xref: /openbmc/linux/net/sunrpc/xprtsock.c (revision 3dc4b6fb)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/net/sunrpc/xprtsock.c
4  *
5  * Client-side transport implementation for sockets.
6  *
7  * TCP callback races fixes (C) 1998 Red Hat
8  * TCP send fixes (C) 1998 Red Hat
9  * TCP NFS related read + write fixes
10  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
11  *
12  * Rewrite of larges part of the code in order to stabilize TCP stuff.
13  * Fix behaviour when socket buffer is full.
14  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
15  *
16  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
17  *
18  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
19  *   <gilles.quillard@bull.net>
20  */
21 
22 #include <linux/types.h>
23 #include <linux/string.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/pagemap.h>
28 #include <linux/errno.h>
29 #include <linux/socket.h>
30 #include <linux/in.h>
31 #include <linux/net.h>
32 #include <linux/mm.h>
33 #include <linux/un.h>
34 #include <linux/udp.h>
35 #include <linux/tcp.h>
36 #include <linux/sunrpc/clnt.h>
37 #include <linux/sunrpc/addr.h>
38 #include <linux/sunrpc/sched.h>
39 #include <linux/sunrpc/svcsock.h>
40 #include <linux/sunrpc/xprtsock.h>
41 #include <linux/file.h>
42 #ifdef CONFIG_SUNRPC_BACKCHANNEL
43 #include <linux/sunrpc/bc_xprt.h>
44 #endif
45 
46 #include <net/sock.h>
47 #include <net/checksum.h>
48 #include <net/udp.h>
49 #include <net/tcp.h>
50 #include <linux/bvec.h>
51 #include <linux/highmem.h>
52 #include <linux/uio.h>
53 #include <linux/sched/mm.h>
54 
55 #include <trace/events/sunrpc.h>
56 
57 #include "sunrpc.h"
58 
59 static void xs_close(struct rpc_xprt *xprt);
60 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
61 		struct socket *sock);
62 
63 /*
64  * xprtsock tunables
65  */
66 static unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
67 static unsigned int xprt_tcp_slot_table_entries = RPC_MIN_SLOT_TABLE;
68 static unsigned int xprt_max_tcp_slot_table_entries = RPC_MAX_SLOT_TABLE;
69 
70 static unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
71 static unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
72 
73 #define XS_TCP_LINGER_TO	(15U * HZ)
74 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
75 
76 /*
77  * We can register our own files under /proc/sys/sunrpc by
78  * calling register_sysctl_table() again.  The files in that
79  * directory become the union of all files registered there.
80  *
81  * We simply need to make sure that we don't collide with
82  * someone else's file names!
83  */
84 
85 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
86 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
87 static unsigned int max_tcp_slot_table_limit = RPC_MAX_SLOT_TABLE_LIMIT;
88 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
89 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
90 
91 static struct ctl_table_header *sunrpc_table_header;
92 
93 /*
94  * FIXME: changing the UDP slot table size should also resize the UDP
95  *        socket buffers for existing UDP transports
96  */
97 static struct ctl_table xs_tunables_table[] = {
98 	{
99 		.procname	= "udp_slot_table_entries",
100 		.data		= &xprt_udp_slot_table_entries,
101 		.maxlen		= sizeof(unsigned int),
102 		.mode		= 0644,
103 		.proc_handler	= proc_dointvec_minmax,
104 		.extra1		= &min_slot_table_size,
105 		.extra2		= &max_slot_table_size
106 	},
107 	{
108 		.procname	= "tcp_slot_table_entries",
109 		.data		= &xprt_tcp_slot_table_entries,
110 		.maxlen		= sizeof(unsigned int),
111 		.mode		= 0644,
112 		.proc_handler	= proc_dointvec_minmax,
113 		.extra1		= &min_slot_table_size,
114 		.extra2		= &max_slot_table_size
115 	},
116 	{
117 		.procname	= "tcp_max_slot_table_entries",
118 		.data		= &xprt_max_tcp_slot_table_entries,
119 		.maxlen		= sizeof(unsigned int),
120 		.mode		= 0644,
121 		.proc_handler	= proc_dointvec_minmax,
122 		.extra1		= &min_slot_table_size,
123 		.extra2		= &max_tcp_slot_table_limit
124 	},
125 	{
126 		.procname	= "min_resvport",
127 		.data		= &xprt_min_resvport,
128 		.maxlen		= sizeof(unsigned int),
129 		.mode		= 0644,
130 		.proc_handler	= proc_dointvec_minmax,
131 		.extra1		= &xprt_min_resvport_limit,
132 		.extra2		= &xprt_max_resvport_limit
133 	},
134 	{
135 		.procname	= "max_resvport",
136 		.data		= &xprt_max_resvport,
137 		.maxlen		= sizeof(unsigned int),
138 		.mode		= 0644,
139 		.proc_handler	= proc_dointvec_minmax,
140 		.extra1		= &xprt_min_resvport_limit,
141 		.extra2		= &xprt_max_resvport_limit
142 	},
143 	{
144 		.procname	= "tcp_fin_timeout",
145 		.data		= &xs_tcp_fin_timeout,
146 		.maxlen		= sizeof(xs_tcp_fin_timeout),
147 		.mode		= 0644,
148 		.proc_handler	= proc_dointvec_jiffies,
149 	},
150 	{ },
151 };
152 
153 static struct ctl_table sunrpc_table[] = {
154 	{
155 		.procname	= "sunrpc",
156 		.mode		= 0555,
157 		.child		= xs_tunables_table
158 	},
159 	{ },
160 };
161 
162 /*
163  * Wait duration for a reply from the RPC portmapper.
164  */
165 #define XS_BIND_TO		(60U * HZ)
166 
167 /*
168  * Delay if a UDP socket connect error occurs.  This is most likely some
169  * kind of resource problem on the local host.
170  */
171 #define XS_UDP_REEST_TO		(2U * HZ)
172 
173 /*
174  * The reestablish timeout allows clients to delay for a bit before attempting
175  * to reconnect to a server that just dropped our connection.
176  *
177  * We implement an exponential backoff when trying to reestablish a TCP
178  * transport connection with the server.  Some servers like to drop a TCP
179  * connection when they are overworked, so we start with a short timeout and
180  * increase over time if the server is down or not responding.
181  */
182 #define XS_TCP_INIT_REEST_TO	(3U * HZ)
183 
184 /*
185  * TCP idle timeout; client drops the transport socket if it is idle
186  * for this long.  Note that we also timeout UDP sockets to prevent
187  * holding port numbers when there is no RPC traffic.
188  */
189 #define XS_IDLE_DISC_TO		(5U * 60 * HZ)
190 
191 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
192 # undef  RPC_DEBUG_DATA
193 # define RPCDBG_FACILITY	RPCDBG_TRANS
194 #endif
195 
196 #ifdef RPC_DEBUG_DATA
197 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
198 {
199 	u8 *buf = (u8 *) packet;
200 	int j;
201 
202 	dprintk("RPC:       %s\n", msg);
203 	for (j = 0; j < count && j < 128; j += 4) {
204 		if (!(j & 31)) {
205 			if (j)
206 				dprintk("\n");
207 			dprintk("0x%04x ", j);
208 		}
209 		dprintk("%02x%02x%02x%02x ",
210 			buf[j], buf[j+1], buf[j+2], buf[j+3]);
211 	}
212 	dprintk("\n");
213 }
214 #else
215 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
216 {
217 	/* NOP */
218 }
219 #endif
220 
221 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
222 {
223 	return (struct rpc_xprt *) sk->sk_user_data;
224 }
225 
226 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
227 {
228 	return (struct sockaddr *) &xprt->addr;
229 }
230 
231 static inline struct sockaddr_un *xs_addr_un(struct rpc_xprt *xprt)
232 {
233 	return (struct sockaddr_un *) &xprt->addr;
234 }
235 
236 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
237 {
238 	return (struct sockaddr_in *) &xprt->addr;
239 }
240 
241 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
242 {
243 	return (struct sockaddr_in6 *) &xprt->addr;
244 }
245 
246 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
247 {
248 	struct sockaddr *sap = xs_addr(xprt);
249 	struct sockaddr_in6 *sin6;
250 	struct sockaddr_in *sin;
251 	struct sockaddr_un *sun;
252 	char buf[128];
253 
254 	switch (sap->sa_family) {
255 	case AF_LOCAL:
256 		sun = xs_addr_un(xprt);
257 		strlcpy(buf, sun->sun_path, sizeof(buf));
258 		xprt->address_strings[RPC_DISPLAY_ADDR] =
259 						kstrdup(buf, GFP_KERNEL);
260 		break;
261 	case AF_INET:
262 		(void)rpc_ntop(sap, buf, sizeof(buf));
263 		xprt->address_strings[RPC_DISPLAY_ADDR] =
264 						kstrdup(buf, GFP_KERNEL);
265 		sin = xs_addr_in(xprt);
266 		snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
267 		break;
268 	case AF_INET6:
269 		(void)rpc_ntop(sap, buf, sizeof(buf));
270 		xprt->address_strings[RPC_DISPLAY_ADDR] =
271 						kstrdup(buf, GFP_KERNEL);
272 		sin6 = xs_addr_in6(xprt);
273 		snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
274 		break;
275 	default:
276 		BUG();
277 	}
278 
279 	xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
280 }
281 
282 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
283 {
284 	struct sockaddr *sap = xs_addr(xprt);
285 	char buf[128];
286 
287 	snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
288 	xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
289 
290 	snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
291 	xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
292 }
293 
294 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
295 				     const char *protocol,
296 				     const char *netid)
297 {
298 	xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
299 	xprt->address_strings[RPC_DISPLAY_NETID] = netid;
300 	xs_format_common_peer_addresses(xprt);
301 	xs_format_common_peer_ports(xprt);
302 }
303 
304 static void xs_update_peer_port(struct rpc_xprt *xprt)
305 {
306 	kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
307 	kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
308 
309 	xs_format_common_peer_ports(xprt);
310 }
311 
312 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
313 {
314 	unsigned int i;
315 
316 	for (i = 0; i < RPC_DISPLAY_MAX; i++)
317 		switch (i) {
318 		case RPC_DISPLAY_PROTO:
319 		case RPC_DISPLAY_NETID:
320 			continue;
321 		default:
322 			kfree(xprt->address_strings[i]);
323 		}
324 }
325 
326 static size_t
327 xs_alloc_sparse_pages(struct xdr_buf *buf, size_t want, gfp_t gfp)
328 {
329 	size_t i,n;
330 
331 	if (!want || !(buf->flags & XDRBUF_SPARSE_PAGES))
332 		return want;
333 	n = (buf->page_base + want + PAGE_SIZE - 1) >> PAGE_SHIFT;
334 	for (i = 0; i < n; i++) {
335 		if (buf->pages[i])
336 			continue;
337 		buf->bvec[i].bv_page = buf->pages[i] = alloc_page(gfp);
338 		if (!buf->pages[i]) {
339 			i *= PAGE_SIZE;
340 			return i > buf->page_base ? i - buf->page_base : 0;
341 		}
342 	}
343 	return want;
344 }
345 
346 static ssize_t
347 xs_sock_recvmsg(struct socket *sock, struct msghdr *msg, int flags, size_t seek)
348 {
349 	ssize_t ret;
350 	if (seek != 0)
351 		iov_iter_advance(&msg->msg_iter, seek);
352 	ret = sock_recvmsg(sock, msg, flags);
353 	return ret > 0 ? ret + seek : ret;
354 }
355 
356 static ssize_t
357 xs_read_kvec(struct socket *sock, struct msghdr *msg, int flags,
358 		struct kvec *kvec, size_t count, size_t seek)
359 {
360 	iov_iter_kvec(&msg->msg_iter, READ, kvec, 1, count);
361 	return xs_sock_recvmsg(sock, msg, flags, seek);
362 }
363 
364 static ssize_t
365 xs_read_bvec(struct socket *sock, struct msghdr *msg, int flags,
366 		struct bio_vec *bvec, unsigned long nr, size_t count,
367 		size_t seek)
368 {
369 	iov_iter_bvec(&msg->msg_iter, READ, bvec, nr, count);
370 	return xs_sock_recvmsg(sock, msg, flags, seek);
371 }
372 
373 static ssize_t
374 xs_read_discard(struct socket *sock, struct msghdr *msg, int flags,
375 		size_t count)
376 {
377 	iov_iter_discard(&msg->msg_iter, READ, count);
378 	return sock_recvmsg(sock, msg, flags);
379 }
380 
381 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
382 static void
383 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
384 {
385 	struct bvec_iter bi = {
386 		.bi_size = count,
387 	};
388 	struct bio_vec bv;
389 
390 	bvec_iter_advance(bvec, &bi, seek & PAGE_MASK);
391 	for_each_bvec(bv, bvec, bi, bi)
392 		flush_dcache_page(bv.bv_page);
393 }
394 #else
395 static inline void
396 xs_flush_bvec(const struct bio_vec *bvec, size_t count, size_t seek)
397 {
398 }
399 #endif
400 
401 static ssize_t
402 xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
403 		struct xdr_buf *buf, size_t count, size_t seek, size_t *read)
404 {
405 	size_t want, seek_init = seek, offset = 0;
406 	ssize_t ret;
407 
408 	want = min_t(size_t, count, buf->head[0].iov_len);
409 	if (seek < want) {
410 		ret = xs_read_kvec(sock, msg, flags, &buf->head[0], want, seek);
411 		if (ret <= 0)
412 			goto sock_err;
413 		offset += ret;
414 		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
415 			goto out;
416 		if (ret != want)
417 			goto out;
418 		seek = 0;
419 	} else {
420 		seek -= want;
421 		offset += want;
422 	}
423 
424 	want = xs_alloc_sparse_pages(buf,
425 			min_t(size_t, count - offset, buf->page_len),
426 			GFP_KERNEL);
427 	if (seek < want) {
428 		ret = xs_read_bvec(sock, msg, flags, buf->bvec,
429 				xdr_buf_pagecount(buf),
430 				want + buf->page_base,
431 				seek + buf->page_base);
432 		if (ret <= 0)
433 			goto sock_err;
434 		xs_flush_bvec(buf->bvec, ret, seek + buf->page_base);
435 		offset += ret - buf->page_base;
436 		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
437 			goto out;
438 		if (ret != want)
439 			goto out;
440 		seek = 0;
441 	} else {
442 		seek -= want;
443 		offset += want;
444 	}
445 
446 	want = min_t(size_t, count - offset, buf->tail[0].iov_len);
447 	if (seek < want) {
448 		ret = xs_read_kvec(sock, msg, flags, &buf->tail[0], want, seek);
449 		if (ret <= 0)
450 			goto sock_err;
451 		offset += ret;
452 		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
453 			goto out;
454 		if (ret != want)
455 			goto out;
