1905  */
1906 static void xs_local_rpcbind(struct rpc_task *task)
1907 {
1908 	xprt_set_bound(task->tk_xprt);
1909 }
1910 
1911 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1912 {
1913 }
1914 
1915 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1916 static struct lock_class_key xs_key[2];
1917 static struct lock_class_key xs_slock_key[2];
1918 
1919 static inline void xs_reclassify_socketu(struct socket *sock)
1920 {
1921 	struct sock *sk = sock->sk;
1922 
1923 	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1924 		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1925 }
1926 
1927 static inline void xs_reclassify_socket4(struct socket *sock)
1928 {
1929 	struct sock *sk = sock->sk;
1930 
1931 	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1932 		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1933 }
1934 
1935 static inline void xs_reclassify_socket6(struct socket *sock)
1936 {
1937 	struct sock *sk = sock->sk;
1938 
1939 	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1940 		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1941 }
1942 
1943 static inline void xs_reclassify_socket(int family, struct socket *sock)
1944 {
1945 	if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1946 		return;
1947 
1948 	switch (family) {
1949 	case AF_LOCAL:
1950 		xs_reclassify_socketu(sock);
1951 		break;
1952 	case AF_INET:
1953 		xs_reclassify_socket4(sock);
1954 		break;
1955 	case AF_INET6:
1956 		xs_reclassify_socket6(sock);
1957 		break;
1958 	}
1959 }
1960 #else
1961 static inline void xs_reclassify_socket(int family, struct socket *sock)
1962 {
1963 }
1964 #endif
1965 
1966 static void xs_dummy_setup_socket(struct work_struct *work)
1967 {
1968 }
1969 
1970 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1971 		struct sock_xprt *transport, int family, int type,
1972 		int protocol, bool reuseport)
1973 {
1974 	struct socket *sock;
1975 	int err;
1976 
1977 	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1978 	if (err < 0) {
1979 		dprintk("RPC:       can't create %d transport socket (%d).\n",
1980 				protocol, -err);
1981 		goto out;
1982 	}
1983 	xs_reclassify_socket(family, sock);
1984 
1985 	if (reuseport)
1986 		xs_sock_set_reuseport(sock);
1987 
1988 	err = xs_bind(transport, sock);
1989 	if (err) {
1990 		sock_release(sock);
1991 		goto out;
1992 	}
1993 
1994 	return sock;
1995 out:
1996 	return ERR_PTR(err);
1997 }
1998 
1999 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
2000 				      struct socket *sock)
2001 {
2002 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2003 									xprt);
2004 
2005 	if (!transport->inet) {
2006 		struct sock *sk = sock->sk;
2007 
2008 		write_lock_bh(&sk->sk_callback_lock);
2009 
2010 		xs_save_old_callbacks(transport, sk);
2011 
2012 		sk->sk_user_data = xprt;
2013 		sk->sk_data_ready = xs_data_ready;
2014 		sk->sk_write_space = xs_udp_write_space;
2015 		sock_set_flag(sk, SOCK_FASYNC);
2016 		sk->sk_error_report = xs_error_report;
2017 		sk->sk_allocation = GFP_NOIO;
2018 
2019 		xprt_clear_connected(xprt);
2020 
2021 		/* Reset to new socket */
2022 		transport->sock = sock;
2023 		transport->inet = sk;
2024 
2025 		write_unlock_bh(&sk->sk_callback_lock);
2026 	}
2027 
2028 	/* Tell the socket layer to start connecting... */
2029 	xprt->stat.connect_count++;
2030 	xprt->stat.connect_start = jiffies;
2031 	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
2032 }
2033 
2034 /**
2035  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
2036  * @transport: socket transport to connect
2037  */
2038 static int xs_local_setup_socket(struct sock_xprt *transport)
2039 {
2040 	struct rpc_xprt *xprt = &transport->xprt;
2041 	struct socket *sock;
2042 	int status = -EIO;
2043 
2044 	status = __sock_create(xprt->xprt_net, AF_LOCAL,
2045 					SOCK_STREAM, 0, &sock, 1);
2046 	if (status < 0) {
2047 		dprintk("RPC:       can't create AF_LOCAL "
2048 			"transport socket (%d).\n", -status);
2049 		goto out;
2050 	}
2051 	xs_reclassify_socket(AF_LOCAL, sock);
2052 
2053 	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
2054 			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2055 
2056 	status = xs_local_finish_connecting(xprt, sock);
2057 	trace_rpc_socket_connect(xprt, sock, status);
2058 	switch (status) {
2059 	case 0:
2060 		dprintk("RPC:       xprt %p connected to %s\n",
2061 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2062 		xprt_set_connected(xprt);
2063 	case -ENOBUFS:
2064 		break;
2065 	case -ENOENT:
2066 		dprintk("RPC:       xprt %p: socket %s does not exist\n",
2067 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2068 		break;
2069 	case -ECONNREFUSED:
2070 		dprintk("RPC:       xprt %p: connection refused for %s\n",
2071 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2072 		break;
2073 	default:
2074 		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2075 				__func__, -status,
2076 				xprt->address_strings[RPC_DISPLAY_ADDR]);
2077 	}
2078 
2079 out:
2080 	xprt_clear_connecting(xprt);
2081 	xprt_wake_pending_tasks(xprt, status);
2082 	return status;
2083 }
2084 
2085 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2086 {
2087 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2088 	int ret;
2089 
2090 	 if (RPC_IS_ASYNC(task)) {
2091 		/*
2092 		 * We want the AF_LOCAL connect to be resolved in the
2093 		 * filesystem namespace of the process making the rpc
2094 		 * call.  Thus we connect synchronously.
2095 		 *
2096 		 * If we want to support asynchronous AF_LOCAL calls,
2097 		 * we'll need to figure out how to pass a namespace to
2098 		 * connect.
2099 		 */
2100 		rpc_exit(task, -ENOTCONN);
2101 		return;
2102 	}
2103 	ret = xs_local_setup_socket(transport);
2104 	if (ret && !RPC_IS_SOFTCONN(task))
2105 		msleep_interruptible(15000);
2106 }
2107 
2108 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2109 /*
2110  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2111  * know that we have exclusive access to the socket), to guard against
2112  * races with xs_reset_transport.
