xref: /linux/net/sunrpc/xprtsock.c (revision 08ec212c0f92cbf30e3ecc7349f18151714041d6)
1 /*
2  * linux/net/sunrpc/xprtsock.c
3  *
4  * Client-side transport implementation for sockets.
5  *
6  * TCP callback races fixes (C) 1998 Red Hat
7  * TCP send fixes (C) 1998 Red Hat
8  * TCP NFS related read + write fixes
9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10  *
11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
12  * Fix behaviour when socket buffer is full.
13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14  *
15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16  *
17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
18  *   <gilles.quillard@bull.net>
19  */
20 
21 #include <linux/types.h>
22 #include <linux/string.h>
23 #include <linux/slab.h>
24 #include <linux/module.h>
25 #include <linux/capability.h>
26 #include <linux/pagemap.h>
27 #include <linux/errno.h>
28 #include <linux/socket.h>
29 #include <linux/in.h>
30 #include <linux/net.h>
31 #include <linux/mm.h>
32 #include <linux/un.h>
33 #include <linux/udp.h>
34 #include <linux/tcp.h>
35 #include <linux/sunrpc/clnt.h>
36 #include <linux/sunrpc/sched.h>
37 #include <linux/sunrpc/svcsock.h>
38 #include <linux/sunrpc/xprtsock.h>
39 #include <linux/file.h>
40 #ifdef CONFIG_SUNRPC_BACKCHANNEL
41 #include <linux/sunrpc/bc_xprt.h>
42 #endif
43 
44 #include <net/sock.h>
45 #include <net/checksum.h>
46 #include <net/udp.h>
47 #include <net/tcp.h>
48 
49 #include "sunrpc.h"
50 
51 static void xs_close(struct rpc_xprt *xprt);
52 
53 /*
54  * xprtsock tunables
55  */
56 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
57 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
58 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
59 
60 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
61 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
62 
63 #define XS_TCP_LINGER_TO	(15U * HZ)
64 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
65 
66 /*
67  * We can register our own files under /proc/sys/sunrpc by
68  * calling register_sysctl_table() again.  The files in that
69  * directory become the union of all files registered there.
70  *
71  * We simply need to make sure that we don't collide with
72  * someone else's file names!
73  */
74 
75 #ifdef RPC_DEBUG
76 
77 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
78 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
79 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
80 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
81 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
82 
83 static struct ctl_table_header *sunrpc_table_header;
84 
85 /*
86  * FIXME: changing the UDP slot table size should also resize the UDP
87  *        socket buffers for existing UDP transports
88  */
89 static ctl_table xs_tunables_table[] = {
90 	{
91 		.procname	= "udp_slot_table_entries",
92 		.data		= &xprt_udp_slot_table_entries,
93 		.maxlen		= sizeof(unsigned int),
94 		.mode		= 0644,
95 		.proc_handler	= proc_dointvec_minmax,
96 		.extra1		= &min_slot_table_size,
97 		.extra2		= &max_slot_table_size
98 	},
99 	{
100 		.procname	= "tcp_slot_table_entries",
101 		.data		= &xprt_tcp_slot_table_entries,
102 		.maxlen		= sizeof(unsigned int),
103 		.mode		= 0644,
104 		.proc_handler	= proc_dointvec_minmax,
105 		.extra1		= &min_slot_table_size,
106 		.extra2		= &max_slot_table_size
107 	},
108 	{
109 		.procname	= "tcp_max_slot_table_entries",
110 		.data		= &xprt_max_tcp_slot_table_entries,
111 		.maxlen		= sizeof(unsigned int),
112 		.mode		= 0644,
113 		.proc_handler	= proc_dointvec_minmax,
114 		.extra1		= &min_slot_table_size,
115 		.extra2		= &max_tcp_slot_table_limit
116 	},
117 	{
118 		.procname	= "min_resvport",
119 		.data		= &xprt_min_resvport,
120 		.maxlen		= sizeof(unsigned int),
121 		.mode		= 0644,
122 		.proc_handler	= proc_dointvec_minmax,
123 		.extra1		= &xprt_min_resvport_limit,
124 		.extra2		= &xprt_max_resvport_limit
125 	},
126 	{
127 		.procname	= "max_resvport",
128 		.data		= &xprt_max_resvport,
129 		.maxlen		= sizeof(unsigned int),
130 		.mode		= 0644,
131 		.proc_handler	= proc_dointvec_minmax,
132 		.extra1		= &xprt_min_resvport_limit,
133 		.extra2		= &xprt_max_resvport_limit
134 	},
135 	{
136 		.procname	= "tcp_fin_timeout",
137 		.data		= &xs_tcp_fin_timeout,
138 		.maxlen		= sizeof(xs_tcp_fin_timeout),
139 		.mode		= 0644,
140 		.proc_handler	= proc_dointvec_jiffies,
141 	},
142 	{ },
143 };
144 
145 static ctl_table sunrpc_table[] = {
146 	{
147 		.procname	= "sunrpc",
148 		.mode		= 0555,
149 		.child		= xs_tunables_table
150 	},
151 	{ },
152 };
153 
154 #endif
155 
156 /*
157  * Wait duration for a reply from the RPC portmapper.
158  */
159 #define XS_BIND_TO		(60U * HZ)
160 
161 /*
162  * Delay if a UDP socket connect error occurs.  This is most likely some
163  * kind of resource problem on the local host.
164  */
165 #define XS_UDP_REEST_TO		(2U * HZ)
166 
167 /*
168  * The reestablish timeout allows clients to delay for a bit before attempting
169  * to reconnect to a server that just dropped our connection.
170  *
171  * We implement an exponential backoff when trying to reestablish a TCP
172  * transport connection with the server.  Some servers like to drop a TCP
173  * connection when they are overworked, so we start with a short timeout and
174  * increase over time if the server is down or not responding.
175  */
176 #define XS_TCP_INIT_REEST_TO	(3U * HZ)
177 #define XS_TCP_MAX_REEST_TO	(5U * 60 * HZ)
178 
179 /*
180  * TCP idle timeout; client drops the transport socket if it is idle
181  * for this long.  Note that we also timeout UDP sockets to prevent
182  * holding port numbers when there is no RPC traffic.
183  */
184 #define XS_IDLE_DISC_TO		(5U * 60 * HZ)
185 
186 #ifdef RPC_DEBUG
187 # undef  RPC_DEBUG_DATA
188 # define RPCDBG_FACILITY	RPCDBG_TRANS
189 #endif
190 
191 #ifdef RPC_DEBUG_DATA
192 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
193 {
194 	u8 *buf = (u8 *) packet;
195 	int j;
196 
197 	dprintk("RPC:       %s\n", msg);
198 	for (j = 0; j < count && j < 128; j += 4) {
199 		if (!(j & 31)) {
200 			if (j)
201 				dprintk("\n");
202 			dprintk("0x%04x ", j);
203 		}
204 		dprintk("%02x%02x%02x%02x ",
205 			buf[j], buf[j+1], buf[j+2], buf[j+3]);
206 	}
207 	dprintk("\n");
208 }
209 #else
210 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
211 {
212 	/* NOP */
213 }
214 #endif
215 
216 struct sock_xprt {
217 	struct rpc_xprt		xprt;
218 
219 	/*
220 	 * Network layer
221 	 */
222 	struct socket *		sock;
223 	struct sock *		inet;
224 
225 	/*
226 	 * State of TCP reply receive
227 	 */
228 	__be32			tcp_fraghdr,
229 				tcp_xid,
230 				tcp_calldir;
231 
232 	u32			tcp_offset,
233 				tcp_reclen;
234 
235 	unsigned long		tcp_copied,
236 				tcp_flags;
237 
238 	/*
239 	 * Connection of transports
240 	 */
241 	struct delayed_work	connect_worker;
242 	struct sockaddr_storage	srcaddr;
243 	unsigned short		srcport;
244 
245 	/*
246 	 * UDP socket buffer size parameters
247 	 */
248 	size_t			rcvsize,
249 				sndsize;
250 
251 	/*
252 	 * Saved socket callback addresses
253 	 */
254 	void			(*old_data_ready)(struct sock *, int);
255 	void			(*old_state_change)(struct sock *);
256 	void			(*old_write_space)(struct sock *);
257 	void			(*old_error_report)(struct sock *);
258 };
259 
260 /*
261  * TCP receive state flags
262  */
263 #define TCP_RCV_LAST_FRAG	(1UL << 0)
264 #define TCP_RCV_COPY_FRAGHDR	(1UL << 1)
265 #define TCP_RCV_COPY_XID	(1UL << 2)
266 #define TCP_RCV_COPY_DATA	(1UL << 3)
267 #define TCP_RCV_READ_CALLDIR	(1UL << 4)
268 #define TCP_RCV_COPY_CALLDIR	(1UL << 5)
269 
270 /*
271  * TCP RPC flags
272  */
273 #define TCP_RPC_REPLY		(1UL << 6)
274 
275 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
276 {
277 	return (struct sockaddr *) &xprt->addr;
278 }
279 
280 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
281 {
282 	return (struct sockaddr_un *) &xprt->addr;
283 }
284 
285 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
286 {
287 	return (struct sockaddr_in *) &xprt->addr;
288 }
289 
290 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
291 {
292 	return (struct sockaddr_in6 *) &xprt->addr;
293 }
294 
295 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
296 {
297 	struct sockaddr *sap = xs_addr(xprt);
298 	struct sockaddr_in6 *sin6;
299 	struct sockaddr_in *sin;
300 	struct sockaddr_un *sun;
301 	char buf[128];
302 
303 	switch (sap->sa_family) {
304 	case AF_LOCAL:
305 		sun = xs_addr_un(xprt);
306 		strlcpy(buf, sun->sun_path, sizeof(buf));
307 		xprt->address_strings[RPC_DISPLAY_ADDR] =
308 						kstrdup(buf, GFP_KERNEL);
309 		break;
310 	case AF_INET:
311 		(void)rpc_ntop(sap, buf, sizeof(buf));
312 		xprt->address_strings[RPC_DISPLAY_ADDR] =
313 						kstrdup(buf, GFP_KERNEL);
314 		sin = xs_addr_in(xprt);
315 		snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
316 		break;
317 	case AF_INET6:
318 		(void)rpc_ntop(sap, buf, sizeof(buf));
319 		xprt->address_strings[RPC_DISPLAY_ADDR] =
320 						kstrdup(buf, GFP_KERNEL);
321 		sin6 = xs_addr_in6(xprt);
322 		snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
323 		break;
324 	default:
325 		BUG();
326 	}
327 
328 	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
329 }
330 
331 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
332 {
333 	struct sockaddr *sap = xs_addr(xprt);
334 	char buf[128];
335 
336 	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
337 	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
338 
339 	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
340 	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
341 }
342 
343 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
344 				     const char *protocol,
345 				     const char *netid)
346 {
347 	xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
348 	xprt->address_strings[RPC_DISPLAY_NETID] = netid;
349 	xs_format_common_peer_addresses(xprt);
350 	xs_format_common_peer_ports(xprt);
351 }
352 
353 static void xs_update_peer_port(struct rpc_xprt *xprt)
354 {
355 	kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
356 	kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
357 
358 	xs_format_common_peer_ports(xprt);
359 }
360 
361 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
362 {
363 	unsigned int i;
364 
365 	for (i = 0; i < RPC_DISPLAY_MAX; i++)
366 		switch (i) {
367 		case RPC_DISPLAY_PROTO:
368 		case RPC_DISPLAY_NETID:
369 			continue;
370 		default:
371 			kfree(xprt->address_strings[i]);
372 		}
373 }
374 
375 #define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
376 
377 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
378 {
379 	struct msghdr msg = {
380 		.msg_name	= addr,
381 		.msg_namelen	= addrlen,
382 		.msg_flags	= XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
383 	};
384 	struct kvec iov = {
385 		.iov_base	= vec->iov_base + base,
386 		.iov_len	= vec->iov_len - base,
387 	};
388 
389 	if (iov.iov_len != 0)
390 		return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
391 	return kernel_sendmsg(sock, &msg, NULL, 0, 0);
392 }
393 
394 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
395 {
396 	struct page **ppage;
397 	unsigned int remainder;
398 	int err, sent = 0;
399 
400 	remainder = xdr->page_len - base;
401 	base += xdr->page_base;
402 	ppage = xdr->pages + (base >> PAGE_SHIFT);
403 	base &= ~PAGE_MASK;
404 	for(;;) {
405 		unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
406 		int flags = XS_SENDMSG_FLAGS;
407 
408 		remainder -= len;
409 		if (remainder != 0 || more)
410 			flags |= MSG_MORE;
411 		err = sock->ops->sendpage(sock, *ppage, base, len, flags);
412 		if (remainder == 0 || err != len)
413 			break;
414 		sent += err;
415 		ppage++;
416 		base = 0;
417 	}
418 	if (sent == 0)
419 		return err;
420 	if (err > 0)
421 		sent += err;
422 	return sent;
423 }
424 
425 /**
426  * xs_sendpages - write pages directly to a socket
427  * @sock: socket to send on
428  * @addr: UDP only -- address of destination
429  * @addrlen: UDP only -- length of destination address
430  * @xdr: buffer containing this request
431  * @base: starting position in the buffer
432  *
433  */
434 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
435 {
436 	unsigned int remainder = xdr->len - base;
437 	int err, sent = 0;
438 
439 	if (unlikely(!sock))
440 		return -ENOTSOCK;
441 
442 	clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
443 	if (base != 0) {
444 		addr = NULL;
445 		addrlen = 0;
446 	}
447 
448 	if (base < xdr->head[0].iov_len || addr != NULL) {
449 		unsigned int len = xdr->head[0].iov_len - base;
450 		remainder -= len;
451 		err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
452 		if (remainder == 0 || err != len)
453 			goto out;
454 		sent += err;
455 		base = 0;
456 	} else
457 		base -= xdr->head[0].iov_len;
458 
459 	if (base < xdr->page_len) {
460 		unsigned int len = xdr->page_len - base;
461 		remainder -= len;
462 		err = xs_send_pagedata(sock, xdr, base, remainder != 0);
463 		if (remainder == 0 || err != len)
464 			goto out;
465 		sent += err;
466 		base = 0;
467 	} else
468 		base -= xdr->page_len;
469 
470 	if (base >= xdr->tail[0].iov_len)
471 		return sent;
472 	err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
473 out:
474 	if (sent == 0)
475 		return err;
476 	if (err > 0)
477 		sent += err;
478 	return sent;
479 }
480 
481 static void xs_nospace_callback(struct rpc_task *task)
482 {
483 	struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
484 
485 	transport->inet->sk_write_pending--;
486 	clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
487 }
488 
489 /**
490  * xs_nospace - place task on wait queue if transmit was incomplete
491  * @task: task to put to sleep
492  *
493  */
494 static int xs_nospace(struct rpc_task *task)
495 {
496 	struct rpc_rqst *req = task->tk_rqstp;
497 	struct rpc_xprt *xprt = req->rq_xprt;
498 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
499 	int ret = -EAGAIN;
500 
501 	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
502 			task->tk_pid, req->rq_slen - req->rq_bytes_sent,
503 			req->rq_slen);
504 
505 	/* Protect against races with write_space */
506 	spin_lock_bh(&xprt->transport_lock);
507 
508 	/* Don't race with disconnect */
509 	if (xprt_connected(xprt)) {
510 		if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
511 			/*
512 			 * Notify TCP that we're limited by the application
513 			 * window size
514 			 */
515 			set_bit(SOCK_NOSPACE, &transport->sock->flags);
516 			transport->inet->sk_write_pending++;
517 			/* ...and wait for more buffer space */
518 			xprt_wait_for_buffer_space(task, xs_nospace_callback);
519 		}
520 	} else {
521 		clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
522 		ret = -ENOTCONN;
523 	}
524 
525 	spin_unlock_bh(&xprt->transport_lock);
526 	return ret;
527 }
528 
529 /*
530  * Construct a stream transport record marker in @buf.