456 	} else if (offset < seek_init)
457 		offset = seek_init;
458 	ret = -EMSGSIZE;
459 out:
460 	*read = offset - seek_init;
461 	return ret;
462 sock_err:
463 	offset += seek;
464 	goto out;
465 }
466 
467 static void
468 xs_read_header(struct sock_xprt *transport, struct xdr_buf *buf)
469 {
470 	if (!transport->recv.copied) {
471 		if (buf->head[0].iov_len >= transport->recv.offset)
472 			memcpy(buf->head[0].iov_base,
473 					&transport->recv.xid,
474 					transport->recv.offset);
475 		transport->recv.copied = transport->recv.offset;
476 	}
477 }
478 
479 static bool
480 xs_read_stream_request_done(struct sock_xprt *transport)
481 {
482 	return transport->recv.fraghdr & cpu_to_be32(RPC_LAST_STREAM_FRAGMENT);
483 }
484 
485 static void
486 xs_read_stream_check_eor(struct sock_xprt *transport,
487 		struct msghdr *msg)
488 {
489 	if (xs_read_stream_request_done(transport))
490 		msg->msg_flags |= MSG_EOR;
491 }
492 
493 static ssize_t
494 xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
495 		int flags, struct rpc_rqst *req)
496 {
497 	struct xdr_buf *buf = &req->rq_private_buf;
498 	size_t want, uninitialized_var(read);
499 	ssize_t uninitialized_var(ret);
500 
501 	xs_read_header(transport, buf);
502 
503 	want = transport->recv.len - transport->recv.offset;
504 	if (want != 0) {
505 		ret = xs_read_xdr_buf(transport->sock, msg, flags, buf,
506 				transport->recv.copied + want,
507 				transport->recv.copied,
508 				&read);
509 		transport->recv.offset += read;
510 		transport->recv.copied += read;
511 	}
512 
513 	if (transport->recv.offset == transport->recv.len)
514 		xs_read_stream_check_eor(transport, msg);
515 
516 	if (want == 0)
517 		return 0;
518 
519 	switch (ret) {
520 	default:
521 		break;
522 	case -EFAULT:
523 	case -EMSGSIZE:
524 		msg->msg_flags |= MSG_TRUNC;
525 		return read;
526 	case 0:
527 		return -ESHUTDOWN;
528 	}
529 	return ret < 0 ? ret : read;
530 }
531 
532 static size_t
533 xs_read_stream_headersize(bool isfrag)
534 {
535 	if (isfrag)
536 		return sizeof(__be32);
537 	return 3 * sizeof(__be32);
538 }
539 
540 static ssize_t
541 xs_read_stream_header(struct sock_xprt *transport, struct msghdr *msg,
542 		int flags, size_t want, size_t seek)
543 {
544 	struct kvec kvec = {
545 		.iov_base = &transport->recv.fraghdr,
546 		.iov_len = want,
547 	};
548 	return xs_read_kvec(transport->sock, msg, flags, &kvec, want, seek);
549 }
550 
551 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
552 static ssize_t
553 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
554 {
555 	struct rpc_xprt *xprt = &transport->xprt;
556 	struct rpc_rqst *req;
557 	ssize_t ret;
558 
559 	/* Look up and lock the request corresponding to the given XID */
560 	req = xprt_lookup_bc_request(xprt, transport->recv.xid);
561 	if (!req) {
562 		printk(KERN_WARNING "Callback slot table overflowed\n");
563 		return -ESHUTDOWN;
564 	}
565 	if (transport->recv.copied && !req->rq_private_buf.len)
566 		return -ESHUTDOWN;
567 
568 	ret = xs_read_stream_request(transport, msg, flags, req);
569 	if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
570 		xprt_complete_bc_request(req, transport->recv.copied);
571 	else
572 		req->rq_private_buf.len = transport->recv.copied;
573 
574 	return ret;
575 }
576 #else /* CONFIG_SUNRPC_BACKCHANNEL */
577 static ssize_t
578 xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
579 {
580 	return -ESHUTDOWN;
581 }
582 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
583 
584 static ssize_t
585 xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
586 {
587 	struct rpc_xprt *xprt = &transport->xprt;
588 	struct rpc_rqst *req;
589 	ssize_t ret = 0;
590 
591 	/* Look up and lock the request corresponding to the given XID */
592 	spin_lock(&xprt->queue_lock);
593 	req = xprt_lookup_rqst(xprt, transport->recv.xid);
594 	if (!req || (transport->recv.copied && !req->rq_private_buf.len)) {
595 		msg->msg_flags |= MSG_TRUNC;
596 		goto out;
597 	}
598 	xprt_pin_rqst(req);
599 	spin_unlock(&xprt->queue_lock);
600 
601 	ret = xs_read_stream_request(transport, msg, flags, req);
602 
603 	spin_lock(&xprt->queue_lock);
604 	if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
605 		xprt_complete_rqst(req->rq_task, transport->recv.copied);
606 	else
607 		req->rq_private_buf.len = transport->recv.copied;
608 	xprt_unpin_rqst(req);
609 out:
610 	spin_unlock(&xprt->queue_lock);
611 	return ret;
612 }
613 
614 static ssize_t
615 xs_read_stream(struct sock_xprt *transport, int flags)
616 {
617 	struct msghdr msg = { 0 };
618 	size_t want, read = 0;
619 	ssize_t ret = 0;
620 
621 	if (transport->recv.len == 0) {
622 		want = xs_read_stream_headersize(transport->recv.copied != 0);
623 		ret = xs_read_stream_header(transport, &msg, flags, want,
624 				transport->recv.offset);
625 		if (ret <= 0)
626 			goto out_err;
627 		transport->recv.offset = ret;
628 		if (transport->recv.offset != want)
629 			return transport->recv.offset;
630 		transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
631 			RPC_FRAGMENT_SIZE_MASK;
632 		transport->recv.offset -= sizeof(transport->recv.fraghdr);
633 		read = ret;
634 	}
635 
636 	switch (be32_to_cpu(transport->recv.calldir)) {
637 	default:
638 		msg.msg_flags |= MSG_TRUNC;
639 		break;
640 	case RPC_CALL:
641 		ret = xs_read_stream_call(transport, &msg, flags);
642 		break;
643 	case RPC_REPLY:
644 		ret = xs_read_stream_reply(transport, &msg, flags);
645 	}
646 	if (msg.msg_flags & MSG_TRUNC) {
647 		transport->recv.calldir = cpu_to_be32(-1);
648 		transport->recv.copied = -1;
649 	}
650 	if (ret < 0)
651 		goto out_err;
652 	read += ret;
653 	if (transport->recv.offset < transport->recv.len) {
654 		if (!(msg.msg_flags & MSG_TRUNC))
655 			return read;
656 		msg.msg_flags = 0;
657 		ret = xs_read_discard(transport->sock, &msg, flags,
658 				transport->recv.len - transport->recv.offset);
659 		if (ret <= 0)
660 			goto out_err;
661 		transport->recv.offset += ret;
662 		read += ret;
663 		if (transport->recv.offset != transport->recv.len)
664 			return read;
665 	}
666 	if (xs_read_stream_request_done(transport)) {
667 		trace_xs_stream_read_request(transport);
668 		transport->recv.copied = 0;
669 	}
670 	transport->recv.offset = 0;
671 	transport->recv.len = 0;
672 	return read;
673 out_err:
674 	return ret != 0 ? ret : -ESHUTDOWN;
675 }
676 
677 static __poll_t xs_poll_socket(struct sock_xprt *transport)
678 {
679 	return transport->sock->ops->poll(transport->file, transport->sock,
680 			NULL);
681 }
682 
683 static bool xs_poll_socket_readable(struct sock_xprt *transport)
684 {
685 	__poll_t events = xs_poll_socket(transport);
686 
687 	return (events & (EPOLLIN | EPOLLRDNORM)) && !(events & EPOLLRDHUP);
688 }
689 
690 static void xs_poll_check_readable(struct sock_xprt *transport)
691 {
692 
693 	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
694 	if (!xs_poll_socket_readable(transport))
695 		return;
696 	if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
697 		queue_work(xprtiod_workqueue, &transport->recv_worker);
698 }
699 
700 static void xs_stream_data_receive(struct sock_xprt *transport)
701 {
702 	size_t read = 0;
703 	ssize_t ret = 0;
704 
705 	mutex_lock(&transport->recv_mutex);
706 	if (transport->sock == NULL)
707 		goto out;
708 	for (;;) {
709 		ret = xs_read_stream(transport, MSG_DONTWAIT);
710 		if (ret < 0)
711 			break;
712 		read += ret;
713 		cond_resched();
714 	}
715 	if (ret == -ESHUTDOWN)
716 		kernel_sock_shutdown(transport->sock, SHUT_RDWR);
717 	else
718 		xs_poll_check_readable(transport);
719 out:
720 	mutex_unlock(&transport->recv_mutex);
721 	trace_xs_stream_read_data(&transport->xprt, ret, read);
722 }
723 
724 static void xs_stream_data_receive_workfn(struct work_struct *work)
725 {
726 	struct sock_xprt *transport =
727 		container_of(work, struct sock_xprt, recv_worker);
728 	unsigned int pflags = memalloc_nofs_save();
729 
730 	xs_stream_data_receive(transport);
731 	memalloc_nofs_restore(pflags);
732 }
733 
734 static void
735 xs_stream_reset_connect(struct sock_xprt *transport)
736 {
737 	transport->recv.offset = 0;
738 	transport->recv.len = 0;
739 	transport->recv.copied = 0;
740 	transport->xmit.offset = 0;
741 }
742 
743 static void
744 xs_stream_start_connect(struct sock_xprt *transport)
745 {
746 	transport->xprt.stat.connect_count++;
747 	transport->xprt.stat.connect_start = jiffies;
748 }
749 
750 #define XS_SENDMSG_FLAGS	(MSG_DONTWAIT | MSG_NOSIGNAL)
751 
752 static int xs_sendmsg(struct socket *sock, struct msghdr *msg, size_t seek)
753 {
754 	if (seek)
755 		iov_iter_advance(&msg->msg_iter, seek);
756 	return sock_sendmsg(sock, msg);
757 }
758 
759 static int xs_send_kvec(struct socket *sock, struct msghdr *msg, struct kvec *vec, size_t seek)
760 {
761 	iov_iter_kvec(&msg->msg_iter, WRITE, vec, 1, vec->iov_len);
762 	return xs_sendmsg(sock, msg, seek);
763 }
764 
765 static int xs_send_pagedata(struct socket *sock, struct msghdr *msg, struct xdr_buf *xdr, size_t base)
766 {
767 	int err;
768 
769 	err = xdr_alloc_bvec(xdr, GFP_KERNEL);
770 	if (err < 0)
771 		return err;
772 
773 	iov_iter_bvec(&msg->msg_iter, WRITE, xdr->bvec,
774 			xdr_buf_pagecount(xdr),
775 			xdr->page_len + xdr->page_base);
776 	return xs_sendmsg(sock, msg, base + xdr->page_base);
777 }
778 
779 #define xs_record_marker_len() sizeof(rpc_fraghdr)
780 
781 /* Common case:
782  *  - stream transport
783  *  - sending from byte 0 of the message
784  *  - the message is wholly contained in @xdr's head iovec
785  */
786 static int xs_send_rm_and_kvec(struct socket *sock, struct msghdr *msg,
787 		rpc_fraghdr marker, struct kvec *vec, size_t base)
788 {
789 	struct kvec iov[2] = {
790 		[0] = {
791 			.iov_base	= &marker,
792 			.iov_len	= sizeof(marker)
793 		},
794 		[1] = *vec,
795 	};
796 	size_t len = iov[0].iov_len + iov[1].iov_len;
797 
798 	iov_iter_kvec(&msg->msg_iter, WRITE, iov, 2, len);
799 	return xs_sendmsg(sock, msg, base);
800 }
801 
802 /**
803  * xs_sendpages - write pages directly to a socket
804  * @sock: socket to send on
805  * @addr: UDP only -- address of destination
806  * @addrlen: UDP only -- length of destination address
807  * @xdr: buffer containing this request
808  * @base: starting position in the buffer
809  * @rm: stream record marker field
810  * @sent_p: return the total number of bytes successfully queued for sending
811  *
812  */
813 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, rpc_fraghdr rm, int *sent_p)
814 {
815 	struct msghdr msg = {
816 		.msg_name = addr,
817 		.msg_namelen = addrlen,
818 		.msg_flags = XS_SENDMSG_FLAGS | MSG_MORE,
819 	};
820 	unsigned int rmsize = rm ? sizeof(rm) : 0;
821 	unsigned int remainder = rmsize + xdr->len - base;
822 	unsigned int want;
823 	int err = 0;
824 
825 	if (unlikely(!sock))
826 		return -ENOTSOCK;
827 
828 	want = xdr->head[0].iov_len + rmsize;
829 	if (base < want) {
830 		unsigned int len = want - base;
831 		remainder -= len;
832 		if (remainder == 0)
833 			msg.msg_flags &= ~MSG_MORE;
834 		if (rmsize)
835 			err = xs_send_rm_and_kvec(sock, &msg, rm,
836 					&xdr->head[0], base);
837 		else
838 			err = xs_send_kvec(sock, &msg, &xdr->head[0], base);
839 		if (remainder == 0 || err != len)
840 			goto out;
841 		*sent_p += err;
842 		base = 0;
843 	} else
844 		base -= want;
845 
846 	if (base < xdr->page_len) {
847 		unsigned int len = xdr->page_len - base;
848 		remainder -= len;
849 		if (remainder == 0)
850 			msg.msg_flags &= ~MSG_MORE;
851 		err = xs_send_pagedata(sock, &msg, xdr, base);
852 		if (remainder == 0 || err != len)
853 			goto out;
854 		*sent_p += err;
855 		base = 0;
856 	} else
857 		base -= xdr->page_len;
858 
859 	if (base >= xdr->tail[0].iov_len)
860 		return 0;
861 	msg.msg_flags &= ~MSG_MORE;
862 	err = xs_send_kvec(sock, &msg, &xdr->tail[0], base);
863 out:
864 	if (err > 0) {
865 		*sent_p += err;
866 		err = 0;
867 	}
868 	return err;
869 }
870 
871 /**
872  * xs_nospace - handle transmit was incomplete
873  * @req: pointer to RPC request
874  *
875  */
876 static int xs_nospace(struct rpc_rqst *req)
877 {
878 	struct rpc_xprt *xprt = req->rq_xprt;
879 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
880 	struct sock *sk = transport->inet;
881 	int ret = -EAGAIN;
882 
883 	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
884 			req->rq_task->tk_pid,
885 			req->rq_slen - transport->xmit.offset,
886 			req->rq_slen);
887 
888 	/* Protect against races with write_space */
889 	spin_lock(&xprt->transport_lock);
890 
891 	/* Don't race with disconnect */
892 	if (xprt_connected(xprt)) {
893 		/* wait for more buffer space */
894 		sk->sk_write_pending++;
895 		xprt_wait_for_buffer_space(xprt);
896 	} else
897 		ret = -ENOTCONN;
898 
899 	spin_unlock(&xprt->transport_lock);
900 
901 	/* Race breaker in case memory is freed before above code is called */
902 	if (ret == -EAGAIN) {
903 		struct socket_wq *wq;
904 
905 		rcu_read_lock();
906 		wq = rcu_dereference(sk->sk_wq);
907 		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
908 		rcu_read_unlock();
909 
910 		sk->sk_write_space(sk);
911 	}
912 	return ret;
913 }
914 
915 static void
916 xs_stream_prepare_request(struct rpc_rqst *req)
917 {
918 	xdr_free_bvec(&req->rq_rcv_buf);
919 	req->rq_task->tk_status = xdr_alloc_bvec(&req->rq_rcv_buf, GFP_KERNEL);
920 }
921 
922 /*
923  * Determine if the previous message in the stream was aborted before it
924  * could complete transmission.