2113  */
2114 static void xs_set_memalloc(struct rpc_xprt *xprt)
2115 {
2116 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2117 			xprt);
2118 
2119 	/*
2120 	 * If there's no sock, then we have nothing to set. The
2121 	 * reconnecting process will get it for us.
2122 	 */
2123 	if (!transport->inet)
2124 		return;
2125 	if (atomic_read(&xprt->swapper))
2126 		sk_set_memalloc(transport->inet);
2127 }
2128 
2129 /**
2130  * xs_enable_swap - Tag this transport as being used for swap.
2131  * @xprt: transport to tag
2132  *
2133  * Take a reference to this transport on behalf of the rpc_clnt, and
2134  * optionally mark it for swapping if it wasn't already.
2135  */
2136 static int
2137 xs_enable_swap(struct rpc_xprt *xprt)
2138 {
2139 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2140 
2141 	if (atomic_inc_return(&xprt->swapper) != 1)
2142 		return 0;
2143 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2144 		return -ERESTARTSYS;
2145 	if (xs->inet)
2146 		sk_set_memalloc(xs->inet);
2147 	xprt_release_xprt(xprt, NULL);
2148 	return 0;
2149 }
2150 
2151 /**
2152  * xs_disable_swap - Untag this transport as being used for swap.
2153  * @xprt: transport to tag
2154  *
2155  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2156  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2157  */
2158 static void
2159 xs_disable_swap(struct rpc_xprt *xprt)
2160 {
2161 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2162 
2163 	if (!atomic_dec_and_test(&xprt->swapper))
2164 		return;
2165 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2166 		return;
2167 	if (xs->inet)
2168 		sk_clear_memalloc(xs->inet);
2169 	xprt_release_xprt(xprt, NULL);
2170 }
2171 #else
2172 static void xs_set_memalloc(struct rpc_xprt *xprt)
2173 {
2174 }
2175 
2176 static int
2177 xs_enable_swap(struct rpc_xprt *xprt)
2178 {
2179 	return -EINVAL;
2180 }
2181 
2182 static void
2183 xs_disable_swap(struct rpc_xprt *xprt)
2184 {
2185 }
2186 #endif
2187 
2188 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2189 {
2190 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2191 
2192 	if (!transport->inet) {
2193 		struct sock *sk = sock->sk;
2194 
2195 		write_lock_bh(&sk->sk_callback_lock);
2196 
2197 		xs_save_old_callbacks(transport, sk);
2198 
2199 		sk->sk_user_data = xprt;
2200 		sk->sk_data_ready = xs_data_ready;
2201 		sk->sk_write_space = xs_udp_write_space;
2202 		sock_set_flag(sk, SOCK_FASYNC);
2203 		sk->sk_allocation = GFP_NOIO;
2204 
2205 		xprt_set_connected(xprt);
2206 
2207 		/* Reset to new socket */
2208 		transport->sock = sock;
2209 		transport->inet = sk;
2210 
2211 		xs_set_memalloc(xprt);
2212 
2213 		write_unlock_bh(&sk->sk_callback_lock);
2214 	}
2215 	xs_udp_do_set_buffer_size(xprt);
2216 
2217 	xprt->stat.connect_start = jiffies;
2218 }
2219 
2220 static void xs_udp_setup_socket(struct work_struct *work)
2221 {
2222 	struct sock_xprt *transport =
2223 		container_of(work, struct sock_xprt, connect_worker.work);
2224 	struct rpc_xprt *xprt = &transport->xprt;
2225 	struct socket *sock;
2226 	int status = -EIO;
2227 
2228 	sock = xs_create_sock(xprt, transport,
2229 			xs_addr(xprt)->sa_family, SOCK_DGRAM,
2230 			IPPROTO_UDP, false);
2231 	if (IS_ERR(sock))
2232 		goto out;
2233 
2234 	dprintk("RPC:       worker connecting xprt %p via %s to "
2235 				"%s (port %s)\n", xprt,
2236 			xprt->address_strings[RPC_DISPLAY_PROTO],
2237 			xprt->address_strings[RPC_DISPLAY_ADDR],
2238 			xprt->address_strings[RPC_DISPLAY_PORT]);
2239 
2240 	xs_udp_finish_connecting(xprt, sock);
2241 	trace_rpc_socket_connect(xprt, sock, 0);
2242 	status = 0;
2243 out:
2244 	xprt_unlock_connect(xprt, transport);
2245 	xprt_clear_connecting(xprt);
2246 	xprt_wake_pending_tasks(xprt, status);
2247 }
2248 
2249 /**
2250  * xs_tcp_shutdown - gracefully shut down a TCP socket
2251  * @xprt: transport
2252  *
2253  * Initiates a graceful shutdown of the TCP socket by calling the
2254  * equivalent of shutdown(SHUT_RDWR);
2255  */
2256 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2257 {
2258 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2259 	struct socket *sock = transport->sock;
2260 	int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2261 
2262 	if (sock == NULL)
2263 		return;
2264 	switch (skst) {
2265 	default:
2266 		kernel_sock_shutdown(sock, SHUT_RDWR);
2267 		trace_rpc_socket_shutdown(xprt, sock);
2268 		break;
2269 	case TCP_CLOSE:
2270 	case TCP_TIME_WAIT:
2271 		xs_reset_transport(transport);
2272 	}
2273 }
2274 
2275 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2276 		struct socket *sock)
2277 {
2278 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2279 	unsigned int keepidle;
2280 	unsigned int keepcnt;
2281 	unsigned int opt_on = 1;
2282 	unsigned int timeo;
2283 
2284 	spin_lock_bh(&xprt->transport_lock);
2285 	keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2286 	keepcnt = xprt->timeout->to_retries + 1;
2287 	timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2288 		(xprt->timeout->to_retries + 1);
2289 	clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2290 	spin_unlock_bh(&xprt->transport_lock);
2291 
2292 	/* TCP Keepalive options */
2293 	kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2294 			(char *)&opt_on, sizeof(opt_on));
2295 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2296 			(char *)&keepidle, sizeof(keepidle));
2297 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2298 			(char *)&keepidle, sizeof(keepidle));
2299 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2300 			(char *)&keepcnt, sizeof(keepcnt));
2301 
2302 	/* TCP user timeout (see RFC5482) */
2303 	kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2304 			(char *)&timeo, sizeof(timeo));
2305 }
2306 
2307 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2308 		unsigned long connect_timeout,
2309 		unsigned long reconnect_timeout)
2310 {
2311 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2312 	struct rpc_timeout to;
2313 	unsigned long initval;
2314 
2315 	spin_lock_bh(&xprt->transport_lock);
2316 	if (reconnect_timeout < xprt->max_reconnect_timeout)
2317 		xprt->max_reconnect_timeout = reconnect_timeout;
2318 	if (connect_timeout < xprt->connect_timeout) {
2319 		memcpy(&to, xprt->timeout, sizeof(to));
2320 		initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2321 		/* Arbitrary lower limit */
2322 		if (initval <  XS_TCP_INIT_REEST_TO << 1)
2323 			initval = XS_TCP_INIT_REEST_TO << 1;
2324 		to.to_initval = initval;
2325 		to.to_maxval = initval;
2326 		memcpy(&transport->tcp_timeout, &to,
2327 				sizeof(transport->tcp_timeout));
2328 		xprt->timeout = &transport->tcp_timeout;
2329 		xprt->connect_timeout = connect_timeout;
2330 	}
2331 	set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2332 	spin_unlock_bh(&xprt->transport_lock);
2333 }
2334 
2335 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2336 {
2337 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2338 	int ret = -ENOTCONN;
2339 
2340 	if (!transport->inet) {
2341 		struct sock *sk = sock->sk;
2342 		unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2343 
2344 		/* Avoid temporary address, they are bad for long-lived
2345 		 * connections such as NFS mounts.