531  */
532 static inline void xs_encode_stream_record_marker(struct xdr_buf *buf)
533 {
534 	u32 reclen = buf->len - sizeof(rpc_fraghdr);
535 	rpc_fraghdr *base = buf->head[0].iov_base;
536 	*base = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | reclen);
537 }
538 
539 /**
540  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
541  * @task: RPC task that manages the state of an RPC request
542  *
543  * Return values:
544  *        0:	The request has been sent
545  *   EAGAIN:	The socket was blocked, please call again later to
546  *		complete the request
547  * ENOTCONN:	Caller needs to invoke connect logic then call again
548  *    other:	Some other error occured, the request was not sent
549  */
550 static int xs_local_send_request(struct rpc_task *task)
551 {
552 	struct rpc_rqst *req = task->tk_rqstp;
553 	struct rpc_xprt *xprt = req->rq_xprt;
554 	struct sock_xprt *transport =
555 				container_of(xprt, struct sock_xprt, xprt);
556 	struct xdr_buf *xdr = &req->rq_snd_buf;
557 	int status;
558 
559 	xs_encode_stream_record_marker(&req->rq_snd_buf);
560 
561 	xs_pktdump("packet data:",
562 			req->rq_svec->iov_base, req->rq_svec->iov_len);
563 
564 	status = xs_sendpages(transport->sock, NULL, 0,
565 						xdr, req->rq_bytes_sent);
566 	dprintk("RPC:       %s(%u) = %d\n",
567 			__func__, xdr->len - req->rq_bytes_sent, status);
568 	if (likely(status >= 0)) {
569 		req->rq_bytes_sent += status;
570 		req->rq_xmit_bytes_sent += status;
571 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
572 			req->rq_bytes_sent = 0;
573 			return 0;
574 		}
575 		status = -EAGAIN;
576 	}
577 
578 	switch (status) {
579 	case -EAGAIN:
580 		status = xs_nospace(task);
581 		break;
582 	default:
583 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
584 			-status);
585 	case -EPIPE:
586 		xs_close(xprt);
587 		status = -ENOTCONN;
588 	}
589 
590 	return status;
591 }
592 
593 /**
594  * xs_udp_send_request - write an RPC request to a UDP socket
595  * @task: address of RPC task that manages the state of an RPC request
596  *
597  * Return values:
598  *        0:	The request has been sent
599  *   EAGAIN:	The socket was blocked, please call again later to
600  *		complete the request
601  * ENOTCONN:	Caller needs to invoke connect logic then call again
602  *    other:	Some other error occurred, the request was not sent
603  */
604 static int xs_udp_send_request(struct rpc_task *task)
605 {
606 	struct rpc_rqst *req = task->tk_rqstp;
607 	struct rpc_xprt *xprt = req->rq_xprt;
608 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
609 	struct xdr_buf *xdr = &req->rq_snd_buf;
610 	int status;
611 
612 	xs_pktdump("packet data:",
613 				req->rq_svec->iov_base,
614 				req->rq_svec->iov_len);
615 
616 	if (!xprt_bound(xprt))
617 		return -ENOTCONN;
618 	status = xs_sendpages(transport->sock,
619 			      xs_addr(xprt),
620 			      xprt->addrlen, xdr,
621 			      req->rq_bytes_sent);
622 
623 	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
624 			xdr->len - req->rq_bytes_sent, status);
625 
626 	if (status >= 0) {
627 		req->rq_xmit_bytes_sent += status;
628 		if (status >= req->rq_slen)
629 			return 0;
630 		/* Still some bytes left; set up for a retry later. */
631 		status = -EAGAIN;
632 	}
633 
634 	switch (status) {
635 	case -ENOTSOCK:
636 		status = -ENOTCONN;
637 		/* Should we call xs_close() here? */
638 		break;
639 	case -EAGAIN:
640 		status = xs_nospace(task);
641 		break;
642 	default:
643 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
644 			-status);
645 	case -ENETUNREACH:
646 	case -EPIPE:
647 	case -ECONNREFUSED:
648 		/* When the server has died, an ICMP port unreachable message
649 		 * prompts ECONNREFUSED. */
650 		clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
651 	}
652 
653 	return status;
654 }
655 
656 /**
657  * xs_tcp_shutdown - gracefully shut down a TCP socket
658  * @xprt: transport
659  *
660  * Initiates a graceful shutdown of the TCP socket by calling the
661  * equivalent of shutdown(SHUT_WR);
662  */
663 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
664 {
665 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
666 	struct socket *sock = transport->sock;
667 
668 	if (sock != NULL)
669 		kernel_sock_shutdown(sock, SHUT_WR);
670 }
671 
672 /**
673  * xs_tcp_send_request - write an RPC request to a TCP socket
674  * @task: address of RPC task that manages the state of an RPC request
675  *
676  * Return values:
677  *        0:	The request has been sent
678  *   EAGAIN:	The socket was blocked, please call again later to
679  *		complete the request
680  * ENOTCONN:	Caller needs to invoke connect logic then call again
681  *    other:	Some other error occurred, the request was not sent
682  *
683  * XXX: In the case of soft timeouts, should we eventually give up
684  *	if sendmsg is not able to make progress?
685  */
686 static int xs_tcp_send_request(struct rpc_task *task)
687 {
688 	struct rpc_rqst *req = task->tk_rqstp;
689 	struct rpc_xprt *xprt = req->rq_xprt;
690 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
691 	struct xdr_buf *xdr = &req->rq_snd_buf;
692 	int status;
693 
694 	xs_encode_stream_record_marker(&req->rq_snd_buf);
695 
696 	xs_pktdump("packet data:",
697 				req->rq_svec->iov_base,
698 				req->rq_svec->iov_len);
699 
700 	/* Continue transmitting the packet/record. We must be careful
701 	 * to cope with writespace callbacks arriving _after_ we have
702 	 * called sendmsg(). */
703 	while (1) {
704 		status = xs_sendpages(transport->sock,
705 					NULL, 0, xdr, req->rq_bytes_sent);
706 
707 		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
708 				xdr->len - req->rq_bytes_sent, status);
709 
710 		if (unlikely(status < 0))
711 			break;
712 
713 		/* If we've sent the entire packet, immediately
714 		 * reset the count of bytes sent. */
715 		req->rq_bytes_sent += status;
716 		req->rq_xmit_bytes_sent += status;
717 		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
718 			req->rq_bytes_sent = 0;
719 			return 0;
720 		}
721 
722 		if (status != 0)
723 			continue;
724 		status = -EAGAIN;
725 		break;
726 	}
727 
728 	switch (status) {
729 	case -ENOTSOCK:
730 		status = -ENOTCONN;
731 		/* Should we call xs_close() here? */
732 		break;
733 	case -EAGAIN:
734 		status = xs_nospace(task);
735 		break;
736 	default:
737 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
738 			-status);
739 	case -ECONNRESET:
740 	case -EPIPE:
741 		xs_tcp_shutdown(xprt);
742 	case -ECONNREFUSED:
743 	case -ENOTCONN:
744 		clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
745 	}
746 
747 	return status;
748 }
749 
750 /**
751  * xs_tcp_release_xprt - clean up after a tcp transmission
752  * @xprt: transport
753  * @task: rpc task
754  *
755  * This cleans up if an error causes us to abort the transmission of a request.
756  * In this case, the socket may need to be reset in order to avoid confusing
757  * the server.
758  */
759 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
760 {
761 	struct rpc_rqst *req;
762 
763 	if (task != xprt->snd_task)
764 		return;
765 	if (task == NULL)
766 		goto out_release;
767 	req = task->tk_rqstp;
768 	if (req == NULL)
769 		goto out_release;
770 	if (req->rq_bytes_sent == 0)
771 		goto out_release;
772 	if (req->rq_bytes_sent == req->rq_snd_buf.len)
773 		goto out_release;
774 	set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
775 out_release:
776 	xprt_release_xprt(xprt, task);
777 }
778 
779 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
780 {
781 	transport->old_data_ready = sk->sk_data_ready;
782 	transport->old_state_change = sk->sk_state_change;
783 	transport->old_write_space = sk->sk_write_space;
784 	transport->old_error_report = sk->sk_error_report;
785 }
786 
787 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
788 {
789 	sk->sk_data_ready = transport->old_data_ready;
790 	sk->sk_state_change = transport->old_state_change;
791 	sk->sk_write_space = transport->old_write_space;
792 	sk->sk_error_report = transport->old_error_report;
793 }
794 
795 static void xs_reset_transport(struct sock_xprt *transport)
796 {
797 	struct socket *sock = transport->sock;
798 	struct sock *sk = transport->inet;
799 
800 	if (sk == NULL)
801 		return;
802 
803 	transport->srcport = 0;
804 
805 	write_lock_bh(&sk->sk_callback_lock);
806 	transport->inet = NULL;
807 	transport->sock = NULL;
808 
809 	sk->sk_user_data = NULL;
810 
811 	xs_restore_old_callbacks(transport, sk);
812 	write_unlock_bh(&sk->sk_callback_lock);
813 
814 	sk->sk_no_check = 0;
815 
816 	sock_release(sock);
817 }
818 
819 /**
820  * xs_close - close a socket
821  * @xprt: transport
822  *
823  * This is used when all requests are complete; ie, no DRC state remains
824  * on the server we want to save.