925  */
926 static bool
927 xs_send_request_was_aborted(struct sock_xprt *transport, struct rpc_rqst *req)
928 {
929 	return transport->xmit.offset != 0 && req->rq_bytes_sent == 0;
930 }
931 
932 /*
933  * Return the stream record marker field for a record of length < 2^31-1
934  */
935 static rpc_fraghdr
936 xs_stream_record_marker(struct xdr_buf *xdr)
937 {
938 	if (!xdr->len)
939 		return 0;
940 	return cpu_to_be32(RPC_LAST_STREAM_FRAGMENT | (u32)xdr->len);
941 }
942 
943 /**
944  * xs_local_send_request - write an RPC request to an AF_LOCAL socket
945  * @req: pointer to RPC request
946  *
947  * Return values:
948  *        0:	The request has been sent
949  *   EAGAIN:	The socket was blocked, please call again later to
950  *		complete the request
951  * ENOTCONN:	Caller needs to invoke connect logic then call again
952  *    other:	Some other error occured, the request was not sent
953  */
954 static int xs_local_send_request(struct rpc_rqst *req)
955 {
956 	struct rpc_xprt *xprt = req->rq_xprt;
957 	struct sock_xprt *transport =
958 				container_of(xprt, struct sock_xprt, xprt);
959 	struct xdr_buf *xdr = &req->rq_snd_buf;
960 	rpc_fraghdr rm = xs_stream_record_marker(xdr);
961 	unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
962 	int status;
963 	int sent = 0;
964 
965 	/* Close the stream if the previous transmission was incomplete */
966 	if (xs_send_request_was_aborted(transport, req)) {
967 		xs_close(xprt);
968 		return -ENOTCONN;
969 	}
970 
971 	xs_pktdump("packet data:",
972 			req->rq_svec->iov_base, req->rq_svec->iov_len);
973 
974 	req->rq_xtime = ktime_get();
975 	status = xs_sendpages(transport->sock, NULL, 0, xdr,
976 			      transport->xmit.offset, rm, &sent);
977 	dprintk("RPC:       %s(%u) = %d\n",
978 			__func__, xdr->len - transport->xmit.offset, status);
979 
980 	if (status == -EAGAIN && sock_writeable(transport->inet))
981 		status = -ENOBUFS;
982 
983 	if (likely(sent > 0) || status == 0) {
984 		transport->xmit.offset += sent;
985 		req->rq_bytes_sent = transport->xmit.offset;
986 		if (likely(req->rq_bytes_sent >= msglen)) {
987 			req->rq_xmit_bytes_sent += transport->xmit.offset;
988 			transport->xmit.offset = 0;
989 			return 0;
990 		}
991 		status = -EAGAIN;
992 	}
993 
994 	switch (status) {
995 	case -ENOBUFS:
996 		break;
997 	case -EAGAIN:
998 		status = xs_nospace(req);
999 		break;
1000 	default:
1001 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1002 			-status);
1003 		/* fall through */
1004 	case -EPIPE:
1005 		xs_close(xprt);
1006 		status = -ENOTCONN;
1007 	}
1008 
1009 	return status;
1010 }
1011 
1012 /**
1013  * xs_udp_send_request - write an RPC request to a UDP socket
1014  * @req: pointer to RPC request
1015  *
1016  * Return values:
1017  *        0:	The request has been sent
1018  *   EAGAIN:	The socket was blocked, please call again later to
1019  *		complete the request
1020  * ENOTCONN:	Caller needs to invoke connect logic then call again
1021  *    other:	Some other error occurred, the request was not sent
1022  */
1023 static int xs_udp_send_request(struct rpc_rqst *req)
1024 {
1025 	struct rpc_xprt *xprt = req->rq_xprt;
1026 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1027 	struct xdr_buf *xdr = &req->rq_snd_buf;
1028 	int sent = 0;
1029 	int status;
1030 
1031 	xs_pktdump("packet data:",
1032 				req->rq_svec->iov_base,
1033 				req->rq_svec->iov_len);
1034 
1035 	if (!xprt_bound(xprt))
1036 		return -ENOTCONN;
1037 
1038 	if (!xprt_request_get_cong(xprt, req))
1039 		return -EBADSLT;
1040 
1041 	req->rq_xtime = ktime_get();
1042 	status = xs_sendpages(transport->sock, xs_addr(xprt), xprt->addrlen,
1043 			      xdr, 0, 0, &sent);
1044 
1045 	dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
1046 			xdr->len, status);
1047 
1048 	/* firewall is blocking us, don't return -EAGAIN or we end up looping */
1049 	if (status == -EPERM)
1050 		goto process_status;
1051 
1052 	if (status == -EAGAIN && sock_writeable(transport->inet))
1053 		status = -ENOBUFS;
1054 
1055 	if (sent > 0 || status == 0) {
1056 		req->rq_xmit_bytes_sent += sent;
1057 		if (sent >= req->rq_slen)
1058 			return 0;
1059 		/* Still some bytes left; set up for a retry later. */
1060 		status = -EAGAIN;
1061 	}
1062 
1063 process_status:
1064 	switch (status) {
1065 	case -ENOTSOCK:
1066 		status = -ENOTCONN;
1067 		/* Should we call xs_close() here? */
1068 		break;
1069 	case -EAGAIN:
1070 		status = xs_nospace(req);
1071 		break;
1072 	case -ENETUNREACH:
1073 	case -ENOBUFS:
1074 	case -EPIPE:
1075 	case -ECONNREFUSED:
1076 	case -EPERM:
1077 		/* When the server has died, an ICMP port unreachable message
1078 		 * prompts ECONNREFUSED. */
1079 		break;
1080 	default:
1081 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1082 			-status);
1083 	}
1084 
1085 	return status;
1086 }
1087 
1088 /**
1089  * xs_tcp_send_request - write an RPC request to a TCP socket
1090  * @req: pointer to RPC request
1091  *
1092  * Return values:
1093  *        0:	The request has been sent
1094  *   EAGAIN:	The socket was blocked, please call again later to
1095  *		complete the request
1096  * ENOTCONN:	Caller needs to invoke connect logic then call again
1097  *    other:	Some other error occurred, the request was not sent
1098  *
1099  * XXX: In the case of soft timeouts, should we eventually give up
1100  *	if sendmsg is not able to make progress?
1101  */
1102 static int xs_tcp_send_request(struct rpc_rqst *req)
1103 {
1104 	struct rpc_xprt *xprt = req->rq_xprt;
1105 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1106 	struct xdr_buf *xdr = &req->rq_snd_buf;
1107 	rpc_fraghdr rm = xs_stream_record_marker(xdr);
1108 	unsigned int msglen = rm ? req->rq_slen + sizeof(rm) : req->rq_slen;
1109 	bool vm_wait = false;
1110 	int status;
1111 	int sent;
1112 
1113 	/* Close the stream if the previous transmission was incomplete */
1114 	if (xs_send_request_was_aborted(transport, req)) {
1115 		if (transport->sock != NULL)
1116 			kernel_sock_shutdown(transport->sock, SHUT_RDWR);
1117 		return -ENOTCONN;
1118 	}
1119 
1120 	xs_pktdump("packet data:",
1121 				req->rq_svec->iov_base,
1122 				req->rq_svec->iov_len);
1123 
1124 	if (test_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state))
1125 		xs_tcp_set_socket_timeouts(xprt, transport->sock);
1126 
1127 	/* Continue transmitting the packet/record. We must be careful
1128 	 * to cope with writespace callbacks arriving _after_ we have
1129 	 * called sendmsg(). */
1130 	req->rq_xtime = ktime_get();
1131 	while (1) {
1132 		sent = 0;
1133 		status = xs_sendpages(transport->sock, NULL, 0, xdr,
1134 				      transport->xmit.offset, rm, &sent);
1135 
1136 		dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
1137 				xdr->len - transport->xmit.offset, status);
1138 
1139 		/* If we've sent the entire packet, immediately
1140 		 * reset the count of bytes sent. */
1141 		transport->xmit.offset += sent;
1142 		req->rq_bytes_sent = transport->xmit.offset;
1143 		if (likely(req->rq_bytes_sent >= msglen)) {
1144 			req->rq_xmit_bytes_sent += transport->xmit.offset;
1145 			transport->xmit.offset = 0;
1146 			return 0;
1147 		}
1148 
1149 		WARN_ON_ONCE(sent == 0 && status == 0);
1150 
1151 		if (status == -EAGAIN ) {
1152 			/*
1153 			 * Return EAGAIN if we're sure we're hitting the
1154 			 * socket send buffer limits.
1155 			 */
1156 			if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
1157 				break;
1158 			/*
1159 			 * Did we hit a memory allocation failure?
1160 			 */
1161 			if (sent == 0) {
1162 				status = -ENOBUFS;
1163 				if (vm_wait)
1164 					break;
1165 				/* Retry, knowing now that we're below the
1166 				 * socket send buffer limit
1167 				 */
1168 				vm_wait = true;
1169 			}
1170 			continue;
1171 		}
1172 		if (status < 0)
1173 			break;
1174 		vm_wait = false;
1175 	}
1176 
1177 	switch (status) {
1178 	case -ENOTSOCK:
1179 		status = -ENOTCONN;
1180 		/* Should we call xs_close() here? */
1181 		break;
1182 	case -EAGAIN:
1183 		status = xs_nospace(req);
1184 		break;
1185 	case -ECONNRESET:
1186 	case -ECONNREFUSED:
1187 	case -ENOTCONN:
1188 	case -EADDRINUSE:
1189 	case -ENOBUFS:
1190 	case -EPIPE:
1191 		break;
1192 	default:
1193 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
1194 			-status);
1195 	}
1196 
1197 	return status;
1198 }
1199 
1200 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1201 {
1202 	transport->old_data_ready = sk->sk_data_ready;
1203 	transport->old_state_change = sk->sk_state_change;
1204 	transport->old_write_space = sk->sk_write_space;
1205 	transport->old_error_report = sk->sk_error_report;
1206 }
1207 
1208 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
1209 {
1210 	sk->sk_data_ready = transport->old_data_ready;
1211 	sk->sk_state_change = transport->old_state_change;
1212 	sk->sk_write_space = transport->old_write_space;
1213 	sk->sk_error_report = transport->old_error_report;
1214 }
1215 
1216 static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
1217 {
1218 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1219 
1220 	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
1221 	clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
1222 	clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
1223 	clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
1224 }
1225 
1226 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
1227 {
1228 	set_bit(nr, &transport->sock_state);
1229 	queue_work(xprtiod_workqueue, &transport->error_worker);
1230 }
1231 
1232 static void xs_sock_reset_connection_flags(struct rpc_xprt *xprt)
1233 {
1234 	smp_mb__before_atomic();
1235 	clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1236 	clear_bit(XPRT_CLOSING, &xprt->state);
1237 	xs_sock_reset_state_flags(xprt);
1238 	smp_mb__after_atomic();
1239 }
1240 
1241 /**
1242  * xs_error_report - callback to handle TCP socket state errors
1243  * @sk: socket
1244  *
1245  * Note: we don't call sock_error() since there may be a rpc_task
1246  * using the socket, and so we don't want to clear sk->sk_err.
1247  */
1248 static void xs_error_report(struct sock *sk)
1249 {
1250 	struct sock_xprt *transport;
1251 	struct rpc_xprt *xprt;
1252 	int err;
1253 
1254 	read_lock_bh(&sk->sk_callback_lock);
1255 	if (!(xprt = xprt_from_sock(sk)))
1256 		goto out;
1257 
1258 	transport = container_of(xprt, struct sock_xprt, xprt);
1259 	err = -sk->sk_err;
1260 	if (err == 0)
1261 		goto out;
1262 	dprintk("RPC:       xs_error_report client %p, error=%d...\n",
1263 			xprt, -err);
1264 	trace_rpc_socket_error(xprt, sk->sk_socket, err);
1265 	xs_run_error_worker(transport, XPRT_SOCK_WAKE_ERROR);
1266  out:
1267 	read_unlock_bh(&sk->sk_callback_lock);
1268 }
1269 
1270 static void xs_reset_transport(struct sock_xprt *transport)
1271 {
1272 	struct socket *sock = transport->sock;
1273 	struct sock *sk = transport->inet;
1274 	struct rpc_xprt *xprt = &transport->xprt;
1275 	struct file *filp = transport->file;
1276 
1277 	if (sk == NULL)
1278 		return;
1279 
1280 	if (atomic_read(&transport->xprt.swapper))
1281 		sk_clear_memalloc(sk);
1282 
1283 	kernel_sock_shutdown(sock, SHUT_RDWR);
1284 
1285 	mutex_lock(&transport->recv_mutex);
1286 	write_lock_bh(&sk->sk_callback_lock);
1287 	transport->inet = NULL;
1288 	transport->sock = NULL;
1289 	transport->file = NULL;
1290 
1291 	sk->sk_user_data = NULL;
1292 
1293 	xs_restore_old_callbacks(transport, sk);
1294 	xprt_clear_connected(xprt);
1295 	write_unlock_bh(&sk->sk_callback_lock);
1296 	xs_sock_reset_connection_flags(xprt);
1297 	/* Reset stream record info */
1298 	xs_stream_reset_connect(transport);
1299 	mutex_unlock(&transport->recv_mutex);
1300 
1301 	trace_rpc_socket_close(xprt, sock);
1302 	fput(filp);
1303 
1304 	xprt_disconnect_done(xprt);
1305 }
1306 
1307 /**
1308  * xs_close - close a socket
1309  * @xprt: transport
1310  *
1311  * This is used when all requests are complete; ie, no DRC state remains
1312  * on the server we want to save.
1313  *
1314  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
1315  * xs_reset_transport() zeroing the socket from underneath a writer.