2346 		 * RFC4941, section 3.6 suggests that:
2347 		 *    Individual applications, which have specific
2348 		 *    knowledge about the normal duration of connections,
2349 		 *    MAY override this as appropriate.
2350 		 */
2351 		kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2352 				(char *)&addr_pref, sizeof(addr_pref));
2353 
2354 		xs_tcp_set_socket_timeouts(xprt, sock);
2355 
2356 		write_lock_bh(&sk->sk_callback_lock);
2357 
2358 		xs_save_old_callbacks(transport, sk);
2359 
2360 		sk->sk_user_data = xprt;
2361 		sk->sk_data_ready = xs_data_ready;
2362 		sk->sk_state_change = xs_tcp_state_change;
2363 		sk->sk_write_space = xs_tcp_write_space;
2364 		sock_set_flag(sk, SOCK_FASYNC);
2365 		sk->sk_error_report = xs_error_report;
2366 		sk->sk_allocation = GFP_NOIO;
2367 
2368 		/* socket options */
2369 		sock_reset_flag(sk, SOCK_LINGER);
2370 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2371 
2372 		xprt_clear_connected(xprt);
2373 
2374 		/* Reset to new socket */
2375 		transport->sock = sock;
2376 		transport->inet = sk;
2377 
2378 		write_unlock_bh(&sk->sk_callback_lock);
2379 	}
2380 
2381 	if (!xprt_bound(xprt))
2382 		goto out;
2383 
2384 	xs_set_memalloc(xprt);
2385 
2386 	/* Tell the socket layer to start connecting... */
2387 	xprt->stat.connect_count++;
2388 	xprt->stat.connect_start = jiffies;
2389 	set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2390 	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2391 	switch (ret) {
2392 	case 0:
2393 		xs_set_srcport(transport, sock);
2394 		/* fall through */
2395 	case -EINPROGRESS:
2396 		/* SYN_SENT! */
2397 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2398 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2399 		break;
2400 	case -EADDRNOTAVAIL:
2401 		/* Source port number is unavailable. Try a new one! */
2402 		transport->srcport = 0;
2403 	}
2404 out:
2405 	return ret;
2406 }
2407 
2408 /**
2409  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2410  *
2411  * Invoked by a work queue tasklet.
2412  */
2413 static void xs_tcp_setup_socket(struct work_struct *work)
2414 {
2415 	struct sock_xprt *transport =
2416 		container_of(work, struct sock_xprt, connect_worker.work);
2417 	struct socket *sock = transport->sock;
2418 	struct rpc_xprt *xprt = &transport->xprt;
2419 	int status = -EIO;
2420 
2421 	if (!sock) {
2422 		sock = xs_create_sock(xprt, transport,
2423 				xs_addr(xprt)->sa_family, SOCK_STREAM,
2424 				IPPROTO_TCP, true);
2425 		if (IS_ERR(sock)) {
2426 			status = PTR_ERR(sock);
2427 			goto out;
2428 		}
2429 	}
2430 
2431 	dprintk("RPC:       worker connecting xprt %p via %s to "
2432 				"%s (port %s)\n", xprt,
2433 			xprt->address_strings[RPC_DISPLAY_PROTO],
2434 			xprt->address_strings[RPC_DISPLAY_ADDR],
2435 			xprt->address_strings[RPC_DISPLAY_PORT]);
2436 
2437 	status = xs_tcp_finish_connecting(xprt, sock);
2438 	trace_rpc_socket_connect(xprt, sock, status);
2439 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2440 			xprt, -status, xprt_connected(xprt),
2441 			sock->sk->sk_state);
2442 	switch (status) {
2443 	default:
2444 		printk("%s: connect returned unhandled error %d\n",
2445 			__func__, status);
2446 		/* fall through */
2447 	case -EADDRNOTAVAIL:
2448 		/* We're probably in TIME_WAIT. Get rid of existing socket,
2449 		 * and retry
2450 		 */
2451 		xs_tcp_force_close(xprt);
2452 		break;
2453 	case 0:
2454 	case -EINPROGRESS:
2455 	case -EALREADY:
2456 		xprt_unlock_connect(xprt, transport);
2457 		return;
2458 	case -EINVAL:
2459 		/* Happens, for instance, if the user specified a link
2460 		 * local IPv6 address without a scope-id.