825  *
826  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
827  * xs_reset_transport() zeroing the socket from underneath a writer.
828  */
829 static void xs_close(struct rpc_xprt *xprt)
830 {
831 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
832 
833 	dprintk("RPC:       xs_close xprt %p\n", xprt);
834 
835 	xs_reset_transport(transport);
836 	xprt->reestablish_timeout = 0;
837 
838 	smp_mb__before_clear_bit();
839 	clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
840 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
841 	clear_bit(XPRT_CLOSING, &xprt->state);
842 	smp_mb__after_clear_bit();
843 	xprt_disconnect_done(xprt);
844 }
845 
846 static void xs_tcp_close(struct rpc_xprt *xprt)
847 {
848 	if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
849 		xs_close(xprt);
850 	else
851 		xs_tcp_shutdown(xprt);
852 }
853 
854 /**
855  * xs_destroy - prepare to shutdown a transport
856  * @xprt: doomed transport
857  *
858  */
859 static void xs_destroy(struct rpc_xprt *xprt)
860 {
861 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
862 
863 	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
864 
865 	cancel_delayed_work_sync(&transport->connect_worker);
866 
867 	xs_close(xprt);
868 	xs_free_peer_addresses(xprt);
869 	xprt_free(xprt);
870 	module_put(THIS_MODULE);
871 }
872 
873 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
874 {
875 	return (struct rpc_xprt *) sk->sk_user_data;
876 }
877 
878 static int xs_local_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
879 {
880 	struct xdr_skb_reader desc = {
881 		.skb		= skb,
882 		.offset		= sizeof(rpc_fraghdr),
883 		.count		= skb->len - sizeof(rpc_fraghdr),
884 	};
885 
886 	if (xdr_partial_copy_from_skb(xdr, 0, &desc, xdr_skb_read_bits) < 0)
887 		return -1;
888 	if (desc.count)
889 		return -1;
890 	return 0;
891 }
892 
893 /**
894  * xs_local_data_ready - "data ready" callback for AF_LOCAL sockets
895  * @sk: socket with data to read
896  * @len: how much data to read
897  *
898  * Currently this assumes we can read the whole reply in a single gulp.
899  */
900 static void xs_local_data_ready(struct sock *sk, int len)
901 {
902 	struct rpc_task *task;
903 	struct rpc_xprt *xprt;
904 	struct rpc_rqst *rovr;
905 	struct sk_buff *skb;
906 	int err, repsize, copied;
907 	u32 _xid;
908 	__be32 *xp;
909 
910 	read_lock_bh(&sk->sk_callback_lock);
911 	dprintk("RPC:       %s...\n", __func__);
912 	xprt = xprt_from_sock(sk);
913 	if (xprt == NULL)
914 		goto out;
915 
916 	skb = skb_recv_datagram(sk, 0, 1, &err);
917 	if (skb == NULL)
918 		goto out;
919 
920 	repsize = skb->len - sizeof(rpc_fraghdr);
921 	if (repsize < 4) {
922 		dprintk("RPC:       impossible RPC reply size %d\n", repsize);
923 		goto dropit;
924 	}
925 
926 	/* Copy the XID from the skb... */
927 	xp = skb_header_pointer(skb, sizeof(rpc_fraghdr), sizeof(_xid), &_xid);
928 	if (xp == NULL)
929 		goto dropit;
930 
931 	/* Look up and lock the request corresponding to the given XID */
932 	spin_lock(&xprt->transport_lock);
933 	rovr = xprt_lookup_rqst(xprt, *xp);
934 	if (!rovr)
935 		goto out_unlock;
936 	task = rovr->rq_task;
937 
938 	copied = rovr->rq_private_buf.buflen;
939 	if (copied > repsize)
940 		copied = repsize;
941 
942 	if (xs_local_copy_to_xdr(&rovr->rq_private_buf, skb)) {
943 		dprintk("RPC:       sk_buff copy failed\n");
944 		goto out_unlock;
945 	}
946 
947 	xprt_complete_rqst(task, copied);
948 
949  out_unlock:
950 	spin_unlock(&xprt->transport_lock);
951  dropit:
952 	skb_free_datagram(sk, skb);
953  out:
954 	read_unlock_bh(&sk->sk_callback_lock);
955 }
956 
957 /**
958  * xs_udp_data_ready - "data ready" callback for UDP sockets
959  * @sk: socket with data to read
960  * @len: how much data to read
961  *
962  */
963 static void xs_udp_data_ready(struct sock *sk, int len)
964 {
965 	struct rpc_task *task;
966 	struct rpc_xprt *xprt;
967 	struct rpc_rqst *rovr;
968 	struct sk_buff *skb;
969 	int err, repsize, copied;
970 	u32 _xid;
971 	__be32 *xp;
972 
973 	read_lock_bh(&sk->sk_callback_lock);
974 	dprintk("RPC:       xs_udp_data_ready...\n");
975 	if (!(xprt = xprt_from_sock(sk)))
976 		goto out;
977 
978 	if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
979 		goto out;
980 
981 	repsize = skb->len - sizeof(struct udphdr);
982 	if (repsize < 4) {
983 		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
984 		goto dropit;
985 	}
986 
987 	/* Copy the XID from the skb... */
988 	xp = skb_header_pointer(skb, sizeof(struct udphdr),
989 				sizeof(_xid), &_xid);
990 	if (xp == NULL)
991 		goto dropit;
992 
993 	/* Look up and lock the request corresponding to the given XID */
994 	spin_lock(&xprt->transport_lock);
995 	rovr = xprt_lookup_rqst(xprt, *xp);
996 	if (!rovr)
997 		goto out_unlock;
998 	task = rovr->rq_task;
999 
1000 	if ((copied = rovr->rq_private_buf.buflen) > repsize)
1001 		copied = repsize;
1002 
1003 	/* Suck it into the iovec, verify checksum if not done by hw. */
1004 	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1005 		UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
1006 		goto out_unlock;
1007 	}
1008 
1009 	UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
1010 
1011 	xprt_adjust_cwnd(task, copied);
1012 	xprt_complete_rqst(task, copied);
1013 
1014  out_unlock:
1015 	spin_unlock(&xprt->transport_lock);
1016  dropit:
1017 	skb_free_datagram(sk, skb);
1018  out:
1019 	read_unlock_bh(&sk->sk_callback_lock);
1020 }
1021 
1022 /*
1023  * Helper function to force a TCP close if the server is sending
1024  * junk and/or it has put us in CLOSE_WAIT
1025  */
1026 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1027 {
1028 	set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1029 	xprt_force_disconnect(xprt);
1030 }
1031 
1032 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
1033 {
1034 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1035 	size_t len, used;
1036 	char *p;
1037 
1038 	p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
1039 	len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
1040 	used = xdr_skb_read_bits(desc, p, len);
1041 	transport->tcp_offset += used;
1042 	if (used != len)
1043 		return;
1044 
1045 	transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
1046 	if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
1047 		transport->tcp_flags |= TCP_RCV_LAST_FRAG;
1048 	else
1049 		transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
1050 	transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
1051 
1052 	transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
1053 	transport->tcp_offset = 0;
1054 
1055 	/* Sanity check of the record length */
1056 	if (unlikely(transport->tcp_reclen < 8)) {
1057 		dprintk("RPC:       invalid TCP record fragment length\n");
1058 		xs_tcp_force_close(xprt);
1059 		return;
1060 	}
1061 	dprintk("RPC:       reading TCP record fragment of length %d\n",
1062 			transport->tcp_reclen);
1063 }
1064 
1065 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
1066 {
1067 	if (transport->tcp_offset == transport->tcp_reclen) {
1068 		transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
1069 		transport->tcp_offset = 0;
1070 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
1071 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1072 			transport->tcp_flags |= TCP_RCV_COPY_XID;
1073 			transport->tcp_copied = 0;
1074 		}
1075 	}
1076 }
1077 
1078 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1079 {
1080 	size_t len, used;
1081 	char *p;
1082 
1083 	len = sizeof(transport->tcp_xid) - transport->tcp_offset;
1084 	dprintk("RPC:       reading XID (%Zu bytes)\n", len);
1085 	p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
1086 	used = xdr_skb_read_bits(desc, p, len);
1087 	transport->tcp_offset += used;
1088 	if (used != len)
1089 		return;
1090 	transport->tcp_flags &= ~TCP_RCV_COPY_XID;
1091 	transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
1092 	transport->tcp_copied = 4;
1093 	dprintk("RPC:       reading %s XID %08x\n",
1094 			(transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
1095 							      : "request with",
1096 			ntohl(transport->tcp_xid));
1097 	xs_tcp_check_fraghdr(transport);
1098 }
1099 
1100 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
1101 				       struct xdr_skb_reader *desc)
1102 {
1103 	size_t len, used;
1104 	u32 offset;
1105 	char *p;
1106 
1107 	/*
1108 	 * We want transport->tcp_offset to be 8 at the end of this routine
1109 	 * (4 bytes for the xid and 4 bytes for the call/reply flag).
1110 	 * When this function is called for the first time,
1111 	 * transport->tcp_offset is 4 (after having already read the xid).
1112 	 */
1113 	offset = transport->tcp_offset - sizeof(transport->tcp_xid);
1114 	len = sizeof(transport->tcp_calldir) - offset;
1115 	dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
1116 	p = ((char *) &transport->tcp_calldir) + offset;
1117 	used = xdr_skb_read_bits(desc, p, len);
1118 	transport->tcp_offset += used;
1119 	if (used != len)
1120 		return;
1121 	transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
1122 	/*
1123 	 * We don't yet have the XDR buffer, so we will write the calldir
1124 	 * out after we get the buffer from the 'struct rpc_rqst'
1125 	 */
1126 	switch (ntohl(transport->tcp_calldir)) {
1127 	case RPC_REPLY:
1128 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1129 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1130 		transport->tcp_flags |= TCP_RPC_REPLY;
1131 		break;
1132 	case RPC_CALL:
1133 		transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
1134 		transport->tcp_flags |= TCP_RCV_COPY_DATA;
1135 		transport->tcp_flags &= ~TCP_RPC_REPLY;
1136 		break;
1137 	default:
1138 		dprintk("RPC:       invalid request message type\n");
1139 		xs_tcp_force_close(&transport->xprt);
1140 	}
1141 	xs_tcp_check_fraghdr(transport);
1142 }
1143 
1144 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
1145 				     struct xdr_skb_reader *desc,
1146 				     struct rpc_rqst *req)
1147 {
1148 	struct sock_xprt *transport =
1149 				container_of(xprt, struct sock_xprt, xprt);
1150 	struct xdr_buf *rcvbuf;
1151 	size_t len;
1152 	ssize_t r;
1153 
1154 	rcvbuf = &req->rq_private_buf;
1155 
1156 	if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
1157 		/*
1158 		 * Save the RPC direction in the XDR buffer
1159 		 */
1160 		memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1161 			&transport->tcp_calldir,
1162 			sizeof(transport->tcp_calldir));
1163 		transport->tcp_copied += sizeof(transport->tcp_calldir);
1164 		transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1165 	}
1166 
1167 	len = desc->count;
1168 	if (len > transport->tcp_reclen - transport->tcp_offset) {
1169 		struct xdr_skb_reader my_desc;
1170 
1171 		len = transport->tcp_reclen - transport->tcp_offset;
1172 		memcpy(&my_desc, desc, sizeof(my_desc));
1173 		my_desc.count = len;
1174 		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1175 					  &my_desc, xdr_skb_read_bits);
1176 		desc->count -= r;
1177 		desc->offset += r;
1178 	} else
1179 		r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1180 					  desc, xdr_skb_read_bits);
1181 
1182 	if (r > 0) {
1183 		transport->tcp_copied += r;
1184 		transport->tcp_offset += r;
1185 	}
1186 	if (r != len) {
1187 		/* Error when copying to the receive buffer,
1188 		 * usually because we weren't able to allocate
1189 		 * additional buffer pages. All we can do now
1190 		 * is turn off TCP_RCV_COPY_DATA, so the request
1191 		 * will not receive any additional updates,
1192 		 * and time out.