1316  */
1317 static void xs_close(struct rpc_xprt *xprt)
1318 {
1319 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1320 
1321 	dprintk("RPC:       xs_close xprt %p\n", xprt);
1322 
1323 	xs_reset_transport(transport);
1324 	xprt->reestablish_timeout = 0;
1325 }
1326 
1327 static void xs_inject_disconnect(struct rpc_xprt *xprt)
1328 {
1329 	dprintk("RPC:       injecting transport disconnect on xprt=%p\n",
1330 		xprt);
1331 	xprt_disconnect_done(xprt);
1332 }
1333 
1334 static void xs_xprt_free(struct rpc_xprt *xprt)
1335 {
1336 	xs_free_peer_addresses(xprt);
1337 	xprt_free(xprt);
1338 }
1339 
1340 /**
1341  * xs_destroy - prepare to shutdown a transport
1342  * @xprt: doomed transport
1343  *
1344  */
1345 static void xs_destroy(struct rpc_xprt *xprt)
1346 {
1347 	struct sock_xprt *transport = container_of(xprt,
1348 			struct sock_xprt, xprt);
1349 	dprintk("RPC:       xs_destroy xprt %p\n", xprt);
1350 
1351 	cancel_delayed_work_sync(&transport->connect_worker);
1352 	xs_close(xprt);
1353 	cancel_work_sync(&transport->recv_worker);
1354 	cancel_work_sync(&transport->error_worker);
1355 	xs_xprt_free(xprt);
1356 	module_put(THIS_MODULE);
1357 }
1358 
1359 /**
1360  * xs_udp_data_read_skb - receive callback for UDP sockets
1361  * @xprt: transport
1362  * @sk: socket
1363  * @skb: skbuff
1364  *
1365  */
1366 static void xs_udp_data_read_skb(struct rpc_xprt *xprt,
1367 		struct sock *sk,
1368 		struct sk_buff *skb)
1369 {
1370 	struct rpc_task *task;
1371 	struct rpc_rqst *rovr;
1372 	int repsize, copied;
1373 	u32 _xid;
1374 	__be32 *xp;
1375 
1376 	repsize = skb->len;
1377 	if (repsize < 4) {
1378 		dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
1379 		return;
1380 	}
1381 
1382 	/* Copy the XID from the skb... */
1383 	xp = skb_header_pointer(skb, 0, sizeof(_xid), &_xid);
1384 	if (xp == NULL)
1385 		return;
1386 
1387 	/* Look up and lock the request corresponding to the given XID */
1388 	spin_lock(&xprt->queue_lock);
1389 	rovr = xprt_lookup_rqst(xprt, *xp);
1390 	if (!rovr)
1391 		goto out_unlock;
1392 	xprt_pin_rqst(rovr);
1393 	xprt_update_rtt(rovr->rq_task);
1394 	spin_unlock(&xprt->queue_lock);
1395 	task = rovr->rq_task;
1396 
1397 	if ((copied = rovr->rq_private_buf.buflen) > repsize)
1398 		copied = repsize;
1399 
1400 	/* Suck it into the iovec, verify checksum if not done by hw. */
1401 	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
1402 		spin_lock(&xprt->queue_lock);
1403 		__UDPX_INC_STATS(sk, UDP_MIB_INERRORS);
1404 		goto out_unpin;
1405 	}
1406 
1407 
1408 	spin_lock(&xprt->transport_lock);
1409 	xprt_adjust_cwnd(xprt, task, copied);
1410 	spin_unlock(&xprt->transport_lock);
1411 	spin_lock(&xprt->queue_lock);
1412 	xprt_complete_rqst(task, copied);
1413 	__UDPX_INC_STATS(sk, UDP_MIB_INDATAGRAMS);
1414 out_unpin:
1415 	xprt_unpin_rqst(rovr);
1416  out_unlock:
1417 	spin_unlock(&xprt->queue_lock);
1418 }
1419 
1420 static void xs_udp_data_receive(struct sock_xprt *transport)
1421 {
1422 	struct sk_buff *skb;
1423 	struct sock *sk;
1424 	int err;
1425 
1426 	mutex_lock(&transport->recv_mutex);
1427 	sk = transport->inet;
1428 	if (sk == NULL)
1429 		goto out;
1430 	for (;;) {
1431 		skb = skb_recv_udp(sk, 0, 1, &err);
1432 		if (skb == NULL)
1433 			break;
1434 		xs_udp_data_read_skb(&transport->xprt, sk, skb);
1435 		consume_skb(skb);
1436 		cond_resched();
1437 	}
1438 	xs_poll_check_readable(transport);
1439 out:
1440 	mutex_unlock(&transport->recv_mutex);
1441 }
1442 
1443 static void xs_udp_data_receive_workfn(struct work_struct *work)
1444 {
1445 	struct sock_xprt *transport =
1446 		container_of(work, struct sock_xprt, recv_worker);
1447 	unsigned int pflags = memalloc_nofs_save();
1448 
1449 	xs_udp_data_receive(transport);
1450 	memalloc_nofs_restore(pflags);
1451 }
1452 
1453 /**
1454  * xs_data_ready - "data ready" callback for UDP sockets
1455  * @sk: socket with data to read
1456  *
1457  */
1458 static void xs_data_ready(struct sock *sk)
1459 {
1460 	struct rpc_xprt *xprt;
1461 
1462 	read_lock_bh(&sk->sk_callback_lock);
1463 	dprintk("RPC:       xs_data_ready...\n");
1464 	xprt = xprt_from_sock(sk);
1465 	if (xprt != NULL) {
1466 		struct sock_xprt *transport = container_of(xprt,
1467 				struct sock_xprt, xprt);
1468 		transport->old_data_ready(sk);
1469 		/* Any data means we had a useful conversation, so
1470 		 * then we don't need to delay the next reconnect
1471 		 */
1472 		if (xprt->reestablish_timeout)
1473 			xprt->reestablish_timeout = 0;
1474 		if (!test_and_set_bit(XPRT_SOCK_DATA_READY, &transport->sock_state))
1475 			queue_work(xprtiod_workqueue, &transport->recv_worker);
1476 	}
1477 	read_unlock_bh(&sk->sk_callback_lock);
1478 }
1479 
1480 /*
1481  * Helper function to force a TCP close if the server is sending
1482  * junk and/or it has put us in CLOSE_WAIT
1483  */
1484 static void xs_tcp_force_close(struct rpc_xprt *xprt)
1485 {
1486 	xprt_force_disconnect(xprt);
1487 }
1488 
1489 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1490 static size_t xs_tcp_bc_maxpayload(struct rpc_xprt *xprt)
1491 {
1492 	return PAGE_SIZE;
1493 }
1494 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1495 
1496 /**
1497  * xs_tcp_state_change - callback to handle TCP socket state changes
1498  * @sk: socket whose state has changed
1499  *
1500  */
1501 static void xs_tcp_state_change(struct sock *sk)
1502 {
1503 	struct rpc_xprt *xprt;
1504 	struct sock_xprt *transport;
1505 
1506 	read_lock_bh(&sk->sk_callback_lock);
1507 	if (!(xprt = xprt_from_sock(sk)))
1508 		goto out;
1509 	dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1510 	dprintk("RPC:       state %x conn %d dead %d zapped %d sk_shutdown %d\n",
1511 			sk->sk_state, xprt_connected(xprt),
1512 			sock_flag(sk, SOCK_DEAD),
1513 			sock_flag(sk, SOCK_ZAPPED),
1514 			sk->sk_shutdown);
1515 
1516 	transport = container_of(xprt, struct sock_xprt, xprt);
1517 	trace_rpc_socket_state_change(xprt, sk->sk_socket);
1518 	switch (sk->sk_state) {
1519 	case TCP_ESTABLISHED:
1520 		if (!xprt_test_and_set_connected(xprt)) {
1521 			xprt->connect_cookie++;
1522 			clear_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
1523 			xprt_clear_connecting(xprt);
1524 
1525 			xprt->stat.connect_count++;
1526 			xprt->stat.connect_time += (long)jiffies -
1527 						   xprt->stat.connect_start;
1528 			xs_run_error_worker(transport, XPRT_SOCK_WAKE_PENDING);
1529 		}
1530 		break;
1531 	case TCP_FIN_WAIT1:
1532 		/* The client initiated a shutdown of the socket */
1533 		xprt->connect_cookie++;
1534 		xprt->reestablish_timeout = 0;
1535 		set_bit(XPRT_CLOSING, &xprt->state);
1536 		smp_mb__before_atomic();
1537 		clear_bit(XPRT_CONNECTED, &xprt->state);
1538 		clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1539 		smp_mb__after_atomic();
1540 		break;
1541 	case TCP_CLOSE_WAIT:
1542 		/* The server initiated a shutdown of the socket */
1543 		xprt->connect_cookie++;
1544 		clear_bit(XPRT_CONNECTED, &xprt->state);
1545 		xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1546 		/* fall through */
1547 	case TCP_CLOSING:
1548 		/*
1549 		 * If the server closed down the connection, make sure that
1550 		 * we back off before reconnecting
1551 		 */
1552 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1553 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1554 		break;
1555 	case TCP_LAST_ACK:
1556 		set_bit(XPRT_CLOSING, &xprt->state);
1557 		smp_mb__before_atomic();
1558 		clear_bit(XPRT_CONNECTED, &xprt->state);
1559 		smp_mb__after_atomic();
1560 		break;
1561 	case TCP_CLOSE:
1562 		if (test_and_clear_bit(XPRT_SOCK_CONNECTING,
1563 					&transport->sock_state))
1564 			xprt_clear_connecting(xprt);
1565 		clear_bit(XPRT_CLOSING, &xprt->state);
1566 		/* Trigger the socket release */
1567 		xs_run_error_worker(transport, XPRT_SOCK_WAKE_DISCONNECT);
1568 	}
1569  out:
1570 	read_unlock_bh(&sk->sk_callback_lock);
1571 }
1572 
1573 static void xs_write_space(struct sock *sk)
1574 {
1575 	struct socket_wq *wq;
1576 	struct sock_xprt *transport;
1577 	struct rpc_xprt *xprt;
1578 
1579 	if (!sk->sk_socket)
1580 		return;
1581 	clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1582 
1583 	if (unlikely(!(xprt = xprt_from_sock(sk))))
1584 		return;
1585 	transport = container_of(xprt, struct sock_xprt, xprt);
1586 	rcu_read_lock();
1587 	wq = rcu_dereference(sk->sk_wq);
1588 	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
1589 		goto out;
1590 
1591 	xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
1592 	sk->sk_write_pending--;
1593 out:
1594 	rcu_read_unlock();
1595 }
1596 
1597 /**
1598  * xs_udp_write_space - callback invoked when socket buffer space
1599  *                             becomes available
1600  * @sk: socket whose state has changed
1601  *
1602  * Called when more output buffer space is available for this socket.
1603  * We try not to wake our writers until they can make "significant"
1604  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1605  * with a bunch of small requests.
1606  */
1607 static void xs_udp_write_space(struct sock *sk)
1608 {
1609 	read_lock_bh(&sk->sk_callback_lock);
1610 
1611 	/* from net/core/sock.c:sock_def_write_space */
1612 	if (sock_writeable(sk))
1613 		xs_write_space(sk);
1614 
1615 	read_unlock_bh(&sk->sk_callback_lock);
1616 }
1617 
1618 /**
1619  * xs_tcp_write_space - callback invoked when socket buffer space
1620  *                             becomes available
1621  * @sk: socket whose state has changed
1622  *
1623  * Called when more output buffer space is available for this socket.
1624  * We try not to wake our writers until they can make "significant"
1625  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1626  * with a bunch of small requests.
1627  */
1628 static void xs_tcp_write_space(struct sock *sk)
1629 {
1630 	read_lock_bh(&sk->sk_callback_lock);
1631 
1632 	/* from net/core/stream.c:sk_stream_write_space */
1633 	if (sk_stream_is_writeable(sk))
1634 		xs_write_space(sk);
1635 
1636 	read_unlock_bh(&sk->sk_callback_lock);
1637 }
1638 
1639 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1640 {
1641 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1642 	struct sock *sk = transport->inet;
1643 
1644 	if (transport->rcvsize) {
1645 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1646 		sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1647 	}
1648 	if (transport->sndsize) {
1649 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1650 		sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1651 		sk->sk_write_space(sk);
1652 	}
1653 }
1654 
1655 /**
1656  * xs_udp_set_buffer_size - set send and receive limits
1657  * @xprt: generic transport
1658  * @sndsize: requested size of send buffer, in bytes
1659  * @rcvsize: requested size of receive buffer, in bytes
1660  *
1661  * Set socket send and receive buffer size limits.
1662  */
1663 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1664 {
1665 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1666 
1667 	transport->sndsize = 0;
1668 	if (sndsize)
1669 		transport->sndsize = sndsize + 1024;
1670 	transport->rcvsize = 0;
1671 	if (rcvsize)
1672 		transport->rcvsize = rcvsize + 1024;
1673 
1674 	xs_udp_do_set_buffer_size(xprt);
1675 }
1676 
1677 /**
1678  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1679  * @xprt: controlling transport
1680  * @task: task that timed out
1681  *
1682  * Adjust the congestion window after a retransmit timeout has occurred.
1683  */
1684 static void xs_udp_timer(struct rpc_xprt *xprt, struct rpc_task *task)
1685 {
1686 	spin_lock(&xprt->transport_lock);
1687 	xprt_adjust_cwnd(xprt, task, -ETIMEDOUT);
1688 	spin_unlock(&xprt->transport_lock);
1689 }
1690 
1691 static int xs_get_random_port(void)
1692 {
1693 	unsigned short min = xprt_min_resvport, max = xprt_max_resvport;
1694 	unsigned short range;
1695 	unsigned short rand;
1696 
1697 	if (max < min)
1698 		return -EADDRINUSE;
1699 	range = max - min + 1;
1700 	rand = (unsigned short) prandom_u32() % range;
1701 	return rand + min;
1702 }
1703 
1704 /**
1705  * xs_set_reuseaddr_port - set the socket's port and address reuse options
1706  * @sock: socket
1707  *
1708  * Note that this function has to be called on all sockets that share the
1709  * same port, and it must be called before binding.