2461 		 */
2462 	case -ECONNREFUSED:
2463 	case -ECONNRESET:
2464 	case -ENETDOWN:
2465 	case -ENETUNREACH:
2466 	case -EHOSTUNREACH:
2467 	case -EADDRINUSE:
2468 	case -ENOBUFS:
2469 		/*
2470 		 * xs_tcp_force_close() wakes tasks with -EIO.
2471 		 * We need to wake them first to ensure the
2472 		 * correct error code.
2473 		 */
2474 		xprt_wake_pending_tasks(xprt, status);
2475 		xs_tcp_force_close(xprt);
2476 		goto out;
2477 	}
2478 	status = -EAGAIN;
2479 out:
2480 	xprt_unlock_connect(xprt, transport);
2481 	xprt_clear_connecting(xprt);
2482 	xprt_wake_pending_tasks(xprt, status);
2483 }
2484 
2485 static unsigned long xs_reconnect_delay(const struct rpc_xprt *xprt)
2486 {
2487 	unsigned long start, now = jiffies;
2488 
2489 	start = xprt->stat.connect_start + xprt->reestablish_timeout;
2490 	if (time_after(start, now))
2491 		return start - now;
2492 	return 0;
2493 }
2494 
2495 static void xs_reconnect_backoff(struct rpc_xprt *xprt)
2496 {
2497 	xprt->reestablish_timeout <<= 1;
2498 	if (xprt->reestablish_timeout > xprt->max_reconnect_timeout)
2499 		xprt->reestablish_timeout = xprt->max_reconnect_timeout;
2500 	if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2501 		xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2502 }
2503 
2504 /**
2505  * xs_connect - connect a socket to a remote endpoint
2506  * @xprt: pointer to transport structure
2507  * @task: address of RPC task that manages state of connect request
2508  *
2509  * TCP: If the remote end dropped the connection, delay reconnecting.
2510  *
2511  * UDP socket connects are synchronous, but we use a work queue anyway
2512  * to guarantee that even unprivileged user processes can set up a
2513  * socket on a privileged port.
2514  *
2515  * If a UDP socket connect fails, the delay behavior here prevents
2516  * retry floods (hard mounts).
2517  */
2518 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2519 {
2520 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2521 	unsigned long delay = 0;
2522 
2523 	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2524 
2525 	if (transport->sock != NULL) {
2526 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2527 				"seconds\n",
2528 				xprt, xprt->reestablish_timeout / HZ);
2529 
2530 		/* Start by resetting any existing state */
2531 		xs_reset_transport(transport);
2532 
2533 		delay = xs_reconnect_delay(xprt);
2534 		xs_reconnect_backoff(xprt);
2535 
2536 	} else
2537 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2538 
2539 	queue_delayed_work(xprtiod_workqueue,
2540 			&transport->connect_worker,
2541 			delay);
2542 }
2543 
2544 /**
2545  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2546  * @xprt: rpc_xprt struct containing statistics
2547  * @seq: output file
2548  *
2549  */
2550 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2551 {
2552 	long idle_time = 0;
2553 
2554 	if (xprt_connected(xprt))
2555 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2556 
2557 	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2558 			"%llu %llu %lu %llu %llu\n",
2559 			xprt->stat.bind_count,
2560 			xprt->stat.connect_count,
2561 			xprt->stat.connect_time,
2562 			idle_time,
2563 			xprt->stat.sends,
2564 			xprt->stat.recvs,
2565 			xprt->stat.bad_xids,
2566 			xprt->stat.req_u,
2567 			xprt->stat.bklog_u,
2568 			xprt->stat.max_slots,
2569 			xprt->stat.sending_u,
2570 			xprt->stat.pending_u);
2571 }
2572 
2573 /**
2574  * xs_udp_print_stats - display UDP socket-specifc stats
2575  * @xprt: rpc_xprt struct containing statistics
2576  * @seq: output file
2577  *
2578  */
2579 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2580 {
2581 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2582 
2583 	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2584 			"%lu %llu %llu\n",
2585 			transport->srcport,
2586 			xprt->stat.bind_count,
2587 			xprt->stat.sends,
2588 			xprt->stat.recvs,
2589 			xprt->stat.bad_xids,
2590 			xprt->stat.req_u,
2591 			xprt->stat.bklog_u,
2592 			xprt->stat.max_slots,
2593 			xprt->stat.sending_u,
2594 			xprt->stat.pending_u);
2595 }
2596 
2597 /**
2598  * xs_tcp_print_stats - display TCP socket-specifc stats
2599  * @xprt: rpc_xprt struct containing statistics
2600  * @seq: output file
2601  *
2602  */
2603 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2604 {
2605 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2606 	long idle_time = 0;
2607 
2608 	if (xprt_connected(xprt))
2609 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2610 
2611 	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2612 			"%llu %llu %lu %llu %llu\n",
2613 			transport->srcport,
2614 			xprt->stat.bind_count,
2615 			xprt->stat.connect_count,
2616 			xprt->stat.connect_time,
2617 			idle_time,
2618 			xprt->stat.sends,
2619 			xprt->stat.recvs,
2620 			xprt->stat.bad_xids,
2621 			xprt->stat.req_u,
2622 			xprt->stat.bklog_u,
2623 			xprt->stat.max_slots,
2624 			xprt->stat.sending_u,
2625 			xprt->stat.pending_u);
2626 }
2627 
2628 /*
2629  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2630  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2631  * to use the server side send routines.