1193 		 * Any remaining data from this record will
1194 		 * be discarded.
1195 		 */
1196 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1197 		dprintk("RPC:       XID %08x truncated request\n",
1198 				ntohl(transport->tcp_xid));
1199 		dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1200 				"tcp_offset = %u, tcp_reclen = %u\n",
1201 				xprt, transport->tcp_copied,
1202 				transport->tcp_offset, transport->tcp_reclen);
1203 		return;
1204 	}
1205 
1206 	dprintk("RPC:       XID %08x read %Zd bytes\n",
1207 			ntohl(transport->tcp_xid), r);
1208 	dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1209 			"tcp_reclen = %u\n", xprt, transport->tcp_copied,
1210 			transport->tcp_offset, transport->tcp_reclen);
1211 
1212 	if (transport->tcp_copied == req->rq_private_buf.buflen)
1213 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1214 	else if (transport->tcp_offset == transport->tcp_reclen) {
1215 		if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1216 			transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1217 	}
1218 }
1219 
1220 /*
1221  * Finds the request corresponding to the RPC xid and invokes the common
1222  * tcp read code to read the data.
1223  */
1224 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1225 				    struct xdr_skb_reader *desc)
1226 {
1227 	struct sock_xprt *transport =
1228 				container_of(xprt, struct sock_xprt, xprt);
1229 	struct rpc_rqst *req;
1230 
1231 	dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1232 
1233 	/* Find and lock the request corresponding to this xid */
1234 	spin_lock(&xprt->transport_lock);
1235 	req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1236 	if (!req) {
1237 		dprintk("RPC:       XID %08x request not found!\n",
1238 				ntohl(transport->tcp_xid));
1239 		spin_unlock(&xprt->transport_lock);
1240 		return -1;
1241 	}
1242 
1243 	xs_tcp_read_common(xprt, desc, req);
1244 
1245 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1246 		xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1247 
1248 	spin_unlock(&xprt->transport_lock);
1249 	return 0;
1250 }
1251 
1252 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1253 /*
1254  * Obtains an rpc_rqst previously allocated and invokes the common
1255  * tcp read code to read the data.  The result is placed in the callback
1256  * queue.
1257  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1258  * connection and return -1.
1259  */
1260 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1261 				       struct xdr_skb_reader *desc)
1262 {
1263 	struct sock_xprt *transport =
1264 				container_of(xprt, struct sock_xprt, xprt);
1265 	struct rpc_rqst *req;
1266 
1267 	req = xprt_alloc_bc_request(xprt);
1268 	if (req == NULL) {
1269 		printk(KERN_WARNING "Callback slot table overflowed\n");
1270 		xprt_force_disconnect(xprt);
1271 		return -1;
1272 	}
1273 
1274 	req->rq_xid = transport->tcp_xid;
1275 	dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1276 	xs_tcp_read_common(xprt, desc, req);
1277 
1278 	if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1279 		struct svc_serv *bc_serv = xprt->bc_serv;
1280 
1281 		/*
1282 		 * Add callback request to callback list.  The callback
1283 		 * service sleeps on the sv_cb_waitq waiting for new
1284 		 * requests.  Wake it up after adding enqueing the
1285 		 * request.
1286 		 */
1287 		dprintk("RPC:       add callback request to list\n");
1288 		spin_lock(&bc_serv->sv_cb_lock);
1289 		list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1290 		spin_unlock(&bc_serv->sv_cb_lock);
1291 		wake_up(&bc_serv->sv_cb_waitq);
1292 	}
1293 
1294 	req->rq_private_buf.len = transport->tcp_copied;
1295 
1296 	return 0;
1297 }
1298 
1299 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1300 					struct xdr_skb_reader *desc)
1301 {
1302 	struct sock_xprt *transport =
1303 				container_of(xprt, struct sock_xprt, xprt);
1304 
1305 	return (transport->tcp_flags & TCP_RPC_REPLY) ?
1306 		xs_tcp_read_reply(xprt, desc) :
1307 		xs_tcp_read_callback(xprt, desc);
1308 }
1309 #else
1310 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1311 					struct xdr_skb_reader *desc)
1312 {
1313 	return xs_tcp_read_reply(xprt, desc);
1314 }
1315 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1316 
1317 /*
1318  * Read data off the transport.  This can be either an RPC_CALL or an
1319  * RPC_REPLY.  Relay the processing to helper functions.
1320  */
1321 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1322 				    struct xdr_skb_reader *desc)
1323 {
1324 	struct sock_xprt *transport =
1325 				container_of(xprt, struct sock_xprt, xprt);
1326 
1327 	if (_xs_tcp_read_data(xprt, desc) == 0)
1328 		xs_tcp_check_fraghdr(transport);
1329 	else {
1330 		/*
1331 		 * The transport_lock protects the request handling.
1332 		 * There's no need to hold it to update the tcp_flags.
1333 		 */
1334 		transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1335 	}
1336 }
1337 
1338 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1339 {
1340 	size_t len;
1341 
1342 	len = transport->tcp_reclen - transport->tcp_offset;
1343 	if (len > desc->count)
1344 		len = desc->count;
1345 	desc->count -= len;
1346 	desc->offset += len;
1347 	transport->tcp_offset += len;
1348 	dprintk("RPC:       discarded %Zu bytes\n", len);
1349 	xs_tcp_check_fraghdr(transport);
1350 }
1351 
1352 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1353 {
1354 	struct rpc_xprt *xprt = rd_desc->arg.data;
1355 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1356 	struct xdr_skb_reader desc = {
1357 		.skb	= skb,
1358 		.offset	= offset,
1359 		.count	= len,
1360 	};
1361 
1362 	dprintk("RPC:       xs_tcp_data_recv started\n");
1363 	do {
1364 		/* Read in a new fragment marker if necessary */
1365 		/* Can we ever really expect to get completely empty fragments? */
1366 		if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1367 			xs_tcp_read_fraghdr(xprt, &desc);
1368 			continue;
1369 		}
1370 		/* Read in the xid if necessary */
1371 		if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1372 			xs_tcp_read_xid(transport, &desc);
1373 			continue;
1374 		}
1375 		/* Read in the call/reply flag */
1376 		if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1377 			xs_tcp_read_calldir(transport, &desc);
1378 			continue;
1379 		}
1380 		/* Read in the request data */
1381 		if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1382 			xs_tcp_read_data(xprt, &desc);
1383 			continue;
1384 		}
1385 		/* Skip over any trailing bytes on short reads */
1386 		xs_tcp_read_discard(transport, &desc);
1387 	} while (desc.count);
1388 	dprintk("RPC:       xs_tcp_data_recv done\n");
1389 	return len - desc.count;
1390 }
1391 
1392 /**
1393  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1394  * @sk: socket with data to read
1395  * @bytes: how much data to read
1396  *
1397  */
1398 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1399 {
1400 	struct rpc_xprt *xprt;
1401 	read_descriptor_t rd_desc;
1402 	int read;
1403 
1404 	dprintk("RPC:       xs_tcp_data_ready...\n");
1405 
1406 	read_lock_bh(&sk->sk_callback_lock);
1407 	if (!(xprt = xprt_from_sock(sk)))
1408 		goto out;
1409 	/* Any data means we had a useful conversation, so
1410 	 * the we don't need to delay the next reconnect
1411 	 */
1412 	if (xprt->reestablish_timeout)
1413 		xprt->reestablish_timeout = 0;
1414 
1415 	/* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1416 	rd_desc.arg.data = xprt;
1417 	do {
1418 		rd_desc.count = 65536;
1419 		read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1420 	} while (read > 0);
1421 out:
1422 	read_unlock_bh(&sk->sk_callback_lock);
1423 }
1424 
1425 /*
1426  * Do the equivalent of linger/linger2 handling for dealing with
1427  * broken servers that don't close the socket in a timely
1428  * fashion
1429  */
1430 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1431 		unsigned long timeout)
1432 {
1433 	struct sock_xprt *transport;
1434 
1435 	if (xprt_test_and_set_connecting(xprt))
1436 		return;
1437 	set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1438 	transport = container_of(xprt, struct sock_xprt, xprt);
1439 	queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1440 			   timeout);
1441 }
1442 
1443 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1444 {
1445 	struct sock_xprt *transport;
1446 
1447 	transport = container_of(xprt, struct sock_xprt, xprt);
1448 
1449 	if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1450 	    !cancel_delayed_work(&transport->connect_worker))
1451 		return;
1452 	clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1453 	xprt_clear_connecting(xprt);
1454 }
1455 
1456 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1457 {
1458 	smp_mb__before_clear_bit();
1459 	clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1460 	clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1461 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1462 	clear_bit(XPRT_CLOSING, &xprt->state);
1463 	smp_mb__after_clear_bit();
1464 	/* Mark transport as closed and wake up all pending tasks */
1465 	xprt_disconnect_done(xprt);
1466 }
1467 
1468 /**
1469  * xs_tcp_state_change - callback to handle TCP socket state changes
1470  * @sk: socket whose state has changed
1471  *
1472  */
1473 static void xs_tcp_state_change(struct sock *sk)
1474 {
1475 	struct rpc_xprt *xprt;
1476 
1477 	read_lock_bh(&sk->sk_callback_lock);
1478 	if (!(xprt = xprt_from_sock(sk)))
1479 		goto out;
1480 	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1481 	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1482 			sk->sk_state, xprt_connected(xprt),
1483 			sock_flag(sk, SOCK_DEAD),
1484 			sock_flag(sk, SOCK_ZAPPED),
1485 			sk->sk_shutdown);
1486 
1487 	switch (sk->sk_state) {
1488 	case TCP_ESTABLISHED:
1489 		spin_lock(&xprt->transport_lock);
1490 		if (!xprt_test_and_set_connected(xprt)) {
1491 			struct sock_xprt *transport = container_of(xprt,
1492 					struct sock_xprt, xprt);
1493 
1494 			/* Reset TCP record info */
1495 			transport->tcp_offset = 0;
1496 			transport->tcp_reclen = 0;
1497 			transport->tcp_copied = 0;
1498 			transport->tcp_flags =
1499 				TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1500 
1501 			xprt_wake_pending_tasks(xprt, -EAGAIN);
1502 		}
1503 		spin_unlock(&xprt->transport_lock);
1504 		break;
1505 	case TCP_FIN_WAIT1:
1506 		/* The client initiated a shutdown of the socket */
1507 		xprt->connect_cookie++;
1508 		xprt->reestablish_timeout = 0;
1509 		set_bit(XPRT_CLOSING, &xprt->state);
1510 		smp_mb__before_clear_bit();
1511 		clear_bit(XPRT_CONNECTED, &xprt->state);
1512 		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1513 		smp_mb__after_clear_bit();
1514 		xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1515 		break;
1516 	case TCP_CLOSE_WAIT:
1517 		/* The server initiated a shutdown of the socket */
1518 		xprt->connect_cookie++;
1519 		xs_tcp_force_close(xprt);
1520 	case TCP_CLOSING:
1521 		/*
1522 		 * If the server closed down the connection, make sure that
1523 		 * we back off before reconnecting
1524 		 */
1525 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1526 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1527 		break;
1528 	case TCP_LAST_ACK:
1529 		set_bit(XPRT_CLOSING, &xprt->state);
1530 		xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1531 		smp_mb__before_clear_bit();
1532 		clear_bit(XPRT_CONNECTED, &xprt->state);
1533 		smp_mb__after_clear_bit();
1534 		break;
1535 	case TCP_CLOSE:
1536 		xs_tcp_cancel_linger_timeout(xprt);
1537 		xs_sock_mark_closed(xprt);
1538 	}
1539  out:
1540 	read_unlock_bh(&sk->sk_callback_lock);
1541 }
1542 
1543 /**
1544  * xs_error_report - callback mainly for catching socket errors
1545  * @sk: socket
1546  */
1547 static void xs_error_report(struct sock *sk)
1548 {
1549 	struct rpc_xprt *xprt;
1550 
1551 	read_lock_bh(&sk->sk_callback_lock);
1552 	if (!(xprt = xprt_from_sock(sk)))
1553 		goto out;
1554 	dprintk("RPC:       %s client %p...\n"
1555 			"RPC:       error %d\n",
1556 			__func__, xprt, sk->sk_err);
1557 	xprt_wake_pending_tasks(xprt, -EAGAIN);
1558 out:
1559 	read_unlock_bh(&sk->sk_callback_lock);
1560 }
1561 
1562 static void xs_write_space(struct sock *sk)
1563 {
1564 	struct socket *sock;
1565 	struct rpc_xprt *xprt;
1566 
1567 	if (unlikely(!(sock = sk->sk_socket)))
1568 		return;
1569 	clear_bit(SOCK_NOSPACE, &sock->flags);
1570 
1571 	if (unlikely(!(xprt = xprt_from_sock(sk))))
1572 		return;
1573 	if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1574 		return;
1575 
1576 	xprt_write_space(xprt);
1577 }
1578 
1579 /**
1580  * xs_udp_write_space - callback invoked when socket buffer space
1581  *                             becomes available
1582  * @sk: socket whose state has changed
1583  *
1584  * Called when more output buffer space is available for this socket.