1710  */
1711 static void xs_sock_set_reuseport(struct socket *sock)
1712 {
1713 	int opt = 1;
1714 
1715 	kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEPORT,
1716 			(char *)&opt, sizeof(opt));
1717 }
1718 
1719 static unsigned short xs_sock_getport(struct socket *sock)
1720 {
1721 	struct sockaddr_storage buf;
1722 	unsigned short port = 0;
1723 
1724 	if (kernel_getsockname(sock, (struct sockaddr *)&buf) < 0)
1725 		goto out;
1726 	switch (buf.ss_family) {
1727 	case AF_INET6:
1728 		port = ntohs(((struct sockaddr_in6 *)&buf)->sin6_port);
1729 		break;
1730 	case AF_INET:
1731 		port = ntohs(((struct sockaddr_in *)&buf)->sin_port);
1732 	}
1733 out:
1734 	return port;
1735 }
1736 
1737 /**
1738  * xs_set_port - reset the port number in the remote endpoint address
1739  * @xprt: generic transport
1740  * @port: new port number
1741  *
1742  */
1743 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1744 {
1745 	dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1746 
1747 	rpc_set_port(xs_addr(xprt), port);
1748 	xs_update_peer_port(xprt);
1749 }
1750 
1751 static void xs_set_srcport(struct sock_xprt *transport, struct socket *sock)
1752 {
1753 	if (transport->srcport == 0)
1754 		transport->srcport = xs_sock_getport(sock);
1755 }
1756 
1757 static int xs_get_srcport(struct sock_xprt *transport)
1758 {
1759 	int port = transport->srcport;
1760 
1761 	if (port == 0 && transport->xprt.resvport)
1762 		port = xs_get_random_port();
1763 	return port;
1764 }
1765 
1766 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
1767 {
1768 	if (transport->srcport != 0)
1769 		transport->srcport = 0;
1770 	if (!transport->xprt.resvport)
1771 		return 0;
1772 	if (port <= xprt_min_resvport || port > xprt_max_resvport)
1773 		return xprt_max_resvport;
1774 	return --port;
1775 }
1776 static int xs_bind(struct sock_xprt *transport, struct socket *sock)
1777 {
1778 	struct sockaddr_storage myaddr;
1779 	int err, nloop = 0;
1780 	int port = xs_get_srcport(transport);
1781 	unsigned short last;
1782 
1783 	/*
1784 	 * If we are asking for any ephemeral port (i.e. port == 0 &&
1785 	 * transport->xprt.resvport == 0), don't bind.  Let the local
1786 	 * port selection happen implicitly when the socket is used
1787 	 * (for example at connect time).
1788 	 *
1789 	 * This ensures that we can continue to establish TCP
1790 	 * connections even when all local ephemeral ports are already
1791 	 * a part of some TCP connection.  This makes no difference
1792 	 * for UDP sockets, but also doens't harm them.
1793 	 *
1794 	 * If we're asking for any reserved port (i.e. port == 0 &&
1795 	 * transport->xprt.resvport == 1) xs_get_srcport above will
1796 	 * ensure that port is non-zero and we will bind as needed.
1797 	 */
1798 	if (port <= 0)
1799 		return port;
1800 
1801 	memcpy(&myaddr, &transport->srcaddr, transport->xprt.addrlen);
1802 	do {
1803 		rpc_set_port((struct sockaddr *)&myaddr, port);
1804 		err = kernel_bind(sock, (struct sockaddr *)&myaddr,
1805 				transport->xprt.addrlen);
1806 		if (err == 0) {
1807 			transport->srcport = port;
1808 			break;
1809 		}
1810 		last = port;
1811 		port = xs_next_srcport(transport, port);
1812 		if (port > last)
1813 			nloop++;
1814 	} while (err == -EADDRINUSE && nloop != 2);
1815 
1816 	if (myaddr.ss_family == AF_INET)
1817 		dprintk("RPC:       %s %pI4:%u: %s (%d)\n", __func__,
1818 				&((struct sockaddr_in *)&myaddr)->sin_addr,
1819 				port, err ? "failed" : "ok", err);
1820 	else
1821 		dprintk("RPC:       %s %pI6:%u: %s (%d)\n", __func__,
1822 				&((struct sockaddr_in6 *)&myaddr)->sin6_addr,
1823 				port, err ? "failed" : "ok", err);
1824 	return err;
1825 }
1826 
1827 /*
1828  * We don't support autobind on AF_LOCAL sockets
1829  */
1830 static void xs_local_rpcbind(struct rpc_task *task)
1831 {
1832 	xprt_set_bound(task->tk_xprt);
1833 }
1834 
1835 static void xs_local_set_port(struct rpc_xprt *xprt, unsigned short port)
1836 {
1837 }
1838 
1839 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1840 static struct lock_class_key xs_key[2];
1841 static struct lock_class_key xs_slock_key[2];
1842 
1843 static inline void xs_reclassify_socketu(struct socket *sock)
1844 {
1845 	struct sock *sk = sock->sk;
1846 
1847 	sock_lock_init_class_and_name(sk, "slock-AF_LOCAL-RPC",
1848 		&xs_slock_key[1], "sk_lock-AF_LOCAL-RPC", &xs_key[1]);
1849 }
1850 
1851 static inline void xs_reclassify_socket4(struct socket *sock)
1852 {
1853 	struct sock *sk = sock->sk;
1854 
1855 	sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1856 		&xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1857 }
1858 
1859 static inline void xs_reclassify_socket6(struct socket *sock)
1860 {
1861 	struct sock *sk = sock->sk;
1862 
1863 	sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1864 		&xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1865 }
1866 
1867 static inline void xs_reclassify_socket(int family, struct socket *sock)
1868 {
1869 	if (WARN_ON_ONCE(!sock_allow_reclassification(sock->sk)))
1870 		return;
1871 
1872 	switch (family) {
1873 	case AF_LOCAL:
1874 		xs_reclassify_socketu(sock);
1875 		break;
1876 	case AF_INET:
1877 		xs_reclassify_socket4(sock);
1878 		break;
1879 	case AF_INET6:
1880 		xs_reclassify_socket6(sock);
1881 		break;
1882 	}
1883 }
1884 #else
1885 static inline void xs_reclassify_socket(int family, struct socket *sock)
1886 {
1887 }
1888 #endif
1889 
1890 static void xs_dummy_setup_socket(struct work_struct *work)
1891 {
1892 }
1893 
1894 static struct socket *xs_create_sock(struct rpc_xprt *xprt,
1895 		struct sock_xprt *transport, int family, int type,
1896 		int protocol, bool reuseport)
1897 {
1898 	struct file *filp;
1899 	struct socket *sock;
1900 	int err;
1901 
1902 	err = __sock_create(xprt->xprt_net, family, type, protocol, &sock, 1);
1903 	if (err < 0) {
1904 		dprintk("RPC:       can't create %d transport socket (%d).\n",
1905 				protocol, -err);
1906 		goto out;
1907 	}
1908 	xs_reclassify_socket(family, sock);
1909 
1910 	if (reuseport)
1911 		xs_sock_set_reuseport(sock);
1912 
1913 	err = xs_bind(transport, sock);
1914 	if (err) {
1915 		sock_release(sock);
1916 		goto out;
1917 	}
1918 
1919 	filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1920 	if (IS_ERR(filp))
1921 		return ERR_CAST(filp);
1922 	transport->file = filp;
1923 
1924 	return sock;
1925 out:
1926 	return ERR_PTR(err);
1927 }
1928 
1929 static int xs_local_finish_connecting(struct rpc_xprt *xprt,
1930 				      struct socket *sock)
1931 {
1932 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
1933 									xprt);
1934 
1935 	if (!transport->inet) {
1936 		struct sock *sk = sock->sk;
1937 
1938 		write_lock_bh(&sk->sk_callback_lock);
1939 
1940 		xs_save_old_callbacks(transport, sk);
1941 
1942 		sk->sk_user_data = xprt;
1943 		sk->sk_data_ready = xs_data_ready;
1944 		sk->sk_write_space = xs_udp_write_space;
1945 		sock_set_flag(sk, SOCK_FASYNC);
1946 		sk->sk_error_report = xs_error_report;
1947 
1948 		xprt_clear_connected(xprt);
1949 
1950 		/* Reset to new socket */
1951 		transport->sock = sock;
1952 		transport->inet = sk;
1953 
1954 		write_unlock_bh(&sk->sk_callback_lock);
1955 	}
1956 
1957 	xs_stream_start_connect(transport);
1958 
1959 	return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, 0);
1960 }
1961 
1962 /**
1963  * xs_local_setup_socket - create AF_LOCAL socket, connect to a local endpoint
1964  * @transport: socket transport to connect
1965  */
1966 static int xs_local_setup_socket(struct sock_xprt *transport)
1967 {
1968 	struct rpc_xprt *xprt = &transport->xprt;
1969 	struct file *filp;
1970 	struct socket *sock;
1971 	int status = -EIO;
1972 
1973 	status = __sock_create(xprt->xprt_net, AF_LOCAL,
1974 					SOCK_STREAM, 0, &sock, 1);
1975 	if (status < 0) {
1976 		dprintk("RPC:       can't create AF_LOCAL "
1977 			"transport socket (%d).\n", -status);
1978 		goto out;
1979 	}
1980 	xs_reclassify_socket(AF_LOCAL, sock);
1981 
1982 	filp = sock_alloc_file(sock, O_NONBLOCK, NULL);
1983 	if (IS_ERR(filp)) {
1984 		status = PTR_ERR(filp);
1985 		goto out;
1986 	}
1987 	transport->file = filp;
1988 
1989 	dprintk("RPC:       worker connecting xprt %p via AF_LOCAL to %s\n",
1990 			xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1991 
1992 	status = xs_local_finish_connecting(xprt, sock);
1993 	trace_rpc_socket_connect(xprt, sock, status);
1994 	switch (status) {
1995 	case 0:
1996 		dprintk("RPC:       xprt %p connected to %s\n",
1997 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
1998 		xprt->stat.connect_count++;
1999 		xprt->stat.connect_time += (long)jiffies -
2000 					   xprt->stat.connect_start;
2001 		xprt_set_connected(xprt);
2002 	case -ENOBUFS:
2003 		break;
2004 	case -ENOENT:
2005 		dprintk("RPC:       xprt %p: socket %s does not exist\n",
2006 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2007 		break;
2008 	case -ECONNREFUSED:
2009 		dprintk("RPC:       xprt %p: connection refused for %s\n",
2010 				xprt, xprt->address_strings[RPC_DISPLAY_ADDR]);
2011 		break;
2012 	default:
2013 		printk(KERN_ERR "%s: unhandled error (%d) connecting to %s\n",
2014 				__func__, -status,
2015 				xprt->address_strings[RPC_DISPLAY_ADDR]);
2016 	}
2017 
2018 out:
2019 	xprt_clear_connecting(xprt);
2020 	xprt_wake_pending_tasks(xprt, status);
2021 	return status;
2022 }
2023 
2024 static void xs_local_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2025 {
2026 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2027 	int ret;
2028 
2029 	 if (RPC_IS_ASYNC(task)) {
2030 		/*
2031 		 * We want the AF_LOCAL connect to be resolved in the
2032 		 * filesystem namespace of the process making the rpc
2033 		 * call.  Thus we connect synchronously.
2034 		 *
2035 		 * If we want to support asynchronous AF_LOCAL calls,
2036 		 * we'll need to figure out how to pass a namespace to
2037 		 * connect.
2038 		 */
2039 		task->tk_rpc_status = -ENOTCONN;
2040 		rpc_exit(task, -ENOTCONN);
2041 		return;
2042 	}
2043 	ret = xs_local_setup_socket(transport);
2044 	if (ret && !RPC_IS_SOFTCONN(task))
2045 		msleep_interruptible(15000);
2046 }
2047 
2048 #if IS_ENABLED(CONFIG_SUNRPC_SWAP)
2049 /*
2050  * Note that this should be called with XPRT_LOCKED held (or when we otherwise
2051  * know that we have exclusive access to the socket), to guard against
2052  * races with xs_reset_transport.
2053  */
2054 static void xs_set_memalloc(struct rpc_xprt *xprt)
2055 {
2056 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt,
2057 			xprt);
2058 
2059 	/*
2060 	 * If there's no sock, then we have nothing to set. The
2061 	 * reconnecting process will get it for us.
2062 	 */
2063 	if (!transport->inet)
2064 		return;
2065 	if (atomic_read(&xprt->swapper))
2066 		sk_set_memalloc(transport->inet);
2067 }
2068 
2069 /**
2070  * xs_enable_swap - Tag this transport as being used for swap.
2071  * @xprt: transport to tag
2072  *
2073  * Take a reference to this transport on behalf of the rpc_clnt, and
2074  * optionally mark it for swapping if it wasn't already.
2075  */
2076 static int
2077 xs_enable_swap(struct rpc_xprt *xprt)
2078 {
2079 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2080 
2081 	if (atomic_inc_return(&xprt->swapper) != 1)
2082 		return 0;
2083 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2084 		return -ERESTARTSYS;
2085 	if (xs->inet)
2086 		sk_set_memalloc(xs->inet);
2087 	xprt_release_xprt(xprt, NULL);
2088 	return 0;
2089 }
2090 
2091 /**
2092  * xs_disable_swap - Untag this transport as being used for swap.
2093  * @xprt: transport to tag
2094  *
2095  * Drop a "swapper" reference to this xprt on behalf of the rpc_clnt. If the
2096  * swapper refcount goes to 0, untag the socket as a memalloc socket.