2632  */
2633 static int bc_malloc(struct rpc_task *task)
2634 {
2635 	struct rpc_rqst *rqst = task->tk_rqstp;
2636 	size_t size = rqst->rq_callsize;
2637 	struct page *page;
2638 	struct rpc_buffer *buf;
2639 
2640 	if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2641 		WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2642 			  size);
2643 		return -EINVAL;
2644 	}
2645 
2646 	page = alloc_page(GFP_KERNEL);
2647 	if (!page)
2648 		return -ENOMEM;
2649 
2650 	buf = page_address(page);
2651 	buf->len = PAGE_SIZE;
2652 
2653 	rqst->rq_buffer = buf->data;
2654 	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2655 	return 0;
2656 }
2657 
2658 /*
2659  * Free the space allocated in the bc_alloc routine
2660  */
2661 static void bc_free(struct rpc_task *task)
2662 {
2663 	void *buffer = task->tk_rqstp->rq_buffer;
2664 	struct rpc_buffer *buf;
2665 
2666 	buf = container_of(buffer, struct rpc_buffer, data);
2667 	free_page((unsigned long)buf);
2668 }
2669 
2670 /*
2671  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2672  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2673  */
2674 static int bc_sendto(struct rpc_rqst *req)
2675 {
2676 	int len;
2677 	struct xdr_buf *xbufp = &req->rq_snd_buf;
2678 	struct rpc_xprt *xprt = req->rq_xprt;
2679 	struct sock_xprt *transport =
2680 				container_of(xprt, struct sock_xprt, xprt);
2681 	struct socket *sock = transport->sock;
2682 	unsigned long headoff;
2683 	unsigned long tailoff;
2684 
2685 	xs_encode_stream_record_marker(xbufp);
2686 
2687 	tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2688 	headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2689 	len = svc_send_common(sock, xbufp,
2690 			      virt_to_page(xbufp->head[0].iov_base), headoff,
2691 			      xbufp->tail[0].iov_base, tailoff);
2692 
2693 	if (len != xbufp->len) {
2694 		printk(KERN_NOTICE "Error sending entire callback!\n");
2695 		len = -EAGAIN;
2696 	}
2697 
2698 	return len;
2699 }
2700 
2701 /*
2702  * The send routine. Borrows from svc_send
2703  */
2704 static int bc_send_request(struct rpc_task *task)
2705 {
2706 	struct rpc_rqst *req = task->tk_rqstp;
2707 	struct svc_xprt	*xprt;
2708 	int len;
2709 
2710 	dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2711 	/*
2712 	 * Get the server socket associated with this callback xprt
2713 	 */
2714 	xprt = req->rq_xprt->bc_xprt;
2715 
2716 	/*
2717 	 * Grab the mutex to serialize data as the connection is shared
2718 	 * with the fore channel
2719 	 */
2720 	if (!mutex_trylock(&xprt->xpt_mutex)) {
2721 		rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2722 		if (!mutex_trylock(&xprt->xpt_mutex))
2723 			return -EAGAIN;
2724 		rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2725 	}
2726 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2727 		len = -ENOTCONN;
2728 	else
2729 		len = bc_sendto(req);
2730 	mutex_unlock(&xprt->xpt_mutex);
2731 
2732 	if (len > 0)
2733 		len = 0;
2734 
2735 	return len;
2736 }
2737 
2738 /*
2739  * The close routine. Since this is client initiated, we do nothing
2740  */
2741 
2742 static void bc_close(struct rpc_xprt *xprt)
2743 {
2744 }
2745 
2746 /*
2747  * The xprt destroy routine. Again, because this connection is client
2748  * initiated, we do nothing
2749  */
2750 
2751 static void bc_destroy(struct rpc_xprt *xprt)
2752 {
2753 	dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2754 
2755 	xs_xprt_free(xprt);
2756 	module_put(THIS_MODULE);
2757 }
2758 
2759 static const struct rpc_xprt_ops xs_local_ops = {
2760 	.reserve_xprt		= xprt_reserve_xprt,
2761 	.release_xprt		= xs_tcp_release_xprt,
2762 	.alloc_slot		= xprt_alloc_slot,
2763 	.rpcbind		= xs_local_rpcbind,
2764 	.set_port		= xs_local_set_port,
2765 	.connect		= xs_local_connect,
2766 	.buf_alloc		= rpc_malloc,
2767 	.buf_free		= rpc_free,
2768 	.send_request		= xs_local_send_request,
2769 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2770 	.close			= xs_close,
2771 	.destroy		= xs_destroy,
2772 	.print_stats		= xs_local_print_stats,
2773 	.enable_swap		= xs_enable_swap,
2774 	.disable_swap		= xs_disable_swap,
2775 };
2776 
2777 static const struct rpc_xprt_ops xs_udp_ops = {
2778 	.set_buffer_size	= xs_udp_set_buffer_size,
2779 	.reserve_xprt		= xprt_reserve_xprt_cong,
2780 	.release_xprt		= xprt_release_xprt_cong,
2781 	.alloc_slot		= xprt_alloc_slot,
2782 	.rpcbind		= rpcb_getport_async,
2783 	.set_port		= xs_set_port,
2784 	.connect		= xs_connect,
2785 	.buf_alloc		= rpc_malloc,
2786 	.buf_free		= rpc_free,
2787 	.send_request		= xs_udp_send_request,
2788 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
2789 	.timer			= xs_udp_timer,
2790 	.release_request	= xprt_release_rqst_cong,
2791 	.close			= xs_close,
2792 	.destroy		= xs_destroy,
2793 	.print_stats		= xs_udp_print_stats,
2794 	.enable_swap		= xs_enable_swap,
2795 	.disable_swap		= xs_disable_swap,
2796 	.inject_disconnect	= xs_inject_disconnect,
2797 };
2798 
2799 static const struct rpc_xprt_ops xs_tcp_ops = {
2800 	.reserve_xprt		= xprt_reserve_xprt,
2801 	.release_xprt		= xs_tcp_release_xprt,
2802 	.alloc_slot		= xprt_lock_and_alloc_slot,
2803 	.rpcbind		= rpcb_getport_async,
2804 	.set_port		= xs_set_port,
2805 	.connect		= xs_connect,
2806 	.buf_alloc		= rpc_malloc,
2807 	.buf_free		= rpc_free,
2808 	.send_request		= xs_tcp_send_request,
2809 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2810 	.close			= xs_tcp_shutdown,
2811 	.destroy		= xs_destroy,
2812 	.set_connect_timeout	= xs_tcp_set_connect_timeout,
2813 	.print_stats		= xs_tcp_print_stats,
2814 	.enable_swap		= xs_enable_swap,
2815 	.disable_swap		= xs_disable_swap,
2816 	.inject_disconnect	= xs_inject_disconnect,
2817 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2818 	.bc_setup		= xprt_setup_bc,
2819 	.bc_up			= xs_tcp_bc_up,