1585  * We try not to wake our writers until they can make "significant"
1586  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1587  * with a bunch of small requests.
1588  */
1589 static void xs_udp_write_space(struct sock *sk)
1590 {
1591 	read_lock_bh(&sk->sk_callback_lock);
1592 
1593 	/* from net/core/sock.c:sock_def_write_space */
1594 	if (sock_writeable(sk))
1595 		xs_write_space(sk);
1596 
1597 	read_unlock_bh(&sk->sk_callback_lock);
1598 }
1599 
1600 /**
1601  * xs_tcp_write_space - callback invoked when socket buffer space
1602  *                             becomes available
1603  * @sk: socket whose state has changed
1604  *
1605  * Called when more output buffer space is available for this socket.
1606  * We try not to wake our writers until they can make "significant"
1607  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1608  * with a bunch of small requests.
1609  */
1610 static void xs_tcp_write_space(struct sock *sk)
1611 {
1612 	read_lock_bh(&sk->sk_callback_lock);
1613 
1614 	/* from net/core/stream.c:sk_stream_write_space */
1615 	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1616 		xs_write_space(sk);
1617 
1618 	read_unlock_bh(&sk->sk_callback_lock);
1619 }
1620 
1621 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1622 {
1623 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1624 	struct sock *sk = transport->inet;
1625 
1626 	if (transport->rcvsize) {
1627 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1628 		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1629 	}
1630 	if (transport->sndsize) {
1631 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1632 		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1633 		sk->sk_write_space(sk);
1634 	}
1635 }
1636 
1637 /**
1638  * xs_udp_set_buffer_size - set send and receive limits
1639  * @xprt: generic transport
1640  * @sndsize: requested size of send buffer, in bytes
1641  * @rcvsize: requested size of receive buffer, in bytes
1642  *
1643  * Set socket send and receive buffer size limits.
1644  */
1645 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1646 {
1647 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1648 
1649 	transport->sndsize = 0;
1650 	if (sndsize)
1651 		transport->sndsize = sndsize + 1024;
1652 	transport->rcvsize = 0;
1653 	if (rcvsize)
1654 		transport->rcvsize = rcvsize + 1024;
1655 
1656 	xs_udp_do_set_buffer_size(xprt);
1657 }
1658 
1659 /**
1660  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1661  * @task: task that timed out
1662  *
1663  * Adjust the congestion window after a retransmit timeout has occurred.
1664  */
1665 static void xs_udp_timer(struct rpc_task *task)
1666 {
1667 	xprt_adjust_cwnd(task, -ETIMEDOUT);
1668 }
1669 
1670 static unsigned short xs_get_random_port(void)
1671 {
1672 	unsigned short range = xprt_max_resvport - xprt_min_resvport;
1673 	unsigned short rand = (unsigned short) net_random() % range;
1674 	return rand + xprt_min_resvport;
1675 }
1676 
1677 /**
1678  * xs_set_port - reset the port number in the remote endpoint address
1679  * @xprt: generic transport
1680  * @port: new port number
1681  *
1682  */
1683 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1684 {
1685 	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1686 
1687 	rpc_set_port(xs_addr(xprt), port);
1688 	xs_update_peer_port(xprt);
1689 }
1690 
1691 static unsigned short xs_get_srcport(struct sock_xprt *transport)
1692 {
1693 	unsigned short port = transport->srcport;
1694 
1695 	if (port == 0 && transport->xprt.resvport)
1696 		port = xs_get_random_port();
1697 	return port;
1698 }
1699 
1700 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1701 {
1702 	if (transport->srcport != 0)
1703 		transport->srcport = 0;
1704 	if (!transport->xprt.resvport)
1705 		return 0;
1706 	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1707 		return xprt_max_resvport;
1708 	return --port;
1709 }
1710 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1711 {
1712 	struct sockaddr_storage myaddr;
1713 	int err, nloop = 0;
1714 	unsigned short port = xs_get_srcport(transport);
1715 	unsigned short last;
1716 
1717 	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1718 	do {
1719 		rpc_set_port((struct sockaddr *)&myaddr, port);
1720 		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1721 				transport->xprt.addrlen);
1722 		if (port == 0)
1723 			break;
1724 		if (err == 0) {
1725 			transport->srcport = port;
1726 			break;
1727 		}
1728 		last = port;
1729 		port = xs_next_srcport(transport, port);
1730 		if (port > last)
1731 			nloop++;
1732 	} while (err == -EADDRINUSE && nloop != 2);
1733 
1734 	if (myaddr.ss_family == AF_INET)
1735 		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1736 				&((struct sockaddr_in *)&myaddr)->sin_addr,
1737 				port, err ? "failed" : "ok", err);
1738 	else
1739 		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1740 				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1741 				port, err ? "failed" : "ok", err);
1742 	return err;
1743 }
1744 
1745 /*
1746  * We don't support autobind on AF_LOCAL sockets
1747  */
1748 static void xs_local_rpcbind(struct rpc_task *task)
1749 {
1750 	xprt_set_bound(task->tk_xprt);
1751 }
1752 
1753 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1754 {
1755 }
1756 
1757 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1758 static struct lock_class_key xs_key[2];
1759 static struct lock_class_key xs_slock_key[2];
1760 
1761 static inline void xs_reclassify_socketu(struct socket *sock)
1762 {
1763 	struct sock *sk = sock->sk;
1764 
1765 	BUG_ON(sock_owned_by_user(sk));
1766 	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1767 		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1768 }
1769 
1770 static inline void xs_reclassify_socket4(struct socket *sock)
1771 {
1772 	struct sock *sk = sock->sk;
1773 
1774 	BUG_ON(sock_owned_by_user(sk));
1775 	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1776 		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1777 }
1778 
1779 static inline void xs_reclassify_socket6(struct socket *sock)
1780 {
1781 	struct sock *sk = sock->sk;
1782 
1783 	BUG_ON(sock_owned_by_user(sk));
1784 	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1785 		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1786 }
1787 
1788 static inline void xs_reclassify_socket(int family, struct socket *sock)
1789 {
1790 	switch (family) {
1791 	case AF_LOCAL:
1792 		xs_reclassify_socketu(sock);
1793 		break;
1794 	case AF_INET:
1795 		xs_reclassify_socket4(sock);
1796 		break;
1797 	case AF_INET6:
1798 		xs_reclassify_socket6(sock);
1799 		break;
1800 	}
1801 }
1802 #else
1803 static inline void xs_reclassify_socketu(struct socket *sock)
1804 {
1805 }
1806 
1807 static inline void xs_reclassify_socket4(struct socket *sock)
1808 {
1809 }
1810 
1811 static inline void xs_reclassify_socket6(struct socket *sock)
1812 {
1813 }
1814 
1815 static inline void xs_reclassify_socket(int family, struct socket *sock)
1816 {
1817 }
1818 #endif
1819 
1820 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1821 		struct sock_xprt *transport, int family, int type, int protocol)
1822 {
1823 	struct socket *sock;
1824 	int err;
1825 
1826 	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1827 	if (err < 0) {
1828 		dprintk("RPC:       can't create %d transport socket (%d).\n",
1829 				protocol, -err);
1830 		goto out;
1831 	}
1832 	xs_reclassify_socket(family, sock);
1833 
1834 	err = xs_bind(transport, sock);
1835 	if (err) {
1836 		sock_release(sock);
1837 		goto out;
1838 	}
1839 
1840 	return sock;
1841 out:
1842 	return ERR_PTR(err);
1843 }
1844 
1845 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1846 				      struct socket *sock)
1847 {
1848 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1849 									xprt);
1850 
1851 	if (!transport->inet) {
1852 		struct sock *sk = sock->sk;
1853 
1854 		write_lock_bh(&sk->sk_callback_lock);
1855 
1856 		xs_save_old_callbacks(transport, sk);
1857 
1858 		sk->sk_user_data = xprt;
1859 		sk->sk_data_ready = xs_local_data_ready;
1860 		sk->sk_write_space = xs_udp_write_space;
1861 		sk->sk_error_report = xs_error_report;
1862 		sk->sk_allocation = GFP_ATOMIC;
1863 
1864 		xprt_clear_connected(xprt);
1865 
1866 		/* Reset to new socket */
1867 		transport->sock = sock;
1868 		transport->inet = sk;
1869 
1870 		write_unlock_bh(&sk->sk_callback_lock);
1871 	}
1872 
1873 	/* Tell the socket layer to start connecting... */
1874 	xprt->stat.connect_count++;
1875 	xprt->stat.connect_start = jiffies;
1876 	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1877 }
1878 
1879 /**
1880  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1881  * @xprt: RPC transport to connect
1882  * @transport: socket transport to connect
1883  * @create_sock: function to create a socket of the correct type
1884  *
1885  * Invoked by a work queue tasklet.
1886  */
1887 static void xs_local_setup_socket(struct work_struct *work)
1888 {
1889 	struct sock_xprt *transport =
1890 		container_of(work, struct sock_xprt, connect_worker.work);
1891 	struct rpc_xprt *xprt = &transport->xprt;
1892 	struct socket *sock;
1893 	int status = -EIO;
1894 
1895 	current->flags |= PF_FSTRANS;
1896 
1897 	clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1898 	status = __sock_create(xprt->xprt_net, AF_LOCAL,
1899 					SOCK_STREAM, 0, &sock, 1);
1900 	if (status < 0) {
1901 		dprintk("RPC:       can't create AF_LOCAL "
1902 			"transport socket (%d).\n", -status);
1903 		goto out;
1904 	}
1905 	xs_reclassify_socketu(sock);
1906 
1907 	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1908 			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1909 
1910 	status = xs_local_finish_connecting(xprt, sock);
1911 	switch (status) {
1912 	case 0:
1913 		dprintk("RPC:       xprt %p connected to %s\n",
1914 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1915 		xprt_set_connected(xprt);
1916 		break;
1917 	case -ENOENT:
1918 		dprintk("RPC:       xprt %p: socket %s does not exist\n",
1919 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1920 		break;
1921 	default:
1922 		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
1923 				__func__, -status,
1924 				xprt->address_strings[RPC_DISPLAY_ADDR]);
1925 	}
1926 
1927 out:
1928 	xprt_clear_connecting(xprt);
1929 	xprt_wake_pending_tasks(xprt, status);
1930 	current->flags &= ~PF_FSTRANS;
1931 }
1932 
1933 #ifdef CONFIG_SUNRPC_SWAP
1934 static void xs_set_memalloc(struct rpc_xprt *xprt)
1935 {
1936 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1937 			xprt);
1938 
1939 	if (xprt->swapper)
1940 		sk_set_memalloc(transport->inet);
1941 }
1942 
1943 /**
1944  * xs_swapper - Tag this transport as being used for swap.