2097  */
2098 static void
2099 xs_disable_swap(struct rpc_xprt *xprt)
2100 {
2101 	struct sock_xprt *xs = container_of(xprt, struct sock_xprt, xprt);
2102 
2103 	if (!atomic_dec_and_test(&xprt->swapper))
2104 		return;
2105 	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE))
2106 		return;
2107 	if (xs->inet)
2108 		sk_clear_memalloc(xs->inet);
2109 	xprt_release_xprt(xprt, NULL);
2110 }
2111 #else
2112 static void xs_set_memalloc(struct rpc_xprt *xprt)
2113 {
2114 }
2115 
2116 static int
2117 xs_enable_swap(struct rpc_xprt *xprt)
2118 {
2119 	return -EINVAL;
2120 }
2121 
2122 static void
2123 xs_disable_swap(struct rpc_xprt *xprt)
2124 {
2125 }
2126 #endif
2127 
2128 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2129 {
2130 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2131 
2132 	if (!transport->inet) {
2133 		struct sock *sk = sock->sk;
2134 
2135 		write_lock_bh(&sk->sk_callback_lock);
2136 
2137 		xs_save_old_callbacks(transport, sk);
2138 
2139 		sk->sk_user_data = xprt;
2140 		sk->sk_data_ready = xs_data_ready;
2141 		sk->sk_write_space = xs_udp_write_space;
2142 		sock_set_flag(sk, SOCK_FASYNC);
2143 
2144 		xprt_set_connected(xprt);
2145 
2146 		/* Reset to new socket */
2147 		transport->sock = sock;
2148 		transport->inet = sk;
2149 
2150 		xs_set_memalloc(xprt);
2151 
2152 		write_unlock_bh(&sk->sk_callback_lock);
2153 	}
2154 	xs_udp_do_set_buffer_size(xprt);
2155 
2156 	xprt->stat.connect_start = jiffies;
2157 }
2158 
2159 static void xs_udp_setup_socket(struct work_struct *work)
2160 {
2161 	struct sock_xprt *transport =
2162 		container_of(work, struct sock_xprt, connect_worker.work);
2163 	struct rpc_xprt *xprt = &transport->xprt;
2164 	struct socket *sock;
2165 	int status = -EIO;
2166 
2167 	sock = xs_create_sock(xprt, transport,
2168 			xs_addr(xprt)->sa_family, SOCK_DGRAM,
2169 			IPPROTO_UDP, false);
2170 	if (IS_ERR(sock))
2171 		goto out;
2172 
2173 	dprintk("RPC:       worker connecting xprt %p via %s to "
2174 				"%s (port %s)\n", xprt,
2175 			xprt->address_strings[RPC_DISPLAY_PROTO],
2176 			xprt->address_strings[RPC_DISPLAY_ADDR],
2177 			xprt->address_strings[RPC_DISPLAY_PORT]);
2178 
2179 	xs_udp_finish_connecting(xprt, sock);
2180 	trace_rpc_socket_connect(xprt, sock, 0);
2181 	status = 0;
2182 out:
2183 	xprt_clear_connecting(xprt);
2184 	xprt_unlock_connect(xprt, transport);
2185 	xprt_wake_pending_tasks(xprt, status);
2186 }
2187 
2188 /**
2189  * xs_tcp_shutdown - gracefully shut down a TCP socket
2190  * @xprt: transport
2191  *
2192  * Initiates a graceful shutdown of the TCP socket by calling the
2193  * equivalent of shutdown(SHUT_RDWR);
2194  */
2195 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
2196 {
2197 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2198 	struct socket *sock = transport->sock;
2199 	int skst = transport->inet ? transport->inet->sk_state : TCP_CLOSE;
2200 
2201 	if (sock == NULL)
2202 		return;
2203 	switch (skst) {
2204 	default:
2205 		kernel_sock_shutdown(sock, SHUT_RDWR);
2206 		trace_rpc_socket_shutdown(xprt, sock);
2207 		break;
2208 	case TCP_CLOSE:
2209 	case TCP_TIME_WAIT:
2210 		xs_reset_transport(transport);
2211 	}
2212 }
2213 
2214 static void xs_tcp_set_socket_timeouts(struct rpc_xprt *xprt,
2215 		struct socket *sock)
2216 {
2217 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2218 	unsigned int keepidle;
2219 	unsigned int keepcnt;
2220 	unsigned int opt_on = 1;
2221 	unsigned int timeo;
2222 
2223 	spin_lock(&xprt->transport_lock);
2224 	keepidle = DIV_ROUND_UP(xprt->timeout->to_initval, HZ);
2225 	keepcnt = xprt->timeout->to_retries + 1;
2226 	timeo = jiffies_to_msecs(xprt->timeout->to_initval) *
2227 		(xprt->timeout->to_retries + 1);
2228 	clear_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2229 	spin_unlock(&xprt->transport_lock);
2230 
2231 	/* TCP Keepalive options */
2232 	kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
2233 			(char *)&opt_on, sizeof(opt_on));
2234 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
2235 			(char *)&keepidle, sizeof(keepidle));
2236 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
2237 			(char *)&keepidle, sizeof(keepidle));
2238 	kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
2239 			(char *)&keepcnt, sizeof(keepcnt));
2240 
2241 	/* TCP user timeout (see RFC5482) */
2242 	kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
2243 			(char *)&timeo, sizeof(timeo));
2244 }
2245 
2246 static void xs_tcp_set_connect_timeout(struct rpc_xprt *xprt,
2247 		unsigned long connect_timeout,
2248 		unsigned long reconnect_timeout)
2249 {
2250 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2251 	struct rpc_timeout to;
2252 	unsigned long initval;
2253 
2254 	spin_lock(&xprt->transport_lock);
2255 	if (reconnect_timeout < xprt->max_reconnect_timeout)
2256 		xprt->max_reconnect_timeout = reconnect_timeout;
2257 	if (connect_timeout < xprt->connect_timeout) {
2258 		memcpy(&to, xprt->timeout, sizeof(to));
2259 		initval = DIV_ROUND_UP(connect_timeout, to.to_retries + 1);
2260 		/* Arbitrary lower limit */
2261 		if (initval <  XS_TCP_INIT_REEST_TO << 1)
2262 			initval = XS_TCP_INIT_REEST_TO << 1;
2263 		to.to_initval = initval;
2264 		to.to_maxval = initval;
2265 		memcpy(&transport->tcp_timeout, &to,
2266 				sizeof(transport->tcp_timeout));
2267 		xprt->timeout = &transport->tcp_timeout;
2268 		xprt->connect_timeout = connect_timeout;
2269 	}
2270 	set_bit(XPRT_SOCK_UPD_TIMEOUT, &transport->sock_state);
2271 	spin_unlock(&xprt->transport_lock);
2272 }
2273 
2274 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
2275 {
2276 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2277 	int ret = -ENOTCONN;
2278 
2279 	if (!transport->inet) {
2280 		struct sock *sk = sock->sk;
2281 		unsigned int addr_pref = IPV6_PREFER_SRC_PUBLIC;
2282 
2283 		/* Avoid temporary address, they are bad for long-lived
2284 		 * connections such as NFS mounts.
2285 		 * RFC4941, section 3.6 suggests that:
2286 		 *    Individual applications, which have specific
2287 		 *    knowledge about the normal duration of connections,
2288 		 *    MAY override this as appropriate.
2289 		 */
2290 		kernel_setsockopt(sock, SOL_IPV6, IPV6_ADDR_PREFERENCES,
2291 				(char *)&addr_pref, sizeof(addr_pref));
2292 
2293 		xs_tcp_set_socket_timeouts(xprt, sock);
2294 
2295 		write_lock_bh(&sk->sk_callback_lock);
2296 
2297 		xs_save_old_callbacks(transport, sk);
2298 
2299 		sk->sk_user_data = xprt;
2300 		sk->sk_data_ready = xs_data_ready;
2301 		sk->sk_state_change = xs_tcp_state_change;
2302 		sk->sk_write_space = xs_tcp_write_space;
2303 		sock_set_flag(sk, SOCK_FASYNC);
2304 		sk->sk_error_report = xs_error_report;
2305 
2306 		/* socket options */
2307 		sock_reset_flag(sk, SOCK_LINGER);
2308 		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
2309 
2310 		xprt_clear_connected(xprt);
2311 
2312 		/* Reset to new socket */
2313 		transport->sock = sock;
2314 		transport->inet = sk;
2315 
2316 		write_unlock_bh(&sk->sk_callback_lock);
2317 	}
2318 
2319 	if (!xprt_bound(xprt))
2320 		goto out;
2321 
2322 	xs_set_memalloc(xprt);
2323 
2324 	xs_stream_start_connect(transport);
2325 
2326 	/* Tell the socket layer to start connecting... */
2327 	set_bit(XPRT_SOCK_CONNECTING, &transport->sock_state);
2328 	ret = kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
2329 	switch (ret) {
2330 	case 0:
2331 		xs_set_srcport(transport, sock);
2332 		/* fall through */
2333 	case -EINPROGRESS:
2334 		/* SYN_SENT! */
2335 		if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2336 			xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2337 		break;
2338 	case -EADDRNOTAVAIL:
2339 		/* Source port number is unavailable. Try a new one! */
2340 		transport->srcport = 0;
2341 	}
2342 out:
2343 	return ret;
2344 }
2345 
2346 /**
2347  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
2348  * @work: queued work item
2349  *
2350  * Invoked by a work queue tasklet.
2351  */
2352 static void xs_tcp_setup_socket(struct work_struct *work)
2353 {
2354 	struct sock_xprt *transport =
2355 		container_of(work, struct sock_xprt, connect_worker.work);
2356 	struct socket *sock = transport->sock;
2357 	struct rpc_xprt *xprt = &transport->xprt;
2358 	int status = -EIO;
2359 
2360 	if (!sock) {
2361 		sock = xs_create_sock(xprt, transport,
2362 				xs_addr(xprt)->sa_family, SOCK_STREAM,
2363 				IPPROTO_TCP, true);
2364 		if (IS_ERR(sock)) {
2365 			status = PTR_ERR(sock);
2366 			goto out;
2367 		}
2368 	}
2369 
2370 	dprintk("RPC:       worker connecting xprt %p via %s to "
2371 				"%s (port %s)\n", xprt,
2372 			xprt->address_strings[RPC_DISPLAY_PROTO],
2373 			xprt->address_strings[RPC_DISPLAY_ADDR],
2374 			xprt->address_strings[RPC_DISPLAY_PORT]);
2375 
2376 	status = xs_tcp_finish_connecting(xprt, sock);
2377 	trace_rpc_socket_connect(xprt, sock, status);
2378 	dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
2379 			xprt, -status, xprt_connected(xprt),
2380 			sock->sk->sk_state);
2381 	switch (status) {
2382 	default:
2383 		printk("%s: connect returned unhandled error %d\n",
2384 			__func__, status);
2385 		/* fall through */
2386 	case -EADDRNOTAVAIL:
2387 		/* We're probably in TIME_WAIT. Get rid of existing socket,
2388 		 * and retry
2389 		 */
2390 		xs_tcp_force_close(xprt);
2391 		break;
2392 	case 0:
2393 	case -EINPROGRESS:
2394 	case -EALREADY:
2395 		xprt_unlock_connect(xprt, transport);
2396 		return;
2397 	case -EINVAL:
2398 		/* Happens, for instance, if the user specified a link
2399 		 * local IPv6 address without a scope-id.
2400 		 */
2401 	case -ECONNREFUSED:
2402 	case -ECONNRESET:
2403 	case -ENETDOWN:
2404 	case -ENETUNREACH:
2405 	case -EHOSTUNREACH:
2406 	case -EADDRINUSE:
2407 	case -ENOBUFS:
2408 		/*
2409 		 * xs_tcp_force_close() wakes tasks with -EIO.
2410 		 * We need to wake them first to ensure the
2411 		 * correct error code.
2412 		 */
2413 		xprt_wake_pending_tasks(xprt, status);
2414 		xs_tcp_force_close(xprt);
2415 		goto out;
2416 	}
2417 	status = -EAGAIN;
2418 out:
2419 	xprt_clear_connecting(xprt);
2420 	xprt_unlock_connect(xprt, transport);
2421 	xprt_wake_pending_tasks(xprt, status);
2422 }
2423 
2424 /**
2425  * xs_connect - connect a socket to a remote endpoint
2426  * @xprt: pointer to transport structure
2427  * @task: address of RPC task that manages state of connect request
2428  *
2429  * TCP: If the remote end dropped the connection, delay reconnecting.
2430  *
2431  * UDP socket connects are synchronous, but we use a work queue anyway
2432  * to guarantee that even unprivileged user processes can set up a
2433  * socket on a privileged port.
2434  *
2435  * If a UDP socket connect fails, the delay behavior here prevents
2436  * retry floods (hard mounts).
2437  */
2438 static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
2439 {
2440 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2441 	unsigned long delay = 0;
2442 
2443 	WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
2444 
2445 	if (transport->sock != NULL) {
2446 		dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2447 				"seconds\n",
2448 				xprt, xprt->reestablish_timeout / HZ);
2449 
2450 		/* Start by resetting any existing state */
2451 		xs_reset_transport(transport);
2452 
2453 		delay = xprt_reconnect_delay(xprt);
2454 		xprt_reconnect_backoff(xprt, XS_TCP_INIT_REEST_TO);
2455 
2456 	} else
2457 		dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2458 
2459 	queue_delayed_work(xprtiod_workqueue,
2460 			&transport->connect_worker,
2461 			delay);
2462 }
2463 
2464 static void xs_wake_disconnect(struct sock_xprt *transport)
2465 {
2466 	if (test_and_clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state))
2467 		xs_tcp_force_close(&transport->xprt);
2468 }
2469 
2470 static void xs_wake_write(struct sock_xprt *transport)
2471 {
2472 	if (test_and_clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state))
2473 		xprt_write_space(&transport->xprt);
2474 }
2475 
2476 static void xs_wake_error(struct sock_xprt *transport)
2477 {
2478 	int sockerr;
2479 	int sockerr_len = sizeof(sockerr);
2480 
2481 	if (!test_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2482 		return;
2483 	mutex_lock(&transport->recv_mutex);
2484 	if (transport->sock == NULL)
2485 		goto out;
2486 	if (!test_and_clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state))
2487 		goto out;
2488 	if (kernel_getsockopt(transport->sock, SOL_SOCKET, SO_ERROR,
2489 				(char *)&sockerr, &sockerr_len) != 0)
2490 		goto out;
2491 	if (sockerr < 0)
2492 		xprt_wake_pending_tasks(&transport->xprt, sockerr);
2493 out:
2494 	mutex_unlock(&transport->recv_mutex);
2495 }
2496 
2497 static void xs_wake_pending(struct sock_xprt *transport)
2498 {
2499 	if (test_and_clear_bit(XPRT_SOCK_WAKE_PENDING, &transport->sock_state))
2500 		xprt_wake_pending_tasks(&transport->xprt, -EAGAIN);
2501 }
2502 
2503 static void xs_error_handle(struct work_struct *work)
2504 {
2505 	struct sock_xprt *transport = container_of(work,
2506 			struct sock_xprt, error_worker);
2507 
2508 	xs_wake_disconnect(transport);
2509 	xs_wake_write(transport);
2510 	xs_wake_error(transport);
2511 	xs_wake_pending(transport);
2512 }
2513 
2514 /**
2515  * xs_local_print_stats - display AF_LOCAL socket-specifc stats
2516  * @xprt: rpc_xprt struct containing statistics
2517  * @seq: output file
2518  *
2519  */
2520 static void xs_local_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2521 {
2522 	long idle_time = 0;
2523 
2524 	if (xprt_connected(xprt))
2525 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2526 
2527 	seq_printf(seq, "\txprt:\tlocal %lu %lu %lu %ld %lu %lu %lu "
2528 			"%llu %llu %lu %llu %llu\n",
2529 			xprt->stat.bind_count,
2530 			xprt->stat.connect_count,
2531 			xprt->stat.connect_time / HZ,
2532 			idle_time,
2533 			xprt->stat.sends,
2534 			xprt->stat.recvs,
2535 			xprt->stat.bad_xids,
2536 			xprt->stat.req_u,
2537 			xprt->stat.bklog_u,
2538 			xprt->stat.max_slots,
2539 			xprt->stat.sending_u,
2540 			xprt->stat.pending_u);
2541 }
2542 
2543 /**
2544  * xs_udp_print_stats - display UDP socket-specifc stats
2545  * @xprt: rpc_xprt struct containing statistics
2546  * @seq: output file
2547  *
2548  */
2549 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2550 {
2551 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2552 
2553 	seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %llu %llu "
2554 			"%lu %llu %llu\n",
2555 			transport->srcport,
2556 			xprt->stat.bind_count,
2557 			xprt->stat.sends,
2558 			xprt->stat.recvs,
2559 			xprt->stat.bad_xids,
2560 			xprt->stat.req_u,
2561 			xprt->stat.bklog_u,
2562 			xprt->stat.max_slots,
2563 			xprt->stat.sending_u,
2564 			xprt->stat.pending_u);
2565 }
2566 
2567 /**
2568  * xs_tcp_print_stats - display TCP socket-specifc stats
2569  * @xprt: rpc_xprt struct containing statistics
2570  * @seq: output file
2571  *
2572  */
2573 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2574 {
2575 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2576 	long idle_time = 0;
2577 
2578 	if (xprt_connected(xprt))
2579 		idle_time = (long)(jiffies - xprt->last_used) / HZ;
2580 
2581 	seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu "
2582 			"%llu %llu %lu %llu %llu\n",
2583 			transport->srcport,
2584 			xprt->stat.bind_count,
2585 			xprt->stat.connect_count,
2586 			xprt->stat.connect_time / HZ,
2587 			idle_time,
2588 			xprt->stat.sends,
2589 			xprt->stat.recvs,
2590 			xprt->stat.bad_xids,
2591 			xprt->stat.req_u,
2592 			xprt->stat.bklog_u,
2593 			xprt->stat.max_slots,
2594 			xprt->stat.sending_u,
2595 			xprt->stat.pending_u);
2596 }
2597 
2598 /*
2599  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2600  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2601  * to use the server side send routines.