2820 	.bc_maxpayload		= xs_tcp_bc_maxpayload,
2821 	.bc_free_rqst		= xprt_free_bc_rqst,
2822 	.bc_destroy		= xprt_destroy_bc,
2823 #endif
2824 };
2825 
2826 /*
2827  * The rpc_xprt_ops for the server backchannel
2828  */
2829 
2830 static const struct rpc_xprt_ops bc_tcp_ops = {
2831 	.reserve_xprt		= xprt_reserve_xprt,
2832 	.release_xprt		= xprt_release_xprt,
2833 	.alloc_slot		= xprt_alloc_slot,
2834 	.buf_alloc		= bc_malloc,
2835 	.buf_free		= bc_free,
2836 	.send_request		= bc_send_request,
2837 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2838 	.close			= bc_close,
2839 	.destroy		= bc_destroy,
2840 	.print_stats		= xs_tcp_print_stats,
2841 	.enable_swap		= xs_enable_swap,
2842 	.disable_swap		= xs_disable_swap,
2843 	.inject_disconnect	= xs_inject_disconnect,
2844 };
2845 
2846 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2847 {
2848 	static const struct sockaddr_in sin = {
2849 		.sin_family		= AF_INET,
2850 		.sin_addr.s_addr	= htonl(INADDR_ANY),
2851 	};
2852 	static const struct sockaddr_in6 sin6 = {
2853 		.sin6_family		= AF_INET6,
2854 		.sin6_addr		= IN6ADDR_ANY_INIT,
2855 	};
2856 
2857 	switch (family) {
2858 	case AF_LOCAL:
2859 		break;
2860 	case AF_INET:
2861 		memcpy(sap, &sin, sizeof(sin));
2862 		break;
2863 	case AF_INET6:
2864 		memcpy(sap, &sin6, sizeof(sin6));
2865 		break;
2866 	default:
2867 		dprintk("RPC:       %s: Bad address family\n", __func__);
2868 		return -EAFNOSUPPORT;
2869 	}
2870 	return 0;
2871 }
2872 
2873 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2874 				      unsigned int slot_table_size,
2875 				      unsigned int max_slot_table_size)
2876 {
2877 	struct rpc_xprt *xprt;
2878 	struct sock_xprt *new;
2879 
2880 	if (args->addrlen > sizeof(xprt->addr)) {
2881 		dprintk("RPC:       xs_setup_xprt: address too large\n");
2882 		return ERR_PTR(-EBADF);
2883 	}
2884 
2885 	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2886 			max_slot_table_size);
2887 	if (xprt == NULL) {
2888 		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2889 				"rpc_xprt\n");
2890 		return ERR_PTR(-ENOMEM);
2891 	}
2892 
2893 	new = container_of(xprt, struct sock_xprt, xprt);
2894 	mutex_init(&new->recv_mutex);
2895 	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2896 	xprt->addrlen = args->addrlen;
2897 	if (args->srcaddr)
2898 		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2899 	else {
2900 		int err;
2901 		err = xs_init_anyaddr(args->dstaddr->sa_family,
2902 					(struct sockaddr *)&new->srcaddr);
2903 		if (err != 0) {
2904 			xprt_free(xprt);
2905 			return ERR_PTR(err);
2906 		}
2907 	}
2908 
2909 	return xprt;
2910 }
2911 
2912 static const struct rpc_timeout xs_local_default_timeout = {
2913 	.to_initval = 10 * HZ,
2914 	.to_maxval = 10 * HZ,
2915 	.to_retries = 2,
2916 };
2917 
2918 /**
2919  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2920  * @args: rpc transport creation arguments
2921  *
2922  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2923  */
2924 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2925 {
2926 	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2927 	struct sock_xprt *transport;
2928 	struct rpc_xprt *xprt;
2929 	struct rpc_xprt *ret;
2930 
2931 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2932 			xprt_max_tcp_slot_table_entries);
2933 	if (IS_ERR(xprt))
2934 		return xprt;
2935 	transport = container_of(xprt, struct sock_xprt, xprt);
2936 
2937 	xprt->prot = 0;
2938 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2939 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2940 
2941 	xprt->bind_timeout = XS_BIND_TO;
2942 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2943 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2944 
2945 	xprt->ops = &xs_local_ops;
2946 	xprt->timeout = &xs_local_default_timeout;
2947 
2948 	INIT_WORK(&transport->recv_worker, xs_local_data_receive_workfn);
2949 	INIT_DELAYED_WORK(&transport->connect_worker,
2950 			xs_dummy_setup_socket);
2951 
2952 	switch (sun->sun_family) {
2953 	case AF_LOCAL:
2954 		if (sun->sun_path[0] != '/') {
2955 			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2956 					sun->sun_path);
2957 			ret = ERR_PTR(-EINVAL);
2958 			goto out_err;
2959 		}
2960 		xprt_set_bound(xprt);
2961 		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2962 		ret = ERR_PTR(xs_local_setup_socket(transport));
2963 		if (ret)
2964 			goto out_err;
2965 		break;
2966 	default:
2967 		ret = ERR_PTR(-EAFNOSUPPORT);
2968 		goto out_err;
2969 	}
2970 
2971 	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2972 			xprt->address_strings[RPC_DISPLAY_ADDR]);
2973 
2974 	if (try_module_get(THIS_MODULE))
2975 		return xprt;
2976 	ret = ERR_PTR(-EINVAL);
2977 out_err:
2978 	xs_xprt_free(xprt);
2979 	return ret;
2980 }
2981 
2982 static const struct rpc_timeout xs_udp_default_timeout = {
2983 	.to_initval = 5 * HZ,
2984 	.to_maxval = 30 * HZ,
2985 	.to_increment = 5 * HZ,
2986 	.to_retries = 5,
2987 };
2988 
2989 /**
2990  * xs_setup_udp - Set up transport to use a UDP socket
2991  * @args: rpc transport creation arguments
2992  *
2993  */
2994 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2995 {
2996 	struct sockaddr *addr = args->dstaddr;
2997 	struct rpc_xprt *xprt;
2998 	struct sock_xprt *transport;
2999 	struct rpc_xprt *ret;
3000 
3001 	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
3002 			xprt_udp_slot_table_entries);
3003 	if (IS_ERR(xprt))
3004 		return xprt;
3005 	transport = container_of(xprt, struct sock_xprt, xprt);
3006 
3007 	xprt->prot = IPPROTO_UDP;
3008 	xprt->tsh_size = 0;
3009 	/* XXX: header size can vary due to auth type, IPv6, etc. */
3010 	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
3011 
3012 	xprt->bind_timeout = XS_BIND_TO;
3013 	xprt->reestablish_timeout = XS_UDP_REEST_TO;