1945  * @xprt: transport to tag
1946  * @enable: enable/disable
1947  *
1948  */
1949 int xs_swapper(struct rpc_xprt *xprt, int enable)
1950 {
1951 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1952 			xprt);
1953 	int err = 0;
1954 
1955 	if (enable) {
1956 		xprt->swapper++;
1957 		xs_set_memalloc(xprt);
1958 	} else if (xprt->swapper) {
1959 		xprt->swapper--;
1960 		sk_clear_memalloc(transport->inet);
1961 	}
1962 
1963 	return err;
1964 }
1965 EXPORT_SYMBOL_GPL(xs_swapper);
1966 #else
1967 static void xs_set_memalloc(struct rpc_xprt *xprt)
1968 {
1969 }
1970 #endif
1971 
1972 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1973 {
1974 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1975 
1976 	if (!transport->inet) {
1977 		struct sock *sk = sock->sk;
1978 
1979 		write_lock_bh(&sk->sk_callback_lock);
1980 
1981 		xs_save_old_callbacks(transport, sk);
1982 
1983 		sk->sk_user_data = xprt;
1984 		sk->sk_data_ready = xs_udp_data_ready;
1985 		sk->sk_write_space = xs_udp_write_space;
1986 		sk->sk_error_report = xs_error_report;
1987 		sk->sk_no_check = UDP_CSUM_NORCV;
1988 		sk->sk_allocation = GFP_ATOMIC;
1989 
1990 		xprt_set_connected(xprt);
1991 
1992 		/* Reset to new socket */
1993 		transport->sock = sock;
1994 		transport->inet = sk;
1995 
1996 		xs_set_memalloc(xprt);
1997 
1998 		write_unlock_bh(&sk->sk_callback_lock);
1999 	}
2000 	xs_udp_do_set_buffer_size(xprt);
2001 }
2002 
2003 static void xs_udp_setup_socket(struct work_struct *work)
2004 {
2005 	struct sock_xprt *transport =
2006 		container_of(work, struct sock_xprt, connect_worker.work);
2007 	struct rpc_xprt *xprt = &transport->xprt;
2008 	struct socket *sock = transport->sock;
2009 	int status = -EIO;
2010 
2011 	current->flags |= PF_FSTRANS;
2012 
2013 	/* Start by resetting any existing state */
2014 	xs_reset_transport(transport);
2015 	sock = xs_create_sock(xprt, transport,
2016 			xs_addr(xprt)->sa_family, SOCK_DGRAM, IPPROTO_UDP);
2017 	if (IS_ERR(sock))
2018 		goto out;
2019 
2020 	dprintk("RPC:       worker connecting xprt %p via %s to "
2021 				"%s (port %s)\n", xprt,
2022 			xprt->address_strings[RPC_DISPLAY_PROTO],
2023 			xprt->address_strings[RPC_DISPLAY_ADDR],
2024 			xprt->address_strings[RPC_DISPLAY_PORT]);
2025 
2026 	xs_udp_finish_connecting(xprt, sock);
2027 	status = 0;
2028 out:
2029 	xprt_clear_connecting(xprt);
2030 	xprt_wake_pending_tasks(xprt, status);
2031 	current->flags &= ~PF_FSTRANS;
2032 }
2033 
2034 /*
2035  * We need to preserve the port number so the reply cache on the server can
2036  * find our cached RPC replies when we get around to reconnecting.
2037  */
2038 static void xs_abort_connection(struct sock_xprt *transport)
2039 {
2040 	int result;
2041 	struct sockaddr any;
2042 
2043 	dprintk("RPC:       disconnecting xprt %p to reuse port\n", transport);
2044 
2045 	/*
2046 	 * Disconnect the transport socket by doing a connect operation
2047 	 * with AF_UNSPEC.  This should return immediately...
2048 	 */
2049 	memset(&any, 0, sizeof(any));
2050 	any.sa_family = AF_UNSPEC;
2051 	result = kernel_connect(transport->sock, &any, sizeof(any), 0);
2052 	if (!result)
2053 		xs_sock_mark_closed(&transport->xprt);
2054 	else
2055 		dprintk("RPC:       AF_UNSPEC connect return code %d\n",
2056 				result);
2057 }
2058 
2059 static void xs_tcp_reuse_connection(struct sock_xprt *transport)
2060 {
2061 	unsigned int state = transport->inet->sk_state;
2062 
2063 	if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED) {
2064 		/* we don't need to abort the connection if the socket
2065 		 * hasn't undergone a shutdown
2066 		 */
2067 		if (transport->inet->sk_shutdown == 0)
2068 			return;
2069 		dprintk("RPC:       %s: TCP_CLOSEd and sk_shutdown set to %d\n",
2070 				__func__, transport->inet->sk_shutdown);
2071 	}
2072 	if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT)) {
2073 		/* we don't need to abort the connection if the socket
2074 		 * hasn't undergone a shutdown
2075 		 */
2076 		if (transport->inet->sk_shutdown == 0)
2077 			return;
2078 		dprintk("RPC:       %s: ESTABLISHED/SYN_SENT "
2079 				"sk_shutdown set to %d\n",
2080 				__func__, transport->inet->sk_shutdown);
2081 	}
2082 	xs_abort_connection(transport);
2083 }
2084 
2085 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2086 {
2087 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2088 	int ret = -ENOTCONN;
2089 
2090 	if (!transport->inet) {
2091 		struct sock *sk = sock->sk;
2092 
2093 		write_lock_bh(&sk->sk_callback_lock);
2094 
2095 		xs_save_old_callbacks(transport, sk);
2096 
2097 		sk->sk_user_data = xprt;
2098 		sk->sk_data_ready = xs_tcp_data_ready;
2099 		sk->sk_state_change = xs_tcp_state_change;
2100 		sk->sk_write_space = xs_tcp_write_space;
2101 		sk->sk_error_report = xs_error_report;
2102 		sk->sk_allocation = GFP_ATOMIC;
2103 
2104 		/* socket options */
2105 		sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
2106 		sock_reset_flag(sk, SOCK_LINGER);
2107 		tcp_sk(sk)->linger2 = 0;
2108 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2109 
2110 		xprt_clear_connected(xprt);
2111 
2112 		/* Reset to new socket */
2113 		transport->sock = sock;
2114 		transport->inet = sk;
2115 
2116 		write_unlock_bh(&sk->sk_callback_lock);
2117 	}
2118 
2119 	if (!xprt_bound(xprt))
2120 		goto out;
2121 
2122 	xs_set_memalloc(xprt);
2123 
2124 	/* Tell the socket layer to start connecting... */
2125 	xprt->stat.connect_count++;
2126 	xprt->stat.connect_start = jiffies;
2127 	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2128 	switch (ret) {
2129 	case 0:
2130 	case -EINPROGRESS:
2131 		/* SYN_SENT! */
2132 		xprt->connect_cookie++;
2133 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2134 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2135 	}
2136 out:
2137 	return ret;
2138 }
2139 
2140 /**
2141  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2142  * @xprt: RPC transport to connect
2143  * @transport: socket transport to connect
2144  * @create_sock: function to create a socket of the correct type
2145  *
2146  * Invoked by a work queue tasklet.
2147  */
2148 static void xs_tcp_setup_socket(struct work_struct *work)
2149 {
2150 	struct sock_xprt *transport =
2151 		container_of(work, struct sock_xprt, connect_worker.work);
2152 	struct socket *sock = transport->sock;
2153 	struct rpc_xprt *xprt = &transport->xprt;
2154 	int status = -EIO;
2155 
2156 	current->flags |= PF_FSTRANS;
2157 
2158 	if (!sock) {
2159 		clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
2160 		sock = xs_create_sock(xprt, transport,
2161 				xs_addr(xprt)->sa_family, SOCK_STREAM, IPPROTO_TCP);
2162 		if (IS_ERR(sock)) {
2163 			status = PTR_ERR(sock);
2164 			goto out;
2165 		}
2166 	} else {
2167 		int abort_and_exit;
2168 
2169 		abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
2170 				&xprt->state);
2171 		/* "close" the socket, preserving the local port */
2172 		xs_tcp_reuse_connection(transport);
2173 
2174 		if (abort_and_exit)
2175 			goto out_eagain;
2176 	}
2177 
2178 	dprintk("RPC:       worker connecting xprt %p via %s to "
2179 				"%s (port %s)\n", xprt,
2180 			xprt->address_strings[RPC_DISPLAY_PROTO],
2181 			xprt->address_strings[RPC_DISPLAY_ADDR],
2182 			xprt->address_strings[RPC_DISPLAY_PORT]);
2183 
2184 	status = xs_tcp_finish_connecting(xprt, sock);
2185 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2186 			xprt, -status, xprt_connected(xprt),
2187 			sock->sk->sk_state);
2188 	switch (status) {
2189 	default:
2190 		printk("%s: connect returned unhandled error %d\n",
2191 			__func__, status);
2192 	case -EADDRNOTAVAIL:
2193 		/* We're probably in TIME_WAIT. Get rid of existing socket,
2194 		 * and retry
2195 		 */
2196 		xs_tcp_force_close(xprt);
2197 		break;
2198 	case -ECONNREFUSED:
2199 	case -ECONNRESET:
2200 	case -ENETUNREACH:
2201 		/* retry with existing socket, after a delay */
2202 	case 0:
2203 	case -EINPROGRESS:
2204 	case -EALREADY:
2205 		xprt_clear_connecting(xprt);
2206 		current->flags &= ~PF_FSTRANS;
2207 		return;
2208 	case -EINVAL:
2209 		/* Happens, for instance, if the user specified a link
2210 		 * local IPv6 address without a scope-id.
2211 		 */
2212 		goto out;
2213 	}
2214 out_eagain:
2215 	status = -EAGAIN;
2216 out:
2217 	xprt_clear_connecting(xprt);
2218 	xprt_wake_pending_tasks(xprt, status);
2219 	current->flags &= ~PF_FSTRANS;
2220 }
2221 
2222 /**
2223  * xs_connect - connect a socket to a remote endpoint
2224  * @task: address of RPC task that manages state of connect request
2225  *
2226  * TCP: If the remote end dropped the connection, delay reconnecting.
2227  *
2228  * UDP socket connects are synchronous, but we use a work queue anyway
2229  * to guarantee that even unprivileged user processes can set up a
2230  * socket on a privileged port.
2231  *
2232  * If a UDP socket connect fails, the delay behavior here prevents
2233  * retry floods (hard mounts).
2234  */
2235 static void xs_connect(struct rpc_task *task)
2236 {
2237 	struct rpc_xprt *xprt = task->tk_xprt;
2238 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2239 
2240 	if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2241 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2242 				"seconds\n",
2243 				xprt, xprt->reestablish_timeout / HZ);
2244 		queue_delayed_work(rpciod_workqueue,
2245 				   &transport->connect_worker,
2246 				   xprt->reestablish_timeout);
2247 		xprt->reestablish_timeout <<= 1;
2248 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2249 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2250 		if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2251 			xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2252 	} else {
2253 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2254 		queue_delayed_work(rpciod_workqueue,
2255 				   &transport->connect_worker, 0);
2256 	}
2257 }
2258 
2259 /**
2260  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2261  * @xprt: rpc_xprt struct containing statistics
2262  * @seq: output file
2263  *
2264  */
2265 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2266 {
2267 	long idle_time = 0;
2268 
2269 	if (xprt_connected(xprt))
2270 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2271 
2272 	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2273 			"%llu %llu %lu %llu %llu\n",
2274 			xprt->stat.bind_count,
2275 			xprt->stat.connect_count,
2276 			xprt->stat.connect_time,
2277 			idle_time,
2278 			xprt->stat.sends,
2279 			xprt->stat.recvs,
2280 			xprt->stat.bad_xids,
2281 			xprt->stat.req_u,
2282 			xprt->stat.bklog_u,
2283 			xprt->stat.max_slots,
2284 			xprt->stat.sending_u,
2285 			xprt->stat.pending_u);
2286 }
2287 
2288 /**
2289  * xs_udp_print_stats - display UDP socket-specifc stats
2290  * @xprt: rpc_xprt struct containing statistics
2291  * @seq: output file
2292  *
2293  */
2294 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2295 {
2296 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2297 
2298 	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2299 			"%lu %llu %llu\n",
2300 			transport->srcport,
2301 			xprt->stat.bind_count,
2302 			xprt->stat.sends,
2303 			xprt->stat.recvs,
2304 			xprt->stat.bad_xids,
2305 			xprt->stat.req_u,
2306 			xprt->stat.bklog_u,
2307 			xprt->stat.max_slots,
2308 			xprt->stat.sending_u,
2309 			xprt->stat.pending_u);
2310 }
2311 
2312 /**
2313  * xs_tcp_print_stats - display TCP socket-specifc stats
2314  * @xprt: rpc_xprt struct containing statistics
2315  * @seq: output file
2316  *
2317  */
2318 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2319 {
2320 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2321 	long idle_time = 0;
2322 
2323 	if (xprt_connected(xprt))
2324 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2325 
2326 	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2327 			"%llu %llu %lu %llu %llu\n",
2328 			transport->srcport,
2329 			xprt->stat.bind_count,
2330 			xprt->stat.connect_count,
2331 			xprt->stat.connect_time,
2332 			idle_time,
2333 			xprt->stat.sends,
2334 			xprt->stat.recvs,
2335 			xprt->stat.bad_xids,
2336 			xprt->stat.req_u,
2337 			xprt->stat.bklog_u,
2338 			xprt->stat.max_slots,
2339 			xprt->stat.sending_u,
2340 			xprt->stat.pending_u);
2341 }
2342 
2343 /*
2344  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2345  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2346  * to use the server side send routines.