2602  */
2603 static int bc_malloc(struct rpc_task *task)
2604 {
2605 	struct rpc_rqst *rqst = task->tk_rqstp;
2606 	size_t size = rqst->rq_callsize;
2607 	struct page *page;
2608 	struct rpc_buffer *buf;
2609 
2610 	if (size > PAGE_SIZE - sizeof(struct rpc_buffer)) {
2611 		WARN_ONCE(1, "xprtsock: large bc buffer request (size %zu)\n",
2612 			  size);
2613 		return -EINVAL;
2614 	}
2615 
2616 	page = alloc_page(GFP_KERNEL);
2617 	if (!page)
2618 		return -ENOMEM;
2619 
2620 	buf = page_address(page);
2621 	buf->len = PAGE_SIZE;
2622 
2623 	rqst->rq_buffer = buf->data;
2624 	rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
2625 	return 0;
2626 }
2627 
2628 /*
2629  * Free the space allocated in the bc_alloc routine
2630  */
2631 static void bc_free(struct rpc_task *task)
2632 {
2633 	void *buffer = task->tk_rqstp->rq_buffer;
2634 	struct rpc_buffer *buf;
2635 
2636 	buf = container_of(buffer, struct rpc_buffer, data);
2637 	free_page((unsigned long)buf);
2638 }
2639 
2640 /*
2641  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2642  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2643  */
2644 static int bc_sendto(struct rpc_rqst *req)
2645 {
2646 	int len;
2647 	struct xdr_buf *xbufp = &req->rq_snd_buf;
2648 	struct sock_xprt *transport =
2649 			container_of(req->rq_xprt, struct sock_xprt, xprt);
2650 	unsigned long headoff;
2651 	unsigned long tailoff;
2652 	struct page *tailpage;
2653 	struct msghdr msg = {
2654 		.msg_flags	= MSG_MORE
2655 	};
2656 	rpc_fraghdr marker = cpu_to_be32(RPC_LAST_STREAM_FRAGMENT |
2657 					 (u32)xbufp->len);
2658 	struct kvec iov = {
2659 		.iov_base	= &marker,
2660 		.iov_len	= sizeof(marker),
2661 	};
2662 
2663 	len = kernel_sendmsg(transport->sock, &msg, &iov, 1, iov.iov_len);
2664 	if (len != iov.iov_len)
2665 		return -EAGAIN;
2666 
2667 	tailpage = NULL;
2668 	if (xbufp->tail[0].iov_len)
2669 		tailpage = virt_to_page(xbufp->tail[0].iov_base);
2670 	tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2671 	headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2672 	len = svc_send_common(transport->sock, xbufp,
2673 			      virt_to_page(xbufp->head[0].iov_base), headoff,
2674 			      tailpage, tailoff);
2675 	if (len != xbufp->len)
2676 		return -EAGAIN;
2677 	return len;
2678 }
2679 
2680 /*
2681  * The send routine. Borrows from svc_send
2682  */
2683 static int bc_send_request(struct rpc_rqst *req)
2684 {
2685 	struct svc_xprt	*xprt;
2686 	int len;
2687 
2688 	dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2689 	/*
2690 	 * Get the server socket associated with this callback xprt
2691 	 */
2692 	xprt = req->rq_xprt->bc_xprt;
2693 
2694 	/*
2695 	 * Grab the mutex to serialize data as the connection is shared
2696 	 * with the fore channel
2697 	 */
2698 	mutex_lock(&xprt->xpt_mutex);
2699 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2700 		len = -ENOTCONN;
2701 	else
2702 		len = bc_sendto(req);
2703 	mutex_unlock(&xprt->xpt_mutex);
2704 
2705 	if (len > 0)
2706 		len = 0;
2707 
2708 	return len;
2709 }
2710 
2711 /*
2712  * The close routine. Since this is client initiated, we do nothing
2713  */
2714 
2715 static void bc_close(struct rpc_xprt *xprt)
2716 {
2717 }
2718 
2719 /*
2720  * The xprt destroy routine. Again, because this connection is client
2721  * initiated, we do nothing
2722  */
2723 
2724 static void bc_destroy(struct rpc_xprt *xprt)
2725 {
2726 	dprintk("RPC:       bc_destroy xprt %p\n", xprt);
2727 
2728 	xs_xprt_free(xprt);
2729 	module_put(THIS_MODULE);
2730 }
2731 
2732 static const struct rpc_xprt_ops xs_local_ops = {
2733 	.reserve_xprt		= xprt_reserve_xprt,
2734 	.release_xprt		= xprt_release_xprt,
2735 	.alloc_slot		= xprt_alloc_slot,
2736 	.free_slot		= xprt_free_slot,
2737 	.rpcbind		= xs_local_rpcbind,
2738 	.set_port		= xs_local_set_port,
2739 	.connect		= xs_local_connect,
2740 	.buf_alloc		= rpc_malloc,
2741 	.buf_free		= rpc_free,
2742 	.prepare_request	= xs_stream_prepare_request,
2743 	.send_request		= xs_local_send_request,
2744 	.wait_for_reply_request	= xprt_wait_for_reply_request_def,
2745 	.close			= xs_close,
2746 	.destroy		= xs_destroy,
2747 	.print_stats		= xs_local_print_stats,
2748 	.enable_swap		= xs_enable_swap,
2749 	.disable_swap		= xs_disable_swap,
2750 };
2751 
2752 static const struct rpc_xprt_ops xs_udp_ops = {
2753 	.set_buffer_size	= xs_udp_set_buffer_size,
2754 	.reserve_xprt		= xprt_reserve_xprt_cong,
2755 	.release_xprt		= xprt_release_xprt_cong,
2756 	.alloc_slot		= xprt_alloc_slot,
2757 	.free_slot		= xprt_free_slot,
2758 	.rpcbind		= rpcb_getport_async,
2759 	.set_port		= xs_set_port,
2760 	.connect		= xs_connect,
2761 	.buf_alloc		= rpc_malloc,
2762 	.buf_free		= rpc_free,
2763 	.send_request		= xs_udp_send_request,
2764 	.wait_for_reply_request	= xprt_wait_for_reply_request_rtt,
2765 	.timer			= xs_udp_timer,
2766 	.release_request	= xprt_release_rqst_cong,
2767 	.close			= xs_close,
2768 	.destroy		= xs_destroy,
2769 	.print_stats		= xs_udp_print_stats,
2770 	.enable_swap		= xs_enable_swap,
2771 	.disable_swap		= xs_disable_swap,
2772 	.inject_disconnect	= xs_inject_disconnect,
2773 };
2774 
2775 static const struct rpc_xprt_ops xs_tcp_ops = {
2776 	.reserve_xprt		= xprt_reserve_xprt,
2777 	.release_xprt		= xprt_release_xprt,
2778 	.alloc_slot		= xprt_alloc_slot,
2779 	.free_slot		= xprt_free_slot,
2780 	.rpcbind		= rpcb_getport_async,
2781 	.set_port		= xs_set_port,
2782 	.connect		= xs_connect,
2783 	.buf_alloc		= rpc_malloc,
2784 	.buf_free		= rpc_free,
2785 	.prepare_request	= xs_stream_prepare_request,
2786 	.send_request		= xs_tcp_send_request,
2787 	.wait_for_reply_request	= xprt_wait_for_reply_request_def,
2788 	.close			= xs_tcp_shutdown,
2789 	.destroy		= xs_destroy,
2790 	.set_connect_timeout	= xs_tcp_set_connect_timeout,
2791 	.print_stats		= xs_tcp_print_stats,
2792 	.enable_swap		= xs_enable_swap,
2793 	.disable_swap		= xs_disable_swap,
2794 	.inject_disconnect	= xs_inject_disconnect,
2795 #ifdef CONFIG_SUNRPC_BACKCHANNEL
2796 	.bc_setup		= xprt_setup_bc,
2797 	.bc_maxpayload		= xs_tcp_bc_maxpayload,
2798 	.bc_num_slots		= xprt_bc_max_slots,
2799 	.bc_free_rqst		= xprt_free_bc_rqst,
2800 	.bc_destroy		= xprt_destroy_bc,
2801 #endif
2802 };
2803 
2804 /*
2805  * The rpc_xprt_ops for the server backchannel
2806  */
2807 
2808 static const struct rpc_xprt_ops bc_tcp_ops = {
2809 	.reserve_xprt		= xprt_reserve_xprt,
2810 	.release_xprt		= xprt_release_xprt,
2811 	.alloc_slot		= xprt_alloc_slot,
2812 	.free_slot		= xprt_free_slot,
2813 	.buf_alloc		= bc_malloc,
2814 	.buf_free		= bc_free,
2815 	.send_request		= bc_send_request,
2816 	.wait_for_reply_request	= xprt_wait_for_reply_request_def,
2817 	.close			= bc_close,
2818 	.destroy		= bc_destroy,
2819 	.print_stats		= xs_tcp_print_stats,
2820 	.enable_swap		= xs_enable_swap,
2821 	.disable_swap		= xs_disable_swap,
2822 	.inject_disconnect	= xs_inject_disconnect,
2823 };
2824 
2825 static int xs_init_anyaddr(const int family, struct sockaddr *sap)
2826 {
2827 	static const struct sockaddr_in sin = {
2828 		.sin_family		= AF_INET,
2829 		.sin_addr.s_addr	= htonl(INADDR_ANY),
2830 	};
2831 	static const struct sockaddr_in6 sin6 = {
2832 		.sin6_family		= AF_INET6,
2833 		.sin6_addr		= IN6ADDR_ANY_INIT,
2834 	};
2835 
2836 	switch (family) {
2837 	case AF_LOCAL:
2838 		break;
2839 	case AF_INET:
2840 		memcpy(sap, &sin, sizeof(sin));
2841 		break;
2842 	case AF_INET6:
2843 		memcpy(sap, &sin6, sizeof(sin6));
2844 		break;
2845 	default:
2846 		dprintk("RPC:       %s: Bad address family\n", __func__);
2847 		return -EAFNOSUPPORT;
2848 	}
2849 	return 0;
2850 }
2851 
2852 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2853 				      unsigned int slot_table_size,
2854 				      unsigned int max_slot_table_size)
2855 {
2856 	struct rpc_xprt *xprt;
2857 	struct sock_xprt *new;
2858 
2859 	if (args->addrlen > sizeof(xprt->addr)) {
2860 		dprintk("RPC:       xs_setup_xprt: address too large\n");
2861 		return ERR_PTR(-EBADF);
2862 	}
2863 
2864 	xprt = xprt_alloc(args->net, sizeof(*new), slot_table_size,
2865 			max_slot_table_size);
2866 	if (xprt == NULL) {
2867 		dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2868 				"rpc_xprt\n");
2869 		return ERR_PTR(-ENOMEM);
2870 	}
2871 
2872 	new = container_of(xprt, struct sock_xprt, xprt);
2873 	mutex_init(&new->recv_mutex);
2874 	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2875 	xprt->addrlen = args->addrlen;
2876 	if (args->srcaddr)
2877 		memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2878 	else {
2879 		int err;
2880 		err = xs_init_anyaddr(args->dstaddr->sa_family,
2881 					(struct sockaddr *)&new->srcaddr);
2882 		if (err != 0) {
2883 			xprt_free(xprt);
2884 			return ERR_PTR(err);
2885 		}
2886 	}
2887 
2888 	return xprt;
2889 }
2890 
2891 static const struct rpc_timeout xs_local_default_timeout = {
2892 	.to_initval = 10 * HZ,
2893 	.to_maxval = 10 * HZ,
2894 	.to_retries = 2,
2895 };
2896 
2897 /**
2898  * xs_setup_local - Set up transport to use an AF_LOCAL socket
2899  * @args: rpc transport creation arguments
2900  *
2901  * AF_LOCAL is a "tpi_cots_ord" transport, just like TCP
2902  */
2903 static struct rpc_xprt *xs_setup_local(struct xprt_create *args)
2904 {
2905 	struct sockaddr_un *sun = (struct sockaddr_un *)args->dstaddr;
2906 	struct sock_xprt *transport;
2907 	struct rpc_xprt *xprt;
2908 	struct rpc_xprt *ret;
2909 
2910 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
2911 			xprt_max_tcp_slot_table_entries);
2912 	if (IS_ERR(xprt))
2913 		return xprt;
2914 	transport = container_of(xprt, struct sock_xprt, xprt);
2915 
2916 	xprt->prot = 0;
2917 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2918 
2919 	xprt->bind_timeout = XS_BIND_TO;
2920 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2921 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2922 
2923 	xprt->ops = &xs_local_ops;
2924 	xprt->timeout = &xs_local_default_timeout;
2925 
2926 	INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
2927 	INIT_WORK(&transport->error_worker, xs_error_handle);
2928 	INIT_DELAYED_WORK(&transport->connect_worker, xs_dummy_setup_socket);
2929 
2930 	switch (sun->sun_family) {
2931 	case AF_LOCAL:
2932 		if (sun->sun_path[0] != '/') {
2933 			dprintk("RPC:       bad AF_LOCAL address: %s\n",
2934 					sun->sun_path);
2935 			ret = ERR_PTR(-EINVAL);
2936 			goto out_err;
2937 		}
2938 		xprt_set_bound(xprt);
2939 		xs_format_peer_addresses(xprt, "local", RPCBIND_NETID_LOCAL);
2940 		ret = ERR_PTR(xs_local_setup_socket(transport));
2941 		if (ret)
2942 			goto out_err;
2943 		break;
2944 	default:
2945 		ret = ERR_PTR(-EAFNOSUPPORT);
2946 		goto out_err;
2947 	}
2948 
2949 	dprintk("RPC:       set up xprt to %s via AF_LOCAL\n",
2950 			xprt->address_strings[RPC_DISPLAY_ADDR]);
2951 
2952 	if (try_module_get(THIS_MODULE))
2953 		return xprt;
2954 	ret = ERR_PTR(-EINVAL);
2955 out_err:
2956 	xs_xprt_free(xprt);
2957 	return ret;
2958 }
2959 
2960 static const struct rpc_timeout xs_udp_default_timeout = {
2961 	.to_initval = 5 * HZ,
2962 	.to_maxval = 30 * HZ,
2963 	.to_increment = 5 * HZ,
2964 	.to_retries = 5,
2965 };
2966 
2967 /**
2968  * xs_setup_udp - Set up transport to use a UDP socket
2969  * @args: rpc transport creation arguments
2970  *
2971  */
2972 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2973 {
2974 	struct sockaddr *addr = args->dstaddr;