3014 	xprt->idle_timeout = XS_IDLE_DISC_TO;
3015 
3016 	xprt->ops = &xs_udp_ops;
3017 
3018 	xprt->timeout = &xs_udp_default_timeout;
3019 
3020 	INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
3021 	INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3022 
3023 	switch (addr->sa_family) {
3024 	case AF_INET:
3025 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3026 			xprt_set_bound(xprt);
3027 
3028 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3029 		break;
3030 	case AF_INET6:
3031 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3032 			xprt_set_bound(xprt);
3033 
3034 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3035 		break;
3036 	default:
3037 		ret = ERR_PTR(-EAFNOSUPPORT);
3038 		goto out_err;
3039 	}
3040 
3041 	if (xprt_bound(xprt))
3042 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3043 				xprt->address_strings[RPC_DISPLAY_ADDR],
3044 				xprt->address_strings[RPC_DISPLAY_PORT],
3045 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3046 	else
3047 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3048 				xprt->address_strings[RPC_DISPLAY_ADDR],
3049 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3050 
3051 	if (try_module_get(THIS_MODULE))
3052 		return xprt;
3053 	ret = ERR_PTR(-EINVAL);
3054 out_err:
3055 	xs_xprt_free(xprt);
3056 	return ret;
3057 }
3058 
3059 static const struct rpc_timeout xs_tcp_default_timeout = {
3060 	.to_initval = 60 * HZ,
3061 	.to_maxval = 60 * HZ,
3062 	.to_retries = 2,
3063 };
3064 
3065 /**
3066  * xs_setup_tcp - Set up transport to use a TCP socket
3067  * @args: rpc transport creation arguments
3068  *
3069  */
3070 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3071 {
3072 	struct sockaddr *addr = args->dstaddr;
3073 	struct rpc_xprt *xprt;
3074 	struct sock_xprt *transport;
3075 	struct rpc_xprt *ret;
3076 	unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3077 
3078 	if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3079 		max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3080 
3081 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3082 			max_slot_table_size);
3083 	if (IS_ERR(xprt))
3084 		return xprt;
3085 	transport = container_of(xprt, struct sock_xprt, xprt);
3086 
3087 	xprt->prot = IPPROTO_TCP;
3088 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3089 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3090 
3091 	xprt->bind_timeout = XS_BIND_TO;
3092 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3093 	xprt->idle_timeout = XS_IDLE_DISC_TO;
3094 
3095 	xprt->ops = &xs_tcp_ops;
3096 	xprt->timeout = &xs_tcp_default_timeout;
3097 
3098 	xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3099 	xprt->connect_timeout = xprt->timeout->to_initval *
3100 		(xprt->timeout->to_retries + 1);
3101 
3102 	INIT_WORK(&transport->recv_worker, xs_tcp_data_receive_workfn);
3103 	INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3104 
3105 	switch (addr->sa_family) {
3106 	case AF_INET:
3107 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3108 			xprt_set_bound(xprt);
3109 
3110 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3111 		break;
3112 	case AF_INET6:
3113 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3114 			xprt_set_bound(xprt);
3115 
3116 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3117 		break;
3118 	default:
3119 		ret = ERR_PTR(-EAFNOSUPPORT);
3120 		goto out_err;
3121 	}
3122 
3123 	if (xprt_bound(xprt))
3124 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3125 				xprt->address_strings[RPC_DISPLAY_ADDR],
3126 				xprt->address_strings[RPC_DISPLAY_PORT],
3127 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3128 	else
3129 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3130 				xprt->address_strings[RPC_DISPLAY_ADDR],
3131 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3132 
3133 	if (try_module_get(THIS_MODULE))
3134 		return xprt;
3135 	ret = ERR_PTR(-EINVAL);
3136 out_err:
3137 	xs_xprt_free(xprt);
3138 	return ret;
3139 }
3140 
3141 /**
3142  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3143  * @args: rpc transport creation arguments
3144  *
3145  */
3146 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3147 {
3148 	struct sockaddr *addr = args->dstaddr;
3149 	struct rpc_xprt *xprt;
3150 	struct sock_xprt *transport;
3151 	struct svc_sock *bc_sock;
3152 	struct rpc_xprt *ret;
3153 
3154 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3155 			xprt_tcp_slot_table_entries);
3156 	if (IS_ERR(xprt))
3157 		return xprt;
3158 	transport = container_of(xprt, struct sock_xprt, xprt);
3159 
3160 	xprt->prot = IPPROTO_TCP;
3161 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
3162 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3163 	xprt->timeout = &xs_tcp_default_timeout;
3164 
3165 	/* backchannel */
3166 	xprt_set_bound(xprt);
3167 	xprt->bind_timeout = 0;
3168 	xprt->reestablish_timeout = 0;
3169 	xprt->idle_timeout = 0;
3170 
3171 	xprt->ops = &bc_tcp_ops;
3172 
3173 	switch (addr->sa_family) {
3174 	case AF_INET:
3175 		xs_format_peer_addresses(xprt, "tcp",
3176 					 RPCBIND_NETID_TCP);
3177 		break;
3178 	case AF_INET6:
3179 		xs_format_peer_addresses(xprt, "tcp",
3180 				   RPCBIND_NETID_TCP6);
3181 		break;
3182 	default:
3183 		ret = ERR_PTR(-EAFNOSUPPORT);
3184 		goto out_err;
3185 	}
3186 
3187 	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3188 			xprt->address_strings[RPC_DISPLAY_ADDR],
3189 			xprt->address_strings[RPC_DISPLAY_PORT],
3190 			xprt->address_strings[RPC_DISPLAY_PROTO]);
3191 
3192 	/*
3193 	 * Once we've associated a backchannel xprt with a connection,
3194 	 * we want to keep it around as long as the connection lasts,
3195 	 * in case we need to start using it for a backchannel again;
3196 	 * this reference won't be dropped until bc_xprt is destroyed.