2347  */
2348 static void *bc_malloc(struct rpc_task *task, size_t size)
2349 {
2350 	struct page *page;
2351 	struct rpc_buffer *buf;
2352 
2353 	BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2354 	page = alloc_page(GFP_KERNEL);
2355 
2356 	if (!page)
2357 		return NULL;
2358 
2359 	buf = page_address(page);
2360 	buf->len = PAGE_SIZE;
2361 
2362 	return buf->data;
2363 }
2364 
2365 /*
2366  * Free the space allocated in the bc_alloc routine
2367  */
2368 static void bc_free(void *buffer)
2369 {
2370 	struct rpc_buffer *buf;
2371 
2372 	if (!buffer)
2373 		return;
2374 
2375 	buf = container_of(buffer, struct rpc_buffer, data);
2376 	free_page((unsigned long)buf);
2377 }
2378 
2379 /*
2380  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2381  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2382  */
2383 static int bc_sendto(struct rpc_rqst *req)
2384 {
2385 	int len;
2386 	struct xdr_buf *xbufp = &req->rq_snd_buf;
2387 	struct rpc_xprt *xprt = req->rq_xprt;
2388 	struct sock_xprt *transport =
2389 				container_of(xprt, struct sock_xprt, xprt);
2390 	struct socket *sock = transport->sock;
2391 	unsigned long headoff;
2392 	unsigned long tailoff;
2393 
2394 	xs_encode_stream_record_marker(xbufp);
2395 
2396 	tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2397 	headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2398 	len = svc_send_common(sock, xbufp,
2399 			      virt_to_page(xbufp->head[0].iov_base), headoff,
2400 			      xbufp->tail[0].iov_base, tailoff);
2401 
2402 	if (len != xbufp->len) {
2403 		printk(KERN_NOTICE "Error sending entire callback!\n");
2404 		len = -EAGAIN;
2405 	}
2406 
2407 	return len;
2408 }
2409 
2410 /*
2411  * The send routine. Borrows from svc_send
2412  */
2413 static int bc_send_request(struct rpc_task *task)
2414 {
2415 	struct rpc_rqst *req = task->tk_rqstp;
2416 	struct svc_xprt	*xprt;
2417 	struct svc_sock         *svsk;
2418 	u32                     len;
2419 
2420 	dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2421 	/*
2422 	 * Get the server socket associated with this callback xprt
2423 	 */
2424 	xprt = req->rq_xprt->bc_xprt;
2425 	svsk = container_of(xprt, struct svc_sock, sk_xprt);
2426 
2427 	/*
2428 	 * Grab the mutex to serialize data as the connection is shared
2429 	 * with the fore channel
2430 	 */
2431 	if (!mutex_trylock(&xprt->xpt_mutex)) {
2432 		rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2433 		if (!mutex_trylock(&xprt->xpt_mutex))
2434 			return -EAGAIN;
2435 		rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2436 	}
2437 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2438 		len = -ENOTCONN;
2439 	else
2440 		len = bc_sendto(req);
2441 	mutex_unlock(&xprt->xpt_mutex);
2442 
2443 	if (len > 0)
2444 		len = 0;
2445 
2446 	return len;
2447 }
2448 
2449 /*
2450  * The close routine. Since this is client initiated, we do nothing
2451  */
2452 
2453 static void bc_close(struct rpc_xprt *xprt)
2454 {
2455 }
2456 
2457 /*
2458  * The xprt destroy routine. Again, because this connection is client
2459  * initiated, we do nothing
2460  */
2461 
2462 static void bc_destroy(struct rpc_xprt *xprt)
2463 {
2464 }
2465 
2466 static struct rpc_xprt_ops xs_local_ops = {
2467 	.reserve_xprt		= xprt_reserve_xprt,
2468 	.release_xprt		= xs_tcp_release_xprt,
2469 	.alloc_slot		= xprt_alloc_slot,
2470 	.rpcbind		= xs_local_rpcbind,
2471 	.set_port		= xs_local_set_port,
2472 	.connect		= xs_connect,
2473 	.buf_alloc		= rpc_malloc,
2474 	.buf_free		= rpc_free,
2475 	.send_request		= xs_local_send_request,
2476 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2477 	.close			= xs_close,
2478 	.destroy		= xs_destroy,
2479 	.print_stats		= xs_local_print_stats,
2480 };
2481 
2482 static struct rpc_xprt_ops xs_udp_ops = {
2483 	.set_buffer_size	= xs_udp_set_buffer_size,
2484 	.reserve_xprt		= xprt_reserve_xprt_cong,
2485 	.release_xprt		= xprt_release_xprt_cong,
2486 	.alloc_slot		= xprt_alloc_slot,
2487 	.rpcbind		= rpcb_getport_async,
2488 	.set_port		= xs_set_port,
2489 	.connect		= xs_connect,
2490 	.buf_alloc		= rpc_malloc,
2491 	.buf_free		= rpc_free,
2492 	.send_request		= xs_udp_send_request,
2493 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
2494 	.timer			= xs_udp_timer,
2495 	.release_request	= xprt_release_rqst_cong,
2496 	.close			= xs_close,
2497 	.destroy		= xs_destroy,
2498 	.print_stats		= xs_udp_print_stats,
2499 };
2500 
2501 static struct rpc_xprt_ops xs_tcp_ops = {
2502 	.reserve_xprt		= xprt_reserve_xprt,
2503 	.release_xprt		= xs_tcp_release_xprt,
2504 	.alloc_slot		= xprt_lock_and_alloc_slot,
2505 	.rpcbind		= rpcb_getport_async,
2506 	.set_port		= xs_set_port,
2507 	.connect		= xs_connect,
2508 	.buf_alloc		= rpc_malloc,
2509 	.buf_free		= rpc_free,
2510 	.send_request		= xs_tcp_send_request,
2511 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2512 	.close			= xs_tcp_close,
2513 	.destroy		= xs_destroy,
2514 	.print_stats		= xs_tcp_print_stats,
2515 };
2516 
2517 /*
2518  * The rpc_xprt_ops for the server backchannel
2519  */
2520 
2521 static struct rpc_xprt_ops bc_tcp_ops = {
2522 	.reserve_xprt		= xprt_reserve_xprt,
2523 	.release_xprt		= xprt_release_xprt,
2524 	.alloc_slot		= xprt_alloc_slot,
2525 	.rpcbind		= xs_local_rpcbind,
2526 	.buf_alloc		= bc_malloc,
2527 	.buf_free		= bc_free,
2528 	.send_request		= bc_send_request,
2529 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
2530 	.close			= bc_close,
2531 	.destroy		= bc_destroy,
2532 	.print_stats		= xs_tcp_print_stats,
2533 };
2534 
2535 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2536 {
2537 	static const struct sockaddr_in sin = {
2538 		.sin_family		= AF_INET,
2539 		.sin_addr.s_addr	= htonl(INADDR_ANY),
2540 	};
2541 	static const struct sockaddr_in6 sin6 = {
2542 		.sin6_family		= AF_INET6,
2543 		.sin6_addr		= IN6ADDR_ANY_INIT,
2544 	};
2545 
2546 	switch (family) {
2547 	case AF_LOCAL:
2548 		break;
2549 	case AF_INET:
2550 		memcpy(sap, &sin, sizeof(sin));
2551 		break;
2552 	case AF_INET6:
2553 		memcpy(sap, &sin6, sizeof(sin6));
2554 		break;
2555 	default:
2556 		dprintk("RPC:       %s: Bad address family\n", __func__);
2557 		return -EAFNOSUPPORT;
2558 	}
2559 	return 0;
2560 }
2561 
2562 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2563 				      unsigned int slot_table_size,
2564 				      unsigned int max_slot_table_size)
2565 {
2566 	struct rpc_xprt *xprt;
2567 	struct sock_xprt *new;
2568 
2569 	if (args->addrlen > sizeof(xprt->addr)) {
2570 		dprintk("RPC:       xs_setup_xprt: address too large\n");
2571 		return ERR_PTR(-EBADF);
2572 	}
2573 
2574 	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2575 			max_slot_table_size);
2576 	if (xprt == NULL) {
2577 		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2578 				"rpc_xprt\n");
2579 		return ERR_PTR(-ENOMEM);
2580 	}
2581 
2582 	new = container_of(xprt, struct sock_xprt, xprt);
2583 	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2584 	xprt->addrlen = args->addrlen;
2585 	if (args->srcaddr)
2586 		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2587 	else {
2588 		int err;
2589 		err = xs_init_anyaddr(args->dstaddr->sa_family,
2590 					(struct sockaddr *)&new->srcaddr);
2591 		if (err != 0) {
2592 			xprt_free(xprt);
2593 			return ERR_PTR(err);
2594 		}
2595 	}
2596 
2597 	return xprt;
2598 }
2599 
2600 static const struct rpc_timeout xs_local_default_timeout = {
2601 	.to_initval = 10 * HZ,
2602 	.to_maxval = 10 * HZ,
2603 	.to_retries = 2,
2604 };
2605 
2606 /**
2607  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2608  * @args: rpc transport creation arguments
2609  *
2610  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2611  */
2612 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2613 {
2614 	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2615 	struct sock_xprt *transport;
2616 	struct rpc_xprt *xprt;
2617 	struct rpc_xprt *ret;
2618 
2619 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2620 			xprt_max_tcp_slot_table_entries);
2621 	if (IS_ERR(xprt))
2622 		return xprt;
2623 	transport = container_of(xprt, struct sock_xprt, xprt);
2624 
2625 	xprt->prot = 0;
2626 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2627 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2628 
2629 	xprt->bind_timeout = XS_BIND_TO;
2630 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2631 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2632 
2633 	xprt->ops = &xs_local_ops;
2634 	xprt->timeout = &xs_local_default_timeout;
2635 
2636 	switch (sun->sun_family) {
2637 	case AF_LOCAL:
2638 		if (sun->sun_path[0] != '/') {
2639 			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2640 					sun->sun_path);
2641 			ret = ERR_PTR(-EINVAL);
2642 			goto out_err;
2643 		}
2644 		xprt_set_bound(xprt);
2645 		INIT_DELAYED_WORK(&transport->connect_worker,
2646 					xs_local_setup_socket);
2647 		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2648 		break;
2649 	default:
2650 		ret = ERR_PTR(-EAFNOSUPPORT);
2651 		goto out_err;
2652 	}
2653 
2654 	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2655 			xprt->address_strings[RPC_DISPLAY_ADDR]);
2656 
2657 	if (try_module_get(THIS_MODULE))
2658 		return xprt;
2659 	ret = ERR_PTR(-EINVAL);
2660 out_err:
2661 	xprt_free(xprt);
2662 	return ret;
2663 }
2664 
2665 static const struct rpc_timeout xs_udp_default_timeout = {
2666 	.to_initval = 5 * HZ,
2667 	.to_maxval = 30 * HZ,
2668 	.to_increment = 5 * HZ,
2669 	.to_retries = 5,
2670 };
2671 
2672 /**
2673  * xs_setup_udp - Set up transport to use a UDP socket
2674  * @args: rpc transport creation arguments
2675  *
2676  */
2677 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2678 {
2679 	struct sockaddr *addr = args->dstaddr;
2680 	struct rpc_xprt *xprt;
2681 	struct sock_xprt *transport;
2682 	struct rpc_xprt *ret;
2683 
2684 	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2685 			xprt_udp_slot_table_entries);
2686 	if (IS_ERR(xprt))
2687 		return xprt;
2688 	transport = container_of(xprt, struct sock_xprt, xprt);
2689 
2690 	xprt->prot = IPPROTO_UDP;
2691 	xprt->tsh_size = 0;
2692 	/* XXX: header size can vary due to auth type, IPv6, etc. */
2693 	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2694 
2695 	xprt->bind_timeout = XS_BIND_TO;
2696 	xprt->reestablish_timeout = XS_UDP_REEST_TO;
2697 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2698 
2699 	xprt->ops = &xs_udp_ops;
2700 
2701 	xprt->timeout = &xs_udp_default_timeout;
2702 
2703 	switch (addr->sa_family) {
2704 	case AF_INET:
2705 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2706 			xprt_set_bound(xprt);
2707 
2708 		INIT_DELAYED_WORK(&transport->connect_worker,
2709 					xs_udp_setup_socket);
2710 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2711 		break;
2712 	case AF_INET6:
2713 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2714 			xprt_set_bound(xprt);
2715 
2716 		INIT_DELAYED_WORK(&transport->connect_worker,
2717 					xs_udp_setup_socket);
2718 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2719 		break;
2720 	default:
2721 		ret = ERR_PTR(-EAFNOSUPPORT);
2722 		goto out_err;
2723 	}
2724 
2725 	if (xprt_bound(xprt))
2726 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2727 				xprt->address_strings[RPC_DISPLAY_ADDR],
2728 				xprt->address_strings[RPC_DISPLAY_PORT],
2729 				xprt->address_strings[RPC_DISPLAY_PROTO]);
2730 	else
2731 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2732 				xprt->address_strings[RPC_DISPLAY_ADDR],
2733 				xprt->address_strings[RPC_DISPLAY_PROTO]);
2734 
2735 	if (try_module_get(THIS_MODULE))
2736 		return xprt;
2737 	ret = ERR_PTR(-EINVAL);
2738 out_err:
2739 	xprt_free(xprt);
2740 	return ret;
2741 }
2742 
2743 static const struct rpc_timeout xs_tcp_default_timeout = {
2744 	.to_initval = 60 * HZ,
2745 	.to_maxval = 60 * HZ,
2746 	.to_retries = 2,
2747 };
2748 
2749 /**
2750  * xs_setup_tcp - Set up transport to use a TCP socket
2751  * @args: rpc transport creation arguments
2752  *
2753  */
2754 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2755 {
2756 	struct sockaddr *addr = args->dstaddr;
2757 	struct rpc_xprt *xprt;
2758 	struct sock_xprt *transport;
2759 	struct rpc_xprt *ret;
2760 
2761 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2762 			xprt_max_tcp_slot_table_entries);
2763 	if (IS_ERR(xprt))
2764 		return xprt;
2765 	transport = container_of(xprt, struct sock_xprt, xprt);
2766 
2767 	xprt->prot = IPPROTO_TCP;
2768 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2769 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2770 
2771 	xprt->bind_timeout = XS_BIND_TO;
2772 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2773 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2774 
2775 	xprt->ops = &xs_tcp_ops;
2776 	xprt->timeout = &xs_tcp_default_timeout;
2777 
2778 	switch (addr->sa_family) {
2779 	case AF_INET:
2780 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2781 			xprt_set_bound(xprt);
2782 
2783 		INIT_DELAYED_WORK(&transport->connect_worker,
2784 					xs_tcp_setup_socket);
2785 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2786 		break;
2787 	case AF_INET6:
2788 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2789 			xprt_set_bound(xprt);
2790 
2791 		INIT_DELAYED_WORK(&transport->connect_worker,
2792 					xs_tcp_setup_socket);
2793 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2794 		break;
2795 	default:
2796 		ret = ERR_PTR(-EAFNOSUPPORT);
2797 		goto out_err;
2798 	}
2799 
2800 	if (xprt_bound(xprt))
2801 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2802 				xprt->address_strings[RPC_DISPLAY_ADDR],
2803 				xprt->address_strings[RPC_DISPLAY_PORT],
2804 				xprt->address_strings[RPC_DISPLAY_PROTO]);
2805 	else
2806 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2807 				xprt->address_strings[RPC_DISPLAY_ADDR],
2808 				xprt->address_strings[RPC_DISPLAY_PROTO]);
2809 
2810 
2811 	if (try_module_get(THIS_MODULE))
2812 		return xprt;
2813 	ret = ERR_PTR(-EINVAL);
2814 out_err:
2815 	xprt_free(xprt);
2816 	return ret;
2817 }
2818 
2819 /**
2820  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2821  * @args: rpc transport creation arguments
2822  *
2823  */
2824 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2825 {
2826 	struct sockaddr *addr = args->dstaddr;
2827 	struct rpc_xprt *xprt;
2828 	struct sock_xprt *transport;
2829 	struct svc_sock *bc_sock;
2830 	struct rpc_xprt *ret;
2831 
2832 	if (args->bc_xprt->xpt_bc_xprt) {
2833 		/*
2834 		 * This server connection already has a backchannel
2835 		 * export; we can't create a new one, as we wouldn't be
2836 		 * able to match replies based on xid any more.  So,
2837 		 * reuse the already-existing one:
2838 		 */
2839 		 return args->bc_xprt->xpt_bc_xprt;
2840 	}
2841 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2842 			xprt_tcp_slot_table_entries);
2843 	if (IS_ERR(xprt))
2844 		return xprt;
2845 	transport = container_of(xprt, struct sock_xprt, xprt);
2846 
2847 	xprt->prot = IPPROTO_TCP;
2848 	xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2849 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2850 	xprt->timeout = &xs_tcp_default_timeout;
2851 
2852 	/* backchannel */
2853 	xprt_set_bound(xprt);
2854 	xprt->bind_timeout = 0;
2855 	xprt->reestablish_timeout = 0;
2856 	xprt->idle_timeout = 0;
2857 
2858 	xprt->ops = &bc_tcp_ops;
2859 
2860 	switch (addr->sa_family) {
2861 	case AF_INET:
2862 		xs_format_peer_addresses(xprt, "tcp",
2863 					 RPCBIND_NETID_TCP);
2864 		break;
2865 	case AF_INET6:
2866 		xs_format_peer_addresses(xprt, "tcp",
2867 				   RPCBIND_NETID_TCP6);
2868 		break;
2869 	default:
2870 		ret = ERR_PTR(-EAFNOSUPPORT);
2871 		goto out_err;
2872 	}
2873 
2874 	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2875 			xprt->address_strings[RPC_DISPLAY_ADDR],
2876 			xprt->address_strings[RPC_DISPLAY_PORT],
2877 			xprt->address_strings[RPC_DISPLAY_PROTO]);
2878 
2879 	/*
2880 	 * Once we've associated a backchannel xprt with a connection,
2881 	 * we want to keep it around as long as long as the connection
2882 	 * lasts, in case we need to start using it for a backchannel
2883 	 * again; this reference won't be dropped until bc_xprt is
2884 	 * destroyed.
2885 	 */
2886 	xprt_get(xprt);
2887 	args->bc_xprt->xpt_bc_xprt = xprt;
2888 	xprt->bc_xprt = args->bc_xprt;
2889 	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2890 	transport->sock = bc_sock->sk_sock;
2891 	transport->inet = bc_sock->sk_sk;
2892 
2893 	/*
2894 	 * Since we don't want connections for the backchannel, we set
2895 	 * the xprt status to connected
2896 	 */
2897 	xprt_set_connected(xprt);
2898 
2899 
2900 	if (try_module_get(THIS_MODULE))
2901 		return xprt;
2902 	xprt_put(xprt);
2903 	ret = ERR_PTR(-EINVAL);
2904 out_err:
2905 	xprt_free(xprt);
2906 	return ret;
2907 }
2908 
2909 static struct xprt_class	xs_local_transport = {
2910 	.list		= LIST_HEAD_INIT(xs_local_transport.list),
2911 	.name		= "named UNIX socket",
2912 	.owner		= THIS_MODULE,
2913 	.ident		= XPRT_TRANSPORT_LOCAL,
2914 	.setup		= xs_setup_local,
2915 };
2916 
2917 static struct xprt_class	xs_udp_transport = {
2918 	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
2919 	.name		= "udp",
2920 	.owner		= THIS_MODULE,
2921 	.ident		= XPRT_TRANSPORT_UDP,
2922 	.setup		= xs_setup_udp,
2923 };
2924 
2925 static struct xprt_class	xs_tcp_transport = {
2926 	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
2927 	.name		= "tcp",
2928 	.owner		= THIS_MODULE,
2929 	.ident		= XPRT_TRANSPORT_TCP,
2930 	.setup		= xs_setup_tcp,
2931 };
2932 
2933 static struct xprt_class	xs_bc_tcp_transport = {
2934 	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2935 	.name		= "tcp NFSv4.1 backchannel",
2936 	.owner		= THIS_MODULE,
2937 	.ident		= XPRT_TRANSPORT_BC_TCP,
2938 	.setup		= xs_setup_bc_tcp,
2939 };
2940 
2941 /**
2942  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2943  *
2944  */
2945 int init_socket_xprt(void)
2946 {
2947 #ifdef RPC_DEBUG
2948 	if (!sunrpc_table_header)
2949 		sunrpc_table_header = register_sysctl_table(sunrpc_table);
2950 #endif
2951 
2952 	xprt_register_transport(&xs_local_transport);
2953 	xprt_register_transport(&xs_udp_transport);
2954 	xprt_register_transport(&xs_tcp_transport);
2955 	xprt_register_transport(&xs_bc_tcp_transport);
2956 
2957 	return 0;
2958 }
2959 
2960 /**
2961  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2962  *
2963  */
2964 void cleanup_socket_xprt(void)
2965 {
2966 #ifdef RPC_DEBUG
2967 	if (sunrpc_table_header) {
2968 		unregister_sysctl_table(sunrpc_table_header);
2969 		sunrpc_table_header = NULL;
2970 	}
2971 #endif
2972 
2973 	xprt_unregister_transport(&xs_local_transport);
2974 	xprt_unregister_transport(&xs_udp_transport);
2975 	xprt_unregister_transport(&xs_tcp_transport);
2976 	xprt_unregister_transport(&xs_bc_tcp_transport);
2977 }
2978 
2979 static int param_set_uint_minmax(const char *val,
2980 		const struct kernel_param *kp,
2981 		unsigned int min, unsigned int max)
2982 {
2983 	unsigned long num;
2984 	int ret;
2985 
2986 	if (!val)
2987 		return -EINVAL;
2988 	ret = strict_strtoul(val, 0, &num);
2989 	if (ret == -EINVAL || num < min || num > max)
2990 		return -EINVAL;
2991 	*((unsigned int *)kp->arg) = num;
2992 	return 0;
2993 }
2994 
2995 static int param_set_portnr(const char *val, const struct kernel_param *kp)
2996 {
2997 	return param_set_uint_minmax(val, kp,
2998 			RPC_MIN_RESVPORT,
2999 			RPC_MAX_RESVPORT);
3000 }
3001 
3002 static struct kernel_param_ops param_ops_portnr = {
3003 	.set = param_set_portnr,
3004 	.get = param_get_uint,
3005 };
3006 
3007 #define param_check_portnr(name, p) \
3008 	__param_check(name, p, unsigned int);
3009 
3010 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3011 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3012 
3013 static int param_set_slot_table_size(const char *val,
3014 				     const struct kernel_param *kp)
3015 {
3016 	return param_set_uint_minmax(val, kp,
3017 			RPC_MIN_SLOT_TABLE,
3018 			RPC_MAX_SLOT_TABLE);
3019 }
3020 
3021 static struct kernel_param_ops param_ops_slot_table_size = {
3022 	.set = param_set_slot_table_size,
3023 	.get = param_get_uint,
3024 };
3025 
3026 #define param_check_slot_table_size(name, p) \
3027 	__param_check(name, p, unsigned int);
3028 
3029 static int param_set_max_slot_table_size(const char *val,
3030 				     const struct kernel_param *kp)
3031 {
3032 	return param_set_uint_minmax(val, kp,
3033 			RPC_MIN_SLOT_TABLE,
3034 			RPC_MAX_SLOT_TABLE_LIMIT);
3035 }
3036 
3037 static struct kernel_param_ops param_ops_max_slot_table_size = {
3038 	.set = param_set_max_slot_table_size,
3039 	.get = param_get_uint,
3040 };
3041 
3042 #define param_check_max_slot_table_size(name, p) \
3043 	__param_check(name, p, unsigned int);
3044 
3045 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3046 		   slot_table_size, 0644);
3047 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3048 		   max_slot_table_size, 0644);
3049 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3050 		   slot_table_size, 0644);
3051 
3052