2975 	struct rpc_xprt *xprt;
2976 	struct sock_xprt *transport;
2977 	struct rpc_xprt *ret;
2978 
2979 	xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries,
2980 			xprt_udp_slot_table_entries);
2981 	if (IS_ERR(xprt))
2982 		return xprt;
2983 	transport = container_of(xprt, struct sock_xprt, xprt);
2984 
2985 	xprt->prot = IPPROTO_UDP;
2986 	/* XXX: header size can vary due to auth type, IPv6, etc. */
2987 	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2988 
2989 	xprt->bind_timeout = XS_BIND_TO;
2990 	xprt->reestablish_timeout = XS_UDP_REEST_TO;
2991 	xprt->idle_timeout = XS_IDLE_DISC_TO;
2992 
2993 	xprt->ops = &xs_udp_ops;
2994 
2995 	xprt->timeout = &xs_udp_default_timeout;
2996 
2997 	INIT_WORK(&transport->recv_worker, xs_udp_data_receive_workfn);
2998 	INIT_WORK(&transport->error_worker, xs_error_handle);
2999 	INIT_DELAYED_WORK(&transport->connect_worker, xs_udp_setup_socket);
3000 
3001 	switch (addr->sa_family) {
3002 	case AF_INET:
3003 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3004 			xprt_set_bound(xprt);
3005 
3006 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
3007 		break;
3008 	case AF_INET6:
3009 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3010 			xprt_set_bound(xprt);
3011 
3012 		xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
3013 		break;
3014 	default:
3015 		ret = ERR_PTR(-EAFNOSUPPORT);
3016 		goto out_err;
3017 	}
3018 
3019 	if (xprt_bound(xprt))
3020 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3021 				xprt->address_strings[RPC_DISPLAY_ADDR],
3022 				xprt->address_strings[RPC_DISPLAY_PORT],
3023 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3024 	else
3025 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3026 				xprt->address_strings[RPC_DISPLAY_ADDR],
3027 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3028 
3029 	if (try_module_get(THIS_MODULE))
3030 		return xprt;
3031 	ret = ERR_PTR(-EINVAL);
3032 out_err:
3033 	xs_xprt_free(xprt);
3034 	return ret;
3035 }
3036 
3037 static const struct rpc_timeout xs_tcp_default_timeout = {
3038 	.to_initval = 60 * HZ,
3039 	.to_maxval = 60 * HZ,
3040 	.to_retries = 2,
3041 };
3042 
3043 /**
3044  * xs_setup_tcp - Set up transport to use a TCP socket
3045  * @args: rpc transport creation arguments
3046  *
3047  */
3048 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
3049 {
3050 	struct sockaddr *addr = args->dstaddr;
3051 	struct rpc_xprt *xprt;
3052 	struct sock_xprt *transport;
3053 	struct rpc_xprt *ret;
3054 	unsigned int max_slot_table_size = xprt_max_tcp_slot_table_entries;
3055 
3056 	if (args->flags & XPRT_CREATE_INFINITE_SLOTS)
3057 		max_slot_table_size = RPC_MAX_SLOT_TABLE_LIMIT;
3058 
3059 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3060 			max_slot_table_size);
3061 	if (IS_ERR(xprt))
3062 		return xprt;
3063 	transport = container_of(xprt, struct sock_xprt, xprt);
3064 
3065 	xprt->prot = IPPROTO_TCP;
3066 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3067 
3068 	xprt->bind_timeout = XS_BIND_TO;
3069 	xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
3070 	xprt->idle_timeout = XS_IDLE_DISC_TO;
3071 
3072 	xprt->ops = &xs_tcp_ops;
3073 	xprt->timeout = &xs_tcp_default_timeout;
3074 
3075 	xprt->max_reconnect_timeout = xprt->timeout->to_maxval;
3076 	xprt->connect_timeout = xprt->timeout->to_initval *
3077 		(xprt->timeout->to_retries + 1);
3078 
3079 	INIT_WORK(&transport->recv_worker, xs_stream_data_receive_workfn);
3080 	INIT_WORK(&transport->error_worker, xs_error_handle);
3081 	INIT_DELAYED_WORK(&transport->connect_worker, xs_tcp_setup_socket);
3082 
3083 	switch (addr->sa_family) {
3084 	case AF_INET:
3085 		if (((struct sockaddr_in *)addr)->sin_port != htons(0))
3086 			xprt_set_bound(xprt);
3087 
3088 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
3089 		break;
3090 	case AF_INET6:
3091 		if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
3092 			xprt_set_bound(xprt);
3093 
3094 		xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
3095 		break;
3096 	default:
3097 		ret = ERR_PTR(-EAFNOSUPPORT);
3098 		goto out_err;
3099 	}
3100 
3101 	if (xprt_bound(xprt))
3102 		dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3103 				xprt->address_strings[RPC_DISPLAY_ADDR],
3104 				xprt->address_strings[RPC_DISPLAY_PORT],
3105 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3106 	else
3107 		dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
3108 				xprt->address_strings[RPC_DISPLAY_ADDR],
3109 				xprt->address_strings[RPC_DISPLAY_PROTO]);
3110 
3111 	if (try_module_get(THIS_MODULE))
3112 		return xprt;
3113 	ret = ERR_PTR(-EINVAL);
3114 out_err:
3115 	xs_xprt_free(xprt);
3116 	return ret;
3117 }
3118 
3119 /**
3120  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
3121  * @args: rpc transport creation arguments
3122  *
3123  */
3124 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
3125 {
3126 	struct sockaddr *addr = args->dstaddr;
3127 	struct rpc_xprt *xprt;
3128 	struct sock_xprt *transport;
3129 	struct svc_sock *bc_sock;
3130 	struct rpc_xprt *ret;
3131 
3132 	xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries,
3133 			xprt_tcp_slot_table_entries);
3134 	if (IS_ERR(xprt))
3135 		return xprt;
3136 	transport = container_of(xprt, struct sock_xprt, xprt);
3137 
3138 	xprt->prot = IPPROTO_TCP;
3139 	xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
3140 	xprt->timeout = &xs_tcp_default_timeout;
3141 
3142 	/* backchannel */
3143 	xprt_set_bound(xprt);
3144 	xprt->bind_timeout = 0;
3145 	xprt->reestablish_timeout = 0;
3146 	xprt->idle_timeout = 0;
3147 
3148 	xprt->ops = &bc_tcp_ops;
3149 
3150 	switch (addr->sa_family) {
3151 	case AF_INET:
3152 		xs_format_peer_addresses(xprt, "tcp",
3153 					 RPCBIND_NETID_TCP);
3154 		break;
3155 	case AF_INET6:
3156 		xs_format_peer_addresses(xprt, "tcp",
3157 				   RPCBIND_NETID_TCP6);
3158 		break;
3159 	default:
3160 		ret = ERR_PTR(-EAFNOSUPPORT);
3161 		goto out_err;
3162 	}
3163 
3164 	dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
3165 			xprt->address_strings[RPC_DISPLAY_ADDR],
3166 			xprt->address_strings[RPC_DISPLAY_PORT],
3167 			xprt->address_strings[RPC_DISPLAY_PROTO]);
3168 
3169 	/*
3170 	 * Once we've associated a backchannel xprt with a connection,
3171 	 * we want to keep it around as long as the connection lasts,
3172 	 * in case we need to start using it for a backchannel again;
3173 	 * this reference won't be dropped until bc_xprt is destroyed.
3174 	 */
3175 	xprt_get(xprt);
3176 	args->bc_xprt->xpt_bc_xprt = xprt;
3177 	xprt->bc_xprt = args->bc_xprt;
3178 	bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
3179 	transport->sock = bc_sock->sk_sock;
3180 	transport->inet = bc_sock->sk_sk;
3181 
3182 	/*
3183 	 * Since we don't want connections for the backchannel, we set
3184 	 * the xprt status to connected
3185 	 */
3186 	xprt_set_connected(xprt);
3187 
3188 	if (try_module_get(THIS_MODULE))
3189 		return xprt;
3190 
3191 	args->bc_xprt->xpt_bc_xprt = NULL;
3192 	args->bc_xprt->xpt_bc_xps = NULL;
3193 	xprt_put(xprt);
3194 	ret = ERR_PTR(-EINVAL);
3195 out_err:
3196 	xs_xprt_free(xprt);
3197 	return ret;
3198 }
3199 
3200 static struct xprt_class	xs_local_transport = {
3201 	.list		= LIST_HEAD_INIT(xs_local_transport.list),
3202 	.name		= "named UNIX socket",
3203 	.owner		= THIS_MODULE,
3204 	.ident		= XPRT_TRANSPORT_LOCAL,
3205 	.setup		= xs_setup_local,
3206 };
3207 
3208 static struct xprt_class	xs_udp_transport = {
3209 	.list		= LIST_HEAD_INIT(xs_udp_transport.list),
3210 	.name		= "udp",
3211 	.owner		= THIS_MODULE,
3212 	.ident		= XPRT_TRANSPORT_UDP,
3213 	.setup		= xs_setup_udp,
3214 };
3215 
3216 static struct xprt_class	xs_tcp_transport = {
3217 	.list		= LIST_HEAD_INIT(xs_tcp_transport.list),
3218 	.name		= "tcp",
3219 	.owner		= THIS_MODULE,
3220 	.ident		= XPRT_TRANSPORT_TCP,
3221 	.setup		= xs_setup_tcp,
3222 };
3223 
3224 static struct xprt_class	xs_bc_tcp_transport = {
3225 	.list		= LIST_HEAD_INIT(xs_bc_tcp_transport.list),
3226 	.name		= "tcp NFSv4.1 backchannel",
3227 	.owner		= THIS_MODULE,
3228 	.ident		= XPRT_TRANSPORT_BC_TCP,
3229 	.setup		= xs_setup_bc_tcp,
3230 };
3231 
3232 /**
3233  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
3234  *
3235  */
3236 int init_socket_xprt(void)
3237 {
3238 	if (!sunrpc_table_header)
3239 		sunrpc_table_header = register_sysctl_table(sunrpc_table);
3240 
3241 	xprt_register_transport(&xs_local_transport);
3242 	xprt_register_transport(&xs_udp_transport);
3243 	xprt_register_transport(&xs_tcp_transport);
3244 	xprt_register_transport(&xs_bc_tcp_transport);
3245 
3246 	return 0;
3247 }
3248 
3249 /**
3250  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
3251  *
3252  */
3253 void cleanup_socket_xprt(void)
3254 {
3255 	if (sunrpc_table_header) {
3256 		unregister_sysctl_table(sunrpc_table_header);
3257 		sunrpc_table_header = NULL;
3258 	}
3259 
3260 	xprt_unregister_transport(&xs_local_transport);
3261 	xprt_unregister_transport(&xs_udp_transport);
3262 	xprt_unregister_transport(&xs_tcp_transport);
3263 	xprt_unregister_transport(&xs_bc_tcp_transport);
3264 }
3265 
3266 static int param_set_uint_minmax(const char *val,
3267 		const struct kernel_param *kp,
3268 		unsigned int min, unsigned int max)
3269 {
3270 	unsigned int num;
3271 	int ret;
3272 
3273 	if (!val)
3274 		return -EINVAL;
3275 	ret = kstrtouint(val, 0, &num);
3276 	if (ret)
3277 		return ret;
3278 	if (num < min || num > max)
3279 		return -EINVAL;
3280 	*((unsigned int *)kp->arg) = num;
3281 	return 0;
3282 }
3283 
3284 static int param_set_portnr(const char *val, const struct kernel_param *kp)
3285 {
3286 	return param_set_uint_minmax(val, kp,
3287 			RPC_MIN_RESVPORT,
3288 			RPC_MAX_RESVPORT);
3289 }
3290 
3291 static const struct kernel_param_ops param_ops_portnr = {
3292 	.set = param_set_portnr,
3293 	.get = param_get_uint,
3294 };
3295 
3296 #define param_check_portnr(name, p) \
3297 	__param_check(name, p, unsigned int);
3298 
3299 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
3300 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
3301 
3302 static int param_set_slot_table_size(const char *val,
3303 				     const struct kernel_param *kp)
3304 {
3305 	return param_set_uint_minmax(val, kp,
3306 			RPC_MIN_SLOT_TABLE,
3307 			RPC_MAX_SLOT_TABLE);
3308 }
3309 
3310 static const struct kernel_param_ops param_ops_slot_table_size = {
3311 	.set = param_set_slot_table_size,
3312 	.get = param_get_uint,
3313 };
3314 
3315 #define param_check_slot_table_size(name, p) \
3316 	__param_check(name, p, unsigned int);
3317 
3318 static int param_set_max_slot_table_size(const char *val,
3319 				     const struct kernel_param *kp)
3320 {
3321 	return param_set_uint_minmax(val, kp,
3322 			RPC_MIN_SLOT_TABLE,
3323 			RPC_MAX_SLOT_TABLE_LIMIT);
3324 }
3325 
3326 static const struct kernel_param_ops param_ops_max_slot_table_size = {
3327 	.set = param_set_max_slot_table_size,
3328 	.get = param_get_uint,
3329 };
3330 
3331 #define param_check_max_slot_table_size(name, p) \
3332 	__param_check(name, p, unsigned int);
3333 
3334 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
3335 		   slot_table_size, 0644);
3336 module_param_named(tcp_max_slot_table_entries, xprt_max_tcp_slot_table_entries,
3337 		   max_slot_table_size, 0644);
3338 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
3339 		   slot_table_size, 0644);
3340