3197 	 */
3198 	xprt_get(xprt);
3199 	args->bc_xprt->xpt_bc_xprt = xprt;
3200 	xprt->bc_xprt = args->bc_xprt;
3201 	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3202 	transport->sock = bc_sock->sk_sock;
3203 	transport->inet = bc_sock->sk_sk;
3204 
3205 	/*
3206 	 * Since we don't want connections for the backchannel, we set
3207 	 * the xprt status to connected
3208 	 */
3209 	xprt_set_connected(xprt);
3210 
3211 	if (try_module_get(THIS_MODULE))
3212 		return xprt;
3213 
3214 	args->bc_xprt->xpt_bc_xprt = NULL;
3215 	args->bc_xprt->xpt_bc_xps = NULL;
3216 	xprt_put(xprt);
3217 	ret = ERR_PTR(-EINVAL);
3218 out_err:
3219 	xs_xprt_free(xprt);
3220 	return ret;
3221 }
3222 
3223 static struct xprt_class	xs_local_transport = {
3224 	.list		= LIST_HEAD_INIT(xs_local_transport.list),
3225 	.name		= "named UNIX socket",
3226 	.owner		= THIS_MODULE,
3227 	.ident		= XPRT_TRANSPORT_LOCAL,
3228 	.setup		= xs_setup_local,
3229 };
3230 
3231 static struct xprt_class	xs_udp_transport = {
3232 	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
3233 	.name		= "udp",
3234 	.owner		= THIS_MODULE,
3235 	.ident		= XPRT_TRANSPORT_UDP,
3236 	.setup		= xs_setup_udp,
3237 };
3238 
3239 static struct xprt_class	xs_tcp_transport = {
3240 	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
3241 	.name		= "tcp",
3242 	.owner		= THIS_MODULE,
3243 	.ident		= XPRT_TRANSPORT_TCP,
3244 	.setup		= xs_setup_tcp,
3245 };
3246 
3247 static struct xprt_class	xs_bc_tcp_transport = {
3248 	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3249 	.name		= "tcp NFSv4.1 backchannel",
3250 	.owner		= THIS_MODULE,
3251 	.ident		= XPRT_TRANSPORT_BC_TCP,
3252 	.setup		= xs_setup_bc_tcp,
3253 };
3254 
3255 /**
3256  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3257  *
3258  */
3259 int init_socket_xprt(void)
3260 {
3261 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3262 	if (!sunrpc_table_header)
3263 		sunrpc_table_header = register_sysctl_table(sunrpc_table);
3264 #endif
3265 
3266 	xprt_register_transport(&xs_local_transport);
3267 	xprt_register_transport(&xs_udp_transport);
3268 	xprt_register_transport(&xs_tcp_transport);
3269 	xprt_register_transport(&xs_bc_tcp_transport);
3270 
3271 	return 0;
3272 }
3273 
3274 /**
3275  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3276  *
3277  */
3278 void cleanup_socket_xprt(void)
3279 {
3280 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
3281 	if (sunrpc_table_header) {
3282 		unregister_sysctl_table(sunrpc_table_header);
3283 		sunrpc_table_header = NULL;
3284 	}
3285 #endif
3286 
3287 	xprt_unregister_transport(&xs_local_transport);
3288 	xprt_unregister_transport(&xs_udp_transport);
3289 	xprt_unregister_transport(&xs_tcp_transport);
3290 	xprt_unregister_transport(&xs_bc_tcp_transport);
3291 }
3292 
3293 static int param_set_uint_minmax(const char *val,
3294 		const struct kernel_param *kp,
3295 		unsigned int min, unsigned int max)
3296 {
3297 	unsigned int num;
3298 	int ret;
3299 
3300 	if (!val)
3301 		return -EINVAL;
3302 	ret = kstrtouint(val, 0, &num);
3303 	if (ret)
3304 		return ret;
3305 	if (num < min || num > max)
3306 		return -EINVAL;
3307 	*((unsigned int *)kp->arg) = num;
3308 	return 0;
3309 }
3310 
3311 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3312 {
3313 	if (kp->arg == &xprt_min_resvport)
3314 		return param_set_uint_minmax(val, kp,
3315 			RPC_MIN_RESVPORT,
3316 			xprt_max_resvport);
3317 	return param_set_uint_minmax(val, kp,
3318 			xprt_min_resvport,
3319 			RPC_MAX_RESVPORT);
3320 }
3321 
3322 static const struct kernel_param_ops param_ops_portnr = {
3323 	.set = param_set_portnr,
3324 	.get = param_get_uint,
3325 };
3326 
3327 #define param_check_portnr(name, p) \
3328 	__param_check(name, p, unsigned int);
3329 
3330 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3331 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3332 
3333 static int param_set_slot_table_size(const char *val,
3334 				     const struct kernel_param *kp)
3335 {
3336 	return param_set_uint_minmax(val, kp,
3337 			RPC_MIN_SLOT_TABLE,
3338 			RPC_MAX_SLOT_TABLE);
3339 }
3340 
3341 static const struct kernel_param_ops param_ops_slot_table_size = {
3342 	.set = param_set_slot_table_size,
3343 	.get = param_get_uint,
3344 };
3345 
3346 #define param_check_slot_table_size(name, p) \
3347 	__param_check(name, p, unsigned int);
3348 
3349 static int param_set_max_slot_table_size(const char *val,
3350 				     const struct kernel_param *kp)
3351 {
3352 	return param_set_uint_minmax(val, kp,
3353 			RPC_MIN_SLOT_TABLE,
3354 			RPC_MAX_SLOT_TABLE_LIMIT);
3355 }
3356 
3357 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3358 	.set = param_set_max_slot_table_size,
3359 	.get = param_get_uint,
3360 };
3361 
3362 #define param_check_max_slot_table_size(name, p) \
3363 	__param_check(name, p, unsigned int);
3364 
3365 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3366 		   slot_table_size, 0644);
3367 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3368 		   max_slot_table_size, 0644);
3369 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3370 		   slot_table_size, 0644);
3371 
3372